Sexual dimorphism in mammalian liver impacts genes affecting hepatic physiology, including inflammatory responses, diseased states, and the metabolism of steroids and foreign compounds. Liver sex specificity is dictated by sex differences in pituitary growth hormone (GH) secretion, with the transcription factor signal transducer and activator of transcription (STAT)5b required for intracellular signaling initiated by the pulsatile male plasma GH profile. STAT5a, a minor liver STAT5 form >90% identical to STAT5b, also responds to sexually dimorphic plasma GH stimulation but is unable to compensate for the loss of STAT5b and the associated loss of sex-specific liver gene expression. A large-scale gene expression study was conducted using 23,574-feature oligonucleotide microarrays and livers of male and female mice, both wild-type and Stat5a-inactivated mice, to elucidate any dependence of liver gene expression on STAT5a. Significant sex differences in expression were found for 2,482 mouse genes, 1,045 showing higher expression in males and 1,437 showing higher expression in females. In contrast to the widespread effects of the loss of STAT5b, STAT5a deficiency had a limited but well-defined impact on liver sex specificity, with 219 of 1,437 female-predominant genes (15%) specifically decreased in expression in STAT5a-deficient female liver. Analysis of liver RNAs from wild-type mice representing three mixed or outbred strains identified 1,028 sexually dimorphic genes across the strains, including 393 female-predominant genes, of which 89 (23%) required STAT5a for normal expression in female liver. These findings highlight the importance of STAT5a for regulation of sex-specific gene expression specifically in female liver, in striking contrast to STAT5b, whose major effects are restricted to male liver.

  • signal transducer and activator of transcription-5a
  • knockout
  • liver sexual dimorphism
  • growth hormone action
  • strain-dependent gene expression

growth hormone (GH) regulates gene expression in several tissues, most notably liver. GH is secreted by the pituitary gland in a sex-dependent manner under the regulation of gonadal steroids (15, 31). This, in turn, leads to substantial sex differences in GH-regulated liver gene expression (21, 37). In the rat, plasma GH levels are highly pulsatile in males, where hormone peaks every ∼3.5 h are followed by a GH-free interval lasting ∼2 h, whereas in females, GH is present in plasma in a more continuous manner (15). Mice also show sexually dimorphic GH secretory patterns, with females characterized by more frequent GH pulses and a distinctly shorter GH-free interpulse interval than males (20). These sex-dependent plasma GH profiles control liver gene expression at the level of transcription, as demonstrated for several GH-regulated genes that encode liver cytochrome P450 (Cyp) enzymes active in the metabolism of steroids, drugs, and environmental chemicals (34).

GH signaling is initiated by the binding of GH to its cell surface receptor, which activates Janus kinase (JAK)2, a GH receptor-associated tyrosine kinase. JAK2, in turn, phosphorylates GH receptor on multiple cytoplasmic domain tyrosine residues, several of which serve as docking sites for signal transducer and activator of transcription (STAT) transcription factors (6, 12), including STAT5a and STAT5b (7, 9, 25). Each STAT is phosphorylated on a single tyrosine residue, which leads to STAT homo- and heterodimer formation, translocation to the nucleus, binding to STAT5 DNA response elements, and stimulation of gene transcription. Both STAT5a and STAT5b are directly activated in male rat liver in response to each incoming plasma GH pulse (4), whereas in livers of female rats, the persistence of plasma GH stimulation leads to partial desensitization of the STAT5 signaling pathway and substantially lower nuclear STAT5b protein than the peak levels seen in males (3, 4, 35). On the basis of these and other findings, STAT5b has been proposed to serve as a mediator of the sex-dependent effects that GH has on liver gene expression (32). This proposal is supported by the characterization of STAT5b-deficient male mice, which display a reduced body growth rate at puberty and a loss of sex-specific liver expression of Cyps and other genes (11, 28, 30). The effects of STAT5b loss on liver gene expression, but not the effects on body growth rates, are recapitulated in a liver-specific STAT5a-STAT5b double-knockout model. Microarray analysis has further demonstrated that the impact of STAT5b on sex-dependent liver gene expression is global, with 90% of male-predominant genes downregulated and 61% of female-predominant genes upregulated in livers of STAT5b-deficient male mice (5). Studies of STAT5a-deficient mice (19) raise the possibility that STAT5a may be required for expression of certain female-specific mouse Cyp proteins (24); however, this has not been investigated in a systematic way.

