Separating the effects of early and later life adiposity on colorectal cancer risk: a Mendelian randomization study

Childhood obesity is a major global public health challenge with increasing prevalence observed in most geographic regions over the past three decades [1]. The rising prevalence of childhood obesity may have important consequences for population health, irrespective of adiposity later in life, given evidence from observational studies linking early life adiposity with elevated risks of chronic diseases, including cancer [1‐4]. Colorectal cancer has a long latency period suggesting that important exposures might have occurred many years ago and therefore is plausible that childhood and adolescence constitute a critical period during which adiposity influences cancer risk during adulthood [5]. Additionally, obesity during childhood and adolescence has been linked with unfavorable metabolic profiles that may influence cancer risk [6]. Therefore, it is possible that the detrimental role of obesity on colorectal cancer risk during adulthood might have started earlier in life.

Two recent meta-analyses of observational studies have reported positive associations between early life body size measures (in adolescence and early adulthood) and later life colorectal cancer risk in both men and women [7, 8]. Despite these associations, causal inference of the health effects of early life adiposity on later life disease risk can be challenging, as individuals who are obese in childhood often remain so during adulthood [9]. Consequently, it is currently unknown if the prior positive associations between early life adiposity and colorectal cancer risk are a direct effect of early life obesity or to what extent they are mediated by later life adiposity.

Mendelian randomization (MR) uses germline genetic variants as proxies to allow causal inference between a given exposure and outcome [10]. Compared to traditional observational analyses, MR analyses should be less susceptible to conventional confounding and reverse causation, given the randomly allocated and fixed nature of genetic variants [11]. Additionally, multivariable MR allows the estimation of independent effects of multiple exposures (e.g., early life and adult body size) on disease outcomes [12‐14]. Univariable MR can be used to estimate the total effect of early life body size on colorectal cancer (Fig. 1A), whereas multivariable MR can be used to estimate the effect of childhood obesity specifically on later life chronic disease risk, independently of adult body size [15, 16]. Under the multivariable framework, we hypothesize three scenarios in which early life body size affects colorectal cancer risk after considering adult body size (Fig. 1B–D): firstly, early life body size solely has direct effects on colorectal cancer, which are not influenced by adult body size (Fig. 1B); secondly, early life body size has only indirect effects on colorectal cancer through adult body size (Fig. 1C); thirdly, early life body size can have both direct and indirect effects on colorectal cancer risk (Fig. 1D). The same approach was applied recently to investigate whether early life body size influences risk of several diseases in later life, including breast and prostate cancer, and whether this effect is mediated by body size in adulthood [15].

We used a two-sample multivariable MR framework to examine potential causal associations between early life body size and colorectal cancer risk, independent of adult body size. We combined genetic variants associated with recalled early life and measured adult body size from a recent genome-wide association study (GWAS) of 453,169 adults [15] and then examined the association of these variants with colorectal cancer risk in a large consortium of up to 125,478 adults (58,131 cases and 67,347 controls) [17, 18].

In the current study, we investigated the effect of early life body size (at age 10 years) on risk of later life colorectal cancer and whether this effect remained robust after accounting for adult body size. Mendelian randomization estimates of the total effect of early life body size on colorectal cancer risk revealed that genetically predicted early life body size was estimated to increase odds of colorectal cancer risk which were strongest for colon and distal colon cancer. However, these associations were generally attenuated following adjustment for adult body size, suggesting no direct effects of early life body size on colorectal cancer risk. One exception was distal colon cancer, where the point estimate for the effect of early life body size on cancer risk remained in the multivariable scenario, though confidence intervals spanned the null. A similar pattern of results was reported in the sex-specific analyses, except for opposing effect estimates for both early life and adult body size on distal colon cancer under the multivariable framework, where greater early life body size appeared protective against distal colon cancer in male and detrimental in females and later life body size protective in females and detrimental in males, though both sets of estimates spanned the null.

Relatively few observational studies have examined the relationship between early life body size and risk of later life colorectal cancer. A joint Nurses’ Health Study (NHS) and Health Professionals Follow-up Study (HPFS) found that higher body fatness in early life (childhood and adolescence) was associated with greater colorectal cancer risk for women but not men with similar results across cancer sub-sites [37], with these associations largely unchanged after multivariable adjustment for adult BMI [37]. A Swedish study of more than 200,000 men who had their height and weight measured in adolescence reported a more than twofold higher risk for later life colorectal cancer for the obese group (BMI ≥ 30 kg/m²) when compared with the normal weight group (BMI 18.5-< 25 kg/m²) [38]. A large Israeli study of almost 1.8 million men and women showed that being overweight and obese at adolescence was associated with higher colon cancer risk for both men (hazard ratio [HR] for overweight, 1.53; 95% CI, 1.28–1.84; HR for obesity, 1.54; 95% CI, 1.15–2.06; statistically significant from a BMI of 23.4 kg/m²) and women (HR for overweight, 1.54; 95% CI, 1.22–1.93; HR for obesity, 1.51; 95% CI, 0.89–2.57; significant from a BMI of 23.6 kg/m²) [39]. However, the Swedish and Israeli studies lacked data on information on important potential confounders including adult body size.

