Elucidating the Role of Gonadal Hormones in Sexually Dimorphic Gene Coexpression Networks

Abstract

We previously used high-density expression arrays to interrogate a genetic cross between strains C3H/HeJ and C57BL/6J and observed thousands of differences in gene expression between sexes. We now report analyses of the molecular basis of these sex differences and of the effects of sex on gene expression networks. We analyzed liver gene expression of hormone-treated gonadectomized mice as well as XX male and XY female mice. Differences in gene expression resulted in large part from acute effects of gonadal hormones acting in adulthood, and the effects of sex chromosomes, apart from hormones, were modest. We also determined whether there are sex differences in the organization of gene expression networks in adipose, liver, skeletal muscle, and brain tissue. Although coexpression networks of highly correlated genes were largely conserved between sexes, some exhibited striking sex dependence. We observed strong body fat and lipid correlations with sex-specific modules in adipose and liver as well as a sexually dimorphic network enriched for genes affected by gonadal hormones. Finally, our analyses identified chromosomal loci regulating sexually dimorphic networks. This study indicates that gonadal hormones play a strong role in sex differences in gene expression. In addition, it results in the identification of sex-specific gene coexpression networks related to genetic and metabolic traits.

Higher order gene interactions differ between sexes and are determined primarily by sex hormones.

Sex differences are observed in most complex genetic disorders, including atherosclerosis, obesity, diabetes, and cancer (1). Males are more susceptible to cardiovascular disease, as are females to higher fat depots and insulin resistance; therefore, sex differences may have a substantial impact on the prevention, diagnosis, and treatment of disease (2,3,4). Sexual differentiation is manifested through various biological factors, stemming from the organizational and activational effects of gonadal hormones to the direct effects of sex-specific genes on the sex chromosomes, such as the testis determining factor Sry (5,6,7). For example, organizational sex differences in brain, liver, and other tissues are induced during development when the testes secrete testosterone, which acts to permanently masculinize the structure and function of tissues.

Later in life, differential exposure of testicular or ovarian hormone secretions results in the reversible, activational sex-specific acute effects of gonadal hormones that disappear after gonadectomy in adults. The organizational and activational effects of testosterone may be mediated by two primary metabolites of testosterone: nonaromatized metabolites such as dihydrotestosterone (DHT), which bind to androgen receptors, and aromatized metabolites such as estradiol (E2), which bind to estrogen receptors. The sex chromosome effects stem from the direct action of the Y-chromosome genes or by the differential dose of X-chromosome genes, resulting from differences in the genomic dose of the gene or the sex difference in the parental genomic imprint of the X-chromosome genes (8).

Sex hormones such as androgens and estrogens cause brain sex differentiation in mammals and affect tissues such as adipose, liver, and skeletal muscle (5,9). For instance, the regulation of body fat distribution is influenced by changes in gonadal hormones, including the adipose-derived hormone leptin (10). Regional sex differences may be required for adipose metabolism that supports the reproduction process, which is dependent on the location of the fat depot (11).

In the liver, metabolism of steroid hormones and drugs has been shown to be sexually differentiated, mainly due to gonadal hormones causing differences in GH. Studies have indicated that sex differences in hepatic steroid metabolism may support a pregnant state when the liver is exposed to high continuous levels of steroid hormones (12). Thus, tissue-specific sex differences may underlie sex-specific roles in reproduction, whereas changes in gonadal hormones contribute to the development of obesity and other related metabolic disorders (10).

The profound effect of sex on metabolic traits has also been observed in chromosomal linkage to these complex biological systems; for example, quantitative trait loci (QTL) have often been observed in one sex but not the other (13,14,15,16,17). These sex-gene interactions imply an underlying genetic network invoked by sex-specific regulation influencing gene expression (13). Indeed, sex differences in the expression of thousands of genes across several tissues has recently been documented; however, the molecular mechanisms driving this large-scale sexual dimorphism of gene expression and whether or not these sex differences are also reflected in their gene interactions remain to be elucidated (18).

In a previous study, the molecular mechanisms that lead to sex differences in the liver tissue were determined by analyzing the effects of the GH-activated transcription factor signal transducer and activator of transcription-5b (STAT5b) on sex-specific gene expression (19). Here, we investigated the activational effects of gonadal hormones and the effect of sex-chromosomes on liver gene expression. We determined the contribution of gonadal hormones and sex chromosomes in driving sexually dimorphic gene expression by comparing gene expression of liver tissue from gonadally intact mice and gonadectomized (GDX) mice that were treated with DHT, E2, or placebo. We also studied GDX four-core genotype (FCG) mice, in which the sex chromosomes are independent of gonadal sex (8). The genes whose expression was affected by hormone treatment were significantly enriched for the sex-specific differentially expressed genes from the previous study by Yang et al. (18). Relative to the gonadal hormones, the sex chromosomes had a minor influence on gene expression changes in the adult mice.

Previous studies of sex differences in gene expression focused mainly on differences in the level of transcript abundance (15). In this study, we asked whether the organization of gene expression networks differ between the sexes. By integrating genetics and gene expression, we sought to examine the genetic architecture of the complex biological systems between male and female. Gene coexpression networks, where nodes and edges reflect gene expression correlation, have been successfully applied to gene expression profiles in many species by investigating the functional relationship of gene coexpression networks in the context of complex biological traits (20,21,22,23,24,25,26,27). We constructed gene coexpression networks in adipose, brain, liver, and muscle of male and female mice from a segregating population of a cross between mouse strains C3H/HeJ and C57BL/6J on a hyperlipidemic apolipoprotein E-null background (BXH · ApoE^−/−). We identified sexually dimorphic modules of these networks corresponding to highly correlated gene expression, in adipose, brain, liver, and muscle as well as a sex-specific network enriched for genes affected by gonadal hormones. In addition, we detected sex-specific modules in adipose and liver correlated with biological traits and the chromosomal regions regulating these networks.

To our knowledge, this is the first microarray-based global study that begins to address how gene network interactions differ between the sexes and how these modules correlate with sexually dimorphic traits as well as sex-specific patterns of genetic linkage. Integrating genomic, gene expression, and biological trait data has allowed us to detect genes and genetic networks influenced by sex that may, in part, explain the differential susceptibility toward diseases between males and females.

Materials and Methods

Mice inbred strains C57BL/6J ApoE^−/− (B6 · ApoE^−/−) and C3H/HeJ ApoE^−/− (C3H · ApoE^−/−) F2 intercross data set

The genotype and gene expression data from adipose, brain, liver, and muscle of the F2 mouse population used in this study have been previously described (17,18). The F2 mouse population consisting of 334 mice (169 female, 165 male) was generated by intercrossing F1 mice of parental strains B6 · ApoE^−/− and C3H · ApoE^−/−. The knockout of the ApoE allele results in hypercholesterolemia, making the mice susceptible to atherosclerosis when they are fed a high-fat diet. Mice were fed chow diet containing 4% fat until 8 wk of age, when they were placed on a Western diet containing 42% fat and 0.15% cholesterol for 16 wk. At 24 wk of age, mice were fasted for 4 h in the morning, then anesthetized with isoflurane for retroorbital sinus blood collection, and then euthanized for collection of adipose, brain, liver, and muscle. The mice were genotyped at 1032 single-nucleotide polymorphisms uniformly distributed over the mouse genome at an average density of 1.5 cM.

Blood sample assays

Plasma glucose, free fatty acids, triglycerides, high-density lipoprotein cholesterol, unesterified and total cholesterol were measured as previously described (17,28). Very-low-density lipoprotein and low-density lipoprotein cholesterol levels were calculated by subtracting high-density lipoprotein cholesterol from total cholesterol levels. Plasma glucose concentrations were measured using a glucose kit (315–100; Sigma Chemical Co., St. Louis, MO). Plasma leptin, adiponectin, and monocyte chemotactic protein-1 levels were measured using mouse ELISA kits (MOB00, MRP300, and MJE00; R&D Systems, Minneapolis, MN).

