Brca2 opposes the role of Pds5 in sister chromatid cohesion

The discovery that Pds5 interacts with Brca2 in a complex that lacks Wapl [10–12] raised the question of whether or not Brca2 plays a role in sister chromatid cohesion. RNAi-mediated depletion of Brca2 in cultured ML-DmBG3-c2 (BG3) cells indicates that it counters the role of Pds5 in establishing or maintaining cohesion. The BG3 clonal cell line derived from larval central nervous system is diploid male, allowing accurate measurement of sister chromatid cohesion defects in metaphase chromosome spreads [14]. Fig 1A shows that as expected, depletion of Pds5 (iPds5) causes precocious sister chromatid separation (PSCS) with ~70% of chromosomes showing partial or complete separation after 5 days of treatment, and Wapl depletion (iWapl) gives less than 10% PSCS, similar to mock-treated cells.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Pds5 and Brca2 oppose each other in sister chromatid cohesion in BG3 cells.

(A) The three micrographs show examples of normal metaphase chromosomes and of precocious sister chromatid separation (PSCS). The bar graph shows the percent of chromosomes showing normal cohesion (blue) or partial or complete PSCS (red) in mock-treated (Mock), and Pds5-depleted (iPds5), Wapl-depleted (iWapl) or Brca2-depleted (iBrca2) BG3 cells after 5 days of RNAi treatment. (B) The graphs show the mean percentage of chromosomes showing PSCS in individual cells after three to five days of RNAi treatment for Pds5 only (iPds5, blue) or for both Pds5 and Brca2 (iPds5 + iBrca2, red). Error bars are standard errors of the mean. A minimum of 50 metaphase nuclei were scored for each individual group and time point. Similar results were obtained in two additional experiments.

https://doi.org/10.1371/journal.pgen.1007225.g001

Brca2 depletion (iBrca2) alone does not alter PSCS frequency (Fig 1A) but if Brca2 is co-depleted with Pds5 (iPds5 iBrca2) PSCS frequency is significantly reduced compared to Pds5 depletion alone (Fig 1B). Pds5 depletion reduces cell proliferation after 5 days of RNAi treatment, presumably reflecting increasing aneuploidy as observed in pds5 mutant larval neuroblasts [7] while there is no discernable effect of Wapl or Brca2 depletion on cell growth. These results indicate that Brca2 opposes the role of Pds5 in sister chromatid cohesion.

Brca2 is unlikely to oppose Pds5 function indirectly through effects on the levels of other sister chromatid cohesion proteins. Western blots of whole cell extracts illustrating Pds5, Wapl and Brca2 depletion are shown in S1 Fig. These westerns also show that Pds5 depletion does not significantly alter the total levels of Nipped-B, Wapl or the cohesin subunits examined. Brca2 depletion does not increase or decrease the levels of Pds5, although Pds5 depletion modestly reduces Brca2 levels.

In RNA-seq experiments described in more detail in a later section, Pds5 depletion increases Brca2 gene transcripts 1.4-fold (S1 Fig, panel I) although Pds5 and cohesin do not bind the Brca2 gene, indicating an indirect effect. The modest increase in Brca2 transcripts indicates that the partial decrease in Brca2 protein caused by Pds5 depletion likely represents destabilization of Brca2 protein in the absence of Pds5. Other possibilities, such as altered transport or translation of Brca2 mRNA cannot be ruled out.

The RNA-seq experiments also show that Pds5 and Brca2 depletion have similar modest effects on RNA levels for some genes encoding sister chromatid cohesion proteins, and indicate that Pds5-Brca2 co-depletion does not suppress sister chromatid cohesion defects by altering expression of these genes. Prior studies show that 80 to 90% depletion of Nipped-B or the SA or Rad21 cohesin subunits in BG3 cells does not measurably alter sister chromatid cohesion [14]. As shown in S1I Fig the effects on cohesion factor transcript levels are all less than 2-fold, making it unlikely that they are involved in the substantial changes in sister chromatid cohesion observed in the Pds5 depletion and their suppression in the Pds5-Brca2 co-depletion. Pds5 and Brca2 depletion both modestly reduce Nipped-B RNA levels by 30 to 40%, and increase transcripts of the Smc1, Smc3, wapl, eco, and dmt (dalmatian, encoding a sororin-shugoshin fusion protein [15, 16]) gene up to 1.5-fold, none of which explains the opposing effects of Pds5 and Brca2 on sister cohesion. The only significant effect of the double Pds5-Brca2 depletion that is not observed in the single depletions is a 40% reduction in SA transcripts, which should reduce, not rescue cohesion if it were of sufficient magnitude to actually have an effect (S1I Fig). No other known cohesion factor gene transcript levels are significantly affected by Pds5 and/or Brca2 depletion (S1I Fig).

The above findings support the idea that Brca2 directly opposes the role of Pds5 in sister chromatid cohesion, possibly reflecting their interaction with each other. One possibility is that Brca2 sequesters the Pds5 needed for cohesion away from cohesin, and that Brca2 depletion allows re-localization of the residual Pds5 remaining after depletion back to cohesin. As shown below in a series of experiments, we find that opposing effects of Pds5 and Brca2 on binding of the SA cohesin subunit to DNA replication origins provides a more likely explanation.

Pds5 and Brca2 chromosome binding patterns differ from cohesin and Wapl

To gain insights into how Pds5, Wapl and Brca2 influence cohesin function and sister chromatid cohesion we mapped them genome-wide in BG3 cells by chromatin immunoprecipitation with high-throughput sequencing (ChIP-seq). As described elsewhere for Nipped-B and cohesin [17, 18] we used at least three biological replicates for each protein, sequencing each replicate to a minimum of 10X genome coverage, normalizing each to input chromatin sequenced to at least 45X genome coverage. In Drosophila, cohesin and Nipped-B occupancy spread broadly over several kilobases, and ultra-deep sequencing and calculating enrichment relative to input chromatin instead of calling peaks are vital for accurate and reproducible results [18]. Enrichment relative to input chromatin was calculated using 250 bp sliding windows to generate values every 50 base pairs, thereby reducing noise and facilitating direct comparisons of replicates and downstream calculations. The biological replicates showed high genome-wide Pearson correlations (0.65 to 0.9) and were averaged for further analysis. Preimmune serum precipitations show no regions with significant enrichment using these procedures, indicating an absence of methodological artifacts. S2A Fig shows examples of the correlations between three SA ChIP-seq biological replicates and S2B Fig shows a genome browser example of ChIP-seq with preimmune serum compared to multiple ChIP-seq experiments.

We compared the Pds5, Wapl and Brca2 binding patterns to the Rad21 cohesin subunit, and prior ChIP-seq data [17] for the SA cohesin subunit and Nipped-B. As illustrated by the kayak gene region in Fig 2A, Wapl displays a broad spreading pattern nearly indistinguishable from cohesin. Nipped-B differs somewhat from cohesin and Wapl with noticeable broad peaks superimposed on the spreading pattern. Pds5 and Brca2 show substantially less spreading, being more restricted to broad peaks that co-localize with Nipped-B peaks (Fig 2A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Cohesin and the accessory factors localize to DNA replication origins in BG3 cells.

