Association of Prediagnostic Blood Metabolomics with Prostate Cancer Defined by ERG or PTEN Molecular Subtypes

The TMPRSS2:ERG gene fusion leading to ERG overexpression is the most common somatic event in primary prostate cancer (1), which is found in approximately 25% of tumors in men of Asian and African descent and approximately 50% of tumors in men of European descent (2). PTEN is the most frequently deleted tumor-suppressor gene in prostate cancer, with approximately 15% of primary prostate cancer cases showing homozygous deletions spanning the PTEN locus (3). Prior studies have suggested that the PTEN deletions are associated with worse outcomes; however, TMPRSS2:ERG fusion is not, but it does appear to be an etiologically distinct subtype (4, 5). Moreover, it has been reported a close biological relationship between ERG overexpression and PTEN loss (6, 7), and may delineate distinct prostate cancer subtypes with different prognosis if they co-occurred (8).

Accumulating evidence suggests that ERG and PTEN define etiologically distinct molecular subtypes. Observational studies have demonstrated an association between risk factors and prostate cancer defined by ERG and PTEN. For example, obesity (9), height (10), tomato sauce consumption (11), and vigorous physical activity (12) were found to be associated with ERG-positive prostate cancer; statin use was found to be associated with lower risk of PTEN-null prostate cancer (13). Furthermore, ERG-positive prostate cancers are characterized by alterations in insulin receptor (IR), insulin growth factor 1 receptor (IGF-1R), other tumor-specific metabolic alterations (14), and several prostate cancer genetic risk variants (15).

Given metabolites are the final downstream products of the genome as well as the intermediates or end products of multiple enzymatic reactions on external exposures, the metabolomic profile before disease onset might provide implications to refine existing risk factors and improve our understanding of etiologically distinct molecular subtypes of prostate cancer. We hypothesized that the systemic metabolic state could contribute to the selective pressure in the prostate environment resulting in outgrowth of malignant clones with genetic alterations that confer increased fitness in a specific metabolic environment. To date, no studies have investigated circulating metabolomic profiles and prostate cancer molecular subtypes. To fill this gap, we leveraged metabolomics data from the prospective Health Professionals Follow-up Study (HPFS) and the Physicians' Health Study (PHS), to investigate the association of prediagnostic blood metabolites with prostate cancer defined by ERG or PTEN status.

Data for this study were derived from two prospective US cohorts, HPFS and PHS. Details on the design and characteristics of the overall HPFS and PHS cohorts have been described elsewhere (16, 17). In brief, the HPFS was initiated in 1986 among 51,529 male health-professionals ages 40 to 75 years at baseline. Participants completed an initial questionnaire at baseline and have been followed up by mailed questionnaires every two years to update exposure information and ascertain incident diseases, including prostate cancer, and the follow-up rates for each biennial cycle have consistently been >90% (18). During 1993 to 1995, blood samples were collected from 18,225 HPFS participants (19). The PHS was initiated in 1982 as a randomized, double-blind, placebo-controlled trial to test the effects of low-dose aspirin and beta-carotene in the primary prevention of cardiovascular diseases and cancer among 22,071 US male physicians, ages 40 to 84 at baseline. Baseline blood specimens were collected and frozen for later analyses from 14,916 participants during 1982–1984 (20). Self-reported blood fasting time was recorded in both cohorts.

For the current study, we included participants who had been previously selected for nested case–control studies of metabolomics in the HPFS and PHS. In HPFS, we included a case–control study of 213 advanced (T3b/T4/N1/M1/fatal) prostate cancer cases and 213 matched controls, matched on age at blood draw, PSA screening, as well as time, season and calendar year of blood draw. We additionally included a case study comprised of 294 non-advanced cases diagnosed between 1993 and 2014 for whom tumor tissue was available. We excluded 1 case for missing clinical TNM stage, 257 cases without ERG or PTEN information, and 13 controls for confirmed diagnosis of prostate cancer using medical records. This resulted in 200 controls, 242 prostate cancer cases with ERG data, and 194 cases with PTEN data from HPFS for this study.

