CORE Metadata, citation and similar papers at core.ac.uk

Provided by Elsevier - Publisher Connector

Genomics 98 (2011) 320–326

Contents lists available at ScienceDirect

Genomics

journal homepage: www.elsevier.com/locate/ygeno

Influence of sex and genetic variability on expression of X-linked in human monocytes

Raphaële Castagné a, Tanja Zeller b,1, Maxime Rotival a, Silke Szymczak c, Vinh Truong a, Arne Schillert c, David-Alexandre Trégouët a, Thomas Münzel b, Andreas Ziegler c, François Cambien a, Stefan Blankenberg b,1, Laurence Tiret a,⁎

a INSERM UMRS 937, Pierre and Marie Curie University (UPMC, Paris 6) and Medical School, 91 Bd de l'Hôpital 75013, Paris, France b II. Medizinische Klinik und Poliklinik, Johannes-Gutenberg Universität Mainz, Universitätsmedizin, Langenbeckstraße 1, 55131 Mainz, Germany c Institut für Medizinische Biometrie und Statistik, Universität zu Lübeck, Universitätsklinikum Schleswig-Holstein, Campus Lübeck, Ratzeburger Allee 160, 23538 Lübeck, Germany

article info abstract

Article history: In humans, the fraction of X-linked genes with higher expression in females has been estimated to be 5% from Received 12 April 2011 microarray studies, a proportion lower than the 25% of genes thought to escape X inactivation. We analyzed Accepted 28 June 2011 715 X-linked transcripts in circulating monocytes from 1,467 subjects and found an excess of female-biased Available online 6 July 2011 − transcripts on the X compared to autosomes (9.4% vs 5.5%, pb2×10 5). Among the genes not previously known to escape inactivation, the most significant one was EFHC2 whose 20% of variability was explained by Keywords: sex. We also investigated cis expression quantitative trait loci (eQTLs) by analyzing 15,703 X-linked SNPs. The X inactivation expression frequency and magnitude of X-linked cis eQTLs were quite similar in males and females. Few genes exhibited a Expression Quantitative Trait Locus stronger genetic effect in females than in males (ARSD, DCX, POLA1 and ITM2A). These genes would deserve Gender further investigation since they may contribute to sex pathophysiological differences. © 2011 Elsevier Inc. All rights reserved.

1. Introduction The exact mechanisms regulating expression status of genes on the inactive X are not fully elucidated. It is known that genes escaping In female mammals, one of the two X is inactivated inactivation are not uniformly distributed along the to maintain equal expression between sexes by a process called X but preferentially lie in the evolutionary younger strata of the X chromosome inactivation (XCI) [1]. XCI is a complex and tightly chromosome [5,6]. It has been hypothesized that long interspersed regulated mechanism governed by the XIST gene which encodes a elements (LINE-1), which are particularly abundant on the X non-coding functional RNA coating the chromosome from which it is chromosome and are enriched in the vicinity of inactivated genes, expressed and triggers inactivation [2]. Although most genes are might control spreading of XCI [7,8]. silenced by this process, it has long been recognized that some genes Although in vitro hybrid systems have greatly contributed to our

escape XCI and are expressed from both the active X (Xa) and inactive knowledge of XCI, they do not reflect the in vivo processes. In X(Xi) chromosomes. Studies using rodent-human hybrid cell lines particular, these experiments do not quantify the difference of retaining an inactive human X chromosome have estimated that expressionbetweenmalesandfemalesforgenesthatescape about 25% of X-linked genes escape inactivation to some degree [3–5]. inactivation. These genes are of primary interest since they are potential contributors to variability between sexes for phenotypic and clinical traits. Recent studies have taken benefit from the microarray Abbreviations: XCI, X Chromosome Inactivation; LCL, Lymphoblastoid Cell Line; technology to compare expression profiles of the X chromosome eQTL, Expression Quantitative Trait Locus; SNP, Single Nucleotide Polymorphism. between males and females [9–12]. The largest study, based on ⁎ Correspondingauthorat:INSERMUMRS937,FacultédeMédecinePitié- lymphoblastoid cell lines (LCLs) from 210 individuals from the four Salpêtrière, 91 bd de l'Hôpital, 75634 PARIS cedex 13, France. Fax: +33 1 40 77 97 28. HapMap populations, reported 5% of X-linked genes showing higher E-mail addresses: [email protected] (R. Castagné), [email protected] (T. Zeller), [email protected] (M. Rotival), [email protected] expression in females than in males [9], a proportion much lower than (S. Szymczak), [email protected] (V. Truong), [email protected] the 25% of genes previously thought to escape XCI [5,6]. This (A. Schillert), [email protected] (D.-A. Trégouët), [email protected] underestimation might be partly explained by an insufficient power (T. Münzel), [email protected] (A. Ziegler), [email protected] to detect small sex differences, since expression from the inactive (F. Cambien), [email protected] (S. Blankenberg), [email protected] (L. Tiret). chromosome may be low or present only in a fraction of females [5]. 1 Present address: Department of General and Interventional Cardiology, University Heart Center, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246 Here, we report the results of a large-scale study including 1467 Hamburg, Germany. population-based subjects in whom the expression profile from

