Exploring the early drivers of pain in Parkinson’s disease

Abstract

Pain is a common and complex non-motor symptom in people with Parkinson’s disease (PWP). Little is known about the genetic drivers of pain in PWP, and progress in its study has been challenging. Here, we conducted two genome-wide association studies (GWAS) to identify genetic variants associated with pain experienced during the earliest stages of Parkinson’s disease. The study population consisted of 4,159 PWP of European ancestry who were mapped to five previously-described, longitudinal pain trajectories. In the first GWAS, the extreme pain trajectories (highest burden versus no significant pain over time) were compared, and in the second GWAS, a multinomial approach was undertaken. While no variant reached genome-wide significance, we identified promising associations, such as rs117108018 (OR_{GWAS−Extreme}=8.96, p_{GWAS−Extreme}=2.5 × 10^− 7), a brain/nerve eQTL for L3MBTL3 and EPB41L2, and rs61881484 (p_{GWAS−Multinomial}=2 × 10^− 7), which intersects a transcription factor peak targeting CREB1, critical in sensory neuron synaptic plasticity and neuropathic pain regulation. Gene-based tests implicated CTNNB1 (p_{GWAS−Extreme}=3.2 × 10^− 5), KLK7 (p_{GWAS−Extreme}=7 × 10^− 5), and SLITRK3 (p_{GWAS−Multinomial}=3.2 × 10^− 5), which have been associated with neurodevelopment. At the pathway-level, there was an enrichment for genes involved in neurotransmitter regulation and opioid dependence. This study implicates neuropathic pain mechanisms as prominent drivers of elevated pain in PWP, suggests potential therapeutic genetic targets for further research.

Introduction

Parkinson’s disease (PD) is the second most prevalent neurodegenerative disease, and pain is a common non-motor symptom that may affect 60–80% of people with Parkinson’s disease (PWP)^1,2,3,4. Pain has a significant impact on quality of life in PWP, and it has been associated with poorer mental health, and sleep and mood disturbances^4,5. While higher levels of pain are commonly indicated among PWP with advanced disease, PWP may also experience pain during the earliest stages of the disease^6,7. To add complexity, pain in PWP demonstrates significant heterogeneity^3,8, and may encapsulate akathisia, dystonia, musculoskeletal pain, neuropathic pain, and primary/central pain that likely vary in severity and intensity as PD progresses^2,4,9,10. The complex etiology, underdiagnosis, and unsystematic treatment of pain among PWP emphasizes the need to better understand the underlying biological mechanisms, with the goal of identifying therapeutic targets for improved pain management¹¹.

Genetic variation plays a prominent role in PD risk, alongside environmental risk factors¹². Over the last two decades, genome-wide association studies (GWAS) have boosted our understanding of the genetic basis of PD, with the identification of ~ 90 genetic risk variants¹³ and prioritization of over 200 genes¹⁴. There have been three modestly-sized genetic studies (Ns < 1,350) that explored cross-sectional definitions of pain in PWP, and collectively their initial findings suggest that genetic variation influences the manifestation of pain in PD^15,16,17. It is difficult to draw robust conclusions from these studies as two focused on candidate genes while one was a GWAS, and their study populations had different continental ancestries (Israel, Taiwan, United Kingdom). But more importantly, their operational definitions of pain were based on different cross-sectional measures. Considering the observed heterogeneity in the severity and accrual of pain in PWP, a genetic investigation of pain should ideally investigate relationships with longitudinal symptom patterns that best reflect an individual’s perception of pain over time. However, such longitudinal data are rare. Patient-reported outcome (PRO) instruments of pain, which are based on the person’s self-report of pain severity, have been shown to be reliable and highly correlated with clinical and observational assessments of pain, even among individuals with cognitive impairment^18,19,20. Generic health-related quality of life instruments, such as the short European Quality of Life (EuroQol) 5 Dimension 5 Level (EQ-5D-5 L) instrument, include a scale for experiences with pain and discomfort and have the added advantage of feasibility for large-scale epidemiologic collection of longitudinal data from PWP.

