Causal effects of blood and urinary biomarkers on cancer risk: A Mendelian randomization study

Abstract

Objective

Observational studies have linked various serum and urinary biomarkers to cancer risk, but confounding and reverse causation limit causal inference. This study aimed to investigate the causal effects of 35 circulating biomarkers on the risk of 13 site-specific cancers.

Methods

We performed a comprehensive two-sample Mendelian randomization analysis using genetic instruments for 35 biomarkers from the UK Biobank (N = 363,228). Summary-level data for 13 cancers were obtained from the FinnGen consortium and IEU Open genome-wide association study. We employed inverse-variance weighted as the primary method, supplemented by sensitivity analyses. Significant associations were further scrutinized using bidirectional Mendelian randomization to rule out reverse causality and multivariable Mendelian randomization to assess independent effects.

Results

Forward Mendelian randomization identified 49 associations. After false discovery rate correction, nine remained: colorectal cancer—nonalbumin protein (odds ratio = 0.86, 95% confidence interval: 0.80–0.93), total protein (odds ratio = 0.85, 95% confidence interval: 0.77–0.93), and uric acid (odds ratio = 0.89, 95% confidence interval: 0.83–0.97); gastric cancer—apolipoprotein B (odds ratio = 0.85, 95% confidence interval: 0.79–0.95) and low-density lipoprotein cholesterol (odds ratio = 0.87, 95% confidence interval: 0.79–0.95); liver cancer—gamma-glutamyl transferase (odds ratio = 1.29, 95% confidence interval: 1.11–1.50); breast cancer—high-density lipoprotein cholesterol (odds ratio = 1.08, 95% confidence interval: 1.04–1.13); and esophageal cancer—total cholesterol (odds ratio = 0.78, 95% confidence interval: 0.68–0.90). Reverse Mendelian randomization suggested no reverse causality except breast cancer on C-reactive protein. Multivariable Mendelian randomization confirmed 12 biomarkers with independent effects: liver cancer (cholesterol, gamma-glutamyl transferase, high-density lipoprotein cholesterol, and low-density lipoprotein); breast cancer (C-reactive protein, high-density lipoprotein cholesterol, and insulin-like growth factor 1); gastric cancer (apolipoprotein B, low-density lipoprotein, and sex hormone-binding globulin); colorectal cancer (insulin-like growth factor 1 and uric acid); plus single markers for thyroid cancer (blood urea nitrogen), cervical cancer (C-reactive protein), ovarian cancer (direct bilirubin), and endometrial cancer (sex hormone-binding globulin).

Conclusions

This large-scale Mendelian randomization study provides robust evidence supporting the tissue-specific causal roles of several circulating biomarkers in the pathogenesis of specific cancers. These findings enhance our understanding of cancer etiology and highlight potential biomarkers for risk assessment and prevention.

Keywords

Serum biomarkers urinary biomarkers cancer risk Mendelian randomization causal inference

Introduction

Cancer is the second leading cause of mortality globally, incurring substantial medical and socioeconomic costs.¹ As a result, cancer prevention and screening are crucial. Despite the rising incidence of cancer, the exact mechanisms and causes underlying tumorigenesis and cancer progression are not fully understood. This highlights the need for a deeper understanding of the etiology, risk factors, and protective factors associated with these diseases.² Serum and urinary biomarkers can act as reliable biological indicators for disease identification and provide insights for diagnosis and prognosis. The detection of relevant positive biomarkers in serum and urine as diagnostic indicators for certain cancers may indicate the likelihood of developing malignancy. Identifying suitable biomarkers holds substantial clinical significance for the diagnosis and treatment of tumors.³

Over the past few decades, epidemiological research has identified a variety of genetic, lifestyle, and environmental factors associated with cancer risk.² However, these factors do not fully explain the etiology of cancer. Furthermore, the complex interplay among these factors adds to the difficulty of determining their potential causal relationships with cancer risk. Previous Mendelian randomization (MR) studies have typically focused on individual biomarkers or single cancer types and few have systematically assessed a broad range of blood and urinary biomarkers across 13 cancers using bidirectional and multivariable approaches. Addressing these gaps is essential to clarify causal pathways and identify biomarkers with independent effects. Therefore, our study provides a comprehensive bidirectional and multivariable MR analysis to systematically evaluate these associations and identify independent causal effects. The UK Biobank (UKB) is a longitudinal cohort study³ that evaluates the genetic determinants of 35 serum and urinary biomarkers in 363,228 participants. Numerous observational studies have investigated the relationship between various UKB biomarkers and cancer, revealing significant associations for some biomarkers. For instance, several MR studies have suggested that genetically predicted levels of serum and urinary biomarkers may be causally associated with cancer risk.⁴ For example, circulating docosahexaenoic acid⁵ has been linked to an increased risk of lung cancer. Moreover, lipids have been associated with urinary tract tumors.³ Nevertheless, observational studies may be susceptible to residual confounding and reverse causality. The aim of this study was to use MR to investigate the potential causal effects of genetically predicted levels of UKB serum and urinary biomarkers on the risk of 13 types of cancer. For the significant causal relationships identified, a series of supplementary analyses was conducted to strengthen their reliability and robustness.

Methods

MR assumptions and study design

The present study performed MR to evaluate the causal effects of 35 serum and urinary biomarkers on cancer. In MR studies, genetic variants are the most effective instrumental variables (IVs). MR theory specifies three fundamental assumptions for eligible IVs:⁶

The IVs are robustly associated with exposure.

