Background

Celiac disease (CeD) is an immune-mediated enteropathy that develops in genetically predisposed individuals as a reaction to gluten ingestion [1]. The global prevalence of CeD is 1.4% according to serological diagnosis, and 0.7% based on biopsy confirmation [2]. As in other autoimmune diseases, the prevalence of CeD is significantly higher in women [3]. Almost all CeD patients carry the Human Leucocyte Antigen (HLA) alleles that encode the HLA-DQ2 and/or HLA-DQ8 molecules. However, HLA is necessary, but not sufficient to develop the disease, and only explains 40% of the overall genetic risk [4, 5]. GWAS and fine-mapping studies like the Immunochip project have identified more than 40 non-HLA loci associated with CeD [6,7,8]. Nevertheless, most of the SNPs located in these loci either map to non-coding regions of the genome, far away from genes, or are in strong linkage disequilibrium (LD) with other associated variants, making it difficult to identify the genes that are functionally involved in the disease [9]. Altogether, HLA and non-HLA variants identified so far explain around 50% of CeD heritability [10]. The missing genetic heritability hypothesis suggests that additional, unidentified genetic and environmental factors are involved in the development of CeD. In this sense, the X chromosome has been historically ignored in most GWAS, or has been analyzed as if it were another autosome, without accounting for male hemizygosity and female X chromosome inactivation (XCI), with only very few studies that take these considerations into account [11,12,13].

Several studies have shown a relationship between the risk of different autoimmune diseases, including systemic lupus erythematosus (SLE), Sjögren syndrome, type 1 diabetes mellitus and CeD, and X chromosome aneuploidies like Klinefelter, Triple X or Turner syndromes [14,15,16,17]. Additionally, many of the approximately 1100 genes on the X chromosome are thought to be related to the immune function [18, 19]. These findings suggest a role for the X chromosome in the biased sex-prevalence of these conditions. In the case of CeD, most risk loci are located on the autosomes but X chromosome genes have also been identified, including TLR7 and TLR8, HCFC1, TMEM187 and IRAK1 [6, 7].

Monocytes are a fundamental part of the innate immune defense against microorganisms [20]. Different studies have also related this type of immune cells to CeD, and gliadin peptides stimulate their production of IL-8 and TNF-α, especially in celiac patients [21, 22]. It has been suggested that the response triggered by gliadin in monocytes is similar to that induced by lipopolysaccharide (LPS) through receptors such as TLR4 [23]. Fairfax et al. analyzed how genetic variants shape gene expression in monocytes, under different in vitro stimuli, including LPS, and showed that more than half of the monocyte eQTLs are specific of the environmental stimulus, but the X chromosome was not included in the analysis [24].

Mendelian randomization (MR), and more specifically the Summary-data-based MR (SMR), integrates GWAS and eQTL summary statistics in order to detect the functional involvement of genes under the GWAS peaks [25]. Particularly, association hits are translated into potentially causal relationships between expression levels of candidate genes and complex traits, in relevant tissues, cell types and context. Again, previous analyses in CeD that have made use of this method to combine GWAS and eQTL summary statistics have omitted the X chromosome [26, 27].

In this study, we hypothesized that the X chromosome could harbor additional susceptibility loci that could help explain both the missing heritability and the higher prevalence in women. Therefore, we aimed to identify genes on the X chromosome that might participate in the pathogenesis and also contribute to the sex bias in CeD, through their specific transcriptional profile in monocytes. For that purpose, we performed an X-chromosome association study (XWAS) in women from the Immunochip project, and constructed a database of X chromosome cis-eQTLs in female monocytes. Finally, we combined the two datasets using SMR in order to find monocyte-specific functional candidates on the X chromosome involved in CeD.

Materials and methods

Immunochip data and X chromosome association analyses

The CeD Immunochip dataset was filtered to include only X chromosome variants with genotyping rate > 95%, minor allele frequency (MAF) > 1% and P-value from Hardy Weinberg equilibrium (PHWE) test > 1 × 10–6 using PLINK1.9 [28]. We removed individuals with call rate < 97% and heterozygosity deviating more than 4 standard deviations from the mean (> 4SD). Genotypes were imputed at the Michigan Imputation Server [29] using HRC r1.1 2016 (GRCh37/hg19) as a reference panel, using only the European population, Minimac4 as the imputation software and phasing with Eagle v2.4. Imputed SNPs with an R2 imputation accuracy above 0.8, MAF > 1% and PHWE > 1 × 10–6 were kept. After applying these filters, 12,508 female samples (7062 cases and 5446 controls) and 9474 males (3712 cases and 5762 controls) and 1611 SNPs were retained. We calculated the first ten principal components of the genotypes using PLINK1.9 to control for potential population stratification in downstream analyses. The top associated SNP was replicated in the Dubois et al. study GWAS dataset [6].

