A comprehensive plasma-based approach to thromboinflammation in bladder cancer: integrating lipidomics, thrombin generation, and NETosis biomarkers

Abstract

Background

Lipid metabolism reprogramming is a hallmark of cancer, yet its systemic manifestations and functional implications in bladder cancer (BC) remain poorly understood. This study aimed to identify plasma lipidomic profiles with diagnostic, staging, and prognostic value in BC, and to explore their associations with thromboinflammatory mechanisms, particularly NETosis and thrombin generation.

Methods

Plasma lipidomics was characterized by ultra-high-performance liquid chromatography high-resolution mass spectrometry from 87 BC patients (non-muscle-invasive BC, NMIBC, and muscle-invasive BC, MIBC) and 30 controls. NETosis markers, active DNaseI concentration, C-reactive protein, fibrinogen, IL-6, prothrombin F1+2, thrombin-antithrombin (TAT) complexes, and thrombin generation tests (TGT) were assessed in parallel. Statistical analyses in R included PCA, ROC curve analysis, elastic net penalized logistic regression, Cox regression, natural cubic spline modeling for survival analyses, and pathway enrichment analyses.

Results

BC patients exhibited increased total lipid load and distinct lipidomic signatures. Phosphatidylcholine (PC) levels decreased with disease severity and were independently associated with poor overall survival (up to 85% reduction in MIBC). Multivariable elastic net models integrating lipidomic and clinical variables achieved high diagnostic accuracy for discriminating BC patients from controls (AUC = 0.943) and for distinguishing MIBC from NMIBC (AUC = 0.860). BC patients displayed increased NETosis, inflammation, and thrombin formation, while decreased active DNaseI. Potentially diagnostic lipids correlated with NETosis, inflammatory markers and TAT, and inversely with active DNaseI and TGT parameters. Given the cross-sectional and observational design, these findings should be interpreted as hypothesis-generating and do not allow causal inference.

Conclusions

Plasma lipidomic profiling identifies biomarkers with diagnostic and prognostic potential in BC and reveals associations with NETosis, active DNaseI dysfunction, and thrombin generation. While causality cannot be inferred from this exploratory study, these findings support lipidomics as a minimally invasive tool for BC stratification and as a window into cancer-associated immunothrombosis, providing a rationale for future longitudinal and mechanistic studies targeting lipid-neutrophil interactions.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12967-026-08570-8.

Background

Bladder cancer (BC) is the most prevalent malignancy of the urinary tract, the ninth most frequently diagnosed worldwide, with approximately 614,000 new cases and 22,000 deaths [1]. Around 75–80% of cases are initially diagnosed as non-muscle-invasive bladder cancer (NMIBC), whereas the remaining 20–25% are muscle-invasive (MIBC), which carries a significantly worse prognosis. The high recurrence rate and additional complications prompt a deeper understanding of the underlying mechanisms in BC. Despite advances in diagnostic biomarkers [2–6], current diagnostic techniques present significant limitations, particularly in the detection of low-grade lesions [7].

Lipid metabolism has gained attention as a hallmark of cancer. Lipids regulate membrane architecture, signal transduction, and tumor-immune interactions, and their reprogramming sustains malignant growth and progression [8]. Lipidomic profiling of plasma provides a minimally invasive readout of these systemic alterations and has demonstrated diagnostic and prognostic utility across several tumor types [8, 9]. Few studies in BC have mainly focused on tumor tissue, leaving the characterization of circulating lipidomic alterations largely unexplored [10–13]. Beyond their structural and metabolic roles, lipids are increasingly recognized as modulators of inflammation and coagulation [14]. Ceramides (Cer) and phosphatidylcholines (PC) can regulate neutrophil activation, cytokine release, platelet function, and modulate the thrombotic risk [14–17]. Furthermore, neutrophil extracellular traps (NETs) have emerged as central mediators of the link between cancer, inflammation, and thrombosis. NETs promote tumor growth, angiogenesis, and metastasis, while promoting a hypercoagulable state [18]. Indeed, our group has demonstrated that BC patients exhibit elevated NETosis and impaired DNaseI-mediated NET degradation, phenomena associated with tumor-promoting processes and adverse clinical outcomes [19].

Given this background, we hypothesized that BC progression may be linked to lipid reprogramming, which could be not only the source of disease biomarkers, but also potential contributors to thromboinflammatory processes, including NETosis and thrombin generation. Thus, we performed an integrated analysis of plasma lipidomics, NETosis markers, and thrombin generation in BC patients and controls. This approach seeks to (i) define circulating lipid signatures associated with disease stage, (ii) evaluate their diagnostic and prognostic potential, and (iii) uncover mechanistic links between lipid metabolism, immunothrombosis, and BC progression.

Methods

Study population and sample collection

Eighty-seven BC patients and 30 age- and sex-matched controls were recruited between July 2016 and March 2022 at La Fe University and Polytechnic Hospital (Valencia, Spain). Patients with a clinical suspicion of BC were enrolled in the study, either at first diagnosis or at recurrence. Controls were patients of the Urology Service who were being examined for non-malignant disorders (e.g. cystitis, benign prostate hyperplasia), which could be a differential diagnosis of BC. These underwent ultrasound examinations to exclude urological malignancies or other abnormalities. All participants were clinically followed until July 2025.

Preoperative clinical staging in BC patients included physical examination, urine cytology, and, for invasive cases, computed tomography (CT) scans of the chest, abdomen, and pelvis. Tumor histological grading was assessed according to the 1973 and 2004 World Health Organization (WHO) classifications [20, 21]. Comprehensive demographic and clinical data were collected. Exclusion criteria included the absence of histological confirmation or the presence of other malignancies.

Peripheral blood samples were obtained from patients before surgery (transurethral resection of bladder tumor or cystectomy), while none were under treatment for BC, and from controls once the motive of consultation was resolved. Blood was collected in 4.5 mL Vacutainer tubes containing 0.5 mL of 0.109 M trisodium citrate (BD Diagnostics). Platelet-free plasma was prepared by centrifugation at 1,811 × g for 30 minutes at 4 °C, aliquoted, and stored at −80 °C until analysis.

