Urine metabolic analysis as a non-invasive method to predict biochemical recurrence in prostate cancer

Abstract

Background

Metabolomics has proven to be a useful science for obtaining biomarkers in prostate cancer. In this work, urine samples were analyzed by nuclear magnetic resonance (NMR) spectroscopy to identify potential urinary biomarkers associated with biochemical recurrence in prostate cancer.

Methods

Urine samples were obtained from patients undergoing transrectal prostate biopsy after prostate massage. Patients were classified as with or without biochemical recurrence after having received prostate cancer treatment. All spectra were acquired using a Bruker Avance III DRX 600 spectrometer. Univariate and multivariate analysis were performed with metabolites and clinical variables to predict tumor presence.

Results

Data were collected from 70 patients treated for prostate cancer, 16 of whom developed biochemical recurrence within 5 years following treatment, with an average time to diagnosed recurrence of 25.68±15.39 months. After establishing a predictive model with the 25 most influential metabolites in Partial Least Squares Discriminant Analysis (PLS-DA analysis), a predictive model of biochemical recurrence was obtained with an area under the curve of 0.95, a sensitivity of 80%, specificity of 98%, positive predictive value (PPV) of 92% and a negative predictive value (NPV) of 96%. Metabolites derived from amino acid metabolism and glycolysis featured most predominantly in this model.

Conclusions

The metabolic profile in urine can be used to construct a model with good discrimination for predicting the development of biochemical recurrence. The molecules highlighted herein frequently belong to amino acid metabolism and glycolysis.

Highlight box.

Key findings

• After establishing a predictive model with the 25 most influential metabolites in Partial Least Squares Discriminant Analysis (PLS-DA analysis), a predictive model of biochemical recurrence was obtained with an area under the curve of 0.95, a sensitivity of 80%, specificity of 98%, positive predictive value (PPV) of 92% and a negative predictive value (NPV) of 96%. Metabolites derived from amino acid metabolism and glycolysis featured most predominantly in this model.

What is known and what is new?

• Metabolomics is a science that allows obtaining biomarkers in a non-invasive way.

• This study used urine nuclear magnetic resonance (NMR) spectroscopy to uncover new data on biomarkers that could predict biochemical recurrence in prostate cancer.

What is the implication, and what should change now?

• The metabolic profile in urine can be used to construct a model with good discrimination for predicting the development of biochemical recurrence. The molecules highlighted herein frequently belong to amino acid metabolism and glycolysis.

Introduction

Prostate cancer remains among the most frequently diagnosed malignancies in men globally, accounting for approximately 11.7% of all new cancer cases, with an incidence of about 19% in developed nations and 5.3% in developing regions (1). Treatment and follow-up of this neoplasia differ considerably according to multiple clinical variables: patient age, serum level of prostate-specific antigen (PSA), local or systemic disease involvement, and patient preferences (2). One phase of the disease, biochemical recurrence, usually precedes clinical recurrence by years. Biochemical recurrence is defined as two consecutive elevations in serum PSA above 0.2 ng/mL after radical prostatectomy or a serum PSA level greater than 2 ng/mL above the nadir PSA after radiotherapy, regardless of whether or not hormone therapy has been received (3).

The growing interest in obtaining non-invasive diagnostic tools for use in prostate cancer management algorithms has driven the discovery of multiple biomarkers from different scientific disciplines in recent years (4). Among these approaches, metabolomics employs nuclear magnetic resonance (NMR) spectroscopy to explore metabolic alterations associated with disease processes, using either tissue specimens or non-invasively collected biofluids such as urine or saliva. Indeed, metabolomics has helped uncover different metabolic pathways implicated in prostate cancer development that could be useful for disease diagnosis, particularly those involving metabolites such as citrate, glycerol-3-phosphocholine, glycine, carnitine, and 0-phosphocholine (5,6).

Our aim in this study was to use urinary NMR spectroscopy to establish possible biomarkers associated with the risk of biochemical recurrence in patients treated for prostate cancer. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-592/rc).

