Regulation of cohesin‐mediated chromosome folding by PDS5 in mammals

Abstract

Cohesin regulates sister chromatid cohesion but also contributes to chromosome folding by promoting the formation of chromatin loops, a process mediated by loop extrusion. Although PDS5 regulates cohesin dynamics on chromatin, the exact function of PDS5 in cohesin‐mediated chromatin looping remains unclear. Two paralogs of PDS5 exist in vertebrates, PDS5A and PDS5B. Here we show that PDS5A and PDS5B co‐localize with RAD21 and CTCF at loop anchors. Rapid PDS5A or PDS5B degradation in liver cancer cells using an inducible degron system reduces chromatin loops and increases loop size. RAD21 enrichment at loop anchors is decreased upon depletion of PDS5A or PDS5B. PDS5B loss also reduces CTCF signals at loop anchors and has a stronger effect on loop enlargement compared with PDS5A. Co‐depletion of PDS5A and PDS5B reduces RAD21 levels at loop anchors although the amount of cohesin on chromatin is increased. Our study provides insight into how PDS5 proteins regulate cohesin‐mediated chromatin looping.

PDS5 proteins stabilize cohesin at loop anchors, facilitate chromatin loop formation and restrict loop expansion in mammals.

Introduction

The mammal interphase genome forms long‐range interactions and is hierarchically folded into topological associating domains (TADs) and chromatin loops. TADs are defined by contiguous genomic segments with highly frequent chromatin interactions inside and insulated from adjacent TADs (Dixon et al, ²⁰¹²; Nora et al, ²⁰¹²), regulating gene transcription by promoting or preventing enhancer‐promoter interactions. Chromatin loops are visible as “corner dots” or “peaks” of high contact frequency between pairs of loci (Rao et al, ²⁰¹⁴). Architectural proteins, notably cohesin and the insulator protein CTCF, enrich at TAD boundaries and loop anchors (Parelho et al, ²⁰⁰⁸; Nasmyth & Haering, ²⁰⁰⁹; Rao et al, ²⁰¹⁴). The recently proposed loop extrusion model suggested that cohesin forms TAD and loops by processively extruding DNA loops until it dissociates from chromatin or encounters the boundary elements such as the convergently oriented CTCF sites (Sanborn et al, ²⁰¹⁵; Fudenberg et al, ²⁰¹⁶). Single‐molecular imaging studies indicate recombinant human cohesin‐NIPLB‐MAU2 complex can extrude DNA and form loops in a manner dependent on cohesin's ATPase activity (Davidson et al, ²⁰¹⁹; Kim et al, ²⁰¹⁹).

Cohesin plays diverse roles in cellular processes, including chromosome segregation and compaction, DNA damage repair, and transcription regulation (Onn et al, ²⁰⁰⁸; Peters et al, ²⁰⁰⁸; Nasmyth & Haering, ²⁰⁰⁹; Haarhuis et al, ²⁰¹⁴). The cohesin complex is composed of an a‐kleisin subunit (RAD21/SCC1), two structural maintenance of chromosomes proteins (SMC3 and SMC1), which form the ring‐shaped structure to topologically entrap DNA within its ring, and one SA (Stromal Antigen) subunit either STAG1 (SCC3) or STAG2 (Haering et al, ²⁰⁰⁸). Cohesin dynamics on chromatin throughout the cell cycle are mainly controlled by load factors NIPBL/MAU2 (SCC2/SCC4), release factor WAPL, PDS5 and sororin (Ciosk et al, ²⁰⁰⁰; Kueng et al, ²⁰⁰⁶; Schmitz et al, ²⁰⁰⁷; Ouyang et al, ²⁰¹⁶). Besides, CTCF is reported to stabilize cohesin by competing for WAPL binding to STAG2‐cohesin complex (Li et al, ²⁰²⁰).

PDS5 is supposed to play important roles in the establishment, maintenance and dissociation of cohesin through forming complexes with different cohesin regulatory proteins. On the one hand, PDS5 promotes SMC3 acetylation mediated by cohesin acetyltransferase ESCO1/2 and subsequent association with sororin to stabilize cohesin on chromatin (Chan et al, ²⁰¹³; Minamino et al, ²⁰¹⁵). On the other hand, PDS5 binds to RAD21 and recruits WAPL through a competing region with sororin to open the gate between SMC1 and SMC3 interface thereby unloading cohesin from chromosome (Nishiyama et al, ²⁰¹⁰; Chan et al, ²⁰¹²; Muir et al, ²⁰¹⁶; Goto et al, ²⁰¹⁷). There are two paralogs of PDS5 in vertebrates, PDS5A and PDS5B, and both of them consist of ~1,400 amino acids, with ~70% identical sequences throughout the protein, including two HEAT clusters separated by a helical insert domain (HID) in the N terminus, and a distinct C terminus ~300 amino acids (Neuwald & Hirano, ²⁰⁰⁰; Sumara et al, ²⁰⁰⁰). It is likely that the conserved N terminus of PDS5A and PDS5B containing binding sites for RAD21, WAPL, and sororin is responsible for their overlapping roles and the differing C terminus determines specific roles. Both PDS5A and PDS5B could interact with STAG1‐cohesin and STAG2‐cohesin and contribute to the chromosome arm cohesion, with STAG1‐PDS5A and STAG1‐PDS5B cohesin complex required for telomere cohesion and STAG2‐PDS5B essential for centromere cohesion (Canudas & Smith, ²⁰⁰⁹; Carretero et al, ²⁰¹³). Knockout mice for either PDS5 gene is embryonic lethality at different development stages (Carretero et al, ²⁰¹³), indicating their function cannot be mutually compensated. The common and specific function of PDS5A and PDS5B on chromosome architecture and gene expression remains to be identified.

Previous studies indicated either depletion of PDS5A or PDS5B has little effect on the amount of cohesin on chromatin, while co‐depletion of PDS5A and PDS5B increased the retention time and the level of chromatin‐bound cohesin according to inverse fluorescence recovery after photobleaching (iFRAP) and chromatin fractionation experiments (Ouyang et al, ²⁰¹⁶; Wutz et al, ²⁰¹⁷; Morales et al, ²⁰²⁰), ostensibly producing phenotypes similar to cohesin releasing factor WAPL. However, chromatin loops were affected strikingly differently in cells depleted of both PDS5 proteins compared to WAPL‐depleted cells. It is reported that co‐depletion of both PDS5A and PDS5B greatly decreased loop numbers, whereas WAPL remarkably increased the number of chromatin loops accompanied by more extended loop anchors (Haarhuis et al, ²⁰¹⁷; Wutz et al, ²⁰¹⁷). How PDS5 proteins perform this paradoxical function is poorly understood. Besides, a recent study showed that Pds5 could restrict chromatin loop expansion in yeast, this was consistent with the observation in mammals that the average size of chromatin loops was increased upon co‐depletion of PDS5A and PDS5B (Wutz et al, ²⁰¹⁷; Dauban et al, ²⁰²⁰). But the mechanisms of how PDS5 regulates loop expansion remain to be investigated.

