MSH3 modifies somatic instability and disease severity in Huntington’s and myotonic dystrophy type 1

1Huntington’s disease and myotonic dystrophy type 1 (DM1) are autosomal dominant disorders caused by CAG·CTG trinucleotide repeat expansions. Huntington’s disease is characterized by a progressive movement disorder, cognitive impairment and psychiatric symptoms, and DM1 by myotonia, muscular dystrophy, cognitive impairment, cardiac conduction defects and endocrine dysfunction. No disease-modifying treatments are available for either.

2Huntington’s disease is caused by a (CAG)n repeat expansion in HTT exon 1 and DM1 by a (CTG)n expansion in the 3′ untranslated region (UTR) of DMPK. In both, inherited repeat length is the major determinant of disease course, correlating inversely with the age at onset and positively with disease severity. The repeat is unstable, and expansion during germline transmission results in genetic anticipation. Repeat tracts are also unstable in somatic cells, tending to expand over time, particularly in Huntington’s disease striatum and DM1 muscle, the most prominently affected tissues in each disease. Such expansion-biased, age-dependent and tissue-specific somatic instability is thought to contribute to disease onset and progression.

3In mouse models, the DNA mismatch repair proteins MSH2 and MSH3 are essential for CAG·CTG repeat expansion, and their inactivation limits expansion events and improves disease phenotype. In patients with DM1, a candidate gene association study reported a coding single nucleotide polymorphism (SNP) (rs26279, p.A1045T) in MSH3 exon 23 that was associated with the rate of somatic expansion. Genome-wide association studies (GWAS) in patients with Huntington’s disease identified variation in DNA repair genes that modify disease course, and pathway analyses in each study further highlighted DNA repair. Such variants also influence onset in other CAG expansion diseases, suggesting a common mechanism operates in conditions caused by repeat expansion. The lead variant in a recent GWAS linking MSH3 with Huntington’s disease progression was the imputed SNP rs557874766, which nominally results in Pro67Ala at the N-terminus.

4However, rs557874766 is located within a 9 bp tandem repeat in exon 1 of MSH3 and the 5′ UTR of the dihydrofolate reductase gene (DHFR) on the opposite strand. This repeat is polymorphic in copy number and sequence, which led us to hypothesize that rs557874766 could be an alignment artefact. Additionally, the 500-bp region flanking the MSH3 repeat is highly polymorphic, containing six SNPs and a 1-bp indel. We conducted targeted Illumina sequencing of the MSH3 exon 1 region in 218 Huntington’s disease and 247 DM1 subjects, which allowed us to obtain accurate haplotype information for the region. Using whole blood RNA-Seq in Huntington’s disease, we investigated whether sequence variation at the MSH3/DHFR locus influences their expression.

5The 218 Huntington’s disease subjects were from TRACK-HD. The DM1OPTIMISTIC cohort of 247 subjects was from OPTIMISTIC and the independent DM1CostaRica cohort of 199 subjects was previously reported in.

6Progenitor pure CAG length for Huntington’s disease was determined by MiSeq sequencing. Five subjects were excluded because they were part of a twin pair (n = 1) or the progenitor CAG length could not be unambiguously identified (n = 4) (Ciosi et al., unpublished results). DM1 progenitor allele length was determined by small pool PCR (; Cumming et al., in press). DM1 patients were tested for CCG repeat interruptions, known cis-modifiers of CTG repeat stability and disease phenotype (, in press).

7Two phenotypes were common to both cohorts: age at onset and rate of somatic expansion of the pathogenic CAG·CTG repeat. Huntington’s disease age at onset represents onset of motor symptoms. DM1 age at onset was subject self-assessment of the first occurrence of symptoms likely related to DM1 (Cumming et al., in press). Somatic CAG·CTG expansion in blood was previously quantified in both cohorts (Ciosi et al., unpublished results; Cumming et al., in press). For Huntington’s disease MiSeq data, the measure of somatic expansion was the proportion of reads in the sample that correspond to somatic expansions (reads with more CAG repeats than the progenitor allele) relative to the number of reads obtained for the progenitor allele (Ciosi et al., unpublished results). For DM1, it was the difference in number of repeats between the modal allele and the estimated progenitor allele length. In both cohorts, relative rate of somatic expansion corresponds to the variation in the measures of somatic expansion that is not explained by age and CAG·CTG repeat length. Positive values reflect a faster rate of somatic expansion.

