Neoadjuvant Pembrolizumab and Radical Cystectomy in Patients with Muscle-Invasive Urothelial Bladder Cancer: 3-Year Median Follow-Up Update of PURE-01 Trial

Muscle-invasive urothelial bladder carcinoma (MIBC) is a deadly disease, for which historical estimates of disease-free survival (DFS) from radical cystectomy (RC) series approximate 50% at 5 years (1). After several decades where no new therapeutic strategies emerged, recent clinical research in the field has led to the approval of newer therapeutic options across various disease stages, and an increasing number of trials at all phases are being offered to patients with MIBC (2, 3). In 2017, we pioneered the use of neoadjuvant immune-checkpoint inhibitors (ICI) in a clinical trial (PURE-01, NCT02736266) testing pembrolizumab as single-agent neoadjuvant therapy in patients with T2–4N0M0 MIBC (4). The study resulted in a proportion of 42% pathological complete responses (ypT0N0). In addition, early-phase single-arm trials of neoadjuvant ICI, as single-agents or combination immunotherapies before RC, reported a proportion of ypT0N0 ranging from 7% to 46% (4–10), but putative predictive biomarkers of response were still inconsistent across trials. Because the initial reports of our study, we have learned that tumors with the highest pre-treatment immune-gene signature scores, those with an expression of programmed-cell-death-ligand-1 (PD-L1), and those harboring the highest tumor mutational burden (TMB) levels yielded the highest probability of benefiting with a ypT0N0 response (11, 12). In addition, neoadjuvant pembrolizumab was medically and surgically safe and did not expose the patients to an excess risk of adverse events (AE) compared with RC alone or neoadjuvant chemotherapy and RC (13). Subsequently, we reported two follow-up updates consolidating the initial findings in terms of ypT0N0 of tumor downstaging proportions (12, 14). In addition, immature event-free survival (EFS) and relapse-free survival (RFS) results suggested the possibility for a considerable proportion of patients to maintain a disease-free status post RC regardless of the pathological response (12). Pending phase III trial results, a critical limitation of the available data is represented by the lack of validation of intermediate endpoints, like ypT0N0, as a surrogate of survival endpoints after ICI.

As a follow-up trial for validation and potential registration purposes, the Keynote-905 trial is currently enrolling patients who are unsuited for cisplatin chemotherapy, randomized toward pembrolizumab monotherapy, or its combination with enfortumab vedotin, as perioperative therapy against upfront RC, and the results are highly awaited (NCT03924895; ref. 15). To date, PURE-01 study yields the longest follow-up duration among the single-agent neoadjuvant ICI trials; therefore, we present the updated survival analysis to highlight stronger outcomes estimates and predictors of sustained clinical benefit.

Patients with a diagnosis of MIBC and eligible for RC were included. Additional inclusion criteria were the presence of predominant urothelial carcinoma histology (UC; subsequently removed in a study amendment to allow inclusion of pure variant histologies) and an Eastern Cooperative Oncology Group performance status (ECOG-PS) ≤2. Patients were enrolled regardless of cisplatin eligibility. The preoperative staging was performed with a thorax-abdomen contrast-enhanced CT scan, a fluorodeoxyglucose PET (FDG-PET) scan, and a multiparametric magnetic resonance imaging (mpMRI) of the bladder. Patients received three courses of pembrolizumab 200 mg, intravenously, every 3 weeks, followed by restaging via the same assessments and by RC. Patients developing a treatment-limiting AE, or those with a radiological non-response to treatment based on the investigator's decision, could receive additional standard chemotherapy of protocol. Post-RC management followed European Association of Urology guidelines regarding indications to receive adjuvant chemotherapy as well as the timing and type of follow-up assessments (1). The primary endpoint was the ypT0N0 proportion in intention-to-treat (ITT) population. The protocol updates in sample size and statistical assumptions have been already reported. Five more patients were enrolled from the last updated report (12); thus, the final study cohort consisted of 155 patients, enrolled between February 2017 and July 2020. The study was approved by local Institutional Review Board and ethical committee. Written informed consent for study participation was obtained from all patients, and the study was conducted in accordance with the Declaration of Helsinki principles.

Study outcomes and their assessment have been previously described (4, 12, 14). The focus of the present analyses was on EFS, representing a secondary endpoint of the trial. EFS was defined as the time from the first pembrolizumab dose to either: (i) radiographic disease progression precluding a curative intent surgery per RECIST v1.1 before RC, (ii) initiation of neoadjuvant chemotherapy preceding RC as per investigator decision, (iii) inability to undergo RC due to the onset of treatment-related side effects, (iv) inability to complete a curative intent surgery determined by the urologist at the time of RC (e.g., unresectable tumor, metastases discovered at RC), (v) local or distant recurrence assessed by cross-sectional imaging and/or biopsy after RC, and (vi) death from any cause. Additional analyses were made on RFS in the subgroup of patients who had received pembrolizumab and RC without additional chemotherapy, along with overall survival (OS) in the ITT cohort. The inverse Kaplan–Meier method calculated the median follow-up duration. Procedures for PD-L1 combined positive score (CPS) assessment, TMB and next-generation sequencing assessments, along with molecular subtype classification, all of these biomarkers being assessed in the pre-treatment transurethral resection of the bladder tumor (TURBT) samples, have been previously described (4, 11).