Gene expression analysis

The RNA and microarray processing were as previously described (29). The custom inkjet microarrays (Agilent Technologies, Palo Alto, CA) contained 2186 control probes and 23,574 noncontrol oligonucleotides extracted from mouse Unigene clusters and combined with RefSeq sequences and RIKEN full-length clones. The number of distinct transcripts and gene models represented on the array were 23,574 and 21,859, respectively. Gene models were created based on clustering of transcripts on the genome. The 18,833 gene models were annotated with a LocusLink identifier, and each gene model was assigned to only one LocusLink identifier. Total RNA was extracted from homogenized mouse tissues with Trizol reagent (Invitrogen, Carlsbad, CA) according to the manufacturer’s protocol. cDNA labeled with either Cy3 or Cy5 was hybridized to at least two microarray slides with fluor reversal and subsequently scanned using a laser confocal scanner. Gene expression changes between two samples were quantified on the basis of spot intensity relative to background, adjusted for experimental variation between arrays using average intensity over multiple channels, and fit to an error model to determine significance (type I error). Gene expression is reported as the ratio of the mean log₁₀ intensity (ml-ratio) relative to the pool derived from 150 mice randomly selected from the F2 population. Full gene and probe information of the array is shown on supplemental Table 2S (published as supplemental data on The Endocrine Society’s Journals Online web site at http://endo.endojournals.org). The microarray data from this study have been deposited to GEO under accession nos. GSE13264 and GSE13265.

Gonadectomy and hormone treatment of mice

Mice of strain C57BL6/J were purchased from The Jackson Laboratory (Bar Harbor, ME) and were placed on a chow diet. The animals were GDX at 8 wk of life, implanted with hormone pellets at 12 wk, and killed at 14 wk. Male and female mice of the hormone-treated groups received sc implants of either 0.5-mg E2 pellet (plasma yield of 300 pg/ml) or a 5-mg DHT pellet (plasma yield of 1–2 ng/ml), designed to release over 21 d (blood levels reported by manufacturer, Innovative Research of America, Sarasota, FL; 17B-E2, catalog item no. E-121, 0.5 mg/pellet; 5α-DHT, catalog item no. A-16). Control mice were treated with the placebo pellet (Innovative Research of America; placebo, catalog item no. C-111) (30). Plasma, whole brain, heart, kidney, skeletal muscle (hamstring), visceral fat, and liver were harvested from the animals. Liver mRNA from five mice was profiled from each group.

Four core genotype (FCG) mice

In mice of the FCG, the Y chromosome is deleted for the testis-determining gene Sry, producing the Y⁻ chromosome (31). The Sry transgene is inserted onto an autosome, so that testis determination is independent of the complement of sex chromosomes. XY⁻Sry gonadal males are bred with XX gonadal females, producing progeny with four different genotypes: two types of gonadal males (XXSry and XY⁻Sry) and two types of gonadal females (XX and XY⁻). The FCG model allows comparison of gonadal males and females independent of their sex chromosome complement, or XX and XY⁻ independent of their gonadal type. The FCG mice for this study were C57BL/6J, produced by backcrossing the Y⁻ chromosome and Sry transgene from MF₁ to C57BL/6J for 10–11 generations. The Y⁻ chromosome derives from strain 129.

C57BL6/J mice (n = 5) from each of the above groups were fed normal chow diet. At 8 wk, the mice were GDX (to eliminate the effect of the gonadal hormones) and killed at 12 wk. Liver mRNA of each of these groups (n = 5 per group) were profiled.

Analysis of differential expression

Gene expression levels of pairs of treatment groups were compared using a one-way ANOVA. To control for false discovery, we used the Q-value software and an estimated 10% false discovery rate (FDR) threshold for all ANOVA results. Genes that passed an FDR of 10% were considered as being differentially expressed between the two groups compared (32,33).

Power calculations to detect gene expression differences between males and females

For each gene that showed differential expression between males and females in the F2 population of the BXH · ApoE^−/− cross, we wanted to compute the number of animals that would be required to identify differential expression in an independent study. The formula for computing n, the number of mice needed to observe differential expression of a gene between any two groups is given by n = (Z_α/2 + Z_β) × (σ_g1+ σ_g2)²/τ², where α and β are the probabilities associated with type I and type II errors, g1 and g2 refer to the two groups (in this case, male and female F2 samples in the BXH · ApoE^−/− cross), σ refers to sd of gene expression in each group, and τ refers to the difference in the means of gene expression values between the two groups. For each of the genes that was differentially expressed between the males and females of the F2 samples in the BXH · ApoE^−/− cross with ANOVA P value of ≤ 0.01, we computed n, taking α to be 0.05 and β to be 0.2, where power = 1 − β = 0.8. Of the 12,789 genes differentially expressed between males and females of the BXH · ApoE^−/− F2 animals, n was found to be ≤5 (study sample size) for 70 genes, indicating that our current study was underpowered to detect many of the sexually dimorphic gene expression patterns observed in the earlier cross where the sample sizes were much larger (34).

Construction of weighted gene coexpression networks and modules

Four thousand probe sets were selected for network analysis based on high variance across the BXH · ApoE^−/− F2 mouse population. Gene expression data from mice with complete genotype data and at least 95% complete phenotype and array data were used.

We used the general framework of weighted gene coexpression network analysis presented in the study by Zhang and Horvath (20). The absolute value of the Pearson correlation coefficient was calculated for all pairwise comparisons of gene expression values across all microarray samples. The Pearson correlation matrix was transformed into an adjacency matrix A, a matrix of connection strengths by using a power function (β). Microarray data can be noisy, and the number of samples is often small, so we weight the Pearson correlations by taking their absolute value and raising them to the power β. Thus, the connection strength (adjacency) a(i,j) between gene expressions x_i and x_j is defined as a(i,j) = [cor(x_i,x_j)]^β.

The network connectivity k_i of the ith gene expression profile x_i is the sum of the connection strengths with all other genes in the network. The network connectivity k_i represents a measure of how correlated the ith gene is with all the other genes in the network.

The resulting weighted network represents an improvement over unweighted networks based on dichotomizing the correlation matrix because 1) the continuous nature of the gene coexpression information is preserved and 2) the results of weighted network analyses are highly robust with respect to the choice of the parameter β, whereas unweighted networks display sensitivity to the choice of the cutoff.

To determine the power β used in the definition of the network adjacency matrix, we made use of the fact that gene expression networks, like virtually all types of biological networks, have been found to exhibit an approximate scale-free topology (20,21,22,23,24,25,26,27). To choose a particular power β, we used the scale-free topology criterion described in Zhang and Horvath (20), which led us to the following choices for the power β: for female and male in adipose, 6 and 6; brain, 5 and 5; liver, 3.3 and 4; and muscle, 11 and 8, respectively.

The next step in network construction is to identify modules, which were defined as groups of genes with high topological overlap (20,21,22,23,24,25,26,27). The use of topological overlap serves as a filter to exclude spurious or isolated connections during network construction. Using the topological overlap dissimilarity measure (1 − topological overlap) in average linkage hierarchical clustering, the modules were defined as branches of the resulting dendrogram.

Analysis of coexpression network correlation to metabolic and genetic traits

To determine coexpression networks trait correlations, the module eigengene was correlated to metabolic traits, and the module gene significance was determined. The module eigengene is the first principal component (pc1) of gene expression of each module and explained most of the variance in gene expression of each module in this study (25,26). The analysis was performed on the same set of genes that clustered together in both males and females. Gene significance was calculated as the average absolute value of the correlation between gene expression profile of each module and clinical trait.

