Impact of 3-dimensional genome organization, guided by cohesin and CTCF looping, on sex-biased chromatin interactions and gene expression in mouse liver

Background Sex differences in the transcriptome and epigenome are widespread in mouse liver and are associated with sex-bias in liver disease. Several thousand sex-differential distal enhancers have been identified; however, their links to sex-biased genes and the impact of any sex-differences in nuclear organization, DNA looping, and chromatin interactions are unknown. Results To address these issues, we first characterized 1,847 mouse liver genomic regions showing significant sex differential occupancy by cohesin and CTCF, two key 3D nuclear organizing factors. These sex-differential binding sites were largely distal to sex-biased genes, but rarely generated sex-differential TAD (topologically associating domain) or intra-TAD loop anchors. A substantial subset of the sex-biased cohesin-non-CTCF binding sites, but not the sex-biased cohesin-and-CTCF binding sites, overlapped sex-biased enhancers. Cohesin depletion reduced the expression of male-biased genes with distal, but not proximal, sex-biased enhancers by >10-fold, implicating cohesin in long-range enhancer interactions regulating sex-biased genes. Using circularized chromosome conformation capture-based sequencing (4C-seq), we showed that sex differences in distal sex-biased enhancer-promoter interactions are common. Sex-differential chromatin interactions involving sex-biased gene promoters, enhancers, and lncRNAs were associated with sex-biased binding of cohesin and/or CTCF. Furthermore, intra-TAD loops with sex-independent cohesin-and-CTCF anchors conferred sex specificity to chromatin interactions indirectly, by insulating sex-biased enhancer-promoter contacts and by bringing sex-biased genes into closer proximity to sex-biased enhancers. Conclusions These findings elucidate how 3-dimensional genome organization contributes to sex differences in gene expression in a non-reproductive tissue through both direct and indirect effects of cohesin and CTCF looping on distal enhancer interactions with sex-differentially expressed genes.

fraction of male-biased than female-biased Lone CTCF sites contained a CTCF motif (66% vs 48%, 141 Fig. S2C), but there was no significant sex difference in normalized ChIP signal or motif score ( Fig. 2A,   142 Fig. S2C). The latter sex differences may be driven by additional factors, such as the inhibitory effect 143 of DNA methylation on CTCF binding [48,49], where the same sequence motif in male and female 144 liver could be preferentially-bound in males due to the hypermethylation of DNA seen in female 145 compared to male mouse liver [50].

147
Mapping of sex-biased binding sites to TAD and intra-TAD boundaries, genes and regulatory 148 elements. We investigated whether the sex-differential binding shown by cohesin and CTCF is linked 149 to a sex-differential segmentation of the genome at the level of TAD and intra-TAD loops (DNA loop 150 domains). We identified 137 CAC sites with significant sex differences in both CTCF and cohesin 151 binding (Fig. 1B), of which 53 are on autosomes (Table S1C). 17 of the 137 sex-differential CAC sites 152 overlap a TAD or intra-TAD loop anchor [32] in either male or female liver, of which 9 are on 153 autosomes (Table S1C). Ten of the 17 CAC sites are associated with an intra-TAD loop predicted to 154 be present in one sex only, of which 5 such loops contained one of more sex-biased genes. One of the 155 5 sex-based intra-TAD loop domains includes 6 sex-biased genes from the Cyp2c gene family and 2 156 sex-biased lncRNA genes (Fig. S2D). Consistent with the low frequency of sex-biased CAC sites at DHS (median distance of 202 bp (in males) or 119 bp (in females)) than were sex-biased CAC sites 169 (median distance = 8.8 kb and 17.5 kb, for male-biased and female-biased CAC, respectively) (Fig. 170 2B, left). Thus, although sex-biased CNC sites are weaker binding than CAC sites ( Fig. 2A), a majority 171 are found at enhancers and may be functional. Female-biased CTCF binding sites tended to be closer 172 to the transcription start site (TSS) than male-biased sites (Fig. 2C), despite similar ChIP-seq sample 173 quality (Table S3B).

175
All classes of sex-biased cohesin and CTCF binding sites were primarily distal from sex-biased genes: 176 < 20% mapped within 20 kb of a sex-biased gene, and only 35 to 53% were found within the same 177 TAD and could therefore be considered potential cis regulators of sex-biased genes (Fig. 2D).

178
Consistent with the association of cohesin with enhancers [26,51], 25-29% of sex-biased CNC sites 179 overlapped a sex-biased enhancer (either a sex-biased DHS or a sex-biased H3K27ac peak), as 180 compared to only 7-11% of sex-biased CAC and Lone CTCF sites (Fig. 2E). Furthermore, 77% (72 of 181 96) of sex-biased CNCs that overlap a sex-biased enhancer are >20 kb from a TSS of sex-biased

186
Sex-biased liver CNC sites, as well as Lone CTCF sites, showed much more tissue-specific binding of 187 CTCF across mouse tissues [52,53] than did sex-biased liver CAC sites (Fig. S3A, lower vs upper 188 panels). In addition, liver-expressed genes that mapped to sex-biased CNC sites showed a more liver-189 specific expression pattern than genes mapping to sex-biased CAC sites, or liver-expressed genes 190 overall (Fig. S3B). This suggests that sex-biased CNC sites participate in tissue-specific transcriptional 191 regulation, as was described for CNC peaks generally [26,29,30]. Significant differences in the tissue-192 specificities of CTCF binding were also seen at male-biased compared to female-biased CAC sites 193 and Lone CTCF sites (Fig. S3A, Fig. S3C).

