Genome folding and nuclear speckles converge to orchestrate fibroblast activation

Abstract

Fibroblasts adopt diverse cell states in response to inductive cues to maintain tissue homeostasis. We leveraged this cell-state plasticity to define how genome organization contributes to changes in cellular identity. We show that TGF-β reconfigures topologically associating domains and chromatin loops at genes upregulated during fibroblast activation into myofibroblasts. Cohesin is required for gene induction during fibroblast activation, and enhanced cohesin stability is sufficient to bypass TGF-β signaling and drive a myofibroblast-like state. Emerging evidence suggests chromatin spatial positioning relative to nuclear speckles can regulate gene expression. Therefore, we examined the role of the critical nuclear speckle component SON and showed that it is required for myofibroblast gene expression. Notably, enhancing genome folding partially rescued gene expression in fibroblasts with SON-depleted nuclear speckles. Together, these findings advance our understanding of fibrosis and support a model in which distinct facets of genome organization converge to orchestrate cell-state transitions.

INTRODUCTION

The functional diversity of human tissues depends on the progressive lineage restriction of progenitor cells into hundreds of specialized cell types, which, in turn, can adopt distinct cell states. The advent of bulk and single-cell transcriptomics has underscored the importance of coordinated gene expression programs in defining cell identity (1, 2). Consequently, substantial efforts have been made to understand changes in cell identity via niche signaling and lineage-specific transcription factors. Indeed, the ectopic expression of transcription factors has enabled the directed differentiation of progenitors in vitro and even the reprogramming of differentiated cells into pluripotent stem cells (3–6). Yet, these conversions are often heterogeneous, inefficient, and the resulting cells frequently retain molecular hallmarks of their original identities (7, 8). Thus, current models focused solely on niche signals and combinations of transcription factors fail to fully reconcile how a limited set of transcriptional regulators and morphogens coordinate gene expression genome-wide to give rise to hundreds of distinct, stable cell types.

Three-dimensional genome organization has emerged as a key mechanism underlying coordinated changes in gene expression (9, 10). One facet of genome organization involves the spatial positioning of chromatin relative to nuclear landmarks (10, 11). Among these landmarks, nuclear speckles—nuclear bodies rich in transcriptional machinery and RNAs—amplify the expression of inducible genes (12–17). However, the functional importance of nuclear speckles in physiologically relevant models remains poorly understood. A second facet of genome organization is the partitioning of chromatin into intervals of increased self-interaction via genome folding. Cohesin—a ring-shaped protein complex—plays an essential role in this process by extruding chromatin through its central opening, giving rise to chromatin loops and topologically associating domains (TADs) (18–25). Human genetics underscores the essential role of cohesin in development. Mutations in genes encoding cohesin subunits or regulators cause Cornelia de Lange Syndrome, yet gene expression is largely stable following cohesin loss in homeostatic cells (22, 26, 27). Thus, it remains an open question whether and how these facets of genome organization converge to coordinate gene expression, particularly during cell-state changes.

Fibroblasts are highly plastic cells that adopt diverse cell states in response to physiologic or pathologic cues (28). Upon TGF-β stimulation, fibroblasts “activate” and acquire a contractile myofibroblast phenotype (29–32). Here, we show that TGF-β stimulation resulted in widespread changes in intra-TAD interaction frequency and chromatin looping, which correlated with changes in gene expression. Loss of NIBPL and BRD4—two key positive regulators of cohesin activity—demonstrated that cohesin activity was required for the induction of myofibroblast gene expression and cellular contractility. Conversely, loss of a negative cohesin regulator WAPL is associated with enhanced cohesin stability and was sufficient to induce a myofibroblast-like state in the absence of TGF-β. Notably, cohesin activity was essential for the induction of genes expressed at low levels in unactivated fibroblasts, prompting us to examine the role of nuclear speckles. Loss of SON, a key structural component of nuclear speckles, disrupted nuclear speckles and impaired myofibroblast gene expression. Strikingly, enhancing cohesin stability via loss of WAPL partially rescued myofibroblast gene expression in SON-depleted fibroblasts, suggesting that the cell can leverage different facets of genome organization in a compensatory manner to tune gene expression. Together, these findings support a model in which distinct facets of genome organization converge to orchestrate coordinated changes in gene expression during a cell-state change.

RESULTS

Fibroblast activation restructures the genome

TGF-β induces transcriptional and phenotypic changes in tissue-resident fibroblasts, promoting their transition into contractile myofibroblasts in vivo. To establish a model for studying genome organization during cell-state changes, we first characterized the in vitro activation kinetics of IMR90 fibroblasts (Fig. 1A). Upon serum starvation and TGF-β stimulation (hereafter ‘activation’), the TGF-β-responsive transcription factors SMAD2 and SMAD3 translocated to the nucleus (Fig. 1B). Accordingly, expression of the myofibroblast genes MEOX1, CCN2, CNN1, and ACTA2 significantly increased 24 hours after activation, as measured by qRT-PCR (Fig. 1C) (33, 34). Consistent with these transcriptional changes, immunofluorescence and western blotting showed a modest accumulation of smooth muscle actin (SMA) protein 24 hours after activation, which rose substantially thereafter (Fig. 1D and fig. S1A). To assess phenotypic changes associated with fibroblast activation in vitro, we seeded fibroblasts in collagen gels and compared matched unactivated and activated groups. Activated fibroblasts compressed the gels more robustly at 48 and 72 hours compared to matched controls, a hallmark of the myofibroblast phenotype (Fig. 1E) (28, 33).

Fig. 1. Fibroblast activation is accompanied by widespread genome reorganization.

(A) Schematic of the fibroblast to myofibroblast transition. Created in BioRender. Gardner, Z. (2026) https://BioRender.com/j0363lj). (B) SMAD2/3 immunostaining in unactivated and activated fibroblasts (24 hours) and quantification (n > 60 images per condition, n = 3 biological replicates). Points, image means; horizontal bars, condition means; scale bar, 5 μm. (C) qRT-PCR of indicated genes in unactivated and activated fibroblasts (24, 48, or 72 hours; n = 3 biological replicates). Points, biological replicates; bars, mean ± SEM. (D) SMA immunostaining in unactivated and activated fibroblasts (24, 48, or 72 hours) and quantification of mean intensity per cell (n > 50 images per condition, n = 2 biological replicates). Points, image means; horizontal bars, condition means; scale bars, 50 μm. (E) Contraction assays of fibroblasts seeded in collagen gels and quantification of relative percent contraction (n = 5 biological replicates). Points, biological replicates; bars, mean ± SEM. (F) Aggregate plots of TADs grouped into pentiles by change in mean separation score between unactivated and activated fibroblasts (24 hours). ∆ = mean_activated-mean_unactivated; color bar, normalized contact enrichment or ratio. (G) PCA of RNA-seq from unactivated and activated fibroblasts (24 hours). Points, biological replicates; axes labeled with percentage of explained variance. (H) Volcano plot of gene expression across unactivated and activated fibroblasts (24 hours). Points, individual genes; shaded rectangles, DEG thresholds (|log₂(fold-change)| > 1.5, Benjamini-Hochberg corrected P value < 0.05. (I) Heatmap of hierarchically clustered DEGs between unactivated fibroblasts and activated fibroblasts (24 hours). Rows, genes; columns, biological replicates; color bar, row-scaled expression. (J) Bar plot of the fraction of TADs in each pentile (F) containing either upregulated (top) or downregulated (bottom) DEGs [(H) and (I)]. [(B) to (E)] Two-sided t test corrected using Benjamini-Hochberg method.

Having established a robust model of fibroblast activation, we tested whether higher-order genome folding is remodeled during this cell-state transition. We performed in situ Hi-C on unactivated fibroblasts and fibroblasts activated for 24 hours, representing an early stage of activation (n = 2 biological replicates). Each replicate was sequenced to an average depth of ∼850 million read pairs. Replicates were highly concordant and pooled in equal proportions for all subsequent analysis (fig. S1B and table S1) (35).

We first assessed changes in genome folding by identifying TADs—contiguous genomic intervals characterized by increased chromatin interactions—in unactivated fibroblasts. For each TAD, we calculated the mean separation score, a measure of intra-TAD contact enrichment, in unactivated and activated fibroblasts (36). TADs were then ranked and sorted into pentiles based on the difference in mean separation score between conditions (table S1) (36). Pentile 1 TADs lost interaction frequency upon activation (weakened), while pentile 5 TADs gained interaction frequency (strengthened) (Fig. 1F). In contrast to the changes in TAD strength, global TAD characteristics such as TAD size, TAD number, and mean separation score across TAD boundaries remained stable (fig. S1, C to F). Since cohesin mediates TAD formation, we also performed ChIP-seq for the core cohesin subunit RAD21 in activated fibroblasts (19, 23, 24, 37, 38). We detected greater RAD21 enrichment over CTCF-bound sites at TAD boundaries—key architectural features that stabilize chromatin topology—and within the interior of strengthened (pentile 5) TADs compared to weakened (pentile 1) TADs in activated fibroblasts (fig. S1G) (19–21, 39, 40).

