CUT&Tag overcomes biases of ChIP and establishes chromatin patterns for repetitive genomic loci

Summary

New in situ chromatin profiling methods, such as CUT&Tag, have streamlined studies of chromatin features by eliminating the need for up-front purification, but we find that some features are not equally detectable when comparing with previous methods. ChIP-Seq and CUT&Tag identify similar chromatin enrichment profiles for genic loci, such as promoters, but major differences are detected at heterochromatin-associated regions. Unlike ChIP-Seq, CUT&Tag detects robust levels of H3K9me3 over a substantial number of repetitive elements, with especially high sensitivity over evolutionarily young retrotransposons. For example, mouse IAPEz-int elements exhibit strong enrichment using CUT&Tag but underrepresentation using ChIP-Seq. Additionally, several euchromatin-associated proteins, such as RUNX1, co-purify with insoluble heterochromatin in ChIP studies, but are detectible at repetitive elements when applying in situ fragmentation methods. Our study reveals that the current understanding of chromatin states is extensively incomplete, and newer in situ chromatin fragmentation-based techniques are preferred for investigating repetitive elements and retrotransposons.

Graphical abstract

Highlights

• •In situ fragmentation overcomes biases produced by ChIP-Seq
• •Heterochromatic regions of the genome are lost to the insoluble pellet during ChIP-Seq
• •CUT&Tag allows for mapping chromatin features at young repetitive elements
• •Some euchromatin-associated factors bind to insoluble heterochromatin

Genomic analysis; Sequence analysis; Methodology in biological sciences

Introduction

Epigenetic marks and chromatin modifications influence chromatin packaging and regulate gene expression.^1,2 Many of these features are known to play crucial roles in organismal development, and misregulation has been associated with a variety of diseases.^3,4,5 CUT&Tag is a relatively new genomics technique that utilizes a Tn5 transposase to map the genomic location of chromatin modifications.⁶ Tn5 allows users to specifically cleave DNA at target genomic locations that are marked by a certain chromatin feature, without the need for crosslinking or sonication.⁷ Prior studies have demonstrated that CUT&Tag offers increased specificity, increased signal to noise ratios, requires fewer cells as input, and can be more cost effective than ChIP-Seq,⁶ making it an attractive alternative in many situations. While both CUT&Tag and ChIP-Seq are capable of mapping most epigenetic marks, prior studies have uncovered inherent biases caused by the application of ChIP-Seq, potentially limiting the investigation of certain chromatin features.^8,9 For example, input material for ChIP-Seq has been found to be biased for open and accessible regions of the genome, and against condensed loci, potentially due to differences in DNA sensitivity to sonication or crosslinking.^10,11 Whether CUT&Tag or CUT&RUN can overcome such biases remains undetermined.

Many heterochromatic regions of the genome contain repetitive elements or retrotransposons, which remain transcriptionally silent in most tissues to prevent the spreading of mobile DNA elements throughout the genome, which can cause mutations and DNA damage.^12,13 With recent advances in technology and the release of the T2T-CHM13 human genome assembly,^14,15 a renewed emphasis has been placed on the investigation of non-coding DNA sequences, including retrotransposons. Various prior studies have demonstrated that certain retrotransposons play important roles in diverse biological processes, including development, immune response, and neurological function.^{16,17,18,19,20} Additionally, the aberrant expression of repetitive elements has recently been linked with disease states, including cancer.²¹ Thus, establishing a deeper understanding of chromatin states at repetitive elements and retrotransposons is central for advancing biological research across a wide range of fields. Accurately interrogating chromatin states over heterochromatic regions is essential to facilitate forthcoming research into repetitive element function.

Chromatin features, including post-translational modifications to histones and histone variants, are known to be involved in regulating chromatin packaging and gene expression patterns in countless biological systems.^1,2 Certain modifications and variants have been associated with condensed chromatin and transcriptional repression, while others have been associated with accessible regions of the genome that are actively expressed. H3K9me3 (Histone H3 Lysine 9 trimethylation), for example, is one of the most well-studied marks known to reside at constitutive heterochromatin (regions of silent, highly compacted DNA), while H3K27me3 (Histone H3 Lysine 27 trimethylation) is primarily found at facultative heterochromatin (regions selectively silenced in specific cell types or developmental stages).^22,23 The majority of repetitive genomic regions are marked by these repressive modifications in differentiated somatic cells, whereas activating histone modifications can occur when repetitive elements become expressed.^16,21 Activating chromatin features, including H3K27ac (Histone H3 Lysine 27 acetylation) and the histone H2A variant H2A.Z, are typically found over actively expressed euchromatic regions of the genome, such as promoters and enhancers.²⁴ While some features have been observed both at euchromatin and heterochromatin loci, whether they have roles in both activation and silencing remains largely unknown, particularly at repetitive loci.^2,25,26

Cognizant of the established limitations of ChIP-Seq,^8,9,10,11 we wondered whether newer chromatin profiling methods, such as CUT&Tag, might be more effective for investigating heterochromatic loci and repetitive elements. To investigate this possibility, we began by analyzing equivalent ChIP-Seq and CUT&Tag datasets, measuring the enrichment and genomic localization patterns of four separate chromatin modifications, H2A.Z, H3K27ac, H3K27me3, and H3K9me3. We found that similar enrichment profiles were present when comparing analogous ChIP-Seq and CUT&Tag datasets measuring H2A.Z, H3K27ac, and H3K27me3, but this was not the case for H3K9me3. Across several distinct mouse and human cell types, measurements of H3K9me3 enrichment were more robust in datasets generated by CUT&Tag than those generated by ChIP-Seq, which facilitated in-depth analysis of repetitive element chromatin states. These initial studies led us to investigate sources of biases in ChIP-based strategies and to assess whether in situ chromatin fragmentation methods could overcome these shortcomings. Our results reveal that the current understanding of chromatin regulation is severely limited due to deficiencies in ChIP-based methods and provide a straightforward route for improved future investigation.

Results

ChIP-Seq is biased in favor of gene promoters and against intergenic regions

To identify genomic loci which might be preferentially enriched in ChIP-Seq datasets, as explored by others previously,¹⁰ we randomly sub-sampled the genome (100,000 1 Kb randomly selected regions) and partitioned regions into quartiles based on normalized enrichment scores (RPKM) from publicly available ChIP-Seq data, generated from input samples (soluble sonicated chromatin extracted prior to immunoprecipitation) (GEO Accession GSE181069)²⁷ purified from mouse embryonic fibroblasts (MEFs). Using standard peak-calling strategies for identifying enriched regions (see STAR Methods), we partitioned the top 20,000 genomic regions possessing the highest ChIP-Seq input enrichment scores, termed “Top Input,” and assessed overall genomic context. Loci with high relative input scores (putative ChIP-Seq false positives) (Quartile 4 and Top Input regions) were located in closer proximity to gene transcription start sites (TSSs) than regions with lower enrichment scores (Quartile 1) (Figure 1A), and these regions included a relatively large number of gene promoters (Figure 1B). They also possessed differing levels of CpG density (Figure S1A) and had highly accessible chromatin, as measured by ATAC-Seq²⁸ (GEO Accession GSE145705) (Figure 1C). In agreement with these observations, enrichment values for ChIP-Seq input were highly correlated with chromatin accessibility measurements at gene promoters (R = 0.76) (Figures 1D and 1E; Figure S1B). Taken together, these results align with prior studies, which report biases from ChIP, with potential artifacts caused by a preferential selection of euchromatin at the expense of heterochromatic loci.¹⁰

Figure 1.

ChIP-Seq input samples are enriched for promoters and open chromatin

(A) Distance to nearest gene transcription start site (TSS) for ChIP-Seq input (100k random 1 Kb regions divided into quartiles based on input enrichment levels), top 20k ChIP-Seq input sites, or 100k random 1 Kb sites.

(B) Genomic annotation for ChIP-Seq input (100k random 1 Kb regions divided into quartiles based on input enrichment levels), top 20k ChIP-Seq input sites, or 100k random 1 Kb sites. Percentages indicate % of promoter regions.

