Methylation Mesa define functional regulatory elements for targeted gene activation

Y V Liu; J Suryatenggara; H Wong; M K Jayasinghe; J P Tang; H K Tan; J Kwon; Q Zhou; J Zhang; S Moein; S Ummarino; M T N Le; J G Doench; T Benoukraf; L Chai; D Hiwase; D Thomas; A Di Ruscio; D G Tenen; M A Bassal

doi:10.1038/s41467-026-76648-4

. 2026 Aug 26;17:10203. doi: 10.1038/s41467-026-76648-4

Methylation Mesa define functional regulatory elements for targeted gene activation

Y V Liu ^1,², J Suryatenggara ¹, H Wong ¹, M K Jayasinghe ³, J P Tang ¹, H K Tan ¹, J Kwon ^1,⁴, Q Zhou ¹, J Zhang ⁵, S Moein ^1,⁴, S Ummarino ⁵, M T N Le ³, J G Doench ², T Benoukraf ^1,⁶, L Chai ⁴, D Hiwase ^7,^8,⁹, D Thomas ^7,^8,⁹, A Di Ruscio ^10,^11,¹², D G Tenen ^5,^11,^✉, M A Bassal ^5,^✉

PMCID: PMC13612829 PMID: 42786170

Abstract

DNA methylation regulates transcription, yet the demethylation of canonical elements like promoter CpG islands exhibits inconsistent correlations with gene activation. We hypothesize that causal regulatory elements are defined by biophysical hypersensitivity. Profiling 24 whole-genome bisulfite sequencing samples across diverse human and murine models, we identify Methylation Mesa, narrow (~45-300 bp), structurally conserved epigenetic regulatory elements. Mesa show enrichment in 5’ untranslated regions and associate with transcriptional activation significantly better than canonical promoters. To investigate causal regulatory dynamics, we develop CRISPR-DiR, an RNA-based targeted demethylation technology offering greater spatial precision, higher potency, and reduced toxicity compared to CRISPR-TET1. While proximal promoter demethylation initiated limited early transcription, focal demethylation of the Mesa seed acts as the primary driver of exponential CDKN2A (p16) reactivation and robust in vivo tumor suppression. We demonstrate that precise demethylation of a Mesa locus triggers localized demethylation, subsequent activation histone mark deposition, and long-range three-dimensional chromatin rewiring. Thus, Methylation Mesa act as precise, dominant epigenetic regulatory hubs, and CRISPR-DiR as a potent high-resolution tool, establishing a structural framework for biomarker discovery and targeted therapies.

Subject terms: DNA methylation, CRISPR-Cas systems, Methylation analysis, Bioinformatics, Cancer genomics

DNA methylation influences gene expression, but the regulatory importance of specific methylated regions is unclear. Here, the authors identify “Methylation Mesa” as focal epigenetic regulatory elements and show that precise Mesa demethylation drives gene activation, chromatin remodeling and tumor suppression.

Introduction

DNA methylation is an epigenetic mechanism capable of transmitting information and regulating gene activity without modifying nucleotide sequence¹. Aberrant DNA methylation, particularly in gene promoters, is considered an epigenetic hallmark of many cancers, contributing to tumorigenesis by transcriptionally silencing tumor-suppressor genes^2–4. Cytosine methylation within CpG dinucleotides is typically enriched in CpG island (CGI) clusters, which overlap with ~40–70% of gene promoters^5–9 (depending on genome build and CGI selection criteria). While promoter-CGIs are the most widely accepted regulatory elements³, evidence shows that non-CGI associated genes also exhibit methylation-dependent transcriptional regulation^10,11.

Despite their widespread use as cancer biomarkers, promoter-CGIs have yielded limited clinical success¹², and many studies report inconsistent or weak associations between their methylation and gene expression^13–15. Advancements in methylome profiling have revealed dynamic destinations beyond CGIs, such as CpG shores¹⁵. However, these canonical destinations are typically defined by broad biologically disjointed “kilobase” (kb) resolutions. We postulate that these macro-regions are actually anchored by much narrower, precise methylation-sensitive “core” regulatory seeds that have not been systematically defined^16,17.

To pinpoint these causal elements, we hypothesized that they must exhibit epigenetic hypersensitivity. Grounded in the principles of DNA methylation turnover, we reasoned that global DNMT depletion via hypomethylating agents (HMAs) would trigger disproportionately rapid, localized demethylation at hypersensitive regulatory loci compared to resistant bulk chromatin. By employing whole-genome bisulfite sequencing (WGBS) across 24 healthy, primary patient, and cell line samples before and after HMA treatment, we identified a uniform, ~45–300 bp structurally conserved demethylation signature, independent of CGI status. We designated this signature the Methylation Mesa (MM).

To accurately benchmark the functional impact of Mesa demethylation, we intentionally restricted our investigations to unambiguously linked promoter-proximal loci to preclude confounding false-positive associations inherent to distal 3D enhancer looping^18–21. This revealed a stronger link between 5’UTR/Exon 1 Mesa demethylation and transcriptional activation than with upstream-TSS promoters (-1kb to TSS). To investigate causal regulatory dynamics, a targeted epigenetic tool was required. However, existing systems like CRISPR dCas9-TET1 suffer from wide editing windows that limit spatial resolution^22,23. Therefore, we engineered a high-resolution targeted demethylation technology with an editing window ten times (10x) narrower.

Utilizing the hypermethylated tumor suppressor CDKN2A (p16) as a model, CRISPR-DiR demonstrated that targeted focal demethylation of the Mesa seed initiates robust gene activation. Crucially, spatial control experiments confirmed that targeting the immediate flanking sequences was significantly less efficient, establishing focal Mesa demethylation as the primary driver for robust transcriptional reactivation and subsequent 3D chromatin looping between proximal and distal Mesa.

In summary, we define MM as highly precise, causal DNA methylation regulatory elements. By uncoupling dynamic epigenetic sensitivity from static sequence density and validating their function with hyper-focal CRISPR-DiR editing, these findings provide a new structural framework for understanding DNA methylation dynamics, biomarker discovery, and targeted epigenetic therapies.

Results

A distinct and narrow-width Mesa signature is consistent across sample types and species

We hypothesized that methylation-sensitive regulatory seed elements would possess a heightened sensitivity to demethylation stimuli than flanking loci. Although DNA hypomethylating agents (HMA), such as Azacytidine (AZA) and Decitabine (DAC), are regarded as nonspecific agents^24,25, their observed effects on the genome are not uniformly distributed²⁶, suggesting a heightened sensitivity of particular loci. To pinpoint these hyper-sensitive sites, we deeply profiled the methylation changes of paired samples before and after HMA treatment. To ensure biological relevance, whole-genomic bisulfite sequencing (WGBS) was performed/obtained for 24 paired samples (pre/post-HMA treatment) across a breadth of cell types with diverse genetic backgrounds ranging from human to mouse, including 4 (pre-)leukemia patients (2 Chronic Myelomonocytic Leukemia (CMML), 1 Refractory anemia with excess blasts-Myelodysplastic Syndromes (RAEB-MDS), 1 Acute Myeloid Leukemia (AML)), 3 healthy cord blood, 15 human cancer line samples from 7 cell types (SNU398, HCT116, K562, KG1a, MDS-L, MOLM13, MOLM14), 1 Microsatellite Stable Colorectal Cancer Patient-Derived Xenograft (PDx) and a murine Colorectal Cancer cell line (CT-26), the last two samples were obtained from GEO accession GSE192745²⁷ (see Methods for sample details). Furthermore, we utilized a DNMT1 knockout HCT116 line to gain insight into which loci require methylation maintenance to remain hypermethylated in the naïve state.

For cell lines, HMA-cultured samples were compared to untreated DMSO controls (Fig. 1a). For patients, diagnostic (Dx) samples were compared to post-HMA treatment (Fig. 1b). Differentially methylated regions (DMRs) were defined by an absolute ≥10% methylation reduction across matched pairs (Fig. 1c, d (gray)). Within these DMRs, we hypothesized that functional regulatory seeds should exhibit greater demethylation sensitivity than their bidirectional flanks. Grounded in the assumption that functional methylation boundaries would revolve around nucleosome units, we evaluated demethylation across 1.47 kb bidirectional flanks binned into 147 bp windows. We isolated “sensitive DMRs” (sDMRs) showing an absolute ≥10% lower average methylation than both flanks. Diagrammatically, this hypersensitivity presents as a “high-low-high” demethylation profile (Fig. 1c, d, (blue)). Conversely, in naïve states, these identical loci consistently exhibit a robust peak of hypermethylation bidirectionally flanked by lower methylation arms (Fig. 1c, d, (red)). Resembling the geographical rock formation, we termed these structurally distinct loci Methylation Mesa (MM).

Spatial profiling across all samples revealed consistent, near-identical Mesa and flanking structural architectures (Fig. 1e-i, Supplementary Fig. 1a). We observed a consistent methylation buffering window at ~294–441 bp (equivalent to ~2–3 nucleosomes in size) that serves as a structural boundary demarcating the Mesa core from surrounding regions. Crucially, WGBS profiling of primary healthy human umbilical cord blood identified robust Mesa signatures matching these exact structural architectures (Fig. 1h, Supplementary Fig. 1a). This supports the notion that Mesa constitute a fundamental, conserved epigenetic architecture present in normal primary tissue (in humans and mice), independent of disease-specific aberrations. Furthermore, sequencing an HCT116 DNMT1^−/− knockout line revealed that in the absence of DNMT1, the characteristic naïve Mesa methylation peak is abolished, indicating these loci actively require DNMT1 for maintenance (Fig. 1i).

To evaluate underlying sequence architecture, we analyzed localized CpG density. While global aggregates showed slight CpG elevation at the Mesa core, stratifying loci by genomic context revealed that this architecture is highly context-dependent. Common Language Effect Size (CLES) analysis demonstrated that while gene-body Mesa show moderate CpG enrichment, 5’UTR Mesa possess CpG densities highly comparable with their surrounding flanks (Supplementary Fig. 1b). Importantly, the distinct hypersensitive boundaries of the Mesa signature are robustly maintained regardless of local CpG density. This effectively uncouples dynamic epigenetic function from static sequence composition, confirming the profile is a conserved epigenetic feature rather than a strict sequence-driven artifact. Finally, size distribution analysis revealed these elements are consistently narrow, predominantly spanning 45–294 bp (Fig. 1j), aligning with our hypothesis that causal regulatory seeds are structurally exiguous.

Mesa enrichment in 5′UTR better associate with gene activation than upstream promoters

We next investigated the genomic distribution of Mesa. Using RefSeq defined genomic features (assembly GCA_000001405.28, release 107), we observed consistent genome-wide distributions across all major functional categories and samples (Fig. 2a). Normalizing Mesa percentages against baseline genomic category percentages revealed that Mesa are most enriched in gene 5′UTRs (155% enrichment, p = 6*10⁻⁴, Bonferroni adjusted), followed by 3′UTRs (151% enrichment, p < 1*10⁻¹⁵, Bonferroni adjusted), introns (145% enrichment, p < 1*10⁻¹⁵, Bonferroni adjusted), and upstream promoters (−1 kb – TSS) (113% enrichment, p = 4*10⁻³, Bonferroni adjusted) (Fig. 2b).

Fig. 2 — a Genome-wide distribution of Mesa loci annotated against RefSeq functional genomic categories. b Relative enrichment of Mesa localization. Percentages of Mesa within functional categories were normalized against natural genomic baseline distributions (shown right). Mesa demonstrate significant maximal enrichment (two-tailed Fisher’s Exact) within 5’UTRs, 3’UTRs, and intronic regions (Bonferroni adjusted p-values annotated). Panel shows mean ± SD. c Profile plot showing the presence of structural signal surrounding coding genes (−5kb to +5 kb relative to TSS and TES). Signal profile demonstrates peak CGI density immediately upstream of the TSS. Mesa elements are significantly depleted at the TSS (Statistical divergence quantified in Supplementary Data 2). **d–f** Paired density distributions and pirate plots map methylation drop (Δ-Methylation) at Mesa compared to canonical 1 kb promoters, core 300 bp promoters, and CGI. Loci filtered for concordant hypomethylation and transcriptional activation (log₂ Fold Change > 1) following HMA treatment. Data shown are representative of MOLM14 treated with 5 μM decitabine. Differences in Δ-methylation between Mesa and canonical genomic features evaluated using a one-sided, paired Wilcoxon signed-rank test. Boxplot center lines indicate median, hinges represent 25th and 75th percentiles. Whiskers extend to the minimum and maximum values 1.5 times the interquartile range from the hinges. Significance denoted as * p < 0.05, ** p < 0.01, and *** p < 0.001. P-values presented in figure. d (left panel) Schematic defining the genomic boundaries of the Upstream promoter region (−1 kb to TSS) evaluated relative to the TSS. (right panel) Paired density distributions for upstream regions of interest. e (left panel) Schematic defining the genomic boundaries of the Downstream region (TSS to +1 kb, encompassing 5’UTR and exon 1 junctions) evaluated relative to the TSS. (right panel) Paired density distributions for downstream regions of interest. f Pirate plots showing the methylation drop (Δ-Methylation) at Mesa compared to canonical 1 kb promoters, core 300 bp promoters, and CGI. Data distribution visualized using a violin plot overlaid with individual data points (# up-regulated/down-regulated genes, respectively noted). Source data are provided in the source data file.

Spatial profiling of the structural signal surrounding coding genes revealed that Mesa elements are significantly depleted at the proximal promoter, or “TSS Core,” peaking instead within downstream gene bodies. This distribution is of note as it rarely coincides with established CGI definitions (Fig. 2c, Supplementary Fig. 2a). This consistent spatial partitioning and TSS-exclusion was further confirmed across our entire cohort via heatmap analysis and regional signal burden quantification (Supplementary Fig. 2b, c). Chi-square testing confirmed that this spatial topology is statistically divergent from canonical CGI distributions (p < 1*10⁻¹⁶, Supplementary Data 2).

Having identified these hyper-sensitive elements, we next investigated their potential to directly regulate expression. We quantified the statistical significance of methylation-dependent gene upregulation using matched WGBS and bulk RNA-Seq data across diverse biological models, including normal primary cord blood, malignant cell lines, patient-derived samples, the Microsatellite Stable Colorectal Cancer Patient-Derived Xenograft (PDx) and the murine Colorectal Cancer cell line (CT-26)²⁷. To ensure analytical rigor, we restricted our statistical analysis specifically to loci exhibiting concordant hypomethylation and transcriptional activation log₂ Fold Change >1, analyzing elements located both upstream (−1 kb to TSS; Fig. 2d) and downstream (TSS to +1 kb, encompassing 5’UTR and exon 1 junctions; Fig. 2e).

Evaluating the methylation-expression variance using non-parametric Wilcoxon rank-sum testing revealed that demethylation specifically at Mesa loci consistently exhibits a statistically significant, and stronger association with target gene activation than demethylation at classical 1 kb promoters, core 300 bp proximal promoters, or CpG Islands. Density distributions and paired pirate plots illustrate a statistically significant shift in the median methylation drop (Δ-Methylation) at MM compared to canonical features (Fig. 2d-f). As representatively shown in the MOLM14 cell line (5 µM DAC), the most statistically powerful association was observed when associating expression with Mesa demethylation within the 1 kb downstream region (Fig. 2e, f). To ensure this observed association was not a statistical artifact of preselecting for upregulated genes, we performed a 10,000-iteration Monte Carlo permutation test. This confirmed that the superior association between Mesa demethylation and gene reactivation significantly deviates from a random null distribution (empirical p < 0.0001), establishing this dynamic as a robust biological feature rather than an artifact of selection bias (Supplementary Fig. 2d).

