Chromosome X-wide association study in multiple system atrophy identifies sex-differential risk loci

Abstract

The human X chromosome accounts for ∼5% of the genome. Despite its size, the role of X-chromosomal variants in human diseases is not well understood, mainly due to its distinct inheritance pattern and the frequent omission of sex chromosomes from genome-wide association studies.

This study used whole-genome sequencing data from 888 multiple system atrophy (MSA) cases and 7128 controls to perform an X-chromosome-wide common variant association study. Our analyses revealed two sex-differential risk loci: one located at Xq28, associated with increased risk for MSA in females [index variant: rs4898389, odds ratio (OR) = 1.69, 95% confidence interval (CI) = 1.40–2.03, P-value = 2.43 × 10⁻⁸], and another at Xp22.2 within the TBL1X gene, identified in males (index variant: rs6638956, OR = 1.48, 95%CI = 1.26–1.73, P-value = 9.92 × 10⁻⁷). In the Xq28 locus, colocalization analyses pointed to FAM3A and PLXNA3 as genes of interest.

These findings emphasize the vital role of sex-differential genetic factors in the pathogenesis of MSA.

Ray et al. identify sex-differential X-chromosomal risk loci for multiple system atrophy, suggesting distinct genetic mechanisms in males and females and highlighting the importance of considering sex chromosomes in adult-onset neurological disorders.

Introduction

Multiple system atrophy (MSA) is a progressive neurodegenerative disease characterized by dysautonomia, parkinsonism, cerebellar ataxia and pyramidal signs.¹ Pathologically, it is classified as a synucleinopathy, marked by extensive α-synuclein-positive oligodendroglial cytoplasmic inclusions and neurodegeneration of the striatonigral and olivopontocerebellar regions of the brain.¹ Despite ongoing research, our understanding of its pathogenesis remains limited, and no disease-modifying treatments are currently available for this fatal condition.¹

Identifying the genetic factors contributing to adult-onset diseases enhances our understanding of the underlying pathophysiology and may lead to improved diagnostic and treatment strategies. Previous candidate gene and genome-wide association studies (GWAS) in MSA have highlighted genes linked to Parkinson’s disease, oxidative stress and inflammation.^2,3 Our recent study, which included whole-genome sequencing data from 888 MSA cases and 7128 controls, identified four genome-wide significant risk loci on autosomal chromosomes.⁴ However, this research did not incorporate the X chromosome, which is enriched for genes expressed in the brain.^5-7 Studies focusing on the X chromosome have uncovered significant risk loci in other synucleinopathies, such as Parkinson’s disease and Lewy body dementia,^8,9 highlighting the need for a focused examination of the X chromosome in MSA.

Here, we describe an X-chromosome-wide association study (XWAS) of MSA using whole-genome sequencing data from a European-ancestry case-control cohort (Fig. 1). Our research identified two sex-differential susceptibility loci. Following the association analyses, we validated our significant findings using long-read sequencing and performed colocalization analyses to prioritize transcriptionally dysregulated genes within the identified risk regions.

Figure 1.

An X-chromosome-wide association study of multiple system atrophy using whole-genome sequencing data from a European-ancestry case-control cohort. Whole-genome sequence data from 888 multiple system atrophy (MSA) cases and 7128 controls were studied. An X-chromosome-wide association study (XWAS) was performed for female-only, male-only and a combined cohort of both sexes. This investigation uncovered two sex-differential signals, one present in females and one in males. Functional genomic investigations were undertaken to prioritize genes of interest within the novel MSA susceptibility loci.

Materials and methods

Study participants and data generation

The study cohort included 888 MSA cases (n = 416 females and n = 472 males) and 7128 unrelated controls (n = 3741 females and n = 3387 males). Participants were recruited from 20 sites/consortia across Europe and North America, as described elsewhere.⁴ The MSA cohort comprised 468 pathologically confirmed cases (47.0%, n = 225 females and n = 243 males) and 420 clinically diagnosed cases (53.0%) who met the Gilman consensus criteria for clinically probable MSA.¹⁰ Detailed information about the case-cohort composition has been described elsewhere.⁴

The controls were derived from two separate cohorts: (i) the DementiaSeq project control cohort, including 1612 females and 1406 males of European ancestry who showed no cognitive decline or neurological deficits on neurological examination (dbGaP accession number: phs001963.v3.p1)^4,11; and (ii) a convenience control cohort consisting of 4110 individuals of European ancestry (2129 females and 1981 males), obtained from the TOPMed Consortium (dbGaP accession numbers: phs001662.v2.p1, phs00974.v5.p4, phs000951.v5.p5).⁴

All study participants had undergone whole-genome sequencing using an Illumina next-generation sequencer, which utilized 150 bp, paired-end sequencing reads at 35× coverage aligned to human genome build 38 (GRCh38DH). The study was approved by the institutional review boards of all participating institutions, and informed consent was obtained from all subjects or their surrogate decision-makers, in accordance with the Declaration of Helsinki.