Two modules in adipose and liver tissue were correlated with metabolic traits. The pc1 of the adipose blue module explained 44% of the gene expression variance in females and 51% in males, and the pc1 of the liver green module explained 39% of the variance in females and 47% in males. The threshold of significance of correlation between the pc1 of the module and metabolic trait was 0.30 at a P value of 0.05 for any module, given n = 135 after correcting for multiple comparisons.

Expression QTL (eQTL) were determined by using the transcript abundance of each gene as a quantitative trait and correlating them to genetic markers to map gene eQTL (25).

Coexpression network t-statistic scores

We used the software Significance Analysis of Microarrays to identify differentially expressed genes (35). The modified t test score was chosen such that the resulting gene list had a median 5% FDR. Specifically, we used the following thresholds for determining differentially expressed genes between males and females: brain, 1.16; liver, 0.76; adipose, 0.58; and muscle, 0.96.

Gene enrichment analysis using DAVID software

Each network module was analyzed separately for pathway enrichment by the use of the Expression Analysis Systemic Explorer (EASE) software, which was downloaded from the Database for Annotation, Visualization, and Integrated Discovery (DAVID) website (36,37). Fisher’s exact test was used to determine whether enrichment in pathway genes within the module is significant; we report Fisher exact P values and the multiple-testing corrected P values (Bonferroni).

Differential network analysis

Differential expression and connectivity was constructed using the method described in Fuller et al. (26). Briefly, the whole-network connectivity in networks 1 and 2 is denoted by for the ith gene, k₁(i) and k₂(i), respectively. To compare the connectivity measures of each network, we divide the connectivity of each gene by the maximum network connectivity: K₁(i) = k₁(i)/max k₁ and K₂(i) = k₂(i)/max k₂.

Next we define a measure of differential connectivity as Diff K(i) = K₁(i) − K₂(i), but other measures of differential connectivity could also be considered. A permutation test was performed to determine statistical significance of observing genes with k > 0.2 differential connectivity by permuting the sex assignment and repeating the permutation process 1000 times (number of permutations = 1000).

Gene coexpression network software

The statistical analysis was carried out with the R software (http://www.R-project.org). The statistical code for generating the weighted gene coexpression network, which also serves as a tutorial, can be found at http://www.genetics.ucla.edu/labs/horvath/CoexpressionNetwork/ (20).

Visualization of network module

The method by Oldham et al. (23) was used to identify which pairs of genes are connected within the network structure. The network structure of approximately 66 genes found on supplemental Table 8S was determined by the pairwise topological overlap value and depicted by VisANT integrative visual analysis tool software (38).

Results

Molecular mechanisms of sexually dimorphic gene expression

We investigated the role of gonadal hormones and sex chromosome complement, XX vs. XY, to determine the molecular mechanisms underlying the profound sexual differences in gene expression observed in our previous study (18). We analyzed the liver gene expression of GDX mice treated with E2, DHT, or placebo and the FCG mice to evaluate the effects of the sex chromosomes in the absence of gonads (Table 1).

Table 1.

Mice treatment groups to determine the molecular basis of sexually dimorphic gene expression

Mice	Treatment
Intact female	None
Intact male	None
GDX female or male	Placebo
	E2
	DHT
FCG Mice	Sex chromosomes
GDX Female	XX
	XY (Sry gene removed from Y chromosome)
GDX Male	XY
	XX (Sry gene on autosome)

First, we determined the degree of overlap between the sex-specific differentially expressed genes of this study and our previous study of the BXH · ApoE^−/− F2 mice cross (18). We detected 2597 genes differentially expressed between gonadally intact males and females compared with our previous study where we observed 17,834 differentially expressed genes between males and females in the liver tissue of the F2 BXH · ApoE^−/− F2 mice cross (supplemental Table 1S) (18).

The difference in gene number is most likely due to the difference in statistical power of the two studies. In this study, we used a sample size of five in each treatment group, in contrast to the sample size of 165 males and 169 females in our previous report. Given the mean and sd of gene expression and sample size from males and females in each study, we calculated that the current study would be powered to detect sex differences in 2834 genes. Indeed, the current study detected 91% of that number, or 2597 genes, which overlapped significantly with the 17,384 differentially expressed genes of the BXH · ApoE^−/− F2 animals (P = 4 × 10⁻¹³⁶) (supplemental Table 1S). The strong overlap of differentially expressed genes between the two studies indicates that the sexual dimorphism in liver gene expression of the F2 mice lacking ApoE in the previous study and in the parental mice with a functional ApoE is comparable.

Effects of gonadectomy on sexually dimorphic gene expression

To study the hormonal effects, we analyzed the liver gene expression of the GDX mice treated with DHT, E2, and placebo. Between the GDX males and females treated with placebo, we detected 12 differentially expressed genes (supplemental Table 1S). This gene number was considerably fewer than the 2597 genes that were differentially expressed between gonadally intact males and females, indicating that the activational effects of gonadal secretions account for most of the sex differences found in gonadally intact mice, and very few expression differences remain in the livers of adult mice when gonadal hormones are absent. The few genes whose expression levels are different between the sexes after gonadectomy may be due to the long-lasting developmental effects of the hormones or the sex chromosomes. Moreover, the near absence of sex differences after gonadectomy does not rule out finding differences due to organizational or sex chromosome effects when those variables are manipulated directly.

In general, genes that were differentially expressed between gonadally intact males and females, between GDX and gonadally intact mice, or between GDX mice treated with placebo or gonadal hormones overlapped significantly with the genes that were sexually dimorphic in the BXH · ApoE^−/− F2 animals (supplemental Table 1S). The strong overlap between the studies reflect the significant role the gonadal hormones play in driving the sexual differences in liver gene expression in the adult, gonadally intact, mice.

Effects of gonadal hormones on sexually dimorphic gene expression in males

To further investigate the effects of gonadal hormones on gene expression in males, we identified genes that were differentially expressed between GDX males treated with placebo and gonadal hormones. Specifically, by comparing the expression changes in GDX males brought about by treatment of either DHT or E2 relative to placebo, we examined whether these treatments reverse the gene expression changes brought about by gonadectomy. There were 4297 genes differentially expressed between intact males and GDX males treated with placebo, 1248 genes differentially expressed between intact males and GDX males treated with DHT, and 1825 genes differentially expressed between intact males and GDX males treated with E2. There was strong overlap of these sets of differentially expressed genes with the sexually dimorphic genes from the BXH · ApoE^−/− F2 mice (P = 10⁻²³ to 10⁻¹⁷³) (supplemental Table 1S).

There were 592 genes differentially expressed between GDX males treated with placebo and GDX males treated with DHT, whereas 64 differentially expressed genes were detected between GDX males treated with placebo and GDX males treated with E2. Of the 4297 genes that are differentially expressed between intact males and GDX males with placebo, 359 overlap with this set of 633 genes that are differentially expressed in GDX males upon DHT or E2 treatment (P = 10⁻⁹⁶).

Effects of gonadal hormones on sexually dimorphic gene expression in females

We identified genes that were differentially expressed between GDX females treated with placebo and gonadal hormones. There were no significantly differentially expressed genes detected between intact females and GDX females with placebo at a 10% FDR. Although this observation is surprising, this could be due to the heterogeneity of gene expression patterns in the intact females as a result of cycling estrogen and progesterone levels. Based on an expression hierarchical clustering of all the animals profiled in this study, the intact females appear more heterogeneous relative to intact males (data not shown). Alternatively, the level of estradiol in the intact females may have been relatively low.