Here, we leveraged the EQ-5D-5 L pain and discomfort instrument and genome-wide data available in 4,159 PWP to identify the genetic determinants of longitudinal pain trajectories in PWP who were early in their disease ($<3$ years). These longitudinal pain trajectories were derived in our previous work²¹, where we observed five distinct subgroups of PWP that have similar longitudinal patterns (trajectories) of pain (Supplementary Fig. S1). For the current study, we leverage our prior PWP clusters and performed GWAS comparing: (1) PWP with a severe pain trajectory to those with no significant pain over time (comparison of the extreme trajectories), and (2) all four trajectories of increasing pain to those with no significant pain over time. Single variant tests of association, gene-based tests of associations, pathway enrichment analyses, and other downstream analyses identified suggestive associations as well as associations consistent with known neuropathic and pain-related mechanisms supporting a genetic basis for the heterogeneous pain trajectories observed in and reported by PWP.

Results

After extensive quality control (QC), the final study population consisted of 4,195 PWP who were within 3 years of onset and assigned to one of five clusters of self-reported pain trajectories, and for whom genetic data were available (Fig. 1). Only 2.3% of PWP were in the severe pain cluster (Class 5) and 22.5% were in the cluster reporting no pain over time (Class 1). The study population was on average 65 years old, male (54%), and were of European descent (Table 1). Increasing proportions of PWP in clusters with greater pain reported OFF episodes, being on PD medications, having arthritis and memory problems, and greater mobility impairments (Table 1).

Fig. 1

Flowchart of the overall study design.

Table 1 Baseline characteristics of study participants ($<3$years disease duration) and their assignment to subgroups of increasing pain impairment. Table 1 summarizes the baseline characteristics of the 4,159 study participants with disease duration $<3$  years. Participants were assigned to five clusters identified from the longitudinal trajectories of patient-reported pain in our prior work, with Class 1 subjects having no pain over time and class 5 subjects having severe pain over time (see supplementary Fig. S1). Abbreviations: SD: standard deviation; BMI: body mass index; IQR: interquartile range. *Patient-reported pain (pain/discomfort component of the EQ-5D-5 L) is an ordinal measure ranging from 0 to 4, indicating no pain to extreme pain. The reported values are the values at the start of the trajectory (see supplementary Fig. S1).

Extreme genome-wide association test results

In this GWAS comparing 97 PWP with a severe pain trajectory to 935 PWP with no pain, there was no evidence of genomic inflation ($\lambda=0.999$, Fig. 2a). While no variant met genome-wide significance ($p<5\times{10}^{-8}$), there were 23 promising associations with $p<5\times{10}^{-6}$ (Fig. 2a; Supplementary Table S1), and there were 7 SNPs with $p<2\times{10}^{-6}$ (Table 2; Supplementary Fig. S5). Associations persisted when adjusting for self-reported arthritis or mobility impairment (Supplementary Tables S2 and S3). The most significant variant, an intergenic variant rs117108018 on chromosome 6, had an exceptionally strong association ($OR=8.96$, $p=2.54\times{10}^{-7}$). This variant maps to a DNase peak and it is 12 kilobases (kb) upstream of Y_RNA, which encodes a small non-coding RNA molecule that is highly expressed dorsolateral prefrontal cortex and tibial nerve²². It is also an eQTL ($p<0.005$) for L3MBTL3 and EPB41L2 in the amygdala and cerebellum brain tissues, respectively (Supplementary Table S4). Suggestive associations also included MAPK8 (rs72794357, $OR=4.61$, $p=1.2\times{10}^{-6}$) and VLDLR (rs4741753, $OR=2.24$, $p=1.2\times{10}^{-6}$) variants, which appear to be eQTLs of their respective genes in brain and tibial nerve tissues. We interrogated LDlink for variants in LD with promising associations ($p<5\times{10}^{-6}$), and there was strong LD between a few of these variants and variants significantly associated with other pain-related traits: shoulder impingement syndrome (rs74821598, ${r}^{2}=0.80$), and sensory perception (rs75941298, ${r}^{2}=0.86$) (Supplementary Table S5). In addition, these 7 variants associated with severe pain in PWP are eQTLs for several genes in the brain/nerve tissues ($p<0.05$) (Supplementary Table S4).

