FGFR3 Destabilizes PD-L1 via NEDD4 to Control T-cell–Mediated Bladder Cancer Immune Surveillance

Bladder cancer is one of the most prevalent cancers, with nearly 573,278 new cases and 212,536 deaths annually on a global scale (1). The relative 5-year overall survival rate of advanced-stage, unresectable or metastatic urothelial carcinoma (UC) of the bladder is low (2). Using historical treatment with second-line chemotherapy, the response rate is approximately 10% to 12% and the median overall survival is approximately 7 to 9 months (3–5). FGFR3 mutations, which have a driver role in carcinogenesis, are very common in UC of the bladder and associated with a low immune signature, which is in accordance with the Cancer Genome Atlas (TCGA) luminal-papillary subtype (6–9).

FGFR3 is an actionable target in UC of the bladder, and erdafitinib, a pan-FGFR inhibitor, is the only targeted drug approved by the FDA for bladder cancer treatment. Although the efficacy of erdafitinib in UC of the bladder with prespecified FGFR alterations is up to 40% (the efficacy of immune checkpoint therapy is only 5%), which means there are still more than half of the patients who do not benefit from these new therapies and end up with conventional treatments including chemotherapy and radiotherapy that are less effective with more side effects (10). The UC of the bladder with FGFR alterations that fail to benefit from erdafitinib are usually more aggressive, and the gloomy scenario is that currently no additional work has explored the underlying mechanism.

Currently, targeting immune checkpoints such as PD-1 and PD-L1 has been approved for treating advanced bladder cancer with durable clinical benefit in a subset of patients (3, 11). Because there is possibly no potential cross-resistance between FGFR inhibitors and ICB, the combination strategy has been investigated (12). Recent studies in advanced UC of the bladder showed that the relevance between the mutation status of FGFR3 and the response to PD-1/PD-L1 inhibitors was controversial (13, 14), and the therapeutic effects of immuno-targeted combined therapy were indefinite. Moreover, it is not yet clear which kind of patients may benefit from the combination treatment, and the potential molecular mechanisms are still obscure.

Most studies have demonstrated that the response to anti-PD-1/PD-L1 treatment correlate with PD-L1 expression levels in UC of the bladder, in which high PD-L1 levels usually mean good response to anti-PD-1/PD-L1 treatment (11). Recent studies have revealed a series of mechanisms of PD-L1 regulation by transcriptional and posttranscriptional means in different cancers (15, 16), but the particularity of PD-L1 regulation in bladder cancer has hardly been well elucidated.

Neuronal precursor cell expressed developmentally downregulated 4 (NEDD4) is a HECT family E3 ubiquitin ligase, which functions in substrate recognition and attachment of ubiquitin to substrates (17, 18). The ubiquitin ligase activity of NEDD4 can be promoted by FGFR1 and EGFR activation, which results in the activation of c-Src (19). However, the immunologic regulatory function of NEDD4 remains largely unknown. Here, we demonstrated that NEDD4 could be phosphorylated by FGFR3 activation and functioned as a regulator of PD-L1 ubiquitination to control CD8⁺ T-cell–mediated immune surveillance in UC of the bladder. The mechanism concerning FGFR3 and PD-L1 links two important molecules and uncovers a novel basis for the application of FGFR3 inhibitors and ICB in UC of the bladder with FGFR3 activation.

Six- to eight-week-old female C57BL/6 mice were purchased from the Laboratory Animal Center of Shandong University. Six-week-old female BALB/c nude mice (CAnN.Cg-Foxn1^nu/Crl) and 4- to 6-week-old Cg-Prkdcscid Il2rgtm1Sug/JicCrl (NOG) mice were purchased from Vital River Laboratory Animal Technology. All mouse strains were maintained under specific pathogen-free conditions. All animal experimental procedures in the study were approved by the Ethics Committee of Qilu Hospital of Shandong University (No. KYLL-2020-264).

Six bladder cancer cell lines (SW780, RT4, MB49, T24, 5637, and UM-UC-3), 293T, SV-HUC-1, SK-NM-C, and KMS-11 were used in this study. The T24, 5637, SW780, RT4, UM-UC-3, 293T, and SK-NM-C cell lines were purchased from ATCC, the MB49 cell line was purchased from Millipore, and the KMS-11 was purchased from JCRB. All cells were cultured in the media supplemented with 10% FBS (Corning), 100 U/mL penicillin and streptomycin (Gibco). All cells used in this study were authenticated by short tandem repeat analysis within 2 years. All newly revived cells were tested free of Mycoplasma contamination by the Mycoalert Detection Kit (Beyotime). The cumulative culture length of the cells between thawing and used in this study was less than 10 passages.

The orthotopic tumor model was established by surgical implantation. In short, all the mice were anesthetized with isoflurane (2.5%) and injected with fLuc-MB49 cells or fLuc-SW780 cells (5 × 10⁵) into the bladder using insulin syringes. Several days after implantation, the mice were randomized into different groups according to the tumor burden and administered regularly. When the tumor size reached a threshold, the tumor-bearing mice were sacrificed and the bladder tumors were dissected for further experiments. For in vivo experiments, erdafitinib and infigratinib were formulated in 20% hydroxypropyl-β-cyclodextrin (HP-β-CD) and orally administered at a dose of 12.5 mg/kg twice daily. Anti-PD-1 antibodies used in animal experiments were given by intraperitoneal injection with a dose of 10 mg/kg every 3 days.

