Multi-omics Integration Analysis Robustly Predicts High-Grade Patient Survival and Identifies CPT1B Effect on Fatty Acid Metabolism in Bladder Cancer

Bladder cancer causes significant morbidity and mortality worldwide. In the United States, there were about 79,000 cases and 17,000 deaths due to bladder cancer in 2017 (1). In the past few decades advances in cancer genomics, transcriptomics, proteomics, and metabolomics resulted in the discovery of potential biomarkers for cancer (2, 3). Unfortunately, most of the biomarkers have failed to demonstrate superior performance characteristics compared with existing clinical tests. Recent advancements in omics technologies have ushered in a new era of targeted cancer therapies by improving our understanding of molecular carcinogenesis. Omics studies have identified therapeutic targets and resulted in the successful development of new drugs to treat various solid tumors like breast cancer and lung cancer (4, 5). For bladder cancer, prognostication and treatment still depend mainly on pathologic and clinical characteristics (6). Patients diagnosed with high-grade invasive disease are difficult to treat effectively and have a relatively low life expectancy despite available multimodal therapies. Therefore, novel prognostic and therapeutic targets against bladder cancer are needed.

Tumor progression relies on the reprogramming of cellular metabolism (7). Our previous metabolomic and lipidomic studies highlighted the significance of altered xenobiotic and fatty acid metabolism in bladder cancer development (8–10). However, we still know little about the altered metabolic pathways during the transition from low-grade to high-grade bladder cancer. The integration of omics facilitates the study of interactions among all classes of biomolecules that can occur in cell, which in turn, determines the cellular physiology and behavior. Treatment for high-grade muscle-invasive bladder cancer (MIBC) has not advanced beyond cisplatin-based combination chemotherapy in the past three decades and very few drugs for the disease have been approved recently (11). The identification of markers to predict poor outcomes among patients with bladder cancer will improve our ability to identify patients who might benefit from adjuvant therapy. Furthermore, a greater understanding of the metabolic molecular mechanisms of bladder cancer progression and the identification of novel therapeutic targets will improve outcomes for patients with high-grade bladder cancer.

We used a robust mass spectrometry (MS) platform to conduct global unbiased metabolomic and lipidomic analyses to identify critical alterations that may contribute to bladder cancer progression. We mapped the altered metabolites and lipids to corresponding genes using the Human Metabolome Database (HMDB) and then a pathway analysis to discover the ways in which metabolic pathways are perturbed in high-grade bladder cancer. We integrated the relevant genes with publicly available transcriptomic data and generated an integrated gene signature to predict the survival of patients with bladder cancer. Furthermore, we found that carnitine palmitoyl transferase 1B (CPT1B) is significantly downregulated in high-grade bladder cancer, which in turn leads to low fatty acid β-oxidation (FAO). Ectopic expression of CPT1B in high-grade bladder cancer cells increased FAO, which in turn reduced epithelial–mesenchymal transition (EMT). Finally, overexpression of CPT1B in high-grade bladder cancer cells reduced cell proliferation and liver metastasis in a chick chorioallantoic membrane (CAM) in vivo model. Our results clearly indicate that CPT1B plays an important role in bladder cancer tumor progression by affecting fatty acid metabolism. CPT1B might therefore have prognostic value and provide a target for therapeutic intervention in high-grade bladder cancer.

High-performance liquid chromatography (HPLC)-grade ammonium acetate, acetonitrile, methanol, chloroform, and water were procured from Burdick & Jackson. MS grade formic acid, standards and internal standards, N-acetyl aspartic acid-d3, tryptophan-15N2, sarcosine-d3, glutamic acid-d5, thymine-d4, gibberellic acid, trans-zeatin, jasmonic acid, anthranilic acid-15N, and testosterone-d3 were purchased from Sigma-Aldrich. Mixture of LPC 17:0/0:0, PG 17:0/17:0, PE 17:0/17:0, PC 17:0/17:0, TAG 17:0/17:0/17:0, SM 18:1/17:0, MAG 17:0, DAG 16:0/18:1, CE 17:0, ceramide 18:1/17:0, PA 17:0, PI 17:0/20:4, and PS 17:0/17:0. were obtained from Avanti polar lipids (Supplementary Table S1).

