Somatic epimutations cap genetic determinism in the human diploid chromatin epigenome

Summary

Diploid human cells contain two non-identical genomes, and differences in their regulation underlie human development and disease. We present Fiber-seq Inferred Regulatory Elements (FIRE) and show that FIRE provides a more comprehensive and quantitative snapshot of the accessible chromatin landscape across the 6 Gbp diploid human genome, overcoming previously unrecognized biases in existing regulatory element catalogs. FIRE enables comprehensive detection of haplotype-selective chromatin accessibility (HSCA), exposing imprinted elements lacking underlying parent-of-origin CpG methylation differences, and gene regulatory modules that permit genes to escape X chromosome inactivation. We uncover that the human leukocyte antigen (HLA) locus harbors the most HSCA in immune cells, where we resolve specific transcription factor (TF) binding events disrupted by disease-associated variants. Finally, we demonstrate that the regulatory landscape of a cell is littered with autosomal somatic chromatin epimutations that are propagated by clonal expansions to create mitotically stable and non-genetically deterministic chromatin alterations.

Graphical Abstract

In Brief:

FIRE is a long-read Fiber-seq method that resolves chromatin accessibility along individual DNA molecules on each parental chromosome, enabling a comprehensive assessment of the diploid epigenome. FIRE reveals that the HLA locus harbors the densest haplotype-selective chromatin in immune cells and resolves regulatory principles of escape from X inactivation. It also uncovers pervasive somatic epimutations, mitotically stable and non-genetically mediated chromatin changes that cap genetic determinism in the human epigenome.

Introduction

Advances in long-read sequencing^1,2 have enabled the routine de novo assembly of diploid human genomes^3,4, resulting in the completion of the first human genome^5–7 and pangenome^8–10. These advances have improved our understanding of genome structure and genetic variation, enabling researchers to investigate genetic variants within their native haplotype context along the 6 Gbp diploid human genome. Our next challenge is studying the function of this 6 Gbp genome, as this can shed light on how non-coding genetic variants contribute to human disease risk, uncover somatic epimutations that selectively arise on only a single haplotype, and illuminate gene regulatory patterns in genetically complex regions of the genome that show the greatest sequence diversity between humans.

Despite the potential for studying the function of genetic variation along the diploid genome, most chromatin assays are reliant on collapsed 3 Gbp representations of a human genome. Specifically, given the inconsistent density of genetic variation within the human genome, short-read-based chromatin methods are poorly suited for uniquely phasing chromatin across the diploid genome, and even when an element happens to overlap a heterozygous single-nucleotide variant (SNV), read-mapping artifacts are known to confound measures of allele-specific chromatin accessibility.

Long-read single-molecule chromatin assays^11–14 have overcome these challenges by leveraging the ability of long reads to correctly map to the diploid genome^12,15,16. For example, Fiber-seq uses a non-specific N⁶-adenine methyltransferase (m6A-MTase) to stencil protein occupancy footprints along DNA molecules in the form of m6A-modified bases (Fig. 1A)¹¹. These m6A-modified DNA molecules are sequenced using a single-molecule platform¹⁷, enabling the synchronous readout of the genetic sequence, CpG methylation status, and chromatin architecture. Prior work has used this data to accurately identify the presence of transcription factors¹⁸; however, despite the promise of single-molecule chromatin assays, current approaches for analyzing this new data type are sensitive to potential experimental batch effects, and are unable to identify putative regulatory elements de novo, resulting in a dearth of new knowledge about how gene regulation is structured across the entire 6 Gbp genome, and how this structure changes across different tissues and disease states. To realize the full potential of long-read single-molecule chromatin assays, we developed a robust machine learning classifier that enables the precise delineation of chromatin architectures across the diploid genome with single-molecule and single-haplotype precision at near-nucleotide resolution. Furthermore, we leverage this approach across multiple cohorts to uncover the basic principles guiding diploid human gene regulation, including the relative contributions of rare and common genetic variants, imprinting, somatic epimutations, and X chromosome inactivation.

Figure 1. Fiber-seq Inferred Regulatory Elements and benchmarking against existing chromatin accessibility measures.

A) A schematic of Fiber-seq experimental and computational processing, including Fiber-seq Inferred Regulatory Elements (FIREs).

B) Genomic locus in GM12878 showing scATAC-seq, DNase-seq, mCpG, FIRE percent actuation, and FIRE peaks. Below are individual Fiber-seq reads (rows) with MSPs (purple), nucleosomes (gray), and FIRE elements (red).

C) Correlation of FIRE score within the peaks of two biological replicates of COLO829BL (two-sided t-test).

D) Correlation of FIRE score with bulk DNase-seq in K562 accessible peaks (two-sided t-test).

E) Venn diagram showing the overlap of FIRE and scATAC peaks in GM12878.

F) (Left) Average count of DNase I reads over FIRE peaks binned by their percent actuation (red-blue color scale). (Right) Percentile-normalized scATAC and DNase I signal for 100 random FIRE peaks across each percent actuation bin.

G) Percent actuation by Fiber-seq versus scATAC-seq, where scATAC-seq accessibility is the fraction of cells with at least one fragment overlapping the peak. Peaks are binned by Fiber-seq actuation.

Results

Single-molecule Fiber-seq inferred regulatory elements

To generate a haplotype-resolved map of human gene regulation, we first sought to develop a machine learning classifier for Fiber-seq that could accurately classify m6A-modified regions (i.e., MTase Sensitive Patches [MSPs]) as actuated single-molecule regulatory elements (Fig. 1A). Importantly, accurately measuring chromatin actuation de novo at the level of single fibers permits the quantitative measurement of chromatin accessibility at a regulatory element as the percent of molecules with chromatin actuation, a potentially more intuitive and quantitative metric of chromatin than is currently possible using traditional chromatin assays. Although existing single-molecule stenciling methods have used MSP size as a simple single-molecule classifier for this process^11,12, this simple classifier has relatively poor performance for the de novo identification of putative regulatory elements genome-wide (Fig. S1D,E), likely owing to its inability to disentangle stochastic, unstable nucleosome occupancy events from actuated regulatory elements. Furthermore, we have found that MSP size is highly sensitive to technical sample-to-sample variation in the total rate of adenine methylation (Fig. S1A,B), which can result in systematic biases in identifying accessible chromatin using single-molecule stenciling methods¹⁹ (Fig. S1C). To resolve this challenge limiting the broader adoption of single-molecule chromatin stenciling methods, we develop a tool for the accurate de novo classification of putative regulatory elements using only Fiber-seq data that does not need any prior knowledge of the underlying reference genome, or the exact primary sequence of the underlying fiber. Importantly, this structure enables this tool to generalize across different cell types that employ unique sets of binding elements to regulate their accessible chromatin^15,16,20,21, complex genomic regions not present within the reference GRCh38^22,23, as well as organisms that have completely distinct genomic architectures²⁴.

To create our training data, we generated multiple Fiber-seq datasets from the reference cell line GM12878 (totaling 136-fold coverage) that differed by over 2.3-fold in their global m6A methylation rate, enabling the final model to extend to analytical fluctuations in the Fiber-seq method (Fig. S1A–C). Although our final tool works irrespective of a reference genome, labels for the training data relied on GRCh38, as this enabled us to use the extensive short-read epigenetic data from GM12878 to assign positive and negative labels to individual MSPs (Methods, Fig. S2, Supplemental Table 1). Notably, regulatory elements identified using bulk methods can exhibit marked heterogeneity in their single-molecule actuation pattern (i.e., can be an actuated element, or merely an internucleosomal linker region at the single-molecule level)^11,12,14 (Fig. S3). As such, MSPs ascribed a positive label based on their genomic location relative to bulk chromatin measurements likely represent a mixture of MSPs arising from actuated regulatory elements and others arising from internucleosomal linker regions. Our training data therefore consist of mixed-positive and clean-negative labels, which we addressed with a semi-supervised classifier adapted from proteomics^25–27 that assigns each MSP an estimated precision for containing a regulatory element (Methods, Fig. S2A–C). We then used the estimated precision to classify 1,959,888,668 MSPs across 24,771,424 chromatin Fiber-seq reads as either Fiber-seq Inferred Regulatory Elements (FIREs; 95% estimated precision or greater; n=32,006,894) or internucleosomal linker regions. Consistent with the model training, MSPs classified as FIREs were markedly enriched in DNase I hypersensitive sites (DHSs) compared to using MSP length alone as a classifier (Fig. S1D).

Aggregating across Fiber-seq inferred regulatory elements.

As no other methods exist for classifying per-molecule chromatin architectures, to benchmark our FIRE classifications, we developed a method for the de novo identification of putative regulatory elements genome-wide using FIREs (Fig. 1B). To accomplish this, we first created a coverage-normalized aggregate FIRE score (Methods) combining molecules covering each base along the genome. These aggregate FIRE scores were calibrated using a null aggregate FIRE score distribution, enabling the calculation of false discovery rates (FDR, Methods, Fig. S4A,B), and identifying genomic positions at the 5% FDR threshold as FIRE peaks (Fig. 1A,B). At each FIRE peak, the “percent actuation” was calculated as the percentage of overlapping Fiber-seq reads with a FIRE. In total, we identified 204,965 FIRE peaks in GM12878, which were robust to underlying changes in our model (Fig. S5), and overlapped 89% of the previously annotated GM12878 DNaseI hypersensitive sites (Supplemental Table 2). Notably, this approach provides an intuitive, biologically meaningful metric of chromatin accessibility and peak width, with the boundaries of each FIRE peak corresponding to the nucleotide-precise median start/stop positions of the underlying single-molecule data, and the intensity reflecting the exact fraction of chromatin fibers with actuation.

Benchmarking FIRE single-molecule measures of chromatin actuation

To test the generalizability of our approach, we generated two biological replicates of Fiber-seq from a different human cell line (COLO829BL) (totaling 319-fold coverage), which demonstrated that we can readily extend our model to other samples and that FIRE scores are highly concordant between replicates (R=0.97) (Fig. 1C) even in the setting of experimental differences in the overall methylation rate between replicates (3.81% vs. 4.49% in replicate 1 versus replicate 2). Furthermore, we showed that FIRE scores were stable after down-sampling our GM12878 data from 135- to 30-fold coverage (Fig. S4C,D). In addition, we generated Fiber-seq data from different cell types (K562, A549, Jurkat, and THP-1 cells) and demonstrated that overall FIRE measures of chromatin actuation are strongly correlated with those obtained using DNase-seq (Figs. 1D, S6A,B, S7A–C). We also benchmarked our FIRE peaks and chromatin actuation measures against single-cell ATAC-seq (scATAC-seq) data, as this data type was not used in our training. Specifically, an entire 26,910-cell ENCODE GM12878 scATAC-seq dataset was pseudo-bulked and peaks called using MACS2. Overall, accessible chromatin peaks identified using Fiber-seq and scATAC-seq were concordant, with 77% of FIRE peaks also called by scATAC-seq (Fig. 1E; Fig. S6A,B). Furthermore, scATAC-seq signal supported even lowly actuated Fiber-seq peaks (10–15%) (Fig. 1F) and scATAC signal increased at peaks with higher Fiber-seq percent actuation (Fig. 1F,G), indicating that Fiber-seq percent actuation scores are in agreement with short-read measures of chromatin accessibility.

