Analysis of DNM3 and VAMP4 as genetic modifiers of LRRK2 Parkinson’s disease

The LRRK2 gene has rare (p.G2019S) and common risk variants for Parkinson’s disease (PD). DNM3 has previously been reported as a genetic modifier of the age at onset in PD patients carrying the LRRK2 p.G2019S mutation. We analyzed this effect in a new cohort of LRRK2 p.G2019S heterozygotes (n = 724) and meta-analyzed our data with previously published data (n = 754). VAMP4 is in close proximity to DNM3, and was associated with PD in a recent study, so it is possible that variants in this gene may be important. We also analyzed the effect of VAMP4 rs11578699 on LRRK2 penetrance. Our analysis of DNM3 in previously unpublished data does not show an effect on age at onset in LRRK2 p.G2019S carriers; however, the inter-study heterogeneity may indicate ethnic or population-specific effects of DNM3. There was no evidence for linkage disequilibrium between DNM3 and VAMP4. Analysis of sporadic patients stratified by the risk variant LRRK2 rs10878226 indicates a possible interaction between common variation in LRRK2 and VAMP4 in disease risk.


Introduction (words: 427)
The p.G2019S coding variant in the LRRK2 gene is the most common high penetrance mutation causing parkinsonism. The mutation occurs in 1-40% of PD cases, varying by ethnicity (1). LRRK2 parkinsonism is broadly similar to "idiopathic" disease in clinical manifestations and age at onset (AAO), generating interest in its potential as a therapeutic target with broader application to PD (2). In addition to rare pathogenic mutations, genome wide association studies (GWAS) show that common variation in LRRK2 is a risk factor for sporadic PD (3) (23).
There is a wide range in PD AAO among p.G2019S carriers. It is estimated that the heritability of PD AAO (~0.11) is lower than the overall heritability of risk for PD (~0.27) (4). Currently the strongest genetic association with AAO is through genetic risk score (GRS). There is overlap between the heritability of risk, as quantified in the GRS, and AAO. As LRRK2 is a functionally complex protein with many protein-protein interaction domains, there may be diverse genetic modifiers that vary in effect size and frequency. Trinh and colleagues (5) presented evidence of modification by a common variant haplotype tag in DNM3 (rs2421947), for which the median onset of LRRK2 parkinsonism for GG homozygotes was 12·5 years younger than that of CC homozygotes (direction of effect GG < GC < CC), suggesting that the C allele or a linked variant may be protective against the development of PD. Similarly, Fernandez-Santiago and colleagues (6) observed a consistent trend for the rs2421947 G allele on AAO in LRRK2 p.G2019S carriers, with median onset 3 years younger in patients with the G allele, though the result was not significant. Foo and colleagues (2018) found no impact of DNM3 in PD AAO in individuals carrying Asian LRRK2 risk alleles (24).
In the largest GWAS meta-analysis of idiopathic PD the nearby VAMP4 gene variant rs11578699, located 113,325 bp from DNM3 rs2421947, emerged as a statistically significant PD risk association, raising the possibility the effect of DNM3 may relate to linkage disequilibrium with VAMP4 (7). LRRK2 has a role in both monogenic and idiopathic/sporadic forms of PD and overlap in genetic modifiers may exist.
We carried out a large replication candidate gene study investigating the impact of reported genetic modifier DNM3 rs2421947 and the nearby GWAS association VAMP4 rs11578699 on AAO. We specifically sought to replicate the discovery finding, to determine whether the association might vary with ethnicity and whether linkage disequilibrium with VAMP4 may be relevant to this association. We analyzed a multi-ethnic cohort of European, North African, and Ashkenazi Jewish LRRK2 p.G2019S carriers and a larger cohort of European patients with and without the LRRK2 common risk allele.

