GBA and APOE Impact Cognitive Decline in Parkinson’s Disease: A 10-Year Population-Based Study

Background: Common genetic variance in apolipoprotein E (APOE), β-glucocerebrosidase (GBA), microtubule-associated protein tau (MAPT), and α-synuclein (SNCA) has been linked to cognitive decline in Parkinson’s disease (PD), although studies have yielded mixed esults. Objectives: To evaluate the effect of genetic variants in APOE, GBA, MAPT, and SNCA on cognitive decline and risk of dementia in a pooled analysis of six longitudinal, non-selective, population-based cohorts of newly diagnosed PD patients. Methods: 1002 PD patients, followed for up to 10 years (median 7.2 years), were genotyped for at least one of APOE-ε4, GBA mutations, MAPT H1/H2, or SNCA rs356219. We evaluated the effect of genotype on the rate of cognitive decline (Mini-Mental State Examanation, MMSE) using linear mixed models and the development of dementia (diagnosed using standardized criteria) using Cox regression; multiple comparisons were accounted for using Benjamini–Hochberg corrections. Results: Carriers of APOE-ε4 (n = 281, 29.7%) and GBA mutations (n = 100, 10.3%) had faster cognitive decline and were at higher risk of progression to dementia (APOE-ε4, HR 3.57, P < 0.001; GBA mutations, HR 1.76, P = 0.001) than non-carriers. The risk of cognitive decline and dementia (HR 5.19, P < 0.001) was further increased in carriers of both risk genotypes (n = 23). No significant effects were observed for MAPT or SNCA rs356219. Conclusions: GBA and APOE genotyping could improve the prediction of cognitive decline in PD, which is important to inform the clinical trial selection and potentially to enable personalized treatment


Introduction
Patients with Parkinson's disease (PD) are more likely to experience cognitive decline than healthy older adults. 1 Problems with cognition affect patients' ability to work and function independently, placing them at higher risk of poor quality of life and nursing home placement. 2 The evolution of cognitive symptoms in PD is very heterogeneous 1 and impedes the recruitment of relevant participants into clinical trials. Hence, the means to identify patients at high risk of cognitive deficits would significantly improve the design and costs of future trials.
Genetic factors are candidate predictors of cognitive decline and dementia in PD (PDD), although heterogeneity in the design of published studies has contributed to inconsistent findings. 3 Among the strongest candidates are variants in the apolipoprotein E (APOE), β-glucocerebrosidase (GBA), microtubule-associated protein tau (MAPT), and α-synuclein (SNCA) loci. The ε4 allele of APOE (APOE-ε4) is the strongest genetic risk factor for sporadic Alzheimer's disease (AD) 4 and the top hit in genome-wide association studies for dementia with Lewy bodies (DLB), 5 and several large studies showed its impact on the progression of cognitive decline in PD. [6][7][8] GBA mutations are the commonest genetic risk factors for PD, 9 and the risk of dementia in carriers of GBA mutations is modulated by the type of mutation. 10 The H1 MAPT haplotype has been linked to tauopathies including AD and also to PD, and common variants in SNCA are established risk factors for sporadic PD. 11,12 Some studies suggest that variants in SNCA and MAPT might also affect the cognitive decline in PD, 13,14 although results are inconsistent. 7,15,16 Despite extensive literature on the impact of genetic variants in APOE, GBA, MAPT, and SNCA on cognitive progression in PD, large studies with prospective follow-up from the time of PD diagnosis are scarce and many studies track patients solely from a clinical environment. In this work we establish the significance of the APOE, GBA, MAPT, and SNCA loci on global cognitive decline and the development of dementia over the natural course of PD in the Parkinson's Incidence Cohorts Collaboration (PICC). Together the six longitudinal, population-based European cohorts form a large sample of patients with deeply characterized disease progression up to 10 years from the time of diagnosis with PD.