The IVs are not associated with any hidden confounders.

The IVs are not directly associated with the outcome and affect the outcome only indirectly through the exposure under investigation.

To validate these assumptions in practice, we performed the following analyses:

Assumption (1) Strong association with exposure. We calculated F statistics for each IV; only IVs with F >10 were retained to avoid weak-instrument bias.

Assumption (2) No confounding. We used the Steiger test to confirm the direction of causality and rule out reverse causation.

Assumption (3) No direct effect on the outcome (no horizontal pleiotropy). We conducted MR-Egger intercept tests and Mendelian Randomization Pleiotropy Residual Sum and Outlier (MR-PRESSO) global tests to detect and correct for horizontal pleiotropy. Leave-one-out analyses were performed to examine whether any single single-nucleotide polymorphisms (SNP) drove the causal estimates.

MR analysis

The causal associations between serum and urinary biomarkers and cancer were estimated using the inverse-variance weighted (IVW), weighted median, and MR-Egger methods:

IVW estimates the causal effect by combining the ratio estimates for each SNP.

The weighted median method can provide unbiased estimates even if up to 50% of the information is derived from invalid IVs.

MR-Egger can estimate the causal effect using the slope coefficient from Egger regression and can also detect small-study bias.

SNP selection and harmonization. SNPs were selected as genetic instruments for each biomarker using the following criteria: genome-wide significance threshold of p <5 × 10⁻⁸ and linkage disequilibrium (LD) clumping with r² <0.001 within a 10,000 kb window. Palindromic SNPs with a minor allele frequency (MAF) >0.42 were removed. Harmonization of the exposure and outcome genome-wide association study (GWAS) summary statistics was performed using the harmonise_data function in the ‘TwoSampleMR’ package, which aligns effect alleles and corrects for strand ambiguity. No proxy SNPs were used.

Sensitivity analysis

A series of sensitivity analyses was used to assess the robustness of the statistically significant causal associations. Cochran’Q statistic was applied to evaluate heterogeneity. The MR-Egger intercept analysis evaluated horizontal pleiotropy, in which IVs affect both the exposure and the outcome through a pathway not mediated by the causal effect. Leave-one-out analysis was conducted by removing SNPs one at a time to evaluate whether a single SNP drove the significant results. MR-PRESSO test was used to detect the influence of outliers.

Reverse MR analysis

We conducted additional reverse MR analyses for 13 types of cancer and 35 serum and urinary biomarkers to explore the possibility of reverse causation. The reverse MR analysis procedure was identical to the MR analysis described earlier.

Multivariable MR

To mitigate the potential pleiotropic effects of serum and urinary biomarkers on tumorigenesis indicated by the positive univariable results, it was necessary to estimate the effect of each positive biomarker on tumor development while accounting for the influence of the other positive biomarkers. We used multivariable Mendelian randomization (MVMR) analysis to address this issue. For each cancer type, we included all biomarkers that showed significant associations in the univariable MR analysis (p < 0.05). Genetic instruments for MVMR were constructed by selecting SNPs associated with any of these biomarkers at genome-wide significance (p < 5 × 10⁻⁸) after LD clumping (r² < 0.001). Using the multivariable extension of the IVW MR method (MVMR), together with sensitivity analyses robust to pleiotropy (MVMR-Egger and MVMR-median), we evaluated the effect of each positive biomarker on the outcome. The odds ratios (ORs) derived from the MR estimates were used to determine the extent to which a one-unit increase in biomarker levels influenced cancer risk. To assess collinearity among the included biomarkers, we calculated the variance inflation factor (VIF); VIF >10 indicated problematic collinearity. Conditional F statistics were computed to evaluate weak-instrument bias in the MVMR model. Correlations among exposures were inherently accounted for by the MVMR model, which estimates the effects of each exposure conditional on the others. Instrument overlap was permitted because the same SNP may be associated with multiple biomarkers; this was accommodated within the MVMR framework. Exposure covariance was estimated using the mv_extract_exposures function in the ‘TwoSampleMR’ package. All participants whose data were included in this study received approval from the relevant institutions and provided written informed consent. A flowchart illustrating the study design is shown in Figure 1.

Figure 1.

Process flowchart of the research methodology.

Statistical analysis

All statistical analyses were conducted using R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria) with the ‘TwoSampleMR’ (v0.5.7) and ‘MR-PRESSO’ (v1.0) packages. The primary analysis used the IVW method. To account for multiple testing across the 35 biomarkers and 13 cancer types, we applied the Benjamini-Hochberg false discovery rate (FDR) correction. Associations with an FDR-adjusted p <0.05 were considered statistically significant. No formal sample size calculation was performed. The sample size was determined by the availability of summary-level GWAS data. We acknowledge that the limited sample size may have affected the statistical significance of some results. This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology Using Mendelian randomization (STROBE-MR) guidelines. The completed STROBE-MR checklist is provided as supplementary material.

Result

Genetic instrument selection

Following stringent quality-control procedures, we identified robust genetic instruments for all 35 serum and urinary biomarkers. A total of 6709 independent SNPs were included in the final MR analysis. The F statistics for all instruments substantially exceeded the threshold of 10, indicating a low risk of weak-instrument bias (Figure 1, Supplementary File 1)

Causal effects of biomarkers on cancer risk (forward MR)

The forward MR analysis, using the IVW method as the primary approach, identified 49 potential causal associations between the 35 biomarkers and 13 site-specific cancers at a significance level of p <0.05 (Figure 2, Tables 1 to 5).

