ARF Confers a Context-Dependent Response to Chemotherapy in Muscle-Invasive Bladder Cancer

Bladder cancer is one of the most commonly diagnosed cancers worldwide, particularly among men (1). While the majority of bladder cancers are non–muscle-invasive tumors that generally have favorable survival outcomes, muscle-invasive bladder cancer (MIBC) has considerably worse prognosis (2, 3). Indeed, despite significant improvements in disease outcomes and treatments for many other cancers, the expected survival for MIBC has not significantly improved in recent years and there are few treatment options, which have relatively poor response (2, 3). The standard-of-care for organ-confined MIBC is cystectomy (i.e., surgical removal of the bladder) with neoadjuvant chemotherapy, which has a 5-year survival of approximately 50%. For patients with metastatic disease, cystectomy is not an option, and the standard-of-care is platinum-based chemotherapy regimens, which provide relatively modest 2-year survival rates (∼15%). A key mechanism for differential response to cisplatin chemotherapy is alterations in genes associated with DNA repair (4, 5). Recent studies have shown promising activity with immune checkpoint blockade in patients with recurrent MIBC (6), particularly in a postchemotherapy setting (7), which has led to recent FDA approval. Therefore, it is imperative to understand the cellular and tumor contexts that account for variable response to chemotherapy.

The Cancer Genome Atlas (TCGA; ref. 8) has identified a spectrum of genomic alterations that are prevalent in MIBC, including a cadre of potential actionable drivers. These and other studies have provided new insights into molecular subtypes of bladder cancer (e.g., refs. 8–12), as well as potential mechanisms for differential drug response (e.g., refs. 4, 5, 13, 14). Notably, these studies have shown that CDKN2A is one of the most frequently altered genes in MIBC (8), while alterations of CDKN2A have been associated with adverse disease outcomes (15). The CDKN2A tumor suppressor locus includes several overlapping transcripts, which encode proteins that regulate the p53 and RB tumor suppressor pathways in response to oncogenic stress (16, 17). One of these transcripts, namely p14^ARF (hereafter referred to as ARF), is a key regulator of the p53 signaling pathway (16, 17). Unexpectedly, we now find that in MIBC, elevated expression of ARF in the nucleolus attenuates response to cisplatin chemotherapy. These findings uncover an unanticipated role for ARF in drug resistance and highlight the importance of understanding tumor context for predicting response to chemotherapy.

Analysis of patient data was performed following protocols approved by the Institutional Review Board of Memorial Sloan Kettering Cancer Center (MSKCC, New York, NY). Tissue microarrays were generated from patients that had undergone radical cystectomy at MSKCC during the period of 2001 to 2013. Clinical follow-up data were obtained for recurrence/metastasis, overall, and disease-specific survival and treatment (neoadjuvant or adjuvant). A subset of the cases represented on the tissue microarrays (TMA) had accompanying genomic sequencing data as reported (8, 15, 18), and a subset of those had undergone treatment with chemotherapy. A description of the clinical characteristics of tumors represented on the TMAs is provided in Table 1. Sections (5 μm) of the TMA blocks were analyzed for H&E and immunohistochemical staining. Each core was scored by estimating the percentage of tumor cells with immunoreactivity, as well as the intensity of staining defined by a 4-tier grade (undetectable = 0, weak = 1, intermediate = 2 strong = 3) and staining localization (cytoplasmic/nuclear/nucleolar). Each case was represented by 3 cores, and the mean scores for the 3 cores were recorded as a single score for each case.

This study used the following publicly available human datasets: (i) TCGA datasets: (a) “TCGA, Nature 2014” includes 131 treatment-naïve MIBC samples and 23 tumor-adjacent histologically normal bladder. Of these, 127 MIBC samples have whole-exome and RNA sequencing data, including clinical outcome data (8); (b) “TCGA provisional” (TCGA 400) contains 413 samples from 412 MIBC patients including 408 samples with RNA sequencing data; (ii) the Sanchez-Carbayo and colleagues dataset consists of expression profiles from 72 MIBC, plus 33 superficial (non–muscle invasive) cancers, and 52 normal bladder, arrayed on an Affymetrix U133A platform (19); (iii) the Rebouissou and colleagues (EMBL-EBI: E-MTAB-1803) dataset consists of expression profiles from 85 cases of MIBC (21 basal and 64 non-basal) arrayed on an Affymetrix U133 Plus 2.0 platform (11); (iv) for molecular subtyping of basal and luminal subtypes, we used gene signatures from the 47-gene predictor “BASE47” (10).

