Urinary Tumor DNA MRD Analysis to Identify Responders to Neoadjuvant Immunotherapy in Muscle-invasive Bladder Cancer

Muscle-invasive bladder cancer (MIBC) is a challenging disease with dismal prognosis. At present, standard of care for resectable MIBC is neoadjuvant chemotherapy followed by radical cystectomy (RC), which suffers from prevalent contraindications and high rates of tumor recurrence (1–3). Because immune checkpoint inhibitors (ICI) have shown durable efficacy and favorable side-effect profiles compared with cytotoxic drugs (4–7), perioperative immunotherapy is emerging as a promising treatment modality, especially in patients who are ineligible or opposed to receive cisplatin. Several ICIs, including pembrolizumab, atezolizumab, durvalumab, and nivolumab, are being evaluated as monotherapy or in combination in the neoadjuvant setting (8–10). However, established criteria are currently lacking to select individuals likely to benefit. A paradigm shift toward biomarker-guided neoadjuvant immunotherapy is particularly relevant to identify and monitor promising candidates for new therapeutic agents or bladder preservation strategies (11–13).

Toripalimab is a humanized IgG4 mAb against programmed cell death 1 (PD-1), which has been approved in China for second-line use in metastatic urothelial carcinoma based on our recently reported POLARIS-03 results (14). To further test the tolerability and effectiveness of toripalimab given to patients with MIBC before surgery, we conducted a single-arm investigator-initiated phase I clinical trial RJBLC-I2N003 (ChiCTR2000029500). This prospective study represented a unique window of opportunity to develop and evaluate novel biomarkers during the course of neoadjuvant immunotherapy.

RJBLC-I2N003, registered at https://www.chictr.org.cn/indexEN.html (ChiCTR2000029500), was an open-label, single-arm clinical trial to evaluate the safety and efficacy of toripalimab (anti-PD-1) as neoadjuvant therapy before RC in patients with MIBC. Eligible patients had a confirmed diagnosis of MIBC with clinical T2-T4N0M0 stage disease. Patients were staged on the basis of MRI, fluorodeoxyglucose PET-CT, and transurethral resection of bladder tumor (TURBT). All the enrolled patients had residual disease after TURBT and were scheduled for RC. Additional inclusion criteria included a predominant (i.e., at least 50%) urothelial carcinoma histology, ages ranging from 18 to 75 years, and Eastern Cooperative Oncology Group performance status of 0–1. Patients were enrolled regardless of their cisplatin eligibility. Key exclusion criteria included documented severe autoimmune or chronic infectious disease and use of systemic immunosuppressive medications.

Patients were treated with preoperative toripalimab at 3 mg/kg every 2 weeks for up to four cycles unless intolerable toxicity or voluntary retreat. RC was planned within 4 ± 2 weeks after the last dose of toripalimab treatment. The primary objectives of RJBLC-I2N003 were the efficacy and safety of toripalimab as neoadjuvant therapy of resectable MIBC. Secondary objectives included patient survival and perioperative complications. The primary efficacy endpoint was pathologic complete response (ypCR, defined by pT0N0) at surgical resection. Safety was assessed at each patient's clinical visit and documented as per the NCI Common Terminology Criteria for Adverse Events, v.4.03. MRI was performed at baseline and every two cycles of toripalimab treatment (please refer to Supplementary Materials and Methods for details). Radiographic images were evaluated by clinical investigators according to Response Evaluation Criteria in Solid Tumors (RECIST) v.1.1 (RECIST1.1). Clinical data were collected from June 23, 2020 through the censor date of February 21, 2022. The median follow-up time was 10.4 months (interquartile range, 8.0–13.2 months). The study was performed following Good Clinical Practice and the Declaration of Helsinki under protocols approved by the Ethics Committee of Ren Ji Hospital. All patients provided written informed consent before enrollment.