There were 30 differentially expressed genes detected between intact females and GDX females treated with E2, and 1858 were differentially expressed between intact females and GDX females treated with DHT. We observed nine genes that were differentially expressed between GDX females treated with placebo vs. E2, whereas 1392 differentially expressed genes between GDX females treated with placebo vs. DHT, which overlapped significantly with the dimorphic genes in the BXH · ApoE^−/− F2 animals (P = 4.4 × 10⁻²⁷).

Comparison of the effects of DHT and E2 on sexually dimorphic gene expression

Gene expression differences between GDX mice treated with placebo and GDX mice treated with gonadal hormones suggest that DHT may have a larger influence on liver gene expression compared with E2 in driving sexually dimorphic gene expression. Analysis of GDX males treated with placebo vs. DHT shows that 592 genes differed in expression, whereas only 64 were differentially expressed between GDX males with placebo vs. E2 (supplemental Table 1S). A similar difference was found in GDX females where nine genes were differentially expressed between GDX females treated with placebo vs. E2 and 1392 genes with DHT. However, in this study, we have compared one dose of DHT and E2, which might not have achieved comparable physiological levels of hormone.

Functional significance of hormone regulated sexually dimorphic genes in liver

We determined the general functions of the sexually dimorphic genes of the male and female mice treated with or without gonadal hormones using the pathway enrichment analysis of Gene Ontology (GO) and Panther Pathways. Most of the comparisons yielded differentially expressed genes enriched for biological pathways involved in lipid and steroid metabolism (P < 0.01) (supplemental Table 1S) (39).

Effects of sex chromosome complement on sexually dimorphic gene expression

To further study the effect of sex chromosomes on gene expression, we examined mice of the FCG, in which the testis-determining Sry gene is genetically moved from the Y chromosome to an autosome by deletion and transgenic insertion. The model produces four groups of gonadally intact mice (XX and XY males; XX and XY females), allowing for comparisons of the independent and interacting roles of gonadal hormones and sex chromosome complement (Table 1) (8).

To eliminate the activational effects of the gonadal hormones on gene expression, the FCG mice were GDX at 8 wk. We compared male XY to male XX and female XX to female XY group, where male and female are defined by gonadal sex at birth. We detected fewer than 10 genes that were differentially expressed in XX vs. XY in both the male and female groups.

These results indicate that few genes are regulated by the complement of sex chromosomes when gonadal hormones are absent. In addition, the sex differences in liver gene expression found in gonadally intact adult mice are mostly explained by genes that show regulation by gonadal hormones rather than by the genes that are regulated by sex chromosome complement in GDX mice.

Even though the FCG animals were GDX before profiling, differences in gene expression were observed between males and females with the same sex chromosome status. We observed 552 genes that were differentially expressed between males and females with the XX genotype, and eight genes differed in expression between males and females with the XY genotype. Because the mice had no gonads, the sex differences reflect either direct effects of Sry on the liver or the enduring organizational effects of gonadal secretions before the time of gonadectomy (12). Interestingly, of the 552 differentially expressed genes between XX males and females, 142 overlapped with the 1880 differentially expressed genes between GDX mice treated with placebo vs. gonadal hormones (P = 1.2 × 10⁻³³). The overlap suggests some of the differences between males and females with same sex-chromosome status are due to the enduring effects of gonadal secretions. The direct effects of Sry on the liver are not tested independently in the FCG model because XX and XY mice of the same gonadal sex do not differ in the presence of Sry.

Gene coexpression networks in males and females

We next explored the large-scale organization of gene expression to characterize the gene network properties that are preserved between the two sexes or are sex specific. In addition, we investigated the sex-specific gene networks in relation to metabolic and genetic traits as well as determined whether these networks are influenced by the same molecular mechanisms that affect sexually dimorphic gene expression, detailed above, namely the gonadal hormones.

We constructed gene coexpression networks to detect the architectural differences underlying gene expression between male and female mice. To generate the networks, we used microarray data from the F2 intercross BXH · ApoE^−/− (17,18). The genetic perturbations regulating gene expression in the cross allowed us to identify genes with common patterns of expression, termed modules, in which the network nodes represent gene transcripts and the edges are the degree of connectivity of transcript abundance among F2 mice. For each gene, the connectivity is defined as the sum of connection strengths (k), determined by the pairwise correlations of gene expression levels across microarray samples, with the other network genes. In coexpression networks, the connectivity measures how correlated a gene is with all other genes; thus, a network module consists of a cluster of highly correlated, connected genes (40,41,42,43,44). For each tissue, we selected 4000 genes for the network analysis exhibiting the highest variance of expression across the mouse population. Male and female networks of adipose, brain, liver, and muscle tissue were constructed in parallel.

A plot of the topological overlap matrix depicting the gene coexpression networks between male and female within adipose, brain, liver, and muscle tissue is shown in Fig. 1. The heat map of the topological overlap matrix plot reflects the degree of gene expression correlation within each module identified by color. The module color of the genes in the female network was assigned to the genes in the corresponding male network to visualize the overall high network preservation between males and females. In general, modules found in one sex were also present in the other sex. For example, the large red module found in female muscle was also found in male muscle (Fig. 1). To further investigate the degree of network preservation between male and female, a percent score was calculated to determine the number of genes clustered within the modules that were retained between male and female (supplemental Fig. 1S). Modules were clustered based on degree within the network, and the percent overlap of the male and female module is shown by increasing color intensity depicting higher percent preservation.

Figure 1.

Topographical overlap matrix plot of genetic coexpression network in males and females. The x-axis and y-axis are modules of genes in the entire network and are identified as blocks of colors at the top or left, marking the rows and columns. The plot represents the degree of correlation of gene expression of each gene with all the other genes as indicated by the color bar where increasing color intensity depicts higher correlation (r). Thus, each row or column of this plot represents the connectivity strength of a single gene within the network, with increasing color intensity depicting higher connectivity (higher correlation with other gene’s expression pattern across animals). To visualize the degree of network preservation between males and females, the genes in the male modules were identified as the color of the genes in the female network. Modules were composed of genes with high topological overlap (similar patterns of correlated expression with other genes). The group of genes with low connectivity is colored beige. The plots show that genes within modules have high correlation of expression across mice, which appear as box-like clusters of high heat (high correlation). A, Adipose tissue; B, brain tissue; C, liver tissue; D, muscle tissue.

Consistent with the results of the overall gene coexpression network of each tissue, the gene content of network modules was largely preserved between males and females, although the degree of preservation varied across tissues. The network preservation as depicted in the pattern of Fig. 1S is reflected in the large number of dark blue boxes depicting modules in which a high percentage of genes in a module of one sex is found also in a corresponding module of the other sex. Here, high preservation is defined as more than 50% overlap of genes in the female module that clustered with the same genes in the male module, and vice versa. In brain and muscle, 88–94% of the modules were highly preserved between males and females. The adipose and liver tissues were the least preserved of all the tissues. Approximately 61–69% of the adipose and 53–84% of the liver modules, respectively, was preserved between males and females (Table 3S).

The threshold of significance of the gene overlap of modules between males and females was based on an enrichment analysis using the Fisher’s exact test (Table 3S). Specifically, we determined for each module which module of one sex was enriched in genes found in a module in the opposite sex. The majority of the modules in all the tissues significantly corresponded to a module in the opposite sex. Significant network preservation of a Bonferroni-corrected P < 10⁻⁵⁰ threshold was observed for 50% of the modules in adipose, 61% in brain, 41% in liver, and 31% in muscle. Although it appears the muscle tissue was not significantly preserved, most of the genes in the male and female network fell into four large modules ranging from 312-1538 genes, with a P < 3.61 × 10⁻⁸⁶ to P < 3.15 × 10⁻³¹⁶, compared with the liver tissue where modules were evenly disbursed among 39–909 genes. These results show that the overall gene coexpression network between males and females are highly preserved across tissues, indicating that the same modules of genes that are highly connected in one sex are also highly connected to each other in the other sex. However, the degree of preservation varied between tissues, being greatest in the brain and muscle tissue and least in adipose and liver (Fig. 1 and supplemental Fig. 1S).