Here, combining in situ Hi‐C, ChIP‐seq for cohesin and its regulators, and RNA‐seq data, we revealed the individual contribution of PDSS5A and PDS5B to chromatin organization at different layers in mammals, providing new perspectives into the roles of PDS5A/B in cohesin‐mediated chromatin looping and the nature of cohesin biology.

Results

Acute degradation of PDS5A or PDS5B in PLC/PRF/5 cells with inducible FKBP^12F36V‐dTAG system

To investigate the direct role of PDS5A and PDS5B in high‐order chromatin organization, we utilized the degradation tag (dTAG) system by knocking in the FLAG‐FKBP^12F36V tag at the N terminus of the endogenous PDS5A or PDS5B locus in a liver cancer cell line ‐ PLC/PRF/5 cells to obtain cells for rapid degradation respectively (Fig 1A; Hu et al, ²⁰²¹). We observed an acute depletion of PDS5A or PDS5B protein upon dTAG‐13 treatment when compared with the control (Figs 1B and EV1A), with no noticeable changes in the protein levels of RAD21, CTCF and the other non‐targeted PDS5 paralog through immunoblotting (Fig EV1B). Besides, the protein levels of tagged PDS5A and PSD5B were similar to wild type (Fig EV1C), indicating no leakage of the dTAG‐FKBP^12F36V system. Then we performed chromatin immunoprecipitation followed by deep sequencing (ChIP‐seq) experiments to detect whether PDS5A/B was removed from the chromatin in the presence of dTAG stimuli. The ChIP‐seq binding profiling of tagged ‐ PDS5A or PDS5B across the genome using antibodies against FLAG, PDS5A or PDS5B showed high similarity and the binding sites were largely overlapped (Fig 1C), indicating FLAG‐FKBP^12F36V fusion did not interfere with the molecular property of PDS5A or PDS5B protein and we could use endogenous knock‐in cells (hereinafter referred to as PDS5A‐dTAG and PDS5B‐dTAG) for downstream functional studies. In addition, the heatmap analysis comparing the genome‐wide signal density of PDS5A/B in cells mock depleted or depleted of PDS5A/B suggested that PDS5A/B degradation was efficient with the dTAG degron technology (Fig 1D), and ChIP‐qPCR was further confirmed that FLAG‐PDS5A or FLAG‐PDS5B was removed from chromatin (Fig 1E and F).

Figure 1. Acute PDS5A or PDS5B degradation with the inducible dTAG technology.

• A
  Schematic of the generation of PSD5A‐dTAG or PDS5B‐dTAG PLC/PRF/5 cells.
• B
  Western blots of whole‐cell extracts showing efficient degradation of PDS5A or PDS5B (left, right) with dTAG treatment (at different time points).
• C
  Genome browser track at a defined region of chromosome 1 for Flag‐PDS5A, PDS5A in PDS5A‐dTAG cells treated with DMSO or dTAG for 24 h (left), or Flag‐PDS5B, PDS5B in PDS5B‐dTAG cells treated with DMSO or dTAG for 24 h (right).
• D
  ChIP‐seq signal for FLAG‐PDS5A and FLAG‐PDSB (ChIP assays were performed with antibodies against Flag and PDS5A/B) centered at TSS of promoters (±3 kb) across all peaks called for each of the proteins in DMSO treated PDS5A‐dTAG or PDS5B‐dTAG cells, showing the drastic reduction in PDS5A/B enrichment on chromatin after degradation. Top: normalized signal intensities for each protein.
• E
  Representative track examples of small genomic regions showing complete depletion of PDSA5A or PDS5B after dTAG treatment (left, right).
• F
  ChIP‐qPCR using Flag antibody demonstrating complete depletion of PDS5A or PDS5B (left, right) in H19/IGF2 imprinting locus and the promoter region of PCDH10, NC represents a negative binding site. Data are represented as mean ± SD from three biological replicates.

Source data are available online for this figure.

Figure EV1. Validation of the PDS5A‐dTAG and PDS5B‐dTAG cells.

• A
  Validation of endogenous knock‐in by PCR for genomic DNA.
• B
  Western blots showing the overall protein levels of PDS5A or PDS5B, RAD21 and CTCF were not affected after PDS5B or PDS5A degradation.
• C
  Western blots showing the protein levels of PDS5A and PDS5B in PDS5A‐dTAG and PDS5B‐dTAG cells respectively (left, right), compared to wild‐type PLC/PRF/5 cells.

Source data are available online for this figure.

PDS5A and PDS5B colocalize extensively with cohesin and CTCF

To characterize the genome‐wide occupancy of PDS5A and PDS5B, we conducted ChIP‐seq assays for FLAG‐PDS5A, FLAG‐PDS5B with antibodies against FLAG in PDS5A‐dTAG or PDS5B‐dTAG cells without dTAG treatment. To further compare the genomic profiles of PDS5A/B with cohesin and CTCF, we also analyzed the ChIP‐seq data of cohesin core component RAD21 and CTCF from PDS5A‐dTAG cells mock depleted of PDS5A. As shown in the genome browser track, these four factors shared a consistent genomic distribution pattern (Fig 2A). We further identified 41,577, 28,766, 53,356, 60,574 specific binding sites for FLAG‐PDS5A, FLAG‐PDS5B, RAD21 and CTCF, respectively, of which all preferentially enriched in promoters and 5' UTR (Fig EV2A). Then, three categories of genomic positions based on the differential binding of PDS5A or PDS5B were defined: PDS5A & PDS5B common sites (59% of all peaks), PDS5A unique sites (39%) and PDS5B unique sites (3%; Figs 2B, and EV2B and C). The overall signal densities of PDS5A, PDS5B, RAD21 and CTCF in PDS5A & PDS5B common sites were stronger than in PDS5A or PDS5B unique sites (Fig 2C). Besides, a large population of PDS5A or PDS5B binding sites were overlapped with RAD21 and CTCF (Fig 2D). Like RAD21 and CTCF, PDS5A/B was enriched at loop anchors and TAD boundaries based on Hi‐C data in the present study (Figs 2E and EV2D). Our data deciphered – to our knowledge for the first time – the global distribution pattern of PDS5A/B and demonstrated that PDS5 proteins co‐localize at numerous sites including loop anchors and TAD boundaries with cohesin and CTCF throughout the genome.

Figure 2. Similar genomic distribution of PDS5A and PDS5B with cohesin and CTCF.

