Dysregulations of DNA and mRNA epigenetic modifications in breast cancer

Dysregulations of DNA and mRNA

epigenetic modifications in breast cancer

Thesis submitted by Clémence AL WARDI

in fulfilment of the requirements of the PhD Degree in biomedical and

pharmaceutical sciences (“Docteur en sciences biomédicales et

pharmaceutiques”)

Academic year 2020-2021

Supervisor: Professor François FUKS

Laboratory of Cancer Epigenetics

Thesis jury :

Carine MAENHAUT (ULB, Chair) François FUKS (ULB, Secretary) Benjamin BECK (ULB)

Isabelle DEMEESTERE (ULB) Denis LAFONTAINE (ULB) Basile STAMATOPOULOS (ULB) Katleen DE PRETER (Ugent)

« No man ever steps in the same river twice, for it's not the same river and he's

not the same man»

Heraclitus; Ve s. av. J.-C.

“What’s your thesis about again?”

1

Résumé

Cette thèse est axée en deux parties : une première partie explorant la dérégulation d’une modification épigénétique localisée sur l’ADN tandis que la seconde porte sur une modification des ARN messagers. Ces deux modifications sont étudiées dans le cadre du cancer du sein.

La première partie concernant l’hydroxymethylation de l’ADN décrit l’influence de l’infiltration immune sur l’expression de l’enzyme TET1, responsable de l’ajout de cette modification. En effet, notre étude met en évidence une anti-corrélation entre l’expression de TET1 et l’infiltration immune (via IHC et analyses in silico). De plus, nous démontrons que l’activation de la voie de signalisation NF-B dans le cancer du sein « basal-like » réduit l’expression de TET1. Finalement, nous avons pu mettre en évidence, grâce à plusieurs expériences dont de l’immunoprécipitation que cette régulation se déroulait (au moins en partie) via liaison d’NF-B au promoteur de TET1. Dans le but d’élargir notre model à d’autres cancers, nous avons pu également observer cette anti-corrélation et l’effet de NF-B dans les cancers de la thyroïde, du poumon et de la peau. Avec le développement récent de l’immunothérapie, ce lien entre modification épigénétique et immunité pourrait s’avérer intéressant à exploiter.

La seconde partie décrit quant à elle la dérégulation de la modification m6A dans le cancer du sein en général et non plus dans un sous-type particulier. Nous avons non seulement été en mesure de décrire une accumulation de la marque dans les tissus cancéreux mais également de proposer une cause et une conséquence à cette dérégulation. En effet, cette augmentation de la présence d’m6A (observée par spectrométrie de masse) serait causée par la baisse d’expression d’une enzyme responsable de sa suppression (l’enzyme FTO). Cette dérégulation, en agissant directement sur les ARNm de la voie de signalisation WNT comme le montre notre m6A-seq, active d’avantage la voie. Cette suractivation des WNT entraine alors une transition épithélio-mésenchymateuse des cellules, étape importante dans le développement de métastases. Ainsi, les patientes présentant une tumeur à basse expression de FTO pourraient être plus sujettes à des récurrences tumorales. Nous avons également observé cette même régulation dans d’autres cancers comme celui de la prostate, à la fois in vitro, in vivo et in silico. De plus, notre recherche met également en évidence une plus grande sensitivité de ces tumeurs à un inhibiteur des WNT. En effet, nos expériences de xénogreffes soulignent que des tumeurs avec peu de FTO, traitées avec l’inhibiteur des WNT ICRT-3, sont bien moins actives et grandes que les tumeurs contrôle et développent moins de métastases. Mesurer la quantité d’FTO tumorale pourrait donc en théorie permettre une meilleure discrimination des patientes pour de tels traitements.

2

Summary

This thesis is divided in two parts: the first part explores the dysregulation of an epigenetic modification localized on DNA and the second part is centred on a mRNA modification. These two marks are being studied in the context of breast cancer (BC).

The first part describes the influence of immune cell infiltration on the expression of the TET1 enzyme, which is responsible for the addition of hydroxymethylation to DNA. Our study indeed first describes an anti-correlation between TET1 expression and immune infiltration (through both IHC and in silico analysis). In addition, we show that NF-B pathway activation in « basal-like » breast cancer cells downregulates TET1 expression. Finally, we highlighted through several experiments including immunoprecipitation, a TET1 promoter binding of NF-B. In an effort to generalize our data to other cancers, we also investigated this anticorrelation and the NF-B regulation in several cancers. Our results indicate that this immune related regulation also occurs in lung, thyroid and skin cancers. With the increasing use of immunotherapy as anti-cancer treatment, this link between an epigenetic modification and immunity might be interesting to explore.

3

Acknowledgments

I would first like to thank my promoter François Fuks who was crazy enough to let me pursue both my master thesis and my PhD in his lab. Special thanks also to Dr. Christos Sotiriou and Dr. Alexandra Van Keymeulen for being part of my PhD comity and everyone in my jury for taking the time to read this work.

I also thank Jana and Evelyne who trained me and taught me everything I can do, from western blots to those annoying TOP FOP. Thank you again Jana for spending your nights reading this! Of course, I want to thank all the people in the lab! I spent 5 wonderful years with you and I love you all. Thank you Your Highness Bouchra for the qPCRs and western blots and all the time I asked for your help. Thank you, Emilie, for the sequencing and the little talks. Thank you, my dear qPCR queen Celine, who I literally submerged through samples (oops). Thank you, Pascale, for the cloning and the chocolates. Thank you, Florence, for all the singing and for being the excuse to have a drink (you know). Thank you, Martin, for his amazing expertise in bioinformatics and the time you spent with me when I was lost with the data. Thank you Jie for your input in my thesis, for being such a nice person even though we made a lot of noise in the office.

I also have to thank my parents (Frédérique and Sémir) for their genetic input which allowed me to be alive and write this thesis. But dad no thanks for the allergy related genes. A special thank you of course for my second mother (Gigi) and her husband (François) who supported me all along. For my brother (Paul) and sisters (Camille, Alexandra and Solène): you did nothing but I still love you anyway. To my friends Kenza and Maïssam. This thesis is also in your delicate hands thanks to my ex-colleague and soon to be husband Eric.

4

Table of contents

Abbreviations ... 7

1.

Introduction ... 13

1.3.3 Other histone modifications ... 21

1.4 DNA modifications ... 21

1.4.1 Cytosine methylation ... 22

1.4.2 The 5mC machinery ... 22

1.4.3 5mC distribution and consequence on gene expression ... 23

1.4.4 Methylome mapping technologies ... 24

1.4.5 Pathological implications of 5mC ... 26

1.4.6 Cytosine hydroxymethylation ... 27

1.4.7 The 5hmC machinery ... 29

1.4.8 5hmC distribution and consequence on gene expression ... 29

1.4.9 Hydroxymethylome mapping technologies ... 32

1.4.10 Pathological implications of 5hmC ... 32

1.4.11 Other covalent DNA modification. ... 34

1.4.12 Methylation of adenine ... 34 1.5 Crosstalks in Epigenetics ... 36 1.6 RNA modifications ... 37 1.6.1 Methylation of adenosine ... 37 1.6.2 m6A machinery ... 39 1.6.3 m6A distribution ... 41

1.6.4 Pathological implications of m6A ... 42

1.6.4.1 m6A in breast cancer ... 43

1.6.5 Other mRNA modifications ... 44

1.7 Breast cancer ... 45

1.7.1 Breast cancer subtypes ... 47

1.7.1.1 Histology-based subtypes ... 47

5

1.7.2 Immune infiltration in breast cancer ... 49

1.7.3 Breast cancer treatments ... 51

1.7.4 Epigenetics in breast cancer ... 51

1.8 The NF-B signaling pathway ... 54

1.8.1 The canonical pathway: ... 54

1.8.2 The non-canonical pathway: ... 55

1.8.3 The NF-B pathway in cancers ... 56

1.8.3.1 Pro-tumorigenic roles of the NF-B pathway: ... 56

1.8.3.2 Anti-tumorigenic roles of the NF-B pathway: ... 56

1.8.3.3 NF-B in breast cancer: ... 57

1.8.4 Therapies targeting the NF-B pathway ... 57

1.9 The WNT signaling pathway ... 58

1.9.1 The canonical pathway ... 59

1.9.2 The non-canonical pathway ... 60

1.9.3 The WNT pathway in cancers ... 60

1.9.3.1 The WNT pathway in breast cancer: ... 61

1.9.3.2 Implications in immunotherapy ... 61

1.9.4 Therapies targeting the WNT pathway ... 62

2.

