A Genome-Wide Synthetic Lethal Screen Identifies Spermidine Synthase as a Target to Enhance Erdafitinib Efficacy in FGFR-Mutant Bladder Cancer Open Access

Bladder cancer is a major source of morbidity and mortality worldwide (1). At the time of initial diagnosis, 25% of patients have muscle-invasive bladder cancer (MIBC) or metastatic disease, and MIBC exhibits a more diverse mutation spectrum compared with non-MIBC (2, 3). Besides, in MIBC, aberrations of FGFR family members are frequent events and associated with a worse prognosis (4). Therefore, targeted therapeutic intervention with selective FGFR inhibitors represents a promising approach for oncogenic activating FGFR alterations in bladder cancer.

Erdafitinib, a pan-FGFR tyrosine kinase inhibitor, was granted accelerated approval by the Food and Drug Administration (FDA) for patients with FGFR2/3-driven urothelial cancer based on the BLC2001 study (5). In another phase II trial, erdafitinib shows a considerable objective response rate of 40% and a median progression-free survival of 5.5 months in urothelial cancer (6). However, the rapid development of acquired resistance remarkably reduces erdafitinib efficacy. Although adipocyte precursors have been reported to mediate resistance against erdafitinib via the NRG1/HER3 axis in bladder cancer (7), the cancer-intrinsic mechanisms underlying erdafitinib resistance remain largely undetermined. Recently, the CRISPR-Cas9 system has enabled unbiased screens for vulnerabilities in cancers, and several CRISPR-Cas9 screens have identified synthetic lethal (SL) targets that enhanced FGFR inhibitor efficacy, including EGFR in liver cancer (8) and PLK1 and MTOR in lung cancer (9, 10). Nonetheless, the efficacy of specific synthetic lethality–based strategies may depend on the genetic background of different tumor types, and no genome-wide screen exploring erdafitinib resistance in bladder cancer has been reported. Therefore, a systematic screen is still required to identify SL targets that enhance erdafitinib-targeted therapy efficacy of patients with bladder cancer.

Metabolic reprogramming is widely recognized as a hallmark of cancers, and dysregulation of polyamine metabolism has been linked to various cancers, with elevated polyamine levels being essential for transformation and tumor progression (11). Previous studies implied that polyamine metabolism may influence growth (12) and chemotherapy sensitivity (13) in bladder cancer; however, the underlying mechanisms remain unclear, and whether polyamine metabolism involves in erdafitinib-targeted therapy is unidentified. As a crucial enzyme in polyamine metabolism, spermidine synthase (SRM) catalyzes the conversion of putrescine to spermidine, which subsequently serves as a substrate for the hypusination of a conserved lysine residue in eukaryotic initiation factor 5A (eIF5A; ref. 14). Furthermore, hypusinated eIF5A (eIF5A^Hyp) supports the efficient translation of a subset of mRNAs with specific motifs (15, 16). Nevertheless, the regulatory functions as well as complexities of SRM and eIF5A in modulating erdafitinib resistance in bladder cancer are yet to be revealed.

To more comprehensively define potential combination strategies, we performed an unbiased whole-genome CRISPR-Cas9 SL screen to identify genes whose deletion augments erdafitinib efficacy and identified SRM as a potential target. More specially, the combination of SRM knockout (KO) and erdafitinib significantly inhibited tumor growth, whereas the KO of SRM alone had little effect on bladder cancer cells. Furthermore, SRM depletion significantly attenuated HMGA2 translation via eIF5A hypusination. Additionally, HMGA2 bound to the promoter of EGFR and HMGA2 knockdown reduced the expression of EGFR. Importantly, pharmacologic inhibition of SRM with trans-4-methylcyclohexylamin (MCHA) significantly enhanced erdafitinib efficacy. Thus, the SL therapy of MCHA with erdafitinib may serve as a promising strategy for increasing the response to erdafitinib-targeted therapy in bladder cancer.

Human bladder cancer cell lines MGH-U3 (RRID: CVCL_9827), SW780 (RRID: CVCL_1728), RT4 (RRID: CVCL_0036), and RT112 (RRID: CVCL_1670), along with human bladder cell line SV-HUC-1 (RRID: CVCL_3798) and human embryonic kidney (HEK) 293T (RRID: CVCL_0063) cells, were purchased from the ATCC. MGU-U3 harbors the Y375C mutation, SW780 carries the FGFR3-BAIAP2L1 fusion, whereas RT4 and RT112 contain FGFR3-TACC3 fusions. MGH-U3 and RT4 cells were cultured in RPMI-1640 medium (Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). SW780 cells were cultured in L15 medium (Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). HEK293T cells were cultured in DMEM (Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). RT112 cells were cultured in Minimum Essential Medium (MEM; Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). All cells were cultured in an incubator at 37°C with humidified atmosphere of 5% CO₂. All cell lines were confirmed within 6 months before use by short tandem repeat profiling and were confirmed negative for Mycoplasma contamination.

As previously described, total RNA from tissue samples and cell lines was extracted using TRIzol reagent (Invitrogen) according to the manufacturer’s instructions (17). cDNA was synthesized by HiScript III RT SuperMix for qPCR (Vazyme). RT-PCR analyses were performed with SYBR Green Master Mix (Vazyme). The primers used are listed in Supplementary Table S1. The results were analyzed using the StepOne Plus Real-Time PCR System (Applied Biosystems), and transcript levels were quantified using the 2^−ΔΔC_t method.