(C) Heatmaps of enrichment scores (RPKM) for ChIP-Seq input and ATAC-Seq datasets over top 20k input sites.

(D) Heatmaps of enrichment scores (RPKM) for ChIP-Seq input and ATAC-Seq datasets over all annotated mouse promoters.

(E) Genome browser enrichment profiles of ChIP-Seq input and ATAC-Seq shows overlap between the methods.

Wilcoxon test was used to calculate p values. Asterisks key: ∗ <0.01, ∗∗ <0.001, ∗∗∗ <0.0001.

ChIP-Seq and CUT&Tag Display Similar Enrichment Patterns for H2A.Z, H3K27ac, and H3K27me3, but Not H3K9me3

Because in situ chromatin profiling methods, such as CUT&Tag, are methodologically distinct from immunoprecipitation-based techniques, we next wondered whether enrichment profiles generated by CUT&Tag differed from profiles generated by ChIP-Seq. Despite the potential biases of ChIP-Seq, effective enrichment scores could be attained by comparing DNA isolated from immunoprecipitated material with DNA isolated from input samples (scored as -log₁₀ p-values from a Poisson distribution). Using this approach, we compared separate enrichment profiles, and began by investigating the activating chromatin modifications H2A.Z²⁹ (GEO Accession GSE51579) and H3K27ac³⁰ (GEO Accession GSE72239), as well as the repressive modifications H3K27me3³⁰ (GEO Accession GSE72239) and H3K9me3³¹ (GEO Accession GSE53939). CUT&Tag replicates were consistent for all four chromatin features (Figure S2A), and similar enrichment patterns were observed over highly enriched regions (peaks) regardless of technique for H2A.Z, H3K27ac, and H3K27me3 (Figures 2A and 2B). However, regions highly enriched for H3K9me3 were largely distinct when comparing ChIP-Seq and CUT&Tag. The most highly enriched regions of the genome for H2A.Z, H3K27ac, and H3K27me3 tended to occur in relatively close proximity to gene transcription start sites (Figure 2C), and these features were found to be preferentially located within gene regulatory regions such as promoters (Figure 2D). Alternatively, regions found to possess H3K9me3 using ChIP-Seq were located in closer proximity to gene transcription start sites than analogous H3K9me3-enriched regions identified by CUT&Tag. Finally, to directly compare enrichment profiles for ChIP-Seq with profiles from CUT&Tag (and overcome potential differences in data processing), we rank normalized and assessed the degree of correlation between datasets. As anticipated, rank scores over highly enriched loci and gene promoters were found to be well correlated for H2A.Z, H3K27ac, and H3K27me3, but scores for H3K9me3 exhibited particularly low correlations at the most highly enriched CUT&Tag and ChIP-Seq sites (R = 0.190 and 0.151, respectively) and a slightly inverse correlation over promoters (R = −0.012). (Figures S2B and S3). These results demonstrate that ChIP-Seq and CUT&Tag perform similarly for the chromatin marks H2A.Z, H3K27ac, and H3K27me3, which are known to frequently reside over gene promoters, but the two techniques differed for H3K9me3, which is known to reside at regions of constitutive heterochromatin.^{2,22,23,24,25}

Figure 2.

ChIP-Seq and CUT&Tag produce similar chromatin landscapes for H2A.Z, H3K27ac, and H3K27me3, but Not H3K9me3

(A) Heatmaps of enrichment scores from H2A.Z, H3K27ac, H3K27me3, and H3K9me3 CUT&Tag and ChIP-Seq datasets. H2A.Z datasets were plotted over the top 30k H2A.Z CUT&Tag and ChIP-Seq sites, H3K27ac datasets were plotted over the top 30k H3K27ac CUT&Tag and ChIP-Seq sites, H3K27me3 datasets were plotted over the top 10k H3K27me3CUT&Tag and ChIP-Seq sites, and H3K9me3 datasets were plotted over the top 30k H3K9me3 CUT&Tag and ChIP-Seq sites. ChIP-Seq enrichment is scored using -log10(p-values), based on enrichment over input in a Poisson distribution, and CUT&Tag enrichment scoring is based on RPKM.

(B) Genome browser enrichment profiles of H2A.Z, H3K27ac, H3K27me3, and H3K9me3 CUT&Tag and ChIP-Seq datasets.

(C) Distance to nearest gene transcription start site for 30k most highly enriched H2A.Z CUT&Tag sites, 30k most highly enriched H2A.Z ChIP-Seq sites, 30k most highly enriched H3K27ac CUT&Tag sites, 30k most highly enriched H3K27ac ChIP-Seq sites, 10k most highly enriched H3K27me3 CUT&Tag sites, 10k most highly enriched H3K27me3ChIP-Seq sites, 30k most highly enriched H3K9me3 CUT&Tag sites, 30k most highly enriched H3K9me3 ChIP-Seq sites, or 100k random 1 Kb sites.

(D) Genomic annotation for 30k most highly enriched H2A.Z CUT&Tag sites, 30k most highly enriched H2A.Z ChIP-Seq sites, 30k most highly enriched H3K27ac CUT&Tag sites, 30k most highly enriched H3K27ac ChIP-Seq sites, 10k most highly enriched H3K27me3 CUT&Tag sites, 10k most highly enriched H3K27me3ChIP-Seq sites, 30k most highly enriched H3K9me3 CUT&Tag sites, 30k most highly enriched H3K9me3 ChIP-Seq sites, or 100k random 1 Kb sites.

Wilcoxon test was used to calculate p values. Asterisks key: ∗ <0.01, ∗∗ <0.001, ∗∗∗ <0.0001.

Crosslinking and sonication create an over-representation of euchromatin and an under-representation of intergenic heterochromatin

Potential biases in ChIP-Seq might arise through increased sensitivity to sonication at euchromatic loci, by a resistance to sonication at heterochromatic loci, or a combination of both factors. To investigate these possibilities, we performed a mock ChIP-Seq experiment on wild-type primary MEFs, sonicating chromatin to varying degrees and isolating DNA from the soluble fraction, which is commonly used for ChIP experiments, and the insoluble fraction, which is typically discarded. We then compared the DNA purified from each mock ChIP-Seq sample. DNA fragments from insoluble pellet samples (from crosslinked minimally sonicated chromatin) exhibited a much larger size (higher molecular weight) than DNA from soluble supernatant fractions (Figure 3A). To establish whether distinct portions of the genome reside within these separate fractions (potentially underlying biases in ChIP-Seq), we performed Illumina sequencing on isolated DNA, including the minimally sonicated soluble chromatin (Crosslinked Sonicated Supernatant 1 – S1), thoroughly sonicated soluble chromatin (Crosslinked Sonicated Supernatant 3 – S3), and insoluble chromatin (Crosslinked Sonicated Pellet 1 – P1). Similar to our observations from ChIP-Seq input sample measurements, DNA isolated from minimally (S1) and thoroughly (S3) sonicated soluble chromatin was found to be enriched over euchromatic gene promoters, while DNA from the insoluble pellet (P1) was more enriched over intergenic regions (Figure 3B). Additionally, highly enriched minimally sonicated supernatant DNA (S1) tended to localize in close proximity to gene TSSs (Figure 3C), and enriched regions from both supernatant samples (S1 and S3) had high CpG densities (Figure 3D). Higher levels of enrichment were also detected over gene promoters when comparing the supernatant and insoluble pellet samples (Figure 3E), as well as regions previously identified (Figure 1) as enriched in ChIP-Seq input datasets (Figure 3F).

Figure 3.

Crosslinked and sonicated soluble chromatin is enriched for promoters, and insoluble chromatin is enriched for heterochromatic regions

(A) Agarose gel shows DNA extracted from supernatant or cellular debris pellet after a mock ChIP-Seq input experiment. Samples were crosslinked or left as non-crosslinked controls, and sonicated for 0–3 cycles.

(B) Genomic annotation of the top 30k most highly enriched P1, S1, or S3 sites, and 100k random 1 Kb regions.

