In vitro and in silico methods to investigate and
overcome drug resistance in cancer
Chiara Facciotto
Research Program in Systems Oncology (ONCOSYS)
Research Programs Unit
Biochemistry and Developmental Biology
Medicum
Faculty of Medicine
University of Helsinki
Finland
Academic dissertation
To be publicly discussed, with the permission of
the Faculty of Medicine of the University of Helsinki,
in Biomedicum Helsinki 1, Lecture Hall 2, Haartmaninkatu 8, Helsinki,
on 13 August 2021, at 12 o’clock noon.
Helsinki 2021
Supervisor
Sampsa Hautaniemi, DTech, Professor
Research Program in Systems Oncology
Research Program Unit
Biochemistry and Developmental Biology
Medicum
Faculty of Medicine, University of Helsinki
Helsinki, Finland
Rainer Lehtonen, PhD, Docent
Research Program in Applied Tumor Genomics
Research Program Unit
Department of Medical and Clinical Genetics
Faculty of Medicine, University of Helsinki
Helsinki, Finland
Reviewers appointed by the Faculty
Kirsi Granberg, PhD, University Lecturer
Faculty of Medicine and Health Technology, Tampere University
Tampere, Finland
Tapio Pahikkala, PhD, Associate Professor
Department of Future Technologies, University of Turku
Turku, Finland
Opponent appointed by the Faculty
Marieke Kuijjer, PhD, Group Leader
Computational Biology and Systems Medicine group
Centre for Molecular Medicine Norway (NCMM), University of Oslo
Oslo, Norway
The Faulty of Medicine uses the Urkund system (plagiarism recognition) to examine
all doctoral dissertations.
Cover illustration created with Biorender.com.
Series No. 41/2021
ISSN 2342-3161 (print)
ISSN 2342-317X (online)
ISBN 978-951-51-7426-0 (print)
ISBN 978-951-51-7427-7 (online)
http://ethesis.helsinki.fi
Unigrafia Oy
Helsinki 2021
To my family and friends, who supported me every step of the way.

Success is not final, failure is not fatal:
it is the courage to continue that counts.
Winston S. Churchill

Abstract
Overcoming drug resistance in cancer is one of the most pressing issues in oncology. The
last century saw a dramatic increase in the discovery of new cancer therapies, so much so
that chemotherapeutic agents and immunotherapies are now, alone or in combination, the
backbone of treatment for many cancers. Despite the increased rate of treatment success
brought by these regimens, cancer patients can become resistant to these drugs. This leads
to disease relapse, hindering patient survival. Drug resistance remains the primary cause of
death in most advanced-stage cancer patients.
The molecular mechanisms responsible for the development of a resistance phenotype in
cancer cells are complex and include both genetic and epigenetic alterations. Since drug
resistance is a multifactorial phenomenon, we used a systems biology approach to investigate
it on different fronts. Specifically, we developed a high-throughput drug screening method to
test new drug combinations, identifying epigenetic inhibitors able to sensitize lymphoma
cells to doxorubicin. We also implemented a bioinformatic pipeline which combines multiple
omics data to identify genes and pathways driving platinum response across multiple cancers.
We then developed a method to compute differential methylation between cancer samples
with varying and unknown tumor purity, which we used to investigate DNA methylation
changes linked to drug resistance in ovarian cancer and lymphoma. Finally, we created a
workflow management system to build complex bioinformatic pipelines and aid researchers
in the analysis of high-throughput biomedical data.
By combining laboratory biology experiments and computational analyses, we gained a
broader understanding of the cellular mechanisms behind immunochemotherapy failure.
Moreover, we were able to identify novel biomarkers associated with platinum response in
multiple cancers, as well as new drug combinations able to overcome immunochemotherapy
resistance in lymphoma cells. The in vitro and in silico methods presented in this thesis can
not only assist researchers in the cancer field, but are broadly applicable to other fields of
biomedical research. Overall, this work is an important stepping stone in both understanding
and overcoming drug resistance in cancer, and has great potential to improve outcomes for
cancer patients in the future.
Sommario
Il superamento delle resistenze alle terapie nei malati di cancro è uno dei problemi più urgenti
in oncologia. Durante lo scorso secolo, il numero delle terapie contro il cancro disponibili é
aumentato considerevolmente, tanto che gli agenti chemioterapici e le immunoterapie sono
ora, da soli o in combinazione, alla base del trattamento di molti tumori. Nonostante questi
regimi terapeutici abbiano un maggiore tasso di successo rispetto all’uso di agenti singoli,
i malati di cancro possono comunque sviluppare una resistenza a queste associazioni di
farmaci. Questo porta alla recidiva della malattia, ostacolando la sopravvivenza del paziente.
La resistenza ai farmaci rimane la principale causa di morte nella maggior parte dei pazienti
oncologici in stadio avanzato.
Complessi meccanismi molecolari, che comprendono alterazioni sia genetiche che epige-
netiche, sono responsabili dello sviluppo delle resistenze alle terapie nelle cellule tumorali.
Poiché la resistenza ai farmaci è un fenomeno multifattoriale, abbiamo utilizzato un ap-
proccio di biologia dei sistemi per indagarlo su diversi fronti. Nello specifico, abbiamo
sviluppato un metodo di high-throughput screening per testare nuove combinazioni di far-
maci, identificando inibitori epigenetici in grado di sensibilizzare le cellule di linfoma alla
doxorubicina. Abbiamo anche implementato una pipeline bioinformatica che combina dati
omici per identificare geni e pathway molecolari che determinano la risposta al platino in
diversi tipi di cancro. Abbiamo quindi sviluppato un metodo per stimare la differenza dello
stato di metilazione del DNA tra campioni di cancro con purezza del tumore variabile e
sconosciuta. Abbiamo poi usato questo metodo per studiare i cambiamenti della metilazione
del DNA legati alla resistenza ai farmaci nel tumore alle ovaie e nel linfoma. Infine, abbiamo
creato un workflow management system per costruire pipeline bioinformatiche complesse e
aiutare i ricercatori nell’analisi di larghe quantitá di dati biomedici.
Combinando esperimenti biologici di laboratorio e analisi computazionali, abbiamo ac-
quisito una più ampia comprensione dei meccanismi cellulari alla base del fallimento
dell’immunochemioterapia. Inoltre, siamo stati in grado di identificare nuovi biomarca-
tori associati alla risposta al platino in diversi tipi di tumori, nonché nuove combinazioni di
farmaci in grado di superare la resistenza all’immunochemioterapia nelle cellule di linfoma.
I metodi in vitro e in silico presentati in questa tesi possono non solo aiutare i ricercatori
nel campo oncologico, ma sono applicabili anche in altri campi della ricerca biomedica.
Nel complesso, questo lavoro è un importante trampolino di lancio sia nella comprensione
che nel superamento delle resistenze ai farmaci antitumorali, così da permettere ai pazienti
oncologici un outcome e una qualità di vita migliori.
Contents
Abbreviations xi
Publications and author’s contributions xiv
1 Introduction 1
2 Cancer 3
2.1 Biology of cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Regulation of gene expression and cellular identity . . . . . . . . . 3
DNA methylation . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Histone modification . . . . . . . . . . . . . . . . . . . . . . . . . 4
RNA interference . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.2 When things go wrong - Cancer initiation and progression . . . . . 6
Gene copy number alterations . . . . . . . . . . . . . . . . . . . . 8
Dysregulation of the epigenome . . . . . . . . . . . . . . . . . . . 8
2.1.3 Cancer heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Diffuse large B-cell lymphoma . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 High-grade serous ovarian cancer . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Cancer therapies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4.1 Treating DLBCL . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.2 Treating HGSOC . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Drug resistance in cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Predicting and overcoming drug resistance in cancer . . . . . . . . . . . . 16
2.6.1 Identification of biomarkers . . . . . . . . . . . . . . . . . . . . . 16
2.6.2 Restoring drug sensitivity via epigenetic reprogramming . . . . . . 16
3 High-throughput biology 19
3.1 High-throughput screenings . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.1 Analyzing dose response data . . . . . . . . . . . . . . . . . . . . 20
3.2 Next-generation sequencing technologies . . . . . . . . . . . . . . . . . . 21
3.2.1 RNA sequencing . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.2 Bisulfite sequencing . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.3 Analyzing sequencing data . . . . . . . . . . . . . . . . . . . . . . 23
3.2.4 Public repositories of sequencing data . . . . . . . . . . . . . . . . 26
3.3 Bioinformatic workflow management systems to analyze high-throughput
data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4 Aims of the study 29
5 Materials and methods 30
5.1 Materials, samples and data . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.2 Drug sensitivity screening . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.3 Anduril workflow management system . . . . . . . . . . . . . . . . . . . . 32
5.4 Differential methylation using maximum-likelihood (DMML)
method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.5 Interactive exploration of multiple data types . . . . . . . . . . . . . . . . 33
5.6 Automatic workflow to integrate multi-omics TCGA data and identify pre-
dictive biomarkers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.7 Survival analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6 Results 35
6.1 Pretreating DLBCL cell lines with epigenetic inhibitors sensitizes them to
immunochemotherapy (I) . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2 Molecular mechanisms responsible for epigenetic sensitization in DLBCL (I) 36
6.3 Identification of potential biomarkers and therapeutic targets to predict and
overcome platinum resistance in multiple cancers (II) . . . . . . . . . . . . 36
6.4 Estimating differential methylation in cancer samples with varying tumor
purity (III) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.5 Deconvoluted DNA methylation from 53 HGSOC samples (Unpublished data) 38
6.6 Anduril 2 workflow framework (IV) . . . . . . . . . . . . . . . . . . . . . 38
7 Discussion 41
Acknowledgements 44
Bibliography 47
Abbreviations
ABC Activated B-cell
aCGH Array comparative genomic hybridization
Anduril ANalysis of Data Using Rapid Integration of aLgorithms
ANXA9 Annexin A9
ARC Activity Regulated Cytoskeleton Associated Protein
ARID1A AT-Rich Interaction Domain 1A
Asc HGSOC sample from ascites
ATP Adenosine triphosphate
Bi-Seq Bisulfite sequencing
BLCA Bladder urothelial carcinoma
BRCA Breast invasive carcinoma
BRCA1 Breast And Ovarian Cancer Susceptibility Protein 1
BRCA2 Breast And Ovarian Cancer Susceptibility Protein 2
BRD Bromodomain
C19orf33 Chromosome 19 Open Reading Frame 33
CAR Chimeric antigen receptor
Cs Cytosine nucleotides
cDNA Complementary DNA
CD20 Cluster of differentiation 20
CESC Cervical squamous cell carcinoma and endocervical adenocarci-
noma
CGI CpG island
CNA Copy number alteration
COAD Colon adenocarcinoma
CREBBP Histone Lysine Acetyltransferase CREB Binding Protein
CpG CG dinucleotide (i.e. a cytosine nucleotide followed by a guanine
nucleotide in 5’→ 3’ direction)
DEG Differentially expressed gene
DLBCL Diffuse large B-cell lymphoma
DMG Differential methylated gene
DMML Differential methylation using a maximum-likelihood
DMP Differential methylated promoter
DMS Differential methylated CpG site
DNA Deoxyribonucleic acid
DNAmeth DNA methylation
DNMT DNA methyltransferase
DSS Dispersion shrinkage for sequencing data, a software to estimate
differential methylation
DSS Drug sensitivity score
ECI2 Enoyl-CoA Delta Isomerase 2
ENCODE Encyclopedia of DNA Elements
EOC Epithelial ovarian cancer
ESCA Esophageal carcinoma
EZH2 Enhancer Of Zeste 2 Polycomb Repressive Complex 2 Subunit
xi
FBXO17 F-Box Protein 17
FDA US Federal Drug Administration
GBM Glioblastoma multiforme
GCB Germinal center B-cell
GDC Genomic data commons
H3K27me3 Trimethylation of the 27th lysine residue of a histone H3 protein
H3K4me3 Trimethylation of the 4th lysine residue of a histone H3 protein
H3K27ac Acetylation of the 27th lysine residue of a histone H3 protein
HAT Histone acetyltransferase
HDAC Histone deacetylase
HDACi Histone deacetylase inhibitor
HDM Histone demethylase
HGSOC High grade serous ovarian cancer
HMT Histone methyltrasferase
HNSC Head and neck squamous cell carcinoma
HR Homologous recombination
HRD Homologous recombination deficiency
HTS High-throughput screening
IC50 Half maximal inhibitory concentration of a compound or drug
LGALS3BP Galectin 3 Binding Protein
LUAD Lung adenocarcinoma
LUSC Lung squamous cell carcinoma
mAb Monoclonal antibody
Mes HGSOC sample from intestine
MESO Mesothelioma
MLF1 Myeloid Leukemia Factor 1
MOABS Model-based Analysis of Bisulfite Sequencing data
mRNA Messenger RNA
miRNA Micro RNA
MRPS21 Mitochondrial Ribosomal Protein S21
MS4A1 Membrane-Spanning 4-Domains, Subfamily A, Member 1
NGS Next generation sequencing technologies
NHEJ Non-homologous end joining
NHL Non-Hodgkin lymphoma
Ome HGSOC sample from omentum
OS Overall survival
OV Ovarian serous cystadenocarcinoma
Ova HGSOC sample from ovary
OvCa Ovarian cancer
PAAD Pancreatic adenocarcinoma
PCR Polymerase chain reaction
Per HGSOC sample from peritoneum
PFS Progression free survival
QC Quality control
R-CHOP Rituximab, cyclophosphamide, doxorubicin, vincristine, and pred-
nisone
READ Rectum adenocarcinoma
xii
RISC RNA-induced silencing complex
RNA Ribonucleic acid
RNA-Seq RNA sequencing
RNAi RNA interference
RRBS Reduced representation bisulfite sequencing
scRNA-Seq Single-cell RNA sequencing
SEC62 SEC62 Homolog, Preprotein Translocation Factor
siRNA Small interfering RNA
SIRT1 Sirtuin 1
SNP Single-nucleotide polymorphism
SOX17 SRY-Box Transcription Factor 17
STAD Stomach adenocarcinoma
STAR Spliced Transcripts Alignment to a Reference
Ts Thymine nucleotides
TBS Targeted bisulfite sequencing
TCGA The Cancer Genome Atlas
TF Transcription factor
TGCT Testicular germ cell tumors
TGS Third generation sequencing technologies
TM4SF1 Transmembrane 4 L Six Family Member 1
TP53 Tumor Protein P53
TSPYL5 Testis-Specific Y-Encoded-Like Protein 5
Tub HGSOC sample from Fallopian tubes
Tum HGSOC sample from unknown location
Us Uracil nucleotides
UCS Uterine carcinosarcoma
UCEC Uterine corpus endometrial carcinoma
WES Whole exome sequencing
WGBS Whole genome bisulfite sequencing
WHO World Health Organization
WMS Workflow management system
xiii
Publications
Publication I Chiara Facciotto*, Julia Casado*, Laura Turunen, Suvi-Katri Leivonen,
Manuela Tumiati, Ville Rantanen, Liisa Kauppi, Rainer Lehtonen, Sirpa
Leppä, Krister Wennerberg, Sampsa Hautaniemi.
Drug screening approach combines epigenetic sensitization with immunochemother-
apy in cancer.
Clinical Epigenetics 11, 192 (2019)
Publication II Chiara Facciotto, Laura Lehtinen, Antti Häkkinen, Julia Casado, Ville
Rantanen, Jaana Oikkonen, Veli-Matti Isoviita, Jani Saarela, Rainer Lehtonen,
Sampsa Hautaniemi.
Identification of novel platinum-resistance biomarkers and therapeutic targets
through integration of multi-omics data from 1,500 cancer patients.
Submitted
Publication III Antti Häkkinen, Amjad Alkodsi, Chiara Facciotto, Kaiyang Zhang, Katja
Kaipio, Sirpa Leppä, Olli Carpén, Seija Grénman, Johanna Hynninen, Sakari
Hietanen, Rainer Lehtonen, Sampsa Hautaniemi.
Identifying differentially methylated sites in samples with varying tumor
purity.
Bioinformatics 34, 18 (2018)
Publication IV Alejandra Cervera, Ville Rantanen, Kristian Ovaska, Marko Laakso, Javier
Nuñez-Fontarnau, Amjad Alkodsi, Julia Casado, Chiara Facciotto, Antti
Häkkinen, Riku Louhimo, Sirkku Karinen, Kaiyang Zhang, Kari Lavikka,
Lauri Lyly, Maninder Pal Singh, Sampsa Hautaniemi.
Anduril 2: upgraded large-scale data integration framework.
Bioinformatics 35, 19 (2019)
* equal contribution
In addition to the publications, unpublished results on DNA methylation in ovarian
cancer are presented in the thesis.
Publications included in other theses
Publication I was included in the thesis of Julia Casado (Proteogenomics methods for
translational cancer research, Helsinki 2021)
xiv
Author’s contributions
Publication I Conceptualized the study, designed and optimized the drug screening proto-
col, implemented the drug screening experiments, built a pipeline to analyze
the results, made the figures and wrote the manuscript (all together with JC).
Processed RNA-Seq data, built the results explorer website.
Publication II Contributed to study conceptualization, designed the bioinformatic pipeline,
integrated TCGA data, performed survival analysis, co-designed and analyzed
the RNAi experiment, made the figures and wrote the manuscript.
Publication III Processed the ovarian cancer DNA methylation data used to test the method.
Contributed to the manuscript.
Publication IV Contributed to the Anduril 2 project by designing, implementing and testing
components, mainly related to the analysis of bisulfite sequencing and drug
screening data. Contributed to the manuscript.
Thesis Contribution
Processed and analyzed the ovarian cancer DNA methylation data.
xv

1 INTRODUCTION
1 Introduction
Cancer is a group of over 100 different diseases, all characterized by the abnormal
growth of cells which, due to mutations or other molecular alterations, have become
neoplastic [1]. The World Health Organization (WHO) reported that worldwide,
18.1 million new cases of cancer were diagnosed and an estimated 9.6 million
people died from cancer in 2018 [2]. Moreover, WHO also estimated that in 2010
the total annual economic cost of cancer at the global level was around 1.16 trillion
US$ [3].
With the advent of modern medicine, the 19th and 20th centuries saw both an
increase in cancer cases (due to improved overall health and an aging population)
and also a boom in the development of new cancer therapies. However, the plastic
nature of cancer allows the disease to develop resistance to treatments such as
chemotherapies and immunotherapies. Hence, overcoming drug resistance in
cancer to improve patient survival, reduce side effects, and decrease the cost of
unsuccessful treatments is one of the major health challenges of the 21st century.