8Two phenotypes were only available for Huntington’s disease; progression score and gene expression. Progression score was derived for 213 TRACK-HD subjects in Ciosi et al. (unpublished results), as described in. It measures typical Huntington’s disease progression that is not explained by age and pure CAG repeat length, with positive scores reflecting faster progression. Blood MSH3 and DHFR expression levels were available for 108 Huntington’s disease subjects.

9MiSeq amplicon sequencing, adapted from, was used to genotype the MSH3 exon 1 repeat and flanking variants (Supplementary Fig. 1). The region was amplified using locus-specific primers incorporating Illumina indexed adaptors (Supplementary Table 1). PCR was carried out using 10 ng of blood genomic DNA, 10% DMSO, 1 µM of each primer, 1× Custom PCR master mix (Thermo Scientific, SM0005), 0.048% (v/v) 2-mercaptoethanol and 0.5 U of Taq polymerase (Sigma) in a total volume of 10 µl. Thermal cycling conditions were: an initial denaturation at 96°C for 5 min, followed by 30 cycles of (96°C for 45 s), (60°C for 45 s) and (70°C for 2 min), with a final extension at 65°C for 1 min followed by 70°C for 10 min. Six hundred sequencing cycles were run 400 nt forward, 200 nt reverse. Quality control confirmed >80% of bases had Phred quality >30.

10Genotyping was conducted on the University of Glasgow Galaxy platform (heighliner.cvr.gla.ac.uk). Paired-end reads were merged and aligned to multiple references corresponding to potential 9 bp repeat alleles (Supplementary material), followed by variant calling. For repeat homozygotes, haplotypes were confirmed from.sam files using Tablet. The Galaxy workflow is available at https://www.myexperiment.org/workflows/5087.html. Conservation analysis used PhastCons and PhyloP (UCSC), with species sequence alignment in Clustal Omega.

11The transcriptome-wide association study (TWAS) method of was used to impute cortical gene expression from 452 dorsolateral prefrontal cortex samples from the CommonMind Consortium into the TRACK-HD GWAS of Huntington’s disease progression (n = 243). Following the approach, we tested association between imputed cortical gene expression and Huntington’s disease progression.

12Linear regression modelling of genotype-phenotype correlation was conducted in R. An additive genetic model was used to score genotypes. For age at onset analysis, we controlled for CAG·CTG repeat length in Huntington’s disease and DM1, and for repeat interruptions in DM1 (Supplementary Table 4). Meta-analysis of somatic expansion and age at onset in Huntington’s disease and DM1 was conducted with METAL. PLINK 1.07 was used to derive allele frequencies, Hardy-Weinberg equilibrium (HWE) and linkage disequilibrium. Haplotype relationships were visualized as a network using median joining on NETWORK.

13Data are available from the corresponding author on request.

14We observed 16 MSH3 repeat alleles, differing in sequence and length from three to nine repeats (Fig. 1A and Supplementary Table 2). Alleles contained combinations of five types of repeat units, with coding potential for proline or alanine (Fig. 1A). They were numbered by repeat length, suffixed alphabetically by frequency i.e. ‘3a’ represents the most common three-repeat allele.

15The most common allele in both cohorts, 6a (Fig. 1B), corresponds to the human reference sequence (NC_000005.10, GRCh38.p12). Illumina sequencing revealed that rs557874766 was not a SNP, but an alignment artefact resulting from the complex 9-bp repeat sequence (Fig. 1C). Individuals with the rs557874766 minor allele instead carry a three-repeat allele, 3a, the second most common allele observed in both cohorts. Two subjects with Huntington’s disease imputed as homozygous for the rs557874766 major allele were determined to be heterozygous for the 3a repeat allele by both Illumina and Sanger sequencing (Supplementary Fig. 2), highlighting the importance of directly genotyping such complex loci. We conclude that rs557874766 does not exist in the form of an SNP and results from incorrect alignment of the 3a allele to the reference 6a allele (Fig. 1C).

16The MSH3 exon 1 repeat region is poorly conserved between species, with mean scores of 0.29 [standard deviation (SD) 0.41] and 0.25 (SD 0.91) in PhastCons and PhyloP, respectively (Supplementary Table 3). Sequence alignment of 20 mammalian reference genomes showed most have two repeats (Supplementary Fig. 3). Together with a four- and a five-repeat allele, the 3a allele has been observed in gorillas and chimpanzees, suggesting 3a is an ancestral allele in humans.