All experiments using animals were performed according to protocols approved by the Institutional Animal Care and Use Committee (IACUC) at Columbia University Medical Center (New York, NY). The p53^f/f; Pten^f/f mice were described previously (20). The conditional p19^Arf allele (Arf^f/f; C57Bl6; 129SJv) has loxP sites flanking Exon1b of Cdkn2a, thereby resulting in deletion of Arf, but not other Cdkn2a transcripts (21). The R26R-Cre^ERT2 allele expresses a tamoxifen-inducible Cre (CreERT2) under the control of the R26R promoter (22), and the R26R-YFP allele (R26R^YFP/+; C57Bl6; 129SJv) has YFP expressed under the control of the R26R promoter (23). Tumor induction was achieved by injection of an adenovirus expressing Cre-recombinase (Adeno-Cre) into the bladder lumen as described previously (20). Ultrasound imaging was performed using a Vevo 2100 Imaging System (Visual Sonics). For drug treatment, cisplatin (Sigma-Aldrich; catalog #479306) was dissolved in PBS and freshly made solution was administered once weekly intraperitoneally at 2 or 4 mg/kg, as indicated in Supplementary Table S1. All mice were monitored for body condition (i.e., muscle tone and weight) and were sacrificed when their body condition score was <1.5, as per guidelines of the IACUC.

At the time of sacrifice, bladder and other relevant tissues were fixed in 10% formalin for histologic and immunostaining analyses, or snap-frozen in liquid nitrogen for Western blot analysis or isolation of RNA. Hematoxylin and eosin (H&E) and immunostaining were done on 3-μm paraffin sections as described previously (20). H&E and immunostained images were captured on an Olympus VS120 whole-slide scanning microscope. Immunofluorescence staining was visualized using a Leica TCS SP5 confocal microscope. Platinum–DNA adducts (Pt-GG) were detected by immunofluorescence after DNA denaturation and protein digestion as described previously (24). Quantification of immunostaining was done using five independent sections from 3 independent mice. Western blot assays were done using total protein lysates extracted in RIPA buffer as described previously (25). Real-time PCR analysis was performed with total RNA using a Quantitech SYBR Green PCR (Qiagen) and quantified as described previously (25). Details of antibodies are provided in Supplementary Table S2; sequences of all primers are provided in Supplementary Table S3.

Gene expression profiling analysis was done using bladder epithelium from wild-type mice (Arf^+/+; p53^+/+; Pten^+/+) or bladder tumors from Arf wild-type (Arf^+/+; p53^f/f; Pten^f/f) or Arf-null (Arf^f/f; p53^f/f; Pten^f/f) mice treated with vehicle or cisplatin (n = 4–6/group). RNA was prepared using MagMAX-96 Total RNA Isolation Kit (Life Technologies), and RNA sequencing was done at the JP Sulzberger Columbia Genome Center at Columbia University Medical Center (New York, NY). A TruSeq RNA Sample Prep Kit v2 (Illumina) was used for library preparation followed by sequencing (30 million reads, single end) on an Illumina HiSeq 2500. Paired-end reads were mapped to mouse genome build M4 using STAR aligner. Raw gene counts were determined using featureCounts then normalized and variance-stabilized using DESeq2 package (Bioconductor) in R-system v3.1.1 (The R Foundation for Statistical Computing, ISBN 3-900051-07-0). The raw and normalized data files are deposited in Gene Expression Omnibus (GEO) with series entry number: GSE89823.

Differentially expressed genes were ranked using t test statistics (Dataset 1). Differentially deregulated biological pathways were identified by gene set enrichment analysis (GSEA) using the Broad Institute GSEA software package (26) with pathways collected in the c2-curated gene sets including KEGG BIOCARTA, and REACTOME biological pathways (http://www.broadinstitute.org/gsea/msigdb/index.jsp), and with the Base47 gene sets (Dataset 2; ref. 10). Mouse genes were mapped to human homologs using biomaRt (27). Statistical significance of enrichment between mouse query signatures and human target pathways was computed with 1,000 gene permutations.

