Lack of Association Between GBA Mutations and Motor Complications in European and American Parkinson’s Disease Cohorts

Study cohorts

In the European dataset, patients with PD were included from the Norwegian ParkWest study [16], the Parkinsonism Incidence in North-East Scotland (PINE) study [17], and the Swedish New Parkinson Patient in Umeå (NYPUM) study [18]. These cohorts provide on-going follow-up of population-based incidence studies of all newly diagnosed PD identified in specific geographic regions, initiated between 2002 and 2009. Briefly, 212 patients were enrolled in the ParkWest study, 211 in the PINE study, and 182 in the NYPUM study with a diagnosis of PD guided by the UK brain bank criteria [19] though not excluding those with a family history of PD. Only those with a confirmed clinical or pathological (if postmortem examination was performed) diagnosis of PD at their latest or final clinical visit were included. Since enrollment, 71 had a diagnosis other than PD during follow-up. Further, 57 declined genotyping, 31 have no available DNA sample or DNA was not extractable, and seven did not consent to follow-up. The remaining 439 patients were eligible for this study. At the time of the study, data from clinical visits for a period of up to ten years were available.

For the American dataset, patients with PD were included from the Parkinson’s Environment Gene (PEG) Study. The PEG cohort provides on-going follow-up of new-onset (up to 5 years after diagnosis) idiopathic PD cases from three rural California counties (Kern, Tulare, Fresno) with participants enrolled in two waves (PEG1 between 2001 and 2007 and PEG2 between 2010 and 2014) [20]. PD patients were all recruited as part of the PEG case-control study through medical groups, neurologists, and public service announcements, and a Parkinson’s disease registry pilot program in these counties. All patients in PEG were seen by movement disorder specialists at least once at baseline, many on multiple occasions and during follow-up, and confirmed as having probable idiopathic PD based on UKBB guidelines except for the family history criterion. Briefly, 849 patients were enrolled in the case-control study, and 525 patients have participated in prospective follow-up visits to assess progression (mean follow-up of 4.7 years (SD = 2.8)). Of those not examined during follow-up (n = 324), 174 were deceased (54%), 42 refused or could not be re-contacted (13%), and 108 are pending examinations (33%). Included in the present study were 445 patients who were recruited within 4.0 years of PD diagnosis, assessed for the presence of motor complications, and had genetic information available. At the time of the study, data from up to three clinical assessments were available (maximum follow up 11.4 years from diagnosis).

Studies were approved by respective ethical committees: The Western Norway Regional Committee for Medical and Health Research Ethics, the Multi-Centre Research Ethics Committee for Scotland, the Regional Ethics Review Board in Umeå, and the UCLA Institutional Review Board. Written informed consent was signed by all participants.

Clinical assessments

The clinical assessments have been described in detail before [16–18, 20]. At baseline, general medical and neurological examinations and semi-structured interviews were performed for all participants to establish medical, drug, and family history (first-degree relative with PD, self-reported). Furthermore, all patients were assessed using Hoehn and Yahr staging [21]. At baseline, patients in ParkWest, PINE, NYPUM and PEG1 were assessed using the Unified Parkinson’s Disease Rating Scale (UPDRS) [22] and in PEG2 using the Movement Disorder Society (MDS)-UPDRS [23]. In PINE, ParkWest and NYPUM, patients were examined on PD medications whenever possible, whilst in PEG patients were examined while functionally off PD medications (overnight medication withdrawal) whenever possible for UPDRS part III.

For the European cohorts and PEG1 at baseline, motor complications were detected using UPDRS part IV (motor fluctuations, score≥1 on UPDRS item 36, 37, 38 or 39; and dyskinesias, score≥1 item 32, 33 or 34). For PEG2 and PEG1 follow up visits, motor complications were detected using MDS-UPDRS part IV (motor fluctuations, score≥1 on UPDRS item 4.3 and 4.4; and dyskinesias, score≥1 item 4.1 and 4.2). Home visits were offered in PEG1 to those unable or unwilling to come to the clinic to minimize attrition bias.

Antiparkinsonian treatment was prescribed and adjusted throughout the study by a study neurologist (European dataset) or treating physician (PEG) according to best clinical judgment. We calculated levodopa-equivalent doses (LED) in accordance with published recommendations [24].

