Translational regulation in
aggressive B-cell lymphomas
Joanna Alicja Krupka
Gonville and Caius College
This dissertation is submitted on Easter Term, 2021
for the degree of Doctor of Philosophy

Declaration
This dissertation 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 that I have submitted, or am concurrently submitting,
for a degree or 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.
I further state that no substantial part of my dissertation has already been submitted, or
is being concurrently submitted, for any such 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. This dissertation does not exceed the prescribed
limit of 60 000 words.
Joanna Alicja Krupka
September, 2021

Abstract
Translational regulation in aggressive B-cell lymphomas
Joanna Alicja Krupka
The Germinal Centre (GC) reaction is a dynamic process where B-cells undergo
recombination and somatic hypermutation of immunoglobulin genes in response to antigen
stimulation. This essential component of the adaptive immune system is associated
with cycles of intensive proliferation and selection, which carries a risk of malignant
transformation. Aggressive lymphomas arising from the GC stage of B-cell development are
the most common haematological malignancies with heterogeneous molecular mechanisms
and clinical presentation. Although the last decade witnessed considerable advances in
the biology of GC reaction and related tumours, the studies focused predominantly on the
network of transcription factors.
The advances in Next Generation Sequencing technologies have opened new possibilities
to explore mechanisms of regulation beyond the level of transcription. Ribo-Seq is a
technique combining ribosome footprinting with deep sequencing of mRNA fragments that
allows to map the position of translating ribosomes with single nucleotide precision.
Here I investigate the mechanisms of translational regulation contributing to lymphoma
development. Firstly, I introduce RiboStream, an automated bioinformatic pipeline
designed to streamline processing of Ribo-Seq datasets while maintaining transparency
and reproducibility of the computational workflow. Then, I provide an overview and
benchmarking of current methods for identifying translationally regulated genes. Based
on these I select a strategy to reveal that overexpression of two B-cell oncogenes, BCL6 or
MYC, is followed by preferential translational of selected transcripts. Next, I show that
loss-of-function mutations in RNA-helicase (DDX3X) promote early development of MYC-
driven lymphoma by bu↵ering the e↵ects of MYC on translation of ribosomal proteins
and the rate of global protein synthesis. Finally, I explore a genome-wide distribution of
translating ribosomes to study the scope of non-canonical translation in lymphoid cells.
Taking advantage of a large dataset of 79 Ribo-Seq libraries I reveal pervasive translation
of ostensibly non-coding regions, and design a knock-down CRISPR screen library to
identify those important for B-cell survival.

Acknowledgements
First and foremost, I would like to express my gratitude to my supervisor, Dr. Daniel
Hodson, who took the risk of having me in his lab and gave me the freedom to discuss
problems I found interesting. His expertise, advice and sense of humour were invaluable. I
would like to thank Dan for being a trusted mentor and friend on my way to becoming a
scientist.
I am also thankful to my second supervisor, Dr. Shamith Samarajiwa, for his enthusiasm
for accompanying me with my first steps in bioinformatics. By sharing all computational
resources with me, I enjoyed an unrestrained opportunity to explore my ideas, for which I
am immensely grateful. I am also indebted to Dr. Martin Turner for inspiring discussions
and encouragement to look ahead into unexplored territories of science.
I would like to also thank all members of the Hodson and the Samarajiwa labs, who
made the last four years a good-humoured and colourful time. Special thanks go to Jie
Gao for her strength of spirit in preparing all the Ribo-Seq libraries and to Chun Gong,
Mata Vorri and Hendrik Runge for fruitful collaboration. My time in Cambridge would not
be the same without Dora Bihary, Shoko Hirosue, Cassandra Kosmidou, David Shorthouse
and Katie Young. Thank you for sharing your time with me and an entire series of pub
events, which helped me get through challenging times.
I would not be where I am today without my dearest friends, who stayed in Poland.
My PhD adventure would not be possible without Dr. Agnieszka Graczyk-Jarzynka, who
one day in November 2015 welcomed me in the Department of Immunology (Medical
University of Warsaw) and became my first mentor. I am also grateful to Julia, Sara,
Kasia and Ania, who, despite being thousands of kilometres away, were always keeping
my spirits up.
All of this would not have been possible without the support of Cancer Research UK
Cambridge Centre and Addenbrooke’s Charitable Trust, who funded my research.
Finally, I would like to thank my family. I am grateful to my parents, brother and
grandparents for believing in me and supporting even the most bizarre of my ideas, to
Alek’s mom and grandma for our weekly chats on FaceTime, and lastly to Alek for his
endless patience and fabulous lemon tarts.

Contents
1 Introduction 15
1.1 Molecular basics of RNA translation . . . . . . . . . . . . . . . . . . . . . 16
1.1.1 Four stages of protein synthesis . . . . . . . . . . . . . . . . . . . . 18
1.1.2 Overview of ribosome biogenesis . . . . . . . . . . . . . . . . . . . . 23
1.2 Translational control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.2.1 Regulation of the translation initiation . . . . . . . . . . . . . . . . 25
1.2.2 Regulation of translation elongation . . . . . . . . . . . . . . . . . . 27
1.2.3 Regulation of translation termination . . . . . . . . . . . . . . . . . 28
1.3 Deregulation of translation in human cancers . . . . . . . . . . . . . . . . . 29
1.3.1 Oncogenic and tumour suppressor pathways converge at controlling
protein synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.3.2 Ribosome biogenesis and its oncogenic potential . . . . . . . . . . . 32
1.3.3 Translation factors are frequently deregulated in cancer . . . . . . . 33
1.3.4 The significance of translational response to stress in cancer . . . . 34
1.4 Elements of B-cell biology from the perspective of lymphoma development 38
1.5 Role of translation in B-cell development and malignancy . . . . . . . . . . 42
1.6 Toolkit to study heterogeneity of translation . . . . . . . . . . . . . . . . . 46
1.7 Project aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2 Materials and methods 53
2.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.1.1 Overview of genomic sequences and annotations used in this study . 53
2.1.2 Overview of external datasets used in this study . . . . . . . . . . . 54
2.1.3 Overview of computational software used in this study . . . . . . . 55
2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2.1 Next Generation Sequencing library preparation and sequencing . . 56
2.2.1.1 RNA-Seq . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2.1.2 Ribo-Seq . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2.2 Processing and quality control of Next Generation Sequencing data 57
2.2.2.1 Adapter trimming and alignment to the reference genome 57
2.2.2.2 Read counting . . . . . . . . . . . . . . . . . . . . . . . . 57
2.2.3 Di↵erential translation analysis of Ribo-Seq data . . . . . . . . . . 57
2.2.4 Di↵erential expression and downstream analysis of RNA-Seq data . 58
2.2.5 Metagene analysis of iCLIP and Ribo-Seq data . . . . . . . . . . . 58
2.2.6 Di↵erential expression analysis of RNA-Seq data . . . . . . . . . . . 58
2.2.7 Downstream data analysis . . . . . . . . . . . . . . . . . . . . . . . 58
2.2.7.1 Individual-nucleotide resolution UV crosslinking and im-
munoprecipitation (iCLIP) . . . . . . . . . . . . . . . . . . 59
2.2.8 Identification of DDX3X mutations from RNA-Seq data . . . . . . . 59
2.2.9 DLBCL Cell-of-origin identification from RNA-Seq data . . . . . . 60
2.2.10 Chromosome Y expression identification from RNA-Seq data . . . . 60
2.2.11 Hierarchical de novo identification of translated regions from Ribo-
Seq data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.2.12 Reanalysis of published mass spectrometry datasets . . . . . . . . . 62
2.2.13 Analysis of proteogenomic data downloaded from OpenProt and
sORFdb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.14 Evolutionary conservation of identified ORFs . . . . . . . . . . . . . 63
2.2.15 CRISPR screen design . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.16 Figures preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.2.17 R and Bioconductor . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3 Genome wide quantification of translation in lymphoid malignancies 67
3.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.1.1 Establishing a bioinformatic pipeline for processing of translatome
and transcriptome data . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.1.2 Quality of translatome profiling in primary GC B-cells . . . . . . . 74
3.1.3 Benchmarking statistical approaches for di↵erential translation analysis 79
3.2 Translational regulation following BCL6 and MYC overexpression in primary
GC B-cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4 Mutations in RNA helicase DDX3X facilitate MYC-driven lymphoma-
genesis 95
4.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.2.1 Examining the prevalence and distribution of DDX3X mutations . . 98
4.2.1.1 DDX3X is preferentially mutated in MYC driven lymphomas 98
4.2.1.2 Context dependent pattern of DDX3X mutation in di↵erent
cancer types . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.2.1.3 DDX3X mutations in B-cell lymphomas cluster within
C-terminal helicase domain . . . . . . . . . . . . . . . . . 101
4.2.1.4 Males with Burkitt Lymphoma and DLBCL are more likely
to have DDX3X mutation . . . . . . . . . . . . . . . . . . 103
4.2.2 DDX3X regulates ribosome biogenesis and global protein synthesis . 105
4.2.2.1 DDX3X binds preferentially to mRNA encoding compo-
nents of core translation machinery . . . . . . . . . . . . . 105
4.2.2.2 DDX3X regulates translation of a subset of expressed tran-
scripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.2.3 Deregulation of MYC in primary GC B-cells increases ribosome
biogenesis and triggers ER stress. . . . . . . . . . . . . . . . . . . . 118
4.2.4 DDX3X mutation interferes with endoplasmic reticulum stress response120
4.2.4.1 DDX3X R475C mutation is associated with suppression of
unfolded protein response in U2932 cells . . . . . . . . . . 120
4.2.4.2 DDX3X mutation is associated with suppression of un-
folded protein response in BL patients . . . . . . . . . . . 123
4.2.5 Up-regulation of DDX3Y in established tumours rescues loss of
DDX3 helicase activity . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5 Elucidating the role of translated micropeptides in Di↵use Large B-cell
Lymphoma 133
5.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.2.1 A systematic approach for de novo identification of noncanonical
translation products in lymphoid cells. . . . . . . . . . . . . . . . . 135
5.2.1.1 An integrated ORF identification workflow . . . . . . . . . 135
5.2.1.2 Pervasive translation of crude non-coding regions in lym-
phoid cells . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
5.2.2 Noncanonical ORFs account for about 10% of proteins detected in
proteomics experiments . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.2.3 Characteristics of noncanonical ORFs producing MHC-bound peptides144
5.2.4 Design of customised knockout CRISPR screen to identify noncanon-
ical ORFs important for B-cells survival . . . . . . . . . . . . . . . 146
5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6 Perspectives 155
Bibliography 159



CHAPTER 1
Introduction
The ‘central dogma’ of molecular biology, proposed by Crick (1958), states that genes
are chemically expressed as proteins in a sequence of molecular procedures. First, a
DNA sequence is rewritten (transcribed) into RNA, then decoded (translated) into an
amino-acids chain, which eventually folds into a protein (Figure 1.1). For a long time,
the translation process and its core components - the ribosomes and translation factors
were viewed as molecular machines passively processing all available transcripts. This focus
has been rewired by the ready avaliability of technology to quantify mRNA abundance,
such as microarray or RNA-Seq.
Figure 1.1: Francis Crick’s first draft of the central dogma of molecular biology
from an unpublished note (1956)
However, the relationship between mRNA and protein levels is far from being simple.
Firstly, it is estimated that steady-state transcript levels in human cells can explain
between 56% and 84% of protein abundance (Lundberg et al., 2010; Liu et al., 2016;
Schwanha¨usser et al., 2011; Jovanovic et al., 2015). The analysis performed by Jovanovic
et al. (2015) showed that mRNA level, translation intensity and protein degradation
rate can explain up to 79% of total protein abundance, with 18 - 26% and 8 - 22% of
this corresponding to translation and protein degradation, respectively. Secondly, the
relationship between mRNA and its product may vary when comparing di↵erent cell types
and conditions. Relative changes in protein abundance during dynamic cell transitions
are explained predominantly by the mRNA abundance, however, the rates of synthesis
15
and degradation vary substantially between individual proteins, which highlights the
importance of regulation other than transcriptional (Jovanovic et al., 2015; Mathieson
et al., 2018).
Therefore, there is a growing appreciation of the importance of regulation imposed
at the point of translation. Protein synthesis consumes the majority of cellular energy
resources, especially if a cell is bio-synthetically active, rapidly growing, or di↵erentiating
(Rolfe and Brown, 1997; Buttgereit and Brand, 1995; Lynch and Marinov, 2015). Hence,
precise regulation of translation seems to be energetically ecient providing an opportunity
to modulate cellular protein levels quickly. It is not surprising that translational control
is especially important for highly energetic processes, such as cell proliferation, hormone
release and stress response (Hershey et al., 2012). Translation is also the primary mechanism
of gene expression regulation in cells lacking active transcription (for example, in oocytes
or red blood cells) or during the early stages of viral infection, when the host transcription
is suppressed (Hershey et al., 2012; Mohr, 2016). Deregulated translation is a hallmark of
cancer, but the exact role of this is complex and context dependent.
1.1 Molecular basics of RNA translation
Translation is an evolutionarily conserved process during which the nucleotide sequence of
a messenger RNA (mRNA) is decoded into the amino-acid sequence of a protein by rules
known as genetic code. A CoDing Sequence region (CDS) is a region of mRNA or genomic
DNA whose nucleotide sequence determines amino-acids sequence. It is usually flanked by
two UnTranslated Regions (UTRs): 50UTR and 30UTR. Every three adjacent nucleotides
of a CDS form a unit called a codon. There are 4 x 4 x 3 = 64 possible combinations
of the three nucleotides, three of which indicate the end of a protein; the remaining 61
corresponds to one out of 20 amino acids. An Open Reading Frame (ORF) is a series
of codons contained between a start and a stop codon. A given RNA sequence can be
decoded in one out of three possible reading frames depending on position of the first
(start) codon.
The codons of a protein coding sequence are deciphered from 50 end to 30 end by
transfer RNA (tRNA). tRNAs are small adaptor molecules that pair with a coding triplet
through a complementary anticodon. The genetic code is redundant, which means that
more than one codon specifies a single amino acid, and more than one tRNA molecule
can match a single codon. Some tRNAs require a complete match between codon and
anticodon, but for some, one mismatch (wobble) at the third position of the codon does
not a↵ect the accurate amino-acid determination. A tRNA carrying a matching amino
acid is called aminoacyl-tRNAs.
The majority of protein coding mRNAs undergoe two specific post-transcriptional
16
modifications: polyadenylation of the 30 end and capping of the 50end with a methylated
guanosine at 7th position (m7G).
The actual reactions of protein synthesis take place inside the ribosome, a large
ribonucleoprotein complex (RNP) composed of multiple ribosomal proteins and ribosomal
RNAs (rRNA). A fully assembled eukaryotic ribosome (80S) consists of a small (40S) and
a large subunit (60S). When not involved in active translation, the ribosomal subunits
are separated. During evolution, the ribosome became larger and richer, but its core
components are so highly conserved that a common ancestor of all living species must have
synthesised its proteins similarly to organisms living today. The human ribosome contains
80 ribosomal proteins and four separate rRNA molecules that account for more than 80 %
of the cellular RNA pool (Cech, 2000). Each ribosome has three tRNA binding sites, that
is A-site for aminoacyl-tRNA, P-site for peptidyl-tRNA and E-site for exit (Figure 1.2).
Each covers three adjacent codons and is essential for the orchestration of the ribosome
movement along the coding sequence. The ribosome reads only one codon at a time, but
several ribosomes (polysomes) can simultaneously translate a single mRNA molecule.
Figure 1.2: Structural model of the eukaryotic 80S ribosome
Three sites for tRNA binding are highlighted: E-site - orange, P-site - violet and A-site -
green. mRNA is shown in red. Estimated localisation of the ribosomal proteins is indicated
by shaded areas: blue for 60S and yellow for 40S subunit; from Schuller and Green (2018)
17
1.1.1 Four stages of protein synthesis
Translation process consists of four phases: initiation, elongation, termination and recycling.
Translation initiation
Translation initiation is a multi-step process, which is considered the principal point of
translational control. There are two main modes of translation initiation: 50cap dependent
and 50cap independent, of which 50cap dependent is the most common.
The canonical pathway of 50cap dependent translation initiation can be divided into 4
stages. The ternary complex (TC), which consists of methionine carrying tRNA (Met-
tRNA or methionyl-tRNA), GTP and eukaryotic Initiation Factor 2 (eIF2) complex is
loaded to the small ribosomal subunit (40S) recycled from the previous round of translation
with bound eIF1, eIF1A and eIF3 factors. TC together with 40S, eIF1, eIF1A and eIF3
form a 43S preinitiation complex (43S PIC) (Figure 1.3, steps 1-2) (Jackson et al.,
2010). Recruitment of 43S PIC onto mRNA requires cooperation of eIF4F complex, eIF4B
and eIF4H. eIF4F complex, consists of a 50cap-binding eIF4E, RNA-helicase eIF4A and
a sca↵old protein eIF4G (Figure 1.3, step 3). Apart from interacting with eIF4A
and eIF4E, eIF4G binds also poly(A) binding protein (PABP), which brings mRNA into
circular shape with 50cap and 30-poly(A) tail close together creating the so-called closed-loop
structure (Mangus et al., 2003; Taylor et al., 2020). 43S PIC starts base-by-base scanning
of 50UTR. It consists of coordinated unwinding of the 50UTR secondary structures and
43S PIC movement towards 30end. Once 43S PIC recognises a start codon (usually AUG
encoding methionine), the scanning process stops (Jackson et al., 2010). Sometimes an
optimal start codon is not necessarily the first AUG encountered. In most eukaryotic
mRNAs, a start codon is framed into a Kozak consensus sequence, an RNA sequence
context ensuring the fidelity of the translation initiation site (TIS) (Kozak, 1987). If the
context of a potential start codon is weak (poorly resembling a Kozak sequence), PIC
may skip it and continue to search for the next codon. This phenomenon is known as
leaky scanning (Zhang et al., 2019). The selection of the optimal TIS is promoted by eIF1,
which discriminates between AUG and weak context codons. The commitment to a TIS
is promoted by eIF5, a GTPase-activating protein targeting eIF2 complex and leading
to partial dissociation of eIF2 from 40S subunit (Jackson et al., 2010). Assembly of 80S
ribosome at the TIS is mediated by eIF5B and followed by release of eIF1, eIF1A, eIF3 and
residual eIF2. (Figure 1.3, step 4). The ribosome is now ready to enter the elongation
stage (Aitken and Lorsch, 2012; Hinnebusch, 2014).
18
Figure 1.3: Diagram showing a simplified process of 50cap- dependent translation
initiation in eukaryotic cells.
The process of translation initiation is divided into four steps:
Step 1. Ternary complex formation from GTP-bound eIF2 and methionyl-tRNA (Met-
tRNA).
Step 2. 43S preinitiation complex (43S PIC) assembly, which includes the ternary complex,
the small ribosome subunit 40S and translation initiation factors (eIF3, eIF1A and eIF1)
Step 3. mRNA binding to 43S PIC promoted by eIF4F complex and eIF4B followed by
50UTR scanning in 50 to 30end direction. An optimal translation initiation site, typically
AUG codon framed by Kozak sequence, is promoted by eIF1 and eIF5
Step 4. Recruitment of the large ribosome subunit 60S leading to 80S ribosome assembly,
and dissociation of the initiation factors mediated by eIF5B.
Adapted from Protein Translation Cascade, by BioRender.com (2021). Retrieved from
https://app.biorender.com/biorender-templates
19
A less common mechanism of translation initiation involves 50cap-independent ribosome
recruitment. Although the exact mechanisms of cap-independent translation are not fully
understood, the two proposed mechanisms involve Internal Ribosome Entry Sites (IRES)
and 50cap-independent translational enhancers (CITEs). Translation through IRES is a
common strategy employed by pathogenic viruses to escape global halt of 50cap mediated
translation in the host cell, but was also found in eukaryotic genes, particularly those
involved in stress or anti-viral response (Yang and Wang, 2019; Jackson et al., 2010). IRES
can interact with canonical initiation factors and recruits 40S ribosomal subunit through
IRES trans-acting factors (ITAFs) (Komar and Hatzoglou, 2011; Yang et al., 2017; Meyer
et al., 2015) or short cis-elements that pairs with 18S rRNA (Dresios et al., 2006; Yang
and Wang, 2019). Because of the lack of conserved IRES sequence, the exact number of
IRES-initiating open reading frames and factors regulating this mode of translation is
unknown.
An alternative mechanism of 50cap-independent translation, also originating from
viruses, involves CITEs - RNA structural elements within mRNAs attracting translation
initiation factors (Shatsky et al., 2018). Proposed mechanism of CITEs-mediated trans-
lation initiation involves reversible N6A-methylation of mRNA within GAC sequence
context recognised by YTHDF1 and followed by direct recruitment of the eIF3 complex
(Meyer et al., 2015; Wang et al., 2015; Shatsky et al., 2018). Although it is still a widely
debated topic, both CITEs and IRES initiating translation may proceed without a full set
of translation initiation factors (Terenin et al., 2013).
It is important to highlight that 50cap-independent translation is still a largely unex-
plored territory and some controversies arose around the accuracy of IRES reporter assays
and so the existence of IRES in eukaryotic cells (Shatsky et al., 2018; Yang and Wang,
2019). Nevertheless the concept of alternative translation initiation adds another layer of
complexity to the cellular translatome potentially broadening proteome diversity.
Translation elongation
Translation elongation takes place between loading the first aminoacyl-tRNA after the
start codon and encountering the first in-frame stop codon. Translation proceeds in the
50 - to - 30 direction, so the N-terminal end of a protein is translated first (Schuller and
Green, 2018). The elongation stage consists of three major steps: tRNA binding, peptide
bond formation and translocation.
After initiation is completed, the tRNA carrying the first N-terminal amino-acid
(usually Met-tRNA) is localised in the middle slot of the ribosome (P-site). Next slot
(A site) exposes the next codon. Once tRNA carrying the next amino-acid binds to the
complementary A site, eukaryotic Elongation Factors 1 alpha 1 (eEF1A1) hydrolyses GTP
to GDP fixing aminoacyl-tRNA within the A-site. The next stage is the formation of a
20
peptide bond between two adjacent amino-acids. The peptidyl transferase centre of the
large ribosomal subunit catalyses the reaction, after which the aminoacyl-tRNA occupying
the P-site releases the attached amino-acid. The ribosome is now in the pre-translocation
state with a peptidyl-tRNA (tRNA with nascent peptide attached) in the A-site and
deacetylated tRNA (tRNA with detached amino-acid) in the P-site (Schuller and Green,
2018). The last step involves the a series of conformational changes that pushes the entire
ribosome three nucleotides towards the 3’ end - a reaction conjugated with another GTP
hydrolysis reaction catalysed by eEF2. During the translocation phase, deacetylated tRNA
from the P-site is transferred to the E-site and released (Ratje et al., 2010). The A-site is
now empty and ready to accept a new aminoacyl-tRNA. The elongation cycle then repeats
until the termination codon appears in the A-site. Each elongation cycle will add one new
amino acid to the growing polypeptide chain (Figure 1.4).
Ecient conformational changes of the ribosome during each elongation cycle are
essential for maintaining the right direction of translation. The energy for this process is
delivered by GTP to GDP hydrolysis, which is performed by the two eucariotic Elongation
Factors 1 and 2 (eEF1 and eEF2).
The accuracy of the ribosome in decoding mRNA into protein is estimated to reach
almost 99,99% (1 misincorporation per 104 amino acids joined). Because the release
of faulty protein product could potentially cause serious consequences for the cell, two
proofreading mechanisms monitor each elongation cycle (Neelagandan et al., 2020). Correct
codon-anticodon matching is favoured due to its high anity, which stabilises the bond
formation, a↵ects rRNA folding around the tRNA-rRNA interaction and triggers GTP
hydrolysis by eEF1. When this is missing, the elongation becomes slow, aminoacyl tRNA
cannot be fixed in the A-site and dissociates o↵ the ribosome. An invalid tRNA in the P-
site increases the risk for further decoding errors, thus decreasing chances for synthesising
a full-length protein. Repetitive amino-acid misincorporation may lead to premature
termination of translation (Neelagandan et al., 2020).
Translation termination
The end of the coding sequence is marked by a stop codon (UAA, UAG or UGA). All
three are recognised by eukaryotic release factor 1 (eRF1). Another release factor, eRF3,
is a GTP-ase that promotes the hydrolysis of the peptidyl-tRNA bond in the P-site. This
releases the C-terminal end of the newly synthesised amino-acid chain from the ribosome.
The nascent polypeptide then completes folding in the cytoplasm. The post-termination
complex (post-TC) disassembles and can be recycled to participate in another round of
translation (Hellen, 2018).
21
Ribosome recycling
The ribosome recycling stage aims to split 80S ribosomes into separate subunits,
preparing them for a new round of translation. This is initiated by the recruitment of
ABCE1 to post-termination ribosomes with eRF1 in the A-site. After 80S disassembly, the
last step involves releasing deacetylated tRNA and mRNA from the 40S ribosome subunit,
which is mediated by translation initiation factors: eIF1, eIF1A and eIF3. The full process
of translation termination and recycling was reviewed comprehensively by Hellen (2018)
Figure 1.4: Diagram showing the process of translation elongation in eukaryotic
cells.
A) Schematic of the translation elongation cycle demonstrating how tRNA moves between
the ribosomal sites:
Amino-acyl tRNA (green) recognises codon in the ribosomal A-site. This is followed by
peptide-bond formation and transfer of the nascent peptide chain to amino-acyl tRNA in
the A-site, now a peptidyl-tRNA. Change in the ribosome conformation after peptide bound
formation are referred to as a hybrid state - a transient conformation, where anticodon
loop of the tRNA remains fixed in in the P and A sites of the small ribosomal subunit, but
the amino-acid site of tRNA (the acceptor stem) is already in the E and P sites of the large
ribosomal subunit (Schuller and Green, 2018) The process of the ribosome translocation
ends with peptidyl-tRNA in the P-site and A-site ready to accept the next amino-acyl
tRNA (Schuller and Green, 2018).
B) Overview of the peptide-transfer reaction catalysed by the ribosome. The peptide-bond
formation occurs through nucleophilic attack of the amino group of the new amino acid
(bound to tRNA in the A-site) on the ester linkage on the peptidyl-tRNA (remaining in
the P-site) (Schuller and Green, 2018). This leaves a deacetylated tRNA in P-site and
peptidyl-tRNA in the A-site longer by one amino-acid. From Schuller and Green (2018)
22
1.1.2 Overview of ribosome biogenesis
Ribosome biogenesis is a complex process involving biosynthesis and assembly of the
complete 80S human ribosome. Briefly, 80S ribosome consists of 2 subunits: small and
large, both composed from ribosomal RNA (rRNA) and ribosomal proteins (RPs). The
small subunit (40S) is responsible for binding, scanning and unwinding of the mRNA,
while the function of the large subunit (60S) is to catalyse peptide bond formation and
check the quality of the nascent peptide (Pelletier et al., 2018).
The nucleolus, a nuclear substructure, is central to the process of ribosome biogenesis.
It is responsible for transcription, processing and modifications of rRNA and the assembly
of precursor ribosomal subunits (Lafontaine et al., 2021). rRNA sequences are organised
in clusters of tandem repeats encoded by ribosomal DNA (rDNA) in nucleolus organiser
regions (NORs). NORs contain 18S, 5.8S and 28S rRNA sequences (47S pre-rRNA)
separated by spacers with regulatory sequences and distributed between the short arms of
five acrocentric chromosomes 13, 14, 15, 21 and 22 (Henderson et al., 1972; Pelletier et al.,
2018). 47S pre-rRNA is transcribed by RNA polymerase I. Another ribosome component,
5S rDNA cluster, is localised on chromosome 1, outside the nucleolus, and transcribed
by RNA polymerase III (Pol III). In contrast, mRNAs of RPs are transcribed by RNA
polymerase II (Pol II), exported to the cytoplasm for translation and then re-imported to
the nucleus to participate in the ribosome assembly (Pelletier et al., 2018). In the nucleolus,
47S pre-rRNA, 5S rRNA, numerous RPs and assembly factors co-transcriptionally form a
90S processome (pre-ribosome) (Figure 1.5).
Next step is 90S maturation, which includes extensive base modifications and cleavage
reactions, resulting in the separation of the pre-40S and pre-60S subunits (Pelletier et al.,
2018; Lafontaine, 2015). This requires the activity of small nucleolar RNAs (snoRNAs)
derived from introns of certain Pol II transcribed genes. Finally, the subunits are exported
to the cytoplasm for the final maturation (incorporation of few additional RPs and
accessory factors) contributing to the assembly of 80S ribosome for protein synthesis
(Figure 1.5) (Tschochner and Hurt, 2003; Pelletier et al., 2018).
23
Figure 1.5: Overview ribosome biogenesis
The components of eukaryotic ribosomes are the product of three RNA polymerases
transcribing di↵erent parts of ribosomal RNA (rRNA), and mRNA for ribosomal proteins
(RP). Polymerase III (Pol III) is responsible for transcription of 5S rRNA cluster in the
nucleus, while polymerase I (Pol I) in nucleolus produces 47S pre-RNA containing 18S,
5.8S and 28S rRNA. As is typical for other protein-coding regions, the RP sequence is
transcribed by Pol II in the nucleus. RP’s mRNA is then exported to the cytoplasm for
translation and imported back to the nucleus. RPs, 47S pre-RNA and 5S rRNA participate
in the assembly of 90S processome, which undergoes chemical modifications and cleavage
resulting in formation of two separate ribosomal subunits (pre-40S and pre 60S). Pre-40S
and pre 60S are exported to the cytoplasm, where after final maturation step, they are
ready to participate in protein synthesis. From Pelletier et al. (2018)
24
1.2 Translational control
Regulation of translation is a broad term covering mechanisms a↵ecting di↵erent stages
of protein synthesis. A given mechanism may a↵ect translation globally or be specific
towards a single mRNA or a transcript group. A further distinction can be made between
processes that modify the function of the core components of the translation machinery
and those that a↵ect RNA directly. From a molecular perspective, translational control
includes a variety of mechanisms such as chemical modifications (e.g. phosphorylation
or methylation), di↵erential expression of translation factors, modulation of the 3D
structure of mRNA, action of trans-activating RNA-binding proteins (RBPs), presence of
cis-regulatory elements in mRNA or recruitment of microRNAs to 30UTR.
For clarity, I will review the most important points of regulation, focusing on each
stage of translation separately.
1.2.1 Regulation of the translation initiation
Translation initiation is considered the main rate-limiting step of eukaryotic protein
synthesis. Many of the regulatory mechanisms target the balance between 50cap dependent
and 50cap independent translation by modulating the activity of core translation initiation
factors.
The most extensively studied mechanisms of translation initiation control involves the
regulation of eIF4E activity through 1) transcription, 2) phosphorylation or 3) sequestration
by binding to a family of translational repressors (Raught and Gingras, 1999). eIF4E is
a direct transcriptional target of many important signalling pathways, including NF-B
and c-MYC (Hariri et al., 2013). Phosphorylation of eIF4E at Ser 209 is mediated by
MAP kinase-interacting kinases MNK1/MNK2 and occurs in response to various mitogenic
and stress factors promoting 50cap dependent translation initiation (Sonenberg, 1996;
Pyronnet et al., 1999). Another arm of regulation is formed by the family of eIF4E
binding proteins (EIF4EBPs), which bind eIF4E and remove it from the pool available
for translation. Binding of 4E-BPs and eIF4G, a part of eIF4F complex, are mutually
exclusive for eIF4E. (Yang et al., 2020). A well characterised regulator of this process is
mTOR signalling, which couples the activity of growth factors with nutrients availability
and the intensity of anabolic processes (Saxton and Sabatini, 2017; Liu and Sabatini, 2020;
Kim and Guan, 2019a). mTOR-mediated phosphorylation of EIF4EBPs disrupts the
4E-BP–eIF4E complex allowing eIF4E to participate in eIF4F initiation complex. EIF4E
is required for translation of all 50capped mRNAs, but not all transcripts are equally
susceptible for changes in eIF4E levels. Transcripts with long 50UTRs and specific RNA
regulatory motifs are particularly sensitive to eIF4E levels. (Smith et al., 2021).
25
The 50UTR scanning requires coordinated unwinding of mRNA structure, which is
performed by RNA helicases, typically eIF4A - a component of the eIF4F complex.
eIF4A activity is promoted by two cofactors: eIF4B or eIF4H, while the avaliability by
programmed cell death 4 (PDCD4). (Yang et al., 2003; Smith et al., 2021). PDCD4 acts
downstream of mTOR and is controlled by inactivation phosphorylation by Ribosomal
Protein S6 Kinase (S6K) (Silvera et al., 2010). Although the net e↵ect of eIF4A inhibition
or sequestration is a mild reduction in global translation rate, the rate and even the
direction of change vary between single transcripts (Modelska et al., 2015; Smith et al.,
2021). eIF4A-sensitive mRNAs are characterised by complex secondary structures in their
50UTRs and the presence of specific RNA motifs (Steinhardt et al., 2014; Modelska et al.,
2015; Wolfe et al., 2014a).
Another important mechanisms of translation initiation control involves phosphoryla-
tion of eIF2↵, the main regulatory subunit of eIF2 complex, which is responsible for joining
methionine-tRNA to the small ribosomal subunit (Smith et al., 2021). Phosphorylated
eIF2↵ has an inhibitory e↵ect on global translation, because it prevents GDP to GTP
restoration of eIF2, another eIF2 subunit, trapping it its inactive form. This shifts the
balance towards 50cap-independent translation initiation. Translation control through
eIF2↵ is central to the Integrated Stress Response (ISR), an adaptive pathway, which
aims to restore cellular homeostatsis or commit the cell to apoptosis following exposure to
unfavourable conditions (Pakos-Zebrucka et al., 2016), see section 1.3.4.
The eciency of translation initiation can also be controlled by numerous RNA-binding
proteins, of which RNA helicases deserves special attention. RNA helicases are multifunc-
tional proteins involved in all aspects of RNA metabolism, such as mRNA processing,
nuclear export, tracking in the cytoplasm, translation, degradation or microRNAs-
mediated RNA silencing. The activity of RNA helicases regulates almost all stages of
protein synthesis (Bourgeois et al., 2016; Linder and Jankowsky, 2011). The eciency of
50cap-dependent translation initiation relies on the ability to unwind structured 50UTRs.
Helicases assisting in this process include DDX48, DDX3X and components of eIF4F
complex (eIF4A1 and eIF4A2). Those are especially important for translating mRNAs with
GC-rich 50UTRs and 30UTRs with microRNA binding sites (Bourgeois et al., 2016; Linder
and Jankowsky, 2011; Sen et al., 2015; Wolfe et al., 2014b). Other roles of RNA helicases
in translation control include mRNA positioning at ribosomal 40S subunit (DHX29), regu-
lation of 40S scanning and ribosome recycling (DHX9), promoting 80S ribosome assembly
(DHX33, DDX3X) and recognition of stop codon (DDX19B). RNA helicases also control
the translation of mRNA regulons; for example, DDX25 is essential for the translation of
mRNAs associated with spermatogenesis, and DDX48 facilitates the translation of several
neuronal mRNAs (Bourgeois et al., 2016; Linder and Jankowsky, 2011). Despite recent
26
advances in the biology of RNA helicases, the full spectrum of their activity remains
elusive.
tRNA-derived small RNAs (tsRNAs), a novel class of regulatory non-coding RNAs,
compose another interesting layer of translational control. tsRNA are cleaved from several
types of tRNA in response to stress. These tsRNAs contains an oligo-G terminal motif
forming G-quadruplexes in the 50UTR of certain transcripts, that suppress translation
initiation by displacing eIF4F complex from 50cap (Ivanov et al., 2014). This mechanism
of translation inhibition by stress-induced tsRNAs was identified in several organisms and
cell types indicating high evolutionary conservation (Ivanov et al., 2011).
1.2.2 Regulation of translation elongation
For a very long time translation elongation was not considered an important rate-limiting
step of protein production. However, variety of factors can influence the elongation speed,
which is especially important in the terms of protein folding and transport. Regulatory
mechanisms of translation elongation involve: ORF codon composition, local sequence
context, post-transcriptional modifications of tRNA, mRNA, and the expression levels of
the core elongation factors (Knight et al., 2020; Schuller and Green, 2018).
The genetic code is redundant (degenerate), so each amino acid can be decoded by two,
four or six codons, with exception for methionine and tryptophan, which are encoded by a
unique codon. Interestingly, the distribution of synonymous codons over the transcriptome
is not uniform (codon bias), and the rate of translation of di↵erent codons is uneven
(codon optimality). Hypotheses put forward to explain an evolutionary origin of such
phenomenon involve varying elongation rate of di↵erent codons, translation accuracy or
selection of splicing enhancers (Richter and Coller, 2015). The speed of tRNA-codon
pairing depends on the abundance of a specific tRNA in the cytoplasm (it takes more
time for a less abundant tRNA to find the codon) and interaction strength (standard
Watson-Crick pairing is quicker than wobble pairing). The relative abundance of di↵erent
types of tRNA in human is not uniform and varies between tissues promoting translation
of cell-specific transcripts (Dittmar et al., 2006). This view has been expanded by findings
in yeast, that demonstrated coordinated changes in the abundance of specific tRNAs
after exposure to various types of stress. In consequence, when tRNA pool is restricted,
transcripts with larger number of rare codons tends to be translated less eciently (Torrent
et al., 2018; Hanson and Coller, 2018). Codon bias correlates with tRNA levels, therefore,
to some extent, it is mirrored by codon optimality (Sabi and Tuller, 2014; Knight et al.,
2020). Both translation eciency and mRNA half-life correlate with codon composition,
but this changes with the concentration of specific tRNAs in the cytoplasm (Presnyak
et al., 2015). Although variations in codon selection and translation dynamics have been
27
observed within and between all species, the majority of studies in this topic came from
lower organisms (Hanson and Coller, 2018).
Certain codon patterns, for example poly-lysine tracks, or mRNA secondary structures
can lead to programmed ribosome stalling and frameshifting (Schuller and Green, 2018).
Local slowdown, or even halt in the elongation rate may facilitate protein folding or signal
recognition particle binding, which essential for secreted protein (Richter et al., 2012).
Ribosome transit can be regulated by several protein interacting with the A-site, for
example the elongation factor EF-G. Programmed frameshifting involves either slipping
back or skipping one nucleotide and occurs in all species (Ketteler, 2012). The scope of
this phenomenon in human is still debated. An example of a gene known to regulate its
translation through frameshifting is ornithine decarboxylase (ODC) (Bekaert et al., 2008).
1.2.3 Regulation of translation termination
Although termination of translation is usually not a rate-limiting step of protein synthesis,
the fidelity of this process is important for maintaining proteome integrity. The biological
activity of truncated or extended proteins, possible products of noncanonical termination
of translation, may be di↵erent from the original mRNA product.
Nonsense mediated decay (NMD) is a translation-coupled process of elimination of
mRNAs harbouring premature stop codons. This is an important surveillance mechanisms
preventing the expression of truncated proteins (Kurosaki and Maquat, 2016; Hellen,
2018). The role of NMD is not limited to aberrant transcript elimination. Transcripts
with upstream ORFs in their 50UTR, products of alternative splicing, intron retention
or auto-regulatory loops may utilise programmed NMD to regulate the pool of mRNAs
available for translation (Kurosaki and Maquat, 2016). It is estimated that the expression
level of about 10% of eukaryotic mRNAs may be modulated by NMD (Kurosaki and
Maquat, 2016)
Finally, near cognate recognition of stop codon (readthrough) may lead to an extended
product, but the full scope of this phenomenon in mammalian cells is still unclear.
Functional readthrough has been shown for three human genes: VEGFA, LDHB, and
MDH1 (Schueren and Thoms, 2016). Interestingly, the extended isoforms show di↵erent
subcellular location (LDHB, MHD1) or have the opposite biological activity to a canonical
product (antiangiogenic instead of proangiogenic in case of VEGFA) (Eswarappa et al.,
2014; Schueren and Thoms, 2016).
28
1.3 Deregulation of translation in human cancers
Enlarged nucleoli, which are the primary site of ribosome biogenesis, was one of the
first hallmarks of malignancy that was later widely applied in diagnostics (Pianese, 1896;
Gani, 1976). Deregulation of translation machinery is frequently observed in spontaneous
cancers as well as in hereditary cancer syndromes. While the latter usually involves point
mutations a↵ecting ribosome biogenesis, the scope of translational reprogramming in the
majority of cancers encompass a variety of mechanisms. Protein synthesis can be hijacked
by malignant cells through aberrant expression, point mutations or post-translational
modifications of translation factors or their regulators, mRNA regulatory elements, RNA
modifications, or preferential codon usage. The mechanisms can include global changes
in protein synthesis and/or increase or decrease in translation intensity of a subset of
transcripts.
1.3.1 Oncogenic and tumour suppressor pathways converge at
controlling protein synthesis
At the global scale, increased protein synthesis rate was shown to accompany high mitotic
activity (Johnson et al., 1976). Indeed, many key oncogenic and tumour suppressor
pathways, such as MYC, PI3K, RAS, PTEN, TP53, converge at the regulation of cellular
translation synchronising proliferation rate with anabolic and catabolic pathways, but this
relationship is far from being simple.
Deregulation of MYC is observed in >50% of human cancers (Meyer and Penn, 2008).
The oncoprotein MYC family consists of three genes: MYC, MYCN, and MYCL, which
have the capacity to regulate about 15% of human genes (Dang, 2012) coordinating
a variety of cellular functions including proliferation, metabolism, di↵erentiation and
immunosurveillance (Chen et al., 2018). MYC regulates protein synthesis through control-
ling the transcription of ribosomal proteins (Boon et al., 2001), ribosomal RNA (rRNA)
(Grandori et al., 2005) (Figure 1.6), and translation initiation factors (eIF4A, eIF4E,
and eIF4G) (Schmidt, 2004; Xu and Ruggero, 2020). Although the role of MYC far
exceeds the regulation of translation, its oncogenic potential is highly dependent on the
translation apparatus: the cell ability to phosphorylate eIF4E (Pourdehnad et al., 2013)
and augment protein synthesis (Barna et al., 2008), see section 1.5. Moreover, MYC
protein abundance is regulated at the level of translation. Firstly, through alternative,
IRES-mediated, translation initiation from a CUG start codon and secondly, by preferential
translation controlled by eIF4A and eIF4E (Schatz et al., 2011; Culjkovic-Kraljacic et al.,
2016).
29
Figure 1.6: Multilevel regulation of ribosome biogenesis by MYC
MYC controls ribosome biogenesis by promoting transcription of ribosome biogenesis
components. It does so by cooperation with other cofactors regulating the recruitment of
RNA polymerases (RNA pol I, II and III) and through chromatin structure remodelling
(Shiue et al., 2009; Van Riggelen et al., 2010). Adapted from Van Riggelen et al. (2010)
Interestingly, some ribosomal proteins were found to inhibit MYC expression creating
a negative feedback loop. For example, RPL11 interferes with MYC binding to 5S rRNA
and tRNA promoters (Dai et al., 2010) and, together with RPL5, jointly binds to MYC
mRNA directing it to RNA-induced silencing complex (RISC) for degradation (Liao et al.,
2014).
The translational control of PTEN/PI3K/AKT and RAS/MEK1/ERK signalling con-
verges at regulating the activity of mTORC1 complex, a key coordinator of cell growth,
survival and metabolism (Figure 1.7) (Kim and Guan, 2019b). Activation of these
pathways leads to inactivating phosphorylation of TSC1/2 complex, which is a nega-
tive regulator of mTORC1. The mTOR pathway promotes 50cap-dependent translation
by controlling eIF4E inhibitory binding proteins, which sequester eIF4E and prevent it
from participating in translation initiation (Silvera et al., 2010). Additional control is
imposed at the level of mTORC1 targets. MNK1 and MNK2, which are downstream
of RAS/MEK1/ERK, phosphorylate eIF4E at a single residue Ser209 independently of
mTORC1 control, thus selectively increasing translation intensity of several protumorigenic
transcripts in mice and human cell lines including anti-apoptotic MCL1 and cyclin D1
(CCND1), MYC, and proangiogenic VEGF and FGF2 (Furic et al., 2010; Sonenberg and
Hinnebusch, 2009; Kevil et al., 1996). Biophysical studies demonstrated that phosphory-
lation at Ser209 promotes 50cap independent translation, but the significance of this for
tumourgenesis is unclear (Zuberek et al., 2003). We know that oncogenic capacity of both
RAS/MEK1/ERK and MYC signalling depends on the ability to phosporylate eIF4E (in
vitro and in vivo evidence) (Yang et al., 2020) suggesting that a specific, prooncogenic,
mode of protein synthesis is a common strategy during cell transformation. Prooncogenic
translation is not limited to malignant cells, though. Robichaud et al. (2018) showed that
mice with mutated phosphorylation site (eIF4ES209A) are resistant to the development
of lung metastases due to decreased translation of MCL1 and BCL2 in prometastatic
neutrophils.
30
Figure 1.7: Signalling pathways converging at regulating protein synthesis
Mitogenic stimulation (growth factors, hormones or cytokines) targets receptor tyrosine
kinases (RTKs), which promotes RAS/ERK and PI3K/PTEN/AKT signalling. ERK
activates 50cap dependent translation either through MNK-mediated phosphorylation of
eIF4E, RSK-mediated phosphorylation of eIF4B or by inhibiting TSC1/TSC2complex, a
negative regulator of mTOR signalling. AKT also regulates translation through mTOR
by inhibiting another negative regulator, PRAS40. From Silvera et al. (2010)
31
Although the contribution of p53 to tumour formation is one of the best studied
mechanisms of cancer, its role in controlling protein synthesis and being regulated by
translation apparatus is less known. Overall, activation of p53 leads to suppression of
ribosome biogenesis a↵ecting translation initiation. Ribosome biogenesis is a complex
process that coordinates the synthesis of rRNA, ribosomal proteins and other auxillary
factors; p53 can control the activity of all of them. Firstly, p53 directly interferes with
the assembly of RNA Pol I complex, which is essential for rRNA transcription (Zhai and
Comai, 2000). Secondly, lessons learnt from studying ribosomopathies show that there is
a bidirectional feedback loop between p53 and the expression of ribosomal proteins. When
the process of ribosome biogenesis is impaired, for example due to mutation in a ribosomal
protein, DNA damage of the ribosomal DNA (rDNA) or insucient transcription of rRNA,
free ribosomal proteins are released to the nucleoplasm. RPL5, RPL11, RPL23a, RPS7,
and RPL26 are known to interact and sequester MDM2, a key inhibitor of p53 (Kampen
et al., 2020) This leads to cell cycle arrest, induction of cellular senescence, or apoptosis,
but it can vary between cell types. For example, haploinsuciency of RPS14 or RPS19 in
erythroid progenitor cells is known to increase p53 and CDKN1A (p21) protein in vivo
to the level similar to those after gamma irradiation (Dutt et al., 2011). p53 can inhibit
transcription of ribosomal proteins and other associated factors, indirectly, by inhibiting
MYC (Ho et al., 2005). The dynamic of p53-related nucleolar stress response and its links
to cancer and neurodegenerative diseases are not fully understood; this topic was reviewed
recently by Lindstro¨m et al. (2018) and Pfister (2019).
1.3.2 Ribosome biogenesis and its oncogenic potential
The involvement of ribosome biogenesis in oncogenesis is well understood in the context
of MYC expression deregulation, see section 1.3.1, but it is not limited to it. Other
ribosome biogenesis regulators with known oncogenic potential include netrin 1 (NTN1)
and epithelial cell-transforming sequence 2 oncogene (ECT2). ECT2 and truncated NTN1
isoform, expressed exclusively in cancer cells, promote rDNA transcription and pre-rRNA
processing. rDNA is one of the most transcriptionally active regions of the genome and
rDNA rearrangements are observed in the majority of lung and colorectal cancers (Stults
et al., 2009). During S phase, when rDNA transcription continues, RNA polymerase I
collides with the replication forks, which facilitates the formation of R-loops (rRNA-rDNA
hybrids) (Pelletier et al., 2018). R-loops are known hot spots of DNA damage (Helmrich
et al., 2011; Skourti-Stathaki and Proudfoot, 2014) and rDNA regions are common fragile
sites (Pelletier et al., 2018).
32
Cancer-specific changes in the ribosome morphology are another interesting concept,
which is sometimes referred to as the oncoribosome hypothesis. Varying expression level
of ribosomal proteins, post-transcriptional rRNA modifications and recurrent mutations
in ribosomal proteins are observed in many malignancies (Babaian et al., 2020; Bastide
and David, 2018; Pelletier et al., 2018)
1.3.3 Translation factors are frequently deregulated in cancer
Regulation of translation initiation is essential for maintaing cell homeostasis during
malignant transformation and under exposure to unfavorable conditions.