Subjects
We used data from PICC, a project pooling data from six PD population-based cohorts in Northern Europe, each designed to collect demographic and clinical data at the point of diagnosis and during prospective follow-up. Each cohort is summarized in Supplementary Table 1 and has been described in detail: Cambridgeshire Incidence of Parkinson's disease from General Practitioner to Neurologist (CamPaIGN), 17 Incidence of Cognitive Impairment in Cohorts with Longitudinal Evaluation-PD (ICICLE-PD), 18 New Parkinson Patient in Umeå (NYPUM), 19 ParkWest, 20 Parkinsonism: Incidence, Cognition and Non-motor heterogeneity in Cambridgeshire (PICNICS), 21 and Parkinsonism Incidence in Northeast Scotland (PINE). 22 Briefly, patients were diagnosed with idiopathic PD using UK Parkinson's Disease Society Brain Bank criteria without using family history as an exclusion criterion. Only those with a confirmed clinical diagnosis at their last clinical visit or autopsy are eligible for the PICC study, which currently has 1107 patients (1035 [93.5%] incident cases); of these, DNA or genetic data were available for 1002 (932 [93.0%] incident cases).

Demographic and Clinical Assessment
Data acquisition has been described in detail. [17][18][19][20][21][22] Age at PD diagnosis, age at symptom onset, and age at baseline were defined as the age when PD diagnosis was made, at first self-reported motor symptoms, and at inclusion in the study, respectively. Patients are reassessed at regular follow-up visits ( Supplementary Fig. 1). Home visits and/or telephone follow-up were offered to minimize attrition bias . The progression and severity of  parkinsonism were evaluated using the Hoehn and Yahr scale 23 and the Unified Parkinson's  Disease Rating Scale (UPDRS) 24 Part III (CamPaIGN, NYPUM, ParkWest, PINE) or the  Movement Disorders Society-UPDRS 25 (MDS-UPDRS) Part III (ICICLE-PD, PICNICS). UPDRS-III scores were converted into MDS-UPDRS-III as described. 26 Table 2).

Genetic Data Collection
Genetic data were available or acquired for this study for APOE, GBA, MAPT, and SNCA using a combination of whole exome sequencing, genotyping arrays, or targeted genotyping using TaqMan genotyping assays. Full details are given in Supplementary methods and the final genetic data set is summarized in Supplementary Table 3.

Statistical Analysis
For primary analysis, patients were grouped by genotype based on previous studies: APOE, carriers of ε4 allele versus non-carriers 6 ; GBA, carriers of any GBA mutation versus noncarriers 6 ; MAPT, carriers of H1/H1 versus H2 haplotype 14 ; and SNCA rs356219, carriers of GG genotype versus A-allele. 30 Primary analyses were corrected for multiple comparisons using the Benjamini-Hochberg false discovery rate (FDR) method at FDR < 0.05. For secondary analysis, carriers of APOE-ε4 were subdivided into carriers of one or two ε4 alleles, and GBA carriers were subdivided into carriers of PD risk or mild mutations, severe mutations, or variants of unknown significance. The classification of GBA mutations was based on pathogenicity in Gaucher disease (GD) 31 and PD (Supplementary Table 4).
Baseline genotype-group comparisons were performed in IBM SPSS 26.0 (Armonk, NY, USA) using t-tests, Mann-Whitney U tests, or χ 2 tests as appropriate, followed by multivariate linear regression for significant findings: age variables were compared with adjustment for study cohort and sex. Median (interquartile range [IQR]) follow-up time and cumulative proportion of dementia were estimated using the Kaplan-Meier method.
Linear mixed-effects regression models were performed in STATA 16.0 using repeated measurements of total MMSE. We performed transformation as described by Philipps et al 32 to minimize bias due to the ceiling/floor effect and curvilinearity of the raw MMSE score. Time in study, genotype, and an interaction between these were included as fixed effects. Analyses were adjusted for study, age at baseline, sex, and education as fixed effects. All models had patients' IDs as random intercepts and random slope of time. Marginal predictions of the decline in MMSE were based on fully adjusted models.
Development of PDD was evaluated using Cox regression in R 4.0.4 (package survival). The date of PDD onset was computed as the midpoint between the study visit at which dementia was diagnosed and the preceding visit, or as the midpoint between the first record of PDD diagnosis in clinical records or death certificates and the preceding study visit. Patients were censored due to death, loss to follow-up, or last recorded visit. Models were adjusted for confounders: age at baseline, sex, and education. In models including only confounders, Akaike Information Criterion was used to decide on the form of continuous confounders (original, natural logarithm [log] or square root [sqrt] transformed), and interaction with time t (t, sqrt(t), log(t), or log(t + 20)) for variables violating the proportional hazard assumption. If significant, time interaction was added also for genetic variables in the fully adjusted models. PDD-free survival was visualised in Kaplan-Meier plots (package survminer). The potential confounding effect of death was investigated using FineGray models allowing for competing risk of death before developing PDD. Data transformed by finegray function was analyzed by weighted Cox regression models adjusted for confounders and time interactions as in the Cox models. All models were stratified for study cohort.