A total of 108 pairs of bladder cancer tissues and corresponding adjacent normal tissues were collected from patients who underwent radical cystectomy at the Department of Urology, Union Hospital, Tongji Medical College, Wuhan, China, between 2014 and 2019. All specimens were independently classified by at least two experienced pathologists according to the sixth edition of the TNM classification system of the International Union Against Cancer. The study was approved by the Ethics Review Committee of Tongji Medical College, Huazhong University of Science and Technology, and all patients provided written informed consent prior to participation. We certify that the study was performed in accordance with the Declaration of Helsinki. The tissues were snap-frozen in liquid nitrogen immediately after collection and stored at −80°C. Patients were followed up regularly, with overall survival defined from the date of surgery to the date of death or the last follow-up for surviving patients.

HEK293T cells were seeded in 10-cm dishes to achieve 85% to 90% confluency at the time of transfection. Transfer plasmids, along with psPAX2 (RRID: Addgene_12260) and pDM2.G (RRID: Addgene_12259), were cotransfected using Lipofectamine 3000. Specifically, 9 μg transfer plasmids, 6 μg psPAX2, 3 μg pDM2.G, and 36 μL of p3000 reagent were added to 1.5 mL of OPTI-MEM medium (Thermo Fisher Scientific). In parallel, 23 μL of Lipofectamine 3000 reagent was diluted in 1.5 mL of OPTI-MEM medium. The plasmid mixture and diluted Lipofectamine were then combined and incubated for 15 minutes at room temperature. Afterward, 3 mL of culture medium were removed, and 3 mL of transfection mix were added. The lentivirus-containing supernatant was harvested at 24 and 56 hours after transfection, then filtered through a 0.45-μm filter, and purified by centrifugation at 19,400 rpm under cold conditions. The virus was resuspended in 100 μL of PBS, aliquoted, and stored at −80°C.

The Human GeCKO (RRID: SCR_009001) v2 CRISPR KO pooled library was used to screen for genes driving erdafitinib resistance in bladder cancer cells (18). MGH-U3 and SW780 cells were transduced with the GeCKO v2 library, which contains 122,411 unique single-guide RNAs (sgRNA) targeting 19,052 human genes and 1,864 miRNAs (6 sgRNAs per gene, 4 sgRNAs per miRNA, and 1,000 nontargeting controls) at a low multiplicity of infection (0.3). Transduced cells were cultured in medium containing 2 μg/mL of puromycin for 7 days to generate a mutant cell pool. The pool was then treated with either vehicle (DMSO) or erdafitinib (100 μmol/L; MedChemExpress, HY-18708) for 4 days. At least 3 × 10⁷ cells were harvested for genomic DNA extraction, ensuring more than 400× coverage of the GeCKO v2 library. The sgRNA sequences were amplified using NEBNextR High-Fidelity 2× PCR Master Mix (NEB, M0541) and subjected to massive parallel amplicon sequencing carried out by Novogene Technology. The sgRNA read count and hit calling were analyzed by the MAGeCK (RRID: SCR_025016) v0.5.7 algorithm (19). The CRISPR screening results were deposited in the Gene Expression Omnibus database (GEO; RRID: SCR_005012): https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE276232.

The sgRNAs targeting SRM (Supplementary Table S2) were synthesized by TSINGKE and cloned into the lentiCRISPR v2 vector (Sigma-Aldrich). Short hairpin RNAs targeting HMGA2 (Supplementary Table S2) were synthesized by TSINGKE and cloned into the pLKO.1 vector (RRID: Addgene_10879). Human HMGA2 and SRM cDNAs were synthesized by TSINGKE and cloned into the pcDNA3.1-3×Flag-C vector to construct HMGA2 and SRM overexpression plasmids. HMGA2_mCherry_WT and HMGA2_mCherry_MUT1-5 were synthesized by TSINGKE. Lipofectamine 2000 (Life Technologies) was used for plasmid transfection according to the manufacturer’s instructions. MGH-U3 and SW780 cells were infected with lentivirus expressing sgRNAs targeting SRM, and stable cell lines were selected using puromycin (Invitrogen).

The EGFR promoter reporter was designed and synthesized into the pGL3-Basic (RRID: Addgene_212936) vector by TSINGKE. The EGFR promoter reporter was transiently cotransfected with a Renilla control plasmid. Dual-Luciferase Reporter Assay Kit (Promega) was used to measure the luciferase activities according to the manufacturer’s protocol.