The PredicineBEACON personalized minimal residual disease (MRD) assay included whole-exome sequencing (WES) of baseline samples using either urine or tumor tissues collected from TURBT, followed by ultra-deep sequencing of subsequent longitudinal urine samples using a personalized MRD panel (personalized mutations plus a fixed panel of actionable/hotspot mutations). Matched peripheral blood mononuclear cell samples were sequenced to obtain high-confidence somatic mutation calls. Up to 50 somatic mutations were selected to design a personalized panel for each patient.

To detect a known variant selected for MRD tracking in the following timepoints, at least one fragment with confident variant support was required. An MRD variant without double-stranded fragment support was categorized as low confidence. To call a sample as MRD positive, one of the following criteria should be met: (i) three or more low-confidence MRD variants were detected, or (ii) two or more MRD variants were detected, and one of them had double-stranded variant support.

The tumor fraction (TF) was estimated according to the somatic mutations detected from the MRD assay ( |$T{F}_{sm}$|⁠) or the copy numbers detected from the low-pass whole-genome sequencing (LP-WGS) assay ( |$T{F}_{cn}$|⁠). Patients were deemed to achieve urinary tumor DNA (utDNA) response when TFsm + TFcn < 10% following toripalimab treatment.

Where TF_b is the tumor fraction of the matched baseline sample, |$i$| is the selected mutation site for MRD tracking, |$n$| is the total number of selected mutation sites for MRD tracking, |$m$| is the number of mutated fragments at the mutation site, and |$t$| is the total number of fragments at the mutation site; |${m}_b$| and |${t}_b$| are mutated and total fragments at the mutation site at the baseline level, respectively.

The |$T{F}_{cn}$| was estimated using ichorCNA as described previously (15). Briefly, LP-WGS with an overall average coverage of 5x was performed on patient samples. The ichorCNA algorithm was applied to GC and mappability-normalized reads to estimate copy-number variations using hidden Markov model. Then, the |$T{F}_{cn}$| prediction was performed using the ichorCNA R software package.

R version 4.0.0 (https://www.R-project.org) was used for statistical analysis and graphic plotting with ggplot2 (RRID:SCR_014601) and ComplexHeatmap (RRID:SCR_017270) packages (16). The R package pROC was used to perform ROC analysis. The optimal cut-off point was defined on the basis of the greatest Youden index (sensitivity + specificity − 1). The copy number G-score was calculated by the GISTIC (RRID:SCR_000151) 2.0 pipeline via GenePattern (RRID:SCR_003201, https://www.broadinstitute.org/cancer/software/genepattern). The Wilcoxon rank-sum test or Student t test was used to compare numeric variables. Fisher exact test was used to compare categorical variables. All tests were two sided and considered statistically significant at P < 0.05.

The analyzed sequencing data generated in this study are available within Supplementary Tables. The raw sequence data reported in this article have been deposited in the Genome Sequence Archive (17) in National Genomics Data Center (18), China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (GSA-Human: HRA005063) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human/browse/HRA005063.

