DNA methylation patterns separate senescence from transformation potential and indicate cancer risk

Summary

Overall shared DNA methylation patterns between senescence (Sen) and cancers have led to the model that tumor promoting epigenetic patterns arise through senescence. We show that transformation-associated methylation changes arise stochastically and independently of programmatic changes during senescence. Promoter-hypermethylation events in transformation involve primarily pro-survival and developmental genes, similarly modified in primary tumors. Senescence-associated hypermethylation mainly involve metabolic regulators, appears early in proliferating “near-senescent” cells which can be immortalized but are refractory to transformation. Importantly, a subset of transformation-associated hypermethylated developmental genes exhibits highest methylation gains at all age-associated cancer risk states across tissue-types. These epigenetic changes favoring cell self-renewal and survival, arising during tissue aging, are fundamentally important for stratifying cancer risk and concepts for cancer prevention.

eTOC Blurb

Xie et al. show that transformation-associated methylation changes arise stochastically and evolve independently of senescence. A subset of transformation-associated hypermethylated genes favoring cell self-renewal and survival exhibits highest methylation gains during aging and early tumorigenesis.

Introduction

DNA methylation abnormalities involving global losses and focal gains at promoter CpG-islands (CpGIs), are a hallmark of cancers (Baylin and Jones, 2011; Baylin and Jones, 2016; Jones and Baylin, 2007; Rodriguez-Paredes and Esteller, 2011; Shen and Laird, 2013). Hundreds of genes are hypermethylated at promoter CpGI in virtually every patient’s cancer and are transcriptionally silenced or, more often, blocked from normal induction compared to the normal cell counterparts. These include key tumor suppressor genes (Baylin and Jones, 2016; Shen and Laird, 2013; Xia et al., 2017). Similar global patterns of hypomethylation and focal CpGI hypermethylation occur during replicative cellular senescence (Cruickshanks et al., 2013; Zane et al., 2014). Senescence is the growth-arrest response to various stress stimuli, the most prominent being telomere shortening during multiple cell division cycles, genotoxic stress and oncogene-induced mitogenic signals (Campisi and d’Adda di Fagagna, 2007). Activating senescence to arrest proliferative cells at risk for neoplastic transformation is a major tumor suppressive mechanism (Campisi, 2001; Kuilman et al., 2010; Michaloglou et al., 2005; Rodier and Campisi, 2011).

Recent reports have highlighted the similarity in epigenetic alterations during tumorigenesis and senescence, and suggested that cancer DNA methylation patterns are similar to those seen in cells undergoing replicative cellular senescence (Cruickshanks et al., 2013; Lowe et al., 2015). It has thus been hypothesized that the epigenetic changes acquired during senescence may facilitate cell transformation if cells escape senescence (Cruickshanks et al., 2013). However, it is puzzling how an epigenetic landscape set for tumor suppression can simultaneously be utilized, or is perhaps of no consequence, for cellular transformation. To dissect these questions, here we have performed comparative genome-wide DNA methylation analysis during cellular transformation and senescence.

Results

Experimental design and characterization of transformation and senescence phenotypes

We used Weinberg’s classical transformation system (Figure 1A) (Hahn et al., 1999) to study the evolution of DNA methylation and gene expression patterns. Early passage (EP) BJ fibroblast cells were sequentially infected with the human telomerase catalytic subunit (hTERT), the simian virus 40 large T antigen (SV40) and the H-Ras oncoprotein (H-rasV12) to establish the immortalization-transformation lineage and, as control, with empty vectors (EV) for the above (Figure 1B). These latter cells also serve, to measure cells entry into senescence (Figures 1A–1B and 1E).

Figure 1. Experimental design and characterization of transformation and senescence phenotypes.

(A) Schema shows the experiment design. EV: empty vector; Near-Sen: near-senescence; Sen: senescence.

(B) Timeline of early transformation, mouse xenografts, Near-Sen and Sen with respect to population doublings (PDs) of Near-Sen and Sen are shown. Dashed lines indicate the time point that cells were early transformed or became near-senescent or mouse xenografts were harvested or cells become fully senescent. The PDs at which cells became near-senescent (PD14) and fully senescent (PD28) were shown in red.

(C) Soft agar assay shows colony formation of Near-Sen, hTERT, SV40 and HRAS cells. Three biological replicates are shown.

(D) The volume of mouse xenografts was measured and plotted with the time after cells injection. hTERT_3 and SV40_3 are negative controls.

(E) Senescence-associated β-galactosidase staining of EP, Near-Sen and Sen cells. Three biological replicates are shown. Scale bar, 200 μm.

See also Figure S1.

As previously shown (Hahn et al., 1999), cells expressing hTERT and SV40 large T antigen become immortalized but are not tumorigenic (Figures 1C–1D). Introducing H-rasV12 as a third gene leads to all three biological replicates becoming fully transformed after roughly one month (HRAS cells) (Figures 1A–1C). These HRAS cells, with varying intensities and frequency, form colonies in soft agar and form tumors as xenografts (HRAS_X) in NOD-SCID immune deficient mice, with HRAS_3 being the most aggressive replicate and HRAS_1 the least (Figures 1C–1D and S1A). Finally, a very important phenotype explored in this system is the state of near-senescence (Near-Sen) (Figures 1A–1B) (Cruickshanks et al., 2013; Zou et al., 2004), wherein the empty vector cells exhibit the senescence feature of β-galactosidase positive staining (Dimri et al., 1995) at about 28 days or approximately 14 population doublings (PD14) but remain in a proliferating state during this period (Figures 1B, 1E, and S1B).

Globally similar but locally distinct DNA methylation patterns in early transformation and senescence

Overall, DNA methylation patterns assayed by the Infinium HumanMethylation450 array (Sandoval et al., 2011), diverge globally in the EP cells, various stages of immortalization-transformation and senescence (Figures S2A-S2F show replicate 2, S2G and S2H show replicate 1 and replicate 3 respectively). The changes begin as early as the introduction of hTERT (immortalization) (Figures S2A-S2C) and continue to accumulate until the cells become transformed (Figures S2D-S2H, panel (a)). Similar changes from EP occur during senescence development with divergent patterns appearing in the near-senescence stage cells as they continue to divide at a similar rate as the EP cells for approximately another 14 PDs, and the patterns are maintained until development of the fully growth-arrested senescence phenotype (Figures S2D-S2H, panels (b-c)). Similar to the cancer epigenome, both senescent and transformed cells globally exhibit widespread losses of methylation (Δβ ≤ −0.2 with respect to EP) coupled with simultaneous increases in proximal promoter CpGIs (Δβ > 0.2 with respect to EP) (Figures S2A-S2F). In addition, gene body and many enhancer regions undergo both gains and losses of DNA methylation with a bias towards losses (Δβ ≤ −0.2 with respect to EP) (Figures S2A-S2F).

