You are here

Article Text

Article menu
Movement disorders
Original research
Genetics of validated Parkinson’s disease subtypes in the Oxford Discovery and Tracking Parkinson’s cohorts
  1. Michael Lawton1,
  2. Manuela MX Tan2,3,
  3. Yoav Ben-Shlomo1,
  4. Fahd Baig4,5,
  5. Thomas Barber4,6,
  6. Johannes C Klein4,6,
  7. Samuel G Evetts4,6,
  8. Stephanie Millin6,7,
  9. Naveed Malek8,
  10. Katherine Grosset9,
  11. Roger A Barker10,
  12. Nigel Williams11,
  13. David J Burn12,
  14. Thomas Foltynie2,
  15. Huw R Morris2,3,
  16. Nicholas Wood2,
  17. Donald G Grosset9,
  18. Michele Tao-Ming Hu4,6
  1. 1 Population Health Sciences, University of Bristol Medical School, Bristol, UK
  2. 2 Department of Clinical and Movement Neurosciences, Queen Square Institute of Neurology, University College London, London, UK
  3. 3 UCL Movement Disorders Centre, University College London, London, UK
  4. 4 Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, UK
  5. 5 Molecular and Clinical Sciences Institute, St. George’s University of London, London, UK
  6. 6 Oxford Parkinson’s Disease Centre, University of Oxford, Oxford, UK
  7. 7 Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, UK
  8. 8 Department of Neurology, Queen's Hospital, Romford, Essex, UK
  9. 9 Department of Neurology, Institute of Neurological Sciences, Queen Elizabeth University Hospital and University of Glasgow, Glasgow, UK
  10. 10 Cambridge Centre for Brain Repair, University of Cambridge, Cambridge, UK
  11. 11 Psychological Medicine and Clinical Neurosciences, Cardiff University, Cardiff, UK
  12. 12 Faculty of Medical Sciences, Newcastle University, Newcastle, UK
  1. Correspondence to Michael Lawton, Population Health Sciences, University of Bristol Medical School, Bristol, Bristol, UK; Michael.Lawton{at}bristol.ac.uk

Abstract

Objectives To explore the genetics of four Parkinson’s disease (PD) subtypes that have been previously described in two large cohorts of patients with recently diagnosed PD. These subtypes came from a data-driven cluster analysis of phenotypic variables.

Methods We looked at the frequency of genetic mutations in glucocerebrosidase (GBA) and leucine-rich repeat kinase 2 against our subtypes. Then we calculated Genetic Risk Scores (GRS) for PD, multiple system atrophy, progressive supranuclear palsy, Lewy body dementia, and Alzheimer’s disease. These GRSs were regressed against the probability of belonging to a subtype in the two independent cohorts and we calculated q-values as an adjustment for multiple testing across four subtypes. We also carried out a Genome-Wide Association Study (GWAS) of belonging to a subtype.

Results A severe disease subtype had the highest rates of patients carrying GBA mutations while the mild disease subtype had the lowest rates (p=0.009). Using the GRS, we found a severe disease subtype had a reduced genetic risk of PD (p=0.004 and q=0.015). In our GWAS no individual variants met genome wide significance (<5×10e-8) although four variants require further follow-up, meeting a threshold of <1×10e-6.

Conclusions We have found that four previously defined PD subtypes have different genetic determinants which will help to inform future studies looking at underlying disease mechanisms and pathogenesis in these different subtypes of disease.

  • PARKINSON'S DISEASE
  • GENETICS

Data availability statement

Data are available on reasonable request. Data from the Oxford Discovery cohort is available on request from https://www.dpag.ox.ac.uk/opdc/research/external-collaborations. Data from the Tracking Parkinsons cohort is available on request from https://www.trackingparkinsons.org.uk/about-1/data/.

https://creativecommons.org/licenses/by/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/.

https://doi.org/10.1136/jnnp-2021-327376

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    WHAT IS ALREADY KNOWN ON THIS TOPIC

    • Data-driven approaches have been used to generate Parkinson’s disease subtypes in many studies but little is known about the genetics of these subtypes.

    WHAT THIS STUDY ADDS

    • We found in previously developed Parkinson’s subtypes that a severe disease subtype had the highest rates of glucocerebrosidase mutation carriers and the lowest genetic risk within a Parkinson’s Genetic Risk Score.

    HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE AND/OR POLICY

    • These results provide some biological validity to our data-driven subtyping approach and will assist in future studies looking at underlying disease mechanisms and pathogenesis.

    Introduction

    Parkinson’s disease (PD) is a common and progressive neurodegenerative disorder encompassing a wide range of motor and non-motor features. There is considerable heterogeneity within these features in terms of presentation and progression which has led many to believe there are different clinically relevant subtypes of the disease. Data-driven approaches have been applied to many PD cohorts to try and delineate these subtypes, the first was in 19991 and three systematic reviews have since been published.2–4 Other hypothesis driven approaches have also been studied in PD,2 3 the most commonly studied is the tremor-dominant (TD) versus postural instability gait difficulty (PIGD) motor subtype5 6 and another of interest is splitting into young-onset versus late-onset PD.7 8

    We previously derived Parkinson’s clinical subtypes in over 2500 early patients with PD recruited from two large cohorts: Oxford Discovery and Tracking Parkinson’s.9 10 These subtypes were derived from the baseline motor and non-motor features using a data-driven approach, which were associated with subsequent motor progression and the medication response. We have recently shown, within the Oxford Discovery cohort, that one of our subtypes had a distinctive biomarker profile with reduced apolipoprotein A1 and increased C reactive protein levels, lending biological validity to our approach.11

    Considering differences in genetics might help determine any difference in the aetiology of the subtypes while also providing a biological confirmation of data-driven clustering approaches. Here, we report on the genetics of our validated PD subtypes using data from the Oxford Discovery and Tracking Parkinson’s cohorts combined. To calculate the genetic risk of PD and related conditions including the atypical parkinsonian disorders and Alzheimer’s disease (AD), we identified a Genome Wide Association Study (GWAS) of disease status (an analysis of case/control status) for each of the diseases. We then looked at whether the genetic risk of PD and related disorders was associated with belonging to a particular disease subtype. We also considered two of the most important mutations in PD, glucocerebrosidase (GBA) and leucine-rich repeat kinase 2 (LRRK2), against our subtypes. Finally, we carry out a GWAS study to see whether any individual genetic variants are associated with belonging to a subtype.

