DPP9 Stabilizes NRF2 to Suppress Ferroptosis and Induce Sorafenib Resistance in Clear Cell Renal Cell Carcinoma

The cell redox balance is the equilibrium between the generation and detoxification of reactive oxygen species (ROS). Cells have several antioxidant defense systems that maintain ROS levels, including the transcription factor nuclear factor erythroid 2-like 2 (NRF2, encoded by NFE2L1). NRF2 orchestrates an elaborate transcriptional program in response to environmental cues arising from oxidants and electrophilic agents, allowing for adaptation and survival under stress conditions (1). Therefore, NRF2 is pivotal for cellular protection against endogenous and exogenous insults. Under unstressed conditions, the NRF2 protein is polyubiquitinated by the Kelch-like ECH-associated protein 1 (KEAP1)–Cullin 3 ubiquitin E3 ligase complex, which maintains the NRF2 protein at a low level by constitutive proteasomal degradation. However, KEAP1-mediated NRF2 degradation is blocked by oxidative and electrophilic stresses. The stabilized NRF2 translocates to the nucleus to drive the transcription of cognate target genes, the protein products of which are involved in cellular antioxidant, detoxification, and metabolic pathways (2, 3). The Cancer Genome Atlas (TCGA) cataloged genetic changes in the KEAP1-NRF2 axis in approximately 6.3% of all patients with cancer (4). These mutations phenotypically converge to constitutively activate NRF2. Otherwise, NRF2 hyperactivation occurs through the noncanonical pathway, where a subset of adaptor proteins (SQSTM1/p62, WTX, ATDC, PALB2, etc.) compete with NRF2 for binding to KEAP1, allowing NRF2 to evade KEAP1-mediated degradation (5).

Cytosolic dipeptidyl peptidase 9 (DPP9) and its close paralog DPP8 are two members of the dipeptidyl peptidase IV enzyme family, characterized by the ability to cleave a dipeptide from the N-terminus of their substrates (preferentially post-proline; refs. 6, 7). Although several strategic screenings for protease substrate profiling have been performed, very few physiologic substrates have been identified. The most well-known molecular function of DPP9 is endogenous inhibition of NLRP1 and CARD8 inflammasomes by directly binding to NLRP1 and CARD8, respectively (8). Inhibiting DPP8/9 with Val-boroPro induces procaspase-1–dependent secretion of cytokines and pyroptotic cell death in monocytes and macrophages (9, 10). In addition to the well-established function of DPP9 in the immune system, DPP9 also plays diverse roles in cell growth, migration, and adhesion. Loss of DPP9 enzymatic activity is lethal in neonatal mice due to suckling defects (11). Moreover, aberrant overexpression of DPP9 has been observed in various types of solid tumors (including hepatocellular, ovarian, colorectal, breast carcinoma, and non–small cell lung cancer) and is usually correlated with unfavorable clinical outcomes and chemoresistance (12–16). These findings suggest that DPP9 represents a promising target for cancer therapy. However, the pathophysiologic functions of DPP9 in solid tumors are still poorly understood.

In this study, we reveal that DPP9 expression is oncogenically upregulated in clear cell renal cell carcinoma (ccRCC). We show that DPP9 promotes NRF2 pathway activation, suppresses ferroptosis, and causes sorafenib resistance in ccRCC cells by directly blocking KEAP1-mediated NRF2 degradation. Dysregulation of the KEAP1-NRF2-SLC7A11 axis partially contributes to DPP9 overexpression–induced ferroptosis suppression and sorafenib resistance.

Caki-1 and Caki-2 were cultivated in McCoy's 5A media (Thermo Fisher Scientific). A498, ACHN, and 293T cells were cultured in DMEM (Thermo Fisher Scientific). SW839, OSRC2, 769-P, 786-O, and CCF-RC1 cells were maintained in RPMI1640 media (Thermo Fisher Scientific). CCF-RC1 cells were acquired (September 24, 2018) from Dr. Ying Hu (Harbin Institute of Technology, Harbin, P.R. China; ref. 17). All the other cell lines were purchased from the Shanghai Cell Bank of the Chinese Academy of Sciences (Shanghai, P.R. China). Media were supplemented with 10% (v/v) FBS (Thermo Fisher Scientific) and 1% antibiotic (100 mg/mL streptomycin and 100 U/mL penicillin). All cells were grown and maintained at 37°C in a humidified incubator with 5% CO₂. The cell lines were routinely tested to exclude contamination by Mycoplasma and not cultured for longer than 15 passages. The genetic identity of the cell lines was confirmed by short tandem repeat. Lipofectamine 2000 (Thermo Fisher Scientific) was used for transient transfection following the manufacturer's instructions. The antibodies used in this study are listed in Supplementary Table S1.

The DPP9-containing protein complexes were purified from 293T cells to detect DPP9-interacting proteins. Briefly, 293T cells were transiently transfected with the pCMV-FLAG-DPP9 constructs. A total of 48 hours after transfection, the 293T cells were washed twice with ice-cold PBS and lysed in ice-cold BC100 buffer (20 mmol/L Tris-HCl, 0.2 mmol/L ethylene diamine tetraacetic acid (EDTA), 100 mmol/L NaCl, 20% glycerol, pH 7.9 0.2% Triton X-100) containing fresh protease inhibitors for 2 hours. The whole-cell lysates (WCL) were centrifuged at 12,000 rpm for 10 minutes, and the supernatants were carefully aspirated without disturbing the pellet and placed in new tubes. The DPP9-containing protein complexes were immunoprecipitated using anti-FLAG antibody-conjugated M2 agarose (Sigma). The bound protein complexes were eluted using FLAG peptides. The eluates were resolved by SDS-PAGE on a 10% gradient gel (Bio-Rad). After staining with Coomassie Blue (CB), the target gel bands were excised and subjected to further analysis.