0888-7543/$ – see front matter © 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.ygeno.2011.06.009 R. Castagné et al. / Genomics 98 (2011) 320–326 321 circulating monocytes was assessed [13].Wefirst examined to what 2.2. Expression level of X-linked transcripts according to sex extent the mechanism of XCI was supported by this dataset including 715 X-linked transcripts. Secondly, because the impact of X-linked For each of the 715 X-linked transcripts, the median and the genetic variability on expression may have profound consequences on variance of expression levels in males were plotted against those in disease traits, as attested by numerous X-linked diseases, we females (Fig. 1). Most of the data points lay on the diagonal investigated whether cis expression quantitative loci (eQTLs) on the suggesting that XCI was globally achieved for the majority of X chromosome differed between sexes. genes. There were however a few points departing from the diagonal, most of them showing a higher expression level in femalesthaninmales. 2. Results We then focused on the transcripts that were differentially expressed between sexes using a Bonferroni-corrected significance The study was conducted in 717 women and 750 men of European threshold to account for the number of transcripts tested (7.0×10−5). origin aged 35 to 74 years that were recruited in the Gutenberg Heart We distinguished transcripts that were expressed at a higher level in Study (GHS), a community-based project conducted in a single centre females (referred to as female-biased) from those that were in Germany [13]. Genome-wide expression profiles of peripheral expressed at a higher level in males (male-biased) [14]. We observed blood monocytes were generated using Illumina HT-12 BeadChip a higher proportion of female-biased than male-biased X-linked expression arrays and genome-wide genotyping was performed using transcripts (9.37% vs 5.59%, p=0.006). We tested whether these the Affymetrix Genome-Wide Human SNP Array 6.0. proportions were different from those observed on autosomes by resampling 106 sets of 715 autosomal transcripts. There was a significant excess of female-biased transcripts on the X chromosome 2.1. X-linked transcripts detected in human monocytes compared to autosomes (9.37% vs 5.46%, pb2×10−5) whereas the difference was not significant for male-biased transcripts (5.59% vs After removing probes with a bad quality score, 1156 probes were 4.65%, p=0.10). The higher proportion of female-biased genes is in unambiguously assigned to the X chromosome, corresponding to 706 X- agreement with the fact that a fraction of X-linked genes are known or linked genes. Analyses were performed at the probe level and, for suspected to escape from silencing [5,9–12]. simplicity, the term of “transcript” was used to denote a unique probe- Because the method of monocyte enrichment used in the present hybridization product (although in few cases the same transcript could study did not yield a 100% purity and even modest heterogeneity of be targeted by several probes). We focused on transcripts whose cell content may induce artefactual associations, we checked whether expression was detected, i.e. higher than the background signal, in≥5% contamination by non-monocyte cells might affect the differences of female samples or≥5% of male samples. These represented 715 X- observed. To estimate the relative amount of RNA coming from the linked (61.8%) transcripts corresponding to 503 unique genes (Supple- different blood cell types, we applied a deconvolution algorithm mentary Table 1). The majority of X-linked transcripts were detected proposed in [15] and used cell-specific expression profiles reported in with similar detection rates in males and females (Supplementary Fig. the HaemAtlas [16] to generate surrogate variables of contamination 1). XIST (X (inactive)-specifictranscript),thegeneresponsiblefor (see Methods). Adjusting for these variables hardly modified the inactivation of one of the X copies in females, was expressed at low difference between female-biased and male-biased X-linked genes levels, yet higher than the background signal, in 62% of males. (9.51% vs 5.73%, p=0.007).