The UMUC3 (CRL-1749) human bladder cancer cells used in this study were obtained directly from ATCC, which provided a certificate of analysis. Cells were purchased in 2010; all studies were done with passage 2 cells. A p14^ARF cDNA was subcloned into the pMXs IRES-RFP (pMXs-IR) vector, which also expresses red fluorescent protein (RFP) to enrich for infected cells by FACS cell sorting. Gene knockdown studies were done using pGIPZ lentiviruses (Dharmacon, GE Healthcare Life Sciences), which express an shRNAmir (miRNA-adapted shRNA) for ITGB3BP (shITGB3BP) as listed in Supplementary Table S3. Two independent shITGB3BP were used; shown are representative data using one. Lentiviruses were generated using second-generation packaging vectors, psPAX2 and pMD2.G (Addgene) in HEK-293T cells (ATCC) and concentrated using the Lenti-X Concentrator reagent (Clonetech) according to the manufacturer's instructions.

To evaluate tumorigenicity and drug response of human bladder cancer cells in vivo, we adapted an orthotopic assay in which cells are implanted into the submucosal layer of the bladder wall of adult male Ncr/Nude mice (6–8 weeks old; Taconic) using ultrasound guidance (28). Briefly, under anesthesia, the lamina propria of the bladder was delaminated from the detrusor muscle by ultrasound-guided delivery of PBS using a 1 mL syringe with a 30 G 0.5 inch needle, followed by delivery of a cell suspension in Matrigel (4 × 10⁴ cells in 20 μl) into the resulting submucosal pocket. Tumors were monitored by ultrasound imaging as above. Mice were treated with cisplatin for 3 weeks (8 mg/kg), and tumors were collected and analyzed as above.

Analysis of protein translation was done using mouse embryonic fibroblasts (MEF) isolated from 13.5-day embryos from Arf wild-type or Arf-null mice having an inducible Cre allele and a YFP reporter both under the control of the ROSA26 locus (Arf^+/+; p53^f/f; Pten^f/f; R26R^CreERT2/YFP or Arf^f/f; p53^f/f; Pten^f/f; R26R^CreERT2/YFP, respectively). MEFs were treated in culture with 0.1 nmol/L 4-hydroxytamoxifen (4-OHT, Sigma-Aldrich) for 24 hours to induce gene recombination. Analysis of protein translation were performed as described previously (29), using the tamoxifen-induced Arf wild-type or Arf-null MEFs treated with 10 μmol/L cisplatin or vehicle (PBS) for 24 hours. For analysis of polysomes, cells (1 × 10⁷) were washed with PBS containing 100 μg/mL cycloheximide (Sigma), and lysed with hypotonic buffer (5 mmol/L Tris-HCl, pH 7.5, 2.5 mmol/L MgCl₂, 1.5 mmol/L KCl, 1× cOmplete protease inhibitor cocktail (EDTA-free, Roche) supplemented with 100 μg/mL cycloheximide, 2 mmol/L dithiothreitol, 100 U/mL RNasin (Promega), 0.5% Triton X-100, and 0.5% sodium deoxycholate, and centrifuged for 7 minutes at 16,000 × g (Eppendorf centrifuge). Equivalent amounts of samples were loaded onto a 10%–50% sucrose gradient in 20 mmol/L HEPES-KOH, pH 7.5, 100 mmol/L KCl and 5 mmol/L MgCl₂, 100 μg/mL cycloheximide, 1 mg/mL heparin (Sigma) and centrifuged at 12,000 × g for 2 hours in an SW40 rotor (Beckman L7-55 Ultracentrifuge). Fractions (0.5 mL) were collected manually, UV absorbance (254 nm) was measured, and data plotted using Prism software (version 5.0c).

For analyses of ³⁵S labeling, Arf wild-type or Arf-null MEFs (1 × 10⁴ per 35-mm dish) were treated with 10 μmol/L cisplatin or PBS for 24 hours. Media was replaced with methionine- and cysteine-free medium (Thermo Fisher Scientific) supplemented with dialyzed FBS (Thermo Fisher Scientific). After 30 minutes, 33 μCi of ³⁵S-methionine (Perkin-Elmer) was added and incubated for 1 hour. Cells were lysed in in RIPA buffer supplemented with cOmplete protease inhibitor cocktail (Roche), and phenylmethylsulfonyl fluoride (PMSF, Sigma; as above). Samples (15 μg) were resolved by SDS-PAGE and visualized by autoradiography or phosphor imaging (Amersham) using a Typhoon 9500 imager analysis (GE Healthcare). Incorporation of ³⁵S labeling was quantified by liquid scintillation counting (Tri-Carb, Perkin-Elmer).

