Background

In men, prostate cancer (PCa) is the most common type of cancer and the second leading cause of cancer-related death [1]. The majority of PCa patients have an indolent clinical course and can be managed with active surveillance or local treatment. However, for patients with aggressive metastatic disease that has failed hormonal therapy, available treatments currently only extend life by a few months [2]. To increase survival, it is crucial to improve patient stratification at diagnosis, as this would prevent overtreatment of indolent disease and place appropriate focus on patients with aggressive and potentially lethal tumors. Analysis of metabolites can highlight deviations from a particular physiologic status and metabolomics is already routinely employed in diagnostics to identify a variety of pathological situations [3, 4]. Metabolites can be measured in bodily fluids such as urine and plasma, which are readily accessible without the need for invasive procedures [5]. The application of metabolomics to PCa analysis has revealed that prostate tumors have metabolic profiles that are distinct from those of normal prostate tissues and other cancers [6,7,8,9]. PCa cells often use fatty acid (FA) metabolism as major source of energy production [10], while other cancer typically use glycolysis to sustain proliferation (Warburg Effect) [11, 12]. It has been proposed that de novo lipid biogenesis and β-oxidation are among the most altered pathways in PCa metabolism and PCa cells exhibit increased uptake of FAs, such as palmitate [13], as well as increased FA β-oxidation [14]. Additionally, PCa is frequently accompanied by the metabolic syndrome which presents altered blood lipid levels and obesity [15, 16]. Thyssle et al. [17] detected high levels of FAs in plasma of PCa patients with metastasis compared to patients without metastasis or with benign prostatic hyperplasia (BPH). Giskeødegård et al. also measured higher levels of plasma FAs (particularly acylcarnitines) in PCa patients compared to BPH controls [18]. Additionally, in a multi-center study [19], comparison of pre-diagnostic plasma from 1077 PCa cases and 1077 controls showed that long-chain acylcarnitines (C14:1, C18:1 and C18:2) were inversely associated with advanced PCa stages, while short-chain acylcarnitine C3 was positively associated with aggressive and lethal disease. Recently, the same authors confirmed these data in another cohort of 3057 matched case-control sets of pre-diagnostic plasma [20]. Crowe et al. [21] similarly found an association between palmitic acid and risk of low-grade PCa. These studies suggest that there is an inverse association between the precursors or early intermediates of FA β-oxidation (i.e. long-chain acylcarnitines) and risk of disease. Conversely, these findings suggest that shorter-chain acylcarnitines are positively correlated with aggressive disease.

In the present study we hypothesized that metabolic profiling of pre-operative plasma collected from individuals diagnosed with PCa will lead to the identification of biomarkers associated to disease progression and stratification. To test this hypothesis, we performed untargeted metabolomics analysis of pre-operative plasma from treatment naïve PCa patients who were classified as low risk, high risk without progression or high risk with progression, according to the PCa stratification criteria of the National Comprehensive Cancer Network (NCCN) [22]. In parallel, we analyzed plasma from PCa metastatic patients and healthy controls, in order to identify metabolites that could be specific for disease status and progression.

Methods

Study design and setting

We conducted a case-control study within a cohort of PCa patients who presented at the Department of Urology, Inselspital Bern University Hospital in Bern, Switzerland between 1996 and 2015. Primary PCa cases were aged between 46 and 74 years, did not present any apparent metastasis and received radical prostatectomy immediately after blood collection. Fasting blood was drawn early in the morning before surgery, processed and separated into components [23]. Metastatic cases were aged between 54 and 86 years at time of blood collection and already presented metastasis. Healthy controls were aged between 29 and 55 at blood collection. For all participants, the plasma was collected in EDTA tubes and stored at − 80 °C. All participants provided informed consent for participation in this study.

