Association of Genetic Variant at Chromosome 12q23.1 With Neuropathic Pain Susceptibility

Key Points Question Are genetic variants associated with neuropathic pain (NP) susceptibility? Findings This genetic association study included a meta-analysis of 3 genome-wide association studies, with 4512 individuals with NP and 428 489 without, all with European descent, and identified a novel genome-wide significant locus at chromosome 12q23.1 near SLC25A3 and a suggestive locus at chromosome 13q14.2 near CAB39L. These mitochondrial phosphate carriers and calcium binding genes are expressed in tissues associated with the generation of NP, including the brain and dorsal root ganglia. Meaning These findings may provide a better understanding of genetic predisposition to NP, and this may inform the development of new treatment strategies.


eMethods 2. Self-completed Questionnaire Data
DOLORisk is an international collaboration involving members of established academic institutions and companies in Europe 4 . They designed a questionnaire based on an agreed approach to NP phenotyping by International consensus 5  Living participants of both the GoDARTS(N=5,236) and the GS:SFHS(N=20,221) who had given consent were contacted by mail with a letter of invitation, a Participant Information Leaflet (PIL) and DOLORisk paper questionnaire in optical character recognition (OCR) format labelled with a unique study code, along with a pre-paid return envelope through the Health Informatics Centre (HIC) (https://www.dundee.ac.uk/hic), a research support unit of the University of Dundee. The self-completed questionnaires from the participants were collected and managed by HIC through their secure mailing system and database. Questionnaire data were scanned, processed, and linked with anonymised participant IDs by HIC services for the DOLORisk study 6 . The confidential personal data in the questionnaire were stored securely and processed and entered into the data entry system. Data handling and delivery were conducted by HIC in a secure safe-haven environment to confirm data security and protection. Data were provided in flat file format and released on secure HIC servers for research purposes. Phenotype information was extracted from the questionnaire data and linked to pre-existing genetic and demographic data. Participants with chronic pain were identified using the following questions in the DOLORisk questionnaire: 1) "Are you currently troubled by pain or discomfort, either all the time or on and off?"; 2) "Are you currently taking medications specifically to treat pain or discomfort?"; and 3) "How long have you been suffering with this pain or discomfort?".
Participants were also asked to specify characteristics of the pain that bothered them the most, using a validated screening tool, the self-complete version of DN4 questionnaire which comprises seven items: burning, painful cold, electric shocks, tingling, pins and needles, numbness and itching. A positive response ("Yes") to each item scored as 1, and negative response ("No") scored as 0. Participants gave positive answers to either the first or second question or both and who reported a pain duration at least three months and scored at least 3 out of 7 on the DN4 questionnaire were classified as possible NP cases. Participants who gave a negative response to the question about current pain at the time of completing the questionnaire, and who were currently not taking pain medications were selected as controls for a case-control GWAS on NP.

