- Open Access
Are females more variable than males in gene expression? Meta-analysis of microarray datasets
© Itoh and Arnold. 2015
Received: 3 August 2015
Accepted: 11 September 2015
Published: 29 October 2015
The majority of preclinical biomedical research involves studies of males rather than females. It is thought that researchers have avoided females based on the idea that female traits are more variable than those of males because of cyclic variation in effects of ovarian hormones.
To test the assumption of inherently greater female variability, we analyzed 293 microarray datasets measuring gene expression in various tissues of mice and humans, comprising analysis of more than 5 million probes.
Meta-analysis showed that on average, male gene expression is slightly more variable than that of females although the difference is small. We also tested if the X chromosome of humans shows greater variability in gene expression in males than in females, as might be expected because of hemizygous exposure of polymorphic X alleles but again found little sex difference.
Our analysis supports and extends previous studies reporting no overall greater phenotypic variability in females.
Recent analysis of published articles indicates that in numerous biomedical fields, male animals are used as subjects more than females [1–4]. Moreover, many studies fail to report the sex of animals, tissues, or cells used in the study so that it is impossible to assess whether the sex of the animal or tissue is an important variable. The male bias raises the concern that scientific findings may be applied with greater certainty to males than females.
Researchers may avoid studying female rodents because they wish to avoid the variability thought to be caused by the estrous cycle. The estrous cycle of female mice is about 4–5 days in length and involves changes in levels of estradiol, progesterone, gonadotrophins, and gonadotrophin releasing hormone . These hormones can have potent effects on gene expression and other phenotypes, including epigenetic changes across the genome [6–8]. On the other hand, group-housed male mice establish a dominance hierarchy leading to individual differences in the level of testosterone, which would also be expected to increase variation in phenotype . Moreover, social status of mice could influence levels of glucocorticoids.
Prendergast et al.  analyzed 293 articles to compare the variability of phenotypes between gonad-intact male and female mice. They found that females are not more variable than males and that under some conditions males are more variable than females. The amount of male variability was reduced in mice housed as individuals, supporting the idea that phenotypic variability in males might be related to social factors, glucocorticoids, and/or testosterone levels.
It has been postulated that in outbred populations such as humans, X genes may have more variable effects on phenotype in males than in females because the effect of each X gene variant in females is averaged with the effect of the X allele on the other X chromosome, whereas the X chromosome variation is not reduced by averaging in males because of hemizygous exposure of the X chromosome. On the other hand, males experience only the maternal imprint on X alleles, whereas females experience the imprints of both parents, which could increase variability of X gene expression in females relative to males.
The development of microarrays revolutionized the global analysis of gene expression in diverse tissues and has often been used to detect patterns of expression across the genome. Microarrays and other high-throughput methods are sensitive, accurate, and suitable for the assessment of effects of sex, genotype, environment, and treatment variables on global patterns of gene expression.
Here, we performed a meta-analysis of online databases of gene expression based on microarray analysis, to compare phenotypic variation in males and females. We analyzed data from human and mice. We looked for evidence that gene expression is more variable in females than in males, either globally throughout the genome or in a subset of genes that would be large enough to be observed as a shift in the distribution of expression variance relative to males. Based on the analysis of more than 5 million probes, we found that variation of gene expression was quite similar between males and females, with slight overall bias towards greater variability in males. We also compared variation in expression of X and autosomal genes in humans and mice and found little evidence for greater variability of expression of X alleles relative to autosomal alleles in both species.
Microarray data analysis
We selected for analysis a total of 293 datasets (103 for human, 190 for mouse) obtained from the GEO database (http://www.ncbi.nlm.nih.gov), which compared male and female human or mouse samples (Additional file 1: Table S1). These datasets report gene expression levels in a variety of tissues (Additional file 2: Figure S1) based on microarray expression profiling using a variety of platforms. The goal was to obtain as much data as possible, to avoid any bias that might be specific to an individual microarray methodological approach. Datasets were included if the study compared males and females, comprised at least three independent samples per sex, and if any treatment or disease condition was applied equally to both sexes (Additional file 1: Table S1). We attempted to analyze all datasets that met these criteria. In a subset of datasets (87 for human, 190 for mouse), we were able to identify probes for genes that are encoded on the X chromosome or autosomes. We analyzed patterns of variation of X and autosomal genes in both sexes. Data from mice came from studies in which the estrous cycle of mice was not monitored by the investigators.