In total, 22.2% of the FIRE peaks were unique to FIRE and not called using either the scATAC-seq or DNase-seq data (i.e., FIRE-specific peaks). These FIRE-specific peaks were on average narrower in terms of their chromatin accessibility (Fig. 2A,B), did not show major changes in GC content (Figs. S7D and S8), and were significantly enriched in REST ChIP-seq peaks that demonstrated single-molecule TF occupancy patterns at these REST sites (Fig. S7E–F). To validate the biological relevance of these FIRE-specific peaks, we overlapped all FIRE peaks with genetic variants in GM12878 cells that show genome-wide significant associations (GWAS) with different human traits and diseases²⁹. We observed that FIRE peaks were significantly enriched for overlapping disease-associated GWAS variants, as expected (p-value = 8.40e-71, two-sided Fisher’s exact test)³⁰. However, unlike scATAC-seq-specific or DNase-specific peaks, FIRE-specific peaks also showed a comparable enrichment for disease-associated GWAS variants (Fig. 2C, p-value = 1.32e-17, two-sided Fisher’s exact test), showing that these FIRE-specific peaks are of similar high quality as Fiber-seq peaks with ATAC-seq and DNase-seq support. These findings demonstrate that FIRE enables the accurate and robust de novo identification of accessible chromatin elements.

Figure 2. Chromatin features within FIRE-specific peaks.

A) Features of FIRE peaks binned by percent actuation.

B) Genomic locus in GM12878 as in Fig. 1B, with CTCF ChIP-seq added. Green highlight (right), a representative CTCF site with greater accessibility in Fiber-seq than scATAC-seq or DNase-seq.

C) Per-base enrichment of GWAS variants in peaks shared between FIRE and DNase/scATAC, FIRE-only peaks, and DNase- or scATAC-only peaks, versus size-matched shuffled windows (p-value = 1.32e-17, two-sided Fisher’s exact test).

D) Correlation of FIRE percent actuation and scATAC-seq signal within FIRE peaks faceted by FIRE peak size (Pearson’s correlation; p < 2.2e–16 two-sided t-test).

E) Prediction of percent FIRE actuation from DNase peak signal using a linear model for different bins of FIRE peak size.

F) Comparison of scATAC-seq signal to FIRE in peaks with (green) and without (red-blue) CTCF ChIP-seq peaks.

G) Genomic locus of NOTCH2NLB comparing the ability to map into repetitive regions between scATAC-seq, DNase-seq, and Fiber-seq.

H) FIRE peaks within segmental duplications, stratified by the sequence identity of the underlying duplication. Peaks shared with scATAC-seq are gray; peaks unique to FIRE are red.

FIRE provides more accurate measures of chromatin accessibility

We next investigated differences between FIRE measures of chromatin accessibility and those of existing short-read chromatin assays. First, we observed that controlling for GC content improved the correlation between short-read chromatin assays and Fiber-seq, likely reflecting GC biases that are induced during the PCR amplification steps of these short-read assays (Fig. S4E). Second, we observed that 78.4% of the GM12878 peaks identified only in Fiber-seq were <200 bp in length, raising the possibility that short-read chromatin assays may be selectively biased against detecting accessibility at shorter regulatory elements (Fig. 2D, S7D). Consistent with this, we observed that although short-read measures of chromatin accessibility are strongly associated with Fiber-seq percent actuation for elements >250 bp in length, this association markedly deteriorates for elements <200 bp in length (Fig. 2D,E), which are enriched in CTCF ChIP-seq peaks (26.5% of all peaks ≤200 bp). This potential size-dependent bias also appeared to extend to scATAC-seq measures of chromatin accessibility. Specifically, elements detected in 10–15% of cells as measured by scATAC-seq can have Fiber-seq actuation ranging from <10% to >90% of fibers (Fig. 1F), with scATAC-seq signal being 3.47-fold lower at CTCF ChIP-seq peaks compared to other scATAC-peaks with a similar Fiber-seq percent actuation (Fig. 2F). These findings suggest that existing short-read maps of chromatin accessibility are reporting inaccurate chromatin accessibility measurements at CTCF binding elements, and consistent with this, we observed that FIRE actuation at a CTCF ChIP-seq peak was markedly better at predicting CTCF ChIP-seq signal at that site than ATAC-seq signal (R² of 0.41 vs. 0.17, respectively) (Fig. S9A). Overall, this size-dependent bias resulted in the width of a peak being more predictive of ATAC-seq Tn5 insertions at a site than the actual number of chromatin fibers on which that peak is actuated (Fig. 2F, S7C). Although the exact cause of this phenomenon is unknown, it is likely driven by a combination of artifacts inherent to short-read chromatin methods, including: (1) available template region for Tn5 transposition/DNaseI nicking; (2) fragment size selection; (3) PCR amplification preferences; and (4) short-read sequencing biases.

FIRE provides a more complete genome-wide map of chromatin accessibility

We found that 8.66% (n=3,933) of the 45,420 GM12878 peaks identified only in Fiber-seq mapped to segmentally duplicated (SD) regions of the human genome (Fig. 2G,H). SDs comprise ~200 Mbp^31,32 that contribute to a variety of human diseases^33,34, but these regions have been challenging to study owing to short-read mapping issues to these highly similar duplicated sequences³⁵. Overall, 45% (3,933/8,700) of all Fiber-seq FIRE peaks in SDs are FIRE-specific peaks, and the SDs with the highest sequence identity contained the highest fraction of FIRE-specific peaks (Fig. 2H). Together, this demonstrates that FIRE is uniquely able to resolve chromatin patterns across both traditionally mappable genomic loci as well as complex genomic regions largely impenetrable to short-read assays.

A haplotype-resolved view of human gene regulation

We next assessed whether haplotype-phasing our Fiber-seq reads could enable us to map chromatin across the 6 Gbp diploid human genome. For GM12878, each ~20kb read on average spans at least one heterozygous variant, and parental short-read sequencing data can bin these reads based on their maternal or paternal origin, enabling the accurate haplotype phasing of 87.9% of reads across GRCh38 (Methods). Using haplotype-phased reads, we pseudo-bulked our per-read chromatin architectures by haplotype, generating haplotype-specific chromatin actuation, CpG methylation, nucleosome positioning, and TF occupancy patterns genome-wide (Fig. 3A). Notably, elements showed more variability in actuation between haplotypes than between Fiber-seq replicates (Fig. S9B), highlighting the reproducibility of FIRE results between replicates and the central role of individual haplotypes in guiding chromatin accessibility patterns. We identified elements with haplotype-selective chromatin accessibility (HSCA) by comparing the percent actuation between the two haplotypes using a Fisher’s exact test with a genome-wide FDR correction (Benjamini-Hochberg). This resulted in the identification of 9,773 elements with nominally significant (p-value <0.05) HSCAs in GM12878 cells, with 1,231 of these elements meeting genome-wide significance along the autosomes (FDR 5%) (Fig. 3A,B). As expected, haplotype-selective peaks were enriched at known imprinted sites, such as the GNAS locus, enabling the precise demarcation of actuated elements and TF binding events that are impacted by imprinting at these sites (Fig. 3A). However, known imprinting sites comprised only 5% of all haplotype-selective peaks (Fig. 3C,D), indicating that other features are the major drivers of haplotype-selective chromatin genome-wide.

Figure 3. Haplotype-selective chromatin accessibility.

A) The GNAS imprinted locus in GM12878 showing GNAS isoforms, scATAC-seq, DNase-seq, CTCF ChIP-seq, NFYA ChIP-seq, mCpG, FIRE percent actuation, FIRE peaks, and maternally (blue) or paternally (red) haplotype-selective FIRE peaks.

B) Paternal minus maternal accessibility for FIRE haplotype-selective peaks, stratified by p-value (two-sided Fisher’s exact test). Dashed line, genome-wide significance at 5% FDR (Benjamini-Hochberg). Red plus marks, known imprinted sites.

C) Stratification of haplotype-selective peaks by imprinting status and the number of genetic variants within each peak.

D) Histogram of the haplotype differences in percent CpG methylation for haplotype-selective peaks within imprinted sites, sites without genetic variants, and non-haplotype-selective peaks.

E) Schematic of sequencing 13 trios with parental short-reads for phasing and Fiber-seq on the probands to identify parent-of-origin effects (POE) in chromatin.

F) Distribution of haplotype-selective chromatin accessibility (HSCA) peaks showing the fraction of fibroblast samples with consistent maternal or paternal bias. Dark red in the histogram indicates previously identified imprinted sites; the pie chart shows the proportion of sites with consistent POE.

G) Browser views of two genomic regions (MAGEL2 and ZDBF2) demonstrating consistent POE in fibroblasts.

H) Relationship between parental bias in CpG methylation (y-axis) and FIRE actuation (x-axis). Purple points represent new imprinted sites, with five showing consistent POE in FIRE without evidence of differential CpG methylation.

To further validate this, we performed Fiber-seq on 13 fibroblast cell lines for which parental genomic sequencing data was available to phase chromatin patterns to the maternal and paternal haplotype (Fig. 3E). Using this, we identified 30 FIRE peaks that showed consistent parent-of-origin haplotype-selective chromatin across the 13 fibroblast cell lines (Fig. 3F), 5 of which were previously unannotated as being within imprinted elements as defined using CpG methylation patterns. Notably, all 5 of these elements are within genomic regions known to be imprinted but do not directly overlap a differentially CpG-methylated region (Fig. 3G,H), indicating that parent-of-origin chromatin accessibility at these sites is independent of local CpG methylation. Overall, this demonstrates that only a fraction of haplotype-selective peaks are caused by previously uncharacterized imprinted elements, and that elements with consistent parent-of-origin chromatin accessibility can be maintained even in the absence of underlying parent-of-origin CpG methylation.