Statistical analyses were done using the Mann–Whitney U test, Student t test, two-way ANOVA test, or Spearman correlation, as appropriate. For comparing quantitative endpoints between two experimental groups, we applied the Mann–Whitney U test (also known as the Wilcoxon rank-sum test). For evaluating the association between two dichotomous variables (such as loss of function status of two genes), we applied the Student t test. For evaluating the influence of two different categorical independent variables on one continuous dependent variable (as for volumetric analysis between treatment groups), we applied the two-way ANOVA test. For evaluating the correlation of expression of two independent groups, we used Spearman correlation. For survival analysis, we computed the Kaplan–Meier curves for two groups, above-median and below-median, based on the expression of the indicated gene, and used the log-rank test to compare survival rates between these two groups. Prism version 5.0c (GraphPad Software) was used for Student t test, Mann–Whitney U test, two-way ANOVA test, and Spearman correlation; and R v3.1.1′s survival package for survival analysis.

Consistent with the causative role of dysregulated p53 in bladder cancer (9, 30), the most prevalent genomic alterations in MIBC are deletions or missense mutations of TP53, which occur in more than 50% of the treatment-naïve primary tumors represented in the TCGA (n = 67/127; Fig. 1A; ref. 8). Among other prevalent alterations, CDKN2A is deleted or mutated in 42% of the cases in TCGA (n = 53/127; Fig. 1A; ref. 8). Considering its relevance as a tumor suppressor gene that regulates p53 activity, the prevalence of CDKN2A alterations further underscores the importance of p53 in MIBC. However, a substantial number of MIBCs have altered TP53 without corresponding alterations of CDKN2A (35%; n = 45/127), with a tendency towards their mutual exclusivity (P = 0.024, log OR = −0.782; ref. 8); this raises a question regarding the status of CDKN2A in contexts in which it is not deleted or mutated. Indeed, we find that CDNK2A (P = 0.002, Mann–Whitney U test; n = 48 normal, 28 papillary, and 81 MIBC; ref. 19), and particularly ARF (P < 0.0001, Mann–Whitney U test; n = 19 normal, 275 MIBC; ref. 8) are significantly overexpressed in MIBC (Fig. 1B and C). This upregulation is over-represented in cases that have mutation or deletion of TP53 (P = 0.006, Mann–Whitney U test; n = 176 MIBC with wild-type p53 and 99 MIBC with mutant p53; Fig. 1C; ref. 8) and in the more aggressive “basal” MIBC subtype (P = 0.008, Mann–Whitney U test; n = 21 basal and 64 non-basal; Fig. 1D; ref. 11). While ARF expression is known to be elevated in cancer as a consequence of p53 loss-of-function (16, 17), at least one study has found that CDKN2A is amplified in bladder cancer, which is associated with poor survival (31). Therefore, we considered the possibility that ARF overexpression might play a causal role in MIBC.

Specifically, we investigated the expression of ARF protein using TMAs comprised of a range of MIBC cases from patients treated at Memorial Sloan Kettering Cancer Center (n = 190; Table 1). Notably, ARF was often robustly expressed (16%; n = 29/179) and localized to the nucleolus (Fig. 1E). To understand the relevance of this expression, we focused on the subset of the cases on the TMAs that had accompanying genomic sequencing data (n = 35; Table 1), such that we could compare ARF protein expression with the status of TP53. Notably, Kaplan–Meier analysis of the relationship of ARF nucleolar expression in p53-mutated MIBC cases with recurrence-free survival revealed that patients expressing ARF in the nucleolus had a significantly worse outcome than those lacking ARF in the nucleolus (P = 0.009, log-rank test; Fig. 1F). Furthermore, among the subset of MIBC within this group that had received treatment (n = 15; Table 1), those with ARF nucleolar expression had a greater tendency to fail chemotherapy (86%, n = 6/7) than those lacking ARF expression (50%, n = 4/8; Fig. 1G). These findings prompted us to investigate whether dysregulation of ARF results in adverse outcome or impaired treatment response in MIBC.

