CAG Repeat Not Polyglutamine Length Determines Timing of Huntington's Disease Onset

Abstract

Variable, glutamine-encoding, CAA interruptions indicate that a property of the uninterrupted HTT CAG repeat sequence, distinct from the length of huntingtin's polyglutamine segment, dictates the rate at which Huntington's disease (HD) develops. The timing of onset shows no significant association with HTT cis-eQTLs but is influenced, sometimes in a sex-specific manner, by polymorphic variation at multiple DNA maintenance genes, suggesting that the special onset-determining property of the uninterrupted CAG repeat is a propensity for length instability that leads to its somatic expansion. Additional naturally occurring genetic modifier loci, defined by GWAS, may influence HD pathogenesis through other mechanisms. These findings have profound implications for the pathogenesis of HD and other repeat diseases and question the fundamental premise that polyglutamine length determines the rate of pathogenesis in the "polyglutamine disorders."

Keywords: CAG repeat; DNA maintenance; DNA repair; Huntington’s disease; age at onset; disease modification; genetic modifier; polyglutamine disease; somatic DNA expansion; trinucleotide repeat.

Graphical abstract

Figure S1

Residual Age at Onset Phenotype for GWA Analysis to Identify Genetic Modifiers of HD, Related to Figure 1 (A) Age at onset data for individuals with HD (y axis) were compared to CAG repeat size based on genotyping. A red line represents our standard CAG-onset phenotype model. Residual age at onset was calculated for each subject by subtracting expected age at onset based on our CAG-onset model from observed age at motor onset. (B) Residual age at onset was our primary phenotype for genetic analysis. Distribution of residual age at onset of individuals with HD (histogram) was compared to a theoretical normal distribution based on the mean and standard deviation of actual data to confirm data normality. (C) For an independent comparison with our previous GWA results (GWA123), we performed mixed effect model GWA analysis of additional HD individuals (GWA45). The Manhattan plot summarizes association signals using residual age at onset for 4,793 HD individuals. Y- and x axis represent -log10(p value) and chromosome number. (D) In addition, we performed and subsequently compared 1) meta-analysis to summarize 5 independent GWA analysis results and 2) mixed effect model combined analysis to avoid statistical artifacts. The Manhattan plot of the meta-analysis shown here and that of combined analysis using the continuous phenotype in Figure 1 were very similar, confirming the lack of batch effect in our GWA analysis results. In order to provide meaningful effect sizes of associated loci, we primarily used the results of the combined analysis. (E) A quantile-quantile (QQ) plot based on the GWA analysis results using continuous phenotype (mixed effect model, combined analysis) and the inflation factor confirmed the lack of statistical inflation in our results. (F) A QQ plot based on the GWA analysis results using dichotomous phenotype (fixed effect model, combined analysis) and the inflation factor confirmed the lack of statistical inflation in our results.

Figure 1

Age at Onset GWAS Signals GWAS results using the residual age at onset phenotype or a dichotomized phenotype, with each circle representing a test SNP and significances shown as –log10(p value) (top) and log10(p value) (bottom), respectively. See also Figures S1, S2, and S3 and Tables S2 and S3–S5.