Sexual dimorphism of liver gene expression is extensive, affecting at least 1,500 genes (5, 38), including many members of the Cyp gene superfamily (34). These sex differences may be impacted by genetic factors that are manifest as strain differences in sex-dependent gene expression. For example, certain Cyp2b steroid 16α-hydroxylase enzymes show strain-dependent differences that affect sex specificity (23). Moreover, Cyp2a4 is subject to strain-specific regulation associated with a recessive mutation that derepresses its expression in males (2). Strain-dependent expression also characterizes sex-specific genes regulated by regulators of sex limitation (Rsl), transcriptional repressors that are deficient in certain mouse strains and modulate the expression of Slp, Mup, and select other male-specific GH-regulated genes (17, 29). Mutations in Rsl1 and Rsl2 serve as examples of strain-specific alleles in transcriptional regulators that impact the sex specificity of GH-responsive genes in the liver (16, 33). It is uncertain whether these or other strain-dependent factors have a widespread effect on the sex specificity of hepatic gene expression.

The present study uses microarray technology to assess the impact of STAT5a deficiency on the sex specificity of liver gene expression. In addition, global assessment of sex specificity was investigated in three different mixed or outbred mouse strains to identify a robust set of strain-independent sex-specific genes, as well as sets of genes whose sex dependence apparently varies with the genetic background. Our findings reveal that Stat5a disruption affects the expression of a discrete subset of sex-dependent genes in female mouse liver, in sharp contrast to Stat5b disruption, which primarily abrogates sex-specific gene expression in male liver.



STAT5a-deficient mice were prepared by targeted disruption of the Stat5a gene (19). Livers were collected from 6- to 9-wk-old male and female 129J × Black/Swiss mice, wild-type (WT) mice, and STAT5a knockout (KO) mice. STAT5a KO livers were previously shown to be devoid of detectable STAT5a RNA (and protein), a result that was confirmed by a four- to sixfold downregulation of STAT5a RNA in the present microarray study (see Supplemental Table S2; supplemental data are available at the online version of this article). Moreover, hepatic levels of STAT5b and STAT5b DNA binding activity are unaffected by the loss of STAT5a (24). Livers were also collected from 8- to 10-wk-old male and female WT mice of the ICR strain (Taconic, Germantown, NY). Livers were snap frozen in liquid nitrogen and stored at −80°C until use. All animal studies were carried out using protocols approved by the Boston University Institutional Animal Care and Use Committee.

RNA isolation.

Total RNA was isolated from ∼0.1 g of frozen mouse liver using Trizol reagent (Invitrogen Life Technologies, Carlsbad, CA) according to the manufacturer's protocol. Liver RNAs prepared from 12 individual 129J × Black Swiss mice (30 μg of RNA per liver, dissolved in diethyl pyrocarbonate-treated water) were used in the present study: three WT males, three STAT5a KO males, three WT females, and three STAT5a KO females. Ten individual ICR mouse liver RNAs (5 WT female livers and 5 WT male livers), kindly provided by Dr. Ekaterina Laz of this laboratory, were used to prepare two pooled liver RNA samples (n = 2 livers and n = 3 livers, respectively). Two of the individual livers for each sex were also used to prepare single liver RNA samples.

Quantitative PCR analysis.

Liver RNA samples were converted to cDNA and used in quantitative PCR (qPCR) assays for individual genes from groups 13 and 14 (see Table 3) using methods detailed elsewhere (11). Amplification of a single specific product during qPCR cycling was verified by examination of dissociation curves of each amplicon. Relative RNA levels were determined after normalization to the 18S RNA content of each sample. Statistical analysis was carried out by Student's t-test using GraphPad Prism software version 4 (San Diego, CA). P values <0.05 were considered significant. qPCR primer design was carried out using Primer Express software (Applied Biosystems), and all primers were verified with respect to their specificity for the target transcript by BLAT [basic local alignment search tool (BLAST)-like alignment tool] analysis of the mouse genome (February 2006 assembly) at http://genome.ucsc.edu/cgi-bin/hgBlat. Primer sequences are shown in Supplemental Table S1.

Microarray analysis.

Global expression analysis was determined using the 23,574-feature mouse Rosetta/Merck Mouse TOE 75k Array 1 [Gene Expression Omnibus (GEO) Platform: GPL 3562; Agilent Technology, Palo Alto, CA], with each feature corresponding to a single 60-mer oligonucleotide antisense probe. Each probe is herein referred to as representing a distinct gene, although the actual number of genes analyzed is likely to be smaller than this number because of the presence of nonannotated sequences, some of which may duplicate results for other genes represented on the chip. Liver RNA samples (n = 3 independent biological replicates for each of 4 sex-genotype combinations) were used in four separate competitive hybridization experiments in a loop design: male (M) WT vs. female (F) WT (M-WT:F-WT), male WT vs. male STAT5a KO (M-WT:M-KO), female WT vs. female STAT5a KO (F-WT:F-KO), and male STAT5a KO vs. female STAT5a KO (M-KO:F-KO). Sample labeling, hybridization, and array scanning were performed as described (13). Briefly, total RNA was reverse transcribed using an oligo(dT) primer, followed by second-strand cDNA synthesis. Labeled cRNA was generated in a two-step process by derivatization of the transcribed products with either Cy3 or Cy5 dyes. For each microarray hybridization, a mixture of equal amounts of a Cy3-labeled and Cy5-labeled sample was used. Each sample pair was hybridized on two separate microarrays, with fluor reversal. All four comparison groups consisted of three biological replicates, giving a total of 24 microarrays for the STAT5a data set. Four fluor-reversed pairs, corresponding to the two pooled RNA and two single liver RNA samples, were hybridized for the WT male and WT female comparison of the ICR mice.