Haplotype-selective chromatin marks disease-associated loci and elements

GM12878 cells contain 3.2 million autosomal heterozygous variants distinguishing each haplotype, and we evaluated whether these genetic variants contribute to haplotype-selective chromatin in GM12878 cells. Overall, autosomal FIRE peaks with HSCA are enriched in directly overlapping heterozygous variants, with 56% of these peaks containing at least one heterozygous variant (Fig. 3C), and some containing over 10 heterozygous variants. To determine whether the encompassed heterozygous variants may in fact be causal of the HSCA, we quantified the overlap of these elements with variants known to have genome-wide significant associations with different human traits and diseases (i.e., lead GWAS variants). Overall, FIRE peaks with HSCA showed a 6-fold enrichment (36 versus 6) in overlapping lead GWAS variants compared to random non-HSCA FIRE peaks (Fig. 4A), indicating that haplotype-selective chromatin at these elements may, in fact, be directly resulting from the effect of these underlying genetic variants.

Figure 4. Haplotype-selective chromatin in the major histocompatibility complex.

A) The number of haplotype-selective peaks (red) or random windows (blue) in GM12878 that overlap GWAS variants that are heterozygous in GM12878.

B) Top, the fraction of lead GWAS variants within a given distance of haplotype-selective peaks with a genetic variant (red), haplotype-selective peaks without genetic variants, and size-matched random FIRE peaks. Bottom, the difference between an HSCA with a genetic variant versus a random set of FIRE peaks of the same size.

C) Histogram of minor allele frequencies for GWAS lead variants that directly overlap haplotype-selective FIRE peaks. Red bars indicate variants within 40 kbp of a haplotype-selective FIRE peak containing a genetic variant.

D) Enrichment of disease-associated variants within 40 kbp of haplotype-selective FIRE peaks for different disease associations. The x-axis shows the log2 fold enrichment, and the y-axis represents the p-value of a two-sided Fisher’s exact test.

E) Haplotype-selective sites in the HLA-DQA1/HLA-DQB1 locus for CD8+ T-cells sequenced to ~30-fold coverage across three individuals.

F) Haplotype-selective Fiber-seq patterns showing disruption of single-molecule Fiber-seq TF occupancy and chromatin actuation in haplotype 1 of Donor 5 (Hap 1. Alt) compared to the reference haplotype, which is haplotype 2 in Donor 5 (Hap 2. Ref).

G) Histogram of the percent of fibers with TF occupancy at the CCAAT box and E-box across both haplotypes of Donor 5. Inaccessible fibers make up the remaining percentage of molecules not shown.

H) Ideogram of the number of haplotype-selective sites in high-coverage CD8+ T-cells and the two-sided Fisher’s exact significance of the enrichment of haplotype-selective chromatin (Methods).

I) Cell-selective enrichment of HLA class I, II, and III for haplotype-selective elements (two-sided Fisher’s exact test).

J) Schematic of testing intra- versus inter-sample cosine similarity between four haplotypes from two donors (GM12878 and COLO829BL) and enrichment of inter-sample similarity within GRCh38 alternative haplotypes (p < 1e-04; permutation test n=10,000; Methods) and segmental duplications (p = 0.0357; permutation test n=10,000; Methods).

However, most lead GWAS variants are not thought to be disease-causal but rather are thought to be tagging a haplotype containing variants that are mediating the disease or trait association, including rare variants with a large effect size³⁶. To evaluate whether these lead GWAS variants are tagging a putatively functional variant along the same haplotype, we quantified the incidence of HSCA FIRE peaks within 500 kbp surrounding lead GWAS variants in GM12878. This demonstrated that lead GWAS variants were enriched for being located within 40 kbp of an HSCA FIRE peak (Fig. 4B). Notably, unlike lead GWAS variants that directly overlap HSCA FIRE peaks, these adjacent HSCA FIRE peaks preferentially contained rare genetic variants with a minor allele frequency (MAF) of less than 5%, indicating that Fiber-seq is identifying rare variants along these haplotypes that are potentially mediating the disease/trait associations (Fig. 4C). Consistent with this, we found that the disease associations of these GWAS lead variants preferentially localized adjacent to GM12878 haplotype-selective chromatin elements were significantly enriched (p < 0.05, two-sided Fisher’s exact test) for diseases or traits related to the biological function of lymphoblast cells (Fig. 4D, Supplemental Table 3).

The per-molecule and near-nucleotide resolution of Fiber-seq permitted the dissection of the specific TF binding elements and variants likely mediating these disease associations. For example, haplotype-selective Fiber-seq patterns within the MOG1/HLA-F locus demonstrated that among the lead SNPs associated with platelet counts, which is frequently an immune-mediated phenotype, only two were present within putative regulatory elements, with one (rs4713235) directly disrupting Fiber-seq TF occupancy at a CTCF binding element (Fig. S10). Furthermore, the 75 kb region surrounding the HLA-DQA1 gene contains 3,199 heterozygous genetic variants that distinguish both haplotypes in GM12878 (Fig. S11A). However, TF footprinting using Fiber-seq in GM12878 cells and primary T-cells enabled the identification of specific variants mediating these haplotype-selective chromatin features (Fig. 4E). For example, TF footprint patterns in the Fiber-seq data exposed that among the 54 variants within a 1 kb region overlapping an HLA-DQA1 promoter-proximal HSCA FIRE peak, rs9271894, which is associated with Celiac disease risk in biobank PheWAS studies³⁷ (OR 2.84, p-value 1e-250), appears to be likely driving this HSCA signal. Specifically, this HLA-DQA1 promoter-proximal HSCA FIRE peak contained two TF footprints with altered occupancy between haplotypes, one encompassing a predicted high-affinity E-box and one encompassing a predicted low-affinity CCAAT box (Fig. 4F,G). However, in these samples with HSCA only the TF footprint encompassing the E-box also contained a heterozygous variant (rs9271894), which disrupts a central base within the E-box. Furthermore, single-molecule protein footprint co-occupancy analyses¹⁵ of these two TF footprints demonstrated that TF occupancy at the CCAAT box is cooperatively dependent upon TF occupancy at the E-box (Fig. 4G), suggesting a central role of this E-box, and potentially of rs9271894 in mediating this HSCA signal. To investigate whether rs9271894 is driving this HSCA signal, we performed plasmid Fiber-seq³⁸ using a 400 bp region encompassing this FIRE peak, as plasmid Fiber-seq is a scalable approach for directly measuring how non-coding variants alter chromatin accessibility. Plasmid Fiber-seq demonstrated that altering rs9271894 alone was sufficient to alter chromatin accessibility of this peak (Fig. S12, Supplemental Table 4). Together, these findings prioritize rs9271894 as the likely functional variant mediating this non-coding HLA-DQA1 Celiac disease association.

The HLA locus contains the highest rate of chromatin epigenomic diversity

Next, we evaluated whether certain genomic loci preferentially harbor HSCA FIRE peaks, irrespective of their disease association. The enrichment of HSCA FIRE peaks was quantified within rolling 100 peak windows (Methods) using Fiber-seq data from lymphoblastoid cells, activated and sorted CD8⁺ T-cells, fibroblasts, and thyroid tissue from multiple unrelated donors. After removing known imprinting regions, this demonstrated that the MHC region on chromosome 6 contained the most haplotype-selective chromatin in the entire human genome in CD8⁺ T-cells (corrected p-value 0.0013, Benjamini-Hochberg corrected FDR < 5%) (Fig. 4H). Notably, whereas this enrichment for haplotype-selective chromatin was selectively localized to the MHC region HLA class II and HLA class I loci in lymphoblastoid cells, CD8⁺ T-cells only showed this localization within the HLA Class II locus (Fig. 4I), reflecting differences in the biological roles of these loci in distinct immune cell types. In contrast, although fibroblasts and thyroid tissue show numerous actuated elements within the MHC region (Fig. S11B,C), these samples did not show any enrichment for haplotype-selective chromatin within the HLA locus (Figs. 4I, S9C), indicating that although many MHC proteins are ubiquitously expressed across various cell types, the marked genetic diversity within the MHC region is only associated with haplotype-selective chromatin patterns in select cell types.

We next evaluated whether the haploid chromatin accessibility within the MHC region showed more variability between both haplotypes within the same individual or between individuals. To accomplish this, we compared haploid Fiber-seq data from the COLO829BL and GM12878 lymphoblastoid cell lines, which were derived from two separate donors of different ages and sex (Fig. 4J). While the far majority of the haploid genome showed greater variability in chromatin accessibility between both donors, as would be expected, we did identify 40 extended genomic loci, including the MHC region, that showed greater chromatin accessible variability between both haplotypes within the same donor. Notably, these 40 genomic loci were significantly enriched for containing alternative haplotypes, like the HLA locus (permutation test n=10,000, p<0.0001) and segmental duplications (permutation test n=10,000, p<0.0357). This demonstrates that genomic regions containing some of the most genetically diverse human haplotypes^39–41 also show the most haplotype-selective chromatin patterns, consistent with the hypothesis that selective pressures and/or consequences of these diverse haplotypes are also at the level of altered gene regulatory patterns⁴².

Somatic autosomal epimutations mark the chromatin landscape of cell lines and tissue

We next sought to understand the mechanistic basis underlying haplotype-selective chromatin at the 44% of GM12878 elements with HSCA that did not contain a genetic variant directly within the peak (i.e., ‘HSCA-noVar’ elements) (Fig. 3C). Notably, unlike GM12878 elements with HSCA that overlap a heterozygous genetic variant (i.e., ‘HSCA-withVar’ elements), HSCA-noVar elements did not show any enrichment for being adjacent to lead GWAS variants (Fig. 4B, S9C), suggesting that haplotype-selective chromatin at HSCA-noVar elements may not be genetically determined. Furthermore, HSCA-noVar elements are overwhelmingly not previously undiscovered imprinting sites (Fig. 3F), raising the possibility that haplotype-selectivity at these sites may arise from somatic epimutations causing autosomal random monoallelic chromatin accessibility. To test this, we asked whether haplotype-selectivity is preserved across cell types from the same donor across peaks with and without variants. Specifically, we performed Fiber-seq on a lymphoblastoid cell line (COLO829BL) and melanoma cell line (COLO829) derived from the same donor (Fig. 5A). To ensure consistent haplotype phasing between these two samples, we leveraged a near telomere-to-telomere assembly of COLO829BL using Fiber-seq, ultra-long Oxford Nanopore, and Hi-C (contig N50 140.01 Mbp with 18 chromosomes having telomere-to-telomere scaffolds)⁴³, enabling us to unambiguously phase 84.2% of all Fiber-seq reads from COLO829BL and 83.9% of Fiber-seq reads from COLO829. Overall, COLO829BL had 710 peaks with genome-wide significant HSCA and haplotype-selectivity of these peaks was in strong agreement across two COLO829BL Fiber-seq replicates, including at HSCA-noVar elements (Pearson’s correlation > 0.9, p-value < 2.2e-16, two-sided t-test) (Fig. 5B). 29% (202/710) of these HSCA elements in COLO829BL also demonstrated some degree of chromatin actuation in the COLO829 cells, enabling us to evaluate the stability of HSCA at these 202 elements (Fig. 5C). We find that for imprinted sites (R=0.87) or COLO829BL HSCA-withVar elements (R=0.76), there is a strong correlation in the HSCA between both cell types (Fig. 5D,E), consistent with imprinted and HSCA-withVar elements largely having deterministic chromatin actuation dependent upon the underlying haplotype. In contrast, COLO829BL HSCA-noVar elements had markedly divergent haplotype-selective actuation between the two cell types (R=0.15) (Fig. 5D,E) indicating that this class of elements overall displays a non-deterministic pattern of chromatin actuation consistent with somatic epimutations.

