Smoking metabolic signature modifies the risk of colorectal cancer and interacts with genetic susceptibility: a large-scale prospective cohort study Authors: Jing Wang, Jiawei Zhang, Zhuoyi Wu, Yuqiang Zhao, Shiyin Meng, Wenjie Xuan, Jing Ni Published: 2026 ABSTRACT BackgroundSmoking tobacco is proven to be associated with higher colorectal cancer (CRC) risk. However, the metabolic pathways of smoking and how they relate to CRC risk are unclear.MethodsA total of 227,898 participants from the UK Biobank were included. A two-stage strategy with a combination of linear regression and Mendelian randomization analysis was employed to identify smoking-related metabolites. Then the elastic net (EN) regression model was used to construct smoking-related metabolic signature. Multivariable Cox regression, Mediation analysis, and interaction analysis were employed to illustrate the associations of smoking and smoking related metabolic signature on the risk of CRC. Furthermore, the interaction between genetic susceptibility and smoking metabolic signature was estimated based on relative excess risk due to interaction (RERI) and multiplicative-scale interaction.ResultsDuring a follow-up of 12.5 years, 3,328 incident CRC patients were observed. We identified 71 smoking-related metabolites and screened 42 of them metabolites by EN model to construct smoking-related metabolic signature. We observed smoking and smoking metabolic signature were both associated with the elevated risk of CRC, with a hazard ratio (HR) value of 1.27 and 1.38, respectively. Mediation analysis revealed metabolic signature can mediate 10.03% in the association between smoking and CRC risk. The additive interaction between genetic susceptibility and metabolic signature was statistically significant (RERI = 0.70, 95% confidence interval [CI] = 0.29–1.10).ConclusionThe detrimental impact of smoking on CRC risk is mediated partially by smoking metabolic signature, which is synergistic with genetic susceptibility.Graphical Abstract Supplementary InformationThe online version contains supplementary material available at 10.1186/s12885-026-16123-x. Introduction Colorectal cancer (CRC) is the third most frequently diagnosed malignancy globally and the second leading cause of cancer-related mortality worldwide [1]. Addressing modifiable risk factors is essential for CRC prevention. Cigarette smoking, a public health issue, has been linked to CRC in several observational studies [2–5]. However, the mechanisms underlying how smoking triggers colorectal tumourigenesis are largely unexplored. Metabolomics profiles have indicated that metabolites differ between smokers and non-smokers [6]. For instance, Xu et al. reported 21 smoking-related metabolites predominantly enriched in amino acid and lipid pathways [7]. Beyond established nicotine metabolites, Gu et al. observed xenobiotic metabolites, involved in amino acids and lipids, associated with smoking behavior [8]. Moreover, a cross-sectional study found that cigarette smoking contributes to the occurrence of squamous cell carcinoma, mediated by glutamine, histidine, and cholic acid [9]. In another nested case–control study, smoking-related metabolic biomarkers showed stronger associations with CRC than self-reported current smoking [10]. Due to the limited sample size and the issues of reverse causality, uncertainties persist regarding the association of smoking-related circulating metabolites with CRC risk. A previous study combined observational study and Mendelian randomization (MR) analysis to investigate the association between smoking, metabolites, and T2D [11]. Therefore, it is imperative to conduct longitudinal investigations alongside MR studies to explore the association of metabolites affected by smoking with CRC risk, as well as the extent to which this association can be attributed to smoking-related metabolites. By leveraging data from a large prospective cohort and genome-wide association study summary statistics, we aimed to establish a metabolic signature of smoking and assess whether this signature mediates the association between smoking and CRC incidence. We also investigated the interactions between the smoking metabolic signature and genetic susceptibility on the incidence of CRC. Methods Study population The participants included in this study were derived from UK Biobank (UKB), a large prospective cohort study that recruited about half a million volunteers aged 37–73 years from 22 assessment centers in the United Kingdom between 2006 and 2010 [12]. The study design is shown in Fig. 1. We excluded participants who had a diagnosed CRC at baseline, those with missing data on smoking status, covariates (including age at recruitment, alcohol consumption, body mass index [BMI], ethnicity, Townsend deprivation index [TDI], physical activity, red and processed meats consumption, history of diabetes, cardiovascular disease [CVD], hypertension and all cancer), missing data on polygenic risk score (PRS) for CRC, individuals of non-white ethnicity, and those with missing data on