Aims of this thesis ... 64

3.0. Chapter I:

Immunity drives TET1 regulation in

cancer through

NF-B. ... 66

3.0.1 Introduction ... 67

3.0.2 Results ... 69

3.0.2.1 TET1 regulation is associated with 5hmC changes in BLBC ... 69

3.0.2.2 In basal-like tumors, high TET1 expression is associated with low levels of immune and defense response markers ... 70

3.0.2.3 TET1 expression is repressed by NF-B activation ... 70

3.0.2.4 TET1 is repressed through binding of NF-B to its promoter ... 72

3.0.2.5 TET1 is down-regulated by NF-B in other cancer types ... 72

6

3.1.1 Introduction ... 75

3.1.2 Results ... 77

3.1.2.1 FTO loss enhances tumor progression... 77

3.1.2.2 FTO loss elicits an EMT program. ... 77

3.1.2.3 FTO controls Wnt signaling in cancer. ... 80

3.1.2.4 FTO-low tumors are sensitive to WNT inhibitor therapy. ... 82

4.0. Discussion ... 84

4.1 Immunity drives TET1 regulation in cancer through NF-B. ... 85

4.2. Downregulation of the m6A RNA demethylase FTO promotes EMT-mediated tumor progression and confers sensitivity to Wnt inhibitors. ... 90

5.0. Conclusion ... 97

6.0. References ... 99

7.0. Appendix

... 137

7.1. Manuscript I ... 138

7

Abbreviations

12 VIM………..………...Vimentin WHO……….. World Health Organisation WNT………..…. Wingless related Integration site WTAP……… Wilms Tumor 1 Associated Protein XIST………..…… X Inactive Specific Transcript YTHD……….…….. YTH Domain-Containing Family ZCCHC4………...………... Zinc Finger CCHC-Type Containing 4 ZO1………... Zona Occludens 1 2-OG………..……. 2-Oleoylglycerol 5CaC………...………. 5-carboxylcytosine 5fC………...…... 5-formylcytosine 5hmC……… 5-Hydroxymethylcytosine 5hmU………..…….. 5-Hydroxymethyluracil 5mC………...…. 5-Methylcytosine 6mA……….……….. DNA N 6-methyladenine

13

14

1.1 Epigenetics

In 1953 James Watson and Francis Crick presented the double-helix structure of DeoxyriboNucleic Acid (DNA), the molecule storing the biological instructions of every living entity. Helped by two other biophysicists: Rosalind Franklin and Maurice Wilkins, their major discovery explained how this molecule of inheritance can replicate itself from one cell to another (WATSON and CRICK 1953). It was finally understood how the information, stored in the form of a four-character molecular code, was transmitted from one generation to the next. However, this genetic code itself was not enough to explain the diversity of cells in an organism. A human body contains trillions of cells, which all carry the same DNA yet an intestinal cell greatly differs in shape and function from a skin cell: one might display specific cilia to absorb nutrients while the other produces a protective pigment.

It is now known that these differences among cells are determined by the activation or repression of specific sets of genes (Eissenberg 2014). Genes are parts of DNA which can (for some of them) act as an instruction unit for the production of proteins. The mechanism responsible of this control of genes (without altering the genetic code itself) has been called “epigenetics” (Feinberg 2008) and defines cell identity. From an evolutionary point of view, epigenetics has been proposed to be partly responsible for the emergence of multicellular organisms by allowing different expression patterns in different cells (Osborne 2017).

15 The first concrete evidence establishing the influence of the environment over epigenetics was described in 2005. It was shown how identical twins, which share the same genome can develop different diseases over time (Fraga et al. 2005). Indeed, even if both twins start their lives with the same epigenetic patterns, as they evolve in different environments their epigenetic patterns will change and this can cause differences in disease susceptibility. Environmental influence on our genes revived an old debate in the scientific community: is the apparition of variants random as Darwin proposed or is it rather guided by environmental need as Lamarck suggested? Not all mutations are the consequence of the influence of the environment, but epigenetic changes such as DNA methylation have been reported to increase the mutations rate (Guerrero-Bosagna 2020). In the light of today’s genomics, it has been suggested that both neo-Darwinian and neo-Lamarckian mechanisms are able to drive evolution. Indeed, epigenetics might increase the ability of the environment to influence natural selection (Skinner 2015).

In addition of its implication in cell differentiation, several biological processes such as aging have been linked to epigenetics. Interestingly, a correlation between methylation of some DNA regions and aging has been described. This observation, which is called the ‘epigenetic clock’, has revived the hopes to decelerate aging (Bell et al. 2019; Unnikrishnan et al. 2019). Accumulating evidences suggest that epigenetic modifications are dysregulated in many diseases including cancer. Several studies highlighted that DNA methylation, histone modifications and RNA based mechanisms are implicated in every step of tumorigenesis, from tumor initiation to the metastasis (Audia and Campbell 2016; Kulis and Esteller 2010; Anastasiadou, Jacob, and Slack 2017). For that reason, intense research activity is currently

Figure 1 : The epigenetics pillars. Epigenetics includes DNA modifications, histone modifications (also called the “histone code”) and the RNA-based mechanisms. Among this last pillar which contains the non-coding RNAs rose a new mechanism:

16 ongoing to discover new therapeutic possibilities, based on epigenetics, for cancers. Nowadays, epigenetic studies already enabled the discovery of new biomarkers useful for diagnosis or prognosis and the description of new therapeutic targets for several cancers (Werner, Kelly, and Issa 2017; Dumitrescu 2018).

Epigenetics is a new and exciting field, which holds great potential for the understanding and treatment of diseases such as cancers. Nevertheless, its complexity as a dynamic phenomenon has to be explored with more studies to comprehend the different aspects of this field. This thesis is segmented in two parts: a first one exploring a dysregulation of the DNA modification called hydroxymethylation and a second part focusing on a mRNA modification called m6A. For both modifications, we studied their dysregulations in breast cancer (BC). I will first introduce basic notions such as chromatin and the epigenetic pillars before explaining breast cancer and finally the pathways involved in this thesis.

1.2 Chromatin organization:

Thanks to the sequencing of the human genome in 2003 we know that DNA is formed by approximately 3 billion base pairs (A-T or G-C) and contains about 20,000 genes (Ezkurdia et al. 2014). If we could stretch once cell’s DNA, it would be around two meters long whereas the size of the nucleus is only 6µm (Alberts et al. 2002). Therefore, in order to be able to be confined in the cell’s nucleus, the DNA is highly packed (around 100,000-fold) by wrapping itself around a core of proteins called histones (Figure 2). This conformation is known as “chromatin” (Chakravarthy et al. 2005). Chromatin is a very dynamic structure which can fold and unfold itself in order to regulate gene expression. In eucaryotes, DNA is divided to form chromosomes and as a diploid specie, Homo sapiens contain two sets of these chromosomes. The basic unit of chromatin is called “nucleosome”. It includes a small piece of DNA (around 147 bp) wrapped around an octamer of histones. This octamer is made of two copies of the

Figure 2: Chromatin basic unit: the nucleosome. DNA is wrapped around a histone protein core composed of two of each histone H2A-H2B-H3 and H4 while another histone called H1 binds DNA outside the core. Adapted from Reuter, S. et al.