17The 3a allele correlated negatively with relative rate of somatic expansion in subjects with Huntington’s disease (P = 0.032) and showed similar effect direction, though above nominal significance, in DM1 (P = 0.053) (Fig. 2 and Supplementary Table 2). Additionally, 3a was associated with delayed age at onset by 1.05 years (P = 0.0029) and slower progression in Huntington’s disease by 0.52 units (P = 3.86 × 10−7), which corresponds to 0.37 and 0.10 units per year on the UHDRS total motor score and total functional capacity, respectively. In DM1, the association between 3a and age at onset showed a consistent effect direction, approaching significance (P = 0.061). In meta-analysis, 3a was significantly associated with relative rate of somatic expansion (P = 0.004) and age at onset (P = 0.003) in Huntington’s disease and DM1. Detailed analysis of the relationship between repeat alleles and phenotypes (Supplementary Table 5) shows that the 3a allele accounts for the reduced somatic expansion rate, delayed onset and slower progression observed in Huntington’s disease. The association with somatic expansion appears to be driven by 3a homozygotes, whereas that with progression seems to follow an additive pattern with the number of 3a alleles. For onset, the pattern of association is unclear. In DM1, the number of seven-repeat alleles was associated with reduced expansion rate (Supplementary Table 5).

18In addition to testing repeat allele effects, we also assessed correlation between flanking SNP genotypes and disease phenotypes. All the flanking variants were in HWE (Supplementary Table 6) and in strong linkage disequilibrium with each other (Fig. 3B). Three variants (rs151182735, rs10168 and rs2250063) were in nearly complete linkage disequilibrium with the 3a allele, and as such were as significantly associated with phenotypes (Fig. 3A and Supplementary Table 6). All three are non-coding variants 5’ to the repeat and their alternative alleles are associated with reduced MSH3 and DHFR expression in the prefrontal cortex and in multiple tissues in GTEx (Supplementary Table 7). Three SNPs, rs1105524, rs1650697 and rs1677658, also correlated with some phenotypes, though not uniformly (Fig. 3A and Supplementary Table 6). Rs1105524 and rs1677658 are non-coding variants, whereas rs1650697 corresponds to Ile79Val. All three are expression quantitative trait loci (eQTL) for MSH3 and DHFR in the prefrontal cortex and in multiple tissues in GTEx (Supplementary Table 7). Previously, in a separate DM1 cohort (DM1CostaRica), reported association between both rs1677658 (P = 0.009) and rs10168 (P = 0.031) and somatic expansion, though neither survived correction for multiple testing for the candidate SNPs analysed. However, the direction of effect for both SNPs was the same as in the present study, and a significant association in meta-analyses with the two DM1 cohorts (rs1677658 P = 0.03, rs10168 P = 0.004) and all three DM1 and Huntington’s disease cohorts (rs1677658 P = 8.85 × 10−4, rs10168 P = 3.37 × 10−4) suggests these variants influence somatic expansion (Supplementary Table 6). reported an association between somatic expansion and age at onset, though the direct effect of MSH3 genotype on age at onset was not found to be significant. In the present study, meta-analyses of the two DM1 cohorts (rs1677658 P = 0.009, rs10168 P = 0.04) and all three DM1 and Huntington’s disease cohorts (rs1677658 P = 8 × 10−4, rs10168 P = 0.003) found the MSH3 genotype was significantly associated with age at onset (Supplementary Table 6). Meta-analyses of the three-repeat allele with all three DM1 and Huntington’s disease cohorts provide further support for its protective effect on somatic expansion (DM1OPTIMISTIC + DM1CostaRicaP = 0.004, DM1OPTIMISTIC + DM1CostaRica + Huntington’s disease P = 3.46 × 10−4) and age at onset (DM1OPTIMISTIC + DM1CostaRicaP = 0.04, DM1OPTIMISTIC + DM1CostaRica + Huntington’s disease P = 0.003) (Supplementary Table 2).

19The associations of SNPs with phenotypes were conditioned on the effects of MSH3 repeat alleles (Supplementary Table 8). As rs151182735, rs10168 and rs2250063 perfectly correlated with 3a, their independent effects could not be determined (Supplementary Table 6). With the exception of rs1677658 (linkage disequilibrium with 3a: r2 = 0.610) and rs1650697 (linkage disequilibrium with 3a: r2 = 0.143), whose alternative alleles were associated with delayed and early age at onset, respectively in the combined Huntington’s disease and DM1 meta-analysis (P = 0.015 and P = 0.029; Supplementary Table 8), there was no significant evidence for association between SNPs and expansion rate, onset or progression independent of repeat alleles.