(A) Genome browser view of Brca2, Pds5, Wapl, Rad21, SA, Nipped-B and Rpb3 ChIP-seq at the kayak locus containing an early DNA replication origin. The scales are log2 enrichment. David MacAlpine provided the processed BG3 early DNA replication data (GEO accession GSE17287) and the scale is MA2C score. The SA, Nipped-B, and Rpb3 ChIP-seq data are published elsewhere [17, 19]. (B) Meta-origin analysis of the ChIP-seq data using the 78 strongest early DNA replication origins (positions in S1 File). The mean enrichment in 10 kb bins was calculated from -100 kb to +100 kb from the origin centers.

https://doi.org/10.1371/journal.pgen.1007225.g002

The restricted Pds5 localization relative to Wapl was unexpected given that Pds5 and Wapl form a complex and can bind together to cohesin [20–22]. Our findings indicate that Wapl also binds cohesin independently of Pds5 in a location-dependent manner. Independent Wapl binding is consistent with the finding the N terminal region of human Wapl binds to the C terminal region of Rad21 in the presence of SA and absence of Pds5 [23].

Pds5 and Brca2 occupancy overlap with each other, as would be expected for them to interact with each other on chromosomes. Using a threshold of enrichment in the 95^th percentile or greater over regions ≥300 bp in length to call binding, there are 6,452 Pds5 euchromatic sites and 6,430 Brca2 euchromatic sites of variable length, 3,600 (56%) of which overlap. The actual overlap is greater because many sites show significant enrichment for the other protein below the 95^th percentile. Examples of binding regions called at the 95^th percentile are shown in S2B Fig. Although Pds5 and Wapl binding strongly overlap, the relative enrichment values differ along the chromosome, indicating that they do not bind stoichiometrically as a complex. For example, at kayak, the strongest Pds5 sites are not the strongest Brca2 sites and vice versa (Fig 2A). Visual inspection in a browser also reveals a minor number of sites enriched only for Brca2 or for Pds5, which are typically short regions with low enrichment (S2B Fig, asterisks). The vast majority of Brca2 and Pds5 binding sites occur within broad regions occupied by cohesin and Wapl, with a few exceptions of small peaks with lower enrichment (S2B Fig, daggers).

Sister chromatid cohesion factor occupancy centers at early DNA replication origins

Visual inspection in a browser revealed that most sites with high Pds5 and Brca2 occupancy in BG3 cells are located within a few kilobases of early DNA replication origins. Sister chromatid cohesion is established during DNA replication, and cohesin is enriched and loaded at replication origins and sites in Drosophila, yeast and vertebrates [24–31]. We thus conducted an origin meta-analysis of all cohesion factor ChIP-seq data, finding that the highest Nipped-B, cohesin, Pds5, Wapl and Brca2 levels all center at origins. The locations of early DNA replication origins in BG3 cells were reported by the modENCODE project and confirmed by mapping pre-replication complex (pre-RC) binding sites [24]. As shown in Fig 2A, a replication origin is located within multiple active genes that bind RNA polymerase II (Rpb3) [19] in the kayak region. We averaged the ChIP-seq enrichment for cohesin and accessory factors in non-overlapping 10 kb bins from -100 kb to +100 kb outward from the centers of the 78 strongest early origins in BG3 cells (positions in S1 File). As revealed in Fig 2B, all show maximal occupancy near the metaorigin center, with the levels decreasing outward in both directions.

The ChIP-seq experiments were conducted with asynchronously dividing cells, and although the cohesion factor binding centers at replication origins, most cells (~85%) are in interphase, in which cohesin binds specifically to active genes and enhancers [32, 33]. Thus the metaorigin analysis reveals that the further an active gene is from the origin, the less likely it is to bind cohesin. For instance, kayak under the origin center binds cohesin, and yata some 20 kb from the origin center binds little, even though it is transcribed (Fig 2A). Inactive genes and intergenic regions located between an active gene that binds cohesin and the origin do not bind cohesin. Thus cohesin binding is discontinuous under most origins and in the origin-flanking regions, with active genes close to the origin binding high cohesin levels, and genes further away binding less or none. The cohesin pattern revealed by the metaorigin analysis thus does not directly measure binding to origins per se, but shows that origins have a strong positive influence on cohesin binding to active genes in their vicinity.

Pds5 and Wapl control cohesin localization around replication origins

Pds5, Brca2 and Wapl all have different effects on sister chromatid cohesion. Sister chromatid cohesion is established during DNA replication, and above we show that binding of these regulatory factors centers at early replication origins. This led us to hypothesize that Pds5, Wapl and Brca2 differentially affect sister chromatid cohesion because they differentially influence cohesin binding at origins. We thus compared their effects on cohesin binding by metaorigin analysis. Cohesin ChIP-seq was performed for BG3 cells depleted for Pds5, Brca2 or Wapl using multiple biological replicates for each depletion. As noted above, depletion of each of these proteins has no significant effects on the levels of each other than a modest decrease in Brca2 with Pds5 depletion (S1 Fig). Also as noted above, Wapl and Brca2 depletion have no measurable effect on cell division, but Pds5 depletion, which causes loss of sister chromatid cohesion, causes cells to stop dividing after five days of RNAi treatment, so chromatin was isolated prior to five days.

These experiments revealed that depletion of Pds5 or Wapl, but not Brca2, increases the distance that the domains of cohesin occupancy extend outward from origins. Fig 3A shows that Pds5 depletion (iPds5) increases spreading of the Rad21 and SA cohesin subunits, Wapl, and Nipped-B in the regions flanking the kayak replication origin. Metaorigin analysis of the Rad21 data shows that this effect is global, with statistically significant increases in occupancy for tens of kilobases in the flanking regions (Fig 3B). Wapl depletion (iWapl) similarly increases cohesin spreading around origins as illustrated for the SA cohesin subunit at kayak locus in Fig 4A and metaorigin analysis (S3A Fig). Brca2 depletion (iBrca2) does not increase the size of cohesin domains (Fig 4A). Although Pds5 depletion reduces chromatid cohesion, Wapl depletion does not. We thus conclude that the expansion of cohesin territories does not reflect loss of sister cohesion. Because Wapl depletion has no measurable effect on cell division, which is reduced by Pds5 depletion, we also conclude that changes in the cell cycle do not cause the cohesin domain expansion.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Depletion of Pds5 causes extension of cohesin and accessory factor binding domains surrounding replication origins.

(A) Genome browser view of the kayak origin region, showing ChIP-seq for Wapl, Rad21, SA, and Nipped-B in control cells and cells depleted for Pds5 (iPds5). Shaded areas show regions with increased binding of cohesin, Wapl and Nipped-B in Pds5-depleted cells. (B) The left panel shows the Rad21 meta-origin analysis in mock control cells (Rad21, blue) and cells depleted for Pds5 (Rad21 iPds5, red). The right panel shows the–log10 p values for differences in Rad21 enrichment in each bin used for the meta-origin analysis. P values were calculated using the Wilcoxon signed rank test.

https://doi.org/10.1371/journal.pgen.1007225.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Pds5 and Brca2 have opposing effects on SA cohesin subunit levels at DNA replication origins.

(A) Genome browser view of SA ChIP-seq at the kayak locus in mock-treated BG3 cells (SA) and BG3 cells RNAi depleted for Pds5 (iPds5), Wapl (iWapl), Brca2 (iBrca2), Pds5 and Wapl (iPds5 iBrca2) and Pds5 and Brca2 (iPds5 iBrca2). (B) Metaorigin analysis of the SA to Rad21 ChIP-seq enrichment ratio in mock control cells (Mock, blue) and cells depleted for Pds5 (iPds5, red). (C) The top panel is the metaorigin plot of SA enrichment in control (SA, blue) and Pds5-depleted cells (SA iPds5, red). The bottom panel is the meta-origin plot of–log10 p values for the difference in SA enrichment calculated using the Wilcoxon signed rank test. (D) Same as C except for cells depleted for Brca2 (iBrca2). An example of the increase in SA at the kayak locus is shown in S2C Fig. (E) Same as C with cells depleted for both Pds5 and Brca2 (iPds5 iBrca2).

https://doi.org/10.1371/journal.pgen.1007225.g004

The expansion of cohesin territories with Pds5 depletion, as measured by Rad21 ChIP-seq, indicates an increase in the total amount of cohesin bound to chromosomes, consistent with in vivo cohesin FRAP data in heterozygous pds5 mutants [9]. Metaorigin analysis, however, indicates that the increase is regional, occurring primarily at locations more distant from replication origins. The increase in cohesin at these locations reflects association of cohesin with active genes that normally bind little or no cohesin, such as yata in Fig 4A. Active genes bind cohesin, and inactive genes or intergenic regions do not, and this holds true for the extended cohesin domains with Pds5 depletion. Skipping of the extended cohesin domains over inactive genes and intergenic regions is illustrated at the string locus in S4A Fig. These findings indicate that many active genes have the potential to bind cohesin, but normally do not because they are too far from an origin.