Microarrays were scanned, and individual feature intensities were preprocessed in a series of steps, consisting of background subtraction, normalization to mean intensities of the Cy3 and Cy5 channels, and detrending to fit a linear relationship between channels (8). Normalized intensities from fluor-reversed pairs of arrays were used to derive expression ratios using the Rosetta error model (8, 36). Probes with intensities resulting from oversaturation were excluded from downstream statistical analysis. The microarray platform and analytical methods used were highly sensitive, and were able to reliably detect gene expression fluorescent signal intensities over a >600-fold dynamic range (0.085–53.2 intensity units), with 18,204 of the 23,574 probes deemed to be reliably expressed in the mouse liver samples analyzed (signal intensities >5 times the whole chip background value for either channel). A similar result was obtained, based on the minimum signal intensities, that reliably allowed for discrimination of sex-specific genes. Thus, for each 129J × Black/Swiss M-WT:F-WT fluorescent reverse pair, the two average intensity values for each gene were compared by t-test to determine whether the two values were significantly different at P < 0.0001. For each channel, the lowest signal intensity at which gene expression was reliably demonstrable (the cut-off intensity value) is given by the minimum value of the greater of the two intensity values for those genes that met the threshold for significant differential expression. A total of 16,876 genes with intensities above the cut-off intensity value for either channel were thus found to be expressed in WT mouse livers of either sex.

Expression ratios obtained in this study are included in Supplemental Table S2. The data are also available for query or download from the GEO website at the National Center for Biotechnology Information (NCBI; http://www.ncbi.nlm.nih.gov/geo) as GEO series GSE7169 and GSE7170. GenBank accession numbers and associated gene names, gene descriptions, and Unigene numbers based on Unigene build 161 were obtained for unassigned features using the BLAT analysis tool located on the University of California Santa Cruz Genome Browser website. Assignments were based on the best matches (identity of ≥56 of 60 nucleotides) to the 60-nucleotide probe sequences of each probe. Probes that returned multiple accession numbers were verified to make sure that all accession numbers represented the same gene. Probe sequences are available on request.

Statistical analysis.

Moderated t-statistics, using standard errors moderated by a simple Bayesian model, were generated using the linear models for microarray data (LIMMA) package as part of the Bioconductor project within the R statistical program (26). This is the same P value used in our prior microarray study on the effect of STAT5b in mouse liver gene expression (5). A filter (P < 0.05) was applied to the P values to determine the statistical significance of each gene's differential expression for each of the four microarray comparisons (M-WT:F-WT, M-WT:M-KO, F-KO:F-WT, F-KO:M-KO). A fold-change filter of 1.5-fold was combined with the above P value filter to reduce the false discovery rate (FDR) to <6%, as follows. Of the 23,574 features represented on the array, 4,597 met the 1.5-fold expression filter for at least one of the four microarray comparisons. The number of genes expected to meet the combined threshold (P < 0.05 and >1.5-fold change in expression) by chance (type I errors) is 0.05 × 4,597, or 230 genes. The actual number of genes meeting the combined threshold was 3,905 genes (see results), corresponding to an FDR of 230/3,905, or 5.9%. Multiple testing correction methods, such as Bonferroni or Holm step-down, were not applied because these options depend heavily on the independence of each gene's expression and thus filter out many bona fide regulated genes (e.g., genes validated by qPCR) to avoid all type I errors; these methods are thus too restrictive in their effort to avoid false positives, as noted elsewhere (1). Fisher's exact test (http://www.exactoid.com/fisher/) was used to compare the effects of STAT5a deficiency on gene expression in males vs. females.

To compare sex specificity across strains, microarrays comparing WT male vs. WT female gene expression in three distinct genetic backgrounds were averaged (n = 10 arrays) to determine the robustness of sex specificity across all three strains and experiments. These three data sets were based on analysis using the same microarray platform for WT males and WT females from the present STAT5a KO study, from the prior study of STAT5b KO mice (5), and from a set of ICR mouse livers, as described above. A differential expression filter (mean ratio >1.5) was applied to the gene expression values deemed statistically significant by the P < 0.05 filter. Analysis of variance (ANOVA) with a Benjamini and Hochberg FDR of 0.001 was used, as implemented in GeneSpring GX 7.3.1 software (Agilent Technologies, Santa Clara, CA), to determine strain specificity for the genes that met the threshold criteria for only one of the three independent data sets for M-WT:F-WT comparison.