Previous reports argue, based on analysis of overall DNA methylation patterns, that aberrantly methylated genes in cancer cells may arise from a senescence intermediate, and thus that senescence-associated DNA methylation may facilitate malignant transformation once senescence is bypassed (Cruickshanks et al., 2013). However, in our analyses, although overall global gains and losses in DNA methylation are similar, at individual genomic regions the methylation patterns are very different for senescence versus transformation. In phyloepigenetic comparisons, changes at promoter CpGI probes evolve very differently between the immortalization-transformation versus the Near-Sen and senescent replicates (Figure 2A). In the latter two states, the methylation patterns evolve away from EP replicates but remain close to each other while the hTERT replicates form independent branches and diverge from each other. Moreover, SV40 and HRAS replicates are widely different from each other and form independent branches. Finally, the xenograft replicates (HRAS_1_X, HRAS_2_X, HRAS_3_X) evolve further from the parental HRAS replicates. Similar phyloepigenetic relationships are observed for the gene body or enhancer probes (data not shown). A heatmap of the most variable methylation probes further highlights the above relationships showing close similarity between the near-senescence and senescence probes as compared to the immortalized-transformed replicates (Figure 2B). Importantly, in addition to previous observations that senescence-associated DNA methylation changes are highly reproducible (Koch et al., 2013), we show virtually identical senescence and near-senescence patterns (Pearson’s r > 0.8) (Figures 2C and S2I-S2P), indicating a potentially defined “program” of methylation changes distinct from stochastic changes during transformation, take effect early during senescence development (Figure 2A). These results starkly contrast with a similar analysis comparing senescence versus hTERT (Figures 2D and S2I-S2P) or the early transformed HRAS cells (Figures 2E and S2I-S2P), wherein methylation changes diverge dramatically as compared to senescence. Thus, immortalization on the path to malignant transformation involves stochastic epigenetic patterns from which cells contributing to transformation may evolve.

Figure 2. Distinct loci-level methylation changes in early transformation compared to senescence.

(A) A phyloepigenetic tree was constructed to infer the evolutionary relationships of replicative cellular senescence lineage and immortalization-transformation lineage. Promoter CpGI probes were used.

(B) Heatmap shows methylation levels of promoter CpGI probes with Standard Deviation (SD) ≥ 0.15 across all samples (top variable probes).

(C) Smoothscatterplot shows comparison of methylation level (β value) between Sen and Near_Sen at promoter CpGIs. Scatterplot shows comparison of methylation deviation (Δβ value) of Near_Sen and Sen from EP at promoter CpGIs. Pearson’s correlation coefficient is shown. Diagonal represents identity line. The 2nd replicate serves as representative.

(D) Same analysis as those in Figure 2C, but compared Sen with hTERT.

(E) Same analysis as those in Figure 2C, but compared Sen with HRAS.

See also Figure S2.

Within the above patterns of methylation changes, a substantial number of CpGI loci that are methylated in hTERT or HRAS have low methylation in senescent cells and vice versa (Figures 2D–2E, S2K, and S2N). Thus, most CpGI loci that gain methylation in hTERT or HRAS cells do not change in senescent cells (Figures 2D–2E, Pearson’s r ≈ 0; S2K, and S2N) or near-senescent cells (Figure S2Q), and vice versa. Similar distinctions are evident between early immortalization-transformation and near-senescence or senescence cells at other genomic regions including, gene bodies and enhancer elements (Figures S2I-S2J, S2L-S2M, S2O-S2P, and S2R-S2S). In summary, while global gains and losses in DNA methylation patterns appear at first glance to be similar between transformation and senescence, they are very different at individual gene loci and CpG site, and thus immortalization and transformation induce a substantially different epigenetic state compared to that in senescence.

Since the earlier study relating senescence methylation changes with cancer had used a different fibroblast cell-type (foetal lung fibroblast, IMR90) (Cruickshanks et al., 2013), we compared the methylation profiles of the senescent states in the IMR90 (IMR90_Sen) and BJ (BJ_Sen) fibroblasts. CpG-site level methylation patterns of the two senescent cells are substantially different across the genome and at promoter CpGI (Figure S2T). However, a significant portion of genes (220 genes) do overlap at their CpGI promoters in IMR90_Sen and BJ_Sen, which as discussed later below, are enriched for genes involved in biosynthetic and metabolic processes (Figure S2V). Further, the IMR90_Sen methylation pattern, as for the BJ_Sen methylation pattern, starkly differ from the methylation patterns in the BJ transformation HRAS cells (BJ_HRAS) (Figure S2U). Thus, although cell type origin-specific methylation differences exist in BJ and IMR90 cells senescence, there is also conservation in the biological processes of genes which overlap in the two cell types.

Dissecting the functional categories of promoter CpGI hypermethylated genes in transformation and senescence

The gene ontology functional categories of genes with promoter CpGI hypermethylation events in transformation and senescence differ dramatically. Thus, a set of 917 genes with hypermethylated promoters, relative to EP cells, in any of the three transformation replicates (HRAS-Specific Methylated, HSM) (Figure 3A) are most enriched for a group of normally inducible development and differentiation regulators (Figure 3C and Tables S1-S4). We and others have previously shown these gene types are highly biased to have cancer-specific promoter CpGI hypermethylation (Easwaran et al., 2012; Gal-Yam et al., 2008; Ohm et al., 2007; Schlesinger et al., 2007; Widschwendter et al., 2007). Importantly, these genes are normally maintained at low poised transcription state in embryonic stem cells (ESC) and tissue-specific progenitor cells to maintain a balance between self-renewal and lineage commitment, but are inducible during differentiation cues (Easwaran et al., 2012; Ohm et al., 2007; Schlesinger et al., 2007; Widschwendter et al., 2007). This poising for induction is not normally controlled by promoter DNA methylation but by polycomb chromatin, which in the ESCs mostly occur in the setting of bivalent chromatin involving both H3K4me3 and H3K27me3 marks at the promoters (Berman et al., 2012; Easwaran et al., 2012; Gal-Yam et al., 2008; Ohm et al., 2007; Schlesinger et al., 2007; Widschwendter et al., 2007). Thus cancer-specific promoter DNA methylation renders these genes difficult to induce, which potentially may contribute to the cancer cell phenotype. In contrast to the above genes, the bulk of 491 near-senescent and senescent-specific methylated (SSM) genes are enriched for processes involved in positive regulation of biosynthetic and metabolic processes (Figures 3A, 3C, and Tables S1-S4). The silencing or prevention of induction of these genes may then be related to a slowing of metabolism as cells enter senescence. In the above analyses, the genes that are methylated across the HRAS replicates ideally represent the potentially different combinations that evolve in sub-clones and dominate the overall phenotypes. Gene enrichment analysis of the methylated genes that are shared amongst all HRAS replicates (40 genes) or the Senescence replicates (442 genes) (Figures 3A and S3A) corroborates the findings that the genes methylated during transformation are enriched for the developmental genes while those enriched in senescence regulate biosynthetic and metabolic processes.

Figure 3. Dissecting the nature of genes with promoter CpGI hypermethylation events in transformation and senescence.