A recent MR study examined the associations between childhood obesity and cancer risk without taking into account adult body size, using a genetic instrument of 15 SNPs from a GWAS of 47,541 children from the Early Growth Genetics (EGG) consortium [40]. Effect estimates from this study did not support a positive relationship between childhood BMI and overall colorectal cancer (OR, 1.11; 95% CI, 0.93–1.32) [40]. Results from our analyses that crucially adjusted for adult body size also reflect little evidence of a positive relationship between early life body size and later life colorectal cancer risk. An additional MR study following a similar approach but using a different dataset for colorectal cancer also did not find any positive relationship between childhood BMI and overall colorectal cancer after adjusting for adult body size [20]. However, this study did not investigate further associations stratified by sex or cancer sub-site. Our findings suggest that positive effect estimates found between early life body size and colorectal cancer may be attributable to participants who were overweight or obese in childhood remaining this weight during adulthood, rather than any distinct harm from childhood adiposity itself. There was weak evidence for a direct effect of early life body size on distal colon cancer risk which warrants further investigation in new and larger studies with measured early life body size. Moreover, multivariable models estimated a greater effect of adult body size on CRC in men compared to women. This finding is consistent with a recent MR study, which found a stronger effect of adult BMI on CRC for men in sex-stratified analyses (OR per 1 standard deviation change: 1.23; 95% CI, 1.08–1.38 in men versus 1.09; 95% CI, 0.97–1.22 in women) [41].

We observed divergent effect estimates for early life body size and distal colon cancer risk for men and women in the multivariable models (P-heterogeneity = 0.08). However, caution is needed when interpreting the results for men as the estimates for early life body size might suffer from weak instrument bias based on the low conditional F statistic.

Strengths of the current study include our use of novel multivariable MR methods to disentangle the effects of early life body size from adult body size on colorectal cancer risk. Importantly, both early life and adult body size data were sourced from the identical participants within UK Biobank belonging to the same generation of individuals contributing data to the colorectal cancer GWAS used in this analysis; this should reduce bias that can be introduced to MR estimates when two samples do not represent the same underlying population [15]. We obtained summary stats for colorectal cancer from the Huyghe et al. GWAS excluding UK biobank participants; thus, independent datasets for the exposure and outcome phenotypes were used, avoiding potential bias due to sample overlap [42]. Finally, the large cancer sample size allowed us to conduct a more detailed analysis by sex and cancer sub-site.

A limitation of our study is that the early life body size phenotype was based on a self-reported questionnaire data than on direct measurements, so we cannot exclude potential recall bias in the measurements. However, the early life body size SNPs were successfully validated in three independent studies with direct measures of childhood BMI, demonstrating their validity as instruments for early life body size [15, 16, 19]. In the ALSPAC cohort, a genetic score based on the SNPs early life body size was found to be a stronger predictor of childhood BMI compared with the score based on the adult variants (area under the curve [AUC]: 0.64 vs 0.61; average age 9.9 year) with a similar finding found in the Young Finns Study (AUC 0.74 vs 0.62; range 3–18 years of age) [15, 19]. In the HUNT study, it was observed that the genetic score for early life body size explained a higher proportion of variance in the measured BMI than the genetic score for adult body size (6.7% vs 2.4%; age group 12–15.9 years) [16]. As exposures investigated were categorical variables with three broad categories, this may limit power to detect small effects; this did not appear to limit site and sex combined analyses, but power was lower in analyses according to sex and subsite, as evidenced by the imprecision of these effect estimates. Moreover, analyzing two correlated phenotypes together like early and adult body size may have introduced collinearity which leads to greater imprecision and possible bias in the multivariable MR effect estimates especially for early life body size; however, the same set of SNPs were strong enough to show independent effects of early life body size on breast cancer (OR: 0.59, 95% CI: 0.50, 0.71) [15]. Additionally, examining the direct of early life body size on colorectal cancer risk in the MVMR analysis may be biased due to mediator adjustment (i.e., induced collider bias due to mediator-outcome confounding). It would be interesting to investigate the role of key confounders of the adult body size relationship; however, an MVMR model including confounders would considerably weaken the power of exposure instruments (as illustrated by the weak conditional F statistic for early life body size and reduced power for stratified analyses in Supplementary Table 4). Furthermore, given the random allocation of genetic variants at conception, the genetic instrument for adult body size should not be correlated with confounding factors. Additionally, it is expected that body size-related variants could contribute to CRC risk via different metabolic pathways, e.g., lipids or glycemic traits. A recent MR study found effects of BMI or WHR on numerous lipids and glycemic traits—some consistent effects were observed for certain lipid metabolites, where levels were raised by BMI and positively associated with the risk of distal colon cancer [41]. A future approach to minimize heterogeneity in instrument selection could be to analyze the association between clusters of genetic determinants of BMI in relation to CRC, informed by phenotypic associations. The lack of ancestral diversity in UK Biobank limits the generalizability of results to diverse populations. Lastly, given the large number of comparisons, we cannot exclude that some of the observed effects might be due to chance.

In conclusion, we found that the suggested effects of early life body size on later life colorectal cancer risk are likely a consequence of individuals who are overweight or obese during childhood remaining this way into adulthood. Our results suggest that being overweight in childhood may not be detrimental in terms of colorectal carcinogenesis providing individuals reach and then maintain normal weight during adulthood. Further research is required to examine any possible role of early life body size on distal colon cancer development.