Cases and controls

Primary cases were men who received a PCa diagnosis following digital rectal examination, prostate specific antigen (PSA) assessment and TNM staging evaluation of prostate biopsies prior to blood collection. The primary cases (n = 33) were selected in order to identify at least 10 age matched patients for each of the PCa stratification levels defined according to the National Comprehensive Cancer Network (NCCN) [24, 25] and to integrate these with clinical follow-up, namely low risk (T1-T2a, Gleason score ≤ 6, PSA < 10 ng/mL), high risk without progression (T ≥ 3a, Gleason score 8–10, PSA > 20 ng/mL, without local or bone progression within 5 years after radical prostatectomy) and high risk with progression (T ≥ 3a, Gleason score 8–10, PSA > 20 ng/mL with local or distant progression at last follow-up within 5 years from radical prostatectomy). PSA progression was defined as two consecutive rising PSA measurements > 0.1 ng/mL. Local progression was confirmed by imaging modality and/or biopsy. Lymph node progression was identified by imaging modalities (MRI, CT-scan or Choline/PSMA-PET CT). Bone progression was identified by skeletal scintigraphy. Of all the primary PCa patients (N = 33), only one (N = 1) received androgen deprivation therapy (Casodex) prior to blood collection.

Metastatic cases were men (n = 9) diagnosed with PCa and who presented evidence of metastatic spreading as documented by skeletal scintigraphy. These are advanced patients, who already received either radiation therapy and/or androgen deprivation treatments. The blood collected from these individuals was not fasting. Controls were healthy individuals (n = 10) with no evidence of disease, who volunteered to participate in this pilot study.

Untargeted metabolomics analysis

Plasma samples were extracted and analyzed by reversed-phase chromatography coupled to high-resolution mass spectrometry (HRMS) as previously described [26]. Metabolic features measured by UHPLC-HRMS were isolated with Progenesis QI (version 2.2, Nonlinear Dynamics, Newcastle, UK) and analyzed as previously described [26]. Potential structures and formulas of the isolated metabolic features were searched against the Human Metabolome Database (HMDB, version 3.6) [27] with a mass accuracy of 8 ppm and against an in-house database containing retention times of a set of 378 compounds. Features identified against the in-house database with a retention time deviation < 0.4 min were accepted as potential identity. The assigned features were manually reviewed for correct peak shape and identity assignment with MS/MS data.

Statistical analysis

Hazard ratios (HRs) for the association between risk of progression and normalized abundances by median cut-off based on the distribution in primary cases were calculated using multivariate Cox proportional-hazard regression model using the Survival [28, 29] and Survminer [30] R package. Normalized abundances data were used to estimate the associations and data were tested for proportional-hazards (PH) assumption and by graphical diagnostics based on the scaled Schoenfeld residuals [31] prior to analysis. The corresponding HRs represent the risk associated with a higher or lower (median cut-off) abundance of the indicated metabolite. Kaplan Meier curves to illustrate the progression of the stratified risk groups in primary PCa cases were calculated using the “survfit” function and the Log-Rank test using the Survival R package [28, 29]. Kaplan Meier curves to identify the association between selected metabolites and patient outcome were calculated using median cut-off normalized values of metabolites abundances with the “survfit” function and differences estimated with the Log-Rank test with the Survival R package [28, 29]. Correlation among the metabolites were calculated using the Pearson parametric correlation test (cut-off p < 0.05) with the Hmisc R package [32]. Differential expression analysis on publically available dataset was conducted with ShinyGEO [33]. Visualization of genomic data was generated with cBioPortal [34, 35]. For calculation of Kaplan Meier curves, TCGA PRAD gene expression data were retrieved from Firehose [36] with RTCGAToolbox R package [37]. Data were downloaded as RSEM normalized values and transformed to log2 in order to achieve normal distribution, genes whose expression was 0 in more than 50% of the samples were removed from analysis. If more than one follow-up was available data were collapsed and the higher information (more recent) maintained. Survival was estimated by calculation of optimal cut points using the Survminer R package [30] and differences identified by Log-Rank test. Principal component analysis (PCA) was conducted on the normalized abundances of the m/z features measured in positive and negative mode by applying log2 transformation and data scaling with the FactoMineR R package [38]. Features with abundances equal to 0 in more than 50% of the sample were excluded from the analysis. Additionally, abundances lower than 250 were removed to reduce noise. Eigenvalues were used to determine the number of principal components to be considered and to display the variability of the data. Confidence interval corresponding to 95% is represented in the ellipses on the PCA plots. For all the data representation, graphical plots were generated using the ggplot2 R package [39]. The analyses were done using RStudio version 1.1.463 [40] and R version 3.5.3 [41].