Quality control

Figure 1 summarizes the study design. The analysis included genome sequence data from 888 MSA cases and 7128 controls. The reference genome GRCh38DH was used for sequence alignment, and variant calling was performed using the GATK Best Practices, as described elsewhere.⁴ In this pipeline, the Y chromosome remains unmasked during alignment for all samples, irrespective of their sex. The pipeline does not require sex information; however, we adjusted ploidy settings based on inferred genetic sex: haploid (ploidy = 1) for males and diploid (ploidy = 2) for females.

Sample-level quality control filtering was conducted according to our previous study on autosomal variants.⁴ Samples were excluded based on the following quality metrics: (i) a high contamination rate (>5% as indicated by the VerifyBamID freemix metric); (ii) excessive heterozygosity (F-statistic outside of the −0.15 to 0.15 range); (iii) a low call rate (≤95%); (iv) inconsistencies between reported and genotypic sex; (v) duplicate samples (identified by pi-hat statistics >0.8); (vi) individuals of non-European ancestry as determined by principal components analysis relative to the HapMap 3 Genome Reference Panel; and (vii) closely related samples (identified by pi-hat statistic >0.125, with one member of each pair excluded). After quality control filtering, 888 MSA cases (n = 416 females and n = 472 males) and 7128 controls (n = 3741 females and n = 3387 males) were available for analysis.

Next, we conducted chromosome-X-specific variant-level quality control checks separately for females and males. We excluded: (i) variants with non-random missingness between cases and controls (P-value ≤ 1.0 × 10⁻⁴); (ii) variants showing haplotype-based non-random missingness (P-value ≤ 1.0 × 10⁻⁴); and (iii) variants with an overall missingness rate of ≥2%. After merging the male and female datasets, we further excluded: (i) variants with >5% difference in minor allele frequency (MAF) between male and female controls; (ii) variants with >5% difference in missingness rates between male and female controls; (iii) variants located in the pseudoautosomal regions (PARs) (chrX:10,001–2,781,479 for PAR1 and chrX:155,701,383–156,030,895 for PAR2) due to their unique recombination patterns and challenging mapping and variant calling; (iv) variants that did not pass the Hardy–Weinberg equilibrium exact test (P-value < 1.0 × 10⁻⁴) in female controls; (v) variants with MAFs significantly different in controls from the reported frequencies in the National Heart, Lung, and Blood Institute (NHLBI) TOPMed database (freeze 5b; https://topmed.nhlbi.nih.gov) or gnomAD database (version 3.1.2; https://gnomad.broadinstitute.org); (vi) variants in the centromeric regions ± 10 kb (due to poor sequence alignment and incomplete resolution of the reference genome assembled at these sites); (vii) variants failing gnomAD filters (version 3.1.2; https://gnomad.broadinstitute.org); (viii) variants with significantly different MAFs between the DementiaSeq control genomes and the TOPMed control genomes [excluding variants with per chromosome false discovery rate (FDR)-corrected P-value ≤ 0.05]; (ix) all small insertions or deletions (<50 bp compared to the reference allele); and (x) variants with poor sequence alignment.

Mapping quality scores and alignment accuracy were assessed to ensure high mapping quality scores at the index variants, and the absence of systematic coverage differences, supported by sex-adjusted ploidy settings and BLAT analyses (https://genome.ucsc.edu/cgi-bin/hgBlat).

Statistical analyses

X-chromosome-wide association study

We performed XWAS analyses using PLINK (version 2.0) and incorporated significant covariates in the logistic regression analyses. Using the ‘–xchr-model 2’ parameter, we coded males as 0/2 and females as 0/1/2 to account for dosage compensation, aligning male genotypes with female genotypes. This approach reflects the hemizygous nature of males and the random X-chromosome inactivation in females. The model incorporates sex as a covariate in regression analyses for X-chromosomal associations, accounting for dosage differences between males and females. The covariates were determined using the step function in R (version 3.5.2; https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/step).

To address any evolutionary differences between autosomes and X chromosomes, genetic principal components were calculated for both autosomes and the X chromosomes in the sex-stratified and combined cohort analyses.^8,9,12-14 Specifically, autosomal and X-chromosomal variants were pruned separately to account for distinct linkage disequilibrium and inheritance patterns. The pruned datasets were then combined, and principal components were calculated using FlashPCA (version 2.0),¹⁵ capturing the population structure from both autosomal and sex-chromosomal variants. The covariates from the model with the lowest Akaike Information Criteria (AIC) were utilized in the association analysis.⁸ For the combined cohort, which included both female and male study participants, the regression analysis was adjusted for six principal components (PC1–PC4, PC7 and PC10). In the sex-stratified analyses, the female-only regression model included four principal components (PC1–PC3 and PC5), while the male-only regression model was adjusted for seven principal components (PC1–PC5, PC8 and PC10). The difference in the number of PCs between females and males may reflect the smaller sample size and hemizygous coding of males on the X chromosome, which can inflate higher-order PCs. For each analysis, we computed the additive logistic regression model using a MAF threshold ≥5% calculated based on the MSA cases. Using the genpwr package (version 1.0.4), the power to detect significant associations at OR > 1.6 with MAF ≥ 1% was 70% in the female-only cohort, 79.3% in the male-only cohort and 99.8% in the combined cohort.