To further ensure these findings were not artifacts of a single treatment intensity, we evaluated transcriptional variance across a HMA titration series (e.g., MOLM14 treated with 1–5 µM DAC; Supplementary Fig. 2k-m). Remarkably, the superior statistical association of Mesa demethylation remained firmly consistent across the entire dosage gradient. This demonstrates an epigenetic hypersensitivity wherein MM reaches regulatory saturation even at low pharmacological doses, mirroring the established efficacy of low-dose HMA therapies. Extending this analysis across our entire expanded cohort, which includes a primary healthy cord blood (CB83), confirmed that this dominant methylation-expression association is rigidly conserved across both normal hematopoiesis and malignant states (Supplementary Fig. 2e-o).

Together, these findings reveal previously undescribed, structurally distinct methylation-sensitive elements that drive target gene reactivation with statistically superior efficacy compared to canonical upstream promoters.

Mesa are extensively marked by primed and active histone marks and show positional conservation

To achieve an unbiased evaluation of chromatin state overlap, we compiled a comprehensive epigenetic footprint mapping 93 curated human ChIP-Seq datasets, encompassing activation-associated (n = 68), repression-associated (n = 25), and structural CTCF marks. Requiring a strict ≥1 bp overlap, we assessed intersection frequencies across distinct genomic partitions. This analysis revealed that promoter and 5’UTR Mesa exhibit robust overlap (up to ~90%) with H3K4me1 and H3K27ac, classical marks of primed and active regulatory elements. Intragenic and intergenic Mesa demonstrated similarly strong overlap (up to ~65%) with these active signatures.

However, clustered heatmap analysis revealed that Mesa do not strictly segregate into uniformly active categories. Instead, they represent dynamic, multivalued regulatory loci capable of harboring poised (bivalent) or strictly repression-associated signatures (such as H3K27me2, H3K27me3, H3K9me3) in a context-dependent manner (Fig. 3a-d). To quantitatively dissect these complex intersections, we constructed UpSet plots mapping combinatorial chromatin states genome-wide (Supplementary Fig. 3a). This multivalent capacity, in conjunction with their robust methylation-expression correlation, suggests that Mesa demethylation may mechanistically prime chromatin by enabling the subsequent deposition of context-specific histone marks. Intriguingly, despite their defined architectural boundaries, Mesa was found not to overlap strongly with CTCF.

Beyond structural architecture, we hypothesized that the genomic coordinates of Mesa (positional conservation) are non-randomly conserved within related biological cohorts. To quantitatively evaluate this spatial positioning, we performed genomic permutation testing against random background models. Pairwise overlap analysis demonstrated highly significant, non-random positional conservation across four patient-derived clinical samples (empirical p < 0.001, average Z-score = 197.5; Fig. 3e, Supplementary Table 3). This spatial concordance was maintained regardless of HMA dosage (MOLM14 treated with 1 μM to 5 μM DAC; average Z-score = 248.1; Fig. 3f) or temporal treatment kinetics (K562 and KG1a lines treated for 48 h and 72 h; average Z-score = 195.3; Fig. 3g).

Collectively, these statistical validations, complemented by macroscopic genome-wide density mapping (Supplementary Fig. 3b), confirm the existence of a core, positionally conserved subset of Mesa. To reconcile this highly significant spatial conservation with the dynamic nature of the epigenome, we note that Mesa coordinates naturally exhibit a duality of function. While the majority of Mesa loci exhibit condition- or patient-specific topological variance to drive context-specific transcriptional programs, a foundational subset remain rigidly anchored across multiple evaluated biological models. Thus, Mesa function as highly versatile regulatory hubs, offering both a stably conserved epigenetic baseline and dynamic, state-specific plasticity.

Development of CRISPR-DiR for precise demethylation to investigate MM causal regulatory dynamics

To investigate Mesa regulatory dynamics, we required a targeted demethylation tool with high resolution, as canonical promoters and TSS-Mesa can be separated by only a few hundred base pairs.

Structural studies demonstrate that the Cas9/dCas9-sgRNA-DNA complex²⁸, including its derivatives (such as CRISPR-SAM²⁹ and CRISPR-Rainbow³⁰), are capable of tolerating RNA aptamer substitutions without compromising stability. Given that DNMT1 actively maintains Mesa methylation (Fig. 1h), we sought to neutralize its local enzymatic activity using DNMT1-interacting RNAs (DiRs)³¹. We substituted the sgRNA tetra- and stem-loops with 23-nucleotide-long R2 and R5 DiR loops from the lncRNA ecCEBPA³¹, constructing eight modified sgDiR scaffolds (Fig. 4a, b). This RNA-based design diverges from conventional CRISPR-protein fusions, bypassing bulky enzymatic domains for a streamlined, non-enzymatic approach.

Fig. 4 — a CRISPR-DiR designs. DNMT1-interacting RNA (DiR) loops R2 and R5³¹ were fused to sgRNA tetra-loop (TL) and/or stem-loop 2 (SL2) regions²⁹. Right: Diagrams representing sgRNA control and sgDiR designs 1-8. b sgRNA control and sgDiRs targeting p16 proximal promoter using G1. Guide sequences, scaffold sgRNA and sgDiRs 1-8 in Supplementary Data 4, 5. c Methylation Sensitive PCR (MSP) demonstrating p16 demethylation in SNU398 72 h post-transfection. Mock: transfection without plasmids; sgRNA: co-transfection dCas9 + sgRNA (no DiR); sgDiR1-8: co-transfection dCas9 + sgDiRs (with DiR); NTC: no template PCR control. d Methylation heatmap showing Bisulfite Sequencing PCR (BSP) results for parental cells, CRISPR-DiR and CRISPR-TET1 stably transduced targeting p16 with G2 for 7 days. Eight colonies of each sample were picked for sanger-sequencing. Heatmap represents average methylation for CpG residues from n = 8 colonies. e Schematic representation of LF, MM, RF at p16 locus. Heatmap represents WGBS results in the LF-MM-RF region of SNU398 naïve and DAC-treated. f Real Time-Quantitative PCR (RT-qPCR) result showing *p16* mRNA expression in SNU398 after CRISPR-DiR targeting. CRISPR-DiR (both dCas9 and sgDiRs) stably transduced by lentivirus. NT1: non-targeting control, LF: guides G3 and G4, MM: guides G2 and G5, RF: guides G6 and G7, LF + RF: guides G3, G4, G6 and G7, LF + MM + RF: guides G3, G4, G2, G5, G6 and G7. sgDiR guides in Supplementary Data 5, guide locations Supplementary Data 7. Data points n = 2 or 3 biological replicates. RT-qPCR results showing functional evaluation of p16 reactivation driven by (g) CRISPR-DiR versus (h) CRISPR-TET1. SNU398 cells stably transduced with either system utilizing matched guide sets to target regulatory regions (LF, MM, RF) or combinatorial (LF + RF) over 30 days. Guide per region: LF: G3, MM: G2, RF: G7, LF + RF: G3 + G7. Data points n = 2 or 3 biological replicates. i Top: LF, MM, RF region for *RAP1GAP2*. MM calculated using WGBS data for SNU398 naïve (N) and DAC (D) cells. Bottom: RT-qPCR results showing mRNA expression over 2 weeks after CRISPR-DiR targeting LF + RF. n = 2/3 biological replicates. Statistics were unpaired one-tailed t-tests. Panels show mean ± SD, * p < 0.05; ** p < 0.01; *** p < 0.001. Source data are provided in the source data file.

To optimize the design, we targeted the hypermethylated CDKN2A (p16) tumor suppressor locus in the SNU398 line^2,32. Utilizing guide G1 (guide first base is −123 bp upstream of the p16 TSS), dCas9 was co-transfected with either unmodified sgRNA or modified sgDiR1-8. Evaluating p16 demethylation at 72 h identified sgDiR6 as the optimal functional structure (Fig. 4c).

We next performed a head-to-head comparison against CRISPR-TET1 using guide G2 (p16 exon 1). Bisulfite sequencing revealed that while CRISPR-TET1 exhibited a broad demethylation footprint of ≥350 bp, CRISPR-DiR restricted demethylation to a highly focal ~30 bp window, achieving >10-fold greater spatial precision (Fig. 4d). Furthermore, competitive co-culture assays demonstrated that CRISPR-TET1 induced substantially greater cellular toxicity and viability defects than, the RNA-based CRISPR-DiR system (Supplementary Fig. 4a).

Although CRISPR-DiR successfully demethylated the p16 proximal promoter, this canonical target yielded only moderate p16 mRNA increases (Supplementary Fig. 4b, c), aligning with WGBS data showing the p16 promoter-CGI remained unresponsive to DAC treatment (Fig. 4e). Instead, our analysis identified a distinct Mesa downstream at the 5’UTR/exon 1 junction (+326 to +358 bp). As guide G1 (guide first base −123bp upstream of TSS) serendipitously fell within the Mesa’s left flank (LF) buffering window (~3 nucleosomes wide), we designed matched guides targeting the LF, the Mesa core (MM), and the right flank (RF) to evaluate regional demethylation that correlates with gene activation (Supplementary Fig. 4d).

Targeting these regions independently revealed that focal demethylation of the Mesa core drove exponentially higher p16 re-expression than targeting the LF or RF (Fig. 4f). While promoter-CGI (LF) targeting initiated early transcription that quickly plateaued, Mesa demethylation drove a sustained, progressive increase to a substantially higher expression ceiling. This strongly suggests that focal seed demethylation serves as the primary initiator for robust, sustained reactivation, acting alongside early promoter-associated effects.

Testing combinatorial targeting, we found that simultaneously targeting both Mesa boundaries (LF + RF) induced maximal p16 expression, surpassing singular region targeting (Fig. 4f). This establishes that combinatorial targeting of the flanking boundaries is the most effective strategy for destabilizing the core regulatory element. This optimal targeting strategy was further validated in the U2OS osteosarcoma line, with specificity confirmed by the absence of off-target activation at adjacent (p14) or distant (CEBPA) loci (Supplementary Fig. 4e-g). Therefore, moving forward, this combinatorial strategy was utilized for subsequent experiments.

In a functional head-to-head comparison, CRISPR-DiR consistently drove stronger and sustained activation than CRISPR-TET1 across both LF and MM targets (Fig. 4g-h). Notably, the imprecise, diffuse footprint of CRISPR-TET1 failed to capture the region-specific kinetic advantages seen with CRISPR-DiR. Furthermore, the p16 expression induced by CRISPR-TET1 rapidly reduced after day 13, likely reflects cumulative cellular toxicity or potential compensatory silencing of the CRISPR-TET1 machinery (Supplementary Fig. 4a). Finally, we validated the universality of this targeting strategy on three additional silenced tumor suppressors in SNU398 cells, namely RHOD and CDX1 (whose Mesa do not overlap upstream promoters), and RAP1GAP2 (which lacks an annotated CGI entirely). In all cases, targeted Mesa demethylation drove robust gene reactivation (Fig. 4i, Supplementary Fig. 4h-i), confirming the superiority of Mesa targeting regardless of canonical sequence annotation.

Tracing the epigenetic kinetics triggered by p16 MM demethylation

When demethylating a Mesa locus using CRISPR-DiR, we observed that p16 mRNA transcription continued to progressively increase over a period of multiple weeks (Fig. 4f). To trace these epigenetic kinetics, we tracked single CpG and locus average methylation, histone mark deposition, mRNA, and protein levels for 53 days following CRISPR-DiR targeting of the p16 LF + RF boundaries in SNU398 cells (Fig. 5a-e).

Kinetic tracking revealed initial demethylation at the LF and RF target sites by Day 8, which progressively spread inward to clear the interjacent Mesa core from Day 13 onwards (Fig. 5a). Crucially, while this progressive demethylation traveled inward to clear the Mesa seed, it remained strictly confined within the targeted boundaries. We observed no outward spreading into the untargeted upstream or downstream flanks over the entire 53-day tracking period (Supplementary Fig. 5a, b). This definitive spatial confinement confirms Mesa operate as discrete, structurally isolated regulatory units, underscoring why precise boundary targeting represents an optimal strategy for driving massive gene re-activation without the indiscriminate epigenetic erasure caused by bulkier tools.

As we followed the kinetic profiling further, we observed a gradual decrease in average methylation across the LF-MM-RF locus (Fig. 5b). Following Day 8 demethylation, active histone marks (H3K4me3, H3K27ac) increased while the silencing mark H3K9me3 decreased (Fig. 5c). Subsequently, p16 mRNA expression surged after Day 13 (Figs. 4f, 5d), leading to visible p16 protein detection after Day 20 (Fig. 5e). This timeline firmly establishes that physical demethylation strictly precedes transcriptional activation, acting as the requisite trigger for histone modifications to accommodate subsequent gene expression.

To evaluate whether this demethylation had a lasting epigenetic memory, we generated a doxycycline-inducible CRISPR-DiR SNU398 cell line. Following just 3 days of induction, p16 demethylation and activation were maintained for over a month (Supplementary Fig. 5c, d). This suggests that Mesa demethylation induces self-sustaining, long-term chromatin rewiring capable of maintaining active gene expression.

Finally, to definitively evaluate the functional superiority of Mesa demethylation over canonical promoter targeting in vivo, we performed expanded, head-to-head mouse xenograft studies. Initial validations showed LF + RF boundary targeting produced significantly smaller tumors over 12 days (Supplementary Fig. 5e-g). In the expanded cohort, mice were engrafted with SNU398 cells transduced with a non-targeting control (NT1), the canonical promoter-CGI guide (G3), or the Mesa boundary guides (LF + RF). Crucially, the in vivo results perfectly mirrored our in vitro observations. Tumors featuring targeted promoter-CGI (G3) demethylation exhibited aggressive growth kinetics and final volumes largely indistinguishable from non-targeting controls (Fig. 5e, f). In contrast, demethylation of the Mesa boundaries (LF + RF) resulted in a dramatic, statistically significant suppression of in vivo tumor growth. This definitive functional validation demonstrates the superior relevance of Mesa over canonical promoter elements in driving true phenotypic changes, highlighting the translatable therapeutic potential of CRISPR-DiR.

Mesa demethylation and gene re-expression initiate long-range interaction with distal Mesa loci

Thus far, we have shown that targeted demethylation of the p16 Mesa was sufficient to trigger potent gene expression. Furthermore, our inducible CRISPR-DiR findings (Supplementary Fig. 5c, d) suggested that this initial physical uncoupling triggers subsequent chromatin rewiring, resulting in lasting epigenetic memory long after the cessation of DiR induction.

Previous studies have reported a p16 enhancer located ~150 kb upstream of the p16 TSS^33–35. Nonetheless, long-range 3D interactions with the p16 locus, and specifically the impact of locus-specific demethylation on enhancer recruitment, remain unexplored. To assess the impact of LF-MM-RF demethylation on the 3D topology surrounding the p16 locus, we performed Circularized Chromosome Conformation Capture (4C) on CRISPR-DiR non-targeting (NT1) and LF + RF targeted samples. We designed two distinct viewpoints (“baits”) anchored near the targeted LF-MM-RF locus (Fig. 6a). Comparing the ~800 bp demethylated region against the control, we detected robust interaction changes between the p16 locus and distal elements, many of which were consistently captured across both viewpoints (Fig. 6b).

Fig. 6 — a Schematic representation of the two viewpoints (generated by restriction enzymes Csp6I and DpnII) surrounding the 800 bp demethylated p16 LF-MM-RF region used for Circularized chromosome conformation capture (4C)-Seq. b 4C-Seq demonstrating the distal interactions initiated by CRISPR-DiR targeted p16 demethylation in SNU398 cells. The interaction changes of p16 LF-MM-RF demethylated samples were normalized to the same time point (Day 13) non-targeting control (NT1) sample. Top: shows distal interactions captured for Viewpoint Csp6I, while the bottom shows Viewpoint DpnII. The MM annotations are indicated with purple bars in the middle. The strong interaction changes demonstrated by the interaction arcs, with color from blue to red, represent the interaction fold change. The potential distal enhancer elements with the strongest interaction were highlighted for both viewpoints on the top, labeled from p16 upstream to downstream (negative orientation, right to left) as E1, E2, E3, E4, E5, and E6. Source data are provided in the source data file.

