DNA hypomethylation represents the most prevalent epigenetic alteration in human cancer, yet its cellular origins remain incompletely understood. In lung adenocarcinoma (LUAD), DNA hypomethylation is associated with upregulation of distinct gene groups, including: cancer-germline (CG) genes normally restricted to testicular germ cells, and stratified epithelium (SE) genes typically expressed in multilayered epithelia. Among the latter, FAM83A has emerged as a pro-tumoral gene implicated in cancer progression and therapy resistance. Here, we asked whether FAM83A activation in LUAD represents an epigenetic abnormality or if it reflects redirection of malignant cells towards an existing epithelial program. By analyzing single-cell RNA sequencing data from normal lung we found that, while FAM83A is undetectable in whole lung tissue homogenates, it is expressed in subsets of airway epithelial cells corresponding to basal/suprabasal and goblet cell populations. In contrast, CG gene MAGEA1 is silent in all lung cell types. Consistently, RT-qPCR detected FAM83A, but not MAGEA1, in primary and immortalized human lung basal cells. Bisulfite sequencing showed that the promoter of FAM83A, but not MAGEA1, is unmethylated in these cells. Analyses of transcriptomic datasets from LUAD cell lines and tissues demonstrated that FAM83A expression is correlated with markers of suprabasal and goblet lineages in malignant cells. They further revealed co-expression of FAM83A with NAPSA, a marker of alveolar AT2 cells, from which LUAD originates. Together, these findings indicate that, whereas MAGEA1 activation in LUAD results from an aberrant process of DNA demethylation, FAM83A upregulation in LUAD reflects redirection of alveolar cells towards an airway differentiation program.

Human cancer-germline (CG) genes are a group of testis-specific genes that become aberrantly activated in various tumors. Ongoing studies aim to understand their functions in order to evaluate their potential as anti-cancer therapeutic targets. Evidence suggests the existence of subcategories of CG genes, depending on location on autosomal or sex chromosomes, reliance on DNA methylation for transcriptional regulation, and profile of expression during gametogenesis and early embryogenesis. To clarify this issue, we developed CTexploreR, a R/Bioconductor package that integrates an up-to-date reference list of human CG genes (n = 146) with multiple bulk and single-cell methylomic and transcriptomic datasets. Based on promoter methylation profiles and responsiveness to a DNA methylation inhibitor, 74% of the CG genes were classified as DNA methylation dependent (Methdep). Intriguingly, most X-linked CG genes (69/70) fell into this category, thereby implicating DNA methylation dependency in the well-documented over-representation of testis-specific genes on the X chromosome. We further observed that, whereas X-linked Methdep CG genes become demethylated and activated in pre-spermatogonia in the fetal testis, most of them resist DNA demethylation in female germ cells and remain therefore silent in fetal and adult oocytes. Importantly, a number of X-linked Methdep CG genes (e.g., FMR1NB, GAGE2A, MAGEB2/C2, PAGE2, VCX3A/B) maintained this maternal-specific imprinting after fertilization, and were expressed exclusively in female preimplantation embryos, which inherit a paternal X chromosome. Together, our study using the CTexploreR package has allowed us to show that X-linked CG genes undergo transient maternal imprinting and contribute therefore to transcriptional sexual dimorphism in early embryos.

Human tumors progress in part by accumulating epigenetic alterations, which include gains and losses of DNA methylation in different parts of the cancer cell genome. Recent work has revealed a link between these two opposite alterations by showing that DNA hypomethylation in tumors can induce the expression of transcripts that overlap downstream gene promoters and thereby induce their hypermethylation. Preliminary in silico evidence prompted us to investigate if this mechanism applies to the locus harboring AGO1, a gene that plays a central role in miRNA biogenesis and RNA interference. Inspection of public RNA-Seq datasets and RT-qPCR experiments show that an alternative transcript starting 13.4 kb upstream of AGO1 (AGO1-V2) is expressed specifically in testicular germ cells, and becomes aberrantly activated in different types of tumors, particularly in tumors of the esophagus, stomach, and lung. This expression pattern classifies AGO1-V2 into the group of “Cancer-Germline” (CG) genes. Analysis of transcriptomic and methylomic datasets provided evidence that transcriptional activation of AGO1-V2 depends on DNA demethylation of its promoter region. Western blot experiments revealed that AGO1-V2 encodes a shortened isoform of AGO1, corresponding to a truncation of 75 aa in the N-terminal domain, and which we therefore referred to as “∆NAGO1”. Interestingly, significant correlations between hypomethylation/activation of AGO1-V2 and hypermethylation/repression of AGO1 were observed upon examination of tumor cell lines and tissue datasets. Overall, our study reveals the existence of a process of interdependent epigenetic alterations in the AGO1 locus, which promotes swapping between two AGO1 protein-coding mRNA isoforms in tumors.