RT4, T24, UMUC3, J82, and TCCSUP were procured from ATCC and maintained as per ATCC instructions. All cell lines were verified using short tandem repeat DNA fingerprinting at the MD Anderson Cancer Center (Houston, TX) and were tested for Mycoplasma contamination every 3 months. All the bladder cancer cell lines were procured from the ATCC less than 2 years before starting the experiments.

All the bladder cancer specimens were procured by a prior written informed consent under Institututional Review Board–approved protocols. Metabolites were extracted from bladder cancer tissues, and mouse liver pool was used as a quality control and followed the extraction procedure described previously (8, 12–14). Briefly, 10 mg of tissue was used for the metabolic extraction. The extraction step starts with addition of 750 μL ice-cold methanol: water (4:1) containing 20 μL spiked internal standards (ISTD). After homogenization, ice-cold chloroform and water were added in a 3:1 ratio for a final proportion of 4:3:2 methanol:chloroform:water. The organic and aqueous layers were collected, dried, and resuspended in methanol: water (1:1). The extract was deproteinized using a 3-kDa molecular filter and the filtrate was dried under vacuum. The dried extracts were resuspended in 100 μL of injection solvent composed of 1:1 methanol:water and subjected to LC-MS.

Lipids were extracted using a modified Bligh–Dyer method (15). The extraction was carried out using 2:2:2 ratio of water:methanol:dichloromethane at room temparature after ISTDs into tissues and quality control pool (12). After homogenization of the samples, the organic layer was collected and completely dried under vacuum. Before MS analysis, the dried extract was resuspended in 100 μL of buffer containing 10 mmol/L NH₄Ac and subjected to LC/MS. The lipidome was separated using reverse-phase (RP) chromatography. To monitor the lipid extraction process, we used a standard pool of tissue samples from aliquots of the same samples.

For metabolomics, 5 μL of the metabolite extract was injected into a 3.5-μm particle 4.6 × 150 mm X bridge amide column, which heats to 60°C. Formic acid (0.1%) in water was used as solvent-A and acetonitrile as solvent-B in positive ionization and 20 mmol/L ammonium acetate in 95% water; 5% acetonitrile and 100% acetonitrile as solvent-B in negative ionization mode. For lipidomics, 10 μL of the lipid extract was injected into a 1.8 μm particle 50 × 2.1 mm column (Acquity HSS UPLC T3), which heats to 55°C. Acetonitrile: water (40:60, v/v) with 10 mmol/L ammonium acetate was solvent-A and acetonitrile: water: isopropanol (10:5:85 v/v) with 10 mmol/L ammonium acetate was solvent-B. Chromatographic elution used a linear gradient over a 20-minute total runtime, with 60% solvent-A and 40% solvent-B gradient in the first 10 minutes. Then, the gradient was ramped in a linear fashion to 100% solvent-B, which was maintained for 7 minutes. After that, the system was switched back to 60% solvent-B and 40% solvent-A for 3 minutes. The flow rate used for these experiments was 0.4 mL/minute. The data acquisition of each sample was performed in both positive and negative ionization modes using a TripleTOF 5600 equipped with a Turbo VTM ion source. The instrument performed one TOF MS survey scan (150 ms) and 15 MS/MS scans with a total duty cycle time of 2.4 milliseconds (ms). The mass range in both modes was 50 to 1,200 m/z. We controlled the acquisition in both MS and MS/MS spectra by data-dependent acquisition function of the Analyst TF software (AB Sciex). Rolling collision energy spread was set whereby the software calculated the collision energy value to be applied as a function of m/z. Mass accuracy was maintained by the use of an automated calibrant delivery system interfaced to the second inlet of the Duo Spray source.

The raw data (.wiff) of metabolomics and lipidomics from AB Sciex are converted to “.mgf” data format using proteoWizard software (16). We used the NIST MS Pep to search the converted files against NIST14 libraries (17, 18) for metabolomics and LipidBlast libraries for lipidomics. The m/z width was determined by the mass accuracy of ISTDs and was set 0.001 for positive mode and 0.005 for a negative mode with an overall mass error of less than 2 parts per million. The minimum match factor used was set to 400 score. The MS/MS identification results from all the files were combined using an in-house software tool to create a library for quantification. All raw data files were searched against this library of identified metabolites with mass and retention time using Multiquant 1.1.0.26 (AB Sciex, USA). We used MS/MS data as an intermediary step to help with identification, but quantification was done using MS1 data only. Relative abundance of peak spectra was used for the analyses. The metabolites identified in both positive and negative ion modes were initially analyzed separately for their relationship with the outcome to ensure persistent results. Identified metabolites were quantified by normalizing against their respective ISTDs. Quality control samples were used to monitor the overall quality of the metabolite extraction and MS analysis.