Consistent with the previously reported p16 enhancer region^33–35, we observed that the targeted p16 Mesa actively interacted with the established E2 distal enhancer (Fig. 6b). Furthermore, we detected several additional interactions at sites >200 kb upstream of p16 (E1), within the adjacent ANRIL-p15-p14 gene locus (E3), and at sites >100 kb downstream of p16 (E4, E5). The strongest of these interactions were independently validated by their overlap across both viewpoints. Crucially, when mapping the 4C interaction landscape against our genome-wide annotations, we observed that the distal interacting loci initiated by p16 Mesa demethylation aligned remarkably well with intragenic and intergenic Mesa (Fig. 6b). This indicates that the Mesa signature demarcates both proximal anchors and distal functional enhancers.

Together, these 4C data demonstrate that CRISPR-DiR-mediated Mesa demethylation successfully captures known enhancer regions while uncovering distinct 3D interplay, including a previously unreported topological connection between p16 and the ANRIL-p15-p14 locus. Intriguingly, both the proximal anchor and the recruited distal loci are highly concordant with Mesa structural signatures. These findings solidify the model that Mesa act as interconnected regulatory nodes. Focal demethylation of the proximal Mesa seed serves as the required initiator, triggering subsequent histone modifications and long-range chromatin rewiring to interact with distal elements that are also structurally marked as Mesa.

Discussion

For decades, the pursuit of causal DNA methylation regulatory elements has been heavily anchored to CpG Islands (CGIs) and their immediate boundaries, including CpG shores, shelves, canyons, and valleys^15,36–38. However, definitive data establishing which specific elements are functionally requisite for triggering gene re-expression remains scarce. We propose that this limitation stems from relying on static, kilobase-scale sequence definitions rather than dynamic biophysical behavior. In this study, we report the discovery of highly precise, structurally conserved methylation-sensitive regulatory elements, Methylation Mesa (MM). By comprehensively profiling over 4 terabytes of WGBS data across diverse human, murine, healthy, and malignant states, we established that Mesa are narrow-width seed loci (predominantly 45–300 bp) (Fig. 1j) characterized by heightened hypersensitivity to demethylation stimuli. Importantly, we demonstrate that while these structural boundaries are globally conserved, their specific genomic coordinates maintain a functional duality; a core subset acts as a stably conserved epigenetic baseline, while the remainder exhibit condition-specific variance to facilitate dynamic, lineage-specific transcriptional rewiring.

This transition towards high-resolution structural profiling necessitated a reevaluation of where causal regulatory elements physically reside genomically. Historically, biomarker discovery has focused on the canonical proximal promoter, operating under the assumption that linear proximity to the transcription start site (TSS) dictates regulatory control. However, our spatial topology analysis revealed that Mesa are significantly depleted at the TSS core (Fig. 2c, Supplementary Fig. 2a–c), instead enriching strongly within gene 5’UTRs (~155% enrichment over baseline) and downstream gene bodies (Fig. 2b). We validated this topographical shift using our engineered, high-resolution CRISPR-DiR system. Utilizing p16 as a model, we demonstrated that while targeted demethylation of the canonical upstream promoter-CGI initiated early expression, it failed to trigger the substantial, sustained exponential gene expression achieved by targeting the Mesa, nor did it alter in vivo tumor kinetics. In contrast, focal demethylation of the 5’UTR/exon 1 Mesa triggered robust, exponential transcriptional reactivation and massive in vivo tumor suppression (Fig. 5). This observation is consistent with our prior published findings^39,40 wherein targeted Mesa uncoupling was sufficient to induce strong gene re-expression.

While validating the p16 locus, we observed that simultaneous targeting of the Mesa Left Flank (LF) and Right Flank (RF) represented the simplest and most potent strategy for triggering robust gene activation (Fig. 4f). Although CRISPR-DiR initially introduced demethylation only at the targeted boundary regions, kinetic tracing revealed a stepwise process whereby demethylation spread progressively inwards to clear the Mesa core, but strictly failed to spread outward beyond the targeted locus over a 53-day period (Fig. 5, Supplementary Fig. 5a, b). Intriguingly, our observations align with the established “neighbor-guided” correction mechanism for DNMT1 methylation maintenance, where methylated flanking regions kinetically guide the re-methylation of imperfectly maintained intervening regions⁴¹. Building upon this principle, we propose a “neighbor-stabilized” structural model, wherein uncoupling of the flanking boundaries deprives DNMT1 of its kinetic guide, triggering and inward collapse of the Mesa core. We hypothesize that the structural integrity of the hypermethylated Mesa core relies on the continuous processivity of DNMT1 recruited to the adjacent flanking nucleosomes. By sterically neutralizing DNMT1 maintenance at these specific boundaries, the local epigenetic architecture destabilizes, triggering targeted erasure of the interjacent seed without risking widespread, off-target epigenetic stripping.

This reliance on DNMT1 maintenance also expands the therapeutic scope of Mesa targeting. While CRISPR-DiR operates via the targeted uncoupling of DNMT1, its utility may not be strictly limited to actively dividing cells undergoing S-phase replication. An emerging body of evidence demonstrates that the epigenetic landscape of post-mitotic neurons is highly dynamic, frequently undergoing base-excision repair and active demethylation driven by high oxidative metabolism^42,43. Consequently, continuous DNMT1 turnover is required to re-methylate these loci, an active cycle critical for regulating neurotransmission and synaptic plasticity⁴⁴. Since the adult epigenome relies on this constant maintenance cycle, we hypothesize that utilizing CRISPR-DiR to sterically block Mesa boundaries will effectively trap the locus in a demethylated state even in non-dividing contexts, providing a highly specific avenue for stable gene reactivation even in post-mitotic tissues without requiring bulky, toxic enzymatic domains.

A related aspect of this study was the long-term kinetics tracing following p16 TSS-proximal Mesa demethylation (Fig. 5). We observed that physical demethylation preceded the acquisition of active histone marks, the removal of repressive marks, and the gradual accumulation of mRNA and protein over two months. Furthermore, our inducible CRISPR-DiR assay illustrated that a brief three-day demethylation pulse was sufficient to maintain an active transcription environment for over a month (Supplementary Fig. 5c).

When annotating Mesa with our compiled histone footprint map, we observed high concordance between Mesa and H3K4me1 across all genomic compartments (Fig. 3a-d). H3K4me1 is recognized as a mark that primes active and bivalent promoters, where regulatory activity is waiting to occur once triggered by activation signals^45–48. These dynamics suggest a model where initial demethylation of the Mesa acts as the requisite primer that subsequently recruits chromatin remodeling complexes to deposit active marks and rewire the local landscape.

To bypass the false-discovery rates associated with linking distal elements based solely on linear genomic proximity^18–21,49, we focused our functional validations on TSS-proximal Mesa. However, subsequent Circularized Chromosome Conformation Capture (4C) profiling revealed a profound 3D topological consequence wherein targeted demethylation of the ~800 bp proximal p16 Mesa actively initiated long-range looping interactions with distal, unannotated enhancer elements located as far as 500 kb away. Intriguingly, these distal interacting loci were also marked as intergenic Mesa. This suggests that Mesa functions as interconnected regulatory nodes, where targeted focal demethylation of the seed acts as the initiator for the 3D architectural rewiring of the entire genomic locus. Further studies are warranted to better understand the recognition mechanisms governing these proximal-distal Mesa interactions.

Finally, in spite of decades of study, it remains unclear exactly how mutations in DNMT3A lead to clonal hematopoiesis, Myelodysplastic Syndrome(s) and Acute Myeloid Leukemia^50–53, or how global hypomethylating agents induce clinical responses in these malignancies. In light of this study, it is plausible that these non-specific agents induce therapeutic responses precisely because functional Mesa exist in a state of continuous, dynamic methylation turnover. This renders them biochemically hypersensitive to HMA-induced DNMT depletion compared to bulk heterochromatin, allowing low-dose HMA to reach regulatory saturation at Mesa sites while sparing the broader genome.

In conclusion, our study presents detailed insights into the structural locations of causal methylation-sensitive regulatory loci. By uncoupling dynamic epigenetic sensitivity from static sequence density and demonstrating the potent in vivo efficacy of spatial boundary targeting, our findings highlight the translatable potential of Mesa targeting. The precise definition of the Mesa regulatory element, paired with the streamlined delivery of CRISPR-DiR, enables the development of highly specific functional biomarkers and the next generation of targeted epigenetic therapies.

Methods

Ethics compliance

This research complies with all relevant ethical regulations. Human sample collection was approved by the Institutional Central Adelaide Local Health Network Ethics Committee (Approval No. R20110304) and the University of Eastern Piedmont (Protocol 939/CE). Umbilical cord blood was collected with written informed consent under Lyell McEwin Hospital human ethics committee approval (HREC/20/WCHN/65). Animal experiments were conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee at Beth Israel Deaconess Medical Center (Protocol ID 061-2022-25) and the National University of Singapore.

Cell lines and patient samples

The cell lines SNU398, K562, KG1a, MOLM13, MOLM14, and MDS-L were cultured in Roswell Park Memorial Institute 1640 medium (RPMI) (Life Technologies, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS) (Invitrogen). Human HEK293T and human osteosarcoma cell line U2OS were maintained in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS). HCT116 was cultured in McCoy’s 5a Medium (Catalog number: 16600082). All cell lines were maintained at 37 °C in a humidified atmosphere with 5% CO₂ and cultured in the absence of antibiotics if not otherwise specified. The ADL samples were mononuclear cells from patients with acute myeloid leukemia and chronic myelomonocytic leukemia. They were obtained from the Myelodysplastic Registry and South Australian Cancer Research Biobank (Institutional Central Adelaide Local Health Network Ethics approval No. R20110304) housed at the South Australian Health and Medical Research Institute. Matched DNA and RNA samples were obtained from the same patient at diagnosis and after four cycles of Azacytidine treatment. Umbilical cord blood (CB) was collected with written consent from scheduled cesarean section deliveries at the Women’s Health Unit, Lyell McEwin Hospital (Adelaide, South Australia), with human ethics committee approval (HREC/20/WCHN/65). The ITL myelodysplastic syndrome sample was obtained from the University of Eastern Piedmont, Novara, Italy (Protocol 939/CE, Study No. CE 154/15). Matched DNA and RNA samples were obtained from the same patient at diagnosis and after four cycles of Azacytidine treatment. Samples were handled in accordance with the Declaration of Helsinki. The CT-26 and PDx data were obtained from GEO Series GSE192745. Additional HCT116 data were obtained from GEO Series GSE97889.

RNA isolation

Total RNA was either extracted using AllPrep DNA/RNA Mini Kit (Qiagen, Valencia, CA) and treated with RNase-free DNase Set (Qiagen) following the manufacturer’s instructions, or isolated with TRIzol (Invitrogen). If the RNA isolation was carried out with the TRIzol method, RNA samples were treated with recombinant RNase-free DNase I (Roche) (10 U of DNase I per 3 mg of total RNA; 37 °C for 1 h; in the presence of RNase inhibitor). After DNase I treatment, RNA samples were extracted with acidic phenol (pH 4.3) to eliminate any remaining traces of DNA.

Genomic DNA extraction

Genomic DNA was extracted by either the AllPrep DNA/RNA Mini Kit (Qiagen, Valencia, CA) for BSP, MSP, and COBRA assays or by phenol–chloroform method if high-quality DNA samples were required for whole-genomic bisulfite sequencing (WGBS). The Phenol–chloroform DNA extraction was performed as described⁵⁴. In the case of the SNU398 samples, 10 cm plates of wild-type SNU398 cells and Decitabine-treated SNU398 cells were washed twice with cold PBS. 2 mL of gDNA lysis buffer (50 mM Tris-HCl pH 8, 100 mM NaCl, 25 mM EDTA, and 1% SDS) was added directly to the cells. The lysates were incubated at 65 °C overnight with 2 mg of proteinase K (Ambion). The lysate was diluted two times with TE buffer before adding 1 mg of RNase A (PureLink) and followed by a 1-h incubation at 37 °C. The NaCl concentration was subsequently adjusted to 200 mM followed by phenol-chloroform extraction at pH 8 and ethanol precipitation. The gDNA pellet was dissolved in 1 mL TE pH 8 buffer and incubated with RNase A with a concentration of 100 μg/mL (Qiagen) for 1 h at 37 °C. The pure gDNA was recovered by phenol-chloroform pH 8 extraction and ethanol precipitation and dissolved in TE pH 8 buffer.

5-aza-2’-deoxycytidine (Decitabine; DAC) treatment

SNU398 cells were treated with 2.5 μM of 5-aza-2’-deoxycytidine (Sigma-Aldrich) according to the manufacturer’s instructions. MOLM13 cells were treated with 1 μM of 5-aza-2’-deoxycytidine. MOLM14 cells were treated with either 1 μM, 2.5 μM, 3.5 μM, or 5 μM 5-aza-2’-deoxycytidine. MDS-L cells were treated with either 0.5 μM, 1 μM, or 2 μM 5-aza-2’-deoxycytidine. Media and drugs were refreshed every 24 h. RNA and genomic DNA were isolated after 3 Days (72 h) of treatment. Details on how the HCT116 sample was treated can be found in the respective publication⁵⁵.

5-azacytidine (Azacytidine; AZA) treatment

K562 and KG1a cells were treated with 2 μM of 5-azacytidine (Sigma-Aldrich) according to the manufacturer’s instructions. Cells were treated for either 48 or 72 h with media and drug refreshed every 24 h. RNA and genomic DNA were isolated thereafter. Healthy cord blood samples were treated with 1 μM of 5-azacytidine for 72 h.

Deoxycytidine (Dox) treatment

In the dCas9 Tet-On SNU398 cells (inducible CRISPR-DiR system shown in Supplementary Fig. 5c, d), the same targeting strategy shown in Fig. 5a (LF + RF) was used, and dCas9 expression was induced following treatment with Deoxycytidine (Dox). Dox was freshly added to the culture medium (1 µM) every day for Dox⁺ sample, while Dox^- samples were cultured in normal medium without Dox. For Dox induced 3 days/8 days samples, 1 µM Dox was added to fresh medium for 3 days and 8 days accordingly, then the cells were kept in medium without Dox until Day 32; for Dox induced 32 days samples, 1 µM Dox was added to fresh medium every other day for 32 days. All treated cells were cultured and assayed at Day 3, Day 8, and Day 32.

Transient transfections

SNU398 cells were seeded at a density of 3.5 × 10⁵ cells/well in 6-well plates 24 h before transfection employing jetPRIME transfection reagent (Polyplus Transfection) as described by the manufacturer. 2 μg mix of sgRNA/sgDiR and dCas9 plasmid(s) (sgRNA/sgDiR: dCas9 molar ratio 1:1) were transfected into each well of cells. The culture medium was changed 12 h after transfection. Alternatively, the Neon™ Transfection System (Thermo Fisher) was used for cell electroporation according to the manufacturer’s instructions. The same plasmid amount and ratios were used in the Neon as in the jetPRIME transfection. Optimal transfection efficiency using the Neon Transfection System was achieved using 0.7–1.5 million cells per 100 µL of suspension, with a single pulse at 1550 V for 35 ms. The culture medium was changed 24 h after transfection. The whole scaffold sequences of sgRNA and sgDiR structures are listed in Supplementary Data 4. The guide RNA sequences are listed in Supplementary Data 5. The plasmids used in this study are listed in Supplementary Data 6.

