Molecular basis of cargo sorting by endosomal trafficking machinery Jonathan Geoffrey Griffiths Kaufman Department of Clinical Biochemistry University of Cambridge This dissertation is submitted for the degree of Doctor of Philosophy King’s College April 2024 Declaration This thesis is the result of my own work and includes nothing which is the outcome of work done in collaboration except as declared in the preface and specified in the text. It is not substantially the same as any work that has already been submitted, or, is being concurrently submitted, for any degree, diploma or other qualification at the University of Cambridge or any other University or similar institution except as declared in the preface and specified in the text. It does not exceed the prescribed word limit for the relevant Degree Committee. Jonathan Geoffrey Griffiths Kaufman April 2024 I would like to dedicate this thesis to my now deceased cat. Thank you for all the support over the years, support for my research (whiskers for seeding) and for sleeping on my clean laundry covering it with your hair. Acknowledgements I would like to acknowledge David Owen, for the investment of his time into helping me and our research. Thank you for the mentorship for the past four years, support through challenging samples and a pandemic and finally enthralling discussion of science and all things endosomal. Furthermore, thank you for trusting me to go off and try my own weird and wacky experiments once in a while, it’s been a blast! I can’t wait to get back to the lab. I would like to thank both past and present members of the Owen lab for their input on my projects. Primarily Zuzana Kadlecova for her formative discussion and guidance, as well as the occasional afterhours glasses of wine. Natalya Leneva for having the patience to try and teach me both BASH and retromer nomenclature. Nathan Zaccai for X-ray crystallography and liposome SPR training. Antoni Wrobel and Veronica Kane-Dickson for support with Cryo- EM. Bernie Kelly for giving honest feedback on some of my more “creative” ideas and teaching me to work with lipids. Sally Gray for her help with cloning and Y2H data included in this thesis. From CIMR Level 5 I would like to particularly thank Scottie Robinson and David Gershlick for helpful discussion of all things from evolution of membrane trafficking to bizarre preprints. Folma Buss for pastoral support and the Buss lab for coffee breaks. I would like to thank Paul Luzio for his time and readiness to discuss any and all aspects of membrane trafficking, the advice and guidance over the past several years. Further thanks (in retrospect) for press ganging me into giving a Monday morning seminar which resulting in me reanalysing my data and the discovering the water network coordinating the dileucine motif. I am also thankful to the wider CIMR community for making it such a fantastic place to do research. In particular; Jess Eden, Maria Pereira, Danielle Stalder, Alex Dagg, Alex Holmes Ioana Sava and Rory Clayton for extended discussion over the years. As well as Adam Rochussen and Martin Limback-Stokin for organising Friday morning journal clubs. For help with structural biology, I would like to thank Airlie McCoy and Phil Evans for their words of wisdom and insight into some of my more difficult crystallographic datasets. I would like to acknowledge the Cryo-EM facility in the department of Biochemistry for screening and data acquisition. Thanks for all the help, Dima, Steve and Lee. I am grateful to Sean Munro for providing both the opportunity to collaborate on such an exciting project and the invaluable discussions. I would like to thank John Briggs for his generosity in screening time and insight into processing of both single particle and tomographic datasets. Furthermore, thank you to the entire Briggs department for being so welcoming during my visits. Becca Taylor for sample prep from nanodiscs to GUVs and help with the microscopes. Both Zunlong Ke and James Stacey for opening my mind to being able to single particle everything. Hui Guo for help with my single particle data and helical reconstruction. Grigory Tagiltsev for his willingness to explain tomography in absolute detail to a complete beginner. I would further like to thank my parents Jim and Gillian for whom there is much to be grateful for. In particular the support in getting me here and continuing to be my scientific inspiration. I couldn’t have done it without you. Finally, I would like to thank Tiff Lai for everything, the coffee and being with me the entire way. Abstract Faithful sorting of transmembrane proteins to their intended steady state localization is crucial for function of cellular organelles and homeostasis of an organism. Yet the vast majority of transmembrane proteins are synthesised in the endoplasmic reticulum (ER) despite the site of their function being within any number of a variety of other distinct organelles. Thus, within each cell there exists mechanism for the “un-mixing” of this otherwise gallimaufry of transmembrane proteins to their intended final destination. The early endosomal system serves as a checkpoint within the cell marking a critical junction between the Golgi apparatus, plasma membrane and endo-lysosomal pathway. It is at these early endosomal organelles that the sorting of a transmembrane protein determines if it is recycled for further use or degraded. Pathophysiological conditions from pathogens, rare genetic disorders and neurodegenerative conditions have all been causally linked to sorting and the endosomal system. Thus, understanding the mechanisms that drive sorting with molecular detail remains of vital importance not only to cellular function but also medicine. This study explores endosomal sorting at a molecular resolution using a combination of in vitro biochemistry and structural biology to understand the mechanisms of sorting of cargoes into and out of the endosomal system. The mechanism by which tethering of endosome-to-Golgi derived vesicles occurs remains incompletely understood. Here I characterize the binding of the tethering component, TBC1D23, directly to cargoes containing a novel acidic-TLY motif, suggesting the role of this tether in the sorting of carriers. While the sorting of acidic dileucine motifs is thought to be well understood, it remains unclear how different cargoes obtained different end point localisations despite containing similar motifs. Here I resolve the interaction of a dileucine motif bound to a fragment of the AP-2 complex at previously unobtained resolution, allowing visualization of not only the dileucine motif but also a previously unresolved water network mediating the reaction. This water network structurally explains the preferential endocytosis of a subset of cargo. Furthermore, the same water network is coordinated differently by the AP-3 complex to increase its affinity to the same subset of cargo, proposing a new mechanism of cargo sorting. Structural studies in understanding the AP-3 complex reveal a surprising lack of regulation in solution. Instead, I suggest the unfunctionalized δ-ear domain may serve to regulate the core through preventing recruitment to membrane by simultaneously breaking interaction sites for phosphoinositide lipids, Arf1 and dileucine cargo. Finally, I show that in vitro reconstitution of the AP-3 complex on synthetic membranes with PI35P2, Arf1 and cargo motifs is sufficient to drive assembly of highly ordered protein coated tubules, which may represent tubular carriers in vivo. Contents Contents .................................................................................................................................... ix List of Figures ........................................................................................................................ xiii List of Tables ......................................................................................................................... xvii Nomenclature .......................................................................................................................... xix Chapter 1 Introduction ............................................................................................................ 2 1.1 Cellular organelles at the ultrastructural level ........................................................... 2 1.2 The Endocytic system ................................................................................................ 7 1.3 Endosomes: An overview .......................................................................................... 9 1.3.1 Architecture, identity and function in endocytic system ................................. 11 1.3.2 Entry into the endocytic system ....................................................................... 11 1.4 Early endosomal formation from endocytic CCVs .................................................. 12 1.5 Maturation into later endosome indicated by Rab markers ..................................... 14 1.6 PIP conversion in endosomal maturation ................................................................ 15 1.7 Endosomal pH .......................................................................................................... 15 1.8 Acquisition of ILVs and fusion with lysosomes ...................................................... 17 1.9 Degradation .............................................................................................................. 17 1.10 Trafficking within the endosomal system ................................................................ 18 1.10.1 Recruitment of membrane scaffolds and cargo adaptors ................................. 18 1.10.2 Cargo selection: Mechanisms of molecular sorting ......................................... 21 1.10.3 Sorting motifs .................................................................................................. 22 1.10.4 Tyrosine-based sorting motifs .......................................................................... 22 1.10.5 Alternative mechanisms of sorting .................................................................. 23 1.11 Membrane remodelling and sculpting ..................................................................... 24 1.12 Carrier scission ......................................................................................................... 27 1.13 Uncoating ................................................................................................................. 28 1.14 Tethering and fusion ................................................................................................ 28 1.15 An overview of AP complexes and their related assemblies ................................... 30 1.15.1 AP-2: A model for AP family members .......................................................... 32 1.15.2 The AP-1 complex: Endosome to Golgi trafficking ........................................ 37 1.15.3 The AP-3 complex: Endosome to late endosomes, lysosomes and LROs ...... 39 Aims of this thesis .................................................................................................................... 44 Chapter 2 Materials and Methods ......................................................................................... 47 2.1 Molecular Biology ................................................................................................... 47 2.1.1 PCR .................................................................................................................. 47 2.1.2 Gibson assembly .............................................................................................. 47 2.1.3 Transformation ................................................................................................. 47 2.2 Bacterial expression and culture .............................................................................. 48 2.3 Bacterial protein purification ................................................................................... 48 2.3.1 Glutathione-S-transferase tagged purification ................................................. 48 2.3.2 Histidine tagged purification ............................................................................ 49 2.4 Insect expression and culture ................................................................................... 49 2.5 Insect protein purification (AP-3 core) .................................................................... 49 2.6 Sodium Dodecyl sulphate polyacrylamide gel electrophoresis (SDS PAGE) ......... 50 2.7 Isothermal Titration Calorimetry (ITC) ................................................................... 50 2.8 Liposome synthesis .................................................................................................. 51 2.8.1 Lipids stocks as solvent ................................................................................... 51 2.8.2 Lipids stocks from powders ............................................................................. 51 2.9 Unilamellar Liposome ............................................................................................. 51 2.10 Multilamellar liposome spin down .......................................................................... 52 2.11 Liposome Surface plasmon resonance (SPR) .......................................................... 52 2.12 Giant unilamellar vesicles (GUVs) .......................................................................... 53 2.13 Lipid Nanodisc reconstitution .................................................................................. 53 2.14 GST pulldowns ........................................................................................................ 54 2.15 Size exclusion chromatography with Multi angle light scattering (SEC-MALS) ... 54 2.16 Single particle Cryo-EM .......................................................................................... 55 2.16.1 Cryo-EM sample preparation ........................................................................... 55 2.16.2 Cryo-EM data collection .................................................................................. 56 2.16.3 Cryo-EM data processing ................................................................................ 57 2.16.4 Helical reconstruction ...................................................................................... 58 2.17 X-ray crystallography .............................................................................................. 59 2.17.1 Protein crystallization ...................................................................................... 59 2.17.2 Crystal mounting and cryoprotection ............................................................... 60 2.17.3 X-ray diffraction data collection ...................................................................... 60 2.17.4 X-ray crystallography data processing ............................................................. 60 2.18 Alphafold2 structural predictions ............................................................................ 61 Chapter 3 TBC1D23: A cargo sorting tether ........................................................................ 62 3.1 Introduction .............................................................................................................. 62 3.2 Characterizing the binding of STX16 and CPD to TBC1D23 ................................. 65 3.3 Crystallization of TBC1D23:STX16 complex ......................................................... 67 3.4 Validation of the TBC1D23:STX16 binding interface ............................................ 77 3.5 Validation of new TLY motif cargo ........................................................................ 81 3.6 TBC1D23 Discussion .............................................................................................. 84 Chapter 4 Differential Dileucine motif sorting ..................................................................... 89 4.1 Introduction .............................................................................................................. 89 4.2 Construction of a stable AP-2 σ2α hemi complex .................................................. 91 4.3 Binding of an array of dileucine sequences to S2A ................................................. 93 4.4 Crystallization of S2A in complex with dileucine motifs ........................................ 97 4.5 A closer examination of S2A:MFSD12 and the ordered water network ............... 101 4.6 Characterizing the X residues of MFSD12 ............................................................ 105 4.7 Characterizing dileucines cargo binding to AP-3 .................................................. 107 4.8 Expression and purification of S3D ....................................................................... 107 4.9 Dileucines have higher affinity for S3D compared to S2A ................................... 108 4.10 Structural basis of dileucine binding to S3D ......................................................... 109 4.11 Characterizing the water network of S3D .............................................................. 114 4.12 Dileucine discussion .............................................................................................. 117 Chapter 5 Structural studies of AP-3 appendages .............................................................. 120 5.1 Introduction ............................................................................................................ 120 5.2 X-ray crystallography of the b3-ear domain .......................................................... 121 5.3 Structural comparison of b3-ear domains .............................................................. 124 5.4 The δ-appendage and possible mechanism of autoinhibition ................................ 125 5.5 Purification of the δ-ear ......................................................................................... 125 5.6 X-ray crystallography of the δ-ear ......................................................................... 127 5.7 Interaction of the δ-ear with the AP-3 core ............................................................ 130 5.8 Structural prediction of the σ3:δ-ear interaction .................................................... 131 5.9 Characterization of the σ3:δ-ear interface ............................................................. 133 Chapter 6 The AP-3 core in solution .................................................................................. 136 6.1 Introduction ............................................................................................................ 136 6.2 Purification of the AP-3 complex .......................................................................... 137 6.3 Optimization of AP-3 core vitrification ................................................................. 138 6.4 Single particle Cryo-EM of the AP-3 core in solution .......................................... 141 6.5 Why is AP-3 open in solution? .............................................................................. 147 6.6 Attempts to fix orientation bias .............................................................................. 149 6.6.1 2D rebalancing ............................................................................................... 149 6.6.2 Tilted collection ............................................................................................. 151 6.6.3 CryoEF ........................................................................................................... 152 6.6.4 Tilted data processing .................................................................................... 152 Chapter 7 Determinants of AP-3 recruitment ..................................................................... 154 7.1 How is the AP-3 complex recruited to the endosome? .......................................... 154 7.2 AP-3 and Phosphoinositides .................................................................................. 155 7.3 Structural basis of Arf1 binding to AP-3 d ............................................................ 158 Chapter 8 The AP-3 complex on membranes ..................................................................... 165 8.1 The AP-3 complex and lipid nanodiscs ................................................................. 165 8.2 The AP-3 complex and Arf1-linked-nanodiscs ..................................................... 167 8.3 AP-3 reconstitution on membranes ........................................................................ 174 8.4 Optimizing AP-3 liposome coating ....................................................................... 177 8.5 Imaging of AP-3 on spherical liposomes. .............................................................. 180 8.6 Occasional tubule ................................................................................................... 180 8.7 Optimization of tubulation ..................................................................................... 182 8.8 Single particle analysis of AP-3 tubules ................................................................ 184 8.9 Cryo-ET and subtomogram averaging of AP-3 tubules ........................................ 188 8.10 Summary ................................................................................................................ 192 Chapter 9 Discussion .......................................................................................................... 196 9.1 Overview ................................................................................................................ 196 9.2 Concluding remarks and outstanding questions .................................................... 197 References .............................................................................................................................. A-1 List of Figures 1.1 Electron microscopy of cellar membranes 2 1.2 Directionality of membrane trafficking 5 1.3 Maturation of the endocytic pathway 8 1.4 Diversity of mechanisms of GTPase recruitment 19 1.5 Coats of all shapes and sizes 25 1.6 Evolution of the human AP family 30 1.7 Early structural understanding of AP2 and clathrin 32 1.8 Domain architecture and conformational change of the AP-2 complex 34 2.1 Helical symmetry 58 3.1 Purification of TBC1D23 PH domain 66 3.2 Binding of TBC1D23 to TLY motifs 67 3.3 Initial TBC1D23 crystallization and diffraction 68 3.4 Low resolution solution of TBC1D23:STX16 complex 70 3.5 SEC-MALS of TBC1D23 PH domain and truncated 71 3.6 Crystallization and diffraction of truncated TBC1D23 PH domain 73 3.7 Novel STX16 peptide density 74 3.8 Structure of the TBC1D23:STX16 complex 75 3.9 Characterization of the TBC1D23 TLY motif binding site 77 3.10 Characterization of the STX16 TLY motif 78 3.11 Characterization the stringency of STX16 TLY motif 80 3.12 Validating the binding of novel TLY 81 3.13 Schematic diagram of the role of TBC1D23 in sorting endosomal derived carrier 88 4.1 Schematic diagram for the sorting of dileucine 90 4.2 Purification of S2A 92 4.3 Characterizing differential affinities of dileucine motifs 95 4.4 Structural comparison of S2A dileucine binding 98 4.5 High resolution structure of S2A:MFSD12 complex 102 4.6 Mapping the dileucine water network of S2A 104 4.7 Characterizing the X residues of MFSD12 106 4.8 Purification of S3D 108 4.9 Binding of MFSD12 dileucine to S3D 109 4.10 Crystallization and structural solution of S3D:MFSD12 complex 111 4.11 Mapping the dileucine water network of S3D 112 4.12 Structural comparison of the binding of dileucine motifs by S2A and S3D 113 4.13 Binding of S3D T70G to MFSD12 dileucine motif 114 4.14 Dileucine and water environment across AP complexes 116 4.15 Schematic diagram for an affinity sorting model of dileucine cargo. 118 5.1 Purification and X-ray crystallography of the β3 appendage 122 5.2 Comparison and structure of the β3 appendage 123 5.3 AP-3 β3 ear does not bind similar motifs to AP-2 β2 ear 124 5.4 Purification and crystallography of δ appendage 126 5.5 Structure of the δ appendage 128 5.6 Binding of the δ appendage to σ3 130 5.7 Alphafold2 prediction of the δ-ear:σ3 interaction 132 5.8 Validation of the predicted σ3:δ-ear interaction 133 5.9 Validation of the predicted σ3 interaction with δ-ear interaction 134 6.1 Purification of the AP-3 core 137 6.2 Optimization of AP-3 vitrification 140 6.3 Single particle cryo-EM of the AP-3 142 6.4 Low resolution docking of the AP-3 core 144 6.5 Insights from low resolution docking of the AP-3 core 145 6.6 AP-3 can bind YXXΦ motifs in solution 146 6.7 Comparison of the AP conformations 148 6.8 2D rebalancing of AP-3 150 7.1 Multilamellar phosphoinositide liposome spin downs 156 7.2 Liposome based SPR of S3D for phosphoinositides 157 7.3 Crystallization and diffraction of S3D:Arf1:MFSD12 159 7.4 Structure of the S3D:MFSD12:Arf1 complex 161 7.5 Comparison of Arf1 binding sites 163 8.1 Nanodisc-link-arf1 theory 167 8.2 Nanodisc-link-Arf1:AP-3 reconstitution 169 8.3 Nanodisc-link-Arf1:AP3 vitrification and 2D classification 170 8.4 Nanodisc-link-Arf1:AP3 3D reconstruction 173 8.5 Optimizing liposome size 176 8.6 Optimizing liposome coating 178 8.7 2D classification of AP-3 on liposomes 180 8.8 AP-3 Coated tubules 181 8.9 Reconstitution of the AP-3 complex on GUVs 183 8.10 2D classification of AP-3 tubules 185 8.11 AP-3 tubules are helical at low resolution 187 8.12 Cryo-electron tomography of AP-3 tubules 189 8.13 Organization of the AP-3 complex on tubules 191 8.14 Schematic diagram of recruitment of the AP-3 complex 195 List of Tables APPENDIX A Table 1 TBC1D23 crystallographic statistics Table 2 S2A crystallographic statistics 1 Table 3 S2A crystallographic statistics 2 Table 4 S2A crystallographic statistics 3 Table 5 S3D crystallographic statistics Table 6 AP-3 appendages crystallographic statistics Table 7 CIMR crystallographic screens APPENDIX B Table 8 Buffers for protein biochemistry Nomenclature 2Fo-Fc electron density map AAGAB Alpha- and Gamma-Adaptin Binding Protein AAK1 AP2-associated protein kinase 1 AFIS aberration-free image shift system ALPs alkaline phosphatase pathway AP adaptor protein AP180 CALM Arf ADP ribosylation factor 1 Arl Arf related ARNO ARF nucleotide-binding site opener ARP2/3 Actin Related Protein 2/3 complex ATP Adenine triphospate BACE1 beta-site amyloid precursor protein cleaving enzyme BAR bin1 amphiphysin rvs BLOC biogenesis of lysosome related organelles complex Ca calcium CALM clathrin assembly lymphoid myeloid leukemia protein CCP clathrin coated pit CCV clathrin coated vesicle CD3 cluster of differentiation 3 CDE caveolin dependent endocytosis CHEVI class C Homologues in Endosome-Vesicle Interaction CIMPR cation independent M6P-receptor CLC1 clathrin light chain CLIC/GEEC clathrin independent carriers/GPI-AP enriched early endosomal compartments CME clathrin mediated endocytosis COPI coatomer COPII Coat protein II CORVET class C core vacuole/endosome tethering CPD carboxypeptidase D cryo-EM Cryogenic sample electron microscopy cryo-ET cryo electron microscopy CrYOLO Cryo You Only Look Once CTF Contrast transfer function CTLA4 cytotoxic t lymphocyte antigen 4 DDM Dodecyl-beta-Maltoside DENN differentially expressed in normal and neoplastic cells DLS Dynamic light scattering DMXL DmX-like protein DUB deubquitinating enzyme EEA1 Early endosome antigen 1 EGFR Epidermal growth factor receptor EHD1 EH domain-containing protein 1 EM electron microscopy ENTH epsin N-terminal homology EOD efficiency Eps15 Epidermal growth factor receptor substrate 15 EPU E pluribus unum ER endoplasmic reticulum ERGIC ER-Golgi intermediary compartment ESCPE1 endosomal sorting complex for promoting exit 1 ESCRT endosomal sorting complexes required for transport FCHO Fer/Cip4 homology domain-only proteins FERRARI Factor for Endosome Recycling And Retromer Interactions FLCN folliculin FNIP folliculin interacting protien Fo-Fc weighted difference (omit) map FYCO FYVE and coiled-coil domain autophagy adaptor 1 FYVE Fab1/YOTB/Vac1/EEA1 GAK Cyclin G-associated kinase GAP GTPase activating protein GARP Golgi associated retrograde protein Gator1 Gap activity towards RAGs GDI GDP-dissociation inhibitor GDP guanidine diphosphate GEF guanine-nucleotide exchange factors GGA Gamma-adaptin ear containing arf binding GPCR G-protein coupled receptor GST Glutathione-S-transferase GTP Guanosine triphosphate GUVs Giant unilamellar vesicle HOPS homotypic fusion and protein sorting Hrb HIV-1 Rev-binding protein hrs Hepatocyte growth factor-regulated tyrosine kinase substrate IDR Intrinsically disordered region IF immunofluorescence ILV intraluminal vesicles ITC Isothermal titration calorimetry KD dissociation constant kV kilo volts LBPA Lysobisphosphatidic acid LDL low density lipoprotein LDLR low density lipoprotein receptor LECA Last eukaryotic common ancestor llps Liquid-liquid phase separation LRO lysosome related organelle LRP9 Low density lipoprotein receptor related protein 9 M6P mannose-6-phosphate MALS multi angle light scattering MES 2-(N-morpholino)ethanesulfonic acid MFSD12 Major facilitator superfamily domain-containing protein 12 Mg magnesium MHC major histocompatibility complex MHD Mu homology domain MOPS 3-(N-morpholino)propanesulfonic acid MSA multi sequence alignment MST multisubunit tethering complex mTOR mechanistic target of Rapamycin MVB multivesicular bodies myr myristoyl NECAP1 Adaptin ear-binding coat-associated protein 1 NPDC1 neuronal proliferation and differentiation control protein 1 NPRL nitrogen permease regulator like NSF N-ethylmaleimide sensitive fusion OCRL Lowe oculocerebrorenal syndrome protein ORP1L Oxysterol-binding protein related proteins 1L PAE predicted alignment error PCH pontocerebellar hypoplasia PEG Polyethylene glycol PH plekstrin homology domain pH potential of hydrogen ions PI Phosphatidylinositol PI35P2 Phosphatidylinositol-3,5-bisphosphate PI3P Phosphatidylinositol-3-phosphate PI45P2 Phosphatidylinositol-4,5-bisphosphate PI4K2A Phosphatidylinositol 4 kinase 2 alpha PI4P Phosphatidylinositol-4-phosphate Pikfyve Phosphatidylinositol kinase with FYVE domain PIP Phosphatidylinositol phosphates pLDDT predicted local distance difference test PLEKHM1 pleckstrin homology and RUN domain containing M1 PQLC2 PQ-loop containing protein 2 Rab Ras-associated binding Racs Ras-related C3 botulinum toxin substrate 1 Rheb Ras homolog enriched in brain RILP Rab7 interacting lysosomal protein RUFY RUN FYVE domain containing protein S2A sigma2-alpha S3D sigma3-delta Sar1 Small COPII coat GTPase SAR1 SDS sodium dodecyl sulfate SDS PAGE SDS polyacrylamide gel electrophoresis SF9 Spodoptera frugiperda SM Sec1/Munc18-like domain SMCR8 Smith-Magenis Syndrome Chromosome Region 8 SNARE SNAP receptor SNX sorting nexin SOC Super Optimal broth with Catabolite repression SPA single particle analysis SPR surface plasmon resonance stam1 Signal transducing adapter molecule 1 STING1 Stimulator of interferon genes 1 STX syntaxin SYNJ Synaptojanin TBC1D Tre2/bub2/cdc16 domain containing protein TBCK tbc and kinase domain containing protein TCR T-cell receptor TGN trans golgi network TM transmembrane TRPML transient receptor potential cation channel mucolipin subfamily TSG101 Tumor susceptibility gene 101 UBD ubiquitin binding domain UIM ubiquitin interacting motif VAMP vesicle associated membrane protein VPS vacuolar protein sorting WASH Wiskott Aldrich and scar homology WDR WD repeat YFP yellow fluorescent protein β-OG β-octyl-glucoside 2 Introduction Chapter 1 Introduction 1.1 Cellular organelles at the ultrastructural level The application of microscopy to biological samples marked a new era into our understanding the inner functioning of cells. Light microscopy first led the way with the discovery of the nucleus and later coupled with chemical staining other membrane bound compartments like the mitochondria and Golgi apparatus (then termed the Golgi complex). However, not until the application of electron microscopy did the truly immense scale of cellular complexity become apparent. Porter, Claude and Fullam published the first electron micrograph of an intact cell, a chicken embryo fibroblast (Figure 1.1.A), which first showed the intertwined “lace-like” nature of mitochondria and endoplasmic reticulum at previously unobtainable resolution (Porter et al., 1945). As technologies developed and methods improved it became only more obvious that the cell did not just contain membranes but was packed with them, forming convoluted and discrete compartments (Figure 1.1.B). Fig. 1.1 Electron microscopy of cellar membranes (A) The first electron micrograph montage of an entire intact cell (chicken embryonic fibroblast) (Porter et al., 1945). (B) Electron micrograph of cellular internal membranes (Farquhar and Palade, 1981). 1.1 Cellular organelles at the ultrastructural level 3 However, visualization alone is not sufficient for understanding. In parallel to these new imaging modalities, in what would now be viewed as classical biochemistry, Claude and colleagues sought to understand the function of cellular components. It was known from the work on cell free fermentation that yeast extracts could catalyse complex chemical reactions and that proteinaceous enzymes from the cell were responsible. Previous attempts to characterize cellular organelles had utilised centrifugation in preparation of the sample, with most of the subsequent assaying consisting of chemical extraction individual species to quantify physical amounts present. Claude was the first to connect these two ideas and with the aid of Hotchkiss and Hogeboom sought to understand where enzymatic activity took place within the cell (Hogeboom et al., 1946). Substantial improvements were made over previous attempts with a combination of gentler homogenisation yielding more intact samples, while differential centrifugation allowed for multiple fractions to be collected over the traditional two fraction approach (supernatant/pellet) and application of quantitative analysis, allowed comparing of concentrations and enzymatic activity of resultant fractions to the initial lysate. The reclassification of fractions to nuclei, mitochondrial, microsomal and supernatant became common place as a growing number of researchers sought to find new enzymatically active fractions. Independent work from both De Duve and Hogeboom found that all fractions assayed contained some distinct enzymatic activity with glucose-6-phosphatase activity being characteristically higher in the microsomal fraction while cytochrome-oxidases were almost exclusive to the mitochondrial fraction. The use of marker-enzymes allowed further dissection of fractionated lysates. In turn De Duve demonstrated that the mitochondrial fraction could be separated further into a heavy and light fraction with the light fraction possessing hydrolase activity that could only be assayed after several days or upon damage to membranes (use of detergents), he termed this new fraction lysosomes. Later, this in turn was also shown to contain another organelle which possessed urate-oxidase activity, now known as peroxisomes (Novikoff et al., 1953). The coupling of these new imaging technologies with partial purifications of cellular lysate proved an effective combination allowing for direct visualisation and assaying of function to be correlated for the first time. This marriage of imaging and biochemistry lead to a rapid 4 Introduction expansion in our understanding of the organisation of cellular biochemistry. The excitement around the potential of this new field spurred yet more technological innovation with the further improvement of electron microscopy but particularly the development of density gradient centrifugation. However, thus far experiments had only yielded static data, in spite of this it hadn’t prevented researchers attempting to fit order to this system, resulting in conclusions with mixed accuracy. It was the first pulse chase style of experiments that radically changed our understanding of dynamics of the endomembrane system. Two strikingly similar yet different papers probed these open questions using a combination of autoradiography and electron microscopy. Roth & Porter studied the uptake of yolk proteins in mosquito oocytes (Roth and Porter, 1964). Noting that the yolk was taken up in pits at the cell surface and coated with “bristles”. These bristles were found on the cytoplasmic face of the pits and could also be found on vesicles adjacent to the plasma membrane; we now know these to be clathrin coated endocytic vesicles. At the longest time points yolk proteins could be found in non-bristle covered membrane. From this Porter and Roth inferred these bristle-coated vesicles were derived by pinching off of bristle coated pits from the plasma membrane and hypothesised that the bristle not only controlled the uptake but also provided the specificity for the yolk protein before being removed as they enter the other cellular compartments. 1.1 Cellular organelles at the ultrastructural level 5 Fig. 1.2 Directionality of membrane trafficking (A) Electron micrograph of bristle coat formation in mosquito oocyte: (DM) dense material, (OOC) oocyte, (FES) interface between oocyte and follicular epithelium (Roth and Porter 1964). (B) A time point in pulse chase providing evidence that delivery of radiolabelled proteins are delivered to the zymogen granules. 6 Introduction Around the same time George Palade, a master of electron microscopy, was the first to probe the kinetics of this system in mammals (Caro and Palade, 1964). In seminal work he tracked the synthesis of proteins in cells isolated from guinea pig pancreas demonstrating that newly synthesised proteins first transited the rough endoplasmic reticulum followed by the Golgi complex and finally ended up in the zymogen storage granule. In the same work he wrote: “Less well established is the role played by the small vesicles. They might function as shuttle carriers between the part rough and part smooth elements of the endoplasmic reticulum and the condensing vacuoles. At any rate, it is reasonably clear that they are an intermediate stage” Caro & Palade (1964) These two papers were not only the earliest studies into the dynamics of cellular trafficking but also the first to suggest vesicles act as carriers between organelles and provided evidence of both anterograde and retrograde movement. Together these landmark discoveries were our initial insight into the molecular mechanisms of protein sorting. 1.2 The Endocytic system 7 1.2 The Endocytic system The mammalian endocytic system is a complex, dynamic entity, which runs from the plasma membrane through a variety of endosomes to the degradative endolysosome. It can be considered as ending with the neutral pH storage lysosome and Lysosome-Related Organelles such as melanosomes and lamella bodies in specialised cell types (Klumperman and Raposo, 2014, Bowman et al., 2019) (Figure 1.3). The endocytic pathway serves as the entry point from the cell surface for TM-proteins, their ligands and many pathogens. The main physiological function of the endosomal systems is to act as a central hub for the sorting of transmembrane (TM) and luminal proteins. In recent years a role in nutrient sensing has revealed the endolysosomal system also acts as an integrative signalling platforms. Aberrant functions along its length are linked to a vast range of pathophysiological states (Azarnia Tehran et al., 2019, Bartuzi et al., 2016, Cui et al., 2018, Maxfield, 2014, Reitz, 2018). As such a mechanistic understanding of how the architecture, composition and trafficking of the endocytic pathway is key to understanding membrane trafficking within the cell. 8 Introduction Fig. 1.3 Maturation of the endocytic pathway A schematic diagram of key events within the process of endosomal maturation and trafficking pathways out of the endosomal system. 1.3 Endosomes: An overview 9 1.3 Endosomes: An overview Although the process of sorting transmembrane and luminal proteins pertains to all organelles, it is a major defining feature of the endosomal system. Whilst not all details of defining an endosome are universally agreed upon, “endosome” is accepted as an umbrella term encompassing a collection of organelles which share similar markers, morphology and/or origin (Solinger and Spang, 2022, Klumperman and Raposo, 2014). Where this differs from other organelles is the temporal and transient nature between the sub categories of endosomes. Generally speaking, endosomes are commonly written about in one of three sub-categories; early endosomes (sometimes called sorting endosomes), late endosomes and recycling endosomes (Figure 1.3). To complicate the matter further, these sub-categories only represent a snapshot of an endosome often only experimentally visualising a single characteristic. It is more likely that endosomes exist on a cline, with specific markers or characteristics weighting an endosome somewhere along this spectrum, from more Golgi and TGN-like to a later endosome or lysosomal compartment. Thus, while a clear-cut definition of what criteria an endosome must meet to be designated in a specific sub-category maybe impossible, the concept that endosomes undergo a maturation process from early endosomes to late endosomes is well accepted by most. The ‘earliest’ endosomes are formed by fusion of endocytic carriers derived from the cell surface. The plasma membrane derived endocytic vesicles that merge into/fuse with pre- existing early endosomes are heterogeneous in origin with clathrin mediated endocytosis (CME), clathrin independent carriers/GPI-AP enriched early endosomal compartments (CLIC/GEEC), caveolin dependent endocytosis (CDE), macropinocytosis and phagocytosis all utilising different mechanisms of uptake (Doherty and McMahon, 2009). As early endosomes mature through becoming sorting endosomes and then late endosomes, often termed multivesicular bodies (MVBs) due to the production of inwardly budded intra luminal vesicles (ILV) in their interiors, they receive additional cargo from the Golgi and other organelles and their luminal pH decreases (Figure 1.3). 10 Introduction Since endosomes lack any exclusive TM marker proteins to define their state of maturation, their status is typically defined by the presence of phosphoinositide (PIP) and small GTPase maturation markers on their cytosolic surfaces (Dickson and Hille, 2019). These in turn recruit effector proteins of a huge variety of function, including trafficking, signalling, fusion- specificity and marker modifying complexes. As an endosome matures, positive amplification and negative feedback loops control the modification of specific PIPs via PI kinases and phosphatases and the exchange of the nucleotide bound state of GTPase markers via GTPase activating proteins (GAPs) and guanine-nucleotide exchange factors (GEFs). All told, many proteins are recruited and exchanged or lost from an endosomes surface to orchestrate this maturation process. Finally, an endolysosome is formed by the heterotypic kiss-and-run fusion of late endosomes/MVBs with lysosomes (a degradative enzymatic storage organelle with a neutral pH), a mature, low pH degradative compartment (Luzio et al., 2009, Bright et al., 2005). Notably these endolysosomes are fusion competent and continue to undergo fusion and content exchange with late endosomes, lysosomes and other endolysosomes, producing a hybrid degradative compartment (Figure 1.3). 1.3 Endosomes: An overview 11 1.3.1 Architecture, identity and function in endocytic system TM protein cargo sorting must occur throughout an endosome’s life to ensure that cargo is dispatched to its intended steady state localisation, such that only cargo destined for destruction are present when the endosome finally becomes the highly degradative endolysosome. Thus, there exist trafficking routes from endosomes to most other organelles, with each route being mediated by a different tubular or vesicular carrier. These include AP-2 with clathrin, AP-1 with clathrin, AP-3±clathrin±BLOC1, retromer with SNX3, retromer with SNX27, retromer with SNXBARs, retriever, ESCPE±SNX27 (Sanger et al., 2019, Chen et al., 2019). While the specific details of each coat and its associated factors vary between coats there are some general commonalities that are shared across all coats. They are formed by recruitment of proteins from the cytosol to form a polymeric assembly that selects cargo and deforms the membrane; the carrier undergoes scission from its donor membrane and is transported to its target membrane; the coat is removed and finally the carrier and the target membrane dock and fuse (Dell'Angelica and Bonifacino, 2019). If more than one type of coated carrier can bud from an endosome of a given maturity, the formation of all coated structures are spatially segregated into subdomains. These subdomains are determined by interplay between PIP and small GTPase markers which often directly recruit adaptors/coats. Inherent in this model of endosomes is the requirement that identity markers can sense the current maturation status of an endosome. This requires detecting, integrating and driving changes in the physiology and environment on both sides of the endosomal limiting membrane including marker presence, luminal pH, intraluminal vesicle production, cytoplasmic face protein complement, carrier coat recruitment and limiting membrane fusion events. 1.3.2 Entry into the endocytic system The primary mechanism for selective cargo uptake from the cell surface is CME, due to our deeper understanding of CME this is the only endocytic process whose specifics will be briefly outlined. 12 Introduction The clathrin-coated vesicles (CCVs) that mediate CME, are formed from clathrin-coated pits (CCPs), which account for ~2% of the PM surface area. Clathrin trimers form an outer mechanical polyhedric lattice composed of hexagons and pentagons. This scaffold only contacts the membrane and its embedded cargo indirectly through clathrin adaptors, the principal ones being heterotetrameric Adaptor Polypeptide 2 (AP-2) complex and the monomeric clathrin assembly lymphoid myeloid leukemia protein (CALM), both of which bind PI45P2 (Kaksonen and Roux, 2018). AP-2 sorts a variety of cargo for incorporation into CCVs through binding YXXΦ and dileucine motifs. AP-2 also coordinates the assembly of a network of 300-400 proteins that undergo a dynamic and finely choreographed process that takes ~2 minutes to form a CCP. CALM plays a critical role in membrane deformation and it also selects SNAREs, which allow the fusion of endocytic CCVs with their target membranes, for incorporation into the CCVs (Miller et al., 2015, Miller et al., 2011). ‘In cell’ AP-2 activation is facilitated by binding to its membrane pre-localised activator FCHO. The FCHO is crosslinked into a liquid-liquid phase separated (LLPS) nanocluster by Eps15 (Ma et al., 2016). This phase-separated nano cluster serves as the initiator for CCP formation by selectively recruiting and concentrating AP-2 and other clathrin adaptors to which it can bind (Day et al., 2021, El Alaoui et al., 2022). The LLPS nanocluster transforms into a central clathrin-crosslinked patch and a LLPS ring. Molecular crowding combined with helical insertion remodels the patch first into a dome followed by an omega shape. The constricted neck acts as a template for BAR domain-containing proteins such as SNX9 and the amphiphysins onto which a large GTPase dynamin is then assembled to drive the final scission of the CCV from the PM. 1.4 Early endosomal formation from endocytic CCVs Uncoating of endocytic CCVs allows access of new protein factors to their membrane. In CME uncoating is initiated by the recruitment of phosphoinositide phosphatase Synaptojanins 1/2, which convert the plasma membrane markers phosphatidylinositol 4,5 bisphosphate (PI45P2) and phosphatidylinositol 3,4,5 bisphosphate (PI345P3) into phosphatidylinositol (PI) and phosphatidylinositol 3 phosphate (PI3P) respectively (McPherson et al., 1996, Posor et al., 2022). Simultaneously the acquisition of an active Rab5-guanine nucleotide exchange factor (GEF) facilitates the other main step in becoming defined as an endosome, the acquisition of 1.4 Early endosomal formation from endocytic CCVs 13 the Rab5:GTP. In the case of CME this is thought to be GAPVD1 (homolog of the nematode protein RME-6) and may be acquired during CME and later activated (Smythe, 2015). Rab5 has been extensively studied and is often described as a primary marker of the early endocytic system. Rab5:GTP can recruit a wide variety of effectors and although many have been identified and characterized, yet the function of a significant number remains elusive (Bucci et al., 1992, Gillingham et al., 2019). Rab5 is also able to recruit lipid phosphatases OCRL and INPP4A/B which further aid in dephosphorylating any remaining PI45P2 and PI345P3 (Bohdanowicz et al., 2012). Another lipid enzyme and early Rab5 effector is the PI3K complex II (Christoforidis et al., 1999). When recruited to the endosome, PI3K phosphorylates PI to PI3P, the other widely acknowledged marker of the early endocytic system. The presence of PI3P and Rab5:GTP cooperatively drive the expansion of the forming early endosome by acquiring further endocytic vesicles through the recruitment of the tether EEA1. EEA1 consists of three domains an N-terminal zinc finger (which binds Rab5:GTP), an extensive coiled-coil region and a C-terminal FYVE domain (which binds PI3P) (McBride et al., 1999). Together, these allow EEA1s to bridge just over 200nm to scavenge for newly endocytosed vesicles before undergoing an entropic collapse allowing for their SNARE mediated fusion (Murray et al., 2016). This is not just true of the endocytic vesicle but can extend to larger endosomal structures, which are under a constant balance of fusion and fission. The coincidence detection of PI3P and Rab5:GTP drives the remodelling of the endosome and the recruitment of a variety of other factors which enable endosomal sorting. Notably the cascade of recruitment of GEFs for specific GTPases from multiple families, result in the creation of patches of Rabs, Arfs, Rheb and Racs among others (Dutta and Donaldson, 2015, Saito et al., 2005, Palamidessi et al., 2008). Through binding their effectors many of these GTPases coordinate the formation of marker-specific subdomains on an endosome’s surface (Figure 1.3). The endosomal system also receives more material from the trans Golgi in clathrin and GGA coated CCVs directly into fully formed early endosomes. These CCVs are enriched in mannose-6-phosphate (M6P) receptors, which transport newly synthesised luminal lysosomal hydrolases due to the modification of these enzymes with M6P tags (Bonifacino, 2004). 14 Introduction 1.5 Maturation into later endosome indicated by Rab markers Although Rab7 is best characterized as a marker of later endosomes, a Rab7:GTP positive subdomains is formed on the surface of early endosomes early in its life . This pioneering pool of Rab7 is recruited and activated by the presence of its Rab5:GTP and PI3P binding GEF the MON1:CCZ1:RMC1 complex. To date the MON1:CCZ1:RMC1 complex is the only identified GEF for Rab7 suggesting it mediates a central role in regulation of endosomal maturation (Dehnen et al., 2020, van den Boomen et al., 2020). Despite its importance only recently has the activity of the yeast MON1:CCZ1 complex been probed, identifying an inhibitory mechanism of the MON1 disordered N-terminus, which when removed allows for efficient binding of both Rab5:GTP and PI3P containing liposomes in order to efficiently recruit Rab7. The role of metazoan RMC1 (yeast Bulli) remains to be determined (Herrmann et al., 2023, Yong et al., 2023). Rab7:GTP, like Rab5:GTP, recruits wide variety of effectors however it has yet to be systematically proven if there is a difference in which effectors are recruited to this pioneering pool as opposed to later Rab7:GTP positive endosomes. Of the potential effectors of this pioneering pool several have been well documented. The most studied is the retromer complex (discussed in more detail subsequently), which has been shown to bind Rab7:GTP directly as part of its mechanism of recruitment (Nakada-Tsukui et al., 2005, Priya et al., 2015). Retromer is a coat-forming complex whose role is to mediate the sorting of a variety of cargoes from endosomes to the Golgi and/or plasma membrane. Notably in recent years interest in retromer and its effectors has skyrocketed as gene association studies discovering a strong link with neurodegenerative disorders like Alzheimer’s and Parkinson’s disease (Williams et al., 2017). Retromer may directly couple the recycling of cargo to endosomal maturation. One binding partner of the VPS29 subunit of retromer is the Rab7 GAP TBC1D5 (Seaman et al., 2009). While the recruitment of TBC1D5 by retromer has been suggested to play a role in uncoating the retromer-coated carriers, Rab7 is not required for the formation of tubules in vitro and no mechanism preventing TBC1D5 from binding forming retromer tubules has been observed (Jia et al., 2016). Thus, these data could be reinterpreted as Rab7:GTP initially concentrates retromer to an endosome’s surface but once localised on the endosomal membrane the presence 1.6 PIP conversion in endosomal maturation 15 of cargo and of PI3P, which bind to retromer accessory coat proteins such as SNX3 and SNX27, is sufficient to stably form a Retromer coated tubule along with its bound Rab7 GAP TBC1D5. However, when Retromer cargo is subsequently depleted from an endosome, both Retromer and TBC1D5 will also be lost from the endosomal surface. This will relieve the inhibition on Rab7:GTP which will subsequently increase, ushering in the early to late conversion of the endosome. Such a molecular crosstalk between trafficking and endosomal maturation as indicated by the Rab and PI complement ensures cargoes that must be recycled are salvaged and not erroneously degraded. 1.6 PIP conversion in endosomal maturation An important step in endosomal maturation is phosphatidylinositol conversion whereby an early endosome’s PI3P is phosphorylated to phosphatidylinostitol 3,5-bisphosphate (PI35P2), since the higher the ratio of PI35P2 to PI3P the later the endosome is marked as being. PI35P2 has fewer identified effectors than PI3P although a key one is likely the TRPML channel, which triggers efflux of Ca2+ from an endosome’s lumen which is reported to aid the fusion of late endosomes with endolysosomes and lysosomes via synaptotagmin VII (Dong et al., 2010). PI3P phosphorylation is mediated by the large multi subunit PIKfyve:Fig4:Vac14 complex (Sbrissa et al., 2004). The current dogma is that PIKFYVE is recruited by the interaction of its FYVE domain with endosomal PI3P and the PIKfyve in turn stabilising Vac14 on the endosomal membrane. Vac14 forms an arching scaffold which counterintuitively also recruits Fig4, a lipid phosphatase which hydrolyses the 5-phosphate of PI35P2 converting it back to PI3P, in essence reversing the action of PIKfyve (Lees et al., 2020). The current explanation for this situation is that the co-recruitment of both a lipid kinase and phosphatase suggest a dynamic process in which both the forward and reverse modifications take place to ensure a balance of PI3P and PI35P2. 1.7 Endosomal pH Another defining feature of endosomes is the luminal pH, which becomes increasingly acidic throughout maturation. This has proven to be important multifaceted aspect of endosomal function, controlling the release of mannose-6-phosphate (M6P) tagged lysosomal hydrolases 16 Introduction from the cation independent M6P-receptor (CIMPR) following delivery from the Golgi in GGA and clathrin coated vesicles as well as the release of endocytosed cargo such as iron from transferrin and LDL particles from LDLR (Brown et al., 1986, Rao et al., 1983, Anderson et al., 1977). Controlled dissipation of this proton gradient also allows for export of metabolites by facilitated diffusion to the cytoplasm (Russnak et al., 2001, Reeves, 1979). The lower pH late in the endocytic system is exploited by many pathogens as a mechanism of entry/release into the cytosol such as Influenza utilising a low pH-induced conformational change in their spike protein to either rupture or fuse with the endosomal membrane (Gruenberg and van der Goot, 2006). The presence of Rab5:GTP triggers the endosomes to begin to acidify and later upon subsequent maturation to Rab7:GTP marked, the endosomes acidify further. Acidification is driven by the V-type ATPase, which hydrolyzes ATP to translocate H+ ions from the cytoplasm to the endosomal lumen. The V-type ATPase is formed of two sub complexes, a transmembrane Vo and a soluble V1 rotating head. The acidification is thought to be regulated by the controllable association/dissociation of the V1 subunit from the Vo (Parra and Kane, 1998, Sava et al., 2024). In yeast acidification has been shown to be dependent on the Regulator of ATPase of the Vacuole and endosome complex (RAVE) complex, which consists of the heterotrimer of rav1, rav2 and skp1. In mammals interestingly, there are three homologs of rav1 (WDR7, DMXL1 and DMXL2), only one homolog of rav2 (ROGDI) and while there are mammalian homologs of skp1 they are not thought to be part of the mammalian regulating complex (Yan et al., 2009). While the exact mechanism of the V1 recruitment by the RAVE complex is not known, it is interesting to note that WDR7 and DMXL1/2 have both been identified as Rab effectors: WDR7 is reported to be recruited by both Rab5 and Rab7 whereas DMXL1/2 is recruited only by Rab7 (Gillingham et al., 2019). These data suggest that WDR7 is recruited at the early endosome while WDR7 and DMXL1 are recruited to maturing endosomes and would also provide a possible explanation to the observed biphasic nature of endosomal acidification (Podinovskaia et al., 2021). Another possible Rab7a-dependent mechanism for affecting endosomal pH via V-type ATPase assembly/function is via the Rab7 effector RILP (Sava et al., 2024, Mulligan et al., 2024). RILP has been proposed to bind V1G1 subunit to influence V1V0 assembly and/or stability. RILP also 1.8 Acquisition of ILVs and fusion with lysosomes 17 influences endosomal positioning in concert with two other Rab7 effectors ORP1L and FYCO via interacting with dynein and kinesin-based motors; suggesting endosome and endolysosome positioning is also coupled to luminal pH (Johnson et al., 2016, Bright et al., 2016). 1.8 Acquisition of ILVs and fusion with lysosomes Although the budding of intraluminal vesicles (ILVs) into endosomes begins early in their lifetime, the number of ILVs increases as the endosome matures (Williams and Urbé, 2007). ILV production is driven by the cytosolic peripheral membranes ESCRT complexes (ESCRT 0, 1, 2 and 3) and the appearance of the lipid lysobisphosphatidic acid (LBPA). LBPA is an inverted cone-shaped lipid and is believed to aid in forming the negative membrane curvature that is critical for inward budding (Matsuo et al., 2004). Being sorted into ILVs, triggered through recognition of ubiquitination signals by ESCRTs 0,1 and 2 is key to TM cargo protein degradation. Being sorted into an ILV also terminates signalling from growth factor receptors and is thus is central to downregulation of cellular stimuli and is often mutated in cancer (Raiborg et al., 2008, Carlton and Baum, 2023). 1.9 Degradation Delivery of neutral pH-inactivated hydrolases in the ‘cheese-like consistency’ lumen of storage lysosomes is via their fusion with late endosomes and endolysosomes (Luzio et al., 2007). The process is critical to TM protein degradation and these heterotypic fusion events can be either direct and complete or transient via ‘kiss and run’ (Bright et al., 2005). A key player in these processes is the heteropentameric, fusion-mediating SNARE-binding HOPS complex (Seals et al., 2000). Despite all this work and many studies probing the mechanism of the HOPS all factors governing its ability to mediate fusion remain unclear. It has been suggested that Rab7, 2 and Arl8 interact with PIPs and proteins including HOPS, RILP and PLEKHM1 to control the fusogenic capacity of late endosomal and lysosomal compartments (van der Beek et al., 2019, Schleinitz et al., 2023). 18 Introduction Finally, a further functionally important role of the endocytic system is mediating autophagy. Autophagy allows for the controlled degradation of cellular components and can provide nutrients for cell survival under starvation conditions (Zhang et al., 2024). However, as this is not a focus of this thesis they will not be discussed in any further detail. 