194
Impact of cohesin depletion on distally-regulated male-biased genes. As noted above, 35-53% of 196 sex-biased cohesin and CTCF binding sites are within the same TAD as at least one sex-biased gene 197 ( Fig. 2D) and could play a role in DNA looping between sex-biased enhancers and sex-biased gene 198 promoters. Examples of sex-biased CTCF and/or cohesin binding sites that were either proximal (< 20 199 kb) or distal to sex-biased genes are shown in Fig. 3A-3C. Fig. 3A shows a highly female-biased 200 enhancer (female-biased DHS and female-biased H3K27ac marks) with an overlapping female-biased 201 CAC site (green arrows) located 33 kb upstream of the female-biased gene Slc22a29 (F/M expression 202 ratio=8.7), the closest TSS. Fig. 3B shows Cml5, a male-specific gene (M/F expression ratio = 20.2) 203 with a male-biased CNC site that overlaps a male-biased DHS ~3 kb upstream of its TSS (Fig. 3A, red 204 arrow). The adjacent gene, Nat8 (M/F = 4.2), has a male-biased CAC site that overlaps a male-biased 205 DHS located ~12 kb upstream of the TSS (Fig. 3B, green arrow). Conceivably, the sex-biased binding 206 of cohesin and CTCF at these sites could contribute to looping of the associated sex-biased enhancers 207 to their correspondingly sex-biased gene targets. Finally, Fig. 3C shows two male-biased complement 208 C8 genes (C8a (M/F = 3.3) and C8b (M/F = 2.8)) that are linearly quite distant (~1.5 Mb) from a cluster 209 of strongly male-biased enhancers near the 5' end of the same TAD. The TAD structure of this 210 genomic region suggests these enhancers are spatially more proximal to the C8 genes than they are 211 to than the linearly much closer Oma1 gene, located just inside the adjacent TAD (also see Fig. S4A).

213
Loss of cohesin binding in male mouse liver, achieved by depletion of the cohesin loading factor, Nipbl, 214 leads to a loss of distal enhancer-promoter contacts and an increase in local ectopic contacts, which 215 can activate proximal genes [22]. We compared the effects of cohesin loss on the expression of male-216 biased genes with proximal sex-biased enhancers versus those that have only distal (>20 kb) sex-217 biased enhancers. Fig. 3D shows the relative changes in gene expression in cohesin-depleted 218 compared to wild-type male mouse liver for the three genes included in Fig. 3C. Expression of C8a 219 and C8b decreased, by 98% and 82%, respectively, upon loss of chromatin-bound cohesin in male 220 liver, while expression of Oma1 increased modestly (+22%), perhaps by an enhancer hijacking 221 mechanism [54]. In contrast, the two male-biased genes with proximal sex-biased enhancers, Cml5 222 and Nat8 (Fig. 3A), showed no significant change in expression following cohesin loss (Fig. S4B).
Next, we examined a set of 61 male-biased genes and verified the requirement of cohesin for 224 expression of the distally-regulated but not the proximally-regulated male-biased genes: male-biased 225 genes with distal (> 20 kb) sex-biased regulatory elements were significantly more sensitive to loss of 226 cohesin than male-biased genes with proximal sex-biased enhancers (median decrease in expression 227 upon cohesin loss: 14.3-fold vs. 1.4-fold; Fig. 3E). This finding likely results from a requirement for 228 cohesin for distal interactions, via either a direct or an indirect looping mechanism. Conceivably, for 229 sex-biased genes with nearby sex-biased regulatory elements, enhancer-promoter loops required for 230 gene expression can be maintained over short genomic distances by transcription factors such as 231 Mediator [26] or YY1 [55], and without a need for cohesin. mono-exonic nuclear-enriched lncRNAs [12] in three clusters across the genomic region displayed.

242
Robust interactions were observed in female but not male livers between the viewpoint enhancer and 243 three genomic regions (red arrows): a strong female-biased enhancer (right arrow), the promoter of 244 A1bg (middle arrow), and a region downstream of A1bg that contains a cluster of four female-specific 245 lncRNAs (lncRNAs 12590-12593; left arrow), where we observed the strongest interactions. The 246 lncRNAs in this cluster are more highly expressed (Fig. S5E) and are more consistently female-biased 247 across various RNA-seq datasets than the other two lncRNA clusters (Fig. S5F). The maximum 248 expression of these 12 lncRNAs ranged from 0.31 to 2.71 fragment per kilobase per million sequence 249 reads (FPKM) in female liver compared to 0 to 0.02 FPKM in male liver (Fig. S5E).

