Classification of liver DHS and mapping to liver-expressed genes.

A. Percentages of each liver DHS subset (x-axis) classified as a weak enhancer, enhancer, insulator or promoter [20]. Enrichments of sex-biased DHS for being an enhancer or weak enhancer, an insulator, or a promoter region were determined by comparing to a background set of sex-independent DHS (66,116 sites). Significance was determined by Fisher Exact Test with Benjamini-Hochberg p-value adjustment: *, p<0.01; **, p<1E-10; ***, p<1E-50. Black asterisks, enrichment; red asterisks, depletion as compared to background DHS set. B. Cumulative frequency distribution of the distance to the nearest transcription start site in the same TAD for male-biased and female-biased enhancer (e), insulator (i), and promoter (p) DHS. C. ChromHMM emission probabilities for each of the 14 chromatin states developed for male and female mouse liver [18], which serves as a reference for data shown in panel D and in Fig. 7. Summary descriptions of the characteristics of each state are shown at the left and below. D. Chromatin state distributions for each sex-biased or sex-independent DHS sets. Chromatin state datasets for male and female liver (Table S5) were used for male-biased and female-biased DHS, respectively (Table S2A).

Discovery and characterization of dynamic and static male-biased DHS.

A. Model showing pulsatile male plasma GH pattern, with mice sampled between GH pulses, when STAT5 is inactive and cytoplasmic (STAT5-low, blue), or at a peak of plasma GH, when liver STAT5 is activated to its homodimeric, nuclear DNA-binding form, which enables STAT5 to open chromatin and bind to its consensus motif, TTCNNNGAA (STAT5-high, red). This intermittent (pulsatile) activation STAT5 leads to chromatin opening and closing at dynamic DHS (top). Static DHS also bind STAT5 intermittently but remain open between plasma GH pulses (bottom). B. EMSA of STAT5 DNA-binding activity in liver extracts prepared from individual male mice. These data represent liver extracts from 1 of 3 separate cohorts of mice; the other 2 cohorts are shown in Fig. S1. Labels at top indicate the DNase-seq library ID for each liver sample (Table S1). Red labels at top indicate STAT5-high activity based on the EMSA patterns displayed, and blue labels indicate STAT5-low activity livers. Numbers at bottom: number of each mouse whose liver was analyzed. Samples were all run on the same gel. See also Fig. S1. C. Principal component (PC) analysis of the distributions of DNase-seq reads per kilobase per million mapped reads for the top 600 or top 200 diffReps-identified sites that are more open in STAT5-high as compared to STAT5-low livers. Eigenvector values for principal component 1 are shown for the individual STAT5-high and STAT5-low liver samples. Dotted red line: empirical cutoff separating STAT5-high from STAT5-low DNase-seq samples; dotted black circles: outlier samples in each dataset. D. Boxplots of DNase-seq activity, in log2(reads per kilobase per million mapped reads), across the top 200 diffReps differential sites that open (as in C) for the STAT5-high DNase-seq libraries (red bars), for the STAT5-low DNase-seq libraries (blue bars), and for DNase-seq libraries for 9 individual male mouse liver ENCODE consortium samples (green bars, replicates 5 to 13, marked on x-axis). Thick dashed black line: arbitrary cutoff used to separate STAT5-high and STAT5-low liver samples. Samples from each group that did not pass the cutoff (red asterisks at bottom) are the same outlier livers circled in panel C. E. Normalized DNase-I cut site aggregate plots for STAT5-high (red) and STAT5-low male livers (blue), and for female livers (black). Peak cut site values are shown. Cut sites were aggregated across the sets of diffReps-identified DHS that open (left) or close (right) in response to endogenous STAT5 pulses in male liver, i.e., sites that show greater diffReps normalized DNase-seq signal intensity in STAT5-high compared to STAT5-low male livers. F. Venn diagram indicating overlap between endogenous STAT5 pulse-opened DHS sets identified by diffReps (2,832 sites that respond to liver STAT5 activity in a dynamic manner, which map to a total of 2,373 of the 70,211 DHS comprising the standard reference DHS set) and the indicated sets of sex-biased and sex-independent DHS that do not respond to a change in liver STAT5 activity, i.e., are static DHS. See Table S2A, columns H and I for full listings.