Figure 5. Deviation of haplotype-selective chromatin across cell types.

A) Schematic of the sequencing of donor 2 of lymphoblast and melanoma cell lines.

B) Replicate concordance of haplotype-selective percent actuation difference across two lymphoblast replicates for imprinted elements (red), elements with genetic variants between haplotypes (orange), and elements without genetic variants (black). Shown is the Pearson’s correlation; all correlations are significant with p < 2.2e–16 (two-sided t-test).

C) The overlap of 710 lymphoblast haplotype-selective peaks with genetic variants and the subset of those peaks (202) that also overlap peaks within the melanoma cell line.

D) Example of shared haplotype-selective elements and unique haplotype-selective elements between lymphoblast and melanoma cells.

E) Concordance of haplotype-selective peaks between lymphoblast and melanoma cells within imprinted sites, sites with genetic variants, and sites without genetic variants (Pearson’s correlation; two-sided t-test: left p = 1e-4, middle p < 2.2e–16, right p = 0.21).

F) Experimental design showing sequencing of liver and lung primary tissues from donor 3.

G) Comparison of haplotype-selective peak concordance between liver and lung primary tissues, analyzed as in panel E.

H) Conceptual model distinguishing haplotype-invariant elements from haplotype-selective elements with deterministic and non-deterministic patterns.

I) Model illustrating how population bottlenecking in cell populations may generate additional haplotype-selective chromatin accessibility.

J) Experimental approach for sequencing nine female fibroblast cell lines, using X-inactivation allelic skewing as a proxy for cell clonality.

K) Correlation analysis between the number of haplotype-selective FIRE peaks without genetic variants and the average allelic skew across all X chromosome promoters, excluding the pseudoautosomal regions (Pearson’s correlation; two-sided t-test p-value = 7.9e–05).

To further evaluate whether autosomal HSCA-noVar elements have non-sequence-deterministic patterns of chromatin accessibility, we used AlphaGenome⁴⁴ to predict haplotype-resolved FIRE signal at HSCA elements within GM12878 cells. For HSCA-withVar elements, AlphaGenome-predicted haplotype DNase differences strongly correlated with measured FIRE actuation differences (R=0.69), with 90.6% of peaks having concordant directionality between predicted and observed haplotype accessibility differences (Fig. S13). In contrast, at HSCA-noVar elements, AlphaGenome predicted minimal DNase differences between the two haplotypes (Fig. S13), consistent with the current sequence-based models being unable to predict epigenetic differences that arise independently of sequence, such as somatic epimutations.

To validate these findings in primary human tissues, we performed Fiber-seq on primary lung and liver tissue that were obtained from a separate single individual. Overall, we found that the lung tissue had 88 peaks with genome-wide significant HSCA (Fig. 5F), with 98.9% (87/88) of these HSCA elements in lung also demonstrating some degree of chromatin actuation in the liver tissue. Consistent with the COLO829 data, imprinted elements (R=0.88) and HSCA-withVar elements (R=0.87) showed a strong correlation in the haplotype-selective chromatin actuation between both tissues. In contrast, lung HSCA-noVar elements showed markedly divergent haplotype-selective chromatin actuation between the two tissues (R=−0.88) (Fig. 5G), consistent with haplotype-selective chromatin at these HSCA-noVar elements arising from somatic epimutations (Fig. 5H).

These findings raise the prospect that some autosomal accessible chromatin elements somatically acquire non-deterministic chromatin actuation patterns that are mitotically stable (Fig. 5I). One prediction of this model would be that bottlenecked samples would have higher rates of elements with HSCA, as a greater proportion of the cells within the Fiber-seq reaction would have arisen from the same founder cell with the somatic epimutation. To test this, we performed Fiber-seq on 46,XX fibroblast cell lines and measured both the number of HSCA-noVar elements in each sample, as well as the degree of skewing of X chromosome inactivation (XCI), which can be used as a proxy for how bottlenecked the sample is. Consistent with this prediction, we observed that fibroblast lines with greater evidence of clonal bottlenecking demonstrated more HSCA-noVar elements (R=0.95) (Fig. 5J,K). Together, these findings indicate that the autosomal accessible chromatin landscape of a cell is molded by somatic epimutations that create mitotically stable and non-genetically deterministic chromatin alterations.

Somatically inactivated elements and domains along the X chromosome

The X chromosome in 46,XX cells is subjected to mitotically stable random XCI, creating an inactive X (Xi) and active X (Xa) chromosome within the same cell that is largely non-genetically deterministic^45,46. XCI represents an extreme case of mitotically stable somatic epimutations, and we sought to utilize Fiber-seq to explore the impact of XCI across every regulatory element along the X chromosome. We applied Fiber-seq to two 46,XX cell lines (GM12878 - previously introduced, and MAN_1877) that have allelic XCI, enabling us to accurately map the chromatin architecture along the Xi and Xa based on parental haplotypes (Fig. 6A). Full-length long-read transcript sequencing of GM12878 cells showed that 99% of the long non-coding RNA XIST transcripts^47–50 originate from the maternal haplotype (Fig. S14), confirming that the Xi is invariantly the maternal haplotype in GM12878 cells. Using the Fiber-seq data, we categorized FIRE elements along the X chromosome as to whether they were preferentially somatically silenced on the Xi (i.e., ‘Xa-selective’ accessibility) or Xa (i.e., ‘Xi-selective’ accessibility), or showed similar chromatin accessibility along both the Xa and Xi (i.e., ‘shared’).

Figure 6. Haplotype-specific features of X chromosome inactivation (XCI).

A) Schematic of culture-derived XCI skewing.

B) Chromosome-wide comparison between percent actuation of the paternal (Xa) and maternal (Xi) haplotypes at each FIRE peak in LCL cells (GM12878). Pseudoautosomal regions (PAR1 & PAR2) are highlighted in orange and gray, respectively.

C) Counts of Xa-specific, Xi-specific, and shared FIRE peaks in LCL (top) and fibroblast (bottom) cells, stratified by location within or outside PAR1 (left), and, for non-PAR1 elements, by overlap with a CTCF site (middle) or TSS (right).

D) Scatterplot of LCL Fiber-seq percent actuation on the Xi (x-axis) and Xa (y-axis) for each TSS. Points are colored by XCI escape annotations from previous studies^56,57.

E) UBA1 promoter region comparing full-length transcript reads, scATAC-seq, CTCF ChIP-seq, mCpG, FIRE percent actuation, FIRE peaks, and representative Fiber-seq reads from the paternal (Xa) and maternal (Xi) haplotypes in LCLs (GM12878).

F) Scatterplot of Fiber-seq percent actuation on the Xi in LCL (x-axis) and fibroblast (y-axis) cells. Points are colored as in panel D.

G) The average number of escaping non-TSS FIRE peaks in LCL (left) and fibroblast (right) cells by absolute distance from TSSs. Counts are displayed separately for escaping TSSs (top, purple) and inactivated TSSs (bottom, blue).

H) Full-length LCL transcript expression differences between the Xa and Xi for genes phased by Fiber-seq and displayed in D. Count differences are displayed as log2 fold-change between the haplotypes. Genes are stratified by the Fiber-seq classifications of their TSS FIRE peaks as in C.

I) The number of escaping LCL non-TSS FIRE peaks within 5 Kb of each TSS in the shared category in H. Shared TSSs were grouped into high or low log2 fold-change in expression, highlighted with blue and purple in H (*p = 0.04031; one-sided Wilcoxon rank sum test).

We observed that chromatin accessibility outside of the pseudoautosomal region 1 (PAR1) was predominantly impacted by XCI, with 63% of elements being Xa-specific, 34% being shared, and only 2.6% being Xi-specific elements (Fig. 6B–D). Aside from the XIST promoter, the majority (73.3%) of ‘Xi-selective’ elements were within ICCE (Inactive-X CTCF-binding Contact Element)⁵¹, DXZ4^52,53, and FIRRE (Functional Intergenic Repeating RNA Element) loci⁵⁴, and 78.9% of these were CTCF elements (Fig. S15), which play a role in organizing three-dimensional mega-domains on the Xi⁵⁵. 66% of ‘shared’ elements were also CTCF elements, with the majority (59%) of X chromosome CTCF sites having similar accessibility along both the Xa and Xi. Notably, boundaries separating chromatin domains subjected to XCI and escaping XCI often contained CTCF elements with protein occupancy on the Xi (Fig. S16). For example, UBA1 has four distinct accessible TSSs in GM12878 cells, with the three upstream TSSs being Xa-specific, and the canonical downstream TSS escaping XCI (Fig. 6E). The canonical XCI escaping UBA1 TSS is bookended by an upstream CTCF element that shows single-molecule TF occupancy on both the Xa and Xi (90% and 59% occupancy, respectively) (Fig. 6E). Overall, these findings demonstrate that CTCF occupancy along the human X chromosome preferentially escapes XCI mediated somatic silencing and that CTCF occupancy along chromosome X may serve as boundary elements insulating chromatin domains from XCI, features that have been previously proposed in mice⁵³.

Escaping promoter-proximal elements enable productive transcription from the Xi

16% of TSSs for transcribed genes in GM12878 cells were also marked by ‘shared’ elements (Fig. 6C). Notably, these TSSs largely corresponded to genes known to escape XCI based on prior transcript sequencing data^56,57 (Fig. 6D), as well as paired GM12878 long-read transcript data. However, the GM12878 transcript sequencing data could only phase 27% of these genes owing to the paucity of heterozygous coding genetic variation along the human chromosome X in GM12878 cells. We sought to directly test the relationship between promoter accessibility and transcript production on the Xi using the paired haplotype-phased long-read transcript and Fiber-seq data. As expected, we observed highly skewed expression in genes with Xa-specific or Xi-specific TSS accessibility (Fig. 6H). However, genes with similar TSS accessibility between the Xa and Xi showed heterogeneous expression, with some genes having similar transcript levels between the Xa and Xi (n=5), and others showing Xa-biased expression to a similar degree to that of genes with Xa-specific TSSs (n=3).