Cell lines and culture conditions

The parental cell lines used in this study have been authenticated using highly polymorphic short tandem repeat (STR) loci. PC-3 M-Pro4 cells originate from serial passage of PC-3 M cells in the prostate of athymic mice [42] and were cultured in DMEM (GibcoBRL) with 4.5 g glucose/L, 10% FCII (Thermo Fisher Scientific), 1% penicillin–streptomycin (PS, Life Technologies). C4–2 [43], C4–2B4 [44] and LNCaP [45] cells were cultured in T-medium DMEM (Sigma-Aldrich) with 20% F-12 K nutrient mixture Kaighn’s modification (GibcoBRL), 10% FCS, 0.125 mg/mL biotin, 1% insulin-transferrin-selenium (ITS), 6.825 ng/mL T3, 12.5 mg/mL adenine, 1% PS. LNCaP cells were derived from PCa lymph node metastasis [46], C4–2 cells was derived from LNCaP cells by passaging in castrated mice [43] and C4–2B from a bone metastasis of LNCaP tumor in nude mice [44]. All cells were cultured at 37 °C and 5% CO2.

Cell proliferation assay

To assess proliferation, cells were seeded at a density of 1500 cells per well and growth monitored for 48 h. AU 490 nm was measured 2 h after incubation with 20 μL of 3-(4,5 dimethylthiazol- 2-yl)- 5 -(3 -carboxymethoxyphenyl)- 2 -(4 -sulfophenyl)- 2 Htetrazolium (MTS, Promega) at 37 °C according to manufacturer’s protocol. Data were normalized for the number of cells seeded and fold change estimated versus control condition (vehicle treatment).

Quantitative real-time PCR (qPCR) analysis

Total RNA was extracted using Trizol (Invitrogen) and cDNA synthesized according to manufacturer’s instructions (Promega) and as previously described [47]. Real-time qPCR was performed with QuantStudio3 (Thermo Fisher Scientific). HPRT was used for normalization and relative expression is calculated with 2−ΔCt formula. Primers sequences are indicated in Additional file 1: Table S1.

Dihydrotestosterone (DHT) and palmitic acid (PA) stimulation

DHT was diluted in ethanol (EtOH) and administered at a final concentration of 10 nmol/L [48]. EtOH was used as control. PA was prepared in albumin fatty acid free (BSA-FFA) [49] (Sigma) and administered at a final concentration of 100 μM [50]. BSA-FFA was used as control. For both treatments, incubation time was 48 h [50]. Before the treatment, in order to wash out the androgens, cells were starved for 48 h in medium containing charcoal stripped serum (CSS). Experimental conditions were replicated (n = 3, biological repeats) and measured (n = 2, technical replicate for each sample).

Results

Characterization of cases and controls

Characteristics of all samples (42 cases and 10 controls) and subgroups of PCa risk categories as classified in the NCCN guidelines (v1.2018) [24] (11 low risk cases, 12 high risk without progression, 10 high risk with progression and 9 Metastatic) are shown in Table 1. All individuals included in the study were Caucasian. The median age of the patients at blood draw was 61.4 years for the low risk group, 62.9 for the high risk without progression group, 63.6 for high risk with progression group and 59.8 for the metastatic group, while the median age of the healthy controls was 38.3 years at blood draw. The PSA median level at blood draw was 5.8 ng/mL for low risk patients, 9.32 ng/mL for high risk patients without progression and 10.9 ng/mL for high risk patients with progression. A median follow-up higher than 5 years (after radical prostatectomy) is available for all the primary cases (7.59 years for the low risk, 7.14 for the high risk without progression and 5.11 for high risk with progression). For the metastatic cases the median follow-up is 3.23 years from blood draw. Among the primary cases, high risk with progression had a significantly lower overall survival compared to low risk and high risk without progression (Log-Rank p = 0.002) (Fig. 1a). Both high risk patients without local and bone progression and high risk patients with local and bone progression experienced PSA progression (following radical prostatectomy) at a median of 5.88 years and 0.28 years, respectively, compared to low risk (no biochemical relapse, log-rank p < 0.0001) (Fig. 1b). High risk patients with progression displayed local recurrence at a median of 4.04 years after radical prostatectomy while high risk patients without progression had local recurrence at 6.98 years compared to low risk cases where no progression was observed (log-rank p = 0.0051) (Fig. 1c). Lymph node progression and bone progression were the two main events discriminating the high and low risk groups. Lymph node progression occurred at a median of 3.16 years in the high risk with progression cases while for high risk without progression group the median was essentially comparable to low risk cases (7.14 versus 7.59 years, respectively, after radical prostatectomy) (Log-Rank p < 0.0001) (Fig. 1d). Bone progression was detected in high risk with progression cases at a median of 5.11 years after radical prostatectomy compared to 7.14 and 7.59 years for high risk without progression and low risk, respectively (Log-Rank p = 0.00027) (Fig. 1e).