1.10 Trafficking within the endosomal system Coats within the endosomal system come in a variety of shapes and size and while not all adaptors and coats are discussed in detail in this thesis, the generalities of their function give us a framework in which to describe the process by which cargo proteins are sorted. Thus, what follows is a short overview of processes shared across adaptor/coat trafficking complexes required for efficient formation and disassembly of coated carriers. 1.10.1 Recruitment of membrane scaffolds and cargo adaptors In the cell, the initiation of any vesicular or tubular trafficking event begins with the recruitment of adaptors to the donor membrane to form the inner layer of the carrier’s coat. Most often this precedes the recruitment of a cross linking scaffold to form an ‘outer coat’, although recruitment of both layers ‘en bloc’ can also occur as is the case for COPI and possibly retromer (Dell'Angelica and Bonifacino, 2019). The adaptors crosslink the membrane and its embedded cargo to the outer mechanical scaffold. One of the most generalized concepts for coats is that their primary mode of recruitment is via small GTPases of the Arf family as is the case for COPI, GGAs, AP-1 AP-3 AP-4 and the highly homologous Sar1 for COPII (Donaldson and Jackson, 2011). These GTPases function as recruiting factors by acting as molecular switches with exchange of GDP for GTP driving a conformational change that allows the recruitment of a coat adaptor as an effector. This provides a tight layer of regulatory control as a GTPase will only be active when its GEF is present to exchange GDP for GTP; The presence of a cognate GAP disfavours adaptor binding. Together the presence of the GEF and absence of the GAP restricts the recruitment of a coat to a discrete subdomain. Arf family members couple the GTP-bound state to membrane insertion thus in their GDP bound conformations they are cytosolic with both the N-terminal 1.10 Trafficking within the endosomal system 19 amphipathic helix (helix 0) and amino-terminal myristoyl moiety buried in a hydrophobic trench. However, when GEF drives GTP exchange they are both ejected allowing for association of the hydrophobic face of helix0 with the membrane and embedding of the myristoyl group into the bilayer to induce positive membrane curvature (Antonny et al., 1997). Fig. 1.4 Diversity of mechanisms of GTPase recruitment (A) Cartoon of Arf family member between cytosolic GDP bound conformation and membrane bound GTP conformation. (B) Cartoon of Rab family member between membrane bound GDP bound conformation and membrane bound GTP conformation. The one exception to the use of Arf family of GTPases in coat formation is the apparent use of Rab7:GTP in recruiting the retromer coat (Nakada-Tsukui et al., 2005). The actual membrane recruitment of Rab7, like all other Rab GTPases, is comparatively insensitive to its nucleotide bound state as the availability of its membrane-inserting palmitoyl group, which is attached via a C-terminal CAAX motif, is unaffected by nucleotide exchange (Gavriljuk et al., 2013). The relative localisation of a Rab’s GEF and GAP determines whether it will be in its GTP form and so capable of recruiting effectors. However, when in the GDP bound conformation, a Rab is more likely to interact with GDI-extraction machinery that removes the Rab from a membrane and facilitating its reinsertion into a new membrane (Hutagalung and Novick, 2011). Consequently, many Rabs that are proposed to act at very specific location are found in a wide range of organelles as evidenced by cellular fractionation and messy IF staining (Itzhak et al., 2016, Cho et al., 2022). As such it is somewhat unusual that a coat is localised by a Rab to drive its assembly on membranes. As previously discussed in section 1.6, Rab7:GTP may 20 Introduction merely act to concentrate retromer on an endosomal subdomain and does not have to directly recruit the retromer coat. Phosphoinositide inositol phospholipid markers of organelle identity also influence the localisation of coats and cargo adaptors. The most striking examples of this is the AP-2 complex which possesses four PI45P2 binding sites and drives CME exclusively at the PM and Sorting Nexin 3 (SNX3), which localizes only on PI3P-positive membranes from where it can drive retromer coat formation (Höning et al., 2005, Xu et al., 2001). While the concept of each compartment having an essentially unique phosphoinositide species is commonplace, there is mounting evidence to suggest that there are multiple minor pools of different phosphoinositide species on various organelles and that these may have distinct functional roles. Finally, it has been suggested that certain combinations of lipids may actually be required for efficient coat recognition; one proposed example of this is the availability clustering of PI45P2 and its height in the membrane, thus availability, being increased in the presence of cholesterol (Lolicato et al., 2022). Cargo presence is generally not thought of as a major determinant of the specificity of coat recruitment as this would risk recruiting a coat to its destination. However, the presence of cargo will greatly increase the strength of its cognate adaptor binding to a given membrane in the presence of its Arf and PIP recruiting factors by avidity effects (Höning et al., 2005, Miller et al., 2011). These seemingly incompatible observations can be reconciled if the adaptor can only ‘see’ cargo once it has been correctly initially positioned by interacting with the correct GTPase and PIP. In the cases of AP-2 and AP-1 (assumed for other adaptors) this occurs through a membrane coupled conformational change in the adaptor (Collins et al., 2002, Heldwein et al., 2004, Jackson et al., 2010, Ren et al., 2013). One further caveat would be the scenario whereby if an adaptor bound multiple cargoes, then their simultaneous presence on only a given organelle or subdomain of an organelle will give the appearance of cargo in defining specificity to coat localisation. Furthermore, in the busy environment of the cytoplasm there will be competition for each factor from other proteins and so “coincidence detection” of all determinants is needed to get physiological levels and speeds of coat recruitment. Most recently, it has been shown that in 1.10 Trafficking within the endosomal system 21 addition to the more well-established mechanisms of coat recruitment described above, the exact site at which coat formation in CME occurs is also influenced by the presence of adaptor concentrating “pioneer factors” – in the case of AP-2 these are a network of FCHO1/2 and Eps15 molecules, which also help to drive the conformational change in AP-2 necessary for cargo recognition (Zaccai et al., 2022). Reports suggest that these pioneer factors behave as a protein condensate which undergo liquid-liquid phase separation (LLPS) and that the partitioning and concentration of adaptors into phase separated nano structures (Ma et al., 2016) plays a key role in initiating CME (Day et al., 2021). 1.10.2 Cargo selection: Mechanisms of molecular sorting Although vesicular/tubular mediated transport explained how a protein may move between different compartments it did not explain how a protein moves to the correct compartment on the route it needs to take. In other words, if all ~6000 human transmembrane proteins and all the secreted proteins synthesised on the endoplasmic reticulum (ignoring the existence of mitochondria and peroxisomes post translational import systems), how does the cell un-mix this otherwise ‘jumbled soup’ to ensure that proteins traverse a complex endomembrane system and arrive at their intended destination? As such the “sorting problem” arose and research into how specificity of cargo was imparted began. The concept of a protein’s localisation being innately encoded by its sequence was first published in symposia of the society for experimental biology, where it was mentioned as little more than a hypothesis. This concept was again predicted the following year by Günter Blobel where he not only postulated that there was not only encoding for active movement of specific transmembrane proteins but also for retention in the ER within protein’s sequences (Blobel et al., 1979, Blobel, 1980). He further proposed that TM or soluble luminal proteins may be able to “piggy-back” to their desired location though interacting with a motif bearing transmembrane protein and that soluble proteins like lysosomal hydrolases may be in fact be shuttled using this process. He was correct on all accounts. Multiple mechanisms of sorting have been discovered which fit into three categories that all require direct interaction of cargo with an adaptor: sorting motifs (short linear motifs found on 22 Introduction many different cargoes), post-translational modification (ubiquitin, phosphorylation, mannose- 6-phosphate) specific cargo:adaptor interactions (AP-3:VAMP7). Recently it has also been suggested that the physiochemical properties (transmembrane helix length and phase separation or selective aggregation) can also act in cargo sorting (Sharpe et al., 2010, Park et al., 2023, Huttner et al., 1991, Parchure et al., 2022). 1.10.3 Sorting motifs Sorting motifs normally consist of a short (3-6 residue) linear peptide sequences located in the intrinsically disordered cytosolic tails of TM proteins of many different functions. To obtain specificity whilst being only 3-6 amino acids in length, sorting motifs employ a combination of electrostatic, hydrogen bonding and van der Waals interactions/hydrophobic packing to maximise the spatial and chemical compatibility of their interface. Yet despite this, the affinity of sorting motifs for their adaptors is considerably weaker than expected for stable complex formation in solution with a KD usually in the 10-100 μM range. This comparatively weak binding is, however enhanced by avidity effects originating from the co-recognition of other cargoes and recruiting determinants (PIPs and small GTPases) located on the same membrane. When considered holistically this results in effective affinities that are in the 0.5-5 μM range. This type of coincidence detection mechanism based on multiple dynamic ‘weak’ individual interactions permits effective regulation of coat:cargo binding by covalent modification of the motif such as phosphorylation as seen for the tyrosine based motif used by the immune system receptor CTLA4 (Miyatake et al., 1998). It also allows for easy and quick reversibility, a factor that is necessary during carrier un-coating and results in coat components being quickly recycled for reuse. 1.10.4 Tyrosine-based sorting motifs The first sorting motif to be identified, and the most studied, is the Tyrosine-based sorting signals, which were subsequently reclassified into two distinct motifs, the NPXY motif and YXXΦ motif (Boll et al., 1996, Chen et al., 1990). Both allow for direct interaction between cargoes and clathrin adaptors and have been most heavily studied in the context of endocytosis. YXXΦ motifs are degenerate with X being any residue and Φ any large hydrophobic residue 1.10 Trafficking within the endosomal system 23 (Commonly FILMV), although other less ideal motifs have been proposed as functional (Sanger et al., 2019). Structurally the YXXΦ motif in complex with the c-μ subunit has been solved for AP-1, AP-2 and AP-3, all of which show similar mechanisms of binding (Owen and Evans, 1998, Mardones et al., 2013). In all cases the tyrosine and the Φ residue bury into respective hydrophobic pockets while the backbone of the X residue hydrogen bonds form a very short β-augmentation. This combination of burying hydrophobic residues and creating hydrogen bonds allows for a higher-than-expected affinity for what would otherwise be a very short peptide sequence. However, the affinity of the motif to a c-μ domain in solution would be considered relatively low affinity compared to other interactions measured in biology. This is thought to be mitigated when an adaptor is held to a membrane as this additional constraint would decrease the degrees of freedom, artificially increasing the adaptor concentration. The coat must also be disassembled and recycled for subsequent use therefore a weak affinity allows dissociation of cargo from the coat-adaptor complex. Notably there is also a difference in affinity of different adaptors for the same motifs, C-μ2 binds TGN38 YXXΦ motif with an affinity of 2.3μM while C-μ3 of the AP-3 complex is much weaker 14 μM (Owen and Evans, 1998, Mardones et al., 2013). As AP-1 and AP-3 are thought to sort from the same compartment yet it is not known if specificity is imparted, and if so how, this favors interaction to only one adaptor over another. Nevertheless, it seems highly likely that there is some form of specificity as TGN38 (which is often used as a marker of the TGN) is almost undetectable at the plasma membrane suggesting it is very quickly re-internalized and returned to the TGN if mistrafficked. 1.10.5 Alternative mechanisms of sorting Another mechanism of cargo selection is recognition of post-translationally ubiquitinated lysine. The two main examples of this are in CME from the cell surface and driving incorporation into ILVs in endosome sand late endosomes. Although initial speculation suggested that during CME cargo (such as active phosphorylated EGFR) underwent multiple mono ubiquitination (Sigismund et al., 2013), then were later demonstrated that cargo could be lys63-linked poly-ubiquitinated in order to be recognised by the ubiquitin interacting motifs (UIMs) of the endocytic adaptors EPS15 and epsin (Hawryluk et al., 2006). The same 24 Introduction conjugated ubiquitins are recognised by the UIMs Hrs STAM of the ESCRT-0, UBDs of ESCRT-I TSG101 and ESCRTII Vps36 to select a cargo for internalisation into ILVs for degradation (Henne et al., 2011). The ubiquitination processes are directly driven through the activity of the E3 ubiquitin ligases and opposed by the action of deubiquitinating enzymes (DUBs), which can reverse the processes. A further post-translation modification involved in cargo selection by adaptors is protein phosphorylation. For example, the phosphorylation of GPCRs like the β1-adrenergic receptor at the PM on multiple sites can mediate their interaction with the CME β-arrestin adaptors (Lohse et al., 1990). This process ties in functionally with GPCRs activity as this allows for desensitisation of the signal even while an agonist is bound. 1.11 Membrane remodelling and sculpting Once a coat has successfully enriched cargo into the nascent membrane subdomain, the membrane subdomain must begin to isolate itself from the organelle of origin to create a vesicular or tubular carrier. As there are many different carrier coats in the cell, it is of no surprise these different types of coats come in a variety of shapes and sizes (Figure 1.5). Every coat discovered thus far is polymeric and forms (pseudo-)symmetry in the packing of its protomers, which plays a role in physically deforming the membrane to generate a carrier. However, membrane deformation can also be driven by direct insertion of a protein from the cytosol into the outer leaflet of the membrane or the lateral pressure derived from tightly packing/concentrating peripheral membrane proteins on the membrane’s cytoplasmic face (Stachowiak et al., 2012). 1.11 Membrane remodelling and sculpting 25 Fig. 1.5 Coats of all shapes and sizes an array of coat forming complexes from within the endocytic system (A) COPI triad assembled on membranes (Dodonova et al., 2015). (B) Retromer:SNX3 complex assembled on membranes (Leneva et al., 2021). (C) Retriever complex in solution (Boesch et al., 2023). (D) AP-1 complex assembled on membranes (Hooy et al., 2022). (E) crystal structure of ESCPE-1 bar domain coat (Lopez-Robles et al., 2023). The most common carrier is the vesicle, defined by its approximately spherical shape. Coats that form vesicles, like COPI and clathrin are polymeric with icosahedral symmetry present. In the case of COPI, trimerization of individual COPI leaves (α, β’, β, γ, δ, ε, ζ) creates the COPI triad which becomes the functional asymmetric unit of the coated vesicle. Each individual triad is relatively rigid providing some initial curvature required for deformation of the membrane. Further higher order packing between triads creates distinct “linkages” which although more flexible further restrict the geometry of the coat causing vesiculation (Taylor et al., 2023). While AP complexes are related to the COPI (β, γ, δ, ζ) F subcomplex, they do not seem to sculpt the membrane, the clathrin scaffold’s is presumed to play a role driving membrane curvature, but how is unclear. Clathrin exists as triskelia, composed of three copies of both clathrin heavy chain and light chain (Ungewickell and Branton, 1981). From ‘unroofing’ experiments and live cell microscopy it has been shown that clathrin forms both flat assemblies variously termed patches or plaques on the plasma membrane or roughly hemispherical domed 26 Introduction structures (Heuser and Anderson, 1989, Mund et al., 2023). The former is thought to exclusively contain planar packing hexagons. To become a spherical coat, the smallest possible structure found corresponding to a ‘C60 buckyball’, pentagons must be added to the flat lattice (Kanaseki and Kadota 1969). How exactly a flat lattice deforms and remodels into a CCV is debated with two competing models. The “constant area” model requires structural contacts to be broken and reformed to allow for the creation of pentagons (Avinoam et al., 2015, Sochacki et al., 2021). By contrast the “constant curvature” model suggests that clathrin is not remodelled per se but that some of the triskelia added are added directly into pentagonal conformers (Kirchhausen, 1993). These two models are debated and the reality probably lies somewhere between the two extremes. Recent work has proposed and demonstrated intermediate structures of “spring-loaded” flat clathrin lattices containing strained pentagons which relax to form a CCV (Tagiltsev et al., 2021). An alternative school argues that the scaffold simply stabilises or moulds to the underlying membrane architecture, which is caused by the presence of peripheral membrane proteins. Most coats are recruited by GTPases which possess an N-terminal alpha helix, termed helix0, whose insertion into the membrane’s cytoplasmic face forced it to bend towards the cytosol as previously discussed in section 1.10.1 (Donaldson and Jackson, 2011). Biophysical studies have demonstrated a role of membrane crowding in driving membrane curvature. In this the membrane is curved because the system seeks to relieve lateral pressure resulting from significantly unstructured but membrane attached proteins interacting with each other. This provides one explanation for why so many endosomal and coat proteins have extensive unstructured portions (Stachowiak et al., 2012, Tesei et al., 2024). Tubular carriers exist and can rely on similar mechanisms of membrane deformation. However, whereas the size and architecture of vesicles is constrained all round by the 3D symmetry of the coat, tubules exhibit 2D (pseudo) helical symmetry and thus can continuously polymerize from the base expanding the size of the tubule. This is seen in BAR domain and retromer arch tubular coats formed both in vitro and found ‘in cell’ (Kovtun et al., 2018, Lopez-Robles et al., 2023). Tubules exhibit a higher surface area to volume ratio, allowing for a higher proportion of transmembrane cargoes compared to soluble cargoes to be incorporated in them and thus may be more common in coats whose cargo is mainly transmembrane cargoes. 1.12 Carrier scission 27 1.12 Carrier scission Once fully formed a carrier must “pinch off” the donor membrane, fully separating itself from its membrane of origin and preventing the diffusion of cargoes away from the carrier. While scission is a critical point within the biogenesis of any carrier there are a variety of mechanisms that mediate this process. In the case of AP-2 the initial bud neck is constricted by the recruitment of BAR domain proteins like SNX9, which in turn acts as a site for the assembly of the large Dynamin GTPase (Lundmark and Carlsson, 2004). This oligomerizes to form a multimeric helical assembly, which upon activation triggers a constriction of the GTPase around the membrane to drive compression of the membrane and thus cause hemi-fusion of two bilayers (Hinshaw and Schmid, 1995, Roux et al., 2006). These structures then collapse and segregate into two independent membranes. However other endosomal Adaptors proteins (AP-1, AP-3 and retromer for example) are supposedly independent of dynamin (Kural et al., 2012). Retromer has instead been proposed to utilise the force generated by the polymerization of branched chain actin to drive fission (Derivery et al., 2009), possibly in connection with action of the ATPase EHD1 (Wunderley et al., 2021). The generation of branched actin networks in mediated by the ARP2/3 complex, however requires an activator. In the case of endosomal membranes, the Wiskott Aldrich and scar homology (WASH) complex has been shown to drive this activation. The WASH complex is directly recruited to sites of retromer driven tubulation through direct bind of the Fam21 subunit of the WASH complex to VPS35 and VPS29 (Guo et al., 2024, Romano-Moreno et al., 2024). In the case of coats like the COPI complex, scission is innately driven by part of the coat itself. The dimerization of Arf1 between triads has been suggested to drive this scission process (Dodonova et al., 2015, Beck et al., 2011). Mutants to the dimer interface of Arf1 resulted in a scission, but not recruitment, defect of COPI. Although the exact nature of how this sits within the coat and creates a budding scar on a COPI coated vesicle remain yet to be visualized. 28 Introduction 1.13 Uncoating To drive the membrane deformation of a vesicle, typically the coat covers nearly the entire surface of the carrier. Yet, to enable efficient fusion of the carrier the coat must first be removed so the membrane itself is physically accessible to the target membrane. In the case of endocytic CCVs, uncoating is a multi-step process. First, the clathrin lattice must be dismantled by J- domain containing adaptors (auxillin or GAK) in concert with the chaperone HSC70 (Ungewickell, 1985, Gall et al., 2000). This is an energy intensive process in which hydrolysis of ATP by HSC70 is thought to generate sufficient force to drive disassembly of the clathrin lattice by disrupting the tripod domain of the individual triskelia. The second step is removal of the PI45P2-binding adaptors including the AP-2 complex and CALM. Dephosphorylation of PI45P2 by lipid phosphatases, such as synaptojanin and OCRL (Posor et al., 2022), is sufficient to remove all CME clathrin adaptors since PI45P2 is the only major binding determinant utilised: This process appears to happen shortly after scission. However, removal of AP-1 from intracellular CCVs is likely largely mediated by Arf1 GAPs and may not be as quick to be uncoated as endocytic CCVs (Nie et al., 2005). In the case of yeast AP-3 (discussed in more detail in section 1.19) the entire AP-3 coated vesicle docks to its target organelle prior to uncoating and fusion (Schoppe et al., 2020). 1.14 Tethering and fusion The final step of any carrier is its attachment to and fusion with its target membrane - a multi- step process with the initial capture of the vesicle by a tether and subsequent membrane fusion mediated by formation of a SNARE (SNAp REceptor) complex (Söllner et al., 1993). Tethering complexes fit loosely into two categories the extended Coil-coil family and the multisubunit tethering complexes (MST). The coil-coil tether family is prolific consisting of proteins such as EEA1 and members of the Rufy and Golgin families (van der Beek et al., 2019, Gillingham and Munro, 2003). They are typically dimeric consisting of a small domain at their N- and C- termini separated by a long coiled-coil domain of up to 1000 residues in some cases. The domains at either terminus typically bind factors such as small GTPases and or phosphoinositides and utilise the long coil to bridge between carriers and their target membranes. The multisubunit tethering complexes including: HOPS, CORVET, EXOCYST, 1.14 Tethering and fusion 29 GARP, BLOC-2, CHEVI and FERARI are in comparison, large folded protein complexes often composed of a combination of solenoids and WD repeats (Spang, 2016). In addition to these domains, they often have an associated Sec1/Munc18-like (SM) family protein, which acts as a catalyst for SNARE fusion (Baker et al., 2015). The MTS family members are often shorter in reach than the coiled-coil tethers and thus can sample less 3D space for finding their cognate carriers. As such these two types of tethers can work in a sequential manner, with a carrier first “caught” by a coiled-coil tether before being brought to a MST to mediate direct fusion. There are over 50 SNAREs known in mammals. They are type II transmembrane or lipid modified tail anchored proteins that directly facilitate membrane fusion through hetero- tetramerization via their ~16 turn alpha helical SNARE motifs. SNAREs fall into one of two categories based on the presence of a conserved Glutamine (Q-SNAREs on target membrane – sometimes termed t-SNAREs) or Arginine (R-SNAREs found on the carrier – sometimes termed v-SNAREs) within their SNARE motifs. SNAREs must be present in both the carrier (selected at the stage of cargo selection) and the target membrane. To mediate fusion these vesicular and target SNAREs must interact (facilitated by SM proteins) to form a bundle of four helical SNARE motifs: three coming from Q-SNAREs and one from an R-SNARE (Sutton et al., 1998). As the SNAREs transmembrane domains exist in different bilayers the complex is initially termed a ‘trans-SNARE complex’. The creation of this trans-snare complex brings the two opposing membranes into close enough proximity to allow for the formation of a hemi- fusion complex. As the membranes fuse the transmembrane domains of the participating SNAREs are in effect found within the same membrane and are thus now called a cis-SNARE complex. A cis-SNARE complex is incapable of mediating further fusion unless unwound by the AAA+ ATPase N-ethylmaleimide sensitive fusion (NSF) and its cofactors driven by ATP hydrolysis (Zhao et al., 2015). Importantly any combination of three Q and an R snare is sufficient for fusion in vitro, ‘in cell’ culture it is likely there is a much higher degree of specificity for combinations of SNAREs to be productive. This enhanced specificity could be due to preferential interactions of specific SNAREs with SM proteins scaffolding proteins during the formation of the initial trans-SNARE complex. This difference is crucial as the concentration and time required to form a SNARE complex in vitro is higher and longer respectively than in cell culture. 30 Introduction The number of SNARE complexes required to drive carrier and target fusion is unknown but early EM demonstrated the presence of 5, 6 and 7 SNARE complex ‘star-like’ structures with a central pore, which resemble the structures predicted (Rickman et al., 2004). This comparatively low number would result in more available 2D membrane space in which canonical cargo can be packed so enhancing the efficiency of vesicular transport. 1.15 An overview of AP complexes and their related assemblies The first scaffolding coat protein discovered, clathrin is ‘shared’ by multiple adaptors within the cell. The first adaptors to be discovered for clathrin were the two AP-1 and AP-2 adaptor complexes, which localised to intracellular membranes and the plasma membrane respectively (Robinson and Pearse, 1986). A third adaptor AP-3 was identified also at the plasma membrane but this was not a complex but a single protein running at ~180kDa on SDS PAGE and so was later renamed AP180 (Simpson et al., 1996). AP in these early cases stood for Assembly Polypeptide as they drove clathrin to assemble into polyhedric lattices at pH 7 (Zaremba and Keen, 1983). Since then, AP has come to mean Adaptor Protein and the number of members of the AP family has grown to include AP-3, AP-4, AP-5 and COPI and most recently with the addition of TSET in plants and some other eukaryotes (figure 1.6) (Hirst et al., 2014). As the AP complexes are arguably the major facilitators of cargo sorting within this thesis what follows is a brief description as to their structure and function. Fig. 1.6 Evolution of the human AP family phylogenetic tree illustrating the order in which the AP complex family diverged. 1.15 An overview of AP complexes and their related assemblies 31 Before ever being structurally resolved AP-1 and AP-2 were initially identified by co- purification (a contaminants) with clathrin and were subsequently separated by Hydroxy- apetite purification. From this it was found that AP-1 and AP-2 were assemblies consisting of small (σ), medium (μ), and two types of large subunits (a β subunit and either a γ (AP-1) or an α subunit). The AP complexes were first visualised at low resolution in 1988 first by negative stain and rotary shadowing electron microscopy (Virshup and Bennett, 1988), which noted “globular protein with one or two knob-like tails”. Later the same year, deep etch electron microscopy (Heuser and Keen, 1988) showed more convincingly that the AP-2 complex consisted of “a central body flanked by two appendages”. The central body and appendages were noted to look like the head of Mickey mouse, with face of Mickey mouse being the AP- 2 core and the ears being the appendages. This likeness of AP-2 to Mickey mouse was so memorable that the appendages became synonymously known as ear domains, a term still used in the literature to this day. Furthermore, these appendages were shown to be sensitive to elastase proteolytic treatment, leading to the hypothesis that the complex was a hexamer consisting of two copies of both the small and medium subunits which were housed in the central body while the one copy of each large subunit consisted of the scaffold of this central body and also encoded the flexible hinge and appendages (Heuser and Keen, 1988). With the benefit of hindsight, while the overall model had some merit, it is now known that the predicted stoichiometry was incorrect and that the complex is heterotetrametric with only one copy of each subunit. 32 Introduction Fig. 1.7 Early structural understanding of AP2 and clathrin (A) Early “Mickey mouse” like AP-2 particles with and without clathrin by deep etch electron microscopy (Heuser and Keen, 1988) (B) Annotated cartoon of the AP-2 complex. 1.15.1 AP-2: A model for AP family members Currently much of what is now known of AP family structurally comes from study of the AP- 2 complex and COPI and to a lesser extent of AP-1. Thus, what follows is an AP-2 centric summary of the structure and function of AP complexes. We now know that the “central” mass, subsequently called the core, is composed of the small and medium subunits and the N-terminal domains of the large subunits (Collins et al., 2002). The latter (β2 and α in the case of AP-2), consist of a curved helical solenoid interacting via their C-terminal helices. Together these solenoids form a bowl-shaped scaffold, within the crook of which the small and medium subunits are housed. When first resolved by X-ray crystallography AP-2 showed that the σ2 subunit adopted a longin domain fold and buried in the crook of the α subunit. Similarly, the N-terminal domain of μ2 subunit was reciprocally buried in the crook of the β2 subunit, while the C-terminal extended barrel like domain called 1.15 An overview of AP complexes and their related assemblies 33 a Mu homology domain (MHD) lay across the β and alpha solenoids in the bowl. However, in this first crystal structure the YXXΦ binding site on C-mu2 was obscured by packing against β2, likewise what was subsequently discovered to be the dileucine binding site was blocked by the N-terminal SKYF sequence of β2 (Owen and Evans, 1998, Kelly et al., 2008). Furthermore, the proposed four PI45P2 binding sites were not coplanar, which they would have to be if they were all to be simultaneously utilised in membrane binding. Thus, it was suggested that this structure occupied an inhibited conformation of the core that would exist in the cytosol and was thus named the “closed” conformation (Collins et al., 2002). However, upon the binding of PI45P2 the AP-2 complex undergoes a colossal conformational change driven by the C-μ2 barrel being ejected from the bowl, enabling C-μ2 to binds the membrane simultaneously and freeing the YXXΦ binding site in the process. Without the C-μ2 barrel to pack against the large subunits (α and β2), they slide apart making a more of a twisted triangular structure also liberating the dileucine cargo binding site in the process. This structure was thus dubbed the open conformation and was suggested to be the conformation adopted on membranes (Jackson et al., 2010). This was later confirmed unambiguously by cryogenic electron tomography (cryo- ET) both on membrane alone and within a clathrin cage (Kovtun et al., 2020). Subsequently a new conformation termed open+ was identified using crystallography in both native and μ2 linker Thr156 phosphorylated forms as it was initially believed to display enhanced conformational flexibility to better explore the surface of the membrane for cargo (Wrobel et al., 2019). 34 Introduction Fig. 1.8 Domain architecture and conformational change of the AP-2 complex (A) Domain architecture of the AP-2 complex. (B) Model depicting the conformation change of AP-2 and how it coordinates the acquisition of cargo and polymerization of clathrin. 1.15 An overview of AP complexes and their related assemblies 35 While by default most adaptors have been assumed to undergo a similar conformational change as AP-2, this may not be appropriate. The COPI F subcomplex adopts a much wider ‘stance’ on the membrane with the large subunits (β and γ) splayed further apart and has yet to be visualised in a closed conformation (Dodonova et al., 2015, Taylor et al., 2023). Moreover AP- 2 is the only AP complex known of that is reliant purely on phospholipid interactions (PI45P2) for causing its conformational change with AP-1 AP-3 AP-4 and COPI all partially depending on Arf1:GTP for membrane binding and to drive their conformational changes (Sanger et al., 2019). Thus, the fidelity of extrapolating our conformational understanding onto other AP complexes remains to be determined. Protruding from the C-terminal ends of the solenoids of the α and β2 subunits are large flexible linker regions known as the “hinges”. The flexibility of the hinge regions allows the appendage domains to sample a large region of space to encounter binding partners. The β2 linker binds directly via a clathrin box (LLNLD for AP-2) to the clathrin heavy chain C-terminal beta- propeller domain (Dell'Angelica et al., 1998). However, the availability of this clathrin box for binding clathrin is regulated by the core itself. In the closed conformation the hinge region including clathrin box binds back onto the core across the back of the μ2 subunit, β and α trunks (Kelly et al., 2014). Once the AP-2 is ‘opened’ by binding to membrane the hinge bind-back binding site on the core is destroyed, releasing the hinge and clathrin box to polymerize clathrin. X-ray crystallography of the C-terminal appendages of the large subunits of AP-2 revealed both α and β2 ears consisted of a closely associated sandwich and platform domains (Owen et al., 1999, Owen et al., 2000). Further work mapped the platform domains of the α appendage as the binding site (top site) for DP[FW] motifs found within the tails of other endocytic adaptors and regulatory/accessory proteins including amphiphysin, epsin and EPS15. Later it was shown that this top site was able to accommodate multiple different motifs including FXDXF (synaptojanin) and the LFGPPL from FCHO but read in the opposite direction to DxF and FxDxF (Jha et al., 2004, Zaccai et al., 2022). An additional side site exits on the sandwich domain which binds (WVX[FW]) motifs which has subsequently also shown to be vital for the binding of accessory factors including AAK1 and NECAP1 (Mishra et al., 2004). 36 Introduction In a similar fashion the β2 ear was shown to also have functional top and side sites which can interact with motifs such as the FXX[FL]XXXR helical motif of accessory adaptors ARH, β- arrestin and epsin and the short linear FXDXF motifs of EPS15 and intersectin (Edeling et al., 2006, Schmid et al., 2006). Thus, both AP-2 ears act as regulatory hubs capable of enriching other cargo adaptors and vesiculation factors into the forming CCV. Despite being crucial for the function of AP-2 and endocytosis some other AP complexes ears show some significant differences. Like AP-2 the β appendages of AP-1, AP-3, AP-5 and COPI as well as the δ (AP- 3), ε (AP-4), γ (COPI) appendages all share the same sandwich platform architecture in humans. While the AP-1 γ appendage only consists of a sandwich domain, conversely the AP- 4 β appendage only a platform domain and AP-5 ζ lacks an ear entirely (Sanger et al., 2019). In the case of AP-1 the β ear is almost identical in sequence to the AP-2 appendage and thus binds similar motifs, while the shorter γ appendage binds [WFY]XX[WYF] motifs such as those on γ-synergin and ArfGAP1 (Page et al., 1999). Likewise, both AP-4 appendages should bind short linear motifs, however, thus far only binding of Tepsin has been demonstrated, which binds to both ears (Frazier et al., 2016). In all these examples, the ear binding determinants are contained within unstructured regions (IDRs), however the γ and β ears of COPI appear to make contacts within folded domains within the coatomer complex. The β COP top site interacts with the γ cop solenoid (Core-like) and side site in binding under the α subunit (clathrin-like). However, recently cryo-ET showed that in the assembled AP-2:clathrin membrane attached complex the β2 ear sandwich subdomain is able to make a number of surface:surface interactions with the clathrin ankle region reinforcing earlier predictions that while the platform site remains free, the side site would contact the clathrin cage so inhibiting its use as a recruiting site (Kovtun et al., 2020). It is yet to be definitely established if there are any interactors that bind the ears of either the AP-3 or AP-5 complexes and if they have binding partners, whether they would be folded domains, IDRs or both. 1.15 An overview of AP complexes and their related assemblies 37 1.15.2 The AP-1 complex: Endosome to Golgi trafficking Although there has been much debate regarding whether AP-1 localises to endosomes, the trans-Golgi, or another intermediate compartment; the recent literature (Navarro Negredo et al., 2017, Shin et al., 2017, Robinson et al., 2024) (and circumstantial evidence from this thesis, discussed in chapter 3) suggests the true localisation of AP-1 is post trans Golgi and most likely to be a tubular recycling/sorting compartment with at least some endosomal characteristics. On its membrane of origin, AP-1 enriches cargo into carriers to mediate their retrograde trafficking to the Golgi stack (Robinson et al., 2024). Like other members of the clathrin adaptor family, AP-1 binds both acidic dileucine [ED]XXXL[LI]and YXXΦ motif containing cargo. Additionally, a subset of cargo with short acidic residue rich sequences, termed ‘acidic clusters’ motifs, can also directly engage a unique site of C-μ1 (Navarro Negredo et al., 2017). However, the case of acidic cluster cargo tails tested, namely Furin, CPD, KIAA0319L and CIMPR this was always in conjunction with either an adjacent YXXΦ or dileucine motif. Furthermore, phosphorylation of serine residues adjacent to Furin’s acidic cluster was shown to increase affinity of this motif for AP1, presumably through increased charge density. However, it remains to be determined if acidic clusters alone can engage with AP-1 in cell. Analogously it has also been suggested that reversible phosphorylation just outside the dileucine motifs of STING1 mediated an interaction with a positive patch of σ1 and so increased the affinity of the interaction, however this interaction was not well resolved by the EM density (Liu et al., 2022). Both of these modifications support the concept that additional specificity of some motifs may be imparted from interactions outside the characterised binding motif and may be inducible by reversible post translational modification. To augment the array of cargoes enriched into vesicles, AP-1 can directly bind other adaptors such as EpsinR and possibly Eps15 (Hirst et al., 2003). EpsinR facilitates the recycling of the Q-SNARE Vti1b in a cis-SNARE (post fusion) complex, back to the Golgi. It achieves this by using a small surface:surface interface between the SNAREs freely exposed hABC domain and epsinR’s ENTH domain (Miller et al., 2007). In line with this observation multiple SNAREs (including STX16, pertinent to this thesis) have been implicated as AP-1 dependent cargo through trafficking defects and are present in AP-1 containing isolated CCVs. Once sufficient cargo has been enriched by the AP-1 and its co-adaptors efficient membrane deformation must occur to allow for budding to form carriers. The mechanism by which AP- 38 Introduction 1 carriers form has been extensively studied and was assumed to be similar to AP-2 as it is found significantly enriched in isolated CCVs (Pearse and Robinson, 1984). Like AP-2, AP-1 also possesses a clathrin box which is required for its ability to polymerize clathrin in vitro. However multiple studies using cellular EM, live and fixed cell light microscopy have also associated AP-1 with tubular structures as well as vesicular structures (Peden et al., 2004, Theos et al., 2005, Stockhammer et al., 2023). In vitro reconstitution of the AP-1 complex on membrane showed a propensity for tubulation of membranes, which could also be observed in cell culture upon knock down of clathrin heavy chain (Hooy et al., 2022). Modelling of the tubules contacts and extrapolation of these results led the authors to suggest a hybrid model that AP-1 may tubulate membranes under the clathrin coat, with clathrin pinching off from the tip of the tubule to ensure efficient trafficking. This is also supported by the observations that clathrin stabilizes curved membranes and preferentially buds off curved or pre-deformed membranes (Cail et al., 2022). Furthermore, the observation that two Arf1-dependent AP adaptors, AP-1 and AP-3, do not appear to require dynamin for scission again suggests a key role of AP complexes carrier scission could be mediates by Arf1 or Sar1 as has been demonstrated in vitro for COPI and COPII respectively. Independently of AP-1, epsinR adaptor and clathrin, the major components of AP-1 containing CCVs, two addditional large complexes have been shown to affect AP-1- based trafficking; the HEATR5B:g-Synergin:Aftiphilin complex and the WDR11:FAM91A1:C17ORF75 complex (WDR11 complex) (Borner et al., 2012, Hirst et al., 2005, Navarro Negredo et al., 2018). Despite direct evidence for the role of both of these complexes in AP-1 trafficking their functions have yet to be well characterized at any level. g- Synergin and Aftiphilin were originally identified as interactors of the AP-1 g ear domain and was subsequently found within CCVs and to form a complex with HEATR5B (Hirst et al., 2005). The majority of both g-Synergin and Aftiphilin are predicted to be intrinsically disordered while HEATR5B is composed of a single alpha-solenoid. A recent report indicated that HEATR5B was able to directly interact with dynein tails and was sensitive to the action of dynactin (Madan et al., 2023). The same study went on to show that this interaction was necessary in drosophila for retrograde transport of AP-1 vesicles back to the Golgi, which agrees with the concept of AP-1 functioning in retrograde traffic (Robinson et al., 2024). 1.15 An overview of AP complexes and their related assemblies 39 In contrast the WDR11 complex was identified in screens for endosome-TGN trafficking defects (Navarro Negredo et al., 2018, Shin et al., 2017). However, despite the similarity in phenotype to relocating either WDR11 or AP-1 complexes, the two have yet to be shown to directly interact. The WDR11 complex was instead shown to overwhelmingly localize to the TGN through its interaction with the Golgi-tether accessory factor TBC1D23. As such it was concluded that the WDR11 complex may act to tether AP-1 derived vesicles via a hand over mechanism. 1.15.3 The AP-3 complex: Endosome to late endosomes, lysosomes and LROs The AP-3 complex, another member of the Adaptor protein family, is composed of the δ, β3, μ3 and σ3 subunits and facilitates sorting to late endocytic, lysosomal and LRO compartments (Peden et al., 2004, Simpson et al., 1997). Like AP-1, AP-3 is an Arf1 dependent adaptor that has been proposed to localized to the endosomes, TGN and tubular vesicular compartments. Although Arf1 localizes both complexes cellular imaging studies have shown that AP-1 and AP-3 display mutually exclusive localization forming their own subdomains; which while in close proximity do not merge or coalesce (Peden et al., 2004, Stockhammer et al., 2023). As is true for AP-1 and AP-2, AP-3 has been shown to bind and traffic both YXXΦ and acidic dileucine motifs-containing cargoes, targeting the cargoes for delivery to the endo-lysosomes and lysosomes or in specialized cell types, lysosome related organelles (LROs) (Marks et al., 1996, Cowles et al., 1997b, Cowles et al., 1997a). Thus, while AP-1 and AP-3 both sort away from endosomes, AP-1 functions in retrograde whereas AP-3 is anterograde. In addition to more typical cargoes, AP-3 also possesses a bespoke mechanism for sorting of the SNARE protein VAMP7 using its δ hinge to directly bind the SNARE’s N-terminal longin domain (Kent et al., 2012). In an analogous manner to the previously mentioned AP-1/VTI1b other SNARE trafficking complexes this is only possible as a post fusion complex (cis-SNARE) with another two Q-SNARE and an R-SNARE and is mechanistically similar to the way VAMP7 is internalized from the PM by the AP-2-associated clathrin adaptor Hrb (Pryor et al., 2008). Interestingly in melanosome biogenesis (a model system for the study of LROs) the mutagenesis of the δ hinge binding site for VAMP7 doesn’t affect VAMP7’s delivery to LROs. 40 Introduction However, loss of an AP-3 and the associated Biogenesis of LRO complex 1 (BLOC1) completely abolishes VAMP7 transport to melanosomes (discussed in more detail later in this section) (Bowman et al., 2021). While mammalian AP-3 possesses a functional clathrin box, the relevance of clathrin to formation of its carriers has been highly contentious. Early studies identified the mammalian AP-3 complex as having a functional clathrin box in the b3 subunit, noting its ability to bind the clathrin terminal domain (Dell'Angelica et al., 1998). Further supporting the use of clathrin, Kornfeld and colleagues separately purified bovine AP-3 and clathrin from brain lysate and were able to reconstitute AP-3 CCVs on synthetic GUVs as well as various cellular membrane fraction (Drake et al., 2000). However, unlike in AP-1 and AP-2, this clathrin box is close to the C-terminal ear domain, thus it remains undetermined if the clathrin box is obscured by adjacent secondary structures preventing any interaction with clathrin. Further, unlike AP-1 and AP-2, AP-3 only partially colocalizes with clathrin in cells by both light and electron microscopy and is not enriched in CCV purifications, further supporting a clathrin-independent role (Peden et al., 2004, Theos et al., 2005, Stockhammer et al., 2023). In pearl mice, deficient in AP3B1, the trafficking of LAMP1 to the lysosome is defective, resulting in an increase of LAMP1 trafficking via the plasma membrane (Le Borgne et al., 1998). Expression of either WT b3 or b3 lacking a clathrin box rescued this phenotype, suggesting that the AP-3 complex as a whole is required, while the clathrin box is dispensable (Peden et al., 2002). Notably, immuno-EM studies of AP-1 and AP-3 show they localise to discrete but adjacent budding sites but with clathrin only clearly present on AP-1 positive buds but not on AP-3 structures (Peden et al., 2004). Model organisms seem to add further confusion to the relevance of clathrin in AP-3 trafficking. Multiple studies on AP-3 trafficking in saccharomyces cerevisiae; often referred to as the Alkaline phosphatase pathway (ALPs pathway), state that the AP-3 complex lacks a clathrin box (Cowles et al., 1997a), and while this is true for the Apl6 (the b3 equivalent), in the subunit Apl5 (the d equivalent) there is a 746LLDLN750 motif which could function as a clathrin box. Other studies in yeast have shown AP-3 trafficking to be unaffected by loss of CLC1 (clathrin light chain) (Black and Pelham, 2000). While APs interact with clathrin heavy chain not clathrin light chain, these data would suggest clathrin is dispensable for trafficking in yeast. 1.15 An overview of AP complexes and their related assemblies 41 Nevertheless, others have shown that clathrin must be uncoated prior to AP-3 vesicle fusion with the vacuole, overtly demonstrating a requirement for clathrin (Schoppe et al., 2020). If the AP-3 complex doesn’t require clathrin for formation of a vesicle coat, how does an AP- 3 vesicle bud from the parental membrane? In an attempt to answer this question, yeast studies sought to address this question by searching for an alternative coat for the AP-3 complex, many of which settled on VPS41 (Rehling et al., 1999). By blocking the fusion of AP-3 coated vesicles, through introduction of a fusion defective vam3 SNARE, a direct interaction between Apl5 and VPS41 was identified. Sequence comparison noted the similarity in domain architecture between VPS41 and CHC1. Both consist of an N-terminal WD repeat followed by a long alpha solenoid domain. Moreover, the genetic perturbation of the AP-3 complex and VPS41 both shared similar phenotypes with defective vacuolar transport. The possibility of using VPS41 as a coat was subsequently supported by mammalian studies showing a reliance on AP-3 trafficking for biogenesis of Large dense core vesicles (LDCV), a proposed LRO, and further still that this process was also dependent on VPS41 (Asensio et al., 2013). The same study went on to show reconstitution of VPS41 alone could self-assemble to “form irregular lattices in vitro with clathrin-like properties”. However, since VPS41’s initial conception as ‘the AP-3 coat’, VPS41 has mainly been characterized as an integral component of the Homotypic fusion and vacuolar protein sorting (HOPS) complex, a hetero-hexameric complex consisting of VPS11, VPS16, VPS18, VPS33, VPS39 and VPS41 (Shvarev et al., 2022). The HOPS complex is a multi-subunit tether on the lysosome that facilitates the capture and fusion of incoming vesicles (including AP-3 vesicles) as well as other organelles such as late endosomes, autophagosomes and other lysosomes. Notably the localization of VPS41 and other members of the HOPS complex in yeast and humans shows an overwhelming colocalization to late endocytic and lysosome proteins and not with the AP-3 complex (Schoppe et al., 2020, Angers and Merz, 2009). In light of these considerations, the similarity in phenotype between AP-3 and VPS41 knock outs, noted by Emr and colleagues prior to the HOPS complex’s discovery can be reinterpreted as the HOPS complex mediating the fusion of AP-3 coated vesicles with lysosome and endolysosomes. Similarly, assembly of irregular invitro “lattices” can also be explained 42 Introduction through a combination of expressing VPS41 in the absence of other components of the HOPS complex and buffer conditions with sub-physiological (50mM) or total absence of salt, creating charge dependent aggregation of VPS41. As previously mentioned since the AP-3 complex has been localised to tubular endosomal networks by both immuno-EM and live cell imagine. However, in a recent preprint, dynamic tubules, which at least partially, associated with AP-3 were identified as being post-trans Golgi, raised the intriguing possibility that AP-3 can directly drive tubular carrier formation akin to the retromer complex (Stockhammer et al., 2023). One of the few complexes proposed to interact with the AP-3 complex is the enigmatic Biogenesis of Lysosome Related Organelles Complex 1 (BLOC1) which is also localized to tubules in cells (Di Pietro et al., 2006, Delevoye et al., 2016). BLOC1 is a hetero-octameric complex consisting of only BLOC1 subunits1-8 (BLOS1-8) all of which are single coiled-coil domains (Falcón-Pérez et al., 2002). BLOC1 is required for the delivery of cargo proteins to LROs such as melanosomes (Starcevic and Dell'Angelica, 2004). As alluded to earlier in this section, VAMP7 (in a cis-SNARE bundle) is still delivered to LROs in the absence of AP-3 but not if both AP-3 and BLOC1 are absent. Thus, it has been proposed BLOC1 can compensate this loss by delivering sufficient VAMP7 through its interaction with the Q-SNARE STX13 (Bowman et al., 2021). 1.15 An overview of AP complexes and their related assemblies 43 Negative stain electron microscopy of purified BLOC1 showed that it took up a bead on a string like architecture (Lee et al., 2012). More recent studies were able to reconstitute BLOC1 and show that it was capable of binding and tubulating membranes in vitro and claimed a direct dependence on PI4P (Jani et al., 2022, Zhu et al., 2022) (anecdotally although not shown in this thesis I have accidentally purified BLOC1, which stabilized bacterial membranes in the form of vesicles). If BLOC1 were to act as a coat with the AP-3 complex, the proposed AP- 3:BLOC1 super complex may be sufficient to tubulate membrane in the absence of clathrin. BLOC1 also further provides specificity to the fusion site of AP-3:BLOC1 tubules by interacting with the BLOC2 complex which partially resembles a multisubunit tethering complex (Di Pietro et al., 2006). Thus, BLOC1 offers a potential additional or alternative coat complex by which AP-3 may deform membranes, with or independently of clathrin while also provide specificity to a target membrane for fusion. 44 Introduction Aims of this thesis This thesis focuses on the theme of understanding the mechanistic basis by which transmembrane cargo molecules are sorted within the endocytic system. As this remains a extremely open ended objective, more specific question regarding each individual project aim to fill gaps within the current literature. The four major aims of this thesis are as follows: 1. To understand and characterize the structural basis of how TBC1D23 is able to interact with a wide array of different cargoes. 2. Why do transmembrane cargoes with the same motif not all have the same steady state localisation? What is the molecular basis for this differential sorting effect? 3. What is the structure of the AP-3 complex and how does it differ from that of other AP complexes? 4. What factors drive the recruitment of the AP-3 complex and how does it assemble on the endosomal membrane? 1.15 An overview of AP complexes and their related assemblies 45 Chapter 2 Materials and Methods 2.1 Molecular Biology All buffers for protein purification and biochemistry are listed in Appendix B (Table 8). 2.1.1 PCR Approximately 10ng of template DNA was incubated with KOD PCR mix: 0.2mM DNTPs, 1.5mM MgSO4, 0.3μM primers, 1X KOD polymerase buffer, 4% Dimethyl sulphoxide (DMSO) and 1 unit of KOD polymerase. For PCR an anneal temperature of 5 °C below lowest melting temperature of template primer was used. The thermocycler was set for 30 iterations of the following: 2.5 minutes of denaturation (95 °C), 10 seconds of annealing and 3.5 minutes of extension (70 °C). 2.1.2 Gibson assembly Amplified fragments for insertion were mixed with digested vector in a 3:1 ratio. The insert:vector mix was subsequently mixed 1:1 with Gibson master mix (New England Biolabs) and heated at 50 °C in a thermocycler for 1 hour. The Gibson assembly mix was transformed into Stellar E. coli (Clontech). 2.1.3 Transformation Competent E. coli were thawed on ice and 12-30 μl of cells was mixed with 1 μL of plasmid DNA (100ng/μL)and incubated on ice for 10 minutes. Transformation mix was heat shocked at 42 °C for 45 seconds before a further 10-minute incubation on ice. Transformed cells were then incubated with 100 μL Super Optimal broth with Catabolite repression (SOC) media (Invitrogen) at 37 °C for 40 mins. The transformed suspension was then plated on TYE plates with specific selective antibiotics. Plates were returned to the 37 °C incubator overnight. 48 2.2 Bacterial expression and culture For protein expression using Bl21(DE3)pLyS (Invitrogen) 10ml 2XTY starter cultures were supplemented with 34 μg/ml chloramphenicol and either 50 μg/ml kanamycin and/or 120μg/ml Ampicillin before being inoculated with single colonies from an overnight transformation. Starter cultures were then incubated at 37 °C shaking until an OD600 of 0.7 was reached. The 10ml starter cultures were then transferred into 1L 2XTY supplemented with 50 ug/ml kanamycin and/or 120ug/ml ampicillin. 1L cultures were incubated at 37 °C shaking until an OD600 of 0.8 was reached before induction using 0.2 mM Isopropyl b-D-1- thiogalactopyranoside (IPTG). Induced cultures were incubated shaking overnight between 16- 22 °C. The following morning cultures were pelleted by centrifugation 6240 RCF for 40 minutes. Pellets were resuspended in a relevant buffer before either immediate purification or frozen for later purification. 2.3 Bacterial protein purification Cells were resuspended in buffer A and supplemented with AEBSF hydrochloride (Santa Cruz biotechnology), 1 mM MnCl2 and 20 ug DNAse1. Cell suspensions were lysed using a cell disruptor (Constant Systems) at 30kPa, before clarification by ultracentrifugation (104350 RCF) for 45 minutes. The supernatants were separated from pellets for further purification. 2.3.1 Glutathione-S-transferase tagged purification The supernatant was batch bound to Glutathione Sepharose resin (Cytiva) for 40 minutes before washing with 400 ml of buffer A at 10 ml/min using a peristatic pump. Subsequently the bound protein was either eluted or tag cleaved off. For elution the resin was incubated in buffer A with 50 mM Glutathione and the flow through collected in 10 ml fractions. For removal of the GST tag the bound resin resuspension was incubated with 500 units of bovine Thrombin (Serva) at room temperature overnight. The following morning the resultant mixture was drained and flow through collected. 2.4 Insect expression and culture 49 2.3.2 Histidine tagged purification The supernatant was batch bound to HisPur NiNTA resin (Thermofischer scientific) for 40 minutes before washing with 400 ml of buffer A supplemented with 15 mM imidazole (pH 8), at a rate of 10ml/min using a peristatic pump. Subsequently the protein was eluted with 50 ml of buffer B supplemented with 300 mM imidazole, in 10ml fractions. The protein concentration of the resultant flow through was measured by Bradford assay (BioRad) and supplemented with AEBSF hydrochloride and 1 mM DTT. The protein was subsequently concentrated for gel filtration into an appropriate buffer for downstream use. 2.4 Insect expression and culture For general culture and protein expression Spodoptera frugiperda 9 (SF9) cells were cultured in suspension at a cell density between 1x106-3x106 cells/ml. Cultures were regularly split with Insect-XPRESS Protein-free Insect Cell Medium with L-Glutamine (Lonza) to maintain optimal cell density. For expression of the AP-3 core 500 ml of Sf9 at a density of 2.5x106 cells/ml before were infected with 5ml of baculovirus containing media. This culture was then made up to 1L returned to the incubator for 20 minutes to ensure adequate mixing of the virus and cells. The culture was then split into two 500 ml cultures. The cell density and YFP fluorescence (510nm wavelength) was monitored over the subsequent days post infection. Cells were maintained between 1x106-3x106 cells/ml post infection. Cells were harvested 3 days post arrest (approximately 80% alive:dead ratio) and stored at -80 for future use. 2.5 Insect protein purification (AP-3 core) Insect cell pellets were thawed on ice in the presence of AP-3 lysis buffer until resuspended. Cells were lysed by being passed three times through sequentially smaller needles (19G, 21G, 23G, 25G). The resultant lysate was clarified by ultracentrifugation at (104350 RCF) for 45 minutes. The supernatant was filtered and loaded onto a 1ml HisTrap Excel column (Cytiva) connected in line to an AKTA pure (Cytiva) using a Sample pump S9H (Cytiva). After loading the HisTrap was washed in Buffer A for 30 column volumes (30 ml) before a step elution using Buffer B fractionating in 0.5 ml fractions. The relevant fraction was pooled and concentrated 50 to 0.5ml for gel filtration. The sample was gel filtered into Buffer C using a Superose 6 10/300 GL column (Cytiva). Relevant peak fractions were analysed by SDS PAGE to confirm the complex was intact. 