Genomic Tumor Correlates of Clinical Outcomes Following Organ-Sparing Chemoradiation Therapy for Bladder Cancer

Radiation therapy with or without chemotherapy remains a pillar of curative-intent treatment in many solid tumors (1), often allowing avoidance of radical surgery and facilitating organ sparing, such as in patients with breast cancer or muscle-invasive bladder cancer (MIBC; refs. 2, 3). Predictive biomarkers for the rational selection of systemic therapies for individual patients have transformed precision oncology (4), but few biomarkers are available clinically to aid therapeutic decisions involving radiotherapy despite the genomic heterogeneity of cancers (5, 6). There exists an urgent need to identify candidate biomarkers for predicting successful outcomes after radiotherapy to help improve selection of patients and potentially allow for de-escalation of therapy in specific clinical circumstances (7).

Promising predictive biomarkers of radiotherapy response include genomic alterations in DNA damage response (DDR) genes (8, 9). Radiation-induced DNA damage is sensed and repaired via a highly coordinated cellular DDR, and tumors that harbor alterations in DDR pathways may have increased sensitivity to radiation. Therefore, there is interest in evaluating DDR genes to determine whether specific DDR alterations are correlated with clinical radiosensitivity of tumors (9, 10). Other studies have identified potential non-DDR tumor biomarkers as predictors for radiotherapeutic efficacy or resistance (11–14), but despite interest (5), this area of precision oncology remains understudied.

Trimodality therapy (TMT), which consists of maximal transurethral resection of bladder tumor (TURBT) followed by chemoradiation, is an effective bladder-preserving treatment for well-selected patients with MIBC (3, 15). Although no randomized trials have been completed comparing TMT with the treatment of radical cystectomy with or without neoadjuvant chemotherapy (2), multiple long-term series have demonstrated comparable outcomes (16–19). MIBC has been previously molecularly characterized, with recurrent mutations identified in known cancer genes such as TP53, KMT2D, and KDM6A. Prior analyses have evaluated gene alterations with clinical outcomes following neoadjuvant chemotherapy prior to cystectomy and identified DDR genes associated with response (20, 21). MIBC molecular subtypes defined by transcriptional profiles have been found to be associated with outcomes in patients with MIBC undergoing radical cystectomy with or without neoadjuvant chemotherapy (22–24). Gene expression profiling of 136 primary tumors from patients with MIBC treated with TMT identified an association of T-cell activation and INFγ signaling expression signatures with improved disease-specific survival in the TMT cohort (25). However, there remains a need for a deeper understanding of the molecular characteristics of MIBC to aid with treatment selection and identify more accurate prognostic and predictive biomarkers of chemoradiation outcomes (26, 27). To this end, we leveraged a well-curated clinical cohort of patients with MIBC homogenously treated with TMT that included patients previously evaluated for gene expression profiling (25), and performed whole-exome sequencing (WES) in combination with gene expression profiling to identify potential molecular biomarkers associated with clinical outcomes of chemoradiation therapy.

Patients with cT2-T4aN0M0 MIBC treated with definitive-intent chemoradiation as part of TMT were retrospectively identified from a single institution between 1986–2015. This study was approved by the Mass General Brigham Institutional Review Board (IRB; Protocol #2008P001128), and studies were conducted in accordance with the Declaration of Helsinki and in accordance with an assurance filed with and approved by the U.S. Department of Health and Human Services. Analyzed tumor tissues were excess discarded clinical specimens previously collected from human subjects as part of usual clinical care under an IRB-approved waiver of consent. Tumors from this cohort had previously been analyzed for whole transcriptome analysis (25), and those with sufficient remaining formalin-fixed, paraffin-embedded (FFPE) tissue available for whole-exome sequencing were included in the study.

All molecular characteristics were correlated with clinical outcomes following TMT. The main clinical outcome used in this analysis was modified bladder-intact event-free survival (mBI-EFS), with an event defined as muscle-invasive recurrence, locoregional nodal progression, distant metastasis, death from bladder cancer, and radical cystectomy for either cancer recurrence or toxicity. For the purposes of this analysis, patients who had a cumulative incidence of an event were defined as having an “unfavorable” long-term mBI-EFS outcome, while those without a cumulative incidence of an event were defined as having a “favorable” long-term mBI-EFS outcome. This long-term outcome classification is a secondary endpoint of the INTACT SWOG/NRG 1806 cooperative group trial of TMT with or without atezolizumab (NCT03775265), and is closely related to bladder-intact event-free survival (BI-EFS), the primary endpoint of the trial, with the only difference being that mBI-EFS does not include all-cause mortality. These endpoints were developed by investigators from ECOG, RTOG/NRG, ALLIANCE, SWOG, and CCTG, in discussion with the NCI GU Steering Committee and the U.S. FDA, as the composite endpoint most clinically relevant for this patient population. Other clinical endpoints examined included cumulative incidence of locoregional failure, cumulative incidence of distant failure, locoregional control, mBI-EFS, BI-EFS, and overall survival (OS).