In parallel, we performed RNA-seq on unactivated fibroblasts and fibroblasts activated for 24 hours and identified 747 downregulated and 544 upregulated genes (|log₂(fold-change)| > 1.5, adjusted p-value <0.05) (Fig. 1, G to I and table S2). To integrate the expression and Hi-C datasets, we stratified TADs based on the expression status of their resident genes. TADs containing upregulated genes were classified as “upregulated,” and those containing downregulated genes as “downregulated.” TADs that harbored both upregulated and downregulated genes were labeled “mixed”, and TADs containing only non-differentially expressed genes (DEGs) or no expressed genes were classified accordingly. A greater proportion of strengthened TADs (pentile 5) contained upregulated genes than weakened TADs (pentile 1, Fig. 1J). Weakened TADs did not show a clear inverse association with downregulated genes (Fig. 1J). Mixed, non-differential, or unexpressed TADs did not show consistent changes in TAD strength, and a small fraction of TADs contained both up- and downregulated genes (fig. S1H). Importantly, the number of expressed genes per TAD was comparable across pentiles, indicating that the observed relationship between increased TAD strength and gene expression was unlikely driven by uneven expressed gene density (fig. S1I). These results show that changes in TAD strength correlate with early gene expression changes during fibroblast activation, prior to measurable changes in contractility.

Tuning cohesin dynamics elicits divergent fibroblast cell states

The above results suggest a connection between fibroblast cell state and genome organization. To further explore this relationship, we modulated cohesin function by depleting BRD4, NIPBL, or WAPL using shRNA (fig. S2, A and B). NIPBL and BRD4 are positive regulators of cohesin; NIPBL is required for cohesin loading and activates its ATPase function to facilitate loop extrusion activity and BRD4 stabilizes NIPBL on chromatin (23, 24, 41–46). In contrast, WAPL promotes cohesin unloading, limiting loop size and cohesin residence time (37, 38, 47–51). To assess the effects of cohesin activity modulation on the fibroblast-to-myofibroblast cell state transition, we performed collagen gel contraction assays. Fibroblasts infected with a shRNA targeting BRD4, NIPBL, WAPL, or a scramble control (shSCR) were embedded in collagen gels and subsequently activated. Depletion of either BRD4 or NIPBL prior to collagen-embedding abrogated gel contraction compared to shSCR, consistent with previous reports using BET inhibition (Fig. 2A) (52). Conversely, WAPL depletion led to robust gel contraction even in the absence of TGF-β relative to shSCR (Fig. 2A). Ki67 immunofluorescence showed only modest changes in the fraction of cycling cells across shRNAs, consistent with partial, as opposed to complete, knockdown of cohesin regulators (fig. S2C) (53–55). These findings demonstrate a critical role for positive and negative regulators of cohesin activity in fibroblast contractility.

Fig. 2. Cohesin regulator depletion elicits divergent fibroblast states.

(A) Contraction assays of fibroblasts treated with the indicated shRNAs and then seeded in collagen gels and quantification of relative percent contraction (n = 5 biological replicates). (B) PCA of RNA-seq data from fibroblasts treated with indicated shRNAs and left unactivated or activated for 24 hours. Points, biological replicates; axes labeled with percentage of explained variance. (C) Volcano plot of gene expression across shSCR-treated unactivated and activated fibroblasts (24 hours). Points, individual genes; shaded points, select fibroblast activation genes; shaded rectangles, DEG thresholds (|log₂(fold-change)| > 1.5, Benjamini-Hochberg corrected P value < 0.05). (D) Heatmap of hierarchically clustered DEGs from shSCR-treated fibroblasts (C) across shRNA-treated fibroblasts. Rows, genes; columns, condition averages; color bar, row-scaled expression. (E) Scatter plots of log₂-transformed, length-scaled and normalized expression of cohesin-sensitive DEGs in shWAPL (top, unactivated) or shNIPBL (bottom, activated) versus shSCR [cluster 1 genes, (D)]. Points, genes; dashed blue line, identity line. Axis limits fixed for visualization only. Density plots comparing the distribution of the expected fraction of genes above identity line–generated by randomly sampling genes from all expressed genes 10,000 times–to the observed fraction of cohesin-sensitive genes (dashed red line) above the identity line (two-sided permutation test). (F) Gene ontology of cohesin-sensitive genes. (G) MEOX qRT-PCR in siNTC (control), siNIPBL, or siWAPL-treated fibroblasts (n = 3 biological replicates). Points, biological replicates; bars, mean ± SEM. (H) Boxplot of expressed genes or cohesin-sensitive genes in shSCR-treated unactivated and activated (24 hours) fibroblasts. Grey box, lower quartile of gene expression. y-axis limit fixed for visualization purposes only. [(A) and (G)] Two-sided t test corrected using Benjamini-Hochberg method.

To understand how knockdown of cohesin regulators with opposing functions yielded divergent contractile phenotypes, we examined fibroblast gene expression using RNA-seq (Fig. 2B). We first identified DEGs during fibroblast activation by comparing shSCR-treated unactivated fibroblasts with shSCR-treated fibroblasts activated for 24 hours. Using our previously defined log-fold change and adjusted p-value thresholds, we identified 635 downregulated and 572 upregulated genes (Fig. 2C). We then examined the expression of these 1,207 DEGs in shNIPBL- and shWAPL-treated fibroblasts. Hierarchical clustering yielded four gene clusters with distinct responses to cohesin regulator depletion (Fig. 2D and table S2). Notably, cluster 1 genes are induced by TGF-β often showed opposing responses to cohesin regulator depletion–shNIPBL treatment attenuated their induction upon activation, whereas shWAPL treatment partially induced their expression in unactivated fibroblasts. This cluster of genes included ACTA1, PADI2, and MEOX1–all implicated in fibroblast activation (28, 33, 34, 56, 57). Quantitative analysis revealed 98% of genes in this cluster failed to become induced upon activation in shNIPBL-treated fibroblasts (P < 1 × 10⁻⁴). Strikingly, 75% of cluster 1 genes were induced in shWAPL-treated unactivated fibroblasts compared to shSCR-treated fibroblasts (P < 1 × 10⁻⁴) (Fig. 2E, top and bottom). Similar effects were observed following BRD4 depletion, with 90% of cluster 1 genes failing to be induced upon activation (fig. S2D). Given these opposing effects on gene expression, which mirrored the observed contractile effects, we designated this cluster of genes as “cohesin-sensitive.” Gene ontology (GO) analysis of this cluster showed enrichment for muscle, cartilage, and connective tissue development, consistent with the collagen contraction phenotype observed at a later timepoint (Fig. 2F). We validated this expression pattern for MEOX1 by qRT-PCR using independent siRNAs (Fig. 2G and fig. S2E). Finally, while cohesin-sensitive genes were expressed at relatively low levels in unactivated fibroblasts, their expression increased to comparable levels following activation (Fig. 2H), suggesting that cohesin may control gene induction, consistent with previous reports (58–60).

Given the parallels in gene expression patterns and collagen contractility observed in activated fibroblasts and unactivated WAPL-depleted fibroblasts, we sought to determine whether WAPL depletion activated canonical TGF-β signaling pathways. Immunofluorescence showed that WAPL depletion did not induce nuclear translocation of SMAD2 or SMAD3 (fig. S2F). In addition, SMAD2 depletion did not significantly affect MEOX1 expression in shWAPL-treated fibroblasts (fig. S2G). These data suggested that WAPL knockdown is, at least partially, sufficient to bypass TGF-β signaling during fibroblast activation. Collectively, these perturbations established an essential role for cohesin in mediating the fibroblast-to-myofibroblast cell-state change.

WAPL depletion results in fibroblast-activation-like chromatin looping

Motivated by the observation that WAPL depletion promotes myofibroblast-like gene expression and contractility, we next examined its impact on genome organization. Chromatin fractionation followed by western blotting demonstrated increased RAD21 levels in the chromatin fraction of shWAPL-treated fibroblasts compared to shSCR-treated controls, consistent with enhanced cohesin stability on chromatin (fig. S3A) (47–50). We next assessed how WAPL depletion affects chromatin architecture by performing in situ Hi-C on unactivated fibroblasts treated with either shSCR or shWAPL. As above, we pooled two concordant biological replicates per condition (Spearman correlation ≥0.88 within conditions; 665 million read pairs on average per replicate, fig. S3B and table S1) (35).

Stratifying TADs into pentiles based on the change in mean separation score revealed widespread shifts in TAD strength following WAPL depletion (fig. S3C). These shifts were accompanied by a modest increase in the number of TADs and a decrease in TAD size (fig. S3, D and E and table S2). Consistent with previous reports, we observed a slight increase in the mean separation score across TAD boundaries following WAPL depletion, suggestive of increased TAD-TAD interactions (fig. S3F) (38, 49, 61). RAD21 occupancy was modestly higher at CTCF peaks associated with strengthened TADs compared to weakened TADs in WAPL-depleted cells, with a more pronounced increase at non-boundary sites compared to boundary sites (fig. S3G). TADs were then classified as containing one or more cohesin-sensitive DEGs, one or more non-cohesin-sensitive DEGs, mixed DEGs, no DEGs, or no expressed genes. TADs strengthened following shWAPL treatment were enriched among TADs containing cohesin-sensitive genes. Conversely, we observed no such correlation when analyzing TADs containing non-cohesin-sensitive DEGs, mixed DEGs, non-DEGs, or unexpressed genes (fig. S3, H and I). Together, these findings suggest that changes in TAD strength following WAPL are linked to cohesin-sensitive gene regulation.

Aggregate TAD analysis and representative Hi-C maps revealed enhanced chromatin looping in WAPL-depleted cells compared to control cells, as indicated by the increased number and intensity of corner dots, consistent with published studies (figs. S3C and S4A) (38, 49). We identified 6897 weakened and 13,071 strengthened loops in shWAPL-treated fibroblasts compared to shSCR-treated fibroblasts (Fig. 3A). To test whether shWAPL-dependent loops were associated with changes in gene expression, we first stratified TADs by the number of differentially weakened or strengthened loops they contained (see methods). TADs containing weakened loops showed a graded enrichment for non-cohesin-sensitive DEGs, whereas TADs containing strengthened loops showed a graded enrichment for cohesin-sensitive genes (Fig. 3B). These results indicated that cohesin-sensitive genes preferentially reside in TADs containing new or strengthened chromatin loops following WAPL-depletion.