All oligonucleotides and gBlocks were ordered from IDT.

Plasmids can be requested by contacting the corresponding authors.

Lentivirus production

All plasmids were transfected into 10 million 293T using TransIT-LT1 reagent (Mirus), lentivector:psPAX2:pMD2.G 9 µg:9 µg:1 µg. The medium was changed 18 h post-transfection, and the virus supernatants were harvested at 48 h and 72 h after transfection. The collected virus was filtered through 0.45 µm PVDF microfilters and stored at −80 °C.

Stable cell line generation

Both SNU398 and U2OS cells were seeded in T75 flasks or 10 cm plates 24 h prior to transduction, and were transduced with lentivirus medium (thawed from −80 °C) together with 4 μg/mL polybrene (Santa Cruz). Once incubated for 24 h at 37 °C in a humidified atmosphere of 5% CO2, the medium with virus was changed to normal culture medium. The dCas9-mCherry positive cells were sorted using an mCherry filter setting with a FACS Aria machine (BD Biosciences) at the Cancer Science Institute of Singapore flow cytometry facility. Other dCas9 and dCas9-TET1 positive cells were selected via puromycin (2 µg/mL) or blasticidin (16 µg/mL) in the culture medium every other day; antibiotic selection was determined by the specific plasmid constructs detailed in Supplementary Table 6. The cells were further cultured and selected for more than a week to obtain stable cell lines. Once the dCas9-based cell lines were generated, sgDiRs virus with different guide RNAs was mixed in equal volume and transduced into dCas9-based stable lines with the same method as described above. The sgDiRs used for generating each stable cell line, the location of each sgDiR, can be found in Supplementary Data 5. The definition of LF, MM and RF can be found in Supplementary Data 7. All sgDiR stable cell lines were either sorted using an EGFP filter with a FACS Aria machine (BD Biosciences) at the Cancer Science Institute of Singapore flow cytometry facility, or selected with puromycin (2 µg/mL), and further assessed in culture by regularly checking the efficiency by microscopy.

Quantitative real-time PCR (qRT-PCR)

1 µg of RNA was reverse transcribed using Qscript cDNA Supermix (QuantaBio). The cDNAs were diluted 3 times for expression analysis. qRT-PCR on cDNA or ChIP-DNA was performed on 384-well plates on a QS5 system (Thermo Scientific) with GoTaq qPCR Master Mix (Promega, Madison, WI). The fold change or percentage input of the samples was calculated using the QuantStudio^TM Design & Analysis Software Version 1.2 (ThermoFisher Scientific) and represented as relative expression (ΔΔCt). Primers used in this study are listed in Supplementary Data 8.

Antibody validation

All primary antibodies utilized for Western blotting and Chromatin Immunoprecipitation (ChIP) were commercially validated by their respective manufacturers for specificity and performance in the indicated applications. Specificity validations, including knockout (KO) validation for CDKN2A/p16 and ChIP-seq verification for histone modifications using SimpleChIP enzymatic assays, are documented on the manufacturer's websites.

Western blot analysis

Total cell lysates were harvested in RIPA buffer (150 mM NaCl, 1% Nonidet P-40, 50 mM Tris, pH8.0, protease inhibitor cocktail) and protein concentrations were determined by Bradford protein assay (Bio-Rad Laboratories, Inc., Hercules, CA, USA) and absorbance was measured at 595 nm on the Tecan Infinite^® 2000 PRO plate reader (Tecan, Seestrasse, Switzerland). Equal amounts of protein from each lysate were mixed with 3× loading dye and heated at 95 °C for 10 min. The samples were resolved by 12% SDS-PAGE (running buffer: 25 mM Tris, 192 mM Glycine, and 0.1% SDS) and then transferred to PVDF membranes (transfer buffer: 25 mM Tris, 192 mM Glycine, and 20% (v/v) methanol (Fischer Chemical)). Membranes were blocked with TBST buffer containing 5% skim milk for 1 h at room temperature with gentle shaking. The blocked membranes were further washed three times with TBST buffer and incubated at 4 °C overnight with primary antibodies CDKN2A/p16 (INK4a) (ab108349, Abcam, 1:1000), β-actin (Santa Cruz β-actin (C4) Mouse monoclonal IgG1 #sc-47778, 1:5000), followed by HRP-conjugated secondary antibody incubation at room temperature for 1 h. Both the primary and secondary antibodies were diluted in 5% BSA-TBST buffer, and all the incubations were performed with gentle shaking. The immune-reactive proteins were detected using the Luminata Crescendo Western HRP substrate (Millipore). Uncropped and unprocessed scans are provided in the source data file.

Bisulfite treatment

The methylation profiles were assessed by bisulfite-conversion-based assays. For DNA bisulfite conversion, 1.6–1.8 μg of genomic DNA of each sample was converted by the EpiTect Bisulfite Kit (Qiagen) following the manufacturer’s instructions.

Methylation-specific PCR (MSP), combined bisulfite restriction analysis (COBRA) and bisulfite sequencing PCR (BSP)

The bisulfite-converted DNA samples were further analyzed using three different PCR-based methods in different assays for the methylation profiles. For Methylation-Specific PCR (MSP), both methylation-specific primers and unmethylation-specific primers of p16 were used for the PCR of the same bisulfite converted sample (the transient transfection samples in the sgRNA and sgDiR1-8 screening assay). The PCR was performed with ZymoTaq PreMix (ZYMO RESEARCH) according to the manufacturer’s instructions, with the program: 95 ⁰C 10 min, 35 cycles (95 ⁰C 30 s, 56 ⁰C 30 s, 72 ⁰C 1 min), 72 ⁰C 7 min, 4 °C hold. Two PCR products of each sample (methylated and unmethylated) were obtained and analyzed in 1.5% agarose gels.

For Combined Bisulfite Restriction Analysis (COBRA), primers specifically amplify both the methylated and unmethylated DNA (primers annealing to specific locus without any CG site) in each region were used for the PCR of the bisulfite converted samples. The PCR was performed with ZymoTaq PreMix (ZYMO RESEARCH) according to the manufacturer’s instructions, with the program: 2 cycles (95 °C 10 min, 55 °C 2 min, 72 °C 2 min), 38 cycles (95 °C 30 s, 55 °C 2 min, 72 °C 2 min), 72 °C 7 min, 4 °C hold. The PCR products were therefore loaded in a 1% agarose gel and the bands with predicted amplification size were cut out and gel-purified. 400 ng purified PCR fragments were incubated in a 20 µl volume for 2.5–3h with 1 µl of the restriction enzymes summarized in Supplementary Data 9. 100 ng of the same PCR fragments were incubated only with the restriction enzyme buffers under the same conditions as the uncut control. The uncut and cut DNA were then separated on a 2.5% agarose gel and stained with ethidium bromide.

For bisulfite sequencing PCR (BSP), primers specifically amplify both the methylated and unmethylated DNA (primers annealing to the specific locus without any CG site) were used for the PCR of the bisulfite converted samples. The PCR was performed with ZymoTaq PreMix (ZYMO RESEARCH) according to the manufacturer’s instructions, with the program: 2 cycles (95 °C 10 min, 55 °C 2 min, 72 °C 2 min), 38 cycles (95°C 30 s, 55 °C 2 min, 72°C 2 min), 72 °C 7 min, 4 °C hold. PCR products were gel-purified (Qiagen) from the 1% TAE agarose gel and cloned into the pGEM-T Easy Vector System (Promega) for transformation. The cloned vectors were transformed into Stbl3 competent cells and miniprep was performed to extract plasmids for Sanger sequencing with either sequencing primer T7 or SP6. Sequencing results were analyzed using QUMA (Quantification tool for Methylation Analysis). Samples with a conversion rate less than 95% and sequence identity less than 90%, as well as clonal variants, were excluded from our analysis. The minimum number of clones for each sequenced condition was 8. To note, the p16 gene in the HCT116 cell line has one mutated and unmethylated allele, and another wild-type and hypermethylated allele. Therefore, for BSP assay in HCT116, the mutated and unmethylated p16 allele were excluded from the analysis.

All the MSP, COBRA, and BSP primers as well as restriction enzymes can be found in Supplementary Data 9.

Whole genomic bisulfite sequencing (WGBS) and RNA-Seq

For samples generated in the DGT lab, WGBS and RNA-Seq were performed by BGI (Beijing Genomics Institute) and NovoGene (Singapore). Both companies performed their respective bisulfite conversion, library construction and sequencing on Illumina Hiseq and NovaSeq instruments. Samples were sequenced with the aim of achieving approximately 30X human genome coverage utilizing 2 × 150 paired-end reads. Sample quality metrics can be found in Supplementary Data 12 – Sheet 1.

Xenograft murine model

Mice experiments related to Fig. 5f-h were conducted according to our protocols approved by the Institutional Animal Care and Use Committee under Beth Israel Deaconess Medical Center, protocol ID 061-2022-25. All mice used for these panels were housed in a sterile barrier facility at the Beth Israel Deaconess Medical Center (BIDMC) with protocols approved by the BIDMC Institutional Animal Care and Use Committee (IACUC). Animals were maintained in standard vivarium conditions with a regulated 12-h light/dark cycle, temperature maintained at 20–24 °C, and humidity controlled between 40 and 60% with ad libitum access to standard chow and water. SNU398 cells stably expressing dCas9 were lentivirally transduced with specific sgDiR constructs targeting: 1) non-targeting control (NT1), 2) p16 promoter (G3), or 3) p16 Mesa-boundaries (G3 + G7; LF + RF). Following selection, cells were trypsinized, neutralized, and washed twice with ice-cold phosphate-buffered saline (PBS). Mice selected for xenografts were of NOD.Cg-Prkdc^scid/J, of the same age (~12 weeks old) and were mixed gender. n = 9 mice were selected for each treatment group. At the experimental endpoint, mice were euthanized via gradual CO₂ asphyxiation, followed by cervical dislocation as a secondary physical method to ensure death, in strict accordance with AVMA guidelines and approved IACUC protocols. For xenograft injections, a master mix was prepared by resuspending the cells in PBS at a density of 1.5 million cells per 50 µL. Immediately prior to injection, the cell suspension was mixed 1:1 with Matrigel (Corning, #354277) on ice. A total volume of 100 µL (containing 1.5 million cells) was injected subcutaneously into the flanks of each mouse using pre-chilled syringes. Following an initial 2.5-week period for tumor engraftment and establishment, tumor volume kinetics were monitored every 2–3 days for 14 days.

Mice experiments related to Supplementary Fig. 5e-g were conducted according to protocols approved by the Institutional Animal Care and Use Committee under National University of Singapore. Animals were maintained in standard vivarium conditions with a regulated 12-h light/dark cycle, temperature maintained at 20–24 °C, and humidity controlled between 40 and 60% with ad libitum access to standard chow and water. Mice of the same age (~12 weeks old) were tagged and grouped randomly for control and test treatments. Mixed gender mice were used in this experiment. At the experimental endpoint, mice were euthanized via gradual CO₂ asphyxiation, followed by cervical dislocation as a secondary physical method to ensure death. 2.5 million SNU398 HCC cells were injected into the flanks of NOD.Cg-Prkdc^scid/J mice after transduction with 1) CRISPR-DiR non-targeting control NT1, or 2) CRISPR-DiR targeting p16 region LF + RF. The tumor sizes were traced and recorded every 2 days for 12 days in total.

Chromatin immunoprecipitation (ChIP)

ChIP was performed as described previously⁵⁶. Briefly, samples of 60 million cells were trypsinized by washing once at room temperature in PBS, then every 50–60 million cells were resuspended in 30 ml of room-temperature PBS. Cells were fixed with 1% formaldehyde for 8 min at room temperature with rotation. Excessive formaldehyde was quenched with 0.25 M glycine. Fixed cells were washed twice with cold PBS supplemented with 1 mM PMSF. After washing with PBS, cells were lysed with ChIP SDS lysis buffer (100 mM NaCl,50 mM Tris-Cl pH8.0, 5 mM EDTA, 0.5% SDS, 0.02% NaN₃, and fresh protease inhibitor cOmplete tablet EDTA-free (5056489001, Roche) and then stored at −80 °C until further processing. Nuclei were collected by spinning down at 3000 rpm at 4 °C for 10 min. The nuclear pellet was resuspended in IP solution (2 volume ChIP SDS lysis buffer plus 1 volume ChIP triton dilution buffer (100 mM Tris-Cl pH8.6, 100 mM NaCl, 5 mM EDTA, 5% Triton X-100), and fresh proteinase inhibitor) with 10 million cells/ml IP buffer concentration for sonication using a Bioruptor (8–10 cycles, 30 s on, 30 s off, High power) to obtain 200 bp to 500 bp DNA fragments. After spinning down to remove debris, 1.2 ml sonicated chromatin was pre-cleared by adding 50 µl washed Dynabeads protein A (Thermo Scientific) and rotating at 4 °C for 2 h. Pre-cleared chromatin was incubated with antibody pre-bound dynabeads protein A (Thermo Scientific) overnight at 4 °C. For histone marker antibodies, 50 μl of Dynabeads protein A was loaded with 3 μg antibody. The next day, magnetic beads were washed through the following steps: buffer 1 (150 mM NaCl, 50 mM Tris-Cl, 1 mM EDTA, 5% sucrose, 0.02% NaN₃, 1% Triton X-100, 0.2% SDS, pH 8.0) two times; buffer 2 (0.1% deoxycholic acid, 1 mM EDTA, 50 mM HEPES, 500 mM NaCl, 1% Triton X-100, 0.02% NaCl, pH 8.0) two times; buffer 3 (0.5% deoxycholic acid, 1 mM EDTA, 250 mM LiCl, 0.5% NP40, 0.02% NaN₃) two times; TE buffer one time. To reverse crosslinks, samples were incubated with 20 μg/ml proteinase K (Ambion) at 65 °C overnight. The samples were then extracted with phenol:chloroform:isoamyl alcohol (25:24:1) followed by chloroform, ethanol precipitated in the presence of glycogen, and re-suspended in 10 mM Tris buffer (pH 8). After reverse crosslinking and purification of DNA, qPCR was performed with the primers listed in Supplementary Data 10. Briefly, p16 primer detecting the enrichments of all histone markers is located in the proximal promoter region within 100 bp around TSS; primers located 50 kb upstream of p16 (Neg 1) and 10 kb downstream of p16 (Neg 2) are the negative control primers located in the regions without enrichment of any of the above proteins. The antibodies used in ChIP assays were: H3K4Me3 (C42D8, #9751, Cell Signaling Technologies), H3K27Ac (ab45173, Abcam), H3K9Me3 (D4W1U, #13969, Cell Signaling Technologies), Rabbit IgG monoclonal (ab172730, Abcam).