Data collection
Patient cohorts: LRRK2 p.G2019S heterozygotes were identified from cohorts in the International Parkinson's Disease Genomics Consortium (IPDGC) and other collaborative centres. Data from 724 LRRK2 p.G2019S heterozygotes was contributed by the National Institute of Health (NIH), University College London (UCL), McGill University (MU) (samples were collected in Columbia University (CU) and Sheba Medical Centre (SMC)), Sorbonne University (SU) (in collaboration with Habib Bourguiba hospital, Sfax, Tunisia, and Blida Hospital, Blida, Algeria), National Institute of Neurological Disorders and Stroke (NINDS), and the University of Tübingen (TU). Data from two published studies analyzing DNM3 rs2421947 and LRRK2 p.G2019S parkinsonism with AAO was also included comprising 754 participants (5,6). All studies willing to participate and currently holding minimum required data for LRRK2 p.G2019S carriers were included: PD onset age or sampling age for asymptomatic carriers; and DNM3 rs2421947 or a high RSQ proxy SNP. Where these studies had also genotyped VAMP4 rs11578699 or a high RSQ proxy, these data were also included in the analysis of the VAMP4 gene (n=786). Patients with PD without a known Mendelian or high risk genetic cause of disease were collated from the UK-wide Tracking Parkinson's study (8) and from IPDGC datasets. Summary statistics for DNM3 rs2421947 and VAMP4 rs11578699 were obtained from the most recent large-scale PD GWAS (7) and PD AAO GWAS (4). All studies were approved by each respective institutional ethics review committee and all participants provided written informed consent. All studies were carried out in accordance with the Declaration of Helsinki (9).

Genetic analyses:
Genotyping was performed on different platforms (Neurochip array, NeuroX array, TaqMan assays, Infinium OmniExpress-24, and Illumina HumanCore Exome array) at participating centres. The LRRK2 p.G2019S mutation was directly genotyped. Sanger sequence or KASPar verification of the DNM3 rs2421947 variant was carried out on data subsets at study specific centres. Where the variant DNM3 rs2421947 was imputed, only RSQ values above 0.85 were used, and only RSQ values above 0.75 were used for VAMP4 rs11578699 (table-e1), as the RSQ quality of available data for variants VAMP4 rs11578699 was lower. 1478 LRRK2 p.G2019S carriers were identified across study cohorts. 2052 samples from patients with idiopathic PD from the Tracking Parkinson's disease study passed quality control and had no identified genetic cause of disease (8). 96 patients with PD were excluded due to missing AAO data; therefore 1956 patients with PD and without LRRK2 p.G2019S were included in further analysis. Summary statistics and patient data from the most recent PD GWAS included over 37.7K cases, 18.6K 'proxy-cases' and 1.4M controls. The most recent PD AAO GWAS included data from 28.6K patients (4,7).

Statistical analyses:
Data from the studies were pooled. We calculated Hardy-Weinberg Equilibrium (HWE) for VAMP4 rs11578699 and DNM3 rs2421947 in p.G2019S carriers. QC HWE for iPD and controls is summarized in the recent large-scale PD GWAS. We assessed linkage disequilibrium (LD) measures RSQ and D' between DNM3 rs2421947 and VAMP4 rs11578699 in different populations to evaluate the possibility that an extended haplotype block might explain the relationship between DNM3 and LRRK2 (table e-2). The allele based Fisher's exact test was used to compare DNM3 rs2421947 and VAMP4 rs11578699 minor allele frequencies (MAF) between LRRK2 p.G2019S heterozygotes of different ethnic backgrounds and with idiopathic PD (table 1). Student's t-tests and ANOVA were used to assess differences in AAO by genotype (table 2). We then analyzed the effect of DNM3 rs2421947 (figure 1) on AAO in idiopathic PD in the Tracking Parkinson's dataset using linear regression of AAO, and Kaplan Meier survival analysis. Next, we analyzed patients with LRRK2 p.G2019S of Ashkenazi Jewish, North African and European ethnicity using linear regression of AAO with available covariates of sex, ethnicity, relatedness and study centre of origin (figure 2). AAO was regressed by DNM3 rs2421947 genotypes, and separately for rs11578699 VAMP4 genotypes. We meta-analyzed DNM3 rs2421947 data from all previously unpublished datasets using linear regression models (figure 2B). We pooled these data with previously published data and metaanalyzed again with the same methods (figure 2C and D). We then analyzed VAMP4 rs11578699 impact on AAO with linear regression (figure 2E and F). Bonferroni or other corrections for multiple testing were not performed as this is a candidate gene-based study. Finally, we analyzed the impact of VAMP4 rs11578699 on AAO in idiopathic PD carrying the LRRK2 rs10878226 variant. LRRK2 rs10878226 has also been implicated in PD risk (odds ratio [OR], 1.20; 95% confidence interval [CI], 1.08-1.33; p = 6.3× 10−4 , n=6129) (3).