After data acquisition, the missing values for metabolites/lipids were imputed using the k-nearest neighbors method. Then the data were log₂ transformed followed by normalization using the median interquantile range normalization. The compound-by-compound t-test was applied, followed by the Benjamin–Hochberg procedure for FDR correction accounting for multiple comparisons. Significance was achieved for FDR < 0.25.

Absolute concentrations of metabolites were quantified by AbsoluteIDQ p180 Kit (Biocrates Life Science AG) using QTRAP 6500 LC/MS/MS system (AB Sciex) equipped with an electrospray ionization (ESI) source, an Agilent G1367B autosampler and the Analyst 1.51 software (AB Sciex). Ten microliters of the homogenized tissue sample, quality control samples, blank, and calibration standards were added to the appropriate wells. The plate was dried and the samples were derivatized with phenylisothiocyanate and dried again. The samples were eluted with 5 mmol/L ammonium acetate in methanol and then were diluted with running solvent provided in kit for flow injection analysis. Twenty microliters of the sample was directly injected into the MS at a flow of 30 μL/minute. Concentrations were calculated and evaluated in the Analyst/MetIQ software by comparing measured analytes in a defined extracted ion count section to those of specific labeled internal standards or nonlabeled, nonphysiologic standards (semiquantitative) provided by Biocrates. Measurement of overall β-oxidation activity was derived by calculating the ratio of short-chain acylcarnitines to free carnitines.

Enriched pathways were determined by using the hypergeometric distribution against the Kyoto encyclopedia of genes and genomes (KEGG) pathway database, as compiled by the MSigDB gene set compendium; significance was achieved for P < 0.05. Pathway networks were visualized using the Cytoscape scientific visualization platform.

Metabolites and lipids were mapped into associated genes/proteins using the HMDB ID. Next, we obtained the transcriptome fingerprint between high and low grade bladder cancer using the Kim cohort as described below. We generated an integrated metabolomics/lipidomics/transcriptomics gene signature, by intersecting the all three gene sets.

To identify the gene signature of high versus low grade bladder cancer, we downloaded the publicly available data from the Kim cohort. Gene significance was assessed by using the Student t test; significance was achieved for P < 0.05 and fold change either above 1.25 or below 0.8. To identify genes associated with low versus high CPT1B gene expression, we used the gene expression data from five independent public cohorts: The Cancer Genomic Atlas (TCGA), Kim (GSE13507), Sjodahl (GSE32894), Lindgren (GSE32548), and Riester (GSE31684). For each cohort, we first stratified the patients by CPT1B, into the top 50% and bottom 50%. Next, we generated a rank file for each expressed gene by the log₂ fold change of low CPT1B samples over high CPT1B samples. We ran Gene Set Enrichment Analysis (GSEA) against the HALLMARK pathway collection, as compiled by Molecular Signature Database (MSigDB). Significance was achieved for FDR-adjusted Q-value < 0.25.

Significant association with survival of an integrated bladder cancer metabolomics/lipidomics/transcriptomic gene signature was evaluated in four gene expression datasets of bladder cancer cohorts from TCGA, Kim, Sjodahl, and Lindgren for which clinical outcomes have been reported. For each gene in the signature and for each specimen, we computed the Z-score for its expression within the cohort, as described previously (9) and computed the sum Z-score for each specimen. Specifically, the Z-scores of genes repressed in the integrated signature were subtracted from the Z-scores of genes induced in the integrated signature, resulting in a corresponding signature activity score for each specimen. In each cohort, the specimens were ranked according to their integrated signature activity score, and association with survival was evaluated via the log-rank test between the bottom 50% of the patients and the top 50% of the patients. Survival significance was assessed by employing the package survival in the R statistical system.