The gel bands were destained three times (500 μL 50 mmol/L NH₄HCO₃/40% methanol) for 1 hour each at room temperature until the color of the gel bands completely faded. Then the gel bands were covered with 75% acetonitrile (ACN), and the gel bands must turn white. Subsequently, the gel bands were washed by HPLC water two times for 1 hour each, and rehydrated gel bands with 50 mmol/L NH₄HCO₃ for 5 minutes. The gel bands were then crushed and added digestion buffer (30 ng trypsin freshly made up in 30 μL 50 mmol/L NH₄HCO₃) to incubate at 37°C for 16 hours. The digestion buffer from the above gel pieces was collected into a new microcentrifuge. Added 0.1% formic acid (FA) to the gel pieces, followed by vortexed for 5 minutes. Then 75% ACN was then added to the gel pieces, followed by vortexed for 5 minutes, and collecting the supernatant into the microcentrifuge. Then peptides were dried using a 60°C vacuum drier.

Peptides were analyzed on a Q Exactive HF-X Hybrid Quadrupole-Orbitrap Mass Spectrometer (Thermo Fisher Scientific) coupled with a high-performance liquid chromatography system (EASY nLC 1200, Thermo). Dried peptide samples dissolved in Solvent A (0.1% FA) were loaded onto a 2-cm self-packed trap column (100 μm inner diameter, 3 μm ReproSil-Pur C18-AQ beads, Dr. Maisch GmbH) using Solvent A and separated on a 150 μm inner-diameter column with a length of 15 cm (1.9 μm ReproSil-Pur C18-AQ beads, Dr Maisch GmbH) over a 75-minute gradient (Solvent A: 0.1% FA in water; Solvent B: 0.1% FA in 80% ACN) at a constant flow rate of 600 nL/minute (0–75 minutes, 0 minute, 4% B; 0–10 minutes, 4%–15% B; 10–60 minutes, 15%–30% B; 60–69 minutes, 30%–50% B; 69–70 minutes, 50%–100% B; 70–75 minutes, 100% B). Mass spectrometry (MS) was performed in data-dependent acquisition mode. For the MS1 spectra full scan, ions with m/z ranging from 300 to 1,400 were acquired by an Orbitrap mass analyzer at a high resolution of 120,000. The automatic gain control (AGC) target value was set to 3E+06. The maximal ion injection time was 80 ms. Top60 precursors were selected for MS2 experiment. The isolation window of selected precursor was 1.6 m/z. Precursor ions were fragmented with higher energy collision dissociation with a normalized collision energy of 27%. Fragment ions were analyzed by an Orbitrap mass analyzer with the resolution of 7,500, AGC target at 5E+04, as well as the maximum ion injection time of MS2 was 20 ms as described previously (18).

The MS raw were searched against the humanRefSeq protein database (27,414 proteins, version 2013-7-4) using Proteome Discoverer (version 2.3.0.523) with a Mascot (version 2.3.01) engine with a percolator (19). Carbamidomethyl cysteine was used as a fixed modification. Oxidation (M) and acetylation (protein N-term) were set as variable modifications. The tolerance for spectral searches a mass tolerance of 20 ppm for the precursor. The maximum number of missing cleavage sites was set at 2. The cutoff of FDR was 1% for both proteins and peptides. The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (20) partner repository with the dataset identifier PXD042187. The identified proteins in the purified DPP9 complex are listed in the Supplementary Table S2.

To generate the KEAP1 and DPP9 knockout (KO) cell lines, single-guided RNAs (sgRNA) were cloned into the pX459 vector (Addgene). Specific sgRNA sequences (listed in Supplementary Table S3) targeting KEAP1 or DPP9 were chosen using the optimized CRISPR design (http://crispr.mit.edu/). Plasmids expressing human Cas9 and KEAP1 sgRNA (KEAP1/pX459), and human Cas9 and DPP9 sgRNA (DPP9/pX459), were prepared by ligating the oligonucleotides into the pX459 BbsI site. To establish the KEAP1 and DPP9 KO cell lines, cells were seeded in a 10 cm dish and transfected with 10 μg of KEAP1/pX459 and DPP9/pX459 plasmid, respectively, using Lipofectamine 2000. A total of 24 hours after transfection, the cells were treated with 1 μg/mL puromycin for 3 days. Surviving cells were seeded in a 96-well plate at a limiting dilution to isolate the monoclonal cell lines. Single-cell clones stably expressing sgRNA were propagated and validated by Western blot (WB) analysis.

The viability of the ccRCC cells was detected using the Cell Counting Kit-8 (CCK-8; Dojindo) according to the manufacturer's instructions. Briefly, the cells were seeded in a 96-well plate at 2,000 cells/well. After the treatments, 10 μL of CCK-8 solution was added to the medium, followed by 2 hours incubation at 37°C. Absorbance was measured at 450 nm using a microplate reader and normalized to the control. Cell viability was calculated using the results from three independent experiments.

Cells were plated in 6-well plates at 3 × 10⁵ cells per well 24 hours before analysis, and ROS levels were detected using the ROS Assay Kit (D6883; Sigma) according to the manufacturer's instructions. Briefly, following two washes with PBS, the cells were stained with 10 μmol/L 2′,7′dichlorodihydrofluorescein diacetate (DCFH-DA) and incubated at 37°C for 20 minutes. DCFH-DA is oxidized and converted into fluorescent 2′,7′-dichlorofluorescein by intracellular ROS. After washing three times with serum-free media to fully remove DCFH-DA that did not enter the cells, the cells were trypsinized, centrifuged, and resuspended in PBS at a density of approximately 1 × 10⁶ cells/mL. The fluorescence signals were detected using a flow cytometer (Beckman).