DNA extraction, whole-exome library preparation, and sequencing were performed at the Broad Institute using methods previously described (12, 28). FFPE TURBT specimens were reviewed by a board-certified genitourinary pathologist to select high-density tumor blocks. Normal (germline) DNA samples were not available. Genomic DNA was purified from FFPE cores using a commercial kit (Qiagen QIAamp DNA FFPE Tissue Kit Quantitation Reagent). Whole-exome capture libraries were constructed from 100 ng of DNA after sample shearing, end repair, and phosphorylation and ligation to barcoded sequencing adaptors. Ligated DNA was size selected for lengths between 200–350bp and subjected to exonic hybrid capture using The Broad Institute Genomics Platform Custom Illumina bait (12). Briefly, the Illumina exome specifically targets approximately 37.7 Mb of primarily exonic territory consisting of all targets from the Agilent exome design (Agilent SureSelect All Exon), all coding regions of Gencode V11 genes, and all coding regions of RefSeq gene and KnownGene tracks from the UCSC genome browser (http://genome.ucsc.edu). The sample was multiplexed and sequenced using Illumina HiSeq/NextSeq technology. Data were analyzed using the Broad Picard Pipeline, which includes demultiplexing and data aggregation.

Data processing and analysis of exome sequence data were performed using Broad Institute pipelines as described previously (12, 28). We used the Getz Lab CGA WES Characterization pipeline developed at the Broad Institute to call, filter, and annotate somatic mutations and copy number variations. The pipeline employs the following tools: MuTect (29), ContEst (30), Strelka (31), Orientation Bias Filter (32), DeTiN (33), AllelicCapSeg (34), MAFPoNFilter (35), RealignmentFilter, ABSOLUTE (36), GATK (37), PicardTools (38), Variant Effect Predictor (39), and Oncotator (40). ContEst (30) was used to evaluate cross-individual contaminations and samples with <4% estimated contamination were included. The MuTect algorithm (29) was applied to identify single-nucleotide variants in targeted exons, and Strelka (31) was used to identify small deletions or insertions. All alterations were annotated with Oncotator (40). Due to the absence of germline DNA, a common variant filtering strategy was implemented to remove likely germline artifacts, using the ExAC database, now a part of gnomAD (41). This tool is also used by AACR Genie (42). Common variants were annotated and soft filtered if the variant appeared in at least 10 alleles across any ancestral subpopulation in ExAC. Mutations were examined for distribution and type and confirmed using Integrative Genomics Viewer (43, 44). Genes comprising known oncogenic signaling pathways (as described in Robertson and colleagues; ref. 45) as well as the following frequently mutated genes in MIBC were analyzed:

TP53/Cell Cycle pathway: ATM, TP53, MDM2, CDKN2A, RB1, CCND1, CDKN1A, PTEN, CCNE1, FBXW7, CDKN1B, CCND1/2/3, CDK4/6

RTK/RAS/PIK3K pathway: PIK3CA, FGFR1/3, ERBB2/3, RAF1, PTEN, TSC1/2, EGFR, AKT1/2, NF1, RAC1, H/N/KRAS, JAK1/2, BRAF

Histone modification pathway: EP300, CREBBP, KMT2C/D, KDM6A, BAP1, ASXL1/2, SETD2

SWI/SNF pathway: ARID1A, ARID1B, ARID2

DDR pathway: ERCC2, BRIP1, ATM, BRCA1/2, RAD21/50, CHEK1

Cohesin complex pathway: STAGE1/2, RAD21, SMC1A/3

Oxidative stress pathway: NFE2L2, KEAP1, CUL3, TXNIP

Alternative splicing pathway: RMB10, SF3B1, U2AF1, CDK12

Other frequently mutated MIBC genes: KMT2A, FAT1, ELF3, SPTAN1

In addition, we examined alterations in specific genes that are being evaluated in on-going clinical trials that incorporated a risk-adapted chemotherapy-based approach to the treatment of MIBC: RETAIN (NCT02710734) and Alliance A031701 (NCT03609216). In RETAIN BLADDER, a mutation-positive patient is defined as any alteration present in ATM, RB1, FANCC, or ERCC2. In the Alliance trial, pretreatment TURBT specimens are evaluated for deleterious DDR gene alternations (truncating mutations including frameshift, nonsense, splice site) in 8 genes (ATM, ATR, FANCC, BRCA1, BRCA2, RECQL4, RAD51C, ERCC5) or any mutation in ERCC2. Loss-of-function mutations in these genes were identified using OncoKB (46).

Mutational signatures (COSMIC Signatures v3.2) were derived from tumor samples using deconstructSigs (https://github.com/raerose01/deconstructSigs) to determine the weights of each mutational signature contributing to the mutational spectrum of a tumor sample (47).