Survival analysis was performed using the survival library in the R statistical system. Univariate analysis used 27 gene signature or the CPT1B with survival, after stratifying patients in top and bottom 50%. For multivariate models, we tested the gene signature and CPT1B, while controlling for other demographic and clinical factors, specifically age, sex, and tumor stage. HRs derived from Cox proportional hazards models are expressed with 95% confidence intervals and P values. For all analyses and each variable, significant association with survival was achieved for P < 0.05.

The CAM was assessed as described previously (19) on embryonic day 7; the eggs were inoculated with 5 × 10⁵ UMUC3 vector control and CPT1B-overexpressing cells per egg. In vivo luminescence imaging was conducted on embryonic days 10, 13, and 15 (3, 5, and 7 days after seeding the tumor cells) by using in vivo imaging system (IVIS). Exposure time was 1 second, 5 minutes after addition of 100 μL of 15 mg/mL d-luciferin, and region of interest was quantitated by the Lumina software. Seventh day after inoculation, the eggs were euthanized as per the AVMA guidelines. The CAM tumors were fixed in 10% formalin and embedded in paraffin for IHC. IHC was performed for CPT1B, E-cad, N-cad, and vimentin to check the expression levels. IHC-stained slides were scanned and analyzed by Aperio CS2 system (Leica Biosystems) and number of positive cells/total cells were plotted.

To analyze metastasis of the tumor into the chick embryo, we harvested embryo visceral organs (i.e., lung, liver, bone, and brain) on day 15 to 18 postfertilization. To achieve this, the eggshell opening was widened to allow excision of the CAM tumor. After the tumor excision, the embryo was humanely euthanized by decapitation with a scissors and the body incised to dissect visceral organs including lung, liver, bone, and brain. Once dissected, the tissues were collected, flash-frozen, and stored at −80°C until further analysis. Quantitative assessment of tumor metastasis into visceral organs was achieved by quantification of human DNA from the metastasized cells in the chicken DNA by performing RT-PCR for human-specific Alu sequence.

To quantify the amount of human DNA from tumor cells that had metastasized to the chick embryo tissues, a standard curve was generated by amplification of genomic DNA isolated from a serial dilution of bladder cancer cells mixed with individual chick visceral tissue DNA (modified from published methods; refs. 20, 21). To detect the human Alu sequences in the chicken DNA background, human Alu-detecting primers and a probe were used (Supplementary Table S2). The standard curve was generated using serial dilutions (0.1 ng–1 μg) of UMUC3 DNA mixed with constant amount of chicken DNA. The RT-PCR was performed in triplicate for each standard as well as the experimental samples and analyzed the DNA content in control and CPT1B overexpression groups.

We characterized the global metabolome and lipidome of low-grade (n = 5) and high-grade (n = 20) bladder cancer tissues using high-resolution LC-MS and clinical information for the patients is shown in Supplementary Table S3. A cocktail of 10 internal standards along with a repetitive analysis of pooled liver extracts served as controls to ascertain the reproducibility and robustness of the profiling platform for metabolomics (Supplementary Figs. S1A and S2A) and lipidomics (Supplementary Figs. S1B and S2B). An extensive NIST database and lipid blast search analysis of the LC-MS data identified a total of approximately 2,000 metabolites and 786 lipids in positive and negative ionization modes (Supplementary Fig. S3). An overview of the study is shown in Fig. 1. A total of 519 metabolites (14 different classes) were significantly altered (FDR < 0.25) between low-grade and high-grade bladder cancer (Fig. 2A; Supplementary Table S4). These include significantly altered levels of metabolites in multiple pathways (Fig. 2D). Likewise, a total of 19 lipids (7 different classes) were significantly altered (FDR < 0.25; Fig. 2B; Supplementary Table S5), that were involved in many pathways (Fig. 2E). Using well-annotated, publicly available Kim transcriptomics data we found 3036 genes that had significantly altered (P < 0.05) between low-grade and high-grade bladder cancer (Fig. 2C). Like the altered lipids and metabolites the altered genes were involved in multiple pathways (Fig. 2F) and 51 common pathways were implicated in all three platforms (Fig. 2G).