In PHS, we first included a case–control study that comprised of 181 prostate cancer cases (100 cT1–3 and Gleason score≥8 cases + 81 cT4/N1/M1cases) and 100 controls free from prostate cancer at the time of diagnosis of matched cases, frequency matched on age at baseline (40–49/50–59/60–69/70+ years) and fasting status (last meal <8 vs. ≥8 hours). We then included a case-only study among localized, low-grade prostate cancers (i.e., cT1–3 and Gleason score = 2–7). This study was comprised of 48 cases who died of prostate cancer within 10 years and 85 cases who survived 10 years and beyond. 46 men in PHS were excluded because of insufficient blood sample remaining to undertake the metabolomics assays. After excluding 239 cases without ERG or PTEN information, 94 controls, 35 prostate cancer cases with ERG data, and 17 cases having PTEN data were included from PHS.

The study protocol was approved by the institutional review boards of the Brigham and Women's Hospital and Harvard T.H. Chan School of Public Health, and those of participating registries as required.

Incident prostate cancers were initially self-reported on questionnaires, followed by cancer diagnosis adjudication by extracting clinical and treatment information from medical records and pathology reports (21). Archival prostate tumor tissue from approximately half of patients with prostate cancer in HPFS was retrieved and underwent central histopathologic review by study pathologists for the standardized tumor grading.

ERG and PTEN status was determined by IHC using previously constructed and validated tumor tissue microarrays (TMA; ref. 22). Tumors were classified as ERG-positive if the case had positive ERG staining (antibody: Clone EPR3864, Epitomics, Inc.) within prostate cancer epithelial cells on at least one TMA core (≥3 cores per subject). ERG IHC status is strongly associated with fusion status as assessed by FISH (23). Tumors were classified as PTEN-loss if PTEN IHC expression (antibody: Clone D4.3 XP; Cell Signaling Technology) was either markedly decreased or entirely lost across >10% of tumor cells compared with surrounding benign glands or stroma (8). PTEN IHC status is strongly associated with PTEN homozygous genetic deletion (8, 24).

Plasma metabolites were profiled at the Broad Institute using untargeted liquid chromatography tandem mass spectrometry methods as described previously (25). Briefly, two methods were applied for the measurement of circulating metabolites: (i) amines and polar metabolites that ionize in the positive ion mode were measured using an LC-MS platform comprised of an Open Accela 1250 U-HPLC coupled with a Q Exactive hybrid quadrupole orbitrap mass spectrometer (Thermo Fisher Scientific); (ii) polar and non-polar lipids were measured using an LC-MS platform comprised of a Shimazu Nexera X2 U-HPLC coupled to an Exactive Plus orbitrap mass spectrometer (Thermo Fisher Scientific). We dropped unknown metabolites, metabolites with coefficient of variation higher than 25% or intraclass correlation coefficient (ICC) less than 0.4 as indicators of interassay reproducibility and within-person reproducibility, metabolites with undetectable levels in >10% of participants, and metabolites that could not be reasonably reproducible in samples with delayed processing in previous pilot tests (25, 26). Metabolite values below the detection limit were assigned a half value of detection limit.

To normalize the distribution of metabolite data, the relative abundance of each metabolite was initially transformed using the natural logarithm, and then scaled to a mean value = 0 and a standard deviation (SD) = 1. To evaluate whether a subset of metabolites was associated with molecular subtype of prostate cancer, metabolites were grouped into categories based on their biological class. First, for all individual metabolites, multinomial logistic regression models were applied using the R package “multinom” to obtain P values for the association between each individual metabolite and risk of prostate cancer by ERG and PTEN status. Second, enrichment analysis was performed using Mean-rank Gene Set Test (GSA) implemented with the R package “geneSetTest.” This analysis ranks individual metabolites by their associated P value (from previous step) and identifies metabolites classes whose metabolites are highly ranked relative to other metabolites (27) in the ranked list. Metabolite classes with nominal P values less than 0.05 were included for the further analyses. Multinomial logistic regression models were applied using the R package “multinom” to estimate odds ratios (OR) and 95% confidence intervals (CI) for the associations of individual metabolites with prostate cancer by ERG and PTEN status. In both the group analyses and individual analyses, we adjusted for the following covariates, assessed immediately before or at the time of blood draw: Age at blood draw, cohort, height, body mass index (BMI), physical activity, smoking status, fasting status, and season.