The Kaplan–Meier method was used to estimate the EFS, RFS and OS. RFS curves were split by pathological response (ypT0N0, ypT_1/a/isN0, ypT2–4N0, and ypT_anyN1–3). For EFS and OS analyses, the 36-month rates were calculated from the date of administration of the first dose of pembrolizumab to the occurrence of an event or last follow-up. For the RFS analyses, the 36-month rates were calculated from the date of RC to the date of recurrence, death, or last follow-up. Tertile split was used to visualize the Kaplan–Meier curves according to both the PD-L1 CPS and the TMB from TURBT. The tested predictors in the univariable Cox regression analyses for EFS included age, gender, a previous clinical history of non-muscle invasive bladder cancer, previous instillations of Bacillus Calmette–Guérin, histology, TMB, CPS, and clinical T-stage. For the multivariable model, due to the limited number of events, pre-specified factors were used (CPS, TMB, and clinical T-stage). All statistical tests were two-sided with a level of significance set at P < 0.05. Analyses were carried out using the R software (version 3.6.1, R Foundation for Statistical Computing).

Raw data for biomarkers of this study were generated at Foundation Medicine Inc. and Veracyte Inc. Derived data supporting the findings of this study are available from the corresponding author upon request.

The median follow-up duration was 39 months (interquartile range, 30–47). The study flow-chart is provided in Supplementary Fig. S1. A total of 134 patients (86.4%) received pembrolizumab and RC without any further perioperative treatment, whereas 12 did not receive RC because of systemic progression or patient refusal. A total of 47 patients (30.3%) developed treatment-related AE (TRAE), with grade 3–4 TRAE observed in 7 patients (4.5%). None of the patients experienced RC-preventing TRAE. The granular distribution of updated safety results is provided in Supplementary Table S1, whereas the updated surgical safety outcomes are depicted in Supplementary Table S2. Eight patients refused RC after the evidence of radiological response on pembrolizumab (N = 7) or pembrolizumab and additional neoadjuvant chemotherapy (N = 1) and were submitted to re-TURBT and surveillance: Six revealed a T0, one a Ta, and one a T2 response, respectively. To date none of them have relapsed. Overall, 14 patients received additional systemic therapy post-pembrolizumab, before RC: Nine of them underwent RC, and three achieved a major pathological response to ypT<2N0. Nine patients received adjuvant cisplatin-based chemotherapy after RC. The remaining cohort characteristics are reported in Table 1. Representativeness of study participants is provided in Supplementary Table S3. At the time of data analyses, 57 patients (39.8%) were treated with pembrolizumab and RC and achieved a ypT0N0 response, and 83 patients (53.5%) achieved a pathologic downstaging to ypT<2N0.

Overall, the 36-month EFS was 74.5% [95% confidence interval (CI), 67.8–81.7; Fig. 1A], and the 36-month OS was 83.8% (95% CI, 77.8–90.2; Fig. 1B). Within the subgroup of patients who did not receive chemotherapy (N = 125), 36-month RFS was 96.3% (95% CI, 91.6–100) for ypT0N0, 96.1% (95% CI, 89–100) for ypT_1/a/is N0, 74.9% (95% CI, 60.2–93) for ypT2–4N0, and 58.3% (95% CI, 36.2–94.1) for ypT_any N1–3 (P = 0.0001; Fig. 2). Univariable Cox regression analyses on EFS showed that higher CPS [hazard ratio (HR), 0.96; 95% CI, 0.95–0.98; P = 0.001] was associated with lower rates of events, whereas clinical stage T3–4 (HR, 2.50; 95% CI, 1.20–5.17; P = 0.01) was associated with higher rates of events. CPS (HR, 0.97; 95% CI, 0.95–0.99; P = 0.003) and clinical stage T3–4 (HR, 2.20; 95% CI, 1.09–4.45; P = 0.03) remained associated with EFS at multivariable analyses (Table 2).