    A recent GWAS study has been published based on the TD and PIGD motor subtypes which found multiple PD risk alleles that might influence the motor subtype.12 We have recently published a GWAS study using data from the Oxford Discovery, Tracking Parkinson’s and PPMI cohorts to look at motor and cognitive progression which found that APOE ε4 influences progressive cognitive impairment.13 This study differs to our previous one as its focus is on data-driven PD subtypes.

    Methods

    Cohorts

    We used data from two large prospective early PD cohorts. The Tracking Parkinson’s cohort includes UK-wide centres, recruited between February 2012 and May 2014. Full details of this cohort along with inclusion/exclusion criteria have been published previously.14 The Oxford Discovery cohort includes patients from 11 hospitals in the Thames Valley region recruited between September 2010 and January 2016.15 In both cohorts, patients were recruited within 3.5 years of diagnosis, and both studies were funded by Parkinson’s UK. Both studies had ethical approval and were undertaken with the understanding and written consent of each subject. Patients are followed up every 18 months collecting a wide range of data in motor, non-motor and cognitive domains. For brevity, we will refer to the Tracking Parkinson’s cohort as Tracking.

    Patient evaluation

    Our data-derived PD subtypes were determined using variables from motor, non-motor and cognitive domains at baseline. Our clustering approach used a factor analysis followed by a k-means cluster analysis where we considered two to five clusters. Individuals were excluded from the cluster analysis if they had been rediagnosed with another condition during follow-up or if they had been given a probability of a diagnosis of PD of <90% at the latest visit as rated by a research neurologist or movement disorder specialist. This was an attempt to exclude those incorrectly diagnosed with PD.

    Our first paper on this subject was based on only the Oxford Discovery cohort (with 769 patients) and we found five clusters gave us the optimal solution.9 In our second paper we used two cohorts where the Tracking cohort (n=1601) was chosen to be the development cohort (as it was larger) and the Oxford Discovery cohort (n=944) was the validation cohort.10 Here, we identified that four clusters were the optimal solution. Comparing the actual and predicted clusters (from a discriminant analysis model fitted to the Tracking clusters) in Oxford Discovery gave us a kappa statistic of 0.58 indicating moderate agreement, providing evidence our cluster approach was moderately stable across the two cohorts. These four clusters (derived using only baseline data) were shown to be associated with different subsequent motor progression over an average of 3 years follow-up and also with medication response using a levodopa challenge. We also found differences in age, gender, Hoehn and Yahr stage as well as TD/PIGD rates between the clusters which were all factors not included in the cluster analysis. The identified clusters were named (1) fast motor progression with symmetrical motor disease, poor olfaction, cognition and postural hypotension; (2) mild motor and non-motor disease with intermediate motor progression; (3) severe motor disease, poor psychological well-being and poor sleep with an intermediate motor progression and (4) slow motor progression with TD, unilateral disease. When we talk about mild/severe disease we are classifying the cross-sectional associations of data at baseline while fast/slow refers to progression rates after baseline so fast/severe and slow/mild can be thought of as different clusters. In this paper, we describe the genetics of the four subtypes (also referred to as clusters since they were developed using cluster analysis) from our development/validation paper.10 Within the Oxford Discovery cohort we report on the predicted clusters since any future research on individuals outside of these cohorts would rely on predictions.

    Genotyping

    In the Tracking Parkinson’s cohort, individuals were genotyped using the Illumina HumanCore Exome array with custom content.14 Within the Oxford Discovery cohort individuals were genotyped on either the Illumina HumanCore Exome-12 V.1.116 or the Illumina InfiniumCore Exome-24 V.1.117 singl-nucleotide polymorphisms (SNP) arrays. The quality control and imputation of this data has been previously described13 and is also described in the online supplemental file 1.

    Supplemental material

    In a principal components (PCs) analysis, 20 genetic PCs were generated from a linkage-pruned SNP set (removing SNPs with an r2 >0.02 in a 1000 kb sliding window shifting 10 SNPs at a time). If an individual was >6 SDs from the mean of one of the first 5 PCs or a clear outlier in a scatter plot they were excluded and then the PCs recalculated and repeated until there were no outliers. The first five PCs were then retained to be included as covariates within the GWAS.

    Our main focus was to look at genetic risk of PD but we also wanted to explore whether they might be shared genetic pathways between other neurodegenerative disorders (progressive supranuclear palsy (PSP), multiple system atrophy (MSA), Lewy body dementia (LBD) and AD) and each subtype while also exploring the potential for selection bias where atypical parkinsonian disorders might have been incorrectly diagnosed as PD. To calculate the genetic risk of each condition we identified an external GWAS of disease status (an analysis of case/control status) applied to separate PD, MSA, PSP, LBD and AD cohorts.18–22 Overlap in genetic pathways and risk has been described previously for LBD, Parkinsons and Alzheimer’s.21 The PD GWAS19 reports that applying a Genetic Risk Score (GRS) using the genome-wide significant hits explained a minimum of 16% of the genetic liability and led to an AUC of 0.651.

    GBA mutations were split into those that are recognised as causing Gaucher’s disease (GD) (the most common being L444P and N370S) and those that are not (E326K and T369M) as previously reported from Tracking.23 For LRRK2, we identified carriers of the G2019S and R1441C mutations, as reported previously from Tracking.24 In Oxford Discovery, carriers of L444P and R1441C mutations were identified by PCR as previously reported25 and the other mutations were identified from the Neurochip26 which is a custom-designed array for the investigation of genetic variation in neurodegenerative diseases and can detect rare variants within the LRRK2 and GBA genes. The Neurochip data underwent similar Quality control to the array data described above and is also described in the online supplemental file 1. In Oxford Discovery, we have carried out Sanger sequencing to confirm the N370S and E326K mutation carriers. All those who underwent Sanger sequencing had the mutation confirmed, however, two of the N370S carriers have not yet had Sanger sequencing. The numbers with other monogenetic forms of PD such as PRKN, SNCA and PINK1 were too small to draw any conclusions, see discussion.