Fig. 3. WAPL depletion induces genome folding patterns mimicking fibroblast activation.

(A) Aggregate plots of chromatin loops in unactivated shSCR- and shWAPL-treated fibroblasts. Color bar, normalized contact frequency or ratio. (B) Bar plot of the fraction of TADs overlapping zero or a low, medium, or high number of shWAPL differential loops (methods) containing either non-cohesin-sensitive (left) or cohesin-sensitive (right) DEGs. (C) Stacked bar plot of fraction of TADs for each loop-count category for activation-strengthened loops (x-axis) that fall into each loop-count category for shWAPL-strengthened loops (fill). (D) Hi-C matrices of the MEOX1 TAD (5 kb resolution). Arrowheads indicate focal enrichment consistent with chromatin looping. DNA FISH probes and MEOX1 annotated, other genes omitted; color bars, normalized contact frequency. (E) MEOX1 TAD DNA FISH [probes indicated in (D)] in shSCR- or shWAPL-treated fibroblasts left unactivated or activated for 24 hours and quantification of probe-to-probe distances (n > 180 alleles imaged per condition, n = 3 biological replicates; two-sided Wilcoxon rank-sum test corrected using Benjamini-Hochberg). Points, individual alleles; dotted line, median distance in shSCR-treated, unactivated condition; scale bar, 5 μm; inset, 2.5 μm. (F) H3K27ac ChIP-seq in unactivated and activated (24 hours) fibroblasts. (G) qRT-PCR of MEOX1 in fibroblasts treated with CRISPRi+sgEmpty (control) or CRISPRi+3sgRNAs targeting the MEOX1 enhancer (sgEnh. 1–3) left unactivated or activated for 24 hours. (H) H3K27ac immunoblot in unactivated shRNA-treated fibroblasts then treated with DMSO or A485 (10 μM, 6 h). (I) MEOX1 qRT-PCR in unactivated shRNA-treated fibroblasts then treated with DMSO or A485 (10 μM, 6 hours). (J) MEOX1 qRT-PCR in fibroblasts treated with CRISPRi+sgEmpty or CRISPRi+3sgRNAs targeting the MEOX1 enhancer (sgEnh. 1–3) followed by shRNA. [(G), (I), and (J)] Two-sided t test corrected using Benjamini-Hochberg method (n = 3 biological replicates). Points, biological replicates; bars, mean ± SEM.

We next compared genome folding changes during activation to those associated with WAPL depletion. We first analyzed chromatin looping in the unactivated versus activated fibroblast Hi-C dataset. Consistent with the analyses above, we observed a stepwise positive association between TADs harboring weakened loops and downregulated genes, and between TADs harboring strengthened loops and upregulated genes (fig. S4, B and C).

Cross-comparison of the Hi-C datasets further revealed substantial overlap in chromatin loop remodeling. TADs with loops weakened upon activation frequently overlapped TADs with loops weakened following WAPL depletion, while TADs with loops strengthened upon activation similarly overlapped TADs with loops strengthened following WAPL depletion (Fig. 3C and fig. S4D). For example, 73% of TADs lacking loops strengthened after activation overlapped TADs with no or few loops strengthened after WAPL depletion, whereas 74% of TADs with a high number of loops strengthened after activation overlapped TADs with a medium or high number of loops strengthened after WAPL depletion (Fig. 3C). These analyses indicate that WAPL depletion phenocopies aspects of activation-like chromatin looping that correlate with gene expression.

Building on our Hi-C analyses, we designed oligo paint DNA fluorescence in situ hybridization (FISH) probes targeting each end of the TAD containing MEOX1, a locus previously shown to require cohesin for proper folding (Fig. 3D) (62–64). MEOX1 encodes a transcription factor implicated in fibroblast activation, is induced upon activation and WAPL depletion, and fails to be induced following NIPBL or BRD4 depletion (Fig. 1C and fig. S2E) (34). Activation of shSCR-treated cells resulted in a significant decrease in the center-to-center distance between the MEOX1 TAD FISH probes, an effect attenuated by depletion of BRD4 or NIPBL (Fig. 3E and fig. S4E). Consistent with our Hi-C findings, MEOX1 probe-probe distances in unactivated shWAPL-treated cells were comparable to those in activated shSCR-treated cells (Fig. 3E). In contrast, DNA FISH probes targeting the boundaries of a TAD that showed minimal change in our Hi-C datasets (MFSD14B TAD) also showed minimal change in probe-probe distance upon activation (Fig. 1F and fig. S4F). These results point to MEOX1 as a representative locus for studying cohesin-mediated genome folding in fibroblast activation.

Cohesin-mediated chromatin loops have been shown to support the function of a subset of cis-regulatory elements (CREs) (65, 66). We therefore tested whether enhancer activity contributes to gene induction following WAPL depletion. MEOX1 expression depends on a TGF-β-responsive enhancer in cardiac fibroblasts (34). Consistent with this, the MEOX1 enhancer showed increased H3K27ac enrichment upon activation, and CRISPRi targeted to this enhancer diminished MEOX1 induction upon activation (Fig. 3, G and H) (67). Inhibition of p300, an acetyl transferase responsible for depositing H3K27ac, with A485 (10 μM, 6 hours) attenuated both H3K27ac levels and MEOX1 expression in unactivated shWAPL-treated fibroblasts compared to DMSO controls, indicating that shWAPL-induced MEOX1 expression required enhancer activity (Fig. 3, H and I and fig. S4H) (68–70). Strikingly, CRISPRi targeting the MEOX1 enhancer alone also attenuated MEOX1 expression in unactivated shWAPL-treated fibroblasts (Fig. 3J and fig. S3H) (67). The levels of H3K27ac enrichment at activation-associated peaks were not substantially different between shSCR- and shWAPL-treated fibroblasts (fig. S4, I and J and table S4). Together, these results are consistent with WAPL depletion stabilizing cohesin-mediated chromatin loops, enabling an otherwise weak enhancer to drive MEOX1 expression in unactivated fibroblasts.

SON depletion impairs fibroblast activation

Prior studies have found correlations between enrichment of enhancer-related histone post-translational modifications and proximity to nuclear speckles, subnuclear bodies enriched in transcriptional machinery and associated with the robust expression of inducible genes (12–17, 71, 72). Given our finding that cohesin is required for enhancer-mediated gene induction during fibroblast activation (figs. S2F, S4H, and Fig. 3H), we sought to understand whether fibroblast gene regulation might occur in proximity to nuclear speckles. To do so, we first integrated published nuclear speckle proximity data (SON TSA-seq) with publicly available H3K4me1 ChIP-seq dataset and our H3K27ac ChIP-seq dataset. Active enhancers (marked by both H3K4me1 and H3K27ac) were relatively enriched in genomic regions with high SON TSA-seq signal—suggestive of speckle proximity—compared to inactive enhancers (marked by H3K4me1 alone) (Fig. 4A). Next, we found that WAPL depletion disrupted chromatin compartmentalization in our datasets, consistent with previous reports (fig. S5A) (37, 38, 49). Notably, A-compartment regions are predicted to associate with nuclear speckles (10, 11). These observations prompted us to test whether cohesin physically interacts with nuclear speckle proteins. RBM25, a nuclear speckle protein, co-immunoprecipitated with NIPBL, BRD4, RAD21, SMC1a and SMC3 in IMR90 fibroblasts (Fig. 4B). We also verified the RAD21-RBM25 interaction in HCT116 cells (fig. S5B).

Fig. 4. Nuclear speckles are required for cohesin-sensitive gene expression.

(A) Line plot of fraction of ChIP-seq peaks (H3K4me1, ENCODE; H3K27ac, this study) in 250 kb bins sorted into vigintiles by SON TSA-seq signal (PRJNA645124). (B) Immunoprecipitation of indicated proteins and western blotting with indicated antibodies in unactivated fibroblasts. (C) SRRM2 immunostaining and MALAT1 RNA FISH in shRNA-treated, unactivated fibroblasts and mean correlation between SRRM2 and MALAT1 signal per image (n > 35 images per condition, n = 2 biological replicates; two-sided t test). Points, image means; horizontal bars, condition means; scale bar, 5 μm. (D) Contraction assays of fibroblasts treated with indicated shRNAs and then seeded in collagen gels and quantification of contraction (n ≥ 4 biological replicates; two-sided t test corrected using Benjamini-Hochberg method). Points, biological replicates; bars, mean ± SEM. (E) Gene expression across shRNA-treated activated fibroblasts (24 hours). Points, individual genes; shaded points, select fibroblast activation genes; shaded rectangles, DEG thresholds (|log₂(fold-change)| > 1.5, Benjamini-Hochberg corrected P value < 0.05. Downregulated DEG gene ontology. (F to I) SON immunostaining and MEOX1 DNA FISH and quantification of probe to speckle-edge distances (two-sided Wilcoxon rank-sum test corrected using Benjamini-Hochberg method. Points, individual alleles; scale bar, 5 μm; inset scale bar, 2.5 μm. (F) Fibroblasts treated with indicated shRNAs and either left unactivated or activated for 24 hours (n > 260 alleles imaged per condition, n = 3 biological replicates). Dotted line, unactivated shSCR-treated, unactivated condition. (G) Fibroblasts treated with CRISPRi+3sgRNAs targeting the MEOX1 enhancer (sgEnh. 1–3) followed by shRNA (n > 240 alleles imaged per condition, n = 3 biological replicates). (H) Fibroblasts treated with shWAPL and then treated with either DMSO or A485 (10 μM, 6 hours; n > 230 alleles imaged per condition, n = 2 biological replicates). (I) Fibroblasts treated with shWAPL and then treated with either DMSO or triptolide (TPL; 100 nM, 1 h; n > 340 alleles imaged per condition, n = 3 biological replicates).