A total of 20 patients (RZ1–20) with newly diagnosed T2–4N0M0 MIBC were recruited from January 2020 to January 2022 (Fig. 1A; Supplementary Table S1). The median age at enrollment was 62.5 years (range, 39–70 years), and 90% were male. The study participant demographics were considered to be representative for the general Chinese population (Supplementary Table S2). Nineteen patients (95%) were eligible to receive cisplatin chemotherapy. Among the 20 cases of urothelial bladder carcinoma, 5 (25%) had concomitant carcinoma in situ, and 4 (20%) had variant histology. One patient (5%) was ineligible to receive cisplatin chemotherapy. After a median interval of 4.0 weeks (interquartile range, 3.1–4.6 weeks) following TURBT, patients were treated with two to four cycles of toripalimab (3 mg/kg) every 2 weeks prior to RC (Fig. 1B). In general, safety signals were consistent with previous findings (Supplementary Table S3). Nineteen (95%) and 6 (30%) patients experienced treatment- and immune-related adverse events (AE), respectively. Grade 3 AEs occurred in 3 patients (15%), which caused toripalimab discontinuation in two cases. No grade 4 or 5 AE was observed. All 20 patients were successfully subjected to RC with a median interval of 5.3 weeks (interquartile range, 4.1–6.5 weeks) since last dose. Three patients (15%) had postponed operations due to toxicities (Fig. 2A). Upon toripalimab usage, most patients displayed extensive tumor regression at radiographic (Fig. 2B; Supplementary Fig. S1A) and histopathologic (Fig. 2C; Supplementary Fig. S1B) examination with few exceptions. Eight patients (40%) achieved ypCR and an additional 8 (40%) were downstaged to non-muscle invasive lesions (pT1 or less). One patient with locoregional progression while receiving toripalimab (RZ12) was able to undergo definitive resection. During a median postoperative follow-up time of 10.2 months (interquartile range, 7.6–13.1 months), only 1 patient (RZ17) relapsed with a urethral mass and all treated patients were alive. We concluded that the two coprimary endpoints of RJBLC-I2N003 including safety and efficacy were both met, recapitulating results from other ICI trials in the field (8, 9, 19).

The clinical advantages of preoperative immunotherapy must be weighed against the potential disadvantages of on-therapy tumor progression and AEs leading to surgical delay. Novel alternatives including targeted agents and antibody–drug conjugates have increasingly become viable options (20, 21). In addition, bladder preservation is associated with better quality of life and may be attempted in well-selected cases with complete eradication of malignant cells (3, 22). Collectively, these considerations emphasize the urgency to develop reliable biomarkers for active disease surveillance and accurate patient stratification throughout the course of neoadjuvant treatment. To this end, prespecified exploratory biomarker discovery was pursued in prospectively collected biospecimens. Measurement of traditional biomarkers at baseline, such as PD-L1 expression, tumor mutation burden, and tertiary lymphoid structures, failed to show predictive value (Supplementary Fig. S2A). Likewise, WES of therapy-naïve neoplastic tissues (Supplementary Table S4) using PredicineWES+, a WES assay with boosted coverage of 600 cancer-related genes, did not find a statistically significant correlation between specific somatic variants and patient outcome (Supplementary Fig. S2B). Moreover, radiologic decreases in tumor size measured by repeated MRI only exhibited marginal utility to discriminate pathologic responders (Fig. 2B; Supplementary Fig. S2C), indicating the pressing need for more robust biomarkers.

Accumulating evidence suggests that liquid biopsy is instrumental to predict and monitor ICI efficacy (23). Primary bladder cancer cells shed DNA directly into the urine, and we recently demonstrated that utDNA outperformed blood-based circulating tumor DNA (ctDNA) to serve as a dependable and precise surrogate of tumor-derived DNA (tDNA) in MIBC (24). In the current work, we assessed the potential utility of tissue-agnostic urinary MRD analysis to identify true responders to neoadjuvant immunotherapy. Baseline utDNA was subjected to PredicineWES+ for personalized MRD test design (Supplementary Table S5). Notably, the mutational landscapes derived from utDNA and tDNA profiling were highly similar (Supplementary Fig. S3A), as were the estimated tumor mutation burdens (Supplementary Fig. S3B). Given the possibly low abundance of utDNA after TURBT and toripalimab treatment, a PredicineBEACON MRD assay was devised that combined ultra-deep sequencing using a bespoke panel to profile up to 50 patient-specific somatic aberrations (Supplementary Table S6), targeted sequencing of a fixed set of 500 actionable/hotspot variants (Supplementary Table S7), and LP-WGS to detect tumor-originated copy-number changes. On the basis of a titration experiment using standard reference materials, PredicineBEACON reached 100% sensitivity with a variant allele frequency of greater than 0.005% and >99% specificity to call MRD-positive events (data not shown). We applied PredicineBEACON to the RJBLC-I2N003 cohort, and obtained TF estimates according to somatic mutations (TFsm) or copy numbers (TFcn; Supplementary Table S8). Consistent with our previous observations (24), utDNA showed superior performance than ctDNA to represent urothelial neoplasms, as reflected by evidently higher TFsm (Supplementary Fig. S4A). In fact, utDNA contained all the variants detected in ctDNA (Supplementary Fig. S4B). In addition, more genome-wide copy-number variations were identified in utDNA compared with ctDNA, as assessed by TFcn scores (Supplementary Fig. S4C) or the GISTIC algorithm (Supplementary Fig. S4D).