Statistical analyses and production of graphs were performed in the statistical environment R . The filtered probes were quantile normalized using the “affy” package from Bioconductor (http://www.bioconductor.org/). The variability of gene expression was measured by coefficient of variation (CV, standard deviation divided by the mean). We compared CV of the two sexes within each dataset. The CV is meant to allow comparison of variation in data with different means because CV compensates for the increase in variation as the mean increases.
In this study, microarray datasets from brain contribute disproportionately to the data of Fig. 1, especially in humans (Additional file 2: Figure S1). Therefore, we asked if the pattern in Fig. 1 reflects mostly brain and whether sex bias in variation might differ in non-brain tissues. Additional file 4: Figure S3 separates the analysis for brain and non-brain tissues in human and mice. All of these analyses show either very slight male bias as in Fig. 1, or sexual equivalence of CV, and therefore do not support the idea that degree of sexual bias in CV differs significantly in brain relative to other tissues.
Number of datasets
Number of probes
p value for sex
Groups of genes from 87 human datasets were categorized according to those that exceeded progressively large CV cutoff values. The percentage of genes in each group that are X-linked is shown in the table, whereas the percentage that was autosomal is 100 minus the number shown. Data for mice is not shown because not many genes showed higher CV values
CV > 1
CV > 2
CV > 3
CV > 4
CV > 5
CV > 6
CV > 7
We analyzed 293 microarray gene expression datasets utilizing more than 5 million probes to address the issue of gene expression variability in male and female humans and mice. We found that the variability of gene expression, measured by CV, was similar in the two sexes. The result provides no support for the hypothesis that female mice or humans are generally more variable in phenotype because of their estrous or menstrual cycles or other variables. Indeed, males were on average slightly more variable in some measures of gene expression. Although one sex was sometimes more variable in overall gene expression in specific tissues, the sex bias was in either direction, depending on the tissue, suggesting that sources of sex-specific variability might differentially affect specific tissues. Our results confirm and extend those of previous studies [3, 10] that found no evidence for generally greater variability of various phenotypes in gonad-intact mice, when, as in the present study, estrous stage was not monitored in females. We found that studying sex differences in gene expression can provide an interesting perspective to separate specific populations of genes for further analysis, for example in the study of Franco et al. , which provided data that urethane causes a female-specific increase in variability of a selected population of genes (Fig. 4). Finally, our results provide little evidence that X-linked gene expression in humans was more variable in males than in females, as might be expected because of the male’s hemizygous X chromosome.
Sex differences in any phenotype are caused by two kinds of mechanisms, ontogenetic (factors that are inherently different between each male and female, which cause sexual differentiation of tissues), and population-level mechanisms (factors that act in a greater proportion of individuals of one sex than the other, leading to average differences among males and females). Ontogenetic factors include the effects of gonadal hormones, both organizational (permanent, differentiating) and activational (reversible) effects . Variation in the effects of gonadal hormones would be expected to produce the largest sex differences in variability of traits, and accordingly, it is temporal or socially induced variation in effects of gonadal hormones that is most often suggested as a source of sex differences in trait variance. The population-level factors include the following: (1) Hemizygous exposure of X alleles. Variation in X alleles induces more variation in phenotypes of males than females because the effect of each variant is averaged across two alleles in females but is expressed fully in individual males. For example, Fragile X syndrome and other types of X-linked mental retardation affect human males more than females [15, 16]. The averaging process should reduce variability among females relative to males. (2) Individual sexually antagonistic autosomal alleles (conferring different fitness effects in males and females) may occur more in one sex than the other because sex-specific deleterious or lethal alleles might drop from the population of one sex more than the other. (3) “Mother’s curse” [17, 18]: because the mitochondrial genome is passed from mother to daughter, male-disadvantageous alleles may build up if they confer advantages to females and disproportionately promote disease in males.
Although sex differences in level of gene expression are normally thought to contribute to sex differences in physiology or disease, sex-biased variation itself, caused by any type of ontogenetic or population-level effect, can cause one sex to reach a threshold for disease or lethality more than the other sex.
The population-level sources of sex-biasing factors will operate only in genetically heterogeneous populations, not in inbred strains. Among the samples analyzed here, therefore, we expected that the human datasets would represent measures of genetically heterogenous individuals, whereas the mouse datasets would often come from inbred lines. This species difference may have contributed to the greater CV of probes in datasets from humans than from mice (Fig. 5). Some evidence suggests that X genes showing high CV values were more likely to occur in males than females (Table 2). Otherwise, we found little evidence that X genes had greater variability in males than females in humans (where it might have occurred) than in mice (where we did not expect it). Because X genes drive and are driven by autosomal genes within gene networks, it is likely that any tendency for greater variation in X gene expression in males is blunted by network feedback or other interactions with autosomal genes, which comprise the vast majority of interacting partners of X genes within gene networks and which would not be expected to show an overall inherent sex bias.