Fig. 2

Manhattan and quantile-quantile plots for the GWAS of extreme trajectories (a) and the multinomial GWAS of all five trajectories (b) Manhattan and quantile-quantile (Q–Q) plots of the results from the genome-wide association studies (GWAS) comparing (a) the two most extreme clusters, which correspond to the least and most impaired groups of persons with Parkinson’s disease, and (b) all five clusters identified from the longitudinal trajectories of patient-reported pain. Both sets of association tests adjusted for age, sex, and the first three multidimensional scaling (MDS) components. The Manhattan plots show $-log10(p-value)$ on the y-axis and the chromosomal position of each genetic variant on the x-axis, with a dashed horizontal line indicating suggestive significance of $2\times{10}^{-6}$. In the Q–Q plots, the observed versus expected p-values from the corresponding GWAS were plotted on the $-log10$ scale. The observed genetic inflation factor was (a) ${\lambda}_{GC}=0.999$ for extreme cluster GWAS, and (b) ${\lambda}_{GC}=1.0409$ for the all cluster GWAS.

Table 2 Suggestive associations identified by comparing the class with the highest pain impairment (class 5; severe pain trajectory) to the class reporting no pain impairment over time (class 1). The results from the genome-wide association study ($p<2\times{10}^{-6}$) comparing participants assigned to the two most extreme pain trajectory classes (class 1 [reference] and class 5), adjusting for age, sex, and the first three multi-dimensional scaling (MDS) components. Abbreviations: OR: odds ratio; MAF: minor allele frequency; CI: confidence interval. *Genes within +/-25 kilobases (kb).

Multinomial genome-wide association test results

There was no evidence of genomic inflation in the multinomial GWAS for pain ($\lambda=1.04$, Fig. 2b). Similar to the extreme GWAS, we did not observe genome-wide significant associations but there were 6 promising associations with $p<2\times{10}^{-6}$ (Table 3; Supplementary Fig. S6; expansive associations [$p<0.001$] are in Supplementary Table S6). Results were unchanged when adjusting for self-reported arthritis or mobility impairment (Supplementary Tables S7 and S8). An intergenic variant (rs61881484) on chromosome 11 less than 2.5 kb upstream of LDLRAD3 had the most significant association ($p=2\times{10}^{-7}$). It is also a brain tissue eQTL of several genes ($p<0.05$; Supplementary Table S9), including: LDLRAD3, FJX1, COMMD9, TRAF6, PRR5L, and PAMR1. We also identified suggestive genic associations, such as SLCO2A1 (rs7653639, $p=1.5\times{10}^{-6}$) and RYK (rs9839609, $p<1.7\times{10}^{-6}$; rs9865808, $p=1.94\times{10}^{-6}$) (Table 3). Evidence suggests that the two RYK variants are eQTLs for this gene in brain and tibial nerve tissues (Supplementary Table S9).

Table 3 Suggestive associations identified by comparing increasing pain trajectory classes (classes 2–5; C2-5) to the no pain over time trajectory (class 1; C1). The results from the genome-wide multinomial logistic regression analysis ($p<2\times{10}^{-6}$) comparing subjects assigned to the four trajectories of increased pain over time (classes 2–5; C2-5) to subjects reporting no pain over time (class 1; C1), adjusting for age, sex, and the first three multi-dimensional scaling (MDS) components. Abbreviations: OR: odds ratio; MAF: minor allele frequency; CI: confidence interval; C1–C5: class 1 (no pain over time) to class 5 (severe pain over time). *Genes within +/-25 kilobases (kb).