Cohesin domain expansion upon Pds5 or Wapl depletion may be linked to changes in chromosome architecture. The ends of the cohesin domain at the Enhancer of split gene complex coincide with the borders of a topologically-associating domain (TAD) [34]. Pds5 or Wapl depletion causes cohesin to extend beyond these borders (S4B Fig). Thus Pds5 or Wapl depletion might alter the locations of the TAD boundaries, or alternatively, reduce the abilities of the boundaries to block cohesin spreading. The Enhancer of split TAD structure does not depend on cohesin as shown by cohesin depletion [34].

The finding that active genes closer to replication origins are more likely to bind cohesin led us to theorize that DNA replication helps determine which genes bind cohesin. DNA replisomes push topologically-bound cohesin in single molecule studies in Xenopus extracts [35] suggesting that cohesin loaded at origins may be pushed outward by replication forks, and data in yeast suggest that cohesin loaded at replication sites travels with replication forks [30]. We reasoned that if DNA replication pushes cohesin, then the cohesin domain expansion caused by Pds5 or Wapl depletion could reflect either a reduced rate of cohesin removal in front of replication forks, or increased speed of replication fork movement. DNA fiber assays show that Pds5 depletion does not alter fork speed (S5 Fig) leading us to favor the idea that cohesin domain expansion is caused by a reduced rate of cohesin removal in front of replication forks.

Pds5 and Brca2 oppositely regulate SA cohesin subunit levels at DNA replication origins

Metaorigin analysis unexpectedly revealed that the SA to Rad21 cohesin subunit ratio is elevated at origin centers compared to flanking regions. The ratio of the mean SA enrichment to mean Rad21 enrichment was calculated for each 10 kb bin for each early origin, and the average ratio for each bin across all origins is plotted in Fig 4B. This is equivalent to normalizing enrichment for a histone modification to total histone occupancy, or a Pol II modification to total Pol II [19]. The high SA to Rad21 enrichment ratio indicates that the SA to Rad21 stoichiometry is higher near origins. The enrichment ratio does not measure actual stoichiometry, or the exact extent to which stoichiometry differs, but indicates the direction in which it varies or changes. It is possible that some or all of the increased SA at origins binds independently of the cohesin complex.

Strikingly, Pds5 depletion reduces SA levels at origin centers (Fig 4C) and equalizes the SA to Rad21 ratio across the metaorigin and flanking regions (Fig 4B). Equalization of the SA to Rad21 ratio across the entire 200 kb metaorigin region upon Pds5 depletion indicates that at origins, Pds5 actively increases SA levels above a default SA to cohesin ratio. In contrast, Wapl depletion does not alter SA levels at origins (S3A Fig) and Brca2 depletion substantially increases SA occupancy in the metaorigin analysis (Fig 4D) and as illustrated specifically at the kayak locus (S2C Fig). When Pds5 and Brca2 are co-depleted, their opposite effects cancel out, leaving little net change in SA levels at origin centers (Fig 4E). This makes the expansion of the SA domains from origins more evident (Fig 4E) giving a pattern similar to that observed with Rad21 (Fig 3B) with statistically significant increases over several kilobases in the flanking regions. Combined, these findings indicate that Pds5 actively increases SA levels at origins, and that Brca2 counters this Pds5 activity.

The opposing roles of Pds5 and Brca2 in controlling SA occupancy at origins parallels their opposing effects on sister chromatid cohesion. The tripartite Smc1-Smc3-Rad21 ring topologically binds chromosomes in the absence of SA or Pds5 [36] both of which are required for sister chromatid cohesion. We thus speculate that Pds5 and Brca2 regulate the SA to cohesin ratio at origins to determine the fraction of cohesin complexes that becomes cohesive behind replication forks. The opposing effect of effects of Pds5 and Brca2 on SA levels occurs at origins, where their occupancy is greatest. We therefore prefer the idea the effects on SA explains their opposing effects on sister chromatid cohesion over the alternative that Brca2 sequesters Pds5 from cohesin to reduce cohesion. Cohesin subunits co-purify with Brca2 along with Pds5 indicating that interaction of Pds5 with cohesin and Brca2 is not mutually exclusive [11].

We also note that the role of Pds5 in increasing SA at replication origins differs from its global effect of reducing total cohesin binding revealed by in vivo FRAP experiments [9] and Rad21 metaorigin analysis above (Fig 3). We theorize that proteins at replication origins modify Pds5 activity, or that this Pds5 activity is limited to when origins are licensed or when they fire.

Pds5 depletion increases Nipped-B levels at origins and extends Nipped-B and Wapl binding domains

Other findings support the idea that Pds5 activity at origins differs from other locations. We considered the possibility that some of the differential effects of Pds5, Wapl, and Brca2 depletion on cohesin levels and distribution around origins could reflect how these factors influence each other’s association with chromosomes, or binding of the Nipped-B cohesin loader. As detailed below, we find that the cohesin accessory factors influence each other’s association with chromosomes, but the changes do not result in the expected effects on cohesin levels at origins.

The effect of Pds5 depletion on Wapl chromosome binding parallels the effects it has on SA. The Wapl to Rad21 ratio, like the SA to Rad21 ratio, is higher at origin centers than in flanking regions, and Pds5 depletion equalizes this ratio across the origin and flanking regions (Fig 5A). Similar to the effect on SA, Pds5 depletion decreases Wapl at origins with moderate increases in flanking regions (Fig 5B). This shows that Pds5 facilitates Wapl binding at origins and confirms that it is not essential for Wapl binding as discussed above. In contrast, Pds5 depletion has only minor effects on Brca2 levels at origins and flanking regions (Fig 5C). Thus, Pds5 is also not required for Brca2 to bind to chromosomes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Pds5 influences Wapl and Nipped-B binding at DNA replication origins with little effect on Brca2.

(A) Metaorigin plot of Wapl to Rad21 ratio in control cells (Mock, blue) and cells depleted for Pds5 (iPds5, red). (B) Left panel is meta-origin analysis of Wapl ChIP-seq enrichment in control cells (Wapl, blue) and cells depleted for Pds5 (Wapl iPds5, red). Right panel is the plot of–log10 p values for the differences in Wapl enrichment in the meta-origin bins calculated using the Wilcoxon signed rank test. (C) Same as B for Brca2 enrichment. (D) Same as A for the Nipped-B to Rad21 ratio. (E) Same as B for Nipped-B ChIP-seq enrichment.

https://doi.org/10.1371/journal.pgen.1007225.g005

The Nipped-B to Rad21 ratio is constant across the 200 kb metaorigin region. The ratio increases at the origin upon Pds5 depletion, accompanied by an overall increase in Nipped-B levels across the metaorigin (Fig 5D and 5E). Pds5 and Nipped-B have very similar HEAT repeat structures [13, 37–41] and compete for binding to cohesin as shown by in vivo FRAP studies [9] and biochemical experiments with C thermophilium orthologs [38]. We thus posit that Pds5 depletion increases Nipped-B occupancy by reducing the competition for binding to cohesin. Wapl depletion increases Pds5 (S3B Fig) and decreases Nipped-B (S3C Fig). We posit that Wapl may facilitate Nipped-B binding and that loss of Nipped-B upon Wapl depletion permits increased Pds5 occupancy.