metabolite measurements [11]. Ultimately, there were 227,898 participants eligible for the current study. The UKB project was approved by the North West Multi-Centre Research Ethical Committee (REC reference: 21/NW/0157), and all participants provided written informed consent.Fig. 1Overall study design. Abbreviation: CRC: Colorectal cancer; BMI: Body mass index; TDI: Townsend index; CVD: Cardiovascular disease; PRS: Polygenic risk score; MR: Mendelian randomization; IV: Instrumental variable; UKB: UK Biobank; GSCAN: GWAS and sequencing consortium of alcohol and nicotine use; GWAS: Genome-wide association study Overall study design. Abbreviation: CRC: Colorectal cancer; BMI: Body mass index; TDI: Townsend index; CVD: Cardiovascular disease; PRS: Polygenic risk score; MR: Mendelian randomization; IV: Instrumental variable; UKB: UK Biobank; GSCAN: GWAS and sequencing consortium of alcohol and nicotine use; GWAS: Genome-wide association study Ascertainment of smoking and covariates Current and past tobacco smoking status (UKB data ID: 20116) at baseline were assessed using touchscreen questionnaires. We categorized the smoking status into smokers (including former smokers and current smokers) and non-smokers based on participants’ responses. The selected covariates include demographic variables (including age at recruitment, sex, TDI, assessment center) and potential confounding factors (including BMI, consumption of alcohol, physical activity, red and processed meats consumption, aspirin use, and the history of diabetes, CVD, hypertension and all cancer) [13]. BMI was calculated by dividing height by body weight and categorized into four classifications: < 25, 25 to < 30, 30 to < 35, and ≥ 35 kg/m2 [14]. Physical activity was defined as low level (< 600 metabolic equivalents [MET]-min/week), intermediate level (≥ 600, < 3000 MET-min/week), and high level (≥ 3000 MET-min/week) [15]. Red and processed meat consumption was categorized into four classifications < 2, 2 to < 3, 3 to < 4, and ≥ 4 occasions per week [16]. The ascertainment of comorbidities was based on medical records (the 10th revision of the International Classification of Diseases [ICD-10] codes: diabetes: E10-E14, CVD: I20-I25, hypertension: I10-I13, and I15). All cancer was identified through the National Health Service (NHS) Central Register (ICD-10 codes: C00-C97). Metabolomics measurement The metabolic biomarkers were derived from the latest July 2023 release. 251 metabolic biomarkers (including 170 original measures and 81 percentage ratios) were measured from EDTA plasma samples at baseline using the high-throughout nuclear magnetic resonance (NMR)-based metabolomics platform developed by Nightingale Health Ltd [17]. The NMR assays covered a range of metabolites, including lipids in 14 subclasses, fatty acids, amino acids, ketone bodies, glycoprotein acetyls, and glycolysis metabolites. We selected the 170 absolute metabolic measurements for the present study. Definition of CRC outcome According to ICD-10 codes from the hospital inpatient records, incident CRC cases were defined as C18 (colon cancer), C19 (tumor of the colorectal junction), and C20 (rectum cancer). The endpoint event was determined as the first occurrence of CRC, loss to follow-up, death, or data censoring. Each participant’s person-years were calculated from the date of recruitment to the date of endpoint event. Identification of smoking-related metabolites We employed a two-stage strategy with a combination of the linear regression model and MR analysis to obtain reliable smoking-related metabolites from 168 original NMR metabolites, in 257,001 participants. Two metabolites (glucose-lactate and spectrometer-corrected alanine) with missing summary-level data were excluded, and a total of 168 metabolites were incorporated for analyses [18]. In the first stage, a linear regression model was utilized to identify which metabolites were associated with smoking. As the measurements of metabolites deviated from normal distribution, we performed a log-transformation (ln[x + 1]) and standardization (z-transformation) on these metabolites. The linear regression model treated smoking status and metabolites as exposure and response variables, respectively, adjusted for age, sex, assessment center, TDI, and the history of diabetes, CVD, and hypertension. Considering the intercorrelations among 168 metabolites, a principal component analysis was conducted. Ultimately, we identified 27 principal components that collectively accounted for more than 99% of the variance among these metabolites. The Bonferroni correction was used for multiple testing correction, with a P-value < 1.85 × 10–3 (0.05/27, 27 principal components explained the most variance among 168 metabolites) as the significance level. In the second stage, MR analysis aimed to further investigate which metabolites were causally affected by smoking. In MR analysis, to satisfy three core assumptions, the selection strategy of instrumental variables (IVs) were: (i) the P-value in GWAS < 5 × 10–8 (relevance assumption); (ii) the