Genomic DNA (gDNA) was extracted from patients' peripheral blood samples and paired frozen tissues or formalin-fixed paraffin embedded sections using a DNeasy Blood & Tissue Kit (Qiagen, 69504) according to the manufacturer's instructions. Raw sequencing data were aligned to the reference human genome (UCSC hg19) through Burrows–Wheeler Aligner. Somatic variants were generated in tumor samples by removing the germline alterations from the matched peripheral blood mononuclear cell gDNA to keep the variants unique to the tumor. Variants were annotated using the ANNOVAR software. A CNV Kit was used to determine the copy-number variations (CNV; https://github.com/etal/cnvkit).The tumor mutation burden of each sample was calculated according to a published and widely used method described by Chalmers and colleagues (20).

All patients or healthy donors involved in the study fully understood and provided informed consent, and all the procedures involved were approved by the Ethics Committee of the Qilu Hospital of Shandong University (No. KYLL-2019-168) and Jinan Maternity and Child Care Hospital (No. KYLL-2020–275). The PDX samples were obtained from surgical resection from an elderly male patient with high-grade UC of the bladder that had FGFR3-TACC3 gene fusion detected by next-generation sequencing (NGS). Human peripheral blood samples used for CD8⁺ T-cell isolation were obtained from patients with bladder cancer prior to surgery. Human umbilical cord blood samples used for CD34⁺ hematopoietic stem cell (HSC) extraction were obtained from consenting donors at Jinan Maternity and Child Care Hospital.

For humanized immune system construction, female NOG mice underwent whole-body irradiation (2.0 Gy), and then huCD34⁺ HSCs (1 × 10⁵ HSCs per mouse) were injected into the tail vein of irradiated NOG mice. After 8 weeks, the HSCs differentiated into the appropriate immune cells, which were monitored by peripheral blood flow cytometry analysis. Fresh bladder cancer specimens were cut into small pieces (3 × 3 × 3 mm³) and implanted subcutaneously into the flanks of 4- to 6-week-old female NOG mice. After the tumor mass expanded to over quintuple its volume, the xenograft tumor was resected and directly retransplanted for expansion in later serial generations using the same procedure. An orthotopic bladder cancer patient-derived xenograft (PDX) model was established via implantation of a single cell suspension from subcutaneous PDXs into the bladder of mouse with humanized immune system. After the PDX model was established, mice were administered regularly.

For bioluminescence living imaging, the anesthetized mice were injected intraperitoneally with d-luciferin (150 mg/kg) and imaged 15 minutes later. Luminescence was detected by an IVIS Lumina system (IVIS Spectrum, PerkinElmer), and image analysis was performed using Living Image Software.

For ultrasound imaging, mouse bladders were catheterized using venous catheters coated with liquid paraffin. Then, phosphate-buffered saline (PBS) was delivered into the lumen using a 1 mL sterile syringe to distend the bladder for imaging. The images were acquired using a Vevo LAZR high-resolution ultrasound system (VisualSonics) with a transducer probe in the B-mode setting.

Short hairpin RNA (shRNA) constructs targeting FGFR1, FGFR2, FGFR3, and FGFR4 were purchased from GeneChem Co., Ltd., and small-interfering RNAs (siRNA) targeting WWP1, ITCH, MDM2, CBL, and NEDD4 were purchased from Gene Pharma Co., Ltd. Generally, to generate cell lines that stably expressed the targeted genes using shRNAs, cells were seeded into six-well plates at a density of 2 × 10⁵ cells per well. After incubation for 24 hours, cells were transfected with lentiviral particles. Then the transfected cells were selected using puromycin at indicated concentrations before use. For siRNA transfection, the seeded cells were transfected with siRNAs according to the standard protocol. The cells were used in the experiments after 48 hours of transfection. NEDD4 and PD-L1 were knocked out by the CRISPR/Cas9 gene editing system. After transfection, the cells were selected with puromycin and diluted to generate single-cell colonies in a 96-well plate. The transfection efficiency was measured by Western blot analysis.

Cells were seeded in plates with different treatment. At indicated time points, cell viability was monitored using sulforhodamine B (SRB) assay protocol. The cell viability was calculated (formula = mean OD_treatment / mean OD_control × 100%) and the time–response curve was plotted to clarify the antiproliferation effect of the inhibitors (21). For crystal violet staining assay, cells were fixed and stained with 0.5% crystal violet solution, and representative images were acquired using a microscope with a digital camera. For quantitative determination, the stained cells were dissolved in methanol, and the absorbance was measured at 570 nm by a microplate reader (22).

Human T cells were obtained from human peripheral blood samples, and mouse T cells were derived from mouse spleens. Briefly, human peripheral blood mononuclear cells were isolated from human peripheral blood by Ficoll density gradient centrifugation. Enrichment of CD8⁺ T cells was performed by magnetic cell sorting (MACS) using the MagCellect Human CD8⁺ T Cell Isolation Kit (R&D Systems) and MagCellect Mouse CD8⁺ T Cell Isolation Kit (R&D Systems) following the manufacturer's instructions. The medium used for human T-cell culture was RPMI1640 supplemented with 10% FBS, 100 U/mL penicillin and streptomycin, 100 IU/mL recombinant human IL2, 2 μg/mL antihuman CD3 antibody, and 1 μg/mL antihuman CD28 antibody. Splenic T cells were cultured in complete RPMI1640 medium in the presence of 2 μg/mL antimouse CD3 antibody and 1 μg/mL antimouse CD28 antibody. After stimulation for 5 days, CD8⁺ T cells were harvested and cocultured with different targeted cells with physical contact. A schematic overview of the procedure is showed in Fig. 3A.

