Dynamic nature of STAT5 binding in male versus female mouse liver. (A) EMSA gel showing STAT5 DNA binding activity status of 8 individual male and 4 individual female mouse livers. STAT5 activity was high (H), low (L), or intermediate (Int) in individual livers. STAT5 activity was substantially above background in low-activity female livers but was undetectable in low-activity male livers. The arrow at the right identifies STAT5 EMSA complex. (B) Continuous infusion of male mice with GH for 7 days abolished the pulsatile (intermittent) nature of liver STAT5 activity seen in untreated males, as indicated by the absence of STAT5 activity-free male livers in the set of 12 individuals shown and in other sets of male livers (data not shown). Typically, 30 to 60% of randomly selected untreated male livers show a STAT5-high or STAT5-intermediate EMSA profile. The two lanes on the left are from untreated males showing low and high STAT5 activity, as indicated. Arrow, STAT5 EMSA complex. (C) Number of STAT5 peaks identified by ChIP-Seq in each of the 4 indicated groups of livers, which differed by sex and by high versus low STAT5 activity status. Detailed peak lists are provided in Table S1 of the supplemental material. (D) UCSC Genome Browser view of the distribution of STAT5 reads around Igf1. The track with 5 red arrows indicates 5 STAT5 binding regions identified by ChIP-PCR (17). Other tracks mark liver DHS sites (39), STAT5 peaks in the merged STAT5 peak set (see Table S1F in the supplemental material), and occurrences of the de novo-discovered STAT5 motif identified in this study. The bottom four tracks show the STAT5 ChIP-Seq reads for the four indicated groups, displayed as wig files generated using MACS software with a sliding window average of 100 bp and a 20-bp step size. STAT5 tracks were normalized to the STAT5-MH sample (see Fig. S1E in the supplemental material).

Male-enriched and female-enriched STAT5 peaks: association with HNF6/CDP motifs and epigenetic marks. (A) An M versus A plot of the number of STAT5-MH reads versus STAT5-FH reads in the set of 15,094 merged STAT5 peaks (see Table S1F in the supplemental material) after normalization using MAnorm (see Materials and Methods). Each peak is associated with an M-value and an A-value: M = log₂[(number of STAT5-MH sequence reads)/(number of STAT5-FH reads); A = 0.5 × [log₂(number of STAT5-MH reads) + log₂(number of STAT5-FH reads)]. Using a fold change of >2 as the cutoff (|M value|, >1; the pair of horizontal black lines), we identified 1,765 STAT5-MH-enriched peaks (blue dots), 1,791 STAT5-FH-enriched peaks (red dots), and 11,538 common peaks (black dots). The total number of STAT5 peaks (15,094) corresponded to the STAT5-MH and STAT5-FH merged peak set (see Materials and Methods). Full details for each peak are shown in Table S1F of the supplemental material. (B and C) Number of de novo-discovered STAT5 motifs (B) and HNF6/CDP family motifs (C) found in each consecutive set of 500 merged STAT5 peaks ranked according to the ratio of male to female STAT5 binding activity, i.e., high M-value (male-enriched STAT5 peaks) to low M-value (female-enriched STAT5 peaks). The M-value scale is shown along the top axis, the STAT5 peak rank is shown along the bottom axis, and the enrichment score is shown along the right axis. (D) Sex ratio of DHS site (red) and H3-K27me3 (blue) sequence reads (log₂ scale) versus sex enrichment of the set of 15,094 merged STAT5 peaks ranked by M-value, as in panels B and C. (E) Sex ratio of H3-K4me1 (green), H3-K4me3 (purple), and H3-K9me3 (blue) sequence reads (log₂ scale) versus sex enrichment of the set of 15,094 merged STAT5 peaks ranked by M-value. (F) Model depicting sex differences in chromatin accessibility, H3-K27me3 marks (black flags), and H3-K4me1 and H3-K4me3 marks (white flags) around sex-dependent STAT5 binding sites, based on the findings shown in panels D and E; see also Fig. S3D to F in the supplemental material. Sex-enriched STAT5 binding correlated positively with gene expression status (forward arrow), chromatin accessibility, and H3-K4me1 and H3-K4me3 marks and negatively with H3-K27me27 marks. However, female-enriched STAT5 binding sites in male liver showed more open chromatin than male-enriched STAT5 binding sites in female liver (compare Fig. S3C and D in the supplemental material), despite the somewhat higher level of H3-K27me3 marks at the female-enriched STAT5 binding sites (compare Fig. S3E with F in the supplemental material).