(C) Distance to the nearest gene transcription start site of the top 30k most highly enriched P1, S1, or S3 sites, and 100k random 1 Kb regions.

(D) CpG density of the top 30k most highly enriched P1, S1, or S3 sites, and 100k random 1 Kb regions.

(E) Heatmaps of P1, S1, and S3 datasets over all the annotated mouse promoters.

(F) Heatmaps of P1, S1, and S3 datasets over the top 20k ChIP-Seq input sites identified in Figure 1.

(G) Heatmaps and profile plots of H3K27ac and H3K9me3 CUT&Tag signal over the most highly enriched P1-specific and S1-specific regions.

(H) Enrichment scores of H3K27ac and H3K9me3 CUT&Tag signal over the most highly enriched P1-specific sites, the most highly enriched S1-specific sites, and 100k random 1 Kb regions.

(I) Genome browser enrichment profiles of P1, S1, and S3 datasets with H3K27ac CUT&Tag, H3K27ac ChIP-Seq, H3K9me3 CUT&Tag, H3K9me3 ChIP-Seq, and ATAC-Seq.

Wilcoxon test was used to calculate p values. Asterisks key: ∗ <0.01, ∗∗ <0.001, ∗∗∗ <0.0001.

We next classified genomic loci based on whether they were purified from the soluble or insoluble pellet samples (termed S1-specific or P1-specific, respectively) (see STAR Methods). In support of our prior observations (Figure 1), we found that the activating histone modification H3K27ac was enriched at S1-specific loci, whereas the repressive histone modification H3K9me3 was enriched at P1-specific loci (Figures 3G–3I). The most highly enriched sites from the soluble chromatin samples were also enriched for H3K27ac, while the most highly enriched sites from the insoluble pellet were enriched for H3K9me3 (Figure S4A). Notably, enrichment scores from samples generated by micrococcal nuclease (MNase)-based genomic fragmentation³² (GEO Accession GSE153939), which is commonly utilized in native ChIP-Seq^7,33 and CUT&RUN³⁴ experiments, were statistically significant but only moderately different in magnitude (median RPKM = 3.113 and 3.468 respectively) comparing between S1 and P1-specific regions, suggesting that MNase-based methods do not suffer from the same biases as standard ChIP-Seq approaches (Figures S4B and S4C). Together, these results indicate that biases in our mock ChIP-Seq experiment arose due to the combination of open/accessible genomic loci being over-represented in the soluble fraction and inaccessible/heterochromatic loci being over-represented in the insoluble pellet fraction.

CUT&Tag Identifies H3K9me3 at young repetitive elements that are undetectable by ChIP-Seq

Our mock ChIP experiments indicated that inaccessible intergenic loci tend to be preferentially excluded from ChIP-Seq assays (Figure 3), potentially explaining the dramatic differences we observed when comparing H3K9me3 patterns obtained from ChIP-Seq with results obtained from CUT&Tag (Figure 2). We next speculated that specific genomic loci might be particularly sensitive to these biases, rendering them undetectable by ChIP-Seq and only detectable by CUT&Tag. To identify such regions, we performed k-means clustering on a combined set of regions identified as enriched in either CUT&Tag or ChIP-Seq. We identified three discrete clusters, including two with higher H3K9me3 enrichment levels from CUT&Tag (Clusters 1 and 2 – C1 & C2) and one with higher H3K9me3 levels from ChIP-Seq (Cluster 3 – C3) (Figure 4A). A similar but less pronounced pattern of enrichment was detected when we assessed H3K27me3 enrichment using an analogous clustering strategy, but no such differences were observed for H3K27ac or H2A.Z (Figure S4D). Prior studies have demonstrated that intergenic repetitive elements and retrotransposons are commonly marked by H3K9me3.^11,12,13,22 Interestingly, both C1 and C2 clusters in our H3K9me3 comparisons (regions with high enrichment levels in CUT&Tag) possessed a high abundance of LTR family retrotransposons (Figure 4B).

Figure 4.

CUT&Tag identifies H3K9me3 at evolutionarily young LTRs

(A) Heatmaps of H3K9me3 CUT&Tag and ChIP-Seq datasets over a union file of all the most highly enriched CUT&Tag and ChIP-Seq H3K9me3 sites, sorted by k-means clustering (C1-C3).

(B) Genomic annotation of the repetitive elements enriched in each cluster (C1-C3).

(C) Rank score plot depicts the enrichment of various repetitive element families in H3K9me3 CUT&Tag and ChIP-Seq. LTR class elements are labeled in blue, and LINE class elements are labeled in orange. Number of repetitive elements identified in either H3K9me3 dataset are depicted with various sized points based on their abundance in these datasets.

(D) Profile plots of H3K9me3 CUT&Tag and ChIP-Seq datasets over all RLTR10-int, RLTR6_Mm, IAPEz-int, and L1_Mus3 elements.

To gain insight into which specific LTR transposons might be impacted by ChIP biases, we next performed rank scoring of all uniquely named repetitive elements in the mouse genome and then subtracted ChIP-Seq rank scores from CUT&Tag rank scores, resulting in a single value for each uniquely named repetitive element subtype. Elements receiving a strong negative score possessed high levels of H3K9me3 specifically in ChIP-Seq datasets, whereas regions with a strong positive score possessed high levels of H3K9me3 specifically in CUT&Tag. We also assessed the evolutionary age of each repetitive element type through the use of milliDiv scoring (base mismatches from the consensus repeat sequence in parts per thousand), with lower scores indicating younger elements.^16,35 Strikingly, we found that the majority of young LTR class repetitive elements exhibited very high levels of H3K9me3 specifically in CUT&Tag, including RLTR10-int, RLTR6_Mm, and IAPEz-int elements. Although many LINEs, such as L1_Mus3, possessed higher ChIP-Seq rank scores, they exhibited a lack of H3K9me3 enrichment in both ChIP-Seq and CUT&Tag (Figures 4C and 4D). To confirm that these observations were due to ChIP-Seq specific chromatin isolation methods, as opposed to antibody-related issues, we next assessed enrichment scores generated from our Mock ChIP-Seq experiments (described Figure 2) and from our ChIP-Seq input samples (described in Figure 1). Notably, none of these samples (ChIP Input or Mock ChIP) relied on antibody-based enrichment scoring, and so any differences in enrichment across the genome could likely be attributed to chromatin fragmentation biases. As anticipated, both ChIP-Seq input samples and supernatant samples (from Mock ChIP) exhibited higher enrichment over gene promoters compared with loci harboring young LTR or LINE elements (Figure S4E). Such biases were not observed in the pellet sample (from Mock ChIP), in the MNase-Seq MEF datasets, or in CUT&Tag datasets lacking primary antibodies (GEO Accession GSE254352) (Figure S4E). Taken together, these results indicate that CUT&Tag is capable of identifying H3K9me3 at specific classes of young repetitive elements, which are underrepresented in ChIP-Seq datasets, likely due to chromatin isolation biases.