Circularized chromosome conformation capture (4C)-Seq

4C-seq was performed as described previously⁵⁷ with modifications⁵⁸. In brief, SNU398 cells with stable CRISPR-DiR treatment for 13 days were used for 4C-Seq. 30 million cells guided by NT1 non-targeting or 30 million cells guided by guides (G3, G4, G6, and G7) targeting region LF + RF were crosslinked in 1% formaldehyde for 10 min at RT with rotation. Then formaldehyde was neutralized by adding 2.5 M glycine to a final concentration of 0.25 M and rotating for 5 min at RT. After washing in cold PBS, cells were resuspended in 9 ml lysis buffer (10 mM Tris-HCl pH8.0, 10 mM NaCl, 5 mM EDTA, 0.5% NP 40), with addition of EDTA-free protease inhibitor (complete tablet, freshly dissolved in nuclease-free water to make a 100X stock, 5056489001, Roche) and lysed multiple times with resuspension every 2–3 min during the 10 min incubation on ice. After lysis, each lysate was split into two 15 ml tubes for Viewpoint 1 (Csp6I) or Viewpoint 2 (DpnII), respectively (4.5 ml/tube, 15 million cells). After spinning down at 3000 rpm for 10 min at 4 °C, each nuclear preparation was washed with 500 µl 1X CutSmart buffer from NEB and spun at 800 g for 10 min at 4 °C, followed by resuspension into 450 µl nuclease free H₂O and transferring exactly 450 µl of the sample into a 1.5 mL tube. To each tube, 60 µl of 10X restriction enzyme buffer provided together with the corresponding restriction enzyme (Viewpoint 1: 10X Buffer B (ER0211, Invitrogen); Viewpoint 2: 10X NEBuffer DpnII (R0543M, NEB)) and 15 µl of 10% SDS buffer were added to the 450 µl sample, followed by an incubation at 37 °C for 1 h with shaking (900 RPM, Eppendorf Thermomixer), followed by adding 75 µl of 20% Triton X-100 to each tube for 1h incubation at 37 °C with shaking (900 RPM). 20 µl samples from each tube were taken out as “undigested” and stored at −20 °C. For Viewpoint 1, 50 µl Csp6I (ER0211, Invitrogen) was added (500 U per tube) together with 5.6 µl 10X Buffer B (ER0211, Invitrogen) for 18 h digestion at 37 °C with shaking (700 RPM); for Viewpoint 2, 10 µl DpnII (R0543M, NEB) (500 U per tube) together with 8 µl nuclease free H₂O and 2 µl 10X NEBuffer DpnII were added for 18 h digestion at 37 °C with shaking (700 RPM). The next day, after removing 20 µl of the sample for de-crosslinking, confirming a digestion efficiency over 80%, and performing PicoGreen DNA quantification to check the DNA concentration in each reaction, 10 µg of the digested DNA was taken out into a new tube and the volume was adjusted to 600 µl with nuclease free H₂O. The samples were heat-inactivated at 65 °C for 20 min. Heat-inactivated chromatin was added to 1X ligation buffer (EL0013, Invitrogen) supplemented with 1% Triton X-100, 0.1 mg/ml BSA, and the volume was adjusted with nuclease free H₂O to 10 ml with a final DNA concentration of 1 ng/µl. After adding 660 U T4 DNA ligase (EL0013, Invitrogen 30 U/µl), samples were incubated at 16 °C in thermal incubator without shaking. The next day, a final concentration of 0.5% SDS and 0.05 mg/ml proteinase K (Ambion) was added to each sample, followed by 65 °C incubation overnight for de-crosslinking. The next day, after adding 30 µl of RNase A (10 mg/ml, PureLink), samples were incubated at 37 °C for 1 h, followed by phenol: chloroform DNA purification. The chromatin was extracted with phenol:chloroform:isoamyl alcohol (25:24:1) followed by chloroform, ethanol precipitated (split to 5 ml/tube and topped up with nuclease free H₂O to 15 ml, then adding 100% ethanol to 68% to avoid SDS precipitation) in the presence of glycogen and dissolved in 10 mM Tris buffer (pH 8). The ligated chromatin was analyzed by agarose gel electrophoresis and the concentration was determined by QUBIT HS DNA kit. 7 µg of ligated chromatin was digested with 10 U specific second cutter NlaIII (R0125S, NEB) in 100 µl system with CutSmart Buffer (NEB), 37 °C overnight without shaking. 5 µl digested chromatin was analyzed by gel electrophoresis prior to heat inactivation. The restriction enzyme was heat-inactivated by incubating the chromatin at 65 °C for 20 min. 7 µg NlaIII digested chromatin was ligated with T4 DNA ligase (EL0013, Invitrogen, 30U/µl) at 20 U/ml in 1X ligation buffer (EL0013, Invitrogen), incubated at 16 °C overnight. The ligated DNA was recovered by phenol:chloroform:isoamyl alcohol (25:24:1) extraction and ethanol precipitated. 100 ng DNA of each sample was used for 4C library preparation. The library was constructed by inverse PCR and nested PCR with KAPA HiFi HotStart ReadyMix (KK2602). The 1st PCR was performed at 100 ng DNA + 1.75 µl 1st PCR primer mix + 12.5 µl KAPA HiFi HotStart ReadyMix + H₂O to 25 µl. The 1st PCR program was 95 °C, 3 min, 15 cycle of (98 °C, 20 s; 65 °C, 15 s; 72 °C, 1 min), 72 °C, 5 min, 4 °C hold. The 1st PCR products were purified with MinElute PCR Purification Kit (28004, Qiagen) and eluted in 13 µl Elution Buffer in the kit. The 2nd PCR was performed at purified 1st PCR product + 1.75 µl 2nd PCR primer mix + 12.5 µl KAPA HiFi HotStart ReadyMix + H₂O to 25 µl. The 2nd PCR program was 95 °C, 3 min, 13 cycle of (98 °C, 20 s; 65 °C, 15 s; 72 °C, 1 min), 72 °C, 5 min, 4 °C hold. The 2nd PCR products were purified using MinElute PCR Purification Kit (28004, Qiagen) and eluted in 10 µl Elution Buffer in the kit. The primer mix was 5 µl 100 M forward primer + 5 µl 100 M reverse primer + 90 µl H₂O. All primer sequences and barcodes are listed in Supplementary Data 11. The libraries were subjected to size selection (250–600 bp) on a 4–20% TBE PAGE gel (Thermo Scientific). The TBE gel was run at 180 V, 55 min, stained with Sybr Safe and visualized with gel-safe, and the libraries were extracted from PAGE using a gel-crush protocol. Picogreen quantification, Bioanalyzer, and KAPA library quantification were performed to check the quality, size and amount of the recovered libraries, and NextSeq 500/550 Mid Output kit V2.5 (150 Cycles) (20024904, Illumina) was used for single-end NextSeq sequencing.

Statistical analysis

Methylation changes of clones analyzed by bisulfite sequencing PCR (BSP) were calculated using the online methylation analysis tool QUMA. Subsequent figures and heatmaps were generated in R⁵⁹ using pheatmap⁶⁰. For mRNA qRT-PCR and ChIP-qPCR, p-values were calculated using paired/unpaired t-tests as indicated in figure legends using GraphPad Prism Software (v9.2 - 10.2.2). P < 0.05 values were considered statistically significant (*P < 0.05; **P < 0.01; ***P < 0.001). Miscellaneous bar graphs and statistical tests were performed using Prism (v9.2 – 10.2.2).

Bioinformatic analysis

Histone marks ChIP-Seq Analysis

The human and mouse histone mark (H3K4me1, H3K4me2(human), H3K4me3, H3K9ac(mouse) H3K14ac, H3K27ac, H3K36me3, H3K27me2, H3K27me3, H3K9me3) enrichments shown in Fig. 3a-d were determined by analyzing ChIP-seq data across the cell lines A549, GM12878, H1-hESC, HCT116, HEPG2, HMEC, HSMM, HUVEC, K562, NHEK, NHLF, and SNU398 obtained from ENCODE and GEO (downloaded Feb/Mar-21). Supplementary Data 13 contains the tables indicating which accession IDs were utilized for each analysis. Once all peak sets were analyzed, in order to evaluate the intersection of genomic features across sample cohorts, consensus binary overlap matrices were segregated into distinct activation and repression subsets. For each category, combination matrices were constructed to calculate the absolute size of all unique peak intersections. Visualization of these multi-set intersections was rendered as UpSet plots utilizing the ComplexHeatmap package^61,62.

ChIP-Seq datasets were processed using ChIP-AP v5.1−5.4⁶³ using the following settings tables for human and mouse, respectively. Called peaks were then overlapped with identified Mesa coordinates using GenomicRanges⁶⁴. Only strict overlaps were accepted (at least 1 bp overlap between the Mesa and the called peak).

Human settings table (ChIP-AP)

See Table 1

Table 1.

Human settings table (ChIP-AP)

program	argument
fastqc1
clumpify	dedupe spany addcount qout=33 fixjunk
bbduk	ktrim=r k = 21 mink=8 hdist=2 hdist2 = 1
trimmomatic	LEADING:20 SLIDINGWINDOW:4:20 TRAILING:20 MINLEN:20
fastqc2
bwa_mem
samtools_view	-q 20
plotfingerprint
fastqc3
macs2_callpeak
gem	-Xmx36G –k_min 8 –k_max 12
sicer2
homer_findpeaks
genrich	-adjustp -v
homer_mergePeaks
homer_annotatePeaks
fold_change_calculator	-normfactor uniquely_mapped

Open in a new tab

This is the settings table utilized for all ChIP-Seq processing using ChIP-AP for human samples. This table is passed as a tab-delimited txt or tsv input into ChIP-AP (along with the raw fq files) to re-process the ChIP-Seq datasets as we did in our manuscript (with some minor deviations only from gem that are unavoidable).

Mouse settings table (ChIP-AP)

See Table 2

Table 2.

Mouse settings table (ChIP-AP)

program	argument
fastqc1
clumpify	dedupe spany addcount qout=33 fixjunk
bbduk	ktrim=r hdist=21 mink=8 hdist=2 hdist2 = 1
trimmomatic	LEADING:20 SLIDINGWINDOW:4:20 TRAILING:20 MINLEN:20
fastqc2	-q
bwa_mem
samtools_view	-q 20
plotfingerprint
fastqc3	-q
macs2_callpeak
gem	-Xmx10G –k_min 8 –k_max 12
sicer2
homer_findpeaks
genrich	-adjustp -v
homer_mergePeaks
homer_annotatePeaks
fold_change_calculator	-normfactor uniquely_mapped

Open in a new tab

This is the settings table utilized for all ChIP-Seq processing using ChIP-AP for mouse samples. This table is passed as a tab-delimited txt or tsv input into ChIP-AP (along with the raw fq files) to re-process the ChIP-Seq datasets as we did in our manuscript (with some minor deviations only from gem that are unavoidable).

Publicly available datasets

Matched WGBS and RNA-Seq data for the PDx and CT26 models were obtained from GEO GSE192745. The PDx model represents the human MSS Colorectal Cancer tissue grafted into NOD-Prkdc^em26Cd52Il2rg^em26Cd22/NjuCrl mice treated with 20 μg decitabine (i.p.) for 5 days. The CT26 murine model represents a mouse colorectal cancer line treated with 1 μM decitabine for 72 h. Full sample details can be obtained from the publishing manuscript²⁷. The HCT116 samples were obtained from GEO GSE97889⁵⁵.

WGBS and MM analysis

Raw WGBS sequenced fastq files were first processed with MSuite (v2)⁶⁵ aligned to hg38 using the flags “-x hg38 -minsize 15 -maxins 1500 -cut-r1-head 8 -cut-r2-head 8”. Complete sequencing quality metrics for all WGBS datasets, including sequencing depth, mapping efficiency, and effective genome coverage, are provided in Supplementary Data 12 – Sheet 1. Custom scripts were then used to reformat the output for Defiant⁶⁶ and Metilene⁶⁷. Defiant was run with the flags “-c 0,2,1 -d 1,3,1 -G147 -CpN5 -p 0.05.” Metilene was run with flags “-m 3 -d 0.1 -M 147 -f 1.” R was then used to merge all outputs, annotations and calculate the average methylation percentages in all regions of interest. For gene annotations, HOMER⁶⁸ was used for either hg38 or mm9. CpG Island annotations and gene definitions were provided by annotatr⁶⁹, an R package from Bioconductor, which uses the AnnotationHub package internally⁷⁰. All figures were generated in R (v4.1.2 - v4.5.2)⁵⁹ using the packages pheatmap⁶⁰, ComplexHeatmap^61,62, ggplot2⁷¹, UpsetR⁷², and Sushi⁷³.

RNA-Seq analysis

Raw fastq files were first processed with clumpify from the BBMap suite⁷⁴, with the flags “dedupe=t spany addcount qout=33 fixjunk.” Next, bbduk⁷⁴ was used for adapter trimming with flags “ktrim=r k=21 mink=8 hdist=2 hdist2=l.” Quality trimming was performed by trimmomatic⁷⁵ with flags “LEADING:24 SLIDINGWINDOW:4:24 TRAILING:24 MINLEN:20.” STAR⁷⁶ was used for alignment with the flags “-readFilesCommand zcat -outFilterScoreMinOverLread 0.25 -outFilterMatchNminOverLread 0.25 -outFilterMultimapNmax 5.” Samtools⁷⁷ was used to sort and index generated bam files. Coverage files were generated using bamCoverage from the deepTools suite⁷⁸ with a bin size of 10 bp. Raw counts tables were generated with featureCounts⁷⁹ with the flags “-g gene_id --extraAttributes gene, name, transcript_id -p -M -O -J -f -d 20” to generate gene-level counts and “-t five_prime_utr -g gene_id --extraAttributes gene_name, transcript_id -p -M -O -J -f -d 20” to generate 5’UTR-level counts. The raw counts were then processed in R using the edgeR package to generate TMM-normalized counts used throughout the manuscript⁶⁸.

Permutation testing

To evaluate the non-random positional conservation of Mesa, first, baseline overlaps between pairwise Mesa sample coordinates (in BED format) were utilized to determine the true observed spatial intersection. To determine the statistical significance of these overlaps against a random background distribution, 10,000 permutations were executed for every unique pairwise comparison. In each permutation iteration, the spatial coordinates of the query sample were randomly shuffled across the reference genome using bedtools⁸⁰. This shuffling was explicitly constrained by established human chromosome sizes (hg38.chrom.sizes) to ensure that the randomly generated background regions preserved the exact quantity and size distribution of the original Mesa intervals while restricting placement to valid genomic bounds. The total base-pair overlap between the permuted query and the fixed target sample was recorded to generate a null distribution. Empirical p-values were calculated by determining the proportion of permuted overlaps that equalled or exceeded the true observed overlap, adjusted with a strict pseudo-count of 1 to prevent zero values. Standardized Z-scores were calculated to quantify the magnitude of positional enrichment above the random expected mean. The resulting cross-sample Z-scores were formatted into a symmetric pairwise matrix and mapped to a divergent color scale for visualization of pairwise sample concordance.

To visualize this genome-wide spatial concordance, circular genomic density profiles were constructed using the circlize package⁸¹. Mesa coordinates from individual patient samples were compiled and a core “conserved” Mesa subset was defined by isolating spatial intervals exhibiting overlap across multiple individual patient cohorts. The regional genomic density of Mesa for each individual patient, alongside the consensus conserved subset, was calculated and plotted as concentric density tracks aligned to hg38 chromosomal ideograms.

All permutation analyses, data scaling, and spatial visualizations were conducted in R (v4.5.x)⁵⁹. Core spatial intersections and genomic shuffling were executed via calls to bedtools⁸⁰. Final pairwise matrix visualizations were rendered utilizing the corrplot package⁸², while circular genomic annotations and legends were generated using circlize⁸¹ and ComplexHeatmap^61,62.

Permutation testing for selection bias

To validate the robustness of the methylation-expression correlation and rule out preselection bias, a Monte Carlo permutation framework was employed. Gene expression labels (Log2FC) were randomly shuffled against spatial methylation coordinates for 10,000 iterations. For each permutation, the identical biological upregulation threshold (>1 Log2FC) was applied, and the median difference in demethylation between the Methylation Mesa and canonical promoter features (1 kb promoter, 300 bp promoter, CpG island) was calculated to establish the expected null distribution. Empirical p-values were determined by the proportion of random permutations exhibiting a demethylation differential as extreme as or greater than the observed biological data.