We conducted sensitivity analyses to evaluate the robustness of our results. First, to minimize the likelihood that latent disease influenced metabolite measures, we excluded cases whose time from blood collection to disease onset was less than 3 years (79 ERG defined cases, 52 PTEN defined cases). Second, to assess potential heterogeneity by diabetes, which has been reported to be inversely associated with risk of prostate cancer (28) and associated with diverse alterations in the metabolome (29), we excluded participants who had self-reported diabetes at the time of blood draw (14 control, 5 ERG-defined cases, 5 PTEN-defined cases). Third, to assess potential heterogeneity by race, we restricted the analysis to white men (excluded 27 control, 20 ERG-defined cases, 10 PTEN-defined cases).

All tests of statistical significance were two-sided. To account for multiple comparisons, the FDR was controlled to 0.05 using the Benjamini and Hochberg approach. All analyses were conducted using SAS 9.4 software (SAS Institute) and R 3.6.1 (cran.r-project.org). P values were considered significant at values <0.05.

Blood-based metabolomics data were available for 277 cases with ERG status, 211 cases with PTEN status, and 294 controls. Compared with men without prostate cancer, men with prostate cancer were less likely to be diabetics and current smokers and included a higher proportion of men who provided fasting blood samples and men who delivered samples in winter (Table 1). As expected, the proportion of PSA level greater than 4 ng/mL at diagnosis and the proportion of advanced- and high-grade cancers were higher for PTEN-loss versus PTEN-intact cases, but similar by ERG status.

We classified 223 known metabolites into 15 categories based on their chemical class, including sphingomyelin, ceramides, unsaturated triacylglycerols, saturated triacylglycerols, unsaturated diacylglycerols, saturated diacylglycerols, phosphatidylcholines, phosphatidylethanolamines, lysophosphatidylcholines, lysophosphatidylethanolamines, cholesterol esters, amino acids, amino acid derivatives, carnitines, as well as others (Supplementary Table S1). Table 2 shows the nominal and FDR-corrected P values for the associations between the metabolite categories and ERG and PTEN-defined prostate cancer, compared with controls. We found an enrichment of sphingomyelins (P: 0.01, FDR-P: 0.15), ceramides (P: 0.04, FDR-P: 0.18), and phosphatidylethanolamines (P: 0.03, FDR-P: 0.18) in serum of men who developed ERG-positive prostate cancer compared with controls. In contrast, there were no nominally significant associations with ERG-negative disease.

For PTEN-defined prostate cancer, unsaturated triacylglycerols were associated with both PTEN-loss (P: 0.001, FDR-P: 0.01) and PTEN-intact cases (P: 0.0001, FDR-P: 0.002) compared with non-cancer controls. Sphingomyelins (P: 0.02, FDR-P: 0.09), ceramides (P: 0.005, FDR-P: 0.04), and amino acids (P: 0.02, FDR-P: 0.09) were associated with PTEN-loss cases, whereas unsaturated diacylglycerols were associated with PTEN-intact cases (P: 0.003, FDR-P: 0.02).

Figure 1 shows associations of individual metabolites in the metabolite classes that nominal P values less than 0.05 with risk of ERG-defined prostate cancer. Among the 10 sphingomyelins metabolites, eight were positively associated with ERG-positive prostate cancer. Specifically, higher risks were observed for men with higher concentrations of C24:0 SM (OR, 1.29; 95% CI, 1.03–1.62), C18:2 SM (OR, 1.26; 95% CI, 1.00–1.58), and C22:1 SM (OR, 1.26; 95% CI, 1.01–1.59). In addition, all (n = 4) ceramide metabolites were positively associated with ERG-positive cancer, whereas 19 of 21 phosphatidylethanolamine were inversely associated with ERG-positive cancer. Compared with ERG-positive prostate cancer cases, the associations of individual metabolites for the risk of ERG-negative prostate cancer in the sphingomyelins, ceramide, and phosphatidylethanolamine classes were weaker and conflicting. For example, except C18:0 SM (OR, 0.81; 95% CI, 0.65–1.00), there was no metabolite significantly associated with the risk of ERG-negative prostate cancer; higher risk of ERG-positive prostate cancer was observed for men with higher concentrations of C22:1 SM, but not for ERG-negative prostate cancer (OR, 0.98; 95% CI, 0.80–1.21). None of the metabolites remained statistically significantly associated with ERG-defined prostate cancer after accounting for multiple comparisons (Supplementary Table S2), and associations of individual metabolites with ERG-defined prostate cancer in the nonsignificant metabolic classes were shown in the Supplementary Table S3.

