Supplementary Materials
GliomaScan data
Cases and controls were genotyped using one of three SNP arrays. For the present study, only data generated with the Illumina 660-Quad array, containing 1,822 SNP probes across the MHC region were used (N=1,844 cases and 2,431 controls).Details regarding case ascertainment, genotyping and quality control have been previously described [1].After excluding samples with call-rates <97%, duplicate samples, related individuals (IBD>0.25), and individuals showing evidence of non-European ancestry, a total of 1,746 cases and 2,312 controls remained for analyses.
Imputation of previously reported glioma risk loci
Genome-wide SNP data were cleaned as described above and then underwent whole-genome imputation using ShapeIT v2 [2] for pre-phasing and Minimac3 for genotype imputation [3]. The cosmopolitan Haplotype Reference Consortium panel, consisting of over 64,000 haplotypes, was used as the imputation reference panel [4]. From the imputed data, we extracted genotypes for 37 SNPs previously associated with glioma risk at genome-wide statistical significance in a prior publication (based on the list appearing in Supplementary Data 1 of Melin, et al. [5]).
Statistical Analyses
We used logistic regression models to analyze the effects of HLA haplotypes on glioma risk. The baseline model assumes a purely additive contribution for a given haplotype, adjusting for the first five principal components, as shown in the following equation:
(1)
Where is the logistic regression intercept, is the additive effect of haplotype, is the dosage of haplotype in individual (0/1/2). This is repeated for each of the nine multigene haplotypes.
The non-additive model includes a dominance term for a given haplotype in addition to the baseline additive model, as shown in the following equation:
(2)
In which the dominance term, , denotes whether individual is heterozygous for haplotype (0/1), thus capturing any deviations from the additive scenario. This is repeated for each of the nine multigene haplotypes.
We tested the significance in the improvement in fit for the dominance model compared to the additive model by calculating the change in deviance (-2 x difference in log likelihood) from the additive model to the dominance model. This value follows a χ2 distribution with 1 degree of freedom. If the dominance term improves the model fit, there is evidence of a deviation from the additive scenario for a given haplotype.
The interaction model includes an additive term for each of two haplotypes being assessed, in addition to an interaction term between the two haplotypes, as shown in the following equation:
(3)
In which and denote the main additive effects of haplotypes and , respectively, and and denote the dosages of haplotypes and , respectively, in individual . denotes the interaction effect of haplotypes and , and denotes the heterozygous status for haplotypes and for individual . This analysis is repeated for HLA-DRB1*1501-DQA1*0102-DQB1*0602 with each of the other eight haplotypes.
We also tested for the general presence of non-additive effects at the HLA-DRB1-DQA1-DQB1 multigene locus by constructing a global logistic regression model including the additive and dominance terms for all 9 haplotype variants simultaneously compared to a model including only the additive terms, based on the equations for analysis of additive and dominance effects by Lenz et al. [6]. The additive model for the HLA-DRB1-DQA1-DQB1 multigene locus is as follows:
(4)
Where 8 of the 9 possible haplotype variants are included in the model, with the 9th variant (arbitrarily selected as the most common allele in the study) set as the reference allele.
The global logistic regression dominance model is as follows:
(5)
Where the dominance terms for all 9 haplotype variants are included in the model.
The change in deviance between the two models follows a χ2 distribution with 9 degrees of freedom.
References:
1. Rajaraman P, Melin BS, Wang Z, et al (2012) Genome-wide association study of glioma and meta-analysis. Hum Genet 131:1877–1888. doi: 10.1007/s00439-012-1212-0
2. Delaneau O, Zagury J-F, Marchini J (2013) Improved whole-chromosome phasing for disease and population genetic studies. Nat Methods 10:5–6. doi: 10.1038/nmeth.2307
3. Das S, Forer L, Schonherr S, et al (2016) Next-generation genotype imputation service and methods. Nat Genet 48:1284–1287. doi: 10.1038/ng.3656
4. McCarthy S, Das S, Kretzschmar W, et al (2016) A reference panel of 64,976 haplotypes for genotype imputation. Nat Genet 48:1279–1283. doi: 10.1038/ng.3643
5. Melin BS, Barnholtz-Sloan JS, Wrensch MR, et al (2017) Genome-wide association study of glioma subtypes identifies specific differences in genetic susceptibility to glioblastoma and non-glioblastoma tumors. Nat Genet 49:789–794.
6. Lenz TL, Deutsch AJ, Han B, et al (2015) Widespread non-additive and interaction effects within HLA loci modulate the risk of autoimmune diseases. Nat Genet 47:1085–1090. doi: 10.1038/ng.3379
Supplementary Figure 1: Flowchart of HLA allele and individual inclusions.
Supplementary Figure 2: Forest plots of glioma risk for a) homozygous haplotype carriers, and b) heterozygous haplotype carriers. Associations are shown as the odds ratios for glioma risk in carriers compared to non-carriers of the haplotype.
a)
b)
Supplementary Figure 3: Scatter plot of glioma riskforhomozygous haplotype carriersplotted on the x-axis, and heterozygous haplotype carriers plotted on the y-axis, for a) DRB1, b) DQA1, and c) DQB1. Associations are shown as the odds ratios for glioma risk in carriers compared to non-carriers of the haplotype. The dashed line represents a purely additive relationship, in which heterozygotes have exactly half the risk of homozygotes (on a log-odds scale).
a.
b.
Supplementary Figure 3 (continued)
c.