Importantly, the increase in Pds5 upon Wapl depletion shows that Wapl does not recruit Pds5 to chromosomes. Instead, Brca2 is more crucial, as Brca2 depletion significantly decreases Pds5 occupancy (S3D Fig). The decrease in Pds5 is significant extending for more than 40 kb in either direction from the origin. This finding argues against the idea that Brca2 depletion suppresses the cohesion defects caused by Pds5 depletion by allowing residual Pds5 to re-localize to cohesin. If this were the case, then we would expect to see increases in Pds5 in the regions flanking origins with Brca2 depletion. Nipped-B and Rad21 depletion also decrease Pds5 binding (S3F and S3G Fig) indicating that interaction of Pds5 with cohesin is also important for Pds5 binding to chromosomes, consistent with prior studies [5, 6, 42, 43].

The key finding that emerges from the above experiments is that, in contrast to the effects on cohesin occupancy in origin-flanking regions, the effects of Pds5 and Wapl depletion on cohesin levels at origins are not explained by their roles in removing cohesin from chromosomes, or their effects on the Nipped-B loading factor.

The ratios of accessory factors to cohesin vary between different types of gene regulatory sequences

The above analysis shows that Pds5 and Wapl globally determine which active genes and regulatory sequences bind cohesin by limiting how far cohesin domains extend outward from replication origins, and that Pds5 and Brca2 control the ratio of SA to cohesin at origins. The metaorigin big picture view, however, does not indicate if association of cohesin and its regulatory factors differ between different types of gene regulatory sequences in the cohesin-binding domains, or how depletion of these factors might influence cohesin binding to different types of gene regulatory elements. Nipped-B and cohesin preferentially bind active promoters with high levels of promoter-proximal pausing, and essentially all transcriptional enhancers and Polycomb Response Elements (PREs) [14, 17, 19, 32–34, 44, 45]. Although the metaorigin analysis indicates that proximity to a replication origin determines the likelihood that an active gene or regulatory sequence will bind cohesin, variations in the levels of cohesin and its accessory factors at different types of regulatory sequences could potentially impact interphase cohesin dynamics and gene expression.

We find that the relative levels of the different cohesin regulatory factors vary substantially between active promoters, enhancers and PREs, indicating that interphase cohesin binding dynamics might differ between these types of regulatory sequences. The mean ChIP-seq enrichment for these proteins at each of the active promoters, extragenic enhancers and PREs in BG3 cells were calculated as previously described [17] and the distributions of the mean enrichments across all active promoters, enhancers and PREs are presented as violin plots in Fig 6. Active promoters, enhancers and PREs were defined as described in Fig 6, and enhancers located within transcribed regions were excluded to minimize the influence of transcription through a sequence on protein binding.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. The relative levels of cohesin, Nipped-B, Pds5, Wapl and Brca2 vary between active promoters, enhancers and Polycomb Response Elements (PREs).

The top diagram summarizes how active promoters (blue) extragenic enhancers (yellow) and PREs (orange) are defined as 500 bp sequences as described elsewhere [17, 19, 33]. There are 7,389 non-heterochromatic active promoters, 523 extragenic enhancers and 195 PREs. There are over 2,500 total active enhancers in BG3 cells but intragenic enhancers are excluded to avoid effects caused by changes in transcription. (A) Violin plots of the distribution of Rad21 ChIP-seq enrichment values (mean enrichment in each 500 bp element) for promoters (PRO, blue), extragenic enhancers (ENH, yellow) and PREs (PRE, orange), and 6,892 random 500 bp sequences as a negative control. White dots show the median values. (B) Same as A for SA. (C) Same as A for Nipped-B. (D) Same as A for Pds5. (E) Same as A for Wapl. (F) Same as A for Brca2. (G) Distribution of SA to Rad21 ChIP-seq enrichment ratios for promoters (PRO) enhancers (ENH) and PREs (PRE). (H) Same as G for the Wapl to SA ratio. (I) Same as G for the Pds5 to Wapl ratio.

https://doi.org/10.1371/journal.pgen.1007225.g006

The Rad21 and SA cohesin subunits show the highest levels at enhancers, followed by PREs, and promoters (Fig 6A and 6B). In contrast, Nipped-B has the highest levels at PREs, followed by enhancers and promoters (Fig 6C). Pds5 is highest at enhancers, followed by promoters, and PREs (Fig 6D) while Wapl shows the same pattern as cohesin, being higher at PREs than promoters (Fig 6E). Unexpectedly, the pattern of relative Brca2 enrichment is more similar to Nipped-B than it is to Pds5, being highest at PREs, followed by enhancers and promoters (Fig 6F). Thus, although Pds5 and Brca2 binding overlap as discussed above, there is a substantial difference in their enrichment at PREs. There are few PREs relative to promoters and enhancers, and it is possible that proteins at PREs recruit Brca2 that is not bound to Pds5. It is important to note that PREs typically do not reside near origins, which occur primarily in regions with active genes.

The variation in the relative amounts of cohesin and accessory factors at different types of regulatory sequences imply that proteins that regulate transcription, and/or differences in chromatin structure at the regulatory sequences differentially influence their ability to bind cohesin and accessory factors. Each type of regulatory sequence also shows a relatively broad distribution of enrichments, which likely reflects differences in the proteins that bind to them in addition to their distance from an origin.

The effects of depleting cohesin accessory factors on cohesin levels and composition supports the idea that cohesin dynamics vary between different types of regulatory sequences. The Pds5 to Wapl ratio is substantially higher at promoters than enhancers or PREs (Fig 6I). Based on in vivo FRAP data showing that a high Pds5 to Wapl ratio destabilizes cohesin binding [9] the high Pds5 to Wapl ratio at promoters predicts that cohesin removal is more efficient at promoters than at enhancers. Supporting this idea, Pds5 depletion substantially increases Rad21 at promoters, with more moderate increases at PREs and very modest increases at enhancers (Fig 7A). All increases are statistically significant (S1 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Pds5, Brca2 and Wapl differentially influence the levels of cohesin subunits and Nipped-B at gene regulatory sequences.

(A) The violin plots show the distributions of Rad21 ChIP-seq enrichment values at promoters (PRO, blue) enhancers (ENH, yellow) and PREs (PRE, orange) in mock-treated control BG3 cells (mock) and cells depleted for Pds5 (iPds5). Red lines indicate the median values for each type of regulatory sequence in mock control cells. (B) Same as A for SA enrichment. (C) Distributions of the SA to Rad21 ratios at promoters (PRO) enhancers (ENH) and PREs (PRE) in control (mock) and Pds5-depleted (iPds5) BG3 cells. The blue line indicates the median ratio at promoters in mock control cells and the red line indicates the median ratio at promoters in Pds5-depleted cells. (D) Same as C for the Wapl to SA ratio. (E) Same as A for SA in control and Brca2-depleted (iBrca2) cells. (F) Same as A for SA in control and Wapl-depleted (iWapl) cells. (G) Same as A for Nipped-B enrichment. (H) Same as A for Nipped-B in mock control and Wapl-depleted (iWapl) cells. Statistical tests of the differences in the distributions of ChIP-seq enrichment after protein depletions in panels A, B, E, F, G and H are provided in S1 Table.

https://doi.org/10.1371/journal.pgen.1007225.g007

Although Pds5 depletion increases Rad21 levels, it also significantly lowers SA at all regulatory sequences (Fig 7B, S1 Table). The combined effect of increasing Rad21 and lowering SA equalizes the SA to Rad21 ratio at all regulatory sequences (Fig 7C). Pds5 depletion also slightly increases the Wapl to SA ratio at all regulatory sequences (Fig 7D). Brca2 depletion significantly increases SA levels at all regulatory sequences as expected from the metaorigin analysis above (Fig 7E, S1 Table). Wapl depletion slightly increases SA at promoters and PREs, with a minor statistically significant decrease at enhancers (Fig 7F, S1 Table).

