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

EPIC is a multi-centre cohort study comprising 520,000 men and women from ten European countries recruited between 1992 and 2000 [12]. It was designed to investigate how diet (intake and biomarkers) is associated with risk of cancer and other diseases. Among other important findings, it has helped establish insulin-like growth factor I as a risk factor for prostate cancer [2, 13].

The 153,400 men in the cohort were recruited from 19 centres in eight countries (Denmark, Germany, Greece, Italy, the Netherlands, Spain, Sweden and the UK). At recruitment, detailed information was collected on dietary intake, lifestyle, anthropometry and previous disease, and 139,600 men also gave a blood sample [12].

All participants gave written informed consent to participate in the EPIC cohort, and the EPIC study protocol was approved by the ethical committees of the International Agency for Research on Cancer (IARC), Lyon, France, and the participating centres. Approval for the current study was obtained from the Internal Review Board of the IARC (Project No. 14-09) and local ethics committees (see Declarations).

For the current analysis, men were eligible if they had blood stored at the central biobank at the IARC (centres in Germany, Greece, Italy, the Netherlands, Spain and the UK), the date of blood collection was known and if no cancer (except non-melanoma skin cancer) had been diagnosed at the time of blood collection.

A standardised protocol for blood collection and processing was followed, and fasting was not required; details are published elsewhere [12]. All plasma samples (citrate anticoagulant) were assayed at the IARC, using the AbsoluteIDQ p180 Kit (Biocrates Life Sciences AG, Innsbruck, Austria) and following the procedure recommended by the vendor. A triple quadrupole mass spectrometer (Triple Quad 4500; AB Sciex, Framingham, MA, USA) was used to quantify a total of 142 metabolites. Samples from matched case-control sets were assayed in the same analytical batch, each of which included six to eleven quality control samples of pooled plasma. Laboratory personnel were blinded to sample category, i.e. case, study control or quality control.

The concentration of total prostate-specific antigen (PSA) at baseline was measured for a previous study [14] (Additional file 1) and was available for 71.1% of men in the current study, including 764 controls, for whom 489 had a concentration below 1 ng/ml, and 768 cases.

Metabolite data were available for 2169 men (Additional file 2: Figure S1A). Metabolites were excluded if more than 15% of men had non-quantifiable assay results (missing data or results outside the measurable range; n = 18) or if the overall coefficient of variation (CV) was higher than 20% (n = 2; Additional file 1; Additional file 2: Figure S1B; Additional file 3: Table S1 shows the completeness of assay results and CVs). This left 122 metabolites for the analysis (7 acylcarnitines, 21 amino acids, 6 biogenic amines, 75 glycerophospholipids (all of which were phosphatidylcholines, the most abundant phospholipid in humans [15], denoted lysoPC or PC; for metabolite nomenclature see Additional file 1), hexose and 12 sphingolipids (all of which were sphingomyelins and denoted SM). Men with missing information on any of the 122 metabolites (n = 1) and men in incomplete case-control sets were excluded (n = 14), leaving 1077 matched case-control sets in the statistical analysis.

Logarithmically transformed metabolite concentrations were used for all analyses.

Partial correlations were calculated between log-transformed concentrations of total PSA and metabolites separately in controls, controls with total PSA <1 ng/ml and cases, adjusting for age at blood collection (<55, 55–59, 60–64, 65–69, ≥70 years), body mass index (BMI; fourths, unknown) and study centre.

We used conditional logistic regression to estimate risk of prostate cancer per standard deviation (SD) increase in metabolite concentrations. Tests for linear trend were computed for metabolite concentrations as continuous variables. Departure from linearity was tested using the likelihood ratio χ² test comparing models with the metabolite concentration as a linear term and as a cubic polynomial, respectively. The analysis was conditioned on the matching variables and further adjusted for exact age (continuously) in one model, and additionally for BMI (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) in a second model. Results from the two models did not materially differ, and only results from the latter model are presented. A model based on fifths of metabolite concentrations was also computed (Additional file 1).