17 four core histones (H2A, H2B, H3 and H4) (Mirabella, Foster, and Bartke 2016). Between these nucleosomes, another histone called H1 binds to DNA and is able to further condense chromatin. Histones are made of two major elements: a globular hydrophobic core and a basic N-terminal tail which can be subject to post translational modifications. These modifications are able to control the compaction level of the chromatin and will be further explored in 1.3 (Histones modifications).

Chromatin can undergo different level of compaction. The first level is called the “beads on a string” conformation while the most compacted form can be observed during metaphase (Mirabella, Foster, and Bartke 2016). Metaphase is a step of mitosis and meiosis when chromosomes are gathered in the centre of the cell before it splits into two cells. In 1928, Emil Heitz observed two different staining of chromatin: a dense staining which he will call

heterochromatin and a lighter staining for euchromatin (Passarge 1979). Those two chromatin

states actually illustrate different levels of compaction (Figure 3). Euchromatin is a less compact state which allows the access of the transcription machinery and therefore gene expression. Nevertheless, not all euchromatin regions are actually transcriptionally active as some genes would need specific transcriptional factors in addition of this open chromatin state. Genes benefitting from a constitutively active transcriptional region are called “housekeeping genes” and are used as controls when analysing gene expression.

Heterochromatin is a highly packed form of chromatin, making gene transcription harder or almost impossible. We can distinguish two subcategories of heterochromatin. Constitutive heterochromatin is a chromatin constantly highly packed while facultative heterochromatin are regions alternating between heterochromatin and euchromatin in certain cells. Regions containing constitutive heterochromatin are usually non-coding, repetitive sequences, centromeres and telomeres to ensure genome integrity (Probst and Almouzni 2011; Postepska-Igielska et al. 2013). A well-known example of facultative heterochromatin is the inactivation of one of the two X chromosomes in female mammals (Chow and Heard 2009). To conclude, chromatin is a very dynamic form where DNA is more or less accessible for transcription. The association of DNA with proteins such as histones allows a complex regulation of its compaction and epigenetic modifications on DNA and histone tails are known to be partly responsible for it.

Figure 3: Different compaction states of chromatin. Chromatin can undergo different levels of packing such as euchromatin which is a loose state usually open to gene expression while heterochromatin is highly compact and therefore rather

18

1.3 Histones modifications:

Post-translational modifications of histone tails can affect gene expression through alteration of chromatin compaction (Audia and Campbell 2016). This can be achieved in two ways: (i) through interaction between the electronic charge of the modification and DNA (which is negatively charged because of phosphate groups) and (ii) through recruitment of chromatin remodelling complex to the modification. Hundreds of histone modifications have been reported with various consequences on gene expression (Figure 4) (Stillman 2018). The combination of these modifications is called the “histone code”.

Mapping of histone marks in the genome can be achieved by using specific antibodies targeting the modification to precipitate only the part of DNA close to this mark. This is called chromatin-immunoprecipitation and it is usually followed by the sequencing of those parts of DNA with high-throughput sequencing (ChIP-Seq) (O’Geen, Echipare, and Farnham 2011). The most characterized modifications of histone tails are acetylation, methylation, phosphorylation and ubiquitination.

1.3.1 Histone acetylation

Acetylation is the transfer of an acetyl group from acetyl-coenzyme A to a lysine of a histone tail (Shahbazian and Grunstein 2007). Lysine is an amino-acid which holds a positive amine group. Because of the presence of several positive charged amino-acid such as lysine, histone tails are globally positively charged which allow them to interact with DNA. This interaction can be weakened by histone acetylation which then neutralizes the positive charge from the lysine. In consequence, chromatin becomes more relaxed and accessible for gene expression (Figure 5). The addition of an acetyl group is catalysed by enzymes called histone acetyltransferases (HATs) (Marmorstein and Roth 2001). In general, enzymes responsible for the addition of a modification are called “writers”.

19 Moreover, histone acetylation can also act as a binding platform and recruit transcription proteins which contain a specific domain such as the bromodomain (Gong, Chiu, and Miller 2016). Those recruited machineries also called “readers” can act directly on chromatin and gene expression. Although some can be described at regulatory regions, most of histone acetylations are present at active genes promoters and occur on histones H3 and H4 (Bannister and Kouzarides 2011). As an example, acetylation of the lysine 4 or 9 of histone 3 (H3K4ac and H3K9ac) can be found in promotor regions.

Histone acetylation is a reversible phenomenon. The removal of an acetyl group (or deacetylation) is ensured by histone deacetylases (HDACs) which can then also be referred as “erasers” of the mark (De Ruijter et al. 2003). In human, we can describe 18 highly conserved HDACs which are divided in four classes depending on their sequence similarities. As HDACs can remove the acetyl group from the lysine of a histone tail and re-establish its positive charge, they are responsible of a more compact chromatin and thus transcription silencing.

1.3.2 Histone methylation

Histone methylation is the addition of a methyl moiety on a histone tail. The modified residue can undergo three different states of methylation depending on the number of added methyl group. If only one methyl group is added the residue is then “monomethylated” but it can also be di and “trimethylated”. Two different writers have been described for methylation depending on the amino-acid: the writers adding a methyl group on arginine (arginine methyltransferases or PRMTs) and the writers for lysine (histone methyltransferases or HMTs) (Greer and Shi 2012).

Figure 5: Histone tails acetylation. The acetylation of histone tails by HAT enzymes is associated with a loose state of chromatin which is favourable for transcription while their deacetylation by HDAC tighten the chromatin and repress gene expression. HATs are considered “writers” of the mark, HDACs “erasers” while protein able to bind the acetylation are called

20 Unlike histone acetylation, histone methylation does not change the electronic charge of the tail. Its effect is then rather mediated through the recruitment of readers. This recruitment can be achieved thanks to specific protein domains able to read methylation such as the “chromo-domain” (Eissenberg 2001). Another difference from histone acetylation is that methylation is not clearly associated with gene expression or gene silencing as it can be present in both situations (Peterson and Laniel 2004). What can distinguish an active or repressive methylation mark is the amino-acid involved, its location and its methylation state. For instance, a methylation mark known to be associated with active gene expression is the trimethylation of histone H3 lysine 4 (H3K4me3) which can be found in promoters. In opposition, trimethylations of lysine 9 or 27 of histone H3 (H3K9me3 and H3K27me3) are rather linked with gene repression (Hyun et al. 2017). Other histone methylation marks have also been reported to be implicated in several mechanisms such as DNA damage response, elongation and genome stability.

Interestingly, some promoters can carry both an active mark such as H3K4me3 and a repressive one as H3K27me3 and are in consequence in a poised state (Voigt, Tee, and Reinberg 2013). Those promoters can be found in embryonic stem cells and are called “bivalent promoters”. Bearing both an active and a repressive mark allows a fast shifting expression of developmental genes and ensures proper cell differentiation. Indeed, in presence of the right differentiation signal, these promoters will either lose the active or the repressive mark and act on gene expression. This loss is catalysed by the erasers of the methylation mark: histone demethylases (HDMs) (Pedersen and Helin 2010). Histone methylation is a major regulator of gene expression and has been shown to often be dysregulated in diseases such as cancer (Varier and Timmers 2011).

