Association Analysis of Chromosome X to Identify Genetic Modifiers of Huntington’s Disease

INTRODUCTION

Huntington’s disease (HD; OMIM #143100) is a dominantly inherited neurodegenerative disease, caused by an expanded unstable CAG trinucleotide repeat in huntingtin (HTT) [1]. CAG repeats longer than 39 are fully penetrant, resulting in cognitive and behavioral manifestations in addition to a characteristic, diagnostic, movement disorder whose timing is best explained by the length of the inherited CAG repeat, with longer repeats typically leading to earlier onset [1–6]. However, the remaining variance in age-at-onset that is not accounted for by the size of the repeat is heritable, which implicates a role for other genetic factors in influencing HD [7–9] by accelerating or decelerating the rate at which the HTT CAG expansion mutation drives disease onset. Thus, identification of genetic modifiers of HD can inform underlying disease mechanisms of the disease and may point to disease-modifying therapeutic strategies [10].

In order to discover genetic modifiers of HD, we have performed a series of genome-wide association studies (GWAS) [11–13] focusing on residual age-at-onset, which represents age-at-onset normalized by individual CAG repeat size [5, 14]. In our standard GWAS testing SNPs (simple nucleotide polymorphisms) on autosomal chromosomes, significant modification signals were detected at many genomic locations that harbor DNA repair genes involved in CAG repeat instability [13, 14]. These data not only suggest an important role for somatic CAG repeat instability in influencing the onset of HD [12, 15– 17] but also support this process as a potential target of disease-delaying therapeutics. Interestingly, although DNA mismatch repair had been proposed as a process important to repeat instability [18–29], the genes implicated by significant age-at-onset modification signals do not completely overlap with canonical mismatch repair genes (https://www.gsea-msigdb.org/gsea/msigdb/geneset_page.jsp?geneSetName=GO_MISMATCH_REPAIR). For example, FAN1, which is better known for inter-strand crosslink repair was the most significant modifier in our recent GWAS of ∼9,000 HD subjects [13, 30]. Only a subset of canonical mismatch repair genes generated genome-wide significant signals (e.g., MLH1, MSH3), whereas others failed to exhibit even suggestive association signals (e.g., MSH2, MSH6) [13, 14]. In addition, our GWAS also implicated non-DNA repair genes (e.g., CCDC82, TCERG1) as modifiers of HD age-at-onset [13]. These genes might act indirectly on CAG repeat instability processes or might influence HD through a different mechanism [13]. Thus, these human genetic studies indicate that HD can be modified by particular genes involved in DNA repair/maintenance, and also by genes not participating directly in DNA repair processes. Our pursuit of GWAS reflects the potential that identification of either type of modifier gene has implications for understanding the pathogenesis of HD and for the development of disease-delaying therapeutic strategies applicable for this and possibly other DNA expansion disorders [19, 28]. Still, the past GWAS [11, 13] remains incomplete because only autosomes were tested. Therefore, to supplement our recent autosomal GWAS (∼9,000 European ancestry) [13], we have now performed an X chromosome-wide association study (XWAS) to determine whether variants in genes on the X chromosome modify the timing of HD onset.

MATERIALS AND METHODS

RESULTS AND DISCUSSION

Following genotype imputation and quality control, we analyzed 214,728 SNPs (minor allele frequency > 0.1%) on chromosome X in 8,963 HD subjects (4,345 males and 4,618 females) who carry 40– 55 CAG repeats (Supplementary Figure 1A). Similar to our autosomal GWAS [13], we focused on a phenotype that represents age-at-onset corrected for individual CAG repeat size (i.e., residual age-at-onset). Consistent with our previous analysis of a relatively small number of HD subjects [31], age-at-onset is similar between male and female HD subjects for a given CAG repeat (Supplementary Figure 1B) indicating no significant effect of the subject’s sex on the timing of onset. As anticipated, residual age-at-onset is also not significantly different between males and females (Supplementary Figure 1C; t-test p-value, 0.5415).