We mapped the panels of altered metabolites, to their corresponding genes using HMDB (10, 22), resulting in the identification of 2,311 genes. A similar analysis of altered lipids led to the identification of 173 genes. Analysis of transcriptomics public data from the Kim cohort resulted in the identification of 3,036 genes that were differentially expressed between low-grade and high-grade bladder cancer. Integration of these three platforms led to 27 intersecting genes between low-grade and high-grade bladder cancer (Fig. 3A).

We next examined the expression of these 27 genes using Kim transcriptomics data. Eleven of these genes were upregulated while the other 16 were downregulated in bladder cancer (Fig. 3B). We examined the association between the 27 genes and survival using publicly available bladder cancer cohorts [Kim (23), Lindgren (24), and Sjodahl (25)]. We used the log-rank test to determine patient outcomes. In all three cohorts, high expression levels of the 11 genes that were upregulated in bladder cancer was associated with poor survival. Similarly, low expression levels of the genes that were downregulated in bladder cancer was associated with poor survival in all three cohorts (Fig. 3C). Interestingly, 11 upregulated gene signature that were upregulated in bladder cancer in the Kim cohort were associated with poor survival in Lindgren and TCGA bladder cancer cohorts (Supplementary Fig. S4).

We analyzed univariate and multivariate Cox proportional hazards regression models using the Kim, Lindgren, Sjodahl, TCGA datasets to assess the associations between the 27 gene signature and the time-to-death. Using multivariate models, we tested the 27 gene signature (top 50% vs. bottom 50%; Supplementary Table S6A–S6D) while controlling for sex, age, and tumor stage. Univariate Cox proportional hazards regression analysis showed that the gene signature was significantly associated with risk of death, with an HR above 1 in the Kim, Lindgren, Sjodahl and TCGA cohorts. Despite that, a multivariate analysis showed that although the HR was greater than 1 in the Kim, Lindgren, and TCGA, the 27 gene signature was not associated with survival independently of the other clinical variables considered. The tumor stage was associated with survival in all four cohorts, while age was associated with survival in Kim, Lindgren, and TCGA cohorts (Supplementary Table S6A—S6D).

To obtain additional insights into the 27 genes, we performed survival analysis based on the individual genes in the same publicly available cohorts. We found that six upregulated genes (PLA2G4C LIPG, PIGS, PIGU, ATP8B2 and ATP8B4) had a significant clinical association with survival (Supplementary Fig. S5). High expression of PLA2G4C and ATP8B2 was associated with poor survival in multiple cohorts (Kim, Lindgren, Sjodahl, and TCGA; ref. 26; Supplementary Fig. S5). Eight of the downregulated genes (CPT1B, PIGB, PLA2G4A, PIGV, INPP5D, PLCD3, PIGZ, and PLA2G10) were associated with poor survival (Supplementary Fig. S6).

Remarkably, low expression of CPT1B was associated with poor survival in TCGA (Fig. 4A) Lindgren, and Sjodahl cohorts (Supplementary Fig. S6), whereas in the Kim cohort there was no association between CPT1B and survival. However, the expression of CPT1B was significantly lower in high grade bladder cancer tumors in all three cohorts (Supplementary Fig. S7). We further tested the univariate and multivariate models for CPT1B (top 50% vs. bottom 50%) by considering for sex, age, and tumor stage using Kim, Lindgren, Sjodahl, and TCGA datasets. Univariate Cox proportional hazards regression analysis showed that CPT1B had an HR below 1 in TCGA, Lindgren, and Sjodahl cohorts and was significantly associated with survival only in the TCGA cohort (Supplementary Table S7A–S7D). In a multivariate Cox proportional hazards regression model adjusted for sex, age, and tumor stage showed that CPT1B had an HR below 1 and was significantly associated with survival in the TCGA cohort (Supplementary Table S7A–S7D). Pathway analysis also indicated that fatty acid metabolism was significantly increased in high-grade bladder cancer. CPT1B is involved in fatty acid transport and acts as a rate-limiting enzyme for FAO, thereby altering the fate of the cancer cells (27). Recent studies showed that cancer metastatic potential is a consequence of an alteration in lipid metabolic pathways (28–30), which led to our hypothesis that CPT1B plays a crucial role in bladder cancer progression.