The molecular mechanisms at the root of drug resistance in cancer are complex
and multivariate, often involving genetic and epigenetic aberrations [4]. Hence, a
multidisciplinary approach is needed to better understand therapy failure, develop
strategies to overcome it, and identify predictive biomarkers to avoid administering
the wrong treatments to cancer patients. Thanks to modern high-throughput
technologies, we are able to generate an unprecedented amount of molecular data,
which represent a great opportunity but also present a major challenge. Correctly
analyzing and interpreting such data to separate meaningful information from noise
constitutes a major bottleneck in advancing our understanding of the molecular
mechanisms behind cancer and other diseases [5]. Hence, tailored computational
tools and pipelines combining multiple data layers are needed to advance our
understanding of the field.
Diffuse large B-cell lymphoma (DLBCL) and high-grade serous ovarian cancer
(HGSOC) are among the most aggressive forms of lymphomas and ovarian cancers,
respectively. Even though standard treatments exist for these cancers, a significant
proportion of patients still develop drug resistance [6, 7]. In this thesis, we
used a systems biology approach to combine laboratory biology experiments
and computational analyses to gain a more comprehensive understanding of the
cellular mechanisms behind immunochemotherapy resistance. In Publication I, we
identified epigenetic inhibitors able to reprogram DLBCL cells, sensitizing them
to doxorubicin and rituximab. In Publication II, we studied the pan-cancer genetic
and epigenetic landscape of patients treated with platinum-based chemotherapy
1
1 INTRODUCTION
and, through a multi-omics integration analysis, we discovered novel biomarkers
associated with platinum response. In Publication III, we developed a computational
method to estimate differential methylation between cancer samples with varying
tumor purities and tested it on DLBCL and HGSOC samples. In Publication
IV, we present a bioinformatic workflow management system to implement easily
reproducible computational pipelines, which was used to systematize and parallelize
all data analyses carried out in Publications I and II.
This thesis proceeds with a literature review of the molecular mechanisms behind
cancer and its resistance to standard therapies, with particular focus on DLBCL
and HGSOC. Additionally, I present an overview of high throughput technologies
used in Publications I-IV, including drug screenings, RNA sequencing and bisulfite
sequencing, followed by a summary of the main methods and results developed to
investigate drug resistance.
2
2 CANCER
2 Cancer
In this section, I explore the molecular mechanisms responsible for the transfor-
mation of healthy cells into cancerous ones, as well as current cancer treatment
options, their limitations and ways to overcome drug resistance in cancer.
2.1 Biology of cancer
Human bodies are made of hundreds of different types of cells, and anomalies in
any of them can lead to different cancers. The biggest risk factor for cancer is
aging. Over half of cancers manifest in people older than 70 [8], because the older
a person is, the higher the chances are that their DNA has accumulated random
mutations that can turn a healthy cell into a cancerous one. Other risk factors
include exposure to UV light, radiation, dangerous chemicals, certain viruses (such
as the human papillomavirus), an unhealthy diet, and smoking: all factors that can
induce mutations in our genome [9].
Every person develops cancer cells throughout their life, but the immune system is
generally very good at detecting and eliminating them. However, when cancer cells
are able to escape immune detection and start to replicate, they can form malignant
masses and spread across the body. In order to uncover the cascade of molecular
changes that lead to cells becoming cancerous, it is important to first understand
how healthy cells develop and maintain their own identity.
2.1.1 Regulation of gene expression and cellular identity
All cells from the same individual contain the same DNA sequence, the blueprint
of the human body. Genes represent the portion of the information encoded in
our genome that can be transcribed into RNA. A subset of genes, accounting for
less than 2% of the human genome, are transcribed into messenger RNA (mRNA)
and then further translated into proteins, the major building blocks of our cells.
However, only a portion of genes are expressed in each cell: this is how the human
body can generate different cell types that carry out very specialized functions and
adapt to environmental stimuli.
A variety of intra- and extra-cellular signals work in concert to regulate which
portion of the genome is transcribed, hence defining the cell’s identity. Some
of these mechanisms can activate or repress the transcription of a gene. These
include the binding of transcription factors (TFs) to regulatory regions [10], as well
as epigenetic mechanisms (such as DNA methylation and histone modifications)
3
2 CANCER
which affect the chromatin structure [11, 12] and regulate which positions of the
genome can be accessed by activating TFs, which in turn initiate transcription. Other
mechanisms, such as RNA interference (RNAi) [13], can fine tune the expression
level of a gene after it has been transcribed.
DNA methylation
Cytosines in the DNA can become methylated when a methyl group (-CH3) binds
to the C5 position of the nucleotide, turning it into a 5-methylcytosine. Although all
cytosines have the potential to acquire a methyl group, in humans DNA methylation
generally occurs at CpG sites, i.e. at cytosines that are followed by a guanine
(Figure 1A). Specific enzymes, known as DNA methyltransferases (DNMTs), are
responsible for adding the methyl group to the cystosine substrate [14]. Specifically,
the DNMT3a and DNMT3b enzymes are responsible for de novo methylation,
while DNMT1 maintains existing DNA methylation patterns during cell replication.
DNA methylation plays a different regulatory role depending on the genomic
region where it is located [15]. Transposable elements, centromeres and other
repetitive regions are usually methylated in order to ensure genome stability. There
are also regions in the genome with a high concentration of CpG sites, which
are known as CpG islands (CGIs) [16]. CGIs are generally unmethylated and
approximately 60-70% of promoter regions in the human genome contain a CGI.
Methylated promoters are usually associated with silencing of the related gene,
while methylation in the body of a gene correlates with increased expression. DNA
methylation can also mediate whether or not transcription factors or other proteins
bind to the DNA [17, 18] and has been linked to regulation of RNA splicing [19].
Histone modification
Histones are proteins involved in the packaging of the DNA into chromatin. DNA
strands wrap around histones, forming structures called nucleosomes [20]. There are
five type of histones (H1, H2A, H2B, H3 and H4) and the N-tail of these proteins can
acquire post-translational chemical modifications such as methylation, acetylation,
phosphorylation, and ubiquitination (Figure 1B). These chemical groups are added
and removed by specific enzymes [21]. For instance, histone methyltransferases
(HMTs) are responsible for adding one or more methyl groups to histone tails, while
histone demethylases (HDMs) remove such chemical groups. In the same way,
histone acetyltransferases (HATs) and deacetylases (HDACs) are responsible for
adding and removing acetyl groups. There are also enzymes that bind to histones
when they recognize specific histone marks. For instance, chromodomain proteins
bind to histones when a methyl mark is present, while bromodomain proteins
(BRDs) recognize acetyl marks.
4
2 CANCER
Figure 1: Cellular mechanisms of gene expression regulation. Caption continues in
the next page. Image created with BioRender.com
5
2 CANCER
Figure 1: (A) DNMT enzymes add methyl groups to the DNA at CpG sites. (B)
Amino acids, such as lysines (Ks), forming the tails of histone proteins can be post-
translationally modified via the addition of chemical groups. HMT and HAT enzymes
add methyl and acetyl groups, respectively. Other enzymes like BRD can bind to
histones, recognizing specific modifications. HDM and HDAC enzymes can remove
methyl and acetyl groups from histone tails. (C) miRNAs are transcribed from DNA
and, together with a set of proteins, form the RISC complex. miRNAs can then bind to
mRNA molecules, which are then degraded by the RISC complex.
Histone modifications can alter the chromatin structure and play a key role in
regulating gene expression. For instance, the trimethylation of the 27th lysine
residue of a histone H3 protein (H3K27me3) is associated with gene silencing when
it occurs in nucleosomes located in a gene promoter, while the trimethylation of the
4th lysine residue of a histone H3 protein (H3K4me3) in the same region correlates
with actively transcribed genes [22]. Acetylation of the 27th lysine residue of a
histone H3 protein (H3K27ac) alters the chromatin structure to make DNA more
accessible to TFs and has also been associated with active enhancers [23].
RNA interference
RNAi is a mechanism of post-transcriptional gene regulation driven by two classes
of small non-coding RNA species: microRNAs (miRNAs) or small interfering
RNAs (siRNAs) [13]. MiRNAs and siRNAs are antisense RNAs, which means that
they are complementary to specific mRNA molecules. These small RNA molecules
can recruit specific enzymes to form the RNA-induced silencing complex (RISC)
and guide it to the small RNA’s complementary mRNA molecule, which is then
cleaved and degraded, thus decreasing its expression level (Figure 1C). Even though
mRNA degradation is the primary mechanism through which miRNAs affect gene
expression, protein production can also be regulated via translational repression
[24]. In this case, miRNAs interact with one of the complex responsible for
mRNAmaturation, preventing the mRNAmolecule from reaching the conformation
required for translation to take place.
Today, artificially designed siRNA molecules can be used to downregulate the
expression of target genes in vitro via RNAi [25]. This allows us to observe how a
cell’s behavior can change with or without a certain protein.
2.1.2 When things go wrong - Cancer initiation and progression
When the mechanisms described above work correctly, healthy cells are in equilib-
rium with each other and the organism thrives. However, alterations in our DNA
can disrupt this equilibrium, causing cells to acquire neoplastic characteristics. The
6
2 CANCER
Figure 2: Dysregulation of gene expression though DNA methylation of cytosines in
the promoter region or alterations in the number of copies of a gene. An increase in
gene expression is generally observed when the promoter of the gene is unmethylated
and/or the portion of the DNA encoding the gene has been amplified (i.e., more than
two copies are present). Conversely, a methylated promoter and/or the deletion of one
or both copies of the gene would lead to a decreased expression of the gene.
term neoplasm was first introduced in the 19th century and comes from the Greek
words neos or "new" and plasma or "formation". Today "neoplasm" indicates a
group of cells displaying abnormal and excessive growth, uncoordinated with the
surrounding tissues. The immune system is generally able to detect neoplastic cells,
but if such cells escape immune detection they continue to replicate and eventually
form a mass, called a tumor. If the growth remains localized, the tumor is considered
benign. However, if the neoplastic cells start to spread to other parts of the body,
damaging other tissues, the tumor is then classified as malignant. Malignant tumors,
also known as cancers, can be fatal for the individual, especially if left untreated.
Every cell type in our body has the potential to become cancerous, and cancers
originating from different tissues may undergo a different progression and require
different treatment strategies. Despite these differences, cancer cells share a set of
mechanisms, known as the hallmarks of cancer [26], which allow them to replicate,
proliferate, and spread around the body.
The majority of cancers (90-95%) are caused by somatic mutations, but a small por-
tion of cancers (5-10%) can be inherited due to germline mutations, i.e. mutations
in a sperm cell or an egg cell, which can be passed from one generation to the next
[27]. The longer cancer cells survive, the more their genomes acquire alterations,
such as point mutations and/or amplifications and deletions of entire chromosomal
sections. This can lead to a chain reaction that affects the mechanisms regulating
gene expression, further advancing tumor progression. Alterations in the number
of copies of a gene, as well as the disruption of epigenetic patterns, are among the
main aberrations that can lead to transcriptional changes, as shown in Figure 2.
7
2 CANCER
Gene copy number alterations
Healthy cells posses two copies of each gene, one inherited from each parent.
However, cancer cells are characterized by genomic instability [28], which is
the accumulation of structural aberrations in the genome. These aberrations can
be as small as the mutation of a single nucleotide [29, 30], but can also affect
broader portions of the DNA via amplifications and deletions of gene-containing
chromosomal regions, or even entire chromosomes. Alterations in the number of
copies of a gene affect its expression [28], leading to changes that can promote tumor
progression. Moreover, copy number alterations (CNAs) of certain genes have also
been associated with resistance to certain cancer drugs. For instance, ERBB2 [31]
and CCNE1 [32] amplifications have been linked to platinum resistance.
Dysregulation of the epigenome
Healthy cells can flexibly adapt to different stimuli by altering their epigenomes
[33]. However, dysregulation of epigenetic patterns is a common feature of cancer
cells, which can alter cells’ proliferation rate as well as gene expression. Epigenetic
changes can even decrease the differentiation state of cancer cells, allowing them to
acquire stemness properties that in turn make them more plastic, increasing their
ability to develop drug resistance mechanisms.
Disruption of DNA methylation patterns is known to be a major feature of cancer
cells, manifesting as a global hypomethylation of the genome, but also as local-
ized hypermethylation of CpG islands and promoter regions [34]. Global DNA
hypomethylation leads to genomic instability, while hypermethylation of gene
promoters might silence tumor suppressor genes. For instance, the silencing of the
BRCA1 gene through promoter methylation is a known mechanism associated to
breast and ovarian cancer [35].
Patterns of histone modifications are also altered in cancer, leading to abnormal
chromatin conformation and to the activation or inactivation of certain genes
[36]. Examples of epigenetic dysregulation affecting histone modifications include
mutations of the HMT gene EZH2 in some forms of lymphoma [37], which leads to
altered patterns of H3K27me3, and overexpression of the HDAC SIRT1 in prostate
cancer [38] which affects histone acetylation.
Nucleosome repositioning [39] and altered networks of non-coding RNAs [40] are
other epigenetic mechanisms commonly disrupted in cancer.
2.1.3 Cancer heterogeneity
Due to the chaotic nature of tumor progression, a tumor mass is a very hetero-
geneous collection of cells (Figure 3). This heterogeneity appears on two levels:
8
2 CANCER
Figure 3: A tumor mass is a heterogeneous collection of cells embedded in extracellular
matrix. Since cancer cells proliferate quickly and accumulate structural aberrations,
different subclones of cancer cells are present in the same tumor. At the same time,
the tumor tissue is infiltrated by other cell types, including endothelial cells forming
blood vessels, immune cells, and cancer-associated fibroblasts. Cells communicate by
releasing extracellular vescicles and other cell signaling molecules. Image created with
BioRender.com
clonal heterogeneity and tumor microenvironment composition.
Clonal heterogeneity occurs when cell replication leads to the accumulation of
genetic and epigenetic changes that create diverging groups of cancer cells, known
as subclones [41, 42]. Factors driving this type of heterogeneity are, for instance,
hypoxia and inflammation.
The other level of heterogeneity is derived from the different cell types that are part
of the tumor microenvironment [43]. Cancer cells can hijack the components of
these other cell types to promote their own growth and survival. For instance, a
tumor requires blood vessels that can deliver nutrients to its cells, so endothelial
cells are recruited during neoangiogenesis (i.e. the new growth of blood vessels)
[44]. Moreover, immune cells are also able to infiltrate the tumor tissue and, by
releasing specific chemicals, can support cancer growth [45]. Cancer-associated
fibroblasts are another important part of the tumor microenvironment [46], where
they provide a structural framework for the neoplasm and synthesize extracellular
matrix, which in turn supports the transport of chemical signals, extracellular
9
2 CANCER
vesicles and nutrients across different cells.
2.2 Diffuse large B-cell lymphoma
Lymphomas are the most common type of blood cancer, specifically affecting a type
of white blood cell known as a lymphocyte. Lymphocytes are the main component
of our immune system. They travel through the blood and the lymphatic system to
defend the body by identifying and neutralizing pathogens, like bacteria and viruses,
clearing the organism of toxic and allergenic substances, and eliminating neoplastic
cells [47]. In particular, B cells are a specific subset of lymphocytes responsible for
the production of antibodies, the proteins that are able to detect both microbes and
unhealthy cells, and hence are fundamental to fighting infections. However, like
every other human cell, B cells can also become cancerous, spreading through the
body and growing in the lymph nodes, bone marrow, spleen, and other organs [48].
Lymphomas are generally classified as either Hodgkin lymphoma or non-Hodgkin
lymphoma (NHL). The latter group includes about 90% of all lymphomas [49].
Worldwide, in 2018 alone 509,590 new NHL cases were diagnosed and 248,724
patients died due to NHL [50]. Diffuse large B-cell lymphomas (DLBCLs) are the
most common type of NHL, accounting for 30-40% of NHL cases [51]. DLBCLs
can be classified into two major subtypes based on the development stage of the
B cells from which they originate: the germinal centre B-cell like (GCB) subtype,
usually with better prognosis, and the activated B-cell like (ABC) subtype [51].
Most DLBCLs belong to one of these two subtypes, but about 10-15% of cases are
unclassifiable.
Even though it can manifest at early age, the occurrence of DLBCL increases with
age and the median age at diagnosis is 70 [52]. The incidence is also slightly higher
in men [52].
Perturbations of the epigenetic landscape are strongly linked to DLBCL [53]. One
example is the dysregulation of histone methylation patterns in DLBCL due to
mutations of the EZH2 gene, common in the GCB subtype [37].
2.3 High-grade serous ovarian cancer
Ovarian cancer is one of the most common and the most deadly type of gynecologic
cancer, encompassing a collection of neoplasms originating from the ovaries or
the Fallopian tubes. In 2018, 295,414 new ovarian cancers were diagnosed, and
184,799 women died of this disease worldwide[50]. About 90% of ovarian cancers
have an epithelial origin, while the remaining 10% originate either from germ or
10
2 CANCER
stromal cells and are less invasive. A small portion (~3%) of epithelial ovarian
cancers (EOC) are mucinous, while about 20% of all EOCs are associated with
endometriosis and are classified as either clear cell or endometrioid carcinomas [54].
The remaining EOC cases originate from epithelial cells of either the Fallopian
tubes or the surface of the ovaries [55], and are divided into high-grade serous
ovarian cancers (HGSOCs) and low-grade serous ovarian cancers.
HGSOC is the most common and aggressive form of EOC (~70% of cases). One of
the causes of HGSOC is hereditary breast-ovarian cancer syndrome, an autosomal
dominant genetic disorder caused by specific genetic mutations in the BRCA1
and BRCA2 genes. BRCA1 and BRCA2 are tumor suppressor genes involved in
DNA repair pathways, including homologous recombination. Mutations in the
TP53 gene are also present in almost all HGSOC [56]. Chromosomal instability,
homologous recombination deficiency (HRD), aneuploidy and somatic mutations
are other typical features of HGSOC.
HGSOCs originate in the ovaries or Fallopian tubes (stage I) and spread first to other
pelvic organs such as the uterus (stage II), and then gradually to organs or tissues
within the peritoneal cavity (stage III) or even beyond, reaching the lungs and
extra-abdominal lymph nodes (stage IV) [57]. The high mortality rate of HGSOC
is due to the unspecific nature of the symptoms at the early stages of the disease, as
well as the lack of accurate screening methods [58]. In fact, over 85% of HGSOC
diagnoses occur at stage III or IV [59] and, by this time, the tumor is already highly
heterogeneous, which increases the probability of developing a treatment-resistant
subclone within the tumor mass.