To determine whether the status of Arf expression altered cisplatin sensitivity in MIBC, we used a genetically engineered mouse model (GEMM) based on combinatorial loss-of-function of p53 and Pten in bladder urothelium (20). Notably, p53^f/f; Pten^f/f mouse bladder tumors display robust expression of Arf at both the protein and transcript levels (>10-fold, P < 0.001 Mann–Whitney U test, n = 8/group), and have accumulation of Arf protein in the nucleolus (Supplementary Fig. S1A and S1B), as we had observed in human MIBC (see Fig. 1E). Furthermore, GSEA comparing an expression signature from these mouse bladder tumors with signatures representing “basal” or “luminal” subtypes of human bladder cancer (10) revealed a significant positive enrichment (NES 2.23; P < 0.001) with the “basal-like” signature and a significant negative enrichment (NES −2.38; P < 0.001) with the “luminal-like” signature (Supplementary Fig. S1C), consistent with the observed greater expression of CDKN2A transcripts in human “basal-like” bladder cancer (see Fig. 1D). On the basis of these observations, we reasoned that this p53^f/f; Pten^f/f GEMM would be informative for studying the functional role of Arf in MIBC in vivo.

Toward this end, we generated GEM mice having a floxed Arf allele (21) in the context of the p53^f/f; Pten^f/fmice (Arf^f/f; p53^f/f; Pten^f/f) such that Arf (but not other CDKN2A transcripts, see ref. 21) could be conditionally deleted in the bladder epithelium together with Pten and p53. We then studied the phenotype and molecular features of bladder tumors arising in Arf wild-type (Arf^+/+; p53^f/f; Pten^f/f) versus Arf-null (Arf^f/f; p53^f/f; Pten^f/f) mice (Supplementary Table S1A). We found that Arf-null bladder tumors displayed a similar phenotype as their Arf wild-type counterparts, albeit with the expected absence of Arf expression (Supplementary Fig. S1A and S1B). Consistent with the known tumor suppressor function of Arf (16, 17), tumor-bearing Arf-null mice displayed a modest, but significant, reduction in tumor latency compared with Arf wild-type mice (P = 0.006, log-rank test; n = 30 Arf wild-type and 37 Arf-null mice; Supplementary Fig. S1D). Furthermore, analysis of their gene expression profiles revealed that the Arf-null and Arf wild-type bladder tumors were overall similar in terms of differentially expressed genes and biological pathways (Datasets 1 and 2). In particular, GSEA comparing an expression signature from Arf-null bladder tumors with signatures representing “basal” or “luminal” subtypes of human bladder cancer (10) revealed a significant positive enrichment (NES 1.90; P < 0.001) with the “basal-like” signature and a significant negative enrichment (NES −2.45; P < 0.001) with the “luminal-like” signature (Supplementary Fig. S1E), as was the case for the Arf wild-type MIBC tumors (Supplementary Fig. S1C).

However, when challenged with cisplatin, a platinum-based chemotherapy that is frequently is used in treatment of human MIBC (2, 3), profound differences in Arf wild-type and Arf-null tumors were unmasked. In particular, Arf wild-type tumors were markedly resistant to treatment with cisplatin compared with the Arf-null tumors (n = 9–15/group; Fig. 2; Supplementary Table S1B). This differential treatment response was evident by assessment of qualitative endpoints, such as tumor morphology (Fig. 2A), as well as several quantitative endpoints, including tumor volume (P = 0.016, two-way ANOVA test; Fig. 2B), tumor weight (P = 0.003, Mann–Whitney U test; Fig. 2C), and cell proliferation (P = 0.004, Mann–Whitney U test; Fig. 2D). Notably, differential response to treatment was not due to differences in uptake of cisplatin, as the Arf-wild-type and Arf-null tumors had similar levels of immunostaining for Pt-GG (Fig. 2A and E), a marker of intra-strand DNA crosslinks induced by cisplatin treatment (24) and similar expression of γH2AX (Fig. 2A), a marker of DNA damage (32). Taken together, these findings demonstrate that Arf status in MIBC in GEMMs has a profound effect on response to cisplatin.

We next investigated whether ARF also affects response to cisplatin treatment in human bladder cancer using UMUC3 cells, which is a highly tumorigenic cell line that lacks endogenous ARF expression (Fig. 3A and B). In particular, we implanted UMUC3 cells orthotopically into the bladder of immunodeficient host mice using ultrasound-mediated guidance (see Materials and Methods and ref. 28), which results in invasive bladder tumors within 2–3 weeks (n = 10–11/group; Fig. 3A and C). These tumors respond to cisplatin as evident by assessment of qualitative endpoints, such as tumor morphology (Fig. 3C), and quantitative endpoints, including reduced tumor volume (Fig. 3C), tumor weight (P = 0.032, Mann–Whitney U test; Fig. 3D), and cellular proliferation (P = 0.014, Mann–Whitney U test; Fig. 3E).