Next, we performed XWAS using the combined (males plus females) data to evaluate the levels of association between residual age-at-onset and SNPs on chromosome X (excluding pseudoautosomal regions due to low genotype imputation quality) as the primary analysis. For a given test SNP, we constructed a linear mixed effects model to correct for familial relationship, using residual age-at-onset as the continuous phenotype variable and a single SNP coded based on X-chromosome inactivation in females [32, 33]. We also performed separate XWAS for males and females. In addition, we performed a separate XWAS with a dichotomized residual age-at-onset phenotype, comparing subjects with the top 30% and bottom 30% residual age-at-onset values using logistic regression models similar to the dichotomous analysis performed for SNPs on autosomes [13]. The X chromosome contains 19 genes annotated as ‘DNA_Repair’ in gene ontology (GO:0006281) [34] (Table 1), but none of these is specifically associated with gene ontology_Mismatch_Repair (GO:0006298). Overall, neither continuous (top panel) nor dichotomous phenotype (bottom panel) XWAS analysis revealed genome-wide significant (p-value < 5E-8) association signals for SNPs near any of the genes involved in DNA repair (Fig. 1, blue circles; Table 1) or elsewhere on the X chromosome (Fig. 1; grey circles).

OPEN IN VIEWER

Fig. 1

XWAS analysis of continuous or dichotomized residual age-at-onset in HD. Linear mixed effects model XWAS analysis was performed to evaluate the levels of association between the test SNP and residual age-at-onset in continuous phenotype analysis (top panel) and dichotomous phenotype analysis (bottom panel). The Y-axis of top and bottom panels represent -log10(p-value) in continuous phenotype analysis and log10(p-value) in dichotomous phenotype analysis, respectively. Blue circles represent SNPs located in 50 KB flanking regions of DNA repair genes (see Table 1). Orange horizontal lines indicate suggestive significance (p-value, 0.00001). Red dots indicate the top SNP (rs59098970) at Xq12 region that generated suggestive significant modification signals. PARs (ChrX: 60001-2699520 and ChrX:154931004-155260560) were excluded from our XWAS due to low quality in genotype imputation.

Despite the absence of genome-wide significant associations, a number of suggestive (p-value < 1E-5) signals were observed (Table 2). For example, the Xq12 region (chrX:64550000-65550000) exhibited four SNPs near MSN (moesin) and VSIG4 (V-set and immunoglobulin domain-containing protein 4) with suggestive association signals in the continuous phenotype XWAS (Fig. 1; Table 2) that showed contribution from both male and female HD subjects (Table 2; Fig. 2; Supplementary Figure 2). Patterns of linkage disequilibrium (Fig. 3) and conditional analysis using the top SNP (rs59098970) (Supplementary Figure 3) revealed two relatively independent signals of suggestive significance in this region. The top SNP (rs59098970) in this region was still suggestive significant (p-value, 2.6E-6) in an association analysis model with additional covariates that included a set of previously discovered autosomal modifier SNPs (tagging modifier haplotype 2AM1, 3AM1, 5AM1-3, 5BM1, 7AM1, 8AM1, 11AM1, 15AM1-4, and 19AM1-3) [13]. A single suggestive signal (rs10284175) was also observed at Xp22.11 in an intron of the non-coding RNA PTCHD1-AS (Patched-domain containing 1 antisense RNA).

OPEN IN VIEWER

Fig. 2

XWAS of residual age-at-onset in males and females. XWAS analysis was performed for males (top panel) and females (bottom panel) separately. Y-axis represent -log10(p-value) for male-specific analysis and log10(p-value) for the female-specific analysis. Orange horizontal lines indicate suggestive significance. Blue circles represent SNPs located in 50 KB flanking regions of DNA repair genes. Orange horizontal lines indicate suggestive significance (p-value, 0.00001). Red dots indicate the top SNP (rs59098970) at Xq12 region.

OPEN IN VIEWER

Fig. 3

Suggestive significant association signals at Xq12. Suggestive significant association signals at Xq12 are displayed. The Y-axis and X-axis represent the levels of association significance (-log10(p-value)) and genomic coordinate, respectively. The secondary Y-axis represents the recombination rate based on HapMap data. Upward and downward triangles represent SNPs whose minor alleles are associated with delayed and hastened age-at-onset, respectively. SNPs are colored based on linkage disequilibrium with the top SNP in this region (i.e., rs59098970). The sizes of triangles are proportional to MAF. Genes in red and blue represent RefSeq Select transcripts on the plus and minus strand, respectively.