Figure S2

Chr 4 GWAS Signals Do Not Correspond to HTT cis-eQTL Signals and Are Dramatically Reduced by Correcting Mis-estimated CAG Repeat Length, Related to Figures 1 and 2 (A) Expression levels of GTEx subjects in various tissues are plotted. The background histogram represents the distribution of HTT mRNA levels across all tissues. TPM (x axis) represents Transcripts Per Million (Li et al., 2010). (B-E) GWAS signals for chromosome 4 (x axis) were compared to GTEx eQTL analysis results for prefrontal cortex BA9 (C), cortex (D), caudate (E), and putamen (F). Upward and downward triangles represent SNPs whose minor alleles were associated with increased and decreased HTT mRNA levels, respectively, in GTEx data. SNPs on the haplotype marked by rs764154313 are infrequent and therefore were filtered out from publicly available GTEx eQTL results. Some SNPs from the haplotype tagged by rs183415333 are in the GTEx eQTL dataset and are compared to HD modifier GWAS data (green triangles) but do not correspond with eQTLs in any of the brain tissues. In BA9, GWAS signals in the 1E-4 range contributed by more frequent SNPs correspond with eQTL signals for decreased HTT expression, but these weak GWAS signals are removed by conditioning for rs183415333. Notably, the rs183415333-tagged haplotype represents a small subset (∼10%) of the chromosomes that bear the more frequent rs13102260 promotor SNP (black triangle here and in Figure 2)A, which was proposed to be involved in regulation of HTT expression levels and subsequent modification of HD (Bečanović et al., 2015) but which failed to yield a strong association signal in either GWA123 or GWA12345. (F) Residual age at onset of heterozygous HD individuals (red line; primary y axis) is qualitatively compared to a histogram distribution of residual age at onset of HD individuals with 2 expanded alleles (30 individuals; secondary y axis). (G) Correcting the mis-estimated CAG repeat length of individuals with rs764154313 or rs183415333 minor alleles dramatically reduced apparent chr 4 signals without a major change in all other GWAS signals. The GWAS data from Figure 1 were re-analyzed using the true uninterrupted CAG repeat from MiSeq sequencing of rs764154313 or rs183415333 minor allele subjects. Of 29 individuals with the rs764154313 minor allele, 28 had DNA available for sequencing: 21 had a CAA-loss allele on the disease chr 4, 2 had CAA-loss allele on the normal chr 4 and 5 had a canonical sequence on both chr 4. Of 102 individuals with the rs183415333 minor allele, 98 had DNA available: 67 had a CAACAG-duplication allele on the disease chr 4, 22 had a CAACAG-duplication allele on the normal chr 4, 2 had a CAACAG-duplication on both the normal and the expanded chr 4 and 7 had a canonical sequence on both chr 4. The 5 individuals for whom no DNA sequence was available, along with one canonical allele individual whose sample showed evidence of mix-up, were excluded from this analysis of 9058 individuals. In this dataset, the length of the CAG repeat accounted for 57% of the variance in age at onset. Inclusion of all SNPs with p < 1E-5 and p < 1E-3 raised the explanatory power to 66% and 90%, respectively. However, to what degree this polygenic modification score is predictive remains to be tested in an independent HD sample. Of note, based on the ancestry of some HD individuals in our current dataset with these CAACAG-duplication alleles, this HTT haplotype tagged by the rs183415333 minor allele is likely to correspond to that reported based upon PCR fragment-based genotyping to be associated with later than expected onset in a Danish HD family (Nørremølle et al., 2009).

Figure 2

Correcting Mis-estimated CAG Repeat Length Removes Significant Signals at HTT (A) Chr 4 signals (–log10(p value); continuous phenotype) plotted versus genomic coordinate (GRCh37/hg19). The dotted red line indicates genome-wide significance. Red and green symbols are SNPs tagging haplotypes with minor alleles of rs764154313 and rs183415333, respectively. In this and subsequent plots, downward and upward triangles are SNPs with minor alleles associated with hastened and delayed onset, respectively. Genes are below in red (plus strand) and blue (minus strand). (B) HTT repeat and adjacent sequence, CAG repeat length estimated by genotyping, true CAG repeat length, and polyglutamine length for chrs with CAA-loss (left), canonical (center), and CAACAG-duplication haplotypes, using 42 uninterrupted CAGs as an example. (C) MiSeq analysis of individuals with rs764154313 and rs183415333 minor alleles identified individuals with CAA-loss (red) and CAACAG-duplication (green) alleles (all others shown in gray), permitting comparison of their age at onset with uninterrupted CAG repeat length (left) or total polyglutamine length (right). The black trend line represents our standard onset-CAG phenotyping model for comparison with trend lines in red and green for those with the CAA-loss and CAACAG-duplication alleles, respectively. (D) A boxplot (plotted as quartiles: whiskers1.5∗IQR(interquartile range)) of residual age at onset for individuals with a rare CAA-loss HD haplotype (red; N = 21), the 8 most frequent canonical (single CAACAG) HD haplotypes (gray; N = 3357 hap.01, 2016 hap.02, 942 hap.03, 272 hap.04, 266 hap.05, 312 hap.06, 257 hap.07, 302 hap.08) or a rare CAACAG-duplication HD haplotype (green; N = 69), ordered by increasing polyglutamine length (given the same CAG repeat length). See also Figure S2 and Table S1.

