An investigation of the function of
adaptor protein complex 4 (AP-4)
Discovering a role for AP-4 in the
spatial control of autophagy
Alexandra Katherine Davies
Department of Clinical Biochemistry
University of Cambridge
This dissertation is submitted for the degree of
Doctor of Philosophy
Corpus Christi College September 2018

To my family, and to Robbie, Molly and all the families affected by AP-4 deficiency.

Declaration
I hereby declare that except where specific reference is made to the work of others, the
contents of this thesis, entitled "An investigation of the function of adaptor protein
complex 4 (AP-4)", are original and have not been submitted in whole or in part for
consideration for any other degree or qualification in this, or any other University. This
dissertation is the result of my own work and includes nothing which is the outcome
of work done in collaboration, except where specifically indicated in the text. This
thesis does not exceed the 60,000 word limit stipulated by the Department of Clinical
Biochemistry.
Alexandra Katherine Davies
September 2018

Acknowledgements
There are many people I wish to thank who have enabled me to conduct this research
and write my thesis. My utmost thanks goes to my supervisor Scottie Robinson for the
opportunity to do my PhD in her lab, for a fascinating research topic and for giving me
an offer I could not refuse. How could I turn down a rotation project that offered the
chance to study a rare disease and a trip to Munich to work in a world-class proteomics
lab? I am so grateful for all the guidance, resources and encouragement you have given
me, and your unbounded enthusiasm for your research is a constant inspiration. A huge
thanks also goes to my second supervisor Georg Borner for sharing the AP-4 project with
me, for very clever proteomics methods and for patiently teaching me everything I now
know about mass spectrometry and statistics.
A big thanks also to everyone in the Robinson Lab for making it a great place to do science
and for being there for the many ups, but also the downs, of my PhD roller coaster ride.
Thanks to Jenny Hirst for lots of advice, shared interest in rare disease and for providing
the AP-4 patient cell lines for this project. The moment I saw the ‘stonking’ accumulation
of ATG9A at the TGN of the patient cells stands out as a high point of my PhD. Thanks to
Paul Manna for interesting discussions, molecular biology advice and for not laughing
(too much) when I made rookie errors. Thanks to James Edgar for the beautiful EM
presented in this thesis, for finding me AP-4 vesicles (another PhD highlight) and for lots
of help when everything got a bit mad with paper submissions. Special thanks to my
fellow Robinson lab PhD student Paloma Navarro for great company in and out of the
lab, and for lots of advice and encouragement, even now from California.
There are lots of other people to thank at the CIMR who make it a wonderful place to work
and I am grateful for you all. In particular, thanks to Guy Pearson and Niko Amin-Wetzel
for being with me for the whole journey of our PhDs. Also special thanks to Zuzana
Kadlecova for always being there for me at the lab, but also outside of it. I also wish to
extend my gratitude to everyone who considered me worthy of the opportunity to study
at CIMR and to Paul Lehner and Stefan Marciniak for excellent rotation projects in their
viii
labs. My PhD work was made possible by funding from the CIMR, the National Institute
for Health Research (NIHR) and an EMBO Short Term Fellowship. I am also grateful to
Corpus Christi College for their support during the time of my PhD.
I am very grateful to Matthias Mann for enabling me to conduct my research with
Georg Borner in his laboratory at the Max Planck Institute for Biochemistry. I have been
spoilt to get to train in such an incredible environment. Thanks also to everyone in the
Department of Proteomics and Signal Transduction for making me welcome. Special
thanks to Daniel Itzhak for teaching me to make maps and to the Cox group gang for lots
of fun and good company. I am also very thankful for an excellent collaboration with
Lauren Parker Jackson, Meredith Frazier and Tara Archuleta (University of Vanderbilt),
work from which is included in this thesis.
I would also like to thank some of the people who got me to where I am today. At Mallaig
High School, thanks to Kate Mundell for encouraging me in my Advanced Higher Biology
class of one. At the University of Glasgow, thanks to all the Genetics crew, especially
Kevin O’Dell and Joe Gray for converting me from Marine Biology. My time at Glasgow
shaped my interest in understanding the molecular mechanisms of disease and gave me
the confidence to pursue it. I am particularly grateful to Tom Van Agtmael for excellent
supervision of my Honours project, encouraging me to go for Cambridge and for a well
timed lunch and pep talk near the end of my thesis writing experience.
A very special mention is deserved by Cure SPG47, particularly Kasey, Chris, Angela and
Kevin, who are working tirelessly to move research on AP-4 deficiency forward. Your
work and your positive approach to life provided me context and inspiration during the
more challenging moments of my final PhD year.
Last, but definitely not least, I would like to say a huge thank you to all my friends and
family. I could not have done this without you. I am sorry there is not space to thank
you all here. I would like to make a special shout out to Kerry and Charlotte for their
friendship and support in Cambridge, particularly over the last year, to my Corpus pals
for good times since 2013, to Steph for camaraderie from Glasgow to AZ to PhD, and
to Sinead and Mairead for being there for me all the way from Mallaig to Glasgow and
beyond. To Margaret and Jim, thanks for making me so welcome in your home during my
time at Mallaig and always since then. To my sister Pippa, thank you for your friendship,
support and for only occasionally making fun of my geekiness. To mum and dad, I
cannot thank you enough for all your love, encouragement and belief in me. Writing
my thesis at Doune, with your support (and lobsters), was a truly enjoyable experience.
Finally, to Sandy, for whom there is too much to thank, thank you for being there.
Abstract
Vesicle trafficking provides the solution to the ‘sorting problem’ – how the eukaryotic cell
maintains the distinct identities, and thus functional properties, of its membrane-bound
organelles. During vesicle trafficking, proteins are selectively sorted into membrane-
bound transport intermediates by vesicle adaptors, which include those of the highly
conserved adaptor protein (AP) complex family. Each AP complex has a distinct sub-
cellular localisation and functions in the sorting of a specific subset of transmembrane
cargo proteins. Adaptor protein complex 4 (AP-4) is one of the more recently identified
AP complexes, whose function has largely remained elusive. In humans, AP-4 deficiency
causes a severe neurological disorder, suggesting an important role in neuronal devel-
opment and homeostasis. However, the pathomechanisms that underly the neuronal
pathology in AP-4 deficiency are currently unknown.
AP-4 is proposed to function in protein sorting at the trans-Golgi network (TGN), so
AP-4 deficiency can be thought of as a disease of missorting. The aim of this study
was to apply unbiased global proteomic approaches to define the composition of AP-4
vesicles and to identify physiological cargo proteins of the AP-4 pathway. Using ‘Dynamic
Organellar Maps’ and comparative analysis of vesicle-enriched fractions from wild-type
and AP-4-depleted cells, three ubiquitously expressed transmembrane cargo proteins,
ATG9A, SERINC1 and SERINC3, were found to be mislocalised in AP-4-deficient cells.
Two novel cytosolic AP-4 accessory proteins, RUSC1 and RUSC2, were also identified.
Further proteomic analyses confirmed the interactions between these proteins. AP-4
deficiency was found to cause missorting of ATG9A in diverse cell types, including patient-
derived cells, as well as dysregulation of autophagy. RUSC2 facilitates the transport of
AP-4-derived, ATG9A and SERINC-positive vesicles from the TGN to the cell periphery.
These vesicles cluster in close association with autophagosomes, suggesting they are the
‘ATG9 reservoir’ required for autophagosome biogenesis. This study uncovers ATG9A
trafficking as a ubiquitous function of the AP-4 pathway. Furthermore, it provides a
potential molecular pathomechanism of AP-4 deficiency, through dysregulated spatial
control of autophagy.

Table of contents
List of figures xvii
List of tables xxi
Abbreviations xxvii
1 Introduction 1
1.1 The sorting problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Coated vesicles - a solution to the sorting problem . . . . . . . . . . . . . . . . . 3
1.2.1 The clathrin coat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.2 Clathrin adaptors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.3 Clathrin-independent AP complexes . . . . . . . . . . . . . . . . . . . . . . 9
1.2.4 Different kinds of coated vesicles . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.5 Additional endosomal protein sorting complexes . . . . . . . . . . . . 12
1.2.6 The solution to the sorting problem . . . . . . . . . . . . . . . . . . . . . . 14
1.3 The stages of vesicular trafficking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 Initiation and cargo selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Vesicle budding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.3 Scission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.4 Uncoating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.5 Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.6 Tethering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.3.7 Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4 Adaptor protein complex 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4.1 The discovery and characterisation of AP-4 . . . . . . . . . . . . . . . . . 26
1.4.2 TEPSIN - an AP-4 ear-binding accessory protein . . . . . . . . . . . . . 29
1.4.3 AP-4 deficiency causes a severe neurological disorder . . . . . . . . . 31
1.4.4 The search for AP-4 cargo proteins . . . . . . . . . . . . . . . . . . . . . . . 32
xii Table of contents
1.5 Project Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2 Materials and Methods 37
2.1 Reagents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.1 Antibodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.2 DNA constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 Generation of polyclonal antibodies against SERINC1 and SERINC3 . . . . 42
2.2.1 Production of antigens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.2 Production and purification of antibodies . . . . . . . . . . . . . . . . . . 42
2.3 Cell culture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.1 Cell lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.2 Patient fibroblasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3.3 SILAC metabolic labelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3.4 Transient transfections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.3.5 Generation of stable cell lines . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4 CRISPR/Cas9-mediated gene editing . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4.1 Design and cloning of guides . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4.2 Knockout of AP4B1 and AP4E1 in HeLa cells . . . . . . . . . . . . . . . . 47
2.4.3 Depletion of AP4B1 and AP4E1 in SH-SY5Y cells . . . . . . . . . . . . . 48
2.4.4 Endogenous tagging of SERINC1 and SERINC3 . . . . . . . . . . . . . . 48
2.5 siRNA-mediated knockdown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.6 Quantitative RT-PCR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.7 Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.7.1 Fluorescence microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.7.2 Automated imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.7.3 Correlative light and electron microscopy (CLEM) . . . . . . . . . . . . 54
2.8 Western blotting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.9 GST pulldowns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.9.1 Preparation of cytosol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.9.2 Protein expression and purification . . . . . . . . . . . . . . . . . . . . . . . 56
2.9.3 GST pulldowns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.10 Immunoprecipitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.10.1 Immunoprecipitation of AP4B1 and AP4E1 . . . . . . . . . . . . . . . . . 57
2.10.2 GFP-trap of GFP-tagged TEPSIN and RUSC2 . . . . . . . . . . . . . . . . 58
2.10.3 Conventional immunoprecipitation of TEPSIN-GFP for MS . . . . . 58
2.10.4 Sensitive immunoprecipitation of TEPSIN-GFP for MS . . . . . . . . 59
2.11 BioID streptavidin pulldowns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Table of contents xiii
2.12 Preparation of vesicle-enriched fractions . . . . . . . . . . . . . . . . . . . . . . . . 60
2.13 Generation of Dynamic Organellar Maps and membrane fractions . . . . . 61
2.14 Mass spectrometry methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.14.1 In-solution digestion of proteins . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.14.2 Peptide purification and fractionation . . . . . . . . . . . . . . . . . . . . . 63
2.14.3 Mass spectrometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.14.4 Processing of mass spectrometry data . . . . . . . . . . . . . . . . . . . . . 64
2.15 Proteomic data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.15.1 Dynamic Organellar Maps statistical analysis . . . . . . . . . . . . . . . 65
2.15.2 Membrane fraction analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.15.3 Whole cell lysate analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.15.4 Vesicle fraction analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.15.5 TEPSIN-GFP immunoprecipitations . . . . . . . . . . . . . . . . . . . . . . 68
2.15.6 BioID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3 Proteomic investigations of adaptor protein complex 4 71
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.1.1 Mass spectrometry-based proteomics . . . . . . . . . . . . . . . . . . . . . 72
3.1.2 Organellar proteomics and vesicle trafficking . . . . . . . . . . . . . . . 74
3.1.3 Spatial proteomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.1.4 Proteomics to study AP-4-mediated vesicle trafficking . . . . . . . . . 78
3.2 AP-4 knockout HeLa cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.2.1 The Double Nickase CRISPR/Cas9 system . . . . . . . . . . . . . . . . . . 80
3.2.2 Guide RNA design and cloning . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.2.3 Generation of AP-4 knockout HeLa cells . . . . . . . . . . . . . . . . . . . 82
3.3 Dynamic Organellar Maps of AP-4 knockout cells . . . . . . . . . . . . . . . . . . 89
3.3.1 Protein translocations in AP-4 knockout cells . . . . . . . . . . . . . . . 89
3.3.2 Further analysis of the Dynamic Organellar Maps . . . . . . . . . . . . 94
3.4 Comparative vesicle profiling of AP-4 depleted cells . . . . . . . . . . . . . . . . 98
3.4.1 Preparation and MS analysis of AP-4-depleted vesicle fractions . . 99
3.5 BioID to search for AP-4 cargo and machinery . . . . . . . . . . . . . . . . . . . . 106
3.5.1 Generation and validation of AP-4 BioID cell lines . . . . . . . . . . . . 106
3.5.2 Identification of AP-4 proximal proteins by mass spectrometry . . . 112
3.5.3 Reciprocal BioID with SERINC1 and SERINC3 as baits . . . . . . . . . 118
3.6 Sensitive immunoprecipitation to reveal AP-4 cargo . . . . . . . . . . . . . . . . 126
3.6.1 Identification of AP-4 interacting partners by MS analysis of TEPSIN-
GFP immunoprecipitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
xiv Table of contents
3.7 Additional proteomic analyses of AP-4 knockout cells . . . . . . . . . . . . . . . 133
3.7.1 Global proteome analysis of AP-4 knockout cells . . . . . . . . . . . . . 133
3.7.2 The membrane fraction of AP-4 knockout cells . . . . . . . . . . . . . . 136
3.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4 Functional characterisation of AP-4 cargo and machinery 147
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
4.1.1 A brief introduction to autophagy . . . . . . . . . . . . . . . . . . . . . . . . 147
4.1.2 Background to ATG9A - the transmembrane autophagy protein . . 151
4.1.3 Background to the SERINC protein family . . . . . . . . . . . . . . . . . . 155
4.1.4 Background to the RUSC protein family . . . . . . . . . . . . . . . . . . . 158
4.2 Interaction between AP-4 and TEPSIN . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.2.1 AP4B1 appendage domain and TEPSIN binding motifs . . . . . . . . 161
4.2.2 The in vivo effect of mutating the AP4B1 TEPSIN-binding motif . . 165
4.2.3 The in vivo effect of mutating the TEPSIN AP4B1-binding motif . . 167
4.3 ATG9A in AP-4-deficient cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.3.1 ATG9A accumulates at the TGN of AP-4-deficient HeLa cells . . . . 172
4.3.2 ATG9A accumulates at the TGN of AP-4-depleted neuroblastoma-
derived SH-SY5Y cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
4.3.3 ATG9A mislocalisation is a ubiquitous phenotype in cells from
AP-4-deficient patients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.3.4 ATG9A levels in AP-4-deficient cells . . . . . . . . . . . . . . . . . . . . . . . 180
4.4 SERINC1 and SERINC3 in AP-4-deficient cells . . . . . . . . . . . . . . . . . . . . 183
4.4.1 Overexpressed SERINC3 loses its dependence on AP-4 for trafficking183
4.4.2 A CRISPR/Cas9-mediated gene-editing approach to localise en-
dogenous SERINC1 and SERINC3 . . . . . . . . . . . . . . . . . . . . . . . . 184
4.4.3 Knock in of Clover tags at the C termini of SERINC1 and SERINC3
in HeLa cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
4.4.4 ATG9A and SERINCs colocalise in AP-4-dependent vesicles . . . . . 193
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles . . . . . . . . . . . . 201
4.5.1 Vesicle fractionation profiles of AP-4-associated proteins . . . . . . . 201
4.5.2 AP-4 cargo proteins accumulate at the periphery of RUSC2 overex-
pressing cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.5.3 The RUSC2-driven accumulation of ATG9A/SERINC-containing
vesicles at the cell periphery depends on AP-4 . . . . . . . . . . . . . . . 209
4.5.4 RUSC2 is pulled down by AP-4 appendage domains . . . . . . . . . . . 213
4.5.5 The peripheral delivery of AP-4 vesicles is microtubule dependent 217
Table of contents xv
4.5.6 The effect of RUSC depletion on ATG9A localisation . . . . . . . . . . . 221
4.6 The role of AP-4 in the spatial control of autophagy . . . . . . . . . . . . . . . . 226
4.6.1 LC3B levels are elevated in AP-4 knockout HeLa cells . . . . . . . . . . 226
4.6.2 LC3B levels are elevated in RUSC-depleted HeLa cells . . . . . . . . . 231
4.6.3 AP-4 vesicles closely associate with autophagosomes . . . . . . . . . . 232
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
5 Discussion 243
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
5.2 Model for AP-4-dependent trafficking . . . . . . . . . . . . . . . . . . . . . . . . . . 245
5.3 AP-4 and the spatial control of autophagy . . . . . . . . . . . . . . . . . . . . . . . 248
5.4 Autophagy and neurological disease . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
5.5 The role of the RUSCs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
5.6 SERINCs in ATG9A vesicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
5.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
References 253
Appendix A Constructs 279
Appendix B PCR primers for cloning 281
Appendix C CRISPR guide sequences 285

List of figures
1.1 Electron microscopy reveals the internal complexity of the cell . . . . . . . . 2
1.2 Roth and Porter’s ‘bristle’-coated vesicle hypothesis . . . . . . . . . . . . . . . . 4
1.3 The structure of the clathrin coat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Adaptor protein complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 COPI and COPII vesicle coats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 An overview of membrane trafficking pathways . . . . . . . . . . . . . . . . . . . 15
1.7 The stages of vesicular trafficking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.8 Adaptor protein complex 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.9 TEPSIN - the AP-4 accessory protein . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1 The SILAC method for relative protein quantification . . . . . . . . . . . . . . . 74
3.2 Double Nickase CRISPR-Cas9 system . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3 Guide RNA design for knockout of AP4B1 and AP4E1 . . . . . . . . . . . . . . . 83
3.4 Generation of AP-4 knockout HeLa cells . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.5 Characterisation of clonal AP-4 knockout cell lines . . . . . . . . . . . . . . . . . 87
3.6 Workflow for Dynamic Organellar Map generation . . . . . . . . . . . . . . . . . 90
3.7 Experimental design for Dynamic Organellar Mapping . . . . . . . . . . . . . . 91
3.8 Protein translocations in AP-4 KO cells . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.9 Workflow for proteomic profiling of AP-4 vesicles . . . . . . . . . . . . . . . . . . 100
3.10 Pairwise comparison of control and AP-4 depleted vesicle fractions . . . . 102
3.11 Other vesicle machinery is unaffected by AP-4 depletion . . . . . . . . . . . . 104
3.12 Principal component analysis of vesicle fractions from AP-4-depleted cells105
3.13 Diagram of the BioID method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.14 AP-4 BioID cell line generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.15 AP-4 BioID fusion proteins colocalise with TEPSIN . . . . . . . . . . . . . . . . . 110
3.16 Promiscuous biotin ligase activity in AP-4 BioID cells . . . . . . . . . . . . . . . 111
3.17 AP-4 BioID fusion proteins biotinylate endogenous AP4E1 and TEPSIN . . 112
xviii List of figures
3.18 AP-4 BioID to screen for AP-4-associated proteins using mass spectrometry114
3.19 Topology of SERINCs and ATG9A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.20 Experimental confirmation of SERINC3 topology. . . . . . . . . . . . . . . . . . . 120
3.21 SERINC BioID cell lines express active biotin ligases . . . . . . . . . . . . . . . . 121
3.22 SERINC BioID to screen for SERINC-associated proteins using MS . . . . . 122
3.23 Hierarchical clustering of AP-4 and SERINC BioID data . . . . . . . . . . . . . 124
3.24 Conventional and sensitive immunoprecipitations of TEPSIN-GFP . . . . . 129
3.25 Global proteome analysis of AP-4 knockout HeLa cells . . . . . . . . . . . . . . 134
3.26 The membrane fraction of AP-4 knockout HeLa cells . . . . . . . . . . . . . . . 139
3.27 Estimated copy numbers of AP-4-associated proteins . . . . . . . . . . . . . . . 144
4.1 An overview of autophagy in mammalian cells . . . . . . . . . . . . . . . . . . . . 149
4.2 ATG9A - the transmembrane autophagy protein . . . . . . . . . . . . . . . . . . . 153
4.3 SERINC1 and SERINC3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.4 RUSC1 and RUSC2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.5 The TEPSIN C-terminal AP4B1 appendage-binding motif . . . . . . . . . . . . 163
4.6 The AP4B1 appendage domain TEPSIN-binding motif . . . . . . . . . . . . . . 164
4.7 Disruption of the AP4B1 appendage domain TEPSIN-binding site greatly
reduces TEPSIN binding to AP-4 in vivo . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.8 Disruption of the AP4B1 appendage domain TEPSIN-binding site does
not prevent TEPSIN recruitment to the membrane . . . . . . . . . . . . . . . . . 168
4.9 Disruption of the TEPSIN AP4B1-binding site reduces, but does not abol-
ish, TEPSIN binding to AP-4 in vivo . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
4.10 Mutant TEPSIN-GFP (L470S/F471S) does not colocalise with AP4E1 . . . . 171
4.11 ATG9A and AP-4 partially colocalise in the perinuclear region . . . . . . . . . 173
4.12 ATG9A accumulates at the TGN of AP-4 knockout HeLa cells . . . . . . . . . . 174
4.13 ATG9A accumulates at the TGN of AP-4 knockdown HeLa cells . . . . . . . . 175
4.14 Quantification of the TGN accumulation of ATG9A in AP-4 knockout HeLa
cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
4.15 ATG9A accumulates at the TGN of AP-4-depleted neuroblastoma-derived
SH-SY5Y cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
4.16 ATG9A mislocalisation is a ubiquitous phenotype in cells from AP-4-deficient
patients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.17 ATG9A expression level is increased in AP-4-deficient patient fibroblasts . 182
4.18 Overexpressed SERINC3 loses its dependence on AP-4 for trafficking . . . 185
4.19 SERINC1 and SERINC3 antibodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
List of figures xix
4.20 CRISPR/Cas9-mediated approach to knock in Clover tags at the C termini
of SERINC1 and SERINC3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4.21 Knock in of Clover tags at the C termini of SERINC1 and SERINC3 in HeLa
cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.22 Localisation of SERINC1 and SERINC3 in knockin HeLa cell lines . . . . . . 192
4.23 Endogenously tagged SERINC1 and SERINC3 are mistrafficked in AP-4-
depleted HeLa cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
4.24 ATG9A and SERINC1/3 colocalisation in peripheral puncta is dependent
on AP-4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
4.25 Quantification of ATG9A and SERINC1/3 colocalisation with and without
AP-4 knockdown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
4.26 SERINC1 and SERINC3 do not accumulate at the TGN of AP-4-depleted cells199
4.27 SERINC1 and SERINC3 partially colocalise with LAMP1 in control and
AP-4-depleted cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
4.28 RUSCs have more similar vesicle fractionation profiles to AP-4 cargo pro-
teins than to AP-4 itself . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
4.29 Generation of HeLa cell lines expressing GFP-tagged RUSC2 . . . . . . . . . . 206
4.30 ATG9A-positive puncta accumulate at the periphery of RUSC2-overexpressing
cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
4.31 SERINC-positive puncta accumulate at the periphery of RUSC2-overexpressing
cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4.32 The effect of RUSC2 overexpression is specific for AP-4 cargo proteins . . . 210
4.33 AP-4 does not accumulate at the periphery of RUSC2-overexpressing cells 211
4.34 CLEM reveals accumulations of small vesicles and tubules at the periphery
of RUSC2-overexpressing cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
4.35 RUSC2-driven accumulation of ATG9A-vesicles at the cell periphery de-
pends on AP-4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.36 ATG9A co-immunoprecipitates with GFP-RUSC2 only in the presence of
AP-4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
4.37 RUSC2 pulls down with AP4E1 and AP4B1 appendage domains . . . . . . . 216
4.38 RUSC2-positive puncta line up along microtubules . . . . . . . . . . . . . . . . 218
4.39 The microtubule network appears normal in RUSC2 overexpressing cells 219
4.40 The peripheral delivery of AP-4 vesicles in RUSC2-overexpressing cells is
microtubule-dependent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
4.41 Depletion of RUSCs by siRNA-mediated knockdown . . . . . . . . . . . . . . . 222
4.42 ATG9A localisation in RUSC-depleted cells . . . . . . . . . . . . . . . . . . . . . . . 224
xx List of figures
4.43 Airyscan enhanced resolution imaging of ATG9A in RUSC-depleted cells . 225
4.44 Loss of AP-4 in HeLa cells causes elevated LC3B levels . . . . . . . . . . . . . . 228
4.45 LC3B puncta are larger in AP-4 knockout cells . . . . . . . . . . . . . . . . . . . . 230
4.46 Loss of RUSCs in HeLa cells causes elevated LC3B levels . . . . . . . . . . . . . 233
4.47 Overexpression of RUSC2 does not affect LC3B levels . . . . . . . . . . . . . . . 234
4.48 RUSC2- and ATG9A-positive vesicles are found in close proximity to LC3B
puncta in starved cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
4.49 RUSC2 and LC3B are found on separate, but closely associated, structures 236
4.50 CLEM reveals that RUSC2-positive vesicles closely associate with autophago-
somes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
5.1 Proposed model of AP-4-dependent trafficking . . . . . . . . . . . . . . . . . . . 246
List of tables
2.1 Primary antibodies used in this study . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2 RUSC2 siRNA oligos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.3 Overview of mass spectrometric analyses . . . . . . . . . . . . . . . . . . . . . . . . 64
3.1 Protein translocations in AP-4 knockout cells . . . . . . . . . . . . . . . . . . . . . 94
3.2 SVM organelle assignment for ATG9A and SERINCs in WT and AP-4 KO cells 96
3.3 Endosomal neighbours of ATG9A and SERINCs in WT and KO cells . . . . . 97
3.4 Proteins lost from the vesicle fraction of AP-4-depleted cells . . . . . . . . . . 101
3.5 Proteins significantly enriched in AP-4 BioID streptavidin pulldowns . . . 116
3.6 AP complex subunits, RUSCs and ATG9A in the SERINC BioID data . . . . 125
3.7 AP-4 interacting proteins from a conventional IP of TEPSIN-GFP . . . . . . 130
3.8 AP-4 interacting proteins from sensitive IPs of TEPSIN-GFP . . . . . . . . . . 132
3.9 Proteins depleted from whole cell lysates of AP-4 KO cells . . . . . . . . . . . . 137
3.10 Proteins enriched in whole cell lysates of AP-4 KO cells . . . . . . . . . . . . . . 138
3.11 Summary of AP-4 proteomic experiments . . . . . . . . . . . . . . . . . . . . . . . 143
4.1 Top 10 similar profiles to RUSC1 from vesicle fractionation profiling . . . . 204
4.2 Top 10 similar profiles to RUSC2 from vesicle fractionation profiling . . . . 204
A.1 DNA constructs used in this study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
C.1 CRISPR guides for knockout of AP4B1 . . . . . . . . . . . . . . . . . . . . . . . . . . 286
C.2 CRISPR guides for knockout of AP4E1 . . . . . . . . . . . . . . . . . . . . . . . . . . 287

Abbreviations
Acronyms / Abbreviations
AAGAB Alpha- and gamma-adaptin-binding protein p34
AKTIP AKT-interacting protein
AP-MS Affinity purification and mass spectrometry
AP1M1 AP-1 complex subunit μ-1
AP4B1 AP-4 complex subunit β-1
AP4E1 AP-4 complex subunit ε-1
AP4M1 AP-4 complex subunit μ-1
AP4S1 AP-4 complex subunit σ-1
APP Amyloid precursor protein
ATG9A Autophagy-related protein 9A
BAR Bin-Amphiphysin-Rvs
BioID Proximity-dependent biotin identification
Cas9 CRISPR-associated endonuclease Cas9
CCDC88B Coiled-coil domain-containing protein 88B
CCV Clathrin-coated vesicle
CI-MPR Cation-independent mannose-6-phosphate receptor
xxiv Abbreviations
CID Collision-induced dissociation
CLEM Correlative light and electron microscopy
CLINT1 Clathrin interactor 1/EpsinR
COP Coat protein complex
crRNA CRISPR RNA
CSP Chemical shift perturbation
Cvt Cytoplasm-to-vacuole targeting
DAB2 Disabled homolog 2
DAGLB Sn1-specific diacylglycerol lipase beta
DFCP1 Double FYVE domain–containing protein
DSB Double-strand break
EGF Epidermal growth factor
EM Electron microscopy
ER Endoplasmic reticulum
EST Expressed sequence tag
FACS Fluorescence-activated cell sorting
FAM160A2 FTS and Hook-interacting protein
FCS Foetal calf serum
FDR False discovery rate
GABARAPL2 Gamma-aminobutyric acid receptor-associated protein-like 2
GAP GTPase-activating protein
GEF Guanine nucleotide exchange factor
GFP Green fluorescent protein
Abbreviations xxv
GGA Golgi-localised, γ ear-containing, ADP-ribosylation factor-binding protein
H Heavy
HOOK1 Protein Hook homolog 1
HSP Hereditary spastic paraplegia
HSQC Heteronuclear single quantum correlation
ILV Intraluminal vesicle
Indels Insertions or deletions
IRES Internal ribosome entry site
ITC Isothermal titration calorimetry
KLC Kinesin ligh chain
L Light
LAMP1 Lysosome-associated membrane glycoprotein 1
LC Liquid chromatography
LC3B Microtubule-associated proteins 1A/1B light chain 3B
LDL Low density lipoprotein
LOPIT Localisation of organelle proteins by isotope tagging
M Movement score
M6PR Cation-dependent mannose-6-phosphate receptor
MRI Magnetic resonance imaging
MS Mass spectrometry
MS/MS Tandem mass spectrometry
NGF Nerve growth factor
NHEJ Non-homologous end-joining
xxvi Abbreviations
NMD Nonsense-mediated decay
NMR Nuclear magnetic resonance
NTRK1 High affinity nerve growth factor receptor
PAGE Polyacrylamide gel electrophoresis
PAM Protospacer adjacent motif
PAS Pre-autophagosomal structure
PCA Principal component analysis
PCC Pearson’s Correlation Coefficient
PCP Protein correlation profiling
PCR Polymerase chain reaction
PE Phosphatidylethanolamine
PI(4,5)P2 Phosphatidylinositol-4,5-biphosphate
PI3K Phosphatidylinositol 3-kinase
PI3P Phosphatidylinositol 3-phosphate
PI4P Phosphatidylinositol-4-phosphate
PLK1 Serine/threonine-protein kinase PLK1
PS Phosphatidylserine
PTM Post-translational modification
R Reproducibility score
RabGAP Rab GTPase-activating protein
RUN RPIP8, UNC-14, and NESCA
RUSC RUN and SH3 domain-containing
SDS-PAGE Sodium dodecyl sulfate–polyacrylamide gel electrophoresis
Abbreviations xxvii
SERINC Serine incorporator
sgRNA short guide RNA
SILAC Stable isotope labelling by amino acids in cell culture
SNARE Soluble N-ethylmaleimide-sensitive factor attachment protein receptor
SSBR Single-strand break repair
SVM Support Vector Machine
TEPSIN AP-4 complex accessory subunit Tepsin
TGN Trans-Golgi network
TMD Transmembrane domain
TMT Tandem mass tagging
tracrRNA Trans-activating CRISPR RNA
V-ATPase Vacuolar-type H+-ATPase

Chapter 1
Introduction
1.1 The sorting problem
What separates us from a bacterium? Well undoubtedly quite a lot. But one answer
to this question is – membranes. One of the defining features of eukaryotic cells is
their compartmentalisation by a complex endomembrane system. The early days of
microscopy in the seventeenth century gave scientists their first glimpses into the cel-
lular world and during this time Robert Hooke coined the term ‘cell’ to describe the
‘infinite company of small Boxes’ that made up the structure of cork (in his 1665 book
Micrographia; Hooke 1665)1. By the end of the nineteenth century many of the now
familiar membrane-bound organelles of the eukaryotic cell had been identified, includ-
ing the nucleus, endoplasmic reticulum (ER), mitochondria and the Golgi apparatus
(Mazzarello, 1999). However, it was the use of the newly invented electron microscope
by scientists including Albert Claude, Keith Porter, and George Palade, in the 1950s,
that revealed the deeply complex internal organisation of eukaryotic cells (Figure 1.1).
The integration of ultrastructural analyses by electron microscopy (EM) and subcellular
fractionation-based biochemistry demonstrated the distinct biochemical properties,
and thus function, of different organelles (Palade, 1975).
The evolution of the endomembrane system brought many benefits, such as an increased
surface area to volume ratio, the ability to compartmentalise biochemical processes that
require diverse reaction conditions, and the selective concentration of reaction com-
ponents. But progress is not without its challenges. Proteins destined for the ER, Golgi
1At a recent Autophagy UK meeting held at Clare College, where I presented the findings of this thesis,
I had the privilege of viewing this amazing book.
2 Introduction
Fig. 1.1 Electron microscopy reveals the internal complexity of the cell. An early trans-
mission electron micrograph of a thin section of a secretory cell from guinea pig pancreas
showing endoplasmic reticulum at the left and Golgi apparatus at the right, with small
coated vesicles in between. Image by George E. Palade and made available by James D.
Jamieson and the Department of Cell Biology, Yale University School of Medicine. Image
citation: George E. Palade (2012) CIL:37239, Cavia porcellus, pancreatic cell. CIL. Dataset.
https://doi.org/doi:10.7295/W9CIL37239.
1.2 Coated vesicles - a solution to the sorting problem 3
apparatus, endosomes, lysosomes, and the plasma membrane, are all synthesised on
ribosomes attached to the membrane of the ER. How then do they get to where they need
to go? Seminal work from George Palade and colleagues during the 1960s demonstrated
the occurrence of vectorial transport between organelles in pancreatic exocrine cells
(Jamieson & Palade, 1967; Palade, 1975). Through a combination of autoradiography,
electron microscopy, fractionation and pulse-chase experiments, newly synthesised
secretory proteins were traced from their site of synthesis on ER-bound ribosomes,
through the Golgi apparatus and onwards to zymogen granules from where they were
destined for secretion. Along the way proteins were found in small membrane-bound
vesicles. This observation gave birth to the ‘vesicle transport hypothesis’, which states
that proteins move between organelles by means of shuttling transport vesicles.
Palade’s vesicle transport hypothesis predicted the budding of vesicles carrying protein
cargoes from a donor compartment, the targeting of vesicles to an acceptor compart-
ment, and the delivery of cargoes to the acceptor compartment via fusion of the limiting
membranes (Palade, 1975). From this hypothesis arose one of the major paradoxes of
cell biology - the ‘sorting problem’. If proteins can move between organelles by vesicular
transport, what stops the contents of different organelles from getting mixed up? How
do organelles retain their distinct identities? It was obvious that there must be a means
of selectivity to ensure only proteins destined for delivery to another organelle are pack-
aged into transport vesicles, whereas resident proteins of the donor organelle are left
behind.
1.2 Coated vesicles - a solution to the sorting problem
Around the same time as Palade’s ground-breaking studies on protein trafficking, Tom
Roth and Keith Porter published a classic paper in which they described ‘coated vesicles’
and ‘coated pits’ in electron micrographs of mosquito oocytes during the internalisation
of yolk proteins (Roth & Porter, 1964). Yolk proteins were taken up in 140 nm pits on
the surface of the oocytes, which had a 20 nm ‘bristle’ coat on their convex cytoplasmic
side (Figure 1.2). They suggested that these pits pinch off from the membrane to form
‘bristle-coated vesicles’, which subsequently uncoat and transport the absorbed yolk
proteins into the cell. Roth and Porter speculated a mechanical role for the bristle coat
and the possibility that it endowed selectivity to the uptake process. So right from this
initial identification of coated vesicles there was an idea that they might hold the key to
solving the sorting problem.
4 Introduction
Fig. 1.2 Roth and Porter’s ‘bristle’-coated vesicle hypothesis. Schematic drawing from Roth &
Porter (1964), summarising findings from electron micrographs of mosquito oocytes during the
uptake of yolk proteins into the cell. (1) A coated pit begins to invaginate from the cell surface;
(2) Coated pits extend until ready to pinch off from the plasma membrane; (3) Coated vesicles
are released into the cell; (4) Vesicles lose their ‘bristle’ coats once within the cell; (5) Vesicles can
fuse with other membrane-bound structures, sometimes leaving a flattened empty sac attached
to the droplet (7); (6) Vesicles continue to coalesce to form the large crystalline proteid yolk
bodies (8) of the oocyte. ER: endoplasmic reticulum; L: lipid.
1.2 Coated vesicles - a solution to the sorting problem 5
1.2.1 The clathrin coat
In 1975 Barbara Pearse, working at the MRC Laboratory of Molecular Biology in Cam-
bridge, purified a subcellular fraction enriched in coated vesicles and found it to be
dominated by a ∼180 kDa protein (Pearse, 1975). She named the protein ‘clathrin’. Her
subsequent high resolution EM analyses of purified coated vesicles, performed in col-
laboration with Tony Crowther and John Finch, revealed a coat structure consisting
of 12 pentagons and a variable number of hexagons (Crowther et al., 1976). This was
in agreement with an earlier study by Toku Kanaseki and Ken Kadota in which they
described coated vesicles as a ‘vesicle in a basket’ and compared the structure to the
shape of a football (Kanaseki & Kadota, 1969; Figure 1.3A). These were in fact the same
type of coated vesicle as those described earlier by Roth and Porter as ‘bristle-coated’
(Roth & Porter, 1964), but the coats appear bristle-like in two-dimensional thin sections.
As clathrin is the major protein in the coated vesicle fraction, the cage-like structures
were presumed to be made out of clathrin. This was formally shown in 1979 by Jim Keen
who demonstrated that it is possible to form basket-like structures using clathrin alone
(Keen et al., 1979). Later structural EM studies revealed that non-assembled clathrin
forms a three-legged structure known as a ‘triskelion’ (Ungewickell & Branton, 1981;
Figure 1.3B). Each triskelion consists of three copies of clathrin heavy chain (the ∼180
kDa protein identified by Pearse) and three copies of a smaller protein now known as
clathrin light chain. In more recent years, x-ray crystallography and cryo-EM studies
have revealed a detailed molecular view of how clathrin triskelia intertwine to form a
clathrin cage (Fotin et al., 2004; Figure 1.3C and D).
The structure of clathrin seems naturally suited to provide a scaffold for membrane
curvature during the formation of a vesicle. In an in vitro lipsosome-based system,
purified clathrin anchored to the membrane is capable of generating spherical buds,
suggesting the assembly of clathrin alone may be enough to produce the membrane
curvature required for vesicle budding (Dannhauser & Ungewickell, 2012). However,
there is still debate surrounding the situation in vivo where densely packed cargo proteins
are an energy barrier to membrane bending (Stachowiak et al., 2013; discussed further
in Section 1.3.2).
Around the same time as Pearse’s work on clathrin, Mike Brown, Joe Goldstein and Dick
Anderson described fibroblasts from a patient with familial hypercholesterolemia that
could bind low density lipoproteins (LDL) on their surface, but not internalise them
(Anderson et al., 1977). Electron microscopy revealed that on the surface of healthy cells,
LDL was concentrated in coated pits and vesicles, whereas on the patient’s cells, LDL
6 Introduction
Fig. 1.3 The structure of the clathrin coat. (A) High-power electron micrographs of negatively-
stained coated vesicles isolated from homogenised guinea pig brain tissue. Inset shows a candi-
date model of the lattice ‘soccer ball-like’ structure. Upper and middle panels labelled A and
B and the inset are from Kanaseki & Kadota (1969). Lower panels labelled C and D are from
T. Roth and show the same image before (C) and after (D) photographic enhancement using
Markam rotation. Image citation: Don W. Fawcett, Toku Kanaseki, Ken Kadota, Thomas Roth
(2011) CIL:11166, Cavia porcellus. CIL. Dataset. https://doi.org/doi:10.7295/W9CIL11166. (B)
Electron micrograph after rotary shadowing of a clathrin triskelion from Ungewickell & Branton
(1981). (C) Model of a clathrin triskelion labelled with the names of the segments of the heavy
chain and with the positions of clathrin light chains marked schematically; from Fotin et al.
(2004). (D) Image reconstruction of a clathrin hexagonal barrel, showing heavy chains only, at
7.9 Å, from Fotin et al. (2004).There are 36 clathrin triskelions in the structure.
1.2 Coated vesicles - a solution to the sorting problem 7
was excluded from these structures. This suggested to Brown, Goldstein and Anderson
that membrane proteins, including the LDL receptor, could be selectively concentrated
into coated pits that are the sites of receptor-mediated endocytosis. They hypothesised
this process would be mediated by ‘unknown peripheral membrane proteins’ that bind
to a ‘recognition site’ within the cytosolic domain of the receptor and postulated that
this could be the function of clathrin. It is now known that this is the job of clathrin-
binding ‘adaptors’ (see below), rather than of clathrin directly, but otherwise this was
a remarkably prescient prediction. This work later led to the first identification of an
endocytic sorting motif as the Brown and Goldstein lab found the sequence NVPY (now
extended to FXNPXY where X is any amino acid) in the cytosolic tail of the LDL receptor
to be required for its internalisation (Chen et al., 1990; Davis et al., 1986). Since then,
short linear sorting motifs in the cytoplasmic domains of transmembrane cargo proteins
have been discovered to be crucial to the selectivity of many membrane trafficking events
(see Section 1.3.1).
1.2.2 Clathrin adaptors
Although clathrin does form cage structures alone in vitro (Keen et al., 1979), it was
soon realised that in vivo clathrin does not work alone. The first hints to this came from
Pearse’s initial clathrin-coated vesicle (CCV) fraction in which she observed ‘Traces of
other proteins are visible on SDS-gel patterns of the final preparation’ (Pearse, 1975).
Pearse speculated that there may be minor vesicle-associated proteins with functions
such as the specification of sites for vesicle formation or the targeting of vesicles to a
specific destination. Keen found that a lower molecular weight fraction eluted from the
coated vesicle preparation greatly enhanced the assembly of clathrin cages when added
to a higher molecular weight fraction containing clathrin alone (Keen et al., 1979). The
lower molecular weight fraction contained one or more proteins of around 100–110 kDa,
which Keen named ‘assembly polypeptides’ or ‘APs’. Purification and characterisation
of these proteins was carried out by Margaret ‘Scottie’ Robinson working with Pearse
(Pearse & Robinson, 1984) and then by Keen (Keen, 1987). This led to the identification of
two different related protein complexes named HA-I and HA-II (for the hydroxylapatite
chromatography method used to separate them) by Robinson and Pearse and AP-1 and
AP-2 (for assembly polypeptides 1 and 2) by Keen. The names AP-1 and AP-2 stuck, but
are now used to stand for adaptor protein complex 1 and 2.
One of the most exciting early findings about AP-1 and AP-2 was that they appeared to
have distinct subcellular localisations; AP-1 is associated with intracellular membranes
8 Introduction
Fig. 1.4 Adaptor protein complexes. (A) Immunofluorescence microscopy of HeLa cells double
labelled for adaptor protein complexes 1 and 2 (AP-1 and AP-2). Image is from Robinson (2015).
(B) Structure of an AP complex: (1) Schematic diagram of AP-2 showing the core, linker regions
and appendage domains (‘ears’). All members of the AP complex family share this overall archi-
tecture; (2) Model of AP-2 from crystal structures of the core (trunk domains of α and β2 subunits,
μ2 and σ2; Collins et al., 2002) and the individual appendage domains (α ear and β2 ear; Owen
et al., 1999, 2000). The linkers are modelled. Interaction sites for clathrin, cargo and accessory
proteins are indicated. Model is by David Owen and reproduced from Robinson (2004).
in the perinuclear region (later shown to be the trans-Golgi network and endosomes)
and AP-2 is associated with coated pits on the plasma membrane (Ahle et al., 1988;
Robinson, 1987; Robinson & Pearse, 1986; Figure 1.4A). The 100–110 kDa AP-containing
fraction was also implicated in the attachment of clathrin to membrane (Unanue et al.,
1981). Together these data suggested that differentially localised clathrin adaptors might
explain how CCVs sort different protein cargoes in different parts of the cell (Robinson,
1987).
AP-1 and AP-2 are heterotetrameric protein complexes consisting of two larges subunits
(corresponding to the 100–110 kDa bands; γ and β1 in AP-1 and α and β2 in AP-2), a
medium subunit (∼50 kDa; μ1/2) and a small subunit (∼20 kDa; σ1/2). The N-terminal
domains of the large subunits, together with μ and σ, form a tightly folded core, while the
C-terminal domains of the large subunits form globular appendages or ‘ears’, which are
attached to the core via a long flexible linker (the ‘Mickey Mouse’ structure; Robinson,
2004; Figure 1.4B). Detailed analyses of the structure and function of AP-1 and AP-
2 complexes have revealed how their structure confers an ability to bind to clathrin,
protein and lipid components of the membrane, and accessory proteins required for
vesicle formation (reviewed in Owen et al., 2004). The core mediates interactions with
transmembrane cargo proteins and lipids, the ears act as binding platforms for vesicle
accessory proteins, and the flexible hinge region of the β subunits contains a clathrin-
1.2 Coated vesicles - a solution to the sorting problem 9
binding motif known as the ‘clathrin box’ (consensus LΦ[D/E]Φ[D/E]; where Φ is a bulky
hydrophobic residue; Dell’Angelica et al., 1998). Thus, adaptor protein complexes act as
master coordinators of vesicle formation and provide selectivity to the process through
binding of specific transmembrane cargo proteins via cytosolic motifs (discussed in
more detail below in Section 1.3.1).
Since the identification of AP-1 and AP-2, a number of alternative adaptors for clathrin-
mediated trafficking have been identified, including the GGA family, the epsinR/epsin
family, DAB2 and ARH (reviewed in Robinson, 2015). These serve to increase the reper-
toire of cargo proteins that can be incorporated into CCVs. The GGAs (named for Golgi-
localised, γ-ear-containing, ARF-binding proteins) are monomeric adaptors which were
cloned based on their homology to the γ ear domain of AP-1 (Dell’Angelica et al., 2000;
Hirst et al., 2000). In mammalian cells the GGAs (of which there are three – GGA1–
3) localise to the trans-Golgi network (TGN) and have partially overlapping function
with AP-1, functioning in the anterograde trafficking of lysosomal hydrolases (Hirst
et al., 2012). EpsinR and epsins1-3 are another family of monomeric adaptors, which
interact with AP-1 and AP-2, respectively. EpsinR is a cargo adaptor for the SNARE (Solu-
ble N-ethylmaleimide-sensitive factor attachment protein receptor) VTI1B (Hirst et al.,
2004), while the epsins are implicated in the sorting of ubiquitinated cargo via ubiquitin-
interacting motifs (UIMs; Chen & De Camilli, 2005). Both families of monomeric adap-
tors have membrane recruitment domains (e.g. GAT in the GGAs and ENTH in the
epsins/epsinR), cargo interaction domains (e.g. VHS in the GGAs and UIMs in the
epsins) and clathrin-binding motifs in their long unstructured linkers (Owen et al., 2004).
DAB2 (disabled homolog 2) and ARH (autosomal recessive hypercholesterolemia pro-
tein) are cargo-selective adaptors for the LDL receptor, binding to its sorting signal
FXNPXY as well as to AP-2 and clathrin (Mishra et al., 2002a,b) – Anderson, Brown and
Goldstein’s mystery ‘unknown peripheral membrane proteins’ (Anderson et al., 1977).
1.2.3 Clathrin-independent AP complexes
AP-1 and AP-2 are not the only members of the adaptor protein complex family. Robinson
and Bonifacino labs searched the expressed sequence tag (EST) database for proteins
with sequence homology to AP-1 and AP-2 subunits and independently identified two
further AP complexes, AP-3 and AP-4 (Dell’Angelica et al., 1999a, 1997a,b; Hirst et al.,
1999; Simpson et al., 1996). For many years it was assumed there were only four AP com-
plexes, but in 2011 Jennifer Hirst in the Robinson Lab identified a fifth AP complex (AP-5)
using structure-based searches, which had greater sequence divergence from the other
10 Introduction
APs (Hirst et al., 2011). Like AP-1 and 2, AP-3 and 4 are also heterotetramers consisting of
two large subunits (δ/ε and β3/4), a medium subunit (μ3/4) and a small subunit (σ3/4).
AP-5 exists as a heterohexamer consisting of the four AP complex subunits (ζ/β5/μ5/σ5)
plus two additional subunits spatacsin (SPG11) and spastizin (SPG15), encoded by the
genes SPG11 and ZFYVE26, respectively (Hirst et al., 2011, 2013a). However, differently
from AP-1 and 2, APs 3–5 appear to be able to function independently of clathrin. There
is no evidence to suggest AP-4 or AP-5 interact with clathrin, but the situation for AP-3
is more controversial. Although initially AP-3 was not found to associate with clathrin
in mammalian cells (Simpson et al., 1996), the β3 subunit was later found to contain
a clathrin box and bind to clathrin (Dell’Angelica et al., 1998). In addition, AP-3 and
clathrin partially colocalise in mammalian cells (Dell’Angelica et al., 1998; Peden et al.,
2004). However, Saccharomyces cerevisiae β3 lacks the clathrin box and AP-3 and clathrin
mutants have non-overlapping phenotypes, suggesting yeast AP-3 functions indepen-
dently of clathrin (Cowles et al., 1997). In β3-deficient mouse fibroblasts, mutant β3
lacking the clathrin box is able to rescue the LAMP1 missorting phenotype, suggesting
clathrin is not essential for AP-3 function (Peden et al., 2002). Therefore, the current
consensus is that AP-3 is able to function independently of clathrin, but that it may work
with clathrin in some organisms (Robinson, 2015).
The clathrin-independent AP complexes all localise to intracellular membranes, AP-3 to
endosomes, AP-4 to the TGN and AP-5 to LAMP1-positive late endosomes/lysosomes
(Hirst et al., 2011; Robinson, 2004). In HeLa cells, AP-3 is almost as abundant as AP-
1 and AP-2, whereas AP-4 and AP-5 are more than 30-fold less abundant (Hirst et al.,
2013b; Itzhak et al., 2016). AP-3 specialises in trafficking from a tubular endosomal
compartment to late endosomes/lysosomes and this is important for the biogenesis
of lysosome-related organelles such as melanosomes (Dell’Angelica et al., 1999b; Feng
et al., 1999). AP-4, the subject of this thesis, is discussed in further detail in Section 1.4.
AP-5 has recently been shown to function in retrograde trafficking of Golgi-associated
proteins away from late endosomes/lysosomes in a partially redundant pathway to
that mediated by retromer from multivesicular bodies (Hirst et al., 2018). This suggests
that AP-5 may provide a backup lysosomal retrieval pathway to retrieve proteins that
have been missed by other retrograde trafficking machinery at earlier organelles of the
endomembrane system.
1.2 Coated vesicles - a solution to the sorting problem 11
1.2.4 Different kinds of coated vesicles
In addition to CCVs, other kinds of coated vesicles have now been identified. Two papers
published by Jim Rothman and Dick Fine in 1980 built momentum for the idea that
CCVs provided the solution to the sorting problem (Rothman et al., 1980; Rothman &
Fine, 1980). They tracked a radiolabelled viral membrane glycoprotein from vesicular
stomatitis virus (VSV), VSV-G, through the secretory pathway by pulse-chase combined
with the isolation of CCVs. At earlier timepoints the VSV-G recovered in the CCVs was in
its high-mannose pre-Golgi form (Endo H-sensitive), whereas at later timepoints it was
in its trimmed post-Golgi form (Endo H-resistant). Rothman and Fine concluded that
CCVs must act on two successive transport routes during secretion, firstly from the ER to
the Golgi and secondly from the Golgi to the plasma membrane. However, clathrin has
since been shown to be non-essential for constitutive secretion and COPI- and COPII-
coated vesicles are now known to mediate transport in the early secretory pathway,
suggesting Rothman and Fine’s VSV-G-containing vesicles were not clathrin-coated, but
simply co-fractionated with CCVs (Robinson, 2015).
COPI (coat protein complex I) was discovered through a series of biochemical studies
in cell-free systems conducted by the Rothman lab (Malhotra et al., 1989) and COPII
(coat protein complex II) coat components were originally identified by genetic screens
in S. cerevisiae for defects in the secretory pathway (the ‘Sec’ mutants; Barlowe et al.,
1994; Novick et al., 1980). COPII functions in ER export and anterograde traffic between
the ER and Golgi, while COPI functions in ER retrieval from the Golgi (retrograde traf-
fic) and in intra-Golgi transport (Gomez-Navarro & Miller, 2016). The COPI coat is a
heteroheptamer known as ‘coatomer’ (Waters et al., 1991) consisting of two subcom-
plexes: the B subcomplex, a trimer of α-COP, β’-COP, and ε-COP, and the F subcomplex,
a tetramer of β-COP, γ-COP, δ-COP and ζ-COP. The F subcomplex is ancestrally related
to the heterotetrameric AP complex family (Schledzewski et al., 1999), while α-β’-COP
has a similar structural organisation to clathrin, consisting of N-terminal β-propellers
followed by extended α-solenoids (Devos et al., 2004). The COPII coat consists of a
heterodimer of SEC23 and SEC24, recruited to the membrane by the small GTPase SAR1,
which subsequently recruits a heterotetramer made up of two SEC13 and two SEC31
subunits (Lederkremer et al., 2001). While there is no sequence similarity between the
COPII components and AP or COPI complexes, SEC13 and SEC31 form an outer coat
with β-propeller and α-solenoid domains, like clathrin and α-β’-COP (Devos et al., 2004).
This suggests a common evolutionary origin for these membrane shaping proteins. Like
clathrin coats, the COPII coat can be divided into two layers - an inner coat consisting of
12 Introduction
SAR1, SEC23 and SEC24, which functions in membrane recruitment and cargo binding
(like the AP complexes), and an outer coat consisting of SEC13 and SEC31, which can self-
assemble into a cage-like structure (similarly to clathrin, but with a distinct geometry;
Stagg et al., 2006; Figure 1.5A and B). It was presumed that a similar organisation would
exist for the COPI coat, but a recent cryo-EM structure from the Briggs and Wieland
groups suggests otherwise (Dodonova et al., 2015). Regions of both COPI subcomplexes
(B and F) contact the membrane and α-β’-COP does not form a cage-like structure, but
instead forms an interconnected assembly with F subcomplexes, made up of a repeated
building block consisting of a ‘triad’ of coatomers (Figure 1.5C and D).
1.2.5 Additional endosomal protein sorting complexes
Additional sorting decisions must be made for proteins within the later stages of the
endocytic pathway - are they destined for degradation in the lysosome, or should they
be rescued from this fate? These sorting decisions are orchestrated by endosomally-
localised multi-protein complexes - the ESCRT (endosomal sorting complexes required
for transport) machinery sort cargo for lysosomal degradation while retrieval complexes
mediate the recycling of cargo back to other membrane compartments (recently re-
viewed in Cullen & Steinberg, 2018; McNally & Cullen, 2018). The ESCRT machinery
consists of four subcomplexes (ESCRT-0, -I, -II, -III) which are sequentially recruited to
the cytosolic side of the endosomal membrane. ESCRT-0, -I and -II function to concen-
trate cargo into a degradative subdomain and then the late-acting ESCRT-III complex
induces membrane remodelling to generate intraluminal vesicles (ILVs) that bud in-
wardly from the membrane, i.e. with the opposite topology to the budding of the coated
vesicles discussed above. Despite this different topology, clathrin is also implicated
in ESCRT-mediated sorting. Flat clathrin lattices on the surface of the endosome are
proposed to stabilise the degradative subdomain by binding to the ESCRT-0 component
HRS (Raiborg et al., 2006), rather than providing a scaffold for membrane reshaping as for
CCVs. The major sorting signal for the lysosomal degradation pathway is ubiquitination
(Katzmann et al., 2001), but there is also emerging evidence for ubiquitin-independent
cargo sorting into ILVs (Cullen & Steinberg, 2018).
Cargo recycling from endosomes was originally thought to be the default pathway
for proteins not tagged for lysosomal degradation and thus to occur in a sequence-
independent manner. However, the identification of the highly conserved retrieval com-
plexes, retromer and retriever, and the discovery of sorting motifs within the cytosolic
domains of recycling cargo has flipped this dogma on its head (Cullen & Steinberg, 2018).
1.2 Coated vesicles - a solution to the sorting problem 13
Fig. 1.5 COPI and COPII vesicle coats. (A) Schematic diagram demonstrating the architecture
of the COPII coat, adapted from Gomez-Navarro & Miller (2016). The COPII coat can be divided
into two layers: an inner coat consisting of heterodimers of SEC23 and SEC24, recruited to
the membrane by the small GTPase SAR1, and an outer coat consisting of heterotetramers
of SEC13 and SEC31. The outer coat assembles into a cage-like structure. (B) Model of the
SEC13/SEC31 cage from cryo-electron microscopy and single particle analysis by Stagg et al.
(2006). (C) Schematic diagram demonstrating the architecture of the COPI coat, adapted from
Gomez-Navarro & Miller (2016). The COPI coat consists of two subcomplexes: the B subcomplex
(α-β’-COP) and the F subcomplex (γ-ζ-β-δ-COP). Distal regions of both subcomplexes form arch-
like structures and the coat assembles from a repeated triangular building block called a ‘triad’.
(D) Model of the COPI coat from the fitting of crystal structures to cryo-electron tomography
and subtomogram averaging data by Dodonova et al. (2015).
14 Introduction
The opposite is now considered true – that the default route from the endosome is to-
wards lysosomal degradation, unless a cargo is rescued via sequence-dependent retrieval.
Retromer was identified in yeast as required for the endosome-to-Golgi retrieval of the
vacuolar protein sorting receptor Vps10 (Seaman et al., 1998). In mammals, retromer
is a heterotrimer consisting of VPS26, VPS35 and VPS29 subunits, which functions in
close association with a heterodimer of sorting nexin (SNX)-Bin/amphiphysin/Rvs (BAR)
proteins (SNX1 or SNX2 with SNX5, SNX6 or SNX32; Van Weering et al., 2012). Retromer
can bind cargo directly, but also acts in concert with cargo adaptors SNX3 and SNX27
for the retrieval of cargo to the TGN and the plasma membrane, respectively (McNally &
Cullen, 2018). Retriever, a recently discovered ‘retromer-like’ complex, is a heterotrimer
consisting of VPS29 (a shared subunit with retromer), DSCR3 (a paralogue of VPS26) and
C16orf62 (McNally et al., 2017). Similarly to retromer, retriever associates with cargo
adaptors SNX17 and SNX31 for sequence-dependent cargo recruitment. Both retromer
and retriever are dependent on the actin-nucleating WASH (Wiskott–Aldrich syndrome
and SCAR homologue) complex and the membrane tubulating activities of SNX-BAR
proteins to generate tubular transport intermediates for cargo export from the endosome.
Retriever additionally requires the CCC (CCDC93, CCDC22, COMMD) complex for its
localisation to endosomes (McNally et al., 2017). Together retromer, retriever, WASH
and CCC complexes are responsible for the correct endosomal recycling of hundreds of
transmembrane proteins (McNally & Cullen, 2018).
1.2.6 The solution to the sorting problem
The solution to the sorting problem is a battery of protein sorting complexes (includ-
ing those introduced above and others), which function in the sequence-dependent
selection of cargo for incorporation into vesicular or tubular transport intermediates
at defined locations in the cell. Figure 1.6 provides a diagrammatic summary of the
different membrane trafficking pathways that occur in a generic eukaryotic cell. The
steps involved in vesicle-mediated trafficking are discussed in further detail below.
1.2 Coated vesicles - a solution to the sorting problem 15
Fig. 1.6 An overview of membrane trafficking pathways. A schematic diagram summarising
the major trafficking pathways in a eukaryotic cell and the protein sorting machinery used to
mediate each step. AP-1 and AP-2, along with alternative adaptors, function in conjunction with
clathrin. The destination of the AP-4-mediated trafficking pathway was unresolved at the start of
this PhD project.
16 Introduction
1.3 The stages of vesicular trafficking
The half century since the identification of coated vesicles has yielded a detailed molec-
ular view of the steps involved in vesicle-mediated trafficking, although there is still a lot
left to learn. The complex process of vesicular trafficking can be broken down into 7 key
stages (summarised in Figure 1.7): (1) Initiation and cargo selection; (2) Vesicle budding
from the donor membrane; (3) Vesicle scission; (4) Vesicle uncoating; (5) Transport to
the acceptor compartment; (6) Tethering to the acceptor organelle; (7) Fusion with the
acceptor membrane. Each stage is discussed in detail below with a focus on coated
vesicles including clathrin-, COPI- and COPII-coated vesicles, but many of the molecular
principles also apply to tubule-mediated protein sorting, e.g. by retromer and retriever
complexes.
1.3.1 Initiation and cargo selection
The first step in the generation of a coated vesicle is the recruitment of the membrane-
proximal coat components (i.e. adaptors in the case of CCVs) to the appropriate intracel-
lular membrane. This involves a combination of activation and membrane-association
of small GTPases from the Ras family, phospholipid interactions and cargo binding. This
multi-step process ensures tight regulation and specificity of adaptor/coat recruitment.
The initiation of COPI, AP-1/3/4, and GGA1-3 vesicle formation begins with the activation
of the small GTPase ARF1, while COPII vesicle formation begins with the activation
of another small GTPase, SAR1 (reviewed in Donaldson & Jackson, 2011). Activation
involves the exchange of GDP for GTP by a guanine nucleotide exchange factor (GEF),
which triggers membrane association of the GTPase by insertion of a myristoyl group
and/or an amphipathic helix into the membrane. The membrane-associated GTP-
bound GTPase then recruits adaptor or coat components. For example, SAR1-GTP
binds to SEC23 to recruit the SEC23-SEC24 heterodimer of COPII to the ER membrane
(Yoshihisa et al., 1993). Similarly, ARF1-GTP binds to AP-1 β and γ subunits to bring
AP-1 onto TGN/endosome membranes (Ren et al., 2013). GTP hydrolysis back to GDP,
catalysed by GTPase-activating proteins (GAPs), triggers release of SAR1 and ARF1 from
the membrane. Thus, the membrane recruitment of these coat protein complexes is
blocked by the GEF inhibitor brefeldin A (Donaldson & Jackson, 2011). In contrast,
AP-2 is not sensitive to brefeldin A as it does not require ARF1-GTP for its membrane
recruitment.
1.3 The stages of vesicular trafficking 17
Fig. 1.7 The stages of vesicular trafficking. A schematic diagram that summaries the 7 key
stages of vesicle trafficking, adapted from Bonifacino (2014). (1) Initiation and cargo selection:
Membrane-proximal coat components are recruited to the donor compartment by binding to a
membrane-associated small GTPase and/or a specific phospholipid. Cargo proteins, including
SNAREs, begin to concentrate at the site of initiation via specific interactions with adaptors. (2)
Vesicle budding: Membrane-distal coat components, e.g. clathrin, bind and polymerise to form
a scaffold, which may drive membrane bending as the vesicle invaginates from the membrane.
(3) Scission: The neck between the vesicle bud and the donor compartment is severed to release
a free vesicle into the cytosol. (4) Uncoating: The vesicle loses its coat and the components are
recycled for further rounds of vesicle budding. (5) Transport: The vesicle is transported to the
acceptor compartment through the action of cytoskeletal transport machinery. (6) Tethering: The
vesicle is tethered to the acceptor compartment by specific tethering factors and Rab GTPases.
(7) Fusion: The vesicle fuses with the membrane of the acceptor compartment in a process
driven by the formation of a trans-SNARE complex, thereby transferring its cargo to the acceptor
compartment.
18 Introduction
The different phospholipid contents of different organellar membranes provides another
regulatory layer to control the specificity of adaptor and coat protein recruitment. AP-
2 preferentially binds to PI(4,5)P2, which is an abundant phospholipid species on the
plasma membrane. AP-2 first binds to PI(4,5)P2 via basic patches on its α and β2 subunits,
and once on the membrane makes a third PI(4,5)P2 interaction with its μ2 subunit
(Jackson et al., 2010). This triggers a large conformational change in AP-2 into its ‘open’
form, exposing binding pockets for its ligands - the endocytic sorting motifs YXXΦ
(where Φ is a bulky hydrophobic residue and X is any amino acid) and [ED]XXXL[LI]
(‘acidic dileucine’ motif). AP-2 is then able to bind its transmembrane cargo proteins
via these motifs, which stabilises the structure in its open conformation leaving it tightly
attached to the membrane via multiple phospholipid and cargo interaction sites. Other
alternative adaptors for endocytosis, for example epsins, also bind to PI(4,5)P2 at the
plasma membrane, while adaptors that localise to other organelles may have preferences
for different phospholipid species (Robinson, 2004). AP-1 has been reported to bind to
phosphatidylinositol-4-phosphate (PI4P) on TGN and endosomal membranes (Wang
et al., 2003b) and the recruitment of AP-3 to endosomes has also been suggested to
require the activity of the phosphatidylinositol 4 kinase PI4KIIα (Craige et al., 2008).
Vesicle cargo itself can also contribute to the initiation of vesicle formation. As discussed
above, the binding of AP-2 to its cargo via YXXΦ and acidic dileucine motifs stabilises its
association with the plasma membrane (Jackson et al., 2010). Although previous studies
had found overexpression of cargo proteins did not increase the number of clathrin-
coated pits (CCPs) at the plasma membrane, members of the Schmid lab found that
enforced clustering of biotinylated transferrin receptor (an endocytic cargo) increased
the rate of CCP maturation and induced the formation of new CCPs (Liu et al., 2010). Also,
at the TGN the overexpression of acidic dileucine motif-containing proteins increased
the membrane recruitment of the GGA family of adaptors (Hirst et al., 2007). These
studies demonstrate that cargo proteins are not merely passive passengers in vesicles,
but have an active role in vesicle coat recruitment and/or stabilisation.
Cargo binding occurs via interactions between adaptors and signals present in the cy-
tosolic tails of transmembrane cargo proteins (reviewed in Traub, 2009). These sorting
signals are diverse but can be grouped into three main classes: (i) short linear motifs;
(ii) post-translational modifications; (iii) folded domains. Short linear motifs, examples
of which include the YXXΦ, acidic dileucine and FXNPXY motifs mentioned above, are
utilised for most of the constitutive sorting events within the secretory and endocytic
systems. They consist of short linear arrays of amino acids (typically four to seven) in dis-
ordered regions of the protein. They are not exactly conserved sequences, but rather are
1.3 The stages of vesicular trafficking 19
degenerate motifs where two or three amino acids (often bulky hydrophobic residues)
are critical for binding to an adaptor. Crystal structures of adaptor subunits bound
to short linear motif-containing peptides have revealed the structural basis for cargo
recognition via the major sorting motifs (Traub, 2009). Reversible post-translational
modifications are employed for selective trafficking events that only occur under certain
conditions. One example is ubiquitination , which as mentioned above is a major sorting
signal for the lysosomal degradation pathway (Katzmann et al., 2001). Ubiquitination is
also used to signal for selective downregulation of receptors from the plasma membrane,
the classic example being that of the epidermal growth factor receptor (Traub, 2009).
Phosphorylation can also play a role in selective endocytosis; for example, hyperphos-
phorylation of G protein coupled receptors (GPCRs) recruits the alternative endocytic
adaptor β-arrestin, which can subsequently bind to the β2 appendage domain of AP-2
and so be incorporated into pre-exisiting CCPs (Scott et al., 2002). Finally, some pro-
teins lack known sorting signals and are instead recognised via features of their tertiary
structure. This is the case for SNAREs such as VAMP7 and VTI1B, which are recognised
via their folded domains by the adaptors HRB/AP-3 and epsinR, respectively (Miller
et al., 2007; Pryor et al., 2008). SNAREs are essential cargo proteins because they are
required for vesicle fusion with the acceptor membrane (see Section 1.3.7 below). This
different mode of cargo recognition for SNAREs may thus ensure them ‘reserved seating’
in the vesicle, as they do not need to compete for binding sites with cargo containing
conventional short linear sorting motifs (Robinson, 2015).
The recruitment of the structural components of the vesicle coat, e.g. clathrin for CCVs
and SEC13-SEC31 for COPII vesicles, generally occurs after assembly of the adaptor
layer (Gomez-Navarro & Miller, 2016; Paczkowski et al., 2015). However, this is not
the case for COPI vesicles where the B and F subcomplexes, which both contribute to
membrane binding and coat structure, are recruited to the membrane en bloc (Hara-
Kuge et al., 1994). Clathrin recruitment is best understood for AP-2 for which a series of
detailed structural studies have been performed by David Owen’s lab. Most recently, Kelly
and colleagues have defined an auto-inhibitory interaction which occurs between the
clathrin box within the flexible hinge of the β2 subunit and the AP-2 core, when AP-2 is
in its cytosolic ‘locked’ form (Kelly et al., 2014). However, the large-scale conformational
change induced by the binding of AP-2 to PI(4,5)P2 at the plasma membrane (Jackson
et al., 2010) releases the clathrin box from this auto-inhibitory interaction, thus allowing
clathrin recruitment to the membrane. Conservation between the region of β2 involved
in this auto-inhibitory interaction and β1 suggest a similar mechanism may regulate
clathrin recruitment to AP-1 vesicles (Kelly et al., 2014).
20 Introduction
To summarise, the initiation of vesicle formation involves multiple layers of regulation to
ensure that it occurs at the right place, the right time and with the right cargo. Once all
the coat components have assembled on the membrane, the process of vesicle budding
can begin.
1.3.2 Vesicle budding
Vesicle budding requires membrane bending, which is an energetically costly process.
As mentioned previously, membrane-anchored clathrin alone is capable of generating
spherical buds from a liposome membrane in vitro (Dannhauser & Ungewickell, 2012).
The membrane buds progress to the narrow neck stage whereupon dynamin is recruited
for fission. This suggests that the polymerisation of the spherical clathrin scaffold, linked
to the membrane via adaptors, is sufficient to drive vesicle budding. However, in vivo
additional factors can increase the energy barrier to membrane bending, including spon-
taneous curvature of the membrane due to differences in lipid composition between
the two leaflets and the presence of proteins embedded in the membrane (reviewed
in Stachowiak et al., 2013). Thus, the mechanisms contributing to membrane bending
during the formation of vesicular and tubular transport intermediates in vivo are still an
area of active debate (reviewed in Kirchhausen, 2012). Possible mechanisms include: (i)
assembly of a scaffold linked to the lipid bilayer, e.g. clathrin or COPI/II coats; (ii) asym-
metrical insertion of lipids, e.g. cholesterol, into the lipid bilayer; (iii) the attachment of
an intrinsically-curved protein to the membrane; (iv) the insertion of a protein ‘wedge’
into the membrane; (v) membrane crowding.
Proteins that may be able to drive membrane curvature during vesicle formation include
accessory proteins with BAR domains, such as FCHO1 and FCHO2 (F-BAR domain only
protein 1 and 2) at CCPs (Henne et al., 2010). The BAR domain forms a curved ‘banana
shaped’ dimer which interacts with negatively charged lipid head groups via positively
charged residues on its concave surface (Peter et al., 2004). Henne and colleagues de-
tected FCHO1/2 at sites of CCP formation prior to the arrival of clathrin and found that
in the absence of FCHO1/2 CCVs failed to form, concluding that FCHO1/2 act as CCP
nucleators (Henne et al., 2010). However, a single molecule live-cell imaging study from
the Kirchhausen lab found that FCHO1/2 arrive after the initiation of coat assembly
(Cocucci et al., 2012). This suggests that they may act to stabilise the growing clathrin
lattice, rather than to provide initial membrane curvature for CCP formation. Similarly,
proteins that insert the hydrophobic part of an amphipathic helix into the outer leaflet of
the membrane, e.g. epsins with their ENTH domain, have been proposed to induce mem-
1.3 The stages of vesicular trafficking 21
brane curvature (Ford et al., 2002). The membrane insertion of the GTP-bound forms
of the small GTPases ARF1 and SAR1 (responsible for recruitment of COPI and COPII
coat components, respectively) has also been suggested to contribute to membrane
curvature (McMahon & Mills, 2004). An alternative hypothesis is that protein-protein
interactions can drive membrane crowding, which can induce membrane bending in
the absence of any membrane-curvature-inducing domains (Stachowiak et al., 2012). It
is possible that several, or even all, of these mechanisms contribute to vesicle budding
in vivo.
1.3.3 Scission
The completion of vesicle formation requires scission from the donor membrane. The
release of CCVs from the plasma membrane requires the activity of the small GTPase
dynamin (Damke et al., 1994; reviewed in Antonny et al., 2016). Dynamin can self-
assemble into a helical oligomer, which acts to constrict the neck of the coated pit
(Hinshaw & Schmid, 1995). Its GTPase activity is required for vesicle scission because
dynamin mutants with defects in GTP binding or hydrolysis have a dominant-negative
effect on endocytosis (Van der Bliek et al., 1993). Exactly how dynamin drives membrane
fission is still an area of active debate, but the consensus is that once dynamin has
polymerised around the neck of a vesicle bud, GTP hydrolysis results in a conformational
change which triggers either constriction or stretching of the membrane to promote
membrane fission (Antonny et al., 2016). In vitro, dynamin alone can constrict and
cut lipid tubules (Roux et al., 2006). However, in vivo it is likely that the process of
vesicle scission is assisted by other factors. These may include: (i) membrane tension
generated by actin polymerisation or the activity of myosin motors; (ii) tension produced
by differences in lipid distribution between the two sides of the membrane constriction;
(iii) degradation of PI(4,5)P2; (iv) destabilisation of the lipid bilayer by the activity of
lipid-binding proteins (Ferguson & De Camilli, 2012). Dynamin contains a proline-rich
domain with which it binds various interaction partners, including the BAR domain-
containing proteins SNX9, endophilin-A2 and amphiphysin (Ferguson et al., 2009).
These proteins helps recruit dynamin and facilitate its polymerisation, but it also possible
they contribute to membrane fission directly via an effect on membrane curvature
(Antonny et al., 2016).
The small GTPases that recruit COPI and COPII coat components, ARF1 and SAR1, are
also ascribed important roles in vesicle scission, although the precise means by which
they do so are poorly defined (Beck et al., 2011; Lee et al., 2005). A study by Frank Adolf
22 Introduction
from Felix Wieland’s lab demonstrated that the scission of COPI and COPII vesicles occurs
independently to GTP hydolysis, in contrast to the activity of dynamin (Adolf et al., 2013).
The authors proposed that the shallow insertion of the amphipathic helices of ARF1
and SAR1 could induce membrane separation during COPI and COPII vesicle budding.
In contrast to the role of dynamin in the scission of CCVs at the plasma membrane,
AP-1 and AP-3 vesicle formation from intracellular membranes has been shown to be
dynamin-independent (Kural et al., 2012). The molecular details of vesicle scission for
these different kinds of coated vesicle await discovery, but it is possible that ARF1 plays
a role here too.
1.3.4 Uncoating
In order to fuse with an acceptor membrane a vesicle must first lose its coat. When
Roth and Porter first described coated vesicles and pits they observed ‘naked vesicles’
lacking the ‘bristle coat’ just inside the plasma membrane, suggesting CCVs lose their
clathrin coat soon after vesicle budding (Roth & Porter, 1964). This has since been
confirmed by live cell imaging studies (Kirchhausen et al., 2014). The uncoating of
CCVs is an active process that requires uncoating machinery including auxilin and
the cytosolic heat-shock cognate protein Hsc70 (Robinson, 2015). Auxilin is recruited
immediately after scission (Taylor et al., 2011). It then recruits the ‘uncoating agent’
Hsc70 via an interaction with its J-domain and Hsc70 clamps onto the C-terminus of
clathrin (Xing et al., 2010). The binding of Hsc70 to clathrin distorts the clathrin lattice
and so leads to clathrin disassembly. The timing of vesicle uncoating is proposed to
involve the recognition of specific lipids on the vesicle membrane by auxilin, via its
PTEN-like domain (Massol et al., 2006). Before vesicle scission any lipids modified by a
coat-associated lipid-modifying enzyme would rapidly diffuse away from the vesicle bud.
After vesicle scission modified lipids would remain on the vesicle membrane and hence
mediate auxilin recruitment. A candidate for a lipid-modifying enzyme in this scheme is
synaptojanin, which is a polyphosphoinositide phosphatase that is recruited around the
time of vesicle scission (Kirchhausen et al., 2014). Synaptojanin defects in Caenorhabditis
elegans cause neuronal phenotypes and an accumulation of CCVs, consistent with an
uncoating defect (Harris et al., 2000). Lipid modification by synaptojanin could have a
dual role in vesicle uncoating, firstly in the recruitment of auxilin and secondly in the
depletion of PI(4,5)P2, which will weaken the affinity of AP-2 for the vesicle membrane
(Kirchhausen et al., 2014).
1.3 The stages of vesicular trafficking 23
For COPI/II vesicles, uncoating is thought to be a simpler process that does not require
the recruitment of specific uncoating factors. Instead GTP hydrolysis is thought to lead
to the dissociation of the small GTPases ARF1 and SAR1 from the vesicle membrane,
triggering the disassembly of the other coat components (Kirchhausen, 2000).
1.3.5 Transport
Once released from the donor membrane, a vesicle must find its way to the correct
acceptor compartment. Although vesicles may travel short distances by diffusion, the
majority of inter-organelle transport (as well as the movement of organelles themselves)
is mediated by motor proteins that travel along cytoskeletal tracks (reviewed in Soldati &
Schliwa, 2006). In animal cells, long-range transport is the realm of kinesin and cytoplas-
mic dynein motors which travel along microtubules, while short-range transport occurs
on actin filaments mediated by myosin motors. In non-polarised cells, microtubules
are orientated with their fast-growing ‘plus ends’ towards the cell periphery and their
slow-growing ‘minus ends’ anchored at the microtubule-organising center (MTOC) in
the centre of the cell. Actin filaments, which are also polarised with plus (‘barbed’) ends
and minus (‘pointed’) ends, form a network close to the cell surface beneath the cell
cortex. Cytoskeletal motors have directionality, functioning either in plus-end-directed
transport (most kinesin and myosin motors) or minus-end-directed transport (cytoplas-
mic dynein, kinesin-14 and myosin VI motors). Mammals have a huge diversity of motor
proteins (over 100) which have class-specific cargo binding domains in order to mediate
specific transport tasks (Soldati & Schliwa, 2006).