Figure 2 shows associations of individual metabolites in the metabolite classes that nominal P values less than 0.05 with PTEN-defined prostate cancer. For the PTEN-loss prostate cancer, among the 10 sphingomyelins metabolites, eight were showed positive association. Specifically, C18:2 SM (OR, 1.64; 95% CI, 1.11–2.42), C22:1 SM (OR, 1.60; 95% CI, 1.07–2.39), and C22:0 SM (OR, 1.49; 95% CI, 1.01–2.22) were positively associated with PTEN-loss prostate cancer diagnosis with nominal significance (Fig. 2A). Meanwhile, all ceramide (n = 4) and unsaturated diacylglycerol (n = 11) metabolites, and 92% metabolites in the amino acids class were also positively associated with PTEN-loss cancer (Fig. 2B–D). However, among unsaturated triacylglycerols, the direction of the association was diverse, for example, C52:3 TAG (OR, 1.54; 95% CI, 1.02–2.32) and C48:3 TAG (OR, 1.52; 95% CI, 1.01–2.29) were positively associated with PTEN-loss prostate cancer, whereas C58:8 TAG was inversely associated with PTEN-loss cancer (OR, 0.65; 95% CI, 0.45–0.95; Fig. 2E). It is interesting that individual metabolites that significantly associated with PTEN-loss prostate cancer showed much weaker associations with PTEN-intact prostate cancer. Such as C58:11 TAG (PTEN-loss: OR, 0.69; 95% CI, 0.50–0.97; PTEN-intact: OR, 0.90; 95% CI, 0.73–1.11). None of the metabolites remained statistically significantly associated with PTEN-defined prostate cancer after accounting for multiple comparisons (Supplementary Table S4), and the associations of individual metabolites with PTEN-defined prostate cancer in the nonsignificant metabolic classes were shown in the Supplementary Table S5.

We performed a number of sensitivity analyses, excluding subjects who had cancer diagnosed within 3 years after blood collection, excluding subjects with a history of diabetes, and restricting the analysis to self-reported white men. We did not observe qualitative differences in associations for the identified metabolites groups. In particular, the enrichment of sphingomyelins for ERG-positive prostate cancer, as well as ceramides and unsaturated triacylglycerols for PTEN-defined prostate cancers were robust in all three sensitivity analyses (Supplementary Tables S6 and S7).

In this population-based case–control study, we investigated whether prediagnostic-circulating metabolites were associated with the development of ERG or PTEN-defined prostate cancer. Among the 15 predefined metabolite groups, sphingomyelins, ceramides, and phosphatidylethanolamines were enriched in ERG-positive prostate cancer cases compared with controls. Unsaturated triacylglycerols were enriched in both PTEN-loss and PTEN-intact prostate cancers, whereas sphingomyelins, ceramides, and amino acids were enriched in PTEN-loss prostate cancer and unsaturated diacylglycerols were enriched in PTEN-intact prostate cancer. Although no individual metabolites were statistically significantly associated with ERG or PTEN-defined prostate cancer after accounting for multiple testing, the patterns of metabolites associated with these subtypes remained. These results suggest that alterations in overall pathways of metabolites, rather than individual metabolites, are better able to define the systemic metabolic state associated with ERG or PTEN prostate cancer.