Gene-bases test results

The gene-based tests involving 18,727 and 18,736 mapped genes for the extreme and multinomial GWAS, respectively, did not reveal associations at a genome-wide significance threshold (defined as p = 0.05/total number of mapped genes). The most significant findings were CTNNB1 ($p=3.2\times{10}^{-5}$) and KLK7 ($p=7\times{10}^{-5}$) for the extreme GWAS (Supplementary Table S10), and SLITRK3 ($p=3.2\times{10}^{-5}$) and LDLRAD3 ($p=1.5\times{10}^{-4}$) for the multinomial GWAS (Supplementary Table S11). Several of these most significant gene-based test candidates were also implicated in Parkinson’s disease, neurodevelopment, and migraine, such as MYLK2, which encompasses an intronic variant rs6060983 associated with PD²³, and FBN2 that underlies suggestive association for migraine susceptibility²⁴ (Supplementary Tables S10 and S11).

Pathway enrichment analysis

The pathway enrichment analysis was also applied to a total of 314 and 247 genes for the two sets of GWAS, as the union of SNP-based and gene-based annotations (Supplementary Tables S12 and S13). Several pathways relevant to the regulation of the neurotransmitter and opioid dependence were enriched in our results (Tables 4 and 5). For example, in extreme GWAS, multiple significant synapse-related pathways were implicated, including KEGG^25,26: serotonergic synapse ($p=1.17\times{10}^{-5}$), dopaminergic synapse ($p=4.54\times{10}^{-5}$), and glutamatergic synapse ($p=7.91\times{10}^{-5}$); opioid signaling, including Reactome: opioid signaling R-HSA-111,885 ($p=4.71\times{10}^{-4}$). As for the multinomial GWAS, there was enrichment of synapses regulation and opioid dependence pathways, including KEGG: morphine addiction ($p=1.70\times{10}^{-5}$) and retrograde endocannabinoid signaling ($p=5.16\times{10}^{-4}$).

Table 4 Pathway Enrichment Analysis results for the GWAS of the extreme pain classes. The pathway enrichment analysis was performed in Enrichr leveraging the union of two gene sets originating from the extreme GWAS: (1) the unique genes in which genic SNPs had an association p-value < 0.0001 (supplementary Table S12), and (2) unique genes with p-value < 0.01 in the gene-based association tests (supplementary Table S12). The table summarizes the results utilizing the KEGG database and the REACTOME database, showing the term (the enriched biological pathway), gene (list of genes overlapping with the enriched pathway), odds ratio, combined score (natural logarithm of the Fisher exact test p-value multiplied by the corresponding z-score), and the p-value. Abbreviations: OR: odds ratio.

Table 5 Pathway Enrichment Analysis results for the Multinomial GWAS of pain classes. The pathway enrichment analysis was performed in Enrichr leveraging the union of two gene sets originating from the multinomial GWAS: (1) the unique genes in which genic SNPs had an association p-value < 0.0001 (supplementary Table S13), and (2) unique genes with p-value < 0.01 in the gene-based association tests (supplementary Table S13). The table summarizes the results utilizing the KEGG database, showing the term (the enriched biological pathway), gene (list of genes overlapping with the enriched pathway), odds ratio, combined score (natural logarithm of the Fisher exact test p-value multiplied by the corresponding z-score), and the p-value. Abbreviations: OR: odds ratio.