    Statistical analysis

    We tabulated the clusters against LRRK2 and GBA status using a Fisher’s exact test (since the frequencies are very small in some cells due to the rarity of these mutations) to determine the strength of any association.

    We calculated the probability of belonging to a cluster from the discriminant analysis model from our validated subtypes paper.10 This probability was converted to log odds to give a more suitable continuous score for linear regression (unbounded range and symmetrical).

    In an attempt to assess the potential for selection bias we compared age (t-test), gender and cluster assignment (χ2 test) for those who did and did not have genetic data from the SNP arrays after quality control.

    We calculated GRS for PD, PSP, MSA, LBD and AD by multiplying the genome wide significant SNPs (p<5×10e-8) by their beta coefficients taken from each external GWAS and then standardising the score. This GRS can be interpreted as an estimate of the contribution of genetics to developing one of these diseases.27 Since the MSA GWAS did not find any genome wide significant SNPs we used those reported at a threshold of <1×10e-6 to calculate the GRS20 and in Alzheimer’s we used two variants that were from previously reported genome-wide significant loci but did not reach significance in the current GWAS.22 The number of SNPs from each GRS are reported in online supplemental table 1, which are the number of SNPs reaching the thresholds specified above in each external GWAS that were also available in our genetic data. Then we used linear regression with log odds of belonging to a cluster as the outcome and each GRS as the exposure. This was carried out separately within each cohort and then the results were combined using a fixed effects meta-analysis. We used a false discovery rate method,28 often called the Benjamini-Hochberg method, to control for multiple comparisons across the four subtypes.11 Using this method in our GRS analyses we have derived q-values. If our significance threshold was 0.05 we would hope to find q values <0.05. These q-values do not have a simple probabilistic interpretation, it is only important whether they reach the chosen threshold. The authors are aware of problems using corrections to p values29 and focusing on statistical significance at an arbitrary 0.05 threshold.30 31 We have tried to not use language like significant and non-significant, instead p values should be viewed by the reader as a continuum where smaller p values represent greater evidence against the null hypothesis and confidence intervals should be examined for the strength of any association. In the results, we have pointed out the direction of some associations and using the derived p values and q-values the reader can decide for themselves the strength of evidence against the null hypothesis. We hope this approach will promote modern thinking that arbitrary p value thresholds are unhelpful.

    We carried out a GWAS with linear regression using the logs odds of belonging to a cluster as the outcome. The first five genetic PCs were used as covariates for each regression. Only SNPs with a minor allele frequency (MAF) >0.05 were included. The data were combined using a fixed effects meta-analysis. We also computed the expected power for our sample size32 for a range of beta and MAFs. The number of SNPs within the GWAS are reported in the online supplemental file 1.

    Palindromic SNPs (where the alleles are nucleotides that pair to each other making it difficult to determine the direction of effect) that had an MAF >0.45 were excluded when calculating the GRS and also from the GWAS.

    Results

    Demographics and potential for selection bias

    After all the quality control procedures, we had genetic data on 1467 derived from 1601 (91.6%) individuals from the original Tracking cluster analysis. Average age (67.2 vs 68.0 with p=0.31) and gender rates (34.2% vs 34.3% female with p=0.97) were similar in those with and without genetic data (respectively). Looking within clusters rates of those included varied from 96.8% (cluster 4) to 88.6% (cluster 1) with a p=0.001. For those with genetic data there were 437, 423, 304 and 303 individuals in clusters 1–4, respectively.

    In the Oxford Discovery cohort, we had genetic data on 807 individuals, out of 944 (85.5%) individuals from the cluster analysis. Within Oxford Discovery average age (67.4 vs 66.1 with p=0.15) and gender rates (34.3% vs 41.6% female with p=0.099) were similar in those with and without genetic data (respectively). Looking within clusters rates of those included varied from 87.5% (cluster 4) to 83.0% (cluster 3) with a p =value 0.53. For those with genetic data there were 261, 145, 185 and 216 individuals in clusters 1–4, respectively.

    Mutation carriers

    Table 1 shows the associations between LRRK2 and GBA mutation carriers against the clusters in both cohorts. In the Tracking cohort the third cluster (severe motor disease and poor psychological well-being) had the largest proportion of LRRK2 carriers (1.9%), however, this is not replicated in Oxford Discovery where the third cluster has no carriers. The combined cohort p value of LRRK2 vs the clusters was p=0.35.

    View this table:
    Table 1

    Data-derived clusters compared with LRRK2 and GBA mutation status

    Within the Tracking cohort the third disease cluster (severe motor disease and poor psychological well-being) had the greatest proportion of GBA carriers (12.9% across both carrier groups) and the second disease cluster (mild motor and non-motor disease) had the lowest proportion of GBA carriers (6.3%). This trend was also seen in Oxford Discovery cohort (11.3% in cluster 3 vs 6.6% in cluster 2). In the combined cohorts a p value for a difference in GBA carrier rates across the clusters was p=0.036, and when combining the two GBA carrier groups the p value was smaller at p=0.009.

    Genetic risk of diseases

    Genetic PD risk (see figure 1) is positively associated with belonging to clusters 2 (mild motor and non-motor disease) (pooled p=0.044 and q=0.059) and 4 (slow motor progression), (pooled p=0.021 and q=0.043), while it is negatively associated with belonging to cluster 3 (severe motor disease and poor psychological well-being) (p=0.004 and q=0.015). For the pooled associations a one SD change in the PD GRS was associated with a 0.2 (95% CI 0.00 to 0.39) increase in the log odds of belonging to cluster 2; 0.2 (95% CI 0.03 to 0.37) increase for cluster 4 and a 0.3 (95% CI 0.10 to 0.51) decrease for cluster 3. We also explored a sensitivity analysis where we adjusted for the GBA mutation carrier groups and found very similar results (see online supplemental figure 1).