Figure 6: Histone tails methylation. Methylation of histone tails can be associated with both repressive and active

chromatin depending on the mark. H3K27me3 is then rather linked to gene silencing while H3K4me3 or H3K36me3 are rather found in active chromatin. Adapted from Frank W. Schmitges et al.2011

21

1.3.3 Other histone modifications

Many more histone modifications have been reported (T. Zhang, Cooper, and Brockdorff 2015). For instance, histone phosphorylation can be found on several residues such as serine, threonine or tyrosine. As histone acetylation, phosphorylation can act on chromatin through the negative charge of its phosphate group. This repulsive force can then relax the chromatin and allow gene expression (Erler et al. 2014). The writers of histone phosphorylation are kinase proteins which use ATP as a cofactor while the erasers are called phosphatases. Histone phosphorylation is also a good example of crosstalk between histone modifications as it can interact with other histone modifications to recruit effector proteins (Zippo et al. 2009). This modification has been linked to various cell functions such as mitosis or DNA repair (Rossetto, Avvakumov, and Côté 2012).

Another example of histone modification is ubiquitination which is the addition of ubiquitin, a 76 amino-acid protein (Weake and Workman 2008). This modification can be added by histone ubiquitin ligases and removed by deubiquitinating enzymes. The most ubiquitinated proteins in the nucleus are the histones H2A and H2B which are mainly monoubiquitinated but can also undergo poly-ubiquitination. H2A and H2B ubiquitination can have different effects on transcription. H2A ubiquitination has been reported to rather be linked with gene repression while H2B ubiquitination is mostly involved in gene expression activation (J. Cao and Yan 2012). Histone ubiquitination is also implicated in several mechanisms such as DNA repair, maintenance of chromatin structure or transcription (Huen et al. 2008)(Chandrasekharan, Huang, and Sun 2009). Although less documented, the other histones H3, H4 and H1 have also been reported to be ubiquitinated (Hengbin Wang et al. 2006)(Thorslund et al. 2015).

1.4 DNA modifications

22

1.4.1 Cytosine methylation

Cytosine methylation is the attachment of a methyl group on the 5’ carbon of a cytosine and is often called 5-methylcytosine (5mC) or the “fifth base” of DNA (Liyanage et al. 2014). About 3 to 5% of all cytosines have been observed to be methylated in mammals, the brain tissue being the most methylated tissue in humans (Moore, Le, and Fan 2013; Pfeifer 2018). 5mC has been described in both prokaryotic and eukaryotic DNA and is implicated in several cell functions such as tissue-specific gene expression, X chromosome inactivation or repression of transposons (Zamudio et al. 2015; Chow and Heard 2009; Newell-Price, Clark, and King 2000). Transposons are indeed sources of genome instability, because of their abilities to translocate into genes, which might alter their functions (Zamudio et al. 2015). DNA methylation is globally maintained in somatic cells to ensure cell specificity but undergoes major reorganisation during development (Y. Zeng and Chen 2019). This reorganization incudes two waves of global demethylation followed by re-methylation: a first one during gametogenesis and a second one after fertilization (Figure 7). It has been suggested that this reshape of methylation patterns avoids the inheritance of environmentally-acquired methylation (Liyanage et al. 2014). Nevertheless, some modifications could escape this reprogramming and still be found in the offspring DNA (Illum et al. 2018).

1.4.2 The 5mC machinery

The writers responsible for cytosine methylation belong to the family of DNA Methyl Transferase (DNMT) which use S-adenosylmethionine (SAM) as a methyl group donor (Figure 8) (N. Zhang 2018). We can describe two types of methylation: the “de novo methylation” and “maintenance methylation”. The de novo methylation corresponds to the establishment of a new DNA methylation pattern in the zygote and is performed by the enzymes DNMT3A and

DNMT3B (Okano et al. 1999). The maintenance methylation ensures the conservation of

methylated sites through cell divisions and is catalysed by DNMT1. Mice with a dysfunctional

Figure 7: Epigenetic reprogramming. Two waves of demethylation and re-methylation are occurring: a first one during gametogenesis and a second one during embryogenesis. Adapted from Zeng and Chen 2019

23

Dnmt3a gene (Dnmt3a knockout) display several developmental anomalies and die shortly

after birth, whereas knockout mice for either Dnmt1 or Dnmt3b die as embryos (Okano et al. 1999). This illustrates how critical DNA methylation patterns are for the normal development of an organism. Proteins able to read 5-methylcytosine can contain a Methyl Binding Domain (MBD) and act on gene expression (Du et al. 2015). For several years, 5mC was described as a stable modification which could potentially undergo passive demethylation through cell divisions (Bhutani, Burns, and Blau 2011). However, in 2009, erasers of cytosine methylation are finally characterized and will be explored in 1.4.6.

1.4.3 5mC distribution and consequence on gene expression

In humans, 5mC is mostly found on cytosines which are followed by a guanine in the DNA sequence (Arand et al. 2012). This conformation is called “CpG” and around 60–80% of them are methylated (Smith and Meissner 2013). CpGs can also be clustered in particular spots in the genome, which are then labelled “CpG islands” or CGIs. Those CpG islands are usually located in the promoter region of genes, but unlike CpGs alone these clusters are generally unmethylated (Deaton and Bird 2011). This absence of methylation on promoters allows for the accessibility of DNA to the transcription machinery. Cytosine methylation is indeed mostly associated with gene repression (Newell-Price, Clark, and King 2000). At least three different phenomena have been described for 5mC mediated gene silencing: 1) Methylation could directly affect transcription factors recruitment to DNA and therefore prevent activation of gene expression (Medvedeva et al. 2014). 2) Another direct effect of 5mC might be its ability to strengthen the DNA duplex. It has indeed been observed that it is physically harder to separate two strands of DNA when methylated (Severin et al. 2011). This could then affect helicases (enzymes responsible for DNA strands dissociation) function. 3) Finally, 5mC might regulate gene expression indirectly, through the recruitment of repression factors (Fujita et al. 2000). Among the highly methylated sites, several regions such as repeated sequences or transposons have been documented (J. Su et al. 2012).

24 Thanks to the emergence of new genome-wide profiling technologies, DNA methylation across the genome or the “methylome” has been more precisely characterized. Nowadays, 5mC locations have been described not only in promoter regions, but also at enhancers or gene bodies, suggesting additional regulating roles (Figure 9) (Ndlovu, Denis, and Fuks 2011). Indeed, those new 5mC locations inside the gene have been shown to inhibit alternative promoter activity and might then be able to regulate isoforms expression (Maunakea et al. 2010). Interestingly, intragenic methylation seems to be mostly catalysed by DNMT3B and is rather positively associated with gene expression (Gagliardi, Strazzullo, and Matarazzo 2018). The mechanism, by which intragenic 5mC might regulate gene expression positively, is not clear yet. However, it has been suggested that it could affect elongation and enhancers (Lorincz et al. 2004) (Y. Song et al. 2019).

