Supplementary Document 1. Methodology
Sample Ascertainment
The study protocol was approved by Griffith University Human Research Ethics Committee. All subjects provided signed, informed consent prior to participation. Data collection procedures have been described in detail elsewhere [1]. In brief, subjects were ascertained based on permanent resident status (not selected on phenotypes of interest), to ensure sampling of individuals from the same genealogical background. In the first instance phenotypic data and biological specimens were obtained from 600 subjects (261 males, 339 females) with a mean age of 50.8 years (standard deviation of 16.4 years). Venous blood specimens were available for 600 individuals from their visit to a temporary research clinic on Norfolk Island, carried out during 2000. Blood samples were collected in EDTA tubes. DNA was isolated from a 10-20ml sample using a standard salting-out procedure [2]. DNA concentration (ng/µl) and purity (260nm:280nm) were determined spectrophometrically using the NanoDrop ND-1000 (NanoDrop Technologies, Inc.). Phenotypic data were obtained via a comprehensive medical questionnaire that included a section specific to migraine. Detailed questions regarding family history, symptoms, triggers, and medication were obtained. Migraine diagnosis was established in accordance with current IHS guidelines [3].
Genealogical Structure
Genealogical data was obtained via questionnaire, and municipal and historical records. These records indicate Pitcairn Island was settled by 9 Isle of man ‘Bounty’ mutineers, 12 Tahitian women and 6 Tahitian men in 1790 [4]. Pedigree reconstruction and validation has confirmed current descendents possess lineages to all 9 ‘Bounty’ mutineers, 6 of the Tahitian women and 2 additional Caucasian sailors who joined the small colony during the early 19th century [1, 5, 6]. A total of 377 individuals were determined to have familial links to these 17 founders and were integrated into heritability analyses. The size and complexity of the genealogical structure (N=6,537) and large volume of missing data prohibited direct use in variance component linkage analysis [7]. To facilitate analysis, the pedigree was split (N=1,078) using a peeling algorithm in the pedigree database management system PEDSYS [8]. This 1,078 member pedigree has been previously employed in genome-wide screens of cardiovascular risk traits [7].
SNP genotyping
DNA samples were genotyped according to the manufacturer’s instructions on Illumina Infinium High Density (HD) Human610-Quad DNA analysis BeadChip version 1. A total of 620,901 genome wide markers were genotyped in a sub-sample of 285 related individuals (135 males; 150 females). Of these related individuals include 76 migraine cases (22 males; 54 females). Markers had a median spacing of 2.7kb (mean = 4.7kb) throughout the genome. Each Human610-Quad DNA analysis BeadChip employed a four-sample format requiring 200ng of DNA per sample. Samples were scanned on the Illumina BeadArray 500GX Reader. Raw data was obtained using Illumina BeadScan image data acquisition software (version 2.3.0.13). Preliminary analysis of raw data was undertaken in Illumina BeadStudio software (version 1.5.0.34) with the recommended parameters for the Infinium assay and using genotype cluster files provided by Illumina. Individuals with a call rate below 95% and SNPs with a call rate below 99%, deviating from Hardy-Wienberg equilibrium (pHWE<1x10-7) or with a minor allele frequency of less than 1% were excluded from analysis. Genotypic data was analysedfor discrepancies, including Mendelian inheritance violationsusing the PEDSYS program INFER [8] and Simwalk2 [9]. The Pedigree RElationship Statistical Test (PREST) was used to verify the pedigree structure and detect relationship misspecification [10]. Discrepant genotypes were blanked prior to analysis. SNPs were annotated using information available from the National Centre for Biotechnology Information (NCBI) Build 36.3.
Statistical analysis: Heritability and Pedigree-Based Association
General characteristics of the subjects in each group were assessed using SPSS version 14.0 for windows (SPSS, Chicago, IL). All statistical analyses on related individuals were conducted using variance components-based methodology implemented in the Sequential Oligonucleotide Linkage Analysis Routines (SOLAR) version 4.0.6 software package. Heritability (h2) estimates were calculated as the ratio of the trait variance that is explained by additive polygenic effects to total phenotypic variance of the trait [11]. The applied polygenic model assumes an infinite number of genetic factors, each with a small additive effect contributing to the trait variance (‘narrow sense’ heritability). Estimates were screened for the covariate effects of age, age-squared, sex and their interactions to allow for differential symptom prevalence in males and females and adjusted for the variable age of onset. Covariates with P-values less than or equal to 0.05 were retained in the final model. Dichotomous trait analysis was enabled by assuming a liability threshold model, with an underlying multivariate normal distribution [12].
Two additional covariates, of potential interest to this study, the inbreeding (F) and ancestry coefficient (Q) were also screened. The ancestry coefficient is a measure of the degree of Polynesian and Caucasian admixture in the Norfolk pedigree. A value of 0 indicates no Polynesian ancestry. A value of 1 signifies full Polynesian ancestry. A significant ancestry-specific effect would warrant further investigation by admixture mapping. In contrast, F reflects the probability that 2 alleles at a locus are identical by descent (IBD). A value of 0 indicates no inbreeding. As the coefficient approaches 1 the level of inbreeding increases.A significant covariate effect would support recessive inheritance and founder effect for migraine. Both these coefficients have been previously described in the Norfolk population [13]. Briefly, coefficients were calculated using PEDIG software assuming the complete founder pedigree that spans more than 200 years and includes the direct descendents of the population founders as well as recent married-in individuals [13, 14]. Specifically, the Meuwissen and Luo method was used to calculate F [15].The covariate effects of ancestry and inbreeding were explored by mixed (polygenic) model analysis.