2.6 Sodium Dodecyl sulphate polyacrylamide gel electrophoresis (SDS PAGE) Samples were mixed with reducing Laemmli buffer and boiled at 95 °C for 5 minutes. Samples were then spun down at 16,000 RCF for 30 seconds using a benchtop centrifuge. Seeblue plus2 prestained ladder (Invitrogen) and samples were loaded onto 4-12% Bis-Tris NuPAGE gel (Invitrogen). In either Bolt MES or MOPS running buffer (Invitrogen) SDS page was ran at 180 V, 400 mA for 40 min with the watts allowed to vary. Gels were stained with instant blue Coomassie protein stain (Sigma-Alderich). 2.7 Isothermal Titration Calorimetry (ITC) Experiments were performed using a Nano ITC machine (TA Instruments). Prior to the experiment all buffers were degassed under constant stirring under vacuum. The ITC protein (analyte) was either gel filtered or desalted into an appropriate ITC buffer. Analyte then was concentrated to 100 μM. Lyophilized peptide (titrant) was resuspended in ITC buffer to make a 5 mM stock. The titrant stock was diluted as appropriate for optimal signal/trace. 50 μL Titrant was titrated into analyte with 20 injections of 2.4 μl each separated by 300 second intervals while stirring at 200 rpm. Experiments were conducted at 20 °C unless stated otherwise. A relevant syringe titrant-into-buffer blank was subtracted from all data. For traces which displayed measurable binding, at least three independent runs that showed clear saturation of binding were used to calculate the mean KD of the reaction, its stoichiometry (n), and their corresponding SEM values. Analysis of results and final figures were carried out using the NanoAnalyzeTM Software (TA Instruments). Both curves and trace data were subsequently exported for figure generation in Prism 5 (GraphPad). 2.8 Liposome synthesis 51 2.8 Liposome synthesis 2.8.1 Lipids stocks as solvent Lipids bought directly as solutions (Avanti polar lipids) in solvents were either diluted in chloroform to make a 1 mg/ml stock or directly aliquoted and stored under inert gas at -20°C. 2.8.2 Lipids stocks from powders Freeze dried powdered lipids with monophosphates were directly resuspended in either chloroform or a 9:1 ratio of chloroform:methanol and left to stand for 1 hour at room temperature with occasional gentle agitation. In the case of lipids like PI35P2, it is not possible to buy this predissolved in solution (low demand) and due to the multiple phosphates it is sparingly soluble if directly resuspended in chloroform methanol, thus the phosphates must be protonated to increase efficiency of solubilisation. 500ug of PI35P2 was suspended in 100 μL of 2:1 chloroform:methanol solvent. After 5 minutes 1 μL of 1 M HCl acid was added after a further 5 minutes 400 μL of 2:1 chloroform:methanol was added. The solution was warmed by hand and vortexed. The solvent was subsequently evaporated under argon to produce a film. The film was then redissolved in 500 μl of 2:1 chloroform:methanol in the absence of HCL. The solvent was again evaporated using argon gas. Before being resuspended in 500 μl of 2:1 chloroform:methanol, to create a stock solution. Stock solutions were aliquoted and stored under inert gas (argon) at -20 °C. 2.9 Unilamellar Liposome Lipids mixes were made from previously described stocks by measuring out to the desired ratio. Lipid mixes were made up to at least 100 μl chloroform, before drying under argon. During evaporation the tube was continuously rotated and heated by hand to produce a thin uniform lipid film. Lipid films were then held overnight in a constant vacuum using a rotary evaporator. The following day the lipid mixes were resuspended in the relevant buffer for 10 minutes with intermittent vortexing. Subsequently lipid mixes were sonicated in an ultrasonic water bath before extrusion. 52 For structural work crude multilamellar liposome suspensions were extruded by passing through 200 nm pore size membrane (Whatman) 21 times using Avanti mini-extruder (avanti polar lipids). For smaller diameter liposome species the extruded 200 nm liposomes were passed through a 100 nm pore size membrane (Whatman) 21 times. For even smaller liposomes the extruded 100 nm liposomes were passed through a 50 nm pore size membrane (Whatman) 21 times. 2.10 Multilamellar liposome spin down Lipid mixes were handled in a 95:5% chloroform:methanol solvent. From this an 80:20% PCPE stock was created. From this stock specific lipid mixes were aliquoted and spiked with either 10% of a specific phosphoinositide species or an additional 10% PC. Lipid mixes were dried down to a lipid film using an Argon gas stream under constant rotation. To further ensure removal of inorganic solvent the dried lipid films were held under constant vacuum for at least 2 hours. Lipid mixes were resuspended in 500μl of spin down buffer and gently vortexed for less than 10 seconds to create multi lamellar crude liposomes. 2 μg of protein was added 50 μl of crude liposome suspension and incubated at 4ºC under gentle rotation for 1 hour. Protein-lipid mixes were spun down at 16,000 RCF for 10 minutes and the supernatant separated from the pellet. Excess Laemmli buffer was added to both 20 μl of supernatant and the entire pellet, these were boiled at 95 ºC for 5 minutes before loading onto SDS PAGE. 2.11 Liposome Surface plasmon resonance (SPR) All experiments were performed using a Biacore T200 (Cytiva) at 30 ºC. Lipid films were synthesised as previously outlined. Lipid films were resuspended in 150 μl 0.3 M sucrose for 1 hour with intermittent gentle vortexing. The subsequently generated sarcosomes diluted in 1 ml of 100mM HEPES pH 7 before pelleting at 16,000 RCF for 30 minutes. The supernatant was again aspirated from the pellet before the pellet was resuspended in 250 μL of core buffer (500 mM NaCl, 50 mM Tris pH8.7, TCEP) before extrusion. The 2.12 Giant unilamellar vesicles (GUVs) 53 liposomes were extruded 11 times through 200 nm Nuclepore Track-etch membrane (Whatman) using an Avanti mini extruder (Avanti polar lipids). Extruded liposomes containing specific phosphoinositide species were bound to different lanes on an L1 chip (Cytiva) in core buffer until a response of 1000 was obtained (Lane1 PCPE, Lane2 PI3P, Lane3 PI4P, Lane4 PI35P2). For all experiments stocks of purified protein were buffer exchanged by gel filtration. For measuring affinity, the protein stocks were serially diluted in SPR buffer to make a range of concentrations for the analyte. Analyte was flowed over all lanes iteratively in increasing concentrations followed by decreasing concentrations (a low to high then high to low scheme). Data was processed using curve fitting Biacore software using an affinity fit model. 2.12 Giant unilamellar vesicles (GUVs) Dry lipid film was dissolved in 25 μL of chloroform:methanol (2:1). 10 μL of the lipid mix was spread across one ITO-coated slide (indium-tin oxide) (Sigma Aldrich). The lipid mix was dried under nitrogen stream. An electroformation chamber was created by placing a silicone gasket around the dried lipid mix and a second ITO-coated slide facing the gasket, the entire assembly was held together using a paper clamp. 650 μL of 0.5 M sucrose was added to the chamber. Plates were warmed to average transition temperature of the lipids (~65 °C) before beginning overnight electrofomation (2.5 V amplitude, 10 Hz, sinusoidal voltage). The following morning the GUVs were detached from the ITO-coated slide and resuspended in 0.5ml of 0.5M glucose and allowed to settle overnight at 4 °C. 2.13 Lipid Nanodisc reconstitution Dried lipid stocks were resuspended in Nanodisc buffer A and were iteratively frozen in liquid nitrogen and sonicated using an ultrasonic water bath. 0.1% Dodecyl-beta-Maltoside (DDM) was added to a nanodisc:lipids mix in a ratio of 1:150 for SPNW15 (and SPNW15-link-Arf1), 1:250 for SPNW25 and 1:600 for SPNW50. This mix was made up to 1.5 ml and incubated on ice for 1 hour. Biobeads (cytiva) were prewashed with 100% methanol, deionised water and Nanodisc buffer A. 300uL of biobeads were added to the nanodisc:lipid:DDM mix and left end-over-end mixing at 4 °C overnight. The following morning the liquid was separated from the biobeads using a gel loading tip and the mix was concentrated to 0.4 ml using a Vivaspin 54 concentrator column (50 kDa cut off)(Amicon). The concentrated protein was gel filtratered into an AP-3 buffer A using Superose 6 increase 10/300gl column (Cytiva). 2.14 GST pulldowns 30 μg of GST-tagged protein (Bait) was added to 30 μl of GST Sepharose resin (Cytiva) in a volume of 100 μl of buffer to create a slurry. The slurry was left to incubate with end-over-end mixing at 4 °C for 30 minutes. The beads were pelleted by centrifugation (16000 RCF for 1 minute) and the supernatant aspirated. The beads and bait were then washed to remove excess unbound bait, this was repeated three times. The beads were again pelleted and to this 100 μg of prey was added and made up to a volume of 0.5 ml. This slurry was then left to incubate with end-over-end mixing at 4 °C for 30 minutes. The slurry was then pelleted, supernatant aspirated and washed with 1ml of pulldown wash buffer (150 mM NaCl, 25 mM HEPES pH 7.2, 1 mM DTT and 2% NP40). This step was iterated a further two times. The beads were then pelleted and aspirated but resuspended in 50 μl of reducing Laemmli buffer and boiled at 95 °C for 5 minutes. The relevant fractions were then visualised by SDS PAGE to assess interaction. 2.15 Size exclusion chromatography with Multi angle light scattering (SEC-MALS) Molecular mass of the full length and truncated C-terminal PH domain of TBC1D23 was determined in solution using SEC MALS. Measurements were performed using a Wyatt Heleos II 18 angle light scattering instrument coupled to a Wyatt Optilab rEX online refractive index detector. Samples for analysis were fractionated in TBC1D23 buffer A using a Superdex S75 increase 10/300 analytical gel filtration column (GE Healthcare) run at 0.5 mL/min on an Agilent 1200 series LC system collecting UV at 280 nm before then passing through the light scattering and refractive index detectors in a standard SEC MALS format. The measured protein concentration and scattering intensity were used to calculate molecular mass from the intercept of a Debye plot using Zimm’s model as implemented in the Wyatt ASTRA software. The instrumental setup was verified using a 2 mg/mL BSA standard run of the same volume as experimental runs. The BSA monomer peak was used to check mass determination 2.16 Single particle Cryo-EM 55 and to evaluate inter-detector delay volumes and band broadening parameters that were then applied to experimental runs during analysis. 2.16 Single particle Cryo-EM 2.16.1 Cryo-EM sample preparation For the AP-3 complex in solution (CIMR). All grids were prepared using a Thermo Fisher Scientific Vitrobot Mark IV, operated at 100% humidity and 10 °C. For Quantifoil holey carbon grids (1.2/1.3 300mesh) grids were first glow discharged using an easiGlow glow discharger (PELCO) using 25 mA for 45 seconds. 3.5 μL of 0.25 mg/ml of purified AP-3 complex (with or without supplementation of detergents or nanodisc) was applied to freshly glow discharged grids and blotted for 3 second blot and a blot force of -10. This was followed by rapid plunge freezing in liquid ethane. For Quantifoil Gold AUltrafoil grids (1.2/1.3 300 mesh and 0.6/2 300 mesh) girds were glow discharged using an easiGlow glow discharger (PELCO) using 25mA for 45 seconds before being flipped and glow discharged for a second time on the opposite face. 3.5 μL of 0.25 mg/ml of purified AP-3 complex (with or without supplementation of detergents) was applied to freshly glow discharged grids and blotted for 3 second blot and a blot force of -10. This was followed by rapid plunge freezing in liquid ethane. For Quantifoil Graphene oxide coated grids (1.2/1.3 300 mesh) were directly used for sample preparation and not glow discharged. 8 μL of 0.15 mg/ml of purified AP-3 complex (with 0.05% B-Octyl glucoside) was applied to a grids and before the first blot. A second application of 8 μl of sample was applied and blotted again prior to by rapid plunge freezing in liquid ethane. Both blots were for a duration of 20 seconds with a blot force of -5. For the AP-3 complex on Liposomes and GUV membranes (MPI Biochemistry). All grids were prepared using a Thermo Fisher Scientific Vitrobot Mark IV, operated at 100% humidity and 18 °C. 56 C-flat CF-2/2-3Cu-50 grids and Quantifoil Multi A (Multiple Shape) Holey Carbon Grids (300 Mesh Gold) were glow discharged using an easiGlow glow discharger (PELCO) using 25 mA for 60 seconds before use. 4 μl of reconstituted AP-3 complex coated sample was applied to freshly glow discharged grids and blotted for 4 second blot and a blot force of 3. This was followed by rapid plunge freezing in liquid ethane. 2.16.2 Cryo-EM data collection Single particle analysis Cryo-EM of AP-3 in solution was collected using the facility at the University of Cambridge Department of Biochemistry. All data for single-particle analysis were collected on a Titan Krios G3 cryo-Electron TEM (TFS) operated at 300 keV equipped with a Gatan K3 direct electron detector. Movies were collected at a nominal magnification of ×130,000 with a resulting pixel size of 0.652 Å and a total accumulated dose of ~50 e−/Å2 at under defocus range of −1.2 to −2.4 μm in steps of 0.2 μm. Data were collected as movies automatically with EPU acquisition software (TFS). Single particle analysis Cryo-EM of AP-3 on membranes was collected using the microscopes within the Briggs department MPI Biochem. All data for single-particle analysis were collected on a Titan Krios G5 cryo-Electron TEM (TFS) operated at 300 keV equipped with a SelectrisX energy filter and Falcon4 direct electron detector. Movies were collected at a nominal magnification of ×130,000 with a resulting pixel size of 1.2 Å and a total accumulated dose of ~40 e−/Å2 at under defocus range of −0.6 to −3.0 μm in steps of 0.2 μm. Data were collected as movies automatically with EPU acquisition software (TFS). Cryo-ET of AP-3 on membranes was collected using the microscopes within the Briggs department MPI Biochem. Tilt series were collected on a Titan Krios G5 cryo-Electron TEM (TFS) operated at 300 keV equipped with a SelectrisX energy filter and Falcon4 direct electron detector. Tilt series were collected between −60 ° and +60 ° in 3-degree angular increment from 0-degree tilt angle using dose symmetric scheme in EPU Tomo5 (TFS). Images were taken with a resulting pixel size of 1.55 Å and a total accumulated dose of ~40 e−/Å2 at under 2.16 Single particle Cryo-EM 57 defocus range of −0.6 to −3.0 μm in steps of 0.2 μm. Data were collected as movies automatically with EPU acquisition software (TFS). 2.16.3 Cryo-EM data processing Single particle Cryo-EM data collected at the University of Cambridge Department of Biochemistry Cryo-EM facility was treated as follows. Dose-fractionated micrographs were motion corrected, CTF corrected, particles picked and extracted using the facility’s in-house installation of WARP. The resultant particle stack was imported into Cryosparc (Structura Bioscience). Iterative rounds of 2D classification selecting for particles of interest with noisy or junk classes discarded. The junk classes were further 2D classified to retrieve any desirable particles that may been misclassified. The particles of interest were pooled and input into a 3D ab initio reconstruction job with multiple classes (often 3-5 classes). Desirable classes were refined first by Homogeneous refinement before the resultant map being input against all particles into heterogeneous refinement. Iterative rounds of homogenous refinement heterogeneous refinement and 3D classification were used to obtain highest resolution 3D volume possible. The resultant particles and volume were then input into non-uniform refinement followed by local refinement and local CTF refinement. Single particle Cryo-EM data collected at the Max Planck institute for Biochemistry was treated as follows. Dose-fractionated micrographs were imported to Relion 5 for motion correction (MotionCorr2) and CTF correction (GCTF). A subset of 100 micrographs were manually annotated and used for training a CrYOLO picking model either single particle single picks or trajectories for filament picking. The trained model was then used to obtain picks from the full dataset. In the case of filament picking a 90% overlap between particle was initially used. Particle picks and previously processed micrographs were imported into Cryosparc V.4 (Strucutura bioscience). Particle picks were used to extract particles from the imported micrographs testing a variety of box sizes. Iterative rounds of 2D classification selecting for particles of interest with noisy or junk classes discarded. The junk classes were further 2D classified to retrieve any desirable particles that may been misclassified. 3D Ab inito proved unsuccessful as tubular volumes would “collapse” 58 into 2D projections, a common problem for hollow objects in cryosparc ab initio (this is thought to be a local minimum preferred by the stochastic gradient descent algorithm). 2.16.4 Helical reconstruction Attempts at helical reconstruction were made using Cryosparc Helix refine job using C1 and no prior helical symmetry inputs. The helical symmetry was then evaluated using the cryosparc symmetry search job. The helical symmetry parameters suggested from the previous job were used for helical reconstruction. Helical pitch values (P) (~80 Å) for the helix were measured from high resolution 2D classes using 3DMOD. Using this pitch value, a 1-Dimensional search was conducted in which non- integer values for the number of subunits (n) were used to calculate an approximate rise (z) and degree rotation (d). The calculated rise (z) and degree rotation (d) were then used for in cryosparc helical refinement job search for helical symmetry. Fig. 2.1 Helical symmetry 2.17 X-ray crystallography 59 2.17 X-ray crystallography 2.17.1 Protein crystallization Samples for crystallization were prepared as follows. Recombinant proteins were purified to the highest possible purity. Gel filtration was used as the final stage of purification for both buffer exchange and removal of aggregates. The buffers used for crystallography are stated for each sample crystallized in Table 8, however generally all adhered to the concept of addition of 1 mM DTT, 0.25 mg/ml AEFBSF and keeping the concentration of the buffer (Tris-HCl) below 20 mM. Protein samples were subsequently concentrated using 20ml Vivaspin concentrators (sartorius) to between 8-20 mg/ml (optimized for individual samples) for crystallization. Peptides (Genscript) for co-crystallization were purchased as lyophilized powders. The peptides were resuspended in crystallization buffer to a concentration of 10 mg/ml and then adjusted to an appropriate pH (7-8.5) using NaOH. Peptides were mixed with protein samples in a defined stoichiometry (optimized for each sample); often resulting in a protein:peptide ratio between 1:1 and 1:5. Protein:protein complexes were mixed stoichiometrically (1:1). Prior to crystallization the sample was centrifuged at 16,000 RCF for 10 minutes to further limit possible aggregation. Initial crystallizations hits were identified using sitting drop vapour diffusion high throughput screening. All crystallization trays for sitting drop were pipetted by Mosquito LV crystal (SPT labtech). Trays from the CIMR1-12 screens (described in Table 7) were used for initial crystallization conditions with 200nl of sample mixed with 200nl of mother liquor. Once made crystallization trays were sealed using crystal clear sealing film (Hampton research) and were kept at 16 degrees in a vibration-controlled incubator. Trays were monitored at 3-, 5- and 14- days after production for nucleation. Initial hits were optimized on a sample-to-sample basis using 96-well sitting drop trays. In broad terms a combination of screening precipitant concentration against buffer pH, dilution screens and finally additive screening (Hampton bioscience) was utilised to optimize crystals for size and morphology. 60 Once optimized in sitting drops the sample was reoptimized by hand for hanging drop crystallization. This optimization adjusted relative ratios of sample and reservoir conditions often with a relative ratio between 1:1 and 1:7 in a final volume between 4-8uL. In addition to this optimization this was often repeated but with the addition of 5% glycerol to reduce the rate of nucleation in hopes of fewer, larger and better ordered crystals. 2.17.2 Crystal mounting and cryoprotection Crystals were mounted by first fishing into myelar of fibre loops of appropriate size (0.025- 0.4mM). Subsequently fished crystals were submerged iteratively (twice) in a cryoprotectant mixture. The cryoprotectant mixture was generated by mixing of a reservoir stock with glycerol to a final concentration of 25-30% (as determined suggested by OPPFcryo) and excess of peptide if applicable. 2.17.3 X-ray diffraction data collection All screening and diffraction data collected for mounted crystals was performed at the Diamond Light Source (DLS) I04 and I24 beamlines. X-ray of a wavelength around the electronic edge of selenomethionine were used. 2.17.4 X-ray crystallography data processing Initial diffraction data was scaled, integrated and merged using the in-house pipeline at the Diamond Light Source. From a selection of Dials, Xia2 -ii, Autoproc and Autoproc+starANISO, an appropriate reflection (.mtz) file was chosen most based upon a combination of its resolution and I/sig(I) in the outer shell. All X-ray crystallography structures solved in this thesis were done so using molecular replacement in the CCP4I or CCP4I2 package. Initial solvent estimations of the unit cell were estimated using Matthew’s coefficient. From this an estimation of the number of copies of the protein within the asymmetric unit (all ensembles) were made. The predicted number of copies was used to determine the search 2.18 Alphafold2 structural predictions 61 parameters for Phaser MR. A phasing model either from previously solved structures or Alphafold2 predicted (with pLDDT values converted to B factor) was input as an ensemble and used for molecular replacement. Once a correct solution was found this model was iteratively refined and manually rebuilt using refmac5 and COOT. Once a stable Rfactor and Rfree was obtained models were optimized for fitting of Ramachandran plots and reduction of clashes and rotamer bias. 2.18 Alphafold2 structural predictions All Alphafold2 structural predictions were made with the using the Alphafold2-multimer with MMseqs2 as hosted on the ColabFold github. All predictions were run with the MSA model of MMseqs2. Alphafold2-multimer was run with 3 rounds of recycling and 1 final round of relaxation. Subsequently five output models were analysed first individually by their cognate Predicted-alignment error (PAE) plot and across all models by alignment and superimposition in UCSF chimera. 62 Chapter 3 TBC1D23: A cargo sorting tether 3.1 Introduction The Golgi is a nexus of sorting and acts as a divide between the endoplasmic reticulum and the endocytic system. The cisternal maturation provides a model which imposes a temporal order from the evolution of an ERGIC-like compartment through to a more TGN-like compartment with the final stages resulting in sorting of proteins either back to the earlier compartments or onwards to endosomes or the plasma membrane. However, at any snapshot in time there always appears to be some type of cis, medial and trans Golgi. As such, despite being so extraordinarily dynamic, a defined complement of transmembrane proteins is required to continue an efficient function of each compartment. However, membrane trafficking is an imperfect system with inevitable mis-sorting events posing a major hurdle to keeping such a complex system in homeostasis. As such the cell has evolved recycling pathways to return lost transmembrane proteins to their correct environments. Furthermore, cycling between the cell surface and internal compartments is vital for the function of some transmembrane proteins. One of the best characterized examples is that of the cation independent mannose-6-phosphate receptor (CIMPR) which is recycled between TGN, plasma membrane and endosomal compartments independently of its ligand. While at the Golgi, the CIMPR is able to bind the mannose-6-phosphate labelled lysosomal hydrolases (M6P-hydrolases). The CIMPR M6P-hydrolase complex is transported to the endocytic system, either via GGA carriers from the Golgi or first transported to the plasma membrane by bulk flow, before internalization by clathrin mediated endocytosis (CME). The incorporation of a CIMPR M6P-hydrolase complex into an endosomal compartment is vital for the release of the hydrolase. Upon maturation the endosome acidifies and in turn destabilises the interaction between CIMPR and its bound hydrolase causing the complex to dissociate. The CIMPR is then recycled back to the Golgi through its interaction with a retromer based coat complex via its WLM motif, leaving the hydrolase in the maturing endosome. The now Golgi 3.1 Introduction 63 localised CIMPR is present in a more neutral compartment and able to bind more M6P- hydrolases. As such the CIMPR exploits the variation in pH between cellular compartments to fulfil its function, of which recycling is one step. As with other trafficking pathways, endosome to Golgi retrograde trafficking thus requires some level of selectivity and therefore like other sorting events has check points to ensure faithful transport. The use of coat proteins and adaptors, such as the Retromer, ESCPE-1 or AP-1 complex, have all been shown at some level to mediate endosome-Golgi sorting and are assumed to be the main drivers of cargo specificity. Specificity of where these coated vesicles fuse however is mediated by a combination of both SNAREs and tethers. The specificity of the majority of sorting processes is thought to be through interactions with either specific lipids, GTPases or directly with coat complexes. In this chapter, I characterize contribution of TBC1D23 in the process of endosome-Golgi trafficking. I further suggest TBC1D23 acts as vesicular tethering factor which directly binds a novel sorting motif to impart selectivity to vesicles through direct interaction with cargo tails. Prior to the work in this chapter, our collaborators from the Munro lab (MRC LMB) had been investigating the mechanism of retrograde trafficking from the endosomal system to the trans Golgi cisternae. Their work had highlighted an essential role for a poorly characterized protein, TBC1D23, and its binding partner, the WDR11 complex, in this process. The conclusion of previous studies suggested TBC1D23 acted as a bridge, connecting incoming endosomal derived carriers (tubules or vesicles) to a subset of long coiled-coiled trans Golgi specific tethers, known as golgins, to facilitate carrier fusion with the Golgi (Wong and Munro, 2014). However, no equivalent of TBC1D23 has been shown to be required by the remaining golgins, so why an additional tethering component was required in this case, and what factors impart specificity to endosomal derived carriers remained unknown. TBC1D23 is unique in its domain architecture. It comprises of four regions: an N-terminal Tre2/Bub2/Cdc16 (TBC) and rhodanese domain in close proximity, followed by an ~130 residue linker separating the C-terminal pleckstrin homology (PH) domain (Shin et al., 2020). Notably only the N-terminus of TBCK/FY-1 is shares partial similarity, possessing a TBC, 64 rhodanese and kinase domains in short succession. An interesting parallel is that TBCK, like TBC1D23, also binds a coiled-coiled tether like protein, Ppp1r21/FY-2 (Quentin et al., 2023, Schuhmacher et al., 2023). However, in spite of these similarities, it is more likely that TBCK and TBC1D23 have different functions despite the high probability of both diverging from a common ancestor. Despite both TBC and rhodanese domains individually being found in several other membrane trafficking components, in TBC1D23 both domains appear to be enzymatically inactive as they lack key catalytic residues. Biochemical and cellular experiments both indicate that together these domains are necessary and sufficient for targeting of TBC1D23 to the Golgi cisternae through their interaction with tethers golgin-97 and golgin-245 (Liu et al., 2020, Shin et al., 2020). The TBC1D23 linker region is intrinsically disordered imparting flexibility and an additional ~38 nm of separation between the N- and C-terminal domains. Although this separation extends the reach of the C-terminal domain to engage with incoming carriers, the distance spanned is marginal when compared to that of golgin-97 and golgin-245 (which span ~100 nm and ~300 nm respectively). The linker has also been suggested to facilitate recruitment of the WDR11 complex though an interaction with its FAM91A1 subunit; however, the role of this complex remains elusive (Navarro Negredo et al., 2018, Zhao et al., 2023, Shin et al., 2020, Shin et al., 2017). The C-terminal PH domain is suggested to selectively capture incoming vesicles; however, the mechanism by which this recruitment takes place remains unclear. Experiments from both the Jia and Munro labs conclude that the C-terminal PH domain can bind FAM21, a component of the Wiskott-Alderich syndrome protein homology (WASH) complex (Huang et al., 2019, Shin et al., 2017, Shin et al., 2020). The WASH complex acts as an endosomal activator of the ARP2/3 complex which creates branched chain actin networks. As the WASH complex directly binds Retromer and has been characterized in the scission stage of retromer tubulation, these observations suggest the possibility that WASH is retained on the vesicles or tubules, licencing them for subsequent recruitment and fusion at the trans Golgi (Derivery et al., 2009, Helfer et 3.2 Characterizing the binding of STX16 and CPD to TBC1D23 65 al., 2013, Huang et al., 2019). However, this data is at odds with prior observations that suggest WASH is only located on the endosome and is not retained on endosome derived retromer tubules (Freeman et al., 2014). In addition to the interaction with FAM21, Jia and colleagues also suggest an additional interaction with PI4P in a similar manner to other PH domains (Huang et al., 2019) concluding that both PI4P and FAM21 can bind simultaneously to opposing interfaces on the same PH domain. Further in vitro pulldowns and mass spectrometry data from Munro and colleagues using the C-terminal PH domain as bait identified an enrichment of proteins containing “acidic cluster” motifs, notably carboxy peptidase D (CPD) and Syntaxin-16 (STX16). Subsequent GST pulldowns showed that the CPD cytoplasmic tail alone was sufficient for an interaction with TBC1D23. This observation suggested that TBC1D23 was able to directly engage with specific cargoes as a “sorting tether”. Triple alanine mutagenesis and sequence alignment revealed that in addition to the acidic clusters, a highly conserved Thr-Leu-Tyr (TLY) was also vital for interaction with the PH domain. To validate and further understand the nature of this interaction, I characterised the interaction of acidic cluster TLY motifs with the TBC1D23 PH domain. 3.2 Characterizing the binding of STX16 and CPD to TBC1D23 Recombinant GST tagged PH domain was bacterially expressed and purified (Figure 3.1) as detailed in methods section 2.3.1. To confirm the results of the previously mentioned pulldown experiments and validate that this was a direct interaction, isothermal titration microcalorimetry (ITC) was used to measure the binding of TBC1D23 C-terminal PH domain to motifs both from CPD and STX16 (Figure 3.2). This revealed that the motifs bound with affinities of 10.5 µM and 1.4 µM for CPD(1371-1380) and STX16(209-221), respectively, which is within the typical range for cargo:adaptor interactions. 66 Fig. 3.1 Purification of TBC1D23 PH domain (A) Gel filtration trace of TBC1D23 on superdex 200 10/300gl. (B) SDS PAGE of fractions from gel filtration. 3.3 Crystallization of TBC1D23:STX16 complex 67 Fig. 3.2 Binding of TBC1D23 to TLY motifs (A) Isothermal titration calorimetry baseline corrected trace and integrated heat release with curve fit of STX16209-221 peptide into TBC1D23 PH domain. (B) Isothermal titration calorimetry baseline corrected trace and integrated heat release with curve fit of CPD1371-1380 peptide into TBC1D23 PH domain. (C) Aligned TLY motifs of STX16 and CPD with their respective calculated affinities and SEM values. 3.3 Crystallization of TBC1D23:STX16 complex To elucidate the mechanism of binding TBC1D23, I attempted to co-crystallize STX16(209-221) and the murine TBC1D23 PH domain. Initial crystallization screens showed promising nucleation which was subsequently optimized by both successive dilution and additive screens before finally being reoptimized for hanging drop vapour diffusion. The final crystals were grown in 0.02 M citric acid, 0.08 M Bis-Tris propane pH 8.8 and 16% w/v PEG3350. The resultant crystals (shown in Figure 3.3.A) were single large crystals with a needle morphology, suggesting they would yield high resolution diffraction. Samples were subsequently fished and cryoprotected in the presence of excess peptide and flash cooled as described in method section 68 2.18.4. Despite being superficially promising, the crystals’ diffraction was weak and did not yield high resolution data, with the best data set collected having a nominal resolution of 3.2 Å in the outer shell (Figure 3.3.B). The diffraction pattern was indexed to space group of P 62 2 2 and phased using molecular replacement of previously published human structure threaded with the murine TBC1D23 sequence (statistics in Appendix A Table 1). Successful phasing resolved four molecules of TBC1D23 within the asymmetric unit. Fig. 3.3 Initial TBC1D23 crystallization and diffraction (A) Optimized TBC1D23:STX16 crystals grown in hanging drops. (B) Weak X-ray diffraction pattern of TBC1D23:STX16 crystals. (C) Symmetry expanded model of 3x3x3 crystals unit cell. 3.3 Crystallization of TBC1D23:STX16 complex 69 Symmetry reconstruction of 3x3x3 unit cells revealed that the protein was packed as tubes in the crystal with large solvent channels, ~147 Å in diameter (Figure 3.3.C). As these large solvent channels contain inherently disordered solvent, precipitant and cryoprotectant, they would not contribute towards X-ray diffraction, providing a possible explanation as to the limited resolution of the dataset. The diffraction pattern’s solution resolved four copies of the PH domain within the asymmetric unit (Figure 3.4.B), subsequently referred to as chains A-D. No robust density for bound peptide was observed. Chain A and D formed a dimer previously observed by (Huang et al., 2019) with either chain D or A from the neighbouring asymmetric unit respectively (Figure 3.4.C). The dimeric structure was formed by the mutual exchange of the C-terminal tail (MKVDALES) via a combination of beta-augmentation, packing of hydrophobic surfaces and charge interactions (Figure 3.4.A). Chain C failed to form the reciprocal dimer but showed no additional density suggesting no bound STX16 motif. Chain B also appeared monomeric as the tail didn’t reciprocally exchange across the asymmetric unit. However, a small cluster of weak unassigned density occupied the same surface bound by the PH domain’s tail (Figure 3.4.D and 2.4.E). This prompted the proposal that the C-terminal tails’ exchange had obscured the motif’s binding site and was favoured in the protein dense environment of a crystal. 70 Fig. 3.4 Low resolution solution of TBC1D23:STX16 complex (A) simplified schematic diagram of the P622121 asymmetric unit and its interactions with neighbouring asymmetric units. (B) Richardson diagram of the solution of the whole asymmetric unit. (C) Previously reported dimer (6JM5) that forms as a contact between asymmetric units by strand exchange of molecules A/D. (D) unbuilt density from the 2fo-fc map found in proximity to molecule B. (E) Density from the 2fo-fc map showing the strand exchange of the MKVDALES tail mediating dimerization of molecules A/D. 3.3 Crystallization of TBC1D23:STX16 complex 71 Thus, to simultaneously increase peptide occupancy and discourage unfavourable lattice formation, I truncated the C-terminal domain by removing the MKVDALES tail (Figure 3.5.A). However, to probe if the dimer was indeed a crystallographic artifact and not the physiologically relevant state, I utilised Size Exclusion Chromatography coupled with Multi Angle Light Scattering (SEC-MALS) to obtain an absorption independent calculation of mass for both the full length and shortened PH domains (Figure 3.5.B). The trace for both full length (FL) and shortened PH domain were overwhelmingly monomeric with no obvious sign of higher order oligomers. As expected, the dN/dC measured a calculated molar mass of 15.2 kDa and 14.3 kDa for the full length and shortened domain respectively, in line with the loss of just the MKVDALES tail (Figure 3.5.B). This further reassured us that our truncation had not affected the fold of the PH domains. Fig. 3.5 SEC-MALS of TBC1D23 PH domain and truncated (A) SDS PAGE of both the full length TBC1D23 PH domain and the shortened truncation. Both used as inputs for SEC-MALS (B) SEC-MALS trace of elution volume and estimated dRi mass fit for both full length TBC1D23 PH domain (black/grey) and the shortened truncation (blue/light blue). 72 This new shortened TBC1D23 showed no difference in its ability to bind STX16 by ITC (data not shown) and was subsequently crystallized in the presence of STX16(209-221). Conditions yielding crystals were optimized in sitting drops to 0.8 M Potassium phosphate dibasic, 0.1 M HEPES and NaOH pH 7.5, 0.8 M sodium phosphate monobasic and 1% 1,2-butandiol before being remade in hanging drops. These crystals, while still rod-like were more angular and had dipyramid prismatic habit compared to the previous crystals (Figure 3.6.A and 3.6.B). As with the previous crystals, these were also fished and cryoprotected in the presence of excess peptide before being flash cooled. 3.3 Crystallization of TBC1D23:STX16 complex 73 Fig. 3.6 Crystallization and diffraction of truncated TBC1D23 PH domain (A) Crystals of truncated TBC1D23 PH domain and STX16. (B) OVA view of litholoop mounted crystal and X-ray beam diameter. (C) X-ray diffraction pattern collected of crystal shown in B. 74 Diffraction data was collected at the Diamond light source I04 beamline. TBC1D23 short crystals diffracted to 2.18 Å although with severe anisotropy (Figure 3.6.C) and as such were indexed and scaled using the diamond light sources installation of Autoproc with Staraniso (statistics in Table 1). The space group was assigned as P 61 2 2 and was subsequently corrected to P 65 2 2 during phasing. The solution from molecular replacement contained two monomeric copies of the PH domain in the asymmetric unit. The previously obstructed site now showed robust density for the STX16 peptide liganding both PH domains (Figure 3.7). Fig. 3.7 Novel STX16 peptide density (A) 2Fo-Fc map of unbuilt and unrefined STX16 peptide density. (B) 2Fo-Fc map of built and refined STX16 peptide density. 3.3 Crystallization of TBC1D23:STX16 complex 75 Fig. 3.8 Structure of the TBC1D23:STX16 complex (A) Interaction of TBC1D23 (blue) with STX16 TLY motif (Gold) and STX16 refined 2Fo-Fc density (grey) (B) Electrostatic interactions of STX16 (gold) and TBC1D23 (blue) (C) Depiction of critical hydrogen bonding (green) and Beta-augmentation of STX16 (gold) and TBC1D23 (blue). (D) symmetrical dimer of STX16 (gold) and TBC1D23 (blue) formed between asymmetric units. (E) hydrophobicity plot of the binding of STX16 (gold) to adjacent hydrophobic pockets (brown) on TBC1D23. 76 After iterative rounds of building and refinement, the final structure refined to a Rfactor/Rfree of 0.20/0.25. The refined model revealed that STX16 extends the beta sheet of TBC1D23 by adopting a twisted beta augmentation with the backbone of TBC1D23 strand 5 while packing against helix 2 (Figure 3.8.A and 3.8.C). Each PH domain STX16 peptide complex also formed a C2 symmetric dimer with a symmetry copy of itself (Figure 3.8.D), using the peptide to form another beta augmentation across asymmetric units. The binding of STX16 positioned Asp209 and Asp210 in close proximity to Lys632, Lys633 and Lys634 which mediates a direct electrostatic interaction (Figure 3.8.B). This presumably aids in binding specificity as well as orienting the motif. STX16 Thr212 packs a methyl group against a hydrophobic surface kinking the mainchain of the peptide while simultaneously positioning the hydroxyl towards the solvent (Figure 3.8.E). The kinking of the mainchain appears to stabilise a strained conformation which splays apart the downstream Leu213 and Tyr214 allowing both sidechains to be buried into neighbouring hydrophobic pockets (Figure 3.8.E). TBC1D23 Lys672 is positioned to take advantage of this splayed conformation and hydrogen bond to the STX16 peptide backbone between the two hydrophobics (Figure 3.8.C). STX16 Arg216 further contributes to binding by forming additional hydrogen bonds with the backbone of strand b5 (Figure 3.8.C). The remaining residues of the STX16 peptide (GFTED) make little contact with TBC1D23 and fold up into the first turn of an alpha helix. 3.4 Validation of the TBC1D23:STX16 binding interface 77 3.4 Validation of the TBC1D23:STX16 binding interface To confirm the validity of this observed binding, a series of point mutations in TBC1D23 were designed to reduce or ablate binding of STX16. Several mutations created largely insoluble material and thus were discounted so that only mutations that didn’t affect folding were considered. These mutations were again assayed by ITC to assess the severity of their effect on STX16 binding motif (Figure 3.9). Fig. 3.9 Characterization of the TBC1D23 TLY motif binding site (A) Isothermal titration calorimetry baseline corrected trace and integrated heat release with curve fit of STX16 titrated into wild type or mutant TBC1D23 PH domain (n=3). (B) schematic diagram of point mutations on TBC1D23. 78 Fig. 3.10 Characterization of the STX16 TLY motif (A) Isothermal titration calorimetry baseline corrected trace and integrated heat release with curve fit of wild type and mutant STX16 TLY motifs titrated into TBC1D23 PH domain (n=3). (B) schematic diagram of point mutations on STX16. KKK632-635EEE was expected to reduce binding of STX16 motif through repulsion of STX16 Asp209 and Asp210. However, this completely abolished all measurable binding of the STX16 peptide, further illustrating the importance of the acidic cluster in cargo binding (Figure 3.9.A). V626D would be predicted to create a charged environment in the pocket bound by STX16 Tyr214 (Figure 3.9.B). As expected, this mutation was detrimental and abolished all binding (Figure 3.9.A). 3.4 Validation of the TBC1D23:STX16 binding interface 79 I675W placed a large aromatic ring in the pocket normally bound by the STX16 Tyr214 (Figure 3.9.B). Structural prediction suggested this would effectively fill the pocket as if it was already liganded and therefore ablate binding. As expected, binding was abolished (Figure 3.9.A). Lys672 forms a hydrogen bond between its amine group and the STX16 Leu213 carboxyl peptide bond (Figure 3.8.C and 3.9.B). Disruption of this interaction significantly weakened (~20-fold) STX16 binding (Lys687A) to 27.5 µM(+/-6.3) (Figure 3.9.A). Reassuringly, the observation that loss of a single hydrogen bond would weaken, but not ablate binding, suggested that overall destabilisation of the PH domain is unlikely validating the structure is of the physiologically correct binding site. Conversely the importance of key STX16 residues was also tested with mutation of Asp209 and Asp210 reduced binding to >200 µM. Mutations to TLY212-214 triad (T212A, L213S or Y214S) all abolished measurable binding (Figure 3.10.A). The more peripheral R216A showed significant reduction (~10-fold) (10.3 µM) in binding affinity but did not abolish binding (Figure 3.10.A). To understand the stringency of the requirements for the TLY motif, point mutations were made to the STX16 motif which were conserved in some chemical properties but with a range of sizes. 80 Fig. 3.11 Characterization the stringency of STX16 TLY motif Isothermal titration calorimetry baseline corrected trace and integrated heat release with curve fit of wild type and mutant STX16 TLY motifs titrated into TBC1D23 PH domain (n=3). T212S kept the hydroxyl moiety of the threonine and was still capable of binding if at a reduced affinity of ~30 μM (Figure 3.11). Taken together with the observation that T212A ablates binding (Figure 3.10), this suggests the presence of the hydroxyl is important but not sufficient to compensate WT binding. Although not performed, a T212V mutation would be informative about the relative contribution of the Thr212 methyl and hydroxyl groups. Leu213 was mutated to either phenylalanine (L213F) or alanine (L213A) in order to preserve the hydrophobic 3.5 Validation of new TLY motif cargo 81 interaction but understand the effect of size. L213A showed very weak binding while L213F was better able to compensate this defect (Figure 3.11). This is indicative of the peptide probably requiring a medium or large sized hydrophobic sidechain (FILMV). Tyr214 was mutated to either Phe (Y214F) or Trp (Y214W) both showed effective compensation for loss of the Tyr suggesting any aromatic ring would be sufficient (Figure 3.11). 3.5 Validation of new TLY motif cargo Utilising our predicted requirements of this motif, [TS][LIMF][YWF] I sought to validate additional cargoes from the literature and published interactomic datasets (Huttlin et al., 2015). As such, a series of new cargoes were selected and tested by ITC. Fig. 3.12 Validating the binding of novel TLY motifs Isothermal titration calorimetry results showing baseline corrected trace, integrated heat release with curve fit and calculated KD (SEM) (A) NPDC1 (B) FAM174c (C) KIAA0319 (D) Furin (E) SLAMF7 (F) CIMPR. (G) Table of aligned TLY motifs and their measured affinities. 82 • NPDC1 is a largely uncharacterised protein suggested to play a role in gastral and neuroendocrine secretion(Lin-Moore et al., 2021, Evrard and Rouget, 2005). The cytoplasmic tail possessed a TVY preceded by acidic cluster and bound to TBC1D23’s PH domain with STX16-like affinity. Sufficient replicates calculated a KD of 2.9 μM (+/- 0.9) (Figure 3.12.A), suggesting this otherwise understudied protein likely utilises the TVY for retention at or return to the Golgi. Fam147c and KIAA0319 are both transmembrane proteins which have been previously suggested to utilise a TBC1D23-dependent mechanism for endosome-to-Golgi trafficking (Shin et al., 2017). Both proteins possess a poly-acidic stretch followed by TVF or TIF respectively. • FAM174 family (A, B and C) is extremely poorly characterized with no defined function however it has been documented to steady state localise at the Golgi apparatus and FAM174A has been genetically linked to Equine metabolic syndrome (Cho et al., 2022, Roy et al., 2020). In ITC, FAM174C DSDEETVFES sequence bound TBC1D23 with ~15.7 μM affinity (Figure 3.12.B). • KIAA0319 and its paralog KIAA0319L have both been suggested to play a role in neuronal guidance and in line with this both have been strongly linked to dyslexia (Cope et al., 2005). In addition to these functions KIAA0319L has been demonstrated to be the entry receptor for human adenovirus and traffic the virus from the cell surface to the trans Golgi transiting the endosomal system (Pillay et al., 2016). At steady state both paralogs localised to the Golgi with a weak punctate signal at the endosome. The ESEFDSDQDTIF motif bound TBC1D23 with ~35 μM affinity (Figure 3.12.C) indicating this as a bona fide endosome to Golgi transported cargo dependent on TBC1D23 for localisation. • Furin is a well characterized protease involved in a variety of cleavage of receptors, hormones and viral spikes (Rehemtulla and Kaufman, 1992). Furin cycles between the Golgi, plasma membrane and endosomal membranes. Furin has a less common poly- 3.5 Validation of new TLY motif cargo 83 acidic followed by TAF which bound TBC1D23 with ~34.3 μM affinity (Figure 3.12.D). • The cation independent mannose-6-phosphate receptor (CIMPR) is a bona fide cargo of the retromer complex and is one of the best characterized receptors known to recycle between the Golgi and endosomes (Seaman, 2018). CIMPR similarly exploits the pH differences across the endocytic system to correctly localise mannose-6-phosphate tagged lysosomal hydrolases to the lysosome. CIMPR possesses an acidic cluster but succeeded by a less obvious WLM motif which has also been implicated as important for its interaction with the Retromer complex (Seaman, 2007, Simonetti et al., 2019). Despite this the predicted motif showed no binding (>200µM KD) (Figure 3.12.E). Likewise, SLAMF7 another possible cargo shared a similar lack of binding indicating that although acidic cluster is important for TBC1D23 binding, the proximity of the TLY motif to the acidic cluster may have an effect (Figure 3.12.F). Taken together this data suggests that TBC1D23 sits atop the Golgi tethers scavenging for incoming vesicles containing acidic cluster motifs which encode a novel Golgi localization signal. The conformity of a protein with any individual acidic cluster motif to the ideal consensus sequence may control the proteins’ distribution throughout the endocytic system, with higher affinity motifs being retained at the Golgi while weaker motifs may cycle between the Golgi, the plasma membrane and endosomes. 84 3.6 TBC1D23 Discussion In this chapter, I have demonstrated that the C-terminal domain of TBC1D23 directly binds to a conserved acidic cluster followed by a TLY motif (subsequently both together are referred to as the TLY motif) present in transmembrane proteins. Moreover, I have elucidated the mechanism of this interaction through X-ray crystallography, delineating the structural requirements for binding via structurally guided mutagenesis. Utilizing these criteria, I have predicted novel cargoes containing this motif, identified through either our own research or published mass spectrometry datasets, and validate the binding of their motifs as cargo and solidifying the TLY motif's status as a novel sorting motif. Given that TBC1D23 is predominantly localized to the Golgi apparatus through its interaction with either golgin-97 or golgin-245, I propose that it functions as a "sorting tether" (notably expression of just the PH domain or overexpression full length TBC1D23 will also give an additional puncate endosomal signal). This suggests that TBC1D23 acts as a mechanism for capturing incoming vesicles by directly binding to cargo motifs within the vesicles. Notably, mutagenesis of the TLY motif of cargo proteins, such as CPD, resulted in the relocalisation of the mutated cargo to the plasma membrane, while other wild-type cargoes remained unaffected. However, when TBC1D23 was mutated to abolish binding, all checked cargoes were relocalised to the plasma membrane, further supporting the role of TBC1D23 in driving cargo selection at the Golgi apparatus endpoint rather than during vesiculation, thus acting as a "sorting tether." This marks the first instance of a tethering complex directly binding cargo, rather than a GTPase (commonly a Rab) or lipid species. Yet, it remains to be established whether this mechanism is unique or if other tethers function similarly. Interestingly, many of the cargoes identified as containing TLY motifs have also been previously identified as present within AP-1 clathrin-coated vesicles (Navarro Negredo et al., 2017). This suggests that the majority of vesicles tethered by TBC1D23 are endosome-derived AP-1 vesicles, as previously suggested by (Navarro Negredo et al., 2018, Shin et al., 2020). 3.6 TBC1D23 Discussion 85 This observation has implications for the role of AP-1, as there has been significant debate regarding whether AP-1 buds from the Golgi, endosomal membrane or both. Further analysis suggests that AP-1 might affect forwards trafficking by facilitating the return of components important for anterograde trafficking, most likely components of GGA vesicles. This tentative conclusion is supported by the observation that STX16, a protein enriched in AP-1 vesicles and a TLY motif-containing TBC1D23 cargo, is one of the SNAREs required to mediate membrane fusion of GGA vesicles with endosomes (Robinson et al., 2024). Other coats besides AP-1 bud vesicles at the Golgi destined for other compartments. However, mistrafficking is a phenomenon known to be corrected by the cell, for example, through mechanisms like KDEL retrieval (Munro and Pelham, 1987). Therefore, a vesicle containing too many TLY motifs might be intercepted on its exit from the Golgi to prevent mis-sorting, provided that the motifs are accessible. The most perplexing aspect of TBC1D23 is its binding partner, the WDR11:Fam91a1:C17orf75 complex. Originally identified as an effector of AP-1 dependent trafficking of tetherin, it seemingly does not colocalize with the AP-1 complex (Navarro Negredo et al., 2018). Instead, it is suggested to have a direct role, possibly in a hand over mechanism, replacing AP-1 with the WDR11 complex which could then interact with TBC1D23, tethering the vesicle. However, the necessity of handing over an adaptor or coat remains unclear as deletion of FAM91A1 doesn’t seem to effect tethering by TBC1D23 and our proposed model of cargo-dependent tethering is almost entirely coat-independent (Shin et al., 2020, Cattin-Ortolá et al., 2024). Both patients and zebrafish models indicate that loss of any subunit of the WDR11 complex causes pontocerebellar hypoplasia (PCH), a disease characterized by the loss of well-defined structures within the pons and brainstem, as well as intellectual disability (Zhao et al., 2023). PCH is also associated with rare variants or genetic loss of TBC1D23, the components of the WDR11 complex, VPS51, VPS53, EXOCS3, VRK1 or an array of tRNA splicing enzymes (TSENS) and charging enzymes (AARS)(Feinstein et al., 2014, Gershlick et al., 2019, Huang et al., 2019, Ivanova et al., 2017, Laugwitz et al., 2020, Namavar et al., 2011). 86 Recent crystal structures of the N-terminus of Fam91a1 bound to a linker sequence of TBC1D23 confirm the binding site shown by Shin and Munro and predicted by AlphaFold2 (Shin et al., 2017, Zhao et al., 2023). This demonstrates that the WDR11 complex binds to the linker region between the N-terminal TBC-rhodanese domains and the PH domain. Mutations to this interface in zebrafish correlate with the severity of PCH. Independently, mouse models implicate loss of WDR11 in driving ciliopathy by affecting Hedgehog signalling and resulting in Kallmann disease (Kim et al., 2010, Kim et al., 2018). Both Kallmann disease and PCH result from underdevelopment of neurons in either the hypothalamus or the pons and brainstem, highlighting the importance of endosome-to-Golgi trafficking in neural development. Previous work failed to identify any homology of the WDR11 complex to other proteins by sequence alignments. However, evolutionary biologists have noted that sequences evolve faster than folds (Chothia and Lesk, 1986). This can be exploited by utilizing the AlphaFold2 EBI database, which allows for genome-wide structural searches, even for experimentally unvalidated models. By aligning proteins based on 3D structure, it is possible to identify structural similarities across an entire family, irrespective of domain assignment. Components of the WDR11 complex were split into individual domains, resulting in sub- models for each member of the complex. All models were mutated to alanine, utilizing only the backbone torsion angles for structural alignment and domain identification. Notably, WDR11's two WD repeats identified many other WD repeat domain-containing proteins, while the solenoid was further identified as a Tetratricopeptide repeat (TPR) domain. Both of these domains are widespread and found throughout all eukaryotes and some archaea. Fam91A1 was split into two domains where naturally it would be separated by a linker region. The N-terminal domain showed some structural similarity to NPRL2 and NPRL3, a component of the Gator1 complex, which acts as a GTPase activating protein for RagA (Bar-Peled et al., 2013, Shen et al., 2018). The C-terminal domain showed alignment to a variety of synthases, most commonly lipid enzymes. 3.6 TBC1D23 Discussion 87 C17orf75 was split into two domains, both of which showed structural similarity to multiple members of the DENN domain-containing family. Notably, the top-ranked alignments were to NPRL2, NPRL3, C9orf72, SMCR8, FLCN, FNIP1, and FNIP2, all of which form three similar complexes (Jansen and Hurley, 2023). Although DENN domains are best characterized for their roles as Rab-GEFs, the previously mentioned domains all form GAP complexes. NPRL2 and NPRL3 are the catalytic module of the GATOR1 complex, which acts as a GAP for RagA/B (Bar-Peled et al., 2013). FLCN, with either FNIP1 or FNIP2, forms the folliculin complex mutated in Birt-Hogg-Dubé syndrome, which has been characterized as a GAP for RagC/D (Tsun et al., 2013). C9orf72 and SMCR8 also form the C9orf72 complex, later renamed the CSW or SCARF complex, which is mutated in amyloid lateral sclerosis and frontal temporal lobe dysplasia (ALS-FTD) and has been demonstrated to be a lysosomal GAP for Arf1 (Su et al., 2020). Previous studies of the RagA/C GTPases have noted that their origins stem from an archaeal Arf1 family, suggesting that this family of GAPs likely diverged alongside the Rag and Arf1 GTPases (Anandapadamanaban et al., 2019). Although speculative, if the WDR11 complex were to act similarly to the Folliculin, CSW/SCARF, or GATOR1 complexes, it would likely act as a GAP for Arf1. After the disassembly of clathrin, the hydrolysis of Arf1 GTP to GDP would result in both Arf1 and the AP-1 complex dissociating from the membrane. This would leave a naked vesicle devoid of coating factors, with only lipids and cargo molecules, which TBC1D23 is directly positioned to bind. Presently all longin domain containing GAP complexes known feature two DENN domain containing proteins; if applied to the WDR11 complex this would suggest that the complex as it currently stands may have an additional unidentified component. 88 Fig. 3.13 Schematic diagram of the role of TBC1D23 in sorting endosomal derived carrier Summary model of the work from this thesis integrated with our current understanding from the literature. Golgi-97 (green) TBC1D23 (blue) STX16 (gold) Fam91A1 (orange) WDR11 (purple) C17orf75 (light sea foam green). 