Given the widespread accessibility that we observed along the Xi, we hypothesized that the escape of promoter-proximal regulatory elements may augment productive transcription from the Xi. To address this, we computed the average number of escaping non-TSS elements within 5–10 Kbp bins away from escaping or inactivated TSSs (Fig. 6G). This revealed enrichment of escaping regulatory elements within 100 Kbp of escaping TSSs, with the largest difference (51-to 71-fold increase) within 5 Kbp of the escaping TSS. Notably, genes that have similar TSS accessibility and transcript levels between the Xa and Xi had significantly more promoter-proximal escape elements compared with genes that have similar TSS promoter accessibility but higher transcription from the Xa (12 vs 0, p-value = 0.04, one-sided Wilcoxon rank sum test; Fig. 6I). Overall, these results demonstrate that promoter accessibility is necessary but often not sufficient for robust gene expression from Xi, and that the simultaneous escape of promoter-proximal regulatory elements may play a pivotal role in maximizing the transcriptional potential of escape genes.

Discussion

We demonstrate how Fiber-seq can enable a comprehensive view of chromatin accessibility across the 6 Gbp diploid human genome with single-molecule and single-haplotype precision at near-nucleotide resolution. Specifically, we present a semi-supervised machine learning tool (FIRE) and show that FIRE provides a more comprehensive and quantitative snapshot of a cell’s accessible chromatin landscape. Importantly, FIRE enables the accurate de novo construction of a cell’s haplotype-resolved chromatin-accessible landscape directly using long-read sequencing data, enabling the advent of synchronous multi-omic long-read sequencing that can fine-map the chromatin mechanisms by which disease-associated haplotypes cause human disease^15,16, as well as resolve chromatin architectures within complex genomic loci like segmental duplications⁵⁸, centromeres^22,23, and transposons²⁴.

Overall, we find that genomic loci marked by the most genetic diversity^39–41,59 also contain the most haplotype-selective chromatin, indicating genetic diversity at these loci is likely directly impacting gene regulatory patterns. As changes in gene regulation are known to play a dominant role in human speciation⁴², it is likely that these haplotype-resolved maps of gene regulation will improve our understanding of the specific genetic variants that make us humans. For example, we show that Fiber-seq can localize the specific regulatory elements and TF binding events that underlie human disease associations, including those within the HLA locus. Although it is largely assumed that only protein-coding variants mediate HLA locus autoimmune disease risk, emerging data have indicated that non-coding variants play a critical role in autoimmunity through altering HLA gene expression^60–62. We show that Fiber-seq can pinpoint specific genetic variants within the HLA locus that drive disease associations, exposing that the Celiac disease-associated SNP rs9271894 disrupts cooperative TF occupancy at an E-box and adjacent CCAAT box immediately upstream of HLA-DQA1 (Fig. 4F,G)⁶³. Although this variant’s impact on immune cell function is unknown, Fiber-seq localized which of the thousands of HLA variants show a molecular phenotype at the chromatin level, an approach that should reshape how non-coding variation in this locus is studied.

It is well known that the establishment of X inactivation differs dramatically between mice and humans, both in terms of the order of events and the epigenetic silencing mechanisms⁴⁶. Although short-read epigenetic maps of the Xi and Xa can be readily established using hybrid mice, this is not readily accomplishable in humans owing to the lack of genetic diversity along chromosome X. We demonstrate that Fiber-seq enables the precise delineation of chromatin elements along the Xa and Xi in humans, revealing widespread chromatin accessibility across the Xi and exposing that promoter accessibility along the Xi is necessary but often not sufficient for robust gene expression from Xi. Epigenetic editing approaches are actively being investigated for the treatment of X-linked diseases⁶⁴, and we anticipate that Fiber-seq will provide a roadmap for these translational efforts.

Outside of the X chromosome, we find that ~99% of autosomal accessible elements show nearly identical levels of chromatin accessibility between the two haplotypes. This demonstrates that, on average, chromatin accessibility is largely deterministic, indicating the potential of computational tools for accurately predicting the vast majority of a cell’s steady-state chromatin accessibility landscape⁴⁴. As transvection is not known to impact human gene regulation, the chromatin patterns along each haplotype are independently formed during a cell’s and organism’s life span, a system ripe for stochastic deviations in the accessibility pattern between both haplotypes as a result of epigenetic noise and subsequent memory⁶⁵. Fiber-seq exposed numerous elements with HSCA that appear to be primarily driven by underlying heterozygous genetic variants. However, ~40% of elements with HSCA lacked an underlying genetic variant. We show that these elements 1) largely have non-deterministic behavior, 2) are not imprinted sites, and 3) can be observed in primary human tissues. Together, these findings demonstrate that autosomal accessible chromatin elements can somatically acquire non-deterministic chromatin actuation patterns that are mitotically stable and divergent between haplotypes, consistent with these being sites of somatic epimutations⁶⁶. Somatic epimutations are increasingly recognized in CpG methylation and transcript sequencing^67,68, but have been challenging to systematically study at the level of chromatin⁶⁹ owing to the inability to accurately haplotype phase short sequencing reads at elements that lack underlying heterozygous variants within a sample. Notably, we find that clonal expansions increase the rate of somatic epimutations, consistent with these somatic epimutations arising from founder events. Together, these findings indicate that the autosomal accessible chromatin landscape of a cell is molded by somatic epimutations that create mitotically stable and non-genetically deterministic chromatin alterations.

Autosomal monoallelic expression is a well-documented feature of normal human tissues⁷⁰ and can influence disease risk⁷¹. While this phenomenon is thought to arise from somatic epimutations, with roles established in both rare⁷² and common⁷³ diseases, current catalogs are largely based on CpG methylation, owing to the haplotype-resolving power of single-molecule methylation assays. Here, we show that autosomal, mitotically stable, haplotype-selective regulatory patterns can arise independently of CpG methylation (Fig. 3E–H), and that most regulatory elements consistent with somatic epimutations lack pronounced haplotype-specific CpG methylation differences (Fig. 3D), though we cannot exclude that differential methylation at distal regulatory elements may contribute to haplotype-selective accessibility at some loci. These findings suggest that epigenetic memory at these loci is primarily encoded in local chromatin architecture rather than CpG methylation or the trans environment. More broadly, our results reveal that somatic epimutations are pervasive and impose an epigenetic ceiling on the deterministic influence of genetic variation in shaping the human diploid epigenome.

Limitations of the Study

Several limitations should be considered when using FIRE or, more broadly, when interpreting Fiber-seq data. First, it is important to ensure appropriate data quality and coverage, especially when comparing multiple datasets. Specifically, the power to detect FIRE peaks (FDR < 0.05) is directly related to coverage. While a strength of Fiber-seq is that molecules are sequenced regardless of accessibility, this also necessitates genome-wide sequencing, which can result in variance in coverage between samples or experiments. Consequently, comparing accessibility signal at a union set of peaks between samples is often more informative than simply comparing the number of intersecting peaks. Of note, deeply sequenced samples often are powered to identify FIRE peaks with less than 15% chromatin actuation, but the biological significance of these lowly actuated peaks remains to be seen. Second, FIRE infers the presence of regulatory elements from m6A modification patterns but cannot directly identify which TFs or protein complexes are bound. Confidently resolving specific-factor occupancy requires complementary approaches, such as ChIP-seq/CUT&RUN/DiMeLo-seq or motif-guided footprint analysis. Third, our characterization of somatic epimutations is inherently conservative: we define HSCA-noVar elements as those lacking any overlapping heterozygous variant and require genome-wide significance, meaning the true prevalence of non-genetically determined haplotype-selective chromatin may be higher than reported. Moreover, while we demonstrate that these elements are mitotically stable, non-deterministic, and amplified by clonal bottlenecking, we do not propose a specific molecular mechanism for their establishment or maintenance. Although we observe somatic epimutations in primary liver and lung tissue, our most detailed characterization relies on GM12878, a clonal cell line in which generation and long-term culture may have reshaped clonal dynamics. Finally, the FIRE model was trained on human GM12878 Fiber-seq data using paired DNase-seq and CTCF ChIP-seq peaks as positive labels. Although we demonstrate generalization to other TFs, cell types, and organisms²⁴, regulatory elements that systematically diverge from human DNase-seq or CTCF ChIP-seq patterns may be underrepresented using FIRE.

Resource availability

Lead Contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Andrew B. Stergachis (absterga@uw.edu).

Materials Availability

No new materials were generated from this study.

Data and Code Availability

Accessions for all ENCODE data used within this manuscript are provided in Supplemental Table 5. Processed epigenetic (FIRE) results for all samples are publicly available through https://doi.org/10.5281/zenodo.14511246, as well as the Fiber-seq data portal https://stergachislab.github.io/Fiber-seq-publication-data/. Per-sample information is provided in Supplemental Table 6. All raw and processed sequencing data generated for GM12878 (PRJNA1233341) and K562 (SRX20077598, SRX20077599, and SRX20077600) have been submitted to the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra/). Restrictions apply to the availability of some genetic data generated or analyzed during this study to preserve subject confidentiality. Genomic data for these samples are accessible to the scientific community through the UDN, GREGoR, All of US, and SMaHT consortia, with per-sample information and accessions provided in Supplemental Table 6. Individuals interested in accessing this should submit a data access request to the relevant consortia.

The code and methodology of FIRE are available on GitHub (https://github.com/fiberseq/FIRE) and Zenodo (https://zenodo.org/records/11075226 and https://doi.org/10.5281/zenodo.20415764). Fibertools is available on GitHub (https://github.com/fiberseq/fibertools-rs) and Zenodo (https://zenodo.org/records/10850620). The code used for phasing long-read data is available as a Snakemake pipeline on GitHub (https://github.com/mrvollger/k-mer-variant-phasing) and Zenodo (https://zenodo.org/records/10655527). The code for making figures and tables is available on GitHub (https://github.com/mrvollger/fire-figures) and Zenodo (https://zenodo.org/doi/10.5281/zenodo.10681988).

STAR Methods

Experimental Model and Study Participant Details

Cell lines used for this study

The GM12878 and K562 cell lines used for this study were derived from female donors. GM12878 cells were obtained from the National Institute of General Medical Sciences Human Genetic Cell Repository at the Coriell Institute for Medical Research. K562 cells were obtained from ATCC. THP-1 and A549 cells were generously donated by the lab of John Stamatoyannopoulos, and Jurkat cells by the lab of John Ray.

Method Details

Cell Culture

GM12878 were purchased from the Coriell Institute for Medical Research and cultured in RPMI 1640 Medium supplemented with 2mM L-glutamine,10% fetal bovine serum, and 100 I.U./mL penicillin/100 μg/mL streptomycin. Cells were maintained in a 37 °C humidified incubator under 5% carbon dioxide. Cell cultures were split every 3–5 days.