Fig. 1. Comparison of expression levels of X-linked transcripts between males and females. (A) Median of expression. The pink line represents the situation where transcripts would completely escape XCI in females (log2Y=log2X+1). The only point departing from this line is XIST, the gene responsible for X inactivation of one chromosome in females. (B) Variance of expression. 322 R. Castagné et al. / Genomics 98 (2011) 320–326

The 54 female-biased genes (67 transcripts) and 35 male-biased chromosomes. For genes lying in PARs, escape from XCI is thought to genes (40 transcripts) of the X chromosome are listed in Supplementary be necessary for dosage compensation between the two female X Table 2. After adjustment for covariates (age, body mass index and chromosomes and the male X and Y. Of the 33 transcripts of the smoking) or exclusion of women with oral contraception or hormone expression array belonging to PARs, 22 were detected in monocytes replacement therapy, only a few genes lost significance, suggesting that corresponding to 14 unique genes, among which 8 had been the classical risk factors did not explain the majority of the sex-biased previously reported to escape XCI [5]. In the present study, five PAR expressions (Supplementary Table 2). Sex explained a much higher genes (35.7%) were found to be male-biased (DHRSX (dehydrogenase/ proportion of the inter-individual variability of expression in female- reductase SDR family X-linked), P2RY8 (purinergic receptor P2Y G- biased than in male-biased genes: in 20 of the 54 female-biased genes, coupled 8), PLCXD1 (phosphatidylinositol-specific phospho- the R2 attributable to sex was higher than 5% whereas it never exceeded lipase C, X domain containing 1), CD99 (CD99 antigen) and CSF2RA this threshold in male-biased genes. About half of the genes found to be (colony stimulating factor 2 receptor, alpha, low-affinity) (Supple- female-biased in our study had been previously reported to escape XCI to mentary Table 3). The proportion of male-biased genes in PARs found some degree [5]. Among the genes not previously known to escape XCI, here is higher than previously reported [9,17], probably owing to the EFHC2 (EF-hand domain (C terminal)-containing protein 2) was the greater power of the present study. Unlike the conclusion drawn from most differentially expressed between males and females, with sex a smaller study in LCLs [9], our study does not support the notion that explaining almost 20% of the inter-individual variability of expression. the majority of PAR genes are fully dosage compensated in females. Among the 35 male-biased genes, five had been previously reported to The lower expression levels in females could be explained by a partial escape XCI to some degree [5].Twoofthem,HCFC1 and MAGEE1,wereno silencing of the genes on the Xi copy or by an upregulation on the Y longer differentially expressed after adjustment for covariates (Supple- copy. mentary Table 2). By contrast, expression levels of SH3KBP1, USP11 and RBBP7 remained significantly higher in males, suggesting that these 2.4. Distribution of sex-biased genes on the X chromosome genes, even though they escape XCI to some extent, do not achieve full compensation in females. Higher levels in males might be explained by As previously reported [5,9,12], female-biased genes were not sex-influenced hormonal control not accounted for by the covariates. uniformly distributed along the X chromosome but preferentially We compared our results with those reported in LCLs from the four clustered on the short arm (16.1% of the transcripts lying on the short HapMap populations [9]. To get robust comparison, we selected the arm were female-biased versus 6.3% on the long arm, p=3.0×10−4), genes that were differentially expressed (pb0.001) in at least three more specifically in regions that more recently stopped recombining populations. No gene was reported as male-biased in LCLs, except with the Y chromosome (strata S3 to S5) (Fig. 2). PAR1 was the region CD99 which is actually located in the PAR1 pseudo-autosomal region the most enriched in male-biased genes. Although an enrichment in (see below). Among the 13 genes that were found female-biased in male-biased genes has been reported in stratum 3 compared to strata LCLs, all except two (FUNDC1 and USP9X) were also found female- 1 and 2 [17], the present data did not support such an evidence. biased in monocytes (Supplementary Table 2). By contrast we found 43 female-biased genes in monocytes that were not reported in LCLs. 2.5. Identification of cis eQTLs on the X chromosome The higher number of sex-biased genes in monocytes might be partly attributable to tissue-specificity of gene expression or different The total number of SNPs located within a 250 kb interval of any X- influence of sex hormones, but it is also likely a consequence of the linked gene (referred to as cis SNPs) was 15,703. After exclusion of greater sample size of the present study. transcripts probed by polymorphic sequences (n=89) and tran- scripts having no cis SNP (n=11), 615 X-linked transcripts (456 2.3. Expression of genes of the pseudo-autosomal regions (PARs) unique genes) were included in the analysis of cis eQTLs. The median number of cis SNPs per gene was 41 (interquartile range: 16–85). At both ends of the X chromosome, the PARs are regions of XY Association between SNP and expression was tested assuming an recombination where genes are essentially equivalent on the X and Y additive model. Given the specificity of the X chromosome with

