Early Detection of Metastatic Relapse and Monitoring of Therapeutic Efficacy by Ultra-Deep Sequencing of Plasma Cell-Free DNA in Patients With Urothelial Bladder Carcinoma

Emil Christensen, PhD; Karin Birkenkamp-Demtröder, PhD; Himanshu Sethi, MPH; Svetlana Shchegrova, PhD; Raheleh Salari, PhD; Iver Nordentoft, PhD; Hsin-Ta Wu, PhD; Michael Knudsen, PhD; Philippe Lamy, PhD; Sia Viborg Lindskrog, BS; Ann Taber, MD; Mustafa Balcioglu, PhD; Søren Vang, PhD; Zoe Assaf, PhD; Shruti Sharma, PhD; Antony S. Tin, PhD; Ramya Srinivasan, MS; Dina Hafez, PhD; Thomas Reinert, PhD; Samantha Navarro, BS; Alexander Olson, BS; Rosalyn Ram, PhD; Scott Dashner, BS; Matthew Rabinowitz, PhD; Paul Billings, MD, PhD; Styrmir Sigurjonsson, PhD; Claus Lindbjerg Andersen, PhD; Ryan Swenerton, PhD; Alexey Aleshin, MD; Bernhard Zimmermann, PhD; Mads Agerbæk, MD; Cheng-Ho Jimmy Lin, MD, PhD, MHS; Jørgen Bjerggaard Jensen, MD, DMSc; and Lars Dyrskjøt, PhD


INTRODUCTION
Urothelial carcinoma is a common malignant disease, with 430,000 new cases diagnosed and 165,000 deaths recorded globally in 2012. 1 Localized, muscleinvasive bladder cancer (MIBC) is treated with radical cystectomy, but 20% of patients with node-negative and 80% with node-positive disease at surgery will experience metastatic relapse. 2 Neoadjuvant chemotherapy improves survival, 3 and treatment with gemcitabine and cisplatin is a commonly used regimen that results in significant downstaging in 40% to 50% of patients. 4,5 Currently, detection of relapse and monitoring of response to treatment in the metastatic setting is performed by standard computed tomography scan. Although imaging techniques offer an assessment of the tumor burden, the monitoring potential is restricted by a suboptimal detection limit and inherent variability in measurements. 6,7 Early detection of metastatic relapse and/or progression and evaluation of treatment efficacy, therefore, are major clinical challenges in this disease setting. Identification of metastatic relapse after cystectomy at an early time point, where relapse is not detectable by radiographic imaging, could aid in the selection of patients who may benefit from early/adjuvant treatment.
The use of circulating tumor DNA (ctDNA) as a biomarker for disease staging at diagnosis, tumor burden, early detection of metastatic relapse, disease surveillance, and therapeutic treatment response is an emerging field in multiple cancer types. [8][9][10][11][12][13][14][15] In bladder cancer, we and others have previously reported proof-of-concept data documenting that ctDNA is detectable in plasma and urine and that high levels of ctDNA are associated with later clinical disease progression and metastatic disease. [16][17][18][19] We therefore hypothesized that longitudinal analysis of ctDNA in patients with MIBC would demonstrate prognostic and predictive power at key time points and provide early evidence of metastatic disease.
To our knowledge, this report is the largest and most comprehensive study of ctDNA in patients with bladder cancer to date. We document that ctDNA is a powerful biomarker for prognosis and early detection of metastatic disease. Furthermore, we show that ctDNA dynamics during treatment is a predictor of chemotherapy response and patient outcome.

PATIENTS AND METHODS
Additional information can be found in the Data Supplement.

Patients and Clinical Samples
Ninety-nine patients diagnosed with MIBC and who were receiving neoadjuvant chemotherapy before cystectomy were prospectively enrolled between 2013 and 2017 at Aarhus University Hospital in Denmark. Treatment and surveillance were done according to Danish national guidelines, which adhere to the European Guidelines for patients with bladder cancer. 20 Blood samples were collected at uniformly scheduled clinical visits and before each chemotherapy cycle. Pathologic downstaging after chemotherapy was defined as Ta,CIS,N0 or less after treatment. Detailed follow-up data were available for all patients; clinical end points were obtained from computed tomography scan results (recurrence-free survival) and from the nationwide civil registry (overall survival). For details, see the Data Supplement. Sixty-eight patients were selected for exome sequencing and ctDNA analysis on the basis of the following criteria: neoadjuvant/first-line chemotherapy for localized MIBC; plasma samples obtained before and during chemotherapy, before and after cystectomy; and available DNA from a tumor biopsy. All patients provided written informed consent, and the study was approved by The National Committee on Health Research Ethics (#1302183).