Sex-biased binding of STAT5 at sex-specific genes. Shown are UCSC Genome Browser screen shots with examples of male-enriched binding of STAT5 at male-specific genes (A and B) and female-enriched binding of STAT5 at a female-specific gene (C). Each panel shows the sex specificity of each gene (male/female or female/male expression ratio) as determined by microarray analysis (61), with horizontal black bars indicating the chromosomal intervals found to contain binding sites for STAT5 in STAT5 high-activity male (MH) and female (FH) livers. The bottom four tracks show the STAT5 ChIP-Seq reads for the four indicated mouse groups, as shown in Fig. 1D.

BCL6 binding in male liver at sites associated with female-enriched STAT5 binding to female-specific genes. Shown are UCSC Browser screen shots of STAT5 and BCL6 binding to the four female-specific genes, Cux2, Bmper, Ntrk2, and Nnmt. (A and B) The same four tracks of STAT5 binding shown in Fig. 5, plus two additional tracks that showed BCL6 binding in male and female liver, as marked above each track. (C and D) Screen shots for the three indicated tracks only, with horizontal bars marking the peak regions for the indicated tracks, as identified with MACS peak-calling software. Lower levels of STAT5 and BCL6 binding than those seen here were apparent in the STAT5-ML, STAT5-FL, and BCL6-F tracks (see data presented in Fig. S1E of the supplemental material). Cux2 (A) belongs to STAT5/BCL6 class V, and the other three genes belong to class VI (Fig. 6D; see also Table S2C and D in the supplemental material). STAT5 binding persists in STAT5-FL liver compared to STAT5-FH liver in the case of Cux2 (A) and other genes (see Fig. S1E), whereas STAT5 binding rarely persists in STAT5-ML liver compared to STAT5-MH liver. The female-to-male expression ratios shown were determined by microarray analysis (61), except for that for Cux2, which was determined by qPCR (35).

Model for regulation of sex-specific liver gene expression by STAT5, BCL6, and factor(s) with an HNF6/CDP motif. Male (pulsatile) plasma GH and female (near-continuous) plasma GH activate distinct temporal patterns of liver STAT5 activity (intermittent STAT5 activity in males versus persistent STAT5 activity in females) (Fig. 1A and B). These sex-differential patterns of liver STAT5 activity are, respectively, associated with male-enriched STAT5 binding at male-biased genes and female-enriched STAT5 binding at female-biased genes (Fig. 4). The transcriptional repressor BCL6 is more highly expressed in male liver, where it preferentially binds to STAT5 sites associated with female-biased genes (Fig. 6 and 7); this competition by BCL6 for STAT5 binding is proposed to augment liver sex differences by repressing the expression of female-biased genes in male liver. In female liver, where the female plasma GH profile suppresses BCL6 (43), the repression of female-biased genes seen in male liver is relieved. Binding site motifs for HNF6/CDP factors are preferentially associated with male-enriched STAT5 binding sites (Fig. 3C), suggesting that an HNF6/CDP factor(s) contributes to the regulation of male-biased genes. Two possible mechanisms for this regulation are illustrated: (i) in male liver, an HNF6/CDP factor, perhaps HNF6 (ONECUT1), cooperates with STAT5 in positively regulating male-biased genes (top); (ii) in female liver, the HNF6/CDP family member CUX2 (CUTL2), which has been characterized as a repressor (21) that is expressed in female but not male liver (35), binds nearby male-enriched STAT5 binding sites and inhibits the expression of male-biased genes (bottom).