The Bonferroni threshold was determined after conducting a haploblock estimation analysis, as detailed elsewhere.⁸ Based on this method, the X-chromosome-wide significance threshold was set at a P-value < 7.94 × 10⁻⁶ (0.05/6296 haploblocks). Variants log₁₀-fold lower than or equal to this threshold (P-value ≤ 7.94 × 10⁻⁵) were considered subsignificant.

For conditional analyses on the XWAS loci identified from the sex-stratified cohorts, the respective index variants were included as covariates. The analyses were conducted using GCTA-COJO¹⁶ (https://yanglab.westlake.edu.cn/software/gcta/#COJO) for the combined and sex-stratified cohorts using the summary statistics from the combined, female-only and male-only XWAS. The local linkage disequilibrium was determined separately for each sex-stratified analysis to capture sex-specific structures.

Sensitivity and validation analyses

A series of sensitivity and validation analyses was performed to test the robustness of the observed association signals as follows:

1. An XWAS analysis was performed on the subset of pathologically confirmed MSA cases (n = 468 in total; n = 225 females and n = 243 males) and all available controls (n = 7128 total; n = 3741 females and n = 3387 males). This sensitivity analysis was carried out for the combined sex-stratified cohorts by applying the additive logistic regression models with an MAF threshold of ≥1%. This MAF threshold was lowered to include the index variants identified in the main analyses. Six covariates were used for the combined cohort (PC1–PC4, PC6 and PC10), five for the females-only cohort (PC1–PC4 and PC8) and six for the male-only cohort (PC1–PC5 and PC8).

2. Similarly, an XWAS analysis was performed on the clinically diagnosed MSA cases (n = 420 in total; n = 191 females and n = 229 males) and all available controls (n = 7128 total; n = 3741 females and n = 3387 males). The same MAF thresholds and covariates were used for the combined cohort, female-only cohort and male-only cohort.

3. To ensure our findings were not confounded by population structure or batch effects, we performed sensitivity analyses on the jointly variant-called cohort (n = 888 MSA cases and 3018 controls of European ancestry⁴), excluding TOPMed data.

4. We performed permutation tests with two steps: (a) resampling with replacement within sex-stratified cohorts (n = 416 female and 472 male cases; n = 3741 female and 3387 male controls) to create 10 000 datasets and generate null distributions through logistic regression; and (b) bootstrapping beta coefficients to derive confidence intervals and P-values, ensuring robust assessment of associations.

5. A meta-analysis was performed using METAL,¹⁷ applying inverse-variance weighting to male-only and female-only summary statistics to assess consistency between sexes.

6. To explore the impact of different X-chromosome inactivation modelling assumptions, we reanalysed the data using a 0/1 coding scheme for males (‘–xchr-model 1’ parameter).

Technical validation of index variants using long-read sequencing

We conducted long-read nanopore whole-genome sequencing of eight female and eight male MSA cases that were previously sequenced with the Illumina HiSeq X10 platform. High-molecular-weight DNA was extracted from frozen cerebellar tissue using the Nanobind HT CBB Kit on a KingFisher Apex instrument (Thermo Fisher Scientific). The extracted DNA was sheared (speed 45, twice) to reach a fragment size of approximately 30 kb, employing a Megaruptor 3 (Diagenode Inc.). To assess the fragment length, we utilized the genomic DNA ScreenTape assay (Agilent) along with the 4200 TapeStation system (Agilent). During library preparation, we performed DNA repair and end prep using the NEBNext^® Companion Module (New England Biolabs), followed by AMPure bead purification (Beckman) and adaptor ligation via the Ligation Sequencing Kit (Oxford Nanopore Technologies). We then loaded 400 ng of the DNA library onto a primed PromethION flow cell (version R9.4.1) and sequenced it for 72 h, flushing (Oxford Nanopore Technologies) and reloading every 24 h to optimize data yield. The data generated ranged from 4.23 to 18.91 million reads, with coverage between 24–49× and an N50 value between 19.27 and 35.86 kb. Super accuracy basecalling was performed using guppy (version 6.2.1). Porechop95 (version 0.2.4) was used for adapter trimming. Read alignment to the human genome build 38 was performed using minimap2 (version 2.2.26). Variant calling was conducted using Epi2me-labs/human-variation (version 2.3.0; https://github.com/epi2me-labs/wf-human-variation). We performed a concordance evaluation for chromosome X-wide significant variants present in the eight females and eight males (n = 48 variants tested) by comparing short-read Illumina genome data against the long-read nanopore data from these cases.

