Predicting Poor Efficacy of Cabozantinib in Advanced Renal Cell Carcinoma by Exploratory Metabolic Pathway Analysis and In Vitro Verification Authors: Takanari Sakai, Yoshihide Kawasaki, Izumi Sakamoto, Eiji Hishinuma, Naomi Matsukawa, Daigo Chiba, Tomonori Sato, Kento Morozumi, Masamitsu Maekawa, Nariyasu Mano, Akihiro Ito Published: 2026 ABSTRACT ABSTRACTBackgroundCabozantinib (CABO) is currently an essential drug for first‐line and sequential therapy in the treatment of advanced renal cell carcinoma (RCC). Predicting its therapeutic efficacy will enable efficient drug selection and improve the prognosis of patients. We conducted an exploratory study to predict the effects of CABO using metabolite analysis with clinical and in vitro samples.MethodsPatients diagnosed with RCC at our hospital between September 2011 and August 2023 who were pathologically diagnosed with clear cell RCC and subsequently received systemic therapy for disease progression or recurrence were included. Patient background, treatment course, and prognosis were retrospectively investigated. Patients were divided into CABO responders and poor responders based on survival analysis. Comprehensive metabolite analysis of serum samples before CABO administration was performed using a non‐targeted analysis system (Orbitrap) with a mass spectrometer. The significance of metabolic pathways related to the poor response of CABO analysis was evaluated by establishing CABO‐resistant renal cancer cell lines and conducting in vitro quantitative analysis.ResultsA total of 19 analyzable cases were identified during the study period, and seven pathways (methylhistamine metabolism, caffeine metabolism, beta‐alanine metabolism, histidine metabolism, glycine and serine metabolism, tryptophan metabolism, and purine metabolism) and five metabolites (methylhistidine, kynurenic acid, xanthine, 7‐methylxanthine, and L‐2‐amino‐3‐oxobutanoic acid) related to the poor efficacy of CABO were identified in a two‐group comparison. Quantitative metabolite analysis of CABO‐resistant cell lines established from three different renal cancer cell lines (786‐O, ACHN, and Caki‐1) confirmed the identified pathways and metabolites in relation to CABO resistance.ConclusionWe identified key metabolites and associated metabolic pathways that may serve as predictive biomarkers for suboptimal cabozantinib efficacy in patients with advanced RCC. While further validation in independent cohorts is warranted, our findings underscore the potential of metabolomic profiling to refine therapeutic selection and improve clinical outcomes. Abbreviations advanced renal cell carcinoma cabozantinib clear cell renal cell carcinoma cabozantinib resistance glutamine glutamate hydrophilic interaction chromatography liquid chromatograph—mass spectrometer overall survival progression‐free survival renal cell carcinoma resistance index tryptophan vascular endothelial growth factor receptor xanthine oxidoreductase Background Cabozantinib (CABO) is a molecularly targeted drug that inhibits multiple receptor tyrosine kinases, including vascular endothelial growth factor receptors (VEGFR) 1, 2, and 3, mesenchymal epithelial transition factor (MET), and AXL [1, 2]. CABO is considered an important drug in both first‐line and second‐line treatment for patients with advanced renal cell carcinoma (aRCC) [3]. Therefore, it is important to properly assess the efficacy of CABO and appropriately modify treatment to improve the prognosis of patients with aRCC. The identification of biomarkers capable of predicting the therapeutic efficacy of CABO would optimize treatment strategies and improve outcomes in aRCC patients [4]. To date, the number of clinically useful biomarkers for predicting response to treatment in RCC remains limited [5]. Metabolic reprogramming is a hallmark of cancer and is reflected in alterations of cellular metabolite profiles. Metabolomics enables comprehensive profiling of metabolites in biological samples and has emerged as a promising approach for identifying biomarkers and therapeutic targets in various cancers. Although CABO is an effective drug, its therapeutic effect varies among patients, and no reliable biomarkers have been established to predict its effectiveness. So far, some studies have reported several molecular mechanisms underlying tyrosine kinase inhibitor (TKI) resistance in renal cell carcinoma (RCC). These include activation of the PI3K/AKT/mTOR pathway, reactivation of alternative angiogenic signaling, and adaptation to hypoxia. However, these mechanisms mainly focus on signal transduction and gene mutations, and these have not yet been established as predictive markers for TKI resistance [6, 7]. Furthermore, the cancer metabolic reprogramming in RCC patients treated with CABO remains poorly understood. Previously, we performed metabolite analysis of clinical samples to identify new biomarkers that could predict the diagnosis and recurrence of RCC [8, 9]. In this study, we used non‐targeted metabolomics to measure serum metabolites before CABO administration to identify the metabolites associated with the poor efficacy of CABO in patients with advanced renal cancer. Furthermore, we quantitatively measured the metabolites identified in established CABO‐resistant renal cancer cells to examine the metabolic changes and significance of CABO resistance. We aimed to verify whether these metabolites could predict the poor efficacy of CABO, the mechanisms of CABO resistance, propose effective therapeutic