Similar conditional logistic regression models were fitted for subgroups by time to diagnosis (≤5 vs. >5 years) and tumour characteristics (low-intermediate vs. high grade, localised vs. advanced stage and non-aggressive vs. aggressive disease), and heterogeneity by subgroups was tested (Additional file 1).

In a sensitivity analysis, we excluded men who were also in a previous analysis conducted in EPIC-Heidelberg [10] (91 cases and 11 controls; personal communication, Tilman Kühn) and resulting in incomplete matched sets. This left 985 matched sets for this analysis.

All tests of statistical significance were two-sided, and to account for multiple testing the false discovery rate was controlled to 0.05 using the Benjamini-Hochberg method [16] (Additional file 1). All analyses were conducted in the Stata Statistical Software Package, version 14 (Stata Corporation, College Station, TX, USA).

The statistical significance of associations between metabolite concentrations and overall prostate cancer risk is shown in Fig. 1, and in Fig. 2 odds ratios (ORs) and 95% confidence intervals (CIs) are shown for metabolites with p < 0.1 (additional results in Additional file 3: Tables S4 and S5). Conventionally statistically significant (p < 0.05) lower risks were seen in men with higher concentrations of the acylcarnitine C18:1 and the amino acid citrulline, while higher risks were seen for the amino acid trans-4-hydroxyproline (t4-OH-Pro), glycerophospholipids PC diacyl (aa) C28:1, PC ae C30:0 and PC ae C30:2 and sphingolipid SM (OH) C14:1, but none of the associations were statistically significant after controlling for multiple testing. The strongest association was with PC ae C30:0, for which the risk was 16% higher per SD increase in concentration (OR_1SD = 1.16; 95% CI 1.04–1.30). There was no evidence of departure from linearity in the association with prostate cancer risk for these metabolites, except for C18:1 for which the test for non-linearity was conventionally significant, but not significant after controlling for multiple testing (Additional file 3: Table S4).

Of the cases, 428 (39.7%) were diagnosed within 5 years of blood collection. A one SD increase in citrulline concentration was associated with a 27% lower risk of prostate cancer diagnosed within 5 years of blood collection but not with prostate cancer diagnosed later after blood collection (OR_1SD = 0.73; 95% CI 0.62–0.86; p _trend = 0.0002, which was significant after controlling for multiple testing, and OR_1SD = 1.02; 95% CI 0.90–1.16, respectively; p _{heterogeneity} = 0.0009; Fig. 3; Additional file 3: Table S6).

Conventionally significant associations were seen in the first 5 years of follow-up, inversely for three acylcarnitines (including C18:1) and positively for two amino acids and a glycerophospholipid, and for follow-up beyond 5 years, for t4-OH-Pro, PC aa C28:1 and PC ae C30:0, although there was no evidence of heterogeneity by time to diagnosis.

Conventionally significant inverse associations with high grade prostate cancer were seen for an acylcarnitine, 24 glycerophospholipids and three sphingolipids, with risk reductions up to 48% (OR_1SD = 0.52, 95% CI 0.35–0.79 for PC aa C32:3; Fig. 4; Additional file 2: Figure S3; Additional file 3: Table S7); however, the associations did not remain after correcting for multiple testing.

After controlling for multiple testing, 12 glycerophospholipids, i.e. lysoPC a C18:0, PC aa C36:2, PC aa C36:3, PC aa C38:3, PC aa C38:5, PC aa C40:2, PC aa C40:3, PC aa C40:4, PC aa C40:5, PC aa C42:4, PC aa C42:5 and PC ae C40:1, were inversely associated with risk of advanced prostate cancer, with risk reductions up to 46% (OR_1SD = 0.54; 95% CI 0.40–0.72 for PC aa C40:3; Fig. 5; Additional file 2: Figure S4; Additional file 3: Table S8). For six of these and three additional glycerophospholipids, the test for heterogeneity by stage was significant after taking multiple testing into account, with inverse associations for advanced disease and no associations for localised disease.

Additionally, conventionally significant associations with advanced disease were seen for four acylcarnitines, two amino acids, a biogenic amine, 20 glycerophospholipids and two sphingolipids; all but the amino acid proline were inversely associated with advanced disease.