Figure 2.

This heatmap visualizes the results of the forward Mendelian randomization analysis based on the inverse-variance weighted (IVW) method. The X-axis represents the 13 site-specific cancers, and the Y-axis corresponds to the 35 blood and urinary biomarkers. The color intensity of each cell indicates the strength of the association's significance: orange color indicates a potential causal association (positive or negative) between the biomarker and cancer risk at a significance level of p <0.05, with a darker shade representing a smaller p value (i.e. greater significance). Green color indicates a nonsignificant association (p ≥ 0.05). This figure provides an intuitive overview of the widespread potential causal associations between various biomarkers and cancer risk prior to multiple testing correction.

Table 1.

MR estimates for lung cancer and melanoma skin cancer biomarkers using the IVW method.

			IVW			MR-Egger		Weighted median
Outcome	Exposure	Number of IVs	OR (95% CI)	p	p_fdr	OR (95% CI)	p	OR (95% CI)	p
Lung cancer
	NAP	293	0.875 (0.794–0.965）	0.008	0.204	0.816（0.673–0.990）	0.0400	0.811（0.689–0.955）	0.012
	TES	89	0.787（0.653–0.948）	0.012	0.204	0.648（0.436–0.964）	0.035	0.789（0.594–1.049）	0.103
	TP	242	0.883（0.789–0.987）	0.029	0.314	0.823（0.646–1.049）	0.117	0.846（0.705–1.015）	0.071
	ALT	190	0.880（0.782–0.992）	0.036	0.314	0.937（0.737–1.191）	0.597	0.937（0.744–1.179）	0.578
	BUN	170	0.868（0.755–0.998）	0.0468	0.328	0.946（0.677–1.321）	0.744	0.891（0.733–1.083）	0.248
Melanoma skin cancer
	ApoB	212	1.001（1.000–1.002）	0.012	0.348	1.001（0.999–1.002）	0.283	1.001（0.999–1.002）	0.209
	LDL-C	217	1.001（1.000–1.002）	0.020	0.348	1.000（0.999–1.002）	0.570	1.001（0.999–1.002）	0.283
	HbA1c	250	0.999（0.997–0.100）	0.035	0.382	0.999（0.997–1.001）	0.207	0.999（0.997–1.000）	0.138
	PHOS	156	1.002（1.000–1.003）	0.044	0.382	1.001（0.998–1.004）	0.475	1.001（0.999–1.004）	0.270

ALT: alanine aminotransferase; ApoB: apolipoprotein B; BUN: blood urea nitrogen; CI: confidence interval; HbA1c: glycated hemoglobin; IV: instrumental variable; IVW: inverse-variance weighted; LDL-C: low-density lipoprotein cholesterol; MR: Mendelian randomization; NAP: nonalbumin protein; OR: odds ratio; p_fdr: false discovery rate-adjusted p value; PHOS: phosphate; TES: testosterone; TP: total protein.

Table 2.

MR estimates for colorectal and gastric cancer biomarkers using the IVW method.

Outcome			IVW			MR-EGGER		Weighted median
Colorectal cancer	Exposure	Number of IVs	OR (95% CI)	p	p_fdr	OR (95% CI)	p	OR (95% CI)	p
	NAP	290	0.862（0.799–0.931）	<0.001	0.005	0.910（0.783–1.058）	0.222	0.982（0.881–1.095）	0.743
	IGF-1	327	1.141（1.062–1.226）	<0.001	0.006	1.203（1.040–1.393）	0.014	1.175（1.052–1.311）	0.004
	TP	240	0.846（0.767–0.932）	0.001	0.009	0.868（0.704–1.071）	0.187	0.818（0.720–0.930）	0.002
	UA	231	0.893（0.825–0.966）	0.005	0.044	0.879（0.773–1.001）	0.052	0.841（0.740–0.957）	0.009
	PHOS	174	1.116（1.015–1.226）	0.024	0.166	1.103（0.933–1.304）	0.254	1.137（0.984–1.315）	0.082
	SHBG	236	1.806（1.005–1.173）	0.037	0.214	1.183（1.023–1.367）	0.024	1.076（0.965–1.200）	0.190
Gastric cancer
	ApoB	237	0.846（0.788–0.948）	0.002	0.036	0.914（0.780–1.071）	0.267	0.847（0.734–0.977）	0.023
	LDL-C	242	0.868（0.794–0.950）	0.002	0.036	0.890（0.759–1.042）	0.149	0.853（0.733–0.994）	0.041
	SHBG	240	1.111（1.012–1.219）	0.027	0.259	1.141（0.953–1.366）	0.151	1.115（0.961–1.249）	0.150
	CHOL	248	0.917（0.847–0.991）	0.030	0.259	0.954（0.836–1.089）	0.490	0.866（0.754–0.995）	0.042

ApoB: apolipoprotein B; CHOL: total cholesterol; CI: confidence interval; IGF-1: insulin-like growth factor 1; IV: instrumental variable; IVW: inverse-variance weighted; LDL-C: low-density lipoprotein cholesterol; MR: Mendelian randomization; NAP: nonalbumin protein; OR: odds ratio; p_fdr: false discovery rate-adjusted p value; PHOS: phosphate; SHBG: sex hormone-binding globulin; TP: total protein; UA: uric acid.