Tissues and cell lines were collected and lysed in RIPA buffer (Thermo Fisher Scientific) supplemented with protease inhibitor cocktail (MedChemExpress). Protein concentration was determined using BCA Protein Assay Kit (HYCEZMBIO). Total protein was subjected to 10% SDS-PAGE gels and transferred to nitrocellulose membranes (Millipore). The membranes were blocked with 5% nonfat milk for 1 hour at room temperature, followed by overnight incubation at 4°C with primary antibodies. Afterward, membranes were incubated with horseradish peroxidase–conjugated secondary antibodies for 1 hour at room temperature. Protein bands were visualized using an enhanced chemiluminescence substrate kit (Millipore), and images were captured using the Bio Spectrum 600 Imaging System (UVP). Antibodies used included primary antibodies against GAPDH (60004-1-Ig, Proteintech, RRID: AB_2263076), SRM (19858-1-AP, Proteintech, RRID: AB_10665555), eIF5A (11309-1-AP, Proteintech, RRID: AB_2262001), eIF5A^Hyp (RGK08101, AntibodySystem SAS, RRID: AB 3665852), HMGA2 (20795-1-AP, Proteintech, RRID: AB_2665377), EGFR (51071-2-AP, Proteintech, RRID: AB_10596476), AKT (10176-2-AP, Proteintech, RRID: AB_2224574), pAKT (80455-1-RR, Proteintech, RRID: AB_2918892), pEGFR(AP0992, ABclonal, RRID: AB_2863885), MYO1B (15012-1-AP, Proteintech, RRID: AB_10642004), LAMC2 (19698-1-AP, Proteintech, RRID: AB_10644139), IL1A (16765-1-AP, Proteintech, RRID: AB_10641044), ZNF185 (31477-1-AP, Proteintech, RRID: AB_3665885), ANXA8L1 (PA5-76232, Thermo Fisher Scientific, RRID: AB_2719959), ANXA3 (ab127924, Abcam, RRID: AB_11143246), SERPINC1 (A11249, ABclonal, RRID: AB_2861534), TGM2 (15100-1-AP, Proteintech, RRID: AB_2202885), RALA (13629-1-AP, Proteintech, RRID: AB_2269251), C1orf116 (14888-1-AP, Proteintech, RRID: AB_2228225), CAV1 (16447-1-AP, Proteintech, RRID: AB_10732595), KRT6A (10590-1-AP, Proteintech, RRID: AB_2134306), SBDS (17618-1-AP, Proteintech, RRID: AB_2184481), NSFL1C (15620-1-AP, Proteintech, RRID: AB_2878158), TXNDC5 (19834-1-AP, Proteintech, RRID: AB_10644285), TOMM34 (12196-1-AP, Proteintech, RRID: AB_2240906), EIF6 (10291-1-AP, Proteintech, RRID: AB_2096515), and SERPINB2 (16035-1-AP, Proteintech, RRID: AB_2186180).

Cell proliferation was assessed using Cell Counting Kit-8 (Dojindo) according to the manufacturer’s instructions. Optical density at 450 nm was measured using a Synergy4 microplate reader (BioTek).

A total of 1,500 to 5,000 cells were seeded in six-well plates and cultured in complete medium supplemented with MCHA, afatinib, or erdafitinib at the indicated dose for approximately 2 weeks. Cells were fixed with ethanol and stained with 0.1% crystal violet.

Tumor xenograft assay was performed following previous protocol (17). All procedures for the animal experiments were approved by the Animal Care Committee of Tongji Medical College. Four-week-old female BALB/c nude mice were chosen for tumor xenografts experiments. All animals were randomly assigned to the experimental or control group (six mice per group), and no blinding was used in the experiments. Bladder cancer cells (5 × 10⁶) were subcutaneously injected into the right axilla of the nude mice. Tumor growth rates were monitored every other week. Tumor volume was calculated according to the following formula: tumor volume = π/6 × length × width². When tumors reached 100 mm³, mice were randomly assigned to 5 days per week treatment with vehicle, erdafitinib (15 or 5 mg/kg, oral gavage), MCHA (5 mg/kg, oral gavage), or combination for 4 weeks. Then, all mice were euthanized by cervical dislocation, and the tumors were photographed and analyzed.

Total RNA was isolated from SRM KO bladder cancer cells and corresponding control cells treated with erdafitinib using TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. Transcriptome sequencing was performed by GeneRead Biotechnology. The RNA sequencing (RNA-seq) data were deposited in the GEO database: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE276411.

As previously described, a RIP assay was conducted following the instructions provided using Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore; ref. 17). Briefly, approximately 1 × 10⁷ cells were harvested and lysed to extract total protein. The proteins were then immunoprecipitated using antibodies against eIF5A (HY-P80653, MCE, RRID: AB_3665877), rabbit IgG (ab172730, Abcam, RRID: AB_2687931), or protein A/G magnetic beads (Life Technologies). After treating with proteinase K, input and coimmunoprecipitated RNAs were extracted using RNeasy Mini Kit (QIAGEN) according to the manufacturer’s instructions and analyzed by qRT-PCR.

Cells were fixed in 1% formaldehyde in PBS for 10 minutes, lysed, and sonicated to fragment the chromatin. The lysates were then immunoprecipitated using HMGA2 antibodies or negative control IgG. The enrichment of HMGA2 protein bound to specific DNA fragments of the EGFR promoter was assessed by PCR. Primers for chromatin immunoprecipitation (ChIP)–PCR amplification are presented in Supplementary Table S1. Chromatin input without immunoprecipitation was used as a positive control.

Cells were treated with 10 μmol/L MCHA or DMSO (control) for 96 hours. The medium was then replaced with labeling medium containing 30 μmol/L O-propargyl-puromycin (OPP; cat# 1407-5, Click Chemistry Tools) in 0.1% DMSO, and cells were incubated for 3 hours. Afterward, cells were washed twice with cold PBS and lysed in RIPA buffer supplemented with 1× protease inhibitor cocktail (cOmplete, EDTA-free, Roche). OPP-tagged proteins were conjugated with biotin via Click Chemistry Tools using biotin picolyl azide (cat# 1167-5) and Click-&-Go Protein Reaction Buffer Kit (cat# 1262, Click Chemistry Tools) from Click Chemistry Tools according to the provided instructions. After biotin–azide conjugation, samples were precipitated by adding 5 volumes of cold acetone and incubating overnight at −20°C. The precipitated proteins were pelleted by centrifugation at 4,000 × g at 4°C for 10 minutes and washed twice with 1 mL cold methanol. Pellets were resuspended in PBS with 1% SDS, and protein concentrations were determined using BCA Assay Kit. A measure of 10 μg of protein were reserved as an input control, whereas 500 μg of protein was incubated with high-capability streptavidin magnetic beads (cat# 1497-1, Click Chemistry Tools) at 4°C overnight with gentle rotation. The beads were washed three times with 1 mL cold PBS with 1% NP-40 and 0.1% SDS. Biotin-labeled nascent proteins were eluted by boiling the beads in 2× Laemmli sample buffer for 10 minutes and subsequently analyzed by Western blotting. The translation rate of HMGA2 protein was assessed by comparing pulldown levels with input.

