The obesity risk variants and their allele frequencies among controls for each ethnic group are described in Supplemental Table 1
. Allele frequencies in whites in the MEC were similar to those observed in populations of European descent.6, 9, 10
Among the MEC controls, we confirmed the positive associations between BMI and 13 variants in 6 loci (BDNF, FTO, KCTD15, NEGR1, NRXN3 and SH2B1) and also between BMI-adjusted waist circumference and 19 SNPs in 10 loci (the 6 loci above plus LYPLAL1, MSRA, SEC16B and TMEM18) (reported separately by the PAGE consortium).
Cases were slightly older than controls (). After controlling for age, cases were more likely than controls to have a higher BMI and personal history of diabetes, to be a former smoker, with higher mean pack-years among smokers, and to have consumed more alcohol. Cases had slightly fewer years of education and were likely to consume less multivitamins, dietary fiber and total calcium. Among women, cases were less likely to use postmenopausal hormone treatment than controls.
Characteristics* of colorectal cancer cases and controls
In the main effect analysis of the obesity risk variants on colorectal cancer, adjusted for age, sex and ethnicity, two out of the 24 SNPs showed a significant association (; results for all SNPs are shown in Supplemental Table 2
). The KCTD15
rs29941 obesity risk allele (C) was associated with a lower colorectal cancer risk, whereas the MC4R
rs17782313 obesity risk allele (C) showed a positive association with colorectal cancer. Further adjustments for BMI or for the risk factors that differed between cases and controls in did not materially change the risk estimates (i.e., difference <10%). However, neither association was statistically significant when corrected for multiple comparisons, either by FDR-adjustment (adjusted p-value
for both rs29941 and rs17782313 = 0.24), by case-control status permutation in sex/race/SNP-strata (permutation p
-value = 0.056 for rs29941, p
= 0.060 for rs17782313) or by the FPRP approach (FPRP = 0.650 for rs17782313 and 0.752 for rs29941, at a prior probability level of 0.01 and power to detect an OR of 1.5). Similarly, haplotypes estimated from the 8 FTO
variants and 2 KCTD15
variants did not show a significant association with CRC (Supplemental Table 3
Association* of selected genetic risk variants for obesity with colorectal cancer, overall and by BMI level
We also examined individuals’ risk score by summing the number of risk alleles for the 15 loci (including two SNPs in low linkage disequilibrium (LD) for BDNF, rs8050136 for FTO, and rs29941 for KCTD15). The risk score, ranging between 4 and 22, showed a significant positive association with BMI (p <0.0001) but not with colorectal cancer (p = 0.47). The MC4R rs17782313 variant, but not KCTD15 rs29941, showed a slightly stronger association with rectal cancer (n = 444 cases; OR = 1.29, 95% CI 1.10–1.51) than with colon cancer (n = 1,369 cases; OR = 1.07, 0.96–1.18; p-heterogeneity = 0.03).
In a subgroup (1,640 cases, 8,878 controls) that had 109 ancestry informative markers (AIMs) data available, the SNP-colorectal cancer associations were examined with and without adjustment for the four principal components that represented the differential AIMs patterns for the 5 race/ethnic groups in the cohort.21
The results were similar, indicating no evidence of population stratification (data not shown).
BMI was positively associated with colorectal cancer risk (OR for a 5 kg/m2
increment = 1.19, 95% CI 1.12–1.26), with a slightly stronger association for colon cancer (OR = 1.21, 95% CI 1.13–1.29) than for rectal cancer (OR = 1.10, 95% CI 0.99–1.23), as reported previously.2
These risk estimates were not changed after adjustment for the rs29941 and rs17782313 variants (data not shown).
Because some of the obesity variants have shown different associations with obesity phenotypes by sex,7
or physical activity,23
we tested for evidence of heterogeneity. Results for FTO
that showed significant heterogeneity by weight status in some or all variants examined are presented in (Phet
<0.05). Three of the 8 variants in FTO
(r2 of 80–95% in whites, 54–98% in the other ethnic groups) and both KCTD15
variants (rs11084753 and rs29941; r2
ranging from 31% in African Americans to 60% in Japanese Americans and Native Hawaiians) had a significant inverse association with colorectal cancer among obese individuals, but not among normal-weight or overweight individuals. As in the main effects analysis, adjusting for BMI did not yield a notable change in the risk estimates (data not shown).
The SNP-cancer associations did not vary across ethnicity (see Supplemental Table 4
), except for the NRXN3
rs10146997 variant, which was inversely associated with colorectal cancer risk in whites only (p-heterogeneity
= 0.02). Similarly, there was no evidence of heterogeneity by sex, age (by median age of 70) or physical activity (by median 1.60 METs; data not shown).