Lipidomic sample preparation and quality control strategy

Fifty µL of plasma were extracted with 150 µL isopropanol, vortexed, and incubated at −20 °C for 20 minutes. Following centrifugation at 13.000 ×g for 10 min at 4 °C, 90 µL supernatant were transferred to a 96-well plate and mixed with 10 µL of a lipidomics internal standard (ISTD) mixture. A pooled quality control (QC) sample was prepared by combining 10 µL from each extract.

To ensure analytical reproducibility, samples were randomly injected. QC samples were run at the beginning of the sequence (n ≥ 5) to condition the system, and subsequently every 5–7 injections in full MS mode. Data-dependent acquisition (DDA, Auto MS/MS) and data-independent acquisition (DIA, All-Ion Fragmentation) were applied to QC samples at the beginning, middle, and end of the batch to support lipid identification and annotation using LipidMS v3 package [22].

Lipidomic detection

Samples were analyzed using an Ultra-Performance Liquid Chromatography (UPLC) system coupled to a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific). Lipid separation was achieved using an Acquity UPLC CSH C18 column (100 × 2.1 mm, 1.8 µm; Waters).

For positive electrospray ionization (ESI⁺), the mobile phases were: (A) 10 mM ammonium formate in acetonitrile:water (60:40, v/v) and (B) 10 mM ammonium formate in isopropanol:acetonitrile (90:10, v/v), delivered at 0.4 mL/min. For negative ionization mode (ESI⁻), ammonium acetate was used as an additive, with a flow rate of 0.6 mL/min. The autosampler and column were maintained at 4 and 65 °C, respectively, and the injection volume was 5 µL.

Lipidomic data preprocessing and analysis

Raw data from full-scan MS, DDA, and DIA acquisitions were processed and annotated using the LipidMS v3 package [22], which applies a rule-based workflow based on experimentally validated MS/MS fragmentation patterns for each lipid class. Lipid annotation was therefore supported by MS/MS evidence. LipidMS provides hierarchical structural annotation levels depending on the available spectral information. When diagnostic fragments enabled identification of fatty acyl chains, these were reported; when positional (sn) information could be inferred, it was included. In cases where such structural detail could not be confidently resolved, lipid species were annotated at the summed composition or subclass level to avoid overinterpretation. Isobaric and isomeric species were evaluated using MS/MS fragmentation rules, chromatographic behavior, and adduct consistency. When these species could not be unequivocally distinguished, they were reported at the summed composition or subclass level. Additionally, predicted retention time trends were used as an orthogonal criterion to support annotation consistency and identify potential misannotations. Peak detection and grouping parameters were optimized based on prior experience with comparable datasets. Data matrices from positive and negative ionization modes were then merged into a single dataset, integrating annotated features from both polarities.

Signal normalization was performed using locally estimated scatterplot smoothing regression (LOESS). QC filtering retained lipid features with a coefficient of variation (CV) <30% in QC samples and detectable in at least 60% of samples within one of the study groups.

To minimize redundancy, when a lipid species was annotated with multiple adducts, only the most abundant adduct per ionization mode was retained. This curation step ensured a representative and non-redundant list of lipid species across modes.

Assessment of NETosis biomarkers, inflammation, and active DNaseI concentration

Different markers of NETosis were analyzed in plasma as previously described [19, 23–27]: cell-free DNA (cfDNA; Quant-iT PicoGreen dsDNA kit, Life Technologies), nucleosomes (Cell Death Detection ELISA^PLUS kit, Roche), calprotectin (Human Calprotectin ELISA kit, Hycult Biotech), myeloperoxidase (MPO; Human MPO ELISA kit, Abnova), neutrophil elastase (PMN Elastase Human ELISA kit, Abcam), and DNA-citrullinated histone H3 (citH3) complexes.

As inflammatory markers, C-reactive protein (CRP, N High Sensitivity CRP, Dade Behring®) and fibrinogen (high-sensitivity standard reagent IL, Instrumentation Laboratory) were measured as part of routine clinical laboratory testing, while interleukin-6 (IL-6, Human IL-6 High Sensitivity ELISA Kit, Diaclone) was measured using ELISA, following the manufacturer’s instructions.

Plasma active DNaseI concentration was measured using the single radial enzyme-diffusion (SRED) assay as previously described [19, 28]. Briefly, agarose gels containing labeled DNA were prepared by dissolving 45 µg/mL salmon sperm DNA (Sigma Aldrich) in a buffer of 20 mM Tris-HCl (pH 7.8), 10 mM MgCl₂, and 2 mM CaCl₂. After heating for 10 min at 50 °C, the solution was mixed with an equal volume of 2% agarose (Condalab) containing 0.08% Safe View (NBS Biologicals) and poured into plates. Following gelification, 1 mm wells were prepared and loaded in duplicate with 2 µL plasma. In each experiment, 2 µL pooled plasma from 19 healthy donors was included in duplicate as an inter-assay calibrator. Plates were incubated for 17 h at 37 °C in a humid chamber. DNA degradation halos were visualized using a fluorimeter (Uvitec), photographed, and quantified with ImageJ software (NIH). Active DNaseI concentration was calculated by interpolation in a standard curve of known concentrations of rhDNaseI (Dornase alfa, Pulmozyme®, Roche).

Assessment of thrombin generation

Prothrombin fragment F1+2 (Enzygnost® F1+2 Mono, Siemens Healthcare GmbH) and thrombin-antithrombin complexes (Enzygnost® TAT Micro, Siemens Healthcare GmbH) were measured to reflect thrombin generated and inhibited in 1:2 diluted plasma using ELISA, following the manufacturer’s instructions.

The thrombin generation test (TGT) was assessed using the Calibrated Automated Thrombogram (CAT) method at 37 °C in a Fluoroskan Ascent microplate reader (Thermo Labsystems) equipped with 390/460 nm filters. Thrombin generation curves and parameters were calculated using Thrombinoscope software (Thrombinoscope BV). Each assay was performed in duplicate and consisted of 60 µL platelet-poor plasma, 20 µL Owren-Koller buffer (Stago), 20 µL of a solution containing 4 pM tissue factor and 4 µM phospholipids (final concentration; Innovin, Dade Behring), and 20 µL fluorogenic substrate (FluCa-Kit, Thrombinoscope BV) in the presence of CaCl₂. In parallel, an additional reaction containing 20 µL thrombin calibrator (Thrombinoscope BV) was run for normalization.