1.4.4 Methylome mapping technologies

The first DNA methylation profiling technologies (e.g. via chromatography or restriction enzymes) allowed for the discovery of promoter methylation (Fouse, Nagarajan, and Costello 2010). Nevertheless, thanks to more recent and advanced technologies, methylation has been located not only at promoters, but also at intragenic and non-coding regions (Maunakea et al. 2010). The most commonly used technology to detect and map DNA methylation is bisulfite sequencing (BS-Seq) (Y. Li and Tollefsbol 2011). This method relies on the use of bisulfite to create a chemical reaction, which converts all unmethylated cytosines to thymine. Treated and untreated DNA samples are then sequenced and compared: all cytosines present in both sequencing reactions are then considered as methylcytosine (Figure 10). The main advantage of this technology is that it is able to detect DNA methylation at single-nucleotide resolution. Unfortunately, this method is not appropriate for large sample cohort, because of its high costs. In addition, it also suffers from technical issues such as DNA degradation (Ziller et al. 2015; Grunau 2001). Incomplete bisulfite conversion can also occur and lead to an overestimation of methylcytosine (Grunau 2001). Another bisulfite-based method is the Infinium array-based technology, which uses fluorescent probes to detect methylated CpGs

Figure 9: Distribution of 5mC across the gene. 5mC is present on promoters where it can repress gene expression, on transcription start sites and on exons where it is associated to gene expression. Adapted from Eunise M. Aquino et al. 2018

25 on an array chip. Its limited costs finally allows for profiling of large cohorts (Pidsley et al. 2016).

Other often-used technologies are based on immunoprecipitation. Specific antibodies against 5mC can immunoprecipitate and enrich methylated DNA fragments before sequencing. This method is called “MeDIP-Seq” and in an adaption of this method, called “MBD-Seq”, reader proteins bound to the modification, and not the modification itself, is immunoprecipitated (Nair et al. 2011). This method achieves good coverage of DNA methylation across the genome, but its costs have to be taken into consideration.

Another method to detect methylation is based on enzymatic reactions. This protocol indeed relies on methylation sensitive and insensitive enzymes, such as the pair HpaII and MspI (Zilberman and Henikoff 2007). Both enzymes can recognize the CCGG restriction site. Nevertheless, MspI will cut the sequence no matter the presence of a methylcytosine as it is methylation- insensitive. HpaII, on the other hand, is unable to cut this sequence if it is methylated. By comparing those two treatments after sequencing, DNA methylation can then be assessed.

Finally, an exciting new technology arose in 2012 to detect 5mC: nanopore sequencing (Figure 11) (Loose 2017). DNA is processed into small pores of a synthetic membrane, which can measure electrolytic current signals that are specific for each nucleotide. When a modified base, e.g. methylcytosine enters through the pore, a specific variation of the nucleotide-specific electric current will be measured and can then be interpreted through bio-informatic analysis.

Figure 10: Bisulfite-sequencing. The bisulfite treatment converts all unmethylated cytosines to uracile which will then be sequenced as thymine after a PCR amplification step. By comparing the DNA with and without bisulfite treatment, the 5mC

26

1.4.5 Pathological implications of 5mC

As DNA methylation has an important role in the regulation of gene expression, its implication in diseases such as Alzheimer, schizophrenia, but especially in cancer, have been extensively explored (Huo et al. 2019; Pries, Gülöksüz, and Kenis 2017; Kulis and Esteller 2010) . Many studies have shown that cancer cells tend to accumulate methylation changes (Kulis and Esteller 2010; Gokul and Khosla 2013; Klutstein et al. 2016). These changes can affect several mechanisms such as apoptosis or cell cycle and therefore disturb normal cell function. Globally, cancer cells undergo a loss of methylation, especially in repetitive sequences, gene poor regions or transposable elements, which can therefore affect genome stability (Klutstein et al. 2016). This 5mC loss can also occur in promoters of proto-oncogenes (genes that can promote cancer) such as the RAS genes and induce their abnormal transcriptional activation (Figure 12) (Maryam and Idrees 2018).

Figure 11: The nanopore technology. In this method, while DNA is going through the pore it can create a voltage variation which is specific for each nucleotide but also any possible modification. Adapted from Michael C Schatz 2017

DOI:10.1038/nmeth.4240

Figure 12: DNA methylation in cancer. Although a global demethylation of DNA is considered as a hallmark in cancer and can induce oncogene expression, hypermethylation can also occur notably in tumor suppressor promoters. Adapted from

27 Although 5mC is globally decreased in cancer cells, local hypermethylations can also take place (Dawson and Kouzarides 2012). Several studies indeed reported promoter hypermethylation of Tumor Suppressor Genes (TSG) and their subsequent silencing (L. Y. Feng, Chen, and Li 2019; Charlet et al. 2017; Manel Esteller 2002). There are multiple underlying causes of DNA methylation changes in cancer. It can be due to mutations or alteration of expression of the 5mC machinery, as it has been observed for DNMTs in Acute Myeloid leukaemia (AML) (Ley et al. 2010). Nevertheless, DNA methylation can also be affected without any disturbance of its machinery and illustrate stochastic events selected by growth advantage (Assenov, Brocks, and Gerhäuser 2018). Not all the genes are affected in the same way among cancers. As an example, the gene GSTP1, implicated in cell detoxification, has been observed as hypermethylated in around 91% of prostate cancers. However, this gene is rather hypomethylated in acute myeloid leukaemia (Singh et al. 2015).

DNA methylation also allowed for the improved characterization of various cancers. Indeed, new clusters for breast or colorectal cancers, unidentified by phenotype or classical molecular observations, have been discovered thanks to methylation profiling (Abeshouse et al. 2015; Dedeurwaerder et al. 2011). In addition, as DNA methylation patterns are cell-type specific, it is now possible to identify cell-type signatures in a tissue sample (Jeschke et al. 2017). This enables the description of the tissue composition with great details and the quantification of features such as the tumor immune infiltration, which is useful for prognosis (Oble et al. 2009; Melichar et al. 2014).

The in-deep analysis of methylation patterns in cancer cells highlighted several potential biomarkers. In 2016, the FDA approved the hypermethylation status of SEPT9 as an early diagnosis marker for colorectal cancer (L. Song et al. 2017). Such methylation anomalies can also be useful as prognosis markers. As an example, promoter hypermethylation of MGMT (O6-methylguanine-DNA methyltransferase) has been linked to a better response to chemotherapy in glioma (M. Esteller 2000). The growing knowledge of histone and DNA methylation patterns in diseases, including cancers, already enabled the production and use of what we call “epi-drugs”. For instance, an HDAC inhibitor called SAHA has been approved to treat some lymphomas while 5-azacytidine (a DNMT inhibitor) is administrated to some AML patients (Eckschlager et al. 2017; Schuh et al. 2017). Currently, several epi-drugs are tested in clinical trials, most frequently in combination with chemo and immunotherapy (ClinicalTrials.gov Identifier: NCT01935947, NCT03220477, NCT02664181).

1.4.6 Cytosine hydroxymethylation

28 oxidation of 5mC to 5hmC. The level of 5hmC across tissues has been determined to be around 0.05 to 0.65 % of all cytosines with a higher amount in cerebral cortex (W. Li and Liu 2011). Of note, although 5hmC is less abundant than 5mC, it corresponds to 40% of methylated cytosines in cerebellum Purkinje cells (Kriaucionis and Heintz 2009).

As 5mC, 5hmC has also been implicated in normal mammalian development and 5hmC/TETs have been proposed to be involved in the demethylation waves, which I previously mentioned (Yamaguchi, Hong, et al. 2013). However, this statement is still under debate as a recent study showed that the hydroxymethylation wave follows the one of demethylation and is not concomitant (Amouroux et al. 2016). Furthermore, hydroxymethylation has also been reported to be involved in the differentiation of several cell types such as neurons, adipocytes or blood cells (Delatte, Deplus, and Fuks 2014; Lio and Rao 2019).