• A
  Genome browser track showing genomic distribution of FLAG‐PDS5A, FLAG‐PDS5B, RAD21, and CTCF at a selected region of chromosome 2. FLAG‐PDS5A and FLAG‐PDS5B ChIP‐seq experiments were performed using Flag antibody from PDS5A‐dTAG or PDS5B‐dTAG cells mock depleted of PDS5A or PDS5B. ChIP‐seq data of RAD21 and CTCF were produced from PDS5A‐dTAG cells without dTAG treatment.
• B
  Heatmaps of signal intensities for FLAG‐PDS5A, FLAG‐PDS5B, CTCF, and RAD21 ChIP‐seq data across all defined FLAG‐PDS5A and FLAG‐PDS5B peaks.
• C
  Averaged ChIP‐seq signal densities of FLAG‐PDS5A, FLAG‐PDS5B, RAD21, and CTCF on PDS5A & PDS5B common sites and PDS5A/PDS5B unique sites.
• D
  Venn Diagram analysis depicting the overlapping binding events of FLAG‐PDS5A or FLAG‐PDS5B with RAD21 and CTCF, or between FLAG‐PDS5A and FLAG‐PDS5B.
• E
  Averaged signal plots displaying the distribution of PDS5A, PDS5B, RAD21 and CTCF at DNA loop anchors and TAD boundaries defined with following Hi‐C data of PDS5A‐dTAG cells without dTAG treatment.
• F
  Reciprocal Re‐ChIP for PDS5A and PDS5B. The co‐binding of PDS5A and PDS5B at PDS5A&PDS5B common sites were assayed by Re‐ChIP‐qPCR. The second ChIP for Re‐ChIP was performed using antibodies against PDS5A, FLAG(PDS5B), RAD21, control IgG. RAD21 served as a positive control. Data are shown as fold enrichment versus IgG (mean ± SD from three biological replicates).
• G
  Heatmaps and averaged signal plots showing the differential PDS5B or PDS5A distribution on sites defined in Fig 2B after PDS5A or PDS5B depletion.

Source data are available online for this figure.

Figure EV2. PDS5A and PDS5B have overlapping and specific genomic binding sites.

• A
  CEAS analysis of the peaks of FLAG‐PDS5A, FLAG‐PDS5B, RAD21 and CTCF.
• B
  CEAS analysis of PDS5A & PDS5B common sites, PDS5A or PDS5B unique sites.
• C
  Representative track examples of PDS5A or PDS5B unique binding sites.
• D
  Hi‐C interaction matrix of the 213.2–215.8 Mb region of chromosome 1 in DMSO treated PDS5A‐dTAG cells. ChIP‐seq profiles for FLAG‐PDS5A, FLAG‐PDS5B, RAD21 and CTCF in corresponding regions are displayed below the Hi‐C interaction matrix.
• E
  ChIP‐seq profiles for FLAG‐PDS5A, FLAG‐PDS5B and RAD21 of the eight positive sites verified by ChIP‐qPCR and Re‐ChIP‐qPCR. The positions of the primers are indicated below the tracks.
• F
  ChIP‐seq profiles of the 43.5–48 Mb region of chr 7 showing the effect of one PDS5 protein depletion on the genomic distribution of the other paralog. The dashed lines represent regions occupied by PDS5A&PDS5B common sites, and the arrows highlights reduced PDS5B signals at common sites upon PDS5A depletion and unchanged PDS5A signals at common sites upon PDS5B depletion.

The ChIP‐seq peaks of PDS5A and PDS5B overlapped at many genomic positions, but it was unclear whether both proteins could co‐bind at a given site or these sites had no preference for PDS5A or PDS5B in a cell population. To address this point, we performed sequential ChIP experiments. Our results indicated that PDS5A and PDS5B could indeed co‐occupy at the same genomic positions in the same cell (Figs 2F and EV2E). Next, we asked whether the depletion of one PDS5 ortholog could have an influence on the other at genomic binding sites. Calibrated ChIP‐seq experiments with PDS5B or PDS5A antibody were performed in cells mock depleted or depleted of PDS5A or PDS5B. After PDS5A depletion, an evident decrease in PDS5B binding signal was observed at both PDS5A & PDS5B common sites and PDS5B unique sites defined in Fig 2B (Figs 2G and EV2F). It is likely that the PDS5B distribution on chromatin was partially PDS5A‐dependent. In contrast, PDS5B loss had little effect on PDS5A binding intensities at both common sites and PDS5A unique sites. Therefore, we conclude that PDS5 proteins could not substitute for each other at genomic binding sites. The interplay between PDS5A and PDS5B on genomic distribution might account for their overlapping functions on chromosome.

The contribution of PDS5A/B to long‐range chromatin interactions

To address the contributions of PDS5A/B to spatial chromatin arrangement, we performed in situ Hi‐C experiments for PDS5A‐dTAG and PDS5B‐dTAG cells with dTAG/DMSO treatment (Fig EV3A and B). When subtracting the individual Hi‐C interaction matrix of chromosome 18 from d‐TAG and DMSO treated cells at a 100 Kb resolution, we observed apparent intra‐chromosomal interactions changes between pairs of loci (Fig 3A). Upon PDS5A or PDS5B loss, short‐range chromosomal interactions (< 0.4 or < 0.6 Mb) decreased while long‐range interactions increased (0.4–6 or 0.6–9 Mb; Fig 3B). This effect was evidently visualized in Hi‐C interaction matrices displaying separately increased and decreased interaction frequencies of an enlarged 15 Mb region of chr18 and other representative regions (Figs 3C and EV3C).

Figure EV3. Depletion of PDS5A or PDS5B does not globally affect large‐scale high‐order chromatin structure.

• A
  Cell cycle profiles of PDS5A/B‐dTAG cells with DMSO or dTAG treatment for 24 h. Cells were harvested at high confluency after treatment.
• B
  Hi‐C data correlation between all replicates for different conditions.
• C
  Zoomed‐in interaction matrices for PDS5A‐dTAG (left) and PDS5B‐dTAG (right) cells with DMSO and dTAG treatments on given regions of chr17 and chr18. Differential Hi‐C matrices (dTAG‐DMSO) split into decreased interactions (blue) and increased interactions (red) are shown below.
• D
  Saddle plots displaying the differential interactions between and within (A and B) compartments after PDS5A and PDS5B loss.
• E
  Density plots representing the size distribution of TAD before and after PDS5A or PDS5B depletion (left, right).
• F
  Insulation scores around TAD boundaries of three classes related to compartments in DMSO and dTAG treated PDS5A‐dTAG or PDS5B‐dTAG cells.

Figure 3. The contribution of PDS5A and PDS5B to long‐range chromosomal interactions.