drug selection methods to overcome resistance, and improve the prognosis of patients with aRCC. Material and Methods Design and Study Cohort Patients who underwent radical or partial nephrectomy at Tohoku University Hospital between September 2011 and August 2023 had a pathological diagnosis of clear cell RCC (ccRCC), subsequently received systemic therapy for disease progression or recurrence, and received CABO as sequential therapy were included in this study. Patients with a history of malignant tumors or those undergoing hemodialysis were excluded from the study. Blood samples were collected before CABO administration. For serum sample collection, patients were kept fasting from 7 PM the previous day, and blood was collected around 6 AM before breakfast on the day of collection. The blood was centrifuged at 1450 × g for 10 min at 4°C, and the resulting supernatant serum was aliquoted into two or more vials and immediately stored at −80°C until analysis. Serum thawing for metabolite analysis was performed only once. Medical records were reviewed to collect clinical information, including age, sex, clinical stage, pathological diagnosis, prior systemic therapy, and survival time. Pathological evaluation was performed by a urological pathologist at our hospital, based on the American Joint Committee on Cancer staging system, 7th edition. To identify significant metabolites that predict the poor efficacy of CABO sequential therapy in aRCC, we compared serum metabolites between poor responders and responders at a median progression‐free survival (PFS) of 9 months after CABO sequential therapy. In our analysis of PFS, the Kaplan–Meier curves showed that half of the analyzed cases experienced early disease progression before the median PFS; therefore, we used the median PFS as the measure of treatment response. This study was approved by the Institutional Review Board of Tohoku University Hospital, and written informed consent was obtained from all study participants. Non‐Targeted Metabolomics Analysis of Serum Samples Methanol (150 μL) was added to 50 μL of the serum sample, mixed, and sonicated for 5 min. The sample was centrifuged at 15,000 × g for 5 min at 4°C, and the supernatant was diluted twofold with water and vortexed. The supernatant was centrifuged again at 15,000 × g for 5 min at 4°C and subjected to analysis. Liquid Chromatograph—Mass Spectrometer (LC–MS/MS) analysis was performed using a Q Exactive mass spectrometer (Thermo Fisher Scientific) connected to a Vanquish UHPLC system, and Xcalibur version 4.2.47 was used for instrument control. The analysis was performed using electrospray ionization in four modes: reversed‐phase, Hydrophilic Interaction Chromatography (HILIC) separation, positive ion, and negative ion modes. In reversed‐phase mode, an Aquity UPLC HSS T3 column (1.8 μm, 2.1 × 150 mm, Waters) was used with a gradient of 0.1% formic acid in water and 0.1% formic acid in acetonitrile as the mobile phase. In HILIC mode, a ZIC‐pHILIC column (5 μm, 2.1 × 100 mm, Merck) was used with a gradient of 10 mM ammonium bicarbonate aqueous solution and acetonitrile as the mobile phase. The column temperature was 40°C in both cases, and 2 μL of the sample was injected. Details of the detection conditions for the mass spectrometer are shown in Table S1, and of the separation conditions for the liquid chromatograph are shown in Table S2. Metabolic pathways and metabolites in relation to the poor efficacy of CABO were identified using metabolite set enrichment analysis of metabolites that were significantly increased or decreased between the effective and poorly responding groups. The metabolite set enrichment analysis is based on a library of predefined metabolic pathways, their metabolite sets, and biochemical functional groups. It statistically verifies whether highly variable listed metabolites are concentrated in specific metabolite sets greater than chance, and ranks all metabolites based on their variability. Cell Culture and Verification of Established CABO‐Resistant Cell Lines Using a WST Assay Human RCC cell lines (786‐O (CVCL_1051), ACHN (CVCL_1067), and Caki‐1 (CVCL_0234)) were purchased directly from the American Type Culture Collection (ATCC, Manassas, VA, USA). All cell lines were certified and characterized according to ATCC's strict quality control standards before distribution. The 786‐O cells were maintained in RPMI‐1640 medium (Nacalai Tesque, Tokyo, Japan) supplemented with 10% FBS in an incubator at 37°C and 5% CO2. ACHN and Caki‐1 cells were cultured in DMEM (Gibco/Life Technologies, Carlsbad, CA, USA) supplemented with 10% FBS. All the cell lines used were routinely treated with anti‐mycoplasma reagents MC‐210 (WakenBtech, Kyoto, Japan) to eliminate potential mycoplasma contamination. Each cell line was seeded at 2 × 106 cells/dish for subsequent analysis and cultured at 37% CO2 and 5% CO2. After 72 h of culture, the cells were washed twice with D‐PBS, harvested with a scraper, and stored at −80°C until assayed. Subsequently, the CABO‐resistant RCC cell lines were established. To generate cabozantinib‐resistant RCC cell lines, parental cells were maintained in medium supplemented with CABO on a continuous basis. For Caki‐1 and ACHN cells, exposure began at 1 μM; for 786‐O cells, it began at 2 μM. Cells were kept at each concentration until proliferation rates returned to a stable baseline, at which point CABO levels were raised in 1 μM increments. This stepwise escalation was carried out over approximately 