The Epidemiologic Study Of BMI-Related Metabolites And Risk Of Colorectal Cancer

Background and Objectives:Colorectal cancer（CRC）is one of the most common malignant tumors that threaten human health and increase society burden worldwide.As the third most common and second most deadly cancer globally,CRC has become a major public health problem.Previous studies have recognized obesity as a leading risk factor for CRC,but its metabolic mechanism is not yet clear.Obesity can alter systemic metabolism,but the evidence about the effect of changes in metabolites caused by obesity on CRC was limited.In addition,identification of CRC at an early stage and interrupting the natural history is important to lower incidence and prolong survival.Although screening（eg,sigmoidoscopy,colonoscopy,fecal occult blood testing）has been demonstrated to reduce both the incidence and mortality of CRC,its diagnostic performance in the general population was not good.Identification of individuals at high risk of CRC is important for early prevention.Although a number of risk prediction models have been developed for CRC,most are based on risk factors collected from routine data or questionnaires and have shown a modest discriminatory ability.Therefore,there is an urgent need to examine the added value of plasma biomarkers for CRC prediction.Metabolomics has emerged as a powerful tool to identify novel biomarkers for metabolic characteristics and reveal mechanisms underlying complex diseases.Therefore,this study aims to（1）use metabolomics data to identify body mass index（BMI）-related metabolites and develop a metabolomic signature for BMI.Then,this study further evaluated the association between the metabolomic signature and CRC risk,so as to provide novel insights into the mechanisms underlying the obesity–CRC association;（2）leverage the summary data of large-scale genome-wide association analysis（GWAS）to conduct a two-sample Mendelian randomization（MR）study to investigate the causal association between identified BMI-related metabolites and CRC risk;（3）screen metabolic biomarkers for CRC.Furthermore,this study developed a CRC risk prediction model that incorporates metabolites and assessed whether adding these metabolites improved the discrimination of models based only on risk factors from questionnaires.Methods:1.Leveraging untargeted metabolomics data from a 1:1 matched,nested case–control study for CRC,including 223 pairs from the US Prostate,Lung,Colorectal,and Ovarian（PLCO）Cancer Screening Trial,this study performed Spearman’s partial correlation to estimate the correlations between baseline BMI and metabolites.Least Absolute Shrinkage and Selection Operator（LASSO）analysis was used to select metabolites that were most informative of BMI.The metabolomic signature was calculated as the weighted sum of the selected metabolites with weights equal to coefficients from the LASSO regression.Multivariable conditional logistic regression models were used to evaluate the associations of the metabolomic signature and BMI-related metabolites with CRC risk.Moreover,the mediating effect of the signature on the association between BMI and CRC was also assessed.In addition,this study externally validated the association of metabolomic signature with CRC risk and mediating effect of the signature on the association between BMI and CRC in another nested case-control study embed in the Jiangsu cohort.2.Two-sample MR study was used to investigate the causal association between BMI-related metabolites and CRC risk.Summary-level data of single nucleotide polymorphisms（SNPs）associated with the BMI-related metabolites as genetic instruments were taken from 3 large-scale GWASs.The summary genetic statistics for CRC were obtained from the most comprehensive GWAS of CRC to date,including 58,221 cases and 67,694 controls.In the main analysis,this study applied the Wald ratio or random-effect inverse-variance weighted（IVW）MR method to estimate the associations between BMI-related metabolites and the risk of CRC.Then,maximum likelihood,fixed-effect IVW,simple median,and weighted median methods were used to check the robustness of findings in the main analysis.Sensitivity analysis was carried out using the leave-one-out approach.The potential directional pleiotropy was assessed via the intercept term in the MR-Egger method.And the heterogeneity of SNP effects was assessed by Cochrane’s Q-statistic.3.Based on a nested case–control study for CRC,including 223 pairs from the PLCO Cancer Screening Trial,this study screened metabolic biomarkers and developed risk prediction models for CRC.The traditional risk factors model included age,sex,BMI,family history of CRC,smoking status,and alcohol consumption.Multivariable conditional logistic regression combined stepwise logistic regression were used to select differential metabolites between CRC cases and controls.This study then incorporated differential metabolites into the traditional risk factor model to assess whether these metabolites improved the predictively of the model.Model discrimination was assessed using C-statistics.Comparison of C-statistics generated from different models was conducted by the Delong test.Meanwhile,the Net Reclassification Index（NRI）was used to evaluate the improvement in prediction performance.In addition,this study externally validated the added value of metabolites for CRC prediction in the nested case-control study embed in the Jiangsu cohort.Results:1.The metabolomic signature of BMI.A total of 27 metabolites were correlated with BMI in PLCO,including 27 lipids,23 amino acids,7 peptides,5 carbohydrates,4 cofactors and vitamins,4 xenobiotics,1 nucleotide,and 1 energy.Among the 72 metabolites,33 