• A
  Differential Hi‐C interaction maps of chromosome 18 after PDS5A or PDS5B degradation (left, right) at 100 kb resolution. The matrices plots are all derived from Juicebox. Middle are Zoomed‐in regions (20–35 Mb) on chromosome 18 from differential Hi‐C interaction matrices of PDS5A/B‐dTAG cells.
• B
  Intra‐chromosomal interaction frequencies plotted against the distance between interaction bins at the maximum of 10 Mb in DMSO or dTAG treated PDS5A/B‐dTAG cells (up). Comparison of the differential interaction frequencies (dTAG‐DMSO) as a function of genomic distance caused by PDS5A or PDS5B depletion (down).
• C
  Differential Hi‐C interaction maps displaying decreased (blue) and increased (red) interaction frequencies of a representative region on chr18 (20–35 Mb) respectively (left, right) after PDS5A or PDS5B depletion (up, down).
• D
  Compartment tracks for the entirety of chromosome 17 at 25 kb resolution as measured by the first principal component (PC1) values of PDS5A‐dTAG or PDS5B‐dTAG cells with DMSO or dTAG treatment.
• E
  Scatter plots of PC1 values in PDS5A‐dTAG or PDS5B‐dTAG cells before and after PDS5A or PDS5B depletion.
• F
  Changes of TAD structure after degradation of PDS5A or PDS5B.
• G
  Differential inter‐TAD and intra‐TAD interactions within or between (A and B) compartments after PDS5A or PDS5B depletion. Boxplots represent the first and third quartiles, with a line at the median. Whiskers extend to data values up to a 1.5‐fold interquartile range. Each group has 23 values. Every value is obtained from one chromosome. P‐values are calculated with the student's t‐test.

Then we analyzed active (A) and inactive (B) nuclear compartments, the genome‐wide compartment scores in 25 Kb bins across all chromosomes were highly correlated with only a small portion of A‐B (1.46 or 1.91%) and B‐A (2.89 or 2.44%) switches occurring after PDS5A or PDS5B depletion (Fig 3D and E). Compartment strength analysis suggested that PDS5A/B loss did not lead to an obvious change in the BA, BB, AB, and AA interaction frequencies except several pair interactions (Fig EV3D).

Looking at the layer of TAD, in PDS5A or PDS5B depleted cells, 27.4 or 17.7% of adjacent TADs were merged into larger TADs, presumably due to the increased long‐range inter‐TAD interaction frequencies, while 7.1 or 4.8% of TADs were split into two or more smaller TADs (Fig 3F), leading to moderately increased TAD size (Fig EV3E). To further illustrate TAD reshuffling, we divided TAD interactions into two groups (inter‐TAD and intra‐TAD) and three categories: within compartment A, within compartment B, between A and B compartments. Long‐range inter‐TAD interactions were significantly increased within or between A and B compartments after PDS5A or PDS5B loss, with more strengthened interactions within A compartments; PDS5A depletion also increased mid‐range intra‐TAD interactions with more enhanced interactions between AB compartments, in contrast, PDS5B depletion decreased intra‐TAD interactions with more weakened interactions within A compartments (Fig 3G). These observations, to some extent, were consistent with our result of chromosomal interactions as a function of genome distance and implied that PDS5A/B could regulate TAD interactions in a compartment‐specific manner. Besides, TAD boundaries strength analyses measured by insulation score showed no notable changes in boundaries located both within A/B compartments and between A and B compartments (Fig EV3F).

PDS5A and PDS5B stabilize chromatin loops and restrict loop expansion

To study the roles of PDS5A/B in chromosome organization at finer scales, we performed a systematic loop calling on the whole‐genome Hi‐C interaction map. In the absence of PDS5A or PDS5B, the number of detectable chromatin loops greatly decreased by 45% (from 1,227 to 672) or 56% (from 1,137 to 499) respectively (Figs 4A and EV4A), and the overall strength of chromatin loops was reduced from aggregate peak analysis (APA) for all loops after either depletion of PDS5 (Fig 4B). Moreover, the average loop size increased by 21% (from 2.86 to 3.45 Mb) or 80% (from 3.43 to 6.14 Mb) separately (Fig 4C). In cells depleted of PDS5A or PDS5B, the strength of common loops, the anchors of which were identical in dTAG and DMSO treated cells was significantly lessened (Fig EV4B and C). Given that loops were formed by cohesin extruding DNA loops between two given CTCF sites (Sanborn et al, ²⁰¹⁵; Fudenberg et al, ²⁰¹⁶), we asked whether the disappearance of chromatin loops caused by PDS5A/B loss coincided with the changes in cohesin or CTCF signals at loop anchors. Quantitative ChIP‐seq experiments for RAD21 and CTCF were performed. Intriguingly, PDS5A/B loss led to a marked decrease of RAD21 occupancy at the anchors of diminished loops (Fig 4D). Meanwhile, PDS5B loss reduced CTCF binding intensities at lost loop anchors while PDS5A had a small effect on that from the signal plot. Upon close inspection, among the anchors of lost loops caused by PDS5A depletion, there was a fraction displaying weakened CTCF signals, and we also noticed some anchors occupied with increased CTCF signals. Figure 4E exemplified the rewiring of chromatin loops at a selected chromosomal region, accompanied by the differential RAD21 and CTCF ChIP‐seq signals in PDS5A‐dTAG or PDS5B‐dTAG cells individually. When looking at a genome‐wide scale, we identified 41% (21,557/52,734) or 34% (16,090/47,566) of all RAD21 binding sites defined in DMSO treated PDS5A‐dTAG or PDS5B‐dTAG cells significantly down‐regulated after PDS5A or PDS5B depletion (Fig EV4D). To confirm this change was not indirectly caused by the change in the amount of cohesin on chromatin following PDS5A/B degradation, we performed chromatin fractionation experiments and found that depletion of PDS5A or PDS5B had little effect on the levels of chromatin‐bound RAD21 and CTCF (Fig EV4E).

Figure 4. PDS5A or PDS5B depletion leads to reduced chromatin loops and enlarged loop size.

• A
  Changes in the number of chromatin loops defined with HICCUPS at 25 kb resolution after PDS5A or PDS5B depletion.
• B
  Aggregate peak analysis (APA) showing average interaction counts around all loops identified in DMSO and dTAG treated PDS5A‐dTAG or PDS5B‐dTAG cells.
• C
  Density plots displaying the length distribution of loops identified in PDS5A‐dTAG or PDS5B‐dTAG cells before and after PDS5A and PDS5B degradation.
• D
  Changes of RAD21 and CTCF binding intensities at lost loop anchors caused by the degradation of PDS5A or PDS5B.
• E
  Representative region of chr 20 at 5 kb resolution showing loop rewiring after PDS5A or PDS5B depletion (left, right). Loops highlighted in black represent common loops that were persistent after PDS5A/B depletion, loops highlighted in red represent lost loops that were only detected in cells mock depleted of PDS5A/B. Below are the corresponding normalized ChIP‐seq signals for RAD21, CTCF and FLAG‐PDS5A and FLAG‐PDS5B.

Figure EV4. Depletion of PDS5A or PDS5B and co‐depletion of PDS5A and PDS5B cause cohesin re‐localization on loop anchors.