Specific motors are thought to be recruited to transport vesicles via large protein com-
plexes that interact with vesicle cargo (reviewed in Akhmanova & Hammer, 2010). These
protein complexes include small GTPases, specific adaptors, scaffolds and regulatory
factors. For example, minus-end-directed myosin VI motors play an important role
in clathrin-mediated endocytosis (Buss & Kendrick-Jones, 2008). The recruitment of
myosin VI to CCVs is mediated by the cargo-selective adaptor DAB2, which binds LDL
receptors, AP-2 and clathrin, and by binding to PI(4,5)P2 (Morris et al., 2002; Spudich
et al., 2007). Upon binding to DAB2 myosin VI forms a membrane-associated dimer,
which functions as a processive motor (Yu et al., 2009). Myosin VI is suggested to be
multi-functional during clathrin-mediated endocytosis, firstly providing actin-generated
force during scission and secondly facilitating the movement of CCVs away from the
cell periphery (Buss & Kendrick-Jones, 2008). Another example is the association of the
plus-end-directed motor kinesin-1 with Trk receptor-containing vesicles for transport
24 Introduction
along the axon (Arimura et al., 2009). An adaptor called SLP1 binds to the cytoplasmic tail
of the Trk receptor and interacts with another adaptor, CRMP-2, and the small GTPase
RAB27 to mediate binding to kinesin-1. The contribution of different motors to different
trafficking pathways and the molecular details of motor recruitment and regulation are
active areas of research.
1.3.6 Tethering
When a vesicle reaches an acceptor membrane it will be captured by a process known
as ‘tethering’. The tethering factors and Rab GTPases that mediate this process provide
the first layer of specificity for vesicle fusion (reviewed in Hong & Lev, 2014). There
are two major classes of tethering factor: (i) homodimeric long coiled-coil proteins;
(ii) multisubunit tethering complexes (MTCs). Coiled-coil tethers have long flexible
structures that allow them to reach out more than 200 nm from the membrane to interact
with vesicles. Examples include the golgin family, which localise to the Golgi apparatus
(Gillingham & Munro, 2016), and the endosomally localised tethering factors EEA1 and
Rabaptin-5 (Gillingham & Munro, 2003). MTCs consist of three to ten subunits and inter-
act with vesicles much closer to the membrane (up to 30 nm). They include the CATCHR
(complexes associated with tethering containing helical rods) family and Class C Vps
complexes such as HOPS (homotypic fusion and protein sorting) and CORVET (class C
core vacuole/endosome tethering), which mediate tethering at endosomes/lysosomes
(Hong & Lev, 2014).
The recruitment of tethering factors themselves to membranes can occur via differ-
ent mechanisms. Some are anchored to the membrane via transmembrane domains,
but most are recruited dynamically via interactions with small GTPases and/or phos-
pholipids (Hong & Lev, 2014). Most tethering factors act as RAB or ARL effectors by
interacting specifically with RAB/ARL GTPases in their GTP-bound form. For example,
EEA1 is localised to early endosomes through a combination of binding to RAB5-GTP
and PI3P (Simonsen et al., 1998). HOPS and CORVET have shared core subunits but their
unique subunits specify their localisation by interacting with RAB7 on late endosomes
or RAB5 on early endosomes, respectively (Plemel et al., 2011). In addition to their role
in vesicle capture, tethering factors have also been implicated in regulating the assembly
of functional SNARE complexes via a network of interaction between tethers, SNAREs
and Sec1/Munc18 (SM) proteins (Hong & Lev, 2014). Thus, vesicle tethering is coupled
to the final stage of vesicle trafficking – fusion.
1.3 The stages of vesicular trafficking 25
1.3.7 Fusion
The final step in the life of a vesicle is its fusion with the membrane of the acceptor
compartment. The machinery that mediates membrane fusion within the secretory and
endocytic pathways is highly conserved throughout eukaryotic evolution. Members of
the SNARE protein family (of which there are at least 36 in humans) are the key compo-
nents of this machinery and since their discovery in the late 1980s a detailed mechanistic
model has emerged to explain how they mediate membrane fusion (reviewed in Jahn &
Scheller, 2006). SNARE proteins present on opposing membranes form a tight four-helix
bundle, bringing the membranes into close proximity and releasing energy that drives
membrane fusion. SNARE proteins have a characteristic SNARE motif consisting of 60–
70 amino acids arranged as heptad repeats and most have a C-terminal transmembrane
domain, although some SNAREs bind membranes via hydrophobic post-translational
modifications (Jahn & Scheller, 2006). SNAREs were originally classified as v-SNAREs
or t-SNAREs based on their presence on vesicles or target membranes (Söllner et al.,
1993b) but it was later realised that v-SNAREs could be found on target membranes
and t-SNAREs on vesicles, plus this could not apply to homotypic fusion events. Now
SNAREs are classified as R-SNAREs (arginine-containing) and Q-SNAREs (glutamine-
containing) based on the identity of a highly conserved residue (Fasshauer et al., 1998).
The formation of a four-helix bundle in a functional SNARE complex requires three
Q-SNAREs and one R-SNARE (described as a trans-SNARE complex).
Other proteins are important for the process of SNARE-mediated membrane fusion. SM
proteins are a small family of soluble proteins which are essential for fusion (Toonen &
Verhage, 2003). SM proteins form an arch-shaped ‘clasp’ structure and can bind to the
four-helix bundles of trans-SNARE complexes (Südhof & Rothman, 2009). This is sug-
gested to stimulate the fusogenic action of the SNARE helix bundle. After fusion has pro-
ceeded to completion, the zippered SNARE complex (now called a cis-SNARE complex) is
disassembled and recycled by the action of the ATPase NSF (N-ethylmaleimide–sensitive
factor), which binds to the SNARE complex via its adaptor SNAP (soluble NSF attachment
protein; Söllner et al., 1993a).
In vitro studies suggest that SNAREs can form complexes relatively promiscuously, but
only specific combinations, which correlate with compartmental localisation, generate a
fusion-competent complex (McNew et al., 2000). This led to the hypothesis that SNAREs
control the specificity of membrane fusion events. However, there are still a lot of open
questions regarding the exact contribution of SNAREs to the specificity of fusion, as
26 Introduction
opposed to roles for the earlier acting tethering factors and small GTPases (Jahn &
Scheller, 2006).
1.4 Adaptor protein complex 4
This section provides a specific introduction to adaptor protein complex 4 (AP-4), the
functional characterisation of which is the subject of this thesis.
1.4.1 The discovery and characterisation of AP-4
AP-4 was discovered in 1999 by the Robinson and Bonifacino labs in two separate at-
tempts to identify novel coat components related to the AP complexes (Dell’Angelica
et al., 1999a; Hirst et al., 1999). Both labs used BLAST searches of EST databases to
identify cDNAs encoding proteins related to subunits of AP-1, AP-2 and AP-3. This re-
vealed a novel heterotetrameric complex consisting of a ∼127 kDa subunit ε (related to
the γ/α/δ subunits of AP-1/2/3), a ∼83 kDa β4 subunit, a ∼50 kDa μ4 subunit (which had
already been identified in an earlier study; Wang & Kilimann, 1997) and a ∼17 kDa σ4
subunit. These proteins are encoded by the genes AP4E1, AP4B1, AP4M1 and AP4S1,
respectively. Immunoprecipitation and yeast two-hybrid experiments suggested that the
four AP-4 subunits formed a complex and structural similarity with their counterparts in
AP-1, AP-2 and AP-3 complexes supported a similar overall organisation for the complex
(Figure 1.8A). More recent studies have demonstrated the obligate nature of the AP-4
complex, as the loss of any single subunit renders the entire complex non-functional.
For example, in fibroblasts from patients with a homozygous loss-of-function mutation
in one of the four subunits, the remaining subunits no longer co-immunoprecipitate and
are present at reduced levels in the cell, suggesting in the absence of complex formation
they are subject to degradation (Hirst et al., 2013b). Like the other AP complexes, AP-4 is
present in the five eukaryotic supergroups, so it was already present by the time of the
last eukaryotic common ancestor (Field & Dacks, 2009; Hirst et al., 2014). However, AP-4
has been lost from several lineages, including yeast, worms and flies, and in these cases
all four AP-4 subunits have been concomitantly lost, supporting the obligate nature of
the complex (Hirst et al., 2013b; Figure 1.8B).
Expression studies at the mRNA and protein level support ubiquitous expression of AP-4
across human tissues (Dell’Angelica et al., 1999a; Hirst et al., 1999, 2013b). Expression
levels are low in all tissues and cells examined. Copy number estimations from quanti-
tative mass spectrometric analyses of whole cell lysates from HeLa cells have revealed
1.4 Adaptor protein complex 4 27
Fig. 1.8 Adaptor protein complex 4. (A) Schematic diagram of AP-4. Like other members of the
AP complex family, AP-4 is a heterotetramer consisting of two large subunits (ε and β4), a medium
subunit (μ4) and a small subunit (σ4). (B) AP-4, and likewise AP-5, has been lost from a number
of model organisms including Saccharomyces cerevisiae (Sc), Schizosaccharomyces pombe (Sp),
Caenorhabditis elegans (Ce) and Drosophila melanogaster (Dm). The other organisms shown
are Arabidopsis thaliana (At), Homo sapiens (Hs) and Mus musculus (Mm). Coulsen plots are
from Hirst et al. (2013b). (C) Copy number estimations of adaptor protein complexes in HeLa
cells from quantitative mass spectrometry data, from Hirst et al. (2013b). The small σ subunits
were not accurately quantified so were excluded from the analysis. Estimated copy numbers
for the remaining three subunits of each complex are shown along with an average (av) for the
three subunits combined (error bars show SD). AP-4 and AP-5 are >30 times less abundant than
AP-1, AP-2 and AP-3. An estimated number of vesicles or buds per cell is also shown for each AP
complex, based on the assumptions that 50% of AP complexes are membrane-associated and
that there are ∼40 AP complexes in a vesicle or bud.
28 Introduction
there to be around 10-20,000 copies of each AP-4 subunit per cell (Hirst et al., 2013b;
Itzhak et al., 2016; Figure 1.8C). This is around 20-30 times fewer than the number of
AP-1, AP-2 or AP-3 subunits, whereas AP-5 is expressed at a comparable level to AP-4.
Quantitative proteomic analysis of primary mouse neurons has shown a similarly low
level of expression for AP-4 (Itzhak et al., 2017). It is possible there is a specific cell type,
developmental stage, or condition in which AP-4 is upregulated, but evidence so far
supports a constitutive low level of expression.
Immunofluorescence studies revealed AP-4 to localise as delicate puncta in the TGN
region in Rat-1 fibroblasts and HeLa cells (Dell’Angelica et al., 1999a; Hirst et al., 1999).
Immuno-EM of cells stably expressing epitope tagged AP4B1 showed labelling of a
perinuclear tubulovesicular post-Golgi compartment, consistent with localisation at the
TGN (Hirst et al., 1999). The steady state localisation of AP complexes indicates their
donor compartment, so this data suggested a role for AP-4 in cargo sorting at the TGN
(Hirst et al., 2013b). As discussed in Section 1.3.1, the membrane recruitment of the
Golgi/TGN/endosome-associated coat complexes COPI, AP-1 and AP-3 is blocked by the
drug brefeldin A, a GEF inhibitor (Donaldson & Jackson, 2011). This occurs because the
membrane association of these complexes is mediated by binding to the small GTPase
ARF1 in its membrane linked GTP-bound form (Ren et al., 2013). In cells treated with
brefeldin A, AP-4 completely redistributes from the TGN to the cytoplasm, suggesting
that the membrane association of AP-4 is similarly dependent on the nucleotide status
of a small GTPase such as ARF1 (Dell’Angelica et al., 1999a; Hirst et al., 1999). A yeast
two-hybrid study by Boehm and colleagues detected GTP-dependent binding between
the AP-4 ε subunit and ARF1 and weaker GTP-independent binding between μ4 and
ARF1 (Boehm et al., 2001). This suggests that like its relatives, AP-4 is also recruited
to the membrane by ARF1-GTP. However, a more recent study proposed a role for an
alternative ARF-like small GTPase called ARL5B (Toh et al., 2017), so this requires further
investigation. Reversible phosphorylation of several serine residues in the AP-4 ε subunit
has also been implicated in the regulation of membrane recruitment of AP-4 (Paolini
et al., 2011). Phosphorylation resulted in reduced membrane association, but the effect
was not as dramatic as that of brefeldin A treatment, suggesting that binding of a small
GTPase is the major determinant for membrane localisation.
AP-4 vesicles do not appear to have a clathrin coat. AP-4 does not colocalise with clathrin
by immunofluorescence nor associate with clathrin-coated structures at the EM level,
and could not be detected in CCV fractions prepared from brain (Hirst et al., 1999). As
mentioned previously, the beta subunits of AP-1, AP-2 and AP-3 have a clathrin-binding
motif within the flexible hinge region known as the ‘clathrin box’ (LΦ[D/E]Φ[D/E];
1.4 Adaptor protein complex 4 29
Dell’Angelica et al., 1998). In contrast, none of the AP-4 subunits contain this con-
sensus motif (Dell’Angelica et al., 1999a; Hirst et al., 1999). Based on this evidence, AP-4
is expected to function independently of clathrin and to date no equivalent scaffolding
protein has been identified for AP-4. This raises the question of whether in the absence
of a scaffold, can AP-4 really function in the formation of vesicles? Evidence from the
Robinson lab suggests AP-4 does generate vesicles: AP-4 subunits are enriched in a
generic vesicle fraction and live cell imaging of a fluorescently tagged AP-4-associated
protein revealed highly mobile puncta that moved rapidly over long distances (Borner
et al., 2012). However, due to the very low expression level of AP-4, immunolocalisation
of endogenous AP-4 by EM or biochemical isolation of AP-4-derived structures has not
been possible.
1.4.2 TEPSIN - an AP-4 ear-binding accessory protein
An array of ear-binding accessory proteins have been identified for AP-1 and AP-2 com-
plexes (Robinson, 2004). So far the only known AP-4 accessory protein is a cytosolic
protein called TEPSIN which was identified in a proteomic study led by Georg Borner
(Borner et al., 2012). Borner developed a multivariate proteomic approach to anal-
yse vesicle-enriched fractions prepared from HeLa cells (discussed in detail in Section
3.1.2). The vesicle-enriched fractions were prepared under different conditions from
cells that had been metabolically labelled using the SILAC (stable isotope labelling of
amino acids in cell culture) method (described in Section 3.1.1). The fractions were
analysed by SILAC-based quantitative mass spectrometry and then the datasets were
combined using principal component analysis. This resulted in the clustering of pro-
teins according to their relative enrichment in the different preparations. Proteins that
form a stable complex clustered together; TEPSIN clustered closely with the four AP-4
subunits, predicting association. In addition, AP-4 and TEPSIN had similar stoichiome-
tries in the vesicle fraction. TEPSIN was found to colocalise with AP-4 at the TGN and
analysis of fibroblasts from AP-4-deficient patients demonstrated that TEPSIN depends
on AP-4 for its recruitment to membranes. Similarly, although total levels of TEPSIN
were unchanged in AP-4-deficient cells, TEPSIN was no longer incorporated into the
vesicle-enriched fraction. Physical association of TEPSIN with AP-4 was demonstrated by
co-immunoprecipitation with the β4 subunit and this interaction was shown to require
functional AP-4 because TEPSIN was not co-immunoprecipitated from AP4M1-deficient
cells, whereas residual AP-4 ε was. This data was corroborated by the observation that
30 Introduction
Fig. 1.9 TEPSIN - the AP-4 accessory protein. Schematic diagram demonstrating the domain
organisation of TEPSIN, a cytosolic protein that binds to AP-4 appendage domains.
TEPSIN is co-ordinately lost from organisms that have lost AP-4, supporting a functional
relationship between the two.
The name TEPSIN stands for ‘tetra-epsin’ because it was named for its epsin N-terminal
homology (ENTH) domain (Rosenthal et al., 1999a), which is most closely related to
the ENTH domain in the AP-1 accessory protein epsinR (Figure 1.9; Borner et al., 2012).
TEPSIN also has another folded domain which is related to both ENTH and VHS (VPS27,
HRS and STAM) domains (Lohi & Lehto, 1998), so has been named ‘VHS/ENTH-like’
(Archuleta et al., 2017). As discussed in Section 1.2.2, epsins function as cargo adaptors
in CCVs and bind phosphoinositides via their ENTH domains (Hirst et al., 2003; Itoh
et al., 2001; Mills et al., 2003). The VHS domain is found in another family of monomeric
adaptors, the GGAs, and is used to bind cargo via acidic dileucine motifs (Owen et al.,
2004). TEPSIN lacks the binding sites for AP-1, AP-2 and clathrin that are present in other
epsins but it is pulled down with the β4 ‘ear’ C-terminal appendage domain (Borner et al.,
2012). This suggests that TEPSIN could have equivalent functions to the epsins in CCVs,
acting as a cargo adaptor for AP-4 vesicles. However, there are some differences between
TEPSIN and the other epsins which suggest it may have a slightly different role. For
example, the ENTH domain in epsins is the primary determinant of their recruitment
to membranes, whereas TEPSIN requires AP-4 for membrane recruitment. To date, the
functional role of TEPSIN in AP-4 vesicles remains to be identified.
1.4 Adaptor protein complex 4 31
1.4.3 AP-4 deficiency causes a severe neurological disorder
A complete loss of AP-1 or AP-2 function is embryonic lethal in mammals (Robinson,
2015). In contrast, humans are viable with loss-of-function mutations in AP-3, AP-4 or
AP-5, although all cause severe disorders: Hermansky-Pudlack syndrome, a disorder
characterised by albinism, bleeding, and pulmonary fibrosis, for AP-3 (Dell’Angelica
et al., 1999b); hereditary spastic paraplegia (HSP) for AP-4 and AP-5 (Hirst et al., 2013b).
The first description of patients with a mutation in an AP-4 complex subunit was in
2009 when Verkerk and colleagues described five siblings with an autosomal recessive
neurological disorder which they likened to cerebral palsy (Verkerk et al., 2009). The dis-
order was characterised by neonatal hypotonia, progressive spastic tetraplegia, delayed
psychomotor development, an inability to walk unaided, severe intellectual disability,
absent or very poor speech development, as well as pseudobulbar signs including stereo-
typic laughter. Onset occurred during early infancy and the disease course was slowly
progressive over 20 years (the oldest patient was 24 at the time of publication). Brain
magnetic resonance imaging (MRI) of three patients and postmortem brain pathology of
one patient revealed a reduction of cerebral white matter and atrophy of the cerebellum.
All five patients were found to have a homozygous mutation at a splice donor site in
intron 14 of AP4M1: c.1137+1G>T (GenBank NM_004722). This was shown to cause
skipping of exon 14 which would introduce a frameshift from residue 342 of the AP4M1
protein onwards (full length 453 amino acids), leading to premature termination of
the protein product after 64 residues of missense (p.Ser342ArgfsTer65). Expression of a
reduced molecular weight protein product of the expected size (44.5 kDa; full-length
protein is 50 kDa) was confirmed by transient expression of V5-tagged wild-type or mu-
tant AP4M1 in HEK293 cells. The C-terminal truncation of the mutant AP4M1 product
occurs within the Mu homology domain (residues 173-453; pfam: O00189) which is
the part of the subunit involved with cargo binding (Owen & Evans, 1998). Although
the truncated protein product is expressed it assembles inefficiently, or unstably, into a
complex with the other AP-4 subunits, as demonstrated by reduced whole cell levels and
co-immunoprecipitation of other AP-4 subunits in fibroblasts from the AP4M1-deficient
patients (Borner et al., 2012). In these cells, AP-4 also fails to localise to the TGN and
TEPSIN is not recruited, demonstrating that the mutation causes loss of AP-4 complex
function.
Since 2009 homozygous loss-of-function mutations in AP4B1, AP4E1 and AP4S1 have
also been identified as causative of similar autosomal recessive neurological disorders
(Abou Jamra et al., 2011; Hirst et al., 2013b; Moreno-De-Luca et al., 2011). This has
32 Introduction
led to definition of a clinically recognisable ‘AP-4 deficiency syndrome’ defined by hy-
potonia progressing to spastic paraplegia, severe intellectual disability and delayed
motor and speech development, with an early age of onset (Abou Jamra et al., 2011;
Ebrahimi-Fakhari et al., 2018). Additional features that are present in many, but not
all patients, include epilepsy, microcephaly, short stature, and episodes of stereotypic
laugter (Ebrahimi-Fakhari et al., 2018). Brain imaging studies have revealed a high
prevalence of brain abnormalities including thinning of the corpus callosum, delayed
myelination or white matter loss and ventriculomegaly (Blumkin et al., 2011; Ebrahimi-
Fakhari et al., 2018; Tüysüz et al., 2014; Verkerk et al., 2009). Most reported mutations are
frameshift or nonsense mutations that introduce premature termination codons, which
are likely to result in transcripts that will be subject to nonsense-mediated decay (Nagy
& Maquat, 1998) or will result in severely truncated non-functional protein products
(Ebrahimi-Fakhari et al., 2018; Hirst et al., 2013b). In all cases where the effect of the
mutation has been investigated in cells, mutations have been shown to result in loss
of AP-4 complex function as demonstrated by all or some of the following criteria: (i)
reduced protein level of the mutant subunit and/or the other AP-4 subunits; (ii) reduced
or no ability to co-immunoprecipitate the AP-4 complex; (iii) lack of TGN localisation
of AP-4; (iv) lack of interaction with TEPSIN (Borner et al., 2012; Hardies et al., 2015;
Hirst et al., 2013b; Kong et al., 2013). Thus, all published evidence strongly suggests the
molecular mechanism of neurological disease caused by AP-4 complex mutations is
loss of function. This supports an important role for AP-4 in neurodevelopment and
homeostasis. However, the aetiology of AP-4 deficiency syndrome is currently unknown.
1.4.4 The search for AP-4 cargo proteins
The very low expression level of AP-4 has made the biochemical characterisation of
AP-4 vesicles challenging and its absence from commonly used model organisms, such
as yeast, worms and flies, has limited genetic approaches to study the pathway. Early
studies focused on the μ4 subunit because the μ subunits of other AP complexes are
implicated in cargo binding. In yeast two-hybrid studies μ4 was able to weakly bind to
tyrosine-based sorting motifs (YXXΦ), including those of the lysosomal proteins CD63,
LAMP1 and LAMP2, but with a lower affinity than the μ subunits of AP-2 or AP-3 (Aguilar
et al., 2001; Hirst et al., 1999; Stephens & Banting, 1998). Aguilar and colleagues also
demonstrated YXXΦ-dependent binding between μ4 and LAMP2 in an in vitro binding
assay (Aguilar et al., 2001). However, knockdown of AP-4 did not alter the subcellular
1.4 Adaptor protein complex 4 33
distribution of these proteins in HeLa cells, so the in vivo relevance of these interactions
is unclear (Janvier & Bonifacino, 2005).
A study conducted by Simmen and colleagues looked for a role for AP-4 in basolateral
sorting in polarised epithelial cells (Simmen et al., 2002). Surface plasmon resonance
(SPR) in vitro binding experiments were used to test for binding of AP-4 to peptides
containing different basolateral-sorting motifs found in furin, the LDL receptor, cation-
dependent mannose-6-phosphate receptor (M6PR) and transferrin receptor. Binding
was detected for all, with the highest affinity for the LDL receptor sorting motif which is
tyrosine-dependent, but not a typical YXXΦ sorting motif. In polarised epithelial cells,
depletion of μ4 resulted in partial missorting of a furin tail chimaera and endogenous
LDL receptor and M6PR to the apical surface. This suggested a role for AP-4 in transport
from the TGN to the basolateral plasma membrane. In another study μ4 was a hit in a
yeast two-hybrid screen for proteins that interact with the C-terminus of the glutamate
receptor δ2, a protein that is expressed only in cerebellar Purkinje cells (Yap et al., 2003).
In vitro and in vivo interaction between exogenously expressed δ2 and HA-tagged μ4 was
dependent on a novel phenylalanine-rich motif in the δ2 C-terminus. This binding was
specific for the δ2 glutamate receptor because binding could not be detected between
μ4 and NMDA or AMPA glutamate receptors. In addition, overexpression of μ4 altered
the subcellular distribution of exogenously expressed δ2 in COS7 cells. Based on these
data and the previous study by Simmen et al. (2002), the authors hypothesised a role for
AP-4 in the somatodendritic sorting of δ2 glutamate receptors in Purkinje neurons.
In a subsequent study by a different group, an Ap4b1 knockout mouse was generated and
the authors looked for missorting of glutamate receptors in Purkinje and hippocampal
neurons (Matsuda et al., 2008). δ2 and AMPA glutamate receptors were mislocalised from
their normal somatodendritic distribution to axon terminals in Ap4b1-deficient neurons.
LDL receptor was similarly mislocalised in axons, whereas NMDA and metabotropic
glutamate receptors were unaffected, supporting a selective role for AP-4 in polarised
trafficking in neurons. In keeping with the study by Yap et al. (2003), which did not
detect direct binding between μ4 and AMPA glutamate receptors, interaction between
μ4 and AMPA receptors was found to be indirect via both binding to the regulatory
transmembrane TARP proteins (Matsuda et al., 2008). This was mediated by another un-
conventional sorting motif containing phenylalanine and tyrosine residues (YRYRF). In-
terestingly, the axon terminals containing AMPA receptors in the AP-4-deficient neurons
were abnormally swollen and EM analysis revealed an accumulation of autophagosome-
like structures. The axon terminals were shown to be immunopositive for the autophagy
marker LC3 and elevated levels of the lipidated, autophagy-competent form of LC3 were
34 Introduction
detected in brain lysates from the Ap4b1 knockout mice. However, the link between AP-4
deficiency and the dysregulation of autophagy was not investigated. The findings of this
study are interesting with regard to the neurological problems suffered by AP-4-deficient
patients as they support a role for AP-4 in protein sorting in neurons. Notwithstanding,
there were some surprising elements to the findings. The mice were reported to be
generally healthy with no overt brain abnormalities or signs of ataxia, in contrast to the
severe symptoms in patients. The only significant phenotype was a poor performance
on a rotorod. This is different from a knockout mouse model of stargazin, a TARP protein,
which also results in AMPA receptor missorting (Jackson & Nicoll, 2011). The stargazin
mouse does suffer from ataxia, and it is unclear why the missorting of TARPs and AMPA
receptors in the Ap4b1 knockout mice does not cause the same phenotype.
A study from the Bonifacino lab garnered particular interest when it implicated AP-4
in the sorting of the amyloid precursor protein (APP), which plays a key role in the
neurodegenerative disorder Alzheimer’s disease (Burgos et al., 2010). The μ4 subunit of
AP-4 was found to specifically interact with the cytosolic tail of APP in a yeast two-hybrid
analysis, whereas the μ subunits of AP-1, AP-2 and AP-3 did not bind APP. The binding
motif in APP was narrowed down to a non-canonical YXXΦ motif, YKFFE. Burgos and
colleagues went on to crystallise the C-terminal domain of μ4 in complex with a peptide
containing this motif and found it to bind μ4 at a completely different binding site from
that used to bind canonical YXXΦ motifs. The effect of AP-4 depletion on the sorting of
exogenously expressed APP was tested in HeLa and H4 neuroglioma cells and found to
result in an accumulation of APP at the TGN. Mutation of the YKFFE sorting motif in
APP also resulted in its mislocalisation to the TGN. The missorting of APP to the TGN
correlated with an increase in the amyloidogenic processing of APP, suggesting that
AP-4-mediated sorting of APP from the TGN to endosomes could have a protective role
in Alzheimer’s disease by reducing the level of beta amyloid production. In a follow up
study, overexpression of μ4 with mutations in the non-canonical tyrosine motif-binding
site had a dominant negative effect, blocking APP transport out of the TGN (Ross et al.,
2014). An independent study from the Gleeson lab also confirmed the effect of AP-4
depletion on the localisation of stably overexpressed APP in HeLa cells (Toh et al., 2017).
However, the physiological relevance of these findings for endogenous APP sorting in
neurons has not been determined.
To summarise, a combination of in vitro binding experiments and candidate-based
localisation studies have suggested roles for AP-4 in the sorting of a diverse array of
cargo proteins, including APP, LDL receptor, AMPA receptors and δ2 glutamate receptors.
The binding of AP-4 to these cargo proteins is reported to be mediated by a diverse
1.5 Project Aims 35
set of non-canonical sorting signals, with no commonality between them, so there is
currently no consensus for an ‘AP-4 sorting motif’. The destination of the AP-4 trafficking
pathway likewise remains controversial, as the APP studies suggest transport to early
endosomes (Burgos et al., 2010; Toh et al., 2017) while the studies in polarised epithelial
cells suggest transport to the basolateral membrane (Simmen et al., 2002). Several of the
functional studies on AP-4 relied on exogenously expressed proteins (e.g. APP), while
the potential endogenous cargoes (e.g. δ2 glutamate receptor) have cell-type limited
expression, unlike AP-4, which is ubiquitously expressed in human tissues (Dell’Angelica
et al., 1999a; Hirst et al., 1999) and in different cell types within the brain (Yap et al.,
2003).
1.5 Project Aims
AP-4 is proposed to function in protein sorting at the TGN, so AP-4 deficiency can be
thought of as a disease of missorting. Therefore, in order to understand the pathomech-
anisms that lead to the severe neurological problems suffered by AP-4-deficient patients,
it is important to identify the endogenous cargoes of the AP-4 trafficking pathway. As
discussed, there is currently no consensus as to which proteins are genuine cargoes of
the AP-4 pathway, and hence no consensus as to its function. Similarly, AP-4 vesicle ma-
chinery is largely uncharacterised; currently the only identified AP-4 accessory protein
is a cytosolic protein of unknown function called TEPSIN. The ubiquitous expression of
AP-4 in human tissues and its conservation throughout eukaryotic evolution support
a ubiquitous and important role for AP-4 in protein sorting. The primary aim of this
study was to use unbiased global proteomic approaches to define the composition of
AP-4 vesicles and to identify physiological cargo proteins of the AP-4 pathway. The
approaches applied were highly sensitive and assayed the localisation of endogenous
proteins, rather than relying on the detection of physical interactions between proteins.
The proteomic approaches and results of these analyses are described in Chapter 3.
Following on from this, the secondary aim was to validate and characterise candidate
AP-4 vesicle cargoes and machinery that were identified in our proteomic screens. This
involved a combination of cell biology and biochemical approaches, work which is
presented in Chapter 4.

Chapter 2
Materials and Methods
2.1 Reagents
All chemicals and reagents were purchased from Sigma-Aldrich unless otherwise stated.
All kits were used according to the manufacturer’s protocol.
2.1.1 Antibodies
The primary antibodies used for Western blotting, immunofluorescence and immuno-
precipitation in this study are listed in Table 2.1.
For Western blotting, horseradish peroxidase (HRP)-conjugated anti-mouse IgG and anti-
rabbit IgG secondary antibodies were used at a dilution of 1:10,000 and HRP-conjugated
anti-chicken IgY secondary antibody was used at a dilution of 1:2000 (all purchased
from Sigma-Aldrich). For immunofluorescence, the following fluorescently labelled
secondary antibodies were purchased from Invitrogen and used at a dilution of 1:500:
Alexa488-labelled goat anti-chicken IgY (A11039), Alexa488-labelled donkey anti-mouse
IgG (A21202), Alexa555-labelled donkey anti-mouse IgG (A31570), Alexa568-labelled
goat anti-mouse IgG (A11031), Alexa594-labelled donkey anti-mouse IgG (A21203),
Alexa488-labelled donkey anti-rabbit IgG (A21206), Alexa555-labelled goat anti-rabbit
IgG (A21429), Alexa594-labelled donkey anti-rabbit IgG (A21207), Alexa647-labelled
donkey anti-rabbit IgG (A31573), Alexa488-labelled donkey anti-sheep IgG (A11015),
Alexa594-labelled donkey anti-sheep IgG (A11016), and Alexa680-labelled donkey anti-
sheep IgG (A21102). Biotinylated proteins were detected with HRP-conjugated strep-
38 Materials and Methods
Table 2.1 Primary antibodies used in this study. WB: Western blotting; IF: immunofluorescence;
IP: immunoprecipitation.
Protein Application
(dilution)
Host species Vendor (ID)
Actin WB (1:3000) Rabbit Sigma-Aldrich
(A2066)
α-tubulin WB (1:10,000) Mouse Sigma-Aldrich
(DM1A, T9026)
AP4B1 WB (1:400); IP
(1:200)
Rabbit In house (Simone;
Hirst et al. (1999))
AP4E1 WB (1:1000) Rabbit In house (Elsie;
Hirst et al. (1999))
AP4E1 IF (1:100) Mouse BD Transduction
Labs (612019)
AP4E1 IP (1:200) Rabbit In house (Enid;
Hirst et al. (1999)
AP4M1 WB (1:400) Rabbit In house (Lupin;
Hirst et al. (1999)
AP4S1 WB (1:250) Rabbit In house (Shona;
Hirst et al. (1999)
ATG9A WB (1:1000); IF
(1:100)
Rabbit Abcam (ab108338)
BirA WB (1:1000) Chicken Abcam (ab14002)
CIMPR IF (1:200) Mouse Abcam (ab2733)
Clathrin HC WB (1:10,000) Rabbit In house (Simpson
et al. (1996))
EEA1 IF (1:500) Mouse BD Transduction
Labs (610457)
GFP IF (1:500) Rabbit Gift from M.
Seaman (Uni of
Cam)
GFP WB (1:1000) Rabbit Abcam (ab6556)
GFP IF (1:500) Chicken Abcam (ab13970)
HA WB (1:2000); IF
(1:2000)
Mouse Covance (16B12)
LAMP1 IF (1:100) Mouse Santa Cruz
Biotechnology
(H4A3)
LC3B WB (1:2000) Rabbit Sigma-Aldrich
(L7543)
LC3B IF (1:400) Mouse MBL international
(M152-3)
Myc IF (1:250) Mouse Merck Millipore
(clone 4A6; 05-724)
TEPSIN WB (1:1000); IF
(1:250)
Rabbit In house (Borner
et al. (2012))
TGN46 IF (1:200) Sheep Bio-Rad (AHP500)
2.1 Reagents 39
tavidin for Western blotting, diluted 1:5000, and streptavidin coupled to Alexa568 for
immunofluorescence, diluted 1:400 (both Invitrogen).
2.1.2 DNA constructs
The DNA constructs created and used in the course of this project are listed in Table
A.1 in Appendix A and details of the cloning methods used in their creation are detailed
below. All enzymes were purchased from New England BioLabs (NEB). Polymerase
chain reaction (PCR) was performed with Phusion High-Fidelity DNA Polymerase and
Phusion HF buffer (NEB), unless otherwise stated. Unless otherwise noted, constructs
were made using Gibson Assembly (Gibson et al., 2009), as described below. Standard
ligations, when performed, were with T4 DNA ligase, overnight at 16 °C or for 1 hour
at room temperature. Transformations were with two microlitres of Gibson Assembly
or ligation reaction using α-Select Silver Competent Cells (Bioline), unless otherwise
noted. Plasmids were purified with a QIAprep Spin Mini or Midi Kit (Qiagen) as necessary.
All cloning was verified by diagnostic digest with appropriate restriction enzymes and
Sanger DNA sequencing (Beckman Coulter Genomics). PCR primers used for cloning
are listed in Appendix B.
Gibson Assembly
Vectors were linearised with a single restriction enzyme and DNA insert fragments were
generated by PCR to include overlapping ends with the linearised vector and/or other
DNA insert fragments as necessary. Linearised vectors and PCR products were separated
by electrophoresis in 1 % w/v agarose and required digestion products were excised and
extracted using a QIAquick Gel Extraction Kit (Qiagen). For the Gibson Assembly itself
a 20 μl reaction containing the Gibson Assembly Master Mix and the linearised vector
and insert fragment(s), typically with a molar ratio of 1:3, was incubated at 50 °C for one
hour.
BioID constructs
Myc-BirA* cDNA was amplified from pcDNA3.1_mycBioID (a gift from Kyle Roux; Ad-
dgene plasmid 35700; Roux et al., 2012). A myc-BirA* tagging construct was made to
allow the C-terminal tagging of any protein with a flexible glycine/serine linker followed
by a myc tag and BirA*. Myc-BirA*, preceded by a G/S linker (10 amino acids) and an
upstream BglII site, was inserted into the HpaI site of a modified pLXIN plasmid (hereon
40 Materials and Methods
called pLXINmod; a gift from Andrew Peden, University of Sheffield). The BglII site was
used to linearise the myc-BirA* tagging construct and cDNAs for AP4B1, AP4M1, AP4S1,
SERINC1 and SERINC3 were added by Gibson Assembly. AP4B1 cDNA was amplified from
a full length IMAGE clone (2906087), and AP4M1 and AP4S1 cDNAs from sequence veri-
fied EST clones. SERINC1 cDNA was amplified from pGEM-T_SERINC1 (Sino Biological)
and SERINC3 from pIRESNeo2_SERINC3-HA-mCherry (a custom synthetic construct
based on the SERINC3 clone AAD22448.1, with several silent nucleotide substitutions;
Genecust). For the AP4E1 BioID construct, myc-BirA* was inserted into the flexible
hinge region of AP4E1, between residues 730 and 731, in the HpaI site of pLXINmod.
AP4E1 cDNA was amplified as two separate fragments from an AP4E1_FKBP knock-
sideways construct (generated by Georg Borner), including a a short linker sequence
(GALVNGGPEPAKNLYT) preceding myc-BirA* in the final construct. The control BioID
construct, pEGFP-myc-BirA*, was made by cloning myc-BirA* cDNA into a pEGFP-N2
vector (Clontech) using BsrGI and XbaI restriction sites, by standard molecular cloning
techniques.
AP4B1 rescue constructs
To generate the constructs for the AP4B1 rescue cell lines, full length (residues 1–739)
and earless (1–612) AP4B1 were amplified from a full length IMAGE clone (2906087)
and cloned into pLXINmod using SalI and NotI restriction sites, by standard molecular
cloning techniques. Full length pLXIN_AP4B1 was modified by site-directed mutage-
nesis by Meredith Frazier (University of Vanderbilt) to give pLXIN_AP4B1[Y682V] and
pLXIN_AP4B1[I669A/A670S].
Wild-type and mutant TEPSIN-GFP constructs
For the TEPSIN-GFP [L470S/F471S] mutant, the L470S and F471S point mutations were
introduced into a TEPSIN-GFP plasmid reported previously (Borner et al., 2012) by site-
directed mutagenesis by Meredith Frazier. Wild-type and mutant TEPSIN-GFP were
subsequently amplified from the wild-type and mutant TEPSIN-GFP plasmids, and were
inserted into the HpaI site of pLXINmod. TEPSIN is GC rich and therefore the PCR
reaction required some optimisation. Phusion GC buffer was used and 10 % dimethyl
sulfoxide (DMSO) was added to the reaction mix (annealing temperature 64 °C).
2.1 Reagents 41
SERINC3 and RUSC2 expression constructs
For the expression of SERINC3 with an extracellular HA tag (pLXINmod_SERINC3_HA
[extloop]), SERINC3 cDNA was amplified as two separate fragments from pIRESNeo2_
SERINC3-HA-mCherry and an HA tag was inserted between residues 311 and 312 within
an extracellular loop of SERINC3, in the HpaI site of pLXINmod. For the constructs for
stable overexpression of GFP-tagged RUSC2, RUSC2 cDNA was amplified from pCMV-
SPORT6_RUSC2 (MHS6278-202800194, Thermo Fisher Scientific) and GFP cDNA was
amplified from pEGFP-N2 (Clontech). As RUSC2 is large (4551 bp coding DNA) it was
amplified as two separate fragments which were inserted along with GFP into the AgeI site
of pQCXIH for pQCXIH_GFP-RUSC2 and the NotI site of pQCXIH for pQCXIH_RUSC2-
GFP. For expression of HA-tagged RUSC2, pQCXIH_HA-RUSC2 was made by cutting out
GFP from pQCXIH_GFP-RUSC2 with NotI and AgeI restriction sites and replacing it with
a triple HA tag.
SERINC antigen GST-fusion constructs
The TMHMM2.0 server (Sonnhammer et al., 1998) was used to predict the membrane
topology of SERINC1 and SERINC3. Based on this we chose the largest predicted inside
(cytoplasmic) and outside (extracellular/luminal) non-transmembrane loops from each
protein to use as antigens:
Serinc1_IN: RTSNNSQVNKLTLTSDESTLIEDGGARSDGSLEDGDDVHRAVDNERDGV
Serinc1_OUT: NEPETNCNPSLLSIIGYNTTSTVPKEGQSV
Serinc3_IN: RTSTNSQVDKLTLSGSDSVILGDTTTSGASDEEDGQPRRAVDNEKEGVQYS
Serinc3_OUT: SNEPDRSCNPNLMSFITRITAPTLAPGNSTAVVPTPTPPSKSGSLLDSDN
The corresponding DNA sequences, with overlapping ends with pGEX-4T-1 (Amersham)
linearised with SmaI, were ordered as double-stranded ‘gBlocks’ from Integrated DNA
Technologies. This allowed each SERINC antigen peptide sequence to be inserted into
pGEX-4T-1 by Gibson Assembly, so as to N-terminally tag each antigen peptide with
GST.
42 Materials and Methods
2.2 Generation of polyclonal antibodies against SERINC1
and SERINC3
2.2.1 Production of antigens
The constructs described above were transformed into α-Select Silver Competent E. coli
and the cells were grown to 500 mL cultures in log phase when expression of the GST-
fusion proteins was induced by the addition of 0.1 mM IPTG to each culture. Cultures
were then incubated at 22 °C overnight. The following day the cells were pelleted (4,000
rpm, JLA-16.250 rotor, 10 minutes, 4 °C), washed in ice-cold PBS (137 mM NaCl, 2.7 mM
KCl, 10 mM Na2HPO4 and 1.76 mM KH2PO4, pH 7.4), pelleted again, resuspended in 5
mL PBS and frozen in 1.5 mL aliquots at −20 °C. To extract the proteins for purification,
aliquots of cells were thawed and sonicated (three times 10–15 seconds at 15 microns).
Triton TX-100 was added to a final concentration of 1% and 200 μM AEBSF protease
inhibitor was added. Lysis was carried out for 10 minutes on ice. Cell debris was pelleted
(16,000 g, 5 minutes, 4 °C) and supernatants (lysates) were transferred to new tubes.
GST-fusion proteins were affinity purified on glutathione-Sepharose (GE Healthcare; 100
μL packed Sepharose per 1 mL lysate; 30 minutes at room temperature). Glutathione-
Sepharose was then pelleted (8,600 g, 1 minute) and washed five times with ice-cold
PBS. Finally the GST-fused antigens were eluted in 10 mM reduced glutathione (in 50
mM Tris pH 8).
2.2.2 Production and purification of antibodies
Antibody production was carried out by Cambridge Research Biochemicals. Two rabbits
were immunised per antigen, each with 500 μg protein at 1 mg/mL plus 500 μL Freund’s
complete adjuvant. An immunisation boost (same protocol) was given on day 14 and
again on day 56. On day 66 sera were harvested and sent to us for affinity purification.
To couple the GST-fusion proteins to Sepharose, GST-fused antigens and GST were
dialysed into coupling buffer (0.1 M NaHCO3 pH 8.3/0.5 M NaCl) so at a concentration
of roughly 1 mg/ml and then coupled to CnBr-activated Sepharose (GE Healthcare; 0.2 g
per 1 mg protein). Protein was incubated with activated Sepharose for 2 hours at room
temperature. The coupled Sepharose was pelleted and the supernatant removed. The
Sepharose was washed in coupling buffer and incubated in 0.1 M Tris-HCl pH 8 for 2
hours at room temperature to deactivate the Sepharose. The Sepharose was then washed
2.3 Cell culture 43
with three cycles of alternating pH (0.1 M C2H3NaO2 pH 4/0.5 M NaCl or 0.1 M Tris-HCl
pH8/0.5 M NaCl). Finally the coupled Sepharose was washed with 200 mM glycine-HCl
pH 2.3 and twice in PBS plus 0.02% sodium azide, in which it was stored at 4 °C until use.
Roughly 10 mL of each serum were depleted of GST-reactive antibodies by incubation
with 1 mL packed GST-Sepharose (overnight at 4 °C). Sera were then incubated with
Sepharose coupled to the appropriate antigen (at least 3 hours at room temperature or
overnight at 4 °C). Sepharose was washed five times with ice-cold PBS, packed into a
column and washed again with PBS. Antibodies bound to the Sepharose were eluted in
eight to ten 500 μL fractions with 200 mM glycine pH 2.3. The pH of each fraction was
restored by the addition of 2 M Tris-HCl pH 8. The protein concentration of each fraction
was estimated with a coomassie dot blot and fractions containing very little protein were
discarded (typically the first two elutions). The remaining fractions were combined,
BSA was added to a final concentration of 1 mg/mL and the purified antibodies were
then dialysed into PBS overnight at 4 °C. The following day dialysed antibodies were
incubated with 500 μL packed GST-Sepharose to further deplete them of GST-reactive
antibodies (1 hour at 4 °C). The GST-Sepharose was pelleted (1,700 g, 1 minute) and the
purified antibodies were transferred to a new tube and sodium azide was added to 0.02%.
Antibodies were stored in 500 µL aliquots at −80 °C for long term storage. Glycerol was
added to one aliquot of each antibody (to a final concentration of 50%) for storage at
−20 °C.
2.3 Cell culture
2.3.1 Cell lines
HeLa M cells and HEK 293ET cells were from ECACC and the human neuroblastoma
cell line SH-SY5Y (Biedler et al., 1978) was from Sigma-Aldrich. The HeLa cells stably
expressing BirA* were a gift from Folma Buss (University of Cambridge) and the HeLa
cells stably expressing EGFP were a gift from Matthew Seaman (University of Cambridge).
The HeLa cells stably expressing TEPSIN-GFP used for the conventional and sensitive
immunoprecipitations for mass spectrometry (Section 3.6.1) were previously generated
in our lab by Georg Borner (Borner et al., 2012).
HeLa M and HEK 293ET cells were maintained in RPMI 1640 (R0883, Sigma-Aldrich).
SH-SY5Y cells were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM) high
glucose (D6546, Sigma-Aldrich). All media was supplemented with 10% v/v foetal calf
44 Materials and Methods
serum (FCS), 4 mM L-glutamine, 100 U/mL penicillin and 100 μg/mL streptomycin and
all cells were cultured at 37 °C under 5% CO2. For antibiotic selection of HeLa cells,
500 μg/mL G418, 150 μg/mL hygromycin or 1 μg/mL puromycin were added to the
culture medium as appropriate. For antibiotic selection of SH-SY5Y cells, 150 μg/mL
hygromycin or 3 μg/mL puromycin were used. Where indicated cells were treated with
10 μg/mL nocodazole in cell culture medium for 2 hours at 37 °C. For starvation during
autophagy assays, cells were washed 3 times with Earle’s balanced salt solution (EBSS;
Sigma-Aldrich) and incubated in EBSS for the specified time. Where indicated cells were
treated with 100 nM Bafilomycin A1 in cell culture medium or EBSS for 2 hours. For AP-4
knocksideways (performed by Dr Georg Borner), cells were treated with rapamycin at
200 ng/mL final concentration (from a 1 mg/mL stock in ethanol) for the specified time.
2.3.2 Patient fibroblasts
The human fibroblasts from AP-4 deficient patients have been previously described and
reduced AP-4 complex formation has been demonstrated for all patient lines. Molecular
details of mutations and references for each are as follows: AP4B1*, GenBank NM_006594:
c.487_488insTAT, p.Glu163_Ser739delinsVal, (Abou Jamra et al., 2011; Borner et al., 2012,
a gift from Laurence Colleaux, Institut Imagine, and Annick Raas-Rothschild, Tel Aviv Uni-
versity); AP4E1*, GenBank NM_007347: c.3313C>T, p.Arg1105* (Kong et al., 2013; a gift
from Xiao-Fei Kong, Jean-Laurent Casanova, and Stephanie Boisson-Dupuis, The Rock-
efeller University); AP4M1*, GenBank NM_004722: c.1137+1G>T, p.Ser342ArgfsTer65
(Borner et al., 2012; Verkerk et al., 2009; a gift from Grazia Mancini, Erasmus Medical
Center); AP4S1*, GenBank NM_007077: c.[289C>T];[138+3_6delAAGT], p.[Arg97*];[?]
(Hardies et al., 2015; the second mutation, which is at a splice donor site, was orig-
inally inaccurately reported as c.138_140delAATG). The control fibroblasts (from a
healthy control donor) were a gift from Craig Blackstone (NIH) and the heterozygous
AP4E1WT/AP4E1* fibroblasts (a gift from X. Kong et al., as above) were from the unaffected
mother of the homozygous AP4E1* patient. Fibroblasts were maintained in DMEM high
glucose as described above.
2.3.3 SILAC metabolic labelling
For metabolic labelling (SILAC method – Ong et al., 2002) in most experiments, HeLa
cells were cultured in DMEM without Arginine, Glutamine, Lysine or Sodium Pyruvate
(A14431-01, Gibco), supplemented with 10% (v/v) dialysed FCS (A11-107, PAA), 1 mM
Sodium Pyruvate (58636, Sigma-Aldrich), 1 x GlutaMAX (35050-061, Gibco), and either
2.3 Cell culture 45
‘Heavy’ amino acids (42 mg/L 13C6, 15N4-L-Arginine HCl and 73 mg/L 13C6, 15N2-L-Lysine
HCl; 201604302 and 211604302, Silantes), or the equivalent ‘Light’ amino acids (Arginine
HCl [A6969] and Lysine HCl [L8662], Sigma-Aldrich). Cells were cultured for at least
seven days in these media before experiments were performed. In the vesicle fraction ex-
periments (Section 3.4), metabolic labelling was performed in SILAC RPMI 1640 medium
(89984, Thermo Fisher Scientific), supplemented with 10% (v/v) dialysed FCS (10,000
MW cut-off; Invitrogen), and either ‘Heavy’ amino acids (50 mg/L 13C6, 15N4-L-Arginine
HCl and 100 mg/L 13C6, 15N2-L-Lysine 2HCl; Cambridge Isotope Laboratories), or the
equivalent ‘Light’ amino acids.
2.3.4 Transient transfections
Unless otherwise stated, transient transfections were carried out in 6-well plates. The
quantities of reagents given in the main text are for 6-well plate transfections; quantities
for a 10 cm dish are included in brackets.
HeLa M cells were transiently transfected using a TransIT-HeLaMONSTER transfection
kit (Mirus Bio LLC). Approximately 16–24 hours before transfection, cells were plated in
2 mL (or 10 mL) antibiotic-free growth medium. Seeding was adjusted to achieve 70–
90% confluency at the time of transfection. For each well, 6 μL (or 24 μL) TransIT-HeLa
reagent were diluted in 200 μL (or 800 μL) OptiMEM I Reduced-Serum Medium (Thermo
Fisher Scientific) and the reaction incubated at room temperature for 10 minutes. Two
micrograms (or 4 μg) of DNA were added to this and mixed by gentle pipetting before
incubation at room temperature for 10 minutes. Two microlitres (or 8 μL) of MONSTER
reagent were added, mixed by gentle pipetting and the reaction incubated at room
temperature for a further 10 minutes. The DNA-liposome complexes were then added
drop-wise to the dish, which was gently rocked to distribute the complexes evenly. Cells
were incubated at 37 °C, 5% v/v CO2 overnight before the transfection mix was replaced
with fresh full growth medium.
2.3.5 Generation of stable cell lines
The majority of stable cell lines were created using retrovirus made in HEK 293ET cells.
HEK 293ET cells were seeded in 6-well plates so as to be at roughly 70–80% confluency
at the time of transfection. Transfection was carried out in 2 mL antibiotic-free medium,
using TransIT-293 Transfection Reagent (Mirus Bio LLC). For each well, 8 μL TransIT-293
Transfection Reagent were mixed with 200 μL OptiMEM and incubated for 10 minutes at
46 Materials and Methods
room temperature. pLXIN or pQCXIH plasmids were mixed with the packaging plasmids
pMD.GagPol and pMD.VSVG in a ratio of 10:7:3 to give a total of 2 μg DNA. The DNA
vector mix was added to the TransIT-293/OptiMEM mix and incubated for 20 minutes at
room temperature, before being added dropwise to the cells. After 48 hours incubation
at 37 °C, the virus-containing supernatant was harvested, filtered through a 0.45 μm
filter, supplemented with 10 μg/mL hexadimethrine bromide (Polybrene, Sigma-Aldrich)
and used to transduce HeLa cells. HeLa cells were seeded in a 6-well plate to reach
a confluency of roughly 80% at the time of transduction. Growth medium was then
removed and replaced with 1 mL viral supernatant. Following a 3 hour incubation at 37
°C, 2 mL fresh growth medium were added and the plate returned to the incubator. The
following day the transduced cells were expanded into a 10 cm dish and then selection
medium was added 48 hours after transduction. When necessary due to variable levels
of transgene expression in mixed populations of stably transduced cells, cell lines were
single cell cloned by serial dilution.
To generate HeLa cells stably expressing GFP-BirA*, HeLa cells in a 10 cm dish were
transfected with pEGFP-myc-BirA* as described above. Selection medium (500 μg/mL
G418) was added 48 hours after transfection. Fluorescence-activated cell sorting (FACS)
was used to enrich for GFP-positive cells. Cells were washed in PBS/10 mM Hepes,
centrifuged (650 g for 8 minutes) and resuspended in sorting solution (2% FCS in PBS/10
mM Hepes) at roughly 5×106 cells/mL. Sorting was carried out on an InfluxTM cell sorter
(BD Biosciences), with gating for live cells and singlets, and then GFP-positive cells
were collected. To check for the maintenance of GFP-BirA* expression, 4× 105 cells
were washed in PBS and fixed in 300 μL fixation buffer (1.1% formaldehyde, in PBS) and
analysed on a FACSCalibur (BD Biosciences).
2.4 CRISPR/Cas9-mediated gene editing 47
2.4 CRISPR/Cas9-mediated gene editing
2.4.1 Design and cloning of guides
The Ensembl Genome Browser (https://www.ensembl.org/) was used to obtain the
genomic sequences of genes of interest and to determine which exons were conserved
in all transcripts of the gene. The Zhang online CRISPR design tool (http://crispr.mit.
edu/; Hsu et al., 2013) was used to identify suitable guide target sites in each gene (20
nucleotides followed by a 3′ protospacer adjacent motif (PAM) with the sequence NGG).
The human U6 promoter, which drives expression of the sgRNA in the Cas9/sgRNA
delivery vectors, requires a guanine base at the transcriptional start site. Thus, where
guides did not already start with a guanine, an extra guanine was added in front of the
guide sequence. Guides were ordered as pairs of complementary oligos (Sigma-Aldrich)
with the sequences 5′-CACCGN19/20-3′ and 5′-AAACN20C-3′. WARNING: The PAM
sequences present in the genome must not be included in the guide sequence1.
Oligo pairs were phosphorylated and annealed in a reaction containing 10 μM of each
oligo and 0.5 μL of T4 Polynucleotide kinase in 1× T4 DNA ligase buffer, incubated at
37 °C for 30 minutes, 95 °C for 5 minutes and then cooled slowly to room temperature.
The Cas9/sgRNA delivery vector (pX330 for wild-type Streptococcus pyogenes Cas9 or
pX335 for the ‘nickase’ mutant form of S. pyogenes Cas9 with a D10A mutation) was
digested with BbsI and gel purified. Each pair of annealed oligos was then ligated into
the plasmid using T4 DNA ligase (one hour at room temperature). Transformation and
plasmid purification were carried out as described above.
2.4.2 Knockout of AP4B1 and AP4E1 in HeLa cells
For the AP4B1 and AP4E1 knockout HeLa cell lines, all copies of the gene were inactivated
using the ‘double nickase’ CRISPR/Cas9 system (Cong et al., 2013; Ran et al., 2013). Three
pairs of guides were designed for each gene, targeting exons 1, 2 and 3 for AP4B1 and
exons 6, 7 and 11 for AP4E1. The guide sequences and quality scores are listed in Tables
C.1 and C.2 in Appendix C. None of the guide pairs had any predicted off-target sites.
Guide oligos were cloned into pX335-U6-Chimeric_BB-CBh-hSpCas9n(D10A) vectors
(a gift from Feng Zhang; Addgene plasmid 4233563; Cong et al., 2013). HeLa M cells
were transfected with paired pX335 plasmids and pIRESpuro (Clontech) in a ratio of
1An common error that can be made by PhD students or even professors!
48 Materials and Methods
2:2:1. Forty-eight hours later, untransfected cells were killed off by a 4 day selection in 1
μg/mL puromycin. Following this transient selection the surviving cells were expanded
for freezing, Western blot analysis of the mixed population and single cell cloning. Single
cell clones were isolated by serial dilution of the cells leading to the plating of ∼10 cells
per 10 cm dish. The single cell clones were expanded and tested for knockout of AP4B1
or AP4E1 by Western blot and immunofluorescence. AP4B1 clone x2A3 and AP4E1
clone x6C3 were negative for AP4B1/AP4E1 expression in both assays and were further
validated by sequencing. Genomic DNA was harvested using a High Pure PCR Template
Purification Kit (Roche) and PCR was used to amplify ∼500 bp regions around the target
sites. The PCR products were then blunt-end cloned (Zero Blunt PCR Cloning Kit;
Invitrogen) and clones were sent for Sanger sequencing with the M13_F primer (24 PCR
clones for AP4B1 clone x2A3 and 17 PCR clones for AP4E1 clone x6C3).
2.4.3 Depletion of AP4B1 and AP4E1 in SH-SY5Y cells
For depletion of AP4B1 and AP4E1 in SH-SY5Y cells, a lentiviral CRISPR/Cas9 system was
used (Timms et al., 2016). In this system wild-type S. pyogenes Cas9 is delivered separately
from the sgRNA (this time a single guide rather than paired guides). Lentivirus was
produced by transfecting HEK 293ET cells with the lentiviral vector plus the packaging
plasmids pCMVΔR8.91 and pMD.G in a ratio of 10:7:3, using TransIT-293 Transfection
Reagent. Harvest of viral supernatants, lentiviral transductions and selection for stable
expression were performed as described above for retroviral transductions. Wild-type
Cas9 was stably introduced into SH-SY5Y cells using the lentiviral vector pHRSIN-PSFFV-
FLAG-Cas9-PPGK-Hygro (a gift from Paul Lehner, University of Cambridge). The guides
that were most effective for the knockout of AP4B1 and AP4E1 in HeLa cells (1A, 2A and 2B
for AP4B1 and 6A, 6B and 7A for AP4E1) were cloned into the lentiviral sgRNA expression
vector pKLV-U6gRNA(BbsI)-PGKpuro2ABFP (a gift from Kosuke Yusa; Addgene 5094666;
Koike-Yusa et al., 2014). Cas9-expressing SH-SY5Y cells were then transduced with these
sgRNA expression vectors. Mixed populations of cells were selected for stable expression
of Cas9 and gRNA and were assessed by Western blotting.
2.4.4 Endogenous tagging of SERINC1 and SERINC3
For CRISPR/Cas9-mediated endogenous tagging of SERINC1 and SERINC3 in HeLa cells,
we used homology-directed repair to introduce a myc-Clover tag at the C terminus of
each protein (see Figure 4.20 in Section 4.4.2). This time a transient wild-type S. pyogenes
Cas9 delivery system was used. Two guides were designed for each gene to enable Cas9
2.4 CRISPR/Cas9-mediated gene editing 49
to cut downstream and proximal to the STOP codon. For SERINC1 the guide sequences
were: ATACACAACTTTACAAAAGT (guide S1_A) and GAACACTGGAGAAGTTACAT (guide
S1_B). For SERINC3 the guide sequences were: GACACCACTGGAACTCACAA (guide
S3_A) and GGTATATGGGTTTTCGGTGA (guide S3_B). Guide oligos were cloned into
pX330-U6-Chimeric_BB-CBh-hSpCas9 vectors (a gift from Feng Zhang; Addgene plasmid
223063; Cong et al., 2013). To generate the homology-directed repair plasmids we made
use of a pDonor_myc-Clover plasmid which was a gift from Dick van den Boomen and
Paul Lehner (University of Cambridge). This was created using a plasmid gifted to them
by Matthew Porteus and Ron Kopito (Stanford University), originally containing a TAP-
tag, which was replaced with myc-Clover cloned from pcDNA3.1-Clover-mRuby2 (a gift
from Kurt Beam; Addgene 4908967; Lam et al., 2012). pDonor_myc-Clover contains a 5′
homology region, followed by myc-Clover, an internal ribosome entry site, a puromycin
resistance gene and then a 3′ homology region. The existing homology regions were
replaced by Gibson Assembly with regions specific for SERINC1 (ENSG00000111897: 5′ –
806 bp preceding the STOP codon; 3′ – 814 bp starting 161 bp after the STOP codon) or
SERINC3 (ENSG00000132824: 5′ – 835 bp preceding the STOP codon; 3′ – 817 bp starting
95 bp after the STOP codon). 3′ homology regions were chosen to avoid the gRNA target
sites. The primers used to clone the homology regions are provided in Appendix B.
HeLa M cells were transfected with the pX330 and pDonor plasmids in a ratio of 1:1.
Forty-eight hours later selection for stable expression of the puromycin resistance gene
(meaning incorporation of the Clover tag) was initiated. Following selection, mixed
populations of knockin cells were analysed by flow cytometry (as described above for the
GFP-BirA* cell line in Section 2.3.5). Single cell clones were isolated (as described above)
and tested for knock in of Clover by Western blotting and immunofluorescence with an
anti-GFP antibody. Clones SERINC1-Clover A3 and SERINC3-Clover B6 were positive
for Clover expression in both assays. To test for correct integration of the tag, genomic
DNA was harvested from each cell line using a High Pure PCR Template Purification Kit
(Roche) and PCR was used to amplify a ∼500 bp region around the position where the
endogenous gene and the myc-Clover tag meet. PCR products were blunt-end cloned
(Zero Blunt PCR Cloning Kit; Invitrogen) and clones were sent for Sanger sequencing
with the M13_F primer.
50 Materials and Methods
2.5 siRNA-mediated knockdown
Transfections of siRNA were carried out with Oligofectamine (Thermo Fisher Scien-
tific) and where indicated cells were mock treated with Oligofectamine without siRNA
or transfected with ON-TARGETplus Non-targeting siRNA #1 (D-001810-01, Dharma-
con). Knockdowns were typically performed in a 6-well plate format, except for vesicle-
enriched fraction experiments where they were scaled up for 15 cm dishes. Cells were
plated approximately 16–24 hours before transfection and seeding was adjusted to
achieve 30–50% confluency at the time of transfection. For each well, 5 μL Oligofec-
tamine was mixed with 35 μL OptiMEM and the reaction incubated at room temperature
for 5 minutes. siRNA oligos were diluted in OptiMEM to 5× the desired final concentra-
tion of siRNA, in a total volume of 160 μL. The oligofectamine and siRNA mixes were
combined and incubated at room temperature for a further 20 minutes. Medium on the
cells was then replaced with 800 μL antibiotic-free medium and the siRNA transfection
mix was added drop-wise to the dish. Cells were incubated at 37 °C, 5% CO2 overnight
before the transfection mix was replaced with fresh full growth medium.
Knockdown of AP-4 was achieved by combined siRNA targeting of AP4E1 and AP4M1
using ON-TARGETplus SMARTpools (AP4E1, L-021474-00; AP4M1, L-011918-01; Dhar-
macon), using a double-hit 96 hours protocol (as described in Borner et al., 2012). For
the first hit the final concentration of siRNA was 40 nM (20 nM AP4M1 + 20 nM AP4E1).
The second hit was performed 48 hours after the first hit with half the final concentra-
tion of siRNA. For knockdown of RUSC1 an ON-TARGETplus SMARTpool (L-020607-01,
Dharmacon) was used. Knockdown of RUSC2 was problematic so two different siRNA
pools and eight individual siRNA oligos were trialled (listed in Table 2.2). The most
effective oligos, Hs_RUSC2_4 and Hs_RUSC2_7 (Qiagen), were used for the RUSC deple-
tion experiments. Both resulted in similar knockdown efficiencies at the mRNA level
(∼70% reduction). After optimisation, a double-hit 72 hours protocol was used for RUSC
knockdowns, with a final siRNA concentration of 20 nM for the first hit and 10 nM for the
second hit, which was performed 36 hours later. For the combined RUSC1 and RUSC2
knockdowns the final concentration of siRNA was 20 nM for each gene (10 nM for the
second hit).
2.5 siRNA-mediated knockdown 51
Ta
b
le
2.
2
R
U
SC
2
si
R
N
A
o
li
go
s.
E
ig
h
ti
n
d
iv
id
u
al
si
R
N
A
o
lig
o
s
an
d
tw
o
si
R
N
A
p
o
o
ls
w
er
e
te
st
ed
fo
r
kn
o
ck
d
ow
n
o
fR
U
SC
2.
O
li
go
Ve
n
d
o
r
C
at
al
o
g
n
u
m
b
er
Se
q
u
en
ce
H
s_
R
U
SC
2_
4
Fl
ex
iT
u
b
e
si
R
N
A
Q
ia
ge
n
SI
00
70
91
28
C
T
G
G
U
U
U
A
A
U
C
A
C
C
U
C
U
A
U
A
A
H
s_
R
U
SC
2_
5
Fl
ex
iT
u
b
e
si
R
N
A
Q
ia
ge
n
SI
03
14
68
22
A
G
G
G
A
C
A
A
G
U
A
U
A
C
A
C
G
A
A
U
A
H
s_
R
U
SC
2_
6
Fl
ex
iT
u
b
e
si
R
N
A
Q
ia
ge
n
SI
04
29
26
73
A
G
G
G
C
A
C
U
U
G
C
U
G
U
A
G
C
C
A
U
A
H
s_
R
U
SC
2_
7
Fl
ex
iT
u
b
e
si
R
N
A
Q
ia
ge
n
SI
04
29
62
64
C
A
G
G
C
G
U
C
U
G
U
U
U
A
U
G
U
A
U
U
U
Fl
ex
iT
u
b
e
G
en
eS
o
lu
ti
o
n
fo
r
R
U
SC
2
(p
o
o
l)
Q
ia
ge
n
G
S9
85
3
T
h
e
fo
u
r
se
q
u
en
ce
s
ab
ov
e.
R
U
SC
2
si
R
N
A
J-
02
61
33
-0
9
D
h
ar
m
ac
o
n
L-
Q
-0
26
13
3-
01
(S
et
o
f4
)
G
G
A
C
A
A
G
U
A
U
A
C
A
C
G
A
A
U
A
R
U
SC
2
si
R
N
A
J-
02
61
33
-1
0
D
h
ar
m
ac
o
n
L-
Q
-0
26
13
3-
01
(S
et
o
f4
)
A
G
U
C
A
U
A
C
C
A
U
G
C
G
C
U
U
C
A
R
U
SC
2
si
R
N
A
J-
02
61
33
-1
1
D
h
ar
m
ac
o
n
L-
Q
-0
26
13
3-
01
(S
et
o
f4
)
G
C
C
C
U
C
A
A
G
U
G
G
C
G
G
G
A
A
U
R
U
SC
2
si
R
N
A
J-
02
61
33
-1
2
D
h
ar
m
ac
o
n
L-
Q
-0
26
13
3-
01
(S
et
o
f4
)
U
C
A
U
C
A
U
C
G
G
G
C
A
G
C
G
U
A
A
O
N
-T
A
R
G
E
T
p
lu
s
H
u
m
an
R
U
SC
2
SM
A
R
T
p
o
o
l
D
h
ar
m
ac
o
n
L-
02
61
33
-0
1
T
h
e
fo
u
r
se
q
u
en
ce
s
ab
ov
e.
52 Materials and Methods
2.6 Quantitative RT-PCR
For quantitative RT-PCR analysis, total RNA was extracted from around 5× 105 cells
using an RNeasy Mini Kit (Qiagen) with on-column DNase digestion using an RNase-
Free DNase Set (Qiagen). A total of 500 ng of total RNA was reverse-transcribed using
TaqMan Reverse Transcription Reagents (Thermo Fisher Scientific) in a 20 μL reaction.
Quantitative-PCR was performed using TaqMan gene expression assays (Thermo Fisher
Scientific), which include two unlabelled PCR primers and one FAM dye-labelled TaqMan
MGB probe; TaqMan Assay IDs were Hs00204904_m1 for RUSC1 and Hs00922017_m1
for RUSC2. The cDNA was diluted 1:5, and 5 μL of this dilution (25 ng RNA equivalent)
was used for each PCR reaction using TaqMan Universal PCR Master Mix (Thermo Fisher
Scientific) and using a 7900HT Fast Real-Time PCR System (Thermo Fisher Scientific),
with Standard mode thermal cycling conditions. Every reaction was carried out in tech-
nical triplicate and for the RUSC depletion experiments knockdowns were performed in
biological triplicate. For each sample, the levels of RUSC1 and RUSC2 mRNAs were nor-
malised using GAPDH as a loading control (TaqMan Assay ID Hs99999905_m1; Thermo
Fisher Scientific). The data were analysed in Microsoft Excel using the ΔΔCt method,
relative to a non-targeting siRNA control. Results are expressed as means ± SEM for
each biological triplicate.
2.7 Microscopy
2.7.1 Fluorescence microscopy
Cells were grown onto 13 mm glass coverslips and fixed in 3% formaldehyde or, for AP4E1
labelling, ice-cold methanol. For formaldehyde fixation, coverslips were washed in PBS
and fixed in 3% formaldehyde in PBS for 20 minutes at room temperature. Coverslips
were washed in PBS and 20 mM glycine in PBS was added for 5 minutes to quench
any residual formaldehyde. Coverslips were then washed again three times in PBS,
permeabilised for 10 minutes in 0.1% saponin in PBS and blocked for 10 minutes in 1%
bovine serum albumin (BSA)/0.01% saponin in PBS. For methanol fixation, coverslips
were washed in PBS and fixed in ice cold 100% methanol for 5 minutes on ice. Methanol
was removed, coverslips washed three times in PBS and blocked for 10 minutes in 0.5%
BSA in PBS. Primary antibody (diluted in BSA block) was added for 45 minutes at room
temperature. Coverslips were washed three times in BSA block and then fluorochrome-
2.7 Microscopy 53
conjugated secondary antibody was added in block for 30 minutes at room temperature.
Coverslips were then washed three times in PBS, followed by a final wash in dH2O,
before being mounted in ProLong Diamond Antifade Reagent with DAPI (Thermo Fisher
Scientific). Widefield images were captured on an Axio Imager II microscope (63×/1.4
NA oil immersion objective; AxioCam 506 camera; ZEISS) and confocal and Airyscan
enhanced resolution images were captured on an LSM880 confocal microscope with
Airyscan (63×/1.4 NA oil immersion objective; ZEISS), both equipped with ZEN software
(ZEISS). Airyscan images were taken in SR (super resolution) mode and raw data were
processed using Airyscan processing in ‘auto strength’ mode (strength = 6.0) with Zen
Black software version 2.3.
Quantification of SERINC1/3-Clover and ATG9A colocalisation was performed on Airyscan
images. Colocalisation was measured using Pearson’s Correlation Coefficient with Costes
thresholding method (Volocity software 6.3; Perkin Elmer), in a peripheral 10 μm2 area
in each cell, selected while viewing the green channel only. A minimum of 19 cells were
analysed for each condition. For statistical analysis data were analysed by a two-tailed
Mann-Whitney U test.
2.7.2 Automated imaging
For automated imaging of ATG9A localisation and LC3B puncta, cells were plated in
96-well microplates (6005182, Perkin Elmer), which were either uncoated or for the
LC3B assay coated with poly-D-lysine. Cells were fixed in 3% formaldehyde in PBS,
permeabilised with 0.1% saponin and labelled for immunofluorescence as described
above. After washing off secondary antibody, cells were stained with HCS CellMask Blue
stain (Thermo Fisher Scientific) diluted 1:5000 in PBS for 30 minutes at room temperature.
Cells were then washed three times in PBS before imaging using a CellInsight CX7 High-
Content Screening Platform (Olympus 20×/0.4 NA objective; Thermo Fisher Scientific)
running HCS Studio 3.0 software. Autofocus was applied using the whole cell mask
channel (channel 1). Experiments were performed in biological triplicate with a technical
triplicate (three separate wells per cell line) within each experiment. Statistical analyses
of imaging data were performed using GraphPad Prism version 5.01 (GraphPad Software).
ATG9A localisation was quantified using the Colocalization Bioapplication V4 (Cellomics,
Thermo Fisher Scientific), using anti-TGN46 to segment the TGN (channel 2; ROI_B).
ROI_A was defined by the whole cell mask (channel 1) minus ROI_B. The average intensity
of anti-ATG9A (channel 3; target 1) was then quantified in the two regions and a ratio
calculated between the two. Ratios were normalised to the mean wild-type ratio. At least
54 Materials and Methods
1400 cells were scored per cell line in each experiment. For statistical analysis data were
log transformed and analysed by one-way ANOVA with Dunnett’s Multiple Comparison
Test.
LC3B puncta were quantified using the Spot Detector Bioapplication V4 (Cellomics,
Thermo Fisher Scientific). Spots were identified with smoothing on (uniform; value = 1),
with the detection method Box (value = 1) and ThreeSigma thresholding (value = 0.012).
Spot total count and average area (in μm2) were measured. At least 500 cells were scored
per cell line in each experiment. For statistical analysis data were analysed by one-way
ANOVA with Dunnett’s Multiple Comparison Test.
2.7.3 Correlative light and electron microscopy (CLEM)
HeLa GFP-RUSC2 (clone 3) cells were mixed with wild-type HeLa cells and seeded on
alpha-numeric gridded glass-bottom coverslips (P35G-1.5-14-C-GRID, MatTek) to be
40–50% confluent at the time of fixation. Preparation for CLEM and imaging were
performed by James Edgar (Robinson Lab, University of Cambridge). Cells were fixed
with 2% formaldehyde/2.5% glutaraldehyde/0.1 M cacodylate buffer for 30 minutes
at room temperature and washed with 0.1 M cacodylate. Cells were then stained with
Hoechst (to stain the nucleus) for 2 minutes, before being washed with 0.1 M cacodylate.
GFP-RUSC2 fluorescence signal was imaged on an LSM780 confocal microscope (63×/1.4
NA oil immersion objective; ZEISS) and the coordinates of cells selected for imaging
were recorded. To prepare for electron microscopy, cells were secondarily fixed with 1%
osmium tetroxide/1.5% potassium ferrocyanide and then incubated with 1% tannic acid
in 0.1 M cacodylate to enhance membrane contrast. Samples were washed with dH2O
and dehydrated using increasing concentrations of ethanol. Epoxy resin (Araldite CY212
mix, Agar Scientific) was mixed at a 1:1 ratio with propylene oxide and this was used
for one hour to infiltrate the samples with resin, following which it was replaced with
neat Epoxy resin. Pre-baked resin stubs were inverted over coordinates of interest, resin
was cured overnight at 65 °C, following which stubs were removed from coverslips using
liquid nitrogen. Areas of interest were identified by alpha-numeric coordinates and 70
nm ultrathin sections were collected using a Diatome diamond knife attached to an
ultracut UCT ultramicrotome (Leica). As areas of interest were at the very basal surfaces
of cells (and so the very top of the resin stub), sections were immediately collected onto
piloform-coated slot grids. Sections were stained with lead citrate before being imaged
on a Tecnai Spirit transmission electron microscope (FEI) at an operating voltage of 80
kV. HeLa GFP-RUSC2 and wild-type cells were imaged, and peripheral accumulations of
2.8 Western blotting 55
uncoated vesicular and tubular structures were only observed in the regions of HeLa
GFP-RUSC2 cells that correlated with the GFP fluorescence.
For the CLEM analysis of RUSC2 overexpressing cells in starvation conditions, HeLa
RUSC2-GFP (clone 1) cells were seeded on alpha-numeric gridded glass-bottom cover-
slips, as above, and incubated in EBSS with 100 nM Bafilomycin A1 for two hours before
fixation. All other steps were performed as described above.
2.8 Western blotting
Estimations of protein concentrations were made using a Pierce BCA Protein Assay Kit
(Thermo Fisher Scientific). Unless otherwise noted, cells were lysed for Western blot
analysis in 2.5% (w/v) SDS/50 mM Tris pH 8. Lysates were passed through a QIAshredder
column (Qiagen) to shred DNA, incubated at 65 °C for 3 minutes and then boiled in
NuPAGE LDS Sample Buffer (Thermo Fisher Scientific). Samples were loaded at equal
protein amounts for SDS–PAGE, performed on NuPAGE 4—12% Bis–Tris gels in NuPAGE
MOPS SDS Running Buffer, or for LC3B blots, on NuPAGE 12% Bis-Tris gels in NuPAGE
MES SDS Running Buffer (all Thermo Fisher Scientific). PageRuler Plus Prestained
Protein Ladder (Thermo Fisher Scientific) was used to estimate the molecular size of
bands. Proteins were transferred to nitrocellulose membrane by wet transfer (Mini-
PROTEAN Tetra Cell, Bio-Rad) at 150 mA for 3 hours or 100 mA overnight. Membranes
were blocked in 5% w/v milk in PBS with 0.1% v/v Tween-20 (PBS-T) for at least 30
minutes at room temperature. Primary antibodies were diluted in 5% milk in PBS-T or
5% BSA in 50 mM Tris-HCl pH 7.4/150 mM NaCl (TBS) with 0.1% v/v Tween-20 (TBS-
T). Primary antibodies were added for at least 1 hour at room temperature, followed
by washing in PBS-T, incubation in secondary antibody (in 5% milk in PBS-T) for 30
minutes at room temperature, washing in PBS-T and finally PBS. For the detection of
biotinylated proteins, membranes were blocked in 5% BSA in TBS-T at room temperature
for 1 hour. HRP-conjugated streptavidin (Invitrogen; 1:5000 in 5% BSA) was added for
1 hour at room temperature with rocking and then membranes were washed as above.
