Genome-wide determinants of mortality and clinical progression in Parkinson’s disease

MDS-UPDRS

p5/ 30   investigating PD progression, and many candidate variant studies.However, no genome-wide studies have reported on survival/mortality in PD.

Added value of this study:
To our knowledge, this is the first GWAS of survival in PD.Our study highlights new loci influencing survival in PD, including TBXAS1 and SYT10.We also conducted GWASs of progression to other clinical milestones, Hoehn and Yahr stage 3 or greater, and cognitive impairment.We show that APOE influences both mortality and cognitive progression in PD, and report an additional locus influencing expression of ADORA2A which affects the rate of motor progression.
Implications of all the available evidence: With the exception of APOE, we report new loci which have not been previously associated with PD progression or for mortality and ageing in the general population.These loci could be investigated in functional studies as potential drug targets to stop or slow progression of PD.In addition, new genetic loci can help to improve our understanding of the biology of PD progression and prediction of progression.Further replication of these loci is also needed in independent, longitudinal PD cohorts.

INTRODUCTION
Parkinson's disease (PD) is a progressive neurodegenerative condition for which there are no treatments to stop or slow disease progression.Large-scale genome-wide casecontrol association studies (GWASs) of PD have identified 90 independent variants associated with disease risk 1 .However, it is also important to study the genetics and biology of disease progression.This will facilitate the development of potential diseasemodifying treatments.There have now been a handful of GWAS which aim to identify genetic variants associated with progression in PD.These have nominated loci in SLC44A1 (encoding choline transporter like protein -1, involved in membrane synthesis) for progression to Hoehn and Yahr (H&Y) stage 3 or greater, 2 APOE for cognitive progression, 3 and RIMS2 (encoding the RAB3 interacting RIMS2 protein, involved in neurotransmitter release) for progression to PD dementia. 4In addition, many candidate gene studies have shown that variants in GBA, APOE, and potentially MAPT, are associated with the rate of PD motor and cognitive progression. 5 progression may be determined by differential cellular susceptibility, related to mitochondrial function or proteostasis, differential cell to cell spread of pathology, or novel pathways and mechanisms.Risk factors determined from case control studies indicate aetiological pathways and guide future preventive trials, these may or may not differ from risk factors that determine disease progression.Currently disease modifying treatment trials focus on intervention in recently diagnosed patients, related to disease progression after diagnosis.Work on large scale longitudinal cohorts over the last ten years has now enabled the collaborative study of large clinico-genetic datasets.Here we have carried out GWAS of progression to three key clinical milestones in PD: mortality, H&Y 3 or greater, and cognitive impairment.We have analysed data from 6,766 PD patients with over 15,340 visits and mean follow-up ranging between 4.2 to 15.7 years.