(A) Top or middle Venn diagram shows comparison of promoter CpGI hypermethylated genes among HRAS_1, HRAS_2 and HRAS_3 or among Sen_1, Sen _2 and Sen _3. Bottom Venn diagram shows comparison of promoter CpGI hypermethylated genes between HRAS_all (all promoter CpGI hypermethylated genes from 3 HRAS replicates) and Sen_All (all promoter CpGI hypermethylated genes from 3 Sen replicates) and p-value was calculated by Fisher’s exact test. HSM, HRAS-Specific Methylated genes; SSM, Senescence-Specific Methylated genes; CM, Common Methylated genes.

(B) Simulation of independent stochastic methylation events shows the ratio of observed overlaps of genes methylated in each replicate of HRAS or HRAS_X and genes methylated in senescence compared to that of the expected overlaps of randomly selected genes with genes methylated in senescence. Slightly different numbers in Figure 3A and Figure 3B is because the overlaps in Figure 3A shows all CpGI genes while in Figure 3B only the CpGI genes present in the gene expression array were used. O:E indicates observed overlaps : expected overlaps.

(C) Gene Ontology (GO) analysis of HSM, SSM and CM genes. Heat map of −log10(p-values) is shown. Green stars indicate biological processes involved in cell development and differentiation.

(D) The methylation level (β value) of probes that are located nearby or within individual genes (CDH1 and SOX17) in EP (black), HRAS and HRAS_X (blue), and Sen (red) are shown. Early transformation-associated hypermethylated sites indicated in red rectangles; senescence-associated ones are in green rectangles.

See also Figure S3, Tables S1-S4.

Intriguingly, there is significant overlap of hypermethylated genes between transformation and senescence. These commonly methylated (CM, 346) genes (Fisher’s test p-value=1.92e-132, odds ratio=7.87, 95% CI= [6.73, 9.2]) (Figure 3A and Tables S2-S3) get hypermethylated more frequently than expected by chance in independent transformation replicates with an average of ≈2.5-fold greater overlap than expected in random sampling (Figure 3B). Even upon long-term growth of the HRAS replicates as xenografts, when stochastic methylation continues to evolve and the number of hypermethylated genes increases 3-fold compared to early HRAS replicates (Figures S3C-S3E), the observed to expected overlap is still ≈2-fold higher (Figure 3B). This above overlap of CM genes may represent promoter methylation events potentially subject to selection in transformation, and in senescence. Indeed, most importantly, this set of overlapping CM genes is significantly enriched for developmental regulators (Figure 3C and Table S4). Included in these are several key cancer methylated genes, such as CDH1, SOX17, MGMT, ATOH1, SFRP4 and WIF1 (Figures 3D and S3F) (Bossuyt et al., 2009; Esteller et al., 2000; Graff et al., 1995; Licchesi et al., 2008; Suzuki et al., 2004; Xia et al., 2017; Zhang et al., 2008).

Human embryonic stem cell (hESC) bivalent genes are methylated and downregulated in early transformation and xenograft replicates

The observations in the previous sections raise the question of relationships between the methylation changes in senescent and transformed/xenograft cells and expression of the involved genes. Interestingly there is marginal, but statistically significant (p-value < 0.05), inverse relationship between methylation and expression in relation to the EP cells only in early transformation and xenograft cells (Figure S4A) but not in the senescent cells (Figure S4B). The latter is consistent with previous work that senescence-associated DNA methylation does not globally regulate gene expression changes (Cruickshanks et al., 2013). However, as stressed earlier, the key relationship may be suppression of induction, rather than reducing basal expression of all these above genes in the setting of the promoter methylation. A key consideration here is the previously discussed relationships of enrichment of cancer methylated genes for genes with polycomb repressive H3K27me3 in the context of bivalent marks in human embryonic (hESC), as well as hematopoietic and mesenchymal stem cells (Bernstein et al., 2006; Easwaran et al., 2012; Kim et al., 2010). Enrichment for genes with bivalent promoters in hESC, and also proliferating BJ fibroblast (ENCODE data), hold true for the CM genes (Figure 4A, p-value = 3.468e-17, odds ratio=2.6, 95% CI= [2.08, 3.25] and Figure S4C) or the HSM genes (Figure 4A, p-value = 1.812e-17, odds ratio=1.88, 95% CI= [1.63, 2.17]; Figure S4C). Importantly, in both early transformation and xenograft populations, hESC and proliferating BJ fibroblast bivalent genes tend to be downregulated by at least 1.5-fold in relation to the EP cells, while this is not observed with senescent cells (Figures 4B–4C and S4D-S4E). Further, the specific sets of CM and HSM genes, but not the SSM genes, that are bivalent marked in hESC or proliferating BJ fibroblast cells are more downregulated in transformation and xenograft cells (Figures 4D–4F and S4F-S4H). The fact that many of these genes have low expression to begin with in the hESC or EP cells (Figure S4I), which is typical of the polycomb marked/bivalent genes, but interestingly are only downregulated in the transformation lineages further emphasizes the potential roles of methylation in driving a tighter and permanent silencing during transformation. Thus, in concordance with our earlier studies (Easwaran et al., 2012), the BJ transformation model involves downregulation of developmental regulator genes that are generally polycomb/bivalent marked in embryonic stem and progenitor cells, and a significant proportion of these genes acquire promoter hypermethylation during transformation.

Figure 4. hESC bivalent genes are methylated and downregulated in the early transformation and xenograft replicates.

(A) Venn diagrams (top, middle or bottom) show comparisons of CM, HSM or SSM genes with hESC bivalent genes. P-values were calculated with Fisher’s exact test.

(B) Heatmap shows the relative expression of 2923 hESC bivalent genes in HRAS, HRAS_X and Sen with respect to EP. The relative expression levels were determined using a log2 scale of fold change.

(C) Quantitative measuring the number of 2923 hESC bivalent genes that are upregulated (1.5-fold increase compared to EP) and downregulated (1.5-fold decrease compared to EP) in Sen, HRAS and HRAS_X.

(D) Quantitative measuring the number of 155 CM-bivalent genes that are upregulated (1.5-fold increase compared with EP) and downregulated (1.5-fold decrease compared with EP) in Sen, HRAS and HRAS_X.

(E) Quantitative measuring the number of 334 HSM-bivalent genes that are upregulated and downregulated in Sen, HRAS and HRAS_X.

(F) Quantitative measuring the number of 182 SSM-bivalent genes that are upregulated and downregulated in Sen, HRAS and HRAS_X.

See also Figure S4.

One key point from above data is that the developmental genes discussed are not consistently methylated in all of the early transformation replicates but yet tend to have downregulated expression. It has been shown that abnormal gene expression in cancer is controlled by Polycomb (PcG) chromatin modifications and DNA methylation, and thus hypothesized that this represents a spectrum of gene silencing along which there is a molecular progression from preceding abnormal PcG control to abnormal promoter DNA methylation (Baylin and Jones, 2011; Jones and Baylin, 2007). Interestingly in this regard, the 182 SSM-hESC bivalent genes or 164 SSM-EP bivalent genes show no bias towards gene upregulation or downregulation in the early transformation and mouse xenografts replicates (Figures 4F and S4H). Thus, key regulator genes controlling cell differentiation and development are downregulated during transformation, and a significant proportion of these genes are promoter methylated, highlighting the importance of stochastic methylation in mediating gene expression programs involved in the cancer phenotype.