Other noteworthy genetic associations

In addition to the single SNP and gene-based discovery efforts described above, we also sought to map previously reported associations to the results of the two GWAS performed here for pain among PWP. To do this, we queried the GWAS Catalog (as of April 2023) for SNPs associated at p-value threshold${10}^{-5}$ with (1) PD risk (526 variants), (2) neuropathic pain (30 variants), (3) chronic pain (146 variants), and (4) response to opioids (50 variants) for lookup in our results of the extreme and multinomial GWAS. In relation to PD risk, 438 of 526 variants were tested in our GWAS, and 11% had marginal associations in the extreme GWAS, including multiple variants within the alpha-synuclein encoding gene, SCNA ($p<0.05$; Supplementary Table S14). Interestingly, there was little evidence of overlap between association for 25 of 30 variants associated with neuropathic pain with findings from the currents analyses (Supplementary Table S15), but a handful of associations were evident amongst 118 of 146 variants associated with chronic pain (Supplementary Table S16) including GABRB2-rs1946247 ($p=5\times{10}^{-8}$, $\beta=0.019$)²⁷, which was marginally significant in both the extreme ($p=2.94\times{10}^{-4}$) and multinomial GWAS ($p=6.53\times{10}^{-5}$). In the final comparison, 29 of 50 variants associated with response to opioids and while only 3 variants were associated with extreme pain in PWP ($p<0.05$; Supplementary Table S17), they implicate putative therapeutic targets. For example, rs61355450C and rs11692586G was associated with higher extreme pain in PWP ($OR=2.39$ and $1.43$) and also with higher opioid analgesic requirements in the treatment of cancer pain²⁸. A variant associated with improved buprenorphine treatment response²⁹, rs7205113T, had a protective association here with extreme pain in PWP ($\text{OR}=0.52$).

Discussion

We report here a large genome-wide association of longitudinal patient-reported pain trajectories for 4,159 PWP recruited by Fox Insight³⁰. In this study, we identified several genetic variants with suggestive significance from two GWAS analyses: (1) a GWAS of the severe pain over time to no experiences of pain over time and (2) a multinomial GWAS comparing four trajectories of increasing pain to no experiences of pain over time. In brief, the results implicate multiple neuropathic pain processes, a possible relationship with dystonia, and synaptic dysfunction as prominent drivers of pain in PD.

Pain is a common non-motor symptom among PWP; however, there has only been one prior GWAS targeting pain in PD. The prior GWAS compared 898 PWP reporting high pain compared with 420 PWP reporting no/low pain who were participants in the UK Parkinson’s Pain Study^28,29. Participants were classified based on their one-point-in-time response to the McGill Pain Questionnaire and the Visual analog scale regarding pain in the past month, and the GWAS identified two strongly correlated TRPM8 variants (rs11563208 [$OR=1.8$] and rs12465950 [$OR=1.7$]; ${r}^{2}=0.85$) associated at genome-wide significance with pain in PD^30,31. However, in the present study there was no evidence for any association for either SNP in either GWAS (rs11563208: ${OR}_{\text{GWAS}-\text{Extreme}}=1.14$, ${p}_{\text{GWAS}-\text{Extreme}}=0.44$ and ${p}_{\text{GWAS}-\text{M}ultinomial}=0.52$; rs12465950: ${OR}_{\text{GWAS}-\text{Extreme}}=0.99$, ${p}_{\text{GWAS}-\text{Extreme}}=0.97$ and ${p}_{\text{GWAS}-\text{M}ultinomial}=0.83$), nor was TRPM8 associated at the gene-level (Supplementary Tables S10 and S11). While both studies examined pain in European-descent PWP, the operational definitions of the pain outcomes were substantially different and this alone may explain the differences in associations.