Colocalization analyses and gene prioritization

We performed colocalization analyses to identify and prioritize variants within risk genes that affect disease risk through differential gene expression in brain tissues. Our workflow incorporated summary statistics from the combined and sex-stratified XWAS alongside expression quantitative trait locus (eQTL) data sourced from the sex-aware GTEx dataset (version 8, https://www.gtexportal.org/home/downloads/adult-gtex/qtl#qtl-gtex_analysis_v8-sex-biased_genes_and_sb-eqtl_data). This analysis included data from thirteen distinct brain regions: amygdala, anterior cingulate cortex (BA24), caudate basal ganglia, cerebellar hemisphere, cerebellum, cortex, frontal cortex (BA9), hippocampus, hypothalamus, nucleus accumbens, putamen, spinal cord (cervical C1) and substantia nigra. We utilized sex-aware eQTL data since around 37% of genes in GTEx brain tissues demonstrate tissue-specific, sex-aware gene expression and cellular composition,¹⁸ including several brain regions pertinent to MSA. We focused solely on X-chromosomal eQTLs, which were further divided into female-only, male-only and combined datasets. The data were harmonized with the XWAS summary statistics.

The colocalization analysis was performed using ‘Coloc’ (version 5.2.3; https://chr1swallace.github.io/coloc/articles/a03_enumeration.html). ‘Coloc’ employs a Bayesian framework (SuSiE) to calculate posterior probabilities, allowing us to pinpoint regions where MSA loci and eQTLs may share a causal variant. We analysed variants within a 1 Mb region surrounding each index variant and evaluated five hypotheses: (i) H₀: no association with either trait; (ii) H₁: association with an MSA XWAS variant, but not with an eQTL variant; (iii) H₂: association with an eQTL variant, but not with an MSA XWAS variant; (iv) H₃: both traits are associated but with different variants; and (v) H₄: a shared causal variant underlies both MSA and eQTL associations. We applied standard priors (P₁ = 10⁻⁴, P₂ = 10⁻⁴ and P₁₂ = 5 × 10⁻⁶) for the analysis. A locus was classified as colocalized if the posterior probability for H₄ (PPH4) was ≥0.80, which indicates a strong possibility of a shared causal variant affecting both traits.

Results

Whole-genome sequencing data from 888 MSA cases and 7128 unrelated controls were available for the X-chromosome-wide association study (XWAS) analyses (Fig. 1). Following standard quality control procedures, we performed sex-stratified XWAS analyses: (i) female participants only (n = 416 cases and n = 3741 controls); (ii) male participants only (n = 472 cases and n = 3387 controls); and (iii) a combined cohort of both sexes (n = 888 cases and n = 7128 controls) using an additive association model. After quality control filtering, a total of 171 652 X-chromosomal variants with a MAF ≥5% were available for XWAS analyses.

XWAS analyses identify two sex-differential susceptibility loci

Our XWAS analyses identified two risk loci with significant associations. The first locus was a gene-rich region spanning 250 kb on Xq28, found in the female-only XWAS (see Fig. 2 for the Manhattan plot and Supplementary Fig. 1 for the regional association plot). The index variant (rs4898389, OR = 1.69, 95% CI = 1.40–2.03, P-value = 2.43 × 10⁻⁸; Table 1), located near the CTAG1A gene, exceeded both the chromosome X-wide (P-value ≤ 7.94 × 10⁻⁶) and genome-wide significance (P-value ≤ 5.0 × 10⁻⁸) thresholds. The second risk locus was found in the male-only XWAS on the short arm of chromosome X (Xp22.2, index variant: rs6638956, OR = 1.48, 95% CI = 1.26–1.73, P-value = 9.92 × 10⁻⁷; Fig. 2 and Table 1). The index variant is located within the intronic sequence of the TBL1X gene (Supplementary Fig. 1). Conditional analyses based on the index variants of both loci indicated no additional signals (Supplementary Fig. 2). The quantile–quantile plots showed no notable population stratification (female-only XWAS λ₁₀₀₀ = 0.926; male-only XWAS λ₁₀₀₀ = 0.995; combined cohort XWAS λ₁₀₀₀ = 0.984; Supplementary Fig. 3).

Figure 2.

Multiple system atrophy X-chromosome-wide association study results in female-only, male-only and combined cohorts. The composite figure illustrates the findings from the X-chromosome-wide association study (XWAS) in multiple system atrophy. (A) Association test results from the female-only XWAS are shown. (B) This plot depicts the results for the male-only XWAS. (C) The results for the combined cohort (female and male study participants). The x-axis represents the chromosomal positions in the human genome build 38. The y-axis displays the association P-values of variants with a minor allele frequency of ≥5% on a −log₁₀ scale. The red dashed horizontal line indicates the significance threshold (P-value ≤ 7.94 × 10⁻⁶) for chromosome X, whereas the grey dashed line represents the genome-wide significance threshold (P-value ≤ 5.0 × 10⁻⁸). Each dot signifies a variant, with red dots indicating X-wide significant variants and orange dots denoting suggestive variants (subsignificance threshold P-value ≤ 7.94 × 10⁻⁵). Black circles indicate variants that surpass the genome-wide significance threshold.

Table 1.