Assessment of GBA status

Genomic DNA was isolated from the peripheral blood of each subject by standard methods. The presence of GBA variants in the European dataset has been described in detail [10]: 188 patients of the ParkWest cohort were characterized by whole exome sequencing and six non-synonymous variants were detected (rs76763715/N370S, rs421016/L444P and rs781152868/Y135C, rs2230288/E326K, rs75548401/T369M, and rs369068553/V460L) and confirmed by direct sequencing of fragments amplified using primers to specifically amplify the functional GBA gene and not the pseudogene (Supplementary Table 1) [10]. These variants were genotyped in all available European samples using TaqMan single nucleotide polymorphism genotyping assay (Thermo Fisher Scientific), using ParkWest samples as controls for each detected genotype. The L444P genotype was determined using restriction fragment length polymorphism (PCR-RFLP) assays and all mutations confirmed by direct sequencing of the polymerase chain reaction product (Supplementary Table 1) [25, 26]. For the PEG dataset, all patients were characterized by a TruSeq custom amplicon panel (Illumina Inc., San Diego, CA), using paired probes designed to hybridize to unique target-specific sequences, including the GBA gene, and PCR amplification to enrich the target regions. 11 non-synonymous variants were detected in the eligible participants (rs150466109/K-27R, rs146774384/R39C, rs144173415/R47Q, rs77834747/I119T, rs409652/G202R, rs78973108/R257Q, rs2230288/E326K, rs75548401/T369M, rs76763715/N370S, rs1064651/D409H, rs421016/L444P) (Supplementary Table 2). All amino acid substitutions are numbered excluding the 39-residue signal peptide.

For the purpose of this study, GBA variants were classified based on published reports of pathogenicity in Gaucher’s disease (GD) or reported associations with PD. GBA mutations were classified as “severe” (L444P, G202R, R257Q, and D409H) if linked to neuropathic type 1 or 2 Gaucher’s disease (GD) or “mild” (N370S) if associated with the non-neuropathic type 1 GD. “Risk” variants have been reported to increase risk of PD [27, 28] but were linked to GD only when occurring in conjunction with other GBA mutations [29]. The remaining variants are of unknown significance. The initial classification of mutation severity in association with GD was based on a published classification [30] supplemented with more recent evidence from The Human Gene Mutation Database [31] (http://www.hgmd.cf.ac.uk/), CliniVar (https://www.ncbi.nlm.nih.gov/clinvar/), and literature searches (summarized in Supplementary Table 3). Each variant not classified in GD was subsequently assessed for association with PD using literature identified in Pubmed.

Statistical methods

Participants were identified as GBA carriers (with one or more GBA mutations) or non-carriers (no GBA mutation). For secondary analysis, we further subdivided the GBA mutations into severe, mild, risk factor or unknown significance (Supplementary Table 2). Because of the small sample size for the GBA mild subgroup, the mild and risk factor categories were combined. Between-group differences were compared using t-tests, Mann-Whitney tests and χ²-tests as appropriate. Non-parametric maximum likelihood estimates (NPMLEs) of the survival distributions were assessed by the expectation-maximization algorithm [32]. Parametric accelerated failure time models were applied for the primary survival analysis, with allowance for interval censoring and with t = 0 at time of diagnosis. The Weibull model was deemed optimal (over other parametric models) for time to MF using both the Akaike and the Bayesian information criteria. For LIDs the Weibull model performed similarly to the Gaussian model and was chosen for consistency and interpretability. These comparisons were made for models adjusted for age and sex. Furthermore, log minus log plots were assessed for the unadjusted models and displayed reasonably straight lines. Coefficients from the Weibull model were transformed into hazard ratios (HR), which were presented with 95%confidence intervals (CI). Further adjustment for motor severity at first visit, or ethnicity (American dataset), or repetition of the models using t = 0 as time of first visit and including adjustment for disease duration at time of first visit did not affect the effect sizes (data not shown). Cox proportional hazards models assuming right-censored data were applied in secondary analysis to allow for models including time-varying covariates (PD medication) (analyzed only for European data). The time of event was set to the first visit MF or LIDs was recorded and censoring was at the last clinical assessment. There were no substantial differences between the HRs obtained using Cox proportional hazards models and those transformed from the Weibull model in unadjusted or adjusted models. Cox proportional hazards models were next applied adjusting for age and sex and the time varying medication variables total daily LED, use of dopamine agonists, or use of levodopa at each annual visit.