Fig. 2. Difference of expression levels between females and males according to the position of transcripts on the X chromosome. PAR1 and PAR2 are the pseudo-autosomal regions. S1 to S5 represent strata that ceased to recombine with the Y chromosome at different times of evolutionary history (S1 is the most ancient). Female-biased transcripts are shown by pink dots and male-biased transcripts by blue dots. R. Castagné et al. / Genomics 98 (2011) 320–326 323 respect to sex, we analyzed males and females separately. The analysis M6B ). In females, departure from the hypothesis of additivity was of eQTLs on the X chromosome is complicated by the fact that, for a tested but never reached significance. same allele effect, the variance explained by a SNP under an additive To further identify cis eQTLs that were differentially regulated in model is expected to be twice lower in females than in males due to males and females, we modeled expression as a function of sex, the different number of X copies (see Appendix). For genes escaping genotype and the interaction between sex and genotype. When XCI, this assumption may not hold but these represent only a minority correcting for the number of genes tested, four genes exhibited a of X-linked genes. As a consequence, the power to detect cis significant interaction with sex (pb8×10−5): ARSD, DCX (doublecor- associations is lower in females than in males for a same sample tin), POLA1 (DNA polymerase alpha 1 catalytic subunit) and ITM2A size. To circumvent this problem, we used the R2 (proportion of (integral protein membrane 2A). Two additional genes, OCRL variance explained by a SNP) rather than the p-value to compare the (responsible of the oculocerebrorenal syndrome of Lowe) and PLS3 frequency of cis eQTLs between males and females, after multiplying (plastin3), showed suggestive evidence for sex interaction (pb10−3) the female R2 by a correction factor to account for the difference of (Supplementary Table 4). In all cases except ARSD, the cis regulatory variances between sexes. Assuming that residual variances are equal effect was significant only in females. For ARSD, the cis effect was in both sexes (a reasonable assumption according to Fig. 1), the observed in both sexes but the allele effect of rs211653 on expression 2 2 correction factor in females is equal to 2/(R XF +1) where R XF is the was more marked in females than in males (β=−0.114±0.008 vs R2 explained by an X-linked SNP in females (see Appendix). −0.076±0.005, p =1.2×10−5) and after correction, the SNP All associations with an R2 N5% in either sex were considered. This explained 37.0% of the variability of ARSD expression in females corresponded to taking a significance threshold of 5.8×10−10 in versus 26.3% in males (Fig. 4). −5 2 males and 9.9×10 in females (or a R XF N2.56% in females before correction). Using this threshold, 71 transcripts were associated with a cis SNP in females (11.5% of the 615 X-linked transcripts) and 56 3. Discussion (9.1%) in males. For genes represented by several SNP-transcript pairs, we retained the SNP-transcript pair associated with the highest R2. In the present study, we found that nearly 10% of X-linked genes This resulted in 60 (13.2%) and 48 (10.5%) X-linked genes that were were female-biased, a proportion higher than previously reported in cis regulated in females and males, respectively (p=0.21). smaller microarray-based studies [9–12]. Female-biased genes were In order to compare cis eQTLs in males and females, we retained for more frequent on the X chromosome than on autosomes, a finding each expression trait the SNP having the smallest combined p-value of consistent with the feminization of the X chromosome during association when combining males and females. Among the 64 cis mammalian evolution [17]. Most of the genes previously known or eQTLs detected on the X chromosome, 44 were common to both sexes, suspected to escape XCI had a female-biased expression in our study. 16 were female-specific and 4 were male-specific (Supplementary In addition to the genes already known, we identified several new Table 4). The R2 of cis eQTLs was roughly comparable in males and genes that were associated with higher expression in females. For females (Fig. 3 and Supplementary Table 4) and allele effects were in most of them, the difference between sexes was relatively small, the same direction in males and females. Six genes had more than 25% suggesting that expression from the Xi copy was reduced compared to of the variability of their expression explained by cis regulation in that from the Xa copy, or that escape from XCI occurred only in a both sexes (ranked by combined p-value: GPR34 (G protein-coupled fraction of females as previously suggested [5]. On the other hand, an receptor 34), TMEM187 (transmembrane protein 187), ARSD (aryl- alternative explanation for the higher female expression could be that sulfatase D), MAOA (monoamine oxidase A), CHST7 (carbohydrate N- acetylglucosamine 6-O sulfotransferase 7) and GPM6B (glycoprotein

Fig. 3. Cis eQTLs of the X chromosome: proportion of variability of expression (R2) explained by the best cis SNP in males and in females. Cis eQTLs associated with an Fig. 4. Box plots of ARSD expression levels according to rs211653 genotype and sex. R2 N5% in either gender were selected (see Supplementary Table 4). Numbers of subjects in each genotype are indicated on the left of the box plots. 324 R. Castagné et al. / Genomics 98 (2011) 320–326