STAT5 binding sites and motifs in STAT5-high and STAT5-low livers. (A) Graph showing the number of STAT5-MH peaks (9,799 peaks ranked along the x axis by MACS peak score in descending order) that overlapped with STAT5-ML peaks in each consecutive set of 100 peaks, using a step size of 1 peak. (Inset) Venn diagram comparing the STAT5-MH and STAT5-ML peak sets, indicating the number of STAT5-MH peaks that were in common with the STAT5-ML peak set (“common peaks”) and the number of peaks that were unique to each peak set. (B) Graph showing the number of STAT5-FH peaks that overlapped with STAT5-FL peaks in each consecutive set of 100 peaks (12,753 STAT5-FH peaks ranked along the x axis), analyzed as described for panel A. (Inset) Venn diagram comparing the STAT5-FH and STAT5-FL peak sets, indicating the number of STAT5-FH peaks that are in common with the STAT5-FL peak set (common peaks) and the number that are unique to each peak set. (C) Sequence logo representing the enriched STAT5 motif identified by de novo motif discovery applied to STAT5-MH peaks (top). This logo is almost identical with that discovered de novo from the STAT5-FH peak set (data not shown) and is very similar to the Transfac STAT5B matrix logo (bottom). (D) Distribution of the STAT5 motif around the summits of STAT5-MH unique and STAT5-ML unique peaks and in peaks common to both sets, as defined in the Venn diagram in panel A. (E) Corresponding distribution for the STAT5-FH unique and the STAT5-FL unique peaks and for the peaks common to both sets, as defined in the Venn diagram in panel B. Shown along the y axis is the total number of motifs in each consecutive window of 60 bp, with a step size of 1 bp, normalized to the total number of peaks in each sample. (F) High-quality STAT5 ChIP-Seq peaks have greater numbers of STAT5 motifs. Shown is the number of STAT5-MH peaks that contained 0, 1, 2, 3, or 4 occurrences of a STAT5 motif in each peak, determined for the three indicated subsets of STAT5-MH peaks and grouped based on the STAT5 ChIP-Seq MACS score. STAT5-FH peaks produced a very similar pattern (data not shown).

Relationship between sex-enriched STAT5 binding and pituitary GH-dependent and sex-biased gene expression. (A) The set of 15,094 merged STAT5 peaks (Fig. 3A) was mapped to STAT5 target genes, defined as RefSeq genes within 10 kb of a merged STAT5 peak. A total of 10,617 (70%) of the 15,094 STAT5 peaks mapped to a total of 5,076 genes, which were classified as male-enriched, female-enriched, or common STAT5 targets, as indicated in the Venn diagram. (B) Enrichment of each of the 3 sets of STAT5 targets shown in panel A for genes that respond to hypophysectomy (Hypox) in either male or female mouse liver. The graph shows the changes in gene expression following hypophysectomy in male versus female liver, and it is based on microarray data (61). Data are graphed as the log₂ expression ratios for hypophysectomized versus untreated controls (UT; i.e., sham-operated mice) in males (x axis) and females (y axis) for genes that passed the microarray significance filter of P < 0.001. Each point represents one gene, and the color indicates its female/male expression ratio in untreated mouse liver, as indicated in the log₂-scale color bar at the right. Shown are ESs and Fisher exact test P values calculated for each of the three STAT5 target gene sets in panel A in the sets of genes that fell into each quadrant, using all RefSeq genes represented on the microarray as background. Blue lettering indicates ES values for STAT5-MH-enriched targets, brown lettering indicates ES values for STAT5-FH-enriched targets, and black lettering indicates ES values for STAT5 male and female common targets. The finding of STAT5 target genes significantly enriched in all but the first quadrant, which represents genes induced by hypophysectomy in both male and female liver, is consistent with GH-activated STAT5 primarily being a positive regulator of gene expression. (C) Genes that respond to hypophysectomy at a P level of <0.001 are enriched in STAT5 targets that contain a STAT5 motif, in both male liver (left) and female (right) liver. Nearly identical results were obtained using other cutoffs to define hypophysectomy-responsive genes (data not shown). (D) Fraction of sex-biased RefSeq genes that are targeted by STAT5 as a function of distance from STAT5 peak to gene body used to identify the gene as a STAT5 target. (E) A total of 2,282 sex-biasedRefSeq genes were ranked based on male/female expression ratio, as shown in the color bar at the bottom (log₂ ratios). For each of the three sets of STAT5 target genes shown in panel A, the fraction of targets present in each consecutive set of 400 sex-biased genes was calculated. The frequency of STAT5 target genes increased dramatically with increasing expression sex bias, most noticeably for the female-specific genes (right). (F) Gene set enrichment analysis showing a significant NES (normalized enrichment score) for STAT5-MH-enriched targets in male-biased genes and for STAT5-FH-enriched targets in female-biased genes. The x axis represents all 21,794 RefSeq genes present on the Agilent 44K_v1 microarray platform used to obtain the male/female expression ratios (61); the y axis presents the running enrichment score. (G) Enrichment of male-specific and female-specific genes in six sets of STAT5 targets, which were identified using six subsets of STAT5 peaks, as defined by the indicated ranges of normalized log₂ male/female STAT5 sequence reads (M-values) and as shown in Fig. S4 of the supplemental material.