Ten nested case–control studies have been conducted to prospectively examine the association between circulating metabolites and prostate cancer risk, including three studies in the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort (30–32), three in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) cohort (33–35), and studies from the Janus Serum Bank Cohort in Norway (36), Prostate, Lung, Colorectal, and Ovarian Cancer Screening (PLCO) Trial (37), Cancer Prevention Study-II Nutrition Cohort (38), and Northern Sweden Health and Disease Study (39). However, none of the studies have examined the associations for molecular subtyped prostate cancers, and some findings could not be replicated in different cohorts. For example, among the two studies that applied GSA similar to our study to measure associations between chemical classes of metabolites and risk of prostate cancer, ATBC reported inverse associations between energy and lipid compounds and prostate cancer (34); however, PLCO found a very different metabolite-risk profile that featured primarily amino acids and peptides (37). The differential findings across studies may reflect differences in metabolomic assay tools, biospecimen processing, study population, and metabolomic approaches for dealing with metabolites and case identification (screening detected vs. clinically detected). However, it is also possible that the molecular subtype of prostate cancer cases might contribute to the interpretation for the distinct patterns in different cohorts.

Large-scale cohort studies have found that high levels of prediagnostic-circulating ligand IGF-1 are associated with an increased risk of prostate cancer (40). In addition, our previous findings suggested that ERG-positive tumors are characterized by altered metabolic signaling pathways, such as higher expression of IR and IGF-1R, compared with ERG-negative tumors (14). Meanwhile, IGF-1 signaling downregulates the expression of PTEN as the PI3K/PTEN/Akt pathway is the downstream of IGF1/IGF-IR (41). Given that insulin sensitivity is strongly influenced by the presence of insulin receptor, IGFs, and IGF receptors (42), we hypothesized that metabolites correlated with insulin sensitivity or IGF signaling might be differently enriched in the prediagnosed samples for patients with prostate cancer with ERG or PTEN subtypes.

Sphingomyelin is generated from ceramide by the transfer of phosphocholine from phosphatidylcholine with the generation of diacylglycerol through sphingomyelin synthase. Accumulation of metabolites in the sphingomyelin signaling pathway, including ceramides, sphingomyelins and diacylglycerols, were found to be involved in the development of insulin resistance (43, 44), and several metabolites in sphingomyelins and ceramides classes were reported to be positively related to the risk of fatal prostate cancer (31, 38), which might contribute to the interpretation for the positive associations of sphingomyelins and ceramides with risk of EGR-positive and PTEN-loss prostate cancer in our study. In addition, although the weak and diverse directions for the associations between metabolites in the unsaturated diacylglycerols class and the risk of fatal or advanced prostate cancer were reported by Wang and colleagues (38), we observed that unsaturated diacylglycerols, were nonsignificantly, positively associated with both PTEN-loss and -intact prostate cancer risk, though, owing to the limited case number, the association needs further validation.

The association between amino acids and prostate cancer risk has been inconsistent. Alanine, lysine, methionine, phenylalanine, arginine, and tryptophan were inversely associated with prostate cancer in ATBC or PLCO (34, 37), whereas these results were not replicated in EPIC (31) or in our study. Moreover, the enrichment of amino acids in PTEN-loss prostate cancer was not replicated in our sensitivity analyses. Phosphatidylethanolamines, an abundant membrane phospholipid that is essential for membrane integrity, cell division, and membrane protein topology, has been identified as one of the biomarkers for diagnosis of prostate cancer with a high sensitivity, specificity, and accuracy (45). Dalmau and colleagues (46) found that when prostate cancer cells undergo epithelial-to-mesenchymal transition induction, which plays a crucial role in cancer metastasis, the level of 12 unsaturated triacylglycerols were increased in cells. Among the 11 metabolites that overlapped in our study, C52:3 TAG was found to be positively associated with PTEN-loss prostate cancer risk with nominal P < 0.05, seven others were found to be positively but nonsignificantly associated both PTEN-loss and -intact prostate cancer risk.