A system of binary and decimal flags was applied for clustering genes based on expression ratios obtained in all four microarrays, as previously described in our STAT5b KO microarray study (5). For the purpose of this clustering, threshold ratios for differential expression were reduced to values of >1.25 and <0.8 for the three arrays that involved STAT5a KO liver RNAs, with retention of the P < 0.05 threshold for statistical significance. Average ratios meeting these threshold and significance criteria contributed to the binary- and decimal-based flag. Thus genes with a M-WT:F-WT microarray ratio meeting the criteria were assigned a binary flag value of 1, while genes meeting the criteria for the M-WT:M-KO, F-KO:F-WT, and F-KO:M-KO microarrays were, respectively, assigned binary flag values of 2, 4, and 8. Genes not meeting these criteria were assigned flag values of 0. The sum of these binary-based flag values defines the whole number portion of the flag and was used as a simple method to identify which of the four microarrays met our criteria for inclusion for any given gene of interest, regardless of the direction (up or down) of the regulation. The flag value was then extended using decimal values of 0.1, 0.01, 0.001, and 0.0001 or 0.2, 0.02, 0.002, and 0.0002 for each of the four microarrays to indicate the direction of regulation between the two conditions on the microarray. Thus average ratios for the M-WT:F-WT microarray >1.5 were assigned a decimal value of 0.2 to indicate upregulation, while average ratios <0.66 were assigned a value of 0.1 to indicate downregulation. The three other microarray ratios were similarly flagged based on threshold values of >1.25 and <0.8, as indicated above, by advancing to a new decimal position for each microarray (i.e., the M-WT:M-KO flag is in the hundredths position, and so on). For each gene, the resulting binary sum describes which microarray ratios met the selection criteria, and the four-digit decimal value describes the direction of regulation (total flagging sum).

Gene Ontology term enrichment analysis.

Gene Ontology (GO) molecular function annotations were analyzed for term enrichment using the GO Browser within GeneSpring GX software. Because the GO term assigned to each gene represents a subclass of its parent class within the ontology, each gene was also assigned all parent GO class terms. For each of the GO categories present among each list of genes with GO annotations, the number of genes assigned that term was calculated along with a count of all genes on the microarray assigned that term. The P value for each category was calculated based on the total number of genes on the microarray within the category and the number of genes in the group being tested within the category. The P value represents the likelihood that at least as many genes would occur if a list of equal size were selected by chance from the total gene count.


Experimental design.

A large-scale expression study was conducted to investigate the role of STAT5a in the sex specificity of mouse liver gene expression. RNA was isolated from livers of male and female mice that were either WT or contained a targeted disruption of the Stat5a gene (KO). RNA samples representing each sex-genotype combination were analyzed in four sets of competitive hybridization to 23,574-feature oligonucleotide microarrays: 1) M-WT vs. F-WT, 2) M-WT vs. M-KO, 3) F-WT vs. F-KO, and 4) M-KO vs. F-KO. Normalized hybridization intensities were used to calculate mean expression ratios based on n = 3 biological replicates for each sex-genotype combination, and P values were calculated using the LIMMA tool set within the R statistical program (26). Probes representing 3,905 genes met the threshold criteria for differential expression (average expression ratio >1.5-fold and a significance level of P < 0.05) for at least one of the four sex-genotype combinations. Thus these 3,905 genes (21% of the 18,204 probes giving a liver expression measurement >5-fold background in at least 1 sex) were expressed in a sex-specific manner in either WT or STAT5a KO mice or responded to the loss of STAT5a in either males or females. Average expression ratios for the 3,905 genes of interest are listed in Supplemental Table S2.

Overview of sex specificity and impact of STAT5a deficiency.

Gene expression differences between WT male and WT female liver were found for 2,482 of the 3,905 genes of interest (64%), indicating that these genes are sexually dimorphic in WT liver. These genes are colored green (M-WT > F-WT) or red (M-WT < F-WT) in Fig. 1, lane 1, where they are displayed at the far ends of a false-color heat map containing all 3,905 genes sorted by average M-WT:F-WT ratio. Expression of 1,045 of the 2,482 genes was male predominant in WT mouse liver (M-WT:F-WT >1.5; Fig. 1, lane 1, top yellow box), while 1,437 genes were female predominant (M-WT:F-WT <0.667; Fig. 1, lane 1, bottom yellow box).

Fig. 1.

False-color heat maps for expression profiles of 3,905 differentially expressed genes from the signal transducer and activator of transcription (STAT)5a knockout (KO) data set and 2,231 differentially expressed genes from the STAT5b KO data set. Genes are depicted based on their mean expression ratios across 4 experimental pairings, as indicated at top (WT, wild type; M, male; F, female). Genes are colored according to the color bar at bottom, ranging from bright green for an expression ratio >4 to bright red for an expression ratio <0.25, with black corresponding to a ratio of 1. In lanes 1–4, 3,905 genes from the present STAT5a data set are sorted according to the mean M-WT:F-WT ratio. The top and bottom yellow boxes enclose genes determined to be male specific and female specific, respectively (M-WT:F-WT >1.5 or <0.66 at P < 0.05). In lanes 5–8, 2,231 differentially expressed genes from the STAT5b KO data set (5) are sorted according to the M-WT:F-WT ratio.