these genes are over-expressed from the female Xa chromosome for defects in neuronal migration [26] and associated diseases such as compared to the male X chromosome, due to sex influenced hormonal subcortical laminar heteropia (SCLH) [27] or chronic refractory control or to environmental factors. For these newly identified genes, temporal lobe epilepsy [28]. Little is known about the POLA1 and the exact mechanism explaining the over-expression in females might ITM2A genes, although the latter one has been suggestively associated be studied using a cell-hybrid assay system that compares expression with height in a meta-analysis [29]. between the Xa and the Xi female chromosomes or between the The present findings open new perspectives of research, such as female Xa chromosome and the male X chromosome. investigating the sex differences in other tissues, confirming by Among the new female-biased genes, EFHC2 deserves special appropriate cell-hybrid functional assays that some of the newly attention as almost 20% of expression variability of this gene was identified sex-biased genes do escape XCI and determining whether explained by sex. EFHC2 is located at Xp11.3 close to a locus sex differences in expression levels translate at the protein level. The implicated in Turner syndrome, a sporadic disorder of women in sex-biased genes identified in monocytes provide a source of which all or part of one chromosome X is deleted [18]. This locus is candidate genes to explain differences for phenotypic traits and known to be subject to maternal imprinting and X inactivation [18] susceptibility to disease between males and females, and among and the higher female expression of EFHC2 in the present study females that are heterogeneous with respect to X inactivation. Finally, suggests that the region submitted to XCI and imprinting might the X-linked cis eQTLs identified here provide a list of genes whose extend up to EFHC2. The EFHC2 gene has been suspected to be expression may mediate the association between X-linked suscepti- involved in fear recognition in women with Turner syndrome [19] bility loci and disease. although this finding has not been further confirmed [20]. The proportion of X-linked male-biased genes was lower than that 4. Conclusions of female-biased genes and was not different from the proportion observed on autosomes. As recently suggested by the evolutionary This study suggests that escape from XCI may be more frequent history of the mammalian X chromosome, X-linked male-biased than previously estimated from population studies and may contrib- genes would be younger genes whereas their autosomal counterparts ute to phenotypic differences between sexes. We also showed that for would be older genes [17]. This is consistent with a progressive the majority of X-linked genes, the impact of genetic variability on cis demasculinization of the X chromosome during mammalian evolution regulation of expression was not substantially different between and the out-of-X traffic of male-biased genes resulting in an sexes, with a few exceptions which might be clinically meaningful. enrichment of these genes on autosomes in the long term [17].It Finally, we indicate how to appropriately correct for the difference of would be interesting to test this hypothesis by dating the sex-biased genetic variability explained by SNPs in the comparison of phenotypic genes found in the present study. effects between sexes. Despite a growing interest for the genetics of gene expression in recent years [21], little is known about the consequence of the allele 5. Subjects and methods dosage imbalance of X-linked SNPs on expression. For loci that are subject to XCI in females, which are the vast majority as seen above, The study has been described in details elsewhere [13]. the proportion of variance attributable to X-linked SNPs is expected to be lower in females than in males [22]. To restore the balance between 5.1. Study population sexes, we derived and applied a correction factor to the female R2. After this correction, the magnitude of most cis eQTLs appeared Study participants of both sexes aged 35–74 years were succes- remarkably similar in both sexes. The frequency of cis eQTLs was sively enrolled into the Gutenberg Heart Study (GHS), a community- slightly higher in females than in males, which might be partly based cohort study based in the Rhein-Main region in western mid- explained by a higher rate of false positives due to the less stringent Germany. All subjects were of European origin and gave written statistical threshold applied to females to be comparable to males. informed consent. Ethical approval was given by the local ethics Due to the large number of cis SNPs per gene and the great power of committee and by the local and federal data safety commissioners. the present study for identifying cis eQTLs [13], it is unlikely that we Individuals for whom we found a discrepancy between the pheno- missed important X-linked cis eQTLs in monocytes. However, the true typic gender and the sex inferred from expression of Y-linked magnitude of eQTLs might be underestimated if the causal variant was transcripts were excluded, leaving 1467 individuals for analysis. not directly genotyped but was in linkage disequilibrium with SNPs on the array. 5.2. Genome-wide expression Four cis eQTLs, ARSD, DCX, POLA1 and ITM2A, exhibited a significant heterogeneity between sexes, all four having a stronger effect in Separation of monocytes was conducted within 60 min after blood females. Notably, ARSD and ITM2A were among the 54 female-biased collection by negative selection using RosetteSep Monocyte Enrich- genes detected in the present study and previously reported as ment Cocktail (StemCell Technologies, Vancouver, Canada). Total RNA escaping XCI [5]. For these genes, the stronger genetic effect in was extracted the same day using Trizol extraction and purification by females might result from a partial or complete activation of the allele silica-based columns. Expression profiles were assessed using the carried by the chromosome escaping XCI. The ARSD gene belongs to a Illumina HT-12 v3 BeadChip. Data were normalized using VST cluster of sulfatase genes that is involved in chondrodysplasia transformation and quantile normalization as implemented in the punctata (CDPX), a severe X-linked skeletal dysplasia [23]. These lumi R package [30,31]. Only probes with a “perfect” or “good” match sulfatase genes are also suspected to be involved in the metabolism of according to ReMOAT [32] were selected and probes with an warfarin, an anticoagulant widely used for preventing thrombosis unknown or ambiguous chromosomal position were discarded, [23]. Interestingly, ARSD has been recently shown to belong to a leaving 1156 probes on the X chromosome. Of these, we selected network of genes trans-regulated by the maternally imprinted KFL14 probes that had a detection p-valueb0.05 in ≥5% of males or ≥5% of locus [24]. Parental imprinting might then be an alternative females (n=715). explanation to the differential genetic effect observed between males and females. Functional assays are required to investigate the 5.3. Genome-wide genotyping potential mechanisms at the origin of this difference. The DCX gene belongs to a superfamily of involved in Genotyping was performed using the Affymetrix Genome-Wide signal transduction and cytoskeletal regulation [25] and is responsible Human SNP Array 6.0 and the Genome-Wide Human SNP NspI/StyI 5.0 R. Castagné et al. / Genomics 98 (2011) 320–326 325