Figure S3

Sex-Specific Association Analysis of Residual Age at Onset, Related to Figure 1 and Table 1 (A) The distribution of age at onset residuals in GWA12345 subjects is presented as a standard boxplot by sex, showing no significant difference between males and females. (B) The Manhattan plot summarizes association signals using residual age at onset for 4,417 male HD individuals. Y- and x axis represent -log10(p value) and chromosome number. (C). The Manhattan plot summarizes association signals using residual age at onset for 4,647 female HD individuals. (D) A QQ plot based on the male-specific GWA analysis is shown, confirming lack of statistical inflation. (E) A QQ plot based on the female-specific GWA analysis is shown, confirming lack of statistical inflation.

Figure 3

Three HD Onset Modification Signals at DHFR/MSH3 (A) SNP significance (–log10(p value) in continuous analysis conditioned on the top 5AM1 (red) SNP compared to GWA12345 significance reveals independent modifier haplotypes with p < 1E-5: 5AM2 (green) and 5AM3 (purple). (B) SNP significance compared to MAF. (C) SNP significance relative to genes in the region. See also Figures S4 and S5.

Figure S4

Onset-Hastening Modifier 5AM1 Is Associated with Somatic CAG Repeat Expansion, Related to Figure 3 (A and B): HD modifier GWAS signals in the MSH3/DHFR region of chr 5 region (x axis, -log10(p value)) were compared to GTEx eQTL signals for MSH3 and DHFR in whole blood (A) and (B), respectively). Upward and downward triangles represent SNPs whose minor alleles were associated with increased and decreased expression levels of the test gene, respectively. Red, green, and purple triangles represent SNPs tagging modifier haplotypes 5AM1, 5AM2, and 5AM3, respectively, as in Figure 3. (C) Method for calculating the peak proportional sum of HTT CAG expansion values from ABI GeneMapper fragment sizing profiles of expanded alleles from our CAG genotyping assay. The bulk of the PCR product for any individual corresponds to the presumptive inherited expanded CAG size and constitutes a floor with respect to which expansion can be examined. PCR products to the left of the main peak are believed to be due largely to PCR artifacts. The peaks to the right of the main peak result from somatic CAG expansions. These are summed and then divided by the size of the main peak to calculate the peak proportional sum of expanded alleles in that individual. The method detects those mosaic individuals with the highest proportion of somatically expanded alleles but is not effective for resolving individuals with reduced somatic expansion. Consequently, it is only applicable to modifier SNPs minor alleles that 1) correspond to blood eQTLs or structural changes, 2) are sufficiently frequent to generate an adequate sample size and 3) are associated with increased somatic expansion (expected to correspond to hastened onset). Of the modifiers in Table 1, only 5AM1 met these criteria. (D) The boxplot shows the peak proportional sum HTT CAG repeat expansion values determined for 7,013 GWA12345 subjects, plotted as quartiles (whiskers1.5^∗IQR) for each CAG repeat size. The proportion of expanded alleles per individual increases with inherited CAG repeat length. The top quartile at each repeat length, representing individuals with the highest proportion of somatic expansions, is distinguished by a wider range of values than the other three quartiles across the CAG repeat lengths. The values for the 1,753 individuals in the highest expansion quartile are plotted as blue circles. Among these individuals, 5AM1-tag SNP rs701383 deviates from Hardy-Weinberg expectation (chi-square 15.80, 2 d.f., p < 0.0004) due to an excess of the A minor allele, associated with hastened onset and increased MSH3 expression.