Network connectivity between males and females

In addition to comparing general network properties between male and female tissues, we also asked whether the gene connectivity within the network is preserved. For each tissue, female gene connectivity was correlated with male gene connectivity. The relationship between connectivity (k) in the female network and that in the male network is depicted in Fig. 2S and Table 4S; genes in the male network are assigned female module colors. Overall correlation (Pearson correlation coefficient) of gene connectivity within each tissue was as follows: adipose tissue r = 0.67, liver r = 0.56, brain r = 0.92, and muscle tissue r = 0.96 (P < 2.2 × 10⁻¹⁶ for all tissues). The greater correlation coefficients in brain and muscle suggest greater similarity between the sexes of gene network interactions in these two tissues relative to adipose and liver. To assess the extent of preservation of individual modules, we also determined the correlation of gene connectivity of the preserved modules in all the tissues. All the preserved modules had high significant correlation (P < 5.7 × 10⁻³) of gene connectivity with the exception of one module (red) in the adipose network and one module (white) in the muscle tissue (Table 5S).

Network hub genes between males and females

To determine whether the most connected genes (network hubs) are preserved, we compared male and female genes with the highest connectivity index (k). A percent score was used to reflect the degree of preservation between male and female hubs. The percent overlap of male and female hub genes across all modules of the top 10 and 25% most connected genes are depicted in supplemental Fig. 3S. In adipose tissue, 45 or 57% of the hub genes (10% most connected and 25% most connected, respectively) were common to male and female. In liver tissue, there was a 45 or 51% preservation of hub genes in both male and female. In muscle, 61 or 66% of the genes were preserved, and in brain, 56 or 70% of the hub genes overlapped between male and female. These results indicate that the hub genes are highly preserved in all the tissues.

Differential network analysis between male and female

To explore how sex bias in gene expression relates to gene connectivity, the differences of gene expression (differential expression) and gene connectivity (differential connectivity) were integrated. To measure differential gene expression between male and female mice, we used the Student’s t test statistic. The difference in gene connectivity (k) between male and female mice was used as a measure of differential connectivity, and a threshold of significance was established. Differential connectivity, defined as the connectivity of each gene in the female network subtracted from the connectivity of the corresponding gene in the male network allowed us to determine which gene is more connected in one sex compared with the other. A permutation test was performed to determine statistical significance of observing genes with k > 0.2 differential connectivity and Student’s t test (q value FDR 5%) was used to determine the statistical significance of differential gene expression (26).

In Fig. 2, the differential connectivity (x-axis) and differential expression (y-axis) are shown for all tissues, along with the differential connectivity P values for each sector. The data points are individual genes that are colored according to their assigned module. The left sectors (1, 7, and 8) indicate higher connectivity in male than female, and the right sectors (3–5) indicate higher connectivity in female, whereas the top sectors (1–3) show genes with higher gene expression in males than females, and the lower sectors (5–7) show great gene expression in females. Thus, the upper left sector (1) reflects male bias for both gene expression and gene connectivity, whereas the lower right sector (5) indicates female bias for both those characteristics.

Figure 2.

Differential expression and differential connectivity profiles between males and females. The difference between male and female connectivity was calculated by subtracting female connectivity (k) from male connectivity (k), and threshold for significance was determined by a permutation test (P values) (vertical lines). Differential gene expression was measured by Student’s t test, threshold for significance based on 5% FDR (q value) (horizontal lines). Differential connectivity is depicted on the x-axis, where male differential connectivity is on the left (negative) and female differential connectivity is on the right (positive). Differential expression is depicted on the y-axis, where male differential expression is on the top (positive) and female differential expression is on the bottom (negative). A, Adipose tissue; B, brain tissue; C, liver tissue; D, muscle tissue.

In adipose tissue, the blue module (sectors 1–7) was enriched for genes with both male and female-biased gene expression, but with high connectivity in males. The green module (sector 3) was primarily enriched for female-biased gene expression and higher female gene connectivity but also contained male-biased gene expression. In liver tissue, there was primarily one module, cyan (sector 1), which was distinctly different from other modules because of its high male bias in gene expression and higher connectivity in males. The blue, red, and yellow liver modules (sectors 3–5) showed higher female gene connectivity as well as high female and male-biased gene expression. Brain tissue consisted of one module (red) spanning sectors 1, 7, and 8, which reflected both female and male biased gene expression but primarily higher male gene connectivity. Enrichment for female gene connectivity was observed in the pink and green modules in brain (sectors 3–5). Lastly, in muscle tissue, modules with high differential connectivity were not observed for either sex.

Integrating the differential gene expression with connectivity allowed us to determine whether gene transcript abundance affects the organization of gene expression. Despite the large-scale differences in gene expression between males and females, gene connectivity was still largely preserved. Nevertheless, the distributions of genes sorted by differential expression and connectivity showed striking tissue-specific patterns and allowed for the identification of modules with sex-specific patterns.

Network correlation with metabolic traits between males and females

We examined how gene networks differ between males and females in relation to biological traits using a similar method described in the study by Ghazalpour et al. (25) using this same dataset. We used the principal component (pc) of gene expression of each module and correlated each of the metabolic traits measured. Briefly, we defined the module eigengene as the first principal component of the module expressions. The module eigengene can be considered the most representative gene expression in a module, because it captures most of the variance explained. In addition, for each clinical trait, we defined a measure of gene significance (gs) as the absolute value of the correlation between gene expression profile of each module and the clinical trait. Module significance was determined as the average gene expression correlation for all genes in a given module. This measure is highly related to the correlation between module eigengene and the trait.

We detected modules of the adipose and liver correlated to metabolic traits but no trait correlations with modules of the brain or muscle tissue were observed (Fig. 3 and supplemental Table 6S). Although the study by Ghazalpour et al. (25) also detected a liver module correlated with metabolic traits, the analysis was performed only in the female network with a different set of genes and primarily focused on the network analyses of one module. We observed a sexually dimorphic module in the adipose and liver network correlated to metabolic traits. The adipose blue module was strongly correlated with traits related to body fat, lipid, and atherosclerosis in both males and females. In adipose, the pc of the blue module was highly correlated to body weight in both males and females (r = 0.63–0.73) as well as total cholesterol, aortic lesions, and the monocyte chemotactic protein-1 (MCP-1), a marker for inflammation (r = 0.31–0.37). Correlations with the adipose blue module and abdominal fat, insulin, and leptin were also observed but were more correlated in females (r = 0.51–0.57) than in males (r = 0.34–0.4). The liver green module was correlated to body fat, glucose, and triglycerides in females but not in males. Abdominal and total fat was strongly correlated with the female liver network (r = 0.5–0.57) compared with the male network (r = 0.1–0.14); glucose and triglycerides was more correlated with the female network (r = 0.34–0.39) than the males network (r = 0.09–0.25). However, the liver green module was highly correlated to weight and leptin (r = 0.4–0.65) as well as total cholesterol and insulin (r = 0.31–0.41) in both males and females.

Figure 3.

Network correlation with metabolic traits between male and female. We measured the correlation of each module with metabolic traits representing body fat, lipid, and atherosclerosis. The principal component of gene expression (pc) of each module was calculated and correlated to each of the metabolic traits. A, Adipose blue module; B, liver green module.