2.4 Cancer therapies
Several treatment strategies are currently available to treat cancer, including surgery,
chemotherapy, and radiotherapy, which are the most common and established
methods. More recent therapies include immunotherapy and targeted therapy.
These treatments can be used alone or in combination, depending on the type of
cancer, its stage and location, and the particular medical history of each patient.
In this thesis, I focus mainly on cancers where chemotherapy is used in combination
with surgery or with immunotherapy. Hence, more information on these three
treatment options can be found below.
Surgery. Surgery is the oldest treatment option for solid cancers [60] and docu-
ments reporting the surgical removal of tumors date back to ancient Egypt. However,
it is only since the middle of the 19th century, when anaesthesia and antisepsis
were introduced, that cancer surgery has become a widespread option to treat solid
11
2 CANCER
malignancies. Not all tumors are operable, but when they are the surgical removal
of the mass before cancer cells spread to other organs leads to a better prognosis
for the patient. In the beginning of the 20th century, surgeons started performing
increasingly radical surgeries, removing large amounts of tissue, but they realized
that when the cancer has already spread to other organs, surgery alone does not
improve patient survival. Since the emergence of other treatment options, surgery
has mostly been used in combination with other therapies.
Chemotherapy. Chemotherapy was first introduced around the middle of the
20th century [61], and since then a plethora of chemotherapeutic agents have been
approved for the treatment of different malignancies. Chemotherapy drugs are
chemical compounds that target fast-growing cells by exploiting certain mecha-
nisms, such as halting the cell replication cycle or damaging DNA. Since cancer
cells are often unable to repair damaged DNA, this triggers a chain reaction leading
to apoptosis.
Multiple chemotherapy agents targeting different molecular mechanisms can be
combined into standard chemotherapy regimens, which specify the drugs to use,
their dosage, and the frequency at which they should be administered to treat a
specific cancer.
Compared to surgery, which is localized to a certain part of the body, chemotherapy
is generally administered intravenously or orally, which means that the drug is
transported throughout the whole body via our circulatory system. This makes
it a more effective option for non-solid cancers or when the tumor has already
metastasized. However, since chemotherapy does not target cancer cells specifically,
this type of treatment is also damaging for healthy tissues and it generally causes
more side effects than other treatment options.
Immunotherapy. Cancer immunotherapy is a branch of oncology that develops a
broad range of therapeutic strategies that, rather than aiming to directly kill cancer
cells, aims to make them recognizable to the immune system of the patient, which
in turn will fight the cancer. Immunotherapy exploits the fact that neoplastic cells
present specific antigens on their cell membranes.
A common and widespread strategy for passive immunotherapy employs artificial
monoclonal antibodies (mAb), designed to recognize tumor antigens and recruit
immune cells by mimicking natural antibodies [62]. Other types of passive
immunotherapies include checkpoint inhibitors [63] and cytokine therapy [64].
Active immunotherapy options, like cancer vaccines [65] and the use of CAR
T-cells [66], have also been developed in recent years.
12
2 CANCER
2.4.1 Treating DLBCL
The standard of care for DLBCL patients consists of the immunochemotherapy
regimen R-CHOP, which combines the monoclonal antibody rituximab with four
chemotherapeutic agents (cyclophosphamide, doxorubicin, vincristine, and pred-
nisone) [67]. Though most patients respond well to R-CHOP, about 30-40% relapse
or continue to have refractory disease, and resistance to R-CHOP is still the most
common cause of mortality among DLBCL patients [6].
Rituximab. Rituximab is a chimeric mouse/human mAb targeting the CD20
protein. CD20, encoded in theMS4A1 gene, is expressed on the surface of B cells,
making it an ideal therapeutic target for malignancies originating from this cell type.
Rituximab was approved in 1997, making it the first therapeutic mAb approved for
cancer treatment [68]. The addition of rituximab to the CHOP regimen led to a
10% increase in the 5-year overall survival of DLBCL patients [69, 70, 71, 72, 73].
Rituximab can recruit T cells [74], which in turn are able to kill cancer cells, or can
directly induce apoptosis via the p38 MAP-kinase signaling pathway [75].
Doxorubicin. Doxorubicin belongs to a class of chemotherapeutic drugs known
as anthracyclines [76], which are extracted from the Streptomyces bacterium.
Doxorubicin is used to treat a wide range of cancers, including DLBCL, and it is
generally used in combination with other chemotherapeutic agents. Doxorubicin
intercalates into DNA molecules, where it inhibits the progression of topoisomerase
II, an enzyme which relaxes supercoils in DNA for transcription and replication [77].
In this way it can halt DNA replication, consequently stopping the proliferation of
cells (including the abnormal growth of cancer cells).
2.4.2 Treating HGSOC
The standard treatment for HGSOC patients consists of a cytoreductive surgery
or "debulking" [78]. The surgeon proceeds to remove the ovaries, fallopian
tubes, and uterus to resect all macroscopic tumor masses. This procedure is
more successful in early-stage patients because the cancer has not yet spread
to more distal parts of the body. To make sure to completely eradicate the disease,
the majority of patients receive adjuvant chemotherapy after surgery. For the
last 20 years, a combination of platinum-based drugs and taxanes has been the
standard chemotherapy administered to HGSOC patients [78]. In recent years,
the administration of neoadjuvant chemotherapy prior to debulking has also been
investigated in the clinic, as an alternative for patients too sick to undergo surgery or
whose tumor is too widespread to be operable [79]. In this case, the patient would
receive neoadjuvant chemotherapy followed by interval debulking and additional
13
2 CANCER
adjuvant chemotherapy.
Even though the platinum/taxane combination is currently the most effective
chemotherapy regimen for HGSOC patients, ≥80% of them will eventually relapse
at some stage [57].
Platinum-based drugs. Currently there are three platinum-based drugs approved
worldwide [80]. Cisplatin was the first platinum compound approved as an anti-
cancer drug in 1978, followed by the less toxic analog carboplatin in 1986 and
oxaliplatin in 2004. Three other platinum compounds have been approved for
clinical use in specific countries (i.e., nedaplatin in Japan, lobaplatin in China,
and heptaplatin in Korea), while many more are currently being developed or are
already undergoing clinical trials.
Platinum-based drugs are absorbed by cells and transported to the nucleus, where
they cause DNA damage by crosslinking with purine bases and inducing apoptosis
[81].
In addition to being the main treatment for HGSOCs, platinum drugs are widely
used in combination therapies for several other cancers. In fact, platinum is
administered to half of patients receiving chemotherapy [82], and to 10-20% of all
cancer patients according to the National Cancer Institute [83].
2.5 Drug resistance in cancer
Despite our increased understanding of the molecular mechanisms behind cancer
and the plethora of treatments developed in the last century, resistance to such
therapies still hinders patient survival.
Drug resistance is defined as intrinsic when the heterogeneous tumor mass con-
tains subclones that are resistant to a certain therapy before the therapy itself is
administered (Figure 4A). This type of resistance is more common in cancers
diagnosed at a later stage as the tumors have had more time to accumulate genetic
and epigenetic changes, making them more likely to develop a resistant phenotype.
Drug sensitivity can also be lost during treatment, when the therapy administered
to the patient drives the formation of resistant subclones (Figure 4B). For instance,
DNA-damaging chemotherapy agents might cause mutations that confer resistance
to that same drug.
The mechanisms behind drug resistance in cancer are complex and still not fully un-
derstood. However, genes responsible for the absorption, distribution, metabolism,
and excretion of drugs have been implicated. Alterations of these genes can lead to
reduced drug intake, increased drug efflux, or even inactivation of the drug itself,
all mechanisms that can lead to a decrease in drug efficacy (Figure 4C). Moreover,
14
2 CANCER
Figure 4: Mechanisms of drug resistance. (A) Intrinsic drug resistance occurs when
one or more subpopulations of cancer cells already developed a resistant phenotype
before the treatment is administered. (B) Acquired drug resistance occurs when the
treatment causes changes in one or more cancer cells, which makes them insensitive
to the effect of the drug. (C) Cellular mechanisms of drug resistance such as
decreased drug influx, increased drug efflux, and drug detoxification via inactivation
or degradation contribute to decreasing the drug concentration inside the cell, hence
reducing its cytotoxic effect. Resistance may also arise when the structure of the drug
target is altered. Image created with BioRender.com
15
2 CANCER
since drugs reach cancer cells via the circulatory system, cells located in hypoxic
regions of the tumor mass might survive treatment simply because the therapy is
not able to reach them [84].
Changes in the structure, abundance or location of a drug target can also affect
the response to a specific therapy (Figure 4C). For instance, downregulation of the
MS4A1 gene (due to genetic or epigenetic factors) or defects in the transport of
CD20 to the cell membrane have been linked to rituximab resistance [85].
Certain cellular mechanisms might also nullify the effect of a drug. For example,
cancer cells with functional DNA repair pathways (such as homologous recom-
bination) are more likely to develop resistance to DNA-damaging agents like
platinum-based drugs [86] and doxorubicin [87].
2.6 Predicting and overcoming drug resistance in cancer
Transitioning from single-agent therapies to combination therapies was one of the
first approaches to overcoming drug resistance in cancer. However, even multi-drug
regimens fail to completely eradicate cancer cells in an individual. Hence, new
strategies to identify patients less likely to benefit from a certain therapy, and to
overcome drug resistance are urgently needed.
2.6.1 Identification of biomarkers
Biomarkers are measurable indicators used to accurately and reproducibly classify
a certain biological condition. To overcome drug resistance in cancer, better
biomarkers are needed to both (i) diagnose tumors at earlier stages (when they are
less likely to have developed drug resistance), and (ii) predict whether a patient
will respond to a certain treatment, so therapies are administered only when the
probability of success is high enough to justify exposing the patient to its side effects
[88]. Different biological samples (e.g. blood, urine, soft tissues, tumor samples,
etc.) and different data types (e.g. genetic [89], epigenetic [89], transcriptomic [90],
proteomic [91], and imaging [92] data) can be used to identify biomarkers. For
instance, mutations in the DNA repair genes BRCA1 and BRCA2 are biomarkers
that can predict the efficacy of PARP inhibitors when treating ovarian cancer [93].
2.6.2 Restoring drug sensitivity via epigenetic reprogramming
In the last few decades, the plasticity of the epigenome has emerged as an appealing
avenue to overcome drug resistance in cancer. Currently, a plethora of epigenetic
inhibitors have been developed in order to block the activity of the main classes of
16
2 CANCER
Figure 5: Different strategies to inhibit DNMT enzymes. (A) Cytidine analogues like
Azacitidine are incorporated in the DNA during cellular replication and, because of
their molecular structure, cannot be methylated, consequently inhibiting the action of
DNMT enzymes. (B) Substrate substitutes targeting the active site of DNMT enzymes
inhibit DNMTs from binding with methyl groups. (C) Antisense oligonicleotides
targeting mRNA molecules encoding DNMTs can halt the production of these enzymes.
Image created with BioRender.com
epigenetic enzymes, including DNMTs, HDACs, HATs, BRDs, HMTs, and HDMs.
So far, only six epigenetic compounds have reached FDA approval [94]: the DNMT
inhibitors azacitidine (in 2004) and decitabine (in 2006), and the HDAC inhibitors
romidepsin (in 2004), vorinostat (in 2006), belinostat (in 2014), and panobinostat
(in 2015). Other drugs, like valporic acid, have also been approved but not for their
epigenetic effect.
Different strategies have been used to design compounds that are able to inhibit
epigenetic enzymes. For instance, when aiming to inhibit DNMTs, one option is to
design nucleoside analogues (e.g. azacitidine and decitabine) containing a modified
cytosine ring that cannot be methylated (Figure 5A) [95]. During cell replication,
these molecules are incorporated into the DNA strand, making it impossible for
DNMTs to methylate the DNA containing the modified nucleosides. Another
strategy used to inhibit DNMT enzymes consists of designing small molecule
inhibitors (e.g. RG108 [96]) that are able to bind directly to the DNMT’s catalytic
region, hence preventing the binding between the DNMT enzyme and a methyl
group (Figure 5B). Decreasing the abundance of a DNMT enzyme is also a way to
inhibit its activity, and can for instance be achieved using antisense oligonucleotides
(e.g. MG98 [97]) that bind to the enzyme’s mRNA and prevent it from being
translated into protein (Figure 5C).
Epigenetic inhibitors are an emerging treatment option for several diseases, in-
cluding diabetes, cardiovascular and immune diseases, mental illnesses, anxiety
disorders, and of course cancer [98]. The addition of epigenetic inhibitors to cancer
regimens is being investigated, as in the case of the HMT inhibitor tazemetostat
tested in combination with R-CHOP to treat newly diagnosed DLBCLs with poor
prognosis [99]. However, recently it has been suggested that using epigenetic
inhibitors as pretreatments might be a useful strategy to sensitize cancers that are
17
2 CANCER
resistant to standard treatment. Successful chemosensitization through epigenetic
reprogramming has already been shown in vitro in blood cancers such as acute
lymphoblastic leukemia [100], but further investigation into this therapeutic option
is needed before testing it in the clinic.
18
3 HIGH-THROUGHPUT BIOLOGY
3 High-throughput biology
In the last few decades, several new high-throughput sequencing technologies
have reshaped the landscape of biomedical research, shifting the scale from single
genes to entire genomes. Consequently, novel computational tools tailored to
process platform-specific data have also been developed. The widespread use of
these technologies has drastically improved our understanding of the molecular
mechanisms regulating healthy cells and driving tumor progression. In turn, this
has led to the identification of several novel targets for cancer treatment, boosting
the drug discovery pipeline.
The number of cancer drugs available on the market has been growing exponentially,
and so has the number of compounds under development. Moreover, repurposing
of existing drugs for cancer treatment requires assays that can screen thousand
of compounds to select the most promising ones. Hence, high-throughput drug
screening protocols were developed to efficiently test drug response in multiple
cancer samples. By integrating information from sequencing and drug response
data, we are now able to investigate therapy resistance in different cancers with an
unprecedented level of accuracy.
3.1 High-throughput screenings
High-throughput screenings (HTSs) have become a standard assay which allows
for the testing of multiple drugs and drug combinations using in vitro models,
such as cell lines or organoids, or ex vivo patient samples [101]. Cells are seeded
in multi-well microtiter plates and the use of liquid-handling robotics allows for
the investigation of tens, hundreds, or even thousands of compounds at varying
concentrations and in a short amount of time. A standard HTS protocol includes
seeding the cells on the plate, adding the compounds to be tested in each well and,
after a certain amount of time, measuring the proportion of cells that were killed by
each compound.
When designing an HTS experiment, it is fundamental to select the right samples,
and to design a comprehensive compound library. Plate effect errors, caused for
instance by increased medium evaporation in the border wells of the plate, can also
be accounted for by randomizing the plate design and by including replicates for
each measurement. Controls are also extremely important when planning a drug
screening experiment, because they allow researchers to normalize and interpret the
effect of the compounds tested. A negative control is a substance that should not
affect cells’ viability. Dimethyl sulfoxide (DMSO) and water, commonly used to
19
3 HIGH-THROUGHPUT BIOLOGY
dissolve drugs, are examples of suitable negative controls. A substance that would
instead achieve maximal cytotoxic effect is used as positive control. An example of
such a compound is benzethonium chloride, a commonly used antiseptic.
Different drug response quantification techniques are currently available to estimate
the viability of cells after treatment. A common assay employs the CellTiter-
Glo® reagent (Promega), which lyses the cells and generates a luminescent signal
proportional to the amount of ATP, an indicator of metabolically active cells. The
luminescence readout is then measured using a microplate reader.
Including multiple concentrations of the same compound in the screening generates
dose response data, which are used to estimate the half maximal inhibitory concen-
tration (IC50) of a compound, i.e. the dose of the compound needed to kill half of
the cells present in the sample. The IC50 is used as the standard measure of the
potency of a drug.
High-throughput screening can also be used to test how the knockdown of a certain
gene affects cells’ viability, their phenotype and/or their response to a specific
therapy. These screenings exploit RNAi mechanisms by transfecting cells with
artificial siRNA [25].
3.1.1 Analyzing dose response data
The standard computational analysis of dose response data obtained from HTS
follows the steps summarized in Figure 6. First, HTS luminescence measurements
generated by a microplate reader are annotated to mark what is measured in each
well. The reliability of the readout is then estimated via standard quality control
(QC) statistics, which include coefficient of variation for both positive and negative
controls, Z’ score, and strictly standardized mean difference [102]. If no severe
bias or plate effect is detected, the luminescence readout is normalized using the
negative controls (0% inhibition) and the positive controls (100% inhibition).
Normalized dose response data for a drug are fitted to a sigmoidal function. A
common model used to fit the curve is the 4-parameter logistic function:
y= a+
d−a
1+(x/c)−b
(1)
where a and d represent the theoretical response when no drug is administered
(x= 0) and when administering an infinite amount of drug (x=+∞), respectively.
c represents the IC50, and b is the slope factor of the curve at x =IC50. The R
package drc [103] performs curve fitting and estimates the IC50 of a drug. Fitted
curves are then used to estimate the effect of the drug on the cells, for instance by
20
3 HIGH-THROUGHPUT BIOLOGY
Figure 6: Standard bioinformatic pipeline describing the analysis of dose response
data obtained from a drug screening. Cell viability is estimated using a microplate
reader which measures the luminescent signal emitted by the fraction of living cells
in each well. The experimental conditions tested in each well are used to annotate
the data. Next, QC statistics are estimated to assess the reliability of the experiment,
and data are normalized using positive and negative controls. Data are then fitted to a
sigmoidal function, which is used to estimate the effect of the drug.
computing drug sensitivity scores (DSSs) [104]. DSSs can then be used to compare
the effect of different compounds and conditions on the same sample.
Software such as the online tool Breeze can analyze dose response data from QC to
DSS calculations [105].
3.2 Next-generation sequencing technologies
The first sequencing technologies were developed by Sanger and his colleagues
at Cambridge University in 1977 [106]. These technologies read DNA and RNA
sequences at a single-base resolution, but were very low-throughput and quite costly.