20Considering variants with minor allele frequency >0.1 and all of the repeat alleles, we observed 25 haplotypes in the region, named Hap1 to Hap25 (Supplementary Table 9). The 3a repeat allele occurs on both Hap1 and Hap2, which differ only in the presence of the rs1677658 alternative allele on the more common Hap2. Hap1 was associated with reduced somatic expansion in DM1 (P = 0.032) and slower progression in Huntington’s disease (P = 0.020), whereas Hap2 was associated with reduced somatic expansion (P = 0.021) and delayed onset (P = 4.03 × 10−5) in both Huntington’s disease and DM1, and with slower progression (P = 1.64 × 10−5) and reduced expression of MSH3 (P = 0.024) and DHFR (P = 1.12 × 10−3) in Huntington’s disease (Supplementary Table 9).

21Overall, this analysis clarifies the sequence and variants present in MSH3 exon 1 and demonstrates that MSH3 repeat variants are associated with disease phenotypes in both Huntington’s disease and DM1.

22Each 3a allele was associated with reduced DHFR expression (P = 2.48 × 10−4; Fig. 4C) and homozygosity for 3a was associated with reduced MSH3 expression (P = 0.0273; Fig. 4B), whereas each 7a or 8a allele was associated with increased MSH3 expression (P = 8.55 × 10−4 and P = 8.26 × 10−3, respectively). The sum of MSH3 repeat lengths on both alleles appeared to correlate with MSH3 (P = 7.00 × 10−3) and DHFR expression (P = 1.76 × 10−3), which would suggest increasing repeat length increases expression of both (Supplementary Fig. 4). However, a more detailed analysis of MSH3 repeat alleles (Supplementary Table 5) shows the number of seven- or eight-repeat alleles is associated with increased expression of MSH3 (P = 4.53 × 10−6), and that this explains the apparent association with the sum of repeat lengths. In this relatively small cohort, MSH3 (age at onset P = 0.446, progression P = 0.440) and DHFR (age at onset P = 0.911, progression P = 0.284) expression in blood were not themselves directly associated with disease phenotype. MSH3 expression was not significantly associated with somatic expansion (P = 0.625), whereas the association of DHFR expression, while nominally significant (P = 0.049), did not survive correction for the number of phenotypes tested.

23In the detailed analysis, the number of three-repeat alleles was associated with reduced DHFR expression (P = 2.33 × 10−4; Fig. 4C), and this was sufficient to explain the apparent association of DHFR expression with other repeat alleles (Supplementary Table 5), including that observed with increasing total repeat length. DHFR and MSH3 expression are correlated (r2 = 0.120, P = 2.06 × 10−4; Fig. 4A). However, association between DHFR and three-repeat alleles remains significant after correcting for MSH3 expression (P = 7.51 × 10−4), and association between MSH3 and seven- or eight-repeat alleles remains significant after correcting for DHFR expression (P = 1.30 × 10−7). In the best-fitting model for DHFR expression, the alternative allele at rs1105524 (linkage disequilibrium with 3a: r2 = 0.192) increases and rs1650697 decreases DHFR expression independently of the three-repeat alleles (Supplementary Table 8). Otherwise, the repeat allele is the major determinant of MSH3 and DHFR expression, and there is no evidence of independent SNP effects.

24In a TWAS, increased expression of both MSH3 and DHFR in prefrontal cortex was associated with faster progression in TRACK-HD at similar levels of significance (P = 2.52 × 10−6 and P = 4.08 × 10−6, respectively; Supplementary Table 10), making it difficult to distinguish which is more functionally relevant. This ties in with the observation that SNPs significantly associated with somatic expansion, age at onset and progression (Supplementary Table 6) were eQTLs for both MSH3 and DHFR in CMC data. Notably, however, increased MSH3 expression was significantly associated with early onset (P = 1.71 × 10−3) in a TWAS of the GeM dataset, while DHFR expression was not significantly associated with onset (Supplementary Table 10). This favours MSH3 over DHFR expression as a modifier of Huntington’s disease course.

25MSH3 has recently been identified as a genetic modifier of somatic instability in DM1, and progression in Huntington’s disease. The MSH3 signal in the GWAS of Huntington’s disease progression was driven by an imputed SNP, rs557874766, located within a 9 bp tandem repeat sequence in exon 1 of MSH3, which is also in the 5′ UTR of DHFR on the opposite strand. MSH3 and DHFR are organized head-to-head, transcribed in opposite directions and are regulated by the same promoter. Here we demonstrate that rs557874766 is an alignment artefact and corresponds to a three-repeat allele, 3a, which was the shortest repeat allele observed and is likely ancestral. At the protein level, in silico modelling predicts that 6a results in the gain of a surface α-helix at the N-terminus of MSH3.