Standard Protocol and Informed Consent
All participants signed written informed consent and regional ethical committees approved each study.

Study Population
The study population is summarized in detail in Supplementary Table 1. A total of 1002 patients were included in the pooled analysis: 139 from CamPaIGN, 146 from ICICLE-PD, 133 from NYPUM, 189 from ParkWest, 250 from PICNICS, and 145 from PINE. The proportion of males was 61.0% (n = 611) and the mean age at diagnosis was 69.1 ± 9.8 years. The median disease duration from diagnosis at baseline was 0.1 years (IQR 0.0-0.2). Patients were followed up for a maximum of 10 years, with a median of 7.2 years (IQR 6.7-10.0). During follow-up, 344 (34.3%) patients died and 177 (17.7%) patients dropped out from the study for reasons other than death. Each of the 1002 patients in the study was genotyped for at least one of APOE, GBA, MAPT, or SNCA loci, and 928 (92.6%) had genotype information for all loci (Table 1 and Supplementary Table 3).

Faster Cognitive Decline in Carriers of APOE-ε4 and GBA Mutations
The rate of annual decline in MMSE score was assessed in linear mixed effects models using data from up to 4477 visits. Both APOE-ε4 and GBA carriers had a faster rate of annual decrease in scores ( Table 2, Fig. 1A and B). For GBA carriers, this remained significant after we excluded 6 carriers with variants of unknown severity (interaction with time, β, −1.19; 95% CI, −2.05 to −0.33; P = 0.007) or 278 patients screened for selected GBA variants using TaqMan and RFLP assays (β, −1.18; 95% CI, −2.16 to −0.19; P = 0.019). The estimated drop in MMSE score over 10 years from diagnosis was from around 29 to 24 points for carriers of APOE-ε4, and from 29 to 23 points for those carrying any GBA mutation, while non-carriers of either APOE-ε4 or GBA mutations were predicted to decline only to 26 points. An association between rs356219-GG and a faster decline in MMSE score ( Table  2; Fig. 1D) marginally missed the threshold for significance after correction for multiple comparisons. No effect on global cognitive decline was shown for the MAPT haplotype (Table 2; Fig. 1C).
Based on the significant findings in the primary analysis, we explored the combined effect of harboring both an APOE-ε4 allele and GBA mutation and found that carriers of both declined faster than carriers of either APOE-ε4 or GBA, compared to non-carriers ( Finally, in secondary analysis, carriers of one APOE-ε4 allele were predicted to have a faster annual change in MMSE score, whilst progression in ε4/ε4 carriers was not significantly different compared to non-carriers, though confidence intervals were wide as only 20 participants harbored ε4/ε4 (Table 2, Fig. 1F). Further, both carriers of risk or mild and carriers of severe GBA mutations were predicted to experience faster decline in MMSE than non-carriers, although these differences were only significant for those with the risk or mild mutations ( Table 2, Fig. 1G).

APOE-ε4 and GBA Mutations Affect the Progression to PDD
At the study end, 290 of 1002 patients had developed PDD and the cumulative proportion of dementia accounting for deaths and losses to follow-up was 46.7%. APOE-ε4 and GBA had a significant impact on the rate of progression to PDD, while no effect was observed for either MAPT or SNCA rs356219 (Table 3, Fig. 2A-D). The model for APOE included a significant time-varying effect of the genetic variable (Table 3), which was also found in a competing risk model with death as a competing outcome (Supplementary Table 5) and indicated that the risk of PDD associated with the ε4 allele decreased over time. Patients carrying any GBA mutation were at 1.8 times higher risk of dementia (P = 0.001), and this remained significant after we excluded the 6 patients with GBA variants of unknown severity (HR 1.68; 95% CI, 1.19 to 2.37; P = 0.003) or the 278 patients screened for selected GBA variants using TaqMan and RFLP assays (HR 1.78; 95% CI, 1.10 to 2.86, P = 0.019). Further, carriers of both APOE-ε4 and GBA mutations were at a 5.2 times higher risk of progressing to PDD than non-carriers (Table 3, Fig. 2E). Szwedo  Secondary analysis showed a dose-dependent risk of developing PDD associated with APOE-ε4: when compared to non-carriers, carriers of one ε4 allele were at 3.1 times higher risk of progressing to dementia, while those who carried two ε4 alleles were at 6.4 times higher risk (Table 3, Fig. 2F). Among the GBA subgroups, the risk of developing PDD was 1.6 times higher in carriers of risk or mild mutations, and 2.7 times higher in carriers of severe mutations, when compared to non-carriers (Table 3, Fig. 2G).