Chemiluminescence detection of HRP-conjugated secondary antibody was carried out
using Amersham ECL Prime Western Blotting Detection Reagent (GE Healthcare) and
X-ray film.
56 Materials and Methods
2.9 GST pulldowns
All steps were performed on ice with pre-chilled ice-cold buffers, unless otherwise noted.
2.9.1 Preparation of cytosol
To prepare cytosol for pulldowns, HeLa cells stably expressing EGFP or GFP-RUSC2
(clone 3), each in two 15 cm plates, were washed in PBS and then cytosol buffer (20 mM
HEPES pH 7.5, 150 mM NaCl) and scraped in a total volume of 800 μL cytosol buffer plus
2 mM DL-Dithiothreitol (DTT, D5545, Sigma Aldrich). Cells were transferred to a 1 mL
Dounce homogeniser (Wheaton) and homogenised with 15 strokes with the tight pestle,
followed by 10 passes through a 30.5 gauge needle. Homogenates were then centrifuged
at 78,400 g (RCF max) in a TLA-110 rotor (Beckman Coulter) for 30 minutes to pellet cell
debris and the supernatants (cytosol) transferred to new tubes, snap frozen in liquid
nitrogen, before storage at −80 °C until use in the pulldown assay.
2.9.2 Protein expression and purification
Expression and purification of GST-tagged AP-4 appendage domains were performed
by Tara Archuleta and Lauren Parker Jackson (University of Vanderbilt). Human AP4E1
(residues 881–1135; Borner et al., 2012) and AP4B1 (residues 612–739; Frazier et al.,
2016) appendage domains were expressed as GST-fusion proteins in BL21(DE3)pLysS
cells (Invitrogen) for 16–20 hours at 22 °C after induction with 0.4 mM IPTG. Proteins
were purified in 20 mM HEPES pH 7.5, 200 mM NaCl, and 2 mM 2-Mercaptoethanol.
Cells were lysed using a disruptor (Constant Systems Limited) and proteins were affinity
purified using glutathione Sepharose (GE Healthcare). Fusion proteins were eluted
in buffer with 30 mM reduced glutathione and further purified by gel filtration on a
Superdex S200 preparative column (GE Healthcare).
2.9.3 GST pulldowns
GST pulldowns were performed by Tara Archuleta and Lauren Parker Jackson (University
of Vanderbilt). For the pulldowns with AP-4 appendage domains, glutathione Sepharose
4B (GE Healthcare) resin was batch equilibrated with 20 mM HEPES pH 7.5, 150 mM
NaCl, 2 mM DTT. The 50% resin slurry (60 μL) was incubated and rotated with 50 μg
of GST, GST-AP4E1, or GST-AP4B1 appendage for 1 hour at 4 °C. Equal total protein
2.10 Immunoprecipitations 57
amounts of HeLa lysates containing GFP-RUSC2 were added independently to each bait
sample. Samples were incubated and rotated for 1 hour at 4° C. Samples were centrifuged
at 2500 g for 5 minutes. Supernatant was removed, and resin was washed with 1 mL of
wash buffer (20 mM HEPES pH 7.5, 150 mM NaCl, 2 mM DTT, 0.5% NP-40) for a total of
3 washes. After final removal of supernatant, 65 μL elution buffer (20 mM HEPES pH 7.5,
150 mM NaCl, 2 mM DTT, 60 mM reduced glutathione) were added to each sample and
incubated for 10 minutes at 4 °C. Samples were centrifuged at 5000 g for 5 minutes, then
supernatant was removed and transferred to a new tube. 10 μL fresh 1M DTT and 25 μL
SDS loading dye were added, and samples were boiled at 95 °C for 10 minutes. Control
HeLa lysates containing GFP were treated in the same way as a control.
2.10 Immunoprecipitations
All steps were performed on ice with pre-chilled ice-cold buffers, unless otherwise noted.
2.10.1 Immunoprecipitation of AP4B1 and AP4E1
For immunoprecipitations of AP4B1 and AP4E1, a 10 cm dish of cells was washed in
PBS and incubated for 10 minutes in 1 mL 1% Triton TX-100 in PBS, supplemented with
200 μM AEBSF protease inhibitor. Lysates were cleared by centrifugation at 16,000 g
for 10 minutes. Roughly 10% of each lysate was retained as an input sample and the
remainder (1 mL) was pre-cleared by incubation with 30 μL packed protein-A-Sepharose
for 1 hour. Input samples were quantified using a Pierce BCA Protein Assay Kit. The
protein-A-Sepharose was removed from the lysates by centrifugation at 8,600 g for 30
seconds. Lysates were adjusted to the same concentration with 1% Triton TX-100 and
then 5 μL antibody were added to each sample. Samples were incubated with antibody
for 3 hours and then 30 μL packed protein-A-Sepharose were added to each for 1 hour.
Following this, protein-A-Sepharose was pelleted (as above) and washed three times with
2 mL 1% Triton TX-100 and three times with PBS. The Sepharose was then resuspended
in 100 μL 1× NuPAGE LDS Sample Buffer, boiled for 5 minutes at 95 °C and pelleted
(16,000 g, 30 seconds) to prepare for analysis by Western blotting.
58 Materials and Methods
2.10.2 GFP-trap of GFP-tagged TEPSIN and RUSC2
Immunoprecipitations of wild-type and mutant TEPSIN-GFP (Section 4.2.3) were per-
formed using GFP-Trap A beads. Wild-type HeLa cells and HeLa cells stably expressing
EGFP, wild-type TEPSIN-GFP or mutant TEPSIN-GFP [L470S/F471S] in 10 cm plates were
washed once in PBS and then scraped in 500 μL GFP-trap lysis buffer (10 mM Tris-HCl pH
7.5, 10 mM NaCl, 0.5 mM EDTA, 0.5% NP-40), supplemented with cOmplete EDTA-free
protease inhibitor (Sigma-Aldrich). Cells were incubated on ice in lysis buffer for 10
minutes and then lysates were cleared by centrifugation at 16,000 g for 10 minutes. A
protein assay was performed and, if required, lysates were adjusted to equal protein
concentrations with lysis buffer. A portion of each lysate was retained as input and
the remainder was incubated with GFP-Trap A beads at 4 °C for 3 hours with rotation.
Beads were washed five times with GFP-trap lysis buffer and then boiled in NuPAGE LDS
Sample Buffer at 75 °C for 10 minutes to prepare for Western blot analysis.
Immunoprecipitations of GFP-RUSC2 (Section 4.5.3) from wild-type and AP4B1 knock-
out HeLa cells stably expressing GFP-RUSC2 (mixed populations), and parental wild-type
HeLa cells, were performed in the same way.
2.10.3 Conventional immunoprecipitation of TEPSIN-GFP for MS
The conventional immunoprecipitations of TEPSIN-GFP for mass spectrometry (MS;
Section 3.6.1) were performed by Dr Georg Borner while working in the Robinson Lab.
The protocol was based on that published in Antrobus & Borner (2011). Wild-type or
TEPSIN-GFP expressing cells (Borner et al., 2012) were SILAC labelled (in biological
duplicate, with label-swap). For harvesting, cells from 2 x 500 cm2 dishes (per IP) were
washed twice in PBS, scraped into 10 mL of PBS-TT (PBS, 0.2% (v/v) Triton X-100, 0.1%
(v/v) Tween-20), and incubated for 25 minutes with rotation. Insoluble material was
removed by centrifugation at 4,800 g for 3 minutes, followed by centrifugation at 21,000
g for 20 minutes. The supernatant was then further cleared by filtration through a 0.22
μm syringe filter. Lysates were pre-absorbed against Protein A Sepharose beads. Approx-
imately 60 μg of rabbit polyclonal anti-GFP antibody (a gift from Matthew Seaman) was
added to cleared lysates. After incubation at 4 °C for 90 minutes with rotation, 50 μL of
Protein A Sepharose beads were added, and samples were incubated for a further 45
minutes. Beads were washed four times in PBS-T, once in PBS, and immunoprecipitates
recovered in 100 μL Soft Elution Buffer (0.2% (w/v) SDS, 0.1% (v/v) Tween-20, 50 mM
Tris-HCl pH 8) by incubation for 7 minutes at 25 °C. Eluates from TEPSIN-GFP express-
2.11 BioID streptavidin pulldowns 59
ing and wild type control cells were then combined prior to acetone precipitation and
analysis by mass spectrometry.
2.10.4 Sensitive immunoprecipitation of TEPSIN-GFP for MS
For the sensitive immunoprecipitations of TEPSIN-GFP (Section 3.6.1), wild-type HeLa
cells (control) and HeLa cells stably expressing TEPSIN-GFP (Borner et al., 2012) were
grown in SILAC media. For two biological replicates (data shown in Figure 3.24) the wild-
type cells were heavy labelled and the TEPSIN-GFP cells light labelled. A third replicate
was performed with a label swap and used to filter out unlabelled proteins from the
data. For each cell line, two 10 cm dishes were washed once in PBS (without CaCl2 and
MgCl2; 14190–094, Thermo Fisher Scientific) and then scraped in 4 mL PBS. Cells were
transferred to a Dounce homogeniser (Sartorius) and homogenised with 20 strokes with
the tight pestle, followed by two passes through a 21 gauge needle. Triton-TX-100 was
added to the cells to a final concentration of 0.01% or 0.025%, cells were incubated at 4 °C
for 20 minutes with rotation, and homogenates were cleared by centrifugation at 4,000 g
for 10 minutes. A portion of each homogenate was retained as input and the remainder
was incubated with GFP-Trap A beads (ChromoTek) at 4 °C for 3 hours with rotation.
Beads were washed five times with 0.01% Triton-TX-100 and then immunoprecipitates
were eluted in 100 μL 2.5% (w/v) SDS/50 mM Tris pH 8 and heated at 65 °C for 5 minutes.
Beads were pelleted and supernatants (immunoprecipitates) transferred to new tubes.
Equal volumes of control and TEPSIN-GFP immunoprecipitates were mixed and in-
solution tryptic digest and peptide purification (single shot) were performed as described
below.
2.11 BioID streptavidin pulldowns
All steps were performed on ice with pre-chilled ice-cold buffers, unless otherwise noted.
Streptavidin pulldowns for the AP-4 BioID experiments were carried out from HeLa cells
stably expressing BirA*-tagged AP4B1/E1/M1/S1 and control wild-type HeLa, HeLa BirA*
and HeLa GFP-BirA* cells, in triplicate (experiments performed on three separate days).
Cells were cultured in the presence of 50 μM biotin for 24 hours prior to performing the
experiment. Cells were harvested by scraping into 5 mL PBS, pelleted (600 g, 5 minutes)
and washed twice in PBS. Lysis was performed in 1 mL RIPA buffer (TBS, 1% NP-40,
0.5% sodium deoxycholate, 1 mM EDTA, 0.1% SDS) supplemented with cOmplete EDTA-
free protease inhibitor. DNA was sheared by running the lysates ten times through a
60 Materials and Methods
25 gauge needle, lysates were incubated at 4 °C for 10 minutes with mixing and then
were sonicated (three times 5 seconds bursts with an amplitude of 10 microns). Lysates
were cleared by centrifugation (16,000 g, 15 minutes) and supernatants transferred to
new tubes and normalised to cell pellet weight with RIPA buffer. Biotinylated proteins
were affinity purified using Pierce High Capacity Streptavidin Agarose (Thermo Fisher
Scientific) by incubating with the lysates for 3 hours with rotation at 4 °C. Beads were
then pelleted and washed three times in RIPA buffer, twice in TBS and three times in 50
mM ammonium bicarbonate (ABC) pH 8 (09830, Sigma Aldrich), before resuspension in
50 mM ABC. Proteins were then reduced by the addition of 10 mM DTT at 56 °C for 30
minutes and alkylated by the addition of 55 mM iodoacetamide (I1149, Sigma Aldrich)
at room temperature for 20 minutes in the dark. Proteins were enzymatically digested
by addition of 1 μg Trypsin (V5280, Promega; stock at 0.1 mg/mL in 1 mM HCl) and
overnight incubation at 37 °C. The following day the beads were pelleted, tryptic peptides
(supernatant) collected, spiked with 1 μL 100% trifluoroacetic acid (TFA), dried almost
to completion in a centrifugal vacuum concentrator (Concentrator 5301, Eppendorf)
and then stored at −20 °C. Later samples were thawed, resuspended in a total volume of
100 μL 1% (v/v) TFA and purified on SDB-RPS StageTips (single shot) as described below.
Streptavidin pulldowns for the SERINC BioID experiments were carried out in an identi-
cal manner from HeLa cells stably expressing BirA*-tagged SERINC1 or SERINC3 and the
same control lines as detailed above. However, pulldowns for SERINC1 and SERINC3
were performed on separate days, each with their own set of control pulldowns (each in
triplicate).
2.12 Preparation of vesicle-enriched fractions
For the comparative proteomic profiling of the vesicle fraction of AP-4-depleted cells (Sec-
tion 3.4), vesicle-enriched fractions were prepared from paired SILAC-labelled control
and AP-4-depleted HeLa cell lines: wild-type versus AP4B1 knockout (two experiments),
wild-type versus AP4E1 knockout (two experiments), wild type (untreated) versus AP-4
knockdown (three experiments), control (untreated) versus AP-4 knocksideways for
10 minutes (three experiments), control (untreated) versus AP-4 knocksideways for 60
minutes (two experiments). Replicate experiments were performed on separate days.
All steps were performed on ice with pre-chilled ice-cold buffers, unless otherwise noted.
For each sample eight confluent 15 cm dishes of cells were washed with PBS and scraped
in a total volume of 7.5 mL Buffer A (0.1 M MES pH 6.5, 0.2 mM EGTA and 0.5 mM MgCl2).
2.13 Generation of Dynamic Organellar Maps and membrane fractions 61
Cells were homogenised with 20 strokes of a motorised Potter-Elvehjem homogeniser
(or a hand-held Dounce homogeniser with tight pestle) and cell debris was removed
by centrifugation at 4,150 g for 32 minutes. Supernatants were treated with 50 μg/mL
ribonuclease A (MP Biomedicals) for 1 hour and then partially digested ribosomes were
pelleted by centrifugation (4,150 g for 3 minutes) and discarded. Membranes were
pelleted by centrifugation at 55,000 rpm (209,900 g RCFmax) for 40 minutes in an MLA-
80 rotor (Beckman Coulter). Membrane pellets were resuspended in 400 μL Buffer A
using a 1 mL Dounce homogeniser, mixed with an equal volume of FS buffer (12.5%
[w/v] Ficoll and 12.5% [w/v] sucrose, in buffer A) and centrifuged at 20,000 rpm (21,700
g RCFmax) for 34 minutes in a TLA-110 rotor, to pellet contaminants. Supernatants
were diluted with four volumes of Buffer A and centrifuged at 40,000 rpm (86,700 g
RCFmax) in a TLA-110 rotor for 30 minutes to obtain the vesicle-enriched fraction
(pellet). Pellets were resuspended in 50 μL 2.5% SDS (in 50 mM Tris pH 8), heated at 65 °C
for 3 minutes and centrifuged to pellet insoluble material (16,000 g, 1 minute). For mass
spectrometry, equal amounts of protein (20–50 μg) from paired SILAC-labelled control
and AP-4-depleted vesicle fractions were mixed and either processed by in-solution or
in-gel tryptic digest as described below.
The vesicle-enriched fractions from the SERINC1 and SERINC3 knockin cells (Section
4.4.4), with and without AP-4 knockdown, were performed in an identical manner but
from four confluent 15 cm dishes of cells per sample.
2.13 Generation of Dynamic Organellar Maps and
membrane fractions
Organellar maps were prepared from wild-type (control), AP4B1 knockout and AP4E1
knockout HeLa cells, in duplicate (six maps in total). The method was previously de-
scribed in detail by Itzhak et al. (2016). Maps were prepared on two separate days, with
a complete set of three on each occasion (one control, one AP4B1 knockout, and one
AP4E1 knockout).
All steps were performed on ice with pre-chilled ice-cold buffers. HeLa cells (1× 15
cm dish SILAC light and 1× 15 cm dish SILAC heavy per map) were washed in PBS
(without CaCl2 and MgCl2), incubated in PBS for 5 minutes, washed in hypotonic lysis
buffer (25 mM Tris-HCl pH 7.5, 50 mM sucrose, 0.5 mM MgCl2, 0.2 mM EGTA), and then
incubated in hypotonic lysis buffer for 5 minutes. Cells were scraped in 4 mL hypotonic
lysis buffer and mechanically lysed with 15 strokes of a Dounce homogeniser (8530700,
62 Materials and Methods
tight pestle; Sartorius). The sucrose concentration was then restored to 250 mM. Lysates
were centrifuged at 1,000 g for 10 minutes to pellet nuclear material and post-nuclear
supernatants were transferred to new tubes. The SILAC light post-nuclear supernatant
was subfractionated into five fractions by a series of differential centrifugation steps:
3,000 g for 10 minutes, 5,400 g for 15 minutes, 12,200 g for 20 minutes, 24,000 g for 20
minutes, 78,400 g for 30 minutes (all speeds RCF max). All pellets were resuspended
in 2.5% SDS/50 mM Tris pH 8 and heated for 5 minutes at 72 °C. In parallel, a single
membrane fraction was obtained from the SILAC heavy post-nuclear supernatant by
centrifugation at 78,400 g (RCF max) for 30 minutes. This fraction served as an internal
reference, by spiking it into each of the ‘light’ subfractions. Analysis by mass spectrometry
provided a ratio of enrichment/depletion for each protein in each subfraction, relative to
the standard. All five ratios combined yielded an abundance distribution profile for each
protein across the subfractions. Principal component analysis revealed which proteins
had similar fractionation profiles (apparent as organellar clusters in Figure 3.8).
The membrane fractions analysed by label free quantification (Section 3.7.2) are the
same as the heavy reference membrane fractions generated during the preparation of
the Organellar maps, but with an additional biological replicate to give three biological
replicates for each cell line.
2.14 Mass spectrometry methods
2.14.1 In-solution digestion of proteins
Protein was precipitated by the addition of 5 volumes of ice-cold acetone, incubated at
−20 °C for 30 minutes and pelleted by centrifugation at 4 °C for 5 minutes at 10,000 g.
Precipitated protein was rinsed in ice-cold 80% acetone and re-pelleted as above. All
subsequent steps were performed at room temperature. Precipitated protein pellets
were air-dried for 5 minutes, resuspended in digestion buffer (50 mM Tris pH 8.1, 8 M
Urea, 1 mM DTT) and incubated for 15 minutes. Protein was alkylated by addition of
5 mM iodoacetamide for 20 minutes and then enzymatically digested by addition of
LysC (V1071, Promega; 1 mg per 50 mg of protein) for at least 3 hours. Digests were then
diluted four-fold with 50 mM Tris pH 8.1 before addition of Trypsin (1 mg per 50 mg of
protein) for an overnight incubation. The peptide mixtures were then acidified to 1%
(v/v) TFA in preparation for peptide purification and fractionation.
2.14 Mass spectrometry methods 63
2.14.2 Peptide purification and fractionation
Several different peptide fractionation and clean-up strategies were used in this study.
For most mass spectrometric experiments, peptides were purified and fractionated on
SDB-RPS (#66886-U, Sigma) StageTips (Kulak et al., 2014). Peptide mixtures in 1% TFA
were loaded onto activated StageTips and washed with Proteomics Wash Buffer (Pre-
omics) and then 0.2% (v/v) TFA. For single shot analysis, peptides were eluted with 60 μL
Buffer X (80% (v/v) acetonitrile, 5% (v/v) ammonium hydroxide). For triple-fractionation,
peptides were eluted successively using 20 μL SDB-RPSx1 (10 mM ammonium formate,
40% (v/v) acetonitrile, 0.5% (v/v) formic acid), then 20 μL SDB-RPSx2 (150 mM ammo-
nium formate, 60% (v/v) acetonitrile, 0.5% formic acid), then 30 μL Buffer X. For six-fold
fractionation, peptides were processed by strong cation exchange (SCX) on StageTips. A
detailed description of these methods has been published by Kulak et al. (2014). Alter-
natively, protein samples were separated by SDS-PAGE, gels were cut into 5–10 slices,
and proteins were digested with trypsin in-gel. Peptide extracts were then cleaned up
on C18-StageTips, before elution in Buffer B (80% (v/v) acetonitrile, 0.5% (v/v) acetic
acid), as described by Rappsilber et al. (2007). Cleaned peptides were dried almost to
completion in a centrifugal vacuum concentrator, and then volumes were adjusted to 10
μL with Buffer A* (0.1% (v/v) TFA, 2% (v/v) acetonitrile) and either immediately analysed
by mass spectrometry, or first stored at −20 °C.
The following techniques were applied for the various mass spectrometric analyses of
this study: single shot SDB-RPS: BioID samples and sensitive IPs; triple-fractionation
SDB-RPS: organellar maps analyses, membrane proteome analysis, vesicle prep analy-
ses (AP-4 knockouts); six-fraction SCX: whole cell lysate full proteome analysis; in-gel
digestion with multiple gel slice fractions: vesicle prep analyses (AP-4 knockdown, AP-4
knocksideways), conventional IPs.
2.14.3 Mass spectrometry
An overview of the mass spectrometric analyses performed during this study is provided
in Table 2.3. This includes information on quantification strategy, sample fractionation
approach, and MS instrument, for each analysis. Two different mass spectrometers were
used (Q-Exactive HF as detailed by Itzhak et al. (2017, 2016) and Q-Exactive as detailed
by Borner et al. (2014); Thermo Fisher Scientific), as indicated in Table 2.3.
64 Materials and Methods
Table 2.3 Overview of mass spectrometric analyses. Summary of the mass spectrometry (MS)
quantification strategies, fractionation approaches and MS instrument used in each proteomic
experiment. KD: knockdown; KS: knocksideways; KO: knockout.
Experiment MS quant strategy Fractionation MS instrument
Maps SILAC 3× SDB-RPS Exactive HF
Whole cell lysates SILAC 6× SCX Exactive HF
Membrane fractions LFQ 3× SDB-RPS Exactive HF
KD/KS vesicle preps SILAC 10 gel slices Exactive
KO vesicle preps SILAC 3× SDB-RPS Exactive HF
Sensitive IPs SILAC 1× SDB-RPS Exactive HF
Conventional IPs SILAC 2×5 gel slices Exactive
BioID LFQ 1× SDB-RPS Exactive HF
2.14.4 Processing of mass spectrometry data
Mass spectrometry raw files were processed in MaxQuant (Cox & Mann, 2008) version
1.5, using the human SwissProt canonical and isoform protein database, retrieved from
UniProt (www.uniprot.org). For SILAC experiments (vesicle fractions; Dynamic Organel-
lar Map subfractions; whole cell lysate analysis; TEPSIN-GFP immunoprecipitations)
multiplicity was set to 2, with Lys8 and Arg10 selected as heavy labels; Re-quantify was
enabled; minimum number of quantification events was set to 1. For label-free experi-
ments (membrane fractions; BioID) multiplicity was set to 1; LFQ was enabled, with LFQ
minimum ratio count set to 1. Membrane fractions were SILAC heavy labelled (Arg10,
Lys8). Matching between runs was enabled. Default parameters were used for all other
settings.
Mass spectrometry proteomics data associated with this project have been deposited to
the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via
the PRIDE partner repository (Vizcaíno et al., 2014) with the dataset identifier PXD010103.
2.15 Proteomic data analysis 65
2.15 Proteomic data analysis
All analyses were performed on the ‘protein groups’ file output from MaxQuant. Data
transformation, filtering and statistical analyses were performed in Perseus software
(Tyanova et al., 2016) version 1.5 and Microsoft Excel. Principal Component Analysis
(PCA) was performed in SIMCA 14 (Umetrics/MKS). Maps PCA plots (Figure 3.8B–D)
show projections along 1st and 3rd principal components, for optimum visualisation.
For all experiments identifications were first filtered by removing matches to the reverse
database, matches based on modified peptides only, and common contaminants (‘stan-
dard filtering’). Further experiment-specific filtering, data transformation and analyses
were performed as described below.
2.15.1 Dynamic Organellar Maps statistical analysis
To identify proteins with shifted subcellular localisation in response to AP-4 knockout, we
applied the rigorous statistical approach developed by Dr Georg Borner and described
in detail by Itzhak et al. (2017, 2016). We adapted the procedure to the experimental
design of the present study as follows. Organellar maps were made in duplicate, from
control, AP4B1 knockout, and AP4E1 knockout cells. Abundance distribution profiles
across all six maps were determined for 3,926 proteins. First, the profiles obtained in the
AP4B1 knockout and AP4E1 knockout cells were subtracted from the profiles obtained
in the cognate control map, protein by protein, to obtain 2× 2 sets of delta profiles
(Con_1-AP4B1_1, Con_2-AP4B1_2; Con_1-AP4E1_1, Con_2-AP4E1_2). For proteins that
do not shift, the delta profile should be close to zero. All delta profile sets were subjected
to a robust multivariate outlier test, implemented in Perseus software (Tyanova et al.,
2016), to identify proteins with delta profiles significantly different from experimental
scatter. The profile distance corresponds to a p-value reflecting how likely it is to observe
this deviation by chance, assuming no real change. For each protein, four such p-values
were hence obtained, two from AP4E1 knockout, and two from AP4B1 knockout. For
maximum stringency, we selected the least significant of these p-values as representative
of a protein’s shift. A shift of equal or greater significance was thus observed in all four
comparisons. We did not treat the four delta maps as completely independent though,
since both knockouts were compared to the same cognate control. Hence, as a very
conservative measure of movement, the selected p-value was only squared (instead of
being raised to the power of four), and then corrected for multiple hypothesis testing
66 Materials and Methods
using the Benjamini Hochberg Method. The negative log10 of the corrected p-value
corresponded to a protein’s movement (M) score.
Next, the reproducibility of observed delta profiles across replicates was determined
as the Pearson correlation (Δ map (Con_1-AP4B1_1) vs Δ map (Con_2-AP4B1_2); and
Δ map (Con_1-AP4E1_1) vs Δ map (Con_2-AP4E1_2)). For maximum stringency, we
chose the lower one of the two obtained correlations as representative of the protein’s
shift reproducibility, corresponding to its R score.
To control the false discovery rate (FDR), we then applied the same analysis to six pre-
viously published untreated wild-type HeLa maps, in which no genuine protein shifts
were expected (Itzhak et al., 2016). In this mock experiment, we designated two maps
as controls, two as ‘mock knockout 1’, and two as ‘mock knockout 2’. As above, we cal-
culated M and R scores from the lowest correlations and p-values of movement. The
estimated FDR at a given set of M and R score cut offs was then calculated as the number
of hits obtained with the mock experiment data, divided by the number of hits obtained
with the AP-4 maps data, scaled by the relative sizes of the datasets (which were almost
identical). At the chosen high stringency cut-offs (M score >4, R score >0.81), not a
single hit was obtained from the mock data. Hence, we estimate the FDR for the three
hits obtained from the AP-4 maps at <1%.
Finally, as an additional criterion, we also evaluated the similarity of identified shifts
across the two different knockouts (i.e. the correlation of Δ map (Con_1-AP4B1_1) vs
Δ map (Con_1-AP4E1_1); and Δ map (Con_2-AP4B1_2) vs Δ map (Con_2-AP4E1_2)).
All three hits showed a very high degree of shift correlation (>0.9) across the two AP-4
knockout lines, thus also passing the additional stringency filter.
2.15.2 Membrane fraction analysis
Relative protein levels in membrane fractions from AP4B1 knockout and AP4E1 knockout
HeLa cells (each in biological triplicate) were compared to those in membrane fractions
from wild-type HeLa cells (in biological triplicate) using LFQ intensity data. The primary
output was a list of identified proteins, and for each protein up to nine LFQ intensities
across the wild-type and AP-4 knockout samples. Following standard data filtering,
proteins were filtered to only leave those with nine LFQ intensities (no missing values
allowed), leaving 6653 proteins. LFQ intensities were then log-transformed and compar-
ison of knockout and wild-type membrane fractions performed with a two-tailed t-test.
2.15 Proteomic data analysis 67
A permutation-based (1000 permutations) estimated FDR of 0.05 and an S0 parameter
of 0.5 were set to define significance cut-offs (Perseus software).
2.15.3 Whole cell lysate analysis
Whole cell lysates from light-labelled AP4B1 knockout and AP4E1 knockout HeLa cells
were compared to lysates from heavy-labelled wild-type HeLa cells by SILAC quantifica-
tion, each in biological triplicate. The primary output was a list of identified proteins,
and for each protein up to six H/L (Heavy/Light) ratios of relative abundance (three
comparing AP4B1 knockout to wild-type and three comparing AP4E1 knockout to wild-
type). Following standard data filtering, proteins were further filtered to require at least
two H/L ratios for each knockout, leaving 6841 proteins. H/L ratios from each replicate
were then normalised to the median H/L ratio for that replicate, log-transformed, and
inverted to L/H so that a protein depleted from the whole cell lysate in the absence of
AP-4 had a negative ratio. A one-sample t-test (two-tailed) was applied to compare the
L/H ratios for each protein to zero (null hypothesis of no change between wild-type and
knockout). To control the FDR, an identical analysis of a mock experiment comparing
light- and heavy-labelled wild-type HeLa lysates was performed by Georg Borner (in
triplicate; no genuine changes were expected here). The FDR was given by the number of
hits observed in the mock experiment divided by the number of hits in the knockout ex-
periment. Using the cut-offs p ≤ 0.02 and a minimum absolute fold change (log2) of 0.45,
the estimated FDR was 25%. The t scores from the mock experiment were calculated
from only three datapoints per protein (whereas up to six were used for the knockout vs
control data), and hence were adjusted to emulate a 6 datapoint experiment. To this
end, we assumed that the observed standard deviations and means had been observed
from 6 datapoints, yielding much lower p values from the same t scores for the mock
data. This procedure thus likely overestimates the number of false positives at a given
cut-off, resulting in highly stringent FDR control.
2.15.4 Vesicle fraction analysis
Paired AP-4-depleted and control vesicle fractions were compared by SILAC quantifi-
cation. The primary output was a list of identified proteins, and for each protein up to
twelve H/L ratios of relative abundance, and the number of quantification events (H/L
ratio count) used to calculate each ratio. Following standard data filtering, proteins were
further filtered to require at least one H/L ratio count in all experiments. This excluded
AP4S1 due to it having a ratio count of 0 in one experiment, so the data for AP4S1 was
68 Materials and Methods
manually added back to the dataset (with one missing datapoint), giving a total of 2848
proteins quantified across all experiments. H/L ratios from each experiment were then
normalised to the median H/L ratio for that experiment. For experiments in which the
control cells were light labelled, normalised H/L ratios were inverted to L/H, and then all
ratios were log transformed for plotting, so that depletion from the vesicle fraction in the
absence of AP-4 was represented by a positive value. Pairwise comparison of replicate
experiments revealed the 10 minutes knocksideways treatment to be insufficient to
cause a shift in the steady state distribution of AP-4 cargo proteins, so this data was
excluded from further analysis. To identify proteins that were consistently lost from the
vesicle fraction from AP-4 depleted cells, the normalised log-transformed SILAC ratios
from the remaining nine experiments were scaled to unit variance and combined by
PCA.
2.15.5 TEPSIN-GFP immunoprecipitations
TEPSIN-GFP immunoprecipitations (conventional and sensitive) were compared to
control immunoprecipitations by SILAC quantification. The primary output for each
was a list of identified proteins, and for each protein one or more (up to three) H/L ratios
of relative abundance between the TEPSIN-GFP and control immunoprecipitations,
and the number of quantification events (H/L ratio count) used to calculate each ratio.
Following standard data filtering, unlabelled proteins were filtered out based on having
an H/L ratio < 0.2 in label-swapped replicates. Proteins were further filtered on a mini-
mum H/L ratio count of 1 in all replicates, leaving 585 proteins for the conventional IP
and 1128 proteins for the sensitive IP. H/L ratios from each replicate were normalised to
the median H/L ratio for that replicate. For replicates in which the control cells were
heavy labelled, normalised H/L ratios were inverted to L/H, and then all ratios were log
transformed for plotting, so that enrichment in the TEPSIN-GFP immunoprecipitation
was represented by positive values. The log SILAC ratios from two replicates were plotted
against each other to reveal proteins enriched in the TEPSIN-GFP immunoprecipitations.
2.15 Proteomic data analysis 69
2.15.6 BioID
For AP-4 BioID, relative protein levels were compared across samples using LFQ intensity
data. LFQ intensities from pulldowns from the control cell lines (HeLa, HeLa BirA*
and HeLa GFP-BirA*) were compressed from nine values (three cell lines in triplicate)
to three using control compression (Lambert et al., 2015). This creates a ‘worst-case
scenario’ control dataset where the three highest LFQ intensities are taken for each
protein. Pulldowns from each AP-4 BioID cell line (in triplicate) were then compared to
the compressed control dataset. Proteins were first filtered on LFQ intensities for valid
values in all three replicate pulldowns from the AP-4 BioID cell line, leaving between
approximately 3100–3700 proteins depending on the subunit. LFQ intensities were then
log-transformed, and missing data points were imputed from a normal distribution with
a downshift of 2.2 and a width of 0.3 standard deviations. Comparisons between control
and AP-4 BioID cell lines were performed with a two-tailed t-test. A permutation-based
(250 permutations) estimated FDR of 0.05 and an S0 parameter of 0.5 were set to define
significance cut-offs (Perseus software).
For the SERINC BioID, the SERINC1 and SERINC3 datasets were combined because
according to the Dynamic Organellar Maps both SERINCs have very similar subcellu-
lar localisations (Section 3.3.2). LFQ intensities from pulldowns from the control cell
lines (HeLa, HeLa BirA* and HeLa GFP-BirA*) were compressed from eighteen values
(three cell lines in triplicate for each of SERINC1 and SERINC3) to six using control
compression. Pulldowns from both SERINC BioID cell lines (each in triplicate; n = 6)
were then compared to the compressed control dataset. Proteins were first filtered on a
minimum MS/MS count of one in all six SERINC BioID pulldowns, leaving 2339 proteins.
Transformation, imputation and statistical analysis were then performed as described
for the AP-4 BioID.

Chapter 3
Proteomic investigations of adaptor
protein complex 4
3.1 Introduction
Proteomics, the large scale identification and analysis of proteins, plays a crucial role
in our quest to understand the complex biology of cells. Genomics has revolutionised
our understanding of biological systems, but genomes are (relatively) static information
storage systems. In contrast, proteomes are highly dynamic. The proteome of a cell is the
functional output of its genome and will define its identity and behaviour. For the biology
of the cell it matters what proteins are expressed and the quantities of these proteins,
when they are expressed, where they are localised, how they are modified with post-
translational modifications (PTMs) and who (i.e. which other proteins) they interact
with. All this may lead a cell biology student to ask, "Why must it be so complicated?".
While nothing will come of dwelling on this last question for too long, mass spectrometry-
based proteomics has now come of age and allows us to address the preceding points.
From the proteome of a cell we can zoom out to a whole organism proteome or zoom in
to an organelle proteome, or even further to analyse the proteins in a specific protein
complex.
Members of the Robinson Lab have applied proteomic methods to study vesicle traf-
ficking, including interaction-based methods involving affinity purification and mass
spectrometry (AP-MS) and comparative proteomic analysis of the clathrin-coated vesicle
(CCV) fraction. The aim of this study was to apply similar techniques to investigate the
function of AP-4, with a particular focus on organellar and spatial proteomic methods.
72 Proteomic investigations of adaptor protein complex 4
3.1.1 Mass spectrometry-based proteomics
The basic level of information revealed by mass spectrometry (MS)-based proteomics is
protein identification. In the most commonly used ‘bottom-up shotgun’ proteomics
workflow (reviewed in Aebersold & Mann, 2016; Zhang et al., 2013) protein identifications
are inferred indirectly from identified peptides, which are released from mixtures of
proteins via proteolysis (typically tryptic digest). The main steps are: 1) Isolation of
proteins from cells or tissues; 2) Digestion of the protein mixture to yield peptides; 3)
Separation of the peptide mixture by liquid chromatography (LC); 4) Analysis of the
peptides by Tandem MS (MS/MS); 5) Identification of peptides via analysis of mass
spectra; 6) Mapping of peptide identifications to proteins. In Tandem MS peptide
samples are ionised to generate a mixture of precursor ions which are selected based on
their specific mass-to-charge ratio (m/z) in a first stage of MS (MS1 scan). The selected
precursor ions are fragmented, typically by collision-induced dissociation (CID), to
generate product ions which are detected in a second stage of MS (MS2 scan). The MS2
scan is used to identify the peptide while MS1 ion intensities are used for quantification.
This process can be extended to further rounds of MS, for example the MS3 workflow
includes an additional ion fragmentation step after MS2 to release reporter ions from
isobaric chemical tags (e.g. TMT labels), which are quantified in a third stage of MS. An
alternative strategy to bottom-up shotgun proteomics is ‘top-down’ proteomics, which
is used to characterise intact proteins (Kelleher et al., 1999). This offers advantages for
specific applications such as PTM positioning and for distinguishing between protein
isoforms. However, technical challenges have prevented the routine use of top-down
proteomic approaches in cell biology.
The next level of information yielded from MS-based proteomics is quantification. This
comes in two main flavours - relative and absolute (reviewed in Chahrour et al., 2015).
Typically when people talk about quantitative proteomics they are referring to relative
quantification - comparing the amount of a protein in one sample to the next or the
relative abundance of different proteins within a sample. However, absolute protein
quantification can be achieved through the use of stable isotope labelled standards of
known concentration. In this study we were concerned with relative quantification, so
absolute quantification approaches are not discussed further here. Relative quantifi-
cation breaks down into two distinct strategies - stable isotope labelling and label-free
approaches. Stable isotope labelling makes use of heavy mass tags to distinguish pep-
tides originating from specific samples via a characteristic shift in the mass spectra.
The tags can be introduced either metabolically to cells (e.g. SILAC, Ong et al., 2002) or
3.1 Introduction 73
chemically to protein or peptide samples (e.g. ICAT, Gygi et al., 1999; iTRAQ, Ross et al.,
2004; TMT, Thompson et al., 2003). In our analyses we made use of the SILAC method, as
well as label-free quantification, so this is discussed in more detail below. Once labelled,
samples can be mixed and processed as a single sample, reducing inter-sample variation
introduced in sample preparation and measurement steps. Metabolic labelling has an
advantage over chemical labelling in this regard because the samples can be mixed
earlier in the experimental procedures. Chemical labels are added after protein purifica-
tion (ICAT) or at the peptide level (iTRAQ/TMT). However, metabolic labelling methods
are most applicable to cell lines that can be passaged in controlled growth medium
containing the heavy isotopic label. It is not possible, for example, to metabolically label
primary tissue samples from patients. Either chemical labelling or label-free approaches
could be used instead. In the label-free approach, samples are measured individually
and compared either by the MS ion intensities (Wang et al., 2003a) or the number of
acquired spectra (spectral counting; Liu et al., 2004). Although label-free methods offer
less accurate quantification than stable isotope labelling, they have a number of advan-
tages including simplicity, low cost, universal applicability, and the ability to compare an
infinite number of samples (stable isotope labelling methods are limited by the number
of available isotopic variants). Where we have employed label-free quantification we
used a label-free quantification algorithm called MaxLFQ (Cox et al., 2014; available in
MaxQuant, Cox & Mann, 2008), which is an MS ion intensity-based method.
The SILAC method for relative protein quantification
The SILAC (stable isotope labelling by amino acids in cell culture) method (Ong et al.,
2002) makes use of stable isotope containing amino acids (usually lysine and/or arginine)
to label proteins in live cells (Figure 3.1). In a typical experiment, control and treated cells
are differentially grown in light (containing normal amino acids) or heavy (containing
13C6
15N2-Lysine and
13C6
15N4-Arginine; mass differences of 8 and 10 D, respectively)
media for a sufficient number of passages to allow as close to 100% label incorporation
as possible. In our lab we have found that for HeLa cells seven days of culture in heavy
medium is sufficient to achieve a mean incorporation efficiency of 97% (Borner et al.,
2012). It is also possible to perform three-way comparisons by using medium isotope-
labelled amino acids (e.g. Lys-4 and Arg-6), but in our analyses we just performed
two-way comparisons with light and heavy amino acids. Cells are harvested and mixed
as early as possible in the experimental procedure, either prior to protein purification or
before the tryptic digest is performed. Trypsin cleaves proteins after arginine and lysine
74 Proteomic investigations of adaptor protein complex 4
Fig. 3.1 The SILAC method for relative protein quantification. Cells are grown in culture
medium containing heavy isotope-labelled amino acids (typically Lys-8 and Arg-10) or the
equivalent unlabelled ‘light’ amino acids, which are incorporated into newly synthesised pro-
teins. Samples can be mixed early in experimental procedures, either before cell lysis (equal
cell numbers) or before proteolytic digest (equal protein quantities). Mixtures of heavy and
light peptides are analysed by liquid chromatography coupled to tandem mass spectrometry
(LC-MS/MS). Paired heavy and light peptides are distinguished by a characteristic mass shift
and the ratio of their intensities provides a measure of relative peptide abundance, from which
relative protein abundances are inferred.
residues, so peptides resulting from a tryptic digest will contain a single labelled amino
acid, resulting in a predictable mass difference and thereby simplifying the subsequent
data analysis. As there are no chemical differences between the isotope labelled and
natural amino acids, heavy and light-labelled peptides will co-elute from the LC column
and so be analysed together in the mass spectrometer. Thus, the peptide peaks of the
heavy and light samples can be very accurately quantified relative to each other to give
the heavy/light (H/L) ratio of a protein between the samples. As the chemistry of the
heavy and light amino acids is the same, there is also not likely to be any biological impact
of growing cells in heavy versus light medium. Nonetheless, performing label-swap
experiments is prudent to control for incomplete labelling efficiency or contamination
of the sample with unlabelled protein (e.g. from the person performing the sample
preparation).
3.1.2 Organellar proteomics and vesicle trafficking
For cell biologists with an interest in membrane trafficking, perhaps the most interesting
question to ask using proteomics is: "Where in the cell are the proteins?". There has been
a natural evolution from the use of MS to study discrete protein complexes to charac-
terise the protein complement of entire organelles (reviewed in Andersen & Mann, 2006).
In the late 1990s researchers began to combine two-dimensional polyacrylamide gel
3.1 Introduction 75
electrophoresis (2D-PAGE) with MS-based protein identification to define the proteome
of various organelles including mitochondria (Rabilloud et al., 1998), chloroplasts (Peltier
et al., 2000) and the Golgi complex (Taylor et al., 2000). Since then, with advances in MS
technology, the number of available organelle proteomes has expanded and they have
become ever more comprehensive and accurate. However, with biochemical isolation
or enrichment of organelles there comes the problem of co-enriched contaminating
proteins. Therefore, it has been important to develop ways of distinguishing these from
genuine components of an organelle. The first attempts to do this were subtractive. For
example, integral membrane proteins of the nuclear envelope were defined by subtract-
ing proteins present in cofractionating organelles from the list of proteins identified in
the nuclear envelope fraction (Schirmer et al., 2003). Here, comparative proteomics
making use of relative quantification methods (e.g. SILAC) came to the forefront, as it
allowed for more direct comparison of related samples. Comparative approaches com-
bined with a functional property of a given organelle also proved useful for increasing the
specificity of the proteomic characterisation of organelles. This approach was applied by
Foster and colleagues to define protein components of cholesterol-dependent lipid rafts
by using SILAC-based quantitative MS to compare lipid rafts isolated from cells with and
without cholesterol depletion (Foster et al., 2003). However, not all cellular structures
are amenable to the levels of biochemical purification or enrichment required for these
comparative approaches. To tackle this problem Andersen, Wilkinson and colleagues
developed a method called protein correlation profiling (PCP) and used it to define the
proteome of the centrosome (Andersen et al., 2003). They took the five sucrose gradi-
ent fractions from the centrosome preparation and analysed them by quantitative MS,
resulting in peptide elution profiles across the fractions. Known centrosomal proteins
had a characteristic profile and this could be used to distinguish novel centrosomal
proteins from contaminating proteins. They demonstrated the accuracy of this method
by confirming the localisation of their novel centrosomal proteins using GFP-tagging
and microscopy.
Organellar proteomic approaches such as these have been adopted by the vesicle traf-
ficking community to characterise the protein components of vesicles. To date, CCVs
are probably the best studied vesicle population. Early studies used MS to characterise
the proteome of CCVs isolated from rat brain (Blondeau et al., 2004) or liver (Girard et al.,
2005). These studies reproducibly identified 209 and 326 proteins in the CCV fractions,
respectively, including a high proportion of known CCV components. However, while
care was taken to make the CCV preparations as pure as possible, they did not incorpo-
rate any means of distinguishing genuine CCV proteins from contaminating proteins.
76 Proteomic investigations of adaptor protein complex 4
In particular CCV preparations suffer from ribosomal contamination, and although they
are highly enriched in CCVs they do also contain other types of vesicles (e.g. AP-4 vesi-
cles), as well as large cytosolic protein complexes. To overcome this limitation, Dr Georg
Borner, working in the Robinson Lab, carried out a comparative proteomic analysis of
CCV fractions prepared from untreated cells and ‘mock’ CCV fractions prepared from
clathrin-depleted cells (Borner et al., 2006). This revealed 63 bona fide CCV proteins
which depended on clathrin for their presence in the CCV fraction. Dr Borner later
extended this comparative approach to the multivariate proteomic profiling of CCVs
(Borner et al., 2012). Modified CCV fractions were prepared under different experimen-
tal conditions and compared with CCV fractions prepared with a standard protocol
(to give a total of 10 binary comparisons). The modified conditions were: (i) siRNA-
mediated depletion of clathrin; (ii) siRNA-mediated depletion of auxilin (involved with
CCV uncoating); (iii) an older CCV preparation protocol with lower enrichment of all
CCV proteins. The 10 resulting datasets were then combined using principal component
analysis (PCA), which is a way of reducing the dimensionality of multi-dimensional
datasets. This revealed distinct clusters for clathrin, AP-1, and AP-2, clearly separated
from most of the other proteins. Known CCV proteins were highly enriched in these
clusters and thus uncharacterised proteins clustering with clathrin, AP-1 or AP-2 were
considered candidate CCV proteins. Automated cluster analysis identified 136 candidate
CCV proteins, including 93% of established CCV coat proteins and 25 predicted novel
CCV coat proteins. Additionally, the separate clustering of other vesicle proteins outside
of the CCV cluster revealed further functional insights including the identification of
novel constituents of the AP-4 and retromer coats.
Comparative approaches to determining vesicle composition are powerful, but rely
on the ability to perform manipulations such as siRNA-mediated knockdowns. As an
alternative approach, which does not reply on chemical or genetic manipulations, Dr
Borner developed ‘fractionation profiling’ of the vesicle-enriched fraction (Borner et al.,
2014). This method builds on the principles of PCP described above. The CCV-enriched
fraction is further fractionated into three sub-fractions and SILAC-based quantitative
proteomics is used to generate protein abundance distribution profiles across these sub-
fractions. More than 3,500 proteins were profiled across the three fractions and PCA was
used to cluster proteins according to their abundance distribution profiles, resulting in
clear segregation of CCV from non-CCV proteins. In addition to distinguishing genuine
CCV proteins from contaminants, this method had the added benefit of clustering other
profiled proteins based on their existence in stable protein complexes, for example AP
complex subunits formed tight clusters. Thus, the profiling could be used to predict
3.1 Introduction 77
the existence of novel protein complexes, including the WDR11/C17orf75/FAM91A1
complex which has since been experimentally confirmed by Dr Paloma Navarro Negredo
(Navarro Negredo et al., 2018).
3.1.3 Spatial proteomics
In spatial proteomics the focus shifts from the in-depth analysis of the constituents of a
single organelle to a global overview of protein localisation in the whole cell (reviewed
in Drissi et al., 2013). This is important because many proteins do not reside in a single
organelle but instead have a mixed steady state distribution between multiple parts
of the cell, often with dynamic movement between them. Spatial proteomics takes
the principles of subcellular fractionation and abundance profiling, which have been
successfully applied to assign proteins to a specific organelle (e.g. Andersen et al., 2003;
Borner et al., 2014), and applies them to the whole cell. The first cell maps were published
in 2006; Dunkley and colleagues in Kathryn Lilley’s group used a method termed ‘LOPIT’
(localisation of organelle proteins by isotope tagging; Dunkley et al., 2004) to generate
an Arabidopsis organelle map (Dunkley et al., 2006), while Foster and colleagues in
Matthias Mann’s group applied their PCP method to produce an organelle map from
mouse liver (Foster et al., 2006). Both studies applied the same basic approach: 1) Gentle
lysis of the cell to preserve organelle integrity; 2) Separation of intact cell components
using density or velocity gradient centrifugation; 3) Measurement of protein abundance
across subcellular fractions by quantitative MS; 4) Comparison of protein abundance
profiles to those of known marker proteins. In the LOPIT method a combination of PCA
and machine learning-based cluster analysis was used to assign >500 proteins to the
endoplasmic reticulum (ER), Golgi apparatus, vacuolar membrane, plasma membrane,
or mitochondria/plastids. In the PCP method the statistical measurement of profile
similarity assigned >1,400 proteins to one of ten different subcellular locations.
Static organelle maps hold a great deal of information about the steady state distribu-
tions of proteins within the cell and so provide important clues about protein function.
However, for the membrane trafficking community the holy grail would be to map the
movement of proteins between different compartments of the cell in response to envi-
ronmental or genetic perturbations. Experiments carried out by Boisvert and colleagues
demonstrated the utility of spatial proteomics for the detection of changes in protein
localisation (Boisvert et al., 2010; Boisvert & Lamond, 2010). In these studies cytoplasmic,
nuclear and nucleolar fractions were generated from differentially metabolically-labelled
cells (SILAC light/medium/heavy) and then recombined to recreate a ‘whole cell extract’
78 Proteomic investigations of adaptor protein complex 4
where the cytoplasm, nucleus and nucleolus are derived from cells with different isotope
labels. MS measurement was used to calculate the ratio of light:medium:heavy isotopic
forms of identified peptides, revealing the relative distribution of >2,000 proteins be-
tween the cytoplasm, nucleus and nucleolus. The authors applied this method to study
the effect of the DNA damage inducing agent etoposide on protein localisation (Boisvert
et al., 2010) and then to look for differences in the DNA damage response between p53-
competent and p53-deficient cells (Boisvert & Lamond, 2010). However, although this
method could be applied to look at the relative distribution of proteins between other
compartments that are amenable to fractionation, it is limited in the number of com-
partments that can be profiled in a single experiment because of the limited number of
SILAC isotope variants. To really analyse protein localisation changes on a global cellular
level requires a method that combines the high spatial resolution of the aforementioned
cell maps with the robustness of this dynamic approach. Serendipitously for the timing
of my PhD such a method was recently developed by Dr Georg Borner and Dr Daniel
Itzhak at the Max Planck Institute for Biochemistry, termed ‘Dynamic Organellar Maps’
(Itzhak et al., 2016). This is described below in Section 3.3.
3.1.4 Proteomics to study AP-4-mediated vesicle trafficking
The aim of this project was to identify cargo and machinery of the AP-4 trafficking
pathway. Progress in the biochemical characterisation of AP-4 vesicles has been impeded
by the very low expression level of AP-4 in tissue culture cells (Hirst et al., 2013b). This
has exacerbated the already significant challenges associated with detecting transient
adaptor protein interactions using traditional affinity purification-based techniques. To
overcome these difficulties we decided to combine highly sensitive MS with unbiased
organellar and spatial proteomic approaches with which we could assay the localisation
of endogenous proteins in AP-4-deficient cells. We applied the Dynamic Organellar
Maps method to detect mislocalisation of proteins in AP-4 knockout cells (see Section 3.2
for generation of the knockout cell lines and Section 3.3 for Dynamic Organellar Maps).
We also applied an orthogonal organellar proteomics method - comparative analysis
of vesicle-enriched fractions prepared from control and AP-4 deficient cells (Section
3.4). These two approaches identified candidate AP-4 cargo and machinery which
were validated using proximity-dependent biotin identification (BioID) and sensitive
immunoprecipitation techniques (Sections 3.5 and 3.6). Finally, further proteomic
investigations of membrane fractions and whole cell lysates from AP-4 knockout cells
3.2 AP-4 knockout HeLa cells 79
revealed further insight into the relationship between AP-4 and our newly identified
AP-4 cargo and machinery (Section 3.7).
Although AP-4 deficiency is a disease of the nervous system, we choose to conduct these
proteomic screens in HeLa cells, rather than in a neuronal cell type, for three key reasons:
1. AP-4 is ubiquitously expressed in human tissues and so we predict it to have a
ubiquitous function in all cells, not just in neurons. We believe that understanding
this ubiquitous role will be important for understanding the pathomechanisms of
AP-4 deficiency.
2. HeLa cells are easy to grow in the large numbers required for biochemical analy-
sis. Particularly the preparation of the vesicle-enriched fraction requires a large
amount of starting material because it makes up just ∼0.1% of total cellular protein,
and AP-4 vesicles are a minor component of this fraction.
3. HeLa cells are amenable to genetic manipulations such as transfection, siRNA-
mediated knockdown, and CRISPR/Cas9-mediated gene editing.
3.2 AP-4 knockout HeLa cells
The effect of AP-4 loss of function was previously studied in the lab using siRNA-mediated
knockdown or the knocksideways system (data discussed in Section 3.4). Experience in
our group of using siRNA to study the roles of other AP complexes has shown that robust
knockdowns (e.g. by around 90% at the protein level) are required to reveal phenotypes,
because any residual AP complex subunit is incorporated efficiently into vesicles (e.g. as
observed for AP-1 complex subunit γ-1 in Hirst et al., 2012). Therefore, we decided that
complete knockout of AP-4 would be preferential for studying the proteomic effects of
AP-4 deficiency. The only AP-4 knockout cells that were available to us at the start of this
project were fibroblasts from patients with AP-4 deficiency. These are difficult to SILAC
label and to grow on the scale required for our proteomic assays, and it is also hard to
have truely matched control cell lines because of individual variability and parameters
such as age and sex. Thus, we decided to make use of CRISPR/Cas9 technology to knock
out AP-4 in HeLa cells. AP-4 is an obligate tetramer and subsequently loss of any one
subunit renders the entire complex non-functional (Borner et al., 2012; Hardies et al.,
2015; Hirst et al., 2013b). Therefore, a criterion for a bona fide AP-4 cargo or accessory
protein is that it is affected in cell lines deficient in different AP-4 subunits. For this
reason, we decided to make AP4B1 and AP4E1 knockout HeLa cell lines.
80 Proteomic investigations of adaptor protein complex 4
3.2.1 The Double Nickase CRISPR/Cas9 system
CRISPR/Cas9-mediated gene editing is now a widely used tool in cell biology and has
become a routine method for studying the effects of protein loss-of-function in cells.
CRISPR/Cas9 systems make use of an RNA-guided CRISPR-associated endonuclease
Cas9 to introduce double-stranded breaks at specific locations in DNA. When these
breaks are repaired by error-prone non-homologous end-joining (NHEJ) there is a
propensity for the introduction of insertions or deletions (indels) in the DNA sequence,
which may cause frameshift of the open reading frame and hence knockout of the gene
(reviewed in Sander & Joung, 2014). The most commonly used CRISPR/Cas9 system
for gene editing is the Type II system from Streptococcus pyogenes which has been engi-
neered for use in human cells (Cong et al., 2013; Mali et al., 2013). The Cas9 nuclease
from S. pyogenes has been human codon-optimised and fused to a nuclear localisation
signal for efficient targeting to the nucleus. Formation of DNA cleavage-competent Cas9
complex also requires CRISPR RNA (crRNA) that contains the DNA-targeting sequence
(spacer) and a trans-activating crRNA (tracrRNA). The crRNA and tracrRNA can either
be expressed individually or as a single chimera termed a short guide RNA (sgRNA). The
Cas9 complex requires sequence complementarity between the spacer in the crRNA
(or guide) and the ‘protospacer’ target sequence in the DNA, and the presence of a
protospacer adjacent motif (PAM) at the 3’ end of the protospacer, for target recognition
and subsequent DNA cleavage to occur (Jinek et al., 2012). The S. pyogenes Cas9 requires
an NGG PAM sequence, where N is any nucleotide.
One downside of the CRISPR/Cas9 system is its potential for off-target effects (much
like siRNA-mediated targeting of mRNA), by inducing double-stranded DNA breaks at
other genomic locations with sequence complementarity with the sgRNA. Studies that
have assessed the activity of Cas9 at potential off-target genomic locations have found
mutations can occur at sites with up to five mismatches with the sgRNA (e.g. Fu et al.,
2013 and reviewed in Sander & Joung, 2014). While this has not prevented wide-scale
adoption of CRISPR/Cas9 methods, care needs to be taken to control for possible off-
target effects. Common approaches are the use of multiple cell lines generated with
independent sgRNAs and rescue experiments. We were particularly concerned about
off-target effects because of our global proteomic approaches - if we identified many
proteomic differences between our control and AP-4 knockout cells, how would we
know which were on-target versus off-target? The large amount of MS time required to
analyse each cell line (e.g. approximately two days per map) made the use of multiple
clones and rescue lines unfeasible. Therefore, we decided to employ a modification of
3.2 AP-4 knockout HeLa cells 81
Fig. 3.2 Double Nickase CRISPR-Cas9 system. Schematic of the use of paired short guide RNAs
(sgRNAs) in combination with the S. pyogenes Cas9 nickase mutant (SpCas9n) to introduce a
compound double-stranded break into DNA. The D10A mutation in SpCas9n causes it to only
cut the DNA strand that is complementary to the sgRNA guide sequence. Cuts are made ∼3
basepairs (bp) upstream of the NGG protospacer adjacent motif (PAM). Paired 20 nucleotide (nt)
DNA target sites on opposite strands of DNA should be separated by 4 to 100 bp. DNA cleavage
activity at off-target sites of single sgRNAs should only result in single-stranded DNA ‘nicks’.
the CRISPR/Cas9 system termed ‘Double Nickase’ CRISPR (Figure 3.2; Mali et al., 2013;
Ran et al., 2013). This uses a mutant form of S. pyogenes Cas9 with a D10A mutation (Sp-
Cas9n) which can only introduce a single-stranded DNA break into the complementary
strand of the target DNA. When the nickase Cas9 is used in parallel to paired sgRNAs
complementary to opposite strands of the DNA target site (separated by 4 to 100 bp),
a compound double-stranded break is generated (and subsequent indels). Off-target
sites should only suffer single-stranded DNA ’nicks’. As single-strand break repair (SSBR)
pathways have much higher fidelity than NHEJ, indels are unlikely to be introduced at
off-target sites.
82 Proteomic investigations of adaptor protein complex 4
3.2.2 Guide RNA design and cloning
The first step in the generation of AP4B1 and AP4E1 knockout cells was the design of
paired sgRNAs to target the genes. For this we used the CRISPR Design Tool developed
by the Zhang group (http://crispr.mit.edu/; Hsu et al., 2013). Their algorithm scans
possible guides against the target genome and provides a score that indicates the on-
target activity of the guide and a list of off-target sites (see Tables C.1 and C.2 in Appendix
C for scores). The likelihood of a CRISPR-induced indel resulting in complete disruption
of gene function is enhanced when the N terminus of the coding region is targeted.
Therefore, guide pairs were selected based on the following criteria: (i) presence in a
coding exon present in all well-supported predicted transcripts of the gene (Ensembl
Transcript Support Level 1); (ii) position of exon as close as possible to the N terminus
of the protein; (iii) guide pair score. Guide pairs were chosen to target exons 1, 2 and
3 of AP4B1 transcript ENST00000369569 (Figure 3.3A) and exons 6, 7 and 11 of AP4E1
transcript ENST00000261842 (Figure 3.3B). All pairs were scored as high quality and had
no off-target sites.
Guides were cloned into the pX335 vector (Figure 3.3C) using a BbsI site that allows
scarless introduction of the guide before the sgRNA scaffold sequence. The U6 promoter
(which drives sgRNA expression) requires a G nucleotide at the start of the transcribed
RNA, so, unless guide sequences already began with G, an additional G was added to
the 5’ end of the guides. Successful cloning was confirmed by diagnostic digest and
sequencing (data not shown). pX335 is a bicistronic vector for co-expression of human
codon-optimised spCas9n and the sgRNA.
3.2.3 Generation of AP-4 knockout HeLa cells
To generate AP4B1 and AP4E1 knockouts, HeLa cells were co-transfected with two pX335
plasmids containing paired sgRNAs and a puromycin resistance plasmid in the ratio 2:2:1.
This ratio was chosen to ensure that cells with the puromycin resistance gene were likely
to also bear the two pX335 plasmids. Forty-eight hours after transfection the cells were
placed under puromycin selection for four days to kill off untransfected cells. Following
this selection the cells were maintained in full growth medium without puromycin until
sufficient numbers were available for analysis. Cells were then harvested and whole
cell lysates analysed by Western blot with antibodies against AP4B1 and AP4E1 (Figure
3.4A). There was a substantial reduction in the level of AP4B1 in all three AP4B1 CRISPR
mixed populations. This was accompanied by a small decrease in AP4E1 levels. This
3.2 AP-4 knockout HeLa cells 83
Fig. 3.3 Guide RNA design for knockout of AP4B1 and AP4E1. (A) Paired sgRNA target se-
quences in exons 1, 2 and 3 of the AP4B1 gene are highlighted in red. PAM sequences at the 3’
ends of the target sequences are underlined. The entire gene spans 10 kb. (B) Paired sgRNA
target sequences in exons 6, 7 and 11 of the AP4E1 gene, as in A. The entire gene spans 97 kb.
(C) Schematic of the pX335 vector. Bicistronic vector for expression of U6 promoter (green)
driven chimeric sgRNA and CBh promoter (yellow) driven human codon-optimised SpCas9n
(red), a mutant form of S. pyogenes Cas9 (D10A) which can only make single-strand breaks in
DNA (’nickase’). The chimeric sgRNA consists of a 20 nucleotide (nt) guide sequence (blue)
and scaffold (purple). NLS: nuclear localisation signal; bGHpA: transcription terminator and
polyadenylation signal.
84 Proteomic investigations of adaptor protein complex 4
is expected because studies of AP-4-deficient patient fibroblasts show that loss of one
AP-4 subunit leads to reduced expression levels of the other AP-4 subunits (Borner et al.,
2012; Hardies et al., 2015; Hirst et al., 2013b). AP4E1 appeared greatly reduced in the
AP4E1 exon 6 population and was reduced to a lesser extent in the exon 7 population. A
decrease in AP4E1 was not detected in the exon 11 population, so this population was
not analysed further.
The AP4B1 populations and AP4E1 exon 6 and 7 populations were single cell cloned
and initially screened by immunofluorescence microscopy (data not shown). There is
no antibody available that works for AP4B1 for immunofluorescence, but an antibody
against AP4E1 can be used as a proxy for all other AP-4 subunits because they form an
obligate complex. Cells were co-labelled with antibodies against AP4E1 and the AP-4
accessory protein TEPSIN (see Figure 3.5B for an example of AP4E1 and TEPSIN labelling
in wild-type HeLa cells). TEPSIN requires AP-4 for its membrane recruitment (Borner
et al., 2012), so loss of TEPSIN from the TGN region is indicative of loss of AP-4 function.
Clones that appeared negative for AP4E1- and TEPSIN-positive puncta in the perinuclear
region were further screened by Western blot with antibodies against AP4B1 or AP4E1
(Figure 3.4B and C). This revealed that many of the cell lines that appeared negative for
AP-4 complex labelling in the immunofluorescence screen still expressed either reduced
levels of the targeted AP-4 subunit (e.g. AP4E1 clone x6A2), or a truncated protein (e.g.
AP4B1 clone x1A2). However, several clones did appear to be negative for expression
of the targeted AP-4 subunit (marked with red boxes), including AP4B1 clone x2A3 and
AP4E1 clone x6C3 which were selected for further validation and characterisation.
AP4B1 clone x2A3 and AP4E1 clone x6C3 were confirmed to be knockouts at the protein
level by Western blot (Figure 3.5A) and by immunofluorescence microscopy (Figure
3.5B). They were further validated by Sanger DNA sequencing across the sgRNA target
sites (Figure 3.5C and D). Genomic DNA was isolated from the cells and ∼500 bp regions
surrounding each target site were amplified by PCR. PCR products were blunt-end cloned
to enable sequencing of individual PCR products. For AP4B1 clone x2A3 the sequencing
of 24 PCR products revealed five different mutant alleles, all of which result in frameshift
and the introduction of a premature termination codon (PTC) in exon 2. For AP4E1 clone
x6C3 the sequencing of 17 PCR products revealed three different mutant alleles, which
each result in frameshift and the introduction of a PTC in exon 6. No wild-type alleles
were recovered from either clone. AP4B1 and AP4E1 are located on chromosomes 1 and
15, respectively. We have not karyotyped our HeLa cell line, but HeLa cells are well-known
for their chromosomal instability (published karyotypes suggest a variable total number
of chromosomes, typically around 80; Landry et al., 2013), so five alleles for AP4E1 is not
3.2 AP-4 knockout HeLa cells 85
Fig. 3.4 Generation of AP-4 knockout HeLa cells. A Double Nickase CRISPR/Cas9 system was
used to target AP4B1 exon 1, 2 or 3 or AP4E1 exon 6, 7 or 11 in HeLa cells. Mixed populations
of cells were screened for loss of protein and then single cell cloned. (A) Mixed populations of
AP4B1 and AP4E1 CRISPR-targeted cells were analysed by Western blot with antibodies against
AP4B1 and AP4E1. AP4B1 was depleted in all three AP4B1 CRISPR populations, while AP4E1
was depleted only by the AP4E1 sgRNAs targeting exons 6 and 7. Wild-type (WT) HeLa cells
were used as a positive control for expression and an antibody against α-Tubulin was used as
a loading control. (B) Selected single cell AP4B1 clones were analysed by Western blot with an
antibody against AP4B1. Wild-type HeLa cell lysates were used as a positive control and CRISPR
clone lanes were overloaded with protein relative to WT lanes. Clones considered to be potential
knockouts in this screen are highlighted in red boxes. (C) Selected single cell AP4E1 clones were
analysed as in B, but with an antibody against AP4E1.
86 Proteomic investigations of adaptor protein complex 4
unexpected. The presence of PTCs early in the mutant transcripts predicts induction
of nonsense-mediated decay (NMD) of the transcribed mRNAs (Nagy & Maquat, 1998).
This was not tested at the mRNA level, but no truncated protein products were detected
in either cell line. Regardless, if they were present they would be so severely truncated
(the largest predicted protein for AP4B1 is 155 amino acids and for AP4E1 is 147 amino
acids) they would be highly unlikely to be functional. Based on this, AP4B1 clone x2A3
and AP4E1 clone x6C3 were used as cell models for AP-4 deficiency in our proteomic
studies.
3.2 AP-4 knockout HeLa cells 87
Fig. 3.5 Characterisation of clonal AP-4 knockout (KO) cell lines. (Full caption on following
page.)
88 Proteomic investigations of adaptor protein complex 4
Fig. 3.5 Characterisation of clonal AP-4 knockout (KO) cell lines. (A) Western blot of whole-
cell lysates from wild-type, AP4B1 KO clone x2A3 and AP4E1 KO clone x6C3, with antibodies
against AP4B1 and AP4E1. An antibody against α-Tubulin was used as a loading control. (B)
Widefield immunofluorescence imaging of AP4E1 and TEPSIN in wild-type and AP-4 KO HeLa
cells. Note that the AP4E1 antibody has some background staining; the signal that is specific to
AP4E1 is the cluster of puncta in the perinuclear region of the cell (inset). AP-4 KO resulted in the
concomitant loss of TEPSIN puncta from this region, indicative of loss of AP-4 function. Scale
bar: 20 μm. (C) Multiple sequence alignment (Jalview) of wild-type AP4B1 exon 2 and the mutant
sequences recovered from clone x2A3. Allele 1 (1 bp deletion) was recovered five times, allele 2
(28 bp deletion) four times, allele 3 (29 bp deletion) six times, allele 4 (26 bp deletion plus 72 bp
insertion) four times and allele 5 (13 bp deletion plus 102 bp insertion) five times. The alignment
shows only the area of exon 2 surrounding the mutations (the rest of exon 2 was the same in the
mutants as the wild-type sequence) and parts of the insertions in alleles 4 and 5 are hidden from
view. The nucleotides are numbered based on the longest allele (allele 5), starting from 1 for
the first nucleotide of exon 2. All mutant alleles result in frameshift and the introduction of a
premature termination codon (PTC) in exon 2. (D) Multiple sequence alignment of wild-type
AP4E1 exon 6 and the mutant sequences recovered from AP4E1 KO clone x6C3. Allele 1 (34 bp
deletion) was recovered seven times, allele 2 (28 bp deletion) six times and allele 3 (compound 53
bp deletion) four times. The alignment shows only the area of exon 6 surrounding the mutations
(the remainder of the exon 6 sequence was identical in the mutants and the wild-type). The
nucleotides are numbered based on the wild-type allele, starting from 1 for the first nucleotide
of exon 6. All mutant alleles result in frameshift and the introduction of a PTC in exon 6.
3.3 Dynamic Organellar Maps of AP-4 knockout cells 89
3.3 Dynamic Organellar Maps of AP-4 knockout cells
Dynamic Organellar Maps is a method that provides protein localisation information at
the whole proteome level (Itzhak et al., 2016). It works on the basic principle of whole cell
fractionation via centrifugation and makes use of the differential fractionation properties
of subcellular organelles; proteins from the same organelle have similar abundance
profiles across the fractions and these profiles can be used to cluster proteins by PCA
according to their subcellular localisation (Figure 3.6). A machine learning algorithm
based on a Support Vector Machine (SVM) model is trained on the abundance profiles
of a large set of organelle marker proteins and can then be used to predict the organelle
localisation of unknown proteins. The method makes used of SILAC-based quantitative
MS to quantify proteins in five membrane fractions relative to an invariant reference
fraction (the ‘high speed spin’). In addition, nuclear and cytosolic fractions are har-
vested and can be used for a global assessment of a protein’s membrane/nuclear/cytosol
split. The method has also been adapted for label-free and tandem mass tagging (TMT)-
based chemical labelling quantification approaches (Itzhak et al., 2017), but was applied
here in its original SILAC form. The method is highly reproducible, allowing the result-
ing ‘cell maps’ to be used comparatively to identify proteins which change subcellular
localisation in one sample relative to another (described as a translocation). In the
original Dynamic Organellar Maps paper (Itzhak et al., 2016), Itzhak and colleagues
demonstrated the utility of the method in this regard by comparing maps from HeLa
cells before and after EGF (epidermal growth factor) stimulation. They were able to
detect the translocation of the EGF receptor from the plasma membrane to endosomes,
the recruitment of known signalling adaptors, and novel translocation events. This
proof-of-principle study suggested that Dynamic Organellar Maps is a method capable
of identifying protein translocation events on a global scale without a requirement for a
priori knowledge of a pathway. As AP-4 is proposed to function in protein sorting, we
expect a subset of proteins that depend on AP-4 for their trafficking from the TGN to
be mislocalised in AP-4-deficient cells. Thus, we decided to apply Dynamic Organellar
Maps as an unbiased screen for the identification of AP-4 cargo proteins.
3.3.1 Protein translocations in AP-4 knockout cells
We prepared organellar maps from wild-type, AP4B1 knockout and AP4E1 knockout HeLa
cells, in biological duplicate (Figure 3.7). For each map, cells were grown in two 15 cm
dishes, one in SILAC light medium and the other in SILAC heavy medium. The cells were
90 Proteomic investigations of adaptor protein complex 4
Fig. 3.6 Workflow for Dynamic Organellar Map generation. (A) Cells are grown in light and
heavy SILAC medium. Light cell lysate is subjected to a series of differential centrifugation steps,
to achieve partial separation of organelles. Heavy cell lysate is subjected to a high speed spin
(HSS) to generate a heavy-labelled reference fraction. Each light sub-fraction is mixed with an
equal protein amount of the heavy reference fraction, digested with trypsin to yield peptides
and analysed by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS).
SILAC-based relative quantification is used to calculate abundance distribution profiles for
proteins across the sub-fractions. (B) Proteins associated with the same organelle have similar
abundance distribution profiles. Median profiles for marker proteins of the Golgi, endosome,
mitochondrion, plasma membrane and endoplasmic reticulum are shown, which are all different.
(C) Proteins can be clustered by organelle according to their abundance distribution profiles.
Clustering can be visualised by principal component analysis (PCA) to give an organellar map.
Diagrams were adapted from Itzhak et al. (2016).
3.3 Dynamic Organellar Maps of AP-4 knockout cells 91
Fig. 3.7 Experimental design for Dynamic Organellar Mapping of wild-type and AP-4 knock-
out (KO) HeLa cells. Maps were made from wild type, AP4B1 KO and AP4E1 KO cell lines, each in
duplicate. Profiles from each KO map were subtracted from the cognate control profiles, to obtain
two AP4E1 delta maps, and two AP4B1 delta maps. Proteins that did not shift had similar profiles
in wild-type and AP-4 KO maps, and hence delta profiles close to zero. To identify significantly
translocating proteins, the magnitude of shift (M) and the reproducibility of shift direction (R)
were scored for each protein and each delta map.
lysed mechanically and post-nuclear supernatant from light cells was subfractionated by
five differential centrifugation steps. In parallel, a single ‘reference’ membrane fraction
was obtained from the heavy post-nuclear supernatant by centrifugation at the top
speed of the gradient. Each light subfraction was then mixed with an equal amount
of the heavy reference fraction and analysed by liquid chromatography-tandem mass
spectrometry (LC-MS/MS). This revealed abundance distribution profiles across the
subfractions, consisting of up to five H/L SILAC ratios per protein. Only proteins with
complete profiles in all six maps were included in the analysis (3,926 proteins after
standard data filtering as described in Section 2.15).
The abundance distribution profiles of all proteins were compared between the wild-
type and AP-4 knockout maps, to search for proteins with an altered localisation in
the AP-4 knockout cells. Ratios were inverted to L/H so they can be interpreted as the
level of enrichment of a protein in the subfraction relative to the reference fraction.
The profiles were then normalised based on the subfraction yields and adjusted so the
ratios sum to 1, meaning they represent the proportion of the protein pelleted in each
subfraction. Normalised profiles from each knockout map were subtracted from their
matched control profiles (maps prepared on the same day), to obtain two AP4E1 delta
92 Proteomic investigations of adaptor protein complex 4
maps and two AP4B1 delta maps (Figure 3.7). Proteins that did not shift had similar
profiles in wild-type and AP-4 knockout maps and hence delta profiles close to zero. To
identify significantly translocating proteins, the magnitude of shift (Movement score or
‘M’) and the reproducibility of shift direction (Reproducibility score or ‘R’) were scored
for each protein and each delta map. M was calculated using a multivariate outlier test
(in Perseus software; Itzhak et al., 2016; Tyanova et al., 2016) to identify proteins with
delta profiles significantly different from experimental scatter and corresponds to a
p-value that represents how likely it is to observe this difference by chance, assuming
the null hypothesis of no change. R was the correlation coefficient of the delta profiles
of replicate maps. To combine data from all four delta maps (two AP4B1 and two AP4E1)
we selected the least significant M score to represent a protein’s movement and the lower
of the two correlation coefficients as a conservative estimate of the reproducibility of the
movement. These values were plotted against each other for every protein to give what
we call an ‘MR’ plot (Figure 3.8A). With stringent cut-offs (M score >4 and R score >0.8),
this revealed three transmembrane proteins as being reproducibly mislocalised in the
AP-4 knockout cells - ATG9A (Autophagy-related protein 9A), SERINC1 and SERINC3
(Serine incorporator 1 and 3). As an additional reproducibility check we calculated the
cross-correlation between the delta profiles of the AP4B1 and AP4E1 maps and for all
three proteins R was >0.9, meaning the protein subcellular localisation change had a
similar direction across both knockout clones.
Statistical analyses that involve multiple comparisons (i.e. the calculation of many
individual p-values, in this case 4 for each of the 3,926 proteins, totalling 15,704) suffer
from the fact that based on probability some p-values are likely to be small by chance.
Therefore, it is important to correct for multiple comparisons so as to avoid a large
number of false positives. There are different ways of doing this, but here we chose to
use the false discovery rate (FDR) approach (Benjamini & Hochberg, 1995). The FDR
is the chance that a statistically significant finding is actually due to a coincidence of
random sampling, i.e. that it is a false positive. The FDR approach controls for a certain
proportion of false positives, for example if the FDR is 0.05 then five percent of statistically
significant findings are expected to be false positives. This is most commonly applied as
a theoretical cut-off based on FDR-corrected p-values (called Q-values). A better method
is to perform a ‘mock’ experiment to empirically define FDR cut-offs. Here we took six
untreated wild-type HeLa maps (previously published in Itzhak et al., 2016) in which
we would expect no genuine protein translocations. Two were designated as controls,
two as ‘mock AP4B1 knockout’ and two as ‘mock AP4E1 knockout’, and an identical
analysis was performed on these maps as that described above for the real experiment.
3.3 Dynamic Organellar Maps of AP-4 knockout cells 93
Fig. 3.8 Protein translocations in AP-4 knockout (KO) cells. (A) MR plot analysis of AP-4 Dy-
namic Organellar Mapping. 3,926 proteins were profiled across all maps. Three proteins whose
subcellular localisation was significantly and reproducibly shifted across the AP-4 KO lines were
identified with very high confidence (FDR <1%). The analysis only covered proteins profiled
across all maps; since AP-4 itself was not present in the KO maps, it was not included. (B-D)
Visualisation of organellar maps by principal component analysis (PCA). PCA scatter plots were
created by Daniel Itzhak. Each scatter point represents a protein; proximity indicates similar
fractionation profiles. Known organellar marker proteins are shown in colour (see legend), and
form clusters. Each plot combines the data from two independent map replicates. B: wild-type;
C: AP4B1 KO; D: AP4E1 KO. The three proteins that undergo significant shifts in AP-4 KOs are
annotated.