RNA extraction from FFPE cores and expression profile generation was previously performed (25). Briefly, transcriptome-wide gene expression profiles were generated using Human Exon Array 1.0 ST oligonucleotide microarrays in a clinic-grade laboratory (Veracyte).

The Genomic subtyping classifier (GSC) subtypes were generated by first identifying neuroendocrine-like (NE-like cases; ref. 48), with the remaining tumors classified using the Seiler 2017 model (23). A T-cell–inflamed immune content signature (25, 49), MIBC consensus class (50), as well as The Cancer Genomic Atlas (TCGA 2017) subtyping models were also used to classify our tumors into various subtypes. Transcriptomic data was available for 67 of the 76 patients (88%) in this cohort.

The bladder cancer cell lines used here have been previously described (51, 52). KU19–19 (RRID: CVCL_1344) and SW1710 (RRID: CVCL_1721) cell lines were originally purchased from DSMZ and were validated by STR profiling. KE1 is a derivative of the KU19–19 bladder cancer cell line and SC14 is a derivative of the SW1710 bladder cancer cell line. Both KE1 and SC14 harbor ERCC2 mutations introduced by CRISPR/Cas9 techniques and have been shown to be nucleotide excision repair (NER) deficient. The KU19–19 and KE1 cell lines were cultured in RPMI media supplemented with 10% FBS. SW1710 and SC14 cell lines were cultured in DMEM supplemented with 10% FBS. Early-passage cells (≤6 passages) were used for all experiments and cell lines were tested monthly to confirm absence of Mycoplasma. For cisplatin and radiation sensitivity assays, cells were plated in 6 well plates (KU19 and KE: 5,000 cells/well; SW1710 and SC14: 8,000 cells/well) and the following day were treated with cisplatin and/or radiation. Cells were then incubated for 10 days, fixed with 4% paraformaldehyde for 20 minutes, stained with 1% crystal violet for 1 hour, and then washed with tap water, allowed to dry, and photographed. For quantification, 1 mL of 1% SDS solution was added to each well to dissolve the crystal violet and the sample solutions were then scanned on a microplate reader at a wavelength of 570 nm. Survival for each condition was calculated as the signal divided by the signal in untreated cells. Survival curves were plotted using Graphpad Prism and the survival fraction was calculated using a linear-quadratic (LQ) model. Combenefit software was used to perform synergy analysis (53).

Differences in single-nucleotide variations and tumor mutational burden (TMB) between those with favorable versus unfavorable clinical outcomes were calculated using χ² analyses and unpaired t tests, respectively. A Benjamini–Hochberg FDR of 0.1 was used to account for multiple hypothesis testing. COSMIC mutational signatures were evaluated using Kruskal–Wallis H and Wilcoxon rank sum tests. Differences in expression profile subtypes between favorable and unfavorable long-term mBI-EFS outcomes were calculated using χ² test. Associations of transcriptomic signature scores and gene mutations versus wild-type (WT) were analyzed using Mann–Whitney U tests. Association of select gene mutations with mBI-EFS, BI-EFS, locoregional control, and OS was assessed using the Kaplan–Meier method and log rank test with P < 0.05 considered to be statistically significant.

The WES data generated in this study are publicly available in dbGAP at phs003402.v1.p1. Expression profile data analyzed in this study were obtained from Gene Expression Omnibus at GSE128701.

There were 92 patients who met initial criteria for inclusion in the study. Seventy-six cases passed quality control metrics for WES and comprise our final cohort (Supplementary Fig. S1). Clinical characteristics of the cohort are summarized in Table 1. The median age of the cohort was 72.3 years, with 24% female patients. All patients had MIBC (86% T2, 14% T3/T4) and the majority were treated with cisplatin-based chemotherapy (66%) concurrent with radiation (median dose 64.3 Gy, interquartile range, 60.7–64.3).

The genomic tumor landscape of the cohort is summarized in Fig. 1 and Supplementary Fig. S2. Figure 1 displays all genes comprising known oncogenic signaling pathways and known frequently mutated MIBC genes (Methods) with at least 5% mutation frequency. Mean target coverage was 109X for somatic tumor DNA (Supplementary Table S1). Mean nonsynonymous TMB for the entire cohort was 12.4 mutations/Mb (range, 3.4–31.5). Overall, TP53 was the most frequently mutated gene with 52.6% of samples having an alteration, followed by KMT2D at 40.8%. Mutational signature analysis identified four active signatures: Signatures 1, 2, 5, and 13. Signature 1, present in 62/76 (81.6%) samples, is associated with age-associated spontaneous or enzymatic deamination of 5-methylcystosine to thymine (54). Signature 5, which is the signature previously associated with ERCC2 mutations in urothelial tumors and with smoking history, was present in 68/76 (89.5%) samples (55). Signatures 2 and 13, present in 44/76 (57.9%) and 39/76 (51.3%) cases, respectively, are associated with APOBEC-mediated mutations (54), and have been described previously in MIBC (45). There were significant differences in the relative contribution of mutations across mutational signatures (Supplementary Fig. S3; Kruskal–Wallis H, P < 0.001), similar to previous observations in MIBC cohorts (56). Signature 5 had a significantly higher contribution than other signatures (Wilcoxon rank sum, P < 0.001), but patients with ERCC2-mutant tumors did not have higher contribution from signature 5 compared with patients with ERCC2 WT tumors (P = 0.813).