Consequently, since the mid-1990s hybridization-based microarrays became the
standard high-throughput assays to quantify gene expression and measure genomic
and epigenomic changes [107, 108]. However, despite being very powerful and
efficient, microarrays are limited in the sense that they do not allow for measurement
of de novo sequences, restricting investigation to those genes and DNA regions
that are already known. In 2005, a new generation of sequencing technologies,
also known as next-generation sequencing (NGS), was developed [109]. These
technologies combined the precision of Sanger sequencing with the ability to
produce a high-throughput readout. Although quite expensive at first, the costs of
NGS assays has steadily decreased in the last decade, drastically changing the way
work in fields like genomics, epigenomics, and transciptomics is conducted. By
allowing researchers to investigate these molecular mechanisms at a single-base
resolution and with unprecedented precision, NGS has undoubtedly increased our
understanding of cellular biology.
Among the most common applications of NGS are whole genome and whole
exome sequencing for the study of mutations and copy number alterations, RNA
sequencing [110] to investigate changes in the transcriptome, bisulfite sequencing
to map methylated cytosines across the genome, chromatin immunoprecipitation
21
3 HIGH-THROUGHPUT BIOLOGY
sequencing (ChIP-Seq) to identify transcription factor binding sites and locate
histone modifications, and an assay for transposase-accessible chromatin using
sequencing (ATAC-Seq).
In the last ten years, a third generation of sequencing technologies (TGS), also
known as long-read sequencing, was introduced [111]. Nanopore technologies
[112] and other TGSs can sequence substantially longer reads than NGSs, producing
genome assemblies of unprecedented quality. Moreover, TGSs can detect DNA
methylation without the need to perform bisulfite conversion.
3.2.1 RNA sequencing
RNA sequencing (RNA-Seq) exploits NGS technologies to quantify the abundance
of RNA transcripts and to estimate differential gene expression between samples. In
addition, it also allows for the investigation of other mechanisms such as alternative
splicing, gene fusion, and mutations.
The RNA-Seq protocol starts with the extraction of RNA molecules from the
cells. Investigation can be done on the whole transcriptome (total RNA) or focus
only on protein coding genes by selecting only mature mRNAs (by capturing
RNA molecules that have a poly-A tail). In both cases, ribosomal RNAs, which
account for over 80% of the RNA molecules inside a cell, are discarded. Size-
selection is then carried out to identify specific RNA molecules, such as small
non-coding RNAs. Next, RNA is fragmented and converted into cDNA to prepare
the sequencing library. Adaptors are added at both ends of each cDNA molecule.
Finally, the library is amplified via polymerase chain reaction (PCR) and sequenced.
In addition to the standard RNA-Seq assay, which measures the average transcript
abundance from all cells in a bulk sample, a single-cell (sc) RNA-Seq protocol has
also been developed [113]. ScRNA-Seq processes each cell individually, allowing
researchers to obtain gene expression signatures from different cell types.
3.2.2 Bisulfite sequencing
Bisulfite sequencing (Bi-Seq) is a technology that allows for the genome-wide
investigation of DNA methylation at a single-base level. Bisulfite conversion uses
sodium bisulfite to deaminate unmethylated cytosines (Cs) into uracils (Us), which
are then converted into thymines (Ts) during the PCR step. Methylated Cs are not
affected by the bisulfite conversion reaction and can hence be detected once the
reads are aligned to a reference genome. A T in the data mapping to a T in the
reference genome is a T, a T mapping to a C is an unmethylated C, and a C mapping
to a C is a methylated C.
22
3 HIGH-THROUGHPUT BIOLOGY
Currently there are three protocols to perform Bi-Seq of bulk samples. Whole
genome bisulfite sequencing (WGBS) [114] was the first method to be developed.
Even though it allows for the analysis of all 29 million CpG sites found in the
human genome, it was initially quite costly and required a high amount of starting
genetic material, especially when aiming to produce data with high sequencing
depth. To address this limitation, the reduced representation bisulfite sequencing
(RRBS) [115] protocol was introduced. RRBS only sequences about 1-10% of the
genome, targeting CpG-rich regions which include most CpG islands, promoters,
gene bodies, other regulatory elements, and repetitive sequences. A third way to
perform Bi-Seq, known as targeted bisulfite sequencing (TBS), is an enrichment-
based procedure able to capture the methylation of CpGs located at specific genomic
regions via hybridization through a probe-based system [116]. In this way, the
experimental costs and the data size are both reduced with respect to WGBS, while
ensuring consistent sequencing of all significant portions of the genome. A standard
array targeting about 3.7 million CpGs is available, covering most promoters and
other relevant genomic regions, but users can also customize the probes to target
specific portion of the genome.
WGBS, RRBS, and TBS follow a very similar protocol. Briefly, DNA is extracted
from the cells, purified from the proteins bound to it, and fragmented. WGBS and
TBS use sonication to randomly fragment the DNA molecule, while RRBS cuts
the genome at specific sites by using a restriction enzyme. The fragments undergo
end repair, A-tailing and adapter ligation, and are then selected based on size via
gel electrophoresis or magnetic beads. The TBS protocol also includes a fragment
selection step via hybridization, which allows for the enrichment of specific target
regions. The surviving fragments are then denaturated and treated with sodium
bisulfite to deaminate unmethylated Cs into Us. PCR amplification converts Us
back to Ts and the fragments are then ready to be sequenced.
Recently, a single-cell bisulfite sequencing protocol was developed, allowing for
more accurate analysis of samples containing heterogeneous cell populations [117].
3.2.3 Analyzing sequencing data
All types of sequencing data require tailored software and pipelines to extract mean-
ingful biological information. However, when performing differential expression
or differential methylation analyses, the main steps are very similar, as shown in
Figure 7 and described below.
Quality control. The computational analysis of RNA-Seq and Bi-Seq data starts
with QC of the reads (usually encoded in fastq format) to assess the proportion
of data that are of sufficiently high quality to be included in the analysis. QC is
23
3 HIGH-THROUGHPUT BIOLOGY
commonly performed using software such as FastQC [118], which produces several
plots to assess, among others, the distribution of quality scores at different positions
in the read, the distribution of the average read quality, the read length distribution
across all reads, the proportion of each nucleotide at different positions in read, and
the the presence of overrepresented sequences.
Trimming and adaptor removal. Reads might contain parts of the adapter
sequence at both ends; moreover, nucleotides at the extremities of a read often
Figure 7: Standard bioinformatic pipeline describing the analysis of RNA-Seq and
bisulfite sequencing data to perform differential expression / methylation analyses.
Steps like quality control, trimming, and pathway enrichment analysis can be performed
using the same software for both RNA-Seq and Bi-Seq data. On the other hand, tools
specific for each data type are needed for alignment, quantification, and identification
of differences between conditions.
24
3 HIGH-THROUGHPUT BIOLOGY
display lower sequencing quality and might lead to misalignment. To minimize
these sources of error, tools such as Trimmomatic [119] or Trim Galore! [120] are
commonly used to trim the reads, removing adapters and low quality end sequences.
Low quality reads or reads that are too short after being trimmed are discarded, and
the quality of the remaining reads is assessed again.
Alignment. Reads passing these first filters are then mapped against a reference
genome using aligner tools designed to address the challenges of each data type.
Hence, aligners for RNA-Seq data should accounts for spliced alignments when
mapping reads to the reference, while aligners for Bi-Seq data should take into
account the T-to-C discrepancy caused by bisulfite conversion. STAR (Spliced
Transcripts Alignment to a Reference) [121] is the most common aligner for RNA-
Seq data. Even though it is memory intensive, STAR is much faster than other
aligners, outperforming them in mapping speed by a factor of over 50. Bismark
[122] and BSMap [123] are are frequently used to align Bi-Seq data. This class
of aligner tools allows not only to map reads to a reference genome, but it also
estimates whether the base mapped to a reference cytosine is a C (i.e. a methylated
C) or a T (i.e. an unmethylated C). The sequence context in which the reference
cytosine is embedded is also estimated, to allow user to separate CpG sites from
CHG and CHH sites (where H stands either A, G, or T nucleotide).
Quantification. Once the reads are mapped, it is possible to estimate the levels
of expression or methylation. Software like eXpress [124] can quantify the
abundance of each gene/transcript. New software such as Kallisto [125] can
perform gene/transcript quantification without first aligning the RNA-Seq reads.
Common measures of gene expression are read counts and FPKM (which stands for
Fragments Per Kilobase of transcript per Million mapped reads). Bi-Seq aligners
such as Bismark usually include a function to extract the methylation values from
aligned reads, but there are also tools such as MethylExtract [126] which can call
both methylation levels as well as genetic variants. The methylation level of a
specific cytosine in the genome is commonly estimated as the proportion between
the number of mapped reads containing a C over the total number of reads aligned
at that same position.
Differential expression/methylation. After the quantification step, different sam-
ples are compared to identify differentially expressed genes (DEGs). Tools such as
the R package DESeq2 [127] are commonly used to estimate differential expression
between conditions. DESeq2 takes as input the unnormalized counts and the
condition to which each sample belongs to, and for every gene it estimates the log2
fold change, p-value, and adjusted p-value, which can then be used to identify genes
with significant changes in abundance between conditions. Differential methylation
to locate DNA methylome changes across conditions can be performed at the single
25
3 HIGH-THROUGHPUT BIOLOGY
CpG level, to identify differentially methylated sites (DMSs), or at the regional level,
to identify for instance differentially methylated genes (DMGs) or differentially
methylated promoters (DMPs). Several bioinformatic tools are available to compute
differential methylation. The R package RnBeads [128, 129] is one of the most
commonly used packages, thanks to its extensive functionalities for filtering and
exploring the data before computing differential methylation via Fisher’s exact
test. Other tools include MOABS [130], and R packages like MethylKit [131],
InfiniumPurify [132], and DSS (dispersion shrinkage for sequencing data) [133].
Pathway and GO enrichment. To interpret the role that DEGs and DMGs/DMPs
play, pathway analysis and gene ontology enrichment are performed using tools
like enrichR [134, 135] or ConsensusPathDB [136], which automatically browse
through public databases (such as KEGG [137], Reactome [138], WikiPathways
[139], and gene ontology [140]) to identify the molecular and biological mecha-
nisms affected by changes in the transcriptome.
When both RNA-Seq and Bi-Seq data are available for the same set of samples, it is
also possible to integrate these two data layers to identify genes whose expression
might be regulated by changes in DNA methylation patterns.
3.2.4 Public repositories of sequencing data
The huge amount of data generated by high throughput technologies created a need
for building public, discipline-specific repositories, where such data can be stored
and shared with the whole scientific community. Examples of such repositories
include the Gene Expression Omnibus (GEO) [141], the Sequence Read Archive
(SRA) [142], the International Human Epigenetic Consortium (IHEC) [143], the
Cancer Cell Line Encyclopedia (CCLE) [144], and the European Genome-phenome
Archive (EGA) [145].
In this thesis we used data from two public repositories: The Cancer Genome Atlas
[146] and The Encyclopedia of DNA Elements [147, 148].
The Cancer Genome Atlas (TCGA) is an program aimed to characterize the
molecular signatures of over 20,000 primary cancer and matched normal samples
spanning 33 cancer types. Using a mix of microarray and sequencing technologies,
TCGA generated genomic, epigenomic, transcriptomic, and proteomic data. These
data are available in the TCGA online repository in three different formats: raw and
controlled data (level 1), processed and controlled data (level 2), and segmented or
interpreted data (level 3). For each sample, clinical data are also available. This
allows researchers to perform pan-cancer integrative studies.
The Encyclopedia of DNA Elements (ENCODE) is a project launched in 2003
26
3 HIGH-THROUGHPUT BIOLOGY
that aims to develop a comprehensive map of functional elements in the human and
mouse genome. ENCODE data are publicly available and include both functional
genomic data of cell lines and tissues generated through microarray and sequencing
technologies, as well as functional characterization data.
3.3 Bioinformatic workflow management systems to analyze
high-throughput data
The large volumes of data generated by high-throughput technologies require multi-
step computational analyses in order to extract biologically meaningful information.
The sequence of tasks to process a set of data is known as workflow and is commonly
represented via a directed network. Workflows can be run manually, but when
dealing with large datasets a fully automated approach is generally more efficient.
Hence, workflow management systems (WMSs) have been developed to provide an
infrastructure to design, execute, and monitor complex pipelines.
Several WMSs are currently available to perform tailored bioinformatic analy-
ses by combining state-of-the-art computational tools. WMSs such as Galaxy
[149], Taverna [150], Chipster [151], and GenePattern [152] are suitable for non-
programmers, while other WMSs, like Anduril [153], Bpipe [154], Nextflow [155],
and Snakemake [156], require coding skills in order to build the workflows.
A schematic representation of a bioinformatic WMS is shown in Figure 8. Bioinfor-
matic WMSs can import multidimensional biomedical data generated by the user or
available in public databases. The user can then design workflows that sequentially
combine different steps in the analysis, each carried out by specific computational
tools and programming languages, to process, visualize, and even integrate multiple
data layers. The WMS will automatically run the pipeline using high-performance
computing and generate results that can be used to draw conclusions regarding the
biomedical questions under investigation.
27
3 HIGH-THROUGHPUT BIOLOGY
Figure 8: Schematic representation of a bioinformatic workflow management system.
The WMS can import biomedical data generated by the user or available in online
public repositories. The data are then processed via a workflow designed by the user,
which employs external state-of-the-art bioinformatic software, but can also include
user-generated scripts. The workflow is then run using high-performance computing
resources to optimize the running time.
28
4 AIMS OF THE STUDY
4 Aims of the study
The overall objective of this thesis is to develop in vitro and in silico methods to
increase our understanding of the mechanisms behind drug resistance in cancer. To
achieve this goal, we used a systems biology approach which combined different
laboratory biology experiments, multivariate statistical methods and bioinformatic
analyses to uncover cellular processes responsible for drug resistance. The main
goals of this project are:
1. Developing a drug screening assay to discover epigenetic inhibitors able to
overcome doxorubicin and rituximab resistance in DLBCL.
2. Integrating multi-omics data to identify predictive biomarkers of platinum
resistance across multiple cancers.
3. Developing a computational method to estimate differential DNAmethylation
across cancer samples of varying tumor purity.
4. Developing a workflow management system to build complex bioinformatic
pipelines to analyze and integrate multiple data types.
29
5 MATERIALS AND METHODS
5 Materials and methods
In this section, the main materials and methods employed in this work are summa-
rized.
5.1 Materials, samples and data
The patient samples and cancer cell lines used in Publications I-IV as well as in
our unpublished results are summarized in Table 1, while the drugs, compounds
and siRNAs used in the biological experiments of Publications I and II are listed
in Table 2. Full details on the experimental designs are available in the appended
manuscripts.
Publication Samples Technology Source
Publication I
Oci-Ly-19, Oci-Ly-3,
Riva-1, and Su-Dhl-4
DLBCL cell lines
Drug screening, RNA-
Seq, WES, immunofluo-
rescence assay
In-house
Publication II
Treatment-naive samples
from 1,503 cancer pa-
tients who received plat-
inum therapy
RNA-Seq, CGH-array,
MethylArray, Clinical
data
TCGA
A549, FaDu, Kuramochi,
and SW-48 cell lines
RNAi experiment
to measure cisplatin
response when 12 genes
are knocked-down
In-house
Publication III
GM12878, K562, and
HepG2 cell lines
WGBS ENCODE
5 HGSOC samples (2
primary and 3 relapse)
and 1 white blood cell
sample
TBS, RNA-Seq, WGS In-house
8 DLBCL samples (4
primary and 4 relapse)
and 2 blood samples
RRBS, WGS In-house
Publication IV
50 OV samples (26 long
response patients and 24
short response ones)
RNA-Seq TCGA
Unpublished data
53 HGSOC samples (27
primary, 17 interval, 9
relapse) from 14 different
patients
WGBS In-house
Table 1: Samples and cell lines used in Publication I-IV and in our unpublished
analysis
30
5 MATERIALS AND METHODS
Publication Class Compound
Publication I
BRD inhibitors I-BET151, JQ1, OTX015, PFI-1, SGC-
CBP30, UNC1215
DNMT inhibitors Azacitidine, Decitabine, Epigallocatechin,
Lomeguatrib, Procainamide, RG108,
Thioguanine
HAT inhibitors C646
HDAC inhibitors AR-42, Belinostat, CI994, CUDC-101,
CUDC-907, Entinostat, Givinostat, JNJ-
26481585, Mocetinostat, Panobinostat,
PCI-24781, PCI-34051, Resminostat,
RGFP966, Rocilinostat, Romidepsin,
SB939, Tubacin, Tubastatin A, Valproic
acid, Vorinostat
HDM inhibitors GSK J4 HCl, IOX-1, Tranylcypromine
HMT inhibitors EPZ005687, GSK343, Pinometostatm,
Tazemetostat, UNC0638, UNC0642
Chemotherapy Doxorubicin
Immunotherapy Rituximab
Publication II
siRNAs siRNAs targeting genes ANXA9, ARC,
C19orf33, ECI2, FBXO17, LGALS3BP,
MLF1, MRPS21, PLK1, SEC62, SOX17,
TM4SF1, and TSPYL5
Chemotherapy Cisplatin
Table 2: Compounds and siRNAs used in Publication I-II
The unpublished data include 53WGBSHGSOC samples extracted from 14 patients
diagnosed at stage III or IV. Samples were collected during primary debulking or
via laparoscopy (primary sample, p), during interval debulking (interval sample,
i), or when the patient relapsed (relapse sample, r). Samples were also obtained
from multiple locations, including ovary (ova), omentum (ome), peritoneum (per),
intestine (meso), Fallopian tubes (tub), ascites (asc), or undefined location (tum).
Samples were collected at the Department of Obstetrics and Gynecology of the
Turku University Central Hospital, as part of the HERCULES project (European
Union’s Horizon 2020 research and innovation program under grant agreement
No. 667403). Ethical approval for the study is stated in the Hospital District of
Southwest Finland ethics committee statement EMTK 145/2015 22.5.2018 § 197.
5.2 Drug sensitivity screening
We used automated liquid-handling robots to implement multi-step drug screenings
to investigate drug resistance in cancer.
31
5 MATERIALS AND METHODS
In Publication I, we designed a drug screening to test whether pretreating DLBCL
cell lines with epigenetic inhibitors can increase the response of doxorubicin and
rituximab. Briefly, 60 epigenetic compounds were pre-seeded on 384-wells plates
and DLBCL cells were then added. A new dose of epigenetic inhibitor therapy
was administered every three days for nine days, to induce cellular reprogramming.
Rituximab and doxorubicin were then administered (at fixed concentrations in the
initial screening, and at varying concentrations in the validation screening). 48
hours later, viability was measured to determine whether the epigenetic treatment
induced sensitization of DLBCL cells to the immunochemotherapy.