We next tested whether cohesin-sensitive genes change their spatial proximity to nuclear speckles during fibroblast activation. Immunofluorescence for the nuclear speckle protein SON combined with DNA FISH targeting the cohesin-sensitive genes MEOX1 and PADI2 revealed that the distance from each locus to the edge of the nearest nuclear speckle decreased upon fibroblast activation (fig. S5, C and D). MEOX1 resides within the chromosome 17q21 R-band, a region previously shown to localize near nuclear speckles in fibroblasts (73). In contrast, a control locus MFSD14B, a gene not upregulated in our datasets, did not change its speckle proximity upon fibroblast activation (fig. S5E). Together, these data suggest that select cohesin-sensitive genes reposition relative to nuclear speckles upon fibroblast activation.

To determine whether nuclear speckles were required for fibroblast activation, we depleted SON, a core component of nuclear speckles (74). SON knockdown altered nuclear speckle morphology and number and resulted in dispersion of the speckle-associated lncRNA MALAT1 throughout the nucleus, consistent with previous reports (Fig. 4C and fig. S5, F and G) (75–77). Functionally, shSON-treated fibroblasts embedded in collagen failed to contract under activating conditions (Fig. 4D). Ki67 immunofluorescence showed minimal changes in the fraction of cycling cells following SON knockdown (fig. S5H). RNA-seq analysis identified 808 downregulated and 462 upregulated genes in shSON- compared to shSCR-treated fibroblasts cultured under activating conditions for 24 hours (Fig. 4E and fig. S5I). Gene ontology terms enriched among cohesin-sensitive genes were also significantly enriched among genes downregulated in shSON-treated, activated fibroblasts (Figs. 2F and 4E). These findings demonstrate that SON depletion disrupts nuclear speckle morphology and impairs fibroblast activation.

Nuclear speckles are enriched in transcriptional machinery, suggesting they may support the expression of nearby genes (78, 79). Indeed, SMAD2/3, p300, and RNA polymerase II (CTD phospho-Ser2) showed increased immunofluorescence signal at and around nuclear speckles in activated fibroblasts (fig. S6A). Therefore, we investigated whether the gene expression changes observed following SON depletion could be explained by disruption of upstream processes. SON depletion did not result in dramatic changes in expression of cohesin complex members, cohesin regulators, or transcriptional activators (fig. S6, B and C). Accordingly, DNA FISH probes targeting the MEOX1 TAD boundaries still gained proximity in activating conditions upon SON depletion and SMAD2/3 nuclear translocation remained intact (fig. S6, D and E). Upon further analysis of our RNA-seq data, we did not observe clear differences in the splicing of MEOX1 or other highly expressed cohesin-sensitive genes in shSON-treated activated fibroblasts (fig. S6F).

Instead, our data suggest that MEOX1 expression correlated with its proximity to the edge of the nearest nuclear speckle (14). MEOX1 was lowly expressed in shBRD4- and shNIPBL-treated fibroblasts, and the locus did not gain proximity to nuclear speckles under activating conditions compared to controls. In contrast, the MEOX1 locus was positioned closer to nuclear speckles in shWAPL-treated fibroblasts in which MEOX1 was expressed in both the unactivated and activated states compared to the unactivated control (Fig. 4F). Although shBRD4 and shNIPBL treatment resulted in modest, but significant changes in nuclear speckle morphology and number, shWAPL treatment did not significantly alter these features compared to shSCR treatment (fig. S7A). These results are consistent with a model in which gene expression during fibroblast activation is regulated by cohesin-mediated genome folding and spatial positioning relative to nuclear speckles.

Genome folding and nuclear speckles are distinct pathways required for fibroblast activation

To dissect the relationship between genome folding, gene expression, and nuclear speckles, we asked whether WAPL depletion alters speckle proximity independent of its effects on transcription. Because WAPL depletion induces MEOX1 expression, we attenuated MEOX1 transcription using CRISPRi targeting of the previously described enhancer and analyzed unactivated shSCR and shWAPL cells under these conditions (Fig. 3J). Under these transcriptionally repressed conditions, MEOX1-speckle edge distances were not significantly different between shSCR- and shWAPL-CRISPRi cells (Fig. 4G). In contrast, the distance between DNA FISH probes marking the boundaries of the MEOX1 TAD remained significantly reduced in shWAPL- compared to shSCR-CRISPRi cells (fig. S7B). To enable measurements at shorter timescales, we next pharmacologically attenuated MEOX1 enhancer function. Treatment of unactivated shWAPL-fibroblasts with A485 increased MEOX1-speckle edge distance compared to controls (Fig. 4H), while MEOX1 TAD probe–probe distance remained unchanged (fig. S7C) (80). These data suggested a link between enhancer-dependent gene expression and nuclear speckle proximity and raised the possibility that either the enhancer itself or MEOX1 transcription was required for nuclear speckle proximity. To distinguish between these possibilities, we inhibited transcription in shWAPL-fibroblasts using triptolide for one hour (fig. S7D). Nuclear speckle morphology was unchanged in triptolide-treated cells (fig. S7E). The distance between MEOX1 and the nuclear speckle edge increased following triptolide treatment in shWAPL cells, consistent with previous reports, while the distance between the probes targeting the MEOX1 TAD boundaries remained unchanged (Fig. 4I and fig. S7F) (81, 82). Taken together, these results suggest that WAPL depletion alone is not sufficient to drive MEOX1 proximity to nuclear speckles and that gene activity is required for speckle proximity.

Given that cohesin-mediated genome folding and SON both contribute to gene expression upon activation, we next asked whether cohesin and SON act through linear or separate pathways. To address this, we performed a dual-depletion analysis. Within the limits of a knockdown approach, we reasoned that if these pathways operate linearly, gene expression in the dual depletion would mirror the dominant single depletion. Alternatively, independent pathways would produce additive effects relative to either perturbation alone. Co-depletion of NIPBL and SON attenuated expression of MEOX1 and PROC in activated fibroblasts beyond either NIPBL or SON depletion alone. Conversely, co-depletion of WAPL and SON led to a partial normalization of MEOX1, PADI2, and PROC expression in activated fibroblasts compared to SON depletion alone, but below the levels observed following WAPL depletion alone (Fig. 5A). To extend this observation genome-wide, we returned to our RNA-seq data and found over 80% of cohesin-sensitive genes demonstrated attenuated expression in shSON-treated, activated fibroblasts (Fig. 5B, top). Additional RNA-seq revealed that shSON-shWAPL co-depletion partially rescued the majority of cohesin-sensitive genes compared to shSON alone (Fig. 5, B, bottom and C). Therefore, these results are consistent with a model in which genome folding and nuclear speckles act through distinct pathways that converge to support the expression of a common set of myofibroblast genes.

Fig. 5. Genome folding and nuclear speckles are convergent pathways.

(A) qRT-PCR of indicated cohesin-sensitive DEGs in shRNA treated fibroblasts left unactivated or activated for 24 hours (n = 3 biological replicates; two-sided t test corrected using Benjamini-Hochberg method). Points, biological replicates; bars, mean ± SEM. (B) Scatter plots of log₂-transformed, length-scaled and normalized expression of cohesin-sensitive DEGs [cluster 1 genes, (D)]. Points, genes; dashed blue line, identity line. Axis limits fixed for visualization only. Density plots comparing the distribution of the expected fraction of genes above identity line–generated by randomly sampling genes from all expressed genes 10,000 times–to the observed fraction of cohesin-sensitive genes (dashed red line) above the identity line (two-sided permutation test). (C) Elbow plot of the ranked log₂(fold-change) in gene expression of cohesin-sensitive. Points, individual genes; shaded boxes, genes with log₂(fold-change) < 0 (grey) or genes with log₂(fold-change) > 0 (red). (D) Proposed model depicting genome folding and nuclear speckles converging to tune the expression of myofibroblast genes. Created in BioRender. Gardner, Z. (2026) https://BioRender.com/j0363lj.

DISCUSSION

Here, we show that cohesin-mediated genome folding and nuclear speckles jointly regulate the robust expression of myofibroblast genes during fibroblast activation (Fig. 5D). Changes in genome organization accompanied fibroblast activation and correlated with changes in gene expression (Fig. 1). This observation prompted us to test whether modulating cohesin activity impacts fibroblast cell state. Indeed, depletion of positive cohesin regulators (BRD4 and NIPBL) was sufficient to impair fibroblast activation. Conversely, depletion of a negative regulator of cohesin (WAPL) was sufficient to increase cohesin levels on chromatin and induce fibroblast activation in the absence of activating conditions (Figs. 2 and 3). Further, we showed depletion of the nuclear speckle structural protein SON impaired fibroblast activation and reduced expression of cohesin-sensitive genes (Fig. 4). Genetic interaction experiments further revealed that cohesin- and SON-dependent gene expression regulation are likely distinct molecular pathways that converge on a shared transcriptional program during fibroblast activation (Fig. 5). Taken together, our data support a model in which cohesin-mediated genome folding and nuclear speckle proximity function as tunable, convergent pathways that coordinate changes in gene expression during fibroblast activation (Fig. 5D).