Dynamic and static male-biased DHS.

A. Normalized DNase-I cut site aggregate plots for STAT5-high (red) and STAT5-low male livers (blue) and female livers (black) across the genomic regions included in the sets of male-biased DHS (plots 1, 2) and sex-independent DHS (plots 3, 4), separated into subsets of DHS that either do (plots 1, 3) or do not bind STAT5 by ChIP-seq analysis (plots 2, 4). Plots 5-8 show corresponding plots for each of the indicated dynamic and static DHS sets. See Table S6 for peak normalized DNase-I site values. B. Bar plots showing number (values above bars) and percent of static and dynamic DHS with one or more occurrences of the indicated STAT5 motif (box), based on FIMO scan of the 70,211 standard reference DHS sequences. Dashed horizontal line: background, genome-wide level of STAT5 motifs at static sex-independent DHS. C. Percentage of dynamic (left) and static (right) DHS sets with zero or more occurrences of the STAT5 motif. X-axis: number of motif occurrences in each individual DHS (n =0-9), y-axis: percentage of full DHS set with the corresponding number of STAT5 motifs. Patterns for dynamic female-biased DHS represent a total of only 7 DHS and are thus not reliable.

Sex-biased DHS

DHS responsive to hypophysectomy, their enrichment for class I and II sex-biased gene targets and responses to GH pulse replacement.

A. Model for impact of ablation of pulsatile GH secretion by hypophysectomy on STAT5-induced chromatin opening. DHS that close following hypophysectomy due to the loss of STAT5 binding reopen within 30 min of exogenous GH treatment, which rapidly reactivates STAT5. B. Numbers of liver DHS that open or close following hypophysectomy (Hx) and in response to a single injection of GH given to hypox male (M) or female (F) mice and euthanized 30, 90 or 240 min later. DHS opening and closing were determined compared to the indicated controls. C. Distributions of sex-biased and sex-independent DHS that open, close or are unchanged (static) by hypophysectomy (Hx) in male and female mouse liver. See Table S4H for full details. D. Enrichment of hypophysectomy-responsive DHS as compared to DHS whose accessibility is unchanged by hypophysectomy for mapping to the four indicated classes of sex-biased genes (see Fig. S4). Class I and II sex-biased genes were identified from RNA-seq gene expression data collected from intact and hypox male and female liver samples (Table S3). The bottom section of the table shows the total number and percentage of sex-biased genes in each of the four indicated sex-biased gene classes that respond to hypophysectomy, as marked (Table S4A, Table S4I). E. Subset of all dynamic DHS (Fig. 2F) that close following hypophysectomy in male mouse liver (n=1,487) and then respond to GH pulse replacement at the indicated time points. Total number of dynamic DHS shown here is lower than the full set of 2,373 dynamic DHS (Fig. 2F), as 70 of these DHS were not identified as DHS in the hypophysectomy study (Table S4H).

DHS activity aggregate plots for hypophysectomy and GH pulse treatment time course.

A. Normalized DNase-I cut site aggregate plots for each of the indicated sets of static and dynamic DHS showing the effects of hypophysectomy of male (MHx) and female mice (FHx) and of GH pulse treatment (MHx+GH) for 30, 90 and 240 min as compared to intact females (c.f., Fig. 3A). Reference values for normalized DNase-I cut site activity in intact male liver are from Fig. 3 and Table S6. B. Plots as in A are shown for the indicated subsets of 2,729 male-biased DHS (plots 1 and 2) and for the indicated subsets of the set of 66,116 sex-independent DHS (plots 3 and 4), i.e., DHS subsets with STAT5 bound (plots 1 and 3) or without STAT5 bound (plots 2 and 4), based on ChIP-seq data for STAT5 in male mouse liver from [15].

Enrichments of sex-biased histone marks at dynamic and static DHS sets.