High levels of eIF4E are common in cancer and in vitro studies revealed that this can
be sucient to induce malignant transformation (Lazaris-Karatzas et al., 1990; Gingras
et al., 1999). Oncogenic properties of eIF4E has been shown also in vivo. Transgenic
mice overexpressing eIF4E have increased risk of several types of cancers, including B-cell
lymphomas (Ruggero et al., 2004) and accelerated B-cell lymphomagenesis when c-Myc is
deregulated as well (Wendel et al., 2004). Oncogenic properties of high levels of eIF4E are
attributed to its ability to regulate translation of specific transcripts rather than control on
global translation. Haploinsuciency of eIF4E is associated with normal translation levels
and development, but still prevents HRAS-induced transformation in mice (Truitt et al.,
2015). eIF4E-sensitive mRNAs (eIF4E regulon) are enriched for CERT motif and complex
RNA structures sequences in their 50UTRs. The latter involves eIF4E-mediated recruitment
of RNA helicase - eIF4A (Smith et al., 2021). These include transcripts associated with cell
proliferation, survival and oxidative stress, such as cyclins, ornithine decarboxylase (ODC),
vascular endothelial growth factor (VEGF), MYC, and phosphoribosyl-pyrophosphate
synthetase 2 (PRPS2) (Bhat et al., 2015). Tumourgenesis following eIF4E overexpression
develops relatively late, suggesting that other prooncogenic events might be required for
full transformation (Ruggero et al., 2004).
Interestingly, the role of eIF4E is not limited to translation initation. A substantial
proportion of it exists withing the nuclei, where eIF4E regulates export of selected mRNA
to the cytoplasm through the nuclear pore. Known transcripts, whose export is promoted
by eIF4E, include MYC, BCL6 and BCL2 (Culjkovic-Kraljacic et al., 2016).
Other translation factors important from the oncogenesis perspective are eIF3 subunits.
Prooncogenic properties have been reported for eIF3A–eIF3D, eIF3G, eIF3H, eIF3M,
eIF3E and eIF3F, but the exact role of individual subunits is not fully understood, see
Smith et al. (2021). Similairly to eIF4E, deregulation of certain eIF3 subunits leads
to preferential translation of specific prooncogenic regulon, which includes JUN, genes
associated with epithelial–mesenchymal transition or inflammation (Smith et al., 2021; Lee
et al., 2016; Desnoyers et al., 2015). Some functions are independent on the role of eIF3
complex in translation initation. For example, the role of eIF3A and eIF3B encompass
33
control of enzymes responsible for reversible N6-methyladenine RNA modification (Smith
et al., 2021). eIF3-complex independent pool of eIF3H acts as a deubiquitylating enzyme
stabilising YAP1, which is important for tumour progression and metastasis (Smith et al.,
2021). An interesting dual role has been shown also for eIF3G, which is cleaved during
apoptosis and translocated to the nucleus, where promotes caspase activation and DNA
degradation (Smith et al., 2021).
When 50cap-dependent translation is inhibited following cellular stress, eIF2A and
eIF5B promote translation of selected transcripts through non-canonical, IRES-mediated
initation of translation. These include few apoptosis inhibitors (XIAP, BIRC2 and
BCL2L1), CDKN1A (p21) and proteins associated with NF-B signalling (Smith et al.,
2021).
The role of translation initation factors in cancer has been summarised in Figure 1.8
1.3.4 The significance of translational response to stress in can-
cer
Uncontrolled cell growth and proliferation combined with limited nutrients supply, hypoxic
environment and deregulated cellular energetics are one of the hallmarks of cancer (Hanahan
and Weinberg, 2011). Adaptation to unfavourable conditions is essential to initiate and
maintain malignant transformation. Key to this process are evolutionarily conserved
stress-signalling pathways that aim to rebalance cell homeostasis or commit it to apoptosis,
if this cannot be achieved. Stress response encompasses a variety of mechanisms of which
the Integrated Stress Response (ISR), Unfolded Protein Response (UPR) and heat shock
response (HSR) are the ones most relevant for the topic of protein translation.
The principal point of ISR is the control of the avaliability of the ternary complex
for translation, thus reprogramming the global translation landscape (Costa-Mattioli and
Walter, 2020). The ternary complex consists of three subunits: eIF2↵, eIF2 and eIF2.
Accumulation of misfolded proteins, amino acid deprivation and other stress signals activate
eIF2↵ kinases: PERK, PKR, HRI, and GCN2 that converge on the phosphorylation of
eIF2↵.
The role of the ternary complex is AUG start codon recognition that triggers GTP
hydrolysis of eIF2 with the aid of eIF5. eIF2-GDP dissociates from the 40S complex and
is recycled for another round of translation by eIF2B, which restores eIF2-bound GDP to
GTP (Pakos-Zebrucka et al., 2016). The ↵ subunit of eIF2 is the main point of control.
In response to various stress stimuli, it can be phosphorylated at Ser 51 (P-eIF2↵) and
act as an inhibitor of eIF2B. In its inactive form (GDP-bound) P-eIF2↵ is limiting for
the ternary complex formation, which results in the reduction of global, 50cap dependent
protein synthesis enabling 50cap independent translation of selected transcripts.
34
Figure 1.8: Overview of 50cap dependent translation initiation in cancer
Numerous components of the translation initation machinery are deregulated in human
cancers. Up to now, the link between oncogenesis and translation initiation has been well
established for the components of eIF4F, eIF3 and eIF2 complex and ribosome biogenesis.
The overexpression of eIF4F components, which include eIF4E, eIF4A and eIF4G results
in preferential translation of selected transcripts. For individual eIF3 subunits similar
mechanisms was observed. eIF2↵ is phosphorylated by di↵erent kinases (PKR, PERK or
GCN2), which are reactive to various types of stress. The net e↵ect of P-eIF2↵ is repression
of 50cap-dependent translation, promotion of non-canonical translation initation and
adaptation to unfavourable conditions. Ribosomal biogenesis is another mechanisms
hacked by cancer cells to drive a malignant phenotype. Many oncogenic signalling pathways
converges at controlling the synthesis of the ribosome components.
Other factors with possible role in tumourgenesis include eIF5A, eIF5B and eIF2A. From
(Silvera et al., 2010)
35
Of these, activating transcription factor 4 (ATF4), activating transcription factor 5
(ATF5), DNA Damage Inducible Transcript 3 (DDIT3 or CHOP) promote transcription
of genes responsible for restoration of cell homeostasis (Hinnebusch, 2005; Vattem and
Wek, 2004; Pakos-Zebrucka et al., 2016).
The presence of misfolded proteins in the endoplasmic reticulum (ER) or the cytosol
can have serious consequences for the cell function. The salient role of UPR is to sense
unfolded proteins in the ER, while HSR respond from the cytosol increasing protein
folding and degradation capacity (Costa-Mattioli and Walter, 2020). UPR consists of three
main sensors: inositol-requiring protein 1 (IRE1↵), protein kinase RNA-like endoplasmic
reticulum (ER) kinase (PERK) and activating transcription factor 6 (ATF6), which
controls the folding capacity of the ER (Hetz, 2012). Early UPR, mediated by PERK-eIF2
and converging with ISR, aims to reduce translation intensity decreasing protein load
to the ER. This is accompanied by IRE1-dependent decay of mRNA and activation of
the autophagy pathway (Hetz, 2012). Then, a group of transcription factors: ATF4,
transcription factor 6 cytosolic fragment (ATF6f) and spliced X box-binding protein 1
(XBP1s) trigger transcription of adaptive genes aiming to alleviate stress and increase
protein folding capacity. If the exposure and duration of stress exceeds the cell’s ability to
restore homeostasis, the cell commits to apoptosis (Hetz, 2012).
Stress response signalling controls pro-survival and apoptotic pathways, so the role
of the stress response in cancer is complex and context specific Figure 1.10, reviewed
by Clarke et al. (2014) and Urra et al. (2016). Briefly, many oncogenic pathways drive
hyperactivation of protein synthesis making the cell susceptible to stress-induced apoptosis.
However, high levels of UPR activation are associated with malignant transformation
(Urra et al., 2016), are known to promote resistance to chemotherapy and radiotherapy
in di↵erent cancer types (Rouschop et al., 2010; Ghaddar et al., 2021), correlate with
poor prognosis in glioblastoma, breast cancer and pre-B acute lymphoblastic leukemia,
promote metabolic reprogramming in prostate cancer and immunosuppressive environment
in ovarian cancer (Clarke et al., 2014; Urra et al., 2016).
36
Figure 1.9: Overview of the cell fate decisions associated with ER stress response
The outcome of ER stress response involves adaptation and restoration of cell’s protein
folding capacity or triggering apoptosis, if the stress is prolonged or not restrained. There
are three signalling arms responsible for the initiation of ER stress response: ATF6, IRE1↵
or PERK-mediated. Of which the latter, is one of the component of the Integrated Stress
Response pathway. The net e↵ect of the ER stress activation is transcriptional upregulation
of several ER chaperones to facilitate protein folding. ER stress also include inhibition of
50cap dependent translation (through eIF2↵) and activation of ER-associated degradation
(ERAD) pathway, autophagy to remove misfolded proteins and regulated IRE1-dependent
decay (RIDD) pathway, which selectively removes mRNAs encoding proteins located in
the endoplasmic reticulum (Hetz, 2012). Above certain threshold, the cell’s ability to
tolerate ER stress may be exceeded and ER stress response signalling will trigger the
apoptosis. In such scenario, cell death is controlled by BCL2 protein family, BAX and
BID and p53-mediated signalling.
From Hetz (2012)
37
1.4 Elements of B-cell biology from the perspective
of lymphoma development
The current World Health Organisation (WHO) classification of B-cell malignancies relies
predominantly on the cell of origin of the tumour. B-cell neoplasms can be divided
into malignancies originating from either progenitor cells, mature B-cells or plasma cells
(Swerdlow et al., 2016). Overall, the majority of Non-Hodgkin Lymphomas (NHL), which
are the most common haematological malignancies in adults, are derived from the germinal
center (GC) stage of B cell development (Mlynarczyk et al., 2019; Swerdlow et al., 2016).
GC-derived lymphomas include: Di↵use Large B-cell Lymphoma (DLBCL), Burkitt
Lymphoma (BL) and Follicular Lymphoma (FL), which all account for about 60% of all
B-cell NHL.
GCs are transient microstructures, which are formed in secondary lymphoid organs
upon activation of mature naive B-cells by T-cell dependent antigens. The GC reaction is
dedicated to selecting and expanding populations of B-cells that will eventually di↵erentiate
into memory B-cells, and plasma cells. This is an essential part of a physiological adaptive
immune response against exogenous pathogens (Klein and Dalla-Favera, 2008; De Silva
and Klein, 2015). Lack of GCs observed in patients with inherited hyper-IgM syndrome
is associated with severe immunodeficiency, which highlights the importance of the GC
reaction (Etzioni and Ochs, 2004).
The GC consists of two functionally and histologically distinct zones: the dark zone
and the light zone. The dark zone consists of a dense population of large B-cells -
centroblasts, that undergo rapid proliferation and somatic hypermutation (SHM). The
aim of the SHM process is to introduce point mutations into the variable regions of the
heavy and light chains of the immunoglobulin genes (IgV), that will generate a population
of mutant subclones with a broad range of B-cell receptor (BCR) anity for the antigen.
This is achieved through expression of activation-induced cytidine deaminase (AICDA)
(Muramatsu et al., 2000; De Silva and Klein, 2015). Centroblasts then migrate to the light
zone becoming centrocytes. Centrocytes compete for survival signal from CD4 T follicular
helper (TFH) cells and Follicular Dendritic Cells (FDC) and only those demonstrating
high-anity towards the immunising antigen evade apoptosis (Nakagawa and Calado, 2021).
The positive selection (survival) stimuli is provided by B-cell receptor (BCR) capturing the
FDC-bound intact antigen, co-stimulatory receptors such as CD40 or tumour-necrosis factor
(TNF)-receptor, adhesive molecules (ICAM-1 and VCAM-1), anti-apoptotic molecules
(BAFF), and a mixture of cytokines secreted by FDC and TFH . BCR-antigen complex
is internalised and the antigen is then presented at the cell surface through MHC II
complex that interacts with CD4 T follicular helper cells. The higher the anity the more
antigen can be acquired from FDC and the stronger the stimulation received from TFH
38
is (Nakagawa and Calado, 2021). Immunoglobulin Class Switch Recombination (CSR)
is another important event in the life of GC B-cells. The mechanism of CSR involves
double stranded breakage and rejoining of the immunoglobulin genes leading to deletional
recombination of the heavy chain region, thus determining the final class of produced
antibodies (IgG, IgE or IgA). GC B-cells undergo several rounds of proliferation and
selection, re-cycling between dark and light zone through a process directed by chemokines
gradients. Selected GC B-cells di↵erentiate into first antibody-secreting plasma cells after
about one week since activation (Basso and Dalla-Favera, 2015; Sha↵er III et al., 2012).
The anity of the antibodies increases with time in the phenomenon known as ”anity
maturation”, which continues over the next weeks. An overview of the most important
steps of B-cell development and maturation were shown in the Figure 1.10.
The family of GC-derived lymphomas is highly heterogeneous comprising a broad
spectrum of clinical presentation and complex molecular context. Each lymphoma subtype
works like a distorting mirror, which resembles a certain stage of B-cell development,
sometimes referred to as cell-of-origin, but aggravates or attenuates specific physiological
mechanisms to drive a malignant phenotype. The oncogenic events may start at earlier
stages than the normal B-cell counterpart and accumulate during di↵erentiation and
maturation (Sha↵er III et al., 2012). For example, although FL resembles GC stage
of development but initial genetic corruption begins at bone marrow pro-B cell stage
(Sha↵er III et al., 2012). In addition to genetic aberrations commonly found in tumours,
such as point mutations, deletions or amplifications that activate potential oncogenes or
deactivate tumour suppressor genes, there are two GC-specific processes, SHM and CSR,
that largely contribute to the uniqueness of the lymphoma genome (Basso and Dalla-Favera,
2015; Klein and Dalla-Favera, 2008; Mlynarczyk et al., 2019). SHM and CSR occur in a
highly hazardous environment, in which DNA damage response and cell cycle checkpoints
are desensitised allowing for remodelling of the immunoglobulin loci and the expansion
of selected clones. An unwanted by-product of CSR is the risk of translocation, which
is common in GC-derived lymphomas. Chromosomal abnormalities in GC-lymphomas
usually involve translocation of a proto-oncogene into an immunoglobulin locus which is
under control of an active enhancer. This drives high-level and sustained transcription
of the translocated oncogene (Ku¨ppers and Dalla-Favera, 2001). Thus translocation of
MYC into the immunoglobulin loci, t(8,14), is a hallmark of Burkitt Lymphoma (occurs
in more than 95% of cases) (Basso and Dalla-Favera, 2015; Klein and Dalla-Favera, 2008;
Mlynarczyk et al., 2019). Other genes with recurrent translocation include BCL2 (80% of
FL patients) and BCL6 (30–40% of DLBCL).
39
Figure 1.10: Overview of B-cell development and maturation
B-cells develop in bone marrow from haematopoietic stem cells (HSC). Recombination of
the immunoglobulin loci is the first step in the generation of a mature immunoglobulin
receptor. This process starts in the pre-pro-B cell and continues to the pro-B cell stage,
leading to a heavy chain protein formation. V(D)J segments recombination requires the
activity of the recombination-activating gene (RAG) complex, which directs cleavage of
the DNA. Mature B-cells (na¨ıve B-cells) leave the bone marrow and migrate to secondary
lymphoid organs (SLOs), where they encounter an antigen. Once activated, with T-cell
dependent antigen, B-cells enter the germinal centre reaction (De Silva and Klein, 2015).
In the dark zone (DZ), activated B-cells, now centroblast, undergo proliferation and
somatic hypermutation of their immunoglobulin loci. Centroblasts migrate to the light
zone (LZ), becoming centrocytes and compete for the survival signal from CD4 T follicular
helper (TFH) cells and Follicular Dendritic Cells (FDC). During germinal centre (GC)
GC reaction B-cells switch their isotype during class switch recombination. Editing of
the immunoglobulin loci during somatic hypermutation and class switch recombination is
mediated by activation-induced cytidine deaminase (AID) activity (De Silva and Klein,
2015). GC B-cells undergo several rounds of proliferation and selection, re-cycling between
LZ and DZ. Finally, GC B-cells di↵erentiate into memory B-cells or antibody secreting
plasma cells (Mlynarczyk et al., 2019; Runte et al., 2019).
40
A massive phenotype shift from quiescent na¨ive B-cells cells to rapidly proliferating
and hypermutated GC B-cells and, finally, to antibody producing plasma cells is directed
by profound reorganisation of the gene expression programme. Transcription factors with
well established links to GC formation and lymphomagenesis include MYC and BCL6.
In the physiological GC reaction, the proto-oncogene MYC has a bimodal pattern of
expression: it is induced early during GC reaction initation and then transiently in the LZ
B-cells undergoing positive selection and DZ re-entry (Calado et al., 2012; Dominguez-Sola
et al., 2012; Basso and Dalla-Favera, 2015). Expression of MYC primes B-cells for massive
proliferation and clonal expansion. BCL6, known as a master regulator of GC-reaction,
is key to GC formation and maintenance (Basso and Dalla-Favera, 2015). It acts as a
transcriptional repressor inhibiting DNA damage response, apoptosis and premature B-cell
activation and terminal di↵erentiation into plasma cells (Basso and Dalla-Favera, 2015).
BCL6 controls the expression of several genes known to drive tumour formation, such as
cell cycle checkpoints (CDKN1A, CDKN1B), DNA damage response regulators (ATR,
TP53) and di↵erentiation factors (eg. IRF4 and PRDM1) (Ci et al., 2009; Basso and
Dalla-Favera, 2015). Desensitised DNA damage response by BCL6 is necessary to sustain
immunoglobulin loci remodelling and anity maturation. The expression of BCL6 is
promoted during the initation of GC reaction by IRF8 and MEF2B, OCT2 and OCA-B
and then downregulated at the GC exit stage (Song et al., 2021).
Despite tremendous progress that has been made in our understanding of processes
governing GC reaction and their links to lymphomagenesis, translation of these findings
into concise subtype classification systems and then further into clinical practice has
been challenging. Of all GC-derived lymphomas, DLBCL has been the most dicult to
subdivide due to its considerable molecular heterogeneity. After years of attempts to
classify DLBCL morphologicaly with little success, the first modern classification utilising
the technology of next generation sequencing was the Cell Of Origin system (Alizadeh
et al., 2000), which distinguished two DLBCL subtypes based on gene expression profile.
Germinal centre B-like DLBCL (GCB-DLBCL) group resembling normal germinal centre
B-cells and activated B-like DLBCL (ABC-DLBCL) similar to activation of peripheral
blood B cells.
In addition to gene expression profile, certain genetic hallmarks of B-cell lymphoma were
also recognised. Identification of BCL2, MYC and/or BCL6 chromosomal rearrangements
has led to distinguishing double/triple hit high grade B-cell lymphoma. This finding was
further complemented by discovery of distinct transcriptomic signatures of MYC-driven
GCB-DLBCL, that overlapped with double hit lymphoma cases, referred to as Molecular
High-Grade DLBCL or DHSig, respectively (Sha et al., 2019; Ennishi et al., 2019).
More recently three genomic studies from Harvard (Chapuy et al., 2018), National
41
Cancer Institute (NCI) (Schmitz et al., 2018) and the UK Haematological Malignancy
Research Network (HMRN) (Lacy et al., 2020) identified converging genetic subclasses
based on the profile of mutations. The current status molecular profiling of DLBCL has
been summarised in the Figure 1.11 and covered in detail by Cutmore, Krupka, Hodson
(2022)
1.5 Role of translation in B-cell development and ma-
lignancy
Although GC reaction and lymphomagenesis have been extensively studied at the level
of transcription, much less is known about post-transcriptional regulation of B-cells
development and maturation. Deregulation of mRNA translation is common in B-cell
lymphoma and typically involves 1) changes in the expression of the core components of
translation machinery or pathways regulating its activity, 2) microRNAs-driven regulation
of protein synthesis, and 3) translational control of key lymphoma oncogenes or tumour
suppressors (Horvilleur et al., 2010).
Aberrant MYC or mTOR signalling are the most studied examples of pathways
a↵ecting the proteins synthesis in B-cell lymphomas. MYC translocation and mutations
are hallmarks of Burkitt Lymphoma and Molecular High-Grade (and Double-Hit) DLBCL
subtype (Schmitz et al., 2012; Sha et al., 2019), while mTOR activation is usually associated
with activation of B-cell receptor (BCR) signalling, PTEN loss (about 14 % of DLBCL),
or mutations in ERK/RAS pathway. (Chapuy et al., 2018; Reddy et al., 2017; Lacy et al.,
2020; Pfeifer et al., 2013)
The oncogenic activity of MYC includes hyperactivation of protein synthesis, but it is
also dependent of the translation machinery. Haploinsuceincy in Rpl24 was shown to ab-
rogate MYC-induced transformation in vivo, while maintaining normal B-cell development
(Barna et al., 2008). This e↵ect was specific to MYC-driven tumours as in Tp53/ mice,
the frequency and latency of tumours was not a↵ected by Rpl24+/ (Barna et al., 2008).
Simultaneous deregulation of c-MYC and eIF4E expression accelerates the development of
B-cell lymphoma in transgenic mice. The molecular mechanism of this cooperation results
from eIF4E’s ability to suppress MYC-induced apoptosis by promoting cellular senescence
(Ruggero et al., 2004). It is interesting in the light of another study, where overexpression
of eIF4E in lymphoma was associated with chemotherapy resistance (Wendel et al., 2004).
Another important pathway converging at regulating protein synthesis is BCR signalling,
which is vital to the survival of normal and malignant B-cells. Depending on whether it
requires antigen engagement or not, BCR signalling in lymphoma is termed chronic active
or tonic.
42
Figure 1.11: Topography of DLBCL molecular subtypes
A conceptual representation of the genetic classification of DLBCL. Coloured hills depict
known genetic subtypes. The genes most commonly altered are indicated in white. Each
subtype is labelled with LymphGen/NCI cluster name (red) (Schmitz et al., 2018), HMRN
(black) (Lacy et al., 2020) and Harvard (blue) equivalent (Chapuy et al., 2018). Briefly,
the MCD/C5/MYD88 subtype covers DLBCL cases with poor prognosis driven by
mutations in B-cell receptor, Toll like receptor (TLR) and NFB pathway accompanied by
immune evading mutations. The EZB/C3/BCL2 group distinguishes cases with recurrent
BCL2, EZH2, CREBBP, KMT2D mutations sharing the pattern with follicular lymphoma
(FL). Interestingly, based on the gene expression signature or MYC amplification status,
LymphGen and HMRN characterise an extra subtype of termed EZB-MYC/MHG. Next,
the BN2/C1/NOTCH2 subtype includes mutation of NOTCH2, TNFAIP3, BCL10,
NFKBIZ. The ST2/C4 group is characterised by activation of the JAK/STAT/ERK
pathway and mutations in SOCS1, DUSP2, STAT3 and BRAF. The HMRN systems
additionally distinguishes two subgroups within ST2/C4, based on SGK1 and SOCS1
mutation status. Lastly, the N1/NOTCH1 subtype is defined by activating mutations
of NOTCH1 gene and the A53/C3 subtype is characterised by enrichment for TP53
mutations and aneuploidy. Whilst patients position on top of the coloured hills will be
reproducibly classified by each classification system, patients in valleys may be unclassified
or classified alternatively across di↵erent classification systems. Uncoloured hills correspond
to unknown DLBCL subtypes which may emerge in the future from currently unclassified
cases. See Cutmore, Krupka, Hodson (2022)
43
Whereas the pro-survival e↵ect of the antigen-independent tonic BCR signalling is
dependent on the activated PI3K/AKT pathway, the antigen stimulated chronic active
BCR signalling engages multiple pathways including PI3K and NF-B (Sha↵er III et al.,
2012). Tonic BCR activity is characteristic to BL and GCB subtype of DLBCL, while the
chronic active BCR signalling is characteristic to ABC-DLBCL. BCR-mediated mTOR
activation promotes both: global 50cap dependent translation initation and preferential
translation of selected transcripts through increased translation factors activity, see section
1.2.1 and 1.3.1 .
Although, core components of the translation machinery are rarely mutated in B-cell
lymphoma (usually in less than 10% of cases), their expression is frequently abnormal
(Taylor et al., 2020). The analysis of mRNA expression of 16 eIFs showed that 12 out of
16 are overexpressed in DLBCL compared to normal tissue samples (Unterluggauer et al.,
2018). Deregulated activity of translation factors may drive pro-oncogenic programmes
of translation. For example, overexpression of eIF4B promotes translation of proteins
associated with tumour cell survival (DAXX, BCL2 and ERCC5) and correlates with poor
clinical outcome (Horvilleur et al., 2014). Another study has shown that translation of key
lymphoma oncogenes, such as MYC, BCL6, BCL2 is under the control of USP11, which
is recruited to the translation initiation complex where it deubiquitinates and stabilises
eIF4B (Kapadia et al., 2018).
In addition to the core components of the translation machinery, there are other RNA-
binding proteins that, when mutated or deregulated, may promote lymphoma development
and progression. Of these a DEAD-box helicase 3, X-linked (DDX3X) has attracted much
attention recently as it been found recurrently mutated in Burkitt Lymphoma (Grande
et al., 2019; Schmitz et al., 2012), Chronic Lymphocytic Leukaemia (CLL) (Ojha et al.,
2015; Takahashi et al., 2018), medulloblastoma (Jones et al., 2012; Pugh et al., 2012;
Robinson et al., 2012), head and neck squamous cell carcinoma (Stransky et al., 2011)
and NK-T cell lymphoma (Jiang et al., 2015). The role of DDX3X has been extensively
studied in medulloblastoma (Jones et al., 2012; Pugh et al., 2012; Robinson et al., 2012;
Patmore et al., 2020; Samir et al., 2019), where it has been classified as tumour suppressor
gene regulating the activity of Wnt signalling. DDX3X function in other cancers, including
lymphoma, is less understood as it has been recognised as both tumour suppressor or
oncogene depending on the cancer type, what is not surprising given ubiquitous role of
DDX3X in RNA biology. DDX3X roles include regulation of transcription, splicing, nuclear
export, stress granule formation and resolution, microRNA biogenesis, mRNA translation
and decay (Linder and Jankowsky, 2011; Mo et al., 2021). It is entirely unknown which of
these functions may be relevant to lymphoma.
44
Table 1.1: Examples of translationally regulated genes relevant to mature B-cell lymphomas,
from Taylor et al. (2020)
Gene(s)
Protein
function
Association with
B-cell neoplasm
Evidence for
translational control
Ref
MYC Cell growth
t(14,8) translocation
in BL
Recurrent mutations
in MHG/Double-Hit
DLBCL
Increased mRNA
translation following
BCR stimulation
Inhibition of
eIF4A reduces
MYC expression in
lymphoma cell lines
Inhibition of
eIF4E reduces MYC
mRNA translation
and nuclear export
in DLBCL cell lines
(Yeomans et al., 2016)
(Culjkovic-Kraljacic et al., 2016)
(Steinhardt et al., 2014)
(Schatz et al., 2011)
(Wilmore et al., 2021)
MCL1
Cell survival
(anti-apoptotic)
Expression induced
following BCR
stimulation
Inhibition of
eIF4A reduces
MCL1 expression in
lymphoma cell lines
(Schatz et al., 2011)
BCL2
Cell survival
(anti-apoptotic)
t(14;18) translocation
in FL
Recurrent translocations
in GCB-DLBCL and
double/triple hit
DLBCL
Inhibition of
eIF4E reduces BCL2
mRNA translation
and nuclear export
in DLBCL cell lines
(Culjkovic-Kraljacic et al., 2016)
CCND1
Cell cycle
progression
t(11;14) translocation
in mantle cell
lymphoma
Inhibition of
eIF4A reduces CCND1
expression in
lymphoma cell lines
(Schatz et al., 2011)
BCL6
Transcriptional
repression
t(14;18) translocation
in double/triple hit
lymphoma
Inhibition of
eIF4E reduces BCL6
mRNA translation
and nuclear export
in DLBCL cell lines
(Culjkovic-Kraljacic et al., 2016)
CARD11
BCL10
MALT1
CBM complex
components
which control
NF-B activation
Recurrent mutations in
subset of ABC-DLBCL
Mediates NF-B
activation downstream
of BCR
Inhibition of
eIF4A reduces CBM
complex expression
in lymphoma cell lines
(Steinhardt et al., 2014)
45
Another interesting topic is aberrant expression of microRNA (miRNA) in B-cell
malignancies. MiRNAs are small (about 22 nucleotides) non-coding RNAs that interact
with a group of mRNA, through fully or partially complementary sequence in the 30UTR,
regulating their translation activity and degradation, see Filipowicz et al. (2008) for a
review. Expression of specific miRNAs signatures in DLBCL predict lymphoma subtype,
event-free survival and the risk of transformation from indolent follicular lymphoma to
DLBCL (Lawrie et al., 2009; Malumbres et al., 2009; Li et al., 2009a). Characteristic
miRNAs patterns have been also observed for Burkitt Lymphoma (Leucci et al., 2008),
Hodgkin Lymphoma (Navarro et al., 2008) and Mantle Cell Lymphoma (Zhao et al.,
2010). Interestingly, even for the same disease, little overlap is observed in miRNAs
expression pattern between individual studies. miRNAs identified as important for one
type of lymphoma do not necessarily have similar e↵ect in another, or their role might be
ambiguous (Horvilleur et al., 2010). The role of individual miRNA in B-cell malignancies
has been reviewed by Sole et al. (2016).
Finally, the expression level of several genes relevant to B-cell malignancies can be
regulated at the level of translation, detailed in Table 1.1.
1.6 Toolkit to study heterogeneity of translation
”A small particulate component of the cytoplasm” was the title of the paper, where
George E. Palade revealed the ribosome to the world (Palade, 1955). Although the
process of translation has been known since 1955 and its biochemical and molecular
nature was comprehensively studied over the last decades, the analysis of the dynamics
of these ”small particles”, especially from a genome-wide perspective, is still challenging.
Technical diculties associated with studying translation are related to the necessity of
studying simultaneously two chemically distinct types of molecules: RNA and protein. The
translation process refers to the very short and dynamic moment when the two families of
molecules physically interact and one (RNA) catalyses the synthesis of another. Therefore
measuring the concentration of total transcript or protein abundance does not reflect true
translation intensity as many factors, not directly related to protein synthesis, may a↵ect
the abundance of both. A simple workaround is to use special separation or labelling
techniques to extract subpopulation of mRNA or protein that are functionally related to
translation. At the protein level, it will be a pool of newly synthesised proteins, ideally
nascent proteins that are still bound by the ribosome. At the transcript level, the evidence
of physical interaction with the translation machinery can be used as a proxy for active
translation.
Another challenge is translation heterogeneity, which involves not only di↵erences
in translation intensity between single cells and tissues, but also varying translation
46
programmes of single or groups of transcripts. This phenomenon addresses the problem of
the scale and resolution of translation quantification.
Available methods can be classified by the number of transcripts or proteins measured
simultaneously into low-throughput and high-throughput, by the level of insight, e.g. bulk,
single cell or cell compartments or by resolution - single nucleotide, transcript wide or
global. The choice of the experimental technique is dictated by the biological question
and the level of throughput required to answer it.
Net-abundance of each protein is a function of both - synthesis and degradation rate,
so that it is safe to assume that only the pool of newly synthesised proteins can mirror
the translation rate (Iwasaki and Ingolia, 2017). Because of relatively large content of
pre-existing proteins in the cell, special separation or labelling techniques are needed to
quantify this subpopulation.
Earlier methods involved radioisotope labelling of newly synthesised proteins, for
example with 35S-methionine or cysteine. Nowadays, non-radioactive, luminescent labelling
is more popular. The most common techniques utilise aminoacyl tRNA analogues, such
as azidohomoalanine (AHA) or alkyne-bearing homopropargylglycine (HPG) (Iwasaki
and Ingolia, 2017), which are incorporated into the C-terminal end of nascent proteins.
The disadvantage of using amino-acids analogues is methionine-depletion step, which can
induce stress response and disrupt a delicate translation pattern. This does not apply to
puromycin analogue - alkyne-bearing puromycin (O-propargyl-puromycin; OPP), which
induces premature translation termination and release of nascent proteins with OPP
attached. Labelled proteins, depending on the exact assay, can be detected by specific
antibodies or secondary labelled with fluorophore by CLICK chemistry (Aviner, 2020).
The protocol can be conjugated with several well-established techniques, such as confocal
microscopy, fluorescence-activated cell sorting (FACS) or classical immunohistochemistry
(Liu et al., 2012; Iwasaki and Ingolia, 2017). Methods involving nascent proteins labeling
allows the measurement of bulk protein translation, or, depending on the exact protocol,
provides single-cell or event cell compartment resolution.
High-throughput techniques have been dominated by two approaches: mass-spectrometry
(MS) or mRNA sequencing based. The first usually combines some type of labelling to
isolate newly synthesised proteins with subsequent measurement with tandem MS. Pulsed
stable isotope labeling by amino acid in cell culture (pSILAC), which is the most popular
technique, involves supplementation of the growing media with amino-acids labelled with
stable isotopes only for a brief moment. Proteins with incorporated heavy amino-acids
(newly synthesised) can be easily distinguished from pre-existing light proteins by their
mass. The abundance of heavy labelled proteins reflect the relative intensity of protein
47
synthesis (Iwasaki and Ingolia, 2017). Other MS techniques, such as BONCAT, QuaN-
CAT or PUNCH-P, relies on isolation of labelled proteins with streptavidin beads. In
bio-orthogonal noncanonical amino acid tagging (BONCAT), AHA-labelled proteins are
tagged with biotin. A variation of BONCAT, quantitative noncanonical amino acid tag-
ging (QuaNCAT), is a combination of BONCAT and SILAC allowing to quantify early
changes of protein synthesis with higher specificity and greater depth than standard BON-
CAT (Howden et al., 2013). Finally, in puromycin-associated nascent chain proteomics
(PUNCH-P) (Aviner et al., 2013), translating ribosomes are isolated from the cell with
ultracentrifugation, next, nascent peptides are labelled with biotin-dC-puromycin, pulled
down with streptavidin beads and analysed by MS (Aviner, 2020). An advantage of MS-
based approaches is their accuracy and high-throughtput. Unfortunately, quantification of
the protein abundance with MS has limited dynamic range and capacity to identify novel
products, isoforms or proteins with post-translational modifications (Iwasaki and Ingolia,
2017).
Polysome fractionation and ribosome footprinting (a.k.a Ribosome Profiling or Ribo-
Seq) are two main high-throughput strategies to analyse translation from the individual
transcript perspective (Figure 1.12). Both are almost equally popular nowadays, but
their technical protocols di↵er substantially and thereby biological interpretation of the
results. Ribo-Seq and polysome fractionation work under the assumption that 1) actively
translated mRNAs are physically associated with the ribosomes, and 2) the eciency
of translation is proportional to the number of ribosomes bound to mRNAs (Piccirillo
et al., 2014). In polysome fractionation, cell lysates containing polysomes (mRNAs with
multiple ribosomes), monosomes (mRNAs with a single 80S ribosome), and free ribosomal
subunits are centrifuged over a sucrose density gradient, which allows for separation
of transcripts according to the number of ribosomes attached. This is referred to as a
polysome profiling step. Each fraction can be distinguished by change in optical density
(absorbance) and subjected to analysis. The composition of it can be evaluated in terms of
the individual mRNA or protein abundance. For mRNA-centric studies, cDNA microarray
(historical approach), RNA-seq (current approach) or RT-PCR (low-throughput) can be
used (Piccirillo et al., 2014; Johannes et al., 1999; Karginov and Hannon, 2013). Typically,
mRNAs translated more eciently will be enriched in the heavy polysome fraction (more
than three ribosomes per mRNA) (Piccirillo et al., 2014; Johannes et al., 1999), while less
translated mRNA will be more abundant in the light fraction, which usually corresponds to
polysomes with three or less ribosomes and the monosome fraction. For better resolution,
each fraction can be quantified separately.
Ribo-Seq is centred on the idea of ribosome footprinting, developed initially by
Wolin and Walter (1988), which allows to determine the position of a ribosome with
single-nucleotide precision (Ingolia et al., 2009). Ribosomes bound to a mRNA protect a
48
fragment (about 30 nucleotides) from RNAse digestion (ribosome-protected fragments,
RPFs). By utilising the advances in next-generation sequencing, it is possible to sequence
those RPFs, thus, to infer the position of each translating ribosome. The information about
the number and the location of ribosomal footprints can be used to compute ribosome
density for a single transcript, which will be a proxy for its translation intensity (Ingolia
et al., 2009). In contrast to Ribo-Seq, polysome profiling-based approaches do not allow
to analyse the precise location of attached ribosomes. It might be enough to judge which
transcripts forms polysomes (are actively translated), but no information about the exact
location and reading frame are available. On the other hand, polysome fractionation
maintains the entire transcript intact, which may be advantageous for studying the e↵ect
of di↵erential splicing or polyadenylation on translation. Polysome fractionation combined
with RNA-Seq reports the total number of ribosomes per mRNA, which depending on
the fraction in which a transcript was found, while Ribo-Seq returns the position of every
ribosome (regardless of the polysome/monosome fraction of the entire transcript). Hence,
Ribo-Seq allows for quantification of the relative ribosome density per transcript, whereas
polysome fractionation-RNA-Seq gives an estimation of the absolute number of ribosomes.
This may prove less ecient when studying translation landscape following large shifts
in global translation, but no systematic comparison has been performed so far (Piccirillo
et al., 2014) Finally, with Ribo-Seq it is possible to annotate de novo, per sample, directly
from the data, which regions of the genome are actively translated at subcodon level, see
chapter 5, which is not possible with polysome profiling.
In a typical Ribo-Seq experiment, treatment with a translation inhibitor briefly
before harvesting aims to freeze the translating ribosomes preventing them from dissociating
from the transcript and maintaining the fidelity of the ribosome footprint localisation.
The combination of ribosomal footprinting with di↵erent translation inhibitors, targeting
di↵erent stages of translation, can broaden the biological application of the Ribo-Seq
data. Cycloheximide (CHX), used in the original Ribo-Seq protocol (Ingolia et al., 2009;
McGlincy and Ingolia, 2017), is the most popular choice. CHX is a small molecule
inhibitor blocking eukaryotic translation elongation by binding to the ribosome E-site
and blocking eEF2-mediated translocation (Obrig et al., 1971; Schneider-Poetsch et al.,
2010; de Loubresse et al., 2014). In contrast, two other popular inhibitors: lactimidomycin
(LTM) and harringtonine (HARR), require empty E-site, thus inhibit specifically only
the first round of elongation capturing the ribosomes at the translation initiation site
(Iwasaki and Ingolia, 2017; de Loubresse et al., 2014). Ribo-Seq combined with HARR
or LTM treatment allows to build a genome-wide map of start codons and is known as
global translation initation site sequencing (GTI-Seq) (Iwasaki and Ingolia, 2017). The
ribosomes not blocked by the translation inhibitor dissociates from the transcripts leaving
only the initiation site footprints for sequencing. A variation of this protocol, quantitative
49
translation initiation sequencing (QTI-seq), involves sequential treatment with LTM and
puromycin, which facilitates run-o↵ of non-initiating ribosomes increasing resolution of
translation initiation site localisation (Gao et al., 2015). The full spectrum of ribosome
footprinting-based techniques has been reviewed by Iwasaki and Ingolia (2017)
Figure 1.12: Comparison of polysome fractionation and Ribo-Seq technique
Polysome fractionation and Ribo-Seq allow analysing the translation intensity of individual
transcripts. Polysome fractionation is based on the technique of polysome profiling, where
a sucrose density gradient separates polysomes, monosomes and individual ribosomal
subunits. Each polysome fraction contains the transcripts with a di↵erent number of
ribosomes attached. By employing a next generation sequencing technology (e.g. RNA-
Seq), it is possible to infer which transcripts are abundant in each fraction. In contrast,
in the Ribo-Seq workflow, transcripts are digested with RNAse generating a mixture of
mRNA fragments protected from RNAse digestion by attached ribosomes. These, so-called
ribosomal footprints can be analysed with next generation sequencing technique.
50
1.7 Project aims
The aim of this thesis is to explore the role of translational control during lymphoma
development.
Firstly, I established an automated bioinformatic pipeline for ecient processing of Ribo-
Seq datasets which allows to keep the computational workflow transparent, reproducible
and flexible. Next, I reviewed and benchmarked current methods developed for identifying
genes regulated at the level of translation. This allowed me to select a strategy to elucidate
that overexpression of two B-cell oncogenes, BCL6 or MYC, is followed by preferential
translation of selected transcripts.
Then, I focus on the role of RNA-helicase (DDX3X) in MYC-driven lymphoma. I reveal
that that loss-of-function mutations in DDX3X are common in MYC-translocated B-cell
lymphomas facilitating early tumour development. By controlling tranlation of selected
transcripts, mutated DDX3X bu↵ers the e↵ects of MYC on translation of ribosomal
proteins and the rate of global protein synthesis.
Finally, I analyse a large dataset of 79 Ribo-Seq libraries to investigate a genome-
wide distribution of translating ribosomes and the extent of non-canonical translation in
lymphoid cells. I show pervasive translation of ostensibly non-coding region, and design a
knock-down CRISPR screen library to identify those important for B-cell survival.
51
52
CHAPTER 2
Materials and methods
2.1 Materials
2.1.1 Overview of genomic sequences and annotations used in
this study
Resource Source Purpose
Gencode v.29 Frankish et al. (2019) Gene and transcript models
GRCh38 Genome Reference Consor-
tium (FASTA file down-
loaded from Gencode v.29)
Nucleotide sequence of the GRCh38
primary genome assembly
H.sapiens rRNA RefSeq (FASTA files) Pre-alignment of Ribo-Seq reads
Reactome Joshi-Tope et al. (2005) Pathway knowledgebase
MSigDb Subramanian et al. (2005) Molecular Signatures Database for
GSEA
sORFdb Olexiouk et al. (2018) Proteogenomic database
OpenProt Brunet et al. (2021) Proteogenomic database
53
2.1.2 Overview of external datasets used in this study
Accession ID Source Description
GSE125966 McCord et al. (2019a) RNA-Seq from GOYA clinical trial
GTEx Consortium et al. (2020) RNA-Seq samples from 54 tissue sites
in non-diseased individuals
TCGA Weinstein et al. (2013) Pan-cancer RNA-Seq dataset
CGCI-BLGSP-
2019
Grande et al. (2019) RNA-Seq from Burkitt Lymphoma pa-
tients
GSE35163 Schmitz et al. (2012) RNA-Seq from Burkitt Lymphoma pa-
tients
EGAS00001003560 Caeser et al. (2019) RNA-Seq from Primary Germinal Cen-
ter B-cells
COSMIC Tate et al. (2019) Database of somatic mutations in can-
cer
MSV000084172 Sarkizova et al. (2020) Mono-allelic MHC-I peptidome from
B721.221 cell line
PXD000332 Deeb et al. (2014) N-glyco FASP and super-SILAC mass
spectrometry from lymphoma patients
PXD002004 Johnston et al. (2018) TMT 10-plex mass spectrometry from
CLL patients and peripheral B-cells
PXD002098 Deeb et al. (2012) Super-SILAC dataset from lymphoma
cell lines
PXD004452 Bekker-Jensen et al. (2017) Deep-proteome dataset of HeLa cells,
tissue samples (colon, prostate, liver)
and 5 human cell lines
PXD004746 Khodadoust et al. (2017) MHC-I and MHC-II peptidomes from
lymphoma patients
PXD010808 Khodadoust et al. (2019) MHC-I peptidome from lymphoma pa-
tients
54
2.1.3 Overview of computational software used in this study
Tool Source Purpose
FastQC Andrews et al. (2017) Quality check: FASTQ files
Cutadapt Martin (2011) Adapter trimming
Bowtie2 Langmead and Salzberg
(2012)
Prealignment of Ribo-Seq to rRNA
STAR Dobin et al. (2013) Sequencing reads alignment
MultiQC Ewels et al. (2016) Summary of QC and alignmnt reports
Samtools Li et al. (2009b) BAM files indexing
GenomicFeatures Lawrence et al. (2013) Manipulating genomic locations
Samtools Morgan et al. (2016) Processing of SAM or BAM files
DESeq2 Love et al. (2014) Di↵erential expression analysis
edgeR Robinson et al. (2010) Di↵erential expression analysis
Deeptools Ramı´rez et al. (2016) Metagene analysis
clusterProfiler Yu et al. (2012) GO and GSEA analysis
GATK Van der Auwera and
O’Connor (2020)
SNV calling from RNA-Seq data
Picard Broad Institute BAM files processing
GSEA Subramanian et al. (2005) GSEA analysis
ORFLine Hu et al. (2021) ORF identification from Ribo-Seq data
ORF-RATER Fields et al. (2015) ORF identification from Ribo-Seq data
RiboCode Xiao et al. (2018) ORF identification from Ribo-Seq data
RibORF Ji (2018) ORF identification from Ribo-Seq data
Salomon Patro et al. (2017) Isoforms quantification
UCSC toolkit Kent et al. (2002) Processing of BigWig files
Chop-chop Labun et al. (2019) gRNAs scoring and selection
MaxQuant Cox and Mann (2008) Mass spectrometry data analysis
Comet Eng et al. (2013) Mass spectrometry data analysis
NewAnce Chong et al. (2020) Mass spectrometry data analysis
ProteoWizard Chambers et al. (2012) Mass spectrometry data analysis
Xtail Xiao et al. (2016) Di↵erential translation analysis
Riborex Li et al. (2017) Di↵erential translation analysis
anota Larsson et al. (2011) Di↵erential translation analysis
anota2 Oertlin et al. (2019) Di↵erential translation analysis
deltaTE Chothani et al. (2019) Di↵erential translation analysis
Ribowaltz Lauria et al. (2018) Ribo-seq QC
55
2.2 Methods
2.2.1 Next Generation Sequencing library preparation and se-
quencing
2.2.1.1 RNA-Seq
Performed by Dr. Jie Gao
Total RNA was extracted using NucleoSpin RNA extraction kit (Machery-Nagel, Cat
No. 740955.250) according to manufacturer’s protocol. 500 ng of total RNA was used to
prepare RNA-seq libraries using NEBNext Poly(A) mRNA magnetic isolation module (Cat
No. NEB E7490) as per the manufacturer’s instruction. Final libraries were amplified by
PCR for 12 cycles, purified with AMPure XP beads and analyzed by Agilent Bioanalyser
before sequencing on an Illumina Hi-seq4000.
2.2.1.2 Ribo-Seq
Performed by Dr. Jie Gao
Ribosome profiling was conducted as previously described (Ingolia et al., 2012) with
minor modifications. 5 million cells per sample were treated with 100 µg/ml of cyclo-
heximide and immediately centrifuged and lysed in 300 µl of bu↵er containing 20 mM
Tris-HCl, pH 7.4, 150 mM NaCl, 5 mM MgCl2, 1% NP40, 1 mM DTT and 100 µg/ml
cycloheximide. 100 µl of the lysate were reserved for RNA-Sequencing and the rest treated
with DNase I and RNase I. Ribosome monomers were purified using Microspin S-400
columns. The ribosome protected RNA fragments (RPF) were extracted using RNA
clean up and concentration kit (23600, Norgen Biotek). RPF were resolved in 15% Novex
TBE-Urea gels, stained with SYBR gold and fragments with 26-34 nt were excised from the
gel. The RNA was extracted from the gel by electrophoresis using D-tube (MWCO 3.5 kDa,
71506-3, Merck Chemical). Precipitated RNA was dephsphorylated by T4 polynucleotide
kinase and ligated to universal miRNA Cloning Linker (NEB) using T4 RNA ligase 2
truncated. cDNA was prepared with SuperScript III and reverse transcription primer
containing a degenerate 5-nucleotides molecular barcode sequence. The cDNA was resolved
by polyacrylamide gel electrophoresis, excised and extracted by dialysis D-tube (71504-3,
Merk). The extracted cDNA was then circularized and PCR amplified. The final library
was separated from PCR primers by electrophoresis and extracted by dialysis using D-tube
(71504-3, Merk) before sequencing on an Illumina Hi-Seq4000 as 50 nt single-end reads.
56
2.2.2 Processing and quality control of Next Generation Se-
quencing data
2.2.2.1 Adapter trimming and alignment to the reference genome
Raw FASTQ files were stripped of adapter sequence using Cutadapt. Reads shorter than
15 nucleotides were discarded. After quality check with FastQC 0.11.5, Ribo-Seq reads
were additionally filtered by rRNA using Bowtie2 2.3.4 with seed length 15. The remaining
reads were then mapped to the human genome (GRCh38) using STAR 2.5.4a with default
parameters. The reference human rRNA index was constructed from RefSeq database.
STAR genome index was built with GENCODE v.29 comprehensive gene annotation set.