Assay kit. Genotypes were called using the Affymetrix Birdseed-V2 Gutenberg Heart Study. The present study was supported by the calling algorithm and quality control was performed [33]. SNPs with a National Genome Network “NGFNplus” (contract A3 01GS0833 and minor allele frequencyb0.01 or deviating from Hardy–Weinberg 01GS0831) and by a joint funding from the Federal Ministry of equilibrium (pb10−4) were excluded. Education and Research, Germany (contract BMBF 01KU0908A) and from the Agence Nationale de la Recherche, France (contract ANR 09 5.4. Statistical analysis GENO 106 01) for the project CARDomics. Statistical analysis benefited from the C2BIG platform supported by the Fondation pour Analysis was performed at the probe level. Difference of mean la Recherche Médicale and the Région Ile-de-France. expression levels between males and females was tested by t-test with sex-specific variances. A Bonferroni threshold was used to Appendix A. Comparison of the genetic R2 in females and males for correct for the number of probes tested (pb7×10−5). To test whether X-linked SNPs the proportion of sex-biased transcripts on the X chromosome significantly differed from that on autosomes, we randomly re- We consider an X-linked SNP with allele effect β on expression. The sampled 106 sets of 715 autosomal transcripts and assessed the allele effect is assumed to be equal in males and in females. If the gene is proportion of these in which the proportion of sex-biased transcripts submitted to XCI, female homozygotes are expected to have the same was higher than the one observed on the X chromosome. expression level as male hemizygotes while female heterozygotes have Contamination of the samples by RNA derived from other blood a level increased by β/2 by comparison to reference homozygotes since cell types than monocytes was assessed by estimating the half of their cells carry an active allele and half carry an inactive allele. proportion of each cellular type using the deconvolution algorithm The genetic variance explained by the SNP is therefore twice higher in proposed by Abbas et al. [15]. To estimate the expression profiles males than in females, that is VXM =2VXF where VXM is the genetic characterizing the specific blood lineages, we used the publicly variance in males and VXF is the genetic variance in females. available HaemAtlas dataset [16]. Applying this method, we found If the residual variance Vres is the same in both sexes, we have: that the samples contained an average of 83.4% of monocyte RNA – 2 = ðÞ= ðÞ (95% coverage interval [69.9 93.4]). Other cell types included RXM =VXM VXM +Vres =2VXF 2VXF +Vres and platelets, T lymphocytes and Natural Killer cells, B lymphocytes R2 =V = ðÞV +V (~5% each) and traces of RNA from erythrocytes and granulocytes XF XF XF res b ( 2%). Differences of expression according to sex were re-tested 2 where R XM is the proportion of variance of expression explained by after adjustment on the proportions of the different cell types 2 the X-linked SNP in males and R XF is the proportion in females. From estimated for each individual. 2 2 2 these formulae we can deduce that R XM =2 R XF/(R XF +1). Sex difference of X-linked gene expression was further tested after 2 For purpose of comparison between males and females, R XF was exclusion of women with oral contraception (n=52) or women with 2 hormone replacement therapy (n=79), and after adjustment for age, then multiplied by a correction factor equal to 2/(R XF +1). body mass index and smoking status (smoker vs non-smoker). For the analysis of cis eQTLs, all SNPs located within 250 kb of Appendix B. Supplementary data either the transcription starting site (TSS) or the transcription ending site (TES) of any X-linked transcript were selected (15,703 cis SNPs). Supplementary data to this article can be found online at doi:10. Probes with a SNP within their sequence were excluded from this 1016/j.ygeno.2011.06.009. analysis to avoid any bias due to a difference of binding between the probe and