A. Enrichments of male-biased histone marks, and B. enrichments of female-biased histone marks. Data is presented as bar graphs showing significant enrichments and significant depletions (negative Y-axis values) for the 6 indicated liver histone marks for each of four DHS sets (see Fig. 2F). Enrichment data are graphed as 6 sets of four bars each, separated by vertical blue lines and ordered from 1 to 4, as shown in the box in A and as marked above select bars. The set of 64,584 static sex-independent DHS was used as the background for the enrichment calculations. Fisher Exact test significance values (log p-values, indicated by bar color) are shown for all values that are significant at p<E-03. Values that did not meet this significance threshold are graphed as ES = 1 (horizontal dashed green line); thus, all bars shown, except those graphed at ES = 1.0, represent statistically significant enrichment or depletion. Full details on the numbers of sites, the source publications used to identify these genomic regions, and corresponding BED files are shown in Table S7. C. Proposed model for chromatin states adopted by dynamic male-biased DHS, by static male-biased DHS, and by static female-biased DHS in male liver (left) and in female liver (right) in response to the stimulatory and/or repressive actions of plasma GH pulses (in male liver) and persistent GH exposure (in female liver). Histone H3 marks are shown by small colored ovals attached to histone tails (see legend in box). H3K27me3 is specifically used to repress chromatin at female-biased DHS in male liver, and H3K9me3 is specifically used to repress chromatin at both classes of male-biased DHS in female liver. Sex-biased H3K36me3 marks are uniquely associated with static male-biased DHS in male liver and with static female-biased DHS in female liver and may serve to keep these DHS constitutively open by inhibiting the introduction of H3K27me3 repressive marks [35, 51] at static female marks in female liver, and perhaps also the introduction of H3K9me3 repressive marks at static male-biased DHS in male liver. Continuous GH infusion in males overrides the stimulatory, chromatin opening effects of GH/STAT5 pulses on dynamic male-biased DHS, leading to extensive closing of dynamic male-biased DHS (Table S2E). DHS with a combination of activating and repressive histone marks in one but not both sexes (i.e., sex-dependent bivalent character) are indicated. The degree of chromatin accessibility is indicated by the relative distance between nucleosomes. Black arrows indicate DHS stimulation of gene transcription upon interaction with a nearby or distal gene promoter.

Enrichments of chromatin states at sex-biased and sex-independent dynamic and static DHS.

A. Enrichments of male liver chromatin states at each of the 4 indicated DHS sets, and B. enrichments of female liver chromatin states at each of the 4 indicated DHS sets. Data are presented as described in Fig. 6, with the set of 64,584 static sex-independent DHS used as background for the enrichment calculations. Many of the background set of DHS are active regulatory regions replete with enhancer marks, hence it is to be expected that the dynamic and static sex-biased DHS sets would show low, albeit significant enrichments for enhancer states E5, E6 and E9-E11. Full details on these analyses including DHS chromatin states and source publications used to identify these genomic regions are provided in Table S7.

STAT5 regulates sex-dependent hepatocyte gene expression at three distinct step.

(1) Sex-biased chromatin opening: GH pulse-induced chromatin opening at dynamic male-biased DHS is driven by pulsatile GH activation of STAT5 in male liver, whereas persistent activation of STAT5 in female liver is associated with static female-biased chromatin opening. (2) Sex-biased transcriptional activation: Sex differences in open chromatin regions and their accessibility enable GH-activated STAT5 to bind chromatin in a sex-biased manner and induce the transcriptional activation of sex-biased genes. (3) Sex-based transcriptional repression: The sex-biased regulatory genes regulated in step (2) include the GH/STAT5-dependent repressor proteins BCL6 (male-biased) and CUX2 (female-specific), which reinforce sex differences in transcription by preferentially suppressing the expression of female-biased and male-biased genes, respectively, as indicated.

EMSA analysis of STAT5 DNA-binding activity in liver extracts prepared from individual male mice.

Shown here are EMSA data for 2 of 3 separate cohorts of mice; the 3rd cohort is shown in Fig. 2B. Livers whose nuclei were used for DNase-seq analysis are marked at the top (Table S1); red labels at top indicate STAT5-high activity based on the EMSA patterns displayed, and blue labels indicate STAT5-low activity livers. Also shown here are EMSA activity for 8 other livers, which have low to intermediate STAT5 EMSA activity and were not included in downstream analysis. Numbers at bottom: liver sample numbers from each mouse cohort.

STAT5 binding is associated with chromatin opening.