94 Proteomic investigations of adaptor protein complex 4
Table 3.1 Protein translocations in AP-4 knockout HeLa cells with an FDR of 26%. Proteins
considered to have undergone significant translocations are listed with associated Movement
(M) and Reproducibility (R) scores. Score cut-offs were M >1 and R >0.8. The three proteins that
met the more stringent score cut-offs of M >4 and R >0.8 (FDR <1%) are shaded green. Cross-R
is the correlation between the ratios from the two different AP-4 knockout lines. Localisation
and copy number are from previously published Dynamic Organellar Maps of wild-type HeLa
cells (Itzhak et al., 2016). LPC: Large Protein Complex.
Gene name Localisation M R Cross-R Copy no.
SERINC3 LPC 24.2 0.96 0.94 28,000
SERINC1 LPC 14.4 0.87 0.94 48,000
ATG9A Endosome 9.6 0.98 0.96 53,000
HMGN1 Nuclear 2.0 0.97 0.99 8,600,000
PTPN9 Endosome 2.0 0.86 0.86 26,000
AHCTF9 Nuclear 1.8 0.86 0.91 330,000
LNPEP Endosome 1.7 0.92 0.89 86,000
CCDC6 LPC 1.0 0.84 0.85 920,000
We could then use the M and R scores from the mock experiment to estimate the FDR at
any given M and R score cut-offs by taking the number of hits obtained with the mock
data and dividing it by the number of hits obtained with the AP-4 maps data. Using this
method we selected high stringency cut-offs of M >4 and R >0.8 at which no hits were
obtained in the mock data. Hence, we can estimate our FDR in the AP-4 knockout maps
experiment to be <1%, giving us very high confidence that the observed translocations
of ATG9A, SERINC1 and SERINC3 are real. In the current study we have focused on these
high confidence hits, but for a more exploratory analysis a less stringent FDR cut-off
could be applied to select a larger number of candidates for follow-up (but with higher
likelihood that some would be false positives). Table 3.1 reports proteins that would be
deemed ‘hits’ with a lower movement cut-off of M >1 (still with a reproducibility cut-off
of R >0.8), which results in an estimated FDR of 26%.
3.3.2 Further analysis of the Dynamic Organellar Maps
Following on from the comparative use of Dynamic Organellar Maps to identify the
translocations of ATG9A, SERINC1 and SERINC3 in the AP-4 knockout cells, further anal-
ysis of the maps with PCA-based visualisation, SVM-based organelle assignment and
neighbourhood analysis allowed for interpretation of the nature of the detected shifts. To
visualise the wild-type and knockout maps the ratio data from replicate maps were com-
bined for principle component analysis - i.e. the input data were 10 SILAC H/L ratios per
3.3 Dynamic Organellar Maps of AP-4 knockout cells 95
map. The resulting cell maps are displayed in Figure 3.8B-D and the positions of ATG9A,
SERINC1 and SERINC3 are marked. All maps had very similar overall structure with
highly resolved organellar clusters, and movements of ATG9A, SERINC1 and SERINC3
were apparent. On the control map all three appeared within the endosomal cluster,
with SERINC1 and SERINC3 very closely juxtaposed and ATG9A further away. The move-
ments on the knockout maps were subtle, with the three proteins still appearing within
the endosomal cluster, but they were very consistent between the AP4B1 and AP4E1
knockouts. Interestingly both SERINCs moved in the same direction in the knockout
cells suggesting that even in cells lacking AP-4 they have similar localisations. However,
the visual interpretation of the shifts is not straightforward so an SVM-based machine
learning algorithm was applied to the ratio data to predict organelle associations for
ATG9A and the SERINCs in the wild-type and knockout cells. Here the ratio data from all
four knockout maps were combined (20 H/L ratios) and the two wild-type maps were
combined (10 H/L ratios). The SVM algorithm is a supervised learning approach for
the automatic classification of data (Noble, 2006). Given a set of classified training data
(in this case a list of organelle marker proteins with their organelle class and associated
map ratios) it can define optimal non-linear boundaries between classes (in this case
organelle clusters). It can then apply these boundaries to assign classes to unlabelled
data - i.e. to predict the organelle assignment of unclassified proteins. Each classification
is associated with an SVM score which reflects the confidence of the classification based
on distance to the classification boundary. Using this method ATG9A was assigned to the
endosomal cluster and the SERINCs to the Large Protein Complex cluster in the wild-type
and all three proteins were assigned to the endosomal cluster in the knockouts (Table
3.2). However, these classifications were mostly associated with low confidence SVM
scores so it is difficult to glean anything meaningful from the assignments. This likely
reflects the fact the proteins are on the edge of the endosomal cluster and have mixed
steady state distributions. The SVM algorithm struggles with mixed profiles because the
marker proteins used to train it mostly localise to a single organelle.
Further information about the translocations in the AP-4 knockout cells came from a
neighbourhood analysis performed by Dr Georg Borner. Ratio data from replicate maps
were combined and the profiles for ATG9A, SERINC1 and SERINC3 were compared to
those of all other proteins assigned as endosomal or lysosomal by the SVM analysis.
Profile similarity was assessed by calculation of the squared Euclidian distance between
profiles as described in detail by Itzhak et al. (2016). The endosomal and lysosomal
proteins with the most similar profiles (i.e. lowest Euclidian distance) were identified
for each of ATG9A, SERINC1 and SERINC3, in the wild-type, AP4B1 knockout and AP4E1
96 Proteomic investigations of adaptor protein complex 4
Table 3.2 SVM organelle assignment for ATG9A, SERINC1 and SERINC3 in wild-type and AP-4
knockout HeLa cells. The SVM prediction scores associated with the organelle assignments are
listed. A score of >1 is high confidence, 0.5–1 is medium confidence, 0–0.5 is low confidence and
<0 should be discounted. LPC: Large Protein Complex.
Gene name
Wild-type AP-4 knockout
Organelle Score Organelle Score
ATG9A Endosome 0.9 Endosome 0.0
SERINC1 LPC −0.8 Endosome 0.5
SERINC3 LPC −0.2 Endosome −0.5
knockout maps. The eight closest endosomal and lysosomal neighbours in each case
are shown in Table 3.3. This revealed a shift in the nearest neighbours of all three
proteins in the knockout cells, which reflects a shift in localisation within the endosomal
cluster. For the SERINCs the nearest neighbours change in the knockout cells but are still
predominated by late endosomal/lysosomal markers, suggesting an intra-endosomal
translocation. In both the wild-type and AP-4 knockout cells the SERINCs feature on
each others list of nearest neighbours demonstrating that they have closely related
subcellular distributions, even in the absence of AP-4. For ATG9A there is a clear shift
from mainly endosomal neighbours in the wild-type cells (e.g. AP-3 complex subunits)
towards neighbours associated with the TGN region (e.g. AP-1 complex subunits and
the SNAREs STX6 and STX16) in the knockout cells, suggesting a shift towards the TGN.
In conclusion, Dynamic Organellar Maps has proved effective for the unbiased detection
of protein mislocalisation in AP-4 knockout cells. This analysis has revealed robust
translocations of three highly conserved transmembrane proteins, ATG9A, SERINC1 and
SERINC3, with a highly stringent FDR control. It also provides some clues as to the nature
of these translocations. Based on this ATG9A and the SERINCs are strong candidates for
novel AP-4 cargo proteins. The next section presents an orthogonal organellar proteomic
approach which provides further support for their candidacy as AP-4 cargo proteins and
identifies additional AP-4-associated proteins.
3.3 Dynamic Organellar Maps of AP-4 knockout cells 97
Table 3.3 Eight nearest endosomal or lysosomal neighbours for ATG9A, SERINC1 and SER-
INC3 in wild-type and AP-4 knockout (KO) cells. Profile distances were calculated using the
squared Euclidian distance; the proteins with the eight lowest distances for each query protein
are listed here. This analysis was performed by Dr Georg Borner. E: Endosome; L: Lysosome; G:
Golgi; TGN: Trans-Golgi network.
Query
Wild-type AP4B1 KO AP4E1 KO
Gene
name
Organelle Gene
name
Organelle Gene
name
Organelle
ATG9A M6PR E AP1G1 TGN/E SCAMP3 E
OCRL E TSG101 E STX16 TGN/E
VAC14 E VPS37B E AP1G1 TGN/E
SCAMP3 E STX6 G/TGN/E STX6 G/TGN/E
AP3D1 E AP1M1 TGN/E SCAMP1 E
IGF2R E SCAMP3 E AP1M1 TGN/E
AP3B1 E AP1S1 TGN/E VAMP3 E
AP3M1 E MVB12A E AP1S1 TGN/E
SERINC1 SERINC3 E RAB9A E RAB9A E
VAMP7 E LAMTOR5 L CLN3 E
RPTOR L PSAP L SERINC3 E
GDAP2 L LAMTOR4 L ATP6V1F E
CLCN3 E CLN3 E LAMTOR5 L
STX12 E IFITM3 E ATP6AP1 E
RAB14 E SERINC3 E ANKFY1 E
VTI1B E CLCN7 L LAMTOR4 L
SERINC3 SERINC1 E RAB9A E RAB9A E
RPTOR L APP E ATP6V1F E
RAB14 E ATP6V1F E RAB7A E
VAMP7 E SERINC1 E SERINC1 E
VTI1A E ATP6AP1 E CLN3 E
VAMP4 TGN PSAP L RAB27A E
CLCN3 E CTSL L IFITM3 E
STX12 E IFITM3 E LAMTOR5 L
98 Proteomic investigations of adaptor protein complex 4
3.4 Comparative vesicle profiling of AP-4 depleted cells
Dr Jennifer Hirst, working in the Robinson Lab, developed the use of CCV isolation as an
assay for the investigation of adaptor function (Hirst et al., 2004). In her initial study she
used Western blotting to compare CCV fractions isolated from control HeLa cells and
HeLa cells depleted of AP1M1 (AP-1 complex subunit μ1) or CLINT1 (Clathrin interactor
1; also known as EpsinR) by siRNA. Dr Georg Borner, also working in the Robinson Lab,
pioneered the combination of CCV isolation and MS-based comparative proteomics
to identify genuine CCV proteins by comparison to a ‘mock’ CCV fraction prepared
from clathrin-depleted cells (as discussed above; Borner et al., 2006). It was then a
logical step to marry the two approaches to investigate the function of clathrin adaptors
(Hirst et al., 2012). Hirst, Borner and colleagues used SILAC-based quantification to
comparatively analyse CCV fractions prepared from HeLa cells in which either AP-1
or the alternative adaptor GGA2 (Golgi-localised, γ ear-containing, ADP-ribosylation
factor-binding protein 2) had been rapidly inactivated using the knocksideways system
(Robinson et al., 2010). In the knocksideways system a protein of interest is tagged with
an FKBP domain (from peptidyl-prolyl cis-trans isomerase FKBP1A) and an FRB domain
(rapamycin binding domain from serine/threonine-protein kinase mTOR) is anchored
in the outer mitochondrial membrane (termed ‘Mitotrap’). The addition of the drug
rapamycin induces heterodimerisation of the FKBP-tagged protein and Mitotrap and
thus a protein of interest can be rapidly sequestered onto mitochondria by the addition
of rapamycin to the culture medium. In just 10 minutes the rerouting is essentially
complete (Hirst et al., 2012) so this makes for much faster protein inactivation than
siRNA-mediated knockdown. This approach revealed a large number of proteins that
are lost from the CCV fraction in the absence of AP-1 (around 100) and a smaller, partially
overlapping set of proteins lost in the absence of GGA2. Careful interpretation of the data
suggested AP-1 acts as a linchpin for intracellular CCV formation, and is involved with
bidirectional trafficking from the TGN to endosomes, while GGA2 acts in conjunction
with AP-1 to traffic hydrolase-receptor complexes in the anterograde direction from the
TGN to endosomes.
As previously mentioned, the CCV fraction is not a pure preparation as it also contains
contaminants and other clathrin-independent vesicle populations. AP-4 and the AP-4
accessory protein TEPSIN are enriched in the CCV fraction (Borner et al., 2012), sug-
gesting that there is a population of AP-4-derived vesicles in the fraction. Therefore, the
comparative vesicle profiling approach described above can be adapted to study the
function of AP-4 (Figure 3.9A). A slight modification of the CCV isolation protocol with an
3.4 Comparative vesicle profiling of AP-4 depleted cells 99
increased final spin speed allows for enhanced enrichment of non-CCVs (Borner et al.,
2012). From this point on this preparation is referred to as the ‘vesicle-enriched fraction’
or the ‘vesicle fraction’. Figure 3.9B outlines the steps involved in the preparation of the
vesicle-enriched fraction. To adapt this method to study AP-4 function we decided to
use four different methods to ablate AP-4 function in HeLa cells: (i) siRNA-mediated
knockdown; (ii) knocksideways; (iii) knockout of AP4B1; (iv) knockout of AP4E1. We
then used SILAC-based quantitative MS to compare the composition of control and
AP-4-depleted vesicle fractions. The reason for using four different methods of AP-4
depletion was to control for any off-target effects caused by any one method and to use
the multivariate proteomic profiling approach previously applied by Dr Borner to CCVs
(Borner et al., 2012). Through the combination of datasets using PCA, candidate AP-4
cargo and machinery were identified by their shared behaviour across all the experiments
(as detailed below).
3.4.1 Preparation and MS analysis of AP-4-depleted vesicle fractions
Vesicle-enriched fractions from AP-4 knockdown and knocksideways cells were prepared
and analysed using LC-MS/MS by Dr Georg Borner. For the knockdown experiments
(performed in triplicate) HeLa cells were grown in SILAC light and heavy media. Knock-
down of AP-4 was performed on the light-labelled cells using two different siRNAs to
simultaneously target AP4E1 and AP4M1, with a 96 hours 2-hit protocol (as described
in Borner et al., 2012). Heavy-labelled control cells were left untreated. For the knock-
sideways experiments HeLa cells stably expressing siRNA-resistant FKBP-tagged AP4E1
were grown in SILAC light and heavy media. Endogenous AP4E1 was depleted using
siRNA and AP-4 was rerouted onto mitochondria with either a standard 10 minutes
rapamycin treatment (experiment performed in triplicate) or an extended 60 minutes
rapamycin treatment (experiment performed in duplicate). Control cells expressed the
FKBP-tagged AP4E1 but were not treated with rapamycin. Efficient rerouting of AP-4
onto mitochondria was verified by microscopy and was complete within the 10 minutes
treatment (data not shown). Each set of experiments included a label swap where the
SILAC-labelling of control and rapamycin-treated cells was reversed. For the AP4B1 and
AP4E1 knockout experiments (each performed in duplicate with a label swap) wild-type
HeLa and AP4B1 or AP4E1 knockout HeLa cells were grown in SILAC light and heavy
media. Vesicle-enriched fractions were prepared in all instances using the protocol
outlined in Figure 3.9B.
100 Proteomic investigations of adaptor protein complex 4
Fig. 3.9 Workflow for proteomic profiling of AP-4 vesicles. (A) Vesicle-enriched fractions were
prepared in pairs from metabolically (SILAC heavy or light) labelled control and AP-4-depleted
cells and compared by quantitative MS. 2,848 proteins were quantified across all experiments.
Principal component analysis (PCA) was used to combine the SILAC ratios from all nine com-
parative AP-4-depleted vesicle fraction experiments. (B) Diagrammatic representation of the
steps involved with preparing the vesicle-enriched fraction. Cells are homogenised mechanically
and cell debris is removed with a low speed spin. Homogenate is treated with ribonuclease A
(RNaseA) and partially digested ribosomes are removed by centrifugation. The supernatant is
then subjected to a high speed spin to pellet membrane components. The membrane fraction is
resuspended in a ficoll/sucrose solution and centrifugation results in pelleting of larger contami-
nants while small vesicles remain in the viscous solution. The vesicle-containing supernatant is
diluted to reduce the viscosity and the vesicles are pelleted with a final high speed spin (40,000
rpm). The vesicle pellet can be resuspended in 2.5% SDS and analysed by Western blotting or
MS.
3.4 Comparative vesicle profiling of AP-4 depleted cells 101
Table 3.4 Proteins lost from the vesicle fraction of AP-4-depleted cells. Proteins that were
depleted in every one of nine comparative vesicle fraction experiments (3× KD: knockdown;
4× KO: knockout; 2× KS: 60 minutes knocksideways) are listed (from a total of 2,848 proteins
quantified in every experiment), sorted by the sum of their ratios. AP4S1 was not identified in
one experiment (ε KO 2) but was included because it is a known component of the AP-4 complex.
Gene
names
Fold depletion in AP-4-depleted vesicle fraction
KD 1 KD 2 KD 3 β4
KO 1
β4
KO 2
ε
KO 1
ε
KO 2
KS 1 KS 2
AP4B1 27.2 24.0 16.5 7.7 10.0 10.1 11.1 9.1 28.4
AP4M1 35.4 23.4 14.4 34.3 7.4 2.7 2.1 9.7 35.4
AP4S1 40.7 18.7 11.8 9.3 16.1 10.6 No ID 12.0 32.1
RUSC1 11.3 12.1 9.6 50.8 83.1 22.4 15.1 1.4 1.9
AP4E1 13.8 20.8 16.7 36.6 12.4 11.8 4.8 4.7 7.1
TEPSIN 10.9 13.8 9.0 1.3 8.9 7.5 9.2 7.4 12.1
RUSC2 2.0 3.1 4.4 9.4 9.3 3.0 4.6 2.4 3.2
SERINC3 5.5 5.5 7.2 1.8 1.6 2.1 2.1 1.7 1.8
SERINC1 5.2 4.9 5.6 1.7 1.4 2.2 2.1 1.6 1.6
ATG9A 1.4 1.6 1.7 1.9 1.8 1.4 1.5 1.3 1.6
DAGLB 1.3 1.3 1.4 1.7 1.4 1.1 1.2 1.3 1.4
TWF2 1.2 1.3 1.1 1.1 1.3 1.4 1.7 1.1 1.1
Paired light and heavy vesicle-enriched fractions from each experiment were mixed,
digested with trypsin to fragment proteins into peptides, and analysed by LC-MS/MS.
The MS raw files from all experiments (a total of 12 SILAC pairs) were processed together
in MaxQuant for protein identification and the calculation of a ratio of relative abundance
(Heavy/Light) for each identified protein in each experiment. After standard data filtering
this resulted in the identification of 5,847 proteins from all experiments combined, each
of which had up to 12 H/L ratios associated with it. Proteins were further filtered to
leave only those which had an H/L ratio from every experiment. This excluded AP4S1
due to one missing datapoint, so the data for AP4S1 was manually added back to the
dataset, giving a total of 2,848 proteins quantified across all experiments. In Figure 3.10
the log2-transformed normalised ratios from replicate experiments are plotted against
each other. For experiments in which the control cells were light labelled, normalised
H/L ratios were inverted to L/H so that depletion from the vesicle fraction in the absence
of AP-4 is always represented by a positive value. Therefore, proteins in the top right
portion of each plot were depleted from the vesicle-enriched fraction in cells lacking
AP-4. Table 3.4 presents the ratio data associated with 12 proteins that were consistently
affected across the experiments.
102 Proteomic investigations of adaptor protein complex 4
Fig. 3.10 Pairwise comparison of control and AP-4 depleted vesicle fractions. SILAC quantita-
tive proteomic profiling of AP-4-depleted vesicle fractions, with AP-4 depletion by knockdown
(three replicates), 10 minutes knocksideways (three replicates), 60 minutes knocksideways (two
replicates), or knockout of AP4B1 or AP4E1 (two replicates each). SILAC ratios (wild-type/AP-4-
depleted) from replicate experiments are plotted against each other. Proteins that are lost from
the vesicle-enriched fraction in the absence of AP-4 are found in the top right sectors of the plots.
Proteins of interest are labelled. The different depletion approaches act on different time scales,
and hence have subtly different effects on vesicle composition. However, all methods show very
similar overall trends.
3.4 Comparative vesicle profiling of AP-4 depleted cells 103
In all 12 experiments the AP-4 subunits themselves and the AP-4 accessory protein
TEPSIN were among the proteins most strongly depleted from the vesicle fraction in the
absence of AP-4. This demonstrated that all methods of AP-4 depletion were effective at
disrupting AP-4 complex function. The reduction of AP-4 in the vesicle fraction did not
affect the levels of other vesicle machinery such as AP-1, AP-2 or AP-3 complex subunits,
or clathrin (Figure 3.11). Next the behaviour of the proteins identified as candidate AP-4
cargo proteins in the Dynamic Organellar Maps experiment (Section 3.3) was assessed.
The SERINCs were strongly depleted from the vesicle fractions prepared from the AP-4
knockdown cells (∼5-fold depletion of SERINC1 and ∼6-fold depletion of SERINC3) and to
a lesser extent from the 60 minutes knocksideways and knockout cells (∼2-fold depletion).
ATG9A was also depleted from the vesicle fractions from these cells, but not as strongly
as the SERINCs (between 1.3- and 1.9-fold depletion). Neither the SERINCs, nor ATG9A,
were depleted from vesicle fractions prepared from the 10 minutes knocksideways cells,
suggesting that 10 minutes is not long enough to see a shift in the steady-state distribution
of AP-4 cargo. Therefore, the data from the 10 minutes knocksideways experiments were
not included in further analyses. To search for additional AP-4-dependent proteins
in the vesicle fraction, information from the other nine datasets was combined using
PCA (Figure 3.12). Plotting the first principle component against either the second
or third principle component resulted in clear separation of AP-4 complex subunits
and TEPSIN from the majority of proteins. The SERINCs were also clear outliers in the
same direction as the core AP-4 complex, reflecting their consistent loss from vesicle
fractions prepared from AP-4-depleted cells. The effect on ATG9A was much more
subtle, but it still separated from the main cluster of proteins, in the direction of the
AP-4 complex. Furthermore, two related cytosolic proteins called RUSC1 and RUSC2
(RUN and SH3 domain-containing protein 1 and 2) were also clearly identified as being
reproducibly lost from the vesicle fractions of AP-4 ablated cells. In fact, they were more
strongly affected than the SERINCs. Returning to the data from the individual vesicle
fraction experiments, RUSC1 was very strongly depleted from vesicle fractions from
AP-4 knockdown and knockout cells, and less affected (but still lost) in the 60 minutes
knocksideways cells. RUSC2 was particularly strongly depleted in the knockouts and
affected to a similar degree by knockdown or knocksideways. Given these strong effects
we were surprised that translocation of the RUSCs was not picked up using Dynamic
Organellar Maps. When we searched the maps datasets for the RUSCs we found that
they were not included in the analysis because they were not consistently profiled across
the subcellular fractions. This is likely due to their very low copy number in HeLa
cells (RUSC1: ∼16,000; RUSC2: ∼3,000; Itzhak et al., 2016). However, both RUSC1 and
104 Proteomic investigations of adaptor protein complex 4
Fig. 3.11 Other vesicle machinery is unaffected by AP-4 depletion. SILAC ratio (control/AP-4-
depleted vesicle fraction) data from 3× knockdown (KD), 4× knockout (KO) and 5× knockside-
ways (KS) experiments. For each method of AP-4 depletion the mean ratios are plotted for each
of AP1M1, AP2M1, AP3M1 and clathrin (CLTC). Error bars represent standard deviation. A ratio
of 1.0 reflects no difference in the amount of protein between control and AP-4-depleted vesicle
fractions.
RUSC2 are highly enriched in the vesicle fraction, suggesting they are vesicle-associated
proteins.
The comparative proteomic profiling of vesicle fractions from AP-4-depleted cells thus
supports that ATG9A, SERINC1 and SERINC3 are AP-4 cargo proteins and has identified
RUSC1 and RUSC2 as candidates for cytosolic AP-4-associated machinery.
3.4 Comparative vesicle profiling of AP-4 depleted cells 105
Fig. 3.12 Principal component analysis of vesicle fractions from AP-4-depleted cells. PCA
combining the SILAC ratios from nine comparative AP-4-depleted vesicle fraction experiments
(3× knockdown; 2× AP4B1 knockout; 2× AP4E1 knockout; 2× 60 minutes knocksideways). (A)
The first and second principal components are plotted, accounting for 30% and 19% of the
variability, respectively. (B) The first and third principal components are plotted, accounting for
30% and 14% of the variability, respectively. Proteins consistently lost from the vesicle fraction of
AP-4-depleted cells are in the top-right section.
106 Proteomic investigations of adaptor protein complex 4
3.5 BioID to search for AP-4 cargo and machinery
The interactions between an AP complex and its cargo and other vesicle machinery are
transient and so have largely proved refractory to standard co-immunoprecipitation ap-
proaches. Therefore, as an alternative approach to search for AP-4 cargo and machinery,
we applied the BioID method (Roux et al., 2012). BioID (which stands for proximity-
dependent biotin identification) uses a promiscuous mutant prokaryotic biotin ligase,
BirA* (R116G), genetically fused to a protein of interest. The R116G mutation causes
the enzyme to have an affinity for activated biotin two orders of magnitude less than
that of the wild-type enzyme (which is highly specific for its substrate). This leads to
premature release of activated biotin from the enzyme, which reacts with the primary
amines (mostly lysines) of nearby proteins, within an estimated range of ∼10 nm (Kim
et al., 2014). This allows an in vivo screen for proteins in close proximity to the protein of
interest. Here, we generated HeLa cell lines that stably expressed each of the four AP-4
complex subunits tagged with BirA*. Following an incubation in excess biotin for 24
hours, biotinylated endogenous proteins were affinity-purified from each cell line and
identified by mass spectrometry (Figure 3.13). The use of appropriate control cell lines is
essential in BioID to control for non-specific biotinylation caused by overexpression or
mislocalisation of the BioID fusion protein. A study from the Gingras Lab explored the
use of multiple controls in BioID experiments, concluding that a combination of controls
with different expression patterns provided better coverage of background biotinylation
than increasing the number of replicates of a single control (Lambert et al., 2015). Based
on this study we decided to use three separate controls in our experiments: (i) parental
HeLa cells to control for endogenously biotinylated proteins and proteins that bind
non-specifically to the streptavidin agarose; (ii) HeLa cells stably expressing untargeted
BirA*, which is predominantly nucleoplasmic; (iii) HeLa cells stably expressing GFP-
BirA*, which is predominantly cytoplasmic. Comparison to these controls identified
proteins specifically enriched in the streptavidin pulldowns from the AP-4 BioID cell
lines, revealing proteins that come into close proximity with the AP-4 complex.
3.5.1 Generation and validation of AP-4 BioID cell lines
The first step in the generation of AP-4 BioID cell lines was to clone each of the four AP-4
subunits with a BirA* tag (38 kDa). Based on prior experience in the lab of tagging adaptor
protein complex subunits with bulky tags, AP4B1, AP4M1 and AP4S1 were C-terminally
tagged, while BirA* was placed within the flexible hinge region of AP4E1 (Figure 3.14A).
3.5 BioID to search for AP-4 cargo and machinery 107
Fig. 3.13 Diagram of the BioID method. Cells stably expressing the BioID-fusion protein are cul-
tured with excess biotin, leading to selective biotinylation of nearby proteins. Following cell lysis,
biotinylated proteins are affinity-purified using streptavidin-conjugated agarose. Biotinylated
proteins can be identified by mass spectrometry (adapted from Roux et al., 2012).
The BirA* tag also included a myc epitope so the tagged proteins could be recognised
using an antibody against either BirA or myc. The constructs were made using Gibson
Assembly and cloned into a pLXIN retroviral vector. An expression vector for GFP-BirA*
was made using standard restriction enzyme-based cloning to introduce myc-BirA* into
pEGFP-N2. Successful cloning of all constructs was confirmed by diagnostic digests and
sequencing (data not shown).
To generate stable cell lines for the AP-4 BioID fusion proteins, HeLa cells were trans-
duced with retrovirus and selected for stable expression of the tagged AP-4 subunits by
addition of G418. To generate the control cell line stably expressing GFP-BirA*, HeLa
cells were transfected with the pEGFP-BirA* expression vector, selected for stable expres-
sion in G418, and then fluorescence-activated cell sorting (FACS) was used to enrich for
GFP-positive cells. This resulted in a mixed population of cells with a stable expression
level of GFP-BirA*, which was monitored using flow cytometry (data not shown). HeLa
cells stably expressing untargeted BirA* were a gift from Tom O’Loughlin (University of
Cambridge). Western blotting of whole cell lysates from these cell lines with antibodies
against BirA or the AP-4 subunits showed expression of fusion proteins of the expected
sizes (Figure 3.14B & C). The expression levels of the AP-4 BioID proteins were roughly
similar to that of free BirA* and GFP-BirA*. Interestingly, the stable overexpression of
AP4B1-BirA* and AP4E1*-BirA appeared to cause a reduction in the levels of endoge-
nous AP4B1 and AP4E1, respectively (Figure 3.14C; endogenous AP4M1 and AP4S1 were
not detected). This suggests that the overexpressed tagged subunit outcompetes the
endogenous protein for incorporation into AP-4 complexes, resulting in destabilisation
of the endogenous protein. This was a promising clue that the tagged AP-4 subunits are
able to form functional AP-4 complexes.
108 Proteomic investigations of adaptor protein complex 4
Fig. 3.14 AP-4 BioID cell line generation. (A) AP4B1 (83 kDa), AP4M1 (50 kDa) and AP4S1 (17
kDa) subunits were C-terminally tagged with BirA* (38 kDa) and AP4E1 (127 kDa) was tagged in
its flexible hinge region. (B) Western blot of whole cell lysates from HeLa cells stably expressing
BirA*, GFP-BirA*, AP4B1-BirA*, AP4E1-BirA*, AP4M1-BirA* or AP4S1-BirA*, with an antibody
against BirA. All expressed a product of the expected size. Wild-type HeLa cells were included as a
negative control. (C) Western blots of whole cell lysates shown in B with antibodies against AP4B1,
AP4E1, AP4M1, or AP4S1 confirmed the identity of the bands seen using anti-BirA in B. Specific
bands for the endogenous or tagged proteins are marked with an arrowhead and non-specific
bands are marked with an asterisk. Endogenous AP4M1 and AP4S1 were not detected.
3.5 BioID to search for AP-4 cargo and machinery 109
The successful application of BioID relies on the correct subcellular targeting of the
BioID fusion protein. To test this the AP-4 BioID cell lines were analysed by immunoflu-
orescence microscopy with an antibody against myc to detect the BioID fusion protein
and an antibody against the AP-4 accessory protein TEPSIN (Figure 3.15). All four AP-4
BioID proteins localised predominantly to the perinuclear TGN area and colocalised
with TEPSIN, confirming that the BirA* tags did not disrupt their subcellular distribution
or interaction with TEPSIN. In the mixed populations of cells the majority of cells were
myc-positive and had similar levels of expression, so it was not necessary to isolate single
cell clones.
To check for promiscuous biotin ligase activity in the BioID cell lines they were cultured
in the presence of excess biotin for 24 hours and then whole cell lysates were harvested
and biotinylated proteins were affinity-purified on streptavidin agarose. Western blots of
whole cell lysates and streptavidin pulldown samples were probed with HRP-conjugated
streptavidin to detect biotinylated proteins (Figure 3.16A). A few bands were detected in
the parental HeLa lysates, likely the small number of endogenous proteins that are natu-
rally biotinylated. A large increase in biotinylated proteins was detected in all the BioID
cell lines, with the highest levels in the cells expressing untargeted BirA* and AP4S1-BirA*.
The band patterns were broadly similar for all the cell lines, suggesting that the BirA*
and GFP-BirA* cell lines would act as good controls for background biotinylation in the
AP-4 BioID lines. Next the localisation of biotinylated proteins was analysed in each
cell line using immunofluorescence microscopy. Cells were labelled with fluorophore-
conjugated streptavidin to detect biotinylated proteins and an antibody against TEPSIN
(Figure 3.16B). As expected, the streptavidin signal was predominantly nucleoplasmic
in the BirA* cells and cytoplasmic in the GFP-BirA* cells. In the AP-4 BioID cells the
streptavidin signal appeared predominantly cytosolic with a concentration in the perin-
uclear region where the TGN is found (HeLa AP4E1-BirA* cells are shown as an example).
Partial colocalisation between the streptavidin signal and TEPSIN was observed. This
provided further evidence that the AP-4 BioID proteins were localised correctly. As a final
validation step, the samples from the streptavidin pulldown experiment were analysed
by Western blotting with antibodies against AP4E1 and TEPSIN (Figure 3.17). Both AP4E1
and TEPSIN were specifically pulled down from the lysates of the AP-4 BioID cell lines
and not from the control cell line lysates. This demonstrated that the AP-4 BioID fusion
proteins were interacting with endogenous proteins as expected and so the cell lines
were used in an MS-based screen to search for novel AP-4-associated proteins.
110 Proteomic investigations of adaptor protein complex 4
Fig. 3.15 AP-4 BioID fusion proteins colocalise with TEPSIN. Widefield imaging of HeLa cells
stably expressing AP4B1-BirA*, AP4E1-BirA*, AP4M1-BirA* or AP4S1-BirA* labelled with anti-myc
(to detect the BioID fusion protein) and anti-TEPSIN. Scale bar: 20 μm.
3.5 BioID to search for AP-4 cargo and machinery 111
Fig. 3.16 Promiscuous biotin ligase activity in AP-4 BioID cells. AP-4 and control BioID cell
lines were cultured in the presence of excess biotin (50 μM) and analysed by Western blotting
and immunofluorescence microscopy. (A) Western blots of whole cell lysates (input lysates) and
streptavidin pulldowns from wild-type HeLa cells or HeLa cells stably expressing BirA*, GFP-
BirA*, AP4B1-BirA*, AP4M1-BirA* or AP4S1-BirA*, probed with HRP-conjugated streptavidin to
detect biotinylated proteins. B) Widefield imaging of wild-type HeLa cells or HeLa cells stably
expressing BirA*, GFP-BirA* or AP4E1-BirA*, labelled with streptavidin-568 and anti-TEPSIN.
Scale bar: 20 μm. Streptavidin labelling was similar in the other AP-4 BioID cell lines (not shown).
112 Proteomic investigations of adaptor protein complex 4
Fig. 3.17 AP-4 BioID fusion proteins biotinylate endogenous AP4E1 and TEPSIN. Wild-type
HeLa cells and HeLa cells stably expressing BirA*, GFP-BirA*, AP4B1-BirA*, AP4E1-BirA*, AP4M1-
BirA* or AP4S1-BirA* were cultured in excess biotin (50 μM) for 24 hours, lysed, and biotinylated
proteins were affinity-purified using streptavidin agarose. Western blots of whole cell lysates
(input) and streptavidin pulldown samples with anti-AP4E1 and anti-TEPSIN. AP4E1 and TEPSIN
were specifically biotinylated in the AP-4 BioID cell lines and not in the control lines.
3.5.2 Identification of AP-4 proximal proteins by mass spectrometry
Following validation of the AP-4 BioID cell lines, they were used in an MS-based screen for
AP-4 machinery and cargo proteins. Wild-type HeLa cells and the control and AP-4 BioID
cell lines were grown in 15 cm dishes and excess biotin (50 μM) was added to the culture
medium for 24 hours before harvest. The cells were lysed in detergent and biotinylated
proteins were affinity-purified on streptavidin agarose. Biotinylated proteins were then
digested to yield peptides by an on-the-bead tryptic digest and peptides were cleaned-up
and analysed by LC-MS/MS. The experiment was performed in triplicate and the MS raw
files from all samples were processed together in MaxQuant to perform relative label-free
quantification across all the samples. Label-free quantification was the method of choice
here, rather than a SILAC approach, so we could compare all samples simultaneously.
The MaxLFQ algorithm (Cox et al., 2014) employed in MaxQuant calculates abundance
profiles for proteins across samples using peptide ion intensity ratios from the MS1
scan, to give ‘LFQ intensities’. We turned on the optional ‘match-between-runs’ feature
which increases the number of peptides used for quantification beyond those that have
been sequenced and identified in an MS2 scan. It allows the transfer of an MS2 peptide
identification from one sample to an unidentified peptide from another sample by
matching their MS1 scan masses and aligned retention times.
3.5 BioID to search for AP-4 cargo and machinery 113
Standard data filtering was applied as described in Section 2.15. Before statistical analysis
of the AP-4 BioID and control samples we performed control compression as described
by Lambert et al. (2015) to increase the stringency of data analysis. Each identified
protein had up to nine associated LFQ intensities from the three control cell lines, each in
triplicate. This control dataset was compressed to create a ‘worst-case scenario’ control
dataset by taking the three highest LFQ intensities for each protein and discarding the
rest. Log-transformed LFQ intensities from each AP-4 BioID pulldown (in triplicate) were
then compared to the compressed control dataset with a two-tailed t-test (in a separate
statistical analysis for each AP-4 subunit). Only proteins with LFQ intensities in all three
replicates of the AP-4 BioID pulldown were included in the analysis, leaving 3,100-3,700
proteins depending on the subunit. This meant that there was no data imputation for
the AP-4 samples and only a small amount for the control samples (only applied to
proteins which had fewer than three LFQ intensities across the nine control samples).
The imputation performed in Perseus software replaces missing values with random
values drawn from a distribution meant to simulate expression below the detection
limit of the MS (Tyanova et al., 2016). The results of the statistical analyses for each AP-4
subunit are represented graphically in Figure 3.18. Proteins are plotted on ‘volcano’ plots
according to their average log2 fold difference in abundance between the AP-4 BioID
pulldown and the control dataset and their t-test− log p -value. To deal with the problem
of multiple hypothesis testing in the t-tests significance cut-off lines corresponding to a
given FDR (here 0.05) were determined by a permutation-based method (Tusher et al.,
2001) implemented in Perseus software. In this approach data are randomly reshuffled
and the t-test is performed on the reshuffled data, and this process is repeated many
times (in this case 250 times). The FDR is estimated as the fraction of accepted hits from
the permuted data over accepted hits from the real data normalised by the total number
of randomisations (Tyanova & Cox, 2018). An additional parameter called s 0 is included
in defining the cut-off line, which controls the relative importance of the t-test p-value
and the difference between means. At s 0= 0 only the p-value matters, while at s 0> 0
the difference between means also plays a role (here s 0= 0.5).
Table 3.5 presents the fold enrichments and t-test p-values for proteins that were either
associated with AP-4 in other proteomic datasets or were called significant in more than
one AP-4 BioID dataset. All four AP-4 subunits showed strong enrichment in all four AP-4
BioID datasets (minimum 32-fold enrichment), validating the approach. TEPSIN was
also strongly enriched for three of the four subunits (minimum 15-fold enrichment), but
not for AP4S1. BioID is a proximity-based method so it is not unexpected to see different
results for the different AP-4 subunits. TEPSIN is known to bind to the AP4B1 and AP4E1
114 Proteomic investigations of adaptor protein complex 4
Fig. 3.18 AP-4 BioID to screen for AP-4-associated proteins using mass spectrometry (MS).
Volcano plots show comparison of protein abundance in affinity purifications of biotinylated
proteins from HeLa cells stably expressing AP4B1-BirA* (A), AP4E1-BirA* (B), AP4M1-BirA* (C),
or AP4S1-BirA* (D), and control cell lines (HeLa, HeLa BirA* and HeLa GFP-BirA), analysed by
label-free quantitative MS. Each experiment was performed in triplicate and the control dataset
was compressed to the three highest LFQ intensities per protein (i.e. high stringency). Data were
analysed with a two-tailed t-test: volcano lines indicate the significance threshold (FDR = 5%;
s 0= 0.5). Proteins of interest are labelled.
3.5 BioID to search for AP-4 cargo and machinery 115
appendage domains (Frazier et al., 2016; Mattera et al., 2015) and in keeping with this it
was most strongly enriched in the AP4B1 and AP4E1 datasets. ATG9A was significantly
enriched in the streptavidin pulldown from the AP4E1-BirA* cells (∼2-fold enrichment),
and was also enriched in the AP4M1-BirA* pulldown, but fell just below the significance
cut-offs. This is consistent with it being an AP-4 cargo protein, and in fact it was the
only transmembrane protein significantly enriched in any of the datasets. In contrast,
SERINC1 and SERINC3 were not present in any of the AP-4 BioID pulldowns. A possible
explanation for this became apparent when we looked at the predicted topologies of the
SERINCs and ATG9A (Figure 3.19). The TOPCONS web server for combined membrane
protein topology and signal peptide prediction (Tsirigos et al., 2015) and the TMHMM
Server for prediction of transmembrane helices in proteins (Sonnhammer et al., 1998)
predicted both SERINCs to have 11 transmembrane domains (TMDs) with luminal N
termini and cytosolic C termini. We went on to confirm these predictions experimentally
by the analysis of overexpressed tagged SERINC3 (see Section 3.5.3 below). Given these
predictions, both SERINC1 and SERINC3 protein sequences are dominated by TMDs
and they only have small cytosolic loops. Lysine residues are the target of activated
biotin and, as AP-4 is cytosolic, only the cytosolic lysines are potential biotinylation sites
for the AP-4 BioID proteins. SERINC1 has only five cytosolic lysines and SERINC3 has
just seven, thus they present a very limited amount of material for biotinylation. Also,
the lysines within the short cytosolic loops may not be accessible. ATG9A is a larger
protein and has 6 TMDs with both its N and C termini in the cytosol (Figure 3.19C). This
topology has been experimentally confirmed by Young et al. (2006). The majority of its
protein sequence is cytosolic and so it presents a lot more material (13 cytosolic lysines)
for potential biotinylation by the AP-4 BioID proteins.
RUSC2, like ATG9A, was significantly enriched in the AP4E1-BirA* streptavidin pulldown
(∼4-fold enrichment), and also fell just below the significance cut off in the AP4M1-BirA*
pulldown. This supports its candidacy as an AP-4 accessory protein. However, RUSC1
was not identified in any of the BioID pulldowns and as it is a cytosolic protein, this
cannot be explained by a lack of cytosolic material for biotinylation. This could suggest
RUSC1 does not directly interact with AP-4, but in general negative results in BioID
should be interpreted with caution. There are other possible explanations for a false
negative result and studies have shown that not all proteins are biotinylated with similar
efficiency (Kim et al., 2014).
In addition to the proteins that were identified as candidate AP-4-associated proteins in
the Dynamic Organellar Maps and vesicle fraction experiments, several other proteins
were consistent hits for more than one AP-4 subunit (Table 3.5). HOOK1 (Protein Hook
116 Proteomic investigations of adaptor protein complex 4
Tab
le
3.5
P
ro
tein
s
sign
ifi
can
tly
en
rich
ed
in
A
P
-4
B
io
ID
strep
tavid
in
p
u
lld
ow
n
s.
Fold
en
rich
m
en
ts
(A
P-4
B
ioID
p
u
lld
ow
n/con
trold
ataset)
an
d
t-test
p
-valu
es
(-lo
g)
fo
r
p
ro
tein
s
th
at
w
ere
eith
er
asso
ciated
w
ith
A
P-4
in
o
th
er
p
ro
teo
m
ic
d
atasets
o
r
w
ere
called
sign
ifi
can
t
in
m
o
re
th
an
o
n
e
A
P-4
B
io
ID
d
ataset.V
alu
es
are
sh
ad
ed
green
w
h
en
th
ey
co
rresp
o
n
d
to
sign
ifi
can
t
en
rich
m
en
t,red
w
h
en
th
e
valu
es
d
id
n
o
t
m
eet
th
e
sign
fi
can
ce
cu
t
o
ffs
(F
D
R
=
5%
;s0
=
0.5),an
d
grey
ifth
e
p
ro
tein
w
as
n
o
t
id
en
tifi
ed
in
th
at
d
ataset.
G
en
e
n
am
es
A
P
4B
1-B
irA
*
A
P
4E
1-B
irA
*
A
P
4M
1-B
irA
*
A
P
4S1-B
irA
*
Fo
ld
en
rich
m
en
t
−
lo
g
p
Fo
ld
en
rich
m
en
t
−
lo
g
p
Fo
ld
en
rich
m
en
t
−
lo
g
p
Fo
ld
en
rich
m
en
t
−
lo
g
p
A
P
4E
1
1852.0
5.3
15248.6
5.1
1009.0
5.2
401.9
4.7
A
P
4M
1
1027.0
4.1
1397.9
4.0
2950.5
4.0
249.1
3.7
A
P
4S1
66.0
3.3
70.4
2.6
32.8
2.9
1530.2
4.1
A
P
4B
1
1419.9
4.8
1063.9
3.9
391.8
4.2
118.3
4.1
H
O
O
K
1
15.0
2.1
52.8
2.4
218.4
3.2
18.2
1.9
T
E
P
SIN
104.1
3.7
66.9
3.0
14.6
3.1
1.2
0.1
A
K
T
IP
N
o
ID
N
o
ID
5.1
1.0
39.2
2.4
1.9
1.1
FA
M
160A
1
0.7
0.5
1.5
0.2
15.6
2.6
1.2
0.6
FA
M
160A
2
N
o
ID
N
o
ID
N
o
ID
N
o
ID
6.6
3.3
N
o
ID
N
o
ID
U
B
R
5
5.5
4.6
0.4
3.0
2.4
3.1
0.8
1.2
U
SP
47
5.3
3.5
0.6
0.4
5.2
2.3
1.2
0.2
C
A
C
Y
B
P
1.7
2.6
4.5
2.1
1.2
0.4
1.1
0.3
R
U
SC
2
N
o
ID
N
o
ID
4.3
2.5
1.3
1.3
N
o
ID
N
o
ID
SU
G
T
1
2.5
2.1
1.9
1.4
3.1
2.7
1.1
0.2
B
IR
C
6
2.6
2.3
0.9
0.1
2.8
1.9
0.6
1.0
D
LG
5
1.6
2.4
1.9
2.4
2.3
2.6
1.1
0.1
AT
G
9A
1.1
0.1
2.1
1.5
1.7
0.9
0.6
0.7
N
C
O
R
2
1.9
1.8
0.1
3.2
2.0
1.7
1.0
0.0
R
AV
E
R
1
1.7
1.3
0.8
0.3
1.8
1.6
0.8
0.8
3.5 BioID to search for AP-4 cargo and machinery 117
Fig. 3.19 Topology of SERINCs and ATG9A. The TOPCONS web server for combined membrane
protein topology and signal peptide prediction (Tsirigos et al., 2015) was used to predict the
topology of (A) SERINC1, (B) SERINC3 and (C) ATG9A. Both SERINCs were predicted to have 11
transmembrane domains (TMDs) with luminal N termini and cytosolic C termini. The TMHMM
Server for prediction of transmembrane helices in proteins (Sonnhammer et al., 1998) returned
very similar results with slight differences in the boundaries to the TMDs. ATG9A was predicted to
have 6 TMDs with both the N and C termini in the cytosol. This matches published experimental
confirmation of ATG9A topology by Young et al. (2006). In this case the TMHMM Server produced
conflicting results, only predicted 5 TMDs, but given the published evidence this is incorrect.
Cytosolic lysine residues (the target of activated biotin) are marked. SERINC1 has only 5, SERINC3
has 7 and ATG9A has 13.
118 Proteomic investigations of adaptor protein complex 4
homolog 1) was strongly enriched in all four AP-4 BioID datasets (minimum 15-fold
enrichment), with the highest level of enrichment in the AP4M1-BirA* pulldown (>200-
fold). HOOK1 is part of a complex called the FHF complex (Xu et al., 2008) which also
includes AKTIP (AKT-interacting protein) and FAM160A2 (FTS and Hook-interacting pro-
tein). AKTIP was significantly enriched in the AP4E1-BirA* and AP4M1-BirA* pulldowns
and FAM160A2, along with a related protein FAM160A1, was significantly enriched in the
AP4M1-BirA* pulldown. This clearly links AP-4 to the FHF complex. This interaction has
also been confirmed by co-immunoprecipitation (see Section 3.6). The other proteins
listed in Table 3.5 have not been identified in any other AP-4-related proteomic datasets,
but the fact they are biotinylated by multiple AP-4 subunits suggests they may be ‘real’
hits (discussed further in Section 3.8).
3.5.3 Reciprocal BioID with SERINC1 and SERINC3 as baits
We hypothesised that SERINC1 and SERINC3 were not biotinylated by BirA*-tagged AP-4
subunits due to their low number of cytosolic lysine residues. Thus, to test for proximity
of AP-4 and SERINCs in the other direction we performed the reciprocal experiment
using SERINC1 and SERINC3 as BioID baits.
In order for SERINCs to be able to biotinylate cytosolic AP-4 complexes the BirA* tags
would need to be positioned on the cytosolic side of the membrane. The topology
predictions discussed above (Figure 3.19) suggested that the SERINCs have their C
termini in the cytoplasm. To confirm this experimentally we analysed overexpressed
tagged SERINC3 (Figure 3.20). HeLa cells stably expressing SERINC3 with a C-terminal
HA tag (predicted to be cytosolic) followed by mCherry (generated by Dr Georg Borner)
were fixed in formaldehyde, with or without saponin permeabilisation, and labelled
with an antibody against HA. Immunofluorescence microscopy detected signals from
both mCherry and anti-HA for the permeabilised cells, but only from mCherry for the
non-permeabilised cells (Figure 3.20A). This showed the C-terminal HA epitope was
inaccessible without permeabilisation, i.e. it was on the cytosolic side of the membrane.
For further confirmation of the topology of SERINC3 we generated a second SERINC3
expression construct, this time placing the tag in a part of the protein that was predicted
to be extracellular. We chose to position an HA tag between residues L311 and A312 in
the predicted fourth luminal/extracellular loop. This position was chosen because the
equivalent position in SERINC5 has been confirmed to be extracellular (Usami et al.,
2015) and there is a conserved predicted glycosylation site (NST) at position 315. Gibson
assembly was used to splice an HA tag between the two segments of SERINC3 and place
3.5 BioID to search for AP-4 cargo and machinery 119
the construct into the retroviral vector pLXIN. Successful cloning was confirmed by
diagnostic digest and sequencing (data not shown). HeLa cells were transduced with
retrovirus and selected for stable expression of HA-tagged SERINC3 by addition of G418.
When these cells were analysed as described above for the cells expressing SERINC3-
HA-mCherry, HA was detected on the surface of unpermeabilised cells (Figure 3.20B).
This demonstrates that the position is indeed on the extracellular side of the membrane.
Based on this, SERINC1 and SERINC3 were C-terminally tagged with myc-BirA* and
cloned into a pLXIN retroviral vector using Gibson Assembly. Successful cloning of the
constructs was confirmed by diagnostic digests and sequencing (data not shown). Stable
HeLa cell lines for SERINC1-BirA* and SERINC3*-BirA were generated as described
for the AP-4 BioID cell lines in Section 3.5.1. Expression and promiscuous biotin ligase
activity of the BirA*-tagged SERINCs was confirmed by immunofluorescence microscopy
with an antibody against the myc epitope in the BirA* tag and fluorophore-conjugated
streptavidin (Figure 3.21). SERINC1-BirA* and SERINC3-BirA* had similar punctate
distributions throughout the cell, but with a concentration in the perinuclear region.
Expression levels were moderate and the majority of cells in the mixed populations
were myc-positive, so as for the AP-4 BioID cell lines it was possible to work with mixed
populations. Streptavidin labelling was also observed throughout the cell, including the
plasma membrane, and was brightest in the perinuclear region.
The SERINC BioID cell lines were then used in an MS-based screen for proximal proteins.
This was performed as described for the AP-4 BioID (Section 3.5.2), with the same control
cell lines (HeLa, HeLa BirA* and HeLa GFP-BirA*). The only difference was that SERINC1
and SERINC3 streptavidin pulldowns were performed on separate days, each with their
own set of three control pulldowns (each in triplicate). The MS raw files from all samples
were processed together in MaxQuant and label-free quantification was performed as
described for the AP-4 BioID. Following standard data filtering (see Section 2.15), log-
transformed LFQ intensities from all six SERINC BioID pulldowns were compared to a
compressed control dataset (consisting of the six highest control LFQ intensities for each
protein) with a two-tailed t-test. The SERINC1 and SERINC3 datasets were combined
for this analysis because according to the Dynamic Organellar Maps both SERINCs have
very similar subcellular localisations (Section 3.3.2), and so should have very similar
proximity interaction partners. For this analysis proteins were first filtered based on
a requirement for a minimum MS/MS count of one in all SERINC BioID pulldowns,
leaving 2,339 proteins. Due to this filtering missing data points only occurred in the
control dataset and missing values were imputed from a normal distribution in Perseus.
The results of this statistical analysis are represented graphically in Figure 3.22.
120 Proteomic investigations of adaptor protein complex 4
Fig. 3.20 Experimental confirmation of SERINC3 topology. (A) Widefield imaging of HeLa cells
stably expressing SERINC3 tagged C-terminally with HA-mCherry, fixed in formaldehyde, with
and without saponin permeabilisation, labelled with anti-HA. The Ha epitope is only accessible
with permeabilisation, i.e. it is on the cytosolic side of the membrane. Scale bar: 20 μm. (B)
Widefield imaging of HeLa cells stably expressing SERINC3 with an HA tag in a predicted extracel-
lular loop (between residues L311 and A312), fixed in formaldehyde, with and without saponin
permeabilisation, labelled with anti-HA. The HA epitope is accessible without permeabilisation,
i.e. it is on the extracellular side of the membrane. Scale bar: 20 μm.
3.5 BioID to search for AP-4 cargo and machinery 121
Fig. 3.21 SERINC BioID cell lines express active biotin ligases. Widefield imaging of HeLa cells
stably expressing (A) SERINC1-BirA* or (B) SERINC3-BirA* labelled with anti-myc (to detect the
BioID fusion protein) and streptavidin-568 (to detect biotinylated proteins). Scale bar: 20 μm.
122 Proteomic investigations of adaptor protein complex 4
Fig. 3.22 SERINC BioID to screen for SERINC-associated proteins using mass spectrometry
(MS). Volcano plot shows a comparison of protein abundance in affinity purifications of bi-
otinylated proteins from HeLa cells stably expressing SERINC1-BirA* or SERINC3-BirA*, and
control cell lines (HeLa, HeLa BirA* and HeLa GFP-BirA*), analysed by label-free quantitative MS.
Each SERINC BioID pulldown was performed in triplicate with its own set of control pulldowns
and the control dataset was compressed to the six highest LFQ intensities per protein (i.e. high
stringency). Data were analysed with a two-tailed t-test (with all six SERINC BioID pulldowns
combined; n = 6): the volcano line indicates the significance threshold (FDR= 5%; s 0= 0.5). AP-4
complex subunits and RUSC1 and RUSC2 were highly enriched in the SERINC BioID pulldowns.
Other AP complex subunits are marked for comparison.
3.5 BioID to search for AP-4 cargo and machinery 123
The first thing that is obvious when looking at Figure 3.22 is that a very large number
of proteins meet the cut off for significance (the same as that used for the AP-4 BioID)
- over 600 proteins were called significant hits for the SERINCs. This is a sign that the
control cell lines do not sufficiently control for background biotinylation by the BirA*-
tagged SERINCs. This became clear when hierarchical clustering was performed on the
LFQ intensities from the SERINC and AP-4 BioID experiments (Figure 3.23). Individual
pulldown samples were grouped blindly according to their LFQ intensities using the
euclidean distance method in Perseus software (Tyanova et al., 2016). For the AP-4 BioID
data, the AP-4 subunit samples clustered closely with the BirA* and GFP-BirA* control
samples, with a large distance to the HeLa samples (Figure 3.23A). Some of the AP4B1,
AP4M1 and AP4S1 samples actually clustered more closely with control samples than
their own replicates. This suggests a large amount of overlap between the proteins
biotinylated in the AP-4 and control BioID cell lines, i.e. background biotinylation is
suitably controlled. In contrast, for the SERINC BioID data there was a large distance
between the SERINC samples, which all clustered together, and the BirA* and GFP-BirA*
controls (Figure 3.23B). This distance was almost as large as that between the SERINC
samples and the HeLa controls. The most likely explanation for this is that the SERINCs
come into close proximity with a very different set of proteins from the cytosolic BioID
control proteins, because of their membrane environment. Unfortunately, this makes it
hard to distinguish proteins that are enriched because they closely associate with the
SERINCs from those that have been biotinylated in passing as the SERINCs are trafficked
around the cell.
Although the controls used in the SERINC BioID experiment were insufficient to identify
SERINC-associated proteins de novo, the data does support the close association of
SERINCs with AP-4 and the RUSCs. AP-1, AP-2, AP-3 and AP-4 complex subunits, and
RUSC1 and RUSC2, are marked on the volcano plot in Figure 3.22 and listed with their
associated fold enrichments and t-test p-values in Table 3.6. While some AP-1 and AP-2
complex subunits were significantly enriched in the SERINC BioID pulldowns (average
enrichment ∼2-fold), the AP-4 complex subunits and the RUSCs were enriched to a much
greater degree (minimum 20-fold for AP-4, ∼25-fold for RUSC1 and >80-fold for RUSC2).
Given the fact that AP-1 and AP-2 are roughly 40 times more abundant in HeLa cells
than AP-4 or the RUSCs (Itzhak et al., 2016), this is a stonking1level of enrichment. This
supports a major role for AP-4 and its associated machinery in the trafficking of the
SERINCs. The lower level of enrichment of AP-1 and AP-2 complex subunits suggests the
1While stonking is not a strictly scientific term, it often employed in the Robinson Lab to emphasise an
exciting result.
124 Proteomic investigations of adaptor protein complex 4
Fig. 3.23 Hierarchical clustering of AP-4 and SERINC BioID data. Hierarchical clustering was
performed in Perseus software using the euclidean distance method on LFQ intensities from (A)
AP-4 and (B) SERINC BioID experiments. The colour gradient goes from black for the lowest LFQ
intensities through blue, green, yellow, orange and red for the highest intensities. AP-4 subunit
samples are highlighted in yellow and SERINC1 and 3 samples are highlighted in blue. The AP-4
subunit samples cluster closely with the BirA* and GFP-BirA* control samples whereas there is a
large distance between the SERINC1 and 3 samples and the controls.
3.5 BioID to search for AP-4 cargo and machinery 125
Table 3.6 AP complex subunits, RUSCs and ATG9A in the SERINC BioID data. Fold enrich-
ments (SERINC BioID pulldowns/control dataset) and t-test p-values (-log) for AP complex
subunits, ATG9A, RUSC1 and RUSC2. TEPSIN was not included in the analysis because it did
not meet the requirement for a minimum MS/MS count of one in all SERINC BioID pulldowns.
Mean log2 LFQ intensities in the SERINC1 (n = 3), SERINC3 (n = 3) and compressed control (n =
6) datasets, and the minimum MS/MS count from the SERINC samples, are shown. Values are
shaded green when they correspond to significant enrichment and red when the values did not
meet the signficance cut offs (FDR = 5%; s 0= 0.5).
Gene
names
Fold
enrichment
−log p Mean
LFQ
SERINC1
Mean
LFQ
SERINC3
Mean
LFQ
control
Min.
MS/MS
count
RUSC2 83.5 3.3 33.8 33.5 27.3 28
AP4M1 27.0 7.0 32.7 31.8 27.5 12
RUSC1 25.4 5.5 30.9 31.6 26.6 7
AP4E1 23.6 6.9 33.7 33.4 29.0 25
AP4B1 21.7 9.4 33.1 32.8 28.5 20
AP4S1 20.4 5.5 28.5 29.4 24.6 2
AP1B1 4.5 5.1 30.8 29.9 28.2 5
ATG9A 3.1 3.0 30.9 30.5 29.1 5
AP2B1 2.8 5.0 33.6 32.9 31.8 23
AP2A1 2.4 3.3 33.4 32.4 31.7 17
AP2M1 2.3 3.6 32.9 32.0 31.2 11
AP1G1 1.5 1.9 30.7 30.1 29.8 8
AP2S1 1.3 0.5 29.5 27.9 28.3 1
AP3D1 0.9 0.3 31.4 31.1 31.4 4
AP3B1 0.9 0.3 31.2 30.9 31.1 3
AP3S1 0.9 0.5 27.9 27.8 28.0 1
AP1S1 0.9 0.4 27.8 27.0 27.6 2
AP3M1 0.8 1.6 30.2 30.0 30.4 5
SERINCs may also be trafficked by AP-1 and AP-2 during other parts of their trafficking
itinerary. Interestingly, SERINC1 and SERINC3 did not biotinylate each other, despite
the evidence for their closely related subcellular distribution (see Section 3.3.2). This
supports the hypothesis that the low number of cytosolic lysine residues in the SERINCs
hinders their identification by BioID. In contrast, ATG9A was significantly enriched in
the SERINC BioID pulldowns, although not as much as the RUSCs and AP-4 subunits
(∼3-fold enrichment; Figure 3.22; Table 3.6).
Taking all the BioID data together, it strongly supports a close association between AP-4,
ATG9A, SERINCs and RUSCs. It also provides evidence for an interaction between the
AP-4 complex and the FHF complex consisting of HOOK1, AKTIP and FAM160A1/2. In
126 Proteomic investigations of adaptor protein complex 4
addition it suggests additional candidates for AP-4-associated proteins which may prove
interesting for future investigations of AP-4 complex function (discussed in Section 3.8).
3.6 Sensitive immunoprecipitation to reveal AP-4 cargo
The combination of affinity purification and mass spectrometry (AP-MS) has provided a
wealth of information about protein complex composition (reviewed in Gingras et al.,
2007). In the AP-MS method, protein complexes are affinity purified from cell lysates and
complex components are identified by MS. Affinity purification can be achieved either
by traditional immunoprecipitation with a specific antibody against an endogenous
protein in a complex or by expression of a protein of interest with an epitope tag, e.g. a
FLAG tag or GFP, which is used as an affinity handle to purify the tagged protein, along
with its interacting partners. The AP-MS method has been successfully applied in a high-
throughput format to generate ‘interactomes’ for several organisms, from Saccharomyces
cerevisiae (e.g. Gavin et al., 2006; Krogan et al., 2006) to Homo sapiens (e.g. Hein et al.,
2015). However, one drawback to this method is that weak or transient protein-protein
interactions are unlikely to survive the lysis and purification conditions. For this reason
weak interactions are largely underrepresented in interactome studies, but weak does not
mean unimportant. Hein and colleagues (Hein et al., 2015) used quantitative proteomics
to survey the human proteome using HeLa cell lines expressing GFP-tagged proteins
under the control of their endogenous proteins. This revealed the protein network to be
dominated by weak interactions characterised by substoichiometric recovery.
The interaction partners of AP complexes, particularly cargo, have largely proved re-
fractory to affinity purification-based biochemical characterisation because they are
predominated by interactions that are weak or transient in nature (reviewed with regard
to AP-2 in Smith et al., 2017). For AP-4, this problem is exacerbated by low expression
levels of both itself and its interacting partners. Therefore, it is unsurprising that a con-
ventional AP-MS approach to identify AP-4 complex interacting partners (conducted by
Dr Georg Borner and discussed in Section 3.6.1) did not reveal interactions between AP-4
and ATG9A, SERINCs or RUSCs. To attempt to preserve these weak interacting partners,
we developed a sensitive immunoprecipitation approach using greatly reduced levels of
detergent and applied it to identify AP-4 interacting partners co-immunoprecipitated
with TEPSIN-GFP.
Detergents are ampiphatic molecules with hydrophilic head groups and hydrophobic
tails. Due to this property, they can insert their tails within the lipid bilayer of cellular
3.6 Sensitive immunoprecipitation to reveal AP-4 cargo 127
membranes, thereby disrupting the membrane and extracting membrane-embedded
proteins. In aqueous solution detergents spontaneously form spherical structures known
as ‘micelles’, to shield their hydrophobic tails. This provides a mimic of the lipid bilayer
environment allowing the extraction of membrane proteins as part of a detergent-protein
complex. Micelles only form above a certain concentration threshold called the critical
micelle concentration (CMC); when a detergent is below its CMC value it will mostly be
present as monomers. Although these properties are favourable for membrane protein
extraction, detergent can disrupt protein-protein interactions, so a compromise must be
found between protein extraction and complex stability. We reasoned at detergent levels
just above the CMC (0.2 mM for Triton TX-100; Tiller et al., 1984) it might be possible
to maintain interactions between AP-4 and its transmembrane cargo proteins, while
still allowing their solubilisation from the membrane. The molecular weight of Triton
TX-100 is 625 g/mol, so 0.2 mM is the equivalent of 0.014% w/v Triton TX-100. Thus,
we decided to trial immunoprecipitations with 0.01% (just below the CMC) and 0.025%
Triton TX-100.
3.6.1 Identification of AP-4 interacting partners by MS analysis of
TEPSIN-GFP immunoprecipitations
AP-MS was previously carried out by Dr Georg Borner to identify AP-4 complex interact-
ing partners co-precipitated with GFP-tagged TEPSIN. Wild-type HeLa cells and HeLa
cells stably expressing TEPSIN-GFP (Borner et al., 2012) were each grown in two 500 cm2
dishes in SILAC light or heavy medium. The experiment was performed in duplicate with
a label-swap. Lysis was performed in a relatively gentle buffer consisting of 0.2% Triton
X-100 and 0.1% Tween-20 (in PBS) and TEPSIN-GFP was immunoprecipitated by the ad-
dition of a polyclonal antibody against GFP, followed by capture on Protein A sepharose
beads. The beads were washed gently in PBS with 0.1% Tween-20 before immunopre-
cipitates were recovered in a soft elution buffer designed to minimise the co-elution
of immunoglobulin (Antrobus & Borner, 2011). Eluates from paired SILAC-labelled
mock and TEPSIN-GFP immunoprecipitations were then combined and relative protein
enrichment in the TEPSIN-GFP samples was measured using SILAC-based quantitative
MS.
The resulting data was filtered in the standard way (Section 2.15), before additional
filtering to remove unlabelled proteins and those without at least one H/L ratio count
in each replicate experiment. This left 585 proteins that had been identified and quan-
tified in both replicates. To identify proteins that were enriched in the TEPSIN-GFP
128 Proteomic investigations of adaptor protein complex 4
immunoprecipitations, log-transformed SILAC ratios (TEPSIN-GFP/mock) from the
two replicate experiments were plotted against each other (Figure 3.24A). The most
strongly enriched proteins were TEPSIN and the AP-4 subunits themselves, followed by
HOOK1, which was also a strong hit in the AP-4 BioID experiments (Section 3.5.2). As
previously discussed, HOOK1 is part of a protein complex called the FHF complex, along
with FAM160A2 which was also enriched in the TEPSIN-GFP immunoprecipitations.
Other proteins that were enriched at least 2-fold in both replicate immunoprecipitations,
mostly heat shock proteins, are listed in Table 3.7. ATG9A, SERINC1/3 or RUSC1/2 were
not identified in this experiment.
For the sensitive, low-detergent immunoprecipitations, wild-type HeLa cells and HeLa
cells stably expressing TEPSIN-GFP were each grown in two 10 cm dishes in SILAC light
or heavy medium. The experiment was performed in triplicate, including a label swap,
for 0.01% and 0.025% Triton TX-100 conditions. Cells were homogenised mechanically
in PBS, before the addition of Triton TX-100 for protein extraction at a final concentration
of 0.01% or 0.025%. TEPSIN-GFP was immunoprecipitated using GFP-Trap A beads
(ChromoTek) and immunoprecipitates were washed in 0.01% Triton TX-100 (for both
conditions) before recovery in an SDS elution buffer (2.5%). Eluates from paired SILAC-
labelled mock and TEPSIN-GFP immunoprecipitations were then combined and relative
protein enrichment in the TEPSIN-GFP samples was measured using SILAC-based
quantitative MS, as for the conventional TEPSIN-GFP immunoprecipitations.
Data filtering was performed as described for the conventional immunoprecipitations.
This left 1128 proteins for the 0.01% condition and 993 proteins for 0.025%. To identify
proteins that were enriched in the TEPSIN-GFP immunoprecipitations, the two deepest
datasets (giving the most robust quantification) for each condition were selected and log-
transformed SILAC ratios (TEPSIN-GFP/mock) were plotted against each other in Figure
3.24B and C. In comparison to the conventional TEPSIN-GFP immunoprecipitations
shown in Figure 3.24A, the sensitive immunoprecipitations had more background, as vi-
sualised by the large clouds of proteins extending into the top-right portions of the plots.
A large number of proteins were enriched up to 4-fold in the TEPSIN-GFP immunopre-
cipitations. Therefore, we decided to just focus on the proteins that were clear outliers
on both plots. These included TEPSIN, the four AP-4 complex subunits and HOOK1,
which were enriched in the conventional immunoprecipitations, but also ATG9A and
SERINC1. SERINC3 was enriched in the 0.01% immunoprecipitations but doesn’t appear
on the plot for the 0.025% immunoprecipitations because it was not identified in one
replicate experiment, so was filtered out from the dataset. However, in the two other
replicates it was enriched (Table 3.8). Only a few other proteins were enriched to a similar
3.6 Sensitive immunoprecipitation to reveal AP-4 cargo 129
Fi
g.
3.
24
C
o
n
ve
n
ti
o
n
al
an
d
se
n
si
ti
ve
im
m
u
n
o
p
re
ci
p
it
at
io
n
s
o
f
T
E
P
SI
N
-G
F
P.
A
P
-4
co
m
p
le
xe
s
an
d
th
ei
r
in
te
ra
ct
in
g
p
ar
tn
er
s
w
er
e
co
-
im
m
u
n
o
p
re
ci
p
it
at
ed
w
it
h
T
E
P
SI
N
-G
F
P
fr
o
m
H
eL
a
ce
lls
.I
m
m
u
n
o
p
re
ci
p
it
at
io
n
s
w
er
e
p
er
fo
rm
ed
w
it
h
(A
)
a
co
n
ve
n
ti
o
n
al
p
ro
to
co
lo
r
(B
)
an
d
(C
)
a
se
n
si
ti
ve
im
m
u
n
o
p
re
ci
p
it
at
io
n
p
ro
to
co
lw
it
h
a
ve
ry
lo
w
le
ve
lo
fd
et
er
ge
n
t,
0.
01
%
T
X
-1
00
in
B
an
d
0.
02
5%
in
C
.E
ac
h
sc
at
te
r
p
lo
t
sh
ow
s
tw
o
re
p
li
ca
te
SI
L
A
C
co
m
p
ar
is
o
n
s
o
fT
E
P
SI
N
-G
F
P
im
m
u
n
o
p
re
ci
p
it
at
es
ve
rs
u
s
m
o
ck
im
m
u
n
o
p
re
ci
p
it
at
es
fr
o
m
p
ar
en
ta
lH
eL
a
ce
lls
.
T
E
P
SI
N
-G
F
P
as
so
ci
at
ed
p
ro
te
in
s
h
av
e
h
ig
h
ra
ti
o
s.
130 Proteomic investigations of adaptor protein complex 4
Table 3.7 AP-4 interacting proteins from a conventional immunoprecipitation (IP) of TEPSIN-
GFP. Proteins that were at least 2-fold enriched in the TEPSIN-GFP IPs relative to mock IPs from
wild-type HeLa cells are listed with their fold enrichment in each replicate and their minimum
H/L ratio count across the two replicates. AP-4 core complex proteins are marked in yellow,
components of the FHF complex are marked in green and heat shock proteins are marked in red.
Gene name
Fold enrichment in TEPSIN-GFP
Min. ratio count
IP 1 IP 2
AP4B1 47.3 13.0 149
AP4E1 46.7 17.4 100
AP4M1 46.4 11.2 75
AP4S1 46.3 14.1 17
TEPSIN 43.7 44.0 118
HOOK1 25.1 24.4 68
HSPA1A 17.4 20.5 246
FAM160A2 14.0 11.4 4
HSPA6 11.3 12.1 5
HSPA8 10.8 9.7 254
NKRF 8.9 4.9 2
HSPA2 8.4 9.4 7
HSPH1 5.0 4.4 36
XRN2 4.6 3.7 4
HSPA4L 4.4 5.3 11
DHX30 3.9 2.1 2
ERGIC1 3.6 3.4 2
HSPA5 3.4 2.5 162
UTS2 3.2 2.7 3
DHX15 3.0 2.5 9
SEC16A 2.7 2.9 14
CUL2 2.2 3.0 1
HSPA4 2.1 5.2 15
3.6 Sensitive immunoprecipitation to reveal AP-4 cargo 131
degree: AAGAB (Alpha- and gamma-adaptin-binding protein p34; only for 0.01%), a
known binder of AP-1 and AP-2 complexes; CCDC88B (Coiled-coil domain-containing
protein 88B), related to the Hook family of proteins; PLK1 (Serine/threonine-protein
kinase PLK1); DAGLB (Sn1-specific diacylglycerol lipase beta); and HSPH1 (Heat shock
protein 105 kDa). CCDC88B was only identified by post-translational modifications, and
would usually be excluded from the analysis on this grounds, but it was left in the dataset
because of its relation to the Hook family of proteins and the fact it was also strongly
enriched in one replicate of the conventional TEPSIN-GFP immunoprecipitation (but
not identifed in the other replicate). These proteins are listed in Table 3.8 along with their
fold enrichments in all replicates and conditions of sensitive immunoprecipitations.
SERINC3 was not enriched in the 3rd replicate of the 0.01% immunoprecipitation, but
this ratio was calculated from a single peptide identification so is less robust than the
quantification from the other two replicates which used three independent peptides.
The 3rd replicates of the 0.01% and 0.025% immunoprecipitations had a tendency for
lower TEPSIN-GFP/mock ratios than the 1st and 2nd replicates. These were the label
swap experiments where the TEPSIN-GFP expressing cells were grown in SILAC heavy
medium. The lower ratios may indicate that the metabolic labelling was incomplete.
There was a trend for the interactions that were not observed in the conventional im-
munoprecipitations (AAGAB, ATG9A, SERINC1/3, DAGLB) to be identified with fewer
peptides, or not identified at all, with the increased amount of detergent in the 0.025%
immunoprecipitations. This demonstrates that even at very low levels of detergent, the
interactions between AP-4 and its cargo proteins begin to be lost.
Thus, our sensitive immunoprecipitation approach revealed candidate AP-4 interact-
ing proteins that were missed by a more conventional AP-MS approach and provided
orthogonal confirmation that ATG9A, SERINC1 and SERINC3 are AP-4 cargo proteins.
The interactions between AP-4 and RUSC1 and RUSC2, on the other hand, were not
identified using this approach.
132 Proteomic investigations of adaptor protein complex 4
Tab
le
3.8
A
P
-4
in
teractin
g
p
ro
tein
s
fro
m
sen
sitive
im
m
u
n
o
p
recip
itatio
n
s
(IP
s)
o
f
T
E
P
SIN
-G
F
P.Sen
sitive
im
m
u
n
o
p
recip
itatio
n
s
w
ere
p
erfo
rm
ed
w
ith
0.01%
o
r
0.025%
T
X
-100
(each
in
trip
licate).
P
ro
tein
s
th
at
w
ere
clear
o
u
tliers
in
F
igu
re
3.24B
an
d
C
are
listed
w
ith
th
eir
fo
ld
en
rich
m
en
t
in
each
exp
erim
en
t
an
d
th
eir
m
in
im
u
m
H
/L
ratio
co
u
n
t
(R
C
)
fro
m
rep
licate
exp
erim
en
ts.
G
rey
sh
ad
in
g
in
d
icates
n
o
id
en
tifi
cation
ofth
e
p
rotein
in
th
atexp
erim
en
t.SE
R
IN
C
3
h
ad
a
low
ratio
in
on
e
rep
licate
ofth
e
0.01%
IP
(sh
ad
ed
red
),b
u
tth
is
w
as
calcu
lated
fro
m
a
sin
gle
p
ep
tid
e
id
en
tifi
catio
n
.
G
en
e
n
am
es
0.01%
IP
0.025%
IP
1
2
3
M
in
.R
C
1
2
3
M
in
.R
C
A
P
4S1
111.2
20.7
16.3
9
29.5
50.5
13.5
8
A
P
4B
1
52.5
74.4
14.8
22
20.5
53.6
11.7
18
A
P
4E
1
37.7
45.5
16.7
25
22.6
42.8
13.6
18
A
P
4M
1
28.7
21.8
12.8
23
35.4
65.1
10.3
22
A
A
G
A
B
19.2
12.5
6.2
1
N
o
ID
N
o
ID
N
o
ID
N
o
ID
T
E
P
SIN
18.2
9.2
10.7
45
10.4
13.0
8.1
36
C
C
D
C
88B
15.9
12.9
16.3
3
6.6
31.5
14.1
3
H
O
O
K
1
10.4
19.6
9.4
20
7.4
18.4
7.8
11
AT
G
9A
10.3
16.9
9.5
8
2.7
13.4
8.8
5
P
LK
1
9.4
24.1
2.9
2
7.1
2.4
3.1
2
SE
R
IN
C
3
9.3
10.6
0.6
1
3.5
4.4
N
o
ID
0
SE
R
IN
C
1
9.1
22.1
3.6
2
7.8
6.2
2.5
1
D
A
G
LB
8.9
15.5
7.6
2
3.6
8.8
4.9
1
H
SP
H
1
7.3
12.3
9.6
7
6.7
6.6
7.2
9
3.7 Additional proteomic analyses of AP-4 knockout cells 133
3.7 Additional proteomic analyses of AP-4 knockout cells
In addition to our organellar, spatial and interaction-focused proteomic studies, we also
conducted quantitative proteomic analyses of the whole cell and membrane-associated
proteomes of AP-4 knockout HeLa cells.