Common language effect size (CLES) calculations

To quantify the magnitude of CpG density enrichment at Mesa relative to its flanks, the number of CpG sites per window of interest was first calculated. Distributions were analyzed across 20 genomic windows flanking the Mesa, extending bidirectionally to ±1.47 kb. CLES values were calculated for every sequenced sample across all genomic compartments of interest based on HOMER annotation. The CLES⁸³ was calculated using the equation CLES = $U / n_{m e s a} * n_{f l a n k}$ where U is the Wilcoxon W statistic (Mann-Whitney U), and n_mesa and n_flank represent the number of observations in the Mesa and flanking windows, respectively. The Wilcoxon rank-sum test was calculated per window to determine significance, with p-values adjusted by the Benjamini-Hochberg (BH) correction. All analyses were conducted in R (v4.5.x)⁵⁹. Data manipulation and visualization were performed using the tidyverse suite⁸⁴. Parallel processing was implemented via doParallel. Statistical testing and plot annotations were integrated using rstatix⁸⁵ and ggpubr⁸⁶. Feature distributions were visualized as pirate plots combining violin plots, boxplots, and raw data jittering via ggbeeswarm⁸⁷.

Methylation-expression association analysis

To evaluate the relationship between demethylation and transcriptional reactivation, first, gene Trimmed Mean of M-values (TMM) was calculated for both naïve and treated samples. To prevent zero-division artifacts, a standard pseudo-count was added to the TMM values, and a Log₂ Fold Change (Log₂ FC) was calculated directly between the treated and baseline (naïve/wild-type) conditions. A core cohort of upregulated genes was defined by a magnitude threshold of Log₂ FC > 1. For each upregulated gene, exact spatial coordinates were mapped for the 5’ UTR, the canonical Promoter-TSS (defined bidirectionally as ±1000 bp from the transcription start site), and any overlapping Mesa. The absolute change in methylation was calculated by subtracting the treated methylation percentage from the baseline. To assess the regulatory power of the Mesa against broader canonical features, mean Mesa methylation was calculated and compared to the average methylation across the corresponding matched canonical Promoter-TSS and 5’ UTR regions. Statistical significance between the demethylation magnitudes of these distinct spatial regions was determined using the non-parametric Wilcoxon rank-sum test, with p-values adjusted via Benjamini-Hochberg (BH) correction. All analyses were conducted in R (v5.4.x)⁵⁹. Data manipulation and spatial intersections were performed using the tidyverse suite⁸⁴. Statistical testing and plot annotations were integrated using rstatix⁸⁵ and ggpubr⁸⁶. Feature distributions were visualized as pirate plots combining violin plots, boxplots, and raw data jittering via ggbeeswarm⁸⁷.

Mesa/CGI signal intensity quantification

To visualize the distribution of DNA methylation across Mesa, flanking regions, and CpG Islands, genome-wide average signal intensity profiles were constructed. Regional DNA methylation values were extracted from sample bigWig files using the computeMatrix module from the deepTools suite⁷⁸ with flags “-referencePoint center -missingDataAsZero”. Methylation intensities were aggregated bidirectionally across a flanking window of ±1.5 kb (-a 1500 -b 1500). Following matrix generation, the spatial data were imported into R⁵⁹ for downstream signal processing and visualization. To reduce technical spatial noise and delineate global trends, the mean column-wise signal intensities across the predefined bins were smoothed utilizing a rolling mean algorithm. Data parsing and spatial smoothing were conducted using the tidyverse suite⁸⁴ and the zoo package⁸⁸, respectively, and the final composite signal intensity profiles were rendered using ggplot2⁷¹.

4C-Seq analysis

For the 4C-seq analysis, the long-range genomic interaction regions generated by the 4C-Seq experiment were processed using the CSI NGS portal⁸⁹. Briefly, raw fq files were aligned to a masked hg19 reference (masked for the gap, repetitive and ambiguous sequences) using bwa mem⁹⁰. Bam files were converted to read coverage files by bedtools genomecov⁸⁰. The read coverage was normalized according to the sequencing depth. BedGraph files of the aligned bams were converted to bigWig format by bedGraphToBigWig. Next, the processed alignment files were analyzed using r3CSeq⁹¹ and using the associated masked hg19 genome (BSgenome.Hsapiens.UCSC.hg19.masked)⁹², from the R Bioconductor repository. Chromosome 9 was selected as the viewpoint, and Csp6I, DpnII were used as the restriction enzymes to digest the genome. Smoothed bam coverage maps were generate using bamCoverage from the deeptools suite⁷⁸ with the flags “-normalizeUsing RPGC -binSize 2000 -smoothLength 6000 -effectiveGenomeSize 2864785220 -outFileFormat bedgraph” and plotted using the Bioconductor package Sushi⁷³ to get the viewpoint coverage depth maps. BigInteract files for UCSC and bedpe files were manually generated with the “score” values being calculated as -log(interaction_q-value_from_r3CSeq + 1*10⁻¹⁰). Sushi was then used to plot the bedpe files to get the 4C looping plots. To identify differential interaction peaks, HOMER’s⁶⁸ get DifferentialPeaks was used with the flag “-F 1.5” after which the corresponding bigInteract and bedpe files were generated as described. LiftOver was used to convert hg38 Mesa coordinates to hg19 coordinates for overlapping.

Software versions utilized

Bioinformatic processing, statistical analyses, and data visualizations were performed using R (v4.1.2–v4.5.2). Short-read sequencing data preprocessing and alignment utilized the BBMap suite (v38.18), Trimmomatic (v0.39), STAR (v2.7.10b), BWA-MEM (v0.7.17-r1188), and Samtools (v1.19.2). Whole-genome bisulfite sequencing (WGBS) and differential methylation analyses were conducted using MSuite (v2.0.0–v2.2.0), metilene (v0.2-8), and Defiant. ChIP-Seq datasets were processed using the ChIP-AP pipeline (v5.1–v5.4). RNA-Seq read quantification was performed with featureCounts (v2.1.1). Genome-wide signal extraction and coordinate intersections utilized deepTools (v3.5.4) and bedtools (v2.31.1). Genomic annotation was performed using HOMER (v5.1), alongside Bioconductor packages GenomicRanges (v1.64.0), annotatr (v1.38.0), and AnnotationHub (v4.2.0). Differential expression and statistical testing were executed using edgeR (v4.10.0), rstatix (v0.7.3), and regioneR (v1.44.0). Data wrangling and spatial visualizations were performed using the tidyverse suite (v2.0.0), pheatmap (v1.0.13), circlize (v0.4.18), ggplot2 (v4.0.3), ggpubr (v0.6.3), UpSetR (v1.4.0), corrplot (v0.95), and zoo (v1.8.15), with parallelization handled by doParallel (v1.0.17). Web-based methylation analyses were conducted using QUMA (version not applicable), and 4C-Seq initial processing was routed through the CSI NGS Portal (retired).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information^{(19.2MB, pdf)}

41467_2026_76648_MOESM2_ESM.pdf^{(118.4KB, pdf)}

Description of Additional Supplementary Files

Supplementary Data 1^{(42.4KB, xlsx)}

Supplementary Data 2^{(28.5KB, xls)}

Supplementary Data 3^{(12.2KB, xlsx)}

Supplementary Data 4^{(10.6KB, xlsx)}

Supplementary Data 5^{(11.1KB, xlsx)}

Supplementary Data 6^{(10.4KB, xlsx)}

Supplementary Data 7^{(9.8KB, xlsx)}

Supplementary Data 8^{(9.5KB, xlsx)}

Supplementary Data 9^{(11.2KB, xlsx)}

Supplementary Data 10^{(9.7KB, xlsx)}

Supplementary Data 11^{(12.6KB, xlsx)}

Supplementary Data 12^{(19.6KB, xlsx)}

Supplementary Data 13^{(39KB, xlsx)}

Reporting Summary^{(104.7KB, pdf)}

Transparent Peer Review file^{(3.3MB, pdf)}

Source data

Source data^{(20.2MB, xlsx)}

Acknowledgements

Y.V.L. would like to thank Wei Liu for constructive discussions that led to the inspiration for this work. We thank Michelle Mok Meng Huang and Chelsia Qiuxia Wang for assistance with flow cytometry experiments; Melissa J. Fullwood and Ying Zhang for providing suggestions and help for the 4C experiments; S. Jha, H.P. Koeffler, and W.J. Chng for their insightful suggestions. We thank Alex Ebralidze for providing technical training and guidance to Y.V.L. We also thank the Tenen lab members and Robert S. Welner for their assistance in the experiments and in reviewing the manuscript. We’d also like to thank our collaborators and their patients who have donated samples in the name of science. J.G.D. would like to thank the Functional Genomics Consortium for their continued support.

Author contributions

D.G.T. initiated the project and provided guidance. Y.V.L conceived the idea and designed CRISPR-DiR. Y.V.L. first observed the sDMR profile. M.A.B. and Y.V.L characterized the MM signature, prepared figures and wrote the manuscript. M.A.B. conceived and wrote the Mesa identification workflow (without AI), performed all bioinformatics data analysis and interrogations. A.D.R. provided technical training and guidance for Y.V.L. Y.V.L., M.A.B., A.D.R., and T.B. designed experiments. Y.V.L. performed biochemical and cellular experiments. M.K.J., J.Z., Y.V.L., L.C., J.P.T., and M.T.N.L. designed and carried out in vivo experiments. A.D.R. and S.U. provided the datasets for ITL-MDS, KG1a and K562. S.U. carried out the AZA treatment of KG1a and K562. J.K. and H.W carried out experiments. D.T. and D.H. provided the Adelaide-origin samples. J.S. and M.A.B. processed the ChIP-Seq datasets used in the histone footprint map. Q.Z. and H.K.T. carried out experiments and provided suggestions. S.M., J.K., L.C., T.B., and J.G.D. provided suggestions for the project.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This research is supported by the National Research Foundation Singapore and the Singapore Ministry of Education under its Research Centres of Excellence initiative, as well as the RNA Biology Center at the Cancer Science Institute of Singapore, as part of funding under the Singapore Ministry of Education’s AcRF Tier 3 grants, Grant number MOE2014-T3-1-006. D.G.T. was funded by the Singapore Ministry of Health’s National Medical Research Council under its Singapore Translational Research (STaR) Investigator Award (MOH-STaR18nov-0002), the AGA/Jenzabar Research Fund, and NIH grants 1R35CA197697 and P01HL131477; A.D.R. by the DOD BMFRP Idea Development Award W81XWH-20-1-0518, NCI R00CA188595, NIH R01DK136116, HIRM Pilot Award 2021, AIRC start up grant 2014-15347, and Fondazione Cariplo N. 2016-0476; L.C. by NIH/NHLBI grant P01HL158688 and the AGA/Jenzabar and Xiu research funds; T.B. by the Canada Research Chairs program. J.G.D. receives funding support from the Functional Genomics Consortium: AbbVie, Bristol Myers Squibb, Janssen, Merck, and Vir Biotechnology. M.T.N.L. is funded by the Target Translation Consortium. D.T. is supported by a Commonwealth Serum Laboratories Centenary Fellowship, Australian Medical Research Future Fund and National Health and Medical Research Council Ideas Grants 1182564 and 1184485 and The Leukemia & Lymphoma Society (www.lls.org), Snowdome Foundation (www.snowdome.org) and the Leukemia Foundation (www.leukemia.org.au) grants 6619-21 and 6650-23.

Data availability

The raw and processed WGBS and RNASeq data generated in this study associated with cell lines have been deposited in the Gene Expression Omnibus (GEO) under accession code GSE331137. The raw patient-derived sample WGBS and RNASeq data are protected and are not available due to ethics compliance restrictions. The processed patient-derived data are available through the GEO repository. For access to the raw sequencing data, contact the corresponding authors.

All publicly available data processed in this study are listed in Supplementary Data 13. Source data are provided with this paper.

Code availability

The methods presented herein list the utilized tools with the required flags to facilitate analysis reproducibility. Code written for analysis and figure generation can be found in our Mesa Identification Workflow (MIW) GitHub (https://github.com/mbassalbioinformatics/MIW - https://doi.org/10.5281/zenodo.21416590).

Competing interests

J.G.D. consults for Microsoft Research, BioNTech, PhenomicAI, Servier, Quotient Therapeutics, Patient Square Capital, Samsung, and Pfizer. J.G.D. receives funding support from the Functional Genomics Consortium, AbbVie, Bristol Myers Squibb, Janssen, and Merck. J.G.D.’s interests are reviewed and managed by the Broad Institute in accordance with its conflict of interest policies.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

These authors jointly supervised this work Tenen, D.G, Bassal, M.A.

Contributor Information

D. G. Tenen, Email: dtenen@bidmc.harvard.edu

M. A. Bassal, Email: mahmoud.bassal@mymail.unisa.edu.au

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76648-4.