251
Matthews and Waxman -2/10/20 10 The precise relationship between the female-biased expression of these lncRNAs and the female-bias 252 in 3D interactions with the distal enhancer is not known. The interaction may be regulatory in nature 253 (e.g., an enhancer-promoter interaction, as with any gene) or it could be facilitated by one or more of 254 the 12 nuclear-enriched, female-biased lncRNAs, as was described for other lncRNAs (Xist [56], Firre

255
[57], Haunt [58]). Alternatively, the female-specific interactions shown may be primarily those of 256 regulatory enhancers driving expression of female-specific genes, including A1bg and multiple 257 lncRNAs. The female-biased CTCF binding seen at both interacting regions (right and left arrows) 258 lends mechanistic support to the latter proposal, with CTCF mediating enhancer-promoter and 259 enhancer-enhancer interactions. As CTCF can interact with lncRNAs in a functional manner, and with 260 high affinity [59], these two mechanisms are not mutually exclusive; one or more of these highly 261 female-specific lncRNAs (Fig. S5E) could function in a cis-acting manner to selectively guide CTCF 262 binding and interactions unique to female liver.

264
Sult3a region: We used 4C-seq to interrogate an enhancer viewpoint proximal to the highly female-265 specific gene Gm4794 (F/M = 704; also known as Sult3a2) and also two female-specific mono-exonic 266 lncRNAs, lnc8820 and lnc8821 (F/M = 34 and 90, respectively) ( Fig. 4B, Fig. S5B). The enhancer 267 viewpoint is distal to two other female-specific protein coding genes, Sult3a1 (F/M = 288) and Rsph4a 268 (F/M = 34). We observed female-biased 4C-seq interactions with the proximal promoter of Gm4794 269 (left red arrow) and with strong female-biased enhancers that overlap lnc8820 and lnc8821. Unlike the 270 enhancer nearby A1bg, these interactions are associated with female-biased CNC peaks present at 271 both the viewpoint enhancer and the downstream interacting enhancer. Despite the absence of CAC 272 insulator elements across this genomic region (and consequently, the absence of intra-TAD loops), the 273 observed focal interactions are all local, within ~35 kb of the viewpoint. However, the viewpoint 274 enhancer also made weak interactions to a broad, ~45 kb region extending from ~15 kb upstream of 275 the promoter of Sult3a1 to ~10 kb beyond the gene body (Fig. 4B, red bracket). This 45-kb region 276 contains several robust female-biased DHS and H3K27ac peaks, but lacks cohesin binding, which 277 may account for the lack of strong, focal interactions with the viewpoint enhancer. Gm4794 and 278 Sult3a1 may both interact with the enhancer that overlaps lnc8820 and lnc8821, but this weak (and perhaps indirect) association cannot be captured by proximity ligation under standard 4C-seq 280 conditions. Reciprocal 4C-seq experiments anchored at these lncRNA TSS or alternative 4C methods 281 with increased sensitivity [60] may be needed to validate these weaker interactions. 282 283 C9 region: We examined a 4C-seq promoter viewpoint placed at the complement factor C9 gene (M/F 284 = 3.5) to investigate whether intra-TAD loops can indirectly coordinate enhancer-promoter contacts 285 preferential to one sex in a genomic region devoid of sex-biased CTCF or cohesin binding. The 3' 286 portion of C9 overlaps an antisense lncRNA that shows a much higher male-bias in expression 287 (lnc12340; M/F = 26) (Fig. 4C, Fig. S5C). A cluster of strongly male-biased enhancers lies ~230 kb 288 upstream of the TSS of C9, and is characterized by male-biased DHS and H3K27ac peaks, whereas 289 the TSS of C9 is only comprised of a male-biased DHS. The far upstream enhancer and the TSS of C9 290 both fall just outside of (< 10 kb from) a nested intra-TAD loop (Fig. 4C, bottom track) that 291 encompasses Dab2, whose expression in male liver is 87-fold lower than C9 (FPKM = 0.7 vs 61). The 292 promoter region of C9 interacts with the cluster of far upstream enhancers (Fig. 4C, red arrow), with 293 stronger interactions seen in male liver. Weaker, mostly non-focal interactions were seen between the 294 C9 promoter and sites within the nested intra-TAD loops. This apparent insulation allows the strong 295 male-biased upstream enhancers to bypass the more proximal Dab2 and drive expression of C9. This 296 insulation-by-looping mechanism also enables the 87-fold higher expression of C9 compared to Dab2 297 in male liver. However, a shorter isoform of Dab2 contained within the larger nested intra-TAD loop 298 does show weak, male-specific interactions despite a lack of sex-bias in expression. The movement of 299 cohesin along chromatin has been linked, at least in part, to transcriptional activity [19,61,62]. Thus, 300 higher levels of transcription in male liver could lead to less pausing of cohesin at loop anchors and 301 thereby increase interactions across these loop boundaries in males. Given the interaction with the far 302 upstream enhancer element, we hypothesized that the expression of C9 and the antisense lnc12340 303 would be sensitive to the loss cohesin of binding. Indeed, both genes showed a 3 to 5-fold decrease in 304 expression upon cohesin depletion in mouse liver (p < 0.01 for both), while the insulated gene Dab2 305 showed no significant change in expression (Fig. 4D). 306 308 intra-TAD loop (M/F = 2.5; FPKM = 132 in male liver). Although there are some weak male-biased 309 DHS within the gene body of Nudt7 (Fig. 4E), there is no apparent sex bias at its shared promoter with 310 the sex-biased lncRNA gene lnc7430 (M/F = 4.0; FPKM = 2.9 in male liver). Approximately 22 kb and 311 39 kb upstream of the TSS of Nudt7 are two male-biased enhancers with male-biased DHS and 312 H3K27ac marks; the latter also is proximal (~2.5 kb upstream) to the TSS of the male-biased lnc7423 313 (M/F = 7.0; FPKM = 2.8 in male liver). This enhancer cluster is one of 503 super-enhancers identified 314 in both male and female liver [32]. We observed male-biased 4C-seq interactions between the 315 enhancer viewpoint and a neighboring male-biased enhancer, and also with the shared Nudt7 and 316 lnc7430 promoter region (Fig. 4E, red arrows; Fig. S5D). In contrast, we did not observe focal 317 interactions in either sex between the enhancer viewpoint and a sex-independent enhancer 15.7 kb 318 upstream of Nudt7. Both Nudt7 and lnc7430 were strongly down regulated in male liver upon cohesin 319 depletion (Fig. 4F), suggesting their expression is dependent on interactions facilitated by the CAC-320 anchored intra-TAD loop that encompasses this genomic region. Expression of lnc7423 was not 321 significantly reduced, perhaps due to its closer proximity to strong male-biased enhancers.