Fiber-seq Library Preparation and Sequencing

Cells were permeabilized and treated with Hia5 enzyme as previously described (75). Specifically, 1 million cells were washed with PBS and then resuspended in 60 μl Buffer A (15 mM Tris, pH 8.0; 15 mM NaCl; 60 mM KCl; 1mM EDTA, pH 8.0; 0.5 mM EGTA, pH 8.0; 0.5 mM Spermidine) and 60 μl of cold 2X Lysis buffer (0.1% IGEPAL CA-630 in Buffer A for GM12878 and GM24385; 0.2% IGEPAL CA-630 in Buffer A for UDN318336 fibroblasts) was added and mixed by gentle flicking then kept on ice for 10 minutes. Samples were then pelleted, the supernatant removed, and then resuspended in 57.5 μl Buffer A and moved to a 25°C thermocycler. 0.5 μl of Hia5 MTase (100 U) and 1.5 μl 32 mM S-adenosylmethionine (NEB B9003S) (0.8 mM final concentration) were added, then carefully mixed by pipetting the volume up and down 10 times with wide bore tips. The reactions were incubated for 10 minutes at 25°C, then stopped with 3 μl of 20% SDS (1% final concentration) and transferred to new 1.5 mL microfuge tubes. High molecular weight DNA was then extracted using the Promega Wizard HMW DNA Extraction Kit A2920. PacBio SMRTbell libraries were then constructed using the Fiber-seq treated gDNA following the manufacturer’s SMRTbell prep kit 3.0 procedure (https://www.pacb.com/wp content/uploads/Procedure-checklist-Preparing-whole-genome-and-metagenome-libraries-using SMRTbell-prep-kit-3.0.pdf).

Nucleosome and MSP (methyltransferase sensitive patch) and calling.

Nucleosome calling is performed by identifying stretches of DNA that are protected from Hia5 (i.e. do not have m6A signal). The rate of false positive m6A calls in nucleosomes is very low when using fibertools (26) allowing for a heuristic to perform as well as or better than our previous HMM caller (75). There are three parameters in our heuristic nucleosome calling that can be adjusted: the minimum nucleosome length (n, default 75), the minimum combined nucleosome length (c, default 100), and the minimum extension to nucleosome length (e, default 25). These three parameters impact the four phases in nucleosome calling.

1. Call all regions that have no m6A events for at least n bases a candidate nucleosome.

2. Call all regions of size c or more that have only one internal m6A (putative false positive) a candidate nucleosome.

3. Extend the length of nucleosomes identified in phases one and two if by spanning one additional m6A e bases of unmodified sequence would be added to the nucleosome length.

4. Recursively apply step three till no nucleosomes change.

After nucleosomes are called MSPs are operationally defined as all the regions between nucleosome footprints in the Fiber-seq data. This process is automated through the fibertools subcommand “add-nucleosomes”.

Training of Fire-seq inferred regulatory elements.

For training data, we generated 21 different GM12878 Fiber-seq experiments with a range of under- and over-methylated experimental conditions to ensure we captured a broad range of percent m6A (Fig. S1; 5.8–13.3%) to ensure our model could generalize to new samples with varying levels of m6A. We merged these sequencing results and randomly selected 10% of the Fiber-seq reads from 100,000 randomly selected DNase I and CTCF ChIP-seq peaks for mixed-positive labels and 100,000 equally sized regions that did not overlap DNase or CTCF Chip-seq for negative labels. We then generated features for each of these MSPs, including length, log fold enrichment of m6A, the A/T content, and windowed measures of m6A around the MSP (Supplemental Table 1) and held out 20% of the Fiber-seq reads to be used as test data.

To carry out semi-supervised training, we used an established method, Mokapot (25), which we summarize below. In the first round of semi-supervised training, Mokapot identifies the feature that best discriminates between our mixed-positive and negative labels and then selects a threshold for that feature such that the mixed-positive labels can be discriminated from the negative labels with 95% estimated precision (defined below). The subset of mixed-positive labels above this threshold is then used as an initial set of positive labels in training an XGBoost model with five-fold cross-validation (28). Then this process is iteratively repeated, using the learned prediction from the previous iteration’s model to create positive labels at 95% estimated precision, until the number of positive identifications at 95% precision in the validation set ceases to increase (15 iterations, Fig. S1). The code and methodology for FIRE are available on GitHub (https://github.com/fiberseq/FIRE) and Zenodo (https://zenodo.org/records/11075226 and https://doi.org/10.5281/zenodo.20415764).

Estimated precision of individual FIRE elements

We cannot compute the precision associated with a particular XGBoost score because we do not have access to a set of clean-positive labels. Instead, we define a notion of “estimated precision” using a balanced held-out test set of mixed-positive and negative labels (20% of the data). We defined the estimated precision (EP) of a FIRE element to be

$$EP=1–\frac{FP+1}{TMP}$$

where TMP is the number of “true” identifications from the mixed positive labels with at least that element’s XGBoost score, and FP is the number of false positive identifications from negative labels with at least that element’s XGBoost score. We add a pseudo count of one to the numerator of false positive identifications so as to prevent liberal estimates for smaller collections of identifications (25).

Aggregate FIRE score calculation

The FIRE score (S_g) for a position in the genome (g) is calculated using the following formula:

$$S_g=\frac{-50}{R_g}\sum _{i=1}^{C_g}lo{g}_{10}(1-min(E{P}_i,0.99))$$

where C_g is the number of FIRE elements at the g^th position, R_g is the number of Fiber-seq reads at the g^th position, and EP_i is the estimated precision of the i^th FIRE element at the g^th position. The estimated precision of each FIRE element is thresholded at 0.99, such that the FIRE score takes on values between 0 and 100. Regions covered by less than four FIRE elements (i.e., if C_g < 4) are not scored and are given a value of negative one (Fig. S2).

Regions of unreliable coverage

Regions with unreliable coverage were defined as regions with Fiber-seq coverage that deviate from the median coverage by five standard deviations, where a standard deviation is defined by the Poisson distribution (i.e., the square root of the mean coverage).

FIRE score FDR calculation

We shuffle the location of an entire Fiber-seq read by selecting a random start position within the chromosome and relocating the entire read to that start position. Fiber-seq reads originating from regions with unreliable coverage (defined above) are not shuffled, and reads from regions with reliable coverage are not shuffled into regions with unreliable coverage (bedtools shuffle -chrom -excl, v2.31.0) (76). We then compute FIRE scores associated with the shuffled genome. Recalling that the FIRE score for the g^th position in the genome is denoted S_g, we divide the number of bases that have shuffled FIRE scores above S_g by the number of bases that have un-shuffled FIRE scores above S_g. This provides an estimate of the FDR associated with the FIRE score S_g.

Peak calling

Peaks are called by identifying FIRE score local-maxima that have FDR values below a 5% threshold and at least 10% actuation (C_g / R_g). Adjacent local maxima that share 50% of the underlying FIRE elements or have 90% reciprocal overlap are merged into a single peak, using the higher of the two local maxima. Then, the start and end positions of the peak are determined by the median start and end positions of the underlying FIRE elements. We also calculated and reported wide peaks by taking the union of the FIRE peaks and all regions below the FDR threshold and then merged the resulting regions that were within one nucleosome (147 bp) of one another.

Short-read accessibility data and comparisons

DNase I hypersensitivity sites and bigWig tracks for GM12878 were downloaded from the ENCODE data portal. We used accession ENCFF762CRQ for DNase peaks and ENCFF960FMM for the bigWig track (Meuleman et al., 2020). For CTCF CHiP-seq peaks, we used the union of ENCFF356LIU and ENCFF960ZGP (77). When intersecting short-read and FIRE peaks, we required a 1 bp overlap, and when measuring the short-read signal within a FIRE peak, we used the maximum values of the short-read signal across the FIRE peak. scATAC-seq data was downloaded in fastq format from the ENCODE portal (Supplemental Table 5). Fastq data from each experiment were processed using cellranger-atac count, and the outputs were aggregated using cellranger-atac aggr (10x Genomics v.2.1.0). Aligned fragments from passing cell barcodes were intersected with FIRE peaks using bedmap (v2.4.41) (78). Shuffled regions were generated using Bedtools shuffle (v2.31.0). scATAC-seq peaks were called using MACS2 callpeak (v2.2.7.1, parameters: -g hs -q 0.01 --nomodel --shift -75 --extsize 150 --keep-dup all -B --SPMR). scATAC-seq percent accessibility values were computed for each element in the FIRE and shuffled peak sets as the percentage of cell barcodes with at least 1 fragment overlapping that respective peak region.

FIRE Peaks Enrichment Transcription Factor Motifs

To determine the transcription factor (TF) binding motifs enriched in FIRE peaks we used Homer v3.12 (http://homer.ucsd.edu/homer/motif/) to discover TF binding sites represented in target sequences and compared them to a chosen background. Using findMotifsGenome.pl, we determined the enrichment of all FIRE peaks and all ATAC peaks to the whole genome as a background, then we compared ATAC-overlapping FIRE peaks and non-ATAC-overlapping FIRE peaks with all FIRE peaks as the background. We determined the REST motif genome coordinates in the non-ATAC-overlapping FIRE peaks for aggregate footprinting analysis using annotatePeaks.pl.

Haplotype-selective peaks

For peaks with at least 10 Fiber-seq reads on both haplotypes, we tested the difference in percent actuation in each haplotype (fraction of reads with FIRE elements) using a two-sided Fisher’s exact test. Specifically, the inputs for the test are the number of FIRE elements in haplotype one, the number of Fiber-seq reads without FIRE elements in haplotype one, the number of FIRE elements in haplotype two, and the number of Fiber-seq reads without FIRE elements in haplotype two. We then apply an FDR correction (Benjamini-Hochberg) to correct for multiple testing and select elements with a correct p-value less than or equal to 0.05 to be haplotype selective peaks.

The FIRE pipeline

The FIRE pipeline (https://github.com/fiberseq/FIRE v0.0.4) was applied to aligned and phased Fiber-seq BAM files with m6A predictions called using Fibertools-rs (v0.4) (26). The FIRE pipeline is a Snakemake (79) workflow that applies the FIRE model to individual reads, calculates the aggregate FIRE scores, computes the FDR, calls peaks, and identifies haplotype-selective peaks. The only required inputs for the FIRE pipeline are an aligned Fiber-seq BAM file and the reference genome used for alignment.

Phasing and identification of genomic variants

Variant calling for single nucleotide variants (SNVs) and insertions/deletions (Indels) was conducted using DeepVariant version 1.5.0 (80), while structural variants (SVs) were identified using pbsv (https://github.com/PacificBiosciences/pbsv). The Fiber-seq reads were haplotype-phased through a specialized pipeline available on GitHub (https://github.com/mrvollger/k-mer-variant-phasing) (16). This pipeline employs SNVs detected by DeepVariant and utilizes the HiPhase (81) variant-based phaser to organize reads into phase blocks. These blocks are then attributed to either the maternal or paternal haplotype using parental short-read genome sequencing in conjunction with the trio k-mer-based phaser meryl (k=31) when available (82).

