Liquid chromatography-mass spectrometry-based metabolic panels characteristic for patients with prostate cancer and prostate-specific antigen levels of 4–10 ng/mL

2.1 Study participants

In this retrospective cross-sectional study, we analyzed the data of 90 participants—30 patients with PCa, 30 with BPH, and 30 with prostatitis—who were admitted to Lishui Central Hospital (the Fifth Affiliated Hospital of Wenzhou Medical College) between August 2021 and December 2023. All participants underwent PSA screening, biopsy, transrectal ultrasound (TRUS), and digital rectal examination (DRE) following the National Comprehensive Cancer Network guidelines for PCa [14]. Written informed consent was obtained from all participants prior to their inclusion in the study. All participants were enrolled by a urologist based on clinical criteria for diagnosing BPH, prostatitis and PCa. PSA levels for all participants ranged 4–10 ng/mL. Prostate cancer was defined as follows: (i) nodules detected through DRE; (ii) hyper- or hypoechoic lesions seen in TRUS; and (iii) biopsy-confirmed PCa with a Gleason score of 6 or above. BPH was diagnosed after a negative DRE, TRUS and biopsy result. The diagnosis of prostatitis had to meet one of the following criteria: (1) infiltration of inflammatory cells within the prostate stroma or (2) infiltration of inflammatory cells in the area surrounding the prostate [15], with the infiltrate primarily comprising lymphocytes, along with plasma cells.

Participants were eligible for inclusion if they were above 50 years of age, had PSA levels between 4 and 10 ng/mL, and had not used 5α-reductase inhibitors or other medications affecting PSA levels in the past year. Additional criteria included no prior prostate biopsy, no previous diagnosis of PCa, no acute urinary tract infection or indwelling catheter within the past 2 weeks, and completion of mpMRI within 2 weeks of the prostate biopsy. Moreover, participants had to be willing to provide blood samples and sign written informed consent. Exclusion criteria included a history of other cancers, severe cardiovascular, hepatic, or renal diseases, psychiatric disorders, use of 5α-reductase inhibitors (e.g., finasteride) for more than three months, incomplete clinical records, or refusal to sign the informed consent.

2.2 Materials and reagents

Fisher Scientific (Fair Lawn, NJ, USA) supplied acetonitrile, methanol, and formic acid (LC-MS grade). Sigma-Aldrich (St. Louis, MO, USA) provided ammonium bicarbonate (LC-MS grade). Internal standards were furnished by Cambridge Isotope Laboratories, Inc. (Tewksbury, MA, USA), Toronto Research Chemicals (Toronto, Canada), and Avanti Polar Lipid, Inc. (Birmingham, AL, USA). The ultrapure water was generated using the Milli-Q water system (Merck KGaA, Darmstadt, Germany).

2.3 Sample collection and preparation

Following the standard operating procedure, blood samples and clinical characteristics were collected from participants. All participants fasted overnight, and their blood was drawn the next morning. To ensure complete coagulation, the samples were transferred into serum gel tubes, gently inverted five times, and left at room temperature for 60 minutes. The samples were then centrifuged at 4 °C for 10 minutes at 1500g to separate serum, which was aliquoted into cryovials and stored at −80 °C.

Polar metabolites from each serum sample were extracted and analyzed using two detection methods, both employing reverse-phase chromatography: technique 1 used positive ionization mode, while technique 2 used negative ionization mode. After thawing the samples at 4 °C, 100 µL of each sample was mixed with 400 µL of a methanol-acetonitrile solution containing isotope-labeled internal standards (ISs). The internal standards included: chenodeoxycholic acid-d4 (1.0 µg/mL), AcCa(12:0)-d9 (0.2 µg/mL), AcCa(18:0)-d3 (0.2 µg/mL), palmitic acid-13C16 (0.2 µg/mL), hippuric acid-d5 (1.0 µg/mL), stearic acid-d35 (1.0 µg/mL), Chlorophenylalanine (1.0 µg/mL), and taurine-d4 (1.0 µg/mL). The mixture was vortexed for 5 minutes and centrifuged at 13,000g for 15 minutes at 4 ℃. Two 200 µL aliquots of the supernatant were transferred to a CentriVap concentrator (centrifugal vacuum concentrator, Labconco Corporation, Kansas, MO, USA) and freeze-dried. The quality of sample preparation was assessed by calculating the relative standard deviation (RSD%) of the raw mass spectrometry response values of the internal standards across all samples, with results ranging from 1.07% to 16.37%, meeting the technical requirements for biological sample analysis. Equal portions of each sample extract were pooled to serve as analytical quality control (QC) samples, and all samples were prepared in parallel for testing.