Statistical analysis

Exploratory analysis of the lipid profiles was performed using principal component analysis (PCA) and heatmaps of clustered observations and variables. Differential lipid composition was assessed considering |FC| ≥2 and false discovery rate (FDR)-adjusted p-value < 0.05. For a more robust estimation, shrunken log₂ fold changes (log₂FC) were estimated from the raw log₂FC estimates using the ashr procedure. Metabolite Set Enrichment Analysis (MSEA) was performed using MetaboAnalyst6.0, while lipid pathway analysis was performed using the Bioinformatics Methodology For Pathway Analysis (BioPAN) tool available at https://www.lipidmaps.org/.

Continuous variables are reported as median and interquartile range (IQR), while categorical variables are expressed as counts and percentages. Replicate experiments are presented as mean ± standard error of the mean (SEM), unless otherwise specified. Group comparisons of continuous variables were performed using the Mann-Whitney U test (two groups) or the Kruskal-Wallis test (three or more groups). Correlations between continuous variables were assessed using Spearman’s rank correlation, considering FDR-adjusted p-value < 0.05.

Survival analyses were performed using univariate Cox proportional hazards regression and Kaplan-Meier curves with log-rank tests, incorporating clinicopathological variables and dichotomized lipid levels. To determine the independent prognostic value of the biomarkers, a multivariable Cox regression model was constructed with a forward stepwise approach, including all variables significant in univariate analyses. To further evaluate the independent prognostic value of PC as a continuous variable, multivariable Cox proportional hazards regression models were constructed for both the entire cohort and the MIBC subgroup. The models included age, MIBC status, and PC levels as covariates, with sex additionally included in the model for the full cohort. To account for potential non-linear associations between PC levels and overall survival (OS), PC was modeled using natural cubic splines. The overall contribution of PC to the model was assessed using a likelihood ratio test, comparing models with and without the spline term. The proportional hazards assumption was evaluated for all covariates using Schoenfeld residuals.

Diagnostic accuracy was evaluated using the area under the receiver operating characteristic (AUC ROC) curves. Two elastic net penalized logistic regression models were adjusted to select lipids able to discriminate BC patients from controls and MIBC from NMIBC. For the BC vs. control comparison, the elastic net model included sex, age, dyslipidemia, diabetes, and statin treatment as clinical covariates. For the MIBC vs. NMIBC comparison, the model included previous treatment as well. The penalization factors for the elastic net were selected by taking the highest lambda at one standard error from the minimum (one-standard-error rule) from 300 repetitions of 10-fold cross-validation. Then, the median of the 300 lambda values was used as the final penalization factor. The elastic net parameter alpha (regulating the mix of L1 and L2 penalization) was set at 0.5. Lipids with coefficients different from zero after the penalization were selected by the analysis and were therefore considered relevant for discriminating between groups.

Statistical analyses were performed using R (version 4.4.2), with p-values < 0.05 considered statistically significant.

Results

Study cohort

The clinical and demographic characteristics of the 87 BC patients and 30 controls are described in Table 1. As the incidence of BC is higher in males than in females [29], the studied cohort was predominantly male, 85% in BC patients and 90% in controls.

Table 1.

Clinical and demographic characteristics of BC patients and controls studied

Characteristics	MIBC patients (n = 34)	NMIBC patients (n = 53)	Controls (n = 30)
Age at diagnosis, median (Q1–Q3)	71 (65.5–75.8)	68 (62–74)	65.5 (61.5–74.5)
Sex, N (%) Male Female	31 (91.2) 3 (8.8)	43 (81.1) 10 (18.9)	27 (90.0) 3 (10.0)
Tumor stage and grade, N (%) pTis TaG1 TaG2 TaG3 T1G3 T1G2 T2G2 T2G3 T3G2 T3G3 T4G2 T4G3	– – – – – – 3 (8.8) 12 (35.3) 2 (5.9) 10 (29.4) 2 (5.9) 5 (14.7)	2 (3.8) 4 (7.6) 23 (43.4) 7 (13.2) 16 (30.2) 1 (1.9) – – – – – –	–
Exitus, N (%) Yes No	19 (55.9) 15 (44.1)	9 (17.0) 44 (83.0)	–
Tumor size, cm2, N (%) <1 1–3 >3	0 (0.0) 14 (41.2) 20 (58.8)	4 (7.5) 32 (60.4) 17 (32.0)	–
Previous treatment, N (%) Doxorubicin Mitomycin BCG Combined treatment None	0 (0.0) 1 (2.9) 4 (11.8) 2 (5.9) 27 (79.4)	1 (1.9) 1 (1.9) 7 (13.2) 2 (3.8) 42 (79.2)	–
Tumor histology, N (%) Urothelial-papillary Urothelial-concomitant CIS	31 (91.2) 3 (8.8)	47 (88.7) 6 (11.3)	–
Smoking habits, N (%) Current Former Never Unknown	17 (50.0) 8 (23.5) 9 (26.5) 0 (0.0)	32 (60.3) 11 (20.8) 9 (17.0) 1 (1.9)	15 (50.0) 4 (13.3) 11 (36.7) 0 (0.0)
Dyslipidemia, N (%)	20 (58.8)	38 (71.7)	15 (50.0)
Lipid levels, median (Q1–Q3) Total cholesterol HDL cholesterol LDL cholesterol Triglycerides	152 (116–201) 43 (37–55) 77 (45–116) 124 (81–145)	179 (156–205) 46 (39–64) 104 (81–104) 134 (101–162)	192 (164–216) 50 (42–57) 132 (108–146) 95 (79–142)
Statin use, N (%)	17 (50.0)	30 (56.6)	14 (46.7)
Diabetes mellitus, N (%)	10 (29.4)	17 (32.1)	4 (13.3)

BC, bladder cancer; BCG, bacillus Calmette-Guérin; CIS, carcinoma in situ; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MIBC, muscle-invasive bladder cancer; NMIBC, non-muscle-invasive bladder cancer

BC patients had higher rates of dyslipidemia and diabetes mellitus than controls. Despite having higher levels of total cholesterol (p = 0.0082), LDL cholesterol (p = 0.0002), and triglycerides (p = 0.0043), statin use was comparable between groups.