We expressed exogenous ARF in UMUC3 cells via lentiviral delivery (Fig. 3A and B). The ARF-expressing UMUC3 cells were implanted orthotopically into the bladder of host mice, and tumor growth was monitored following treatment with vehicle or cisplatin. Although the ARF-expressing UMUC3 cells develop muscle-invasive bladder tumors with similar latency, morphology, and histology as the parental UMUC3 cells (which lack ARF), ARF dampened response to cisplatin, which was evident from both quantitative (e.g., tumor weight and proliferation) and qualitative (e.g., morphology) endpoints (n = 11–14/group; Fig. 3C–E). Notably, exogenous ARF was localized to the nucleolus of the ARF-expressing UMUC3 tumors (Fig. 3C), as was the case for endogenous ARF in human MIBC (see Fig. 1E) and in the GEMM of MIBC (see Fig. 2A). These findings demonstrate that the accumulation of ARF in the nucleolus attenuates response to cisplatin in human bladder cancer models in vivo.

To study the mechanisms by which Arf impairs response to cisplatin, we examined expression profiles from Arf wild-type and Arf-null mouse bladder tumors treated with vehicle or cisplatin (Datasets 1 and 2). Among differentially expressed genes, we focused on ITGB3BP, which was the top-most significantly changed gene following cisplatin-treatment of Arf-null versus Arf-wild-type mouse tumors (P = 1.4 × 10⁻⁹; t test; Supplementary Fig. S2A and S2B; Dataset 1). ITGB3BP (a.k.a β3-endonexin, NRIF3, CENPR) is a relatively uncharacterized component of integrin signaling that was discovered (as β3-endonexin) based on its ability to interact with the integrin β3 subunit, presumably to influence intracellular integrin signaling (33, 34). We now find that ITGB3BP expression is upregulated in several solid tumors in human cancer, and in particular in bladder cancer, where its expression is well-correlated with expression of CDKN2A (Spearman correlation, r = 0.2433; P < 0.0001; Supplementary Fig. S2C).

To examine its potential relevance for modulating response to cisplatin in human bladder cancer, we examined the consequences of depleting ITGB3BP for tumor growth and response to cisplatin in the UMUC3 cell model (n = 9–10/group; Fig. 3A and B). Specifically, we infected UMUC3 cells with lentiviruses expressing shRNA that target ITGB3BP (Supplementary Table S3) and monitored the consequences for tumor growth and drug response as above. Strikingly, ITGB3BP-depleted UMUC3 cells phenocopied the consequences of ARF gain-of-function in this orthotopic model of human bladder cancer (Fig. 3C–E). In particular, the vehicle-treated ITGB3BP-depleted UMUC3 cells developed muscle-invasive bladder tumors of similar size, morphology, and histology as the control UMUC3 cells, as well as the ARF-expressing UMUC3 cells. However, these tumors failed to respond to cisplatin treatment, as had been observed for the ARF-expressing UMUC3 cells, which was evident from both quantitative (e.g., tumor weight and proliferation) and qualitative (e.g., morphology) endpoints (Fig. 3C–E). These findings suggest that the effects of ARF on response to cisplatin treatment are mediated, in part, through the differential expression of ITGB3BP, and highlight a novel role for this integrin pathway gene in resistance to cisplatin.

The prominent expression of ARF in the nucleolus, which is the main site of ribosome biosynthesis, together with the known relevance of ARF for protein translation (35, 36) prompted us to investigate whether the adverse consequences of ARF for response to cisplatin may be due to impaired protein translation. Indeed, GSEA showed that among the most significant differentially regulated biological pathways in cisplatin-treated Arf-null versus Arf wild-type bladder tumors were those associated with protein translation, such as “Reactome peptide chain elongation” (NES 2.42, P < 0.001) and “KEGG ribosome” (NES 2.32, P < 0.001; Fig. 4A; Dataset 2). This was further evident by heatmap representation of genes in the leading edge of the GSEA, which showed upregulation of genes associated with ribosome production (Fig. 4B). In particular, expression of RPLP0 (for ribosomal protein lateral stalk subunit P0, a.k.a L10E, LP0, P0, RPP0), which encodes a major structural component of ribosomes, was significantly changed in Arf-null versus Arf wild-type tumors treated with vehicle (P = 0.04, t test), and even more so following treatment with cisplatin (P = 0.006, t test; Supplementary Fig. S2B). Notably, we found that RPLP0 is dysregulated in human bladder cancer in TCGA (8), where it is inversely correlated with expression levels of CDKN2A (Spearman correlation, r = −0.2148; P < 0.0001; Supplementary Fig. S3C).