Sex-stratified X-chromosome-wide association study analyses identify two multiple system atrophy risk loci

Chr. position (SNP ID)	Genea	EA/OA	REF/ALT	gnomAD EAF	EAF (cases/controls)	OR	95% CI	P-value	XWAS study design
Index variant in female-differential XWAS locus on Xq28
chrX:154,599,376 (rs4898389)	CTAG1A	G/A	A/G	0.14	0.21/0.14	1.69	1.40–2.03	2.43 × 10−8	Female-only
					0.20/0.15	1.32	1.19–1.46	1.35 × 10−7	Combined females and males
					0.19/0.15	1.20	1.06–1.36	4.56 × 10−3	Male-only
Index variant in male-differential XWAS locus on Xp22.2
chrX:9,721,087 (rs6638956)	TBL1X	C/A	A/C	0.06	0.12/0.06	1.48	1.26–1.73	9.92 × 10−7	Male-only
					0.10/0.07	1.36	1.19–1.56	1.11 × 10−5	Combined females and males
					0.07/0.07	1.08	0.81–1.43	0.61	Female-only

Chromosomal positions are represented according to the human genome build hg38. The effect allele frequency refers to the allele that increases disease risk. The X-wide level of significance threshold was ≤7.94 × 10⁻⁶, with significant values highlighted using bold font. The effect allele frequencies for gnomAD (version 4.1.0) refer to the frequencies in the non-Finnish European ancestry population. Alt = alternate allele; CI = confidence interval; Chr. = chromosome; EA = effect allele; EAF = effect allele frequency; OA = other allele; OR = odds ratio; REF = reference allele; XWAS = chromosome X-wide association study.

^aThe gene in closest proximity to the index variant is shown.

In the combined XWAS of both sexes, we confirmed that the Xq28 and Xp22.2 risk loci are significantly associated with MSA (Fig. 2). The Xq28 index variant, rs4898389, identified in the female-only analysis, also surpassed the chromosome X-wide significance threshold in the combined cohort (OR = 1.32, 95% CI = 1.19–1.46, P-value = 1.35 × 10⁻⁷; Table 1), though the association was weaker. Similarly, the male-predominant Xp22.2 index variant, rs6638956, demonstrated a chromosome X-wide subsignificant association in the combined cohort (OR = 1.36, 95% CI = 1.19–1.56, P-value = 1.11 × 10⁻⁵; Table 1), which was also weaker.

To compare the associations across the sex-stratified cohorts, we generated beta–beta plots of the logistic regression models using the standardized β-values (or z-scores) from the male-only and female-only XWAS (Supplementary Fig. 4). Variants at the CTAG1A locus in the female-only XWAS showed a significant positive correlation (r = 0.76, P-value = 2.4 × 10⁻⁶²) with variants at the same locus in the male-only analysis; however, variants at the TBL1X locus in the male-only XWAS indicated a nominally significant negative correlation with variants at this locus in the female-only XWAS (r = −0.3, P-value = 0.016). Based on these data, we conclude that CTAG1A is a female-predominant risk locus, while TBL1X is male-predominant.

Sensitivity analyses confirm the chromosome X-associated loci

We evaluated the strength of the association through sensitivity analyses on: (i) the subset with pathologically confirmed MSA (n = 225 females and n = 243 males); and (ii) the clinically diagnosed subset (n = 189 females and n = 227 males) against healthy controls (n = 3741 females and n = 3387 males). The index variant rs4898389 surpassed the X-wide significance level in the clinically diagnosed female-only and combined XWAS analyses (OR = 1.92, 95% CI = 1.50–2.45, P = 1.80 × 10⁻⁷). Although no other sex-stratified XWAS loci achieved X-wide significance, their effect directions were consistent with previous research, indicating the robustness of these associations (Supplementary Table 1). Furthermore, no significant difference was observed between the sexes when analysing the MSA-Cerebellar and MSA-Parkinsonism clinical subtypes (Supplementary Table 2).

Additional sensitivity analyses, including modelling X-chromosome inactivation with different dosage assumptions of hemizygosity (Supplementary Table 3), permutation tests (Supplementary Fig. 5), a meta-analysis of male-only and female-only XWAS data (Supplementary Table 4) and examination of the association signals in the case-control cohorts that excluded the TOPMed convenience controls (Supplementary Table 5), confirmed robust associations and effect directions at two risk loci. Moreover, none of the index variants identified in recent XWAS studies of Alzheimer’s disease,^19,20 Parkinson’s disease⁹ and Lewy body dementia⁸ were linked to MSA risk, reinforcing the notion that our signals are specific to MSA (Supplementary Table 6). Details of these analyses are in the Supplementary material.

Validation of significant signals through long-read sequencing

To validate our XWAS findings, we compared the variants within the two association signals identified by the XWAS analyses of the short-read whole-genome sequence data with long-read nanopore sequencing data from a subset of female (n = 8) and male (n = 8) MSA cases. This analysis acted as a technical replication of genotyping accuracy rather than an independent association replication. Our results revealed almost complete agreement on the genotyping calls of the significant variants within these cases (in females, 6/8 had 100% concordance and the remaining two cases had 98% concordance; in males, 6/8 had 100% concordance and two cases had 98% concordance; Supplementary Table 7). In addition, we examined the nanopore sequencing data for any structural variants within the associated loci but found none.