To detect lipid peroxides, cells cultured in 6-cm dishes were exposed to Liperfluo (Dojindo) at a final concentration of 10 μmol/L for 30 minutes at 37°C, according to the manufacturer's instructions. Following two washes with PBS, the cells were harvested and analyzed using a flow cytometer.

The relative malondialdehyde (MDA) levels in cell lysates were measured using a Lipid Peroxidation Assay Kit (Colorimetric/Fluorometric; ab118970; Abcam), according to the manufacturer's instructions. Briefly, thiobarbituric acid (TBA) was added to the supernatant, which reacted with MDA in the cell lysates to generate an MDA-TBA adduct for 1 hour at 95°C. The absorbance of the MDA-TBA adduct was measured at 532 nm using a microplate reader and the result was normalized to that of the control. Each assay was performed in triplicate.

Total RNAs were extracted using TRIzol reagent (Tiangen) according to the manufacturer's instructions. The amount and quality of total RNAs were determined using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific). Total RNAs were reverse-transcribed into cDNA using a Superscript RT kit (TOYOBO) with random primers, and qPCR amplification was carried out using the SYBR Green PCR Master Mix Kit (TOYOBO). Gene expression changes were normalized to the level of endogenous GAPDH and calculated using the 2^−ΔΔCT method. The primer sequences used for qRT-PCR are listed in Supplementary Table S3.

A total of 500 ng of isolated RNA per sample was used as input material. Sequencing libraries were generated using a Ribo-off rRNA Depletion Kit (Vazyme) and VAHTS Universal V6 RNA sequencing (RNA-seq) Library Prep Kit (Vazyme) for the Illumina platform following the manufacturer's instructions. Index codes were added to attribute the sequences to each sample. The libraries were sequenced on the Illumina platform, and 150 bp paired-end reads were generated.

As described previously (21), RNA-seq raw data quality was assessed using FastQC (v0.11.9), and the adaptors were trimmed with Trim_Galore (version 0.6.6) before any data filtering criteria were applied. Reads were mapped onto the human reference genome (GRCh38.p13 assembly) using STAR software (v2.7.7a). The mapped reads were assembled into transcripts or genes using StringTie software (v2.1.4) and the genome annotation file (hg38_ucsc.annotated.gtf). The relative abundance of the transcripts was measured using normalized fragments per kilobase of transcript per million mapped reads metrics.

The glutathione (GSH) levels were evaluated using the Total Glutathione Assay Kit (S0052; Beyotime), according to the manufacturer's instructions. Cultured 786-O and CCF-RC1 cells were trypsinized, deproteinized solution S was added, and the cells were vortexed. The cells were subjected to two quick freeze-thaw cycles using liquid nitrogen and a 37°C water bath. After 5 minutes in an ice bath, the cell lysates were centrifuged at 10,000 × g for 10 minutes at 4°C. The supernatants were loaded into a 96-well plate to determine total GSH. Next, 150 μL of the total GSH detection working solution was added to each well and incubated for 5 minutes at room temperature. A 50 μL aliquot of NADPH (0.5 mg/mL) solution was rapidly added to the samples and the plate was mixed briefly on a shaker. The plates were incubated for 1 hour at 37°C. Absorbance was measured at 412 nm using a microplate reader. Each assay was performed in triplicate.

GPX4 enzymatic activity was detected using the Total Glutathione Peroxidase Assay Kit with NADPH (Beyotime), following the manufacturer's instructions. As described previously (22), the cells were harvested and washed with cold PBS. Then, the cells were lysed with 200 μL assay buffer and centrifuged at 10,000 × g for 15 minutes at 4°C. The supernatants were transferred to new tubes. Next, GSH peroxidase detection buffer, GPX detection working buffer, and the samples were loaded into a 96-well plate. The samples were mixed thoroughly and incubated at room temperature for 15 minutes. A 10 μL aliquot of peroxide solution (30 mmol/L) was added to the wells, the plate was incubated at 25°C for 5 minutes, and the resulting color was measured at 340 nm using a microplate reader. Each assay was performed in triplicate.

Cells on round cover glasses were fixed for cell staining in 4% paraformaldehyde at room temperature for 15 minutes. The fixed cells were washed three times with PBS. The cells were permeabilized with 0.5% Triton X-100 in PBS for 10 minutes, and then in a blocking solution for 30 minutes at room temperature. The samples were stained with primary antibodies at 4°C overnight in antibody reaction buffer (PBS plus 1% BSA and 0.3% Triton X-100, pH 7.4). After washing three times with PBS (10 minutes each), fluorescence-labeled secondary antibodies were incubated for 30 minutes. The cells were washed three times (10 minutes each) after incubation with the secondary antibodies and then counterstained with DAPI (1 μg/mL) for 10 minutes at room temperature. The cells were subjected to confocal microscopy (LSM880; Carl Zeiss).

Four to 6 weeks old BALB/c nude mice (weight 20–25 g) were obtained from Shanghai SLAC Laboratory Animal Co., Ltd for in vivo xenografts. Cultured 786-O or CCF-RC1 cells (5 × 10⁶) were subcutaneously heterotransplanted into the right flank of all mice. The mice were maintained under specified conditions. When tumor volume increased to 50 mm³, the sorafenib treatment groups were intravenously injected with sorafenib (10 mg/kg) every 2 days. Then, the tumor volume was calculated using the following formula: tumor volume = (long × wide²) × 1/2, every 7 days. At the end of the experiment, the tumors were imaged, measured, and weighed after euthanizing the mice. All procedures were performed with the approval of the Animal Care Committee of Fudan University (Shanghai, P.R. China).