Results
DNM3 rs2421947 and VAMP4 rs11578699 genotypes were in Hardy-Weinberg equilibrium in controls.
There was no LD between DNM3 rs2421947 and VAMP4 rs11578699, as assessed through RSQ and D', within or across population samples, indicating that these are independent variants (table e-2) We investigated the possibility that ethnic variation in allele frequency might explain a variable effect of DNM3 on affected LRRK2 penetrance. Aside from the Norwegian cohort DNM3 rs2421947 MAF (C) of 0.69 (vs. other Europeans p=0.00012), all DNM3 rs2421947 MAF in the study were between 0.43-0.51 (table 1). There was no difference between Ashkenazi Jews and European frequencies (p=0.94); or between Ashkenazi Jews and North Africans (p=0.28); nor between Europeans and North Africans (p=0.54) at this locus. Publicly available allele frequencies (http://gnomad.broadinstitute.org) were similar (table 1). VAMP4 rs11578699 MAF were between 0.15-0.30 in different populations.
We assessed possible association between these variants and p.G2019S genotype using the allele based Fisher's exact test. There was no association between DNM3 rs2421947 and p.G2019S genotype in all PD cases (p=0.42). There was no significant difference in DNM3 allele frequencies between patients with PD and asymptomatic heterozygotes (asymptomatic p.G2019S heterozygotes were of Ashkenazi Jewish, North African, and European non-Jewish ethnicity), as assessed through Fisher's Exact test (p=0.056). When the previous data was meta-analyzed with the current data there was a significant association between DNM3 rs2421947 and PD affected status in p.G2019S heterozygotes (p=3.1x10 -5 ) by the Fisher's Exact test, although there were differences between populations (Table 1). Mean sampling age for p.G2019S heterozygotes without PD was 56 years across all ethnicities; 63 years in AJ subjects; and 56 years for Europeans.
In the most recent PD GWAS there was no genome-wide significant association between DNM3 rs2421947 and PD by logistic regression, (p=0.0051). There was no association between VAMP4 rs11578699 and LRRK2 p.G2019S status in European PD (p=0.60) and no significant difference in VAMP4 rs11578699 MAF between LRRK2 p.G2019S patients and asymptomatic carriers by the Fisher's Exact test (p=0.64). We then assessed the relationship between AAO and DNM3 rs2421947 and VAMP4 rs11578699 (table 2) using t-tests and ANOVA. There was no effect of DNM3 rs2421947 on AAO as assessed through ANOVA (p=0.55, F=0.59, df=2, n=708). When meta-analysed with data from previous cohorts there was a nominal effect of DNM3 on AAO (2 years difference between G and C genotypes), (p=0.021, F=3.88, df=2, n=1304).  Figure 1A), and LRRK2 p.G2019S carriers ( Figure 1B-D). The same method was also used to visualise risk for VAMP4 rs11578699 genotypes in individuals with LRRK2 p.G2019S ( Figure 1E and F). We carried out a multi-ethnic meta-analysis of newly contributed data of DNM3 rs2421947 GG versus CC and CG genotypes against onset age in 724 LRRK2 p.G2019S carriers, using a random effects model on disease-free survival. Linear regression meta-analysis on PD AAO was not significant (beta = -1.19, p = 0.55, n =708), though I 2 heterogeneity was 86.9%, p < 0.01. The impact of GG versus CC and CG was not significant in sub-group analyses (figure 2). When our new data was pooled with the two previous studies and meta-analysed, linear regression meta-analysis of LRRK2 p.G2019S AAO was not significant (beta = -2.21, p = 0.083, n = 1304). I 2 total heterogeneity for linear regression meta-analysis was 82.3%, p<0.0001 (figure 2).
We evaluated the possibility of ethnic specific effects using linear regression of AAO (figure 2). Though in each subgroup analysis confidence intervals overlapped zero, there appeared to be a trend towards earlier onset in G allele carriers of Ashkenazi Jewish ethnicity, consistent with previously reported data, and in the discovery data effects were strongest in Arab Berbers suggesting ethnic specific effects.
Using cox proportional survival analysis, we studied the discovery data and subsequent cohorts separately (GG versus CC and CG as described previously). There was a strong association between rs2421947 and PD AAO in discovery data (hazard ratio We meta-analyzed the samples studied in this paper with the discovery data ( Figure 2). Finally, we investigated the possibility that might be an interaction between the GWAS defined common LRRK2 risk allele (rs10878226) and VAMP4. We analyzed 4882 cases with PD carrying the LRRK2 risk variant minor allele using linear regression of AAO and plotting of the Kaplan Meier curve, as shown in Figure 3. TT versus TC and CC VAMP4 rs11578699 genotype was nominally significantly associated with AAO risk in this cohort (beta=1.68, se=0.81, p=0.037, n=4882). VAMP4 rs11578699 was not significantly associated with AAO risk in idiopathic PD cases without the LRRK2 rs10878226 risk variant (beta=-0.28, se=0.48, p=0.56, n=14970). Far fewer patients with idiopathic PD had been genotyped for DNM3 rs2421947 in the PD GWAS cohort (table 1), meaning we were not able to assess interaction between DNM3 and the LRRK2 risk allele.