Table 3.

MR estimates for liver, esophageal, and pancreatic cancer biomarkers using the IVW method.

Outcome			IVW			MR-EGGER		Weighted median
Hepatic cancer	Exposure	Number of IVs	OR (95% CI)	p	p_fdr	OR (95% CI)	p	OR (95% CI)	p
	GGT	278	1.290（1.109–1.499）	0.001	0.032	1.213（0.925–1.591）	0.164	1.101（0.843–1.436）	0.480
	CHOL	247	0.837（0.712–0.985）	0.032	0.348	0.920（0.701–1.208）	0.548	0.807（0.627–1.038）	0.095
	HDL	256	0.862（0.747–0.994）	0.041	0.348	1.221（0.981–1.520）	0.075	1.009（0.806–1.262）	0.939
	LDL-C	243	0.834（0.695–0.100）	0.049	0.348	0.785（0.573–1.075）	0.133	0.767（0.586–1.004）	0.053
Esophageal cancer
	CHOL	249	0.779（0.676–0.898）	0.001	0.020	0.816（0.646–1.030）	0.088	0.934（0.757–1.153）	0.525
	LDL-C	244	0.830（0.718–0.958）	0.011	0.193	0.781（0.619–0.985）	0.038	0.920（0.729–1.160）	0.481
	ALB	197	0.781（0.622–0.982）	0.034	0.285	0.502（0.299–0.843）	0.010	0.809（0.573–1.142）	0.228
	GGT	277	0.853（0.730–0.998）	0.047	0.285	0.838（0.630–1,.113）	0.223	0.812(0.635–1.039)	0.099
Pancreatic cancer
	UA	232	1.243（1.034–1.495）	0.020	0.470	1.153（0.856–1.553）	0.350	1.174（0.854–1.613）	0.324

ALB: albumin; CHOL: total cholesterol; CI: confidence interval; GGT: gamma-glutamyl transferase; HDL: high-density lipoprotein cholesterol; IV: instrumental variable; IVW: inverse-variance weighted; LDL-C: low-density lipoprotein cholesterol; MR: Mendelian randomization; OR: odds ratio; p_fdr: false discovery rate-adjusted p value; UA: uric acid.

Table 4.

MR estimates for cervical, ovarian, and endometrial cancer biomarkers using the IVW method.

Outcome			IVW			MR-EGGER		Weighted median
Cervical cancer	Exposure	Number of IVs	OR (95% CI)	p	p_fdr	OR (95% CI)	p	OR (95% CI)	p
	CRP	134	1.001（1.000–1.002）	0.005	0.152	1.000（0.998–1.001）	0.825	1.001（0.996–1.003）	0.144
	AST	136	1.001（1.000–1.003）	0.009	0.152	0.999（0.997–1.001）	0.566	1.001（0.999–1.003）	0.412
	TRIG	127	1.001（1.000–1.002）	0.017	0.183	1.001（0.999–1.002）	0.260	1.001（1.000–1.002）	0.146
	TP	142	1.001（1.000–1.002）	0.033	0.250	1.003（1.000–1.006）	0.046	1.001（0.999–1.003）	0.238
	NAP	173	1.001（1.000–1.002）	0.038	0.250	1.002（1.000–1.004）	0.103	1.001（1.000–1.002）	0.206
Malignant neoplasm of ovary
	BILD	77	1.3655（1.047–1.780）	0.0215	0.529	1.424（1.035–1.959）	0.033	1.299（0.882–1.915）	0.186
	IGF-1	300	0.762（0.588–0.987）	0.0399	0.529	0.820（0.486–1.385）	0.459	0.850（0.547–1.322）	0.471
	URK	10	11.330（1.051–122.121）	0.045	0.529	11.159（0.000–9163903）	0.737	39.130（1.808–847.028）	0.019
Endometrial cancer
	SHBG	241	0.810（0.708–0.927）	0.002	0.078	0.678（0.527–0.872）	0.003	0.733（0.572–0.939）	0.014
	ALT	192	1.212（1.001–1.469）	0.049	0.496	0.926（0.632–1.359）	0.696	1.033（0.735–1.453）	0.850

ALT: alanine aminotransferase; AST: aspartate aminotransferase; BILD: direct bilirubin; CI: confidence interval; CRP: C-reactive protein; IGF-1: insulin-like growth factor 1; IV: instrumental variable; IVW: inverse-variance weighted; MR: Mendelian randomization; NAP: non-albumin protein; OR: odds ratio; p_fdr: false discovery rate-adjusted p value; SHBG: sex hormone-binding globulin; TP: total protein; TRIG: triglycerides; URK: urinary potassium.

Table 5.

MR estimates for breast, thyroid, and bladder cancer biomarkers using the IVW method.