Conventionally significant associations were seen for 13 metabolites with aggressive prostate cancer. Positive associations were observed with an acylcarnitine, an amino acid, a glycerophospholipid, hexose and a sphingomyelin, while inverse associations were seen for seven glycerophospholipids and a sphingolipid, but none remained after controlling the false discovery rate (Fig. 6; Additional file 2: Figure S5; Additional file 3: Table S9). The strongest association was with glycerophospholipid lysoPC a C16:0, for which the estimated risk reduction was 48% (OR_1SD = 0.52, 95% CI 0.31–0.86).

During follow-up, 144 men died of prostate cancer, and after excluding matched sets in which the control had died before the case (n = 13) or vital status was unknown for the control (n = 4), 127 matched sets were available for analysis. Seven metabolites were conventionally significantly associated with death from prostate cancer. Men with higher concentrations of an acylcarnitine (C3), two amino acids (methionine and t4-OH-Pro), a biogenic amine (ADMA), hexose and a sphingolipid (SM (OH) C14:1) were at higher risk, while an inverse association was found for a glycerophospholipid (PC aa C42:4; Fig. 7; Additional file 2: Figure S6; Additional file 3: Table S10). However, the associations did not remain after controlling for multiple testing.

Results were not materially changed after excluding the overlap in participants between the EPIC-Heidelberg study [10] and our analysis (Additional file 3: Table S11).

In this prospective study of 122 plasma metabolite concentrations and prostate cancer risk, seven metabolites were associated with risk of overall prostate cancer at a conventional level of significance but not after correction for multiple testing; these were from several metabolite classes, suggesting that dysregulation of many metabolic pathways may be related to prostate cancer. The results stratified by time to diagnosis suggested that lower concentrations of citrulline might be a marker of subclinical prostate cancer, as the association was observed for disease diagnosed in the first few years after blood draw only. In contrast, the associations with others, including t4-OH-Pro, SM (OH) C14:1, PC ae C30:0 and other glycerophospholipids, did not vary by time to diagnosis, and these might thus provide insights into aetiology. Subgroup analysis indicated a possible link between higher concentrations of several glycerophospholipids and decreased risk of high risk tumour subtypes, especially advanced stage prostate cancer. Similar to the results for overall prostate cancer, suggested associations were observed between death from prostate cancer and seven metabolites from several metabolite classes.

Besides the current study, four smaller prospective studies of metabolomics and prostate cancer risk have been published; two were nested within the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (ATBC) [8, 9], one in the EPIC-Heidelberg study [10] and one in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) [11] (with 74, 200, 310 and 380 cases, respectively). Comparison of results from metabolomics studies is not straightforward owing to differences in population characteristics, biological medium [17, 18], the technological assay and statistical tools, including procedures for dealing with the large number of metabolites. Nonetheless, replication of possible associations both with specific metabolites and also more globally with metabolite classes is essential for interpretation of results.

In the two ATBC analyses, mass spectrometry was used to measure 420 and 626 metabolites, respectively, samples were fasting serum samples from smokers and stage and grade were combined to define aggressive tumours (as opposed to our more restricted definition of aggressive disease based on tumour stage, nodes and metastasis but not grade) [8, 9]. Among other metabolite classes, amino acids, glycerophospholipids and sphingolipids were measured and with some overlap of specific metabolites with the current study (19 amino acids and two biogenic amines). The first ATBC analysis [8] did not find a significant association between citrulline and prostate cancer, but for overall prostate cancer the OR per SD increase (0.87, 95% CI 0.62–1.23) was similar to the observation in our study (0.90, 95% CI 0.82–0.99). We did not replicate the ATBC results indicative of a lower risk of prostate cancer (p < 0.05) in men with higher concentrations of some amino acids, i.e. alanine (for overall and aggressive prostate cancer), lysine (for overall and non-aggressive prostate cancer), methionine (for overall and aggressive prostate cancer) or phenylalanine (for overall and non-aggressive prostate cancer). In contrast, we observed a conventionally significant higher risk of death from prostate cancer with higher methionine concentration. The main finding of the first ATBC study was a strong inverse association between 1-stearoylglycerol, a product of lipid breakdown, and risk of overall and aggressive prostate cancer, but we did not have data on this metabolite. In line with the first ATBC analysis, the second analysis [9], with no overlap in participants, showed inverse associations of lipids (including glycerophospholipids) and energy metabolites (involved in the Krebs cycle) with risk of aggressive prostate cancer. Similarly, we found inverse associations for several glycerophospholipids with risk for high grade, aggressive (defined based on stage) and especially advanced stage prostate cancer.