Discussion
Using the largest pooled study of population-based cohorts to date, we have provided data on the impact of genetic variants in APOE, GBA, MAPT and SNCA on cognitive decline over the first 10 years of PD. We have shown that both APOE-ε4 and GBA mutations have an independent and additive effect on cognitive outcomes, adding to mounting evidence that these variants are key drivers of cognitive decline in PD, whilst common variants in SNCA and MAPT showed no significant impact. Understanding the role of common variants on the pattern of cognitive decline over the natural course of PD will support more accurate disease prognosis and should be considered when designing clinical trials.
Carriers of APOE-ε4 were predicted to experience faster cognitive decline, measured using both the annual change in MMSE and the time to a clinical diagnosis of dementia, reinforcing earlier evidence that APOE is a major risk factor for cognitive decline in PD [6][7][8] as well as other dementias. 4,5 Further, APOE-ε4 had a dose-dependent effect, with about a threefold increased risk of dementia per ε4 allele, comparable to results from a small cohort of neuropathologically confirmed PD cases (HR per ε4 allele 1.82, 95% CI 1.16 to 2.83). 14 Interestingly, we found that the risk of PDD associated with APOE-ε4 was not constant over the course of the disease: carrier status had a large impact on the development of dementia in the early stages of the disease, while no effect of APOE-ε4 was seen by 10 years from diagnosis. This may explain why a recent report failed to show an association of APOE-ε4 and the development of dementia in a cohort that the authors note was underrepresented for early dementia cases. 33 Similar observations have also been made in AD, showing the APOE genotype-related risk for AD decreases significantly with age, 34 and would support the importance of early initiation of neuroprotective treatment (when available) in APOE-ε4 carriers.
As previously reported, 6 also by cohorts included in this study, 35,36 GBA carriers were at twofold increased risk of progressing to dementia. This was also reflected in the faster rate of global cognitive decline in patients carrying GBA mutations. Further, we show that patients with GBA mutations enriched in neuropathic GD progress to dementia faster than patients with GBA mutations linked to non-neuropathic GD or risk of PD, compared to non-carriers. Previous studies have used different classification schemes for GBA variants, but consistently show that the risk of cognitive impairment increases with the severity of GBA mutation. 35,37,38 Contrary to our findings, several longitudinal studies showed no association with the GBA PD-risk mutations, although these were either small studies 39 or grouped E326K and T369M together with synonymous or intronic variants, 7,40 potentially diluting their effect. Our study reinforces previous findings 35,41,42 that not only severe GBA mutations, but also the PD-risk variants are important players in modifying the cognitive progression in PD and contribute to the clinical heterogeneity among GBA carriers. Given that E326K/T369M are the commonest GBA mutations, this knowledge is relevant for a large GBA-PD subpopulation. We also show that the severe (neuropathic GD) GBA mutations were associated with a greater degree of cognitive impairment at baseline, which indicates that the clinical continuum linked to GBA mutations is apparent already at diagnosis.
Lastly, we observed that the risk of cognitive impairment was further increased in individuals harboring both a GBA mutation and the APOE-ε4 allele. This small subgroup was at fivefold increased risk of progressing to PDD compared to non-carriers. A similar trend was previously observed in a study of 298 patients with PD, where 3 of 6 carriers of both APOE-ε4 and GBA severe mutations progressed to dementia (HR 2.95; 95% CI 0.80 to 10.90). 7 A faster decline in MMSE in GBA carriers with the APOE-ε4 allele was also recently shown in 100 Ashkenazi Jewish patients with DLB. 43 The increased risk of cognitive impairment observed in carriers of both APOE-ε4 and GBA mutations is possibly due to the combination of neurodegenerative mechanisms mediated by these genotypes. The compromised activity of mutated β-glucocerebrosidase facilitates the accumulation and aggregation of α-synuclein and APOE-ε4 exacerbates the brain accumulation and subsequent deposition of amyloid-β. 1 Interestingly, two recent studies indicated a novel role for APOE-ε4 in enhancing the α-synucleinopathy, and particularly the spread of Lewy body pathology, 44,45 reinforcing the importance of APOE-ε4 as a potential therapeutic target in PDD.
Besides its potential impact on cognitive decline in PD, the direct influence of APOE-ε4 on α-synuclein pathology could contribute to the earlier age of PD onset in the ε4 carriers observed in our study, and reported previously. 46,47 Both of these studies used time of self-reported onset of cardinal PD symptoms, and our study reaffirms this finding of a younger age at clinical PD diagnosis in APOE-ε4 carriers in a population-based cohort.