Focusing on quantitative metrics derived from the urine analyte, we found that TFsm was selectively diminished in toripalimab responders (Supplementary Fig. S5A), as was TFcn, albeit to a modest extent (Supplementary Fig. S5B), implicating utDNA reduction as a potential biological marker of tumor remission. At baseline, the values of area under the ROC curve (AUC) were 0.708, 0.760, and 0.531 for TFsm, TFcn, and MRI measurements, respectively, in predicting ypCR (Fig. 3A). Following completion of neoadjuvant therapy, the AUC values were 0.969, 0.854, and 0.729, respectively (Fig. 3B). Using the optimal cut-off points defined by ROC analysis, TFsm and TFcn levels in posttreatment utDNA, as compared with those in pretreatment utDNA (Supplementary Fig. S5C), were significantly correlated with pathologic outcome (Supplementary Fig. S5D), and were superior to correlates observed with MRI. Finally, we determined the MRD status of utDNA samples according to the PredicineBEACON test. All patients were inferred to be MRD-positive after TURBT, while 3 (RZ10, RZ15, and RZ20) became MRD-negative before bladder removal and invariably achieved ypCR at the end of the study (Fig. 3C; Supplementary Fig. S5E). These findings lay the foundation for tissue-agnostic urinary MRD assessment to identify exceptional responders to neoadjuvant checkpoint blockade who may be candidates for bladder preservation. Such a paradigm-shifting clinical approach will presumably alleviate the significant burden of unnecessary RC.

We reasoned that serial utDNA monitoring would aid to gauge the optimal duration of neoadjuvant immunotherapy. Indeed, utDNA kinetic profiles suggested that at least three to four cycles of presurgical toripalimab were likely required in most cases to yield dramatic decreases of both TFsm and TFcn (Supplementary Fig. S6A). Importantly, 7 of 17 (41%) MRD-positive patients actually showed utDNA response with minimal TFsm and TFcn (Fig. 3C; Supplementary Fig. S6B), and might need continued toripalimab to fully eradicate the remaining cancer cells. Of interest, utDNA response was observed in 8 of 8 (100%) of patients that achieved ypCR. Furthermore, to explore the possibility of treating MRD with alternative drugs, we investigated the dynamic changes of actionable variants included in PredicineBEACON. For example, the recurrent FGFR3 S249C mutation was identified in utDNA samples from patient RZ12 (Supplementary Fig. S6C), with increased mutant allele frequency observed following neoadjuvant toripalimab (Supplementary Fig. S6D), in keeping with on-treatment progressive disease. It was worth noting that FGFR3 S249C was not detected in bulky tissue analysis, implying that a subclonal lesion was responsible for tumor growth. These results implicated patient RZ12 as a candidate for subsequent FGFR-targeted therapy, such as erdafitinib. In total, 4 of 17 MRD-positive subjects (RZ01, RZ06, RZ11, RZ12) could possibly benefit from genotype-matched FDA-approved regimens targeting FGFR3 variants (Fig. 3C; Supplementary Table S7). Collectively, 12 of 20 patients in this study (60%, 12/20) were potential candidates for bladder preservation programs. Taken together, we propose that longitudinal urinary MRD analysis can be considered for prospective testing as an integrated biomarker for further development to allow for adaptive management of individual patients (Fig. 4).