While neither the extreme nor multinomial pain GWAS revealed genome-wide significant associations, the genetic variants that were marginally associated with pain in PD were independent of arthritis and mobility limitations, and implicate genes that appear to be related to relevant biological mechanisms, and therefore suggest promising processes for further investigation. For example, the most significant finding from the extreme pain GWAS (rs1178108018) with a large magnitude of effect ($OR=9$) is intergenic and a brain/nerve tissue eQTL for L3MBTL3, which has been associated with multiple sclerosis risk. This variant regulates Notch, and Notch activation is essential for the development of neuropathic pain^31,32. Variant rs1178108018 is also a brain/nerve tissue eQTL for EPB41L2, and in mice loss of the encoded membrane skeletal protein results in myelin abnormalities in the peripheral nervous system³³. There were also genic MAPK8 and VLDLR variants with suggestive associations in the extreme GWAS that were brain tissue eQTLs for MAPK8 and VLDLR. Implication of MAP kinases is of interest since they may have a role in increasing pain sensitivity³⁴, and MAPK8 may be associated with opioid dependence ($p=5\times{10}^{-7}$)³⁵. VLDLR is Reelin receptor, and Reelin plays a critical role in neurodevelopment and synaptic plasticity^36,37. There were also strong associations ($OR>3$; $p=5\times{10}^{-6}$) for ANO3 variants in the extreme GWAS (Supplementary Table S1), and this gene has been suggestively associated with altered neohesperidin dihydrochalcone taste sensitivity³⁸ (taste impairment is another common non-motor symptom in PD³⁹), cannabis dependence⁴⁰, and alcohol use disorder⁴¹. ANO3 itself is intriguing as mutations within the gene cause monogenic autosomal dominant dystonia (Dystonia-24)⁴², and upwards of 30% of PWP may also have dystonia⁴³. Lastly, these ANO3 variants are brain tissue eQTLs for BDNF-AS (Supplementary Table S4), which is up-regulated in a mouse model of PD where is may promote apoptosis and autophagy⁴⁴.

The multinomial pain GWAS also highlighted several other genes that merit further investigation. For example, the most significant finding (rs61881484) intersects a transcription factor peak that targets CREB1⁴⁵, a gene critical in synaptic plasticity in sensory neurons⁴⁶ and is involved in the regulation of neuropathic pain⁴⁷. Variant rs61881484 is also a brain tissue eQTL of several genes with varied but relevant functions (e.g. FJX1 which regulates dendritic extensions⁴⁸, COMMD9 has a critical role in endosomal sorting of Notch family members⁴⁹ and Notch signaling activation is a key driver of neuropathic pain³², TRAF6 is involved in maintaining neuropathic pain⁵⁰ and chronic visceral pain⁵¹, and PAMR1 is associated with muscular dystrophy and predicted to be associated with migraine pain^52,53), further supporting its role as a likely driver of pain in PD. Other suggestive genic (RYK, HDAC7, GFRA1) associations ($p<5\times{10}^{-6}$) further implicate a prominent role for neuropathic pain processes in PD. For example RYK-rs9839609 is a brain/nerve tissue eQTL for this gene, which encodes a receptor that actively participates in Wnt signaling pathway known to be essential for neurite outgrowth along with many other neuronal development activities⁵⁴. In rats, blocking Wnt/Ryk signaling inhibits induction of neuropathic pain without affecting normal pain sensitivity⁵⁵. HDAC7 and other histone deacetylases may merit closer investigation as HDAC inhibitors were shown to attenuate hypersensitivity in a neuropathic pain rat model⁵⁶. Lastly, there is evidence that GFRA1, which encodes a GDNF receptor, may play a key role in thermal hyperalgesia in a neuropathic pain rat model^57,58.

The gene-based tests and pathway enrichment analyses further highlight likely biological processes relevant for understanding PD pain. In general, genes involved in neurodevelopment and opioid signaling were enriched in these pain GWAS. For example, CTNNB1 is among the most significant gene-based result for both GWAS (${p}_{\text{GWAS}-\text{Extreme}}=3.2\times{10}^{-5}$; ${p}_{\text{GWAS}-\text{M}ultinomial}=2.7\times{10}^{-4}$), and its key function includes promoting neurogenesis⁵⁹. CTNNB1 de novo mutations have been associated with polyneuropathy in lower limbs⁶⁰. KLK7, a gene with marginal significance in the extreme GWAS ($p=7\times{10}^{-5}$) encodes an astrocyte-derived amyloid-β degrading enzyme important for the pathogenesis of Alzheimer’s disease⁶¹. We also observed genes (e.g. SLITRK3 and LDLRAD3) in the multinomial GWAS involved in neurite outgrowth and modulating amyloid precursor protein function^62,63. Moreover, gene-based tests for the extreme GWAS prioritized genes (e.g. MYLK2 and PTK2B) that have been associated with neurodegenerative disorders in prior studies^23,64. Noteworthy, the extreme pain GWAS was enriched for pathways relevant to serotonergic, dopaminergic, and glutamatergic synapse signaling. Among the most significantly enriched pathways in the multinomial pain GWAS was morphine addiction, which is intriguing as morphine is an opioid analgesic for managing chronic pain, the long-term use of which could lead to tolerance and severe side effects, including addiction^65,66. The “retrograde endocannabinoid signaling” pathway is also of interest as it plays an essential role in modulating the function of both the excitatory and inhibitory synapses⁶⁷. When comparing the current associations with prior GWAS for general pain or PD risk, none of the genome-wide significant findings in prior GWAS²⁷ for several PD-related outcomes were amongst the most significant findings in the two GWAS reported here. Still, lookups of these prior associations suggest consistent findings with the current results, e.g., GABRB2-rs1946247 is associated with multisite chronic pain⁶⁵ and it was suggestively associated with PD pain in both GWAS conducted here. Our results also suggest that SCNA, a well-established PD risk locus, may also play a role in PD pain^13,68. To recap, these gene-based and enrichment results, and relationships with other phenotypes suggest that future research on the role of synaptic processes in PD and PD pain may lead to promising therapeutic targets for PWP.