Independent H3K9me3 ChIP-Seq datasets differ in their ability to interrogate repetitive loci

Having found that CUT&Tag can overcome putative ChIP-Seq biases, we next speculated that separate publicly available datasets may differ in their ability to identify H3K9me3 enrichment over retroviral DNA sequences and repetitive loci, potentially resulting from differences in crosslinking and/or sonication techniques. To investigate this possibility, we analyzed two additional publicly available H3K9me3 ChIP-Seq datasets, acquired independently from separate laboratories. One dataset was generated from MEFs (GEO Accession GSE181069), and the other was generated from day 13.5 mouse embryonic limb tissue, which is the primary tissue source from which MEFs are typically generated (ENCODE ENCSR022DED). All three ChIP-Seq datasets were found to be more correlated with one another than with the H3K9me3 CUT&Tag dataset (Figure S5A). We next compared H3K9me3 enrichment across all three datasets, along with the MEF CUT&Tag dataset. Similar to our prior results (Figure 2), we observed low/modest levels of enrichment for H3K9me3 within the GSE53939 dataset, and almost no enrichment above background for H3K9me3 from the GSE181069 dataset (Figure 5A). For the ENCODE dataset, on the other hand (ENCSR022DED), we detected much better enrichment than the other two ChIP-Seq datasets, exhibiting moderate to high levels of H3K9me3 over regions previously found to be enriched CUT&Tag, as well as young LTR elements (Figure S5B). Notably, however, data from ENCSR022DED also exhibited higher non-specific enrichment levels over randomly selected intergenic background regions, as compared with the other two ChIP-Seq datasets (Figures S5B and 5B). As in our prior analysis, we next performed rank scoring of enrichment over uniquely named repetitive element subtypes, in order to assess potential differences in enrichment compared with CUT&Tag. Here again, CUT&Tag outperformed ChIP-Seq measurements and successfully identified H3K9me3 enrichment over numerous young LTR elements, including RLTR10-int, RLTR6_Mm, and IAPEz-int (Figure 5C). Despite performing better than the other ChIP-Seq datasets (even exhibiting modest enrichment over these repetitive loci) (Figure S5B), the ENCODE dataset (ENCSR022DED) remained less effective for measuring H3K9me3 levels over young LTRs than the CUT&Tag dataset (Figure 5C - bottom).

Figure 5.

ENCODE ChIP-Seq Datasets Identify Low H3K9me3 at Repetitive Elements

(A) Heatmaps and profile plots of H3K9me3 CUT&Tag and three H3K9me3 ChIP-Seq (GSE53939, GSE181069, and ENCSR022DED) datasets over the 30k most highly enriched H3K9me3 CUT&Tag sites.

(B) Genome browser enrichment profiles of H3K9me3 CUT&Tag and three H3K9me3 ChIP-Seq (GSE53939, GSE181069, and ENCSR022DED) datasets.

(C) Rank score plots depict the enrichment of various repetitive element families in H3K9me3 CUT&Tag and ChIP-Seq (GSE181069 and ENCSR022DED) datasets. LTR class elements are labeled in blue, and LINE class elements are labeled in orange. Number of repetitive elements identified in either the H3K9me3 dataset is depicted with various sized points based on their abundance in these datasets.

CUT&Tag identifies H3K9me3, H2A.Z, and H3K27me3 at young repetitive elements in various mouse and human cell lines

To establish additional support for our findings (Figures 4 and 5), we next compared H3K9me3 CUT&Tag data from MEFs with H3K9me3 CUT&Tag data from mouse embryonic stem cells (mESCs) and H3K9me3 CUT&RUN data from MEFs. Here again, specific classes of evolutionarily young repetitive elements, particularly LTRs, were more highly enriched for H3K9me3 than many of the evolutionarily older elements (Figures 6A and 6B). As in our prior results, enrichment for H3K9me3 over RLTR10-int, RLTR6_Mm, and IAPEz-int elements was particularly high in the MEF and mESC CUT&Tag datasets, as well as the MEF CUT&RUN dataset. Additionally, we compared CUT&Tag³⁶ (GEO Accession GSE213350) and ChIP-Seq (ENCODE ENCSR000APZ) data generated from human H1 stem cells and again found that evolutionarily young LTR class transposons possessed higher levels of H3K9me3 in the CUT&Tag datasets than in data generated by ChIP-Seq (Figure S6). Taken together, these results indicate that in situ fragmentation-based methods (such as CUT&Tag or CUT&RUN) can efficiently map many repetitive elements across a variety of cell types, and deficiencies from ChIP-Seq can be effectively overcome with these more recently developed techniques.

Figure 6.

CUT&Tag and CUT&RUN identify H3K9me3, H2A.Z, and H3K27me3 at young repetitive elements in multiple cell types

(A) Scatterplots compare average milliDiv scores with average enrichment scores for repetitive elements marked in H3K9me3 datasets from MEF CUT&Tag, MEF CUT&RUN, and mESC CUT&Tag. LTR class elements are labeled in blue, and LINE class elements are labeled in orange. Number of repetitive elements throughout the entire genome is depicted with various sized points.

(B) Profile plots of H3K9me3 MEF CUT&RUN and H3K9me3 mESC CUT&Tag datasets over all RLTR10-int, RLTR6_Mm, IAPEz-int, and L1_Mus3 elements.

(C) Profile plots of H2A.Z, H3K27ac, and H3K27me3 CUT&Tag datasets over all RLTR10-int, RLTR6_Mm, IAPEz-int, and L1_Mus3 elements.

(D) Genome browser enrichment profile of H2A.Z MEF CUT&Tag, H2A.Z MEF ChIP-Seq, H3K27me3 MEF CUT&Tag, H3K27me3 MEF ChIP, H3K9me3 MEF CUT&Tag, H3K9me3 MEF ChIP-Seq (Pederson), H3K9me3 MEF CUT&RUN, and H3K9me3 mESC CUT&Tag.

While discrepancies between ChIP-Seq and CUT&Tag methods were initially identified through measurements of H3K9me3, the possibility remained that additional chromatin features may be present over repetitive elements, such as young LTR transposons, but they have been largely unexplored due to biases of ChIP-Seq. To investigate this possibility, we returned to our prior measurements of H3K27ac, H2A.Z, and H3K27me3. Remarkably, we found that IAPEz-int possessed moderate levels of H2A.Z in CUT&Tag datasets, while RLTR10-int elements were moderately enriched in both H2A.Z and H3K27me3 CUT&Tag datasets (Figures 6C and 6D). Taken together, these results provide compelling evidence that heterochromatic loci and repetitive elements are restricted to the insoluble chromatin fraction during standard ChIP-Seq experiments, that chromatin profiling methods that utilize in situ chromatin fragmentation are able to overcome these biases, and that our current knowledge of DNA binding proteins or chromatin modifications localized within heterochromatin regions (such as LTR elements) is decidedly incomplete.

Many factors traditionally thought to bind euchromatin co-purify with insoluble heterochromatin

Having demonstrated a clear under-representation of heterochromatic repetitive elements within ChIP-based assays (Figure 6), we next speculated that proteins bound at heterochromatin loci might be unknowingly excluded from downstream analyses. To investigate this possibility, we prepared crosslinked and sonicated chromatin in a manner similar to the aforementioned mock ChIP-Seq experiments, but rather than investigating the DNA portion of supernatant and pellet fractions, we performed mass spectrometry and identified enriched proteins. Here, we identified 834 soluble proteins significantly enriched in the supernatant (p-value <0.05, Log2FC > 0.5) and 1509 proteins significantly enriched in the insoluble pellet (Figure 7A). Intriguingly, gene ontology (GO) analysis revealed an enrichment for proteins involved in nucleic acid binding and chromatin modification in the pellet-enriched fraction, while transmembrane and transporter-associated proteins tended to be enriched in the supernatant (Figure 7B; Tables S1 and S2). Further inspection revealed several proteins with known function in centromeres or the nucleolus to be enriched within the pellet fraction (Figures 7C and 7D), likely due to the highly compact nature of these separate nuclear compartments/structures.^37,38 Several zinc-finger family proteins, which are known to function in heterochromatin binding and repetitive element silencing, were also enriched within pellet samples (Figure 7E).^39,40 In addition to these somewhat expected results, we identified many enriched factors involved in epigenetic silencing or transcriptional activation within the pellet fraction. These included well established silencing factors, such as ATRX, DNMT1, DNMT3A, SIRT6, and UHRF1,^41,42 as well as several factors typically thought to function in gene activation and reside within euchromatin, such as BRD4, JMJD6, KAT2B, and NSD1/2 (Figure 7F).^43,44,45 Perhaps most surprisingly, many well-studied transcription factors with known binding capacity at gene regulatory regions were found to be significantly enriched in the pellet (Figure 7G), including ELF1, YY1, RUNX1, and ETV6.^46,47,48,49 Remarkably, our re-analysis of a publicly available RUNX1 CUT&RUN⁵⁰ dataset (GEO Accession GSE153281) confirmed that RUNX1 does indeed bind to various repetitive element subtypes in human cancerous acute myeloblastic leukemia cells (Kasumi-1). As with our measurement of H3K9me3 genomic localization, evolutionarily young LTR elements were found to be preferentially enriched (Figure 7H). High enrichment occurred over LTR12E, LTR12C, LTR13, and LTR77 elements (Figures 7I and 7J). Taken together, these results indicate that several commonly studied proteins, including several epigenetic components and transcription factors that are traditionally studied in the context of genic euchromatin, are depleted from ChIP-based assays and may have unknown auxiliary functions within heterochromatic repetitive portions of mammalian genomes.