Briefly, after different treatments for adequate time periods, cells were collected for further examination. Tumors from the mouse model were dissected out and disrupted to a single-cell suspension by type IV collagenase and DNase.

For the analysis of PD-L1 levels, cells were harvested and stained with human or mouse anti-PD-L1 antibodies and run on a cytometer for analysis.

To assess the function of CD8⁺ T cells after treatment, cells were harvested and fixed with 4% paraformaldehyde at 4°C for 1 hour. For permeabilization, the fixed cells were washed twice with PBS and resuspended in 500 μL of Triton X-100 for 15 minutes. Then, the cells were washed and stained with Ki67, TNFα, granzyme B, and perforin antibodies for 30 minutes at 37°C. Fluorescence intensity was analyzed by flow cytometry.

In the early stage of treatment, the bladder tumors were removed for tumor infiltrating T-cell analysis. The collected single-cell suspension was stained with the indicated antibodies, then analyzed by flow cytometry. In response to various specific analyses of T-cell activity, CD8⁺ T cells were sorted out using a magcellect mouse or human CD8⁺ T Cell Isolation Kit for further analysis.

The above data analysis was carried out using FlowJo or Cytoexpert software. Antibodies used can be found in Supplementary Table S1.

For hematoxylin and eosin staining, the mice were sacrificed at the endpoint, and the tumor tissue was carefully removed. After fixation with 4% paraformaldehyde, the tissue was embedded and stained using a standard protocol. Images were acquired using a VS120 Olympus microscope with OlyVIA software (Olympus).

Cells were plated on glass coverslips and treated differently. After fixation, the slides were blocked with immunofluorescence blocking fluid and incubated with the primary antibody overnight at 4°C. After washing, the corresponding fluorescent secondary antibodies were added to label the primary antibodies. DAPI was used to stain the nuclear structure. For tumor tissue immunofluorescence analysis, paraffin sections (4 μm) were deparaffinized in xylene and dehydrated in graded ethanol, followed by boiling with 10 mmol/L citrate antigen retrieval solution (Beyotime). Then, the specimens were blocked in goat serum and incubated with primary antibodies at 4°C overnight. After rinsing, the slides were incubated with secondary antibodies at room temperature for 1 hour. Following several washes, the slides were stained with DAPI reagent and mounted with antifade solution. Immunofluorescence was imaged using a laser confocal microscope and analyzed with ZEN Imaging Software by ZEISS.

IHC was performed following the standard protocol: after xylene and graded ethanol treatment, the sections were subjected to antigen retrieval by boiling in citrate buffer. Then the sections were blocked and incubated with the primary antibodies in a humidified chamber overnight. Immunodetection was visualized with diaminobenzidine (DAB), and the sections were counterstained with hematoxylin. Images were acquired using the scanner described above.

For quantification of PD-L1 mRNA expression, total RNA of treated cells was extracted using TRizol reagent according to the protocol, and complementary DNAs (cDNA) were synthesized with a cDNA Synthesis Kit. qPCR reaction was conducted with SYBR Green master mix using a fluorescence ratio PCR instrument (Bio-Rad, CFX Connect). The fold change in PD-L1 mRNA expression between different groups was determined by the 2^−ΔΔCt method. The PD-L1 primers were as follows: human PD-L1 forward 5′-ATTTGGAGGATGTGCCAGAG-3′ and reverse 5′-CCAGCACACTGAGAATCAACA-3′; human actin forward 5′-GCAAAGACCTGTACGCCAACA-3′ and reverse 5′-TGCATCCTGTCGGCAATG-3′; mouse PD-L1 forward 5′-GCATTATATTCACAGCCTGC-3′ and reverse 5′-CCCTTCAAAAGCTGGTCCTT-3′; mouse actin forward 5′-ACCTTCTACAATGAGCTGCG-3′ and reverse 5′-CTGGATGGCTACGTACATGG-3′.

For coimmunoprecipitation assay, the transfected and treated cells were scraped and lysed in IP buffer containing protease and phosphatase inhibitor cocktails on ice, followed by centrifugation at 14,000 × g for 10 minutes at 4°C to remove insoluble materials. The supernatant of samples was incubated with the protein A/G beads with the indicated antibodies overnight at 4°C. After washing with IP buffer, the beads were eluted by boiling with SDS buffer. The coimmunoprecipitations were detected by immunoblotting with the indicated antibodies (Supplementary Table S1).

For immunoblotting, the protein lysate was run on an SDS-PAGE gel and transferred to a polyvinylidene difluoride membrane. After blocking with nonfat milk, the membranes were incubated with specific primary antibodies at 4°C overnight. After washing, the membranes were stained with HRP-conjugated secondary antibodies for 1 hour at 25°C. Protein band detection was performed with ECL Detection Kits (EMD Millipore), and images were obtained using a chemiluminescence apparatus. Quantification of the band intensity was analyzed using ImageJ software.