The present study has several strengths and limitations. This study is the first GWAS of longitudinal pain trajectories in PWP, and one of the few studies of a longitudinal pain phenotype. In addition to the large sample size, we used two complimentary approaches, a multinomial GWAS and a traditional GWAS of a binary outcome, alongside a comprehensive strategy for investigating pain in PD. While this study was large, statistical power was limited (i.e. there were only 97 PWP in the severe pain trajectory) as we did not identify genome-wide significant associations. This underscores challenges in collecting sufficient data for the subjective and evolving outcome of pain in PWP. Further, the PWP involved in this study were limited to those early in their disease duration ($<3$ years), which we perceive as a key strength, but pain among PWP increases with disease duration⁶⁸ and therefore, we may have increased power by examining pain in PWP with longer disease durations (this is an active future direction). The study was also limited to PWP of European ancestry, limiting the generalizability to non-European populations. Also, our longitudinal definitions of pain were based on a readily implemented generic PRO that will allow for comparisons to non-PD population, a perceived strength, and while the findings do implicate relevant biological processes, more granular insights may be gained by using more sensitivity tools that capture various pain sensations. While the study population represents one of the largest patient-driven PD registries, a key limitation is the that it is a convenience sample of volunteers reporting a diagnosis of PD, though a diagnosis of PD was confirmed by clinicians in validation study of 165 participants⁶⁹. Overall, the framework presented in this study serves as a blueprint for forthcoming research examining the intersection of PROs and genotypic variation, providing a versatile approach with great potential to tackle a broad spectrum of neurological conditions and other complex phenotypes, and demonstrate the potential of PROs for elucidating biological insights.

To summarize, our findings also shed light on likely drivers of pain experienced by PWP, who are early in their disease, through the two GWAS, by leveraging a person-centered approach to modelling a pain PRO and coupling these outcomes with comprehensive genetic data. We identified genetic variants with suggestive significance implicated in pain sensitivity^47,55 and regulating neuropathic pain^8,69, along with multiple eQTLs associated with neuropathic pain and other pain mechanisms (i.e. synapse signaling), including a potential role for a genetic driver of dystonia. The findings, however, emphasize the significance of processes related to neuropathic pain contributing to pain severity in PWP^8,70. Previous research suggests that the pain treatment for PWP mainly comprises nonsteroidal anti-inflammatory drugs⁷⁰, but the evidence that GABRB2 may also contribute to pain in PWP suggests that possible targets for future research might include anti-convulsants (i.e. gabapentine)⁷¹. We also observed tentative relationships for opioid and endocannabinoid related pathways, which suggests other potential therapeutic targets. In general, our findings further fuel a broader discourse on the significance of pain management within the realm of PD and warrants interdisciplinary collaborations from a variety of disciplines to enable a holistic treatment paradigm that addresses both the underlying condition and the intricate pain profiles it gives rise to.