2.2.2.2 Read counting
Trimmed gene models were built using GENCODE v.29 comprehensive gene annotation
set with GenomicFeatures R package. The first 30nt and last 30nt of each CDS region
were removed to reflect translation elongation intensity, as described previously. Trimmed
CDS models representing di↵erent transcript isoforms were merged by gene. The number
of footprints for 50UTR and 30UTR was obtained using a similar logic: the UTRs regions
were trimmed by 5 nucleotides adjacent to the annotated start (for 50UTR) or stop codon
(for 30UTR). As is expected from Ribo-seq, 28-30 nt long fragments with evident triplet
nucleotide periodicity relative to start and stop codon were the most abundant and were
selected for further analysis. Localisation of ribosomal P-site was determined by o↵set
between 5’end of fragments spanning translation start site and annotated start codon,
which was 12 nt for read lengths of 28-30 nt. Per sample per gene P-sites counts matrices
were build using Genomic Ranges R package allowing for assignment of a read to more
than one overlapping features. At least 25 overlapped bases were required to assign a read
to a gene. Corresponding RNA-Seq samples were counted using the same gene models.
Genes with low counts were filtered out with a threshold of minimum 128 counts
2.2.3 Di↵erential translation analysis of Ribo-Seq data
Di↵erential translation analysis was performed as previously described (Sendoel et al., 2017;
Hsieh et al., 2012). Briefly, di↵erentially expressed genes were identified using DESeq2.
RNA-Seq and Ribo-Seq derived read counts were analysed separately. Log2 fold change in
translation eciency (TE) was computed by subtracting Log2 fold change in mRNA from
log2 fold change in ribosome footprints abundance. Both values were obtained from the
DESeq2 output.
The following rules were applied to define di↵erentially translated genes:
57
1. Statistically significant change in ribosomal footprints abundance (evaluated with
standard DESeq2 workflow), FDR < 0.1,
2. Absolute mean log2 fold change in mRNA abundance < 0.3,
3. Absolute mean log2 fold change in TE abundance > 0.3.
In addition, two other gene regulatory classes were identified dependent on the rela-
tionship between Ribo-Seq and RNA-Seq measured expression changes. For genes with
statistically significant change in mRNA abundance (FDR < 0.1 and absolute log2 fold
change in mRNA abundance > 0.3) and concordant change in Ribo-Seq (mean log2 fold
change in ribosomal footprints abundance > 0 or < 0 depending on the direction of change
and regardless of the statistical significance) are classified as ‘Homodirectional’. When the
direction of change is opposite, the genes are classified as either ‘bu↵ered mRNA up’ or
‘bu↵ered mRNA down’.
2.2.4 Di↵erential expression and downstream analysis of RNA-
Seq data
2.2.5 Metagene analysis of iCLIP and Ribo-Seq data
A metagene analysis for scaled density of Ribo-Seq reads or iCLIP hits relative to start
and stop codon was performed using deeptools. The coverage of sequencing reads was
normalized per sample by the total number of uniquely mapped reads (CPM) excluding sex
chromosomes. Scaled coverage per each transcript was computed using computeMatrix func-
tion with parameters as follows: scale-regions -m 10000 -bs 20 -m 5000 -b
3000 -a 3000 -p 40 --metagene --exonID CDS --transcriptID tran-
script --skipZeros. Transcripts models were built using GENCODE v.29 basic
gene annotation set.
2.2.6 Di↵erential expression analysis of RNA-Seq data
Uniquely mapped reads were assigned to genes using featureCounts function from Rsubread
package allowing for assignment of a read to more than one overlapping features. At
least 25 overlapped bases were required to assign a read to a gene. Di↵erential expression
analysis was performed using standard workflow from DESeq2 package.
2.2.7 Downstream data analysis
Identified up-regulated and down-regulated genes were used to perform gene ontology
analysis with enrichGO function from clusterProfiler package. Enrichment scores were
58
computed using in-house scripts by taking the ratio between the number of di↵erentially
expressed genes overlapping with a gene ontology set and the number of background genes
assigned to this gene set. Pathway analysis was performed using browser-based Reactome
Pathway Database. A list of all expressed genes detected in RNA-Seq was used as a
background set for over-representation testing. Gene Set Enrichment Analysis (GSEA)
was performed using the GSEA function from clusterProfiler package. Gene expression
measurements were normalized using variance stabilising transformation, as implemented
in DESeq2 package, and analysed for enrichment in hallmark gene set from MSigDB v. 7.0
2.2.7.1 Individual-nucleotide resolution UV crosslinking and immunoprecip-
itation (iCLIP)
iCLIP libraries were performed by Dr Chun Gong in the Hodson lab using two lymphoma
cell lines as well as primary human germinal centre B cells purified from donor tonsil.
iCLIP data preprocessing (alignment, trimming and peaks calling) was performed by
Dr Igor Ruiz De Los Mozos using iMaps web server. The downstream analysis and data
visualisation was performed by me. Briefly, Unique Molecular Identifiers were used to
distinguish and remove PCR duplicates before removing experimental barcodes and Solexa
adapters. The trimmed reads were mapped to GENCODE GRCh38 v.29 using STAR
with default parameters. First nucleotide after the UMI was assigned as the crosslink site
defined by the truncated cDNA. Crosslink significant sites were determined by the iCount
peaks finding algorithm (False Discovery Rate < 0.05), by weighting the enrichment of
crosslinks versus shu✏ed random positions. Neighbouring cDNA start position less than
15 nt apart were joined to form high confidence crosslink clusters with iCount clusters
function. Genes with more than 4 cross-linking peaks in at least one experiment were
considered as valid DDX3X targets.
2.2.8 Identification of DDX3X mutations from RNA-Seq data
I used published RNA-Seq data to identify cases with mutation of the DDX3X gene.
Single Nucleotide Variant (SNV) calling was performed according to GATK guideline
using paired-end RNA-Seq data from diagnostic biopsies of 553 patients enrolled in the
GOYA trial - a phase III of 1st line treatment in DLBCL. Briefly, RNA-Seq paired-end
reads were mapped to the reference genome using STAR 2.5.4a 2-pass mode (Dobin et al.,
2013). Read groups were added with AddOrReplaceReadGroups and duplicated reads
were identified with MarkDuplicates script from Picard tools. Sequences overhanging
intronic regions were hardclipped and STAR mapping qualities were reassigned to match
GATK software. Variant calling was performed with HaplotypeCaller GATK script, with
phredScore 20 as minimal variant calling confidence. Variants clusters (at least 3 valid
59
variants in a window of 35 bases) were removed to diminish the e↵ect of RNA-Seq mapping
errors. Standard variants quality filtering was applied with VariantFiltration GATK script:
Fisher Strand values > 30.0 and Qual By Depth < 2.0 with. Individual SNVs were then
annotated with gene names and their predicted consequences on protein function using
VariantAnnotation Bioconductor package and gene models from GENCODE comprehensive
gene annotations set (v.28). In order to identify samples with potential DDX3X mutation,
ENSG00000215301 (gene ID) and ENST00000644876 (transcript ID) DDX3X models were
used as a reference. All nonsense, frameshift and non-synonymous SNVs with the ratio of
variant coverage : reference coverage > 0.2, localised in DDX3X helicase domain and not
previously reported in the ExAC database of common population variants were considered
as valid hits.
2.2.9 DLBCL Cell-of-origin identification from RNA-Seq data
Classification of DLBCL biopsies into four transcriptomic subtypes: ABC, GCB, Unclassi-
fied and Molecular High Grade (MHG) DLBCL was performed as previously described
(Reddy et al., 2017; Sha et al., 2019).
2.2.10 Chromosome Y expression identification from RNA-Seq
data
Chromosome Y expression identification in human RNA-Seq data was performed using
decision tree algorithm implemented in rpart R package. The model was trained using
RNA-Seq data from Genotype-Tissue Expression (GTEx) Project, which includes gene
expression samples from 54 tissue sites in non-diseased individuals (11688 samples in
total)(2013). The GTEx datasets (V8) were obtained from https://gtexportal.org/home/.
Raw RNA-Seq counts were filtered and TMM normalised. Per gene scaled gene expression
values were used as an input. The GTEx data were randomly split into a training and
test set comprising, respectively, 80% (9333 samples) and 20% (2355) of the data. The
algorithm running on default parameters achieved high performance: F1 score = 0.9973,
AUC = 0.9973, 8 males were misclassified. KDM5D, DDX3Y, USP9Y, RPS4Y1, TXLNGY,
XIST were identified as classifying genes. In order to assess the ability of the algorithm to
classify cancer samples, GTEx trained model was benchmarked against cancer dataset.
TCGA gene expression data (RNA-Seq only)(Weinstein et al., 2013) were downloaded
using TCGAbiolinks, the Bioconductor package for integrative analysis with GDC data.
Similarly, the dataset was split into a training (80%, 9198 samples) and test set (20%,
2336 samples). The algorithm achieved lower performance than in GTEx data: F1 score =
0.9624, AUC = 0.9779, 46 males and 37 females were misclassified on default parameters.
When the GTEx trained model was tested on the TCGA test dataset, the number of
60
misclassified females and males was 2 and 295, respectively. The 295 males classified as
females showed remarkably lower expression of genes localised on chromosome Y, which
could reflect previously reported loss of chromosome Y during oncogenesis (Dunford
et al., 2017). In order to obtain high-confidence set of male DLBCL patient samples with
chromosome Y expression, the GTEx trained model was used to classify samples in the
GOYA dataset.
2.2.11 Hierarchical de novo identification of translated regions
from Ribo-Seq data
Ribo-Seq and RNA-Seq FASTQ files were processed and a standard quality check was
performed as described earlier in section 2.2.2. P-site location was determined automatically
with Plastid suite using psite and phase by size functions with default settings. Only
reads of length between 25 and 33, with clear three nucleotide pattern, were used to call
ORFs. Aligned sequencing reads were used to run ORFLine (Hu et al., 2021), ORF-
RATER (Fields et al., 2015), RibORF (Ji, 2018) and RiboCode (Xiao et al., 2018) with
default settings. Harringtonine treated samples, if available, were used as an additional
input for ORF-RATER run to increase precision of locating translation initation sites.
Identified ORFs were annotated over the reference transcriptome that was customised
for lymphoid cells, so that it contains only one the most highly expressed representative
isoform per gene. Three di↵erent RNA-Seq datasets were queried for that purpose: 1) 50
base pair single-end RNA-seq data matching Ribo-Seq dataset (77 samples), 2) 50 base
pair paired end RNA-Seq data from GOYA clinical trial (553 samples) (McCord et al.,
2019a). Transcript-level quantification of expression was performed with Salmon following
the recommendations from the user manual (Patro et al., 2017). Salmon .sf files were
loaded to R with listings package. Scaled TPMs (transcript expression abundance scaled to
library size) were used to determine a mean expression of each isoform. The isoform with
the highest mean expression was selected as representative. For protein coding transcripts,
only protein coding isoforms were considered.
Identified ORFs were classified as canonical if both, start and stop codon were present
in Gencode V.29 annotations. All the remaining ORFs were assigned to one out of seven
types as defined in Table 2.1.
61
Table 2.1: Unified naming system for classifying identified ORFs
Unified type Definition
canonical Matching start and stop codon
uORF Initiating and terminating downstream of the canonical ORF.
dORF Initiating and terminating upstream of the canonical ORF.
novel Localised in non-coding transcript.
internal Contained within known CDS but out-of-frame with canonical product.
truncation Matching stop codon but initiating downstream of the canonical start site.
extension Matching stop codon but initiating upstream of the canonical start site.
readthrough Matching start codon but terminating downstream of the canonical termination site.
overlap uORF
Initiating upstream of the canonical start codon and
terminating upstream of the canonical termination site.
overlap dORF
Initiating upstream of the canonical termination site but
terminating downstream of the canonical termination site.
2.2.12 Reanalysis of published mass spectrometry datasets
ProteomeXchange database (Deutsch et al., 2016) was searched for proteomic experiments
performed in lymphoid cells (keywords: B-cells, lymphoma, lymphocyte) covering variety
of experimental conditions and mass spectrometry (MS) techniques. Experiments with
incomplete submission, targeted experiments (eg. co-immunoprecipitation of a single
protein) or with very few replicates were excluded. RAW files were downloaded from
PRIDE (Perez-Riverol et al., 2019) or MassIVE repository.
To avoid forced misanotation of MS/MS spectra to noncanonical products, a customised
reference FASTA file was build by merging a database of human canonical proteins
downloaded from UniProt with amino-acid sequences of predicted micropeptides. To
decrease search space noncanonical ORFs shorter than 6 amino-acids and showing more
than 20 % of in-frame overlap with known ORFs were filtered out.
Andromeda search engine implemented in MaxQuant 1.6.3.4 software (Cox and Mann,
2008) was used to process all collected MS datasets. The parameters were set as fol-
lows: precursor mass tolerance 20 ppm, MS/MS fragment tolerance 0.5 Da; PSM FDR
and Protein FDR threshold 1%. Noncanonical reference sequences were provided as
proteogenomic fasta file, that allows for separate FDR calculations for canonical and
noncanonical peptides. For immunopepditomics data, NewAnce workflow was followed
(Chong et al., 2020), that required additional search with Comet engine (Eng et al., 2013).
To match Comet input requirements, RAW files were converted to mzXML format with
the MsConvert from ProteoWizard tool (Chambers et al., 2012). Search parameters for
immunopeptidomics data were set as described above with small modification: peptide
length of 8–15 amino-acids was set when searching MHC I data and 8-25 for MHC II.
Variable and fixed modifications were kept as in the original publication.
62
2.2.13 Analysis of proteogenomic data downloaded from Open-
Prot and sORFdb
FASTA files containing the sequence of peptides deposited in sORFdb and OpenProt 1.6
databases were downloaded from https://www.openprot.org (Brunet et al., 2021)
and http://sorfs.org (Olexiouk et al., 2018), respectively. Amino-acid sequences of
candidate ORFs identified in this study were queried against those with the databases
using BLASTP (Altschul et al., 1990) with default settings. ORFs matching the queried
database in at least 95% were defined as valid matches.
2.2.14 Evolutionary conservation of identified ORFs
Evolutionary conservation of each identified ORF was evaluated with PhastCons score,
which represents the probability that each nucleotide belongs to an evolutionary conserved
element. Per base values for GRCh38/hg38 were downloaded from UCSC. BigWig
files obtained from three multiple alignments were selected: 1) 100 way: representing
conservation between 99 vertebrate genomes and the human genome 2) 20 way: 29
vertebrates genomes, and 3) 7 way: 6 vertebrate genomes, including 3 primates.
Averaged scores over ORF regions were computed with a command line tool bigWigAv-
erageOverBed from UCSC toolkit (Kent et al., 2010).
2.2.15 CRISPR screen design
To identify ORFs fit for CRISPR screening, entries fulfilling the following criteria were
removed: 1) ORFs shorter than 30 nucleotides, 2) with mean expression below 20th
percentile, 3) sharing more than 50% of sequence with known protein coding region, and
4) having a unique part (not overlapping any CDS) shorter than 30 nucleotides. The
remaining ORFs were collapsed by their genomic coordinated, extended 16 nucleotides
in 50 direction and used to generate a BED file. Because CRISPR-Cas9 enzyme cuts
3-4 nucleotides upstream of the PAM sequence, extension of the queried region allows to
maximise the number of candidate gRNAs targeting the translation start site. BED file
was used as an input for Chop Chop, a Python tool to search for all gRNA given genomic
coordinates (Labun et al., 2019). A customised wrapper script was used to search for all
regions at once. Chop chop parameters were set as follows:
chopchop . py T 1 M NGG maxMismatches 3 g 20
scoringMethod ALL f NN n N G hg38 o temp
t WHOLE Target [REGION]
Obtained gRNAs were then filtered based on their GC content and the number of
predicted o↵-target locations. All gRNAs with any predicted o↵-targets with 0 mismatches
63
or having GC content lower than 40% or higher than 70% were removed. gRNAs were
reassigned back to ORFs and ORFs with fewer than 5 gRNAs were considered untargetable
and filtered out. All sequences were prepended with a G at 50 end to facilitate transcription
from the U6 promoter. gRNAs were scored as implemented in the CHOP-CHOP workflow.
gRNAs with a G at position 20 (upstream of PAM were prioritised) and on-target activity
scores were computed as described in (Moreno-Mateos et al., 2015; Xu et al., 2015; Doench
et al., 2014; Dutt et al., 2011).
Remaining ORFs were then scored and ranked based on the features increasing their
biological importance. Following characteristics were used to compute the mean score per
ORF:
1. Mean scaled ORF score (Bazzini et al., 2014): reflecting the accumulation of ribosomal
footprints in the first reading frame: ORFscore = log2(RRF1+RRF2+RRF3), where
RRF is the proportion of reads in given reading frame; scaled so that it takes values
from 0 to 1, when 1 indicates a situation when all ribosomal footprints belong to the
first reading frame and 0 means no first frame preference.
2. Expression level: measured as percentile of mean expression across all unmerged
samples included in the study.
3. Identification in the reanalysed mass spectrometry studies: 2 points for ORFs
identified in both immunopeptidomics and full proteome studies, 1 - identified in
one and 0 - no found in proteomics.
4. Present in Uniprot or any of the two proteogenomic databases (sORFdb or OpenProt):
1 - if present, 0 - if not.
5. Mean conservation score computed by averaging three phasCons scores (100way,
20way and 7way) computed as described above in section 2.2.14, takes values from 0
(no evolutionary conservation) to 1 (full evolutionary conservation).
6. The number of ORF finding tools that has identified an ORF as actively translated
scaled, so that it takes values from 0.25 (detected by 1 tool) to 1 (detected by all
tools)
7. The number of samples in which an ORF was detected by at least one tool scaled so
it takes values from 0.0084 (detected in 1 sample) to 1 (detected in all samples)
8. The proportion of di↵erential expression analyses, where an ORF reached statistical
significance threshold.
The library was constructed to target top scoring ORFs with 4-5 top scoring gRNAs
per each and synthesised as a pool of 6000 single-stranded oligos (Twist Bioscience).
64
2.2.16 Figures preparation
Figures were prepared using R package ggplot2 (Wickham, 2016) and composed into panels
with Adobe Illustrator or Adobe Photoshop.
2.2.17 R and Bioconductor
Statistical analysis was performed with R 4.0.2. The version of Bioconductor was 3.12.
65
66
CHAPTER 3
Genome wide quantification of
translation in lymphoid
malignancies
3.1 Background
Deregulation of protein synthesis is a hallmark of malignant transformation, however the
molecular mechanisms underlying those processes still remain elusive. The transcription
factors BCL6 and MYC are essential for induction and maintenance of GC reaction
(De Silva and Klein, 2015) and, when deregulated, become drivers of lymphomagenesis
(Basso and Dalla-Favera, 2015). During the GC reaction, both proteins drive extensive
reprogramming of B-cell gene expression landscape, which has been studied mainly at
the level of transcription. The contribution of other levels, including post-transcriptional
regulation, is poorly understood. Sendoel et al. (2017) showed that overexpression of
SOX2 in skin epidermis is followed by broad changes in translation programme. SOX2 is a
transcription factor that is highly expressed in fully developed squamous cell carcinoma as
well as in cells considered to be cancer stem cells for this type of tumour. I hypothesised that
similar reprogramming of protein synthesis may follow overexpression of the transcription
factors BCL6 and MYC in GC B-cells.
We took advantage of a primary GC B-cells culture system, that has been developed in
the Hodson lab (Caeser et al., 2019). Primary GC B-cells were isolated from tonsil tissue
discarded after elective tonsillectomy and cultured with Follicular Dendritic Cells (FDCs)
expressing CD40L and IL 21, which mimic the supportive role of GC microenvironment.
The co-culture system has been described in detail by Caeser et al. (2019). To test whether
BCL6 and MYC, in addition to their transcriptional e↵ects, are responsible for changes
at the level of translation, we decided to perform Ribo-Seq and RNA-Seq experiments in
67
primary GC B-cells overexpressing either MYC-t2A-BCL2 or BCL6-t2A-BCL2 or BCL2
alone (Figure 3.1 A). The co-expression of BCL2 and BCL6 or MYC from the same viral
construct with T2A linker was essential to keep B-cells alive in culture and perform the
experiments (Figure 3.1 B). Transduction of either MYC-t2A-BCL2 or BCL6-t2A-BCL2
is associated with malignant transformation of the primary GC B-cells inducing their
prolonged expansion and survival (Caeser et al., 2019).
Cell culture, transduction and sequencing libraries were prepared by Dr. Jie Gao as
described in section 2.2.1.1.
Figure 3.1: Studying translational regulation in primary GC B-cells overexpressing
common lymphoma oncogenes
A) Flowchart showing the experimental design for investigating translational changes following
BCL6 or MYC overexpression. Primary GC B-cells are cultured with a monolayer of
immortalised Folicular Dendritic Cells-like feeder cells. The feeders express CD40L and IL
21 providing the essential survival signal for cultured primary cells.
B) Diagram illustrating the design of a viral construct for simultaneous expression of BCL2
with BCL6 or MYC using t2A linker. Thosea asigna virus 2A (t2A) is a small peptide
(19-22 amino acids), which undergoes self-cleavage resulting in simultaneous expression of
two or more proteins in equivalent amounts.
In this chapter:
1. I introduce RiboStream, a bioinformatic pipeline developed to process Ribo-Seq and
RNA-Seq data simultaneously,
2. I provide an overview and comparison of available tools for identifying di↵erentially
translated genes. Despite growing popularity of Ribo-Seq in studies on protein
translation, the analytic workflow is not well established. I use an external dataset
of 54 Ribo-Seq and RNA-Seq samples to evaluate the available tools and choose the
approach for further analysis,
3. I apply my optimised strategy to analyse changes in the translation as non-malignant
GC B-cells undergo transformation driven by 2 di↵erent oncogene combinations.
68
3.1.1 Establishing a bioinformatic pipeline for processing of trans-
latome and transcriptome data
The analysis of large genomic datasets usually involves sequential application of various
command line tools, software packages and custom scripts. The design of such workflows
depends on the exact next-generation sequencing technology and is usually tailored to
available computational resources. In order to facilitate processing of dataset for this and
future projects, there was a need to develop an in-house computational pipeline.
I developed a pipeline, which I named RiboStream, for parallel processing of Ribo-Seq
and RNA-Seq datasets; https://github.com/ashakru/RiboStream_bpipe.git. Subse-
quent stages of the workflow were linked using Bpipe, a Groovy language platform for
managing bioinformatic jobs (Sadedin et al., 2012). A workflow management system, such
as Bpipe, can aid to achieve reproducibility, and transparency of the analysis. Bpipe
pipeline architecture has numerous advantages over running all the jobs manually or
through a series of Shell scripts.
The central idea of a workflow management system is to automate, orchestrate and
monitor the execution of subsequent stages of the data processing. Automatic log of
executed commands and the outputs management increase readability and control on the
workflow, that facilitate troubleshooting and customisation, if needed.
It is not surprising that workflow management systems have been widely adopted
by the bioinformatic community and they are now considered a part of a good research
practice. The choice of Bpipe has been dictated by the simplicity of Bpipe architecture,
compatibility with a cluster resource manager, and the avaliability of comprehensive
documentation and users community support.
Briefly, RiboStream takes raw FASTQ files with Ribo-Seq or RNA-Seq reads, as received
from a sequencing facility, and returns: aligned BAM files, various quality check metrics
and read count matrices, which ready for downstream analysis tailored to the biological
question of interest. Optionally, Sequence Read Archive (SRA) accession numbers can be
provided instead of FASTQ files. In such scenario the pipeline downloads FASTQ files
from SRA, the largest publicly available repository of next generation sequencing data.
Flexibility and control over each stage is provided by a single configuration file, where all
the parameters of executed tools can be adjusted. Sample-specific parameters, such as
the localisation of the FASTQ files (or SRA accession numbers), type of the experiment
(Ribo-Seq or RNA-Seq) or the sequence of the adapters for trimming, are contained in a
sample sheet and parsed by the pipeline when needed. An outline of the pipeline is shown
in Figure 3.2
69
Figure 3.2: Overview of RiboStream, a Bpipe pipeline for Ribo-Seq data analysis
Flowchart showing a basic bioinformatic workflow that I developed for Ribo-Seq and RNA-Seq
data processing. A typical run performs automatic preprocessing of raw FASTQ files, alignment
to the reference genome and routine quality check focused on ribosomal footprints data. The
output, in a form of count matrices, BAM and BigWig files, is ready for a downstream analysis.
Bioinformatic tools implemented in the workflow at each stage are indicated.
SRA - Sequence Read Archive; QC - quality check
While most of the stages are the same for both data types, there are three steps specific
and for Ribo-Seq samples: prealignment to non-coding RNA sequences, extended quality
control and modified reads counting strategy.
The first involves removal of common contaminating sequences derived from ribosomal
RNA (rRNA), tRNA or small nuclear RNA (Ingolia et al., 2012). The advantage of this
extra data purification step is an increase in the proportion of true ribosome-derived
mRNA footprints in the final BAM file, containing reads aligned to the human genome.
This facilitates the interpretation of the alignment pattern and improves the estimation of
the ribosomal P-site location. An additional benefit is a smaller size of the final output
file, that accelerates further processing.
Quality control is a critical step of all next generation sequencing experiments and
usually involves evaluation of raw reads parameters (e.g. total number of reads, base quality
scores, GC content), alignment eciency and reproducibility of biological or technical
replicates. In addition there are several metrics characterising a good quality Ribo-Seq
70
experiment, that should be checked before performing further analyses. Ribo-Seq specific
quality measures include the proportion of uniquely mapped reads mapping to various
genomic locations (introns, exons, known CDS or untranslated regions), and transcript
biotypes (e.g. protein coding, lncRNAs, pseudogenes). We anticipate the majority of
ribosomal footprints to map to CDS regions of known protein-coding genes. This step was
performed with Ribowaltz package and custom R scripts utilising basic functions of the
versatile GenomicFeatures and GenomicAlignments packages. Another important quality
check is the pattern of 50ends of the footprints around annotated start and stop codons.
This should reveal enrichment of ribosomal footprints in CDS comparing to 50UTR or
30UTR and clear three-nucleotide periodicity of alignment reflecting single frame preference
of translated ribosomes (Figure 3.3 A). These patterns usually are evident for reads of
length between 26 and 30 nucleotides, which most likely correspond to true ribosomal
footprints.
A similar metric, which also investigates the characteristic distribution of the footprints,
is a metagene profile. RiboStream performs a metagene analysis to examine the distribution
of aligned Ribo-Seq reads over all known protein-coding transcripts. The aggregated
(metagene) profile visualises general patterns in read coverage over sets of genomic regions
of interest, for example promoters, transcripts or CDS (Figure 3.3 B). A metagene
analysis examining the density of Ribo-Seq reads mapping between known start and stop
codons is vital to ensure data quality and may reveal biologically important changes
in ribosome occupation between experimental conditions (Gerashchenko and Gladyshev,
2014). The expected pattern characterising a successful Ribo-Seq experiment is the
footprint density at the first 30 to 40 codons that is far greater than in the rest of CDS. In
the literature the mechanism of this pattern was widely discussed. One proposed biological
explanation is that it reflects the speed of translation initation, which is slower than
translation elongation. Specific codon context around the start codon may explain the
accumulation of the ribosomal footprints mirroring local translation rate (Ingolia et al.,
2009; Tuller et al., 2010).
An alternative hypothesis involves treatment with translation inhibitors. Cycloheximide,
which was used in this and numerous other studies including the original Ribo-Seq paper
(Ingolia et al., 2009), inhibits translation elongation by interrupting ribosome translocation.
By the time all cells are saturated with the inhibitor, some ribosomes keep initiating, which
may explain the observed accumulation of footprints in the first codons (Gerashchenko
and Gladyshev, 2014).
71
Figure 3.3: Illustration of critical steps in data processing specific to Ribo-Seq
datasets
A) Diagram showing the mechanism explaining periodical pattern of ribosome footprints
alignment, which suggest involvement in active translation.
B) Flowchart of metagene analysis implemented in RiboStream. The workflow consists of three
steps in which read coverage profiles are scaled to the same length, and then aggregated to
create a metagene profile. The final plot represents an average mapping density over all
known CDS. Deeptools commands (bamCoverage, computeMatrix or plotProfile) used at
each stage are indicated.
C) Diagram showing the position of estimated ribosomal P-site in a 28 nucleotide long Ribo-Seq
read.
D) Diagram showing the strategy to assign mapped reads to di↵erent genomic regions: 50UTR,
CDS or 30UTR.
CDS - Coding Sequence Region, 50UTR - 50 Untranslated Region, 30UTR - 30 Untranstaled
Region
72
An initiation peak also occurs in untreated Ribo-Seq samples (Ingolia et al., 2009;
Weinberg et al., 2016), but they are much smaller than the peaks in samples treated
with translation inhibitor. Small peaks can be also observed at the stop codon, but the
mechanism of this is unknown. Regardless, the mechanism, the accumulation of Ribo-Seq
reads around the start codon has real consequences for the data interpretation. Firstly,
it is a useful quality measure of the purity of Ribo-Seq signal. Secondly, it allows to
identify translation initiation sites directly from the data, which is utilised by various ORF
identification algorithms (Calviello and Ohler, 2017). Lastly, this requires a di↵erent read
counting strategy to the approach established for RNA-Seq data.
Similairly, as di↵erential expression analysis is a core application of RNA-Seq technique,
di↵erential translation is one of the main aims of a typical Ribo-Seq experiment. Counting
how many fragments have aligned to each gene is an essential step of such analysis.
Ribo-Seq read counting in the Bpipe pipeline was implemented to follow the original
protocol from Ingolia et al. (2009) with minor modifications. Ribosomal footprints were
assigned to genomic regions (CDS, intron, 50UTR or 30UTR) based on the position of
the estimated P-site (Figure 3.3 C). I built the models of the genomic regions with
GenomicFeatures R package using GENCODE v.29 comprehensive gene annotation set
as reference. The first 15 and last 15 nucleotides of each CDS region were trimmed, so
that the assigned footprints reflect the baseline translation intensity rather than the rate
of initiation or termination (Figure 3.3 D). Trimmed CDS models representing di↵erent
transcript isoforms were merged by gene. Only fragments with evident triplet nucleotide
periodicity relative to start and stop codon, typically between 27 and 30 nucleotides long,
were selected. The pipeline determines the position of the ribosomal P-site with psite
script from Plastid toolkit that looks at the o↵set between 50 end of fragments spanning
translation start site and annotated start codon, which, usually, is 12 nucleotides for reads
in the range between 27 and 30 nucleotides (Figure 3.3 C). Finally, per sample and per
gene counts matrices are built using Genomic Ranges R package allowing for assignment
of a read to more than one overlapping features (Figure 3.3 D). Corresponding RNA-Seq
samples are counted using the same gene models.
Processing of a small Ribo-Seq experiment (9-12 samples) with RiboStream installed
on High Performance Computing Cluster (18 nodes, 176 CPUs) takes about 3 hours.
73
3.1.2 Quality of translatome profiling in primary GC B-cells
Overall the average number of uniquely mapped reads in Ribo-Seq samples was almost
80 million reads of which about 70% corresponded to rRNA sequences (Figure 3.4 A).
Although the amount of rRNA-mapping reads depends on the organisms, footprint isolation
method, and translational status, they usually account for the majority of sequenced reads
(McGlincy and Ingolia, 2017). Given the fact that a ribosome-footprint complex consists
of several kilobases of rRNA but only about 28 bases of ribosome protected mRNA, this
is not unexpected (Ingolia et al., 2012). Higher rRNA content may also be observed in
conditions associated with global change in translation rate, when the ratio of actively
translated ribosomes to other RNAs is a↵ected (Ingolia et al., 2012). In contrast to a
routine rRNA-depletion library strategy for RNA-Seq experiment, only a few specific
rRNA fragments account for the large proportion of contamination observed in Ribo-Seq.
This may be related to the RNAse digestion of rRNA at specific and reproducible positions
(Ingolia et al., 2012). Our study adopted the rRNA removal protocol introduced in the
original Ribo-Seq workflow, where contaminating sequences are hybridised to biotinylated
oligonucleotides and depleted with streptavidin beads (Ingolia et al., 2012).
Of non-rRNA reads, the majority was about 29 nucleotides long corresponding to the
length of the ribosome protected mRNA fragments. The average GC content of uniquely
mapped reads was about 50 %. In about 60% mapping destination was known coding
sequence region. (Figure 3.4 B-C).
The analysis of a mean distribution of the reads 50 end around the start and stop codons,
revealed that the 50 end starts aligning about 12 nucleotides upstream from the annotated
start codon and shows strong 3-nucleotide periodicity (Figure 3.5 A). The strongest 3
nucleotide pattern of alignment was seen for reads between 26 and 29 nucleotides long,
which is in line with previous work. (Ingolia et al., 2009, 2012). The 12 nucleotide o↵set
between 50 end of a read and TIS allowed to estimate the position of the ribosome P-site,
as was described in the section 3.1.1. The distribution of estimated P-sites around start
codon presented clear frame preference and expected accumulation in the first 20 bases of
the CDS (Figure 3.5 B). The same analysis applied to samples treated with di↵erent
translation inhibitor, harringtonine, which immobilises ribosomes shortly after translation
initiation, revealed distinct pattern. While samples treated with cycloheximide showed
evident 30nucleotide periodicity over the entire length of CDS, estimated P-sites from the
harringtonine group were found almost exclusively in the first 25 nucleotides (Figure 3.5
B).
74
Figure 3.4: Basic mapping statistics metrics for Ribo-Seq reads for the BCL6/MYC
experiment in human GC B-cells
A) The fate of Ribo-Seq reads after the two rounds of alignment shown for each replicate. The
left panel shows the percentage of reads in each of the five group (aligned to rRNA, removed
during trimming stage, mapped to multiple locations, unmapped or aligned uniquely). The
right panel depicts the total number of uniquely mapped reads that was eligible for the
further analysis.
B) Histogram of read length of uniquely mapped Ribo-Seq reads stratified by experimental
condition.
C) Distribution of mapped reads from Ribo-Seq and RNA-Seq experiments to gene features,
showing the expected restriction of Ribo-Seq reads to the CDS with only a small portion
mapping to UTRs.
75
On a global scale, a metagene plot of cycloheximide treated samples showed a char-
acteristic peak at the expected translation initation site (TIS) and abrupt drop-o↵ the
signal after translation termination site (TTS). The pattern of ribosomes occupancy was
similar for all experimental conditions (Figure 3.5 C). When P-sites were stratified by
transcript regions (50UTR, CDS or 30UTR) and read length, they, again, showed a strong
frame restriction in known CDS regions (Figure 3.5 D).
Even in a good quality experiment, genes with low expression level have typically higher
variance which may decrease the sensitivity of di↵erential expression analysis. With only
few read counts assigned, it is dicult to distinguish biological di↵erences from technical
(sampling) noise. Therefore, a common practice is to remove those lowly expressed genes
before attempting di↵erential expression analysis. I estimated the threshold of this filtering
adopting the strategy introduced by Ingolia et al. (2009), removing all genes with number
of RNA-Seq reads lower than 128 in more than 25% of samples. I also filtered out histone
genes, which are not polyadenylated and therefore underrepresented in our RNA-Seq data
with poly(A) enrichment library strategy (Figure 3.6 A).
Next, I evaluated the reproducibility of gene expression measurements obtained from
Ribo-Seq and RNA-Seq. Between-replicate measurement error distribution was relatively
narrow: on average normalised expression values from the same experimental condition
were lower than 1.22 fold for 90 % of the genes (Figure 3.6 B). Translation eciency
(TE), which is the ratio of normalised ribosomal footprints abundance to mRNA level,
showed almost 100-fold dynamic range (Figure 3.6 C), which is similar to the original
Ribo-Seq protocol (Ingolia et al., 2009). Lastly, the values for both techniques were highly
reproducible between biological and technical replicates, average Pearson’s product of
correlation was 0.965 for both Ribo-Seq and RNA-Seq (Figure 3.6 D). Interestingly, when
Ribo-Seq measured expression levels are juxtaposed with the RNA-Seq counterparts two
observations can be made. Firstly, ribosomal footprints abundance has smaller dynamic
range that mRNA abundance and, secondly, the relationship between Ribo-Seq and RNA-
Seq signal seems to be intensity dependent (Figure 3.6 D). This means that the high
correlation between the number of ribosomal footprints and mRNA abundance can be
spurious, particulairly for lowly expressed transcripts, and a translation intensity of a
mRNA needs to be interpreted with care.
I conclude that all the quality measures and the reproducibility of gene expression
measurements meet the expectations for Ribo-Seq data. By comparison to previous
Ribo-Seq studies, this experiment has larger number of replicates (4 versus usual 2) and
shows overall high reproducibility and good quality of the libraries. This confirms that
ribosomal footprints obtained from primary GC B-cells: 1) correspond to the ribosomes
involved in active translation, and 2) can be used for further analyses.
76
Figure 3.5: Reading frame restriction of Ribo-Seq reads for the BCL6/MYC exper-
iment in human GC B-cells
A) Heatmap of Ribo-Seq read frame usage by read length, showing read frame restriction in
the 28-31 nucleotide ribosome protected fragments.
B) Histogram of mean fold changes in TMM normalised ribosome footprints density between
replicates of the same experimental condition.
C) Metagene plot showing distribution of mapped Ribo-Seq reads to regions of the transcript.
D) Heatmap of Ribo-Seq read frame usage by gene feature, showing how the characteristic
read frame bias within the CDS but not the UTRs.
77
Figure 3.6: Low counts filtering and reproducibility of Ribo-Seq and RNA-Seq
samples for the BCL6/MYC experiment in human GC B-cells
A) Scatter plot showing trimmed mean of M-values (TMM) normalised ribosome footprints
density against mRNA abundance. Genes classified as low counts and histons were excluded
from further analysis.
B) Histogram of mean log2 fold changes in ribosome footprints abundance between samples
from the same experimental condition. This is similar to the reproducibility reported by
Ingolia et al. (2009).
C) Histogram of mean Translation Eciency (TE) values obtained from paired Ribo-Seq
and RNA-Seq samples from the same experimental condition. Similar dynamic range was
shown by Ingolia et al. (2009).
D) Scatter plots showing consistency between four replicates of Ribo-Seq and RNA-Seq.
Pearson correlation coecients are shown for each comparison.
78
3.1.3 Benchmarking statistical approaches for di↵erential trans-
lation analysis
The aim of di↵erential translation analysis is to identify genes regulated predominantly at
the level of translation. Although many computational tools has been developed to perform
the statistical analysis of di↵erentially translated genes, very few have been used beyond
their initial publication. The popularity of the Ribo-Seq technique, however, is growing.
In 2021 the original paper (Ingolia et al., 2009) reached 3050 citation and, according to
PubMed, there are almost 1000 papers containing Ribo-Seq or ribosome-profiling in the
title or abstract. An opposite trend has been observed for the computational methods.
Only two out of seven most popular tools for di↵erential translation analysis, that will be
reviewed in this section, exceeds 100 citation. For comparison, the number of citations for
DESeq2 and edgeR, which are the two most popular packages for di↵erential expression
analysis of RNA-Seq experiment go over 27 000 and 20 000, respectively (Love et al., 2014;
Robinson et al., 2010).
To make an informed decision on how to identify translationally regulated genes
associated with BCL6 or MYC overexpression, I reviewed previous approaches to di↵erential
translation analysis and evaluated their performance using an external dataset of 54 Ribo-
Seq/RNA-Seq pairs from Yoruba lymphoblastoid cell lines (Battle et al., 2015). The
choice of this dataset has been dictated by the large number of replicates available, which
allows me to benchmark tools using various experimental designs. I will return to the
BCL6/MYC data at the end of this chapter.
Translation eciency
Ribo-Seq relies on the assumption that the density of ribosomal footprints (Ribo-Seq
reads) mapping to a gene is a proxy for its translation intensity (Ingolia et al., 2009).
Because the more abundant mRNAs produce more ribosomal footprints, the Ribo-Seq
signal is dependent on a transcript abundance. In other words, the number of ribosomal
footprints mapping to a gene is positively correlated with the number of RNA-Seq reads
assigned to the same gene. Therefore, each Ribo-Seq sample is always paired with its
corresponding RNA-Seq sample in order to enable the calculation of mRNA-specific
translation rate. Translation Eciency (TE) is a measure introduced first by Ingolia et al.
(2009) to compute translation rate relative to mRNA abundance. By definition, TE is the
ratio of ribosomal footprints abundance to mRNA abundance, typically presented as log
ratios. Thus TE for a gene g is:
TEg = log(
RFg
mRNAg
),
where RF and mRNA stand for ribosomal footprints or mRNA abundance, respectively.
79
Di↵erent normalisation methods can be used to calculate these values. A simple approach
is to normalise the number of ribosomal footprints per gene by the library size and the
length of the CDS with Reads Per Kilobase Million (RPKM) or just by the library size,
using e.g. Counts Per Million (CPM). Normalisation methods developed initially for
di↵erential expression analysis of RNA-Seq data, such as Trimmed Mean of M-values
(TMM) (Robinson et al., 2010) or Relative Log Expression (RLE) (Love et al., 2014), are
also applicable.
Fold change in TE has been widely adopted to identify genes with translation load that
is disproportionate to the transcript abundance. However, as was pointed by Larsson et al.
(2010), the TE fails to control for the e↵ect of mRNA level fully. Bias of TE ratio can be
explained mathematically by spurious correlation, described by Pearson in 1897 (Pearson,
1897). It refers to a situation when dependent variable (TE in this case) correlates with
the mean value of the independent variable (mRNA abundance) even when mRNA and
ribosomal footprints abundance are uncorrelated, which is illustrated by the following
equation:
r(YZ),Z =
sZp
s2Y + s
2
Z
,
where Y is a vector of ribosomal footprints abundance, Z is a vector of paired mRNA
abundance, r is the Pearson correlation coecient and s is the sample standard deviation.
Here a correlation between TE and log mRNA abundance is a function of the standard
deviation of TE and mRNA abundance vector from experimental replicates. Larsson et al.
(2010) conclude that this property gives rise to false positives and false negatives when
TE is the only metric used to identify translationally regulated genes, especially if large
shifts in mRNA abundance are expected.
Brief overview of current methods for di↵erential translation analysis
The lack of established analytic workflow, as this is the case for RNA-Seq data analysis,
suggests that the di↵erential translation analysis is not straightforward. Despite di↵erent
biological origin Ribo-Seq and RNA-Seq data share common features: both represent
sequence of mRNA fragments that can be summarised as count matrices of gene counts
per sample. Most Ribo-Seq specific tools take advantage of these similarities tailoring
transcriptomic workflows to di↵erential translation analysis. Comprehensive review and
benchmarking of developed strategies has been missing, which complicates an informed
strategy selection. In order to bridge this gap and pick an optimal strategy to analyse the
data for this project, I reviewed five tools:
1. Xtail (Xiao et al., 2016),
2. Riborex (Li et al., 2017),
80
3. anota (Larsson et al., 2011),
4. anota2 (Oertlin et al., 2019),
5. deltaTE (Chothani et al., 2019),
in terms of their statistical approaches, performance, and the impact of experimental
design on the accuracy and ability to identify di↵erentially translated genes.
Because the majority of tools either run DESeq2 in the background or utilise similar
strategies to analyse counts data, I will first outline the basics of DESeq2 analysis. In
DESeq2 raw counts for each gene are modelled with negative binomial distribution (NB)
using a generalized linear model (GLM). Mean values for each gene are normalised taking
into account di↵erences in the total number of reads between samples and NB parameters
(mean and dispersion) are estimated from the data. With a simple experiment design
(for example control and treatment), a GLM regression model will have one coecient
corresponding to the log2 fold change between experimental conditions. The significance
of this e↵ect size is typically tested with Wald statistics, as detailed by Love et al. (2014).
A straightforward implementation of the DESeq2 method for di↵erential translation
analysis is deltaTE, which takes a combined matrix of Ribo-Seq and RNA-Seq counts
and models the di↵erence between the two, that is specific to the experimental condition.
DeltaTE models this e↵ect introducing an interaction term to the GLM formula and test
for statistical significance with Wald test. A potential problem with this approach is that
using both data types to estimate NB parameters may lead to biased results.
Very similar strategy (combined analysis with interaction term) is employed by Ri-
borex. However, in addition to DESeq2-based workflow, two more methods are available:
edgeR-based and Voom-based. All three process Ribo-Seq and RNA-Seq reads together.
It is important to mention that edgeR (Robinson et al., 2010) and DESeq2, in their
principles are very similar: both use GLM methods, model counts data with NB and
assume most genes are not di↵erently expressed. Di↵erences can be mainly attributed
to processing outliers, dealing with lowly expressed genes, and normalization approach
which may a↵ect the estimation of NB parameters. Voom is based on limma, which was
originally developed to analyse microarrays data (Law et al., 2014). Voom estimates
the mean-variance relationship of the log-counts, generates a precision weight for each
observation and enters these into the an empirical Bayes model (Law et al., 2014).
Xtail also uses NB and estimates its parameters running DESeq2 in the background
separately for RNA-Seq and Ribo-Seq counts. Then, it builds joint probability matrix of
two distributions: one for the di↵erence between experimental conditions (separately for
Ribo-Seq and RNA-Seq), and second, reflecting the overall di↵erence in gene expression
measured by Ribo-Seq and RNA-Seq (separately for experimental conditions). Finally,
81
Xtail compares the log2 fold changes of ribosomal footprints with log2 fold changes of
mRNA abundance OR the di↵erence in the disproportion in translational and mRNA
response between two conditions. The final p-value is selected dependent on which of the
two approaches returns more conservative results.
The last family of tools, anota/anota2 employs a di↵erent approach. While the ma-
jority of methods focus on a simple di↵erence in TE between two conditions, anota/anota2
rely on the analysis of partial variance (APV) and linear regression to control for cytosolic
mRNA levels evaluating changes that are independent on transcripts abundance.
To sum up, all tools designed for di↵erential translation analysis (except anota/anota2)
assume read counts to follow NB, model the expression with GLM and test for the di↵erence
in TE. However, only two (Riborex and deltaTE) allows for experimental design more
complex than just two conditions, e.g. including time series, sequencing batches or di↵erent
biological models. Di↵erences in the strategies are also related to the estimation of NB
parameters, null hypothesis and statistical approaches to identify deferentially translated
genes.
An alternative approach to identify di↵erentially translated genes, used successfully by
other groups (Sendoel et al., 2017; Hsieh et al., 2012) is to combine a standard di↵erential
expression analysis performed separately for translatome and transcriptome data with
heuristic system of (arbitrary) thresholds to extract genes with biologically relevant e↵ect
sizes. An example set of rules for calling a gene di↵erentially translated is as follows:
1. Statistically significant change in ribosomal footprints abundance (evaluated with
standard DESeq2 workflow), FDR < 0.1,
2. Absolute mean log2 fold change in mRNA abundance < 0.3,
3. Absolute mean log2 fold change in TE abundance > 0.3.
This heuristic approach can be naturally extended to classify genes into distinct regulatory
subtypes based on the direction and coordination between translation and mRNA levels.
Genes with significant change in mRNA level, with concordant direction of translational
change, are classified as homodirectional (up or down), while genes with opposite direction
of change (e.g. mRNA downregulated with ribosomal footprints abundance upregulated)
are considered ‘bu↵ered’, see section 2.2.3. I refer to this strategy as DESeq2T and,
for the purpose of this benchmarking analysis, focus only on the di↵erentially translated
genes.
82
Comparison of strategies to perform di↵erential translation analysis
To compare the performance of existing tools in their ability to accurately identify
di↵erentially translated genes, I generated a series of Ribo-Seq analysis sets with varying
number of replicates and di↵erentially expressed genes. For that purpose I used data
from a real translatome study of 54 Yoruba lymphoblastoid cell lines (Battle et al., 2015).
Previous performance analyses, associated with the initial publication of the di↵erential
translation tools, used predominantly artificial data. Here, by using true experimental
data we are closer to mimicking a real-life user experience. I chose to use the Battle et al.
(2015) data because of the large number of paired Ribo-Seq and RNA-Seq samples of the
same cell type. I downloaded raw FASTQ files from SRA and processed them using my
RiboStream pipeline obtaining CDS count matrices.
First, I evaluated the performance of each tool under a NULL model. From the set of
all 54 samples, I generated a series of analysis sets with varying number of replicates. As
they all come from repeated sequencing of the same cell type, I do not expect any genes
to be identified as di↵erentially translated. This allowed me to estimate the frequency of
type I of error (false positives). Half of the randomly selected samples in each analysis
set was labelled as ‘treatment’, the remaining half as ‘control’. No separation between
‘treatment’ and ‘control’ was observed in the PCA plot (Figure 3.7 A). The tools were
applied to datasets with 2-20 replicates, 50 runs per tool per replicate group in total.
Figure 3.7: Comparison of available tools for di↵erential translation analysis
A) Representative PCA plot showing a NULL model dataset with 7 replicates per condition.
Treatment and control samples were randomly assigned from a set of 54 samples of the
same cell type.
B) Heatmap showing the median number of di↵erentially translated genes per tool per
replicate group in the NULL model test (assuming no di↵erence in expression between two
conditions). For each replicate group, 50 count matrices were generated and each tool was
applied.