ccRCC patient-derived organoids (PDO) were cultured as reported previously (23). Briefly, fresh tumor tissue samples derived from patients with ccRCC were washed with ice-cold PBS supplemented with 2% penicillin/streptomycin (TBD science) three times and were cut into pieces with diameters around 1 mm. The tumor tissue pieces were washed three times again before digestion. After the last wash, the tumor tissue pieces were suspended with digestion buffer about 20 times the volume of the tissue pieces in a 50 mL tube [two-third of 250 U/mL collagenase IV (Solarbio), one-third of 0.05% Trypsin-EDTA (TBD science), 10% penicillin/streptomycin, and 10 μmol/L Y-27632 (Selleck)], all diluted with DMEM. The tumor tissue pieces were incubated with the digestion buffer in a water bath at 37°C until the mixture becomes cloudy, and then the same volume of DMEM containing 10% FBS was added and mixed gently to abort the digestion process. The suspension was added to a 100 μm cell strainer and centrifuged at 200 × g for 5 minutes to collect the cell clusters. The cell clusters were then washed with 1% BSA/PBS three times to wash away residual digestive enzymes. After the last wash, the cell cluster pellet was suspended by Matrigel (356231, Corning) and a 50 μL drop was added to the center of a well in a 24-well plate. The plate was put into the cell incubator to polymerize the gel at 37°C for 15 minutes. After the polymerization, 550 μL complete medium prepared with Kidney Cancer Organoid Kit (bioGenous) supplemented with 10 μmol/L Y-27632 was added to the well. The complete medium was changed approximately every 3–5 days.

For ccRCC PDO stable line establishment, organoids were blown into cell clusters using pipette. After centrifugation at 200 × g, 4°C for 5 minutes, the cell clusters were suspended by complete media (bioGenous) with control or DPP9 KO lentivirus particles (also expressing eGFP, Genechem), respectively. The mixture was incubated in room temperature for 30 minutes and then on ice for 5 minutes. After the incubation, Matrigel of equal volume was added into the mixture, and 50 μL drops were planted in a 24-well plate. After gel polymerization, a complete medium supplemented with 10 μmol/L Y-27632 was added to the well. Organoids expressing eGFP were selected for expansion, and the efficiency of DPP9 KO was verified by WB analysis.

To evaluate the response of organoids to sorafenib treatment, parental and DPP9 KO organoids were blown into cell clusters by pipetting. After centrifugation collection, the cell clusters were suspended with Matrigel. 5 μL droplets were planted to the wells of a 96-well plate. After 3 days, the images of organoids were captured using Nexcope NIB610FL microscope system (Nexcope) and a complete medium with sorafenib (3 μmol/L) was added for medium change (sorafenib treatment day 1). After 3 days, the complete medium with sorafenib was refreshed. In day 7, the images of organoids treated with sorafenib were captured again and the viability of the organoids was evaluated using CellTiter-Glo three-dimensional (3D) Cell Viability Assay kit (Promega) according to the manufacturer's guidance.

A total of 207 patients with localized ccRCC, who underwent radical nephrectomy or partial nephrectomy between March 2005 and December 2009 at Fudan University Shanghai Cancer Center (FUSCC), were included in this study. None of the patients received preoperative or postoperative adjuvant treatment. The patients underwent regular postoperative reviews and had long-term follow-up data. Seven pairs of fresh ccRCC tissues were obtained from October 2020 to December 2020 at FUSCC. The tumor tissues and adjacent noncancerous tissues were collected and subjected to WB analysis. 26 metastatic ccRCC specimens from patients who underwent cytoreductive surgery and sorafenib treatment between April 2006 and July 2009 were also included in the study. Patients with progressive disease during 1 year of the sorafenib treatment were defined as sorafenib-resistant. This study also included 110 patients with advanced ccRCC, who were treated with tyrosine kinase inhibitors (TKI) at the Department of Urology of FUSCC from Jan 2008 to Mar 2020. This study was in accordance with the recommendations of the Research Ethics Committee of FUSCC according to the provisions of the Declaration of Helsinki (as revised in Fortaleza, Brazil, October 2013). The protocol was approved by the Research Ethics Committee of FUSCC. Written informed consent was obtained from all the patients recruited in this study.

Tissue microarray (TMA) slides consisting of 207 localized and 26 metastatic ccRCC specimens were obtained from patients with FUSCC. Briefly, the TMA slides were deparaffinized, rehydrated, incubated with 3% H₂O₂, and blocked with goat serum. They were incubated with primary antibodies against DPP9 (ab62025; Abcam) or NRF2 (16396-1-AP; Proteintech) overnight at 4°C. After three washes in PBS, the TMA slides were incubated with peroxidase-conjugated secondary antibody for 1 hour at room temperature and stained with diaminobenzidine to detect the chromogens. All IHC staining intensities were assessed by two independent pathologists blinded to the clinical data. A 4-point scale (0, negative; 1, weak; 2, moderate; 3, intense) was used, and the quantity of immunostaining was classified (0, none; 1, 1%–30%; 2, 31%–60%; 3, >60%). The total score was the product of the two scores. Patients were divided into high and low DPP9 expression groups according to the median IHC score. To confirm the specificity of the anti-DPP9 antibody, we conducted genetic controls for the IHC analysis using an anti-DPP9 antibody in both parental and DPP9 KO 786-O cells.