METHODS
We studied 11 cohorts from Europe and America, and included cohorts in each analysis who had sufficient data on the outcomes of interest (see Supplementary Materials).Genotyping, quality control, and imputation was performed in each cohort separately but following the same steps.Only variants with high imputation quality scores (INFO/R2 > 0.8) and minor allele frequency > 1% were retained for analysis.
We assessed the following clinical outcomes: mortality, H&Y stage 3 or greater, and cognitive impairment.26, dementia using the DSM classification, or dementia using the MDS criteria, or study withdrawal due to reported dementia or cognitive impairment.These criteria were based on previous studies in these cohorts 2,6,7 .Cohorts were excluded if less than 20 individuals met the outcome of interest during the follow-up period, or < 5% of the total cohort size.Progression to each clinical milestone from the starting time point of PD symptom onset was assessed using Cox proportional hazard models.We adjusted for age at onset, gender, and the first five genetic principal components to adjust for population stratification.Meta-analysis was performed in METAL (version 2011-03-25) 8 , using an inverse variance weighted fixed effects model.GWASs with a genomic inflation factor above 1.2 were excluded from the meta-analysis.Only SNPs that were present in > 1,000 individuals were included in the final results.SNPs with heterogeneous effects across cohorts were also excluded (p-value < 0.05 for Cochran's Q-test for heterogeneity, and/or I 2 > 80).The null hypothesis was tested with the standard GWAS significance level of 5 x 10 -8 .Results were uploaded to Functional Mapping and Annotation of GWAS (FUMA; https://fuma.ctglab.nl/) 9 to annotate, prioritise, and visualize GWAS results and perform gene-set analysis with MAGMA.Forest plots for the top SNPs were generated in R v3.6 using the forestplot package.Further details are provided in the Supplementary Materials.
We also performed candidate variant analysis of the 90 PD risk variants from the most recent PD case-control GWAS, 1 and the cumulative PD genetic risk score (GRS).We also examined associations for other candidate variants that have been reported in PD and Progressive Supranuclear Palsy (PSP) progression: SLC44A1, RIMS2, WWOX, TMEM108, APOE ε2 allele, MAPT H1 haplotype, and rs2242367 adjacent to the LRRK2 locus.We analysed the Alzheimer's disease (AD) GRS in relation to PD progression.38 loci passing genome-wide significance from the latest AD GWAS were used to create the AD GRS. 10 The APOE region was excluded from the GRS (19:40,000,000-50,000,000). 10 To clarify whether our genetic results for mortality were related to PD or non-PD causes (e.g.general immunity, cardiovascular disease, or COVID), we performed a competing risk analysis using the Fine-Gray method for each of the top SNPs to evaluate whether the SNPs were related to PD death or non-PD death (Supplementary Materials).

RESULTS
Overall 6,766 participants with PD were analysed with mean follow-up between 4.2 and 15.7 years (Table 1).We did not have data from regular follow-up visits for all studies as some studies only contributed mortality data.However in the studies that had regular follow-up visit data available, over 15,340 visits were analysed (Table 1).p8/30

GWAS of mortality
One study (Cambridge PD Research Clinic) was excluded from the meta-analysis of mortality because the study-specific genomic inflation factor was above 1.2 (Supplementary Table 1).The PPMI study was also excluded because less than 20 individuals reached this endpoint.Two loci passed genome-wide significance and were identified to determine mortality in PD (Figure 1).The top SNP was rs429358 in Chromosome 19 (p=4.0x10-10 ), which tags the APOE ε4 allele (Table 2).One other loci in Chromosome 7 in TBXAS1 also reached significance (p<5 x 10 -8 ), and another loci in Chromosome 12 near SYT10 was nominally associated (p=5.3x10 - ).Regional association plots are shown in Figure 1 and Supplementary Figures 1-3.The top ten independent SNPs identified from FUMA and the nearest genes are reported in Table 2.
In the MAGMA gene-based test, APOE was significantly associated with mortality (p=1.9x10-10 ), and SYT10 was associated just below genome-wide significance (p=3.6x10-6 ).There was no significant association of any gene-sets or tissues in the MAGMA gene-set analysis and gene property analysis for tissue specificity.
In TBXAS1, there were two SNPs associated with mortality, rs4726467 and rs144889025, identified as independent in FUMA with r 2 <0.6 (Table 2).However in LDpair these two SNPs are correlated with D'=0.82 and r 2 =0.57, in non-Finnish European populations (https://ldlink.nci.nih.gov/).To determine whether these were independently associated with mortality, we performed conditional analysis by including both SNPs in the model.When we did this, the SNPs were no longer significantly associated (both p-values > 0.05), indicating they are not independent signals.
The top SNP in Chromosome 12, rs10437796, is not directly within SYT10 but increases SYT10 expression in the testis and decreases expression in the tibial nerve.
The SNP also increases expression of the long noncoding RNA (lncRNA) RP11-438D14.2(ENSG00000259937) in the brain.This is a 'sense intronic' transcript to SYT10, a long non-coding transcript which is within an intron of a coding gene and does not overlap any exons.Brain cis-eQTL data from MetaBrain showed that the effect allele A of rs10437796 significantly increased expression of SYT10 in cortex but not other brain regions.This SNP was not an eQTL or sQTL for any genes in blood in eQTLGen.There were no coding variants in LD with this SNP in LDproxy within a 500kb window.
We also performed colocalization analysis to determine whether the association signals for PD mortality and gene expression are driven by a shared causal variant (see Supplementary Materials; Supplementary Table 2).We used cis-eQTL data from PsychENCODE and eQTLGen to examine gene expression in whole brain or blood, respectively.However, no PD mortality loci showed evidence of colocalization with eQTLs (PP.H4<0.75).