Genetic regulation of male and female networks

In previous studies, sex differences in gene eQTL have been observed reflecting sex-specific gene regulation (13,14,15,16,17). To examine the chromosomal loci regulating the networks between male and female mice , we screened the sex-specific modules for eQTL (Fig. 4, A and B, and supplemental Fig. 4S). The sex-biased modules consisted of genes that were highly connected in one sex and not the other as detected by the differential network connectivity analysis (Fig. 2). The transcript abundance of each gene in the sex-biased module was then used to detect its corresponding QTL. We also calculated the principal component of the transcript abundance of the sex-biased module to determine how the module relates to the QTL as a whole.

Figure 4.

eQTL of the blue module in the adipose network and the cyan module in the liver network. The gene transcript abundance was used as a trait to detect QTL and the number of eQTL with a LOD of 3 or higher are plotted. The red curve indicates the eQTL count in the female network, and the blue curve indicates the eQTL count male network. The adipose blue module consisted of 1090 genes and the cyan consisted of 82 genes. A, Adipose blue module; B, liver cyan module.

In adipose tissue, the blue module of the female network consisted of striking eQTL hotspots on chromosome 1 and 7, whereas in the male network, eQTL hotspots were observed on chromosomes 8 and 11 (Fig. 4A). A similar result was found when the LOD score of the principal component of the same modules was determined (supplemental Fig. 4S); the female blue adipose module had LOD higher than 6 for chromosome 1, and the same module in males showed LOD higher than 4 for chromosome 8. In the liver tissue, the cyan module of the male network showed enrichment for eQTL on the X chromosome but no strong eQTL were found for the equivalent module in the female network (Fig. 4B).

In brain and muscle tissue, strong sex-specific eQTL hotspots were not observed. The sex differences in the QTL detected of the adipose and liver module provide evidence for genetic locations that perturb networks in a sex-specific manner.

Networks enriched for genes driven by sex hormones

In addition to determining whether sexually dimorphic coexpression network modules may be involved in certain biological pathways, we examined modules for enrichment of genes whose expression was differentially affected by gonadal hormones. Differentially expressed genes between the GDX mice treated with placebo compared with GDX mice treated with E2 and DHT (P < 0.01) were screened for module enrichment in the liver tissue. Genes belonging to the cyan module of the liver tissue with the highest connectivity in male (shown in Fig. 2, sector 1) were largely the same set of genes significantly affected by both the presence of the testes and DHT (P = 5.6 × 10⁻²⁸) (Tables 2 and 3). Similar network enrichment was also observed in the GDX female mice treated with DHT. Genes affected by the presence of ovaries and E2 were not strongly enriched for any modules, which was not surprising because highly connected female modules were not observed in the liver tissue.

Table 2.

Sexually dimorphic network enriched for genes affected by gonadal hormones

Experiment	Experimental variable				Sector	EASE score	Bonferroni P value
	Testis	Ovaries	E2	DHT
Castrated male placebo vs. castrated male DHT				X	1	1.91 × 10−27	1.71 × 10−26
Castrated male placebo vs. castrated male E2			X		NS	NS	NS
Castrated male placebo vs. ovariectomized female placebo					NS	NS	NS
Intact female vs. intact male	X	X	X	X	1	3.12 × 10−34	2.81 × 10−33
Ovariectomized female placebo vs. ovariectomized female DHT				X	1	1.86 × 10−13	1.67 × 10−12
Ovariectomized female placebo vs. ovariectomized female E2			X		NS	NS	NS
Intact male vs. castrated male placebo	X			X	1	5.64 × 10−28	5.07 × 10−27
Intact female vs. ovariectomized female placebo		X	X		NS	NS	NS
Intact male vs. castrated male DHT	X				1	2.47 × 10−16	2.23 × 10−15
Intact female vs. ovariectomized female E2		X			1	NS	NS

The genes in a male sex-specific liver module of the BXH · ApoE^−/− F2 cross were screened for liver genes differentially expressed between mice with or without gonadal hormones (P < 0.01). Degree of enrichment was measured by the Fisher exact test within the EASE software. NS, Not significant. 

Table 3.

Differentially connected genes of the liver ‘cyan’ module enriched for differentially expressed genes between gonadectomized males treated with placebo vs. DHT

Gene symbol	Locus link ID	Chromosome	Official gene name	Fold change	Differential connectivity (k)	Sex differential expression (t score)	Hormone differential expression (P value)
Ankrd13b	268445	11	Ankyrin repeat domain 13b	1.28	−0.72	6.26	7.20 × 10−4
Bmf	171543	2	Bcl2 modifying factor	1.45	−0.62	6.02	1.14 × 10−3
C8b	110382	4	Complement component 8	1.19	−0.31	3.80	4.78 × 10−4
Chad	12643	11	Chondroadherin	1.14	−0.78	4.90	4.84 × 10−4
Chst11	58250	10	Carbohydrate sulfotransferase 11	1.12	−0.58	3.95	2.16 × 10−3
Csf3	12985	11	Colony stimulating factor 3	1.08	−0.67	4.50	2.54 × 10−3
Cyp26b1	232174	6	Cytochrome P450	1.26	−0.55	5.51	1.10 × 10−3
Egfr	13649	11	Epidermal growth factor receptor	1.40	−0.30	5.89	1.72 × 10−4
Es31	382053	8	Esterase 31	1.17	−0.24	6.02	1.18 × 10−3
F11	109821	8	Coagulation factor XI	0.95	−0.15	−2.55	2.11 × 10−3
Gab3	210710	X	Growth factor receptor	1.18	−0.74	5.75	2.13 × 10−4
Gpx1	14775	9	Glutathione peroxidase 1	1.06	−0.19	2.98	2.19 × 10−3
Hydin	244653	8	Hydrocephalus inducing	1.16	−0.57	5.71	2.14 × 10−3
Ighg	16061	12	Immunoglobulin heavy chain	1.23	−0.76	5.89	5.21 × 10−4
Kcne2	246133	16	Potassium voltage-gated channel	1.08	−0.45	3.91	8.86 × 10−4
Krt15	16665	11	Keratin 15	1.19	−0.69	4.84	7.25 × 10−4
Lgtn	16865	1	Ligatin	1.24	−0.53	5.36	9.75 × 10−4
Ly9	17085	1	Lymphocyte antigen 9	1.10	−0.58	4.49	7.19 × 10−4
Mars	216443	10	Methionine-tRNA synthetase	1.27	−0.73	6.00	5.20 × 10−4
Mllt6	246198	11	Myeloid/lymphoid	1.10	−0.61	5.44	1.59 × 10−3
Mug1	17836	6	Murinoglobulin 1	1.09	−0.24	3.54	1.33 × 10−3
Myh2	17882	11	Myosin	1.36	−0.23	4.80	1.77 × 10−3
Ppme1	72590	7	Protein phosphatase methylesterase 1	1.35	−0.72	5.30	2.25 × 10−3
Rabif	98710	1	RAB interacting factor	1.38	−0.71	5.39	2.70 × 10−3
Rad51l1	19363	12	RAD51-like 1 (S. cerevisiae)	0.88	−0.56	−3.48	6.49 × 10−4
Ruvbl2	20174	7	RuvB-like protein 2	1.17	−0.76	5.58	1.64 × 10−3
Serpina3k	20714	12	Serine (or cysteine) peptidase inhibitor	1.27	−0.43	5.47	1.28 × 10−3
Serpinb3c	381286	1	Serine (or cysteine) peptidase inhibitor	1.16	−0.82	4.80	6.04 × 10−4
Slc22a16	70840	10	Solute carrier family 22	1.15	−0.67	4.83	2.84 × 10−3
Slco1a1	28248	6	Solute carrier organic anion transporter	1.63	−0.51	6.12	1.27 × 10−3
Smyd5	232187	6	SET and MYND domain containing 5	1.16	−0.52	5.37	1.10 × 10−3
Spag16	66722	1	Sperm associated antigen 16	1.17	−0.80	5.15	2.73 × 10−3
Spdyb	74673	5	Speedy homolog B (Drosophila)	1.11	−0.63	4.82	2.27 × 10−3
Spsb1	74646	4	splA/ryanodine receptor/SOCS box	1.20	−0.72	4.61	2.26 × 10−3
Sult5a1	57429	8	Sulfotransferase family 5A	1.14	−0.42	3.65	4.57 × 10−4
Susd4	96935	1	Sushi domain containing 4	1.65	−0.19	6.32	3.93 × 10−4
Tmc3	233424	7	Transmembrane channel-like	1.15	−0.76	5.76	1.29 × 10−3
V1ra2	22297	6	Vomeronasal 1 receptor	1.10	−0.23	4.04	2.92 × 10−3
Wif1	24117	10	Wnt inhibitory factor 1	1.09	−0.66	4.29	7.90 × 10−4
Zp3	22788	5	Zona pellucida glycoprotein 3	1.17	−0.82	5.39	2.58 × 10−4