To ensure analytical integrity, four QC samples were evenly distributed at the beginning, middle, and end of each data collection sequence. Quality control results were assessed using principal component analysis (PCA), time-series plotting of the first principal component, and Spearman correlation analysis between the initial and final QC samples. To rehydrate the freeze-dried samples, 120 µL of a 1:1 methanol/water solution was added to each sample.

2.4 Ultra-high-performance liquid chromatography high-resolution mass spectrometry (UHPLC-HRMS) analytical methods

An Ultimate™ 3000 UPLC system coupled with a Q Exactive™ Quadrupole-Orbitrap mass spectrometer (Thermo Scientific, San Jose, CA, USA) was used in the measurements of the metabolites. In method 1, metabolites were separated by an ACE C18-PFP column and then detected using positive electrospray. As mobile phases, water (A) and acetonitrile (B) with 0.1% formic acid were used. Over a period of 10 minutes, the linear gradient increased from 2% to 98% organic mobile phase. For method 2, the metabolome was analyzed in negative ion mode using an Acquity™ HSS C18 column provided by Waters Corporation (Milford, USA). A mixture of water and acetonitrile/methanol with 5 mM ammonium bicarbonate (NH₄HCO₃) was utilized as the mobile phase. A gradient was established to start at 2% organic phase, ramping up to 100% over 10 minutes, followed by an additional 5 minutes for column washing and equilibration. Both methods used the same flow rate (0.4 mL/min), injection volume (5 µL), and column temperature (50 °C).

Quadrupole-Orbitrap mass spectrometers functioned with identical ionization parameters except for the ionization voltage, using heated electrospray ionization (ESI) sources. The parameters included sheath gas at 45 arbitrary units (arb), auxiliary gas at 10 arb, heater temperature at 355 °C, capillary temperature at 320 °C, and S-Lens RF level at 55%. Using Automatic Gain Control (AGC) set to 1E6 and a max injection time of 200 ms, 70,000 full widths at half maximum resolution were applied to metabolome extracts in full scan mode. The scan range was 70–1000 m/z. QC samples were repeatedly injected to obtain the top 10 data-dependent MS2 spectra (full scan-ddMS2) for detailed metabolite structural annotation. A resolution setting of 17,500 Full Width at Half Maximum (FWHM) was used for the acquisition of full MS/MS data. Stepped normalized collision energy was used for the collision-induced dissociation of metabolites, with ultrapure nitrogen serving as the fragmentation gas.

2.5 Metabolomic data processing

Metabolic peaks were processed for component extraction using Compound Discoverer 2.1 software (Thermo Scientific, San Jose, CA, USA). Structural annotation was further achieved by comparing the acquired MS2 data with the proprietary iPhenome™ SMOL high-resolution MS/MS spectrum local library, generated with the mzCloud library, as well as authentic standards. Additionally, exact m/z values (±5 ppm) from MS1 spectra were matched against the Chemical metabolite database based on the Human Metabolome Database (HMDB) [16]. Separate manual inspections of the proposed identification results were required to eliminate false positives. The metabolomic data were normalized, trimmed, and log-transformed for further processing [17]. SIMCA-P software (version 14.1, Sartorius AG Umetrics, Goettingen, NI, Germany), which comprises orthogonal partial least squares discriminant analysis (OPLS-DA) and PCA, was used for multivariate analyses. Additionally, univariate analyses, such as independent sample t-tests and false discovery rate adjustments using the Benjamini-Hochberg method, as well as data visualization and Chemical Similarity Enrichment Analysis (ChemRICH) chemical enrichment analysis, were conducted on a proprietary R-based cloud computing platform, Integrated Pathway-Oriented Scoring (IPOS) [18]. Metabolic pathway analysis of differential metabolites was executed by the MetaboAnalyst platform, and metabolite-metabolite interaction networks were constructed [19].

2.6 LASSO feature selection and cross-validation for prostate cancer classification

We performed 5-fold cross-validation on two classification tasks: distinguishing between PCa and BPH and between prostate cancer and prostatitis. First, using differential metabolites as features, we applied LASSO regression for feature selection, incorporating L1 regularization to reduce redundant features and minimize the risk of overfitting. Additionally, we selected differential metabolites with an area under the curve (AUC) greater than 0.7 as input features. Next, in K-fold cross-validation, we trained and compared the performance of Support Vector Machine (SVM) and logistic regression models, evaluating metrics such as accuracy, precision, recall, F1-score, and receiving operating characteristic (ROC)-AUC. Finally, based on the cross-validation results, we selected the best-performing model and retrained it on the entire dataset filtered by LASSO to ensure robust prediction accuracy and generalizability in real-world applications.