Outcome			IVW			MR-EGGER		Weighted median
Breast cancer	Exposure	Number of IVs	OR (95% CI)	p	p_fdr	OR (95% CI)	p	OR (95% CI)	p
	HDL	238	1.084（1.040–1.129）	<0.001	0.004	1.100（1.031–1.174）	0.004	1.055（1.002–1.110）	0.040
	IGF-1	298	1.062（1.015–1.110）	0.009	0.078	1.068（0.974–1.170）	0.161	1.033（0.974–1.095）	0.280
	TRIG	208	0.944（0.904–0.986）	0.009	0.078	0.910（0.852–0.971）	0.005	0.921（0.876–0.969）	0.001
	URK	8	1.511（1.094–2.085）	0012	0.078	3.713（0.475–29.035）	0.258	1.376（0.928–2.041）	0.112
	ApoA	214	1.056（1.012–1.103）	0.013	0.078	1.081（1.008–1.159）	0.030	1.031（0.978–1.087）	0.253
	ALP	274	0.960（0.925–0.995）	0.027	0.135	0.919（0.867–0.974）	0.005	0.949（0.907–0.994）	0.026
	CHOL	234	1.040（1.002–1.080）	0.040	0.175	0.999（0.942–1.060）	0.982	1.060（1.008–1.114）	0.022
Thyroid cancer
	BUN	171	0.691（0.517–0.924）	0.013	0.386	0.563（0.286–1.107）	0.097	0.635（0.410–0.984）	0.042
	HDL	256	0.820（0.691–0.972）	0.022	0.386	0.945（0.722–1.235）	0.677	1.019（0.768–1.351）	0.898
	UA	232	1.248（1.016–1.532）	0.035	0.386	1.070（0.768–1.490）	0.691	1.294（0.934–1.792）	0.122
Bladder cancer
	BUN	144	1.001（1.000–1.002）	0.038	0.993	1.002（1.000–1.004）	0.101	1.001（1.000–1.003）	0.096

ALP: alkaline phosphatase; ApoA: apolipoprotein A; BUN: blood urea nitrogen; CHOL: total cholesterol; CI: confidence interval; HDL: high-density lipoprotein cholesterol; IGF-1: insulin-like growth factor 1; IV: instrumental variable; IVW: inverse-variance weighted; MR: Mendelian randomization; OR: odds ratio; p_fdr: false discovery rate-adjusted p value; TRIG: triglycerides; UA, uric acid; URK: urinary potassium.

After applying a FDR correction to account for multiple testing, nine associations remained statistically significant (p_fdr < 0.05), highlighting particularly robust findings: Colorectal cancer. Nonalbumin protein, total protein (TP), and uric acid were associated with a decreased risk, whereas insulin-like growth factor 1 (IGF-1) was associated with an increased risk. Gastric cancer. Apolipoprotein B (ApoB) and low-density lipoprotein (LDL) cholesterol were associated with a decreased risk. Liver cancer. Gamma-glutamyl transferase (GGT) was associated with an increased risk. Breast cancer. High-density lipoprotein cholesterol (HDL) cholesterol was associated with an increased risk. Esophageal cancer. Total cholesterol was associated with a decreased risk. We note that some nominally significant associations had ORs very close to 1 (e.g. C-reactive protein (CRP), aspartate aminotransferase (AST), and TP for cervical cancer, with ORs of 1.001). Although these associations reached statistical significance, their effect sizes were so small that they are unlikely to have clinical or biological relevance. Therefore, we focused our interpretation on associations with larger effect sizes and those that remained significant after FDR correction.

Sensitivity analyses and assessment of reverse causality

Sensitivity analyses confirmed the robustness of these findings. MR-Egger regression intercept tests and MR-PRESSO global tests found no significant evidence of horizontal pleiotropy for the identified associations (all p > 0.05). Leave-one-out sensitivity analyses demonstrated that no single SNP drove the causal estimates. To evaluate the possibility of reverse causation, we performed bidirectional MR analyses. With the exception of a potential reverse causal effect of breast cancer on CRP levels (OR = 1.013, 95% confidence interval (CI): 1.001–1.025, p = 0.034), we found no evidence that genetic susceptibility to cancer influenced biomarker levels.

Independent causal effects assessed by multivariable MR

To determine whether the identified biomarkers exerted direct effects independent of other related metabolites, we conducted MVMR analyses. After adjusting for other significant biomarkers within each cancer type, 12 biomarkers retained significant independent causal effects (MVMR-IVW p < 0.05). Key findings include (Table 6):

Table 6.

Multivariable MR estimates for independent causal effects of biomarkers on cancer risk.

Outcome			MVMR-IVW		MVMR-EGGER		MVMR-median
Hepatic cancer	Exposure	Number of IVs	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p
	CHOL	412	4.713（1.407–15.784）	0.012	4.627（1.375–15.565）	0.013	5.633（1.545–20.533）	0.009
	GGT		1.257（1.051–1.504）	0.012	1.276（1.065–1.529）	0.008	1.219（0.942–1.577）	0.132
	HDL		0.593（0.414–0.848）	0.043	0.595（0.415–0.853）	0.005	0.527（0.351–0.791）	0.002
	LDL-C		0.175（0.053–0.585）	0.046	0.178（0.053–0.594）	0.005	0.137（0.038–0.493）	0.002
Cervical cancer	CRP	262	1.0013（1.0002–1.0024）	0.023	1.0013（1.0002–1.0025）	0.020	1.0015（0.999–1.003）	0.063
Thyroid cancer	BUN	366	0.635（0.456–0.883）	0.007	0.588（0.382–0.905）	0.016	0.632（0.391–1.021）	0.061
Malignant neoplasm of ovary	BILD	313	1.361（1.045–1.771）	0.022	1.415（1.071–1.869）	0.014	1.239（0.828–1.854）	0.298
Colorectal cancer	IGF-1	509	1.151（1.054–1.256）	0.002	1.166（1.039–1.309）	0.009	1.184（1.052–1.334）	0.005
	UA		0.903（0.822–0.991）	0.031	0.903（0.822–0.991）	0.0317	0.903（0.797–1.024）	0.111
Breast cancer	CRP	474	1.107（1.031–1.189）	0.005	1.099（1.022–1.182）	0.011	1.075（0.996–1.159）	0.063
	HDL		1.240（1.008–1.527）	0.042	1.240（1.007–1.526）	0.043	1.147（0.948–1.389）	0.159
	IGF-1		1.067（1.009–1.129）	0.023	1.065（1.007–1.127）	0.027	1.065（1.005–1.129）	0.033
Gastric cancer	ApoB	371	0.531（0.310–0.911）	0.022	0.534（0.311–0.919）	0.024	0.821（0.446–1.511）	0.526
	LDL-C		2.398（1.053–5.457）	0.037	2.403（1.055–5.477）	0.037	1.006（0.409–2.473）	0.990
	SHBG		1.115（1.005–1.236）	0.040	1.134（1.004–1.235）	0.042	1.124（0.975–1.296）	0. 108
Endometrial cancer	SHBG	297	0.784（0.679–0.905）	0.001	0.778（0.670–0.903）	<0.001	0.774（0.633–0.946）	0.108