ELISA assays of polyamine metabolism, including putrescine and polyamine, were performed with commercially available kits from HYCEZMBIO according to the manufacturer’s protocol.

Whole blood samples were stored overnight at 4°C. Afterward, the samples were centrifuged at 3,000 rpm for 15 minutes at 4°C, and the supernatant were collected for immediate analysis. The blood alanine aminotransferase (ALT), aspartate aminotransferase (AST), TBIL, DBIL, ALB, ALP, γGT, TBA, blood urea nitrogen (BUN), creatinine (CREA), uric acid (UA), creatine kinase (CK), CK-MB, lactate dehydrogenase (LDH), LDH1, and phosphorus of mice were measured by using a fully automated biochemical analyzer (Chemray 800, Rayto).

Previously published ChIP-seq data for HMGA2 (20) in A549 cells were reanalyzed using WashU Epigenome Browser and mapped into human genome hg38 using default parameter.

Two- to three-micrometer sections were prepared from formalin-fixed, paraffin-embedded tissues. Antigen retrieval was performed by microwaving the sections in boiling 10 mmol/L citrate buffer (pH 6.0) for 10 minutes at 600 W. The sections were then incubated with anti-Ki67 antibody (ab15580, Abcam, RRID: AB_443209), EGFR (51071-2-AP, Proteintech, RRID: AB_10596476), and pAKT (80455-1-RR, Proteintech, RRID: AB_2918892) incubation.

Briefly, 6 hours after polarization, the cells were lysed and centrifuged through a 10% to 50% sucrose gradient and fractionated using a piston gradient fractionator (Bio-Comp Instruments). The fractions were collected, and total RNA from the fractions was reverse-transcribed and analyzed by qPCR. Primers for HMGA2 were used for qRT-PCR.

Polyamines were measured by high-performance liquid chromatography (HPLC). An Agilent liquid chromatograph equipped with a VWD detector (UltiMate3000, Thermo Fisher Scientific/HPLC 1200, Agilent Technologies) was used to profile the metabolites content by Norminkoda Biotechnology Co., Ltd. The HPLC separation was performed on a ZORBAX ODS C18 guard column (3.9 mm × 20 mn, 5 μm) at 25°C. The solvents for the mobile phase were A (0.1% aqueous acetic) and B (methanol with 0.2% acetic acid). The flowrate was 1 mL/minutes, and the injection volume was 200 μL (1 × 10⁷ cells). The polyamine content was expressed as μg/mL.

The proteomics analysis was performed by Shanghai Bioprofile Technology Co., Ltd. Protein was extracted from 1 × 10⁸ wild-type (WT) and SRM KO MGH-U3 cells treated with DMSO or erdafitinib using SDT lysis buffer (4% SDS, 100 mmol/L DTT, and 100 mmol/L Tris-HCl, pH 8.0). The samples were boiled for 5 minutes, ultrasonicated, and boiled again for an additional 5 minutes. Undissolved cellular debris were removed by centrifugation at 16, 000 g for 15 minutes. The supernatant was collected and quantified using BCA Protein Assay Kit (Bio-Rad). Detergent, dithiothreitol, and idoacetamide were added in UA buffer to block reduced cysteines. The protein suspension was then digested with trypsin (Promega) at a 50:1 ratio overnight at 37°C. Peptides were collected by centrifugation at 16,000 × g for 15 minutes and desalted using a C18 StageTip device for further LC/MS analysis. Peptide concentrations were determined at OD280 using a NanoDrop One device. LC-MS/MS was performed on a Q-Exactive HF-X mass spectrometer coupled with an Easy 1200 nLC system (Thermo Fisher Scientific). Peptides were first loaded onto a trap column (100 μm × 20 mm, 5 μm, C18, Dr. Maisch, GmbH) in buffer A (0.1% formic acid in water). Reverse-phase HPLC separation was carried out using a self-packed column (75 μm × 150 mm; 3 μm ReproSil-Pur C18 beads, 120 Å, Dr. Maisch, GmbH) at a flow rate of 300 nL/minutes. The reverse-phase HPLC mobile phase A was 0.1% formic acid in water, and mobile phase B was 0.1% formic acid in 95% acetonitrile. Peptides were eluted for more than 60 minutes with the following linear gradient of buffer B: 0 to 2 minutes, 2% to 8%; 2 to 45 minutes, 8% to 28%; 45 to 50 minutes, 28% to 40%; 50 to 52 minutes, 40% to 100%; and 52 to 60 minutes, 100%. Mass spectrometry (MS) data were acquired using a data-dependent top-20 method, dynamically selecting the most abundant precursor ions from the survey scan (350–1,800 m/z) for HCD fragmentation. The instrument was operated with peptide recognition mode enabled, and a lock mass of 445.120025 Da was used for mass calibration. Full MS scans were acquired at a resolution of 60,000 at m/z 200, with MS/MS scans at 4,500 at m/z 200. The maximum injection time was set to 50 ms for both MS and MS/MS. Normalized collision energy was 27, with an isolation window of 1.6 Th and a dynamic exclusion duration of 60 seconds. The MS proteomics data have been deposited to the ProteomeXchange Consortium (https://proteomecentral.proteomexchange.org) via the iProX partner repository (21, 22) with the dataset identifier PXD055531.