In addition to the multiple effects on cohesin, Pds5 and Wapl also influence binding of the Nipped-B cohesin loader at gene regulatory sequences. Pds5 depletion significantly increases Nipped-B at all regulatory sequences (Fig 7G, S1 Table) supporting the idea that Pds5 and Nipped-B compete for binding to cohesin. Wapl depletion substantially reduces Nipped-B at enhancers and PREs, but unexpectedly increases Nipped-B at promoters (Fig 7H, S1 Table). Wapl is low at promoters, so this suggests that regulatory sequences with high Wapl levels might sequester Nipped-B, and that proteins at promoters, potentially cohesin, or the TBPH RNA-binding protein [17] recruit Nipped-B released from enhancers and PREs by Wapl depletion. The idea that Wapl sequesters Nipped-B is consistent with the metaorigin analysis described above indicating that Wapl facilitates Nipped-B binding.

Pds5, Brca2 and Wapl have similar effects on gene expression as Nipped-B and cohesin

The finding that the levels of the different cohesin accessory factors and their ratios relative to cohesin vary between different types of regulatory sequences led us to hypothesize that their effects on gene expression would differ. Surprisingly, however, as described below, RNA-seq experiments reveal that despite differences in magnitude, the effects of Pds5, Brca2, and Wapl on gene expression significantly overlap those of Nipped-B and cohesin, largely affecting the same genes in the same directions. Our laboratory reported expression microarray and precision run-on sequencing (PRO-seq) experiments showing that Nipped-B and cohesin have very similar genome-wide effects on RNA accumulation and transcription in BG3 cells [14, 33]. For the studies here we performed RNA-seq in BG3 cells depleted for Nipped-B, Rad21, Pds5, Wapl, and Brca2, and compared their genome-wide effects on RNA levels (Fig 8). Three biological replicates were used for each depletion, and were compared to six mock control replicates. The Pearson correlation coefficients between replicates for genome-wide expression levels were all >0.95. The expression data are provided in S1 Data.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Pds5 and Brca2 have similar effects on gene expression in BG3 cells that overlap the effects of Nipped-B and cohesin.

(A) Genome-wide Pearson correlation coefficients for the log2 fold-changes in RNA levels caused by depletion of Nipped-B (iNipped-B), Rad21 (iRad21), Pds5 (iPds5), Wapl (iWapl), Brca2 (iBrca2), Brca2 and Pds5 (iBrca2 iPds5) and Brca2 and Wapl (iBrca2 iWapl). Gene expression values used for the analysis are in S1 Data. (B) Dot plot of log2 fold-changes in RNA levels caused by Nipped-B depletion versus the changes caused by Rad21 depletion. Red dots show statistically significant changes in gene expression caused by Nipped-B depletion (q ≤ 0.05). (C) Dot plot of log2 fold-changes in RNA levels caused by Pds5 depletion versus changes caused by Brca2 depletion. (D) Dot plot of log2 fold-changes in RNA levels caused by Nipped-B depletion versus the changes caused by Pds5 depletion. Red dots show genes significantly altered by Nipped-B depletion (q ≤ 0.05). (E) Overlap in the genes that increase and decrease in expression with the indicated depletions at p ≤ 0.05. P values were used instead of the more stringent q values to obtain larger groups of genes. Numbers in red indicate genes that increase in expression and numbers in blue are genes that decrease. The numbers in the overlap boxes show the number that change with both depletion treatments. Red indicates genes that increase with both and blue indicate genes that decrease with both. Brown indicates genes that increase with one treatment, and decrease with the other. All overlaps in genes that increase or decrease in expression are statistically significant by Fisher’s exact test (S1 Data).

https://doi.org/10.1371/journal.pgen.1007225.g008

The genome-wide Pearson correlation between the fold-change changes in RNA levels caused by Nipped-B and Rad21 depletion is 0.63 (Fig 8A and 8B) consistent with prior studies showing that Nipped-B and cohesin have very similar effects on gene expression and transcription [14, 33]. The correlation between Pds5 and Brca2 depletion is 0.61, similar to the correlation between Nipped-B and Rad21 (Fig 8A and 8C). In contrast, the Pearson correlations between the effects of Nipped-B or Rad21 depletion versus Pds5 or Brca2 depletion are lower, ranging from 0.21 to 0.43 (Fig 8A and 8D). The correlations between Wapl depletion with Nipped-B, Rad21, Pds5 or Brca2 depletion range from 0.24 to 0.38 (Fig 8A).

The high correlation between the effects of Pds5 and Brca2 depletion on RNA levels was unexpected because Pds5 depletion reduces sister chromatid cohesion and Brca2 depletion does not. We thus conclude that most effects of Pds5 depletion on gene expression are not caused by sister chromatid cohesion defects. Because Pds5 depletion reduces cell division and Brca2 depletion does not, we also conclude that most changes in gene expression are not caused by changes in the cell cycle. It is possible, however, that some differences between the effects of Pds5 and Brca2 depletion reflect changes in sister chromatid cohesion or cell cycle effects.

Although the correlations between the effects of Nipped-B or Rad21 depletion with those of Pds5, Brca2 and Wapl depletion are relatively low, close examination indicates that the same genes are affected. Most genes affected by Nipped-B depletion are altered in the same direction by Pds5 depletion as illustrated by the dot plot in Fig 8D, although the magnitudes of the changes tend to be smaller with Pds5 depletion. Agreeing with prior studies [14] gene ontology analysis shows that genes involved in neurogenesis and imaginal disc development are increased by all depletions, and that genes involved in protein translation are reduced (S1 Data). Moreover, there is significant overlap in the genes that increase in expression (p ≤ 0.05) with Nipped-B or Rad21 depletion and the genes that increase with Pds5, Wapl, or Brca2 depletion, as shown in Fig 8E. All overlaps in genes that increase with depletion of any of the cohesion factors are significant by Fisher’s exact test (S1 Data) and few genes show changes in the opposite directions. There is also statistically significant overlap in the genes that decrease with Nipped-B or Rad21 depletion and the genes that decrease with depletion of Pds5, Wapl, or Brca2, with only a few opposite effects, primarily with Wapl depletion (Fig 8E).

Although the effects of individually depleting Pds5 and Brca2 correlate modestly with Wapl depletion (0.38, 0.28) co-depletion of Pds5 and Brca2 strongly correlates with Wapl depletion (0.73) and Wapl-Brca2 co-depletion (0.68) (Fig 8A). The Pds5-Brca2 and Wapl-Brca2 double depletions show more modest correlations with Nipped-B or Rad21 depletion (0.12–0.33) or with depletion of Pds5 or Brca2 alone (0.20–0.47) than with Wapl depletion. There is still significant overlap in the genes that increase in expression with Nipped-B or Rad21 depletion, but more genes that decrease in expression with Nipped-B or Rad21 depletion show increased expression in the double depletions (Fig 8E). Because Wapl depletion has virtually the same effects as Pds5-Brca2 or Wapl-Brca2 co-depletion, we conclude that loss of Wapl is epistatic to loss of Pds5 or Brca2, suggesting that much of the influence of Pds5 and Brca2 on gene expression requires Wapl. Wapl depletion more frequently has an opposite effect on those genes that decrease in expression upon Nipped-B or cohesin depletion, suggesting that Wapl counteracts activation by Nipped-B and cohesin at some genes.