The CeD XWAS was performed separately for men and women using the newml method implemented in SNPTEST (version 2.5.6), assuming a complete inactivation of one chromosome in females and equal effect size in both sexes [30]. Specifically, this method uses a logistic regression model encoding genotypes in males as 0/1 and in females as 0 / ½ / 1. The analysis was performed assuming an additive genetic model and the first ten principal components of the genotype data were included as covariates. The P-value threshold for statistical significance was set at P < 8.68 × 10–5 after Bonferroni correction according to the number of independent tests, as determined with the SimpleM method [31]. Results were plotted on a Manhattan plot generated with the qqman R package [32].

A gene-based association analysis was carried out in women using the FastBAT method available in the Genome-wide Complex Trait Analysis (GCTA) software package [33, 34]. This method integrates GWAS summary statistics and LD information to calculate the P-value of a set of variants within a preset distance from a gene. The analysis was performed using the default settings suggested by GCTA-FastBAT: gene regions extended 50 kb away from both the 3′ and 5′ UTRs of the genes, and SNPs in strong LD (r2 ≥ 0.9) were pruned. We conducted the analysis for 2393 genes and 1611 SNPs and the P-value significance threshold was set at P < 0.05 after False Discovery Rate (FDR) correction. A regional association plot combining XWAS and gene-based association results was generated with an open-source R script (https://github.com/Geeketics/LocusZooms).

Monocyte cis-eQTLs

A catalog of naïve and LPS-stimulated monocyte cis-eQTLs of the X chromosome was constructed in women using SNP genotype and expression data from a general population study by Fairfax et al. [24]. X chromosome variants and individuals were filtered and imputed as described above. Additionally, only those SNPs that were homozygous for the minor allele in at least 5 samples were retained. After the QC, 233 female samples and 165,648 variants remained for subsequent analyses.

Monocyte expression data that had already undergone data normalization, transformation with Variance Stabilizing Transformation (VST), batch effect correction and removal of outliers, were subjected to additional QC steps using the IluminaHumanv4.db [35] to remove those probes that matched more than one locus, those on the autosomes, Y chromosome or without chromosome reported, those containing SNPs with MAF > 0.1, and those described as Bad or No match. This resulted in 1347 probes mapping to 1250 stable gene IDs in the X chromosome.

We combined genotype (165,648 SNPs) and gene expression (1250 genes) information of 226 samples from the naïve monocyte female dataset using the QTLtools software [36]. The associations between SNP genotypes and gene expression levels were tested with simple linear regressions assuming normal distribution of the data, and the first 10 principal components of the genotype were included as covariates. Only SNPs within a 1-Mb window from the transcription start site (TSS) of a gene were analyzed. PeQTL values were calculated with the nominal pass option and a PeQTL < 5 × 10–8 threshold was set to identify significant results. The same analysis was performed with expression data from LPS-stimulated monocytes in women (n = 130).

SMR analyses

SMR was performed combining the summary statistics from the XWAS and the eQTLs of naïve and LPS-stimulated monocytes using the SMR software [25]. Briefly, SMR uses cis-eQTLs as instrumental variables, gene expression as the exposure, and CeD as the outcome, to infer genes pleiotropically or causally associated with CeD. SMR results were subjected to the heterogeneity in dependent instruments (HEIDI) test to detect the presence of LD. In this test, a significant P-value suggests that the association detected could be the result of two genetic variants in strong LD, whereas a non-significant P-value indicates that a single variant is associated with both gene expression and the disease. We used the following default parameters suggested by SMR: the cis-window was set at 2 Mb, the threshold PeQTL for the SMR analysis was set at 5 × 10–8, the threshold PeQTL for the HEIDI test was set at 1.57 × 10–3 and SNPs with a LD r2 > 0.9 or r2 < 0.05 with the top associated eQTL were pruned. SMR results with an FDR q-value < 0.05 and PHEIDI > 0.05, were considered as pleiotropic or causal associations. We plotted the TMEM187 region (2000 kb) on the X chromosome using the SMRLocusPlot script available in the SMR website.