Methods

Study design and inclusion criteria

This prospective, observational study collected consecutive cases of male patients attending at the University Clinic Hospital of Valencia to undergo medically-indicated transrectal prostate biopsy over 24 months. Individuals aged over 18 years who provided written informed consent were invited to participate. For the present work, we selected the subgroup with prostate cancer diagnosis subsequently treated by radical prostatectomy or radiotherapy with curative intent, and this sample was subdivided into those with or without biochemical recurrence (following the definition set forth in the introduction). Figure 1 shows the flow chart of patient selection; of the 201 patients included for tissue sample collection, 70 had received treatment with surgery or radiotherapy.

Figure 1.

Flow chart of patient selection for the study.

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by institutional ethics board of INCLIVA, Health Research Institute (No. 05.2014). Informed consent was taken from all the patients.

Sample collection

A transrectal massage was first performed, and urine was collected after the subsequent spontaneous urination, immediately prior to biopsy. Then, ten transrectal biopsy cores were obtained from each participant for standard histopathological evaluation. Collected urine samples were promptly transferred to the Molecular Imaging and Metabolomics Laboratory of the Central Medical Research Unit, University of Valencia, where they were frozen at −80 ℃ and stored until analysis. The biopsy cores for histological assessment were fixed in formalin immediately after sampling and delivered to the Pathology Department. Based on histological findings, patients were classified as having or not having prostate cancer. Liquid urine samples were prepared in 5 mm high-throughput NMR SampleJet tubes. The general sample preparation procedure consisted of adding 80 µL of a solution of NaH₂PO₄·H₂O (39 mM)/Na₂HPO₄·7H₂O (62 mM)/trimethylsilyl propanoic acid-d4 (TSP, 31.54 mM) in D₂O in 425 µL of urine and introducing the entire mixture into the 5 mm NMR tube. The final concentration of TSP in the measuring tube was 5.05 mM.

The procedure for acquiring and processing NMR spectra can be found in Appendix 1.

The database was updated as cases were collected, so that patients’ baseline characteristics were recorded at the time of study inclusion, while anatomopathological characteristics were added one month after specimen collection. In all cases with a tumor diagnosis, we recorded disease management (or loss to follow-up, depending on the case) and presence or absence of biochemical recurrence (according to the definitions in the introduction) during the 5 years following treatment.

Statistical analysis

Statistical analyses were conducted using SPSS software version 20 (IBM SPSS Statistics for Windows, 2011; IBM Corp., Armonk, NY, USA). Initially, a comparative analysis was performed between patients with and without biochemical recurrence following treatment. The clinical variables included in this comparison were age, PSA, PSA density, PSA ratio, prostate volume, suspicious digital rectal exam (DRE), ultrasound suspicion of tumor, and International Society of Urological Pathology (ISUP) tumor grade (obtained from transrectal biopsy in cases treated with radiotherapy and from the prostatectomy specimen in those treated with surgery). For this purpose, we carried out univariate analyzes (Student’s t-test, median test and chi-squared test), followed by multivariate analysis (binary logistic regression) if a trend towards statistical significance was reached in univariate (P<0.3). In this comparative analysis, the dependent variable was biochemical recurrence. Statistically significant differences were set at P<0.05.

Multivariate analysis of metabolomic profiles was carried out using MATLAB R2019b (MathWorks, Natick, MA, USA), incorporating in-house scripts developed in the Molecular Imaging and Metabolomics Laboratory alongside the PLS Toolbox package (Eigenvector Research, Inc., WA, USA). In the search for global correlations, we used advanced multivariate statistical techniques such as principal component analysis (PCA) to detect patterns and/or groupings within the sample groups, and partial least squares discriminant analysis (PLS-DA) to force separation between groups and build discriminant models.

Finally, the predictive performance of the PLS-DA model, based on the most relevant metabolites [those with variable importance in projection (VIP) score >1], was evaluated through receiver operating characteristic (ROC) curve analysis and examination of its component matrix.

Results

Sample characteristics

The study cohort comprised 70 treated patients, among whom 16 experienced biochemical recurrence, with a mean time to recurrence of 25.68±15.39 months. Comparative analysis of the clinical variables of the two study groups (cases with and without biochemical recurrence) is shown in Table 1. The only variables with a tendency towards statistical significance were PSA ratio (0.10 in cases with recurrence versus 0.12 in cases without, P=0.20); suspicious DRE (56.2% in the recurrence group versus 32.9% in the non-recurrence group, P<0.08), and ultrasound suspicion of tumor (75% in the recurrence group vs. 57.4% in the non-recurrence group, P=0.21). ISUP grade 4–5 was more frequent in the biochemical recurrence group (75% vs. 29.6% in the group without recurrence, P=0.04).