9 months, culminating in the respective final maintenance concentrations. The 786‐O and Caki‐1 cells were resistant to CABO (ChemScene, Monmouth Junction, MJ, USA) at a final concentration of 20 μM. These cell lines were designated 786‐O‐CR and Caki‐1‐CR, respectively. The parental ACHN cell line was treated in the same manner, but was resistant to CABO at a final concentration of 15 μM and designated ACHN‐CR. After resistance was established, the cell lines were maintained in medium containing CABO at the respective final concentrations. Parental cell lines that were not exposed to TKIs were designated as CABO‐sensitive cell lines (786‐O‐WT, ACHN‐WT, and Caki‐1‐WT). The cytotoxicity of CABO was evaluated using a lactate dehydrogenase (LDH) Water‐Soluble Tetrazolium (WST) assay kit (Dojindo Molecular Technologies, Tokyo, Japan). The 786‐O and 786‐O‐CR cells were seeded at 2000 cells/well, Caki‐1 and Caki‐1‐CR at 5000 cells/well, and ACHN and ACHN‐CR at 3000 cells/well in 96‐well plates. CABO was added at various concentrations, and the cells were cultured for 96 h. Cell counts were determined using Cell Count Reagent SF (Nacalai Tesque, Tokyo, Japan), and absorbance was measured at 450 nm using a microplate reader (Apollo LB911 SD, Berthold Japan, Tokyo, Japan). Cell viability was calculated by normalizing the blank‐corrected absorbance of CABO‐treated wells to that of vehicle‐treated control wells, using the following formula: Cell viability (%) = [(A450 of CABO‐treated wells−A450 of blank wells)/(A450 of vehicle‐control wells−A450 of blank wells)] × 100. IC50 values were calculated using the dose–response curve [10]. The concentration–response curves were fitted using a four‐parameter logistic model. IC50 values were calculated from the fitted curves. The four‐parameter logistic model was defined as Y = Bottom + (Top − Bottom)/[1 + 10((LogIC50 − X) × HillSlope)], where Y represents cell viability, and X represents the logarithm of CABO concentration. The resistance index (RI) was calculated as follows: RI = IC50 of CABO‐resistant cells/IC50 of parental cells. CABO resistance was considered to be established when the resistant cells showed a rightward shift of the concentration–response curve and/or an RI > 1 compared with the corresponding parental cells. For cell lines in which IC50 could not be determined within the tested concentration range, increased residual viability at high CABO concentrations and increased AUC were used as supportive validation metrics. Most experiments using cell lines were performed within 24 months and 10 passages after purchase from the ATCC. Cellular Sample Preparation for LC–MS/MS Ice‐cold isopropanol (40 μL) was added to each 1 × 106 cell pellet, and the mixture was vortexed and sonicated for 5 min. After centrifugation at 15,000 × g and 4°C for 5 min, 10 μL of the supernatant was used for LC–MS/MS analysis. Targeted metabolite quantification was performed using an Xevo TQ‐XS (Waters, LCC) and an AQUITY UPLC I class (Waters, MA, USA) system according to previously published methods [11, 12, 13]. Statistical Analysis Clinical variables are presented as median values. Statistical analyses were performed using GraphPad Prism 10 (GraphPad Software LLC) and MetaboAnalyst 6.0. Survival curves were estimated using the Kaplan–Meier (KM) method, and differences between groups were assessed using the log‐rank test. Statistical significance was defined as p < 0.05. For non‐targeted metabolomics analysis, metabolites showing significant differences between CABO responders and poor responders were extracted using volcano plot analysis based on fold change (> 2 or < 0.5) and p‐value (< 0.05). Metabolite set enrichment analysis was subsequently performed using MetaboAnalyst 6.0 with curated metabolic pathway libraries, including Kyoto Encyclopedia of Genes and Genomes (KEGG)‐related metabolite sets. Enrichment results were visualized using pathway‐associated bar plots and metabolic pathway mapping to facilitate biological interpretation of metabolomic alterations associated with poor CABO response. A predictive model for poor CABO efficacy was constructed using multiple logistic regression analysis based on the identified metabolites. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curve analysis according to the area under the curve (AUC), sensitivity, and specificity. A predictive model for the poor efficacy of CABO was created using metabolites identified using non‐targeted metabolomics and multiple logistic regression analysis. Results Patient Characteristics and Efficacy of Cabozantinib in Sequential Therapy We identified 19 patients with ccRCC who progressed after first‐line therapy and received CABO as a sequential therapy. Blood samples were collected before CABO administration, and prognostic analysis was performed. Table 1 shows the characteristics of the 19 patients. The median age was 59 years, 84% were male, and 69% were classified as intermediate or poor according to the International Metastatic RCC Database Consortium (IMDC) risk classification. Histopathological diagnosis revealed sarcomatoid features in 26% of the cases. TABLE 1Clinical characteristics of patients in the CABO‐responder and ‐poor responder groups.All (N = 19)Responders (N = 9)Poor responders (N = 10)Age (median)59 (49–78)66 (49–82)59 (50–68)Sex (M/F)16/3 (84%)7/2 (77.8%)9/1 (90%)cT stage1, 27 (36.8%)4 (44.4%)3 (30%)3, 412 (63.2%)5 (55.5%)7 (70%)Visceral metastasis8 (42%)44IMDC riskFavorable6 (31%)51Intermediate10 (53%)37Poor3 (16%)12Pathology with sarcomatoid5 (26%)231st line therapyIO combination10 (52.6%)46Other9 (47.3%)54Line2nd9 (47%)453rd≦10 (53%)55BORPR, SD11 (57.9%)91PD8 (42.1%)09OS (month)47 (10–142)79 (31–142)40 (10–63)PFS (month)9 (0–41)29 (13–41)4 (0–9) Clinical characteristics of patients in the CABO‐responder and ‐poor responder groups. In all patients, no complete response (CR) was observed, and 57.9% achieved stable disease or better. As shown in the survival analysis using KM curves (Figure 1), the median PFS from the start of CABO administration was 9 months. When patients were divided into two groups based on the median PFS [a CABO‐poor responder group (10 patients) and a CABO‐responder group (nine patients)], the median PFS was 4 months in the CABO‐poor responder group and 29 months in the responder group. Progression‐free survival (PFS) of patients with advanced renal cancer who received cabozantinib (CABO) as sequential therapy. (A) PFS after CABO administration in all patients. (B) Comparison of PFS between the CABO‐poor responder group and CABO‐responder group; (C) overall survival (OS) from the start of systemic therapy in all patients; (D) comparison of OS from the start of systemic therapy between the CABO‐poor responder and CABO‐responder group. The black line indicates CABO responders and the red line indicates poor CABO responders. CABO, cabozantinib. Identification of Metabolic Pathways and Metabolites Involved in Poor Efficacy of Cabozantinib in Sequential Therapy Serum metabolites before CABO administration were measured in all patients using non‐targeted metabolomics. Volcano plots were used to compare the two groups in four modes (Tables S1 and S2): C18 (positive and negative) and HILIC (positive and negative). Twenty‐eight metabolites were detected in the CABO‐poor responder group that showed a fold change of ≥ 2 or ≤ 0.5 and were significantly different (p < 0.05) from those in the CABO‐responder group (Figure 2). Table 2 lists the 28 metabolites identified from the metabolite database. Non‐targeted metabolomics were performed in the cabozantinib (CABO)‐responder and non‐responder groups using four modes. After data acquisition and cleaning, multivariate analysis was performed for each mode. Metabolites showing significant differences (red dots) were extracted from the volcano plots in (A) C18 (positive), (B) C18 (negative), (C) HILIC (positive), and (D) HILIC mode (negative) based on fold change > 2 or < 0.5 and p < 0.05. CABO, cabozantinib; HILIC, hydrophilic interaction chromatography. TABLE 2List of serum metabolites showing significant changes in the CABO‐poor responders.MetaboliteFold change p FDRMode(Carbamoylamino) (4‐hydroxyphenyl) acetic acid5.555240.04210.048769HILIC NegaNitrosoguvacoline4.28740.04950.049582HILIC PosiMethylhistidine3.36800.03950.050431HILIC Nega7‐Methylxanthine3.30830.03620.051539HILIC Posi2,6‐Dideoxy‐3‐O‐methyl‐L‐arabino‐hexopyranose3.12240.00400.037626HILIC NegaIsobutyrylglycine2.97000.01770.038562HILIC NegaOxalosuccinic acid2.80080.00760.056555HILIC NegaVal‐pro2.77000.0230.047982HILIC PosiXanthine2.67660.04070.050225HILIC NegaN (2)‐succinyl‐L‐ornithine2.65970.03860.05106HILIC NegaHistidinol2.61000.01580.036576HILIC Posi2‐Methyl‐3‐methoxypyrazine2.55040.04820.049629HILIC PosiL‐2‐Amino‐3‐oxobutanoic acid2.50160.04200.050169HILIC Nega6‐Acetyl‐2‐(hydroxymethyl)‐2,3‐dihydro‐4 (1H)‐pyridinone2.47570.01480.042333HILIC Nega13 (S)‐HpOTrE2.39720.03050.053824HILIC NegaProclavaminic acid2.38880.03160.050971HILIC Posiγ‐Guanidinobutyrate2.38230.04670.050843HILIC PosiKynurenic acid2.29340.04810.050858HILIC Nega3‐Thiotolene2.23600.01080.044474HILIC NegaAceturic acid2.22430.01500.039679C18 PosiSuccinic anhydride2.22430.01500.037034C18 PosiEntadamide A2.20250.00020.008495HILIC Posi(4S)‐4‐[(2E)‐2‐Octenoyloxy]‐4‐(trimethylammonio)butanoate2.18250.04640.052104HILIC NegaGamma‐Hydroxybutyric acid2.09810.00800.049637HILIC NegaHydroxylysine2.03090.00860.045817C18 Posi4‐Methyleneglutamine0.43030.00260.049833HILIC PosiBut‐2‐ynedioic acid0.30180.02420.047295HILIC Nega2‐(4‐Methylthiazol‐5‐yl) ethyl butyrate0.22200.03810.052339C18 Posi List of serum metabolites showing significant changes in the CABO‐poor responders. The metabolite set enrichment analysis of these 28 metabolites using MetaboAnalyst 6.0 identified seven pathways (methylhistamine, caffeine, beta‐alanine, histidine, glycine and serine, tryptophan, and purine metabolism) in relation to the ineffectiveness of CABO (Figure 3A). Methylhistamine metabolism is most likely involved in the CABO‐poor response. Further extraction of metabolites in relation to these seven metabolic pathways using MetaboAnalyst 6.0 ultimately identified five metabolites (methylhistidine, kynurenic acid, xanthine, 7‐methylxanthine, and L‐2‐amino‐3‐oxobutanoic acid) involved in the poor efficacy of CABO (Figure 3B). MetaboAnalyst 6.0 identified metabolic pathways and metabolites associated with poor CABO efficacy through the metabolite set enrichment analysis. (A) Important metabolic pathways were identified through the metabolite set enrichment analysis, integrating metabolites showing significant differences for each of the four categories. Seven pathways involving metabolites were extracted. The longer the bar in the graph, the greater the metabolic pathway relates to poor CABO efficacy. (B) Of the metabolites in relation to the seven identified pathways, five (methylhistidine, kynurenic acid, xanthine, 7‐methylxanthine, and L‐2‐amino‐3‐oxobutanoic acid) were significantly