Statistical differences between groups were determined using two tailed Student t test, one-way ANOVA, and two-way ANOVA. Kaplan–Meier survival curves were used and analyzed by the log-rank (Mantel–Cox) test in mouse experiments using GraphPad Prism Software. The P value in the figure legend indicates *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001 (*, P < 0.05 was regarded as statistically significant).

The data generated in this study are available upon request from the corresponding author.

To determine whether the immune system influenced the efficacy of FGFR inhibition, we inoculated MB49 cells into immunocompetent (C57BL/6) and immunodeficient (CAnN.Cg-Foxn1nu/Crl) mouse bladders (Fig. 1A and B). Interestingly, drug intervention with erdafitinib or infigratinib inhibited tumor progression in immunocompetent mice (the inhibition rate was approximately 50%) but exerted a better suppression effect in immunodeficient mice (the inhibition rate was approximately 75%; Fig. 1C–J). These findings suggest the existing relevance between the immune system and the therapeutic effect of FGFR inhibition. We also evaluated the stability of the therapeutic effect, and found that the erdafitinib and infigratinib had persistent tumor suppression effect (Supplementary Figs. S1A–S1F). Moreover, ultrasound imaging and HE staining of tumor tissue further confirmed the tumor inhibition effect of erdafitinib and infigratinib in tumor-bearing mice (Fig. 1K and L). Both erdafitinib and infigratinib administration conspicuously decreased Ki67 and inhibited the downstream FGFR signaling pathways in nude mice, including phosphorylation of ERK1/2, AKT, and MEK. (Fig. 1K and L; Supplementary Fig. S1G; refs. 23, 24).

To better understand the relationship between FGFR inhibition and tumor immunology, we detected the level of PD-L1 and the proportion of CD8⁺ T cells in tumors treated with erdafitinib and infigratinib by flow cytometry. Unexpectedly, the PD-L1-positive tumor cell ratio increased substantially in the tumors treated with erdafitinib or infigratinib, accompanied by a significant decrease in CD8⁺ T-cell infiltration in tumor tissues (Fig. 1M–P). TNFα and granzyme B were reported to be acknowledged markers for CD8⁺ T cells activation (25, 26). We found that erdafitinib and infigratinib also decreased the proportion of granzyme B and TNFα positive CD8⁺ T cells in MB49 tumors (Fig. 1Q–T). We further evaluated the immune checkpoint PD-L2 expression level of tumors in different groups, as well as PD-1, Tim-3, and LAG-3 expression levels on CD8⁺ T cells, the results showed that erdafitinib and infigratinib made no obvious effect on these markers in vivo (Supplementary Figs. S2A–S2E). We also detected the proportion of immunosuppressive cells including MDSCs and Tregs in tumors treated with erdafitinib and infigratinib, and no differences were displayed from the results (Supplementary Figs. S2F and S2G). Furthermore, genetic deletion of PD-L1 further improves tumor suppression efficacy of erdafitinib and infigratinib (Supplementary Figs. S2H–S2L). Together, our results in bladder cancer mouse model with different immunologic functions led us to discover that the tumor inhibition effect of FGFR inhibitors was accompanied by PD-L1 upregulation and CD8⁺ T-cell inhibition.

FGFR inhibition-mediated PD-L1 upregulation was further verified in multiple bladder cancer cell lines. Strikingly, incubation with erdafitinib and infigratinib induced approximately 40% proliferation inhibition of SW780, RT4, and MB49 cells in which FGFR3 was activated by FGFR3 mutations or high expression (27, 28, 29). In contrast, erdafitinib or infigratinib only induced approximately 20% proliferation inhibition of T24, 5637, and UM-UC-3 cells in which FGFRs were normal (Supplementary Fig. S3A). In addition, cell apoptosis and cell-cycle assessment using flow cytometry revealed that erdafitinib and infigratinib arrested the cell cycle in the G₀–G₁ stage but did not induce cell apoptosis (Supplementary Figs. S3B and S3C). Interestingly, bladder cancer cell PD-L1 levels were significantly upregulated in line with the strong proliferation inhibition rates in SW780, RT4, and MB49 cells with erdafitinib or infigratinib treatment, in contrast to PD-L1 expression levels in T24, 5637, and UM-UC-3 cells, which have normal FGFR levels, that did not increase obviously (Fig. 2A and B; Supplementary Figs. S4A and S4B). The results also showed that erdafitinib or infigratinib could not affect PD-L1 expression on SV-HUC-1 human normal bladder epithelium cell line (Supplementary Figs. S4A and S4B). Because erdafitinib and infigratinib inhibit several FGFRs, including FGFR1–4, it is of interest to identify which FGFR is involved in the process of FGFR inhibition-mediated PD-L1 upregulation. To this end, using an RNA interference screen of FGFR1–4, we identified FGFR3 as a crucial regulator of PD-L1 in bladder cancer cells. The results revealed that FGFR3 interference could upregulate PD-L1 expression to the greatest extent in SW780 and MB49 bladder cancer cells (Fig. 2C and D; Supplementary Figs. S4C and S4D). Fluorescence confocal analysis showed more obvious results that FGFR3 interference could also inhibit FGFR3 phosphorylation and increase PD-L1 levels in human and mouse bladder cancer cells (Fig. 2E). Furthermore, both erdafitinib and infigratinib upregulated PD-L1 in SW780 and MB49 cells by inhibiting the phosphorylation of FGFR3 (Fig. 2F). In addition, we also verified the basal level of FGFR family proteins and PD-L1 in the panel of multiple bladder cancer cell lines used in this study and we found that SW780, RT4, and MB49 cells had significantly high FGFR3 phosphorylation levels accompanied with relatively low PD-L1 expression (Supplementary Figs. S4E and S4F). Together, our findings offer compelling evidence that FGFR3 plays a critical role in PD-L1 regulation among bladder cancer cells with FGFR3 activation. To verify this phenomenon in other tumors, we selected another two cancer cell lines with FGFR3 activating mutations including Ewing sarcoma cell line SK-N-MC and multiple myelomas cell line KMS-11 (30, 31). The results showed that erdafitinib and infigratinib induced moderate PD-L1 expression in SK-N-MC and KMS-11 cells (Supplementary Fig. S4G). These results provide further evidence that the phenomenon we found in bladder cancer has the potential to be applied to certain tumors. In clinical practice, both chemotherapy and targeted therapy are commonly used in bladder cancer treatment. We found that both cisplatin and paclitaxel could elevate the PD-L1 expression in MB49 and T24 cells (Supplementary Fig. S4H).