Lipid metabolism reprogramming in BC involves PC depletion with prognostic implications

The UPLC-HRMS analysis identified 313 lipid species across 16 classes in the samples studied. The relative abundance of lipid classes, defined as the proportion of each class within the total detected lipid pool, gradually varies, showing a reprogramming of lipid metabolism with tumor severity (Fig. 1A).

Fig. 1.

Plasma lipidomic alterations in BC patients reveal decreased PC levels associated with poor prognosis. (A) Stacked plot showing the relative abundance (%) of lipid classes in plasma samples from controls, patients with NMIBC, and patients with MIBC. Lipid class abundance was calculated as the percentage of each class relative to the total lipid signal. (B) Boxplot illustrating PC levels in controls, NMIBC, and MIBC patients. Kruskal-Wallis test with post hoc Dunn’s test: *p < 0.05, ***p < 0.001, ****p < 0.0001. (C-D) Kaplan-Meier survival curves showing that lower plasma PC levels (below median) are significantly associated with worse OS in the global cohort (NMIBC + MIBC) (C) and the MIBC group (D). (E-G) Lipid pathway analysis using BioPAN comparing: (E) NMIBC vs. controls, (F) MIBC vs. controls, and (G) MIBC vs. NMIBC. Green and purple edges represent activated and suppressed enzymatic steps, respectively. BC patients exhibit suppressed PC synthesis from PE, LPC, and DG, accompanied by enhanced conversion of PC into LPC and DG. BC: bladder cancer; CE: cholesteryl esters; Cer: ceramide; CerP: ceramide 1-phosphate; DG: diacylglycerol; FA: fatty acids; LPC: lysophosphatidylcholine; LPE: lysophosphatidiletanolamine; MG: monoacylglycerol; MIBC: muscle invasive bladder cancer; NMIBC: non-muscle invasive bladder cancer; OS: overall survival; PC: phosphatidylcholine; PCo|PCp: plasmanyl-glycerophosphocolines/plasmenyl glycerophosphocolines; PE: phosphatidylethanolamine; PEo|Pep: plasmanyl-glycerophosphoethanolamines/plasmenyl-glycerophosphoethanolamines; PI: phosphatidylinositol; SM: sphingomyelin; SphP: sphingosine 1-phosphate; TG: triacylglycerol

The relative abundance of PC significantly decreased with disease severity, being higher in plasma from controls (27.2%) than in NMIBC (25.2%) and MIBC (23.8%) (Fig. 1B). In contrast, Cer, diacylglycerol (DG), phosphatidylethanolamine (PE), sphingosine-1-phosphate (SphP), and phosphatidylinositol (PI) were significantly increased in BC plasma samples compared to those from controls (Supplementary Figure S1). Additionally, MIBC plasma exhibited significantly higher relative abundance of ceramide-1-phosphate (CerP) and lysophosphatidylethanolamine (LPE) than both NMIBC patients and controls, and higher relative abundance of free fatty acids (FA) and lysophosphatidylcholine (LPC) than controls (Supplementary Figure S1). Analysis of lipid saturation states (Supplementary Figure S2) revealed a decrease in polyunsaturated lipids (p < 0.05) while an increase in monounsaturated lipids (p < 0.01) in NMIBC compared to controls.

To assess whether these lipid changes were associated with patient prognosis, we performed Cox regression and Kaplan-Meier analyses. Cox regression analysis revealed that patients with higher absolute PC levels (>114.13 AU, median) had a significantly reduced risk of death. In the global cohort (NMIBC + MIBC), patients with high PC levels showed a 57% lower risk of death compared with those with low levels (Hazard Ratio [HR]=0.431, 95% Confidence Interval [CI]: 0.198–0.941, p = 0.035). This protective effect was more pronounced in MIBC, where high PC levels associated with a 73% lower risk of death (HR = 0.269, 95% CI: 0.098–0.741, p = 0.011). Kaplan-Meier curves corroborated these findings, showing significantly shorter OS in patients with low PC levels, both in the global cohort (median OS: 82.5 vs. 109.2 months, p = 0.03; Fig. 1C) and more markedly in MIBC (16.0 vs. 109.2 months, p = 0.007; Fig. 1D). A multivariable analysis including PC levels, tumor stage, and grade revealed that tumor stage and PC levels (in the global cohort), as well as PC levels (in MIBC cohort), were independently associated with survival (Table 2). In addition, modeling PC as a continuous variable using natural cubic splines revealed a non-linear association with OS in both the entire cohort and the MIBC subgroup (Supplementary Figure S3). Specifically, higher PC levels were associated with a progressively lower risk of death. In multivariable Cox regression analysis, PC remained significantly associated with OS when modeled as a non-linear term (likelihood ratio test p = 0.03 for the entire cohort, p = 0.005 for the MIBC subgroup), indicating that it provides prognostic information beyond standard clinical variables. MIBC status was independently associated with worse survival (HR = 3.33, 95% CI: 1.42–7.84, p = 0.006) (Supplementary Table S1). Our lipid pathway analysis using BioPAN suggested a pattern consistent with reduced PC synthesis from PE, LPC, and DG, together with increased conversion of PC into LPC and DG in BC patients compared to controls (Fig. 1E–G).

Table 2.

Multivariable Cox regression model for overall survival of BC patients

	Global cohort (n = 87)			MIBC cohort (n = 34)
	HR	95% CI	p-value	HR	95% CI	p-value
Tumor stage (MIBC vs. NMIBC)	4.085	1.737–9.607	0.001	-	-	-
PC levels (High vs. Low)	0.312	0.133–0.730	0.007	0.230	0.078–0.678	0.008

Global cohort refers to all BC patients. CI: confidence interval; HR: hazard ratio; MIBC: muscle-invasive bladder cancer; NMIBC: non-muscle-invasive bladder cancer; PC: phosphatidylcholine

Specific lipid signatures in plasma display potential diagnostic and staging ability for BC

In terms of absolute abundance, plasma from MIBC patients showed a significantly higher total lipid load (521.54 ± 159.83 AU) than NMIBC (446.54±170.92 AU) and controls (297.92±66.66 AU; p < 0.0001) (Fig. 2A). This increase was consistent across all lipid classes (Fig. 2B).

