Untargeted metabolomic profiling of serum and urine in kidney cancer: a non-invasive approach for biomarker discovery

Abstract

Introduction

Kidney cancer (KC) is a significant global health burden. Early diagnosis remains challenging due to the limited sensitivity and specificity of existing biomarkers. Metabolomics enables the detection of disease-specific metabolic alterations, offering potential for improved non-invasive biomarker discovery.

Objectives

This study aims to characterize metabolic signatures distinguishing KC patients from non-cancer controls and evaluate the diagnostic potential of annotated metabolites in serum and urine.

Methods

An untargeted metabolomic analysis was performed on serum and urine samples from 56 KC patients and 200 controls using ultra-high-resolution mass spectrometry coupled with ultra-high-performance liquid chromatography (UHPLC-UHRMS in both positive and negative ionization modes with vacuum insulated probe heated electrospray ionization (VIP-HESI)). Samples were collected from the same individuals, which helped minimize inter-individual variability and enabled cross-biofluid comparison of metabolic profiles. Multivariate statistical techniques were applied to detect metabolic differences, including principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA). An external validation strategy using training and validation subsets was employed to assess the robustness of candidate metabolite biomarkers matched in the discovery dataset.

Results

Distinct metabolic signatures were observed between KC patients and controls, with key metabolic pathways involving lipid metabolism, amino acid biosynthesis, and glycerophospholipid metabolism. 19 serum and 12 urine metabolites showed high diagnostic potential (AUC > 0.90), demonstrating strong sensitivity and specificity.

Conclusion

These findings support the application of metabolomics for RCC detection and highlight the metabolic alterations associated with kidney cancer. Further validation in larger cohorts is necessary to confirm the clinical utility of these potential biomarkers.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11306-025-02294-4.

Introduction

Kidney cancer (KC) constitutes approximately 2.2% of all cancers globally and ranks as the third most common malignancy of the urinary tract, following prostate and bladder cancer (Bray et al., 2024). In 2022, more than 434,000 new KC cases were diagnosed worldwide, and 155,702 deaths were documented (Bray et al., 2024).

KC is a heterogeneous group of malignancies, with renal cell carcinoma (RCC) as the most common type, arising from renal tubular epithelial cells and accounting for ~ 90% of cases. RCC includes several histopathological subtypes, most notably clear cell RCC (ccRCC), which represents ~ 80% of cases, followed by papillary (pRCC) and chromophobe RCC (chRCC), each with distinct molecular and histological features (Young et al., 2024). Rare subtypes include aggressive collecting duct carcinoma (CDC) and low-grade tubulocystic RCC (tcRCC). Benign tumors, such as oncocytoma (OCT) and angiomyolipoma (AML), must also be differentiated due to differing clinical implications. Prognosis strongly depends on tumor subtype, grade, and stage (Hancock & Georgiades, 2016).

Early detection is a major focus in KC research, as diagnosing the disease at an early stage greatly improves outcomes. Stage I cases have a 5-year survival rate of 88% compared to just 14% for stage IV (Miller et al., 2022). About 60% of KC cases are found incidentally through abdominal imaging such as computed tomography (CT), ultrasound, or magnetic resonance imaging (MRI) (Cirillo et al., 2024). The classic triad of flank pain, hematuria, and a palpable mass is rare and usually indicates advanced disease. RCC is notably resistant to both radiation and many drugs (Da Silva Prade et al., 2023). The most effective treatment for localized RCC is radical nephrectomy, with nephron-sparing surgery preferred at early stages. However, 30 to 70% of patients experience recurrence after surgery, and approximately one-third are diagnosed with metastatic disease at the time of presentation (Xiang et al., 2022).

Various protein-based biomarkers have been proposed to improve the prognosis of metastatic RCC patients (Li et al., 2021; Sanguedolce et al., 2024). Although many candidates have been proposed, their low sensitivity and specificity limit their clinical applicability. Currently, no validated biomarkers are available to reliably guide diagnosis, predict treatment response, or distinguish between RCC subtypes. Therefore, further research is essential to discover and validate novel biomarkers that could support early detection, patient stratification, monitoring of disease progression, and personalized therapy (Farber et al., 2017).

Metabolomics has emerged as a powerful tool for biomarker discovery in KC, offering insight into real-time biochemical changes associated with tumor development and progression (Johnson et al. 2016). Untargeted profiling captures global metabolic shifts, revealing disease-specific signatures that may improve early diagnosis, prognosis, and treatment strategies (Dinges et al. 2019; M. S. Monteiro et al. 2014).

Over the past decade, highly sensitive methods such as gas chromatography–mass spectrometry (GC-MS) (Kordalewska et al. 2022; M. Monteiro et al. 2017) liquid chromatography–mass spectrometry (LC-MS) (Nizioł et al. 2018; Sato et al. 2020), laser desorption/ionization MS (LDI-MS) (Arendowski et al. 2020a, 2020b), mass spectrometry imaging (MSI) (Arendowski, Nizioł, Ossoliński, et al. 2018; Nizioł, Copié, et al. 2021; Nizioł, Sunner, et al., 2020; Płaza et al. 2022), and proton nuclear magnetic resonance (¹H NMR) (Kordalewska et al. 2022; Nizioł, Ossoliński, et al. 2020; Nizioł, Copié, et al. 2021) have been employed to discover KC biomarkers (Kordalewska et al. 2022; Nizioł et al. 2020a, 2020b). LC-MS, compared to NMR, detects a much more compounds of molecules with much higher sensitivity, resolution, and accuracy, using much less material (Pan & Raftery 2007).

KC metabolomics has focused mainly on serum (Deja et al. 2021; O. S. Falegan et al., 2017; Wang et al. 2022; F. Zhang et al. 2017), urine (Bifarin et al. 2021; O. Falegan et al., 2017; Liu et al. 2019), and renal tissue (Hájek et al. 2018; Jing et al. 2019; Nizioł, Copié, et al. 2021) samples. Urine and serum are more accessible than tissue, with urine offering a non-invasive alternative despite variability due to dilution.

One of the first applications of high-performance liquid chromatography–mass spectrometry (HPLC-MS) in RCC serum metabolomics appeared in 2011, using both reversed-phase (RP) and hydrophilic interaction chromatography (HILIC) combined with multivariate analysis (Lin et al. 2011). The study included 33 RCC patients and 25 healthy controls, demonstrating the potential of metabolomics as a noninvasive diagnostic tool. Later, a study based on 42 RCC patients and 66 healthy individuals proposed four serum biomarkers with prognostic relevance (F. Zhang et al. 2017). In 2021 Manzi et al. (2021), a lipidomic analysis using ultra-performance liquid chromatography quadrupole time-of-flight mass spectrometry (UPLC-QToF-MS) on 112 RCC patients and 52 controls led to the development of two lipid panels: one for detection (95.7% accuracy) and one for staging (82.1%). In 2022 Wang et al. (2022), another serum-based study analyzed 86 RCC patients and 45 controls with UPLC-MS, detecting 240 RCC-specific metabolites and 64 capable of distinguishing RCC from benign tumors, including four highly relevant diagnostic markers.