A. Bar graph of the number and percentage of dynamic and static DHS that have one or more STAT5 binding sites, based on STAT5 ChIP-seq data for mouse liver. BEDTools was used to determine the overlap between the merged list of 15,094 STAT5 binding sites [1] and the standard reference set of 70,211 liver DHS used in this study. Dashed horizontal line: background, genome-wide level of STAT5 binding at static sex-independent DHS. B. Distribution of STAT5 ChIP-seq signal intensity values for the sets of dynamic and static DHS. For each DHS that contains a STAT5 binding site (numbers shown in A), the corresponding normalized STAT5 ChIP-seq read count (average of STAT5 male-high samples and STAT5 female-high liver samples; [1]) was obtained and used to calculate the indicated distributions. C. Normalized DNase-I cut site aggregate plots for male livers at STAT5-high, STAT5-low, and female liver for the indicated sets of male-biased, female-biased, and sex-independent DHS, analyzed separately for dynamic DHS subsets (plots 1A, 1B, 4A, 4B) and static DHS subsets (plots 2A, 2B, 3A, 3B, 5A. 5B). Top row: STAT5-bound subsets; bottom row: non-STAT5-bound subsets. See Table S6 for DNase-seq aggregate plot peak values.

TF motifs discovered de novo from sets of DHS sequences.

De-novo motif discovery was carried out on the sets of dynamic and static DHS sequences using the MEME and DREME algorithms of MEME-ChIP [2]. Listed are the de-novo discovered motifs and their associated p-values for: A, dynamic male-biased DHS (834 sites); B, dynamic sex-independent DHS (1,532 sites); C, static male-biased DHS (1,895 sites); D, static female-biased DHS (1,359 sites); and E, static sex-independent DHS (64,584 sites). Shown are the top most significant motifs. No motifs were discovered from the very small number of dynamic female-biased DHS (7 sites). The presence of the STAT5B motif is a distinguishing characteristic that separates the dynamic from static DHS, which supports our hypothesis that direct binding of STAT5 plays a key role in chromatin opening in response to GH pulses at these dynamic DHS.

Sex-biased genes can be classified based on their pituitary hormone-dependence.

A. Model for plasma GH profile dependence and responses of class I and class II male-biased and female-biased genes to pituitary hormone ablation in each sex by hypophysectomy (‘hypox’). Figure from [3]. B. Sex-specific genes were identified from RNA-seq data from intact male and female mouse liver based on a gene list comprised of 24,197 RefSeq and 3,152 multi-exonic lncRNA genes. First, sex-specific genes were identified with |fold-change| > 1.5, adjusted p-value < 0.05 (for RefSeq genes), and |fold-change| > 2.0, adjusted p-value < 0.05 (for lncRNA genes), with FPKM > 0.25 for the sex with greater signal intensity for both RefSeq and lncRNA datasets (see Methods). Sex-specific genes that were responsive to hypophysectomy (|fold-change| > 2 and an adjusted p-value < 0.05) were further classified into class I, II and corresponding subclasses (IA, IB, IC, IIA, and IIB) based on RNA-seq data from intact and hypox male and female liver samples. Shown are the full list of RefSeq and lncRNA genes. See Table S3 for full gene listings.

DNase cut site aggregate plots for the GH time course DHS data.