linkage disequilibrium (LD) < 0.01 in 1000 windows (independent assumption). And the frequency of effect allele of SNP > 0.45 and < 0.55 was excluded. If IVs were absence in outcome GWAS, a proxy SNP (LD > 0.8) was searched in LDlink website (https://ldlink.nih.gov/). We extracted 156 and 112 significantly independent single nucleotide polymorphisms (SNPs) as IVs for lifetime smoking index from UK Biobank and smoking initiation from GWAS and sequencing consortium of alcohol and nicotine use (GSCAN) consortium, respectively [19, 20]. Metabolites GWAS summary statistics were obtained from the UK Biobank [18]. Detailed information of GWAS summary statistics source and IVs was presented in Tables S1-S3. Four methods (inverse-variance weighted, MR-Egger, weighted median, weighted mode) were employed in two-sample MR analysis. Among them, the inverse-variance weighted method as the main MR method to estimate the causal relationships between smoking and metabolites. MR analyses were adhere to the STROBE-MR guidelines [21]. The statistically significant threshold of P-value was set as 0.05. A smoking-related metabolite should simultaneously meet the significant threshold and exhibit the same direction of correlation in both stages. Construction of smoking-related metabolic signature The smoking-related metabolic signature (referred to as “metabolic signature” below) was constructed based on the identified smoking-related metabolites using the elastic net (EN) regression model. The EN regression integrates the characteristics of Ridge regression and Least absolute shrinkage and selection operator (LASSO) regression and exhibits robust prediction performance when multicollinearity exists [22]. We treated the smoking status as the response variable and used the “caret” package to randomly split participants into training and testing sets in a 7:3 ratio. Then we used a 10-fold cross-validation approach to screen non-zero features in the training set at α = 0.5 and the minimum value of λ. Subsequently, we calculated the metabolic signature by the weights sum of selected metabolites using corresponding regression coefficients as weights [23]. The metabolic signature was stratified into two groups (low and high level) by the median value. Polygenic risk score construction To assess the genetic prediction of CRC, we constructed the PRS based on 90 independent SNPs identified in the previous study that reached the genome-wide significance threshold (P < 5 × 10–8) in the European population [24]. Detailed information of these SNPs is provided in Table S4. The PRS was calculated by the sum of the number of risk alleles multiplied by the weight of each SNP using the following formula:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${PRS}_{i}=\sum_{l=1}^{n}{X}_{il}{\beta }_{l}$$\end{document}where n is the number of independent significant SNPs, xil is the number of the risk alleles (0, 1, or 2) for i-th participant at the l-th SNP, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{l}$$\end{document} is weight of l-th SNP. The PRS was quintile stratified (Q1-Q5) and then defined as low (Q1), intermediate (Q2-Q4), and high (Q5) risk groups. Statistical analysis The missing value of metabolites was attributed to assay values below the limit of detection. If the metabolites’ missing rate was above 10%, these metabolites were excluded. If metabolites’ missing rate was below 10%, their missing value was imputed by half of the minimum of these measured values. Cox proportional hazards regression models were utilized to estimate the association of smoking status, smoking-related metabolites, and metabolic signature on the risk of CRC. The Cox models were fitted with follow-up time as time scale, adjusted for age, sex, assessment center, TDI, BMI, consumption of alcohol, physical activity, red and processed meats consumption, aspirin use, and the history of diabetes, CVD, hypertension, and all cancer. Furthermore, the mediation analysis was performed to assess the mediation proportion of smoking-CRC association mediated through metabolic signature. Due to metabolic signature and the status of CRC outcome were both binary variables, therefore, generalized linear regression model were employed to estimate the direct and indirect effect. The confidential interval of the effect of mediation was estimated through 5000 simulations. The above-mentioned analyses were performed in the subgroup analyses. We conducted the joint and interaction analyses to explore whether the coexistence of genetic susceptibility and smoking status/metabolic signature could enhance their overall impact on the risk of CRC. The relative excess risk due to interaction (RERI) and the attributable proportion due to interaction (AP) were calculated to examine the statistical significance of additive interaction by Cox models [25]. The 95% CIs of the RERI and AP were estimated by randomly 5000 bootstraps [26]. The multiplicative interaction between metabolic signature/smoking status and PRS was estimated by the Cox model with the multiplicative