Urinary metabolomics has also been widely explored as a non-invasive approach. A 2011 untargeted profiling study reported quinolinate, 4-hydroxybenzoate, and gentisate as metabolites associated with altered amino acid and energy metabolism, reflecting tumor activity (Kim et al. 2011). Targeted analysis using high-performance liquid chromatography tandem mass spectrometry (HPLC-MS/MS) later enabled simultaneous detection and quantification of urinary RCC biomarkers (Chen et al. 2015). In a larger study involving 100 RCC patients, 34 patients with benign tumors, and 129 healthy controls, a nine-metabolite panel showed strong discriminatory power for RCC (Liu et al. 2019). Additional urinary studies conducted in 2020 supported these findings, though with smaller sample sizes (Oto et al. 2020; Sato et al. 2020; M. Zhang et al. 2020). In 2022, two separate experiments analyzed urine from 30 RCC and 38 ccRCC patients, further validating urine metabolomics as a promising diagnostic tool (Kordalewska et al. 2022).

Collectively, these studies show steady progress in metabolomic profiling of serum and urine for RCC detection, staging, and prognosis. While standardization and clinical validation are still needed, current evidence strongly supports the value of metabolomics as a non-invasive approach for improving RCC diagnosis and patient management.

In this study, we performed untargeted metabolomic profiling of serum and urine samples collected from the same KC patients, using ultra-high-performance liquid chromatography coupled with ultra-high-resolution mass spectrometry (UHPLC-UHRMS) and an external validation strategy based on training and validation sets. Analyses were conducted in both positive and negative ion modes, using a highly sensitive vacuum insulated probe heated electrospray ionization (VIP-HESI) source. This integrated ion source improves desolvation efficiency, enhances ionization stability, and provides higher sensitivity compared to conventional ESI or HESI sources, allowing for the detection of a broader range of metabolites, including low-abundance and less readily ionizable compounds. The VIP-HESI ion source was used in this work as it is currently the most efficient solution for metabolomics compatible with Bruker’s mass spectrometer, according to manufacturer’s tests and independent reports (Midha et al., 2023). This setup enabled intra-individual comparison of systemic and renal-localized metabolic alterations, minimizing inter-individual variability. Our goal was to discover clinically relevant, non-invasive biomarkers for early KC detection by comparing the metabolic profiles of ~ 60 KC patients and 200 non-cancer controls. Using both unsupervised and supervised multivariate techniques on the training and validation sets, we annotated distinct metabolic signatures that differentiate KC from controls with high confidence and strong predictive performance.

Materials and methods

All chemicals used were of LC-MS or analytical reagent grade. Locally produced deionized water (18 MΩ cm) was utilized. Methanol of LC-MS grade was purchased from Sigma Aldrich (St. Louis, MO, USA).

Collection of human urine and serum samples

Serum and urine samples were collected from patients at John Paul II Hospital in Kolbuszowa, Poland. The study population had an average age of 74 and was of white ethnicity. A total of 56 patients diagnosed with kidney cancer provided both blood and urine samples, while an additional three patients contributed only urine samples. In total, 56 serum samples and 59 urine samples were analyzed. The control group consisted of age- and sex-matched individuals admitted to the Urology Department for surgical treatment of benign urological conditions, including benign prostatic hyperplasia (BPH), urinary stones, phimosis, ureteropelvic junction obstruction (UPJO), and stress urinary incontinence. In total, 197 urine samples and 200 serum samples were collected from control patients for analysis. Urine samples were collected as spot urine samples. All samples (both urine and blood) were obtained one day before surgery, as part of the preoperative evaluation, in the morning and in a fasting state, to ensure standardization and reduce circadian and hydration-related variability. No pre-analytical normalisation to creatinine, osmolality, or specific gravity was applied to minimise processing steps and avoid additional manipulation of the samples. As part of the standard preoperative protocol, each patient underwent a comprehensive laboratory assessment the day before surgery, including a complete blood count, electrolyte analysis, coagulation panel, creatinine measurement, glomerular filtration rate (GFR) assessment, urinalysis, lung X-ray, and abdominal ultrasound. Following thorough clinical evaluation and laboratory testing, all cancer patients underwent transurethral resection of bladder tumor (TURBT). The study was approved by the local Bioethics Committee of the University of Rzeszów (Poland, Permit Number 2018/04/10) and adhered to all relevant ethical guidelines and regulations. Written informed consent was obtained from all participants after they were provided with detailed information about the study’s objectives and methodology. The control group (NCs) included patients hospitalized in the Urology Department for benign urological conditions such as urolithiasis, benign prostate hyperplasia, testicular hydrocele, varicocele, phimosis, ureteropelvic junction stenosis, urinary incontinence, and urethral stricture. To enhance clinical relevance, the control cohort was composed of patients with benign urological conditions representing common differential diagnoses in routine urological practice. This design reflects real-world diagnostic settings and supports the translational applicability of the matched candidate biomarkers. To exclude the presence of malignancies, all control patients underwent at least one abdominal ultrasound, with those diagnosed with urolithiasis often undergoing a CT scan. In addition, they underwent routine laboratory tests required for urological surgery to rule out inflammation. Control group patients provided written consent to donate leftover urine samples for research purposes after being informed about the study. Each participant provided 10 ml of urine, which was centrifuged at 3000 rpm for 10 min at room temperature. Approximately 2.6 ml of blood was drawn from each subject and centrifuged under the same conditions. The separated serum was then stored at − 80 °C until further analysis. The clinical characteristics of the patients are detailed in Supplementary Information 1, Table S1.

Sample preparation

As outlined in our recent study, metabolites with medium-to-high polarity were extracted from urine and serum samples (Ossoliński et al., 2022, 2023). In brief, the samples were thawed at 4 °C and centrifuged at 12,000×g for 5 min at the same temperature. To each 200 µL aliquot of the resulting supernatant, 600 µL of acetone was added. The mixture was then vortexed for 1 min, incubated at room temperature for 20 min, followed by an additional 20-minute incubation at − 20 °C, and subsequently centrifuged at 6000×g for 5 min at 4 °C. A supernatant volume of 550 µL (serum) or 750 µL (urine) was collected and vacuum-dried using a SpeedVac concentrator. The dried residues were reconstituted in 130 µL (serum) or 900 µL (urine) of methanol, vortexed, and centrifuged again (9000×g for 5 min). Finally, 100 µL of the processed serum supernatant was transferred into an HPLC vial insert (130 µL capacity), while 800 µL of the processed urine supernatant was transferred into an HPLC vial (1500 µL capacity). Pooled quality control (QC) samples were prepared separately for urine and serum matrices by combining equal aliquots from representative samples within each biological group (cancer and control). All prepared samples were placed in the Elute autosampler for further analysis.

Instrumentation

Untargeted analysis was conducted using a Bruker Elute UHPLC system operated with Hystar 3.3 software, coupled with an ultra-high-resolution Bruker Impact II mass spectrometer (60000 + resolution version; Bruker Daltonik GmbH, Bremen, Germany). The system featured an ESI QToF-MS, utilizing Data Analysis™ v.4.2 (Bruker Daltonik GmbH) and MetaboScape^® (v.2022b, Bruker Daltonics GmbH) for data processing. Metabolite separation was performed using a gradient of mobile phases on a Waters UPLC ACQUITY BEH column (C18 silica, 1.7 μm particles, 50 × 2.1 mm) with a compatible column guard. Additional methodological details can be found in our previous publication (Nizioł et al., 2022, 2023) and Supplementary Information 1 (Section S1-3), where we describe data processing, filtering, quality control procedures, and system stability monitoring. All procedures were performed in accordance with established guidelines for untargeted metabolomics workflows (Beger et al., 2019; Broadhurst et al., 2018; Dudzik et al., 2018; Kirwan et al., 2022).