Shown are normalized DNase-I cut site aggregate plots for livers from hypophysectomized male mice (MHx), hypophysectomized male mice treated with GH (MHx+GH) then euthanized after 30, 90 or 240 min, and intact female and hypophysectomized female mice (FHx) across various sets of dynamic and static male-biased DHS, static female-biased DHS, and static sex-independent DHS (see Fig. S1C). Each DHS set was separated into subsets based on STAT5 binding, as determined by Chip-seq (top row: STAT5-bound DHS subsets; bottom row: corresponding non-STAT5-bound DHS subsets). Key results: MHx mice treated with GH for either 30 or 90 min show the largest degree of chromatin opening in the set of 710 dynamic male-biased DHS that bind STAT5. Chromatin opening decreased substantially after 240 min (plot 1A), at which time the activation of liver STAT5 DNA binding activity is fully reversed [4]. Smaller increases in chromatin opening were observed with GH pulse treatment at the subset of 124 dynamic male-biased DHS that did not bind STAT5 (plot 1B), consistent with their dynamic responses to endogenous GH pulsation, and suggesting that chromatin opening at these sites proceeds by a distinct mechanism than at the STAT5-bound dynamic male-biased sites. Static male-biased DHS with STAT5 bound (597 sites) showed a more modest increase in chromatin opening with GH pulse treatment (c.f., higher basal level in MHx control and lower induced level with GH pulse; plot 2A), while static male-biased DHS without STAT5 binding (1,298 sites) showed little or no GH pulse responsiveness (plot 2B). Importantly, the increase in chromatin opening at the STAT5-bound static male-biased DHS largely persists at 240 min (plot 2A), in contrast to the more substantial decline in chromatin opening seen for at STAT5-bound dynamic male-biased DHS (plot 1A). This suggests that, while STAT5 can open chromatin at the static male-biased DHS, it is not required to maintain chromatin accessibility between the naturally occurring endogenous plasma GH pulses. Chromatin opening at female-biased DHS was decreased by hypophysectomy, both at the 258 sites bound by STAT5 and at the 1101 sites that did not show STAT5 binding (plots 3A and 3B). Further, GH pulse treatment of MHx mice stimulated a modest decrease in liver chromatin accessibility at both sets of female-biased DHS. Finally, the dynamic, but not the static sex-independent DHS showed large increases in chromatin opening following GH pulse treatment (plot 4A and plot 5A), similar to the dynamic male-biased DHS. Overall, DHS that bind STAT5 showed higher levels of chromatin opening than DHS that do not bind STAT5 (top row vs. bottom row). Notably, this difference in chromatin accessibility is preserved in hypophysectomized male and female liver, even though liver STAT5 is inactive and cannot bind DHS under these conditions due to the absence of GH stimulation, indicating a role for other pituitary-determined factors in chromatin opening.

Impact of hypophysectomy and GH pulse replacement on set of 2,373 dynamic DHS.

S6A. Four-oval Venn diagram showing the number of STAT5-high DHS that were close following hypophysectomy and/or open in response to a single exogenous GH pulse given to hypophysectomized male mice. The set of STAT5-high (dynamic) DHS (n = 2,373 sites; Fig. 2E) was analyzed to determine their overlap with the set of DHS that close following hypophysectomy (n = 1,487 sites) or DHS that open when a single exogenous GH pulse is given to hypophysectomized male mice and liver tissue collected after 30 min (n = 1, 475 sites), 90 min (n = 1,393 sites) or 240 min (n= 547 sites) (see Fig. 4 and Table S2: column I = “dynamic” combined with columns AK-AM = “dDHS_open”). S6B. Flowchart of STAT5-high DHS (i.e., n = 2,373 dynamic DHS) identifying hypox-responsive and exogenous GH pulse-responsive DHS in male mouse liver. Subsections of the 4-oval Venn diagram (shown in A) are used to illustrate the DHS subsets defined by the flowchart (7 subsets), including a set of 399 DHS that do not undergo chromatin closing following hypophysectomy and also do not undergo chromatin opening following an exogenous GH pulse. S6C. Boxplot analysis showing the magnitude of chromatin opening between STAT5-high and STAT5-low male livers for the 7 DHS subsets. The x-axis shows the DHS subsets numbered 1-7 with the corresponding number of DHS sites (as defined in B). RiPPM normalized DNase-seq read counts were obtained from the STAT5-high and STAT5-low male liver samples for the standard reference set of 70,211 DHS (Table S2). The relative magnitude of chromatin opening (i.e., fold-change value) was calculated from the ratio of STAT5-high/STAT5-low RiPPM-normalized DNase-seq read counts in the DHS region. Shown are the distributions of fold-change values for DHS subsets numbered 1-7 defined from the flowchart in B. A Wilcoxon rank-sum test with Benjamini–Hochberg p-value adjustment was used to determine significant differences between distributions of fold-change values (*P < 0.05, **P < 1e-03, ***P < 1e-10). See Table S6 for DNase-seq aggregate plot peak values. S6D. DNase-I cut site aggregate plots for the 7 DHS subsets defined in B. See Table S6 for DNase-seq aggregate plot peak values. Key findings shown in B, C and D: Another dynamic DHS subset, comprised of 399 DHS, was unresponsive to hypophysectomy and to GH pulse treatment (B). These DHS showed weaker mean chromatin accessibility and lower differential accessibility between STAT5-high and STAT5-low livers than the hypophysectomy and exogenous GH responsive sites (C, D). Of the STAT5-high DHS (n=2,373 sites), the subset that responded to hypox and/or a single exogenous GH pulse (set 3, n=1,974) showed significantly greater chromatin opening induced by an endogenous GH/STAT5 pulse than the non-responsive subset (n=399) in pituitary-intact male livers (set 3 vs. set 2). Data in D validate the large increase in chromatin accessibility in the set of responsive DHS compared to non-responsive DHS (set 3 vs. set 2). Of the responsive DHS (n=1,974 sites), the subset that was opened by an exogenous GH pulse (set 5, n=1,620 DHS) showed greater chromatin opening induced by an endogenous GH/STAT5 pulse than the DHS subset not opened by a single exogenous GH pulse (n=354) (set 4), a finding that was confirmed by the DNase-I aggregate cutting profiles shown in D. Of the GH pulse opened DHS (n=1,620 sites), the subset that was closed following hypophysectomy (n=1,133 sites) showed greater chromatin opening due to an endogenous GH/STAT5 pulse than the subset that was not closed following hypophysectomy (n=487 sites) (set 7 vs. set 6), as was also confirmed by DNase-I aggregate cutting profiles in D. Further, the DHS that closed after hypophysectomy and then reopen following an exogenous GH pulse at any of the three time points (30, 90, or 240 min) (n=1,133 sites; see Fig. 4D) showed the largest magnitude (C) and relative levels of chromatin opening (D) due an endogenous GH/STAT5 pulse when compared to the other DHS subsets (set 7 vs. sets 1-6).