its transcript. References Association between expression and SNP was tested by linear regression model with genotype coded (0, 2) in males and (0, 1, 2) in [1] M.F. Lyon, Gene action in the X-chromosome of the mouse (Mus musculus L.), – females according to the number of minor alleles. This model assumed Nature 190 (1961) 372 373. [2] J. Chow, E. Heard, X inactivation and the complexities of silencing a sex additive allele effects. Departure from additivity was tested in females chromosome, Curr. Opin. Cell Biol. 21 (2009) 359–366. by comparing the additive model to the general model (χ2 with 1 df). [3] C.J. Brown, L. Carrel, H.F. Willard, Expression of genes from the human active and – For every expression trait, a p-value of association was computed for inactive X chromosomes, Am. J. Hum. Genet. 60 (1997) 1333 1343. [4] L. Carrel, A.A. Cottle, K.C. Goglin, H.F. Willard, A first-generation X-inactivation profile each cis SNP-transcript pair, separately in males and females. We of the human X chromosome, Proc. Natl. Acad. Sci. U. S. A. 96 (1999) 14440–14444. focused on cis eQTLs explaining N5% of expression variability in either [5] L. Carrel, H.F. Willard, X-inactivation profile reveals extensive variability in X- males or females. In females, the genetic R2 was corrected to account linked gene expression in females, Nature 434 (2005) 400–404. 2 [6] Y. Yasukochi, O. Maruyama, M.C. Mahajan, C. Padden, G.M. Euskirchen, V. Schulz, for the difference of genetic variance between sexes and make the R et al., X chromosome-wide analyses of genomic DNA methylation states and gene statistics comparable between males and females (see Appendix). expression in male and female neutrophils, Proc. Natl. Acad. Sci. U. S. A. 107 Homogeneity of allele effects between males and females was tested (2010) 3704–3709. [7] L. Carrel, C. Park, S. Tyekucheva, J. Dunn, F. Chiaromonte, K.D. Makova, Genomic by introducing a sex×genotype interaction in the linear regression environment predicts expression patterns on the human inactive X chromosome, model, with genotype coded as outlined above. We computed a PLoS Genet. 2 (2006) e151. combined p-value for males and females using Fisher's method and [8] J.C. Chow, C. Ciaudo, M.J. Fazzari, N. Mise, N. Servant, J.L. Glass, et al., LINE-1 – activity in facultative heterochromatin formation during X chromosome inacti- selected for each expression trait the SNP transcript pair with the vation, Cell 141 (2010) 956–969. lowest combined p-value. [9] C.M. Johnston, F.L. Lovell, D.A. Leongamornlert, B.E. Stranger, E.T. Dermitzakis, All analyses were performed in R v. 2.12.0 [34]. M.T. Ross, Large-scale population study of human cell lines indicates that dosage compensation is virtually complete, PLoS Genet. 4 (2008) e9. [10] I.W. Craig, J. Mill, G.M. Craig, C. Loat, L.C. Schalkwyk, Application of microarrays to Acknowledgments the analysis of the inactivation status of human X-linked genes expressed in lymphocytes, Eur. J. Hum. Genet. 12 (2004) 639–646. [11] Z. Talebizadeh, S.D. Simon, M.G. Butler, X chromosome gene expression in human The Gutenberg Heart Study is funded through the government of – “ ” tissues: male and female comparisons, Genomics 88 (2006) 675 681. Rheinland-Pfalz ( Stiftung Rheinland Pfalz für Innovation , contract [12] A.F. McRae, N.A. Matigian, L. Vadlamudi, J.C. Mulley, B. Mowry, N.G. Martin, et al., AZ 961-386261/733), the research programs “Wissen schafft Zukunft” Replicated effects of sex and genotype on gene expression in human lympho- and “Schwerpunkt Vaskuläre Prävention” of the Johannes Gutenberg- blastoid cell lines, Hum. Mol. Genet. 16 (2007) 364–373. [13] T. Zeller, P. Wild, S. Szymczak, M. Rotival, A. Schillert, R. Castagne, et al., Genetics University of Mainz and its contract with Boehringer Ingelheim and and beyond—the transcriptome of human monocytes and disease susceptibility, PHILIPS Medical Systems including an unrestricted grant for the PLoS ONE 5 (2010) e10693. 326 R. Castagné et al. / Genomics 98 (2011) 320–326