Multivariate statistical analysis

Before statistical analysis, the data for urine and serum extracts were separately divided into training and validation sets. A total of 30% of the data was allocated for validation, while the remaining 70% was used for training. The selection of the validation subset was performed proportionally to key data characteristics, including patient sex, age, cancer stage, and lesion type, ensuring a balanced distribution. Statistical analysis was conducted separately on the four datasets. The training set was used to select diagnostic urine and serum discriminant metabolite feature distinguishing cancer patients from the control group. In contrast, the validation set was employed to independently assess the diagnostic performance of the selected urine and serum discriminating metabolite features.

Statistical analysis was conducted using MetaboAnalyst 6.0 (Pang et al., 2024), an online platform for metabolomics data processing. Within MetaboAnalyst, data were log-transformed and auto-scaled before analysis to ensure normalization. The resulting metabolite profiles were initially examined using unsupervised Principal Component Analysis (PCA) to explore patterns and variations. The separation observed in the PCA score plots between kidney cancer (KC) patients and non-cancer (NC) controls was further evaluated using supervised multivariate statistical methods, specifically Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA). To assess the quality of the OPLS-DA models, both the goodness of fit (R²Y) and predictive capability (Q²) were evaluated. Variable Importance in Projection (VIP) plots were generated to select metabolites contributing most significantly to group separation, with VIP values above 1.0 considered as potential biomarker candidates. To assess whether the observed class separation was statistically significant and not due to overfitting or random chance, 2000-fold permutation tests were performed (P value < 0.05 threshold). For individual metabolite comparisons, the non-parametric Wilcoxon rank-sum test (Mann–Whitney U test) with False Discovery Rate (FDR) correction was applied. Differences were considered statistically significant if both the raw P value and FDR-adjusted P value were below 0.05. The diagnostic performance of selected metabolites was assessed through Receiver Operating Characteristic (ROC) curve analysis and random forest modeling. The area under the curve (AUC), along with the 95% confidence interval, specificity, and sensitivity, was calculated to determine the reliability of each metabolite as a potential biomarker. AUC values were interpreted as follows: > 0.9—highly reliable biomarker, 0.7–0.9—moderately reliable, 0.5–0.7—low reliability, 0.5—equivalent to random chance. Only metabolites with AUC > 0.70 were considered meaningful. To enhance robustness and reduce the risk of false positives, we applied a two-step discovery/replication strategy. Independent multivariate models (OPLS-DA) were built separately on the training and validation sets. Candidate biomarkers were selected within each subset using consistent statistical criteria (VIP > 1.0, P < 0.05, FDR-adjusted, AUC > 0.7, FC > 2 or < 0.5), and only those metabolites that met all thresholds in both subsets were retained. Additionally, a metabolic pathway impact analysis was conducted using MetaboAnalyst 6.0 to select biochemical pathways associated with kidney cancer. This analysis utilized the Kyoto Encyclopedia of Genes and Genomes (KEGG) and the Small Molecule Pathway Database (SMPD) pathway library for Homo sapiens. Each affected pathway was evaluated based on raw P values, Holm-adjusted P values (Holm–Bonferroni correction), and FDR-adjusted P values, computed via pathway topology analysis. Furthermore, enrichment analysis was performed at the sub-class level of chemical compounds, categorizing the annotated metabolites into sub-chemical classes, to assess their distribution within specific structural classifications.

Results

Distinguishing between kidney cancer and control serum samples

A total of 14,154 m/z features were detected in each serum sample under positive ion mode, of which 1,468 were annotated as specific chemical compounds, based on database and spectral matching. Similarly, 5978 m/z features were detected in negative ion mode, with 725 compounds successfully annotated. For further analysis, data from both positive and negative ion modes were combined to provide a comprehensive metabolic profile. Unsupervised 2D Principal Component Analysis (PCA) score plots for both datasets effectively distinguished cancer patients from controls.

In the training set, Principal Components 1 and 2 (PC1 and PC2) accounted for 11.5% and 6.1% of the total variance, respectively, providing the best separation between groups. Within the central 95% confidence region, only a few outliers were observed (Fig. 1A). Similarly, in the validation set, PC1 (12.9%) and PC2 (6.7%) demonstrated the strongest differentiation between cancer and control serum samples (Fig. S1A, Supplementary Information 1).

Fig. 1.

Metabolomic analysis of serum samples from KC and NC groups in the training set. A PCA and B OPLS-DA score plots distinguishing tumor (violet) and control (orange) serum samples. C Receiver operating characteristic (ROC) curves illustrating the diagnostic performance. D–G Box-and-whisker plots of selected metabolites differentiating KC and control samples. Y-axis values represent MS signal intensities after log₁₀ transformation and autoscaling (z-score normalization). ***P value < 0.001. PC principal component, AUC area under the curve, CI confidence interval

To explore metabolic differences between the KC group and NC, a supervised multivariate OPLS-DA analysis was conducted. The score plot from the training set demonstrated a clear separation between the two groups (Fig. 1B). To test whether the model’s discrimination between cancer and control samples could have arisen by chance (i.e., overfitting), we performed 2000 permutation tests (Table S2, Supplementary Information). The low P values (Q² = 0.960, R²Y = 0.989, P value = 5E−04 (0/2000)) confirmed that the model performance was unlikely to be due to random classification. These results highlight significant metabolic differences between cancer and control serum samples. The high R²Y and Q² values indicate that the OPLS-DA model possesses strong interpretability and predictive power. A similar trend was observed in the validation dataset, where the OPLS-DA model effectively distinguished KC patients from NCs (Table S2). The observed separation was statistically validated by 2,000 permutation tests (Q² = 0.919, R²Y = 0.989, P value = 5E−04 (0/2000)), indicating that the model’s performance was unlikely to result from random chance.

The VIP plot generated from the OPLS-DA model was used to select potential serum biomarkers for KC. To further evaluate the diagnostic performance of these discriminating metabolite features, univariate ROC analysis was conducted on both the training and validation datasets. The area under the ROC curve (AUC), a key metric for assessing model performance, was used to determine the sensitivity and specificity of each potential biomarker. Only m/z features with an AUC > 0.70 were considered relevant. To ensure robust biomarker candidate selection, candidates were selected based on a combination of VIP values (> 1.0), AUC scores (> 0.7), and independent Wilcoxon rank-sum tests with FDR correction (P value and FDR < 0.05). For robust discriminating metabolite features selection, we first assigned significant metabolites in the training set and independently verified these in the validation set. Only those meeting significance criteria in both datasets were retained as potential biomarkers. Ultimately, 103 m/z features common to both datasets were retained, each of which was assigned to a specific chemical compound (Table 1, Supplementary Information 2).

Table 1.

Differential serum metabolites distinguishing KC patients from NCs, selected based on statistical significance (P value < 0.001, FDR < 0.001), variable importance (VIP > 1), fold change (FC < 0.5 or > 2), and strong diagnostic performance (AUC > 0.9)