2.7 Statistical analysis

Figures were created using R software 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria; available at http://www.R-project.org). Statistical significance was set at p < 0.05 or an adjusted p < 0.05. Differences between the two groups were evaluated using the non-parametric Mann-Whitney U test and the parametric Student’s t-test. The adjusted p-value was calculated using the Benjamini-Hochberg correction, and categorical variables were assessed using the chi-square test.

3.1 Clinical characteristics of the study participants

The clinical features of the 90 enrolled patients are summarized in Table 1. No significant differences were observed in age or PSA levels among the three clusters. However, significant differences were found in the ratio of free PSA to total PSA, prostate volume, PSA density, Prostate Imaging-Reporting and Data System (PI-RADS) score, and the proportion of positive findings from digital rectal examination (DRE), ultrasonography, and contrast-enhanced ultrasonography (Table 1). Specifically, the ratio of free-to-total PSA and prostate volume was lower in PCa patients than in those with prostatitis or BPH. In contrast, PSA density and PI-RADS scores were higher in PCa patients than in those with BPH or prostatitis. Additionally, the proportion of patients with positive findings from DRE, ultrasonography, or contrast-enhanced ultrasonography was higher in the PCa group compared to those with BPH or prostatitis.

3.2 Quality control

Analytical quality control was maintained using a pooled QC strategy. In the PCA, the QC samples clustered tightly (Fig. 1), indicating the experimental platform’s stable performance throughout the analytical process and the reliability of the test data. Additionally, the minimal variation in principal component 1 (PC1) for the QC samples further confirmed the stability of the analysis (Fig. 2). Spearman’s correlation analysis was performed between the first and last QC data matrices, with the results shown in a Spearman correlation scatter plot (Fig. 3). Although the correlation coefficients were smallest between these two groups of QC samples, the lowest correlation was still 0.998, highlighting the excellent data quality. Furthermore, isotope-labeled internal standards were added to the extracts, and sample pretreatment was evaluated by the RSD% of the raw mass spectral response of the internal standards in each sample. The RSD% ranged from 1.07% to 16.37%, which meets the technical requirements for biological sample analyses.

Fig. 1.QC samples clustered tightly in the unsupervised PCA score plot, indicating stable analytical performance throughout the batch and confirming the high quality of the metabolomic data. Blue dots: QC pooled batch QC samples; red dots: samples tested in this project. Pareto scaling for PCA. PCA, principal component analysis; QC, quality control.

Fig. 2.Time series plot of principal component 1 during the analytical batch is shown, with PCA analysis performed using Pareto scaling. PCA, principal component analysis; PC1, principal component 1; std. dev, standard deviation.

Fig. 3.First and last QC samples in the analytical batch showed a high correlation in Spearman correlation analysis, confirming the high quality of the acquired untargeted metabolomic data. QC, quality control.

3.3 Metabolomic profiling differences between PCa and BPH

PCA analysis showed no significant differences in the metabolic profiles between PCa and benign prostatic hyperplasia (BPH) (Fig. 4A). However, the OPLS-DA score scatter plot clearly distinguished PCa from BPH (Fig. 4B), with R²Y and Q² values of 0.728 and 0.0138, respectively, indicating the model’s reliability. Through OPLS-DA, metabolites with Variable Importance in Projection (VIP) scores greater than 1.0 and p-values less than 0.05 were identified as differential metabolites. After excluding exogenous metabolites, 12 key metabolites that distinguish PCa from BPH were identified using the LASSO regression algorithm (Table 2), with six detected in ESI+ mode and six in ESI− mode.

Fig. 4.PCA and OPLS-DA of LC-MS metabolite profiles of patients with PCa or BPH. (A) The PCA model’s score scatter plot for patients with PCa and BPH is displayed. The X-axis represents the first principal component (t[1]), and the Y-axis represents the second principal component (t[2]). (B) The OPLS-DA model’s score scatter plot for patients with PCa and BPH is displayed. The X-axis represents the predictive direction (t[1]), and the Y-axis represents the orthogonal direction (to[1]). Abbreviations: BPH, benign prostatic hyperplasia; LC-MS, liquid chromatography-mass spectrometry; PCa, prostate cancer; PCA, principal component analysis; OPLS-DA, orthogonal projections to latent structures-discriminate analysis.