Figure S5

Correspondence between Onset Modifier Signals and GTEx eQTL Signals for MSH3 and DHFR, Related to Figure 3 (A B) HD modifier GWAS signals in the MSH3/DHFR region of chr 5 region (x axis, -log10(p value)) were compared to GTEx eQTL signals for MSH3 and DHFR (A) and (B), respectively) in BA9, cerebral cortex, caudate and putamen whole blood. Upward and downward triangles represent SNPs whose minor alleles were associated with increased and decreased expression levels of the test gene, respectively. Red, green, and purple triangles represent SNPs tagging modifier haplotypes 5AM1, 5AM2, and 5AM3, respectively, as in Figure 3. There was no sex difference in the expression of MSH3 in any of the tissues (blood p = 0.555; BA9 p = 0.586; cortex p = 0.993; putamen p = 0.542; or caudate p = 0.054). Interaction between sex and the top 5AM1 SNP in influencing expression was not nominally significant in any tissue (blood p = 0.799; BA9 p = 0.867; cortex p = 0.498; putamen p = 0.985) except caudate (p = 0.025) but the latter significance was eliminated by multiple testing correction. Larger samples sizes will be required for a definitive conclusion.

Figure 4

Four HD Onset Modification Signals at FAN1 (A) SNP significance (–log10(p value) in continuous analysis conditioned on the top 15AM1 (red) SNP compared to GWA12345 significance reveals independent modifier haplotypes: 15AM2 (green), 15AM3 (purple), and 15AM4 (gold). Because SNPs tagging 15AM4 have alleles of close to equal frequency, the direction of the arrow varies depending on whether the minor or major allele is on the 15AM4 haplotype. (B) SNP significance compared to MAF. (C) SNP significance relative to genes in the region. See also Figure S6.

Figure S6

Correspondence between Onset Modifier Signals and GTEx eQTL Signals for FAN1 and LIG1, Related to Figures 4 and 5 (A) Chromosome 15 onset modifier signals depicted (as in Figure 4) in red (15AM1), green (15AM2), purple (15AM3), and gold (15AM4) (x axis) were compared to GTEx eQTL signals (y axis) for brain regions and whole blood. Upward and downward triangles represent SNPs whose minor alleles were associated with increased and decreased expression levels of FAN1, respectively. Because SNPs tagging 15AM4 have alleles of close to equal frequency, the direction of the arrow varies depending on whether the minor or major allele is on the 15AM4 haplotype. (B) Chromosome 19 onset modifier signals depicted in red (19AM1) and green (19AM2) (x axis) were compared to GTEx eQTL signals (y axis) for LIG1 expression in brain regions and whole blood. The tag SNP for 19AM3 is too infrequent to appear in GTEx eQTL results. Upward and downward triangles represent SNPs whose minor alleles were associated with increased and decreased expression levels of LIG1, respectively.

Figure 5

Three Onset Modification Signals at LIG1 (A) SNP significance (–log10(p value) in continuous analysis conditioned on the top 19AM1 (red) SNP compared to GWA12345 significance reveals independent modifier haplotypes: 19AM2 (green) and 19AM3 (purple). (B) SNP significance compared to MAF. (C) SNP significance relative to genes in the region. See also Figure S6.

Figure 6

An Onset Modification Signal at CCDC82 (A) SNP significance in dichotomous analysis conditioned on the top 11AM1 (red) SNP compared to GWA12345 significance reveals the presence (with subsequent analyses) of a related haplotype (pink triangles). (B) SNP significance compared to MAF, revealing a slight difference for tag SNPs of the closely related haplotypes. (C) SNP significance relative to genes in the region. (D) SNP significance in dichotomous analysis is compared to cis-eQTL signals for CCDC82 in GTEx consortium data for prefrontal cortex BA9, cortex, caudate, and putamen. In this case, the direction of the triangle indicates association of the minor allele with increased (upward) or decreased (downward) expression. Both related haplotypes (red and pink) show correlation with cis-eQTL SNPs associated with increased expression of CCDC82 in all 4 brain regions. See also Figure S7.

Figure S7

Association Signals in Continuous Phenotype Analysis for Chromosome 11 Region and Correspondence to GTEx eQTL Signals, Related to Figure 6 (A) Association analysis for SNPs in the chromosome 11 region using the continuous phenotype. (B). These continuous phenotype analysis results were compared to GTEx eQTL signals for CCDC82 in brain regions. (C and D) GTEx eQTL signals for MAML2 (C) and JRKL (D) in selected brain regions (y axis) were compared to chromosome 11 onset modifier signals based on the dichotomous phenotype (from Figure 6). Red and pink triangles represent SNPs marking the 11AM1 and a closely related haplotype.