Methods

Study population

This study of de-identified data was deemed non-human subject research by the institutional review boards at Case Western Reserve University and The MetroHealth System, Cleveland, Ohio. These data were provided by participants in Fox Insight (https://foxinsight.michaeljfox.org), an ongoing, virtual, longitudinal study of people aged 18 years and older focused on empowering discovery, validation, and reproducibility in PD PRO research³⁰. We accessed these data through the Fox Insight Data Exploration Network (Fox DEN) on 10/14/2021 (for up-to-date information visit https://foxinsight-info.michaeljfox.org/insight/explore/insight.jsp). Fox Insight includes routine assessment of PRO measures, other health-related data, and environmental exposures, and participants are invited to provide biological samples to be used for generating genome-wide genetic data by 23andMe. Launched in 2015, Fox Insight has enrolled $>\,54{,}000$ participants, including ~38,000 PWP who are on average 66 years of age, of European descent (96%), and 65% male.

Fox Insight participants are invited to complete routine PD-related self-assessments via an online portal every three months. A telehealth pilot study of a subset of participants confirmed a diagnosis in 95.1% of 167 individuals⁷². As previously described, we applied latent class growth analysis⁷¹, a data-driven mixture model approach, to longitudinal data for EQ-5D-5 L pain/discomfort item (5 level Likert scale where 0 = no pain or discomfort and 4 = extreme pain or discomfort) available in 8,612 PWP early in their disease course ($<3$ years)⁷². We identified five distinct clusters of PWP will similar pain trajectories over time (Supplementary Fig. S1)⁷³. The five clusters were used in the subsequent genetic analyses as our outcomes of interest. Detailed genetic data were generated from saliva samples of PWP participating in genetic sub-study, which underwent detailed quality control assessment and genetic imputation (Supplementary Methods).

Two sets of association tests

We performed GWAS to compare the clusters identified from the longitudinal trajectories of patient-reported pain (Fig. 1): (1) a logistic GWAS that compared the two most extreme clusters (Supplementary Fig. S1), Class 1 (no pain) and Class 5 (severe pain) using PLINK1.9, which correspond to the least and most impaired groups of PWP, respectively, and aimed to identify genetic variations associated with extreme pain among PWP, and (2) a multinomial GWAS that individually compared Classes 2–5 to Class 1 using the frequentist multinomial logistic regression approach implemented in Trinculo⁷⁴. All analyses were adjusted for age, sex, the first three MDS components, and the genotyping arrays. To characterize the associations identified in our GWAS, we utilized the NHGRI-EBI GWAS Catalog⁷⁵, a comprehensive database of published GWAS findings, along with RegulomeDB⁷⁶, FIVEx⁷⁷, LDlink⁷⁸, and Bgee⁷⁹, the web-based tools to explore the functional potential, expression quantitative trait loci (eQTL), and linkage disequilibrium (LD) patterns.

We also performed several gene-related analyses for protein-coding genes using MAGMA v1.6⁸⁰ embedded in FUMA 1.5.3⁸¹. In brief, the summary statistics from each GWAS were used to perform the gene-based test, gene-set analyses, and tissue expression analysis. A Bonferroni correction was used to determine genome-wide significance for these gene-based analyses.

Enrichment analysis

For each GWAS, we annotated SNPs using SeattleSeq Annotation 138^82,83. Unique genes were selected using the p-value threshold of 10^− 4, resulting in 142 and 110 genes for the extreme and multinomial GWAS, respectively. We also generated additional gene annotation subsets based on the MAGMA gene-based tests described above. The genes were selected using a gene-based test p-value threshold of 0.01, resulting in 192 and 154 genes for the two analyses. Per GWAS, gene sets were combined to create the final gene list, resulting in 314 and 247 genes for the extreme and multinomial GWAS, respectively. We employed Enrichr^80,81, an efficient gene set enrichment analysis tool with the interactive web server, to perform the pathway enrichment analysis.