References

1.Jones, P. A. Functions of DNA methylation: islands, start sites, gene bodies and beyond. Nat. Rev. Genet.13, 484–492 (2012). [DOI] [PubMed] [Google Scholar]
2.Esteller, M., Corn, P. G., Baylin, S. B. & Herman, J. G. A gene hypermethylation profile of human cancer. Cancer Res.61, 3225–3229 (2001). [PubMed] [Google Scholar]
3.Herman, J. G. & Baylin, S. B. Gene silencing in cancer in association with promoter hypermethylation. N. Engl. J. Med.349, 2042–2054 (2003). [DOI] [PubMed] [Google Scholar]
4.Feinberg, A. P. & Tycko, B. The history of cancer epigenetics. Nat. Rev. Cancer4, 143–153 (2004). [DOI] [PubMed] [Google Scholar]
5.Gardiner-Garden, M. & Frommer, M. CpG islands in vertebrate genomes. J. Mol. Biol.196, 261–282 (1987). [DOI] [PubMed] [Google Scholar]
6.Larsen, F., Gundersen, G., Lopez, R. & Prydz, H. CpG islands as gene markers in the human genome. Genomics13, 1095–1107 (1992). [DOI] [PubMed] [Google Scholar]
7.Takai, D. & Jones, P. A. Comprehensive analysis of CpG islands in human chromosomes 21 and 22. Proc. Natl. Acad. Sci. USA99, 3740–3745 (2002). [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Wang, Y. & Leung, F. C. An evaluation of new criteria for CpG islands in the human genome as gene markers. Bioinformatics20, 1170–1177 (2004). [DOI] [PubMed] [Google Scholar]
9.Saxonov, S., Berg, P. & Brutlag, D. L. A genome-wide analysis of CpG dinucleotides in the human genome distinguishes two distinct classes of promoters. Proc. Natl. Acad. Sci. USA103, 1412–1417 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Nunez, J. K. et al. Genome-wide programmable transcriptional memory by CRISPR-based epigenome editing. Cell184, 2503–2519 e2517 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Han, H. et al. DNA methylation directly silences genes with non-CpG island promoters and establishes a nucleosome-occupied promoter. Hum. Mol. Genet.20, 4299–4310 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Koch, A. et al. Analysis of DNA methylation in cancer: location revisited. Nat. Rev. Clin. Oncol.15, 459–466 (2018). [DOI] [PubMed] [Google Scholar]
13.van Vlodrop, I. J. et al. Analysis of promoter CpG island hypermethylation in cancer: location, location, location!. Clin. Cancer Res.17, 4225–4231 (2011). [DOI] [PubMed] [Google Scholar]
14.Hodges, E. et al. Directional DNA methylation changes and complex intermediate states accompany lineage specificity in the adult hematopoietic compartment. Mol. Cell44, 17–28 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Irizarry, R. A. et al. The human colon cancer methylome shows similar hypo- and hypermethylation at conserved tissue-specific CpG island shores. Nat. Genet.41, 178–186 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Zinn, R. L., Pruitt, K., Eguchi, S., Baylin, S. B. & Herman, J. G. hTERT is expressed in cancer cell lines despite promoter DNA methylation by preservation of unmethylated DNA and active chromatin around the transcription start site. Cancer Res.67, 194–201 (2007). [DOI] [PubMed] [Google Scholar]
17.van Vlodrop, I. J. et al. Prognostic significance of Gremlin1 (GREM1) promoter CpG island hypermethylation in clear cell renal cell carcinoma. Am. J. Pathol.176, 575–584 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Li, G. et al. Extensive promoter-centered chromatin interactions provide a topological basis for transcription regulation. Cell148, 84–98 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Mifsud, B. et al. Mapping long-range promoter contacts in human cells with high-resolution capture Hi-C. Nat. Genet.47, 598–606 (2015). [DOI] [PubMed] [Google Scholar]
20.Sanyal, A., Lajoie, B. R., Jain, G. & Dekker, J. The long-range interaction landscape of gene promoters. Nature489, 109–113 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
21.Fulco, C. P. et al. Systematic mapping of functional enhancer-promoter connections with CRISPR interference. Science354, 769–773 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Xu, X. et al. A CRISPR-based approach for targeted DNA demethylation. Cell Discov.2, 16009 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Morita, S. et al. Targeted DNA demethylation in vivo using dCas9-peptide repeat and scFv-TET1 catalytic domain fusions. Nat. Biotechnol.34, 1060–1065 (2016). [DOI] [PubMed] [Google Scholar]
24.Derissen, E. J., Beijnen, J. H. & Schellens, J. H. Concise drug review: azacitidine and decitabine. Oncologist18, 619–624 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Stresemann, C. & Lyko, F. Modes of action of the DNA methyltransferase inhibitors azacytidine and decitabine. Int. J. Cancer123, 8–13 (2008). [DOI] [PubMed] [Google Scholar]
26.Kelly, T. K. et al. Genome-wide mapping of nucleosome positioning and DNA methylation within individual DNA molecules. Genome Res.22, 2497–2506 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Yu, G. et al. Low-dose decitabine enhances the effect of PD-1 blockade in colorectal cancer with microsatellite stability by re-modulating the tumor microenvironment. Cell Mol. Immunol.16, 401–409 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Jinek, M. et al. Structures of Cas9 endonucleases reveal RNA-mediated conformational activation. Science343, 1247997 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
29.Konermann, S. et al. Genome-scale transcriptional activation by an engineered CRISPR-Cas9 complex. Nature517, 583–588 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Ma, H. et al. Multiplexed labeling of genomic loci with dCas9 and engineered sgRNAs using CRISPRainbow. Nat. Biotechnol.34, 528–530 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Di Ruscio, A. et al. DNMT1-interacting RNAs block gene-specific DNA methylation. Nature503, 371–376 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Shim, Y. H., Yoon, G. S., Choi, H. J., Chung, Y. H. & Yu, E. p16 Hypermethylation in the early stage of hepatitis B virus-associated hepatocarcinogenesis. Cancer Lett.190, 213–219 (2003). [DOI] [PubMed] [Google Scholar]
33.Visel, A. et al. Targeted deletion of the 9p21 non-coding coronary artery disease risk interval in mice. Nature464, 409–412 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Li, Q. et al. FOXA1 mediates p16(INK4a) activation during cellular senescence. EMBO J.32, 858–873 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Liu, Y. T. et al. Identification of de novo enhancers activated by TGFbeta to drive expression of CDKN2A and B in HeLa cells. Mol. Cancer Res.17, 1854–1866 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Jeong, M. et al. Large conserved domains of low DNA methylation maintained by Dnmt3a. Nat. Genet.46, 17–23 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Sandoval, J. et al. Validation of a DNA methylation microarray for 450,000 CpG sites in the human genome. Epigenetics6, 692–702 (2011). [DOI] [PubMed] [Google Scholar]
38.Xie, W. et al. Epigenomic analysis of multilineage differentiation of human embryonic stem cells. Cell153, 1134–1148 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Kwon, J. et al. Pseudogene-mediated DNA demethylation leads to oncogene activation. Sci. Adv.7, eabg1695 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
40.Liu, Y.-C. et al. Demethylation and up-regulation of an oncogene after hypomethylating therapy. N. Engl. J. Med.386, 1998–2010 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Wang, Q. et al. Imprecise DNMT1 activity coupled with neighbor-guided correction enables robust yet flexible epigenetic inheritance. Nat. Genet.52, 828–839 (2020). [DOI] [PubMed] [Google Scholar]
42.Brooks, P. J., Marietta, C. & Goldman, D. DNA mismatch repair and DNA methylation in adult brain neurons. J. Neurosci.16, 939–945 (1996). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Ma, D. K. et al. Neuronal activity-induced Gadd45b promotes epigenetic DNA demethylation and adult neurogenesis. Science323, 1074–1077 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Nelson, E. D., Kavalali, E. T. & Monteggia, L. M. Activity-dependent suppression of miniature neurotransmission through the regulation of DNA methylation. J. Neurosci.28, 395–406 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Bae, S. & Lesch, B. J. H3K4me1 distribution predicts transcription state and poising at promoters. Front Cell Dev. Biol.8, 289 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
46.Calo, E. & Wysocka, J. Modification of enhancer chromatin: what, how, and why? Mol. Cell49, 825–837 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.Creyghton, M. P. et al. Histone H3K27ac separates active from poised enhancers and predicts developmental state. Proc. Natl. Acad. Sci. USA107, 21931–21936 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
48.Rada-Iglesias, A. et al. A unique chromatin signature uncovers early developmental enhancers in humans. Nature470, 279–283 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Whalen, S., Truty, R. M. & Pollard, K. S. Enhancer-promoter interactions are encoded by complex genomic signatures on looping chromatin. Nat. Genet.48, 488–496 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Ley, T. J. et al. DNMT3A mutations in acute myeloid leukemia. N. Engl. J. Med.363, 2424–2433 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Steensma, D. P. & Ebert, B. L. Clonal hematopoiesis as a model for premalignant changes during aging. Exp. Hematol.83, 48–56 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Thiede, C. Mutant DNMT3A: teaming up to transform. Blood119, 5615–5617 (2012). [DOI] [PubMed] [Google Scholar]
53.Yang, L., Rau, R. & Goodell, M. A. DNMT3A in haematological malignancies. Nat. Rev. Cancer15, 152–165 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Dimitrov, M. R. et al. Successive DNA extractions improve characterization of soil microbial communities. PeerJ5, e2915 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Ponnaluri, V. K. C. et al. NicE-seq: high resolution open chromatin profiling. Genome Biol.18, 122 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Gonzalez, D. et al. ZNF143 protein is an important regulator of the myeloid transcription factor C/EBPalpha. J. Biol. Chem.292, 18924–18936 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Matelot, M. & Noordermeer, D. Determination of high-resolution 3D chromatin organization using circular chromosome conformation capture (4C-seq). Methods Mol. Biol.1480, 223–241 (2016). [DOI] [PubMed] [Google Scholar]
58.Krijger, P. H. L., Geeven, G., Bianchi, V., Hilvering, C. R. E. & de Laat, W. 4C-seq from beginning to end: A detailed protocol for sample preparation and data analysis. Methods170, 17–32 (2020). [DOI] [PubMed] [Google Scholar]
59.R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2020).
60.pheatmap: Pretty Heatmaps v. R package verion v1.0.12 (2019).
61.Gu, Z. Complex heatmap visualization. Imeta1, e43 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Gu, Z., Eils, R. & Schlesner, M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics32, 2847–2849 (2016). [DOI] [PubMed] [Google Scholar]
63.Suryatenggara, J., Yong, K.J., Tenen, D.E., Tenen, D.G. & Bassal, M.A. ChIP-AP: an integrated analysis pipeline for unbiased ChIP-seq analysis. Brief Bioinform.23, 10.1093/bib/bbab537 (2022). [DOI] [PMC free article] [PubMed]
64.Lawrence, M. et al. Software for computing and annotating genomic ranges. PLoS Comput. Biol.9, e1003118 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Li, L. et al. Msuite2: All-in-one DNA methylation data analysis toolkit with enhanced usability and performance. Comput. Struct. Biotechnol. J.20, 1271–1276 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Condon, D. E. et al. Defiant: (DMRs: easy, fast, identification and ANnoTation) identifies differentially methylated regions from the iron-deficient rat hippocampus. BMC Bioinform.19, 31 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Juhling, F. et al. metilene: fast and sensitive calling of differentially methylated regions from bisulfite sequencing data. Genome Res.26, 256–262 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell38, 576–589 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
69.annotatr: genomic regions in context v. R package version 1.22.0 (2017). [DOI] [PMC free article] [PubMed]
70.AnnotationHub: Client to access AnnotationHub resources v. R package version 3.4.0 (2022).
71.ggplot2: Elegant Graphics for Data Analysis (2016).
72.Conway, J. R., Lex, A. & Gehlenborg, N. UpSetR: an R package for the visualization of intersecting sets and their properties. Bioinformatics33, 2938–2940 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
73.Sushi: Tools for visualizing genomics data v. R package version 1.34.0 (2022).
74.BBMap: A Fast, Accurate, Splice-Aware Aligner (USA, 2014).
75.Bolger, A. M., Lohse, M. & Usadel, B. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics30, 2114–2120 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
76.Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics29, 15–21 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
77.Li, H. et al. The sequence alignment/map format and SAMtools. Bioinformatics25, 2078–2079 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
78.Ramirez, F. et al. deepTools2: a next-generation web server for deep-sequencing data analysis. Nucleic Acids Res.44, W160–W165 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
79.Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general-purpose program for assigning sequence reads to genomic features. Bioinformatics30, 923–930 (2014). [DOI] [PubMed] [Google Scholar]
80.Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics26, 841–842 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
81.Gu, Z., Gu, L., Eils, R., Schlesner, M. & Brors, B. circlize implements and enhances circular visualization in R. Bioinformatics30, 2811–2812 (2014). [DOI] [PubMed] [Google Scholar]
82.Wei, T. & Simko, V. R Package ‘Corrplot’: Visualization of a Correlation Matrix (2024).
83.Wong, S. & McGraw, K. A common language effect size statistic. Psychol. Bull.111, 361–365 (1992). [Google Scholar]
84.Wickham, H. et al. Welcome to the Tidyverse. J. Open Source Softw.4, 1686 (2019). [Google Scholar]
85.Kassambara, A. rstatix: Pipe-Friendly Framework for Basic Statistical Tests (2023).
86.Kassambara, A. ggpubr: ‘ggplot2’ Based Publication Ready Plots (2026).
87.Clarke, E., Sherrill-Mix, S. & Dawson, C. ggbeeswarm: categorical scatter (Violin Point) plots (2025).
88.Zeileis, A. & Grothendieck, G. zoo: S3 infrastructure for regular and irregular time series. J. Stat. Softw.14, 1–27 (2005). [Google Scholar]
89.An, O. et al. CSI NGS portal: an online platform for automated NGS data analysis and sharing. Int. J. Mol. Sci.21, 10.3390/ijms21113828 (2020). [DOI] [PMC free article] [PubMed]
90.Li, H. & Durbin, R. Fast and accurate long-read alignment with Burrows-Wheeler transform. Bioinformatics26, 589–595 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
91.Thongjuea, S., Stadhouders, R., Grosveld, F. G., Soler, E. & Lenhard, B. r3Cseq: an R/Bioconductor package for the discovery of long-range genomic interactions from chromosome conformation capture and next-generation sequencing data. Nucleic Acids Res.41, e132 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
92.BSgenome.Hsapiens.UCSC.hg19.masked: full masked genome sequences for Homo sapiens (UCSC version hg19, based on GRCh37.p13). (2020).

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Information^{(19.2MB, pdf)}

41467_2026_76648_MOESM2_ESM.pdf^{(118.4KB, pdf)}

Description of Additional Supplementary Files

Supplementary Data 1^{(42.4KB, xlsx)}

Supplementary Data 2^{(28.5KB, xls)}

Supplementary Data 3^{(12.2KB, xlsx)}

Supplementary Data 4^{(10.6KB, xlsx)}

Supplementary Data 5^{(11.1KB, xlsx)}

Supplementary Data 6^{(10.4KB, xlsx)}

Supplementary Data 7^{(9.8KB, xlsx)}

Supplementary Data 8^{(9.5KB, xlsx)}

Supplementary Data 9^{(11.2KB, xlsx)}

Supplementary Data 10^{(9.7KB, xlsx)}

Supplementary Data 11^{(12.6KB, xlsx)}

Supplementary Data 12^{(19.6KB, xlsx)}

Supplementary Data 13^{(39KB, xlsx)}

Reporting Summary^{(104.7KB, pdf)}

Transparent Peer Review file^{(3.3MB, pdf)}

Source data^{(20.2MB, xlsx)}

Data Availability Statement

All publicly available data processed in this study are listed in Supplementary Data 13. Source data are provided with this paper.

[CR1] 1.Jones, P. A. Functions of DNA methylation: islands, start sites, gene bodies and beyond. Nat. Rev. Genet.13, 484–492 (2012). [DOI] [PubMed] [Google Scholar]

[CR2] 2.Esteller, M., Corn, P. G., Baylin, S. B. & Herman, J. G. A gene hypermethylation profile of human cancer. Cancer Res.61, 3225–3229 (2001). [PubMed] [Google Scholar]

[CR3] 3.Herman, J. G. & Baylin, S. B. Gene silencing in cancer in association with promoter hypermethylation. N. Engl. J. Med.349, 2042–2054 (2003). [DOI] [PubMed] [Google Scholar]

[CR4] 4.Feinberg, A. P. & Tycko, B. The history of cancer epigenetics. Nat. Rev. Cancer4, 143–153 (2004). [DOI] [PubMed] [Google Scholar]

[CR5] 5.Gardiner-Garden, M. & Frommer, M. CpG islands in vertebrate genomes. J. Mol. Biol.196, 261–282 (1987). [DOI] [PubMed] [Google Scholar]

[CR6] 6.Larsen, F., Gundersen, G., Lopez, R. & Prydz, H. CpG islands as gene markers in the human genome. Genomics13, 1095–1107 (1992). [DOI] [PubMed] [Google Scholar]

[CR7] 7.Takai, D. & Jones, P. A. Comprehensive analysis of CpG islands in human chromosomes 21 and 22. Proc. Natl. Acad. Sci. USA99, 3740–3745 (2002). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR8] 8.Wang, Y. & Leung, F. C. An evaluation of new criteria for CpG islands in the human genome as gene markers. Bioinformatics20, 1170–1177 (2004). [DOI] [PubMed] [Google Scholar]

[CR9] 9.Saxonov, S., Berg, P. & Brutlag, D. L. A genome-wide analysis of CpG dinucleotides in the human genome distinguishes two distinct classes of promoters. Proc. Natl. Acad. Sci. USA103, 1412–1417 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Nunez, J. K. et al. Genome-wide programmable transcriptional memory by CRISPR-based epigenome editing. Cell184, 2503–2519 e2517 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR11] 11.Han, H. et al. DNA methylation directly silences genes with non-CpG island promoters and establishes a nucleosome-occupied promoter. Hum. Mol. Genet.20, 4299–4310 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Koch, A. et al. Analysis of DNA methylation in cancer: location revisited. Nat. Rev. Clin. Oncol.15, 459–466 (2018). [DOI] [PubMed] [Google Scholar]

[CR13] 13.van Vlodrop, I. J. et al. Analysis of promoter CpG island hypermethylation in cancer: location, location, location!. Clin. Cancer Res.17, 4225–4231 (2011). [DOI] [PubMed] [Google Scholar]