Exome Sequencing and Bioinformatics Analysis
Libraries of tumor and matching germline DNA were prepared using 100 to 500 ng DNA and captured by SeqCap EZ MedExomeV1_hg19 or MedExomePlusV1_hg19 panel (Roche, Basel, Switzerland). Sequencing data were processed according to Genome Analysis Toolkit Best Practices, and single nucleotide variants and insertions and deletions were called using MuTect2 (Broad Institute, Cambridge, MA). Exome sequencing metrics are listed in the Data Supplement.

Plasma Multiplex Polymerase Chain Reaction Next-Generation Sequencing
For each patient, 16 patient-specific somatic variants were selected as previously described. 21 Cell-free DNA (cfDNA) was extracted from a median of 7.5 mL of plasma using QIAamp Circulating Nucleic Acid Kit (QIAGEN, Hilden, Germany). Libraries were created and sequenced as previously described. 14 Quality control was performed throughout the workflow (Data Supplement). In total, 651 (99%) of 656 plasma samples passed the sample quality control process. A plasma sample with at least two variants with a confidence score above a predefined algorithm threshold (0.97) was defined as ctDNA positive. Mutation calls from plasma samples are listed in the Data Supplement.

Statistical Analyses
Assessment of statistical significance was performed using Wilcoxon rank sum or Kruskal-Wallis test for continuous variables and Fisher's exact test for categorical variables. Survival analyses were carried out in R using packages survminer, survival, and coxphf (https://cran.r-project.org). Recurrence assessment was not available for patients 4519 and 3889, and these patients were excluded from analyses where recurrence status is considered. Furthermore, cystectomy was not completed for patients 4175 and 4250. Recurrence status after cystectomy was therefore not possible to evaluate, and these patients were similarly excluded. Recurrence rates 12 months after cystectomy were based on imaging data up to 14 months after cystectomy to allow for variability in scheduling of imaging.

Patient Characteristics and Primary Tumor Analysis
In total, 68 patients with localized MIBC fulfilled the inclusion criteria, with a median follow-up of 21 months after cystectomy. We observed a recurrence rate of 20% (n = 13) among the 64 patients with available recurrence evaluation. Whole-exome sequencing (WES) of tumor and matched germline DNA was performed at a mean target coverage of 1043 (range, 313 to 2513) for tumor samples and 663 (range, 353 to 1203) for germline samples, which identified an average of 488 mutations (range, 11 to 3,536 mutations) per patient (Data Supplement). A summary of mutation frequency, mutational signatures, frequently mutated genes, and clinical data is shown in Figure 1.

ctDNA Monitoring by Ultra-Deep Multiplex Polymerase Chain Reaction-Based Next-Generation Sequencing
A predefined and previously validated ctDNA analysis pipeline was applied to 656 plasma samples procured from the 68 patients. 14 In brief, unique patient-specific assays were designed for 16 highly ranked somatic mutations, and multiplex polymerase chain reaction next-generation sequencing was performed on plasma cfDNA. A sample was called ctDNA positive if two or more target variants were detected, as previously described. 14 Sample-level analytic sensitivity was previously determined to be greater than 95% at a 0.01% ctDNA concentration level. 25 Plasma samples were sequenced to a median target coverage of 105,0003 (error rate: transitions, 0.0063%; transversions, 0.0033%; Data Supplement).