Figure 7.

Pelleted fraction from crosslinked and sonicated samples contains well known euchromatic factors

(A) Scatterplot shows significant pellet-enriched (red) and supernatant-enriched (blue) proteins from a mass spectrometry experiment.

(B) Gene ontology analysis of the annotated and significant pellet-enriched and supernatant-enriched proteins.

(C) Scatterplot shows centromere associated proteins (yellow) that were significantly enriched in the pellet fraction.

(D) Scatterplot shows ribosomal and nucleolus associated proteins (purple) that were significantly enriched in the pellet fraction.

(E) Scatterplot shows zinc finger proteins (orange) that were significantly enriched in the pellet fraction.

(F) Scatterplot shows epigenetic factors (green) that were significantly enriched in the pellet fraction.

(G) Scatterplot shows transcription factors (cyan) that were significantly enriched in the pellet fraction.

(H) Scatterplot compares average milliDiv scores with average enrichment scores for repetitive elements marked in a RUNX1 CUT&RUN dataset. LTR class elements are labeled in blue, and LINE class elements are labeled in orange. Number of repetitive elements throughout the entire genome is depicted with various sized points.

(I) Profile plots of RUNX1 CUT&RUN over all LTR12E, LTR13, and LTR77 elements.

(J) Genome browser enrichment tracks of RUNX1 CUT&RUN.

Discussion

As proposed in prior studies,^10,11 we find ChIP-based strategies to be biased toward accessible regions of the genome. Since we did not observe such biases in datasets generated by CUT&Tag and CUT&RUN, which utilize in situ enzymatic methods to fragment chromatin, it is plausible that the shortcomings of ChIP are due to chromatin purification, crosslinking, and sonication steps.^6,34 It is noteworthy that biases of ChIP-Seq seem to be marginal (and/or mitigated by input normalization) when interrogating activating chromatin modifications, such as H3K27ac, which exhibited similar enrichment patterns for both CUT&Tag and ChIP-Seq datasets in our analyses. Unlike open and accessible genomic regions, the vast majority of loci enriched for H3K9me3 exhibited highly dissimilar enrichment patterns when we compared data generated from ChIP-Seq with CUT&Tag. Since we found that chromatin within the discarded pellet of ChIP samples tended to have higher levels of H3K9me3, as measured by CUT&Tag, we find it likely that many repetitive elements and retrotransposons are missed in many published ChIP-Seq studies, potentially because repetitive loci are more compacted and thus less sensitive to sonication. It is also possible that repetitive genomic loci occupy separate nuclear compartments compared with euchromatic loci, and/or these elements exist in non-soluble (or phase separated) nuclear states that are not easily purified through standard cross-linking and sonication methods. These inferences align with previous reports that genomic regions containing H3K9me3 are difficult to purify and are resistant to sonication.¹¹ There may also be subtle differences when comparing MNase-based techniques, such as CUT&Run or Native ChIP, with Tn5-based techniques, such as CUT&Tag, as MNase is a multi-turnover endo-exonuclease and Tn5 is a single-turnover transposase. Such differences were not explored in this study.

Through comparisons of several distinct H3K9me3 ChIP-Seq datasets, we found that differences in implementation may help to mitigate some biases of ChIP-Seq, but CUT&Tag still outperformed ChIP-Seq for measuring chromatin features at repetitive regions, regardless of the dataset. Additionally, CUT&Tag does not require the same level of optimization that ChIP-Seq does in order to effectively map repetitive elements. We also note that prior research on repetitive element function took advantage of older sequencing technologies, which cannot accommodate longer read lengths or paired-end sequencing, and often utilized distinct antibodies from newer in situ-based fragmentation methods. For these reasons, we suspect that longer sequencing read technologies, combined with CUT&Tag (or CUT&RUN) will further improve our ability to assess chromatin features over repetitive loci, due to improved enrichment and mappability. It will also be critical for future researchers to carefully select antibodies based on empirical verification against known controls and/or established standards. We have provided a supplemental table detailing read length and antibodies used (if available) for datasets analyzed within the current study (Table S3).

While most repetitive elements in the genome are bound by silencing factors, preventing their expression and subsequent spread throughout the genome, at particular times during development, a subset of elements, including evolutionarily young retrotransposons, can function as transcriptional regulators and potentially influence proximal gene expression patterns.^16,21 Here, we demonstrate that CUT&Tag overcomes biases of ChIP-Seq strategies and allows for the investigation of chromatin modifications at what would otherwise be undetectable repetitive regions. These results indicate that our current understanding of chromatin regulation at repetitive elements, or even repetitive element function, may be severely limited. Our measurements of ChIP enrichment discrepancies focused mainly on the repressive mark H3K9me3, which is typically present at silent repetitive elements, but we also observed the presence of H2A.Z and H3K27me3 at LTR elements. Whether additional chromatin features that are typically associated with euchromatin (such as H2A.Z) are also bound at repetitive loci remains an intriguing and unexplored possibility. Prior studies have indicated that chromatin modifications such as H3K27ac and H3K4me3 can function in the activation of certain repetitive loci,^16,21 but it remains unknown how widespread or common this type of regulation takes place. Subsequent research studies are necessary to address this unknown.

As a scientific community, our current understanding of repetitive element regulation and function, as well as protein binding with heterochromatin, has been largely gleaned from decades of ChIP-based studies. With further adoption of in situ chromatin fragmentation methods, we now have the opportunity to expand the knowledge base from which new hypotheses, mechanisms, and models are formulated. We find our mass spectrometry results to be particularly interesting in this regard. While we did identify several proteins with known heterochromatic function within the pellet fraction of our experiment, such as DNMT1 and SIRT6, we also uncovered numerous factors that are not known to bind heterochromatin or influence its transcription, including KAT2B, BRD4, and RUNX1. Importantly, it should be noted that our ability to co-purify repetitive genomic elements along with specific proteins is not sufficient to conclude that co-purified proteins are truly bound to repetitive loci, and additional genomics studies are required. We did, however, find that once such a protein, RUNX1 (using publicly available CUT&RUN dataset) is able to bind at specific repetitive elements in human cancer cells. These results are further supported by prior studies in which RUNX1 loss impacted chromatin accessibility levels across various repetitive elements. Unlike the prior study, our study provides direct evidence for RUNX1 binding to LTR class retrotransposons.⁵¹ It is quite possible that many of the proteins we identified within the pellet fraction are depleted from ChIP studies, especially when bound to insoluble portions of the genome. Thus, the function of these seemingly euchromatic factors within heterochromatin has remained unknown – due to technical limitations. It is our hope that future researchers take note of the ChIP biases we uncovered and revisit the function of proteins that were previously considered to function exclusively within euchromatin loci.

For the vast majority of prior studies that investigated genomic patterns of chromatin features, ChIP-Seq has been the preferred method. Consequently, our results suggest that much of what we know about chromatin regulation over repetitive elements is incomplete, and many unknown factors could be involved in repetitive element or heterochromatin regulation. In addition to extending our knowledge of basic mechanisms, further investigation of repetitive loci using in situ methods could have translational impacts in the context of both development and disease. For example, abnormal H3K9me3 levels have been observed in several cancer types, but the inability to adequately map the landscape of healthy and diseased tissues has made it difficult to precisely determine the role of H3K9me3 in disease.^4,52 Moreover, the use of CUT&Tag and CUT&RUN should enable the research community to achieve a more complete understanding of repetitive element function, and potentially better target chromatin machinery therapeutically. In addition to expanding the assayable portion of the genome, our study offers an approach that could allow forthcoming researchers to investigate the role of what would otherwise be considered euchromatic proteins within more compacted gene-poor genomic loci. With emerging technologies such as CUT&Tag, along with recent efforts to assemble more complete genomes,^14,15 we foresee an impending “golden age” of repetitive element research, which will undoubtedly reveal novel roles for proteins and repetitive elements in a wide range of critical biological processes.