Tumoral transformation processes sometimes include activation of unscheduled gene expression programs in the cancer cells. This is best exemplified by the so-called cancer-testis (CT) genes, a group of genes expressed in testicular germ cells that become activated in tumors of various somatic origins, through a process of DNA demethylation. Here, we explored the possibility that other tissue-specific gene clusters may become ectopically activated in tumors. Lung adenocarcinoma (LUAD) was used as a model, as all necessary transcriptomic datasets were available, including that of AT2 cells, the cell-of- origin of LUAD. We found that besides CT genes, an abundant group of genes expressed in the brain (CB genes, n=63) or in both brain and testis (CBT genes, n=28) become aberrantly activated in LUAD cell lines and tissues. Interestingly, activation of CB and CBTgene clusters was also detected in various other tumor types. Most CB/CBT genes appeared to exert neuronal functions. Moreover, a significant number of them encode antigens involved in neurological paraneoplastic syndromes. Neither neuroendocrine transdifferentiation, which occurs in 10-20% LUAD, nor DNA demethylation appeared to be responsible of the ectopic activation of CB and CBTgene clusters. Instead, prediction tools and depletion experiments identified the REST repressor as a regulator of a number of CB/CBT genes. Conclusion: Together, our data indicate that tumor development is associated with aberrant activation of a brain gene expression program, supporting the assumption that acquisition of neuronal functions might contribute to malignancy.

Genome-wide loss of DNA methylation is commonly observed in human cancers, but its impact on the tumor transcriptome remains ill-defined. Previous studies demonstrated that this epigenetic alteration causes aberrant activation of a germline-specific gene expression program. Here, we examined if DNA hypomethylation in tumors also leads to de-repression of gene clusters with other tissue specificities. To this end, we explored transcriptomic and methylomic datasets from human lung adenocarcinoma (LUAD) cell lines, normal lung, and lung alveolar type II cells, considered as the origin of LUAD. Interestingly, DNA demethylation in LUAD cell lines was associated with activation of not only germline-specific (CG) genes, but also gene clusters displaying specific expression in the gastrointestinal tract (GI), or in stratified epithelia (SE). Consistently, genes from all three clusters showed highly specific patterns of promoter methylation among normal tissues and cell types, and were generally sensitive to induction by a DNA demethylating agent. Analysis of TCGA datasets confirmed that demethylation and activation of CG, GI and SE genes also occurs in vivo in LUAD tumor tissues, in association with global genome hypomethylation. For genes of the GI cluster, we demonstrated that HNF4A is a necessary factor for transcriptional activation following promoter demethylation. Interestingly, expression of several SE genes, in particular FAM83A, correlated with both tumor grade and reduced patient survival. Together, our study uncovers novel cell-type specific gene clusters that become aberrantly activated in LUAD tumors in association with genome-wide hypomethylation.

Metabolic plasticity in cancer cells makes use of metabolism-targeting agents very challenging. Drug-induced metabolic rewiring may, however, uncover vulnerabilities that can be exploited. We report that resistance to glycolysis inhibitor 3-bromopyruvate (3-BrPA) arises from DNA methylation in treated cancer cells and subsequent silencing of the monocarboxylate transporter MCT1. We observe that, unexpectedly, 3-BrPA-resistant cancer cells mostly rely on glycolysis to sustain their growth, with MCT4 as an essential player to support lactate flux. This shift makes cancer cells particularly suited to adapt to hypoxic conditions and resist OXPHOS inhibitors and anti-proliferative chemotherapy. In contrast, blockade of MCT4 activity in 3-BrPA-exposed cancer cells with diclofenac or genetic knockout, inhibits growth of derived spheroids and tumors in mice. This study supports a potential mode of collateral lethality according to which metabolic adaptation of tumor cells to a first-line therapy makes them more responsive to a second-line treatment.