ApoB: apolipoprotein B; BILD: direct bilirubin; BUN: blood urea nitrogen; CHOL: total cholesterol; CI: confidence interval; CRP: C-reactive protein; GGT: gamma-glutamyl transferase; HDL: high-density lipoprotein cholesterol; IGF-1: insulin-like growth factor 1; IV: instrumental variable; LDL-C: low-density lipoprotein cholesterol; MR: Mendelian randomization; MVMR-EGGER: multivariable Mendelian randomization-Egger; MVMR-IVW: multivariable Mendelian randomization inverse-variance weighted; MVMR-median: multivariable Mendelian randomization weighted median; OR: odds ratio; SHBG: sex hormone-binding globulin; UA: uric acid.

Four biomarkers (cholesterol (CHOL), GGT, HDL-C, and low density lipoprotein cholesterol (LDL-C)) were independently associated with liver cancer risk. Three biomarkers (CRP, HDL-C, and IGF-1) were independently associated with breast cancer risk. Three biomarkers (ApoB, LDL-C, and sex hormone-binding globulin (SHBG)) were independently associated with gastric cancer risk. Two biomarkers (IGF-1 and uric acid (UA)) were independently associated with colorectal cancer risk.

Single biomarkers showed independent effects for thyroid cancer (blood urea nitrogen (BUN)), cervical cancer (CRP), ovarian cancer (direct bilirubin), and endometrial cancer (SHBG). The independent causal effects confirmed by MVMR indicate that these biomarkers are not merely surrogates for other metabolic factors. From a clinical perspective, markers such as GGT (liver cancer) and IGF-1 (colorectal and breast cancer) could potentially refine risk prediction models, although further validation in prospective cohorts is needed before clinical application.

Discussion

In this comprehensive bidirectional and multivariable MR study, we systematically investigated the causal relationships between 35 circulating biomarkers and 13 major cancers. Moving beyond mere association, our study design robustly addressed confounding and reverse causation, with MVMR analysis helping to identify biomarkers with independent effects. We identified 49 potential causal associations, of which 9 withstood stringent FDR correction. Crucially, 12 biomarkers were identified as having direct, independent causal effects on liver, breast, gastric, and colorectal cancers through MVMR. Our findings provide insight into the complex and tissue-specific roles of metabolic, inflammatory, and hormonal pathways in cancer etiology.

Tissue-specific paradox of lipid metabolism

Our results challenge the conventional, one-dimensional view of lipids in cancer risk, revealing a compelling tissue-specific paradox. Counterintuitively, genetically predicted higher levels of ApoB and LDL-C were associated with a reduced risk of gastric cancer. This inverse association could be explained, at least in part, by reverse causality, as advanced gastric cancer is often accompanied by cachexia and malnutrition, which lead to hypolipidemia.^7,8 However, we acknowledge that this interpretation remains speculative in the absence of longitudinal clinical data. Prospective studies with detailed nutritional and cancer staging information are needed to clarify the direction of causality. Similarly, CHOL, LDL-C, and HDL-C appeared to be protective against liver cancer. A previous MR study supports a causal relationship between higher LDL-C levels and a lower risk of hepatocellular carcinoma (HCC),⁹ whereas epidemiological evidence suggests a complex, nonlinear relationship for HDL.¹⁰ In stark contrast, HDL-C was a significant risk factor for breast cancer.¹¹ This finding aligns with the “obesity paradox” in metabolism and may reflect tumor-specific lipid reprogramming, whereby cholesterol is shunted to support membrane synthesis in rapidly proliferating cancer cells.¹² This dichotomy underscores that the role of a biomarker is highly context-dependent and shaped by the unique metabolic environment of each organ.

Systemic inflammation, liver function, and nutritional status

Biomarkers reflecting systemic inflammation and organ function constituted another major axis of causality. CRP, a key inflammatory marker, was independently associated with an increased risk of cervical and breast cancers. This is consistent with clinical studies linking elevated CRP levels to a poor prognosis in patients with cervical cancer¹³ and with its role in fostering a pro-tumorigenic microenvironment.¹⁴ The bidirectional MR finding suggests that breast cancer elevates CRP levels, raising the possibility of a vicious cycle in which the tumor both creates and thrives in an inflammatory milieu. However, this hypothesis requires experimental validation. Furthermore, GGT, a marker of liver function and oxidative stress, was a strong independent risk factor for liver cancer, corroborating its well-established role as a diagnostic and prognostic biomarker, often used alongside alpha-fetoprotein (AFP).¹⁵ Conversely, low levels of proteins such as albumin (ALB) and TP, indicative of poor systemic nutritional status and cachexia, were associated with an increased risk of lung and esophageal cancers,^16–18 reinforcing their established value as prognostic indicators across oncology.