Statistical analyses were mainly performed by using GraphPad Prism software. Data are presented as mean ± SD for experiments performed independently at least three times. Differences between the two groups were analyzed by two-way analysis of variance. The test for trend was performed with a polynominal contrast procedure by using Statistical Package for the Social Sciences (SPSS) 20.0 software. Survival analysis was performed using the Kaplan–Meier method and log-rank test. A P value < 0.05 was considered significant. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Pan-cancer gene expression analysis based on tumor and normal samples were derived from TCGA (transcriptome datasets, http://gepia2.cancer-pku.cn/) and Clinical Proteomic tumor Analysis Consortium (CPTAC, proteomics data, https://ualcan.path.uab.edu/analysis-prot.html). Public databases (TCGA and CPTAC) were used to analyze the correlations of DPP9 expression with clinical risk factors. The available RNA-seq data of 534 patients with ccRCC from TCGA were obtained from the UCSC Xena website (https://xenabrowser.net/). We acquired quantitative proteomics data for 110 patients with ccRCC from the CPTAC database (https://proteomics.cancer.gov/programs/cptac). We also obtained RNA-seq data of 91 patients with ccRCC from the International Cancer Genome Consortium (ICGC) database (https://icgc.org/). The proteomic data of the 213 Chinese patients with ccRCC were obtained from the Supplementary Data S3 in Qu and colleagues’ study (24). An additional publicly available RNA-seq dataset (GSE87121) provided 5 sorafenib-resistant and 5 sorafenib-sensitive patients with ccRCC was derived from Gene Expression Omnibus (GEO). The raw RNA-seq data have been deposited in NCBI's GEO and are accessible through GEO Series accession number GSE242367 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE242367).

Previous studies indicated that DPP9 promotes tumorigenesis in several types of cancer (13–16, 25–27). However, it is unknown whether DPP9 is aberrantly expressed in ccRCC, the most frequent histologic subtype of RCC. To address this issue, we investigated the expression pattern of DPP9 in ccRCC by analyzing the publicly available TCGA ccRCC dataset. As shown in Fig. 1A, DPP9 mRNA expression was markedly upregulated in ccRCC tissues compared with normal kidney tissues. In contrast, DPP9 expression was moderately altered in kidney chromophobe and kidney renal papillary cell carcinoma (KIRP), which are both relatively uncommon histologic subtypes of RCC. The upregulation of DPP9 in ccRCC was also validated in the ICGC ccRCC cohort (Supplementary Fig. S1A). The CPTAC dataset showed that DPP9 was most markedly upregulated at the protein levels among the tumor types examined (Fig. 1B). A similar trend was observed in a large-scale proteomic analysis of 213 ccRCC specimens from the FUSCC cohort (Supplementary Fig. S1B; ref. 24).

Analysis of TCGA ccRCC dataset revealed that DPP9 mRNA expression was positively correlated with several clinical parameters, including T stage (Fig. 1C), tumor–node–metastasis (TNM) stage (Fig. 1D), and pathologic grade (Fig. 1E). Positive correlations between DPP9 protein expression and advanced TNM stage/pathologic grade were independently validated by analyzing the CPTAC dataset (Fig. 1F and G). We confirmed that DPP9 protein expression was upregulated in five of seven fresh ccRCC tissues compared with adjacent noncancerous tissues as assessed by WB analysis (Fig. 1H and I). Following validation of the antibody specificity for IHC analysis in parental and DPP9 KO 786-O cells (Supplementary Fig. S1C), we further evaluated DPP9 protein expression in ccRCC by IHC analysis of a TMA consisting of 207 localized ccRCC and adjacent noncancerous tissues from the FUSCC cohort. As shown in Fig. 1J, DPP9 staining intensity in tumor tissues was classified as negative, low, or high. A strong upregulation of DPP9 protein was observed in ccRCC tissues compared with adjacent noncancerous tissues (Fig. 1K). Furthermore, survival analysis indicated that high DPP9 expression was significantly correlated with shorter overall survival (OS) in TCGA cohort (Fig. 1L). In the FUSCC ccRCC cohort, high DPP9 expression was associated with worse OS (Fig. 1M). Taken together, our results indicate that DPP9 expression aberrantly increased in ccRCC and was correlated with a poor prognosis in patients with ccRCC.

To identify potential molecular mediators of DPP9’s oncogenic function, we isolated DPP9 protein complexes from 293T cells overexpressing FLAG-DPP9 and determined the proteins present in the complexes by MS (Fig. 2A). KEAP1 was present in the complexes and ranked with high confidence in the interaction hit list (Supplementary Table S2). Considering the pivotal roles of the KEAP1-NRF2 pathway in cancer, we conducted further analyses to determine the pathophysiologic significance of the DPP9–KEAP1 interaction. We first detected DPP9 protein levels in a panel of RCC cell lines with different subtypes (ccRCC or KIRP) and genetic backgrounds (Supplementary Table S4). Generally, DPP9 is ubiquitously expressed in the RCC cell lines detected (Supplementary Fig. S2A). We then verified the binding of KEAP1 to DPP9 using coimmunoprecipitation assays by overexpressing KEAP1 and DPP9 (Fig. 2B and C). Endogenous interaction between DPP9 and KEAP1 was observed in ccRCC (786-O, CCF-RC1, Caki-1, and OSRC2) and KIRP (Caki-2 and ACHN) cell lines, supporting the notion that the DPP9-KEAP1 interaction is not lineage specific (Fig. 2D and E; Supplementary Fig. S2B–S2F). Next, we determined the protein sequence mediated the mutual interaction between KEAP1 and DPP9. Previous studies have demonstrated that KEAP1-associated proteins (e.g., NRF2, IKKβ, and PGAM5) contain consensus ESGE, ETGE and/or DLG motifs that are responsible for binding to the KELCH domain of KEAP1 (2, 3). DPP9 contained a perfectly matched ETGE and ESGE motif, but no DLG motif was found. Notably, the ESGE motif was highly conserved among different species, from humans to zebrafish. In contrast, the ETGE motif was not conserved and not present in the frog or zebrafish DPP9 proteins (Fig. 2F). Deletion of the ESGE motif, but not the ETGE motif, completely abolished the KEAP1–DPP9 interaction (Fig. 2G; Supplementary Fig. S2G). KEAP1 contained a CUL3-binding BTB-BACK domain, an IVR domain, and a KELCH domain that is responsible for binding to substrates such as NRF2 (Fig. 2H). We demonstrated that the KELCH domain, but not the BTB or IVR domain of KEAP1, mediated its interaction with DPP9 (Fig. 2I). Finally, we demonstrated impaired binding of cancer-derived KEAP1 mutants in the KLECH domain to DPP9, analogous to the KEAP1–NRF2 interaction (Fig. 2J; Supplementary Fig. S2H). Taken together, our results indicate that DPP9 interacts with KEAP1 in cells via the ESGE motif.