interaction term. To examine the robustness of the primary findings, we conducted the following sensitivity analyses: (1) excluded the CRC patients in the first two years during follow-up; (2) filled up the missing information of covariates by “mice” package (for the covariates missing percentage < 1%); (3) excluded the inflammatory bowel disease (IBD) patients at baseline. All of the above-mentioned statistical analyses were performed in R software (version 4.3.3). The P-value < 0.05 was considered statistically significant unless otherwise stated. Results Baseline characteristics Among the 227,898 participants included in this study (46.5% men), the mean age was 56.68 years. Over a median follow-up duration of 12.5 years, 3,328 incident CRC cases were documented. The baseline characteristics of the study participants are presented in Table 1. Compared to participants without CRC, those who developed CRC were older, more likely to be men, current smokers, had a higher BMI, and ate more red and processed meat. Furthermore, they tend to have a history of aspirin use, hypertension, diabetes, CVD, and a family history of cancer.Table 1Baseline characteristics of the study populationCharacteristicsTotal (N = 227,898)Non-incident CRC (n = 224,570)Incident-CRC (n = 3,328)PSmokers (%)103,687 (45.5)101,858 (45.4)1,829 (55.0)< 0.001Age (year)56.68 ± 8.0456.62 ± 8.0460.67 ± 6.70< 0.001Man (%)105,999 (46.5)104,085 (46.3)1,914 (57.5)< 0.001TDI−1.58 ± 2.93−1.58 ± 2.93−1.56 ± 2.970.816Drinking status (%)0.105 Never6,815 (3.0)6,726 (3.0)89 (2.7) Former7,275 (3.2)7,150 (3.2)125 (3.8) Current213,808 (93.8)210,694 (93.8)3,114 (93.6)Physical activity (%)†0.409 Low43,643 (19.2)42,976 (19.1)667 (20.0) Intermediate115,609 (50.7)113,945 (50.7)1,664 (50.0) High68,646 (30.1)67,649 (30.1)997 (30.0)BMI (kg/m2, %)< 0.001 < 2577,159 (33.9)76,264 (34.0)895 (26.9) 25 to < 3098,204 (43.1)96,646 (43.0)1,558 (46.8) 30 to < 3538,584 (16.9)37,951 (16.9)633 (19.0) ≥ 3513,951 (6.1)13,709 (6.1)242 (7.3)Diet (%) Red meats consumption< 0.001 < 1 occasion per week21,822 (9.6)21,590 (9.6)232 (7.0) 1 to 2 occasion per week17,831 (7.8)17,561 (7.8)270 (8.1) 2 to 3 occasion per week72,995 (32.0)71,987 (32.1)1,008 (30.3) ≥ 3 occasion per week115,250 (50.6)113,432 (50.5)1,818 (54.6) Processed meats consumption< 0.001 < 1 occasion per week88,336 (38.8)87,191 (38.8)1,145 (34.4) 1 to 2 occasion per week67,685 (29.7)66,674 (29.7)1,011 (30.4) 2 to 3 occasion per week63,011 (27.6)61,984 (27.6)1,027 (30.9) ≥ 3 occasion per week8,866 (3.9)8,721 (3.9)145 (4.4)Aspirin use (%)31,011 (13.6)30,415 (13.5)596 (17.9)< 0.001Comorbidities (%) Diabetes10,484 (4.6)10,252 (4.6)232 (7.0)< 0.001 CVD12,655 (5.6)12,410 (5.5)245 (7.4)< 0.001 Hypertension53,278 (23.4)52,276 (23.3)1,002 (30.1)< 0.001Family history of all cancer (%)82,510 (36.2)81,121 (36.1)1,389 (41.7)< 0.001Data are present as mean ± SD for continuous variables and n (%) for categorical variablesThe P-value was calculated by t test (continous variables) or χ2 test (category variables)Abbreviations: CRC Colorectal cancer, TDI Townsend deprivation index, BMI Body mass index, CVD Cardiovascular disease, SD Standard deviation†Physical activity was defined as low (< 600 metabolic equivalents (MET)-min/week), intermediate (≥ 600, < 3000 MET-min/week), high (≥ 3000 MET-min/week) Table 1Baseline characteristics of the study populationCharacteristicsTotal (N = 227,898)Non-incident CRC (n = 224,570)Incident-CRC (n = 3,328)PSmokers (%)103,687 (45.5)101,858 (45.4)1,829 (55.0)< 0.001Age (year)56.68 ± 8.0456.62 ± 8.0460.67 ± 6.70< 0.001Man (%)105,999 (46.5)104,085 (46.3)1,914 (57.5)< 0.001TDI−1.58 ± 2.93−1.58 ± 2.93−1.56 ± 2.970.816Drinking status (%)0.105 Never6,815 (3.0)6,726 (3.0)89 (2.7) Former7,275 (3.2)7,150 (3.2)125 (3.8) Current213,808 (93.8)210,694 (93.8)3,114 (93.6)Physical activity (%)†0.409 Low43,643 (19.2)42,976 (19.1)667 (20.0) Intermediate115,609 (50.7)113,945 (50.7)1,664 (50.0) High68,646 (30.1)67,649 (30.1)997 (30.0)BMI (kg/m2, %)< 0.001 < 2577,159 (33.9)76,264 (34.0)895 (26.9) 25 to < 3098,204 (43.1)96,646 (43.0)1,558 (46.8) 30 to < 3538,584 (16.9)37,951 (16.9)633 (19.0) ≥ 3513,951 (6.1)13,709 (6.1)242 (7.3)Diet (%) Red meats consumption< 0.001 < 1 occasion per week21,822 (9.6)21,590 (9.6)232 (7.0) 1 to 2 occasion per week17,831 (7.8)17,561 (7.8)270 (8.1) 2 to 3 occasion per week72,995 (32.0)71,987 (32.1)1,008 (30.3) ≥ 3 occasion per week115,250 (50.6)113,432 (50.5)1,818 (54.6) Processed meats consumption< 0.001 < 1 occasion per week88,336 (38.8)87,191 (38.8)1,145 (34.4) 1 to 2 occasion per week67,685 (29.7)66,674 (29.7)1,011 (30.4) 2 to 3 occasion per week63,011 (27.6)61,984 (27.6)1,027 (30.9) ≥ 3 occasion per week8,866 (3.9)8,721 (3.9)145 (4.4)Aspirin use (%)31,011 (13.6)30,415 (13.5)596 (17.9)< 0.001Comorbidities (%) Diabetes10,484 (4.6)10,252 (4.6)232 (7.0)< 0.001 CVD12,655 (5.6)12,410 (5.5)245 (7.4)< 0.001 Hypertension53,278 (23.4)52,276 (23.3)1,002 (30.1)< 0.001Family history of all cancer (%)82,510 (36.2)81,121 (36.1)1,389 (41.7)< 0.001Data are present as mean ± SD for continuous