Several studies have investigated the dynamics of active demethylation processes (Figure 13) (Yamaguchi, Hong, et al. 2013; Kohli and Zhang 2013; X. Hu et al. 2014). The direct removal of the methyl group from 5mC is considered as an unlikely event. Indeed, the presence of the methyl group on 5mC makes it thermodynamically highly stable and such a process would require a high amount of energy (X. Wu and Zhang 2017). For that reason, intermediates are needed to achieve demethylation. A possibility is to break the glycosyl bond between the ribose and the base. This event can be catalysed by a DNA glycosylase and would result in an abasic site as the intermediate, which can then be removed and further replaced (J.-K. Zhu 2009). The replacement with a new, unmodified cytosine is performed by the base excision repair (BER) machinery. Studies have suggested that the glycosylases can recognize different intermediates. A first one is created when 5hmC undergoes deamination by AID/APOBEC proteins and to become 5-hydroxymethyluracile (5hmU) (J. U. Guo et al. 2011). Another possibility is that 5hmC might be further oxidized by TET enzymes to become 5-formylcytosine (5fC) and 5-carboxycytosine (5caC), which can then be targeted by glycosylase (Y. F. He et al. 2011). 5hmC has also been suggested to promote passive demethylation through the

Figure 13: DNA demethylation process. Passive demethylation can occur through cell divisions: the modification is not maintained and is lost over time. Active demethylation involves enzymes such as the TETs which oxidize 5mC. Bhutani et al.

29 impairment of DNMT1 activity (Valinluck and Sowers 2007). However, 5hmC seems to be more than just an intermediate to demethylation as several reader proteins, which affect gene expression, have been described (R. Chen et al. 2017; Sequeira and Vermeulen 2020).

1.4.7 The 5hmC machinery

The enzymes that catalyse the oxidation of 5mC to 5hmC are called ten-eleven translocation (TET) methylcytosine dioxygenases and include TET1, TET2 and TET3 (Delatte, Deplus, and Fuks 2014). TETs can also oxidise 5hmC into further forms: formylcytosine (5fC) and 5-carboxylcytosine (5caC) (X. Wu and Zhang 2017). Structurally, TET enzymes contain two (for TET2) or three domains and two binding sites for co-factors. The C-terminal core catalytic domain is constituted by a double-stranded β-helix (DSBH) domain, a cysteine-rich domain and binding site for Fe (II) and 2-oxoglutarate (2-OG), which are cofactors for the oxidation (Figure 14).

The isocitrate dehydrogenases 1 and 2 (IDH1 and 2) are responsible for the formation of 2-OG. Frequently mutated in myeloid disorders such as acute myeloid leukaemia, these mutated enzymes produce an oncometabolite called 2-hydroxyglutarate (2-HG) (Hui Yang et al. 2012). This product directly competes with the correct substrate of dioxygenases such as TET enzymes and inhibit their functions in consequence.

Several studies suggested that the core catalytic domain binds to cytosines in a CpG context with higher affinity, but that surrounding DNA sequences were not determined yet (L. Hu et al. 2015, 2013; Rasmussen and Helin 2016). Both TET1 and TET3 contain a third domain able to directly bind DNA: the N-terminal CXXC zinc finger domain (Kohli and Zhang 2013). Although TET2 does not contain any CXXC domain, a gene coding for a CXXC-containing protein called

IDAX (or CXXC4) is actually present next to the TET2 gene. It has been suggested that TET2 and IDAX used to be one gene, which has been evolutionary split (Dunican, Pennings, and Meehan

2013).

30 (Deplus et al. 2013a). Moreover, TET1 is also involved in gene repression by the recruitment of repressor factors such as SIN3A (Williams et al. 2011).

TET1 has been extensively studied for its role in neurogenesis (R. R. Zhang et al. 2013; Kim et

al. 2016; Choi et al. 2019). It has indeed been shown that TET1 is required for the establishment of the neural progenitor cell (NPC) pool in the hippocampus, which is critical for memory (R. R. Zhang et al. 2013). Furthermore, in psychotic patients, TET1 has been found upregulated in the parietal cortex, highlighting its important role in the adult brain (Dong et al. 2012). About its developmental role, TET1 has been shown to be involved in the massive DNA demethylation process of embryonic germ cells (Yamaguchi, Shen, et al. 2013). Interestingly, TET1 has also been suggested to participate to promoter demethylation of the “master” pluripotency gene NANOG (Ito et al. 2010). Nevertheless, the implication of TET1 in the maintenance of pluripotency remains controversial and is still in debate (Dawlaty et al. 2011).

TET2 has been particularly studied for its role in haematopoiesis, because in this gene several

mutations have been found in hematopoietic diseases (Quivoron et al. 2011; Delhommeau et al. 2009; Moran-Crusio et al. 2011). Furthermore, genetic depletion of Tet2 in T cells seems to affect the differentiation of CD4+ T cells towards T helper (Th) cells (Ichiyama et al. 2015). Studies of TET3 highlighted its role in brain functions and zygote development (Shen et al. 2014; T. Li et al. 2014; H. Yu et al. 2015). Although all three TETs are present in the brain, TET3 seems to be the highest expressed and is essential for neurone differentiation (T. Li et al. 2014).

Concerning TET3’s implication in development, it has been observed to be especially expressed in oocytes and the zygote, but its expression rapidly decreases through cell divisions

Figure 14: TET enzymes structure. All three TETs contain a catalytic domain made of a cysteine rich domain, co-factors binding sites and a double-stranded β-helix domain. In addition, in the exception of TET2, TETs present a CXXC domain to

31 (Tan and Shi 2012). Of note, TET3 is responsible for the hydroxymethylation in the paternal genome at the beginning of embryogenesis (T. P. Gu et al. 2011).

1.4.8 5hmC distribution and consequence on gene expression

5hmC has been located in several regions of the genome. In genes, it can be found at promoters, regulatory sites such as enhancers and exons of highly expressed genes (Figure 15) (Stroud et al. 2011; Jeschke, Collignon, and Fuks 2016). The relationship between the presence of 5hmC and gene expression is not always straightforward. Overall, hydroxymethylation has been associated with active transcription, the level of 5hmC being particularly high in gene bodies of active genes (D. Q. Shi et al. 2017). Nevertheless, this correlation is not always verified and can vary among cell types. For instance, in ES cells and neural progenitor cells, highly expressed genes do not display 5hmC in promoters while lowly expressed genes show the opposite (Tan et al. 2013). In addition, when we focus on gene bodies, 5hmC accumulation positively correlates with active transcription in ESCs, but negatively in NPCs. Interestingly, TET1 has been suggested as especially responsible for the enrichment of 5hmC at the promoter/TSS sites while TET2 might rather catalyse 5hmC in gene bodies of active genes (Yun Huang et al. 2014). Furthermore, 5hmC is also enriched at the “bivalent” promoters previously mentioned (Verma et al. 2018).

In conclusion, the impact of hydroxymethylation on gene expression is still unclear and not as straightforward as DNA methylation. It seems to be differentially regulated among tissues and could then depend on each cell-type specific regulatory network (D. Q. Shi et al. 2017).

32

1.4.9 Hydroxymethylome mapping technologies

Unfortunately, the previously-described BS-Seq technique, developed to map DNA methylation, cannot be used to map hydroxymethylation (Y. Li and Tollefsbol 2011). This method is indeed unable to distinguish 5mC from 5hmC as both modifications protect from the conversion of C to T after bisulfite treatment. This questions the reliability of this method to map 5mC, however 5mC is much more abundant in the genome than 5hmC and thus, BS-seq is still commonly used as DNA methylome mapping technology. Moreover, an optimisation of this method now allows to distinct 5mC and 5hmC after bisulfite conversion.