ctDNA Detection for Prognosis and Relapse Detection
Throughout the disease courses, presence or absence of ctDNA was strongly correlated with patient outcomes (Fig 2; Figs 3A to 3C). Of note, ctDNA-positive samples obtained during follow-up and outside of treatment were generally followed by additional ctDNA-positive samples (Fig 2). However, at very-low ctDNA levels, we observed two  whole-exome sequencing. Of note, the tumor of patient 3857 was hypermutated with a mutational burden of 126 mutations/megabase (Mb) and displayed a POLD1 mutation that previously has been associated with hypermutators. 22   exceptions (patients 3731 and 4415; approximately 0.6 copies/mL plasma) and here, subsampling may influence ctDNA detection and repeatability. 26 Three time points are of significant interest. First, ctDNA status at diagnosis before chemotherapy was strongly prognostic (Fig 3A). In bladder cancer, the first intervention is transurethral resection of bladder tumor; plasma ctDNA status after transurethral resection of bladder tumor may serve as a proxy to measure minimal residual disease. For patients who were ctDNA positive at this time point, we observed overall and 12-month recurrence rates of 46% (11 of 24 patients) and 42% (10 of 24 patients), respectively. Of note, only 3% of patients (one of 35) who were ctDNA negative at this first time point experienced a recurrence during the study (P , .001; 12 months, 0% [zero of 35 patients; P , .001]). The detection of ctDNA at this early time point is therefore a strong prognostic factor for the long-term clinical outcome after chemotherapy and cystectomy (HR, 29.1; P = .001; Data Supplement).
The second time point (after chemotherapy and before cystectomy) was also prognostic of patient outcome ( Fig 3B). In ctDNA-positive patients, we observed an overall and 12-month recurrence rate of 75% (six of eight patients). In ctDNA-negative patients, the overall and 12month recurrence rates were 11% (six of 55 patients; P , .001) and 7% (four of 55 patients; P , .001), respectively. Presence of ctDNA before cystectomy was associated with pathology at cystectomy as 100% of ctDNA-positive patients at this time point had residual tumor (stage $ T1) and/or lymph node metastases identified at cystectomy (Fig 3D). Furthermore, 100% of patients (36 of 36) with pT0 at cystectomy were ctDNA negative. For this second time point, we observed an HR of 12.0 (P , .001; Data Supplement).
Third, and most significantly, plasma ctDNA status during disease surveillance after cystectomy was highly prognostic (Fig 3C). We observed an overall recurrence rate of 76% (13 of 17 patients) and a 12-month recurrence rate of 59% (10 of 17 patients) in ctDNA-positive patients. In ctDNAnegative patients, the recurrence rate was 0% at both time points (zero of 47 patients; P , .001). The status of ctDNA at any time point after cystectomy was stronger than any other predictive factor, such as lymph node status before cystectomy and pathologic downstaging (Figs 3E and 3F). In addition, in multivariable Cox proportional hazards regression analysis, ctDNA status was the strongest predictor of recurrence-free survival after cystectomy (HR, 129.6; P , .001; Data Supplement).