variables and n (%) for categorical variablesThe P-value was calculated by t test (continous variables) or χ2 test (category variables)Abbreviations: CRC Colorectal cancer, TDI Townsend deprivation index, BMI Body mass index, CVD Cardiovascular disease, SD Standard deviation†Physical activity was defined as low (< 600 metabolic equivalents (MET)-min/week), intermediate (≥ 600, < 3000 MET-min/week), high (≥ 3000 MET-min/week) Baseline characteristics of the study population Data are present as mean ± SD for continuous variables and n (%) for categorical variables The P-value was calculated by t test (continous variables) or χ2 test (category variables) Abbreviations: CRC Colorectal cancer, TDI Townsend deprivation index, BMI Body mass index, CVD Cardiovascular disease, SD Standard deviation †Physical activity was defined as low (< 600 metabolic equivalents (MET)-min/week), intermediate (≥ 600, < 3000 MET-min/week), high (≥ 3000 MET-min/week) Smoking-related metabolites and signature A total of 137 (96 positive and 41 negative) metabolites were significantly associated with smoking in the line regression model (P < 1.85 × 10–3). Genetically predicted levels of 85 and 74 metabolites were associated with lifetime smoking index and smoking initiation, respectively (Tables S5-S8). Ultimately, 71 smoking-related metabolites were validated. The results of heterogeneity and horizontal pleiotropy test were shown in Tables S9-10. Among them, 48 metabolites increased by smoking included glycoprotein acetyls (GlycA), fatty acids (total fatty acids [FA], monounsaturated fatty acids [MUFA], saturated fatty acids [SFA]), and various lipoprotein molecules. Meanwhile, 23 metabolites decreased by smoking including albumin, docosahexaenoic acid (DHA), and some lipids (Tables S5-S8). The associations of those smoking-related metabolites on the risk of incident CRC were presented in Fig. 2 and Tables S11-S12. The EN regression model screened 42 of 71 metabolites to create a metabolic signature, comprising 20 smoking metabolites and 22 smoking-decreased metabolites (Fig. 3 and Table S13). Notably, the top five metabolites that contributed to metabolic signature were esterified cholesterol in very large VLDL, total FA, MUFA, free cholesterol in very large VLDL, and SFA.Fig. 2The association between smoking-related metabolites and the risk of CRC. Abbreviation: XS_: Very small; S_: Small; M_: Medium; L_: Large; XL_: Very large; XXL_: Extremely large; VLDL: Very-low-density lipoprotein; LDL: Low-density lipoprotein; HDL: High-density lipoprotein; IDL: Intermediate density lipoprotein; TG: Triglycerides; SFA: Saturated fatty acids; MUFA: Monounsaturated fatty acids; FA: Total fatty acids; DHA: Docosahexaenoic acid; GlycA: Glycoprotein AcetylsFig. 3The weights of smoking-related metabolic signature identified by EN model. Abbreviations: XS_: Very small; S_: Small; M_: Medium; L_: Large; XL_: Very large; XXL_: Extremely large; VLDL: Very-low-density lipoprotein; LDL: Low-density lipoprotein; HDL: High-density lipoprotein; IDL: Intermediate density lipoprotein; TG: Triglycerides; _L: Lipids; _P: Particle concentrations of lipoproteins; _CE: Esterified cholesterol; _FC: Free cholesterol; _C: Cholesterol; SFA: Saturated fatty acids; MUFA: Monounsaturated fatty acids; FA: Total fatty acids; DHA: Docosahexaenoic acid; GlycA: Glycoprotein Acetyls. The image was generated by the prism 8.0 and Adobe Illustrator 2020 software The association between smoking-related metabolites and the risk of CRC. Abbreviation: XS_: Very small; S_: Small; M_: Medium; L_: Large; XL_: Very large; XXL_: Extremely large; VLDL: Very-low-density lipoprotein; LDL: Low-density lipoprotein; HDL: High-density lipoprotein; IDL: Intermediate density lipoprotein; TG: Triglycerides; SFA: Saturated fatty acids; MUFA: Monounsaturated fatty acids; FA: Total fatty acids; DHA: Docosahexaenoic acid; GlycA: Glycoprotein Acetyls The weights of smoking-related metabolic signature identified by EN model. Abbreviations: XS_: Very small; S_: Small; M_: Medium; L_: Large; XL_: Very large; XXL_: Extremely large; VLDL: Very-low-density lipoprotein; LDL: Low-density lipoprotein; HDL: High-density lipoprotein; IDL: Intermediate density lipoprotein; TG: Triglycerides; _L: Lipids; _P: Particle concentrations of lipoproteins; _CE: Esterified cholesterol; _FC: Free cholesterol; _C: Cholesterol; SFA: Saturated fatty acids; MUFA: Monounsaturated fatty acids; FA: Total fatty acids; DHA: Docosahexaenoic acid; GlycA: Glycoprotein Acetyls. The image was generated by the prism 8.0 and Adobe Illustrator 2020 software The association of smoking status/metabolic signature on the risk of CRC One thousand eight hundred twenty-nine of 3,328 newly diagnosed CRC patients were smokers. The incidence rate of CRC was 0.98 per 1000 person-years among non-smokers, compared with 1.47 per 1000 person-years among smokers. In the multivariate-adjusted Cox model, smoking was associated with a 27% elevated risk of CRC, with a hazard ratio (HR) value of 1.27 (95%CI = 1.18–1.36). When controlling for the metabolic