323
Sex-independent, nested intra-TAD loops restrict Nox4 to proximal enhancer-promoter 324 interactions. NADPH oxidase 4 (Nox4) exhibits male-biased expression in mouse liver (M/F = 7.7) 325 and may contribute to a number of liver pathologies whose incidence or severity is male-biased [63, 326 64]. Nox4 is highly up regulated in tumor compared to healthy liver tissue of mice that spontaneously 327 develop liver tumors (Fig. S7A), and in humans, Nox4 is up regulated in hepatocellular carcinoma and 328 other cancers (Fig. S7B). Mouse Nox4 is located within a pair of nested intra-TAD loops (Fig. 5A,   329 bottom). Mouse Nox4 also has a strong male-biased enhancer 11.5 kb upstream of the TSS and a 330 strong male-biased DHS 125 kb downstream of the TSS (green vertical highlight). However, only the 331 upstream region has H3K27ac (active enhancer) marks (Fig. 5A). We placed 4C-seq viewpoints at 332 both the upstream region (viewpoint VP1, at -11.5 kb; red highlight in Fig. 5A) and the downstream 333 region (viewpoint VP2, at +125 kb; green highlight) to investigate chromatin interactions with each 334 putative regulatory region. Interactions with the downstream DHS at VP2 were limited to the domain defined by the pair of 3' anchors of the nested intra-TAD loops, consistent with these loops insulating 336 from distal interactions (Fig. 5A, green bracket at bottom). As a result, VP2 did not interact with the 337 promoter of Nox4 or with the -11.5 kb enhancer in either male or female liver. Furthermore, VP2 did 338 not show any consistent male-biased interactions, despite its location at a strong male-biased DHS.

340
In contrast, the -11.5 kb enhancer at VP1 showed male-biased interactions with several genomic

358
These findings support the predicted model of two nested intra-TAD loops, with the shorter enclosed 359 loop insulated from the larger enclosing loop. Domain predictions for other mouse tissues, based on 360 computational methods and experimentally-observed looping in mouse embryonic stem cells [27,52], 361 support the conclusion that the genomic regions defined by VP1 and VP2 are in separate domains 362 (Fig. 5A, bottom). Accordingly, only the -11.5 kb enhancer at VP1 would be predicted to interact with the Nox4 promoter. Generally, the cis regulatory elements relevant for the regulation of Nox4 appear to 364 be contained within the shorter intra-TAD loop. It is less clear what regulatory function the male-biased 365 DHS at VP2 plays, as it does not interact with Nox4 or with the downstream gene, Tyr, which is not 366 expressed in liver (FPKM < 0.01 in both sexes).

369
We investigated sex differences in autosomal 3D genome organization in the mouse liver model,

370
focusing on sex-based differences in chromatin binding and interactions involving cohesin and CTCF, 371 which mediate long-range DNA looping interactions that segment mammalian genomes into 372 megabase-scale TAD domains and their shorter intra-TAD domains. We identified 1,847 binding sites 373 for cohesin and/or CTCF that show significant differential occupancy between male and female mouse 374 liver; however, very few of these sites were associated with sex differences in TAD or intra-TAD loop 375 anchors. A majority of the sex-biased binding sites classified as cohesin-non-CTCF (CNC) sites (but 376 only a minority of cohesin-and-CTCF (CAC) and Lone CTCF sites) mapped to distal enhancers, and a 377 major subset of these overlapped sex-biased distal enhancers (median distance 238 kb to TSS of a 378 sex-biased gene). These findings are consistent with the general role of cohesin in mediating distal 379 enhancer−promoter interactions [26,28], and more specifically, indicate a role for sex-biased cohesin 380 binding in sex-biased enhancer activity. We also found that male-biased genes with distal but not 381 proximal sex-biased enhancers were much more sensitive to cohesin depletion than genes with 382 proximal sex-biased enhancers, implicating cohesin in long-range enhancer interactions regulating 383 these sex-biased genes. Finally, by applying 4C-seq to sex-biased enhancer viewpoints in five 384 genomic regions, we established that sex differences in chromatin interactions are a common feature 385 of sex-biased gene expression in the liver, and we elucidated how chromatin interactions link sex-386 biased genes to distal sex-biased enhancers, guided both directly and indirectly by cohesin and/or 387 CTCF looping.