In Publication II, we tested how cisplatin response changes when genes we iden-
tified as related to platinum resistance were knocked-down. siRNAs targeting 12
biomarker genes were transfected in cancer cell lines and, after 24 hours, cisplatin
was administered. Cell viability was measured after 48 hours and the effect of gene
knockdown on platinum response was estimated.
5.3 Anduril workflow management system
The open source component-based workflow framework Anduril (ANalysis of
Data Using Rapid Integration of aLgorithms) [153, 157] developed by the Systems
Biology of Drug Resistance in Cancer group (University of Helsinki) was used to
build the computational pipelines in Publications I-IV.
Pipelines implemented using the first version of Anduril were coded using the
programming language AndurilScript, while Anduril 2.0 uses the Scala language
to build workflows. Each pipeline runs a sequence of components, which can
be created using a variety of programming languages, including R, Matlab, Java,
Bash, and Python. Components are divided into bundles based on the type of
data/analysis they are designed for. The pipelines in Publications I-IV combine
existing components with newly developed ones, and a separate bundle of Anduril
components was created to analyze dose response data from Publication I.
5.4 Differential methylation using maximum-likelihood (DMML)
method
Bulk tumor samples contain a mixture of cell types that comprise both cancerous
and non-cancerous cells. Thus, when comparing the methylomes of two cancer
samples, it is important to account for their differing tumor purity levels. To address
this issue, we developed a computational method to perform differential methylation
using maximum-likelihood (DMML). The DMML algorithm estimates the purity of
32
5 MATERIALS AND METHODS
two or more cancer samples, extracts the tumor-specific signal from bulk samples,
and computes differential methylation more accurately. As methylation patterns
might span several neighboring CpG sites, DMML models co-methylation for a
user-defined number of adjacent CpG sites that are measured in the same read.
A detailed description of the DMML method can be found in Publication III.
Briefly, the method uses an expectation–maximization (EM) algorithm to estimate
the latent variables (in our case the DNA methylation levels in both cancer and
normal cells) using the observed DNA methylation data (which comes from a
mixture of tumor and non-tumor cells). DMML can take a non-tumorous control
sample as input, but this is not required when estimating the purity of the samples
or assessing differential methylation. Controls are only required when determining
the methylome state for each cell type.
We compared the performance of DMML against other existing methods, including
Fisher’s exact test, MOABS, a mixture-adjusted Fisher’s exact test, InfiniumPurify
and DSS methods. To evaluate their performance we used simulated data generated
by mixing WGBS data from three ENCODE cell lines (GM12878, K562, and
HepG2). We also estimated tumor purity and differential methylation in patient-
derived samples, including TBS data from HGSOCs and RRBS data for DLBCLs.
We used DMML to estimate tumor purity in patient samples, and compared the
obtained values to the purity estimated from genome sequencing data from the
same samples.
DMML is suitable for pairwise comparison of two or a few samples, but the
co-methylation modeling needs to be dropped to scale the method and allow
simultaneous comparison of many (»10) samples. In this way, we were able to use
DMML to extract the patient-, location-, and treatment stage-specific methylation
profiles, as well as a profile from the non-cancerous cells, from samples in our
unpublished HGSOC dataset.
5.5 Interactive exploration of multiple data types
Visualizing and interpreting biomedical datasets can be challenging due to the huge
amounts of information they contain. To address this issue, we used the R packages
shiny [158] and plotly [159] to create interactive visualizations of our datasets in
Publication I. Shiny allows users to subset data and visualize different portions of
the entire dataset, while plotly provides interactivity with the plot itself by hovering
over it and zooming in and out.
Using these packages we developed a website (http://app.anduril.org/DLBCL_DSRT)
where different data types (including sequencing data, drug response curves, and
33
5 MATERIALS AND METHODS
immunofluorescence images) could be loaded and explored.
5.6 Automatic workflow to integrate multi-omics TCGA data and
identify predictive biomarkers
In order to integrate multi-omics data from pan-cancer TCGA patients, we designed
a pipeline that would download expression, DNA methylation and gene copy
number data from the GDC data portal, and identify genes (9,976) and samples
(1,503) for which all three data layers are available. DNA methylation and copy
number data were then binarized to match the input format required by the R
package CNAmet [160], which was then used to perform data integration and
identify genes whose expression is regulated by DNA methylation or gene copy
number.
5.7 Survival analysis
In biomedical research, survival analysis is an extremely useful tool to assess how
certain biological and molecular features, or even certain treatment options, improve
or diminish survival in a cohort of patients across time. Typical time measures are
progression free survival (PFS) and overall survival (OS). PFS measures the time
from diagnosis until the first recurrence, the last follow-up or death. OS measures
instead the time from diagnosis until the last follow-up or death.
The survival function estimates the proportion of patients still alive at a given time
and it is visualized using Kaplan-Meyer plots. The non-parametric log-rank test
is commonly used to test whether the difference between two survival functions
is significant. In Publication II and IV, survival analysis was performed in R via
Kaplan-Meier analysis and log-rank test using the package survival [161, 162] and
survival curves were visualized using the package survminer [163]. In Publication
II, patients with the 20% highest and lowest expression of each gene were included
in the analysis, and PFS was used as the the time variable. In Publication IV, OS
was instead used to investigate the effect that certain mutations had on the survival
of HGSOC patients.
34
6 RESULTS
6 Results
In this thesis, I present different in vitro and in silico methods designed and imple-
mented to investigate the molecular mechanisms responsible for drug resistance
in cancer. The main results from Publications I-IV, as well as some unpublished
data on DNA methylation changes occurring in HGSOC patients, are summarized
below.
6.1 Pretreating DLBCL cell lines with epigenetic inhibitors sensi-
tizes them to immunochemotherapy (I)
In order to systematically assess the ability of epigenetic inhibitors to reverse
R-CHOP resistance in DLBCL, we designed and implemented a multi-step drug
screening. Four DLBCL cell lines (Riva-I, Su-Dhl-4, Oci-Ly-3, and Oci-Ly-19)
were pretreated for 9 days using 60 different epigenetic inhibitors, and were then
administered rituximab and doxorubicin (the main components of the R-CHOP
regimen). Cell viability was estimated 48h later and the response of cells receiving
only the epigenetic pretreatment was compared to the response of cells receiving
the pretreatment followed by rituximab and doxorubicin. HDAC inhibitors induced
sensitization in all cell lines, while BRD and HMT inhibitors sensitized three of
the four cell lines. Oci-Ly-3 was the most responsive cell line (20/60 inhibitors
inducing sensitization), followed by Su-Dhl-4 (10/60), Oci-Ly-19 (9/60) and Riva-I
(3/60). It is important to note that the concentration of the epigenetic inhibitors
required to induce cellular reprogramming was lower than the concentration needed
to induce a cytotoxic effect, which indicates that using epigenetic inhibitors as a
pretreatment instead of in combination with R-CHOP might lead to less side effects.
To validate the sensitizing effect of the most potent inhibitors, we conducted a
second drug screening to estimate the synergy between the epigenetic compounds
and the immunochemotherapy. In this assay, we tested varying concentrations
of both the epigenetic inhibitor and the rituximab-doxorubicin combination. Ten
compounds were included in this validation: belinostat, entinostat, and I-BET151
were tested in all four cell lines, while vorinostat, resminostat, givinostat, pinome-
tostat, tazemetostat, SGC0946, and OTX015 were tested only in the cell line(s)
that showed a reprogramming effect from that drug in the first screening. HDAC
inhibitors (vorinostat, entinostat,resminostat, and belinostat), and HMT inhibitors
(pinometostat, tazemetostat, and SGC0946) showed high synergy with rituximab
and doxorubicin. Among these, entinostat and tazemetostat displayed the most
potent sensitization effects. BRD inhibition showed instead lower synergy.
35
6 RESULTS
6.2 Molecular mechanisms responsible for epigenetic sensitiza-
tion in DLBCL (I)
To investigate which genes and pathways are linked to epigenetic sensitization,
we compared the trascriptomes of the four untreated DLBCL cell lines with
their transcriptomes after pretreatment with entinostat, vorinostat, belinostat and
tazemetostat. Interestingly, differential expression followed by pathway analysis
identified disruption of DNA repair, cell cycle, cell adhesion and apoptosis as
potential mechanisms behind epigenetic sensitization.
We further investigated the link between impaired DNA repair pathways and epige-
netic sensitization to doxorubicin using an immunofluorescence assay. DLBCL cells
treated with HDAC inhibitors (entinostat, belinostat, vorinostat) showed impaired
homologous recombination (HR) and upregulation of the non-homologous end
joining (NHEJ) pathway (which was expected since NHEJ is often active when HR
is not functioning correctly).
When we looked into genetic mutations which might affect the efficacy of epigenetic
reprogramming, we observed that the Su-Dhl-4 cell line has a missense mutation in
the EZH2 gene, which encodes an HMT enzyme. This mutation has been shown to
increase sensitivity to tazemetostat [164]. Other notable mutations that might affect
the efficacy of HDAC inhibitors were a truncating mutation in the CREBBP gene
identified in Riva-I and Oci-Ly-19, and a truncating mutation in the ARID1A gene
in Riva-I.
6.3 Identification of potential biomarkers and therapeutic targets
to predict and overcome platinum resistance in multiple can-
cers (II)
The integration of gene expression, DNA methylation, and gene copy number data
of 9,976 genes across 1,503 pan-cancer patients treated with platinum chemotherapy
revealed 333 upregulated and 48 downregulated genes (FDR ≤ 0.01). Survival
analysis showed that 164 out of these 381 genes have a significant association to
progression free survival (PFS), and are hence promising candidate biomarkers
to predict platinum response. A spreadsheet listing these genes is included in
supplementary table S4 of Publication II. Among these 164 putative biomarkers we
found genes involved in pathways related to drug resistance, cancer progression
and metastasis, apoptosis, and immune-related mechanisms. Some of these genes,
like ARC, were detected at high and low expression in most of the 17 cancer types
included in the study. Other genes, like SOX17, were expressed only in one or a few
36
6 RESULTS
cancers. To identify the most promising pan-cancer biomarkers, we used different
ranking systems to prioritize genes that had (i) both high and low expression across
most cancer types, and (ii) whose expression was strongly regulated by gene copy
number or DNA methylation.
To investigate whether our biomarkers could also serve as therapeutic targets to
increase platinum response in multiple cancers, we performed an RNAi screening
using cell lines obtained from lung (A549), head and neck (FaDu), ovarian (Ku-
ramochi), and colorectal (SW-48) cancers. ANXA9, C19orf33, ECI2, FBXO17,
LGALS3BP,MLF1, MRPS21, SEC62, TM4SF1, TSPYL5 were selected as the top
pan-cancer candidates, while SOX17 was included in the screening because it is
highly expressed in ovarian cancer. Cisplatin response in the untreated cell lines
was compared to cisplatin response when these 12 genes were downregulated.
MRPS21 was the only gene to mildly but consistently increase cisplatin response in
all cell lines. ARC, ANXA9, C19orf33, and TSPLY5 showed synergy with cisplatin
treatment in two cell lines. Knockdown of MLF1 enhanced cisplatin inhibitory
power in FaDu cells, while knockdown of SOX17 improved cisplatin response in
SW-48 cells but not in Kuramochi cells (the HGSOC cell line).
6.4 Estimating differential methylation in cancer samples with
varying tumor purity (III)
Bulk tumor samples contain a mixture of cancerous and non-cancerous cells. Since
DNA methylation patterns are cell type specific, the methylome observed by
sequencing bulk samples is not the actual cancer methylome but a mixture of
cancer and normal methylomes. Our DMML method allows for accurate estimation
the tumor purity of a sample even without a matched control, and for performance
of differential methylation while taking into account the proportion of cancer cells
contained in each sample.
To test DMML’s tumor purity estimation we used Monte Carlo simulations to
generate tumor samples with varying levels of purity by mixing the WGBS methy-
lation profiles of cancer cell lines (K562 and HepG2) and a non-cancerous cell
line (GM12878). The accuracy of DMML purity estimation in low purity samples
exceeds that of existing methods (like Fisher’s exact test, MOABS, DSS, and
InfiniumPurify). Moreover, when reads are long enough to contain multiple CpG
sites, DMML can model co-methylation, further increasing the method’s accuracy
even in low coverage settings.
We then used DMML to perform differential methylation using five TBS HGSOC
samples and eight RRBS DLBCL samples, to test whether the method is effective
37
6 RESULTS
in different cancer types as well as when using different bisulfite sequencing
technologies. We compared DMML’s differential methylation calls with the ones
obtained from the other methods. Any two of these methods shared at least 91%
of the calls in the HGSOC dataset and 83% of the calls in the DLBCL dataset, but
DMML identified novel candidate differentially methylated sites even without a
control sample.
We also investigated the relation between methylation and gene expression in the
HGSOC dataset. Differential promoter methylation obtained with DMML showed
significant anticorrelation with differential expression. The anticorrelation between
methylation and expression was stronger when using differentially methylated sites
identified by DMML than those identified by other methods, suggesting once again
that DMML can more accurately recover the cancer methylation signal in bulk
samples.
6.5 Deconvoluted DNA methylation from 53 HGSOC samples (Un-
published data)
Heterogeneity in a tumor sample occurs at different levels, including inter-patient
heterogeneity, clonal evolution of the metastatic lesions driven by the microenviron-
ment, clonal evolution due to treatment exposure, and presence of non-cancerous
cells within the tumor mass. When analyzing DNA methylation data obtained from
bulk tumor samples, it is important to remember that the measured signal is mixture
of all these heterogeneous components.
In order to estimate how much patient variability, tumor location, sampling stage
and non-cancerous methylomes contribute to the measured methylation signal, we
used a modified version of the DMML method to deconvolute the methylome of
each sample into these four signals. According to our results shown in Figure 9,
patient-specific variability contributes the most to the methylation signal observed
in our samples. This is not surprising since DNA methylation patterns are affected
by environment and lifestyle. The sample location also affects the methylation
landscape quite considerably, especially in the five samples obtained during the
interval surgery. This could indicate that the selective pressure of the tumor
microenvironment at the metastatic site has an effect on the DNA methylation
patterns of the cancer cells.
6.6 Anduril 2 workflow framework (IV)
The use of Anduril 2 as a WMS to implement complex computational analyses
presents several advantages. The downloadable version of Anduril 2 already in-
38
6 RESULTS
Figure 9: Heatmap showing the contribution of deconvoluted DNA methylation signals
from 53 HGSOC samples. For each sample we estimated the proportion of the observed
methylation signal that is due to inter-patient variability (patient-specific component),
microenvironment-driven evolution (location-specific component), treatment stage at
which the sample was obtained (stage-specific component), and proportion of the signal
due to non-cancerous cells present in the sample (non-cancerous component).
cludes over 400 fully-documented components and functions, suitable for analyzing
all main NGS data types. New components can be easily developed by users,
providing a flexible and modular approach to design tailored workflows. Moreover,
component dependencies are automatically detected and independent parts are
39
6 RESULTS
parallelized to reduce processing time. If the workflow execution is interrupted or
the user wants to change parts of the script, only the modified/unprocessed portion
of the analysis needs to be re-run. This is particularly useful in bioinformatic
analyses processing large datasets with run-times of several days.
40
7 DISCUSSION
7 Discussion
Overcoming drug resistance in cancer is one of the major health challenges of the
21st century. Modern high throughput screening and sequencing technologies allow
us to explore the molecular mechanisms of cancer cells at an unprecedented level
of detail. This, in turn, aids the search for effective treatment strategies to overcome
drug resistance, as well as the discovery of biomarkers able to predict patients’
response to specific treatments. Moreover, the increased availability of large cancer
datasets in public repositories provides scientists with invaluable resources needed
to develop novel methods and software to analyze these data, giving rise to the field
of computational oncology [165].
The work presented in this thesis exploits high throughput technologies to develop
in vitro and in silico methods to investigate R-CHOP resistance in DLBCL, as well
as platinum resistance across multiple cancers, and more specifically in HGSOC.
In Publication I, we developed a drug screening method to measure the effect of non-
simultaneous drug combinations on cancer cell lines. We used this method to test
the ability of 60 epigenetic inhibitors to sensitize DLBCL cell lines to rituximab and
doxorubicin. Since the experiment was conducted in vitro, we could only investigate
how epigenetic reprogramming affects the cytotoxic effect of rituximab [75] and
not its ability to induce an immune response towards the cancer cells. HDAC and
HMT inhibitors were the most promising classes of compounds to sensitize DLBCL
cells to the toxic effect of rituximab and doxorubicin. The inhibitor concentration
required to achieve sensitization was lower than then one required to produce a
cytotoxic effect, suggesting that the use of epigenetic inhibitors as a pretreatment
before R-CHOP administration in clinical use could result in less side effects than
if they were used simultaneously. When investigating the mechanisms that might
be involved in the episensitization process, we uncovered that HDAC inhibitors
are able to dysregulate the homologous recombination (HR) pathway, which in
turn increases the efficacy of doxorubicin. This finding is in line with other studies
showing a link between HDAC inhibition and downregulation of HR[166, 167],
suggesting that HDAC inhibitors could be effective sensitizers for DNA damaging
treatements, including radiotherapy and chemotherapeutics like platinum-based
drugs.
The use of epigenetic inhibitors, alone or in combination with immunochemothera-
pies, has already shown promising results in several clinical trials [168]. However,
undesired side effects on non-cancerous tissues need to be minimized, and targeted
nanoscale delivery systems promise to improve the effectiveness of epigenetic
treatments in cancer [169].
41
7 DISCUSSION
In Publication II, we took advantage of public cancer data from The Cancer
Genome Atlas (TCGA) repository to investigate platinum resistance in a pan-cancer
context. We hypothesized that combining evidence from three different data levels
(expression, copy-number and DNA methylation) would enable us to overcome
spurious findings and identify genes related to platinum resistance. We performed
multi-omics data integration followed by survival analysis to identify genes whose
expression is related to platinum response. We identified 164 putative biomarkers
predictive of platinum response. Further investigation of these candidates could
lead to the development of a routine test to select which patients should receive
platinum treatment and which patients should be administered a different therapeutic
regimen. We also conducted an RNAi experiment to test whether the most promising
candidate biomarkers could also serve as therapeutic targets to overcome platinum
resistance, and we were able to find an association between the knockdown of
the gene encoding the mitochondrial ribosomal proteinMRPS21 and an increased
response to cisplatin in four different cancers. The exact function of MRPS21 is
still poorly understood, however other mitochondrial ribosomal proteins have been
reported to affect apoptotic mechanisms [170].