    Figure 1
    Figure 1

    Genetic risk of Parkinson’s disease (PD) versus likelihood of belonging to a cluster.

    We can see in figure 2 that within the Oxford Discovery cohort genetic PSP risk is negatively associated with cluster 2 (mild motor and non-motor disease) (p=0.006 and q=0.024) and positively associated with cluster 3 (severe motor disease and poor psychological well-being) (p=0.014 and q=0.027). However within the Tracking cohort the association between PSP with cluster 2 (mild motor and non-motor disease) is much smaller (-0.04 vs −0.42) and for cluster 3 (severe motor disease) it is within the opposite direction (-0.12 vs 0.41). When compared with the Oxford Discovery cohort the pooled p values and q values are much larger for both cluster 2 (p=0.046 and q=0.18) and cluster 3 (p=0.38 and q=0.70). Also within figure 2, we can see that genetic MSA risk is negatively associated with belonging to cluster 4 (slow motor progression) (pooled p=0.020 and q=0.079) where a 1 SD change in the GRS was associated with a 0.20 (95% CI 0.03 to 0.37) decrease in the log odds of belonging to cluster 4.

    Figure 2
    Figure 2

    Genetic risk of atypical Parkinson’s: progressive supranuclear palsy (PSP) and multiple system atrophy (MSA).

    In figure 3, we can see that the associations of the clusters with genetic risk of LBD and AD look very similar (especially for clusters 1, 2 and 4). Cluster 2 (mild motor and non-motor disease) is inversely associated with both LBD and AD in Tracking but not within Oxford Discovery. Within the AD GWAS the APOE genetic variant has an effect size much higher than all the others (OR of 3.32 compared with an average of 1.27 when the direction of effect is coded as positive) so we also explored what would happen when that variant is removed (see online supplemental figure 2). When removing this variant cluster 1 (fast motor progression) is positively associated with AD (pooled p=0.063 and q=0.25) where a one SD change in the GRS was associated with a 0.13 (−0.01 to 0.28) increase in the log odds of belonging to cluster 1.

    Figure 3
    Figure 3

    Genetic risk of dementia: Alzheimer’s disease and Lewy body dementia (LBD). PD, Parkinson’s disease.

    Genome Wide Association Study

    There was little evidence of population stratification since within the four GWAS analyses from Tracking, the genomic inflation factor lambda varied from 1.001 to 1.008, while within Oxford Discovery they were all 1.0.

    We highlight the power we have to detect a genome wide significant variant given our sample size in online supplemental table 2. Generally our power is small to detect rare variants with high effect sizes or common variants with small effect sizes. Since we found no genome wide significant variants in table 2 we highlight (non-independent) variants that reached a threshold of <1×10e-6, similar to the MSA GWAS study.20 At this threshold we identified 3 SNPs that were associated with cluster 1. The QQ-plot for this cluster (see online supplemental figure 3) shows a hump at the upper end which implies an excess of genetic variants associated with phenotypic cluster 1 at lower p value levels (0.0001–0.000001). We had one SNP at the reduced threshold for cluster 3 and none for clusters 2 and 4. None of the other QQ-plots (see online supplemental figures 4–6) show evidence of there being an excess of variants associated with any phenotypic cluster. The cohort specific results from table 2 can be found in online supplemental table 3). In the online supplemental file 1 the biological relevance of the identified SNPs are reported along with some network analyses (none of which met a threshold of Bonferroni adjusted-value of 0.05 shown in online supplemental figures 7–10).

    View this table:
    Table 2

    SNPs meeting a threshold of 1×10e-6 from the genome wide association study meta-analysis for each data-driven cluster

    Discussion

    The associations between GBA and the phenotypic clusters, with a severe disease cluster having the greatest proportion of carriers and a mild disease cluster having the smallest proportion, are what would be expected given the observational evidence that GBA mutations are associated with higher Hoehn and Yahr stage and worse cognition.33–36 GD-causing and GBA risk variants such as E365K (E326K) have also been associated with more rapid motor and cognitive impairment in PD in other studies.37 This has been hypothesised to relate to lysosomal dysfunction and the more rapid accumulation of pathogenic alpha-synuclein species in patients with carrying GBA variants.38 However, there are also reports that GBA mutations are associated with earlier disease onset while cluster 3 has the most GBA mutations and a higher than average age at diagnosis and cluster 2 has the least GBA mutations and the lowest average age at diagnosis.10 This highlights that there is still heterogeneity of disease onset within the clusters and that GBA mutation carriers are only a small proportion (~12%) of even the cluster with the highest carrier rate. We hypothesise that other similar genetic variants are associated with the severe disease cluster that may relate to impaired proteostasis and/or lysosomal dysfunction.

    There is also heterogeneity of clinical phenotype within LRRK2 carriers which would make it difficult to correlate them with clusters. One study showed that mutations of the LRRK2 gene are associated with less cognitive impairment compared with iPD39 while others have failed to confirm this.40 41 A study of LRRK2 found a slower decline in UPDRS scores42 and another found no discernible effect on rate of motor disease progression.43

    There are several possible explanations for the negative association between genetic risk of PD and the third, severe disease cluster. The first is that the individuals in this cluster have a more environmental and less genetically driven disease aetiology. The second is that this cluster is enriched with non-PD cases although the MSA and PSP genetic risk pooled associations do not support this, and it would also require that the PD GWAS studies had no enrichment of other similar conditions. The third is one of selection bias, in that these severe disease cases are less likely to participate in the PD cohorts that supply cases to the PD GWAS study we used, as compared with Oxford Discovery and Tracking cohorts which offered local clinical review for the majority of research participants. This PD GWAS study used data from 17 different datasets.19 Note that the GRS came from the imputed genetic data which excludes rarer genetic variants such as those within the GBA gene. The severe disease cluster has low genetic risk of PD looking at common variants yet the rare GBA variants have the highest frequency within this cluster.