Identification of CpG methylation

Base-level CpG methylation was called using jasmine (PacBio), and the percent CpG methylation at each genomic position was identified from a pileup of reads using pb-CpG-tools (https://github.com/PacificBiosciences/pb-CpG-tools).

Imprinted loci

Imprinted loci were defined using the 143 differentially methylated regions identified in 12 B-lymphocyte cell lines by Akbari et al. (83). Novel imprinted elements in the 13 fibroblast samples were identified by taking the union of all haplotype-selective peaks across all 13 samples and identifying sites where at least 10 of the samples showed significant chromatin actuation. Subsequently, we analyzed these elements for sites where all samples showed the same direction of maternal or paternal skew and classified these as putative novel imprinted elements.

Identifying regions enriched in haplotype-selective elements

To test for regions enriched for haplotype-selective elements, we took consecutive windows containing 100 FIRE peaks (sliding by 10 peaks) and compared the number of nominally significant haplotype-selective elements in the 100-peak window to the number of haplotype-selective peaks genome-wide using a two-sided Fisher’s exact test.

Intra- versus Inter-sample sample similarity in FIRE actuation

To measure the similarity between two haplotypes, we took genomic regions containing 100 peaks and measured the cosine similarity of the percent actuation between the two haplotypes. We then repeated this process across the genome in 100 peak windows, sliding 10 peaks at a time and comparing every unique pair of haplotypes across our two samples.

Testing for enrichment of inter-sample similarity within SDs and alternative haplotypes

To test for the enrichment of regions with greater inter-sample similarity, we compared the percent of sites intersecting with SDs or alternative haplotypes to 10,000 random shufflings of windows of the same size to calculate an empirical p-value.

Genome assembly of COLO829BL and estimation of the number of T2T scaffolds.

To create a de novo assembly for COLO829BL, we used Verkko (v1.4.1) with down-sampled PacBio HiFi reads (60-fold) and ONT ultra-long reads (60-fold) where ultra-long reads were down-sampled in descending order based on read length to retain the longest reads (4). Additionally, we used 30-fold Illumina Hi-C data within Verkko to allow for long-range phasing of the assembly. To estimate the number of telomere-to-telomere (T2T) scaffolds, we required that a single scaffold covered more than 95% of a specific chromosome based on T2T-CHM13v2.0 alignment and that there were more than 20 occurrences of telomere sequences with 500bp of both ends of the contig.

Genome assembly of ST001.

To create a de novo assembly for ST001, we used hifiasm (v0.19.9) with down-sampled PacBio HiFi reads selected to retain the longest 70-fold of reads. The initial PacBio HiFi dataset combined the Fiber-seq liver and lung reads from ST001. We then created consistent phasing by aligning the Fiber-seq reads (pbmm2 v1.13.0) back to the diploid assembly of ST001 and assigned haplotypes based on alignment to either the H1 or H2 assembly.

Plasmid construction

To generate plasmid constructs, the pEF-GFP backbone was linearized by PCR using Repliqa HiFi ToughMix (Quantabio, 95200) with an annealing temperature of 61°C for 30 cycles. The PCR product was purified using a Monarch PCR & DNA Cleanup Kit (NEB, T1130) and treated with DpnI (NEB, R0176) for 1 hour at 37°C to remove methylated template, followed by a second column purification. Gene fragments (gBlocks) encoding the H1 wild-type, 1-edit, 3-edit, and H2 reference sequences were synthesized by IDT with terminal homology arms compatible with the linearized backbone. Constructs were assembled by Gibson Assembly using Gibson Assembly Master Mix (NEB, E2611) at a 3:1 molar ratio of insert to backbone and incubated at 50°C for 1 hour.

Bacterial transformation and plasmid preparation

Gibson assembly reactions were transformed into NEB 5-alpha competent cells (NEB, C2987) following the manufacturer’s protocol and plated on LB agar supplemented with 100 μg/mL carbenicillin. Single colonies were inoculated into 5 mL LB with 100 μg/mL carbenicillin and grown overnight at 37°C with shaking at 250 rpm. Plasmid DNA was isolated using the Monarch Spin Plasmid Miniprep Kit (NEB, T1110) and sequence-verified by Plasmidsaurus. To remove endogenous m6A methylation, verified plasmids were re-transformed by electroporation into the methyltransferase-deficient strain ER2796 (1.8 kV) and recovered in SOC medium. Transformants were selected on LB with 100 μg/mL carbenicillin, cultured overnight, miniprepped, and sequence-verified as above. Verified constructs were pooled at equimolar concentrations.

Cell culture and transfection

2×10⁶ HEK293T cells were seeded in a 10 cm dish in 12 mL DMEM (Thermo Fisher, 11965076) supplemented with 10% FBS (Sigma-Aldrich, F0926) and 1% penicillin-streptomycin (Thermo Fisher, 15070063). Twenty-four hours after seeding, cells were transfected with 10 μg of the equimolar plasmid pool using Lipofectamine LTX (Thermo Fisher, 15338100) in Opti-MEM (Thermo Fisher, 31985070) according to the manufacturer’s protocol. Forty-eight hours post-transfection, cells were harvested by trypsinization (0.25% trypsin-EDTA; Thermo Fisher, 25200072), counted, and pelleted at 350 × g for 5 minutes.

Plasmid Fiber-seq

Cell pellets were resuspended in Buffer A (15 mM Tris-HCl pH 8.0, 15 mM NaCl, 60 mM KCl, 1 mM EDTA pH 8.0, 0.5 mM EGTA pH 8.0, 0.5 mM spermidine) at a concentration of 2×10⁶ cells per 60 μL. Nuclei were extracted by addition of an equal volume of 2x lysis buffer (Buffer A supplemented with 0.05% IGEPAL CA-630) and incubated on ice for 10 minutes. Nuclei were pelleted at 350 × g for 5 minutes at 4°C, resuspended in 60 μL Fiber-seq Master Mix (Buffer A, 0.8 mM SAM, 200 U Hia5), and incubated at 25°C for 10 minutes. The reaction was quenched by addition of SDS to a final concentration of 1% (38).

Plasmid DNA isolation and library preparation

Methylated plasmid DNA was isolated from treated nuclei using the Monarch Plasmid Miniprep Kit and eluted in 35 μL water. Contaminating genomic DNA was removed by treatment with Exonuclease V (NEB, M0345) at 37°C for 1 hour, followed by heat inactivation at 70°C for 30 minutes and column purification. Plasmids were linearized with KpnI-HF (NEB, R3142), and sequencing libraries were prepared using the SMRTbell Prep Kit 3.0 (PacBio) according to the manufacturer’s protocol. Libraries were sequenced on a PacBio Revio at the University of Washington PacBio Sequencing Services.

Differential nucleosome occupancy analysis

To compare chromatin architecture across plasmid constructs, we performed pairwise differential footprint analyses on Fiber-seq data from four plasmid samples (H1_wild, H2_ref, 1_edit, 3_edit, Supplemental Table 4). Each sample was processed using fibertools to obtain m6A methylation data, followed by FiberHMM to call nucleosome footprints (≥80 bp) on individual molecules. Reads were aligned to their respective plasmid reference sequences using pbmm2.

To normalize for chromatin state, reads were subsampled across all four samples to a shared nucleosome-per-read (NPR) distribution by taking the minimum read count at each NPR value. Per-position nucleosome occupancy was then computed as the fraction of subsampled reads with a nucleosome footprint overlapping each base in the plasmid sequence. To account for a 3 bp internal deletion in H1_wild relative to the other constructs, the corresponding positions were removed from the H1_wild occupancy array to restore positional alignment. To identify regions of differential nucleosome positioning, we subtracted the per-position occupancy of each sample from H2_ref (the reference haplotype), and separately compared H1_wild, 1_edit, and 3_edit to one another to assess equivalence among constructs carrying the functional variant.

X chromosome peaks.

FIRE peaks used in XCI analyses were filtered to remove peaks overlapping ENCODE 2020 hg38 blacklist regions (accession: ENCFF356LFX). Transcriptional start site (TSS) FIRE peaks were identified as peaks intersecting both GENCODE v45 TSSs (padded to 20 bp) and ENCODE CAGE-seq peaks (Supplemental Table 5). CTCF FIRE peaks were identified as peaks intersecting ENCODE CTCF ChIP-seq peaks (Supplemental Table 5). Intersections were performed using Bedtools intersect (v2.31.0). Peaks included in XCI analyses were filtered to meet the following phasing and coverage criteria: <= 25% variance from the mean coverage, < 35% of reads assigned as unphased, and at least 10X coverage from each haplotype. In addition, GM12878 peaks surrounding the MTMR1 TSS (chrX:150,631,401–150,738,995) and a fibroblast sample peak at the PDHA1 TSS (chrX:19,343,661–19,343,996) were manually filtered due to substantial disagreement in between the variant and k-mer phasing methods. Peaks were required to be actuated on at least 30% of Fiber-seq reads on the Xa and/or Xi.

XCI Classifications.

X chromosome FIRE peaks with a percent actuation difference of >= 50% between haplotypes were classified as “Xa-specific” or “Xi-specific” if the peaks were actuated on >= 30% of paternal or maternal Fiber-seq reads, respectively. FIRE Peaks were classified as “shared” if the peaks were actuated on >= 30% of Fiber-seq reads on both haplotypes or >= 30% of reads on one haplotype with a percent difference of < 50% between haplotypes. Transcriptional start site (TSS) FIRE peaks actuated on less than 30% of maternal Fiber-seq reads were classified as inactivated. TSS FIRE peaks were classified as fully escaping XCI if >= 30% of maternal Fiber-seq reads were actuated and the percent actuation difference between haplotypes was less than 25%.

Iso-seq analysis

Full-length 3’ transcriptomic iso-seq data was generated in a previous study (16). Reads were clustered by isoform using Isoseq cluster 2 (PacBio v3.99.99) and aligned to hg38 using pbmm2 (PacBio v1.11.99, parameters: --sort --min-gap-comp-id-perc 95.0 --min-length 50 --sample --report-json mapping_stats.report.json --preset ISOSEQ). Phased variant calls were used to haplotag clustered isoforms using WhatsHap v1.6 (84). Clustered isoforms were then collapsed using Isoseq collapse v4.0.0 (PacBio). Isoforms were annotated using Pigeon v1.2.0 (PacBio). Isoforms and genes with >= 35% of transcripts not assigned to a haplotype were classified as unphased.