Expression analysis in pediatric CeD patients

TMEM187 expression was quantified in peripheral blood mononuclear cells (PBMCs) from CeD children at onset (20 females, 8 males), patients on gluten-free diet (GFD) (6 females, 9 males), potential CeD patients (5 females, 6 males) and non-celiac controls (17 females, 10 males). CeD was diagnosed at the Pediatric Gastroenterology Unit of Cruces University Hospital. The study was approved by the Clinical Research Board of Cruces University Hospital. Samples (2.5 ml in EDTA-containing tubes) were collected after informed consent had been obtained from their parents or guardians and transferred to the Basque Biobank for Research. PBMCs were isolated using the MACSprep™ PBMC Isolation kit (Miltenyi Biotec SL, Madrid, Spain; cat. no. 130–115-169), RNA was purified using the NucleoSpin® RNA mini kit (Macherey–Nagel, Düren, Germany; cat. no. 740955.250 4,392,653) and stored at – 80 °C until use.

The expression of TMEM187 was quantified by RT-qPCR using the TaqMan RNA-to-Ct 1-Step Kit (Thermo Fisher Scientific Inc., Waltham, MA, USA; cat. no. 4392653) and a commercially available TaqMan Gene Expression assay (Thermo Fisher Scientific Inc., cat. no. Hs01920894_s1) on a Bio-Rad CFX Real Time PCR system (Bio Rad Scientific, Hercules, CA, USA). The housekeeping gene RPLPO was simultaneously measured and used as an endogenous control. Relative expression in each sample was calculated using the 2−ΔΔCt method and therefore, the expression of each sample was normalized to both the expression of RPLPO and the average of the controls. Differences in gene expression levels were analyzed with Mann–Whitney U-test using GraphPad Prism v.8.0.1. Finally, the expression of TMEM187 in different immune cell types (B, NK, T cells, and monocytes) was analyzed and sex-stratified expression plots were constructed using the default settings of the Database of Immune Cell Expression eQTL Epigenomics (DICE, https://dice-database.org/) online browser [37].

Results

XWAS and gene-level association analyses in CeD

To determine whether the sex bias in CeD prevalence is related to the X chromosome, we performed the XWAS of CeD in women and men separately (Additional file 1). We identified a single association peak on Xq28 including 13 significant SNPs (PXWAS < 8.68 × 10–5), with rs78237385 (PXWAS = 2.30 × 10–5, OR = 1.20 ± 0.10) as the top SNP (Additional file 2). The male XWAS did not detect any significant association (Additional file 3). The top SNP was also suggestively associated with CeD in females from the Dubois et al. study (P-value = 1.86 × 10–4) but was not significant (P-value = 0.19) in men from the same study (Additional file 4).

We also performed a gene-based association analysis in women, that takes into account the aggregated effect of sets of SNPs. We introduced 2392 genes and 1611 SNPs in the analysis and defined each gene region as ± 50 kb from both 3′ and 5′ UTRs. As a result, 276 association tests were performed and 25 candidate genes in the X chromosome were identified to be associated with CeD in women (FDR q-value < 0.05) (Additional file 5). Out of the 25 genes, 22 are novel genes associated with CeD, although ARHGAP4, RENBP, NAA10, AVPR2 or MECP2 had been previously identified in other autoimmune diseases [38,39,40]. XWAS and gene-level analysis results are summarized in Fig. 1.

Fig. 1
figure 1

Locus zoom plot 200 kb upstream and downstream of the XWAS top SNP. The XWAS top SNP, rs78237385 is represented with a purple diamond. In the top panel, the color schema represents the LD between the top SNP and the SNPs included in the region. In the bottom panel, the color schema represents the P-value from the gene-based association analysis

Full size image

Cis-eQTL analysis of the X chromosome in monocytes of women

In order to obtain an additional layer of information to interpret the XWAS signal, we calculated the X chromosome eQTLs in monocytes from women. We analyzed 226 samples with informative genotype and gene expression data from naïve monocytes, and performed multiple linear regressions between 165,648 SNPs and 1250 probes corresponding to 819 genes. Applying a threshold PeQTL value of 5 × 10–8, we identified 1097 cis-eQTLs involving 1054 SNPs, 19 probes and 18 independent genes (Additional file 6). The analysis in 130 LPS-stimulated female monocyte samples revealed 150 cis-eQTLs, corresponding to 94 SNPs, 9 probes and 8 genes (Additional file 7). Two genes (ZNF185 and TMEM187) were common to both situations, 16 were unique to naïve female monocytes, and 6 to female monocytes after LPS stimulation.