    We have data on other monogenetic forms of Parkinson’s (SNCA, PRKN and PINK1) and have published this data from the Tracking cohort.24 However, the numbers are too small to draw any conclusions against our clusters. Only one individual from the Tracking clusters have a biallelic PINK1 mutation, none had a biallelic PRKN mutation and only one a SNCA mutation. In the Oxford Discovery cohort, we have data within these genes from the Neurochip but again the numbers are too small to draw any conclusions, no one from the cluster analysis had a SNCA or a biallelic PRKN mutation and only one individual had a biallelic PINK1 mutation.

    The negative association between genetic risk of PSP and cluster 2 and the positive association with cluster 3 in the Oxford Discovery cohort is what we would expect to see if there was enrichment of PSP cases. That is, PSP cases are more likely to belong to a severe motor disease cluster than a mild motor and non-motor disease cluster. However, this is not backed up by the associations within the larger Tracking cohort. This could represent a chance finding in Oxford Discovery. Alternatively, it could reflect the procedure we used to exclude patients from the analysis, that is dropping those with probability of diagnosis of PD of <90% at the latest clinic visit. In Tracking 367/1975 (18.6%) were dropped, while in Oxford Discovery only 76/1022 (7.4%) were dropped using this criterion (see online supplemental figure 1 in the original paper10). Since a greater proportion were dropped in Tracking it is more likely that we have excluded PSP cases from this cohort. The reported PD disease probability would, in all likelihood, be reduced if the clinician documented features consistent with atypical parkinsonism during the clinical review, including the presence of symmetrical motor disease, early onset falls, suboptimal levodopa response, a supranuclear gaze palsy or early autonomic failure.

    In previous research, we found cluster 3 was associated with a higher proinflammatory baseline profile (raised CRP, reduced apolipoprotein A1). This is interesting, as it suggests that in PD subtype 3—who have greater rates of cognitive dysfunction—early immune modulation might improve clinical outcomes, for example, by reducing future dementia risk if commenced early enough in the disease process. The lower overall genetic risk of PD and a higher pro-inflammatory profile in this cluster, are consistent with a hypothesis that the aetiology of this cluster is more driven by environmental rather than genetic risk factors.

    Although none of our individual variants met the GWAS p value significance threshold the ones that we highlight might be interesting for future follow-up and research. It could be that the variants, or closest genes to these variants, are a reason that a person develops a particular subtype of Parkinson’s.

    In previous research, we used multinomial logistic regression to look at how blood biomarkers are associated with an individual belonging to one of the clusters.11 For this genetic analysis, we decided to simplify the analyses by carrying out four separate analyses using the probability of belonging to the cluster as the outcome. This made the GWAS easier to run and interpret with fewer variables to estimate.

    The strengths of this study are we have used two large early in the disease course and well-phenotyped PD cohorts. Our subtypes were created using large amounts of phenotypic data incorporating 21 variables across 12 important domains and these subtypes were developed and validated in over 2500 subjects. These subtypes were shown to be associated with both motor progression and medication response in a levodopa challenge. The limitations of this study are that in terms of searching for individual genetic variants it is still too small to find any that reach genome wide significance, assuming that such variants exist. Also there is the possibility of selection bias as rates of those with genetic data varied by cluster within the Tracking cohort. The frequency of PD subtypes in our cohorts may be different to that in the general PD population if belonging to a subtype was related to agreeing to take part in our cohorts or our cohorts failed to identify specific individuals during recruitment. However, to bias our estimates of genetics versus the clusters, it would require that selection into our cohorts was also related to an individual’s genetics. Diagnosis of PD will not be perfect and some patients will turn out to have other parkinsonian disorders, although we have attempted to mitigate this by excluding individuals with a diagnostic probability of PD <90% at the latest visit.

    There are other subtypes that have been defined by a data-driven cluster analysis on motor and non-motor symptomatic data. Currently, it is difficult to determine whether the cluster definition we have used is more robust or superior to other definitions. However, in a recent systematic review our paper was rated (among 25 other data-driven studies) along with two others as having the highest methodological quality and clinical applicability.2 What sets our cluster definition apart is our use of an external validation.

    Future work is now ongoing to understand the underlying disease pathophysiology driving these different clinical clusters in early PD, and their subsequent progression. This will use a mechanistic approach comparing lysosomal, mitochondrial, inflammatory function, α-synuclein (α-syn) seeding amplification44 and α-omics profiles across the four PD clinical clusters.

    The differences in genetics between these clusters lends biological validity to our data-driven clustering approach while also providing evidence that the different subtypes can inform on underlying disease mechanisms and pathogenesis, as well as informing individual disease trajectories in PD.

    Data availability statement

    Data are available on reasonable request. Data from the Oxford Discovery cohort is available on request from https://www.dpag.ox.ac.uk/opdc/research/external-collaborations. Data from the Tracking Parkinsons cohort is available on request from https://www.trackingparkinsons.org.uk/about-1/data/.

    Ethics statements

    Patient consent for publication

    Ethics approval

    This study involves human participants and was approved by the Oxford Discovery cohort was approved by NRES Committee, South Central Oxford A Research Ethics Committee, Reference number 16/SC/0108 The Tracking Parkinsons cohort was approved by West of Scotland Research Ethics Service (WoSRES) reference 11/AL/0163. Participants gave informed consent to participate in the study before taking part.

    Acknowledgments

    The Oxford Discovery study was funded by the Monument Trust Discovery Award from Parkinson’s UK and supported by the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre based at Oxford University Hospitals NHS Trust and University of Oxford, and the NIHR Clinical Research Network: Thames Valley and South Midlands

    The Tracking Parkinson’s study was funded by Parkinson’s UK and supported by the National Institute for Health Research (NIHR) DeNDRoN network, the NIHR Newcastle Biomedical Research Unit based at Newcastle upon Tyne Hospitals NHS Foundation Trust and Newcastle University, and the NIHR funded Biomedical Research Centre in Cambridge (Grant number:146281).