After validating the interaction between KEAP1 and DPP9, we explored whether KEAP1 targets DPP9 for ubiquitin-dependent degradation, similar to its authentic substrate, NRF2. However, KEAP1 overexpression did not alter the protein levels of ectopically coexpressed DPP9 (Fig. 3A). To examine the effect of KEAP1 on endogenous DPP9, we generated Tet-on-inducible 786-O cells that conditionally express FLAG-KEAP1. The induction of exogenous KEAP1 led to a time-dependent reduction in the NRF2 protein levels, whereas DPP9 protein level remained unchanged (Fig. 3B). Furthermore, depleting KEAP1 by CRISPR/Cas9-mediated KO in 293T and 786-O cells did not change DPP9 protein levels, but markedly upregulated the NRF2 protein levels (Fig. 3C).

Previous studies revealed that the ETGE motif-harboring KEAP1-interacting proteins, such as SQSTM1, iASPP, and PALB2, sequester KEAP1, leading to increased stability and activation of NRF2 (17, 28, 29). We investigated whether DPP9 affects the NRF2 protein levels through its interaction with KEAP1. Depleting DPP9 with siRNAs in 293T cells downregulated the NRF2 protein levels (Fig. 3D). DPP9 KO in 786-O cells and CCF-RC1 cells, led to a marked decrease in the protein levels of NRF2, HMOX1, and SLC7A11 (Fig. 3E). HMOX1 and SLC7A11 are two representative transcriptional targets of NRF2 (30). Depleting DPP9 with siRNAs also caused downregulation of NRF2, HMOX1, and SLC7A11 protein levels in OSRC2 and 769-P cells (Supplementary Fig. S3A). DPP9 KO shortened the half-life of NRF2 in 786-O cells (Fig. 3F and G), whereas DPP9 overexpression prolonged the half-life of NRF2 (Supplementary Fig. S3B and S3C), indicating that DPP9 positively regulates NRF2 protein stability. Overexpression of DPP9-WT, but not the ΔESGE mutant, increased the protein levels of NRF2, HMOX1, and SLC7A11 in a dose-dependent manner (Fig. 3H). As DPP9 is an intracellular prolyl peptidase, we generated a catalytically dead mutant DPP9-S759A to evaluate whether the NRF2 protein levels were affected by the enzymatic activity of DPP9. However, reintroducing DPP9-S759A mutant into DPP9 KO 786-O cells reversed NRF2 protein to a level similar to that of DPP9-WT (Fig. 3I). Treating 786-O cells with a potent DPP9 enzymatic inhibitor 1G244 did not affect the NRF2 protein levels, indicating that the peptidase activity of DPP9 is dispensable for regulating NRF2 (Supplementary Fig. S3D).

We evaluated the role of DPP9 in protecting NRF2 from KEAP1-mediated degradation. DPP9-WT, but not the ΔESGE mutant, reduced the KEAP1-NRF2 binding affinity and NRF2 ubiquitination (Fig. 3J and K). As the DPP9–KEAP1 interaction enhanced the stability of NRF2, human tumors overexpressing DPP9 may upregulate the NRF2 protein levels. Indeed, the IHC analysis showed that their expression was positively correlated in the FUSCC ccRCC cohort (Fig. 3L and M). Taken together, these results indicate that DPP9 stabilizes NRF2 in ccRCC cells by competitively binding to KEAP1.

To evaluate the outcome of DPP9-mediated NRF2 regulation, we investigated the global transcriptomic changes induced by DPP9 KO in 786-O cells by RNA-seq. Among the 322 reported transcriptional targets of NRF2 (Supplementary Table S5; refs. 31, 32), multiple genes were significantly downregulated in DPP9 KO 786-O cells, including SLC7A11, HMOX1, AIFM2, TXNRD1, and TXN (Fig. 4A). The decrease in HMOX1 and SLC7A11 mRNA levels was confirmed by qRT-PCR (Fig. 4B). Conversely, overexpression of DPP9-WT, but not the ΔESGE mutant, increased the HMOX1 and SLC7A11 mRNA levels (Fig. 4C). Taken together, these results indicate that DPP9 positively regulates the transcriptional outputs of the NRF2 pathway.

NRF2 is a central hub that neutralizes ROS and restores the cellular redox balance. As DPP9 promotes NRF2 protein stability, we explored whether DPP9 affects redox balance in ccRCC cells. As a result, DPP9 KO led to significant ROS accumulation. Reintroducing DPP9-WT into DPP9 KO 786-O cells reversed this effect, whereas reintroducing the ΔESGE mutant failed to do so (Fig. 5A and B). Exposure to H₂O₂ is widely used to trigger oxidative stress in cellular models. Much weaker nuclear localization of NRF2 was observed in DPP9 KO 786-O cells than parental cells after H₂O₂ treatment (Fig. 5C). A stronger binding between KEAP1 and DPP9 was also observed after H₂O₂ treatment (Fig. 5D). Moreover, H₂O₂ treatment triggered much weaker induction of NRF2, HOMX1, and SLC7A11 in DPP9 KO 786-O cells than those in parental cells. Reintroducing DPP9-WT into DPP9 KO 786-O cells reversed these effects, but the ΔESGE mutant did not (Fig. 5E and F). Consistent with their effects on the ROS levels, DPP9 KO 786-O and CCF-RC1 cells were more sensitive to H₂O₂-induced cell death than parental cells. Reintroducing DPP9-WT into DPP9 KO 786-O or CCF-RC1 cells reversed these effects, but the ΔESGE mutant did not (Fig. 5G and H). Taken together, these results indicate that overexpressing DPP9 reduces cellular ROS levels, thereby leading to resistance to oxidative stress–induced cell death in ccRCC cells.