To investigate the status of protein translation in Arf wild-type versus Arf-null cellular contexts, we used mouse embryo fibroblasts (MEF) derived from the corresponding mutant mice (see Materials and Methods). First, we analyzed the relative levels of actively transcribed polysomes in Arf wild-type versus Arf-null MEFs by comparing the polysome profiles following sucrose density gradient fractionation (Fig. 4C). These analyses showed greater abundance of polysomes in Arf-null MEFs relative to Arf wild-type MEFs (1.5- to 2-fold increase in three independent experiments; Fig. 4C), which is consistent with the increased levels of protein translation in the former. We then measured the relative levels of protein translation by analysis of S³⁵ incorporation into total protein, which revealed significantly greater levels in Arf-null versus Arf wild-type MEFs (P = 0.0004, t test; Fig. 4D). Furthermore, cisplatin reduced translation in the Arf-null MEFs (P = 0.002, t test), while there was a modest increase in translation in the Arf wild-type MEFs (P = 0.041, t test; Fig. 4D). These findings suggest Arf expression impairs protein translation, while cellular contexts lacking Arf have reduced translation following treatment with cisplatin. Furthermore, reduced expression of RPLP0 in human bladder cancer is associated with adverse outcome, as evident in Kaplan–Meier analyses comparing MIBC patients with low versus high expression levels of RPLP0 (P = 0.018, log-rank test; Fig. 4E), which extends the relevance of our findings from the mouse cell culture models to human bladder cancer.

Our findings have uncovered a novel mechanism of drug resistance that involves the repurposing of a tumor suppressor in a new role as an adverse effector of drug response. In particular, we find that bladder tumors having deletions or mutations of TP53 without corresponding loss of CDKN2A typically express high levels of ARF that accumulates in the nucleolus. However, while dysregulated ARF expression may not be detrimental for bladder tumorigenesis, when challenged with cisplatin, ARF protein in the nucleolus impairs drug response, which involves context-dependent alterations in protein translation (Fig. 4F). Our findings suggest that ARF status may predict response to chemotherapy, and furthermore, that human bladder tumors with elevated ARF may benefit from the addition of agents that affect protein translation together with chemotherapy.

Our findings also reveal a new role for the integrin β3–binding protein ITGB3BP (aka β3-endonexin, NRIF3, CENPR) as a mediator of cisplatin response in human bladder cancer. Although ITGB3BP was initially identified on the basis of its ability to interact with integrin (33, 34), ITGB3BP has multiple transcripts and, as evident from its various aliases, has been described as having several independent functions (e.g., refs. 33, 37). In particular, ITGB3BP was independently identified as a coactivator of nuclear hormone receptors (NRIF3 for nuclear receptor interacting factor 3) and shown to promote apoptosis in breast cancer cells (37). Besides this interesting link, ITGB3BP has not otherwise been widely studied in cancer. Notably, in other contexts, integrin signaling has been shown to be critical for chemoresistance (38), and has been linked to p53 status and ARF function (39). These observations, combined with the known role of ITGB3BP in breast cancer, which has similar molecular features as bladder cancer (8, 9), warrants its further study as a basis for understanding the role of integrin signaling in response to chemotherapy in bladder cancer.

ARF is known to function as a tumor suppressor that regulates p53 pathway activity in response to cellular stress, although ARF also has p53-independent functions (16, 17). However, in addition to its well-characterized tumor-suppressive role, ARF has been shown to have tumor-promoting or antiproliferative activities in some contexts (40–43). Moreover, ARF is upregulated following drug treatment in lung and ovarian cancer (44, 45), and its nucleolar expression is associated with adverse outcome in lymphoma (46). Therefore, it has become increasingly apparent that the functions of ARF and its relationship to p53 status are relatively complex.

Our current findings expand the complexity of ARF functions in cancer by demonstrating that it has an adverse effect on drug response. Importantly, we have found that the consequences of ARF expression were not manifested in tumor models that were unchallenged by cisplatin. Instead, this drug treatment unmasked an otherwise occult tumor-promoting role of ARF in mediating drug resistance. Thus, our results are not discordant with the known function of ARF in tumor suppression, nor are they inconsistent with previous studies that have attributed ARF overexpression in tumors to be a secondary consequence of p53 mutation; in fact, our current findings may explain why it is necessary for ARF expression to be so tightly regulated in normal cellular contexts (21).