CM genes show maximal gains in promoter methylation during aging and cancer

The methylation patterns we outline raise an intriguing question of how they relate to aging, the biggest demographic risk factor for cancer (Klutstein et al., 2016). Genes with promoter hypermethylation in cancers have been shown to acquire methylation during aging (Ahuja et al., 1998; Issa et al., 1994; Rakyan et al., 2010; Teschendorff et al., 2010) and senescence is well characterized to be related to aging (Campisi, 2013; van Deursen, 2014). Surprisingly, however, we find that genes which gain promoter DNA methylation during aging are mostly not SSM genes but rather are more likely those linked to transformation, intriguingly the CM genes. This is shown in studies below when the methylation status of transformation versus senescence specific genes are analyzed across various datasets of primary tumors and aging tissues.

We reasoned that the set of 346 CM genes, common between senescence and transformation, and enriched for developmental regulators, might be prone to hypermethylation in aging and cancer. Indeed, in comparison to either HSM or SSM genes, the CM genes have the highest tendency to be hypermethylated in different primary tumor tissues, but not corresponding normal tissues, and HSM genes are second in this regard (Figures 5A). In peripheral blood mononuclear cells (PBMNC) and skin samples from healthy elderly people (Heyn et al., 2012; Vandiver et al., 2015), CM and HSM genes again have the highest likelihood of hypermethylation (Figures 5B–5E and Table S5). Most importantly, comparison of methylation of the three gene groups across a panel of normal tissues collected in TCGA studies shows that CM genes starkly stand out the most for their gains of promoter methylation not only as tissues age but also as a function of aging-related risk for cancer (Figures 5F–5G and S5A). In accordance with these observations, CM genes accumulate maximal methylation at earliest tumor stages. Thus, in a breast cancer data set that has mapped methylation changes in normal breast and progressive ductal carcinoma in situ (DCIS) to invasive carcinoma (Fleischer et al., 2014), the CM genes get most methylated early on, and this pattern remains throughout tumor progression (Figure S5B). Majority methylation gains occur at early DCIS stages and continue to be maintained at the same levels during progressive tumor stages indicating that these methylation gains are early events during tumorigenesis. Various observations indicate that although the statistically significant global absolute mean differences of DNA methylation in the normal-tumor, young-old and progressive tumorigenesis stages comparisons are subtle (Δβ < 0.05), these differences are meaningful with biological implications. First, random sampling does not give the same difference (Figures 5B and 5D). Second, the methylation β value in these datasets measures average methylation in a heterogenous pool of cells collected from each individual normal human subject. Thus as a whole the average methylation only shows a slight increase (Klutstein et al., 2017) in such analyses but for our data, a significant change indicates the existence of subpopulations of cells in the tissue with evolving methylation in these HSM/CM probe regions. Third, corroborating our observations, recently it was shown that normal samples have few alleles/cells with methylation of consecutive CpGs in the CpGIs, while in majority of alleles/cells the CpGs are not methylated, in these same category of genes we have studied (Klutstein et al., 2017). Thus the average methylation only shows a slight increase. Fourth, low levels of methylation cause substantial gene silencing (Curradi et al., 2002; Hsieh, 1994). These observations support our hypothesis that subpopulations of cells with a combination of epigenetic events at critical tumor suppressors and developmental regulator genes may be more permissive to tumorigenesis (Easwaran et al., 2014). Taken together, these analyses show that CM genes are hotspots for DNA hypermethylation not only in primary tumors but also in aging tissues, and important for tracking cancer risk. Moreover, the genes in precursor cancer cells that most likely evolve such methylation patterns in aging tissue are those that may mediate retention of proliferating, aging cells and not those only facilitating senescence.

Figure 5. Relationships between hypermethylated senescence and transformation genes, aging and cancer.

(A) Beanplots show the methylation level (β values) of the three groups of genes (HSM, SSM and CM) for primary samples (T) in the Cancer Genome Atlas (TCGA) for Lung Adenocarcinoma (TCGA_LUAD), Lung Squamous Cell Carcinoma (TCGA_LUSC) and Colon Adenocarcinoma (TCGA_COAD) as well as their adjacent normal samples (N). P-values were calculated with Wilcoxon rank sum test.

(B) Beanplots show the methylation level (β values) of HSM, SSM and CM genes in newborn and peripheral blood (PBMNC) of individuals as a function of aging. 7000 promoter CpGI probes were randomly selected (Random sampling) and serve as a negative control. P-values were calculated with Wilcoxon rank sum test.

(C) Dotplots show the degree of overlapping between blood aging methylated genes and the three groups of genes. Fisher’s exact test was used to determine the p-value of overlapping between blood aging methylated genes and the HSM, SSM, and CM genes (Table S5). Minus log10 scale of p-values were plotted. Wilcoxon rank sum test p-values are shown (***P < 0.001; ****P < 0.0001; n.s., not significant).

(D) Beanplots show the methylation level (β values) of HSM, SSM and CM genes in epidermis samples from young (age < 35) versus elderly (with age > 60) individuals. 7000 promoter CpGI probes were randomly selected (Random sampling) and serve as a negative control. P-values were calculated with Wilcoxon rank sum test.

(E) Dotplots show the degree of overlapping between epidermis aging methylated genes and the three groups of genes. Fisher’s exact test was used to determine the p-value of overlapping between epidermis aging methylated genes and the HSM, SSM, and CM genes (Table S5). Minus log₁₀ scale of p-values were plotted. Wilcoxon rank sum test p-values are shown (*P < 0.05; **P < 0.01; n.s., not significant).

(F) Boxplot shows mean β value of probes in the promoter CpGIs of HSM, SSM and CM genes for every individual in each cancer risk group for normal human colon (other tissues shown in Figure S5A). The cancer risk indicates the percentage of individuals diagnosed with cancers in colon according to increasing age by decades. Wilcoxon rank sum test p-values are shown (*P < 0.05).

(G) Heatmaps show mean β value of probes in the promoter CpGIs of HSM, SSM and CM genes for all individuals in each age-associated cancer risk state across various tissues. The cancer risk (%) and corresponding age (risk/age) of the various tissues are shown at the right side of heatmaps. Color keys show the β value of probes.

See also Figure S5, Table S5.