The aforementioned results show that overexpressing DPP9 induced the upregulation of SLC7A11, which imports cystine for GSH biosynthesis, thereby protecting cells from oxidative stress and ferroptosis; this is a regulated form of nonapoptotic cell death driven by the accumulation of lipid-based ROS, particularly lipid hydroperoxides (33, 34). Gene set enrichment analysis (GSEA) of the RNA-seq data indicated that the DPP9 KO-affected genes were enriched in the ferroptosis process (Fig. 6A; Supplementary Table S6). Electron microscopic analysis showed that DPP9 KO 786-O cells had prominent swollen mitochondria and increased mitochondrial membrane density, and lacked mitochondrial cristae, which are typical morphologic changes during ferroptosis (Fig. 6B; refs. 35–37). We used Liperfluo staining to monitor lipid peroxidation during ferroptosis and found that the Liperfluo signals were more prominent in DPP9 KO 786-O cells; this was reversed by reintroducing of DPP9-WT, but not the ΔESGE mutant (Fig. 6C).

The classic ferroptotic inducer erastin elicited more cell death in DPP9 KO 786-O cells than that in parental cells; this effect was largely eliminated by cotreatment with ferrostatin (Ferr-1) or deferoxamine (DFO), which are selective inhibitors of ferroptosis. Reintroducing DPP9-WT into DPP9 KO 786-O cells reversed their vulnerability to erastin, but the ΔESGE mutant did not (Fig. 6D). Furthermore, overexpression of DPP9-WT, but not the ΔESGE mutant, resulted in resistance of 786-O cells to erastin-induced cell death (Fig. 6E). To further delineate whether DPP9 KO promotes ferroptosis by inhibiting SLC7A11 expression, we stably overexpressed SLC7A11 in DPP9 KO 786-O cells. The restoration of SLC7A11 expression in DPP9 KO 786-O cells reversed erastin-induced cell death (Fig. 6F). Moreover, restoring SLC7A11 expression largely reversed DPP9 KO-induced changes in GPX4 enzymatic activity (Fig. 6G; Supplementary Fig. S4A), and the levels of MDA (Fig. 6H; Supplementary Fig. S4B) and cellular GSH (Fig. 6I; Supplementary Fig. S4C) in 786-O or CCF-RC1 cells. Taken together, these results indicate that DPP9 suppresses ferroptosis in a SLC7A11-dependent manner.

Sorafenib is a multikinase inhibitor used to treat advanced ccRCC and hepatocellular carcinoma (HCC; ref. 38). It was first proposed to act as a potent inducer of ferroptosis by inhibiting SLC7A11 activity (39). Therefore, we suspected that aberrant overexpression of DPP9 in ccRCC may cause intrinsic sorafenib resistance by suppressing ferroptosis. We found that overexpressing DPP9 in 786-O and CCF-RC1 cells suppressed sorafenib-induced cell death (Fig. 7A). Sorafenib treatment elicited more cell death in DPP9 KO 786-O than parental cells, and this effect was largely eliminated by cotreatment with Ferr-1 or DFO. Reintroducing DPP9-WT into DPP9 KO 786-O cells reversed their vulnerability to sorafenib, but the ΔESGE mutant did not (Fig. 7B). Furthermore, overexpression of DPP9-WT, but not the ΔESGE mutant, resulted in resistance of 786-O cells to sorafenib (Fig. 7C).

We examined whether DPP9 induces sorafenib resistance in xenograft tumor models. DPP9 KO markedly suppressed 786-O and CCF-RC1 xenograft tumor growth and potentiated sorafenib-induced tumor suppression (Fig. 7D–F; Supplementary Fig. S4D–S4F). The use of cultivated organoids obtained from resected primary human ccRCC tissues suggested that organoids lacking DPP9 were more susceptible to sorafenib-induced cell death than parental cells (Fig. 7G–I). By analyzing DPP9 expression in 26 sorafenib-treated metastatic ccRCC specimens in FUSCC detected by IHC, we found that a higher expression of DPP9 in sorafenib-resistant patients as compared with sorafenib-sensitive patients (P < 0.05, Fig. 7J). Moreover, we also noticed a higher expression of DPP9 in sorafenib-resistant patients as compared with sorafenib-sensitive patients by analyzing a publicly available RNA-seq dataset (GSE87121); however, this trend did not achieve statistical significance, possibly due to the limited samples (5 sorafenib-resistant and 5 sorafenib-sensitive patients with ccRCC, P = 0.0814; Supplementary Fig. S4G). Analysis of the RNA-seq dataset from 110 TKI-treated patients suggested that high DPP9 expression was correlated with inferior progression-free survival and OS in the FUSCC cohort (P < 0.05; Fig. 7K and L). Taken together, these results indicate that DPP9 overexpression contributes to sorafenib resistance in ccRCC.

Hyperactivation of the NRF2 pathway confers cancer cell resistance to anticancer drugs and oxidative stress, and directs cancer cells toward metabolic rewiring of tumors (which promotes tumorigenesis and progression). Therefore, targeting the NRF2 pathway is a promising therapeutic strategy for various cancers, including ccRCC (40). The current study provides evidence that DPP9 binds to and sequesters KEAP1, upregulates the NRF2 protein level and transcriptional outputs, reduces cellular ROS levels, and suppresses ferroptosis, thereby driving ccRCC tumorigeneses and sorafenib resistance.