3.7.1 Global proteome analysis of AP-4 knockout cells
A SILAC-based quantitative proteomics approach was taken to look for global changes
in protein abundance in the AP-4 knockout HeLa cells. Whole cell lysates were harvested
from AP4B1 and AP4E1 HeLa cells grown in SILAC light medium and wild-type HeLa
cells grown in SILAC heavy medium, in triplicate for each cell line. Each light knockout
lysate was mixed with an equal protein amount of heavy wild-type lysate and subjected
to in-solution tryptic digest. Peptides were then purified and fractionated for deep MS
analysis. MaxQuant processing of the raw files generated a list of identified proteins
with up to six H/L ratios of relative abundance (three for wild-type/AP4B1 knockout
and three for wild-type/AP4E1 knockout). Following standard data filtering (Section
2.15), proteins were further filtered on a requirement for at least two H/L ratios for each
knockout, leaving 6,841 proteins. Normalised and log-transformed ratios were inverted
to L/H so depletion from the knockout is represented by a negative ratio. A one-sample
t-test was then applied to look for proteins with significant depletion or enrichment
in the knockout cells. The results of this statistical analysis are represented graphically
in Figure 3.25. To control for multiple hypothesis testing, the FDR was estimated by
applying the same statistical analysis to a mock experiment (light labelled HeLa versus
heavy labelled HeLa lysates) in which no real changes in abundance were expected
(mock experiment analysis was performed by Dr Georg Borner). This is conceptually
similar to the approach used for protein translocation in Section 3.3.1. At a cut-off of
p ≤ 0.02 and a minimum absolute log2 fold change of 0.45 the FDR was 25%. This is
relatively high but because the statistical power is low in a deep experiment with small
fold changes, reducing the tolerance for false discoveries would come at the cost of
missing real discoveries. As we were mostly interested in the behaviour of our newly
identified AP-4-associated proteins, we could tolerate a high FDR.
AP4E1 and AP4M1 were among the most depleted proteins in the AP-4 knockout cells.
There are two possible explanations why we see SILAC ratio for AP4E1 from the AP4E1
knockout cells, when we would expect no light-labelled AP4E1 peptides at all. One
possibility is that the heavy labelling in the wild-type cells is not 100% efficient, meaning
134 Proteomic investigations of adaptor protein complex 4
Fig. 3.25 Global proteome analysis of AP-4 knockout (KO) HeLa cells. Whole cell lysates from
light SILAC-labelled AP4B1 and AP4E1 KO HeLa cells were analysed by SILAC-based quantitative
mass spectrometry, in comparison to lysates from heavy SILAC-labelled wild-type cells (in
triplicate for each KO line). More than 6,500 proteins were quantified. Data were analysed with
a two-tailed one sample t-test. Significance cut-offs were defined as p ≤ 0.02 and a minimum
absolute fold change (log2) of 0.45, with an estimated FDR of 25 %. Proteins marked with black
or coloured circles meet these criteria, except ATG9A (pink) and the FHF complex proteins (red).
3.7 Additional proteomic analyses of AP-4 knockout cells 135
that some light proteins originate from the wild-type cells. Alternatively, Maxquant has
a feature called ‘re-quantify’ which is employed when only one isotope pattern from a
SILAC pair is identified. In these instances the MaxQuant algorithm integrates the inten-
sity from the position in the m/z retention time plane where the SILAC partner would
be expected to be, to avoid losing all quantification information about the identified
protein. If the SILAC partner is really absent, then the computed intensity will just be
the background noise level. In this case the latter explanation applies, because all the
ratios for AP4E1 from the AP4E1 knockout were quantified with the re-quantify method.
AP4S1 and RUSC2 were not covered by the analysis. AP4B1 was filtered out from the
dataset because it only had one H/L ratio for each knockout cell line, but those ratios
showed strong depletion in both cell lines. TEPSIN was also filtered out of the dataset
because it was only identified in one sample, and only by a single peptide. RUSC1 was
also among the most depleted proteins (∼3-fold depletion), but this was a lower level of
depletion than that from the vesicle-enriched fraction of the AP-4 knockout cells (mini-
mum 15-fold depletion; see Table 3.4). However, this could be due to ratio compression
because of use of the re-quantify method. The quantification from the vesicle-enriched
fraction should be more accurate as RUSC1 is highly enriched in the fraction. SERINC1
and SERINC3 were both depleted around 1.5-fold in the AP-4 knockout cells. In this
case the majority of their depletion from the vesicle fraction of AP-4 knockout cells may
be explained by depletion at the whole cell level. However, AP-4 knockdown caused
greater loss of SERINCs from the vesicle fraction than AP-4 knockout, and data from
cells expressing endogenously tagged SERINCs suggests that with AP-4 knockdown the
loss of SERINCs from the vesicle fraction occurs without an obvious difference in ex-
pression at the whole cell level (see Section 4.4.4). The expression level of ATG9A was
not significantly altered in the AP-4 knockout cells, but it was slightly enriched in the
AP4B1 knockout cell lysates, suggesting if anything its loss from the vesicle fraction
may be slightly underestimated. Components of the FHF complex (HOOK1, AKTIP and
FAM160A1) showed no differences in expression level between the wild-type and AP-4
knockout cells.
Other proteins that met the cut-offs for significant depletion or enrichment in the knock-
out cells are listed in Tables 3.9 and 3.10, respectively. These proteins were additionally
filtered to remove proteins that were also variable in the wild-type versus wild-type mock
experiment (variable defined as a median absolute fold change of ≥ 1.2). As the FDR
in this experiment was 25%, it is likely these lists include a number of false positives.
However, some proteins are of particular interest. TRIM33 (E3 ubiquitin-protein lig-
ase TRIM33) was one of the most strongly depleted proteins from the AP-4 knockout
136 Proteomic investigations of adaptor protein complex 4
cells. The TRIM family have an emerging role in autophagy as dual-function autophagy
receptors and regulators (reviewed in Kimura et al., 2016). Although TRIM33 has not
been directly linked to autophagy this could be an interesting connection. There is also
an over-representation of mitochondrial proteins among the list of depleted proteins
- MFF (Mitochondrial fission factor) and nine components of the mitochondrial large
ribosomal subunit (MRPL16/18/20/21/33/34/35/42/52). Two proteins that jump out
from the table of enriched proteins are MAP1LC3B (Microtubule-associated proteins
1A/1B light chain 3B), the autophagy marker protein from here on referred to as ‘LC3B’,
and GABARAPL2 (Gamma-aminobutyric acid receptor-associated protein-like 2). Both
are homologs of yeast Atg8 which is an ubiquitin-like protein required for the formation
of autophagosomal membranes. The elevated level of LC3B in the AP-4 knockout HeLa
cells was the first indicator that the mislocalisation of ATG9A may have downstream
effects on autophagy. This is investigated further in Chapter 4 Section 4.6.
3.7.2 The membrane fraction of AP-4 knockout cells
Total membrane fractions from AP-4 knockout cells were analysed by MS with label-free
quantification, in comparison to total membrane fractions from wild-type cells. The
membrane fractions were prepared during the preparation of the Dynamic Organellar
Maps (Section 3.3.1; described as ‘reference’ membrane fractions), but with an extra
replicate to give three biological replicates for each cell line (3 x AP4B1 knockout, 3 x
AP4E1 knockout and 3 x wild-type). MaxQuant processing of the raw MS data files yielded
a list of identified proteins, each with up to nine associated LFQ intensity measurements
of relative protein abundance. Following standard data filtering (Section 2.15), the data
was filtered stringently to leave only proteins that had the full nine LFQ intensities
(no missing datapoints were allowed). This left a total of 6,653 proteins that had been
identified and quantified in every sample. The log-transformed LFQ intensities from
the six AP-4 knockout membrane fractions were compared to those from the three
wild-type membrane fractions using a two-tailed t-test. The results of this statistical
analysis are represented graphically in Figure 3.26. This time a permutation-based FDR
estimation was used to control for multiple hypothesis testing, as was used for the BioID
experiments (see Section 3.5.2) and with the same cut-off of an FDR of 5% with s 0= 0.5.
The only significantly affected protein using this cut-off was RUSC1, which was depleted
around four-fold from the membrane fraction of the AP-4 knockout cells. This is greater
than the level of depletion seen for RUSC1 from the AP-4 knockout whole cell lysates.
However, the quantification of RUSC1 in the whole cell proteome of the knockout cells
3.7 Additional proteomic analyses of AP-4 knockout cells 137
Table 3.9 Proteins depleted from whole cell lysates of AP-4 knockout (KO) cells. A two-tailed
one-sample t-test was used to identify proteins that were significantly depleted from light-
labelled AP4B1 and AP4E1 KO cells in comparison to heavy-labelled wild-type cells (in triplicate
for each KO). Proteins that met the t-test significance cut-offs of p≤ 0.02 and a minimum log2 fold
depletion of 0.45 (FDR = 25%) are listed with their fold depletion (non-log) in each experiment,
median fold depletion and t-test − log p -value. AP-4-associated proteins are shaded yellow. Grey
shading indicates no identification of the protein in that experiment.
Gene
names
Fold depletion: B1 KO Fold depletion: E1 KO Median −log p
1 2 3 1 2 3
AP4M1 6.2 4.6 1.7 3.3 3.9 1.7
RUSC1 2.7 4.1 3.9 2.7 3.2 2.7
AP4E1 2.0 1.9 2.4 5.2 4.7 2.4 2.1
TRIM33 5.3 2.7 1.1 1.8 1.6 3.9 2.2 1.8
SNCA 2.2 2.3 1.8 2.1 1.4 1.3 1.9 2.8
FOLR1 2.2 2.1 2.6 1.5 1.8 1.8 1.9 3.6
HLA-B 1.9 6.6 6.1 1.8 1.7 1.7 1.8 1.8
PSMB10 2.8 1.4 1.7 1.7 1.9 1.7 2.2
TFE3 1.2 1.2 1.8 1.7 2.0 1.7 1.7
MFF 1.4 1.5 1.6 1.8 2.0 2.1 1.7 3.4
PDE3A 2.5 2.1 1.9 1.4 1.3 1.1 1.6 2.0
SERINC1 1.2 1.5 1.6 1.7 2.0 1.8 1.6 2.8
ATAD3B 2.0 2.0 2.1 1.3 1.3 1.1 1.6 2.0
MBNL3 2.3 1.9 2.0 1.2 1.4 1.3 1.6 2.2
MYLK3 1.5 1.4 1.5 2.1 1.5 1.2 1.5 2.6
NQO1 1.4 1.4 1.2 1.5 1.5 1.7 1.5 3.4
MRPL16 1.3 1.4 1.6 1.2 2.3 1.5 1.5 2.1
MRPL42 1.6 1.4 1.6 1.4 1.3 1.5 1.4 4.0
MRPL34 1.8 1.2 1.4 1.5 1.4 1.5 1.4 3.0
HLA-A 1.7 1.9 1.8 1.1 1.2 1.2 1.4 1.8
SERINC3 1.1 1.5 1.4 1.0 1.6 1.8 1.4 1.7
PLIN2 1.2 1.7 1.6 1.2 1.9 1.1 1.4 1.9
MRPL33 1.4 2.0 1.4 1.3 1.2 1.5 1.4 2.4
MRPL35 1.4 1.5 1.2 1.2 1.5 1.4 2.4
LIFR 1.3 1.5 1.8 1.1 1.2 1.4 1.4 2.5
CSRP2 1.5 1.6 1.1 1.4 1.1 1.4 1.4 2.3
B2M 1.6 1.6 1.5 1.1 1.1 1.3 1.4 2.0
MRPL21 1.5 1.4 1.5 1.1 1.2 1.3 1.4 2.6
ABHD14B 1.3 1.4 1.2 1.4 1.6 1.4 1.4 3.5
LAMA1 1.5 1.4 1.5 1.2 1.3 1.0 1.4 2.2
FTHL17 1.3 1.4 1.8 1.4 1.4 1.9
MRPL20 1.4 1.3 1.5 1.3 1.1 1.5 1.4 3.1
MRPL18 1.4 1.5 1.4 1.1 1.1 1.4 1.4 2.6
MRPL52 1.6 1.3 1.4 1.4 1.1 1.4 1.4 2.7
138 Proteomic investigations of adaptor protein complex 4
Tab
le
3.10
P
ro
tein
s
en
rich
ed
in
w
h
o
le
celllysates
o
fA
P
-4
kn
o
cko
u
t(K
O
)cells.A
tw
o-tailed
on
e-sam
p
le
t-testw
as
u
sed
to
id
en
tify
p
rotein
s
th
at
w
ere
sign
ifi
can
tly
en
rich
ed
in
ligh
t-lab
elled
A
P
4B
1
an
d
A
P
4E
1
K
O
cells
in
co
m
p
ariso
n
to
h
eavy-lab
elled
w
ild
-typ
e
cells
(in
trip
licate
fo
r
each
K
O
).P
ro
tein
s
th
at
m
et
th
e
t-test
sign
ifi
can
ce
cu
t-o
ffs
o
fp≤
0.02
an
d
a
m
in
im
u
m
lo
g
2
fo
ld
en
rich
m
en
t
o
f0.45
(F
D
R
=
25%
)
are
listed
w
ith
th
eir
fold
en
rich
m
en
t(n
on
-log)
in
each
exp
erim
en
t,m
ed
ian
fold
en
rich
m
en
tan
d
t-test−
lo
g
p
-valu
e.G
rey
sh
ad
in
g
in
d
icates
n
o
id
en
tifi
catio
n
o
fth
e
p
ro
tein
in
th
atexp
erim
en
t.
G
en
e
n
am
es
Fo
ld
en
rich
m
en
t:B
1
K
O
Fo
ld
en
rich
m
en
t:E
1
K
O
M
ed
ian
−
lo
g
p
1
2
3
1
2
3
A
R
H
G
D
IB
1.8
2.2
1.9
2.7
2.4
2.0
2.1
4.1
N
T
5E
1.3
1.9
2.1
1.5
3.8
1.5
1.7
2.0
SE
LH
2.1
1.5
1.6
2.0
1.3
1.6
2.2
C
N
N
M
2
1.6
1.9
1.8
1.2
1.6
1.4
1.6
2.8
N
M
E
4
1.5
1.7
1.5
2.0
1.2
1.5
2.3
M
A
P
1LC
3B
1.7
1.6
1.5
1.3
1.4
1.5
3.1
LR
B
A
1.5
1.5
1.7
1.5
1.5
3.2
SE
Z
6L2
1.5
1.6
1.3
1.2
1.5
1.5
2.6
SLC
7A
2
1.5
1.6
1.5
1.3
1.4
1.5
1.5
4.1
G
P
R
C
5A
1.5
1.6
1.4
1.5
1.4
1.2
1.4
3.6
M
A
R
V
E
LD
1
1.5
1.7
1.6
1.3
1.1
1.0
1.4
2.0
G
A
B
A
R
A
P
L2
1.6
1.6
1.7
1.2
1.2
1.1
1.4
2.1
SLF
N
5
1.4
1.5
1.3
1.4
1.3
2.5
1.4
2.1
C
D
IP
T
1.5
1.6
1.8
1.1
1.3
1.1
1.4
2.0
D
D
B
2
1.5
1.4
1.6
1.0
1.3
1.1
1.4
1.7
W
N
K
2
1.2
1.3
1.6
1.2
1.4
2.0
1.4
2.2
Z
FA
N
D
5
1.5
1.2
1.2
1.4
1.4
1.4
1.4
3.2
3.7 Additional proteomic analyses of AP-4 knockout cells 139
Fig. 3.26 The membrane fraction of AP-4 knockout (KO) HeLa cells. Comparison of protein
abundance in total membrane fractions prepared from AP-4 KO (AP4B1 and AP4E1, each in
triplicate, n = 6) and wild-type HeLa cells (in triplicate, n = 3), analysed by label-free quantitative
MS. More than 6,600 proteins were quantified; RUSC1 was the only protein significantly depleted
from the membrane fraction of AP-4 KO cells (RUSC2 and AP-4 subunits were not consistently
detected). Data were analysed with a two-tailed t-test: volcano lines indicate the significance
threshold (FDR = 5%; s 0= 0.5).
140 Proteomic investigations of adaptor protein complex 4
relied on the re-quantify algorithm, so the level of depletion may be underestimated as no
light-labelled RUSC1 peptides were identified by MS. Regardless, this evidence suggests
that RUSC1 requires AP-4 for its recruitment to the membrane and is destabilised at the
whole cell level in the absence of AP-4, supporting its candidacy as an AP-4 accessory
protein. RUSC2 was not identified, perhaps reflecting its lower whole cell copy number
(Itzhak et al., 2016). The AP-4 accessory protein TEPSIN is known to require AP-4 for its
recruitment to membrane (Borner et al., 2012). In this experiment it was depleted from
the membrane fraction from the AP-4 knockout cells by ∼1.4-fold, but did not meet the
cut-off for significance. It is possible this level of depletion is underestimated though,
because quantification in the AP-4 knockout membrane fractions relied on the match-
between-runs feature. No peptides for TEPSIN were actually identified by MS/MS in the
knockout membrane fractions. In contrast, the quantification of the membrane proteins
SERINC1, SERINC3 and ATG9A was robust, with a minimum MS/MS ratio count of
six, four and twelve, respectively. SERINC1 was slightly depleted from the membrane
fraction of AP-4 knockout cells (by ∼1.2-fold), while SERINC3 was unaffected and ATG9A
was slightly enriched (1.6-fold). However, these changes did not meet the cut-off for
significance.
The most enriched protein in the AP-4 knockout membrane fractions was MAGEB2
(Melanoma-associated antigen B2), which was enriched more than 4-fold. Although
this fell just outside the cut-off for significance, the quantification was based on robust
MS/MS peptide identifications. MAGEB2 belongs to the melanoma antigen gene (MAGE-
I) family of tumour-specific antigens and has been reported to enhance the activity of
E2F transcription factors, thereby promoting tumour cell proliferation (Peche et al.,
2015). The gene family is poorly characterised functionally but has been implicated in
binding to E3 RING ubiquitin ligases including members of the TRIM family, which, as
mentioned before, have an emerging role in autophagy (Doyle et al., 2010). Also among
the most strongly enriched proteins was ARHGDIB (Rho GDP-dissociation inhibitor
2; >2-fold enrichment). This was enriched to a similar degree in the AP-4 knockout
whole cell lysates, suggesting this difference is real, and Rho family GTPases have been
identifed as regulators of autophagy (Aguilera et al., 2012). One of the most strongly
depleted proteins from the knockout membrane fractions was MME (Neprilysin; >5-
fold depletion), but again this did not meet the cut-off for significance. MME was
not quantified in the whole cell lysate, but its loss from the membrane fraction of the
knockout cells look real because it was identified by multiple peptides in all the wild-type
samples but only by a single peptide in one knockout sample (the rest of its knockout
LFQ intensities were by match-between-runs). MME is an integral membrane protein
3.7 Additional proteomic analyses of AP-4 knockout cells 141
endopeptidase which acts on numerous neuropeptides including β-amyloid (Howell
et al., 1995). Mutations in MME have been described to cause the neurodegenerative
diseases Charcot-Marie-Tooth disease Type 2 (Auer-Grumbach et al., 2016; Higuchi et al.,
2016) and Spinocerebellar ataxia 43 (Depondt et al., 2016).
142 Proteomic investigations of adaptor protein complex 4
3.8 Summary
Multiple orthogonal proteomic approaches were applied to investigate AP-4 function in
HeLa cells, with the goal of identifying AP-4 cargo proteins and AP-4 vesicle machinery.
The approaches applied were as follows:
1. Dynamic Organellar Maps - a spatial proteomics method to identify endogenous
proteins mislocalised in AP-4 knockout HeLa cells.
2. Comparative vesicle profiling - a SILAC-based quantitiative comparison of vesicle-
enriched fractions prepared from control and AP-4-depleted HeLa cells, to identify
proteins lost from the vesicle fraction in the absence of AP-4.
3. AP-4 BioID - a screen for proteins in close proximity to the AP-4 complex to identify
possible interacting partners.
4. SERINC BioID - a reciprocal screen for proteins in close proximity to SERINC1 and
SERINC3, to test for their proximity to AP-4.
5. Sensitive immunoprecipitation - immunoprecipitation of AP-4 complexes via
GFP-tagged TEPSIN, under sensitive low-detergent conditions, to identify weak or
transient interaction partners.
6. Whole cell and membrane proteomes - global quantitiative analyses of whole cell
lysates and membrane fractions from AP-4 knockout HeLa cells, to learn more
about the relationship between AP-4 and candidiate AP-4-associated proteins.
Prior to this project the only known AP-4 accessory protein was TEPSIN and there was
no consensus as to which proteins are genuine cargoes of the AP-4 pathway. From the
proteomics experiments performed in this study we have identified three ubiquitously
expressed AP-4 cargo proteins, ATG9A, SERINC1 and SERINC3, and two cytosolic AP-4-
associated proteins RUSC1 and RUSC2. The data also supports an interaction between
AP-4 and a complex containing HOOK1, AKTIP and FAM160A1/2 (the FHF complex;
Xu et al., 2008). The results of the different proteomic experiments with regard to these
proteins are summaried in Table 3.11. The results of the vesicle profiling experiment
suggest that the RUSCs, ATG9A and SERINCs are present in AP-4-derived vesicles, as
they are all enriched in the vesicle fraction and lost from the fraction in cells lacking
AP-4. In contrast, the components of the FHF complex are not enriched in the vesicle
fraction indicating they may interact with AP-4 but without becoming incorporated into
AP-4-derived vesicles themselves. Therefore, the remainder of this thesis focuses on
3.8 Summary 143
Table 3.11 Summary of AP-4 proteomic experiments. Known AP-4 complex proteins and AP-
4-associated proteins identified in this project are listed with a summary of their behaviour
in the different AP-4 proteomic experiments. Green shading indicates a result that supports
AP-4-association. n.i. = not identified.
Gene
name
Organellar
maps
Vesicle
profiling
AP-4
BioID
SERINC
BioID
Sensitive
TEPSIN-
GFP IP
Conventional
TEPSIN-
GFP IP
AP4B1 n.i. Depleted Enriched Enriched Enriched Enriched
AP4E1 n.i. Depleted Enriched Enriched Enriched Enriched
AP4M1 n.i. Depleted Enriched Enriched Enriched Enriched
AP4S1 n.i. Depleted Enriched Enriched Enriched Enriched
TEPSIN n.i. Depleted Enriched n.i. Enriched Enriched
ATG9A Moves Depleted Enriched Enriched Enriched n.i.
SERINC1 Moves Depleted n.i. n.i. Enriched n.i.
SERINC3 Moves Depleted n.i. n.i. Enriched n.i.
RUSC1 n.i. Depleted n.i. Enriched n.i. n.i.
RUSC2 n.i. Depleted Enriched Enriched n.i. n.i.
HOOK1 n.i. n.i. Enriched Enriched Enriched Enriched
FAM160A1 n.i. n.i. Enriched n.i. n.i. n.i.
FAM160A2 n.i. n.i. Enriched n.i. n.i. Enriched
AKTIP n.i. n.i. Enriched n.i. n.i. n.i.
the RUSCs, ATG9A and the SERINCs. Importantly, they are all low abundance proteins,
expressed at comparable levels to AP-4 and TEPSIN in HeLa cells (Figure 3.27; Itzhak
et al., 2016) and in primary mouse neurons (Itzhak et al., 2017). This is in keeping with
our hypothesis that they are all AP-4 vesicle-associated proteins, and it also highlights
the sensitivity of our proteomic approaches. The identification of ATG9A as a potential
AP-4 cargo is particularly intriguing, given the published report of aberrant autophagy
in an Ap4b1 knockout mouse model (Matsuda et al., 2008).
The use of multiple orthogonal approaches was powerful because every method missed
some of the AP-4-associated proteins. Dynamic Organellar Maps was particularly pow-
erful for detecting the mislocalisation of cargo proteins - ATG9A, SERINC1 and SERINC3
were very robust outliers on the maps from the AP-4 knockout cells. The maps data
also begin to tell us something about the behaviour of the AP-4 cargo proteins in cells
that lack AP-4; scrutiny of their nearest neighbours in wild-type and AP-4 knockout
cells suggested movement towards the TGN for ATG9A and intra-endosomal movement
for the SERINCs. The cytosolic AP-4-associated proteins did not appear on the maps
because they were not identified across the different membrane subfractions - probably
partly because they are only transiently recruited to membranes and partly due to their
144 Proteomic investigations of adaptor protein complex 4
Fig. 3.27 Estimated copy numbers of AP-4-associated proteins. Estimated copy numbers of
AP4B1, AP4E1, AP4M1, TEPSIN, RUSC1, RUSC2, ATG9A, SERINC1 and SERINC3, in HeLa cells,
from previously published data (Itzhak et al., 2016). Data are displayed as mean ± SD (n = 6.)
low expression levels. Vesicle profiling, on the other hand, was very good at detecting cy-
tosolic AP-4-associated machinery. TEPSIN and the RUSCs were very strongly depleted
from the vesicle-enriched fractions prepared from AP-4-depleted cells. This is because
vesicle-associated proteins are highly enriched in the fraction. Loss of ATG9A and the
SERINCs from the vesicle-enriched fraction was detected, but it was much less dramatic
than that of TEPSIN and the RUSCs. This is not surprising because it is known that
cargo proteins are sometimes less strongly affected in the vesicle fraction, as they may
exist in several different vesicle populations (Borner et al., 2012). Of note, the SERINCs
were more strongly affected by AP-4 knockdown than knockout, suggesting some com-
pensation may occur in cells that lack AP-4 over a long period of time. BioID detected
proximity between AP-4 and its cargo and cytosolic machinery, with the exception of
RUSC1 and the SERINCs. This highlights the utility of the method for detecting transient
interactions that are missed by conventional immunoprecipitation-based approches,
but also the potential problems of false negatives in BioID. The reciprocal BioID experi-
ment performed with SERINC1 and SERINC3 baits demonstrated the SERINCs do come
into close proximity with AP-4, RUSCs and ATG9A, so it seems likely they were missed
in the AP-4 BioID experiment due to low levels of accessible lysines for biotinylation.
Our sensitive immunoprecipitation approach was also able to preserve the interactions
3.8 Summary 145
between AP-4 and its cargo proteins, which were lost in the higher detergent conven-
tional immunoprecipitation. However, the RUSCs were not identified in the sensitive
immunoprecipitation, suggesting that if they do interact with AP-4 (a hypothesis that is
supported by the BioID data) the interactions are very weak or transient. The proteomic
analyses of whole cell lysates and membrane fractions from AP-4 knockout cells indi-
cated the RUSCs may rely on AP-4 for their recruitment to membrane and may also be
destabilised at the whole cell level in the absence of AP-4.
The proteins discussed above are those most strongly implicated in AP-4-mediated
trafficking due to their identification in multiple experiments, and thus follow-up work
in Chapter 4 focuses on these proteins. Nonetheless, there may be additional inter-
esting candidates for AP-4-associated proteins in the individual experiments. AAGAB
(Alpha- and gamma-adaptin-binding protein p34) was enriched in the sensitive TEPSIN-
GFP immunoprecipitation and as a known binder of AP-1 γ and AP-2 α subunits (Page
et al., 1999) it is a candidate for an AP-4 ε binder. In support of this hypothesis, AAGAB
was enriched in the streptavidin pulldowns for AP4E1 and AP4S1 baits, although this
did not make the cut-off for significance. Another protein that was enriched in the
sensitive immunoprecipitation, DAGLB (Sn1-specific diacylglycerol lipase beta), was
also consistently depleted from vesicle fractions from AP-4-ablated cells, to a similar
degree as ATG9A. DAGLB is a lipase that catalyses the hydrolysis of diacylglyerol to
2-arachidonoylglycerol (2-AG), an abundant endocannabinoid, and is required for neu-
rogenesis in the hippocampus (Gao et al., 2010). It did not undergo a translocation on
the Dynamic Organellar Maps of AP-4 knockout cells but in fractionation profiling of the
vesicle-enriched fraction performed by Georg Borner, DAGLB (Borner et al., 2014) has a
simlar profile to RUSC1, suggesting related distribution within the vesicle fraction (see
Section 4.5.1). There are also a number of proteins that were significantly enriched in
BioID pulldowns for more than one AP-4 subunit, which may be of interest. In particular
there were multiple proteins implicated in ubiquitin-mediated degradation: the E3
ubiquitin-protein ligases UBR5 and BIRC6 (Baculoviral IAP repeat-containing protein
6), the ubiquitin protease USP47 (Ubiquitin carboxyl-terminal hydrolase 47), CACYBP
(Calcyclin-binding protein) and SUGT1 (Protein SGT1 homolog). As ubiquitin signalling
is known to play important roles in autophagy (reviewed in Grumati & Dikic, 2018), it is
possible these proteins are linked to our discovery that ATG9A is an AP-4 cargo protein.
In fact, BIRC6 was recently identified in a screen for autophagy regulators and was shown
to be required for autophagosome-lysosome fusion (Ebner et al., 2018). Thus, some of
these proteins may be worth following up in future studies to determine whether they
do have a connection to AP-4-mediated trafficking.

Chapter 4
Functional characterisation of AP-4
cargo and machinery
4.1 Introduction
This chapter presents follow-up cell biological studies of AP-4 cargo and machinery. The
majority of the work concerns the proteins that were at the intersection of the proteomic
analyses described in Chapter 3 - the AP-4 cargo proteins ATG9A, SERINC1 and SERINC3,
and the cytosolic AP-4 accessory protein RUSC2. Here we investigate how these proteins
are connected to AP-4 and to each other and what happens to them in the absence of
AP-4. The final part of the chapter (Section 4.6) focuses on the downstream effect that
ATG9A missorting has on autophagy and presents data that supports a role for AP-4
in the spatial control of autophagy. In addition, Section 4.2 describes work conducted
in collaboration with Dr Lauren Parker Jackson and Dr Meredith Frazier (University of
Vanderbilt, Tennessee) on the interaction between AP-4 and its previously identified
accessory protein TEPSIN.
4.1.1 A brief introduction to autophagy
Macroautophagy (which is referred to as autophagy from here onwards) is one of two
major cellular degradation pathways, the other being the ubiquitin-proteasome system.
Autophagy specialises in the clearance of cytoplasmic contents including long-lived
proteins and organelles, which are delivered by large double-membrane bound vesicles
called autophagosomes for degradation in the lysosome. It occurs under basal condi-
148 Functional characterisation of AP-4 cargo and machinery
tions as a mechanism of homeostasis but can be upregulated in response to nutrient
starvation or other cellular stresses. Defects in autophagy have been implicated in a
wide range of disorders affecting many different organ systems, including a large num-
ber of neurodegenerative disorders (recently reviewed in Menzies et al., 2017), cancer,
infectious diseases and cardiovascular diseases (reviewed in Levine & Kroemer, 2008).
Hence there is huge interest in the physiological processes and regulation of autophagy.
The process of autophagy is highly complex and is the subject of many detailed reviews
(e.g. Bento et al., 2016; Wen & Klionsky, 2016; Yu et al., 2018). Here the major steps and
machinery involved in mammalian autophagy are briefly summarised to provide con-
text for our studies of the impact of AP-4 deficiency on the trafficking of the autophagy
protein ATG9A and the downstream effect of ATG9A missorting on autophagy.
Autophagy is a conserved process in all eukaryotic cells and much of our understanding
of the molecular mechanisms of autophagy has come from studies in yeast. However,
there are differences between yeast and mammalian autophagy (Bento et al., 2016)
which is why the focus here is on the pathway in mammalian cells. The core proteins
required for the process of autophagy are designated as autophagy-related (ATG) proteins
(Klionsky et al., 2003), many of which were originally identified in pioneering yeast
genetic screens performed in the lab of Yoshinori Ohusumi (Tsukada & Ohsumi, 1993).
The steps involved with autophagy in mammalian cells are outlined in Figure 4.1 and can
be broadly separated into three main stages: (1) Nucleation/initiation; (2) Formation
and elongation of the phagophore to generate an autophagosome; (3) Docking/fusion
with the lysosome for the degradation of autophagic contents in the autolysosome.
Initiation of the phagophore begins with activation and translocation of the ULK kinase
complex (consisting of ULK1, ULK2, ATG13, RB1CC1 and ATG101) to sites of autophago-
some nucleation. What defines these nucleation sites in mammalian cells remains a sub-
ject of debate, but the consensus is that they are specialised sub-domains of the ER and
they have been shown to correspond to ER-mitochondria contact sites (Hamasaki et al.,
2013) and autophagy-specific ER exit sites (Karanasios et al., 2016). A defining feature of
these sites is the clustering of small vesicles containing the transmembrane autophagy
protein ATG9A, which is thought to contribute to the nucleation of the phagophore
(Karanasios et al., 2016). Following this there is activation of the class III phosphatidyli-
nositol 3-kinase (PI3K) VPS34 complex (consisting of VPS34, VPS15, ATG14 and Beclin-1)
which stimulates the synthesis of phosphatidylinositol 3-phosphate (PI3P) and the for-
mation of a membrane platform known as an ‘omegasome’ (Axe et al., 2008). PI3P drives
the recruitment of PI3P-binding proteins of the WIPI family, including WIPI2B which in
4.1 Introduction 149
Fig. 4.1 An overview of autophagy in mammalian cells. The three main stages of autophagy
are: (1) Initiation at sites where clusters of ATG9-containing vesicles are closely associated with
the endoplasmic reticulum (ER). The ULK1/2 kinase complex is activated and translocates to
the site of nucleation, followed by recruitment of the class III phosphatidylinositol 3-kinase
(PI3K) complex; (2) Formation and elongation of the phagophore is driven by proteins recruited
via binding to the phosphatidylinositol 3-phosphate (PI3P) generated by the PI3K complex.
WIPI2B binds and recruits the ATG12–ATG5-ATG16L1 complex, which catalyses the conjugation
of ATG8 proteins, including LC3, to the phagophore membrane. Cytosolic cargoes for degrada-
tion, e.g. damaged organelles or proteins, are captured as the phagophore extends to form a
complete autophagosome; (3) The completed autophagosome fuses with a lysosome to generate
an autolysosome in which the autophagic cargo is degraded.
turn recruits autophagy effectors to form the double-membrane phagophore (Dooley
et al., 2014).
The next stage of phagophore formation and elongation involves the WIPI2B-binding
protein ATG16L1 and its interaction partners ATG12–ATG5 (Dooley et al., 2014). ATG12 is
a ubiquitin-like protein which covalently attaches to its substrate ATG5 (Mizushima et al.,
1998). ATG12–ATG5-ATG16L1 acts as the equivalent of an E3 ubiquitin-protein ligase to
catalyse the covalent conjugation of a second family of ubiquitin-like autophagy proteins,
the Atg8 family, to the lipid phosphatidylethanolamine (PE; Hanada et al., 2007). There
are at least six mammalian orthologs to yeast Atg8, MAP1LC3A/B/C, GABARAP, and
GABARAPL1/2 (Yu et al., 2018). The most widely studied Atg8 mammalian homolog is
MAP1LC3B (Microtubule-associated proteins 1A/1B light chain 3B), commonly referred
to as ‘LC3B’ with the unlipidated form designated as LC3B-I and the lipidated form
designated as LC3B-II. LC3B-II tightly associates with both inner and outer membranes of
the phagophore and autophagosome and so it commonly used as a marker for autophagy
(Klionsky et al., 2016). Work in yeast suggests Atg8 proteins are required for phagophore
elongation and autophagosome maturation (Xie et al., 2008), although their exact role
remains unclear. They also function in the recruitment of autophagic cargo during
150 Functional characterisation of AP-4 cargo and machinery
selective autophagy by interacting with autophagy receptors such as p62 (SQSTM1) via
LC3-interacting region ‘LIR’ motifs (reviewed in Birgisdottir et al., 2013).
Following completion and maturation of the double-membrane bound autophagosome
it fuses with the lysosome to generate an autolysosome in which the autophagic contents
are degraded. This process involves Dynein-dependent retrograde microtubule-based
transport of autophagosomes towards the perinuclear region where lysosomes are con-
centrated (Kimura et al., 2008). Next the autophagosome and lysosome are tethered via
the HOPS complex (VPS11, VPS16, VPS18, VPS33A, VPS39, VPS41) which also interacts
with the Q-SNARE STX17 to facilitate formation of the trans-SNARE complex that medi-
ates autophagosome-lysosome fusion (Jiang et al., 2014; Takats et al., 2014). The small
GTPase RAB7 and various adaptors that link autophagosomal and lysosomal machinery
are also involved with this tethering process (Yu et al., 2018). Finally, SNARE-mediated fu-
sion of the lysosomal and outer autophagosomal membranes occurs allowing lysosomal
enzymes to degrade the inner autophagosomal membrane and the cargo within.
Despite the progress that has been made in the identification of the core autophagy
machinery, the exact function of many of the ATG proteins remains unclear. There
are also many complex regulatory mechanisms and membrane trafficking steps that
feed into the autophagy pathway. The membrane source(s) for the formation of the
autophagosomal membrane is a particular point of controversy; there are three com-
peting models: (i) maturation of an existing membrane platform, e.g. part of the ER;
(ii) assembly from different membrane sources; (iii) a combination model where the
autophagosomal membrane is derived from a preexisiting membrane platform but
receives membrane from other sources during maturation (Bento et al., 2016). Many
studies have demonstrated a close relationship between the ER and early autophagic
structures (e.g. Axe et al., 2008; Hayashi-Nishino et al., 2009), leading to suggestions
that the autophagosomal membrane is derived from the ER. However, it is not yet clear
whether the omegasome is a sub-domain of the ER or an independent compartment in
very close association with the ER. In addition to the ER, many other organelles have
been shown to make membrane contact sites with autophagosomal membranes (Biazik
et al., 2015) and several different organelles have been suggested to supply membranes
to forming phagophores (Bento et al., 2016). ATG9 is the only transmembrane core
ATG protein and thus its identification led to much anticipation that it might hold the
key to identifying the elusive autophagosomal membrane donor compartment. How-
ever, despite its essential role at an early stage of autophagosome biogenesis, its precise
molecular function is still unknown.
4.1 Introduction 151
4.1.2 Background to ATG9A - the transmembrane autophagy protein
Autophagy-related protein 9A (ATG9A) is encoded by the gene ATG9A which is a homolog
of yeast ATG9. ATG9 was originally identified in a screen for autophagy-defective yeast
mutants performed by Miki Tsukada and Yoshinori Ohusumi (Tsukada & Ohsumi, 1993)
and in a separate screen for mutants of the closely related yeast cytoplasm-to-vacuole
targeting (Cvt) pathway (Harding et al., 1995). Vertebrates have two homologous genes
to yeast ATG9, named ATG9A and ATG9B in humans. Both genes have been shown to
encode protein products that are functionally orthologous to yeast Atg9, but ATG9A is
ubiquitously expressed while ATG9B is enriched in the placenta and the pituitary gland
(Yamada et al., 2005).
Initial characterisation of Atg9-deficient yeast revealed an absence of autophagic vesi-
cle formation and sensitivity to nutrient starvation, which are both characteristics of
autophagy mutants (Lang et al., 2000; Noda et al., 2000). Localisation studies based on
the expression of fluorescently tagged Atg9 (overexpressed and endogenous) showed
it to localise to peripheral punctate structures in the cytoplasm as well as to the pre-
autophagosomal structure (PAS), the site of autophagosome formation in yeast (Lang
et al., 2000; Tucker et al., 2003). Yeast Atg9 cycles between these locations in an Atg1-
Atg13 complex-dependent manner and this cycling is essential for autophagy (Reggiori
et al., 2004). Atg9 appears at the PAS in the earliest stages of autophagosome biogenesis
and in its absence many downstream Atg proteins fail to localise to the PAS (Suzuki et al.,
2007). The peripheral Atg9 puncta are often seen in close proximity to mitochondria, ini-
tially leading to speculation that a population of Atg9 resides on mitochondria (Reggiori
et al., 2005). However, immunoelectron microscopy studies revealed that the puncta
do not correspond to the membranes of any known organelle and instead represent a
novel compartment consisting of clusters of vesicles and tubules, named as the ‘Atg9
reservoir’ (Mari et al., 2010). In this study Mari and colleagues found that Atg9 reservoirs
translocate to close proximity with the vacuole to generate the PAS and act in the nucle-
ation of the cup-shaped membrane known as the ‘phagophore’, which extends to form
the autophagosome. This study used overexpressed GFP-tagged Atg9 to define the Atg9
reservoir whereas another study used endogenously tagged Atg9 and found the majority
to reside on highly mobile cytoplasmic vesicles which did not form clusters (Yamamoto
et al., 2012). The authors of this study suggested the clusters of tubules and vesicles ob-
served by Mari and colleagues could thus be an artefact of overexpression. Despite this
disagreement, both studies agreed on the size of the Atg9 vesicles (Mari et al. reported a
mean diameter of 30-40 nm and Yamamoto et al. estimated a diameter of 30-60 nm),
152 Functional characterisation of AP-4 cargo and machinery
on the origin of the vesicles being the Golgi apparatus, and on their involvement in the
nucleation of the phagophore.
Exactly how the Atg9 vesicles facilitate phagophore formation is unclear and although
it is commonly assumed they deliver membrane, this has not been formally shown
(Wen & Klionsky, 2016). Even if they do, based on the small size of Atg9 vesicles and the
quantities in which they appear at the PAS, it is expected that an additional membrane
source is required for expansion of the autophagosome, the nature of which is a subject
of much debate (Yamamoto et al., 2012). Also controversial is exactly when yeast Atg9 is
retrieved from the autophagosome membrane. It is known to be incorporated into the
phagophore but is not present on the autophagosomes that accumulate in the vacuole
when autophagosome degradation is blocked with the proteinase B inhibitor PMSF
(Noda et al., 2000).
Human ATG9A is an 839 amino acid protein with six transmembrane domains (Figure
4.2A). It shares a highly conserved APG9 domain (Pfam ID, PF04109) with yeast Atg9
(amino acids 176-530), while its N- and C-terminal cytosolic regions are more divergent
(Young et al., 2006). Depletion of ATG9A by siRNA-mediated knockdown in human cell
lines causes a large reduction in autophagosome formation (Yamada et al., 2005; Young
et al., 2006). Atg9a knockout mice die within one day of birth due to an inability to survive
the neonatal starvation period (Saitoh et al., 2009). This phenotype is shared with other
autophagy-deficient mouse models, e.g. Atg5- and Atg7-deficient mice, indicating an
important role in starvation-induced autophagy. As in yeast, ATG9A acts at an early stage
in autophagosome biogenesis because knockdown of ATG9A in HEK293 cells reduces
the formation of omegasomes, the precursor of autophagosomes in mammalian cells
(Winslow et al., 2010). However, a few autophagosomes still form under conditions of
starvation in mouse embryonic fibroblasts (MEFs) derived from the Atg9a knockout mice,
suggesting that, unlike in yeast, mammalian Atg9 is not strictly essential for autophagy
(Orsi et al., 2012).
In human cells under basal conditions ATG9A localises to the TGN as well as to peripheral
cytoplasmic puncta (Young et al., 2006). The peripheral population of ATG9A has been
described to co-localise with various organelle markers including of early, recycling and
late endosomes (Orsi et al., 2012; Young et al., 2006). Starvation induces a redistribution
of ATG9A away from the TGN towards the peripheral pool, and peripheral ATG9A is
recycled back to the TGN when cells are returned to nutrient-rich conditions (Young
et al., 2006). Under starvation conditions colocalisation of ATG9A with LC3B-positive
autophagosomal structures can also be observed. However, in contrast to the situa-
4.1 Introduction 153
Fig. 4.2 Autophagy-related protein 9A (ATG9A) - the transmembrane autophagy protein. (Full
caption on following page.)
154 Functional characterisation of AP-4 cargo and machinery
Fig. 4.2 Autophagy-related protein 9A (ATG9A) - the transmembrane autophagy protein. (A)
Schematic diagram demonstrating the topology of ATG9A. (B) Correlative light and electron
microscopy (CLEM) analysis of HEK293 cells expressing GFP-DFCP1 (an early autophagosome
marker) and mRFP-ATG9 from Orsi et al. (2012). Top left shows confocal microscopy of mRFP-
ATG9 and GFP-DFCP1-positive structures (boxes 1-3). Electron micrographs of sections from
areas 1-3 are shown. Black arrowheads indicate a double-membrane phagophore. Asterisks mark
autophagosomes. The mRFP-ATG9 signal corresponds to clusters of small vesicles and tubules.
Scale bar: 50 μm. (C) Summary model of ATG9A trafficking in mammalian cells, based on Noda
(2017). ATG9A localises to the trans-Golgi network, endosomes and the plasma membrane
(PM), as well as the tubulovesicular ‘ATG9 reservoir’. Adaptor protein complex 2 (AP-2) mediates
the retrieval of ATG9A from the plasma membrane to recycling endosomes. TBC1D14 and the
TRAPPIII complex have been shown to be involved with transport from recycling endosomes
to the Golgi apparatus. SNX18 and Dynamin-2 also act at the recycling endosome to generate
ATG9A and ATG16L1-containing tubules which contribute to autophagosome biogenesis. The
Golgi is also considered to be a major source of the ATG9-containing vesicles and tubules that
constitute the ATG9 reservoir, but the machinery involved remains to be identified.
tion in yeast, ATG9A does not appear to be incorporated into the growing phagophore,
and rather ATG9A-containing vesicles make transient contact with the autophagosome
membrane (Orsi et al., 2012). These transient contacts can be seen with early omega-
some structures marked by DFCP1 (double FYVE domain–containing protein) and later
structures marked by LC3B, suggesting ATG9A vesicles play a role throughout autophagy
formation. Despite this, ATG9A vesicles in human cells appear morphologically very
similar to those in yeast. Correlative light and electron microscopy (CLEM) and cryo-
immunoelectron microscopy studies revealed ATG9A to localise to clusters of tubules
and vesicles, which are often found in close proximity to autophagosomes (Figure 4.2B;
Orsi et al., 2012).
ATG9A has a complex (and controversial) trafficking itinerary which encompasses all
major components of the endocytic pathway and disruption at various stages has been
shown to affect autophagy (summarised in Figure 4.2C and recently reviewed in Noda,
2017). This has led to the idea that the continuous cycling of ATG9A between com-
partments provides an ‘ATG9A reservoir’ that is ready to be mobilised upon autophagy
induction. In addition to its localisation at the TGN and in various flavours of endosome,
several reports have shown that ATG9A travels via the plasma membrane, from which
it returns to recycling endosomes via AP-2- and clathrin-mediated endocytosis (Imai
et al., 2016; Popovic & Dikic, 2014; Puri et al., 2013). Depletion of AP-2 results in ATG9A
accumulation at the plasma membrane and a clear reduction in LC3 lipidation. The
trafficking of ATG9A from RAB11-positive recycling endosomes to the Golgi apparatus is
regulated by TBC1D14 in conjunction with the TRAPPIII complex (Lamb et al., 2016).
4.1 Introduction 155
Depletion of the TRAPPIII subunit TRAPPC8 diminishes the juxta-nuclear pool of ATG9A
and inhibits autophagosome formation at an early stage. Also at the recycling endosome,
SNX18 (Sorting nexin-18) and DNM2 (Dynamin-2) play a role in the generation of ATG9A
and ATG16L1- containing tubules that become autophagosome precursors (Søreng et al.,
2018). ATG9A accumulates in recycling endosomes in SNX18-deficient cells and au-
tophagic flux is compromised. The starvation-induced redistribution of ATG9A from the
TGN to the peripheral pool is dependent on the serine/threonine-protein kinase ULK1,
which is the mammalian homolog of yeast Atg1 (Young et al., 2006). This is mediated
at least in part by the phosphorylation of ATG9A by ULK1 at Ser-14 (Zhou et al., 2017).
ATG9A still reaches DFCP-1-positive omegasome structures in ULK1-deficient cells, but
these early autophagosomal structures are restricted to the juxtanuclear region and fail
to mature (Orsi et al., 2012).
Despite this wealth of trafficking information, it is hard to disentangle direct from in-
direct contributions to ATG9A trafficking. The exact nature of the ‘ATG9A reservoir’, its
origin, and how it interacts with autophagosomes during their biogenesis remains to be
understood.
4.1.3 Background to the SERINC protein family
The function of the SERINCs is even more mysterious than that of ATG9A. The SER-
INC family is highly conserved across all eukaryotes and consists of five members in
mammalian cells (SERINC1-5; Inuzuka et al., 2005). However, our HeLa cell line only
expresses SERINC1 and SERINC3 (Itzhak et al., 2016), raising the possibility that addi-
tional SERINC family members may be AP-4 cargo proteins in other human cell types.
All members of the family are roughly the same size (423-518 amino acids), consist of a
high proportion of hydrophobic amino acids (>50 %) and are predicted to have between
10 and 11 transmembrane domains (Inuzuka et al., 2005). As discussed in Section 3.5.2,
SERINC1 and SERINC3 are both predicted to have 11 transmembrane domains, with
their N termini being luminal/extracellular and their C termini being cytosolic (Figure
4.3). The mammalian family members share 31-58 % sequence identity and have 25-30
% sequence identity with the single yeast homolog Tms1 (Inuzuka et al., 2005). Saccha-
romyces cerevisiae Tms1 deletion strains were viable and showed no obvious growth
defects or sensitivities to a range of stresses including metal ions, temperature, osmotic
stress, oxidative stress and different carbon or nitrogen sources (Grossman et al., 2000).
However, the high degree of conservation of SERINC proteins in eukaryotic cells suggests
they must have an important biological function.
156 Functional characterisation of AP-4 cargo and machinery
Fig. 4.3 Serine incorporator 1 and 3 (SERINC1 and SERINC3). Schematic diagram demonstrat-
ing the topologies of SERINC1 and SERINC3.
Mammalian genes encoding SERINC family proteins were initially cloned in a screen
for genes that are differentially expressed in tumours in mice, and so were named TDE
for ‘testicular tumor differentially expressed’ (Lebel & Mes-Masson, 1994). An early
functional study of murine SERINC1 and SERINC3 (named in the paper as TMS-2 and
TMS-1, respectively) described expression in the brain with particular enrichment in
areas associated with glutamatergic neurons such as the hippocampus and cerebral
cortex (Grossman et al., 2000). This led the authors to hypothesise that they might be
vesicular glutamate transporters but their attempts to detect glutamate uptake in cells
were unsuccessful. The name SERINC was coined for ‘serine incorporator’ based on a
proposed role in the incorporation of serine into phosphatidylserine and sphingolipids
(Inuzuka et al., 2005). In this study Serinc1 and Serinc2 were up-regulated in the rat brain
following kainate-induced seizures, whereas Serinc5 was down-regulated. As well as
their differential activity-dependent expression, neural localisation of the mRNAs was
different - Serinc1 and Serinc2 were enriched in neuronal cells of the hippocampus and
cerebellar cortex (in agreement with the study from Grossman and colleagues) while Ser-
inc5 was predominantly expressed in the myelin-rich white matter. In yeast two-hybrid
experiments with SERINC1 as bait, Inuzuka and colleagues identified interactions with
serine biosynthesis enzymes including Ser3 (3-phosphoglycerate dehydrogenase) and
Spt (a subunit of serine palmitoyltransferase). Based on this finding the authors investi-
gated the effect of SERINCs on cellular serine. Serine transport across the membrane of
Escherichia coli and mammalian COS cells was unaffected by expression of SERINC1,
suggesting the SERINCs do not function as serine transporters. However, increased in
vitro incorporation of radiolabelled serine into phosphatidylserine and sphingolipids
was detected in lysates from Escherichia coli and COS cells that overexpressed SERINC1.
Similar effects were observed for SERINC2 and SERINC5. Additionally, a Tms1-deficient
yeast strain had reduced levels of serine incorporation into sphingolipids. Based on
4.1 Introduction 157
these findings the authors proposed a model of dual-function for the SERINCs, firstly in
acting as a scaffold for enzymes involved in serine biosynthesis and secondly to carry
polar serine molecules into the lipid bilayer. Neither function has been experimentally
confirmed.
More recently members of the SERINC family have been identified as restriction factors
for human immunodeficiency virus (HIV-1) and have been found to be excluded from
HIV-1 virions by the activity of the HIV-1 accessory protein Nef (Rosa et al., 2015; Usami
et al., 2015). Usami and colleagues carried out a proteomic comparison of virions pro-
duced by T lymphoid cells infected with wild-type or Nef-deficient HIV-1. SERINC3 was
the only host protein that was reproducibly identified in Nef-deficient virions and not in
the wild-type virions. The authors went on to show that both SERINC3 and SERINC5 are
excluded from HIV-1 virions in a Nef-dependent manner and that their overexpression in
producer cells reduced the infectivity of virions. Rosa and colleagues used an orthogonal
transcriptomics approach to identify differentially expressed genes that correlated with
Nef-responsiveness. SERINC5 was the most correlated gene with the requirement of
Nef for HIV-1 infectivity. They also found that both SERINC5 and SERINC3 shared the
ability to restrict HIV-1. Overexpressed SERINC5 localised predominantly to the plasma
membrane when expressed alone but was also present as bright puncta in the perinu-
clear region. Co-expression of Nef resulted in loss of the plasma membrane localisation
and an accumulation in perinuclear puncta which were positive for the late endosomal
marker Rab7-RFP (Rosa et al., 2015; Usami et al., 2015). This suggests Nef redirects
SERINC3/5 from the plasma membrane to avoid their incorporation into HIV-1 virions
as they bud from the surface of the cell. In line with this the knockdown of AP-2 reduced
the ability of Nef to target SERINC5 for removal from the plasma membrane, suggesting
this is mediated by clathrin- and AP-2-dependent endocytosis (Rosa et al., 2015). Exactly
how SERINC3 and SERINC5 exert their effect on HIV-1 infectivity is unknown. Their
presence on HIV-1 virions impairs fusion with the target cell, but this effect does not
seem large enough to account for their entire impact on infectivity (Rosa et al., 2015;
Usami et al., 2015). HIV-1 virions are enriched in phosphatidylserine and sphingolipids
and their lipid composition is important for their infectivity, leading to speculation that
the SERINCs exert their effect by modifying the lipid content of the virion (Fackler, 2015).
However, a quantitative MS-based lipidomics study found no significant differences in
the lipid composition of virions or producer cells with and without SERINC5 expression
(Trautz et al., 2017). Similarly another study found no significant alteration of membrane
lipid composition in macrophages, T cells or B cells from a SERINC1-deficient mouse
158 Functional characterisation of AP-4 cargo and machinery
model (Chu et al., 2017). Therefore further work is required to understand both the
cellular role of SERINCs and how they restrict the infectivity of HIV.
4.1.4 Background to the RUSC protein family
The RUSC (RUN and SH3 domain-containing) protein family is also poorly characterised
functionally. The family consists of two members, RUSC1 and RUSC2, and so far or-
thologs have only been identified in vertebrates (Jin et al., 2016). RUSC1 and RUSC2 have
35 % sequence identity and both contain a RUN domain and a C-terminal SH3 domain
(Figure 4.4). The RUSC1 gene is subject to alternative splicing to give four different
RUSC1 isoforms. Our MS data is consistent with isoform 2 (433 amino acids) being the
major isoform expressed in our HeLa cells and it also appears to be the predominant
isoform expressed in the mouse brain (MacDonald et al., 2012). RUSC2 on the other
hand appears to exist as a single isoform consisting of 1,516 amino acids.
RUN (named from RPIP8, UNC-14, and NESCA proteins) domains have a globular
structure consisting of six conserved blocks and are found in a number of proteins
that have been linked to the function of Rap and Rab family GTPases (Callebaut et al.,
2001; reviewed in Yoshida et al., 2011). Several RUN domain containing proteins are
implicated in vesicular trafficking and have been shown to associate with cytoskeletal
components including kinesins and microtubules. For example, a protein called SKIP1
(SifA and kinesin-interacting protein) interacts with kinesin-1 via an interaction between
its RUN domain and kinesin light chain (KLC; Dumont et al., 2010). Overexpression of
SKIP1 resulted in a microtubule- and kinesin-1-dependent redistribution of LAMP1-
positive organelles to the periphery of the cell. This may be a conserved function of
RUN domains because the C. elegans protein UNC-14 also binds to kinesin-1 via its RUN
domain (Sakamoto et al., 2004). In C. elegans UNC-14 is required for the proper axonal
Fig. 4.4 RUN and SH3 domain-containing protein 1 and 2 (RUSC1 and RUSC2). Schematic
diagram demonstrating the domain organisation of RUSC1 and RUSC2. There are additional
RUSC1 isoforms, but our mass spectrometry data identified the isoform shown as predominant
in our samples.
4.1 Introduction 159
transport of synaptic vesicles and neurite extension. The SH3 (Src homology 3) domain
was first identified as a conserved region outside the catalytic part of several cytoplasmic
protein tyrosine kinases including SRC, ABL1, LCK (Mayer et al., 1988). SH3 domains are
small domains (approximately 60 amino acids) found in around 300 human proteins and
are thought to mediate protein-protein interactions via binding to proline-rich regions
on interaction partners (Morton & Campbell, 1994).
RUSC1 was originally named NESCA for ‘new molecule containing SH3 at the carboxy-
terminus’ (Matsuda et al., 2000). It is ubiquitously expressed but enriched in the brain
and expressed in neurons throughout the cerebellum, cortex, hippocampus, brain stem
and spinal cord (MacDonald et al., 2012, 2004). RUSC1 was originally identified in a
yeast two-hybrid screen with NTRK1 (High affinity nerve growth factor receptor, also
known as Trk-A) as bait (MacDonald et al., 2004). In rat PC12 cells (a neuronal cell
model) overexpressed RUSC1 was cytoplasmic but NGF stimulation resulted in the re-
distribution of RUSC1 from the cytoplasm to the nuclear envelope. In contrast RUSC1
did not show any nuclear localisation in non-neuronal human HEK293 cells, even in the
presence of exogenous NTRK1. This study also implicated RUSC1 in the regulation of
neurite outgrowth because PC12 cells that overexpressed RUSC1 had enhanced neurite
outgrowth in response to NGF, whereas RUSC1 depletion with siRNA had the opposite
effect. Another study identified RUSC1 as an interaction partner for a protein called
NEMO (NF-kappa-B essential modulator), also in a yeast two-hybrid screen (Napoli-
tano et al., 2009). NEMO is the regulatory subunit of the IKK core complex which is
required for the activation of NF-kappa-B. NF-kappa-B is a transcription factor which
regulates numerous genes in response to diverse biological stimuli including, in the
nervous system, NGF stimulation (Wood, 1995). RUSC1 was also found to interact with
TRAF6 (TNF receptor-associated factor 6), an E3 ubiquitin ligase which is also an NK-
kappa-B activator, suggesting it acts as a central adapter in NGF-induced NF-kappa-B
signalling (Napolitano et al., 2009). More recently, a follow-up study by MacDonald and
colleagues described a role for RUSC1 as an adapter for neuronal vesicular transport
(MacDonald et al., 2012). They identified interactions between RUSC1, cytoskeletal com-
ponents including microtubules and actin, kinesin-1 and the synaptic vesicle protein
syntaxin-1. Endogenous RUSC1 had a punctate distribution pattern in primary mouse
neurons, including in axons and dendrites, and overlapped with labelling for kinesin-1
and syntaxin-1. Based on these findings the authors hypothesised that RUSC1 provides
a link between syntaxin-1 and cytoskeletal transport machinery for the anterograde
transport of pre-synaptic membrane components. This has parallels to the roles of SKIP1
and C. elegans UNC-14 discussed above.
160 Functional characterisation of AP-4 cargo and machinery
RUSC2 was identified in a yeast two-hybrid screen for proteins that interact with the
Ras-related small GTPase RAB1 in its active GTP-bound form (Bayer et al., 2005). RAB1
and its functional homolog in yeast, Ypt1, are required for vesicle transport between the
ER and the Golgi apparatus (Segev, 1991; Tisdale et al., 1992) but also play an important
role in autophagosome formation (reviewed in Ao et al., 2014). Ypt1/RAB1 have also
been identified as components of Atg9/ATG9A-containing vesicles (Kakuta et al., 2017,
2012), suggesting a possible link between RUSC2 and ATG9A. However, the functional
relevance of the association between RUSC2 and RAB1 is unknown. Yeast two-hybrid
screening suggests RUSC2 may also bind to RAB35 and RAB41 via its RUN domain
(Fukuda et al., 2011). RUSC1 on the other hand did not exhibit any Rab-binding activity
in this screen. Interestingly, overexpression of the RAB35-binding domain of RUSC2
has a dominant negative effect on activated RAB35 and this inhibits neurite outgrowth
in PC12 cells (Fukuda et al., 2011). This suggests both RUSC1 and RUSC2 are involved
in the regulation of neurite outgrowth. Like RUSC1, RUSC2 is ubiquitously expressed
but appears to be enriched in the brain (Bayer et al., 2005). It has been shown to be
membrane-associated and in HeLa cells overexpressed HA-tagged RUSC2 had a punctate
distribution throughout the cytosol, with a concentration in the perinuclear region of the
cell (Bayer et al., 2005). Recently homozygous loss-of-function mutations in RUSC2 were
identified in three patients (from two separate families) with a neurological disorder
characterised by congenital hypotonia, motor delay, severe intellectual disability, limited
speech and secondary microcephaly (Alwadei et al., 2016). Two of the three patients,
who were siblings, additionally presented with epilepsy during infancy and had brain
abnormalities including thinning of the corpus callosum, while the third unrelated
patient was less severely affected. These symptoms have clear overlap with those of AP-4
deficiency, and the age of onset during early infancy was similar. This is consistent with
our hypothesis that RUSC2 functions in AP-4-mediated trafficking.
4.2 Interaction between AP-4 and TEPSIN 161
4.2 Interaction between AP-4 and TEPSIN
Prior to this project, TEPSIN was the only known AP-4 accessory protein. As previously
discussed, TEPSIN was identified by Dr Georg Borner using a comparative proteomic
vesicle profiling approach (Borner et al., 2012). Vesicle-enriched fractions were prepared
under different conditions and TEPSIN had a very similar pattern of enrichment across
the preparations as the AP-4 subunits, predicting association. In this study GFP-tagged
TEPSIN was found to pull down specifically with the AP4B1 subunit C-terminal ap-
pendage domain (residues 612-749), but not with the AP4E1 subunit appendage domain.
Dr Lauren Parker Jackson and Dr Meredith Frazier, working at the University of Vander-
bilt in Tennessee, led efforts to further characterise the interaction between AP-4 and
TEPSIN. They defined binding sites within the C terminus of TEPSIN and the AP4B1
appendage domain that mediate the interaction and demonstrated their importance for
binding in vitro (discussed below in Section 4.2.1). Working in collaboration with Jackson
and Frazier, I investigated the importance of these binding motifs in vivo (Sections 4.2.2
and 4.2.3). This work has been published in Frazier et al. (2016).
4.2.1 AP4B1 appendage domain and TEPSIN binding motifs
TEPSIN is a member of the epsin family of proteins, which also includes epsins1-3
(EPN1-3) and EpsinR (CLINT1), accessory proteins for AP-2 and AP-1 vesicles, respec-
tively. Interactions between epsin family members and AP complex appendages are
mediated by short amino acid motifs which are harboured within the unstructured C-
terminal regions of the epsins (Mills et al., 2003; Owen et al., 1999; Rosenthal et al., 1999b).
Based on this, Jackson and Frazier generated GST-fusion constructs of unstructured
regions of TEPSIN and used them in pulldown experiments with the AP4B1 appendage
domain to narrow down the binding region to residues 450-500 in the TEPSIN C termi-
nus. Sequence alignment of this region across TEPSIN homologs from multiple species
revealed a highly conserved hydrophobic sequence of eight amino acids with the con-
sensus sequence LFxG[M/L]x[L/V] (Figure 4.5A). L470, F471, G473 and M474 were the
most highly conserved residues. They then used a technique called isothermal titra-
tion calorimetry (ITC) to test for direct binding between this region of TEPSIN and the
AP4B1 appendage domain (Figure 4.5B). ITC works by directly measuring the heat that
is released or absorbed during a molecular binding event (described in Leavitt & Freire,
2001). A ligand (in this case the TEPSIN region) is gradually titrated into a reaction cell
containing the possible binding partner (the AP4B1 appendage domain) and a sensitive
162 Functional characterisation of AP-4 cargo and machinery
calorimeter measures the change in temperature, which is directly proportional to the
amount of ligand that binds. In this way ITC can be used both as a means to detect a
binding event and to measure the affinity of binding. In this case the binding affinity
was in the low micromolar levels (KD = 2.8 μM +/- 0.8 μM), which is similar to other AP
complex subunit interactions (Edeling et al., 2006). In contrast no binding was detected
between the TEPSIN region and either AP1B1 or AP2B1 appendage domains (Figure
4.5B). Site-directed mutagenesis revealed L470 and F471 to be critical for binding as
mutation of both to serine residues completely abolished binding (4.5C). Similarly, re-
placing the highly conserved G473 with the larger isoleucine residue and the conserved
M474 with a polar glutamine residue also abrogated measurable binding.
Jackson and Frazier additionally used a nuclear magnetic resonance (NMR) chemical
shift perturbation experiment (reviewed in Williamson, 2013) to identify key residues
on the AP4B1 appendage domain for the binding of the TEPSIN motif (Figure 4.6A).
The AP4B1 appendage domain was labelled with 15N and unlabelled TEPSIN (residues
450-500) was titrated into it. In an NMR chemical shift perturbation experiment the
15N-1H heteronuclear single quantum correlation (HSQC) spectrum of the 15N-labelled
protein alone is compared to the HSQC spectra of the protein complexed with increasing
amounts of its unlabelled binding partner. The chemical shift changes that are detected
using NMR are highly sensitive to the electronic environment of the nucleus, so chemical
shift perturbations (CSPs) can be caused by non-covalent interactions with a binding
partner. Usually the largest CSPs will belong to amino acids close to the interaction
surface. When the largest CSPs were mapped onto the NMR structure of AP4B1 (protein
data bank: 2MJ7; deposited by the Northeast Structural Genomics Consortium) this
revealed residues on the surface of the AP4B1 appendage domain that were involved with
TEPSIN binding (Figure 4.6B). Comparison of AP4B1 and AP2B1 appendage domains
revealed a highly conserved binding patch that contained many of the AP4B1 residues
identified in the NMR chemical shift experiment. This suggested that the TEPSIN motif
binds to the AP4B1 appendage domain in an equivalent position to where epsin1 binds
to the AP2B1 appendage domain. Based on this hypothesis and the NMR data, Jackson
and Frazier selected several candidate AP4B1 residues for site-directed mutagenesis and
used a combination of pulldowns and ITC to test their importance for binding to TEPSIN.
This revealed the most important residues for TEPSIN binding to be I669, A670 and Y682
(Figure 4.6C). Double mutation of I669 to an alanine residue and A670 to a serine residue
completely abrogated binding to TEPSIN in the ITC experiment. Y682 was replaced by a
valine residue but this construct was not produced with a high enough yield to use for
ITC. Nonetheless, the mutation eliminated binding to TEPSIN in a pulldown experiment.
4.2 Interaction between AP-4 and TEPSIN 163
Fig. 4.5 The TEPSIN C-terminal AP4B1 appendage-binding motif. (A) Sequence alignment
of TEPSIN from major eukaryotic super groups revealed a conserved hydrophobic region of
eight residues in the unstructured C-terminus. Highly conserved amino acids are marked by
asterisks. (B) Isothermal titration calorimetry (ITC) was used to test for binding between wild-type
AP4B1 appendage domain and a recombinant TEPSIN fragment (residues 450–500) containing
the conserved hydrophobic sequence. Binding was detected with an equilibrium dissociation
constant (KD) of 2.9 ± 0.8 μm. In contrast, the TEPSIN fragment did not bind to AP1B1 or AP2B1
appendage domains. (C) ITC was used to test for binding between wild-type AP4B1 appendage
domain and TEPSIN mutants with mutations of highly conserved residues marked with asterisks
in A. Representative ITC traces are shown. All mutants either reduced or completely abolished
detectable binding to AP4B1 appendage domain. L470, F471, G473 and M474 were the most
important residues for binding. Experiments were performed by Meredith Frazier and Lauren
Jackson. From Frazier et al. (2016).
164 Functional characterisation of AP-4 cargo and machinery
Fig. 4.6 The AP4B1 appendage domain TEPSIN-binding motif. (A) A nuclear magnetic reso-
nance (NMR) chemical shift perturbation (CSP) experiment was used to identify residues on the
surface of the AP4B1 appendage domain involved in binding to TEPSIN. 15N-1H heteronuclear
single quantum correlation (HSQC) spectra of the initial (black) and final (red) titration points are
overlaid. This reveals chemical shifts resulting from binding of unlabelled recombinant TEPSIN
(residues 450–500) to 15N-labelled AP4B1 appendage domain. Residues identified for further
analysis are highlighted in B. (B) Residues that exhibited large CSPs in the NMR CSP experiment
were mapped onto an NMR structure of AP4B1 (PDB: 2MJ7). These were E632, W635, L636, I669,
A670 and Y682. These residues mapped to a highly conserved binding patch, also present on
AP2B1. A close-up view of this binding surface is shown. (C) ITC was used to test for binding
between the TEPSIN fragment (residues 450-500) and AP4B1 appendage domain mutants. The
AP4B1 mutant I669A/A670S exhibited no measurable binding to the TEPSIN motif by ITC (a
representative trace is shown). (D) Mutation of Y682V in the AP4B1 appendage domain abolishes
binding to GST-TEPSIN (residues 450–500) in a GST-pulldown experiment. Experiments were
performed by Meredith Frazier and Lauren Jackson. From Frazier et al. (2016).
4.2 Interaction between AP-4 and TEPSIN 165
4.2.2 The in vivo effect of mutating the AP4B1 TEPSIN-binding motif
Following on from the in vitro characterisation of the TEPSIN-binding motif on the
AP4B1 appendage domain, we sought to test its functional relevance to the interaction
between AP-4 and TEPSIN in cultured human cells. In order to do this we made use of the
AP4B1 knockout HeLa cell line described in Section 3.2.3 to provide a clean background
for the expression of structure-based point mutants. In these cells there is no change
in the TEPSIN expression level (Figure 4.7A), as is also the case in fibroblasts from AP-4
deficient patients (Borner et al., 2012). To test whether AP4B1 residues I669, A670 and
Y682 are important for TESPIN binding in vivo, stable rescue cell lines were created
on the AP4B1 knockout background using either full-length wild-type AP4B1 or one
of several mutant constructs: (i) earless AP4B1 lacking the entire appendage domain;
(ii) full-length AP4B1 with I669A and A670S substitutions; (iii) full-length AP4B1 with a
Y682V substitution. To generate the wild-type and earless rescue constructs, full-length
(residues 1-739) or earless (residues 1-612) AP4B1 were cloned into a pLXIN retroviral
vector using restriction enzyme-based cloning. Successful cloning was confirmed by
diagnostic digests and sequencing (data not shown). Residue 612 was chosen as the
cut-site for the earless construct based on the NMR structure of the AP4B1 appendage
(protein data bank: 2MJ7; deposited by the Northeast Structural Genomics Consortium)
which shows that residues 618-739 comprise the appendage domain. The construct
used to generate the NMR structure started at residue 610, but the residues prior to 618
were disorded in the NMR structure, suggesting they are part of the flexible hinge region.
Meredith Frazier used side-directed mutagenesis to modify full-length pLXIN_AP4B1 to
give pLXIN_AP4B1[Y682V] and pLXIN_AP4B1[I669A/A670S].
To generate stable cell lines, AP4B1 knockout HeLa cells were transduced with retrovirus
and selected for stable expression of the AP4B1 constructs by addition of G418. Western
blotting of whole cell lysates from these cell lines with an antibody against AP4B1 con-
firmed expression of proteins of the expected size (Figure 4.7A). The expression levels of
the four rescue constructs were similar, but higher than that of endogenous AP4B1. All
four AP4B1 constructs rescued the level of endogenous AP4E1 (which is reduced in the
AP4B1 knockout cells) to wild-type levels. This suggests that the mutant AP4B1 proteins,
including the earless mutant, are able to form stable AP-4 complexes. The expression
level of TEPSIN was the same in all cell lines. Despite the higher level of expression of
the AP4B1 rescue constructs, when AP-4 was immunoprecipitated with an antibody
against AP4B1 the same amount of AP4E1 was co-immunoprecipitated from all cell lines,
suggesting that the amount of AP-4 complex formation is unaffected by the presence of
166 Functional characterisation of AP-4 cargo and machinery
Fig. 4.7 Disruption of the AP4B1 appendage domain TEPSIN-binding site greatly reduces
TEPSIN binding to AP-4 in vivo. (A) Western blot of whole-cell lysates from control (wild-type
HeLa), AP4B1 knockout (KO) and AP4B1 KO cells stably rescued with full-length wild-type or mu-
tant (earless, Y682V or I669A/A670S) AP4B1, with antibodies against AP4B1, AP4E1 and TEPSIN
(N.B. TEPSIN has two isoforms). An antibody against clathrin heavy chain was used as a loading
control. In the rescued cell lines AP4B1 is overexpressed relative to the level of expression in
the control wild-type HeLa cells. (B) Western blots of immunoprecipitates of AP4B1 or AP4E1
from extracts of the cells shown in A, with antibodies against AP4B1, AP4E1 and TEPSIN (marked
by arrow heads). No TEPSIN could be detected in immunoprecipitates from the KO cells, but
the immunoprecipitates from the KO cells rescued with wild-type AP4B1 contained a similar
amount of TEPSIN to immunoprecipitates from the wild-type control cells. A small amount of
TEPSIN was detected in the immunoprecipitates from the KO cells rescued with mutant AP4B1.
one subunit in excess (Figure 4.7B). While the amount of AP4E1 co-immunoprecipitated
with AP4B1 was unchanged, substantially less TEPSIN co-immunoprecipitated with
AP4B1 from all three mutant cell lines. Very similar results were seen when using an anti-
body against AP4E1 to immunoprecipitate AP-4 (Figure 4.7B). The amount of TEPSIN in
the immunoprecipitates from the I669A/A670S and Y682V mutant cells was comparable
to that from the earless cells, providing evidence that these residues are critical for the
direct interaction between TEPSIN and the AP4B1 appendage domain in vivo. However,
a small amount of TEPSIN still came down with AP-4 from all three cells lines rescued
with mutant AP4B1. This is in contrast to the AP4B1 knockout cell line where no TEPSIN
could be detected in the AP-4 immunoprecipitates. This strongly suggests there must be
at least one additional TEPSIN binding site elsewhere on the AP-4 complex.
4.2 Interaction between AP-4 and TEPSIN 167
Given that disruption of the AP4B1 appendage domain TEPSIN-binding site had a severe
effect on TEPSIN binding in the immunoprecipitation assays, we hypothesised that the
residual TEPSIN binding observed would likely be insufficient to mediate recruitment
of TEPSIN to TGN membranes. To test this, wild-type HeLa cells and the three mutant
AP4B1 rescue cell lines were analysed by immunofluorescence microscopy with anti-
bodies against AP4E1 and TEPSIN (Figure 4.8). The AP4E1 labelling was similar in all
cell lines, confirming that all three mutant AP4B1 constructs could successfully form
AP-4 complexes on the membrane. Surprisingly, punctate TEPSIN labelling with clear
colocalisation with AP4E1 was also seen in all cell lines. This provided further evidence
that other parts of the AP-4 complex are involved with TEPSIN binding and recruitment
to the membrane.
4.2.3 The in vivo effect of mutating the TEPSIN AP4B1-binding motif
We next sought to test the role of the C-terminal TEPSIN AP4B1 appendage-binding motif
(LFxG[M/L]x[L/V]) in vivo. To do this we created HeLa cell lines that stably overexpressed
either wild-type TEPSIN-GFP or a TEPSIN-GFP construct carrying two point mutations,
L470S and F471S. These mutations abolished binding to the AP4B1 appendage domain
in vitro (Section 4.2.1). The mutations were introduced into a construct containing
C-terminally GFP-tagged TEPSIN (Borner et al., 2012) by site-directed mutagenesis by
Meredith Frazier. Wild-type and mutant [L470S/F471S]TEPSIN-GFP were then amplified
by PCR and inserted in a pLXIN retroviral vector using Gibson Assembly. Successful
cloning was confirmed by diagnostic digests and sequencing (data not shown). Then
stable cell lines were generated by the transduction of wild-type HeLa cells with retrovirus
for the wild-type and mutant TEPSIN-GFP, followed by selection for stable expression by
the addition of G418. Flow cytometry of mixed populations of cells stably expressing
the wild-type and mutant TEPSIN-GFP constructs demonstrated very similar levels of
expression (data not shown) and this was confirmed by Western blotting with antibodies
against GFP and TEPSIN (Figure 4.9A). The Western blotting with anti-TEPSIN revealed
that the TEPSIN-GFP constructs were greatly overexpressed relative to the endogenous
expression level of TEPSIN.
AP-4 complexes were immunoprecipitated from each cell line with an antibody against
AP4B1 and co-immunoprecipitation of TEPSIN-GFP was assayed by probing West-
ern blots with an antibody against GFP (Figure 4.9B). This showed greatly reduced
co-immunoprecipitation of mutant TEPSIN GFP [L470S/F471S] relative to wild-type
TEPSIN-GFP, supporting an important role for the C-terminal TEPSIN motif in binding to
168 Functional characterisation of AP-4 cargo and machinery
Fig. 4.8 Disruption of the AP4B1 appendage domain TEPSIN-binding site does not prevent
TEPSIN recruitment to the membrane. Widefield imaging of AP4B1 knockout HeLa cells stably
rescued with wild-type or mutant (earless, Y682V or I669A/A670S) AP4B1, labelled with anti-
AP4E1 and anti-TEPSIN. TEPSIN colocalises with AP-4 in the perinuclear region (insets) in all
four rescued cell lines. Scale bar: 20 μm.
4.2 Interaction between AP-4 and TEPSIN 169
Fig. 4.9 Disruption of the TEPSIN AP4B1-binding site reduces, but does not abolish, TEPSIN
binding to AP-4 in vivo. (A) Western blot of whole-cell lysates from control (wild-type HeLa) and
HeLa cells stably overexpressing wild-type or mutant [L470S/F471S]TEPSIN-GFP, with antibodies
against GFP and TEPSIN. TEPSIN-GFP is marked with an asterisk and endogenous TEPSIN (of
which there are two isoforms) is marked by arrow heads. The control cells were only blotted for
TEPSIN, not GFP. TEPSIN-GFP is overexpressed a great deal relative to the endogenous expression
level of TEPSIN. (B) Western blot of immunoprecipitates of AP4B1 from extracts of HeLa cells
stably expressing either wild-type or mutant [L470S/F471S] TEPSIN-GFP, with antibodies against
GFP, AP4E1 and AP4B1. Mutant TEPSIN-GFP co-immunoprecipitates with AP-4 less than wild-
type TEPSIN-GFP. (C) As in B, but of immunoprecipitates of AP4E1. AP4B1 was not detected in
the input lysates. (D) Western blot of imumunoprecipitates of GFP (GFP-Trap) from extracts of
control (wild-type HeLa) and HeLa cells stably expressing GFP alone, or wild-type or mutant
[L470S/F471S] TEPSIN-GFP (TG), with antibodies against AP4E1, AP4B1 (marked with an arrow
head) and GFP. Less AP-4 co-immunoprecipitates with mutant TEPSIN-GFP than with wild-type
TEPSIN-GFP.