Genome-wide association testing was performed using measured genotype analysis [16], embedded in a variance components-based linkage model [17]. This assumed an additive model of allelic effect, where SNP genotypes AA, AB and BB were coded as -1, 0 and 1, respectively and used as a linear predictor of phenotype [17]. A total of 544,590 SNPs across chromosomes 1 to 22 were available for analysis. The genome-wide significance threshold was estimated empirically for this population by calculating the minimum number of effective tests (Meff) using the simple M program [18]. SNP results were annotated using the Whole Genome Association Study Viewer (WAGViewer) program [19] and NCBI Build 37.1.
Replication Cohort
Important findings were further assessed in an independent case-control cohorts from the Women’s Genome Health Study (WGHS) [20, 21].Ascertainment and ethical approval of these cohorts are described in detail elsewhere [20-22]. The WGHS cohort included 23,294 female participants aged 45 and older of European ancestry for which whole blood samples and phenotypic data is available. Migraine phenotypic data is available for 23,230 of these participants. In total, genotypic information was available for 21,008 controls and 4,705 migraineurs.
A candidate gene association analysis was also undertaken as part of this study. The same statistical approach was applied for association analysis of candidate loci, however a local type I error of α = 0.05 was applied. A Bonferroni adjustment was not required to protect against type I error inflation as the application of selection criteria for candidate genes negates the global null hypothesis. This approach has been successfully implemented in a GWAS to identify novel loci influencing serum cholesterol levels [23]. This approach was implemented to identify novel gene candidates for future evaluation in the Norfolk pedigree.
Page 1 of 8
References
1.Bellis, C., et al., Phenotypical characterisation of the isolated norfolk island population focusing on epidemiological indicators of cardiovascular disease. Hum Hered, 2005. 60(4): p. 211-9.
2.Miller, S.A., D.D. Dykes, and H.F. Polesky, A simple salting out procedure for extracting DNA from human nucleated cells. Nucleic Acids Res, 1988. 16(3): p. 1215.
3.ICHD-II, International classification of headache disorders, 2nd edition. Cephalalgia, 2004. 24(suppl 1): p. 1–160.
4.Hoare, M., Norfolk Island : a revised and enlarged history 1774-1998. 5th Edition ed. 1999, Rockhampton: Central Queensland University Press. 228.
5.McEvoy, B.P., et al., European and Polynesian admixture in the Norfolk Island population. Heredity, 2009.
6.Macgregor, S., et al., Legacy of mutiny on the Bounty: founder effect and admixture on Norfolk Island. Eur J Hum Genet, 2010. 18: p. 67-72.
7.Bellis, C., et al., Linkage mapping of CVD risk traits in the isolated Norfolk Island population. Human Genetics, 2008. 124(5): p. 543-552.
8.Dyke, B., PEDSYS: A pedigree data management system, 2.0 edn. 1996: Population Genetics Laboratory, Department of Genetics, Southwest Foundation for Biomedical Research, San Antonio.
9.Sobel, E., J.C. Papp, and K. Lange, Detection and Integration of Genotyping Errors in Statistical Genetics. The American Journal of Human Genetics, 2002. 70(2): p. 496-508.
10.McPeek, M.S. and L. Sun, Statistical Tests for Detection of Misspecified Relationships by Use of Genome-Screen Data. The American Journal of Human Genetics, 2000. 66(3): p. 1076-1094.
11.Göring, H.H.H., J.D. Terwilliger, and J. Blangero, Large Upward Bias in Estimation of Locus-Specific Effects from Genomewide Scans The American Journal of Human Genetics, 2001. 69(6): p. 1357-1369.
12.Duggirala, R., et al., A variance component approach to dichotomous trait linkage analysis using a threshold model. Genetic Epidemiology, 1997. 14(6): p. 987-992.
13.Macgregor, S., et al., Legacy of mutiny on the Bounty: founder effect and admixture on Norfolk Island. Eur J Hum Genet, 2009.
14.Boichard, D., Pedig: A fortran package for pedigree analysis suited to large populations. 2002: p. 7th World Congress on Genetics Applied to Livestock Production, Montpelier, France, pp. 28:13.
15.Meuwissen, T.H.E. and Z. Luo, Computing inbreeding coefficients in large populations. Genetics Selection Evolution, 1992. 24: p. 305–313.
16.Boerwinkle, E., R. Chakraborty, and C.F. Sing, The use of measured genotype information in the analysis of quantitative phenotypes in man. Annals of Human Genetics, 1986. 50(2): p. 181-194.
17.Blangero, J., et al., Quantitative Trait Nucleotide Analysis Using Bayesian Model Selection. Human Biology, 2005. 77(5): p. 541-559.
18.Gao, X., et al., Avoiding the high Bonferroni penalty in genome-wide association studies. Genetic Epidemiology, 2010. 34(1): p. 100-105.
19.Ge, D., et al., WGAViewer: Software for genomic annotation of whole genome association studies. Genome Research, 2008. 18(4): p. 640-643.
20.Kurth, T., et al., Migraine and Risk of Cardiovascular Disease in Women. JAMA, 2006. 296(3): p. 283-291.
21.Schürks, M., J.E. Buring, and T. Kurth, Agreement of self-reported migraine with ICHD-II criteria in the Women's Health Study. Cephalalgia, 2009. 29(10): p. 1086-1090.
22.Anttila, V., et al., Genome-wide association study of migraine implicates a common susceptibility variant on 8q22.1. Nat Genet, 2010. [Epub ahead of print].
23.Igl, W., et al., Modeling of Environmental Effects in Genome-Wide Association Studies Identifies SLC2A2 and HP as Novel Loci Influencing Serum Cholesterol Levels. PLoS Genet, 2010. 6(1): p. e1000798.
Page 1 of 8