    References

    1. ↵
      2. Graham JM ,
      3. Sagar HJ
      . A data-driven approach to the study of heterogeneity in idiopathic Parkinson’s disease: identification of three distinct subtypes. Mov. Disord. 1999;14:10–20.doi:10.1002/1531-8257(199901)14:1&lt;10::AID-MDS1005&gt;3.0.CO;2-4
      OpenUrlCrossRefPubMedWeb of Science
    2. ↵
      2. Mestre TA ,
      3. Fereshtehnejad S-M ,
      4. Berg D , et al
      . Parkinson’s Disease Subtypes: Critical Appraisal and Recommendations. J Parkinsons Dis 2021;11:395–404.doi:10.3233/JPD-202472
      OpenUrlCrossRef
    3. ↵
      2. Thenganatt MA ,
      3. Jankovic J
      . Parkinson disease subtypes. JAMA Neurol 2014;71:499–504.doi:10.1001/jamaneurol.2013.6233 pmid:http://www.ncbi.nlm.nih.gov/pubmed/24514863
      OpenUrlPubMed
    4. ↵
      2. van Rooden SM ,
      3. Heiser WJ ,
      4. Kok JN , et al
      . The identification of Parkinson’s disease subtypes using cluster analysis: A systematic review. Mov. Disord. 2010;25:969–78.doi:10.1002/mds.23116
      OpenUrlCrossRefPubMed
    5. ↵
      2. Jankovic J ,
      3. McDermott M ,
      4. Carter J
      . Variable expression of Parkinson’s disease: a base-line analysis of the DATATOP cohort. The Parkinson Study Group, Neurology 1990;40:1529–34.
      OpenUrlCrossRef
    6. ↵
      2. Eisinger RS ,
      3. Martinez-Ramirez D ,
      4. Ramirez-Zamora A
      . Parkinson’s disease motor subtype changes during 20 years of follow-up, Parkinsonism Relat. Disord 2020;76:104–7.
      OpenUrl
    7. ↵
      2. Niemann N ,
      3. Jankovic J
      . Juvenile parkinsonism: differential diagnosis, genetics, and treatment, parkinsonism relat. Disord 2019;67:74–89.
      OpenUrl
    8. ↵
      2. Mehanna R ,
      3. Jankovic J
      . Young-onset Parkinson’s disease: Its unique features and their impact on quality of life, Parkinsonism Relat. Disord 2019;65:39–48.
      OpenUrl
    9. ↵
      2. Lawton M ,
      3. Baig F ,
      4. Rolinski M , et al
      . Parkinson’s Disease Subtypes in the Oxford Parkinson Disease Centre (OPDC) Discovery Cohort. J Parkinsons Dis 2015;5:269–79.doi:10.3233/JPD-140523
      OpenUrlCrossRefPubMed
    10. ↵
      2. Lawton M ,
      3. Ben-Shlomo Y ,
      4. May MT , et al
      . Developing and validating Parkinson’s disease subtypes and their motor and cognitive progression. J Neurol Neurosurg Psychiatry 2018;89:1279–87.doi:10.1136/jnnp-2018-318337
      OpenUrlAbstract/FREE Full Text
    11. ↵
      2. Lawton M ,
      3. Baig F ,
      4. Toulson G , et al
      . Blood Biomarkers With Parkinson’s Disease Clusters and Prognosis: The Oxford Discovery Cohort. Mov Disord 2020;35:279–87.doi:10.1002/mds.27888
      OpenUrlCrossRef
    12. ↵
      2. Alfradique-Dunham I ,
      3. Al-Ouran R ,
      4. von Coelln R , et al
      . Genome-Wide association study meta-analysis for Parkinson disease motor subtypes. Neurol Genet 2021;7:e557.doi:10.1212/NXG.0000000000000557 pmid:http://www.ncbi.nlm.nih.gov/pubmed/33987465
      OpenUrlAbstract/FREE Full Text
    13. ↵
      2. Tan MMX ,
      3. Lawton MA ,
      4. Jabbari E , et al
      . Genome‐wide association studies of cognitive and motor progression in Parkinson’s Disease. Mov Disord 2021;36:424–33.doi:10.1002/mds.28342
      OpenUrlPubMed
    14. ↵
      2. Malek N ,
      3. Swallow DMA ,
      4. Grosset KA , et al
      . Tracking Parkinson’s: Study Design and Baseline Patient Data. J Parkinsons Dis 2015;5:947–59.doi:10.3233/JPD-150662
      OpenUrlCrossRefPubMed
    15. ↵
      2. Szewczyk-Krolikowski K ,
      3. Tomlinson P ,
      4. Nithi K
      . The influence of age and gender on motor and non-motor features of early Parkinson’s disease: Initial findings from the Oxford Parkinson Disease Center (OPDC) discovery cohort, Parkinsonism Relat. Disord 2013;20:99–105.
      OpenUrl
    16. ↵
      1. Illumina
      . HumanCoreExome 12 v1.1 support files, 2021. Available: https://emea.support.illumina.com/downloads/humancoreexome-12v1-1_product_support_files.html [Accessed 01/10/2021].
    17. ↵
      1. Illumina
      . Infinium CoreExome 24 v1.1 support files, 2021. Available: https://emea.support.illumina.com/downloads/infinium-coreexome-24-v1-1-support-files.html [Accessed 01/10/2021].
    18. ↵
      2. Chen JA ,
      3. Chen Z ,
      4. Won H , et al
      . Joint genome-wide association study of progressive supranuclear palsy identifies novel susceptibility loci and genetic correlation to neurodegenerative diseases. Mol Neurodegener 2018;13:41.doi:10.1186/s13024-018-0270-8
    19. ↵
      2. Nalls MA ,
      3. Blauwendraat C ,
      4. Vallerga CL , et al
      . Identification of novel risk loci, causal insights, and heritable risk for Parkinson's disease: a meta-analysis of genome-wide association studies. Lancet Neurol 2019;18:1091–102.doi:10.1016/S1474-4422(19)30320-5 pmid:http://www.ncbi.nlm.nih.gov/pubmed/31701892
      OpenUrlCrossRefPubMed
    20. ↵
      2. Sailer A ,
      3. Scholz SW ,
      4. Nalls MA , et al
      . A genome-wide association study in multiple system atrophy. Neurology 2016;87:1591–8.doi:10.1212/WNL.0000000000003221 pmid:http://www.ncbi.nlm.nih.gov/pubmed/27629089
      OpenUrlPubMed
    21. ↵
      2. Chia R ,
      3. Sabir MS ,
      4. Bandres-Ciga S
      . Genome sequencing analysis identifies new loci associated with Lewy body dementia and provides insights into the complex genetic architecture, (2020).
    22. ↵
      2. Kunkle BW ,
      3. Grenier-Boley B ,
      4. Sims R
      . Genetic meta-analysis of diagnosed Alzheimer’s disease identifies new risk loci and implicates Abeta, tau, immunity and lipid processing, Nat. Genet 2019;51:414–30.