Sixty-seven samples had transcriptomic data available (Fig. 1). Among the GSC subtypes (23, 48), Luminal was most common (21/67, 31.3%), while the Luminal papillary subtype was most common among TCGA subtypes (21/67, 31.3%), and the Basal squamous subtype was most common among the MIBC consensus class (19/67, 28.4%). The distribution of TCGA subtypes in our cohort was not significantly different from the distribution in the TCGA MIBC cohort (Supplementary Fig. S4; ref. 45). Overall, the mean T-cell–inflamed signature score was 0.23 (range, −0.04 to 0.98).

To verify that the genomic data correlates with the transcriptomic data, we examined how a few genomic alterations associated with corresponding transcriptomic signature scores. FGFR3 mutations are common in non–MIBC as well as in tumors of the upper tract, but are also present at lower frequencies in MIBC (57). In this cohort, 6 of 76 tumors (7.9%) harbored FGFR3 mutations (Supplementary Fig. S5A). A lncRNA-based genomic classifier (58) revealed significant enrichment for FGFR3+ (“FGFR3 active”) cases in FGFR3 mutant tumors (4/6 cases; 66.7%) compared with FGFR3 WT tumors (4/61 cases; 6.6%, P < 0.0001). Similarly, tumors with FGFR3 mutations also had higher FGFR3 transcriptional signature scores (59) compared with FGFR3 WT tumors (median 0.528 versus 0.191; P < 0.001; Supplementary Fig. S5B).

Mutations in TP53 were most common in the cohort (40/76 samples mutated). Gene expression profiling revealed higher hallmark p53 pathway scores (60) among those with TP53 WT tumors compared with mutant (median 0.3593 versus 0.1171; P = 0.0001; Supplementary Fig. S6A). Separately, upon evaluation of a previously described novel TP53 mutation signature in prostate cancers (61), tumors with TP53 mutations had higher signature scores compared with WT tumors (median 0.9945 versus 0.9786; P = 0.020; Supplementary Fig. S6B), consistent with prior associations (61).

Median follow-up for the cohort was 74.6 (range, 20.0–266.4) months in alive patients. Thirty-seven of 76 patients (49%) were characterized as having favorable long-term outcomes for mBI-EFS, and 39 (51%) as having unfavorable long-term outcomes for mBI-EFS (see study endpoints in Methods). Five-year mBI-EFS for the entire cohort was 45.4% (Supplementary Fig. S7A). Five-year BI-EFS for the entire cohort was 37.9% (Supplementary Fig. S7B), and 5-year OS for the entire cohort was 48.6% (Supplementary Fig. S7C). There were 19 locoregional failures and 22 distant failures.

TMB has been shown to be associated with response to immune checkpoint inhibitors (62, 63), and TMB has also been suggested to be associated with radiation response in other cancers (64). We therefore evaluated whether there were differences in TMB in patients with MIBC with favorable compared with unfavorable clinical outcomes after TMT in our cohort. Interestingly, there was no significant difference in TMB between those with favorable (13.3 mutations/Mb, range 3.4–31.5) and unfavorable mBI-EFS outcomes (11.5 mutations/Mb, range 6.4–26.8), P = 0.1584 (Supplementary Fig. S8).

We then performed an exploratory analysis of DDR genes as defined in the Alliance A031701 (NCT03609216) and RETAIN (NCT02710734) clinical trials of risk-adapted MIBC treatment to determine whether mutations in any of these 10 DDR genes correlated with clinical outcomes after TMT (Fig. 2A). The following mutations were identified in our cohort: ATM (n = 13), RB1 (n = 17), FANCC (n = 4), ERCC2 (n = 8), ATR (n = 9), BRCA1 (n = 7), BRCA2 (n = 11), RECQL4 (n = 0), RAD51C (n = 0), and ERCC5 (n = 2; Supplementary Table S2). TMB was higher among DDR mutant tumors (13.4 mutations/Mb, range 3.4–31.5) compared with DDR WT tumors (10.8 mutations/Mb, range 5.5–24.7, P = 0.04). In the TCGA cohort which included patients with MIBC who underwent surgical resection (45), the presence of these DDR mutations was not significantly associated with improved survival [median months survival in patients with DDR mutations: 41.72 (95% confidence interval (CI), 26.91–92.9) versus 33.11 (95% CI, 22.86–54.86); P = 0.442; Supplementary Fig. S9]. However, in our TMT cohort, having any DDR gene mutation present was associated with improved mBI-EFS (HR, 0.51; 95% CI, 0.27–0.97; P = 0.040; Fig. 2B), as well as improved locoregional control (HR, 0.31; 95% CI, 0.12–0.80; log-rank P = 0.009; Fig. 2C). This was also consistent when evaluating cumulative incidence of locoregional failures (n = 19); patients without locoregional failure were enriched for the presence of any DDR gene mutation present [39/57 (68.4%) with mutations] compared with patients with locoregional failure [7/19 (37%) with mutations; χ²P = 0.014]. There were no associations between DDR gene alterations and distant failures (n = 22 total distant failures, 11 among patients with DDR gene-altered tumors). In addition, there was no association between patients with TP53 mutations and favorable or unfavorable mBI-EFS.