signature in the above model, the effect of smoking on the CRC risk was slightly attenuated (HR = 1.24, 95%CI = 1.16–1.33) (Table 2). CRC patients were categorized into two groups based on the median value of metabolic signature level. The CRC incidence rates in the metabolic signature were similar to those in smoking status. Smoking metabolic signature was also positively associated with the risk of CRC, with an adjusted HR of 1.38 (95%CI = 1.25–1.51) per SD increment of metabolic signature level. Compared with the low metabolic signature level, the multivariable HR in the high level was 1.24 (95%CI = 1.15–1.34). The metabolic signature mediated 10.03% of the effect of smoking on CRC risk (Fig. 4).Table 2The effect of smoking and smoking-related metabolic signature on the risk of CRCSmoking statusMetabolic signatureNon-smokersSmokersLowHighPer SD increaseCases1,4991,8291,3252,0033,328Person years1,523,8161,247,6631,399,1491,372,3302,771,478incidence rate (per 1000 person-years)0.981.470.951.461.20HR (95%CI)a1.001.27 (1.18–1.36)1.001.24 (1.15–1.34)1.38 (1.25–1.51)HR (95%CI)b1.001.24 (1.16–1.33)1.00--Abbreviations: CRC Colorectal cancer, SD Standard deviation, HR Hazard ratio, CI Confidence intervalaCox models were fitted with follow-up time as time scale, adjusted for age, sex, assessment center, Townsend Deprivation Index, body mass index, consumption of alcohol, physical activity, red and processed meats consumption, aspirin use, and the history of diabetes, cardiovascular disease, hypertension, and all cancerbCox models were further adjusted for metabolic signature on the basis of model aFig. 4The mediation effect of smoking-related metabolic signature in the association between smoking and CRC risk. Abbreviation: CRC: Colorectal cancer Table 2The effect of smoking and smoking-related metabolic signature on the risk of CRCSmoking statusMetabolic signatureNon-smokersSmokersLowHighPer SD increaseCases1,4991,8291,3252,0033,328Person years1,523,8161,247,6631,399,1491,372,3302,771,478incidence rate (per 1000 person-years)0.981.470.951.461.20HR (95%CI)a1.001.27 (1.18–1.36)1.001.24 (1.15–1.34)1.38 (1.25–1.51)HR (95%CI)b1.001.24 (1.16–1.33)1.00--Abbreviations: CRC Colorectal cancer, SD Standard deviation, HR Hazard ratio, CI Confidence intervalaCox models were fitted with follow-up time as time scale, adjusted for age, sex, assessment center, Townsend Deprivation Index, body mass index, consumption of alcohol, physical activity, red and processed meats consumption, aspirin use, and the history of diabetes, cardiovascular disease, hypertension, and all cancerbCox models were further adjusted for metabolic signature on the basis of model a The effect of smoking and smoking-related metabolic signature on the risk of CRC Abbreviations: CRC Colorectal cancer, SD Standard deviation, HR Hazard ratio, CI Confidence interval aCox models were fitted with follow-up time as time scale, adjusted for age, sex, assessment center, Townsend Deprivation Index, body mass index, consumption of alcohol, physical activity, red and processed meats consumption, aspirin use, and the history of diabetes, cardiovascular disease, hypertension, and all cancer bCox models were further adjusted for metabolic signature on the basis of model a The mediation effect of smoking-related metabolic signature in the association between smoking and CRC risk. Abbreviation: CRC: Colorectal cancer Interaction with genetic susceptibility Compared with participants in the low PRS level, the HR for developing CRC was 1.68 (95%CI = 1.51–1.88) for those in the intermediate PRS level, 2.66 (95%CI = 2.37–2.99) for those in the high PRS level, respectively (Figure S1). Individuals with the combination of high metabolic signature level and high PRS level had an HR of 3.40 (95%CI = 2.84–4.08) compared to those with low levels of the metabolic signature and PRS, with significant additive interaction (RERI = 0.70, 95%CI = 0.29–1.10) (Table 3). Individuals with the combination of smoking and high PRS level had an HR of 3.44 (95%CI = 2.89–4.09) compared to those with never smoking and low PRS level, without evidence of additive or multiplicative interaction (Table S14).Table 3Joint and interaction analyses of smoking-related metabolic signature and genetic susceptibility with incident CRCPRSMetabolic signatureCases/Sample sizeHR (95%CI)Additive interactionMultiplicative interactionRERI (95%CI)AP (95%CI)HR (95%CI)LowLow155/22,655ReferenceIntermediateLow791/67,4981.71 (1.44–2.03)HighLow379/22,4712.48 (2.06–2.99)LowHigh230/22,5381.22 (1.00–1.50)IntermediateHigh1,144/67,3052.04 (1.72–2.42)0.10 (−0.20, 0.35)0.05 (−0.09, 0.19)0.97 (0.78, 1.22)HighHigh629/22,1033.40 (2.84–4.08)0.70 (0.29, 1.10)0.21 (0.08, 0.32)1.12 (0.88, 1.43)Cox models were fitted with follow-up time as time scale, adjusted for age, sex, assessment center, Townsend Deprivation Index, body mass index, consumption of alcohol, physical activity, red and processed meats consumption, aspirin use, and the history of diabetes, cardiovascular disease, hypertension and all cancerAbbreviations: CRC