Oncogene induced senescence has minimal DNA methylation changes

We find a fundamental difference in the epigenetic state of replicative cellular senescence and oncogene induced senescence (OIS). The latter senescence mode is highly relevant to the choice between resistance versus becoming addicted to oncogenic signaling (Bartkova et al., 2006; Lowe et al., 2004; Michaloglou et al., 2005; Sarkisian et al., 2007). OIS results from the acute expression of genes promoting oncogenesis which triggers, in about 10 days, an immediate growth arrest thus blocking replication and preventing oncogenic addiction to key mutations (Lowe et al., 2004; Michaloglou et al., 2005; Serrano et al., 1997). When we induce OIS by expressing H-rasV12 directly in early passage BJ cells (Figure S6), very minimal changes in DNA methylation patterns ensue with respect to either early transformation cells (Pearson’s r < 0.2) (Figure 6A) or replicative cellular senescence (Pearson’s r ≈ 0, Figure 6B). Thus cells that subsequently do acquire oncogenic addiction and enter tumor progression are those selected to bypass OIS, and possibly those with age-related DNA methylation patterns, which are extended during further transformation.

Figure 6. Oncogene induced senescence has minimal DNA methylation changes.

(A) Scatterplots show comparison of methylation alterations (Δβ value) of early transformed cells (HRAS) and oncogenes-induced senescent cells (OIS) with respect to EP at promoter CpGIs. Pearson’s correlation coefficients are shown. Diagonal represents identity line. Three biological replicates are shown.

(B) Same analysis as those in Figure 6A, but compare methylation alterations (Δβ value) of senescent cells (Sen) and oncogenes-induced senescent cells (OIS) with respect to EP.

See also Figure S6.

The Near-senescent cell state is permissive to immortalization but resistant to transformation

Replicative cellular senescence is believed to be an important mechanism for suppressing tumorigenesis (Campisi, 2001; Kuilman et al., 2010; Michaloglou et al., 2005; Rodier and Campisi, 2011). However, as we have introduced earlier, recent reports have suggested that senescence may also be a state from which cells can escape and participate in tumor promotion (Cruickshanks et al., 2013). Our data outlined in the above sections strongly suggests against this senescence bypass hypothesis. In this regard, selection of near-senescent cells in these studies is highly relevant and critical as it has been reported previously that growth-arrested phenotype of fully senescent cells cannot be reversed by hTERT (Beausejour et al., 2003). We find that near-senescent cells, while they can be forced to escape senescence to become immortalized, are not only resistant to transformation but continue to harbor the majority of DNA methylation patterns seen in senescent cells. When the same regimen used for achieving tumor progression (Figure 1A) is applied to the near-senescent cells, with good resultant protein expression of telomerase (Near-Sen-hTERT), SV40 (Near-Sen-SV40), and H-rasV12 (Near-Sen-HRAS) (Figure 7A), the growth rate of these cells increases with introduction of each of the above genes (Figure 7B). Yet, none of these cell populations, even those with the final addition of H-rasV12 (Near-Sen-HRAS), showed anchorage independent growth in soft agar (Figures 7C and 7D), which is a hallmark of H-rasV12 transformed EP BJ-fibroblasts (Figure 1C). Furthermore, at both genome-wide and promoter CpGI regions, the immortalized cells generated from the Near-Sen cells maintain the DNA methylation patterns of the near-senescent and the senescent cells (Figures 7E–7F and S7A-S7B). Across the genome, 73.5%, 80.7% and 80.7% of senescence-associated hypermethylated probes are in Near-Sen-hTERT, Near-Sen-SV40 and Near-Sen-HRAS (Figure 7E). Similarly, at promoter CpGI regions, 71.8%, 82.3% and 81.8% of hypermethylated probes in senescence are shared with Near-Sen-hTERT, Near-Sen-SV40 and Near-Sen-HRAS, respectively (Figure 7F) and this includes both the anti-proliferative (SSM) and the pro-survival (CM) genes (Figures S7C-S7F). On the other hand, compared to the above conservation of methylation with the senescent cells, there is a substantial difference in the overlap of the methylated probes between Near-Sen-hTERT, Near-Sen-SV40 or Near-Sen-HRAS and early transformation associated hTERT/SV40/HRAS at both whole genome and promoter CpGI regions (Figures S7G and S7H). Furthermore, only ~10% of HSM genes corresponding probes in any replicate of early transformation are present in Near-Sen-hTERT, Near-Sen-SV40 or Near-Sen-HRAS cells (Figures S7I-S7N). Thus, the retention of the majority of senescence-associated hypermethylation and the lack of transformation-associated hypermethylation in engineered near-senescent cells suggest that senescence-associated DNA methylation patterns may contribute to resistance against transformation.

Figure 7. The Near-senescent cell state is permissive to immortalization but resistant to transformation.

(A) Western blots show ectopic expression of hTERT, SV40 and H-RasV12 in near-senescent cells.

(B) Population doublings of near-senescent cells infected with retrovirus encoding EV, hTERT, hTERT and SV40 or hTERT and SV40 and H-RasV12 were measured and plotted with time (days).

(C) Soft agar assays show colony formation of near-senescent or EP cells infected with retrovirus encoding EV, hTERT, hTERT and SV40 or hTERT and SV40 and H-RasV12.

(D) Quantitatively measuring the colony number of Near_Sen_HRAS and EP_HRAS in Figure 7C.

(E) Venn diagram showing the comparison of hypermethylated probes (probes with Δβ > 0.2 when compared to EP) between Near_Sen_EV and Sen, Near_Sen_hTERT and Sen, Near_Sen_SV40 and Sen, Near_Sen_HRAS and Sen across whole genome.

(F) Same analysis as those in Figure 7E, but at promoter CpGIs.

See also Figure S7.

Discussion

We have outlined various key differences in DNA methylation changes during cellular transformation versus replicative cellular senescence that involve important biological processes. Senescence involves programmed methylation changes affecting CpGI-promoter hypermethylation of cellular biosynthesis and metabolism regulators, which may favor a gradual shutdown of biosynthetic processes. In contrast, promoter hypermethylation in transformation evolve stochastically and predominantly involve developmental regulators, which if not properly induced, would favor anti-differentiation and self-renewal mechanisms (Orkin and Zon, 2008). The programmed DNA methylation alterations in senescence could provide a key epigenetic contribution to the long observed Hayflick phenomenon through which human fibroblasts in culture inevitably enter the timed replicative senescent state (Hayflick and Moorhead, 1961). Most importantly, transformation-associated DNA methylation changes evolve completely independent of OIS or replicative senescence intermediate, in contrast to previous hypotheses (Cruickshanks et al., 2013) suggesting senescence epigenetic changes are not the route to eventual oncogene addiction and transformation.

A key finding is that senescence-associated methylation does not facilitate transformation, although near-senescent cells can be immortalized. In the two months the immortalized near-senescent cells were cultured, longer than the process taken to immortalize and transform EP cells, the degree of variation of methylation change was minimal and the patterns mostly overlapped those for near-senescent (Near-Sen-hTERT/SV40/HRAS) and senescent cells. Thus, there is either no change from the senescent methylation pattern or markedly slower rate of change than that accompanying EP immortalization-transformation. It will be interesting to see whether during more prolonged growth of the immortalized near-senescent cells, there is a gradual divergence of the near-senescence methylation patterns and whether this is accompanied by capacity to be transformed by H-rasV12. Relevant to this, we recently showed that immortalized bronchial epithelial cells exposed daily to cigarette smoke condensate (CSC) take about ten months to acquire similar epigenetic changes defined in the current immortalized-transformed cells. Most importantly, unlike for control bronchial cells not exposed to CSC, long-term CSC exposure resulted in immediate transformation upon expressing KRAS^G12D mutant protein (Vaz et al., 2017).