Mutations in KEAP1, CUL3, and NRF2 have been reported in a relatively small fraction (6.6%) of ccRCC cases (41). However, several studies have demonstrated a generally strong effect of NRF2 hyperactivation in ccRCC, indicating that other mechanisms may regulate the NRF2 pathway. In addition to direct canonical regulation of the NRF2 pathway by KEAP1, noncanonical regulation of the NRF2 pathway by proteins (p62/SQSTM1, PLAB2, WTX, ADTC, and DPP3) can disrupt the KEAP1-NRF2 interaction, as reported in several cancer-related studies (30, 42–45). Most of these proteins harbor the “ETGE” or “ESGE” motif, which mediates competitive binding with KEAP1 from NRF2. For example, ATDC interacts with KEAP1 to drive NRF2-mediated tumorigenesis and chemoresistance in pancreatic cancer (44). DPP3 abrogates KEAP1-mediated NRF2 degradation to support breast cancer survival under oxidative stress (45). ADTC and DPP3 are overexpressed in pancreatic and breast cancers, respectively, and their expression is correlated with poor survival (44, 46). Similarly, DPP9 binds to KEAP1 through an ESGE motif that competes with the ETGE motif, thereby stabilizing NRF2 in ccRCC. In the current study, we established that the ESGE motif-harboring protein DPP9 is a novel noncanonical regulator of the NRF2 pathway (Fig. 8). Importantly, DPP9 is overexpressed in multiple types of cancer. We analyzed the RNA-seq dataset of TCGA cancer cohorts deposited in the GEPIA2 database (http://gepia2.cancer-pku.cn/#general) and found that DPP9, but no other reported noncanonical regulators of the NRF2 pathway, was significantly overexpressed in ccRCC (data not shown), indicating that DPP9 may play a substantial role in NRF2 hyperactivation in ccRCC. Moreover, we demonstrated that DPP9 peptidase activity was not required for regulation of the KEAP1-NRF2 pathway, indicating the diverse roles of DPP9 in cellular processes. We also suggest that blocking the interaction between DPP9 and KEAP1 using small molecules is a possible therapeutic strategy to reduce activation of the NRF2 pathway in ccRCC.

Sorafenib is an orally administered multitarget kinase inhibitor that inhibits certain receptor tyrosine kinases (RTKs), including VEGFR1/2, PDGFR, KIT, FLT3, RET, and CSF1R (47). Studies have revealed that SLC7A11, a key mediator of ferroptosis, is inhibited by sorafenib (48). Sorafenib is the first targeted drug approved by the FDA for treating metastatic ccRCC, which paved the way for the approval of other targeted agents and revolutionized the treatment of ccRCC (49). Unfortunately, the high prevalence of sorafenib resistance has become a major obstacle to sorafenib therapy for patients with ccRCC. Approximately 22% of patients with ccRCC are intrinsically resistant to sorafenib, and most (if not all) of the remaining patients exhibit sorafenib resistance and tumor progression after 6–15 months of therapy (20). Underlying mechanisms of sorafenib resistance in ccRCC have been proposed, including altered expression of efflux and influx transporters, the epithelial–mesenchymal transition, self-renewal of cancer stem cells, changes in the tumor microenvironment and redox reprogramming (47). Fully elucidating the molecular mechanism would help predict and monitor the clinical efficacy of sorafenib in patients with ccRCC. The increase of NRF2 activity in cases of drug resistance is well established. The role of the KEAP1-NRF2 pathway in sorafenib resistance has attracted considerable attention. Gao and colleagues showed that the protein levels of NRF2 and its targets are markedly induced in sorafenib-resistant HCC cells, and the malignant phenotypes of the cells are negated following NRF2 KO. YAP/TAZ and ATF4 drive sorafenib resistance by upregulating SLC7A11 (50). Overall, the results of this study suggest that overexpression of DPP9 in ccRCC likely contributes to intrinsic sorafenib resistance by upregulating the NRF2-SLC7A11 axis. This finding may help guide sorafenib use in ccRCC treatment.

In addition to the tumor cell–intrinsic effects, such as ferroptosis suppression, sorafenib likely works clinically in ccRCC treatment through multiple molecular mechanisms, with its antiangiogenic effects considered pivotal (51). While our study demonstrated that DPP9 overexpression is involved in sorafenib resistance, it should be noted that this effect cannot be solely attributed to ferroptosis. The KEAP1-NRF2 pathway also plays critical roles in tumor angiogenesis and vascular diseases (52). Further work is necessary to fully elucidate how DPP9-mediated NRF2 activation contributes to tumorigenesis and drug resistance in ccRCC.

No disclosures were reported.

K. Chang: Conceptualization, data curation, formal analysis, methodology, writing–original draft, writing–review and editing. Y. Chen: Formal analysis. X. Zhang: Methodology. W. Zhang: Validation, methodology. N. Xu: Methodology. B. Zeng: Methodology. Y. Wang: Validation. T. Feng: Validation. B. Dai: Resources, data curation. F. Xu: Writing–original draft, project administration. D. Ye: Supervision, funding acquisition. C. Wang: Conceptualization, writing–original draft, project administration, writing–review and editing.

This work was supported in part by the National Natural Science Foundation of China (no. 32370726, 91957125, 81972396, and 81672558 to C. Wang; no. 81902614 to K. Chang; no. 82172703 to B. Dai), and the Natural Science Foundation of Shanghai (22ZR1406600 to C. Wang.), and Science and Technology Research Program of Shanghai (no. 9DZ2282100), State Key Development Programs of China (no. 2022YFA1104200 to C. Wang). This study was also partly supported by the Shanghai Sailing Program (19YF1408600 to K. Chang). This study was also partly supported by a grant from the Natural Science Foundation of Science and Technology Commission of Shanghai Municipality (20ZR1412300 to B. Dai).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.