No.	Name	Structure	m/za	RT [s]	VIPb	FCc	P valued	FDRe	AUC	Spec. [%]e	Sens. [%]e
1	Glycerol 1-stearatef,h,i	C21H42O4	359.3156	329	2.68	4.91	5.02E−21	5.24E−19	0.998	100	99
2	2-Hydroxy-2-methylbutyric acidf,g,h	C5H10O3	160.0968	79	2.32	0.30	9.19E−21	8.77E−19	0.995	97	98
3	Leucinef,i	C9H17NO3	188.1284	103	2.60	0.23	1.65E−20	1.29E−18	0.992	100	96
4	Octadecanedioic acidf,g	C18H34O4	356.2800	242	2.72	3.84	5.71E−20	3.91E−18	0.985	97	99
5	2-Propylphenolf,h,i	C9H12O	137.0961	186	2.22	4.24	1.03E−19	5.64E−18	0.981	97	93
6	Myristic acidf,g,h	C14H28O2	270.2429	273	1.86	2.61	1.13E−19	6.07E−18	0.981	100	95
7	SM(d33:1)f,h,i	C38H77N2O6P	689.5592	315	2.10	8.42	8.91E−19	3.68E−17	0.969	87	96
8	LPE(16:0)f,h,i	C21H44NO7P	452.2779	290	1.81	6.76	9.31E−19	3.78E−17	0.966	95	96
9	Benzoic acidf,g,h	C7H6O2	121.0295	131	2.46	3.12	1.90E−18	6.96E−17	0.964	97	93
10	Methyl 2-methoxybenzoatef,h,i	C9H10O3	165.0557	179	2.49	3.25	2.87E−18	9.82E−17	0.962	100	93
11	LPI(18:1)f,h,i	C27H51O12P	597.3036	377	1.38	4.11	4.80E−18	1.53E−16	0.958	92	91
12	4-Hydroxybutanoic acid lactonef,i	C4H6O2	87.0441	441	1.86	3.25	2.12E−17	5.68E−16	0.949	92	92
13	4-Deoxytetronic acidf,g	C4H6O2	85.0298	53	2.01	2.04	3.08E−17	8.15E−16	0.947	87	92
14	2-(5-Oxovaleryl)phosphatidylcholinef,h,i	C29H56NO9P	594.3765	277	2.25	5.23	1.35E−16	3.17E−15	0.938	90	85
15	2,2-Dimethylsuccinic acidf,g,h	C6H10O4	145.0507	109	2.24	4.00	2.04E−16	4.57E−15	0.935	90	88
16	Succinic acidf,g	C4H6O4	141.0160	25	1.89	2.10	2.56E−15	4.79E−14	0.919	90	94
17	LPC(16:0/0:0)f,h,i	C24H50NO7P	496.3398	379	2.36	3.20	1.14E−14	1.91E−13	0.909	92	88
18	2-Hydroxyglutaric acidf,g	C5H8O5	171.0265	25	2.18	2.77	2.51E−14	3.84E−13	0.904	90	91
19	SM(d41:1)f,h,i	C46H93N2O6P	801.6836	201	1.08	2.41	3.25E−14	4.82E−13	0.902	90	88

AUC area under the curve, FC fold change, FDR false discovery rate, m/z mass-to-charge ratio, RT retention time, Sens. Sensitivity, Spec. specificity, VIP variable influence on projection

^aExperimental monoisotopic mass of ion; ^bVIP scores derived from OPLS-DA model; ^cfold change between cancer and control serum calculated from the abundance mean values for each group—cancer-to-normal ratio; ^dvalues determined from non-parametric t-test; ^eROC curve analysis for individual biomarkers; ^fthe metabolites annotated by high precursor mass accuracy; ^gthe metabolites annotated by matching retention time; ^hthe metabolites annotated by matching isotopic pattern; ⁱthe metabolites annotated by matching MS/MS fragment spectra

In serum samples, the results demonstrated that 19 out of the previously selected 103 metabolites exhibited exceptionally high AUC values (> 0.9), with specificity and sensitivity exceeding 87% and 85%, respectively, in both the training and validation sets (Tables 1 and S1, Supplementary Information 2). The combined mass features from both datasets served as a highly effective discriminator between KC and NC serum samples (AUC > 0.991), as illustrated in Figs. 1C and S1C.

Distinguishing between kidney cancer and control urine samples

In positive ionization mode, a total of 8,590 m/z features were detected in each urine sample, of which 1,129 were annotated as specific chemical compounds. Unsupervised PCA effectively distinguished urine samples from KC patients and healthy controls. In the training set, PC 1 and 2 accounted for 13.7% and 8.2% of the total variance, respectively, providing optimal separation between the groups. Within the 95% central confidence region, only a few outliers were observed (Fig. 2A). Similarly, in the validation set, PC1 (20.9%) and PC2 (8.6%) demonstrated the most significant differentiation between KC and control urine samples (Fig. S2A, Supplementary Information 1).

Fig. 2.

Metabolomic analysis of urine samples from KC and NC groups in the training set. A PCA and B OPLS-DA score plots distinguishing tumor (violet) and control (orange) urine samples. C Receiver operating characteristic (ROC) curves illustrating the diagnostic performance. D–G Box-and-whisker plots of selected metabolites differentiating KC and control samples. Y-axis values represent MS signal intensities after log₁₀ transformation and autoscaling (z-score normalization). ***P value < 0.001. PC principal component, AUC area under the curve, CI confidence interval

To further investigate metabolic differences between the KC and NC groups, a supervised multivariate OPLS-DA was conducted. The score plot from the training set revealed a clear separation between the two groups (Fig. 2B). The statistical significance of the OPLS-DA model was evaluated using 2,000 permutation tests (Table S2, Supplementary Information) to assess the risk of overfitting. The low P value (P value = 5E−04; 0/2000 permutations exceeded the original model’s Q²) indicates that the observed group separation (Q² = 0.990, R²Y = 0.949) is unlikely to have occurred by chance. These findings highlight significant metabolic distinctions between cancer and control urine samples. The high R²Y and Q² values indicate that the OPLS-DA model possesses strong interpretability and predictive accuracy. A similar trend was observed in the validation dataset, where the OPLS-DA model effectively distinguished KC patients from NCs (Table S2). This separation was statistically supported by 2,000 permutation tests (Q² = 0.980, R²Y = 0.923, P value = 5E−04 (0/2000)), indicating that the model’s performance was unlikely to be due to chance and that the observed group discrimination was statistically robust.

Potential KC biomarkers were selected using a VIP plot generated from the OPLS-DA model. To further evaluate their diagnostic potential, univariate receiver operating characteristic (ROC) analysis was performed on both the training and validation datasets. The area under the ROC curve (AUC), a key metric for assessing model performance, was used to determine the sensitivity and specificity of each biomarker candidate. Only m/z features with an AUC > 0.70 were considered relevant. To ensure a rigorous discriminating metabolite features selection process, candidates were selected based on a combination of VIP values (> 1.0), AUC scores (> 0.7), and independent Wilcoxon rank-sum tests with false discovery rate (FDR) correction (P value and FDR < 0.05). This approach led to the selection of 229 differential variables in urine samples from KC and NC patients in the training set, while 309 significant variables were found in the validation set. Ultimately, 84 m/z features common to both datasets were retained and assigned to specific chemical compounds (Table 2, Supplementary Information 2).

Table 2.