XWAS and cis-eQTL SMR and gene expression analyses

The SMR and HEIDI analyses of the summary statistics of the female CeD XWAS and the naïve monocyte cis-eQTL dataset identified two SNPs (rs7350355 and rs5945386) that were associated with two expression probes (ILMN_2198185 and ILMN_3242211, respectively) corresponding to the same gene, TMEM187, with a q-value < 0.05 and PHEIDI > 0.05. The minor alleles rs7350355*G and rs5945386*G are both the CeD risk alleles and were negatively correlated with TMEM187 expression (Fig. 2; Additional file 8). The SMR analysis was replicated with cis-eQTLs from LPS-stimulated female monocytes and revealed a single SNP (rs80208125) that was associated with the same two TMEM187 probes (Additional file 9; Additional file 10).

Fig. 2
figure 2

SMR locus plot of the results of the SMR analysis between the CeD XWAS and the naïve female monocyte eQTLs. In the top panel, grey dots represent − log10(P-values) for the female XWAS SNPs. Diamonds represent − log10(P-values) for probes from the SMR analysis and filled diamonds show those that pass the HEIDI test. In the middle panel, the red crosses represent − log10(P-values) for gene probes in the eQTL analysis. In the bottom panel, the location of the probes on the X chromosome is shown

Full size image

We investigated the expression of TMEM187 in PBMCs from female CeD patients at diagnosis, on GFD, potential CeD patients and non-celiac female controls. TMEM187 showed a significantly higher expression in active CeD patients compared to controls (P-value = 0.0417) and no differences were observed in potential CeD and GFD-treated individuals (Fig. 3A). We also studied TMEM187 expression in men but no significant differences were found (Fig. 3B). On the other hand, the expression of TMEM187 varied among different immune cell types (Additional file 11).

Fig. 3
figure 3

Results of the expression analysis. A Expression of TMEM187 in PBMCs of female samples. B Expression of TMEM187 in PBMCs of male samples. Both female and male samples were classified into four groups: non-celiac controls, celiac patients at onset, potential celiac patients and celiac patients on GFD represented by white, light grey, dark grey and black circles, respectively

Full size image

Discussion

The prevalence of CeD is significantly higher in women, as in the case of other autoimmune diseases [3]. A higher prevalence of immune-mediated disorders has also been observed in individuals with syndromes related to X chromosome aneuploidies [14, 17], suggesting an implication of the X chromosome in the risk of autoimmune diseases, probably also in the risk of CeD. In the current study, we have focused on gene expression in female monocytes in order to identify genes on the X chromosome that are involved in CeD and could explain the sex bias in the disease prevalence. For that purpose, we conducted a sex-specific XWAS in women from the Immunochip dataset and identified 25 genes associated with CeD in women, of which 22 have not been previously reported. Remarkably, 24 out of the 25 genes that were significant in the analysis are located on chromosome region Xq28, which has been previously associated with different autoimmune conditions such as SLE [39], rheumatoid arthritis [40], systemic sclerosis [38] or CeD [7].

On the other hand, it has been suggested that gliadin triggers an innate response in monocytes similar to that produced by LPS [23]. We performed an X chromosome cis-eQTL analysis in female naïve and LPS-stimulated monocytes and identified 6 eQTL-genes unique to LPS-stimulated female monocytes, some of which have been previously associated with autoimmune disorders, including PLXNA3 [41].

We then integrated the XWAS with cis-eQTL data from female naïve and LPS-stimulated monocytes using an MR approach. To our knowledge, this is the first SMR analysis of the X chromosome in CeD. We identified two SNPs (rs7350355 and rs5945386) that regulate the expression of TMEM187 in naïve monocytes and a different SNP (rs80208125) associated with TMEM187 in LPS-stimulated monocytes, suggesting that the genetic background may be more important than LPS stimulation in the regulation of TMEM187 expression. rs7350355 is a missense variant located within exon 2 of TMEM187, rs5945386 is an intergenic variant located 21 kb downstream of TMEM187, and rs80208125 is located on the 5’ UTR of TMEM187. All these variants exhibit strong LD (> 0.87) in the British population in England and Scotland, from the 1000 Genomes Project [42]. In all three cases, the CeD risk alleles (rs7350355*G, rs5945386*G and rs80208125*G) were the minor alleles (MAF around 0.2) and were negatively correlated with TMEM187 expression. TMEM187 encodes a multipass transmembrane protein of unknown function [43], and it has been proposed as a putative candidate gene in CeD together with IRAK1 and HCFC1, located in the same locus [7]. As far as we know, the only study that observed TMEM187 dysregulation in CeD is the one carried out by Pascual et al. in duodenal biopsies [44]. The TMEM187 locus has also been associated with other autoimmune disorders such as SLE or rheumatoid arthritis [39, 45], supporting the hypothesis of a shared genetic background in autoimmune disorders.