Next, because missense mutations in DDR genes are often non-functional passenger events, we further evaluated whether the presence of known loss of function (LOF) mutations in DDR genes (as annotated in OncoKB; ref. 46) impacts outcomes after TMT (Supplementary Tables S3 and S4). We found that patients without locoregional failure continued to be enriched for the presence of any DDR (LOF) mutation present [26/57 (66.7%) with mutations] compared with patients with locoregional failure [2/19 (11%) with mutations; χ²P = 0.006]. Patients with favorable mBI-EFS outcomes were enriched for the presence of an ERCC2 LOF mutation [4/37 (11%) with mutations] compared with patients with unfavorable mBI-EFS outcomes [1/39 (3%) with mutations; χ²P < 0.001].

We then evaluated the proportion of patients with favorable mBI-EFS outcomes after TMT who would have met DDR gene mutation-specific inclusion criteria for either of the risk-adapted bladder-sparing clinical trials that are testing whether chemotherapy alone is sufficient to omit upfront local therapy (cystectomy or radiation) in patients with MIBC with DDR gene mutations (Supplementary Fig. S10A and S10B). Upon applying these inclusion criteria [Alliance A031701 included any truncating mutations including frameshift, nonsense, splice site in 8 genes (ATM, ATR, FANCC, BRCA1/2, RECQL4, RAD51C, ERCC5) or any mutation in ERCC2; RETAIN included any alteration present in ATM, RB1, FANCC, or ERCC2], it was found that 20/37 (54%) patients with favorable mBI-EFS had DDR gene mutations necessary for eligibility for one or both trials (9 RETAIN only, 2 Alliance only, 9 both trials), while 17/37 (46%) patients did not meet any of the mutation eligibility criteria for these trials (Supplementary Fig. S10A; Supplementary Table S4). Thus, nearly half of patients who had favorable mBI-EFS outcomes after bladder-sparing TMT would not have been eligible for participation in chemotherapy-only–based risk-adapted bladder preservation clinical trials.

Of the DDR gene alterations evaluated, ERCC2 mutations stood out as being significantly associated with favorable mBI-EFS outcomes (n = 7) as compared with unfavorable outcomes (n = 1, locoregional failure; Fisher exact P = 0.027; Fig. 2A). ERCC2 is a DNA helicase that plays an important role in the NER pathway (65). Upon examination of the distribution of 8 ERCC2 mutations, 7 were located within one of the two conserved helicase domains (Fig. 3A; Supplementary Table S5). The distribution of ERCC2 mutations in this cohort reflects the overall mutation distribution present in the TCGA bladder cohort (Supplementary Fig. S11; ref. 45). Six of the 8 patients with ERCC2 mutations received cisplatin-based chemotherapy concurrently with radiation (all with favorable mBI-EFS outcomes); of the 2 that did not, 1 received mitomycin C, which also creates DNA lesions that require the NER pathway for repair of DNA damage, and was also associated with favorable mBI-EFS outcome. The 1 patient with an unfavorable mBI-EFS outcome received concurrent 5-FU. Therefore, among patients who received cisplatin-based chemoradiotherapy, all 6 patients with an ERCC2 mutation had favorable mBI-EFS outcomes. Patients with ERCC2 mutations had significantly improved mBI-EFS (HR, 0.15; 95% CI, 0.06–0.37; P = 0.030; Fig. 3B), and improved BI-EFS, an endpoint that includes all-cause mortality (HR, 0.33; 95% CI, 0.15–0.68; P = 0.044; Fig. 3C). A trend was also noted toward improved OS in patients with ERCC2 mutations (HR, 0.36; 95% CI, 0.17–0.80; P = 0.075; Fig. 3D). There was no significant difference in TMB among ERCC2 mutation tumors (14.0 mutations/Mb, range 4.2–24.3) versus ERCC2 WT tumors (12.2 mutations/Mb, range 3.4–31.5; P = 0.38).