Fig. 2.

BC patients exhibit increased plasma lipid load with tumor severity and distinct lipidomic signatures with diagnostic potential. (A) Violin plot showing total lipid abundance in plasma from controls, NMIBC, and MIBC patients. Kruskal-Wallis test with post hoc Dunn’s test: *p < 0.05, ***p < 0.001, ****p < 0.0001. (B) Heatmap displaying the average relative abundance of each lipid class across the study groups. (C) Principal component analysis (PCA) of plasma lipid profiles in the clinical groups studied. (D) Volcano plot comparing the lipid profiles of BC patients vs. controls. Annotated lipids are colored by lipid class, revealing 58 lipid species significantly upregulated in BC (|log2FC|>1; FDR < 0.05). (E) Bar plot of the pathway enrichment analysis representing the enrichment ratio of upregulated lipids in BC. (F) Unsupervised hierarchical clustering based on the lipid species selected by the elastic net model, illustrating the separation between BC patients and controls. (G) Predicted probability of BC for each individual based on the elastic net model. The vertical dashed line represents the optimal probability threshold (69.83%) for discriminating BC patients (NMIBC in orange and MIBC in red) from controls (green), determined using the youden index. CE: cholesteryl esters; Cer: ceramide; CerP: ceramide 1-phosphate; DG: diacylglycerol; FA: fatty acids; LPC: lysophosphatidylcholine; LPE: lysophosphatidiletanolamine; MG: monoacylglycerol; MIBC: muscle invasive bladder cancer; NMIBC: non-muscle invasive bladder cancer; PC: phosphatidylcholine; PCo|PCp: plasmanyl-glycerophosphocolines/plasmenyl glycerophosphocolines; PE: phosphatidylethanolamine; PEo|Pep: plasmanyl-glycerophosphoethanolamines/plasmenyl-glycerophosphoethanolamines; PI: phosphatidylinositol; SM: sphingomyelin; SphP: sphingosine 1-phosphate; TG: triacylglycerol

To further characterize specific lipid profiles, we performed an unsupervised PCA of precise lipid species, including all samples. The PCA score plot revealed that control samples were more homogeneous than those from BC patients (Fig. 2C). Differential lipid composition analysis identified 58 lipid species significantly increased in BC patients compared to controls (|log₂FC|>1; FDR < 0.05) (Fig. 2D, Supplementary Table S2). Pathway enrichment analysis showed that these increased lipids were associated with sphingolipid, glycerolipid, and FA metabolism, as well as the biosynthesis of phospholipids, plasmalogens, steroids, and bile acids (Fig. 2E). Comparisons between subgroups revealed 122 lipid species significantly elevated in MIBC vs. controls (Supplementary Figure S4 and Table S3) and 44 species elevated in NMIBC vs. controls (Supplementary Figure S5 and Table S4). In contrast, no different lipids were identified between MIBC and NMIBC (Supplementary Figure S6).

To assess the diagnostic potential of individual lipid species, we performed AUC ROC analyses. Cer (d16:1/17:0), Cer(d18:1/12:0), Cer(d34:1), Cer(d18:1/18:0), PC(35:0), and PC(36:0) showed the strongest discrimination ability between BC patients and controls (AUC > 0.85; Table 3, Supplementary Figure S7 and Table S5). To develop a multivariable model integrating clinical covariates, we fitted an elastic net penalized logistic regression model to discriminate BC patients from controls. The model selected eleven lipids and diabetes mellitus status as predictive variables (Fig. 2F). No additional clinical variables were retained by the model. After internal cross-validation, the model achieved an AUC of 0.943 (95% CI: 0.904–0.982, p = 5.48×10⁻¹³), with a positive predictive value (PPV) of 100% and a negative predictive value (NPV) of 68.2% in the study cohort (Fig. 2G).

Table 3.

Diagnostic performance of selected plasma lipid species and multivariable elastic net models for BC detection and staging

BC vs. Control
Lipid species	AUC	95% CI	p-value	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)
Cer(d18:1/12:0)	0.92	0.87–0.97	<2.22 × 10−16	75.9	100.0	100	57.7
Cer(d34:1)	0.92	0.87–0.97	<2.22 × 10−16	85.1	93.3	97.3	66.7
Cer(d18:1/18:0)	0.88	0.82–0.94	<2.22 × 10−16	67.8	100.0	100	50.8
Cer(d16:1/17:0)	0.88	0.82–0.94	<2.22 × 10−16	81.6	86.7	94.6	60.5
PC(35:0)	0.87	0.81–0.94	<2.22 × 10−16	79.3	83.3	93.2	56.8
PC(36:0)	0.85	0.78–0.92	<2.22 × 10−16	65.5	90.0	94.9	46.6
Elastic net model (BC vs. Control)*	0.94	0.90–0.98	5.48 x 10−13	83.9	100.0	100.0	68.2
MIBC vs. NMIBC
Lipid species	AUC	95% CI	p-value	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)
PC(40:6)	0.77	0.66–0.87	6.18 × 10−07	79.4	64.2	58.7	82.9
PCo(44:5)| PCp(46:7)	0.75	0.64–0.86	3.93 × 10−06	82.4	60.4	57.1	84.2
PC(37:4)	0.75	0.64–0.86	5.94 × 10−06	79.4	66.0	60.0	83.3
iAuthor PCp(44:5)	0.74	0.63–0.85	7.00 × 10−06	88.2	54.7	55.6	87.9
SM(d40:3)	0.74	0.63–0.85	8.36 × 10−06	82.4	60.4	57.1	84.2
PC(39:5)	0.72	0.61–0.84	4.87 × 10−05	79.4	62.3	57.4	82.5
Elastic net model (MIBC vs. NMIBC)*	0.86	0.78–0.94	1.69 x 10−08	64.7	92.5	86.6	80.3