4.1 Introduction 89 Chapter 4 Differential Dileucine motif sorting 4.1 Introduction Acidic dileucine motifs are defined by the consensus motif [ED]XXXL[LI] in the case of Adaptor protein complexes (and DXXLL-termini in the case of GGAs). They are widely utilised sorting signals that were originally characterized for their role in enabling efficient endocytosis of cargo proteins. These motifs enable endocytosis of cargo by directly binding to the AP-2 complex (Heilker et al., 1996). In light of the discovery that other AP complexes and adaptors (AP-1, AP-3, AP-4 and GGAs) were capable of also binding dileucine motifs, it was further hypothesised that such a motif may also encode directions to enrich cargoes in distinct subcellular compartments (Marks et al., 1996). But while the concept of a dileucine “code” has been hypothesised and investigated, the results remain unclear (Marks et al., 1996, Sitaram et al., 2012, Sitaram et al., 2008, Doray et al., 2007). Although the minimal requirement for a dileucine motif appears to be [ED]XXXL[LI], previous studies have noted some flexibility in the stringency of this motif and suggested a role of the X residues controlling the cargoes preferred final destination and thus the cargoes’ apparent steady state localisation by differentially affecting the affinities for different AP complexes (Doray et al., 2008, Sitaram et al., 2012). Despite being a sorting signal critical for endocytosis, at a molecular level the dileucine motif is currently far more enigmatic compared to functionally similar motifs, such as the YXXΦ motif. As the YXXΦ motif bound C-μ2, the binding site was more easily mapped and C-μ2 could be expressed independently to the rest of the AP-2 complex, thus the YXXΦ motif was structurally characterized much earlier than the dileucine motif (Owen and Evans, 1998). A decade later the first structural characterization of the dileucine motif bound to the AP-2 core lacking the N-terminal SKYF sequence of b2 (known as the unlatched core) demonstrated that the binding site lay across both the σ2 and α subunits (Kelly et al., 2008), dispelling earlier “demonstrated” binding sites in the b-subunit of AP-1 (Rapoport et al., 1998). In addition to this, the structure showed how the specificity of the dileucine motif was imparted with the acidic residue coordinated by Arg21 on AP2 α and both leucines buried into adjacent hydrophobic pockets on σ2. 90 Fig. 4.1 Schematic diagram for the sorting of dileucine cargo (A) Structure of unlatched AP-2 bound to mutated CD4 dileucine sequence. (B) Diagram of dileucine motif containing proteins at steady state localisations and adaptor/coat complexes that traffics them. 4.2 Construction of a stable AP-2 σ2α hemi complex 91 Since the first structures of a dileucine motif bound to AP-2, the binding site of multiple other dileucine-like (non-canonical) virus motifs have been resolved which show similar modes of binding (Liu et al., 2022, Morris et al., 2018, Kwon et al., 2020). Despite this, the vast majority of these structures are resolved at resolutions >3 Å with the exception of a 2.34 Å structure of AP-1 bound to the phosphorylated STING1 dileucine motif (7R4H), albeit with >3.5 Å local resolution around the dileucine binding site (Liu et al., 2022). Thus, little is known about this interaction at higher resolution and a direct comparison between dileucine motif affinities and localisation of transmembrane proteins cargo has never been published. To understand how different proteins, all bearing similar motifs, have drastically different localisations within the cell, high resolution structural models and comparative binding measurements would be needed. In an attempt to better characterise the molecular mechanism of dileucine cargo binding to AP complexes, this chapter aims to understand how variation between different dileucine motifs and mutants impacts the mechanism of cargo sorting via changes in affinity with the AP-2 complex. 4.2 Construction of a stable AP-2 σ2α hemi complex As all prior work on dileucine binding had been carried out on the whole “unlatched” AP-2 core, this significantly restricted the total yield obtainable to ~20 mg from 12 litres of E. coli culture. Past expression of individual domains of AP complexes had been met with varying success, but in the case of AP-2 when expressed in either E. coli or eukaryotic cells, an incomplete core complex was either insoluble or degraded. Additionally, the binding of dileucine peptides would require two subunits to make up the interface which is similarly impacted by expression. Previous work from the Hurley and Sinning labs had managed to obtain AP-2 and COPI truncated hemi-complexes in E. coli and SF9 cultures by co-expressing the σ2 and a truncated α (or the coatomer equivalent, ζ and γ). Initial attempts to recreate truncated core constructs based on previous studies (Ren et al., 2014) found the yield to be similar to that of the complete AP-2 core which, while sufficient for crystallography, was still limiting for ITC. Addition of a linker between the C-terminus of σ2 and N-terminus of α created a single polypeptide chain, which I call S2A. The linker proved advantageous resulting in a much-improved yield of S2A, yielding 50-160 mg of purified protein from 6 litres of E. coli 92 culture (outlined in methods section 2.3). Although not empirically proven, our hypothesis is that, as the σ2 longin domain is translated, the nascent protein can either become insoluble and then subsequently degraded or bind an already present α trunk. However, the addition of a linker between the σ2 and α holds them in close proximity after translation increasing the likelihood of an interaction. As such, the α trunk effectively acts as a purpose-built scaffold for the otherwise insoluble longin domain, thereby improving the stability of the construct. This new S2A construct could be purified easily to high concentration for downstream biochemical and structural analysis (Figure 4.2). Fig. 4.2 Purification of S2A SDS PAGE of purified S2A fractions from gel filtration 4.3 Binding of an array of dileucine sequences to S2A 93 4.3 Binding of an array of dileucine sequences to S2A We first sought to understand if different dileucine cargoes had different affinities for our S2A construct. I opted to characterise this binding using isothermal titration calorimetry (ITC) as it had previously been used to study the binding of motifs to cargo adaptors in membrane trafficking and the yield of the S2A construct was high enough to allow for many repeats. For this approach, six previously reported dileucine motifs from the proteins Bace1, Cd3g, LRP9, MFSD12, PI4K2a and PQLC2 were selected and ordered as chemically synthesised peptides. BACE1 encodes the β-secretase, a protease notoriously involved with the cleavage of the amyloid precursor protein in Alzheimer’s disease (Vassar et al., 2009). Although the exact localisation of the BACE1 is disputed it is most frequently cited as localizing somewhere between the late Golgi, plasma membrane and early endocytic compartments. This observation probably corresponds to cycling between these compartments. At the C-terminus of BACE1 is a dileucine motif “DDISLL” which has been reported to affect BACE1 localisation in vitro cell culture. Cluster of differentiation 3g (CD3g) is a component of the extensively studied T-cell receptor complex (TCR) and was the first dileucine motif to be identified (Letourneur and Klausner, 1992). As the TCR is activated at the plasma membrane, when a T-cell finds an appropriately stimulating peptide bound MHC complex, the steady state localisation of CD3g is at the plasma membrane. In its extended C-terminal tail, CD3g contains a “DKQTLL” demonstrated to be required for its endocytosis. Low density lipoprotein receptor related protein 9 (LRP9) is a murine equivalent of the human LRP10. The role of these proteins is poorly understood. The understanding of LRP9 sorting is also complicated by the extended C-terminal tail which contains multiple sorting motifs. However, in the literature one of the LRP9 dileucines has stood as a model cargo motif due to its ability to bind to both AP complexes and GGAs (Doray et al., 2008). Consistent with these observations, the “EDEPLL” motif has been shown to be essential for the TGN and late endosomal localisation of LRP9 (Doray et al., 2008). 94 Major facilitator superfamily domain-containing protein 12 (MFSD12) is a poorly characterised lysosomal transporter associated with the condition cystinosis and skin pigmentation (Crawford et al., 2017, Adelmann et al., 2020). Counterintuitively, MFSD12 has been shown to import cysteine from the cytoplasm into the lysosome and melanosome. This transport was suggested to act as a “reducing currency” for the synthesis of compounds such as pheomelanin and to regulate the activity of lysosomal hydrolases (Adelmann et al., 2020). Unlike many other dileucine cargoes, the “EHTPLL” motif of MFSD12 is located in a cytoplasmic loop between TMD6 and TMD7, instead of in a terminal tail. PQ-loop repeat-containing protein 2 (PQLC2) is another lysosomal transporter which has been implicated in the pH dependent export of positively charged amino acids (His, Arg and Lys) as well as mixed disulphide compounds (including cystine) out of the lysosome (Zhu et al., 2021, Leray et al., 2021). Alongside this bona fide role as a transporter, PQLC2 has been implicated to recruit the CSW/SCARF complex in the absence of amino acids (Amick et al., 2020). Also implicated in cystinosis, PQLC2 stands in contrast with MFSD12, in that it removes the product of oxidised cysteine (such as cystine) from the lysosome to the cytoplasm. This further supports the conclusions of (Adelmann et al., 2020) suggesting that cysteine acts as a redox currency, with a blockage in influx or efflux causing cystinosis. As with other cargoes, it is localised by a C-terminal “ELEPLL” motif. Phosphatidyl-inositol-4-kinase 2 α (PI4K2 α) is responsible for synthesis of PI4P on the endosomal membranes. Several reports have claimed its localisation is dependent on the interaction of the “ERQPLL” motif with the AP-3 complex (Salazar et al., 2005). However, as far as I am aware, this is a soluble cytoplasmic protein. As such, it remains unclear as to why a cytoplasmic protein would bind a dileucine cargo binding site or why it would only bind to AP-3 complex and not the AP-1, AP-2 or AP-4 complexes. As these motifs were derived from cargoes with varying localisations, I would expect them to have a variety of affinities for the AP complexes. The motif peptides were titrated into S2A for ITC as outlined in method section 2.7. 4.3 Binding of an array of dileucine sequences to S2A 95 Fig. 4.3 Characterizing differential affinities of dileucine motifs Isothermal titration calorimetry baseline corrected trace and integrated heat release with curve fit of S2A titrated with peptide motifs of (A) BACE1 (B) CD3g (C) LRP9 (D) MFSD12 (E) PI4K2α (F) PQLC2 (G) MFSD12 and MFSD12 P(-1)G (H) BACE1 S(-1)P. (I) Aligned sequence of dileucine motifs used for ITC. 96 As expected, and in concordance with previous reports, different dileucines showed a range of affinities (Figure 4.3). These data exhibited a general trend showing a preference for Glu over Asp at (-4) and a Pro at (-1), both of which had previously been implied from studies in cells (Doray et al., 2008, Sitaram et al., 2012, Kozik et al., 2010). However, as both the Pro and Glu co-vary in our list of peptides, it is impossible from this data alone to conclude to what extent each individual residue is imparting affinity. Additional problems arise from the possibility of differences in solubilities of each peptide changing their effective concentration and consequently their calculated KD. Additionally, in retrospect, there was insufficient sampling of motifs from a variety of localizations which further confounded difficulty in interpreting differences of the motif sequence with localization. Thus, to confirm the importance of the Pro (-1), variants of MFSD12 and BACE1 motifs were produced. In the case of MFSD12 the Pro was exchanged with a Gly; this imparted significant flexibility to the motif that would normally be far more restrained. This resulted in a reduced affinity of the motif for S2A construct to below measurable for curve fitting and KD calculation (Figure 4.3.G). This result suggests that the gain of flexibility upon loss of the Pro (-1) is inhibitory for cargo enrichment into a forming vesicle. Conversely, the BACE1 motif, which lacks a Pro (-1) was exchanged from Ser(-1)Pro. As predicted, the affinity of the BACE1 mutant motif increased to ~9 μM as compared to the WT sequence (Figure 4.3.H), suggesting that a Pro (-1) is sufficient to increase the rate at which a cargo is endocytosed. 4.4 Crystallization of S2A in complex with dileucine motifs 97 4.4 Crystallization of S2A in complex with dileucine motifs To understand the molecular basis for these differences in affinity, I sought to elucidate the structure of several motifs bound to S2A by X-ray crystallography, to determine whether there were differences in interaction between the cargo and adaptor through bonds made by different X residues or the remodelling of binding site on σ2. S2A was concentrated to 14 mg/ml and supplemented with each dileucine peptide in a 1:1.5 ratio. Subsequently the S2A:dileucine complex was used for the initial screening of crystallization conditions. Of these S2A:dileucine complexes, LRP9, MFSD12, PI4K2a, BACE1 and CD3g yielded initial hits. These hits were subsequently optimised by a combination of dilution and additive screens before being regrown in hanging drops to yield larger diffracting crystals. These crystals were cryoprotected, flash cooled and mounted as outlined in method section 2.18.2. The raw diffraction images were scaled, indexed and merged with the Diamond Light Sources in-house installation of Autoproc, Dials Xia2. The S2A:dileucine complexes crystallized in a variety of space groups and a range of resolutions, (statistics in Table 2 and 3 in Appendix A). A Matthews coefficient was used to estimate the number of copies of S2A per asymmetric unit which guided parameters for solution of the diffraction patterns by molecular replacement. Molecular replacement was run with Phaser MR searching for a specified number of S2A copies (ensemble1) using a coordinate model derived by truncating from previously published open AP-2 structure (2XA7) in the absence of a ligand. Once a correct solution was found by Phaser MR, iterative rounds of refinement (refmac5) and manual building (COOT) were used until a stable Rfactor/Rfree was obtained for final structures. All structural solutions showed unassigned density for the dileucine peptide in the previously reported binding site across σ2 and α from “unlatched” AP-2 core (Kelly et al., 2008), into which the dileucine motifs were built before additional rounds of refinement. 98 Fig. 4.4 Structural comparison of S2A dileucine binding Close up of S2A (sigma2 in cyan, α blue) binding to the dileucine peptides (gold) from (A) LRP9 (B) PI4K2a (C) MFSD12 (D) CD3γ (E) BACE1 (F) Superimposition of dileucine peptides backbone forming warped hook-shape. 4.4 Crystallization of S2A in complex with dileucine motifs 99 The best LRP9 containing crystals diffracted to 2.9 Å in space group C 1 2 1. Within the asymmetric unit, there were five copies of S2A, all coordinating LRP9 dileucine motifs. Although this is intermediate resolution, four of the five peptides could be built with confidence. The LRP9 peptides appeared to take up a similar position with the Leu(0), Leu(+1) and Pro(-1) adopting identical conformations. Likewise, the acidic residues at -2, -3 and -4 all pointed in a similar direction across all LRP9 peptides. However, there was some degree of flexibility, with regards to the placement of their carboxylic acid group (Figure 4.4.A). The resolution of diffraction obtained from PI4K2α containing crystals was 2.6 Å. This was only solvable in the P1 space group with eight copies of S2A to the asymmetric unit. The density around the peptides was generally harder to interpret than that of the LRP9 crystals and only six of the eight copies of S2A apparently coordinated the peptide. As previously, the Leu and Pro (+1, 0 and -1) were positioned identically between models while the -2, -3 and -4 were more varied. In particular, the Arg (-3) was poorly resolved suggesting it was highly mobile within the crystal as shown by (Figure 4.4.B). Unlike LRP9 however, the residues upstream of the core motif (essentially -5) were visible, adopting a variety of conformations but seemingly making no additional interactions. The crystals containing MFSD12 diffracted to a resolution of 1.7 Å and the structure was solved in space group P 1 21 1. Two copies of the S2A construct were resolved within the asymmetric unit, both liganding an MFSD12 peptide. At this much higher resolution, the placement of side chains within the crystal could be determined with high confidence. As with previous dileucine motifs, little difference was observed between them aside from flexibility of the (-3) residue. It should also be noted that the His (-3) and Glu (-4) were less well resolved in (Figure 4.4.C). Again, no additional interactions were observed between side chains and the adaptor. Like the PI4K2A containing crystals, the CD3g containing crystals diffracted to 3 Å and the structure was solved in P1 with eight copies of S2A to an asymmetric unit. From these eight S2A, only two showed any buildable density for a CD3g dileucine. While the density was on the whole much less defined, Leu (+1, 0) and Thr (-1) occupied similar positions to the previous peptides with Thr side chain pointing away from the adaptor. The Gln at (-2) was resolved in 100 two conformations: the first similar to that resolved in previous structures where the Gln projects away from the mainchain, and the second where the side chain reached over towards (-4) in (Figure 4.4.D), although again no additional contacts were made. The Lys (-3) and Glu (-4) appeared to adopt similar conformations to those previously found for previous peptides with Lys (-3) projecting away from the Adaptor and Glu (-4) pointing towards α Arg21 (Figure 4.4.D). Diffraction of S2A:BACE1 crystals was also limited to 3 Å resolution. The complex was solved in the space group P 43 21 2 and contained two copies of the complex per asymmetric unit. Both peptides bound in nearly identical fashion to previous structures with Leu (0) and Leu (+1) bound in their respective pockets; Ser (-1), Ile (-2) and Asp (-3) all pointing away from the adaptor and finally Asp (-4) projecting pointing towards AP2 α Arg21 (Figure 4.4.E). As comparison of several dileucine motifs revealed little difference in side chain interaction, the overall conformations of all peptides were compared. From all the previously discussed structures, the best resolved peptides from each of the LRP9, MFSD12, PI4K2a, Bace1 and Cd3g structures were compared by superimposition of the σ2 subunit. This comparison showed that all peptides adopted nearly identical backbone conformations, forming a warped hook shape (Figure 4.4.F). This observation suggested that there was an ideal conformation assumed by a dileucine motif in binding an AP complex. Additionally, as previously discussed, the sidechains of the X residues all pointed away from the adaptor showing no obvious contacts with the adaptor. 4.5 A closer examination of S2A:MFSD12 and the ordered water network 101 4.5 A closer examination of S2A:MFSD12 and the ordered water network As the S2A:MFSD12 crystals diffracted to 1.7 Å, the structure allowed not only for greater certainty of sidechain positions and interactions but also revealed additional water molecules in and around the dileucine binding site (Figure 4.5). As previously published, the LL (0,+1) packed into a hydrophobic pocket on σ2; this interaction is crucial for binding (Kelly et al., 2008). The Glu (-4) projected towards α Arg21 forming a direct electrostatic interaction. His (-3) sticks up away from σ2 forming a crystal contact with an S2A in the neighbouring asymmetric unit, but with no obvious interaction with its bound adaptor. The Thr (-2) sidechain lies along the plane of the peptide making no contact to the adaptor. The Pro (-1) rigidifies the motif by sterically restricting the backbone and lowering flexibility. This also locks the F torsion angle of the upstream peptide bond, positioning the carbonyl downwards towards the adaptor. Despite the proline having been shown in multiple studies to affect the trafficking of cargo (Doray et al., 2008, Sitaram et al., 2012), there were no additional contacts with the adaptor, raising the question of why dileucine cargo bearing a “PLL” sequence are preferentially endocytosed. 102 Fig. 4.5 High resolution structure of S2A:MFSD12 complex (A) Orthoslice of S2A:MFSD12 complex. (B) Cartoon diagram of S2A:MFSD12 complex. (C) Close up of MFSD12 bound to S2A:MFSD12. 4.5 A closer examination of S2A:MFSD12 and the ordered water network 103 The role of ordered water molecules in scaffolding interactions becomes visible with the higher resolution data as molecules that had previously been unresolved throughout the structure could now be reliably placed. Notably several water molecules appeared in and around the dileucine binding site, as well as in the interface between the σ2 and α subunits. When analysed, several of the water molecules in the dileucine binding site appeared to make crucial hydrogen bonds with both the dileucine motif and the adaptor, effectively scaffolding the interaction (Figure 4.5). Water molecules assumed critical for binding were numbered 1-5 as shown in schematic (Figure 4.6). Water molecules 1 and 2 appear to coordinate the peptide bond carbonyls of the (-5), (-4) and (-2) residues as well as the residues of AP-2 σ2 (Figure 4.6). This causes the peptide to curve into the warped hook shape previously observed for all dileucine peptides in (Figure 4.4.F). Water molecule 3 connected a separate water network to the carbonyl backbone of residue (-6) suggesting that length and accessibility may play a role in cargo affinity (Figure 4.6). The backbone of L (+1), conserved across all peptides, also formed hydrogen bonds to water molecules 4 and 5 which presumably aids binding of the motif. Although prior data did not offer an explanation for why there was an increased affinity for Pro (-1) containing dileucine motifs, in light of this new water network an explanation can be proposed. Pro imparts significant inflexibility to the peptide backbone directly locking the F torsion angle of the upstream (-2) peptide bond; this conformation orients the (-2) carbonyl downwards towards the water molecule, in an ideal position for hydrogen bonding. Although dynamics can’t be directly observed within a crystal structure, one can hypothesise that this decreased flexibility imparted by Pro (-1) results in an increased frequency with which water 1 is coordinated by the dileucine motif. Conversely, this suggests more flexible residues, such as Gly, are disfavoured within the motif as demonstrated in (Figure 4.3.G), thus offering a possible insight into the mechanisms of preferential dileucine cargo sorting. 104 Fig. 4.6 Mapping the dileucine water network of S2A Schematic diagram of hydrogen bonding water network mediating dileucine binding. Sigma2 (cyan) Α (blue), water molecules (red), hydrogen bonds (green) and dileucine peptide (gold). 4.6 Characterizing the X residues of MFSD12 105 4.6 Characterizing the X residues of MFSD12 To better untangle the contribution of individual residues to sorting, a series of variant peptides were synthesised to the MFSD12 dileucine sequence. This approach had the advantage of being more comparable than the previous set of ITC data as it limited the effects of different solubilities, lengths and sequences of various peptides. The series of mutations chosen had either obvious structural effects or had been proposed to have effects from the literature. These mutants were again characterised by a combination of ITC and X-ray crystallography. Both cell biological experiments and biochemistry had previously indicated that a dileucine motif only bound if position (0) was a leucine but there was less stringency for the second leucine (+1) which can also accommodate an isoleucine. Thus, to confirm these interpretations in our assay, WT MFSD12 as well as a single (LI) and double (II) isoleucine mutation were tested by ITC (Figure 4.7.A). Reassuringly, the binding of WT and I (+1) variants gave a similar KD (~20μM) while Ile Ile (0, +1) abolished all measurable binding. co-crystals of S2A:MFSD12I) grown in the same conditions as the WT MFSD12 motif yielded crystals very similar to the WT. These new crystals diffracted to 1.9 Å and showed little variation in either the peptide or water network binding the adaptor. Ile (+1) packing its Cg1 and Cd in place of the Cd1 and Cd2 for Leu (+1) (Figure 4.7.B) (statistics in Table 4). To test the previously reported effect of the X residues on dileucine binding, the X residues were exchanged to alanine and their affinities assessed by ITC. The P(-1)A, as previously speculated, decreased the binding of the dileucine motif to S2A (Figure 4.7.C), but not to the extent as previously observed with the P(-1)G variants; the substitution of T(-2)A and H(-3)A had little no observable effect (Figure 4.7.C). These results suggested that P (-1) was the only X residue that determines the dileucine motifs affinity to AP-2 through its interaction with the ordered water network. While residues X (-2) and X (-3) which point away from the adaptor were dispensable. Crystallization of MFSD12 P(-1)S did not prevent high resolution yielding a 2.1 Å diffraction pattern (statistics in Table 4). In this structure the dileucine peptide adopted an identical conformation to the wild type and with the same water network visible (Figure 4.7.D). These observations shows that the water network is present even in the absence of the Pro (-1). 106 To further test the importance of the (-4) position I sought to compare the effects of Asp or Glu on affinity of the MFSD12 motif. This position had been previously suggested to affect the favourability of motifs for either the adaptors AP-1 or AP-3. While the difference of a between Asp and Glu is only a single CH2, the first two helices of AP-2 α (and the S2A construct), which harbour Arg21 and coordinates the acidic at (-4), appear to be the most mobile and poorly resolved even in high resolution structures (Figure 4.7.E). Thus, I sought to understand if there was a preference in motif binding between Asp or Glu at (-4) by ITC and crystallography. Unsurprisingly, there was no decisive difference in affinity to AP-2 between Asp (-4) and Glu (-4) by ITC, and X-ray crystallography (2.3 Å) showed no difference between the conformation or coordination of the two residues (Figure 4.7.F) (statistics in Table 4). Fig. 4.7 Characterizing the X residues of MFSD12 (A) Isothermal titration calorimetry baseline corrected trace, integrated heat release with curve fit of MFSD12 II (0, +1) (red) and LI (0, +1) titrated into S2A. (B) X-ray crystallography of S2A (blue/cyan) and MFSD12 LI (0, +1) (gold) . (C) Isothermal titration calorimetry baseline corrected trace, integrated heat release with curve fit of MFSD12 P(-1)A, T(-2)A and H(-3)A titrated into S2A. (D) X-ray crystallography of S2A (blue/cyan) and MFSD12 P(-1)G (gold). (E) Isothermal titration calorimetry baseline corrected trace, integrated heat release with curve fit of MFSD12 E(- 4)D titrated into S2A. (F) X-ray crystallography of S2A (blue/cyan) and MFSD12 E(-4)D (gold). 4.7 Characterizing dileucines cargo binding to AP-3 107 4.7 Characterizing dileucines cargo binding to AP-3 Due to the success of S2A construct from the AP-2 complex, I sought to understand to what extent the mechanism of dileucine binding was conserved across the AP family. Thus, I chose to create a similar construct to study the binding of dileucine to the AP-3 complex, which had previously never been structurally characterized. Moreover, this construct (subsequently referred to as S3D) would be predicted to bind the GTPase Arf1 and could be used to study this interaction at high resolution. Using similar domain boundaries as before, two constructs were designed: the first consisted of full length σ3 (1-193) followed by a 12-residue linker, then δ (1-365), a 4 residue cloning scar (AMGK) and finally a His6 tag, and the second similar but with a shortened σ3 (1-70) and with a shorter 6 residue linker to δ (1-365). 4.8 Expression and purification of S3D Both S3D constructs were expressed in E. coli and purified as detailed in method section 2.3. Unlike the S2A constructs, the S3D constructs showed multiple proteolysis events in a reproducible laddering pattern, more severely for the full length σ3 S3D constructs. Degradation of both constructs was substantially reduced but not totally eliminated by the addition of 100 μl of fresh 4-(2-aminoethyl) benzenesulfonyl fluoride hydrochloride (AEBSF) (25 mg/ml) at every step of the purification. Due to the shorter S3D being on average less proteolyzed during expression and purification, this construct was used for subsequent experiments. Despite the inherent degradation, S3D seemed to remain relatively intact and travel as a single species by gel filtration (Figure 4.8) suggesting that the complex was still folded and possibly functional. 108 Fig. 4.8 Purification of S3D (A) Gel filtration trace of S3D on superdex 200 10/300gl. (B) SDS PAGE of fractions from gel filtration. 4.9 Dileucines have higher affinity for S3D compared to S2A To confirm functionality of the protein, the purified S3D was concentrated to 100 μM (6.2 mg/ml) for ITC. Titration of the previously characterized MFSD12 dileucine peptide motif with S3D showed remarkably higher release of energy than the same concentration of peptide with S2A. Thus, the concentration of peptide was lowered to give a finer sampling of the curve and a more accurate curve fit of the data (Figure 4.9). From three independent experiments, MFSD12 was calculated to have a KD of 3.6 μM (+/-0.26). This is strikingly stronger than the previously measured ~24 μM binding of the same peptide to the S2A construct. There were no obvious differences in the in structure of AP-2 and AP-3 that would offer explanation for an ~6.5-fold stronger affinity. 4.10 Structural basis of dileucine binding to S3D 109 Fig. 4.9 Binding of MFSD12 dileucine to S3D Isothermal titration calorimetry results showing baseline corrected trace, integrated heat release with curve fit and calculated KD (SEM) of MFSD12 peptides titrated into S3D. 4.10 Structural basis of dileucine binding to S3D Thus, to better understand the differences between the binding of MFSD12 to AP-2 and AP-3, I attempted to crystallize an S3D:MFSD12 peptide complex. S3D was concentrated to 11.3 mg/ml (182 μM) and mixed with MFSD12 in a 1:1.1 ratio before screening for crystals. One crystallization of the conditions yielded small crystals in under 2 hours (reagent alcohol and bicine pH 8). Thus, the condition was recreated and optimized by diluting the precipitant (reagent alcohol) which slowed initial nucleation from 2-5 hours to 1 day and yielded larger, more uniform crystals. Subsequent optimizations trialed range of pHs and buffers (Tris, bicine and HEPES) against reagent alcohol as well as individual alcohols (methanol, ethanol and isopropanol). Several conditions were reoptimized for hanging drop vapor diffusion which yielded large single crystals with a plate-like morphology. The most promising crystals were cryoprotected and flash cooled as outlined in method section 2.18.2 before screening and data collection at the Diamond Light Source I04 Beamline. Data was collected with full 360 ° of rotation at a wavelength of 0.99987 Å. The crystal that yielded the best data set was grown in 20% methanol and 0.1 M Tris-HCl pH 8.6 which with mild anisotropy diffracted to 2.3 Å (Figure 4.10.A). This dataset was integrated, scaled and merged with the Diamond Light Source’s in-house installation of autoPROC+STARANISO. 110 The crystal was solved by molecular replacement in C 2 2 21 using a threaded model of the sequence of AP-3 σ3 and δ on the backbone of the previously solved S2A construct. The solution in agreement with the Matthew’s coefficient showed a single copy of S3D in the asymmetric unit. Due to differences in angle between the helices in S2A (α) and S3D (δ), the initial phasing model solution required significant rebuilding and refinement which was iterated until all density was satisfied. The presence of unassigned density around the dileucine binding site allowed for unambiguous building of the MFSD12 peptide. The final structure refined to with a Rfactor/Rfree of 0.20/0.23 indicative of a correct solution (statistics in Table 5). The structure showed a typical AP architecture with the longin domain of σ3 wrapped in the solenoid of the δ-trunk. Unlike S2A, the helices of the δ-trunk in S2D appeared to be less regular and not as evenly ordered; in particular, the helices 9-11 were splayed apart creating a large hydrophobic groove between them (Figure 4.10.B). The acidic dileucine binding site appears highly conserved between the AP-2 and AP-3 with MFSD12 adopting a similar hook shaped conformation as previously observed in both this thesis and the literature (Figure 4.10.C). Both Leu (0, +1) buried into respective hydrophobic pockets (Figure 4.10.C) with their backbone coordinating the equivalent of water molecules 3 and 4 bridging an interaction with σ3. Pro (-1) coordinated a water molecule 1 which in turn scaffolded hydrogen bonding to the peptide backbone in upstream residues (Figure 4.10.D). Both the Thr (-2) and His (-3) peptide backbone formed identical hydrogen bonds to σ3 as seen in the S2A, while the side chains made no meaningful contribution to binding as they projected out towards the solvent. Glu (- 4) made a direct electrostatic interaction with δ Arg26 (Figure 4.10.D). This similarity offered no real insight into the difference in affinity for the same dileucine motif between the two adaptors. 4.10 Structural basis of dileucine binding to S3D 111 Fig. 4.10 Crystallization and structural solution of S3D:MFSD12 complex (A) X-ray diffraction pattern of S3D:MFSD12 crystals. (B) Hydrophobic groove between δ helices 9-12 in the context of the S3D. (C) Structural overview of the S3D:MFSD12 complex depicted as orthoslice. (D) Interaction of MFSD12 and hydrogen bonding network bound to S3D. 112 When comparing the hydrogen bonding networks of both S2A and S3D constructs, one striking difference was in the coordination of the water molecule network (Figure 4.11). The interactions of water molecules 3 and 4 was relatively well conserved. Likewise, water molecules 1 had similar interactions but water molecule 2 was absent entirely. S3D water molecule 1 as previously mentioned formed the same set of hydrogen bonds as with S2A, coordinating the dileucine peptide into the hooked conformation and effectively acting as a “glue” between the peptide and adaptor. However, water molecule 2 in S2A was replaced with the hydroxyl of σ3 Thr70 sidechain in S3D (Figure 4.12). Thr70 creates a direct interaction between water molecule 1 and the adaptor resulting in one less water molecule being required for binding of the dileucine motif, which in turn is more entropically favorable (due to a lower entropic penalty than S2A as an extra molecule of water is contributed to bulk solvent). The presence of water molecule 2 in S2A is therefore due to σ2 Gly68 which is structurally equivalent to S3D σ3 Thr70. Fig. 4.11 Mapping the dileucine water network of S3D Schematic diagram of hydrogen bonding water network mediating dileucine binding. σ3 (grey), δ (light blue), water molecules (red), hydrogen bonds (green) and dileucine peptide (gold). 1 4 5 -2 -1 -3 -4 -5 0 +1 4.10 Structural basis of dileucine binding to S3D 113 Fig. 4.12 Structural comparison of the binding of dileucine motifs by S2A and S3D (A) Close up of coordination of water molecule 1 by S3D. (B) Close up of coordination of water molecule 1 and 2 by S2A. 114 4.11 Characterizing the water network of S3D As the coordination of water molecule 1 was the only major structural difference between S2A and S3D, I attempted to test the contribution of Thr70 to the binding of dileucine peptide. Thus, I generated S3D construct with T70G to remove the Thr and mimic the S2A binding site. If Thr70 was mediating the shift in affinity from this change, I would expect that the affinity of MFSD12 dileucine would shift from the previously measured ~3.6 μM KD of S3D to the lower ~24 μM binding of S2A. Both the S3D (WT) and S3D (T70G) were expressed and purified in parallel, before being assayed for binding by ITC consecutively (using the same aliquot of peptide). As predicted the ITC showed a clear difference between the two constructs with the T70G mutation reducing the affinity to ~23 μM (+/-2) (Figure 4.13). This is in line with the T70G mutation mimicking the water molecule environment of the binding site for AP-2 and demonstrates that hydrogen bonding of water molecule 1 by T70 is causative for the higher affinity. Fig. 4.13 Binding of S3D T70G to MFSD12 dileucine motif Isothermal titration calorimetry results showing baseline corrected trace, integrated heat release with curve fit of MFSD12 peptides titrated into S3D (black) and S3D T70G (navy). 4.11 Characterizing the water network of S3D 115 Taken together, the previous observation from literature that stronger affinity cargoes are preferentially sorted by an adaptor and that the Pro (-1) increases affinity to the adaptor by coordinating water molecule 1 more frequently, this observation would logically lend itself to the model that dileucine motifs bearing the Pro (-1) are favoured cargoes of the AP-3 complex and thus when at an endosome are preferentially sorted onwards to late endosomes, lysosomes and LROs. However, I can’t conclusively draw this conclusion in the absence of affinity data for the AP-1 complex. Modelling of the other AP complexes allows for prediction of how the water network may vary between AP complexes and thus their affinities for a dileucine cargo. Notably the AP-1 σ1 complex has a serine (S64) in place of σ3 Thr70 and σ2 Gly68, which could form a similar hydrogen bond as Thr70 of σ3 (Figure 4.14.A); however, I would hypothesise this may have a weaker affinity than σ3 as serine has a wider distribution of favorable rotamers, whereas threonine is far more restricted. The AP-4 complex has yet to be experimentally demonstrated to bind dileucine motifs. In place of σ3 Thr70, σ4 possesses an alanine (Ala64) (Figure 4.14.D). While Ala64 is not polar and thus can’t hydrogen bond to water molecule 1, the methyl group hydrogens occupy the same position of water molecule 2 (preventing its binding). This also breaks any potential link to the neighboring water network coordinated by Glu100 found in AP-2 (Figure 4.14.B). Furthermore, while σ4 Asp92 is capable of coordinating water molecules 3 and 4, Arg97 is too long to provide additional hydrogen bonds. Taken together, these observations suggest that while AP-4 has the potential to bind dileucine motifs, it is likely that they will have a weaker affinity than that of other AP complexes, although experimental testing would be required to confirm this. Finally, while AP-5 was modelled its dileucine binding site was severely different to other AP complexes and lacked the Leu (0) pocket. Thus, it was concluded that the AP-5 complex can’t bind dileucine motifs at this site. 116 Fig. 4.14 Dileucine and water environment across AP complexes proposed binding of dileucine motifs to (A) AP-1 (AF-2 modelled), (B) AP-2 (experimental crystal structure), (C) AP-3 (experimental crystal structure), (D) AP-4 (AF-2 modelled). 4.12 Dileucine discussion 117 4.12 Dileucine discussion In summary, in this section I aimed to better understand the molecular underpinnings of dileucine sorting. From this study I find that despite different dileucine motifs harboring different affinities to the same AP complex in vitro, all visualized by X-ray crystallography adopt a similar warped-hook conformation and similar interactions. High resolution structure of the MFSD12 dileucine motif revealed a previously undescribed network of water molecules that mediate the binding of dileucine motif to the AP-2 complex. Structurally guided mutagenesis supported by the literature shows that this water molecule network is actively utilized to preferentially select dileucine motifs with Pro (-1). However, in contrast to the current literature, no role was seen for the X residues at the (-2) and (-3) positions. Attempts to understand the extent of the conservation of this unique mechanism of cargo binding revealed MFSD12 dileucine motif bound to the AP-3 complex with higher affinity. Structural comparison revealed that the AP-3 complex coordinated the water molecule shared by the proline differently from AP-2 (by substituting a water molecule with Thr70). Mutagenesis of Thr70 to Gly in AP-3 showed complete reversal of affinity back to the affinity of the AP-2 complex. Thus, this water network actively drives the preferential binding and endocytosis of [ED]XXPLL dileucines. Furthermore, this water network is exploited by the AP-3 complex to select [ED]XXPLL for onward trafficking to lysosomes and lysosome related organelles. Finally, this model has several implications for the differential sorting of dileucine cargo. From these data, I hypothesise that despite previous reports of dileucine motifs encoding a discrete end point localization (positive selection), affinity measurements reveal dileucines interacting with AP complexes have a purely “get away from the plasma membrane” signal (negative selection). This explains why dileucines with weaker affinities, like CD4, CD3 and BACE1 are restricted to the plasma membrane and early endocytic system, while cargoes with stronger affinities like MFSD12, PQLC2 and tyrosinase exhibit a late endosomal, lysosomal and LRO localization. Thus, the proline (-1) encodes a “depth control” for how far a cargo is to be found from the plasma membrane. 118 Fig. 4.15 Schematic diagram for an affinity sorting model of dileucine cargo. Moreover, this is the first instance of water molecules being key features in the binding of a sorting motif. In retrospect this is not surprising. As all cargo motifs are short linear sequences, in order to generate significant affinity to their adaptor, their binding often utilises their backbone to create many hydrogen bonds. In the case of YXXΦ, NPXY, TLY, KKXX, KXKX and WLM/ELYLL motifs, the peptide backbone beta-augments or becomes pseudo alpha- helical in order to maximise hydrogen bonding with its cognate adaptor (Owen and Evans, 1998, Zhang et al., 1997, Cattin-Ortolá et al., 2024, Ma and Goldberg, 2013, Simonetti et al., 2019). However, in the case of the previously resolved CD4 dileucine (Kelly et al., 2008), the majority of the backbone makes few hydrogen bonds at all, suggesting there was a missing component to the interaction, which we now know to be the water network. Late endosome Lysosome LRO Endosome AP-3AP-1 AP-2 Golgi [ED]XXAL[LI] [ED]XXPL[LI] 4.12 Dileucine discussion 119 Of note, although this is the first documentation of a water network coordinating the dileucine motif, I am not the first to observe it. The density for these water molecules can be clearly discerned in previously published maps for 4NEE, 4P6Z and to a lesser extent 7R4H (Ren et al., 2014, Jia et al., 2014, Liu et al., 2022). This density was simply not reported as water molecules and were not built in these aforementioned mentioned structures. 120 Chapter 5 Structural studies of AP-3 appendages 5.1 Introduction The necessity of clathrin for function of AP-3 carriers (tubules and/or vesicles) has been controversial (Simpson et al., 1997, Dell'Angelica et al., 1998). The AP-3 complex localises to non-clathrin coated buds and is not copurified with clathrin coated vesicles (CCVs). While metazoan AP-3 possesses a highly conserved clathrin box in its b3 hinge it is significantly closer to the predicted domain boundaries of the b3 ear than would be found for b1 or b2 (Simpson et al., 1996, Dell'Angelica et al., 1998). Thus, despite an ideal sequence for binding clathrin, the clathrin box may be obscured or even incorporated as part of the folded ear domain. Other AP complexes like the AP-1, AP-2 and AP-4 recruit additional vesiculation factors and cargo adaptors using the ear domains (Sanger et al., 2019). As such the ear domains serve as binding platforms interacting with short linear motifs on either the top site (platform domain) or side site (sandwich domain). To date there are no short motifs known to interact with the b3 ear domain. 5.2 X-ray crystallography of the 3-ear domain 121 5.2 X-ray crystallography of the b3-ear domain To investigate the availability of the AP-3 clathrin box I sought to crystallize the b3 ear domain. Due to a miscommunication the entire hinge-ear was expressed and purified as outlined in 2.3. The purified protein was concentrated to 15 mg/ml and screened for crystallization conditions (Figure 5.1.A and 5.1.B). Unsurprisingly, there was no nucleation noted at any point within the first two weeks, most likely due to the ~188-residue disordered hinge region preventing efficient crystal contact formation. However, by chance the plates were examined again 354 days later, where a single crystal was found having grown in 0.2 M calcium acetate hydrate with 20% w/v PEG3,350 at pH 7.5. This crystal was cryo-protected in 25% glycerol and flash cooled (Figure 5.1.C). Weak diffraction patterns were collected at the I04 beamline likely due to the size and age of the crystal (Figure 5.1.D). The diffraction pattern was indexed, merged and integrated using Diamond Light Sources in house installation of Dials, yielding a 3.3 Å dataset in space group P 21 21 21. Initial content estimation was analysed using Mathew’s coefficient, which indicated four copies of the b3 ear per asymmetric unit. The diffraction pattern was solved using molecular replacement (Phaser MR) searching four copies of the AP- 2 ear (1E42) split into two independent ensembles (the platform and sandwich domain). Initial solutions confidently placed the backbone of both the folded sandwich and platform domains. The loops between the folded domains were subsequently rebuilt into the density before one round of refinement (refmac5). After this first round of refinement, the output coordinates files were manually edited to join the cognate platform and sandwich domains into a single chain. This edited coordinate file was used for subsequent rounds of refinement and manual building. A final solution with an RFactor/RFree of 0.208/0.271 (statistics in Appendix A Table 6). 122 Fig. 5.1 Purification and X-ray crystallography of the b3 appendage (A) Gel filtration trace of b3 appendage on superdex 200 10/300gl. (B) SDS PAGE of fractions from gel filtration. (C) OVA view of litholoop mounted b3 appendage crystal and X-ray beam diameter. (D) X-ray diffraction pattern of b3 appendage crystal shown in (C). The final model revealed four copies of the b3 appendage in the asymmetric unit with large solvent pockets separating individual asymmetric units. Despite the limited resolution of the structure, due to the highly rigid fold of the ear domains, the density was extremely well ordered with almost all residues being buildable with high confidence (Figure 5.2.B). However, within the structure only the b3 appendage alone was seen with no associated density indicative of the clathrin box. Taken together with the large solvent pockets this suggests the clathrin box is freely accessible (likely disordered within the solvent pockets). 5.2 X-ray crystallography of the 3-ear domain 123 Fig. 5.2 comparison and structure of the β3 appendage (A) Comparative domain architecture of the β subunits of AP-1, AP-2 and AP-3 with annotated clathrin boxes (B) Structure of β-ear (green)with associated 2Fo-Fc electron density (left, pink). 124 5.3 Structural comparison of b3-ear domains Examining the surface of the structure revealed conserved hydrophobic patches between the AP-2 b2 ear and AP-3 b3 ear. Thus, I sought to see if the same motifs found in the endocytic adaptors might function in the same manner for the AP-3 complex. Using ITC I tested both the side site using EPSIN DPW motif and the top site with DDXXFXXLXXXR motif from ARH and DDXXFXXFXXXR from b-arrestin respectively (Figure 5.3). No binding was found to any motif suggesting that if the b3 ear does recruit additional vesicular factors it does not do so using the same motifs as the b2 ear. Fig. 5.3 AP-3 b3 ear does not bind similar motifs to AP-2 b2 ear Isothermal titration calorimetry results showing baseline corrected trace for titration of (A) Epsin DPW motif (B) b-arrestin DDXXFXXFXXXR motif and (C) ARH DDXXFXXLXXXR motif into b3 ear. 5.4 The δ-appendage and possible mechanism of autoinhibition 125 5.4 The δ-appendage and possible mechanism of autoinhibition Unlike other adaptor proteins, the appendage domains of AP-3 have no known or validated interactors. Previously, the Bonifacino group had suggested that, instead of recruiting additional adaptors like AP-2, AP-1 and AP-4, the AP-3 appendages may instead have a role in regulating the recruitment of AP-3 to the endosomal membrane. The mechanism they proposed suggested that the δ-ear bound σ3, thereby preventing the binding of the GTPase Arf1; furthermore, they also noted that this did not prevent binding of dileucine motifs (Lefrançois et al., 2004). To understand the molecular basis of the proposed ear-core interactions, I first sought to structurally characterise the differences between the δ appendage and other AP appendages. 5.5 Purification of the δ-ear The GST-δ-ear was expressed in BL21 DE3 plysS E. coli and purified as in methods section 2.3. After excessive washing of the protein bound, GST resin beads were divided equally between two columns. The first column was eluted using 50 mM reduced glutathione while the second was cleaved with thrombin overnight to remove the GST tag (Figure 5.4.A). Both stocks were subsequently concentrated for gel filtration and ran on SDS-PAGE to confirm they were intact with little contamination (Figure 5.4.B). The GST-tagged δ-ear was flash frozen for later use, and the cleaved δ-ear was concentrated to 15 mg/ml for crystallization. 126 Fig. 5.4 purification and crystallography of δ appendage (A) SDS PAGE of gst-tagged (eluted) and cleaved (thrombin) fractions of GST- δ ear (B) Gel filtration trace of δ appendage on superdex 200 10/300gl. (C) OVA view of litholoop mounted δ appendage crystal and X-ray beam diameter (red). (D) X-ray diffraction pattern of δ appendage crystal shown in (C). 5.6 X-ray crystallography of the δ-ear 127 5.6 X-ray crystallography of the δ-ear Twelve crystallization trays were set up and checked for nucleation on days 1, 3 and 5. The initial hits were all microcrystals that appeared to be multinucleate and were clumped. These hits were optimized using dilution screens yielding small single bipyramidal shaped crystals (Figure 5.4.C). This condition was reoptimized further for hanging drop vapour diffusion with and without 5% glycerol to reduce nucleation and increase size. The final crystals were grown in a 2:2.5 ratio of protein to mother liquor (0.2M sodium malonate, 20% PEG3350, pH6 and 5% glycerol). The δ-ear crystals were mounted cryoprotected and flash cooled as stated in methods section 2.18.2. Sixteen crystals were screened at the Diamond I04 beamline and diffraction pattern collected using a wavelength 0.9999Å. The best data set diffracted to 2.7 Å resolution (Figure 5.4.D) and was integrated, scaled and merged by Diamond Light Sources in house installation of autoPROC (statistics in Table 6). The structure was solved by molecular replacement using an independent ensemble of either the platform or sandwich domain of the α ear threaded with the sequence of the δ-ear. The crystal was solved in P 65 2 2 with two molecules in the asymmetric unit. After several rounds of refinement and rebuilding the final refinement produced a model with a Rfactor/Rfree of 0.24/0.30. During the building, it became clear there was additional unmodeled density in the weighted difference map (Fo-Fc map). As the density appeared to roughly resemble the architecture of a β sheet it was assumed this was an additional copy of the δ-ear. This additional copy appeared to have dual occupancy in which only one copy of a protein could be present without inducing a clash. As such, the additional copy was left unmodeled. 128 Fig. 5.5 structure of the δ appendage (A) structure of the δ ear (blue) (B) Comparison of the δ platform and the α platform domains with motif binding FXDXF motif (black) (1W80). (C) Comparison of the δ sandwich domain domains with the α sandwich domain domains binding FXXW motif (black) (3HS8). 5.6 X-ray crystallography of the δ-ear 129 As predicted the δ-ear adopted the typical architecture of an adaptin appendage domain, composed of an N-terminal sandwich domain closely packed against a C-terminal platform domain. Compared to a previously published structure of the α-ear, it adopted a thinner and more elongated profile (Figure 5.5.A). Comparison of the platform domains showed that the interface for the top site was not well conserved. When compared with α-ear bound to FXDXF motif of either Synaptojanin (1W80) or Amphiphysin (1KY7) it appeared the δ platform still contained a pocket for the first phenylalanine although it is not as deep or hydrophobic, the charge distribution is almost inverted across the entire interface which offers no obvious binding site for the aspartic acid, while the second phenylalanine site is missing entirely (Figure 5.5.B) (Traub, 2009). However, the conservation of the first phenylalanine pocket suggests that the ear may be able to bind DPF motifs such as shown for the α-ear (1KYF). The other binding site located on the appendage domains is the side site situated on the sandwich domain. The δ sandwich domain showed a chemically rather similar conservation to the α sandwich domain. However, superimposion with the structure of the α sandwich bound to the FXXW motif of intersectin-1 (3HS8) showed that the once again the pockets on the δ sandwich were not deep enough to accommodate either the phenylalanine or tryptophan (Figure 5.5.C)(Pechstein et al., 2010). This suggests that the FXXW motif is unlikely to bind to AP-3 nearly as well as AP-2, if at all. The observation that the top site of the δ appendages platform domain was shallower than that of AP-2, suggested that the ear may be more likely to interact with another surface rather than binding a disordered motif. This was further implied by the most hydrophobic regions of the domain projecting out suggesting they may bury in another domain and given the speculation in the literature, possibly that of σ3. 130 5.7 Interaction of the δ-ear with the AP-3 core To substantiate the finding of (Lefrançois et al., 2004), I designed Yeast-Two hybrid screens using the δ-ear against the σ3 alone (all Y2H experiments were performed by Sally Gray using the pGBT9/pGAD-C system, with the HF7c reporter strain (James et al., 1996), essentially as previously described (Harbour et al., 2010); growth on plates lacking histidine (−His) indicated that interaction had occurred). Since the S3D construct (as detailed in Chapter 3) is encoded as a single polypeptide chain but has both σ3 and the first 19 helices of δ this was also crossed this with the δ-ear (result shown in Figure 5.6.A). Unexpectedly the δ-ear bound σ3 alone but not S3D. As a positive control to confirm that the S3D was indeed expressed, folded and functional, this was crossed against both constitutively active Arf1 Q71L and the dileucine of MFSD12 in tandem (x3) (Figure 5.6.A). Fig. 5.6 Binding of the δ appendage to σ3 (A) yeast-2-hybrid of interactions between the δ appendage against various components of the AP-3 core (B) cartoon diagram of proposed inhibition of the AP-3 core by the δ appendage. (C) Schematic diagram of truncated S3D construct. 5.8 Structural prediction of the σ3:δ-ear interaction 131 In light of these surprising results, I wondered if the helices of δ trunk were in some way blocking access to binding site on σ3. As (Lefrançois et al., 2004) had also emphasised how the C-terminal tail region (153-193) of σ3 might be necessary for the binding of the δ-ear I wondered if our observations were an artefact of the S3D construct, due to the C-terminus of σ3 being obscured. Thus, a series of truncations were made to the δ trunk in the S3D construct producing σ3 followed by (δ1, δ1-122, δ1-250, δ1-365) detailed in (Figure 5.6.C). In addition to this approach, I tried to reproduce the claim that σ3 153-193 was required for binding of the δ-ear and created a σ3 1-154 to test the binding. Of all truncations designed only the addition of the first residue of the δ-trunk (S3D δ1) bound the δ-ear (Figure 5.6). This result suggested that not only is the C-terminus of σ3 not required to be free for binding δ-ear, but that the binding site for δ-ear is masked by helices in δ1-122. Contrary to the previously published report the last 40 residues of σ3 were not required for binding as σ3 (1-154) still bound the δ-ear and in fact mildly enhanced it (Figure 5.6.A). 5.8 Structural prediction of the σ3:δ-ear interaction As the yeast-2-hybrid experiments had suggested that the δ-ear would bind only σ3 I utilised Alphafold2-multimer with MMseqs2 (Google Collab) to try and model the interaction (Jumper et al., 2021). Five models were generated with three rounds of recycling and one round of relaxation. All models were almost identical (<0.2Å RMSD) at the interface of interaction with no major differences between sidechain positions and an extremely robust predicted alignment error (PAE) plot throughout (Figure 5.7.A and 5.7.B). The models showed helix 3 of σ3 packing into the top site of the δ-ear platform domain (Figure 5.7.A) with σ3 tail (153-193) not forming any contacts suggested to be disordered by the PAE plot (Figure 5.7.B). In a complete AP-3 complex, helix 3 of σ3 is normally inaccessible as it is blocked by the first 4 helices of the δ-trunk; as such the δ-trunk and δ-ear would be unable to bind simultaneously (Figure 5.7.C) explaining why δ-ear was able to bind σ3 but not S3D as previously observed in (Figure 5.6). The interface was composed of three hydrophobic residues of σ3 (L107, I110 and F111) packing into shallow hydrophobic groove and the previously mentioned ‘F pocket’ utilised by the DPF and FXDXF motifs that bind the α-ear (Figure 5.7.D). Surrounding this interface is an 132 array of electrostatic residues which pair with complementary residues on the opposing interface (Figure 5.7.E). Fig. 5.7 Alphafold2 prediction of the δ-ear:σ3 interaction (A) Top ranked AF-2 prediction of δ-ear:σ3 (light blue:grey) (B) Predicted alignment error plot of top ranked AF-2 model. (C) Superimposition of top ranked AF-2 prediction and δ-trunk (navy). (D) Binding of σ3 helix to δ-ear top site pocket. (E) interaction between σ3 helix and δ-ear top site pocket. 5.9 Characterization of the σ3:δ-ear interface 133 5.9 Characterization of the σ3:δ-ear interface To confirm the validity of this model, I designed specific point mutants to δ-ear that I would expect to abolish binding to σ3. The mutation of C1117W and D1130W both aimed to fill the hydrophobic pocket in which σ3 F111 docked. Only C1117W ablated binding while D1130W had no effect (Figure 5.8). Less extreme mutations such as L1062S and L1119S which make up the edges of the σ3 F111 also ablated all binding when compared to a WT control (Figure 5.8). More peripheral charged contacts such as K1132A also showed marked reduction of binding but did not abolish it, in line with the model that this contact was not as critical for the interaction as suggested by (Figure 5.7.E). Fig. 5.8 Validation of the predicted σ3:δ-ear interaction Yeast-2 hybrid of specific point mutations to the predicted σ3:δ-ear interface. 134 Conversely mutations were made to the cognate interface on σ3 to confirm this disrupted binding. Mutation of σ3 helix 3 from 107LDLIFH112 to 107SDLSAA112 was able to prevent all measurable binding to δ-ear (Figure 5.8) and likewise for less drastic mutations of D108A and F111S (Figure 5.9) suggesting that the model produced in (Figure 5.7) was indeed correct. Finally, to see if σ3 helix 3 alone could drive interaction with the δ-ear, σ3 (98-117) was crossed against the δ-ear and while weak growth was observed after 5 days; this result was also seen when crossed against just the empty vector indicative of autoactivation, so that no conclusion can be drawn (Figure 5.9). Fig. 5.9 Validation of the predicted σ3 interaction with δ-ear interaction Yeast-2 hybrid of specific point mutations to σ3 to ablate the predicted σ3:δ-ear interface. 