associated with poor CABO efficacy. CABO, cabozantinib. The asterisk indicates p < 0.05. Based on the identified metabolites and metabolic pathways, a metabolic map (Figure 4) that correlated with the poor efficacy of CABO was created. The map was condensed into four metabolic pathways: amino acid, indole, purine, and lactate metabolism. Cases in which these metabolic pathways fluctuate may result in a CABO‐poor response. The metabolic pathways and metabolites in the metabolic pathway map derived from non‐targeted metabolite analysis. In particular, the amino acid metabolic, indole metabolic, purine metabolic, and lactate metabolic pathways were shown to be important metabolic pathways related to poor efficacy of CABO. CABO, cabozantinib. Establishment of Cabozantinib‐Resistant Renal Cancer Cell Lines Three different renal carcinoma cell lines (786‐O, ACHN, and Caki‐1) were designated as CABO‐sensitive (786‐O‐WT, ACHN‐WT, and Caki‐1‐WT) and used to generate CABO‐resistant cell lines (786‐O‐CR, ACHN‐CR, and Caki‐1‐CR). For 786‐O, the IC50 exceeded the maximum test concentration (60 μM) and was right‐censored for both WT and CR cell lines. However, the concentration‐response curves for CR cell lines were generally shifted upward (rightward) compared with those of the WT cell lines. The survival rate at the highest concentration and area under the survival curve from 0 to 60 μM was high for CR cell lines (Figure 5A). The lower limit estimates of the four‐parameter logistic (4PL) fit also tended to be higher for the CR cell lines. This concentration‐dependent increase in the survival rate of CR cell lines indicated the establishment of CABO‐resistant cell lines. Cytotoxicity analysis using the lactate dehydrogenase (LDH) Assay Kit‐WST in three renal cancer cell lines. The horizontal axis represents cabozantinib (CABO) concentration and the vertical axis represents cell survival rate. For each cell line ((A) 786‐O, (B) ACHN, (C) Caki‐1), the CABO‐sensitive cell line (WT) represents a red line, and the CABO‐resistant cell line (CR) represents a blue line. CABO, cabozantinib; CR, CABO‐resistant; LDH, lactate dehydrogenase. Although ACHN cells tended to be resistant to CABO, the survival curve measured using the WST method shifted to the right in the CR group (Figure 5B). The IC50 was higher for ACHN‐CR (7.49 μM) than that for ACHN‐WT (5.96 μM), resulting in a RI of 1.26. Furthermore, for Caki‐1, CR did not achieve 50% inhibition within the assay range (Figure 5C). Linear interpolation estimated the IC50 to be 12.6 μM for WT and IC50 > 23.05 μM for CR, exceeding the maximum concentration tested. The RI was calculated as 23.05/12.6 > 1.83. Furthermore, the lower limit estimates of the 4PL fit tended to be high for ACHN‐CR and Caki‐1‐CR (i.e., a lower maximum inhibition rate), suggesting an increased survival rate at higher concentrations. We concluded that CABO‐resistant cell lines had been established in ACHN and Caki‐1 because the RI exceeded 1 in the CR groups and the IC50 difference was statistically significant. A comparison of IC50 values between RCC cell lines is shown in Table 3. TABLE 3Comparison of IC50 among RCC cell lines.IC50 Cabozantinib sensitiveCabozantinib resistance786‐O786‐O: 10.2 μM (95% CI: 9.7–10.7)786‐O CR: > 60 μMACHNACHN: 6.0 μM (95% CI: 5.2–6.8)ACHN CR: 8.2 μM (95% CI: 6.2–10.1)Caki‐1Caki‐1: 10.4 μM (95%: CI 9.1–11.8)Caki‐1 CR: 67.0 μM (95% CI: 60.2–74.7) Comparison of IC50 among RCC cell lines. In Vitro Analysis of Identified Metabolites Associated With Poor Efficacy of Cabozantinib Twenty‐eight metabolites identified by non‐targeted metabolomics from clinical samples (Table 2) were quantitatively measured, and seven metabolites, including glutamine (Gln) and glutamate (Glu), associated with the four fluctuated metabolic pathways based on the metabolite set enrichment analysis and our previous studies [8, 9], were compared between WT and CR cell lines. All three CR cell lines showed changes in metabolites, suggesting enhanced metabolic pathways, which were observed in clinical samples of patients with a poor response to CABO (Figure 6). Quantitatively measured metabolites in the CABO‐resistant cell lines (CR) compared with those in their respective parent lines (CT). (A) 786‐O, (B) ACHN, and (C) Caki‐1 changes in concentrations of seven metabolites related to metabolic pathways identified by non‐targeted metabolomics of clinical samples. The unspecified values for glutamate concentrations in ACHN and Caki‐1 indicate that they significantly exceed the maximum value of the calibration curve, 800 μM. 1‐Met‐His, 1‐methylhistidine; 3‐Met‐His, 3‐methylhistidine; CABO, cabozantinib; CR, CABO‐resistant; CT, control; Gln, glutamine; Glu, glutamate; Trp, tryptophan.