Table 1. Univariate analysis. Differences between clinical characteristics of the two study groups (cases with and without biochemical recurrence).

Variable	Recurrence (n=16)	No recurrence (n=54)	P
Age, years	67.53±9.69	68.76±8.70	0.49†
PSA, ng/mL	11.5 [7–1,446]	9 [3–237]	0.78‡
PSA density	0.35 (0.13–3.20)	0.26 (0.06–24.80)	0.95‡
PSA ratio	0.10±0.08	0.12±0.06	0.20†
Prostate volume, cc	34.92±13.70	39.19±19.32	0.43†
Suspicious DRE	9 (56.2)	17 (32.9)	<0.08§
Suspicion on ultrasound	12 (75.0)	31 (57.4)	0.21§
ISUP 4–5 in biopsy/prostatectomy specimen	12 (75.0)	16 (29.6)	0.04†

Data are shown as median [range], mean ± SD, or n (%). ^†, Mann-Whitney U; ^‡, Median test; ^§, Chi-squared. DRE, digital rectal examination; ISUP, International Society of Urological Pathology; PSA, prostate-specific antigen; SD, standard deviation.

The subsequent multivariate analysis can be seen in Table 2. Patients with unfavorable histology in the biopsy or prostatectomy specimen (ISUP 4–5) were 2.5-fold more likely to develop biochemical recurrence than those with ISUP 1–3. No other clinical variable was related to developing biochemical recurrence.

Table 2. Multivariate analysis.

Variable	OR	95% CI	P value†
PSA ratio	0.05	0.01–263.43	0.49
Suspicious DRE	2.35	0.60–9.21	0.21
Suspicion on ultrasound	1.14	0.24–5.24	0.86
ISUP 4–5	2.5	1.25–3.45	0.04

Binary logistic regression including clinical variables with a tendency towards statistically significant differences in the previous univariate analysis (cases with and without biochemical recurrence). ^†, binary logistic regression. CI, confidence interval; DRE, digital rectal examination; ISUP, International Society of Urological Pathology; OR, odds ratio; PSA, prostate-specific antigen.

NMR determination of the metabolic fingerprint of prostate cancer with future biochemical recurrence in a urine sample obtained by spontaneous urination after transrectal massage

A total of 75 metabolites were identified. Figure 2 shows the average ¹H PRESAT spectra of urine samples for cases with vs. without biochemical recurrence. Each signal or peak unambiguously corresponds to a type of molecule or metabolite. A metabolite can present several signals in different parts of the spectrum, one for each magnetically active group. The area contained under each peak is directly proportional to the concentration of the metabolite in the sample. Table S1 shows the average and standard deviation of the same results in urine in cases with and without biochemical recurrence.

Figure 2.

Assignment of the most representative signals in a ¹H-NMR PRESAT spectrum, acquired at 600 MHz, for a urine sample from a subject without biochemical recurrence (black) and with biochemical recurrence (red) of prostate cancer at the time of diagnosis, measured at 310 K. (A) Aliphatic region of the spectrum (δ=0.5–4.5 ppm) and (B) aromatic region (δ=5.50–9.50 ppm). The most intense peaks in the spectrum were assigned as follows: 1, 3-methyl-2-oxovalerate; 2, isoleucine; 3, leucine; 4, valine; 5, isobutyrate; 6, methylsuccinate; 7, hydroxybutyrate; 8, fucose; 9, methylmalonate; 10, threonine; 11, lactate; 12, 2-hydroxybutyrate; 13, 2-phenylpropionate; 14, alanine; 15, arginine; 16, lysine; 17, N-acetylserotonin; 18, acetate; 19, proline; 20, glutamine; 21, glutamate; 22, O-acetylcarnitine; 23, acetoin; 24, acetoacetate; 25, succinate; 26, citrate; 27, sarcosine; 28, N-methylhydantoin; 29, creatinine; 30, cysteine; 31, cis-aconitate; 32, ethanolamine; 33, O-phosphocholine; 34, betaine; 35, trimethylamine N-oxide; 36, taurine; 37, myo-inositol; 38, wisteria; 39, ribose; 40, mannitol; 41, hippurate; 42, urea; 43, 3-hydroxykynurenine; 44, gentisate; 45, histidine; 46, indole-3-lactate; 47, phenylalanine; 48, phenylacetylglycine.