Cause of death analysis
To clarify whether our genetic results for all-cause mortality were related to PD or non-PD causes (e.g.general immunity, cardiovascular disease), we performed a competing risk analysis for PD death and non-PD death.In the UKB and QSBB cohorts which collected cause of death data, we classified the primary cause of death as either 1) related to PD and end of life (e.g.pneumonia, aspiration pneumonia, bronchopneumonia), or group.For the top APOE SNP rs429358, increased mortality appeared to be driven by PD deaths (HR=1.47,p-value=2.6x10 - ) (Figure 2A, Table 3).However for the TBXAS1 SNPs, rs4726467 and rs144889025, there was no significant effect of the SNPs on PD deaths (p-values>0.05)(Figure 2B, 2C).These two TBXAS1 SNPs were nominally associated with increased non-PD deaths (HR=1.63,p-value=0.01,and HR=1.54, p-value=0.03respectively).The SYT10 SNP, rs10437796, was nominally protective against PD death (HR=0.71,p=0.05) and increased non-PD deaths (HR=1.50,p=1.8x10 -4 ) (Figure 2D).This suggests that APOE ε4 is having a specific effect on PD mortality but the other SNPs may not be PD specific.

PD Progression GWAS
Tan and colleagues p13/30 GWAS of H&Y stage 3 or greater 3,299 individuals were analysed for progression to H&Y stage 3 or greater.The Oslo cohort was excluded as the genomic inflation factor was greater than 1.2.After exclusion of the Oslo dataset, 872 individuals (26.4%) met the outcome of H&Y stage 3 or greater, with a median time to H&Y 3 of 5.5 years.6,553,504 SNPs passed filtering for heterogeneity and MAF variability.The genomic inflation factor of the meta-analysis was 1.01.The top ten independent SNPs from FUMA are reported in Table 4.One locus in Chromosome 22 was significantly associated with progression to H&Y stage 3 or greater, with the lead SNP rs112809886 (p=1.9x10-9 ), close to GGT5 (Figure 3).The regional association plot is shown in Supplementary Figure 4.
The top SNP, rs112809886, is an eQTL for ADORA2A, with the alternate allele increasing gene expression in the tibial nerve and cerebellar hemisphere of the brain.It also decreases gene expression of UPB1 and ADORA2A in whole blood (GTEx).In the eQTLGen data for blood eQTLs, it is a significant cis-eQTL for UPB1, SUSD2, GSTT1, and AP000351.10.Brain QTL data from MetaBrain showed that the effect allele A of rs112809886 increased expression of ADORA2A in the but not any other brain regions.

GWAS of cognitive impairment
3,565 individuals were analysed for cognitive impairment.Of those, 1,081 (30.3%) met the outcome of cognitive impairment with a median time of 5.0 years.6,713,896 SNPs passed filtering for heterogeneity and MAF variability.The genomic inflation factor was 1.02.The top ten independent SNPs from FUMA are reported in Table 5.
The top SNP was rs429358 in Chromosome 19, which is the same APOE ε4 tagging SNP as identified in the mortality GWAS (Figure 4).The minor allele C was associated with more rapid progression to cognitive impairment (HR=1.54,p=5.5x10 -13 ).
Figure 4. GWAS meta-analysis of progression to cognitive impairment.