In Publication III, we developed a new bioinformatic software (DMML) to correctly
estimate differential methylation in cancer samples with varying tumor purity.
DMML can be used to analyze different types of bisulfite sequencing data (including
WGBS, RRBS, and TBS), outperforms existing alternatives, produces robust results
with and without non-cancerous controls, does not require prior purity estimates,
and can be used to to compute tumor purity and differential methylation in any
cancer type. Since DNA methylation has been shown to be a contributing factor
in the development of drug resistance, understanding the link between methylome
changes and drug response can help us to develop strategies to overcome treatment
failure. We also used a modified version of the DMML algorithm to estimate the
origin of inter-sample variability, as shown in our unpublished results. According
to our analysis, the differences in methylation patterns observed in our 53 HGSOC
samples are patient-specific, likely due to adaptation to diverse environmental
stimuli accumulated throughout the patient’s life, while changes driven by treatment
stage and sample location (i.e. the pressure to adapt to the microenvironment)
contribute less to the observed methylation signal.
In publication IV, we develop Anduril 2, a workflow management system able han-
dle the processing of big data obtained from high-throughput biology experiments.
This WMS has been used to implement the data analysis pipelines of publication
I and II, while the software described in publication III has been included as a
standard component in Anduril 2 and can easily be included in future pipelines.
Taken together, the methods presented in this thesis provide new tools to investigate
42
7 DISCUSSION
the molecular mechanisms behind cancer progression and drug resistance. More-
over, these methods are flexible and can be applied to biomedical investigation
beyond the cancer research field.
With new and continuously emerging high-throughput technologies, like single-
cell and third generation sequencing, computational oncology is a fast-evolving
field which will gain more and more of a central role in the fight against cancer
in the coming years. The integration of multi-level cancer data holds the key to
unlocking the complexity of drug resistance, generating new insights to improve
cancer treatment strategies, and even opening the doors to the era of precision
oncology [171].
43
Acknowledgements
Research is always the result of teamwork, and the studies described in this thesis
would not have been possible without the help and support of many people. First of
all, I want to thank professor Sampsa Hautaniemi for the guidance and mentorship
provided throughout my PhD. Thank you for the opportunity to dive into the world
of epigenetics, and for the chance to work on both in vitro and in silico projects,
which really helped me broaden my understanding of the scientific method. Thank
you also for allowing me to explore the academic environment beyond the research
field, and for giving me the opportunity to be a teaching assistant in your courses.
I am grateful to the members of my thesis committee, professor Merja Heinäniemi
and doctor Miina Ollikainen, for their time and their useful feedback and guidance
during our yearly meetings. I also thank my thesis pre-examiners, doctor Kirsi
Granberg and professor Tapio Pahikkala, for the time spent revising this work and
for the helpful suggestions on how to improve it.
I am very thankful to the Doctoral Programme in Biomedicine (DPBM) and the
Doctoral School in Health Sciences (DSHealth) for the financial and academic
support provided during my doctoral studies. In particular, I want to thank Alma,
Anita, Eeva, Nina, and Hannu for their support to the PhD student community. I
am also grateful for the funding provided by the Academy of Finland, the Sigrid
Jusélius Foundation, and the Cancer Foundation Finland, as well as the European
Union’s Horizon 2020 research and innovation programme under grant agreement
No 667403 for HERCULES), which were instrumental to conduct this work.
This work was carried out in the Systems Biology of Drug Resistance in Cancer
Laboratory at the Faculty of Medicine, University of Helsinki during 2014-2021. I
am grateful to all my colleagues at the lab for creating a very collaborative research
environment. In particular, I want to thank Julia for sharing the long hours and
all the ups and downs of the journey that led to Publication I. Ville, thank you
for being a very patient teacher and for your help with all sorts of coding issues.
Antti, thank you for being the statistics wizard of the lab! Rainer, thank you for
always reminding me that biology doesn’t function in binary mode, for helping
me to make sense of the long gene and pathway lists, and for hosting the annual
Salmari competition: making salmiakkikossu is a skill I wasn’t expecting to acquire
during my PhD, but which might prove useful while living in Finland. Tiia and
Karen, thank you for all the invaluable support you provide, the lab could not
function without you! A big thank you also to the past and current members of the
lab, including Alejandra, Alexandra, Amjad, Chengyu, Emilia, Ingrid, Kaiyang,
Kari, Katherine, Kristian, Jaana, Javier, Juha, Lilli, Mai, Melanie, Mikko, Oskari,
44
Acknowledgements
Pekka, Ping, Riku, Sanaz, Valeria, Veli-Matti, and Yilin. Thank you for all the
coding camps, the Anduril days, the Monday cookie breaks, and the interesting
discussions.
I am really thankful to all the research collaborators I had the opportunity to
work with and learn from throughout my PhD. I thank professor Sirpa Leppä
and doctor Suvi-Katri Leivonen for the interesting collaboration in the epigenetic
reprogramming project. I also want to thank Anne Aarnio and Marika Tuukkanen
for teaching me the basics of wet lab work. Professor Krister Wennerberg, Laura
Turunen, Jani Saarela, Swapnil Potdar, and the team at the High Throughput
Biomedicine Unit in FIMM, thank you for your help in implementing the drug
screenings and in analyzing and interpreting the results. I am grateful to professor
Liisa Kauppi and doctor Manuela Tumiati for their contribution in the epigenetic
reprogramming study. A big thank you goes also to doctor Laura Lehtinen for
helping me with the RNAi experiment. To professor Benno Schwikowski, thank
you for welcoming me in your lab during my visit at Institute Pasteur, and for the
many interesting discussions. I also want to thank all the people in HERCULES
and ONCOSYS for providing a very stimulating research environment.
The research community at the University of Helsinki is very dynamic, and I am
really lucky to have been able to collaborate with many scientists interested in
bridging the gap between science and society. I want to thank Shishir, Elina, Mridul,
Tuomo, Kul, Siggi, Hannu, and the whole TEDxHelsinkiUniversity team. I had so
much fun working with you to bring this event to life, and I found many friends in
the process. A big thank you also to all the past and current members of The Science
Basement, and in particular to Lea, Ekaterina, Erika, Stephany, Katja, Eleanna,
Alok, Susanna, and Vesa, who have really been driving this group! You are all
incredibly creative and resourceful, and I always have a great time experimenting
with you different ways of communicating science. I am also grateful to the
members of the DPBM student council, for bringing our community closer.
Throughout my PhD, I really enjoyed the opportunity to design new courses and I
had a great time working with experts that were ready to share their knowledge. I
thank all the students I had the pleasure to work with, as well as Miina Ollikainen,
Nina Kaminen-Ahola, Vanessa Fuller, Brendan Battersby, Vincenzo Cerullo, Oana
Velcu-Laitinen, Tarja Kahkonen, Asta Pirskanen, Jussi Heinonkoski, Hannu Sariola,
and all other people I collaborated with in different teaching activities. A special
thank you goes to Vincenzo and Vanessa, who have been friends and mentors in
these last few years.
A huge thank you to Jonna Katajisto and Olli Silvennoinen, for taking a chance on
me and welcoming me into the HiLIFE team. I also want to thank Pia Runeberg,
45
for helping me get started with Y Science.
I definitely would not have survived this roller-coaster without good friends, always
there to share the happy times as well as the hard one. To Ceci, Greta, Ausi, Pepe,
Davide, Sid, Leo, and Masha, thank you for being a constant source of joy and
laughter, for pampering me every time I come back to Italy, and for always being
close despite the distance. To my Italians here in Helsinki, Elisabetta, Rossella,
Sara, Manlio, and Stephany, thank you for always being ready to discuss and cook
Italian food, and for helping me be a little less homesick (especially during this
pandemic). To Evisa, Ville, and Lea, thank you for keeping me sane (as much as
my three neurons allowed) and for listening to my endless talks (especially those
about epigenetics). To Tiia, thank you for understanding my color-coordination
OCD and for getting me to do at least a bit of exercise during my PhD years. I
also want to thank Dima for being a constant source of support and a lunch buddy
since our MSc studies up to the writing of this thesis. Mårten, thank you for being
my bridge to the Swedish-speaking community in Helsinki, for the very colorful
risottos, and for bringing Fabian into my life. To Anja and Antonia Maria, thank
you for sharing the best flat in Helsinki, and for all the awesome parties. Axel,
thank you for allowing me to join you and Fabian in your Survivor evenings. To all
the other wonderful friends I found here in Helsinki, thank you for all the hugs and
the fun times.
Together with my friends, my family has also been a source of love and endless
support. I want to thank Outi, Kasper, Jaffi, Mario, Robin, Vicky, and Liam for
welcoming me among the Nybergs. To my whole family in Italy (which also
includes the Carollos and the El Filaouis), thank you for being big, warm, and
cuddly.
I thank my parents, Mariolina and Ottavio, for always being just a phone call away,
for encouraging me to jump, being there to catch me if I fall, and helping me get
back up again. Thank you for all the opportunities you gave me, and for the loving
family you built. I am grateful to you in more ways than I can count. Lucy, my
sister, my friend, my travel companion. I am proud of the woman you have become,
and I know we can always count on each other.
Fabian, thank you for being with me throughout this whole roller-coaster, for being
kind, patient, positive, brave, and loving, for letting me sing to my heart’s content,
and for pushing me to explore the world in unexpected directions.
Chiara Facciotto
Helsinki, 2021
46
REFERENCES
Page(s)References
[1] National Cancer Institute. (2015) What is cancer? (https://www.cancer.gov/
about-cancer/understanding/what-is-cancer#types). 1
[2] World Health Organization. (2018) Latest global cancer data (https://www.who.int/
cancer/PRGlobocanFinal.pdf). 1
[3] World Health Organization. (2018) Cancer - key facts (https://www.who.int/
news-room/fact-sheets/detail/cancer). 1
[4] You, J. S & Jones, P. A. (2012) Cancer genetics and epigenetics: two sides of the same
coin? Cancer Cell 22, 9–20. 1
[5] Muir, P, Li, S, Lou, S, Wang, D, Spakowicz, D. J, Salichos, L, Zhang, J, Weinstock,
G. M, Isaacs, F, Rozowsky, J, & Gerstein, M. (2016) The real cost of sequencing: scaling
computation to keep pace with data generation. Genome Biol 17, 53. 1
[6] Friedberg, J. W. (2011) Relapsed/refractory diffuse large B-cell lymphoma. Hematology
Am Soc Hematol Educ Program 2011, 498–505. 1, 13
[7] Torre, L. A, Trabert, B, DeSantis, C. E, Miller, K. D, Samimi, G, Runowicz, C. D, Gaudet,
M. M, Jemal, A, & Siegel, R. L. (2018) Ovarian cancer statistics, 2018. CA Cancer J Clin
68, 284–296. 1
[8] Aunan, J. R, Cho, W. C, & Søreide, K. (2017) The Biology of Aging and Cancer: A Brief
Overview of Shared and Divergent Molecular Hallmarks. Aging Dis 8, 628–642. 3
[9] Wu, S, Zhu, W, Thompson, P, & Hannun, Y. A. (2018) Evaluating intrinsic and non-intrinsic
cancer risk factors. Nat Commun 9, 3490. 3
[10] Renkawitz, R. (2006) Transcription Factors and Regulation of Gene Expression. (Springer
Berlin Heidelberg, Berlin, Heidelberg), pp. 1886–1890. 3
[11] Jaenisch, R & Bird, A. (2003) Epigenetic regulation of gene expression: how the genome
integrates intrinsic and environmental signals. Nat Genet 33 Suppl, 245–254. 4
[12] Gibney, E. R & Nolan, C. M. (2010) Epigenetics and gene expression. Heredity (Edinb)
105, 4–13. 4
[13] Hannon, G. J. (2002) RNA interference. Nature 418, 244–251. 4, 6
[14] Lyko, F. (2018) The DNA methyltransferase family: a versatile toolkit for epigenetic
regulation. Nat Rev Genet 19, 81–92. 4
[15] Jones, P. A. (2012) Functions of DNA methylation: islands, start sites, gene bodies and
beyond. Nat Rev Genet 13, 484–492. 4
[16] Illingworth, R. S & Bird, A. P. (2009) CpG islands–’a rough guide’. FEBS Lett 583,
1713–1720. 4
47
REFERENCES
[17] Yin, Y, Morgunova, E, Jolma, A, Kaasinen, E, Sahu, B, Khund-Sayeed, S, Das, P. K, Kivioja,
T, Dave, K, Zhong, F, Nitta, K. R, Taipale, M, Popov, A, Ginno, P. A, Domcke, S, Yan,
J, Schübeler, D, Vinson, C, & Taipale, J. (2017) Impact of cytosine methylation on DNA
binding specificities of human transcription factors. Science 356.4
[18] Medvedeva, Y. A, Khamis, A. M, Kulakovskiy, I. V, Ba-Alawi, W, Bhuyan, M. S, Kawaji,
H, Lassmann, T, Harbers, M, Forrest, A. R, & Bajic, V. B. (2014) Effects of cytosine
methylation on transcription factor binding sites. BMC Genomics 15, 119.4
[19] Shukla, S, Kavak, E, Gregory, M, Imashimizu, M, Shutinoski, B, Kashlev, M, Oberdoerffer,
P, Sandberg, R, & Oberdoerffer, S. (2011) CTCF-promoted RNA polymerase II pausing
links DNA methylation to splicing. Nature 479, 74–79.4
[20] Mariño-Ramírez, L, Kann, M. G, Shoemaker, B. A, & Landsman, D. (2005) Histone
structure and nucleosome stability. Expert Rev Proteomics 2, 719–729.4
[21] Lu, D. (2013) Epigenetic modification enzymes: catalytic mechanisms and inhibitors. Acta
Pharmaceutica Sinica B 3, 141 – 149.4
[22] Schuettengruber, B, Chourrout, D, Vervoort, M, Leblanc, B, & Cavalli, G. (2007) Genome
regulation by polycomb and trithorax proteins. Cell 128, 735–745.6
[23] Spicuglia, S & Vanhille, L. (2012) Chromatin signatures of active enhancers. Nucleus 3,
126–131.6
[24] Huntzinger, E & Izaurralde, E. (2011) Gene silencing by microRNAs: contributions of
translational repression and mRNA decay. Nat Rev Genet 12, 99–110.6
[25] Mohr, S. E, Smith, J. A, Shamu, C. E, Neum?ller, R. A, & Perrimon, N. (2014) RNAi
screening comes of age: improved techniques and complementary approaches. Nat. Rev.
Mol. Cell Biol. 15, 591–600.6, 20
[26] Hanahan, D & Weinberg, R. A. (2011) Hallmarks of cancer: the next generation. Cell 144,
646–674.7
[27] Garber, J. E & Offit, K. (2005) Hereditary cancer predisposition syndromes. J Clin Oncol
23, 276–292.7
[28] Negrini, S, Gorgoulis, V. G, & Halazonetis, T. D. (2010) Genomic instability–an evolving
hallmark of cancer. Nat Rev Mol Cell Biol 11, 220–228.8
[29] Loewe, L & Hill, W. G. (2010) The population genetics of mutations: good, bad and
indifferent. Philos Trans R Soc Lond B Biol Sci 365, 1153–1167.8
[30] Martínez-Jiménez, F, Muiños, F, Sentís, I, Deu-Pons, J, Reyes-Salazar, I, Arnedo-Pac, C,
Mularoni, L, Pich, O, Bonet, J, Kranas, H, Gonzalez-Perez, A, & Lopez-Bigas, N. (2020) A
compendium of mutational cancer driver genes. Nat Rev Cancer 20, 555–572.8
[31] Huang, D, Duan, H, Huang, H, Tong, X, Han, Y, Ru, G, Qu, L, Shou, C, & Zhao, Z. (2016)
Cisplatin resistance in gastric cancer cells is associated with HER2 upregulation-induced
epithelial-mesenchymal transition. Sci Rep 6, 20502.8
48
REFERENCES
[32] Gorski, J. W, Ueland, F. R, & Kolesar, J. M. (2020) CCNE1 Amplification as a Predictive
Biomarker of Chemotherapy Resistance in Epithelial Ovarian Cancer. Diagnostics (Basel)
10. 8
[33] Ilango, S, Paital, B, Jayachandran, P, Padma, P. R, & Nirmaladevi, R. (2020) Epigenetic
alterations in cancer. Front Biosci (Landmark Ed) 25, 1058–1109. 8
[34] Kulis, M & Esteller, M. (2010) DNA methylation and cancer. Adv Genet 70, 27–56. 8
[35] Esteller, M, Silva, J. M, Dominguez, G, Bonilla, F, Matias-Guiu, X, Lerma, E, Bussaglia, E,
Prat, J, Harkes, I. C, Repasky, E. A, Gabrielson, E, Schutte, M, Baylin, S. B, & Herman,
J. G. (2000) Promoter hypermethylation and BRCA1 inactivation in sporadic breast and
ovarian tumors. J Natl Cancer Inst 92, 564–569. 8
[36] Zhao, Z & Shilatifard, A. (2019) Epigenetic modifications of histones in cancer. Genome
Biol 20, 245. 8
[37] Li, B & Chng, W. J. (2019) EZH2 abnormalities in lymphoid malignancies: underlying
mechanisms and therapeutic implications. J Hematol Oncol 12, 118. 8, 10
[38] Huffman, D. M, Grizzle, W. E, Bamman, M. M, Kim, J. S, Eltoum, I. A, Elgavish, A, &
Nagy, T. R. (2007) SIRT1 is significantly elevated in mouse and human prostate cancer.
Cancer Res 67, 6612–6618. 8
[39] Farman, F. U, Iqbal, M, Azam, M, & Saeed, M. (2018) Nucleosomes positioning around
transcriptional start site of tumor suppressor (Rbl2/p130) gene in breast cancer. Mol Biol
Rep 45, 185–194. 8
[40] Anastasiadou, E, Jacob, L. S, & Slack, F. J. (2018) Non-coding RNA networks in cancer.
Nat Rev Cancer 18, 5–18. 8
[41] McGranahan, N & Swanton, C. (2017) Clonal Heterogeneity and Tumor Evolution: Past,
Present, and the Future. Cell 168, 613–628. 9
[42] Hinohara, K & Polyak, K. (2019) Intratumoral Heterogeneity: More Than Just Mutations.