Tumor development involves alterations in DNA methylation patterns, which include both gains (hypermethylation) and losses (hypomethylation) in different genomic regions. The mechanisms underlying these two opposite, yet co-existing, alterations in tumors remain unclear. While studying the human MAGEA6/GABRA3 gene locus, we observed that DNA hypomethylation in tumor cells can lead to the activation of a long transcript (CT-GABRA3) that overlaps downstream promoters (GABRQ and GABRA3) and triggers their hypermethylation. Overlapped promoters displayed increases in H3K36me3, a histone mark deposited during transcriptional elongation and known to stimulate de novo DNA methylation. Consistent with such a processive mechanism, increases in H3K36me3 and DNA methylation were observed over the entire region covered by the CT-GABRA3 overlapping transcript. Importantly, experimental induction of CT-GABRA3 by depletion of DNMT1 DNA methyltransferase, resulted in a similar pattern of regional DNA hypermethylation. Bioinformatics analyses in lung cancer datasets identified other genomic loci displaying this process of coupled DNA hypo/hypermethylation, and some of these included tumor suppressor genes, e.g. RERG and PTPRO. Together, our work reveals that focal DNA hypomethylation in tumors can indirectly contribute to hypermethylation of nearby promoters through activation of overlapping transcription, and establishes therefore an unsuspected connection between these two opposite epigenetic alterations.

Telomeric repeat-containing RNA (TERRA) molecules play important roles at telomeres, from heterochromatin regulation to telomerase activity control. In human cells, TERRA is transcribed from subtelomeric promoters located on most chromosome ends and associates with telomeres. The origin of mouse TERRA molecules is, however, unclear, as transcription from the pseudoautosomal PAR locus was recently suggested to account for the vast majority of TERRA in embryonic stem cells (ESC). Here, we confirm the production of TERRA from both the chromosome 18q telomere and the PAR locus in mouse embryonic fibroblasts, ESC, and various mouse cancer and immortalized cell lines, and we identify two novel sources of TERRA on mouse chromosome 2 and X. Using various approaches, we show that PAR-TERRA molecules account for the majority of TERRA transcripts, displaying an increase of two to four orders of magnitude compared to the telomeric 18q transcript. Finally, we present a SILAC-based pull-down screen revealing a large overlap between TERRA-interacting proteins in human and mouse cells, including PRC2 complex subunits, chromatin remodeling factors, DNA replication proteins, Aurora kinases, shelterin complex subunits, Bloom helicase, Coilin, and paraspeckle proteins. Hence, despite originating from distinct genomic regions, mouse and human TERRA are likely to play similar functions in cells.

The human genome harbors many duplicated segments, which sometimes show very high sequence identity. This may complicate assignment during genome assembly. One such example is in Xq28, where the arrangement of 2 recently duplicated segments varies between genome assembly versions. The duplicated segments comprise highly similar genes, including MAGEA3 and MAGEA6, which display specific expression in testicular germline cells, and also become aberrantly activated in a variety of tumors. Recently, a new gene was identified, CT-GABRA3, the transcription of which initiates inside the segmental duplication but extends far outside. According to the latest genome annotation, CT- GABRA3 starts near MAGEA3, with which it shares a bidirectional promoter. In an earlier annotation, however, the duplicated segment was positioned in the opposite orientation, and CT-GABRA3 was instead coupled with MAGEA6. To resolve this discrepancy, and based on the contention that genes connected by a bidirectional promoter are almost always co-expressed, we decided to compare the expression profiles of CT-GABRA3, MAGEA3, and MAGEA6. We found that in tumor tissues and cell lines of different origins, the expression of CT-GABRA3 was better correlated with that of MAGEA6. Moreover, in a cellular model of experimental induction with a DNA demethylation agent, activation CT-GABRA3 was associated with that of MAGEA6, but not with that of MAGEA3. Together these results support a connection between CT-GABRA3 and MAGEA6 and illustrate how promoter-sharing genes can be exploited to resolve genome assembly uncertainties.

Adult cardiac progenitor cells (CPCs) display a low capacity to differentiate into cardiomyocytes in injured hearts, strongly limiting the regenerative capacity of the mammalian myocardium. To identify new mechanisms regulating CPC differentiation, we used primary and clonally expanded Sca-1+ CPCs from murine adult hearts in homotypic culture or coculture with cardiomyocytes. Expression kinetics analysis during homotypic culture differentiation showed downregulation of Wnt target genes concomitant with increased expression of the Wnt antagonist, Wnt inhibitory factor 1 (Wif1), which is necessary to stimulate CPC differentiation. We show that the expression of the Wif1 gene is repressed by DNA methylation and regulated by the de novo DNA methyltransferase Dnmt3a. In addition, miR-29a is upregulated early during CPC differentiation and downregulates Dnmt3a expression, thereby decreasing Wif1 gene methylation and increasing the efficiency of differentiation of Sca-1+ CPCs in vitro. Extending these findings in vivo, transient silencing of Dnmt3a in CPCs subsequently injected in the border zone of infarcted mouse hearts improved CPC differentiation in situ and remote cardiac remodeling. In conclusion, miR-29a and Dnmt3a epigenetically regulate CPC differentiation through Wnt inhibition. Remote effects on cardiac remodeling support paracrine signaling beyond the local injection site, with potential therapeutic interest for cardiac repair.