Candidate variant analysis
We did not find that any of the 90 PD risk SNPs were associated with PD progression at genome-wide significance (Supplementary Materials; Supplementary Table 4).Only one variant, rs35749011, near KRTCAP2 but tagging the GBA p.E326K variant, was associated with mortality (p=3.6x10 - ) below the analysis-wide significance threshold (pvalue threshold 0.05/88=0.00057)and nominally associated with cognitive impairment (p=0.02).There was also no association between the PD GRS and any of the progression outcomes (Supplementary Materials).In the candidate variant analysis, only the PSP survival SNP rs2242367 was associated with more rapid progression to mortality in PD (HR=1.13[95% CI 1.04 to 1.21], p=0.002) (Supplementary Table 5).

Power calculations
The power to detect a signal in a survival GWAS depends on a number of factors, including effect size, allele frequency of the effect allele, and the proportion of individuals meeting the outcome of interest.Using the 'survSNP' package 13 , we estimate that this study had 92% power to detect a significant effect (p<5x10 -8 ) for our top APOE SNP rs429358 in the mortality GWAS, given an allele frequency of 16%, Hazard Ratio of 1.34, event/death rate of 32.1% and median time to death of 10.6 years.Supplementary Figure 7 shows how power changes with different event rates and allele frequencies.Clearly power for progression studies will increase with longer follow-up as more individuals meet the outcomes.

DISCUSSION
We have conducted a large meta-analysis GWAS of progression to clinical milestones in PD.We have identified the APOE, TBXAS1, SYT10, and ADORA2A loci as relevant to survival, motor, and cognitive progression in PD.

APOE:
The APOE SNP rs429358 was associated with both mortality and cognitive impairment.APOE is the strongest genetic risk factor for AD, 10,14 and is also associated with cardiovascular disease (including coronary heart disease / coronary artery disease), and cholesterol levels. 15 PD, APOE has been associated with age at onset, 16 cognition and dementia, and potentially motor progression, 17 but not PD risk. 1,18In the GWAS of PD age at onset, the effect of APOE was shown to be similar between age at onset in cases and age of entry of controls. 16This suggests that the effect of APOE on PD age at onset is more generally related to ageing, and not specific to PD age at onset.Indeed, GWASs of longevity and survival in the general population have identified APOE as the strongest PD Progression GWAS Tan and colleagues p16/30 genetic factor, with the same ε4 (rs429358) allele associated with increased mortality 19 and found less frequently in long-living individuals. 20However, our analysis of competing cause of death showed that the APOE SNP significantly increased PDrelated deaths and not non-PD deaths.This suggests that although APOE influences survival in general, it may also have a specific effect in PD.Another possibility is that APOE increases risk of frailty, both physical and cognitive, leading to death but this is attributed to PD as it would not be possible to disentangle these two influences on mortality.
We also analysed AD GRSs excluding APOE.This showed that non-APOE AD genetic risk influences mortality but not cognitive impairment in PD, providing some support for a specific non-AD pathology related role of ApoE in PD cognitive impairment.There is some preclinical support for this in that transgenic mice carrying the ApoE e4 allele have increased alpha-synuclein pathology both in a double transgenic (ApoEe4 x SNCA A53T) and a striatal preformed fibril injection model. 21 TBXAS1: TBXAS1 encodes Thromboxane A Synthase 1, which catalyses the conversion of prostaglandin H2 to thromboxane A2, which acts as a platelet aggregator and vasoconstrictor.3][24][25] We showed that a locus in TBXAS1 was associated with all-cause mortality in PD cohorts, but our competing risk analysis in a subset of cases suggests this may be driven by non-PD deaths rather than being specific to PD progression.Thus, it is possible that the association between TBXAS1 and PD mortality may be due to increased coronary artery disease or other non-PD causes.
However, there are a few important factors to consider.Firstly, the locus we identified for mortality in PD cohorts is different from those associated with coronary artery disease, so it may be that different SNPs within this gene have different effects.
Secondly, the competing risk analysis was performed in a limited sample that had cause of death data available, and UKB and QSBB patients may not be representative of the wider PD population.There was a more rapid time to death in these cohorts compared to our other observational cohorts.This may be due to recruitment of more clinically atypical PD cases in the QSBB, 26,27 and late 'detection' of PD in the UKB through hospital records (patients may experience onset of PD symptoms several years beforehand but are only identified with PD in the UKB when they present in hospital).Thus further analyses of cause of death in larger and more representative cohorts are needed.Thirdly, it is also possible that patients with PD are more susceptible to adverse effects of a SNP than healthy controls, so PD increases vulnerability to the SNP effect on coronary artery disease but the cause of death is not able to be classified as 'PD-p17/30 related'.This has also been seen in previous non-genetic studies of PD mortality. 28inally, it is also important to consider that PD individuals are less likely to be smokers and hence any effect of the SNP on mortality through coronary artery disease may be less pronounced than in smokers.Thus we need to further clarify the mechanism through which TBXAS1 variants influence mortality in PD patients, and whether this is specific to PD or more general.
Our search of the most recent longevity GWASs 29 suggests that variation at TBXAS1 does not influence mortality and longevity in the general population (Supplementary Materials).