In ESI+ mode, levels of hexenoylcarnitine (AcCa(6:1)) and lysophosphatidylcholine (LysoPC(20:5/0:0)) were significantly higher in PCa patients than in BPH patients, while phytosphingosine levels were lower. In ESI− mode, levels of (4E,15Z)-bilirubin IXa and bilirubin were elevated in PCa patients compared to BPH patients, while levels of C-mannosyl leucine and 3-hydroxymethylglutaric acid were significantly lower in PCa patients than in BPH patients. Overall, these results reveal significant metabolic differences between PCa and BPH.

Fig. 5 presents the ROC analysis results, demonstrating that these five key metabolites effectively differentiate BPH from PCa, with AUC values greater than 0.7 and p-values less than 0.05. Specifically, phytosphingosine (AUC = 0.703, sensitivity = 73.3%, specificity = 63.3%), LysoPC(0:0/20:4) (AUC = 0.722, sensitivity = 66.7%, specificity = 70.0%), (4E,15Z)-bilirubin IXa (AUC = 0.724, sensitivity = 86.7%, specificity = 53.3%), C-mannosyl leucine (AUC = 0.712, sensitivity = 73.3%, specificity = 63.3%), and dihydrothymine (AUC = 0.729, sensitivity = 70.0%, specificity = 70.0%) showed excellent performance in differentiating PCa from BPH.

Fig. 5.ROC curve analysis of the selected essential metabolites for distinguishing PCa from BPH. Assessment of the diagnostic performance of phytosphingosine (A), LysoPC (0:0/20:4) (B), (4E,15Z)-bilirubin IXa (C), C-mannosyl leucine (D), dihydrothymine (E), and the combination of these six metabolites (F). Abbreviations: AUC, area under the ROC curve; BPH, benign prostatic hyperplasia; PCa, prostate cancer; ROC, receiver operating characteristic; LysoPC, lysophosphatidylcholine.

In 5-fold cross-validation, we trained and compared the performance of SVM and logistic regression models, ultimately selecting logistic regression, which had an average ROC-AUC of 0.78. The final model achieved an AUC of 0.852, with a sensitivity of 80.0% and a specificity of 83.3%. These findings suggest that the metabolic panel of differential metabolites and their combination have strong diagnostic potential for distinguishing BPH from PCa.

3.4 Metabolomic profiling differences between prostatitis and PCa

PCA analysis showed no significant difference in sample distribution trends between patients with prostatitis and those with prostate cancer (PCa) (Fig. 6A). However, OPLS-DA clearly demonstrated a significant distinction between prostatitis and PCa patients (Fig. 6B). The R²Y and Q² values for the OPLS-DA model were 0.709 and 0.166, respectively. VIP >1 and p < 0.05 were used to identify differentiating metabolites. After excluding exogenous metabolites, we used the LASSO regression algorithm to select 19 key metabolites (Table 3), including seven detected in ESI+ mode and 12 in ESI− mode. In ESI+ mode, N-acetylcadaverine levels were lower in prostatitis patients than in PCa patients. In ESI− mode, glycohyocholic acid and oxalic acid levels were higher in prostatitis patients, while palmitoleic acid (Free Fatty Acid (FFA(16:1))) levels were lower.

Fig. 6.PCA and OPLS-DA analyses of LC-MS metabolite profiles from patients with prostatitis and PCa. (A) Score scatter plot of the PCA model for patients with prostatitis and PCa. The X-axis represents the first principal component (t[1]), and the Y-axis represents the second principal component (t[2]). (B) Score scatter plot of the OPLS-DA model for patients with prostatitis and PCa. The X-axis represents the predictive direction (t[1]), and the Y-axis represents the orthogonal direction (to[1]). Abbreviations: LC-MS, liquid chromatography-mass spectrometry; PCa, prostate cancer; PCA, principal component analysis; OPLS-DA, orthogonal projections to latent structures-discriminate analysis.