• A
  Changes in the number of loop anchors upon PDS5A or PDS5B loss.
• B
  Venn diagram showing the numbers of common loops defined both in DMSO and dTAG treated PDS5A‐dTAG or PDS5B‐dTAG cells, lost loops defined only in DMSO treated cells and the newly gained loops existing only in dTAG treated cells.
• C
  Aggregate peak analysis showing the weakened interaction strength around common loops after PDS5A or PDS5B degradation.
• D
  Heatmaps displaying the ChIP‐seq signal densities for significantly changed peaks of RAD21 upon depletion of PDS5A or PDS5B.
• E
  Immunoblotting analysis of the chromatin‐bound fraction and the nuclear supernatant fraction from PDS5A‐dTAG or PDS5B‐dTAG cells (left, right) treated with DSMO and dTAG.

Source data are available online for this figure.

Taken together, PDS5A/B regulated the formation or maintenance of chromatin loops via stabilizing cohesin association at loop anchors and restricted the expansion of DNA loops.

Co‐depletion of PDS5A and PDS5B reduces cohesin levels at loop anchors

A previous study indicated co‐depletion of PDS5A and PDS5B by siRNA decreased the number of detectable loops by 42.8% in HeLa cells through increasing the retention time and the amounts of cohesin on chromatin (Wutz et al, ²⁰¹⁷), without detailed characterization on how co‐depletion of PDS5 proteins affected cohesin dynamics on chromatin. To elucidate how the genomic distribution of cohesin re‐localized and whether the cohesin enrichment at loop anchors were decreased after joint depletion of PDS5A and PDS5B, we constructed PDS5A;PDS5B‐dTAG cells permitting simultaneous depletion of PDS5A and PDS5B by integrating another FKBP^12F36V to the N terminus of the endogenous PDS5B locus in PDS5A‐dTAG cells (Fig 5A). In agreement with previous studies, the level of chromatin‐bound RAD21 was increased upon co‐depletion of PDS5A and PDS5B (Fig 5B). We then performed calibrated ChIP‐experiments for RAD21 and found nearly 50% (24,718/50,111) of all RAD21 peaks identified in DMSO treated PDS5A;PDS5B‐dTAG cells significantly down‐regulated after both PDS5A and PDS5B loss (Fig 5C). Of them, 48% were originally detected as non‐significantly down‐regulated sites upon PDS5A or PDS5B loss, as exemplified in Fig 5D. Moreover, cohesin signals at all loop anchors (the positions were derived from DMSO treated PDS5A‐dTAG cells) were even more strongly reduced in cells co‐depleted of PDS5A and PDS5B (Fig 5E). We also identified there were 13,360 RAD21 sites significantly up‐regulated after joint depletion of PDS5A and PDS5B, either from new gained peaks or peaks that originally existed in DMSO treated cells, both with relatively low binding intensities (Fig 5F). This might be the reason why the amount of chromatin‐bound cohesin was increased upon PDS5A and PDS5B co‐depletion. Taken together, our results confirmed that PDS5 proteins could play dual roles in regulating cohesin dynamics on chromatin, and PDS5 proteins are required for proper cohesin establishment at loop anchors.

Figure 5. Co‐depletion of PDS5A and PDS5B reduces the levels of cohesin at loop anchors.

• A
  Validation of PDS5A;PDS5B‐dTAG cells for simultaneous depletion of PDS5A and PDS5B.
• B
  Chromatin fractionation assay showing the change of chromatin‐bound level of RAD21 or CTCF after co‐depletion of PDS5A and PDS5B.
• C
  Heatmaps displaying the ChIP‐seq binding intensities for significantly changed RAD21 peaks after joint PDS5 proteins depletion (left). Venn diagram showing RAD21 peaks identified in cells before and after co‐depletion of PDS5A and PDS5B.
• D
  Differential RAD21 ChIP‐seq signals of representative regions showing significantly down‐regulated peaks after co‐depletion of PDS5A and PDS5B.
• E
  Changes of RAD21 signals at all loop anchors defined in PDS5A‐dTAG cells or PDS5B‐dTAG cells with DMSO treatment after PDS5A and PDS5B co‐depletion, or either depletion. For PDS5A;PDS5B‐dTAG cells, we used the genomic positions of loop anchors identified in DMSO treated PDS5A‐dTAG cells.
• F
  Differential RAD21 ChIP‐seq signals of representative regions showing significantly up‐regulated peaks after co‐depletion of PDS5A and PDS5B.

Source data are available online for this figure.

PDS5A and PDS5B display a moderate and non‐redundant effect on gene expression

Our analyses of contribution to 3D genome folding implied that PDS5A/B might affect gene expression. This prompted us to identify differentially expressed genes (DEGs) after PDS5A/B degradation. We identified 435 misregulated genes in PDS5A depleted cells (Padjust < 0.05), with 229 being up‐regulated and 206 being down‐regulated (Fig 6A). In contrast, PDS5B depletion induced 47 DEGs, with 19 being up‐regulated and 28 being down‐regulated (Fig 6B). There were 24 overlapped DEGs with coherent change following PDS5A or PDS5B depletion (Fig 6C). Gene ontology enrichment analysis of PDS5A up‐regulated genes indicated a preferential effect on extracellular structure organization, while down‐regulated genes were related with RNA catabolic process, as for PDS5B depletion, down‐regulated genes indicated a preferential effect on vesicle docking, while up‐regulated genes had no preference (Fig 6D). Our findings indicated PDS5A and PDS5B could regulate gene expression and function through overlapping and discrete ways.

Figure 6. PDS5A and PDS5B regulate shared and distinct gene sets.

• A, B
  Volcano plots showing gene expression changes after PDS5A (A) and PDS5B (B) degradation. Genes with significant changes (Padj < 0.05) are colored blue (upregulated) or red (downregulated).
• C
  The overlap and discrepancy of PDS5A and PDS5B misregulated genes (left). Heatmap displaying Benjamini–Hochberg adjusted P‐value of shared genes down‐regulated and up‐regulated after PDS5A or PDS5B depletion (right).
• D
  Gene ontology enrichment analysis of PDS5A/B differentially expressed genes.
• E, F
  Boxplot quantifying misregulated genes caused by PDS5A depletion around differential loops (E) or reshuffled TAD regions (±250 kb) (F). Boxplots represent the first and third quartiles, with a line at the median. Whiskers extend to data values up to 1.5‐fold interquartile range. Each group has 100 values. Each value is obtained with a random sampling with replacement. P‐values are calculated with the student's t‐test.

Then, we attempted to link misregulated gene expression with the altered chromatin structure described above. We did not find a generalized relationship between the DEGs and differential chromatin loops or reshuffled TADs, and we could not readily distinguish the direct transcriptional effect of PDS5A or PDS5B depletion from the indirect one because of the limitations of our study. But we observed that down‐regulated genes had a higher probability to locate proximal to the disappeared chromatin loops than any unchanged gene sets, and up‐regulated genes were significantly enriched in regions suffering either disrupted or gained TAD boundaries (Fig 6E and F).