Colocalization analyses prioritize PLXNA3 and FAM3A

To identify putative causative genes within the XWAS risk loci, we analysed sex-aware cis-eQTL data from the Genotype-Tissue Expression Project (GTEx; version 8; https://gtexportal.org), which is based on bulk expression data generated using tissue from 13 brain regions. This dataset takes sex-aware gene expression into account. We evaluated variants in a 1 Mb region surrounding the index variant across the three XWAS analyses (female-only, male-only and the combined analysis) to assess the probability that these variants are simultaneously connected to disease risk and gene expression alterations.

In the combined cohort, the colocalization evaluation of the Xq28 locus identified two key genes: PLXNA3, which showed a high posterior probability (PPH4) in the cervical spinal cord (eQTL ID: rs5987271, PPH4 = 0.986; Table 2), and FAM3A in the hypothalamus (eQTL ID: rs7611, PPH4 = 0.999; Table 2). Notably, the GTEx dataset lacked information for many other genes within the Xq28 locus (e.g. G6PD is only present in the putamen), precluding colocalization studies. In the smaller female-only or male-only cohorts, analysing the index variant and the surrounding variants in the 1 Mb region did not detect MSA risk through differential gene expression, likely due to limited statistical power. Furthermore, no significant colocalizations were detected at the Xp22.2 locus.

Table 2.

Colocalization analyses identify genes of interest within the multiple system atrophy Xq28 locus

Tissue	eQTL position (SNP ID)	Gene	z-Score (XWAS)	z-Score (eQTL)	PPH4	XWAS index variant
Spinal corda	chrX:154,503,673:G:A (rs5987271)	PLXNA3	4.27	8.39	0.99	rs4898389
Hypothalamus	chrX:154,506,456:A:T (rs7611)	FAM3A	5.26	−6.47	1.00	rs4898389

X-chromosomal eQTLs mediating multiple system atrophy risk through differential gene expression. Significant eQTLs with colocalization PPH4 values ≥0.80 are shown. The eQTL ID refers to the eQTL with the highest PPH4 values in a given locus. eQTL = expression quantitative trait locus; PPH4 = posterior probability of hypothesis 4; XWAS = X-chromosome-wide association study.

^aSpinal cord tissue was collected at the C1 cervical region.

We further examined an RNA sequencing dataset comparing gene expression in the cerebellum of 47 MSA cases to 47 healthy controls (GEO: GSE199715).²¹ PLXNA3 showed a log₂-fold change of +0.025 (FDR P-value = 0.95), consistent with our colocalization analysis, while FAM3A had a log₂-fold change of +0.016 (FDR P-value = 0.98), which was opposite to our colocalization results in the hypothalamus. Neither change was statistically significant, supporting the conclusion that these genes are not dysregulated in the cerebellar tissue of MSA patients.

Discussion

Our research is the first XWAS performed for MSA, a poorly understood adult-onset neurodegenerative disease. Our findings provide new genetic insights into the sex differences observed in the disease, thereby enhancing our understanding of this rare and disabling neurological disorder. The XWAS analyses revealed two susceptibility loci that differ by sex: one identified at Xq28 in females and the other at Xp22.2 within the TBL1X gene in males. The Xq28 locus is female-driven and not detected in the male-XWAS, whereas the Xp22.2 locus is male-driven and absent in the female-only XWAS. We confirmed these loci using sensitivity analyses and long-read nanopore sequencing of selected samples. Furthermore, colocalization analyses indicated that the PLXNA3 and FAM3A genes likely play a role at the Xq28 locus for female cases. Our findings suggest the involvement of sex-related genetic factors in the development of MSA, potentially accounting for some of the clinical disparities observed between men and women with the disease.^22-24

Research on sex differences in MSA is limited. Recent reviews indicate that men mainly exhibit autonomic symptoms, while women more frequently display Parkinsonian features.²⁴ Biases in data collection may influence these findings; for example, erectile dysfunction has long been used as an indicator of urogenital failure in men, whereas there is no equivalent focus for female sexual dysfunction. Autonomic blood pressure issues, such as supine hypertension and more pronounced drops when standing, are linked to higher mortality and can lead to earlier referrals and detection in men, potentially further skewing the observed ratios. These differences emphasize the importance of understanding genetic factors and sex-specific modifiers in MSA. Our identification of sex-differential risk loci supports the need for larger, prospective studies to better explore disease mechanisms and the relationship between genotype and phenotype. Proteogenomic research further reveals sex-biased regulation, showing that 13% of brain proteins and 1.5% of sex-biased protein QTLs vary by sex, along with similar patterns in brain and immune expression QTLs, indicating sex-differential genetic effects.²⁵