      OpenUrl
    23. ↵
      2. Malek N ,
      3. Weil RS ,
      4. Bresner C , et al
      . Features of GBA -associated Parkinson’s disease at presentation in the UK Tracking Parkinson’s study. J Neurol Neurosurg Psychiatry 2018;89:702–9.doi:10.1136/jnnp-2017-317348
      OpenUrlAbstract/FREE Full Text
    24. ↵
      2. Tan MMX ,
      3. Malek N ,
      4. Lawton MA , et al
      . Genetic analysis of Mendelian mutations in a large UK population-based Parkinson's disease study. Brain 2019;142:2828–44.doi:10.1093/brain/awz191 pmid:http://www.ncbi.nlm.nih.gov/pubmed/31324919
      OpenUrlPubMed
    25. ↵
      2. Barber TR ,
      3. Lawton M ,
      4. Rolinski M , et al
      . Prodromal parkinsonism and neurodegenerative risk stratification in REM sleep behavior disorder. Sleep 2017;40.doi:10.1093/sleep/zsx071
    26. ↵
      2. Blauwendraat C ,
      3. Faghri F ,
      4. Pihlstrom L
      . NeuroChip, an updated version of the NeuroX genotyping platform to rapidly screen for variants associated with neurological diseases. Neurobiol Aging 2017;57:247.doi:10.1016/j.neurobiolaging.2017.05.009 pmid:28602509
      OpenUrlCrossRefPubMed
    27. ↵
      2. Igo RP ,
      3. Kinzy TG ,
      4. Cooke Bailey JN
      . Genetic risk scores. Curr Protoc Hum Genet 2019;104 doi:10.1002/cphg.95 pmid:http://www.ncbi.nlm.nih.gov/pubmed/31765077
    28. ↵
      2. Benjamini Y ,
      3. Hochberg Y
      . Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B 1995;57:289–300.doi:10.1111/j.2517-6161.1995.tb02031.x
      OpenUrlCrossRefWeb of Science
    29. ↵
      2. Perneger TV
      . What’s wrong with Bonferroni adjustments. BMJ 1998;316:1236–8.doi:10.1136/bmj.316.7139.1236
      OpenUrlFREE Full Text
    30. ↵
      2. Sterne JAC ,
      3. Smith GD
      . Sifting the evidence - what’s wrong with significance tests? Bmj-Brit Med J 2001;322:226-+.
      OpenUrl
    31. ↵
      2. Wasserstein RL ,
      3. Lazar NA
      . The ASA’s Statement on p-Values: Context, Process and Purpose. Am Stat 2016;70:129–31.
      OpenUrlCrossRefPubMed
    32. ↵
      2. Visscher PM ,
      3. Wray NR ,
      4. Zhang Q , et al
      . 10 years of GWAS discovery: biology, function, and translation. The American Journal of Human Genetics 2017;101:5–22.doi:10.1016/j.ajhg.2017.06.005
      OpenUrlCrossRefPubMed
    33. ↵
      2. Liu G ,
      3. Boot B ,
      4. Locascio JJ
      . Specifically neuropathic Gaucher’s mutations accelerate cognitive decline in Parkinson’s. Ann Neurol 2016;80:674–85.
      OpenUrlCrossRefPubMed
    34. ↵
      2. Winder-Rhodes SE ,
      3. Evans JR ,
      4. Ban M , et al
      . Glucocerebrosidase mutations influence the natural history of Parkinson's disease in a community-based incident cohort. Brain 2013;136:392–9.doi:10.1093/brain/aws318 pmid:http://www.ncbi.nlm.nih.gov/pubmed/23413260
      OpenUrlCrossRefPubMedWeb of Science
    35. ↵
      2. Alcalay RN ,
      3. Caccappolo E ,
      4. Mejia-Santana H , et al
      . Cognitive performance of GBA mutation carriers with early-onset PD: the CORE-PD study. Neurology 2012;78:1434–40.doi:10.1212/WNL.0b013e318253d54b pmid:http://www.ncbi.nlm.nih.gov/pubmed/22442429
      OpenUrlCrossRefPubMed
    36. ↵
      2. Setó-Salvia N ,
      3. Pagonabarraga J ,
      4. Houlden H , et al
      . Glucocerebrosidase mutations confer a greater risk of dementia during Parkinson’s disease course. Mov. Disord. 2012;27:393–9.doi:10.1002/mds.24045
      OpenUrlCrossRefPubMed
    37. ↵
      2. Davis MY ,
      3. Johnson CO ,
      4. Leverenz JB , et al
      . Association of GBA mutations and the E326K polymorphism with motor and cognitive progression in Parkinson disease. JAMA Neurol 2016;73:1217–24.doi:10.1001/jamaneurol.2016.2245 pmid:http://www.ncbi.nlm.nih.gov/pubmed/27571329
      OpenUrlPubMed
    38. ↵
      2. Menozzi E ,
      3. Schapira AHV
      . Exploring the Genotype–Phenotype correlation in GBA-Parkinson disease: clinical aspects, biomarkers, and potential modifiers. Front Neurol 2021;12:694764.doi:10.3389/fneur.2021.694764
    39. ↵
      2. Somme JH ,
      3. Molano Salazar A ,
      4. Gonzalez A
      . Cognitive and behavioral symptoms in Parkinson’s disease patients with the G2019S and R1441G mutations of the LRRK2 gene, Parkinsonism Relat. Disord 2015;21:494–9.
      OpenUrl
    40. ↵
      2. Alcalay RN ,
      3. Mirelman A ,
      4. Saunders-Pullman R , et al
      . Parkinson disease phenotype in Ashkenazi jews with and without LRRK2 G2019S mutations. Mov Disord. 2013;28:1966–71.doi:10.1002/mds.25647
      OpenUrlCrossRefPubMed
    41. ↵
      2. Marras C ,
      3. Alcalay RN ,
      4. Caspell‐Garcia C , et al
      . Motor and nonmotor heterogeneity of LRRK2 ‐related and idiopathic Parkinson’s disease. Mov Disord. 2016;31:1192–202.doi:10.1002/mds.26614
      OpenUrl
    42. ↵
      2. Saunders-Pullman R ,
      3. Mirelman A ,
      4. Alcalay RN , et al
      . Progression in the LRRK2-Asssociated Parkinson disease population. JAMA Neurol 2018;75:312–9.doi:10.1001/jamaneurol.2017.4019 pmid:http://www.ncbi.nlm.nih.gov/pubmed/29309488
      OpenUrlPubMed
    43. ↵
      2. Yahalom G ,
      3. Orlev Y ,
      4. Cohen OS , et al
      . Motor progression of Parkinson’s disease with the leucine-rich repeat kinase 2 G2019S mutation. Mov Disord. 2014;29:1057–60.doi:10.1002/mds.25931
      OpenUrlPubMed
    44. ↵
      2. Poggiolini I ,
      3. Gupta V ,
      4. Lawton M
      . Diagnostic value of cerebrospinal fluid alpha-synuclein seed quantification in synucleinopathies. Brain 2021.