Differential urine metabolites distinguishing KC patients from ncs, selected based on statistical significance (P value < 0.001, FDR < 0.001), variable importance (VIP > 1), fold change (FC < 0.5 or > 2), and strong diagnostic performance (AUC > 0.9)

No.	Name	Structure	m/za	RT [s]	VIPb	FCc	P valued	FDRe	AUC	Spec. [%]e	Sens. [%]e
1	2,5-Furandicarboxylic acidf,g,h	C6H4O5	157.0132	79	2.06	0.14	2.23E−16	4.75E−15	0.927	83	94
2	2-Furoylglycinef,g,h	C7H7NO4	170.0448	98	2.10	0.15	1.25E−16	2.71E−15	0.930	83	89
3	Benzamidef,i	C7H7NO	122.0601	380	2.20	0.30	9.51E−21	8.19E−19	0.986	95	94
4	12-Methyltridecanoic acidf,h,i	C14H28O2	246.2427	181	1.75	2.21	4.25E−19	1.72E−17	0.964	93	89
5	Cis, cis-Muconic acidf,g,h	C6H6O4	143.0335	94	1.81	0.24	3.53E−15	6.03E−14	0.909	83	89
6	Indoxyl sulfatef,g,h	C8H7NO4S	231.0435	112	2.12	0.15	4.04E−15	6.52E−14	0.908	88	89
7	PC(16:1_18:1)f,h,i	C42H80NO8P	758.5702	310	1.76	11.71	1.45E−14	2.13E−13	0.900	88	86
8	Sebacic acidf,g,h	C10H18O4	185.1172	135	2.08	0.36	1.80E−15	3.23E−14	0.914	80	91
9	1-Octadecanaminef,h,i	C18H39N	270.3156	250	2.93	11.32	6.80E−22	1.98E−19	1.000	100	100
10	Oleamidef,h,i	C18H35NO	282.2792	307	2.13	2.26	2.98E−21	3.72E−19	0.992	95	99
11	Sphinganinef,g,h	C18H39NO2	302.3054	223	1.96	4.15	1.51E−20	1.13E−18	0.983	95	91
12	Sphingosinef,g,h	C18H37NO2	256.3001	232	2.04	970.05	5.29E−19	1.92E−17	0.963	93	91

AUC area under the curve, FC fold change, FDR false discovery rate, m/z mass-to-charge ratio, RT retention time, Sens. Sensitivity, Spec. specificity, VIP variable influence on projection

^aExperimental monoisotopic mass of ion; ^bVIP scores derived from OPLS-DA model; ^cfold change between cancer and control urine calculated from the abundance mean values for each group—cancer-to-normal ratio; ^dvalues determined from non-parametric t-test; ^eROC curve analysis for individual biomarkers; ^fthe metabolites annotated by high precursor mass accuracy; ^gthe metabolites annotated by matching retention time; ^hthe metabolites annotated by matching isotopic pattern; ⁱthe metabolites annotated by matching MS/MS fragment spectra

Among the analyzed urine samples, 12 out of the previously selected 84 metabolites exhibited exceptionally high AUC values (> 0.9), with specificity and sensitivity exceeding 80% and 86%, respectively, in both the training and validation datasets (Tables 2 and S2, Supplementary Information 2). The combined m/z features from both datasets served as a highly effective discriminator between KC and NC urine samples (AUC = 1), as illustrated in Figs. 2C and S2C.

Pathway analysis of potential biomarkers

Pathway and enrichment analyses were conducted separately on the 103 and 84 most differentiating metabolites annotated in serum and urine, respectively, using UHPLC-UHRMS analysis.

Pathway analysis (Tables S3 and S4) performed on the differential compounds distinguishing kidney cancer patients from the control group revealed several significantly enriched pathways in both serum and urine samples. In serum, ten metabolic pathways were significantly impacted (P value): linoleic acid metabolism; valine, leucine and isoleucine biosynthesis; glycerophospholipid metabolism; biosynthesis of unsaturated fatty acids; propanoate metabolism; alanine, aspartate and glutamate metabolism; alpha-linolenic acid metabolism; glycine, serine and threonine metabolism; ketone body metabolism; butanoate metabolism (Fig. 3A, Table S3, Supplementary Material 1). The most notable pathways included linoleic acid metabolism, valine, leucine, and isoleucine biosynthesis, and glycerophospholipid metabolism, each characterized by low raw P values and retaining significance following multiple comparison corrections (Holm P, FDR). Furthermore, the high “impact” value for linoleic acid metabolism (1.0000) indicates a substantial disruption of this pathway’s metabolic profile in kidney cancer patients.

Fig. 3.

Pathway and chemical class analysis of differentiating metabolites in KC. A, B Pathway analysis of statistically significant metabolites in serum (A) and urine (B), with bubble size indicating pathway impact and color representing significance (red = highest, white = lowest). Numbered pathways correspond to the most enriched results: in serum (A) 1—linoleic acid metabolism; 2—valine, leucine and isoleucine biosynthesis, 3—biosynthesis of unsaturated fatty acids 4—glycerophospholipid metabolism; 5—propanoate metabolism; 6—alanine, aspartate and glutamate metabolism; 7—alpha-linolenic acid metabolism; 8—glycine, serine and threonine metabolism; and in urine (B) 1—linoleic acid metabolism; 2—glycerophospholipid metabolism; 3—tyrosine metabolism; 4—sphingolipid metabolism; 5—histidine metabolism (C, D) Distribution of significantly altered metabolites in serum and urine samples, categorized by pathway (C) and main-chemical subclass (D). In C, bars indicate the number of upregulated (dark green = serum, brown = urine) and downregulated (light green = serum, light brown = urine) metabolites in KC samples, mapped to each enriched KEGG or SMPDB pathway. In panel D, metabolites are grouped by chemical subclass, with color intensity indicating direction and sample origin (dark blue = serum upregulated, gray = urine upregulated; light blue = serum downregulated, light gray = urine downregulated). The x-axis in both panels shows the number of differential metabolites per category

In urine, when comparing BC to NCs, five metabolic pathways were significantly affected (P value): linoleic acid metabolism, glycerophospholipid metabolism, tyrosine metabolism, sphingolipid metabolism, and histidine metabolism (Fig. 3B, Table S4, Supplementary Material 1). The highest enrichment was observed primarily in linoleic acid metabolism and glycerophospholipid metabolism, suggesting a pivotal role of disrupted fatty acid and phospholipid metabolism in disease pathogenesis. The low raw P values and significance retained after multiple comparison corrections (Holm P, FDR) underscore the involvement of these pathways in kidney cancer. Notably, the elevated “impact value” highlights the critical role of these altered metabolites in the functionality of each pathway. Taken together, these findings emphasize the importance of lipid metabolism perturbations in kidney cancer.

Figure 3C presents all metabolic pathways enriched among the differentiating compounds from urine and serum, highlighting the shared pathways, as well as the number of metabolites that are upregulated and downregulated within these pathways. This analysis highlights key pathways such as glycerophospholipid metabolism, amino acid metabolism, and fatty acid metabolism that exhibit significant differences between kidney cancer patients and controls in both serum and urine.

The most differentiating compounds between KC patients and NCs were classified into specific chemical sub-classes, with an indication of which were upregulated and downregulated in both urine and serum (Fig. 3D). Notably, classes such as phenylacetic acids, very long-chain fatty acids, and purines exhibit distinct upregulation or downregulation patterns between serum and urine samples, suggesting potential differences in metabolic dysregulation pathways.

However, it is important to note that pathway enrichment analysis in untargeted metabolomics has inherent limitations, including annotation uncertainty, incomplete coverage of biochemical pathways, and potential statistical bias due to feature correlation or database composition. Therefore, our pathway results should be interpreted as hypothesis-generating and will require further validation using orthogonal approaches and larger datasets (Wieder et al., 2021).

Discussion

Over the past decade, metabolomics research has provided valuable insights into the metabolic profiles of individuals affected by various diseases, including cancer. These studies have also proposed potential markers for disease progression and recurrence. Given that rapidly proliferating cancer cells often reprogram their metabolism to meet increased energy demands, monitoring fluctuations in metabolite concentrations within cancer cells or bodily fluids could offer a promising avenue for discovering novel cancer potential biomarkers. Numerous studies have highlighted the significant potential of metabolomic markers in cancer diagnosis, as well as in deepening our understanding of the molecular mechanisms underlying cancer initiation and progression (Han et al., 2021).