All data are indicated as the mean ± SD processed by GraphPad Prism 8.0 (RRID: SCR_002798). The Student t test was used to assess the group difference. Kaplan–Meier survival and ROC curves, the log-rank test, and Cox regression analysis were used to evaluate survival difference. P < 0.05 was considered statistically significant.

The CRISPR screening results were deposited in the GEO database: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE276232. The RNA-seq data were deposited in the GEO database: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE276411. The MS proteomics data have been deposited to the ProteomeXchange Consortium (https://proteomecentral.proteomexchange.org) via the iProX partner repository with the dataset identifier PXD055531. All other raw data generated in this study are available upon request from the corresponding author.

To search for genes whose deletion enhances erdafitinib efficacy, we performed genome-wide CRISPR/Cas9 KO library screen on two FGFR-mutant bladder cancer cell lines (MGH-U3 and SW780) treated with erdafitinib-targeted therapy. The human GeCKOv2 CRISPR library that contains 122,411 unique sgRNAs targeting 19,052 protein-coding genes and 1,864 miRNAs was utilized to generate a mutant cell pool (Fig. 1A). Infected cells were treated with DMSO or 100 nmol/L erdafitinib for 4 days, and then cells were harvested and subjected to next-generation sequencing to analyze the differential sgRNA representation. Negative screen analysis was performed based on the MAGeCK algorithm (Fig. 1B; ref. 19). Shared and unique SL genes were identified (P < 0.05, log2 fold change < −0.5 and good sgRNAs ≥5) in negative screen analysis (Fig. 1C; Supplementary Table S3). Then we analyzed 108 pairs of bladder cancer and paired normal bladder tissues. We found that SRM was significantly upregulated in bladder cancer tissues (Fig. 1D), and Kaplan–Meier survival analysis revealed that high expression of SRM was remarkably associated with poor prognosis in patients with bladder cancer (Fig. 1E; Supplementary Fig. S1A). Moreover, ROC curve analysis (Supplementary Fig. S1B) and Cox regression analysis (Supplementary Fig. S1C) further confirmed the association between high SRM expression and unfavorable clinical outcomes in these patients. Thereby, we identified SRM as a potential SL gene.

To validate the effects of SRM on erdafitinib efficacy, we first generated stable SRM KO cells in FGFR-mutant bladder cancer cell lines (MGH-U3 and SW780) with specific sgRNAs (Fig. 2A). SRM KO significantly sensitized MGH-U3 and SW780 cells to erdafitinib treatment, which was confirmed by IC₅₀ values (Fig. 2B; Supplementary Fig. S2A). Notably, erdafitinib treatment at a dose of 10 nmol/L showed no effect on bladder cancer cells but yielded a strong synergistic effect in reducing cell viability in SRM KO cells (Fig. 2C and D; Supplementary Fig. S2B and S2C). To gain further insights into the function of SRM in vivo, we evaluated the antitumoral effect of erdafitinib using a xenograft mouse model of WT and SRM KO MGH-U3 cells (Fig. 2E). The combination of SRM deletion and erdafitinib-targeted therapy elicited a complete inhibition of tumor growth, whereas SRM KO alone showed little effect (Fig. 2F and G). IHC analyses also revealed a profound decrease in the proliferation marker (Ki67) in mice with SRM KO MGH-U3 xenograft after erdafitinib treatment (Supplementary Fig. S2D). Taken together, these integrated in vitro and in vivo experiments confirmed that SRM KO significantly augmented erdafitinib efficacy in FGFR-mutant bladder cancer.

Having established the essential SL role for SRM in erdafitinib treatment, we next search for the mediator responsible for the enhanced erdafitinib efficacy. To confirm the polyamine metabolites levels, we performed HPLC on WT and SRM KO MGH-U3 cells. In line with previous studies (14), we observed a decrease in spermidine levels in SRM KO cells, accompanied by an increase in the upstream metabolite putrescine and downstream metabolite spermine (Fig. 3A; Supplementary Fig. S3A). Additionally, ELISA analysis of MGH-U3 and SW780 cells confirmed a decrease in spermidine levels and an increase in putrescine levels (Supplementary Fig. S3B). Similar effect was observed when bladder cancer cells were exposed to MCHA, an inhibitor of SRM (Fig. 3B; Supplementary Fig. S3C; refs. 23, 24). It is well established that spermidine posttranslationally modifies the translation factor eIF5A, which is essential for the efficient translation of select mRNA subsets (16). As expected, SRM KO significantly decreased eIF5A^Hyp levels (Fig. 3C; Supplementary Fig. S3D). Hence, we hypothesized that SRM deletion augmented erdafitinib efficacy via reduced translation of specific proteins induced by eIF5A^Hyp. Then quantitative MS was performed on WT and SRM KO MGH-U3 cells treated with either erdafitinib or DMSO (Fig. 3D). We found 239 downregulated proteins in KO versus WT groups and 324 downregulated proteins in KO + erdafitinib versus WT + erdafitinib groups (Supplementary Table S4). Among these, 93 overlaid proteins were identified, and 20 of them exhibited high abundance in MGH-U3 WT cells (Fig. 3E). Then, we confirmed the protein and mRNA levels of these 20 genes and found that only C1orf116 and EGFR mRNA levels were reduced upon SRM depletion, suggesting a reduction in translation of other proteins (Fig. 3F; Supplementary Fig. S3E). Furthermore, RIP experiments revealed that eIF5A binds to HMGA2 mRNAs but not to others (Fig. 3G; Supplementary Fig. S3F). These data suggested that SRM and eIF5A might regulate the translation of HGMA2.