Hierarchical clustering of the four microarray data sets grouped the male-female comparisons for WT and KO mice together (Pearson correlation coefficient = 0.726). Thus genes found to be sex specific in WT mice generally retained sex specificity in the STAT5a KO mice. Additionally, no discernable overall correlation was found between sex specificity and the loss of STAT5a in either sex. For 930 of the 3,905 genes of interest (24%), the threshold criteria for significant change in expression was met when comparing WT and STAT5a KO mice of the same sex (Table 1). For 579 of these 930 genes (62%), the loss of STAT5a affected gene expression in female liver only, whereas for 289 of the 930 genes (31%), the loss of STAT5a affected gene expression in males only (Table 1). In male mouse liver, STAT5a deficiency had no effect on 93–95% of the sex-predominant genes. STAT5a deficiency also had no effect on 93% of male-specific genes in female liver (Table 2). By contrast, in female liver, 219 of 1,437 female genes (15%) were decreased in expression in the absence of STAT5a (Table 2). The impact of STAT5a deficiency on female-specific gene expression in female liver was significantly greater than the effects of STAT5a deficiency on either 1) male-predominant or female-predominant gene expression in male liver (P < 10−12; Fisher's exact test) or 2) male-predominant gene expression in female liver (P < 10−10; Fisher's exact test). The impact of STAT5a deficiency on these female-predominant genes is discussed further, below.

View this table:
Table 1.

Changes in gene expression in STAT5a KO mice compared with WT mice of the same sex

View this table:
Table 2.

Impact of STAT5a deficiency on expression of male-specific and female-specific gene expression

Clustering by significance and differential expression.

The 3,905 genes that met the threshold criteria for at least one of the four sex-genotype expression data sets were clustered into subgroups using a “flagging” system (5), whereby each gene is assigned to a specific category based on its expression ratio in each of the four sex-genotype combinations investigated (see materials and methods); 3,820 of the 3,905 genes were thus classified into 28 groups of coexpressed genes, as shown in Table 3. Table 3 also presents the distribution of gene counts in each of the corresponding categories determined earlier based on sex specificity and the loss of STAT5b (5). Groups comprised of <15 genes in both the present study and the previous study of STAT5b KO mice (5) are collected into a single category named “other.” Striking differences in these two profiles are apparent from the comparisons presented in Supplemental Fig. S1. Six of the seven largest gene groups identified in the present study, comprising 2,975 of the 3,905 genes (76%) (groups 4A, 4B, 5A, 5B, 7A, and 7B), did not show a significant change in expression in either male or female STAT5a KO mice (i.e., no regulation by STAT5a in either males or females; Table 3). The remaining 930 genes (24%) met the threshold criteria for a response to STAT5a deficiency in either males or females, as noted above. These latter genes include group 13B, the sixth largest gene group, which is comprised of 164 female-predominant genes that were downregulated with STAT5a deficiency in female but not male liver, leading to the loss of sex specificity in the STAT5a KO strain (Table 4). A corresponding group of 26 male-predominant genes (group 13A) was specifically upregulated in STAT5a-deficient female liver, leading to a loss of sex specificity.

View this table:
Table 3.

Distribution of the 3,905 from the STAT5a KO study and 2,231 differentially expressed genes from the STAT5b KO study within 28 flag-based coexpression groups

View this table:
Table 4.

Sex-specific genes that are up- or downregulated in STAT5a-deficient females but not males and are not sex-specific in the STAT5a KO strain

Female-specific genes that respond to STAT5a deficiency in females.