The nucleolus is known to be a target of cellular stress and is a key target for therapeutic intervention in cancer. Our current findings provide a novel connection among several important observations that have been made previously in the literature; namely, that nucleolar ARF has an inhibitory role in protein translation (35, 36), that chemotherapy inhibits ribosome biosynthesis (47–49), and that the nucleolus is important for modulating DNA damage in response to chemotherapy (50, 51). In particular, we find that nucleolar ARF inhibits ribosome biosynthesis, consistent with previous observations (35, 36). Unexpectedly, however, this novel activity of ARF counteracts the inhibitory effects of cisplatin for tumor growth; in particular, although cisplatin inhibits protein translation in Arf-null contexts, inhibition of protein translation is augmented by cisplatin in the context of nucleolar Arf. Notably, we have found that expression of the ribosomal protein RPLP0 is associated with worse disease outcome, which further emphasizes the relevance of ribosomal proteins and cancer (52, 53). The importance of translational control in MIBC is evident from prevalent alteration of genes associated with protein translation, including PI3CA (∼20%) TSC1 (∼9%), and PTEN (∼6%; ref. 8). Furthermore, cisplatin sensitivity has also been shown to correlate with alterations in DNA repair in patients with bladder cancer (4, 5). Given that ARF is a sensor for regulating DNA damage response (54, 55), it will be of interest to assess the relationship of ARF to DNA damage in these mutational contexts.

The broader implication of our findings is that tumor contexts in which ARF is expressed may benefit from the combination of chemotherapy with agents that affect protein translation. Indeed, the importance of protein translation and the benefits of therapeutic targeting of translation are now well established. In future studies, these concepts can be assessed in coclinical trials using preclinical models similar to those that we have described here.

No potential conflicts of interest were disclosed.

Conception and design: T.B. Owczarek, T. Kobayashi, L. Rong, C. Abate-Shen

Development of methodology: T.B. Owczarek, T. Kobayashi, L. Rong, J. Wang, D.B. Solit, H.A. Al-Ahmadie, C. Abate-Shen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.B. Owczarek, T. Kobayashi, L. Rong, A.M. Puzio-Kuter, M.Y. Teo, D.B. Solit, H.A. Al-Ahmadie

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.B. Owczarek, T. Kobayashi, R. Ramirez, G. Iyer, F. Sanchez-Vega, N. Schultz, T. Zheng, D.B. Solit, H.A. Al-Ahmadie, C. Abate-Shen

Writing, review, and/or revision of the manuscript: T.B. Owczarek, T. Kobayashi, G. Iyer, M.Y. Teo, T. Zheng, D.B. Solit, H.A. Al-Ahmadie, C. Abate-Shen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Iyer, M.Y. Teo

Study supervision: C. Abate-Shen

We thank Cathy Mendelsohn, Antonina Mitrofanova, Kenneth Olive, and Michael Shen for helpful comments on the manuscript, and Charles Sherr (St. Jude Children's Research Hospital) for the gift of the Arf mice.

This work was supported by grants from the JP Sulzberger Columbia Genome Center and the Small-Animal Imaging Facility, which are shared resources of the Herbert Irving Comprehensive Cancer Center at Columbia University, supported in part by NIH/NCI grant #P30 CA013696. Core facilities at Memorial Sloan Kettering Cancer Center are supported in part by NIH/NCI grant #P30 CA008748. This work was supported in part by NIH grants CA193442 (C. Abate-Shen) and CA182587 (D.B. Solit), funding from the TJ Martell Foundation for Leukemia, Cancer and AIDS Research (C. Abate-Shen), and an innovator award from the Bladder Cancer Advocacy Network (C. Abate-Shen and D.B. Solit). T.B. Owczarek was supported by a post-doctoral research fellowship from the Urology Care Foundation Research Scholars Program and Dornier MedTech. T. Kobayashi was supported by post-doctoral training grants from the American Urological Association Foundation and the American Association for Cancer Research-Amgen Clinical and Translational Cancer Research Fellowship. C. Abate-Shen is an American Cancer Society Research Professor supported in part by a generous gift from the F.M. Kirby Foundation.

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.