To better understand the role of CPT1B in high-grade bladder cancer, we examined the expression of CPT1B in low-grade and high-grade bladder cancer. CPT1B expression was significantly lower in high-grade bladder cancer than in low-grade bladder cancer (Fig. 4B) in an independent validation cohort (Supplementary Table S8). Histochemical analysis of a tissue micro array (TMA; Fig. 4C) confirmed the findings of the transcriptomic cohorts analyses (Supplementary Fig. S7). It is known smoking is associated with high-grade bladder cancer (31). In comparison with nonsmokers, smokers with bladder cancer had reduced expression of CPT1B (Fig. 4D). Likewise, the levels of CPT1B were also reduced in high-grade compared with low-grade bladder cancer cell lnes (Fig. 4E). Consistent with the CPT1B expression, high-grade bladder cancer cell lines displayed low FAO activity compared to low-grade bladder cancer cell lines (Fig. 4F). We measured the levels of acylcarnitines, which are indicators of FAO, and found that the levels of palmitoyl and octanoyl carnitines were significantly lower in high-grade bladder cancer patients (Fig. 4G). Taken together, our results demonstrated the co-occurrence of CPT1B downregulation and low FAO levels in high-grade bladder cancer.

To obtain additional insights into the effects of CPT1B on metabolism in high-grade bladder cancer, an in silico analysis was performed on data divided into CPT1B high versus CPT1B low signature using public gene expression data sets. For each transcriptomics cohort that profiled the CPT1B gene (TCGA, Kim, Sjodahl, and Lindgren), we stratified the patients based on CPT1B gene expression. Next, we divided the patients based on the median CPT1B value into top 50% and bottom 50%. We then computed a gene signature using a parametric t-test and a minimum difference in expression of 1.25 fold between the patients in the top 50% of CPT1B expression and bottom 50% of CPT1B expression. For each cohort, the signature consisted of the genes that were differentially expressed by at least 1.25-fold between the patients with high and low CPT1B expression, respectively. We referred the obtained signatures as “CPT1B high vs. CPT1B low” signatures. We used GSEA method to determine the enriched pathways of the CPT1B high versus CPT1B low signatures in each cohort. A total of 17 pathways were uniformly enriched (FDR-adjusted Q-value < 0.25) in all five datasets (Fig. 5A). These pathways include inflammation/immune system, metabolism, and oncogenic pathways (Fig. 5A). Notably, E2F targets and the EMT pathway were enriched with a strict Q-value < 0.0001, further supporting the biological relevance of CPT1B in bladder cancer progression (Fig. 5B).

To determine the effects of CPT1B, in high-grade bladder cancer, we ectopically expressed the CPT1B in high-grade UMUC3 bladder cancer cells (Fig. 6A). The CPT1B overexpressed cells displayed increased FAO activity (Fig. 6B), with increased levels of acyl carnitines and reduced levels of fatty acids compared with control cells (Fig. 6C). EMT is one of the major hallmarks of the metastasis in high-grade cancer. The CPT1B overexpressed cells displayed decreased levels of the EMT markers vimentin and snail compared with vector control (Fig. 6D).

Given that CPT1B loss and FAO disruption led to significantly increased activity in pathways associated with EMT and invasion (Fig. 5), we hypothesized that CPT1B overexpression leads to a reduction in the metastatic capabilities of cancer cells. To functionally test that hypothesis, we performed in vivo CAM xenograft assay (32) using CPT1B overexpressing cells. Compared to vector control UMUC3 cells, CPT1B overexpression cells demonstrated reduced tumor growth in the xenograft assays (Fig. 6E and F). IHC confirmed that the CPT1B overexpression CAM tumors had high E-cad levels, but low N-cad and vimentin levels (Fig. 6G). We tested the metastatic potential of the CPT1B overexpression cells by Alu PCR. There was decreased metastasis to liver of the chick embryo from CPT1B overexpression tumors than controls (Fig. 6H). Overall, our results confirmed the role of CPT1B in tumor growth and metastasis in high-grade bladder cancer. Taken together, our results suggested that CPT1B downregulation impairs FAO and increases the tumor growth and metastasis in bladder cancer (Fig. 6I).