Hormonal and growth factor signaling pathways

Our study provides supportive genetic evidence for the causal involvement of specific hormonal and growth factor pathways in site-specific cancers. IGF-1 emerged as a potentially important factor in colorectal and breast cancer risk, consistent with its fundamental role in promoting cellular proliferation and inhibiting apoptosis. This is highly plausible given its fundamental role in promoting cellular proliferation and inhibiting apoptosis. Our finding is strongly supported by studies showing elevated IGF-1 expression in colorectal cancer tissues¹⁹ and the established role of the insulin/IGF system in cancer development and therapy resistance.²⁰ The role of sex hormones was highlighted by SHBG, which exhibited contrasting effects: it was a protective factor for endometrial cancer²⁰ but a risk factor for gastric cancer.²¹ This suggests complex, tissue-specific interactions among sex hormone signaling, SHBG, and carcinogenesis, potentially mediated by differences in receptor expression or local hormone metabolism.

Comparison with previous MR studies

Our results are consistent with previous MR findings for GGT and liver cancer¹⁵ and for LDL-C and gastric cancer.⁸ However, the association between HDL-C and breast cancer differs from some earlier reports,¹¹ possibly because of differences in study populations or adjustment for confounders. Unlike previous single-exposure MR studies, our work provides a systematic evaluation of 35 biomarkers across 13 cancers and uses MVMR to identify independent effects.

Potential clinical implications

Although our findings are based on genetic analyses and require further validation, they suggest several potential clinical applications. First, biomarkers with robust independent causal effects, such as GGT for liver cancer and IGF-1 for colorectal and breast cancer, could be considered for inclusion in risk prediction models, particularly for high-risk populations. Second, the tissue-specific effects of lipids (e.g. HDL-C increasing breast cancer risk but decreasing liver cancer risk) highlight that risk assessment should be cancer site–specific. Third, our results may help prioritize biomarkers for future interventional studies or for repurposing existing drugs (e.g. lipid-lowering agents). However, we emphasize that these are hypothesis-generating observations; prospective clinical cohorts and randomized trials are needed before clinical application.

Strengths and limitations

The primary strengths of our study include the use of large-scale, high-quality GWAS data, the application of bidirectional MR to mitigate reverse causation, and, most importantly, the implementation of MVMR to disentangle independent causal effects—a significant methodological advance that provides greater clarity than previous univariable MR studies.

However, several limitations merit consideration. First, the predominant use of European-ancestry data limits the generalizability of our findings to other populations, where genetic architectures and environmental exposures may differ. Second, although we employed rigorous sensitivity analyses to detect pleiotropy, the possibility of residual horizontal pleiotropy cannot be entirely ruled out. Third, MR estimates reflect the lifelong effects of genetic predisposition, which may not fully mirror the effects of modifying these biomarkers during adulthood through interventions. Finally, the lack of granular data on cancer subtypes, stage, and molecular characteristics in the GWAS summary statistics prevented us from exploring heterogeneity within cancer types, representing an important avenue for future research. Additionally, although we used F statistics >10 to exclude weak instruments, residual weak-instrument bias cannot be completely ruled out, particularly for exposures with a limited number of genetic variants. Furthermore, the cancer GWAS summary statistics were obtained from different consortia (FinnGen and IEU Open GWAS), which may differ in phenotype definitions, genotyping platforms, and covariate adjustments, potentially introducing heterogeneity across cancer types. This is an in silico study based on publicly available GWAS summary data. Therefore, the findings should be considered hypothesis-generating and require validation in independent cohorts, experimental models, and clinical trials.

Conclusion

In conclusion, this large-scale MR study provides supportive evidence for causal relationships between several circulating biomarkers and site-specific cancer risks, with distinct tissue-specific patterns. Biomarkers showing independent effects (e.g. GGT for liver cancer and IGF-1 for colorectal and breast cancer) warrant further investigation. Future prospective cohort studies and functional studies are needed to validate these findings and explore their potential clinical applications, such as risk stratification and early detection.

Supplemental Material

sj-csv-1-imr-10.1177_03000605261466576 - Supplemental material for Causal effects of blood and urinary biomarkers on cancer risk: A Mendelian randomization study

Supplemental material, sj-csv-1-imr-10.1177_03000605261466576 for Causal effects of blood and urinary biomarkers on cancer risk: A Mendelian randomization study by Lina Leng and Ying Li in Journal of International Medical Research

Footnotes

Acknowledgments

We would like to acknowledge the participants and investigators of the FinnGen and UK Biobank studies.

ORCID iD

Lina Leng

Ethics approval and consent to participate

In this study, we utilized extensive GWAS summary data from original studies, where all participants provided informed consent. Given our reliance solely on aggregate statistical data, no further ethical clearance is required.

Consent for publication

Not applicable.