The loss of STAT5a in female liver decreased the sex specificity of 275 genes, with 219 female-predominant genes decreased and 56 male-predominant genes increased (Table 2). Of these 275 genes, 190 (69%) belong to groups 13A and 13B (Table 3), indicating a loss of sex specificity in the KO mice; 79 of the 275 genes (29%) comprise groups 14A and 14B, whose genes display the same pattern of response to STAT5a deficiency in female liver as groups 13A and 13B, albeit with partial retention of sex specificity in the STAT5a KO strain (Table 4 vs. Table 5). The quantitative relationships between M-WT:F-WT ratios and F-KO:F-WT ratios for the 190 genes in groups 13A and 13B and for the 79 genes in groups 14A and 14B are shown in Fig. 2, where mean log2 ratios for each comparison are plotted on a log-log scale. A linear relationship between the magnitude of sex specificity and the response to STAT5a deficiency in female liver is evident for the genes in groups 13A and 13B (Fig. 2A; slope = 0.815, y-intercept = 0.0547, r = 0.924), indicating that STAT5a plays a significant role in the sex specificity of these genes. A linear relationship between sex specificity and response to the loss of STAT5a in female liver was also seen for the genes in groups 14A and 14B (Fig. 2B; slope = 0.33, y-intercept = −0.007, r = 0.905), where the slope of 0.33 highlights the partial loss of sex specificity in the absence of STAT5a. These expression profiles were confirmed by qPCR analysis for select genes from groups 13A, 13B, 14A, and 14B, as shown in Fig. 3 and in Supplemental Table S3. A majority of the sex-specific genes responsive to the loss of STAT5a in female liver that are presently annotated with GO terms (194 of the 275 genes in this category) are associated with six primary molecular function GO categories: binding, catalytic activity, signal transduction, transporter, enzyme regulation, and transcription regulation (Table 6). Four subcategories of molecular function, signal transducer activity, receptor activity, sugar binding, and carbohydrate binding, were significantly enriched (P ≤ 0.0004) among the 194 genes and encompass 68 unique genes. Of note, groups 13A and 14A both include multiple members of the Serpina proteinase inhibitor family in the listing of top genes (Tables 4 and 5).

Fig. 2.

Scatterplots for quantitative correlation of M-WT:F-WT and F-KO:F-WT ratios for genes in groups 13 and 14. A: log2 ratios for M-WT:F-WT (x-axis) and the corresponding F-KO:F-WT data (y-axis) plotted on a log-log scale for all genes in groups 13A and 13B (Table 3; n = 190 genes). The 2 sets of ratios for the 2 experimental comparisons are highly correlated (r = 0.924), and the best-fit line (y = 0.815x − 0.0547), shown in solid, has a slope close to 1 and an intercept near 0. B: log2 ratios for M-WT:F-WT (x-axis) and the corresponding F-KO:F-WT data (y-axis) plotted on a log-log scale for all genes in groups 14A and 14B (Table 3; n = 79 genes). The 2 sets of ratios for the 2 experimental comparisons are highly correlated (r = 0.905), and the best-fit line (y = 0.33x + 0.007), shown in solid, has an intercept near 0 but a slope near 1/3. In both A and B, genes found in the top right quadrant (I) are male predominant and were increased in expression in female liver in the absence of STAT5a, while genes found in the bottom left quadrant (III) are female predominant and were decreased in expression in STAT5a-deficient female liver; 95% prediction boundaries are shown as dashed lines.

Fig. 3.

Quantitative PCR analysis of select genes in groups 13A, 13B, 14A, and 14B. RNA samples prepared from individual male and female livers, wild type (MWT and FWT, respectively) and STAT5a knockout (MKO and FKO, respectively), as indicated, were assayed for the indicated RNAs from each group as described in materials and methods. Significant differences are indicated as follows. For F-WT vs. M-WT, *P < 0.05 and **P < 0.01; for M-KO vs. M-WT and F-KO vs. F-WT, +P < 0.05 and ++P < 0.01. Results confirm the general trends in expression that are characteristic of each group (c.f., Tables 4 and 5).

View this table:
Table 5.

Sex-specific genes that are up- or downregulated in STAT5a-deficient females but not males and are sex specific in the STAT5a KO strain

View this table:
Table 6.

Distribution of molecular function-related GO categories and GO enrichment for 194 sex-specific genes that show a dependence on STAT5a in female mouse liver

Comparison of STAT5a deficiency with STAT5b deficiency.

No overall correlation between sex-specific gene expression and the effect of STAT5a deficiency was seen, either in males or in females, in contrast to our earlier finding with STAT5b-deficient mice (5). Qualitatively, when the regulated genes in the present STAT5a KO study were sorted by sex specificity, only the M-KO:F-KO comparison showed similarity to M-WT:F-WT in the false-color heat map (Fig. 1, lane 2 vs. lane 1), whereas an extensive similarity in the heat map was apparent for the M-WT:F-WT and M-WT:M-KO comparisons in the STAT5b KO study (Fig. 1, lane 8 vs. lane 5). The distribution of genes among the coexpressed gene groups also exhibited marked differences in the requirement of STAT5a and STAT5b for sex-specific gene expression (Table 3 and Supplemental Fig. S1). STAT5b was required for the expression in male liver of 1,205 sex-specific genes, or 75% of the 1,603 sex-specific genes identified in the STAT5b study (5) (genes comprising STAT5b KO study groups 1A, 1B, 2A, 2B, 3A, 3B, 6A, and 6B). By contrast, the corresponding eight groups comprised only ∼4% of the 2,482 sex-specific genes in the present STAT5a study (Table 3). Six of the seven largest groups of coexpressed genes identified in the present study are made up of sex-specific genes that did not respond to the loss of STAT5a in either males or females (groups 4A, 4B, 5A, 5B, 7A, and 7B). Together, these six groups comprise 76% of the genes of interest in the STAT5a study vs. only 23% of the corresponding set of genes in the STAT5b study. Major differences in the impact of STAT5a vs. STAT5b deficiency are thus apparent.