Treatments for high-grade muscle-invasive and metastatic bladder cancer have not advanced beyond cisplatin-based combination chemotherapy (33). The median survival time of patients with high-grade MIBC is only 14 to 15 months with cisplatin-based chemotherapy (15), highlighting the need for novel therapeutic approaches. Genomic integrative studies showed that genes involved in the PI3-kinase/AKT/mTOR, RTK/RAS, and CDKN2A/CDK4/CCND1 pathways are associated with high-grade bladder cancer (26). We used integrated approach combining unbiased metabolomics, lipidomics, and transcriptomics to investigate the progression of high-grade bladder cancer and to delineate opportunities for therapeutic and prognostic intervention.

Alteration of metabolism is one of the hallmarkers for cancer progression (34). Metabolic alterations enable cancer cells to meet the increased energy demands for proliferation and survival in the nutrient-deprived tumor microenvironment (35). Our metabolomics analysis identified 519 differentially metabolites that were differentially produced in high-grade bladder cancer, including glycolysis, TCA cycle, PPP pathway, polyamines, bile acids, carnitine, prostaglandins, and nucleotides indicating that metabolism is a highly dynamic process in bladder cancer. Unbiased lipidomics analysis identified 19 differentially produced lipids in high-grade bladder cancer, including triglycerides, phosphatidylcholines, phosphatidylethanolamines, and phosphatidyl inositols. The higher levels of triglycerides and low levels of phosphatidylcholines and phosphatidylethanolamines are in line with previous studies (10). We used transcriptomics data from the Kim cohort (26) to identify genes that were differentially expressed in high-grade bladder cancer. We mapped the altered metabolites and lipids to their corresponding genes and then to enriched pathways. The results from metabolomics, lipidomics, and transcriptomics suggest that there were common deregulated metabolic pathways implicated the arachidonic acid, phosphatidylinositol, and glycerophospholipid metabolism. These pathways are important in various cancers (36, 37) (38, 39) and will be focal points in future bladder cancer research. Clinical trials of drugs targeting these metabolic alterations in bladder cancer patients are warranted.

Our integrative multi-omics analysis identified 27 common genes that were implicated by metabolomics, lipidomics, and transcriptomics data. Each of these genes has a profound impact on the cellular metabolism, and particularly lipid metabolism highlighting the importance of fatty acids in cell proliferation and tumorigenesis (40). The upregulated genes (PLTP, PLA2G4C, LIPG, PIGS, PIGT, PIGU, PLCXD1, ATP8B2, PIGY, PTDSS1, ATP8B4) and downregulated genes (PIGB, PIGG, PIGN, PLCE1, PLA2G4A, CPTIB,PIGV, ATP8B1, CEL, PLA2G4F, PLCG2, PLA2G4B, INPP5D, PLCD3, PIGZ, PLA2G10) provided a gene signature that was associated with poor survival in Lindgren, Kim, and Sjodhal cohorts. Further characterization and validation of this gene signature may provide a prognostic marker for high-grade bladder cancer. Some of these genes in the signature have been reported to be involved in the metastasis of other cancers; for example, PLTP, and LIPG upregulation in breast cancer (41, 42), ATP8B1 in colorectal cancer (43), and PLCG2 in chronic lymphocytic leukemia (44). So far, the role of these genes in high-grade bladder cancer remains unclear. Further characterization and functional validation of the genes could provide insights into the molecular events that are related to progression of bladder cancer from low-grade to high-grade.

Our integrated analysis of these omics datasets revealed the previously unappreciated importance of CPT1B in high-grade bladder cancer. CPT1B resides at the outer mitochondrial membrane and regulates intracellular lipid metabolism by transporting long-chain fatty acids into the mitochondria for β-oxidation (45, 46). Overexpression of fatty acid synthase (47), leads to greater accumulation of fatty acids and low expression of CPT1B in high-grade bladder cancer. That, in turn, supports the requirement of actively proliferating cells for membrane structural and signaling lipids. FAO and fatty acid synthesis are mutually antagonistic process (48). The low abundance of palmitoyl and octanoyl carnitine and low FAO activity observed in patients with high-grade bladder cancer are consistent with low CPT1B expression observed in patient with high-grade bladder cancer and cell lines. GSEA of genes associated with low CPT1B expression in five distinct public bladder cancer cohorts revealed alterations of many immunological, metabolic, and oncogenic pathways that are known hallmarks of cancer. Low CPT1B expression was associated with high levels of EMT which regulates metastasis in cancer by conferring an invasive phenotype (49), is the salient feature of high-grade bladder cancer. CPT1B over expression reduced the cell proliferation and EMT markers supporting the relevance of FAO in cellular transformation and bladder cancer progression towards high grade disease. Furthermore, CPT1B over expression in in vivo significantly reduced tumor growth, and metastasis, suggesting that a strategy to increase the CPT1B levels in high grade bladder cancer might help to inhibit tumor growth and metastasis.