Brca2 influences gene expression in developing wings and facilitates growth

The finding that Brca2 influences gene expression in BG3 cells raised the question of whether or not it also influences gene expression during in vivo development. We conducted RNA-seq in 3^rd instar wing imaginal discs from different control genetic backgrounds, and two different brca2 null mutants, which revealed that many genes increase and decrease in expression. Using a statistical threshold of q ≤ 0.05 (5% false discovery rate) 208 genes increase in expression and 606 decrease, with most changes less than 2-fold (Fig 9A, S1 Data). Gene ontology analysis indicates that the decreasing genes are involved in morphogenesis and development (S1 Data). In addition to many genes that regulate development, the genes that decrease in expression include broadly-acting transcription factors such as Mediator subunits and Pol II kinases, cell cycle control genes, and genes encoding DNA repair proteins (Fig 9B). Unlike pds5 or wapl null mutant flies [7, 8] brca2 null mutant flies are viable, although females are sterile because of defective meiosis [46].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Brca2 influences gene expression in developing wings.

(A) The log2 fold-change in gene expression in brca2 null mutant 3^rd instar wing imaginal discs is plotted versus the log2 expression level in control wing discs for all active genes. Active genes are defined as those that are expressed at or above the median level in control discs plus those expressed at or above the control median level in brca2 mutant discs. Red dots indicate statistically significant changes in gene expression (q ≤ 0.05). The dot representing the brca2 gene is labeled. The blue line indicates no change, and the two red lines indicate 2-fold increases or decreases. (B) Examples of genes down-regulated in brca2 mutant discs in the indicated categories. This is not a comprehensive list, which can be generated from the expression data provided in S1 Data.

https://doi.org/10.1371/journal.pgen.1007225.g009

There are no overt structural mutant phenotypes in brca2 null adult flies, but it seems likely that the many modest changes in gene expression could cause subtle growth or developmental deficits that might be revealed by close examination, or in different genetic backgrounds. We measured the sizes of adult male and female brca2 mutant wings, to look for possible changes in growth. Male brca2 mutant wings are some 9% smaller than controls, and female wings are 4% smaller (S6 Fig). These reductions are similar in magnitude to the dominant effects of null mutations in the fly myc (diminutive, dm) and Tor genes encoding critical dosage-sensitive growth regulators [47]. They are also similar to the effects of heterozygous Nipped-B mutations [47]. Thus, Brca2 facilitates wing growth, and we hypothesize that this reflects the role of Brca2 in gene expression.

Cohesin localization

The finding that the maximal levels of cohesin and cohesin regulatory factors center at replication origins and decrease extending outward for many kilobases strongly suggests that DNA replication plays a role in positioning cohesin. This hypothesis is consistent with single molecule studies in Xenopus extracts showing that DNA replication causes cohesin to translocate unidirectionally [35]. The single-molecule studies also show that cohesin translocation is suppressed by Pds5 and Wapl [35]. This agrees with our finding that depletion of Pds5 or Wapl increases cohesin levels for tens of kilobases surrounding replication origins. We thus hypothesize that pushing of cohesin by replication forks is a key determinant of cohesin localization and which genes bind cohesin, and that reduction of Pds5 or Wapl slows the rate of cohesin removal in front of replication forks, leading to increased cohesin domain size. Recent experiments in yeast shows that cohesin loaded in G1 prior to replication can remain associated with chromosomes during S phase [48] and thus is available to be pushed.

In Xenopus oocyte extracts, association of the Nipped-B orthologs with chromatin and cohesin loading require replication origin licensing and formation of the complete pre-replication complex (pre-RC) containing the origin recognition complex (ORC) the minichromosome maintenance (MCM) helicase complex, Cdc6 (cell division cycle 6) and Cdt1 (cyclin-dependent transcript 1) [25–28]. This supports the idea that replication origins are major loading sites for cohesin. Cohesin is also enriched at origins in HeLa cells [29, 31]. However, studies in Drosophila cultured cells depleted for a pre-RC component indicate that origin licensing is not essential for stable association of cohesin with chromatin, although these studies did not examine cohesin localization [24]. Xenopus oocytes do not have active gene transcription, and we think it likely that a combination of both licensed replication origins and active genes determine cohesin binding in Drosophila cells.

Taken together, our findings lead us to envision that DNA replication exerts global control over the extent of cohesin domains, and that within these domains, proteins binding to active genes and regulatory sequences define the detailed cohesin localization and dynamics. Once an active gene or regulatory sequence binds cohesin and accessory factors, cohesin is continuously loaded and removed from these sites during interphase [9]. In this scenario, the cohesin dynamics at different sites is fine-tuned by the ratios of Nipped-B, Pds5, Wapl and Brca2, which as shown here, vary between promoters, enhancers and PREs. A key question this idea raises is how cohesin is positioned on chromosomes in early G1 after mitosis and before DNA replication. In addition to loading at licensed origins, we speculate that book-marking factors remain bound to active cohesin-binding genes through mitosis and facilitate cohesin re-loading in early G1. The cohesin and accessory protein levels for each gene and regulatory sequence are then refreshed during DNA replication as necessitated by duplication of the genome.

Studies in yeast implicate DNA replication in temporarily positioning cohesin [32] and cohesin is enriched at origins in HeLa cells [29, 31] but a potential role for DNA replication in cohesin localization in mammalian cells remains to be investigated. Recent studies show that transcription, the CTCF architectural protein, and Wapl position cohesin in mammalian cells [49, 50]. Absence of CTCF causes cohesin to localize to transcription start sites [49] similar to the pattern observed in Drosophila. Unlike mammalian CTCF, Drosophila CTCF does not interact directly with the SA cohesin subunit [51] and cohesin does not significantly co-localize with CTCF [34]. Thus the pattern observed in CTCF-deficient mammalian cells is similar to the normal pattern in Drosophila.

In mammalian cells that lack both CTCF and Wapl, cohesin accumulates in large islands several kilobases in size at the 3’ ends of genes [49] suggesting that transcription pushes cohesin to the ends of genes, as seen in yeast [52–54]. Although we detect modest increases in cohesin in some intergenic regions upon Pds5 or Wapl depletion in Drosophila cells, we do not see new substantial intergenic domains, suggesting that there may be additional factors in mammalian cells that trap cohesin at the 3’ ends of genes. Although it remains to be investigated, we envision that DNA replication also positions cohesin in mammalian cells, and that CTCF and transcription refine cohesin positioning, similar to our finding that the cohesin extending out from Drosophila replication origins skips over inactive genes to associate with active genes and regulatory sequences.

Sister chromatid cohesion

It is poorly understood why Pds5, which removes cohesin from chromosomes, is also required for sister chromatid cohesion [55]. A recent study in budding yeast intimately links the role of Pds5 in cohesion to DNA replication by showing that mutations affecting the Elg1 protein that unloads the PCNA replication clamp, or alternatively overexpression of PCNA, suppress cohesion defects of pds5 mutants [56].