Previous studies have reported on the global similarities in gains and losses of DNA methylation in senescent cells and cancer samples from TCGA and other studies (Berman et al., 2011; Cruickshanks et al., 2013; Hansen et al., 2011). The cells used in the previous study, IMR90 foetal lung myofibroblasts, are different from our work with BJ foreskin, fibroblasts. The epigenome in the senescent states of these two cell types are different (Figure S2U), indicating cell-type specific programmed changes in the epigenome during senescence. In spite of the different cell types used, our conclusions that epigenetic changes during tumorigenesis evolve independent of senescent epigenetic state hold. Both the current work and our analyses of the data from previous work reveal significant methylation differences in isogenic senescent and transformed cells, as well as independent cancer data sets. The previous study, in contrast to ours, compared IMR90 senescent epigenome to non-isogenic cancer epigenomes from other studies. Importantly, although methylation patterns for senescence differ in BJ and IMR90 cells, in both cases they differ starkly from cancer associated promoter CpGI hypermethylated genes. Moreover, in both BJ and IMR90 data, there is ~20-30% overlap of senescence and transformation/cancer-associated hypermethylated genes/regions. Finally, we show that the senescence-associated methylation does not promote tumorigenesis. Thus, a series of experiments show that methylation alterations in transformed cells are acquired without need for a senescence intermediate, and that the senescence-related methylation does not promote transformation.

Our studies have important bearings on relationships between cellular aging, senescence, and cancer risk. Other studies note epigenetic similarities between aging and senescence and suggest the latter contributes to and is a hallmark of aging in a tissue context (Lopez-Otin et al., 2013; van Deursen, 2014). Both genome-wide losses and CpGI promoter methylation gains are well known to be associated with cancer risk, especially for colon (Issa et al., 1994). Our data suggest we might re-interpret how DNA methylation alterations associate with both aging and its relationship to cancer risk. In the present study, DNA methylation alterations in cancer appear to stochastically evolve during continued cell divisions but also closely track with previous observations for aging (Fraga et al., 2005; Issa, 2014). Accordingly, increased methylation at the CM and HSM genes during aging and early stages of tumorigenesis appears tightly linked, and the CM genes, especially emerge as potential biomarkers of cancer risk. Panels involving these latter genes may help stratify, for every age group, individuals with the highest risk for cancer development. In this context a recent study showed that variation in DNA methylation across tissues best explains the variability in lifetime cancer risk for different tissues (Klutstein et al., 2017), better than the previously shown stem cell division numbers (Tomasetti et al., 2017; Tomasetti and Vogelstein, 2015). Lastly, our work highlights the conceptual distinction that chronological aging and immortalization-transformation methylation (CM and HSM) are more similar to each other than to the programmed senescence-related methylation. Thus, while both senescent and proliferative aging cells could coexist in the same tissues, the risk for cancer appears to come from the latter.

STAR★Methods

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact Hariharan Easwaran (Heaswar2@jhmi.edu).

Experimental Model and Subject Details

Cell culture, transformation and senescence

Human foreskin fibroblast BJ cells were obtained from William C. Hahn’s lab. For transformation, we followed the procedures described previously (Hahn et al., 1999). Early passage or near-senescent BJ cells were sequentially infected with retrovirus packaged with pBabe-hygro-hTERT (Addgene: 1773), pBabe-zeo-SV40 large T genomic DNA (LT and ST antigens) (Addgene: 1778) and pBabe-puro-HrasV12 (Addgene: 1768). Infected cells were selected with 80 ug/ml hygromycin, 800ug/ml zeocin and 1 ug/ml puromycin for 3-5 days respectively. For near-senescence and senescence, early passage BJ cells were infected with mock retrovirus as a control and were cultured accordingly with the transformation processes. When we got the early transformed cells, early passage BJ cells roughly underwent 14 population doublings and become near-senescent (Figure 1B). These near-senescent cells were continuously cultured until they ceased to proliferate and got fully senescent. At this point, they roughly grew another 14 population doublings (Figure 1B). Near-senescent and senescent cells were tested with β-galactosidase staining assay (Cell Signaling Technology: 9860). Induction of oncogene-induced senescence (OIS) was performed as previously described (Narita et al., 2003). Briefly, early passage BJ cells were infected with retrovirus encoding HrasV12. Infected cells were selected with 1 ug/ml puromycin for 3 days, and roughly 10 days post-infection OIS cells were tested with β-galactosidase staining or harvested for DNA methylation analysis. All cell lines were cultured in medium with 66.3% Knockout DMEM (Thermo Fisher: 10829-018), 16.5% Medium 199 (Thermo Fisher: 11150-059), 14.5% FBS, 3.5mM glutamine and Pen/strep. We designated population doubling (PD) as 0 when we received cells from Dr. Hahn.

Mice

We used male NOD-SCID immune deficient mice (4-5 weeks old) (Jackson laboratories) for mouse xenograft assays. Mice were housed at the Johns Hopkins Animal care facility. All animal care and protocols followed were in accordance with guidelines of the institutional Animal Care and Use Committee (IACUC). All animal experiments were approved by the Johns Hopkins Animal Care and Use Committee. Mice were injected with 1×10⁶ early transformed cells and immortalized cells (hTERT_3 and SV40_3 as negative control) and tumor volumes were measured every week. 8 weeks after cell injection, all the mice were sacrificed and mice xenografts were harvested.

Method Details

Soft agar colony formation, tumorsphere formation

For soft agar colony formation assay, 10000 early transformed cells were seeded in 0.4% soft agar medium and cultured for 14-18 days. Cells were then fixed with 4% formaldehyde and stained with 1ug/ml ethidium bromide. For tumorsphere formation assay, 5000 early transformed cells were cultured in low attachment 6-well plates for 10 days.