5.9 Characterization of the σ3:δ-ear interface 135 In summary, while I was able to reproduce the interaction proposed by (Lefrançois et al., 2004), I suggest that the interface of interaction is not the σ3 tail, but helix 3 of the σ3 longin domain and that the interaction is mutually exclusive with the binding of the first helices of the δ-trunk. I was unable to definitively reconcile the role of this interaction within a complete AP-3 complex, but a number of theories can be suggested. One plausible explanation could be that the interaction of the δ-ear and σ3 plays a role in complex assembly akin to the suggested function of Alpha- and Gamma-Adaptin Binding Protein (AAGAB) for the AP-1 and AP-2 complexes (Gulbranson et al., 2019). Another possibility is that the first several helices of δ- trunk have some inherent flexibility and can reversibly associate and dissociate from σ3. When dissociated, this would free the interface for δ-ear to bind back onto what would otherwise be an occluded interface. A change in conformation of the first several helices would also break the Arf1 binding site preventing AP-3 recruitment to the membrane. This in turn would fit with the observations that the δ-ear and Arf1 are mutually exclusive in their binding to the AP-3 core. Taken together, these interpretations could suggest that the δ-ear stabilises a conformation of the AP-3 core where the N-terminal helices of the δ-trunk are pried away from σ3 preventing interaction of AP-3 with Arf1. The solution structure of the AP-3 core obtained by single particle Cryo-EM (Chapter 6), the density for the first several helices of δ-trunk are absent supporting the possibility that the binding site on σ3 is available, however no density of the δ-ear is observed either. These observations are caveated further by the limited resolution of the reconstruction. 136 Chapter 6 The AP-3 core in solution 6.1 Introduction The core components of AP-3 complex are phylogenetically closely related in sequence to core components of the other, well characterized AP complexes (AP1, AP2 and COPI). However, although there is shared sequence similarity, there is a surprising difference in 3D structure, with COPI taking up a wider stance when in the “super open” conformation on a membrane whereas AP-2 and AP-1 have been shown to adopt a “closed” conformation in solution and a narrower “open” structure on membranes (Dodonova et al., 2015, Collins et al., 2002, Heldwein et al., 2004, Ren et al., 2013, Jackson et al., 2010). All these characterized adaptors were all present in the last eukaryotic common ancestor (LECA) with AP-1, AP-2 and COPI the most divergent from each other, thus likely representing extreme examples within the family. To understand where AP-3 falls within this spectrum of conformation, I aimed to resolve the structure of the AP-3 core complex both in solution and on membranes by cryogenic electron microscopy (Cryo-EM). Over many years, efforts to characterize AP-3 core by the Owen lab (Helen Kent, Lena Wartosch and Veronica Kane-Dickson) had proven difficult and encountered several major issues in obtaining suitable samples. Unlike AP-1 and AP-2, which can be expressed in E. coli, the AP-3 core is heavily proteolyzed when bacterially expressed with significant degradation to the μ3-linker and hinge regions. Such samples were unusable for structural studies partially as the heterogeneity, and potential loss of domains would severely limit subsequent observations as to the conformation or regulation. Therefore, the AP-3 core was re-cloned into baculoviral vectors for expression in insect cells (SF9), which yielded less protein but of a significantly higher quality. Grids made with this newly expressed AP-3 core in turn had proven problematic with samples suffering from a combination of aggregation and complex disassembly when vitrified. As the previous samples had behaved as an intact complex by SEC- MALS and Dynamic light scattering (DLS), the poor sample quality was attributed to the 6.2 Purification of the AP-3 complex 137 complex’s interaction with the air-water interface. As such, the first aim was to purify the complex and attempt to increase stability by breaking interactions with the air-water interface. 6.2 Purification of the AP-3 complex Insect cells were grown, infected and harvested as outlined in material and methods section 2.4. AP-3 expressing insect cell stocks were lysed and purified using nickel affinity purification as outlined in material and methods section 2.5. The resulting protein yielded a single large peak with a smaller shoulder both containing all four subunits of the AP-3 complex (Figure 6.1). Fig. 6.1 Purification of the AP-3 core (A) Gel filtration trace of AP-3 with full length δ on superose 6 10/300gl. (B) SDS PAGE of fractions from gel filtration. 138 6.3 Optimization of AP-3 core vitrification Initial attempts to improve the sample preparation by optimizing for salinity and buffer pH found no effect. Micrographs of the grid showed severe aggregation across the grid holes with some areas resulting in large fibrillar clumps (Figure 6.2.A). In less dense areas particles still had some small aggregation however particles still appeared irregular and indicative of complex disassembly (Figure 6.2.B). The severity of the aggregation was found to be reduced as the time between purification and grid making was reduced. As such all subsequent purifications and grid production was preformed within 24 hours. However, there was still prevalent “yolking” on the grid where high concentration patches of protein fail to spread across the grid, leading to locally varied protein concentration. This often manifests with protein dense patches (termed “yolks” due to the resembles to the yolk of an egg) which are too thick and densely packed with particles to image through while neighbouring areas are devoid of particles (Figure 6.2.C). Historically, this has been attributed to insufficient glow-discharging causing the surface of the grid to remain hydrophobic and leading to inadequate spreading. However, from testing the same batch of grids with a variety of different protein samples (Dept of Biochemistry), it has been concluded the only factor that correlated with the presence of yolks was the addition of proteins that are predisposed to forming yolks. Thus, these results are probably due an inherent property of the AP-3 complex when vitrified. In some previously reported instances, the use of detergents such as CHAPSO, trehalose and β-octyl-glucoside had proven helpful in stabilizing complexes. Such treatments have been suggested to work by reducing effects of air-water interface interactions with the sample (Li et al., 2021). In subsequent sample preparations, all were trialled. While CHAPSO and trehalose showed little effect on complex stability, addition of 0.05% β-octyl-glucoside reduced the degree of heterogeneity within a sample compared to previous grids (Figure 6.2.D). Despite this apparent increase in complex stability, particles still appeared to cluster and aggregate although not as severely as previously seen. In parallel to these optimizations, it was suggested that a change of chemical environment on the grid may also help to resolve particle integrity. 6.3 Optimization of AP-3 core vitrification 139 Screening of different grids types in the presence of β-octyl-glucoside showed that for both quantifoil AUfoil grids (Figure 6.2.E) and quantifoil graphene oxide grids, (Figure 6.2.F) the yolking was now less severe and monodisperse single particles could be found in the edges of the “yolk”. 140 Fig. 6.2 Optimization of AP-3 vitrification (A) 73000x magnification micrograph of thread like aggregate of purified AP-3 (Scale=100nm). (B) 73000x magnification micrograph of heterogenous AP-3 particles (Quantifoil holey carbon) (Scale=100nm). (C) 700x magnification micrograph of grid square showing with AP-3 dense “yolking” effect (Scale=20µm). (D) 73000x magnification micrograph of AP-3 +0.05% Beta2- Octylglucoside (Quantifoil holey carbon) (Scale=100nm). (E) 73000x magnification micrograph of AP-3 +0.05% Beta2-Octylglucoside (Quantifoil Graphene oxide) (Scale=100nm). (F) 73000x magnification micrograph of AP-3 +0.05% Beta2-Octylglucoside (Quantifoil gold grid) (Scale=100nm). A C E B D F 6.4 Single particle Cryo-EM of the AP-3 core in solution 141 6.4 Single particle Cryo-EM of the AP-3 core in solution Single particle Cryo-EM data around the edges of yolks was collected first on a 200 kV Talos (University of Cambridge). Micrographs were processed on the fly using the Department of Biochemistry’s in-house installation of WARP (Tegunov and Cramer, 2019). This program was used for initial motion correction, CTF correction and particle picking. The subsequently generated particle stack of 63868 particles were extracted using Cryosparc (Structura Bioscience) (Punjani et al., 2017). 2D classification showed a significant difference between the core of the AP-3 complex and the AP-2 complex which had previous been collected by Veronica Kane-Dickson; however, due to the limited number of particles collected on the Talos, the data were relatively low resolution and thus it was difficult to definitively confirm what differences existed between the two complexes. Thus, the dataset was collected for the AP-3 core using the 300kV Krios. This new dataset was processed as previously outlined for the Talos dataset yielding a particle stack of 1353983 particles. 2D classification, now with significantly increased particle number, was able to resolve high resolution secondary structure such as alpha helices within the solenoids of the large subunits of the AP-3 core. However, unlike the previously collected data for AP-2, no density for the C-μ3 domain could be seen in the core’s bowl; thus, unlike AP-2, AP-3 was open in solution (Figure 6.3). In line with this ab initio reconstruction of the AP-3 core produces several models, all in the open conformation. Iterative rounds of classification and refinement of the ab initio models produced resultant volumes that could be refined to 3.27 Å resolution (Figure 6.3). At low resolution, this model resembled a similar overall profile of previously solved open AP-2 (2XA7) or AP-1 (4HMY) structures (but without the Arf1) (Jackson et al., 2010, Ren et al., 2013). However, high resolution and model building proved impossible as severe orientation bias of the particles had resulted in a highly anisotropic map (Figure 6.3). 142 Fig. 6.3 Single particle cryo-EM of the AP-3 core (A) 2D-classification of the AP-3 core. (B) orientation distribution of 3D reconstruction of the AP-3 core. (C) gold standard Fourier shell correlation (GS-FSC) plot of 3D reconstruction of the AP-3 core. (D) Sharpened map of the AP-3 core 3D reconstruction. 6.4 Single particle Cryo-EM of the AP-3 core in solution 143 The docking of an Alphafold2 model of the AP-3 complex into the density allowed for comparison and subsequently four major differences were noted (Figure 6.4). 1. The C-μ3 domain was not resolved in the final map, but as N-μ3 was visible in the map and the band for μ3 ran at full size on SDS page this suggested that the C-terminal domain was present in the sample but was disordered compared to the rest of the complex and not been cleaved during purification (Figure 6.5). 2. The N-terminal helices of AP-3 d and most of s3 were not well resolved unlike the other parts of the b3 μ3 and the d trunk (Figure 6.5). This suggested inherent flexibility of the d helices leading to poorly resolved s3 and d N-terminus. 3. Although the d:σ hemi complex of AP-3 appears to be flexible thus poorly resolved, no new density corresponding to the N-terminal helices of the d trunk was not observed. However, in addition to this no density fitting the d-ear could be discerned within the structure. In line with our hypothesis presented in chapter 5, if the ear did bind within this structure, it is not obvious and likely transient or lost in alignment or classification (Figure 6.5). 4. A large region of the b3 trunk was not accounted for in the map that was in the model. Halfway through the solenoid of b3 trunk was an intervening flexible loop (Figure 6.5). As this appears to be unique to AP-3 and has not been observed in other AP complexes its function is unknown. Data from phosphoproteomic screens reveals that this loop contains the most commonly identified phosphosite detected in the AP-3 complex (Figure 6.5), thus may serves as a regulatory site to recruit additional vesiculation factors. Supporting this interpretation further, the loop is conserved with other more distant eukaryotes like yeasts and trypanosomes, so it likely serves a core purpose to the AP-3 complex. 144 Fig. 6.4 Low resolution docking of the AP-3 core (A) filtered volume of the AP-3 core coloured by subunit σ3 (grey), N-μ3 (crimson), b3(green) and d (steel blue). (B) coordinates model produced by AF-2 and built into single particle density. (coloured as previously stated with EM map in ghostly white). 6.4 Single particle Cryo-EM of the AP-3 core in solution 145 Fig. 6.5 Insights from low resolution docking of the AP-3 core (A) Complete AF2 model docked into experimental density σ3 (grey), N-μ3 (crimson) b3 (forest green), b3 loop phosphosite residues (yellow) and d (steel blue) (B) Sequence of b3 loop not accounted for by density and citations table from phosphosite plus showing observed phosphorylation sites throughout the b3 subunit. (C) Close-up of N-terminal helices of AP-3 d-trunk is not resolved within EM map. 146 As other all other previous solution structures of AP complexes in solution have been shown to be closed this result was as much a puzzling as it was interesting. To understand if the complex was really open in solution and not an artifact of vitrification or proteolysis, I sought to assess if the YXXΦ binding site was accessible in the absence of membrane or Arf1:GTP. GST pulldowns with either GST-linker-TGN38 or GST-linker without a YXXΦ motif were conducted as outlined in method section 2.14. After coincubation with AP-3 core (without the d hinge and ear) and three rounds of washing, only the construct containing the YXXΦ motif was able to pull down AP-3 while the GST-linker alone did not (Figure 6.6). This suggested that in solution the YXXΦ binding site was accessible, indicative of an open but not closed conformation. By comparison fluorescence anisotropy has shown that in solution AP-2 are ~99.9% in the closed conformation (Jackson et al., 2010). Fig. 6.6 AP-3 can bind YXXΦ motifs in solution Pulldown of GST-linker or GST-linker-YXXΦ (bait) against recombinant AP-3 core (prey). 6.5 Why is AP-3 open in solution? 147 6.5 Why is AP-3 open in solution? As all evidence suggested that the AP-3 core was open in solution, I wanted to understand what differences there were compared to the cores of AP-1 and AP-2. Since I had been unable to resolve a high-resolution structure of the AP-3 core, I compared the determinants of the closed conformation for AP-2 and AP-1 and compared to predicted models from the Alphafold2 database (EBI). For both AP-1 and AP-2 in the closed conformation two points of contact are made between the C-μ domain and the rest of the core. The first point of contact is the non- specific charge interaction between the C-μ barrel and the side of the α/γ trunk, while the second is the packing of a valine (Val365 in both AP-2 and AP-1) from the β subunit which packs into the Φ pocket of YXXΦ binding site of the C-μ subunit to hold the AP in a closed conformation (Figure 6.7). Modelling of the AP-3 core in a closed conformation by Alphafold2 (AF2) proved impossible as AF2 is incapable of modelling any AP complexes in the closed conformation. This is presumably, at least in part, due to the fact there are more open structures available on the PDB, biasing the AF2 training dataset. Thus, comparison was made by aligning the previously published structure of C-μ3 (4IKN) and the predicted model of β3 and d (AP-3) against a closed AP-2 core (2VGL) (Mardones et al., 2013, Jackson et al., 2010). Whilst the interaction of C- μ3 against d showed no obvious interface, there was a clear difference in the packing of Φ pocket of C-μ3 against β3, as instead of V365. AP-3 possesses a Gln614. Gln614 is both too long and polar to satisfy the packing of the Φ pocket (Figure 6.7). Thus, while the Φ pocket is conserved in C-μ3, the chemistry imparted by Gln614 may be the key determinant preventing the adoption of a closed conformation by the AP-3 complex. It would be interesting to express a mutant complex with Q416V to see if this can reverse the observation and close the AP-3 complex. 148 Fig. 6.7 Comparison of the AP conformations (A) Depiction of phylogenetic analysis of the evolutionary origins of the AP complexes indicating emergence of closed conformations. (B) Sequence comparison of β subunits of human AP complexes. (C) Structure of “latching” interface mediating the AP-2 closed structures b2 (green), μ2 (pink), σ2(cyan), α (blue), hydrophobics (brown) and hydrophilic (turquoise). (D) Structural comparison of “latching” interface of different AP complex family members b1 (pink), b2 (green), b3 (forest green), b4 (mauve), b5 (violet). 6.6 Attempts to fix orientation bias 149 Additionally, to see how widespread this phenomenon was across the whole AP family, sequence alignment of AP-1/2/3/4/5 and COPI were conducted. Interestingly, the Val was conserved in AP-1/2/4 but not AP-3/5 and COPI. This trend obviously mimicked the order in which the AP complexes diverged from each other; with COPI AP-3 and AP-5 diverging earliest with AP-1/2/4 diverging most recently (Figure 6.7.A). In the case of both AP-5 and COPI, the Val is replaced by Gly457 or Asp375 respectively, which both also fail to mimic the size or hydrophobicity of a Val that would pack into a Φ pocket. Notably, when the predicted model of AP5B1 is superimposed structurally, a Pro458 takes the place of the Val, possibly enabled by the flexibility of the upstream glycines. Thus, although both share a hydrophobic characteristic, it is unlikely that a proline substitutes for a valine. Although this hypothesis would need to be experimentally verified and is possibly at risk of over-interpretation, the data at present could suggest that a closed form may not exist for COPI, AP-5 and AP-3 complexes, while predicting its existence for the AP-4 complex. This interpretation would be in line with the gain or loss of closed form some time before the LECA. However, these results are limited by our lack of understanding of AP-5 and the apparent incompatibility of vitrifying COPI in the absence of membrane (personal communication with Kat Ciazynska and Rebecca Taylor). 6.6 Attempts to fix orientation bias The high anisotropy of AP-3 in solution made the map interpretations difficult, limiting model building and the conclusions that could be drawn. Thus, in an attempt to obtain more interpretable maps by single particle cryo-EM I aimed to fix the orientation bias to obtain a more isotropic dataset. 6.6.1 2D rebalancing Software packages, such as cryosparc, offer a 2D rebalance job type that aim to reduce the effects of orientation bias of a volume by limiting the number of oversampled orientations within a set of 2D classes. In practice, this works by taking 2D classified particles and removing particles in the most common classes with the most particles. The number of particles removed is specified as a rebalance factor which is approximately the ratio of particles in the smallest 150 class to the largest class. The subsequently modified particle stack is then able to be used to create a volume using ab inito reconstruction which can then be refined; such an approach could be applied to the previously obtained dataset. A range of rebalance factors were trailed with little success. From a starting dataset of 244396 particles, 10%, 20%, 30%, 40% and 50% 2D rebalances were performed. Notably a 30% rebalance resulted 138401 particles remaining which still exhibited severe orientation bias, while a 50% rebalance resulted in 4400 particles which were not sufficient for a reconstruction and still had some mild bias (Figure 6.8). This result was attributed to the fact that I have effectively only have a single orientation and no other views were present so the normalisation was against a value of zero. Fig. 6.8 2D rebalancing of AP-3 orientation biased particle distribution with (A) no rebalancing (B) 30% rebalancing (C) 50% rebalancing. 6.6 Attempts to fix orientation bias 151 6.6.2 Tilted collection A common but imperfect solution to orientation bias is tilted data collection. This approach involves physically tilting the grid relative to the path of the beam during data collection, resulting in the particle orientation distribution being randomised about the tilt axis in a manner proportional to the angle of degree tilt. In some cases, this approach has proven beneficial. However, tilting in itself brings a series of practical limitations and complication. While this approach does offer increase the spread of particle orientations, it doesn’t in itself allow for the observation of new orientations but is instead restricted to the orientations already present on the grid. This results in improvement but still may produce a map that is difficult to build or interpret. Additionally, tilting increases the thickness of the sample reducing signal to noise ratio due to the increased number inelastic collisions between the electron beam and sample. While the effects of inelastic scatter can be mitigated by collection of more data and a well calibrated energy filter, the Krios in the Department of Biochemistry lacked an energy filter. In addition to tilted data collection requiring more data be collected, collecting data with a tilted grid is also slower. This requirement is partially due to the need for the microscope to find eucentric height at each position and then shift away from it at every new grid square, but this problem is further confounded by inability to use aberration-free image shift system (AFIS). AFIS speeds up data collection by allowing automated beam shifting to move between holes (preselected in EPU software) within a grid. Beam shifting is desirable where possible as it doesn’t require the movement of the stage, which induces vibration that takes time to stabilise before an image may be taken. Therefore, more images can be taken within a set timespan when beam shifting between holes as compared to moving the stage. Sadly, beam shifting relies on EPU being able to fit a repeating lattice of circles over the grid squares in order to find the holes and focus before imaging, but when tilted relative to the beam, grid holes appear not as circles but ellipses, which EPU can’t recognise. Hence all holes must be assigned manually and navigated between by stage movement, limiting the number of images that can be taken in a given time. 152 6.6.3 CryoEF To assess the degree of orientation bias, I used CryoEF software from the Russo group (Naydenova and Russo, 2017). CryoEF scores the severity of the orientation bias of a single particle structure by determining the “efficiency” (EOD) of a specimen. This efficiency calculates to what extent the sample orientation alone limits the isotropic resolution of the reconstruction. An EOD=1 implies no limitation from orientation while EOD=0.5 being approximately the threshold at which a structure will successfully reach high resolution within ~105 particles. Based on this efficiency value it is possible to then determine roughly how much more data would need to be collected and at a defined tilt angle to provide a more buildable structure. From our previously collected data sets, it was found that across the map the data ranged from 2.1-5.38 Å resolution depending on the direction with an EOD=0.56. As this efficiency is borderline CryoEF recommended tilted data collection either 2.1x more data at a 28 ° tilt or ~3.3x more data at a 43 ° tilt. 6.6.4 Tilted data processing With these caveats in mind, a single particle dataset of AP-3 core on quantifoil AUltrafoil grids was collected at a tilt of 30 ° from eucentric. Micrographs were motion corrected and CTF corrected using WARP and from these 365,499 particles were picked. The particle stack was processed in cryosparc with iterative rounds of 2D classification, followed by ab inito reconstruction. 2D classes showed significantly less detail compared to the non-tilted data collection. Promising volumes were iteratively refined by heterogeneous refinement, yet despite this approach all maps generated were not significantly improved in interpretability with N-terminal trunk of β3 disappearing entirely. However, at this time 9-13 Å solution structures of the AP-3 core from yeast was published alongside homology models. The researchers, from the Ungermann & Raunser labs, also noted the core complex was open in solution and there was significant flexibility between the apl5/aps3 (δ/σ3) hemicomplex and the rest of the core (Schoppe et al., 2020). Despite it not being resolved within the map, using crosslinking mass-spectrometry they also claimed that the apl5 ear bound back to the core in an autoinhibitory fashion to prevent binding of Arf1:GTP, 6.6 Attempts to fix orientation bias 153 in a similar manner as suggested for the mammalian complex by Bonifacino and colleagues (Lefrançois et al., 2004). However, it is worth noting that while the yeast AP-3 complex bears homology to the metazoan complex, the yeast complex lacks ear domains entirely, suggesting that some reconsideration of the conclusions drawn may be warranted. 154 Chapter 7 Determinants of AP-3 recruitment 7.1 How is the AP-3 complex recruited to the endosome? Although originally thought to bud from the trans Golgi due to a phenotype in a Saccharomyces cerevisiae genetic screen (Cowles et al., 1997a), the localisation of AP-3 remains controversial. Substantial immuno-EM and live cell imaging experiments of mammalian cells localised the AP-3 complex to a tubulovesicular endosomal compartment (Peden et al., 2004, Theos et al., 2005, Di Pietro et al., 2006, Dell'Angelica et al., 1998). However, recent studies, with lower resolution immuno-florescence data, claim that AP-3 can be at the Golgi in some instances in neuronal cultures, suggesting its localization may not only be different between yeast and higher eukaryotes but also between different cell types within an organism (Xu et al., 2022). Thus, there is still no generally accepted consensus, and a lack of certainty as to the true localization of the AP-3 surrounds the topic. The AP-3 complex, like AP-1, AP-4 and COPI, has been shown to be dependent on the GTPase Arf1 (Simpson et al., 1996, Simpson et al., 1997). Arf1 localizes to membranes when in the GTP bound conformation; GTP causes the displacement of the N-terminally myristoylated amphipathic helix which drives the interaction with lipid bilayers (Cherfils, 2014). As Arf1 recruitment to organellar membrane is dependent on the presence of ARF-GEFs it can be found across the trans Golgi and early endocytic system, making any individual effector localisation difficult to assign. Despite there being a wealth of published structural and biochemical information on the recruitment for both AP-1 and COPI, very little has been characterized for AP-3 (Yu et al., 2012, Ren et al., 2013, Dodonova et al., 2015, Hooy et al., 2022). In contrast, AP-2 has been demonstrated to localize primarily in a PI45P2 dependent manner, although additional reports have suggested a more minor Arf6 dependency (Höning et al., 2005, Krauss et al., 2003). Conversely, GTPase-dependent adaptors like AP-1 and AP-5 have been suggested to have a more minor preference for other phosphatidylinositol lipids (Wang et al., 2003, Hirst et al., 2021). 7.2 AP-3 and Phosphoinositides 155 In order to understand the cellular determinants for localization of the AP-3 complex I aimed to biochemically and structurally characterized the role for both Arf1 GTPase and phosphatidylinositol lipids. 7.2 AP-3 and Phosphoinositides The AP-2 complex harbours four PI45P2 binding sites; two on C-μ2, one on α and finally one on b3. Previous work from the Hurley and Bonafacino labs had shown that C-μ3 was not as charged as C-μ2 and concluded it was less phosphatidylinositol dependent for localization (Mardones et al., 2013). As such I sought to study the other major site at the N-terminus of δ. As S3D had proven a tractable construct for and validated structurally I utilized this to dissect the interaction with membranes. As a simple test to see if there was any interaction with phosphatidylinositol lipids, a multilamellar liposome spin down experiment was conducted as specified in method section 2.10. A stock phosphatidylcholine phosphatidylethanolamine (PCPE) lipid mix was made at a 80:20 ratio. Subsequent phosphatidylinositol phosphate (PIP) lipid mixes were made up with a 70:20:10 ratio of PC:PE:PIP. Of the lipids tested, PI3P is considered the primary phosphatidylinositol of the early endosomal system, while PI4P is predominantly trans Golgi, PI35P2 is a scarcer lipid most commonly found on later endosomes and lysosomes, and PI45P2 the most abundant phosphatidylinositol in the cell is a defining hallmark of the plasma membrane. Unexpectedly, in multilamellar liposome spindowns S3D there was significant pelleting for the PI35P2 (Figure 7.1). This result supports the observation and previous reports of the AP-3 complex being recruited to endosomal membranes; however, it suggests the recruitment is either on far later in the maturation than would have been expected or on a highly specialized subdomain. Although surprising, as there was no pelleting with PCPE negative control I can confidently say that aggregation was not a factor in pelleting and the lack of pelleting in PI45P2 suggests that it is not simple charge mediated interaction but specific binding of the PI35P2 lipid. 156 Fig. 7.1 Multilamellar phosphoinositide liposome spin downs SDS PAGE of 10% phosphoinositide liposome or PCPE liposome spindowns experiments pelleting S3D. In a more refined experiment to confirm the proposed interaction, liposome Surface Plasmon Resonance (SPR) was used to compare phosphoinositides in parallel. Similar lipid mixes as before, however now spiked with only 4% of phosphoinositide (PC:PE:PI 76:20:4), were extruded as unilamellar liposomes. After subtraction of PCPE background, the S3D showed a clear preference to PI35P2 liposomes over other phosphoinositides, in concordance with the previous spindown data (Figure 7.2). Moreover, the PI35P2 the response units (RU) appeared to be concentration dependent for S3D further supporting that this was a real interaction. From four technical replicates across two independent experiments binding data was collected and surface affinity data suggested a KD of 2.1µM (+/-0.62) between S3D and PI35P2 in SPR buffer (Figure 7.2.B). As such it was concluded that PI35P2 is sufficient to recruit S3D to a membrane in vitro. 7.2 AP-3 and Phosphoinositides 157 Fig. 7.2 Liposome based SPR of S3D for phosphoinositides (A) Sensogram showing response over time for varying concentrations of S3D flowed over 4% PI35P2 liposomes with control PCPE trace subtracted. (B) Binding curve obtained by affinity fit of data from sensogram in (A). 158 7.3 Structural basis of Arf1 binding to AP-3 d An interaction with phosphoinositides only contributes a single binding site, so it is unlikely to be sufficient to recruit the entire AP-3 complex in vivo. A previously mentioned GTPase, Arf1, is most likely the primary determinant of AP-3 localisation. However, Arf1 is localized across the later Golgi cisternae and the early endocytic system. Thus, the discrete localization of AP- 3 is most likely on a subdomain of an Arf1 positive compartment and may involve cross-talk between these two factors, a concept termed coincident detection. Previous structural work on COPI and AP-1 had identified two binding sites for Arf1:GTP, located at similar sites of both the large subunits (Yu et al., 2012, Ren et al., 2013, Dodonova et al., 2015, Hooy et al., 2022). To visualize the binding of Arf1 to AP-3 on the d hemicomplex I sought to co-crystallize the interaction. Previously frozen aliquots of both Arf1 Q71L and S3D were gel filtered and concentrated to 14 mg/ml. The two stocks were mixed with MFSD12 dileucine peptide in a 1:1:1.5 stoichiometric ratio before being used for crystallization screens (outlined in more detail in method section 2.18). Similar to previously crystallized S3D:MFSD12 complex the initial nucleation was rapid with first signs of crystals could be found as early as one hour. The initial crystallization conditions were also similar to that of S3D:MFSD12 with initial hits being a combination of alcoholic conditions and buffers 8-9 pH, yielding thin plate like crystals. This condition was subsequently optimized by screening of different alcohols and buffers at various pH and dilutions with several of the most promising of conditions taken forwards for reoptimization for hanging drop crystallization. Most of the crystals grown via hanging drop displayed a similar plate like morphology as before but much thicker (Figure 7.3.A). Sixteen crystals were mounted and cryoprotected as outlined in method section 2.18.2. Diffraction patterns were collected through 360 ° of rotation at a wavelength of 0.9796 Å. The best diffraction patterns were collected on crystals being grown in; 25% methanol, 0.1 M Bicine pH 8.5 and 5% glycerol, where spots could be processed out to an isotropic resolution of 2.4 Å (Figure 7.3.B). This data set was integrated, scaled and merged into the space group of P 21 21 21 with the Diamond Light Sources in-house installation of Xia2- Dials. 7.3 Structural basis of Arf1 binding to AP-3 159 Fig. 7.3 Crystallization and diffraction of S3D:Arf1:MFSD12 (A) OVA view of litholoop mounted crystal and X-ray beam diameter. (B) X-ray diffraction pattern collected of crystal shown in A. The diffraction pattern was solved by molecular replacement (Phaser MR) using the previously solved S3D structure and the predicted coordinates of Arf1 from the Alphafold2-EBI database as separate ensembles. As the Mathews coefficient estimated three S3D:Arf1:MFSD12 assemblies (40% solvent) within the asymmetric unit, three copies of both the S3D and Arf1 ensembles were searched for. However, Phaser MR was only able to find two copies of S3D and a single copy of Arf1. The initial building and refinement (Coot and Refmac5) showed unambiguous placement of previously mentioned copies of S3D and Arf1; additionally, the unmodelled density corresponding to either of the two copies of MFSD12 dileucine or the GTP could be built. This structure was iteratively corrected and refined reaching a final Rfactor/Rfree of 0.23/0.28. Notably only one copy of S3D was bound to Arf1; however, at the same interface on the unliganded copy of S3D there was blurry density (Figure 7.4.A). Superimposition of S3D:Arf1 with over the unliganded model to produce two S3D:Arf1 in the asymmetric unit, could also be refined and produced a valid solution with 0.24/0.28 Rfactor/Rfree, with a better resolved but still unbuildable density around this second Arf1. Unlike the well resolved Arf1, this second Arf1 sits facing a large solvent pocket rather than forming a crystal contact; taken with the weak density, this location suggests that this second copy is only partially present within the crystal and not vital to lattice formation. 160 The structure of S3D:Arf1 revealed a similar arrangement as had been seen with previous AP complexes and COPI. S3D with Arf1 was almost identical in conformation to the previously crystallized S3D. Arf1 bound to the outer curve of the d-solenoid interfacing with helices 2-7. As with other adaptors helices 4 and 6 formed the major contact with Phe77, Ile80, Val108, Met110 and Leu111 creating a hydrophobic wedge projecting out of the δ solenoid (Figure 7.4.B) and burying into a hydrophobic cavity on the switch I & II interface (Figure 7.4.C. On the edges of d-solenoid’s hydrophobic wedge, several charged residues created a peripheral charged surface capable of interacting with oppositely charged side chains on Arf1. Although the charges on either protein are in close proximity to one another, they are not close enough to be considered as a direct interaction it may suggest that these sidechains may form more minor electrostatic interactions which may guide and orient the interaction between the Arf1 and AP-3. 7.3 Structural basis of Arf1 binding to AP-3 161 Fig. 7.4 Structure of the S3D:MFSD12:Arf1 complex (A) Structural overview of the S3D:Arf1:MFSD12 complex σ3 (grey), d (steel blue), MFSD12 dileucine peptide (gold), Arf1 (yellow) and GTP (coloured by atom). (B) Arf1 binding site on d coloured by hydrophobicity and electrostatic. Divide between the displays annotated by black line. (C) enlarged view of Arf1 bound to “hydrophobic wedge” on d. 162 This interaction is comparable with other adaptor protein complexes with a central hydrophobic wedge being the critical mediator of binding but, as previously mentioned, the Arf family of GTPases recruits numerous effectors with AP complexes only making up one cohort of those effectors. As such, it is not surprising that even within the adaptors, there is variation on how an interaction can be scaffolded. Superimposition of Arf1 bound to murine AP-1g (6CM9), human AP-3d (this thesis) and bovine COPIg (3TJZ) showed minor differences between the three interactions (Figure 7.5) (Morris et al., 2018, Yu et al., 2012). Core to all three interactions was the positioning of two successive, medium sized hydrophobic residues (AP-1g Leu101/Leu102, AP-3d Met110/Leu111 and COPIg Ile103/Ile104) into a hydrophobic pocket on Arf1. Likewise, AP-1g Leu71, AP-3d Ile80 lined a hydrophobic pocket, but COPI diverges slightly by using the methyl group of Thr74 in its stead. Structurally AP-1g Leu68, AP-3d Phe77 and COPIg Phe71 all occupy the same position on helix 4; however, their packing against Arf1 is not identical with COPIg Phe71 spanning a central area of the Arf1 interface and AP-1g Leu68 making a more peripheral contact. Despite the conservation of the Phe between AP-3 and COPI, AP-3d mimics the conformation of AP-1g Leu68 rather than making a more central contact. This result can be explained by the presence of AP-1g Val99 and AP- 3d Val108 which occupy the same interface with Arf1 as COPIg Phe71; importantly, there is no equivalent to either of these valines in COPIg suggesting these may be the reason for the discrepancy between the orientations of AP-3d Phe77 and COPIg Phe71. Additional, to the aforementioned interactions COPI also utilizes Phe77 to make another large hydrophobic interaction with Arf1; while this Phe is only partially conserved to AP-1g Ala75, both chemically very different to AP-3d Ser84. Although relatively well conserved and all are capable of binding Arf1 in vitro, these small differences in modes of binding likely dictate preference of association with different members of the Arf family in vivo. 7.3 Structural basis of Arf1 binding to AP-3 163 Fig. 7.5 Comparison of Arf1 binding sites (A) Superimposition of all models listed subsequently against Arf1 coloured by hydrophobicity (B) Arf1 interaction with AP-1 g (mauve) (6CM9). (C) Arf1 interaction with AP-3 d (steel blue) (this thesis) (steel blue). (D) Arf1 interaction with bovine COPI g (green)(3TJZ). 164 The orientation of Arf1 when bound positions the N-terminus facing towards the membrane, allowing for the amphipathic helix and N-terminal myristoylation to insert into the bilayer and taken together with the previously suggested PI35P2 binding site on the N-terminus of the d- trunk, I can suggest both these processes, although independent can happen simultaneously and thus they may work together to hold the AP-3 complex to the membrane. Likewise, this also suggest that AP-3 adopts a similar conformation on a membrane as previously shown for other adaptors such as COPI, AP-2 and AP-1 (Dodonova et al., 2015, Kovtun et al., 2020, Hooy et al., 2022). In summary the d solenoid helices 3-6 form a central hydrophobic wedge surrounded by charged residues that pack into the Switch I and II regions of Arf1. As previously described, the interface is only accessible when Arf1 is in the GTP bound state, as when in the GDP bound state, strand 2 caps the beta-sheet occluding the interaction site. While similarities exist between the modes of binding for adaptor complexes, there is a large degree of sequence degeneracy, with binding tolerating a range of hydrophobic side chains. 0 165 Chapter 8 The AP-3 complex on membranes 8.1 The AP-3 complex and lipid nanodiscs As the structure of the AP-3 core in solution has thus far proved intractable for high resolution, I moved to trying to understand the structure of the AP-3 core on membranes. Unlike AP-1 and AP-2 which we know undergo a conformational change upon binding to the membrane, it is unclear whether the AP-3 complex would take up the same conformation observed in solution or an even more extreme open conformation similar to that of AP-2 open plus or COPI (Wrobel et al., 2019, Dodonova et al., 2015). Lipid nanodiscs have been widely utilised as a structural tool to observe transmembrane proteins in a near-native membrane environment (Denisov and Sligar, 2016). Lipid nanodiscs are lipid proteins complexes in which the protein, which contains repeats from the ApoA1 gene, form a single continuous alpha helix which circularizes to encompass a single leaflet of the lipid bilayer. Together two copies of the nanodisc protein are capable of encompassing a patch of membrane to stabilise a lipid bilayer. However, nanodiscs have been under-utilised as a tool for understanding the structure of peripheral membrane proteins, such as coats. Thus, I aimed to understand the structure of AP-3 reconstituted onto nanodisc membranes. The majority of transmembrane proteins visualized using lipid nanodiscs have had relatively compact transmembrane regions, thus the diameter of the lipid nanodiscs has rarely been more than 12 nm. As AP-1 and AP-2 in the open conformation are ~12 nm in diameter I opted for the spNW nanodiscs constructs which utilise a SpyTag:SpyCatcher to stabilise larger discs with diameters upwards of 12 nm (Zhang et al., 2021). The spNW15, spNW25 and spNW50 constructs encode 15 nm, 25 nm and 50 nm respectively were expressed and purified as outlined in (Zhang et al., 2021). In addition to this protocol, a gel filtration step was added to remove contaminants. Both 15 nm and 25 nm nanodiscs were reconstituted with lipid mix of 43% DOPC, 28% DOPE, 10% DOPS, 10% Cholesterol, 5% PI35P2, 3% PI3P and 1% DiI. The peak of the fraction was 166 concentrated and frozen for storage prior to use (courtesy of Rebecca Taylor, Briggs lab). Several attempts to reconstitute the complex were made with a variety of optimizations to time, order that the reagents were added and their concentration. In brief, the protocol used went as follows: aliquots of purified N-terminally mysistoylated-Arf1 (courtesy of Marius Boicu, Briggs lab) were incubated with reconstituted lipid nanodiscs, purified AP-3, excess GTP and EDTA (to remove Mg2+ liganding the GDP phosphates); after a short incubation (10-30 minutes), this preparation was supplemented with excess MgCl2 to quench the EDTA and hopefully lock Arf1 in a GTP bound conformation. To assess the success of these reconstitutions, the sample was analysed by mass photometry to obtain a per-particle mass measurement and an insight of the heterogeneity within the sample. While the reconstituted sample showed a species of increased mass as compared with the individual components, there was still a significant number of particles which were smaller than the expected AP-3:Arf1:Nanodisc sample. This was trialled with all sizes of nanodisc. In an effort to directly visualise the reconstituted assembly, the same sample was vitrified on quantifoil holey carbon (1.2/1.3 300 mesh) grids and screened. Despite the increased size observed by mass photometry, grids appeared mostly devoid of AP-3:nanodisc particles with AP-3 particles aggregated and with significant yolking. Despite significant efforts, this approach proved unsuccessful. 8.2 The AP-3 complex and Arf1-linked-nanodiscs 167 8.2 The AP-3 complex and Arf1-linked-nanodiscs While attempting to troubleshoot this procedure, I speculated as to why it had not been successful: two main hypotheses were considered. First was the myr-Arf1 was non-functional, although this seemed less likely as previously this batch had been tested for GTP-dependent tubulation of liposomes. Another more onerous hypothesis was that the EDTA induced chelation of metal ions for nucleotide exchange was incompatible with either the nanodiscs (most likely the lipids) or the AP-3 protein itself. One alternative would be to use the use of a Arf1 GEF (ARNO Sec7 domain) to load the GTP, nevertheless this would not account for non- functional myr-Arf1. A different solution would be to eliminate the need for nucleotide exchange altogether. GTP locking mutations of Arf1 have been well characterized, but use of these poses a different problem, how to specifically attach this to a nanodisc as in the GTP bound state any N-terminal lipid moiety would be constantly exposed and prone to aggregate. Thus, a radical solution was proposed: I directly cloned a constitutively active Arf1 (Q71L) onto the C-termini of the lipid nanodisc construct to create an automatic AP-3 recruiting nanodisc (Figure 8.1). Fig. 8.1 Nanodisc-link-Arf1 theory (A) Domain architecture of modified pSPNW vectors to make Nanodisc-link-Arf1 protein. (B) Diagram of circularized and reconstituted nanodisc-link-Arf1 construct. (C) Diagram of AP-3 bound to reconstituted nanodisc-link-Arf1. 168 While unorthodox, this approach would also offer some distinct advantages over the use of soluble Arf1. Firstly, as previously, mentioned nucleotide exchange was no longer required, removing any incompatibility issues of EDTA. Secondly, each fully formed nanodisc could only have two copies of Arf1, fixing the stoichiometry to prevent the possibility of the surface of the nanodisc from being overcrowded by excess Arf1 which would inhibit subsequent the recruitment of AP-3. Finally, as the nanodisc protein forms dimers, this effectively makes the nanodisc and both Arf1 molecules intramolecular, lowering the degrees of freedom, in turn causing the dimeric nanodisc-Arf1 complex to behave as a higher affinity interaction for the AP-3 complex through an avidity effect. A construct was created from the spNW15 construct with a C-terminal additional 20 amino acid flexible linker (4xSGAGS) followed by D13-Arf1 (Q71L), subsequently this is referred to as the spNW15-link-Arf construct (Figure 8.1.A). This construct was expressed and purified as before and reconstituted with a simplified lipid mix (PC:PE:PI35P2 80:16:4%). As a trial, once reconstituted, this preparation was incubated with bacterially expressed AP-3 missing the C-µ3 domain for one hour. In addition to the previous protocol, the reconstituted sample was purified by gel filtration in an attempt to assess heterogeneity of the reconstitution, as well as obtain a more homogeneous sample for single particle analysis. Individual fractions of the gel filtration were resolved by SDS-PAGE alongside the input spNW15-link-Arf1 nanodiscs and AP-3 (Figure 8.2). The prominent bands for both the spNW15-link-Arf1 and AP-3 were of similar size and thus overlapped, complicating interpretation of the trace. However, some of the minor bands were unique to each species and coeluted in fractions 19-23 (Figure 8.2.B). 8.2 The AP-3 complex and Arf1-linked-nanodiscs 169 Fig. 8.2 Nanodisc-link-Arf1:AP-3 reconstitution (A) Gel filtration trace of bacterially expressed AP-3 core (µ3 3 1-150) incubated with reconstituted Nanodisc-link-Arf1 superose 6 10/300gl. (B) SDS PAGE of fractions from gel filtration. 170 Fig. 8.3 Nanodisc-link-Arf1:AP3 vitrification and 2D classification (A) 73000x magnification micrograph of fraction 19 with unique classes shown outlined in blue. (B) 73000x magnification micrograph of fraction 21. (C) 2D classification of particles collected from both particles collected from both fractions 19 and fractions 21. 8.2 The AP-3 complex and Arf1-linked-nanodiscs 171 Samples from both fractions 19 and 21 were vitrified on Quantifoil holey carbon grids for single particle cryo-EM. Initial screening proved promising with the observation of new “box- shaped” particles as well as circular particles assumed to be free nanodiscs (Figure 8.3.A and 8.3.B). A single particle cryo-EM dataset was collected from the grids made with fraction 19 and 21 using a Titan Krios G3 equipped with Gatan K3 detector. Collected micrographs were motion corrected, CTF corrected, and picked using the (University of Cambridge) Department of Biochemistry’s installation of WARP (Tegunov and Cramer, 2019). Subsequent particle stacks were imported into Cryosparc and pooled for further processing. Initial 2D classes showed the typical C-shaped AP-3 class as well as several new classes. Encouragingly, several of these new classes showed high resolution features such as helices as well as less discernible blurry features (Figure 8.3.C). Ab initio reconstruction of these particles yielded a volume which was obviously different to the previous solution AP-3 complex (Figure 8.4.A and Figure 8.4.B). Through a series of refinements and further selections the initial volume was subsequently refined to a 2.9 Å map (Figure 8.4.C and 8.4.D). This new map still showed some level of preferred orientation but to far lesser extent than the previous solution structure (Figure 8.4.E). Interpretation of that map proved difficult as there was marked variability in local resolution. At the centre of the structure, in what was assigned as part of the δ solenoid, individual helices and side chains could be clearly discerned (Figure 8.4.C), whereas in more peripheral regions such as the N- termini of either of the large subunits and their interaction with Arf1 was less well resolved (Figure 8.4.B). Fitting of previously derived Alphafold2 models of the AP-3 open core likewise proved near impossible without rebuilding segments of the model. Overall, it appeared that the patch of membrane supported by the nanodisc was smaller than anticipated and that the AP-3 complex appeared “squashed” on the membrane surface. Interpretation of this map, suggested that like other AP complexes on a membrane, the b3 and N-µ3 subunits lay mostly flat on the membrane with the interaction site on d forming a small arch. Likewise, the N-terminal s3d hemi-complex contacted the membrane but appeared to be twisted into a conformation that had previously not been observed. Due to the poor local resolution and taken together with previous observation as to the flexibility of the s3d hemi-complex, it was suggested that the spNW15- link-Arf1 had stabilised a highly strained artefactual conformation of the AP-3 core. Further supporting this theory was the observation that the Arf1 densities appeared to be axial in 172 position compared to one another (Figure 8.4.B). It had been assumed that the two copies of the spNW15-link-Arf1 that form a complete nanodisc would be able to rotate independently of one another once reconstituted to facilitate AP-3. This was clearly not the case as AP-3 had strained to fit these presumably less favourable geometries. Thus, while the reengineering of nanodiscs had solved some of the apparent incompatibly of the AP-3 complex with the air water interface, this also introduced significant artefacts complicating interpretation. While this had not proved successful in this instance, further optimization such as using a larger sized nanodisc (spNW25) and coupled with longer linkers connecting the Arf1 would likely eliminate these issues making this process far more tractable approach. 8.2 The AP-3 complex and Arf1-linked-nanodiscs 173 Fig. 8.4 Nanodisc-link-Arf1:AP3 3D reconstruction (A) Refined EM map of Nanodisc-link-Arf1:AP-3 3D reconstruction. (B) Gaussian filtered map coloured by interpreted chains. (C) high resolution features of side chains on solenoid helices. (D) Gold standard Fourier shell correlation (GS-FSC) of Nanodisc-link-Arf1:AP- 3 3D reconstruction. (E) orientation distribution plot of 3D reconstructed Nanodisc-link-Arf1:AP-3. 174 8.3 AP-3 reconstitution on membranes To understand the conformation of AP-3 on membranes, I turned to reconstituting the whole AP-3 Arf1:GTP complex with cargo peptides and phosphoinositides on proteoliposomes. Unlike lipid nanodiscs, this also brought the added possibility of resolving the AP-3 complex in context with neighbouring complexes and hopefully less likely to produce artefacts. Previous attempts with Katarzyna Ciazynska to solve the structure of AP-3 on membranes by cryo-electron tomography (cryo-ET) had proven difficult due to a combination of factors. We had successfully managed to dock the AP-3 complex to membranes using myr-ARF1 with TGN38 cargo peptide and obtained fuzzy coated liposomes with irregular morphology. Yet ultimately despite a lot of work, subtomogram averaging proved impossible as a reference volume proved unobtainable. In recent years, cryo-ET coupled with subtomogram averaging has been hailed as the solution to understanding protein in complex samples in a more native environment (Dodonova et al., 2015, Kovtun et al., 2018, Gemmer et al., 2023). One interesting phenomenon has been the view that samples on membranes are too heterogenous for single particle analysis (SPA) and insufficient particle number along with inadequate viewing angles makes structural solution impossible without tomography. Thus, many ground-breaking publications of structures on membranes have been solved by tomography. However, recently there has been an increase in publications challenging this assumption with complex and heterogenous samples being solved to high resolution on membranes using SPA (Falzone and MacKinnon, 2023, Stacey et al., 2023, Claire et al., 2024). Thus, I aimed to solve the structure of the AP-3 complex on proteoliposomes but this time using SPA or cryo-ET. Although reconstitution of AP-3 on synthetic membranes has been achieved before, it had only been used for limited numbers of tomograms or images (Drake et al., 2000). For structural studies, a sufficiently high yield and ideally uniform protein liposome complex is needed for averaging; accordingly, this requires serial optimization and screening of the sample. 8.3 AP-3 reconstitution on membranes 175 As I was only interested in proteins bound on a membrane, I sought to first optimize the size and morphology of the liposomes to ensure I had an adequate number of positions for collecting. However, as liposomes behave differently when coated with protein and proteins can also affect the morphology of the liposome, identical coating reactions were carried out on liposomes of different sizes with a 1:2 stoichiometry of AP-3 complex to myr-Arf1. The reactions were incubated at room temperature on the bench for 1 hour before being applied to freshly glow discharged C-flat holey carbon grids. Screening of these initial samples showed that coating had been successful for all sizes of liposomes, although in all cases there was significant background, uneven coating and occasional aggregation of protein. Likewise, as is common with liposomes, the sample exhibited clustering of liposomes in patches and had a preferential interaction with the carbon support layer instead of the grid holes. However, from this screen, it was clear that serial extrusion had proven successful in generating a range of liposome sizes (Figure 8.5). This range of sizes followed a general trend with the pore size of the extrusion filter, 50 nm liposomes being the smallest on average and spread the most evenly between across the grid and within grid holes (Figure 8.5.A). The liposomes generated by 100 nm were generally smaller than those generated using 200 nm extrusion filters; however, in both cases the larger liposomes from each condition were attached to the carbon support preferentially to the grid hole (Figure 8.5.C). Additionally, a more diverse range of liposome morphologies was observed for the 200 nm liposomes. As the smaller 50 nm liposomes were more frequently found in the grid hole and more regular in size, this extrusion regime was used for subsequent experiments. These results suggested there was an optimum size for spreading of liposomes and for entering the grid hole. Thus, the coating for 50 nm liposomes was optimized further. 176 Fig. 8.5 Optimizing liposome size AP-3 coating reactions performed on liposomes extruded with a pore size of (A) 50nm (B) 100nm and (C) 200nm. (scale bar = 100nm) 8.4 Optimizing AP-3 liposome coating 177 8.4 Optimizing AP-3 liposome coating In addition to preferential clustering of liposomes on the grid (Figure 8.6.A), in comparison to previously achieved AP-2 samples, the coating of liposomes by AP-3 was patchy with some liposomes devoid of any coating at all; some with incomplete coating and some where the coat appeared to be overly thick and heterogenous subsequently assigned as “over-coating” (Figure 8.6.B). To optimise production of suitable coated liposomes, several independent lines of enquiry were followed. Firstly, to limit the possibility of aggregation of the AP-3 complex, AP-3 alone was spun at 16,000 RCF for 15 min prior to coating reactions. While it was difficult to tell if this had significant impact on the final grids, it was not obviously detrimental; thus, it was repeated before each coating reaction. The incubation time of the coating reaction was varied to include both shorter and longer time points. Variation of incubation time seemed to have a net negative effect with shorter time points (15 or 30 minutes) generally having less complete coating and higher background, while longer time points (1.5 or 2 hours) showed little difference in coating as compared to before. From this result, it was concluded that the coating reaction only needed to incubate to give sufficient time for recruitment. Accordingly, all subsequent coating reactions were incubated for ~1 hour. To prevent overcoating of AP-2 on liposomes, it had been observed that limiting the incubation time before making grids had yielded more even coating. However, it should be noted that AP- 2 only requires PI45P2 to bind membrane in vitro, whereas AP-3 also requires myr-Arf1:GTP as well as PI35P2. Incubation of myr-Arf:GTP alone with liposomes for 45 minutes followed by addition of the AP-3 complex also failed to recruit the AP-3 complex uniformly with a range of coating efficiencies throughout the sample. 178 Fig. 8.6 Optimizing liposome coating (A) Square view of liposomes coating reactions on grid. (scale bar = 12.5 uM) AP-3 coating reactions performed with (B) equal Arf1 and AP-3 stoichiometry (2:1) (B) stoichiometric excess of Arf1 to AP-3 (5.2:1) and (C) stoichiometric excess of Arf1 on a Multi-A grids. (scale bar = 100nm) 8.4 Optimizing AP-3 liposome coating 179 The ratio of AP-3 to myr-ARF1 was changed so that myr-Arf1 was in stoichiometric excess. This increased stoichiometry of myr-Arf1 showed a drastic effect in reducing background and increasing the evenness of coating on liposomes (Figure 8.6.B and 8.6.C). A 1:5.2 ratio was settled on a sufficient. This result is possibly due to the recruitment of AP-3, at least in vitro, being a cooperative process where the presence of an AP-3 complex aids the recruitment of the second complex creating a positive feedback loop. Finally, gold Multi-A grids were trialled alongside C-flats in order to evaluate if different hole shapes and sizes facilitated better spreading of liposomes. In contrast to more standard C-flat grids, gold Multi-A grids feature an array of different hole sizes, shapes (elliptical vs spherical) and densities at which the holes are packed together. Qualitatively, most smaller liposomes were retained in the smaller spherical holes while larger liposomes were more frequently found on the edge of elliptical holes. In addition to this coarse segregation, it was generally observed that there was less overcrowding of coated liposomes on Multi-A grids (Figure 8.6.D). 180 8.5 Imaging of AP-3 on spherical liposomes. A single particle cryo-EM dataset of 3975 micrographs was collected from spherical liposomes on multi-A grids. Micrographs were motion and CTF corrected using Relions implementation of MotionCor2 and CTFind4. 