* represents p < 0.05, and ** represents p < 0.01. In the primary tumor‐derived cell line 786‐O, significant differences were observed in Gln and methylhistidine (1−/3‐Met‐His), indicating the involvement of amino acid metabolic pathways in CABO resistance. The differences in tryptophan (Trp) suggested the involvement of the indole metabolic pathway, and the difference in hypoxanthine indicated the influence of the purine degradation pathway, including xanthine/7‐methylxanthine. Glu and hypoxanthine in the metastatic tumor‐derived cell lines, ACHN and Caki‐1 cells, exhibited opposite behaviors to those of 786‐O cells. Although an increasing trend of Glu and hypoxanthine was observed in 786‐O cells, no significant difference was observed in ACHN cells. In Caki‐1 cells, significant differences were observed for 3‐Met‐His and xanthine. Like 786‐O, the difference in Trp in ACHN cells and Caki‐1 cells suggested the involvement of the indole metabolic pathway, and the difference in hypoxanthine suggested the influence of purine degradation (xanthine/7‐methylxanthine). In WT and CR 786‐O and Caki‐1 cells, substantial changes were observed in each metabolite; however, in ACHN cells, the changes were not substantial, indicating that the survival rate of CR was not notably higher than that of WT. As described above, the behavior of serum metabolites associated with CABO resistance is not always consistent, depending on the characteristics of each cell line. However, we found that changes in multiple amino acid metabolic pathways (Glu, Met‐His), indole metabolic pathways (Trp), and purine metabolic pathways (xanthine, 7‐methylxanthine) dynamically fluctuated with CABO resistance. Therefore, in vitro analysis demonstrated that measurement of these metabolites may be able to predict the poor clinical efficacy of CABO. Prediction of Poor Efficacy of Cabozantinib Using Serum Metabolites A predictive model for the poor efficacy of CABO was created using five metabolites identified using non‐targeted metabolomics and multiple logistic regression analysis (Figure 7A–E). Each ROC curve in Figure 7E was generated by combining two (red), three (green), four (blue), or five (light blue) metabolites to maximize the AUC. The combination of five metabolites showed the largest AUC; the ROC curve [logit(P) = log (P/(1−P)) = 2.318 + 0.826 7‐methylxanthine−1.423 kynurenic acid + 3.922 xanthine + 2.424 methylhistidine + 2.598 L‐2‐amino‐3‐oxobutanoic acid] yielded an AUC of 0.723 (95% CI: 0.304–1) (Figure 7G). The PFS was compared using KM curves between the high‐ and low‐metabolite groups and normalized using the median values of each of the five metabolites identified in non‐targeted metabolomics (Figure 8A–E). Measurement of the five metabolites in serum showed that a poor response to CABO could be predicted in the high‐level group, but due to the small number of cases in this analysis, no statistically significant difference was observed (Figure 8F). The ROC predictive model for poor CABO efficacy for each of the identified metabolites: (A) 7‐methylxanthine, (B) kynurenic acid, (C) L‐2‐amino‐3‐oxobutanoic acid, (D) xanthine, and (E) methylhistidine. (F) Each ROC curve was created by combining two (red), three (green), four (blue), or five (light blue) metabolites to maximize the AUC. (G) Using five metabolites identified by non‐targeted metabolomics, a multiple logistic regression analysis was performed to generate a receiver operating characteristic curve (ROC) [logit(P) = log(P/(1−P)) = 2.318 + 0.826 7‐methylxanthine−1.423 kynurenic acid + 3.922 xanthine + 2.424 methylhistidine + 2.598 L‐2‐amino‐3‐oxobutanoic acid] predicting poor response to CABO. AUC, area under the curve; CABO, cabozantinib; ROC, receiver operating characteristics. F: PFS was calculated by dividing the median value of each metabolite into high (red line) and low (black line) values (A) 7‐methylxanthine, (B) kynurenic acid, (C) L‐2‐amino‐3‐oxobutanoic acid, (D) xanthine, (E) methylhistidine. L‐2‐amino‐3‐oxobutanoic acid alone significantly predicted poor CABO response, but no significant differences were observed for the other four metabolites. (F) PFS was compared between the high and low groups using the KM curves for the normalized and additive medians of the five metabolites identified using non‐targeted metabolomics. CABO, cabozantinib; KM, Kaplan–Meier; PFS, progression‐free survival. Discussion The strength of this study is that serum metabolite measurements before CABO administration revealed the metabolic pathways and metabolites involved in the poor efficacy of CABO in sequential therapy for patients with advanced renal cancer. Additionally, we used in vitro analysis of CR lines to establish the significance of these metabolites in relation to CABO resistance. CABO was approved for the treatment of metastatic RCC based on the METEOR [14] and CABOSUN trials [15]. The efficacy of CABO in sequential therapy is maintained regardless of the line of therapy, from the first to the fourth line of therapy [16]. Furthermore, the combination of nivolumab and CABO has been established as a first‐line treatment, demonstrating significant advantages over sunitinib in terms of PFS and overall survival (OS) [17, 18]. Therefore, CABO has become a key drug in the treatment of advanced renal cancer; assessing the efficacy of CABO and changing treatment regimens are crucial for improving the prognosis of patients with aRCC. Metabolic abnormalities are characteristic findings of cancer and manifest as changes in metabolites. Metabolomics, a technology with the potential to identify biomarkers and therapeutic targets by comprehensively screening metabolites in patients with various cancers, is expected to be useful for RCC treatment. To date, since clinically useful biomarkers for early diagnosis and pathogenesis of RCC are lacking, RCC is commonly diagnosed using imaging test results. Therefore, measurement of metabolites may be useful in the diagnosis and prognosis of RCC. Several studies have demonstrated the usefulness of metabolite‐based diagnostic and prognostic models. For von Hippel–Lindau syndrome, a rare autosomal dominant disease in which RCC is a major cause of death, a high‐performance diagnostic model was developed using the top 10 of 23 specific