The mBI-EFS endpoint includes cystectomy due to treatment toxicity as an event, and in our cohort, 2 patients had their bladders removed due to toxicity (Supplementary Table S4). To evaluate whether excluding these events would affect our analysis, we evaluated a new endpoint that does not include cystectomy for toxicity (called “re-modified BI-EFS”). We found that having any DDR mutation was significantly associated with improved re-modified BI-EFS (HR, 0.51; 95% CI, 0.26–0.99; P = 0.040; Supplementary Fig. S12) and patients with ERCC2 mutations had significantly improved re-modified BI-EFS (HR, 0.15; 95% CI, 0.06–0.37; P = 0.037; Supplementary Fig. S13).

We then evaluated the transcriptional landscape of ERCC2 mutant tumors compared with ERCC2 WT tumors. Patient samples with ERCC2 mutations (n = 6 with transcriptomic data available) had differential enrichment of several transcriptomic signature scores and molecular subtypes compared with ERCC2 WT tumors (n = 61; Fig. 4A). Samples with ERCC2 mutations had higher hallmark DNA repair scores (ref. 60; median 0.110 versus −0.318; P = 0.020), consistent with dysregulation of DDR (Supplementary Fig. S14). In addition, ERCC2 mutant tumors were enriched for TCGA Luminal papillary subtype compared with ERCC2 WT (5/6, 83% versus 26% Luminal papillary; P = 0.004; Fig. 4B). These same patients were also enriched for GSC Luminal subtype compared with ERCC2 WT tumors (4/6, 67% versus 28% Luminal; P = 0.050, Supplementary Fig. S15). ERCC2 mutant tumors were not associated with higher or lower T-cell–inflamed signature scores compared with ERCC2 WT tumors.

We next tested the functional impact of ERCC2 mutations on cisplatin and radiation sensitivity in bladder cancer cell lines. KU19–19 and SW1710 are bladder cancer cell lines with WT ERCC2, while KE1 and SC14 are derivatives of KU19–19 and SW1710, respectively, with ERCC2 mutations introduced by CRISPR-mediated gene editing (52). We have previously shown that the ERCC2-mutant KE1 and SC14 lines are NER-deficient and have increased cisplatin sensitivity compared with their respective WT ERCC2 parental lines (51, 52). To model the impact of ERCC2 functional loss on cisplatin-based chemoradiotherapy response, we compared the sensitivity of the KU19–19/KE1 and SW1710/SC14 cell pairs to varying doses of combined cisplatin and radiation treatment. As expected, the ERCC2-mutant lines were significantly more sensitive to cisplatin monotherapy than the WT ERCC2 parental lines (Fig. 5A–D). The addition of sublethal doses of cisplatin to radiation did not significantly impact sensitivity in the parental KU19–19 and SW1710 cell lines; however, there was a significant increase in sensitivity with dose enhancement factors of 1.55 and 1.40 when cisplatin was added to radiation in the ERCC2-mutant SC14 and KE1 cell lines, respectively (Fig. 5E–J). The ERCC2 mutant clones demonstrated mild increase in radiation sensitivity compared with WT, but this was much less pronounced than the increase in cisplatin sensitivity (Fig. 5E and F). Taken together, these data demonstrate that the presence of a deleterious ERCC2 mutation is sufficient to drive sensitivity to combined radiation and cisplatin treatment in bladder cancer preclinical models.

Biomarkers of treatment outcomes to guide precision oncology are needed to select patients for radiation or chemoradiation-based organ-sparing treatment approaches. In MIBC, TMT can be an effective alternative to radical cystectomy for bladder preservation, and there is an urgent unmet need for biomarkers to guide treatment selection. In this study, we performed integrated genomic and transcriptomic analyses to evaluate genomic determinants of clinical outcomes following TMT in the largest MIBC cohort with such data available. When evaluating molecular associations with clinical outcomes, including BI-EFS and mBI-EFS which are clinically relevant endpoints for this patient population (66), it was found that having any DDR gene alteration as per genes defined in the Alliance A031701 (NCT03609216) and RETAIN BLADDER (NCT02710734) trials was associated with improved locoregional control as well as mBI-EFS. Of these, ERCC2 mutations stood out as being significantly associated with improved mBI-EFS and BI-EFS, with a trend towards improved OS. This is the first study to associate molecular biomarkers with BI-EFS and mBI-EFS, clinical trials endpoints previously defined by researchers and stakeholders invested in relevant outcomes for patients with MIBC.