100 corrected micrographs were manually picked for training a machine learning based picker in CrYOLO. CrYOLO identified 352,817 particles. Micrographs were imported into cryosparc and picked particle coordinates extracted. Multiple rounds of 2D classification showed mostly membrane classes. After several rounds of classification, an unresolved blur above the membrane was resolved into a discrete blob (Figure 8.7). However, 3D reconstruction of these particles yielded spurious densities. Fig. 8.7 2D classification of AP-3 on liposomes Cryosparc 2D classification of AP-3 on spherical liposomes 8.6 Occasional tubule While screening for ideal coating conditions, in several reactions it was noted that in rare instances large liposomes were tubulated and appeared to be protein coated (Figure 8.8). The coated tubules appeared to be of an even diameter of ~30-40 nm with a regular ordered arrangement. Further inspection revealed that a slight helical twist could be observed in coats. Although Arf1 is a known to tubulate membranes in vitro, these tubules are much smaller in diameter (~8-10 nm) than the observed tubes. Measuring of the thickness of the coat (distance 0 181 of the coat from the edges of the membrane) suggested it was too thick to be accounted for by Arf1 alone. As AP-3 has previously been localized to tubular endocytic compartments by immuno-EM and more recently by stimulated emission depletion microscopy (Peden et al., 2004, Theos et al., 2005, Stockhammer et al., 2023), it was suggested that AP-3 may tubulate membrane as an alternative method of membrane deformation to the previously debated clathrin dependent vesiculation. Taken together this might suggest that tubulated membranes coated by AP-3 in vitro may mimic tubular carriers in cells. Yet with only three instances of tubules being observed in these samples more tubules would be required to generate sufficient particles for structural analysis and averaging. Fig. 8.8 AP-3 Coated tubules (A) tubule found to be deforming large liposome. (B) AP-3 coated liposome completely tubulated. (scalebar=100nm) 182 8.7 Optimization of tubulation Since tubules had only been observed from larger liposomes preparations and in (Figure 8.8.A) the tubule can clearly be seen emanating from a large coated liposome, this suggested one or both of the following; A) larger liposomes were preferentially tubulated as they exhibit less membrane tension, lowering the energy barrier required to be overcome for deformation from a sphere to a tube. B) Alternatively, there needed to be a sufficient number of adaptors present on a single membrane to allow for enough force to be generated for efficient membrane deformation. Regardless of which scenario was really taking place, both scenarios could be simultaneously addressed by docking more adaptors onto fewer, larger membranes. Thus, giant unilamellar vesicles (GUVs) were generated as described in method section 2.9, using the same lipid mix as previously described (PC:PE:PI35P2:DiI:cholesterol). Coating reactions were performed with stoichiometric excess myr-Arf1 to AP-3 (1:5.2). As the resultant GUVs are filled with sucrose, they are dense and not buoyant in buffer, hence settle during a coating reaction. Therefore, both the supernatant and settled GUV pellet were applied to freshly glow discharged C-flat and gold Multi-A grids. As predicted, there was a significant increase in the number of coated tubules across the grid (Figure 8.9). The supernatant seemed to contain more even and uniform tubules while the pellet contained less evenly coated tubules mixed with untubulated membrane and larger electron dense guvs. Logically this makes sense as tubules have a higher ratio of membrane to soluble contents thus there is less dense sucrose in comparison to buoyant membranes, hence the tubules but not the unbudded GUVs, float in the supernatant. Although not pertinent to the structure of the AP-3 coat, this does illustrate a critical role for tubules in preferentially sorting and delivering transmembrane cargoes over soluble cargo, by virtue of their high surface area to volume ratio, they are able to enrich more transmembrane proteins per carrier than a spherical vesicle. 8.7 Optimization of tubulation 183 Fig. 8.9 Reconstitution of the AP-3 complex on GUVs (A) 17500x magnification micrograph of tubules formed by reconstitution of the AP-3 complex on GUVs. (scalebar=500nm) (B) 73000x magnification micrograph of AP-3 tubules. (scalebar=100nm) 184 8.8 Single particle analysis of AP-3 tubules A single particle dataset of 15,298 micrographs was collected of AP-3 coated tubules using a Titan Krios G5 (MPI Biochemistry). The dataset was processed as described in method section 2.16.3. In brief, micrographs were motion corrected using MotionCorr2 and CTF corrected with GCTF (Both Relion4 implementation). 100 corrected micrographs containing AP-3 were manually picked using CrYOLO 1.4. Outputted picks were used to train a machine learning model for unsupervised training. The subsequently trained model was used to pick particles across all 15,298 micrographs using CrYOLO 1.9.4. The subsequent particles picks were exported from Relion4, both corrected micrographs and particle picks were subsequently imported into Cryosparc (Structura biosciences). The picked particle locations were initially extracted from the imported micrographs using a variety of box sizes and Fourier cropped for computational simplicity. Extracted particle stacks were initially classified by iterative rounds of 2D classification. It was found that larger box sizes (588 pixels/705.6Å) were able to contain the entire ensemble on a segment of coated tube resulting in 2D classes while smaller particles struggled to align different views across the adaptor with no obvious top views and poorly aligning side views. The 2D classes for large segments of a tubes seemed to show well-ordered lattice with some classes displaying arch-like assembly presumed to be copies of the AP-3 complex (Figure 8.10). 8.8 Single particle analysis of AP-3 tubules 185 Fig. 8.10 2D classification of AP-3 tubules Segments of AP-3 coated tubules classified by cryosparc 2D classification. 186 As previous screening images of the tubules appeared to have a pronounced helical twist and 2D classes contained arches on both sides which were exactly offset by half a repeat of an arch, it was proposed that the tubules may have helical symmetry. To check if the tubes were indeed helical, all particles belonging to a 2D class were converted to power spectra and then averaged against all other particles within the same 2D class using average power spectra job (Cryosparc V4.4). The resulting averaged power spectra showed multiple symmetric layer lines with regularly ordered points (contains information about axial density and spacing of a tube) above and below an equatorial line (contains information as to the density distribution of a tube) all indicative of helicity at least at low resolution (Figure 8.11.A). After multiple stringent rounds of 2D classification, particles were used to generate an initial C1 symmetrical “helical” volume without the imposition of helical symmetry. This process resulted in a tubular volume with four lines of arch-like ridges running the length of the tube. To try to obtain an initial estimate of helical parameters (rise (DZ), pitch (p)and degrees of rotation) the C1 tube was used as input for Cryosparc’s symmetry search utility (Figure 8.11.B). Searching for just rise (DZ) & rotation (f), the volume initially showed two possible helical parameters of different confidences. The first was predicted to have a small pitch with a rise (DZ) of ~15-20 Å and rotation of ~90 °, thus an estimated pitch of ~60-80Å. However, the estimated confidence in this was limited as due to its Mean score error (MSE) score. The second helix was far more confident but predicted a rise (DZ) of ~80 Å and rotation (f) of ~1.5 ° producing a colossal pitch estimation, suggesting a full rotation of the tubule would take several hundred adaptors. As a helical array may be described by more than one symmetry operation, both of the previously suggested symmetries (and near symmetries) were searched with little success. 8.8 Single particle analysis of AP-3 tubules 187 Fig. 8.11 AP-3 tubules are helical at low resolution (A) Average power spectra of 2D classes from cryosparc. (B) Predicted analysis of helical symmetry from cryosparc symmetry search. 188 8.9 Cryo-ET and subtomogram averaging of AP-3 tubules An alternative method of obtaining helical parameters is by directly observing the tubes in real space. Previously, the Briggs lab had used Cryo-electron tomography (Cryo-ET) and subtomogram averaging to obtain helical parameters by directly reconstructing 3D volumes. In the event that the tubes were only helical at low resolution and not “helical enough” for enforcing of helical symmetry, tomography would also allow for an independent method of obtaining a 3D volume. Thus 32 tilt series were collected from which a single tomogram was reconstructed as a pilot to establish the pipeline. Due to the lack of fiducials the tilt series was reconstructed in Bin4 using patch tracking alignment in IMOD (Mastronarde and Held, 2017). Within the tomogram a total of 9 tubes were picked and used for reference free alignment (Figure 8.12.A and 8.12.B). On the assumption that the previously estimated helical symmetry of four asymmetric units per helical turn was correct, the tubes were sampled at double the expected frequency (8 overlapping subunits per turn with a rise of 80 Å). The randomly sampled positions were extracted as subtomograms in subtom MATLAB R2021b (v911) and used for subtomogram averaging. Initially, subtomograms were aligned vertically (Z-plane) to separate the membrane and protein layers. Following this a tubular cut-out mask was generated (as outlined in (Peukes et al., 2020)) and given with increased freedom to sample laterally in the X and Y planes but with limited movement in the Z and no angular rotation. An increasingly helical map appeared with corkscrew like features. The same mask as before was reiterated but with the addition of sampling +/-45 ° angular rotation in Y/Q which resulted in a map with similar appearance to the previously observed 2D classes (Figure 8.11.C and 8.11.E). Of immediate note, the new map reproduced the arches seen by 2D classification, each individual arch was assumed to correspond to the joining of the b3 and d subunits and thus be one AP-3 complex. Cross section orthoslices of whole tubes and lattice maps both clearly showed six copies of the AP-3 complex were required to form one helical turn (Figure 8.12.D). Thus, all previously helical parameter estimations searched for were incorrect. However, the limited resolution combined with the pseudo 2-fold symmetry of the AP-3 complex complicated direct interpretation (Figure 8.12.A 8.12.B). 8.9 Cryo-ET and subtomogram averaging of AP-3 tubules 189 Fig. 8.12 Cryo-electron tomography of AP-3 tubules (A) Slice of lowpass filtered tomogram of AP-3 coated tubes (scalebar=100nm). (B) Subtomogram averaged volume simulated over extracted positions from the tomogram (ArtiaX). (C) Oblique view of subtomogram averaged volume. (D) Orthoslice cross section of tubule of from subtomogram averaging. (E) Close up of AP-3 filament on side of averaged tubule. 190 At the feet of the arches a strong round density can be seen which separates from the main arch at high thresholding, which albeit at risk of overinterpretation, I would interpret to be Arf1. Around this Arf1 forms a C2-like interface with another Arf1 and large AP-3 subunit (either d or b3) (Figure 8.13.B). However, the nature of this C2-like interface is vital to understanding the helicity of the tube and explicitly whether it possesses processivity. At this stage, the low resolution combined with AP-3 integral 2-fold symmetry makes identifying which large subunit impossible. However, three obvious solutions present themselves: 1) The C2-like interface is composed of a heterotypic d:b3 interaction mediated by two Arf1 molecules. This interpretation would result in a directionality of the helix and would suggest the helix is continuous and should be solvable by helical processing with 6 asymmetric units per rotation. In other words, each AP-3 complex is one asymmetric unit (Figure 8.12.C). 2) The C2-like interface is composed of a homotypic interaction of either d:d or b3:b3 mediated by two Arf1 molecules and both interfaces would have to be present to produce the resultant tube. Likewise, this should be solvable by helical processing, if truly helical; however, the two unique linkages would require two AP-3 complexes per asymmetric unit resulting in a pseudo 6-fold which would in reality be a 3-fold helical symmetry consisting of a trimer of dimers (Figure 8.12.D) 3) The tube is not processive and the C2-like interface is composed of a mixture of d:b3, d:d and b3:b3 contacts resulting in a tube that appears helical at low resolution but not at high resolution. The resulting tube is thus not solvable by helical processing and each subunit must be aligned individually. This is most likely solvable only in C1 by subtomogram averaging. Thus far attempts to enforce 6-fold and 3-fold helical symmetry have not been successful, possibly due to either the tube not being processive (hypothesis 3) or the input particles not being helical enough for high resolution solution. 8.9 Cryo-ET and subtomogram averaging of AP-3 tubules 191 Since all of these three hypotheses should be directly provable by subtomogram averaging, future work will focus to increase the resolution of subtomogram averaging by reconstruct more tomograms, with more particles and CTF corrections of tilt series (nova CTF or WARP M) which will likely yield a higher resolution structure. Fig. 8.13 organization of the AP-3 complex on tubules (A) Close up of three arch-like structures assumed to be AP3. (B) top view of linkage between AP-3 dimers with potential AP3 subunits labelled and two Arf1 (yellow) forming a C2-like interface. (C) Potential asymmetric unit of a helix with 6-fold symmetry mediated by d:b3 linkages. (D) Potential asymmetric unit of a helix with pseudo 6-fold symmetry mediated by d:d and b3:b3 linkages. 192 8.10 Summary and discussion Although thus far a high-resolution structure of the whole AP-3 core has remained elusive, taken together the data in this investigation suggest that the regulation of the AP-3 complex is unique from that of other AP complexes and reveals the intricate details of conserved mechanisms from recruitment to cellular membranes, cargo selection and membrane deformation. Strikingly the first observation that the AP-3 core is open in solution, stands in stark contrast to what has been shown thus far for the AP-1 and AP-2 complexes. As AP-3 had previously assumed closed solution in solution this begs several questions: • Why is there not a closed form? • What prevents aberrant binding of sequences that are not sorting motif that may conform to YXXΦ and dileucine cargo in the cytoplasm? As previously stated, the comparisons of other AP-1 and AP-2 show that β1 and β2 (which are almost identical in sequence) Val365 packs into the Φ pocket of μ1 and μ2 respectively, while β3 Gln614 is too large and polar to make such a contact, although this is yet to be experimentally verified. While assessing the ability of a cargo adaptors to bind cargo motifs in solution was performed in vitro (Figure 5.6), performing such an experiment in the cytoplasm of a living cell is generally not performed and would likely be fraught with difficulties in validation. Circumstantial evidence work from (Riera-Tur et al., 2022) may however support this assertion. By expressing synthetically designed amyloid proteins in cortical neurones, the amyloids were able to bind μ3, most likely by forming a YXXΦ in solution, which is preconformed into a beta sheet (as virtue of being an amyloid). Furthermore, swollen non- functional lysosomes were observed, which was at least partially attributed to lack of AP-3 function. Yet in spite of this, quantification of the other subunits showed that they had 8.10 Summary and discussion 193 decreased in cells expressing these synthetic amyloids suggesting this may be the only the μ3 subunit binding in the absence of a complete complex. By yeast-2-hybrid assays, I reproduced and further characterized the binding of the δ ear to σ3 found by (Lefrançois et al., 2004). Using alphafold2 prediction guided Y2H experiments I was able to show that the δ ear and the N-termini of δ trunk compete for binding of σ3. I suggest this acts as a mechanism of inhibition by simultaneously breaking the binding sites for of phosphoinositide, Arf1 and dileucine cargo, creating a stable open conformation. However, this does not yet account for the position of the C-μ3 subunit within this open complex. Subsequent work aimed to understand the mechanistic underpinnings of recruitment of the AP- 3 complex to endosomal membranes. Firstly, the binding of Arf1 to the δσ3 hemicomplex was directly visualized by X-ray crystallography showing an interaction interface conserved across the AP family. Notably while this agreed with previous literature, published immunoEM data showed that both AP-3 and AP-1 sort from the same endosome but form discrete subdomains, despite being both being recruited by Arf1 (Peden et al., 2004). This suggested that either self- association or another factor was vital in driving segregation of these adaptors. 194 As other members of the AP family bind to phosphoinositides for recruitment, I considered whether AP-1 and AP-3 utilise different phosphoinositides to fine tune their localisation. The major phosphoinositide binding site at the N-terminus of δ was tested using the S3D construct, which bound strongest to the late endocytic lipid, PI35P2. This was surprising as PI3P is the most abundant phosphoinositide on the early endosome (from which AP-3 buds) and had previously been said to bind (Baust et al., 2008). Similarly, this binding was tested by floatation assay in yeast which found that both PI3P and PI4P had a similar affect in recruiting the AP-3 complex; however, this was only marginal when compared to Arf1 and the authors did not test other lipids (Schoppe et al., 2020). The binding of PI35P2 has raised three interesting scenarios for trafficking by the AP-3 complex: 1) The AP-3 complex buds from a later compartment However, this is unlikely as AP-3 has been localised to the same compartment as AP-1 trafficking and would disagree with the colocalization of Arf1 with early endocytic structures (and not on later compartments). 2) The AP-3 complex binds directly to a small pool of PI35P2 produced on the early endosome prior to or during endosomal maturation. This would require the PI35P2 synthesising enzymes (PIKfyve:Fig4:Vac14 complex) to be recruited to the early endosome. Despite the mechanism of recruitment not being totally understood, the fyve domain of Pikfyve suggests that PI3P maybe necessary and possibly sufficient for recruitment. 3) PI35P2 production only occurs as an endosome matures thus the AP-3 complexes budding is both spatially and temporally regulated through coincidence detection of both Arf1 and PI35P2. Interestingly, this model parallels the Retromer complex which uses a combination of an early endocytic lipid PI3P and the late endocytic GTPase Rab7 (Nakada- Tsukui et al., 2005, Priya et al., 2015, Jia et al., 2016). As both of these occur as an endosome matures, one would predict that a “burst” of sorting events to occur as maturation takes place (most likely temporally intertwined with the Rab5 to Rab7 conversion). 8.10 Summary and discussion 195 Furthermore I show that AP-3 can be reconstituted on synthetic membranes with PI35P2, myr- Arf1:GTP and cargo peptides. Moreover, when sufficient copies of the AP-3 complex are present on a membrane they are able to tubulate to form continuous tubules, which low resolution subtomogram averaging shows is formed of interlocking filaments by interacting with other neighbouring copies of the AP-3 coat. As AP-3 has been localised to endocytic tubules and alone appears sufficient for membrane bending, this may explain the lack of total dependency with clathrin. However, the physiological relevance of reconstituted tubes to the tubules seen in cells remains to be determined. Reassuringly the diameter of AP-3 coated tubules seen protruding from endosomes is ~50 nm while invitro reconstituted tubes are ~40 nm which is not totally dissimilar in diameter but different enough to affect stoichiometry. Even if the reconstituted tubules themselves are not identical to tubules formed in cells, the contacts between the AP-3 complexes maybe functionally important in vivo. Additionally, there is no reason for a tubule formed in cells to be perfectly coated; thus, the in vitro sample may represent a more idealised version of the tubules observed in cells. Thus, once intercomplex contacts have been determined, specific point mutation can be generated to adversely affect the packing of AP-3 on a tubule and determine if this affects the sorting of cargo to lysosomes. Fig. 8.14 Schematic diagram of recruitment of the AP-3 complex 196 Chapter 9 Discussion 9.1 Overview The use of the endocytic system for uptake of nutrients, signalling factors, pathogens has been the focus of research for over 60 years. While the formation of the endosome is transient in nature and the endosome’s maturation and trafficking into and out of the endosome are well- documented phenomena, the molecular basis by which sorting events are intimately linked to endosomal maturation remains only partially understood. In this thesis I aimed to understand how cargo is sorted both into and out of the endosomal system at the molecular level. For this I utilised a number of biochemical, biophysical and structural approaches to study interaction of cargo motifs and their adaptors and adaptors. My main findings can be summarised: The characterization of a novel endosome-to-Golgi sorting motif (TLY) and its interaction with the PH domain of TBC1D23. Currently, this is the only known instance of cargo dependent tethering to an organelle and thus TBC1D23 has been dubbed a “sorting tether”. Further structural characterization into the molecular specificity of the acidic dileucine sorting signal to its adaptors elucidated a previously undiscovered ordered water network, which mediates the interaction. I validated the relevance of this water network further demonstrating its role in preferentially selecting “[ED]XXPL[LI]” cargo motifs, providing molecular context to previous findings which observed differential sorting of acidic dileucine motifs into and out of the endosomal system. Finally structural studies into the molecular architecture of the AP-3 complex which mediates endosome-to-lysosome trafficking, have revealed a surprising absence of conformational regulation; previously thought to be ubiquitous across the AP complexes. As the AP-3 complex exists in the open conformation in solution I propose that the δ-ear functions as an autoinhibition mechanism to simultaneously break the binding sites for Arf1, PI35P2 and acidic 9.2 Concluding remarks and outstanding questions 197 dileucine cargo. Finally, I demonstrate that when reconstituted onto membranes with Arf1, PI35P2 and cargo motifs the AP-3 is capable of oligomerizing into coated tubules which may mimic coated structures in vivo. 9.2 Concluding remarks and outstanding questions While the work of this thesis has made a small advance in our understanding of the mechanisms of sorting individual cargo motifs at the endosome and trafficking machinery that expedites these processes, much of the dynamics of how sorting of a specific coat-adaptor complex is integrated with regards to all endosomal sorting events holistically remain to be determined. Moreover, understanding of the dynamics of these structures in cells remains and how this influences a cargo’s steady state localisation leaves outstanding questions for future research. Open and unanswered questions from this study for future lines of research are suggested as follows: TBC1D23 • What is the function of the WDR11 complex? • Does the TBC1D23 or WDR11 complex have a specificity to a specific coated carrier and if so which ones? (AP1, Retromer, etc) • Are there other vesicular factors outside of TLY motifs that the TBC1D23 and WDR11 complex bind that drive carrier tethering? • Are there other adaptors that can directly associate with the TLY motif? • Are there other proteins that can act as sorting tethers? 198 Sorting of Dileucine motifs • Is the Pro(-1) sufficient to localise an acidic dileucine motif to the lysosome in cells? • Does the AP-3 Thr70 encode a specificity towards [ED]XXPL[LI] motifs over non Pro containing motifs or are dileucines motifs sorted to later endocytic compartments by bulk affinity? • Are there sequences outside of the acidic dileucine motif that convey specificity to either AP-1 or AP-3 which bias a cargo’s sorting to the earlier endocytic (Golgi) or later endocytic (Late endosomes, Lysosome and LRO) compartments? The AP-3 complex • Why is AP-3 open in solution? Is there a need for regulation of the AP-3 core? • Is the δ-ear sufficient to provide inhibition of the AP-3 core in vivo. • Does the binding of PI35P2 influence when, in an endosome maturation, the AP-3 complex is recruited? • Both AP-1 and AP-3 are recruited by Arf1 and yet form adjacent but discrete microdomains, what factors allow for this specificity? • How does the AP-3 drive vesiculation? Is tubulation sufficient or do BLOC-1 or clathrin significantly impact vesicle formation. References ADELMANN, C. H., TRAUNBAUER, A. K., CHEN, B., CONDON, K. J., CHAN, S. H., KUNCHOK, T., LEWIS, C. A. & SABATINI, D. M. 2020. MFSD12 mediates the import of cysteine into melanosomes and lysosomes. Nature, 588, 699-704. AMICK, J., THARKESHWAR, A. K., TALAIA, G. & FERGUSON, S. M. 2020. PQLC2 recruits the C9orf72 complex to lysosomes in response to cationic amino acid starvation. J Cell Biol, 219. ANANDAPADAMANABAN, M., MASSON, G. R., PERISIC, O., BERNDT, A., KAUFMAN, J., JOHNSON, C. M., SANTHANAM, B., ROGALA, K. B., SABATINI, D. M. & WILLIAMS, R. L. 2019. Architecture of human Rag GTPase heterodimers and their complex with mTORC1. Science, 366, 203-210. ANDERSON, R. G., BROWN, M. S. & GOLDSTEIN, J. L. 1977. Role of the coated endocytic vesicle in the uptake of receptor-bound low density lipoprotein in human fibroblasts. Cell, 10, 351-64. ANGERS, C. G. & MERZ, A. J. 2009. HOPS interacts with Apl5 at the vacuole membrane and is required for consumption of AP-3 transport vesicles. Mol Biol Cell, 20, 4563- 74. ANTONNY, B., BERAUD-DUFOUR, S., CHARDIN, P. & CHABRE, M. 1997. N-terminal hydrophobic residues of the G-protein ADP-ribosylation factor-1 insert into membrane phospholipids upon GDP to GTP exchange. Biochemistry, 36, 4675-84. ASENSIO, C. S., SIRKIS, D. W., MAAS, J. W., JR., EGAMI, K., TO, T. L., BRODSKY, F. M., SHU, X., CHENG, Y. & EDWARDS, R. H. 2013. Self-assembly of VPS41 promotes sorting required for biogenesis of the regulated secretory pathway. Dev Cell, 27, 425-37. AVINOAM, O., SCHORB, M., BEESE, C. J., BRIGGS, J. A. & KAKSONEN, M. 2015. ENDOCYTOSIS. Endocytic sites mature by continuous bending and remodeling of the clathrin coat. Science, 348, 1369-72. AZARNIA TEHRAN, D., LOPEZ-HERNANDEZ, T. & MARITZEN, T. 2019. Endocytic Adaptor Proteins in Health and Disease: Lessons from Model Organisms and Human Mutations. Cells, 8, 1345-1397. BAKER, R. W., JEFFREY, P. D., ZICK, M., PHILLIPS, B. P., WICKNER, W. T. & HUGHSON, F. M. 2015. A direct role for the Sec1/Munc18-family protein Vps33 as a template for SNARE assembly. Science, 349, 1111-4. BAR-PELED, L., CHANTRANUPONG, L., CHERNIACK, A. D., CHEN, W. W., OTTINA, K. A., GRABINER, B. C., SPEAR, E. D., CARTER, S. L., MEYERSON, M. & SABATINI, D. M. 2013. A Tumor suppressor complex with GAP activity for the Rag GTPases that signal amino acid sufficiency to mTORC1. Science, 340, 1100-6. BARTUZI, P., BILLADEAU, D. D., FAVIER, R., RONG, S., DEKKER, D., FEDOSEIENKO, A., FIETEN, H., WIJERS, M., LEVELS, J. H., HUIJKMAN, N., KLOOSTERHUIS, N., VAN DER MOLEN, H., BRUFAU, G., GROEN, A. K., ELLIOTT, A. M., KUIVENHOVEN, J. A., PLECKO, B., GRANGL, G., MCGAUGHRAN, J., HORTON, J. D., BURSTEIN, E., HOFKER, M. H. & VAN DE SLUIS, B. 2016. CCC- and WASH-mediated endosomal sorting of LDLR is required for normal clearance of circulating LDL. Nat Commun, 7, 10961. A-2 Crystallographic statistics and screens BAUST, T., ANITEI, M., CZUPALLA, C., PARSHYNA, I., BOUREL, L., THIELE, C., KRAUSE, E. & HOFLACK, B. 2008. Protein Networks Supporting AP-3 Function in Targeting Lysosomal Membrane Proteins. Molecular Biology of the Cell, 19, 1942- 1951. BECK, R., PRINZ, S., DIESTELKÖTTER-BACHERT, P., RÖHLING, S., ADOLF, F., HOEHNER, K., WELSCH, S., RONCHI, P., BRÜGGER, B., BRIGGS, J. A. & WIELAND, F. 2011. Coatomer and dimeric ADP ribosylation factor 1 promote distinct steps in membrane scission. J Cell Biol, 194, 765-77. BLACK, M. W. & PELHAM, H. R. 2000. A selective transport route from Golgi to late endosomes that requires the yeast GGA proteins. J Cell Biol, 151, 587-600. BLOBEL, G. 1980. Intracellular protein topogenesis. Proc Natl Acad Sci U S A, 77, 1496- 500. BLOBEL, G., WALTER, P., CHANG, C. N., GOLDMAN, B. M., ERICKSON, A. H. & LINGAPPA, V. R. 1979. Translocation of proteins across membranes: the signal hypothesis and beyond. Symp Soc Exp Biol, 33, 9-36. BOESCH, D. J., SINGLA, A., HAN, Y., KRAMER, D. A., LIU, Q., SUZUKI, K., JUNEJA, P., ZHAO, X., LONG, X., MEDLYN, M. J., BILLADEAU, D. D., CHEN, Z., CHEN, B. & BURSTEIN, E. 2023. Structural organization of the retriever-CCC endosomal recycling complex. Nat Struct Mol Biol. BOHDANOWICZ, M., BALKIN, D. M., DE CAMILLI, P. & GRINSTEIN, S. 2012. Recruitment of OCRL and Inpp5B to phagosomes by Rab5 and APPL1 depletes phosphoinositides and attenuates Akt signaling. Mol Biol Cell, 23, 176-87. BOLL, W., OHNO, H., SONGYANG, Z., RAPOPORT, I., CANTLEY, L. C., BONIFACINO, J. S. & KIRCHHAUSEN, T. 1996. Sequence requirements for the recognition of tyrosine-based endocytic signals by clathrin AP-2 complexes. Embo j, 15, 5789-95. BONIFACINO, J. S. 2004. The GGA proteins: adaptors on the move. Nat Rev Mol Cell Biol, 5, 23-32. BORNER, G. 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. J Cell Biol, 197, 141- 60. BOWMAN, S. L., BI-KARCHIN, J., LE, L. & MARKS, M. S. 2019. The road to lysosome- related organelles: Insights from Hermansky-Pudlak syndrome and other rare diseases. Traffic, 20, 404-435. BOWMAN, S. L., LE, L., ZHU, Y., HARPER, D. C., SITARAM, A., THEOS, A. C., SVIDERSKAYA, E. V., BENNETT, D. C., RAPOSO-BENEDETTI, G., OWEN, D. J., DENNIS, M. K. & MARKS, M. S. 2021. A BLOC-1-AP-3 super-complex sorts a cis-SNARE complex into endosome-derived tubular transport carriers. J Cell Biol, 220. BRIGHT, N. A., DAVIS, L. J. & LUZIO, J. P. 2016. Endolysosomes Are the Principal Intracellular Sites of Acid Hydrolase Activity. Curr Biol, 26, 2233-45. BRIGHT, N. A., GRATIAN, M. J. & LUZIO, J. P. 2005. Endocytic delivery to lysosomes mediated by concurrent fusion and kissing events in living cells. Curr Biol, 15, 360-5. BROWN, W. J., GOODHOUSE, J. & FARQUHAR, M. G. 1986. Mannose-6-phosphate receptors for lysosomal enzymes cycle between the Golgi complex and endosomes. J Cell Biol, 103, 1235-47. A.1 TBC1D23 A-3 BUCCI, C., PARTON, R. G., MATHER, I. H., STUNNENBERG, H., SIMONS, K., HOFLACK, B. & ZERIAL, M. 1992. The small GTPase rab5 functions as a regulatory factor in the early endocytic pathway. Cell, 70, 715-728. CAIL, R. C., SHIRAZINEJAD, C. R. & DRUBIN, D. G. 2022. Induced nanoscale membrane curvature bypasses the essential endocytic function of clathrin. Journal of Cell Biology, 221. CARLTON, J. G. & BAUM, B. 2023. Roles of ESCRT-III polymers in cell division across the tree of life. Current Opinion in Cell Biology, 85, 102274. CARO, L. G. & PALADE, G. E. 1964. PROTEIN SYNTHESIS, STORAGE, AND DISCHARGE IN THE PANCREATIC EXOCRINE CELL. AN AUTORADIOGRAPHIC STUDY. J Cell Biol, 20, 473-95. CATTIN-ORTOLÁ, J., KAUFMAN, J., G, G.,, GILLINGHAM, A., K.,, WAGSTAFF, J., L.,, PEAK-CHEW, S., STEVENS, T., J.,, BOULANGER, J., OWEN, D., J., & MUNRO, S. 2024. Cargo selective vesicle tethering: The structural basis for binding of specific cargo proteins by the Golgi tether component TBC1D23. Science Advances. CHEN, K. E., HEALY, M. D. & COLLINS, B. M. 2019. Towards a molecular understanding of endosomal trafficking by Retromer and Retriever. Traffic, 20, 465-478. CHEN, W. J., GOLDSTEIN, J. L. & BROWN, M. S. 1990. NPXY, a sequence often found in cytoplasmic tails, is required for coated pit-mediated internalization of the low density lipoprotein receptor. J Biol Chem, 265, 3116-23. CHERFILS, J. 2014. Arf GTPases and their effectors: assembling multivalent membrane- binding platforms. Current Opinion in Structural Biology, 29, 67-76. CHO, N. H., CHEVERALLS, K. C., BRUNNER, A. D., KIM, K., MICHAELIS, A. C., RAGHAVAN, P., KOBAYASHI, H., SAVY, L., LI, J. Y., CANAJ, H., KIM, J. Y. S., STEWART, E. M., GNANN, C., MCCARTHY, F., CABRERA, J. P., BRUNETTI, R. M., CHHUN, B. B., DINGLE, G., HEIN, M. Y., HUANG, B., MEHTA, S. B., WEISSMAN, J. S., GOMEZ-SJOBERG, R., ITZHAK, D. N., ROYER, L. A., MANN, M. & LEONETTI, M. D. 2022. OpenCell: Endogenous tagging for the cartography of human cellular organization. Science, 375, eabi6983. CHOTHIA, C. & LESK, A. M. 1986. The relation between the divergence of sequence and structure in proteins. The EMBO Journal, 5, 823-826-826. CHRISTOFORIDIS, S., MIACZYNSKA, M., ASHMAN, K., WILM, M., ZHAO, L., YIP, S. C., WATERFIELD, M. D., BACKER, J. M. & ZERIAL, M. 1999. Phosphatidylinositol-3-OH kinases are Rab5 effectors. Nat Cell Biol, 1, 249-52. CLAIRE, E. C., RYAN, K., STEPHANIE, A. B., YINGKE, L., GAUTIER, M. C., JUSTIN, M. D. T., CASSANDRA, J. W., RAYAN, S., JI-YOUNG, Y., LU-YANG, W. & JOHN, L. R. 2024. High resolution cryo-EM of V-ATPase in native synaptic vesicles. bioRxiv, 2024.04.01.587493. COLLINS, B. M., MCCOY, A. J., KENT, H. M., EVANS, P. R. & OWEN, D. J. 2002. Molecular architecture and functional model of the endocytic AP2 complex. Cell, 109, 523-35. COPE, N., HAROLD, D., HILL, G., MOSKVINA, V., STEVENSON, J., HOLMANS, P., OWEN, M. J., O'DONOVAN, M. C. & WILLIAMS, J. 2005. Strong evidence that KIAA0319 on chromosome 6p is a susceptibility gene for developmental dyslexia. Am J Hum Genet, 76, 581-91. A-4 Crystallographic statistics and screens COWLES, C. R., ODORIZZI, G., PAYNE, G. S. & EMR, S. D. 1997a. The AP-3 adaptor complex is essential for cargo-selective transport to the yeast vacuole. Cell, 91, 109- 18. COWLES, C. R., ODORIZZI, G., PAYNE, G. S. & EMR, S. D. 1997b. The AP-3 Adaptor Complex Is Essential for Cargo-Selective Transport to the Yeast Vacuole. Cell, 91, 109-118. CRAWFORD, N. G., KELLY, D. E., HANSEN, M. E. B., BELTRAME, M. H., FAN, S., BOWMAN, S. L., JEWETT, E., RANCIARO, A., THOMPSON, S., LO, Y., PFEIFER, S. P., JENSEN, J. D., CAMPBELL, M. C., BEGGS, W., HORMOZDIARI, F., MPOLOKA, S. W., MOKONE, G. G., NYAMBO, T., MESKEL, D. W., BELAY, G., HAUT, J., ROTHSCHILD, H., ZON, L., ZHOU, Y., KOVACS, M. A., XU, M., ZHANG, T., BISHOP, K., SINCLAIR, J., RIVAS, C., ELLIOT, E., CHOI, J., LI, S. A., HICKS, B., BURGESS, S., ABNET, C., WATKINS-CHOW, D. E., OCEANA, E., SONG, Y. S., ESKIN, E., BROWN, K. M., MARKS, M. S., LOFTUS, S. K., PAVAN, W. J., YEAGER, M., CHANOCK, S. & TISHKOFF, S. A. 2017. Loci associated with skin pigmentation identified in African populations. Science, 358. CUI, Y., YANG, Z. & TEASDALE, R. D. 2018. The functional roles of retromer in Parkinson's disease. FEBS Lett, 592, 1096-1112. DAY, K. J., KAGO, G., WANG, L., RICHTER, J. B., HAYDEN, C. C., LAFER, E. M. & STACHOWIAK, J. C. 2021. Liquid-like protein interactions catalyse assembly of endocytic vesicles. Nat Cell Biol, 23, 366-376. DEHNEN, L., JANZ, M., VERMA, J. K., PSATHAKI, O. E., LANGEMEYER, L., FRÖHLICH, F., HEINISCH, J. J., MEYER, H., UNGERMANN, C. & PAULULAT, A. 2020. A trimeric metazoan Rab7 GEF complex is crucial for endocytosis and scavenger function. J Cell Sci, 133. DELEVOYE, C., HEILIGENSTEIN, X., RIPOLL, L., GILLES-MARSENS, F., DENNIS, M. K., LINARES, R. A., DERMAN, L., GOKHALE, A., MOREL, E., FAUNDEZ, V., MARKS, M. S. & RAPOSO, G. 2016. BLOC-1 Brings Together the Actin and Microtubule Cytoskeletons to Generate Recycling Endosomes. Curr Biol, 26, 1-13. DELL'ANGELICA, E. C. & BONIFACINO, J. S. 2019. Coatopathies: Genetic Disorders of Protein Coats. Annu Rev Cell Dev Biol, 35, 131-168. DELL'ANGELICA, E. C., KLUMPERMAN, J., STOORVOGEL, W. & BONIFACINO, J. S. 1998. Association of the AP-3 Adaptor Complex with Clathrin. Science, 280, 431- 434. DENISOV, I. G. & SLIGAR, S. G. 2016. Nanodiscs for structural and functional studies of membrane proteins. Nature Structural & Molecular Biology, 23, 481-486. DERIVERY, E., SOUSA, C., GAUTIER, J. J., LOMBARD, B., LOEW, D. & GAUTREAU, A. 2009. The Arp2/3 activator WASH controls the fission of endosomes through a large multiprotein complex. Dev Cell, 17, 712-23. DI PIETRO, S. M., FALCÓN-PÉREZ, J. M., TENZA, D., SETTY, S. R., MARKS, M. S., RAPOSO, G. & DELL'ANGELICA, E. C. 2006. BLOC-1 interacts with BLOC-2 and the AP-3 complex to facilitate protein trafficking on endosomes. Mol Biol Cell, 17, 4027-38. DICKSON, E. J. & HILLE, B. 2019. Understanding phosphoinositides: rare, dynamic, and essential membrane phospholipids. Biochem J, 476, 1-23. DODONOVA, S. O., DIESTELKOETTER-BACHERT, P., VON APPEN, A., HAGEN, W. J. H., BECK, R., BECK, M., WIELAND, F. & BRIGGS, J. A. G. 2015. A structure of A.1 TBC1D23 A-5 the COPI coat and the role of coat proteins in membrane vesicle assembly. Science, 349, 195-198. DOHERTY, G. J. & MCMAHON, H. T. 2009. Mechanisms of endocytosis. Annu Rev Biochem, 78, 857-902. DONALDSON, J. G. & JACKSON, C. L. 2011. ARF family G proteins and their regulators: roles in membrane transport, development and disease. Nat Rev Mol Cell Biol, 12, 362-75. DONG, X. P., SHEN, D., WANG, X., DAWSON, T., LI, X., ZHANG, Q., CHENG, X., ZHANG, Y., WEISMAN, L. S., DELLING, M. & XU, H. 2010. PI(3,5)P(2) controls membrane trafficking by direct activation of mucolipin Ca(2+) release channels in the endolysosome. Nat Commun, 1, 38. DORAY, B., KNISELY, J. M., WARTMAN, L., BU, G. & KORNFELD, S. 2008. Identification of Acidic Dileucine Signals in LRP9 that Interact with Both GGAs and AP-1/AP-2. Traffic, 9, 1551-1562. DORAY, B., LEE, I., KNISELY, J., BU, G. & KORNFELD, S. 2007. The γ/σ1 and α/σ2 Hemicomplexes of Clathrin Adaptors AP-1 and AP-2 Harbor the Dileucine Recognition Site. Molecular Biology of the Cell, 18, 1887-1896. DRAKE, M. T., ZHU, Y. & KORNFELD, S. 2000. The assembly of AP-3 adaptor complex- containing clathrin-coated vesicles on synthetic liposomes. Mol Biol Cell, 11, 3723- 36. DUTTA, D. & DONALDSON, J. G. 2015. Rab and Arf G proteins in endosomal trafficking. Methods Cell Biol, 130, 127-38. EDELING, M. A., MISHRA, S. K., KEYEL, P. A., STEINHAUSER, A. L., COLLINS, B. M., ROTH, R., HEUSER, J. E., OWEN, D. J. & TRAUB, L. M. 2006. Molecular switches involving the AP-2 beta2 appendage regulate endocytic cargo selection and clathrin coat assembly. Dev Cell, 10, 329-42. EL ALAOUI, F., CASUSO, I., SANCHEZ-FUENTES, D., ARPIN-ANDRE, C., RATHAR, R., BAECKER, V., CASTRO, A., LORCA, T., VIAUD, J., VASSILOPOULOS, S., CARRETERO-GENEVRIER, A. & PICAS, L. 2022. Structural organization and dynamics of FCHo2 docking on membranes. eLife, 11, e73156. EVRARD, C. & ROUGET, P. 2005. Subcellular localization of neural-specific NPDC-1 protein. J Neurosci Res, 79, 747-55. FALCÓN-PÉREZ, J. M., STARCEVIC, M., GAUTAM, R. & DELL'ANGELICA, E. C. 2002. BLOC-1, a novel complex containing the pallidin and muted proteins involved in the biogenesis of melanosomes and platelet-dense granules. J Biol Chem, 277, 28191-9. FALZONE, M. E. & MACKINNON, R. 2023. Gβγ activates PIP2 hydrolysis by recruiting and orienting PLCβ on the membrane surface. Proc Natl Acad Sci U S A, 120, e2301121120. FARQUHAR, M. G. & PALADE, G. E. 1981. The Golgi apparatus (complex)-(1954-1981)- from artifact to center stage. J Cell Biol, 91, 77s-103s. FEINSTEIN, M., FLUSSER, H., LERMAN-SAGIE, T., BEN-ZEEV, B., LEV, D., AGAMY, O., COHEN, I., KADIR, R., SIVAN, S., LESHINSKY-SILVER, E., MARKUS, B. & BIRK, O. S. 2014. VPS53 mutations cause progressive cerebello-cerebral atrophy type 2 (PCCA2). J Med Genet, 51, 303-8. FRAZIER, M. N., DAVIES, A. K., VOEHLER, M., KENDALL, A. K., BORNER, G. H., CHAZIN, W. J., ROBINSON, M. S. & JACKSON, L. P. 2016. Molecular Basis for A-6 Crystallographic statistics and screens the Interaction Between AP4 β4 and its Accessory Protein, Tepsin. Traffic, 17, 400- 15. FREEMAN, C. L., HESKETH, G. & SEAMAN, M. N. 2014. RME-8 coordinates the activity of the WASH complex with the function of the retromer SNX dimer to control endosomal tubulation. J Cell Sci, 127, 2053-70. GALL, W. E., HIGGINBOTHAM, M. A., CHEN, C.-Y., INGRAM, M. F., CYR, D. M. & GRAHAM, T. R. 2000. The auxilin-like phosphoprotein Swa2p is required for clathrin function in yeast. Current Biology, 10, 1349-1358. GAVRILJUK, K., ITZEN, A., GOODY, R. S., GERWERT, K. & KÖTTING, C. 2013. Membrane extraction of Rab proteins by GDP dissociation inhibitor characterized using attenuated total reflection infrared spectroscopy. Proceedings of the National Academy of Sciences, 110, 13380-13385. GEMMER, M., CHAILLET, M. L., VAN LOENHOUT, J., CUEVAS ARENAS, R., VISMPAS, D., GRÖLLERS-MULDERIJ, M., KOH, F. A., ALBANESE, P., SCHELTEMA, R. A., HOWES, S. C., KOTECHA, A., FEDRY, J. & FÖRSTER, F. 2023. Visualization of translation and protein biogenesis at the ER membrane. Nature, 614, 160-167. GERSHLICK, D. C., ISHIDA, M., JONES, J. R., BELLOMO, A., BONIFACINO, J. S. & EVERMAN, D. B. 2019. A neurodevelopmental disorder caused by mutations in the VPS51 subunit of the GARP and EARP complexes. Hum Mol Genet, 28, 1548-1560. GILLINGHAM, A. K., BERTRAM, J., BEGUM, F. & MUNRO, S. 2019. In vivo identification of GTPase interactors by mitochondrial relocalization and proximity biotinylation. Elife, 8. GILLINGHAM, A. K. & MUNRO, S. 2003. Long coiled-coil proteins and membrane traffic. Biochim Biophys Acta, 1641, 71-85. GRUENBERG, J. & VAN DER GOOT, F. G. 2006. Mechanisms of pathogen entry through the endosomal compartments. Nature Reviews Molecular Cell Biology, 7, 495-504. GULBRANSON, D. R., CRISMAN, L., LEE, M., OUYANG, Y., MENASCHE, B. L., DEMMITT, B. A., WAN, C., NOMURA, T., YE, Y., YU, H. & SHEN, J. 2019. AAGAB Controls AP2 Adaptor Assembly in Clathrin-Mediated Endocytosis. Dev Cell, 50, 436-446.e5. GUO, Q., CHEN, K.-E., GIMENEZ-ANDRES, M., JELLETT, A. P., GAO, Y., SIMONETTI, B., LIU, M., DANSON, C. M., HEESOM, K. J., CULLEN, P. J. & COLLINS, B. M. 2024. Structural basis for coupling of the WASH subunit FAM21 with the endosomal SNX27-Retromer complex. bioRxiv, 2023.08.15.553351. HARBOUR ME, BREUSEGEM SY, ANTROBUS R, FREEMAN C, REID E & SEAMAN MN. (2010) The cargo-selective retromer complex is a recruiting hub for protein complexes that regulate endosomal tubule dynamics. J Cell Sci 123, 3703–17. HAWRYLUK, M. J., KEYEL, P. A., MISHRA, S. K., WATKINS, S. C., HEUSER, J. E. & TRAUB, L. M. 2006. Epsin 1 is a Polyubiquitin-Selective Clathrin-Associated Sorting Protein. Traffic, 7, 262-281. HEILKER, R., MANNING‐KRIEG, U., ZUBER, J. F. & SPIESS, M. 1996. In vitro binding of clathrin adaptors to sorting signals correlates with endocytosis and basolateral sorting. The EMBO Journal, 15, 2893-2899-2899. HELDWEIN, E. E., MACIA, E., WANG, J., YIN, H. L., KIRCHHAUSEN, T. & HARRISON, S. C. 2004. Crystal structure of the clathrin adaptor protein 1 core. Proc Natl Acad Sci U S A, 101, 14108-13. A.1 TBC1D23 A-7 HELFER, E., HARBOUR, M. E., HENRIOT, V., LAKISIC, G., SOUSA-BLIN, C., VOLCEANOV, L., SEAMAN, M. N. J. & GAUTREAU, A. 2013. Endosomal recruitment of the WASH complex: active sequences and mutations impairing interaction with the retromer. Biol Cell, 105, 191-207. HENNE, W. M., BUCHKOVICH, N. J. & EMR, S. D. 2011. The ESCRT pathway. Dev Cell, 21, 77-91. HERRMANN, E., SCHÄFER, J. H., WILMES, S., UNGERMANN, C., MOELLER, A. & KÜMMEL, D. 2023. Structure of the metazoan Rab7 GEF complex Mon1-Ccz1- Bulli. Proc Natl Acad Sci U S A, 120, e2301908120. HEUSER, J. E. & ANDERSON, R. G. 1989. Hypertonic media inhibit receptor-mediated endocytosis by blocking clathrin-coated pit formation. J Cell Biol, 108, 389-400. HEUSER, J. E. & KEEN, J. 1988. Deep-etch visualization of proteins involved in clathrin assembly. J Cell Biol, 107, 877-86. HINSHAW, J. E. & SCHMID, S. L. 1995. Dynamin self-assembles into rings suggesting a mechanism for coated vesicle budding. Nature, 374, 190-2. HIRST, J., BORNER, G. H., HARBOUR, M. & ROBINSON, M. S. 2005. The aftiphilin/p200/gamma-synergin complex. Mol Biol Cell, 16, 2554-65. HIRST, J., HESKETH, G. G., GINGRAS, A.-C. & ROBINSON, M. S. 2021. Rag GTPases and phosphatidylinositol 3-phosphate mediate recruitment of the AP-5/SPG11/SPG15 complex. Journal of Cell Biology, 220. HIRST, J., MOTLEY, A., HARASAKI, K., PEAK CHEW, S. Y. & ROBINSON, M. S. 2003. EpsinR: an ENTH domain-containing protein that interacts with AP-1. Mol Biol Cell, 14, 625-41. HIRST, J., SCHLACHT, A., NORCOTT, J. P., TRAYNOR, D., BLOOMFIELD, G., ANTROBUS, R., KAY, R. R., DACKS, J. B. & ROBINSON, M. S. 2014. Characterization of TSET, an ancient and widespread membrane trafficking complex. Elife, 3, e02866. HOGEBOOM, G. H., CLAUDE, A. & HOTCH-KISS, R. D. 1946. The distribution of cytochrome oxidase and succinoxidase in the cytoplasm of the mammalian liver cell. J Biol Chem, 165, 615-29. HÖNING, S., RICOTTA, D., KRAUSS, M., SPÄTE, K., SPOLAORE, B., MOTLEY, A., ROBINSON, M., ROBINSON, C., HAUCKE, V. & OWEN, D. J. 2005. Phosphatidylinositol-(4,5)-bisphosphate regulates sorting signal recognition by the clathrin-associated adaptor complex AP2. Mol Cell, 18, 519-31. HOOY, R. M., IWAMOTO, Y., TUDORICA, D. A., REN, X. & HURLEY, J. H. 2022. Self- assembly and structure of a clathrin-independent AP-1:Arf1 tubular membrane coat. Sci Adv, 8, eadd3914. HUANG, W., LIU, Z., YANG, F., ZHOU, H., YONG, X., YANG, X., ZHOU, Y., XUE, L., ZHANG, Y., LIU, D., MENG, W., ZHANG, W., ZHANG, X., SHEN, X., SUN, Q., LI, L., MA, C., WEI, Y., BILLADEAU, D. D., MO, X. & JIA, D. 2019. Structural and functional studies of TBC1D23 C-terminal domain provide a link between endosomal trafficking and PCH. Proc Natl Acad Sci U S A, 116, 22598-22608. HUTAGALUNG, A. H. & NOVICK, P. J. 2011. Role of Rab GTPases in membrane traffic and cell physiology. Physiol Rev, 91, 119-49. HUTTLIN, E. L., TING, L., BRUCKNER, R. J., GEBREAB, F., GYGI, M. P., SZPYT, J., TAM, S., ZARRAGA, G., COLBY, G., BALTIER, K., DONG, R., GUARANI, V., VAITES, L. P., ORDUREAU, A., RAD, R., ERICKSON, B. K., WUHR, M., CHICK, J., ZHAI, B., KOLIPPAKKAM, D., MINTSERIS, J., OBAR, R. A., A-8 Crystallographic statistics and screens HARRIS, T., ARTAVANIS-TSAKONAS, S., SOWA, M. E., DE CAMILLI, P., PAULO, J. A., HARPER, J. W. & GYGI, S. P. 2015. The BioPlex Network: A Systematic Exploration of the Human Interactome. Cell, 162, 425-440. HUTTNER, W. B., GERDES, H. H. & ROSA, P. 1991. The granin (chromogranin/secretogranin) family. Trends Biochem Sci, 16, 27-30. ITZHAK, D. N., TYANOVA, S., COX, J. & BORNER, G. H. H. 2016. Global, quantitative and dynamic mapping of protein subcellular localization. eLife, 5, e16950. IVANOVA, E. L., MAU-THEM, F. T., RIAZUDDIN, S., KAHRIZI, K., LAUGEL, V., SCHAEFER, E., DE SAINT MARTIN, A., RUNGE, K., IQBAL, Z., SPITZ, M. A., LAURA, M., DROUOT, N., GERARD, B., DELEUZE, J. F., DE BROUWER, A. P. M., RAZZAQ, A., DOLLFUS, H., ASSIR, M. Z., NITCHKE, P., HINCKELMANN, M. V., ROPERS, H., RIAZUDDIN, S., NAJMABADI, H., VAN BOKHOVEN, H. & CHELLY, J. 2017. Homozygous Truncating Variants in TBC1D23 Cause Pontocerebellar Hypoplasia and Alter Cortical Development. Am J Hum Genet, 101, 428-440. JACKSON, L. P., KELLY, B. T., MCCOY, A. J., GAFFRY, T., JAMES, L. C., COLLINS, B. M., HÖNING, S., EVANS, P. R. & OWEN, D. J. 2010. A Large-Scale Conformational Change Couples Membrane Recruitment to Cargo Binding in the AP2 Clathrin Adaptor Complex. Cell, 141, 1220-1229. JAMES P, HALLADAY J & CRAIG EA (1996) Genomic libraries and a host strain designed for highly efficient two-hybrid selection in yeast. Genetics 144, 1425–36. JANI, R. A., DI CICCO, A., KEREN-KAPLAN, T., VALE-COSTA, S., HAMAOUI, D., HURBAIN, I., TSAI, F. C., DI MARCO, M., MACÉ, A. S., ZHU, Y., AMORIM, M. J., BASSEREAU, P., BONIFACINO, J. S., SUBTIL, A., MARKS, M. S., LÉVY, D., RAPOSO, G. & DELEVOYE, C. 2022. PI4P and BLOC-1 remodel endosomal membranes into tubules. J Cell Biol, 221. JANSEN, R. M. & HURLEY, J. H. 2023. Longin domain GAP complexes in nutrient signalling, membrane traffic and neurodegeneration. FEBS Lett, 597, 750-761. JHA, A., AGOSTINELLI, N. R., MISHRA, S. K., KEYEL, P. A., HAWRYLUK, M. J. & TRAUB, L. M. 2004. A novel AP-2 adaptor interaction motif initially identified in the long-splice isoform of synaptojanin 1, SJ170. J Biol Chem, 279, 2281-90. JIA, D., ZHANG, J.-S., LI, F., WANG, J., DENG, Z., WHITE, M. A., OSBORNE, D. G., PHILLIPS-KRAWCZAK, C., GOMEZ, T. S., LI, H., SINGLA, A., BURSTEIN, E., BILLADEAU, D. D. & ROSEN, M. K. 2016. Structural and mechanistic insights into regulation of the retromer coat by TBC1d5. Nature Communications, 7, 13305. JIA, X., WEBER, E., TOKAREV, A., LEWINSKI, M., RIZK, M., SUAREZ, M., GUATELLI, J. & XIONG, Y. 2014. Structural basis of HIV-1 Vpu-mediated BST2 antagonism via hijacking of the clathrin adaptor protein complex 1. Elife, 3, e02362. JOHNSON, D. E., OSTROWSKI, P., JAUMOUILLÉ, V. & GRINSTEIN, S. 2016. The position of lysosomes within the cell determines their luminal pH. J Cell Biol, 212, 677-92. JUMPER, J., EVANS, R., PRITZEL, A., GREEN, T., FIGURNOV, M., RONNEBERGER, O., TUNYASUVUNAKOOL, K., BATES, R., ŽÍDEK, A., POTAPENKO, A., BRIDGLAND, A., MEYER, C., KOHL, S. A. A., BALLARD, A. J., COWIE, A., ROMERA-PAREDES, B., NIKOLOV, S., JAIN, R., ADLER, J., BACK, T., PETERSEN, S., REIMAN, D., CLANCY, E., ZIELINSKI, M., STEINEGGER, M., PACHOLSKA, M., BERGHAMMER, T., BODENSTEIN, S., SILVER, D., VINYALS, O., SENIOR, A. W., KAVUKCUOGLU, K., KOHLI, P. & HASSABIS, A.1 TBC1D23 A-9 D. 2021. Highly accurate protein structure prediction with AlphaFold. Nature, 596, 583-589. KAKSONEN, M. & ROUX, A. 2018. Mechanisms of clathrin-mediated endocytosis. Nature Reviews Molecular Cell Biology, 19, 313-326. 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. Journal of Cell Biology, 42, 202-220. KELLY, B. T., GRAHAM, S. C., LISKA, N., DANNHAUSER, P. N., HÖNING, S., UNGEWICKELL, E. J. & OWEN, D. J. 2014. Clathrin adaptors. AP2 controls clathrin polymerization with a membrane-activated switch. Science, 345, 459-63. KELLY, B. T., MCCOY, A. J., SPÄTE, K., MILLER, S. E., EVANS, P. R., HÖNING, S. & OWEN, D. J. 2008. A structural explanation for the binding of endocytic dileucine motifs by the AP2 complex. Nature, 456, 976-979. KENT, H. M., EVANS, P. R., SCHÄFER, I. B., GRAY, S. R., SANDERSON, C. M., LUZIO, J. P., PEDEN, A. A. & OWEN, D. J. 2012. Structural basis of the intracellular sorting of the SNARE VAMP7 by the AP3 adaptor complex. Dev Cell, 22, 979-88. KIM, H. G., AHN, J. W., KURTH, I., ULLMANN, R., KIM, H. T., KULHARYA, A., HA, K. S., ITOKAWA, Y., MELICIANI, I., WENZEL, W., LEE, D., ROSENBERGER, G., OZATA, M., BICK, D. P., SHERINS, R. J., NAGASE, T., TEKIN, M., KIM, S. H., KIM, C. H., ROPERS, H. H., GUSELLA, J. F., KALSCHEUER, V., CHOI, C. Y. & LAYMAN, L. C. 2010. WDR11, a WD protein that interacts with transcription factor EMX1, is mutated in idiopathic hypogonadotropic hypogonadism and Kallmann syndrome. Am J Hum Genet, 87, 465-79. KIM, Y. J., OSBORN, D. P., LEE, J. Y., ARAKI, M., ARAKI, K., MOHUN, T., KANSAKOSKI, J., BRANDSTACK, N., KIM, H. T., MIRALLES, F., KIM, C. H., BROWN, N. A., KIM, H. G., MARTINEZ-BARBERA, J. P., ATALIOTIS, P., RAIVIO, T., LAYMAN, L. C. & KIM, S. H. 2018. WDR11-mediated Hedgehog signalling defects underlie a new ciliopathy related to Kallmann syndrome. EMBO Rep, 19, 269-289. KIRCHHAUSEN, T. 1993. Coated pits and coated vesicles — sorting it all out. Current Opinion in Structural Biology, 3, 182-188. KLUMPERMAN, J. & RAPOSO, G. 2014. The complex ultrastructure of the endolysosomal system. Cold Spring Harb Perspect Biol, 6, a016857. KOVTUN, O., DICKSON, V. K., KELLY, B. T., OWEN, D. J. & BRIGGS, J. A. G. 2020. Architecture of the AP2/clathrin coat on the membranes of clathrin-coated vesicles. Science Advances, 6, eaba8381. KOVTUN, O., LENEVA, N., BYKOV, Y. S., ARIOTTI, N., TEASDALE, R. D., SCHAFFER, M., ENGEL, B. D., OWEN, D. J., BRIGGS, J. A. G. & COLLINS, B. M. 2018. Structure of the membrane-assembled retromer coat determined by cryo- electron tomography. Nature, 561, 561-564. KOZIK, P., FRANCIS, R. W., SEAMAN, M. N. & ROBINSON, M. S. 2010. A screen for endocytic motifs. Traffic, 11, 843-55. KRAUSS, M., KINUTA, M., WENK, M. R., DE CAMILLI, P., TAKEI, K. & HAUCKE, V. 2003. ARF6 stimulates clathrin/AP-2 recruitment to synaptic membranes by activating phosphatidylinositol phosphate kinase type Igamma. J Cell Biol, 162, 113- 24. A-10 Crystallographic statistics and screens KURAL, C., TACHEVA-GRIGOROVA, S. K., BOULANT, S., COCUCCI, E., BAUST, T., DUARTE, D. & KIRCHHAUSEN, T. 2012. Dynamics of intracellular clathrin/AP1- and clathrin/AP3-containing carriers. Cell Rep, 2, 1111-9. KWON, Y., KAAKE, R. M., ECHEVERRIA, I., SUAREZ, M., KARIMIAN SHAMSABADI, M., STONEHAM, C., RAMIREZ, P. W., KRESS, J., SINGH, R., SALI, A., KROGAN, N., GUATELLI, J. & JIA, X. 2020. Structural basis of CD4 downregulation by HIV-1 Nef. Nat Struct Mol Biol, 27, 822-828. LAUGWITZ, L., BUCHERT, R., GROESCHEL, S., RIESS, A., GRIMMEL, M., BECK- WODL, S., STURM, M., GOHLA, G., DOBLER-NEUMANN, M., KRAGELOH- MANN, I. & HAACK, T. B. 2020. Pontocerebellar hypoplasia type 11: Does the genetic defect determine timing of cerebellar pathology? Eur J Med Genet, 63, 103938. LE BORGNE, R., ALCONADA, A., BAUER, U. & HOFLACK, B. 1998. The mammalian AP-3 adaptor-like complex mediates the intracellular transport of lysosomal membrane glycoproteins. J Biol Chem, 273, 29451-61. LEE, H. H., NEMECEK, D., SCHINDLER, C., SMITH, W. J., GHIRLANDO, R., STEVEN, A. C., BONIFACINO, J. S. & HURLEY, J. H. 2012. Assembly and architecture of biogenesis of lysosome-related organelles complex-1 (BLOC-1). J Biol Chem, 287, 5882-90. LEES, J. A., LI, P., KUMAR, N., WEISMAN, L. S. & REINISCH, K. M. 2020. Insights into Lysosomal PI(3,5)P(2) Homeostasis from a Structural-Biochemical Analysis of the PIKfyve Lipid Kinase Complex. Mol Cell, 80, 736-743.e4. LEFRANÇOIS, S., JANVIER, K., BOEHM, M., OOI, C. E. & BONIFACINO, J. S. 2004. An ear-core interaction regulates the recruitment of the AP-3 complex to membranes. Dev Cell, 7, 619-25. LERAY, X., CONTI, R., LI, Y., DEBACKER, C., CASTELLI, F., FENAILLE, F., ZDEBIK, A. A., PUSCH, M. & GASNIER, B. 2021. Arginine-selective modulation of the lysosomal transporter PQLC2 through a gate-tuning mechanism. Proc Natl Acad Sci U S A, 118. LETOURNEUR, F. & KLAUSNER, R. D. 1992. A novel di-leucine motif and a tyrosine- based motif independently mediate lysosomal targeting and endocytosis of CD3 chains. Cell, 69, 1143-1157. LI, B., ZHU, D., SHI, H. & ZHANG, X. 2021. Effect of charge on protein preferred orientation at the air–water interface in cryo-electron microscopy. Journal of Structural Biology, 213, 107783. LIN-MOORE, A. T., OYEYEMI, M. J. & HAMMARLUND, M. 2021. rab-27 acts in an intestinal pathway to inhibit axon regeneration in C. elegans. PLoS Genet, 17, e1009877. LIU, D., YANG, F., LIU, Z., WANG, J., HUANG, W., MENG, W., BILLADEAU, D. D., SUN, Q., MO, X. & JIA, D. 2020. Structure of TBC1D23 N-terminus reveals a novel role for rhodanese domain. PLoS Biol, 18, e3000746. LIU, Y., XU, P., RIVARA, S., LIU, C., RICCI, J., REN, X., HURLEY, J. H. & ABLASSER, A. 2022. Clathrin-associated AP-1 controls termination of STING signalling. Nature, 610, 761-767. LOHSE, M. J., BENOVIC, J. L., CODINA, J., CARON, M. G. & LEFKOWITZ, R. J. 1990. beta-Arrestin: a protein that regulates beta-adrenergic receptor function. Science, 248, 1547-50. A.1 TBC1D23 A-11 LOLICATO, F., SALEPPICO, R., GRIFFO, A., MEYER, A., SCOLLO, F., POKRANDT, B., MÜLLER, H. M., EWERS, H., HÄHL, H., FLEURY, J. B., SEEMANN, R., HOF, M., BRÜGGER, B., JACOBS, K., VATTULAINEN, I. & NICKEL, W. 2022. Cholesterol promotes clustering of PI(4,5)P2 driving unconventional secretion of FGF2. J Cell Biol, 221. LOPEZ-ROBLES, C., SCARAMUZZA, S., ASTORGA-SIMON, E. N., ISHIDA, M., WILLIAMSON, C. D., BAÑOS-MATEOS, S., GIL-CARTON, D., ROMERO- DURANA, M., VIDAURRAZAGA, A., FERNANDEZ-RECIO, J., ROJAS, A. L., BONIFACINO, J. S., CASTAÑO-DÍEZ, D. & HIERRO, A. 2023. Architecture of the ESCPE-1 membrane coat. Nat Struct Mol Biol, 30, 958-969. LUNDMARK, R. & CARLSSON, S. R. 2004. Regulated membrane recruitment of dynamin- 2 mediated by sorting nexin 9. J Biol Chem, 279, 42694-702. LUZIO, J. P., BRIGHT, N. A. & PRYOR, P. R. 2007. The role of calcium and other ions in sorting and delivery in the late endocytic pathway. Biochem Soc Trans, 35, 1088-91. LUZIO, J. P., PARKINSON, M. D., GRAY, S. R. & BRIGHT, N. A. 2009. The delivery of endocytosed cargo to lysosomes. Biochem Soc Trans, 37, 1019-21. MA, L., UMASANKAR, P. K., WROBEL, A. G., LYMAR, A., MCCOY, A. J., HOLKAR, S. S., JHA, A., PRADHAN-SUNDD, T., WATKINS, S. C., OWEN, D. J. & TRAUB, L. M. 2016. Transient Fcho1/2⋅Eps15/R⋅AP-2 Nanoclusters Prime the AP-2 Clathrin Adaptor for Cargo Binding. Dev Cell, 37, 428-43. MA, W. & GOLDBERG, J. 2013. Rules for the recognition of dilysine retrieval motifs by coatomer. Embo j, 32, 926-37. MADAN, V., ALBACETE-ALBACETE, L., JIN, L., SCATURRO, P., WATSON, J. L., MUSCHALIK, N., BEGUM, F., BOULANGER, J., BAUER, K., KIEBLER, M. A., DERIVERY, E. & BULLOCK, S. L. 2023. HEATR5B associates with dynein- dynactin and promotes motility of AP1-bound endosomal membranes. Embo j, 42, e114473. MARDONES, G. A., BURGOS, P. V., LIN, Y., KLOER, D. P., MAGADÁN, J. G., HURLEY, J. H. & BONIFACINO, J. S. 2013. Structural basis for the recognition of tyrosine-based sorting signals by the μ3A subunit of the AP-3 adaptor complex. J Biol Chem, 288, 9563-71. MARKS, M. S., WOODRUFF, L., OHNO, H. & BONIFACINO, J. S. 1996. Protein targeting by tyrosine- and di-leucine-based signals: evidence for distinct saturable components. Journal of Cell Biology, 135, 341-354. MASTRONARDE, D. N. & HELD, S. R. 2017. Automated tilt series alignment and tomographic reconstruction in IMOD. J Struct Biol, 197, 102-113. MATSUO, H., CHEVALLIER, J., MAYRAN, N., LE BLANC, I., FERGUSON, C., FAURÉ, J., BLANC, N. S., MATILE, S., DUBOCHET, J. & SADOUL, R. 2004. Role of LBPA and Alix in multivesicular liposome formation and endosome organization. Science, 303, 531-534. MAXFIELD, F. R. 2014. Role of endosomes and lysosomes in human disease. Cold Spring Harb Perspect Biol, 6, a016931. MCBRIDE, H. M., RYBIN, V., MURPHY, C., GINER, A., TEASDALE, R. & ZERIAL, M. 1999. Oligomeric complexes link Rab5 effectors with NSF and drive membrane fusion via interactions between EEA1 and syntaxin 13. Cell, 98, 377-86. MCPHERSON, P. S., GARCIA, E. P., SLEPNEV, V. I., DAVID, C., ZHANG, X., GRABS, D., SOSSINI, W. S., BAUERFEIND, R., NEMOTO, Y. & DE CAMILLI, P. 1996. A presynaptic inositol-5-phosphatase. Nature, 379, 353-357. A-12 Crystallographic statistics and screens 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, 570-4. MILLER, S. E., MATHIASEN, S., BRIGHT, N. A., PIERRE, F., KELLY, B. T., KLADT, N., SCHAUSS, A., MERRIFIELD, C. J., STAMOU, D., HÖNING, S. & OWEN, D. J. 2015. CALM regulates clathrin-coated vesicle size and maturation by directly sensing and driving membrane curvature. Dev Cell, 33, 163-75. MILLER, SHARON E., SAHLENDER, DANIELA A., GRAHAM, STEPHEN C., HÖNING, S., ROBINSON, MARGARET S., PEDEN, ANDREW A. & OWEN, DAVID J. 2011. The Molecular Basis for the Endocytosis of Small R-SNAREs by the Clathrin Adaptor CALM. Cell, 147, 1118-1131. MISHRA, S. K., HAWRYLUK, M. J., BRETT, T. J., KEYEL, P. A., DUPIN, A. L., JHA, A., HEUSER, J. E., FREMONT, D. H. & TRAUB, L. M. 2004. Dual engagement regulation of protein interactions with the AP-2 adaptor alpha appendage. J Biol Chem, 279, 46191-203. MIYATAKE, S., NAKASEKO, C., UMEMORI, H., YAMAMOTO, T. & SAITO, T. 1998. Src family tyrosine kinases associate with and phosphorylate CTLA-4 (CD152). Biochem Biophys Res Commun, 249, 444-8. MORRIS, K. L., BUFFALO, C. Z., STÜRZEL, C. M., HEUSINGER, E., KIRCHHOFF, F., REN, X. & HURLEY, J. H. 2018. HIV-1 Nefs Are Cargo-Sensitive AP-1 Trimerization Switches in Tetherin Downregulation. Cell, 174, 659-671.e14. MULLIGAN, R. J., MAGAJ, M. M., DIGILIO, L., REDEMANN, S., YAP, C. C. & WINCKLER, B. 2024. Collapse of late endosomal pH elicits a rapid Rab7 response via V-ATPase and RILP. bioRxiv. MUND, M., TSCHANZ, A., WU, Y.