In addition, 36-month EFS was 59.7% (95% CI, 47.5–75.1), 76.7% (95% CI, 66.0–89.2), and 89.8% (95% CI, 81.7–98.6) in patients with low (N = 51), intermediate (N = 52), and high (N = 49) PD-L1 CPS, respectively (P = 0.001; Fig. 3). Kaplan–Meier EFS curves according to baseline TMB are shown in Supplementary Fig. S2: 36-month EFS was 71.8% (95% CI, 60.3–85.5), 63.0% (95% CI, 50.8–78.2) and 87.3% (95% CI, 78.4–97.3) in patients with low (N = 50), intermediate (N = 49), and high (N = 49) TMB, respectively (P = 0.001). In the biomarker-evaluable population, molecular subtyping was used to classify the PURE-01 cohort into basal squamous (Ba/Sq, N = 38), luminal nonspecified (N = 17), luminal papillary (LumP; N = 16), luminal unstable (LumU; N = 20), stroma-rich (N = 8), and NE-like (N = 3) subtypes according to the Consensus model, with similar patterns for The Cancer Genome Atlas (TCGA) and the Genomic Subtyping Classifier (GSC) models (16–18). Claudin-low tumors, being consistently classified as a “basal squamous” or “Ba/Sq” tumors by the TCGA and Consensus classifiers, respectively (Supplementary Tables S4 and S5), had the best RFS outcome and NE-like the worst one, confirming our previous findings (12). In particular, the claudin-low (GSC) subtype showed exceptional RFS rates, with one event out of 14 patients at 36-month after pembrolizumab. Likewise, basal (GSC) and basal/squamous (Consensus and TCGA) patients showed improved RFS versus the other subtypes (Supplementary Fig. S3). In addition, Supplementary Fig. S4 illustrates the range of PD-L1 staining (CPS %) within each subtype for the GSC, Consensus and TCGA models.

Although the follow-up is maturing, the PURE-01 study steadily confirms that a substantial proportion of patients presenting with nonmetastatic MIBC can be safely managed with ICI monotherapy preoperatively. Compared with the original publication (4), very few additional AE were recorded, and all the listed types of events are currently below the 10% frequency. These data reinforce the reliability of strategies that are aimed to overcome the limitations of cisplatin-based chemotherapy. Such limitations are also highlighted by the initial results reported to date by the phase II trials testing chemo-immunotherapy combinations in MIBC. In these trials, a proportion varying from 7% to 45.2% ypT0N0 responses has been reported (19–25), therefore, not substantially different from neither the proportions of ypT0N0 reported in immunotherapy trials (4–12) nor the modern estimates of ypT0N0 responses recently reported with standard chemotherapy from prospective trials like COXEN and VESPER (26, 27). However, when looking at the amplitude of ypT0N0 results from neoadjuvant ICI trials, it becomes evident that a biomarker selection is needed to consider a potential translation of this strategy into standard practice. In this regard, we acknowledge a consistent increase in ypT0N0 proportions obtained with ICI from different studies in the subgroup of PD-L1–expressing tumors (4–12). Despite the inherent limitations related to the identification of optimal cutoff points for biomarkers analyzed as continuous variables, in our study we were able to identify a significant increase in EFS in patients with the highest CPS levels. The tertile of patients with the highest CPS values revealed a 3-year EFS of 90%, a result that favorably compares with RC outcomes, at least in the population of patients who are ineligible to receive standard-of-care chemotherapy. In addition, there is a long-dated discussion in perioperative trials regarding the role of intermediate endpoints like DFS or RFS as surrogate of OS, and about the magnitude of DFS increase over the benchmark results that would be considered worth of potential change in clinical practice.

In PURE-01 study, we relied on the EFS as an endpoint that could best include the type of events that may occur in the preoperative setting. These events include patient refusal to undergo RC, that is, an emerging issue for patients who are found to have achieved a clinical response to treatment, and the possibility for the investigators to predict the patients as being nonresponders via radiological assessments, thereby administering sequential chemotherapy with the aim to reconvert them to clinical responders. This conduct yields the advantages to minimize the risk of losing patients because of disease progression preventing RC, and may inspire newer consolidation or de-escalation strategies for patients who are complete responders and may spare the bladder any radical local therapy.

In general, a few important messages have been delivered by this update. First, we can cure a substantial proportion of patients with MIBC with a short and well-tolerated therapy course, without compromising their possibilities of receiving the standard treatments. Second, we can reach EFS probabilities that are quite compelling and certainly deserve validation by selecting those patients whose tumors reveals high levels of PD-L1 expression, higher than the 10% CPS cutoff values identified in metastatic disease (28). CPS (analyzed as a continuous value), together with an increasing clinical T-stage, was also independently associated with EFS on multivariable analyses.