Genomic DNA extraction and genome-wide DNA methylation analysis

Genomic DNA was extracted and purified from cell pellets using the Promega Wizard Genomic DNA Purification Kit according to the manufacturer’s instructions. Genomic DNA quality was assessed by low concentration agarose gel (0.6%) electrophoresis and spectrometry of OD260/280 and OD 260/230 ratio. DNA bisulfite conversion was carried out using EZ DNA Methylation Kit (Zymo Research) by following manufacturer’s manual with modifications for Illumina Infinium Methylation Assay. Briefly, 0.5 – 1.0 ug of genomic DNA was first mixed with 5 ul of M-Dilution Buffer and incubate at 37°C for 15 minutes and then mixed with 100 ul of CT Conversion Reagent prepared as instructed in the kit’s manual. Mixtures were incubated in a thermocycler with 16 thermal cycles at 95°C for 30 seconds and 50°C for one hour. Bisulfite-converted DNA samples were loaded onto 96-column plates provided in the kit for desulphonation and purification. Concentration of eluted DNA was measured using Nanodrop-1000 spectrometer. Bisulfite-converted DNA was analyzed using Illumina’s Infinium Human Methylation450 Beadchip Kit (WG-314-1001) by following manufacturer’s manual. Beadchip contains 485,577 CpG loci in human genome. Briefly, 4 ul of bisulfite-converted DNA was added to a 0.8 ml 96-well storage plate (Thermo Scientific), denatured in 0.014N sodium hydroxide, neutralized and amplified with kit-provided reagents and buffer at 37°C for 20-24 hours. Samples were fragmented using kit-provided reagents and buffer at 37°C for one hour and precipitated by adding 2-propanol. Re-suspended samples were denatured in a 96-well plate heat block at 95°C for 20 minutes. 12 ul of each sample was loaded onto a 12-sample chip and the chips were assembled into hybridization chamber as instructed in the manual. After incubation at 48°C for 16-20 hours, chips were briefly washed and then assembled and placed in a fluid flow-through station for primer-extension and staining procedures. Polymer-coated chips were image-processed in Illumina’s iScan scanner.

R (http://www.r-project.org) and the Bioconductor minfi package (Aryee et al., 2014; Maksimovic et al., 2012) were used to preprocess and normalize the raw data. β-values were computed based on following definition: β value = (signal intensity of methylation-detection probe)/(signal intensity of methylation-detection probe + signal intensity of non-methylation-detection probe + 100), ranging from 0 to 1.0 for each CpG site (methylation level from 0-100%, respectively). Probes with poor signals (p-value > 0.01) were not included. All probes were mapped to the Human Genome Assembly GRCh37/hg19, annotated with UCSC CpGI track and NCBI’s Reference Sequence (refseq) database. Chromosome X- and Y-linked probes were removed from subsequent analyses. The probes located within CpGIs and from −1500 bp to +1500 bp around transcription start sites (TSS) were called promoter CpGIs probes. Body probes were obtained according to the IlluminaHumanMethylation450k.db annotation package (Triche T and Jr. IlluminaHumanMethylation450k.db: Illumina Human Methylation 450k annotation data. R package version 2.0.9). The enhancer probes were from (Yao et al., 2015).

Identification of the differentially methylated probes (HSM, SSM probes): Since the HRAS clones per se are different/divergent (based on clustering/PCA), we used a filtering approach to obtain the HSM and SSM probes. For hypermethylated probes, we compared individual HRAS and Senescence replicates with EP and called any probe satisfying the following condition to be hypermethylated: (a) Δβ_i >= mean(EP_i) + 2*(SD(EP_i)); (b) Δβ_i >= 0.2; (c) mean(EP_i) <= 0.2, where Δβ_i is the difference in beta-values (β) between each HRAS replicate and the mean of the three EP replicates for every probe i (mean(EP_i)), SD(EP_i) is the standard deviation of the probe in the three EP replicates. For hypomethylated probes following conditions were used: (a) Δβ_i <= mean(EP_i) − 2*(SD(EP_i)); (b) Δβ_i <= −0.2; (c) mean(EP_i) >= 0.5, where Δβ_i is the difference in beta-values (β) between each Senescence replicate and the mean of the three EP replicates for every probe i.

Simulating random sampling of the promoter CpGI genes

Monte Carlo simulations to estimate the expected overlap with genes methylated in each of the senescence replicates for a randomly picked number of genes were performed as below. The number of genes selected randomly corresponds to the mean number of genes methylated in the HRAS replicates. The probability of picking each gene was dictated by the probability distribution function (PDF) of the expression levels in EP (basal expression level) of genes hyper-methylated in senescence. Thus, the probability was weighted such that low expressing genes have higher chance of getting picked compared to high expressing genes, which is the same probability of expression levels of genes hyper-methylated in senescence. To obtain the PDF of the expression levels in EP (basal expression level) of genes hyper-methylated in senescence, the mean expression values in EP from the expression data was obtained. PDF of these genes was estimated using approxfun function from the stats package in R. All these analyses were done using all CpGI genes which is present on the Agilent expression array.

TCGA primary tumor and aging tissues DNA methylation data analysis

TCGA lung adeonocarcinoma (LUAD), lung squamous carcinoma (LUSC) and colon adenoicarcinoma (COAD) DNA methylation data was obtained from the data matrix portal. The β-values of probes corresponding to HSM, SSM and CM genes in LUAD, LUSC and COAD were compared by Wilcoxon test.

Human young and elderly tissues (blood and epidermis) DNA methylation data were obtained from Gene Expression Omnibus (GEO) (GSE30870, GSE51954) (Heyn et al., 2012; Vandiver et al., 2015). For blood data, the peripheral blood mononuclear cells (PBMNC) from cord blood of newborn and whole blood of nonagenarian were used. For epidermis data, the sun protected groups were selected. The β-values of probes corresponding to HSM, SSM and CM genes in blood (PBMNC) (Newborn, Nonagenarian) and epidermis (Young, Elderly) were compared by Wilcoxon test. Blood (PBMNC) aging methylated genes refer to promoter CpGI genes that with β value < 0.2 in newborn cord blood (PBMNC) but with Δβ value > 0.2 in nonagenarian whole blood (PBMNC) with respect to the newborn. Epidermis aging methylated genes refer to promoter CpGI genes that with β value < 0.2 in young epidermis but with Δβ value > 0.2 in elderly epidermis with respect to the young. Fisher’s exact test was used to determine the p-values of overlapping between blood (PBMNC) or epidermis aging methylated genes and the three groups of genes (HSM, SSM and CM) (Table S5).

For the pan-tissue analysis of relationship of DNA methylation in HSM, SSM and CM genes to aging and cancer risk (obtained from https://seer.cancer.gov/archive/csr/1975_2010/browse_csr.php), normal tissue data from TCGA were obtained and normalized together by the functional normalization (funnorm) function in the Minfi Bioconductor/R package (Aryee et al., 2014). The risk of being diagnosed with cancers of the various tissues in the next ten years at different decades of age was obtained. The mean methylation beta value of probes in the promoter CpGIs of HSM, SSM and CM genes were computed for each patient in that age group. All patient samples in a certain decade of age were grouped together (for example, patients in their 30’s were all considered in the 30 age group, and so on). The mean beta values were then plotted as heatmaps and boxplots.

As above, the mean β value of probes in the promoter CpGIs of HSM, SSM and CM genes for the progressive breast cancer stages (Fleischer et al., 2014) were computed for the different groups of normal, ductal carcinoma in situ (DCIS), mixed DCIS-invasive and invasive carcinoma.

IMR90 bisulfite sequencing (BS-seq) data analysis

The bisulfite sequencing raw data for proliferating and senescent IMR90 cells were obtained from GEO (GSE48580) and analyzed using Bismark pipeline (Krueger and Andrews, 2011) according to the manual. The commands were: 1) fastq-dump –split-files <file_name.sra>, 2) trim_galore –paired –trim1 <file1.fq> <file2.fq>, 3) bismark -n 1 –bowtie2 <path_to_bowtie2_genome> <path_to_file_trimmed.fq>, 4) bismark_methylation_extractor -p –no_overlap <path_to_file.bam>, 5) bismark2bedGraph –CX <path_tofile_bismark_bt2_pe.txt> -o <path_to_output_file.bed>.