[14] H. Ellegren, J. Parsch, The evolution of sex-biased genes and sex-biased gene [26] J.S. Liu, Molecular genetics of neuronal migration disorders, Curr. Neurol. expression, Nat. Rev. Genet. 8 (2007) 689–698. Neurosci. Rep. 11 (2011) 171–178. [15] A.R. Abbas, K. Wolslegel, D. Seshasayee, Z. Modrusan, H.F. Clark, Deconvolution of [27] V. des Portes, F. Francis, J.M. Pinard, I. Desguerre, M.L. Moutard, I. Snoeck, et al., blood microarray data identifies cellular activation patterns in systemic lupus doublecortin is the major gene causing X-linked subcortical laminar heterotopia erythematosus, PLoS ONE 4 (2009) e6098. (SCLH), Hum. Mol. Genet. 7 (1998) 1063–1070. [16] N.A. Watkins, A. Gusnanto, B. de Bono, S. De, D. Miranda-Saavedra, D.L. Hardie, [28] L. D'Alessio, H. Konopka, E.M. López, E. Seoane, D. Consalvo, S. Oddo, et al., et al., A HaemAtlas: characterizing gene expression in differentiated human blood Doublecortin (DCX) immunoreactivity in hippocampus of chronic refractory cells, Blood 113 (2009) e1–e9. temporal lobe epilepsy patients with hippocampal sclerosis, Seizure 19 (2010) [17] Y.E. Zhang, M.D. Vibranovski, P. Landback, G.A.B. Marais, M. Long, Chromosomal 567–572. redistribution of male-biased genes in mammalian evolution with two bursts of [29] D.F. Gudbjartsson, G.B. Walters, G. Thorleifsson, H. Stefansson, B.V. Halldorsson, P. gene gain on the X chromosome, PLoS Biol 8 (2010). Zusmanovich, et al., Many sequence variants affecting diversity of adult human [18] D.H. Skuse, R.S. James, D.V. Bishop, B. Coppin, P. Dalton, G. Aamodt-Leeper, et al., height, Nat. Genet. 40 (2008) 609–615. Evidence from Turner's syndrome of an imprinted X-linked locus affecting [30] P. Du, W.A. Kibbe, S.M. Lin, lumi: a pipeline for processing Illumina microarray, cognitive function, Nature 387 (1997) 705–708. Bioinformatics 24 (2008) 1547–1548. [19] L.A. Weiss, S. Purcell, S. Waggoner, K. Lawrence, D. Spektor, M.J. Daly, et al., [31] S.M. Lin, P. Du, W. Huber, W.A. Kibbe, Model-based variance-stabilizing Identification of EFHC2 as a quantitative trait locus for fear recognition in Turner transformation for Illumina microarray data, Nucleic Acids Res. 36 (2008) e11. syndrome, Hum. Mol. Genet. 16 (2007) 107–113. [32] N.L. Barbosa-Morais, M.J. Dunning, S.A. Samarajiwa, J.F.J. Darot, M.E. Ritchie, A.G. [20] A.R. Zinn, H. Kushner, J.L. Ross, EFHC2 SNP rs7055196 is not associated with fear Lynch, et al., A re-annotation pipeline for Illumina BeadArrays: improving the recognition in 45, X Turner syndrome, Am. J. Med. Genet. B Neuropsychiatr. Genet. interpretation of gene expression data, Nucleic Acids Res. 38 (2010) e17. 147B (2008) 507–509. [33] A. Ziegler, Genome-wide association studies: quality control and population- [21] W. Cookson, L. Liang, G. Abecasis, M. Moffatt, M. Lathrop, Mapping complex based measures, Genet Epidemiol. 33 (Suppl 1) (2009) S45–S50. disease traits with global gene expression, Nat. Rev. Genet. 10 (2009) 184–194. [34] Development Core Team, R: A Language and Environment for Statistical [22] D.G. Clayton, Sex chromosomes and genetic association studies, Genome Med. 1 Computing, R Foundation for Statistical Computing, Vienna, Austria3-900051- (2009) 110. 07-0, 2005. [23] B. Franco, G. Meroni, G. Parenti, J. Levilliers, L. Bernard, M. Gebbia, et al., A cluster of sulfatase genes on Xp22.3: in chondrodysplasia punctata (CDPX) and implications for warfarin embryopathy, Cell 81 (1995) 15–25. [24] K.S. Small, A.K. Hedman, E. Grundberg, A.C. Nica, G. Thorleifsson, A. Kong, et al., Identification of an imprinted master trans regulator at the KLF14 locus related to WEB REFERENCES multiple metabolic phenotypes, Nat. Genet. 43 (2011) 561–564. [25] O. Reiner, F.M. Coquelle, B. Peter, T. Levy, A. Kaplan, T. Sapir, et al., The evolving The URLs for data presented herein are as follows:Reannotation and Mapping of doublecortin (DCX) superfamily, BMC Genomics 7 (2006) 188. Oligonucleotide Arrays Technologies (ReMOAT), http://remoat.sysbiol.cam.ac.uk/.