Our data for the first time indicate that neoadjuvant toripalimab followed by RC is feasible and efficacious in patients with localized MIBC. Specifically, neoadjuvant toripalimab had a low incidence of immune-related AEs, and was not coupled with notable delays or complications in subsequent surgical procedures. The ypCR occurred in 40% of enrolled patients, in accordance with the latest clinical investigations of other ICIs in this setting (8, 9). Therefore, the RJBLC-I2N003 pilot study supports further exploration of neoadjuvant toripalimab in randomized controlled trials.

As a predefined objective in designing RJBLC-I2N003, the current proof-of-concept biomarker analyses lay unprecedented groundwork for the incorporation of actionable utDNA MRD testing into preoperative MIBC management in upcoming clinical trials and conceivably routine practice. Such a noninvasive approach is poised to not only facilitate individually tailored therapy selection, but also enable real-time surveillance of various treatments including the emerging neoadjuvant immunotherapy, targeted agents, and antibody–drug conjugates, among others. We envision that the tissue-agnostic, urine-based MRD assay described here holds enormous promise to inform clinical decision-making around organ-preserving opportunities and may fundamentally transform health care guidelines for patients with MIBC.

Limitations of the current study include the single-arm design with relatively small number of patients, despite that the liquid biopsy assays profile many longitudinal samples derived from these cases. In addition, we use ypCR at surgery as the efficacy endpoint in line with previous literature (8, 9), and long-term follow-up is ongoing to confirm the benefit of neoadjuvant toripalimab. It will also be important to evaluate the combination of ICIs with chemotherapy and other new therapeutics, as exemplified by KEYNOTE-866 and KEYNOTE-905/EV-303 (25). Finally, our envisagement of utDNA MRD-guided selective bladder preservation is of unknown clinical value, which warrants future larger prospective investigations.

Y. Zhang reports employment with Huidu Shanghai Medical Sciences Ltd. Z. Zhao reports employment with and stock ownership of Predicine. P. Du reports other support from Predicine outside the submitted work. S. Jia reports employment with and stock ownership of Predicine and Huidu Shanghai Medical Sciences Ltd. No disclosures were reported by the other authors.

R. Zhang: Conceptualization, data curation, funding acquisition. J. Zang: Formal analysis, investigation, visualization, writing–original draft. D. Jin: Conceptualization, data curation. F. Xie: Software, formal analysis, methodology. A. Shahatiaili: Data curation. G. Wu: Data curation, methodology. L. Zhang: Data curation. L. Wang: Data curation. Y. Zhang: Software, methodology. Z. Zhao: Software, methodology. P. Du: Software, methodology. S. Jia: Conceptualization, supervision, methodology, writing–review and editing. J. Fan: Supervision. G. Zhuang: Conceptualization, supervision, funding acquisition, investigation, writing–review and editing. H. Chen: Conceptualization, supervision, funding acquisition, investigation, writing–review and editing.

This work was supported by the National Natural Science Foundation of China (82172596, to G Zhuang; 82173076, to H. Chen; 81902562, to R. Zhang), Shanghai Natural Science Foundation (20Y11904900, to H. Chen), Shanghai Municipal Education Commission-Gaofeng Clinical Medicine Grant Support (20161313, to G Zhuang), Shanghai Hospital Development Center (SHDC2022CRD034, to H. Chen), Ren Ji Hospital Research Funding Projects (PYII20-07, to H. Chen; RJPY-LX-001, to R. Zhang), Collaborative Innovation Center for Clinical and Translational Science by Ministry of Education & Shanghai (CCTS-2022203, to G. Zhuang), the Fundamental Research Funds for the Central Universities (YG2023LC03, to H. Chen), innovative research team of high-level local universities in Shanghai (SHSMU-ZLCX20210200, to G. Zhuang), and 111project (no. B21024, to G. Zhuang).