-L., FREY, F., MEHL, J. L., KAKSONEN, M., AVINOAM, O., SCHWARZ, U. S. & RIES, J. 2023. Clathrin coats partially preassemble and subsequently bend during endocytosis. Journal of Cell Biology, 222. MUNRO, S. & PELHAM, H., R,B. 1987. <1-s2.0-0092867487900869-main.pdf>. Cell. MURRAY, D. H., JAHNEL, M., LAUER, J., AVELLANEDA, M. J., BROUILLY, N., CEZANNE, A., MORALES-NAVARRETE, H., PERINI, E. D., FERGUSON, C., LUPAS, A. N., KALAIDZIDIS, Y., PARTON, R. G., GRILL, S. W. & ZERIAL, M. 2016. An endosomal tether undergoes an entropic collapse to bring vesicles together. Nature, 537, 107-111. NAKADA-TSUKUI, K., SAITO-NAKANO, Y., ALI, V. & NOZAKI, T. 2005. A Retromerlike Complex Is a Novel Rab7 Effector That Is Involved in the Transport of the Virulence Factor Cysteine Protease in the Enteric Protozoan Parasite Entamoeba histolytica. Molecular Biology of the Cell, 16, 5294-5303. NAMAVAR, Y., BARTH, P. G., KASHER, P. R., VAN RUISSEN, F., BROCKMANN, K., BERNERT, G., WRITZL, K., VENTURA, K., CHENG, E. Y., FERRIERO, D. M., BASEL-VANAGAITE, L., EGGENS, V. R., KRAGELOH-MANN, I., DE MEIRLEIR, L., KING, M., GRAHAM, J. M., JR., VON MOERS, A., KNOERS, N., SZTRIHA, L., KORINTHENBERG, R., CONSORTIUM, P. C. H., DOBYNS, W. B., BAAS, F. & POLL-THE, B. T. 2011. Clinical, neuroradiological and genetic findings in pontocerebellar hypoplasia. Brain, 134, 143-56. 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. Nat Commun, 9, 596. A.1 TBC1D23 A-13 NAVARRO NEGREDO, P., EDGAR, J. R., WROBEL, A. G., ZACCAI, N. R., ANTROBUS, R., OWEN, D. J. & ROBINSON, M. S. 2017. Contribution of the clathrin adaptor AP-1 subunit mu1 to acidic cluster protein sorting. J Cell Biol, 216, 2927-2943. NAYDENOVA, K. & RUSSO, C. J. 2017. Measuring the effects of particle orientation to improve the efficiency of electron cryomicroscopy. Nature Communications, 8, 629. NIE, Z., FEI, J., PREMONT, R. T. & RANDAZZO, P. A. 2005. The Arf GAPs AGAP1 and AGAP2 distinguish between the adaptor protein complexes AP-1 and AP-3. J Cell Sci, 118, 3555-66. NOVIKOFF, A. B., PODBER, E., RYAN, J. & NOE, E. 1953. Biochemical heterogeneity of the cytoplasmic particles isolated from rat liver homogenate. J Histochem Cytochem, 1, 27-46. OWEN, D. J. & EVANS, P. R. 1998. A Structural Explanation for the Recognition of Tyrosine-Based Endocytotic Signals. Science, 282, 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 alpha-adaptin appendage domain. Cell, 97, 805-15. 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. Embo j, 19, 4216- 27. PAGE, L. J., SOWERBY, P. J., LUI, W. W. & ROBINSON, M. S. 1999. Gamma-synergin: an EH domain-containing protein that interacts with gamma-adaptin. J Cell Biol, 146, 993-1004. PALAMIDESSI, A., FRITTOLI, E., GARRÉ, M., FARETTA, M., MIONE, M., TESTA, I., DIASPRO, A., LANZETTI, L., SCITA, G. & DI FIORE, P. P. 2008. Endocytic Trafficking of Rac Is Required for the Spatial Restriction of Signaling in Cell Migration. Cell, 134, 135-147. PARCHURE, A., TIAN, M., STALDER, D., BOYER, C. K., BEARROWS, S. C., ROHLI, K. E., ZHANG, J., RIVERA-MOLINA, F., RAMAZANOV, B. R., MAHATA, S. K., WANG, Y., STEPHENS, S. B., GERSHLICK, D. C. & VON BLUME, J. 2022. Liquid-liquid phase separation facilitates the biogenesis of secretory storage granules. J Cell Biol, 221. PARK, S., LEVENTAL, I., PASTOR, R. W. & IM, W. 2023. Unsaturated Lipids Facilitate Partitioning of Transmembrane Peptides into the Liquid Ordered Phase. J Chem Theory Comput, 19, 5303-5314. PARRA, K. J. & KANE, P. M. 1998. Reversible association between the V1 and V0 domains of yeast vacuolar H+-ATPase is an unconventional glucose-induced effect. Mol Cell Biol, 18, 7064-74. PEARSE, B. M. & ROBINSON, M. S. 1984. Purification and properties of 100-kd proteins from coated vesicles and their reconstitution with clathrin. Embo j, 3, 1951-7. PECHSTEIN, A., BACETIC, J., VAHEDI-FARIDI, A., GROMOVA, K., SUNDBORGER, A., TOMLIN, N., KRAINER, G., VORONTSOVA, O., SCHÄFER, J. G., OWE, S. G., COUSIN, M. A., SAENGER, W., SHUPLIAKOV, O. & HAUCKE, V. 2010. Regulation of synaptic vesicle recycling by complex formation between intersectin 1 and the clathrin adaptor complex AP2. Proc Natl Acad Sci U S A, 107, 4206-11. 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. J Cell Biol, 164, 1065-76. A-14 Crystallographic statistics and screens PEDEN, A. A., RUDGE, R. E., LUI, W. W. & ROBINSON, M. S. 2002. Assembly and function of AP-3 complexes in cells expressing mutant subunits. J Cell Biol, 156, 327-36. PEUKES, J., XIONG, X., ERLENDSSON, S., QU, K., WAN, W., CALDER, L. J., SCHRAIDT, O., KUMMER, S., FREUND, S. M. V., KRÄUSSLICH, H. G. & BRIGGS, J. A. G. 2020. The native structure of the assembled matrix protein 1 of influenza A virus. Nature, 587, 495-498. PILLAY, S., MEYER, N. L., PUSCHNIK, A. S., DAVULCU, O., DIEP, J., ISHIKAWA, Y., JAE, L. T., WOSEN, J. E., NAGAMINE, C. M., CHAPMAN, M. S. & CARETTE, J. E. 2016. An essential receptor for adeno-associated virus infection. Nature, 530, 108- 12. PODINOVSKAIA, M., PRESCIANOTTO-BASCHONG, C., BUSER, D. P. & SPANG, A. 2021. A novel live-cell imaging assay reveals regulation of endosome maturation. Elife, 10. PORTER, K. R., CLAUDE, A. & FULLAM, E. F. 1945. A STUDY OF TISSUE CULTURE CELLS BY ELECTRON MICROSCOPY : METHODS AND PRELIMINARY OBSERVATIONS. J Exp Med, 81, 233-46. POSOR, Y., JANG, W. & HAUCKE, V. 2022. Phosphoinositides as membrane organizers. Nature Reviews Molecular Cell Biology, 23, 797-816. PRIYA, A., KALAIDZIDIS, I. V., KALAIDZIDIS, Y., LAMBRIGHT, D. & DATTA, S. 2015. Molecular Insights into Rab7-Mediated Endosomal Recruitment of Core Retromer: Deciphering the Role of Vps26 and Vps35. Traffic, 16, 68-84. PRYOR, P. R., JACKSON, L., GRAY, S. R., EDELING, M. A., THOMPSON, A., SANDERSON, C. M., EVANS, P. R., OWEN, D. J. & LUZIO, J. P. 2008. Molecular basis for the sorting of the SNARE VAMP7 into endocytic clathrin-coated vesicles by the ArfGAP Hrb. Cell, 134, 817-27. PUNJANI, A., RUBINSTEIN, J. L., FLEET, D. J. & BRUBAKER, M. A. 2017. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nature Methods, 14, 290-296. QUENTIN, D., SCHUHMACHER, J. S., KLINK, B. U., LAUER, J., SHAIKH, T. R., HUIS IN 'T VELD, P. J., WELP, L. M., URLAUB, H., ZERIAL, M. & RAUNSER, S. 2023. Structural basis of mRNA binding by the human FERRY Rab5 effector complex. Mol Cell, 83, 1856-1871 e9. RAIBORG, C., MALERØD, L., PEDERSEN, N. M. & STENMARK, H. 2008. Differential functions of Hrs and ESCRT proteins in endocytic membrane trafficking. Exp Cell Res, 314, 801-13. RAO, K., VAN RENSWOUDE, J., KEMPF, C. & KLAUSNER, R. D. 1983. Separation of Fe+3 from transferrin in endocytosis. Role of the acidic endosome. FEBS Lett, 160, 213-6. RAPOPORT, I., CHEN, Y. C., CUPERS, P., SHOELSON, S. E. & KIRCHHAUSEN, T. 1998. Dileucine‐based sorting signals bind to the β chain of AP‐1 at a site distinct and regulated differently from the tyrosine‐based motif‐binding site. The EMBO Journal, 17, 2148-2155-2155. REEVES, J. P. 1979. Accumulation of amino acids by lysosomes incubated with amino acid methyl esters. Journal of Biological Chemistry, 254, 8914-8921. REHEMTULLA, A. & KAUFMAN, R. J. 1992. <1-s2.0-095816699290086X-main.pdf>. REHLING, P., DARSOW, T., KATZMANN, D. J. & EMR, S. D. 1999. Formation of AP-3 transport intermediates requires Vps41 function. Nat Cell Biol, 1, 346-53. A.1 TBC1D23 A-15 REITZ, C. 2018. Retromer Dysfunction and Neurodegenerative Disease. Curr Genomics, 19, 279-288. REN, X., FARÍAS, 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, 755-67. REN, X., PARK, S. Y., BONIFACINO, J. S. & HURLEY, J. H. 2014. How HIV-1 Nef hijacks the AP-2 clathrin adaptor to downregulate CD4. Elife, 3, e01754. RICKMAN, C., MEUNIER, F. A., BINZ, T. & DAVLETOV, B. 2004. High affinity interaction of syntaxin and SNAP-25 on the plasma membrane is abolished by botulinum toxin E. J Biol Chem, 279, 644-51. RIERA-TUR, I., SCHÄFER, T., HORNBURG, D., MISHRA, A., PADILHA, M. D. S., FERNÁNDEZ-MOSQUERA, L., FEIGENBUTZ, D., AUER, P., MANN, M., BAUMEISTER, W., KLEIN, R., MEISSNER, F., RAIMUNDO, N., FERNÁNDEZ- BUSNADIEGO, R. & DUDANOVA, I. 2022. Amyloid-like aggregating proteins cause lysosomal defects in neurons via gain-of-function toxicity. Life Science Alliance, 5, e202101185. ROBINSON, M. S., ANTROBUS, R., SANGER, A., DAVIES, A. K. & GERSHLICK, D. C. 2024. The role of the AP-1 adaptor complex in outgoing and incoming membrane traffic. Journal of Cell Biology, 223. ROBINSON, M. S. & PEARSE, B. M. 1986. Immunofluorescent localization of 100K coated vesicle proteins. J Cell Biol, 102, 48-54. ROMANO-MORENO, M., ASTORGA-SIMÓN, E. N., ROJAS, A. L. & HIERRO, A. 2024. Retromer-mediated recruitment of the WASH complex involves discrete interactions between VPS35, VPS29 and Fam21. bioRxiv, 2024.01.22.576619. ROTH, T. F. & PORTER, K. R. 1964. YOLK PROTEIN UPTAKE IN THE OOCYTE OF THE MOSQUITO AEDES AEGYPTI. L. J Cell Biol, 20, 313-32. ROUX, A., UYHAZI, K., FROST, A. & DE CAMILLI, P. 2006. GTP-dependent twisting of dynamin implicates constriction and tension in membrane fission. Nature, 441, 528- 531. ROY, M. M., NORTON, E. M., RENDAHL, A. K., SCHULTZ, N. E., MCFARLANE, D., GEOR, R. J., MICKELSON, J. R. & MCCUE, M. E. 2020. Assessment of the FAM174A 11G allele as a risk allele for equine metabolic syndrome. Anim Genet, 51, 607-610. RUSSNAK, R., KONCZAL, D. & MCINTIRE, S. L. 2001. A family of yeast proteins mediating bidirectional vacuolar amino acid transport. J Biol Chem, 276, 23849-57. SAITO, K., ARAKI, Y., KONTANI, K., NISHINA, H. & KATADA, T. 2005. Novel Role of the Small GTPase Rheb: Its Implication in Endocytic Pathway Independent of the Activation of Mammalian Target of Rapamycin. The Journal of Biochemistry, 137, 423-430. SALAZAR, G., CRAIGE, B., WAINER, B. H., GUO, J., DE CAMILLI, P. & FAUNDEZ, V. 2005. Phosphatidylinositol-4-kinase type II alpha is a component of adaptor protein-3-derived vesicles. Mol Biol Cell, 16, 3692-704. SANGER, A., HIRST, J., DAVIES, A. K. & ROBINSON, M. S. 2019. Adaptor protein complexes and disease at a glance. J Cell Sci, 132. SAVA, I., DAVIS, L. J., GRAY, S. R., BRIGHT, N. A. & LUZIO, J. P. 2024. Reversible assembly and disassembly of V-ATPase during the lysosome regeneration cycle. Molecular Biology of the Cell, 35, ar63. A-16 Crystallographic statistics and screens SBRISSA, D., IKONOMOV, O. C., STRAKOVA, J., DONDAPATI, R., MLAK, K., DEEB, R., SILVER, R. & SHISHEVA, A. 2004. A mammalian ortholog of Saccharomyces cerevisiae Vac14 that associates with and up-regulates PIKfyve phosphoinositide 5- kinase activity. Mol Cell Biol, 24, 10437-47. SCHLEINITZ, A., PÖTTGEN, L. A., KEREN-KAPLAN, T., PU, J., SAFTIG, P., BONIFACINO, J. S., HAAS, A. & JESCHKE, A. 2023. Consecutive functions of small GTPases guide HOPS-mediated tethering of late endosomes and lysosomes. Cell Rep, 42, 111969. SCHMID, E. M., FORD, M. G., BURTEY, A., PRAEFCKE, G. J., PEAK-CHEW, S. Y., MILLS, I. G., BENMERAH, A. & MCMAHON, H. T. 2006. Role of the AP2 beta- appendage hub in recruiting partners for clathrin-coated vesicle assembly. PLoS Biol, 4, e262. SCHOPPE, J., MARI, M., YAVAVLI, E., AUFFARTH, K., CABRERA, M., WALTER, S., FRÖHLICH, F. & UNGERMANN, C. 2020. AP-3 vesicle uncoating occurs after HOPS-dependent vacuole tethering. Embo j, 39, e105117. SCHUHMACHER, J. S., TOM DIECK, S., CHRISTOFORIDIS, S., LANDERER, C., DAVILA GALLESIO, J., HERSEMANN, L., SEIFERT, S., SCHAFER, R., GINER, A., TOTH-PETROCZY, A., KALAIDZIDIS, Y., BOHNSACK, K. E., BOHNSACK, M. T., SCHUMAN, E. M. & ZERIAL, M. 2023. The Rab5 effector FERRY links early endosomes with mRNA localization. Mol Cell, 83, 1839-1855 e13. SEALS, D. F., EITZEN, G., MARGOLIS, N., WICKNER, W. T. & PRICE, A. 2000. A Ypt/Rab effector complex containing the Sec1 homolog Vps33p is required for homotypic vacuole fusion. Proc Natl Acad Sci U S A, 97, 9402-7. SEAMAN, M. N. 2007. Identification of a novel conserved sorting motif required for retromer-mediated endosome-to-TGN retrieval. J Cell Sci, 120, 2378-89. SEAMAN, M. N., HARBOUR, M. E., TATTERSALL, D., READ, E. & BRIGHT, N. 2009. Membrane recruitment of the cargo-selective retromer subcomplex is catalysed by the small GTPase Rab7 and inhibited by the Rab-GAP TBC1D5. J Cell Sci, 122, 2371- 82. SEAMAN, M. N. J. 2018. Retromer and the cation-independent mannose 6-phosphate receptor-Time for a trial separation? Traffic, 19, 150-152. SHARPE, H. J., STEVENS, T. J. & MUNRO, S. 2010. A comprehensive comparison of transmembrane domains reveals organelle-specific properties. Cell, 142, 158-69. SHEN, K., HUANG, R. K., BRIGNOLE, E. J., CONDON, K. J., VALENSTEIN, M. L., CHANTRANUPONG, L., BOMALIYAMU, A., CHOE, A., HONG, C., YU, Z. & SABATINI, D. M. 2018. Architecture of the human GATOR1 and GATOR1-Rag GTPases complexes. Nature, 556, 64-69. SHIN, J. J. H., CROOK, O. M., BORGEAUD, A. C., CATTIN-ORTOLA, J., PEAK-CHEW, S. Y., BRECKELS, L. M., GILLINGHAM, A. K., CHADWICK, J., LILLEY, K. S. & MUNRO, S. 2020. Spatial proteomics defines the content of trafficking vesicles captured by golgin tethers. Nat Commun, 11, 5987. SHIN, J. J. H., GILLINGHAM, A. K., BEGUM, F., CHADWICK, J. & MUNRO, S. 2017. TBC1D23 is a bridging factor for endosomal vesicle capture by golgins at the trans- Golgi. Nat Cell Biol, 19, 1424-1432. SHVAREV, D., SCHOPPE, J., KÖNIG, C., PERZ, A., FÜLLBRUNN, N., KIONTKE, S., LANGEMEYER, L., JANULIENE, D., SCHNELLE, K., KÜMMEL, D., FRÖHLICH, F., MOELLER, A. & UNGERMANN, C. 2022. Structure of the HOPS tethering complex, a lysosomal membrane fusion machinery. eLife, 11, e80901. A.1 TBC1D23 A-17 SIGISMUND, S., ALGISI, V., NAPPO, G., CONTE, A., PASCOLUTTI, R., CUOMO, A., BONALDI, T., ARGENZIO, E., VERHOEF, L. G., MASPERO, E., BIANCHI, F., CAPUANI, F., CILIBERTO, A., POLO, S. & DI FIORE, P. P. 2013. Threshold- controlled ubiquitination of the EGFR directs receptor fate. Embo j, 32, 2140-57. SIMONETTI, B., PAUL, B., CHAUDHARI, K., WEERATUNGA, S., STEINBERG, F., GORLA, M., HEESOM, K. J., BASHAW, G. J., COLLINS, B. M. & CULLEN, P. J. 2019. Molecular identification of a BAR domain-containing coat complex for endosomal recycling of transmembrane proteins. Nat Cell Biol, 21, 1219-1233. SIMPSON, F., BRIGHT, N. A., WEST, M. A., NEWMAN, L. S., DARNELL, R. B. & ROBINSON, M. S. 1996. A novel adaptor-related protein complex. Journal of Cell Biology, 133, 749-760. SIMPSON, F., PEDEN, A. A., CHRISTOPOULOU, L. & ROBINSON, M. S. 1997. Characterization of the adaptor-related protein complex, AP-3. J Cell Biol, 137, 835- 45. SITARAM, A., DENNIS, M. K., CHAUDHURI, R., DE JESUS-ROJAS, W., TENZA, D., SETTY, S. R. G., WOOD, C. S., SVIDERSKAYA, E. V., BENNETT, D. C., RAPOSO, G., BONIFACINO, J. S. & MARKS, M. S. 2012. Differential recognition of a dileucine-based sorting signal by AP-1 and AP-3 reveals a requirement for both BLOC-1 and AP-3 in delivery of OCA2 to melanosomes. Molecular Biology of the Cell, 23, 3178-3192. SITARAM, A., PICCIRILLO, R., PALMISANO, I., HARPER, D. C., DELL'ANGELICA, E. C., SCHIAFFINO, M. V. & MARKS, M. S. 2008. Localization to Mature Melanosomes by Virtue of Cytoplasmic Dileucine Motifs Is Required for Human OCA2 Function. Molecular Biology of the Cell, 20, 1464-1477. SMYTHE, E. 2015. Role of the rab5 guanine nucleotide exchange factor, Rme-6, in the regulation of clathrin-coated vesicle uncoating. Methods Mol Biol, 1298, 283-93. SOCHACKI, K. A., HEINE, B. L., HABER, G. J., JIMAH, J. R., PRASAI, B., ALFONZO- MÉNDEZ, M. A., ROBERTS, A. D., SOMASUNDARAM, A., HINSHAW, J. E. & TARASKA, J. W. 2021. The structure and spontaneous curvature of clathrin lattices at the plasma membrane. Developmental Cell, 56, 1131-1146.e3. SOLINGER, J. A. & SPANG, A. 2022. Sorting of cargo in the tubular endosomal network. Bioessays, 44, e2200158. SÖLLNER, T., WHITEHEART, S. W., BRUNNER, M., ERDJUMENT-BROMAGE, H., GEROMANOS, S., TEMPST, P. & ROTHMAN, J. E. 1993. SNAP receptors implicated in vesicle targeting and fusion. Nature, 362, 318-324. SPANG, A. 2016. Membrane Tethering Complexes in the Endosomal System. Front Cell Dev Biol, 4, 35. STACEY, J. C. V., TAN, A., LU, J. M., JAMES, L. C., DICK, R. A. & BRIGGS, J. A. G. 2023. Two structural switches in HIV-1 capsid regulate capsid curvature and host factor binding. Proc Natl Acad Sci U S A, 120, e2220557120. STACHOWIAK, J. C., SCHMID, E. M., RYAN, C. J., ANN, H. S., SASAKI, D. Y., SHERMAN, M. B., GEISSLER, P. L., FLETCHER, D. A. & HAYDEN, C. C. 2012. Membrane bending by protein–protein crowding. Nature Cell Biology, 14, 944-949. STARCEVIC, M. & DELL'ANGELICA, E. C. 2004. Identification of snapin and three novel proteins (BLOS1, BLOS2, and BLOS3/reduced pigmentation) as subunits of biogenesis of lysosome-related organelles complex-1 (BLOC-1). J Biol Chem, 279, 28393-401. A-18 Crystallographic statistics and screens STOCKHAMMER, A., ADARSKA, P., NATALIA, V., HEUHSEN, A., KLEMT, A., HAREL, S., RODILLA-RAMIREZ, C., SPALT, C., ÖZSOY, E., LEUPOLD, P., GRINDEL, A., FOX, E., MEJEDO, J. O., PUCHKOV, D., HAUCKE, V. & BOTTANELLI, F. 2023. Multi-functional ARF1 compartments serve as a hub for short-range cargo transfer to endosomes. bioRxiv, 2023.10.27.564143. SU, M. Y., FROMM, S. A., ZONCU, R. & HURLEY, J. H. 2020. Structure of the C9orf72 ARF GAP complex that is haploinsufficient in ALS and FTD. Nature, 585, 251-255. SUTTON, R. B., FASSHAUER, D., JAHN, R. & BRUNGER, A. T. 1998. Crystal structure of a SNARE complex involved in synaptic exocytosis at 2.4 A resolution. Nature, 395, 347-53. TAGILTSEV, G., HASELWANDTER, C. A. & SCHEURING, S. 2021. Nanodissected elastically loaded clathrin lattices relax to increased curvature. Sci Adv, 7. TAYLOR, R. J., TAGILTSEV, G. & BRIGGS, J. A. G. 2023. The structure of COPI vesicles and regulation of vesicle turnover. FEBS Letters, 597, 819-835. TEGUNOV, D. & CRAMER, P. 2019. Real-time cryo-electron microscopy data preprocessing with Warp. Nature Methods, 16, 1146-1152. TESEI, G., TROLLE, A. I., JONSSON, N., BETZ, J., KNUDSEN, F. E., PESCE, F., JOHANSSON, K. E. & LINDORFF-LARSEN, K. 2024. Conformational ensembles of the human intrinsically disordered proteome. Nature, 626, 897-904. THEOS, A. C., TENZA, D., MARTINA, J. A., HURBAIN, I., PEDEN, A. A., SVIDERSKAYA, E. V., STEWART, A., ROBINSON, M. S., BENNETT, D. C., CUTLER, D. F., BONIFACINO, J. S., MARKS, M. S. & RAPOSO, G. 2005. Functions of adaptor protein (AP)-3 and AP-1 in tyrosinase sorting from endosomes to melanosomes. Mol Biol Cell, 16, 5356-72. TRAUB, L. M. 2009. Tickets to ride: selecting cargo for clathrin-regulated internalization. Nat Rev Mol Cell Biol, 10, 583-96. TSUN, Z. Y., BAR-PELED, L., CHANTRANUPONG, L., ZONCU, R., WANG, T., KIM, C., SPOONER, E. & SABATINI, D. M. 2013. The folliculin tumor suppressor is a GAP for the RagC/D GTPases that signal amino acid levels to mTORC1. Mol Cell, 52, 495-505. UNGEWICKELL, E. 1985. The 70‐kd mammalian heat shock proteins are structurally and functionally related to the uncoating protein that releases clathrin triskelia from coated vesicles. The EMBO Journal, 4, 3385-3391-3391. UNGEWICKELL, E. & BRANTON, D. 1981. Assembly units of clathrin coats. Nature, 289, 420-2. VAN DEN BOOMEN, D. J. H., SIENKIEWICZ, A., BERLIN, I., JONGSMA, M. L. M., VAN ELSLAND, D. M., LUZIO, J. P., NEEFJES, J. J. C. & LEHNER, P. J. 2020. A trimeric Rab7 GEF controls NPC1-dependent lysosomal cholesterol export. Nat Commun, 11, 5559. VAN DER BEEK, J., JONKER, C., VAN DER WELLE, R., LIV, N. & KLUMPERMAN, J. 2019. CORVET, CHEVI and HOPS - multisubunit tethers of the endo-lysosomal system in health and disease. J Cell Sci, 132. VASSAR, R., KOVACS, D. M., YAN, R. & WONG, P. C. 2009. The beta-secretase enzyme BACE in health and Alzheimer's disease: regulation, cell biology, function, and therapeutic potential. J Neurosci, 29, 12787-94. VIRSHUP, D. M. & BENNETT, V. 1988. Clathrin-coated vesicle assembly polypeptides: physical properties and reconstitution studies with brain membranes. J Cell Biol, 106, 39-50. A.1 TBC1D23 A-19 WANG, Y. J., WANG, J., SUN, H. Q., MARTINEZ, M., SUN, Y. X., MACIA, E., KIRCHHAUSEN, T., ALBANESI, J. P., ROTH, M. G. & YIN, H. L. 2003. Phosphatidylinositol 4 phosphate regulates targeting of clathrin adaptor AP-1 complexes to the Golgi. Cell, 114, 299-310. WILLIAMS, E. T., CHEN, X. & MOORE, D. J. 2017. VPS35, the Retromer Complex and Parkinson's Disease. J Parkinsons Dis, 7, 219-233. WILLIAMS, R. L. & URBÉ, S. 2007. The emerging shape of the ESCRT machinery. Nature Reviews Molecular Cell Biology, 8, 355-368. WONG, M. & MUNRO, S. 2014. Membrane trafficking. The specificity of vesicle traffic to the Golgi is encoded in the golgin coiled-coil proteins. Science, 346, 1256898. WROBEL, A. G., KADLECOVA, Z., KAMENICKY, J., YANG, J. C., HERRMANN, T., KELLY, B. T., MCCOY, A. J., EVANS, P. R., MARTIN, S., MÜLLER, S., SALOMON, S., SROUBEK, F., NEUHAUS, D., HÖNING, S. & OWEN, D. J. 2019. Temporal Ordering in Endocytic Clathrin-Coated Vesicle Formation via AP2 Phosphorylation. Dev Cell, 50, 494-508.e11. WUNDERLEY, L., ZHANG, L., YARWOOD, R., QIN, W., LOWE, M. & WOODMAN, P. 2021. Endosomal recycling tubule scission and integrin recycling involve the membrane curvature-supporting protein LITAF. J Cell Sci, 134. XU, H., CHANG, F., JAIN, S., HELLER, B. A., HAN, X., LIU, Y. & EDWARDS, R. H. 2022. SNX5 targets a monoamine transporter to the TGN for assembly into dense core vesicles by AP-3. Journal of Cell Biology, 221. XU, Y., HORTSMAN, H., SEET, L., WONG, S. H. & HONG, W. 2001. SNX3 regulates endosomal function through its PX-domain-mediated interaction with PtdIns(3)P. Nat Cell Biol, 3, 658-66. YAN, Y., DENEF, N. & SCHÜPBACH, T. 2009. The vacuolar proton pump, V-ATPase, is required for notch signaling and endosomal trafficking in Drosophila. Dev Cell, 17, 387-402. YONG, X., JIA, G., LIU, Z., ZHOU, C., YI, J., TANG, Y., CHEN, L., CHEN, L., WANG, Y., SUN, Q., BILLADEAU, D. D., SU, Z. & JIA, D. 2023. Cryo-EM structure of the Mon1-Ccz1-RMC1 complex reveals molecular basis of metazoan RAB7A activation. Proc Natl Acad Sci U S A, 120, e2301725120. YU, X., BREITMAN, M. & GOLDBERG, J. 2012. A structure-based mechanism for Arf1- dependent recruitment of coatomer to membranes. Cell, 148, 530-42. ZACCAI, N. R., KADLECOVA, Z., DICKSON, V. K., KOROBCHEVSKAYA, K., KAMENICKY, J., KOVTUN, O., UMASANKAR, P. K., WROBEL, A. G., KAUFMAN, J. G. G., GRAY, S. R., QU, K., EVANS, P. R., FRITZSCHE, M., SROUBEK, F., HÖNING, S., BRIGGS, J. A. G., KELLY, B. T., OWEN, D. J. & TRAUB, L. M. 2022. FCHO controls AP2’s initiating role in endocytosis through a PtdIns(4,5)P2-dependent switch. Science Advances, 8, eabn2018. ZAREMBA, S. & KEEN, J. H. 1983. Assembly polypeptides from coated vesicles mediate reassembly of unique clathrin coats. J Cell Biol, 97, 1339-47. ZHANG, S., LI, L., LIU, X. & ZHONG, Q. 2024. The hookup model of the HOPS complex in autophagosome-lysosome fusion. Autophagy, 20, 714-715. ZHANG, S., REN, Q., NOVICK, S. J., STRUTZENBERG, T. S., GRIFFIN, P. R. & BAO, H. 2021. One-step construction of circularized nanodiscs using SpyCatcher-SpyTag. Nature Communications, 12, 5451. A-20 Crystallographic statistics and screens ZHANG, Z., LEE, C. H., MANDIYAN, V., BORG, J. P., MARGOLIS, B., SCHLESSINGER, J. & KURIYAN, J. 1997. Sequence-specific recognition of the internalization motif of the Alzheimer's amyloid precursor protein by the X11 PTB domain. Embo j, 16, 6141-50. ZHAO, L., DENG, H., YANG, Q., TANG, Y., ZHAO, J., LI, P., ZHANG, S., YONG, X., LI, T., BILLADEAU, D. D. & JIA, D. 2023. FAM91A1-TBC1D23 complex structure reveals human genetic variations susceptible for PCH. Proc Natl Acad Sci U S A, 120, e2309910120. ZHAO, M., WU, S., ZHOU, Q., VIVONA, S., CIPRIANO, D. J., CHENG, Y. & BRUNGER, A. T. 2015. Mechanistic insights into the recycling machine of the SNARE complex. Nature, 518, 61-7. ZHU, H., LI, Q., LIAO, T., YIN, X., CHEN, Q., WANG, Z., DAI, M., YI, L., GE, S., MIAO, C., ZENG, W., QU, L., JU, Z., HUANG, G., CANG, C. & XIONG, W. 2021. Metabolomic profiling of single enlarged lysosomes. Nat Methods, 18, 788-798. ZHU, Y., LI, S., JAUME, A., JANI, R. A., DELEVOYE, C., RAPOSO, G. & MARKS, M. S. 2022. Type II phosphatidylinositol 4-kinases function sequentially in cargo delivery from early endosomes to melanosomes. J Cell Biol, 221. A.1 TBC1D23 A-21 Appendix A Crystallographic statistics and screens A.1 TBC1D23 Table 1 TBC1D23 (559-684) TBC1D23 (559-677) STX16 (206-221) PDB:8QQF Autoproc-STARANISO Autoproc-STARANISO Space group P 62 2 2 P 65 2 2 Resolution range (Å) 194.38-3.17 116.19-2.19 InnerShell 112.394-11.533 116.19-8.280 OuterShell 3.814-3.172 2.479-2.187 Rmerge (all I+ & I-) 0.939 (2.988) 0.133 (2.607) Rmerge (within I+ & I-) 0.989 (2.998) 0.141 (2.710) Rmeas (all I+ & I-) 0.951 (3.028) 0.136 (2.651) Rmeas (within I+ & I-) 1.013 (3.073) 0.145 (2.791) Rpim (all I+ & I-) 0.151 (0.485) 0.024 (0.472) Rpim (within I+ & I-) 0.220 (0.667) 0.034 (0.668) Number of observations 352792 (17313) 486437 (23072) Number of unique observations 9118 (457) 14895 (745) <(I)/sd(I)> 7.5 (1.9) 23.0 (1.9) CC(1/2) 0.972 (0.627) 0.999 (0.707) Completeness % (spherical) 90.8 (47.7) 43.2 (7.0) Completeness % (ellipsoidal) 90.8 (47.7) 94.8 (82.9) Multiplicity 38.7 (37.9) 32.7 (31.0) RMS bond lengths (Å) 0.0063 0.0077 RMS bond angles (°) 1.644 1.605 Reflections (All/Free) 8038 / 356 14895/752 Rfactor/Rfree 0.209 / 0.274 0.186 / 0.259 Ramachandran (%) (favoured/outliers) 97.11/2.89 99.240.76 A-22 Crystallographic statistics and screens A.2 Sigma2-Alpha (S2A) Table 2 S2A:LRP9 S2A:PI4K2a S2A:MFSD12 Autoproc- STARANISO Autoproc Autoproc Space group C 1 2 1 P 1 P 1 21 1 Resolution range (Å) 75.626-2.889 189.459-2.665 67.552-1.756 InnerShell 75.626-9.105 189.459-7.234 67.552-4.765 OuterShell 3.198-2.889 2.711-2.665 1.786-1.756 Rmerge (all I+ & I-) 0.305 (1.265) 0.106 (1.005) 0.060 (0.978) Rmerge (within I+ & I-) 0.293 (1.187) 0.101 (1.122) 0.056 (0.923) Rmeas (all I+ & I-) 0.331 (1.369) 0.131 (1.318) 0.069 (1.152) Rmeas (within I+ & I-) 0.346 (1.403) 0.142 (1.587) 0.072 (1.229) Rpim (all I+ & I-) 0.126 (0.519) 0.075 (0.843) 0.033 (0.604) Rpim (within I+ & I-) 0.183 (0.743) 0.101 (1.122) 0.045 (0.804) Number of observations 346844 (17264) 386901 (15238) 427260 (17902) Number of unique observations 50690 (2534) 140164 (7088) 102187 (5052) <(I)/sd(I)> 6.9 (1.9) 4.2 (0.8) 12.3 (1.5) CC(1/2) 0.986 (0.464) 0.993 (0.353) 0.998 (0.354) Completeness % (spherical) 66.7 (12.8) 97.1 (98.0) 97.6 (96.8) Completeness % (ellipsoidal) 81.6 (3.6) - - Multiplicity 6.8 (6.8) 2.8 (2.1) 4.2 (3.5) RMS bond lengths (Å) 0.0186 0.0166 0.01 RMS bond angles (°) 2.522 2.408 1.79 Reflections (All/Free) 50690/2480 136402/5265 92119/4723 Rfactor/Rfree 0.173/0.284 0.181/0.245 0.182/0.229 Ramachandran (%) (favoured/outliers) 93.91/1.28 96.73/1.10 96.00/1.12 A.3 A-23 Table 3 S2A:CD3gamma S2A:BACE1 Autoproc-STARANISO Dials Xia2 Space group P 1 C 1 2 1 Resolution range (Å) 117.127-3.110 190.5-3.02 InnerShell 117.127-9.944 190.5-8.18 OuterShell 3.479-3.110 3.07-3.02 Rmerge (all I+ & I-) 0.129 (0.319) 1.553 (12.027) Rmerge (within I+ & I-) 0.174 (0.746) 1.547 (12.009) Rmeas (all I+ & I-) 0.169 (0.441) 1.593 (12.342) Rmeas (within I+ & I-) 0.246 (1.055) 1.629 (12.651) Rpim (all I+ & I-) 1.055 (0.303) 0.357 (2.761) Rpim (within I+ & I-) 0.174 (0.746) 0.508 (3.966) Number of observations 96921 (3969) 1030706 (49087) Number of unique observations 47140 (2353) 52129 (2485) <(I)/sd(I)> 2.1 (1.5) 3.2 (0.4) CC(1/2) 0.986 (0.674) 0.972 (0.309) Completeness % (spherical) 54.3 (9.5) 99.7 (95.0) Completeness % (ellipsoidal) 78.4 (40.6) - Multiplicity 2.1 (1.7) 19.8 (19.8) RMS bond lengths (Å) 0.0255 0.016 RMS bond angles (°) 2.663 2.197 Reflections (All/Free) 47141/2430 64067/3257 Rfactor/Rfree 0.233/0.31 0.205/0.266 Ramachandran (%) (favoured/outliers) 93.7/1.59 94.2/1.15 A-24 Crystallographic statistics and screens Table 4 S2A:MFSD12 L(+1)I S2A:MFSD12 P(-3)S S2A:MFSD12 E(-4)D GEHTPLIAPAT GEHTSLLAPAT GDHTPLLAPAT Dials Xia2 Autoproc Autoproc- STARANISO Space group P 1 21 1 P 1 21 1 P 1 21 1 Resolution range (Å) 50.21-1.77 96.661-2.066 90.773-2.293 InnerShell 50.23 - 4.80 96.661-5.514 90.773-7.052 OuterShell 1.80 - 1.77 2.066-2.031 2.538-2.293 Rmerge (all I+ & I-) 0.076 (2.362) 0.103 (2.964) 0.103 (0.840) Rmerge (within I+ & I-) - 0.096 (2.623) 0.096 (0.759) Rmeas (all I+ & I-) 0.082 (2.554) 0.111 (3.302) 0.113 (0.960) Rmeas (within I+ & I-) - 0.113 (3.232) 0.115 (0.980) Rpim (all I+ & I-) 0.031 (0.963) 0.042 (1.427) 0.045 (0.454) Rpim (within I+ & I-) - 0.060 (1.862) 0.063 (0.613) Number of observations 701526 (34755) 392995 (4922) 293703 (9914) Number of unique observations 100830 (5022) 56992 (973) 47832 (2393) <(I)/sd(I)> 17.5 (0.6) 11.3 (0.5) 11.7 (1.6) CC(1/2) 0.999 (0.307) 0.999 (0.243) 0.998 (0.529) Completeness % (spherical) 100 (99.7) 83.9 (29.1) 70.4 (13.5) Completeness % (ellipsoidal) - - 94.1 (63.6) Multiplicity 7.0 (6.9) 6.1 (4.1) 6.1 (4.1) RMS bond lengths (Å) 0.021 0.0093 0.0091 RMS bond angles (°) 2.606 1.676 1.638 Reflections (All/Free) 82762/4107 49916/2389 47832/2390 Rfactor/Rfree 0.191/0.256 0.187/0.255 0.187/0.248 Ramachandran (%) (favoured/outliers) 96.33/0.94 96.62/0.85 94.95/1.50 A.4 Sigma3-Delta (S3D) A-25 A.4 Sigma3-Delta (S3D) Table 5 S3D:MFSD12 S3D:MFSD12:Arf1 Autoproc-STARANISO Dials Xia2 Space group C 2 2 21 P 2 21 21 Resolution range 99.706-2.262 68.01-2.35 InnerShell 99.706-7.391 - OuterShell 2.533-2.264 2.39-2.35 Rmerge (all I+ & I-) 0.117 (1.581) 0.101 (3.087) Rmerge (within I+ & I-) 0.115 (1.554) - Rmeas (all I+ & I-) 0.124 (1.673) 0.105 (3.199) Rmeas (within I+ & I-) 0.130 (1.741) - Rpim (all I+ & I-) 0.042 (0.539) 0.028 (0.838) Rpim (within I+ & I-) 0.060 (0.776) - Number of observations 294156 (15631) 1187535 (60312) Number of unique observations 34064 (1703) 85873 (4198) <(I)/sd(I)> 12.1 (1.7) 13.5 (0.4) CC(1/2) 0.998 (0.501) 0.999 (0.336) Completeness % (spherical) 62.1 (10.9) 99.43 (98.15) Completeness % (ellipsoidal) 93.9 (58.5) - Multiplicity 8.6 (9.2) 13.83 (14.37) RMS bond lengths (Å) 0.0098 0.007 RMS bond angles (°) 1.876 1.522 Reflections (All/Free) 34064/1742 85811/4330 Rfactor/Rfree 0.194/0.232 0.229/0.278 Ramachandran (%) (favoured/outliers) 94.46/1.78 94.75/1.61 A-26 Crystallographic statistics and screens A.5 AP-3 Appendages Table 6 Delta ear Beta ear Autoproc Dials Xia2 Space group P 65 2 2 P 21 21 21 Resolution range 100.790-2.718 87.09-3.37 InnerShell 100.790-7.376 87.12-9.14 OuterShell 2.764-2.718 3.43-3.37 Rmerge (all I+ & I-) 0.896 (14.039) 0.353 (1.911) Rmerge (within I+ & I-) 0.969 (14.006) 0.342 (1.856) Rmeas (all I+ & I-) 0.911 (14.285) 0.369 (1.993) Rmeas (within I+ & I-) 1.002 (14.480) 0.372 (2.021) Rpim (all I+ & I-) 0.168 (2.624) 0.106 (0.563) Rpim (within I+ & I-) 0.252 (3.667) 0.146 (0.793) Number of observations 511605 (25562) 303380 (14815) Number of unique observations 17686 (871) 24920 (1196) <(I)/sd(I)> 8.6 (1.6) 2.7 (0.4) CC(1/2) 0.990 (0.358) 0.989 (0.147) Completeness % (spherical) 100 (100) 100.0 (99.3) Completeness % (ellipsoidal) - - Multiplicity 28.9 (29.3) 12.2 (12.4) RMS bond lengths (Å) 0.0079 0.0079 RMS bond angles (°) 1.559 1.616 Reflections (All/Free) 17682/926 22172/1066 Rfactor/Rfree 0.241/0.303 0.228/0.272 Ramachandran (%) (favoured/outliers) 94.5/1.9 95.4/1.2 A.6 CIMR crystallization screens A-27 A.6 CIMR crystallization screens Table 7 Screen Components Supplier CIMR1 Structure Screen 1 Structure Screen 2 Molecular Dimensions CIMR2 Wizard Screen I Wizard Screen II Emerald Biosystems CIMR3 Wizard Screen III Wizard Screen IV Emerald Biosystems CIMR4 Cryo I Cryo II Emerald Biosystems CIMR5 PEG/Ion Screen PEG/Ion Screen 2 Hampton Research CIMR6 Index Screen Hampton Research CIMR7 PACT Premier Molecular Dimensions CIMR8 JCSG-plus Molecular Dimensions CIMR9 (NH4)2SO4 Grid Screen PEG 6000 Grid Screen PEG/LiCl Grid Screen MPD Grid Screen Hampton Research CIMR10 MemStart MemSys Molecular Dimensions CIMR11 Morpheus Molecular Dimensions CIMR12 Stura Screen Macrosol Screen Molecular Dimensions CIMR13 Proplex Molecular Dimensions Appendix B Buffers for protein biochemistry Table 8 Protein Buffer name Composition AP3 core Lysis Buffer 200mM NaCl, 20mM Tris (pH 8), 2mM 2- mecaptoethanol, 10mM imidazole (pH 8), 1mM MnCl2, 50μg DNAse1, 6.25mg AEBSF hydrochloride, 100μl PI Mix III, 100μL Aprotinin, 100μl Soy trypsin inhibitor Buffer A 200mM NaCl, 20mM Tris (pH 8.5), 2mM 2-mecaptoethanol, 10mM imidazole (pH 8), 25μg/ml AEBSF hydrochloride Buffer B 200mM NaCl, 20mM Tris (pH 8.5), 2mM 2-mecaptoethanol, 300mM imidazole (pH 8), 25μg/ml AEBSF hydrochloride AP-3 Cryo-EM buffer 200mM NaCl, 20mM Tris (pH 8), 0.5mM TCEP SASC/S3D Buffer A 300mM NaCl, 50mM Tris (pH 7.4), 2mM 2-mecaptoethanol, 10mM imidazole (pH 8) Buffer B 300mM NaCl, 50mM Tris (pH 7.4), 300mM 2-mecaptoethanol, 10mM imidazole (pH 8) SASC ITC buffer 300mM NaCl, 100mM Tris (pH 7.4), 0.5mM TCEP SASC Crystallography buffer 250mM NaCl, 20mM Tris (pH 7.4), 1mM DTT TBC1D23 Lysis Buffer 250 mM NaCl, 20 mM Tris (pH 7.4), 1 mM DTT, 25μg/ml AEBSF hydrochloride, MnCl2 and DNAse I. Buffer A 250 mM NaCl, 20 mM Tris (pH 7.4), 1 mM DTT, 25μg/ml AEBSF hydrochloride TBC1D23 ITC buffer 100mM NaCl, 100mM Tris (pH 7.4), 0.5mM TCEP TBC1D23 Crystallography buffer 250 mM NaCl, 20 mM Tris (pH 7.4), 1 mM DTT, 25μg/ml AEBSF hydrochloride Arf1 (Q71L) Lysis Buffer 250 mM NaCl, 20 mM Tris (pH 7.4), 1 mM DTT, 25μg/ml AEBSF hydrochloride, 1mM MgCl2, 0.5mM MnCl2, DNAse I. Buffer A 250 mM NaCl, 20 mM Tris (pH 7.4), 1 mM DTT, 25μg/ml AEBSF hydrochloride, 1mM MgCl2. B-2 Buffers for protein biochemistry Table 8 (continued) Beta ear and Delta ear Lysis Buffer 200 mM NaCl, 20 mM Tris (pH 7.4), 1 mM DTT, 25μg/ml AEBSF hydrochloride, 0.5mM MnCl2, DNAse I. Buffer A 250 mM NaCl, 20 mM Tris (pH 7.4), 1 mM DTT, 25μg/ml AEBSF hydrochloride. Nanodisc SPNW15 SPNW25 SPNW30 SPNW50 Lysis Buffer 100 mM NaCl, 50 mM Tris-HCl (pH 8), 5% glycerol, 2 mM β-mercaptoethanol, DNAse1, 25μg/ml AEBSF hydrochloride. Buffer A 100 mM NaCl, 50 mM Tris-HCl (pH 8), 5% glycerol, 2 mM β-mercaptoethanol. Buffer B 400 mM NaCl, 50 mM Tris-HCl (pH 8), 5% glycerol, 2 mM β-mercaptoethanol, 300mM imidazole. Nanodisc EM buffer 200 mM NaCl, 20 mM Tris (pH 7.4), 0.5 mM TCEP. Nanodisc SPNW15- link-Arf1 Lysis Buffer 100 mM NaCl, 50 mM Tris-HCl (pH 8), 5% glycerol, 2 mM β-mercaptoethanol, 2mM MgCl2, DNAse1, 25μg/ml AEBSF hydrochloride. Buffer A 100 mM NaCl, 50 mM Tris-HCl (pH 8), 5% glycerol, 2mM MgCl2, 2 mM β-mercaptoethanol. Buffer B 400 mM NaCl, 50 mM Tris-HCl (pH 8), 5% glycerol, 0.5mM MgCl2, 2 mM β-mercaptoethanol, 300mM imidazole. Nanodisc EM buffer 200 mM NaCl, 20 mM Tris (pH 7.4), 1mM MgCl2 0.5 mM TCEP. A.6 CIMR crystallization screens C-3 Appendix C Publications from this thesis Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 1 of 15 C E L L B I O L O G Y Cargo selective vesicle tethering: The structural basis for binding of specific cargo proteins by the Golgi tether component TBC1D23 Jérôme Cattin- Ortolá1†‡, Jonathan G. G. Kaufman2†, Alison K. Gillingham1†, Jane L. Wagstaff1, Sew- Yeu Peak- Chew1, Tim J. Stevens1, Jérôme Boulanger1, David J. Owen2*, Sean Munro1* The Golgi- localized golgins golgin- 97 and golgin- 245 capture transport vesicles arriving from endosomes via the protein TBC1D23. The amino- terminal domain of TBC1D23 binds to the golgins, and the carboxyl- terminal domain of TBC1D23 captures the vesicles, but how it recognizes specific vesicles was unclear. A search for binding partners of the carboxyl- terminal domain unexpectedly revealed direct binding to carboxypeptidase D and syntaxin- 16, known cargo proteins of the captured vesicles. Binding is via a threonine- leucine- tyrosine (TLY) sequence present in both proteins next to an acidic cluster. A crystal structure reveals how this acidic TLY motif binds to TBC1D23. An acidic TLY motif is also present in the tails of other endosome- to- Golgi cargo, and these also bind TBC1D23. Structure- guided mutations in the carboxyl- terminal domain that disrupt motif binding in vitro also block vesicle capture in  vivo. Thus, TBC1D23 attached to golgin- 97 and golgin- 245 captures vesicles by a previously unde- scribed mechanism: the recognition of a motif shared by cargo proteins carried by the vesicle. INTRODUCTION The transport of proteins between the organelles of the secretory and endocytic pathways is mediated by tubular/vesicular carriers that bud off donor organelles and then fuse with destination or- ganelles. The process by which the vesicles recognize their cor- rect destination involves tethering factors that mediate the initial capture of the vesicle before the SNARE proteins in the vesicle and destination membranes form a complex that then drives membrane fusion (1, 2). Depending on the transport step, these tethering factors are long coiled- coil proteins or large protein complexes that are located to a specific organelle (3–5). In the case of the Golgi apparatus, these tethers include a set of long coiled- coil proteins called “golgins.” They are located to specific parts of the Golgi stack via C- terminal domains that either bind small guanosine triphosphatases (GTPases) or form a single trans- membrane domain that spans the Golgi bilayer (6–8). Their role as tethers has been clearly demonstrated by the finding that replacing their C- terminal domains with one that directs targeting to mito- chondria results in the accumulation of specific Golgi- destined carriers at this ectopic location (9–11). The ability of golgins to capture vesicles raises the question of how these and other tethers can recognize the correct vesicles. In the case of the golgins, conserved N- terminal regions are neces- sary and sufficient to mediate vesicle capture (10). Three golgins, golgin- 97, golgin- 245, and GCC88, can capture endosome- to- Golgi carriers, and the first two share a closely related vesicle capture motif at the N terminus. This motif binds to a cytosolic protein, TBC1D23, a member of the Tre2/Bub2/Cdc16 (TBC) family of Rab GTPase- activating proteins (GAPs) (Fig.  1A). However, TBC1D23 is unlikely to have GAP activity as it lacks the conserved catalytic residues necessary for stimulating guano- sine 5′- triphosphate hydrolysis (12, 13). TBC1D23 is located to the Golgi in a manner that is dependent on the presence of gol- gin- 97 and golgin- 245, and it can, by itself, capture vesicles when relocated to mitochondria, indicating that it forms a bridge be- tween the golgins and endosome- derived vesicles (12). In addi- tion to the TBC domain, TBC1D23 has a rhodanese domain and a C- terminal domain that is related to pleckstrin homology (PH) domains (13, 14). The TBC/rhodanese N- terminal region of the protein binds directly to the N terminus of golgin- 97 or golgin- 245, and the C- terminal domain is necessary and sufficient for vesicle capture (12, 14). Between these two regions, there is a linker region, part of which binds to a complex of three proteins FAM91A1, WDR11, and C17orf75 (12, 15, 16). The function of this complex is unknown, and the part of TBC1D23 that it binds is not required for vesicle capture (12). The ability of the C- terminal domain of TBC1D23 to capture a specific class of vesicles means that it must recognize a feature shared by these vesicles. Previous studies have shown that the C terminus can bind to the endosomal Wiskott- Aldrich syndrome protein and scar homolog (WASH) complex via the FAM21 sub- unit, which has a long C- terminal tail that also binds several other WASH interactors and regulators (12, 17, 18). However, WASH is thought to be involved in retromer- and retriever- based carrier for- mation at endosomes, including carriers that are destined for the Golgi and others destined for the plasma membrane (19, 20). This led to the suggestion that TBC1D23 is able to bind to an additional factor that allows it to specifically capture carriers destined for the Golgi (12). Here, we describe the application of an unbiased affinity chromatography approach to identifying interaction partners of the C- terminal domain of TBC1D23 and report that it can bind directly to the cytoplasmic tails of several of cargo proteins that are found in endosome- to- Golgi carriers. This provides a mechanism by which TBC1D23 and its associated golgins can capture and tether endo- some–derived carriers as they arrive at the Golgi. 1MRC Laboratory of Molecular Biology, Francis Crick Avenue, Cambridge CB2 0QH, UK. 2Cambridge Institute for Medical Research, Cambridge Biomedical Campus, Hills Road, Cambridge CB2 0XY, UK. *Corresponding author. Email: djo@ 30cam. ac. uk (D.J.O.); sean@ mrc- lmb. cam. ac. uk (S.M.) †These authors contributed equally to this work. ‡Present address: Transition Bio, Eastbrook, Cambridge CB2 8DU, UK. Copyright © 2024 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution License 4.0 (CC BY). D ow nloaded from https://w w w .science.org on A pril 22, 2024 C-4 Publications from this thesis Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 2 of 15 RESULTS TBC1D23 binds directly to the cytoplasmic tail of the endocytic cargo carboxypeptidase D To identify binding partners for the C- terminal PH- like domain of TBC1D23, we immobilized on beads a glutathione S- transferase (GST) fusion to the domain and performed affinity chromatography of lysates from 293T cells. Mass spectrometry (MS) of the bound proteins revealed, as expected, that the C- terminal domain could selectively enrich the subunits of the FAM91A1 complex along with the WASH complex subunit FAM21, while the N- terminal domain bound to its partner golgins (Fig. 1B). As is typical for GST fusion– based purifications, the proteins bound to the C- terminal domain −10 −5 0 5 10 2 3 4 Log2 fold change GST-CPD vs. GST-CIMPR −L og (P v al ue ) TBC1D23 GGA2 CLINT1 CLTA GGA1 CLTB CLTC SNX6 SNX2 SNX5 SNX1 Clathrin and adaptors ESCPE-1 complex 0 50 100 150 0 50 100 150 200 FAM91A1/WDR11 C17orf75 complex WASH complex Golgin-97/245 WDR11 FAM91A1 C17ORF75 FAM21A Golgin-245Golgin-97 CPD 1 684 15 295 316 460 (1–331) Golgin-97/245 (514–558) FAM91A1 complex (559–684) Vesicles FAM21 TBC PH-likeRhodanese Unstructured TBC1D23 domain structureA GST-TBC1D23 (1–513) G ST -T BC 1D 23 (5 14 –6 84 ) C GST-CPD (1321–1380) GST-CIMPR(2327-2491) B D His6 _35 _25 _40 _70 _50 _35 _25 _40 _70 _50 kDaFuri n ( 73 9– 79 4) CPD (1 32 1– 13 80 ) (−) CIM PR (2 32 7– 24 91 ) TGN46 (4 03 –4 37 ) + TB C 1D 23 -H is 6 Input (3%) Be ad s al on e GST- TB C 1D 23 -H is 6 _70 E GST-TBC1D23 Inp ut (3% ) His-MBP- CPD (1321–1380) − + − + − + − + − + − + _70 H is 6 -M PB-C PD (1321–1380) 1–559514–684 _50 _40 _35 _25 kDa 514–5581–513 559–684 _100 GST Fig. 1. The C- terminal domain of TBC1D23 forms a complex with the cytoplasmic tail of CPD. (A) Domain structure of mouse TBC1D23 (UniProt Q8K0F1), the version of the protein used in this study. Labeling indicates the regions found by deletion mapping to bind the golgins, the FAM91A1 complex, the WASH complex, and vesicles (12, 14). (B) MS analysis from affinity chromatography of 293T cell lysates using GST- TBC1D23 fragments. The plot compares the average spectral counts from two inde- pendent replicates of GST- TBC1D23 (1 to 513) versus GST- TBC1D23 (514 to 684). Values are in data S1. (C) Volcano plot of MS analysis comparing the eluates from affinity chromatography of 293T cell lysates using the cytoplasmic tails of CPD or CIMPR. The plot compares the spectral intensities from proteins bound to each bait, using data from three independent biological replicates. Endosomal sorting complex for promoting exit 1 (ESCPE- 1). Values are in data S1. (D) Coomassie- stained gel and anti- His6 immunoblot showing that TBC1D23- His6 binds directly and specifically to the cytoplasmic tail of CPD. GST- tagged tails of indicated endocytic cargoes were immobilized on beads and incubated with bacterial lysate containing TBC1D23- His6. Representative of three repeats. (E) Coomassie- stained gel showing that the C- terminal domain of TBC1D23 is necessary and sufficient for binding to CPD. GST- tagged fragments of TBC1D23 were immobilized on beads and incubated with lysate from bacteria expressing the cytoplasmic tail of CPD [His6- MBP- CPD (1321 to 1380)]. Representative of two repeats. D ow nloaded from https://w w w .science.org on A pril 22, 2024 A.6 CIMR crystallization screens C-5 Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 3 of 15 contained several abundant chaperones and cytosolic enzymes that are likely to reflect nonspecific interactions, but in addition to these, we identified carboxypeptidase D (CPD). This metalloprotease is broadly expressed and removes C- terminal basic residues following the action of furin and related proteases on diverse substrates in- cluding neuropeptides and growth factors (21, 22). CPD is known to be primarily localized to the trans- Golgi network (TGN) and to recycle through endosomes and the plasma membrane. CPD is a type I protein with a 60- residue cytoplasmic tail, and so to validate this interaction, we used a GST fusion to the tail for affinity chroma- tography from 293T cell lysates. This showed that TBC1D23 has a strong preference for the tail of CPD over that of a control protein— the cation- independent mannose 6- phosphate receptor (CIMPR), another abundant protein that recycles between endosomes and the TGN (Fig. 1C). The interaction between TBC1D23 and CPD could be recapitulated with proteins expressed in Escherichia coli, demon- strating that it is direct (Fig. 1D), and use of truncated forms con- firmed that it is the C- terminal domain of TBC1D23 that binds to the CPD tail (Fig.  1E). Together, these results show that the cytoplasmic tail of CPD can form a stable complex with residues 559 to 684 of TBC1D23, the C- terminal domain that mediates vesi- cle capture. TBC1D23 is required for the normal trafficking of CPD To test whether TBC1D23 is required for the trafficking of CPD, we used CRISPR- Cas9 to remove the TBC1D23 gene from both human embryonic kidney (HEK) 293 cells and the rat insulinoma line INS- 1 (Fig. 2A and fig. S1A). CPD cycles between endosomes and the Golgi, and for some such proteins, it has been found that perturbation of retrieval from endosomes results in their destabili- zation (12, 21, 23). In both knockout cell lines, the steady- state level of CPD was reduced, and this could be rescued by expression of TBC1D23–green fluorescent protein (GFP) from a stably transfected gene under the control of a cumate- inducible promotor (Fig. 2, A and B). In rat INS- 1 cells, it was possible to detect CPD by immuno- fluorescence, and the levels of the protein in the TGN were reduced in the absence of TBC1D23 (Fig. 2C). Again, this phenotype was rescued by expression of TBC1D23- GFP but not a form lacking the C- terminal domain (Fig. 2, D and E, and fig. S1, B to D). The C- terminal domain of TBC1D23 has been previously reported to bind to the FAM21 subunit of the WASH complex (12, 13). How- ever, removal of FAM21 from HEK- 293 cells did not affect the steady- state levels of CPD (Fig. 2F). This indicates that the role of TBC1D23 in CPD traffic does not depend on the presence of FAM21. In contrast, deletion of the two golgins that bind TBC1D23 did reduce overall CPD levels as expected (fig. S1E). Together, these results show that TBC1D23 is required to maintain steady- state levels of CPD in the TGN and that performing this function involves an activity of TBC1D23 that is distinct from its ability to bind FAM21. Conserved residues in CPD mediate its interaction with TBC1D23 To gain more understanding of the interaction between TBC1D23 and CPD, we mapped the region of the CPD tail that is required for binding and found that a C- terminal 16- residue region is necessary and sufficient (Fig. 3, A and B). The residues Leu1375 and Tyr1376 are particularly important, with upstream conserved acidic residues also contributing which becomes more apparent when several are mutated (Fig.  3, B and C). Nuclear magnetic resonance (NMR) spectroscopy of the CPD tail showed that the presence of TBC1D23 affected the residues in the C- terminal region around the acidic threonine- leucine- tyrosine (TLY) motif, confirming it binding to this region in solution (fig. S2). TBC1D23 binds to the cytoplasmic domain of multiple cargoes of endosome- derived vesicles The finding that TBC1D23 can bind directly to the cytoplasmic tail of CPD, a protein that is known to recycle from endosomes back to the Golgi, raised the possibility that this interaction could mediate the capture of endosome- derived vesicles by TBC1D23. However, the capture of these vesicles by TBC1D23 ectopically relocated to mitochondria still occurred in cells from which the CPD gene had been deleted, indicating that binding to CPD is not sufficient to account for vesicle capture (fig. S3, A to C). To find further factors that might contribute to vesicle capture in addition to CPD, we used GST- TBC1D23 (559 to 684) for affinity chromatography of 293T cell lysates. As expected, we identified CPD and subunits of the WASH complex among the interacting proteins (Fig.  3D). Also enriched was the SNARE syntaxin- 16, a known cargo of endosome- to- Golgi carriers, along with one of its interacting partners, the Sec1/Munc18- family protein VPS45 (24). We confirmed that syn- taxin- 16 is present in vesicles captured by either TBC1D23 or golgin- 97 when they are relocated to mitochondria in HeLa cells (fig. S3, D and E). The syntaxin- 16 cytoplasmic domain bound directly to the TBC1D23 C- terminal domain in vitro, and residues 151 to 225 of syntaxin- 16 are both necessary and sufficient for the interaction (Fig. 3E). CPD binds to TBC1D23 via the sequence EEETLY, and syntaxin- 16 contains a similar sequence: 209DDNTLY in the ex- posed linker between the SNARE domain and the Habc domain (Fig. 3F). Mutating the TLY sequence in syntaxin- 16 to AAA abol- ished its binding to TBC1D23 (Fig. 3G). While the steady- state lev- els of syntaxin- 16 were not detectably altered in ∆TBC1D23 cells (fig. S3F), the Golgi- localized pool of the protein was reproducibly reduced, and this could be rescued by expression of TBC1D23- GFP (Fig.  3, H and I, and fig.  S3G). Together, our results show that TBC1D23 binds to the cytoplasmic tail of at least two proteins in the vesicles that it captures and thus raise the possibility that this capture could occur via direct and specific interaction with a subset of vesicle cargoes. X- ray crystallography of a complex between TBC1D23 and syntaxin- 16 illuminates their interaction To understand how the C- terminal domain of TBC1D23 recognizes the tails of vesicle cargo, we used a combination of x- ray crystallog- raphy and biochemistry. Peptides corresponding to the relevant regions of syntaxin- 16 and CPD showed robust enthalpic- dominated binding to the C- terminal domain by isothermal titration calorim- etry (ITC) with affinities of ~1.4 and 10.6 μM, respectively, and a stoichiometry of 1:1 (Fig.  4A). Crystallization of the C- terminal domain in the presence of the syntaxin- 16 peptide yielded crystals that diffracted to 3.3- Å resolution and were solved by molecular replacement using the previously published apo structure (6JM5) in space group P6222 (table  S1). Four C- terminal domains were present within the asymmetric unit (fig.  S4, A to C). Two of the domains dimerized via strand exchange in the same manner as that seen in the peptide- free structure with reciprocal C- terminal tails (VLDALES) inserting between strand 5 and the helix 2 of the other domain but with no sign of the peptide (13). In the third domain, D ow nloaded from https://w w w .science.org on A pril 22, 2024 C-6 Publications from this thesis Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 4 of 15 the electron density was unclear, but for the final domain, density corresponding to a single syntaxin- 16 peptide was visible (fig. S4D). This peptide density was in the same position that was occupied by the cross- dimerizing, C- terminal tail in the first two molecules of the asymmetric unit. Given the above results, we generated a version of the C- terminal domain lacking the last seven residues (VLDALES). Crystals of this truncated domain in the presence of the syntaxin- 16 peptide diffracted to higher resolution (2.2 Å) and were solved in space group P6522 again by molecular replacement with unliganded 6JM5 (table S1). The _70 _150 _100 _50 0.0 0.5 1.0 1.5 C PD in te ns ity (r at io to W T) WT TB1D 23 (+) C um ate TBC1D23 Merge CPD α-Tubulin WT ∆TBC1D 23 (−) C um ate (+) C um ate WT ∆TBC1D 23 (−) C um ate (+) C um ate W T ∆T BC 1D 23 CPD GM130 CPD GM130 TBC1D23-GFP ∆T BC 1D 23 293 293 kDa _70 _150 _50 WT TB1D 23 (+) C um ate 0.0 0.5 1.0 1.5 C PD in te ns ity (r at io to W T) A C D TBC1D23-GFP TBC1D23-GFP TBC1D23 CPD α-Tubulin α-Tubulin WT ∆TBC1D 23 (−) (+) (−) (+) GFP TBC1D23-GFP 1– 55 8 Full Le ng th kDa E WT TB1D 23 WT (+ ) 1– 55 9 ( +) 0.0 0.5 1.0 C PD in te ns ity (r at io to W T) ∆TBC1D 23 TBC1D23 CIMPR FAM21A/C Res cu e ( +) Cum ate WT ∆FA M21 A/C ∆TBC1D 23 , ∆ FA M21 A/C CPD WASHC1 WT TB1D 23 ; FA M21 A/C FA M21 A/C 0.0 0.5 1.0 1.5 C PD in te ns ity (r at io to W T) F INS-1 INS-1 TBC1D23-GFP kDa _70 _150 _50 B 0.002 0.03 n.s. 0.001 Merge Fig. 2. Normal CPD trafficking requires the C- terminal domain of TBC1D23 but does not require the WASH complex. (A) Immunoblots comparing TBC1D23, CPD, and α- tubulin in whole- cell lysates of HEK- 293 and Ins- 1 cells that were either wild type (WT), ∆TBC1D23, or the latter rescued with TBC1D23- GFP stably expressed under a cumate- inducible promoter as indicated. (B) The intensities of the bands in (A) were quantified and normalized to wild type (shown are mean and SEM, P values calculated using two- tailed unpaired t test); numbers of biological replicates are 6 (HEK- 293) or 3 (INS- 1). (C) Confocal micrographs of wild- type or ∆TBC1D23 INS- 1 cells stained for endogenous CPD and GM130 (Golgi marker). Scale bars, 10 μm. Representative of three repeats. (D) Confocal micrographs of ∆TBC1D23 INS- 1 cells transiently expressing mouse TBC1D23- GFP and labeled for the GFP and endogenous CPD and GM130. Scale bar, 10 μm. Representative of three repeats. (E) Immunoblots of the indicated HEK- 293 cell lines as in (A) but with the addition of HEK- 293 cells expressing TBC1D23- GFP (1 to 558) and with quantification as in (B) showing mean and SEM of two independent replicates. Camera- based imaging of enhanced chemiluminescence from the secondary antibodies can result in very low background outside of the relevant bands. (F) Immunoblots of whole- cell lysates from wild- type HEK- 293 cells (WT) or the indicated mutants, some of which expressed TBC1D23- GFP from a cumate- inducible promotor. CPD levels were quantified as in (B) for blots from three independent clones of ∆FAM21A/C and ∆TBC1D23;∆FAM21A/C, with mean values and SEM shown and P values calculated using two- tailed unpaired Welch’s t test. CPD is destabilized by removal of TBC1D23 but not by removal of FAM21A/C. Source data for (B), (E), and (F) are in data S2. n.s., not significant. D ow nloaded from https://w w w .science.org on A pril 22, 2024 A.6 CIMR crystallization screens C-7 Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 5 of 15 C Red: Residues required for binding CPD 1321 WT Inp ut (3% ) 13 21 –1 35 0 13 48 –1 38 0 13 21 –1 37 8 13 21 –1 37 6 13 21 –1 37 4 13 21 –1 37 2 13 21 –1 37 0 13 21 –1 36 8 13 21 –1 36 6 GST-CPD TB C 1D 23 -H is 6 WTT13 74 A L1 37 5A Y13 76 A S13 77 A S13 78 A K13 79 A H13 80 A GST-CPD (1321–1380) TB C 1D 23 -H is 6 Transmembrane 13 61 –1 38 0 13 70 –1 38 0 13 74 –1 38 0 kDa _70 _35 _25 _40 _50 kDa _70 _35 _25 E13 73 A E13 72 A E13 71 A T13 70 A D13 69 A T13 68 A D13 66 A E13 67 A EEE->A AA _40 _50 WT 1361 1380 −10 −5 0 5 10 2 4 6 Log2 fold change GST vs. GST-TBC1D23 (559–684) −L og (P v el ue ) VPS45 STX16 CPD FAM21A WASHC5 WASHC4 WASH1 FAM21C FKBP15 CCDC53 WASH complex Others A B D F STX16 (1 –2 81 ) STX16 (1 –1 50 ) STX16 (1 51 –2 25 ) STX16 (1 –2 81 ) STX16 (1 –1 50 ) STX16 (1 51 –2 25 ) -MBP-His6G ST-TBC 1D 23 (559–684) 40_ kDa 50_ 70_ 100_ Inputs (3%) E G CPD Inputs (3%) GST GST-TBC1D23 (559–684) _40 kDa _35 _25 _70 _50 His6-MPB: STX 16 STX 16 (T LY -> AAA) H6-MBP-CPD (1321–1380) STX16-MBP-H6 (1–281) CPD STX 16 STX 16 (T LY -> AAA) CPD STX 16 STX 16 (T LY -> AAA) W T ∆T BC 1D 23 GM130STX16 R es cu e (+ )c um at e H I GFP booster TBC1D23-GFP 0.0 0.5 1.0 1.5 0.00 0.05 0.10 0.15 Bin center R el at iv e fre qu en cy (f ra ct io ns ) WT ∆TBC1D23 Rescue (+) cumate WT ∆TBC1D 23 Res cu e ( +) cu mate 0.0 0.5 1.0 1.5 Pi xe l in te ns ity S TX 16 /G M 13 0 ns TMDSNARE motifHa Hb HbNStx16 206..MDDGDDNTLYHR..215 Fig. 3. TBC1D23 binds to C- terminal conserved residues of the cytoplasmic tail of CPD. (A) The cytoplasmic tail of human CPD. (B and C) Coomassie- stained gels of the eluates from immobilized GST- CPD truncations or mutants incubated with a bacterial lysate containing TBC1D23- His6. Each representative of two repeats. (D) Volcano plot of the MS analy- sis from affinity chromatography of 293T cell lysates using bacterially expressed GST- TBC1D23 (559 to 684) or GST alone. Shown are mean spectral intensities of bound proteins from three independent experiments. Values are in data S1. (E) Coomassie- stained gel showing that the eluates from immobilized GST- TBC1D23 (559 to 684) incubated with lysates from bacteria expressing the indicated fragments of syntaxin- 16 (STX16)–MBP–His6 (numbering as in UniProt O14662- 2). Representative of two repeats. (F) Cartoon of syntaxin- 16 show- ing the location of the key structural features and the acidic TLY motif related to that of CPD. (G) Coomassie- stained gel showing that the eluates from immobilized GST- TBC1D23 (559 to 684) incubated with lysates from bacteria expressing His6- MBP- CPD (1321 to 1380), syntaxin- 16 (1 to 281)–MBP–His6 or syntaxin- 16 (1- 281 with T191A, L192A, and Y193A)– MBP–His6. Representative of two repeats. (H) Confocal micrographs of the indicated INS- 1 cells (wild type, ∆TBC1D23, and ∆TBC1D23 stably expressing TBC1D23- GFP under a cumate promoter in the presence of cumate for 24 to 36 hours). Cells stained for GFP and endogenous syntaxin- 16 and GM130. (I) Scatter plot (left) showing the ratio of the Golgi fluores- cence intensity of syntaxin- 16 over the GM130- positive regions (Golgi). The horizontal bar is the mean. For wild type, n = 530, for ∆TBC1D23, n = 628, and for the stable rescue, n = 725. ****P < 0.0001 [ordinary one- way analysis of variance (ANOVA) followed by a Sidak’s multiple comparison tests with a single pooled variance]. Ratios were also plotted as frequency distributions with a bin width of 0.05 (right). Values are in data S2. D ow nloaded from https://w w w .science.org on A pril 22, 2024 C-8 Publications from this thesis Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 6 of 15 asymmetric unit contained two copies of the domain with both having density corresponding unambiguously to the syntaxin- 16 peptide in the previously proposed binding site (fig. S4E). The pre- viously observed strand- exchanged dimer was absent with the homodimeric C- terminal tail–mediated interface being replaced by the peptide. Thus, the structure elucidates how the syntaxin- 16 peptide binds to the C- terminal domain of TBC1D23 (Fig.  4B). The peptide extends TBC1D23’s β sheet by adopting a twisted β augmentation mode of interaction with strand 5 and buries ~670 Å2 of the C- terminal domain’s solvent accessible surface area, a simi- lar sized interface to those seen previously for other trafficking motif–binding proteins that have a similar KD of interaction (25, 26). The Thr212 in the acidic TLY motif of syntaxin- 16 packs its methyl group against the alkyl chain of Lys628 that orients the hydroxyl group toward the solvent to stabilize a somewhat strained backbone conformation (Fig. 4C). This allows the side chains of the leucine and tyrosine residues to be splayed apart and simulta- neously buried into neighboring hydrophobic pockets in the domain (Fig. 4D). As a result, the acidic residues Asp209 and Asp210 of syn- taxin- 16 are in close proximity to three basic residues (Lys632 to Lys634) in the C- terminal domain of TBC1D23, which would allow a direct electrostatic interaction (Fig. 4E). Mutations in the TBC1D23 C- terminal domain and the syntaxin- 16 peptide perturb their interaction The structure indicates that recognition of the acidic cluster and the TLY triad are the critical factors in driving peptide binding and specificity. To validate this, a series of point mutations was designed in the C- terminal domain and initially assessed for folding by yield and circular dichroism. Two of the residues that form parts of the hydrophobic pockets that bind the leucine and tyrosine of the pep- tide (Ile629 and Ile639) are also part of the domain’s core hydrophobic residues, and their mutation caused misfolding (I629S and I639S). The remaining mutants were assayed for binding to the syntaxin- 16 peptide by ITC (Fig. 5, A to C). Lys672 of TBC1D23 hydrogen bonds via its amine group to the carboxyl peptide bond of Lys213 in the TLY in syntaxin- 16 (Fig.  4C); mutation to alanine (K672A) weakened binding ~20 fold (KD  ~  30 μm). Mutation of Val626, which packs against the aromatic ring of reside Tyr214, to aspartate (V626D) abolishes peptide binding. The binding pocket in TBC1D23 for Asp210 Lys634 Lys633 Lys632 Asp209 Lys672 Tyr214 Leu213 HydrophilicHydrophobic Thr121 BA C ED DDNTLYHRGFTED EEETLYSSKH CPDSyntaxin-16 Fig. 4. Structure of TBC1D23 C- terminal domain bound to a syntaxin- 16 peptide. (A) ITC trace and fitted curve of the indicated peptides from syntaxin- 16 (left) and CPD (right) binding to TBC1D23 C- terminal domain. Source data are in data S2. (B) Structure of the TBC123 C- terminal domain (sky blue) and syntaxin- 16 (209 to 217) (gold) with the corresponding electron density map of the latter. (C) Close- up of the hydrogen bonding network between syntaxin- 16 (gold) and TBC1D23 (sky blue). Lys672 hydrogen bonds to the backbone of syntaxin- 16. Hydrogen bonds are in green. (D) Hydrophobicity surface plot of TBC1D23 showing the packing of residues of syntaxin- 16 (gold) into two hydrophobic pockets. (E) Close- up of Asp209 and Asp210 in syntaxin- 16 (gold) interacting with Lys632 to Lys634 in TBC1D23 (sky blue). D ow nloaded from https://w w w .science.org on A pril 22, 2024 A.6 CIMR crystallization screens C-9 Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 7 of 15 Tyr214 was obstructed by mutating one of its lining residues (I675W), and, again, this reduced binding to below detectable levels (KD > 300 μM). Last, mutation to alanine of Lys632, Lys633 and Lys634, which are adjacent to the TLY- binding hydrophobic grove, also abolished peptide binding, indicating that their interaction with the acidic cluster in syntaxin- 16 is important. Thus, mutations that either re- move interactions with the peptide or protrude into the binding pocket strongly affect peptide binding, confirming that the interac- tions seen in the crystal also occur in solution. Mutations were also made in the syntaxin- 16 peptide to further validate the interactions observed in the crystal structure (Fig. 5D). ITC assays confirmed a role in binding for all three of the residues in the acidic TLY motif and for the two adjacent acidic residues, with Arg216 making a smaller contribution (Fig.  5, C to E). The acidic TLY motif is absolutely conserved between syntaxin- 16 and CPD, and so to determine whether conservative changes in the motif can be tolerated, we tested variants of these three residues (fig. S4F). Mu- tation of Thr212, which holds the leucine and tyrosine in a strained position to allow engagement with their pockets, to serine (T212S) reduced binding ~20- fold to KD ~ 30 μM. Changing Lys213 to the smaller alanine (L213A) or the larger phenylalanine (L213F) both caused a reduction of binding of >100- fold, consistent with the side- chain fitting into a size- specific hydrophobic pocket in the structure, i.e., a medium- sized hydrophobic residue is required at this posi- tion. Last, Tyr214 is more tolerant of conservative mutation with Y214W and Y214F peptides having KD values of ~7 and ~15 μM, respectively (fig. S4F). Further binding partners for TBC1D23 are identified by use of the C- terminal domain–binding motif Although both syntaxin- 16 and CPD bind to the C- terminal do- main through a closely related motif, the only functional property that they share is the fact that they are both in vesicles that recycle between endosomes and Golgi and can be captured by TBC1D23. We thus wondered whether there were further proteins in these vesicles that also have an acidic TLY motif that can bind TBC1D23. Peptide motifs that are recognized by proteins are rarely invariant, and so other versions of the motif are likely to be functional. From STX16/TBC1D23 KD (µM) Wild type 1.4 ± 0.1 TBC1D23 mutants KKK632-634AAA No binding V626D No binding K672A 27.5 ± 6.3 I675W No binding STX16 mutants DD209-210AA >200 T212A No binding L213S No binding Y214S No binding R216A 10.3 ± 0.1 Wild type K672A Wild type K672A Others Wild type DD209AA R216A Wild type DD209AA Others R216A B E A C D K672A V626D KKK632-634AAA I675W Y214A DD209-210AA T212A R216A L213S Fig. 5. Mutational analysis of the interaction between TBC1D23 and syntaxin- 16. (A) Schematic view of the mutations in TBC1D23 that were tested by ITC. (B) ITC of relative binding of syntaxin- 16 peptide to the TBC1D23 C- terminal domain mutants. Mutations are indicated by the color coding in (A), with wild type in black. (C) Summary of ITC results. Source data are in data S2. (D) Schematic view of the mutations in the syntaxin- 16 peptide that were tested by ITC. (E) ITC of relative binding of the mutant syntaxin- 16 peptides to the TBC1D23 C- terminal domain. Mutations are indicated by the color coding in (D), with wild type in black. D ow nloaded from https://w w w .science.org on A pril 22, 2024 C-10 Publications from this thesis Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 8 of 15 the above mutagenesis experiments and the nature of the interac- tions seen in the structure, we chose [DE]> = 3/5[TS][LIV][YFW] as being a loose but plausible definition of the core binding motif for the TBC1D23 C- terminal domain and used this to search the pre- dicted cytoplasmic tails of all membrane proteins in the human pro- teome. In addition, we looked for membrane proteins among the interactors reported for TBC1D23 in the BioPlex protein interaction screen (27). The latter approach identified four membrane proteins, three of which are reported to be in the endosome/Golgi system and one of unknown location. The former approach identified 19 mem- brane proteins of which 10 are reported to be endosomal or Golgi, with two proteins found by both approaches (table  S2). We thus tested the ability of the TBC1D23 C- terminal domain to bind to peptides corresponding to the TLY- like region from several of these proteins (Fig. 6, A to C, and fig. S5). NPDC1 is a largely uncharacterized protein suggested to play a role in neuropeptide secretion and dense core vesicle traffic (28, 29), and the peptide from its tail has an affinity for TBC1D23 similar to that of syntaxin- 16 at ~3 μM (fig. S5). KIAA0319 is a neuronal pro- tein linked to dyslexia, and although its function remains unknown, it has been reported to recycle from the surface back to the Golgi region (30). The peptide from KIAA0319 showed weak but detect- able binding with KD ~ 35 μM (fig. S5). KIAA0319 is only expressed in neuronal cells but KIAA0319L, its non- neuronal paralog, is enriched in vesicles captured by golgin- 97 and its localization is per- turbed in human cells lines and zebrafish lacking TBC1D23 (31, 32). The presence of the TBC1D23- binding motif in these neuronal pro- teins may be relevant to human mutations in TBC1D23 causing defects in neurodevelopment (33, 34). FAM174A and its paralogs FAM174B and FAM174C are small- membrane proteins of unknown function that have been reported to be in the Golgi (35). All have a TLY- like motif and the peptide from FAM174C bound with an affin- ity of 16 μM. Last, TVP23B/C is a small polytopic membrane pro- tein known to recycle between endosomes and Golgi and to be enriched in vesicles captured by golgin- 97 (32, 36), and its 34- residue cytoplasmic tail bound to the TBC1D23 C- terminal domain with this interaction disrupted by mutating the three lysine residues in TBC1D23 that we have shown to be required for binding CPD and syntaxin- 16 (Fig. 6C). In contrast, no binding was seen with pep- tides from CIMPR, which also traffics between Golgi and endo- somes and has an acidic region in its tail but no TLY- like motif, and SLAMF7 that has a TLY- like motif but a smaller acidic cluster that contains a proline (Fig.  6B). Thus, multiple proteins that recycle between endosomes and Golgi have sequences in their cytoplasmic tails that can bind directly to the C- terminal domain of TBC1D23. Residues in TBC1D23 that are required for binding the acidic TLY motif are also required for vesicle capture in vivo The presence in endosome- to- Golgi vesicles of diverse proteins with a TLY- like motif in their cytoplasmic tails raises the possibility that TBC1D23 can capture these vesicles by binding to the tails protrud- ing from the vesicle. TBC1D23 is sufficient to capture vesicles when relocated to mitochondria (12), and so we tested the effect on this capture of mutations that disrupt peptide binding in vitro. Relocation of vesicles by the constructs was quantified with proximity biotinyl- ation (BioID) using a promiscuous biotin ligase fused to TBC1D23 along with a mitochondrial targeting signal (Fig. 7A). When wild- type TBC1D23 was expressed on mitochondria, vesicle cargo was efficiently biotinylated as expected (Fig. 7B). However, mutation of the residues that disrupt peptide binding in vitro resulted in a near complete loss of biotinylation of vesicle cargo proteins, indicating that vesicle capture was also disrupted. This is not simply due to the C A B Cargo Sequence KD (µM) STX16 1.4 ± 0.2 CPD 10.6 ± 3.5 NPDC1 2.9 ± 0.5 KIAA0319 35.2 ± 11.7 FAM174C 15.7 ± 3.4 CIMPR No binding SLAMF7 No binding TMD TMD TMD TMD NPDC1 TVP23B/C KIAA0319 281..EDGDFTVYEC..290 9..DTEDVSLFDA..18 118..DSDEETVFES..127 1034..DSDQDTIFSR..1043 325 276 132 1072 1 1 1 1 SP SP SP FAM174C Inp ut (3% ) K63 2E ,K 63 3E ,K 63 4E W T GST I63 9A T6 40 A GST-TBC1D23 (559–684) His6-MBP- _TVP23C (1–34) _GST-TBC1D23 (559–684) kDa 50_ 40_ 35_ 25_ 70_ 100_ Fig. 6. Identification of membrane proteins containing a TBC1D23- binding acidic TLY motif. (A) Schematic of four human proteins that have an acidic TLY motif in their cytoplasmic tail and that have been reported to recycle from endosomes to the Golgi or be localized to the Golgi (SP, signal peptide; TMD transmembrane domain). (B) Binding affinities of the indicated peptides containing the acidic TLY motifs from the proteins shown in (A), as determined by ITC (fig. S5; source data are in data S2.). (C) Coomassie- stained gel showing the binding of the N- terminal cytoplasmic tail of TVP23C [His6- MBP- TVP23C (1 to 34)] to beads coated with the C- terminal domain of TBC1D23 fused to GST. Mutations in the acidic TLY motif disrupt the interaction. Representative of two repeats. D ow nloaded from https://w w w .science.org on A pril 22, 2024 A.6 CIMR crystallization screens C-11 Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 9 of 15 vesicle being docked by a different mechanism, as immunofluores- cence demonstrated that the accumulation of vesicle cargo proteins on the TBC1D23- coated mitochondria was greatly reduced (Fig. 7, C and D). Together, these results demonstrate that the same part of TBC1D23 that binds to the tails of vesicle cargo in vitro is required for vesicle capture in vivo, consistent with a model in which TBC1D23 can capture the incoming vesicle by recognizing its cargo proteins. DISCUSSION The C- terminal domain of TBC1D23 was known to be sufficient to capture one or more of the classes of carrier that mediate transport of membrane proteins from endosomes to Golgi (12, 13). Our search for binding partners of this domain has now revealed that it can bind directly to the cytoplasmic tails of several of the cargo proteins that are present in these vesicles via a conserved acidic TLY motif. BirA* biotin ligase Mitochondrion Biotinoyl-AMP Empty A B MAOHABirA* TBC1D23 (514–684) TBC1D23 (514–684)–BirA*–HA–MAO V62 9D I67 5W KKKtoE EE I62 9W W T 50 35 40 C Syntaxin-16 HA tag (TBC1D23) Streptavidin * S treptavidin bead eluate Syntaxin-16 (vesicle cargo) GM130 (Golgi) TBC1D23-HA-MAO (mitochondria) Merge Wild type V626D KKKtoEEE * * * D -ve co ntr ol Wild ty pe V62 9D KKKtoE EE 0 0.5 1.0 Sy nt ax in -1 6 m ito /to ta l TBC1D23-HA-MAO (mitochondria) **** **** **** Fig. 7. Residues in TBC1D23 required for peptide binding in vitro are also required for vesicle capture in vivo. (A) Ectopic relocation and biotinylation: a chimera comprising the TBC1D23 C- terminal domain attached to the BirA* promiscuous biotin ligase and the mitochondrial targeting signal from monoamine oxidase (MAO). (B) Immunoblots of streptavidin precipitations from whole- cell lysates of 293T cells transfected with the indicated variants of the mitochondrially targeted C- terminal domain of TBC1D23 fused to BirA*. Mutation of key residues does not affect total biotinylation or the stability of the chimera but greatly reduces biotinylation of syntaxin- 16 indicating loss of vesicle capture. The lower band (asterisk) indicates some clipping of the chimera apparently between the TBC1D23 part and BirA*. Representative of three repeats. (C) Immunofluorescence of cells expressing the indicated forms of mitochondrial full- length TBC1D23 and immunolabeled for the TBC1D23 chimera [hemagglutinin (HA) tag], syntaxin- 16, and the Golgi marker GM130. TBC1D23 is sufficient to cause the accumulation of vesicles at this ectopic location, and, hence, syntaxin- 16 accumulates on mitochondria coated in wild- type TBC1D23 (arrows), rather than being predominantly in the Golgi in untransfected cells (asterisks). The mutations in TBC1D23 that disrupt peptide binding in vitro greatly reduce mitochondrial accu- mulation of syntaxin- 16. Representative of three repeats. Scale bars, 10 μm. (D) Quantitation of the degree of mitochondrial relocation of syntaxin- 16 in (C). The relocation induced by the wild- type protein, and the reductions in this relocation caused by the mutations are statistically significant (****P < 0.0001, unpaired, two- tailed t tests). D ow nloaded from https://w w w .science.org on A pril 22, 2024 C-12 Publications from this thesis Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 10 of 15 This interaction provides a mechanism by which TBC1D23 dis- played on the ends of golgin- 97 and golgin- 245 can recognize spe- cific vesicles and thus tether them to the trans- Golgi before subsequent SNARE engagement and vesicle fusion. For both CPD and syntaxin- 16, their predicted structures indicate that the acidic TLY motif is likely to be located at some distance from the trans- membrane domain and, hence, the vesicle membrane. The unstruc- tured tail of CPD or the α- helical bundle of syntaxin- 16 are predicted to place the motif ~15 nm away from the membrane which would allow recognition to be unimpeded by the vesicle surface. The direct binding of a tethering factor to the cargo carried by a vesicle has not been previously proposed as a mechanism for vesicle capture, but it would certainly ensure that only a specific set of vesi- cles is recognized (37). Like all such tethering interactions, the mechanism requires that the tether binds only to the vesicle and not to the organelle from which the cargo- laden vesicle budded. It is possible to imagine several ways in which this could be achieved. TBC1D23 is primarily localized at the Golgi as determined by both immunofluorescence and organelle fractionation, indicating that the interaction of its N- terminal domain with golgin- 97 and golgin- 245 is the dominant factor in determining its localization in  vivo (12, 32, 33). In addition, the golgins are homodimers, and so each golgin is likely to display two copies of TBC1D23 that would in- crease the avidity of the interaction with the tails in the vesicle over that of a monomer of TBC1D23 that was free in the cytoplasm. It may be that interaction between TBC1D23 and the cargo proteins is only strong enough to capture a vesicle when an array of TBC1D23 presented on multiple golgins can interact with multiple TLY- containing proteins that are concentrated in a single vesicle, and, hence, the avidity of the interaction is increased even further. Another mechanism that could account for TBC1D23 only rec- ognizing cargo proteins that are present in vesicles is that the acidic TLY motif in the cargo proteins could be masked in some way when the cargo is in endosomes and other compartments. This could occur if the coat proteins that sort them into the retrograde pathway rapidly sequester them into forming carriers and if the interaction with the coat prevents recognition of the acidic TLY motif by TBC1D23 until the carrier has budded and uncoated. The identity of the coat that makes the carriers captured by TBC1D23 is as yet unresolved as there are multiple routes back from endosomes to the Golgi, with both clathrin coats with the activating protein 1 (AP- 1) adaptor and sorting nexins with retromer proposed to be involved in generating carriers (19, 38). Nonetheless, note that the acidic TLY motif includes a stretch of acidic residues, and acidic clusters can act as sorting signals for packaging into clathrin/AP- 1–coated carriers (39, 40). Both CPD and syntaxin- 16 are known to be enriched in AP- 1–dependent clathrin–coated vesicles, although whether these vesicles are delivering cargo to or from the trans- Golgi remains to be resolved (41). There are known to be multiple classes of vesicle arriving at the Golgi from endosomes, and although TBC1D23 appears to respon- sible for the capture of at least some of these, the mechanism by which the others are captured is unknown. The golgin GCC88 cap- tures at least one such set of vesicles although it does not bind TBC1D23, but how this is achieved is still unclear (9, 12). This multiplicity of endosome- to- Golgi routes with only some carriers being captured by TBC1D23 is consistent with human mutation of TBC1D23 resulting primarily in neurodevelopmental defects despite being widely expressed in most tissues (33, 34). It may be that when TBC1D23- dependent tethering is lost, other capture mechanisms can partially compensate, or some cargo proteins can return to the Golgi by some of the alternative routes. The enrichment of a particular set of proteins in each specific type of transport vesicle is only one type of distinguishing feature that would allow the specific recognition necessary to ensure the fidelity of intracellular traffic. For most trafficking steps, vesicle rec- ognition remains poorly understood. In some cases, Arf or Rab GT- Pases are believed to play a role, although they are also likely to be present on donor organelles resulting in the same issues of vesicle versus donor organelle identity (4, 37, 42). Our finding that a tether- ing factor can recognize a set of cargo proteins in the carriers that it captures not only sheds light on endosome- to- Golgi traffic but could also illuminate other membrane trafficking routes where the mechanism of vesicle recognition remains to be found. MATERIALS AND METHODS Plasmids Details of the plasmids used in this report, together with the cloning methods used to generate them, are provided in data S3. Note that the cDNA for TBC1D23 used throughout is from mouse (Q8K0F1) with 684 residues [corresponding to the 684- residue human isoform (UniProt Q9NU8- 2)], and residues are numbered accordingly. Peptides for binding partners are all based on the human proteins. Antibodies Full lists of primary and secondary antibodies used for Western blotting and immunofluorescence are provided in data S3. Mammalian cell culture The full list of cell lines used are in data S3. HeLa (American Type Culture Collection), HEK293T (American Type Culture Collection, CRL- 3216; referred to throughout the paper as 293T), and HEK- 293 Flp- In T- REx 293 cells stably expressing Cas9 (referred to through- out the paper as HEK- 293) were cultured in Dulbecco’s modified Eagle’s medium GlutaMAX (Gibco) supplemented with 10% fetal bovine serum (FBS) and penicillin/streptomycin at 37°C and 5% CO2. The insulinoma INS- 1–derived 832/13 rat cell line was ob- tained from C. Newgard (Duke University School of Medicine) via M. Ailion (University of Washington, Seattle) (referred to through- out the paper as INS- 1) (43). INS- 1 cells were grown with RPMI- 1640 GlutaMAX (Gibco), supplemented with 10% FBS, 1 mM sodium pyruvate, 10 mM Hepes, and penicillin/streptomycin at 37°C and 5% CO2. For transient transections, unless noted, we used polyethyleni- mine (PEI; Polyscience, 24765) dissolved in phosphate- buffered saline (PBS) to 1 mg/ml. The ratio of PEI (in microliters) to DNA (in micrograms) used was 3:1. PEI was dissolved in Opti- MEM (Gibco) and incubated at room temperature for 5 min. DNA was added and incubated for another 20 min at room temperature before drop- wise addition onto cells that had been seeded the day before. Cells were free of mycoplasma as determined by routine testing using MycoAlert (Lonza). Generation of knockout cells using CRISPR- Cas9 Precise information about the cell lines generated and guide RNA (gRNA) sequences is provided in data S3. ∆TBC1D23 (in HEK293 Flp- In T- REx 293 cell line stably expressing Cas9, HEK- 293) cells were generated as follows: cells grown to ~70% confluence in six- well D ow nloaded from https://w w w .science.org on A pril 22, 2024 A.6 CIMR crystallization screens C-13 Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 11 of 15 plates and cotransfected with a gRNA to TBC1D23 (12) and pmaxGFP using Fugene6 (Promega) according to the manufacturer’s instruc- tions. Forty- eight hours later, cells were trypsinized, and single GFP- positive cells were sorted into each well of 96- well plates using a cell sorter. Single colonies that grew were expanded and analyzed by immunoblotting. All other knockouts made in HEK- 293 and INS- 1 cells were gen- erated as follows: gRNA with high- efficiency and high- specificity scores was chosen using the UCSC genome browser and CRISPOR (44). To improve the knockout efficiency, each gene was targeted with two gRNA, one targeting an early exon and the second target- ing a late exon, so a large part of the coding region could be removed (data S3). All gRNAs were cloned onto the pX459 vector, which also expresses Cas9 and a puromycin resistance gene (45). Cells were grown to ~70% confluence in six- well plates and transfected with a mixture of 6 μl of PEI and an equal amount of each vector coding for gRNAs for a total of 2 μg of DNA in 100 μl of Opti- MEM. Forty- eight hours later, the cells were trypsinized and replated in complete medium containing puromycin (1.5 μg/ml). Forty- eight to 72 hours later, cells were diluted to one cell per two wells in three 96- well plates and grown in complete medium. Clones that grew were ex- panded and screened by immunoblotting or immunofluorescence. When possible, several clones were assayed (data S3). For rescue of knockout cell lines, the relevant genes were cloned into a piggyBac vector containing a puromycin resistance gene and the protein of interest expressed under a cumate- inducible promoter. Cells grown to ~70% confluence in 10- cm plates were cotransfected with the plasmid coding for the rescue constructs and the plasmid coding the piggyBac transposase at 5:1 molar ratio using 6 μg of DNA and 18 μl of PEI in 1 ml of Opti- MEM. Forty- eight hours after transfection, cells were trypsinized and replated in selection medium containing puromycin (1.5 μg/ml). The medium was changed every 48 to 72 hours while keeping the puromycin selection until conflu- ence was reached (usually 10 days), and expression of the rescue con- struct was verified by Western blotting and immunofluorescence. Rescue lines were maintained as a polyclonal population. Immunofluorescence Cells were transfected in 24- well plates (300 ng of DNA and 1 μl of PEI in 50 μl Opti- MEM) or in 6- well plates (1 μg of DNA and 3 μl of PEI in 100 μl of Opti- MEM). The next day, cells were dissociated using trypsin and seeded onto polytetrafluoroethylene- coated mul- tiwell slides (Hendley- Essex). About 36 to 48 hours after transfec- tion, cells were fixed with 4% paraformaldehyde in PBS (10 min), permeabilized in 0.5% (v/v) of Triton X- 100 in PBS (5 min), and blocked for 1 hour in PBS containing 20% FBS and 0.25% Tween 20. Primary and secondary antibodies were applied sequentially in blocking buffer for 1 hour at room temperature. After washing, cells were mounted using ProLong Gold antifade mountant (Thermo Fisher Scientific). Slides were imaged using a Leica TCS SP8 laser scanning confocal microscope and a 63× lens. Quantification of fluorescent micrographs For each condition in each independent replicate, three multichan- nel fluorescence micrographs at 1024 × 1024 resolution were ac- quired. The images were then automatically processed using a custom ImageJ macro to measure the intensity in all channels for regions of interest (ROIs) defined by segmenting the second channel (Golgi marker GM130). Briefly, the second channel was processed with a rolling ball and a median filter. A threshold based on the mean and SD of the image allowed the separation of the background and foreground, while a filter on minimum area removed spurious detection. ROIs were enlarged to include the nearby region. Last, the macro reports the area of the ROI and the mean and maximum value of each channel. For each condition, over 100 individual cells were quantified. For each ROI, the mean fluorescence intensity of the protein of interest was divided by the mean fluorescence inten- sity of the GM130 channel. The data of the ratio of mean fluores- cence intensities were then plotted as frequency distributions using GraphPad Prism 9 with a bin width of 0.05. For each marker, the experiment was repeated twice with similar results. Determination of protein levels by immunoblotting HEK- 293 or INS- 1 cells were seeded in 10- cm plates, and where appropriate, the medium was supplemented with 1× cumate (Sys- tem Biosciences, QM159A- 1) to induce expression of the rescue constructs. After 24 to 36 hours, cells were resuspended by scraping, washed twice with ice- cold PBS, and pelleted by a 5- min 500g spin at 4°C. Cells were resuspended in lysis buffer [50 mM tris (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% Triton X- 100, and protease inhibi- tor cocktail (cOmplete, Roche)], and incubated on ice for 5  min. Lysates were clarified at 17,000g for 5 min at 4°C, and the total pro- tein concentration was determined using the bicinchoninic acid assay (Thermo Fisher Scientific, 23227). Clarified lysates were then supplemented with 4× NuPAGE LDS sample buffer containing 100 mM dithiothreitol (DTT; Invitrogen, NP0007). Equal amount of total protein was separated by SDS–polyacrylamide gel electro- phoresis (PAGE) using precast gels (Invitrogen, XP04205) and analyzed by immunoblotting. Immunoblotting Protein samples in 1× NuPAGE LDS sample buffer containing 25 mM DTT were boiled at 95°C, loaded on to SDS- PAGE gels, and transferred to nitrocellulose membranes. Membranes were blocked in 5% (w/v) of milk in PBS- T [PBS with 0.1% (v/v) of Tween 20] for 1 hour, incubated overnight at 4°C with primary antibody in the same blocking solution, washed three times with PBS- T for 5 min, incubated with horseradish peroxidase (HRP)– conjugated secondary antibody in 5% (w/v) of milk in PBS- T for 1 hour and, washed three times with PBS- T for 5 min. Blots stained with HRP- conjugated secondary antibodies were imaged using a Bio- Rad ChemiDoc imager with SuperSignal West Pico PLUS (Thermo Fisher Scientific, 34577). The “gels” analysis tool in ImageJ was used for the quantification of blots, with raw data normalized to the wild- type band (data S2). In vitro binding assays using recombinant proteins Recombinant proteins for binding assays were expressed as follows: Plasmids were transformed into E. coli BL21–CodonPlus (DE3)– RIL (Agilent, 230245). From an overnight starter culture, cells were grown in 2× TY medium containing ampicillin [100 μg/ml; or kana- mycin (50 μg/ml) when appropriate] and chloramphenicol (34 μg/ml) at 37°C in a shaking incubator. When the culture reached optical density at 600 nm (OD600) = 0.6 to 0.8, the temperature was lowered to 16°C, and protein expression was induced with 100 μM isopropyl- β- d- 1- thiogalactopyranoside (IPTG), and incubated overnight. Bacteria cells were harvested by centrifugation at 4000g at 4°C for 15  min and were mechanically resuspended on ice in lysis buffer D ow nloaded from https://w w w .science.org on A pril 22, 2024 C-14 Publications from this thesis Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 12 of 15 containing 50 mM tris (pH 7.4), 150 mM NaCl, 1 mM EDTA, 5 mM 2- mercaptoethanol, 1% Triton X- 100, and protease inhibitor cock- tail (cOmplete, Roche). Cells were lysed by sonication, and the lysates were clarified by centrifugation at 20,000g at 4°C for 15 min. Clarified lysates were flash- frozen in liquid nitrogen and stored at −80°C until needed. For binding to beads, saturating amounts of clarified bacterial lysates containing GST- tagged baits were added to glutathione- Sepharose beads previously washed with lysis buffer [50 mM tris (pH 7.4), 150 mM NaCl, 1 mM EDTA, 5 mM 2- mercaptoethanol, and 1% Triton X- 100] and incubated at 4°C for 1 hour on a tube roller. Beads were washed once with lysis buffer, once with lysis buf- fer supplemented with 500 mM NaCl, and once again with lysis buf- fer and incubated with clarified bacterial lysates containing the recombinant prey for 2 hours at 4°C on a rotator. Beads were washed three times with lysis buffer and eluted by boiling in lysis buffer supplemented with a 4× solution of NuPAGE LDS sample buffer containing 100 mM DTT. Boiled slurry was separated on SDS- PAGE gels and analyzed by Coomassie blue stain (Abcam, ab119211) or by immunoblot using an anti- His6 HRP- conjugated antibody. TBC1D23 affinity chromatography from 293T cell lysate GST- TBC1D23 (1 to 513) and GST- TBC1D23 (514 to 684) were ex- pressed in the E. coli strain BL21- GOLD (DE3; Agilent Technolo- gies). Bacteria were grown at 37°C to an OD600 of 0.7 and induced with 100 μM IPTG overnight at 16°C. Cells were harvested by cen- trifugation, resuspended in lysis buffer [25 mM tris- HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% (v/v) Triton X- 100, and one EDTA- free cOmplete protease tablet/50 ml and 1 mM phenylmethylsulfo- nyl fluoride (PMSF)], dounce- homogenized and sonicated on ice. The lysates were clarified by centrifugation at 12,000g for 15 min at 4°C and applied at saturating levels to glutathione- Sepharose beads. The beads were then incubated with cell lysates prepared from two confluent 175- cm2 flasks of 293T cells. 293T cells were collected by centrifugation at 500g for 3 min, washed once in ice- cold PBS, and lysed in lysis buffer [25 mM tris- HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% (v/v) Triton X- 100, 1 mM PMSF, and one cOmplete pro- tease inhibitor tablet/50 ml) at 4°C for 30 min, followed by clarifica- tion by centrifugation at 17,000g for 10 min. Lysates were precleared on empty glutathione- Sepharose beads for 30 min before a 2- hour incubation with TBC1D23- coated beads. Beads were washed exten- sively in lysis buffer, and proteins eluted first in a high- salt elution buffer (25 mM tris- HCl, 1.5 M NaCl, and 1 mM EDTA) to release interacting proteins and then in SDS sample buffer to release the GST- fusion proteins. Proteins were precipitated from the high- salt elution sample by chloroform/methanol precipitation and resus- pended in SDS sample buffer containing 1 mM 2- mercaptoethanol. Eluates were separated by SDS- PAGE and analyzed by MS. Affinity chromatography of 293T cell lysate with GST- tail fusions The approach used was as described previously (46), as follows. Clarified lysates from 450 ml of 2× TY cultures containing bacteria expressing recombinant GST, GST- CPD (1321 to 1380), GST- CIMPR (2327 to 2491), and GST- TBC1D23 (559 to 684) were thawed. For each GST- tagged bait, 100 μl of glutathione- Sepharose 4B bead slurry (GE Healthcare) was used. Clarified bacterial lysates were added to the empty glutathione- Sepharose beads and incubated at 4°C for one hour on a roller. 293T cells (from four confluent T175 flasks per GST- tagged bait) were collected by scraping, washed twice with PBS, and lysed with lysis buffer [50 mM tris (pH 7.4), 150 mM NaCl, 1 mM EDTA, 5 mM 2- mercaptoethanol, 1% Triton X- 100, and EDTA- free cOmplete protease inhibitor]. The lysate was clari- fied by centrifugation for 5  min at 17,000g and precleared with 100 μl of bead slurry for 1 hour at 4°C on a tube roller. Beads loaded with recombinant GST- tagged baits were washed once with ice- cold lysis buffer, once with lysis buffer supplemented with 500 mM NaCl, and once again with lysis buffer. Beads were incubated with the pre- cleared 293T cell lysate for 2 to 4 hours on a roller at 4°C. Beads were washed twice with lysis buffer, transferred to 0.8  ml centrifuge columns (Pierce, 89869B), and washed twice more. Columns were brought to room temperature and eluted five times with 100 μl of elution buffer (1.5 M NaCl in lysis buffer) by centrifugation at 100g for 1 min; for the final elution, the sample was centrifuged at 17,000g for 1 min. Eluates were pooled together and concentrated to ~75 μl using a centrifugal filter (Amicon Ultra 0.5- ml 3000, Millipore, UFC500324), supplemented with 25 μl of NuPAGE 4× LDS sample buffer (Invitrogen, NP0007) containing 100 mM DTT. The eluate (40%) was separated by SDS- PAGE and stained with InstantBlue Coomassie stain (Abcam, ab119211). Each lane was cut into eight gel slices, transferred into a 96- well microtiter plate, and subjected to MS analysis. MS analysis of eluates from affinity chromatography For samples prepared with GST fusions to parts of TBC1D23, slices were destained with 50% (v/v) of acetonitrile and 50 mM ammonium bicarbonate, reduced with 10 mM DTT, and alkylated with 55 mM iodoacetamide. After alkylation, proteins were digested with trypsin (Promega, UK) overnight at 37°C at an enzyme to protein ratio of 1:20. The resulting peptides were separated by nanoscale capillary liquid chromatography (LC)–MS/MS using an Ultimate U3000 high- performance LC (HPLC) (Thermo Fisher Scientific, Dionex, San Jose, USA) to deliver a flow of ~300  nl/min. A C18 Acclaim PepMap100 5 μm, 100- μm × 20- mm nanoViper (Thermo Fisher Scientific, Dionex, San Jose, USA), trapped the peptides before separation on a 25- cm PicoCHIP nanospray column packed with Reprosil- PUR C18 AQ (New Objective Inc., Littleton, USA). Pep- tides were eluted with a 60- min gradient of acetonitrile [2 to 80% (v/v)]. The analytical column outlet was directly interfaced via a nanoflow electrospray ionization source, with a hybrid dual pres- sure linear ion trap mass spectrometer (Orbitrap Velos, Thermo Fisher Scientific, San Jose, USA). Data- dependent analysis was carried out, using a resolution of 30,000 for the full MS spectrum, followed by 10 MS/MS spectra in the linear ion trap. MS spectra were collected over a mass/charge ratio (m/z) range of 300 to 2000. MS/MS scans were collected using a threshold energy of 35 for collision induced dissociation. LC- MS/MS data were then searched against a protein database (UniProtKB) using the Mascot search engine programme (Matrix Science) (47). Database search parameters were set with a precursor tolerance of 5 parts per million (ppm) and a fragment ion mass tolerance of 0.8 Da. Two missed enzyme cleavages were allowed, and variable modifications for oxi- dized methionine, carbamidomethyl cysteine, pyroglutamic acid, phosphorylated serine, threonine, and tyrosine were included. MS/ MS data were validated using the Scaffold programme (Proteome Software Inc.) (48). For samples prepared with GST fusions to tails of cargo proteins, gel slices (1 to 2 mm) were placed in 96- well microtiter plates and D ow nloaded from https://w w w .science.org on A pril 22, 2024 A.6 CIMR crystallization screens C-15 Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 13 of 15 destained with 50% (v/v) of acetonitrile and 50 mM ammonium bicarbonate, reduced with 10 mM DTT, and alkylated with 55 mM iodoacetamide. After alkylation, proteins were digested with trypsin (6 ng/μl; Promega, UK) overnight at 37°C. The resulting peptides were extracted in 2% (v/v) of formic acid and 2% (v/v) acetonitrile. The digests were separated by nanoscale capillary LC- MS/MS using an Ultimate U3000 HPLC (Thermo Fisher Scientific, Dionex, San Jose, USA) to deliver a flow of ~300  nl/min. A C18 Acclaim PepMap100 5 μm, 100- μm × 20- mm nanoViper (Thermo Fisher Scientific, Dionex, San Jose, USA), trapped the peptides before sepa- ration on a C18 BEH130 1.7 μm, 75- μm × 250- mm analytical ultrahigh- performance LC column (Waters, UK). Peptides were eluted with a 60- min gradient of acetonitrile (2 to 80%). The ana- lytical column outlet was directly interfaced via a nanoflow elec- trospray ionization source, with a quadrupole Orbitrap mass spectrometer (Q- Exactive HFX, Thermo Fisher Scientific). MS data were acquired in data- dependent mode using a top 10 method, where ions with a precursor charge state of 1+ were excluded. High- resolution full scans (R = 60,000; m/z, 300 to 1800) were recorded in the Orbitrap, followed by higher energy collision dissociation (26% normalized collision energy) of the 10 most intense MS peaks. The fragment ion spectra were acquired at a resolution of 15,000, and dynamic exclusion window of 20 s was applied. Raw data files from LC- MS/MS data were processed using Proteome Discoverer v2.1 (Thermo Fisher Scientific) and then searched against a human protein database (UniProtKB, reviewed) using the Mascot search engine programme. Database search parameters were set with a pre- cursor tolerance of 10 ppm and a fragment ion mass tolerance of 0.2 Da. One missed enzyme cleavage was allowed, and variable modifications for oxidized methionine, carbamidomethyl cysteine, pyroglutamic acid, phosphorylated serine, threonine, and tyrosine were included. MS/MS data were validated using Scaffold. For the analysis of mass spectral intensities, all raw files were processed with MaxQuant v1.5.5.1 using standard settings and searched against UniProt with the Andromeda search engine integrated into the MaxQuant software suite (49). Enzyme search specificity was Trypsin/P for both endoproteinases. Up to two missed cleavages for each pep- tide were allowed. Carbamidomethylation of cysteines was set as fixed modification with oxidized methionine and protein N- acetylation considered as variable modifications. The search was performed with an initial mass tolerance of 6 ppm for the precursor ion and 0.5 Da for MS/MS spectra. The false discovery rate was fixed at 1% at the peptide and protein level. Statistical analysis was carried out using the Perseus module of MaxQuant (50). Peptides mapped to known contaminants and reverse hits were removed, and only protein groups identified with at least two peptides, one of which was unique, and two quantitation events were considered for data analysis. Each protein had to be detected in at least two of the three replicates. Missing values were imputed by values simulating noise using the Perseus’ default settings. To calculate P values, t tests were performed. Expression of 15N- labeled CPD(1321 to 1380) for NMR analysis The GST- CPD (1321 to 1380) fusion contains a PreScission protease cleavage site downstream of the GST tag. An overnight 2× TY starter culture was inoculated into 1- liter flasks of NH4Cl- free M9 medium containing yeast nitrogen base (1.7 g/liter; Sigma- Aldrich, Y1251), 15NH4Cl (1 g/liter; Sigma- Aldrich, 299251), and glucose (4 g/liter). Protein expression was induced using 300 μM IPTG and incubated overnight at 16°C for 16 hours, and bacterial lysate was prepared as described above for nonisotopically labeled GST fusions. Excess of glutathione- Sepharose beads (500 μl of bead slurry per liter of 2× TY culture) was incubated with the clarified bacterial lysate for 1 hour at 4°C. Beads were then washed twice with lysis buffer and incubated with the clarified lysate for 1  hour at 4°C on a rotator. Beads were then washed twice with lysis buffer [50 mM tris (pH 7.4), 150 mM NaCl, 1 mM EDTA, and 1% Triton X- 100] and twice with NMR buffer [50 mM tris (pH 7.4), 150 mM NaCl, and 5 mM 2- mercaptoethanol]. Beads were then incubated with GST- tagged PreScission protease (GE Healthcare, GE27- 0843- 01) at a concen- tration of 50 μl of enzyme per 1 ml of bead slurry and incubated from 5 hours to overnight at 4°C on a rotator. The supernatant was collected, concentrated using an Amicon 0.5- ml 3 KDa centrifugal filter (Millipore, UFC500324), and flash- frozen in liquid nitrogen. Protein concentration was measured using Bradford reagent (Bio- Rad, 5000006). NMR data collection and analysis for CPD All CPD NMR datasets for were collected at 278 K using a 600- MHz Bruker Avance III spectrometer with a 5- mm TCI triple resonance cryoprobe. All samples were prepared with 5% D2O as a lock solvent, at pH 7.4, 50 mM tris, 150 mM NaCl, and 5 mM 2- mercaptoethanol. 1H- 15N band- selective excitation short transient- transverse relax- ation optimized spectroscopy (BEST- TROSY) was collected for all samples using an optimized pulse sequence (51). The assignment of backbone NH, N, Cα, Cβ, and C′ resonances of the 86 μM 15N- 13C CPD (residues 1321 to 1380) sample was completed using standard three- dimensional datasets acquired as pairs to provide own and pre- ceding carbon connectivities and between 20 and 40% nonuniform sampling to aid faster data acquisition. Both the HNCO and HN(CA) CO experimental pair and the CBCA(CO)NH and HNCACB pair were collected with 1024, 64, and 96 complex points in the proton, nitrogen, and carbon dimensions, respectively. Nitrogen connectivities were established using (H)N(COCA)NNH and (H)N(CA)NNH experiments with 2048, 64, and 96 complex points in the proton, direct, and indirect nitrogen dimensions, respectively. All data were processed using Topspin versions 3.2 or 4 (Bruker) or, if required, NMRPipe (52), with compressed sensing for data reconstruction (53), and analyzed using NMRFAM- Sparky. The backbone assign- ment was completed using the Mars program (54). Binding of TBC1D23 to the CPD construct was observed by 1H- 15N BEST- TROSY. Unlabeled TBC1D23 was added to 15N- labeled CPD with a final concentration of both proteins at 38 μM. The 1H- 15N BEST- TROSY spectrum was compared to a second spectrum recorded for a 38 μM sample of free 15N CPD. While some small- peak perturba- tions were observed, the main differences caused by TBC1D23 bind- ing were line broadening. This was quantified by taking the peak height ratios of the bound and free spectra. To analyze the chemical shift perturbations, the following equation was used to report calcu- lated the distance between each peak in the bound spectrum when compared to the assigned, free CPD: Δδ“total” = ((δH)2+ (δN/5)2)0.5, with the smallest value reported as the minimal chemical shift perturbation. Isothermal titration microcalorimetry Experiments were performed using a Nano ITC machine from TA Instruments. TBC1D23 C- terminal domains were gel- filtered into D ow nloaded from https://w w w .science.org on A pril 22, 2024 C-16 Publications from this thesis Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 14 of 15 buffer composed of 100 mM NaCl, 100 mM tris (pH 7.4), and 1 mM tris(2- carboxyethyl)phosphine (TCEP). Peptides were dissolved in the same batch of buffer. Both the wild- type and mutant C- terminal domains were concentrated to 100 μM (1.4 mg/ml) with peptide concentration varying between 0.8 and 5 mM (depending on the peptide). Peptides were titrated into TBC1D23 C- terminal domains with 20 injections of 2.4 μl each separated by 300- s intervals. Ex- periments were conducted at 20°C. A relevant syringe- peptide- into- buffer blank was subtracted from all data, and for constructs that displayed measurable binding, three independent runs that showed clear saturation of binding were used to calculate the mean KD of the reaction, the stoichiometry (n), and the SEM calculated using NanoAnalyze. Data were exported to GraphPad Prism for figure generation. Protein expression and purification for structural analysis Recombinant proteins were expressed in BL21 plyS E. coli grown in shaking 2× TY medium at 37°C. Expression was induced with 0.2 mM IPTG, and cells were grown overnight at 22°C. Cells were resuspended in buffer A [250 mM NaCl, 20 mM tris (pH 7.4), and 1 mM DTT] supplemented with 4- benzenesulfonyl fluoride hydro- chloride, MnCl2 and deoxyribonuclease I. Cell pellets were lysed using a cell disruptor (Constant Systems) before clarification by ultracentrifugation (104,350 relative centrifugal force) for 45 min. The supernatant was batch bound to glutathione- Sepharose resin (Cytiva) in buffer A before washing with 400 ml of buffer A and overnight cleavage of the GST- tag with thrombin at room tempera- ture. The resultant flow through was concentrated for gel filtration Superdex 200 column (GE Healthcare) into either buffer A for crys- tallization or into ITC buffer [100 mM NaCl, 100 mM tris (pH 7.4), and 1 mM TCEP]. X- ray crystallography TBC1D23 C- terminal domains was concentrated to either 11 mg/ ml (full length) or 14 mg/ml (- VLDALES`) and mixed with 1.5 times molar excess syntaxin- 16209–221 peptide. High- throughput sitting drops were used to obtain crystallization conditions, which were further optimized before being repeated in hanging drop with protein:peptide complex being mixed 1:1 ratio with crystallization mother liquor. TBC1D23 full- length C- terminal domain was crys- tallized in 0.02 M citric acid, 0.08 M bis- tris propane (pH 8.8), and 16% (w/v) of PEG3350. TBC1D23 (- VLDALES) C- terminal domain was crystallized in 0.8 M potassium phosphate dibasic, 0.1 M Hepes/ NaOH (pH 7.5), 0.8 M sodium phosphate monobasic, and 1% of 1,2- butandiol. Crystals were cryoprotected by soaking in mother liquor supplemented with 35% glycerol and syntaxin- 16209–221 pep- tide (1 mg/ml) and flash- cooled in liquid nitrogen. Diffraction data were collected at the Diamond Light Source on the IO4 beamline at 100 K and processed with Autoproc. Structures were solved by molecular replacement with the PHASER- MR pro- gram using as a search model a single copy of the TBC1D23 C- terminal domain (6JM5) with zinc and water molecules removed. REFMAC5 was used for iterative rounds of refinement interspersed by manual rebuilding of the model using Coot. Crystallographic programs were run using the CCP4i2 package, and figures were ren- dered using UCSF ChimeraX (55). Data collection and refinement statistics are summarized in table S1. Electron density of the full- length TBC1D23 domain structure in the presence of syntaxin- 16 peptide clearly showed C- terminal tails bound in two molecules and syntaxin- 16 bound in a third with the fourth being somewhat ambiguous (fig. S4). Deletion of the C- terminal tail (VLDALES) and inclusion of syntaxin- 16 peptide resulted in a structure whose elec- tron density, when solved by molecular replacement using 6JM5 minus the VLDALES as the search model, unambiguously showed the syntaxin- 16 peptide bound in all molecules. Coordinates and reflection data are deposited in the Protein Data Bank (PDB) under accession code 8QQF. Supplementary Materials This PDF file includes: Figs. S1 to S5 Tables S1 and S2 Legends for data S1 to S3 Other Supplementary Material for this manuscript includes the following: Data S1 to S3 REFERENCES AND NOTES 1. R. W. Baker, F. M. Hughson, Chaperoning SNARE assembly and disassembly. Nat. Rev. Mol. Cell Biol. 17, 465–479 (2016). 2. J. S. Bonifacino, B. S. Glick, The mechanisms of vesicle budding and fusion. Cell 116, 153–166 (2004). 3. C. Ungermann, D. Kümmel, Structure of membrane tethers and their role in fusion. Traffic 20, 479–490 (2019). 4. V. Szentgyörgyi, A. Spang, Membrane tethers at a glance. J. Cell Sci. 136, jcs260471 (2023). 5. A. K. Gillingham, S. Munro, Transport carrier tethering - How vesicles are captured by organelles. Curr. Opin. Cell Biol. 59, 140–146 (2019). 6. M. Lowe, The physiological functions of the golgin vesicle tethering proteins. Front. Cell Dev. Biol. 7, 94 (2019). 7. P. Z. C. Chia, P. A. Gleeson, The regulation of endosome- to- Golgi retrograde transport by tethers and scaffolds. Traffic 12, 939–947 (2011). 8. N. Muschalik, S. Munro, Golgins. Curr. Biol. 28, R374–R376 (2018). 9. M. Wong, S. Munro, Membrane trafficking. The specificity of vesicle traffic to the Golgi is encoded in the golgin coiled- coil proteins. Science 346, 1256898 (2014). 10. M. Wong, A. K. Gillingham, S. Munro, The golgin coiled- coil proteins capture different types of transport carriers via distinct N- terminal motifs. BMC Biol. 15, 3 (2017). 11. K. Sato, P. Roboti, A. A. Mironov, M. Lowe, Coupling of vesicle tethering and Rab binding is required for in vivo functionality of the golgin GMAP- 210. Mol. Biol. Cell 26, 537–553 (2015). 12. J. J. H. Shin, A. K. Gillingham, F. Begum, J. Chadwick, S. Munro, TBC1D23 is a bridging factor for endosomal vesicle capture by golgins at the trans- Golgi. Nat. Cell Biol. 19, 1424–1432 (2017). 13. W. Huang, Z. Liu, F. Yang, H. Zhou, X. Yong, X. Yang, Y. Zhou, L. Xue, Y. Zhang, D. Liu, W. Meng, W. Zhang, X. Zhang, X. Shen, Q. Sun, L. Li, C. Ma, Y. Wei, D. D. Billadeau, X. Mo, D. Jia, Structural and functional studies of TBC1D23 C- terminal domain provide a link between endosomal trafficking and PCH. Proc. Natl. Acad. Sci. U.S.A. 116, 22598–22608 (2019). 14. D. Liu, F. Yang, Z. Liu, J. Wang, W. Huang, W. Meng, D. D. Billadeau, Q. Sun, X. Mo, D. Jia, Structure of TBC1D23 N- terminus reveals a novel role for rhodanese domain. PLOS Biol. 18, e3000746 (2020). 15. P. Navarro Negredo, J. R. Edgar, P. T. Manna, R. Antrobus, M. S. Robinson, The WDR11 complex facilitates the tethering of AP- 1- derived vesicles. Nat. Commun. 9, 596 (2018). 16. G. H. H. Borner, M. Y. Hein, J. Hirst, J. R. Edgar, M. Mann, M. S. Robinson, Fractionation profiling: A fast and versatile approach for mapping vesicle proteomes and protein- protein interactions. Mol. Biol. Cell 25, 3178–3194 (2014). 17. T. S. Gomez, D. D. Billadeau, A FAM21- containing WASH complex regulates retromer- dependent sorting. Dev. Cell 17, 699–711 (2009). 18. D. Jia, T. S. Gomez, D. D. Billadeau, M. K. Rosen, Multiple repeat elements within the FAM21 tail link the WASH actin regulatory complex to the retromer. Mol. Biol. Cell 23, 2352–2361 (2012). 19. Y. Tu, L. Zhao, D. D. Billadeau, D. Jia, Endosome- to- TGN trafficking: Organelle- vesicle and organelle- organelle interactions. Front. Cell Dev. Biol. 8, 163 (2020). 20. D. Jia, T. S. Gomez, Z. Metlagel, J. Umetani, Z. Otwinowski, M. K. Rosen, D. D. Billadeau, WASH and WAVE actin regulators of the Wiskott- Aldrich syndrome protein (WASP) family are controlled by analogous structurally related complexes. Proc. Natl. Acad. Sci. U.S.A. 107, 10442–10447 (2010). D ow nloaded from https://w w w .science.org on A pril 22, 2024 A.6 CIMR crystallization screens C-17 Cattin-Ortolá et al., Sci. Adv. 10, eadl0608 (2024) 29 March 2024 S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E 15 of 15 21. F. J. Eng, O. Varlamov, L. D. Fricker, Sequences within the cytoplasmic domain of gp180/ carboxypeptidase D mediate localization to the trans- Golgi network. Mol. Biol. Cell 10, 35–46 (1999). 22. O. Varlamov, L. D. Fricker, Intracellular trafficking of metallocarboxypeptidase D in AtT- 20 cells: Localization to the trans- Golgi network and recycling from the cell surface. J. Cell Sci. 111, 877–885 (1998). 23. E. P. Roquemore, G. Banting, Efficient trafficking of TGN38 from the endosome to the trans- Golgi network requires a free hydroxyl group at position 331 in the cytosolic domain. Mol. Biol. Cell 9, 2125–2144 (1998). 24. M. S. Struthers, S. G. Shanks, C. MacDonald, L. N. Carpp, A. M. Drozdowska, D. Kioumourtzoglou, M. L. M. Furgason, M. Munson, N. J. Bryant, Functional homology of mammalian syntaxin 16 and yeast Tlg2p reveals a conserved regulatory mechanism. J. Cell Sci. 122, 2292–2299 (2009). 25. S. C. Harrison, Peptide–surface association: The case of PDZ and PTB domains. Cell 86, 341–343 (1996). 26. D. J. Owen, P. R. Evans, A structural explanation for the recognition of tyrosine- based endocytotic signals. Science 282, 1327–1332 (1998). 27. E. L. Huttlin, R. J. Bruckner, J. Navarrete- Perea, J. R. Cannon, K. Baltier, F. Gebreab, M. P. Gygi, A. Thornock, G. Zarraga, S. Tam, J. Szpyt, B. M. Gassaway, A. Panov, H. Parzen, S. Fu, A. Golbazi, E. Maenpaa, K. Stricker, S. Guha Thakurta, T. Zhang, R. Rad, J. Pan, D. P. Nusinow, J. A. Paulo, D. K. Schweppe, L. P. Vaites, J. W. Harper, S. P. Gygi, Dual proteome- scale networks reveal cell- specific remodeling of the human interactome. Cell 184, 3022–3040.e28 (2021). 28. C. Evrard, P. Rouget, Subcellular localization of neural- specific NPDC- 1 protein. J. Neurosci. Res. 79, 747–755 (2005). 29. A. T. Lin- Moore, M. J. Oyeyemi, M. Hammarlund, rab- 27 acts in an intestinal pathway to inhibit axon regeneration in C. elegans. PLOS Genet. 17, e1009877 (2021). 30. C. Levecque, A. Velayos- Baeza, Z. G. Holloway, A. P. Monaco, The dyslexia- associated protein KIAA0319 interacts with adaptor protein 2 and follows the classical clathrin- mediated endocytosis pathway. Am. J. Physiol. Cell Physiol. 297, C160–C168 (2009). 31. L. Zhao, H. Deng, Q. Yang, Y. Tang, J. Zhao, P. Li, S. Zhang, X. Yong, T. Li, D. D. Billadeau, D. Jia, FAM91A1–TBC1D23 complex structure reveals human genetic variations susceptible for PCH. Proc. Natl. Acad. Sci. U.S.A. 120, e2309910120 (2023). 32. J. J. H. Shin, O. M. Crook, A. C. Borgeaud, J. Cattin- Ortolá, S. Y. Peak- Chew, L. M. Breckels, A. K. Gillingham, J. Chadwick, K. S. Lilley, S. Munro, Spatial proteomics defines the content of trafficking vesicles captured by golgin tethers. Nat. Commun. 11, 5987 (2020). 33. I. Marin- Valencia, A. Gerondopoulos, M. S. Zaki, T. Ben- Omran, M. Almureikhi, E. Demir, A. Guemez- Gamboa, A. Gregor, M. Y. Issa, B. Appelhof, S. Roosing, D. Musaev, B. Rosti, S. Wirth, V. Stanley, F. Baas, F. A. Barr, J. G. Gleeson, Homozygous mutations in TBC1D23 lead to a non- degenerative form of pontocerebellar hypoplasia. Am. J. Hum. Genet. 101, 441–450 (2017). 34. E. L. Ivanova, F. T. Mau- Them, S. Riazuddin, K. Kahrizi, V. Laugel, E. Schaefer, A. de Saint Martin, K. Runge, Z. Iqbal, M.- A. Spitz, M. Laura, N. Drouot, B. Gérard, J.- F. Deleuze, A. P. M. de Brouwer, A. Razzaq, H. Dollfus, M. Z. Assir, P. Nitchké, M.- V. Hinckelmann, H. Ropers, S. Riazuddin, H. Najmabadi, H. van Bokhoven, J. Chelly, Homozygous truncating variants in TBC1D23 cause pontocerebellar hypoplasia and alter cortical development. Am. J. Hum. Genet. 101, 428–440 (2017). 35. N. H. Cho, K. C. Cheveralls, A.- D. Brunner, K. Kim, A. C. Michaelis, P. Raghavan, H. Kobayashi, L. Savy, J. Y. Li, H. Canaj, J. Y. S. Kim, E. M. Stewart, C. Gnann, F. McCarthy, J. P. Cabrera, R. M. Brunetti, B. B. Chhun, G. Dingle, M. Y. Hein, B. Huang, S. B. Mehta, J. S. Weissman, R. Gómez- Sjöberg, D. N. Itzhak, L. A. Royer, M. Mann, M. D. Leonetti, OpenCell: Endogenous tagging for the cartography of human cellular organization. Science 375, eabi6983 (2022). 36. I. S. Stein, A. Gottfried, J. Zimmermann, G. Fischer, von Mollard, TVP23 interacts genetically with the yeast SNARE VTI1 and functions in retrograde transport from the early endosome to the late Golgi. Biochem. J. 419, 229–236 (2009). 37. I.- M. Yu, F. M. Hughson, Tethering factors as organizers of intracellular vesicular traffic. Annu. Rev. Cell Dev. Biol. 26, 137–156 (2010). 38. K. E. McNally, P. J. Cullen, Endosomal retrieval of cargo: Retromer is not alone. Trends Cell Biol. 28, 807–822 (2018). 39. P. Navarro Negredo, J. R. Edgar, A. G. Wrobel, N. R. Zaccai, R. Antrobus, D. J. Owen, M. S. Robinson, Contribution of the clathrin adaptor AP- 1 subunit μ1 to acidic cluster protein sorting. J. Cell Biol. 216, 2927–2943 (2017). 40. J. Stöckli, S. Höning, J. Rohrer, The acidic cluster of the CK2 site of the cation- dependent mannose 6- phosphate receptor (CD- MPR) but not its phosphorylation Is required for GGA1 and AP- 1 binding. J. Biol. Chem. 279, 23542–23549 (2004). 41. J. Hirst, G. H. H. Borner, R. Antrobus, A. A. Peden, N. A. Hodson, D. A. Sahlender, M. S. Robinson, Distinct and overlapping roles for AP- 1 and GGAs revealed by the “knocksideways” system. Curr. Biol. 22, 1711–1716 (2012). 42. Y. Homma, S. Hiragi, M. Fukuda, Rab family of small GTPases: An updated view on their regulation and functions. FEBS J. 288, 36–55 (2021). 43. I. Topalidou, J. Cattin- Ortolá, B. Hummer, C. S. Asensio, M. Ailion, EIPR1 controls dense- core vesicle cargo retention and EARP complex localization in insulin- secreting cells. Mol. Biol. Cell 31, 59–79 (2020). 44. J.- P. Concordet, M. Haeussler, CRISPOR: Intuitive guide selection for CRISPR/Cas9 genome editing experiments and screens. Nucleic. Acids Res. 46, W242–W245 (2018). 45. F. A. Ran, P. D. Hsu, J. Wright, V. Agarwala, D. A. Scott, F. Zhang, Genome engineering using the CRISPR- Cas9 system. Nat. Protoc. 8, 2281–2308 (2013). 46. J. Cattin- Ortolá, L. G. Welch, S. L. Maslen, G. Papa, L. C. James, S. Munro, Sequences in the cytoplasmic tail of SARS- CoV- 2 spike facilitate expression at the cell surface and syncytia formation. Nat. Commun. 12, 5333 (2021). 47. D. N. Perkins, D. J. C. Pappin, D. M. Creasy, J. S. Cottrell, Probability- based protein identification by searching sequence databases using mass spectrometry data. Electrophoresis 20, 3551–3567 (1999). 48. A. Keller, A. I. Nesvizhskii, E. Kolker, R. Aebersold, Empirical statistical model to estimate the accuracy of peptide identifications made by MS/MS and database search. Anal. Chem. 74, 5383–5392 (2002). 49. J. Cox, M. Mann, MaxQuant enables high peptide identification rates, individualized p.p.b.- range mass accuracies and proteome- wide protein quantification. Nat. Biotechnol. 26, 1367–1372 (2008). 50. S. Tyanova, T. Temu, P. Sinitcyn, A. Carlson, M. Y. Hein, T. Geiger, M. Mann, J. Cox, The Perseus computational platform for comprehensive analysis of (prote)omics data. Nat. Methods 13, 731–740 (2016). 51. A. Favier, B. Brutscher, Recovering lost magnetization: Polarization enhancement in biomolecular NMR. J. Biomol. NMR 49, 9–15 (2011). 52. F. Delaglio, S. Grzesiek, G. W. Vuister, G. Zhu, J. Pfeifer, A. Bax, NMRPipe: A multidimensional spectral processing system based on UNIX pipes. J. Biomol. NMR 6, 277–293 (1995). 53. K. Kazimierczuk, V. Y. Orekhov, Accelerated NMR spectroscopy by using compressed sensing. Angew. Chem. Int. Ed. Engl. 50, 5556–5559 (2011). 54. Y.- S. Jung, M. Sharma, M. Zweckstetter, Simultaneous assignment and structure determination of protein backbones by using NMR dipolar couplings. Angew. Chem. Int. Ed. Engl. 43, 3479–3481 (2004). 55. M. D. Winn, C. C. Ballard, K. D. Cowtan, E. J. Dodson, P. Emsley, P. R. Evans, R. M. Keegan, E. B. Krissinel, A. G. W. Leslie, A. McCoy, S. J. McNicholas, G. N. Murshudov, N. S. Pannu, E. A. Potterton, H. R. Powell, R. J. Read, A. Vagin, K. S. Wilson, Overview of the CCP4 suite and current developments. Acta Crystallogr. D Biol. Crystallogr. 67, 235–242 (2011). Acknowledgments: We thank M. Ailion, D. Billadeau, L. Fricker, and O. Perisic for providing reagents; S. Gray for technical assistance; and A. McCoy for invaluable help compiling the structural data. We thank Diamond Light Source for beamtime and the staff of beamline IO4 for assistance with crystal testing and data collection. Funding: Funding was provided by the Medical Research Council, as part of U.K. Research and Innovation file reference number MC_ U105178783 (to S.M.) and Wellcome Trust grant number 207455/Z/17/Z (to D.J.O.). Author contributions: Conceptualization: S.M., J.C.- O., and A.K.G. Methodology: S.M., J.C.- O., T.J.S., A.K.G., and D.J.O. Investigation: J.C.- O., J.G.G.K., A.K.G., J.L.W., S.- Y.P.- C., and D.J.O. Validation: S.M., J.C.- O., A.K.G., D.J.O., and J.G.G.K. Visualization: S.M., J.C.- O., A.K.G., D.J.O., J.L.W., and J.G.G.K. Resources: J.B., J.C.- O., and D.J.O. Formal analysis: S.M., J.B., J.C.- O., D.J.O., J.G.G.K., and S.- Y.P.- C. Supervision: S.M. and D.J.O. Project administration: S.M. and D.J.O. Funding acquisition: S.M. and D.J.O. Writing—original draft: S.M., D.J.O., and J.C.- O. Writing—review and editing: S.M., D.J.O., and J.C.- O. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. The structure of TBC1D23 C terminus in complex with the syntaxin- 16 peptide is deposited at PDB (8QQF). Submitted 26 September 2023 Accepted 26 February 2024 Published 29 March 2024 10.1126/sciadv.adl0608 D ow nloaded from https://w w w .science.org on A pril 22, 2024