Global loss of DNA methylation is frequently observed in the genome of human tumors. Although this epigenetic alteration is clearly associated with cancer progression, the way it exerts its pro-tumoral effect remains incompletely understood. A remarkable consequence of DNA hypomethylation in tumors is the aberrant activation of "cancer-germline" genes (also known as "cancer-testis" genes), which comprise a diverse group of germline-specific genes that use DNA methylation as a primary mechanism for repression in normal somatic tissues. Here we review the evidence that such cancer-germline genes contribute to key processes of tumor development. Notably, several cancer-germline genes were found to stimulate oncogenic pathways involved in cell proliferation (SSX, DDX43, MAEL, PIWIL1), angiogenesis (DDX53), immortality (BORIS/CTCFL), and metastasis (CT-GABRA3). Others appear to inhibit tumor suppressor pathways, including those controlling growth inhibition signals (MAGEA11, MAGEB2), apoptosis (MAGEA2, MAGEC2), and genome integrity (HORMAD1, NXF2). Cancer-germline genes were also implicated in the regulation of tumor metabolism (MAGEA3/MAGEA6). Together, our survey substantiates the concept that DNA hypomethylation promotes tumorigenesis via transcriptional activation of oncogenes. Importantly, considering their highly restricted pattern of expression, cancer-germline genes may represent valuable targets for the development of anti-cancer therapies with limited side effects.

BACKGROUND:Many human tumors show aberrant activation of a group of germline-specific genes, termed cancer-germline (CG) genes, several of which appear to exert oncogenic functions. Although activation of CG genes in tumors has been linked to promoter DNA demethylation, the mechanisms underlying this epigenetic alteration remain unclear. Two main processes have been proposed: awaking of a gametogenic program directing demethylation of target DNA sequences via specific regulators, or general deficiency of DNA methylation activities resulting from mis-targeting or down-regulation of the DNMT1 methyltransferase.RESULTS:By the analysis of transcriptomic data, we searched to identify gene expression changes associated with CG gene activation in melanoma cells. We found no evidence linking CG gene activation with differential expression of gametogenic regulators. Instead, CG gene activation correlated with decreased expression of a set of mitosis/division-related genes (ICCG genes). Interestingly, a similar gene expression signature was previously associated with depletion of DNMT1. Consistently, analysis of a large set of melanoma tissues revealed that DNMT1 expression levels were often lower in samples showing activation of multiple CG genes. Moreover, by using immortalized melanocytes and fibroblasts carrying an inducible anti-DNMT1 small hairpin RNA (shRNA), we demonstrate that transient depletion of DNMT1 can lead to long-term activation of CG genes and repression of ICCG genes at the same time. For one of the ICCG genes (CDCA7L), we found that its down-regulation in melanoma cells was associated with deposition of repressive chromatin marks, including H3K27me3.CONCLUSIONS:Together, our observations point towards transient DNMT1 depletion as a causal factor of CG gene activation in vivo in melanoma.

We previously demonstrated that APP epigenetically regulates Egr1 expression both in cultured neurons and in vivo. Since Egr1 is an immediate early gene involved in memory formation, we wondered whether other early genes involved in memory were regulated by APP and we studied molecular mechanisms involved. By comparing prefrontal (PF) cortex from wild type (APP+/+) and APP knockout mice (APP-/-), we observed that APP down regulates expression of four immediate early genes, Egr1, c-Fos, Bdnf and Arc. Down regulation of Egr1, c-Fos and Bdnf transcription resulted from a decreased enrichment of acetylated histone H4 on the corresponding gene promoter. Further characterization of H4 acetylation at Egr1 and c-Fos promoters revealed increased acetylation of H4K5 and H4K12 residues in APP-/- mice. Whereas APP affected Egr1 promoter activity by reducing access of the CREB transcription factor, its effect on c-Fos appeared to depend on increased recruitment of HDAC2 histone deacetylase to the gene promoter. The physiological relevance of the epigenetic regulation of Egr1 and c-Fos gene transcription by APP was further analyzed following exposure of mice to novelty. Although transcription of Egr1 and c-Fos was increased following exposure of APP+/+ mice to novelty, such an induction was not possible in APP-/- mice with a high basal level of expression of these immediate early genes. Altogether, these results demonstrate that APP-mediated regulation of c-Fos and Egr1 by different epigenetic mechanisms is needed for their induction during exposure to novelty.