The complexity invoked by this model may reflect a feature—not a flaw—of gene expression regulation during cell-state changes. The dual requirement for proper genome folding and spatial positioning may buffer against noisy or ambiguous inductive cues, enabling the integration of multiple regulatory inputs before the cell commits to a new gene expression program. This multifaceted control may be especially important in particularly plastic cell types, such as fibroblasts, and during progressive lineage restriction when cellular identity is unstable (28, 34). Notably, recent work has highlighted a supportive role for cohesin in promoting proper gene expression during embryonic stem cell differentiation, building on earlier studies demonstrating its role in gene induction (83). Maintaining these regulatory mechanisms may impose additional energetic costs on the cell, but these costs may be advantageous compared to the risk of unstable or inappropriate cell-fate decisions. Indeed, the consequences of aberrant cell-state changes can be profound; excessive or unresolved fibroblast activation underlies fibrotic disease across organ systems, while loss of cellular identity is a hallmark of oncogenic transformation (28, 29, 31, 84). Given the importance of balancing cellular plasticity with stable cellular identity, several key questions remain.

How is specificity achieved? Our data demonstrate that genome folding is both necessary and sufficient to induce fibroblast activation, yet the mechanisms conferring specificity to this response remain unclear. While WAPL depletion drove fibroblasts towards a myofibroblast-like state, it is conceivable that enhanced cohesin stability could have pushed cells towards other states. In addition, WAPL depletion likely only partially recapitulates fibroblast activation, suggesting that additional inputs are required for full activation. For example, shWAPL depletion may amplify the effects of basally weak, TGF-β–responsive enhancers, as observed at MEOX1, while other features of activation may depend on cohesin-independent pathways or non-transcriptional effects of TGF-β. This, in turn, raises the possibility that the basal enhancer landscape shapes the transcriptional response to genome folding. Moreover, the feedback between cohesin-mediated genome folding and enhancers may not be unidirectional. Our previous work has shown that BRD4, which “reads” acetylated proteins including at enhancers, stabilizes NIPBL on chromatin to orchestrate genome folding (46). More recent work suggests that active enhancers can strengthen CTCF binding via the resulting RNA, thereby strengthening TAD boundaries (85). It is tempting to speculate that enhancer activity may establish local feed-forward circuits to promote local changes in genome folding, which then amplify an enhancer’s ability to boost gene expression. Such a model does not necessitate that cohesin loads at enhancers, merely that the local enhancer landscape influences cohesin activity or stability.

How do expressed genes localize to nuclear speckles? Several mechanisms could explain the tendency of actively transcribed genes to reside near nuclear speckles. Transcription—or the resultant RNA—may nucleate the formation of nuclear speckles around transcriptionally active regions. Nuclear speckles are enriched in RNAs and RNA-binding proteins, and many transcription factors have been shown to bind RNA (82, 85–88). These RNA-protein interactions could recruit nuclear speckle components to sites of active transcription. Alternatively, transcription—or the resultant RNA—could mediate directed motion of loci towards pre-existing nuclear speckles. While live imaging studies of reporters support such directed motion, more detailed studies of endogenous loci in physiologically relevant contexts are needed (13). Finally, it is possible that specific proteins, such as those harboring recently described “speckle-targeting motifs,” may facilitate nuclear speckle-locus association (16, 17, 71). Importantly, these mechanisms are not mutually exclusive; local accumulation of speckle-associated factors could, in turn, promote directed motion of transcribed genes towards larger or pre-existing nuclear speckles. Our findings support a model in which cohesin is necessary to induce gene expression, and the nuclear speckle is required to amplify it during cell-state changes (Fig. 5D).

How do nuclear speckles, in turn, promote gene expression? Previous studies have shown inducible genes depend on nuclear speckles for proper expression (12–17). Our work extends those findings by demonstrating that a key nuclear speckle protein SON is required for robust gene expression during a physiologic cell-state change (Fig. 4, D and E). However, the mechanisms by which nuclear speckles amplify gene expression remain incompletely understood. Nuclear speckles are enriched in transcriptional machinery, and our data show increased SMAD2/3, p300, and RNA polymerase II Ser2p signal surrounding nuclear speckles (fig. S6A). These observations raise the possibility that nuclear speckles may promote or stabilize active transcription. Importantly, not all genes require intact nuclear speckles for their expression. When comparing RNA-seq data from shSCR- and shSON-treated fibroblasts under activating conditions, the majority of genes remained unchanged following SON depletion, with a roughly 2:1 ratio of downregulated to upregulated differentially expressed genes. Indeed, what determines a gene’s dependence on nuclear speckles remains an open and exciting question. Speckles have also been proposed to serve as reservoirs for RNA processing factors that enhance splicing efficiency or fidelity (89, 90). While we cannot rule out this contribution entirely, it is likely that nuclear speckles facilitate multiple distinct steps of gene expression.

Further mechanistic studies into the interplay between chromatin architecture and spatial positioning are required to answer these and other outstanding questions. It is becoming clear that multiple facets of genome organization can tune gene expression. It is therefore exciting to speculate that ‘drugging’ different facets of genome organization may provide unique therapeutic opportunities to prevent or reverse diseases rooted in aberrant cellular identity.

MATERIALS AND METHODS

Cell culture

IMR90 fibroblasts (CCL-186 70045937, ATCC) were maintained in DMEM (Gibco, 11965084) supplemented with 10% FBS (SH30910.03 AH29836628, Cytiva) and 1% Penicillin-Streptomycin (Gibco, 15140122) at 37°C with 5% O₂ and 5% CO₂. Experiments were carried out using fibroblasts below passage 10 after receipt from the vendor. Fibroblasts were activated by serum starvation (media as above, with 0.1% FBS) and concurrent treatment with human TGF-β (2 ng/mL; 100-21C, Peprotech). Lenti-X 293 T cells (632180, Takara) were cultured in DMEM (Gibco, 11965084) supplemented with 10% FBS (SH30910.03 AH29836628, Cytiva) and 1% Penicillin-Streptomycin (Gibco, 15140122) at 37°C with 20% O₂ and 5% CO₂. HCT116 cells (CCL-247, ATCC) were cultured in McCoy’s 5A Medium (16600082, Gibco) supplemented with 10% FBS (SH30910.03 AH29836628, Cytiva) and 1% Penicillin-Streptomycin (Gibco, 15140122) at 37°C with 20% O₂ and 5% CO₂. All cells were passaged using trypsin-EDTA (0.05%; 25300054, Gibco).

Lentiviral production

Lentivirus was generated by co-transfecting the lentiviral vector with psPAX2 and pMD2.G (12260 and 12259 respectively, Addgene) into Lenti-X 293 T cells using Lipofectamine 2000 (11668019, Invitrogen) according to the manufacture’s protocol. The media was changed 24 hours after transfection. Viral supernatant was collected 48 hours and 72 hours after transfection, and the media was changed between collections. Viral collections were pooled, passed through a 0.45 μm PES filter, and stored at −80°C. The lentiviral vector used for CRISPRi was pLenti-(BB)-EF1a-KRAB-dCas9-P2A-BlastR (118154, Addgene). Three guides targeting the MEOX1 enhancer were individually cloned into pLenti-(BB)-EF1a-KRAB-dCas9-P2A-BlastR (hereafter, ‘backbone’) by digesting the backbone with BsmbI-v2 (R0739S, New England Biolabs), annealing oligos containing the desired sgRNAs and compatible overhangs, and ligating the backbone and the annealed oligos together using T7 DNA ligase (M0318S, New England Biolabs). The ligation reaction was then transformed into chemically competent E. coli (C3040, New England Biolabs) and plated on LB agar with ampicillin. Colonies were picked into liquid culture and grown overnight at 37°C with shaking. Plasmid DNA was then isolated from liquid culture using the QIAprep Spin Miniprep Kit (27104, Qiagen) according to the manufacturer’s protocol and sequence verified by nanopore sequencing (Plasmidsaurus) or sanger sequencing (Azenta). The lentiviral vector used for shRNA expression was pLKO.1-puro (8453, Addgene). The shRNA targeting SON was generated by University of Pennsylvania’s High Throughput Sequencing Core. sgRNA and shRNA sequences are provided in the supplemental information (table S4).

Lentiviral transduction

To generate polyclonal CRISPRi cell lines, IMR90 fibroblasts were reverse transduced by simultaneous plating and treatment with polybrene (8 μg/mL; TR-1003-G, Sigma-Aldrich) and either a 1:1:1 volumetric pool of lentiviral supernatant harboring the CRISPRi machinery and one of three sgRNAs targeting the MEOX1 enhancer or an equivalent volume of lentivirus harboring the CRISPRi machinery and an empty sgRNA. Fibroblasts were incubated with viral supernatant for 24 hours and then recovered in maintenance media for 24 hours. Fibroblasts were selected with blasticidin (2.5 μg/mL; 203350, Sigma-Aldrich) for 72 ≥ hours prior to expansion. For shRNA mediated knockdown, IMR90 fibroblasts were reverse transduced by simultaneous plating and treatment with filtered pLKO.1-puro lentiviral supernatant. Fibroblasts were incubated with viral supernatant for 24 hours and then recovered in maintenance media for 24 hours. Fibroblasts were selected with puromycin (1–1.5 μg/mL; A1113803, Gibco) for 48 hours before the media was changed to either activation media or maintenance media for 24 hours prior to harvest.

siRNA transfection

siRNAs (Control ‘siNTC’ D-001210-05-20, siNIPBL L-012980-00-0010, siWAPL L-026287-01-0010, siSON M-012983-01-0005; Dharmacon) were transfected into IMR90 fibroblasts using Lipofectamine RNAiMAX (13778150, Invitrogen) according to the manufacturer’s protocol. The next day, the media was changed. After 24 hours, fibroblasts were either activated (as above) or maintained in basal media for an additional 24 hours prior to harvest.