Discussion
We have analyzed the effect of DNM3 rs2421947 on AAO in LRRK2 p.G2019S parkinsonism.
We have not replicated the association between DNM3 and LRRK2 p.G2019S AAO in this study or in meta-analysis with all available data. In the original discovery analysis, there was a difference of 12.5 years between DNM3 rs2421947 GG and CC genotypes (meta-analysis HR 1·61, 95% CI 1·15-2·27, p=0·02). In our data the AAO difference between GG and CC genotypes was 1.4 years, which was not significant. Using AAO regression, meta-analysis of independent sample series in newly genotyped samples did not identify a significant difference in AAO between genotypes (beta = -1.19, p = 0.55, n =708). Similarly, when discovery and replication data was analyzed together there was no significant effect of DNM3 rs2421947 on AAO (beta = -2.21, p = 0.083, n = 1304). However, there was significant heterogeneity in replication and combined DNM3 regression meta-analyses.
In "sporadic" PD, consistent with the recent AAO GWAS (4), our study indicated that DNM3 rs2421947 does not affect AAO in non-p.G2019S disease.
We did not identify linkage disequilibrium between DNM3 and VAMP4, nor an independent effect of VAMP4 on LRRK2 penetrance. Analysis of carriers of a common LRRK2 risk SNP provided nominally significant support for a potential interaction between VAMP4 and LRRK2, which requires replication in larger sample sizes. Gene expression analysis indicated that VAMP4 rs11578699 is an eQTL for VAMP4 expression.
One possibility for the lack of replication of DNM3 modification of p.G2019S AAO is that heterogeneity in the sample may limit the observed effect in the meta-analysis; this has previously led to non-replication in independent samples in genome-wide association studies (13) (14). Significant heterogeneity was observed in the linear regression meta-analysis of replication and combined data in this study, which may be explained by ethnicity effects. The impact of DNM3 may vary between ethnicities or be population specific, and the strongest effects were seen in Ashkenazi Jewish p.G2019S carriers although not reaching significance ( Figure 2). Further studies in large numbers of Ashkenazi Jewish and Arab patients are needed.
Due to its role in innate immunity LRRK2 may have been under different selective pressures in human evolution, relating to differences in environmental pathogens. Interestingly, LRRK2 p.G2019S penetrance varies across ethnic groups. Hentati  These data imply that there may be potential protective factors in the relatively homogeneous Norwegian and Ashkenazi Jewish population and that ethnic and population effects are likely to be very important in analyzing this variant. Our analysis and comparison of background allele frequencies in this study suggest that this is unlikely to primarily relate to VAMP4 and DNM3. However, there may be other rare variant effects that will emerge with fine mapping of this region.
The LRRK2 common risk variant analysis is of interest; the variant is not a proxy for p.G2019S in Europeans, so this represents an independent marker of PD risk. Around a quarter (24.6%) of sporadic PD cases carry the common LRRK2 risk variant rs10878226, which is associated with PD (combined odds ratio [OR], 1.20, 95% CI, 1.08-1.33, p=6.3×10 -4 , n=6129) (3). Our common LRRK2 variant analysis indicates a possible interaction between common variation in LRRK2 and VAMP4, which requires further study. VAMP4 and LRRK2 are both involved in synaptic vesicle dynamics, which has relevance to the etiology of PD. Underscoring this possibility, analysis of VAMP4 and DNM3 expression indicated that both are highly expressed in the brain, though these are not in LD and likely represent independent signals.
Modifiers of the penetrance of LRRK2 p.G20129S are likely to be therapeutic targets and may be important in genetic counselling. Our large study, aggregating new and previously published data has indicated significant heterogeneity across studies, but not provided robust replication of an interaction between DNM3 and LRRK2 p.G2019S. Further genome-wide studies in different populations are needed, to resolve the determinants of the variable penetrance seen in individuals with p.G2019S.