To further validate our findings in clinical UC of bladder specimens, we analyzed the correlations between FGFR3 activating mutations and PD-L1 expression in 98 human UC bladder specimens using gene sequencing and immunohistochemical staining (Fig. 2G; Supplementary Table S2). We found that high PD-L1 expression was detected in 20 (37.7%) of the 53 specimens with wild-type FGFR3 or nonactivating mutations but in only 2 (4.4%) of the 45 specimens with FGFR3 activating mutations (Fig. 2G). These findings supported the concept that FGFR3 activation associated with PD-L1 downregulation had a reliable clinical correlation. A previous study among 1,000 chemotherapy-naive radical cystectomy patients showed that FGFR3 mutations significantly prolonged cancer-specific survival (7). Most less aggressive cases of nonmuscle-invasive bladder cancer are characterized by activating mutations in FGFR3 (32). In addition, the luminal-papillary subtype of UC, which has a higher percentage of alteration in FGFR3 genes, has better prognosis than basal subtype (13). Similarly, we found that FGFR3 mutations were also associated with low-grade and early-stage disease in 476 patients with bladder cancer from the TCGA database (Fig. 2H). These clinical data supported the notion that FGFR3 activating mutation was consistent with low PD-L1 expression and relatively good cancer prognosis.

To our knowledge, the effect of checkpoint inhibition therapy lies in its ability to potentiate T-cell–centered cancer cell immune destruction. Activated T cells can play a powerful role in tumor killing in vivo and in vitro (33). On the basis of this theory, we used cocultivation experiments with bladder cancer cells to evaluate the effects of FGFR3 inhibition or silencing on T-cell responses. We isolated and stimulated human CD8⁺ T cells from peripheral blood and mouse CD8⁺ T cells from splenocytes to coculture with bladder cancer cells (Fig. 3A). Consistent with the model in which FGFR3 inhibition by inhibitors and knockdown by shRNA upregulated PD-L1 protein levels in bladder cancer cells, the cocultivation results revealed that PD-L1 upregulation in bladder cancer cells largely protected them from being killed by T cells (Fig. 3B and C). In line with bladder cancer cell resistance to the killing effect of T cells, flow cytometry analysis revealed that the killing effect of T cells was seriously inhibited by upregulation of PD-L1 in tumor cells. Moreover, cocultivation with FGFR3-inhibited bladder cancer cells significantly reduced the proportion of activated CD8⁺ T cells indicated by Ki67, TNFα, granzyme B, and perforin positive ratios (Fig. 3D–G).

To investigate the mechanism under FGFR3 regulating PD-L1 levels, we first assessed the relationship between FGFR3 inhibition and PD-L1 mRNA levels by RT-PCR. The results showed that the levels of PD-L1 mRNA with FGFR3 inhibition were not substantially upregulated in accordance with the protein levels (Fig. 4A). Clinical patient data from the TCGA/GTEx (n = 404) and Metabric (n = 1,904) databases revealed that FGFR3 expression has no obvious correlation with PD-L1 expression in cancer at mRNA level (Supplementary Fig. S4I). On the basis of this observation, we sought to determine whether the posttranslational pathway was the most relevant to PD-L1 upregulation in bladder cancer cells by screening common protein degradation pathways using a proteasome inhibitor (MG132) and several autophagy inhibitors. Our results revealed that only MG132 significantly increased PD-L1 in SW780 and MB49 cells (Fig. 4B and C). Broadly speaking, these results confirmed that FGFR3 might participate in PD-L1 degradation mediated by the ubiquitin–proteasome pathway (UPP). PD-L1 in SW780 cells was then pulled down by immunoprecipitation and subjected to ubiquitination analysis. Although PD-L1 had notable basal ubiquitination, FGFR3 interference abolished the ubiquitination of PD-L1 in SW780 cells (Fig. 4D, left). In concert with this, erdafitinib and infigratinib treatment also significantly inhibited PD-L1 ubiquitination (Fig. 4D, right). These results supported the notion that FGFR3 activation induced PD-L1 ubiquitination in bladder cancer cells. To identify the regulatory factor that serves as a bridge between FGFR3 and PD-L1, we first analyzed the protein–protein interaction (PPI) between FGFR3 and common E3 ubiquitin ligases in tumors through the STRING interaction network (Supplementary Table S3; refs. 34, 35). Among the 70 E3 ubiquitin ligases, only five were related to FGFR3 (Fig. 4E). To better identify which E3 ubiquitin ligase participates in the process of PD-L1 ubiquitination, we used the UbiBrowser database, a website that can predict human ubiquitin ligase (E3) substrate interactions, and displayed the top 20 E3 ubiquitin ligases that might participate in PD-L1 ubiquitination (Fig. 4F; ref. 36). To pinpoint which E3 ubiquitin ligase controls PD-L1 ubiquitination in bladder cancer cells, we used siRNA to reduce the relevant E3 ubiquitin ligase levels based on the five E3 ubiquitin ligases predicted to interact with FGFR3. To this end, we found that only NEDD4 interference could upregulate PD-L1 levels, similar to FGFR3 inhibition in human and mouse bladder cancer cells. (Fig. 4G and H; Supplementary Figs. S4J and S4K). This result was consistent with the intersection of the protein interaction predictions and E3 ubiquitin ligase predictions.