Collagen gel contraction assays

Wild-type or shRNA-treated IMR90 fibroblasts (prepared as described above) were released using 0.05% trypsin-EDTA (25300054, Gibco), pelleted, and resuspended in PureCol EZ Gel solution (5074, Advanced Biomatrix) to a final concentration of 1 × 10⁶ cells/mL. To each of 10 wells in a 24-well plate, 300 μL of the collagen–cell mixture was added. The plate was then incubated under standard culture conditions for 90 minutes to allow for gel polymerization. After polymerization, 500 μL of maintenance media was gently added to each well, and the plate was incubated overnight. The following day, gels designated for activation were switched to starvation media, while control gels remained in maintenance media. The next day activated gels were maintained in starvation media and treated with 2 ng/mL TGF-β (100-21C, Peprotech); control gels remained in maintenance media. Gels were then gently detached from the sides of the well and imaged at 0, 24, 48, and 72 hours post-release. Gel area at each time point was measured using the circle tool in ImageJ/FIJI (v 2.9.0) (91). Percent contraction was calculated by dividing the gel area at each time point by the area at 0 hours, subtracting the result from 1, and multiplying by 100.

qRT-PCR

RNA was harvested from approximately 1 × 10⁶ IMR90 fibroblasts, cultured as detailed above, using the RNEasy Plus Kit (Qiagen, 74136) according to the manufacturer’s protocol. The resultant RNA was quantified using a Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific) and normalized across samples. The RNA was then used as input for cDNA synthesis carried out using the iScript cDNA Synthesis Kit (1708891, Bio-Rad). The resultant cDNA was then used as input with PowerUp SYBR Green Master Mix (A25742, Applied Biosystems). Primer sequences available upon request. Four technical replicates were set up per biological replicate in 384-well plates and run using a QuantStudio 5 apparatus (Applied Biosystems). Relative expression was calculated using the delta-delta Ct method (92).

Subcellular fractionation

Subcellular fractionations were carried out as previously described (46). Briefly, IMR90 fibroblasts or HCT116 cells were harvested, washed with PBS, and collected into cold buffer A (10 mM HEPES pH 7.9, 1.5 mM MgCl₂, 10 mM KCl, 340 mM sucrose, 10% glycerol) with protease inhibitor (11836170001, Roche). Triton X-100 was added to a final concentration of 0.1% and the cell suspension was briefly vortexed and then incubated on ice. Nuclei were isolated by centrifugation and the supernatant (cytosolic fraction) was collected. Nuclei were resuspended in cold buffer B [3 mM EDTA, 0.2 mM EGTA, 1 mM dithiothreitol (DTT)] with protease inhibitor (11836170001, Roche). Chromatin was then isolated by centrifugation and the supernatant (nucleoplasmic fraction) was collected. Chromatin was then solubilized in lysis buffer (10 mM HEPES pH 7.9, 3 mM MgCl₂, 5 mM KCl, 140 mM NaCl, 0.1 mM EDTA, 0.5% NP-40, 0.5 mM DTT) supplemented with protease inhibitor (11836170001, Roche) and benzonase (E1014, Sigma-Aldrich) and incubated at 4°C for 45 minutes. For whole cell lysate, cells were collected directly into lysis buffer with benzonase (E1014, Sigma-Aldrich). For whole nuclear lysate, nuclei were resuspended directly into lysis buffer with benzonase (E1014, Sigma-Aldrich). For immunoprecipitations, the chromatin fraction was collected as detailed above. For determining the relative abundance of RAD21 on chromatin, buffer B was replaced with lysis buffer without benzonase (E1014, Sigma-Aldrich) and the incubation was extended to 20 minutes on ice. Samples were then used for immunoprecipitation or denatured in LDS sample buffer (NP0007, Invitrogen) with sample reducing agent (NP0009, Invitrogen) by boiling and used directly for western blotting.

Immunoprecipitation

Immunoprecipitations were carried out as previously described (46). In Brief, Protein G Dynabeads (10003D, Thermo Fisher) were washed with PBST before being conjugated with antibody at room temperature. The antibodies used for immunoprecipitation were rabbit IgG (L27A9, Cell signaling; 4 μg), anti-RAD21 (ab992, Abcam; 4 μg), anti-SMC1a (A300-055A, Bethyl; 4 μg), anti-SMC3 (ab9263, Abcam; 4 μg), anti-NIPBL (A301-779A, Bethyl; 4 μg), and anti-BRD4 (A700-004CF, Bethyl; 4 μg). Bead-antibody complexes were then resuspended in sample lysate (see above) and incubated overnight at 4°C. Bead-antibody-protein complexes were washed with lysis buffer before resuspension in NuPage LDS buffer (NP0007, Invitrogen) with sample reducing agent (NP0009, Invitrogen). Samples were then denatured by boiling and used for western blotting.

Western blotting

Western blotting was performed as previously described (46). Samples were run on 4–12% BIS-Tris protein gels (WG1403, Invitrogen) in MOPS running buffer (NP0001, Invitrogen). Gels were wet-transferred to a PVDF membrane in transfer buffer (NP00061, Invitrogen). Blots were blocked with 5% BSA (w/v) in TBS with 0.1% Tween-20 for 1 hour at room temperature, blots were trimmed to the region of interest and probed with primary antibody overnight at 4°C. The primary antibodies used were anti-RAD21 (ab992, Abcam; 1:1000), anti-SMC1a (A300-055A, Bethyl; 1:1000), anti-SMC3 (ab9263, Abcam; 1:1000), anti-NIPBL (A301-779A, Bethyl; 1:1000), and anti-BRD4 (A700-004CF, Bethyl; 1:1000), anti-GAPDH (14C10, Cell Signaling; 1:1000), anti-SMA (ab5694, Abcam; 1:500), anti-RBM25 (HPA003025, Atlas Antibodies; 1:1000), anti-SON (HPA023535, Atlas Antibodies; 1:500), anti-HDAC2 (3F3, Cell Signaling; 1:2000), and anti-H3K27ac (D5E4, Cell Signaling; 1:1000). Blots were washed with TBS with 0.1% Tween-20 and then incubated with HRP-linked secondary antibodies. The secondary antibody used for immunoprecipitations was anti-rabbit IgG (L27A9, Cell Signaling; 1:2000). For other western blots, the antibodies used were anti-rabbit IgG (7074, Cell Signaling; 1:2000) and anti-mouse IgG (7076, Cell Signaling; 1:2000). Blots were incubated with SuperSignal West Pico (34080, Thermo Fisher) or Femto (34095, Thermo Fisher) substrates and then imaged on either an ImageQuant LAS 500 or ImageQuant LAS 4000. Band densitometry was performed using ImageJ/FIJI (v 2.9.0). Brightness and contrast were adjusted uniformly across images to aid visualization.

Immunofluorescence (IF)

IMR90 fibroblasts were cultured as detailed above on uncoated glass coverslips in multiwell plates. Media was removed, and samples were fixed with 4% formaldehyde (5710, Electron Microscopy Sciences) in PBS for 10–12 minutes at room temperature. The fixative was removed and samples were washed with PBS twice for 10 minutes at room temperature. The washing solution was removed, and samples were permeabilized with 0.5% Triton X-100 in PBS for 5 minutes at room temperature. The permeabilization solution was removed and samples were washed with PBS twice for 10 minutes at room temperature. Samples were incubated with 5% BSA (w/v) in PBS with 0.05% Tween-20 (hereafter, ‘blocking solution’) for 1 hour at room temperature. The blocking solution was removed and samples were incubated overnight in primary antibody diluted in blocking solution at 4°C. The primary antibodies used were anti-SON (HPA023535, Atlas Antibodies; 1:500), anti-RNA pol II CTD phospho Ser2 (AB 2687450, Active Motif), anti-p300 (ab14984, Abcam; 1:500), anti-smad2/3 (610842, BD; 1:500), anti-SRRM2 (PA5–68827, Invitrogen; 1:500), anti-Ki67 (ab15580, Abcam; 1:500), and anti-SMA (ab5694, Abcam; 1:500). The primary antibody was removed and samples were washed with PBS with 0.05% Tween-20 three times for 10 minutes at room temperature. The washing solution was removed and samples were incubated with fluorescent secondary antibody diluted in blocking solution for 2 hours at room temperature. The secondary antibodies used were anti-rat IgG AF-488 (A21470, Invitrogen; 1:500), anti-rabbit IgG AF-488 (A21206, Invitrogen; 1:500), anti-mouse IgG AF-555 (A31570, Invitrogen; 1:1000), and anti-rabbit IgG AF-647 (A21245, Invitrogen; 1:1000 or 1:2000) The secondary antibody was removed and samples were washed first with PBS with 0.05% Tween-20 two times for 10 minutes and then with PBS two times for 10 minutes all at room temperature. The washing solution was removed and samples were incubated with DAPI in PBS for 5–10 minutes at room temperature. The DAPI solution was removed, and samples were placed in PBS prior to mounting. Coverslips were mounted on glass slides with ProLong Gold Antifade Mountant (P36930, Invitrogen) and sealed with clear nail polish.