In our study, we interrogated the expression of TMEM187 in PBMCs from celiac patients at onset, celiac patients on GFD, potential celiac patients and non-celiac controls. PBMCs are a mixture of immune cells that contain monocytes (10–20%), together with lymphocytes (70–90%) and dendritic cells (1–2%), among others [46]. We observed a higher expression of TMEM187 in PBMCs from female pediatric patients at disease onset compared to non-celiac children. These results are consistent with a study published in 2016 by Pascual et al. that showed an upregulation of TMEM187 expression in biopsies of celiac adults [44]. The upregulation of TMEM187 in female CeD patients was not replicated in male PBMCs, suggesting possible role of TMEM187 in the sex bias of CeD that nevertheless needs to be confirmed with additional investigations.

The overexpression reported for TMEM187 in female CeD patients at onset is apparently contradictory to the fact that CeD risk alleles correlate with lower expression. This observed divergence could be due to different reasons: first, the CeD risk eQTLs could have an effect at the protein level, taking into account that rs7350355 is a missense variant that could alter the function of TMEM187, regardless of mRNA quantity. On the other hand, we are unable to definitively ascertain which of the eQTLs is the causal SNP, given the strong LD. Another reason could be that PBMCs contain a relatively modest proportion of monocytes, and the SNP could have different effects on the gene expression in other cell types, therefore explaining the apparent contradiction. The highly variable expression of TMEM187 in different immune cell types warrants further research on its role in the immune system. In addition, we have to bear in mind that the present expression analyses have been carried out in a pediatric cohort of diagnosed celiac children, while female donors in the monocyte expression study were non-celiac adults. It has been well reported that disease and immunogenic insult can sometimes surpass the genotypic effect, and lead to this kind of apparently contradictory situations [47]. Finally, Pascual et al. observed differences in the gene expression profile of susceptibility genes in CeD between children and adults, including TMEM187 [44].

Last but not least, it is worth mentioning that the main objective in the present work was not to identify SNPs with a functional involvement in CeD, nor to ascertain the mechanism by which they exert their function, but to highlight potentially causal genes that participate in the pathogenesis of the disease through their expression in monocytes. We carried out both the XWAS and the eQTL calculations with the aim of obtaining instruments to perform downstream analyses such as SMR, being aware that in our results SNPs will lose their relevance and will be replaced by functional candidate genes.

One limitation of this study is that both the lack of association as well as the absence of significant differential expression between CeD cases and controls observed in men could be due to the smaller sample size of the male cohort. This is partly a consequence of the higher incidence of CeD in women and an important factor to consider. However, we have studied the association of the top SNP of our female XWAS (rs78237385) in an independent dataset [6] and the P-values in men and women are 0.19 and 1.86 × 10–4, respectively. Additionally, the SNPs with significant results in the SMR analysis (rs5945386, rs7350355 and rs80208125) have P-values in women of 8.29 × 10–5, 4.66 × 10–3 and 3.34 × 10–3, respectively, while in men, they show P-values of 0.50, 0.49 and 0.57, respectively. We consider these P-values unlikely to become significant even with a higher number of male samples, and this lack of a significant association in men could imply a possible divergent mechanism of pathogenesis between sexes, that could explain the increased prevalence of CeD, and other autoimmune diseases observed in women [48].

Perspectives and significance

This is the first SMR approach in the X chromosome in CeD. We have identified TMEM187 as a candidate gene in CeD in monocytes and validated its differential expression in PBMCs from female CeD patients at onset. The fact that both the genetic association and the differential gene expression are not found in male patients suggests a role for TMEM187 in the sex bias observed in CeD. SMR appears as a useful approach to identify potentially causal genes under association peaks, including the X chromosome. Further studies are needed to identify the function of TMEM187 and understand its behavior in different cell types and disease status, and to clarify its role in CeD pathogenesis and the sex bias.