Colorectal cancer, PRS Polygentic risk score, HR Hazard ratio, CI Confidence interval, RERI Relative excess risk due to interaction (additive), AP Attributable proportion due to interaction Table 3Joint and interaction analyses of smoking-related metabolic signature and genetic susceptibility with incident CRCPRSMetabolic signatureCases/Sample sizeHR (95%CI)Additive interactionMultiplicative interactionRERI (95%CI)AP (95%CI)HR (95%CI)LowLow155/22,655ReferenceIntermediateLow791/67,4981.71 (1.44–2.03)HighLow379/22,4712.48 (2.06–2.99)LowHigh230/22,5381.22 (1.00–1.50)IntermediateHigh1,144/67,3052.04 (1.72–2.42)0.10 (−0.20, 0.35)0.05 (−0.09, 0.19)0.97 (0.78, 1.22)HighHigh629/22,1033.40 (2.84–4.08)0.70 (0.29, 1.10)0.21 (0.08, 0.32)1.12 (0.88, 1.43)Cox models were fitted with follow-up time as time scale, adjusted for age, sex, assessment center, Townsend Deprivation Index, body mass index, consumption of alcohol, physical activity, red and processed meats consumption, aspirin use, and the history of diabetes, cardiovascular disease, hypertension and all cancerAbbreviations: CRC Colorectal cancer, PRS Polygentic risk score, HR Hazard ratio, CI Confidence interval, RERI Relative excess risk due to interaction (additive), AP Attributable proportion due to interaction Joint and interaction analyses of smoking-related metabolic signature and genetic susceptibility with incident CRC Cox models were fitted with follow-up time as time scale, adjusted for age, sex, assessment center, Townsend Deprivation Index, body mass index, consumption of alcohol, physical activity, red and processed meats consumption, aspirin use, and the history of diabetes, cardiovascular disease, hypertension and all cancer Abbreviations: CRC Colorectal cancer, PRS Polygentic risk score, HR Hazard ratio, CI Confidence interval, RERI Relative excess risk due to interaction (additive), AP Attributable proportion due to interaction Subgroup analyses and sensitivity analyses In the stratified analyses, the positive association between metabolic signature and incident CRC was pronounced among men (Pinteraction = 0.025). No significant risk heterogeneity between smoking status/metabolic signature and other covariates in relation to the risk of CRC was detected (Pinteraction > 0.05) (Table S15). Additionally, the main results remained unchanged in all sensitivity analyses restricted to control participants at baseline (Table S16). Discussion In this large population-based epidemiological study, we found a variety of circulating metabolite biomarkers that are causally influenced by smoking and subsequently constructed a comprehensive smoking-related metabolic signature. Both smoking and its metabolic signature were associated with increased risk of CRC. Furthermore, the metabolic signature mediated 10.03% of the association between smoking and CRC. Additionally, participants with low genetic risk and low smoking metabolic level exhibited the lowest risk of incident CRC events across different smoking metabolic groups and PRS categories. We also observed an additive interaction between smoking metabolic signature and genetic susceptibility to CRC. In line with our findings, the detrimental effect of smoking on CRC has been reported in several previous studies [10, 27, 28]. We detected that smoking had a positive impact on the levels of GlycA, lipoproteins in small HDL, and some fatty acids (MUFA, FA, and SFA), while it had a negative impact on the levels of lipoproteins in very large HDL particles and some fatty acids (unsaturated fatty acid and DHA). Most of them have reported evidence showing a connection to tobacco smoke exposure. For example, Harada et al. reported that extensive amino acids and organic acids metabolic alteration in current cigarette smokers. Higher glutamate concentrations were observed in smokers, suggesting that smoking is associated with glutamate pathway and may induce endothelial dysfunction [29]. Previous studies have reported that smokers had significantly higher plasma GlycA and arachidonic acid (AA) levels compared to nonsmokers [30, 31]. A possible explanation is the elevated concentration of sPLA259 (a pivotal metabolic enzyme that could hydrolyze AA from phospholipids and release unoxidized AA into the blood) in tobacco smokers [32]. Gu et al. discovered that xenobiotics (such as tobacco, benzoate, and xanthine), amino acids, lipids, and vitamins were associated with current smoking status compared to non-smoking status. The underlying biological mechanism is cigarette smoking may induce enzymes that metabolize smoking-related metabolites, such as CYP1A2, xanthine oxidase, and N-acetyltransferase 2, leading to increases in xanthine metabolites [8]. Through a two-stage analytic strategy, we identified smoking-related metabolic profiles and prioritized 42 metabolites to establish a metabolic signature by a regression model. The metabolic reprogramming was observed in previous study [33]. The selected metabolic biomarkers were found to be correlated with smoking in both linear regression and MR analyses, which