Limitations of the study

There are a series of limitations in this study that should be noted. These include differences in antibodies used for generating separate datasets and our direct comparison between these somewhat distinct datasets. In some cases, specific antibody information for particular datasets used was not available or missing (see Table S3). There are also known differences in sequencing protocols across analyzed datasets (e.g., read length, paired-end vs. single-end), which may affect detection sensitivity across samples and contribute additional confounding factors. As noted, our study does not explore differences in enrichment due to enzymatic chromatin fragmentation methods, such as technical differences between MNase-based (e.g., CUT&RUN) and Tn5-based (e.g., CUT&Tag) methods. Additionally, our study is somewhat limited in scope, having focused only on a handful of chromatin features (H3K9me3, H3K27ac, H2A.Z, and so forth). Whether additional chromatin factors are subject to ChIP-based biases remains only partially understood. In sum, these limitations should be considered when interpreting our results, and offer appreciable potential for follow-up experimentation or deeper investigation.

Resource availability

Lead contact

Further information and requests for reagents or resources should be directed to (and will be fulfilled by) the Lead Contact Patrick J. Murphy (pjm249@cornell.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Sequencing data generated in this study have been deposited at NCBI GEO and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. Mass-Spec data generated in this study can be accessed via ProteomeXchange, the accession number is listed in the key resources table.

Acknowledgments

This work was funded by grants from the National Institutes of Health, R35GM137833 (PJM) and R35GM133462 (MRO).

Author contributions

CUT&Tag datasets were generated by KDC, SH, KEM, and MCA. CUT&RUN datasets were generated by EC. Mock ChIP-Seq input pellet and supernatant datasets were generated by BJP and RFL. Data analysis was done by the BJP. PMV provided conceptual guidance throughout the project. The initial article was drafted by BJP. Edits to the article were made by PJM and MRO. The entire project was jointly supervised by PJM and MRO.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCES	SOURCE	IDENTIFIER
Antibodies
H2A.Z	Active Motif	Cat# 39113; RRID:AB_2615081
H3K27ac	Active Motif	Cat# 39133; RRID:AB_2561016
H3K27me3	Active Motif	Cat# 39155; RRID:AB_2561020
H3K9me3	Active Motif	Cat# 39161; RRID:AB_2532132
Experimental models: Organisms/cell lines
Wild Type Mouse Embryonic Fibroblasts	This study	WT C57B6 e10.5 embryos
Mouse Embryonic Stem Cells	ATCC	Cat# SCRC-1002
Deposited data
H2A.Z CUT&Tag (MEF)	This study	GSE288759
H3K27ac CUT&Tag (MEF)	This study	GSE288759
H3K27me3 CUT&Tag (MEF)	This study	GSE288759
H3K9me3 CUT&Tag (MEF)	This study	GSE288759
H3K9me3 CUT&RUN (MEF)	This study	GSE288759
H3K9me3 CUT&Tag (mESC)	This study	GSE288759
Chromatin fractionation (MEF)	This study	GSE288757
ATAC-seq (MEF)	Murphy et al., 2020	GSE145705
H2A.Z ChIP-Seq (MEF)	Obri et al., 2014	GSE51579
H3K27me3 ChIP-Seq (MEF)	Xie et al., 2017	GSE72239
H3K9me3 ChIP-Seq (MEF)	Pederson et al., 2014	GSE53939
H3K9me3 ChIP-Seq (MEF)	Ghimire et al., 2021	GSE181069
H3K9me3 ChIP-Seq (H1)	Wang et al., 2023	GSE213350
H3K9me3 ChIP-Seq (H1)	Zhang et al., 2020	GSM1003585
MNase-Seq (MEF)	Tartour et al., 2022	GSE153939
RUNX1 CUT&RUN (Kasumi-1 cells)	Stengel et al., 2020	GSE153281
H3K27ac ChIP-Seq (MEF)	Xie et al., 2017	GSE72239
ChIP-Seq Input (MEF)	Ghimire et al., 2021	GSE181069
H3K9me3 ChIP-Seq (Mouse Embryonic Limb Tissue)	ENCODE 2012	GSE82358
No Antibody Control CUT&Tag	Wen et al. 2024	GSE254352
Mass-Spec Fractionation	This Study	ProteomeXchange: PXD067760
Software and algorithms
Bowtie2 (v2.4.1)	Langmead & Salzberg, 2012	https://github.com/BenLangmead/bowtie2
Picard (v2.5.0)	Broad Institute	https://broadinstitute.github.io/picard/
Deeptools (v3.5.4)	Ramírez et al., 2016	https://github.com/deeptools/deepTools
Dplyr (v1.1.4)	R package	https://github.com/tidyverse/dplyr
R studio (2025.05.1+513)	R core team	https://www.r-project.org/
Homer annotatePeaks.pl	Heinz et al., 2010	http://homer.ucsd.edu/homer/
UCSC Exe Utilities	UCSC Genome Browser	https://hgdownload.soe.ucsc.edu/downloads.html#source_downloads
Macs2 (v2.2.6)	PyPI	https://pypi.org/project/MACS2/
Cutadapt (v2.6)	Martin, 2011	https://github.com/marcelm/cutadapt/?tab=readme-ov-file
pheatmap (v1.0.13)	Ravio Kolde R Package	https://cran.r-project.org/web/packages/pheatmap/pheatmap.pdf
ClusterProfiler (v4.16.0)	Yu et al., 2012	https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html
ggplot2 (v3.5.2)	Wickham et al., 2016	https://ggplot2.tidyverse.org/
Featurecounts (v2.0.3)	Liao et al., 2014	https://academic.oup.com/bioinformatics/article/30/7/923/232889?login=false

Experimental model and study participant details

Murine cell culture

Primary WT C57B6 MEFs were obtained from embryonic day 10.5 mouse embryos. Cells were tested for mycoplasma and no additional authentication was performed. Cells were grown in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin and were cultured at 37°C.

mESCs were purchased from ATCC (cat. #SCRC-1002) and no additional authentication was performed. They were cultured on mitomycin C-inactivated MEFs at 37°C. Mitomycin C-inactivated MEFs were grown on 0.1% gelatin-coated plates for 1 day prior to mESC seeding. mESCs were grown in KO DMEM supplemented with 15% KO Serum Replacement, 1% Gluta-MAX, 1% NEAA, 0.1% LIF, and 0.1% β-mercaptoethanol.

Method details

CUT&Tag

pA-Tn5 was purified and loaded with sequencing adaptors as previously described.⁶ CUT&Tag experiments were conducted using the protocol developed by Kaya-Okur et al.⁶, described at https://www.protocols.io/view/bench-top-cut-amp-tag-kqdg34qdpl25/v3. Samples were not crosslinked, and nuclei were extracted with Nuclear Extraction Buffer. Following nuclear extraction, samples were placed in a Cryo 1°C Freezing Container (Nalgene #5100-0001) and stored at -80°C until use. Each sample was resuspended in 50 μL antibody buffer, and 1 μL of primary antibody was added to each sample. The samples were placed on nutator and incubated overnight at 4°C. On the second day, each sample was resuspended in 100 μL Dig-wash buffer and 1 μL of secondary antibody were used for all samples. The samples were placed on nutator and incubate at room temperature for one hour. Each sample was resuspended in Dig-300 buffer and 1 μL of pA-Tn5 (157 μg/mL) loaded with sequencing adaptors was used to tagment each sample with one hour incubation at room temperature. Following chloroform extraction and ethanol precipitation, each DNA pellet was resuspended in 25 μL of RNase Solution (400 μL of autoclaved dH₂O + 1 μL of RNase A (20 mg/mL, PureLink #12091-021)) and incubated for 10 minutes at 37°C. Purified DNA samples were then stored at -20°C until PCR amplification and sequencing.