389
Although TADs and intra-TADs are largely conserved across tissues, 20-30% of all such CAC-390 mediated loops are cell type-specific [32]. Nevertheless, when comparing male and female mouse liver, which show extensive growth hormone-regulated differences in epigenetic state [8,9,14], we 392 did not find evidence for widespread formation of sex-specific intra-TAD loops. Rather, we found that 393 intra-TAD loops in mouse liver are largely sex-independent and devoid of sex-biased CTCF or 394 cohesin binding at their CAC anchors. These loops do have the ability, however, to indirectly facilitate 395 sex-dependent chromatin interactions. Thus, a sex-independent intra-TAD loop was shown to 396 insulate the super-enhancer-associated male-biased gene Nudt7, and nested intra-TAD loops 397 insulating Nox4 restricted the promoter of this male-biased gene from an intronic enhancer while 398 enabling interactions with a cluster of upstream enhancers. Furthermore, our analysis of female-399 biased gene regions revealed female-biased proximal enhancer−promoter interactions in the Sult3a 400 gene region associated with female-biased cohesin binding, as well as female-biased interactions 401 between the A1bg promoter, a far distal (>100 kb) enhancer, and distal female-biased CTCF binding 402 sites. Together, these findings support the proposal that CTCF and cohesin contribute in both direct 403 and indirect ways to the formation of sex-biased enhancer-promoter contacts in mouse liver (Fig. 6).

405
Our analysis of the male-biased complement factor gene C9 provides an interesting example of sex-406 independent CAC-looping that indirectly facilitates sex-biased enhancer-promoter contacts. C9 407 interacts strongly with a distal (>200 kb upstream) male-biased enhancer while bypassing the weakly 408 expressed (and sex-independent) Dab2 gene region, which is insulated by a nested pair of intra-TAD 409 loops. These nested loops, in turn, bring the TSS of C9 into much closer proximity of a cluster of far 410 upstream male-biased enhancers than would be achieved based on linear genomic distance alone 411 (Fig. 6B). Furthermore, we observed more frequent contacts between C9 and the far upstream 412 enhancers in male compared to female liver, despite the absence of any male-biased binding of CTCF 413 or cohesin to help explain sex differences in contact frequency. Conceivably, the male-biased DHS

433
While CAC sites at TAD and intra-TAD boundaries have a well-established role as anchors that enable 434 loop domain-level nuclear organization [18,27,28,32,33], non-anchor CAC sites may directly link 435 enhancers to promoters or to other enhancers, and thereby contribute to interactions governing tissue-436 specific gene expression [36,42]. A majority of sex-differential CAC binding occurs at non-anchor CAC 437 sites (Table S1), a subset of which may mediate long-distance interactions involving sex-biased 438 enhancers and gene promoters. Specific examples described here include the enhancer−enhancer 439 and enhancer−promoter contacts that we identified by 4C-seq for A1bg in the context of female-biased 440 CTCF binding. Similarly, more than half of male-biased liver CTCF binding occurred at Lone CTCF 441 sites, which we found are closer than CAC sites to gene TSS, and can also play a non-canonical role 442 in looping between enhancers and promoters [36].

444
4C-seq interactions between sex-biased enhancer viewpoints and distal sex-biased lncRNAs [11,12] 445 were found in three of the five sex-biased genomic regions we investigated. In one example, the highly female-biased gene A1bg is nearby three clusters of strongly female-biased, nuclear-enriched mono-447 exonic lncRNAs, several of which are transcribed from genomic loci that show female-specific 448 interactions with the distal female-specific enhancer viewpoint that we interrogated. Enhancer-449 associated lncRNAs have been defined as intergenic transcripts with enhancer chromatin marks 450 whose expression is tissue-restricted and is associated with increased expression of nearby expressed 451 protein coding genes [73,74]. Based on our findings, we propose that functional enhancer-associated 452 lncRNAs might be identified by their looping interactions with enhancer sequences, which can be 453 determined globally using high throughput interaction methods, such as Hi-C [75].

455
The sex-biased cohesin and CTCF binding sites described here were discovered using livers from 456 adult (8 week) mice, and likely encompass only a subset of all sex-differential cohesin and CTCF 457 binding sites across the lifespan of a mouse, given the dramatic changes in sex-biased gene 458 expression that occur during prenatal and especially postnatal liver development [76,77]. The sex-459 biased binding of cohesin and CTCF to liver chromatin in adult liver is expected to be regulated by 460 pituitary growth hormone secretion, which is sex-dependent and produces the sex-dependent plasma 461 growth hormone profiles that regulate the vast majority of sex differences in the adult liver, including 462 differences in gene expression [14,78], transcription factor binding [79,80], and chromatin states [8, 9, 463 14]. Given that CTCF binding to DNA can be inhibited by DNA methylation [48,49,81], the 464 hypomethylation of enhancer sequences seen in male compared to female liver [50] could contribute 465 to male-specific CTCF binding at such sites. Such an effect is expected to become more pronounced 466 with age, given the increased male hypomethylation reported in older mice [50,82].