Promoter-proximal escape quantification

TSS FIRE peaks were grouped by escape status (fully escaping or inactivated). For each group, non-TSS FIRE peaks within 100 Kb in either direction (upstream and downstream) of each TSS were counted in 5 Kb distance bins. Counts within each bin were then normalized by the number of TSSs in the respective group.

CTCF footprinting

X chromosome CTCF FIRE peaks (overlapping ENCODE CTCF ChIP-seq peaks) that also fully overlap a FIMO-predicted CTCF motif (85) (v5.5) were used for footprinting analyses. We decided to limit CTCF footprinting to the most highly bound portions of CTCF, modules two and three (86). FIRE elements that completely overlapped modules two and three of a motif were centered using fibertools ft center (26), and the number of m6A observed within the motif was quantified. FIRE-contained motifs with <= 1 m6A were classified as bound.

Supplementary Material

2

Table S1. Features used in the FIRE XGBoost model, related to Figure 1.

3

Table S2. Statistics on GM12878 FIRE peaks, related to Figures 1 and 2.

4

Table S3. Diseases enriched for GWAS variants near HSCA, related to Figure 4.

5

Table S4. Sequences used for Plasmid Fiber-seq, related to Figure 4.

6

Table S5. Manifest of ENCODE short-read accessions, related to data availability.

7

Table S6. Manifest of Fiber-seq samples, related to data availability.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
GM12878 lymphoblast cell line	Coriell	GM12878
K562 erythroleukemia cell line	ATCC	K562
BCH_1877-01 fibroblasts	GREGoR	N/A
BCH_2410-01 fibroblasts	GREGoR	N/A
BCH_0252-01 fibroblasts	GREGoR	N/A
BCH_1199-01 fibroblasts	GREGoR	N/A
UDN052264 fibroblasts	Undiagnosed disease network	N/A (UDN numbers)
UDN374570 fibroblasts	Undiagnosed disease network	N/A
UDN212054 fibroblasts	Undiagnosed disease network	N/A
UDN687128 fibroblasts	Undiagnosed disease network	N/A
UDN355246 fibroblasts	Undiagnosed disease network	N/A
UDN966087 fibroblasts	Undiagnosed disease network	N/A
UDN305049 fibroblasts	Undiagnosed disease network	N/A
UDN121217 fibroblasts	Undiagnosed disease network	N/A
UDN204349 fibroblasts	Undiagnosed disease network	N/A
UDN318336 fibroblasts	Undiagnosed disease network	N/A
UDN633333 fibroblasts	Undiagnosed disease network	N/A
ST-01475643 CD8+ T cells	All of Us	N/A
ST-01401189 CD8+ T cells	All of Us	N/A
ST-02053298 CD8+ T cells	All of Us	N/A
ST-02097424 CD8+ T cells	All of Us	N/A
ST001 liver tisssue	SMaHT	N/A
ST001 lung tisssue	SMaHT	N/A
COLO829BL lymphoblast (Blood) cell line	SMaHT	N/A
COLO829 melanoma cell line	SMaHT	N/A
R5211 thyroid tisssue	GREGoR	N/A
R4640 thyroid tisssue	GREGoR	N/A
STR 4/4* thyroid tisssue	GREGoR	N/A
Jurkat cells	John Ray lab	Jurkat
THP-1 cells	Stamatoyannopoulos lab	THP-1
A549 cells	Stamatoyannopoulos lab	A549
Critical commercial assays
Gibco RPMI 1640 Medium	ThermoFisher Scientific	11875093
Gibco Penicillin-Streptomycin	ThermoFisher Scientific	15140122
Gibco Fetal Bovine Serum (FBS)	ThermoFisher Scientific	16-140-071
Promega Wizard HMW DNA Extraction kit	Promega	A2920
PacBio SMRTbell prep kit 3.0	PacBio	102-182-700
Deposited data
Fiber-seq GM12878	This paper	NCBI BioProject: PRJNA1346268
Fiber-seq K562	Jha et al. (ref. 26)	NCBI BioProject: PRJNA956114
Fiber-seq Jurkat	This paper	SRA accessions: SRX32023674
Fiber-seq THP-1	This paper	SRA accessions: SRX32023673
Fiber-seq A549	This paper	SRA accessions: SRX32023674
Fiber-seq BCH 1877-01 fibroblast	GREGoR	GREGoR
Fiber-seq BCH 2410-01 fibroblast	GREGoR	GREGoR
Fiber-seq BCH 0252-01 fibroblast	GREGoR	GREGoR
Fiber-seq BCH 1199-01 fibroblast	GREGoR	GREGoR
Fiber-seq UDN052264 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN052264), deposited in AnVIL, accession number phs003047
Fiber-seq UDN374570 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN374570), deposited in AnVIL, accession number phs003047
Fiber-seq UDN212054 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN212054), deposited in AnVIL, accession number phs003047
Fiber-seq UDN687128 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN687128), deposited in AnVIL, accession number phs003047
Fiber-seq UDN355246 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN355246), deposited in AnVIL, accession number phs003047
Fiber-seq UDN966087 fibroblast	Undiagnosed disease network(Heinz et al. 2010)	Undiagnosed disease network (UDN966087), deposited in AnVIL, accession number phs003047
Fiber-seq UDN305049 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN305049), deposited in AnVIL, accession number phs003047
Fiber-seq UDN121217 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN121217), deposited in AnVIL, accession number phs003047
Fiber-seq UDN204349 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN204349), deposited in AnVIL, accession number phs003047
Fiber-seq UDN318336 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN318336), deposited in AnVIL, accession number phs003047
Fiber-seq UDN633333 fibroblast	Undiagnosed disease network	Undiagnosed disease network (UDN633333), deposited in AnVIL, accession number phs003047
Fiber-seq ST-01475643 CD8+ T cell	All of Us	All of Us
Fiber-seq ST-01401189 CD8+ T cell	All of Us	All of Us
Fiber-seq ST-02053298 CD8+ T cell	All of Us	All of Us
Fiber-seq ST-02097424 CD8+ T cell	All of Us	All of Us
Fiber-seq ST001 liver tisssue	SMaHT	Deposited in the SMaHT data portal, accession SMAFIAK7UKIU and dbGaP (phs004193)
Fiber-seq ST001 lung tisssue	SMaHT	Deposited in the SMaHT data portal, accessions SMAFI49TN6HY, SMAFI1Y69GQN and dbGaP (phs004193)
Fiber-seq COLO829BL lymphoblast (Blood) cell line	SMaHT	Deposited in the SMaHT data portal, accession SMAFIEQ93MZL and dbGaP (phs004193)
Fiber-seq COLO829BL lymphoblast (Blood) cell line	SMaHT	Deposited in the SMaHT data portal, accession SMAFIFRLAQ54 and dbGaP (phs004193)
Fiber-seq COLO829 melanoma cell line	SMaHT	Deposited in the SMaHT data portal, accessions SMAFIO69TOMB, SMAFIURDW1P6, SMAFIJPL47BQ and dbGaP (phs004193)
Fiber-seq R5211 thyroid tisssue	GREGoR	GREGoR; Published PMCID: PMC11472772
Fiber-seq R4640 thyroid tisssue	GREGoR	GREGoR; Published PMCID: PMC11472772
Fiber-seq STR 4/4* thyroid tisssue	GREGoR	GREGoR; Published PMCID: PMC11472772
Fiber-seq Jurkat cell line	This paper	NCBI BioProject: PRJNA1417779
Fiber-seq THP-1 cell line	This paper	NCBI BioProject: PRJNA1417779
Fiber-seq A549 cell line	This paper	NCBI BioProject: PRJNA1417779
CAGE GM12878 nucleolus fraction	ENCODE Portal	ENCODE project: ENCSR000CIT
CAGE GM12878 cytosolic fraction	ENCODE Portal	ENCODE project: ENCSR000CIR
CAGE GM12878 nuclear fraction	ENCODE Portal	ENCODE project: ENCSR000CJY
CAGE GM12878 cytosolic fraction	ENCODE Portal	ENCODE project: ENCSR000CJZ
CAGE GM12878 nuclear fraction	ENCODE Portal	ENCODE project: ENCSR000CIS
CTCF ChIP-seq GM12878	ENCODE Portal	ENCODE project: ENCSR000AKB
CTCF ChIP-seq GM12878	ENCODE Portal	ENCODE project: ENCSR000DZN
snATAC-seq GM12878	ENCODE Portal	ENCODE project: ENCSR371ZNU
snATAC-seq GM12878	ENCODE Portal	ENCODE project: ENCSR635OKA
snATAC-seq GM12878	ENCODE Portal	ENCODE project: ENCSR680NPV
DNase-seq GM12878	ENCODE Portal	ENCODE project: ENCSR000EMT
DNase-seq K562	ENCODE Portal	ENCODE project: ENCSR000EOT
Software and algorithms
FIRE	This paper	https://fiberseq.github.io/fire/fire.html
Fibertools	Jha et al.	https://fiberseq.github.io/fibertools/fibertools.html
Mokapot	Fondrie et al.	https://github.com/wfondrie/mokapot
XGBoost	Chen et al.	https://github.com/dmlc/xgboost
Bedtools	Quinlan et al.	https://github.com/arq5x/bedtools2
10X cellranger-atac	10X Genomics	https://www.10xgenomics.com/support/software/cell-ranger-atac
Bedmap	Neph et al.	https://bedops.readthedocs.io/en/latest/content/reference/statistics/bedmap.html
MACS2	MACS2	https://github.com/macs3-project/MACS/wiki/Install-macs2
Homer	Heinz et al.	http://homer.ucsd.edu/homer/motif
Snakemake	Mölder et al.	https://snakemake.readthedocs.io/en/stable/
DeepVariant	Poplin et al.	https://github.com/google/deepvariant
pbsv	PacBio	https://github.com/PacificBiosciences/pbsv
HiPhase	Holt et al.	https://github.com/PacificBiosciences/HiPhase
Meryl	Rhie et al.	https://github.com/marbl/meryl
Jasmine	PacBio	https://github.com/PacificBiosciences/pb-CpG-tools
Verkko	Rautiainen et al.	https://github.com/marbl/verkko
Hifiasm	Cheng et al.	https://github.com/chhylp123/hifiasm
Pbmm2	PacBio	https://github.com/PacificBiosciences/pbmm2
Iso-seq	PacBio	https://isoseq.how/
FIMO	Grant et al.	https://meme-suite.org/meme/doc/fimo.html

Highlights:

• FIRE delivers single-molecule, haplotype-resolved chromatin maps of the diploid genome

• Somatic epimutations create mitotically stable haplotype-selective chromatin

• HLA harbors the densest haplotype-selective chromatin in immune cells

• Promoter-proximal escape elements drive productive transcription from the inactive X