metabolites common to both plasma and tissue, focusing on key pathways, such as arginine and proline metabolism [19]. In our previous studies, we performed metabolomic profiling of 138 pairs of ccRCC and matched normal tissues, finding that metabolites and their pathways could distinguish between the early diagnosis and malignant state of RCC using a non‐targeted metabolomics protocol [20]. We also found that ccRCC is characterized by extensive alterations in central carbon metabolism, one‐carbon metabolism, and antioxidant responses, including the glutathione and methionine metabolic pathways [20]. Additionally, RCC progression and metastasis are associated with increased levels of metabolites in the glutathione and cysteine/methionine metabolic pathways [21]. In our previous studies, we constructed highly sensitive and specific diagnostic and clinical stage prediction models for RCC using four or five metabolites [8]. We also constructed a postoperative recurrence prediction model using five urinary metabolites (lactate, glycine, 2‐hydroxyglutaric acid, succinic acid, and kynurenic acid) with a high AUC, sensitivity, and specificity (0.894%, 88.9%, and 88.0%, respectively) [22]. To date, there have been few models for predicting TKI resistance, including CABO, using metabolite measurements in the treatment of aRCC. In this study, we constructed a prediction model for the poor efficacy of CABO using qualitative measurements of five metabolites, including Gln, methylhistidine, kynurenic acid, xanthine, 7‐methylxanthine, and L‐2‐amino‐3‐oxobutanoic acid, which were identified from four metabolic pathways that were determined to be important by the metabolite set enrichment analysis. Although this model lacked excellent sensitivity and specificity, its AUC was approximately 0.723, suggesting that it is capable of predicting CABO‐poor responders to a certain degree. Furthermore, when ROC curves were constructed by combining two, three, four, and five of the identified metabolites to achieve the highest AUC, the combination of five showed the best AUC (Figure 7). Thus, the construction of various predictive models based on metabolite analysis could help improve the prognosis of patients with advanced renal cancer. Belzutifan, a recently approved anticancer drug, demonstrated significant benefits over everolimus in PFS and objective response rates in patients with advanced ccRCC who received immune checkpoint and antiangiogenic therapies including CABO [23], making it a potential alternative treatment option for CABO‐poor responders. In this study, we review the four identified metabolic pathways that are associated with the poor efficacy of CABO. Increased methylhistidine and histidine metabolism are important for understanding the pathophysiology of cancer‐related cachexia. Methylhistidine is produced during the degradation of muscle proteins, such as actin and myosin, resulting in elevated methylhistidine concentrations in the blood and urine [24]. Based on our analysis results showing that methylhistidine metabolism could be most involved in CABO resistance, this suggests that CABO renal cancer cells may proliferate uncontrollably under CABO. Increased Trp metabolism allows cancer cells to consume large amounts of Trp, leading to nutrient deficiency and impaired function of the surrounding T cells, facilitating cancer cell evasion from the immune system [25]. Furthermore, increased levels of xanthine, an intermediate in the purine metabolic pathway, indicate increased caffeine and purine metabolism, which ultimately converts xanthine to uric acid. Xanthine oxidoreductase (XOR) catalyzes this reaction. Cancer cells control reactive oxygen species levels by regulating XOR activity, thereby enhancing survival [26]. Increased levels of L‐2‐amino‐3‐oxobutanoic acid, which indicates increased beta‐alanine metabolism, glycine, and serine metabolism, are metabolic intermediates of L‐aspartic acid β‐semialdehyde or oxaloacetate and are notably involved in the metabolic activity of cancer cells in pyruvate metabolism and the TCA cycle. Resistance to CABO, a key drug, is a major challenge in the treatment of aRCC. Resistance to targeted drugs could be classified as either intrinsic (primary) or acquired (secondary). Intrinsic resistance occurs when a drug rapidly becomes ineffective. This type of resistance may result from the presence of resistant tumor clones before the initiation of treatment, due to genetic resistance or evolutionary clonal selection. Acquired resistance is characterized by tumor growth after initial tumor regression while the patient is still receiving treatment [27]. The results of this study showed that the metabolic pathways that were altered in the CABO‐resistant cell lines were also changed in the CABO‐poor responders, but it was not possible to assess whether similar metabolic pathways were involved in acquired and intrinsic resistance. Resistance to molecular‐targeted therapies is regulated by complex and multifactorial mechanisms beyond single metabolic alterations. Recent studies have demonstrated that tumor microenvironment remodeling, hypoxia adaptation, intercellular communication, and exosome‐mediated signaling contribute to therapeutic resistance in various solid tumors, including RCC and hepatocellular carcinoma [28]. In particular, metabolic reprogramming interacts dynamically with angiogenic signaling, immune regulation, and stromal cell communication, thereby forming complex resistance networks. Therefore, the