170 Functional characterisation of AP-4 cargo and machinery
AP-4. When the experiment was repeated using an antibody against AP4E1 to immuno-
precipitate AP-4 the results were similar, but the reduction in the amount of mutant
TEPSIN-GFP that was co-immunoprecipitated was less than in the AP4B1 immuno-
precipitation (Figure 4.9C). We confirmed these effects by performing the reciprocal
immunoprecipitation experiment. Wild-type or mutant TEPSIN-GFP were immuno-
precipitated using GFP-trap beads and Western blots of the immunopreciptates were
probed with antibodies against AP4B1 and AP4E1 (Figure 4.9D). In line with the other
immunoprecipitation experiments, less AP-4 was co-immunoprecipitated with the mu-
tant TEPSIN-GFP than with the wild-type TEPSIN-GFP. The failure of the L470S/F471S
mutations to completely abolish binding of TEPSIN-GFP to AP-4 in vivo provides further
evidence to support the existence of one or more additional TEPSIN-AP-4 interaction
sites. Finally, we tested whether the L470S/F471S mutations prevented membrane re-
cruitment of TEPSIN-GFP by using immunofluorescence microscopy (Figure 4.10). HeLa
cells stably expressing wild-type or mutant [L470S/F471S] TEPSIN-GFP were fixed and
labelled with an antibody against AP4E1. As expected, wild-type TEPSIN-GFP colo-
calised with AP4E1 as delicate puncta in the TGN region. However, we did not detect
colocalisation between mutant TEPSIN-GFP and AP4E1 and instead the mutant TEPSIN-
GFP appeared purely cytosolic. This was surprising given that mutation of the AP4B1
appendage domain TEPSIN-binding site did not prevent recruitment of endogenous
TEPSIN to the membrane (Figure 4.8). This difference may be due to the high level of
overexpression of TEPSIN-GFP in this experiment. Even in the cells expressing the wild-
type TEPSIN-GFP it was hard to detect the perinuclear puncta against the high cytosolic
background. It is possible that there is some recruitment of the mutant TEPSIN-GFP to
the membrane, but that it is not enough to see above the background. Alternatively, be-
cause the cells are expressing endogenous as well as GFP-tagged TEPSIN, the wild-type
endogenous TEPSIN may outcompete the mutant TEPSIN-GFP for binding to AP-4.
Taking all the in vivo experiments together we conclude that the interaction between
the TEPSIN C-terminal motif (LFxG[M/L]x[L/V]) and the AP4B1 appendage domain is
important for binding between TEPSIN and AP-4 in vivo. However, it appears there must
be at least one additional interaction site to explain the residual binding and membrane
recruitment of TEPSIN that occurs even in the absence of the AP4B1 appendage domain.
4.2 Interaction between AP-4 and TEPSIN 171
Fig. 4.10 Mutant TEPSIN-GFP (L470S/F471S) does not colocalise with AP4E1. Widefield imag-
ing of HeLa cells stably expressing either wild-type or mutant TEPSIN-GFP, labelled with anti-
AP4E1. Wild-type TEPSIN-GFP has a cytosolic distribution but is also found to colocalise with
AP4E1 in the perinuclear region (insets). Colocalisation could not be detected between mutant
TEPSIN-GFP and AP4E1. Scale bar: 20 μm.
172 Functional characterisation of AP-4 cargo and machinery
4.3 ATG9A in AP-4-deficient cells
The proteomics data presented in Chapter 3 strongly support that the highly conserved
transmembrane protein ATG9A is an AP-4 cargo protein. It was robustly identified as
mislocalised in AP-4 knockout HeLa cells using Dynamic Organellar Maps (Section 3.3)
and was depleted in vesicle-enriched fractions prepared from AP-4 ablated cells (Section
3.4). In addition, ATG9A was found in close proximity to AP-4 using BioID (Section 3.5)
and was enriched in sensitive low detergent immunopreciptations of AP-4 via TEPSIN-
GFP (Section 3.6). In this section, data is presented that further supports our proteomic
findings and provides further information about the effect of AP-4 deficiency on ATG9A.
4.3.1 ATG9A accumulates at the TGN of AP-4-deficient HeLa cells
At steady state in unstressed cells ATG9A is known to localise to the TGN (Young et al.,
2006). AP-4 also localises at steady state to the TGN (Dell’Angelica et al., 1999a; Hirst
et al., 1999) and so is presumed to mediate cargo sorting at the TGN. Based on this
we hypothesised that AP-4 may sort ATG9A into vesicles at the TGN, and that in AP-4-
deficient cells, ATG9A may accumulate at the TGN because of a block in this transport
step. Our Dynamic Organellar Maps experiment supports this hypothesis because
neighbourhood analysis of ATG9A in wild-type and AP-4 knockout HeLa cells suggested
a shift from mainly endosomal neighbours towards neighbours that localise to the TGN
(Section 3.3.2). To investigate this hypothesis we used immunofluorescence microscopy
to characterise the localisation of ATG9A in wild-type and AP-4 knockout HeLa cells.
In wild-type HeLa cells ATG9A was detected as fine puncta throughout the cell with
increased density in the juxtanuclear region (Figure 4.11), consistent with previous data
(Young et al., 2006). AP4E1-positive puncta were found in the same juxtanuclear region
and there was partial colocalisation between ATG9A and AP4E1 in this region. In contrast,
there was a striking accumulation of ATG9A in the perinuclear region of the knockout
cells. Co-labelling of wild-type and AP-4 knockout HeLa cells with antibodies against
ATG9A and the TGN marker protein TGN46 confirmed that ATG9A accumulates at the
TGN in both knockout cell lines (Figure 4.12). ATG9A similarly accumulated at the TGN
of HeLa cells depleted of AP-4 by siRNA-mediated knockdown (Figure 4.13). Importantly,
the mislocalisation of ATG9A was completely rescued in the AP4B1 knockout HeLa
cell line stably rescued with wild-type AP4B1, confirming that it is caused by lack of
functional AP-4 complex (Figure 4.12B).
4.3 ATG9A in AP-4-deficient cells 173
Fig. 4.11 ATG9A and AP-4 partially colocalise in the perinuclear region. Widefield imaging of
wild-type and AP4E1 knockout HeLa cells, labelled with anti-ATG9A and anti-AP4E1. ATG9A and
AP4E1 localise to delicate puncta which partially colocalise in the perinuclear region (inset). In
the AP4E1 knockout cells ATG9A accumulates in the perinuclear region. Scale bar: 20 μm.
174 Functional characterisation of AP-4 cargo and machinery
Fig. 4.12 ATG9A accumulates at the TGN of AP-4 knockout HeLa cells. (A) Widefield imaging
of wild-type and AP4E1 knockout HeLa cells, labelled with anti-ATG9A and anti-TGN46. Scale
bar: 20 μm. (B) Widefield imaging of wild-type, AP4B1 knockout and AP4B1 knockout HeLa cells
stably expressing AP4B1 (rescue), as in A.
4.3 ATG9A in AP-4-deficient cells 175
Fig. 4.13 ATG9A accumulates at the TGN of AP-4 knockdown HeLa cells. Widefield imaging of
HeLa cells mock treated with oligofectamine (no siRNA) or treated with siRNA to knock down
AP-4, labelled with anti-ATG9A and anti-TGN46. Scale bar: 20 μm.
176 Functional characterisation of AP-4 cargo and machinery
In order to quantify the AP-4 missorting phenotype, we developed a TGN translocation
assay on an automated microscope system (CellInsight CX7 High-Content Screening
Platform, Thermo Fisher Scientific). Wild-type, AP4E1 knockout, AP4B1 knockout and
AP4B1 knockout HeLa cells stably rescued with AP4B1, plated in 96-well glass-bottomed
microplates, were labelled with anti-ATG9A, anti-TGN46 and a whole cell mask. The
relative overlap of the ATG9A signal with the TGN46 signal was quantified with an adapta-
tion of the Colocalization Bioapplication Version 4 (Cellomics, Thermo Fisher Scientific).
The whole cell mask defined the total area of the cell and the anti-TGN46 signal was used
to segment the TGN. The average intensity of the ATG9A signal was quantified in the
TGN region and in the rest of the cell (whole cell minus TGN), and a ratio was calculated
between the intensities of the two areas. The experiment was performed in biological
triplicate and more than 1,400 cells were scored per cell line in each replicate. Ratios
were normalised to the mean wild-type ratio and were log-transformed for statistical
analysis by one-way ANOVA with Dunnett’s Multiple Comparison Test for significant
differences from the wild-type cells. The results of this analysis are displayed in Figure
4.14. There was a significant increase in the relative amount of ATG9A that overlapped
with the TGN46 signal in both AP-4 knockout cell lines (p = 0.0005 for AP4E1 and p <
0.0001 for AP4B1), whereas there was no significant difference between the amount of
ATG9A/TGN46 overlap between the wild-type and AP4B1 rescued cells (p = 0.63).
4.3.2 ATG9A accumulates at the TGN of AP-4-depleted
neuroblastoma-derived SH-SY5Y cells
Following on from our finding of ATG9A accumulation at the TGN of AP-4-deficient HeLa
cells, we wanted to determine whether ATG9A would be similarly affected by loss of AP-4
in a cell line more relevant to the neuronal phenotypes of AP-4 deficiency. For this we
chose to use the neuroblastoma-derived human cell line SH-SY5Y (Biedler et al., 1978).
The SH-SY5Y cell line is a popular ‘neuronal’ cell model and is a sub-clone of the SK-N-
SH cell line, which originates from a bone marrow biopsy from a four year old female
with a metastatic neuroblastoma. SH-SY5Y cells retain biochemical and morphological
properties of neurons and can be differentiated into a mature neuron-like phenotype
by the addition of retinoic acid and brain-derived neurotrophic factor (Encinas et al.,
2000). Initially we attempted to knock out AP4B1 and AP4E1 in SH-SY5Y cells using the
same double nickase CRISPR/Cas9 system we had used to generate the AP-4 knockout
HeLa cells (Section 3.2). However, SH-SY5Y cells are not as amenable to transfection
as HeLa cells, nor do they tolerate single cell cloning well, so we could not recover
4.3 ATG9A in AP-4-deficient cells 177
Fig. 4.14 Quantification of the TGN accumulation of ATG9A in AP-4 knockout HeLa cells. The
ratio of ATG9A labelling intensity between the TGN and the rest of the cell was quantified using
an automated microscope. Ratios were normalised to the mean wild-type ratio. The experiment
was performed in biological triplicate (mean indicated, n = 3) and > 1,400 cells were scored
per cell line in each replicate. Log-transformed ratios were subjected to one-way ANOVA with
Dunnett’s Multiple Comparison Test for significant differences from the wild-type: *** p ≤ 0.001;
ns p > 0.05.
any AP-4-deficient clones using this method. Therefore, we decided to use a lentiviral
CRISPR/Cas9 system (as used by the Lehner Lab in Timms et al., 2016) to deplete AP4B1
and AP4E1 in mixed populations of SH-SY5Y cells. For this we made use of some of the
same guides that were used in the double nickase approach but cloned them into the
BbsI site of the lentiviral sgRNA expression vector pKLV-U6gRNA(BbsI)-PGKpuro2ABFP
(Koike-Yusa et al., 2014). This is a bicistronic vector that allows for co-expression of an
sgRNA with an expression cassette which consists of a puromycin resistance gene and a
blue fluorescent protein (BFP). Successful cloning was confirmed by diagnostic digests
and sequencing (data not shown). In this lentiviral CRISPR/Cas9 system the Cas9 is
delivered via a separate vector, pHRSIN-SFFV-FLAG-Cas9-PPGK-Hygro, a bicistronic
vector for the co-expression of wild-type Cas9 nuclease with a hygromycin resistance
gene. SH-SY5Y cells were first transduced with lentivirus for delivery of the Cas9 vector
and were selected for stable expression by addition of hygromycin. Cas9-expressing
SH-SY5Y cells were then transduced a second time with lentivirus for the delivery of the
sgRNA vectors targeting AP4B1 or AP4E1 and were selected for stable expression by the
addition of puromycin. By this process we generated mixed populations of SH-SY5Y
178 Functional characterisation of AP-4 cargo and machinery
cells that stably co-expressed the Cas9 nuclease and a sgRNA to target either AP4B1 or
AP4E1.
Western blotting of whole cell lysates from these mixed populations of cells with anti-
bodies against AP4B1 and AP4E1 revealed varying degrees of AP-4 depletion depending
on the guide that was used (Figure 4.15A). The most effective guide for targeting AP4B1
was 2A. This resulted in a reduction of AP4B1 below the level of detection by Western
blot, with no visible truncation products. The level of AP4E1 was also clearly reduced in
this population, demonstrating loss of AP-4 complex formation in the cells. In contrast
residual AP4B1 was detected in the cells expressing AP4B1 guides 1A and 2B and the
AP4E1 level was accordingly reduced to a lesser extent in these cells. The best AP4E1
guide was 6B as cells expressing this guide had the lowest level of AP4E1 and also a clear
reduction in the level of AP4B1. Guide 6A was almost as good, whereas guide 7A had
very little effect on the level of AP4E1 or AP4B1. Based on this, AP4B1 population 2A and
AP4E1 population 6B were chosen for further analysis. Immunofluorescence microscopy
with antibodies against ATG9A and TGN46 showed that in comparison to SH-SY5Y cells
that expressed only Cas9, depletion of AP-4 resulted in a striking accumulation of ATG9A
at the TGN (4.15B).
4.3.3 ATG9A mislocalisation is a ubiquitous phenotype in cells from
AP-4-deficient patients
We next wanted to test if ATG9A missorting also occurs in individuals with AP-4 defi-
ciency. To test this we analysed fibroblasts from patients with homozygous mutations
in one of the four AP-4 genes (Abou Jamra et al., 2011; Hardies et al., 2015; Kong et al.,
2013; Verkerk et al., 2009). The fibroblasts lines were acquired or established by Dr
Jennifer Hirst in the Robinson Lab and loss of AP-4 function has been confirmed in each
case by Western blotting, immunofluorescence and immunoprecipitation experiments
(Borner et al., 2012; Hardies et al., 2015; Hirst et al., 2013b; Kong et al., 2013). Immunoflu-
orescence microscopy analysis with antibodies against ATG9A and TGN46 was used
to compare fibroblasts from the AP-4-deficient patients with those from two healthy
control individuals (Figure 4.16). In the control fibroblasts ATG9A was found as puncta
throughout the cell, with only a subtle concentration in the TGN area. In contrast there
was a dramatic accumulation of ATG9A at the TGN of cells from patients with mutations
in any of the four AP-4 subunits. In addition we analysed fibroblasts from an individual
with a heterozygous loss-of-function mutation in AP4E1 - the phenotypically normal
mother of the homozygous AP4E1 patient. In this case ATG9A localisation appeared nor-
4.3 ATG9A in AP-4-deficient cells 179
Fig. 4.15 ATG9A accumulates at the TGN of AP-4-depleted neuroblastoma-derived SH-SY5Y
cells. SH-SY5Y cells stably expressing Cas9 were transduced with sgRNAs to AP4B1 or AP4E1.
Mixed populations were selected for sgRNA expression and parental Cas9-expressing SH-SY5Y
cells were used as a control. (A) Western blot of whole cell lysates, with antibodies against AP4E1
and AP4B1 (marked with arrow heads). The asterisk marks a non-specific band detected by the
AP4E1 antibody. Two different exposures of the blot are shown. (B) Widefield imaging of SH-SY5Y
cells stably expressing Cas9 only (control), AP4B1-targeted population 2A and AP4E1-targeted
population 6B, labelled with anti-ATG9A and anti-TGN46. Scale bar: 20 μm.
180 Functional characterisation of AP-4 cargo and machinery
mal, which suggests that mislocalisation of ATG9A is a cellular phenotype that correlates
with disease in AP-4 deficiency.
4.3.4 ATG9A levels in AP-4-deficient cells
The TGN accumulation of ATG9A in the fibroblasts from AP-4 deficient patients was so
strong that it indicated a possible increase in the whole cell expression level of ATG9A
(Figure 4.16). Indeed, Western blotting revealed a large increase in the overall ATG9A
signal in the four patient cell lines, relative to fibroblasts from a healthy control individual
(Figure 4.17A). Interestingly, there also appeared to be an intermediate level of ATG9A
expression in the cells from the unaffected individual with a heterozygous mutation
in AP4E1. This was confirmed by label-free quantitative MS-based analysis of whole
cell lysates from the AP4B1, AP4E1 and AP4M1 patient cells, a control cell line and
the AP4E1 heterozygous cells, performed by Dr Georg Borner. The LFQ intensity for
ATG9A was between seven- and nine-fold higher in the AP-4-deficient patient cell lines
than in the control cell line, whereas there was a roughly two-fold increase in ATG9A in
the heterozygous line. This quantification was only calculated from a single replicate
experiment, but it is consistent with the results of the Western blot. The control cell
line shown in the Western blot in Figure 4.17A was the ‘CB’ line, whereas the control
in the quantitative MS experiment was the ‘GM’ line (both are shown in Figure 4.16).
However, Western blotting has shown both lines to express very similar levels of ATG9A
(Figure 4.17A-2). These results indicate that fibroblasts from AP-4-deficient patients
may compensate for the missorting of ATG9A by increasing its expression.
The greatly elevated ATG9A level in the AP-4 patient fibroblasts is different from the
situation in the AP-4 knockout HeLa cells where SILAC-based quantitative MS showed
no significant alteration in ATG9A level (see Section 3.7.1). The proteomic analysis did
indicate a slight enrichment of ATG9A in the AP4B1 knockout whole cell lysates. This
was confirmed by Western blotting with an antibody against ATG9A (Figure 4.17B) but
it is a very subtle increase in comparison to the strong increase in the AP-4 patient
fibroblasts. ATG9A levels in whole cell lysates from the AP4B1- and AP4E1-depleted
SH-SY5Y cells were also compared by Western blotting to the ATG9A level in the control
Cas9-expressing parental SH-SY5Y cells (Figure 4.17C). Similarly to the AP-4 knockout
HeLa cells, AP-4 depletion did not strongly affect the level of ATG9A in SH-SY5Y cells.
This suggests that different cells respond differently to the effects of AP-4 deficiency and
subsequent missorting of ATG9A. Therefore, it will be important to study this further in
more relevant cell types such as primary neurons.
4.3 ATG9A in AP-4-deficient cells 181
Fig. 4.16 ATG9A mislocalisation is a ubiquitous phenotype in cells from AP-4-deficient pa-
tients. Widefield imaging of fibroblasts from two healthy control individuals, patients with
homozygous loss-of-function mutations in one of the four AP-4 genes, and an individual with a
heterozygous (Het) mutation in AP4E1 (phenotypically normal mother of the AP4E1* patient),
labelled with anti-ATG9A and anti-TGN46. In the merged image, DAPI labelling of the nucleus is
also shown (blue). Scale bar: 20 μm.
182 Functional characterisation of AP-4 cargo and machinery
Fig. 4.17 ATG9A expression level is increased in AP-4-deficient patient fibroblasts. (A) Western
blot of whole cell lysates from a healthy control individual (line ‘CB’), patients with homozygous
mutations in one of the four AP-4 genes, and an individual with a heterozygous (Het) mutation
in AP4E1 (phenotypically normal mother of the AP4E1* patient), with antibodies against AP4E1,
AP4B1 and ATG9A. An antibody against clathrin heavy chain was used as a loading control. (A-2)
Western blot of whole cell lysates from two different healthy control individuals (‘GM’ and ‘CB’
lines), with an antibody against ATG9A. An antibody against clathrin heavy chain was used as a
loading control. The ATG9A expression level was similar in both lines. (B) Western blot of whole
cell lysates from wild-type, AP4B1 knockout and AP4E1 knockout HeLa cells, with an antibody
against ATG9A. An antibody against clathrin heavy chain was used as a loading control. (C)
Western blot of whole cell lysates from parental Cas9-expressing, AP4B1-depleted and AP4E1-
depleted SH-SY5Y cells, with antibodies against AP4E1, AP4B1 and ATG9A. An antibody against
alpha-tubulin was used as a loading control.
In conclusion, we have found that ATG9A localisation depends on AP-4 not only in
HeLa cells, but also in neuroblastoma-derived SH-SY5Y cells and in fibroblasts from
AP-4-deficient patients. The additional finding of highly elevated ATG9A levels in AP-4
patient cells further supports the importance of AP-4 in the regulation of AP-4 trafficking.
This suggests that trafficking of ATG9A from the TGN is a ubiquitous function of AP-4.
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 183
4.4 SERINC1 and SERINC3 in AP-4-deficient cells
In addition to ATG9A, our proteomics analyses identified the related and highly con-
served transmembrane proteins SERINC1 and SERINC3 as AP-4 cargo proteins. Like
ATG9A, SERINC1 and SERINC3 underwent translocations on Dynamic Organellar Maps
of AP-4 knockout cells (Section 3.3), were depleted from vesicle-enriched fractions pre-
pared from AP-4-ablated cells (Section 3.4), and were enriched in sensitive low-detergent
immunoprecipitations of AP-4 complexes (Section 3.6). In addition, when SERINC1 and
SERINC3 were used as BioID baits, AP-4 subunits were highly enriched in pulldowns
of biotinylated proteins, indicating their close proximity to AP-4 in the cell (Section
3.5.3). Following on from these findings we sought to further investigate the effect of
AP-4 deficiency on SERINC localisation. This follow-up proved more challenging than
that of ATG9A due to a lack of suitable reagents. Unfortunately, we have not been able to
find any commercial antibodies that are able to detect endogenous SERINC1 or SER-
INC3. Overexpressed SERINC3 no longer exhibited AP-4-dependent trafficking (see
Section 4.4.1) so we needed to find a way to localise endogenous SERINC1 and SERINC3.
After unsuccessful attempts to generate our own antibodies for immunofluorescence,
CRISPR/Cas9-mediated gene-editing was used to tag endogenous SERINC1 and SER-
INC3 with a fluorescent protein (Section 4.4.3). This enabled us to study the localisation
of endogenous SERINCs and their relationship with ATG9A and AP-4.
4.4.1 Overexpressed SERINC3 loses its dependence on AP-4 for
trafficking
Previous studies of the subcellular localisation of SERINC family members have relied
on overexpressed tagged constructs. Inuzuka and colleagues transiently transfected
COS cells (a cell line derived from monkey kidney tissue) with GFP-tagged SERINC1
and found it to localise to the ER (Inuzuka et al., 2005). Rosa and colleagues transiently
expressed GFP-tagged SERINC5 in the immortalised T lymphyocyte cell line Jurkat-TAg
and found it to localise almost exclusively to the plasma membrane (Rosa et al., 2015).
Usami and colleagues similarly found mCherry-tagged SERINC5 to localise to the plasma
membrane when transiently expressed in HeLa and U2-OS cells, but it was additionally
localised to puncta in the perinuclear region (Usami et al., 2015). It is not known whether
these findings accurately reflect the localisation of endogenous SERINC proteins but
in the absence of any way to detect the endogenous proteins, we initially sought to
validate the dependence of SERINCs on AP-4 for trafficking by analysing a HeLa cell
184 Functional characterisation of AP-4 cargo and machinery
line that stably overexpressed SERINC3 C-terminally tagged with HA-mCherry. These
experiments were performed by Dr Georg Borner while working in the Robinson Lab.
Vesicle-enriched fractions were prepared from these cells, with and without knockdown
of AP-4 by siRNA. Western blotting of whole cell homogenates (input material) and the
vesicle-enriched fractions revealed robust depletion of AP-4 from both (Figure 4.18A).
As expected, TEPSIN was also lost from the vesicle-enriched fraction prepared from
the AP-4-depleted cells, demonstrating loss of AP-4 function. However, the level of
SERINC3-HA-mCherry (detected via anti-HA) in both the whole cell homogenate and
the vesicle-enriched fraction was unaffected by AP-4 knockdown. This is in contrast to
the effect of AP-4 knockdown on endogenous SERINC3; our SILAC-based proteomic
profiling of vesicle fractions revealed a minimum five-fold depletion of SERINC3 from
vesicle-enriched fractions prepared from AP-4 knockdown cells (see Table 3.4 in Section
3.4.1). Similarly, AP-4 knockdown had no obvious effect on the localisation of SERINC3-
HA-mCherry. In fixed control and AP-4-depleted cells it localised both to the plasma
membrane and to bright perinuclear puncta, similar to the localisation described for
exogenously expressed SERINC5 (Figure 4.18B). These data suggest that overexpression
of SERINC3 results in an alteration of its trafficking and loss of its dependence on AP-4
for its presence in the vesicle-enriched fraction.
4.4.2 A CRISPR/Cas9-mediated gene-editing approach to localise
endogenous SERINC1 and SERINC3
Following the finding that overexpressed SERINC3 is no longer trafficked in an AP-
4-dependent manner, it was obvious that we would need to find a way to study the
localisation of endogenous SERINC1 and SERINC3. Our first approach was to generate
our own SERINC1 and SERINC3 antibodies, and we tried both rabbit polyclonal and
mouse monoclonal antibodies. Unfortunately neither attempt produced antibodies that
were capable of detecting the endogenous proteins for immunofluorescence, although
one polyclonal SERINC3 antibody did detect a band on a Western blot of the expected
size (53 kDa), which was absent in whole cell lysates from SERINC3 knockdown cells
(Figure 4.19A). None of the SERINC1 antibodies detected a band of the expected size
that knocked down, but two polyclonal SERINC1 antibodies (raised against the same
antigen) detected a much smaller band on a Western blot (∼20 kDa), which seemed to
be specific for SERINC1 because it was not present in whole cell lysates from SERINC1
knockdown cells (Figure 4.19B). This suggests the possibility that there is either a much
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 185
Fig. 4.18 Overexpressed SERINC3 loses its dependence on AP-4 for trafficking. (A) Vesicle-
enriched fractions were prepared from HeLa cells stably overexpressing SERINC3-HA-mCherry,
with and without knockdown of AP-4 by siRNA. Western blots of input material (whole cell
homogenates) and vesicle-enriched fractions, with antibodies against HA (to detect SERINC3-
HA-mCherry), AP4E1 and TEPSIN. An antibody against clathrin heavy chain was used as a loading
control. Loss of TEPSIN from the vesicle-enriched fraction prepared from the AP-4 knockdown
cells confirmed loss of AP-4 function. Unlike endogenous SERINCs, overexpressed SERINC3-HA-
mCherry was not depleted from the AP-4 knockdown vesicle-enriched fraction. (B) Widefield
imaging of HeLa cells stably expressing SERINC3-HA-mCherry, with and without knockdown of
AP-4 by siRNA (the same cells used in A). The cells were fixed in formaldehyde and the mCherry
signal was imaged without antibody amplification. There was no obvious difference between
the localisation of SERINC3-HA-mCherry in control and AP-4-depleted cells. These data were
generated by Dr Georg Borner.
186 Functional characterisation of AP-4 cargo and machinery
Fig. 4.19 SERINC1 and SERINC3 antibodies. Rabbit polyclonal antibodies were raised against
predicted cytosolic (IN) and extracellular (OUT) antigens for SERINC1 and SERINC3.(A) Western
blots of whole cell lysates from control HeLa cells (untreated) and HeLa cells treated with siRNA to
knock down (KD) SERINC3, probed with each SERINC3 antibody. Only SERINC3_IN_1 detected
a band of the expected size (53 kDa) that was not present in the KD lysate. (B) Western blots of
whole cell lysates from control HeLa cells (untreated) and HeLa cells treated with siRNA to KD
SERINC1, probed with each SERINC1 antibody. None of the antibodies detected a band of the
expected size (50 kDa) but SERINC1_IN_1 and SERINC1_IN_2 detected a small band (∼20 kDa)
that was absent in the KD lysate.
smaller splice isoform of SERINC1 (none are predicted) or a cleavage or degradation
product of some kind.
As our antibody approach was unsuccessful we designed a CRISPR/Cas9-mediated
gene-editing approach to knock in fluorescent tags at the endogenous loci for SERINC1
and SERINC3. When we used CRISPR/Cas9-mediated gene editing to create our AP-4
knockout cells we exploited the tendency of the non-homologous end-joining (NHEJ)
repair pathway to introduce errors such as indels, which cause frameshift of the opening
read frame and thus knock out the targeted gene. Although this is the most frequently
used double-strand break (DSB) repair pathway in mammalian cells (Mao et al., 2008),
cells do have a much more precise method for repairing DSBs called homology-directed
repair (HDR). HDR uses a repair template homologous to the damaged site, which would
naturally be the sister chromatid or homologous chromosome. This process can be
manipulated to perform precise gene-editing events (e.g. the addition of a tag or the
replacement of a mutant by a wild-type allele) through the supply of an exogenous repair
template (e.g. Ran et al., 2013).
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 187
Based on our knowledge of SERINC topology we decided to place the tag at the cytosolic
C termini of SERINC1 and SERINC3. The chosen tag was Clover, an engineered form
of Aequorea victoria GFP (Lam et al., 2012), as we hoped its increased brightness might
enable us to detect the SERINCs without antibody amplification and thus allow live cell
imaging. Given the low expression level of SERINCs (Figure 3.27 in Section 3.8) and
the likelihood that we would achieve only monoallelic insertion of the tag (HDR is an
inefficient process) we suspected this would be unlikely, but Clover is similar enough to
GFP that antibodies to GFP can be used for its detection. As HDR is a much less efficient
repair pathway than NHEJ, we decided to use wild-type Cas9 to induce DSBs, rather than
the more accurate, but less efficient, Double Nickase system that we used for generating
the AP-4 knockouts.
The knockin approach is outlined in Figure 4.20. The first step in the tagging process
was to design suitable sgRNAs that would introduce a DSB break as close as possible
to the C terminus of SERINC1 and SERINC3. As before, we made use of the Zhang
lab CRISPR Design Tool (http://crispr.mit.edu/; Hsu et al., 2013) and identified guides
that would target sites 20-100 bp downstream from the termination codon of each
gene (see Section 2.4 for guide sequences). By cutting the DNA downstream of the
termination codon we would avoid damaging the open reading frame of the gene in
any alleles that were not repaired by HDR. Two guides were chosen to target exon 10 of
SERINC1 transcript ENST00000339697 and two to target exon 10 of SERINC3 transcript
ENST00000342374. Guides were cloned into the pX330 vector using a BbsI site that
allows scarless introduction of the guide before the sgRNA scaffold sequence. Similar
to the pX335 vector used for the Double Nickase system (Figure 3.3C in Section 3.2.2),
pX330 is a bicistronic vector for co-expression of human codon-optimised wild-type
S. pyogenes Cas9 and a chimeric sgRNA. As in pX335, sgRNA expression is driven by
a U6 promoter so, unless guide sequences already began with G, an additional G was
added to the 5’ end of the guides. Successful cloning of the sgRNA/Cas9 constructs was
confirmed by diagnostic digests and sequencing (data not shown).
Next we needed to design repair templates that contained the Clover tag flanked by
5’ and 3’ homology arms for SERINC1 or SERINC3, which would enable homologous
recombination between the target site and the repair template (Figure 4.20). For this
we adapted the pDonor_myc-Clover tagging plasmid, which was a kind gift from Dick
van den Boomen in the Lehner Lab. This vector contains a 5’ homology region (∼800
bp) followed by cDNA for a myc epitope tag fused to Clover, an internal ribosome entry
site (IRES), a puromycin resistance gene, and finally a 3’ homology region (∼800 bp).
Homology regions for SERINC1 and SERINC3 were amplified by PCR from genomic DNA
188 Functional characterisation of AP-4 cargo and machinery
Fig. 4.20 CRISPR/Cas9-mediated approach to knock in Clover tags at the C termini of SER-
INC1 and SERINC3. Both SERINC1 and SERINC3 genes have 10 exons with the termination
codon (STOP) within exon 10. Short guide RNAs (sgRNAs) were designed to target sites 20-100
bp downstream from the termination codons of SERINC1 and SERINC3. A tagging construct
(pDonor_myc-Clover) was designed to contain a myc-Clover tag followed by an internal ribosome
entry site (IRES) and a puromycin resistance gene (PuroR), flanked by two homology regions
(HR) for either SERINC1 or SERINC3. Each homology region was roughly 800 bp and the 5’ HR
was the genomic DNA (gDNA) sequence immediately preceding the termination codon while
the 3’ HR was designed to be downstream of the sgRNA target sites. Co-transfection of cells
with an sgRNA, Cas9 and a tagging construct allowed introduction of a double-strand break and
incorporation of the myc-Clover_IRES_PuroR by homologous recombination.
harvested from HeLa cells. The 5’ homology regions were roughly 800 bp of sequence
immediately upstream of the STOP codons and the 3’ homology regions were roughly 800
bp starting downstream of the sgRNA target sites, so the repair template could not be cut
by Cas9. The existing homology regions in pDonor_myc-Clover were removed and the
SERINC1 and SERINC3 homology regions were assembled into the vector using Gibson
Assembly. Successful cloning was confirmed by diagnotic digests and sequencing. The
inclusion of the IRES and puromycin resistance gene following myc-Clover means that
it is possible to select for successful recombination events using puromycin. When the
myc-Clover_IRES_PuroR is correctly integrated at the targeted genomic locus, expression
of the endogenous gene will result in the generation of a single transcript encoding both
the myc-Clover-tagged protein of interest and the puromycin resistance gene. This
provides a mean to overcome the low efficiency of HDR because the only cells that will
survive selection should express a tagged protein.
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 189
4.4.3 Knock in of Clover tags at the C termini of SERINC1 and
SERINC3 in HeLa cells
To generate the SERINC1 and SERINC3 knockin cell lines, HeLa cells were co-transfected
with a pX330 plasmid containing the sgRNA and Cas9 and a pDonor_myc-Clover tag-
ging plasmid in a 1:1 ratio. Forty-eight hours later the transfected cells and a dish of
untransfected control cells were placed under puromycin selection until all the control
cells had been killed off, to select for genomic integration of the tagging construct. One
population of the SERINC1 knockin cells (guide B) succumbed to an infection and had to
be discarded. Two weeks after transfection there were sufficient cells to analyse by flow
cytometry, freeze some stocks of the mixed populations and seed the cells for single cell
cloning. Analysis by flow cytometry revealed around 50 % of the cells that had survived
selection to be GFP-positive (Figure 4.21A). Theoretically the cells should not be able to
express the puromycin resistance gene without also expressing the Clover tag, suggesting
some unexpected recombination may have occurred. The GFP signal from the SERINC1
knockin cells was brighter than that from the SERINC3 knockin cells, as expected based
on proteomic estimation of their copy numbers (Itzhak et al., 2016; see Figure 3.27 in Sec-
tion 3.8). Single cell clones were picked and screened for expression of tagged SERINC1
or SERINC3 by Western blot with an antibody against GFP (Figure 4.21B and C). SERINC1
clones A2 and A3 expressed a product of the expected size, as did SERINC3 clones A1,
A3, A5 and B6. Interestingly, the SERINC1 knockin clones had a second smaller GFP
positive band of roughly 45 kDa in size. Two of our SERINC1 antibodies detected a band
on a Western blot of around 20 kDa which knocked down with siRNA targeting SERINC1
(Figure 4.19B). As the myc-Clover tag is approximately 27 kDa, the ∼45 kDa band in the
SERINC1 knockin cells could represent this smaller SERINC1 product, plus the tag. To
look for further evidence of the existence of a small SERINC1 product, Dr Georg Borner
analysed MS data from 10 vesicle fraction experiments in which the vesicle fractions
were run on SDS-PAGE gels to separate proteins according to size and the gels were
cut into 10 individual slices for peptide extraction and MS analysis. The distribution of
peptides corresponding to a given protein over the 10 gel slices gives an estimation of
protein size, relative to proteins of known molecular weight (Figure 4.21D). As expected,
SERINC1 peptides peaked around the 50 kDa mark (similar to AP4M1), but there was
also a separate smaller peak at a lower molecular weight, similar to the peak for AP4S1
which has a molecular weight of 17 kDa. This supports the existence of a lower molecular
weight SERINC1 product, the nature of which invites further investigation.
190 Functional characterisation of AP-4 cargo and machinery
Fig. 4.21 Knock in of Clover tags at the C termini of SERINC1 and SERINC3 in HeLa cells. (Full
caption on following page.)
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 191
Fig. 4.21 Knock in of Clover tags at the C termini of SERINC1 and SERINC3 in HeLa cells. (A)
Mixed populations of SERINC1-Clover (guide A), SERINC3-Clover (guide A) and SERINC3-Clover
(guide B) knockin (KI) HeLa cells were analysed by flow cytometry. Approximiately 50% of each
population were positive for Clover fluorescent signal, relative to non-expressing control HeLa
cells. SERINC1-Clover was brighter than SERINC3-Clover, as expected from proteomic copy
number estimations (Itzhak et al., 2016). (B) The SERINC1-Clover KI cells were single cell cloned
and clones were analysed by Western blot with an antibody against GFP (to detect Clover). Wild-
type HeLa cell lysates were used as a negative control. Clones A2 and A3 expressed a band of
the expected size (∼78 kDa), but also expressed a smaller GFP-positive band of ∼45 kD in size.
(C) The SERINC3-Clover KI cells were single cell cloned and analysed as in B. Clones A1, A3, A5
and B6 expressed a band of the expected size (∼80 kDa). (D) Proteomics analysis carried out
by Georg Borner to estimate the molecular weight of proteins identified in mass spectrometric
measurements of 10 vesicle fractions. Proteins in the vesicle fractions were separated according
to size by SDS-PAGE and the gels were cut into 10 individual slices where fraction 1 contained the
lowest molecular weight proteins and fraction 10 contained the highest. The relative distribution
of peptides for a given protein across the gel fractions provides an estimation of size, relative
to proteins of known molecular weight (AP4M1, AP4S1 and CLTC were used in this instance).
SERINC1 and SERINC3 both peak around the 50 kDa mark as expected, but SERINC1 has a
bimodal distribution with a second peak around the 20 kDa mark.
The SERINC knockin clones were further screened by immunofluorescence microscopy
with an antibody against GFP (4.22A and B). HeLa SERINC1-Clover clone A3 and HeLa
SERINC3-Clover B6 both had uniform expression across the cell population so were
selected for further analysis. Both SERINC1-Clover and SERINC3-Clover localised as fine
puncta throughout the cell, with a concentration in the perinuclear region. As indicated
by the flow cytometry (Figure 4.21A), SERINC1-Clover was brighter than SERINC3-Clover.
Without antibody amplification SERINC1-Clover was only just detectable above back-
ground autofluorescence and SERINC3-Clover was not detectable above the background.
This suggests the expression levels are probably too low to allow for live cell imaging,
unless a microscope with much higher sensitivity can be used. However, the anti-GFP
antibody amplification works well, so we can now localise endogenous SERINC pro-
teins. Correct integrations of the tag at the endogenous genomic loci were confirmed
by isolating genomic DNA from the cells and amplifying the region surrounding the
integration site by PCR with a forward primer to the C terminus of SERINC1 or SERINC3
and a reverse primer within the Clover tag. PCR products were blunt-end cloned and
sequenced, confirming in-frame fusion of myc-Clover to the C termini of SERINC1 and
SERINC3 (data not shown).
192 Functional characterisation of AP-4 cargo and machinery
Fig. 4.22 Localisation of SERINC1 and SERINC3 in knockin HeLa cell lines. (A) Widefield imag-
ing of HeLa SERINC1-Clover clone A3, labelled with anti-GFP to detect the Clover fluorescent
tag. A clone that was negative for Clover expression by Western blot (clone A1) was used as a
negative control. The SERINC1-Clover fluorescent signal without antibody amplification is only
just above the background of autofluorescence in the negative control. Scale bar: 20 μm. (B)
Widefield imaging of HeLa SERINC3-Clover clone B6, labelled with anti-GFP to detect the Clover
fluorescent tag. A clone that was negative for Clover expression by Western blot (clone B5) was
used as a negative control. The SERINC3-Clover fluorescent signal without antibody amplifi-
cation could not be detected above the background level of autofluorescence in the negative
control. Scale bar: 20 μm.
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 193
4.4.4 ATG9A and SERINCs colocalise in AP-4-dependent vesicles
Our SERINC knockin cells allowed us to avoid overexpression artefacts, but there was still
the possibility that the introduction of the C-terminal tag itself disrupted the trafficking
of the SERINCs. To test for this HeLa SERINC1-Clover and HeLa SERINC3-Clover cells
were mock transfected or transfected with siRNA to knock down AP-4 (using the same
96 hours 2 hit knockdown protocol as in the proteomic vesicle profiling experiment in
Section 3.4.1). Vesicle-enriched fractions were then prepared from mock transfected and
AP-4 knockdown cells and were analysed by Western blotting with antibodies against GFP
(to detect the Clover-tagged SERINCs) and AP4E1 (Figure 4.23A). Both SERINC1-Clover
and SERINC3-Clover were depleted in the AP-4 knockdown vesicle fractions, suggesting
that the C-terminal tags do not disrupt the AP-4-dependent trafficking of SERINC1 and
SERINC3. Additionally, in the Western blots of whole cell homogenates (input) from
the AP-4 knockdown cells some lower molecular weight GFP-positive bands appeared,
suggesting that the post-translational modification or processing of the SERINCs may
be altered in the absence of AP-4. Having verified that endogenously tagged SERINC1
and SERINC3 are trafficked like the wild-type proteins, we used our SERINC knockin
cells to investigate SERINC localisation in control and AP-4-depleted cells.
The distribution of SERINC1 and SERINC3 to fine puncta throughout the cell was rem-
iniscent of the distribution pattern of ATG9A (Figure 4.11) and, as with ATG9A, there
was partial colocalisation between SERINC1/3 and AP4E1 in the perinuclear region
(Figure 4.23B). This led to the idea that the SERINCs might be present in the same vesicle
population as ATG9A. Indeed, confocal immunofluorescence microscopy analysis of
HeLa SERINC1-Clover and HeLa SERINC3-Clover cells with antibodies against GFP and
ATG9A revealed considerable overlap between ATG9A and the SERINCs in peripheral
puncta (Figure 4.24A and B). Strikingly, AP-4 knockdown resulted in a reduction in the
colocalisation between ATG9A and the SERINCs in the peripheral puncta, suggesting
this is an AP-4-dependent vesicle population.
We next sought to quantify the degree of colocalisation between ATG9A and the SERINCs
in the control and AP-4 knockdown cells. As the peripheral ATG9A- and SERINC-positive
puncta are very small, this required higher resolution and better signal-to-noise ratio
than standard confocal microscopy could offer. Thus we decided to use the Airyscan
feature on our LSM880 confocal microscope (ZEISS; Huff, 2015). In standard confocal
microscopy a pinhole aperture is used to block out-of-focus light from reaching the
detector. This provides increased spatial resolution but comes with the cost of decreased
brightness because the signal from the out-of-focus light is discarded. As the pinhole is
194 Functional characterisation of AP-4 cargo and machinery
Fig. 4.23 Endogenously tagged SERINC1 and SERINC3 are mistrafficked in AP-4-depleted
HeLa cells. (A) Vesicle-enriched fractions were prepared from HeLa SERINC1-Clover and HeLa
SERINC3-Clover knockin cells, with and without knockdown of AP-4 by siRNA (control cells
were untreated). Western blots of input material (whole cell homogenates) and vesicle-enriched
fractions with antibodies against GFP (to detect Clover) and AP4E1. An antibody against clathrin
heavy chain was used as a loading control. As expected, SERINC1- and SERINC3-Clover were
depleted from the vesicle-enriched fractions prepared from AP-4 knockdown cells, indicating
that the tagged SERINCs are trafficked like the endogenous proteins. (B) Widefield imaging
of HeLa SERINC1-Clover clone A3 and SERINC3-Clover clone B6, labelled with anti-GFP and
anti-AP4E1. The SERINCs and AP4E1 partially colocalise in the perinuclear region (insets). Scale
bar: 20 μm.
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 195
Fig. 4.24 ATG9A and SERINC1/3 colocalisation in peripheral puncta is dependent on AP-4. (A)
Confocal microscopy was used to image SERINC1-Clover (via anti-GFP) and anti-ATG9A in HeLa
SERINC1-Clover clone A3, with and without knockdown of AP-4 by siRNA. Control cells were
mock transfected without siRNA. SERINC1 and ATG9A colocalised in peripheral puncta in the
mock transfected cells, and this colocalisation appeared reduced in the AP-4 knockdown cells.
Scale bar: 10 μm. (B) Confocal microscopy was used to image SERINC3-Clover (via anti-GFP)
and anti-ATG9A in HeLa SERINC3-Clover clone B6, with and without knockdown of AP-4 by
siRNA, as in A.
196 Functional characterisation of AP-4 cargo and machinery
made smaller to improve resolution, less light can reach the detector and so there is a
trade-off between optimising spatial resolution and the signal-to-noise ratio. Airyscan-
ning is a detection method that employs a concentric hexagonal array of 32 detector
elements, which act as very small pinholes, to collect all the photons emitted from the
excited sample, not just the in-focus light. The signals from each detector are then
computationally reassigned to their correct position, producing increased resolution
without a compromise to the signal-to-noise ratio. ZEISS estimate a 1.7-fold increase in
resolution in x, y and z dimensions, accompanied with four-to-eight times the signal-
to-noise ratio, compared with the standard confocal mode with a pinhole set to one
Airy unit (Huff, 2015). HeLa SERINC1-Clover and HeLa SERINC3-Clover cells, with and
without AP-4 knockdown and labelled with anti-GFP and anti-ATG9A, were imaged on
the LSM880 in confocal mode with Airyscanning in Superresolution mode (Figure 4.25A
and C). Twenty individual cells were used for quantification of colocalisation for each
condition for SERINC1-Clover and 19 cells were used for each condition for SERINC3-
Clover1 (Figure 4.25B and D). A peripheral 10× 10 μm area was selected in each cell
viewing only the SERINC channel. Colocalisation was measured in these areas using
Pearson’s Correlation Coefficient (PCC) with Costes’ thresholding method (performed in
Volocity software 6.3; Perkin Elmer). PCC measures the strength of the linear relationship
between the fluorescent intensities of pixels in the two channels and has values ranging
from 1 to −1, where PCC = 1 represents perfect correlation and PCC = −1 represents per-
fect anti-correlation (Manders et al., 1992). The Costes’ thresholding method limits the
pixels included in the calculation of the PCC to just those above an intensity threshold
set for each channel (Barlow et al., 2010). The thresholds are automatically set at levels
at which pixels below both threshold values have no or negative correlation and are thus
interpreted as background noise (Costes et al., 2004). The mean thresholded PCC for
ATG9A and SERINC1-Clover in control cells was 0.65 and in AP-4 knockdown cells it was
reduced to 0.31. Likewise, the mean thresholded PCC for ATG9A and SERINC3-Clover
in control cells was 0.54 and in AP-4 knockdown cells it was reduced to 0.26. Statisti-
cal analysis of the difference between the PCC values for control and AP-4 knockdown
cells was performed using a two-tailed Mann-Whitney U test and for both SERINC1
and SERINC3 the depletion of AP-4 significantly reduced their degree of colocalisation
with ATG9A (p ≤ 0.001). A Mann-Whitney U test is a non-parametric alternative to the
t-test and was used in this instance because, unlike the t-test, it does not require the
assumption of a normal distribution.
1Because the tired microscopist couldn’t count to 20.
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 197
Fig. 4.25 Quantification of ATG9A and SERINC1/3 colocalisation with and without AP-4
knockdown. (A) Confocal microscopy with Airyscanning was used to image SERINC1-Clover (via
anti-GFP) and anti-ATG9A in HeLa SERINC1-Clover clone A3, with and without knockdown of
AP-4 by siRNA. Control cells were mock transfected without siRNA. Representative images show
a confocal slice of the whole cell, and a peripheral 10 x 10 μm square imaged with Airyscanning
in Superresolution mode, used for the quantification of colocalisation between SERINC1-Clover
and ATG9A. Note the images are shown here with increased brightness to aid viewing, but quan-
tification was performed on the raw data and care was take to avoid saturation when images were
captured. Scale bar: 10 μm. (B) Quantification of colocalisation between SERINC1-Clover and
ATG9A in peripheral regions of mock treated and AP-4 knockdown (KD) cells, using Thresholded
Pearson’s Correlation Coefficient. 20 cells were quantified per condition. Data show mean (n
= 20) and results of a two-tailed Mann-Whitney U test: ***p ≤ 0.001. (C) Confocal microscopy
with Airyscanning was used to image SERINC3-Clover (via anti-GFP) and anti-ATG9A in HeLa
SERINC3-Clover clone A3, with and without knockdown of AP-4 by siRNA, as in A. (D) Quan-
tification of colocalisation between SERINC3-Clover and ATG9A in peripheral regions of mock
treated and AP-4 KD cells, using Thresholded Pearson’s Correlation Coefficient. 19 cells were
quantified per condition. Data show mean (n = 19) and results of a two-tailed Mann-Whitney U
test: ***p ≤ 0.001.
198 Functional characterisation of AP-4 cargo and machinery
In AP-4-depleted cells colocalisation between the SERINCs and ATG9A in peripheral
puncta was reduced and the SERINCs appeared to accumulate in the perinuclear region.
To test if, like ATG9A, the SERINCs accumulate at the TGN in AP-4-depleted cells, con-
trol and AP-4 knockdown HeLa SERINC1-Clover and HeLa SERINC3-Clover cells were
labelled with antibodies against GFP and TGN46 and imaged by confocal microscopy
(Figure 4.26). Although SERINC1 and SERINC3 did seem to become more concentrated
in the perinuclear region in the AP-4 knockdown cells, there was not an obvious overlap
with the TGN46 labelling. This suggests that the SERINCs accumulate in another juxtanu-
clear compartment, perhaps recycling endosomes, and warrants further investigation.
This is in keeping with our Dynamic Organellar Maps neighbourhood analysis (Section
3.3.2), which supported intra-endosomal translocations for SERINC1 and SERINC3.
As exogenously expressed SERINC5 was previously shown to colocalise with the late
endosomal marker Rab7-RFP in Nef-expressing cells (Rosa et al., 2015), we also looked
at the late endosomal protein LAMP1 (Lysosome-associated membrane glycoprotein 1)
in the SERINC knockin cells (Figure 4.27). In both control and AP-4-depleted cells there
was partial colocalisation between the SERINCs and LAMP1, although higher resolution
imaging is required to ascertain whether they are present on the same compartment or
on closely apposed compartments. Particularly in the control cells the SERINC labelling
appears as finer puncta than the LAMP1-positive structures, so it is possible that the
partial colocalisation represents clusters of SERINC-positive puncta in close association
with LAMP1-positive endosomes. The degree of colocalisation between the SERINCs
and LAMP1 was not obviously altered by AP-4 knockdown, but higher resolution imaging
and quantification are required to confirm this.
To conclude this section, CRISPR/Cas9-mediated gene-editing has provided us with a
means to study the subcellular localisation of SERINC1 and SERINC3. They are found in
a population of small ATG9A-positive vesicles and the colocalisation between ATG9A and
the SERINCs in these vesicles is dependent on AP-4. However, unlike ATG9A, SERINC1
and SERINC3 do not accumulate at the TGN in AP-4-deficient cells but instead appear
to accumulate in another juxtanuclear compartment. Further studies are required to
learn more about the subcellular distribution of the SERINCs and what happens to them
in the absence of AP-4.
4.4 SERINC1 and SERINC3 in AP-4-deficient cells 199
Fig. 4.26 SERINC1 and SERINC3 do not accumulate at the TGN of AP-4-depleted cells. (A)
Confocal imaging of anti-GFP and anti-TGN46 in HeLa SERINC1-Clover clone A3 cells, with or
without knockdown of AP-4 by siRNA (control was mock transfected without siRNA). Scale bar:
10 μm. (B) HeLa SERINC3-Clover clone B6 cells analysed as in A. SERINCs accumulate in the
perinuclear region of AP-4-depleted cells, but do not show obvious overlap with TGN46.
200 Functional characterisation of AP-4 cargo and machinery
Fig. 4.27 SERINC1 and SERINC3 partially colocalise with LAMP1 in control and AP-4-
depleted cells. (A) Confocal imaging of anti-GFP and anti-LAMP1 in HeLa SERINC1-Clover
clone A3 cells, with or without knockdown of AP-4 by siRNA (control was mock transfected
without siRNA). Scale bar: 10 μm. (B) HeLa SERINC3-Clover clone B6 cells analysed as in A.
Partial colocalistion between SERINCs and LAMP1 was observed in control and AP-4-depleted
cells.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 201
4.5 Role of RUSC2 in the peripheral delivery of AP-4
vesicles
The cytosolic proteins RUSC1 and RUSC2 were strongly affected in our proteomic vesicle
profiling experiments presented in Section 3.4. RUSC2 was not identified in our global
or membrane proteome analyses of AP-4 knockout cells (Section 3.7), probably due to its
very low copy number, but RUSC1 was significantly depleted from both the membrane
fractions and the whole cell lysates prepared from AP-4 knockout cells. This suggests
that in the absence of AP-4, RUSCs are not recruited to the membrane, and this leads to
their destabilisation at the whole cell level. The fact that homozygous loss-of-function
mutations in RUSC2 cause a similar neurological disorder to AP-4 deficiency provides
further evidence that the RUSCs are AP-4 accessory proteins. Additionally, RUSC2 was
enriched in the AP-4 BioID experiments when AP4E1 and AP4M1 were used as baits
(Section 3.5), suggesting it may directly interact with the AP-4 complex. For this reason
we focused on follow-up of the interactions between RUSC2, AP-4 and AP-4-associated
proteins (data presented in this section), but it will be important in the future to also
study RUSC1 in the context of AP-4-dependent trafficking.
4.5.1 Vesicle fractionation profiles of AP-4-associated proteins
Dr Georg Borner previously performed fractionation profiling of the vesicle-enriched
fraction of HeLa cells, briefly discussed in Section 3.1.2 (Borner et al., 2014). He prepared
vesicle-enriched fractions from SILAC light- and heavy-labelled HeLa cells. The heavy
vesicle-enriched fraction acted as a reference fraction while the light-labelled vesicle
fraction was further fractionated by differential centrifugation to yield three sub-fractions
of roughly equal total protein content. The process was repeated with a label-swap and
slightly different spin parameters (to improve profile discrimination) and SILAC-based
quantitative MS was then used to generate protein abundance distribution profiles across
the six sub-fractions. These distribution profiles can be used to predict protein-protein
interactions, because proteins which are part of a stable complex have very similar
distributions to other members of the complex. For example, the AP-4 accessory protein
TEPSIN has a similar abundance distribution profile to that of the four AP-4 complex
subunits, whereas AP-1 complex subunits have a distinct fractionation profile from
that of AP-4 (Figure 4.28A). Figure 4.28B shows the abundance distribution profiles for
RUSC1, RUSC2, SERINC1, SERINC3 and ATG9A, in comparison to that of AP4M1. Unlike
TEPSIN, RUSC1 and RUSC2 do not have a similar profile to AP-4, suggesting that they do
202 Functional characterisation of AP-4 cargo and machinery
not form a stable complex with AP-4. This is in keeping with our immunoprecipitation
data (Section 3.6) because the RUSCs were not co-immunoprecipitated with AP-4, even
under sensitive low-detergent conditions. However, the fractionation profiles of the
RUSCs were more similar to those of the AP-4 cargo proteins SERINC1, SERINC3 and
ATG9A. Dr Georg Borner created ‘The Predictor’ which is a database of the profiling data
from the vesicle fractionation profiling experiment (Borner et al., 2014). The Predictor
can be used to compare the profile of a query protein to that of the other proteins in the
dataset by profile similarity calculations based on distances in multidimensional log
space. Once the distances have been calculated between a query protein profile and
every other protein in the database, proteins are ranked according to their similarity
with the query.
Tables 4.1 and 4.2 show the proteins that were ranked as the top 10 most similar profiles
to that of RUSC1 and RUSC2, respectively. For RUSC1, ATG9A and SERINC1 were in the
top 10 with average squared Euclidian distances from RUSC1 similar to that between
TEPSIN and AP-4 (∼0.03). Interestingly, another protein in the list is DAGLB (Sn1-specific
diacylglycerol lipase beta), which has popped up in other AP-4 proteomic experiments.
DAGLB is enriched in sensitive low-detergent immunoprecipitations of AP-4 complexes
(Section 3.6) and is depleted from vesicle-enriched fractions prepared from AP-4-ablated
cells (Section 3.4), suggesting it may be worth following up as an AP-4-associated protein.
For RUSC2 the most similar profile was of SERINC3 and SERINC1 was also within the top
10 most similar profiles. In this case the average squared Euclidian distances were larger
and did not meet the cut-off for our definition of similarity based on subunits of known
protein complexes (see Borner et al., 2014). However, it seems it must be more than
a coincidence that the SERINCs are among the most similar proteins to RUSC2 in the
vesicle fractionation experiment. These data suggest that unlike TEPSIN which remains
closely associated with AP-4 in the cell (as shown by their tight colocalisation), the RUSCs
may interact more transiently with AP-4 and may spend more time in association with
AP-4 cargo proteins than with AP-4 itself.
4.5.2 AP-4 cargo proteins accumulate at the periphery of RUSC2
overexpressing cells
Similarly to the SERINCs, the subcellular distribution of RUSC2 is not well studied and
suffers from a lack of reagents. We have been unable to find any antibodies capable
of detecting the endogenous protein for immunofluorescence. Bayer and colleagues
transiently transfected HeLa cells with HA-tagged RUSC2 and found that it had a punctate
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 203
Fig. 4.28 RUSCs have more similar vesicle fractionation profiles to AP-4 cargo proteins than
to AP-4 itself. Vesicle fractionation profiling data from Borner et al. (2014). Vesicle-enriched
fractions were fractionated by differential centrifugation to give three sub-fractions. This was
performed in duplicate, with slight variations in spin parameters, to give a total of six sub-
fractions which were analysed by SILAC-based quantitative mass spectrometry. (A) Proteomic
profiles of TEPSIN, AP-4 complex subunits (AP4B1/AP4E1/AP4M1/AP4S1) and AP-1 complex
subunits (AP1B1/AP1G1/AP1M1/AP1S1), showing the relative abundance distribution across the
six sub-fractions on a log2 scale. TEPSIN and the four AP-4 subunits have very similar profiles,
predicting their existence in a stable protein complex. (B) Proteomic profiles of AP4M1, RUSC1,
RUSC2, SERINC1, SERINC3, and ATG9A, as in A. Unlike TEPSIN, the RUSCs do not have similar
profiles to AP-4, but rather have similarity to the profiles of AP-4 cargo proteins.
204 Functional characterisation of AP-4 cargo and machinery
Table 4.1 Top 10 similar profiles to RUSC1 from vesicle fractionation profiling data from
Borner et al. (2014). RUSC1 was used to query ‘The Predictor’ database which calculates profile
similarity between a query protein profile and all proteins in the database and returns a list of
proteins ranked according to their similarity with the query. The proteins ranked as the top 10
most similar profiles to RUSC1 are listed with the average squared Euclidean distance from the
RUSC1 profile (measure of similarity) and a similarity rating from a scoring system based on
distances between subunits of known protein complexes: ** = very similar; * = similar.
Rank Gene name Avg. squared distance Similarity rating
1 GNL3L 0.009 **
2 ATG9A 0.026 *
3 DAGLB 0.036 *
4 CNO 0.041 *
5 SERINC1 0.044 *
6 BLOC1S3 0.045 *
7 DKFZp762I1415 0.045 *
8 AP3M2 0.047 *
9 NSUN2 0.047 *
10 BACE2 0.049 *
Table 4.2 Top 10 similar profiles to RUSC2 from vesicle fractionation profiling data from
Borner et al. (2014). RUSC2 was used to query ‘The Predictor’ database which calculates profile
similarity between a query protein profile and all proteins in the database and returns a list of
proteins ranked according to their similarity with the query. The proteins ranked as the top 10
most similar profiles to RUSC2 are listed with the average squared Euclidean distance from the
RUSC2 profile (measure of similarity) and a similarity rating from a scoring system based on
distances between subunits of known protein complexes: F = below cut-off for similarity.
Rank Gene name Avg. squared distance Similarity rating
1 SERINC3 0.147 F
2 SRP19 0.187 F
3 IFRD1 0.223 F
4 MRFAP1 0.280 F
5 EDF1 0.341 F
6 WDR47 0.374 F
7 SERINC1 0.379 F
8 GMIP 0.404 F
9 DKFZp762I1415 0.404 F
10 TMEM184A 0.409 F
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 205
distribution throughout the cell, with an enrichment in the perinuclear area (Bayer et al.,
2005). Another study looked at overexpressed GFP-tagged RUSC2 in the rat neuronal cell
line PC12 and found a similar punctate distribution but in addition to the perinuclear
enrichment of RUSC2, there was also enrichment at the distal ends of neurites where it
colocalised with RAB35 (Fukuda et al., 2011).
In order to investigate the relationship between RUSC2 and AP-4 and its cargo proteins,
we generated HeLa cells lines that stably overexpress GFP-tagged RUSC2. Constructs
for N- and C-terminal tagged RUSC2 were generated by Gibson Assembly and cloned
into a pQCXIH retroviral vector. RUSC2 is a large gene (coding DNA of 4,551 bp) so it
was amplified by PCR as two separate fragments which were then joined seamlessly in
the Gibson Assembly reaction. Successful cloning of both constructs was confirmed by
diagnostic digests and sequencing (data not shown). To generate stable cell lines, HeLa
cells were transduced with retrovirus and selected for stable expression of GFP-tagged
RUSC2 by the addition of Hygromycin. Western blotting of whole cell lysates with an
antibody against GFP showed expression of fusion proteins of the expected size (189
kDa) and the expression level of N-terminally tagged RUSC2 was higher than that of
the C-terminally tagged construct (Figure 4.29A). The GFP signal from both constructs
was visible by fluorescence microscopy without a need for anti-GFP amplification and
revealed a punctate distribution with some perinuclear brightness but also a concentra-
tion in clusters at the periphery of the cell (Figure 4.29B). This was similar to previously
published localisations of exogenously expressed RUSC2 (Bayer et al., 2005; Fukuda et al.,
2011), and the peripheral clusters of RUSC2 puncta were reminiscent of the enrichment
of RUSC2 at the distal ends of neurites that was observed by Fukuda and colleagues. The
expression levels of GFP-tagged RUSC2 were very variable in the mixed populations of
cells so they were single cell cloned and clones with uniform expression were selected
for further imaging experiments.
The vesicle fractionation profiling data (Section 4.5.1) suggested a possible association
between the RUSCs and AP-4 cargo proteins. To investigate this, wild-type HeLa cells
and clonal GFP-tagged RUSC2-expressing cells were mixed on coverslips and analysed
by immunofluorescence microscopy with an antibody against ATG9A (Figure 4.30). The
overexpression of RUSC2 resulted in a dramatic relocation of ATG9A puncta from the
perinuclear region to the cell periphery where they colocalised with the GFP-tagged
RUSC2. There was also clear colocalisation between ATG9A and GFP-tagged RUSC2 in
puncta away from the periphery of the cell. To test if RUSC2 overexpression affected
SERINC1 and SERINC3 in a similar manner, we made use of our SERINC1- and SERINC3-
Clover knockin cell lines. The GFP tag in the pQCXIH_GFP-RUSC2 construct was cut out
206 Functional characterisation of AP-4 cargo and machinery
Fig. 4.29 Generation of HeLa cell lines expressing GFP-tagged RUSC2. (A) Western blot of
whole cell lysates from HeLa cells stably expressing N- or C-terminally GFP-tagged RUSC2,
with an antibody against GFP. Wild-type HeLa cells were included as a negative control. Bands
are of the expected size (189 kDa). Note the 250 kDa marker runs lower on our gel system than
expected - clathrin heavy chain, which has a molecular weight of 192 kDa, also runs on the 250
kDa marker. (B) Widefield imaging of HeLa cells stably expressing N- or C-terminally GFP-tagged
RUSC2, without antibody amplification. Both constructs had a punctate distribution with partic-
ular accumulations in clusters at the periphery of the cells. Expression levels were variable in
both populations. Scale bar: 20 μm.
and replaced by a triple HA tag using flanking NotI and AgeI restriction sites. Successful
cloning was confirmed by diagnostic digests and sequencing. HeLa SERINC1-Clover
clone A3 and HeLa SERINC3-Clover clone B6 were transiently transfected with the HA-
RUSC2 expression construct. The cells were then analysed by immunofluorescence
microscopy with antibodies against GFP (to detect the Clover tags) and HA (Figure 4.31A).
Like ATG9A, SERINC1 and SERINC3 were both relocated from their mainly perinuclear
distribution to the cell periphery where they colocalised with the HA-tagged RUSC2.
RUSC2 overexpression also had the same effect on overexpressed SERINC3-HA (Figure
4.31B). These experiments support a close relationship between the subcellular distribu-
tions of RUSC2 and AP-4 cargo proteins and suggest RUSC2 is involved with delivering
AP-4 cargo proteins to the cell periphery.
Next we investigated the nature of the compartments that accumulated at the periphery
of the RUSC2-overexpressing cells by testing for the presence of other organelle markers
(Figure 4.32). HeLa cells stably expressing GFP-RUSC2 were mixed on coverslips with
parental HeLa cells and analysed by immunofluorescence microscopy with anti-TGN46
(a marker of the TGN), anti-LAMP1 (a marker of late endosomes), anti-EEA1 (a marker
of early endosomes) and anti-RAB11 (a marker of recycling endosomes). In each case
GFP-RUSC2 accmulated at the periphery of the cells without a corresponding accumu-
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 207
Fig. 4.30 ATG9A-positive puncta accumulate at the periphery of RUSC2-overexpressing cells.
Widefield imaging of HeLa GFP-RUSC2 clone 3 and HeLa RUSC2-GFP clone 1, mixed on coverslips
with parental HeLa cells (marked with asterisks), labelled with anti-ATG9A. The insets show
accumulation of RUSC2- and ATG9A-positive puncta at the periphery of the cell. Scale bar: 20
μm.
208 Functional characterisation of AP-4 cargo and machinery
Fig. 4.31 SERINC-positive puncta accumulate at the periphery of RUSC2-overexpressing cells.
(A) Widefield imaging of HeLa cells expressing endogenously tagged SERINC1-Clover or SERINC3-
Clover, transiently transfected with HA-RUSC2 and double labelled with anti-GFP (to detect
Clover) and anti-HA. Non-transfected cells are marked with asterisks. The insets show accu-
mulation of HA-RUSC2- and SERINC-positive puncta at the periphery of the cell. Scale bar: 20
μm. (B) Widefield imaging of a mixed population of HeLa cells selected for stable expression of
GFP-RUSC2 and HA-tagged SERINC3, labelled with anti-HA. Cells only positive for HA-tagged
SERINC3 are marked by an asterisk. The insets show accumulation of GFP-RUSC2 and HA-
positive puncta at the periphery of the cell. Scale bar: 20 μm.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 209
lation of the co-labelled protein, demonstrating the effect of RUSC2 overexpression is
specific for AP-4 cargo proteins. AP-4 itself does not accumulate on the peripheral AP-4
cargo-containing structures (Figure 4.33A). Likewise, the localisation of AP-1 (labelled via
anti-AP1G1) and its cargo protein CI-MPR (Cation-independent mannose-6-phosphate
receptor) were unaffected by RUSC2 overexpression, further demonstrating the speci-
ficity of RUSC2 for AP-4 cargo proteins (Figure 4.33B).
In order to gain a more detailed view of the RUSC2- and AP-4 cargo-containing struc-
tures at the periphery of RUSC2 overexpressing cells, Dr James Edgar, a post-doc in the
Robinson Lab, used correlative light and electron microscopy (CLEM). HeLa cells stably
expressing GFP-RUSC2 (clone 3) were mixed with parental HeLa cells and seeded on
special alpha-numeric gridded glass-bottom coverslips. The alpha-numeric grid allows
the same cells imaged by fluorescence microscopy to be identified for imaging on the
electron microscope. The GFP-RUSC2 fluorescent signal was imaged on a confocal mi-
croscope and a series of images at different focus depths were captured. Following this,
ultrathin 70 nm sections were prepared for imaging by electron microscopy. Alignment
of the fluorescent and electron micrographs revealed the peripheral GFP-RUSC2-positive
puncta to correspond to clusters of small uncoated vesicles and tubules at the very basal
surface of the cells (Figure 4.34). These were not found in regions of the cells which were
negative for GFP-RUSC2 fluorescent signal, nor were they observed in equivalent regions
in the parental HeLa cells. We hypothesise that these structures are the elusive AP-4
vesicle population and, of note, they appear similar to previously published descriptions
of the tubulovesicular ‘ATG9A reservoir’ (see Section 4.1.2; Mari et al., 2010; Orsi et al.,
2012).
4.5.3 The RUSC2-driven accumulation of ATG9A/SERINC-containing
vesicles at the cell periphery depends on AP-4
We hypothesised that the ATG9A, SERINC and RUSC2-containing structures at the pe-
riphery of the RUSC2-overexpressing cells are AP-4-derived vesicles. To test this hypoth-
esis siRNA was used to knock down AP-4 in HeLa cells stably expressing GFP-RUSC2.
When the cells were analysed by immunofluorescence microscopy with an antibody
against ATG9A this revealed that in the absence of AP-4, RUSC2 did not accumulate
at the periphery of the cell or colocalise with ATG9A which accumulated in the TGN
region (Figure 4.35A). The same was observed when AP4B1 and AP4E1 knockout cells
were transduced with retrovirus, selected for the stable expression of GFP-RUSC2, and
analysed in the same manner (Figure 4.35B and C). Importantly, both the peripheral
210 Functional characterisation of AP-4 cargo and machinery
Fig. 4.32 The effect of RUSC2 overexpression is specific for AP-4 cargo proteins. Widefield
imaging of HeLa cells stably expressing GFP-RUSC2 (clone 3), mixed on coverslips with parental
HeLa cells (marked with asterisks), labelled with anti-TGN46, anti-LAMP1, anti-EEA1 or anti-
RAB11. The insets show accumulation of GFP-RUSC2-positive puncta at the periphery of the
cells, without redistribution of the co-labelled protein, demonstrating that the effect of RUSC2
overexpression is specific for AP-4 cargo proteins. Scale bar: 20 μm.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 211
Fig. 4.33 AP-4 does not accumulate at the periphery of RUSC2-overexpressing cells. (A) Wide-
field imaging of HeLa cells stably expressing GFP-RUSC2 (clone 3), mixed on coverslips with
parental HeLa cells (marked with asterisks), labelled with anti-AP4E1. The insets show accu-
mulation of GFP-RUSC2-positive puncta at the periphery of the cells, without redistribution of
AP-4. Scale bar: 20 μm. (B) Widefield imaging of the same cells as in A, labelled with anti-AP1G1
or anti-CI-MPR. Neither AP-1 nor its cargo protein CI-MPR (Cation-independent mannose-6-
phosphate receptor) accumulated at the periphery of the RUSC2-overexpressing cells (see insets).
Scale bar: 20 μm.
212 Functional characterisation of AP-4 cargo and machinery
Fig. 4.34 CLEM reveals accumulations of small vesicles and tubules at the periphery of RUSC2-
overexpressing cells. Correlative light and electron microscopy (CLEM) of HeLa cells stably
expressing GFP-RUSC2. The peripheral GFP-RUSC2 puncta corresponded to accumulations
of small uncoated vesicular and tubular structures (electron micrograph (EM) areas A and B),
which were not found in peripheral regions negative for GFP-RUSC2 (area C) or in wild-type
HeLa cells. Scale bars: fluorescence, 10 μm; EM, 500 nm. Sample preparation and imaging by Dr
James Edgar.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 213
distribution of GFP-RUSC2 and its colocalisation with ATG9A were restored in the AP4B1
knockout by transient expression of AP4B1 (Figure 4.35C). This demonstrates that the
peripheral RUSC2/ATG9A/SERINC-positive structures are indeed AP-4-dependent com-
partments. The fact that AP-4 itself does not accumulate on these structures (Figure
4.33A) suggests it is released soon after vesicle budding, as is typical of most known
vesicle coats (Robinson, 2015).
Given the tight colocalisation between RUSC2 and ATG9A, and its dependence on AP-4,
we decided to test whether ATG9A physically interacts with GFP-RUSC2 and whether
this depends on AP-4. Lysates were prepared from wild-type and AP4B1 knockout HeLa
cells stably expressing GFP-RUSC2, and parental HeLa cells, using a relatively mild lysis
buffer (0.5 % NP-40 detergent; 100 mM NaCl). GFP-RUSC2 was immunoprecipitated
using GFP-trap beads and the immunoprecipitates were analysed by Western blotting
with antibodies against GFP, ATG9A and AP4B1 (Figure 4.36). ATG9A and AP4B1 both
co-immunoprecipitated with GFP-RUSC2 from the wild-type cells, but ATG9A was not
co-immunoprecipitated with GFP-RUSC2 from the AP4B1 knockout cells. This demon-
strates that RUSC2 and AP-4 are not only found in close proximity (as indicated by our
BioID data presented in Section 3.5.2), but actually physically interact. Likewise there is
a physical interaction between RUSC2 and ATG9A. These interactions could be indirect,
but the requirement of AP-4 for the interaction between RUSC2 and ATG9A suggests
that these proteins come together transiently during the formation of AP-4 vesicles.
4.5.4 RUSC2 is pulled down by AP-4 appendage domains
The AP-4 BioID data (Section 3.5.2) and the co-immunopreciptation of AP4B1 with GFP-
RUSC2 (Figure 4.36) demonstrated proximity and physical interaction between RUSC2
and AP-4. To investigate this interaction further we worked in collaboration with Dr
Lauren Parker Jackson and Dr Tara Archuleta at the University of Vanderbilt (Tennessee).
Most AP complex accessory proteins interact with one or both C-terminal ‘ear’ domains
of the large subunits. This includes the AP-4 accessory protein TEPSIN, for which our
data (presented in Section 4.2; Frazier et al., 2016) and work from the Bonifacino lab
(Mattera et al., 2015) has shown direct binding to both AP4B1 and AP4E1 appendage
domains. Therefore, we decided to test whether GFP-tagged RUSC2 could be pulled
down from cytosol using either of the AP-4 appendage domains.
Cytosol was prepared by mechanical homogenisation from two 15 cm dishes each of
HeLa cells stably expressing either GFP-RUSC2 (clone 3) or EGFP alone as a control.
The homogenates were cleared of cell debris and membrane material by 30 minutes
214 Functional characterisation of AP-4 cargo and machinery
Fig. 4.35 RUSC2-driven accumulation of ATG9A-vesicles at the cell periphery depends on AP-
4. (A) Widefield imaging of HeLa cells stably expressing GFP-RUSC2 (clone 3) treated with siRNA
to knock down AP-4, labelled with anti-ATG9A. GFP-RUSC2 puncta do not accumulate at the
cell periphery, nor do they colocalise with ATG9A. Scale bar: 20 μm. (B) Widefield imaging of
AP4E1 knockout (KO) HeLa cells stably expressing GFP-RUSC2, labelled with anti-ATG9A. As with
AP-4 knockdown, GFP-RUSC2 puncta neither accumulate at the cell periphery, nor colocalise
with ATG9A. Scale bar: 20 μm. (C) Widefield imaging of AP4B1 KO HeLa cells stably expressing
GFP-RUSC2, labelled with anti-ATG9A, with or without rescue by transient expression of AP4B1.
Scale bar: 20 μm.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 215
Fig. 4.36 ATG9A co-immunoprecipitates with GFP-RUSC2 only in the presence of AP-4. West-
ern blots of immunoprecipitates of GFP-RUSC2 from extracts of wild-type or AP4B1 knockout
HeLa cells stably expressing GFP-RUSC2, probed with antibodies against GFP, ATG9A and AP4B1.
Parental wild-type HeLa cells were used as a negative control. An antibody against clathrin heavy
chain was used as a loading control.
centrifugation at 78,400 g (RCF max), the same spin parameters as used to generate
the membrane fraction in the Dynamic Organellar Maps workflow. Drs Jackson and
Archuleta expressed AP4B1 (residues 612-739) and AP4E1 (residues 881-1135) appendage
domains as GST-fusion proteins and affinity purified them on glutathione Sepharose,
followed by elution from the Sepharose and further purification by gel filtration (Figure
4.37A). They then used the purified recombinant proteins as baits in pulldown assays
from the HeLa GFP-RUSC2 and HeLa EGFP cytosols, along with GST as a negative control
bait. Equal amounts (50 μg) of the fusion proteins immobilised on glutathione Sepharose
were incubated with cytosol for one hour, the Sepharose resin was washed and then the
GST-fusion proteins plus binding partners were eluted from the Sepharose and analysed
by Western blotting with an antibody against GFP (Figure 4.37B). GFP-RUSC2, but not
GFP alone, was pulled down with both AP-4 appendage domains, and not with the GST
control. Notably, more GFP-RUSC2 pulled down with the AP4E1 appendage domain
than the AP4B1 appendage domain. This is in keeping with our BioID data (Section
3.5.2) because RUSC2 was most strongly enriched in the streptavidin pull downs from
the AP4E1 BioID cell line. This suggests the interaction between RUSC2 and the AP4E1
appendage domain may be direct, but further work with purified RUSC2 is necessary to
test this.
216 Functional characterisation of AP-4 cargo and machinery
Fig. 4.37 RUSC2 pulls down with AP4E1 and AP4B1 appendage domains. (A) Coomassie-
stained SDS-PAGE gel with purified GST-fusion proteins used as bait for the pulldown assays:
GST (negative control), GST-ε (AP4E1 residues 883-1137) and GST-β4 (AP4B1 residues 612-739).
Gel shows input bait and pulldown samples eluted from glutathione Sepharose and analysed
by Western blotting in B. (B) Western blot of GST pulldowns using GST-ε, GST-β4 or GST, from
cytosol from HeLa cells stably expressing GFP-RUSC2, or GFP alone as a control, with an antibody
against GFP. A Western blot of input cytosol is also shown. GFP-RUSC2 pulled down specifically
with both AP-4 appendage domains and not with GST alone. Pulldowns were performed by Dr
Lauren Parker Jackson and Dr Tara Archuleta at the University of Vanderbilt (Tennessee).