Approximately 16% of the cyan module consisted of genes that were highly differentially connected and expressed in males (P < 0.05), and 79% of these genes were enriched for differentially expressed genes affected by DHT. Fold change indicates the difference of the transcript abundance between GDX males treated with DHT, and the hormone differential expression is the P value between the treatment groups. Differential connectivity (k) is the difference between male and female connectivity. Sex differential expression (t score) reflects the difference between the means of the transcript abundance between males and females. Degree of enrichment was measured by the Fisher exact test within the EASE software. 

Discussion

Males and females are affected by different patterns of disease manifestation and progression, but how these differences are biologically determined remains to be elucidated. In our recent study, we identified thousands of genes differentially expressed across tissues under the control of sex-specific QTL hotspots (18). In this report, we investigated sexual dimorphism of gene expression at the molecular level and explored gene coexpression networks between males and females. To our knowledge, this is the first study to ask whether sexual dimorphisms can be detected at the level of global gene expression networks, rather than simply at the level of expression of individual genes. These analyses provide new perspectives on sex-specific organization of gene networks and their regulation and relationship to clinical traits.

The major conclusions that emerged from our study are: 1) in the liver tissue, many differences in gene expression between males and females can be attributed to the acute effects of gonadal hormones acting in adulthood; 2) relatively few sex differences in liver are explained by direct effects of the sex chromosome genes; 3) although the difference in gene expression typically does not influence the overall pattern of gene connectivity in the two sexes, in some cases, they result in dramatic effects on modules in coexpression networks, particularly in liver and adipose and to a lesser extent in muscle and brain; and 4) certain sex-specific modules are clearly associated with clinical traits as well as chromosomal loci regulating these networks; 5) We have identified at least one module in liver (cyan) that shows male-specific expression and connectivity and is largely composed of genes that respond directly to androgens in adulthood.

We explored the molecular basis of sexually dimorphic gene expression by investigating the effects of gonadal hormones and sex chromosomes on gene expression. We observed 2597 genes differentially expressed between gonadally intact males and females in liver tissue that were highly enriched for the sexually dimorphic genes observed in the liver tissue of the BXH · ApoE^−/− F2 mice cross (P = 4.1 × 10⁻¹³⁶) (18). In addition, the genes that were differentially expressed between the intact and the GDX mice, treated with or without gonadal hormones, overlapped significantly with the genes that were sexually dimorphic in the BXH · ApoE^−/− F2 animals, indicating that gonadal hormones play a strong role in sexual dimorphism of gene expression. The gonadal hormone effects were observed to be larger with the treatment of DHT in either sex, and thus the hormone effects are likely mediated by androgen receptors. The number of genes that were differentially expressed between GDX mice treated with placebo vs. DHT were much higher than the number of genes that were differentially expressed between GDX mice treated with placebo vs. E2. The genes affected by DHT were primarily enriched for genes involved in pathways for lipid and steroid metabolism (supplemental Table 1S).

In addition to studying the effects of gonadal hormones on sexually dimorphic gene expression, we measured the role of the sex chromosomes on gene expression. We detected eight differentially expressed genes between XX and XY males, and no genes were differentially expressed between XX and XY females. These results suggest that compared with the gonadal hormones, the sex chromosomes play a modest role in driving differential gene expression in liver between adult males and females. Because the sex chromosome effects were measured in GDX mice, it is conceivable that more sex chromosome effects would occur in mice that have higher levels of gonadal hormones. For example, some effects of X or Y genes might occur only when androgens or estrogens are present. This analysis provides some evidence for interactions between sex chromosome complement and gonadal sex, suggesting that some hormone effects differ in XX and XY mice.

Next, we investigated the organization of gene expression between males and females. This analysis allowed us to identify genes as well as gene coexpression networks that are highly connected in one sex but not the other, providing evidence for a sex-specific functional relationship. Sexually dimorphic gene coexpression networks were identified by the degree of differential gene connectivity between males and females and subsequently characterized in relation to genetic and physiological traits. Although sexually dimorphic modules were observed in all tissues, the coexpression networks of adipose and liver tissue were highly enriched for modules with striking sex differences compared with brain and muscle. These results are not unexpected, because both adipose and liver are endocrine-responsive organs that interact with sex hormones. For example, Cyp19a1, the aromatase involved in estrogen regulation in both adipose and liver tissue, was identified as a hub in the adipose network in females but not in males (9). The high degree of network preservation of brain tissue was somewhat unexpected, because it is a well-documented sexually dimorphic organ (6). Sexual dimorphism has been reported in specific brain regions, such as the hypothalamus (45). Because the brain is very heterogeneous with respect to gene expression, examination of the entire brain may have resulted in a loss of power to identify sex-specific expression patterns. Among all tissue, the muscle network showed the greatest preservation of module composition between the sexes, suggesting similar overall function in male and female muscle. These observations indicate that the gene clustering of modules of these coexpression networks are highly preserved between males and females but that the degree of preservation is tissue-specific (Fig. 1 and supplemental Fig. 1S).

We determined whether the preservation observed in the overall tissue network extended to gene connectivity and the hub genes. We observed a high number of sexually dimorphic hub genes in adipose and liver compared with brain and muscle tissue. Approximately 50% of the hub genes in adipose and liver were identified as a hub gene in one sex but not the other compared with 34% of the hub genes in brain and 30% in muscle (supplemental Fig. 3S).

Our analyses based on network connectivity and differential expression allowed us to detect three sex-specific gene coexpression modules in adipose (male, blue; female, blue and green), five sex-specific modules in liver (male, green and cyan; female, blue, red, and yellow), three sex-specific modules in brain (male, red; female, pink and green) but no sex-specific modules in muscle (Fig. 2). Some modules had high sexually dimorphic gene connectivity and high differential expression, but the majority showed sex-biased differential expression but not differential connectivity (females, k < 0.2; males, k > −0.2; Fig. 2). These results indicate that gene coexpression networks are preserved between males and females despite a higher or lower abundance of gene transcript between the sexes. Therefore, a significant sex difference in gene expression does not necessarily lead to a difference in the organization of the gene expression network between males and females.

Although the majority of gene coexpression networks consisted of the same genes that clustered together in both males and females, modules with highly differentially connected genes were detected. The preserved modules consisting of genes that are more connected in one sex compared with the other imply that the network connectivity is sexually dimorphic. To illustrate this observation, we used the VisANT visual analysis tool to depict the network structure of a module made up of correlated genes in both males and females but are sex-specifically differentially connected (Fig. 5). This module consists of the same group of correlated genes in both the male and female liver network (Fig. 2, sector 1, and supplemental Table 8S); however, the male module is more connected than the female module as reflected in the number of genes with at least 40 connections in the male network compared with the female network consisting primarily of genes with fewer than 40 connections.