83
Overall, the median number of false positive findings, at FDR < 0.05 with absolute
log2 fold change in TE larger than 0.3, was the highest for Xtail (122 genes), which is in
line with previous study Oertlin et al. (2019) and suggest that the performance of Xtail in
terms of false positive findings may be inferior in comparison to other tools (Figure 3.7
B). For other tools, the median number of false positive findings was higher in comparisons
with low number of replicates. When the number of replicates was higher than 3, the
number of false positives in the NULL model was negligible (with exception of Xtail).
Next, I applied the tools to a di↵erential translation model with simulated fold changes.
This time, randomly selected genes from a ”treatment” group were assigned artificial
fold changes sampled from a normal distribution. The fold changes were synchronised
between RNA-Seq and Ribo-Seq samples, so that there is a 0.4 Pearson correlation between
fold changes observed in RNA-Seq and Ribo-Seq, which resembles a natural relationship
between mRNA and ribosomal footprints abundance (Figure 3.8 A). The count matrices
with designed fold changes were generated using seqgendi↵, R package for adding a known
amount of signal to a real read count matrix. Clear separation between treatment and
control group was observed in PCA plot (Figure 3.8 B). Similairly to the NULL model
analysis, the tools were applied to datasets with 2 - 20 replicates, 50 runs per tool per
replicate group (350 comparisons in total). For each tool and each run I defined a ground
truth, which was the result of di↵erential translation analysis performed using 20 replicates.
The first striking di↵erence between the tools was the total number of di↵erentially
translated genes (DTGs) identified with 20 replicates (Figure 3.8 C). While the TE-
based strategies (TEdelta, Riborex and Xtail) identified more than 5000 DTGs, DESeq2T
returned almost five times fewer genes. For anota and anota2 the mean number of DTGs
was 1950 and 3600, respectively. This may reflect di↵erent definitions of di↵erential
translation between the tools. For TE-based strategies, statistically significant change in
TE (which may be dependent on either change in ribosomal footprints, mRNA abundance or
both) is sucient to classify a gene as di↵erentially translated. Whereas, for anota/anota2
and DESeq2T, the change in ribosomal footprints abundance must be significant AND
independent on mRNA abundance (anota/anota2) or with negligible change at mRNA
level (DESeq2T).
Next, I compared the robustness of the DTGs identification between the tools. For
each tool and replicate group I computed the following metrics: false positive rate (FPR),
false negative rate (FNR), true positive rate (TPR), accuracy, precision and the F1 score
(Figure 3.8 D). Similairly as for NULL model test, the highest rate of false positives was
seen for Xtail.
84
Figure 3.8: Comparison of available tools for di↵erential translation analysis
A) Representative scatter plot of log2 fold change distribution for simulation of Ribo-Seq and
RNA-Seq di↵erential expression analysis.
B) Representative PCA plot showing a model di↵erential translation dataset used for bench-
marking.
C) Boxplot showing the total number of identified di↵erentially translated genes (DTGs) by
tool when 20 replicates were used for comparisons. Each tool has been applied to a 20
analysis sets.
D) Performance of 9 algorithms for identifying DTGs by the number of replicates used for the
comparison. For each replicate group mean values of the performance metrics over 50 runs
were computed.
85
Overall, TE-based strategies had higher TPR, lower FNR and higher precision (positive
predictive value), but this was accompanied by higher FPR. In contrast, the accuracy (the
total number of correct findings), and F1 score (weighed average of precision and TPR)
was the highest for DESeq2T, while TE-based showed inferior performance. The number of
replicates a↵ected the performance of all methods, but with di↵erent strength. The general
trend was inferior performance for lower number of replicates (2-3) with improvement for
comparisons with 8-10 replicates. Interestingly, FPR was less dependent on the number of
replicates: it increased steadily between 2 and 5, and, with the exception of Riborex and
deltaTE, stabilised for 5 and more. For Riborex and deltaTE, FPR started growing again
when the number of replicates exceeded 10.
The results of benchmarking analysis of methods for identifying di↵erentially translated
genes demonstrate that no single method is superior, but taking into account experimental
design and biological question of interest, it is possible to di↵erentiate between better and
worse approaches. Firstly, a striking feature of a di↵erential translation analysis, regardless
the method, is its relatively low recall (true positive rate) and high FNR when the number
of replicates is low. It seems that at least 8 replicates are needed to reach the level of
50-70 % true positive findings in the group of identified DTGs. Secondly, with FDR <
0.05, the rate of false positive findings is well controlled for almost all TE-based tools, with
the exception of Xtail, which shown high FPR in the NULL model and in the di↵erential
translation model. The lowest rate of false positive findings across all replicates groups was
observed for DESeq2T. Lastly, the flexibility to analyse more complex experimental design,
such as time series or sequencing batches, might be key for method selection. Among the
benchmarked tools, this is possible in Riborex, DESeq2T and TEdelta.
I conclude that for the analysis of small to medium size translatome experiments with
complex experimental design DESeq2T and Riborex are the preferential choice. However,
when the primary interest of the analysis are the changes ribosomal footprints abundance
that are independent of the mRNA level, the analysis performed with DESeq2T may
provide better answer due to its lower rate of false positive findings. Therefore this is the
approach I adopted to identify di↵erentially translated genes in this and the next chapter.
86
3.2 Translational regulation following BCL6 andMYC
overexpression in primary GC B-cells
Having settled on an analysis strategy to identify di↵erentially translated genes, I returned
to the experiment with primary GC B-cells. I analysed gene expression responses following
MYC and BCL6 overexpression in primary GC B-cells following the DESeq2T method
tested above and detailed in section 2.2.3. Briefly, I performed a standard di↵erential
expression analysis with DESeq2, separately for Ribo-Seq and RNA-Seq data. This
revealed that, for both BCL6 and MYC, the fold changes in mRNA abundance and
ribosomal footprint density were highly correlated (Pearson’s product of correlation, 0.71
and 0.82, respectively) suggesting that mRNA-driven changes are the dominant mechanism
of regulation (Figure 3.9 A-B). Overall, in both experiment types, I identified 946
di↵erentially expressed genes for BCL6 and more than four times more, 4247, for MYC,
of which 31.8% and 43.7% overlapped between transcriptome and translatome analysis
(Figure 3.9 C).
Next, I classified di↵erentially expressed genes into 6 regulatory groups, depending
on the direction (up or down) and the type of expression response: translation only,
coordinated (homodirectional) or bu↵ered between translation and mRNA level, see
section 2.2.3.
To characterise each of the potential regulatory programmes, I performed a gene ontol-
ogy (GO) and pathway analysis. Top statistically significant findings are shown in (Figure
3.9 D-E). In BCL6 overexpressing cells, genes showing coordinated downregulation at
the level of transcription and translation were enriched for numerous terms associated
with unfolded protein response and inflammation. Indeed, the expression of many core
components of the ER stress pathway, including ERN1, XBP1, or DDIT3 (CHOP) was
significantly reduced in both RNA-Seq and Ribo-Seq data. Pathways related to extracel-
lular matrix organisation and signalling through receptor tyrosine kinases were enriched
in the upregulated group. For MYC, I observed strong upregulation of terms related
to translation and ribosome biogenesis, which is in line with previous studies on MYC
regulatory network. The downregulated group consisted of terms related to inflammatory
responses, G protein-coupled receptors (GPCR) and PD-L1 signalling. The latter included
mainly genes related to HLA presentation. Interestingly, the decrease in mRNA abundance
of CD274 (Programmed death-ligand 1, PD-L1) in MYC overexpressing cells was bu↵ered
by increase in the ribosome footprint density (TE log2 fold change = 1.280). In BCL6
overexpressing cells, a similar trend was observed - mild mRNA downregulation bu↵ered by
the ribosome abundance (TE log2 fold change = 0.42). The expression of PD-L1 on the cell
surface is known to inhibit T cell-mediated immune responses, which is a well established
mechanism of maintaining physiological self-tolerance, but also promoting immune escape
87
during tumour formation (Pardoll, 2012). Maintenance of PD-L1 translation combined
with translational downregulation of HLA-related genes by MYC, could mediate permissive
environment for aberrant clonal expansion of GC B-cells, thus, facilitate tumour formation.
However, the average expression of PD-L1 in primary GC B-cells was low, just above 10th
percentile of expression.
The mean di↵erence in TE had narrow dynamic range. Only about 1% of genes
had absolute log2 fold changes larger than 1. Translation-level changes were dominated
by the regulation of synthesis of housekeeping proteins, such as the components of the
mitochondrial respiratory chain (MYC only) or ribosomal proteins (MYC and BCL6)
(Figure 3.9 E). 4 out of 7 mitochondrially encoded subunits of NADH dehydrogenase
(complex I) were translationally upregulated in MYC overexpressing cells (TE log2 fold
change between 0.4 and 1) as well as CYCS gene (cytochrome c), and 3 components of
the cytochrome c oxidase (Figure 3.9 F). This may complement a known function of
MYC in promoting mitochondria biogenesis (Morrish and Hockenbery, 2014).
When it comes to ribosomal proteins (RPs), BCL6 or MYC overexpression was
associated with translational control of few of them, but in opposite directions (Figure
3.9 E). RPL28, RPL32, RPL36AL, RPLP0 and RPS9 were translationally suppressed
in MYC overexpressing cells, while translational upregulation of RPL5, RPL24, RPL18,
RPS27L and RPL30 was associated with BCL6 overexpression. Translational regulation of
RPs is typically associated with mTOR signalling and preferential translation of transcripts
with 50terminal oligopyrimidine (50TOP) motifs. However, since MYC overexpression
is associated with an increase in the RPs mRNA abundance, changes in TE can reflect
mRNA rather than translation-driven changes.
Overall, I conclude that the overexpression of MYC or BCL6 is associated with
relatively minor changes in translation intensity that are independent on mRNA-driven
reprogramming. Translational control a↵ects only selected transcripts, predominantly those
involved in highly energetic processes such as ribosome biogenesis or respiratory chain. In
addition, translational bu↵ering of decreasing levels of PD-L1 mRNA were revealed, which
may suggest the involvement of translational control in immune surveillance mechanisms.
Although, the data presented here do not allow me to determine the significance of such
adaptive mechanisms, they provide an interesting insight into the scope of changes that
can accompany deregulation of MYC or BCL6 in B-cell lymphoma.
88
Figure 3.9: Di↵erential translation analysis of GC B-cells overexpressing BCL6 or
MYC
A) Scatter plot showing log2 fold changes in ribosome footprints occupancy against log2 fold
changes in mRNA level for BCL6-t2A-BCL2 versus BCL2 comparison. Colours represent
regulatory groups.
B) Scatter plot showing log2 fold changes in ribosome footprints occupancy against log2 fold
changes in mRNA level for MYC-t2A-BCL2 versus BCL2 comparison. Colours represent
regulatory groups.
C) Venn Diagrams showing the overlap in di↵erentially expressed genes identified by trans-
latome (Ribo-Seq) or transcriptome (RNA-Seq) data.
D) Reactome Pathway analysis for genes with coordinated change (homodirectional) in
expression. Three top enriched pathway are shown per group (FDR < 0.1).
E) Gene ontology analysis of di↵erentially translated genes (FDR < 0.1).
F) The network of protein-protein interactions of translationally upregulated genes following
MYC overexpression. Obtained from STRING 1.1.5.
89
3.3 Discussion
Main findings
Ribo-seq, also known as ribosome profiling, is a Next Generation Sequencing technique
that has transformed studies on mRNA translation. Here, I introduce RiboStream,
a bioinformatic pipeline that I developed to process translatome data eciently and
accurately. RiboStream was applied to process a large dataset of 54 Ribo-Seq and RNA-
Seq samples that allowed me to systematically benchmark available tools for di↵erential
translation analysis. Based on this, I selected DESeq2T as preferential for analysing Ribo-
Seq datasets in this study. DESeq2T method combines a standard di↵erential expression
analysis using DESeq2 software with a set of rules classifying genes subject to distinct
mechanisms of regulation, DESeq2T showed superior accuracy, low false positive rate and
flexibility to analyse experiments with complex experimental design. A similar strategy,
applied in previous studies (Hsieh et al., 2012; Sendoel et al., 2017) has lead to biologically
relevant results (confirmed experimentally), despite using a lower number of replicates
than in our study. I then used DESeq2T to dissect the translational consequences of
BCL6 or MYC overexpression in the primary GC B-cells model. This revealed preferential
translational of selected transcripts encoding certain ribosomal proteins and the components
of the respiratory chain. While the concordant translational response followed changes in
the mRNA level in the majority of genes, there were few exceptions from that rule. A
particularly interesting example was translational bu↵ering of CD274 expression (PD-L1).
Limitations and artefacts of translatome studies
Given the dynamic nature of translation, the process of Ribo-Seq library preparation is
a critical factor in the accurate measurement of cellular translatome. In order to keep
the position of translating ribosomes undisturbed during cell harvesting, the standard
practice is to treat the cells with translation inhibitors just before the procedure. The
most common drug used to freeze the ribosomes before harvesting is cycloheximide (CHX),
but the fidelity and precision of ribosomal footprints mapping in samples treated with
CHX has been questioned (Duncan and Mata, 2017; Gerashchenko and Gladyshev, 2014;
Santos et al., 2019; Hussmann et al., 2015). Analyses performed in lower eukaryotes
(Schizosaccharomyces sp. and Saccharomyces sp.) showed that CHX induces artefacts in
ribosome coverage profile in cells subjected to di↵erent types of stress. CHX artefacts were
responsible for skew in codon occupancy towards CGA and CGG (Hussmann et al., 2015),
accumulation of the ribosomal footprints in 50UTR and in the first 100–200 nucleotides of
the coding sequence, so-called 50 translation ramp. (Gerashchenko and Gladyshev, 2014;
Tuller et al., 2010). Highly expressed genes, such as ribosomal proteins, were more prone to
90
experience CHX-induced artefacts (Santos et al., 2019) and the e↵ect was dose-dependent.
However, a similar study validating these findings in yeasts and mammalian cells showed
that this e↵ect is species specific (Sharma et al., 2019). Sharma et al. (2019) showed that
CHX not only did not disturb the codon occupancy pattern in human cells but also did not
a↵ect the gene-level translation intensity measurements. Important factors determining the
quality of the experiment included: the type of RNAse, its concentration (Sharma et al.,
2019; Gerashchenko and Gladyshev, 2017) and the time between the onset of harvesting
and flash-freezing (Sharma et al., 2019; Rooijers et al., 2013) These results suggest that
the experiments performed in yeast cannot be simply extrapolated to mammalian systems.
Although we cannot exclude that di↵erential translation analysis performed in human cells
is free from artefacts and biases, it is unlikely that the results presented here are caused
by the usage of CHX during the Ribo-Seq library preparation.
A group of housekeeping genes undergo translational regulation following
BCL6 or MYC overexpression
The analysis of cellular translatome and transcriptome following BCL6 or MYC overexpres-
sion revealed a profound correlation between mRNA and ribosomal footprint abundance,
which suggest domination of mRNA-driven changes. Given that, during dynamic cell
transitions, even up to 92 % of per mRNA-translation rates can be explained by mRNA
levels (Jovanovic et al., 2015), this is not a surprising finding for proteins, which are
bona fide transcription factors. Interestingly, the same study (Jovanovic et al., 2015),
utilising a combination of pulsed-SILAC and RNA-Seq, showed that while the majority of
changes in protein level is, indeed, driven by mRNA abundance, a group of housekeeping
genes, including ribosomal proteins (RPs) and mitochondrion-related were more dependent
on translation and protein degradation, respectively. The hypothesis put forward to
explain this was that translational regulation is reactive to dynamic changes of cellular
states by finely tuning the rate of specific metabolic processes (Jovanovic et al., 2015).
Here, I identified RPs and oxidative phosphorylation related genes as translationally
regulated. MYC overexpression resulted in decreased translation rate of selected RPs
and increased synthesis of several key enzymes of the respiratory chain. On the contrary,
BCL6 overexpression was associated with an increase in the translation of a few ribosomal
proteins.
Translational control of ribosomal proteins expression
Translational regulation of RPs was reported previously. Strong repression of translation
of RPs was observed during di↵erentiation of mouse embryonic stem cells (Ingolia et al.,
2011), and after treatment with mTOR inhibitor of PC3 human prostate cancer cells
(Hsieh et al., 2012). RPs were also translationally upregulated in heart tissue samples of
91
patients with dilated cardiomyopathy compared to normal hearts (van Heesch et al., 2019).
Two patterns of regulation can be distinguished in these studies: global and selective.
The first mechanism refers to 50TOP mediated control of mTOR downstream signalling
on the translation of the core components of protein synthesis machinery, which includes
RPs and few translation factors (Thoreen et al., 2012; Hsieh et al., 2012; Philippe et al.,
2020). While expression of MYC is known to correlate with mTOR activation (Lu et al.,
2021; Pourdehnad et al., 2013; Liu et al., 2017), no such pattern has been shown for BCL6.
While inhibition of mTOR signalling is associated with translational repression of all
RPs (Hsieh et al., 2012), in other studies (van Heesch et al., 2019; Ingolia et al., 2011),
including this one, di↵erential translation was shown only for individual proteins. There is
a possibility that forced expression of MYC or BCL6 induces adaptive mTOR-mediated
adjustments of the RPs synthesis, but only a few single RPs reached the detection level.
Ribosomal proteins are highly expressed; there are millions of ribosomes present in every
cell, so even small changes in their abundance can have a profound e↵ect on translation
(Genuth and Barna, 2018).
Selective regulation of translation of specific RPs may be explained in the context of
ribosomal heterogeneity. Tissue specific or developmental stage specific pattern of the
core ribosomal proteins expression has been reported in human and other organisms, as
reviewed by (Genuth and Barna, 2018; Shi and Barna, 2015). Changes in the abundance
of selected RPs and the assembly of specialised ribosomes could drive a specific programme
of translation that could facilitate the oncogenic function of MYC or BCL6. For example,
knockdown of RPL28, which I found translationally downregulated in MYC overexpressing
cells, is not lethal to the cell but promotes an MHC I peptides presentation of non-canonical
peptides from non-AUG start codons and noncoding regions of the transcriptome. Thus,
regulation of MHC I peptide presentation may facilitate immune surveillance (Wei et al.,
2019). An additional level of complexity is provided by di↵erences in the stoichiometry
of RPs in free ribosomal subunits and translationally active ribosomes, or the extra-
ribosomal roles of certain RPs (Shi et al., 2017). Mutations in RPs are well known in
the context of ribosomopathies, hereditary disorders with increased risk of malignancy,
including lymphoma. Abnormalities in the ribosome biogenesis process can trigger p53
activation and DNA damage response (Lindstro¨m et al., 2018). Therefore fine tuning of
the abundance of individual ribosomal proteins may be essential for tumour initiation and
maintenance.
An alternative explanation is that the disproportionate translation of specific ribosomal
proteins is a marker of profound transcriptional reorganisation following BCL6 or MYC
overexpression. Both Ribo-Seq and RNA-Seq analysis provides only a relative quantification
of gene expression, which may mask global shifts in translation intensity. It would be
interesting to investigate the changes in polysome fractions of individual transcripts and
92
compare them with the results from our Ribo-Seq experiment, as this could provide better
resolution of translational response during dynamic reprogramming of the transcriptome.
MYC controls PD-L1 translation in primary GC B-cells
Translational control of the immune checkpoints and HLA surface molecules has attracted
much attention in the context of immunosurveillance of developing Here, I observed MYC-
induced downregulation of mRNA abundance of 24 out of 26 genes encoding HLA class I
and II molecules combined with translational control of PD-L1, an important component
of an immune checkpoint. PD-L1 is a ligand of PD1 receptor located on activated T-cells
and expressed during persistent antigen stimulation to constrain immune response (Sharpe
and Pauken, 2018) MYC promotes the immune escape of tumours through a variety
of mechanisms. Its ability to downregulate HLA expression has been known for a long
time (Versteeg et al., 1988; God et al., 2015; Staege et al., 2002), so this not comes as a
surprise in MYC-overexpressing GC B-cells. MYC-induced immunosuppressive phenotype
is further enhanced by transcriptional control of two immune checkpoints: PD-L1 and
CD47 (Casey et al., 2016, 2018). Inactivation of Myc lead to abrupt decrease in PD-L1
and CD47 mRNA and protein level in multiple cancer types (in vitro and in vivo evidence)
(Casey et al., 2016, 2018), which may explain an increased recruitment of immune cells to
the tumour tissue following depletion of MYC (Rakhra et al., 2010). It is important to
note that PD-L1 mRNA levels in cancer, including B-cell lymphoma, are also regulated
by MYC-independent mechanisms, including CD274 gene amplification, translocation
under active promoter or truncation of 30UTR, which stabilised PD-L1 mRNA (Ansell
et al., 2015; Green et al., 2010; Twa et al., 2014; Kataoka et al., 2016). In contrast, the
overexpression of MYC in primary GC B-cells, studied here, was not associated with an
increase in PD-L1 and CD47 mRNA level but almost two-fold decrease for PD-L1. PD-L1
expression level was bu↵ered by an increased ribosome footprints abundance suggesting
the contribution of post-transcriptional control.
PD-L1 mRNA is known to contain 2 upstream Open Reading Frames (uORF), which
interrupt an e↵ective translation of the PD-L1 protein. MYC-induced activation of the
integrated stress response and subsequent phosphorylation of eIF2↵ was shown to alter
the uORF/PD-L1 balance increasing PD-L1 expression (Xu et al., 2019). Although the
ribosomal footprints abundance in 50UTR of PD-L1 was evident in our data, there were
no di↵erences in the occupation of it between the experimental conditions. This does not
allow me to exclude uORF-mediated regulation of PD-L1 translation, but it is enough to
speculate that, if such exists, it is not mediated by a binary on/o↵ translation of uORF.
Interestingly, PD-L1 translation in cancer cells requires a certain setup of the translational
apparatus. PD-L1 translation was promoted upon eIF2↵ phosphorylation during stress
response, while the ablation of eIF4E phosphorylation at Serine 209 Xu et al. (2019) and
93
depletion of eIF5B reduced PD-L1 expression (Suresh et al., 2020).
So far, the PD-1/PD-L1 inhibition therapy in Non-Hodgkin Lymphoma showed limited
eciency, as most patients do not respond well to monotherapy or the response is temporary
(Zhang et al., 2018b). The role of PD1/PD-L1 axis in aggressive B-cell lymphoma is
complex. T follicular helper cells express high levels of PD1 and are considered important
in regulating B-cell di↵erentiation in GC B-cells and the formation of long-lived plasma
cells (Goodman et al., 2017; Good-Jacobson et al., 2010). However, cell surface expression
of PD-L1 was found only in 11-30% of DLBCL patients (depending on the study). PD-L1
expression was found associated with EBV positive tumours, which had inferior overall
survival (Goodman et al., 2017; Kiyasu et al., 2015). A discrepancy between the direction
of change upon changing MYC levels in established tumours (Casey et al., 2016), including
B-cell lymphoma, and the primary GC B-cells, studied here, may suggest that the nature
of PD-L1 control by MYC may change in the course of tumour development. It would be
interesting to elucidate this cross-talk between the regulation of transcription, translation
and mRNA stability of PD-L1, which could shed light on the susceptibility of lymphoma
cells to PD-1/PD-L1 inhibition.
94
CHAPTER 4
Mutations in RNA helicase DDX3X
facilitate MYC-driven
lymphomagenesis
4.1 Background
Burkitt Lymphoma (BL) is a highly aggressive form of non-Hodgkin lymphoma (NHL)
with 3:1 male:female incidence ratio (Smith et al., 2015; Morton et al., 2006). It presents
in three distinct forms: endemic, sporadic and immunodeficiency-associated. The endemic
Burkitt Lymphoma (eBL) is the most common childhood cancer in sub-Saharan Africa
accounting for nearly half of all paediatric cancers there. The estimated number of new
eBL diagnoses in Africa was 3900 in 2018 (Ha¨mmerl et al., 2019). BL not associated with
immunodeficiency and occurring outside endemic Africa is defined as sporadic Burkitt
Lymphoma (sBL). It is a rare disease accounting for only 2% of all newly diagnosed NHL.
The annual incidence rate of BL in Europe is about 0.36 per 100,000 (Smith et al., 2015).
sBL occurs with two age peaks - 10 and 75 years.
With a growth fraction approaching 100 % and a mass doubling time around 25 hours,
BL is one of the most aggressive tumours in humans. Despite such fulminant onset, it
is potentially curable with intensive chemotherapy. 5-year survival is 87% for patients
younger than 20 (Costa et al., 2013). The toxicity of such regimes poses a substantial risk
for older individuals, thus treatment intensity must often be reduced. 5–year survival for
patients 60 and older is only 25-33%. Moreover, full compliance of multi-agent treatment,
which consist of sequential administration of four to six drugs, is dicult to achieve in
countries with limited access to medical care. It is estimated that the long term survival of
paediatric BL in sub-Saharan Africa is between 30% and 50% and has remained unchanged
since 1970s (Ozuah et al., 2020).
95
Recent advances in molecular mechanisms of lymphoma development open new oppor-
tunities to improve the therapy of BL patients. From a molecular perspective, BL arises
from the germinal centre (GC) stage of B-cell development, which is dedicated to selecting
and expanding mature lymphocytes producing high-anity antibodies (De Silva and Klein,
2015; Klein and Dalla-Favera, 2008; Basso and Dalla-Favera, 2015). This process was
explained in detail in section 1.4. The oncoprotein MYC is essential for the successful
outcome of the GC reaction, but its expression is transient and limited only to a small
portion of GC B cells undergoing positive selection (Calado et al., 2012; Dominguez-Sola
et al., 2012). MYC is a transcription factor that regulates many key cell functions, such
as proliferation, DNA replication, protein biosynthesis, and metabolism. DNA-strand
breaks and non-homologous end joining are the principal mechanisms involved in the
immunoglobulin genes recombination. A side e↵ect of this is the risk of translocation
involving highly expressed immunoglobulin. Translocation between an oncogene MYC and
immunoglobulin heavy or light chain loci is an almost universal feature of BL, observed in
more than 95 % of cases (Swerdlow et al., 2016).
However, sustained MYC upregulation alone is not sucient to drive lymphomagenesis.
MYC overexpression in non-cancer cells triggers apoptosis mediated by both p53-dependent
and p53-independent pathways and amplifies the apoptotic signal in mitochondria through
inhibition of anti-apoptotic BCL2 expression (McMahon, 2014). A mouse model of
human lymphoma revealed that further co-operating mutational mechanisms, such as
PI3K activation, are needed for MYC-induced malignant transformation (Sander et al.,
2012). Genes and pathways recurrently mutated in BL include transcription factors,
such as TCF3/ID3, or FOXO1, SWI/SNF chromatin remodelling complex (ARID1A,
SMARCA4), genes related to apoptosis (TP53, USP7, CDKN2A), GPCR signalling, B-cell
receptor/PI3K signalling and epigenetic regulators (Grande et al., 2019; Schmitz et al.,
2012; Bouska et al., 2017; Grande et al., 2019; Lo´pez et al., 2019; Richter et al., 2012).
Interestingly, in the most recent genomic study of BL (Grande et al., 2019), an RNA
binding protein, DDX3X, was the most frequently mutated gene after MYC, when both
point mutations and copy number changes were considered.
Despite such high recurrence rate of DDX3X mutations, little is known about their role
in BL. Whilst DDX3X mutations have been also reported in chronic lymphocytic leukaemia
(Ojha et al., 2015; Takahashi et al., 2018), medulloblastoma (Jones et al., 2012; Pugh et al.,
2012; Robinson et al., 2012), head and neck squamous cell carcinoma (Stransky et al.,
2011) and NK-T cell lymphoma (Jiang et al., 2015), the function of DDX3X in cancer
remains puzzling and conflicting as it has been classified both as a tumour suppressor
and an oncogene (Soto-Rifo et al., 2012; He et al., 2018). Its dual function has been
reported not only for di↵erent types of cancer but also within the same type (He et al.,
96
2018), which underscores the context-specific e↵ect of DDX3X in malignancy. DDX3X
role in human diseases is not limited to cancer. Heterozygous mutations in DDX3X have
been previously linked to neurodevelopmental disorders associated with autism spectrum,
intellectual disability and seizures (Johnson-Kerner et al., 2020; Lennox et al., 2020a;
Kellaris et al., 2018; Iossifov et al., 2014).
DDX3X is a highly conserved ATP-dependent RNA helicase involved in various aspects
of RNA biology: transcription, splicing, nuclear export, stress granule formation and
resolution, microRNA biogenesis, mRNA translation and decay (Linder and Jankowsky,
2011; Mo et al., 2021). Known RNA-independent functions include regulation of WNT,
and NFB signalling (Pugh et al., 2012; Xiang et al., 2016). DDX3X is located on the
non-pseudoautosomal region of chromosome X and is known to escape chromosome X
inactivation in wide range of tissues (Berletch et al., 2011; Cotton et al., 2015). The Y-
chromosome paralogue, DDX3Y, shares 92% amino acid similarity with DDX3X. Although
widely transcribed, DDX3Y protein is expressed exclusively in spermatogonia (Ditton
et al., 2004; Rauschendorf et al., 2011; Foresta et al., 2000a).
Given the high frequency of DDX3X mutations in BL, I set out to establish the
contribution of this gene to lymphomagenesis in Burkitt lymphoma.
In this chapter:
1. I examine the frequency and distribution of DDX3X mutations in BL,
2. I perform a bioinformatic analysis of multi-omic datasets to elucidate the molecular
role of DDX3X in BL
This project was conducted as a collaborative project with Dr Chun Gong in the Hodson
lab. Dr Gong performed the wet lab experiments whilst I performed all computational
analysis.
97
4.2 Results
4.2.1 Examining the prevalence and distribution of DDX3X mu-
tations
4.2.1.1 DDX3X is preferentially mutated in MYC driven lymphomas
In order to establish the frequency of point mutations in BL in the UK cohort, I reviewed
the results of a 293-gene targeted sequencing panel of 39 cases of previously untreated
sporadic BL. Consistently with previous reports, MYC, ID3, TP53, CCND3, DDX3X,
ARID1A, FOXO1 and SMARCA4 were the most frequently mutated genes. Mutation in
DDX3X was found in 30.8% (12/39) patients and was much more common in males than
females, 11 versus 1, respectively (Figure 4.1 A). . The same analysis was applied to
928 cases of DLBCL (Lacy et al., 2020). Of these, only 5.2% had DDX3X mutation, which
was in line with other recent sequencing studies (Figure 4.1 B) (Chapuy et al., 2018;
Reddy et al., 2017; Schmitz et al., 2012).
Figure 4.1: Frequency of DDX3X mutations in BL
A) Barplot showing mutation frequency (%) for the indicated genes detected using a 293-gene
panel applied to 39 cases of Burkitt lymphoma (sequencing, mutation calling and filtering
performed by Dr. Peter Campbell and Dr. Philip Beer).
B) Barplot showing the frequency of DDX3X mutation across published sequencing studies of
BL and DLBCL.
98
Unlike BL, DLBCL is a highly heterogeneous disease. The recently described Molecular-
High Grade (MHG) subtype of DLBCL shares several similarities with BL which are
reflected by BL-like gene expression signature involving high expression of genes related
to cell cycle, TCF3 signalling and ribosome biogenesis (Sha et al., 2019). In order to
establish a link between DDX3X mutation and MYC-driven lymphomagenesis, I compared
the relationship of DDX3X mutation frequency with MYC status in targeted sequencing
data from the published UK cohort of 550 DLBCL cases with available fluorescence-in-situ
hybridisation (FISH) data for MYC (Cucco et al., 2020). The frequency of MYC locus
rearrangement was significantly enriched in cases with DDX3X mutation (Figure 4.2 A,
Chi-square test, p-value = 0.001): out of 34 patients with DDX3X mutation, 16 (47.06%)
showed MYC translocation, comparing with only 21.04 % (109/516) in DDX3X wild-type
group. In the same study, a comparison of DDX3X mutation frequencies between DLBCL
subtypes revealed remarkable enrichment of DDX3X mutation in MHG group (Figure
4.2 B). Out of 558 cases with available gene expression profile, that was used to identify
DLBCL transcriptomic subtype, DDX3X was mutated in 16.7% of MHG versus 4.3%
and 2.2% of GCB and ABC DLBCL, respectively. The enrichment of DDX3X mutations
in the MHG subtype was significantly higher than in other subtypes (p-value = 0.001,
Chi-squared test).
To validate this finding, I re-analysed RNA-Seq data from a large, publicly available
dataset of 553 DLBCL patients from GOYA clinical trial (McCord et al., 2019b): I
downloaded the raw FASTQ files from Sequence Archive Database (SRA), performed
quality control, alignment to the reference genome and read counting, as detailed in
section 2.2.2. The aim of this analysis was to 1) classify the cases by their transcriptional
subtypes, 2) identify those with DDX3X either mutated or not expressed and 3) compare
the frequency of DDX3X alteration between the subtypes.
I combined two previously developed DLBCL classifiers to, firstly, segregate cases
into ABC, GCB, Unclassified (Reddy et al., 2017) and then, distinguish the MHG group
among samples belonging to GCB-DLBCL subtype, as described by Sha et al. (2019). The
frequency of each identified subtype in GOYA dataset was as expected with the MHG
group accounted for about 10 % of total cases (Figure 4.2 C).
Next, I performed Single Nucleotide Variants (SNVs) calling from paired-end RNA-Seq
data according to the GATK Best Practices Guideline. Because of the lack of germline
control for each sequenced tumour, I imposed rigid criteria to establish DDX3X mutation
status for each sample. I defined DDX3X mutant samples as those with either nonsense,
frameshift or non-synonymous SNVs, localised in evolutionary conserved DDX3X helicase
domain, with the ratio of variant coverage to reference coverage > 0.2. I filtered out all
common population variants reported in the ExAC database.
99
Figure 4.2: DDX3X mutations are enriched in MYC-driven DLBCL
A) Barplot showing the proportion of cases with MYC rearrangement detected by FISH from
a cohort of 550 cases of DLBCL from Cucco et al. (2020) stratified by DDX3X mutation
status.
B) Barplot showing the frequency of DDX3X mutation in 558 cases of DLBCL from Cucco
et al. (2020) stratified by transcriptional subtype.
C) Barplot showing the frequency of DDX3X mutation in 554 cases of DLBCL from GOYA
study McCord et al. (2019b) stratified by transcriptional subtype.
D) Heatmap showing DDX3X mutation status by transcriptional subtype in 553 DLBCL cases
from GOYA trial. Rows represent gene expression signatures which were used to assign
transcriptional subtypes (obtained from Reddy et al. (2017) and Sha et al. (2019)).
100
5 samples with disproportional low expression of DDX3X were also classified as loss-
of-DDX3X. See the section 2.2.8 for a detailed description of the analytic workflow.
Concordant with previous result, DDX3X mutation or absent expression was significantly
enriched in the MHG subtype (19.0%), compared to GCB (8.1%) and ABC (3.0%) DLBCL
(p-value < 105, Chi-squared test) (Figure 4.2 C-D).
The computational evidence for MYC-DDX3X interaction was subsequently tested
in the lab by Dr. Chun Gong in a culture system using transduced primary human
germinal centre B cells. This revealed a competitive advantage in ex vivo GC B cells when
co-transduced with both MYC and helicase-mutant DDX3X. This competitive advantage
was not seen in cells co-transduced with MYC and WT DDX3X or cells co-transduced with
mutant DDX3X and BCL6. These functional experiments confirmed the computational
prediction of co-operation between MYC and helicase-mutant DDX3X.
4.2.1.2 Context dependent pattern of DDX3X mutation in di↵erent cancer
types
Since DDX3X was reported to be both a tumour suppressor and an oncogene, I wondered
if there are any di↵erences in the types of point mutations between cancer types suggesting
the di↵erential role of DDX3X. Hence, I examined the characteristics of DDX3X mutations
downloaded from COSMIC, a Catalogue of Somatic Mutations in Cancer. Although the
total number of mutations reported in COSMIC is biased by the availability of sequencing
datasets corresponding to each tissue type, an interesting observation can be made by
looking at the ratio of disrupting mutations such as nonsense or frameshift. While the
majority of cancers had a varying mixture of disrupting and non-disrupting mutations,
almost all mutations reported in Central Nervous System (CNS), thyroid and pancreas
were of missense type (Figure 4.3 A) This supports the observation that the role of
DDX3X mutations is context-dependent and may di↵er between cancer types.
4.2.1.3 DDX3X mutations in B-cell lymphomas cluster within C-terminal
helicase domain
As a member of the DEAD-box RNA helicase family, DDX3X (and DDX3Y) contains an
evolutionarily conserved helicase core of two RecA-like domains (Linder and Jankowsky,
2011; Mo et al., 2021). The two core domains contain 12 sequence motifs that are involved
in either RNA binding (motifs Q, I, II/DEAD, VI) or ATP binding and hydrolysis (Ia,
Ib, Ic, IV, IVa, V, VI). The helicase core is surrounded by two Low Complexity Domains
(LCDs) that are known to participate in the assembly of RNA-protein aggregations, such
as stress granules, through liquid-phase separation (Molliex et al., 2015; Valentin-Vega
et al., 2016). I investigated the distribution of DDX3X point mutations across the protein
domains in patients diagnosed with BL or DLBCL.
101
Figure 4.3: Types of DDX3X mutations in di↵erent cancer types
A) Barplot showing distribution of DDX3X mutation types in di↵erent cancer types included
in COSMIC database (v.89)
B) Lollipop plot with the distribution of DDX3X mutations identified in this and published
studies of BL and DLBCL over the functional domains and motifs of DDX3X protein.
102
This revealed the accumulation of mutations within the helicase domains, especially
in the C-terminal helicase domain. The mutational hot-spots in BL and DLBCL were:
R488, R475, R311, and R528, R534 (Figure 4.3 B). Some of them are shared with
medulloblastoma and are known to abolish helicase activity. In previous studies, RNA
unwinding assays showed complete loss of helicase activity for R475 mutation (Lennox
et al., 2020a) and almost 100-fold reduction in activity for R534 (Floor et al., 2016). In line
with the mutation pattern observed in COSMIC data and frequent deletions of DDX3X
loci reported in the recent whole-genome sequencing study of BL (Grande et al., 2019),
there were multiple disrupting mutations, many of them localised near N-terminus.
These findings suggest that loss of RNA helicase function is the predominant conse-
quence of DDX3X mutation in B-cell lymphoma.
4.2.1.4 Males with Burkitt Lymphoma and DLBCL are more likely to have
DDX3X mutation
Given the strong male skew in DDX3X mutation frequency observed in the targeted
sequencing panel of BL patients, I attempted to establish whether there is any relationship
between sex and the probability of DDX3X mutation in B-cell lymphoma. In order to and
improve confidence and precision of the e↵ect size estimation, I performed meta-analysis
of the sex skew in DDX3X mutation ratio using previously published sequencing studies
of BL and DLBCL with available sex data. I collected data from 7 published studies with
available DDX3X mutation status and patient sex. Data from the targeted sequencing of
39 patients from this study were also included.
Because patient sex was unavailable for 553 DLBCL patients from GOYA clinical
trial, I decided to classify RNA-Seq samples into male and female using the decision tree
algorithm implemented in rpart R package. In total, the number of patients included in
the analysis was 395 for BL (6 studies) and 2180 for DLBCL (3 studies). Pooled male
to female ratios were 4 for BL and 1.2 for DLBCL. The median percentage of DDX3X
mutated samples was 30.385% for BL and 5.967 % for DLBCL. Single data points and
sequencing type used to call variants shown in (Figure 4.4 A - B, Table 4.1).
The analysis revealed that males diagnosed with BL or DLBCL have approximately
1.23 times the risk of DDX3X mutation compared to females ((Figure 4.4 A-B), Random
e↵ects model p-value = 0.0002).
The allele frequencies suggested that DDX3X mutations were predominantly clonal in
both females and males. In females only one copy of DDX3X was mutated (Figure 4.4
C).
103
Figure 4.4: Metaanalysis of DDX3X mutation associated gender skew in published
DLBCL and Burkitt Lymphoma studies
A) Forrest plot showing the e↵ect sizes for each study. The red bar represents the prediction
interval around the pooled e↵ect shown by the grey diamond.
B) L’Abbe´ plot showing sex skew of DDX3X across this and other studies of BL and DLBCL.
C) Dot plot showing DDX3X mutant allele frequency by sex. Data are taken from this study
and three other BL sequencing studies for which sex and MAF was available (Grande
et al., 2019; Lo´pez et al., 2019; Zhou et al., 2019).
Mantel-Haenszel random-e↵ects model was used to calculate the overall Relative Risk
(RR) and 95% CI. The RR in all studies had a range of 1.06–1.45 and RR of 1.23. The
heterogeneity across studies was assessed by Cochran’s Q test and Tau-squared (p-value
0.8338, 2 = 0.0033). All computations were performed using meta R package.
104
Table 4.1: An Overview of BL and DLBCL datasets used for meta-analysis of sex skew of
DDX3X mutation occurrence
Study Disease
Number of
patients
Male:Female
ratio
Sequencing
type
Abate et al. (2015) BL 20 5.67 RNA-Seq (2x75 bp)
Gong, Krupka et al. (2021) BL 39 5.50 Targeted panel)
Kaymaz et al. (2017) BL 28 2.50 RNA-Seq (2x100 bp)
Lo´pez et al. (2019) BL 21 6.00 WGS
Zhou et al. (2019) 167 20 2.41 WES
Grande et al. (2019) BL 120 1.79 WGS
Cucco et al. (2020) DLBCL 337 1.15 Targeted panel
McCord et al. (2019b) (GOYA) DLBCL 553 0.93 RNA-Seq (2x50 bp)
Reddy et al. (2017) DLBCL 998 1.30 WES
4.2.2 DDX3X regulates ribosome biogenesis and global protein
synthesis
4.2.2.1 DDX3X binds preferentially to mRNA encoding components of core
translation machinery
The existing literature suggests that DDX3X has a versatile role in cell biology encompassing
many aspects of RNA biology, regulation of cell proliferation, stress response and apoptosis.
In order to uncover which of these may be relevant in lymphoma, Dr. Chun Gong performed
immunoprecipitation of the endogenous DDX3X and SILAC mass spectrometry of the
interacting proteins (Figure 4.4 A). Gene ontology analysis of proteins interacting with
DDX3X in at least one cell line revealed a strong enrichment for the proteins participating
in the translation initiation including almost all components of the eIF3 complex, eIF4A,
eIF4E, and eIFG4 (Figure 4.4 B). This is in line with previous works (Lee et al., 2008a;
Soto-Rifo et al., 2012; Shih et al., 2008) reporting association of DDX3X protein with
translation initiation complex. Among interacting proteins, there were also 7 components
of stress granules (SGs). SGs are membraneless assemblies of messenger ribonucleoproteins
(mRNPs) that form from mRNAs stalled in translation initiation in response to stress
(Protter and Parker, 2016; Buchan and Parker, 2009; Jain et al., 2016). SGs-associated
proteins included: DDX1, ATXN2L, NUFIP2, PDCD4, USP10, UPF1, and EWSR1.
These findings suggest that the role of DDX3X in lymphoma cell lines focuses on
translation and protein synthesis, either through participation in translation initiation or
stress granules assembly.
105
Figure 4.5: DDX3X co-immunoprecipitates with essential components of translation
machinery
A) DDX3X-interacting proteins were identified by SILAC-MS following immunoprecipitation
of endogenous DDX3X in U2932 (RRID: CVCL 1896) and Mutu (RRID: CVCL ZY05).
Scatter plot shows log2 SILAC ratios of interacting proteins. Proteins significantly enriched
in both cell lines are labelled. The experiment was performed by Jade Gong.
B) Venn diagram showing overlap of DDX3X-interacting proteins in two lymphoma cell lines -
Mutu and U2932
C) Barplot showing gene Ontology (GO) enrichment of DDX3X-interacting proteins identified
in both cell lines.
106
As an RNA-helicase, DDX3X can bind directly to RNA a↵ecting its fate and function.
In order to identify transcripts bound by DDX3X, we decided to perform individual
nucleotide resolution crosslinking immunoprecipitation (iCLIP). This technique combines
immunoprecipitation of UV-crosslinked protein-RNA complexes with Next Generation
Sequencing, which allows mapping the localisation of protein-RNA complexes with single
nucleotide precision (Hafner et al., 2021). Although a similar technique was used previously
by other groups, they all used HEK293T cells transfected to expressed FLAG-tagged
DDX3X (Valentin-Vega et al., 2016; Oh et al., 2016; Calviello et al., 2021). To investigate
the binding profile of DDX3X at physiological expression levels and take into account
the context-specific functions of DDX3X, it was crucial to perform the experiment with
endogenous DDX3X protein in lymphoid cells. iCLIP was conducted using two lymphoma
cell lines: U2932 and Mutu, as well as non-malignant human GC B cells purified from
discarded tonsil tissue (Caeser et al., 2019). The experiment was performed by Dr. Chun
Gong in at least two biological replicates per condition with an isotype control with IgG
antibody that does not recognise DDX3X. Details regarding the experimental protocol are
described in (Gong, Krupka et al., 2021).
Overall, the number of uniquely mapped reads for DDX3X iCLIP was between 4-24
million which, as expected from this technique, accounted for about 30-40% of the total
number of reads aligned. In contrast, the number of uniquely mapped reads in the IgG
control samples was less than 15,800 suggesting high signal-to-noise ratio. Because the
density of crosslinking sites per gene was highly consistent between replicates (Pearson’s
Correlation Coecient between 0.7 - 0.96), I pooled the data from the same cell type
together to increase the sensitivity of DDX3X binding analysis, see methods 2.2.7.1.
First, I examined the distribution of DDX3X crosslinking sites across genomic regions.
In all cell types DDX3X bound ubiquitously to mature coding transcripts (Figure 4.6
A).
To visualise the precise location of the DDX3X binding site from a mature mRNA per-
spective, I performed a metagene analysis where the aggregated coverage of all crosslinking
sites over all expressed transcripts is plotted relative to the known translation start and
termination site. This showed strong enrichment in DDX3X binding at translation initia-
tion sites (TIS) and further into the open reading frame with another peak in crosslinking
sites density at the translation termination site (TTS) (Figure 4.6 B). These findings
are consistent with our previously demonstrated association of DDX3X with the proteins
of translation initiation machinery (Figure 4.5).
107
Figure 4.6: Binding profile of endogenous DDX3X in lymphoid cells
A) Barplot showing density of iCLIP cross-link sites mapping to the indicated genetic features
is shown for the indicated cell types.
B) Venn diagram showing the overlap between DDX3X-bound transcripts detected in iCLIP
experiments in lymphoma cell lines and primary human GC B cells.
C) Metagene summary of cross-link density across DDX3X-bound mRNA transcripts, showing
length-scaled coding region and 3kb of the 5’ and 3’ untranslated regions. TIS = translation
initiation site, TTS = translation termination site, ORF = open reading frame.
108
As an RNA-helicase DDX3X is known to facilitate translation of transcripts with
complex secondary structures in their 50UTR. To address this behaviour I investigated the
preference in DDX3X binding with regards to the adjacent sequence context, GC content
and RNA secondary structure. De novo motif search revealed no consensus binding motif,
neither the analysis of folding energy profile and GC content around the binding site
showed no pattern. This suggests that the binding of DDX3X to mRNA is not directly
related to any particular RNA sequence context.
Next, I examined whether DDX3X binds to a specific family of transcripts. In order to
define a list of high-confidence DDX3X targets, I filtered out first all binding peaks with
less than 10 crosslinking sites and then, discarded genes with less than 3 iCLIP peaks.
I found substantial overlap between DDX3X targets in all three cell types - 45.73 % of
all high-confidence targets were detected in at least 2 cell types (Figure 4.6 C). The
majority of cell type-specific targets could be explained by the di↵erence in the expression
level of the transcript and by the di↵erence in the total number of uniquely mapped reads
between the groups. Although I had previously hypothesised that the R475 mutation
situated in the RNA-binding domain may abolish interaction with RNA, the strong overlap
of iCLIP targets between Mutu (DDX3X R475S) and the WT cells (U2932, primary GC
B-cells) is evidence that this is not the case. However, I acknowledge that subtle binding
di↵erences are not excluded.
Figure 4.7: DDX3X protein binds predominantly to mRNA of ribosomal proteins
Barplot showing Gene Ontology (GO) enrichment of DDX3X-bound transcripts identified by
iCLIP in the indicated cell types.
BP = Biological Process, CC = Cell component, MF = Molecular Function.
109
Gene ontology analysis of mRNA targets that were shared between at least two cell
types (441 genes) revealed a strong enrichment for mRNAs encoding components of the
core translation machinery, in particular, 45 ribosomal proteins, 9 translation initiation
factors, 67 genes associated with cellular response to stress and 23 30UTR binding genes
(Figure 4.7). High expression level of the core components of translation machinery
enriched here may be considered a confounding factor suggesting that the iCLIP signal is
non-specific. However, the iCLIP peaks spans the broad range of expression levels and
these highly expressed mRNAs do not come up in other iCLIP experiments using identical
protocol (personal communication with Dr. Martin Turner), which supports the specificity
of the DDX3X binding.