minimized the bias from reverse causality and confounders. Several potential mechanisms are involved in the correlation between metabolic alteration caused by smoking and the risk of CRC. Bai et al. have demonstrated that cigarette smoking influences dysbiosis of gut microbiotas. Compared to the smoke-free control mice, they found that mice exposed to cigarette smoke exhibited significant enrichment of Eggerthella lenta and enhanced oncogenic MAPK/ERK signaling pathway [28]. Moreover, the microbiome can induce excessive weight gain in cigarette smoke-exposed mice, concertedly with host to modulate the depletion of N-acetylglycine, a weight-lowering metabolite [34]. These findings highlight the importance of gut microbiota in the association between smoking and colorectal tumorigenesis. Our study is the first to quantify the mediation role of smoking-related metabolic signature in the association between smoking and CRC. A similar study of smoking and T2D found that smoking-related metabolic signatures mediated 38.3% of the excess risk [11]. The 10.03% mediation proportion in our study implied that other pathophysiological mechanisms triggered by smoking may play a role in CRC development. Compared with non-smokers, extensive DNA methylation pattern changes have been observed in smokers, indicating DNA methylation may play an essential role in smoking-induced colorectal neoplasms [35]. Moreover, the carcinogens in cigarette smoke (e.g. polycyclic aromatic hydrocarbons) can cause irreversible damage in DNA by forming DNA adducts [36]. Additionally, we detected a sex difference in the relation of smoking metabolic signature with CRC, which might be attributed to the greater dose of tobacco exposure in males. Our results align with previous studies that show PRS was associated with a gradual increased risk of CRC [37–39]. We found a synergistic effect between smoking metabolic signature and PRS in the development of CRC. As far as we are aware, accumulating evidence has shown several lifestyles interact with genetic components to modify the risk of CRC. For example, Choi et al. observed a significant additive interaction between healthy lifestyle scores and PRS. Individuals with a high genetic susceptibility are more likely to reduce CRC risk through lifestyle modifications than those with low genetic susceptibility [40]. Focusing on the interaction between gene and lifestyle factors, some pioneering studies have identified that these interactions can increase or reduce the risk of CRC. Aglago et al. discovered a locus within FMN1/GERM1 gene region which interacts with BMI and associates with CRC risk, implying the necessity of investigating the etiologic link between obesity and CRC [41]. Another gene-environment interaction study suggested that fiber and fruit related genetic variation might modify the association between these dietary factors and CRC risk, and the underlying mechanisms may relate to chronic inflammatory intestinal disorders and bowel function [42]. We need to acknowledge several limitations. Firstly, despite including 168 metabolites sourced from the NMR platform in UK Biobank, the number of metabolites analyzed was still restricted. Consequently, the mediation proportion of smoking-related metabolic signature on CRC risk might be underestimated, given that only 42 metabolites were considered in the analysis. Secondly, we calculated the metabolic signature for smoking exposure and estimated its impact on CRC risk in the same cohort. Although we utilized a 10-fold cross-validation method to control the effect of overfitting, it couldn’t fully eliminate potential bias. Further work is warranted to validate our findings in other independent cohorts. Thirdly, in the MR analysis, the influence of confounding was not consideration. However, two smoking-related IVs were used in our study, and the results remained stable, suggesting that the impact of confounding factors were limited. Fourthly, metabolomic biomarkers were measured at baseline, which can not reflect a trajectory of change in these biomarkers during the follow-up period. Our study simultaneously employed both a linear model and MR analysis to filter out reliable metabolites influenced by smoking. Finally, the current analyses were restricted to the European population, data from a wider range of populations are needed to generalize to other ethnicities. Conclusions Collectively, our study verified that greater exposure to tobacco smoking can lead to alterations in metabolites and demonstrated a detrimental impact on CRC development. The metabolic signature of smoking might mediate the heightened CRC risk in smokers, and these metabolite alterations exacerbate the CRC risk particularly in individuals genetically predisposed to the disease. These findings will contribute to deepening our understanding of the intricate connections among smoking, metabolites, and colorectal tumorigenesis, and provide new insight into strategies for CRC prevention. Supplementary Information Supplementary Material 1. Supplementary Material 1.