*Elastic net models include lipid species selected by penalized logistic regression, integrating lipidomic and clinical variables. Predicted probabilities were used to calculate diagnostic performance metrics. AUC: area under the ROC curve; BC: bladder cancer; Cer: ceramide; CI: confidence interval; MIBC: muscle-invasive bladder cancer; NMIBC: non-muscle-invasive bladder cancer; NPV: negative predictive value; PC: phosphatidylcholine; PCo|PCp: plasmanyl-glycerophosphocolines/plasmenyl glycerophosphocolines; PPV: positive predictive value; SM: sphingomyelin

To distinguish MIBC from NMIBC, PC (37:4), PC(39:5), PC(40:6), PCo(44:5)|PCp(46:7), PCo(44:6)|PCp(44:5), and SM(d40:3) displayed the highest discriminatory capacity (AUC > 0.725; Table 3, Supplementary Figure S8 and Table S6). To develop a multivariable model integrating clinical covariates, we fitted an elastic net penalized logistic regression model to discriminate MIBC from NMIBC patients. The model selected twelve lipids as predictive variables (Supplementary Figure S9). After internal cross-validation, the model achieved an AUC of 0.860 (95% CI: 0.781–0.940, p = 1.69×10⁻⁰⁸), a PPV of 84.6% and an NPV of 80.3% (Supplementary Figure S9).

BC-associated plasma lipidome correlates with NETosis, reduced active DNaseI, and coagulation markers

Next, we analyzed the association of BC-related lipid dysregulation with NETosis, inflammation, and coagulation markers. BC patients showed significantly elevated levels of cfDNA, nucleosomes, elastase, and calprotectin, along with a reduced concentration of active DNaseI, compared to controls (Fig. 3A–E). MIBC patients additionally exhibited higher CRP, fibrinogen, and IL-6 levels than both NMIBC and controls (Fig. 3F–H). Regarding hypercoagulability, BC patients had increased prothrombin F1+2 and TAT than controls (Fig. 3I–J), despite having similar AT levels (data not shown). However, BC patients showed similar thrombin generation capacity, MPO levels, and DNA-citH3 levels compared to controls (Supplementary Figures S10 and S11).

Fig. 3.

Bladder cancer-related lipidomic profiles are associated with a NETosis-related immunothrombotic dysregulation. (A-J) Boxplots showing significantly elevated levels of NETosis, inflammatory, and thrombin generation markers, along with reduced active DNaseI in BC patients compared to controls. Kruskal-Wallis test with Dunn’s post hoc: **p < 0.01, ***p < 0.001, ****p < 0.0001. (K) Spearman correlation matrix including NETosis, inflammatory, and thrombin generation markers, the elastic net model (BC vs. control), and the selected lipid species. Only FDR-adjusted significant correlations are shown (p < 0.05). BC: bladder cancer; Cer: ceramide; cfDNA: cell-free DNA; citH3: citrullinated histone 3; CRP: C-reactive protein; ETP: endogenous thrombin potential; F1+2: prothrombin F1+2 fragment; IL-6: interleukin-6; MG: monoacylglycerol; MIBC: muscle-invasive bladder cancer; MPO: myeloperoxidase; NMIBC: non-muscle-invasive bladder cancer; PC: phosphatidylcholine; PCo: plasmanyl-glycerophosphocolines; PCp: plasmenyl glycerophosphocolines; SM: sphingomyelin; TAT: thrombin-antithrombin complexes; TG: triacylglycerol; ttPeak: time to thrombin peak; VelIndex: velocity index (slope of the thrombin generation curve)

Strong positive correlations among NETosis markers were consistent with a neutrophil origin for these signals (Fig. 3K and Supplementary Table S7). NETosis markers positively correlated with CRP, fibrinogen and IL-6, while these inversely correlated with active DNaseI concentration (Fig. 3K and Supplementary Table S7). The coagulation markers F1+2 and TAT, and the TGT parameters positively correlated with several NETosis markers, fibrinogen, IL-6, and among themselves.

Regarding the lipid species with diagnostic potential, strong positive correlations were observed among most lipids, particularly within lipid classes. In contrast, SM (d35:1) inversely correlated with PCo(16:1/20:5)|PCp(16:0), TG(54:0), and MG(22:2). Most diagnostic lipids positively correlated with NETosis and inflammatory markers, and inversely correlated with active DNaseI. With respect to coagulation-related markers, SM(d35:1) and Cer(d18:1/12:0) positively correlated with TAT, whereas PCo(16:1/20:5)|PCp(16:0) inversely correlated with TGT parameters (Fig. 3K and Supplementary Table S7).

Consistently, the variables retained in the BC vs. control elastic net model positively correlated with NETosis and inflammatory markers, inversely with active DNaseI levels, and strongly correlated with the lipid species included in the model (Fig. 3K and Supplementary Table S7).

Discussion

Aberrant lipid metabolism is increasingly recognized as a hallmark of cancer, driving tumor growth, metastatic spread, and therapy resistance [30, 31]. To sustain malignant progression and adapt to microenvironmental pressures, cancer cells reprogram key metabolic pathways, including de novo lipogenesis, lipid uptake, and phospholipid remodeling. However, the contribution of lipid metabolism to BC pathogenesis remains poorly characterized and limited to tumor tissue [10–13].

This study provides an integrated analysis of plasma lipidomics and the implication of lipid dysregulation in modulating thromboinflammation in BC. We found that BC patients exhibit a systemic reprogramming of lipid metabolism, characterized by an increased total lipid load and a redistribution of lipid classes and species, more profusely in MIBC patients. This systemic enrichment was consistent across all lipid classes, highlighting that a global expansion of the circulating lipid pool characterizes BC. Tissue-based lipidomic studies confirmed a higher total phospholipid and FA content than their benign counterparts [10]. Interestingly, Lee et al. [32] reported ~170% higher total lipid levels in cisplatin-resistant BC cells relative to sensitive cells, linking elevated lipid production to aggressiveness and therapy resistance. Particularly, our data reveal a progressive decrease of PC from controls to NMIBC and MIBC, together with increases in Cer, DG, PE, SphP, and PI. The decrease in PC abundance is particularly noteworthy, since PCs are among the most abundant membrane phospholipids. Nizioł et al. [33] already highlighted lipids (especially glycerophospholipids) as the most discriminant metabolites in serum, with variations in PC/LPC/LPE consistent with membrane remodeling and pathways related to the inflammatory state. In our cohort, the progressive decrease in PC (controls > NMIBC >MIBC) and its worse prognostic association reinforces its systemic and clinically relevant role. A progressive loss of PCs and PEs with advancing stages, accompanied by the accumulation of DGs, has also been documented in BC tissue [10]. Mechanistically, the observed pattern aligns with an increase in PC catabolism and a reduction in the replenishment of PC via synthesis/remodeling routes, as suggested by our BioPAN analysis. This scheme is consistent with (i) increased activity of phospholipases A2 (PLA2), which hydrolyze PC into LPC+FA and have been implicated in tumor inflammation, epithelial-to-mesenchymal transition, and progression; (ii) activation of PC-specific phospholipase C and/or phospholipase D, generating DG; and (iii) a mismatch in the “Lands’ cycle” which normally reacylates LPC to reconstitute PC [34–36]. Furthermore, consistently with other tumors [37], low PC levels associated with worse survival in BC. Vantaku et al. [11] showed that alterations in FA metabolism were strongly linked to poor survival in high-grade BC. Similarly, Feng et al. [12] identified differential lipidomic landscapes between basal and luminal MIBC, where free FAs and sulfatides were enriched in more aggressive basal tumors.