In conclusion, for the first time, we identified critical alterations of metabolite and lipid profiles that contribute to bladder cancer progression by levaraging high resolution LC-MS based unbiased metabolomics, and lipidomics in combination with bioinformatics analysis. Our integrated study of low-grade versus high-grade bladder cancer provided numerous novel insights into disease biology and delineated pathways that provide potential opportunities for therapeutic intervention. The low levels of FAO and downregulation of CPT1B in high-grade bladder cancer indicate that the impairment of β-oxidation plays an important role in bladder cancer tumor progression. These results may have prognostic value and provide targets for therapeutic intervention in the future. Taken together, our results suggest that interventions that increase the CPT1B levels may have therapeutic value for high-grade bladder cancer patients that are notably dependent on FAO metabolism.

A.G. Sikora is an employee of Ovodex, reports receiving commercial research grants from Tessa Therapeutics, other commercial research support from Advaxis, and holds ownership interest (including patents) in patents related to CAM model. No potential conflicts of interest were disclosed by the other authors.

Conception and design: V. Vantaku, C.R. Ambati, S.R. Donepudi, V. Putluri, F.-C. von Rundstedt, H. Villanueva, C. Coarfa, N. Putluri

Development of methodology: V. Vantaku, J. Dong, C.R. Ambati, D. Perera, S.R. Donepudi, V. Putluri, B. Karanam, C. Coarfa, N. Putluri

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.R. Ambati, S.R. Donepudi, V. Putluri, M. Villanueva, F.-C. von Rundstedt, L.Y. Ballester, M.K. Terris, R.J. Bollag, S.P. Lerner, A.B. Apolo, H. Villanueva, A.G. Sikora, Y. Lotan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Vantaku, J. Dong, D. Perera, C.S. Amara, M.J. Robertson, D.W.B. Piyarathna, M. Villanueva, M.K. Terris, H. Villanueva, M. Lee, A.G. Sikora, Y. Lotan, C. Coarfa, N. Putluri

Writing, review, and/or revision of the manuscript: V. Vantaku, J. Dong, C.R. Ambati, D. Perera, S.R. Donepudi, C.S. Amara, F.-C. von Rundstedt, L.Y. Ballester, S.P. Lerner, A.B. Apolo, H. Villanueva, A.G. Sikora, Y. Lotan, A. Sreekumar, C. Coarfa, N. Putluri

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Villanueva, H. Villanueva, A. Sreekumar, C. Coarfa

Study supervision: Y. Lotan, C. Coarfa, N. Putluri

Other (in vitro/in vivo experiments): S.S. Ravi

This research was fully supported by American Cancer Society (ACS) Award 127430-RSG-15-105-01-CNE (N. Putluri), NIH/NCI R01CA220297 (N. Putluri), and NIH/NCI R01CA216426 (N. Putluri), partially supported by the following grants: NIH/NCI U01 CA167234 (A.S.K), CPRIT RP170295 (C.C.), as well as funds from Alkek Center for Molecular Discovery (A. Sreekumar). This project was also supported by the Agilent Technologies Center of Excellence (COE) in Mass Spectrometry at Baylor College of Medicine, Metabolomics Core, Human Tissue Acquisition and Pathology at Baylor College of Medicine with funding from the NIH (P30 CA125123), CPRIT Proteomics and Metabolomics Core Facility (N. Putluri; RP170005), and Dan L. Duncan Cancer Center. CAM assay was supported by the Patient-Derived Xenograft and Advanced in vivo Models Core Facility at Baylor College of Medicine with funding from the Cancer Prevention and Research Institute of Texas (CPRIT) grant #170691. We would like to thank the team of the Georgia Cancer Center Biorepository/BRAG-Onc for biospecimen collection and annotation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.