DNA oligo paint fluorescence in situ hybridization (DNA FISH)

Samples were fixed, permeabilized, and washed as outlined in ‘Immunofluorescence.’ DNA FISH was performed as previously described (63). Briefly, the washing solution was removed and samples were treated with 2x SSCT (0.3 M NaCl, 0.03 M sodium citrate, and 0.1% Tween-20) for 5 minutes at room temperature. The 2x SSCT was removed and samples were treated with 50% formamide/2x SSCT solution (combine 4x SSCT with an equivalent volume of formamide) for 5 minutes at room temperature. The 50% formamide/2x SSCT solution was removed and samples were treated with 50% formamide/2x SSCT solution heated to 95°C and incubated on a metal block in a 95°C water bath for 5 minutes. The 95°C 50% formamide/2x SSCT solution was removed, and samples were treated with 50% formamide/2x SSCT heated to 65C and incubated on a metal block in a 65°C water bath for 20 minutes. Samples were allowed to reach room temperature before proceeding. The now room temperature 50% formamide/2x SSCT solution was removed and samples were treated with a hybridization mix consisting of primary probe(s), bridges when applicable, dNTPs (5.6 μM), 50% formamide, 10% dextran sulfate, 4% polyvinylsulfonic acid, in 2x SSCT and then incubated on a metal block in a 95°C water bath for 5 minutes. Samples were incubated overnight at 37°C. The next day, samples were washed with 2x SSCT for 5 minutes at room temperature. The room temperature 2x SSCT was removed, and samples were treated with 2x SSCT heated to 65°C and incubated on a metal block in a 65°C water bath for 20 minutes. Samples were allowed to reach room temperature before proceeding. The now room temperature 2x SSCT was removed, and samples were treated with 0.2x SSC for 10 minutes at room temperature. The 0.2x SSC was removed and samples were treated with 2x SSC while secondary hybridization mix was prepared. Samples were treated with secondary hybridization mix consisting of fluorescent secondary probes, 10% formamide, 10% dextran sulfate, 4% polyvinylsulfonic acid, in 2x SSCT for 2 hours at room temperature. Samples were washed with 2x SSCT heated to 65°C and incubated on a metal block in a 65°C water bath for 20 minutes. Samples were allowed to reach room temperature before proceeding. The now room temperature 2x SSCT was removed, and samples were treated with 0.2x SSC for 10 minutes at room temperature. The 0.2x SSC was removed and samples were treated with 2x SSC with DAPI at room temperature for 5–10 minutes. The DAPI solution was removed, and samples were placed in 2x SSC prior to mounting. Coverslips were mounted on glass slides with ProLong Gold Antifade Mountant (P36930, Invitrogen) and sealed with clear nail polish. Probe coordinates are included in the supplemental information (table S4).

Immunofluorescence followed by DNA oligo paint fluorescence in situ hybridization

To simultaneously visualize protein and DNA targets, IF was performed as indicated above. Prior to starting the DNA FISH protocol, samples were fixed with 4% formaldehyde (5710, Electron Microscopy Sciences) in PBS for 10–12 minutes at room temperature. The fixative was removed and samples were washed with PBS twice for 10 minutes at room temperature. The DNA FISH protocol was then performed as indicated above.

MALAT1 RNA oligo paint fluorescence in situ hybridization followed by immunofluorescence

IMR90 fibroblasts were cultured as detailed above on uncoated glass coverslips in multiwell plates. RNA FISH was performed as previously described (63). In brief, media was removed, and samples were fixed with 4% formaldehyde (5710, Electron Microscopy Sciences) in PBS for 10–12 minutes at room temperature. The fixative was removed and samples were washed with PBS twice for 10 minutes at room temperature. The wash solution was removed, and samples were permeabilized with 70% ethanol and 1% SDS overnight at 4°C. The next day, the permeabilization solution was removed and samples were washed with 2x SSCT twice for 5 minutes at room temperature. Samples were treated with a hybridization mix consisting primary probes targeting the common exons of the lncRNA MALAT1, 50% formamide, 10% dextran sulfate, 4% polyvinylsulfonic acid, and 1% SDS in 2x SSCT. Samples were placed on a metal block in a 60°C water bath for 3 minutes. Samples were incubated overnight at 37°C. The next day, samples were washed with 2x SSCT twice for 5 minutes at room temperature. Samples were then treated with a hybridization mix consisting fluorescent secondary probes, 50% formamide, 10% dextran sulfate, 4% polyvinylsulfonic acid, and 1% SDS in 2x SSCT. Samples were washed with 2x SSCT twice for 5 minutes at room temperature. Samples were washed with PBS with 0.05% Tween-20 twice for 5 minutes at room temperature. To limit exposure to RNAse, BSA was omitted from the proceeding abbreviated immunofluorescence protocol. The wash solution was removed, and samples were treated with primary antibody diluted in PBS with 0.05% Tween-20 supplemented with RNAseOUT (10777019, Invitrogen) for 1 hour at room temperature. The primary antibody was removed and samples were washed with PBS with 0.05% Tween-20 twice for 5 minutes at room temperature. The washing solution was removed and samples were treated with fluorescent secondary antibody diluted in PBS with 0.05% Tween-20 for 1 hour at room temperature. The secondary antibody was removed and samples were washed with PBS with 0.05% Tween-20 twice for 5 minutes at room temperature. The secondary antibody was removed and samples were treated with DAPI in PBS for 5–10 minutes at room temperature. The DAPI solution was removed, and samples were placed in PBS prior to mounting. Coverslips were mounted on glass slides with ProLong Gold Antifade Mountant (P36930, Invitrogen) supplemented with RNAseOUT (10777019, Invitrogen) and sealed with clear nail polish.

Confocal image capture and analysis

All microscopy was performed on a Leica SP8 confocal microscope using a Leica 40x or 63x oil objective and HyD detectors in standard mode with 100% gain. Regions were selected for imaging at random using the LAS X navigator screen. Image capture was automated. DNA FISH and IF with DNA-FISH image sets were captured as z-stacks, while IF only image sets were captured as a single plane unless otherwise noted. Out of plane or empty images were discarded. Splitting multi-channel images, max projecting, as well as background subtraction were automated using an ImageJ/FIJI (v 2.9.0) macro. Single channel image sets were imported into CellProfiler (v 4.2.8) for further analysis (63, 93). Images were segmented algorithmically using the IdentifyPrimaryObjects module with either 2- or 3-state Otsu thresholding. To ensure measurements were made within the same nucleus, objects were assigned to parent nuclei using the RelateObjects module. Distances between DNA FISH probes were calculated as the 2D Euclidean distance between object centroids using x-y coordinates exported from CellProfiler. For DNA FISH probe-to-probe measurements, distances above 0.75 μm were excluded from analysis. For DNA FISH probe-to-speckle measurements, segmented nuclear speckles were converted to binary masks. A distance transform was then applied to the nuclear speckle mask to generate a per-pixel map of the Euclidean distance to the nearest nuclear speckle edge. The distance for each DNA FISH probe was defined as the mean value of this distance map over the probe, corresponding to the average distance between the probe and the edge of the nearest nuclear speckle. For DNA FISH probe-to-speckle measurements, distances above 1.5 μm were excluded from analysis. For sample images, brightness and contrast were adjusted uniformly to aid visualization.

RNA-sequencing preparation and analysis

RNA was harvested from approximately 1 × 10⁶ IMR-90 fibroblasts, cultured as detailed above, using the RNEasy Plus Kit (74136, Qiagen) according to the manufacturer’s protocol. The resultant RNA was quantified using a Nanodrop 2000 spectrophotometer (Thermo Scientific). RNA quality was assessed by agarose gel electrophoresis. Subsequently, 1 μg of RNA was used as input for the NEBNext Poly(A) mRNA Magnetic Isolation Module (E7490L, New England Biolabs) and subsequently the NEBNext Ultra II RNA Library Prep kit (E7770L, New England Biolabs) according to the manufacturer’s protocol. Libraries were then sequenced on either a NextSeq1000 SE122 (Illumina) or NovaSeq X Plus PE150 (Illumina, Novogene). When paired-end sequencing was performed, only read-one was used for downstream analyses and all samples were treated as single-end to maintain consistency.

Adaptors and low-quality bases were clipped using Trimmomatic (v 0.39) (94). Prior to pseudoalignment, a reference index was first built from a cDNA reference transcriptome (Homo_sapiens.GRCh38.cdna.all.fa) using kallisto index (v 0.46.2) (95). Reads were pseudoaligned to the cDNA reference transcriptome using kallisto quant (v 0.46.2). The resulting abundance files were imported into R (v 4.4.1) using the tximport (v 1.32.0) package and condensed into length-normalized gene-level quantifications (96). Transcript-to-gene mapping was performed using the EnsDb.Hsapiens.v86 (v 2.99.0) package. Genes with counts per million (CPM) > 1 in ≥3 samples were retrained and libraries were normalized using the trimmed mean of M-values (TMM) method and log₂ transformed using the edgeR (v 4.2.2) package (97). Differential expression analysis was performed using the limma (v 3.60.6) package (98). Genes with a Benjamini-Hochberg adjusted P value < 0.05 and an absolute log₂(fold-change) > 1.5 were designated as differentially expressed. Hierarchical clustering was used to identify modules of coordinately regulated DEGs across conditions.

To assess shifts in gene expression between conditions, scatter plots were generated by plotting log₂-transformed, normalized expression values for each gene, and the proportion of genes above the identity line (y = x) was calculated. Permutation testing was used to evaluate whether the observed shift exceeded that which was expected by chance. For each iteration, a gene set equal in size to the observed set was randomly sampled from all expressed genes (defined as CPM > 1 in ≥3 samples), and the proportion of genes above the identity line was calculated. This process was repeated 10,000 times to generate a null distribution, which was used to calculate a two-tailed p-value.

Gene ontology (GO) analysis was performed using the clusterProfiler (v 4.12.6) package (99, 100). Enrichment for biological process (BP) terms was tested using the enrichGO function with the org.Hs.eg.db annotation database (v 3.19.1). GO analysis was conducted using gene symbols as input and the Benjamini–Hochberg method to adjust p-values. Terms with adjusted P value and q value <0.001 were considered significantly enriched.

For splicing analysis, reads from biological replicates were pooled by condition and aligned to the GRCh38 reference genome using HISAT2 (v 2.2.1) with the --dta flag (101). Only primary alignments were retained using samtools view (v 1.20) with the -F 260 flag (102). To account for differences in sequencing depth, alignment files were down-sampled to match the lowest read count using samtools view with the --subsample flag (v 1.20). After sorting and indexing, the down sampled alignment files were loaded into the Integrative Genomics Viewer (v 2.18.4) and sashimi plots were generated (103). Intron-spanning read thresholds were adjusted based on local coverage to facilitate interpretation.