To confirm whether the effects of SRM on erdafitinib efficacy were mediated via HMGA2, we overexpressed HMGA2 in SRM KO bladder cancer cells (Fig. 3H; Supplementary Fig. S3G). The enhancement of erdafitinib efficacy via SRM KO could be reversed by overexpression of HMGA2 (Fig. 3I and J; Supplementary Fig. S3H–S3I). In line with the in vitro results, HMGA2 overexpression abolished the effects of SRM KO to erdafitinib treatment in FGFR-mutant bladder cancer, indicated by increased tumor volume (Fig. 3K and L) and tumoral Ki67 expression (Fig. 3M). To sum, these data suggested that SRM depletion enhances erdafitinib-targeted therapy efficacy probably via reduced translation of HMGA2.

We further address how SRM and eIF5A modulate the translation of HMGA2. SRM KO cells were supplemented with spermidine, and the reduction of eIF5A^Hyp and HMGA2 protein levels caused by SRM KO was rescued (Fig. 4A; Supplementary Fig. S4A). Then we treated bladder cancer cells with MCHA or the DHPS inhibitor N1-guanyl-1,7-diaminoheptane. Both treatments significantly reduced eIF5A^Hyp and HMGA2 protein levels but had little effect on HMGA2 mRNA levels (Fig. 4B and C; Supplementary Fig. S4B and S4C). To directly visualize the translation of HMGA2 mRNA, an OPP-mediated pulldown assay was performed to detect nascent synthesized HMGA2 protein (Fig. 4D). Indeed, SRM KO and MCHA treatment significantly reduced newly synthesized HMGA2 proteins (Fig. 4E; Supplementary Fig. S4D). Furthermore, we performed polyribosome profiling on WT and SRM KO bladder cancer cell lines, and RNA was isolated from polyribosome profiling sedimentation fractions and reversed-transcribed for RT-PCR analysis. Deletion of SRM did not affect the levels of HMGA2 transcript; however, there was a leftward shift in HMGA2 transcript in polyribosome profiles compared with WT MGH-U3 cells (Fig. 4F; Supplementary Fig. S4E), indicating a translation initiation block in SRM KO cells.

eIF5A has been previously reported to alleviate translational stalling of the ribosome at specific amino acid sequences (16). Based on these, we searched the sequence of HMGA2 protein and identified four potential ribosome-pausing motifs: Arginine-Proline-Arginine (RPR) at positions 31 to 33, 47 to 49, 75 to 77, and 79 to 81. Then we generated five different constructs and evaluated the sensitivity of these constructs to SRM depletion by monitoring HMGA2-mCherry translation (Fig. 4G). Intriguingly, these mutations partially rescued the expression of HMGA2 in the absence of SRM and eIF5a^Hyp (Fig. 4H; Supplementary Fig. S4F), suggesting that the RPR motif may confer HMGA2 with the specific requirement to rely on eIF5A^Hyp for smooth translation. These data indicate that eIF5A^Hyp facilitates the synthesis of HMGA2 via the RPR motif.

Functioning as an architectural transcription factor, HMGA2 directly binds to DNA sequences, thereby modifying DNA structure and regulating the transcription of target genes (25, 26). In our LC/MS datasets, protein-protein interaction (PPI) network analysis revealed that EGFR is a hub gene and that HMGA2 is highly connected to EGFR (Supplementary Fig. S5A). We have demonstrated that SRM regulates HMGA2 translation via eIF5A^Hyp. To delve deeper into the underlying molecular mechanism, transcriptome analysis was conducted on WT and SRM KO MGH-U3 cells treated with erdafitinib (Fig. 5A; Supplementary Fig. S5B). As expected, SRM KO significantly reduced the mRNA levels of EGFR (Fig. 5B; Supplementary Fig. S5C). Subsequent analysis of publicly available ChIP-seq data indicated that HMGA2 may bind to the EGFR promoter (20). ChIP experiments revealed that HMGA2 binds to the EGFR promoter (Fig. 5C and D; Supplementary Fig. S5D and S5E). Moreover, SRM KO could reduce the luciferase activity of the EGFR promoter (Fig. 5E; Supplementary Fig. S5F). Previous studies have reported that the EGFR–PI3K–AKT signaling pathway is an important adaptive feedback mechanism of FGFR inhibitors (27, 28). Herein, SRM KO significantly reduced pAKT levels in the presence of erdafitinib, whereas SRM KO alone had no effect on pAKT levels, likely due to an adaptive feedback mechanism (Fig. 5F; Supplementary Fig. S5G). Of note, whereas 100 nmol/L erdafitinib reduced pAKT levels and 10 nmol/L erdafitinib showed no effect on WT cells, 10 nmol/L erdafitinib induced a remarkable reduction in p-AKT levels in SRM KO cells (Fig. 5G; Supplementary Fig. S5H). These data indicated that combination treatment achieves more inhibition of AKT phosphorylation compared with each individual treatment alone. Besides, the knockdown of HMGA2 markedly decreased EGFR expression (Fig. 5H; Supplementary Fig. S5I), and overexpression of HMGA2 could reverse the reduction of EGFR and pAKT levels upon SRM depletion (Fig. 5I; Supplementary Fig. S5J). Thereby, SRM supports the efficient translation of HMGA2, which subsequently promotes the expression of EGFR.