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 217
4.5.5 The peripheral delivery of AP-4 vesicles is microtubule
dependent
The peripheral accumulation of AP-4 vesicles in the RUSC2-overexpressing cells sug-
gested targeting to the plus-ends of microtubules. RUSC1 has been implicated in
microtubule-based transport via the anterograde microtubule motor protein kinesin-1
(MacDonald et al., 2012). As discussed in Section 4.1.4, other RUN domain containing
proteins have also been implicated in kinesin-1-mediated microtubule-based transport,
for example SKIP1 (Dumont et al., 2010) and the C. elegans protein UNC-14 (Sakamoto
et al., 2004). Thus, we decided to investigate a possible role for microtubule-based
transport in the peripheral delivery of AP-4 vesicles.
Firstly confocal immunofluorescence microscopy was used to look at the position of
GFP-RUSC2 puncta in relation to the microtubule network in HeLa cells stably expressing
GFP-RUSC2 and labelled with an antibody against alpha-tubulin (Figure 4.38). The GFP-
RUSC2-positive puncta appeared to line up along the lengths of microtubules, consistent
with a role for the microtubule network in their distribution. Similar colocalisation
with microtubules has been observed for RUSC1 (MacDonald et al., 2012) and SKIP1
(Dumont et al., 2010). To test whether the altered distribution of AP-4 vesicles in RUSC2-
overexpressing cells might be caused indirectly by alterations to the microtubule network,
alpha-tubulin labelling was compared between wild-type and RUSC2 overexpressing
HeLa cells (Figure 4.39). No obvious differences were apparent in the organisation of the
microtubule network, suggesting this is not the cause of the peripheral accumulation of
AP-4 vesicles in RUSC2-overexpressing cells.
To test whether the peripheral delivery of AP-4 vesicles requires the microtubule network,
the GFP-RUSC2 overexpressing cells were treated with the microtubule-depolymerising
agent Nocodazole for two hours before fixation. Immunofluoresence microscopy with
an antibody against ATG9A revealed that the nocodazole treatment prevented the pe-
ripheral localisation of GFP-RUSC2 and ATG9A-positive puncta, which were instead
distributed evenly throughout the cell (Figure 4.40). However, GFP-RUSC2 and ATG9A
still colocalised in the nocodazole-treated cells, unlike in AP-4-deficient cells. These
data suggest that the distribution of AP-4 vesicles requires microtubule-based transport,
whereas their formation does not. Future studies are required to identify the microtubule
transport machinery involved.
218 Functional characterisation of AP-4 cargo and machinery
Fig. 4.38 RUSC2-positive puncta line up along microtubules. Confocal imaging of HeLa cells
stably expressing GFP-RUSC2 (clone 3), labelled with anti-alpha-tubulin. Scale bar: 10 μm.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 219
Fig. 4.39 The microtubule network appears normal in RUSC2 overexpressing cells. Confocal
imaging of wild-type HeLa and HeLa cells stably expressing GFP-RUSC2 (clone 3) or RUSC2-GFP
(clone 1), labelled with anti-alpha-tubulin. Scale bar: 10 μm.
220 Functional characterisation of AP-4 cargo and machinery
Fig. 4.40 The peripheral delivery of AP-4 vesicles in RUSC2-overexpressing cells is
microtubule-dependent. Widefield imaging of HeLa cells stably expressing GFP-RUSC2 (clone
3), cultured with or without nocodazole (10 μg/ml, 2 hours), labelled with anti-ATG9A. Insets
show how disruption of microtubules with nocodazole resulted in loss of the peripheral local-
isation of GFP-RUSC2 puncta, but their colocalisation with ATG9A remained. Scale bar: 20
μm.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 221
4.5.6 The effect of RUSC depletion on ATG9A localisation
Our discovery that RUSC2 functions in the peripheral delivery of ATG9A-containing
vesicles suggested that ATG9A localisation might be altered in the absence of RUSCs.
Based on our data we would predict that without RUSCs ATG9A would be packaged into
vesicles at the TGN, but that these vesicles would not be distributed to the periphery of
the cell and would instead accumulate in the perinuclear region.
In order to test this hypothesis we first needed to validate a robust method of depleting
RUSC1 and RUSC2 from cells. We decided to use siRNA-mediated knockdown but
as there are no antibodies that can detect endogenous RUSC proteins we had to rely
on quantification at the mRNA level by quantitative reverse transcription-polymerase
chain reaction (qRT-PCR) to validate the knockdowns. RUSC1 proved relatively simple
to knock down as the use of an siRNA Smartpool (Dharmacon) resulted in a robust
reduction in RUSC1 transcript of around 70 % (see Figure 4.41C). RUSC2 on the other
hand proved difficult to knock down efficiently. The Dharmacon Smartpool that was
initially tested only reduced RUSC2 transcript levels by roughly 50 % (data not shown).
To identify a more effective siRNA against RUSC2, eight different siRNA oligos were
trialled on HeLa cells overexpressing GFP-RUSC2 and the efficiency of the knockdowns
was gauged by Western blotting with an antibody against GFP (Figure 4.41A and B). Two
different transfection protocols were tested: (A) One-hit transfection and harvest 48
hours later; (B) Two-hit transfection with the second hit after 48 hours and harvest 96
hours after the first hit. Like the Dharmacon Smartpool, the four individual Dharmacon
oligos did not efficiently reduce the level of GFP-RUSC2. However, Qiagen oligos 4,
5 and 6, and a pool of all the Qiagen oligos combined, were much more effective at
depleting GFP-RUSC2, particularly using the two-hit 96 hours transfection protocol
(Figure 4.41B). The efficiency of Qiagen oligo 7 could not be estimated from this assay
because it targets the 3’ untranslated region (UTR) of RUSC2, which is not present in the
expression construct for GFP-RUSC2. qRT-PCR was then used to assess the efficiency
of knockdown of endogenous RUSC2 mRNA in wild-type HeLa cells by Qiagen oligos
4, 6 and 7 using the two-hit 96 hours transfection protocol (data not shown). The most
effective oligos were 4 and 7, which both reduced the level of RUSC2 mRNA by around
70 %. However, after 96 hours of RUSC2 knockdown there was a lot of cell death, which
didn’t occur at 48 hours using the same oligos. Therefore, we decided to use a two-hit 72
hours knockdown protocol (with the second hit 36 hours after the first) for the ATG9A
localisation experiment, which resulted in efficient knockdown of RUSC1 (Figure 4.41C)
and RUSC2 (Figure 4.41D) with less cell death than the 96 hours protocol.
222 Functional characterisation of AP-4 cargo and machinery
Fig. 4.41 Depletion of RUSCs by siRNA-mediated knockdown. (A) Western blots of whole cell
lysates from HeLa cells stably expressing GFP-RUSC2 (clone 3) transfected with different RUSC2-
targeting siRNAs, individually or as a pool, using a one-hit 48 hours transfection protocol, with
an antibody against GFP. Cells were mock treated with oligofectamine (no siRNA) or transfected
with siRNA to knock down AP-4 as controls. An antibody against actin was used as a loading
control. Note the Qiagen 7 oligo targets the 3’ UTR of RUSC2 mRNA, which is not present in the
GFP-RUSC2 expression construct, thus it would not be expected to knock down GFP-RUSC2. (B)
Western blots as in A, but cells were transfected using a two-hit 96 hours transfection protocol.
The most effective oligos from the GFP-RUSC2 test were Qiagen 4 and 6, and knockdowns
were more efficient with the 96 hours protocol than the 48 hours protocol. (C) Quantification
by qRT-PCR of RUSC1 mRNA levels in HeLa cells treated with siRNA to knock down RUSC1
(Smartpool), RUSC2 (Q4 or Q7 oligos), or both together, using a two-hit 72 hours transfection
protocol, relative to cells transfected with a non-targeting siRNA. The experiment was performed
in biological triplicate. Data are displayed as mean ± SEM (n = 3). The same cells were used for
the experiments shown in Figures 4.42, 4.43 and 4.46. (D) Quantification of RUSC2 mRNA levels
in the same cells, as described in C.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 223
Having established a robust method to deplete cells of RUSC1 and RUSC2 we tested
whether siRNA-mediated knockdown of either RUSC1, RUSC2, or both together, affects
the localisation of ATG9A. RUSC2 knockdown was performed with Q4 and Q7 siRNA
oligos to control for off-target effects. Transcript levels from both genes were robustly
reduced by around 70 % (Figure 4.41C and D). A non-targeting siRNA was used as a
control to ensure any observed effects were due to RUSC depletion and not due to the
delivery of siRNA to the cells. Immunofluorescence microscopy with antibodies against
ATG9A and TGN46 was used to compare ATG9A localisation in RUSC-depleted and
control treated cells (Figure 4.42). RUSC1 knockdown did not have an obvious effect
on ATG9A localisation. In the cells treated with the Q4 RUSC2 siRNA oligo there ap-
peared to be an increased concentration of ATG9A around the TGN and a corresponding
decrease in peripheral ATG9A labelling. This effect was more subtle than the effect of
AP-4 depletion on ATG9A localisation (see Section 4.3.1), and ATG9A did not appear to
accumulate within the TGN itself. Combined knockdown of RUSC1 and RUSC2 (with the
Q4 oligo) enhanced the effect on ATG9A localisation, suggesting that there may be some
functional redundancy between RUSC1 and RUSC2 in the peripheral delivery of ATG9A
vesicles. Confocal imaging with Airyscan enhanced resolution revealed a clustering of
ATG9A-positive puncta around the TGN in the cells treated with the Q4 RUSC2 siRNA
(Figure 4.43A), which was different from the TGN accumulation of ATG9A observed
in AP-4 knockdown cells (Figure 4.43B). However, the Q7 RUSC2 siRNA oligo did not
replicate this phenotype clearly (Figure 4.42). The cells treated with the Q7 oligo either
looked similar to control cells (see insets) or they were rounded up and had aberrant
TGN morphology (cells marked by asterisks). Thus it is unclear whether the clustering
of ATG9A puncta in the Q4 RUSC2 knockdown cells is specific for loss of RUSC2 or an
off-target effect.
In conclusion, our studies of RUSC2 have shown that it directly interacts with AP-4
and AP-4 cargo proteins and that overexpression of RUSC2 drives AP-4 vesicles to the
cell periphery in an AP-4 and microtubule-dependent manner. In addition our data
suggests that RUSC depletion may result in the perinuclear clustering of ATG9A vesicles,
but further work is required to confirm this phenotype. The CLEM analysis of GFP-
RUSC2-overexpressing cells has provided us with the first ultrastructural view of the
AP-4-dependent vesicle compartment and we hypothesise that this is the tubulovesicular
‘ATG9A reservoir’ that has previously been described in the literature (Mari et al., 2010;
Orsi et al., 2012).
224 Functional characterisation of AP-4 cargo and machinery
Fig. 4.42 ATG9A localisation in RUSC-depleted cells. Widefield imaging of HeLa cells treated
with siRNA to knock down RUSC1 (Smartpool), RUSC2 (Q4 or Q7 oligos), or both together, using
a two-hit 72 hours transfection protocol, labelled with anti-ATG9A and anti-TGN46. HeLa cells
transfected with a non-targeting (NT) siRNA were used as a control. DAPI staining of the nucleus
is also shown. Insets show clustering of ATG9A-positive puncta around the TGN in cells treated
with the Q4 RUSC2 siRNA but this effect was not obvious with the Q7 RUSC2 siRNA. Scale bar:
20 μm.
4.5 Role of RUSC2 in the peripheral delivery of AP-4 vesicles 225
Fig. 4.43 Airyscan enhanced resolution imaging of ATG9A in RUSC-depleted cells. (A) Confo-
cal microscopy with Airyscanning was used to image anti-ATG9A and anti-TGN46 in HeLa cells
treated with siRNA to knock down RUSC1 (Smartpool), RUSC2 (Q4 oligo), or both together, using
a two-hit 72 hours transfection protocol. HeLa cells transfected with a non-targeting (NT) siRNA
were used as a control. Scale bar: 5 μm. (B) Confocal microscopy with Airyscanning was used to
image anti-ATG9A and anti-TGN46 in HeLa cells treated with siRNA to knock down AP-4. Scale
bar: 5 μm.
226 Functional characterisation of AP-4 cargo and machinery
4.6 The role of AP-4 in the spatial control of autophagy
A previously published study of an Ap4b1 knockout mouse model reported an aberrant
accumulation of autophagosomes in swollen axon terminals of Purkinje and hippocam-
pal neurons (Matsuda et al., 2008). There was also elevated levels of the autophagic
marker protein LC3B (unlipidated and lipidated forms) and decreased levels of the
autophagy substrate p62, suggesting an increased level of autophagic activity in the
Ap4b1 knockout mice. The accumulated autophagosomes were immuno-positive for
AMPA receptors so the authors suggested that AP-4 was involved with the sorting of
AMPA receptors to the somatodendritic domain in neurons. However, they did not
investigate the link between AP-4 deficiency and the dysregulation of autophagy. Given
our identification of the core autophagy protein ATG9A as an AP-4 cargo protein in
this study, we hypothesised that the missorting of ATG9A might be the mechanistic
cause underlying the aberrant autophagy. In this section data is presented that supports
this hypothesis and we also demonstrate a close association between the AP-4-derived
ATG9A/SERINC/RUSC-positive compartments and autophagosomes, suggesting AP-4
plays an important role in the spatial control of autophagy.
4.6.1 LC3B levels are elevated in AP-4 knockout HeLa cells
The PE-conjugated form of LC3 (LC3-II) is the most commonly used marker for au-
tophagosomes because it is the only protein that reliably associates with completed
autophagosomes (Klionsky et al., 2016) and the amount of LC3-II closely correlates with
the number of autophagosomes (Kabeya et al., 2000). The most widely used method to
monitor autophagy is a Western blotting-based assay for detection of LC3 (typically the
LC3B isoform). LC3 is detected as two bands on a Western blot - LC3-I (the cytosolic
unlipidated form) runs at around 16 kDa and LC3-II (the membrane-associated lipidated
form) runs lower at around 14 kDa, despite its larger mass, probably due to its increased
hydrophobicity (Klionsky et al., 2016). During a short induction of autophagy, e.g. by a
short period of starvation, LC3-I levels will decrease and LC3-II levels will increase as
LC3-I is converted to LC3-II. However, LC3-II is also degraded by autophagy so longer
inductions of autophagy can result in decreased levels of both LC3-I and LC3-II. Addi-
tionally, LC3-II levels may increase not as a result of autophagy induction, but rather as
a result of reduced lysosomal degradation of autophagosomes. This means it is difficult
to interpret steady state levels of LC3-II. Therefore, when monitoring autophagy it is
important to have a method that measures the flux through the pathway rather than just
4.6 The role of AP-4 in the spatial control of autophagy 227
a steady state measurement of LC3-II levels (Klionsky et al., 2016). Comparison of LC3-II
levels between cells with and without pharmacological inhibition of autophagosome
degradation allows estimation of the amount of autophagic degradation activity occur-
ing in the cells. A commonly used drug for this purpose is the vacuolar-type H+-ATPase
(V-ATPase) inhibitor bafilomycin A1, which blocks autophagosome-lysosome fusion
(Yamamoto et al., 1998). A further point of confusion during the interpretation of LC3
blots can come from the fact that the sensitivity of detection of LC3-II by anti-LC3 is
often higher than that of LC3-I (Klionsky et al., 2016). Thus, it is not appropriate to
gauge autophagy flux by LC3-II/LC3-I or LC3-II/total LC3 ratios, but instead LC3-II
levels should be directly compared between samples relative to a loading control.
To investigate whether loss of AP-4 in HeLa cells caused similar effects on autophagy as
observed in neurons from Ap4b1 knockout mice, autophagy flux was monitored using
Western blotting with an antibody against LC3B (Figure 4.44). To induce autophagy cells
were subjected to amino acid and serum starvation by incubation in Earle’s balanced
salt solution (EBSS) for one or two hours. Cells in complete and starvation medium were
grown with or without the addition of bafilomycin A1 (100 nM for two hours) to block
autophagosome degradation. In untreated cells, there were increased levels of both
unlipidated LC3B-I and lipidated LC3B-II in the AP4E1 knockout relative to wild-type
HeLa cells (Figure 4.44A). In addition, the ratio of LC3B-II to LC3B-I was decreased.
Under starvation conditions (when LC3B-I is mostly converted into LC3B-II), there
was also increased LC3B-II in the AP4E1 knockout. The elevated level of LC3B in the
knockout persisted with bafilomycin treatment suggesting it was not due to a block in
degradation. The same trends were seen when comparing AP4B1 knockout cells with
the AP4B1 rescued cell line (Figure 4.44B). This demonstrates the aberrant autophagy
was due to a lack of AP-4 and not due to differences between clonal cell lines. Our SILAC-
based quantitative analysis of whole cell proteomes of AP-4 knockout cells presented in
Section 3.7.1 corroborated these findings as LC3B was found to be significantly enriched
in whole cell lysates from both knockout cell lines. In addition, another Atg8 homolog,
GABARAPL2 was also significantly enriched in the AP-4 knockout cells.
In yeast, elevated Atg8 levels have been shown to lead to an increase in autophagosome
size, whereas the number of autophagosomes is unaffected (Xie et al., 2008). A simi-
lar correlation between total LC3 level and the size of autophagosomes has also been
observed in mammalian cells (Nakagawa et al., 2004). Group A Streptococcus infection
induces higher levels of LC3 and larger autophagosomes than starvation conditions.
To test whether the elevated level of LC3B in our AP-4 knockout cells reflects an in-
crease in autophagosome size we used immunofluorescence microscopy to visualise
228 Functional characterisation of AP-4 cargo and machinery
Fig. 4.44 Loss of AP-4 in HeLa cells causes elevated LC3B levels. (A) Western blots of whole-cell
lysates from wild-type and AP4E1 knockout HeLa cells, cultured in full medium or starved for one
or two hours in EBSS, with or without the addition of bafilomycin A1 (Baf A1; 100 nM, 2 hours),
with antibodies against LC3B, AP4E1 and AP4B1. An antibody against clathrin heavy chain was
used as a loading control. Note that the phosphatidylethanolamine-conjugated from of LC3B
(LC3B-II) runs lower than the unlipidated form (LC3B-I), despite its larger mass, probably as a
consequence of increased hydrophobicity. (B) Western blots of whole-cell lysates from AP4B1
knockout and AP4B1 knockout HeLa cells rescued with stable expression of AP4B1, as described
in A.
4.6 The role of AP-4 in the spatial control of autophagy 229
LC3B-positive structures in wild-type, AP-4 knockout and AP4B1 rescued HeLa cells
under basal and starvation conditions (Figure 4.45). Under starvation conditions the
LC3B labelling was clearly brighter in the AP-4 knockout cells than in the wild-type or
rescued cells (Figure 4.45A). In the untreated cells the LC3B puncta also seemed slightly
brighter in the AP-4 knockout cells, but the difference was much more subtle (Figure
4.45B). By eye it was hard to judge whether the brighter LC3B labelling constituted a
greater number of LC3B-positive puncta or larger/brighter puncta. Therefore, we used
automated imaging to quantify the number and size of the LC3B puncta in the four
different cell lines (Figure 4.45C-F). The cells were grown in 96-well glass-bottomed
microplates and labelled with anti-LC3B and a whole cell mask. Cells were imaged using
a CellInsight CX7 High-Content Screening Platform (Thermo Fisher Scientific) and LC3B
puncta were quantified using the Spot Detector Bioapplication V4 (Cellomics, Thermo
Fisher Scientific) to measure spot total count and average area (in μm2). The experiment
was performed in biological triplicate and more than 500 cells were scored per cell line
in each replicate. Statistical analysis was by one-way ANOVA with Dunnett’s Multiple
Comparison Test for significant differences from the wild-type cells. This assay revealed
a significant increase in the size of the LC3B puncta in both AP-4 knockout cell lines
under starvation conditions, whereas they were rescued to normal size in the AP4B1
rescued cells (Figure 4.45C). In contrast there were no significant changes to the number
of LC3B puncta in the AP-4 knockout cells (Figure 4.45D). The results from the assay
when the cells were grown in full medium (basal condition) were not as consistent. LC3B
puncta were significantly larger in the AP4B1 knockout cells, and this was rescued by
AP4B1 expression, but there was not a significant change in the AP4E1 knockout cells.
Under basal conditions there appeared to be a decrease in the number of LC3B puncta
in the AP-4 knockout cells, but this was only significant for the AP4E1 knockout cell line.
These data suggest that the elevated level of LC3B in the AP-4-deficient cells reflects
larger autophagic structures, rather than a greater number of autophagosomes.
Collectively, these data suggest that lack of AP-4 in HeLa cells causes dysregulation of
autophagy, similar to the previously reported observation of aberrant autophagy in
neurons from AP-4-deficient mice (Matsuda et al., 2008). Based on our identification
of ATG9A as an AP-4 cargo protein we hypothesise that the mechanistic cause of the
dysregulated autophagy is missorting of ATG9A.
230 Functional characterisation of AP-4 cargo and machinery
Fig. 4.45 LC3B puncta are larger in AP-4 knockout cells. (Full caption on following page.)
4.6 The role of AP-4 in the spatial control of autophagy 231
Fig. 4.45 LC3B puncta are larger in AP-4 knockout cells. (A) Widefield imaging of wild-type,
AP4E1 knockout (KO), AP4B1 KO, and AP4B1 KO HeLa cells rescued with stable expression of
AP4B1 (Rescue), starved for two hours in EBSS, labelled with anti-LC3B. Scale bar: 20 μm. (B)
Widefield imaging of the cells described in A, in basal conditions, labelled with anti-LC3B. Scale
bar: 20 μm. (C) Quantification of the apparent size (in μm2) of LC3B puncta in the cells described
in A (starvation conditions) using an automated microscope. The experiment was performed in
biological triplicate (mean indicated, n = 3), and over 500 cells were scored per cell line in each
replicate. Data were subjected to one-way ANOVA with Dunnett’s Multiple Comparison Test
for significant differences to the wild-type: * p ≤ 0.05; ** p ≤ 0.01; *** p ≤ 0.001; ns p > 0.05. (D)
Quantification of the total number of LC3B-positive spots per cell, as in C. (E) Quantification of
the apparent size (in μm2) of LC3B puncta in the cells described in B (basal conditions) using an
automated microscope. The experiment was performed in biological triplicate (mean indicated,
n = 3), and over 500 cells were scored per cell line in each replicate. Data were subjected to
one-way ANOVA with Dunnett’s Multiple Comparison Test for significant differences to the
wild-type: * p ≤ 0.05; ** p ≤ 0.01; *** p ≤ 0.001; ns p > 0.05. (F) Quantification of the total number
of LC3B-positive spots per cell, as in E.
4.6.2 LC3B levels are elevated in RUSC-depleted HeLa cells
Given the aberrant autophagy caused by loss of AP-4 and our discovery of the role of
the RUSCs in the peripheral delivery of AP-4-derived ATG9A-containing vesicles, we
investigated whether depletion of the RUSCs might also affect autophagy. HeLa cells
depleted of RUSC1, RUSC2, or both together (the same cells as used in the ATG9A
localisation experiments in Section 4.5.6) were grown in complete or starvation medium
(to induce autophagy), with or without bafilomycin A1 (which blocks autophagosome
degradation). Transcript levels from both genes were robustly reduced by around 70 %
by the knockdowns (Figure 4.41D and E). Whole cell lysates were harvested and analysed
by Western blotting with an antibody against LC3B (Figure 4.46A). Knockdown of RUSC1
alone had no effect on the level of LC3B in HeLa cells grown in complete or starvation
medium. In contrast, knockdown of RUSC2 resulted in clear alterations to the level of
LC3B in both basal and starvation conditions, with very similar effects for both RUSC2
siRNA oligos (Q4 and Q7). In untreated cells there was an increased level of LC3B-I in
the RUSC2-depleted cells. Under starvation conditions there was elevated LC3B-I and
LC3B-II in the RUSC2 knockdown cells and there was a further increase with bafilomycin
treatment, suggesting there is not a block in autophagosome degradation. These effects
are comparable to those caused by knockdown of AP-4 (Figure 4.46B). Loss of both RUSC
proteins had an additive effect, resulting in further elevated LC3B-I and LC3B–II levels
and a more severe effect than that caused by AP-4 knockdown (Figure 4.46A and B).
This provides further evidence that RUSC1 and RUSC2 are at least partially functionally
232 Functional characterisation of AP-4 cargo and machinery
redundant (they also appeared to have an additive effect on ATG9A localisation - see
Section 4.5.6).
4.6.3 AP-4 vesicles closely associate with autophagosomes
Following on from our finding that the loss of RUSCs causes aberrant autophagy in HeLa
cells we tested whether overexpression of RUSC2 also affects LC3B levels. HeLa cells
stably expressing GFP-RUSC2 or RUSC2-GFP and parental HeLa cells were grown in full
or starvation medium for one or two hours, with or without the addition of bafilomycin
A1 (100 nM for 2 hours). Whole cell lysates were then analysed by the LC3B Western blot-
ting assay (Figure 4.47A and B). In contrast to the effect of RUSC2 depletion, LC3B levels
were unaffected by overexpression of RUSC2, in both basal and starvation conditions.
Immunofluorescence microscopy with an antibody against LC3B also showed similar
LC3B labelling in wild-type and RUSC2-overexpressing HeLa cells grown in full medium
(Figure 4.47C). However, when immunofluorescence microscopy was used to analyse
RUSC2-overexpressing cells that had been starved for two hours prior to fixation, we
observed a change in the localisation of RUSC2-positive puncta, away from the periphery
of the cell (Figure 4.48A). Despite this change in the localisation of RUSC2 puncta, they
were still positive for ATG9A, so they represent the same compartment that is found at
the very periphery of RUSC2-overexpressing cells under basal conditions. These RUSC2-
and ATG9A-positive puncta were often found in very close proximity to LC3B-positive
puncta, and the latter appeared larger in the RUSC2-overexpressing cells than in wild-
type HeLa cells. The presence of bafilomycin did not appear to affect the localisation
of RUSC2/ATG9A-positive puncta in starved cells, but LC3B accumulated under these
conditions and again larger LC3B puncta were observed in RUSC2-overexpressing cells
(Figure 4.48B). The resolution of the widefield microscope was not sufficient to determine
whether the closely positioned LC3B puncta were on the same or different structures to
RUSC2 and ATG9A. Thus, HeLa cells stably expressing GFP-RUSC2 or RUSC2-GFP were
starved for two hours in EBSS and then analysed by confocal microscopy with Airyscan-
ning with an antibody against LC3B (Figure 4.49). This revealed the RUSC2-positive and
LC3B-positive puncta to be separate structures, which were often closely associated. In
some cases it appeared that RUSC2 and LC3B were present on two separate structures
that were almost intertwined with each other (see zoomed images for examples). This
suggested to us a possible relationship between the RUSC2/ATG9A-positive puncta and
autophagosome positioning.
4.6 The role of AP-4 in the spatial control of autophagy 233
Fig. 4.46 Loss of RUSCs in HeLa cells causes elevated LC3B levels. (A) HeLa cells were treated
with siRNA to knock down RUSC1, RUSC2 (Q4 and Q7 oligos), both together, or were transfected
with a non-targeting siRNA (two-hit 72 hours knockdown protocol), and were cultured in full
medium or starved for two hours in EBSS, with or without the addition of bafilomycin A1 (100
nM, 2 hours), before lysis. Western blots of whole cell lysates with an antibody against LC3B. An
antibody against actin was used as a loading control. Knockdowns were confirmed by qRT-PCR
(Figure 4.41D and E). Q4 and Q7 RUSC2 siRNAs had very similar effects on LC3B levels. (B) HeLa
cells were treated with siRNA to knock down AP-4, RUSC1, RUSC2 (Q7 oligo), RUSC1 and RUSC2
together, or were transfected with a non-targeting siRNA (two-hit 72 hours knockdown protocol),
and were cultured in full medium or starved for two hours in EBSS before lysis. Western blots of
whole cell lysates with an antibody against LC3B. An antibody against actin was used as a loading
control. RUSC2 knockdown had a similar effect on LC3B levels to AP-4 knockdown, and double
knockdown of RUSC1 and RUSC2 had a more severe effect.
234 Functional characterisation of AP-4 cargo and machinery
Fig. 4.47 Overexpression of RUSC2 does not affect LC3B levels. (A) Western blots of whole
cell lysates from wild-type HeLa cells and HeLa cells stably expressing GFP-RUSC2 (clone 3),
cultured in full medium or starved for one or two hours in EBSS, with or without the addition of
bafilomycin A1 (100 nM, 2 hours), with antibodies against LC3B and GFP. An antibody against
actin was used as a loading control. (B) Western blots of whole cell lysates from wild-type HeLa
cells and HeLa cells stably expressing RUSC2-GFP (clone 1), as described in A. The wild-type
samples are the same as the samples shown in A. (C) Widefield imaging of HeLa cells stably
expressing GFP-RUSC2 (clone 3) or RUSC2-GFP (clone 1), mixed on coverslips with parental
HeLa cells (marked with asterisks), cultured in full medium, labelled with anti-LC3B. Scale bar:
20 μm.
4.6 The role of AP-4 in the spatial control of autophagy 235
Fig. 4.48 RUSC2- and ATG9A-positive vesicles are found in close proximity to LC3B puncta in
starved cells. (A) Widefield imaging of HeLa cells stably expressing GFP-RUSC2 (clone 3) or
RUSC2-GFP (clone 1), mixed on coverslips with parental HeLa cells (marked with asterisks),
starved for two hours in EBSS, labelled with anti-ATG9A and anti-LC3B. The insets show RUSC2-
and ATG9A-positive puncta in very close proximity to LC3B puncta. Scale bar: 20 μm. (B)
Widefield imaging of HeLa cells stably expressing GFP-RUSC2 or RUSC2-GFP, mixed on coverslips
with parental HeLa cells (marked with asterisks), starved for two hours in EBSS with 100 nM
bafilomycin A1, as in A.
236 Functional characterisation of AP-4 cargo and machinery
Fig. 4.49 RUSC2 and LC3B are found on separate, but closely associated, structures. Confocal
imaging with Airyscanning of HeLa cells stably expressing GFP-RUSC2 (clone 3) or RUSC2-GFP
(clone 1), starved for two hours in EBSS, labelled with anti-LC3B. The zoomed area is roughly 10
μm2. Scale bar: 5 μm.
4.6 The role of AP-4 in the spatial control of autophagy 237
To view the RUSC2-positive and LC3B-positive structures at ultrastructural resolution,
Dr James Edgar performed CLEM of the RUSC2-GFP overexpressing HeLa cells grown un-
der starvation conditions, in the presence of bafilomycin A1 (Figure 4.50). Bafilomycin
was included to accumulate LC3B-positive structures to give us the best chance of
identifying them on the EM. The CLEM was performed exactly as for the unstarved
RUSC2-overexpressing cells shown previously in Figure 4.34. We could not label for
LC3B because the conditions required for immunofluorescence labelling are not com-
patible with preserving ultrastructure for EM analysis. Therefore we had to rely on
identifying RUSC2-GFP by light microscopy and based on our immunofluorescence
microscopy analyses (Figure 4.48 and 4.49) we assumed LC3B-positive structures would
often be present close by. In the starved cells, the RUSC2-GFP signal observed by light
microscopy corresponded to clusters of small, uncoated vesicular structures (Figure
4.50), as we had previously observed at the periphery of RUSC2-overexpressing cells
grown in full medium. Importantly, we frequently observed double membrane-bound
autophagosomes juxtaposed to the RUSC2-positive vesicle clusters. The RUSC2-positive
vesicles were present in several serial sections, so appeared to cluster around the au-
tophagosomes in 3D. This suggests a very close relationship between the AP-4-derived
RUSC2/ATG9A/SERINC-containing vesicle compartment and autophagosomes and
supports our hypothesis that AP-4 functions in the generation of the tubulovesicular
‘ATG9A reservoir’, which functions in biogenesis of autophagosomes.
238 Functional characterisation of AP-4 cargo and machinery
Fig. 4.50 CLEM reveals that RUSC2-positive vesicles closely associate with autophagosomes.
Correlative light and electron microscopy (CLEM) of HeLa cells stably expressing RUSC2-GFP,
starved for two hours in EBSS with 100 nM bafilomycin A1. The RUSC2-GFP puncta correspond
to clusters of small uncoated vesicular and tubular structures (marked with asterisks), often in
close proximity to autophagosomes (marked with arrows). Serial EM sections are shown. Scale
bars: fluorescence and EM, 5 μm; Zoom, 500 nm. Sample preparation and imaging by Dr James
Edgar.
4.7 Summary 239
4.7 Summary
Based on the proteomics data presented in Chapter 3 we hypothesised that the highly
conserved transmembrane proteins ATG9A, SERINC1 and SERINC3 are AP-4 cargo
proteins and the cytosolic proteins RUSC1 and RUSC2 are AP-4 accessory proteins. The
follow-up studies of ATG9A, SERINC1, SERINC3 and RUSC2 presented in this chapter
support these hypotheses. From these studies we can conclude the following:
1. The autophagy transmembrane protein ATG9A accumulates at the TGN in AP-4-
deficient HeLa cells, SH-SY5Y neuroblastoma cells, and fibroblasts from patients
with AP-4 deficiency. This supports the hypothesis that ATG9A is a ubiquitous
cargo protein of the AP-4 trafficking pathway.
2. The mislocalisation of ATG9A to the TGN is a phenotype that correlates with disease
in AP-4 deficiency because it occurs in patients with homozygous loss-of-function
mutations in each of the four AP-4 subunits, whereas ATG9A localisation is normal
in a heterozygous individual.
3. SERINC1 and SERINC3 colocalise with ATG9A in peripheral puncta and this colo-
calisation depends on AP-4, suggesting the puncta are AP-4-derived structures.
4. RUSC2 is also found on SERINC- and ATG9A- positive structures and overexpres-
sion of RUSC2 results in the accumulation of these structures at the periphery of
the cell. CLEM revealed these structures to correspond to clusters of small un-
coated tubules and vesicles, which fit published descriptions of the ‘ATG9 reservoir’.
Both the colocalisation of RUSC2 with ATG9A and the peripheral distribution of
the puncta rely on AP-4, providing further evidence in support of the hypothesis
that these are AP-4-derived vesicles.
5. GFP-RUSC2 pulls down from cytosol with both AP-4 appendage domains (β4 and
ε). This supports the hypothesis that it is an AP-4 ear-binding accessory protein,
although currently we do not know whether this interaction is direct.
6. ATG9A can be co-immunoprecipitated with GFP-RUSC2 only in the presence of
AP-4, suggesting the three proteins come together transiently during the formation
of AP-4 vesicles.
7. The peripheral delivery of AP-4 vesicles involving RUSC2 is microtubule-dependent.
240 Functional characterisation of AP-4 cargo and machinery
8. Loss of AP-4 or the RUSCs results in an aberrant increase in LC3B which does not
seem to be due to a block in autophagosome degradation. This indicates that the
mislocalisation of ATG9A has a downstream effect on autophagy regulation.
9. AP-4 vesicles containing ATG9A and RUSC2 (and presumably the SERINCs) are
found in very close proximity to autophagosomes. This suggests they are the
tubulovesicular ATG9 reservoir implicated in the biogenesis of autophagosomes.
RUSC1 was not followed up specifically, but there does appear to be at least partial func-
tional redundancy between RUSC1 and RUSC2 with regard to their effect on autophagy.
In contrast to RUSC2 depletion, the depletion of RUSC1 alone did not affect LC3B levels.
However, RUSC1 knockdown had an additive effect on LC3B levels when combined with
RUSC2 knockdown. The effects of RUSC1/2 depletion on ATG9A localisation require fur-
ther investigation because two different siRNA oligos against RUSC2 produced different
results. The preliminary data suggests that in line with the effect on LC3B levels, RUSC1
knockdown alone may not affect ATG9A localisation, whereas RUSC2 depletion with
the Q4 oligo resulted in an apparent acccumulation of ATG9A-positive puncta around
the TGN. This effect appeared exaggerated when combined with RUSC1 depletion. The
generation of RUSC1 and RUSC2 knockout cell lines and/or the use of additional siRNA
oligos is necessary to confirm these results.
During the preparation of the work presented in this thesis for publication (Davies et al.,
2018), the Bonifacino lab published a paper in which they independently identified
ATG9A as an AP-4 cargo protein (Mattera et al., 2017). In a second paper published during
the revision of our manuscript they demonstrated that ATG9A is mistrafficked in primary
neurons from an Ap4e1 knockout mouse model (De Pace et al., 2018). Together these
studies support an important role for AP-4 in the sorting of ATG9A and the regulation of
autophagy, which is discussed further in Chapter 5.
Also presented in this Chapter was work performed in collaboration with Lauren Jackson
and Meredith Frazier at the University of Vanderbilt on the interaction between AP-4
and its accessory protein TEPSIN (published in Frazier et al., 2016). This resulted in
the identification of a binding site for TEPSIN on the β4 ear domain and a conserved
AP-4-binding motif at the C terminus of TEPSIN (LFxG[M/L]x[L/V]). Our in vivo experi-
ments indicated there must be one or more additional interaction sites between AP-4
and TEPSIN and this was confirmed by another study published by the Bonifacino lab
while our publication was in preparation (Mattera et al., 2015). Mattera and colleagues
described the same interaction between the C terminus of TEPSIN and the β4 ear, plus
an additional interaction between another conserved motif in the TEPSIN C terminus
4.7 Summary 241
(S[A/V]F[S/A]FLN) and the ε ear. This suggests the possibility that TEPSIN contributes to
the assembly of the AP-4 coat by cross-linking multiple AP-4 complexes. As this work was
a separate project from the rest of the work conducted for this thesis, it is not discussed
further in Chapter 5.

Chapter 5
Discussion
5.1 Overview
AP-4-deficiency in humans results in severe neurological problems, including spastic
paraplegia and intellectual disability. As AP complexes function in protein sorting, AP-4-
deficiency is a disease of missorting, and so the question of which proteins are missorted
in the absence of AP-4 is key to understanding the pathomechanisms of the disease.
The nature of AP-4 vesicles and their role in membrane trafficking has largely remained
elusive for the two decades since their discovery. Thus, the goal of this PhD project
was to apply orthogonal global proteomic tools to delineate the function of the AP-4
pathway. The results of these proteomic analyses are presented in Chapter 3 and sum-
marised in Section 3.8. Through the intersection of these analyses we identified three
transmembrane cargo proteins, ATG9A, SERINC1 and SERINC3, and two novel AP-4
accessory proteins, RUSC1 and RUSC2. The power of our approach derived from the
combined investigation of subcellular localisation information and quantitative mass
spectrometry, in the form of Dynamic Organellar Mapping and vesicle composition
profiling. Importantly, this approach was unbiased, hypothesis-free and analysed the
subcellular distribution of endogenous proteins. The latter is critical for assessing the
role of trafficking pathways, especially in the case of AP-4, which is of comparatively
low abundance (ca. 15,000-20,000 complexes per HeLa cell, compared with 500,000-
1,000,000 each of AP-1 or AP-2; Itzhak et al., 2016). In fact, overexpressed SERINC3 was
no longer trafficked in an AP-4-dependent manner (Section 4.4.1), suggesting that inves-
tigations based on overexpressed candidate cargo proteins are likely to lead to spurious
results. The AP-4-associated proteins identified in this study all have low expression
244 Discussion
levels similar to those of AP-4 itself, highlighting the sensitivity of our approach (Figure
3.27). They are also, like AP-4, expressed ubiquitously. In the follow-up cell biology
studies presented in Chapter 4 and summarised in Section 4.7, ATG9A localisation was
found to depend on AP-4 not only in HeLa cells, but also in neuroblastoma-derived
SH-SY5Y cells and in fibroblasts from AP-4-deficient patients (Section 4.3). This suggests
that trafficking of ATG9A from the TGN is a ubiquitous function of AP-4.
As mentioned, the Bonifacino lab have independently identified ATG9A as an AP-4
cargo protein, showing accumulation of ATG9A at the TGN in AP4E1 knockout HeLa
and HAP1 cells, and mouse embryonic fibroblasts (Mattera et al., 2017). In agreement
with our findings of aberrant autophagy in AP-4-depleted HeLa cells (Section 4.6.1),
they also observed increased levels of LC3B and enlarged autophagosomal structures.
However, the increased sensitivity of our proteomic approaches allowed us to reveal
additional interactions between AP-4 and the SERINCs and RUSCs, which were missed
by conventional affinity purification approaches. This highlights the limitations of
interaction-based approaches for identifying vesicle-associated proteins, due to the
transient nature of vesicle protein interactions. A further paper from the Bonifacino lab
published this year (De Pace et al., 2018) and a pre-print from the Kittler lab (Ivankovic
et al., 2017) both describe an Ap4e1 knockout mouse model (the same one - C57BL/6J
Ap4e1-/- [Ap4e1tm1b(KOMP)Wtsi]). The mice exhibit a number of neurological motor
deficits and brain abnormalities reminiscent of those observed in AP-4-deficient patients.
Both groups independently reported an accumulation of Atg9a at the TGN of primary
neurons from the Ap4e1-deficient mice, distal axonal swellings, and reduced retrograde
movement of autophagosomes, suggesting a defect in autophagosome maturation.
Collectively these studies and the work presented in this thesis strongly support the
role of AP-4 in the post-Golgi sorting of ATG9A and establish ATG9A missorting as a
candidate for the cause of neuronal pathology in AP-4 deficiency.
5.2 Model for AP-4-dependent trafficking 245
5.2 Model for AP-4-dependent trafficking
From the data presented in this thesis a model for AP-4-dependent trafficking emerges
(Figure 5.1). In this model, AP-4 packages the transmembrane proteins ATG9A, SERINC1
and SERINC3 into vesicles at the TGN, which associate via the RUSCs with machinery
for microtubule plus-end-directed transport to the cell periphery. This is supported by
the fact that ATG9A accumulates at the TGN in AP-4-deficient cells (Section 4.3) and that
RUSC2 overexpression results in an accumulation of AP-4-derived vesicles containing
ATG9A and SERINCs at the periphery of the cell (Section 4.5). As we observed very
close proximity of these vesicles to autophagosomes in starved cells (Section 4.6.3) we
propose that these AP-4-derived vesicles are the tubulovesicular ‘ATG9 reservoir’, which
is proposed to play an important role in autophagosome biogenesis (Duke et al., 2014;
Mari et al., 2010; Nair et al., 2011; Orsi et al., 2012). The AP-4-derived ATG9A-positive
vesicles and tubules we have observed at the periphery of RUSC2-overexpressing cells
by CLEM fit the aforementioned published descriptions of this compartment (compare
Figures 4.34, 4.50 and 4.2B).
Neuronal deficiency of Atg9a in mice leads to progressive axonal degeneration, ataxia and
convulsions (Yamaguchi et al., 2018). Thus, provided that AP-4 has equivalent functions
in neurons and HeLa cells, which our neuroblastoma data (Section 4.3.2) and the publi-
cations from the Bonifacino and Kittler labs (De Pace et al., 2018; Ivankovic et al., 2017)
support, a hypothesis for neuronal AP-4 pathology emerges. Neurons require efficient
long-range transport, especially towards the distal axon, rendering them susceptible to
disturbances in membrane trafficking (Blackstone, 2012). Furthermore, microtubules
are unipolar in axons, with distally localised plus-ends (Baas et al., 1988), and the distal
axon is an important site of autophagosome biogenesis (Maday & Holzbaur, 2014; Maday
et al., 2012). In Caenorhabditis elegans neurons, ATG-9-containing vesicles are trans-
ported towards the distal axonal microtubule plus-ends, and this is critical for axonal
autophagosome biogenesis (Stavoe et al., 2016). The transport of ATG-9 along the axon
is mediated by the synaptic vesicle kinesin UNC-104/KIF1A. C. elegans has lost the genes
for AP-4 so ATG-9-containing vesicles must be generated by a different mechanism.
Nonetheless, our model suggests that in mammalian neurons, AP-4-derived vesicles
carrying ATG9A perform a functionally equivalent role (Figure 5.1). In neurons lacking
AP-4, ATG9A will not be packaged correctly at the TGN, and will hence not efficiently
reach the distal axon. This may interfere with autophagosome formation at the axon
terminal and/or with other functions of ATG9A, disrupting neuronal homeostasis. This
hypothesis does not preclude the possibility that AP-4 also has neuron-specific cargo and
246 Discussion
Fig. 5.1 Proposed model of AP-4-dependent trafficking. (1) AP-4 and its accessory proteins,
TEPSIN, RUSC1 and RUSC2, are recruited to the trans-Golgi network (TGN) membrane where
they concentrate their transmembrane cargo proteins, ATG9A, SERINC1 and SERINC3, into a
vesicle bud; (2) A vesicle carrying ATG9A and SERINCs, coated by AP-4 and its accessory proteins,
buds from the TGN membrane; (3) AP-4 and TEPSIN fall off the vesicle membrane and are
available for further rounds of vesicle budding at the TGN, while RUSCs remain associated with
the vesicle; (4) The vesicle associates with microtubule transport machinery, via the RUSCs, for
plus-end-directed transport to the cell periphery; (5) The peripheral ATG9A-containing vesicles
act in the nucleation of autophagosomes. In neurons, microtubules are polarised with their
plus-ends at the distal axon, which is an important site of autophagosome biogenesis. Thus,
in AP-4-deficient neurons ATG9A may not be delivered efficiently to the distal axon, thereby
disrupting the spatial control of autophagy. Diagram was jointly created with Scottie Robinson.
5.2 Model for AP-4-dependent trafficking 247
it will be important to apply unbiased proteomic approaches such as Dynamic Organel-
lar Maps and vesicle profiling to neuronal cells to address this possibility. This could
also reveal whether the accumulation of glutamate receptors within autophagosomes in
axon terminals of AP-4-deficient neurons (De Pace et al., 2018; Matsuda et al., 2008) is
due to a direct role for AP-4 in their trafficking or a block in their autophagy-mediated
degradation (Shehata et al., 2012).
An important aspect to developing this model will be to identify the sequences within
ATG9A and the SERINCs required for AP-4-mediated sorting. Yeast two-hybrid analyses
conducted in the Bonifacino lab suggest that binding between μ4 and ATG9A is mediated
by a YXXΦE motif (YQRLE) in the cytosolic N-terminus of ATG9A (Mattera et al., 2017).
This is the same consensus sequence as the sorting motif in APP that they previously
described to bind to μ4 (Burgos et al., 2010). Mutation of the tyrosine residue within
the motif abolished binding of AP-4 to GFP-tagged ATG9A in vivo but the localisation
of the mutant ATG9A was not determined. The same sorting motif has previously been
implicated in binding to AP-1 and AP-2 and mutation resulted in reduced TGN localisa-
tion and retention at the plasma membrane (Imai et al., 2016; Zhou et al., 2017). Thus,
it is possible that the lack of co-immunoprecipitation of AP-4 with ATG9A lacking the
tyrosine motif is due to a reduction in their colocalisation at the TGN rather than a
direct effect on binding. Further work is required to investigate this and the sorting
determinants within the SERINCs. The generation of a mouse model with mutation of
the ATG9A AP-4 sorting motif would then allow assessment of the contribution of ATG9A
missorting to the phenotypes of AP-4 deficiency.
248 Discussion
5.3 AP-4 and the spatial control of autophagy
The identification of ATG9A as an AP-4 cargo provides a potential mechanistic basis for
the aberrant accumulation of autophagosomes previously observed in neuronal axons of
Ap4b1 knockout mice (Matsuda et al., 2008). There is a similar dysregulation of autophagy
in AP-4 knockout HeLa cells, which contain enlarged autophagosomes and increased
levels of LC3B (Section 4.6.1). Basal elevation of LC3B-I and LC3B-II has been observed
in ATG9A knockout HeLa cell lines (Lu et al., 2017; Nezich et al., 2015; Tsuboyama et al.,
2016), so our data are consistent with the notion of ATG9A mistrafficking leading to
impaired ATG9A function. However, the effect of ATG9A mislocalisation may well be
different from that caused by its complete loss. The role of ATG9A in autophagy is poorly
defined, but it is thought to contribute to autophagosome nucleation, without becoming
incorporated into the autophagosome membrane itself (Karanasios et al., 2016; Orsi
et al., 2012). Atg9 is necessary for mice to survive early neonatal starvation, a process
which depends on autophagy, and Atg9 knockout mouse embryonic fibroblasts have
greatly reduced levels of autophagosome formation (Saitoh et al., 2009). In yeast the
phagophore assembly site originates from Atg9-positive clusters of vesicles and tubules
(the ‘Atg9 reservoir’; Mari et al., 2010). Likewise, in mammalian cells Atg9 localises to a
tubulovesicular compartment, distinct from other organellar markers (Orsi et al., 2012).
Depletion of ULK1, a serine/threonine-protein kinase involved with the initiation of
autophagy, results in the clustering of ATG9 in the perinuclear region of HEK293 cells and
a concomitant clustering of early autophagosomal structures in the same region (Orsi
et al., 2012). This suggests the spatial positioning of the ATG9 reservoir compartment
may influence the positions at which autophagosomes nucleate. This is consistent
with a recent high resolution imaging study from the Ktistakis lab which showed early
autophagosome structures to emerge from regions where ATG9 vesicles align with the
ER (Karanasios et al., 2016). Similarly, work in C. elegans indicates that autophagy is
spatially regulated in neurons through the delivery of ATG-9 vesicles to the distal axon
(Stavoe et al., 2016). If this is the case, then our findings implicate AP-4 as an important
part of the machinery involved with the spatial control of autophagy, via the peripheral
delivery of ATG9A-containing vesicles.
In AP-4-deficient cells, ATG9A trafficking may be stalled at the TGN and the peripheral
‘ATG9A reservoir’ depleted. Further investigation is necessary to understand how this
may lead to the observed effects on autophagosomes in AP-4-deficient cells. ATG9A
mistrafficking has been linked previously to an accumulation of enlarged immature
autophagosomes in Niemann-Pick type A patient fibroblasts (Corcelle-Termeau et al.,
5.4 Autophagy and neurological disease 249
2016). The increased size of LC3B puncta in our AP-4 knockout HeLa cells may point
to a similar maturation defect, and future CLEM analysis will be important to investi-
gate this further. In support of this hypothesis, in neurons from Ap4e1 knockout mice
there is reduced retrograde movement of autophagosomes along the axon towards the
soma, which is suggestive of an altered maturation state (De Pace et al., 2018; Ivankovic
et al., 2017). The effect of ATG9A mislocalisation may be cell context-specific. Our data
showing a dramatic increase in ATG9A levels in AP-4-deficient patient fibroblasts but
not HeLa or SH-SY5Y cells suggests different cell types may compensate in different
ways (Section 4.3.4). Atg9 levels were also increased in the brain and in hippocampal
neurons from Ap4e1 knockout mice (De Pace et al., 2018; Ivankovic et al., 2017). In
contrast to the elevated LC3B levels observed in AP-4-deficient HeLa cells (Section 4.6.1
and Mattera et al., 2017), no alterations to LC3B levels were observed in cortical neurons
or different brain regions from the Ap4e1-deficient mice under basal or starvation condi-
tions, perhaps due to the compensatory increase in Atg9 (De Pace et al., 2018). Despite
this, mutant huntingtin, which is an autophagy substrate, had an increased tendency to
form aggregates in axons of Ap4e1-deficient neurons, suggesting that in the absence of
AP-4, autophagy-dependent aggregate clearance is impaired. Finally, it is worth noting
that ATG9A has functions independent of autophagy (Goodwin et al., 2017; Saitoh et al.,
2009; Yamaguchi et al., 2018), which may also be relevant to potential pathogenic effects
caused by its missorting.
5.4 Autophagy and neurological disease
Constitutive autophagy is known to play a critical role in neuronal homeostasis and
dysregulation of autophagy has been implicated in diverse neurodegenerative disor-
ders (reviewed in Menzies et al., 2017; Vijayan & Verstreken, 2017). Complete lack of
core autophagy genes result in embryonic or neonatal lethality, but neuron-specific
loss of the autophagy proteins Atg5 and Atg7 during embryonic development causes
neurodegeneration in mice (Hara et al., 2006; Komatsu et al., 2006). A hypomorphic
allele of ATG5 that leads to reduced autophagy causes ataxia, intellectual disability and
developmental delay in human patients (Kim et al., 2016). ‘Congenital disorders of
autophagy’ are an emerging group of single gene disorders that affect the autophagy
pathway, including EGF5-related Vici syndrome, SNX14-associated autosomal-recessive
cerebellar ataxia, and three forms of hereditary spastic paraplegia, SPG11, SPG15 and
SPG49 (Ebrahimi-Fakhari et al., 2016). This group of disorders has been proposed to
have the following unifying clinical features:
250 Discussion
• Onset of disease in childhood or adolescence
• Prominent nervous system involvement with multiple brain regions affected, often
including long white-matter tracts (e.g. corpus callosum and cortico-spinal tract)
and the cerebellum
• Neurodegenerative phenotypes including developmental regression and cognitive
decline
• Delayed development, intellectual disability, hypotonia, seizures and movement
disorders are common
• Storage disease phenotype
• Often multisystem disease including non-neuronal phenotypes such as myopathy
and ophthalmic manifestations
• Progressive disease course which is often fatal
AP-4 deficiency shares a number of these features including an early age of onset, thin
corpus callosum, developmental delay, intellectual disability, hypotonia progressing
to spasticity and often seizures. This supports the hypothesis that dysregulation of
autophagy caused by ATG9A mistrafficking is involved with the pathogenesis of AP-4
deficiency. It will now be important to learn more about the dynamics of autophagy
within AP-4-deficient cells in order to ascertain whether pharmacological intervention
targeting the autophagy pathway could be helpful for AP-4 deficiency.
5.5 The role of the RUSCs
Loss-of-function mutations in RUSC2 cause a neurological disorder with considerable
overlap with AP-4 deficiency phenotypes, including congenital hypotonia, motor delay,
severe intellectual disability, limited speech and secondary microcephaly (Alwadei et al.,
2016). This is in keeping with our identification of RUSCs as AP-4 accessory proteins
and the very similar effects of AP-4 and RUSC2 depletion on autophagy (Section 4.6.2).
The additive effect of RUSC1 and RUSC2 combined knockdown suggests partial func-
tional redundancy and links them both clearly to the AP-4 trafficking pathway. This
suggests that the neurological disorder caused by loss of RUSC2 may also be caused by
mislocalisation of ATG9A. Overexpression of RUSC2 clearly alters ATG9A localisation
(Section 4.5.2) but the effect of RUSC depletion on ATG9A localisation requires further
investigation. In this regard it will be interesting to look for ATG9A and SERINC sorting
5.6 SERINCs in ATG9A vesicles 251
defects in cells from RUSC2-deficient patients. The RUSCs are poorly characterised but
implicated in vesicular transport (Bayer et al., 2005; MacDonald et al., 2012). RUSC1 has
been proposed to act as a vesicle-transport adaptor by linking syntaxin-1 to kinesin-1
motors (MacDonald et al., 2012). Another RUN domain containing protein, SKIP1, has
similarly been shown to interact with kinesin-1 and overexpression of SKIP1 resulted in
an accumulation of LAMP1-positive compartments at the periphery of the cell (Dumont
et al., 2010). This phenotype is very reminiscent of the effect of RUSC2 overexpression
on the ATG9A compartment, and both are microtubule-dependent. This suggests a role
for RUSCs in linking AP-4-derived vesicles to microtubule transport machinery.
The additive effect of RUSC1 and RUSC2 knockdown on LC3B levels was actually more
severe than AP-4 knockdown (Figure 4.46), suggesting the RUSCs may have additional
roles in autophagy regulation beyond the AP-4-dependent trafficking of ATG9A. Interac-
tions with proteins of the ATG8 family are mediated by a degenerate short linear amino
acid motif known as an LC3-interacting region (LIR; reviewed in Birgisdottir et al., 2013).
The iLIR database (http://repeat.biol.ucy.ac.cy/iLIR/; Jacomin et al., 2016) identified
sequences fitting the consensus LiR motif in both RUSC1 and RUSC2 (residues 340-345
SDWLIV in RUSC1; residues 1407-1412 SDWLSL in RUSC2). Both are found within dis-
ordered segments with the potential to become ordered following protein interaction
(anchors), as is common for known LiRs. They also both had high position-specific scor-
ing matrix (PSSM) scores of 22, where the median score for a test group of known LiRs
was 18 (Kalvari et al., 2014). Thus, it will be interesting to test whether these predicted
LiR motifs mediate interaction between the RUSCs and LC3B.
5.6 SERINCs in ATG9A vesicles
To our knowledge, this is the first time endogenous SERINCs have been localised at
the subcellular level. The Dynamic Organellar Maps data suggests the SERINCs have a
mixed steady state distribution in more than one organelle (Section 3.3.2). In addition to
their presence in ATG9A-positive vesicles we also detected partial colocalisation of en-
dogenously tagged SERINC1 and SERINC3 with LAMP1 (Figure 4.27) and overexpressed
SERINC3 was also detected at the plasma membrane (Figure 3.20). The presence of SER-
INC1 and SERINC3 in the ATG9A tubulovesicular compartment (Section 4.4.4) warrants
further investigation as it suggests a functional relationship. Very little is known about
the function of SERINCs, despite their recent identification as HIV restriction factors
(Rosa et al., 2015; Usami et al., 2015) and their high degree of conservation in all branches
of eukaryotes. They were originally proposed to mediate the incorporation of serine
252 Discussion
into membrane lipids (Inuzuka et al., 2005), but recent functional studies have shown
no effect on membrane composition (Chu et al., 2017; Trautz et al., 2017). However,
functional redundancy between the five members of the SERINC protein family may be
an issue and requires consideration in future efforts to assign protein function to the
SERINCs. Also, HeLa cells only express SERINC1 and SERINC3 (Itzhak et al., 2016), so it is
possible that AP-4 plays a similar role in the trafficking of other SERINC family members
in other cell types. The possibility that SERINCs could have a role in the regulation of
autophagy is intriguing. What is their role on the ATG9 reservoir compartmemt? Lipids
play a central role in the membrane sculpting process of forming an autophagosome,
although the lipid composition of the autophagosome is poorly defined (Martens et al.,
2016). If SERINCs are involved with the biosynthesis of phosphatidylserine (PS) and
sphingolipids, could this influence the properties of the autophagosome membrane?
Although phosphatidylethanolamine (PE) is the major target for ATG8 protein conju-
gation in vivo, PS can also serve as an acceptor for ATG8 in vitro (Sou et al., 2006) and
its negative charge may contribute to membrane binding by ATG12–ATG5-ATG16L1,
the E3 ligase-like complex that catalyses the conjugation of ATG8 family members to
the membrane (Romanov et al., 2012). Finally, it is interesting to note that autophagy
has been identified as a major host immune defence mechanism against HIV infection,
which is counteracted by the HIV accessory protein Nef (reviewed in Espert et al., 2015).
To end this thesis on a highly speculative note – could the role of the SERINCs as HIV
restriction factors be linked in part to a role in autophagy?
5.7 Conclusion
In conclusion, this study has greatly expanded our understanding of the AP-4 pathway
and provides evidence that AP-4-derived vesicles play an important role in the spatial
control of autophagy. ATG9A, SERINC1 and SERINC3 are the first ubiquitously expressed
endogenous proteins that have been demonstrated to be missorted in AP-4-deficient
cells. This provides opportunity for development of AP-4 sorting assays which can be
used to screen for additional players in AP-4-mediated trafficking or to test the efficacy of
therapeutic interventions for AP-4 deficiency. Importantly there is now a well supported
hypothesis that the pathomechanism of AP-4 deficiency involves the missorting of
ATG9A vesicles, and this marks a significant step towards the development of possible
treatments.
References
Abou Jamra R. et al. (2011), ‘Adaptor protein complex 4 deficiency causes severe
autosomal-recessive intellectual disability, progressive spastic paraplegia, shy charac-
ter, and short stature.’, The American Journal of Human Genetics 88(6), 788–795.
Adolf F., Herrmann A., Hellwig A., Beck R., Brügger B., Wieland F. T. (2013), ‘Scission of
COPI and COPII vesicles is independent of GTP hydrolysis’, Traffic 14(8), 922–932.
Aebersold R., Mann M. (2016), ‘Mass-spectrometric exploration of proteome structure
and function’, Nature 537(7620), 347–355.
Aguilar R. C., Boehm M., Gorshkova I., Crouch R. J., Tomita K., Saito T., Ohno H., Bonifa-
cino J. S. (2001), ‘Signal-binding specificity of the mu4 subunit of the adaptor protein
complex AP-4.’, The Journal of biological chemistry 276(16), 13145–13152.
Aguilera M. O., Berón W., Colombo M. I. (2012), ‘The actin cytoskeleton participates in
the early events of autophagosome formation upon starvation induced autophagy’,
Autophagy 8(11), 1590–1603.
Ahle S., Mann A., Eichelsbacher U., Ungewickell E. (1988), ‘Structural relationships
between clathrin assembly proteins from the Golgi and the plasma membrane.’, The
EMBO journal 7(4), 919–929.
Akhmanova A., Hammer J. A. (2010), ‘Linking molecular motors to membrane cargo’,
Current Opinion in Cell Biology 22(4), 479–487.
Alwadei A. H., Benini R., Mahmoud A., Alasmari A., Kamsteeg E.-J., Alfadhel M. (2016),
‘Loss-of-function mutation in <i>RUSC2</i> causes intellectual disability and sec-
ondary microcephaly’, Developmental Medicine & Child Neurology 58(12), 1317–1322.
Andersen J. S., Mann M. (2006), ‘Organellar proteomics: Turning inventories into in-
sights’, EMBO Reports 7(9), 874–879.
Andersen J. S., Wilkinson C. J., Mayor T., Mortensen P., Nigg E. A., Mann M. (2003),
‘Proteomic characterization of the human centrosome by protein correlation profiling’,
Nature 426, 570–574.
Anderson R. G., Goldstein J. L., Brown M. S. (1977), ‘A mutation that impairs the ability of
lipoprotein receptors to localise in coated pits on the cell surface of human fibroblasts’,
Nature 270(5639), 695–699.
254 References
Antonny B. et al. (2016), ‘Membrane fission by dynamin: what we know and what we
need to know’, The EMBO Journal 35(21), 2270–2284.
Antrobus R., Borner G. H. (2011), ‘Improved elution conditions for native co-
immunoprecipitation’, PLoS ONE 6(3), e18218.
Ao X., Zou L., Wu Y. (2014), ‘Regulation of autophagy by the Rab GTPase network’, Cell
Death and Differentiation 21(3), 348–358.
Archuleta T. L., Frazier M. N., Monken A. E., Kendall A. K., Harp J., McCoy A. J., Creanza N.,
Jackson L. P. (2017), ‘Structure and evolution of ENTH and VHS/ENTH-like domains
in tepsin’, Traffic 18(9), 590–603.
Arimura N. et al. (2009), ‘Anterograde Transport of TrkB in Axons Is Mediated by Direct
Interaction with Slp1 and Rab27’, Developmental Cell 16(5), 675–686.
Auer-Grumbach M. et al. (2016), ‘Rare Variants in MME, Encoding Metalloprotease
Neprilysin, Are Linked to Late-Onset Autosomal-Dominant Axonal Polyneuropathies’,
American Journal of Human Genetics 99(3), 607–623.
Axe E. L., Walker S. A., Manifava M., Chandra P., Roderick H. L., Habermann A., Griffiths
G., Ktistakis N. T. (2008), ‘Autophagosome formation from membrane compartments
enriched in phosphatidylinositol 3-phosphate and dynamically connected to the
endoplasmic reticulum’, Journal of Cell Biology 182(4), 685–701.
Baas P. W., Deitch J. S., Black M. M., Banker G. A. (1988), ‘Polarity orientation of mi-
crotubules in hippocampal neurons : Uniformity in the axon and nonuniformity in
the dendrite’, Proceedings of the National Academy of Sciences of the United States of
America 85(November), 8335–8339.
Barlow A. L., MacLeod A., Noppen S., Sanderson J., Guérin C. J. (2010), ‘Colocalization
Analysis in Fluorescence Micrographs : Verification of a More Accurate Calculation of
Pearson ’ s Correlation Coefficient’, Microscopy and Microanalysis 16, 710–724.
Barlowe C. et al. (1994), ‘COPII: A membrane coat formed by Sec proteins that drive
vesicle budding from the endoplasmic reticulum’, Cell 77(6), 895–907.
Bayer M., Fischer J., Kremerskothen J., Ossendorf E., Matanis T., Konczal M., Weide T.,
Barnekow A. (2005), ‘Identification and characterization of Iporin as a novel interaction
partner for rab1.’, BMC cell biology 6(1), 15.
Beck R. et al. (2011), ‘Coatomer and dimeric ADP ribosylation factor 1 promote distinct
steps in membrane scission’, Journal of Cell Biology 194(5), 765–777.
Benjamini Y., Hochberg Y. (1995), ‘Controlling the False Discovery Rate : A Practical and
Powerful Approach to Multiple Testing’, Journal of the Royal Statistical Society. Series B
(Methodological) 57(1), 289–300.
Bento C. F., Renna M., Ghislat G., Puri C., Ashkenazi A., Vicinanza M., Menzies F. M.,
Rubinsztein D. C. (2016), ‘Mammalian Autophagy: How Does It Work?’, Annual Review
of Biochemistry 85(1), 685–713.
References 255
Biazik J., Ylä-Anttila P., Vihinen H., Jokitalo E., Eskelinen E. L. (2015), ‘Ultrastructural
relationship of the phagophore with surrounding organelles’, Autophagy 11(3), 439–
451.
Biedler J. L., Roffler-Tarlov S., Schachner M., Freedman L. S. (1978), ‘Multiple Neuro-
transmitter Synthesis by Human Neuroblastoma Cell Lines and Clones’, CANCER
RESEARCH 38, 3751–3757.
Birgisdottir A. B., Lamark T., Johansen T. (2013), ‘The LIR motif - crucial for selective
autophagy’, Journal of Cell Science 126(15), 3237–3247.
Blackstone C. (2012), ‘Cellular pathways of hereditary spastic paraplegia.’, Annual review
of neuroscience 35, 25–47.
Blondeau F. et al. (2004), ‘Tandem MS analysis of brain clathrin-coated vesicles reveals
their critical involvement in synaptic vesicle recycling’, Proceedings of the National
Academy of Sciences of the United States of America 101(11), 3833–3838.
Blumkin L., Lerman-Sagie T., Lev D., Yosovich K., Leshinsky-Silver E. (2011), ‘A new
locus (SPG47) maps to 1p13.2-1p12 in an Arabic family with complicated autoso-
mal recessive hereditary spastic paraplegia and thin corpus callosum’, Journal of the
Neurological Sciences 305(1-2), 67–70.
Boehm M., Aguilar R. C., Bonifacino J. S. (2001), ‘Functional and physical interactions of
the adaptor protein complex AP-4 with ADP-ribosylation factors (ARFs).’, The EMBO
journal 20(22), 6265–6276.
Boisvert F.-M., Lam Y. W., Lamont D., Lamond A. I. (2010), ‘A Quantitative Proteomics
Analysis of Subcellular Proteome Localization and Changes Induced by DNA Damage’,
Molecular & Cellular Proteomics 9(3), 457–470.
Boisvert F.-M., Lamond A. I. (2010), ‘p53-Dependent subcellular proteome localization
following DNA damage’, Proteomics 10, 4087–4097.
Bonifacino J. S. (2014), ‘Vesicular transport earns a Nobel’, Trends in Cell Biology 24(1), 3–
5.
Borner G. H. H., Antrobus R., Hirst J., Bhumbra G. S., Kozik P., Jackson L. P., Sahlender D. a.,
Robinson M. S. (2012), ‘Multivariate proteomic profiling identifies novel accessory
proteins of coated vesicles.’, The Journal of cell biology 197(1), 141–160.
Borner G. H. H., Harbour M., Hester S., Lilley K. S., Robinson M. S. (2006), ‘Comparative
proteomics of clathrin-coated vesicles’, Journal of Cell Biology 175(4), 571–578.
Borner G. H. H., Hein M. Y., Hirst J., Edgar J. R., Mann M., Robinson M. S. (2014), ‘Frac-
tionation profiling: a fast and versatile approach for mapping vesicle proteomes and
protein-protein interactions’, Molecular Biology of the Cell 25, 3178–3194.
Burgos P. V., Mardones G. A., Rojas A. L., DaSilva L. L. P., Prabhu Y., Hurley J. H., Bonifacino
J. S. (2010), ‘Sorting of the Alzheimer’s disease amyloid precursor protein mediated by
the AP-4 complex.’, Developmental cell 18, 425–436.
256 References
Buss F., Kendrick-Jones J. (2008), ‘How are the cellular functions of myosin VI regulated
within the cell?’, Biochemical and Biophysical Research Communications 369(1), 165–
175.
Callebaut I., Jean, De Gunzburg, Goud B., Mornon J. P. (2001), ‘RUN domains: A new fam-
ily of domains involved in Ras-like GTPase signaling’, Trends in Biochemical Sciences
26(2), 79–83.
Chahrour O., Cobice D., Malone J. (2015), ‘Stable isotope labelling methods in mass
spectrometry-based quantitative proteomics’, Journal of Pharmaceutical and Biomed-
ical Analysis 113, 2–20.
Chen H., De Camilli P. (2005), ‘The association of epsin with ubiquitinated cargo along the
endocytic pathway is negatively regulated by its interaction with clathrin.’, Proceedings
of the National Academy of Sciences of the United States of America 102(8), 2766–2771.
Chen W.-J., Goldstein J. L., Brown M. S. (1990), ‘NPXY, a sequence often found in cyto-
plasmic tails, is required for coated-pit mediated internalization of the low density
lipoprotein receptor’, Journal of Biological Chemistry 265(6), 3116–3123.
Chu E. P. et al. (2017), ‘Disruption of Serinc1, which facilitates serine-derived lipid syn-
thesis, fails to alter macrophage function, lymphocyte proliferation or autoimmune
disease susceptibility’, Molecular Immunology 82, 19–33.
Cocucci E., Aguet F., Boulant S., Kirchhausen T. (2012), ‘The first five seconds in the life
of a clathrin-coated pit’, Cell 150(3), 495–507.
Collins B. M., McCoy A. J., Kent H. M., Evans P. R., Owen D. J. (2002), ‘Molecular architec-
ture and functional model of the endocytic AP2 complex’, Cell 109(4), 523–535.
Cong L. et al. (2013), ‘Multiplex Genome Engineering Using CRISPR/Cas Systems’, Science
339, 819–823.
Corcelle-Termeau E. et al. (2016), ‘Excess sphingomyelin disturbs ATG9A trafficking and
autophagosome closure’, Autophagy 12(5), 833–849.
Costes S. V., Daelemans D., Cho E. H., Dobbin Z., Pavlakis G., Lockett S. (2004), ‘Auto-
matic and quantitative measurement of protein-protein colocalization in live cells’,
Biophysical Journal 86(6), 3993–4003.
Cowles C. R., Odorizzi G., Payne G. S., Emr S. D. (1997), ‘The AP-3 Adaptor Complex Is
Essential for Cargo-Selective Transport to the Yeast Vacuole’, Cell 91, 109–118.
Cox J., Hein M. Y., Luber C. A., Paron I., Nagaraj N., Mann M. (2014), ‘Accurate Proteome-
wide Label-free Quantification by Delayed Normalization and Maximal Peptide Ratio
Extraction , Termed MaxLFQ’, Molecular & Cellular Proteomics 13(9), 2513–2526.
Cox J., Mann M. (2008), ‘MaxQuant enables high peptide identification rates, individual-
ized p.p.b.-range mass accuracies and proteome-wide protein quantification.’, Nature
biotechnology 26(12), 1367–1372.
References 257
Craige B., Salazar G., Faundez V. (2008), ‘Phosphatidylinositol-4-Kinase Type II Alpha
Contains an AP-3–sorting Motif and a Kinase Domain That Are Both Required for
Endosome Traffic’, Molecular biology of the cell 19, 1415–1426.
Crowther R. A., Finch J. T., Pearse B. M. F. (1976), ‘On the structure of coated vesicles’,
Journal of Molecular Biology 103(4), 785–798.
Cullen P. J., Steinberg F. (2018), ‘To degrade or not to degrade: mechanisms and signifi-
cance of endocytic recycling’, Nature Reviews Molecular Cell Biology .
Damke H., Baba T., Warnock D. E., Schmid S. L. (1994), ‘Induction of mutant dy-
namin specifically blocks endocytic coated vesicle formation’, Journal of Cell Biology
127(4), 915–934.
Dannhauser P. N., Ungewickell E. J. (2012), ‘Reconstitution of clathrin-coated bud and
vesicle formation with minimal components’, Nature Cell Biology 14(6), 634–639.
Davies A. K., Itzhak D. N., Edgar J. R., Archuleta T. L., Jackson L. P., Robinson M. S.,
Borner G. H. H. (2018), ‘AP-4 vesicles contribute to spatial control of autophagy via
RUSC-dependent peripheral delivery of ATG9A’, Nature Communications 9, 3958.
Davis C. G., Lehrman M. A., Russell D. W., Anderson R. G., Brown M. S., Goldstein J. L.
(1986), ‘The J. D. mutation in familial hypercholesterolemia: Amino acid substitution
in cytoplasmic domain impedes internalization of LDL receptors’, Cell 45(1), 15–24.
De Pace R. et al. (2018), ‘Altered distribution of ATG9A and accumulation of axonal
aggregates in neurons from a mouse model of AP-4 deficiency syndrome’, PLoS genetics
14(4), e1007363.
Dell’Angelica E. C., Klumperman J., Stoorvogel W., Bonifacino J. S. (1998), ‘Association
of the AP-3 adaptor complex with clathrin’, Science 280(5362), 431–434.
Dell’Angelica E. C., Mullins C., Bonifacino J. S. (1999a), ‘AP-4, a Novel Protein Complex
Related to Clathrin Adaptors’, Journal of Biological Chemistry 274(11), 7278–7285.
Dell’Angelica E. C., Ohno H., Ooi C. E., Rabinovich E., Roche K. W., Bonifacino J. S.
(1997a), ‘AP-3: an adaptor-like protein complex with ubiquitous expression.’, The
EMBO journal 16(5), 917–928.
Dell’Angelica E. C., Ooi C. E., Bonifacino J. S. (1997b), ‘β3A-adaptin, a subunit of the
adaptor-like complex AP-3’, Journal of Biological Chemistry 272(24), 15078–15084.
Dell’Angelica E. C., Puertollano R., Mullins C., Aguilar R. C., Vargas J. D., Hartnell L. M.,
Bonifacino J. S. (2000), ‘GGAs: A family of ADP ribosylation factor-binding proteins
related to adaptors and associated with the Golgi complex’, Journal of Cell Biology
149(1), 81–93.
Dell’Angelica E. C., Shotelersuk V., Aguilar R. C., Gahl W. A., Bonifacino J. S. (1999b),
‘Altered trafficking of lysosomal proteins in Hermansky-Pudlak syndrome due to
mutations in the β3A subunit of the AP-3 adaptor’, Molecular Cell 3(1), 11–21.
258 References
Depondt C., Donatello S., Rai M., Wang F. C., Manto M., Simonis N., Pandolfo M. (2016),
‘MME mutation in dominant spinocerebellar ataxia with neuropathy (SCA43)’, Neu-
rology Genetics 2(5), e94.
Devos D., Dokudovskaya S., Alber F., Williams R., Chait B. T., Sali A., Rout M. P. (2004),
‘Components of coated vesicles and nuclear pore complexes share a common molec-
ular architecture’, PLoS Biology 2(12), e380.
Dodonova S. O., Diestelkoetter-Bachert P., Von Appen A., Hagen W. J., Beck R., Beck M.,
Wieland F., Briggs J. A. (2015), ‘A structure of the COPI coat and the role of coat proteins
in membrane vesicle assembly’, Science 349(6244), 195–198.
Donaldson J. G., Jackson C. L. (2011), ‘ARF family G proteins and their regulators: Roles
in membrane transport, development and disease’, Nature Reviews Molecular Cell
Biology 12(6), 362–375.
Dooley H. C., Razi M., Polson H. E. J., Girardin S. E., Wilson M. I., Tooze S. A. (2014),
‘WIPI2 Links LC3 Conjugation with PI3P, Autophagosome Formation, and Pathogen
Clearance by Recruiting Atg12-5-16L1’, Molecular Cell 55(2), 238–252.
Doyle J. M., Gao J., Wang J., Yang M., Potts P. R. (2010), ‘MAGE-RING protein complexes
comprise a family of E3 ubiquitin ligases’, Molecular Cell 39(6), 963–974.
Drissi R., Dubois M.-l., Boisvert F.-M. (2013), ‘Proteomics methods for subcellular pro-
teome analysis’, The FEBS journal 280, 5626–5634.
Duke E. M. H. et al. (2014), ‘Imaging endosomes and autophagosomes in whole mam-
malian cells using correlative cryo-fluorescence and cryo-soft X-ray microscopy (cryo-
CLXM)’, Ultramicroscopy 143, 77–87.
Dumont A., Boucrot E., Drevensek S., Daire V., Gorvel J. P., Poüs C., Holden D. W., Méresse
S. (2010), ‘SKIP, the host target of the salmonella virulence factor SifA, promotes
kinesin-1-dependent vacuolar membrane exchanges’, Traffic 11(7), 899–911.
Dunkley T. P. J. et al. (2006), ‘Mapping the Arabidopsis organelle proteome’, Proceedings
of the National Academy of Sciences 103(17), 6518–6523.
Dunkley T. P. J., Watson R., Griffin J. L., Dupree P., Lilley K. S. (2004), ‘Localization of
Organelle Proteins by Isotope Tagging ( LOPIT )’, Molecular & Cellular Proteomics
3(11), 1128–1134.
Ebner P., Poetsch I., Deszcz L., Hoffmann T., Zuber J., Ikeda F. (2018), ‘The IAP family
member BRUCE regulates autophagosome-lysosome fusion’, Nature communications
9, 599.