468
The sex-specific patterns of pituitary growth hormone secretion regulating sex-differences in the liver 469 emerge at puberty, and have been implicated in the dynamic regulation of liver chromatin states in 470 both male and female adult mouse liver [9,14]. We do not know when during mouse liver development 471 the sex-differential chromatin interactions described here are first established, whether they constitute 472 a relatively fixed (static) 3D framework governing transcription in male and female nuclei, or 473 alternatively, whether they respond dynamically to the temporal changes in plasma growth hormone profiles that regulate sex differences in liver chromatin states. The potential for dynamic, reversible

496
We employed the mouse liver model with its extensive sex differences in gene expression to study sex 497 differences in nuclear organization and DNA looping interactions in a non-reproductive tissue exposed 498 to sex-unique patterns of hormone stimulation. We determined that male-biased genes with distal but 499 not proximal sex-biased enhancers are particularly sensitive to the loss of cohesin binding.

500
Furthermore, while most sex-biased binding sites for CTCF and cohesin were found to be distal from 501 sex-biased genes, a subset likely contributes to sex-biased looping between regulatory elements in cis, as exemplified by the female-biased DNA looping interactions observed for A1bg. In addition, sex-503 independent CAC-looping may indirectly provide sex specificity to chromatin interactions by insulating 504 male-biased genes such as Nudt7, or by bringing a sex-biased gene into closer proximity to a cluster 505 of sex-biased enhancers, as demonstrated for C9. Together, these findings illustrate the direct and 506 indirect contributions that cohesin and CTCF can make to sex-biased gene expression in the liver, and 507 may be broadly applicable to other biological systems where distal regulation of gene expression is of 508 interest.  Gel Extraction Kit (Qiagen, cat. # 28706) and quantified with a Qubit HS DNA kit (Invitrogen, cat. # 532 Q32854). All samples were processed using the same protocol and conditions. Sequencing was 533 performed for a total of eight CTCF ChIP-seq samples (n=4 individual male and n=4 individual female 534 livers) and a total of six Rad21 ChIP-seq samples (n=3 male, n=3 female livers). The male liver ChIP-535 seq samples were those reported previously [32] and are available at GSE102997.

537
Impact of cohesin depletion sex-biased gene expression. RNA-seq data for wild-type and cohesin 538 depleted (Nipbl-deficient) mouse liver (GSE93431) [22]. was analyzed for n=4 wild-type (control) and 539 n=4 Nipbl-depleted male mouse liver replicates. Data was FPKM-normalized, and reads were 540 expressed relative to the mean of the wild-type group, which was set = 1 on a per gene basis by 541 dividing the expression value for each individual replicate by the mean of the wild-type group plus a 542 small pseudo count (1e-6) to avoid dividing by zero. Data is presented as mean relative expression ± 543 SD for all plots. To determine the global effects of cohesin depletion on male-biased genes, the set of 544 all expressed, strongly male-biased genes (FPKM > 1, and male/female (M/F) expression ratio > 3) 545 was divided into two groups based on distance from their TSS to the nearest male-biased DHS or 546 male-biased H3K27ac-marked region. Male-biased genes with proximal sex-biased enhancers (n=29) 547 were defined as having their TSS < 20 kb from the nearest male-biased DHS and from the nearest 548 male-biased H3K27ac peak; and male-biased genes with distal sex-biased enhancers (n=32) were 549 those with TSS > 20 kb from both such regions. The underlying expression values for all genes are 550 provided in Table S2A and Table S2B.   (Table S3A). Candidate viewpoints were selected based on the following criteria.

579
First, we only considered viewpoints that are in the same TAD as at least one protein-coding or 580 lncRNA gene showing >3-fold sex bias in its expression. Second, the viewpoint must be within 1 kb of 581 the transcription start site (TSS) of a sex-biased gene, or it must overlap a sex-biased enhancer 582 (minimum 2-fold sex-bias in the sex bias in normalized DHS opening or H3K27ac mark intensity).

583
Third, the non-reading primer (Table S3A) was required to map to the genome uniquely, while the 584 reading primer was more stringently required to have > 89% unique sequence identity (i.e., no 18-mer 585 within a 20 nt primer sequence that maps elsewhere in the genome). Inverse PCR amplification of 1 microgram of each 4C template was performed using Platinum Taq DNA polymerase (Invitrogen, cat.

605
Computational analysis of ChIP-seq datasets. Sequence reads were split by barcode and mapped 606 to mouse genome assembly mm9 using Bowtie2 (v2.2.9). All reads not uniquely mapped to the 607 genome were excluded from downstream analyses. Peak calling was performed using MACS2 (v2.1.1) 608 with default parameters, and peaks that overlapped blacklisted genomic regions 609 (www.sites.google.com/site/anshulkundaje/projects/blacklists) were filtered out. Additionally, we 610 removed spurious peaks that exclusively contained PCR duplicated reads, defined as 5 or more 611 identical sequence reads that do not overlap any other reads. All BigWig tracks used to visualize 612 sequencing data in a genome browser were normalized for both sequencing depth and sample quality, 613 expressed as reads in peaks per million mapped reads (RIPM). In practice, the browser y-axis displays the read count from a given sample divided by the total number of reads in peaks (reads that overlap a 615 peak identified in any sample, or the union peak list), per million. Normalization was performed 616 separately for the CTCF and cohesin datasets. This approach is functionally similar to the quality 617 control metric known as Fraction of Reads in Peaks (FRiP) used by the ENCODE consortium [88].