We identified 58 lipid species significantly elevated in BC patients compared to controls, particularly in pathways related to sphingolipid, glycerolipid, FA, and phospholipid metabolism, along with plasmalogen, steroid, and bile acid biosynthesis. These findings are consistent with established hallmarks of lipid reprogramming in cancer, which include upregulated sphingolipid rheostat signaling (Cer/S1P), glycerolipid remodeling (DG/TG turnover), and activation of the steroid/bile acid pathway [38].

BC patients also displayed an increased NETosis, diminished active DNaseI, increased inflammation, and in vivo coagulation markers (F1+2 and TAT). Surprisingly, the ex vivo thrombin generation parameters did not differ between patients and controls. This discrepancy has been previously described for cancer patients [39], suggesting that ex vivo assays may be less sensitive for detecting hypercoagulability in cancer patients than in vivo markers. One plausible explanation could be the depletion of coagulation proteins in plasma samples due to an increased in vivo thrombin generation occurring in the acute phase of the disease. In addition, TGT conditions should be carefully designed to capture subtle procoagulant changes driven by lipids.

Regarding potential diagnostic utility, four Cers and two PCs achieved strong discrimination between BC patients and controls (AUC > 0.85), comparable to the performance of lipid species in other cancers [8]. Additionally, we fitted an elastic net penalized logistic regression model to discriminate BC patients from controls, which achieved an AUC of 0.943 and reached a PPV of 100%, outperforming previously proposed plasma and urine metabolite panels that often lack sufficient specificity [40, 41]. The increase in Cers is consistent with reports showing ethnic-specific sphingolipid alterations in BC tissues, together with reductions in PCs and elevations in DGs [13]. Regarding potential staging utility, three PCs, two glycerophosphocholines, and SM(d40:3) were able to discriminate MIBC from NMIBC (AUC~0.72–0.78). Although the identified lipid profile showed promising staging performance (AUC = 0.8602), with high specificity (92.5%) and a high positive predictive value (84.6%), supporting its potential for confirming MIBC presence, its moderate sensitivity underscores the need for further refinement, which represents a limitation of this study. This finding highlights the challenge of plasma-based discrimination between invasive and non-invasive disease states, which may be confounded by shared systemic alterations. Future studies may improve specificity and overall performance by increasing sample size, stratifying patients according to tumor grade, and implementing multivariable models that integrate multiple lipid species with additional molecular layers, such as microRNA profiles [4] and other omics approaches, including transcriptomics or proteomics [12], as well as relevant clinical variables and cytology data. Such integrative strategies could enable the development of more comprehensive diagnostic and staging scores, potentially enhacing sensitivity while preserving the high specificity observed. Direct comparisons between the proposed lipid models and voided urinary cytology were not feasible as cytology data were available for only 5% of patients, representing an additional limitation of this study. Overall, these integrative approaches may improve risk stratification, facilitate early detection of MIBC, and support more personalized therapeutic decision-making in BC. Although key clinical variables (sex, smoking status, metabolic comorbidities, statin use, and prior treatment) were incorporated into multivariable penalized regression models to mitigate confounding, residual confounding cannot be fully excluded. In addition, the sex imbalance in our cohort, reflecting the epidemiology of BC, limits the generalizability of the findings to female BC patients, given potential sex-related differences in lipid metabolism, hormonal regulation, and immune responses. Cross-validation was applied to reduce overfitting, but the models were both developed and internally validated within the same cohort. Moreover, no prior sample size calculation was performed, which may affect the robustness of both diagnostic and survival analyses. Thus, independent validation in larger, multicenter, prospective cohorts with adequate representation across sex and treatment strata should be a primary objective of future research. This would enable sufficiently powered subgroup analyses (e.g., sex-stratified and BCG-stratified models) and help confirm whether these lipid signatures are independently and consistently associated with BC biology. In addition, future studies incorporating patients with suspected BC or other urological malignancies as control groups would allow a more rigorous evaluation of the identified lipid signatures.