UKBB
At the time of this study, direct neuropathic pain phenotyping information is not available in the UKBB. We therefore used the self-reported medical history information records as a proxy for NP phenotype. Dispensed medications information was captured from the answers given by the participants at an assessment centre through an interview with a trained nurse. Hospital admissions data, including the diagnosis associated with the reason for any admission, were extracted by linking to the available nation-wide participants' electronic records. On the basis of NeuPSIG guidelines for NP treatment 7 , the most relevant medications to include for case identification were gabapentin, pregabalin and duloxetine to identify individuals with likely NP. Duloxetine is used to treat depression, but it is not the first choice of drug for depression disorders treatment (National Institute for Health and Care Excellence. First-choice antidepressant use in adults with depression or generalised anxiety disorder. 2013;1-4). We did not have records of other commonly used medicines, capsaicin, and lidocaine plasters, for peripheral NP in the UKBB. Individuals who had no recorded history of having been prescribed any of these drugs were selected as controls for the GWAS. Apart from these drugs, subjects with a recorded history of amitriptyline, other tricyclic anti-depressants or tramadol were excluded from controls or cases, as these drugs are used to treat nonneuropathic pain and is not specific to NP. As gabapentin and pregabalin are used for epilepsy treatment 8 , subjects were excluded from cases or controls if they had been admitted to hospital and formally diagnosed with epilepsy or if they had been recorded as receiving any of the following anti-epileptic medications: clobazam, clonazepam, eslicarbazepine, ethosuximide, lamotrigine, levetiracetam, lacosamide, perampanel, phenytoin, phenobarbital, sodium valproate, topiramate, and zonisamide. The International Classification of Diseases 10 (ICD-10) diagnosis codes, G40.0, G40.1, G40.2, G40.3, G40.4, G40.5, G40.6, G40.7, G40.8, G40.9, G41, and R56, were used to classify epilepsy and recurrent seizures in the hospital admissions records. These diagnosis codes were used to identify subjects with epilepsy in addition to their prescription history of gabapentin and pregabalin. Therefore, we have applied exclusion criteria to avoid possible misclassification bias. Moreover, cases and controls were matched for ancestry, and principal components to address any differences. eMethods 4. Genotyping, Quality Control, and Imputation Genotype data for the GoDARTS 1 , GS:SFHS 9 and UKBB 10 study populations were preexisting and linked to the phenotype data. Blood samples were collected from the GoDARTS participants and used for genotyping by either Affymetrix 6.0 or Illumina Omni Express chips or Illumina Infinimum Broad chips. Samples were excluded based on the following criteria: samples with a call rate less than 95%, the mismatch between clinical data and genotypic gender, batch effects, ancestry outliers using principal components, sample duplicates (IBD score > 0.8). The poor-quality markers were identified and excluded on the basis of monomorphism, Hardy-Weinberg Equilibrium (HWE) p-value less than 1×10 -6 and call rate less than 95%. PLINK 1.07 11 was used to perform the quality assessment for genotyping data from all platforms. Blood or saliva were collected from the GS:SFHS participants to extract DNA. The samples were genotyped on the Illumina Human Omni Express Exome-8 v1.0 Bead chip, and Illumina Omni Express Exome-8 v1.2 Bead Chip. Quality control assessment was performed for genotyping data using GenABEL 1.7-6 12 and PLINK 1.07 11 . Samples were removed if they met the following criteria: samples with a call rate less than 98%, sample duplicates, and samples with gender discrepancies between reported and genotype data. SNVs with a call rate less than 98% HWE p-value less than 1×10 -6 . and MAF < 1%. Ancestry outliers were identified by applying a six standard deviation cut-off in a principal component analysis using genotyping data from the GS:SFHS participants merged with 1,092 individuals from the 1,000 Genomes project 13 . For the UKBB cohort, blood samples were collected to extract DNA from the participants on their visit to the UKBB assessment centre. Genome-wide genotyping was performed using two similar custom-designed genotyping arrays including UK Biobank Axiom (438,427 participants) and UK BiLEVE Axiom Affymetrix array (49,950 participants) 10 . UKBB's genotyping, QC, PCA and imputation methodology are described in detail elsewhere 3 . We selected individuals of European ancestry in the UKBB based on principal component analysis (PCA) and self-reporting ancestry information.
The genotype data from all three cohorts were imputed against a haplotype reference consortium (HRC r1.1) reference panel in NCBI build 37 14 . Post-imputation QC checks were conducted in all individual studies; monomorphic markers or those with imputation quality score < 0.4 were excluded. The genomic position of the markers is based on the NCBI human genome build 37.

eMethods 5. Genome-Wide Association Analyses and Meta-analyses
We conducted genome-wide association analyses in each of the three cohorts (GoDARTS, GS:SFHS and UKBB) separately. Both genotypic and imputed markers were tested for their association with NP using a linear mixed non-infinitesimal model in BOLT-LMM software to account for relatedness and population structure 15 . This model assumes an additive genetic model that was corrected for age and gender. The beta estimates and SEs were converted and approximated to traditional odds ratios using the formula below (https://data.broadinstitute.org/alkesgroup/BOLT-LMM/).
We conducted the meta-analysis of GWAS (GoDARTS and GS:SFHS) in stage1 using a fixed effect inverse variance weighted meta-analysis in GWAMA 16 . The genomic control inflation factor lambda was 1.023. To increase study power, we combined the summary results from all three cohorts in stage2. We calculated genomic inflation factors (ʎ) in individual data sets for population stratification and applied genomic control. Prior to the meta-analysis, SNVs with low minor allele frequency (< 0.001), low imputation quality score (<0.4) and deviation from Hardy-Weinberg equilibrium (P<1×10 -6 ) were removed from the summary GWAS results. The presence of heterogeneity between these studies was examined with the I 2 statistic. Manhattan, Quantile-quantile (QQ) and forest plots were generated to visualize the GWAS results using R 3.4 and metafor R package 17 .
Regional association plots were created using LocusZoom 18 . ScatterShot is a web application which was used to generate cluster plot images for directly typed variants in the from the UKBB dataset 19 . of eQTL for brain tissues from GTEx v6 and the most significant SNV from this study using the R package "coloc" which is based on Bayesian statistical methods and generates five posterior probabilities (PP0, PP1,PP2,PP3,PP4) for each locus 29 . We report the gene with the highest probability score (PP4) of being correlated with the most significant signal.