A major strength of this study lies in the comparative analysis of serum and urine, both of which are readily accessible in clinical settings and provide complementary metabolic insights. Serum metabolite changes often reflect systemic alterations in energy balance, amino acid metabolism, and lipid turnover, while urine captures localized events within the renal microenvironment and excretory processes. Several metabolomic studies have demonstrated the value of analyzing both biofluids for RCC classification, staging, and biomarker candidate discovery. For instance, O. Falegan et al. (2017) utilized NMR and GC-MS to differentiate benign renal masses from RCC and distinguish between tumor stages based on metabolic signatures in serum and urine. Similarly, Arendowski et al. (2021) employed gold nanoparticle-assisted laser desorption/ionization mass spectrometry to suggest metabolic markers capable of distinguishing RCC subtypes and disease progression. More recently, Xu et al. (2024) demonstrated that integrating serum and urine metabolic fingerprints enhances the classification of RCC subtypes and improves early diagnostic accuracy​. These differences highlight complementary methodological perspectives: while Xu et al.’s study offers an efficient, high-throughput fingerprinting platform for tumor classification and risk stratification, our work prioritizes chemical annotation, biological interpretability, and validation across independent patient sets. Together, such diverse approaches are essential for building a robust and translatable KC biomarker pipeline.

The present study demonstrates that untargeted metabolomic profiling of both serum and urine can robustly discriminate between KC patients and NCs, underscoring the potential of metabolomics as a non-invasive approach for potential biomarker discovery. Using UHPLC-HRMS in both positive and negative ion modes and applying strict statistical criteria, we selected a set of metabolites and metabolic pathways strongly associated with kidney cancer development. Our conservative selection strategy, requiring candidate metabolites to meet all statistical criteria independently in both training and validation subsets, was designed to prioritize reproducibility and reduce the risk of false positives. This approach, described in detail in the Methods section, aligns with current recommendations for metabolomic biomarker discovery (Xia et al., 2013). Our data reveal overlapping metabolic pathways, such as linoleic acid and glycerophospholipid metabolism, but also distinct metabolic alterations in each biofluid, with amino acid biosynthesis predominating in serum and sphingolipid and histidine metabolism in urine. These results reinforce the notion that combining serum- and urine-based metabolite panels can provide a more comprehensive characterization of the metabolic derangements associated with kidney cancer, ultimately improving diagnostic and prognostic precision.

Lipid metabolism plays a crucial role in renal carcinoma progression, influencing tumor aggressiveness, immune evasion, and metastatic potential (Heravi et al., 2022). Our findings reveal significant alterations in fatty acid, phospholipid, and sphingolipid metabolism, with distinct yet complementary changes observed in serum and urine. Our findings align with earlier tissue-based studies that reported profound alterations in lipid composition in RCC (Hoffmann et al., 2005), including evidence of membrane lipid remodeling and disrupted linoleic-acid metabolism. Despite differences in sample type and analytical focus, these studies consistently highlight changes in phosphatidylethanolamine and phosphatidylserine classes, supporting the notion of altered membrane dynamics as a hallmark of tumor biology.

Linoleic acid metabolism was disrupted in both biofluids, consistent with its essential role in membrane biosynthesis and the production of lipid mediators that modulate inflammation. Increased linoleic acid metabolism has been linked to cancer progression, as tumoral and stromal cells enhance pro-inflammatory eicosanoid production, promoting a tumor-supportive microenvironment (Currie et al., 2013; Nagarajan et al., 2021). Maslov et al. (2023) further reported linoleic acid metabolism as a key dysregulated pathway in RCC progression, particularly in late-stage disease, supporting its role in tumor-associated inflammation. Manzi et al. (2021) also observed altered lipid profiles in RCC, highlighting disruptions in fatty acid metabolism that contribute to tumor progression​. Similarly, Liu et al. (2019) highlighted alterations in linoleate metabolism as a potential urinary biomarker for RCC detection.

Beyond linoleic acid, we also observed dysregulation in glycerophospholipid metabolism, which is essential for membrane integrity, signaling pathways, and lipoprotein assembly (Cheng et al., 2022). Similar findings were reported by Wang et al. (2022), who observed significant alterations in glycerophospholipid metabolism and phosphatidylinositol signaling in RCC, reinforcing the role of lipid remodeling in tumor biology. Additionally, Maslov et al. (2023) demonstrated disruptions in phosphatidylcholine (PC) and phosphatidylethanolamine (PE) metabolism, correlating with disease progression and tumor proliferation​.

Tumor cells upregulate fatty acid synthesis and desaturation enzymes to sustain rapid cell growth and membrane expansion. Myristic acid (C14:0), another dysregulated fatty acid, is involved in protein myristoylation, a post-translational modification crucial for signal transduction, membrane targeting, and cellular trafficking (Resh, 2016). Manzi et al. (2021) reported significant alterations in lipid metabolism in RCC, including changes in saturated fatty acid levels, supporting the hypothesis that myristic acid dysregulation may reflect tumor-driven metabolic reprogramming. Furthermore, Maslov et al. (2023) observed dysregulation in the biosynthesis of unsaturated fatty acids in RCC, further supporting the role of lipid reprogramming in tumor growth. Notably, our findings revealed elevated serum myristic acid levels in RCC patients compared to controls, suggesting its potential as a biomarker for disease progression. This aligns with previous lipidomic analyses demonstrating substantial shifts in lipid composition in RCC, particularly involving saturated and unsaturated fatty acids, which contribute to altered cell membrane fluidity and oncogenic signaling pathways (Manzi et al., 2021). Additionally, Liu et al. (2019) emphasized the metabolic reprogramming in RCC, highlighting dysregulated fatty acid oxidation and ketone body metabolism, which may further contribute to altered serum myristic acid levels​.

Lipid metabolism also plays a critical role in immune modulation and apoptosis regulation. Lysophosphatidylcholines (LPCs), sphingomyelins (SMs), and other complex lipids showed significant dysregulation in our data. Elevated LPCs are associated with enhanced inflammatory responses and immune cell recruitment, consistent with findings from Wang et al. (2022), who also reported lysophosphatidylcholine (LPC(19:2) as a key metabolite distinguishing RCC from benign kidney tumors​. Similarly, Jonsson et al. (2024) reported increased PC and PE levels in RCC urine samples, further supporting the role of lipid metabolism dysregulation in tumor development​.

Sphingolipids, including ceramides, sphinganine, and sphingosine, regulate apoptosis, autophagy, and stress responses (P. Liu et al. 2020). In contrast, sphingomyelins and their metabolites (e.g., ceramide) participate in pathways controlling apoptosis and cell cycle arrest (Ogretmen, 2017). Their dysregulation shifts the balance toward tumor survival by suppressing pro-apoptotic pathways. Liu et al. (2019) reported sphingolipid metabolism alterations in RCC urine, further emphasizing their role as potential non-invasive biomarkers​. Maslov et al. (2023) also demonstrated that sphingolipid pathway disruptions were significantly correlated with RCC progression, reinforcing their importance in tumor immune evasion and apoptosis modulation​. Jonsson et al. (2024) also demonstrated significant lipid metabolism alterations in RCC urine, further emphasizing the potential of sphingolipid-related markers as non-invasive indicators of renal metabolic processes​. Notably, sphingolipid alterations were particularly prominent in urine samples, underscoring their potential as non-invasive biomarkers reflecting localized renal metabolic processes.

Beyond lipid perturbations, our data reveal significant alterations in amino acid metabolism, with distinct changes observed in serum and urine. In serum, the most affected pathways include branched-chain amino acid (BCAA) metabolism (valine, leucine, and isoleucine), alanine, aspartate, and glutamate metabolism, as well as glycine, serine, and threonine metabolism. In urine, notable disruptions were observed in histidine and tyrosine metabolism, highlighting the metabolic reprogramming associated with RCC progression.