There are several potential limitations in our study. First, the sample size was modest, particularly for the PTEN-loss prostate cancer cases. Although differential associations of metabolites were observed according to ERG and PTEN status, no associations reached statistical significance after accounting for multiple comparisons. Given that, it is more difficult to explore the metabolomic profiles for prostate cancer cases co-occurred ERG-positive and PTEN-loss. However, we conducted an exploratory analysis and divided cases into PTEN-intact/ERG-negative (n = 182) and PTEN-loss and ERG-positive (n = 22) groups. The results showed that for the PTEN-loss and ERG-positive groups, the significant enrichment of ceramides (P < 0.001), but not sphingomyelins (P = 0.12) was observed; for the PTEN intact/ERG-negative group, unsaturated triacylglycerols (P < 0.001) and unsaturated diacylglycerols (P = 0.004) were enriched. A larger epidemiological study with access to both prediagnostic bloods and tumor tissue materials for molecular subtyping will be needed to confirm what we observed in this study. Such effort may require the pooling of data across cohorts. Second, more than 90% of the men in our study are of European descent, which may limit the generalization of the results to other populations. This is particularly important given that the prevalence of TMPRSS2:ERG is lower in men of African and Asian ancestry. The strengths of our study include the combination of detailed longitudinal data on external exposures, prediagnostic metabolite profiling, and molecular defined outcomes together for the innovative analysis. The inclusion of well captured information on BMI, smoking status, fasting status, and the season of blood draw allowed us to evaluate associations independent of these factors.

In summary, the present study demonstrates the distinct metabolomic profiles associated with ERG and PTEN-defined prostate cancer in men nested from two prospective cohorts with long-term follow-up. Metabolites in the sphingomyelin signaling pathway, especially sphingomyelins and ceramides, appear to be positively associated with ERG-positive and PTEN-loss prostate cancer, and might contribute to the development of prostate cancer in different molecular subtypes. Larger studies are needed to confirm our findings, and further our understanding of the environmental factors that contribute to the etiology of distinct molecular subtypes of prostate cancer.

C.K. Zhou reports grants from NIH during the conduct of the study. B.A. Dickerman reports grants from National Institutes of Health during the conduct of the study. S.P. Finn reports personal fees from Roche and nonfinancial support from MSD and Pfizer outside the submitted work. M.G. Vander Heiden reports grants from NCI during the conduct of the study, as well as personal fees from Agios Pharmaceuticals, Iteos Therapeutics, Auron Therapeutics, Faeth Therapeutics, and Aeglea Biotherapeutics outside the submitted work. L.A. Mucci reports grants from National Cancer Institute during the conduct of the study. No disclosures were reported by the other authors.

X. Feng: Formal analysis, methodology, writing–original draft, writing–review and editing. C.K. Zhou: Formal analysis, methodology, writing–review and editing. C.B. Clish: Data curation. K.M. Wilson: Writing–review and editing. C.H. Pernar: Methodology. B.A. Dickerman: Writing–review and editing. M. Loda: Data curation, writing–review and editing. S.P. Finn: Data curation, writing–review and editing. K.L. Penney: Writing–review and editing. D.R. Schmidt: Data curation, writing–review and editing. M.G. Vander Heiden: Data curation, writing–review and editing. E.L. Giovannucci: Supervision, writing–review and editing. E.M. Ebot: Conceptualization, formal analysis, writing–review and editing. L.A. Mucci: Conceptualization, supervision, writing–review and editing.

We would like to thank the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY. The authors assume full responsibility for analyses and interpretation of these data. The Health Professionals Follow-up Study is supported by U01 CA 167552 from the National Cancer Institute. The tissue microarrays were constructed by the Tissue Microarray Core Facility at the Dana-Farber/Harvard Cancer Center (P30 CA 006516). This project was supported in part by the Dana-Farber/Harvard Cancer Center SPORE in Prostate Cancer (P50 090381) and the US Army Prostate Cancer Research Program. X. Feng was supported by the program of China Scholarships Council (No.201806210455). L.A. Mucci, K.L. Penney, S.P. Finn, and K.M. Wilson were Prostate Cancer Foundation Young Investigators. D.R. Schmidt is supported by the Harvard Catalyst/Harvard Clinical and Translational Science Center (NIH Award KL2 TR002542). B.A. Dickerman is supported by National Institutes of Health Grant (K99 CA248335).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.