These findings suggest that DDX3X binds preferentially to mature transcripts encoding
proteins linked to various aspects of protein translation.
4.2.2.2 DDX3X regulates translation of a subset of expressed transcripts
Links between DDX3X and protein synthesis machinery were revealed by: 1) co- immuno-
precipitation of DDX3X with several translation initiation factors and stress granules
components, and 2) the observation of preferential binding to a subset of mature tran-
scripts, prompted me to form a hypothesis, that DDX3X may alter the translation of its
mRNA targets. To determine which transcripts are sensitive to DDX3X depletion, we
performed transcriptome-wide translational profiling (Ribo-Seq) in lymphoma cell lines.
The basic assumptions, strengths and limitations of Ribo-Seq technique were discussed
in detail in Chapter 3. Sequencing libraries were prepared by Dr. Jie Gao as detailed in
section 2.2.2.2 . The experiment was performed in two lymphoma cell lines: Mutu and
U2932 with two di↵erent DDX3X shRNA and a scrambled shRNA as control. For each
sample, a paired sequencing (RNA and Ribo-Seq) was performed at two time points (24h
and 48h) (Figure 4.8).
Figure 4.8: Identification of DDX3X sensitive transcripts: Experimental setting
Diagram showing experimental design of translational profiling for DDX3X shRNA cells
110
The average number of uniquely mapped reads (non-rRNA ) in the Ribo-Seq samples
was 9,623,015. In all samples, the profile of aligned sequencing reads meets expectations
for Ribo-Seq experiment:
1. fragment length between 26 and 32 nucleotides with a peak at 28, nucleotides,
2. enrichment of fragments in CDS,
3. evidence of three-nucleotide periodicity in the frame preference (Figure 4.9 A-C),
4. characteristic pattern shown in the metagene analysis of the estimated P-site location
showing enrichment of footprints at the start codon and abrupt drop-o↵ at the stop
codon (Figure 4.9 D), performed as described in section 3.1.1
The gene expression measurement with Ribo-Seq and RNA-Seq were highly reproducible
between experimental replicates. The Pearson correlation coecient was higher than 0.97
for all the samples from the same sequencing type. mRNA levels were also strongly
correlated with the ribosome abundance (Pearson correlation coecient 0.92-0.95). All
this together supports a good quality of the experiment.
I sought to classify transcripts into distinct regulatory profiles. The extent to which
change in expression occurs at the level of translation, transcription or both can be
distinguished by juxtaposing the di↵erence in ribosomal footprints with the di↵erence in
mRNA abundance. Because mRNA abundance positively correlates with the number of
ribosomal footprints mapping to this region, the ribosomal footprints density for each
transcript should be normalised by the transcript abundance. The obtained metric reflects
the relative density of ribosomes per transcripts and is known as Translation Eciency
(TE). Interpretation and limitation of TE has been reviewed in section 3.1.3. I dissected
mRNA and translation driven changes using the method described in section 3.1.3. This
is the same strategy that was applied to elucidate translational consequences of BCL6 and
MYC overexpression in the chapter 3.
This analysis revealed that the changes in expression profile after DDX3X depletion
were mainly limited to the level of translation. In DDX3X WT cell line U2932, out of
200 di↵erentially expressed genes, 70 and 90 showed decreased or increased translation
rates respectively (Figure 4.10 A). Because the number of genes di↵erentially expressed
at the level of mRNA was relatively small (40 genes), I decided to simplify the original
classification described in section 2.2.3. The remaining 40 genes di↵erentially expressed in
RNA-Seq data were classified into two groups: mRNA down or mRNA up, depending on
the direction of change.
111
Figure 4.9: Quality control of Ribo-Seq dataset examining translational conse-
quences of DDX3X depletion in U2932 and Mutu
A) Histogram showing the distribution of read length in Ribo-Seq and RNA-Seq experiments.
Ribo-Seq shows a characteristic peak at 28-29 nt, which corresponds to the length of the
ribosome protected mRNA fragment.
B) Heatmap of Ribo-Seq read frame usage by read length, showing read frame restriction
in the 28-31nt ribosome protected fragments, characteristic of ribosome position on the
mRNA transcript.
C) Heatmap of Ribo-Seq read frame usage by gene feature, showing how the characteristic
read frame bias within the CDS but not the UTRs.
D) Metagene plot showing distribution of mapped Ribo-Seq reads to regions of the transcript,
with the characteristic peak at the translation initiation site (TIS) and abrupt drop-o↵ at
translation termination site (TTS).
112
Translationally downregulated genes were enriched for components of the core transla-
tional machinery, in particular, protein constituents of the ribosome (Figure 4.10 B-C).
This was specific to the cytosolic ribosome as no significant di↵erence in TE was observed
for transcripts encoding mitochondrial ribosomal proteins (Figure 4.11 B p-value =
2.2 · 1016, Kolmorgorov-Smirnov test). The group of translationally downregulated genes
also included ODC1, a known gene translationally regulated by DDX3X (Calviello et al.,
2021). No significantly enriched terms were found in the translationally upregulated mRNA
up or mRNA down groups. Genes in TE down group were more likely to be identified
as iCLIP targets (Figure 4.11 A-C, p-value=3 · 1014, Chi-squared test). Identified
translationally regulated genes spanned broad range of mRNA expression levels.
Figure 4.10: DDX3X-sensitive transcipts are enriched for the components of the
cytosolic ribosome
A) Scatter plot comparing changes in mRNA abundance (RNA-Seq) with changes in ribosome
footprint density (Ribo-Seq) following shRNA depletion of DDX3X in U2932. Data are
from eight replicate knockdowns using two di↵erent shRNAs. Transcripts with altered
translational eciency (TE) or mRNA abundance are indicated by color.
B) Cumulative distribution of TE change in U2932 following shRNA depletion of DDX3X is
plotted for genes encoding cytosolic or mitochondrial ribosome proteins, or all other genes.
P-value calculated using Kolmogorov-Smirnov test.
C) Barplot showing Gene Ontology (GO) enrichment of genes with reduced TE following
DDX3X depletion.
113
Figure 4.11: Di↵erentially translated mRNAs following DDX3X depletion in U2932
are more likely to be bound by DDX3X
A) Scatter plot showing changes in TE plotted against mRNA abundance. Genes identified as
di↵erentially translated are coloured. Ribosomal proteins and ODC1 (a known DDX3X-
regulated gene) are indicated.
B) Scatter plot showing changes in TE plotted against cross-lining density from iCLIP
experiments.
C) Barchart showing the proportion of transcripts with di↵erential translation identified as
direct targets of DDX3X in iCLIP experiments. The number of genes within each category
is indicated. Adjusted p-values (Fisher test) are shown and reflect the comparison of each
category with stable genes.
D) Violin plot showing GC content distribution across di↵erent categories of di↵erentially
expressed genes. Adjusted p-values (Wilcoxon test) are shown and reflect the comparison
of each category with stable genes.
E) Scatter plot showing the log2 fold changes in TE against 50TOP score, which reflects the
strength of the 50TOP motif. Canonical 50TOP genes are indicated.
114
The helicase activity of DDX3X may be linked to preferential translation of mRNAs
with complex 50UTR structure. I compared GC content and RNA folding energy of
50UTR sequence in translationally regulated mRNAs to explore this hypothesis. Although
translationally downregulated genes had slightly higher GC content (adjusted p-value =
0.0092, Wilcoxon rank sum test), the RNA folding energy was not significantly di↵erent
between the regulatory groups (Figure 4.11 D). Because translational control of ribosomal
proteins is linked to certain RNA motifs, such as 5 terminal oligopyrimidine (50TOP) motif,
I explored also this option. I downloaded the list of canonical 50TOP and 50TOP scores
from a recent study surveying 50TOP sequences in the human transcriptome (Philippe
et al., 2020). A striking finding was that all canonical 50TOP mRNAs, which include
ribosomal proteins and few translation factors, showed a decreased translation eciency
following DDX3X depletion (p-value < 2.2e-16, Wilcoxon rank sum test), suggesting that
DDX3X may regulate translation of its targets through 50TOP motif (Figure 4.11 E).
Next, we validated the prediction of Ribo-Seq in three ways:
1. Mass spectrometry of DDX3X shRNA U2932 cells:
The results of Ribo-Seq experiments were confirmed at the protein level with a tandem
mass tag (TMT) mass spectrometry. In line with the Ribo-Seq results, the abundance
of almost all cytosolic ribosomal proteins was reduced after shRNA depletion of
DDX3X in U2932 cells (Figure 4.12 A). GO analysis of proteins with reduced
abundance in the DDX3X shRNA group showed enrichment for terms associated with
protein synthesis, and the protein constituents of the ribosome, therefore reiterating
the Ribo-Seq findings (Figure 4.12 B-C). Interestingly, the depletion of R475S
mutant DDX3X from Mutu had minimal e↵ect on translation or mRNA abundance
of specific transcripts, which supports the hypothesis of loss-of-function nature of
this mutation(Figure 4.12 B).
2. ProteomeHD database of protein co-regulatory groups:
To examine the DDX3X interaction in a broader context, I queried Proteome HD
database (Kustatscher et al., 2019). This repository integrates data from 5,288 mass-
spectrometry runs spanning diverse human tissue types and biological conditions to
create a map of the functional associations between co-expressed proteins. Examining
co-regulation landscape of DDX3X showed 81 proteins, of these 27 were parts of
the GO Cytosolic Ribosome group (adjusted p-value = 7.2 · 1042) and 8 were
components of the core translation machinery (Figure 4.13). Overall, GO terms
with the strongest enrichment among DDX3X-interacting proteins were: RNP
complex, mRNA metabolism, translation initiation and the cytosolic ribosome.
115
Figure 4.12: Validation of Ribo-Seq results with proteomic profiling of DDX3Xsh
cells.
A) Heatmap showing altered abundance of ribosomal proteins in mass spectrometry analysis
performed following shRNA DDX3X depletion in U2932.
B) Heatmap showing fold change in RNA-Seq, RiboSeq and TE across eight replicate knock-
downs for all di↵erentially translated genes. Protein abundance changes are shown (Mass
Spec). DDX3X targets identified from iCLIP and genes encoding ribosomal proteins (RPs)
are indicated by purple or red bands respectively.
C) GO terms enriched amongst proteins with reduced abundance in proteomic profiling (MS)
following DDX3X depletion. Results from RiboSeq experiments are included for comparison.
116
Figure 4.13: Validation of Ribo-Seq results with ProteomeHD analysis of proteins
co-regulated with DDX3X and OPP-assay of DDX3X depleted cells
A) Map of the co-regulated human proteome plotted using data downloaded from Proteome-HD
(Kustatscher et al., 2019). The 81 proteins identified as being statistically co-regulated with
DDX3X are coloured. Ribosome proteins identified as being co-regulated with DDX3X are
shown in red.
B) Barplot showing global protein synthesis quantified by OPP incorporation at the indicated
time points following shRNA depletion of DDX3X in U2932 and normalized to control
shRNA. Data shows mean+SEM, * p < 0.05, *** p < 0.001, ANOVA with multiple
comparison testing, n=4 replicate experiments. Performed by Dr. Chun Gong.
117
3. O-propargyl-puromycine (OPP) assay for measuring global translation
rate:
Translational downregulation of the ribosomal proteins after shRNA depletion of
DDX3X in U2932 was associated with a reduction in global synthesis rate (Figure
4.13 B). A similar e↵ect was observed in primary GC B-cells transduced with
helicase-dead DDX3X mutants (Gong, Krupka et al., 2021).
Taken together, these results reveal that DDX3X promotes translation of transcripts
encoding core components of translation machinery, in particular ribosomal proteins. The
net e↵ect of this is the regulation of global protein synthesis capacity.
4.2.3 Deregulation of MYC in primary GC B-cells increases ri-
bosome biogenesis and triggers ER stress.
The observation that loss-of-function mutations in DDX3X a↵ect ribosomal proteins
synthesis, and hence decrease global translation load, contrasts sharply with the known
role of MYC in promoting ribosome biogenesis. Relevant experimental system is key to
understand molecular mechanism of lymphoma. GC B-cells are the cell-of-origin of BL, so
it was essential to investigate the the transcriptional consequences of MYC overexpression
in this system. Previous diculties with genetic manipulations of ex vivo human B-cells
means that this experiment has not been performed before. For this I analysed the
RNA-Seq data I generated in chapter 3 (section 3.2) Transduction of MYC alone into
primary GC B-cells triggers apoptosis, which is not the case for MYC transduction into
established lymphoma cell lines. Therefore an experiment where MYC is overexpressed
together with BCL2 (MYC-t2A-BCL2), should allow me to infer transcriptional response
to MYC in the context of cell-of-origin of BL. Di↵erential expression analysis comparing
MYC-BCL2 with BCL2 alone revealed massive upregulation of a ribosome biogenesis
signature. This was specific to MYC, as no such pattern was seen for BCL6-BCL2 cells. In
addition, Gene Set Enrichment Analysis (GSEA) revealed strong upregulation of Unfolded
Protein Response (UPR) signature. Moreover, ERN1, a key sensor of UPR, was one of
the most significantly upregulated genes in MYC transduced GC B-cells.
UPR is an adaptive response to disturbance of the Endoplasmic Reticulum (ER)
homeostasis, which results in the accumulation of misfolded proteins in the ER lumen. It
triggers the expression of ER chaperones aiming to contain misfolded proteins through
ER-associated degradation (ERAD) or commit the cell to apoptosis (Walter and Ron,
2011; Ruggiano et al., 2014; Hetz et al., 2015; Zhang et al., 2020). Molecular markers
of UPR involve alternative splicing of transcription factor XBP1 initiated by ERN1 and
phosphorylation of eIF2↵ by EIF2AK3 (PERK) (Ron and Walter, 2007).
118
Figure 4.14: Deregulation of MYC in primary GC B-cells increases ribosome
biogenesis and triggers Unfolded Protein Response
A) Heatmap showing mRNA expression level of genes belonging to the gene set Ribosome
Biogenesis (GO: 0042254) in human GC B-cells transduced with BCL2, BCL6-2A-BCL2,
or MYC-2A-BCL2
B) Gene set enrichment analysis (GSEA) of RNA-seq from human GC B-cells transduced
with MYC-2A-BCL2 compared to BCL2 alone, showing enrichment of gene sets related to
MYC, UPR, and mammalian target of rapamycin complex 1 (mTORC1) signalling. Genes
ordered according to DESeq2 test statistics (decreasing order).
C) Volcano plot (log2 fold change against -log10(FDR) from di↵erential expression analysis
MYC-2A-BCL2 compared to BCL2 alone). The position of MYC and ERN1 indicated.
119
Elevated levels of phosphorylated eIF2↵ and spliced isoform of XBP1 in primary GC
B-cells transduced with MYC-t2A-BCL2 were confirmed experimentally by Dr. Chun
Gong. UPR turned out to be, at least partly, related to MYC-induced apoptosis. Primary
GC B-cells transduced with MYC alone and treated with rapamycin, an allosteric inhibitor
of mTORC1, had significantly lower protein synthesis associated with a modest reduction
in apoptosis rate. This experiment was performed by Dr Chun Gong (Gong, Krupka et al.,
2021).
These results show that deregulation of MYC in primary GC B-cells is associated with
a higher protein synthesis rate and induction of ER stress response. Loss of DDX3X,
by limiting global protein synthesis might, protect cells from proteotoxic stress, thus
alleviating MYC-induced apoptosis. Support for this hypothesis comes from a series of
experiments described in the next section.
4.2.4 DDX3X mutation interferes with endoplasmic reticulum
stress response
4.2.4.1 DDX3X R475C mutation is associated with suppression of unfolded
protein response in U2932 cells
A combination of computational and experimental evidence allowed us to hypothesise that
loss-of-function mutations in DDX3X may play a similar role to rapamycin treatment in
protecting cells from MYC-induced apoptosis. If deregulation of MYC in primary GC
B-cells is associated with proteotoxic stress and apoptosis, DDX3X mutations may lower
translation load, alleviate ER stress, thus decreasing apoptosis rate.
To interrogate the regulatory network of DDX3X mutation in lymphoid cells, I analysed
RNA-Seq data from an experiment performed by Dr. Gong comparing CRISPR–edited
clones of U2932 cells expressing R475C helicase mutant DDX3X. Control clones with a syn-
onymous mutation in endogenous DDX3X were created in parallel. Di↵erential expression
analysis comparing CRISPR-Cas9 DDX3X clones revealed strong downregulation of the
key regulators of ER stress response - ERN1 and XBP1. Moreover, the mRNA expression
pattern of R475C-edited clones resembled the profile of samples with the strongest deple-
tion of DDX3X following shRNA treatment. GSEA analysis of DDX3X-mutant clones and
RNA-Seq of shDDX3X depleted cells also revealed striking overlap. In both comparisons
‘MYC targets V1’, ‘MTORC1 signalling’ and ‘Unfolded Protein Response’ terms were
among the most downregulated gene sets.
120
Figure 4.15: DDX3X R475C mutation is associated with downregulation of genes
associated with unfolded protein response in U2932 cells
A) GSEA analysis of RNA-seq data comparing DDX3X R475C-edited or control clones (left),
and DDX3X shRNA knockdown experiments (right). Genes ordered according to DESeq2
test statistics (decreasing order).
B) Heatmap showing genes that are di↵erentially expressed between control and homozygous
R475C edited clones (left). The same genes are shown for shRNA knockdown (right).
For shRNA experiments samples are ordered left to right by DDX3X mRNA expression.
The top bar indicates the expression of DDX3X mRNA showing how stronger knockdown
recapitulates the signature seen in R475C-edited clones.
C) Boxplot showing relative expression of the Unfolded Protein Response (UPR) marker
transcripts ERN1 (encoding IRE1) and XBP1 mRNA in RNA-seq from DDX3X R475C-
edited clones. Statistical significance from di↵erential expression analysis (DESeq2).
121
Figure 4.16: DDX3X R475C mutation is associated with downregulation of proteins
associated with ER stress in U2932 cells
A) Heatmap showing proteins with altered abundance in proteomic profiling of DDX3X
R475C-edited clones. Proteins included in the Gene Ontology (GO) terms Endoplasmic
Reticulum (ER), GO: 0005783, and ER-associated protein degradation pathway (ERAD),
GO: 0036503, are indicated by red and orange highlighting, respectively.
B) Barplot showing the statistical significance of top GO terms enriched among proteins with
decreased expression in DDX3X R475C-edited clones. BP, biological process; CC, cellular
component; MF, molecular function.
122
Given the evidence of DDX3X driving gene expression changes predominantly at the
level of translation, the R475C mutant and control clones were subjected to proteomic
profiling. In line with previous experiments, GO terms linked to protein processing in
the ER and ER stress were enriched in protein downregulated in R475C mutant clones.
Nearly one-third of downregulated proteins were related to GO terms associated with the
ER or ER stress.
I conclude that the R475C mutation can be considered loss-of-function and is associated
with suppression of expression of genes related to unfolded protein response.
4.2.4.2 DDX3X mutation is associated with suppression of unfolded protein
response in BL patients
In an orthogonal approach, I examined gene expression profiles in human biopsy samples
obtained from two published BL datasets (Grande et al., 2019; Schmitz et al., 2012).
I downloaded raw FASTQ files from Sequence Read Archive, performed quality check,
alignment to the reference genome and di↵erential expression and GSEA analysis comparing
DDX3X mutant to wild-type samples. Expression pattern of DDX3X mutant cases mirrored
the changes observed in our CRISPR-edited clones: reduced expression of ‘MYC targets’,
‘MTORC1 signalling’ and ‘Unfolded Protein Response’ sets in GSEA accompanied by
strong downregulation of ERN1 and XBP1 mRNA levels.
The link between DDX3X mutation and ER stress response was confirmed experimen-
tally by Dr. Chun Gong. Co-transduction of primary GC B-cells with MYC and DDX3X
mutant recapitulated the pattern observed after rapamycin treatment: loss-of-function
DDX3X mutations (R475C and K230E) abrogated both the MYC-induced apoptosis as
well as the increase in the global translation rate caused by MYC. DDX3X mutation
was also able to alleviate ER stress response induced by treatment with thapsigargin
(Gong, Krupka et al., 2021), which inhibits the sarco/endoplasmic reticulum Ca2+ ATPase
(SERCA) thereby inducing ER stress .
Taken together, I conclude that loss-of-function DDX3X mutations can counteract the
e↵ects of MYC to drive global protein synthesis and trigger proteotoxic stress. This also
reveals a potential vulnerability of MYC-driven lymphomas to drugs inducing ER stress.
123
Figure 4.17: Downregulation of genes associated with unfolded protein response in
BL patients with DDX3X mutation
A) GSEA analysis of RNA-seq data from the indicated studies reanalyzed to compare cases of
sporadic BL with either WT or mutant DDX3X. Genes ordered according to DESeq2 test
statistics (decreasing order).
B) Relevant gene sets downregulated in the presence of mutant DDX3X or the relative
abundance of the Unfolded Protein response (UPR) transcripts ERN1 and XBP1. Statistical
significance is from DESeq2.
C) Heatmap showing mRNA expression in RNA-Seq of GSEA core enrichment genes in the
gene set Hallmark Unfolded Protein Response in BL biopsies in two published Burkitt
lymphoma RNA-Seq data sets and in U2932 DDX3X R475C edited clones.
124
4.2.5 Up-regulation of DDX3Y in established tumours rescues
loss of DDX3 helicase activity
Context-dependent e↵ect of DDX3X activity was a recurrent finding in this and previous
studies. While loss-of-function DDX3X mutation was beneficial for primary GC B-cells
co-transduced with MYC, it triggered apoptosis in established lymphoid cell lines (Gong,
Krupka et al., 2021). Because upregulated translation rate is a feature of established
tumours, and the ability to increase global protein synthesis is essential for MYC-driven
lymphomas (Barna et al., 2008), we hypothesised that there must be a way to compensate
for the reduced translation capacity during the latter stages of lymphomagenesis.
Both BL and MHG subtypes of DLBCL have strongly skewed sex ratios in favour of
males and DDX3X mutations are more frequent in male-skewed cancers (Alkallas et al.,
2020; Dunford et al., 2017). DDX3X shares almost 92% amino–acid and 70% nucleotide
sequence identity with the Y–chromosome homologue, DDX3Y. At the functional level, it
is redundant with DDX3X in regulating protein synthesis (Venkataramanan et al., 2020).
Firstly, DDX3Y can rescue a decrease in total translation after depleting DDX3X and
secondly, the translation profile of single transcript in male-derived colorectal cancer HCT
116 cells is indistinguishable (Venkataramanan et al., 2020). DDX3Y is widely transcribed
in many adult tissue types, but it is not expressed at the protein level anywhere except
spermatogonia (Ditton et al., 2004; Foresta et al., 2000a; Rauschendorf et al., 2011). To
establish the status of DDX3Y expression in lymphoid cells, I examined changes in mRNA
abundance in a previously published RNA-Seq dataset including di↵erent states of GC
B-cells (processed by me from FASTQ files) (Caeser et al., 2019). I did not observe
significant changes in DDX3X mRNA comparing freshly isolated GC B-cells, established
lymphoma cell lines, as well as transduced and transformed GC B-cells. Finally, no
significant di↵erence in DDX3X mRNA level was observed between DDX3X mutant and
wild–type samples in male BL patients (Grande et al., 2019). However, the coverage
of ribosomal footprints in the DDX3Y region from Mutu cells was similar to known
protein coding regions with similar expression level. In line with that, immunoblotting
for DDX3Y in male lymphoma cell lines performed by Dr. Chun Gong revealed strong
protein expression. The same was observed in patient-derived BL xenografts and 5 primary
biopsies from male patients (Gong, Krupka et al., 2021). This is consistent with previous
works proposing that DDX3Y protein expression in testis is regulated predominantly at
the level of translation through alternative usage of translation initiation site in 50UTR of
DDX3Y (Jaroszynski et al., 2011). I questioned whether DDX3Y translation might be
directly influenced by DDX3X. However, I found no evidence from the iCLIP or Ribo-Seq
experiments that the DDX3Y transcript was a direct target of DDX3X. Nevertheless,
the link between loss-of-function mutation in DDX3X and protein expression of DDX3Y
125
remains strong. In vivo tumorigenesis experiment performed by Dr. Chun Gong to validate
this observation showed that primary GC B-cells with deleted DDX3X, transduced with
MYC-2A-BCL2 and implanted subcutaneously in Matrigel into immunodeficient mice,
forms tumours and start to express DDX3Y protein seven weeks after injection. This
suggests that DDX3Y expression at the protein level is unique to transformed B-cells.
Although indirectly, the expression of DDX3Y is linked to loss of DDX3X, but the exact
mechanisms of this is unclear.
Figure 4.18: Expression of DDX3Y and DDX3X in primary GC B-cells and
lymphoma cell lines
A) Boxplot showing the mRNA expression level of DDX3X and DDX3Y in Primary GC B-cells,
GC B-cells transduced with BCL6-BCL2 or MYC-BCL2 constructs and in lymphoid cell
lines (left panel) and in BL patients for males and females separately (right panel). VST,
Variance Stabilising Transformation
B) Scatter plot of mean mRNA and ribosomal footprints abundance (VST transformed)
showing the expression level of DDX3Y in comparison to other genes. DDX3X, DDX41
(ubiquitously expressed RNA helicase), FOXO1 (BL oncogene) and EIF3J (translation
initation factor) were indicated for comparison.
C) Coverage plot of Ribo-Seq and RNA-Seq reads in the DDX3Y region in representative
samples from Mutu cells
126
4.3 Discussion
Main findings
Although recurrent DDX3X mutations have been identified in a variety of malignancies, the
molecular role of DDX3X in malignancy remains puzzling. The versatile function of DDX3X
in regulating multiple stages of RNA biogenesis is mirrored by its complex and context-
specific role in tumorigenesis - both oncogenic and tumour suppressor activity has been
reported. This study discusses the function of DDX3X in MYC-driven B-cell lymphoma
revealing functional cooperation between MYC and mutant DDX3X. Deregulation of MYC
expression, through translocation to highly expressed immunoglobulin loci or mutations
increasing protein stability, is central to the development of Burkitt Lymphoma and Double
Hit or Molecular High-Grade subtype of DLBCL. All three diseases share the germinal
centre origin and are associated with poor clinical prognosis. The model proposed here
highlight the vulnerability of MYC-driven lymphoma to proteotoxic stress opening an
attractive opportunity for therapeutic intervention. By integrating data from di↵erent
high-throughput sequencing techniques I show that the e↵ect of DDX3X mutations
is mainly mediated by changes in translation of selected transcripts, predominantly
ribosomal proteins, which in turn controls global translation intensity. The requirements
for translation load change during lymphomagenesis, so that the state of decreased
translation capacity might not meet the demands of a fully established tumour. By
sequential deregulation of DDX3X and DDX3Y, it is possible to bu↵er global protein
synthesis to support these changing needs. Therefore drugs that disrupt this delicate
balance of translation load and proteotoxic stress may prove e↵ective against MYC-driven
lymphoma.
Context specific role of DDX3X in malignant cells
DDX3X mutations have been widely studied in medulloblastoma, the most common brain
tumour in children, where they are associated with Wingless (WNT) and Sonic hedghog
(SHH) subtypes (Northcott et al., 2017; Patmore et al., 2020). The pattern of DDX3X
mutations in medulloblastoma di↵ers from what I have observed in BL, which suggests
a di↵erent molecular role of DDX3X mutant in the two tumours. While nonsense and
frameshifts mutations are frequent in BL, they are never observed in medulloblastoma.
Moreover, amplification of MYC family genes, which is a hallmark of Group 3 and Group 4
medulloblastoma subtype is never accompanied by mutations in DDX3X (Northcott et al.,
2017), indicating that the cooperative e↵ect between DDX3X and MYC is also context-
specific. DDX3X has been shown to regulate brain development and mediates tumour
suppressing stress and inflammasome response (Patmore et al., 2020). Medulloblastoma-
127
associated DDX3X mutations promote oncogenic WNT signalling directly in helicase-
independent fashion through an association with CSNK1E protein (Cruciat et al., 2013;
Pugh et al., 2012) and indirectly, through prevention of pyroptosis, inflammatory apoptosis
mediated by inflammasome activation coupled with WNT signalling (Huang et al., 2020;
Samir et al., 2019; Patmore et al., 2020). I did not find compelling evidence of the
association between DDX3X and CSNK1E in B-cells. Other previously reported individual
targets of DDX3X include ODC1 (Van Steeg et al., 1991; Calviello et al., 2021), KLF4
(Cannizzaro et al., 2018) and MITF (Phung et al., 2019). Of which, only ODC1 was found
to be DDX3X sensitive in lymphoid cells highlighting the strong tissue-specific context of
DDX3X activity. In contrast, I collected multiple orthogonal lines of evidence to support
the regulation of ER stress and protein synthesis as the dominant phenotype of DDX3X
mutations in MYC-driven lymphomas. This, in light of these data, is achieved through
the regulation of translation of selected transcripts encoding the core components of the
translation machinery. DDX3X–DDX3Y axis maintains the balance between translation
load and proteotoxicity that is stage-specific and allows the germinal centre B-cells for
adaptation to deregulated MYC expression. This study, however, does not allow me to
exclude the possibility that loss-of-function mutations in DDX3X exert additional functions
that may converge with functions of DDX3X reported in other tissue types.
Role of DDX3X as auxiliary translation factor
DDX3X is reported to play multiple roles in RNA biology, however, this study indicates
that in primary GC B-cells and lymphoma cell lines, DDX3X acts predominantly at the
level of translation. The role of DDX3X in regulating protein synthesis has been studied
previously revealing both activating and repressing functions. Although this study does
not define the precise mechanism of translational repression upon loss of DDX3X in B-cell
lymphoma, at least two models could potentially explain this phenomenon.
Firstly, DDX3X is known to act as an auxillary translation factor directly interacting,
independently on RNA, with eIF3, and 40S subunit of the ribosome (Geissler et al.,
2012; Lee et al., 2008b). DDX3X induces conformational changes of the 43S pre-initiation
complex promoting the release of translation initiation factors and joining the 60S ribosomal
subunit leading to 80S complex formation (Geissler et al., 2012). In line with this study,
the role of DDX3X in translation seems to be limited mainly to the initiation stage. In
lymphoid cells, we observed that DDX3X protein co-immunoprecipitates with almost all
components of the eIF3 complex and DDX3X iCLIP binding profile resembles the pattern
of known translation initiation factors (Calviello et al., 2021). In fact, DDX3X has not been
found in polysome fractions of human hepatoma cells by Geissler et al. (2012) reinforcing
translation initiation as the main point of DDX3X mediated translational control.
Although DDX3X depletion a↵ects translation intensity globally, it is worth mentioning
128
that Geissler et al. (2012) found that only about 50 % of the newly assembled ribosomes
contained DDX3X, which suggests that regulation of translation by DDX3X might be
limited only to a subset of transcripts. In fact, as an RNA-helicase, DDX3X can participate
in the unwinding of highly structured regions within 50UTRs, thus, facilitating translation
initiation, but the evidence underlying this activity is puzzling. Soto-Rifo et al. (2012)
and Calviello et al. (2021) work on DDX3X showed that transcripts with complex 50UTR,
originating from viral or host transcription respectively, are sensitive to DDX3X depletion.
This is in contrast to Geissler et al. (2012), who argue that depletion of DDX3X had
a substantial impact on the translation of all studied viral transcripts regardless the
complexity of the leader sequence.
Almost complete loss of helicase activity for BL-associated DDX3X mutant suggests
that the molecular function of DDX3X in BL is helicase-dependent. This would argue
in favour of the 50UTR structure as a key determinant for DDX3X-sensitive transcript.
Although, DDX3X-sensitive transcripts had slightly higher GC content, their RNA folding
energy of their 50UTR was not significantly di↵erent from other genes suggesting that
other factors may be involved.
An alternative hypothesis linking DDX3X with translation regulation of the core compo-
nents of translation machinery is 50TOP-mediated regulation. 50 terminal oligopyrimidine
(50TOP) motifs are situated in the 50UTR of selected transcripts. The translation of
50TOP mRNAs is inhibited by LARP1, an RNA-binding protein, which is inactivated by
mTORC1 complex. There is a possibility that DDX3X competes with LARP1 to control
translation of 50TOP transcripts or acts as 50TOP translation modulator. Apparently,
almost all canonical 50TOP mRNAs have negative TE changes after DDX3X knock-down
in U2932, but further experiments are needed to confirm if translational control by DDX3X
is mediated through 50TOP and/or mTOR control.
Di↵erent DDX3X expression levels may promote di↵erent cellular responses
The observation of dose-dependent mRNA expression profile of DDX3Xsh cells, which
resembles the profile of R475C cells only when depletion is the strongest could suggest that
even small amount of DDX3X is sucient to maintain translation of target transcripts.
Dual, dose-dependent role of DDX3X in translation has been reported previously. While
moderate overexpression of DDX3 stimulated translation Geissler et al. (2012), massive
overexpression lead to a halt in global translation. Given the versatile function of DDX3X
in regulating di↵erent aspects of RNA biology, there is a possibility that di↵erent functions
require precise titration of DDX3X levels in the cell.
In addition to the regulation of translation initiation, studies of the yeast homologue
of DDX3X, Ded1, show that it is able to nucleate stress granules (SGs) formation (Hilliker
et al., 2011). While sequestration of selected transcripts in SGs does not require Ded1
129
helicase activity, the reverse process of releasing mRNA into the cytoplasm does so,
thereby promoting their translation (Hilliker et al., 2011). Indeed, SGs formation in human
cells was reported in studies using overexpressed mutant DDX3X (Lennox et al., 2020a;
Valentin-Vega et al., 2016). Valentin-Vega et al. (2016) argue that low complexity domains
flanking the helicase core of DDX3X are the main e↵ector of SGs assembly. This could
explain the observed toxicity of supraphysiological levels of either wild-type or mutant
DDX3X (Lennox et al., 2020a; Valentin-Vega et al., 2016).
Pro-oncogenic properties of decreased translation load
The process of oncogenesis is typically associated with an increased translation load
that meet the demands of deregulated cell cycle and growth (Ruggero, 2013). However,
in several malignancies, an opposite strategy has evolved. In neuroblastoma, the most
common extracranial tumour in children, adaptation to limited nutrient supply is mediated
by the activation of eEF2-kinase, which suppress translation elongation (Leprivier et al.,
2013). It is also interesting that this e↵ect was found predominantly in MYC-N-driven
tumour model (Delaidelli et al., 2017) suggesting that tight regulation of translation
load might be a general feature of malignancies deregulation of the MYC gene family.
RUNX1 deficiency, which is common in myelodysplastic syndrome and acute myelogenous
leukaemia (AML), reduces the rate of ribosome biogenesis making the haematopoetic
stem cells (HSCs) resistant to stress, thus, providing a competitive advantage over normal
HSCs (Cai et al., 2015). Another example comes from a mouse model of medulloblastoma,
where an activating mutations in PERK are essential for premalignant cells to decrease
translation load, which is restored at later stages of tumour formation (Ho et al., 2016).
PERK (EIF2AK3) is a key activator of the integrated stress response, which target 50cap
translation initiation. Similarly to our model, lowering of global translation at early stages
of tumorigenesis decreases ER stress and protects cells from apoptosis (Ho et al., 2016).
The conclusion that suppressed translation is advantageous for MYC-induced lym-
phomagenesis might be inconsistent with studies using existing mouse models of MYC-
driven lymphoma. Haploinsuciency for Rpl24 in Eµ-Myc mouse, which overexpress Myc
in the B-cell compartment, decreases global protein synthesis and delayed the lymphoma
onset (Barna et al., 2008). This is radically di↵erent to the two-stage model proposed
in this study. Genetic deletion of Rpl24 induces permanent suppression of translation,
whilst loss of DDX3X can be rescued by expression of DDX3Y protein. Thus the level of
translation can be tuned to changing requirements of the newly forming tumour. This
hypothesis relies strongly on the assumption that DDX3X and DDX3Y are redundant
in their molecular function. Germline heterozygous mutations in DDX3X has long been
linked to cortical malformations, intellectual disability and autism-spectrum disorders
in females (Lennox et al., 2020b). Males with DDX3Y mutation, in turn, present with
130
infertility, but no cognitive dysfunction has been observed suggesting that DDX3X cannot
compensate for loss of DDX3Y (Ferlin et al., 2003; Foresta et al., 2000b). This might be
attributed to the di↵erence in expression pattern between tissue types, however, I cannot
exclude the possibility that the ectopic expression of DDX3Y in transformed B-cells is
more than just rescuing DDX3X insuciency. Nevertheless, the evidence supporting the
overlap in molecular role of DDX3X and DDX3Y is rich. Firstly, DDX3X and DDX3Y are
92% identical in the amino-acid sequence. Secondly, recently published work reveal that
substitution of DDX3X with DDX3Y completely recapitulates gene expression profile at the
level of transcription and translation (Venkataramanan et al., 2020). Finally, we observed
that DDX3Y bound the same mRNA targets as DDX3X and had a similar influence on
regulating global protein synthesis (Gong, Krupka et al., 2021). The ability to reactivate
DDX3Y protein expression may explain the observed sex bias towards males in BL and
other cancers with recurrent DDX3X mutations. Given the importance of restored protein
synthesis capacity in mature tumours, a natural question arises: what compensatory
mechanism exists in female lymphomas, who also (but at much lower frequencies) present
with DDX3X mutation. A review of mutation co-occurrence in BL exome sequencing
data (Grande et al., 2019) shows that mutations in SIN3A are enriched (24 % vs. 6%;
p-value < 0.05, Fisher Exact test) in the presence of DDX3X mutation. SIN3A is a
transcriptional corepressor that antagonises MYC directly interacting with MXD1-MAX
heterodimers (Nascimento et al., 2011). This observation allows me to speculate that
SIN3A inactivation at the latter stages of lymphomagenesis could increase MYC activity
and thereby global translation capacity. Other complementary mechanisms, genetic or
non-genetic, restoring protein synthesis or increasing MYC activity are also possible. In
contrast, this problem does not exist in males. Ectopic expression of DDX3Y has been
found in all investigated cell lines. Moreover there are almost no mutations in DDX3Y
across published lymphoma datasets, which suggests that functional DDX3Y is important
for successful lymphomagenesis.
The importance of DDX3X/DDX3Y axis in MYC-driven lymphoma opens
new therapeutic opportunities
The results presented here reinforce therapeutic relevance of drugs targeting ER stress for
the treatment of MYC-driven lymphoma. Firstly, they support the vulnerability of the
lymphomagenesis process to proteotoxic stress. This is especially interesting in the light
of the recent results from the REMoDEL-B study examining bortezomib as an addition to
standard R-CHOP therapy in DLBCL. An unexpected outcome of a post-hoc analysis was
that the addition of the proteasome inhibitor appeared to benefit patients of the MHG
subtype. This study also proposes DDX3X as an attractive therapeutic target. DDX3Y
is essential for maintaining the translational capacity of transformed B-cells, but is not
131
expressed at the protein level in normal cells (Ditton et al., 2004; Foresta et al., 2000a;
Rauschendorf et al., 2011), which allows us to speculate that the toxicity of potential
DDX3Y inhibitor will be low. Indeed, males with germline DDX3Y mutation present with
infertility, but no other phenotype, except azoospermia, has been observed (Foresta et al.,
2000a; Rauschendorf et al., 2011). Up to now, there are two small molecule inhibitors of
DDX3 available for pre-clinical studies (Bol et al., 2015; Brai et al., 2016). Both target the
core helicase domain of DDX3, which is almost identical between DDX3X and DDX3Y.
Undoubtedly, there is an urgent need for the development of more DDX3Y-specific agents.
An interesting solution could be the use of Proteolysis Targeting Chimeras (PROTAC)
technology to target unique parts of DDX3Y to direct the protein to ubiquitin-proteasome
system for degradation (Gao et al., 2020).
132
CHAPTER 5
Elucidating the role of
translated micropeptides in
Diffuse Large B-cell Lymphoma
5.1 Background
The human genome is thought to contain approximately 20,000 protein-coding sequences.
The annotation of these Open Reading Frames (ORFs) has been based on a set of rules
predicting translation of stable and functional proteins, for example requiring minimum
length of 100 amino-acids (aa), methionine encoded translation start site, biased codon
usage and high sequence conservation between species (Brent, 2005; Ingolia et al., 2011).
The hypothesis of one gene, one enzyme (or one polypeptide in latter variants), formed
initially in 1941 (Beadle and Tatum, 1941), also largely shaped our understanding of the
organisation of the protein coding regions. However, in the light of recent studies, there
are several exceptions from those rules suggesting that the human proteome might be
much more complex and dynamic than we have thought.
Firstly, the number of variants and isoforms, or as has been recently proposed proteo-
forms (Smith and Kelleher, 2013), of known proteins may exceeds a fixed, gene-centric,
set of 20,000 sequences. Some estimates put the figure of 70,000 or even few millions
proteoforms produced in the human genome, if all post-translational modification are
included (Aebersold et al., 2018a). Another layer of proteome complexity is provided
by the discovery of new coding regions that were missed by the consensus annotations
due to, for example, their small size or unexpected localisation. The existence of those
noncanonical proteins, also referred to as micropeptides or cryptic peptides, have already
been confirmed by several studies, which demonstrated wide range of biological activity
of these products. For instance, the chemokine family, which are secreted by cells to
133
coordinate the immune response, comprise more than 40 known proteins with length
between 67–127 amino-acids (Moser and Willimann, 2004). Other examples of biologically
relevant micropeptides include the the family of defensins (18-45 aa), ribosomal protein
L24 (RPL24, 25 aa) and three calcium transporter regulators: phospholamban (PLN, 52
aa), sarcolipin (SLN, 31 aa), and myoregulin (MLN, 46 aa). The ability of new sequencing
techniques, such as Ribo-Seq to reveal the position of every translating ribosome in the
cell opens new opportunities to re-examine the translation landscape. Ribo-Seq technique
proved to be highly e↵ective to identify new ORFs in bona fide non-coding regions of
the genome, such as long non-coding RNAs (lncRNAs) or 50 and 30 untranslated regions,
50UTR and 30UTR respectively, flanking known ORFs (Chen et al., 2020; Chong et al.,
2020; Jackson et al., 2018; Calviello and Ohler, 2017). However, only a handful of these
have been validated to encode functional micropeptides with potent biological functions.
The scope of noncanonical translation and its exact role are still unclear. In attempt to
bridge this gap, I took advantage of a large translatome dataset generated in the Hodson
lab involving 79 Ribo-Seq libraries with matching RNA-Seq samples obtained from primary
B-cells and lymphoma cell lines.
In this chapter:
1. I introduce a systematic approach to annotate de novo actively translated regions
directly from Ribo-Seq data,
2. I query a set of mass-spectromentry datasets downloaded from public repositories
for identified noncanonical ORFs,
3. I rank putative micropeptides according to their biological relevance,
4. I use the top scoring entries to the design a knockdown CRISPR screen to identify
ORFs essential for B-cell growth and survival.
134
5.2 Results
5.2.1 A systematic approach for de novo identification of non-
canonical translation products in lymphoid cells.
5.2.1.1 An integrated ORF identification workflow
To annotate regions of noncanonical translation, I developed a bioinformatic pipeline
that integrates translatome (Ribo-Seq) and transcriptome (RNA-Seq) data creating a
non-redundant database of putative peptides and proteins (Figure 5.1 A).
The core idea of this workflow is to analyse the phasing pattern of ribosomal footprints
that is expected to be enriched in actively translated regions. It is important to point
out that a simple alignment of Ribo-Seq reads to a genomic region is not an evidence
of active translation. Of RNAse protected mRNA fragments, sequenced in Ribo-Seq,
only in about 85 % correspond to the ribosomes position. The remaining portion of
footprints comes from other RNA-protein complexes (Ji et al., 2016). The two situations
can be easily distinguished by looking at the distribution of Ribo-Seq reads over a queried
region. Ribosome-derived reads (1) have length about 28-29 nucleotides, (2) span the
entire translated region (signal uniformity), and (3) show 3-nucleotide periodicity in the
signal shape (frame preference), which corresponds to the ribosome decoding only three
nucleotides at time. In contrast, nonribosomal footprints are usually highly localised and
produce reads of varying lengths (Ji et al., 2016).
To find genomic regions showing features of active translation I have combined four
independent computational algorithms: ORFLine (Hu et al., 2021), ORF-RATER (Fields
et al., 2015), RibORF (Ji, 2018) and RiboCode (Xiao et al., 2018) to (Table 5.1) that were
developed specifically for this task. Overall, all four tools benefit from a regular pattern of
alignment of ribosome-derived footprints, but adopt di↵erent statistical models to rank
and select regions, which have the highest probability of being translated. Some tools
incorporate also additional metrics that reflect signal uniformity, such as Inside/outside
ratio (I/O ratio), sequence coverage or Ribosome Release Score (RRS) quantifying the
abrupt drop in Ribo-Seq signal after the stop codon.
Because translation can sometimes begin from codons other than AUG (Kearse and
Wilusz, 2017), all possible reading frames constrained by any combination of start (AUG,
UUG, GUG or CUG) and stop codons (UAA, UGA, or UAG) were considered. All
parameters of ORF-finding algorithms were kept at default values, as recommended in the
original publication, see section 2.2.11 for details.
135
Figure 5.1: Systematic approach to identify translated ORF in lymphoid cells
A) Flow chart showing the workflow for identifying ORFs from Ribo-Seq datasets
B) Diagram explaining the strategy of hierarchical merging adapted from Ouspenskaia et al.
(2020). Samples are represented as leaves that are merged based on the biological similarity
to create clades and finally one large root file containing all sequenced reads. Accumulation
of Ribo-Seq reads over multiple samples should increase the strength of frame preference
(increase sensitivity), while maintaining tissue specificity, as individual samples are also
used for ORF identification.
136
Table 5.1: An Overview of ORF-finding algorithms used in this study
Tool Input
Start
codons
Translation metrics Statistical evaluation
ORFLine
Ribo-Seq
RNA-Seq
NUG
Frame preference,
Ribosome release score,
Inside/outside read ratio,
sequence coverage
Log-scaled chi-squared
goodness of fit test
for frame preference
ORF-RATER
Ribo-Seq
RNA-Seq
NUG
Frame preference,
aggregate profiles over
start/stop codons
Linear regression model
RibORF Ribo-Seq NUG
Frame preference,
signal uniformity
Logistic regression model
Support Vector Machine classifier
RiboCode Ribo-Seq NUG Frame preference
Modified Wilcoxon
signed-rank test
First, the RiboStream pipeline was applied to a large dataset of 79 paired Ribo-Seq
and RNA-Seq libraries covering a variety of lymphoid cell types including 12 samples from
primary germinal centre B-cells, 5 samples from DLBCL tumour biopsies and 60 samples
from established lymphoma cell lines.
To maximise sensitivity and specificity of detection of translated regions, I adopted
a strategy of hierarchical ORF detection that was initially introduced by Ouspenskaia
et al. (2020). If the number of aligned footprints fall below certain level, the 3-nucleotide
periodicity become indistinguishable from non-translational noise and may be missed by
ORF identification algorithm.
However, if we aggregate the Ribo-Seq signal across multiple samples and run the
tools on the merged set of footprints, truly translated regions should accumulate enough
mapped reads to show a clear phasing pattern, thus, reaching detection level of ORF
finding algorithms. Hierarchical merging of Ribo-Seq and RNA-Seq BAM files (aligned
reads) is guided by the biological similarity, which allows to increase sensitivity of detection
while maintaining the specificity of cell type expression (Figure 5.1 B). For example,
from the data generated in chapter 3 (3 experimental conditions, 4 replicates each) 4
merged files can be built and 16 ORF identification runs can be performed at three levels:
1. Leaves level: individual samples runs (12 samples)
2. Group (clades) level: files merged by experimental condition (3 files, one per condi-
tion))
3. Root level: all 12 individual samples merged into 1 file
The same logic has been applied to all 79 Ribo-Seq samples giving rise to 177 files to run
ORF identification tools.
137
5.2.1.2 Pervasive translation of crude non-coding regions in lymphoid cells
In total, I found 13 483 canonical ORFs and 43 537 noncanonical ORFs identified by
at least two algorithms (Figure 5.2 A). Interestingly, only 420 (0.7%, 420/57,020 )
noncanonical ORFs were picked by all four tools what suggests substantial di↵erences
between the ORF finding strategies and reinforces the need of using more than one tool for
comprehensive identification. As expected, the number of ORFs was positively correlated
with the number of ribosomal footprints used as an input. The merged files, generated
through hierarchical merging strategy, brought almost one order of magnitude more ORFs
than would be from using from just raw, unmerged, files (Figure 5.2 B).
Although all tools classify identified ORFs into distinct types routinely, the strategies
to do so and the exact definitions of those ORF types vary between the algorithms. To
make the results comparable, I reassigned the ORFs to a representative transcript isoform
and unified ORF classification specifying nine ORF categories based on their genomic
location and reading frame (Figure 5.2 C), see section 2.2.11 for detailed description of
the classification strategy.
