Pre-diagnostic metabolite concentrations and prostate cancer risk in 1077 cases and 1077 matched controls in the European Prospective Investigation into Cancer and Nutrition

Background Little is known about how pre-diagnostic metabolites in blood relate to risk of prostate cancer. We aimed to investigate the prospective association between plasma metabolite concentrations and risk of prostate cancer overall, and by time to diagnosis and tumour characteristics, and risk of death from prostate cancer. Methods In a case-control study nested in the European Prospective Investigation into Cancer and Nutrition, pre-diagnostic plasma concentrations of 122 metabolites (including acylcarnitines, amino acids, biogenic amines, glycerophospholipids, hexose and sphingolipids) were measured using targeted mass spectrometry (AbsoluteIDQ p180 Kit) and compared between 1077 prostate cancer cases and 1077 matched controls. Risk of prostate cancer associated with metabolite concentrations was estimated by multi-variable conditional logistic regression, and multiple testing was accounted for by using a false discovery rate controlling procedure. Results Seven metabolite concentrations, i.e. acylcarnitine C18:1, amino acids citrulline and trans-4-hydroxyproline, glycerophospholipids PC aa C28:1, PC ae C30:0 and PC ae C30:2, and sphingolipid SM (OH) C14:1, were associated with prostate cancer (p < 0.05), but none of the associations were statistically significant after controlling for multiple testing. Citrulline was associated with a decreased risk of prostate cancer (odds ratio (OR1SD) = 0.73; 95% confidence interval (CI) 0.62–0.86; p trend = 0.0002) in the first 5 years of follow-up after taking multiple testing into account, but not after longer follow-up; results for other metabolites did not vary by time to diagnosis. After controlling for multiple testing, 12 glycerophospholipids were inversely associated with advanced stage disease, with risk reduction up to 46% per standard deviation increase in concentration (OR1SD = 0.54; 95% CI 0.40–0.72; p trend = 0.00004 for PC aa C40:3). Death from prostate cancer was associated with higher concentrations of acylcarnitine C3, amino acids methionine and trans-4-hydroxyproline, biogenic amine ADMA, hexose and sphingolipid SM (OH) C14:1 and lower concentration of glycerophospholipid PC aa C42:4. Conclusions Several metabolites, i.e. C18:1, citrulline, trans-4-hydroxyproline, three glycerophospholipids and SM (OH) C14:1, might be related to prostate cancer. Analyses by time to diagnosis indicated that citrulline may be a marker of subclinical prostate cancer, while other metabolites might be related to aetiology. Several glycerophospholipids were inversely related to advanced stage disease. More prospective data are needed to confirm these associations. Electronic supplementary material The online version of this article (doi:10.1186/s12916-017-0885-6) contains supplementary material, which is available to authorized users.

Medicine, Lund University, Skåne University Hospital, Malmö, Sweden, for a previous study [2]. Intraand inter-assay coefficients of variation were less than 9%. PSA concentration was available for 71.1% of men in the current study, including 764 controls, 489 of which had a concentration below 1 ng/ml, and for 768 cases.

Metabolite measurements outside the measurable range
After excluding 18 metabolites for which more than 15% of participants had measurements outside the measurable range (Additional file 3: Table S1), the remaining measurements outside the measurable range were imputed. Measurements below the limit of detection (applicable to 12 metabolites for 1 to 307 men) and quantification (applicable to 7 metabolites for 1 to 296 men) were set to half the lowest measured concentration and to half the limit of quantification, respectively. Measurements above the highest concentration calibration standards were set to the highest standard concentration (applicable to 1 metabolite for 1 man).

Coefficients of variation for metabolite concentrations
Overall coefficients of variations were calculated as the standard deviation divided by the mean (Additional file 3: Table S1). For the 122 included metabolites, the median (range) was 12.3% (7.0-17.2) for acylcarnitines, 8.8% (6.0-12.6) for amino acids, 10.6% (4.3-17.2) for biogenic amines, 11.2% (7.3-Julie A. Schmidt et al. Pre-diagnostic metabolite concentrations and prostate cancer risk in 1077 cases and 1077 matched controls in the European Prospective Investigation of Cancer and Nutrition. BMC Medicine 2 17.7) for glycerophospholipids and 10.4% (8.0-19.9) for sphingolipids, and the coefficient of variation for hexose was 6.5%.

Nomenclature of metabolites
Fatty acid side chains in acylcarnitines, glycerophospholipids and sphingolipids were labelled "Cx:y," where x and y denote the total number of carbon atoms and double bonds, respectively, in each molecule [3]. Acylcarnitines were abbreviated according to the fatty acid side chain. All glycerophospholipids were phosphatidylcholines, and subclasses were separated by the number and type of fatty acids side chains. "LysoPC a" denotes phosphatidylcholines with one acyl fatty acid side chain, "PC aa" denotes two acyl side chains, and "PC ae" denotes one acyl and one alkyl side chain.
Sphingolipids were all sphingomyelins with a hydroxyl group (SM (OH)) or without a hydroxy group attached and were also labelled according to the fatty acid side chain. Hexose is the sum of a range of monosaccharides with six carbon atoms, including glucose, fructose and galactose.

Conditional logistic regression by fifths of metabolite concentrations
Conditional logistic regression was used to estimate risk of prostate cancer by fifths of metabolite concentrations (based on the distribution among controls). Tests for linear trend were computed across the median concentrations in the fifths. Like in the main model presented in the paper, this model was conditioned on the matching variables and further adjusted for exact age (continuously), body mass index (fourths; unknown), smoking (never; past; current; unknown), alcohol intake (<10; 10-19; 20-39; ≥40 g of alcohol per day; unknown), education (primary; secondary; degree level; unknown) and marital status (married or cohabiting; not married or cohabiting; unknown).

Test for heterogeneity
Tests for heterogeneity in the associations between metabolite concentrations and prostate cancer risk by subgroups (i.e. time to diagnosis and tumour characteristics) were done using the likelihood ratio χ 2 test, which compared models with and without an interaction term between the linear trend variables and the outcome variable of interest.

Multiple testing
The Benjamini-Hochberg false discovery rate controlling procedure was used to account for multiple testing in all analyses of metabolite concentrations.
First, the p-values were sorted and ranked from the lowest p(1) to the highest p(m). Let k be the largest rank (i) for which p(i) < (i/m) × α is true; α is the significance level and set to 0.05 in all analyses. Then all the null hypotheses for p-values from p(1) to p(k) were rejected [4]. This method allows 5% of positive findings to be false, on average [5].
Additionally, an adjusted p-value (p adj sometimes referred to as a q-value by other authors) was computed. p adj (i) was defined as the minimum of p(n) x (m/n) for n being i, i+1, …, m. tests with a p adj < 0.05 were declared significant after controlling the FDR at 5% [6].