ChIP-sequencing preparation and analysis

Chromatin fragmentation and immunoprecipitation of wild-type samples was performed using the SimpleChIP Enzymatic Chromatin IP Kit (9003, Cell Signaling) according to the manufacturer’s protocol. Chromatin fragmentation and immunoprecipitation of shRNA-treated samples was performed using the SimpleChIP Plus Sonication Chromatin IP Kit (56383, Cell Signaling) according to the manufacturer’s protocol. Chromatin was sheared using a Covaris S220 apparatus (Covaris). The antibodies used for chromatin immunoprecipitation were anti-H3K27ac (D5E4, Cell Signaling) and anti-RAD21 (992, Abcam). Sequencing libraries were prepared from the resultant immunoprecipitated chromatin, and respective inputs, using the NEBNext Ultra II DNA Library Prep Kit for Illumina (E7645L, New England Biolabs) according to the manufacturer’s protocol. The resultant wildtype derived libraries were sequenced on either a NextSeq 550 PE38 (Illumina) or NextSeq 1000 PE60 (Illumina). The resultant shRNA-treated derived libraries were sequenced on a NovaSeq X Plus PE150 (Illumina, Novogene). Reads were aligned to the GRCh38 reference genome using Bowtie2 (v 2.5.3) in local alignment mode (104). Alignments were restricted to the single best match per read and mixed or discordant read pairs were discarded. Alignments with mapping scores below 30 as well as PCR duplicates were discarded using Samtools (v 1.20).

Browser track files for H3K27ac were generated using deepTools (v 3.5.5). CPM normalized signal intensity files were generated using bamCoverage and blacklist regions were excluded. The resulting signal intensity files were visualized in the Integrative Genomics Viewer (v 2.18.4).

RAD21 metaplots were generated using deepTools (v 3.5.5) (105). Signal intensity files representing the CPM normalized ChIP-to-input ratio were produced with bamCompare, excluding ENCODE blacklist regions. CTCF ChIP-seq peaks (ENCODE accession ENCFF203SRF) were assigned to TAD pentiles (see ‘Hi-C preparation and analysis’) and classified as boundary or non-boundary using the rtracklayer (v 1.64.0) and GenomicRanges (v 1.56.2) packages in R (v 4.4.1) (106, 107). A peak was classified as a “boundary” peak if it was located within 10 kb of the start or end of its assigned TAD. Metaplots were then generated using computeMatrix in scale-regions mode and visualized with plotProfile.

H3K27ac peaks were identified in wild-type samples using MACS2 (v 2.2.9.1) with matched input controls. Differential peak analysis between unactivated and activated conditions was performed in R (v 4.4.1). Peak sets were merged across replicates, filtered by requiring detection across biological replicates, and filtered by overlap with blacklist regions using the rtracklayer (v 1.64.0), GenomicRanges (v 1.56.2), and plyranges (v 1.24.0) packages (108). Read counts over the non-redundant peak set were quantified using the Rsamtools (v2.20.0) and GenomicAlignments (v 1.4.0) packages (107, 109). Differential peaks were identified using the edgeR (v 4.2.2) and limma (v 3.60.6) packages. Peaks with a Benjamini-Hochberg adjusted P value < 0.01 and an absolute log₂(fold-change) > 1 were considered differentially enriched/depleted. Hierarchical clustering was used to identify modules of coordinately regulated peaks. deepTools (v 3.5.5) was used to visualize these peak sets as heatmaps across wild-type and shRNA-treated samples using computeMatrix in scale-regions mode and plotHeatmap.

To examine the relationship between enhancer status and speckle proximity, H3K27ac peaks (this study; unactivated condition) and H3K4me1 peaks (ENCODE accession ENCSR831JSP) were identified as described above. Publicly available SON TSA-Seq data (SRA accession SRP271102) were aligned to the GRCh38 reference genome as described above. Subsequent analyses were performed in R (v 4.4.1). SON TSA-Seq read counts were quantified within 250 kb bins (excluding blacklist regions) using the GenomicAlignments (v 1.4.0) and edgeR (v 4.2.2) packages. Bins were then ranked by normalized SON signal and divided into vigintiles. H3K4me1 peaks were classified as either H3K4me1-only or H3K4me1 & H3K27ac based on genomic overlap with the H3K27ac peak. The total fraction of each peak type falling within each SON TSA-Seq vigintile was then calculated.

Hi-C preparation and analysis

Chromatin was prepared using the Arima-HiC Kit (A510008, Arima) according to the manufacturer’s protocol. Sequencing libraries were generated using the Arima Library Prep Module (A303011, Arima) and sequenced on a NovaSeq X Plus PE150 (Illumina, Novogene). Reads were processed according to the HiC-Pro pipeline (v 3.1.0) (35). Biological replicates were uniformly down sampled. The Spearman correlation between biological replicates was determined using hicCorrelate with the --log1p flag from HiCExplorer (v 3.7.5) (36). Biological replicates were then merged. Subsequent analyses were performed using the merged matrices.

Compartments were identified using Juicer Tools (110). Knight-Ruiz (KR) normalized, 250 kb resolution contact matrices were passed to eigenvector from Juicer Tools. PCA scores were calculated for individual chromosomes and the sign was adjusted based on gene density and publicly available H3K36me3 ChIP-seq data (ENCODE accession ENCSR437ORF). Chromosome-specific eigenvector tracks were concatenated, and megabase-scale blacklist regions were removed. The distribution of sign-corrected PC1 values was visualized in R. Compartment-scale, KR normalized matrices were visualized with HiCExplorer (v 3.7.5) using hicTransform with the --method obs_exp flag and hicPlotMatrix.

Topologically associating domains (TADs) were identified using HiCExplorer (v 3.7.5). Uncorrected 10 kb contact matrices were normalized using hicCorrectMatrix with the --correctionMethod ICE flag. TADs were called using these ICE corrected matrices with hicFindTADs, and the resulting domain, boundary, and separation score files were filtered to exclude megabase-scale blacklist regions. For each TAD, the mean separation score was calculated using the control condition TAD calls (i.e., unactivated or shSCR-treated) and the separation score tracks from both control and treatment conditions (unactivated vs. activated; shSCR vs. shWAPL). TADs were then ranked by the change in mean separation score between conditions (Activated - Unactivated or shWAPL - shSCR) and stratified into pentiles. Aggregate TAD maps were generated for each pentile using coolpuppy (111). TAD sets were passed to coolpuppy.py using the --local and --rescale options, and visualized with plotpup.py. Ratio maps were generated using dividepups.py and similarly visualized with plotpup.py. To examine the relationship between changes in TAD strength and gene expression, genes were mapped to control-condition TADs by their transcription start sites in R (v 4.4.1) using the EnsDb.Hsapiens.v86 (v 2.99.0) and GenomicRanges (v 1.56.2) packages. A single transcription start site was used per gene. TADs were classified based on the gene expression patterns of their constituent genes as outlined in the main text, and the fraction of TADs in each expression category within each TAD pentile was calculated.

Chromatin loops were identified using Mustache (112). Uncorrected 5 kb contact matrices were normalized using hicCorrectMatrix from HiCExplorer (v 3.7.5) with the --correctionMethod ICE flag. Differential loops were called from ICE-corrected matrices using diff_mustache.py. Consensus and differential loop sets were assembled in R (v 4.4.1) using the GenomicRanges (v 1.56.2) package by filtering out condition specific loops, not flagged as differentially strengthened as well as loops identified as differentially strengthened in both conditions. Pileup analysis was performed using coolpuppy. Briefly, differential loops were passed to coolpuppy.py and visualized with plotpup.py. Ratio plots were computed with dividepups.py and visualized using plotpup.py. Loops and TADs were visualized on ICE normalized, 5 kb matrices with HiCExplorer (v 3.7.5) using hicPlotMatrix. To examine the relationship between differential chromatin looping and gene expression, differential loops were represented as loop domains—continuous intervals from the start of the first loop anchor to the end of the second anchor. The number of loop domains overlapping control condition TADs was then determined in R (v 4.4.1) using the GenomicRanges (v 1.56.2) package. TADs were classified into zero-, low-, medium-, or high-loop categories using a successive median approach: after removing zero-loop TADs, the median loop count of the remaining TADs defined the low/medium cutoff, and the process was repeated to determine the medium/high cutoff. The fraction of TADs in each gene expression category was calculated for each loop category. To determine the relationship between differential chromatin looping across experimental conditions, each TAD identified in shSCR-treated fibroblasts was mapped to a TAD in the WT, unactivated fibroblasts by its midpoint in R (v 4.4.1) using the GenomicRanges (v 1.56.2) package. TADs which could not be uniquely assigned were removed. We then compared loop categories across these pairs of TADs.

Statistical analyses

Statistical analyses unless otherwise noted were performed in R (v 4.4.1) using either base R functions, measures intrinsic to the above packages, or the rstatix (v 0.7.2) package (113). Data plotting, unless otherwise indicated, was performed in R using the Tidverse (v 2.0.0) suite of packages and the ggpubr (v 0.6.1) package (114, 115). The following cut-offs were used for assigning significance to P values: ****P < 1 × 10⁻⁴, 1 × 10⁻⁴ ≤ ***P < 0.001, 0.001 ≤ **P < 0.01, **; 0.01 ≤ *P < 0.05, *; P ≥ 0.05, ns.

Supplementary Materials

The PDF file includes:

Figs. S1 to S7

Legends for tables S1 to S5

Other Supplementary Material for this manuscript includes the following:

Tables S1 to S5