To understand how NEDD4 influenced bladder cancer cell PD-L1 levels, we first examined the relationship between upstream FGFR3 and NEDD4. Confocal immunofluorescence analysis revealed that p-FGFR3 colocalized with NEDD4 at the cell surface of bladder cancer cells (Fig. 5A and B). To assess whether FGFR3 activation was necessary for NEDD4 binding, we examined the effect of FGF1, the FGFR3-specific ligand, in 293T cells transfected with plasmids expressing Myc-tagged FGFR3 and Flag-tagged NEDD4. The immunoprecipitation and immunofluorescence results only showed a detectable FGFR3–NEDD4 combination and NEDD4 phosphorylation after FGF1 exposure (Fig. 5C; Supplementary Fig. S5A), whereas FGFR3 activity inhibited by erdafitinib prevented FGF1-induced FGFR3 and NEDD4 binding, consistent with no NEDD4 phosphorylation (Fig. 5D). It has been reported that NEDD4 can be phosphorylated to greatly improve its ubiquitination capacity (19). In the bladder cancer cell line SW780 with FGFR3-activating mutations, both FGFR3 and NEDD4 were phosphorylated at high levels, which were substantially reduced when FGFR3 was inhibited by FGFR inhibitors or shRNAs (Fig. 5E and F). These results revealed that the activation of FGFR3 was necessary for NEDD4 binding and phosphorylation. To further verify whether NEDD4 was critical for PD-L1 degradation in bladder cancer cells, we knocked out NEDD4 using a specific CRISPR/Cas9-sgRNA in SW780 and MB49 cells. As expected, depletion of NEDD4 led to a remarkable elevation of PD-L1 levels in both human and mouse bladder cancer cells (Fig. 5G and H). These results revealed that NEDD4 was a critical regulator for PD-L1 levels in bladder cancer with FGFR3 activation.

A hypothesis arising from the above results is that NEDD4 and PD-L1 can interact molecularly. To confirm this hypothesis, we performed fluorescence confocal localization of NEDD4 and PD-L1. As expected, NEDD4 colocalized with PD-L1 in bladder cancer cells (Fig. 6A and B). We knocked out PD-L1 from SW780 and MB49 cells and found that the NEDD4 was still distributed in the cell, but PD-L1 antibody only remains little nonspecific staining (Supplementary Figs. S5B and S5C). To further determine whether NEDD4 interacts with PD-L1, we expressed Myc-tagged FGFR3, Flag-tagged NEDD4, and V5-tagged PD-L1 in 293T cells. The levels of NEDD4 coimmunoprecipitated with PD-L1 significantly increased in cells in which FGFR3 was activated by FGF1 and NEDD4 was phosphorylated (Fig. 6C). These results indicated that the interaction between NEDD4 and PD-L1 was obviously enhanced in cells after NEDD4 was phosphorylated by FGFR3 activation. In addition, we also found that NEDD4 significantly coimmunoprecipitated with PD-L1 in SW780 and MB49 cells (Supplementary Fig. S5D). To further verify whether NEDD4 regulated PD-L1 protein levels through its E3 ligase activity, we assessed the NEDD4-mediated ubiquitination process of PD-L1 in bladder cancer cells. NEDD4-mediated polyubiquitination of PD-L1 was clearly detectable in SW780 cells transfected with plasmids expressing V5-tagged PD-L1 and hemagglutinin (HA)-tagged ubiquitin in the presence of a plasmid expressing Flag-tagged NEDD4 (Fig. 6D). To investigate the type of NEDD4-mediated polyubiquitination of PD-L1, we used vectors expressing HA-tagged mutant ubiquitin (K48) or ubiquitin (K63), which contain substitution of arginine for all lysine residues except the lysine at position 48 or 63, respectively. NEDD4 catalyzed the polyubiquitination of PD-L1 in the presence of HA-tagged wild-type ubiquitin [HA–ubiquitin (WT)] and HA–ubiquitin (K48) but not in the presence of HA–ubiquitin (K63) in SW780 cells (Fig. 6D). These data suggested that NEDD4 catalyzed the K48-linked polyubiquitination of PD-L1. In addition to the above findings, we identified that NEDD4 was necessary for PD-L1 ubiquitination by using SW780 cells with NEDD4 knockout. The ubiquitination of PD-L1 was significantly inhibited after NEDD4 knockout or treatment with erdafitinib (Fig. 6E). Pulse-chase analysis using cycloheximide revealed that NEDD4 deletion significantly prolonged the half-life of the PD-L1 protein in SW780 cells, whereas NEDD4 overexpression significantly shortened the PD-L1 protein half-life (Fig. 6F and G). The cocultivation results revealed that NEDD4 deletion in bladder cancer cells protected them from being killed by T cells to a great extent (Fig. 6H). Moreover, cocultivation with NEDD4-knockout bladder cancer cells significantly diminished the proportion of Ki67, TNFα, granzyme B, and perforin positive CD8⁺ T cells (Fig. 6I and J). To further determine the effects of NEDD4 on bladder cancer control in vivo, we used NEDD4 WT/KO MB49 cells to establish bladder cancer models in immunocompetent C57BL/6 mice (Supplementary Fig. S6A). The tumor growth speed in mice bearing NEDD4-KO MB49 tumors was significantly increased compared with that in mice bearing NEDD4-WT MB49 tumors (Supplementary Figs. S6B–S6F). The gap in tumor growth speed between the two groups was largely narrowed after anti-PD-1 antibody treatment (Supplementary Figs. S6B–S6F). Notably, tumors with NEDD4 deletion had markedly increased PD-L1 levels and decreased proportions of CD8⁺ T cells and granzyme B expression, which were reversed by treatment with an anti-PD-1 antibody (Supplementary Fig. S6G). These findings revealed that NEDD4 could downregulate the PD-L1 in MB49 cells and that NEDD4 deletion could help tumor cells subvert immune surveillance.