Ebrahimi-Fakhari D. et al. (2018), ‘Clinical and genetic characterization of AP4B1-
associated SPG47’, American Journal of Medical Genetics. Part A. 176A, 311–318.
Ebrahimi-Fakhari D., Saffari A., Wahlster L., Lu J., Byrne S., Hoffmann G. F., Jungbluth
H., Sahin M. (2016), ‘Congenital disorders of autophagy: An emerging novel class of
inborn errors of neuro-metabolism’, Brain 139(2), 317–337.
References 259
Edeling M. A., Smith C., Owen D. (2006), ‘Life of a clathrin coat: Insights from clathrin
and AP structures’, Nature Reviews Molecular Cell Biology 7(1), 32–44.
Encinas M., Iglesias M., Liu Y., Wang H., Muhaisen A., Ceña V., Gallego C., Comella
J. X. (2000), ‘Sequential Treatment of SH-SY5Y Cells with Retinoic Acid and Brain-
Derived Neurotrophic Factor Gives Rise to Fully Differentiated, Neurotrophic Factor-
Dependent, Human Neuron-Like Cells’, Journal of Neurochemistry 75(3), 991–1003.
Espert L., Beaumelle B., Vergne I. (2015), ‘Autophagy in Mycobacterium tuberculosis and
HIV infections’, Frontiers in Cellular and Infection Microbiology 5(June), Article 49.
Fackler O. T. (2015), ‘Spotlight on HIV-1 Nef: SERINC3 and SERINC5 identified as restric-
tion factors antagonized by the pathogenesis factor’, Viruses 7, 6730–6738.
Fasshauer D., Sutton R. B., Brunger A. T., Jahn R. (1998), ‘Conserved structural features
of the synaptic fusion complex: SNARE proteins reclassified as Q- and R-SNAREs’,
Proceedings of the National Academy of Sciences 95(26), 15781–15786.
Feng L. et al. (1999), ‘The β3A subunit gene (Ap3b1) of the AP-3 adaptor complex is
altered in the mouse hypopigmentation mutant pearl, a model for Hermansky-Pudlak
syndrome and night blindness’, Human Molecular Genetics 8(2), 323–330.
Ferguson S. et al. (2009), ‘Coordinated Actions of Actin and BAR Proteins Upstream of
Dynamin at Endocytic Clathrin-Coated Pits’, Developmental Cell 17(6), 811–822.
Ferguson S. M., De Camilli P. (2012), ‘Dynamin, a membrane-remodelling GTPase’,
Nature Reviews Molecular Cell Biology 13(2), 75–88.
Field M. C., Dacks J. B. (2009), ‘First and last ancestors: reconstructing evolution of
the endomembrane system with ESCRTs, vesicle coat proteins, and nuclear pore
complexes’, Current Opinion in Cell Biology 21(1), 4–13.
Ford M. G. J., Mills I. G., Peter B. J., Vallis Y., Praefcke G. J. K., Evans P. R., McMahon H. T.
(2002), ‘Curvature of clathrin-coated pits driven by epsin’, Nature 419(6905), 361–366.
Foster L. J., Hoog C. L. D., Mann M. (2003), ‘Unbiased quantitative proteomics of lipid
rafts reveals high specificity for signaling factors’, Proceedings of the National Academy
of Sciences of the United States of America 100(10), 5813–5818.
Foster L. J., Hoog C. L. D., Zhang Y., Zhang Y., Xie X., Mootha V. K., Mann M. (2006), ‘A
Mammalian Organelle Map by Protein Correlation Profiling’, Cell 125, 187–199.
Fotin A., Cheng Y., Sliz P., Grigorieff N., Harrison S. C., Kirchhausen T., Walz T. (2004),
‘Molecular model for a complete clathrin lattice from electron cryomicroscopy’, Nature
432(7017), 573–579.
Frazier M. N., Davies A. K., Voehler M., Kendall A. K., Borner G. H. H., Chazin W. J.,
Robinson M. S., Jackson L. P. (2016), ‘Molecular Basis for the Interaction Between AP4
beta4 and its Accessory Protein, Tepsin’, Traffic 17, 400–415.
Fu Y., Foden J. a., Khayter C., Maeder M. L., Reyon D., Joung J. K., Sander J. D. (2013),
‘High-frequency off-target mutagenesis induced by CRISPR-Cas nucleases in human
cells.’, Nature biotechnology 31(9), 822–826.
260 References
Fukuda M., Kobayashi H., Ishibashi K., Ohbayashi N. (2011), ‘Genome-wide Investigation
of the Rab Binding Activity of RUN Domains: Development of a Novel Tool that
Specifically Traps GTP-Rab35’, Cell Structure and Function 36(2), 155–170.
Gao Y. et al. (2010), ‘Loss of Retrograde Endocannabinoid Signaling and Reduced Adult
Neurogenesis in Diacylglycerol Lipase Knock-out Mice’, Journal of Neuroscience
30(6), 2017–2024.
Gavin A. C. et al. (2006), ‘Proteome survey reveals modularity of the yeast cell machinery’,
Nature 440(7084), 631–636.
Gibson D. G., Young L., Chuang R.-Y., Venter J. C., Hutchison C. A., Smith H. O. (2009),
‘Enzymatic assembly of DNA molecules up to several hundred kilobases.’, Nature
methods 6(5), 343–345.
Gillingham A. K., Munro S. (2003), ‘Long coiled-coil proteins and membrane traffic’,
Biochimica et Biophysica Acta - Molecular Cell Research 1641(2-3), 71–85.
Gillingham A. K., Munro S. (2016), ‘Finding the Golgi: Golgin Coiled-Coil Proteins Show
the Way’, Trends in Cell Biology 26(6), 399–408.
Gingras A.-C., Gstaiger M., Raught B., Aebersold R. (2007), ‘Analysis of protein complexes
using mass spectrometry’, Nature Reviews Molecular Cell Biology 8, 645–654.
Girard M., Allaire P. D., Mcpherson P. S., Blondeau F. (2005), ‘Non-stoichiometric Relation-
ship between Clathrin Heavy and Light Chains Revealed by Quantitative Comparative
Proteomics of Clathrin-coated Vesicles from Brain and Liver’, Molecular & Cellular
Proteomics 4(8), 1145–1154.
Gomez-Navarro N., Miller E. A. (2016), ‘COP-coated vesicles’, Current Biology 26(2), R54–
R57.
Goodwin J. M. et al. (2017), ‘Autophagy-Independent Lysosomal Targeting Regulated by
ULK1/2-FIP200 and ATG9’, Cell Reports 20, 2341–2356.
Grossman T. R., Luque J. M., Nelson N. (2000), ‘Identification of a ubiquitous family of
membrane proteins and their expression in mouse brain.’, The Journal of experimental
biology 203(Pt 3), 447–457.
Grumati P., Dikic I. (2018), ‘Ubiquitin signaling and autophagy’, Journal of Biological
Chemistry 293(15), 5404–5413.
Gygi S. P., Rist B., Gerber S. A., Turecek F., Gelb M. H., Aebersold R. (1999), ‘Quantita-
tive analysis of complex protein mixtures using isotope-coded affinity tags.’, Nature
biotechnology 17, 994–999.
Hamasaki M. et al. (2013), ‘Autophagosomes form at ER-mitochondria contact sites’,
Nature 495(7441), 389–393.
Hanada T., Noda N. N., Satomi Y., Ichimura Y., Fujioka Y., Takao T., Inagaki F., Ohsumi Y.
(2007), ‘The Atg12-Atg5 conjugate has a novel E3-like activity for protein lipidation in
autophagy’, Journal of Biological Chemistry 282(52), 37298–37302.
References 261
Hara-Kuge S., Kuge O., Orci L., Amherdt M., Ravazzola M., Wieland F. T., Rothman J. E.
(1994), ‘En Bloc incorporation of coatomer subunits during the assembly of COP-
coated vesicles’, Journal of Cell Biology 124(6), 883–892.
Hara T. et al. (2006), ‘Suppression of basal autophagy in neural cells causes neurodegen-
erative disease in mice’, Nature 441, 885–889.
Hardies K. et al. (2015), ‘Recessive loss-of-function mutations in AP4S1 cause mild fever-
sensitive seizures, developmental delay and spastic paraplegia through loss of AP-4
complex assembly’, Human Molecular Genetics 24(8), 2218–2227.
Harding T. M., Morano K. A., Scott S. V., Klionsky D. J. (1995), ‘Isolation and characteriza-
tion of yeast mutants in the cytoplasm to vacuole protein targeting pathway’, Journal
of Cell Biology 131(3), 591–602.
Harris T. W., Hartwieg E., Horvitz H. R., Jorgensen E. M. (2000), ‘Mutations in synaptojanin
disrupt synaptic vesicle recycling’, Journal of Cell Biology 150(3), 589–599.
Hayashi-Nishino M., Fujita N., Noda T., Yamaguchi A., Yoshimori T., Yamamoto A. (2009),
‘A subdomain of the endoplasmic reticulum forms a cradle for autophagosome for-
mation’, Nature Cell Biology 11(12), 1433–1437.
Hein M. Y. et al. (2015), ‘A Human Interactome in Three Quantitative Dimensions Orga-
nized by Stoichiometries and Abundances’, Cell 163, 712–723.
Henne W. M., Boucrot E., Meinecke M., Evergren E., Vallis Y., Mittal R., McMahon H. T.
(2010), ‘FCHo proteins are nucleators of Clathrin-Mediated endocytosis’, Science
328(5983), 1281–1284.
Higuchi Y. et al. (2016), ‘Mutations in MME cause an autosomal-recessive Charcot-Marie-
Tooth disease type 2’, Annals of Neurology 79(4), 659–672.
Hinshaw J. E., Schmid S. L. (1995), ‘Dynamin self-assembles into rings suggesting a
mechanism for coated vesicle budding’, Nature 374(6518), 190–192.
Hirst J., Barlow L. D., Francisco G. C., Sahlender D. A., Seaman M. N. J., Dacks J. B., Robin-
son M. S. (2011), ‘The fifth adaptor protein complex.’, PLoS biology 9(10), e1001170.
Hirst J., Borner G. H. H., Antrobus R., Peden A. A., Hodson N. A., Sahlender D. A., Robin-
son M. S. (2012), ‘Distinct and overlapping roles for AP-1 and GGAs revealed by the
"knocksideways" system’, Current Biology 22(18), 1711–1716.
Hirst J., Borner G. H. H., Edgar J., Hein M. Y., Mann M., Buchholz F., Antrobus R., Robinson
M. S. (2013a), ‘Interaction between AP-5 and the hereditary spastic paraplegia proteins
SPG11 and SPG15’, Molecular Biology of the Cell 24(16), 2558–2569.
Hirst J., Bright N. A., Rous B., Robinson M. S. (1999), ‘Characterization of a fourth adaptor-
related protein complex.’, Molecular biology of the cell 10(8), 2787–2802.
Hirst J. et al. (2014), ‘Characterization of TSET, an ancient and widespread membrane
trafficking complex’, eLife 3, e02866.
262 References
Hirst J., Irving C., Borner G. H. H. (2013b), ‘Adaptor protein complexes AP-4 and AP-
5: new players in endosomal trafficking and progressive spastic paraplegia.’, Traffic
14, 153–164.
Hirst J., Itzhak D. N., Antrobus R., Borner G. H., Robinson M. S. (2018), ‘Role of the
AP-5 adaptor protein complex in late endosome-to-Golgi retrieval’, PLoS Biology
16(1), e2004411.
Hirst J., Lui W. W., Bright N. A., Totty N., Seaman M. N., Robinson M. S. (2000), ‘A family
of proteins with γ-adaptin and VHS domains that facilitate trafficking between the
trans-golgi network and the vacuole/lysosome’, Journal of Cell Biology 149(1), 67–79.
Hirst J., Miller S. E., Taylor M. J., Fischer von Mollard G., Robinson M. S. (2004), ‘EpsinR
Is an Adaptor for the SNARE Protein Vti1b’, Molecular biology of the cell 15, 5593–5602.
Hirst J., Motley A., Harasaki K., Chew S. Y. P., Robinson M. S. (2003), ‘EpsinR: an ENTH
Domain-containing Protein that Interacts with AP-1’, Molecular Biology of the Cell
14, 625–641.
Hirst J., Seaman M. N., Buschow S. I., Robinson M. S. (2007), ‘The role of cargo proteins
in GGA recruitment’, Traffic 8(5), 594–604.
Hong W. J., Lev S. (2014), ‘Tethering the assembly of SNARE complexes’, Trends in Cell
Biology 24(1), 35–43.
Hooke R. (1665), Micrographia: or Some physiological descriptions of minute bodies
made by magnifying glasses : With observations and inquiries thereupon, printed by Jo.
Martyn, and Ja. Allestry, printers to the Royal Society, and are to be sold at their shop
at the Bell in S. Paul’s Church-yard, London.
Howell S., Nalbantoglu J., Crine P. (1995), ‘Neutral endopeptidase can hydrolyze beta-
amyloid(1-40) but shows no effect on beta-amyloid precursor protein metabolism’,
Peptides 16(4), 647–652.
Hsu P. D. et al. (2013), ‘DNA targeting specificity of RNA-guided Cas9 nucleases.’, Nature
biotechnology 31(9), 827–832.
Huff J. (2015), ‘The Airyscan detector from ZEISS: confocal imaging with improved
signal-to-noise ratio and super-resolution’, Nature Methods 12(1205), i–ii.
Imai K. et al. (2016), ‘Atg9A trafficking through the recycling endosomes is required for
autophagosome formation’, Journal of Cell Science 129(20), 3781–3791.
Inuzuka M., Hayakawa M., Ingi T. (2005), ‘Serinc, an activity-regulated protein family,
incorporates serine into membrane lipid synthesis’, Journal of Biological Chemistry
280(42), 35776–35783.
Itoh T., Koshiba S., Kigawa T., Kikuchi A., Yokoyama S., Takenawa T. (2001), ‘Role of the
ENTH Domain in Phosphatidylinositol-4,5-Bisphosphate Binding and Endocytosis’,
Science 291(5506), 1047–1051.
References 263
Itzhak D. N. et al. (2017), ‘A Mass Spectrometry-Based Approach for Mapping Protein
Subcellular Localization Reveals the Spatial Proteome of Mouse Primary Neurons’,
Cell Reports 20, 2706–2718.
Itzhak D. N., Tyanova S., Cox J., Borner G. H. H. (2016), ‘Global, quantitative and dynamic
mapping of protein subcellular localization’, eLife 5, e16950.
Ivankovic D., López-Doménech G., Drew J., Tooze S. A., Kittler J. T. (2017), ‘AP-4 mediated
ATG9A sorting underlies axonal and autophagosome biogenesis defects in a mouse
model of AP-4 deficiency syndrome’, bioRxiv pre-print.
Jackson A. C., Nicoll R. A. (2011), ‘The Expanding Social Network of Ionotropic Glutamate
Receptors: TARPs and Other Transmembrane Auxiliary Subunits’, Neuron 70(2), 178–
199.
Jackson L. P. et al. (2010), ‘A large-scale conformational change couples membrane
recruitment to cargo binding in the AP2 clathrin adaptor complex’, Cell 141(7), 1220–
1229.
Jacomin A. C., Samavedam S., Promponas V., Nezis I. P. (2016), ‘iLIR database: A web
resource for LIR motif-containing proteins in eukaryotes’, Autophagy 12(10), 1945–
1953.
Jahn R., Scheller R. H. (2006), ‘SNAREs - Engines for membrane fusion’, Nature Reviews
Molecular Cell Biology 7(9), 631–643.
Jamieson J. D., Palade G. E. (1967), ‘Intracellular transport of secretory proteins in the
pancreatic exocrine cell. I. Role of the peripheral elements of the Golgi complex.’, The
Journal of cell biology 34(2), 577–596.
Janvier K., Bonifacino J. S. (2005), ‘Role of the Endocytic Machinery in the Sorting of
Lysosome-associated Membrane Proteins’, Molecular biology of the cell 16, 4231–4242.
Jiang P., Nishimura T., Sakamaki Y., Itakura E., Hatta T., Natsume T., Mizushima N. (2014),
‘The HOPS complex mediates autophagosome-lysosome fusion through interaction
with syntaxin 17’, Molecular Biology of the Cell 25(8), 1327–1337.
Jin Z., Schwend T., Fu J., Bao Z., Liang J., Zhao H., Mei W., Yang J. (2016), ‘Members of
the Rusc protein family interact with Sufu and inhibit vertebrate Hedgehog signaling’,
Development 143(21), 3944–3955.
Jinek M., Chylinski K., Fonfara I., Hauer M., Doudna J. A., Charpentier E. (2012), ‘A
Programmable Dual-RNA-Guided DNA Endonuclease in Adaptive Bacterial Immunity’,
Science 337, 816–821.
Kabeya Y. et al. (2000), ‘LC3, a mammalian homologue of yeast Apg8p, is localized in
autophagosome membranes after processing’, The EMBO Journal 19(21), 5720–5728.
Kakuta S., Yamaguchi J., Suzuki C., Sasaki M., Kazuno S., Uchiyama Y. (2017), ‘Small
GTPase Rab1B is associated with ATG9A vesicles and regulates autophagosome for-
mation’, The FASEB Journal 31(9), 3757–3773.
264 References
Kakuta S., Yamamoto H., Negishi L., Kondo-Kakuta C., Hayashi N., Ohsumi Y. (2012),
‘Atg9 vesicles recruit vesicle-tethering proteins Trs85 and Ypt1 to the autophagosome
formation site’, Journal of Biological Chemistry 287(53), 44261–44269.
Kalvari I., Tsompanis S., Mulakkal N. C., Osgood R., Johansen T., Nezis I. P., Promponas
V. J. (2014), ‘iLIR: A web resource for prediction of Atg8-family interacting proteins’,
Autophagy 10(5), 913–925.
Kanaseki T., Kadota K. (1969), ‘The "vesicle in a basket". A morphological study of the
coated vesicle isolated from the nerve endings of the guinea pig brain, with special
reference to the mechanism of membrane movements.’, The Journal of cell biology
42(1), 202–220.
Karanasios E. et al. (2016), ‘Autophagy initiation by ULK complex assembly on ER
tubulovesicular regions marked by ATG9 vesicles’, Nature Communications 7, 12420.
Katzmann D. J., Babst M., Emr S. D. (2001), ‘Ubiquitin-dependent sorting into the mul-
tivesicular body pathway requires the function of a conserved endosomal protein
sorting complex, ESCRT-I’, Cell 106(2), 145–155.
Keen J. H. (1987), ‘Clathrin assembly proteins: Affinity purification and a model for coat
assembly’, Journal of Cell Biology 105(5), 1989–1998.
Keen J. H., Willingham M. C., Pastan I. H. (1979), ‘Clathrin-Coated Vesicles : Isolation,
Dissociation and Factor-Dependent Reassociation of Clathrin Baskets’, Cell 16, 303–
312.
Kelleher N. L., Lin H. Y., Valaskovic G. A., Aaserud D. J., Fridriksson E. K., Mclafferty
F. W. (1999), ‘Top Down versus Bottom Up Protein Characterization by Tandem High-
Resolution Mass Spectrometry’, Journal of the American Chemical Society 121(4), 806–
812.
Kelly B. T., Graham S. C., Liska N., Dannhauser P. N., Höning S., Ungewickell E. J., Owen
D. J. (2014), ‘AP2 controls clathrin polymerization with a membrane-activated switch’,
Science 345(6195), 459–463.
Kim D. I., Kc B., Zhu W., Motamedchaboki K., Doye V., Roux K. J. (2014), ‘Probing nuclear
pore complex architecture with proximity-dependent biotinylation’, Proceedings of the
National Academy of Sciences of the United States of America 111(24), E2453–E2461.
Kim M. et al. (2016), ‘Mutation in ATG5 reduces autophagy and leads to ataxia with
developmental delay’, eLife 5, e12245.
Kimura S., Noda T., Yoshimori T. (2008), ‘Dynein-dependent movement of autophago-
somes mediates efficient encounters with lysosomes.’, Cell structure and function
33(1), 109–122.
Kimura T., Mandell M., Deretic V. (2016), ‘Precision autophagy directed by receptor regu-
lators – emerging examples within the TRIM family’, Journal of Cell Science 129(5), 881–
891.
References 265
Kirchhausen T. (2000), ‘Three ways to make a vesicle’, Nature Reviews Molecular Cell
Biology 1(3), 187–198.
Kirchhausen T. (2012), ‘Bending membranes’, Nature Cell Biology 14(9), 906–908.
Kirchhausen T., Owen D., Harrison S. C. (2014), ‘Molecular Structure, Function, and
Dynamics of Clathrin-Mediated Membrane Traffic’, Cold Spring Harbor perspectives
in biology 6, a016725.
Klionsky D. D. J. et al. (2016), ‘Guidelines for the use and interpretation of assays for
monitoring autophagy (3rd edition).’, Autophagy 12(1), 1–222.
Klionsky D. J. et al. (2003), ‘A unified nomenclature for yeast autophagy-related genes’,
Developmental Cell 5(4), 539–545.
Koike-Yusa H., Li Y., Tan E.-P., Velasco-Herrera M. D. C., Yusa K. (2014), ‘Genome-wide
recessive genetic screening in mammalian cells with a lentiviral CRISPR-guide RNA
library’, Nature Biotechnology 32(3), 267–273.
Komatsu M. et al. (2006), ‘Loss of autophagy in the central nervous system causes neu-
rodegeneration in mice’, Nature 441, 880–884.
Kong X. F. et al. (2013), ‘A Novel Homozygous p.R1105X Mutation of the AP4E1 Gene
in Twins with Hereditary Spastic Paraplegia and Mycobacterial Disease’, PLoS ONE
8(3), e58286.
Krogan N. J. et al. (2006), ‘Global landscape of protein complexes in the yeast Saccha-
romyces cerevisiae’, Nature 440(7084), 637–643.
Kulak N. A., Pichler G., Paron I., Nagaraj N., Mann M. (2014), ‘Minimal, encapsulated
proteomic-sample processing applied to copy-number estimation in eukaryotic cells.’,
Nature methods 11(3), 319–324.
Kural C., Tacheva-grigorova S. K., Boulant S., Cocucci E., Baust T., Duarte D., Kirch-
hausen T. (2012), ‘Report Dynamics of Intracellular Clathrin/AP1- and Clathrin/AP3-
Containing Carriers’, Cell Reports 2(5), 1111–1119.
Lam A. J. et al. (2012), ‘Improving FRET dynamic range with bright green and red fluo-
rescent proteins’, Nature Methods 9(10), 1005–1012.
Lamb C. A., Nühlen S., Judith D., Frith D., Snijders A. P., Behrends C., Tooze S. A. (2016),
‘TBC1D14 regulates autophagy via the TRAPP complex and ATG9 traffic’, The EMBO
Journal 35(3), 281–301.
Lambert J.-P., Tucholska M., Go C., Knight J. D. R., Gingras A.-C. (2015), ‘Proximity
biotinylation and affinity purification are complementary approaches for the inter-
actome mapping of chromatin-associated protein complexes’, Journal of Proteomics
118, 81–94.
Landry J. J. M. et al. (2013), ‘The Genomic and Transcriptomic Landscape of a HeLa Cell
Line’, G3: GENES, GENOMES, GENETICS 3, 1213–1224.
266 References
Lang T., Reiche S., Straub M., Bredschneider M., Thumm M. (2000), ‘Autophagy and the
cvt pathway both depend on AUT9’, Journal of Bacteriology 182(8), 2125–2133.
Leavitt S., Freire E. (2001), ‘Direct measurement of protein binding energetics by isother-
mal titration calorimetry’, Current Opinion in Structural Biology 11(5), 560–566.
Lebel M., Mes-Masson A.-M. (1994), ‘Sequence analysis of a novel cDNA which is over-
expressed in testicular tumors from polyomavirus large T-antigen transgenic mice’,
DNA Sequence 5(1), 31–39.
Lederkremer G. Z., Cheng Y., Petre B. M., Vogan E., Springer S., Schekman R., Walz T.,
Kirchhausen T. (2001), ‘Structure of the Sec23p/24p and Sec13p/31p complexes of
COPII’, Proceedings of the National Academy of Sciences of the United States of America
98(19), 10704–10709.
Lee M. C., Orci L., Hamamoto S., Futai E., Ravazzola M., Schekman R. (2005), ‘Sar1p
N-terminal helix initiates membrane curvature and completes the fission of a COPII
vesicle’, Cell 122(4), 605–617.
Levine B., Kroemer G. (2008), ‘Autophagy in the Pathogenesis of Disease’, Cell 132(1), 27–
42.
Liu A. P., Aguet F., Danuser G., Schmid S. L. (2010), ‘Local clustering of transferrin recep-
tors promotes clathrin-coated pit initiation’, Journal of Cell Biology 191(7), 1381–1393.
Liu H., Sadygov R. G., Yates J. R. (2004), ‘A model for random sampling and estimation of
relative protein abundance in shotgun proteomics’, Analytical Chemistry 76(14), 4193–
4201.
Lohi O., Lehto V.-P. (1998), ‘VHS domain marks a group of proteins involved in endocyto-
sis and vesicular trafficking’, FEBS Letters 440(3), 255–257.
Lu S. L. et al. (2017), ‘Endothelial cells are intrinsically defective in xenophagy of Strepto-
coccus pyogenes’, PLoS Pathogens 13(7), e1006444.
MacDonald J. I. S., Dietrich A., Gamble S., Hryciw T., Grant R. I., Meakin S. O. (2012),
‘Nesca, a novel neuronal adapter protein, links the molecular motor kinesin with the
pre-synaptic membrane protein, syntaxin-1, in hippocampal neurons’, Journal of
Neurochemistry 121(6), 861–880.
MacDonald J. I. S., Kubu C. J., Meakin S. O. (2004), ‘Nesca, a novel adapter, translocates to
the nuclear envelope and regulates neurotrophin-induced neurite outgrowth’, Journal
of Cell Biology 164(6), 851–862.
Maday S., Holzbaur E. L. F. (2014), ‘Autophagosome biogenesis in primary neurons
follows an ordered and spatially regulated pathway’, Developmental Cell 30(1), 71–85.
Maday S., Wallace K. E., Holzbaur E. L. F. (2012), ‘Autophagosomes initiate distally and
mature during transport toward the cell soma in primary neurons’, Journal of Cell
Biology 196(4), 407–417.
References 267
Malhotra V., Serafini T., Orci L., Shepherd J. C., Rothman J. E. (1989), ‘Purification of a
novel class of coated vesicles mediating biosynthetic protein transport through the
Golgi stack’, Cell 58(2), 329–336.
Mali P., Yang L., Esvelt K. M., Aach J., Guell M., Dicarlo J. E., Norville J. E., Church G. M.
(2013), ‘RNA-Guided Human Genome Engineering via Cas9’, Science 339, 823–826.
Manders E. M. M., Stap J., Brakenhoff G. J., Van Driel R., Aten A. (1992), ‘Dynamics
of three-dimensional replication patterns during the S-phase, analysed by double
labelling of DNA and confocal microscopy’, Journal of cell science 103(Pt 3), 857–862.
Mao Z., Bozzella M., Seluanov A., Gorbunova V. (2008), ‘Comparison of nonhomologous
end joining and homologous recombination in human cells’, DNA Repair 7(10), 1765–
1771.
Mari M., Griffith J., Rieter E., Krishnappa L., Klionsky D. J., Reggiori F. (2010), ‘An Atg9-
containing compartment that functions in the early steps of autophagosome biogene-
sis’, Journal of Cell Biology 190(6), 1005–1022.
Martens S., Nakamura S., Yoshimori T. (2016), ‘Phospholipids in Autophagosome Forma-
tion and Fusion’, Journal of Molecular Biology 428, 4819–4827.
Massol R. H., Boll W., Griffin A. M., Kirchhausen T. (2006), ‘A burst of auxilin recruitment
determines the onset of clathrin-coated vesicle uncoating’, Proceedings of the National
Academy of Sciences 103(27), 10265–10270.
Matsuda S. et al. (2000), ‘Molecular cloning and characterization of a novel human gene
(NESCA) which encodes a putative adapter protein containing SH3.’, Biochimica et
biophysica acta 1491(1-3), 321–326.
Matsuda S., Miura E., Matsuda K., Kakegawa W., Kohda K., Watanabe M., Yuzaki M. (2008),
‘Accumulation of AMPA receptors in autophagosomes in neuronal axons lacking
adaptor protein AP-4.’, Neuron 57(5), 730–745.
Mattera R., Guardia C. M., Sidhu S. S., Bonifacino J. S. (2015), ‘Bivalent motif-ear interac-
tions mediate the association of the accessory protein tepsin with the AP-4 adaptor
complex’, Journal of Biological Chemistry 290(52), 30736–30749.
Mattera R., Park S. Y., De Pace R., Guardia C. M., Bonifacino J. S. (2017), ‘AP-4 mediates
export of ATG9A from the trans-Golgi network to promote autophagosome formation’,
Proceedings of the National Academy of Sciences 114(50), E10697–E10706.
Mayer B. J., Hamaguchi M., Hanafusa H. (1988), ‘A novel viral oncogene with structural
similarity to phospholipase C’, Nature 332(6161), 272–275.
Mazzarello P. (1999), ‘A unifying concept: The history of cell theory’, Nature Cell Biology
1(1), E13–E15.
McMahon H. T., Mills I. G. (2004), ‘COP and clathrin-coated vesicle budding: Different
pathways, common approaches’, Current Opinion in Cell Biology 16(4), 379–391.
268 References
McNally K. E., Cullen P. J. (2018), ‘Endosomal Retrieval of Cargo: Retromer Is Not Alone’,
Trends in Cell Biology 28(10), 807–822.
McNally K. E. et al. (2017), ‘Retriever is a multiprotein complex for retromer-independent
endosomal cargo recycling’, Nature Cell Biology 19(10), 1214–1225.
McNew J. A., Parlatl F., Fukuda R., Johnston R. J., Paz K., Paumet F., Söllner T. H., Roth-
man J. E. (2000), ‘Compartmental specificity of cellular membrane fusion encoded in
SNARE proteins’, Nature 407(6801), 153–159.
Menzies F. M. et al. (2017), ‘Autophagy and Neurodegeneration : Pathogenic Mechanisms
and Therapeutic Opportunities’, Neuron 93, 1015–1034.
Miller S. E., Collins B. M., McCoy A. J., Robinson M. S., Owen D. J. (2007), ‘A SNARE-
adaptor interaction is a new mode of cargo recognition in clathrin-coated vesicles.’,
Nature 450(7169), 570–574.
Mills I. G. et al. (2003), ‘Epsinr: An AP1/clathrin interacting protein involved in vesicle
trafficking’, Journal of Cell Biology 160(2), 213–222.
Mishra S. K., Keyel P. A., Hawryluk M. J., Agostinelli N. R., Watkins S. C., Traub L. M.
(2002a), ‘Disabled-2 exhibits the properties of a cargo-selective endocytic clathrin
adaptor’, The EMBO Journal 21(18), 4915–4926.
Mishra S. K., Watkins S. C., Traub L. M. (2002b), ‘The autosomal recessive hypercholes-
terolemia (ARH) protein interfaces directly with the clathrin-coat machinery’, Pro-
ceedings of the National Academy of Sciences 99(25), 16099–16104.
Mizushima N. et al. (1998), ‘A protein conjugation system essential for autophagy’, Nature
395(6700), 395–398.
Moreno-De-Luca A. et al. (2011), ‘Adaptor protein complex-4 (AP-4) deficiency causes a
novel autosomal recessive cerebral palsy syndrome with microcephaly and intellectual
disability.’, Journal of medical genetics 48(2), 141–144.
Morris S. M., Arden S. D., Roberts R. C., Kendrick-Jones J., Cooper J. A., Luzio J. P., Buss
F. (2002), ‘Myosin VI binds to and localises with Dab2, potentially linking receptor-
mediated endocytosis and the actin cytoskeleton’, Traffic 3(5), 331–341.
Morton C., Campbell I. (1994), ‘SH3 Domains: Molecular ‘Velcro’’, Current Biology
4(7), 615–617.
Nagy E., Maquat L. E. (1998), ‘A rule for termination-codon position within intron-
containing genes: When nonsense affects RNA abundance’, Trends in Biochemical
Sciences 23(6), 198–199.
Nair U. et al. (2011), ‘SNARE proteins are required for macroautophagy’, Cell 146(2), 290–
302.
Nakagawa I. et al. (2004), ‘Autophagy defends cells against invading group A Streptococ-
cus’, Science 306(5698), 1037–1040.
References 269
Napolitano G., Mirra S., Monfregola J., Lavorgna A., Leonardi A., Ursini M. V. (2009),
‘NESCA: A new NEMO/IKKgamma and TRAF6 interacting protein’, Journal of Cellular
Physiology 220(November 2008), 410–417.
Navarro Negredo P., Edgar J. R., Manna P. T., Antrobus R., Robinson M. S. (2018), ‘The
WDR11 complex facilitates the tethering of AP-1-derived vesicles’, Nature Communi-
cations 9(596).
Nezich C. L., Wang C., Fogel A. I., Youle R. J. (2015), ‘MiT/TFE transcription factors are
activated during mitophagy downstream of Parkin and Atg5’, Journal of Cell Biology
210(3), 435–450.
Noble W. S. (2006), ‘What is a support vector machine ?’, Nature biotechnology
24(12), 1565–1567.
Noda T. (2017), ‘Autophagy in the context of the cellular membrane-trafficking system:
the enigma of Atg9 vesicles.’, Biochemical Society transactions p. BST20170128.
Noda T., Kim J., Huang W. P., Baba M., Tokunaga C., Ohsumi Y., Klionsky D. J. (2000),
‘Apg9p/Cvt7p is an integral membrane protein required for transport vesicle formation
in the Cvt and autophagy pathways’, Journal of Cell Biology 148(3), 465–479.
Novick P., Field C., Schekman R. (1980), ‘Identification of 23 complementation groups
required for post-translational events in the yeast secretory pathway’, Cell 21(1), 205–
215.
Ong S.-E., Blagoev B., Kratchmarova I., Kristensen D. B., Steen H., Pandey A., Mann M.
(2002), ‘Stable Isotope Labeling by Amino Acids in Cell Culture, SILAC, as a Simple
and Accurate Approach to Expression Proteomics’, Molecular & Cellular Proteomics
1(5), 376–386.
Orsi A., Razi M., Dooley H. C., Robinson D., Weston A. E., Collinson L. M., Tooze S. A.
(2012), ‘Dynamic and transient interactions of Atg9 with autophagosomes, but not
membrane integration, are required for autophagy’, Molecular biology of the cell
23(10), 1860–1873.
Owen D. J., Collins B. M., Evans P. R. (2004), ‘ADAPTORS FOR CLATHRIN COATS: Struc-
ture and Function’, Annual Review of Cell and Developmental Biology 20(1), 153–191.
Owen D. J., Evans P. R. (1998), ‘A structural explanation for the recognition of tyrosine-
based endocytotic signals’, Science 282(5392), 1327–1332.
Owen D. J., Vallis Y., Noble M. E., Hunter J. B., Dafforn T. R., Evans P. R., McMahon H. T.
(1999), ‘A structural explanation for the binding of multiple ligands by the α- adaptin
appendage domain’, Cell 97(6), 805–815.
Owen D. J., Vallis Y., Pearse B. M., McMahon H. T., Evans P. R. (2000), ‘The structure and
function of the beta 2-adaptin appendage domain.’, The EMBO journal 19(16), 4216–
4227.
270 References
Paczkowski J. E., Richardson B. C., Fromme J. C. (2015), ‘Cargo adaptors: Structures
illuminate mechanisms regulating vesicle biogenesis’, Trends in Cell Biology 25(7), 408–
416.
Page L. J., Sowerby P. J., Lui W. W. Y., Robinson M. S. (1999), ‘γ-Synergin: An EH Domain –
containing Protein that Interacts with γ-Adaptin’, Journal of Cell Biology 146(5), 993–
1004.
Palade G. (1975), ‘Intracellular Aspects of the Process of Protein Synthesis’, Science
189(4200), 347–358.
Paolini L., Radeghieri A., Civini S., Caimi L., Ricotta D. (2011), ‘The epsilon hinge-ear
region regulates membrane localization of the AP-4 complex.’, Traffic 12(11), 1604–
1619.
Pearse B. M. (1975), ‘Coated vesicles from pig brain: purification and biochemical char-
acterization.’, Journal of molecular biology 97(1), 93–98.
Pearse B. M., Robinson M. S. (1984), ‘Purification and properties of 100-kd proteins from
coated vesicles and their reconstitution with clathrin.’, The EMBO journal 3(9), 1951–
1957.
Peche L. Y., Ladelfa M. F., Toledo M. F., Mano M., Laiseca J. E., Schneider C., Monte M.
(2015), ‘Human MageB2 protein expression enhances E2F transcriptional activity,
cell proliferation, and resistance to ribotoxic stress’, Journal of Biological Chemistry
290(49), 29652–29662.
Peden A. A., Oorschot V., Hesser B. A., Austin C. D., Scheller R. H., Klumperman J. (2004),
‘Localization of the AP-3 adaptor complex defines a novel endosomal exit site for
lysosomal membrane proteins’, Journal of Cell Biology 164(7), 1065–1076.
Peden A. A., Rudge R. E., Lui W. W. Y., Robinson M. S. (2002), ‘Assembly and function of AP-
3 complexes in cells expressing mutant subunits’, Journal of Cell Biology 156(2), 327–
336.
Peltier J. B., Friso G., Kalume D. E., Roepstorff P., Nilsson F., Adamska I., van Wijk K. J.
(2000), ‘Proteomics of the chloroplast: systematic identification and targeting analysis
of lumenal and peripheral thylakoid proteins.’, The Plant cell 12, 319–341.
Peter B. J., Kent H. M., Mills I. G., Vallis Y., Butler P. J. G., Evans P. R., McMahon H. T. (2004),
‘BAR Domains as Sensors of Membrane Curvature: The Amphiphysin BAR Structure’,
Science 303(5657), 495–499.
Plemel R. L., Lobingier B. T., Brett C. L., Angers C. G., Nickerson D. P., Paulsel A., Sprague
D., Merz A. J. (2011), ‘Subunit organization and Rab interactions of Vps-C protein
complexes that control endolysosomal membrane traffic’, Molecular Biology of the
Cell 22(8), 1353–1363.
Popovic D., Dikic I. (2014), ‘TBC1D5 and the AP2 complex regulate ATG9 trafficking and
initiation of autophagy’, EMBO Reports 15(4), 392–401.
References 271
Pryor P. R. et al. (2008), ‘Molecular Basis for the Sorting of the SNARE VAMP7 into Endo-
cytic Clathrin-Coated Vesicles by the ArfGAP Hrb’, Cell 134(5), 817–827.
Puri C., Renna M., Bento C. F., Moreau K., Rubinsztein D. C. (2013), ‘Diverse autophago-
some membrane sources coalesce in recycling endosomes’, Cell 154(6), 1285–1299.
Rabilloud T. et al. (1998), ‘Two-dimensional electrophoresis of human placental mito-
chondria and protein identification by mass spectrometry: toward a human mito-
chondrial proteome.’, Electrophoresis 19, 1006–1014.
Raiborg C., Wesche J., Malerod L., Stenmark H. (2006), ‘Flat clathrin coats on endosomes
mediate degradative protein sorting by scaffolding Hrs in dynamic microdomains’,
Journal of Cell Science 119(12), 2414–2424.
Ran F. A. et al. (2013), ‘Double nicking by RNA-guided CRISPR cas9 for enhanced genome
editing specificity’, Cell 154(6), 1380–1389.
Rappsilber J., Mann M., Ishihama Y. (2007), ‘Protocol for micro-purification, enrichment,
pre-fractionation and storage of peptides for proteomics using StageTips.’, Nature
protocols 2(8), 1896–1906.
Reggiori F., Shintani T., Nair U., Klionsky D. J. (2005), ‘Atg9 cycles between mitochondria
and the pre-autophagosomal structure in yeasts.’, Autophagy 1(2), 101–109.
Reggiori F., Tucker K. A., Stromhaug P. E., Klionsky D. J. (2004), ‘The Atg1-Atg13 complex
regulates Atg9 and Atg23 retrieval transport from the pre-autophagosomal structure’,
Developmental Cell 6(1), 79–90.
Ren X., Farias G. G., Canagarajah B. J., Bonifacino J. S., Hurley J. H. (2013), ‘Structural
basis for recruitment and activation of the AP-1 clathrin adaptor complex by Arf1’,
Cell 152(4), 755–767.
Robinson M. S. (1987), ‘100-kD coated vesicle proteins: Molecular heterogeneity and
intracellular distribution studied with monoclonal antibodies’, Journal of Cell Biology
104(4), 887–895.
Robinson M. S. (2004), ‘Adaptable adaptors for coated vesicles.’, Trends in cell biology
14(4), 167–174.
Robinson M. S. (2015), ‘Forty Years of Clathrin-coated Vesicles’, Traffic 16, 1210–1238.
Robinson M. S., Pearse B. M. (1986), ‘Immunofluorescent localization of 100K coated
vesicle proteins’, Journal of Cell Biology 102(1), 48–54.
Robinson M. S., Sahlender D. a., Foster S. D. (2010), ‘Rapid inactivation of proteins by
rapamycin-induced rerouting to mitochondria.’, Developmental cell 18, 324–331.
Romanov J., Walczak M., Ibiricu I., Schüchner S., Ogris E., Kraft C., Martens S. (2012),
‘Mechanism and functions of membrane binding by the Atg5-Atg12/Atg16 complex
during autophagosome formation’, EMBO Journal 31(22), 4304–4317.
Rosa A. et al. (2015), ‘HIV-1 Nef promotes infection by excluding SERINC5 from virion
incorporation’, Nature 526(7572), 212–217.
272 References
Rosenthal J. A., Chen H., Slepnev V. I., Pellegrini L., Salcini A. E., Di Fiore P. P., De Camilli
P. (1999a), ‘The epsins define a family of proteins that interact with components of
the clathrin coat and contain a new protein module’, Journal of Biological Chemistry
274(48), 33959–33965.
Rosenthal J. A., Chen H., Slepnev V. I., Pellegrini L., Salcini A. E., Di Fiore P. P., De Camilli
P. (1999b), ‘The epsins define a family of proteins that interact with components of
the clathrin coat and contain a new protein module’, Journal of Biological Chemistry
274(48), 33959–33965.
Ross B. H., Lin Y., Corales E. A., Burgos P. V., Mardones G. A. (2014), ‘Structural and func-
tional characterization of cargo-binding sites on the µ4-subunit of adaptor protein
complex 4.’, PloS one 9(2), e88147.
Ross P. L. et al. (2004), ‘Multiplexed Protein Quantitation in <i>Saccharomyces cere-
visiae</i> Using Amine-reactive Isobaric Tagging Reagents’, Molecular & Cellular
Proteomics 3(12), 1154–1169.
Roth T., Porter K. (1964), ‘Yolk Protein Uptake in the Oocyte of the Mosquito Aedes
Aegypti.’, The Journal of cell biology 20, 313–332.
Rothman J. E., Bursztyn-Pettegrew H., Fine R. E. (1980), ‘Transport of the membrane
glycoprotein of vesicular stomatitis virus to the cell surface in two stages by clathrin-
coated vesicles.’, The Journal of cell biology 86(1), 162–171.
Rothman J. E., Fine R. E. (1980), ‘Coated vesicles transport newly synthesized membrane
glycoproteins from endoplasmic reticulum to plasma membrane in two successive
stages.’, Proceedings of the National Academy of Sciences of the United States of America
77(2), 780–784.
Roux A., Uyhazi K., Frost A., De Camilli P. (2006), ‘GTP-dependent twisting of dynamin
implicates constriction and tension in membrane fission’, Nature 441(7092), 528–531.
Roux K. J., Kim D. I., Raida M., Burke B. (2012), ‘A promiscuous biotin ligase fusion
protein identifies proximal and interacting proteins in mammalian cells.’, The Journal
of cell biology 196(6), 801–810.
Saitoh T. et al. (2009), ‘Atg9a controls dsDNA-driven dynamic translocation of STING
and the innate immune response’, Proceedings of the National Academy of Sciences of
the United States of America 106(49), 20842–20846.
Sakamoto R., Byrd D. T., Brown H. M., Hisamoto N., Matsumoto K., Jin Y. (2004), ‘The
Caenorhabditis elegans UNC-14 RUN Domain Protein Binds to the Kinesin-1 and
UNC-16 Complex and Regulates Synaptic Vesicle Localization’, Molecular Biology of
the Cell 16(2), 483–496.
Sander J. D., Joung J. K. (2014), ‘CRISPR-Cas systems for editing, regulating and targeting
genomes’, Nature biotechnology 32(4), 347–355.
Schirmer E. C., Florens L., Guan T., Yates J. R., Gerace L. (2003), ‘Nuclear Membrane
Proteins with Subtractive Proteomics’, Science 301(5638), 1380–1382.
References 273
Schledzewski K., Brinkmann H., Mendel R. R. (1999), ‘Phylogenetic analysis of compo-
nents of the eukaryotic vesicle transport system reveals a common origin of adaptor
protein complexes 1, 2, and 3 and the F subcomplex of the coatomer COPI’, Journal of
Molecular Evolution 48(6), 770–778.
Scott M. G., Benmerah A., Muntaner O., Marullo S. (2002), ‘Recruitment of activated G
protein-coupled receptors to pre-existing clathrin-coated pits in living cells’, Journal
of Biological Chemistry 277(5), 3552–3559.
Seaman M. N., McCaffery J. M., Emr S. D. (1998), ‘A membrane coat complex essential for
endosome-to-Golgi retrograde transport in yeast’, Journal of Cell Biology 142(3), 665–
681.
Segev N. (1991), ‘Mediation of the attachment or fusion step in vesicular transport by
the GTP-binding Ypt1 protein’, Science 252(5012), 1553–1556.
Shehata M., Matsumura H., Okubo-Suzuki R., Ohkawa N., Inokuchi K. (2012), ‘Neuronal
Stimulation Induces Autophagy in Hippocampal Neurons That Is Involved in AMPA
Receptor Degradation after Chemical Long-Term Depression’, Journal of Neuroscience
32(30), 10413–10422.
Simmen T., Höning S., Icking A., Tikkanen R., Hunziker W. (2002), ‘AP-4 binds basolateral
signals and participates in basolateral sorting in epithelial MDCK cells.’, Nature cell
biology 4(2), 154–159.
Simonsen A. et al. (1998), ‘EEA1 links PI(3)K function to Rab5 regulation of endosome
fusion’, Nature 394(6692), 494–498.
Simpson F., Bright N. A., West M. A., Newman L. S., Darnell R. B., Robinson M. S. (1996),
‘A novel adaptor-related protein complex.’, The Journal of cell biology 133(4), 749–760.
Smith S. M., Baker M., Halebian M., Smith C. J. (2017), ‘Weak Molecular Interactions in
Clathrin-Mediated Endocytosis’, Frontiers in Molecular Biosciences 4, Article 72.
Soldati T., Schliwa M. (2006), ‘Powering membrane traffic in endocytosis and recycling’,
Nature Reviews Molecular Cell Biology 7(12), 897–908.
Söllner T., Bennett M. K., Whiteheart S. W., Scheller R. H., Rothman J. E. (1993a), ‘A
protein assembly-disassembly pathway in vitro that may correspond to sequential
steps of synaptic vesicle docking, activation, and fusion’, Cell 75(3), 409–418.
Söllner T., Whiteheart S. W., Brunner M., Erdjument-Bromage H., Geromanos S., Tempst
P., Rothman J. E. (1993b), ‘SNAP receptors implicated in vesicle targeting and fusion’,
Nature 362(6418), 318–324.
Sonnhammer E. L., von Heijne G., Krogh A. (1998), ‘A hidden Markov model for predicting
transmembrane helices in protein sequences.’, Proceedings of the Sixth International
Conference on Intelligent Systems for Molecular Biology 6, 175–182.
Søreng K., Munson M. J., Lamb C. A., Bjørndal G. T., Pankiv S., Carlsson S. R., Tooze S. A.,
Simonsen A. (2018), ‘SNX18 regulates ATG9A trafficking from recycling endosomes by
recruiting Dynamin-2’, EMBO reports 19(4), e44837.
274 References
Sou Y. S., Tanida I., Komatsu M., Ueno T., Kominami E. (2006), ‘Phosphatidylserine in
addition to phosphatidylethanolamine is an in vitro target of the mammalian Atg8
modifiers, LC3, GABARAP, and GATE-16’, Journal of Biological Chemistry 281(6), 3017–
3024.
Spudich G., Chibalina M. V., Au J. S. Y., Arden S. D., Buss F., Kendrick-Jones J. (2007),
‘Myosin VI targeting to clathrin-coated structures and dimerization is mediated by
binding to Disabled-2 and PtdIns(4,5)P2’, Nature Cell Biology 9(2), 176–183.
Stachowiak J. C., Brodsky F. M., Miller E. A. (2013), ‘A cost-benefit analysis of the physical
mechanisms of membrane curvature’, Nature Cell Biology 15(9), 1019–1027.
Stachowiak J. C. et al. (2012), ‘Membrane bending by protein-protein crowding’, Nature
Cell Biology 14(9), 944–949.
Stagg S. M., Gürkan C., Fowler D. M., LaPointe P., Foss T. R., Potter C. S., Carragher B.,
Balch W. E. (2006), ‘Structure of the Sec13/31 COPII coat cage’, Nature 439(7073), 234–
238.
Stavoe A. K. H., Hill S. E., Hall D. H., Colon-Ramos D. A. (2016), ‘KIF1A/UNC-104 Trans-
ports ATG-9 to Regulate Neurodevelopment and Autophagy at Synapses’, Develop-
mental Cell 38(2), 171–185.
Stephens D. J., Banting G. (1998), ‘Specificity of interaction between adaptor-complex
medium chains and the tyrosine-based sorting motifs of TGN38 and Igp120’, The
Biochemical journal 335, 567–572.
Südhof T. C., Rothman J. E. (2009), ‘Membrane Fusion : Grappling with SNARE and SM
proteins’, Science 474(5913), 474–478.
Suzuki K., Kubota Y., Sekito T., Ohsumi Y. (2007), ‘Hierarchy of Atg proteins in pre-
autophagosomal structure organization’, Genes to Cells 12(2), 209–218.
Takats S. et al. (2014), ‘Interaction of the HOPS complex with Syntaxin 17 mediates
autophagosome clearance in Drosophila’, Molecular Biology of the Cell 25(8), 1338–
1354.
Taylor M. J., Perrais D., Merrifield C. J. (2011), ‘A high precision survey of the molecular
dynamics of mammalian clathrin-mediated endocytosis’, PLoS Biology 9(3), e1000604.
Taylor R. S., Wu C. C., Hays L. G., Eng J. K., Yates J. R., Howell K. E. (2000), ‘Proteomics of
rat liver Golgi complex: Minor proteins are identified through sequential fractionation’,
Electrophoresis 21, 3441–3459.
Thompson A., Schäfer J., Kuhn K., Kienle S., Schwarz J., Schmidt G., Neumann T., Hamon
C. (2003), ‘Tandem mass tags: A novel quantification strategy for comparative analysis
of complex protein mixtures by MS/MS’, Analytical Chemistry 75(8), 1895–1904.
Tiller G. E., Mueller T. J., Dockter M. E., Struve W. G. (1984), ‘Hydrogenation of Triton
X-100 eliminates its fluorescence and ultraviolet light absorption while preserving its
detergent properties’, Analytical Biochemistry 141(1), 262–266.
References 275
Timms R. T. et al. (2016), ‘Genetic dissection of mammalian ERAD through comparative
haploid and CRISPR forward genetic screens’, Nature Communications 7, 11786.
Tisdale E. J., Bourne J. R., Khosravi-far R., Channing J. D., Balch W. E. (1992), ‘GTP-
Binding Mutants of Rabl and Rab2 are Potent Inhibitors of Vesicular Transport from
the Endoplasmic Reticulum to the Golgi Complex’, Journal of Cell Biology 119(4), 749–
761.
Toh W. H., Tan J. Z. A., Zulkefli K. L., Houghton F. J., Gleeson P. A. (2017), ‘Amyloid
precursor protein traffics from the Golgi directly to early endosomes in an Arl5b- and
AP4-dependent pathway’, Traffic 18(3), 159–175.
Toonen R. F., Verhage M. (2003), ‘Vesicle trafficking: Pleasure and pain from SM genes’,
Trends in Cell Biology 13(4), 177–186.
Traub L. M. (2009), ‘Tickets to ride: Selecting cargo for clathrin-regulated internalization’,
Nature Reviews Molecular Cell Biology 10(9), 583–596.
Trautz B. et al. (2017), ‘The host-cell restriction factor SERINC5 restricts HIV-1 infectivity
without altering the lipid composition and organization of viral particles’, Journal of
Biological Chemistry 292(33), 13702–13713.
Tsirigos K. D., Peters C., Shu N., Käll L., Elofsson A. (2015), ‘The TOPCONS web server
for consensus prediction of membrane protein topology and signal peptides’, Nucleic
Acids Research 43(Webserver issue), W401–W407.
Tsuboyama K., Koyama-Honda I., Sakamaki Y., Koike M., Morishita H., Mizushima N.
(2016), ‘The ATG conjugation systems are important for degradation of the inner
autophagosomal membrane’, Science 354(6315), 1036–1041.
Tsukada M., Ohsumi Y. (1993), ‘Isolation and characterization of autophagy-defective
mutants of Saccharomyces cerevisiae’, FEBS Letters 333(1.2), 169–174.
Tucker K. A., Reggiori F., Dunn W. A., Klionsky D. J. (2003), ‘Atg23 Is Essential for the
Cytoplasm to Vacuole Targeting Pathway and Efficient Autophagy but Not Pexophagy’,
Journal of Biological Chemistry 278(48), 48445–48452.
Tusher V. G., Tibshirani R., Chu G. (2001), ‘Significance analysis of microarrays applied
to the ionizing radiation response’, Proceedings of the National Academy of Sciences of
the United States of America 98(9), 5116–5121.
Tüysüz B. et al. (2014), ‘Autosomal recessive spastic tetraplegia caused by AP4M1 and
AP4B1 gene mutation: Expansion of the facial and neuroimaging features’, American
Journal of Medical Genetics, Part A 164(7), 1677–1685.
Tyanova S., Cox J. (2018), Perseus: A bioinformatics platform for integrative analysis of
proteomics data in cancer research, in L. von Stechow, ed., ‘Cancer Systems Biology.
Methods in Molecular Biology.’, Vol. 1711, Humana Press, New York, pp. 133–148.
Tyanova S., Temu T., Sinitcyn P., Carlson A., Hein M. Y., Geiger T., Mann M., Cox J. (2016),
‘The Perseus computational platform for comprehensive analysis of (prote)omics
data’, Nature Methods 13(9), 731–740.
276 References
Unanue E. F., Ungewickell E., Branton D. (1981), ‘The binding of clathrin triskelions to
membranes from coated vesicles.’, Cell 26(3 Pt 1), 439–446.
Ungewickell E., Branton D. (1981), ‘Assembly units of clathrin coats’, Nature 289, 420–
422.
Usami Y., Wu Y., Göttlinger H. G. (2015), ‘SERINC3 and SERINC5 restrict HIV-1 infectivity
and are counteracted by Nef’, Nature 526(7572), 218–223.
Van der Bliek A. M., Redelmeier T. E., Damke H., Tisdale E. J., Meyerowitz E. M., Schmid
S. L. (1993), ‘Mutations in human dynamin block an intermediate stage in coated
vesicle formation’, Journal of Cell Biology 122(3), 553–563.
Van Weering J. R. T. et al. (2012), ‘Molecular basis for SNX-BAR-mediated assembly of
distinct endosomal sorting tubules’, EMBO Journal 31(23), 4466–4483.
Verkerk A. J. M. H. et al. (2009), ‘Mutation in the AP4M1 gene provides a model for neuroax-
onal injury in cerebral palsy.’, The American Journal of Human Genetics 85(1), 40–52.
Vijayan V., Verstreken P. (2017), ‘Autophagy in the presynaptic compartment in health
and disease.’, The Journal of cell biology 216(7), 1–12.
Vizcaíno J. A. et al. (2014), ‘ProteomeXchange provides globally coordinated proteomics
data submission and dissemination’, Nature Biotechnology 32(3), 223–226.
Wang W. et al. (2003a), ‘Quantification of proteins and metabolites by mass spectrometry
without isotopic labeling or spiked standards’, Analytical Chemistry 75, 4818–4826.
Wang X., Kilimann M. W. (1997), ‘Identification of two new µ-adaptin-related proteins,
µ-ARP1 and µ-ARP2’, FEBS Letters 402(1), 57–61.
Wang Y. J. et al. (2003b), ‘Phosphatidylinositol 4 phosphate regulates targeting of clathrin
adaptor AP-1 complexes to the Golgi’, Cell 114(3), 299–310.
Waters M. G., Serafini T., Rothman J. E. (1991), ‘’Coatomer’: A cytosolic protein com-
plex containing subunits of non-clathrin-coated Golgi transport vesicles’, Nature
349(6306), 248–251.
Wen X., Klionsky D. J. (2016), ‘An overview of macroautophagy in yeast’, Journal of
molecular biology 428(9A), 1681–1699.
Williamson M. P. (2013), ‘Using chemical shift perturbation to characterise ligand bind-
ing’, Progress in Nuclear Magnetic Resonance Spectroscopy 73, 1–16.
Winslow A. R. et al. (2010), ‘α-Synuclein impairs macroautophagy: Implications for
Parkinson’s disease’, Journal of Cell Biology 190(6), 1023–1037.
Wood J. N. (1995), ‘Regulation of NF-kappa B activity in rat dorsal root ganglia and
PC12 cells by tumour necrosis factor and nerve growth factor.’, Neuroscience Letters
192(1), 41–44.
Xie Z., Nair U., Klionsky D. J. (2008), ‘Atg8 Controls Phagophore Expansion during Au-
tophagosome Formation’, Molecular Biology of the Cell 19, 3290–3298.
References 277
Xing Y., Böcking T., Wolf M., Grigorieff N., Kirchhausen T., Harrison S. C. (2010), ‘Struc-
ture of clathrin coat with bound Hsc70 and auxilin: Mechanism of Hsc70-facilitated
disassembly’, EMBO Journal 29(3), 655–665.
Xu L., Sowa M. E., Chen J., Li X., Gygi S. P., Harper J. W. (2008), ‘An FTS/Hook/p107FHIP
Complex Interacts with and Promotes Endosomal Clustering by the Homotypic Vac-
uolar Protein Sorting Complex’, Molecular Biology of the Cell 19, 5059–5071.
Yamada T., Carson A. R., Caniggia I., Umebayashi K., Yoshimori T., Nakabayashi K.,
Scherer S. W. (2005), ‘Endothelial nitric-oxide synthase antisense (NOS3AS) gene
encodes an autophagy-related protein (APG9-like2) highly expressed in trophoblast’,
Journal of Biological Chemistry 280(18), 18283–18290.
Yamaguchi J. et al. (2018), ‘Atg9a deficiency causes axon-specific lesions including neu-
ronal circuit dysgenesis’, Autophagy 14(5), 764–777.
Yamamoto A., Tagawa Y., Yoshimori T., Moriyama Y., Masaki R., Tashiro Y. (1998),
‘Bafilomycin A1 Prevents Maturation of Autophagic Vacuoles by Inhibiting Fusion
between Autophagosomes and Lysosomes in Rat Hepatoma Cell Line, H-4-II-E Cells.’,
Cell Structure and Function 23(1), 33–42.
Yamamoto H. et al. (2012), ‘Atg9 vesicles are an important membrane source during
early steps of autophagosome formation’, Journal of Cell Biology 198(2), 219–233.
Yap C. C., Murate M., Kishigami S., Muto Y., Kishida H., Hashikawa T., Yano R. (2003),
‘Adaptor protein complex-4 (AP-4) is expressed in the central nervous system neurons
and interacts with glutamate receptorδ2’, Molecular and Cellular Neuroscience 24, 283–
295.
Yoshida H., Kitagishi Y., Okumura N., Murakami M., Nishimura Y., Matsuda S. (2011),
‘How do you RUN on?’, FEBS Letters 585(12), 1707–1710.
Yoshihisa T., Barlowe C., Schekman R. (1993), ‘Requirement for a GTPase-activating
protein in vesicle budding from the endoplasmic reticulum’, Science 259(5100), 1466–
1468.
Young A. R. J. et al. (2006), ‘Starvation and ULK1-dependent cycling of mammalian Atg9
between the TGN and endosomes’, Journal of cell science 119(18), 3888–3900.
Yu C., Feng W., Wei Z., Miyanoiri Y., Wen W., Zhao Y., Zhang M. (2009), ‘Myosin VI
Undergoes Cargo-Mediated Dimerization’, Cell 138(3), 537–548.
Yu L., Chen Y., Tooze S. A. (2018), ‘Autophagy pathway: Cellular and molecular mecha-
nisms’, Autophagy 14(2), 207–215.
Zhang Y., Fonslow B. R., Shan B., Baek M.-c., Yates J. R. (2013), ‘Protein Analysis by
Shotgun/Bottom-up Proteomics’, Chemical Reviews 113, 2343–2394.
Zhou C. et al. (2017), ‘Regulation of mATG9 trafficking by Src- and ULK1-mediated
phosphorylation in basal and starvation-induced autophagy’, Cell Research 27(2), 184–
201.

Appendix A
Constructs
The DNA constructs created and used in the course of this project are listed in Table A.1.
280 Constructs
Tab
le
A
.1
D
N
A
co
n
stru
cts
u
sed
in
th
is
stu
d
y.
C
o
n
stru
ct
In
sert
Vecto
r
b
ackb
o
n
e
So
u
rce
p
LX
IN
_A
P
4E
1_F
K
B
P
A
P
4E
1_F
K
B
P[in
h
in
ge]
p
LX
IN
m
o
d
G
.B
o
rn
er
(D
avies
et
al.,2018)
p
E
G
F
P-B
irA
*
M
yc-B
irA
*
p
E
G
F
P-N
2
A
.D
avies
(D
avies
et
al.,2018)
p
LX
IN
m
o
d
_m
yc-B
irA
*
(C
-term
taggin
g
co
n
stru
ct)
M
yc-B
irA
*
p
LX
IN
m
o
d
A
.D
avies
(D
avies
et
al.,2018)
p
LX
IN
m
o
d
_A
P
4M
1-B
irA
*
A
P
4M
1
p
LX
IN
m
o
d
_m
yc-B
irA
*
(C
-term
taggin
g
co
n
stru
ct)
A
.D
avies
(D
avies
et
al.,2018)
p
LX
IN
m
o
d
_A
P
4S1-B
irA
*
A
P
4S1
p
LX
IN
m
o
d
_m
yc-B
irA
*
(C
-term
taggin
g
co
n
stru
ct)
A
.D
avies
(D
avies
et
al.,2018)
p
LX
IN
m
o
d
_A
P
4B
1-B
irA
*
A
P
4B
1
p
LX
IN
m
o
d
_m
yc-B
irA
*
(C
-term
taggin
g
co
n
stru
ct)
A
.D
avies
(D
avies
et
al.,2018)
p
LX
IN
m
o
d
_A
P
4E
1-B
irA
*
[in
h
in
ge]
A
P
4E
1-B
irA
*
(tag
in
h
in
ge)
p
LX
IN
m
o
d
A
.D
avies
(D
avies
et
al.,2018)
p
LX
IN
m
o
d
_SE
R
IN
C
1-B
irA
*
SE
R
IN
C
1
p
LX
IN
m
o
d
_m
yc-B
irA
*
(C
-term
taggin
g
co
n
stru
ct)
A
.D
avies
(u
n
p
u
b
lish
ed
)
p
LX
IN
m
o
d
_SE
R
IN
C
3-B
irA
*
SE
R
IN
C
3
p
LX
IN
m
o
d
_m
yc-B
irA
*
(C
-term
taggin
g
co
n
stru
ct)
A
.D
avies
(u
n
p
u
b
lish
ed
)
p
IR
E
SN
eo
2_SE
R
IN
C
3-H
A
-
m
C
h
erry
SE
R
IN
C
3-H
A
-m
C
h
erry
p
IR
E
SN
eo
2
G
en
ecu
st(cu
sto
m
;D
avies
etal.,
2018)
p
LX
IN
m
o
d
_SE
R
IN
C
3_H
A
[extlo
o
p]
SE
R
IN
C
3_H
A
[extlo
o
p]
p
LX
IN
m
o
d
A
.D
avies
(D
avies
et
al.,2018)
p
LX
IN
m
o
d
_A
P
4B
1
A
P
4B
1
(W
T
)
p
LX
IN
m
o
d
A
.D
avies
(Frazier
et
al.,2016)
p
LX
IN
m
o
d
_E
arless_A
P
4B
1
E
arless
A
P
4B
1
p
LX
IN
m
o
d
A
.D
avies
(Frazier
et
al.,2016)
p
LX
IN
m
o
d
_A
P
4B
1_m
u
t
[Y
682V
]
M
u
tan
t
A
P
4B
1
[Y
682V
]
p
LX
IN
m
o
d
_A
P
4B
1
M
.Frazier
(Frazier
et
al.,2016)
p
LX
IN
m
o
d
_A
P
4B
1_m
u
t
[I668A
;A
670S]
M
u
tan
t
A
P
4B
1
[I669A
;A
670S]
p
LX
IN
m
o
d
_A
P
4B
1
M
.Frazier
(Frazier
et
al.,2016)
Appendix B
PCR primers for cloning
All primers used in the generation of the constructs used in this study are listed below.
pEGFP-myc-BirA* (cloning sites BsrGI and XbaI)
Amplification of myc-BirA* from pcDNA3.1_mycBioID:
F: CGGTAGCTGTACAAGATGGAACAAAAACTCATCT (contains BsrGI site)
R: GGTCGTTCTAGATCAGCGGTTTAAACTTAAG (contains XbaI site)
C-terminal myc-BirA* tagging construct (pLXINmod)
Amplification of myc-BirA* from pcDNA3.1_mycBioID:
F: TCTAGGCGCCGGAATTCGTTAGATCTGGCAGCGGCGGCAGCGGCAGCGGCATGGAA-
CAAAAACTCATCTCAG (contains GS linker)
R: GATCCCTCGAGGTCGACGTTTTACTTCTCTGCGCTTCTCAG
AP4E1 BioID (pLXINmod_AP4E1_myc-BirA*)
Amplification of myc-BirA* from pcDNA3.1_mycBioID:
F: CCTGTACACCATGGAACAAAAACTCATCTC
R: GCAGAGCTCCGCGGTTTAAACTTAAGCTTG
Amplification of N-terminal part of AP4E1 (and linker) from pLXINmod_AP4E1_FKBP:
F: TCTAGGCGCCGGAATTCGTTGCGGCGATGAGCGACATA
R: TTTGTTCCATGGTGTACAGGTTCTTGGCGG
Amplification of C-terminal part of AP4E1 from pLXINmod_AP4E1_FKBP:
F: TTTAAACCGCGGAGCTCTGCCTGTTCCTC
R: GATCCCTCGAGGTCGACGTTCTAGGATCCCTCCATCACC
AP4B1 BioID (pLXINmod_AP4B1-myc-BirA*)
Amplification of AP4B1 from Image clone 2906087:
282 PCR primers for cloning
F: CTAGGCGCCGGAATTCGTTAGCCACCATGCCGTACCTTGGCTCC
R: GCTGCCGCCGCTGCCAGATCCTGATTTTATTTCTTCAATTGTTCCAATC
AP4M1 BioID (pLXINmod_AP4M1-myc-BirA*)
Amplification of AP4M1 from a sequence verified EST clone:
F: CTAGGCGCCGGAATTCGTTAGCCACCATGATTTCCCAATTCTTCATTCTG
R: GCTGCCGCCGCTGCCAGATCCGATCCGAATGACATAGGCG
AP4S1 BioID (pLXINmod_AP4S1-myc-BirA*)
Amplification of AP4S1 from a sequence verified EST clone:
F: CTAGGCGCCGGAATTCGTTACATAACTTTTGAACTGTATTTGG
R: GCTGCCGCCGCTGCCAGATCCGCTTTCTGACATCTTATCAAG
SERINC1 BioID (pLXINmod_SERINC1-myc-BirA*)
Amplification of SERINC1 from pGEM-T_SERINC1:
F: CTAGGCGCCGGAATTCGTTAGCCACCATGGGGAGCGTCCTGGGG
R: TGCCGCTGCCGCCGCTGCCAGATCCGTCAAAATCACGATTTGTAAGAACAAGTG
SERINC3 BioID (pLXINmod_SERINC3-myc-BirA*)
Amplification of SERINC3 from pIRESNeo2_SERINC3-HA-mCherry:
F: CTAGGCGCCGGAATTCGTTAATAGGATCCGCCACCATG
R: TGCCGCTGCCGCCGCTGCCAGATCCGCTGAAGTCCCGACTGGTG
pLXINmod_AP4B1 (cloning sites SalI and NotI)
Amplification of WT AP4B1 from Image clone 2906087: F: ATTACTGTCGACGCCACCAT-
GCCGTACCTTGGCTCC (contains SalI site)
R: GGTCGTGCGGCCGCTTATGATTTTATTTCTTCAATTGTTCCAATC (contains NotI site)
pLXINmod_Earless_AP4B1 (cloning sites SalI and NotI)
Amplification of Earless AP4B1 (residues 1–612) from Image clone 2906087:
F: ATTACTGTCGACGCCACCATGCCGTACCTTGGCTCC (contains SalI site)
R: GGTCGTGCGGCCGCTTAGAGTTCTTGTACCCTCTCCTTGTTC (contains NotI site)
WT/mutant TEPSIN-GFP constructs in pLXINmod
Amplification of WT/mutant [L470S/F471S]TEPSIN-GFP from pTEPSIN-GFP WT/mutant:
F: TCTAGGCGCCGGAATTCGTTTCGAGGCACGAGGCTGGATC
R: GATCCCTCGAGGTCGACGTTTTACTTGTACAGCTCGTCCATGCCG
pLXINmod_SERINC3_HA[extloop]
Amplification of N-terminal part of SERINC3 from pIRESNeo2_SERINC3-HA-mCherry:
F: TCTAGGCGCCGGAATTCGTTACCGGTGCCACCATGGGGGCTGTG (contains AgeI site)
283
R: GGAACATCGTATGGGTACAGGGTTGGTGCAGTTATGCG
Amplification of C-terminal part of SERINC3 from pIRESNeo2_SERINC3-HA-mCherry:
F: ACCAACCCTGTACCCATACGATGTTCCAGATTACGCTGCTCCTGGAAATTCAACTGC
(contains HA tag)
R: GATCCCTCGAGGTCGACGTTTTATTAGCTGAAGTCCCGACTGGTG
pQCXIH_GFP-RUSC2
Amplification of GFP from pEGFP-N2:
F: GGAATTGATCCGCGGCCGCAGCCACCATGGTGAGCAAG
R: ACCGGTGCCAGATCCCTTGTACAGCTCGTCCATGC
Amplification of N-terminal part of RUSC2 from pCMV-SPORT6_RUSC2:
F: GCTGTACAAGGGATCTGGCACCGGTGGCAGCGGCAGCGGCATGGATAGTCCCCCAAAGC
R: CCTGTTCCATGGAGCCGGTACTCAGTCAAG
Amplification of C-terminal part of RUSC2 from pCMV-SPORT6_RUSC2:
F: TACCGGCTCCATGGAACAGGAAGCTTGC
R: AATTAAGCGTACGAGGCCTATCAGTTTTGGCTGCTTCC
pQCXIH_RUSC2-GFP
Amplification of GFP from pEGFP-N2:
F: CAGCCAAAACGGATCTGGCACCGGTGGCAGCGGCAGCGGCATGGTGAGCAAGGGC-
GAG
R: CGTACGAGGCCTACCGGTGCTTACTTGTACAGCTCGTCCATG
Amplification of N-terminal part of RUSC2 from pCMV-SPORT6_RUSC2:
F: CGCTGCAGGAATTGATCCGCGGCCGCCACCATGGATAGTCCCCCAAAGC
R: CCTGTTCCATGGAGCCGGTACTCAGTCAAG
Amplification of C-terminal part of RUSC2 from pCMV-SPORT6_RUSC2:
F: TACCGGCTCCATGGAACAGGAAGCTTGC
R: ACCGGTGCCAGATCCGTTTTGGCTGCTTCCAGG
pDonor_myc-Clover_SERINC1 (SERINC1 endogenous tagging donor)
Amplification of 5′ homology region of SERINC1 from HeLa genomic DNA:
F: CTCCCCGGGCGCGACTAGTGAATTCGAGTGCAGTGGCTTGATC
R: GAGTTTTTGTTCAGAACCCGATCGGTCAAAATCACGATTTGTAAG
Amplification of 3′ homology region of SERINC1 from HeLa genomic DNA:
F: CATTATACGAAGTTATTTAATTAACAAGTGCATTGATATGTGAAGTAG
R: ATAATCAGCATCATGATGTGGTACCCTGCAACCTCTGCCTCCT
pDonor_myc-Clover_SERINC3 (SERINC3 endogenous tagging donor)
Amplification of 5′ homology region of SERINC3 from HeLa genomic DNA:
284 PCR primers for cloning
F: CTCCCCGGGCGCGACTAGTGAATTCTCCACCTCATGCTCTGCTT
R: GAGTTTTTGTTCAGAACCCGATCGGCTGAAGTCCCGACTGGTG
Amplification of 3′ homology region of SERINC3 from HeLa genomic DNA:
F: CATTATACGAAGTTATTTAATTAAGTGAATGCTTTGCAAGTTTG
R: ATAATCAGCATCATGATGTGGTACCGCATCCTGACTAGAAGCG
Appendix C
CRISPR guide sequences
Guide sequences used for knockout of AP4B1 and AP4E1 are listed in Table C.1 and C.2,
respectively.
286 CRISPR guide sequences
Tab
le
C
.1
C
R
ISP
R
gu
id
es
fo
r
kn
o
cko
u
to
fA
P
4B
1.
Paired
gu
id
es
u
sed
in
th
e
d
o
u
b
le
n
ickase
C
R
ISP
R
/C
as9
ap
p
ro
ach
to
kn
o
cko
u
tA
P
4B
1
in
H
eL
a
cells.T
h
e
n
u
m
b
er
fo
llow
in
g
‘x’in
th
e
gu
id
e
n
am
e
refers
to
th
e
exo
n
targeted
.
N
o
n
e
o
fth
e
gu
id
e
p
airs
h
ad
an
y
co
m
b
in
ed
o
ff-target
sites.G
u
id
es
x1_A
,x2_A
an
d
x2_B
w
ere
also
u
sed
w
ith
w
ild
-typ
e
C
as9
to
d
ep
lete
A
P
4B
1
in
SH
-SY
5Y
cells.
G
u
id
e
Seq
u
en
ce
Q
u
ality
sco
re
O
ff-targetsites
to
tal
(in
gen
e)
P
air
sco
re
A
P
4B
1_x1_A
A
C
G
T
C
C
T
C
G
G
A
G
C
C
A
A
G
G
TA
84
67
(13)
63
A
P
4B
1_x1_B
AT
C
C
T
C
A
C
AT
T
C
A
A
G
C
T
G
AT
74
190
(26)
A
P
4B
1_x2_A
G
A
C
C
C
C
A
AT
C
C
A
AT
G
G
T
G
C
G
95
37
(5)
85
A
P
4B
1_x2_B
T
G
C
A
C
A
G
C
G
TAT
T
G
AT
G
G
C
C
90
58
(8)
A
P
4B
1_x3_A
T
C
T
C
A
AT
G
G
T
C
T
G
C
G
G
G
ATA
90
84
(7)
68
A
P
4B
1_x3_B
T
G
TATATA
C
T
C
C
T
G
C
A
C
A
C
C
75
140
(10)
287
Ta
b
le
C
.2
C
R
IS
P
R
gu
id
es
fo
r
kn
o
ck
o
u
to
fA
P
4E
1.
Pa
ir
ed
gu
id
es
u
se
d
in
th
e
d
o
u
b
le
n
ic
ka
se
C
R
IS
P
R
/C
as
9
ap
p
ro
ac
h
to
kn
o
ck
o
u
tA
P
4E
1
in
H
eL
a
ce
lls
.
T
h
e
n
u
m
b
er
fo
llo
w
in
g
‘x
’i
n
th
e
gu
id
e
n
am
e
re
fe
rs
to
th
e
ex
o
n
ta
rg
et
ed
.
N
o
n
e
o
ft
h
e
gu
id
e
p
ai
rs
h
ad
an
y
co
m
b
in
ed
o
ff
-t
ar
ge
t
si
te
s.
G
u
id
es
x6
_A
,x
6_
B
an
d
x7
_A
w
er
e
al
so
u
se
d
w
it
h
w
ild
-t
yp
e
C
as
9
to
d
ep
le
te
A
P
4E
1
in
SH
-S
Y
5Y
ce
lls
.
G
u
id
e
Se
q
u
en
ce
Q
u
al
it
y
sc
o
re
O
ff
-t
ar
ge
ts
it
es
to
ta
l
(i
n
ge
n
e)
P
ai
r
sc
o
re
A
P
4E
1_
x6
_A
C
T
T
G
AT
TA
G
G
A
G
C
A
AT
G
A
G
A
60
25
6
(1
1)
38
A
P
4E
1_
x6
_B
G
C
A
C
T
T
T
G
T
G
A
C
A
G
A
G
AT
G
T
64
23
5
(1
6)
A
P
4E
1_
x7
_A
C
T
G
T
G
G
TA
AT
T
G
A
AT
T
C
TA
C
74
14
4
(8
)
47
A
P
4E
1_
x7
_B
A
AT
T
C
A
G
C
T
C
T
T
G
A
G
A
AT
A
C
64
27
3
(1
9)
A
P
4E
1_
x1
1_
A
G
TA
A
AT
AT
T
C
A
A
G
C
AT
T
T
T
C
41
50
0
(2
4)
36
A
P
4E
1_
x1
1_
B
A
G
A
G
TA
T
G
T
C
AT
C
G
T
C
A
AT
T
87
11
3
(9
)