CUT&RUN

CUT&RUN experiments were conducted using the CUTANA™ ChIC/CUT&RUN Kit from Epicypher according to the manufacturer’s protocol, as described at https://www.epicypher.com/wp-content/uploads/2024/10/cutana-cutrun-protocol-2.1.pdf. Bead activation buffer was used to wash Concanavalin A beads. Approximately 500,000 WT MEFs were harvested and centrifuged for 3 minutes at 600xg at room temperature. Each sample was resuspended in 100 μL wash buffer for total of two washes. Combined 100 μL washed cells with 10 μL bead slurry with gentle vortex. After removing the supernatant, 50 μL antibody buffer and 1 μL antibody were added separately to each sample with gentle vortex, and the samples were incubated on nutator at 4°C overnight. On the second day, leave the beads on magnetic rack, 200 μL digitonin buffer was added to wash beads for total of two washes. 50 μL cold digitonin buffer and 2.5 μL pAG-MNase were added to each sample with gentle vortex, and the samples were incubated at room temperature for 10 minutes. Digitonin buffer was used to wash each sample for total of two washes. 50 μL digitonin buffer and 1 μL 100mM CaCl2 were added to each sample with gentle vortex, and samples were incubated on nutator at 4°C for 2 hours. Stop buffer was added to each sample and incubate at 37°C for 10 minutes in thermocycler. The samples were placed on magnet stand and transfer supernatant to 1.5mL tube. CUTANA DNA Purification Kit was used to purify DNA and 12 μL elution buffer was used to elute DNA. Qubit fluorometer was used to measure DNA concentration. The NEBNext Ultra_TM II Library Prep Kit for Illumina was used to prepare Illumina library and Bioanalyzer was performed to analyze DNA size range.

Chromatin fractionation and library construction for mock ChIP sequencing

MEFs were grown to confluency in a 10 cm plate. Cells were washed with 5 mL of PBS and then crosslinked with 1% paraformaldehyde (Pierce #28906) for 10 minutes at RT in PBS. 125 mM glycine was used to quench crosslinking. Cells were then washed with 5 mL of PBS and suspended in 2 mL of 1% SDS Lysis Buffer (83 mM Tris-HCl; 167 mM NaCl; 1.1% Triton X-100; 0.05% SDS; 1x Protease Inhibitor (Pierce #A32963); in autoclaved dH₂O) for 10 minutes at RT. Aliquots of samples were then sonicated for 0, 1, 2, or 3 cycles (pulse = 10s; rest = 20s; amplitude = 30%; 5 min on), keeping tubes on ice between cycles. Samples were centrifuged at 3,000xg for 10 minutes at RT and then separated into supernatant and pellet fractions. 5 μL of Proteinase K (>600 U/mL, ∼20 mg/mL, Thermo Scientific #EO0491) + 20 mM EDTA was added to each supernatant sample, and each pellet was resuspended in 500 μL of 1% SDS Lysis Buffer + 5 μL of Proteinase K (>600 U/mL, ∼20 mg/mL, Thermo Scientific #EO0491) + 20 mM EDTA. Pellet samples were then broken up with a 20-gauge syringe. All supernatant and pellet samples were vortexed to mix and incubated for 1 hour at 50°C. 1% SDS was added to each sample, and the pellet samples were again broken up with a 20-gauge syringe. All samples were then incubated overnight at 65°C. DNA was then purified using standard phenol-chloroform extraction followed by alcohol precipitation. Purified DNA samples were stored at -20°C until sequencing adaptors were added. For supernatant samples, 10 ng of DNA was mixed with dH₂O up to 25 μL, and sequencing adapters with barcodes (NEBNext Multiplex Oligos for Illumina #E7600S) were added using the NEBNext library preparation kit (cat. #E7645), as per the manufacturer’s instructions described at https://www.neb.com/en-us/products/e7645-nebnext-ultra-ii-dna-library-prep-kit-for-illumina. For DNA isolated from pellet samples, we applied a rapid tagmentation-based library construction approach, as described previously by others,⁵³ with minor modifications. Libraries were constructed from purified DNA in 50 μL of 1X Tagmentation Buffer (10 mM Tris; 5 mM MgCl₂; 2.5% dimethylformamide; 33% PBS; 0.1% Tween20; in autoclaved dH₂O) + 1 μL Tn5 and incubated at 37°C for 30 minutes at 1000 RPM. 0.2% SDS was then added to quench tagmentation, and samples were incubated at room temperature for 5 minutes. PCR was then used for adding i5 and i7 barcodes.

Next-generation sequencing

CUT&Tag, CUT&RUN, and mock ChIP samples were pooled and sequenced either by NovoGene or the UR-Genomics Research Center, using short-read Illumina next generation sequencing platforms. Raw and processed sequencing data generated in this study can be found at NCBI GEO with the accession numbers GSE288757 (Mock ChIP, chromatin fractionation) and GSE288759 (CUT&Tag and CUT&RUN).

Mass spectrometry

MEFs were prepared for mass spectrometry similarly to the mock ChIP experiments described above. All samples were sonicated for 2 cycles, and Proteinase K was not included in any of the buffers. 200 μM NaCl was added to all samples during the overnight 65°C incubation to reverse crosslinks, and pellet samples were again broken up with a 20-gauge syringe following this overnight incubation. Samples were concentrated by adding 6x volumes of ice-cold acetone and incubating for 30 minutes. Samples were centrifuged at 15,000xg for 5 minutes. Supernatant was discarded and samples were air dried for 5 minutes. Samples were then solubilized and run on a 4-12% SDS-PAGE gel. The gel was stained with SimplyBlue SafeStain (Invitrogen) and washed overnight. Gel slices were excised, cut into 1mm cubes, and destained. The destained gel slices were reduced with DTT (Sigma) and alkylated with IAA (Sigma), and then dehydrated with acetonitrile. Trypsin (Promega) was diluted to 10 ng/μL in 50 mM ammonium bicarbonate and used to cover the dehydrated gel slices. The slices were incubated in the trypsin for 30 minutes at room temperature. Additional ammonium bicarbonate was added until the gel pieces were completely submerged, and the gel pieces were then incubated overnight at 37°C. The next day, peptides were extracted from the gel slices by adding 50% acetonitrile and 0.1% TFA, and then dried using a CentriVap concentrator (Labconco). Desalting was performed with homemade C18 spin columns, followed by drying, and reconstitution in 0.1% TFA. A fluorometric peptide assay (Thermo Fisher) was used to determine the final peptide concentrations. The extracted peptides were then used for mass spectrometry experiments. Peptides were injected onto a 75 μm x 2 cm trap column (Thermo Fisher) and then refocused on an Aurora Elite 75 μm x 15 cm C18 column (IonOpticks) using a Vanquish Neo UHPLC (Thermo Fisher) attached to an Orbitrap Astral mass spectrometer (Thermo Fisher). Solvent A used for these experiments was 0.1% formic acid in water, and solvent B was 0.1% formic acid in 80% acetonitrile. Ions were added to the mass spectrometer with an Easy-Spray source operating at 2 kV. The solvent gradient started at 1% solvent B and increased to 5% solvent B over 0.1 minutes. The solvent gradient further increased to 30% solvent B in 12.1 minutes, 40% solvent B in 0.7 minutes, and finally 99% solvent B in 0.1 minutes. The gradient was held at 99% solvent B for 2 minutes to wash the column (total runtime 15 minutes). The column was re-equilibrated with 1% solvent B between each mass spectrometry run. The Orbitrap Astral was used in data-independent acquisition (DIA) mode, and MS1 scans were acquired in the Orbitrap at a resolution of 240,000. The maximum injection time was 5 ms covering a range of 380-980 m/z. DIA MS2 scans were acquired in the Astral mass analyzer using a 6 ms maximum injection time with variable windowing (4 Da from 380-750 m/z and 6 Da from 750-980 m/z). The HCD collision energy was 28%, and the normalized AGC was 500%. Fragment ions were acquired from 150-2000 m/z with a cycle time of 0.6 seconds.