Table 1 Characteristics of the cases and controls
Full size table
Fig. 1
figure 1

Characterization of disease progression in PCa cases. a Overall survival calculated by the Kaplan Meier method for high-risk with progression (HR + P, blue solid line), high-risk without progression (HR-P, red dotted line) and low-risk (LR, black dashed line) PCa cases. P-value estimated with Log-Rank test. b PSA progression free survival calculated by the Kaplan Meier ethod for HR + P, HR-P and LR PCa cases. P-value estimated with Log-Rank test. c Local progression free survival calculated by the Kaplan Meier method for HR + P, HR-P and LR PCa cases. P-value estimated by the Log-Rank test. d Lymph node progression-free survival calculated by the Kaplan Meier method for HR + P, HR-P and LR PCa cases. P-value estimated by the Log-Rank test. e Bone progression-free survival calculated by the Kaplan Meier method for HR + P, HR-P and LR PCa cases. P-value estimated with the Log-Rank test. For statistical details see Methods section

Full size image

Metabolomics analysis

Principal component analysis of all the m/z features measured in positive and negative ionization mode was able to distinguish primary versus metastatic cases and healthy controls (Fig. 2a and Additional file 2: Figure S1A). However, a heatmap of all the m/z features demonstrated that hierarchical clustering could not differentiate between the primary subgroups (low risk versus high risk without progression versus high risk with progression) (Fig. 2b).

Fig. 2
figure 2

Sample classification by principal component analysis and overview of metabolomics profiling. a Principal component analysis (PCA) of all the m/z features measured in positive and negative ionization mode. The variation retained by PC1 (16.3%) is represented of the X axis and the variation retained by the PC2 (9.3%) is represented on the Y axes. Ellipses represent the 95% confidence interval for each group. b Heatmap generated with scale and centered normalized m/z abundances measured in positive and negative ionization mode. Euclideian distance between groups and Minkowski distance between metabolites was used and clustering was calculated with the Ward (Ward.D2) method for minimum variance between m/z values. c Principal component analysis (PCA) of the annotated m/z features in positive and negative ionization mode. The variation retained by PC1 (16.7%) is represented of the X axis and the variation retained by the PC2 (13.5%) is represented on the Y axis. Ellipses represent the 95% confidence interval for each group. d Heat map generated with scale and centered normalized m/z abundances of the annotated metabolites. Same methods used for Fig. 2b were applied. For statistical details see Methods section

Full size image

Principal component analysis of the annotated metabolites was able to distinguish healthy controls versus patients, but not between the primary risk subgroups (Fig. 2c, Additional file 2: Figure S1B and Fig. 2d).

Characterization of acylcarnitines distribution and prognostic associations

A complete panel of acylcarnitines (from C2 to C16) was detected in patients and controls (Table 2, values normalized against L-carnitine (C0) levels). We analyzed the impact of acylcarnitines on disease associated risks by multivariate Cox regression models, for all the primary patients, given that the median age of the individuals in these groups was highly comparable. Medium- chain acylcarnitines (C6-C12) were positively associated with risk of PSA progression (p = 0.036, median cut-off) (Additional file 3: Figure S2A) and long-chain acylcarnitines were inversely associated with local and bone progression risk (p = 0.034 and p = 0.033 respectively, median cut-off) (Additional file 3: Figure S2B and S2D). No association with lymph node progression risk was detected by short-, medium-, or long-chain acylcarnitines (using pooled values for class of molecules, namely C2-C5 (short-chain), C6-C12 (medium-chain), C14-C16 (long-chain), normalized against C0, median cut-off) (Additional file 3: Figure S2C). When we investigated the contribution of the single molecules, we found an inverse association between acetylcarnitine (C2) and risk of PSA progression (p = 0.016, global Log-Rank p = 0.05 with multivariate Cox regression analysis for the panel of acylcarnitine associated with risk of PSA progression tested for proportional hazard assumption, median cut-off) (Fig. 3a). Isovalerylcarnitine (C5) and hexanoylcarnitine (C6) were inversely and positively respectively associated with risk of PSA progression (p = 0.035 and p = 0.01, respectively, normalized against C0, median cut-off) (Fig. 3a). However, calculation of Kaplan-Meier curves for PSA progression (using Log-Rank test), did not display a significant difference for C2 and C5 (High versus Low, median cut-off), while higher C6 was significantly associated with PSA progression (Log-Rank p = 0.014) (Fig. 3b, c and d).