Trends Cell Biol 29, 569–579. 9
[43] Baghban, R, Roshangar, L, Jahanban-Esfahlan, R, Seidi, K, Ebrahimi-Kalan, A, Jaymand,
M, Kolahian, S, Javaheri, T, & Zare, P. (2020) Tumor microenvironment complexity and
therapeutic implications at a glance. Cell Commun Signal 18, 59. 9
[44] Lugano, R, Ramachandran, M, & Dimberg, A. (2020) Tumor angiogenesis: causes,
consequences, challenges and opportunities. Cell Mol Life Sci 77, 1745–1770. 9
[45] Whiteside, T. L & Parmiani, G. (1994) Tumor-infiltrating lymphocytes: their phenotype,
functions and clinical use. Cancer Immunol Immunother 39, 15–21. 9
49
REFERENCES
[46] Sahai, E, Astsaturov, I, Cukierman, E, DeNardo, D. G, Egeblad, M, Evans, R. M, Fearon, D,
Greten, F. R, Hingorani, S. R, Hunter, T, Hynes, R. O, Jain, R. K, Janowitz, T, Jorgensen, C,
Kimmelman, A. C, Kolonin, M. G, Maki, R. G, Powers, R. S, Puré, E, Ramirez, D. C, Scherz-
Shouval, R, Sherman, M. H, Stewart, S, Tlsty, T. D, Tuveson, D. A, Watt, F. M, Weaver,
V, Weeraratna, A. T, & Werb, Z. (2020) A framework for advancing our understanding of
cancer-associated fibroblasts. Nat Rev Cancer 20, 174–186.9
[47] Chaplin, D. D. (2010) Overview of the immune response. J Allergy Clin Immunol 125,
3–23.10
[48] Yuen, G. J, Demissie, E, & Pillai, S. (2016) B lymphocytes and cancer: a love-hate
relationship. Trends Cancer 2, 747–757.10
[49] Armitage, J. O, Gascoyne, R. D, Lunning, M. A, & Cavalli, F. (2017) Non-Hodgkin
lymphoma. Lancet 390, 298–310.10
[50] Bray, F, Ferlay, J, Soerjomataram, I, Siegel, R. L, Torre, L. A, & Jemal, A. (2018) Global
cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36
cancers in 185 countries. CA Cancer J Clin 68, 394–424.10
[51] Li, S, Young, K. H, & Medeiros, L. J. (2018) Diffuse large b-cell lymphoma. Pathology 50,
74 – 87. 50th anniversary review issue.10
[52] Smith, A, Howell, D, Patmore, R, Jack, A, & Roman, E. (2011) Incidence of haematological
malignancy by sub-type: a report from the Haematological Malignancy Research Network.
Br J Cancer 105, 1684–1692.10
[53] Jiang, Y & Melnick, A. (2015) The epigenetic basis of diffuse large B-cell lymphoma.
Semin Hematol 52, 86–96.10
[54] Momenimovahed, Z, Tiznobaik, A, Taheri, S, & Salehiniya, H. (2019) Ovarian cancer in
the world: epidemiology and risk factors. Int J Womens Health 11, 287–299.11
[55] Zhang, S, Dolgalev, I, Zhang, T, Ran, H, Levine, D. A, & Neel, B. G. (2019) Both
fallopian tube and ovarian surface epithelium are cells-of-origin for high-grade serous
ovarian carcinoma. Nat Commun 10, 5367.11
[56] Bell, D, Berchuck, A, Birrer, M, Chien, J, Cramer, D, Dao, F, Dhir, R, DiSaia, P, Gabra,
H, Glenn, P, Godwin, A, Gross, J, Hartmann, L, Huang, M, Huntsman, D, Iacocca, M,
Imielinski, M, Kalloger, S, Karlan, B, Levine, D, Mills, G, Morrison, C, Mutch, D, Olvera,
N, Orsulic, S, Park, K, Petrelli, N, Rabeno, B, Rader, J, Sikic, B, Smith-McCune, K, Sood,
A, Bowtell, D, Penny, R, Testa, J, Chang, K, Dinh, H, Drummond, J, Fowler, G, Gunaratne,
P, Hawes, A, Kovar, C, Lewis, L, et al. (2011) Integrated genomic analyses of ovarian
carcinoma. Nature 474, 609–615.11
[57] Lisio, M. A, Fu, L, Goyeneche, A, Gao, Z. H, & Telleria, C. (2019) High-Grade Serous
Ovarian Cancer: Basic Sciences, Clinical and Therapeutic Standpoints. Int J Mol Sci 20.11, 14
50
REFERENCES
[58] Buys, S. S, Partridge, E, Black, A, Johnson, C. C, Lamerato, L, Isaacs, C, Reding, D. J,
Greenlee, R. T, Yokochi, L. A, Kessel, B, Crawford, E. D, Church, T. R, Andriole, G. L,
Weissfeld, J. L, Fouad, M. N, Chia, D, O’Brien, B, Ragard, L. R, Clapp, J. D, Rathmell,
J. M, Riley, T. L, Hartge, P, Pinsky, P. F, Zhu, C. S, Izmirlian, G, et al. (2011) Effect of
screening on ovarian cancer mortality: the Prostate, Lung, Colorectal and Ovarian (PLCO)
Cancer Screening Randomized Controlled Trial. JAMA 305, 2295–2303. 11
[59] Narod, S. (2016) Can advanced-stage ovarian cancer be cured? Nat Rev Clin Oncol 13,
255–261. 11
[60] Wyld, L, Audisio, R. A, & Poston, G. J. (2015) The evolution of cancer surgery and future
perspectives. Nat Rev Clin Oncol 12, 115–124. 11
[61] Chabner, B. A & Roberts, T. G. (2005) Timeline: Chemotherapy and the war on cancer.
Nat Rev Cancer 5, 65–72. 12
[62] Weiner, L. M, Dhodapkar, M. V, & Ferrone, S. (2009) Monoclonal antibodies for cancer
immunotherapy. Lancet 373, 1033–1040. 12
[63] Robert, C. (2020) A decade of immune-checkpoint inhibitors in cancer therapy. Nat
Commun 11, 3801. 12
[64] Berraondo, P, Sanmamed, M. F, Ochoa, M. C, Etxeberria, I, Aznar, M. A, Pérez-Gracia,
J. L, Rodríguez-Ruiz, M. E, Ponz-Sarvise, M, Castañón, E, & Melero, I. (2019) Cytokines
in clinical cancer immunotherapy. Br J Cancer 120, 6–15. 12
[65] Hollingsworth, R. E & Jansen, K. (2019) Turning the corner on therapeutic cancer vaccines.
NPJ Vaccines 4, 7. 12
[66] Mohanty, R, Chowdhury, C. R, Arega, S, Sen, P, Ganguly, P, & Ganguly, N. (2019) CAR T
cell therapy: A new era for cancer treatment (Review). Oncol Rep 42, 2183–2195. 12
[67] Nowakowski, G, Frontzek, F, & Schmitz, N. (2019) in Aggressive Lymphomas. Hematologic
Malignancies, eds. Lenz, G & Salles, G. (Springer, Cham), pp. 145–155. 13
[68] Pierpont, T. M, Limper, C. B, & Richards, K. L. (2018) Past, Present, and Future of
Rituximab-The World’s First Oncology Monoclonal Antibody Therapy. Front Oncol 8, 163.
13
[69] Dotan, E, Aggarwal, C, & Smith, M. R. (2010) Impact of Rituximab (Rituxan) on the
Treatment of B-Cell Non-Hodgkin’s Lymphoma. P T 35, 148–157. 13
[70] Lee, L, Crump, M, Khor, S, Hoch, J. S, Luo, J, Bremner, K, Krahn, M, & Hodgson, D. C.
(2012) Impact of rituximab on treatment outcomes of patients with diffuse large b-cell
lymphoma: a population-based analysis. Br J Haematol 158, 481–488. 13
[71] Vose, J. M, Link, B. K, Grossbard, M. L, Czuczman, M, Grillo-Lopez, A, Gilman, P, Lowe,
A, Kunkel, L. A, & Fisher, R. I. (2001) Phase II study of rituximab in combination with chop
chemotherapy in patients with previously untreated, aggressive non-Hodgkin’s lymphoma.
J Clin Oncol 19, 389–397. 13
51
REFERENCES
[72] Sehn, L. H, Donaldson, J, Chhanabhai, M, Fitzgerald, C, Gill, K, Klasa, R, MacPherson,
N, O’Reilly, S, Spinelli, J. J, Sutherland, J, Wilson, K. S, Gascoyne, R. D, & Connors,
J. M. (2005) Introduction of combined CHOP plus rituximab therapy dramatically improved
outcome of diffuse large B-cell lymphoma in British Columbia. J Clin Oncol 23, 5027–5033.
13
[73] Pfreundschuh, M, Trümper, L, Osterborg, A, Pettengell, R, Trneny, M, Imrie, K, Ma, D,
Gill, D, Walewski, J, Zinzani, P. L, Stahel, R, Kvaloy, S, Shpilberg, O, Jaeger, U, Hansen,
M, Lehtinen, T, López-Guillermo, A, Corrado, C, Scheliga, A, Milpied, N, Mendila, M,
Rashford, M, Kuhnt, E, & Loeffler, M. (2006) CHOP-like chemotherapy plus rituximab
versus CHOP-like chemotherapy alone in young patients with good-prognosis diffuse large-
B-cell lymphoma: a randomised controlled trial by the MabThera International Trial (MInT)
Group. Lancet Oncol 7, 379–391.13
[74] Weiner, G. J. (2010) Rituximab: mechanism of action. Semin Hematol 47, 115–123.13
[75] Singh, V, Gupta, D, Arora, R, Tripathi, R. P, Almasan, A, & Macklis, R. M. (2014) Surface
levels of CD20 determine anti-CD20 antibodies mediated cell death in vitro. PLoS One 9,
e111113.13, 41
[76] Tacar, O, Sriamornsak, P, & Dass, C. R. (2013) Doxorubicin: an update on anticancer
molecular action, toxicity and novel drug delivery systems. J Pharm Pharmacol 65, 157–
170.13
[77] Thorn, C. F, Oshiro, C, Marsh, S, Hernandez-Boussard, T, McLeod, H, Klein, T. E, &
Altman, R. B. (2011) Doxorubicin pathways: pharmacodynamics and adverse effects.
Pharmacogenet Genomics 21, 440–446.13
[78] Lisio, M. A, Fu, L, Goyeneche, A, Gao, Z. H, & Telleria, C. (2019) High-Grade Serous
Ovarian Cancer: Basic Sciences, Clinical and Therapeutic Standpoints. Int J Mol Sci 20.13
[79] Vergote, I, Tropé, C. G, Amant, F, Kristensen, G. B, Ehlen, T, Johnson, N, Verheijen, R. H,
van der Burg, M. E, Lacave, A. J, Panici, P. B, Kenter, G. G, Casado, A, Mendiola, C, Coens,
C, Verleye, L, Stuart, G. C, Pecorelli, S, & Reed, N. S. (2010) Neoadjuvant chemotherapy
or primary surgery in stage iiic or iv ovarian cancer. New England Journal of Medicine 363,
943–953. PMID: 20818904.13
[80] Dilruba, S & Kalayda, G. V. (2016) Platinum-based drugs: past, present and future. Cancer
Chemother Pharmacol 77, 1103–1124.14
[81] Dasari, S & Tchounwou, P. B. (2014) Cisplatin in cancer therapy: molecular mechanisms
of action. Eur J Pharmacol 740, 364–378.14
[82] Johnstone, T. C, Park, G. Y, & Lippard, S. J. (2014) Understanding and improving platinum
anticancer drugs–phenanthriplatin. Anticancer Res 34, 471–476.14
[83] Institute, N. C. (2014) The "accidental" cure—platinum-based treatment for cancer: The
discovery of cisplatin.14
[84] Minassian, L. M, Cotechini, T, Huitema, E, & Graham, C. H. (2019) Hypoxia-Induced
Resistance to Chemotherapy in Cancer. Adv Exp Med Biol 1136, 123–139.16
52
REFERENCES
[85] Tomita, A. (2016) Genetic and Epigenetic Modulation of CD20 Expression in B-Cell
Malignancies: Molecular Mechanisms and Significance to Rituximab Resistance. J Clin
Exp Hematop 56, 89–99. 16
[86] Damia, G & Broggini, M. (2019) Platinum Resistance in Ovarian Cancer: Role of DNA
Repair. Cancers (Basel) 11. 16
[87] Spencer, D. M, Bilardi, R. A, Koch, T. H, Post, G. C, Nafie, J. W, Kimura, K.-I, Cutts, S. M,
& Phillips, D. R. (2008) Dna repair in response to anthracycline–dna adducts: A role for both
homologous recombination and nucleotide excision repair. Mutation Research/Fundamental
and Molecular Mechanisms of Mutagenesis 638, 110–121. 16
[88] Buonaguro, F. M, Caposio, P, Tornesello, M. L, De Re, V, & Franco, R. (2019) Cancer
Diagnostic and Predictive Biomarkers 2018. Biomed Res Int 2019, 3879015. 16
[89] Coppedè, F, Lopomo, A, Spisni, R, &Migliore, L. (2014) Genetic and epigenetic biomarkers
for diagnosis, prognosis and treatment of colorectal cancer. World J Gastroenterol 20, 943–
956. 16
[90] van IJzendoorn, D. G. P, Szuhai, K, Briaire-de Bruijn, I. H, Kostine, M, Kuijjer, M. L,
& Bovée, J. V. M. G. (2019) Machine learning analysis of gene expression data reveals
novel diagnostic and prognostic biomarkers and identifies therapeutic targets for soft tissue
sarcomas. PLoS Comput Biol 15, e1006826. 16
[91] Li, D & Chan, D. W. (2014) Proteomic cancer biomarkers from discovery to approval: it’s
worth the effort. Expert Rev Proteomics 11, 135–136. 16
[92] Dregely, I, Prezzi, D, Kelly-Morland, C, Roccia, E, Neji, R, & Goh, V. (2018) Imaging
biomarkers in oncology: Basics and application to MRI. J Magn Reson Imaging 48, 13–26.
16
[93] McCabe, N, Turner, N. C, Lord, C. J, Kluzek, K, Bialkowska, A, Swift, S, Giavara, S,
O’Connor, M. J, Tutt, A. N, Zdzienicka, M. Z, Smith, G. C, & Ashworth, A. (2006)
Deficiency in the repair of DNA damage by homologous recombination and sensitivity to
poly(ADP-ribose) polymerase inhibition. Cancer Res 66, 8109–8115. 16
[94] Ganesan, A, Arimondo, P. B, Rots, M. G, Jeronimo, C, & Berdasco, M. (2019) The timeline
of epigenetic drug discovery: from reality to dreams. Clin Epigenetics 11, 174. 17
[95] Derissen, E. J, Beijnen, J. H, & Schellens, J. H. (2013) Concise drug review: azacitidine
and decitabine. Oncologist 18, 619–624. 17
[96] Lyko, F & Brown, R. (2005) DNA methyltransferase inhibitors and the development of
epigenetic cancer therapies. J Natl Cancer Inst 97, 1498–1506. 17
[97] Amato, R. J. (2007) Inhibition of DNA methylation by antisense oligonucleotide MG98 as
cancer therapy. Clin Genitourin Cancer 5, 422–426. 17
[98] Tollefsbol, T. O, ed. (2018) in Epigenetics in Human Disease (Second Edition), Translational
Epigenetics. (Academic Press) Vol. 6, Second edition edition, p. ii. 17
53
REFERENCES
[99] Sarkozy, C, Morschhauser, F, Dubois, S, Molina, T, Michot, J. M, Cullières-Dartigues, P,
Suttle, B, Karlin, L, Le Gouill, S, Picquenot, J. M, Dubois, R, Tilly, H, Herbaux, C, Jardin,
F, Salles, G, & Ribrag, V. (2020) A LYSA Phase Ib Study of Tazemetostat (EPZ-6438) plus
R-CHOP in Patients with Newly Diagnosed Diffuse Large B-Cell Lymphoma (DLBCL)
with Poor Prognosis Features. Clin Cancer Res 26, 3145–3153.17
[100] Bhatla, T, Wang, J, Morrison, D. J, Raetz, E. A, Burke, M. J, Brown, P, & Carroll, W. L.
(2012) Epigenetic reprogramming reverses the relapse-specific gene expression signature
and restores chemosensitivity in childhood B-lymphoblastic leukemia. Blood 119, 5201–
5210.18
[101] Cronk, D. (2013) in Drug Discovery and Development (Second Edition), eds. Hill, R &
Rang, H. (Churchill Livingstone), Second edition edition, pp. 95 – 117.19
[102] Zhang, X. D. (2011) Illustration of ssmd, z score, ssmd*, z* score, and t statistic for hit
selection in rnai high-throughput screens. Journal of Biomolecular Screening 16, 775–785.
PMID: 21515799.20
[103] Ritz, C, Baty, F, Streibig, J. C, & Gerhard, D. (2015) Dose-response analysis using r. PLOS
ONE 10.20
[104] Yadav, B, Pemovska, T, Szwajda, A, Kulesskiy, E, Kontro, M, Karjalainen, R, Majumder,
M. M, Malani, D, Murumägi, A, Knowles, J, Porkka, K, Heckman, C, Kallioniemi, O,
Wennerberg, K, & Aittokallio, T. (2014) Quantitative scoring of differential drug sensitivity
for individually optimized anticancer therapies. Sci Rep 4, 5193.21
[105] Potdar, S, Ianevski, A, Mpindi, J. P, Bychkov, D, Fiere, C, Ianevski, P, Yadav, B,
Wennerberg, K, Aittokallio, T, Kallioniemi, O, Saarela, J, & Östling, P. (2020) Breeze: an
integrated quality control and data analysis application for high-throughput drug screening.
Bioinformatics 36, 3602–3604.21
[106] Sanger, F, Nicklen, S, Coulson, A. R, Sanger, F, Nicklen, S, & Coulsen, A. R. (1992) DNA
sequencing with chain-terminating inhibitors. 1977. Biotechnology 24, 104–108.21
[107] Brewster, J. L, Beason, K. B, Eckdahl, T. T, & Evans, I. M. (2004) The microarray
revolution: Perspectives from educators. Biochem Mol Biol Educ 32, 217–227.21
[108] Hoheisel, J. D. (2006) Microarray technology: beyond transcript profiling and genotype
analysis. Nat Rev Genet 7, 200–210.21
[109] Goodwin, S, McPherson, J. D, & McCombie, W. R. (2016) Coming of age: ten years of
next-generation sequencing technologies. Nat Rev Genet 17, 333–351.21
[110] Stark, R, Grzelak, M, & Hadfield, J. (2019) RNA sequencing: the teenage years. Nat Rev
Genet 20, 631–656.21
[111] van Dijk, E. L, Jaszczyszyn, Y, Naquin, D, & Thermes, C. (2018) The Third Revolution in
Sequencing Technology. Trends Genet 34, 666–681.22
[112] Wang, Y, Yang, Q, & Wang, Z. (2014) The evolution of nanopore sequencing. Front Genet
5, 449.22
54
REFERENCES
[113] Hwang, B, Lee, J. H, & Bang, D. (2018) Single-cell RNA sequencing technologies and
bioinformatics pipelines. Exp Mol Med 50, 1–14. 22
[114] Frommer, M, McDonald, L. E, Millar, D. S, Collis, C. M, Watt, F, Grigg, G. W, Molloy,
P. L, & Paul, C. L. (1992) A genomic sequencing protocol that yields a positive display
of 5-methylcytosine residues in individual DNA strands. Proc Natl Acad Sci U S A 89,
1827–1831. 23
[115] Meissner, A, Gnirke, A, Bell, G. W, Ramsahoye, B, Lander, E. S, & Jaenisch, R.