Data preparation, descriptive and between group comparisons were performed in SPSS. NPMLEs of survival distributions, parametric survival analysis for interval-censored data and Cox regression with time-dependent covariates were performed in R v. 4.0.2 with package survival, functions survfit, survreg and coxph. The main plots of survival curves were created with function ggsurvplot of package survminer.

Data availability

Anonymized data are available on request by any qualified investigator for purposes of replicating procedures and results.

Baseline profile of PD-GBA carriers

This study included 884 patients with PD recruited to either one of three European cohorts (ParkWest, PINE or NYPUM; n = 439) or the American PEG cohort (n = 445). The baseline characteristics of the cohorts are listed in Supplementary Table 4. The median age at diagnosis was 70.7 (14.0) years and 70.0 (13.0) years in the European cohorts and PEG cohort, respectively. The European cohorts comprised 60.8%(267) males and the PEG cohort comprised 61.3%(273) males.

In the European cohorts, GBA variants were identified in 53 (12.1%) patients [10] and in the PEG cohorts, GBA variants were identified in 36 (8.1%) patients (Table 1). The median age of diagnosis in the GBA carrier group was younger (EUR 66.7 years, USA 66.0 years) than in the non-carriers (EUR 71.1 years, USA 70.0 years) and the distributions in the two groups differed significantly in both the European (Mann–Whitney U p = 0.01) and PEG cohorts (Mann–Whitney U p = 0.03). No further differences were identified between carriers of a GBA mutation and non-carriers for demographic or clinical variables assessed at the first clinical visit (Table 1).

Effect of GBA on the development of motor complications

We first examined the development of motor complications in 439 participants of the European cohorts. Of these, 9 patients were only assessed at one clinical visit and were excluded, leaving 430 patients in the survival analysis. By 10 years of follow-up, 36 (67.9%) of the 53 carriers of any GBA mutation had developed MF compared to 222 (58.9%) of the 377 non-carriers, and 18 (34.0%) of 53 GBA mutation carriers had developed LIDs compared to 150 (39.8%) of the 377 non-carriers. Parametric accelerated failure time models were applied to assesses the impact of GBA carrier status on the time to develop motor complications. The risk of developing MF or LIDs in carriers of a GBA-mutation was not different compared to non-carriers in unadjusted analysis (MF: HR 1.22, 95%CI 0.86 to 1.74, p = 0.27; LIDs: HR 0.78, 95%CI 0.48 to 1.28, p = 0.33) or when controlling for age and sex (Table 2; Fig. 1). Repetition of the model excluding those GBA variants of unknown significance did not affect the effect sizes (data not shown). Further, analysis of the impact of either severe GBA mutations or mild and risk factor GBA mutations on the time to develop motor complications compared to non-carriers revealed no statistically significant association with either MF or LIDs (Table 2). Finally, we applied Cox proportional hazards regression models including time-varying covariates to account for the possible effects of medication regime over the course of PD. Inclusion of time varying total LED, use of levodopa or use of dopamine agonist did not alter the lack of association of GBA carrier status with the development of either MF or LIDs (data not shown).

OPEN IN VIEWER

Fig. 1

Nonparametric maximum likelihood estimates of the survival distributions for onset of motor fluctuations (MF) and levodopa-induced dyskinesias (LIDs) in the European data set (A and B) or the American dataset (C and D). Time from Parkinson’s disease diagnosis is shown in years. Subjects were grouped into carriers of a GBA variant (blue) or non-carriers (yellow).

To validate these findings, we next assessed the development of motor complications in 445 participants from the PEG cohort. 440 patients were assessed for MF and 445 patients for LIDs during the study. 10 patients reported MF and 3 LIDs at the time of diagnosis and were excluded from the survival analysis. Parametric accelerated failure time models were then applied to assess the association of GBA status with the development of motor complications from the time of PD diagnosis, and showed that in the American cohorts GBA-carriers were not at increased risk of developing MF or LIDs in comparison to the non-carriers in unadjusted analysis (MF: HR 1.18, 95%CI 0.72 to 1.96, p = 0.51; LIDs: HR 1.08, 95%CI 0.54 to 2.13, p = 0.83) or when controlling for age, sex (Table 2, Fig. 1C, D). Similarly, the effect sizes did not change when the analysis was repeated after excluding the variants of unknown significance, or when analysing the impact of either severe GBA mutations or mild and risk factor GBA mutations compared to non-carriers (data not shown).