SYT10:
We also found evidence that a locus near SYT10 (synaptotagmin 10) in Chromosome 12 was associated with mortality.This protein is a calcium sensor involved in regulation of calcium-dependent exocytosis, specifically related to peptidergic vesicles. 302][33] Our competing risk analysis suggested that SYT10 largely influenced non-PD deaths, and appeared to be nominally protective against PD deaths.However, this locus does not have a significant effect on longevity in the general population, although there was nominal association in the most recent longevity GWAS, 29 (beta=-0.008,p=0.003) (Supplementary Materials).This suggests that the effect of the SNP is not specific to PD mortality, however our competing risk analysis is underpowered as discussed above.Further cause of death analysis in both PD patients and healthy controls are needed to clarify the effect of this locus.

GGT5 / UPB1 / ADORA2A:
A locus near GGT5 (gamma-glutamyltransferase 5) was associated with progression to H&Y stage 3 or greater.The gene encodes a protein that cleaves the gamma-glutamyl peptide bond in the process of metabolising gammaglutamyl compounds such as antioxidants, and inflammatory molecules. 34The top SNP rs112809886 is an eQTL for ADORA2A in the brain and tibial nerve, and this variant increases the expression of ADORA2A in human cerebellum.ADORA2A encodes the adenosine A2A receptor, which is highly expressed in the basal ganglia in enkephalin expressing GABA-ergic striatal-pallidal neurons, so there is preliminary evidence that increased adenosine neurotransmission may increase motor progression in PD.The A2A receptor antagonist istradefylline has been licensed as a treatment of PD in Japan since 2013 and in the US since 2019, and has been shown to decrease daily OFF time. 35,36This is a promising locus but further work is required to test for replication, as this SNP showed some heterogeneity of effects across cohorts.