Next, we aimed to unravel the association of BC-related lipid dysregulation with NETosis, inflammation, and coagulation markers. Lipid composition critically modulates neutrophil function and the formation of NETs [15], while both lipids and NETs contribute to coagulation [42] and thrombosis [25, 43, 44]. We recently reported that BC patients exhibit increased NETosis together with a reduced active DNaseI concentration, which impaired NET degradation [19], findings that were confirmed herein. Interestingly, lipid species with potential diagnostic value showed direct correlations with NETosis markers (DNA-citH3, calprotectin, elastase, and calprotectin) and inflammatory markers (CRP, fibrinogen, and IL-6), and inverse correlations with active DNaseI. Nevertheless, given that both lipid metabolism and NETosis are influenced by inflammatory states, the observed lipid-NET associations may be partially confounded, particularly in controls with prior urological conditions. However, the markedly higher inflammatory burden observed in BC patients likely mitigates this potential bias. In line with our findings, changes in the SM/Cer balance have been shown to induce NET formation [15]. Moreover, sphingolipid pathways are known regulators of neutrophil trafficking, survival, and activation [45]. Specifically, S1P-S1PR signaling prolongs neutrophil lifespan and promotes recruitment, while ceramide-pathway enzymes influence neutrophil effector functions, including ROS production and NETosis [17, 46]. Collectively, our data support a model in which systemic lipid remodeling in BC could foster a NET-permissive, pro-inflammatory state. Nevertheless, the observational nature of this study causal inference. Future mechanistic studies are warranted to ascertain whether the identified lipid species actively regulate NETosis and DNaseI activity. For instance, co-culture systems combining BC cells with primary human neutrophils could be used to evaluate whether specific Cer or PCs directly modulate NET formation and degradation. Furthermore, pharmacological targeting of the axis PC-LPC through inhibition of Lp-PLA₂ with darapladib [47], blockade of glucosylceramide synthesis with PDMP [48], or manipulation of the SM/Cer balance via sphingomyelinase [15] could help delineate whether these lipid alterations act as upstream drivers of thromboinflammatory signaling or represent downstream consequences of tumor-associated metabolic reprogramming. Ultimately, in vivo models and longitudinal patient studies will be required to establish temporal relationships and confirm the functional relevance of lipid-neutrophil interactions to BC progression.

Regarding hypercoagulability, several NETosis markers correlated with coagulation and thrombin generation markers. In particular, SM(d35:1) and Cer(d18:1/12:0) positively correlated with TAT, whereas PCo(16:1/20:5)|PCp(16:0) inversely correlated with the TGT parameters. Our results align with previous findings showing a modulatory effect on coagulation of lipids like glucosylCer [43] and acylcarnitines [44]. Specific clotting assays and the TGT should be conducted in the presence or absence of dysregulated lipids to further confirm their functional involvement in mediating a hypercoagulable state in BC patients.

A key limitation of this study is its cross-sectional and observational nature, which precludes causal inference between lipidomic alterations, disease progression, and thromboinflammatory activation. The associations identified herein should therefore be interpreted as hypothesis-generating. The observed lipid changes may reflect tumor-driven metabolic reprogramming, systemic inflammation, or host responses to cancer progression. However, given the well-described ability of specific lipid species to regulate neutrophil activation, NETosis, and coagulation pathways, it remains biologically plausible that these alterations contribute functionally to the pro-thromboinflammatory state observed in BC. Future longitudinal and mechanistic studies will be required to determine whether these lipid alterations actively contribute to disease progression or represent downstream consequences of tumor-host interactions.

Conclusions

Our study demonstrates that BC is associated with systemic lipid metabolism reprogramming, characterized by increased total plasma lipid load and a progressive depletion of PC, which correlates with poor prognosis. Specific lipid species, particularly Cers and PCs, show compelling diagnostic and staging potential, enabling minimally invasive disease stratification. Importantly, the dysregulated lipid profile could be functionally linked to enhanced NETosis and inflammation, reduced active DNaseI, and thrombin generation, suggesting a role in promoting a pro-inflammatory, thromboinflammatory state. These findings support a model in which lipid remodeling could contribute to immune dysfunction in BC, offering novel insights into disease biology.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

AU

arbitrary units

AUC

area under the curve

BC

bladder cancer

BCG

Bacillus Calmette-Guerin

CE

cholesteryl esters

Cer

ceramide

CerP

ceramide 1-phosphate

cfDNA

cell-free DNA

CIS

carcinoma in situ

citH3

citrullinated histone 3

DG

diacylglycerol

ETP

endogenous thrombin potential

FA

fatty acids

IL-6

Interleukin-6

LPC

lysophosphatidylcholine

LPE

lysophosphatidiletanolamine

MG

monoacylglycerol

MIBC

muscle-invasive bladder cancer

MPO

myeloperoxidase

NETs

Neutrophil extracellular traps

NMIBC

non-muscle invasive bladder cancer

NPV

negative predictive value

OS

Overall survival

PC

phosphatidylcholine

PCA

principal component analysis

PCo|PCp

plasmanyl-glycerophosphocolines/plasmenyl glycerophosphocolines

PE

phosphatidylethanolamine

PEo|Pep

plasmanyl-glycerophosphoethanolamines/plasmenyl-glycerophosphoethanolamines

PI

phosphatidylinositol

PLA2

phospholipases A2

PPV

positive predictive value

ROC

receiver operating characteristic

SRED

single radial enzyme-diffusion

SM

sphingomyelin

SphP

sphingosine 1-phosphate

TG

triacylglycerol

TGT

Thrombin generation test

ttPeak

time to thrombin peak

VelIndex

velocity index

Author contributions

Conception and design: PM, JO, and MR. Acquisition of data: JO, MR, JP-A, CDV-D, MM-S, and SB. Analysis and interpretation of data: JO, AH-P, MR, and PM. Drafting of the manuscript: JO, AH-P, and PM. Critical revision of the manuscript for important intellectual content: PV, RH, EP, MR, P-A, CDV-D, M-S, and SB. Statistical analysis: JO, AH-P, and MR. Obtaining funding: JO, AH-P, EP, CDV-D, and PM. Administrative, technical or material support: PV, RH, and EP. Supervision: PM.

All authors read and approved the final version of the manuscript. PM had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Funding

This research was supported by research grants from Instituto de Salud Carlos III (ISCIII) (PI20/00075, FI21/00171, PI23/00449, FORT23/00021, PI24/01607, CD25/00146) co-funded by the European Regional Development Fund “A way to make Europe” or co-funded by the European Union. J.O. is supported by a postdoctoral grant from Sociedad Española de Trombosis y Hemostasia (SETH 2023). A. H.-P- is supported by the Scientific Foundation of the Spanish Association against Cancer and the European Union through the AECC Talent Programme (GA101081298) (TALEN246729HERR).

Data availability

The MS lipidomics data have been deposited in the public repository Zenodo (https://zenodo.org) with the data set identifier DOI 10.5281/zenodo.17158525.

Declarations

Ethics approval and consent to participate

The study was approved by the institutional ethics committee of La Fe University and Polytechnic Hospital (refs. 2014/0314 and 2020–101-1). Written informed consent was obtained from all participants. The study was conducted in accordance with the principles of the Declaration of Helsinki (Edinburgh amendment, 2000).

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.