We incorporated colocalization analysis, a fine-mapping method that utilizes eQTL statistics to determine whether a shared causal variant is predicted to drive association signals for MSA risk and gene expression. Significant colocalizations reflect both the shared linkage disequilibrium structure and possible coordinated functional effects. Our colocalization analyses using sex-aware eQTL data from 13 brain tissue regions in the GTEx database identified differential expression of FAM3A in the hypothalamus and PLXNA3 in the cervical spinal cord as contributors to the Xq28 locus. Although MSA primarily impacts the cerebellum, brainstem and basal ganglia, changes in the hypothalamus have been observed in MSA patients.^26,27  FAM3A encodes a cytokine-like protein that affects glucose and lipid metabolism through the activation of peroxisome proliferator-activated receptor gamma.²⁸ Additionally, FAM3A enhances mitochondrial metabolism by increasing ATP production in healthy mice and is part of a protein family that protects neurons from oxidative stress and apoptosis.²⁹ Although significant neuronal loss is evident in MSA, it is primarily classified as an oligodendrogliopathy.³⁰ Given the high expression of FAM3A in oligodendrocytes and the evidence supporting its neuroprotective role,³¹ further research is warranted to clarify the precise mechanism by which FAM3A may be involved in MSA. In particular, the limited size of samples in GTEx meant these analyses may have lacked the power to detect effects of FAM3A outside the hypothalamus.

The PLXNA3 gene, which was also identified by the colocalization analysis of the Xq28 locus, is located 28.4 kb upstream of FAM3A. PLXNA3 encodes plexin A3, part of a family of cell surface receptors for semaphorins.³² This protein is essential for axonal growth and guidance,^32,33 playing a key role in the direction of axons from branchiomotor and motor neurons, along with cytoskeletal remodelling. This interaction is crucial for fetal brain development, with PLXNA3 being instrumental in cell motility and synapse formation.^32,34 In the adult central nervous system, PLXNA3 is expressed in oligodendrocyte precursor cells, influencing their migration and myelin repair.^35,36  In vivo models of cellular senescence indicate an upregulation of PLXNA3 in the brains of wild-type and APP/PS1 mutant mice, a model for Alzheimer’s disease.³⁷ This observation points to a possible involvement of PLXNA3 in ageing and neurodegeneration beyond the spinal cord.

The association signal found in the Xp22.2 locus among males is in an intron of the TBL1X gene, responsible for encoding the Transducin Beta-Like 1 X-Linked protein. This regulatory protein serves as a coactivator alongside beta-catenin for Wnt target genes³⁸ and plays a critical role in regulating embryonic development. TBL1X is classified as an X-degenerate gametologue, denoting that it has a counterpart on the Y chromosome, TBL1Y, which has partially lost its function over the course of evolution.³⁹ Further, TBL1Y shows substantially lower expression than TBL1X,⁴⁰ which may explain why TBL1X mutations in male patients cannot be compensated for by TBL1Y. Skewed X-chromosome inactivation may lessen the severity of TBL1X mutations in females.⁴¹ Our dataset reflects this trend, noting a negative correlation between z-scores in the female and male groups within the beta–beta plots (Supplementary Fig. 4). A deeper understanding of the molecular mechanisms linked to this risk locus is crucial and necessitates further investigation.

Our study has limitations. The rarity of MSA presents a significant challenge in gathering sufficiently large case cohorts required for unbiased, well-powered genome-wide analyses. The limited number of available cases reduces statistical power, particularly for detecting genetic variants with modest effect sizes. Consequently, this issue limits our ability to perform replication studies. However, our study is based on unbiased whole-genome sequencing data rather than imputed genotype data, ensuring comprehensive variant coverage and reducing potential biases arising from imputation inaccuracies. Nonetheless, our sensitivity analysis, which included pathologically confirmed cases—the gold standard for MSA diagnosis—showed consistently aligned directions of effect for the two sex-differential risk loci. Additionally, we validated our leading signals using long-read sequencing, achieving highly concordant genotype calls and ruling out structural variants in our regions of interest that could distort our association results.

Another limitation of our study was the use of different covariates in the sex-stratified analysis, which might have masked the influence of incomplete X-inactivation. To mitigate the potential effects of skewed X-inactivation, we used a sex-aware gene expression QTL dataset for the colocalization analyses, as X-inactivation is linked to tissue- and cell-type-specific gene expression.⁴² Nonetheless, we could not determine differential gene expression in the female-only and the male-only cohorts, likely due to insufficient statistical power. We were further restricted in conducting functional genomic integrative analyses due to the limited representation of X-chromosomal genes in the sex-aware gene expression QTL dataset, highlighting the need for larger tissue- and cell-specific transcriptomic catalogues. For example, the lack of transcript information for G6PD in the sex-aware GTEx dataset, a gene that has previously been linked to parkinsonism,⁴³ meant we could not test for its involvement in MSA.

Finally, our study focused on individuals of European ancestry, as this population had substantial collections of biosamples readily accessible for genome sequencing. Genetic associations may vary across different populations. Looking forward, the availability of larger and more diverse cohorts will facilitate a more thorough understanding of MSA’s pathogenesis. To support this research, we have made our summary statistics publicly available.

In conclusion, our study contributes to the existing research on the genetic architecture of MSA by including the X chromosome. These findings will support the development of a multifactorial predictive model that integrates genetic risk factors, biomarkers and clinical information, ultimately assisting in formulating personalized treatment strategies for both women and men affected by MSA.