PD risk variants and candidate variants:
3][4] We showed that one variant, rs35749011, linked to GBA p.E326K (also known as p.E365K) was associated with mortality and cognitive impairment.Interestingly this variant does not cause Gaucher disease or have a major effect on glucosylceramide levels suggesting a dissociation between glucosylceramide and the role of GBA in PD progression.This is consistent with the recently reported trial data reporting a lack of effect of the glucosylceramide inhibitor venglustat in modifying PD progression.
We were not able to replicate findings for other candidate variants nominated from previous PD progression GWASs.We examined results for rs382940 in SLC44A1 for progression to H&Y stage 3 or greater, 2 however this was not associated with progression in any of our results.
We also did not replicate findings for variants in RIMS2, WWOX, and TMEM108 which have been reported for PD dementia. 4The p-values for these variants were all > 0.3 in our GWASs (Supplementary Table 4).This may be due to different criteria for defining the cognition outcomes, as the Liu study used a more stringent MDS criteria to define PD dementia whereas we have used more liberal criteria thus capturing cognitive impairment less severe than dementia.Further work is needed to replicate nominated variants both from the current study and previous studies.We did not find evidence to support APOE ε2 and MAPT H1 haplotype as factors for mortality or cognitive progression.We found some evidence suggesting the PSP mortality SNP, rs2242367, 37 was also associated with more rapid mortality in PD.This SNP was shown to increase expression of LRRK2, although there was no colocalisation with LRRK2. 37This finding could indicate that there is some 'contamination' of PSP cases in our PD cohorts, as PSP can be frequently misdiagnosed as PD and we did not have pathological diagnosis data on the majority of cases.
Limitations: This study has some limitations.This study is one of the largest GWASs of PD progression and the first large-scale GWAS of PD mortality.However larger sample sizes and longer follow-up are needed to detect variants with smaller effects or lower allele frequencies.Secondly, more data is needed on postmortem pathological diagnosis to improve power for cause of death analyses.Another limitation is the heterogeneity between cohorts and PD case selection.Our cohorts tend to be recruited from specialist clinics and groups of patients, and this may lead to a tendency to recruit more atypical patients, or rapidly progressing patients.More population-based studies are needed to improve generalisability of these results.Several of our cohorts are also non-incident, with a delay between symptom onset and study entry, and this mean that we are not able to capture the most rapidly progressing patients.

PD Progression GWAS
Tan and colleagues p19/30 In addition, we nominated genes from the top SNPs based on physical proximity and eQTL databases, however additional fine-mapping and annotation is needed to prioritise causal variants and genes for each locus.Finally, the interpretation of GWAS for neurological disease remains limited by the resolution of the effects of genomic variants on gene expression from bulk RNA sequencing studies.Rapidly increasing sample sizes, and the development of single cell resources will enable a more direct interpretation of the relationship between genomic variants and disease biology.

Conclusions:
We conducted three large-scale GWASs of PD progression, including the first GWAS of mortality in PD.We identified three genome-wide significant signals, including TBXAS1.We also showed that the genetic factors influencing progression in PD are largely different to those influencing PD risk, emphasising the need for further studies of progression.This work will help us to better understand the biology of PD progression and develop new disease-modifying treatments.

GWAS
Cognitive impairment was defined as a Montreal Cognitive Assessment (MoCA) score of ≤ 21, a Mini Mental State Examination (MMSE) score of ≤

Figure 2 .
Figure 2. Competing risk cumulative incidence function plots for the top 4 mortality SNPs.
5,744 patients were included in the meta-analysis of mortality.Of these, 1,846 (32.1%) individuals had died with a median time to death of 10.6 years from PD onset.7,696,389 SNPs were present in at least 1,000 individuals and 7,313,918 SNPs passed meta-analysis filtering for heterogeneity and MAF variability.The genomic inflation value of the meta-analysis after filtering was 1.04.
Figure 1.GWAS meta-analysis of mortality.(A) The Manhattan plot showing two GWAS significant loci after meta-analysis.The blue dashed line indicates the threshold for genome-wide significance, p=5x10 -8 .SNPs highlighted in red have p-value<5x10 -9 .

Table 1 .
Trondheim Parkinson's Disease Study (Trondheim) data was provided on request (https://doi.org/10.14802/jmd.21029).The Oslo Parkinson's Disease data was provided on request (Lasse Pihlstrom/ Mathias Toft, lasse.pihlstrom@medisin.uio.no).Cohort demographics.Means (SD) are shown unless otherwise indicated.Data shown are only in individuals who had both clinical and genetic data, after quality control filters have been applied within each cohort.The number of individuals in each cohort with complete data for each outcome of interest (and covariates) may be less than the total number.Follow-up time is calculated as the time from study entry to last visit, death, or last known status date (censoring), whichever is the latest.For QSBB and Trondheim, we did not have data on age at study entry so follow-up is calculated as the time from onset to death.

Table 2 .
Top SNPs from meta-analysis of progression to mortality.

Table 5 .
Top SNPs from meta-analysis of progression to cognitive impairment.