Table 2 Geometric mean (95% confidence interval) normalized abundances of acylcarnitines in Cases and Controls
Full size table
Fig. 3
figure 3

Hazard ratio (95% confidence interval) and Kaplan Meier curves for the association of acylcarnitines with PSA progression. a Hazard ratios and 95% confidence interval (CI) for the association of the complete panel of acylcarnitines with the risk of PSA progression. Groups (lower and higher risk) were separate by median cut-off (suffix “_med”) of the normalized abundances for each molecule. C4 = butyrylcarnitine, C4_M = methylmalonylcarnitine. b PSA progression-free survival calculated by the Kaplan Meier method for acetylcarnitine (C2). Groups are defined by median-cut off of normalized abundances (vs L-carnitine) of high acetylcarnitine (C2 High, blue solid line) vs low acetylcarnitne (C2 Low, red dotted line). P-value was estimated with Log-Rank test. c PSA progression free survival calculated by the Kaplan Meier method for isovalerylcarnitine (C5). Groups are defined by median-cut off of normalized abundances (versus L-carnitine) of High isovalerylcarnitine (C5 High, red dot line) versus low isovelrylcarnitne (C5 Low, blue solid line). P-value was estimated with Log-Rank test. d PSA progression-free survival calculated by the Kaplan Meier method for hexanoylcarnitine (C6). Groups are defined by median-cut off of normalized abundances (vs L-carnitine) of high hexanoylcarnitine (C6 High, blue solid line) versus low hexanoylcarnitne (C6 Low, red dotted line). P-value was estimated with Log-Rank test. For statistical details see Methods section

Full size image

While no association with bone progression was established, we found an inverse association between L-carnitine (C0) and isovalerylcarnitine (C5) and risk of lymph node progression (p = 0.006 and p = 0.027, respectively, median cut-off, Log-Rank p = 0.04 with multivariate Cox regression analysis) (Fig. 4a). Calculation of Kaplan-Meier curves for lymph node progression displayed a significant difference for C0 (High versus Low, median cut-off, log-rank p = 0.044), but not for C5. (Fig. 4b and c). We did not find a significant association between the classes of acylcarnitines and the risk of bone progression.

Fig. 4
figure 4

Hazard ratio (95% confidence interval) and Kaplan Meier curves for the association of acylcarnitines with lymph node progression. a Hazard ratios and 95% confidence interval (CI) for the association of the complete panel of acylcarnitines with the risk of lymph node progression. Groups (lower and higher risk) were separate by median cut-off (suffix “_med”) of the normalized abundances for each molecule. C4 = butyrylcarnitine, C4_M = methylmalonylcarnitine. b Lymph node progression-free survival calculated by the Kaplan Meier method for L-carnitine (C0). Groups are defined by median-cut off of normalized abundances of high L-carnitine (C0 High, red dotted line) versus low L-carnitne (C0 Low, blue solid line). P-value was estimated with Log-Rank test. c Lymph node progression free survival calculated by the Kaplan Meier method for isovalerylcarnitine (C5). Groups are defined by median-cut off of normalized abundances (versus L-carnitine) of high isovalerylcarnitine (C5 High, red dotted line) vs low isovelrylcarnitne (C5 Low, blue solid line). P-value was estimated with Log-Rank test. For statistical details see Methods section

Full size image

As depicted in Fig. 5a we found a strongly and positive correlation between C2 and medium- (C6-C12) and long-chain (C14-C16) acylcarnitines in primary cases (cut-off p < 0.05), compared to control cases (Fig. 5b). We also detected a strong and positive correlation between the classes of medium- and long-chain acylcarnitines in primary and control cases, compared to metastatic cases (Fig. 5d).