Our findings lead us to speculate that Pds5 activity is modified at replication origins, where it actively increases the levels of the SA subunit, and that the high level of SA at origins is important for establishing sister cohesion during replication. SA is required for sister cohesion although it is not part of the ring that encircles DNA, and cohesin topologically binds chromosomes without SA [36]. It is proposed that SA can link two cohesin rings to mediate a handcuff mechanism for sister chromatid cohesion [57, 58] or alternatively that SA may act as a chaperone that facilitates topological binding of one cohesin ring around both sisters at replication forks [36]. These reports, together with the correlation between the loss of high SA levels at replication origins and loss of sister chromatid cohesion upon Pds5 depletion, lead us to propose that maintaining high levels of SA at origins is the critical Pds5 function for establishing sister chromatid cohesion. This idea is further supported by the finding that Brca2 depletion suppresses both the loss of cohesion and the loss of SA at origins caused by Pds5 depletion. Wapl depletion does not alter SA levels at origins, and does not reduce sister chromatid cohesion. An origin-specific role for Pds5 in facilitating SA binding can thus resolve the paradox that Pds5 is required for both sister chromatid cohesion and cohesin removal.

We do not know how Pds5 facilitates SA association with origins or how Brca2 counteracts this activity. We can envision, however, multiple possible direct mechanisms. SA, like Pds5, Wapl and Nipped-B contains several HEAT armadillo-like repeats, and has a structure similar to these cohesin regulators [13, 23, 37–41, 59, 60]. We thus speculate that when brought into proximity of each other on chromosomes, Brca2 might also interact with SA, or even Nipped-B. Although Wapl did not co-purify with Brca2 from Drosophila extracts, significant sub-stoichiometric amounts of SA and Nipped-B and other cohesin subunits were detected [11].

Pds5 and Nipped-B proteins form similar hook-like structures and interact with multiple cohesin subunits. Brca2 interacts with a region in the “handle” of Pds5 adjacent to the N terminal region that interacts with Wapl and facilitates Pds5 dimerization [11, 13]. Thus one idea is that Pds5 aids SA binding through interactions with Rad21 and SA, but Pds5 dimers formed by Brca2 do not. Brca2 depletion would reduce Pds5 dimerization and thereby increase SA binding. The caveat to this argument, however, is that Brca2 depletion also reduces Pds5 binding, and thus should also reduce SA binding. An alternative idea is that Brca2 might interact with SA on chromosomes and reduce its affinity for cohesin. It is also possible that the pre-RC at origins facilitates SA binding, and that Brca2 inhibits this binding via interaction with the pre-RC or SA. In this case, Pds5 could facilitate SA loading by interacting with Brca2 and preventing it from interacting with SA. Testing these ideas may be possible using Xenopus extracts that couple origin licensing with cohesin loading [25–28] and by examining the direct interactions of SA, Pds5 and Brca2 with each other and with replication factors.

As described below, although Pds5 and Brca2 have opposing effects on sister chromatid cohesion and SA levels at replication origins, they have very similar effects on gene expression. This apparent paradox can be explained if Pds5 activity is altered at origins or in early G1 during origin licensing. In other words, we suspect that the roles of Pds5 in sister chromatid cohesion and gene expression are separated by chromosomal location and/or phase of the cell cycle.

The Pds5-Brca2 complex is important for mitotic DNA repair and homologous recombination during meiosis, corresponding with the known role of Brca2 in DNA repair [10–12]. The finding that Brca2 can oppose Pds5 function in sister chromatid cohesion shows that Brca2 plays an additional role in regulating genome stability. The anti-cohesion role of Brca2 is opposite to its role in DNA repair in the sense that reduced cohesion decreases genome stability and DNA repair increases stability. It remains to be seen if hypermorphic brca2 mutations or overexpression of Brca2 can reduce sister chromatid cohesion sufficiently to alter chromosome segregation, but this is potentially relevant to the increased cancer susceptibility in some individuals with BRCA2 missense mutations. It is also of interest that the human BRCA2 gene neighbors PDS5B, with some cancer-associated mutations likely altering both genes [61]. This raises the possibility that some mutations could alter the BRCA2-PDS5B ratio in a way that disfavors sister chromatid cohesion.

Gene regulation

Since the original discovery that Nipped-B mutations alter enhancer-dependent gene expression [62] an attractive concept has been that cohesin holds enhancers and promoters together by a topological mechanism similar to how it holds sister chromatids together. This idea has been expanded in loop extrusion models in which chromatin is threaded through topologically bound cohesin to form intra-chromosomal loops important for topological domain (TAD) formation and gene regulation [50, 63]. These models thus also suggest cohesion-like mechanisms to hold together two regions of the same chromosome.

Not all TADs in Drosophila cells, however, require cohesin for their formation [34, 64–66] and data presented here argue against the intra-chromosomal cohesion model for facilitating enhancer-promoter loops. The Pds5 to Wapl ratio is substantially higher at promoters than at enhancers, which based on in vivo FRAP [9] predicts that cohesin binding is less stable at promoters. Indeed, as shown here, Pds5 depletion substantially increases cohesin levels at promoters but has much less effect at enhancers. The intra-chromosomal cohesion model predicts that the cohesin dynamics will be same at enhancers and promoters, either because they are the same cohesin complexes at both sequences in the embrace cohesion model, or are tightly linked to each other in the handcuff model. A caveat to this argument is that if there are several cohesin rings at these regulatory sequences, as few as one may mediate intra-chromosomal cohesion, while the majority differ in their composition and binding dynamics.

The finding that Pds5 and Brca2 depletion have very similar genome-wide effects on gene expression provides further evidence against the intra-chromosomal cohesion model because Pds5 depletion strongly reduces sister cohesion and Brca2 depletion does not. If sister chromatid cohesion is lost, then intra-chromosomal looping should also be substantially reduced. It is also possible, however, that only a small fraction of the effects of Pds5 depletion on gene expression involve looping deficits, and that changes in cohesin’s other roles in gene regulation are responsible for most effects on gene expression. For instance, cohesin influences transcription of active genes by recruiting the PRC1 Polycomb repressive complex to the promoter region [19, 44]. Nipped-B and cohesin bind to essentially all enhancers, however, and there are many instances where Nipped-B or cohesin depletion reduces transcription of genes activated by known enhancers [33]. We thus currently prefer the alternative idea that interactions between Nipped-B or cohesin with other proteins, such as the Mediator complex [67] facilitate enhancer-promoter looping.

The finding that Pds5 and Brca2 influence expression of the same genes as Nipped-B and cohesin supports the idea that they mediate their effects on gene expression in large part through their effects on cohesin binding and dynamics. It is counterintuitive, however, that their effects are largely in the same direction as Nipped-B or cohesin because Pds5 depletion increases cohesin at promoters and Nipped-B at all regulatory sequences. It is also unexpected that Pds5 and Brca2 depletion, which have opposite effects on SA levels at regulatory sequences, have the same effect on gene expression. These findings thus argue that optimal cohesin dynamics are more important than absolute cohesin levels for gene expression. It must be kept in mind that most genes that bind cohesin do not show large changes in transcription upon cohesin depletion, and that only a subset show substantial and statistically significant changes [14, 33]. Both increases or decreases in cohesin levels or dynamics at these more sensitive genes could have similar effects on their transcription.

Another possibility we considered is that the level of Pds5, and not Nipped-B or cohesin, is more crucial for gene regulation. Nipped-B and Brca2 depletion reduce Pds5 at all regulatory sequences except PREs, and thus have similar effects on Pds5 binding levels as depletion of Pds5. Against this idea, Wapl depletion increases Pds5 levels on gene regulatory elements, and has similar effects on gene expression as Pds5 depletion. We thus currently favor the idea that cohesin binding dynamics is the key factor that drives the majority of the significant effects of cohesin on transcription. This idea can be tested in future experiments through more direct measurements of transcription such as nascent RNA sequencing, and overexpression of cohesin regulators.