In order to compare the Illumina 450K array with the bisulfite-Seq data we counted all the methylated and unmethylated reads at the area between probe position and 50 bases on the left of the probe. We only selected the areas with total reads more than 5. The percentage of methylation at the probe position is: (Number of Methylated reads on 50 bases)/(Total number of reads on 50 bases). IMR90 promoter CpGI hypermethylated probes and genes are listed in Tables S1-S2.

BJ early passage (EP) bivalent genes analysis

The chromatin data for early passage BJ fibroblast cells are from the ENCODE project: H3K27me3 ChIP-seq data: https://www.encodeproject.org/experiments/ENCSR000DQG/; H3K4me3 ChIP-seq data: https://www.encodeproject.org/experiments/ENCSR000DQH/; Control ChIP-seq data: https://www.encodeproject.org/experiments/ENCSR000DQE/. EP bivalent genes were identified as previously described (Easwaran et al., 2012) and listed in Table S6.

Genome-wide genes expression analysis

RNA was extracted from EP, early transformed cells (HRAS_1, HRAS_2, HRAS_3), mouse xenograft (HRAS_1_X, HRAS_2_X, HRAS_3_X) and senescent cells (Sen_1, Sen_2) using TRIzol Reagent (Thermo Fisher Scientific: 15596026) and RNeasy Mini Kit (Qiagen: 74104) with DNase digestion. RNA was quantified with NanoDrop ND-1000 followed by quality assessment with 2100 Bioanalyzer (Agilent Technologies) according to manufacturer’s protocol. Sample amplification and labeling procedures were carried out by using Quick RNA Amplification and Labeling Kit (Cat# 5190-0447, Agilent Technologies) with minor modifications. Briefly, 400 ng total RNA was reverse-transcribed into cDNA by MMLV-RT using oligo dT primers (Cat# RA300A-2, System Bioscience) that incorporate T7 promoter sequence. The cDNA is then used as a template for in vitro transcription in the presence of T7 RNA polymerase and Cyanine-labeled CTP (Perkin Elmer). The labeled cRNA is purified using RNeasy mini kit (Qiagen). RNA spike-in controls (Agilent Technologies) are added to RNA samples before amplification and labeling according to manufacturer’s protocol. Agilent human GE 4x 44K v2 microarrays (G4845A) were used, which contain 41,000 unique probes targeting 27,958 Entrez gene RNAs. 825 ng of each Cyanine-labeled samples was used for hybridization at 65°C for 17 hours in a hybridization oven with rotation. After hybridization, microarrays are washed and dried according to Agilent microarray processing protocol in a walk-in ozone-controlled enclosure. Microarrays were scanned using an Agilent G2505C Scanner controlled by Agilent Scan Control 7.0 software. Raw data were extracted with Agilent Feature Extraction 9.5.3.1 software and were preprocessed and normalized using R and the Bioconductor limma package (Ritchie et al., 2015) and Loess and Aquantile normalization were performed for within-array and between-arrays normalization (Smyth and Speed, 2003). Median of the M values (M value = log₂ [Treated/EP]) were calculated for multiple probes that target the same gene. Genes with median M value ≥ 0.585 are considered as upregulated with respect to EP, whereas genes with median M value ≤ −0.585 are marked as downregulated genes in this study. The log2 cutoff of 0.585 (fold change = 1.5) was used as this has previously been shown to be a good cutoff for identifying methylated genes that undergo re-expression upon treatment with the demethylating drug 5-Azacytidine (Schuebel et al., 2007).

Genes class enrichment analysis

The promoter CpGI hyper-methylated genes from each genesets (CM, HSM, SSM) are loaded to DAVID (Huang da et al., 2009a; Huang da et al., 2009b) to conduct Gene Ontology (GO) analysis. The most enriched (top1-3) biological process (BP) clusters are selected based on their Enrichment Score. Within the most enriched clusters, the top 15 biological processes with p-value < 0.01 were picked. The relative ranking order of biological processes were determined using a log₁₀ scale of their p-values. The analysis results for each genes set are listed in Table S4.

Western blots and antibodies

Cells were lysed with 4% SDS and supernatant was collected after lysate centrifuged in Qiashredder (Qiagen). Protein concentrations were determined with BCA assay (Thermo Fisher Scientific: 23225). Equal amount of proteins (30 ug) were loaded into polyacrylamide SDS page gels. Candidate proteins were detected with antibodies against hTERT (Thermo Fisher Scientific: PA5-11447), SV40 T Ag (Santa Cruz Biotechnology: sc-147) and H-Ras (Santa Cruz Biotechnology: sc-520). GAPDH (Sigma: G9545) serves as a loading control.

Statistical analysis

All analysis were performed in R statistical software (http://www.r-project.org), and Fisher’s exact test (gene overlaps) and Wilcoxon rank sum test (gene expression differences) were used to determine the p-values.

Supplementary Material

7

Document S1: Figures S1-S7.

Table S1, Related to Figures 3: Promoter CpGI hyper-methylated probes list in each state. Provided as an Excel File.

Table S2, Related to Figures 3: Promoter CpGI hyper-methylated genes list in each state. Provided as an Excel File.

Table S3, Related to Figure 3: HSM, SSM and CM genes list. Provided as an Excel File.

Table S4, Related to Figures 3: HSM, SSM, CM, HRAS_40_Com_genes, Sen_442_Com_genes and BJ_Sen_IMR90_Sen_220_Com_genes GO analysis. Biological Processes (BP) for each gene set are shown. Provided as an Excel File.

Table S5, Related to Figure 5: Statistically determining the degree of overlapping between aging hyper-methylated genes and HSM, SSM or CM genes. Provided as an Excel File.

Table S6, Related to STAR Methods: BJ EP bivalent genes list. Provided as an Excel File.

Highlights.

• Globally similar but locally distinct DNA methylation in Sen versus transformation

• Sen resistance to transformation involves programmed methylation of metabolism genes

• Stochastic promoter methylation in transformation affects developmental genes

• Common Sen and transformation methylated developmental genes track aging/cancer risk

Significance.

The origins of DNA methylation alterations in cancers are incompletely understood. Based on similarities in overall methylation patterns in replicative-senescence and cancers, it is hypothesized that tumor-promoting DNA methylation in cancers derive from cells escaping senescence. We show that the tumor-associated methylation changes evolve independent of senescence, and are pro-survival events with functional implications contrasting that in senescence. Further, senescence-associated methylation may prevent tumorigenesis. During aging, which is the best-known cancer-risk factor, and during early tumorigenesis, methylation abnormalities accumulate maximally at a subset of these transformation-associated methylated genes. Our findings have implications on the origins of DNA methylation during tumorigenesis, and for developing prevention strategies and refining biomarkers to define cancer risk.