Previous reports on MAPT and SNCA and cognition in PD are mixed. We show a small effect of rs356219 on global cognitive decline before adjustment for multiple comparisons, replicating our previous finding using a subset of 443 patients included in this current work. 30 A larger, independent cohort will be required to validate this result, but given the small effect size, the impact of this variant is unlikely to be clinically meaningful. Few studies have reported the impact of MAPT or rs356219 on the progression to dementia. In keeping with our findings, a longitudinal study of 298 Spanish patients followed retrospectively found no association between MAPT or rs356219 and the time to dementia, 7 and a study of 514 patients showed no association between MAPT haplotype and the years from PD motor onset until PDD. 33 Contrary to these and our results, an association between MAPT H1/H1 genotype and dementia onset was found in two previous survival analyses. 14,17 Both of these were high-quality studies but were notably smaller (129 patients from the CamPaIGN population-based study, included in this work, and an independent sample of 152 neuropathologically confirmed cases). Many others have shown no association between cognitive test performance, cognitive diagnosis, or rate of cognitive decline and the H1 haplotype 15,16,[48][49][50][51] or rs356219, 15,52 and our study provides additional Szwedo et al. Page 8 evidence that the MAPT H1 haplotype and SNCA rs356219 do not play a key role in cognitive decline in PD.
Longitudinal studies represent a gold standard for tracking disease progression but are traditionally hampered by small sample size, short follow-up time, and losses to followup. Considering this, our study has notable strengths, presenting data from the largest, population-based sample of mostly incident PD patients to date, with regular follow-up over up to 10 years. Further, the cohorts had uniform design, used standardized diagnostic criteria for PD and PDD, and attrition (18%) was very low. An additional strength is the applied time-to-event analyses that were also performed allowing for competing risk of death. Potential limitations include incomplete screening for GBA mutations, although most participants were genotyped for the common variants and the number of rare mutations or complex GBA variants, such as RecNCiI, undetected is expected to be modest. Further, we chose MMSE scores to assess global cognition, and alternatives such as the Montreal Cognitive Assessment (MoCA) are more sensitive to mild and domain-specific changes 53 and may have revealed subtle effects for some loci. Lastly, it will be important to expand this work to consider promising new candidates, such as the RIMS2 locus from the first genome-wide survival analysis. 6 Evidence for the impact of common genetic variants on dementia in the general PD population is important for predicting the prognosis of newly diagnosed patients. In our study, 36% of patients were carriers of either APOE-ε4 or GBA, which places many individuals at risk of a more severe disease course, and the importance of these results is augmented by the additional effect of GBA and APOE carrier status on reducing the age at disease onset. Use in clinical practice will necessitate more precise estimates of when patients develop dementia and it will be important to establish the success of combining GBA and APOE mutations with other predictors. More immediately, the impact of genetic loci on the rate of cognitive decline can be useful in improving clinical trials of putative cognitive neuroprotective agents in PD populations. MMSE is a popular outcome measure in clinical trials due to its short application time and sensitivity to the effect of treatment, 54 although trials have largely been ineffective, in part attributed to heterogeneity in patient selection and variability in disease progression. Our models estimate that carriers of either GBA or APOE variants will decline on average 0.5 MMSE points per year, whilst carriers of both variants would average 1.1 points decline per year. Although modest, this is substantially higher than estimates of 0.1-0.2 points per year in other longitudinal studies of unselected cohorts, [55][56][57] and suggests that inclusion of genetic variants in trial inclusion criteria could improve homogeneity and trial power.
In conclusion, our findings provide evidence for the role of both GBA and APOE in the rate of cognitive decline in the general PD population. We show that both APOE-ε4 and GBA mutations are risk factors for cognitive impairment, and the effect of APOE-ε4 on PDD risk is greater in early disease, which should be considered when interpreting the current literature and designing future trials. This knowledge may further improve the accuracy of disease prognosis, especially for those with a younger age at onset who are not traditionally identified as of high risk of rapid cognitive decline.