Of the 43 537 noncanonical ORFs, the largest group (31232/43537, 71.8%) encompassed
ORFs situated in the noncoding parts of coding transcripts, or overlapping the known
CDS out-of-frame with the annotated product. The noncanonical ORFs were found in
8586 protein-coding transcripts and each of these contained more than one additional
ORF, 2.8 on average. 14.7% (6386) of noncanonical ORFs showed substantial, in-frame,
overlap with annotated ORFs and were considered new isoforms of known ORFs with
either truncated or extended amino-acid sequence. Among the noncanonical ORFs, 5919
(5919/43537, 13.6%) were localised on transcripts annotated as noncoding, predominantly
in long noncoding RNAs (Figure 5.2 C).
Consistent with previous studies (Chen et al., 2020; Cuevas et al., 2021), noncanonical
ORFs were much shorter than the canonical proteins: the length of almost 50 % was
lower than 100 amino-acids (Figure 5.2 D). Noncanonical ORFs showed strong frame
preference, accumulation of footprints at the start codon and sudden drop in signal after
estimated termination site (Figure 5.2 E), which all characterise active translation and
resembles the pattern observed in known protein coding regions.
138
Figure 5.2: Ribo-Seq reveals pervasive translation of noncanonical ORFs in B-cells.
A) Venn Diagram showing the total number of ORFs identified and the overlap between the
ORF-finding algorithms. Only ORFs detected by more than 1 tool (numbers in black)
were selected for further analysis.
B) Scatter plot showing the relationship between the mean number of ORFs detected per
sample and the total number of ribosomal footprints used as an input. The colour indicates
which input files were obtained from merging the original FASTQ files together.
C) Barplot showing the percentage contribution of each ORF type to the total of 57 020 ORFs
identified. ORF types were divided into three classes: known ORFs and their variants
(blue), noncanonical ORFs localised on coding transcripts (violet) and novel ORFs in
noncoding transcripts (pink).
D) Histogram showing ORF length distribution for canonical and noncanonical ORFs.
E) Barplot showing average three nucleotide periodicity and single frame preference in canoni-
cal and noncanonical ORFs.
139
5.2.2 Noncanonical ORFs account for about 10% of proteins
detected in proteomics experiments
The analysis of the Ribo-Seq dataset revealed pervasive translation of thousands of
noncanonical regions in lymphoid cells. The most burning questions arising in this context
are: 1) are those products of noncanonical translation detectable at the protein level, 2) is
this a reproducible finding, and 3) are there any additional evidence, other than Ribo-Seq,
that those regions are biologically relevant?
I addressed the first question by reanalysing 13 publicly available mass spectometry (MS)
datasets, covering various cell types and conditions, and querying them for the presence of
peptides originating from predicted ORFs. Searching for the products of noncanonical
translation in proteomic data is a nontrivial challenge. Firstly, their small molecular
weight can impede detection by standard mass spectrometry techniques, thus, even less
popular methods should be explored. Recent studies demonstrated that noncanonical
proteins are enriched in Major Histocompatibility Complex (MHC) bound peptides (MAPs)
(Chong et al., 2020; Chen et al., 2020; Ouspenskaia et al., 2020). Mass-spectrometry based
identification of MAPs (immunopeptidomics) has attracted much attention in the context
of tumour neoantigens and numerous datasets are deposited in public repositories. Another
interesting technique is, so called, deep proteome or high-resolution MS, which allows for
higher protein coverage and broader dynamic range of detection (Bekker-Jensen et al.,
2017). Therefore the collected datasets included SILAC or tandem mass tag (TMT)
techniques, as well as immunopeptidomics and deep-proteome experiments, see section
2.2.12 for details regarding the bioinformatic workflow.
A typical analysis of a proteomic dataset involves the assignment of collected spectra to
the reference sequence. This poses another diculty, because an inflated search database
used for mass spectra survey can lead to spurious matches increasing the number of false
positive findings (Blakeley et al., 2012; Nesvizhskii, 2014). To decrease the search space
for MS analysis 1) I narrowed down a list of putative ORFs by removing all ORFs shorter
than 18 nucleotides (6 amino-acids), and 2) filtered out all in-frame ORFs with more than
20% overlap with known ORFs focusing my attention on unique ORFs that are unlikely to
be variants of known peptides. After filtering my list of putative micropeptides had 30,188
unique sequences that were used to build a customised search database for MS analysis.
To avoid forced assignment of the mass spectra to noncanonical ORFs, the final search
database contained the noncanonical peptides sequences and a set of reference proteins
downloaded from UniProtKB. I queried the MS datasets using MaxQuant software or
NewAnce workflow, designed specifically to process the immunopeptidomics data, see
Chong et al. (2020).
140
From all MS datasets surveyed, I identified 564,764 peptides assigned to 12,518 unique
proteins (PSM FDR 1%, Protein FDR 1%) (Figure 5.3 A). Of these, between 1 and 5.5
% were assigned to noncanonical ORFs (Figure 5.3 B) accounting for 1,311 unique ORFs
in total. The total number of peptides identified in MS data and the percentage of those
derived from noncanonical ORFs varied slightly between datasets and MS techniques, the
di↵erence was on the border of statistical significance (Kruskall-Wallis p = 0.1). The highest
proportion of peptides derived from noncanonical ORFs was in MHC-I immunopeptidomics
experiments and in a dataset utilising a deep proteome technique (Bekker-Jensen et al.,
2017) (Figure 5.3 B). This may suggest that certain properties of immunopeptidomics
and deep proteome experiments may facilitate the detection of noncanonical proteins.
It can be attributed to technical as well as biological properties, such as preferential
access to the antigen presentation pathway, diculties with detecting small or short living
proteins by standard techniques. MHC complex pull-down, multi-fractionation or the
usage of multiple proteases in deep-proteome studies may facilitate detection of otherwise
undetectable proteins (Bekker-Jensen et al., 2017).
Figure 5.3: Analysis of mass spectrometry datasets in search of peptides matching
predicted noncanonical ORFs
A) Barplot showing the total number of unique peptides identified in each mass spectrometry
dataset. Colour of the bar indicates whether a peptide has been assigned to noncanonical
(red) or canonical (grey) protein.
B) Boxplot showing the percentage of unique peptides matching predicted noncanonical
proteins. Statistical significance of the di↵erence in the proportion of peptides derived from
noncanonical ORFs in each MS group was determined with Kruskal-Wallis test, p-value =
0.1.
141
Next, I evaluated the accuracy of noncanonical proteins identification using four
metrics. In all datasets the distribution of mass measurement error (the di↵erence between
an individual measurement and an expected value for a peptide) and the Andromeda
score (probabilistic score reflecting the accuracy of peptide to MS spectrum match) were
indistinguishable for noncanonical and canonical proteins (Figure 5.4 A-C). The observed
Retention Time (RT) of eluted peptides was highly correlated with the chromatographic
Hydrophobicity Index (HI) predicted from the amino-acid composition of a peptide (Figure
5.4 C) (Krokhin and Spicer, 2010). The correlation for peptides derived from canonical
and noncanonical ORFs was similar. Lastly, the distribution of MHC-bound peptides
was similar in both groups with the majority of peptides having length between 9 and 10
nucleotides for MHC-I and 11-17 nucleotides for MHC-II (Figure 5.4 E).
Finally, I addressed the question of reproducibility of ORF identification. I aligned
the noncanonical ORFs against the extended database of human proteins downloaded
from UniProt (SwissProt and TrEMBL sets) and from two public repositories, OpenProt
and sORFdb, containing the sequences of new proteins predicted by other proteogenomic
studies. 5774 (19.12%, 5774/30188) ORFs matched UniProt, OpenProt or sORFdb with
at least 95% sequence similarity (Figure 5.4 F). Almost half of the ORFs (47.52%,
623/1311) with matching peptides from the reanalysed MS experiments were confirmed in
at least one of the databases (Figure 5.4 F).
In summary, although relatively small proportion of noncanonical ORFs was identified
in proteomic experiments, they accounted for about 10% of all unique proteins quantified
and showed the accuracy of detection no di↵erent to known proteins. In total, 21.40% of
noncanonical ORFs (6462/30188) had evidence of protein levels expression (UniProt or
proteomics) or have been identified as actively translated by other groups, which reinforces
the authenticity of our Ribo-Seq workflow. The remaining 78.6% were unique to Ribo-Seq
data from this study (no protein level evidence and not identified in other translatome
studies).
142
Figure 5.4: Features of peptides from canonical and noncanonical proteins identified
in MS experiments
A) Histogram of mass error for identified canonical and noncanonical peptide-spectrum
matches (PSM). The p-value < 1016, calculated with Kolmogorov–Smirnov test (Number
of canonical PSM = 7378711, number of noncanonical PSM = 80829).
B) Histogram of Andromeda Score, which mirrors the accuracy of peptide-spectrum match
of identified canonical and noncanonical peptides. The p-value < 1016, calculated with
Kolmogorov–Smirnov test (Number of canonical PSM = 7378711, number of noncanonical
PSM = 80829).
C) Pearson correlations between observed and SSRCalc predicted retention times of peptides
derived from canonical and noncanonical proteins.
D) Total number of unique canonical and noncanonical proteins and mean sequence coverage
of matched peptides.
E) Histogram of length of MAPs derived from canonical and noncanonical proteins.
F) Venn Diagram showing the overlap between ORFs identified in MS datasets, found in
UniProt and proteogenomic databases.
143
5.2.3 Characteristics of noncanonical ORFs producing MHC-
bound peptides
Next, I wished to find the features of noncanonical proteins that increase the probability
of protein level expression. In contrast to known proteins, there was only a little overlap
in noncanonical ORFs detected in immunopeptidomics and full proteome MS data: only
4.3% were identified by both techniques (Figure 5.5 A-B). The distribution of the ORF
types between various proteomic techniques was also remarkably di↵erent (Figure 5.5
C). MHC-bound peptides were more likely to come from uORF, while peptides derived
from noncoding transcripts (pseudogenes or lncRNAs) were more common in full proteome
data (Chi-square test for independence, p-value < 2.2 · 106). Interestingly, there was
no di↵erence in the distribution of ORF types between standard MS experiments and
deep-proteome studies (Chi-square test for independence, p-value = 0.59). This suggests
that there might be a relationship between the ability to detect ORF-derived peptides
and their biological properties.
I integrated di↵erent ORFs characteristics into a machine-learning algorithm based on
Random Forests to predict MS detection. Because of little overlap between immunopep-
tidomics and full proteome studies in detecting peptides originating from noncanonical
ORFs, I built separate single-class classification models to distinguish MHC-detected vs.
not detected and full proteome (deep proteome or standard MS) detected vs. not detected.
Sequence characteristics incorporated into the models included: the percentage of the
overlap with known CDS, ribosomal footprints density, translation eciency, gene type
(protein-coding, lncRNA or pseudogene), GC content, length, amino-acid composition and
isoelectric point (pI). Variants or isoforms of known genes (ORFs classified as canonical,
truncated, readthrough or extended) were filtered out and all noncanonical ORFs with
mean mRNA expression and ribosome footprints density > 0 were included. Of these
randomly selected 75% were used as a training set, the remaining 25% was used for testing.
The performance varied substantially between the two suggesting that ORF features are
not always sucient to predict detection in proteomics. ORF characteristics predicted
well the identification of MHC-bound peptides (80% recall at 40% precision, AUPRC
= 0.4919). Chromosome location, ORF score and the number of ORF identifying tools
recognising an ORF as translated were the most important factors for accurate prediction.
ORFs with detected MHC-bound peptides were significantly enriched in chromosome 4,
20 and 6 (z-test of two proportions, adjusted p-value < 0.01), had higher ORF scores
(Kolmogorov-Smirnov test, p-value < 1016) and were more likely to be detected by more
than one tool (Chi-squared test < 1016). Surprisingly, despite similar number of ORFs
detected in full proteome studies, the prediction did not work eciently for this group
(AUPRC = 0.0264).
144
Figure 5.5: Ribo-Seq reveals pervasive translation of noncanonical ORFs in B-cells.
A) Venn Diagram showing the overlap between MHC and full proteome derived peptides
derived from noncanonical proteins.
B) Venn Diagram showing the overlap between MHC and full proteome derived peptides
derived from canonical proteins.
C) Barplot showing the proportion of ORF types in peptides identified by di↵erent mass
spectrometry techniques. Chi-square test for independence, p-value < 2.2 · 106
D) Performance of machine-learning-based classifiers in predicting detection of noncanonical
ORFs derived peptides in mass spectrometry. Random forest classifiers were trained on
the set of noncanonical ORFs characteristics and performance was assessed in a tenfold
cross-validation (CV) mode.
E) Feature importance of random forest classifier predicting MAPs detection represented as
mean decrease in accuracy.
F) Percentage of noncanonical ORFs with MAPs per chromosome. Highlighted points indicate
chromosomes with statistically significant enrichment of MAPs. Z-test of two proportions,
adjusted p-value < 0.01.
G) Violin plot showing the di↵erence in ORF score between noncanonical ORFs with and
without detected MAPs. Kolmogorov-Smirnov test, p-value < 1016
H) Percentage of noncanonical ORFs detected with more than one tool compared for ORFs
with without detected MAPs, Chi-squared test for independence < 1016
I) Scatter plot showing feature importance of random forest regression model predicting the
value of the ORF score; measured with mean decrease in accuracy and mean increase in
mean squared error (MSE).
145
I also investigated the factors determining ORF score value, which directly translates
into the strength of three-nucleotide periodicity, and so is the key determinant of active
translation. Overall ORF features explained 61.29 % of ORF score variance with overlap
with known CDS, ribosome footprint density, mRNA expression, conservation score and
AUG as a start codon being the most influential predictive features. High importance of
expression related measures (mRNA expression or ribosomal footprints density) resonates
well with known relationship between the number of ORFs annotated and the number of
mapped ribosomal footprints.
This analysis suggests that, firstly, certain biological factors, sequence characteristics
may determine MHC presentation of peptides derived from noncanonical ORFs and,
secondly, low sequence coverage or low expression level may be the primary limiting factor
for ecient detection and validation of noncanonical ORFs.
5.2.4 Design of customised knockout CRISPR screen to identify
noncanonical ORFs important for B-cells survival
To identify the noncanonical ORFs that are essential for B-cell survival in a systematised
manner, I wished to screen the most promising candidates using a customised CRISPR
library which contains 6000 gRNAs targeting 1,625 not overlapping ORFs.
I designed a knockout CRISPR screen to target selected ORFs with features indicating
an important biological function in lymphoid cells. I divided the process of selecting ORFs
for screening into two stages: negative selection aiming to discard all ORFs that are not
fit for CRISPR-Cas9 targeting and positive selection enriching for regions with interesting
biological features.
The first stage discards all ORFs which are either too short to be targeted with sucient
number of gRNAs, have low expression in the primary cells or cell lines chosen for the
screen, or share too large overlap with known CDS region, so that the observed phenotype
could be easily explained by disruption in the canonical coding region (Figure 5.6 A-B).
Next I screened the remaining ORFs for compatible gRNAs cutting within a predicted
ORF. I filtered out all untargetable ORFs with the number of good quality gRNAs below
5. By good quality I understand gRNAs with GC content between 40 and 70 %, with no
direct o↵-target locations and without homopolymers of 4 or more consecutive Ts. The
presence of homopolymers in gRNA sequence can decrease cutting activity and TTTT is
known to act as minimal T-stretch termination signal for RNA polymerase III (Gao et al.,
2018). Indeed, when I compared the eciency of dropout for genes identified as essential
in knockout CRISPR-Cas9 screen in lymphoid cell lines (Phelan et al., 2018), I observed
significant decrease for gRNAs containing homopolymers, especially series of Ts (Figure
5.6 C).
146
Figure 5.6: Design of a CRISPR-Cas9 CRISPR screen library to study noncanonical
ORFs in lymphoid cells
A) Barplot showing the proportion of noncanonical ORF excluded from the screen. In total
about 68.49 % ORFs were excluded because of low expression level, too large overlap with
known CDS or too being too short.
B) Violin plots showing mean expression level of noncanonical ORFs in cells selected for
the screen. Red dashed line corresponds to the 20th percentile, which was considered a
threshold for low expression.
C) Barplots showing dropout eciency of gRNAs targeting essential genes from Phelan et al.
(2018) knockout CRISPR-Cas9 screen stratified by the presence of homopolymers in gRNAs
sequence. Statistical significance (compared to ‘None’ group) determined with two sample
Wilcoxon test, adjusted p-values computed with Benjamini and Hochberg method. **** <
0.0001, *** < 0.001, ** < 0.01
D) Violin plot showing RNA folding energy of gRNA included and not included in the final
CRISPR-Cas9 screen library.
E) Violin plot showing one of the gRNA eciency scores for gRNAs included and not included
in the final CRISPR-Cas9 screen library.
F) Barplot showing the percentage of di↵erent ORF types targeted with the library of gRNAs.
G) Histograms showing the ORFs targeted with the gRNA library have overall higher evolu-
tionary conservation, higher ORF score and higher expression level.
The p-values in all groups (folding energy, DOENCH 2016 score, evolutionary conservation,
ORF score and expression level) < 1010, calculated with Kolmogorov–Smirnov test
147
The eciency of all gRNAs was evaluated using a number of eciency scores, including
prioritising a G at position 20 (upstream of PAM) and few multi-factor scores predicting
gRNA stability and activity (Moreno-Mateos et al., 2015; Xu et al., 2015; Doench et al.,
2014, 2016) (Figure 5.6 D). Given reports about inferior on-target activity of gRNAs
with internal harpins and regions of self-complementarity, I also prioritised gRNAs with
lower folding energy (Thyme et al., 2016) (Figure 5.6 E).
9514 ORFs that left after the stage of negative selection were then ranked based
on the features that make them interesting from a biological perspective. I took intro
account the following characteristics: mean ORF score (reflecting the strength of frame
preference), mRNA expression level, the fact of being detected at the protein level (in
immunopeptidomics or full proteome data) or matching predicted proteins from external
databases, evolutionary conservation, the number of tools that identified an ORF as actively
translated, the number of samples where this happened and the di↵erential expression
pattern shown in our large Ribo-Seq dataset. The final library targets 1,625 top scoring
ORFs with gRNAs showing the highest predicted on-target activity, on average 4-5 gRNAs
per ORF (Figure 5.6 F-G).
As negative controls I used 250 non-targeting gRNAs from an established Brunello
CRISPR-Cas9 library (Sanson et al., 2018), which do not recognise any sequence in
the human genome. This will help me to distinguish the e↵ect of neutral drift from
selective disadvantage caused by disruption of biologically important ORFs. For positive
control I selected two known oncogenes, MYC and POU2AF1, which knock-out should
be disadvantageous for established lymphoma cell lines and primary GC B-cells (Phelan
et al., 2018; Caeser et al., 2019) allowing me to estimate the expected dropout capacity of
the screen. This library was ordered as an oligo pool from Twist Bioscience and cloned
into a lentiviral backbone, maintaining representation > 100 colonies per guide. It was
then introduced into six Cas9-expressing lymphoma cell lines and ex vivo human GC B
cells maintaining representation of >1,000 transduced cells per guide. Cells were harvested
at day 21 and sequencing libraries prepared. At the time of writing these libraries are
waiting to be sequencing. These CRISPR screening experiments were conducted by Dr.
Stamatia Vori, a Masters student in the Hodson lab.
148
5.3 Discussion
The last four years have witnessed an increase in studies investigating the topic of
noncanonical translation. Our extensive collection of 79 B-cell translatomes generated
in primary GC and malignant B-cell cells provides a powerful resource to investigate
this question in the context of physiology and pathology. To our knowledge, this is the
largest study exploring this topic in B-cells so far. The proteogenomic method described
here integrates Ribo-Seq, RNA-Seq, mass spectrometry (MS) and external databases to
comprehensively evaluate the scope of noncanonical translation in lymphoid cells and
identify the most promising targets for high-throughput knockout CRISPR screen. The
initial results of the systematic analysis of B-cells translatomes revealed 43 537 noncanonical
Open Reading Frames (ORFs) and 13 483 canonical (known) ORFs. Noncanonical ORFs
were typically situated in ostensibly noncoding regions of the genome, accounted for about
10% of all proteins detected in the analysis of external mass spectrometry experiments, and
almost 20% was also found in other proteogenomic databases. To identify noncanonical
ORFs essential to B-cell survival, I designed a customised knockdown CRISPR screen
library targeting top ORFs with the highest likelihood of protein level expression and
biological relevance.
Evidence behind pervasive translation of noncanonical ORFs
This and previous studies (Chen et al., 2020; Chong et al., 2020; Cuevas et al., 2021;
van Heesch et al., 2019) employed Ribo-Seq to provide evidence for widespread ribosome
occupancy in noncoding regions of the transcriptome. This has been observed in a range of
model organisms (Zhang et al., 2019; Mackowiak et al., 2015), in studies utilising di↵erent
translation inhibitors for Ribo-Seq library preparation (Ingolia et al., 2011; Zhang et al.,
2018a; Lee et al., 2012), in polysome-centric studies (Poly-Ribo-Seq) (Aspden et al., 2014),
and in vivo with RiboTag RNA sequencing (Jackson et al., 2018; Sanz et al., 2009). This
is a striking finding that puts some rules of eukaryotic translation into question, including
monocistronic organisation of eukaryotic transcripts or the purely noncoding nature of
several long-noncoding RNAs.
Limitations of proteomic-based approaches for novel proteins discovery
Although noncanonical translation seems to be a recurrent observation in translatome
studies, validation of their products at the protein level is lagging behind. In a most
popular approach - shotgun proteomics, proteins extracted from the cell are fragmented
with a protease (typically trypsin), and the mixture of digested peptides is analysed with
tandem mass spectrometry (MS/MS). Identification of the mass spectra occurs through
comparison with the reference database of in silico generated spectra from a provided
149
sequence of reference proteins (Perez-Riverol et al., 2018). This poses diculties for
pinpointing peptides originating from proteins not included in the reference sequence
database, with unexpected post-translational modifications (PTMs) or with low signal-to-
noise ratio (Griss et al., 2016). On the other hand, an inflated reference database leads to
a high number of false positive findings (Blakeley et al., 2012; Nesvizhskii, 2014), so the
reference must be prepared with care. Here, I adopted a parsimonious strategy combining
the sequence of known proteins with the sequence of ORFs predicted with Ribo-Seq, so
that using the smallest possible reference database to search for novel proteins.
Only about 4% of noncanonical ORFs predicted in our study had evidence of protein
level expression. Are those remaining 96% of noncanonical ORFs a technical noise, or we
just failed to detect it with proteomics? Both scenarios are possible. The arguments in
favour of the first are: relatively low reproducibility of the assay between di↵erent ORF
identification algorithms and the small overlap with external proteogenomic databases
as only about 20% of noncanonical ORFs were observed in other studies. Another issue
is that we cannot exclude the possibility that the Ribo-Seq signal, even when showing
the periodical pattern of alignment, does not always correspond to actively translating
ribosomes. For example, could the non-coding regions of the genome work as a ribosome
sponge by sequestrating them or directing their transport? Noncanonical translation could
also correspond to mRNA quality check or maturation process, e.g. during a pioneering
round of translation (Maquat et al., 2010).
On the other hand, although orthogonal evidence for about 20% of noncanonical
ORFs seems like a low recall, it may be a lot given that we know very little about
noncanonical translation biology. Even in the group of 20,386 manually curated human
proteins (Swiss-Prot), almost 20% (3,989/20,386) lacks strong protein level evidence
(Perdiga˜o et al., 2015). Indeed, out of about 20,000 canonical proteins, only half was
identified in our MS data analysis. On average, 75% of mass spectra reported in a typical
MS experiment remains unidentified (Griss et al., 2016). Many of these have high quality
and are likely to emerge from real proteins (Chick et al., 2015). The life cycle (synthesis
and degradation rate), a pattern of PTM or subcellular localisation of noncanonical ORFs
may be di↵erent to canonical proteins, which may poses diculties with detection using
standard proteomic techniques. For example, not all noncanonical micropeptides contain
a tryptic cleavage site, which may complicate their identification in assays using trypsin
for protein fragmentation. Enrichment of noncanonical translation products in the MHC-
bound peptidome (immunopeptidome) is an unprecedented finding (Chong et al., 2020;
Ouspenskaia et al., 2020; Chen et al., 2020; Cuevas et al., 2021). Firstly, it reinforces
the concept that the search for micropeptides, ‘hidden’ in the cellular proteome, should
include a broad range of proteomic techniques. And secondly, it suggests that the role
of micropeptides may be linked to immunity and the generation of peptides directed for
150
MHC presentation. Finally, the MS data analysed here were downloaded from external
repositories, so there is a possibility that our ability to detect mass spectra matching
predicted micropeptides might be lower than in studies, where MS and Ribo-Seq data
were generated from the same biological model. For example, matching mass spectra were
found for about 7% of noncanonical ORF identified by Chen et al. (2020).
Potential role of micropeptides in immunity and immune surveillance
Noncanonical translation has drawn recently much attention in the context of antigen
presentation, and prospective targets for cancer immunotherapy as a considerable fraction
of MHC-derived peptides (MAPs) has been attributed to noncanonical proteins (Chong
et al., 2020; Ouspenskaia et al., 2020; Cuevas et al., 2021). In line with those studies, I also
observed a sizeable proportion of MAPs (between 2 and 5%) originating from noncanonical
ORFs. Of 9,242 proteins identified in all reanalysed immunopeptidomics datasets, 652
(7.05%) were noncanonical proteins encoded by ORFs from noncoding regions of the
genome. MHC genes are one of the most polymorphic genes, and each allele (allomorphs)
binds a distinct set of peptides. This may limit our ability to identify MAPs accurately
if the immunopeptidomics data were generated from di↵erent cells than the Ribo-Seq
samples used for ORF prediction.
A fascinating concept linking noncanonical translation to immunopeptidome and
defective ribosomal products (DRiPs) was formulated by Jonathan Yewdell. DRiPs
are short-lived peptides (half-lives of minutes), which originate from invalid rounds of
translation, e.g. due to mutations, synthesis errors, misfolding, truncation etc. (Dersh
et al., 2021). Initially, the DRiPs hypothesis was put forward to explain the rapid
presentation of antigens derived from stable viral proteins (Yewdell et al., 1996). DRiPs
can also arise in abundance from noncanonical translation (e.g. ORFs with near-cognate
start codons), especially when 50-cap-dependent translation is shut down during stress
or viral infection. Interestingly, the composition of MAPs is poorly reflected by the
transcriptome and proteome, i.e. most abundant mRNAs or proteins do not necessarily
produce the largest number of MAPs (Pearson et al., 2016). Certain genomic regions
are ‘hot spots’ of MAPs despite relatively small contribution to the cellular proteome or
transcriptome (Pearson et al., 2016). Preferential access of certain peptide groups to the
MHC presentation pathway could be a part of the immunosurveillance process, which is
especially important in the context of tumour formation. Immunosurveillance involves a
complex interplay between the tumour immunogenicity, immune cells infiltration, cytotoxic
T-cells activation, immune checkpoints and the microenvironment (Dersh et al., 2021). A
better understanding of cancer-specific antigens could aid in designing cancer vaccines,
personalised CAR T cell therapy or drugs that increase peptide generation in cancer cells,
thus immune visibility of the tumour.
151
Possible roles of upstream Open Reading Frame (uORFs)
Translation of thousands of upstream Open Reading Frames (uORFs) is a recurrent finding
in global translatome studies, including this one. The regulatory properties of uORFs
have been known for a long time, but the evidence was mainly anecdotal, limited to a
handful of transcripts, such as ATF4, MDM2, CEBPA and CEBPB (Wethmar et al., 2014).
However, current estimates place the percentage of mammalian protein-coding genes with
potentially functional uORFs at around 40-50% (Johnstone et al., 2016; Lee et al., 2012).
In our study, I identified uORFs in 37% (5034/13483) of protein-coding genes, between 2 -
3 uORFs per gene.
uORFs are usually permissive for translation of the main coding sequence, but the
eciency of re-initiation of the downstream translation may be reduced Smith et al.
(2021); Zhang et al. (2019); Hinnebusch et al. (2016). Despite a possible deleterious
e↵ect of uORFs on canonical ORF translation and observed depletion of population
variants creating polymorphic uORFs, uORFs present in the genome show a higher level
of conservation than expected from neutral evolution (Zhang et al., 2019; Churbanov
et al., 2005; Zhang et al., 2021). A recent systematic analysis of 16,907,129 upstream
AUGs (uAUGs) in 478 eukaryotic species showed strong purifying selection for the vast
majority of uORFs suggesting the biological importance of these regions (Zhang et al.,
2021). In line with these studies, I observed many uORFs localised in the regions of high
evolutionary conservation.
Up to now, regulation of translation of the downstream (canonical) ORF is the main
role of uORF. The amino-acid sequence of putative, uORF-encoded micropeptides shows
a smaller degree of conservation, suggesting that the primary function of the majority of
uORFs might be a fine tuning of downstream ORF translation rather than the encoding
of stable proteins (Zhang et al., 2021). The balance between uORF and main ORF
translation is influenced by a number of factors. Proposed determinants include: the total
number of uORFs in 50UTR (Zhang et al., 2018a; Chew et al., 2016), uORF position with
respect to canonical start codon (distance, out-of-frame/in-frame) (Johnstone et al., 2016;
Chew et al., 2016; Calvo et al., 2009), the activity of certain translation factors, Kozak
sequence context (Rogozin et al., 2001) or the adjacent mRNA secondary structure (Chew
et al., 2016; Zhang et al., 2019).
A well studied mechanism of preferential uORFs translation is stress and immune
response. The frequency of uORF leaky scanning (start codon skipping) negatively
correlates with the availability of the ternary complex (Orr et al., 2020). Stress response-
associated programme of uORF translation has been associated with the preferential
translation of the immune regulator programmed cell death ligand-1 (PD-L1), and a set of
genes related to the development of squamous cell carcinoma (Sendoel et al., 2017).
152
Active translation of long non-coding RNAs
The concept of actively translated long noncoding RNAs (lncRNAs) is still a controversial
and widely debated topic. The main critics of this hypothesis argue that the evidence of
lncRNAs translation comes almost exclusively from Ribo-Seq studies. Given that RNAse
protected fragments can originate from nonribosomal footprints (Ji et al., 2016), the
abundance of footprints observed in certain lncRNAs might be just an artefact. Aebersold
et al. (2018b) argue that among all possible amino-acid sequences that could be produced
from lncRNAs, only 69 shown proteomic evidence, which, for the majority, was limited to
a single peptide match or could be explained by pseudogene missanotation or overlapping
exons from adjacent protein coding transcripts. Moreover, most candidate lncRNA-encoded
peptides: lack detectable functional protein domain, show lower expression level and lower
evolutionary conservation than known protein coding genes (Ji et al., 2015). This, however,
provides only a glimpse of a complex system. LncRNAs belongs to a heterogeneous group
of 9,640 transcripts (according to ENCODE) with a broad range of biological activity
including regulation of transcription, mRNA splicing, sequestration of certain mRNAs
or chromatin remodelling (Aebersold et al., 2018b; Statello et al., 2021; Derrien et al.,
2012). Annotation of lncRNA has been based predominantly on cDNA alignment to
the genome, chromatin signature indicating active transcription and lack of an ORF
meeting strict protein-coding criteria (Guttman et al., 2009). Although they do not
encode, by convention, canonical proteins, lncRNAs biogenesis is almost indistinguishable
from coding mRNAs: they are capped, spliced and polyadenylated (Statello et al., 2021).
Pervasive mapping of Ribo-Seq reads to lncRNAs has been reported in several works
(Aspden et al., 2014; Bazzini et al., 2014; Ruiz-Orera et al., 2014; Chong et al., 2020;
Ouspenskaia et al., 2020; Fields et al., 2015). Indeed, the ribosome occupancy or translation
eciency value is not sucient to distinguish between true coding and noncoding regions
(Guttman et al., 2013; Xiao et al., 2016), but other metrics, such as ORFscore or Ribosome
Release Score (RRS), have been developed to assist with this task. Both ORFscore (the
strength of frame preference) and RRS (ribosome clearance after stop codon) are highly
dependent on the ORF coverage and may underperform for transcripts with low expression
values. I expect that, by combining multiple ORF identification algorithms with the
strategy of hierarchical merging, it is possible to, firstly, decrease dependency on one
metric to score putative ORFs, and secondly, benefit from a large Ribo-Seq dataset to
increase sensitivity and specificity of ORF prediction for transcripts with a broad range
of expression values. Many lncRNAs-encoded proteins turned out to be functional and
have been validated experimentally in humans, fly and mouse (Jackson et al., 2018; Chen
et al., 2020). Translated lncRNAs were more hydrophobic with predicted alpha-secondary
structure and Kozak sequence around the translation initiation site, but these features
varied between studies suggesting strong context or methodology dependency (Ji et al.,
153
2015; Li and Liu, 2019). Ji et al. (2015) showed that lncRNAs with evident features of
active translation were almost exclusively cytoplasmic, had higher conservation scores and
evidence of purifying selection of amino-acid sequence.
A consensus model explaining the observations of coding properties of lncRNAs can
be proposed. First of all, there is a possibility that in the group of current lncRNAs,
there are true protein coding transcripts that have been missed in annotations because
of their unusual features. Some coding lncRNAs could also form a group of bifunctional
RNAs: encoding a protein sequence and biologically important noncoding transcript. A
handful of such transcripts has already been observed in human, bacteria, and few model
organisms, including Xenopus, Drosophila, Zebrafish (Aebersold et al., 2018b; Hube et al.,
2011; Chooniedass-Kothari et al., 2004; Kondo et al., 2010; Ingolia et al., 2011; Kumari
and Sampath, 2015).
Dark side of human proteome may shed light on noncanonical translation
The topic of noncanonical translation converges with the concept of dark proteome. The
dark proteome refers to the portion of the cellular proteome with unknown structure,
encoded by bona-fide noncoding transcripts or with atypical folding structure (Perdiga˜o
et al., 2015). Our preliminary analysis showed that the noncanonical ORFs are enriched
for intrinsically disordered regions (IDRs), which is an intriguing observation. IDRs are
regions with compositional bias in the amino-acid sequence containing more hydrophilic
amino acids and proline residues than structured regions (Dyson and Wright, 2005). IDRs
provide a large surface of interaction with frequent short linear motifs (SLIMs), including
peptide or nucleic acids binding motifs or sites for post-translational modifications (Tompa
et al., 2014; Dinkel et al., 2014). Given their ability to bind and interact with other
molecules, IDR-containing proteins, including identified noncanonical ORFs, may act as
chaperones or co-factors complementing the function of other proteins (Dyson and Wright,
2005).
154
CHAPTER 6
Perspectives
Ribo-Seq as a tool to study genome-wide translation in lymphoid cells
There is no doubt that the technology to study ribosome occupancy with single-nucleotide
precision has been a breakthrough for translation studies. Ribo-Seq has been applied to
analyse protein synthesis quantitatively, by estimating the translation intensity of a chosen
region, or qualitatively, by exploring which portions of the transcriptome undergo active
translation.
All Ribo-seq data presented in this thesis has achieved expected quality and repro-
ducibility, similar to in the original protocol (Ingolia et al., 2012). This would not be
possible without an ecient bioinformatic workflow. Therefore, RiboStream pipeline,
which I have developed for parallel and transparent data processing, was the core stage of
this project.
Despite the widespread application of Ribo-Seq in research, a computational workflow
has been poorly standardised and the bioinformatic tools developed, specifically for Ribo-
Seq, has rarely been used beyond the initial publication. In this thesis, I performed a basic
benchmarking of available tools for di↵erential translation analysis to make an informed
decision on the strategy for analysing our Ribo-Seq data. The most striking derivative of
this analysis was the strong dependency of the performance of the tools on the experimental
design. With the most common experimental design in the literature (2 experimental
conditions, 2 replicates each) on average only 20% of truly di↵erentially translated genes
could be recalled. The number of replicates as high as 8 was just enough to rise the true
positive rate to about 70% and stabilise the false positive rate around the desired level of
5%. This suggests that to utilise the full power of Ribo-Seq for di↵erential translation
analysis, at least 8 replicates is recommended. Our benchmarking would benefit from
an additional set of Ribo-Seq data, that could validate the aforementioned observations.
It would be also interesting to dissect characteristics of the transcripts arising as false
positives or false negatives, that could assist in the di↵erential translation interpretation.
155
Translational landscape of GC B-cells malignant transformation
In chapter 3 I have also studied the translational consequences of deregulated expression
of the two transcription factors, BCL6 and MYC, in primary GC B-cells. With similar ex-
perimental design and biological motivation to find di↵erentially translated genes following
the overexpression of an oncogenic transcription factor, I have not observed such profound
translational reprogramming as Sendoel et al. (2017). A possible limitation of this study
is that Ribo-Seq provides only relative quantification of the ribosome footprint abundance,
which may be challenging to interpret when massive changes in gene expression landscape
are expected. It would be interesting to perform the same experiment but using polysome
fractionation combined with RNA-Seq instead. Up to now, a systematic comparison of
Ribo-Seq with polysome fractionation based di↵erential translation analysis has not been
performed.
While mRNA levels have directed the majority of changes in gene expression, the
di↵erentially translated genes were mainly associated with cellular housekeeping functions,
such as ribosome biogenesis or oxidative phosphorylation. The adaptive role of translational
control in fine tuning of highly energetic metabolic processes and to what extent this may
a↵ect the process of malignant transformation is not well understood.
The concept of ribosome heterogeneity is another fascinating topic worth deeper
exploration. If BCL6 or MYC-induced changes in the translation intensity of individual
ribosomal proteins translates into the stoichiometry of ribosomal proteins incorporated
into the ribosome, this could be an essential mechanism of post-transcriptional regulation.
The role of DDX3X in facilitating MYC-driven lymphomagenesis
In chapter 4 I revealed that DDX3X loss-of-function promotes lymphomagenesis by
bu↵ering MYC-driven increase in global protein synthesis and proteotoxic stress. I show
that DDX3X controls the translation of ribosomal proteins, thus global translation load.
Although this involves direct binding of DDX3X to transcripts encoding ribosomal proteins,
the exact molecular mechanism is still unclear: whether this is related to RNA unwinding
activity of DDX3X, associated with mTOR/LARP1/5’TOP axis or facilitated by one of
the other versatile functions of DDX3X.
Another question that remained relating to the role of DDX3X in MYC-driven lym-
phomas is the mechanism of DDX3Y protein upregulation in transformed B-cells and
redundancy of biological activity of DDX3Y and DDX3X in lymphoid cells.
156
Exploring the role of noncanonical proteome in immunity and tumour im-
munosurveillance
Finally, in chapter 5 I revealed pervasive translation of noncanonical ORF in lymphoid
cells, which is a captivating finding. Although this is not the only study uncovering
widespread ribosome occupancy of bona-fide noncoding regions, this is one of the largest
studies and the first investigating this topic in primary GC B-cells.
An introductory analysis presented in this thesis undoubtedly has a hypothesis-
generating flavour. The most burning questions related to noncanonical translation
involve the extension of the protein level expression, the interplay between uORF and main
ORF translation, and the frequency of potential functional protein domains and short
linear motifs, which could shed light on the biological role of synthesised micropeptides.
The analysis of three external immunopeptidomics data revealed that between 2 and
5% of MHC-bound peptides (MAPs) could originate from predicted noncanonical ORFs.
To what extent this refers to the immunogenic function of noncanonical translation or
reflects the inferior performance of other proteomic methods, remains to be addressed.
157
158
Bibliography
Abate, F., Ambrosio, M. R., Mundo, L., Laginestra, M. A., Fuligni, F., Rossi, M., Zairis,
S., Gazaneo, S., De Falco, G., Lazzi, S., et al. (2015). Distinct viral and mutational
spectrum of endemic Burkitt lymphoma. PLoS pathogens, 11(10):e1005158.
Aebersold, R., Agar, J. N., Amster, I. J., Baker, M. S., Bertozzi, C. R., Boja, E. S.,
Costello, C. E., Cravatt, B. F., Fenselau, C., Garcia, B. A., et al. (2018a). How many
human proteoforms are there? Nature Chemical Biology, 14(3):206.
Aebersold, R., Agar, J. N., Amster, I. J., Baker, M. S., Bertozzi, C. R., Boja, E. S.,
Costello, C. E., Cravatt, B. F., Fenselau, C., Garcia, B. A., et al. (2018b). How many
human proteoforms are there? Nature Chemical Biology, 14(3):206–214.
Aitken, C. E. and Lorsch, J. R. (2012). A mechanistic overview of translation initiation in
eukaryotes. Nature Structural and Molecular Biology, 19(6):568–576.
Alizadeh, A. A., Eisen, M. B., Davis, R. E., Ma, C., Lossos, I. S., Rosenwald, A., Boldrick,
J. C., Sabet, H., Tran, T., Yu, X., et al. (2000). Distinct types of di↵use large b-cell
lymphoma identified by gene expression profiling. Nature, 403(6769):503–511.
Alkallas, R., Lajoie, M., Moldoveanu, D., Hoang, K. V., Lefranc¸ois, P., Lingrand, M.,
Ahanfeshar-Adams, M., Watters, K., Spatz, A., Zippin, J. H., et al. (2020). Multi-
omic analysis reveals significantly mutated genes and DDX3X as a sex-specific tumor
suppressor in cutaneous melanoma. Nature Cancer, 1(6):635–652.
Altschul, S. F., Gish, W., Miller, W., Myers, E. W., and Lipman, D. J. (1990). Basic local
alignment search tool. Journal of Molecular Biology, 215(3):403–410.
Andrews, S. et al. (2017). FastQC: a quality control tool for high throughput sequence
data. 2010.
Ansell, S. M., Lesokhin, A. M., Borrello, I., Halwani, A., Scott, E. C., Gutierrez, M.,
Schuster, S. J., Millenson, M. M., Cattry, D., Freeman, G. J., et al. (2015). PD-1
blockade with nivolumab in relapsed or refractory Hodgkin’s lymphoma. New England
Journal of Medicine, 372(4):311–319.
159
Aspden, J. L., Eyre-Walker, Y. C., Phillips, R. J., Amin, U., Mumtaz, M. A. S., Brocard,
M., and Couso, J.-P. (2014). Extensive translation of small open reading frames revealed
by Poly-Ribo-Seq. eLife, 3:e03528.
Aviner, R. (2020). The science of puromycin: From studies of ribosome function to
applications in biotechnology. Computational and Structural Biotechnology Journal,
18:1074–1083.
Aviner, R., Geiger, T., and Elroy-Stein, O. (2013). Novel proteomic approach (PUNCH-P)
reveals cell cycle-specific fluctuations in mRNA translation. Genes & development,
27(16):1834–1844.
Babaian, A., Rothe, K., Girodat, D., Minia, I., Djondovic, S., Milek, M., Miko, S. E. S.,
Wieden, H.-J., Landthaler, M., Morin, G. B., et al. (2020). Loss of m1acp3 ribosomal
RNA modification is a major feature of cancer. Cell Reports, 31(5):107611.
Barna, M., Pusic, A., Zollo, O., Costa, M., Kondrashov, N., Rego, E., Rao, P. H.,
and Ruggero, D. (2008). Suppression of Myc oncogenic activity by ribosomal protein
haploinsuciency. Nature, 456(7224):971–975.
Basso, K. and Dalla-Favera, R. (2015). Germinal centres and B cell lymphomagenesis.
Nature Reviews Immunology, 15(3):172–184.
Bastide, A. and David, A. (2018). The ribosome, (slow) beating heart of cancer (stem)
cell. Oncogenesis, 7(4):1–13.
Battle, A., Khan, Z., Wang, S. H., Mitrano, A., Ford, M. J., Pritchard, J. K., and Gilad, Y.
(2015). Impact of regulatory variation from RNA to protein. Science, 347(6222):664–667.
Bazzini, A. A., Johnstone, T. G., Christiano, R., Mackowiak, S. D., Obermayer, B.,
Fleming, E. S., Vejnar, C. E., Lee, M. T., Rajewsky, N., Walther, T. C., et al. (2014).
Identification of small ORF s in vertebrates using ribosome footprinting and evolutionary
conservation. The EMBO journal, 33(9):981–993.
Beadle, G. W. and Tatum, E. L. (1941). Genetic control of biochemical reactions in
Neurospora. Proceedings of the National Academy of Sciences of the United States of
America, 27(11):499.
Bekaert, M., Ivanov, I. P., Atkins, J. F., and Baranov, P. V. (2008). Ornithine decarboxylase
antizyme finder (OAF): fast and reliable detection of antizymes with frameshifts in
mRNAs. BMC Bioinformatics, 9(1):1–10.
160
Bekker-Jensen, D. B., Kelstrup, C. D., Batth, T. S., Larsen, S. C., Haldrup, C., Bramsen,
J. B., Sørensen, K. D., Høyer, S., Ørntoft, T. F., Andersen, C. L., et al. (2017). An
optimized shotgun strategy for the rapid generation of comprehensive human proteomes.
Cell Systems, 4(6):587–599.
Berletch, J. B., Yang, F., Xu, J., Carrel, L., and Disteche, C. M. (2011). Genes that escape
from X inactivation. Human Genetics, 130(2):237–245.
Bhat, M., Robichaud, N., Hulea, L., Sonenberg, N., Pelletier, J., and Topisirovic, I.
(2015). Targeting the translation machinery in cancer. Nature Reviews Drug discovery,
14(4):261–278.
Blakeley, P., Overton, I. M., and Hubbard, S. J. (2012). Addressing statistical biases in
nucleotide-derived protein databases for proteogenomic search strategies. Journal of
Proteome Research, 11(11):5221–5234.
Bol, G. M., Vesuna, F., Xie, M., Zeng, J., Aziz, K., Gandhi, N., Levine, A., Irving, A.,
Korz, D., Tantravedi, S., et al. (2015). Targeting DDX3 with a small molecule inhibitor
for lung cancer therapy. EMBO Molecular Medicine, 7(5):648–669.
Boon, K., Caron, H. N., Van Asperen, R., Valentijn, L., Hermus, M.-C., Van Sluis, P.,
Roobeek, I., Weis, I., Voute, P., Schwab, M., et al. (2001). N-myc enhances the expression
of a large set of genes functioning in ribosome biogenesis and protein synthesis. The
EMBO Journal, 20(6):1383–1393.
Bourgeois, C. F., Mortreux, F., and Auboeuf, D. (2016). The multiple functions of RNA
helicases as drivers and regulators of gene expression. Nature Reviews Molecular Cell
Biology, 17(7):426–438.
Bouska, A., Bi, C., Lone, W., Zhang, W., Kedwaii, A., Heavican, T., Lachel, C. M., Yu,
J., Ferro, R., Eldorghamy, N., et al. (2017). Adult high-grade b-cell lymphoma with
Burkitt lymphoma signature: genomic features and potential therapeutic targets. Blood,
130(16):1819–1831.
Brai, A., Fazi, R., Tintori, C., Zamperini, C., Bugli, F., Sanguinetti, M., Stigliano, E.,
Este´, J., Badia, R., Franco, S., et al. (2016). Human DDX3 protein is a valuable target
to develop broad spectrum antiviral agents. Proceedings of the National Academy of
Sciences, 113(19):5388–5393.
Brent, M. R. (2005). Genome annotation past, present, and future: how to define an ORF
at each locus. Genome research, 15(12):1777–1786.
161
Brunet, M. A., Lucier, J.-F., Levesque, M., Leblanc, S., Jacques, J.-F., Al-Saedi, H. R.,
Guilloy, N., Grenier, F., Avino, M., Fournier, I., et al. (2021). OpenProt 2021: deeper
functional annotation of the coding potential of eukaryotic genomes. Nucleic Acids
Research, 49(D1):D380–D388.
Buchan, J. R. and Parker, R. (2009). Eukaryotic stress granules: the ins and outs of
translation. Molecular Cell, 36(6):932–941.
Buttgereit, F. and Brand, M. D. (1995). A hierarchy of ATP-consuming processes in
mammalian cells. Biochemical Journal, 312(1):163–167.
Caeser, R., Di Re, M., Krupka, J. A., Gao, J., Lara-Chica, M., Dias, J. M., Cooke, S. L.,
Fenner, R., Usheva, Z., Runge, H. F., et al. (2019). Genetic modification of primary
human B cells to model high-grade lymphoma. Nature Communications, 10(1):1–16.
Cai, X., Gao, L., Teng, L., Ge, J., Oo, Z. M., Kumar, A. R., Gilliland, D. G., Mason, P. J.,
Tan, K., and Speck, N. A. (2015). Runx1 deficiency decreases ribosome biogenesis and
confers stress resistance to hematopoietic stem and progenitor cells. Cell Stem Cell,
17(2):165–177.
Calado, D. P., Sasaki, Y., Godinho, S. A., Pellerin, A., Ko¨chert, K., Sleckman, B. P., De
Albora´n, I. M., Janz, M., Rodig, S., and Rajewsky, K. (2012). The cell-cycle regulator
c-Myc is essential for the formation and maintenance of germinal centers. Nature
Immunology, 13(11):1092–1100.