Figure 5.

Visualization of sex-specific differentially connected genes of the liver cyan module (Fig. 2, sector 1). These genes were also enriched for differentially expressed genes affected by DHT. Genes with 40 or more connections are identified by the larger nodes as seen primarily in the male network, and genes with fewer than 40 connections are identified by the smaller nodes as seen primarily in the female network.

We characterized these sexually dimorphic gene coexpression networks further by measuring the correlation of genetic and metabolic traits in males and females (Fig. 3 and supplemental Table 6S). The adipose blue module was significantly correlated with metabolic traits such as body weight, leptin, and insulin (r > 0.4) (Fig. 3) and genetic loci in both sexes (Fig. 4A). These results are not surprising because the blue module is highly enriched for genes known to be involved in metabolism and obesity, including leptin, lipin, USF1, and PPAR-α, as well as genes involved the inflammatory response, biopolymer metabolism, catalytic and GTPase regulator activity (Table 4) (46).

Table 4.

Functional significance of sexually dimorphic network modules using pathway enrichment of DAVID software

Module, sex specificity	Gene enrichment category	P value	Bonferroni P value
Adipose tissue
Blue, male	Lysosome	1.26 × 10−6	9.40 × 10−4
	Catalytic activity	1.55 × 10−6	2.00 × 10−3
	Phosphorylation	1.49 × 10−5	1.10 × 10−2
	GTPase regulator activity	2.04 × 10−5	2.60 × 10−2
	Hemopoiesis	5.87 × 10−5	1.30 × 10−1
	Immune response	7.96 × 10−5	1.70 × 10−1
Green, female	Cell communication	1.54 × 10−6	3.50 × 10−3
	Sensory perception of chemical stimulus	4.89 × 10−6	1.10 × 10−2
	Olfactory receptor activity	1.02 × 10−5	1.30 × 10−2
	Signal transduction	1.32 × 10−5	3.00 × 10−2
	Neurophysiological process	1.82 × 10−5	4.10 × 10−2
Brain tissue
Green, female	RRM	1.83 × 10−4	7.30 × 10−2
	Cell organization and biogenesis	5.99 × 10−4	7.20 × 10−1
	Intracellular organelle	7.13 × 10−4	2.30 × 10−1
Pink, female	Intracellular organelle	1.26 × 10−3	3.70 × 10−1
	Organelle	1.30 × 10−3	3.80 × 10−1
Red, male	KRAB	1.11 × 10−18	4.60 × 10−16
	Zinc finger, C2H2-subtype	6.70 × 10−12	1.30 × 10−8
	Regulation of cellular metabolism	2.14 × 10−7	4.50 × 10−4
	Nucleus	2.21 × 10−6	8.00 × 10−4
	Transcription, DNA-dependent	6.03 × 10−7	1.30 × 10−3
Liver tissue
Blue, female	Membrane	7.65 × 10−18	8.70 × 10−15
	Transmembrane	1.86 × 10−15	2.20 × 10−12
	Signal	3.13 × 10−15	3.50 × 10−12
	Immune response	1.18 × 10−14	4.30 × 10−11
Cyan, male	Endoplasmic reticulum	4.90 × 10−4	3.10 × 10−1
	Heme	1.40 × 10−3	6.60 × 10−1
	Serine esterase	2.00 × 10−3	7.80 × 10−1
	Serine esterase activity	1.85 × 10−3	9.00 × 10−1
	Carboxylesterase activity	1.85 × 10−3	9.00 × 10−1
	Cytochrome P450	1.61 × 10−3	9.60 × 10−1
	Transport	1.75 × 10−3	9.80 × 10−1
Green, male	Cell adhesion	2.11 × 10−8	4.80 × 10−5
	Extracellular matrix	1.44 × 10−7	1.10 × 10−4
	Signal	3.85 × 10−6	2.90 × 10−3
	Glycoprotein	1.86 × 10−5	1.40 × 10−2
	Metal ion binding	7.61 × 10−5	9.10 × 10−2
	Ion binding	7.61 × 10−5	9.10 × 10−2
Red, female	Organelle	1.51 × 10−5	9.20 × 10−3
	Intracellular	1.64 × 10−5	1.00 × 10−2
	Nuclear protein	2.06 × 10−5	2.30 × 10−2
	Kinase	5.88 × 10−5	6.50 × 10−2
Yellow, female	Alcohol metabolism	8.09 × 10−6	2.90 × 10−2
	Monosaccharide biosynthesis	3.40 × 10−5	1.20 × 10−1

RRM, RNA recognition motif; KRAB, Krüppel-associated box. 

The QTL detected for this adipose module in males and females were on different chromosomes (Fig. 4A). These QTL results indicate that the blue module, which is made up of highly differentially connected genes and is strongly correlated to metabolic traits, is also regulated by genes on different chromosomes in the male and female adipose network. These QTL sex difference results are further evidence that by integrating sex as a covariate when evaluating QTL may increase the power to detect genes underlying complex genetic traits (13,14,15,16,17).

Our analysis also uncovered a sex-specific gene network in the liver tissue. The cyan module showed strong differential expression and connectivity in males (Fig. 2, sector 1, and Fig. 5) and was highly enriched for genes affected by gonadal hormones. The module was not significantly correlated with any metabolic traits nor did it have any module QTLs (data not shown), but it did consist of eQTLs on the X chromosome in males (Fig. 4B). These results reiterate the findings from the sex chromosome complement analysis in our study, where genes on the sex chromosome could influence or be influenced by gonadal secretions. In addition, genetic variation on the X chromosome may cause differences in the expression of its genes in cis (genetic variation in one locus that drives differences in expression of genes on or near that locus), which subsequently act in trans (genetic variation in one locus that drives differences in expression of genes distant from that locus) to affect the autosomal genes of this module. The regulation by the X chromosome in males but not females may reflect the hemizygous exposure of X alleles in males, so that variation in the effects of X alleles is more pronounced in males than in females.

The cyan module was enriched for genes involved in carboxyl and serine esterase activity and endoplasmic reticulum, which included genes involved in spermatogenesis, such as the activator of CREM (Fhl5), zona pellucida glycoprotein 3 (Zp3), sperm-associated antigen 16 (Spag16), and Y-box protein 2 (Ybx2) as well as hormone-related genes such as esterase 31 (Es31), hydroxysteroid dehydrogenase-5 (Hsd3b5), and cytochrome P450 (Cyp26b1 and Cyp7b1), which are known to be regulated by the sexually differentiated GH (Table 3) (19). Although most of the genes in this network have not been well annotated, we speculate that the genes may be regulated by a common transcription factor influenced by gonadal hormones. Thus, the cyan module in liver is an androgen-sensitive group of genes that show male bias in expression and network connectivity, which are driven more in males than females by genetic variation on the sex chromosomes. The X chromosome is expected under some conditions to be enriched for genes that play a role in male-specific functions (47).

In conclusion, using a constellation of global network and linkage analysis tools, we identified sexually dimorphic gene networks in both adipose and liver tissue associated with sex-specific metabolic traits and regulated by different genetic loci between males and females. Investigation of sexually dimorphic gene coexpression networks as they relate to biological traits allowed us to obtain a better understanding of the interactions that may play a significant role in sex-biased disease (48,49). Genes whose expression and interactions are affected as a response to the gonadal hormones and are highly correlated to biological traits related to atherosclerosis, diabetes, and obesity may help uncover key sex-specific metabolic therapeutic targets.