Anna Klimova

Purpose:Reliable prognostic tools in ovarian cancer are urgently needed to guide risk-adapted treatment decisions, yet the clinical utility of urinary metabolites for noninvasive risk stratification remains largely undefined. Here, we define a clinically relevant urinary metabolite signature that enables noninvasive prognostic risk stratification in ovarian cancer.Experimental Design:We used targeted 1H nuclear magnetic resonance spectroscopy to profile 149 metabolites related to energy metabolism, oxidative stress, mitochondrial function, nitrogen metabolism, amino acid degradation, gut microbiome activity, and inflammation. Metabolites were measured in preoperative urine samples from 199 consecutive patients with newly diagnosed ovarian cancer treated in routine clinical practice between 2013 and 2022.Results:Unsupervised clustering revealed biologically heterogeneous subgroups but lacked prognostic resolution and alignment with overt clinical phenotypes. However, single-metabolite analysis identified a condensed three-metabolite prognostic signature comprising glycine, alanine, and citrate. A final parsimonious model integrating this metabolite signature with clinical covariates outperformed established risk factors alone (Fédération Internationale de Gynécologie et d'Obstétrique stage and surgical outcome), accurately predicted 60-month overall survival (AUC = 0.839), and stratified risk. Patients in the highest-risk quartile (Q4) had markedly shorter progression-free survival [Δmedian ≈ 56 months; HR, 2.63; 95% confidence interval (CI), 1.54–4.52; P < 0.001] and overall survival (Δmedian ≈ 86 months; HR, 2.49; 95% CI, 1.39–4.46; P = 0.009) compared with the lowest-risk group (Q1).Conclusions:We define a urinary three-metabolite signature that enables noninvasive identification of patients with high-risk ovarian cancer beyond established clinical factors. This signature may support molecular stratification and risk-adapted clinical decisions, thereby underscoring the clinical scalability of urine as a matrix for metabolic risk profiling in ovarian cancer.

Supplementary Data 1 from A Urinary Three-Metabolite Signature Enables Noninvasive Identification of Patients with High-Risk Ovarian Cancer

Identification of an optimized three-metabolite signature. A, Representation of stepwise model construction, comparing nested Cox models and ROC performance. B, Time-dependent ROC curves of the clinical model (model 1), extended model (model 2: clinical +5 metabolites), and optimized model (model 3: clinical + glycine, alanine, citrate). AUC values indicate discriminatory performance for 60-month OS. To account for individuals censored before 60 months, AUC estimates were calculated using the inverse probability of censoring weighting method. Analysis was restricted to the complete-case cohort. (A, Image created with BioRender.com. Schwarz, F. [2026] https://BioRender.com/jd4n0h0.)

Prognostic stratification using the three-metabolite signature in combination with clinical covariates (model 3). Kaplan–Meier curves for (A) PFS and (B) OS across quartiles of the composite risk score derived from model 3. The log-rank test was used to assess statistical significance.

Patient characteristics at primary diagnosis.

Urine-based metabolomics profiling in patients with ovarian cancer at the cluster level. A, Metabolites with ≥ 50% nonzero values across urine samples from 199 patients with ovarian cancer. B, Principal component (PC) analysis of urinary metabolites, identifying two distinct metabolic clusters (cluster 1: red; cluster 2: green). C, Mean metabolite abundances across clusters. D, PFS and (E) OS by metabolic cluster assignment. F, Cancer antigen 125 (CA-125) levels and (G) surgical outcome [macroscopic complete resection (R = 0) vs. residual disease (R > 0)] stratified by cluster. Statistical significance was assessed by Wilcoxon or χ2 test, as appropriate; exact P values are shown in the panels.

Supplementary Figures from A Urinary Three-Metabolite Signature Enables Noninvasive Identification of Patients with High-Risk Ovarian Cancer

Identification of prognostically informative urinary metabolites. A, Overview of the analytic workflow used to identify urinary metabolites with prognostic relevance. B, Regularization path of Cox-LASSO regression across λ-values. C, Nonzero β-coefficients at optimal λ (λ_min) for retained metabolites. Here, β denotes the Cox regression model coefficient and is shown as the log HR for each metabolite, not the metabolite concentration. D, Pairwise Pearson correlations among retained metabolites (n = 190); all r < 0.28. (A, Image created with BioRender.com. Schwarz, F. [2026] https://BioRender.com/7oq101y.)

Comparative analysis of urinary versus plasma-based metabolomic signature. A, Pairwise correlations between metabolite concentrations in matched plasma and urine samples (n = 161); r denotes Pearson’s coefficient. Asterisks indicate P < 0.05. B, Schematic of model comparisons using urinary metabolites and a previously established plasma signature. C, Time-dependent ROC curves of the urinary model, plasma signature, and combined model. AUC values indicate discriminatory performance for 60-month OS. To account for censoring before 60 months, AUC estimates were calculated using inverse probability of censoring weighting. Analysis was restricted to the complete-case cohort. (B, Image created with BioRender.com. Schwarz, F. [2026] https://BioRender.com/dy8x8k4.)