Kaplan–Meier survival analysis revealed that high expression of HMGA2 and EGFR was remarkably associated with poor prognosis in patients with bladder cancer (Supplementary Fig. S6A and S6B). Besides, correlation analysis indicated that SRM and HMGA2 are positively correlated with the expression of EGFR mRNAs as SRM and HMGA2 regulate the transcription of EGFR. Besides, SRM does not correlate with HMGA2 mRNA levels, which aligns with the role of SRM in regulating the translation of HMGA2 rather than its transcription (Supplementary Fig. S6C–S6E). To further confirm the regulatory role of SRM on erdafitinib efficacy via HMGA2, we overexpressed SRM in SRM KO MGH-U3 cells and found that the levels of eIF5A^Hyp, HMGA2, and EGFR can be restored (Supplementary Fig. S6F). Besides, the enhancement of erdafitinib efficacy via SRM KO could be reversed by overexpression of SRM (Supplementary Fig. S6G). Similar results were obtained in SW780 cells (Supplementary Fig. S6H and S6I). Furthermore, when we supplemented SRM KO MGH-U3 cells with spermidine, EGFR signaling was rescued (Supplementary Fig. S6J) and enhanced erdafitinib efficacy upon SRM KO was reversed (Supplementary Fig. S6K). Comparable effects were also observed in SW780 cells (Supplementary Fig. S6L and S6M). After all, SRM could regulate the expression of EGFR and erdafitinib efficacy via HMGA2.

To explore potential therapeutic strategies for FGFR-mutant bladder cancer, we treated WT and SRM KO MGH-U3 or SW780 cells with erdafitinib alone or in combination with MCHA. MCHA treatment significantly sensitized bladder cancer cells to erdafitinib (Fig. 6A and B; Supplementary Fig. S7A). Similar effects were observed when we exposed RT112 and RT4 bladder cancer cells to erdafitinib alone or in combination with MCHA (Supplementary Fig. S7B and S7C). Furthermore, because MCHA has not been tested clinically, we evaluated the therapeutic potential of combining EGFR inhibitor with erdafitinib for the treatment of patients with FGFR-mutant bladder cancer. We observed a strong synergistic effect in reducing cell viability upon combination treatment of EGFR and FGFR inhibitors (Fig. 6C and D; Supplementary Fig. S7D). Moreover, we evaluate the toxicity of the combination of MCHA and erdafitinib both in vitro and in vivo. The combined therapy of MCHA and erdafitinib did not attenuate the cell growth of normal bladder cell line SV-HUC-1 (Supplementary Fig. S7E). Besides, we evaluated the biochemical indicators of cardiac, hepatic, and renal function of mice. First, hepatic function markers, such as ALT and AST, increased during the combination treatment but remained within the normal range. Second, BUN levels increased slightly, whereas CREA levels remained unchanged and UA levels decreased. Third, only LDH1 exhibited an increase, whereas CK, CK-MB, and LDH levels remained unchanged. Lastly, no significant changes in body weight were observed. Collectively, these findings suggest that combination therapy exhibits acceptable toxicity (Supplementary Fig. S7F; Supplementary Table S5). The combination of MCHA and erdafitinib also elicited a complete inhibition of tumor growth in mice xenografted with human MGH-U3 cells, whereas MCHA treatment alone showed little effect (Fig. 6E and F). Similar effect was observed when we exposed bladder cancer cells to afatinib and erdafitinib (Fig. 6G and H). Additionally, IHC analysis not only demonstrated a significant reduction of Ki67 after combination treatment (Fig. 6I and J) but also showed a decrease in EGFR and pAKT levels upon combination treatment of MCHA and erdafitinib (Supplementary Fig. S7G and S7H).

In general, SRM deletion could attenuate the translation of HMGA2, which subsequently reduced EGFR transcription, thus enhancing the sensitivity of FGFR-mutant bladder cancer cells to erdafitinib-targeted treatment. Furthermore, treatment with MCHA demonstrated synergistic lethality when combined with erdafitinib-targeted therapy in FGFR-mutant bladder cancer cells. These findings suggest that the SRM inhibitor MCHA could be potentially utilized for the therapeutic intervention of FGFR-mutant bladder cancer.

Despite the impressive clinical benefits from erdafitinib-targeted therapy in FGFR-mutant bladder cancer, therapeutic resistance inevitably develops after a period of disease stabilization or regression. Therefore, clarifying the molecular events underlying erdafitinib resistance is crucial for developing new therapeutic strategies to maximize its clinical benefit. Notably, breakthroughs in genome-editing technologies and high-throughput screening have renewed interest in synthetic lethality, unveiling novel strategies to overcome therapeutic resistance in oncology (29). In this study, we used a whole-genome CRISPR SL screen on FGFR-mutant bladder cancer cell lines and identified SRM as target whose deletion enhanced erdafitinib efficacy. As a result, genetic and pharmacologic inhibition of SRM, in combination with erdafitinib, synergistically inhibited bladder cancer growth both in vitro and in vivo. Furthermore, we found that SRM promoted the translation of HMGA2 via eIF5a hypusination, thereby activating EGFR transcription and reducing bladder cancer sensitivity to erdafitinib. These data supported the exploration of the combination of polyamine metabolism inhibition and erdafitinib treatment as a strategy to extend benefit to patients with FGFR-mutant bladder cancer.