metabolomic alterations identified in this study may represent one aspect of a broader adaptive response associated with CABO resistance. Efforts to overcome treatment resistance are progressing through ongoing research in various fields, including the identification of biomarkers and the exploration of new pathways, which relate to the development of targeted therapies that could effectively address resistance in ccRCC [7]. The high efficacy of CABO, which demonstrates its antitumor activity even in tumors with poor angiogenic potential [29], highlights the importance of the metabolic pathways involved in enhanced VEGFR signaling in CABO resistance, suggesting the restoration of alternative angiogenic pathways in regulating resistance [30]. Methylhistidine and xanthine metabolic pathways are also involved in angiogenesis [31]. Our previous study demonstrated that controlling Gln metabolism inhibits the restoration of VEGFR signaling [10] and is useful for overcoming CABO resistance [9]. Therefore, controlling methylhistidine, xanthine, and Gln and their metabolic pathways may be a target for overcoming CABO resistance through the regulation of VEGFR signaling. This study has several limitations. First, the subjects in this study had diverse backgrounds, including comorbidities, which may introduce selection bias into the group assignments. Second, because this study was designed as an exploratory metabolomics analysis for hypothesis generation, priority was given to identifying candidate metabolites and metabolic pathways. Therefore, overly strict corrections for multiple comparisons may increase false negative results and potentially overlook biologically significant candidates. Furthermore, because we focused on exploratory metabolite analysis related to the poor efficacy of CABO, we deliberately did not include factors that may affect oncological outcomes, such as a history of radiation therapy or previous treatment. Many patients who experienced poor efficacy of CABO also had a poor IMDC risk. Third, although high levels of L‐2‐amino‐3‐oxobutanoic acid significantly predicted poor response to CABO in the survival analysis, this study was a retrospective analysis of a small number of patients; therefore, the predictive model for poor response to CABO did not have high sensitivity or specificity (Figure 8). Finally, due to the lack of comparison with metabolites in other TKIs' resistance and a lack of a validation cohort, it remains unclear whether the metabolic pathways identified in this study could predict CABO‐poor responders in clinical practice. Because the AUC confidence interval is wide and there is no independent validation cohort, caution is required when evaluating the reliability of the predictive model. Due to these limitations, the present findings should be interpreted as results for hypothesis generation rather than conclusive predictive evidence. Further validation in larger, independent cohorts will be required before these candidate metabolite biomarkers can be translated into clinical practice. Nevertheless, our findings provide preliminary evidence that the identified serum metabolite biomarkers may contribute to optimizing therapeutic selection for patients with advanced renal cell carcinoma. Conclusion We identified the important metabolites (methylhistidine, kynurenic acid, xanthine, 7‐methylxanthine, and L‐2‐amino‐3‐oxobutanoic acid) and their metabolic pathways that could predict the poor efficacy of CABO in aRCC patients. Although further validation studies are needed for clinical application, metabolomic analysis is expected to improve the prognosis of patients with aRCC by assessing drug efficacy and selecting therapeutic drugs. Author Contributions Takanari Sakai: methodology, data curation, writing – original draft, investigation. Yoshihide Kawasaki: investigation, validation, writing – original draft, writing – review and editing, project administration, data curation, supervision, resources, formal analysis. Izumi Sakamoto: investigation, formal analysis, methodology. Eiji Hishinuma: methodology, investigation, formal analysis. Naomi Matsukawa: investigation, formal analysis. Daigo Chiba: formal analysis, data curation. Tomonori Sato: resources, data curation. Kento Morozumi: resources, data curation. Masamitsu Maekawa: investigation, formal analysis, methodology. Nariyasu Mano: supervision. Akihiro Ito: supervision, data curation. Funding This work was supported by Japan Society for the Promotion of Science Grants‐in‐Aid for Scientific Research (KAKENHI) 20K07582, 24K12478. Ethics Statement This study was approved by the Institutional Ethics Review Board of Tohoku University School of Medicine. The registry and the Registration No. of the study are 2019‐1‐749. Consent This study was conducted in accordance with the principles of the Declaration of Helsinki, and all informed consent was obtained from the subjects. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: Mass Spectrometer conditions for serum non‐targeted metabolomics. First, the ion with the highest intensity in the entire scan was selected as the precursor ion, and the product ion spectrum obtained by decomposing this ion was measured. Then, based on the intensity of the chosen ion, the next precursor ion was selected. This process of decomposition and measurement was repeated in the same way, efficiently analyzing various compounds in the sample. Table S2: Liquid Chromatograph conditions for serum non‐targeted metabolomics.