Chromatin state analysis of static and dynamic male-biased DHS.

A. Distribution of static and dynamic DHS in the defined classes of enhancer DHS, weak enhancer DHS, insulator DHS and promoter DHS, based on histone mark patterns [5]. B. Chromatin state distributions in male mouse liver of dynamic and static male-biased DHS, based on the 14 chromatin state model developed from the combination of six active and repressive histone marks and DHS, which segment the mouse genome into inactive, bivalent, enhancer-like, promoter-like, or transcribed-like states [6]. Chromatin state data are shown in Table S5. Right side of figure shows the emission probabilities for the six histone marks and DHS for each of the 14 chromatin states (reproduced from Fig. 1C). Key results: Overall, 92-96% of dynamic and static male-biased DHS were classified as enhancers, with a larger fraction being weak enhancers [5] in the case of static male-biased DHS. Promoter DHS and insulator DHS comprised the balance of each male-biased DHS set (2-5% each) (A). Similarly, 86% of female-biased DHS and 90% of dynamic sex-independent DHS were classified as enhancers or weak enhancers, unlike static sex-independent DHS, where insulator and promoter DHS designations were much more common (15-20% each versus 3-8% for dynamic sex-independent DHS (A; also see Fig. 1A). We also did not observe major differences in chromatin state distributions between dynamic and static male-biased DHS based on a 14-state map developed for male and female mouse liver [6]. Thus, dynamic and static male-biased DHS both showed a high frequency (59-72%) of chromatin state E6, whose emission parameters indicate a high frequency of DHS and of the activating chromatin marks H3K27ac and H3K4me1 (B). Much smaller percentages of both male-biased DHS subsets were in other chromatin states, primarily enhancer states E5 and E11, promoter state E7, and bivalent state E12, which is characterized by the presence of both activating marks (H3K4me1) and repressive marks (H2K27me3).

Top enriched Gene Ontology (GO) terms identified by GREAT analysis of the predicted gene targets of each of the indicated 4 DHS sets.

Mouse mm9 genomic coordinates for each DHS set were entered into the web-based tool GREAT (v.4.0.4) ( and target genes identified under default conditions (Association rule: Basal+extension: 5000 bp upstream, 1000 bp downstream, 1 million bp max extension). Shown are the top enriched GO Biological Process terms for each DHS set using the whole genome background setting, with –log10 Binomial p-values shown. Complete data output from GREAT, including listings of gene targets and the DHS associated with each DHS are shown in Table S8.