Calviello, L. and Ohler, U. (2017). Beyond read-counts: Ribo-seq data analysis to
understand the functions of the transcriptome. Trends in Genetics, 33(10):728–744.
Calviello, L., Venkataramanan, S., Rogowski, K. J., Wyler, E., Wilkins, K., Tejura, M.,
Thai, B., Krol, J., Filipowicz, W., Landthaler, M., et al. (2021). DDX3 depletion represses
translation of mRNAs with complex 5 UTRs. Nucleic Acids Research, 49(9):5336–5350.
Calvo, S. E., Pagliarini, D. J., and Mootha, V. K. (2009). Upstream open reading frames
cause widespread reduction of protein expression and are polymorphic among humans.
Proceedings of the National Academy of Sciences, 106(18):7507–7512.
Cannizzaro, E., Bannister, A. J., Han, N., Alendar, A., and Kouzarides, T. (2018). DDX3X
RNA helicase a↵ects breast cancer cell cycle progression by regulating expression of
KLF4. FEBS Letters, 592(13):2308–2322.
Casey, S. C., Baylot, V., and Felsher, D. W. (2018). The MYC oncogene is a global
regulator of the immune response. Blood, The Journal of the American Society of
Hematology, 131(18):2007–2015.
162
Casey, S. C., Tong, L., Li, Y., Do, R., Walz, S., Fitzgerald, K. N., Gouw, A. M., Baylot,
V., Gu¨tgemann, I., Eilers, M., et al. (2016). MYC regulates the antitumor immune
response through CD47 and PD-L1. Science, 352(6282):227–231.
Cech, T. R. (2000). The ribosome is a ribozyme. Science, 289(5481):878–879.
Chambers, M. C., Maclean, B., Burke, R., Amodei, D., Ruderman, D. L., Neumann, S.,
Gatto, L., Fischer, B., Pratt, B., Egertson, J., et al. (2012). A cross-platform toolkit for
mass spectrometry and proteomics. Nature Biotechnology, 30(10):918–920.
Chapuy, B., Stewart, C., Dunford, A. J., Kim, J., Kamburov, A., Redd, R. A., Lawrence,
M. S., Roemer, M. G., Li, A. J., Ziepert, M., et al. (2018). Molecular subtypes of
di↵use large B cell lymphoma are associated with distinct pathogenic mechanisms and
outcomes. Nature Medicine, 24(5):679–690.
Chen, H., Liu, H., and Qing, G. (2018). Targeting oncogenic Myc as a strategy for cancer
treatment. Signal transduction and targeted therapy, 3(1):1–7.
Chen, J., Brunner, A.-D., Cogan, J. Z., Nun˜ez, J. K., Fields, A. P., Adamson, B., Itzhak,
D. N., Li, J. Y., Mann, M., Leonetti, M. D., et al. (2020). Pervasive functional translation
of noncanonical human open reading frames. Science, 367(6482):1140–1146.
Chew, G.-L., Pauli, A., and Schier, A. F. (2016). Conservation of uORF repressiveness and
sequence features in mouse, human and zebrafish. Nature Communications, 7(1):1–10.
Chick, J. M., Kolippakkam, D., Nusinow, D. P., Zhai, B., Rad, R., Huttlin, E. L., and
Gygi, S. P. (2015). A mass-tolerant database search identifies a large proportion of
unassigned spectra in shotgun proteomics as modified peptides. Nature Biotechnology,
33(7):743–749.
Chong, C., Mu¨ller, M., Pak, H., Harnett, D., Huber, F., Grun, D., Leleu, M., Auger, A.,
Arnaud, M., Stevenson, B. J., et al. (2020). Integrated proteogenomic deep sequencing
and analytics accurately identify non-canonical peptides in tumor immunopeptidomes.
Nature Communications, 11(1):1–21.
Chooniedass-Kothari, S., Emberley, E., Hamedani, M., Troup, S., Wang, X., Czosnek, A.,
Hube, F., Mutawe, M., Watson, P., and Leygue, E. (2004). The steroid receptor RNA
activator is the first functional RNA encoding a protein. FEBS Letters, 566(1-3):43–47.
Chothani, S., Adami, E., Ouyang, J. F., Viswanathan, S., Hubner, N., Cook, S. A., Schafer,
S., and Rackham, O. J. (2019). deltaTE: detection of translationally regulated genes
by integrative analysis of Ribo-seq and RNA-seq data. Current Protocols in Molecular
Biology, 129(1):e108.
163
Churbanov, A., Rogozin, I. B., Babenko, V. N., Ali, H., and Koonin, E. V. (2005).
Evolutionary conservation suggests a regulatory function of AUG triplets in 5-UTRs of
eukaryotic genes. Nucleic Acids Research, 33(17):5512–5520.
Ci, W., Polo, J. M., Cerchietti, L., Shaknovich, R., Wang, L., Yang, S. N., Ye, K., Farinha,
P., Horsman, D. E., Gascoyne, R. D., et al. (2009). The BCL6 transcriptional program
features repression of multiple oncogenes in primary B cells and is deregulated in DLBCL.
Blood, 113(22):5536–5548.
Clarke, H. J., Chambers, J. E., Liniker, E., and Marciniak, S. J. (2014). Endoplasmic
reticulum stress in malignancy. Cancer Cell, 25(5):563–573.
Consortium, G. et al. (2020). The GTEx consortium atlas of genetic regulatory e↵ects
across human tissues. Science, 369(6509):1318–1330.
Costa, L. J., Xavier, A. C., Wahlquist, A. E., and Hill, E. G. (2013). Trends in survival of
patients with Burkitt lymphoma/leukemia in the USA: an analysis of 3691 cases. Blood,
121(24):4861–4866.
Costa-Mattioli, M. and Walter, P. (2020). The integrated stress response: From mechanism
to disease. Science, 368(6489).
Cotton, A. M., Price, E. M., Jones, M. J., Balaton, B. P., Kobor, M. S., and Brown, C. J.
(2015). Landscape of DNA methylation on the X chromosome reflects CpG density,
functional chromatin state and x-chromosome inactivation. Human Molecular Genetics,
24(6):1528–1539.
Cox, J. and Mann, M. (2008). MaxQuant enables high peptide identification rates,
individualized ppb-range mass accuracies and proteome-wide protein quantification.
Nature Biotechnology, 26(12):1367–1372.
Crick, F. H. (1958). On protein synthesis. In Symp Soc Exp Biol, volume 12, page 8.
Cruciat, C.-M., Dolde, C., De Groot, R. E., Ohkawara, B., Reinhard, C., Korswagen,
H. C., and Niehrs, C. (2013). RNA helicase DDX3 is a regulatory subunit of casein
kinase 1 in Wnt–-catenin signaling. Science, 339(6126):1436–1441.
Cucco, F., Barrans, S., Sha, C., Clipson, A., Crouch, S., Dobson, R., Chen, Z., Thompson,
J. S., Care, M. A., Cummin, T., et al. (2020). Distinct genetic changes reveal evolutionary
history and heterogeneous molecular grade of DLBCL with MYC/BCL2 double-hit.
Leukemia, 34(5):1329–1341.
164
Cuevas, M. V. R., Hardy, M.-P., Holly`, J., Bonneil, E´., Durette, C., Courcelles, M., Lanoix,
J., Coˆte´, C., Staudt, L. M., Lemieux, S., et al. (2021). Most non-canonical proteins
uniquely populate the proteome or immunopeptidome. Cell Reports, 34(10):108815.
Culjkovic-Kraljacic, B., Fernando, T. M., Marullo, R., Calvo-Vidal, N., Verma, A., Yang, S.,
Tabbo`, F., Gaudiano, M., Zahreddine, H., Goldstein, R. L., et al. (2016). Combinatorial
targeting of nuclear export and translation of RNA inhibits aggressive b-cell lymphomas.
Blood, 127(7):858–868.
Cutmore, Krupka, Hodson (2022). Molecular profiling in di↵use large b cell lymphoma –
challenges and opportunities. Under review.
Dai, M.-S., Sun, X.-X., and Lu, H. (2010). Ribosomal protein L11 associates with c-Myc
at 5S rRNA and tRNA genes and regulates their expression. Journal of Biological
Chemistry, 285(17):12587–12594.
Dang, C. V. (2012). Myc on the path to cancer. Cell, 149(1):22–35.
de Loubresse, N. G., Prokhorova, I., Holtkamp, W., Rodnina, M. V., Yusupova, G., and
Yusupov, M. (2014). Structural basis for the inhibition of the eukaryotic ribosome.
Nature, 513(7519):517–522.
De Silva, N. S. and Klein, U. (2015). Dynamics of B cells in germinal centres. Nature
Reviews Immunology, 15(3):137–148.
Deeb, S. J., Cox, J., Schmidt-Supprian, M., and Mann, M. (2014). N-linked glycosylation
enrichment for in-depth cell surface proteomics of di↵use large B-cell lymphoma subtypes.
Molecular & Cellular Proteomics, 13(1):240–251.
Deeb, S. J., D’Souza, R. C., Cox, J., Schmidt-Supprian, M., and Mann, M. (2012). Super-
SILAC allows classification of di↵use large B-cell lymphoma subtypes by their protein
expression profiles. Molecular & Cellular Proteomics, 11(5):77–89.
Delaidelli, A., Leprivier, G., and Sorensen, P. H. (2017). eEF2K protects MYCN-amplified
cells from starvation. Cell Cycle, 16(18):1633.
Derrien, T., Johnson, R., Bussotti, G., Tanzer, A., Djebali, S., Tilgner, H., Guernec, G.,
Martin, D., Merkel, A., Knowles, D. G., et al. (2012). The GENCODE v7 catalog of
human long noncoding rnas: analysis of their gene structure, evolution, and expression.
Genome Research, 22(9):1775–1789.
Dersh, D., Holly`, J., and Yewdell, J. W. (2021). A few good peptides: MHC class I-
based cancer immunosurveillance and immunoevasion. Nature Reviews Immunology,
21(2):116–128.
165
Desnoyers, G., Frost, L. D., Courteau, L., Wall, M. L., and Lewis, S. M. (2015). Decreased
eIF3e expression can mediate epithelial-to-mesenchymal transition through activation of
the TGF signaling pathway. Molecular Cancer Research, 13(10):1421–1430.
Deutsch, E. W., Csordas, A., Sun, Z., Jarnuczak, A., Perez-Riverol, Y., Ternent, T.,
Campbell, D. S., Bernal-Llinares, M., Okuda, S., Kawano, S., et al. (2016). The
ProteomeXchange consortium in 2017: supporting the cultural change in proteomics
public data deposition. Nucleic Acids Research, page gkw936.
Dinkel, H., Van Roey, K., Michael, S., Davey, N. E., Weatheritt, R. J., Born, D., Speck,
T., Kru¨ger, D., Grebnev, G., Kuban´, M., et al. (2014). The eukaryotic linear motif
resource ELM: 10 years and counting. Nucleic Acids Research, 42(D1):D259–D266.
Dittmar, K. A., Goodenbour, J. M., and Pan, T. (2006). Tissue-specific di↵erences in
human transfer RNA expression. PLoS genetics, 2(12):e221.
Ditton, H., Zimmer, J., Kamp, C., Rajpert-De Meyts, E., and Vogt, P. (2004). The AZFa
gene dby (DDX3Y) is widely transcribed but the protein is limited to the male germ
cells by translation control. Human Molecular Genetics, 13(19):2333–2341.
Dobin, A., Davis, C. A., Schlesinger, F., Drenkow, J., Zaleski, C., Jha, S., Batut, P.,
Chaisson, M., and Gingeras, T. R. (2013). STAR: ultrafast universal RNA-seq aligner.
Bioinformatics, 29(1):15–21.
Doench, J. G., Fusi, N., Sullender, M., Hegde, M., Vaimberg, E. W., Donovan, K. F.,
Smith, I., Tothova, Z., Wilen, C., Orchard, R., et al. (2016). Optimized sgRNA design to
maximize activity and minimize o↵-target e↵ects of CRISPR-Cas9. Nature Biotechnology,
34(2):184–191.
Doench, J. G., Hartenian, E., Graham, D. B., Tothova, Z., Hegde, M., Smith, I., Sullender,
M., Ebert, B. L., Xavier, R. J., and Root, D. E. (2014). Rational design of highly
active sgRNAs for CRISPR-Cas9–mediated gene inactivation. Nature Biotechnology,
32(12):1262–1267.
Dominguez-Sola, D., Victora, G. D., Ying, C. Y., Phan, R. T., Saito, M., Nussenzweig,
M. C., and Dalla-Favera, R. (2012). The proto-oncogene MYC is required for selection
in the germinal center and cyclic reentry. Nature Immunology, 13(11):1083–1091.
Dresios, J., Chappell, S. A., Zhou, W., and Mauro, V. P. (2006). An mRNA-rRNA
base-pairing mechanism for translation initiation in eukaryotes. Nature structural &
Molecular Biology, 13(1):30–34.
166
Duncan, C. D. and Mata, J. (2017). E↵ects of cycloheximide on the interpretation
of ribosome profiling experiments in Schizosaccharomyces pombe. Scientific Reports,
7(1):1–11.
Dunford, A., Weinstock, D. M., Savova, V., Schumacher, S. E., Cleary, J. P., Yoda, A.,
Sullivan, T. J., Hess, J. M., Gimelbrant, A. A., Beroukhim, R., et al. (2017). Tumor-
suppressor genes that escape from X-inactivation contribute to cancer sex bias. Nature
Genetics, 49(1):10–16.
Dutt, S., Narla, A., Lin, K., Mullally, A., Abayasekara, N., Megerdichian, C., Wilson,
F. H., Currie, T., Khanna-Gupta, A., Berliner, N., et al. (2011). Haploinsuciency for
ribosomal protein genes causes selective activation of p53 in human erythroid progenitor
cells. Blood, 117(9):2567–2576.
Dyson, H. J. and Wright, P. E. (2005). Intrinsically unstructured proteins and their
functions. Nature Reviews Molecular Cell biology, 6(3):197–208.
Eng, J. K., Jahan, T. A., and Hoopmann, M. R. (2013). Comet: an open-source MS/MS
sequence database search tool. Proteomics, 13(1):22–24.
Ennishi, D., Jiang, A., Boyle, M., Collinge, B., Grande, B. M., Ben-Neriah, S., Rushton, C.,
Tang, J., Thomas, N., Slack, G. W., et al. (2019). Double-hit gene expression signature
defines a distinct subgroup of germinal center b-cell-like di↵use large b-cell lymphoma.
Journal of Clinical Oncology, 37(3):190.
Eswarappa, S. M., Potdar, A. A., Koch, W. J., Fan, Y., Vasu, K., Lindner, D., Willard,
B., Graham, L. M., DiCorleto, P. E., and Fox, P. L. (2014). Programmed translational
readthrough generates antiangiogenic VEGF-Ax. Cell, 157(7):1605–1618.
Etzioni, A. and Ochs, H. D. (2004). The hyper IgM syndrome—an evolving story. Pediatric
research, 56(4):519–525.
Ewels, P., Magnusson, M., Lundin, S., and Ka¨ller, M. (2016). MultiQC: summarize
analysis results for multiple tools and samples in a single report. Bioinformatics,
32(19):3047–3048.
Ferlin, A., Moro, E., Rossi, A., Dallapiccola, B., and Foresta, C. (2003). The human Y
chromosome’s azoospermia factor b (azfb) region: sequence, structure, and deletion
analysis in infertile men. Journal of Medical Genetics, 40(1):18–24.
Fields, A. P., Rodriguez, E. H., Jovanovic, M., Stern-Ginossar, N., Haas, B. J., Mertins, P.,
Raychowdhury, R., Hacohen, N., Carr, S. A., Ingolia, N. T., et al. (2015). A regression-
based analysis of ribosome-profiling data reveals a conserved complexity to mammalian
translation. Molecular Cell, 60(5):816–827.
167
Filipowicz, W., Bhattacharyya, S. N., and Sonenberg, N. (2008). Mechanisms of post-
transcriptional regulation by microRNAs: are the answers in sight? Nature Reviews
Genetics, 9(2):102–114.
Floor, S. N., Condon, K. J., Sharma, D., Jankowsky, E., and Doudna, J. A. (2016).
Autoinhibitory interdomain interactions and subfamily-specific extensions redefine the
catalytic core of the human dead-box protein ddx3. Journal of Biological Chemistry,
291(5):2412–2421.
Foresta, C., Ferlin, A., and Moro, E. (2000a). Deletion and expression analysis of AZFa
genes on the human Y chromosome revealed a major role for DBY in male infertility.
Human Molecular Genetics, 9(8):1161–1169.
Foresta, C., Ferlin, A., and Moro, E. (2000b). Deletion and expression analysis of AZFa
genes on the human Y chromosome revealed a major role for DBY in male infertility.
Human Molecular Genetics, 9(8):1161–1169.
Frankish, A., Diekhans, M., Ferreira, A.-M., Johnson, R., Jungreis, I., Loveland, J., Mudge,
J. M., Sisu, C., Wright, J., Armstrong, J., et al. (2019). GENCODE reference annotation
for the human and mouse genomes. Nucleic Acids Research, 47(D1):D766–D773.
Furic, L., Rong, L., Larsson, O., Koumakpayi, I. H., Yoshida, K., Brueschke, A., Petroulakis,
E., Robichaud, N., Pollak, M., Gaboury, L. A., et al. (2010). eIF4E phosphorylation
promotes tumorigenesis and is associated with prostate cancer progression. Proceedings
of the National Academy of Sciences, 107(32):14134–14139.
Gani, R. (1976). The nucleoli of cultured human lymphocytes: I. nucleolar morphology in
relation to transformation and the DNA cycle. Experimental Cell Research, 97(2):249–
258.
Gao, H., Sun, X., and Rao, Y. (2020). PROTAC technology: opportunities and challenges.
ACS Medicinal Chemistry Letters, 11(3):237–240.
Gao, X., Wan, J., Liu, B., Ma, M., Shen, B., and Qian, S.-B. (2015). Quantitative profiling
of initiating ribosomes in vivo. Nature Methods, 12(2):147–153.
Gao, Z., Herrera-Carrillo, E., and Berkhout, B. (2018). Delineation of the exact transcrip-
tion termination signal for type 3 polymerase III. Molecular Therapy-Nucleic Acids,
10:36–44.
Geissler, R., Golbik, R. P., and Behrens, S.-E. (2012). The DEAD-box helicase DDX3
supports the assembly of functional 80s ribosomes. Nucleic Acids Research, 40(11):4998–
5011.
168
Genuth, N. R. and Barna, M. (2018). The discovery of ribosome heterogeneity and its
implications for gene regulation and organisms life. Molecular Cell, 71(3):364–374.
Gerashchenko, M. V. and Gladyshev, V. N. (2014). Translation inhibitors cause abnormal-
ities in ribosome profiling experiments. Nucleic Acids Research, 42(17):e134–e134.
Gerashchenko, M. V. and Gladyshev, V. N. (2017). Ribonuclease selection for ribosome
profiling. Nucleic Acids Research, 45(2):e6–e6.
Ghaddar, N., Wang, S., Woodvine, B., Krishnamoorthy, J., van Hoef, V., Darini, C.,
Kazimierczak, U., Ah-Son, N., Popper, H., Johnson, M., et al. (2021). The integrated
stress response is tumorigenic and constitutes a therapeutic liability in KRAS-driven
lung cancer. Nature Communications, 12(1):1–15.
Gingras, A.-C., Raught, B., and Sonenberg, N. (1999). eIF4 initiation factors: e↵ectors
of mRNA recruitment to ribosomes and regulators of translation. Annual review of
biochemistry, 68(1):913–963.
God, J. M., Cameron, C., Figueroa, J., Amria, S., Hossain, A., Kempkes, B., Bornkamm,
G. W., Stuart, R. K., Blum, J. S., and Haque, A. (2015). Elevation of c-MYC disrupts
HLA class II–mediated immune recognition of human B cell tumors. The Journal of
Immunology, 194(4):1434–1445.
Gong, Krupka, Gao, J., Grigoropoulos, N. F., Screen, M., Usheva, Z., Cucco, F., Barrans,
S., Painter, D., Mohammed, M., et al. (2021). Sequential inverse dysregulation of the
RNA helicases DDX3X and DDX3Y facilitates MYC-driven lymphomagenesis.
Good-Jacobson, K. L., Szumilas, C. G., Chen, L., Sharpe, A. H., Tomayko, M. M.,
and Shlomchik, M. J. (2010). PD-1 regulates germinal center B cell survival and the
formation and anity of long-lived plasma cells. Nature Immunology, 11(6):535–542.
Goodman, A., Patel, S. P., and Kurzrock, R. (2017). PD-1–PD-L1 immune-checkpoint
blockade in B-cell lymphomas. Nature Reviews Clinical Oncology, 14(4):203–220.
Grande, B. M., Gerhard, D. S., Jiang, A., Griner, N. B., Abramson, J. S., Alexander,
T. B., Allen, H., Ayers, L. W., Bethony, J. M., Bhatia, K., et al. (2019). Genome-wide
discovery of somatic coding and noncoding mutations in pediatric endemic and sporadic
Burkitt lymphoma. Blood, 133(12):1313–1324.
Grandori, C., Gomez-Roman, N., Felton-Edkins, Z. A., Ngouenet, C., Galloway, D. A.,
Eisenman, R. N., and White, R. J. (2005). c-Myc binds to human ribosomal DNA and
stimulates transcription of rRNA genes by RNA polymerase I. Nature Cell Biology,
7(3):311–318.
169
Green, M. R., Monti, S., Rodig, S. J., Juszczynski, P., Currie, T., O’Donnell, E., Chapuy,
B., Takeyama, K., Neuberg, D., Golub, T. R., et al. (2010). Integrative analysis reveals
selective 9p24. 1 amplification, increased PD-1 ligand expression, and further induction
via JAK2 in nodular sclerosing Hodgkin lymphoma and primary mediastinal large b-cell
lymphoma. Blood, 116(17):3268–3277.
Griss, J., Perez-Riverol, Y., Lewis, S., Tabb, D. L., Dianes, J. A., Del-Toro, N., Rurik,
M., Walzer, M., Kohlbacher, O., Hermjakob, H., et al. (2016). Recognizing millions of
consistently unidentified spectra across hundreds of shotgun proteomics datasets. Nature
Methods, 13(8):651–656.
Guttman, M., Amit, I., Garber, M., French, C., Lin, M. F., Feldser, D., Huarte, M., Zuk, O.,
Carey, B. W., Cassady, J. P., et al. (2009). Chromatin signature reveals over a thousand
highly conserved large non-coding RNAs in mammals. Nature, 458(7235):223–227.
Guttman, M., Russell, P., Ingolia, N. T., Weissman, J. S., and Lander, E. S. (2013).
Ribosome profiling provides evidence that large noncoding RNAs do not encode proteins.
Cell, 154(1):240–251.
Hafner, M., Katsantoni, M., Ko¨ster, T., Marks, J., Mukherjee, J., Staiger, D., Ule, J.,
and Zavolan, M. (2021). CLIP and complementary methods. Nature Reviews Methods
Primers, 1(1):1–23.
Ha¨mmerl, L., Colombet, M., Rochford, R., Ogwang, D. M., and Parkin, D. M. (2019).
The burden of Burkitt lymphoma in Africa. Infectious Agents and Cancer, 14(1):1–6.
Hanahan, D. and Weinberg, R. A. (2011). Hallmarks of cancer: the next generation. cell,
144(5):646–674.
Hanson, G. and Coller, J. (2018). Codon optimality, bias and usage in translation and
mRNA decay. Nature reviews Molecular cell biology, 19(1):20.
Hariri, F., Arguello, M., Volpon, L., Culjkovic-Kraljacic, B., Nielsen, T. H., Hiscott, J.,
Mann, K. K., and Borden, K. L. (2013). The eukaryotic translation initiation factor
eif4e is a direct transcriptional target of NF-B and is aberrantly regulated in acute
myeloid leukemia. Leukemia, 27(10):2047–2055.
He, Y., Zhang, D., Yang, Y., Wang, X., Zhao, X., Zhang, P., Zhu, H., Xu, N., and Liang,
S. (2018). A double-edged function of DDX3, as an oncogene or tumor suppressor, in
cancer progression. Oncology reports, 39(3):883–892.
Hellen, C. U. (2018). Translation termination and ribosome recycling in eukaryotes. Cold
Spring Harbor perspectives in biology, 10(10):a032656.
170
Helmrich, A., Ballarino, M., and Tora, L. (2011). Collisions between replication and
transcription complexes cause common fragile site instability at the longest human genes.
Molecular cell, 44(6):966–977.
Henderson, A., Warburton, D., and Atwood, K. (1972). Location of ribosomal DNA in
the human chromosome complement. Proceedings of the National Academy of Sciences,
69(11):3394–3398.
Hershey, J. W. B., Sonenberg, N., and Mathews, M. B. (2012). Principles of Translational
Control: An Overview. Cold Spring Harbor Perspectives in Biology, 4(12):a011528–
a011528.
Hetz, C. (2012). The unfolded protein response: controlling cell fate decisions under ER
stress and beyond. Nature reviews Molecular cell biology, 13(2):89–102.
Hetz, C., Chevet, E., and Oakes, S. A. (2015). Proteostasis control by the unfolded protein
response. Nature Cell Biology, 17(7):829–838.
Hilliker, A., Gao, Z., Jankowsky, E., and Parker, R. (2011). The DEAD-box protein Ded1
modulates translation by the formation and resolution of an eIF4F-mRNA complex.
Molecular Cell, 43(6):962–972.
Hinnebusch, A. G. (2005). Translational regulation of GCN4 and the general amino acid
control of yeast. Annu. Rev. Microbiol., 59:407–450.
Hinnebusch, A. G. (2014). The scanning mechanism of eukaryotic translation initiation.
Annual Review of Biochemistry, 83:779–812.
Hinnebusch, A. G., Ivanov, I. P., and Sonenberg, N. (2016). Translational control by
5-untranslated regions of eukaryotic mRNAs. Science, 352(6292):1413–1416.
Ho, J. S., Ma, W., Mao, D. Y., and Benchimol, S. (2005). p53-dependent transcriptional
repression of c-myc is required for G1 cell cycle arrest. Molecular and Cellular Biology,
25(17):7423–7431.
Ho, Y., Li, X., Jamison, S., Harding, H. P., McKinnon, P. J., Ron, D., and Lin, W. (2016).
PERK activation promotes medulloblastoma tumorigenesis by attenuating premalignant
granule cell precursor apoptosis. The American Journal of Pathology, 186(7):1939–1951.
Horvilleur, E., Sbarrato, T., Hill, K., Spriggs, R., Screen, M., Goodrem, P., Sawicka, K.,
Chaplin, L., Touriol, C., Packham, G., et al. (2014). A role for eukaryotic initiation
factor 4B overexpression in the pathogenesis of di↵use large b-cell lymphoma. Leukemia,
28(5):1092–1102.
171
Horvilleur, E., Wilson, L. A., and Willis, A. E. (2010). Translation deregulation in B-cell
lymphomas.
Howden, A. J., Geoghegan, V., Katsch, K., Efstathiou, G., Bhushan, B., Boutureira, O.,
Thomas, B., Trudgian, D. C., Kessler, B. M., Dieterich, D. C., et al. (2013). QuaNCAT:
quantitating proteome dynamics in primary cells. Nature Methods, 10(4):343–346.
Hsieh, A. C., Liu, Y., Edlind, M. P., Ingolia, N. T., Janes, M. R., Sher, A., Shi, E. Y.,
Stumpf, C. R., Christensen, C., Bonham, M. J., Wang, S., Ren, P., Martin, M., Jessen,
K., Feldman, M. E., Weissman, J. S., Shokat, K. M., Rommel, C., and Ruggero, D.
(2012). The translational landscape of mTOR signalling steers cancer initiation and
metastasis. Nature, 485(7396):55–61.
Hu, F., Lu, J., Munoz, M. D., Saveliev, A., and Turner, M. (2021). ORFLine: a
bioinformatic pipeline to prioritise small open reading frames identifies candidate secreted
small proteins from lymphocytes. bioRxiv.
Huang, L., Luo, R., Li, J., Wang, D., Zhang, Y., Liu, L., Zhang, N., Xu, X., Lu, B., and
Zhao, K. (2020). -catenin promotes NLRP3 inflammasome activation via increasing
the association between NLRP3 and ASC. Molecular Immunology, 121:186–194.
Hube, F., Velasco, G., Rollin, J., Furling, D., and Francastel, C. (2011). Steroid receptor
RNA activator protein binds to and counteracts SRA RNA-mediated activation of MyoD
and muscle di↵erentiation. Nucleic Acids Research, 39(2):513–525.
Hussmann, J. A., Patchett, S., Johnson, A., Sawyer, S., and Press, W. H. (2015). Un-
derstanding biases in ribosome profiling experiments reveals signatures of translation
dynamics in yeast. PLoS Genetics, 11(12):e1005732.
Ingolia, N. T., Brar, G. A., Rouskin, S., McGeachy, A. M., and Weissman, J. S. (2012).
The ribosome profiling strategy for monitoring translation in vivo by deep sequencing of
ribosome-protected mRNA fragments. Nature Protocols, 7(8):1534–1550.
Ingolia, N. T., Ghaemmaghami, S., Newman, J. R., and Weissman, J. S. (2009). Genome-
wide analysis in vivo of translation with nucleotide resolution using ribosome profiling.
Science, 324(5924):218–223.
Ingolia, N. T., Lareau, L. F., and Weissman, J. S. (2011). Ribosome profiling of mouse
embryonic stem cells reveals the complexity and dynamics of mammalian proteomes.
Cell, 147(4):789–802.
Iossifov, I., O’roak, B. J., Sanders, S. J., Ronemus, M., Krumm, N., Levy, D., Stessman,
H. A., Witherspoon, K. T., Vives, L., Patterson, K. E., et al. (2014). The contribution
of de novo coding mutations to autism spectrum disorder. Nature, 515(7526):216–221.
172
Ivanov, P., Emara, M. M., Villen, J., Gygi, S. P., and Anderson, P. (2011). Angiogenin-
Induced tRNA Fragments Inhibit Translation Initiation. Molecular Cell, 43(4):613–623.
Ivanov, P., O’Day, E., Emara, M. M., Wagner, G., Lieberman, J., and Anderson, P. (2014).
G-quadruplex structures contribute to the neuroprotective e↵ects of angiogenin-induced
trna fragments. Proceedings of the National Academy of Sciences, 111(51):18201–18206.
Iwasaki, S. and Ingolia, N. T. (2017). The growing toolbox for protein synthesis studies.
Trends in biochemical sciences, 42(8):612–624.
Jackson, R., Kroehling, L., Khitun, A., Bailis, W., Jarret, A., York, A. G., Khan, O. M.,
Brewer, J. R., Skadow, M. H., Duizer, C., et al. (2018). The translation of non-canonical
open reading frames controls mucosal immunity. Nature, 564(7736):434–438.
Jackson, R. J., Hellen, C. U., and Pestova, T. V. (2010). The mechanism of eukaryotic
translation initiation and principles of its regulation. Nature Reviews Molecular cell
biology, 11(2):113–127.
Jain, S., Wheeler, J. R., Walters, R. W., Agrawal, A., Barsic, A., and Parker, R. (2016).
ATPase-modulated stress granules contain a diverse proteome and substructure. Cell,
164(3):487–498.
Jaroszynski, L., Zimmer, J., Fietz, D., Bergmann, M., Kliesch, S., and Vogt, P. (2011).
Translational control of the AZFa gene DDX3Y by 5 UTR exon-T extension. Interna-
tional journal of andrology, 34(4pt1):313–326.
Ji, Z. (2018). RibORF: identifying genome-wide translated open reading frames using
ribosome profiling. Current Protocols in Molecular Biology, 124(1):e67.
Ji, Z., Song, R., Huang, H., Regev, A., and Struhl, K. (2016). Transcriptome-scale
rnase-footprinting of rna-protein complexes. Nature Biotechnology, 34(4):410–413.
Ji, Z., Song, R., Regev, A., and Struhl, K. (2015). Many lncRNAs, 5’UTRs, and pseudo-
genes are translated and some are likely to express functional proteins. elife, 4:e08890.
Jiang, L., Gu, Z.-H., Yan, Z.-X., Zhao, X., Xie, Y.-Y., Zhang, Z.-G., Pan, C.-M., Hu, Y.,
Cai, C.-P., Dong, Y., et al. (2015). Exome sequencing identifies somatic mutations of
DDX3X in natural killer/T-cell lymphoma. Nature Genetics, 47(9):1061–1066.
Johannes, G., Carter, M. S., Eisen, M. B., Brown, P. O., and Sarnow, P. (1999). Iden-
tification of eukaryotic mRNAs that are translated at reduced cap binding complex
eIF4F concentrations using a cDNA microarray. Proceedings of the National Academy
of Sciences, 96(23):13118–13123.
173
Johnson, L. F., Levis, R., Abelson, H. T., Green, H., and Penman, S. (1976). Changes
in RNA in relation to growth of the fibroblast. iv. alterations in the production and
processing of mRNA and rrna in resting and growing cells. The Journal of Cell Biology,
71(3):933–938.
Johnson-Kerner, B., Blok, L. S., Suit, L., Thomas, J., Kleefstra, T., and Sherr, E. H.
(2020). DDX3X-related neurodevelopmental disorder. GeneReviews R[Internet].
Johnston, H. E., Carter, M. J., Larrayoz, M., Clarke, J., Garbis, S. D., Oscier, D., Stre↵ord,
J. C., Steele, A. J., Walewska, R., and Cragg, M. S. (2018). Proteomics profiling of CLL
versus healthy B-cells identifies putative therapeutic targets and a subtype-independent
signature of spliceosome dysregulation. Molecular & Cellular Proteomics, 17(4):776–791.
Johnstone, T. G., Bazzini, A. A., and Giraldez, A. J. (2016). Upstream ORF s are prevalent
translational repressors in vertebrates. The EMBO journal, 35(7):706–723.
Jones, D. T., Ja¨ger, N., Kool, M., Zichner, T., Hutter, B., Sultan, M., Cho, Y.-J., Pugh,
T. J., Hovestadt, V., Stu¨tz, A. M., et al. (2012). Dissecting the genomic complexity
underlying medulloblastoma. Nature, 488(7409):100–105.
Joshi-Tope, G., Gillespie, M., Vastrik, I., D’Eustachio, P., Schmidt, E., de Bono, B., Jassal,
B., Gopinath, G., Wu, G., Matthews, L., et al. (2005). Reactome: a knowledgebase of
biological pathways. Nucleic Acids Research, 33(suppl 1):D428–D432.
Jovanovic, M., Rooney, M. S., Mertins, P., Przybylski, D., Chevrier, N., Satija, R.,
Rodriguez, E. H., Fields, A. P., Schwartz, S., Raychowdhury, R., et al. (2015). Dynamic
profiling of the protein life cycle in response to pathogens. Science, 347(6226).
Kampen, K. R., Sulima, S. O., Vereecke, S., and De Keersmaecker, K. (2020). Hallmarks
of ribosomopathies. Nucleic acids research, 48(3):1013–1028.
Kapadia, B., Nanaji, N. M., Bhalla, K., Bhandary, B., Lapidus, R., Beheshti, A., Evens,
A. M., and Gartenhaus, R. B. (2018). Fatty acid synthase induced S6Kinase facilitates
USP11-eIF4B complex formation for sustained oncogenic translation in DLBCL. Nature
Communications, 9(1):1–15.
Karginov, F. V. and Hannon, G. J. (2013). Remodeling of Ago2–mRNA interactions upon
cellular stress reflects mirna complementarity and correlates with altered translation
rates. Genes & development, 27(14):1624–1632.
Kataoka, K., Shiraishi, Y., Takeda, Y., Sakata, S., Matsumoto, M., Nagano, S., Maeda,
T., Nagata, Y., Kitanaka, A., Mizuno, S., et al. (2016). Aberrant PD-L1 expression
through 3-UTR disruption in multiple cancers. Nature, 534(7607):402–406.
174
Kaymaz, Y., Oduor, C. I., Yu, H., Otieno, J. A., Ong’echa, J. M., Moormann, A. M., and
Bailey, J. A. (2017). Comprehensive transcriptome and mutational profiling of endemic
Burkitt lymphoma reveals EBV type–specific di↵erences. Molecular Cancer Research,
15(5):563–576.
Kearse, M. G. and Wilusz, J. E. (2017). Non-AUG translation: a new start for protein
synthesis in eukaryotes. Genes & Development, 31(17):1717–1731.
Kellaris, G., Khan, K., Baig, S. M., Tsai, I.-C., Zamora, F. M., Ruggieri, P., Natowicz,
M. R., and Katsanis, N. (2018). A hypomorphic inherited pathogenic variant in
DDX3X causes male intellectual disability with additional neurodevelopmental and
neurodegenerative features. Human Genomics, 12(1):1–9.
Kent, W. J., Sugnet, C. W., Furey, T. S., Roskin, K. M., Pringle, T. H., Zahler, A. M.,
and Haussler, D. (2002). The human genome browser at UCSC. Genome research,
12(6):996–1006.
Kent, W. J., Zweig, A. S., Barber, G., Hinrichs, A. S., and Karolchik, D. (2010). BigWig
and BigBed: enabling browsing of large distributed datasets. Bioinformatics, 26(17):2204–
2207.
Ketteler, R. (2012). On programmed ribosomal frameshifting: the alternative proteomes.
Frontiers in genetics, 3:242.
Kevil, C. G., De Benedetti, A., Payne, D. K., Coe, L. L., Laroux, F. S., and Alexander,
J. S. (1996). Translational regulation of vascular permeability factor by eukaryotic
initiation factor 4E: implications for tumor angiogenesis. International Journal of
Cancer, 65(6):785–790.
Khodadoust, M. S., Olsson, N., Chen, B., Sworder, B., Shree, T., Liu, C. L., Zhang, L.,
Czerwinski, D. K., Davis, M. M., Levy, R., et al. (2019). B-cell lymphomas present
immunoglobulin neoantigens. Blood, 133(8):878–881.
Khodadoust, M. S., Olsson, N., Wagar, L. E., Haabeth, O. A., Chen, B., Swaminathan, K.,
Rawson, K., Liu, C. L., Steiner, D., Lund, P., et al. (2017). Antigen presentation profiling
reveals recognition of lymphoma immunoglobulin neoantigens. Nature, 543(7647):723–
727.
Kim, J. and Guan, K. L. (2019a). mTOR as a central hub of nutrient signalling and cell
growth. Nature Cell Biology, 21(1):63–71.
Kim, J. and Guan, K.-L. (2019b). mTOR as a central hub of nutrient signalling and cell
growth. Nature Cell Biology, 21(1):63–71.
175
Kiyasu, J., Miyoshi, H., Hirata, A., Arakawa, F., Ichikawa, A., Niino, D., Sugita, Y., Yufu,
Y., Choi, I., Abe, Y., et al. (2015). Expression of programmed cell death ligand 1 is
associated with poor overall survival in patients with di↵use large b-cell lymphoma.
Blood, 126(19):2193–2201.
Klein, U. and Dalla-Favera, R. (2008). Germinal centres: role in B-cell physiology and
malignancy. Nature Reviews Immunology, 8(1):22–33.
Knight, J. R., Garland, G., Poyry, T., Mead, E., Vlahov, N., Sfakianos, A., Grosso, S.,
De-Lima-Hedayioglu, F., Mallucci, G. R., Von Der Haar, T., Smales, C. M., Sansom,
O. J., and Willis, A. E. (2020). Control of translation elongation in health and disease.
DMM Disease Models and Mechanisms, 13(3).
Komar, A. A. and Hatzoglou, M. (2011). Cellular IRES-mediated translation: the war of
ITAFs in pathophysiological states. Cell Cycle, 10(2):229–240.
Kondo, T., Plaza, S., Zanet, J., Benrabah, E., Valenti, P., Hashimoto, Y., Kobayashi, S.,
Payre, F., and Kageyama, Y. (2010). Small peptides switch the transcriptional activity
of shavenbaby during drosophila embryogenesis. Science, 329(5989):336–339.
Kozak, M. (1987). At least six nucleotides preceding the aug initiator codon enhance
translation in mammalian cells. Journal of Molecular Biology, 196(4):947–950.
Krokhin, O. V. and Spicer, V. (2010). Predicting peptide retention times for proteomics.
Current Protocols in Bioinformatics, 31(1):13–14.
Kumari, P. and Sampath, K. (2015). cncRNAs: Bi-functional RNAs with protein coding
and non-coding functions. In Seminars in Cell & Developmental Biology, volume 47,
pages 40–51. Elsevier.
Ku¨ppers, R. and Dalla-Favera, R. (2001). Mechanisms of chromosomal translocations in B
cell lymphomas. Oncogene, 20(40):5580–5594.
Kurosaki, T. and Maquat, L. E. (2016). Nonsense-mediated mRNA decay in humans at a
glance. Journal of Cell Science, 129(3):461–467.
Kustatscher, G., Grabowski, P., Schrader, T. A., Passmore, J. B., Schrader, M., and
Rappsilber, J. (2019). Co-regulation map of the human proteome enables identification
of protein functions. Nature Biotechnology, 37(11):1361–1371.
Labun, K., Montague, T. G., Krause, M., Torres Cleuren, Y. N., Tjeldnes, H., and Valen,
E. (2019). CHOPCHOP v3: expanding the CRISPR web toolbox beyond genome editing.
Nucleic Acids Research, 47(W1):W171–W174.
176
Lacy, S. E., Barrans, S. L., Beer, P. A., Painter, D., Smith, A. G., Roman, E., Cooke,
S. L., Ruiz, C., Glover, P., Van Hoppe, S. J., et al. (2020). Targeted sequencing in
DLBCL, molecular subtypes, and outcomes: a haematological malignancy research
network report. Blood, 135(20):1759–1771.
Lafontaine, D. L. (2015). Noncoding RNAs in eukaryotic ribosome biogenesis and function.
Nature Structural & Molecular Biology, 22(1):11–19.
Lafontaine, D. L., Riback, J. A., Bascetin, R., and Brangwynne, C. P. (2021). The
nucleolus as a multiphase liquid condensate. Nature Reviews Molecular Cell Biology,
22(3):165–182.
Langmead, B. and Salzberg, S. L. (2012). Fast gapped-read alignment with Bowtie 2.
Nature Methods, 9(4):357–359.
Larsson, O., Sonenberg, N., and Nadon, R. (2010). Identification of di↵erential translation
in genome wide studies. Proceedings of the National Academy of Sciences, 107(50):21487–
21492.
Larsson, O., Sonenberg, N., and Nadon, R. (2011). anota: Analysis of di↵erential
translation in genome-wide studies. Bioinformatics, 27(10):1440–1441.
Lauria, F., Tebaldi, T., Bernabo`, P., Groen, E. J., Gillingwater, T. H., and Viero, G.
(2018). ribowaltz: optimization of ribosome p-site positioning in ribosome profiling data.
PLoS computational biology, 14(8):e1006169.
Law, C. W., Chen, Y., Shi, W., and Smyth, G. K. (2014). voom: Precision weights unlock
linear model analysis tools for RNA-seq read counts. Genome Biology, 15(2):1–17.
Lawrence, M., Huber, W., Pages, H., Aboyoun, P., Carlson, M., Gentleman, R., Morgan,
M. T., and Carey, V. J. (2013). Software for computing and annotating genomic ranges.
PLoS Computational Biology, 9(8):e1003118.
Lawrie, C. H., Chi, J., Taylor, S., Tramonti, D., Ballabio, E., Palazzo, S., Saunders, N. J.,
Pezzella, F., Boultwood, J., Wainscoat, J. S., et al. (2009). Expression of microRNAs
in di↵use large b cell lymphoma is associated with immunophenotype, survival and
transformation from follicular lymphoma. Journal of Cellular and Molecular Medicine,
13(7):1248–1260.
Lazaris-Karatzas, A., Montine, K. S., and Sonenberg, N. (1990). Malignant transfor-
mation by a eukaryotic initiation factor subunit that binds to mRNA 5’cap. Nature,
345(6275):544–547.
177
Lee, A. S., Kranzusch, P. J., Doudna, J. A., and Cate, J. H. (2016). eIF3d is an
mRNA cap-binding protein that is required for specialized translation initiation. Nature,
536(7614):96–99.
Lee, C.-S., Dias, A. P., Jedrychowski, M., Patel, A. H., Hsu, J. L., and Reed, R. (2008a).
Human DDX3 functions in translation and interacts with the translation initiation
factor eIF3. Nucleic Acids Research, 36(14):4708–4718.
Lee, C.-S., Dias, A. P., Jedrychowski, M., Patel, A. H., Hsu, J. L., and Reed, R. (2008b).
Human DDX3 functions in translation and interacts with the translation initiation
factor eIF3. Nucleic Acids Research, 36(14):4708–4718.
Lee, S., Liu, B., Lee, S., Huang, S.-X., Shen, B., and Qian, S.-B. (2012). Global mapping of
translation initiation sites in mammalian cells at single-nucleotide resolution. Proceedings
of the National Academy of Sciences, 109(37):E2424–E2432.
Lennox, A. L., Hoye, M. L., Jiang, R., Johnson-Kerner, B. L., Suit, L. A., Venkataramanan,
S., Sheehan, C. J., Alsina, F. C., Fregeau, B., Aldinger, K. A., et al. (2020a). Pathogenic
DDX3X mutations impair RNA metabolism and neurogenesis during fetal cortical
development. Neuron, 106(3):404–420.
Lennox, A. L., Hoye, M. L., Jiang, R., Johnson-Kerner, B. L., Suit, L. A., Venkataramanan,
S., Sheehan, C. J., Alsina, F. C., Fregeau, B., Aldinger, K. A., et al. (2020b). Pathogenic
DDX3X mutations impair RNA metabolism and neurogenesis during fetal cortical
development. Neuron, 106(3):404–420.
Leprivier, G., Remke, M., Rotblat, B., Dubuc, A., Mateo, A.-R. F., Kool, M., Agnihotri,
S., El-Naggar, A., Yu, B., Somasekharan, S. P., et al. (2013). The eEF2 kinase confers
resistance to nutrient deprivation by blocking translation elongation. Cell, 153(5):1064–
1079.
Leucci, E., Cocco, M., Onnis, A., De Falco, G., Van Cleef, P., Bellan, C., Van Rijk,
A., Nyagol, J., Byakika, B., Lazzi, S., et al. (2008). MYC translocation-negative
classical Burkitt lymphoma cases: an alternative pathogenetic mechanism involving
mirna deregulation. The Journal of Pathology: A Journal of the Pathological Society of
Great Britain and Ireland, 216(4):440–450.
Li, C., Kim, S.-W., Rai, D., Bolla, A. R., Adhvaryu, S., Kinney, M. C., Robetorye, R. S.,
and Aguiar, R. C. (2009a). Copy number abnormalities, MYC activity, and the genetic
fingerprint of normal B cells mechanistically define the microRNA profile of di↵use large
B-cell lymphoma. Blood, 113(26):6681–6690.
178
Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N., Marth, G., Abecasis,
G., and Durbin, R. (2009b). The sequence alignment/map format and SAMtools.
Bioinformatics, 25(16):2078–2079.
Li, J. and Liu, C. (2019). Coding or noncoding, the converging concepts of RNAs. Frontiers
in genetics, 10:496.
Li, W., Wang, W., Uren, P. J., Penalva, L. O., and Smith, A. D. (2017). Riborex: fast
and flexible identification of di↵erential translation from Ribo-seq data. Bioinformatics,
33(11):1735–1737.
Liao, J.-M., Zhou, X., Gatignol, A., and Lu, H. (2014). Ribosomal proteins L5 and
L11 co-operatively inactivate c-Myc via RNA-induced silencing complex. Oncogene,
33(41):4916–4923.
Linder, P. and Jankowsky, E. (2011). From unwinding to clamping the DEAD box RNA
helicase family. Nature Reviews Molecular Cell Biology, 12(8):505–516.
Lindstro¨m, M. S., Jurada, D., Bursac, S., Orsolic, I., Bartek, J., and Volarevic, S. (2018).
Nucleolus as an emerging hub in maintenance of genome stability and cancer pathogenesis.
Oncogene, 37(18):2351–2366.
Liu, G. Y. and Sabatini, D. M. (2020). mTOR at the nexus of nutrition, growth, ageing
and disease. Nature Reviews Molecular Cell Biology, 21(4):183–203.
Liu, J., Xu, Y., Stoleru, D., and Salic, A. (2012). Imaging protein synthesis in cells and
tissues with an alkyne analog of puromycin. Proceedings of the National Academy of
Sciences, 109(2):413–418.
Liu, P., Ge, M., Hu, J., Li, X., Che, L., Sun, K., Cheng, L., Huang, Y., Pilo, M. G., Cigliano,
A., et al. (2017). A functional mammalian target of rapamycin complex 1 signaling is
indispensable for c-Myc-driven hepatocarcinogenesis. Hepatology, 66(1):167–181.