Unsupervised analysis: PCA, NMR metabolic fingerprinting of prostate cancer with future biochemical recurrence in a urine sample obtained by spontaneous urination after transrectal massage

The PCA (Figure 3) corresponding to the two groups is shown below, showing no spontaneous separation between their distribution. Cases with recurrence were heterogeneously distributed throughout the scatter plot, as were cases without biochemical recurrence.

Figure 3.

Scatter plot obtained from PCA analysis of 1H PRESAT spectra of urine samples for cases with biochemical recurrence versus cases without recurrence. The green squares represent cases without recurrence, and the purple triangles represent cases with recurrence. There is no spontaneous separation of the two samples. PC, principal component; PCA, principal component analysis.

Supervised analysis: PLS-DA, NMR metabolic fingerprint of prostate cancer with future biochemical recurrence in a urine sample obtained by spontaneous urination after transrectal massage

Supervised analysis was carried out to test for differential distribution by biochemical recurrence in the plot, checking for any tendency towards separation between the pre-established groups. The same cases were included as for PCA. The scatter plot is shown in Figure 4, revealing that supervised analysis also failed to identify differential distribution between the two groups. Although there was a certain tendency for subjects with recurrence to occupy the area of the upper right quadrant, they were distributed heterogeneously throughout all quadrants, and there was some overlap between cases of the two groups. Therefore, among all the metabolites studied in urine, the ability to differentiate between cases with and without recurrence was not proven.

Figure 4.

Scatter plot obtained from PLS-DA analysis of 1H PRESAT spectra of urine samples for cases with versus without biochemical recurrence. The green squares represent cases without recurrence, and the purple triangles represent cases with recurrence. LV, latent variable; PLS-DA, Partial Least Squares Discriminant Analysis.

Data cross-validation was carried out to assess the performance of the multivariate PLS-DA model in distinguishing patients with and without biochemical recurrence. The result of the validation is represented graphically with a ROC curve, shown in Figure 5. The ROC curve derived from PLS-DA showed an area under the curve of 0.43, demonstrating the poor discriminatory capacity of the model. With the results obtained so far, we proceeded to model optimization, selecting the molecules with a VIP score above 1 in PLS-DA to determine whether the most influential metabolites could better discriminate between the two groups.

Figure 5.

ROC curve obtained from cross-validation of the PLS-DA model that discriminates between cases with and without biochemical recurrence, in subjects who received some type of treatment for prostate cancer. The blue line corresponds to the treated data and the green line corresponds to the cross-validation. AUC, area under the curve; C, calibration; CV, cross-validation; PLS-DA, Partial Least Squares Discriminant Analysis; ROC, receiver operating characteristic.

Metabolic profile and metabolic pathways related to biochemical recurrence

After constructing the PLS-DA model, the most influential metabolites (VIP score >1) were examined (Table 3). The most striking impact was found in amino acid metabolism and glucose metabolism. Finally, logistic regression was performed to determine the influence of a series of metabolic variables on a dichotomous response variable, such as occurrence or not of biochemical recurrence. Therefore, all metabolites with a VIP score greater than 1 (a total of 25) were included as molecules with weight in the between-group discrimination.

Table 3. Metabolites identified with VIP score >1 and associated metabolic pathways in urine samples, cases with biochemical recurrence (they were selected to achieve model optimization).