Based on extensive analysis of microarray datasets measuring gene expression in both sexes of mice and humans, we found no evidence that variability of gene expression is generally greater in females than males.
This work was supported by the NIH grants DK083561, HD076125, and R56119886.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Beery AK, Zucker I. Sex bias in neuroscience and biomedical research. Neurosci Biobehav Rev. 2011;35:565–72.PubMed CentralView ArticlePubMedGoogle Scholar
- Klein SL, Schiebinger L, Stefanick ML, Cahill L, Danska J, De Vries GJ, et al. Opinion: sex inclusion in basic research drives discovery. Proc Natl Acad Sci U S A. 2015;112:5257–8.PubMed CentralView ArticlePubMedGoogle Scholar
- Mogil JS, Chanda ML. The case for the inclusion of female subjects in basic science studies of pain. Pain. 2005;117:1–5.View ArticlePubMedGoogle Scholar
- Zucker I, Beery AK. Males still dominate animal studies. Nature. 2010;465:690.View ArticlePubMedGoogle Scholar
- Caligioni CS. Assessing reproductive status/stages in mice. Curr Protoc Neurosci. 2009; Appendix 4: Appendix. doi:10.1002/0471142301.nsa04is48.
- Kubarek L, Kozlowska A, Przybylski M, Lianeri M, Jagodzinski PP. Down-regulation of CXCR4 expression by tamoxifen is associated with DNA methyltransferase 3B up-regulation in MCF-7 breast cancer cells. Biomed Pharmacother. 2009;63:586–91.View ArticlePubMedGoogle Scholar
- Schwarz JM, Nugent BM, McCarthy MM. Developmental and hormone-induced epigenetic changes to estrogen and progesterone receptor genes in brain are dynamic across the life span. Endocrinol. 2010;151:4871–81.View ArticleGoogle Scholar
- Nugent BM, Schwarz JM, McCarthy MM. Hormonally mediated epigenetic changes to steroid receptors in the developing brain: implications for sexual differentiation. Horm Behav. 2011;59:338–44.PubMed CentralView ArticlePubMedGoogle Scholar
- Machida T, Yonezawa Y, Noumura T. Age-associated changes in plasma testosterone levels in male mice and their relation to social dominance or subordinance. Horm Behav. 1981;15:238–45.View ArticlePubMedGoogle Scholar
- Prendergast BJ, Onishi KG, Zucker I. Female mice liberated for inclusion in neuroscience and biomedical research. Neurosci Biobehav Rev. 2014;40:1–5.View ArticlePubMedGoogle Scholar
- R Development Core Team. R: A language and environment for statistical computing. In: R Foundation for Statistical Computing. Vienna, Austria: 2006.Google Scholar
- Zahn JM, Poosala S, Owen AB, Ingram DK, Lustig A, Carter A, et al. AGEMAP: a gene expression database for aging in mice. PLoS Genet. 2007;3:e201.PubMed CentralView ArticlePubMedGoogle Scholar
- Franco MD, Colombo F, Galvan A, Cecco LD, Spada E, Milani S, et al. Transcriptome of normal lung distinguishes mouse lines with different susceptibility to inflammation and to lung tumorigenesis. Cancer Lett. 2010;294:187–94.View ArticlePubMedGoogle Scholar
- Arnold AP. The organizational-activational hypothesis as the foundation for a unified theory of sexual differentiation of all mammalian tissues. Horm Behav. 2009;55:570–8.PubMed CentralView ArticlePubMedGoogle Scholar
- Hagerman RJ. Fragile X syndrome. Curr Probl Pediatr. 1987;17:621–74.PubMedGoogle Scholar
- Howard-Peebles PN. Non-specific X-linked mental retardation: background, types, diagnosis and prevalence. J Ment Defic Res. 1982;26(Pt 4):205–13.PubMedGoogle Scholar
- Camus MF, Clancy DJ, Dowling DK. Mitochondria, maternal inheritance, and male aging. Curr Biol. 2012;22:1717–21.View ArticlePubMedGoogle Scholar
- Gemmell NJ, Metcalf VJ, Allendorf FW. Mother’s curse: the effect of mtDNA on individual fitness and population viability. Trends Ecol Evol. 2004;19:238–44.View ArticlePubMedGoogle Scholar