Appendix 1.

International MSA genomics consortium

Full details are provided in the Supplementary material.

Anindita Ray, Ruth Chia, Memoona Rasheed, Ece Bayram, Jinhui Ding, Ronald L. Walton, Regina H. Reynolds, Monica Diez-Fairen, Ignacio Alvarez, Alessandra Fanciulli, Nadia Stefanova, Klaus Seppi, Susanne Duerr, Fabian Leys, Florian Krismer, Victoria Sidoroff, Alexander Zimprich, Walter Pirker, Olivier Rascol, Alexandra Foubert-Samier, Wassilios G. Meissner, François Tison, Anne Pavy-Le Traon, Maria Teresa Pellecchia, Paolo Barone, Maria Claudia Russillo, Juan Marín-Lahoz, Jaime Kulisevsky, Soraya Torres, Pablo Mir, Maria Teresa Periñán, Christos Proukakis, Viorica Chelban, Lesley Wu, Yee Y. Goh, Laura Parkkinen, Michele T. Hu, Christopher Kobylecki, Jennifer A. Saxon, Sara Rollinson, Emily Garland, Italo Biaggioni, Irene Litvan, Ileana Rubio, Roy N. Alcalay, Kimberly T. Kwei, Steven J. Lubbe, Qinwen Mao, Margaret E. Flanagan, Rudolph J. Castellani, Vikram Khurana, Alain Ndayisaba, Andrea Calvo, Gabriele Mora, Antonio Canosa, Gianluca Floris, Ryan C. Bohannan, Anni Moore, Lucy Norcliffe-Kaufmann, Jose-Alberto Palma, Horacio Kaufmann, Ted M. Dawson, Liana S. Rosenthal, Alexander Pantelyat, Marilyn S. Albert, Olga Pletnikova, Juan C. Troncoso, Jon Infante, Carmen Lage, Pascual Sánchez-Juan, Geidy E. Serrano, Thomas G. Beach, Pau Pastor, Huw R. Morris, Diego Albani, Jordi Clarimon, Zane Jaunmuktane, Karl Frontzek, Tamás Révész, Tom Warner, John A. Hardy, Mina Ryten, Luigi Ferrucci, Eric Topol, Ali Torkamani, Alex R. DeCasien, Adriano Chiò, Philip A. Low, Wolfgang Singer, William P. Cheshire, Zbigniew K. Wszolek, Dennis W. Dickson, Bryan J. Traynor, J. Raphael Gibbs, Clifton L. Dalgard, Owen A. Ross, Henry Houlden and Sonja W. Scholz.

Data availability

The summary statistics of the MSA XWAS have been deposited in the GWAS catalogue (https://www.ebi.ac.uk). The individual-level sequence data for a subset of the MSA genomes (n = 683 cases) reported in this paper are available in dbGaP (www.ncbi.nlm.nih.gov/gap/; accession number: phs001963.v3.p1). The remaining 205 MSA genomes could not be publicly shared, and access will be provided using appropriate collaboration agreements. The TOPMed control genome data are available in dbGaP (accession numbers: phs001662.v2.p1, phs00974.v5.p4 and phs000951.v5.p5). The control genome data from 1980 subjects from the DementiaSeq project are available in dbGaP (accession number: phs001963.v3.p1), and the remaining control genomes are available upon request from the Wellderly study team (contact: atorkama@scripps.edu).

Funding

This research was supported by the Intramural Research Program of the National Institutes of Health (NIH, program #: 1ZIAAG000935, ZIANS003154). The contributions of the NIH authors were made as part of their duties as NIH federal employees, are in compliance with agency requirements and are considered works of the United States Government. However, the findings and conclusions presented in this paper are those of the authors and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services.

Competing interests

B.J.T. holds US, EU and Canadian patents on the clinical testing and therapeutic intervention for the hexanucleotide repeat expansion of C9orf72. B.J.T. and S.W.S receive research support from Cerevel Therapeutics. B.J.T. is an editorial and advisory board member for Brain, eClinicalMedicine and the Journal of Neurology, Neurosurgery, and Psychiatry. The Mayo Clinic brain bank is supported by the Mangurian Foundation, CurePSP, the Tau Consortium and the Robert E. Jacoby Professorship. O.A.R. and D.W.D. are supported by the NINDS Tau Center without Walls Program (U54-NS100693) and NIH (UG3-NS104095). S.W.S. is an editorial board member for JAMA Neurology and the Journal of Parkinson’s Disease. S.W.S. is a scientific advisory board member for the Lewy Body Dementia Association, Mission MSA and the GBA1 initiative. O.A.R. is supported by the NIH (P50-NS072187; R01-NS078086; U54-NS100693; U54-NS110435; R01-AG087165), DOD (W81XWH-17-1-0249), The American Brain Foundation, The Michael J. Fox Foundation, The Little Family Foundation, Ted Turner and family, the Mayo Clinic Foundation and the Center for Individualized Medicine. D.W.D. receives research support from the NIH (P50-AG016574; P30-AG062677; U54-NS100693; P01-AG003949). All other authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.