Genomic alterations in DDR genes are prevalent in bladder cancer, and these alterations have been previously studied and associated with improved responses to chemotherapy and immune checkpoint inhibitors (20, 21, 67, 68). DDR gene mutations have been previously found to be associated with higher TMB (69, 70), and our cohort similarly demonstrated higher TMB among DDR mutated tumors versus DDR WT tumors. Although TMB was not associated with outcomes after TMT, DDR gene alterations were associated with favorable long-term outcomes in our TMT cohort. Interestingly, DDR gene mutations are not associated with improved survival in the TCGA cohort, which is comprised of patients treated with radical cystectomy, indicating that DDR gene mutations are not inherently prognostic. ERCC2 mutations have been associated with sensitivity to cisplatin-based chemotherapy, and functional profiling of clinically observed ERCC2 missense mutations has demonstrated that nearly all mutations arising within the conserved helicase domains are unable to support normal cellular NER, including mutations at Y24, N238, G606, and G665 as observed in this cohort (52, 65). While ERCC2-dependent NER has no known role in the repair of radiation-induced DNA damage (71), our functional data demonstrate that the presence of an ERCC2 mutation is sufficient to sensitize bladder cancer cells to combination treatment with cisplatin and radiation. Consistent with this notion, 75% (6/8) of the patients with ERCC2 mutations received concurrent cisplatin-based chemotherapy in our cohort. The one patient with unfavorable mBI-EFS in the cohort with an ERCC2 mutation did not receive cisplatin-based chemotherapy. Although we cannot parse out the relative contributions of chemotherapy versus radiation to the favorable mBI-EFS outcomes observed in patients with DDR or ERCC2 mutations, contemporary TMT includes both concurrent radiation and low-dose radiosensitizing cisplatin (or other DNA damaging) chemotherapy. Future investigations could evaluate relative contribution of these mutations in neoadjuvant full-dose cisplatin cohorts compared with TMT cohorts receiving concurrent low-dose radiosensitizing cisplatin, or by comparing with outcomes in chemotherapy-only bladder-sparing trials. Nevertheless, our data and data from others (72) suggest that ERCC2 could serve as a potential biomarker for TMT treatment selection in cisplatin-eligible patients.

Our analysis also evaluated patients who had transcriptomic signature/subtype data available, and we observed higher hallmark DNA repair scores (60) among samples with ERCC2 mutations. These tumors were also enriched for GSC Luminal subtype and TCGA Luminal papillary subtype, compared with ERCC2 WT tumors. Prior data indicate that the Luminal/Luminal papillary subtypes are associated with heterogeneous responses to neoadjuvant chemotherapy (73), while a separate analysis did not find any association between molecular subtype and rate of complete response following TMT (25). In the TCGA cohort, ERCC2 mutant tumors were not enriched for Luminal papillary subtype (74). Therefore, our findings must be validated in larger cohorts.

Due to the prevalence of DDR mutations in MIBC, trials are currently underway which use a “risk-adapted” bladder preservation approach based on DDR gene mutational status to guide treatment strategies. The RETAIN BLADDER phase II trial, the Alliance A031701 trial, and the HCRN GU 16–257 (ref. 75; NCT03558087) are several examples exploring such an approach, but these bladder-sparing trials omit upfront local therapy (cystectomy or radiation) and employ chemotherapy-only regimens. In our cohort, we found that tumors with any DDR alteration were associated with improved locoregional control and mBI-EFS following standard-of-care TMT. In addition, 46% of patients with favorable mBI-EFS outcomes after TMT did not meet mutation criteria for inclusion in either of these chemotherapy-only trials.

To our knowledge, this is the largest study to date of integrated molecular profiling of samples from a cohort of TMT patients, and one of few studies leveraging both genomic and transcriptomic data, along with functional characterization of our findings. There are several limitations to this study, including its retrospective nature and small numbers. In addition, the lack of matched normal tissue could lead to overcalling alterations, although we used several approaches to account for this (41, 42). The mean TMB in the TCGA bladder cohort is 7.8 mutations/Mb (range 0.03–118.2; ref. 45), and although the mean TMB in our cohort [12.4 mutations/Mb (range, 3.4–31.5)] is significantly different (P < 0.001), it falls well within the range reported by TCGA and is not near the upper end of the range. Therefore, our computational approaches helped control for the lack of matched normal sample, and the molecular characteristics of our cohort are consistent with other large MIBC cohorts. We also lack a validation cohort to confirm our findings; however, a cohort of patients with MIBC who underwent TMT at multiple institutions from the NRG Oncology/Radiation Therapy Oncology Group trials will be analyzed as a separate study. Finally, tumor spatial heterogeneity cannot be inferred from this study as we do not have data from multiple areas of each tissue sample. Our study is strengthened by the use of a homogenously treated TMT cohort that is well-representative of the general MIBC patient population; the frequency of gene alterations in our TMT cohort were consistent with other large MIBC cohorts, including alterations in TP53 (n = 40, 52.6%), KMT2D (n = 31, 40.8%), and KMT2C (n = 16, 21.1%). The frequency of FGFR3 alterations was relatively low in this cohort (n = 6, 7.9%), consistent with a lower frequency in MIBC compared with non-invasive bladder cancer (76). We found mutation signatures previously associated with MIBC including Signatures 1, 2, 5, and 13 (77). The distribution of ERCC2 mutations in this cohort reflects the overall mutation distribution in cancers of the urogenital tract per TCGA bladder, supportive of the representative nature of our MIBC cohort. Therefore, our findings can be broadly applicable to patients with MIBC.

