Background

Human studies of the metabolic basis of prostate cancer (PCa) have yet to determine which metabolic changes associated with PCa are a cause or a consequence of tumour development and progression.

Methods

Observational analyses

We performed a nested case-control study of nuclear magnetic resonance (NMR)-measured metabolites (log-transformed and scaled to z-scores) and PCa occurence in a cross-sectional sample of 5,040 localized PCa cases (2335) and controls (2705) and an exploration of the effect of PCa on metabolites in the UK-based, National Institute for Health Research–supported Prostate testing for cancer and Treatment (ProtecT) trial. Statistical models were adjusted for age and family history of prostate cancer and non-uniform variance accounted for with the use of robust standard errors, clustering on study center.

Bi-directional, two-sample Mendelian randomization

Then, identifying SNPs previously associated with metabolites (metab-QTLs) from Kettunen et al. (2016), we pulled the summary data for the association of metab-QTLS and PCa incidence in the Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome (PRACTICAL) consortium (44,825 PCa cases 27,904 controls). Next, extracting SNPs associated with PCa from the publicly available GWAS Catalog and using the summary data for the association of these SNPs with metabolites, we performed bi-directional, two-sample Mendelian randomization (MR), interrogating the causality of the metabolome on PCa incidence and the causality of PCa incidence on the metabolome.

Results

Table 1 (ProtecT)

Selected baseline characteristics (medians and interquartile ranges, or percents) for cases and controls

Characteristic Case (n=2335) Control (n=2705) p value*
Age 62.84 (58.69-66.59) 62.79 (58.56-66.40) 0.684
Family history of prostate cancer (%)** 174 (7) 129 (5) <0.001
BMI# 26.50 (24.45-28.85) 26.58 (24.38-28.73) 0.7387

* p-value based on Chi-squared tests (for categorical variables) and Wilcoxon rank sum tests (for continuous variables)
** Family history data available on only 89% of these subjects
# BMI data available on only 74% of these subjects

Volcano plots

Volcano plots of metabolites on prostate cancer (left) and prostate cancer on metabolites (right). Labeled metabolites are Bonferroni signficiant.

Volcano plots of metabolites on prostate cancer (left) and prostate cancer on metabolites (right). Labeled metabolites are Bonferroni signficiant.

Bi-directional MR results of metabolites and prostate cancer

Metabolites are labeled if the p-values for the MR are <0.05 or if the p-values for the observational association are Bonferroni significant (p<0.000219282). Cyan coloring indicates a discordance between the observational and MR estimates, where discordance is defined as opposite direction of effect: i.e, for the comparison of odds ratios (above Left, Metabolites on Prostate Cancer), if the effect estimate for the observational association is <1 and the effect estimate for the causal association is >1, or if the effect estimate for the observation association is >1 and the effect estimate for the causal association is <1, then cyan (discordant); and, for the comparison of linear regressions (above Right, Prostate Cancer on Metabolites), if the effect estimate for the obervational association is <0 and the effect estimate for the causal association is >0 or if the effect estimate for the observation association is >0 and the effect estimate for the causal association is <0, then cyan (discordant).

Metabolites are labeled if the p-values for the MR are <0.05 or if the p-values for the observational association are Bonferroni significant (p<0.000219282). Cyan coloring indicates a discordance between the observational and MR estimates, where discordance is defined as opposite direction of effect: i.e, for the comparison of odds ratios (above Left, Metabolites on Prostate Cancer), if the effect estimate for the observational association is <1 and the effect estimate for the causal association is >1, or if the effect estimate for the observation association is >1 and the effect estimate for the causal association is <1, then cyan (discordant); and, for the comparison of linear regressions (above Right, Prostate Cancer on Metabolites), if the effect estimate for the obervational association is <0 and the effect estimate for the causal association is >0 or if the effect estimate for the observation association is >0 and the effect estimate for the causal association is <0, then cyan (discordant).

ProtecT logistic regression results

ProtecT linear regression results

MR results for metabolites on PCa

The top finding is for creatinine. Some evidence of heterogenity, but the SNP that appears to be driving the association is located nearest a gene (SLC22A2) linked to creatinine production and secretion [2]. Serum creatinine has been previously observed to increase risk of prostate cancer in a prospective study (OR=2.23, 95% CI 1.33–3.75) [3]. Weinstein et al. (2009) note that homocysteine is produced during the synthesis of creatine (creatinine’s precursor) and proffer that levels of circulating creatinine might mark homocysteine status and one-carbon metabolism, where a reduction in the availability of one-carbon groups possibly impacts DNA methylation, synthesis, and repair [3].

Results for the Mendelian randomization of metabolites on prostate cancer incidence (p<0.05)

Metabolite Estimate SE Pval Method
98 Crea 1.410868 0.0825445 0.0000305 Maximum likelihood
75 Gp 1.167327 0.0567604 0.0064149 Maximum likelihood
4 Pyr 1.292758 0.1165183 0.0275415 Maximum likelihood
43 Ala 1.198443 0.0837828 0.0307243 Maximum likelihood
85 Lac 1.379156 0.1520822 0.0345320 Maximum likelihood
109 FAw79S 1.119108 0.0539872 0.0371225 Maximum likelihood

Complete list of MR results for metabolites on prostate cancer incidence

References

  1. Kettunen, J. et al. Genome-wide study for circulating metabolites identifies 62 loci and reveals novel systemic effects of LPA. Nat Commun 7, 1–9 (2016).

  2. Köttgen, A. et al. Multiple loci associated with indices of renal function and chronic kidney disease. Nat. Genet. 42, 376–384 (2010).

  3. Weinstein, S. J. et al. Serum creatinine and prostate cancer risk in a prospective study. Cancer Epidemiol Biomarkers Prev. 18, 2630–2649 (2009).