Fig. 5
figure 5

Correlation matrix for all metabolites identified in PCa cases and controls. Correlations are shown for the different acylcarnitines in primary PCa cases (a) and controls (b). c Insert with plot A displays the correlation among the acylcarnitines and all the annotated metabolites in primary cases. d Correlation among acylcarnitines in metastatic cases. The sizes of the circles are dependent on the Pearson correlation coefficient. Blue circles correspond to positive correlations and red circles correspond to negative correlations. Correlations that do not reach significance (p > 0.05) are indicated by an empty square box. For statistical details see methods section

Full size image

Suberic acid and indoleacetic acid are associated with PSA progression

When the analysis was extended to all the annotated metabolites, we found a positive and significant correlation in primary patients between medium- and long-chain acylcarnitines (namely C6, C12 and C14) and suberic acid, an unsaturated dicarboxylic acid that is frequently elevated in patients with fatty acid oxidation disorders [51] (Fig. 5c). This association was stronger in the metastatic cases, for hexanoylcarnitine (C6), L-octanoylcarnitine (C8) and decanoylcarnitine (C10) (all belonging to the medium-chain acylcarnitines class) (Additional file 4: Figure S3). In the metastatic cases we also detected a remarkable strong and significant inverse correlation between medium-chain acylcarnitines and tryptophan detected in positive and negative ionization mode (this effect was consistent also in primary cases). When we calculated the Kaplan-Meier curves (Log-Rank test by median cut-off), we found that high levels of suberic acid were significantly associated with PSA progression (p = 0.032, Log-Rank test) (Fig. 6a). Furthermore, high levels of indoleacetic acid, a breakdown product of tryptophan metabolism directly downstream indoleacetaldehyde, were significantly associated with PSA and lymph node progression (p = 0.0081 and p = 0.025, respectively, Log-Rank test) (Fig. 6b and c). Notably, indoleacetaldehyde is metabolized to indoleacetic acid by aldehyde dehydrogenase 7 family member A1 (ALDH7A1), which is associated with aggressive PCa [52, 53]. All these assigned features were manually reviewed for correct peak shape and identity assignment with MS/MS data (Additional file 5: Table S2).

Fig. 6
figure 6

Association between annotated metabolites and disease progression and analysis of gene expression data. PSA progression-free survival calculated by the Kaplan Meier method for (a) suberic acid and (b) indoleacetic acid and lymph node progression-free survival for indoleacetic acid is represented in (c). Groups are defined by median-cut off of normalized abundances (low abundance, blue solid line) vs high abundance (red dotted line). P-value was estimated with log-rank test. IDO1 expression data of primary and androgen ablation resistant metastasis are shown from GSE6752 (d) and GSE6919 (e-f). ACADM expression data in the same set of samples are displayed in (g-h-i). Fold change (FC) is calculated versus the normal or primary tumor samples and p-value (P) for significance between two groups estimated by t-test. For statistical details see Methods section

Full size image

Indoleamine 2,3-dioxygenase (IDO1) and medium-chain acyl-CoA dehydrogenase (ACADM) are elevated in metastatic prostate cancer

Indoleamine 2,3-dioxygenase (IDO1) was recently shown to enhance carnitine palmitoyltransferase I (CPT1) activity and fatty acid oxidation by degrading L-tryptophan [54]. This prompted us to search for a possible association between IDO1 and the medium-chain acylcarnitines pattern that we detected in primary and metastatic cases. To do this, we investigated the expression of IDO1 and medium-chain Acyl-CoA Dehydrogenase (ACADM) in primary and metastatic PCa by analyzing gene expression profiles from datasets containing androgen-ablation resistant PCa metastatic samples [55] and normal and tumor adjacent tissues [55, 56]. We found a significant increase of IDO1 in metastasis VS primary PCa (p = 0.014) (Fig. 6d and Additional file 6: Figure S4A) and a similar upregulation was detected in tumor and tumor adjacent tissues compared to normal tissue (Fig. 6e, f and Additional file 6: Figure S4B). While ACADM was similarly higher in primary PCa tissue compared to metastasis tissue (p = 0.001) (Fig. 6g and Additional file 6: Figure S4C), there was no difference in ACADM expression in tumor tissue compared to tumor adjacent tissues (Fig. 6h, i and Additional file 6: Figure S4D). When we looked at the distribution of IDO1 and ACADM genomic alteration in a panel of PCa studies, we found a substantial distribution of deletions and amplifications of IDO1 while ACADM also appeared to be mutated in a considerable fraction of the samples (Additional file 7: Figure S5A and S5B). However, higher transcriptional levels of IDO1 were not associated with disease progression (in TCGA PRAD) [36] and while higher ACADM expression displayed a trend toward disease progression, this was not significant (p = 0.065, Log-Rank test with maximized cut-off groups) (Additional file 7: Figure S5C and S5D).