The PLCO analysis used the same metabolomic assay as the ATBC analyses, although on non-fasting rather than fasting serum samples, but in contrast to any of the other studies, participants were all screened for prostate cancer using an annual PSA test and digital rectal examination [11]. A lower risk of prostate cancer was suggested (p < 0.05) in relation to higher concentrations of some amino acids and their derivatives, including arginine (with aggressive prostate cancer; defined by combining stage and grade information) and tryptophan (with overall and aggressive prostate cancer). We did not replicate these findings. The reported positive associations with lipids for overall and aggressive prostate cancer were inconsistent with the results from ATBC and our findings. These differences may be due to the screening of the PLCO population [11], as screening was not common in ATBC [9] or EPIC.

The EPIC-Heidelberg analysis [10] mostly investigated the same metabolites as we did and used a similar assay. The strongest finding for overall prostate cancer in both studies was a suggested (p < 0.05) positive association with the glycerophospholipid PC ae C30:0 (EPIC-Heidelberg: OR_{top vs. bottom fourth} = 1.89, 95% CI 1.06–3.36). Our results were not materially changed after excluding the small overlap in participants between the two analyses. In both EPIC-Heidelberg and our analyses, the results for PC ae C30:0 did not differ by time to diagnosis. Additionally, the EPIC-Heidelberg analysis suggested inverse associations with three glycerophospholipids (all lysophosphatidylcholines) and positive associations with alanine, proline and methionine sulphoxide, the latter of which was not measured in our analysis. Our results for these metabolites were mainly in the same direction but less strong for overall prostate cancer, while we observed stronger inverse associations for two of the lysophosphatidylcholines with aggressive disease.

The strongest findings in our study were an inverse association of citrulline with risk of prostate cancer diagnosis within the first 5 years of follow-up, suggested positive associations of prostate cancer risk with t4-OH-Pro, SM (OH) C14:1 (for overall and death from prostate cancer for both metabolites) and PC ae C30:0 (for overall and aggressive prostate cancer), and the lower risk of advanced prostate cancer in relation to glycerophospholipid concentrations. However, relatively little is known about the potential biological role of these specific metabolites in carcinogenesis of the prostate.

Citrulline has antioxidant functions and has been shown to protect DNA and enzymes from reactive oxygen species [19], which might otherwise promote progression of prostate cancer via continuous proliferation and impaired apoptosis [20]. In line with our results, lower urinary concentrations of citrulline have been observed in patients with prostate cancer compared to healthy controls [21], perhaps because of altered citrulline metabolism in tumour cells. However, whether citrulline might be useful as a marker of subclinical prostate cancer needs to be confirmed.

It is not clear why t4-OH-Pro (one variant of hydroxyproline) might be related to risk of prostate cancer, but urinary excretion of hydroxyproline has been previously recognised as an early marker of bone metastases in patients with prostate cancer [22, 23], as hydroxyproline is released from collagen in tumour invasion [24]. While 4-hydroxyproline has also been suggested as a marker of red and processed meat [25, 26], the evidence does not suggest an association between these foods and prostate cancer risk [27].

Standard amino acids have often been reported to differ between controls and patients with various cancer types [4, 28], but it is not clear if the associations with the non-standard amino acids citrulline and t4-OH-Pro are specific to prostate cancer or not.