Supplementary methods
For the meta-analyses data was divided into the same ethnicity subgroups (Ashkenazi Jewish, North African, and European), and study centre specific groups. Where there were <60 datapoints from a particular study centre for DNM3 rs2421947, data was pooled, and study centre of origin was used as a covariate. Cox proportional hazards model analyses and linear regressions for p.G2019S carriers were carried out on GG versus CC and CG genotypes, as exploratory data analysis indicated a dominant protective effect of the minor C allele. For the VAMP4 rs11578699 variant these survival methods were carried out on CC versus TC and TT due to the smaller number of TT genotypes available. Family relatedness, sex, and principal components of ancestry were used as covariates, where available, in each analysis.
A random-effects model was used in meta-analyses. Natural logs of hazard ratios for GG versus CC and CG genotypes were taken for DNM3 rs2421947, and for CC versus TC and TT for VAMP4 rs11578699. The log of standard errors was calculated using outputted hazard ratio confidence intervals, using the Taylor expansion with leading term (also known as the delta method) (17).

Supplementary results
Cox proportional hazard analysis DNM3 did not influence time to the development of disease in idiopathic PD by the Cox proportional hazards model (hazard ratio [HR] 0.98, 95% CI 0.89-1.07, p = 0·60, n=1956) for GG versus CC and CG carriers. DNM3 rs2421947 did not significantly affect the age associated hazards of developing p.G2019S parkinsonism through random effects meta-analysis in our previously unpublished data (hazard ratio [HR] 1.09, 95% CI 0.95-1.25, p = 0.20, n=724), as shown in Figure 2 . I 2 heterogeneity was <0.01% (p = 0.42). When our new data was pooled with the two previous studies and meta-analysed, there was a nominally significant effect of DNM3 rs2421947 on AAO in LRRK2 p.G2019S parkinsonism ([HR] 1.14, 95% CI, 1.02-1.27, p = 0.025, n = 1478). I 2 total heterogeneity was 22.8%, p=0.37. VAMP4 rs11578699 was nominally associated with disease AAO in idiopathic PD cases carrying the LRRK2 risk variant rs10878226 (TT versus CC and TC (hazard ratio [HR] 0.86, 95% CI 0.76-0.98, p=0.023), and not associated with AAO in idiopathic PD cases not carrying the LRRK2 risk variant rs10878226 (hazard ratio [HR] 1.02, 95% CI 0.94-1.10, p=0.47). figure e-1. Regional association plot of PD cases versus control from the largest PD GWAS, identifying rs11578699 as the lead SNP Uncorrected for multiple testing c There are insufficient p.G2019S carriers in these categories for the t test to be calculated.