Because FGFR3 inhibition could both inhibit tumor growth and regulate PD-L1 stability, we sought to determine whether FGFR3 inhibition combined with ICB would enhance antitumor immunity accompanied by tumor growth inhibition. We established three mouse models including MB49 in a C57BL/6 mouse model, SW780 in a humanized immune reconstruction mouse model, and PDX mouse models bearing tumors with FGFR3 activating mutations from clinical patients. C57BL/6 mice were randomly grouped on the basis of tumor fluorescence intensity and treated with each formulation from 3 days post-inoculation with Luci+ MB49 cells (Supplementary Fig. S7A). In the initial 11 days after drug administration, remarkable tumor regression was observed in the combination therapy group compared with that in the monotherapy groups (Supplementary Figs. S7B–S7E). Combination therapy also significantly prolonged the survival of tumor-bearing mice. (Supplementary Fig. S7F). Ultrasound imaging and tumor tissue section staining further confirmed the excellent tumor regression effect and Ki67 inhibition of combination administration (Supplementary Fig. S7G). Interestingly, erdafitinib combined with anti-PD-1 antibody therapy increased the total proportion and activated CD8⁺ T cells compared with monotherapy administration, which indicated that combination therapy could fully activate the tumor-killing effect of the mouse immune system (Supplementary Fig. S7H). The safety and tolerability evaluation of the combination therapy showed that no significant relevant changes existed (Supplementary Figs. S8A–S8C).

To verify the therapeutic effect of combination therapy on human bladder cancer, we established a humanized immune system in NOG mice using human CD34⁺ HSC transplantation after radiation to achieve human bladder cancer cell implantation (Fig. 7A). Hu-HSC-NOG mice were randomly grouped and treated with each formulation from 5 days post-inoculation with Luci+ SW780 cells (Fig. 7B). Combination therapy achieved the best effect on suppressing tumor growth and prolonging survival time among all the treated groups (Fig. 7C–G). The results from ultrasound imaging and flow cytometric analysis showed that SW780 tumor-bearing mice treated with combination therapy could suppress tumor growth by inhibiting cell proliferation and stimulating antitumor immune responses centered around CD8⁺ T cells (Fig. 7H–K).

To be closer to the clinical application, we established PDX mouse models on hu-HSC-NOG mice using a UC of the bladder specimen from an elderly male patient with high-grade UC of the bladder that had FGFR3–TACC3 gene fusion detected by NGS. Bladder cancer PDX mice with a humanized immune system were randomly grouped on the basis of ultrasound imaging assessment and treated with each formulation from 5 days post-inoculation (Fig. 7L). The combination treatment inhibited tumor growth and prolonged median overall survival more effectively than that in control mice or mice treated with monotherapy (Fig. 7M–P). We examined the PD-L1 expression levels in the PDX tumors, and the results indicated that PD-L1-positive cells ratio conspicuously increased in the tumors treated with erdafitinib (Fig. 7Q). Besides, exhaustion markers including Tim-3, LAG-3, and PD-1 on CD8⁺ T cell showed no obvious difference among different treated group (Fig. 7R). Strikingly, these results demonstrated that our research had important potential value that could be translated into clinical application.

Previous studies on molecular subtypes and ICB therapy have demonstrated that there is a relatively low immune signature and decreased expression of PD-L1 in UC of the bladder with FGFR3 mutation or high expression (6, 9, 13, 37), but the regulatory mechanism is obscure. Our study reveals that the activation of FGFR3 pathways leads to the downregulation of PD-L1 and that the PD-L1 level is elevated after administration of FGFR inhibitors in UC of the bladder with FGFR3 activation. More importantly, we uncovered the underlying mechanism about the significant number of patients with FGFR3-activated UC of the bladder could not benefit from FGFR inhibition and proposed the combination therapy to address this challenge. Most studies have revealed that the therapeutic effect of PD-1/PD-L1 inhibitors is closely related to PD-L1 expression levels in various cancers (38, 39). However, the specific mechanism underlying which PD-L1 is regulated in UC of the bladder remains unclear. Herein, our study illustrates the intrinsic pathway of PD-L1 regulation using a series of in vivo and in vitro experiments in which PD-L1 can be destabilized via NEDD4 phosphorylation caused by FGFR3 activation.