618
These analyses were facilitated by a ChIP-seq analysis pipeline described elsewhere [89]. All samples 619 used in this study were judged to be of good quality, with a mean FRiP value of 0.217 and ranging 620 from 0.103 to 0.344. A full listing of samples sequenced and sequencing statistics is provided in Table   621 S3B.

623
To identify sex-differential ChIP-seq peaks, diffReps (v1.55.4) [90] was used with default parameters 624 and a window size of 200 bp to identify in-peak differential sites, i.e., diffReps sites that overlap a 625 MACS2 peak, defined below. The diffReps output list of sites was filtered to remove diffReps-identified 626 sites that did not meet the following conditions: overlap with at least one of the peaks in the union peak 627 list for the relevant factor (Table S1F, Table S1G for CTCF; Table S1H, Table S1I for cohesin), 628 contains at least 10 sequence reads, shows >2-fold sex difference, and has an FDR < 0.05. The

653
Those ΔCoh peaks that overlapped 2 of the 4 CTCF replicates were excluded from downstream 654 analyses. Similarly, sex-differential CTCF peaks (ΔCTCF) were designated CAC(ΔCTCF) peaks if they 655 overlapped a cohesin peak found in 2 or 3 of the three available cohesin ChIP-seq biological 656 replicates. Sex-differential CTCF peaks were designated Lone CTCF(ΔCTCF) peaks if they 657 overlapped peaks in either 0 or 1 of the three cohesin ChIP-seq biological replicates. 137 sex-658 differential CTCF peaks overlapped sex-differential cohesin peaks, and were thus CAC(ΔCoh/ΔCTCF); 659 50 of these 137 CAC peaks were autosomal (Table S1C). Sex-independent cohesin peaks, and sex-660 independent CTCF peaks, were respectively defined by ranking each peak based on the following 661 ratio: (RIPM-normalized ChIP signal for the merged male sample) / (RIPM-normalized ChIP signal for 662 the merged female sample), performed separately for CTCF and for cohesin. The 1,000 peaks whose 663 ratios were closest to 1 were defined as the set of sex-independent CTCF, and cohesin, peaks.

665
Discovery of intra-TAD loops. CTCF motif discovery was performed using the FIMO option from 666 MEME Suite (v4.10.0), and presence of a motif was defined as a motif score > 10. Intra-TAD loops for 667 female mouse liver were identified using the computational method described previously for male liver 668 [32]. The analysis pipeline was run for all CAC sites and with an initial loop count of 20,000, using the set of default parameters reported for male liver [32]. This analysis yielded 9,724 intra-TAD loops with 670 10,273 loop anchors in female liver; this compares to 9,543 intra-TAD loops and 9,052 loop anchors 671 identified in male liver [32]. The redundancy in loop anchors is a reflection of nested CAC-mediated 672 loop structures, as was described in other studies using experimentally-measured loop identification 673 compared to the computational approach used here; these studies include ChIA-PET analysis of the 674 cohesin subunit SMC1A in mouse embryonic stem cells [27] and Hi-C analysis in human GM12878 675 cells, where 9,448 loops were associated with 12,903 loop anchors [18,33]. Reciprocal overlap 676 between loops was analyzed using bedtools (bedtools intersect -wa -u -r -f 0.8), as described [32]. A 677 total of 2,527 intra-TAD loops were unique to either male or female liver; however, very few had 678 anchors that overlapped a sex-differential CAC site, suggesting that most are not biologically relevant.

679
Supporting this, the loops that were unique to either male or female liver were weaker than the loops 680 shared between male and female livers, and in many cases the loops narrowly met the significance 681 cutoff in one sex but not the other. This finding is similar to our earlier finding that tissue-specific loops 682 are often weaker than those predicted in multiple tissue types [32]. Intra-TAD loops for male and 683 female mouse liver are listed in Table S1J, and female intra-TAD loop anchors are listed in Table S1K; 684 a comparable listing for male liver is available in [32].

686
Computational analysis of 4C-seq datasets. Biological replicates were demultiplexed by index read 687 barcode. As the fastq files for each biological replicate contained sequence reads from multiple 688 viewpoints, the reads in each file were further split based on matches to the reading primer for each 689 viewpoint (Table S3A). Then, prior to mapping, we used FASTX-Toolkit (v0.0.14) to remove the first 20

796
A. Distribution of male/female ratios for all diffReps-identified sex-differential sites that overlap a 797 MACS2 peak for binding of cohesin (left) and CTCF (right). The y-axis shows the number of binding 798 sites per bin, and the x-axis shows the sex difference in binding, expressed as log2(Male/Female) fold-799 change. Gray bars represent binding sites below the 10 read minimum count threshold, which were 800 filtered out, and black bars represent sites that were statistically significant, but showed a |fold change| 801 < 2 (values between -1 and 1 on the graph). Pink and blue bars respectively represent female-biased 802 and male-biased sites above these thresholds.