Several different suggestive signals were observed in the sex-specific XWAS. Three Xq25 SNPs with common minor alleles in GRIA3, which encodes subunit 3 of the AMPA type ionotropic glutamate receptor, achieved suggestive significance in males but were not even nominally significant (p < 5E-2) in females (Table 2). By contrast, three SNPs with low frequency minor alleles in Xq21.22 and one in Xp22.31, both regions representing large intergenic segments, all reached suggestive significance only in females (Table 2).

Any of these suggestive association signals may represent genuine modifier effects that will become genome-wide significant with larger XWAS sample sizes or be revealed by complementary evidence. Alternatively, they may be due to chance. If the former, they would more likely be acting through the mechanism that causes neuronal toxicity in HD rather than through the CAG repeat expansion that leads to triggering of that toxicity mechanism when a threshold CAG length is reached through somatic expansion of the repeat in vulnerable neurons [14]. None of the association signals is near any genes reported to be involved in DNA repair, being instead near genes that could conceivably be participating in causing or protecting the cell from toxic damage. For example, excitotoxicity via glutamate receptor activation has long been postulated to play a role in HD making GRIA3 an interesting candidate [35]. Similarly, the level of expression of moesin (‘membrane-organizing extension spike protein’), which links integral membrane proteins to the cytoskeleton and regulates signaling from surface receptors [36] has recently been implicated in neuronal viability [37]. Whether these or other genes in the vicinity of the suggestive signals are modifiers of HD will require additional evidence that we hope to gain from future genetic studies or from other experimental approaches.

To facilitate engagement of the research community with our autosomal GWAS data [13], we previously created a web site that allows interested investigators to browse the association data to view SNP location and identity, minor allele frequency, association p-value and effect size by SNP or by gene, as well as providing a link to a custom track in the UCSC Genome Browser to permit direct integration with the multitude of other regional annotations available [14]. The GWAS web site also allows download of the full autosomal data set. For additional useful functionality, the web site provides the capacity to test user-defined gene sets for significant association with HD age-at-onset [14]. Consequently, to ensure that investigators using complementary approaches can capitalize on the information provided by this XWAS and potentially develop evidence that assists in defining autosomal or X-linked modifiers, we have updated our GEM Euro 9K web site by combining the autosomal and X-chromosome data, with all of the functionality described above. Our updated ‘GEM Euro 9K’ website is available at https://www.hdinhd.org/.

Our autosomal GWAS identified multiple loci involved in DNA repair processes that implicated somatic expansion of the CAG repeat as the rate driver leading to disease onset and other loci that might be indirectly involved in this process or act later, on the toxicity mechanism. On the assumption that the toxicity mechanism in human HD involves the expression of huntingtin protein with an expanded polyglutamine segment, which remains uncertain, huntingtin lowering strategies are now being tested as possible therapeutic interventions to slow the worsening of HD symptoms. The identification of multiple DNA maintenance genes as modifiers of onset has subsequently led to preclinical efforts to interfere with somatic CAG expansion as a potential treatment. Identification of additional candidate modifiers of the toxicity driver mechanism might similarly help to define the nature of that mechanism and to provide additional routes to treatments validated directly by human data. Although the current XWAS did not reveal genome-wide significant modification signals or add additional loci to the DNA maintenance category of modifiers, our data remain consistent with the possibility that age-at-onset in HD can be modified by non-DNA repair mechanisms in some X chromosome genes. As revealed by our GWAS, identification of functionally-related sets of modifier genes and thereby informing the underlying mechanisms of modification can significantly contribute to the development of effective disease modifying treatments for HD. Consequently, we are currently in the process of further expanding the GWAS and XWAS datasets with the goal of further dissecting the rate driving process and delineating the mechanism(s) involved in causing and modifying the toxic damage.