Roles of Brca2 in gene expression and development

The finding that Brca2 has significant effects on gene expression in both cultured cells and in vivo has implications for the role of BRCA2 mutations in cancer susceptibility. Many of the genes whose expression is altered in brca2 mutant wing discs are involved in development, cell cycle control and DNA repair, or are broadly active transcriptional regulators. Effects on gene expression caused by BRCA2 mutations could conceivably contribute to cancer etiology even in the absence of DNA damage.

Although Drosophila brca2 null mutants are viable [46] mouse Brca2 null mutations are homozygous lethal during embryogenesis [68] and thus are also likely early recessive lethal in humans. However, some mice with Brca2 hypomorphic mutations survive and are smaller than littermates, with multiple developmental abnormalities [69]. Importantly, biallelic human BRCA2 mutations expressing truncated or otherwise altered BRCA2 proteins cause Fanconi anemia type D1 [70, 71]. This type of Fanconi anemia is associated with highly penetrant and fatal early childhood leukemia, but also multiple congenital abnormalities including slow growth and microcephaly, which are phenotypes that also occur in Cornelia de Lange syndrome, caused by dominant mutations altering cohesin function [72–77]. The sources of the developmental deficits in Brca2 mutant mice and Fanconi anemia D1 are unknown, but our results showing effects on gene expression and growth in developing Drosophila wings raise the likelihood that they reflect multiple changes in gene expression, similar to those caused by heterozygous Nipped-B and NIPBL mutations.

Cell culture, RNAi treatment and metaphase spreads

BG3 cells were cultured and RNAi treatment and metaphase spreads were performed as previously described [14]. At least 50 metaphase nuclei were scored for each treatment group and time point in each experiment.

Pds5 and Brca2 antibodies

Full length Pds5 (residues 1–1218) was expressed as a His6 fusion in E. coli, purified by nickel chromatography, and used to immunize a guinea pig at Pocono Rabbit Farm and Laboratory, Inc. Brca2 residues 1 to 404 of were expressed as a His6 fusion peptide in E. coli, purified and used to immunize a rabbit and a guinea pig at Josman, LLC.

ChIP-seq

ChIP-seq was performed as described before [17, 18]. The Nipped-B, Rad21, SA and Wapl antibodies and their validation were described previously [32, 78]. The Pds5 and Brca2 antibodies were validated by RNAi westerns (S1 Fig). Multiple independent biological replicate experiments were performed for each experimental ChIP-seq group, sequencing to at least 10X genome coverage per replicate, and normalizing to input chromatin sequencing of at least 45X genome coverage. Mean enrichment at active promoters, enhancers and PREs was calculated as described before [17, 19] using custom R [79] scripts. Meta-origin analysis was performed using custom R scripts and the locations of early DNA replication origins in BG3 cells (GEO Accession GSE17281, S1 File). Meta-origin plots were made using Microsoft Excel. Violin plots were made using the wvioplot.R package (https://github.com/cran/wvioplot). The Integrated Genome Browser (IGB) [80] was used to inspect ChIP-seq data and prepare figures. The thresholding function in IGB was used to generate bed files for defining regions of significant enrichment, and the bedtools intersectbed program [81] was used to find overlaps in binding regions. ChIP-seq data was deposited in the NCBI GEO database, accession number GSE109387.

DNA fiber analysis

DNA fibers were spread on microscope slides by adapting a published procedure [82]. Mock or RNAi-depleted BG3 cells were labeled with 20 micromolar IdU (ThermoFisher, cat no. 10035701) in the media for 20 min at 25°C, washed twice with phosphate buffered saline (PBS) and then labeled for 20 min with 200 micromolar CldU in the culture media (Sigma Aldrich, cat no. C6891) at 25°C. Labeled cells were washed twice with PBS, trypsinized, washed twice with PBS and brought to 7,000 cells per microliter in PBS and placed on ice. For experiments with hydroxyurea (HU) treatment, cells were washed twice with PBS after CldU incorporation and incubated at 25°C with 4 mM HU for 4 hours prior to cell collection. Labeled cells were diluted with unlabeled control BG3 cells in a 1:5 or 1:10 ratio before spreading to decrease the number of overlapping labeled fibers [82]. Two microliters of diluted cells were mixed with eight microliters of lysis buffer (200 mM Tris^.HCl pH 7.5, 50 mM EDTA, 0.5% SDS) on top of a microscope slide (VWR, cat no. 48311–703) for eight min. After lysis, the slides were tilted at a 20 to 40^o angle to allow the DNA to spread. Slides were air-dried at room temperature for 1 hour and fixed in 3:1 v/v methanol:acetic acid solution for eight min. Fixed slides were dried and stored at 4°C before immunostaining.

For immunostaining, fixed slides were washed twice for five min with PBS and treated with 2.5 N HCl for one hour. Acid-treated slides were washed three times for five min with PBS, blocked with 5% (w/v) bovine serum albumin (BSA) in PBS for one hour at 37^o and rinsed in PBST (0.05% Tween 20 in PBS) for five to ten sec just prior to adding primary antibodies diluted in PBST containing 1% BSA. The primary antibodies used were 1:20 mouse anti-BrdU B44 (BD Biosciences, cat no. 347580) which recognizes IdU and 1:100 rat anti-BrdU monoclonal antibody BU1/75 (ThermoFisher, cat no. MA1-82088) which recognizes CldU. Forty microliters of the primary antibody mix was added to each slide, covered with a coverslip, and the slides were incubated in a dark humid chamber for 1.5 hours at room temperature. Coverslips were removed in PBS, and the slides washed three times for five min with PBST and kept in PBS until addition of the secondary antibodies. The secondary antibodies were each diluted 1:125 in PBST containing 1% BSA. The secondary antibodies used were Alexa Fluor 546 goat anti-mouse IgG1 (Invitrogen, cat no. A21123) and Alexa Fluor 488 chicken anti-rat IgG (Invitrogen, cat no. A21470). Forty microliters of diluted secondary antibody mix was added to each slide, covered with a coverslip, and the slides were incubated for one hour in a dark humid chamber at room temperature. Coverslips were removed in PBS, and the slides washed three times for five min in PBST and left in PBS before mounting. For mounting, slides were air-dried at room temperature in the dark, and mounted with 20 microliters of ProLong Gold antifade reagent (Invitrogen, cat no. P36930) under a coverslip. Slides were dried at room temperature in the dark and stored at 4^o before imaging. Fluorescent micrographs were digitally captured using a Leica SP5 laser scanning confocal microscope with a 60x objective, and the lengths of connected red and green fibers were measured using NIH ImageJ software. Statistical analysis was conducted using R and violin plots generated using wvioplot.R.

RNA-seq

Total RNA-seq seq with BG3 cells was performed using ribosomal depletion and processed as described previously [19, 47] using three independent biological replicates for each depletion and six previously reported mock control samples [19] (GEO Accession GSE100548). Genome-wide expression values between replicates showed correlations >0.95. Gene expression data in mean normalized nucleotide coverage per gene, and gene ontology analysis is provided in S1 Data. Total RNA-seq with dissected late third instar wing discs from female larvae was conducted as described before using ribosomal depletion [47]. To minimize genetic background effects, five wild-type controls consisted of two Canton S samples and three y w samples from independent stocks. The five brca2 -/- replicates consisted of two isolations of GFP negative larvae from a brca2^KO / CyO, Kr-GFP stock and three isolations from GFP negative larvae from a brca2^56e / CyO, Kr-GFP stock. All replicates gave genome-wide correlation values >0.95. Both brca2 null alleles were the gift of Trudi Schüpbach. The expression data in mean normalized nucleotide coverage per gene, and gene ontology analysis is provided in S1 Data. RNA-seq data was deposited in the NCBI GEO database, accession number GSE109387.

Wing area measurements

Wings were collected from adult flies from crosses conducted at 25°C, mounted, photographed and measured as previously described [47].