Introduction:
Common variants in the gene encoding insulin receptor substrate 1 (
IRS1
) and nearby on 2q36.3 have been associated with levels of fasting insulin (FI). We hypothesized that a greater burden of rare variants in these regions is associated with higher FI.
Methods:
CHARGE-S sequenced (average coverage >60x) the
IRS1
and 2q36.6 regions (totaling 185 kb) in 3,539 individuals on the SOLiD platform. FI information among non-diabetics was available in 3 studies: Framingham Heart Study (
N
=811), Cardiovascular Heart Study (
N
=967) and Atherosclerosis Risk in Communities Study (
N
=1761). We analyzed rare variants (MAF < 1%) using a weighted sum test, similar to Madsen-Browning (powerful to detect an association if effects of casual rare variants are in the same direction), and the SKAT test (preferred method if variant effects are in opposite directions). Meta-analyses of weighted rare variants results used the inverse-variance method while SKAT results used a similar approach. For multi-variant tests, the threshold for significance was considered to be α = 0.05. Coding annotation predictions were obtained from the dbNSFP database which includes functional predictions from SIFT, MutationTaster, Polyphen-2, Phylo-P and LRT. Non-coding annotation information (protein binding regions, transcription factor binding sites, DNase hypersensitivity sites, conservation scores) was obtained from ENCODE and ORegAnno databases. From these annotations, we grouped different types of variants together (possible loss of function; possibly regulatory) in order to determine specific variants contributing most to the effect.
Results:
Sequencing found 4,534 variants in two regions, 86.7% of which were rare and novel, not seen in 1000 genomes or dbSNP. Approximately 20% of variants had annotation information available; of these, 34 variants were possibly damaging. We found suggestive association with FI (
p
=0.03) for all rare variants in the meta-analysis of weighted-sum tests at 2q36.3 but not at
IRS1
. At
IRS1
(but not at 2q36.3), SKAT meta-analysis tests showed evidence for all rare variants associated with FI (
p
=0.03). SKAT tests restricted to
N
=365 possibly damaging variants at
IRS1
suggested an association with FI in coding (
p
=0.06) and in non-coding (
p
=0.02) variants.
Conclusion:
Large scale deep sequencing in the
IRS1
and 2q36.3 regions found very large numbers of new, rare variants. Multi-variant tests suggest that rare variation in these regions influence FI levels, with individuals with more and rarer variants having higher FI. Further investigation is warranted to address why weighted sum and SKAT tests provide different levels of evidence for association in the two regions. Also, conditional analyses will test whether new rare variants at
IRS1
or 2q36 explain observed GWAS associations.