Supplementary Table 1: Glioma risk estimates for nominally significant (P<0.05) HLA alleles
HLA allele / HG18 position / Non-ref / Ref / Freq / INFO / ORa / SE / PHLA-DPB1*0101 / 33157346 / Present / Absent / 0.0558 / 1.0232 / 0.7756 / 0.1030 / 0.01358
HLA-DRB1*1302 / 32660042 / Present / Absent / 0.0538 / 1.0077 / 0.7777 / 0.1035 / 0.01515
HLA-B*4001 / 31431272 / Present / Absent / 0.0720 / 0.9826 / 0.8281 / 0.0925 / 0.04137
Models adjusted for top five principal components
aPer-allele odds ratio
Supplementary Table 2: Glioma risk estimates for multigene haplotype associations based on additive models.
Haplotype / Freqa / ORb / 95% LCI / 95% UCI / P-valueDRB1*0101-DQA1*0101-DQB1*0501 / 531 / 1.01 / 0.83 / 1.22 / 0.95
DRB1*0301-DQA1*0501-DQB1*0201 / 734 / 0.99 / 0.84 / 1.18 / 0.94
DRB1*0401-DQA1*0301-DQB1*0301 / 284 / 0.85 / 0.66 / 1.11 / 0.23
DRB1*0401-DQA1*0301-DQB1*0302 / 362 / 0.94 / 0.75 / 1.19 / 0.61
DRB1*0404-DQA1*0301-DQB1*0302 / 279 / 1.02 / 0.78 / 1.34 / 0.87
DRB1*0701-DQA1*0201-DQB1*0202 / 363 / 1.14 / 0.91 / 1.43 / 0.25
DRB1*1101-DQA1*0501-DQB1*0301 / 285 / 0.95 / 0.74 / 1.23 / 0.72
DRB1*1301-DQA1*0103-DQB1*0603 / 433 / 1.03 / 0.83 / 1.27 / 0.81
DRB1*1501-DQA1*0102-DQB1*0602 / 907 / 1.03 / 0.88 / 1.20 / 0.75
Models adjusted for top five principal components
aHaplotype frequency among 2098 individuals (x 2 haplotypes per individual = out of 4196 haplotypes total)
bPer-allele odds ratio
Supplementary Table 3: Additive and non-additive effect sizes of glioma risk for multigene haplotypes across the DRB1-DQA1-DQB1 loci.
Additive Model additive term / Non-additive model dominance term / Improvement in model / Stratified modelsOR / 95% CI / OR / 95% CI / P-valuea / Hom. Effect / Het. effect
OR / 95% CI / OR / 95% CI
DRB1*0101-DQA1*0101-DQB1*0501 / 1.01 / 0.83, 1.22 / 0.94 / 0.65,1.36 / 0.723 / 1.13 / 0.60, 2.12 / 0.99 / 0.79, 1.24
DRB1*0301-DQA1*0501-DQB1*0201 / 0.99 / 0.84, 1.18 / 0.88 / 0.64,1.21 / 0.432 / 1.15 / 0.67, 1.97 / 0.95 / 0.78, 1.16
DRB1*0401-DQA1*0301-DQB1*0301 / 0.85 / 0.66, 1.11 / 1.52 / 0.69,4.12 / 0.314 / 0.35 / 0.05, 1.57 / 0.89 / 0.68, 1.18
DRB1*0401-DQA1*0301-DQB1*0302 / 0.94 / 0.75, 1.19 / 0.72 / 0.36,1.39 / 0.327 / 1.57 / 0.45, 5.73 / 0.90 / 0.70, 1.15
DRB1*0404-DQA1*0301-DQB1*0302 / 1.02 / 0.78, 1.34 / 1.42 / 0.63,3.89 / 0.418 / 0.57 / 0.08, 2.63 / 1.06 / 0.80, 1.42
DRB1*0701-DQA1*0201-DQB1*0202 / 1.14 / 0.91, 1.43 / 0.93 / 0.56,1.52 / 0.767 / 1.47 / 0.61, 3.71 / 1.12 / 0.87, 1.44
DRB1*1101-DQA1*0501-DQB1*0301 / 0.95 / 0.74, 1.23 / 0.91 / 0.4,2.14 / 0.810 / 1.05 / 0.20, 5.02 / 0.95 / 0.72, 1.24
DRB1*1301-DQA1*0103-DQB1*0603 / 1.03 / 0.83, 1.27 / 1.02 / 0.65,1.64 / 0.923 / 1.02 / 0.44, 2.29 / 1.03 / 0.81, 1.31
DRB1*1501-DQA1*0102-DQB1*0602 / 1.03 / 0.88, 1.20 / 1.53 / 1.16,2.03 / 0.002 / 0.64 / 0.40, 1.01 / 1.23 / 1.01, 1.49
aP-value indicates the significance of improvement in fit by including the non-additive term compared to the additive model.
Supplementary Table 4:Additive and non-additive effect sizes of glioma risk for DRB1*1501, DQA1*0102, and DQB1*0602.
Additive Model additive term / Non-additive model dominance term / Improvement in model / Stratified modelsOR / 95% CI / OR / 95% CI / P-valuea / Hom. Effect / Het. effect
OR / 95% CI / OR / 95% CI
DRB1*1501 / 1.03 / 0.89-1.19 / 1.43 / 1.10, 1.88 / 7.7x10-3 / 0.65 / 0.41-1.01 / 1.12 / 0.97-1.30
DQA1*0102 / 0.91 / 0.82-1.02 / 1.35 / 1.11, 1.64 / 2.0x10-3 / 0.61 / 0.45-0.83 / 1.06 / 0.93-1.22
DQB1*0602 / 1.05 / 0.91-1.21 / 1.54 / 1.18, 2.02 / 1.3x10-3 / 0.62 / 0.39-0.97 / 1.15 / 0.99-1.33
aP-value indicates the significance of improvement in fit by including the non-additive term compared to the additive model.