The abnormal regulation of NEDD4 plays a critical role in oncogenesis and tumor progression (40, 41). NEDD4, which is frequently overexpressed in most cancers, is associated with chemotherapy and targeted therapy resistance (40), whereas there are no studies concerning the influence of NEDD4 on immunotherapy. Our findings reveal potentially critical roles of NEDD4 in the regulation of PD-L1 expression. The expression level and activating status of NEDD4 may play roles in the response to PD-1/PD-L1 inhibitors. In our study, we found that although NEDD4 knockout led to rapid tumor growth, the response to anti-PD-1 antibody was more sensitive than that in NEDD4 WT tumors. Hence, the NEDD4 E3 ligase may be a promising predictive biomarker and therapeutic target in advanced UC of the bladder with immunotherapy or combined therapy.

Mutations and high expression of FGFR3 have different effects on bladder cancer prognosis. Patients enrolled in clinical trials with different FGFR inhibitors had different inclusion criteria for mutations or high expression, and the clinical trial results were also different (42). Our study included both bladder cancer with high expression and mutations of FGFR3, and we conclude that the activation of FGFR3 is the key issue rather than expression or mutation. An aberrant FGFR signaling axis was found to function in oncogenesis, tumor progression, and resistance to anticancer therapy across different tumor types (43). FGFR3 mutation or fusion was also found in breast cancer, glioblastoma, intrahepatic cholangiocarcinoma, cervical cancer, and so on, but the mutation frequency was lower than that of UC of the bladder (44). Numerous FGFR inhibitors are currently being assessed in preclinical, phase 1, phase 2, and phase 3 clinical trials in different cancer types (37). It has been reported that erdafitinib inhibits PD-L1 expression in human lung tumor and FGFR2 promoted expression of PD-L1 in colorectal cancer (45, 46). In this study, we showed that erdafitinib and infigratinib induce PD-L1 expression in bladder cancer cells. Furthermore, we verified this phenomenon in another two FGFR3 activating mutations cancer cell lines and found that FGFR3 inhibition also induced PD-L1 expression in these cells. On the basis of this, we speculated the phenomenon that erdafitinib and infigratinib inducing PD-L1 expression in bladder cancer cells may be cancer selective. Future studies should be conducted to extend this study to other cancer types and generalize our hypothesis.

Recent studies have demonstrated that transcriptional and posttranscriptional modifications of PD-L1 through several signaling pathways including JAK/STAT3, GSK3β, SPOP, and STT3 have emerged as important regulatory mechanisms that modulate immunosuppression in cancer (15, 16). Previous studies have shown that chemotherapy treatment can also induce PD-L1 expression in lung cancer (47, 48). However, chemotherapy and targeted therapy differ greatly in their mechanisms of action and therapeutic indication. The particularity of PD-L1 regulation in bladder cancer, especially in certain subtypes of bladder cancer, has hardly been well elucidated in previous studies. Here, we identified FGFR3 as a new regulator of PD-L1 protein levels through NEDD4 in FGFR3-activated bladder cancer. On the basis of the available data, we conclude that the activation of FGFR3 in bladder cancer recruits NEDD4 and phosphorylates it, whereas NEDD4 phosphorylation can activate its ability to ubiquitinate PD-L1 protein (Fig. 7S). Furthermore, the observations that FGFR3 activating mutation in clinical data is inversely correlated with PD-L1 levels suggest a model in which FGFR3 is correlated with bladder cancer immunology by influencing PD-L1. In line with this, it can be speculated that FGFR3 and NEDD4 in bladder cancer may fulfill an immunoregulatory role by influencing the ubiquitination of the immune checkpoint molecule PD-L1. Finally, multiple mouse models, including PDX mice bearing an FGFR3–TACC3 gene fusion bladder tumor from clinical excision, showed that erdafitinib and anti-PD-1 antibody combination treatment significantly enhanced the effectiveness of the current ICB and targeted therapy. These results may change our understanding of FGFR3 targeted therapy in UC of the bladder.

No disclosure was reported by the authors.

W. Jing: Conceptualization, data curation, software, formal analysis, visualization, methodology, writing–original draft. G. Wang: Software, formal analysis, visualization, methodology, writing–original draft. Z. Cui: Formal analysis, methodology. G. Xiong: Methodology. X. Jiang: Formal analysis, investigation. Y. Li: Methodology. W. Li: Resources. B. Han: Formal analysis. S. Chen: Resources, supervision, writing–review and editing. B. Shi: Resources, supervision, funding acquisition, project administration, writing–review and editing.

This research was supported by the National Natural Science Foundation of China (grants 81670687 and 81970661 to B. Shi; grant 81800672 to S. Chen), the Tai Shan Scholar Foundation (ts201511092 to B. Shi); Primary Research & Development Plan of Shandong Province (2019GSF108123 to S. Chen) and CSCO Clinical Oncology Research Foundation (Y-2019AZQN-0557 to S. Chen). All schematic figures were created with BioRender.com.

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.