To do so, a first chemical oxidation of 5hmC (into 5fC) is performed before bisulfite conversion. This allows the conversion of both unmodified C and previous 5hmC to be converted to T. The optimized method is called oxidative bisulfite sequencing (oxBS-Seq) (Figure 16) (Booth et al. 2013). By comparison of BS-Seq and oxBS-Seq profiles, it is possible to distinguish between 5mC and 5hmC. Another variation of the BS-Seq is the Tet-assisted bisulfite sequencing (TAB-Seq) where 5mC is the residue made sensitive to bisulfite-conversion. 5mC is indeed oxidized by TET enzymes into 5CaC while 5hmC is protected beforehand by a glycosylation step (Skvortsova et al. 2017).

Another commonly used technique is based on immunoprecipitation, similar to the previously-described MeDIP-seq for mapping of DNA methylation. 5hmC containing sequences can be targeted through different methods: (i) by an anti-5hmC antibody (hMeDIP), (ii) by the addition of a biotin-tag on 5hmC to perform a biotin pulldown (hMe-Seal) or finally (iii) by bisulfite treatment, which will convert 5hmC into cytosine 5-methylenesulphonate (CMS), which is then targeted by a specific antibody (Skvortsova et al. 2017; Rauluseviciute,

Figure 16: Bisulfite-based methods to map 5mC and/or 5hmC. The classical bisulfite-sequencing cannot distinguish 5mC and 5hmC. To do so, pre-treatments can either sensitize 5hmC (OxBS-seq) or 5mC (TAB-seq) to conversion. Adapted from

33 Drabløs, and Rye 2019). The backdraw of these methods are their resolution, which is lower than that of bisulfite-based sequencing protocols. Nevertheless, their costs and ready-to-use kits are clear advantages, especially for processing large sample cohorts.

Finally, the nanopore-based technology is suitable to map hydroxymethylation or other DNA modifications, however this method requires complex bioinformatics approaches (Qian Liu et al. 2019).

1.4.10 Pathological implications of 5hmC

When DNA hydroxymethylation was discovered as a new DNA modification, several studies explored its implications in cancers (Takai et al. 2014; Uribe-Lewis et al. 2015; Yuji Huang et al. 2016). As 5hmC has been linked to cell differentiation for several cell-types, the hypothesis that dysregulations of the 5hmC machinery could lead to cancer was logical. A lot of these studies indeed highlighted the dysregulation of 5hmC in cancers and they mostly observed a global loss of 5hmC in tumors (Figure 17) (Jeschke, Collignon, and Fuks 2016). This drop of hydroxymethylation in cancer can be explained by different mechanisms. TETs can undergo mutations as observed in leukaemia or melanoma, but might also be downregulated through promoter hypermethylation (as described in head and neck cancer) or microRNAs interference (Delhommeau et al. 2009; H. Yang et al. 2013; Chuang et al. 2015). Other possibilities include post-translational modifications, which can alter TETs stabilities or localization, dysregulation of co-factors or finally inhibition of their catalytic activities (Jeschke, Collignon, and Fuks 2016).

The global decrease of hydroxymethylation has been well described in several cancers such as melanoma, lung cancer or breast cancer and has been linked to silencing of tumor suppressor genes (TSG) (Lian et al. 2012; K. Chen et al. 2016; Thienpont et al. 2016). Without hydroxymethylation to induce demethylation, some TSGs are repressed. Since several studies have shown that TETs can inhibit metastasis and tumour proliferation in cancers such as brain, colorectal or liver cancers, TETs are often described as tumour suppressors themselves (M. Sun et al. 2013; C. H. Hsu et al. 2012; Duan et al. 2017). However, more recent studies also suggested the opposite: TET enzymes can also be observed as oncogenes. For instance, TET1

34 can be upregulated by hypoxia, which is known to promote tumorigenesis and can induce the expression of hypoxia-responsive genes (Tsai et al. 2014). Moreover, some cancers have also been reported to display an increase of 5hmC (Hao Huang et al. 2013; Navarro et al. 2014).

TET2 is one of the most mutated genes in hematopoietic malignancies and this seems to be

an early event in tumorigenesis (Yimei Feng et al. 2019; Ferrone, Blydt-Hansen, and Rauh 2020). Genomic alterations of TET2 have been frequently linked to the loss of 5hmC. For instance, genetic alterations of TET2 are present in around 30% of myelodysplastic syndrome (MDS) and 10 % of acute myeloid leukaemia (AML) patients (Ferrone, Blydt-Hansen, and Rauh 2020). However, TETs can also be affected in other ways than through mutations in hematopoietic diseases. As an example, TET1 has been observed to be silenced by hypermethylation in non-Hodgkin B cell lymphoma (B-NHL) (Cimmino et al. 2015a). Dysregulation of the TETs can further be mediated by their interaction with oncometabolites such as 2-hydroxyglutarate in leukaemia, as previously mentioned in 1.4.7 (W. Xu et al. 2011). Indeed, the oncometabolite 2-hydroxyglutarate can bind the TET catalytic domain in place of the substrate 2OG and thereby decrease TET activity. Interestingly and in line with TETs dual role, TET1 has been described as an oncogene in MLL rearranged leukaemia where it is genetically fused with the protein Mixed Lineage Leukaemia (MLL) after translocation. In this pathology, TET1 is then directed by MLL to oncogenes (Hao Huang et al. 2013).

5hmC levels or TET enzymes expression has been suggested as diagnostic and prognosis markers. For instance, low levels of 5hmC in glioma have been associated with poor survival (Orr et al. 2012). Another example is the expression of TET3 in head and neck carcinomas where poor survival is linked with hypermethylation of TET3 promoter (Misawa et al. 2018). While I just described several examples of hydroxymethylation dysregulations in cancer, it is important to stress that aberrant 5hmC has also been observed in other diseases than cancers such as psychosis, depression or allergies (Dong et al. 2012; Gross et al. 2017; H. Li et al. 2019). Therefore, they might provide new therapeutic opportunities in these diseases as well.

1.4.11 Other covalent DNA modification

Only a few DNA modifications such as 5mC and 5hmC have been well characterized in humans. Nowadays, the other DNA modifications for which we have subsequent information are the other two oxidized forms of 5hmC (5fc and 5Cac) and finally very recently, 6mA (Q. Zhu, Stöger, and Alberio 2018). To date, their functions and dynamics are poorly understood and thus, further investigations are warranted.

35 and differentiation of mouse ES cells. This modification has indeed been observed as enriched at promoters and exons of active genes during differentiation of ES cells (Raiber et al. 2012; C. Zhu et al. 2017). Also, 5fC has been observed to display a highly tissue-specific pattern, suggesting a role in development (Iurlaro et al. 2016). In comparison of 5fC, 5Cac is less documented. However, 5CaC might display a role in differentiation as its levels seem to increase during neural and hepatic cell differentiation (Wheldon et al. 2014; Lewis et al. 2017). Interestingly, a recent study implicated both 5fC and 5CaC in prostate cancer. 5fC is described as more abundant in a subtype of prostate cancer called ERG+ where it might be associated to good prognosis (Storebjerg et al. 2018). Concerning 5caC, higher levels have been observed - in both ERG− and ERG+ subtypes. Moreover, 5fC dysregulations have also been described in early liver cancer where it seems to be decreased (Jiao Liu et al. 2019).

Notably, it is still challenging to discriminate functions for 5fC and 5CaC independent of their role in active demethylation and more studies are needed to unravel their specific roles.

