UKBB GWAS and genetic correlation estimates
In UKBB GWAS, we identified 998, 1,014 and 4 significant independent SNPs at 901, 718, 4 loci for BMI, WHRadjBMI and pancreatic cancer respectively (Fig. 1). None of the obesity signals were shared with pancreatic cancer in the UKBB. However, 3 of the 22 established pancreatic cancer loci by Klein et al5 were shared with WHRadjBMI in UKBB and had same direction of effect. These were NR5A2, ETAA1 and ZNRF3. Conversely, only ETAA1 from Klein et al5 was shared with BMI in the UKBB. Additionally, there was positive genetic correlation between both obesity measures and pancreatic cancer, but the estimates did not meet statistical significance (rGBMI=0.472, P = 0.479, rGWHRadjBMI=0.098, P = 0.671) (Supplementary table 1). Similarly, the genetic correlation between T2D and pancreatic cancer in the UKBB was underpowered and did not meet statistical significance (rG=-0.0139, P = 0.961) (Supplementary Table 1).
Effects of obesity variants on pancreatic cancer via polygenic scores
We identified a significant (Bonferroni multiple testing corrected P = 0.05/2 tests = 0.025) direct association between BMI PGS and pancreatic cancer (OR[95%CI] = 1.077[1.023–1.13], P = 0.0052). We also identified a direct association between WHRadjBMI PGS and pancreatic cancer, however, this association was not statistically significant (OR[95%CI] = 1.047[0.93–1.103], P = 0.088) (Table 1). To determine if the association between adiposity PGS and pancreatic cancer was driven by T2D, we adjusted for T2D in the association tests. After T2D adjustment, the significance of the association for both BMI and WHRadjBMI PGS declined suggesting that T2D could be acting via adiposity in pancreatic cancer risk (ORBMI_PGS[95%CI] = 1.069[1.015–1.079],P = 0.012); ORWHRadjBMI_PGS[95%CI] = 1.039[0.99–1.079], P = 0.15) (Table 1).
Table 1
Association between adiposity polygenic scores and pancreatic cancer
| Unadjusted model | T2D adjusted model |
Adiposity trait | OR (95%CI) | P | OR (95%CI) | P |
BMI | 1.077 (1.023–1.13) | 0.0052 | 1.069 (1.015–1.079) | 0.012 |
WHRadjBMI | 1.047 (0.93–1.103) | 0.088 | 1.039 (0.99–1.079) | 0.15 |
Legend: OR(95%CI) = Odds ratio of association and the lower and upper 95% confidence intervals (CI) |
Causality results using Mendelian randomization
We report a causal effect of WHRadjBMI on pancreatic cancer at nominal significance (OR[95%CI] = 1.00095[1.00011–1.0018], P = 0.027) based on the IVW method, indicating a weak but positive causal effect estimate (Fig. 2). However, none of the other MR tests for this direction were significant. The Cochran’s Q test indicated absence of heterogeneity among the genetic instruments (QIVW=258.08, P = 0.787). On the contrary, we have not identified any causal effect (Bonferroni P = 0.05/4 tests = 0.0125) of BMI on pancreatic cancer in either of the MR tests performed (Supplementary Table 2). There was no evidence of a causal effect from pancreatic cancer to WHRadjBMI (ORIVW(P) = 0.143(0.604). The results from pancreatic cancer to BMI were less informative with large standard errors despite nominal significance in some of the sensitivity MR tests (ORWeightedMedian[95%CI] = 58.105[3.997–844.69], P = 0.003) (Supplementary Table 2).