Metabolite	Metabolic pathway
Creatine	Amino acid metabolism
3-hydroxykynurenine	Amino acid metabolism
2-phenylpropionate	Fatty acid metabolism
Proline	Amino acid metabolism
Trimethylamine N-oxide	Amino acid metabolism
N-acetylserotonin	Amino acid metabolism
Trigonelline	Nicotinic acid/carboxylic acid metabolism
Mannose	Glycolysis metabolism
Glucose	glucose metabolism
Arabinose	Glycolysis metabolism
Urea	Urea cycle
Sucrose	Glycolysis metabolism
Acetoacetate	Glycerophospholipid metabolism
Methylmalonate	Fatty acid oxidation
2-hydroxybutyrate	Fatty acid metabolism
Betaine	Glycerophospholipid metabolism
Myo-inositol	Inositol metabolism
Fucose	Glycolysis metabolism
Acetoin	TCA cycle
3-hydroxybutyrate	Fatty acid metabolism
N-methylhydantoin	urea cycle
Tyrosine	Amino acid metabolism
Methanol	Glycolysis metabolism
Taurine	Amino acid metabolism
Carnosine	Amino acid metabolism

TCA, tricarboxylic acid; VIP, variable importance in projection.

The ROC curve derived from the model with 25 metabolites is shown in Figure 6. The area under the curve was 0.95, with P<0.01 and a 95% confidence interval of 0.89–0.99. Furthermore, the sensitivity of the model was 80% and the specificity was 98%. The confusion matrix of the model is shown in Table 4. Upon optimization, the model had a great predictive capacity, as shown by the ROC curve, and it also presents high sensitivity, specificity and predictive values, as seen in Table 3.

Figure 6.

ROC curve obtained from the model with metabolites in urine with VIP score above 1 (blue curve), and analysis of biochemical recurrences. The green line represents the diagonal line that divides the ROC space. ROC, receiver operating charactersistic; VIP, variable importance in projection.

Table 4. Confusion matrix for the model derived from metabolites in urine with VIP score above 1 (disease vs. health state analysis).

Group	Sensitivity	Specificity	PPV	NPV
Biochemical recurrence	80%	98%	92%	96%

NPV, negative predictive value; PPV, positive predictive value; VIP, variable importance in projection.

Discussion

Changes in the concentration of metabolites in different body fluids reflect physiological and pathological states. The metabolome in any biological sample represents the final product of the so-called “omics cascade” and indicates the health or disease state of the subject in question. Therefore, obtaining metabolic signatures from biofluids can be a useful approach to identify non-invasive biomarkers and characterize the molecular mechanisms associated with pathological conditions (7).

Urine samples offer advantages for carrying out metabolomic studies, since they can be collected non-invasively (8). However, several limitations have been reported in the metabolic analysis of urine, principally the presence of copious urinary components in varying concentrations, which can result in poorly represented metabolites passing undetected in the sample or false positives in the identification of certain molecules (8,9). As urine is the final vehicle that transports metabolites from abundant metabolic pathways, which are modified depending on diet, pharmacological treatments or genetic profile, this can complicate the interpretation of metabolic profiles. Nonetheless, several studies have demonstrated the utility of metabolomics for detecting biomarkers in prostate cancer. These studies characteristically use high quantities of metabolites to build predictive models, given their nature as the final vehicle of many metabolic pathways. This fact differentiates metabolic studies in urine from those carried out in tissue, in which the models normally have a low number of molecules (10).

The present study is the first to use urine metabolomics as a non-invasive tool to predict biochemical recurrence in prostate cancer. The first study using metabolomics to predict biochemical recurrence, by Maxeiner et al., was carried out using tissue (11). These authors included 16 cases of biochemical recurrence, the same sample size as ours. By identifying a total of 27 spectral regions, they were able to predict biochemical recurrence with an accuracy of 78% (11). Our study differs in the use of a control group with up to 54 cases, and in the good results in terms of specificity (98%) and predictive values [positive predictive value (PPV) 92%, negative predictive value (NPV) 96%] obtained using an optimized model with 25 metabolites.

Two other publications in this line are by Braadland and Vandergrift (12,13). The first group collected a sample of 110 patients undergoing radical prostatectomy, of whom 50 developed biochemical recurrence. They verified through survival analysis that high intratumoral concentrations of spermine and citrate were related to cases free of biochemical recurrence, while high ratios of (total choline + creatine)/spermine and (total choline + creatine)/citrate were associated with short time to recurrence. Furthermore, spermine and the (total choline + creatine)/spermine ratio were independently associated in their multivariate analysis with development of recurrence (12).