The seven classes of STAT5 and BCL6 binding sites: motifs, targets, and enrichment for sex-specific genes. (A) Venn diagrams depicting the number of peaks that overlapped between the indicated STAT5 and BCL6-M peak sets, as determined by comparing the genomic coordinates (see Table S1 in the supplemental material). (B) The DNA sequence logos (motifs) for the seven different classes of STAT5 and BCL6 binding sites defined by the Venn diagram in panel A (labeled I to VII), identified by de novo motif discovery. Motif discovery was carried out using peaks trimmed to a total width of 600 nt centered around the peak summit. The motif shown for class I sites is almost identical to the motifs for classes II and III, and the motif shown for class VI sites is almost identical to the motif for class V, as indicated (see also Fig. S5B in the supplemental material). (C) Cumulative distribution of the distance from the BCL6 motif to the nearest STAT5 motif for the set of STAT5 and BCL6 common peaks (i.e., classes IV, V, and VI in panel A) that contained both motifs. (D) Venn diagrams depicting the number of gene targets that overlapped between the indicated STAT5 and BCL6-M target gene sets. The target gene sets identified by this Venn diagram are designated I to VII and correspond to the seven peak sets shown in panel A. (E) Enrichment of male-specific genes versus female-specific genes in each of the indicated seven classes of STAT5 and BCL6 targets, as marked and based on the target gene sets shown in panel D. ESs and Fisher exact test P values are shown on the top of each bar set.

Transcription factor motif (A) and gene ontology (B) enrichment analysis for 7 sets of STAT5 and BCL6 peak targets. (A) Heat map showing ES values of all motifs enriched (P < 1E − 10; ES, >1.7) in at least 1 set of the 7 peak sets shown in Fig. 6A, using as background the motifs present in all mouse liver DHS sites (39) that showed no overlap with STAT5-MH or STAT5-FH peaks or BCL6-M peaks. The motif families used for enrichment analysis were the set of 97 motif families described previously (41), supplemented with additional motifs as described in Materials and Methods. (B) Heat map showing ES values for GO terms meeting the combined threshold of ES of >2.5 and P of <0.01, and the number of genes in each gene list containing >8 in at least 1 of the 7 gene sets defined in Fig. 6D, using all RefSeq genes as background. Genes in each of the 7 gene sets that had enriched GO terms are listed in Table S3 of the supplemental material. Dark red represents highly enriched GO terms, white indicates no enrichment, and blue indicates depletion of the GO term from the indicated gene set, as specified by the ES scale color bar at the right. The vertical pink bar at the right delineate GO terms associated with BCL6 and STAT5 common targets (classes IV to VI).