Discussion

In this study, we first reveal the genomic distribution pattern of PDS5A/B, which is highly correlated with cohesin and CTCF binding sites thereby providing the regulatory foundation for cohesin dynamic and cohesin‐mediated chromosome folding. Then we use the inducible dTAG degron system to perform loss‐of‐function studies on PDS5A and PDS5B. We investigate the individual contributions of PDS5A and PDS5A on chromatin structure at different layers. In the absence of PDS5A or PDS5B, long‐range interactions increase resulting in fewer but larger TADs, and nearly half of the chromatin loops disappeared accompanied by increased loop size, while the A/B compartment remains largely stable. This is consistent with the notion that compartments arise from different mechanisms independent of cohesin and CTCF (Nora et al, ²⁰¹⁷; Schwarzer et al, ²⁰¹⁷). Our findings in PDS5A or PDS5B deficient cells, to some degree, are in agreement with the phenotypes observed in PDS5A and PDS5B co‐depleted cells from a different context or in Pds5 deficient yeast (Wutz et al, ²⁰¹⁷; Dauban et al, ²⁰²⁰). We set out to explore the mechanism and find that PDS5A or PDS5B loss causes cohesin establishment defects on loop anchors. A marked decrease of RAD21 enrichment especially at loop anchors originally co‐bound by RAD21 and PDS5A/B is observed after PDS5A or PDS5B degradation, leading to less loop formation and weakened loop strength. PDSA5/B is thought to play both negative and positive roles in regulating the dynamic association of cohesin on chromatin through partnering with WAPL and sororin (Nishiyama et al, ²⁰¹⁰; Chan et al, ²⁰¹²). The influence of either depletion of PDS5 protein on the Esco1/2‐sororin axis may be dominant over the influence on WAPL‐mediated release activity. Hence depletion of PDS5A/B tends to dissociate cohesive cohesin from chromatin. The molecular determinants are yet to be understood. Furthermore, our findings provide a possible explanation for the previously reported paradoxical phenotypes observed in cells co‐depleted of PDS5A and PDS5B. Though the retention time and the amount of cohesin on chromatin are increased, but cohesin signals at loop anchors are strongly reduced, resulting in less detectable loops in PDS5A and PDS5B co‐depleted cells. The studies on co‐depletion of PDS5A and PDS5B reinforce our understanding of the mechanism by how PDS5 proteins regulate cohesin‐mediated chromatin loops. In these processes mentioned above, the functions of PDS5A and PDS5B are overlapping. The interplay between PDS5A and PDS5B can partially account for their overlapping function, as PDS5A depletion significantly decreases PDS5B enrichment on chromatin, while PDS5B has a mild influence on PDS5A.

WAPL limits cohesin's residence time on chromatin thereby counteracts loop extension (Haarhuis et al, ²⁰¹⁷; Wutz et al, ²⁰¹⁷). In Saccharomyces cerevisiae, it is reported that Esco1 inhibits cohesin translocation to form long loops and Pds5 negatively regulates loop size via EscoI‐dependent and Wapl‐dependent manners (Dauban et al, ²⁰²⁰). The mechanisms regulating the dynamic of the cohesin‐mediated loop are very conserved between mammals and yeast. Our findings demonstrate that PDS5 can restrict loop expansion in mammals. We speculate that the loop expansion detected in PDS5A or PDS5B depleted cells, or in previously reported PDS5A and PDS5B co‐depleted cells (Wutz et al, ²⁰¹⁷), could be due to the reduced retention of cohesin on loop anchors, which in turn serves as physical barriers for loop translocation beyond the loop extruding factor. Other possibilities may be the competition between PDS5 and SCC2. A previous study indicates the replacement of PDS5 bound cohesin by SCC2 bound promotes cohesin's ATPase activity required for loop translocation energy (Petela et al, ²⁰¹⁸). Furthermore, as has been pointed out, CTCF defines loop anchors and restricts loop expansion (Wutz et al, ²⁰¹⁷). Degradation of PDS5A or PDS5B increases loop size by 21 or 80% on average respectively, we attribute the more remarkable phenotype of PDS5B depletion on loop expansion to the decreased CTCF signals at loop anchors. As for PDS5A depletion, we find there are some cases of lost loop anchors displaying weakened CTCF signals, and some are occupied with increased signals, making the average intensities of CTCF at loop anchors barely changed. And as discussed in‐depth by the authors, PDS5 may define loop anchors together with CTCF considering the phenotypes of PDS5A and PDS5B co‐depletion on loop formation are quite similar to CTCF depletion (Wutz et al, ²⁰¹⁷). Our study provides further evidence for this concept and emphasizes the contribution of PDS5B more than PDS5A to cooperate with CTCF in setting loop anchors. A recent study indicates the CTCF N terminus could directly interact with PDS5A/B via a conserved motif shared with WAPL and sororin, implying an underlying function interplay between PDS5 and CTCF (Nora et al, ²⁰²⁰). The molecular and structural basis that directs this remains to be determined. Altogether, our findings decipher the regulation of PDS5 in cohesin‐mediated chromosome loops and provide new perspectives on the nature of cohesin biology and DNA loops.

Finally, our studies show that PDS5A and PDS5B have a moderate and non‐redundant effect on gene expression, they can regulate the transcription of shared and different gene sets. Among the DEGs that responded to PDS5A depletion, there are many genes related to RNA catabolic process, implying that PDS5A can regulate gene expression via post‐transcriptional paths. The relationship between 3D genome structure and transcription is controversial. Mounting studies suggest dramatically altered genome topology only plays a modest role in transcription, the misappropriate enhancer – promoter interaction event directly influenced by the depletion of architectural regulators like cohesin and CTCF is rare in a given cell type (Nora et al, ²⁰¹⁷; Rao et al, ²⁰¹⁷; Ghavi‐Helm et al, ²⁰¹⁹). Prolonged depletion of these factors with the degron system may cause a more noticeable effect on gene expression indirectly. In addition, it's proposed that a combination of 3D genome, epigenome, and transcription factors together determine the transcriptional activation state of a target gene (Fulco et al, ²⁰¹⁹; Kim & Shendure, ²⁰¹⁹).

Materials and Methods

Cell culture and reagents

PLC/PLF/5 was purchased from ATCC, and cell lines expressing FKBP12^F36V‐tagged PDS5A and PDS5B were cultured in DMEM (BI, 01‐052‐1A), supplemented with 10% fetal bovine serum (Hyclone, GE Healthcare SH30068.03) and 1% Pen‐Strep. Antibodies and reagents used in this study are: PDS5A (Bethyl, A300‐089A; Proteintech, 17485‐1‐AP), PDS5B (Bethyl, A300‐537A), FLAG (Sigma F1804, Abmart, M20008M), RAD21 (Abcam, ab992), CTCF (Millipore, 07‐729), Beta Tubulin (Proteintech, 10094‐1‐AP), Histone H3 (ABclonal, A2348), Rabbit IgG (CST, 2729S), biotin‐14‐dATP (picoHelix, 312421929; Life Technologies, 19524016), dTAG‐13 (Tocris, 6605).