Comparison of sex specificity among three mouse strains.

M-WT:F-WT expression ratios, obtained in the present analysis of 129J × Black Swiss mouse livers, were compared with the corresponding ratios determined using the same microarray platform for two sets of male and female mouse livers from different genetic backgrounds: 129J × BALB/c mice [genetic background used in the STAT5b KO mouse study (5)] and ICR mice, an outbred strain. Normalized M-WT:F-WT expression ratios for a total of 10 microarrays (3 from 129J × Black Swiss mice, 3 from 129J × BALB/c mice, and 4 from ICR mice) were combined to obtain a single data set, of which 1,028 genes (635 male predominant, 393 female predominant) met the threshold of >1.5-fold differential expression at P < 0.001 (Supplemental Table S4). These 1,028 genes include 109 of the 275 of the genes (40%) whose sex specificity decreased in female liver in the absence of STAT5a (c.f., Table 2). Furthermore, 523 of the 1,028 genes (370 male predominant, 153 female predominant) met the threshold of >1.5-fold differential expression and P < 0.05 in all three independent data sets, indicating robust sex-specific expression across strains and experiments (Supplemental Fig. S2 and Supplemental Table S5).

Further examination of the three microarray data sets revealed genes that met the criteria for sex specificity in only one of the data sets (Supplemental Fig. S2). ANOVA analysis identified 54 genes showing sex specificity in 129J × BALB/c mice but not in the other two strains at P < 0.001 (Supplemental Table S6). Similarly, 49 genes were identified as sex specific in ICR mice only, while 551 genes were sex specific in 129J × Black Swiss mice only. Top candidates for these strain-dependent, sex-specific genes are presented in Supplemental Table S6.


The present study investigates the role of STAT5a in the sexual dimorphism of mouse liver gene expression, which was previously shown to be extensive and highly dependent on STAT5b, the most abundant liver STAT5 form. Whereas a majority of the sex-specific hepatic genes identified here were unaffected by the loss of STAT5a, expression of a unique subset of sex-specific genes was specifically altered in STAT5a-deficient female mouse liver. These include four Cyp genes, Cyp2a12, Cyp2d22, Cyp3a16, and Cyp8b1, each of which responds to the loss of STAT5a in female but not male liver (gene groups 13A and 14B; Table 3). Overall, 219 genes, corresponding to 15% of female-predominant genes, were decreased in expression, whereas 5% of male-predominant genes were increased in expression in livers of female but not male mice deficient in STAT5a. The portion of female-predominant genes that required STAT5a for expression increased to 23% (89 genes) when a strain-independent set of 393 female-specific genes was considered. Thus STAT5a is likely to exhibit transcriptional effects in female mouse liver that are not compensated for by STAT5b. This finding contrasts with the extensive dependence of sex-specific gene expression on STAT5b seen in male liver, where nearly 90% of all sex specificity is lost or significantly reduced in the absence of STAT5b (5). This finding is highlighted by the markedly different distribution of sex-specific genes among the major STAT5a- and STAT5b-coregulated gene groups (Supplemental Fig. S1). The precise physiological significance of this specific role of STAT5a in female liver is difficult to ascertain. Of note, however, is that four molecular functions as annotated by GO were overrepresented in the set of STAT5a-dependent, sex-specific genes in female liver: signal transduction, receptors, carbohydrate binding, and sugar binding (P < 0.0001).

Earlier studies suggested a role for STAT5a in the expression of a female-specific Cyp2b family member, based on the loss of a female-specific Cyp2b protein in STAT5a-deficient female mouse liver (24). In that case, however, the Cyp2b protein was also decreased in STAT5b-deficient females, a pattern of regulation distinct from that of the major group of STAT5a-dependent female genes described here. The role of STAT5a as a feminizing factor in female mouse liver (positive regulation of female genes and negative regulation of male genes) is analogous to that of STAT5b in male liver, where STAT5b serves as a masculinizing factor (positive regulation of male genes and negative regulation of female genes). STAT5a but not STAT5b also plays the major role in another female-specific function, mammary gland differentiation and lactogenesis in response to prolactin stimulation (19).

The present study includes an analysis of sex-specific mouse liver genes across three mixed or outbred mouse strains. Combined analysis of the data from all three strains (10 microarrays) taken as a single data set identified 1,028 genes showing sex-dependent expression at P < 0.001 across all three studies (Supplemental Table S4). These include liver-expressed Y-linked genes such as Ddx3y, Eif2s3y, and Jarid1d as well as 26 Cyp genes and 10 Sult (sulfotransferase) genes. A subset comprised of 523 genes met the threshold for sex specificity in all three mouse strain backgrounds and thus corresponds to a robust core of sex-specific hepatic genes. Forty-seven genes annotated as transcriptional regulators by GO were included among the robust sex-specific genes, including cut-like 2 (Cutl2), SRY-box containing gene 15 (Sox15), and myogenic factors 5 and 6 (Myf5, Myf6), a subset of which could potentially mediate the sex-specific actions of GH-activated STAT5a or STAT5b, as discussed elsewhere (34). Cutl2, in particular, has recently been characterized as a highly female-specific, continuous GH-regulated nuclear factor in both mouse and rat liver (F > M ∼100) that could contribute to the establishment or enforcement of liver sex specificity (18).