metabolites were positively and 39 metabolites were inversely correlated with BMI.Using LASSO,this study identified that 39 of the 72 BMI-related metabolites were independent predictors of BMI.Then a metabolomic signature was created based on the coefficients generated in LASSO and the concentrations of selected metabolites.The signature was highly correlated with BMI（r=0.73,p<0.0001）and explained 53%of the variation in BMI.The signature mainly involved in amino acid and lipid metabolism.With regard to amino acid metabolism,the metabolomic signature of a higher BMI was characterized by increased levels of tryptophan metabolites（kynurenine and C-glycosyltryptophan）,glutamate,N-acetylalanine,tyrosine,and creatine,as well as decreased levels of glycine,serine and threonine metabolites（N-acetylglycine and betaine）,glutamine,asparagine,serotonin,and histidine.Among lipids,The metabolomic signature of a higher BMI was featured with increased levels of bile acid derivatives（deoxycholate and 7-HOCA）,carnitine metabolites（carnitine and butyrylcarnitine）,and androsteroid monosulfate 2,as well as decreased levels of fatty acids（docosahexaenoate and2-ethylhexanoate）,glycerophospholipid metabolites（1-palmitoylglycerophosphoethanolamine,2-palmitoylglycerophosphoethanolamine,and 1-stearoylglycerophosphoethanolamine）,steroids（androsterone sulfate,epiandrosterone sulfate,and cortisol）,and palmitoleoyl sphingomyelin.2.The association of the metabolomic signature of BMI and BMI-related metabolites with CRC risk.There was a positive linear association between the signature and CRC risk in both cohorts.Per 1-SD increment of the signature was associated with 38%（95%CI:9-75%）and 28%（95%CI:2-62%）higher risks of CRC in the PLCO and Jiangsu cohorts,respectively.The mediation proportion of the signature on the association between BMI and CRC was 35.7%and 17.0%,respectively.For the association between each BMI-related metabolite and CRC risk,this study found that glutamine was inversely associated with CRC risk in both cohorts,with OR per 1-SD increment=0.72（95%CI:0.57-0.92）and OR=0.50（95%CI:0.28-0.90）in the PLCO and Jiangsu cohorts,respectively.3.BMI-related metabolites instruments.SNPs that were identified to be associated with BMI-related metabolites at the genome-wide significance level(p value<5×?10^-8)in the published GWASs and were not in linkage disequilibrium（LD）with other SNPs（r²<?0.1 within a clumping window of 500?kb）were used as instruments for these metabolites.In addition,palindromic SNPs and SNPs with F-statistic less 10 were excluded.Finally,222 SNPs for 43BMI-related metabolites were selected as instruments.4.Association of BMI-related metabolites with CRC risk.Main analysis showed that each SD increase in genetically determined glutamine（OR:0.77,95%CI:0.65-0.89）,and 1,5-AG（OR:0.59,95%CI:0.36-0.96）was associated with a lower risk of CRC,while each SD increase in genetically determined mannose（OR:1.88,95%CI:1.20-2.93）,and gamma-glutamylvaline（OR:1.25,95%CI:1.05-1.48）was associated with a higher risk of CRC.Multiple MR models suggested that the causal association between BMI-related metabolites and CRC was very robust.Sensitivity analysis showed that there was no heterogeneity or horizontal pleiotropy,the causal relationship between BMI-related metabolites and CRC risk was not driven by a single SNP.5.Model discrimination in PLCO.Applying the predictors in the traditional risk factors model to CRC cases and controls showed a C-statistic of 0.58（95%CI:0.51-0.61）in PLCO.Through multivariable logistic regression combined stepwise regression,this study identified 13differential metabolites between cases and controls.Adding differential metabolites into the model,the C-statistic increased to 0.70（0.65-0.74）.This study also compared the two models,and the P-value<0.001,indicating that adding metabolites significantly improve the model performance.In reclassification analyses,NRI increased by 0.31（95%CI:0.22-0.60）when comparing the model with metabolic biomarkers to the model with conventional risk factors.6.Model discrimination in Jiangsu.Applying the predictors in the traditional risk factors model to CRC cases and controls showed a C-statistic of 0.55（95%CI:0.50-0.67）in Jiangsu.Adding differential metabolites into the model,the C-statistic increased to 0.62（0.56-0.67）.This study also compared the two models,and the P-value was 0.03,indicating that adding metabolites also significantly improve the model performance in Jiangsu.In reclassification analyses,NRI increased by 0.09（95%CI:0.04-0.35）when comparing the model with metabolic biomarkers to the model with conventional risk factors.Conclusions:1.This study identified a metabolomic signature of BMI involving multiple metabolic pathways and demonstrated its positive association with CRC risk.It provides novel insights into the mechanisms underlying the obesity–CRC association and informs future research to better identify individuals at high risk of CRC.Future studies are warranted to uncover metabolic targets and approaches for improved prevention of CRC.2.The two sample MR study showed that genetically determined glutamine and 1,5-AG were associated with a lower risk of CRC,while genetically determined mannose and gamma-glutamylvaline were associated with a higher risk of CRC.These four metabolites may be the main drivers of the positive association between obesity and CRC.3.This study developed and externally validated a risk prediction model with metabolites for CRC that showed improved discrimination over models based on risk factors from questionnaires only.These results provide evidence for the inclusion of metabolites in CRC risk assessment to promote early detection of the disease.