the potential influence of SLE susceptibility variants in our tests of
association with LN. To account for potential population substructure,
principal component analyses were computed using SNPs that
passed quality control.35 All tests of association were computed and
adjusted for five principal components in the set I samples and four
principal components in the set II and set III samples. Within each
GWAS set, autosomal SNPs that passed quality control filters and
exhibited a minor allele frequency .0.01 were used to impute the
approximately 2.5 million SNPs in HapMap2 (Build 35) using IMPUTE.36
The reference set for imputation was the Utah residents with
ancestry from northern and western Europe available via the HapMap.
Within each GWAS set, logistic regression models included the abovestated
principal components and accounted for the imputation uncertainty
via the posterior genotype probabilities.36 Dominant, additive,
and recessive genetic models were computed assuming at least 10 and
20 individuals homozygous for the minor allele were observed. Because
of the modest sample size, this requirement caused many SNPs
to not have a test computed in set III. The evidence of association from
each of the three GWAS sets was weighted by sample size and combined
using the weighted inverse normal method as implemented in
METAL.37 To account for any inflation, the genomic control-adjusted
P value was reported.