1.4.12 Methylation of adenine

DNA N6-methyladenine (6mA) has been mostly studied in prokaryotes where it is highly abundant (Messer and Noyer-Weidner 1988). However, its presence in human was not confirmed until recently (C. Le Xiao et al. 2018). Although 6mA seems to be lower abundant than 5mC, the 2018 study observed that around 0.05% of adenines were methylated in autosomal chromosomes. The X and Y chromosomes contained only 0.02% of 6mA, while the most enriched DNA was mitochondrial with almost 0.2%. About the 6mA distribution, this same study localized the modification in introns and intergenic regions in particular (26.19% and 70.61%, respectively) (C. Le Xiao et al. 2018). However, when they focused on genes, 6mA was enriched in exons and associated with gene expression. The writer of 6mA has been identified as N6AMT1, while ALKBH1 has been proposed to be the eraser of the mark. Biological functions of 6mA have just began to be investigated, but already uncovered several implications. It has first been suggested that this mark was involved in transposon silencing (T. P. Wu et al. 2016). In addition, 6mA seems indeed to be, as 5mC and 5hmC, implicated in developmental processes as 6mA levels increase after fertilization (to 0.2% 6mA/A) and decrease (< 0.001%) during embryogenesis (M. Zhang et al. 2020). Moreover, environmental stress seems to elevate 6mA in the mouse brain. Interestingly, 6mA has already been shown to be dysregulated in cancer as its global level seems to be decreased in liver cancer. This downregulation of 6mA might promote tumorigeneses and has been associated with poor prognosis (C. Le Xiao et al. 2018). Of note, higher 6mA levels have also been observed in the highly malignant brain cancer glioblastoma (Xie et al. 2018).

36

1.5 Crosstalks in Epigenetics

Many studies have demonstrated that different epigenetic layers engage in complex crosstalk. For instance, some histone marks such as H3K27me3 recruit DNMTs to induce methylation (Figure 18) (Fuks 2005). Another study has suggested that H3K9 methylation promotes DNA methylation (Q. Zhao et al. 2016). This crosstalk between histone and DNA modifications allows for regulation of local gene repression through tightening of chromatin. The opposite has also been observed: 5mC can shape histone marks. For example, 5mC can be bound by MBDs proteins, such as MBD1, which can recruit the SETDB1 histone methyltransferase and repress transcription (Mahmood and Rabbani 2019).

Another example of epigenetic crosstalk has been revealed by our lab in 2013 (Deplus et al. 2013b). This study highlighted the interaction between the TET enzymes and the histone glycosylase OGT. TETs can indeed act as scaffold proteins to increase OGT activity and guide this writer to promoters where it can glycosylate H2B and other substrates. Among these targeted substrates is also HCF1, a subunit of an H3K4 methyltransferase called the SET1/COMPASS complex. Thanks to its glycosylation, HCF1 stabilizes the complex and induces H3K4 tri-methylation, which is an active mark of transcription.

In conclusion, epigenetic modifications are highly dynamic and can regulate each other in order to regulate gene expression. As the number of such modifications (for histone or DNA) and enzymes involved is consequent, the probabilities of these crosstalks are considerable and currently highly investigated.

Figure 18: Examples of epigenetic crosstalk. DNA methylation can regulate histone modifications for instance through recruitment of enzymes such as HDACs or DNMTs. The opposite is also possible: histone modifications can induce DNA

37

1.6 RNA modifications

Analogous to DNA and histone modifications, another mechanism able to regulate gene expression has recently been uncovered: RNA epigenetics, also called epitranscriptomics (N. Liu and Pan 2015). This new and emerging field explores the function of RNA modifications and sustains a novel level of regulation, i.e. the fine-tuning of gene expression. Over 160 RNA modifications have been described affecting all kinds of RNAs such as ribosomal RNAs (rRNAs) or transfer RNAs (tRNAs). This thesis will focus on modifications present on messenger RNA (mRNAs) (Figure 19) (Boccaletto et al. 2018). Up to date, at least 7 mRNA modifications, including mrC, hmrC and m6A, and have been described in humans. These mRNA modifications have been shown to impact cell functions such as transcription, translation, splicing, localization and stability (B. S. Zhao, Roundtree, and He 2016).

This new epigenetics pillar further illustrates the complexity gene regulation and is currently attracting a lot of attention in research due to its therapeutic potential.

1.6.1 Methylation of adenosine

The first internal mRNA modification discovered in humans was the methylation of adenosine (m6A) (Gilbert, Bell, and Schaening 2016). It has been found in approximately 25% of mRNAs (with a mean of 2 to 3 modified residues per transcript) and represents at 0.1 to 0.4% of all adenosines (Ries et al. 2019; M. Chen and Wong 2020). As the most prevalent mRNA modification in eukaryotes, it is also the most studied so far. Those studies uncovered many consequences of m6A on mRNAs: m6A can influence their stability, transport from the nucleus, translation or splicing (Figure 20). Concerning its impact on physiological processes,

38 m6A has been involved in embryonic development, adipogenesis and stem cell regulation among others (M. Chen and Wong 2020).

Although to a lesser extent, m6A can also be found in other RNA types. Only two sites containing m6A have been observed in rRNA (A1832 in 18S rRNA and A4220 in 28S) (van Tran et al. 2019). Tran et al. also identified the writers METTL5 for 18S and ZCCHC4 for 28S rRNA. However, little is known about the rest of the machinery and the functions. Some small nuclear and nucleolar RNAs have been reported to be m6A modified, especially several small nuclear RNAs involved in splicing (Knuckles and Bühler 2018). In addition, several long-non coding RNAs (lncRNAs) can bear m6A modifications, especially those that undergo alternative splicing, which implies a role in isoform regulation (S. Xiao et al. 2019). Interestingly, two well-known lncRNAs have been identified as decorated with m6A: X-inactive specific transcript (XIST) and metastasis-associated lung adenocarcinoma transcript 1 (MALAT1) (Dominissini et al. 2012). For MALAT1, a consequence of m6A has been described as “m6A switch”. The presence of m6A might indeed inhibit the binding of proteins to MALAT1 by alterating its secondary structure (N. Liu et al. 2015; K. I. Zhou et al. 2016).

m6A has also been described on microRNA (miRNAs) where it can promote miRNA maturation (Alarcón, Lee, et al. 2015). Of note, it has further been suggested that some miRNAs might be responsible for the deposition of m6A on mRNA through binding with m6A writers (T. Chen et al. 2015). Finally, transfer RNA (tRNAs), which are known to be heavily modified, showed surprisingly only one m6A modified site in Escherichia coli, while almost nothing is known in eukaryotes (B. S. Zhao, Roundtree, and He 2016).

39

1.6.2 m6A machinery

The methylation of adenosine on mRNA is co-transcriptionally added by a writer complex formed by methyltransferase like 3 (METTL3), METTL14, WTAP (an adaptor protein) and possible regulatory subunits such as RBM15 (Figure 21) (L. He et al. 2019). Thanks to the SAM binding domain of METTL3 and 14, the complex is able to transfer a methyl group to the N-6 position of the adenosine. m6A can be removed by two erasers belonging to the ALKBH enzyme family. The first m6A eraser that was identified was the fat mass and obesity-associated protein (FTO) in 2011 (Jia et al. 2011). In a similar manner as TETs, FTO might also be able to catalyse the formation of further oxidized forms of m6A such as 6-hydroxymethyladenosine and 6-formyladenosine (Fu et al. 2013). The name of FTO has been assigned after the observation that FTO isoforms were associated with childhood obesity (Farooqi 2011). Indeed, FTO might promote obesity through increasing of food intake and adipogenesis while decreasing energy consumption (Fischer et al. 2009; Church et al. 2010). In 2013 a second m6A eraser has been identified: the alkB homolog 5 protein (ALKBH5) (Zheng et al. 2013). Interestingly, FTO and ALKBH5 tissue expression can vary as FTO is higher expressed in brain and adipose tissues, while ALKBH5 is particularly present in the testes (X.