In summary, our study leverages integrated molecular profiling of a large cohort of patients with MIBC homogenously treated with TMT to evaluate for biomarkers of response. Alterations in a DDR gene, ERCC2, associated with improved mBI-EFS following standard-of-care TMT, as well as a trend toward improved OS. These data require validation but demonstrate early promise in elucidating novel biomarkers to guide precision oncology for radiation-based management in patients with MIBC.

C.-L. Wu reports other support from GoPath Diagnostics, as well as personal fees from OrigiMed Co. Ltd. outside the submitted work. A.S. Feldman reports research funding from Convergent Genomics. R.J. Lee reports personal fees from Exelixis, Bayer, Dendreon, and Blue Earth and grants and personal fees from Janssen outside the submitted work. E. Van Allen reports personal fees from Serinus Bio, Tango Therapeutics, Genome Medical, Genomic Life, Monte Rosa Therapeutics, Manifold Bio, Illumina, Enara Bio, Foaley & Hoag, and Riva Therapeutics; grants and personal fees from Novartis and Janssen; and grants from BMS and Sanofi outside the submitted work. In addition, E. Van Allen has a patent for Institutional patents filed on chromatin mutations and immunotherapy response, and methods for clinical interpretation pending and issued. T.S. Hong reports consulting for Synthetic Biologics, Novocure, Boston Scientific, Neogenomics, Merck, GSK, and NextCure; scientific advisory board member of PanTher Therapeutics (equity) and Lustgarten; and research funding (clinical trials) from Taiho, AstraZeneca, BMS, GSK, IntraOp, and Ipsen. Y. Liu reports personal fees from Veracyte during the conduct of the study, as well as personal fees from Veracyte outside the submitted work. E. Davicioni reports other support from Veracyte during the conduct of the study, as well as other support from Veracyte outside the submitted work; in addition, E. Davicioni has a patent for US201762448921 pending. E.A. Gibb reports other support from Veracyte during the conduct of the study, as well as other support from Veracyte outside the submitted work. K.W. Mouw reports personal fees from Riva Therapeutics, as well as personal fees from UroGen outside the submitted work. J.A. Efstathiou reports personal fees from Blue Earth Diagnostics, Boston Scientific, AstraZeneca, Merck, Roivant Pharma, Myovant Sciences, Janssen, Bayer Healthcare, Progenics Pharmaceuticals, Pfizer, Genentech, Gilead, Lantheus Medical Imaging, Angiodynamics, IBA, and Estellas outside the submitted work. D.T. Miyamoto reports grants from NCI, Radiation Oncology Institute, and George E. Safiol Foundation during the conduct of the study, as well as other support from Cardiff Oncology outside the submitted work. No disclosures were reported by the other authors.

S.C. Kamran: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. Y. Zhou: Formal analysis, investigation, methodology. K. Otani: Investigation. M. Drumm: Data curation, investigation. Y. Otani: Investigation. S. Wu: Investigation, writing–review and editing. C.-L. Wu: Investigation, writing–review and editing. A.S. Feldman: Investigation, writing–review and editing. M. Wszolek: Investigation, writing–review and editing. R.J. Lee: Investigation, writing–review and editing. P.J. Saylor: Investigation, writing–review and editing. J. Lennerz: Investigation, writing–review and editing. E. Van Allen: Investigation, writing–review and editing. H. Willers: Investigation, project administration, writing–review and editing. T.S. Hong: Funding acquisition, investigation, project administration, writing–review and editing. Y. Liu: Data curation, formal analysis, investigation, visualization. E. Davicioni: Data curation, software, supervision, investigation. E.A. Gibb: Data curation, software, investigation, visualization, writing–review and editing. W.U. Shipley: Funding acquisition, investigation, project administration. K.W. Mouw: Conceptualization, resources, formal analysis, supervision, investigation, project administration, writing–review and editing. J.A. Efstathiou: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing. D.T. Miyamoto: Conceptualization, resources, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We gratefully acknowledge Drs. David B. Solit and Sizhi P. Gao for providing the ERCC2-mutant KE1 and SC14 cell lines. This work was supported by grants from the NIH (C06 CA059267 to J.A. Efstathiou; U01CA220714 to H. Willers; R0125737 to K.W. Mouw; R01CA259007 to D.T. Miyamoto), the Radiation Oncology Institute (ROI2021–9151 to D.T. Miyamoto), and the George E. Safiol Foundation (to D.T. Miyamoto).

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.