Liu, Y., Beyer, A., and Aebersold, R. (2016). On the dependency of cellular protein levels
on mRNA abundance. Cell, 165(3):535–550.
Lo´pez, C., Kleinheinz, K., Aukema, S. M., Rohde, M., Bernhart, S. H., Hu¨bschmann,
D., Wagener, R., Toprak, U. H., Raimondi, F., Kreuz, M., et al. (2019). Genomic and
transcriptomic changes complement each other in the pathogenesis of sporadic Burkitt
lymphoma. Nature communications, 10(1):1–19.
Love, M. I., Huber, W., and Anders, S. (2014). Moderated estimation of fold change and
dispersion for RNA-seq data with DESeq2. Genome Biology, 15(12):1–21.
179
Lu, J., Cannizzaro, E., Meier-Abt, F., Scheinost, S., Bruch, P.-M., Giles, H. A., Lu¨tge,
A., Hu¨llein, J., Wagner, L., Giacopelli, B., et al. (2021). Multi-omics reveals clinically
relevant proliferative drive associated with mTOR-MYC-OXPHOS activity in chronic
lymphocytic leukemia. Nature Cancer, 2(8):853–864.
Lundberg, E., Fagerberg, L., Klevebring, D., Matic, I., Geiger, T., Cox, J., A¨lgena¨s,
C., Lundeberg, J., Mann, M., and Uhlen, M. (2010). Defining the transcriptome and
proteome in three functionally di↵erent human cell lines. Molecular Systems Biology,
6(1):450.
Lynch, M. and Marinov, G. K. (2015). The bioenergetic costs of a gene. Proceedings of
the National Academy of Sciences, 112(51):15690–15695.
Mackowiak, S. D., Zauber, H., Bielow, C., Thiel, D., Kutz, K., Calviello, L., Mastrobuoni,
G., Rajewsky, N., Kempa, S., Selbach, M., et al. (2015). Extensive identification and
analysis of conserved small ORFs in animals. Genome Biology, 16(1):1–21.
Malumbres, R., Sarosiek, K. A., Cubedo, E., Ruiz, J. W., Jiang, X., Gascoyne, R. D.,
Tibshirani, R., and Lossos, I. S. (2009). Di↵erentiation stage–specific expression of
microRNAs in B lymphocytes and di↵use large B-cell lymphomas. Blood, 113(16):3754–
3764.
Mangus, D. A., Evans, M. C., and Jacobson, A. (2003). Poly (A)-binding proteins:
multifunctional sca↵olds for the post-transcriptional control of gene expression. Genome
Biology, 4(7):1–14.
Maquat, L. E., Tarn, W.-Y., and Isken, O. (2010). The pioneer round of translation:
features and functions. Cell, 142(3):368–374.
Martin, M. (2011). Cutadapt removes adapter sequences from high-throughput sequencing
reads. EMBnet. Journal, 17(1):10–12.
Mathieson, T., Franken, H., Kosinski, J., Kurzawa, N., Zinn, N., Sweetman, G., Poeckel,
D., Ratnu, V. S., Schramm, M., Becher, I., et al. (2018). Systematic analysis of protein
turnover in primary cells. Nature Communications, 9(1):1–10.
McCord, R., Bolen, C. R., Koeppen, H., Kadel, E. E., Oestergaard, M. Z., Nielsen, T.,
Sehn, L. H., and Venstrom, J. M. (2019a). PD-L1 and tumor-associated macrophages
in de novo DLBCL. Blood Advances, 3(4):531–540.
McCord, R., Bolen, C. R., Koeppen, H., Kadel, E. E., Oestergaard, M. Z., Nielsen, T.,
Sehn, L. H., and Venstrom, J. M. (2019b). PD-L1 and tumor-associated macrophages
in de novo DLBCL. Blood Advances, 3(4):531–540.
180
McGlincy, N. J. and Ingolia, N. T. (2017). Transcriptome-wide measurement of translation
by ribosome profiling. Methods, 126:112–129.
McMahon, S. B. (2014). MYC and the control of apoptosis. Cold Spring Harbor perspectives
in medicine, 4(7):a014407.
Meyer, K. D., Patil, D. P., Zhou, J., Zinoviev, A., Skabkin, M. A., Elemento, O., Pestova,
T. V., Qian, S.-B., and Ja↵rey, S. R. (2015). 5 UTR m6A promotes cap-independent
translation. Cell, 163(4):999–1010.
Meyer, N. and Penn, L. Z. (2008). Reflecting on 25 years with MYC. Nature Reviews
Cancer, 8(12):976–990.
Mlynarczyk, C., Fonta´n, L., and Melnick, A. (2019). Germinal center-derived lymphomas:
The darkest side of humoral immunity. Immunological Reviews, 288(1):214–239.
Mo, J., Liang, H., Su, C., Li, P., Chen, J., and Zhang, B. (2021). DDX3X: structure,
physiologic functions and cancer. Molecular cancer, 20(1):1–20.
Modelska, A., Turro, E., Russell, R., Beaton, J., Sbarrato, T., Spriggs, K., Miller, J., Gra¨f,
S., Provenzano, E., Blows, F., et al. (2015). The malignant phenotype in breast cancer is
driven by eif4a1-mediated changes in the translational landscape. Cell death & disease,
6(1):e1603–e1603.
Mohr, I. (2016). Virology: Closing in on the causes of host shuto↵. eLife, 5:e20755.
Molliex, A., Temirov, J., Lee, J., Coughlin, M., Kanagaraj, A. P., Kim, H. J., Mittag, T.,
and Taylor, J. P. (2015). Phase separation by low complexity domains promotes stress
granule assembly and drives pathological fibrillization. Cell, 163(1):123–133.
Moreno-Mateos, M. A., Vejnar, C. E., Beaudoin, J.-D., Fernandez, J. P., Mis, E. K.,
Khokha, M. K., and Giraldez, A. J. (2015). CRISPRscan: designing highly ecient
sgRNAs for CRISPR-Cas9 targeting in vivo. Nature Methods, 12(10):982–988.
Morgan, M., Page`s, H., Obenchain, V., and Hayden, N. (2016). Rsamtools: Binary
alignment (BAM), FASTA, variant call (BCF), and tabix file import. R package version,
1(0):677–689.
Morrish, F. and Hockenbery, D. (2014). MYC and mitochondrial biogenesis. Cold Spring
Harbor perspectives in medicine, 4(5):a014225.
Morton, L. M., Wang, S. S., Devesa, S. S., Hartge, P., Weisenburger, D. D., and Linet,
M. S. (2006). Lymphoma incidence patterns by WHO subtype in the United States,
1992-2001. Blood, 107(1):265–276.
181
Moser, B. and Willimann, K. (2004). Chemokines: role in inflammation and immune
surveillance. Annals of the rheumatic diseases, 63(suppl 2):ii84–ii89.
Muramatsu, M., Kinoshita, K., Fagarasan, S., Yamada, S., Shinkai, Y., and Honjo,
T. (2000). Class switch recombination and hypermutation require activation-induced
cytidine deaminase (AID), a potential RNA editing enzyme. Cell, 102(5):553–563.
Nakagawa, R. and Calado, D. P. (2021). Positive selection in the light zone of germinal
centers. Frontiers in Immunology, 12:1053.
Nascimento, E. M., Cox, C. L., MacArthur, S., Hussain, S., Trotter, M., Blanco, S., Suraj,
M., Nichols, J., Ku¨bler, B., Benitah, S. A., et al. (2011). The opposing transcriptional
functions of Sin3a and c-Myc are required to maintain tissue homeostasis. Nature Cell
Biology, 13(12):1395–1405.
Navarro, A., Gaya, A., Martinez, A., Urbano-Ispizua, A., Pons, A., Balague´, O., Gel, B.,
Abrisqueta, P., Lopez-Guillermo, A., Artells, R., et al. (2008). MicroRNA expression
profiling in classic Hodgkin lymphoma. Blood, 111(5):2825–2832.
Neelagandan, N., Lamberti, I., Carvalho, H. J., Gobet, C., and Naef, F. (2020). What
determines eukaryotic translation elongation: recent molecular and quantitative analyses
of protein synthesis. Open biology, 10(12):200292.
Nesvizhskii, A. I. (2014). Proteogenomics: concepts, applications and computational
strategies. Nature Methods, 11(11):1114–1125.
Northcott, P. A., Buchhalter, I., Morrissy, A. S., Hovestadt, V., Weischenfeldt, J., Ehren-
berger, T., Gro¨bner, S., Segura-Wang, M., Zichner, T., Rudneva, V. A., et al. (2017).
The whole-genome landscape of medulloblastoma subtypes. Nature, 547(7663):311–317.
Obrig, T. G., Culp, W. J., McKeehan, W. L., and Hardesty, B. (1971). The mechanism
by which cycloheximide and related glutarimide antibiotics inhibit peptide synthesis on
reticulocyte ribosomes. Journal of Biological Chemistry, 246(1):174–181.
Oertlin, C., Lorent, J., Murie, C., Furic, L., Topisirovic, I., and Larsson, O. (2019).
Generally applicable transcriptome-wide analysis of translation using anota2seq. Nucleic
Acids Research, 47(12):e70–e70.
Oh, S., Flynn, R. A., Floor, S. N., Purzner, J., Martin, L., Do, B. T., Schubert, S.,
Vaka, D., Morrissy, S., Li, Y., et al. (2016). Medulloblastoma-associated DDX3 variant
selectively alters the translational response to stress. Oncotarget, 7(19):28169.
182
Ojha, J., Ayres, J., Secreto, C., Tschumper, R., Rabe, K., Van Dyke, D., Slager, S.,
Shanafelt, T., Fonseca, R., Kay, N. E., et al. (2015). Deep sequencing identifies genetic
heterogeneity and recurrent convergent evolution in chronic lymphocytic leukemia. Blood,
The Journal of the American Society of Hematology, 125(3):492–498.
Olexiouk, V., Van Criekinge, W., and Menschaert, G. (2018). An update on sORFs.
org: a repository of small orfs identified by ribosome profiling. Nucleic Acids Research,
46(D1):D497–D502.
Orr, M. W., Mao, Y., Storz, G., and Qian, S.-B. (2020). Alternative ORFs and small
ORFs: shedding light on the dark proteome. Nucleic Acids Research, 48(3):1029–1042.
Ouspenskaia, T., Law, T., Clauser, K. R., Klaeger, S., Sarkizova, S., Aguet, F., Li, B.,
Christian, E., Knisbacher, B. A., Le, P. M., Hartigan, C. R., Keshishian, H., Ap↵el, A.,
Oliveira, G., Zhang, W., Chow, Y. T., Ji, Z., Shukla, S. A., Bachireddy, P., Getz, G.,
Hacohen, N., Keskin, D. B., Carr, S. A., Wu, C. J., and Regev, A. (2020). Thousands
of novel unannotated proteins expand the MHC I immunopeptidome in cancer. bioRxiv.
Ozuah, N. W., Lubega, J., Allen, C. E., and El-Mallawany, N. K. (2020). Five decades of
low intensity and low survival: adapting intensified regimens to cure pediatric Burkitt
lymphoma in Africa. Blood Advances, 4(16):4007–4019.
Pakos-Zebrucka, K., Koryga, I., Mnich, K., Ljujic, M., Samali, A., and Gorman, A. M.
(2016). The integrated stress response. EMBO reports, 17(10):1374–1395.
Palade, G. E. (1955). A small particulate component of the cytoplasm. The Journal of
Cell Biology, 1(1):59–68.
Pardoll, D. M. (2012). The blockade of immune checkpoints in cancer immunotherapy.
Nature Reviews Cancer, 12(4):252–264.
Patmore, D. M., Jassim, A., Nathan, E., Gilbertson, R. J., Tahan, D., Ho↵mann, N., Tong,
Y., Smith, K. S., Kanneganti, T.-D., Suzuki, H., et al. (2020). DDX3X suppresses the
susceptibility of hindbrain lineages to medulloblastoma. Developmental Cell, 54(4):455–
470.
Patro, R., Duggal, G., Love, M. I., Irizarry, R. A., and Kingsford, C. (2017). Salmon
provides fast and bias-aware quantification of transcript expression. Nature methods,
14(4):417–419.
Pearson, H., Daouda, T., Granados, D. P., Durette, C., Bonneil, E., Courcelles, M.,
Rodenbrock, A., Laverdure, J.-P., Coˆte´, C., Mader, S., et al. (2016). MHC class I–
associated peptides derive from selective regions of the human genome. The Journal of
clinical investigation, 126(12):4690–4701.
183
Pearson, K. (1897). On a form of spurious correlation which may arise when indices are
used in the measurement of organs. In Royal Soc., London, Proc., volume 60, pages
489–502.
Pelletier, J., Thomas, G., and Volarevic´, S. (2018). Ribosome biogenesis in cancer: new
players and therapeutic avenues. Nature Reviews Cancer, 18(1):51–63.
Perdiga˜o, N., Heinrich, J., Stolte, C., Sabir, K. S., Buckley, M. J., Tabor, B., Signal, B.,
Gloss, B. S., Hammang, C. J., Rost, B., et al. (2015). Unexpected features of the dark
proteome. Proceedings of the National Academy of Sciences, 112(52):15898–15903.
Perez-Riverol, Y., Csordas, A., Bai, J., Bernal-Llinares, M., Hewapathirana, S., Kundu,
D. J., Inuganti, A., Griss, J., Mayer, G., Eisenacher, M., et al. (2019). The PRIDE
database and related tools and resources in 2019: improving support for quantification
data. Nucleic Acids Research, 47(D1):D442–D450.
Perez-Riverol, Y., Vizca´ıno, J. A., and Griss, J. (2018). Future prospects of spectral
clustering approaches in proteomics. Proteomics, 18(14):1700454.
Pfeifer, M., Grau, M., Lenze, D., Wenzel, S.-S., Wolf, A., Wollert-Wulf, B., Dietze, K.,
Nogai, H., Storek, B., Madle, H., et al. (2013). PTEN loss defines a PI3K/AKT pathway-
dependent germinal center subtype of di↵use large B-cell lymphoma. Proceedings of the
National Academy of Sciences, 110(30):12420–12425.
Pfister, A. S. (2019). Emerging role of the nucleolar stress response in autophagy. Frontiers
in Cellular Neuroscience, 13:156.
Phelan, J. D., Young, R. M., Webster, D. E., Roulland, S., Wright, G. W., Kasbekar, M.,
Sha↵er, A. L., Ceribelli, M., Wang, J. Q., Schmitz, R., et al. (2018). A multiprotein
supercomplex controlling oncogenic signalling in lymphoma. Nature, 560(7718):387–391.
Philippe, L., van den Elzen, A. M., Watson, M. J., and Thoreen, C. C. (2020). Global
analysis of LARP1 translation targets reveals tunable and dynamic features of 5TOP
motifs. Proceedings of the National Academy of Sciences, 117(10):5319–5328.
Phung, B., Cies´la, M., Sanna, A., Guzzi, N., Beneventi, G., Ngoc, P. C. T., Lauss,
M., Cabrita, R., Cordero, E., Bosch, A., et al. (2019). The X-linked DDX3X RNA
helicase dictates translation reprogramming and metastasis in melanoma. Cell Reports,
27(12):3573–3586.
Pianese, G. (1896). Beitrag zur histologie und aetiologie des carcinoms, volume 1. G.
Fischer.
184
Piccirillo, C. A., Bjur, E., Topisirovic, I., Sonenberg, N., and Larsson, O. (2014). Transla-
tional control of immune responses: from transcripts to translatomes. Nature Immunology,
15(6):503–511.
Pourdehnad, M., Truitt, M. L., Siddiqi, I. N., Ducker, G. S., Shokat, K. M., and Ruggero,
D. (2013). Myc and mTOR converge on a common node in protein synthesis control that
confers synthetic lethality in Myc-driven cancers. Proceedings of the National Academy
of Sciences, 110(29):11988–11993.
Presnyak, V., Alhusaini, N., Chen, Y.-H., Martin, S., Morris, N., Kline, N., Olson, S.,
Weinberg, D., Baker, K. E., Graveley, B. R., et al. (2015). Codon optimality is a major
determinant of mRNA stability. Cell, 160(6):1111–1124.
Protter, D. S. and Parker, R. (2016). Principles and properties of stress granules. Trends
in Cell Biology, 26(9):668–679.
Pugh, T. J., Weeraratne, S. D., Archer, T. C., Krummel, D. A. P., Auclair, D., Bochicchio,
J., Carneiro, M. O., Carter, S. L., Cibulskis, K., Erlich, R. L., et al. (2012). Medul-
loblastoma exome sequencing uncovers subtype-specific somatic mutations. Nature,
488(7409):106–110.
Pyronnet, S., Imataka, H., Gingras, A.-C., Fukunaga, R., Hunter, T., and Sonenberg, N.
(1999). Human eukaryotic translation initiation factor 4G (eIF4G) recruits mnk1 to
phosphorylate eIF4E. The EMBO journal, 18(1):270–279.
Rakhra, K., Bachireddy, P., Zabuawala, T., Zeiser, R., Xu, L., Kopelman, A., Fan, A. C.,
Yang, Q., Braunstein, L., Crosby, E., et al. (2010). CD4+ T cells contribute to the
remodeling of the microenvironment required for sustained tumor regression upon
oncogene inactivation. Cancer Cell, 18(5):485–498.
Ramı´rez, F., Ryan, D. P., Gru¨ning, B., Bhardwaj, V., Kilpert, F., Richter, A. S., Heyne,
S., Du¨ndar, F., and Manke, T. (2016). deeptools2: a next generation web server for
deep-sequencing data analysis. Nucleic Acids Research, 44(W1):W160–W165.
Ratje, A. H., Loerke, J., Mikolajka, A., Bru¨nner, M., Hildebrand, P. W., Starosta, A. L.,
Do¨nho¨fer, A., Connell, S. R., Fucini, P., Mielke, T., et al. (2010). Head swivel on the
ribosome facilitates translocation by means of intra-subunit tRNA hybrid sites. Nature,
468(7324):713–716.
Raught, B. and Gingras, A.-C. (1999). eIF4E activity is regulated at multiple levels. The
international journal of biochemistry & cell biology, 31(1):43–57.
185
Rauschendorf, M.-A., Zimmer, J., Hanstein, R., Dickemann, C., and Vogt, P. (2011).
Complex transcriptional control of the AZFa gene DDX3Y in human testis. International
Journal of Andrology, 34(1):84–96.
Reddy, A., Zhang, J., Davis, N. S., Mott, A. B., Love, C. L., Waldrop, A., Leppa, S.,
Pasanen, A., Meriranta, L., Karjalainen-Lindsberg, M.-L., et al. (2017). Genetic and
functional drivers of di↵use large b cell lymphoma. Cell, 171(2):481–494.
Richter, J., Schlesner, M., Ho↵mann, S., Kreuz, M., Leich, E., Burkhardt, B., Rosolowski,
M., Ammerpohl, O., Wagener, R., Bernhart, S. H., et al. (2012). Recurrent mutation
of the ID3 gene in Burkitt lymphoma identified by integrated genome, exome and
transcriptome sequencing. Nature Genetics, 44(12):1316.
Richter, J. D. and Coller, J. (2015). Pausing on polyribosomes: make way for elongation
in translational control. Cell, 163(2):292–300.
Robichaud, N., Hsu, B. E., Istomine, R., Alvarez, F., Blagih, J., Ma, E. H., Morales, S. V.,
Dai, D. L., Li, G., Souleimanova, M., et al. (2018). Translational control in the tumor
microenvironment promotes lung metastasis: Phosphorylation of eif4e in neutrophils.
Proceedings of the National Academy of Sciences, 115(10):E2202–E2209.
Robinson, G., Parker, M., Kranenburg, T. A., Lu, C., Chen, X., Ding, L., Phoenix, T. N.,
Hedlund, E., Wei, L., Zhu, X., et al. (2012). Novel mutations target distinct subgroups
of medulloblastoma. Nature, 488(7409):43–48.
Robinson, M. D., McCarthy, D. J., and Smyth, G. K. (2010). edgeR: a Bioconductor
package for di↵erential expression analysis of digital gene expression data. Bioinformatics,
26(1):139–140.
Rogozin, I. B., Kochetov, A. V., Kondrashov, F. A., Koonin, E. V., and Milanesi, L. (2001).
Presence of ATG triplets in 5 untranslated regions of eukaryotic cDNAs correlates with
a ‘weak’context of the start codon. Bioinformatics, 17(10):890–900.
Rolfe, D. and Brown, G. C. (1997). Cellular energy utilization and molecular origin of
standard metabolic rate in mammals. Physiological Reviews, 77(3):731–758.
Ron, D. and Walter, P. (2007). Signal integration in the endoplasmic reticulum unfolded
protein response. Nature Reviews Molecular cell biology, 8(7):519–529.
Rooijers, K., Loayza-Puch, F., Nijtmans, L. G., and Agami, R. (2013). Ribosome profiling
reveals features of normal and disease-associated mitochondrial translation. Nature
Communications, 4(1):1–8.
186
Rouschop, K. M., Van Den Beucken, T., Dubois, L., Niessen, H., Bussink, J., Savelkouls,
K., Keulers, T., Mujcic, H., Landuyt, W., Voncken, J. W., et al. (2010). The unfolded
protein response protects human tumor cells during hypoxia through regulation of
the autophagy genes MAP1LC3B and ATG5. The Journal of Clinical Investigation,
120(1):127–141.
Ruggero, D. (2013). Translational control in cancer etiology. Cold Spring Harbor Perspec-
tives in Biology, 5(2):a012336.
Ruggero, D., Montanaro, L., Ma, L., Xu, W., Londei, P., Cordon-Cardo, C., and Pandolfi,
P. P. (2004). The translation factor eIF-4E promotes tumor formation and cooperates
with c-Myc in lymphomagenesis. Nature medicine, 10(5):484–486.
Ruggiano, A., Foresti, O., and Carvalho, P. (2014). Er-associated degradation: Protein
quality control and beyond. Journal of Cell Biology, 204(6):869–879.
Ruiz-Orera, J., Messeguer, X., Subirana, J. A., and Alba, M. M. (2014). Long non-coding
RNAs as a source of new peptides. eLife, 3:e03523.
Runte, F., Renner IV, P., and Hoppe, M. (2019). Kuby immunology.
Sabi, R. and Tuller, T. (2014). Modelling the Eciency of Codon–tRNA Interactions
Based on Codon Usage Bias. DNA Research, 21(5):511–526.
Sadedin, S. P., Pope, B., and Oshlack, A. (2012). Bpipe: a tool for running and managing
bioinformatics pipelines. Bioinformatics, 28(11):1525–1526.
Samir, P., Kesavardhana, S., Patmore, D. M., Gingras, S., Malireddi, R. S., Karki, R.,
Guy, C. S., Briard, B., Place, D. E., Bhattacharya, A., et al. (2019). DDX3X acts as
a live-or-die checkpoint in stressed cells by regulating NLRP3 inflammasome. Nature,
573(7775):590–594.
Sander, S., Calado, D. P., Srinivasan, L., Ko¨chert, K., Zhang, B., Rosolowski, M., Rodig,
S. J., Holzmann, K., Stilgenbauer, S., Siebert, R., et al. (2012). Synergy between PI3K
signaling and MYC in Burkitt lymphomagenesis. Cancer Cell, 22(2):167–179.
Sanson, K. R., Hanna, R. E., Hegde, M., Donovan, K. F., Strand, C., Sullender, M. E.,
Vaimberg, E. W., Goodale, A., Root, D. E., Piccioni, F., et al. (2018). Optimized libraries
for CRISPR-Cas9 genetic screens with multiple modalities. Nature Communications,
9(1):1–15.
Santos, D. A., Shi, L., Tu, B. P., and Weissman, J. S. (2019). Cycloheximide can
distort measurements of mRNA levels and translation eciency. Nucleic acids research,
47(10):4974–4985.
187
Sanz, E., Yang, L., Su, T., Morris, D. R., McKnight, G. S., and Amieux, P. S. (2009). Cell-
type-specific isolation of ribosome-associated mRNA from complex tissues. Proceedings
of the National Academy of Sciences, 106(33):13939–13944.
Sarkizova, S., Klaeger, S., Le, P. M., Li, L. W., Oliveira, G., Keshishian, H., Hartigan,
C. R., Zhang, W., Braun, D. A., Ligon, K. L., et al. (2020). A large peptidome dataset
improves HLA class I epitope prediction across most of the human population. Nature
Biotechnology, 38(2):199–209.
Saxton, R. A. and Sabatini, D. M. (2017). mTOR Signaling in Growth, Metabolism, and
Disease. Cell, 168(6):960–976.
Schatz, J. H., Oricchio, E., Wolfe, A. L., Jiang, M., Linkov, I., Maragulia, J., Shi, W.,
Zhang, Z., Rajasekhar, V. K., Pagano, N. C., et al. (2011). Targeting cap-dependent
translation blocks converging survival signals by AKT and PIM kinases in lymphoma.
Journal of Experimental Medicine, 208(9):1799–1807.
Schmidt, E. V. (2004). The role of c-myc in regulation of translation initiation. Oncogene,
23(18):3217–3221.
Schmitz, R., Wright, G. W., Huang, D. W., Johnson, C. A., Phelan, J. D., Wang, J. Q.,
Roulland, S., Kasbekar, M., Young, R. M., Sha↵er, A. L., et al. (2018). Genetics
and pathogenesis of di↵use large b-cell lymphoma. New England Journal of Medicine,
378(15):1396–1407.
Schmitz, R., Young, R. M., Ceribelli, M., Jhavar, S., Xiao, W., Zhang, M., Wright, G.,
Sha↵er, A. L., Hodson, D. J., Buras, E., et al. (2012). Burkitt lymphoma pathogenesis
and therapeutic targets from structural and functional genomics. Nature, 490(7418):116–
120.
Schneider-Poetsch, T., Ju, J., Eyler, D. E., Dang, Y., Bhat, S., Merrick, W. C., Green,
R., Shen, B., and Liu, J. O. (2010). Inhibition of eukaryotic translation elongation by
cycloheximide and lactimidomycin. Nature Chemical Biology, 6(3):209–217.
Schueren, F. and Thoms, S. (2016). Functional translational readthrough: a systems
biology perspective. PLoS Genetics, 12(8):e1006196.
Schuller, A. P. and Green, R. (2018). Roadblocks and resolutions in eukaryotic translation.
Nature Reviews Molecular Cell Biology, 19(8):526–541.
Schwanha¨usser, B., Busse, D., Li, N., Dittmar, G., Schuchhardt, J., Wolf, J., Chen, W.,
and Selbach, M. (2011). Global quantification of mammalian gene expression control.
Nature, 473(7347):337–342.
188
Sen, N. D., Zhou, F., Ingolia, N. T., and Hinnebusch, A. G. (2015). Genome-wide analysis
of translational eciency reveals distinct but overlapping functions of yeast DEAD-box
RNA helicases Ded1 and eIF4A. Genome Research, 25(8):1196–1205.
Sendoel, A., Dunn, J. G., Rodriguez, E. H., Naik, S., Gomez, N. C., Hurwitz, B., Levorse,
J., Dill, B. D., Schramek, D., Molina, H., et al. (2017). Translation from unconventional
5 start sites drives tumour initiation. Nature, 541(7638):494–499.
Sha, C., Barrans, S., Cucco, F., Bentley, M. A., Care, M. A., Cummin, T., Kennedy, H.,
Thompson, J. S., Uddin, R., Worrillow, L., et al. (2019). Molecular high-grade b-cell
lymphoma: defining a poor-risk group that requires di↵erent approaches to therapy.
Journal of Clinical Oncology, 37(3):202.
Sha↵er III, A. L., Young, R. M., and Staudt, L. M. (2012). Pathogenesis of human b cell
lymphomas. Annual review of immunology, 30:565–610.
Sharma, P., Nilges, B. S., Wu, J., and Leidel, S. A. (2019). The translation inhibitor
cycloheximide a↵ects ribosome profiling data in a species-specific manner. bioRxiv, page
746255.
Sharpe, A. H. and Pauken, K. E. (2018). The diverse functions of the PD1 inhibitory
pathway. Nature Reviews Immunology, 18(3):153–167.
Shatsky, I. N., Terenin, I. M., Smirnova, V. V., and Andreev, D. E. (2018). Cap-independent
translation: What’s in a name? Trends in Biochemical Sciences, 43(11):882–895.
Shi, Z. and Barna, M. (2015). Translating the genome in time and space: specialized
ribosomes, RNA regulons, and RNA-binding proteins. Annual Review of Cell and
Developmental Biology, 31:31–54.
Shi, Z., Fujii, K., Kovary, K. M., Genuth, N. R., Ro¨st, H. L., Teruel, M. N., and Barna, M.
(2017). Heterogeneous ribosomes preferentially translate distinct subpools of mRNAs
genome-wide. Molecular cell, 67(1):71–83.
Shih, J., Tsai, T., Chao, C.-H., and Lee, Y. W. (2008). Candidate tumor suppressor DDX3
RNA helicase specifically represses cap-dependent translation by acting as an eIF4E
inhibitory protein. Oncogene, 27(5):700–714.
Shiue, C. N., Berkson, R. G., and Wright, A. P. (2009). c-Myc induces changes in higher
order rDNA structure on stimulation of quiescent cells. Oncogene, 28(16):1833–1842.
Silvera, D., Formenti, S. C., and Schneider, R. J. (2010). Translational control in cancer.
Nature Reviews Cancer, 10(4):254–266.
189
Skourti-Stathaki, K. and Proudfoot, N. J. (2014). A double-edged sword: R loops as threats
to genome integrity and powerful regulators of gene expression. Genes & Development,
28(13):1384–1396.
Smith, A., Crouch, S., Lax, S., Li, J., Painter, D., Howell, D., Patmore, R., Jack, A., and
Roman, E. (2015). Lymphoma incidence, survival and prevalence 2004-2014: Sub-type
analyses from the UK’s Haematological Malignancy Research Network. British Journal
of Cancer, 112(9):1575–1584.
Smith, L. M. and Kelleher, N. L. (2013). Proteoform: a single term describing protein
complexity. Nature methods, 10(3):186–187.
Smith, R. C., Kanellos, G., Vlahov, N., Alexandrou, C., Willis, A. E., Knight, J. R., and
Sansom, O. J. (2021). Translation initiation in cancer at a glance. Journal of Cell
Science, 134(1).
Sole, C., Larrea, E., Manterola, L., Goicoechea, I., Armesto, M., Arestin, M., M Ca↵arel,
M., M Araujo, A., Fernandez-Mercado, M., Araiz, M., et al. (2016). Aberrant expression
of microRNAs in B-cell lymphomas. Microrna, 5(2):87–105.
Sonenberg, N. (1996). mRNA 5’cap-binding protein eIF4E and control of cell growth.
Translational Control.
Sonenberg, N. and Hinnebusch, A. G. (2009). Regulation of translation initiation in
eukaryotes: mechanisms and biological targets. Cell, 136(4):731–745.
Song, S., Cao, C., Choukrallah, M.-A., Tang, F., Christofori, G., Kohler, H., Wu, F., Fodor,
B. D., Frederiksen, M., Willis, S. N., et al. (2021). OBF1 and Oct factors control the
germinal center transcriptional program. Blood, The Journal of the American Society
of Hematology, 137(21):2920–2934.
Soto-Rifo, R., Rubilar, P. S., Limousin, T., De Breyne, S., Decimo, D., and Ohlmann,
T. (2012). DEAD-box protein DDX3 associates with eif4f to promote translation of
selected mRNAs. The EMBO journal, 31(18):3745–3756.
Staege, M. S., Lee, S. P., Frisan, T., Mautner, J., Scholz, S., Pajic, A., Rickinson, A. B.,
Masucci, M. G., Polack, A., and Bornkamm, G. W. (2002). MYC overexpression imposes
a nonimmunogenic phenotype on Epstein–Barr virus-infected B cells. Proceedings of the
National Academy of Sciences, 99(7):4550–4555.
Statello, L., Guo, C.-J., Chen, L.-L., and Huarte, M. (2021). Gene regulation by long
non-coding RNAs and its biological functions. Nature Reviews Molecular Cell Biology,
22(2):96–118.
190
Steinhardt, J. J., Peroutka, R. J., Mazan-Mamczarz, K., Chen, Q., Houng, S., Robles,
C., Barth, R. N., DuBose, J., Bruns, B., Tesoriero, R., et al. (2014). Inhibiting
CARD11 translation during BCR activation by targeting the eIF4A RNA helicase.
Blood, 124(25):3758–3767.
Stransky, N., Eglo↵, A. M., Tward, A. D., Kostic, A. D., Cibulskis, K., Sivachenko,
A., Kryukov, G. V., Lawrence, M. S., Sougnez, C., McKenna, A., et al. (2011). The
mutational landscape of head and neck squamous cell carcinoma. Science, 333(6046):1157–
1160.
Stults, D. M., Killen, M. W., Williamson, E. P., Hourigan, J. S., Vargas, H. D., Arnold,
S. M., Moscow, J. A., and Pierce, A. J. (2009). Human rRNA gene clusters are
recombinational hotspots in cancer. Cancer Research, 69(23):9096–9104.
Subramanian, A., Tamayo, P., Mootha, V. K., Mukherjee, S., Ebert, B. L., Gillette,
M. A., Paulovich, A., Pomeroy, S. L., Golub, T. R., Lander, E. S., et al. (2005). Gene
set enrichment analysis: a knowledge-based approach for interpreting genome-wide
expression profiles. Proceedings of the National Academy of Sciences, 102(43):15545–
15550.
Suresh, S., Chen, B., Zhu, J., Golden, R. J., Lu, C., Evers, B. M., Novaresi, N., Smith, B.,
Zhan, X., Schmid, V., et al. (2020). eIF5B drives integrated stress response-dependent
translation of PD-L1 in lung cancer. Nature Cancer, 1(5):533–545.
Swerdlow, S. H., Campo, E., Pileri, S. A., Harris, N. L., Stein, H., Siebert, R., Advani, R.,
Ghielmini, M., Salles, G. A., Zelenetz, A. D., et al. (2016). The 2016 revision of the world
health organization classification of lymphoid neoplasms. Blood, 127(20):2375–2390.
Takahashi, K., Hu, B., Wang, F., Yan, Y., Kim, E., Vitale, C., Patel, K. P., Strati,
P., Gumbs, C., Little, L., et al. (2018). Clinical implications of cancer gene muta-
tions in patients with chronic lymphocytic leukemia treated with lenalidomide. Blood,
131(16):1820–1832.
Tate, J. G., Bamford, S., Jubb, H. C., Sondka, Z., Beare, D. M., Bindal, N., Boutselakis,
H., Cole, C. G., Creatore, C., Dawson, E., et al. (2019). COSMIC: the catalogue of
somatic mutations in cancer. Nucleic Acids Research, 47(D1):D941–D947.
Taylor, J., Yeomans, A. M., and Packham, G. (2020). Targeted inhibition of mRNA
translation initiation factors as a novel therapeutic strategy for mature B-cell neoplasms.
Exploration of targeted anti-tumor therapy, 1:3.
191
Terenin, I. M., Andreev, D. E., Dmitriev, S. E., and Shatsky, I. N. (2013). A novel
mechanism of eukaryotic translation initiation that is neither m7G-cap-, nor IRES-
dependent. Nucleic Acids Research, 41(3):1807–1816.
Thoreen, C. C., Chantranupong, L., Keys, H. R., Wang, T., Gray, N. S., and Sabatini,
D. M. (2012). A unifying model for mTORC1-mediated regulation of mRNA translation.
Nature, 485(7396):109–113.
Thyme, S. B., Akhmetova, L., Montague, T. G., Valen, E., and Schier, A. F. (2016). Internal
guide RNA interactions interfere with Cas9-mediated cleavage. Nature Communications,
7(1):1–7.
Tompa, P., Davey, N. E., Gibson, T. J., and Babu, M. M. (2014). A million peptide motifs
for the molecular biologist. Molecular Cell, 55(2):161–169.
Torrent, M., Chalancon, G., de Groot, N. S., Wuster, A., and Madan Babu, M. (2018).
Cells alter their tRNA abundance to selectively regulate protein synthesis during stress
conditions. Science Signaling, 11(546).
Truitt, M. L., Conn, C. S., Shi, Z., Pang, X., Tokuyasu, T., Coady, A. M., Seo, Y.,
Barna, M., and Ruggero, D. (2015). Di↵erential requirements for eIF4E dose in normal
development and cancer. Cell, 162(1):59–71.
Tschochner, H. and Hurt, E. (2003). Pre-ribosomes on the road from the nucleolus to the
cytoplasm. Trends in Cell Biology, 13(5):255–263.
Tuller, T., Carmi, A., Vestsigian, K., Navon, S., Dorfan, Y., Zaborske, J., Pan, T., Dahan,
O., Furman, I., and Pilpel, Y. (2010). An evolutionarily conserved mechanism for
controlling the eciency of protein translation. Cell, 141(2):344–354.
Twa, D. D., Chan, F. C., Ben-Neriah, S., Woolcock, B. W., Mottok, A., Tan, K. L.,
Slack, G. W., Gunawardana, J., Lim, R. S., McPherson, A. W., et al. (2014). Genomic
rearrangements involving programmed death ligands are recurrent in primary mediastinal
large b-cell lymphoma. Blood, 123(13):2062–2065.
Unterluggauer, J. J., Prochazka, K., Tomazic, P. V., Huber, H. J., Seeboeck, R., Fechter,
K., Steinbauer, E., Gruber, V., Feichtinger, J., Pichler, M., et al. (2018). Expression
profile of translation initiation factor eIF2B5 in di↵use large B-cell lymphoma and its
correlation to clinical outcome. Blood Cancer Journal, 8(9):1–5.
Urra, H., Dufey, E., Avril, T., Chevet, E., and Hetz, C. (2016). Endoplasmic reticulum
stress and the hallmarks of cancer. Trends in Cancer, 2(5):252–262.
192
Valentin-Vega, Y. A., Wang, Y.-D., Parker, M., Patmore, D. M., Kanagaraj, A., Moore,
J., Rusch, M., Finkelstein, D., Ellison, D. W., Gilbertson, R. J., et al. (2016). Cancer-
associated DDX3X mutations drive stress granule assembly and impair global translation.
Scientific reports, 6(1):1–16.
Van der Auwera, G. A. and O’Connor, B. D. (2020). Genomics in the Cloud: Using
Docker, GATK, and WDL in Terra. O’Reilly Media.
van Heesch, S., Witte, F., Schneider-Lunitz, V., Schulz, J. F., Adami, E., Faber, A. B.,
Kirchner, M., Maatz, H., Blachut, S., Sandmann, C.-L., et al. (2019). The translational
landscape of the human heart. Cell, 178(1):242–260.
Van Riggelen, J., Yetil, A., and Felsher, D. W. (2010). MYC as a regulator of ribosome
biogenesis and protein synthesis. Nature Reviews Cancer, 10(4):301–309.
Van Steeg, H., Van Oostrom, C. T., Hodemaekers, H. M., Peters, L., and Thomas,
A. A. (1991). The translation in vitro of rat ornithine decarboxylase mRNA is blocked
by its 5 untranslated region in a polyamine-independent way. Biochemical Journal,
274(2):521–526.
Vattem, K. M. and Wek, R. C. (2004). Reinitiation involving upstream ORFs regulates
ATF4 mRNA translation in mammalian cells. Proceedings of the National Academy of
Sciences, 101(31):11269–11274.
Venkataramanan, S., Calviello, L., Wilkins, K., and Floor, S. N. (2020). DDX3X and
DDX3Y are redundant in protein synthesis. Biorxiv.
Versteeg, R., Noordermeer, I. A., Kru¨se-Wolters, M., Ruiter, D. J., and Schrier, P. I.
(1988). c-myc down-regulates class I HLA expression in human melanomas. The EMBO
journal, 7(4):1023–1029.
Walter, P. and Ron, D. (2011). The unfolded protein response: from stress pathway to
homeostatic regulation. Science, 334(6059):1081–1086.
Wang, X., Zhao, B. S., Roundtree, I. A., Lu, Z., Han, D., Ma, H., Weng, X., Chen, K.,
Shi, H., and He, C. (2015). N6-methyladenosine modulates messenger RNA translation
eciency. Cell, 161(6):1388–1399.
Wei, J., Kishton, R. J., Angel, M., Conn, C. S., Dalla-Venezia, N., Marcel, V., Vincent,
A., Catez, F., Ferre´, S., Ayadi, L., et al. (2019). Ribosomal proteins regulate MHC class
I peptide generation for immunosurveillance. Molecular cell, 73(6):1162–1173.
193
Weinberg, D. E., Shah, P., Eichhorn, S. W., Hussmann, J. A., Plotkin, J. B., and Bartel,
D. P. (2016). Improved ribosome-footprint and mRNA measurements provide insights
into dynamics and regulation of yeast translation. Cell Reports, 14(7):1787–1799.
Weinstein, J. N., Collisson, E. A., Mills, G. B., Shaw, K. R. M., Ozenberger, B. A., Ellrott,
K., Shmulevich, I., Sander, C., and Stuart, J. M. (2013). The cancer genome atlas
pan-cancer analysis project. Nature Genetics, 45(10):1113–1120.
Wendel, H.-G., De Stanchina, E., Fridman, J. S., Malina, A., Ray, S., Kogan, S., Cordon-
Cardo, C., Pelletier, J., and Lowe, S. W. (2004). Survival signalling by Akt and eIF4E
in oncogenesis and cancer therapy. Nature, 428(6980):332–337.
Wethmar, K., Barbosa-Silva, A., Andrade-Navarro, M. A., and Leutz, A. (2014). uORFdb
— a comprehensive literature database on eukaryotic uorf biology. Nucleic Acids Research,
42(D1):D60–D67.
Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New
York.
Wilmore, S., Rogers-Broadway, K.-R., Taylor, J., Lemm, E., Fell, R., Stevenson, F. K.,
Forconi, F., Steele, A. J., Coldwell, M., Packham, G., et al. (2021). Targeted inhibition
of eIF4A suppresses b-cell receptor-induced translation and expression of MYC and
MCL1 in chronic lymphocytic leukemia cells. Cellular and Molecular Life Sciences,
pages 1–13.
Wolfe, A. L., Singh, K., Zhong, Y., Drewe, P., Rajasekhar, V. K., Sanghvi, V. R., Mavrakis,
K. J., Jiang, M., Roderick, J. E., Van der Meulen, J., et al. (2014a). Rna g-quadruplexes
cause eif4a-dependent oncogene translation in cancer. Nature, 513(7516):65–70.
Wolfe, A. L., Singh, K., Zhong, Y., Drewe, P., Rajasekhar, V. K., Sanghvi, V. R., Mavrakis,
K. J., Jiang, M., Roderick, J. E., Van der Meulen, J., Schatz, J. H., Rodrigo, C. M., Zhao,
C., Rondou, P., de Stanchina, E., Teruya-Feldstein, J., Kelliher, M. A., Speleman, F.,
Porco, J. A., Pelletier, J., Ra¨tsch, G., and Wendel, H. G. (2014b). RNA G-quadruplexes
cause eIF4A-dependent oncogene translation in cancer. Nature, 513(7516):65–70.
Wolin, S. L. and Walter, P. (1988). Ribosome pausing and stacking during translation of
a eukaryotic mRNA. The EMBO journal, 7(11):3559–3569.
Xiang, N., He, M., Ishaq, M., Gao, Y., Song, F., Guo, L., Ma, L., Sun, G., Liu, D., Guo,
D., et al. (2016). The DEAD-box RNA helicase DDX3 interacts with nf-b subunit p65
and suppresses p65-mediated transcription. PloS one, 11(10):e0164471.
194
Xiao, Z., Huang, R., Xing, X., Chen, Y., Deng, H., and Yang, X. (2018). De novo
annotation and characterization of the translatome with ribosome profiling data. Nucleic
Acids Research, 46(10):e61–e61.
Xiao, Z., Zou, Q., Liu, Y., and Yang, X. (2016). Genome-wide assessment of di↵erential
translations with ribosome profiling data. Nature Communications, 7(1):1–11.
Xu, H., Xiao, T., Chen, C.-H., Li, W., Meyer, C. A., Wu, Q., Wu, D., Cong, L., Zhang,
F., Liu, J. S., et al. (2015). Sequence determinants of improved CRISPR sgRNA design.
Genome Research, 25(8):1147–1157.
Xu, Y., Poggio, M., Jin, H. Y., Shi, Z., Forester, C. M., Wang, Y., Stumpf, C. R., Xue, L.,
Devericks, E., So, L., et al. (2019). Translation control of the immune checkpoint in
cancer and its therapeutic targeting. Nature Medicine, 25(2):301–311.
Xu, Y. and Ruggero, D. (2020). The role of translation control in tumorigenesis and its
therapeutic implications. Annual Review of Cancer Biology, 4:437–457.
Yang, H.-S., Jansen, A. P., Komar, A. A., Zheng, X., Merrick, W. C., Costes, S., Lockett,
S. J., Sonenberg, N., and Colburn, N. H. (2003). The transformation suppressor Pdcd4
is a novel eukaryotic translation initiation factor 4A binding protein that inhibits
translation. Molecular and Cellular Biology, 23(1):26–37.
Yang, X., Zhong, W., and Cao, R. (2020). Phosphorylation of the mRNA cap-binding
protein eIF4E and cancer. Cellular Signalling, page 109689.
Yang, Y., Fan, X., Mao, M., Song, X., Wu, P., Zhang, Y., Jin, Y., Yang, Y., Chen,
L.-L., Wang, Y., et al. (2017). Extensive translation of circular RNAs driven by N
6-methyladenosine. Cell Research, 27(5):626–641.
Yang, Y. and Wang, Z. (2019). IRES-mediated cap-independent translation, a path leading
to hidden proteome. Journal of Molecular Cell Biology, 11(10):911–919.
Yeomans, A., Thirdborough, S. M., Valle-Argos, B., Linley, A., Krysov, S., Hidalgo, M. S.,
Leonard, E., Ishfaq, M., Wagner, S. D., Willis, A. E., et al. (2016). Engagement of the
b-cell receptor of chronic lymphocytic leukemia cells drives global and MYC-specific
mRNA translation. Blood, 127(4):449–457.
Yewdell, J. W., Anto´n, L. C., and Bennink, J. R. (1996). Defective ribosomal products
(DRiPs): a major source of antigenic peptides for MHC class I molecules? The Journal
of Immunology, 157(5):1823–1826.
195
Yu, G., Wang, L.-G., Han, Y., and He, Q.-Y. (2012). clusterprofiler: an R package for
comparing biological themes among gene clusters. Omics: a journal of integrative biology,
16(5):284–287.
Zhai, W. and Comai, L. (2000). Repression of RNA polymerase i transcription by the
tumor suppressor p53. Molecular and Cellular Biology, 20(16):5930.
Zhang, H., Dou, S., He, F., Luo, J., Wei, L., and Lu, J. (2018a). Genome-wide maps of
ribosomal occupancy provide insights into adaptive evolution and regulatory roles of
uORFs during Drosophila development. PLoS Biology, 16(7):e2003903.
Zhang, H., Wang, Y., and Lu, J. (2019). Function and evolution of upstream ORFs in
eukaryotes. Trends in Biochemical Sciences, 44(9):782–794.
Zhang, H., Wang, Y., Wu, X., Tang, X., Wu, C., and Lu, J. (2021). Determinants of
genome-wide distribution and evolution of uORFs in eukaryotes. Nature Communications,
12(1):1–17.
Zhang, J., Medeiros, L. J., and Young, K. H. (2018b). Cancer immunotherapy in di↵use
large b-cell lymphoma. Frontiers in oncology, 8:351.
Zhang, T., Li, N., Sun, C., Jin, Y., and Sheng, X. (2020). MYC and the unfolded protein
response in cancer: synthetic lethal partners in crime? EMBO Molecular Medicine,
12(5):e11845.
Zhao, J.-J., Lin, J., Lwin, T., Yang, H., Guo, J., Kong, W., Dessureault, S., Moscinski,
L. C., Rezania, D., Dalton, W. S., et al. (2010). microRNA expression profile and
identification of miR-29 as a prognostic marker and pathogenetic factor by targeting
CDK6 in mantle cell lymphoma. Blood, 115(13):2630–2639.
Zhou, P., Blain, A. E., Newman, A. M., Zaka, M., Chagaluka, G., Adlar, F. R., O↵or,
U. T., Broadbent, C., Chaytor, L., Whitehead, A., et al. (2019). Sporadic and endemic
Burkitt lymphoma have frequent FOXO1 mutations but distinct hotspots in the AKT
recognition motif. Blood Advances, 3(14):2118–2127.
Zuberek, J., Wyslouch-Cieszynska, A., Niedzwiecka, A., Dadlez, M., Stepinski, J., Au-
gustyniak, W., Gingras, A.-C., Zhang, Z., Burley, S. K., Sonenberg, N., et al. (2003).
Phosphorylation of eIF4E attenuates its interaction with mRNA 5 cap analogs by elec-
trostatic repulsion: intein-mediated protein ligation strategy to obtain phosphorylated
protein. RNA, 9(1):52–61.
196