Third, the standard selection of patients for neoadjuvant chemotherapy according to their eligibility of cisplatin chemotherapy could be revisited. In our study, in which pembrolizumab was offered to all patients regardless of cisplatin eligibility, we did not observe cases for whom a potential advantage in chemotherapy administration was overtly prevented by pembrolizumab administration, either in terms of pathological responses or as regards EFS outcomes (and immature OS). Therefore, preventing a population of about 50% of patients diagnosed with MIBC to receive a chemotherapy-free regimen within a clinical trial does not seem to be appropriate. In fact, following the PURE-01 findings, next-generation trials like the Keynote-B15 (NCT04700124) are testing pembrolizumab administration neoadjuvantly in combination with enfortumab vedotin, an antibody–drug conjugate targeting Nectin-4, versus standard-of-care chemotherapy, in cisplatin-eligible patients (29). Pending validation of the predictive role of PD-L1 expression toward pembrolizumab efficacy, one further step could consist of overcoming the limitations in the use of cisplatin ineligibility to test chemotherapy-free combinations by comparing a pembrolizumab monotherapy against standard-of-care chemotherapy in patients with high CPS tumors. Fourth, the pathological response categories, originally developed in patient populations treated with RC alone or preceded by cisplatin-based chemotherapy, seem to be also able to identify discrete prognostic categories in the setting of neoadjuvant ICI. Although concerns still remain about the role of pathological response as an intermediate endpoint in ICI trials, our data are similar to that of ABACUS study, a study testing two courses of atezolizumab neoadjuvantly before RC, reporting a ypT0N0 of 31% and an associated 2-year DFS of 85% in such patients, discriminating them from the nonresponders (5). Finally, besides the PD-L1 expression, putative tumor biomarkers for patient selection have been thoroughly investigated. A continuous-scale analysis of TMB showed an overall positive association between increasing TMB values and EFS, and the results favored those patients yielding the highest TMB values. However, the association was weaker in multivariable analyses compared with CPS, and there were crossing Kaplan–Meier curves of EFS between TMB intermediate and low tertiles. This association expands from previous findings, reporting conflicting results regarding the association between TMB and OS in patients with advanced-stage disease receiving ICI within clinical trials or as real-world practice (30–33). In addition, following our primary focus on pre- and post-therapy tumor biomarkers, we were able to confirm the initial findings associating the baseline GSC claudin-low subtype with the best RFS outcomes, whereas NE-like tumors had the worst RFS rates according to Consensus classification, TCGA, and Decipher classification. NE-like tumor behavior of the PURE-01 cohort was consistent with those reported for this subtype in other MIBC cohorts. Nevertheless, the RFS difference between molecular subtypes was statistically significant only for Decipher/GSC classification. Apart from the inherent biases related to tumor heterogeneity and sampling biases of the TURBT, the major limitation is still represented by the lack of a prospective comparison with the outcomes in cohorts of patients treated with standard chemotherapy, as overlapping effect of ICI and chemotherapy is anticipated, for example, in the population of patients with a basal subtype MIBC. The integration of tumor biomarker in composite models (e.g., integration of CPS and molecular subtypes) is a major goal of follow-up clinical research, deserving larger sample size, higher power for statistical analyses and the potential for introducing artificial intelligence and machine learning into the data analysis strategy.

In conclusion, the present results further support the strategy of single-agent pembrolizumab as a reliable and efficacious treatment preceding RC in MIBC. They also confirm that we can build upon the original study design, aiming at sequencing new therapies or introducing new strategies of bladder preservation in patients who achieve a clinical response. These results also strengthen the role of PD-L1 expression and molecular subtypes as putative biomarkers of EFS and RFS, possibly anticipating the results from ongoing randomized trials like the Keynote-905 study.

E.A. Gibb reports employment at Veracyte, Inc. J.S. Ross reports personal fees from Foundation Medicine during the conduct of the study. A. Necchi reports grants and personal fees from Merck and AstraZeneca, and personal fees from Roche, Janssen, BMS, Astellas, Bayer, Incyte, and Seattle Genetics during the conduct of the study; and grants and personal fees from Merck, AstraZeneca, and BMS, and personal fees from Astellas, Bayer, Incyte, Janssen, Roche, and Seattle Genetics outside the submitted work. No disclosures were reported by the other authors.

G. Basile: Writing–original draft. M. Bandini: Writing–original draft. E.A. Gibb: Resources, formal analysis, writing–original draft. J.S. Ross: Data curation, writing–review and editing. D. Raggi: Formal analysis. L. Marandino: Formal analysis. T. Costa de Padua: Data curation, writing–review and editing. E. Crupi: Data curation, writing–review and editing. R. Colombo: Formal analysis. M. Colecchia: Data curation. R. Lucianò: Data curation. L. Nocera: Data curation. M. Moschini: Data curation. A. Briganti: Data curation. F. Montorsi: Data curation. A. Necchi: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This work was supported by the Associazione Italiana per la Ricerca sul Cancro (AIRC), grant number: MFAG: 2017 Id.20617. This work was supported by Merck & Co., Inc. (Kenilworth, NJ).

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.