803
B. Venn diagram indicating 137 sex-differential peaks are common between cohesin and CTCF.

804
Overlap is based on all sex-biased peaks, including male-biased and female-biased peaks on sex 805 chromosomes (autosomal sex-biased peak numbers are shown in parenthesis, at the right). This 806 pattern of limited overlap was also seen when the full set of unfiltered diffReps regions was examined 807 (Fig. S1A). In total, 1,847 unique peaks exhibited significant sex bias in liver chromatin binding of 808 CTCF and/or cohesin.

818
Rather, it may be due to differences in co-binding of transcription factors in male or female liver 819 displacing cohesin and/or CTCF within 1 nucleosome length.     Table S1     Methods). Green arrows indicate CTCF sex-differential CAC, and red arrows indicate CTCF sex-877 differential CNC and Lone peaks. Male/Female stranded polyA+ RNA-seq gene expression ratios [11] 878 are indicated above each panel.

900
E. Loss of cohesin binding has a 10-fold greater suppressive effect on male-biased genes with distal 901 sex-biased enhancers than those with proximal sex-biased enhancers. Shown is the mean expression for cohesin-depleted versus wild-type liver, such that a value of 0.1 represents a 10-fold reduction in 903 expression after cohesin loss. The median relative expression for DHS/H3K27ac-proximal genes is 904 0.69 (representing a modest suppressive effect of cohesin loss) and the corresponding median for 905 DHS/H3K27ac-distal genes is 0.07, indicating a >10-fold greater reduction in gene expression (p = 906 0.0087; M-W). Similar results were obtained when the definition of proximally-regulated genes was 907 relaxed to include genes with a TSS < 20 kb from either a male biased DHS or a male biased DHS 908 (median of 0.45 versus 0.042 for distal genes). Also see Table S2.

943
E. Nudt7 lacks male-biased DHS or H3K27ac at its promoter, but interacts with a distal male-biased 944 enhancer within the same intra-TAD loop. This viewpoint is anchored at a male-biased DHS (red

961
Chromatin states are colored: green indicates an enhancer-like state, blue indicates a promoter-like 962 state, and purple a transcribed-like state. Red indicates an inactive chromatin state (see Fig. S7C for 963 further details). Regions Y1 and Y2 are in chromatin state E13 in male liver but in inactive stateE2 in 964 female liver. Y1 and Y2 both include a short enhancer state region in male liver (E11 within region Y1; 965 E10 in region Y2). The absence of focal interactions between VP1 and VP2 supports the model of two 966 nested and insulated intra-TAD loops shown at the bottom. All tracks were normalized and are 967 presented as described in Fig. 4.

968
B. Expression of Nox4, but not the neighboring gene Tyr, is cohesin-dependent. Although Nox4 is 969 primarily regulated by proximal enhancers within the shorter intra-TAD loop, its full expression is 970 nevertheless dependent on cohesin. This may be due to the need for the intra-TAD loop structure; 971 however, loss of this insulation did not increase expression of Tyr. Expression of Nox4 was reduced by 972 62% (p<0.0001). Data presentation is as described in Fig. 4D. (1) All sex-differential sites output by diffReps without filtering for 990 overlap of a MACS2 ChIP-seq peak; (2) All sex-differential diffReps sites (rather than peaks, which 991 may contain multiple sites) from diffReps that overlap a MACS2 peak for a given factor; (3) All sex-992 differential hotspots identified by diffReps, which is an alternate method in the diffReps software 993 package to identify differential sites. Specifically, this last approach looks for clusters of differentially 994 co-regulated sites that might be missed by simple overlap analysis. Overlap for all Venn diagrams is 995 defined as ≥ 1 bp overlapping using bedtools. In some cases, two Rad21 peaks overlap a CTCF peak, 996 or vice versa, and therefore, the number of overlapping cohesin (Rad21) sites does not necessarily 997 equal the number of CTCF sites (hence, two numbers in the Venn overlap). shown in blue. The total number of differential peaks in each group is indicated below each chart.

1003
Overall, female-biased sites are comprised of a higher percentage of CAC sites than male-biased 1004 sites. Consequently, a larger percentage of male-biased peaks are CNC peaks (for ΔCohesin peaks) 1005 and Lone CTCF peak (for ΔCTCF peaks). Peak numbers here differ slightly from Fig. 1B for cohesin 1006 differential peaks, but not CTCF, because of our approach to categorizing peaks as CNC or CAC for 1007 cohesin peaks (see Methods). For CTCF we defined CAC peaks as genomic regions bound by CTCF 1008 that were also bound by cohesin in a majority of individual cohesin replicates (2 or 3 out of a total n=3 1009 per sex). Using the same approach for cohesin, we defined CAC peaks as genomic regions bound by motif. In contrast, a larger fraction of male-biased Lone CTCF peaks contain a CTCF motif, despite no

1122
The gene tracks and sex-biased sites are as described in Fig. 4. These bed file tracks are as follows