    Supplementary materials

    • Supplementary Data

      This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Footnotes

    • DGG and MT-MH are joint senior authors.

    • Twitter @foltynie

    • DGG and MT-MH contributed equally.

    • Contributors ML: analysis and interpretation of the data, writing of the manuscript. ML acts as guarantor and accepts full responsibility for the finished work and/or the conduct of the study, had access to the data, and controlled the decision to publish. MT: analysis and interpretation of the data, revision of the manuscript. YB-S: study concept and design, analysis and interpretation of the data, revision of the manuscript. FB: acquisition of data, revision of the manuscript. TB: acquisition of data, revision of the manuscript. JCK: acquisition of data, revision of the manuscript. SGE: analysis and interpretation of data, revision of the manuscript. SM: analysis and interpretation of data, revision of the manuscript. NM: acquisition of data, revision of the manuscript. KG: study concept and design, acquisition of data, revision of the manuscript. RAB: study concept and design, acquisition of data, revision of the manuscript. NW: study concept and design, revision of the manuscript. DJB: study concept and design, acquisition of data, revision of the manuscript. TF: study concept and design, acquisition of data, revision of the manuscript. HRM: study concept and design, acquisition of data, revision of the manuscript. NW: study concept and design, revision of the manuscript. DGG: study concept and design, acquisition of data, revision of the manuscript. MT-MH: study concept and design, acquisition of data, revision of the manuscript.

    • Funding Both the Oxford Discovery (grant reference J-1403) and Tracking Parkinson’s cohorts (grant reference J-1101) were funded by Parkinson’s UK

    • Disclaimer The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.

    • Competing interests ML: received fees for advising on a secondary analysis of an RCT sponsored by North Bristol NHS trust. MMXT: reports no disclosures. YB-S: reports no disclosures. FB: reports no disclosures. TB: reports no disclosures. JCK: reports no disclosures. SGE: reports no disclosures. SM: reports no disclosures. NM: reports no disclosures. KG: reports no disclosures. RAB: received grants from Parkinson’s UK, NIHR, Cure Parkinson’s Trust, Evelyn Trust, Rosetrees Trust, MRC and EU along with payment for advisory board attendance from Oxford Biomedica, Aspen Neuroscience, UCB, BlueRock Therapeutics, Novo Nordisk and LCT, and honoraria from Wiley and Springer. NW: reports no disclosures. DJB: received grants from NIHR, Wellcome Trust, GlaxoSmithKline, Parkinson’s UK, and Michael J Fox Foundation. TF: grants from National Institute of Health Research, Michael J Fox Foundation, John Black Charitable Foundation, Cure Parkinson’s Trust, Innovate UK, Janet Owens Research Fellowship, Van Andel Research Institute and Defeat MSA. Advisory Boards for Peptron, Voyager Therapeutics, Handl therapeutics, Living Cell Technologies, Bial grants from Welsh Assembly Government, personal fees from Teva, personal fees from Abbvie, personal fees from Teva, personal fees from UCB, personal fees from Boehringer-Ingelheim, personal fees from GSK, non-financial support from Teva, grants from Ipsen Fund, non-financial support from Medtronic, grants from MNDA, grants from PSP Association, grants from CBD Solutions, grants from Drake Foundation, personal fees from Acorda, outside the submitted work; In addition, HRM has a patent. HRM is a coapplicant on a patent application related to C9ORF72 - Method for diagnosing a neurodegenerative disease (PCT/GB2012/052140) pending. NW: reports Funding from Aligning science against Parkinsons (ASAP). He has also received consultancy fees from GSK. DGG: received payment for advisory board attendance from Merz Pharma, Vectura plc, and consultancy fees from the GM clinic. Grant support from Parkinson’s UK, the Neurosciences Foundation, and Michael’s Movers. MT-MH: received payment for Advisory Board attendance/consultancy for Biogen, Roche, CuraSen Therapeutics, Evidera, Manus Neurodynamica. She received funding/grant support from Parkinson’s UK, Oxford NIHR BRC, University of Oxford, CPT, Lab10X, NIHR, Michael J Fox Foundation, H2020 European Union, GE Healthcare and the PSP Association. MT-MH is a coapplicant on a patent application related to smartphone predictions in Parkinson’s (PCT/GB2019/052522) pending.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.