(2005) Reduced representation bisulfite sequencing for comparative high-resolution DNA
methylation analysis. Nucleic Acids Res 33, 5868–5877. 23
[116] (2012) Agilent SureSelect Human Methyl-Seq for the Quantitative Analysis of DNA
Methylation with Single-Base Resolution. Agilent Technical Overview, Publication Number
5991-0166EN. 23
[117] Smallwood, S. A, Lee, H. J, Angermueller, C, Krueger, F, Saadeh, H, Peat, J, Andrews,
S. R, Stegle, O, Reik, W, & Kelsey, G. (2014) Single-cell genome-wide bisulfite sequencing
for assessing epigenetic heterogeneity. Nat Methods 11, 817–820. 23
[118] Andrews, S. (2010) FastQC: a quality control tool for high throughput sequence data.
(http://www.bioinformatics.babraham.ac.uk/projects/fastqc). 24
[119] Bolger, A. M, Lohse, M, & Usadel, B. (2014) Trimmomatic: a flexible trimmer for Illumina
sequence data. Bioinformatics 30, 2114–2120. 25
[120] Krueger, F. (2010) Trim Galore! (http://www.bioinformatics.babraham.ac.uk/
projects/trim_galore/). 25
[121] Dobin, A, Davis, C. A, Schlesinger, F, Drenkow, J, Zaleski, C, Jha, S, Batut, P, Chaisson,
M, & Gingeras, T. R. (2013) STAR: ultrafast universal RNA-seq aligner. Bioinformatics
29, 15–21. 25
[122] Krueger, F & Andrews, S. R. (2011) Bismark: a flexible aligner and methylation caller for
Bisulfite-Seq applications. Bioinformatics 27, 1571–1572. 25
[123] Xi, Y & Li, W. (2009) BSMAP: whole genome bisulfite sequence MAPping program. BMC
Bioinformatics 10, 232. 25
[124] Roberts, A & Pachter, L. (2013) Streaming fragment assignment for real-time analysis of
sequencing experiments. Nat Methods 10, 71–73. 25
[125] Bray, N. L, Pimentel, H, Melsted, P, & Pachter, L. (2016) Near-optimal probabilistic
RNA-seq quantification. Nat Biotechnol 34, 525–527. 25
[126] Barturen, G, Rueda, A, Oliver, J. L, & Hackenberg, M. (2013) MethylExtract: High-Quality
methylation maps and SNV calling from whole genome bisulfite sequencing data. F1000Res
2, 217. 25
[127] Love, M. I, Huber, W, & Anders, S. (2014) Moderated estimation of fold change and
dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550. 25
55
REFERENCES
[128] Assenov, Y, Müller, F, Lutsik, P, Walter, J, Lengauer, T, & Bock, C. (2014) Comprehensive
analysis of DNA methylation data with RnBeads. Nat Methods 11, 1138–1140.26
[129] Müller, F, Scherer, M, Assenov, Y, Lutsik, P, Walter, J, Lengauer, T, & Bock, C. (2019)
RnBeads 2.0: comprehensive analysis of DNA methylation data. Genome Biol 20, 55.26
[130] Sun, D, Xi, Y, Rodriguez, B, Park, H. J, Tong, P, Meong, M, Goodell, M. A, & Li, W. (2014)
MOABS: model based analysis of bisulfite sequencing data. Genome Biol 15, R38.26
[131] Akalin, A, Kormaksson, M, Li, S, Garrett-Bakelman, F. E, Figueroa, M. E, Melnick,
A, & Mason, C. E. (2012) methylKit: a comprehensive R package for the analysis of
genome-wide DNA methylation profiles. Genome Biol 13, R87.26
[132] Zheng, X, Zhang, N, Wu, H. J, & Wu, H. (2017) Estimating and accounting for tumor
purity in the analysis of DNA methylation data from cancer studies. Genome Biol 18, 17.26
[133] Feng, H & Wu, H. (2019) Differential methylation analysis for bisulfite sequencing using
DSS. Quant Biol 7, 327–334.26
[134] Chen, E. Y, Tan, C. M, Kou, Y, Duan, Q, Wang, Z, Meirelles, G. V, Clark, N. R, & Ma’ayan,
A. (2013) Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool.
BMC Bioinformatics 14, 128.26
[135] Kuleshov, M. V, Jones, M. R, Rouillard, A. D, Fernandez, N. F, Duan, Q, Wang, Z, Koplev, S,
Jenkins, S. L, Jagodnik, K. M, Lachmann, A, McDermott, M. G, Monteiro, C. D, Gundersen,
G. W, & Ma’ayan, A. (2016) Enrichr: a comprehensive gene set enrichment analysis web
server 2016 update. Nucleic Acids Res 44, W90–97.26
[136] Kamburov, A, Pentchev, K, Galicka, H, Wierling, C, Lehrach, H, & Herwig, R. (2011)
ConsensusPathDB: toward a more complete picture of cell biology. Nucleic Acids Res 39,
D712–717.26
[137] Kanehisa, M, Furumichi, M, Tanabe, M, Sato, Y, & Morishima, K. (2017) KEGG: new
perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res 45, D353–D361.
26
[138] Jassal, B, Matthews, L, Viteri, G, Gong, C, Lorente, P, Fabregat, A, Sidiropoulos, K, Cook,
J, Gillespie, M, Haw, R, Loney, F, May, B, Milacic, M, Rothfels, K, Sevilla, C, Shamovsky,
V, Shorser, S, Varusai, T, Weiser, J, Wu, G, Stein, L, Hermjakob, H, & D’Eustachio, P.
(2020) The reactome pathway knowledgebase. Nucleic Acids Res 48, D498–D503.26
[139] Slenter, D. N, Kutmon, M, Hanspers, K, Riutta, A, Windsor, J, Nunes, N, M?lius, J, Cirillo,
E, Coort, S. L, Digles, D, Ehrhart, F, Giesbertz, P, Kalafati, M, Martens, M, Miller, R,
Nishida, K, Rieswijk, L, Waagmeester, A, Eijssen, L. M. T, Evelo, C. T, Pico, A. R,
& Willighagen, E. L. (2018) WikiPathways: a multifaceted pathway database bridging
metabolomics to other omics research. Nucleic Acids Res 46, D661–D667.26
56
REFERENCES
[140] Carbon, S, Douglass, E, Good, B. M, Unni, D. R, Harris, N. L, Mungall, C. J, Basu, S,
Chisholm, R. L, Dodson, R. J, Hartline, E, Fey, P, Thomas, P. D, Albou, L. P, Ebert, D,
Kesling, M. J, Mi, H, Muruganujan, A, Huang, X, Mushayahama, T, LaBonte, S. A, Siegele,
D. A, Antonazzo, G, et al. (2021) The Gene Ontology resource: enriching a GOld mine.
Nucleic Acids Res 49, D325–D334. 26
[141] Clough, E & Barrett, T. (2016) The Gene Expression Omnibus Database. Methods Mol
Biol 1418, 93–110. 26
[142] Leinonen, R, Sugawara, H, & Shumway, M. (2011) The sequence read archive. Nucleic
Acids Res 39, 19–21. 26
[143] Stunnenberg, H. G, Hirst, M, Abrignani, S, Adams, D, de Almeida, M, Altucci, L, Amin, V,
Amit, I, Antonarakis, S. E, Aparicio, S, Arima, T, Arrigoni, L, Arts, R, Asnafi, V, Esteller,
M, Bae, J. B, Bassler, K, Beck, S, Berkman, B, Bernstein, B. E, Bilenky, M, Bird, A, Bock,
C, Boehm, B, Bourque, G, Breeze, C. E, Brors, B, Bujold, D, et al. (2016) The International
Human Epigenome Consortium: A Blueprint for Scientific Collaboration and Discovery.
Cell 167, 1145–1149. 26
[144] Barretina, J, Caponigro, G, Stransky, N, Venkatesan, K, Margolin, A. A, Kim, S, Wilson,
C. J, Lehár, J, Kryukov, G. V, Sonkin, D, Reddy, A, Liu, M, Murray, L, Berger, M. F,
Monahan, J. E, Morais, P, Meltzer, J, Korejwa, A, Jané-Valbuena, J, Mapa, F. A, Thibault, J,
Bric-Furlong, E, Raman, P, Shipway, A, Engels, I. H, Cheng, J, Yu, G. K, Yu, J, Aspesi, P,
de Silva, M, Jagtap, K, Jones, M. D, Wang, L, Hatton, C, Palescandolo, E, Gupta, S, Mahan,
S, Sougnez, C, Onofrio, R. C, Liefeld, T, MacConaill, L, Winckler, W, Reich, M, Li, N,
Mesirov, J. P, Gabriel, S. B, Getz, G, Ardlie, K, Chan, V, Myer, V. E, Weber, B. L, Porter, J,
Warmuth, M, Finan, P, Harris, J. L, Meyerson, M, Golub, T. R, Morrissey, M. P, Sellers,
W. R, Schlegel, R, & Garraway, L. A. (2012) The Cancer Cell Line Encyclopedia enables
predictive modelling of anticancer drug sensitivity. Nature 483, 603–607. 26
[145] Lappalainen, I, Almeida-King, J, Kumanduri, V, Senf, A, Spalding, J. D, Ur-Rehman, S,
Saunders, G, Kandasamy, J, Caccamo, M, Leinonen, R, Vaughan, B, Laurent, T, Rowland, F,
Marin-Garcia, P, Barker, J, Jokinen, P, Torres, A. C, de Argila, J. R, Llobet, O. M, Medina,
I, Puy, M. S, Alberich, M, de la Torre, S, Navarro, A, Paschall, J, & Flicek, P. (2015) The
European Genome-phenome Archive of human data consented for biomedical research. Nat
Genet 47, 692–695. 26
[146] Tomczak, K, Czerwin´ska, P, &Wiznerowicz, M. (2015) The Cancer Genome Atlas (TCGA):
an immeasurable source of knowledge. Contemp Oncol (Pozn) 19, 68–77. 26
[147] ENCODE Project Consortium. (2004) The ENCODE (ENCyclopedia Of DNA Elements)
Project. Science 306, 636–640. 26
57
REFERENCES
[148] ENCODE Project Consortium, Moore, J. E, Purcaro, M. J, Pratt, H. E, Epstein, C. B,
Shoresh, N, Adrian, J, Kawli, T, Davis, C. A, Dobin, A, Kaul, R, Halow, J, Van Nostrand,
E. L, Freese, P, Gorkin, D. U, Shen, Y, He, Y, Mackiewicz, M, Pauli-Behn, F, Williams, B. A,
Mortazavi, A, Keller, C. A, Zhang, X. O, Elhajjajy, S. I, Huey, J, Dickel, D. E, Snetkova,
V, Wei, X, Wang, X, Rivera-Mulia, J. C, Rozowsky, J, Zhang, J, Chhetri, S. B, Zhang, J,
Victorsen, A, White, K. P, Visel, A, Yeo, G. W, Burge, C. B, Lécuyer, E, Gilbert, D. M,
Dekker, J, Rinn, J, Mendenhall, E. M, Ecker, J. R, Kellis, M, Klein, R. J, Noble, W. S,
Kundaje, A, Guigó, R, Farnham, P. J, Cherry, J. M, Myers, R. M, Ren, B, Graveley, B. R,
Gerstein, M. B, Pennacchio, L. A, Snyder, M. P, Bernstein, B. E, Wold, B, Hardison, R. C,
Gingeras, T. R, Stamatoyannopoulos, J. A, & Weng, Z. (2020) Expanded encyclopaedias of
DNA elements in the human and mouse genomes. Nature 583, 699–710.26
[149] Afgan, E, Baker, D, Batut, B, van den Beek, M, Bouvier, D, Cech, M, Chilton, J, Clements,
D, Coraor, N, Grüning, B. A, Guerler, A, Hillman-Jackson, J, Hiltemann, S, Jalili, V, Rasche,
H, Soranzo, N, Goecks, J, Taylor, J, Nekrutenko, A, & Blankenberg, D. (2018) The Galaxy
platform for accessible, reproducible and collaborative biomedical analyses: 2018 update.
Nucleic Acids Res 46, W537–W544.27
[150] Wolstencroft, K, Haines, R, Fellows, D, Williams, A, Withers, D, Owen, S, Soiland-Reyes,
S, Dunlop, I, Nenadic, A, Fisher, P, Bhagat, J, Belhajjame, K, Bacall, F, Hardisty, A, Nieva
de la Hidalga, A, Balcazar Vargas, M. P, Sufi, S, & Goble, C. (2013) The Taverna workflow
suite: designing and executing workflows of Web Services on the desktop, web or in the
cloud. Nucleic Acids Res 41, W557–561.27
[151] Kallio, M. A, Tuimala, J. T, Hupponen, T, Klemelä, P, Gentile, M, Scheinin, I, Koski, M,
Käki, J, & Korpelainen, E. I. (2011) Chipster: user-friendly analysis software for microarray
and other high-throughput data. BMC Genomics 12, 507.27
[152] Reich, M, Liefeld, T, Gould, J, Lerner, J, Tamayo, P, & Mesirov, J. P. (2006) GenePattern
2.0. Nat Genet 38, 500–501.27
[153] Ovaska, K, Laakso, M, Haapa-Paananen, S, Louhimo, R, Chen, P, Aittomäki, V, Valo, E,
Núñez-Fontarnau, J, Rantanen, V, Karinen, S, Nousiainen, K, Lahesmaa-Korpinen, A. M,
Miettinen, M, Saarinen, L, Kohonen, P, Wu, J, Westermarck, J, & Hautaniemi, S. (2010)
Large-scale data integration framework provides a comprehensive view on glioblastoma
multiforme. Genome Med 2, 65.27, 32
[154] Sadedin, S. P, Pope, B, & Oshlack, A. (2012) Bpipe: a tool for running and managing
bioinformatics pipelines. Bioinformatics 28, 1525–1526.27
[155] Di Tommaso, P, Chatzou, M, Floden, E. W, Barja, P. P, Palumbo, E, & Notredame, C. (2017)
Nextflow enables reproducible computational workflows. Nat Biotechnol 35, 316–319.27
[156] Köster, J & Rahmann, S. (2018) Snakemake-a scalable bioinformatics workflow engine.
Bioinformatics 34, 3600.27
58
REFERENCES
[157] Cervera, A, Rantanen, V, Ovaska, K, Laakso, M, Nuñez-Fontarnau, J, Alkodsi, A, Casado,
J, Facciotto, C, Häkkinen, A, Louhimo, R, Karinen, S, Zhang, K, Lavikka, K, Lyly, L,
Pal Singh, M, & Hautaniemi, S. (2019) Anduril 2: upgraded large-scale data integration
framework. Bioinformatics 35, 3815–3817. 32
[158] Chang, W, Cheng, J, Allaire, J, Xie, Y, & McPherson, J. (2020) shiny: Web Application
Framework for R. R package version 1.5.0. 33
[159] Sievert, C. (2020) Interactive Web-Based Data Visualization with R, plotly, and shiny.
(Chapman and Hall/CRC). 33
[160] Louhimo, R & Hautaniemi, S. (2011) CNAmet: an R package for integrating copy number,
methylation and expression data. Bioinformatics 27, 887–888. 34
[161] Therneau, T. M. (2020) A Package for Survival Analysis in R. R package version 3.2-7. 34
[162] Terry M. Therneau & Patricia M. Grambsch. (2000)Modeling Survival Data: Extending
the Cox Model. (Springer, New York). 34
[163] Kassambara, A, Kosinski, M, & Biecek, P. (2020) survminer: Drawing Survival Curves
using ’ggplot2’. R package version 0.4.7. 34
[164] Bisserier, M & Wajapeyee, N. (2018) Mechanisms of resistance to EZH2 inhibitors in
diffuse large B-cell lymphomas. Blood 131, 2125–2137. 36
[165] Lefor, A. T. (2011) Computational oncology. Jpn J Clin Oncol 41, 937–947. 41
[166] Groselj, B, Sharma, N. L, Hamdy, F. C, Kerr, M, & Kiltie, A. E. (2013) Histone deacetylase
inhibitors as radiosensitisers: effects on DNA damage signalling and repair. Br J Cancer
108, 748–754. 41
[167] Kachhap, S. K, Rosmus, N, Collis, S. J, Kortenhorst, M. S, Wissing, M. D, Hedayati, M,
Shabbeer, S, Mendonca, J, Deangelis, J, Marchionni, L, Lin, J, Höti, N, Nortier, J. W,
DeWeese, T. L, Hammers, H, & Carducci, M. A. (2010) Downregulation of homologous
recombination DNA repair genes by HDAC inhibition in prostate cancer is mediated through
the E2F1 transcription factor. PLoS One 5, e11208. 41
[168] Lu, Y, Chan, Y. T, Tan, H. Y, Li, S, Wang, N, & Feng, Y. (2020) Epigenetic regulation in
human cancer: the potential role of epi-drug in cancer therapy. Mol Cancer 19, 79. 41
[169] Roberti, A, Valdes, A. F, Torrecillas, R, Fraga, M. F, & Fernandez, A. F. (2019) Epigenetics
in cancer therapy and nanomedicine. Clin Epigenetics 11, 81. 41
[170] Kim, H. J, Maiti, P, & Barrientos, A. (2017) Mitochondrial ribosomes in cancer. Semin
Cancer Biol 47, 67–81. 42
[171] Madhavan, S, Beckman, R. A, McCoy, M. D, Pishvaian, M. J, Brody, J. R, & Macklin, P.
(2021) Envisioning the future of precision oncology trials. Nature Cancer 2, 9–11. 43
59