Whereas the loss of growth control in cancer cells predisposes transformed cells to be more sensitive to polyamine depletion than normal cells (11), no more than 15% of the spermidine normally present in cancer cells is required for cell proliferation (30). In our model, both SRM KO and low-dose MCHA treatment did not result in an 80% inhibition of spermidine, and indeed, they failed to induce considerable cell death in bladder cancer cells. Moreover, polyamine metabolism inhibitors have been found to inhibit cancer development in several rodent models (12, 31). However, their effectiveness as a therapeutic agent has been modest with treatment-limiting toxicity at high doses (11, 31). Intriguingly, the combination therapy of low-dose MCHA and erdafitinib significantly reduced the growth of FGFR-mutant bladder cancer cells. These findings suggested a previously unappreciated role for polyamine metabolism as a SL target of erdafitinib and highlighted a novel drug combination for treating FGFR-mutant bladder cancer. Additionally, previous studies have reported a significant 81% risk reduction of metachronous adenoma for patients in a relatively low dietary polyamine intake quartile when combined with polyamine metabolism inhibitors (32). Therefore, further research is required to evaluate the potential for clinical application of combining dietary polyamine control, polyamine metabolism inhibition, and erdafitinib-targeted therapy.

In our LC/MS dataset, SRM KO resulted in 656 and 938 differentially expressed proteins when alone or combined with erdafitinib treatment, respectively. Although there were some similarities, most of these differentially expressed proteins did not overlap. Besides, the majority of these proteins have not been identified in previous studies upon eIF5a^Hyp inhibition. Based on the observations, we speculate that environmental context and cell system may impact protein translation regulated by polyamine metabolism. Furthermore, eIF5a^Hyp is thought to be particularly important for elongation through certain motifs in the mRNAs. In our study, although SRM KO significantly decreased eIF5a^Hyp levels, it still led to the upregulation of 244 proteins, indicating the involvement of other protein expression regulation mechanisms not directly dependent of eIF5A^Hyp. Indeed, spermidine could directly bind to its target molecules and impact their functions (33). Another study also revealed a cross-talk between eIF5A and epigenetic remodeling (34). These results have supported the complicated roles of polyamine metabolism in regulation of multiple biological processes. Moreover, we demonstrated that HGMA2 is a direct downstream effector of eIF5A^Hyp-mediated protein translation control, which subsequently served as a transcriptional factor activating its target genes. Thereby, our results not only add to the knowledge of protein expression control of eIF5A, but also shed light into the mechanism by which polyamine metabolism induces erdafitinib resistance.

Although erdafitinib has shown promising effects against FGFR-mutant bladder cancer, the rapid development of resistance and the limited dosing due to toxicity has greatly restricted its clinical efficacy (35). Recently, a functional genetic screen using a short hairpin RNA library identified PI3K as a SL target of FGFR inhibitor (36), and combination of erdafitinib and PI3K inhibitors showed synergistic effects in bladder cancer (28). Herein, we used a genome-wide CRISPR screen and identified SRM as a SL gene, providing a promising therapeutic strategy for patients with erdafitinib-resistant bladder cancer. Although SRM selective inhibitors have not yet undergone clinical trials, our study provides preclinical evidence that SRM-specific inhibitors have strong apparent potential for use as efficient antitumor drugs through synergistic enhancement of the antitumor effects of erdafitinib. Additionally, although EGFR-dependent adaptive feedback activation has been reported as an important mechanism limiting the efficacy of FGFR inhibitors, our results indicated that 96 hours of erdafitinib treatment at a high dose of 100 nmol/L did not activate the EGFR signaling pathway (Fig. 5F; Supplementary Fig. S5G). Previously, 1 μmol/L erdafitinib treatment also failed to stimulate EGFR signaling, further confirming the reliability of our results (28). Despite this, our study demonstrated that the EGFR inhibitor afatinib induces strong antitumor effects when combined with erdafitinib, consistent with previous studies (27, 28). Intriguingly, SRM was confirmed to regulate the expression of EGFR via HGMA2, which may explain why SRM deletion is SL with erdafitinib. More importantly, further clinical trials are warranted to evaluate the combination therapy of SRM or EGFR inhibitors and erdafitinib.

In summary, the insights gained from this study enhance our understanding of erdafitinib resistance in FGFR-mutant bladder cancer. The exploration of the SL target that enhances erdafitinib efficacy provides future potential combination strategies. After all, combining SRM inhibitor MCHA with erdafitinib holds promise for overcoming erdafitinib-targeted therapy resistance for patients with FGFR-mutant bladder cancer.

No disclosures were reported.

Y. Yu: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. X. Gao: Software. H. Zhao: Formal analysis. J. Sun: Supervision. M. Wang: Software. X. Xiong: Methodology. J. Li: Formal analysis. C. Huang: Funding acquisition. H. Zhang: Conceptualization, supervision, project administration, writing–review and editing. G. Jiang: Supervision, funding acquisition, validation, project administration. X. Xiao: Supervision, funding acquisition, validation, writing–original draft, project administration, writing–review and editing.

The authors wish to express their sincere gratitude to the patients who generously participated in this study. The authors express their appreciation to Union Hospital of Tongji Medical College of Huazhong University of Science and Technology (Wuhan P.R. China) for their assistance during this study. This work was supported by the National Natural Science Foundation of China (nos. 82072840, 82373053, 82102734, 82473179, and 82473301).