26A total of 16 MSH3 repeat alleles were observed, varying in sequence and length from three to nine repeats. Repeat alleles 6a and 3a are the first and second most common in this European cohort, though previous studies suggest a seven-repeat allele may be second most common in East Asian populations. In Huntington’s disease, 3a was associated with reduced somatic expansion, delayed onset and slower progression. In DM1, each 3a allele showed a trend towards reduced somatic expansion and delayed onset but was significantly associated with both measures in meta-analysis of Huntington’s disease and DM1. Longer seven-repeat alleles were associated with reduced somatic expansion only in DM1. Whether this reflects a subtle difference in MSH3 biology between the two disorders, or simply a sampling error, remains undetermined.

27The MSH3 repeat lies between binding domains for PCNA and EXO1, both of which are involved in mismatch repair (MMR). PCNA is a sliding clamp that participates in DNA replication, but in MMR it delivers MSH proteins to mismatches and increases binding specificity. EXO1 excises the daughter strand after mismatch recognition, as well as being involved in end resection during homologous recombination. The MSH3 repeat region is poorly conserved between species, with other mammals having between zero and five repeats. This lack of evolutionary constraint suggests functional redundancy in the MMR pathway and a lack of a major effect of N-terminal MSH3 variation outside the context of repeat expansion disease. Unlike other MMR components, germline heterozygous MSH3 mutations are not associated with increased risk of cancer, most likely because MSH2/MSH6 can also initiate repair at replication errors.

28Three non-coding variants 5’ of the repeat were in near complete linkage disequilibrium with 3a, so it is not possible to determine their independent effects on disease phenotypes. All three are associated with reduced MSH3 expression in multiple tissues, including cortex (CMC and GTEx). Controlling for repeat alleles, no SNPs were significantly associated with phenotypes, except the intronic rs1677658 and the exon 1 rs1650697 variants, which contributed to delayed or early onset, respectively in the combined Huntington’s disease and DM1 dataset. Rs1677658 was associated with reduced MSH3 and DHFR expression (CMC and GTEx), whereas rs1650697 was associated with increased DHFR in Huntington’s disease blood, as well as multiple tissues in GTEx. Hap2, the MSH3 haplotype most significantly linked with reduced somatic expansion and delayed onset in Huntington’s disease and DM1, and with slower progression in Huntington’s disease, contains the 3a allele, along with alternative alleles of non-coding variants rs151182735, rs10168 and rs2250063, which are in complete linkage disequilibrium with it, and rs1677658. It is thus difficult to assess which (if any) MSH3 variants (repeats or SNPs) are driving associations with disease phenotypes, and further investigation in a larger sample is warranted.

29Whole blood transcriptomic analysis in a subset of the Huntington’s disease patients found the 3a allele was associated with reduced expression of MSH3 and DHFR, and seven- or eight-repeat alleles with increased MSH3 expression. DHFR, which shares a promoter with MSH3, is a ubiquitously expressed enzyme involved in purine, thymidylic acid and amino acid synthesis, but has not previously been implicated in Huntington’s disease pathogenesis. Our TWAS found that increased expression of MSH3 and DHFR in cortex are associated with faster Huntington’s disease progression. While MSH3 expression was significantly associated with early onset in our GeM TWAS (P = 1.71 × 10−3), DHFR expression was not associated with disease course. This is consistent with Huntington’s disease mouse brain, in which expression of MSH3, but not DHFR, correlates with somatic expansion. Neither MSH3 nor DHFR expression in blood was significantly associated with somatic expansion, onset or progression in this sample. However, investigation in a larger sample, or in a more relevant tissue, such as striatum, would be of interest.

30Collectively, our results suggest the MSH3 3a repeat allele reduces somatic expansion and improves phenotype in both Huntington’s disease and DM1, potentially through altering MSH3 expression levels. However, given the proximity of the repeat region to MMR protein binding domains, the 3a allele could also alter MSH3 function in the recognition and repair of insertion-deletion loops, double-strand breaks or single-strand annealing. Repetitive DNA sequences form unusual secondary structures such as slipped strands, hairpin loops, G-quadruplexes and R-loops, the stability of which correlates with expansion. MSH3 may recognize these structures and initiate repair, during which out of register synthesis could result in repeat expansion. This preliminary study elucidates variation in MSH3 that modifies Huntington’s disease and identifies the same signal in an independent trinucleotide repeat disease. Though beyond the scope of the present study, in the future it will be important to replicate these findings in additional independent cohorts for each disease. Together, these results suggest a common mechanism, involving somatic expansion, operates in vivo in distinct trinucleotide repeat diseases to influence disease course. Therefore, modulation of MSH3 has significant therapeutic potential in a range of diseases caused by repeat expansions.