Fig. 7 shows the ROC analysis results, indicating that the differentiation between prostatitis and PCa had an AUC greater than 0.7 and p-values less than 0.05. The eight key metabolites identified included pregnenolone sulfate (AUC = 0.724, sensitivity = 76.7%, specificity = 63.3%), bilirubin (AUC = 0.779, sensitivity = 76.7%, specificity = 76.7%), lysophosphatidylethanolamine (LysoPE(0:0/22:5n6)) + LysoPE(22:5n3/0:0) (AUC = 0.721, sensitivity = 63.3%, specificity = 70.0%), 5-hydroxyindole sulfate (AUC = 0.712, sensitivity = 66.7%, specificity = 80.0%), N-acetylcadaverine (AUC = 0.722, sensitivity = 50.0%, specificity = 83.3%), LysoPC(20:5/0:0) (AUC = 0.743, sensitivity = 73.3%, specificity = 76.7%), dihydrothymine (AUC = 0.749, sensitivity = 70.0%, specificity = 73.3%), and LysoPC(0:0/20:5) (AUC = 0.707, sensitivity = 63.3%, specificity = 76.7%). In 5-fold cross-validation, we trained and compared the performance of SVM and logistic regression models, ultimately selecting the SVM model, which achieved an average ROC-AUC of 0.867. The final model had an AUC of 0.891, with a sensitivity of 76.7% and a specificity of 90.0%. These results indicate that the metabolic panel of differential metabolites and their combinations have strong diagnostic potential for distinguishing between prostatitis and PCa patients.

Fig. 7.Displayed is the ROC curve analysis of the selected essential metabolites for distinguishing prostatitis from PCa. Assessment of the diagnostic performance of pregnenolone sulfate (A), bilirubin (B), LysoPE(0:0/22:5n6) + LysoPE(22:5n3/0:0) (C), 5-hydroxyindole sulfate (D), N-acetylcadaverine (E), LysoPC(20:5/0:0) (F), dihydrothymine (G), LysoPC(0:0/20:5) (H), and the combination of these eight metabolites (I). Abbreviations: AUC, area under the ROC curve; BPH, benign prostatic hyperplasia; PCa, prostate cancer; ROC, receiver operating characteristic; LysoPE, lysophosphatidylethanolamine; LysoPC, lysophosphatidylcholine.

3.5 Metabolic pathway analysis of differential metabolites and construction of the metabolite-metabolite interaction network

Class analysis of the chemical properties of metabolites that differed between groups was conducted based on the structural annotation results from untargeted metabolomics with in-depth coverage (Fig. 8). A total of 712 metabolites were identified across 21 major classes, with 120 fatty acid species, accounting for 16.8% of the total. Metabolic phenotypes were then differentiated using PCA and OPLS-DA. Of the 109 differential metabolites identified in comparisons between prostatitis and PCa, fatty acid species were the most abundant, with 30 species. Fatty acid metabolites represented the largest proportion of all differential metabolites when comparing PCa and prostatitis patients. To explore the pathway-based metabolic features, we performed a metabolic pathway analysis on the differential metabolites (Fig. 9). Our findings suggest that several metabolic pathways, including the biosynthesis of unsaturated fatty acids, play key roles in PCa. Additionally, the metabolite-metabolite interaction network (Fig. 10) highlighted hypoxanthine, oxalic acid, and succinic acid as central nodes for potential functional relationships among a wide range of annotated metabolites.

Fig. 8.Classification and quantification of significantly different metabolites (p < 0.05). (A) Differential metabolites in the comparison between PCa and BPH. (B) Differential metabolites in the comparison between PCa and prostatitis. The red boxes indicate that fatty acid metabolites represent a significantly larger proportion in both comparisons. Abbreviations: BPH, benign prostatic hyperplasia; PCa, prostate cancer; LysoPC, lysophosphatidylcholine; LysoPI, lysophosphatidylinositol; LysoPE, lysophosphatidylethanolamine; LysoPA, lysophosphatidic acid.

Fig. 9.The size of the bubbles is indicative of the pathway impacts, while the depth of the bubbles corresponds to the p value. (A) Pathway analysis with MetaboAnalyst of metabolites differing between PCa and BPH. (B) Pathway analysis with MetaboAnalyst of metabolites differing between PCa and prostatitis. Abbreviations: BPH, benign prostatic hyperplasia; PCa, prostate cancer.

Fig. 10.Construction of the metabolite interaction network reveals the metabolic processes in prostate cancer, highlighting the interactions between metabolites such as succinic acid and hypoxanthine. FAD, flavin adenine dinucleotide; NADP, nicotinamide adenine dinucleotide phosphate; NADH, nicotinamide adenine dinucleotide (reduced); ADP, adenosine diphosphate.