One of the most notable alterations in serum was the dysregulation of BCAA metabolism, particularly involving leucine and isoleucine. BCAAs play a crucial role in multiple cancer types, providing anaplerotic substrates for the tricarboxylic acid (TCA) cycle and serving as building blocks for protein synthesis (Arendowski et al., 2018b; Liu et al., 2024). In kidney cancer, increasing evidence suggests that BCAA uptake and catabolism support rapid tumor cell proliferation and may modulate mTOR signaling pathways (Ericksen et al., 2019).

Our findings are consistent with previous research, such as the work of Gao et al. (2008), who reported reduced leucine levels in RCC tumor tissues while observing elevated leucine levels in serum, suggesting systemic metabolic alterations​. Similarly, several studies (Hakimi et al. 2016; Jing et al. 2019; Nizioł, Copié, et al. 2021) revealed downregulated leucine levels in RCC tissues, reinforcing the notion that tumor cells actively deplete leucine from their microenvironment​. Additionally, Ganti et al. (2012) observed a significant reduction in leucine excretion in urine, indicating that RCC cells not only consume leucine locally but also alter its systemic metabolism.

However, reported leucine levels in serum remain inconsistent across studies. Some findings indicate elevated leucine levels in the serum of RCC patients, which contrasts with our observation of decreased serum leucine (Zira et al., 2010a). This discrepancy may be attributed to differences in tumor burden, disease stage, or metabolic adaptation strategies. Additionally, methodological differences could also play a role, as previous studies employed NMR spectroscopy, whereas our study utilized UHPLC-HRMS. These techniques differ in sensitivity, specificity, and the ability to resolve complex metabolomic profiles, which may contribute to variations in detected leucine levels.

We also observed significant disruptions in glycine and serine metabolism in serum, reinforcing the role of these amino acids in RCC metabolic adaptation. Zhang et al. (2017) reported similar findings, showing that alterations in glycine and serine pathways contribute to tumor progression​. These amino acids serve as precursors for nucleotide biosynthesis and one-carbon metabolism, pathways critical for sustaining high proliferation rates in cancer cells.

In urine, phenylalanine and tyrosine metabolism were significantly altered, which aligns with Schaeffeler et al. (2019), who demonstrated shifts in aromatic amino acid metabolism across RCC subtypes​. Jonsson et al. (2024) also reported disruptions in phenylalanine and tyrosine metabolism as key metabolic alterations in RCC urine samples. These changes suggest that tumor metabolism affects systemic aromatic amino acid turnover, influencing neurotransmitter and hormone biosynthesis.

Additionally, we detected histidine metabolism dysregulation in urine, further supported by Nizioł, Ossoliński, et al. (2021), who observed reduced histidine-containing peptides in RCC urine samples​. Gao et al. (2012) and Zhang et al. (2017) reported decreased histidine levels in RCC tissues and serum, respectively, emphasizing its role in cancer metabolism​.

Finally, we found alterations in tryptophan metabolism, with increased kynurenine pathway activity suggesting immune evasion by the tumor. Liu et al. (2019) and Zira et al. (2010b) reported similar findings, demonstrating that RCC patients exhibit upregulated tryptophan metabolism linked to immunosuppression​. These findings highlight the compartmentalized nature of metabolic reprogramming in RCC, where serum alterations reflect systemic metabolic shifts, while urinary changes indicate localized tumor effects.

Among the annotated metabolites, 19 in serum (Table 1) and 12 in urine (Table 2) exhibited high diagnostic accuracy (AUC > 0.90), demonstrating strong specificity and sensitivity in both training and validation cohorts. These findings emphasize the potential of metabolomics in distinguishing RCC patients from healthy individuals based on systemic and localized metabolic changes.

In serum, notable differentiating compounds included benzoic acid and methyl 2-methoxybenzoate, both aromatic derivatives with potential roles in detoxification and immune regulation. Their altered levels may reflect changes in microbial metabolism, oxidative stress, or inflammatory signaling, processes that are frequently dysregulated in RCC (Mao et al., 2019). Prior research suggests that shifts in benzoic acid metabolism, particularly its conjugation with amino acids, may reflect broader metabolic reprogramming in cancer progression, further reinforcing its relevance as a potential biomarker (Holbrook et al., 2024). Additionally, dicarboxylic acids, such as octadecanedioic acid and 2,2-dimethylsuccinic acid, were significantly modulated, pointing toward enhanced ω-oxidation and alternative fatty acid degradation pathways that could support tumor energy demands. These observations align with previous metabolomic studies, suggesting that fatty acid oxidation and detoxification pathways are reprogrammed in RCC to support rapid proliferation (Jin et al., 2021).

In urine, notable discriminatory metabolites included sphinganine and sphingosine, reinforcing the importance of sphingolipid metabolism in RCC progression. These changes reflect disturbances in cell membrane integrity, apoptosis regulation, and tumor immune evasion mechanisms. The significant upregulation of sphinganine observed in our study is consistent with findings from metabolomic analyses of RCC, where sphinganine was found to be among the most altered metabolites, suggesting a disruption in sphingolipid homeostasis (Kordalewska et al., 2022). This alteration aligns with broader metabolic reprogramming characteristic of renal cancer, where shifts in lipid metabolism play a crucial role in tumor growth and survival. Furthermore, comprehensive metabolic profiling of ccRCC has demonstrated widespread dysregulation of central carbon metabolism, amino acid metabolism, and antioxidant responses (Hakimi et al., 2016). Specifically, their study highlighted significant alterations in lipid metabolism pathways, which may contribute to the observed increase in sphinganine. As sphingolipid metabolism is closely linked to cellular stress responses, apoptosis resistance, and membrane signaling, its dysregulation may facilitate ccRCC progression and metastasis.

Indoxyl sulfate, a gut-derived uremic toxin, is primarily excreted by the kidneys. Its significantly reduced urinary levels in renal cancer patients suggest impaired tubular secretion, potentially due to tumor-related dysfunction. This may lead to systemic accumulation, promoting local inflammation, oxidative stress, and fibrosis, which could further contribute to tumor progression and renal damage. These findings align with evidence linking IS to chronic kidney disease progression and nephrovascular toxicity (Wakabayashi & Marumo, 2022).

Furthermore, fatty amides such as 1-octadecanamine and oleamide, along with dicarboxylic acids like sebacic acid, suggest alternative lipid catabolism pathways and immune modulation in KC patients. Previous studies reported elevated urinary oleamide in renal cancer patients, reinforcing its potential as a metabolic biomarker (Arendowski et al., 2020a). As a signaling lipid and neuromodulator, oleamide influences GABA-A receptors and voltage-gated sodium channels and has been linked to altered calcium regulation in renal and bladder cancer cells (Verdon et al., 2000). Its increased excretion may reflect dysregulated lipid metabolism, tumor-induced metabolic shifts, or impaired renal processing.

Among the most distinguishing compounds between the cancer and control groups were dicarboxylic acids. In serum samples, elevated concentrations of octadecanedioic acid and 2,2-dimethylsuccinic acid were observed in KC patients. Additionally, urine analyses revealed higher levels of 2,5-furandicarboxylic acid and cis, cis-muconic acid in the KC cohort. The elevation of octadecanedioic acid, a long-chain dicarboxylic acid, may suggest disruptions in fatty acid metabolism associated with RCC. Dicarboxylic acids are organic compounds containing two carboxyl functional groups, playing a crucial role in various metabolic pathways, including fatty acid oxidation and the tricarboxylic acid (TCA) cycle. Altered lipid metabolism has been implicated in cancer progression, potentially due to increased energy demands and membrane biosynthesis in proliferating tumor cells (Guerra et al., 2024).