Fatty acid stimulation enhances proliferation and transcription of AR responsive genes

Finally, we investigated the effect of fatty acid treatment on proliferation in C4–2B4 and PC-3 M-Pro4 cells, and regulation of androgen receptor (AR) responsive genes in AR positive LNCaP and its derivative C4–2 cells. Treatment of C4–2B4 and PC-3 M-Pro4 cells with palmitic acid (PA) (range between 50 and 200 μM) resulted in increased proliferation (Fig. 7a, b, c, d and Fig. 7e, f, g and h, respectively). Administration of PA alone in C4–2 cells was not able to induce expression of AR-responsive genes (Fig. 7i, j, k and l). However, administration of PA in combination with DHT, resulted in an additive effect tha was significant for KLK3 and TMPRSS2 (Fig. 7j and l). These results were confirmed in LNCaP cells (Fig. 7m, n, o, and p) where the additive effect was consistent for KLK3 (Fig. 7n) but not for TMPRSS2 (Fig. 7p).

Fig. 7
figure 7

In vitro functional characterization of DHT and palmitate (PA) stimulation on AR positive and negative PCa cells. a Proliferation assessed in C4–2B by MTS assay upon PA (range 50–400 μM) or vehicle (control) stimulation for 48 h. b-d Representative bright field images of cultured cells under indicated experimental conditions. e Proliferation assessed in PC-3 M-Pro4 by MTS assay upon PA (range 50–400 μM) or vehicle (control) stimulation for 48 h. f-h Representative bright field images of cultured cells upon experimental conditions. i-l Relative expression or AR responsive genes upon dihydrotestosterone (DHT) or control (EtOH) stimulation, and palmitate (PA) or vehicle (BSA fatty acid free-FFA) or combination (PA + DHT) in AR positive C4–2 and (M-P) LNCaP cells. P-value indicated in the plots is relative to ANOVA. Multiple comparison significance between experimental condition is indicated by * (** p < 0.01, *** p < 0.001). Details related to quantification and normalization are included in the Methods section

Full size image

Discussion

In this study, we have provided evidence that preoperative plasma fatty acid metabolites can inform risk of prostate cancer progression. Specifically, we find that medium-chain acylcarnitines are positively associated with the risk of PSA progression and long-chain acylcarnitines are inversely associated with progression of PCa locally and to the bone. Additionally, in primary and metastatic cases, medium-chain acylcarnitines were positively associated with suberic acid, which also correlated with the risk of PSA progression. Two multi-center studies conducted by Schmidt et al. [19, 20] have also identified a correlation between acylcarnitines and PCa disease by measuring pre-diagnostic plasma metabolites and searching for those that appeared to be associated with a certain PCa grade or stage. In these studies, men with higher plasma concentrations of long-chain acylcarnitines (C18:1 and C18:2) had lower risk of advanced stage prostate cancer at diagnosis, while propionylcarntine (C3) and methionine metabolism appeared to be positively associate with aggressive disease and PCa death [19].

The main difference between these studies and our study is that Schmidt et al. [19, 20] employed pre-diagnostic plasma to identify molecules associated with risk of developing PCa, while we analyzed pre-operative-plasma, with the goal of linking specific metabolite profile with disease progression and risk stratification groups as defined by the NCCN guidelines. We found that reduced incidence of lymph node and bone progression correlated with higher levels of long-chain acylcarnitines. Despite looking at a different end points, employing a reduced sample size and starting from different samples, it is remarkable that our data are aligned with the results of Schmidt et al. [19, 20] and seem to suggest that better patient outcome is associated with elevated levels of long-chain acylcarnitines in both pre-diagnostic and pre-operative plasma.