Genome editing for dTAG endogenous knock‐in

The CRISPR/Cas9 system was used to genetically engineer PLC/PLF/5 cells. Oligos coding guide RNA targeting the exon 1 of PDS5A or PDS5B were cloned into the CRISPR/Cas9 plasmid (PX459) as described (Ran et al, ²⁰¹³). PITCH donor vector with sgRNA cutting sites and microhomology‐containing dTAG repair template was constructed. To generate PDS5A‐dTAG or PDS5B‐dTAG endogenous tagged cell lines, 3ug sgRNA/Cas9 and PITCH donor plasmids were transfected into 1 × 10⁶ PLC/PLF/5 cells with lipofectamine 3000 (Invitrogen). Cells were plated in a 10 cm culture dish with a medium containing 2 μg/ml puromycin at 72 h post‐transfection. After approximately 10 days of cell growth and continuous selection with puromycin, individual colonies were picked using a stereoscope and seeded into a 96‐well plate. Homogeneous knock‐in clones were verified by PCR using genomic DNA as a template, the protein level of target genes and degradation efficiency was tested by western blotting. To induce PDS5A/B degradation, 200 nM dTAG‐13 (final concentration) dissolved in DMSO was added to the cells. For all experiments related to next‐generation sequencing, PDS5A‐dTAG or PDS5B‐dTAG cells were harvested 24 h after the treatment of dTAG and DMSO.

To generate PDS5A;PDS5B‐dTAG cells permitting simultaneous depletion of PDS5A and PDS5B, we first modified the original PITCH vector for PDS5B by replacing the puromycin resistance with hygromycin resistance using homologous recombination. SgRNA/Cas9 and the modified PTICH vector for tagging PDS5B were transfected into PDS5A‐dTAG cells. The following procedures were the same as the construction of cells depleted of PDS5A or PDS5B except for hygromycin at a concentration of 300 μg/ml was used for resistance selection instead of puromycin.

The specific gRNA sequences used for targeting the exon 1 of PDS5A or PDS5B and primers sequences for genomic DNA PCR are listed in Table EV1.

Chromatin fractionation

Collected cell pellets were resuspended and swelled with low salt buffer containing 10 mM HEPES (pH 7.5), 1.5 mM MgCl₂, 10 mM KCl, 0.5 mM DTT, 1.0 mM PMSF, and protease inhibitor cocktail on ice for 15 min. NP‐40 was added to a final concentration of 0.25% for 5 min to break the plasm membrane. Nuclei pellets were collected by centrifugation at 400 g for 5 min. Then equal volume high salt buffer consisting of 20 mM Hepes (pH 7.5), Glycerol 10% (V/V), 0.42 M KCl, 4 mM MgCl₂, 0.2 mM EDTA, 0.5 mM DTT, 1.0 mM PMSF and protease inhibitor cocktail as low salt buffer was added to the pellet, left in ice for 20 min and vortexed every 5 min. Insoluble chromatin pellets were separated and collected by centrifugation at 16,000 g for 15 min. The supernatant was the soluble nuclear fraction.

ChIP, ChIP‐seq and data analyses

ChIP assays were carried out as described previously (Lan et al, ²⁰⁰⁷). Briefly, indicated growing cells were collected and crosslinked with 1% formaldehyde for 10 min at room temperature, followed by quenching with 125 mM glycine for 5 min, washed twice with 1× phosphate‐buffered saline (PBS), resuspended and lysed in ChIP lysis buffer consisting of 500 mM NaCl, 50 mM HEPES (pH 7.5), 1% Triton X‐100, 0.1% SDS, 0.1% Na‐Deoxycholate, 1 mM EDTA, 1.0 mM PMSF and protease inhibitor cocktail (Complete EDTA‐free, Roche). Chromatin was sheared to an average length of 300–700 bp with a Qsonica sonicator. We performed calibrated ChIP‐seq experiments for RAD21, CTCF, PDS5A and PDS5B in PDS5A‐dTAG or PDS5B‐dTAG cells with different treatments by spiking‐in sonicated chromatin from mouse cells at a proportion of 1:3 initially. After overnight incubation with specific antibodies and Dynabeads™ Protein A and G beads (Invitrogen), precipitated chromatin was washed, de‐crosslinked and digested with proteinase K. The resulting DNA was then purified using a PCR extraction kit (QIAGEN) and eluted in TE buffer containing 10 mM Tris–HCl (pH 8.0). The DNA elutes were either used for ChIP‐qPCR or processed for sequencing. ChIP‐qPCR was performed with ChamQ Universal SYBR qPCR Master Mix (Vazyme) on ABI7500. ChIP‐qPCR primer sequences for H19/IGF2 locus and the promoter region of PCDH10 are listed in Table EV1. DNA sequencing libraries were prepared with VAHTS™ Universal DNA Library Prep Kit for Illumina (Vazyme, ND607). After amplification and size selection with Agencourt AMPure XP beads (Beckman Coulter), the size distributions were determined with Agilent 2100 Bioanalyzer. Equimolar pools of libraries were sequenced with Illumina HiSeq X Ten (150 bp, pair‐end; Annoroad Gene Technology, Beijing, China).

For ChIP‐seq data processing, the data quality of raw reads was qualified by FastQC v0.11.5, then FASTQ data were mapped to the human genome (hg19) and mouse genome (mm9) using the Bowtie v1.2.3 (−y ‐k 1 ‐M 1 ‐‐chunkmbs 2048 ‐S ‐‐best –shmem; Langmead et al, ²⁰⁰⁹). Duplicates were removed with samtools v1.9 (Li et al, ²⁰⁰⁹). Regular peak calling was performed using MACS2 with a threshold q‐value < 0.01 (−q 0.01; Zhang et al, ²⁰⁰⁸). Overlap between peaks from different proteins was performed using the “intersect” option of BEDTools v2.27 with a minimum of 1 nt overlap (Quinlan & Hall, ²⁰¹⁰). Genomic distribution of ChIP‐seq peaks relative to reference genes was performed using the Cis‐regulatory Element Annotation System (CEAS; Shin et al, ²⁰⁰⁹). To unbiasedly compare the enrichment of each protein under different treatments, the spike‐in strategy was used and scaleFactor was calculated. A normalized bigwig file was generated using “bamCompare” from deepTools v3.3.0 and all heatmaps were generated with deepTools (Ramírez et al, ²⁰¹⁶). To define significantly changed RAD21 peaks after the depletion of PDS5A or PDS5B, or co‐depletion of PDS5A and PDS5B, peaks identified in DMSO and dTAG treated cells were merged, then log2 fold change of averaged binding intensities for each peak was calculated, abs (log2 fold change) > 1 was used as a threshold. The detailed ChIP‐seq information is present in Table EV2.