Quantifications and statistical analyses

Bioinformatic analysis

Raw mass spectrometry data were processed using DIA-NN version 1.8.1 (https://github.com/vdemichev/DIA-NN) using library-free analysis mode.⁵⁴ The Mus musculus UniProt ‘one protein sequence per gene’ database (UP000000589_10009, downloaded 4/7/2021) was used to annotate the dataset while enabling ‘deep learning-based spectra and RT prediction’. Precursor ion generation settings included a maximum of 1 missed cleavage, a maximum of 1 variable modification for Ox(M), a peptide length range of 7-30, a precursor charge range of 2-4, a precursor m/z range of 380-980, and a fragment m/z range of 150-2000. Quantification was performed with ‘Robust LC (high precision)’ mode, using RT-dependent normalization, MBR enabled, protein inferences set to ‘Genes’, and ‘Heuristic protein inference’ turned off. Mass tolerances and scan window sizes were automatically determined by the software. Precursors were filtered at a library precursor q-value of 1%, a library protein group q-value of 1%, and a posterior error probability of 50%. Protein quantification was performed using the MaxLFQ algorithm in the DIA-NN R package (https://github.com/vdemichev/diann-rpackage). The number of peptides in each protein group was counted with the DiannReportGenerator Package (https://github.com/URMC-MSRL/DiannReportGenerator).⁵⁵

Publicly available datasets were downloaded from ENA. CutAdapt⁵⁶ was used to trim the adaptor sequences from CUT&Tag datasets with parameters -m 1 -a CTGTCTCTTATA -A CTGTCTCTTATA. Fasta files were aligned to the mouse (mm10) and human (hg38) genomes with Bowtie2.⁵⁷ PICARD MergeSamFiles was used to convert .sam files to .bam files with parameters SO= coordinate CREATE_INDEX=true. PICARD MarkDuplicates was used to remove duplicate reads from all .bam files with parameters REMOVE_DUPLICATES=true CREATE_INDEX=true. Deeptools⁵⁸ BAMcoverage was used to convert .bam files to .bw files with parameters --normalizeUsing RPKM --binSize 10 --extendReads 100. UCSC bigwigtobedgraph was used to convert .bw files to .bedgraph files. UCSC bigWigMerge was used to merge all replicates from each experiment MACS2 bdgcmp was used to calculate ChIP-Seq enrichment scores above background input levels with parameters -m ppois. MACS2 bdgcmp was also used to calculate enrichment scores of pellet and supernatant samples over one another with parameters -m logFE -p 10. MACS2 bdgpeakcall was used to call peaks on all datasets with parameters -g 100 -l 100. Various -c values were used with MACS2 bdgpeakcall to generate roughly 10k or 30k peaks, and the resulting peak sets were trimmed to exactly the top 10k or 30k locations in R based on RPKM or ppois enrichment scores. For S1-specific and P1-specific peaksets, MACS2 bdgpeakcall was used with parameters -g 100 -l 100 and c = 0.4 on the S1/P1 and P1/S1 bdgcmp files. Genome browser enrichment profiles were generated with IGV. HOMER⁵⁹ annotatePeaks was used to determine genomic annotations for the most highly enriched regions in each dataset, as well as distance to nearest TSS and CpG density using parameters -CpG and mm10 or hg38 genomes downloaded from HOMER. Deeptools multiBigwigSummary was used with parameters BED-file and --outRawCounts to calculate enrichment scores that were then assigned a rank, and ranks were used along with the pHeatmap R package to generate rank-normalized heatmaps with the parameters cluster_rows= FALSE, cluster_cols = FALSE, col = colorRampPalette(c(“lightblue”, “darkblue”))(256). Genomic loci were sorted in these heatmaps based on average rank scores across the two datasets analyzed. Deeptools multiBigwigSummary was also used with parameters BED-file and --outRawCounts to calculate enrichment scores that were used in making scatter plots and RPKM boxplots in R. Scatter plots were made using the R packages ggplot2 and ggpointdensity with the parameters geom_pointdensity(alpha=0.1, size = 3) + scale_color_gradient(low=“#041370”, high=“#FFFF00”) + theme_bw() + theme(panel.grid.major = element_blank(), panel.grid.minor = element_blank(). Deeptools computeMatrix was used to generate .gz files for downstream analysis, and the scale-regions parameter was added for analysis of repetitive elements. Deeptools plotHeatmap was used to generate standard heatmaps using parameters --missingDataColor white --colorList “white,red,maroon,purple” --yMin 0, as well as desired –yMax and –zMax values. Deeptools plotProfile was used to generate profile plots over specific repetitive elements. Deeptools plotCorrelation was used to generate the spearman correlation plot with parameters --corMethod spearman --whatToPlot heatmap –plotNumbers. Bedtools intersect was used to determine overlapping regions of datasets with parameters -wa | uniq. The dplyr and scales R packages were used to filter datasets by repeat name, as well as calculate number of repeats, average milliDiv score, and average ranks for each repeat family using parameters filter, group_by, and summarise plot was used in R to generate rank score difference scatter plots with parameters pch=16, col=rgb(0,0,0,0.2), cex=AdjN. Adjusted N values were calculated based on number of repeats calculated by dplyr, with n < 10 = 0.1, 10<n<200 = n/100, and n<200 = 2. LTRs were labeled with points in R using parameters pch=16, col=rgb(0,0,1,0.6), cex=AdjN. LINEs were labeled with points in R using parameters pch=16, col=alpha(“darkorange”, 0.6), cex=AdjN. featureCounts⁶⁰ in the Rsubread package was used to generate the counts for the average milliDiv vs log10(Average Reads per Copy) plots using the parameters -O -p and -a (RepeatMasker), and then adding a pseudocount of 1. Gene ontology analysis was conducted using Gene IDs for proteins that were significantly (p.value < 0.05) enriched in the pellet or supernatant fractions (log2 fold change > 0.5). Analysis was conducted in R using the clusterProfiler⁶¹ package, with parameters OrgDb = “org.Mm.eg.db”, ont = “MF”, readable = TRUE, fun = enrichGO, qvalueCutoff = 0.05. Gene ontology plots were produced in R with the ggplot2 (Wickham et al., 2016 ) package with parameters aes(Count, fct_reorder(Description, Count))) + facet_grid(“∼Cluster”) + geom_segment(aes(xend=0,yend=Description)) + geom_point(aes(color=p.adjust, size=GeneRatio∗100)) + scale_color_gradientn(colours=c(“#f7ca64”, “#46bac2”, “#7e62a3”), trans=“log10”, guide = guide_colorbar(reverse = TRUE, order = 1)) + theme(panel.border = element_blank(), panel.grid.major = element_line(linetype = ‘dotted’, colour = ‘#808080’), panel.grid.major.y = element_blank(), panel.grid.minor = element_blank(), axis.line.x = element_line()) + scale_size_continuous(range=c(1,5)) + guides(size = guide_legend(override.aes = list(shape=1))).

Statistical methods

For the boxplots, asterisks indicate the significant differences between samples. Wilcoxon tests were used to calculate p values (p<0.0001). Asterisks key: ∗ < 0.01, ∗∗ < 0.001, ∗∗∗ < 0.0001. Boxes represent interquartiles (25%-75%), whiskers represent outer quartiles (0-25% and 75-100%), and solid black bars depict median values. The number of genomic loci included within boxplots is indicated in the accompanying heatmaps (provided either in the panel title, or along the y-axis), as well as within text descriptions, found throughout the results section. The statistical parameters are also described in figure legends. Either Pearson or Spearman correlation test were performed for correlational analysis and rank normalization heatmaps, as specified. Gene ontology was analyzed using ClusterProfiler with default setting and adjusted p values provided. For all differential enrichment tests or gene ontology tests adjusted p values were used for multiple test correction.

Published: October 13, 2025