Also identified here are genes whose liver sex specificity was apparent in only one of the three mouse genetic backgrounds investigated. Some of these genes may be subject to strain-dependent genetic regulation. Genes exhibiting significant discrepancies in expression across the three strain backgrounds are listed in Supplemental Table S6. Only a subset of these genes is likely to exhibit true strain-dependent sex specificity. Technical factors inherent in microarray analysis, as well as the fact that individual livers, rather than pools of livers, comprised the biological triplicates for one of the microarray studies, could contribute to the apparent strain-dependent loss of sex specificity for some of these genes. Nevertheless, these findings raise the possibility that the strain dependence of mouse liver sex specificity may be more extensive than was previously recognized. Earlier studies identified Rsl alleles that conferred strain-specific regulation of certain sex-specific genes, notably Slp, Cyp2d9, and several Mup family members (17). In other studies, strain-specific regulation of the female-predominant Cyp2a4 was associated with a recessive mutation at the GH-dependent repression (Gdr) locus that abolishes repression of Cyp2a4 expression in males (2). Further investigation using larger numbers of arrays representing several pure outbred and inbred mouse strains, including strains containing the Rsl and Gdr mutations, will be necessary to isolate effects due to strain differences from those due to individual variability that is strain independent, as well as from technical limitations of microarray technology.

It is apparent that STAT5a plays a unique role in sexually dimorphic gene expression in female mouse liver. Presumably, these actions of STAT5a are GH regulated, insofar as the vast majority of liver sexual dimorphism, including at least some of the STAT5a-dependent genes presently identified, is dictated by plasma GH profiles (34). It is not known, however, whether the unique actions of STAT5a in female liver reported here reflect a direct action of STAT5a on the genes in groups 13 and 14 (Table 3), or whether the requirement for STAT5a is indirect. It is also unclear why STAT5b is unable to compensate for the loss of STAT5a, given its 10- to 20-fold higher expression in liver (24) and the very close similarities in biochemical and gene regulatory activities of these two closely related (>90% identical) STAT5 proteins (7, 14). Conceivably, the unique actions of STAT5a could reflect a requirement for tetrameric STAT5a complexes for trans-activation via paired STAT5 response elements, insofar as STAT5a, but not STAT5b, readily forms tetrameric STAT-DNA complexes (27). Formation of DNA-bound STAT5a tetramers could be less efficient in male than in female liver owing to competition for DNA binding by the much more abundant dimeric STAT5b complexes that form in response to each incoming plasma GH pulse (3, 4). Further study will be required to investigate these and other possible mechanisms.

STAT5b has been proposed to serve as an upstream transcriptional regulator of a much larger downstream transcriptional network that leads to activation of many of the male-predominant genes and repression of the many female-predominant genes whose hepatic expression is dependent on STAT5b (34). Conceivably, STAT5a could play such a role, albeit on a much smaller scale, in female liver. Of note, 11 of the sex-specific genes dependent on STAT5a in female liver were annotated as transcriptional regulators by GO; these include inhibitor of DNA binding 2 (Id2) and interferon regulatory factor 2 (Irf2). The apparent STAT5a-dependent repression of male-predominant genes in female liver might not be a direct function of STAT5a. For example, an inhibitory nuclear factor, such as Id2, could serve as a direct target of STAT5a and contribute to STAT5a inhibition of select male-specific genes in female liver. Interestingly, Id2 and STAT5a show cooperative activity in development and maturation of mouse mammary tissue, where the loss of Id2 decreased active tyrosine-phosphorylated STAT5a levels by 40% (22). A similar pathway, involving STAT5a and other transcriptional regulators, could be responsible for the limited number of genes that exhibit STAT5a dependence in female liver. Further studies will be required to elucidate the role of STAT5a in female gene expression in liver and other tissues, including mammary gland, where STAT5a is the major STAT5 form.


This work was supported in part by National Institutes of Health (NIH) Grant DK-33765 (to D. J. Waxman). K. H. Clodfelter and G. D. Miles received training core support from the Superfund Basic Research Center at Boston University, NIH Grant 5-P42-ES07381, and V. Wauthier received fellowship support from the Belgian American Educational Foundation.


  • Address for reprint requests and other correspondence: D. J. Waxman, Dept. of Biology, Boston Univ., 5 Cummington St., Boston, MA 02215 (e-mail: djw{at}bu.edu).

    Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).


View Abstract