Author contributions

The authors listed in our statement were all involved in this study. LNL and YL provided guidance and guidance in the manuscript, and Y Land LN L wrote the manuscript. YL and LN L carried out chart making and literature searching. They are the drafters and revisers of the final paper. At the end of the study, all authors read and approved the manuscript.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declare no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Availability of data and materials

All data used in this study are publicly available. Summary statistics for the 35 serum and urinary biomarkers were obtained from the UK Biobank (GWAS ID: ukb-bub35), as described by Sinnott-Armstrong et al.⁶ Cancer GWAS summary data were obtained from the FinnGen consortium and IEU Open GWAS using the following identifiers:

Colorectal cancer: ebi-a-GCST90018808

Gastric cancer: ebi-a-GCST90018849

Lung cancer: ebi-a-GCST90018875

Esophageal cancer: ebi-a-GCST90018841

Liver cancer: ebi-a-GCST90018858

Cervical cancer: ukb-b-8777

Thyroid cancer: ebi-a-GCST90018929

Pancreatic cancer: ebi-a-GCST90018893

Endometrial cancer: ebi-a-GCST90018838

Breast cancer: ieu-a-1126

Malignant neoplasm of ovary: finn-b-C3_OVARY_EXALLC

Melanoma skin cancer: ieu-b-4969

Bladder cancer: ieu-b-4874

Publicly available datasets were analyzed in this study. No additional permission was required.

Supplemental material

Supplemental material for this article is available online.

References

Siegel

Giaquinto

Jemal

. Cancer statistics, 2024. CA Cancer J Clin 2024; 74: 12–49.

Maomao

Dianqin

, et al. Current cancer burden in China: epidemiology, etiology, and prevention. Cancer Biol Med 2022; 19: 1121–1138.

Zhao

Wang

, et al. Identifying causal relationships between 35 blood and urine biomarkers and urologic cancers: mR-meta combined with Bayesian colocalization Mendelian randomization analysis. Discov Oncol 2024; 15: 617.

Yun

Guo

, et al. Genetically predicted 486 blood metabolites in relation to risk of colorectal cancer: a Mendelian randomization study. Cancer Med 2023; 12: 13784–13799.

Zhao

Luo

, et al. The association between circulating docosahexaenoic acid and lung cancer: a Mendelian randomization study. Clin Nutr 2022; 41: 2529–2536.

Sinnott-Armstrong

Tanigawa

Amar

, et al. Genetics of 35 blood and urine biomarkers in the UK Biobank. Nat Genet 2021; 53: 185–194.

Soffer

Marston

Maki

, et al. Role of apolipoprotein B in the clinical management of cardiovascular risk in adults: an expert clinical consensus from the national lipid association. J Clin Lipidol 2024; 18: e647–e663.

Kim

Cho

, et al. Association between total cholesterol levels and all-cause mortality among newly diagnosed patients with cancer. Sci Rep 2024; 14: 58.

Cao

Wang

Zhu

, et al. Low-density lipoprotein cholesterol and risk of hepatocellular carcinoma: a Mendelian randomization and mediation analysis. Lipids Health Dis 2023; 22: 110.

10.

Zhang

Song

, et al. Association between blood lipid levels and the risk of liver cancer: a systematic review and meta-analysis. Cancer Causes Control 2024; 35: 943–953.

11.

Pavlova

Zhu

Thompson

. The hallmarks of cancer metabolism: still emerging. Cell Metab 2022; 34: 355–377.

12.

Feingold

. Introduction to lipids and lipoproteins. In: Feingold KR, Adler RA, Ahmed SF, et al. (eds) Endotext [Internet]. South Dartmouth: MDText.com, Inc., 2024.

13.

Wang

Zhou

, et al. Prognostic values of LDH and CRP in cervical cancer. Onco Targets Ther 2020; 13: 1255–1263.

14.

Liu

Zhang

Liu

, et al. Immune and inflammation: related factor alterations as biomarkers for predicting prognosis and responsiveness to PD-1 monoclonal antibodies in cervical cancer. Discov Oncol 2022; 13: 96.

15.

Liu

Luo

Chen

, et al. Diagnostic value of 5 serum biomarkers for hepatocellular carcinoma with different epidemiological backgrounds: a large-scale, retrospective study. Cancer Biol Med 2021; 18: 256–270.

16.

Liu

Zhang

Ruan

, et al. One-year mortality in patients with cancer cachexia: association with albumin and total protein. Cancer Manag Res 2021; 13: 6775–6783.

17.

Scott

McMillan

Forrest

, et al. The systemic inflammatory response, weight loss, performance status and survival in patients with inoperable non-small cell lung cancer. Brit J Cancer 2002; 87: 264–267.

18.

Zhang

, et al. Relationships between serum albumin and urea level and the clinical pathological characteristics and survival time in patients with lung cancer. Zhongguo Fei Ai Za Zhi. 2017; 20; 175–186.

19.

Vigneri

Tirrò

Pennisi

, et al. The insulin/IGF system in colorectal cancer development and resistance to therapy. Front Oncol 2015; 5: 230.

20.

Mullee

Dimou

Allen

, et al. Testosterone, sex hormone-binding globulin, insulin-like growth factor-1 and endometrial cancer risk: observational and Mendelian randomization analyses. Br J Cancer 2021; 125: 1308–1317.

21.

Liu

Zhang

Lagergren

, et al. Circulating sex hormone levels and risk of gastrointestinal cancer: systematic review and meta-analysis of prospective studies. Cancer Epidemiol Biomarkers Prev 2023; 32: 936–946.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

3.25 MB