Our study revealed significantly elevated levels of SM(d41:1) and SM(d33:1) in the serum of kidney cancer patients compared to controls. These odd-chain sphingomyelins are uncommon in human lipid metabolism, as fatty acid biosynthesis typically favors even-numbered carbon chains. Their accumulation suggests tumor-induced alterations in sphingolipid metabolism, potentially linked to lipidomic reprogramming in kidney cancer. This reprogramming affects sphingomyelin synthesis and turnover, contributing to the observed changes (Mullen et al., 2012; Saito et al., 2016). Additionally, modifications in fatty acid elongation pathways or exogenous lipid uptake could support their presence in circulation. Another possible source is tumor-derived extracellular vehicles (EVs), which play a role in cancer progression and intercellular communication. EVs from KC cells have been shown to carry altered sphingomyelin compositions, potentially leading to selective enrichment of SM(d41:1) and SM(d33:1) in serum (Skotland et al., 2019). Furthermore, the systemic lipidomic impact of KC must be considered. The hepatic lipidome significantly influences circulating sphingolipid levels, with hepatic ceramides correlating with specific sphingomyelin species in LDL (Lahelma et al., 2022). Disruptions in liver sphingolipid metabolism in response to KC-related metabolic shifts may further contribute to the observed increase in these sphingomyelins. Notably, SM(d41:1) has been reported as a discriminative lipid in RCC patients, supporting its potential as a biomarker (Wolrab et al., 2021) while SM(d33:1) requires further validation.

These results emphasize that serum and urine metabolomic profiles, although distinct, provide complementary perspectives on RCC-related metabolic changes. While serum metabolites reflect systemic metabolic shifts, urinary metabolites capture local tumor-driven metabolic alterations, reinforcing the importance of integrating multi-biofluid approaches for RCC biomarker candidate discovery.

In our study, we assigned a set of metabolites that showed significant discriminatory power between cancer and control groups. However, according to the Metabolomics Standards Initiative (MSI) classification (Sumner et al., 2007), these metabolites should be considered as putatively annotated compounds (Level 2), as we did not confirm their identity using authentic reference standards. Annotations were based on exact mass, isotopic pattern fit (mSigma), MS/MS fragmentation spectra, and retention times available in the Bruker HMDB 2.0 library. Although this multifactorial approach improves confidence in annotation, we recognize that full MSI Level 1 identification requires confirmation with reference compounds. We plan to address this in follow-up studies by validating key candidate biomarkers using targeted assays and authentic standards. However, these metabolites should be considered as candidate biomarkers, pending validation in larger, independent cohorts and across broader clinical settings, in line with recommended biomarker development frameworks (Koulman et al., 2009).

Our study reveals promising metabolomic alterations in kidney cancer, however, their full clinical significance warrants further investigation and validation. A critical factor influencing diagnostic performance is the nature of the control group. In our study, the non-cancer control cohort consisted of patients with benign urological conditions such as benign prostatic hyperplasia and urolithiasis representing common differential diagnoses in urological practice. This design was intentionally chosen to reflect the real-world clinical setting in which patients present with symptoms or radiological findings that require distinction between malignant and benign etiologies. Such an approach enhances the translational applicability of the assigned metabolic signatures, as it tests the potential biomarkers under diagnostically relevant conditions rather than idealized comparisons with healthy asymptomatic individuals (Beger et al., 2019; Cameron et al., 2023). Nonetheless, this strategy may introduce a trade-off in terms of specificity for detecting cancer in asymptomatic populations, and it remains possible that some metabolic alterations are shared across malignant and non-malignant urological conditions.

Since biomarker performance is influenced by disease prevalence, symptom burden, and comorbidity profiles, future studies will aim to confirm the diagnostic utility of these candidate biomarkers by expanding the scope of comparator groups to include healthy individuals, patients with systemic inflammatory conditions, and individuals undergoing cancer screening. Additionally, further research will focus on validating these findings in larger, independent cohorts and on assessing confounding factors such as diet, comorbidities, and treatment, as well as exploring causal links between these metabolic changes and tumor progression.

A potential methodological limitation of the present study is the lack of pre-analytical normalization of urinary metabolites to dilution markers such as creatinine or specific gravity (SG). Although all samples were collected under tightly standardized conditions (morning, fasting, preoperative), minor variability in hydration may still influence metabolite concentrations. Our decision not to normalize was supported by previous work in a similar clinical setting, where SG measurements showed minimal inter-individual variability and had a negligible impact on the interpretation of untargeted metabolomic data. Furthermore, recent studies have emphasized that no single normalization method performs universally better, and that over-correction may introduce bias, especially when sample collection is well controlled and biological variability is low (Bouatra et al., 2013; Cook et al., 2020).

To assess whether sex or age influenced the observed metabolic differences, we performed PCA on the full dataset. No clear clustering by sex or age was observed, suggesting that cancer status was the dominant source of variation. These results align with previous studies reporting only minor metabolic effects of sex and age in RCC patients (Deja et al., 2021; KIm et al., 2009). Another important source of biological variability in our dataset is the clinical and histopathological heterogeneity of the RCC cohort, which included multiple subtypes, grades, and TNM stages. While this diversity limited our ability to perform stratified analyses, it was deliberately retained to identify metabolic perturbations common to RCC irrespective of subtype. In future studies, larger and stratified cohorts will be required to validate subtype-, grade-, and stage-specific candidate biomarkers.

Future research will focus on expanding the study to larger, independent cohorts to capture disease heterogeneity and account for metabolic variations influenced by diet, comorbidities, and treatment. Additionally, further studies will be essential to establish causal relationships between these metabolic changes and tumor progression.

Conclusions

Our comprehensive metabolomic analysis of serum and urine has revealed distinct metabolic signatures associated with kidney cancer, highlighting significant alterations in lipid metabolism, amino acid biosynthesis, and other interconnected biochemical pathways. The ability to robustly distinguish kidney cancer patients from non-cancer controls, supported by high AUC values for key metabolites, underscores the clinical potential of these findings. The inclusion of both serum and urine, collected from the same individuals, provided complementary insights into systemic and renal-localized metabolic alterations. This study demonstrates the application of minimally invasive sampling combined with advanced analytical and statistical methods to support metabolomics-based strategies for early detection, prognosis, and disease monitoring. To translate these discoveries into clinical practice, further large-scale validation and mechanistic studies are necessary to refine these potential biomarkers and establish their role in guiding personalized treatment strategies for kidney cancer patients.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

A.O: Formal analysis; methodology, investigation; resources; data curation, writing—original draft; A.P-A.: Methodology, investigation; data curation; T.R.: Methodology, resources, data curation, writing—review and editing. K.O.: Resources, J.N.: Conceptualization, methodology, formal analysis, investigation, resources, data curation, writing—original draft, writing—review & editing, visualization, supervision, project administration, funding acquisition. T.O.: Resources.

Data availability

The corresponding author’s data supporting this study’s findings are available upon reasonable request.

Declarations

Conflict of interest

The authors declare no competing financial and/or non-financial interests.

Consent to participate

The patients provided written consent to participate in research.

Consent for publication

The patients provided written informed consent for the publication of any associated data.

Ethical approval

The local Bioethics Committee approved the study protocol at the University of Rzeszow (Poland) (permission no. 2018/04/10).

Research involving human and/or animal participants

This article does not contain any studies with human and/or animal participants performed by either of the authors.