[CR14] 14.Hodges, E. et al. Directional DNA methylation changes and complex intermediate states accompany lineage specificity in the adult hematopoietic compartment. Mol. Cell44, 17–28 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR15] 15.Irizarry, R. A. et al. The human colon cancer methylome shows similar hypo- and hypermethylation at conserved tissue-specific CpG island shores. Nat. Genet.41, 178–186 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR16] 16.Zinn, R. L., Pruitt, K., Eguchi, S., Baylin, S. B. & Herman, J. G. hTERT is expressed in cancer cell lines despite promoter DNA methylation by preservation of unmethylated DNA and active chromatin around the transcription start site. Cancer Res.67, 194–201 (2007). [DOI] [PubMed] [Google Scholar]

[CR17] 17.van Vlodrop, I. J. et al. Prognostic significance of Gremlin1 (GREM1) promoter CpG island hypermethylation in clear cell renal cell carcinoma. Am. J. Pathol.176, 575–584 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Li, G. et al. Extensive promoter-centered chromatin interactions provide a topological basis for transcription regulation. Cell148, 84–98 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR19] 19.Mifsud, B. et al. Mapping long-range promoter contacts in human cells with high-resolution capture Hi-C. Nat. Genet.47, 598–606 (2015). [DOI] [PubMed] [Google Scholar]

[CR20] 20.Sanyal, A., Lajoie, B. R., Jain, G. & Dekker, J. The long-range interaction landscape of gene promoters. Nature489, 109–113 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR21] 21.Fulco, C. P. et al. Systematic mapping of functional enhancer-promoter connections with CRISPR interference. Science354, 769–773 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR22] 22.Xu, X. et al. A CRISPR-based approach for targeted DNA demethylation. Cell Discov.2, 16009 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR23] 23.Morita, S. et al. Targeted DNA demethylation in vivo using dCas9-peptide repeat and scFv-TET1 catalytic domain fusions. Nat. Biotechnol.34, 1060–1065 (2016). [DOI] [PubMed] [Google Scholar]

[CR24] 24.Derissen, E. J., Beijnen, J. H. & Schellens, J. H. Concise drug review: azacitidine and decitabine. Oncologist18, 619–624 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR25] 25.Stresemann, C. & Lyko, F. Modes of action of the DNA methyltransferase inhibitors azacytidine and decitabine. Int. J. Cancer123, 8–13 (2008). [DOI] [PubMed] [Google Scholar]

[CR26] 26.Kelly, T. K. et al. Genome-wide mapping of nucleosome positioning and DNA methylation within individual DNA molecules. Genome Res.22, 2497–2506 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Yu, G. et al. Low-dose decitabine enhances the effect of PD-1 blockade in colorectal cancer with microsatellite stability by re-modulating the tumor microenvironment. Cell Mol. Immunol.16, 401–409 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.Jinek, M. et al. Structures of Cas9 endonucleases reveal RNA-mediated conformational activation. Science343, 1247997 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR29] 29.Konermann, S. et al. Genome-scale transcriptional activation by an engineered CRISPR-Cas9 complex. Nature517, 583–588 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR30] 30.Ma, H. et al. Multiplexed labeling of genomic loci with dCas9 and engineered sgRNAs using CRISPRainbow. Nat. Biotechnol.34, 528–530 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR31] 31.Di Ruscio, A. et al. DNMT1-interacting RNAs block gene-specific DNA methylation. Nature503, 371–376 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR32] 32.Shim, Y. H., Yoon, G. S., Choi, H. J., Chung, Y. H. & Yu, E. p16 Hypermethylation in the early stage of hepatitis B virus-associated hepatocarcinogenesis. Cancer Lett.190, 213–219 (2003). [DOI] [PubMed] [Google Scholar]

[CR33] 33.Visel, A. et al. Targeted deletion of the 9p21 non-coding coronary artery disease risk interval in mice. Nature464, 409–412 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR34] 34.Li, Q. et al. FOXA1 mediates p16(INK4a) activation during cellular senescence. EMBO J.32, 858–873 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR35] 35.Liu, Y. T. et al. Identification of de novo enhancers activated by TGFbeta to drive expression of CDKN2A and B in HeLa cells. Mol. Cancer Res.17, 1854–1866 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR36] 36.Jeong, M. et al. Large conserved domains of low DNA methylation maintained by Dnmt3a. Nat. Genet.46, 17–23 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR37] 37.Sandoval, J. et al. Validation of a DNA methylation microarray for 450,000 CpG sites in the human genome. Epigenetics6, 692–702 (2011). [DOI] [PubMed] [Google Scholar]

[CR38] 38.Xie, W. et al. Epigenomic analysis of multilineage differentiation of human embryonic stem cells. Cell153, 1134–1148 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR39] 39.Kwon, J. et al. Pseudogene-mediated DNA demethylation leads to oncogene activation. Sci. Adv.7, eabg1695 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR40] 40.Liu, Y.-C. et al. Demethylation and up-regulation of an oncogene after hypomethylating therapy. N. Engl. J. Med.386, 1998–2010 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR41] 41.Wang, Q. et al. Imprecise DNMT1 activity coupled with neighbor-guided correction enables robust yet flexible epigenetic inheritance. Nat. Genet.52, 828–839 (2020). [DOI] [PubMed] [Google Scholar]

[CR42] 42.Brooks, P. J., Marietta, C. & Goldman, D. DNA mismatch repair and DNA methylation in adult brain neurons. J. Neurosci.16, 939–945 (1996). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR43] 43.Ma, D. K. et al. Neuronal activity-induced Gadd45b promotes epigenetic DNA demethylation and adult neurogenesis. Science323, 1074–1077 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR44] 44.Nelson, E. D., Kavalali, E. T. & Monteggia, L. M. Activity-dependent suppression of miniature neurotransmission through the regulation of DNA methylation. J. Neurosci.28, 395–406 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR45] 45.Bae, S. & Lesch, B. J. H3K4me1 distribution predicts transcription state and poising at promoters. Front Cell Dev. Biol.8, 289 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR46] 46.Calo, E. & Wysocka, J. Modification of enhancer chromatin: what, how, and why? Mol. Cell49, 825–837 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR47] 47.Creyghton, M. P. et al. Histone H3K27ac separates active from poised enhancers and predicts developmental state. Proc. Natl. Acad. Sci. USA107, 21931–21936 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR48] 48.Rada-Iglesias, A. et al. A unique chromatin signature uncovers early developmental enhancers in humans. Nature470, 279–283 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR49] 49.Whalen, S., Truty, R. M. & Pollard, K. S. Enhancer-promoter interactions are encoded by complex genomic signatures on looping chromatin. Nat. Genet.48, 488–496 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR50] 50.Ley, T. J. et al. DNMT3A mutations in acute myeloid leukemia. N. Engl. J. Med.363, 2424–2433 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR51] 51.Steensma, D. P. & Ebert, B. L. Clonal hematopoiesis as a model for premalignant changes during aging. Exp. Hematol.83, 48–56 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR52] 52.Thiede, C. Mutant DNMT3A: teaming up to transform. Blood119, 5615–5617 (2012). [DOI] [PubMed] [Google Scholar]

[CR53] 53.Yang, L., Rau, R. & Goodell, M. A. DNMT3A in haematological malignancies. Nat. Rev. Cancer15, 152–165 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR54] 54.Dimitrov, M. R. et al. Successive DNA extractions improve characterization of soil microbial communities. PeerJ5, e2915 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR55] 55.Ponnaluri, V. K. C. et al. NicE-seq: high resolution open chromatin profiling. Genome Biol.18, 122 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR56] 56.Gonzalez, D. et al. ZNF143 protein is an important regulator of the myeloid transcription factor C/EBPalpha. J. Biol. Chem.292, 18924–18936 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR57] 57.Matelot, M. & Noordermeer, D. Determination of high-resolution 3D chromatin organization using circular chromosome conformation capture (4C-seq). Methods Mol. Biol.1480, 223–241 (2016). [DOI] [PubMed] [Google Scholar]

[CR58] 58.Krijger, P. H. L., Geeven, G., Bianchi, V., Hilvering, C. R. E. & de Laat, W. 4C-seq from beginning to end: A detailed protocol for sample preparation and data analysis. Methods170, 17–32 (2020). [DOI] [PubMed] [Google Scholar]

[CR59] 59.R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2020).

[CR60] 60.pheatmap: Pretty Heatmaps v. R package verion v1.0.12 (2019).

[CR61] 61.Gu, Z. Complex heatmap visualization. Imeta1, e43 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR62] 62.Gu, Z., Eils, R. & Schlesner, M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics32, 2847–2849 (2016). [DOI] [PubMed] [Google Scholar]

[CR63] 63.Suryatenggara, J., Yong, K.J., Tenen, D.E., Tenen, D.G. & Bassal, M.A. ChIP-AP: an integrated analysis pipeline for unbiased ChIP-seq analysis. Brief Bioinform.23, 10.1093/bib/bbab537 (2022). [DOI] [PMC free article] [PubMed]

[CR64] 64.Lawrence, M. et al. Software for computing and annotating genomic ranges. PLoS Comput. Biol.9, e1003118 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR65] 65.Li, L. et al. Msuite2: All-in-one DNA methylation data analysis toolkit with enhanced usability and performance. Comput. Struct. Biotechnol. J.20, 1271–1276 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR66] 66.Condon, D. E. et al. Defiant: (DMRs: easy, fast, identification and ANnoTation) identifies differentially methylated regions from the iron-deficient rat hippocampus. BMC Bioinform.19, 31 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR67] 67.Juhling, F. et al. metilene: fast and sensitive calling of differentially methylated regions from bisulfite sequencing data. Genome Res.26, 256–262 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR68] 68.Heinz, S. et al. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell38, 576–589 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR69] 69.annotatr: genomic regions in context v. R package version 1.22.0 (2017). [DOI] [PMC free article] [PubMed]

[CR70] 70.AnnotationHub: Client to access AnnotationHub resources v. R package version 3.4.0 (2022).

[CR71] 71.ggplot2: Elegant Graphics for Data Analysis (2016).

[CR72] 72.Conway, J. R., Lex, A. & Gehlenborg, N. UpSetR: an R package for the visualization of intersecting sets and their properties. Bioinformatics33, 2938–2940 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR73] 73.Sushi: Tools for visualizing genomics data v. R package version 1.34.0 (2022).

[CR74] 74.BBMap: A Fast, Accurate, Splice-Aware Aligner (USA, 2014).

[CR75] 75.Bolger, A. M., Lohse, M. & Usadel, B. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics30, 2114–2120 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR76] 76.Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics29, 15–21 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR77] 77.Li, H. et al. The sequence alignment/map format and SAMtools. Bioinformatics25, 2078–2079 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR78] 78.Ramirez, F. et al. deepTools2: a next-generation web server for deep-sequencing data analysis. Nucleic Acids Res.44, W160–W165 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR79] 79.Liao, Y., Smyth, G. K. & Shi, W. featureCounts: an efficient general-purpose program for assigning sequence reads to genomic features. Bioinformatics30, 923–930 (2014). [DOI] [PubMed] [Google Scholar]

[CR80] 80.Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics26, 841–842 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR81] 81.Gu, Z., Gu, L., Eils, R., Schlesner, M. & Brors, B. circlize implements and enhances circular visualization in R. Bioinformatics30, 2811–2812 (2014). [DOI] [PubMed] [Google Scholar]

[CR82] 82.Wei, T. & Simko, V. R Package ‘Corrplot’: Visualization of a Correlation Matrix (2024).

[CR83] 83.Wong, S. & McGraw, K. A common language effect size statistic. Psychol. Bull.111, 361–365 (1992). [Google Scholar]

[CR84] 84.Wickham, H. et al. Welcome to the Tidyverse. J. Open Source Softw.4, 1686 (2019). [Google Scholar]

[CR85] 85.Kassambara, A. rstatix: Pipe-Friendly Framework for Basic Statistical Tests (2023).

[CR86] 86.Kassambara, A. ggpubr: ‘ggplot2’ Based Publication Ready Plots (2026).

[CR87] 87.Clarke, E., Sherrill-Mix, S. & Dawson, C. ggbeeswarm: categorical scatter (Violin Point) plots (2025).

[CR88] 88.Zeileis, A. & Grothendieck, G. zoo: S3 infrastructure for regular and irregular time series. J. Stat. Softw.14, 1–27 (2005). [Google Scholar]

[CR89] 89.An, O. et al. CSI NGS portal: an online platform for automated NGS data analysis and sharing. Int. J. Mol. Sci.21, 10.3390/ijms21113828 (2020). [DOI] [PMC free article] [PubMed]

[CR90] 90.Li, H. & Durbin, R. Fast and accurate long-read alignment with Burrows-Wheeler transform. Bioinformatics26, 589–595 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR91] 91.Thongjuea, S., Stadhouders, R., Grosveld, F. G., Soler, E. & Lenhard, B. r3Cseq: an R/Bioconductor package for the discovery of long-range genomic interactions from chromosome conformation capture and next-generation sequencing data. Nucleic Acids Res.41, e132 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR92] 92.BSgenome.Hsapiens.UCSC.hg19.masked: full masked genome sequences for Homo sapiens (UCSC version hg19, based on GRCh37.p13). (2020).

PERMALINK

Methylation Mesa define functional regulatory elements for targeted gene activation

Y V Liu

J Suryatenggara

H Wong

M K Jayasinghe

J P Tang

H K Tan

J Kwon

Q Zhou

J Zhang

S Moein

S Ummarino

M T N Le

J G Doench

T Benoukraf

L Chai

D Hiwase

D Thomas

A Di Ruscio

D G Tenen

M A Bassal

Abstract

Introduction

Results

A distinct and narrow-width Mesa signature is consistent across sample types and species

Fig. 1. Characterization of Methylation Mesa signatures following hypomethylating agent (HMA) induced demethylation.

Mesa enrichment in 5′UTR better associate with gene activation than upstream promoters

Fig. 2. Genomic distribution of Methylation Mesa and its superior correlation with transcriptional activation.

Mesa are extensively marked by primed and active histone marks and show positional conservation

Fig. 3. Regulatory potential and spatial consistency of Methylation Mesa (MM).

Development of CRISPR-DiR for precise demethylation to investigate MM causal regulatory dynamics

Fig. 4. Development of CRISPR-DiR to validate causal regulatory elements for gene activation.

Tracing the epigenetic kinetics triggered by p16 MM demethylation

Fig. 5. Kinetic tracing of CRISPR-DiR mediated epigenetic rewiring and expanded in vivo functional validation.

Mesa demethylation and gene re-expression initiate long-range interaction with distal Mesa loci

Fig. 6. CRISPR-DiR-induced local demethylation in p16 initiates strong interactions with distal elements marked as MM.

Discussion

Methods

Ethics compliance

Cell lines and patient samples

RNA isolation

Genomic DNA extraction

5-aza-2’-deoxycytidine (Decitabine; DAC) treatment

5-azacytidine (Azacytidine; AZA) treatment

Deoxycytidine (Dox) treatment

Transient transfections

Lentivirus production

Stable cell line generation

Quantitative real-time PCR (qRT-PCR)

Antibody validation

Western blot analysis

Bisulfite treatment

Methylation-specific PCR (MSP), combined bisulfite restriction analysis (COBRA) and bisulfite sequencing PCR (BSP)

Whole genomic bisulfite sequencing (WGBS) and RNA-Seq

Xenograft murine model

Chromatin immunoprecipitation (ChIP)

Circularized chromosome conformation capture (4C)-Seq

Statistical analysis

Bioinformatic analysis

Histone marks ChIP-Seq Analysis

Human settings table (ChIP-AP)

Table 1.

Mouse settings table (ChIP-AP)

Table 2.

Publicly available datasets

WGBS and MM analysis

RNA-Seq analysis

Permutation testing

Permutation testing for selection bias

Common language effect size (CLES) calculations

Methylation-expression association analysis

Mesa/CGI signal intensity quantification

4C-Seq analysis

Software versions utilized

Reporting summary

Supplementary information

Source data

Acknowledgements

Author contributions