Vandergrift et al.’s study retrospectively collected data from 185 patients, from which 365 specimens of prostatectomies for carcinoma were extracted, 15 corresponding to benign hyperplasia and 14 to benign specimens from prostate biopsy. The authors attempted to establish a differential metabolic profile according to tumor aggressiveness (comparing ISUP grades) and studied biochemical recurrence prediction through metabolomics. In addition to finding elevated concentrations of myo-inositol, glycerophosphocholine, phosphocholine, and valine in samples with high ISUP grades, they were able to establish a model to predict recurrence. With a true positive rate of 83% for pinpointing recurrence, the most influential metabolites were choline, phosphocholine, glycerophosphocholine and glutamate, while high levels of myo-inositol showed a significantly lower risk of recurrence during follow-up (13).

Regarding the most influential pathways in PLS-DA, the two key pathways in our study were amino acid metabolism and glucose metabolism, both previously implicated in prostate cancer development (5,6). Tumor cells typically display increased glycolytic activity to support both energy generation and biosynthetic demands. While aerobic glycolysis is not an exclusive metabolic signature of prostate cancer, enhanced glycolytic activity has been documented in its advanced stages. The main enzymes involved in this pathway—GLUT1, HK1, HK2, and PFK2—are regulated by the androgen receptor (14).

The role of GLUT1, a transmembrane glucose transporter overexpressed in prostate cancer and regulated by the androgen receptor, is also especially important. Several factors, such as hypoxia, can increase GLUT1 expression in cell membranes and consequently, promote glucose absorption. It has been shown that GLUT1 is more frequently overexpressed in prostate cancer and in cases with a high Gleason score. Likewise, loss of tumor protein 53, common in prostate cancer, also favors the expression of GLUT1 (15). Several metabolites presented herein, such as sucrose or mannose, could be useful in future studies to enhance understanding of the prostate cancer metabolome and for potential translation of this knowledge into clinical practice.

Regarding amino acid metabolism, this represents a source of carbon and nitrogen in the tricarboxylic acid cycle, producing NADPH. Glutamine is converted to glutamate via glutaminase, and then to α-ketoglutarate, which enters the tricarboxylic acid cycle, producing energy in the form of cellular ATP (16). Additionally, glutamine contributes to mitochondrial carbon accumulation to maintain mitochondrial membrane potential and for nucleotide, protein, and lipid synthesis (17). The androgen receptor in tumor cells regulates glutamine transporters and drives glutamine metabolism toward glutaminolysis, which leads to the accumulation of α-ketoglutarate. This receptor therefore promotes the absorption of glutamine, while drugs such as rapamycin decrease the absorption mediated by this receptor. Furthermore, some oncogenes, such as MYC, are regulated in prostate cancer, with the function of modulating glutaminolysis in mitochondria. Overexpression of MYC in prostate cancer induces the conversion of glutamine to glutamic acid, which is used as a carbon source for NADPH production (18).

Certain limiting aspects of our study are the single-center design and the sample size collected. Furthermore, a 5-year follow-up period in patients who received treatment may be short, considering the natural history of prostate cancer. Despite these limitations, we believe that our data can enhance understanding of the pathophysiology of prostate cancer and the predisposition towards biochemical recurrence, in addition to being useful for future management approaches in this neoplasia. As a distinguishing feature, ours is the first study attempting to predict biochemical recurrence via metabolic study in urine (a non-invasive method). Urinary biomarkers may be valid in future studies seeking to diagnose or monitor prostate cancer using non-invasive methods, for example, by performing NMR on tissue. Furthermore, our biochemical recurrence prediction model could gain predictive capability by adding clinical variables, thus opening avenues for future research in this setting.

Conclusions

The metabolic profile in urine can be used to build a model with good discrimination for predicting the development of biochemical recurrence. The molecules highlighted herein frequently belong to amino acid metabolism and glycolysis.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by institutional ethics board of INCLIVA, Health Research Institute (No. 05.2014). Informed consent was taken from all the patients.

Data Sharing Statement

Available at https://tau.amegroups.com/article/view/10.21037/tau-2025-592/dss

tau-14-11-3708-dss.pdf ^{(67.4KB, pdf)}

DOI: 10.21037/tau-2025-592