Our extended analysis of all annotated metabolites indicates that the perturbation of the acylcarnitine metabolism in primary and metastatic cases is consistent with that of other related metabolites. For example, we detected a strong and significant positive correlation between medium-chain acylcarnitines and suberic acid, which is present in the urine of patients with fatty acid oxidation disorders and was elevated in individuals with perturbations in ACADM [51]. The fact that medium-chain acylcarnitines and suberic acid were both elevated in the same samples in this study, might reflect the metabolic state of more aggressive prostate cancers. In addition to their correlation, it is remarkable that medium-chain acylcarnitines and suberic acid both display the same inverse association with tryptophan. In a study conducted on serum from 64 PCa cases and 50 healthy controls, Zang et al. shown that tryptophan is a specific discriminatory metabolite of PCa [57]. It has been shown that cancer increases the consumption of tryptophan to evade immune control [58] and this has been linked to IDO1 activity. Degradation of L-tryptophan by IDO1 promotes CPT1 activity, thereby enhancing fatty acid metabolism [54]. In PCa, PTEN-deficient tumors have been associated with an immunosuppressive microenvironment mediated by increased expression of IDO1 [59].

The patterns of metabolites that we observed in primary versus metastatic samples and the transcriptional analysis of tumor tissue versus normal tissue, are consistent with the elevated levels of IDO1 that we found in metastatic compared with primary cases and that we found in androgen-ablation resistant PCa metastatic samples [55]. Increased levels of IDO1 in patient with metastatic disease compared with patients with primary tumors would lead to a strong reduction in L-tryptohpan and a corresponding increase in fatty acid metabolism products (C2 and medium-chain acylcarnitines), as we observed in our cohort. However, the precise association with ACADM activity should be further investigated, since its increased levels in metastatic versus primary cases might t also suggest a decrease in the products of the enzymatic reactions (i.e. medium-chain acylcarnitines), rather than an accumulation, as documented here.

Levels of other products of tryptophan metabolism support the link between tryptophan and medium-chain acylcarnitines. In a metastatic setting, a reduction of L-tryptophan levels due to high IDO1 expression, would lead to reduced levels of indoleacetaldehyde, which is a substrate of ALDH7A1. Further experimentation should be conducted to elucidate whether the augmented levels of ALDH7A1, already reported to be a marker of advanced PCa [52, 53], can be supported by these metabolic alterations and eventually represent an adaptation mechanism resulting in higher levels of indoleacetic acid, thus supporting its correlation with PSA and lymph node progression.

In addition to correlating fatty acid metabolism with risk or PCa progression, we have also shown that fatty acids impact PCa cell behavior. We find that PA and DHT have an additive effect on the expression of AR responsive genes in LNCaP and C4–2 cells, which normally grow in medium without DHT supplementation and are therefore capable of proliferating under castration conditions. These data are in line with recent findings that interference in fatty acid metabolism via inhibition of de novo lipogenesis targets androgen receptor signaling in castration-resistant prostate cancer [60]. Additionally, they reinforce previous data that PCa cell lines have fatty acid and fatty acid β-oxidation altered [61]. Moreover, these observations are supported by the increase in proliferation that we observed upon PA stimulation of PC-3 M-Pro4 cells, which lack AR expression and are by definition a castration resistant model. While our observations are supported by recent evidence that suppressing fatty acid uptake has therapeutic effects in preclinical models of PCa [62], additional experimentation should be conducted to elucidate whether increased fatty acid levels affect AR signaling to promote an aggressive phenotype.

While the number of patients allocated in each of the classification groups (low risk, high risk without progression and high risk with progression) is quite small, the total number of PCa cases (n = 33) is in line with other studies conducted in single centers [17, 18, 57, 63, 64] and, importantly, includes age-matched primary cases. Our results are primarily associated with disease progression rather than disease status and the fact that our control group displayed a significantly lower age at blood draw compared to PCa cases (primary and metastatic), does not impact with the association that we conduct within the group of primary cases. Furthermore, it should be noted that this is a pilot study, and our findings are confirmatory of previously published work, as elaborated in this discussion. The principal discriminative factor between the groups of our primary cases was the progression to lymph node or bone metastasis. However, even with this extreme end point, we were not able to identify metabolites specific for a particular risk group. Tissue samples from the PCa cases included in this study cohort are currently being retrieved to further validate the metabolic association with stratified risk groups.

Conclusions

In conclusion, our pilot study confirms and extends previous findings that fatty acid metabolism is associated with advanced PCa and risk of disease progression and our results suggest that evaluation of acylcarnitines profile might contribute to improve patient stratification. Our findings further demonstrate that analysis of metabolites in plasma is a valid method for generating novel hypothesis worth to be validated in an independent cohort and to identify patients who are at risk of disease progression. Finally, our data suggest that this metabolomics strategy could be employed to assist patient stratification at diagnosis.