A role of sphingomyelins (the type of sphingolipid investigated here) in carcinogenesis might be explained by their involvement in cell proliferation, migration and autophagy [29]. Higher sphingomyelin concentrations in prostate tumour tissue and patients’ plasma than in benign prostatic hyperplasia tissue and control participants, respectively, have been reported [30, 31]. SM (OH) C14:1 has also been suggested as a marker of cream intake [25], and intake of dairy products might be related to prostate cancer risk [32].

All glycerophospholipids investigated here were phosphatidylcholines, the homeostasis of which (including PC ae C30:0) plays a critical role in cell regulation, with increased synthesis leading to proliferation [33]. Higher plasma phosphatidylcholine concentrations in patients with prostate cancer than in controls have been reported [31], which is in line with the positive association of PC ae C30:0 with risk. While possible mechanisms for the inverse associations between several phosphatidylcholines and advanced prostate cancer are not clear, positive associations have been reported between concentrations of diacyl-phosphatidylcholines (denoted PC aa Cx:y in the current paper) and type 2 diabetes [34], which is linked to a lower prostate cancer risk [35, 36]. The association between phosphatidylcholines and risk might apply to malignancies in general rather than being prostate cancer specific. Changes in phosphatidylcholine concentrations have been reported in patients with cancers of the bladder, brain, breast, kidney, liver, lung and ovaries [4, 28, 37], and in the prospective analysis in the EPIC-Heidelberg subcohort, associations with breast and colorectal cancer were also suggested [10]. Altered circulating phosphatidylcholine concentrations in men who are subsequently diagnosed with cancer may be due to altered lipid uptake and metabolism by rapidly proliferating cancer cells in subclinical tumours; these metabolites are required for membrane synthesis and lipid-based cell signalling [38].

To date, this is the largest study of metabolite concentrations and risk of prostate cancer. The relatively large number of participants has enabled us to investigate not only total risk of prostate cancer but also to conduct exploratory analyses by time to diagnosis and prostate tumour characteristics, and of death from prostate cancer. Furthermore, the detailed information on covariates has reduced the risk of confounding driving the results.

The choice of metabolomics assay was determined by coverage of metabolites of a priori interest (amino acids) and cost, but use of a targeted assay limits our analysis to the metabolites covered by the assay.

The study has a number of other limitations. Only one blood sample was available per participant, which will attenuate the results if a single measure does not represent long-term exposure. Reproducibility of metabolites over 4 months to 2.3 years has been reported to be moderate to high (median interclass correlation coefficients: 0.54–0.70), suggesting that a single measurement may be adequate for most metabolites [18, 39, 40], although the reproducibility was lower for some acylcarnitines and a few amino acids [18, 39]. Secondly, pre-analytical conditions, e.g. food consumption prior to blood collection and use of anticoagulant in blood sampling tubes, could potentially diminish our ability to detect associations [41‐43]. However, fasting status has been found to explain only a small amount of the variability of metabolite concentrations [41, 44‐46], and cases and controls were matched on time since last food or drink to minimise any risk of bias due to fasting status. Similarly, cases and controls were matched by centre as a proxy for sample handling. Like other anticoagulants, use of citrate in plasma samples can affect measurements of metabolites, including amino acids, glycerophospholipids and sphingomyelins [42, 43]. Thus, the metabolite concentrations reported here might not be directly comparable to concentrations measured in serum, or in plasma treated with EDTA or heparin. However, relative risk estimates are unlikely to be affected, as citrate was used in all samples. Thirdly, variation in the histological grading and stage classification between pathologists and over time could result in some misclassification of tumours by stage and grade category [47‐50], which in turn might lead to attenuation of risk estimates in subgroup analyses. Finally, although this is the largest study to date of metabolites and prostate cancer risk, the numbers and thus the statistical power are still limited, especially for analyses of tumour subtypes and death. Furthermore, a relatively conservative controlling procedure for multiple testing which does not account for correlations between metabolites was used.

The inclusion of data from further incident prostate cancer cases and matched controls in future analyses will increase the reliability of estimates of the associations between metabolite concentrations and prostate cancer risk, overall and for tumour subtypes. Dimension-reduction approaches that allow the investigation of patterns in metabolite profile may also provide further insights into the role of plasma metabolites in prostate cancer development.