Sequential ChIP (Re‐ChIP)

Chromatin for the first ChIP of Re‐ChIP was prepared as described above. We used 5 μg antibodies against PDS5A or FLAG (PDS5B) to perform the first ChIP experiments from PDS5B‐dTAG cells. 2 × 10⁷ cells were used per ChIP reaction. ChIP DNA was eluted in re‐ChIP elution buffer consisting of 50 mM Tris–HCl (pH 8.0), 10 mM EDTA, and 2% SDS for 30 min at 37°C to denature the antibodies and the elutes were diluted 10 times in 1× ChIP lysis buffer. Then 2 μg antibodies against PDS5A, FLAG (PDS5B), RAD21 or IgG were added to the elutes and samples were performed for the second ChIP experiment. RAD21 served as a positive control. Fold enrichment over IgG was used to quantify the enrichment of the indicated antibody at a specific site. A list of primer sequences for ChIP‐qPCR can be found in Table EV1.

In situ Hi‐C and data analyses

The in situ Hi‐C libraries were generated as previously described with a few modifications (Rao et al, ²⁰¹⁴). Briefly, for each cell sample, 2–3 × 10⁶ cells were harvested and crosslinked using 1% formaldehyde, followed by quenching with 125 mM glycine for 5 min. Cell pellets were collected by centrifugation, washed with PBS and immediately frozen in liquid nitrogen and stored at 80°C. After thawing, cells were penetrated and nucleus restriction reaction was performed overnight using DpnI instead of MboI enzyme, digested DNA fragments were filled with biotin at 37°C for 1 h, and proximally ligated at RT for 4 h. Then DNA was de‐crosslinked and purified with ethanol precipitation. After shearing to 300–700 bp fragments with Qsonica sonicators, the biotin‐tagged DNA was captured by streptavidin T1 beads (Life Technologies) followed by end repairing, adaptor ligation and PCR amplification on beads. The resulting DNA libraries were sequenced to > 200 million reads pairs on Illumina Nova 6000 system (150 bp, pair‐end; Annoroad Gene Technology, Beijing, China).

For data analyses, raw sequencing reads were processed using HiC‐Pro v3.1.0 (Servant et al, ²⁰¹⁵). Positions of di‐tags were mapped against the human genome (hg19). Experimental artifacts, like self‐circles, re‐ligations, and duplicates, were filtered out. Two biological replicates were taken for each condition, and correlation between replicates was done using “hicCorrelate” function from HiCExplorer (v2.2.1). After ensuring replicate reproducibility, qualified reads from the same treatment were merged for all downstream analyses. The number of valid pairs for each condition was randomly sampled to the same. All the analyses were performed using 25 kb resolution except explicitly mentioned. AllValidPairs from HiC‐Pro output were transferred into .hic file using “hicpro2juicebox.sh” option from HiC‐Pro. Juicebox v1.11.08 was applied to visualize interaction matrix with the input of .hic file (Durand et al, ^2016a). Interaction frequency as a function of genomic distance curve was plotted by “analyzeHiC” from Homer (−res 25,000 ‐vsGenome ‐nomatrix ‐ifc –chopify; Heinz et al, ²⁰¹⁸). To segregate A and B compartments, we performed eigenvalue decomposition of the Hi‐C matrix with “runHiCpca.pl” from Homer (−res 25,000 ‐window 50,000). TADs were called using “hitad” function from TADlib and topological domains with level 0 were defined as TADs (Wang et al, ²⁰¹⁷). Insulation scores were calculated using “findTADsAndLoops.pl” from Homer (−res 25,000 ‐window 125,000). Loops were called by “hiccups” from juicer v1.22.01 (−m 1,024 ‐r 25,000 ‐k KR ‐f 0.1 ‐p 1 ‐i 3 ‐d 50,000 ‐t 0.02,1.5,1.75,2; Durand et al, ^2016b). Common loops were defined if both the anchors of the loops were located within 50 kb in DMOS and dTAG‐treated cells. Aggregate peak analysis (APA) was performed by counting the average interaction frequencies for any given 2D coordinates. The interaction frequency matrices were generated by extending 250 kb up and downstream for each of the two anchors. The detailed information for Hi‐C data is present in Table EV3.

RNA‐seq, and data analyses

Total RNA was extracted from cells with Trizol reagent (Life Technologies). The RNA quality was assessed using the Agilent 2100 Bioanalyzer and mRNA was purified with VAHTS mRNA capture beads (Vazyme). Strand‐specific libraries were prepared with VAHTS Universal V6 RNA‐seq Library Prep Kit for Illumina (Vazyme), and the libraries were sequenced with Illumina HiSeq X10 (150 bp, pair‐end).

For bioinformatics analyses, raw reads from RNA‐seq were mapped to the human genome (hg19) and GENECODE v38 using hisat2 v2.2.1 (Kim et al, ²⁰¹⁵). Quantification for each expressed gene was performed using featureCounts from Rsubread v2.0.1 (−p ‐B ‐C) (Liao et al, ²⁰¹⁹). Differentially expressed genes were calculated by DESeq2 v1.32.0 with cutoffs as follows: Padj < 0.05 (Love et al, ²⁰¹⁴). Functional annotations were done using enrichGO from clusterProfiler package (Yu et al, ²⁰¹²).

To evaluate if the differentially expressed genes caused by PDS5A depletion were enriched in altered chromatin structures like differential loops or TAD boundaries. We used a bootstrapping strategy, in brief, 100 random sampling with replacements were performed for gene sets that were up‐regulated or down‐regulated, then we calculated the number of genes that were located in the regions suffering loop and TAD rearrangements in each round. Unchanged gene sets were processed as the same procedure and used as control (expected). We did not apply this approach to DEGs caused by PDS5A depletion, because the number of misregulated genes was too small.

The detailed information for RNA‐seq data is present in Table EV2. The differentially expressed genes upon PDS5A or PDS5B depletion are listed in Datasets EV1 and EV2.

Author contributions

Dingdang Yu: Conceptualization; investigation; writing – original draft; writing – review and editing. Guoyu Chen: Data curation; writing – original draft; writing – review and editing. Yuci Wang: Data curation; visualization. Yining Wang: Investigation. Risheng Lin: Investigation. Nanbo Liu: Visualization. Ping Zhu: Resources; funding acquisition. Hang Liu: Visualization. Tao Hu: Resources. Rui Feng: Investigation. Haizhong Feng: Resources. Fei Lan: Resources. Jiabin Cai: Resources; supervision; project administration. Hao Chen: Supervision; funding acquisition; writing – review and editing.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Supporting information

Expanded View Figures PDF

Table EV1

Table EV2

Table EV3

Dataset EV1

Dataset EV2

Source Data for Expanded View

PDF+

Source Data for Figure 1

Source Data for Figure 2

Source Data for Figure 5

EMBO reports (2022) 23: e54853

Data availability

All next‐generation sequencing data sets generated in this study have been deposited in NCBI Gene Expression Omnibus (GEO) database with accession number GSE209849 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE209849).