Epigenetic programming within the cardiac stem cell (CSC) niche controls CSC specification and differentiation, but the identity of putative mediators is poorly characterized. To dissect these mechanisms, we used primary and clonally expanded Sca-1+/c-kit+ CSC from murine adult hearts and stimulated cardiomyocyte differentiation upon culture in a differentiation medium (DIFF) containing 5'-Azacytidine and transforming growth factor β1 or co-cultured with rat cardiomyocytes. With this model, we previously showed that inducible deletion of β-catenin enhanced CSC differentiation in vitro and in vivo. Accordingly, we detected a constitutive activity of Wnt/b-catenin pathway in undifferentiated CSC and a downregulation of Wnt target genes (in % of ctl: Axin2: 36±4%; Snai2: 46±7%; n=3; p<0.01) in CSC treated with DIFF medium, associated with an upregulation of Wnt antagonist, Wif-1 (310±57%; n=3, p<0.05). Like several Wnt/b-catenin repressor genes, Wif-1 gene expression is DNA methylation sensitive and susceptible to be regulated by the de novo DNA methyltransferases Dnmt3. Indeed, Dnmt3a was downregulated in CSC treated with DIFF, while siRNA targeting Dnmt3a upregulated Wif-1 expression in non-differentiated cells (425±254%; n=4, p<0.05) and promoted CSC cardiac differentiation in co-culture assay (assessed by quantitative expression of cTnT; 162±7% vs. siRNA-scramble; n=7, p<0.001). In parallel, we found an early upregulation of miR-29a (444±43%; n=3, p<0.01), a well-known regulator of Dnmt3a, in DIFF-treated CSC. Indeed, early treatment of CSC with miR-29a mimic decreased Dnmt3a protein level and increased Wif-1 expression (839±257%; n=4, p<0.05). Reciprocally, CSC treatment with anti-miR-29a (LNA-miR-29a; vs scramble ctl) upregulated Dnmt3a and downregulated Wif-1 expression (51±9%; n=7, p<0.01). Importantly, this treatment also significantly decreased CSC differentiation in co-culture assay (63±4% vs. LNA-scramble; n=11, p<0.001). Altogether, this suggests that upregulation of miR-29a promotes differentiation (and blockade of miR-29a inhibits differentiation) through Dnmt3a-dependent regulation of the Wnt inhibitor, Wif-1. In support of this, co-treatment with anti-miR-29a and siRNA-Dnmt3a abrogates the above effect both on Wif-1 expression and CSC differentiation (compared to treatment with respective scramble, n=7). We conclude that differentiation of CSC involves the epigenetic regulation of canonical Wnt/β-catenin activity through miR-29a and Dnmt3a-dependent control of expression of the endogenous inhibitor, Wif-1, possibly amenable to therapeutic modulation for cardiac repair.

Genome hypomethylation is a common epigenetic alteration in human tumors, where it often leads to aberrant activation of a group of germline-specific genes, commonly referred to as "cancer-germline" genes. The cellular functions and tumor promoting potential of these genes remain, however, largely uncertain. Here, we report identification of a novel cancer-germline transcript (CT-GABRA3) displaying DNA hypomethylation-dependent activation in various tumors, including melanoma and lung carcinoma. Importantly, CT-GABRA3 harbors a microRNA (miR-105), which has recently been identified as a promoter of cancer metastasis by its ability to weaken vascular endothelial barriers following exosomal secretion. CT-GABRA3 also carries a microRNA (miR-767) with predicted target sites in TET1 and TET3, two members of the ten-eleven-translocation family of tumor suppressor genes, which are involved in the conversion of 5-methylcytosines to 5-hydroxymethylcytosines (5hmC) in DNA. Decreased TET activity is a hallmark of cancer; here, we provide evidence that aberrant activation of miR-767 contributes to this phenomenon. We demonstrate that miR-767 represses TET1/3 mRNA and protein expression and regulates genomic 5hmC levels. Additionally, we show that high CT-GABRA3 transcription correlates with reduced TET1 mRNA levels in vivo in lung tumors. Together, our study identified a cancer-germline gene that produces microRNAs with oncogenic potential. Moreover, our data indicate that DNA hypomethylation in tumors can contribute to reduced 5hmC levels via activation of a TET-targeting microRNA.