Serial ctDNA Measurements for Disease Surveillance
ctDNA dynamics (ie, changes in ctDNA levels measured in consecutive samples) and detection of relapse during disease courses are shown for selected patients in Fig 4A. For example, for patient 4251, ctDNA was detected 64 days after cystectomy, and clinical relapse was detected 309 days after cystectomy (lead time, 245 days). Similarly, for patient 4189, ctDNA was detected 273 days after cystectomy, and clinical relapse was detected 369 days after cystectomy (lead time, 96 days). Overall, for patients with metastatic relapse and detectable ctDNA, we found ctDNA analysis to have a median lead time of 96 days (283 to 245 days; P = .023) over conventional imaging (Fig 4B). Restriction of analyses to patients with simultaneous plasma and radiographic imaging identified eight patients, five of whom showed a lead time in recurrence detection for ctDNA analyses. The remaining three patients showed simultaneous recurrence detection, and the resulting median lead time for all eight patients was 107 days (0 to 186 days; P = .059). Disease courses and ctDNA detection and dynamics are shown in the Data Supplement for all 68 patients.
For evaluation of potential clinical performance of the ctDNA test, we calculated sensitivity and specificity measures by restricting our analysis to only include plasma samples where 180 days or more (approximately two times the median lead time) of follow-up was available for patients with nonmetastatic disease. Using these criteria, serial analysis of ctDNA during surveillance after cystectomy identified metastatic relapse with 100% (13 of 13 patients) sensitivity and 98% (48 of 49 patients) specificity.
To assess the impact of heterogeneity between primary tumors and metastases on serial ctDNA measurements, we performed WES of cfDNA in four plasma samples from three patients. Samples were sequenced to a mean target coverage of 3073 (2723 to 3403), and between 508 and 1,294 mutations were identified. We compared all mutations identified in the plasma WES data to the associated WES data from the primary tumor to assess mutational changes acquired during metastatic evolution (Fig 4C). We found a high degree of similarity between the mutational landscapes of the primary tumors and the cfDNA, which indicated a limited clonal evolution during the disease course of the selected patients. On average, we identified 62 mutations in ctDNA present at the time of metastases, which had not been detected in the primary tumors (0.5% to 61.2% increase in number of mutations compared with the primary tumors).  that pathologic downstaging is suboptimal for evaluating treatment efficacy. Our ctDNA results demonstrated that the presence and dynamics of ctDNA during chemotherapy were correlated to pathologic downstaging. In total, 85% of ctDNA-negative patients (35 of 41) showed pathologic downstaging. Patients who were initially ctDNA positive but with subsequent clearance of ctDNA (ie, ctDNA no longer measurable) showed a response rate of 53% (nine of 17 patients), whereas patients without clearance of ctDNA showed a response rate of 0% (zero of eight patients; Figs 5A and 5B). Of note, for patients who were ctDNA positive before or during treatment, the dynamics of ctDNA during chemotherapy was significantly associated with disease recurrence, whereas pathologic downstaging was not significantly associated with disease recurrence (Fig 5C), which indicates that ctDNA measurements may be a better tool for evaluating treatment efficacy. Overall, our results suggest that the presence of ctDNA identifies patients with a high risk of developing metastatic spread and that the dynamics of ctDNA during treatment further inform both chemotherapy response and outcome. Of note, although pathologic downstaging is a strong predictor of outcome, ctDNA informs chemotherapy response and outcome during treatment and before cystectomy.
We also assessed possible predictive biomarkers of treatment response in the primary tumors. Analysis of mutational processes 27 showed a significantly higher contribution of the trinucleotide mutational signature 5 (P = .024) in patients who responded to chemotherapy (Figs 1 and 5D). A high contribution of mutational signature 5 was significantly associated with ERCC2 mutation status (Fig 5E), which indicated a correlation to DNA damage response mechanisms as previously described. 28 Patients with ERCC2 mutations were associated with a higher rate of response to chemotherapy, although not significantly ( Fig 5F). Finally, transcriptional analysis of the tumors (n = 46) showed that molecular subtypes and immune signatures were not significantly associated with response to chemotherapy and ctDNA status (Data Supplement). In conclusion, clinical parameters and molecular features of the primary tumor were associated with treatment response and outcome, but ctDNA monitoring remained the strongest predictor of outcome and therapy response in high-risk patients (ctDNA positive; Data Supplement).

DISCUSSION
This study documents several important findings for ctDNA analysis for patients with bladder cancer: (i) ctDNA serves as a prognostic biomarker already before chemotherapy, (ii) ctDNA dynamics during chemotherapy reflect response to treatment and patient outcome, and (iii) ctDNA identifies disease recurrence in the postsurgery setting with high sensitivity and specificity and a positive lead time compared with radiographic imaging. On the basis of these findings, new paradigms for ctDNA-guided patient management should be investigated in future clinical trials. Suggestions for ctDNA-guided management concepts are presented in the Data Supplement. Patients who are ctDNA negative before chemotherapy seem to have a low risk of recurrence after cystectomy when treated with the current standard of care (recurrence rate only 3% in this study). Because of the low risk of micrometastatic spread, these patients may be eligible for cystectomy without neoadjuvant chemotherapy (Data Supplement). Patients who are ctDNA positive before chemotherapy seem to be at high risk of recurrence (recurrence rate in this study, 46%). ctDNA might be an indicator of early disease dissemination with micrometastases, and assessment of response to treatment in this patient group is therefore crucial. In our study, we observed ctDNA before or during chemotherapy in 43% of patients (27 of 63). For patients with clearance of ctDNA during treatment, we observed pathologic downstaging in 53%, whereas for patients without clearance of ctDNA, none were found to be downstaged. Of note, for patients who were ctDNA positive before or during chemotherapy, ctDNA dynamics during chemotherapy showed a superior association with patient outcome compared with pathologic downstaging. We therefore propose that ctDNA-positive patients be monitored using ctDNA analysis during chemotherapy to assess treatment efficacy. Patients with ctDNA clearance, which suggests responsiveness to chemotherapy, may be offered additional cycles of chemotherapy before cystectomy. For patients without clearance of ctDNA, the potential benefits of other therapeutic strategies can be explored (Data Supplement). 29 Detection of ctDNA after cystectomy serves as direct evidence of occult carcinoma cells and thus remnant disease. ctDNA was detected in 17 patients after cystectomy, and 13 of these were diagnosed with a recurrence. ctDNAbased recurrence detection displayed a lead time of up to 245 days (median, 96 days) compared with radiographic imaging. Of note, for three of four ctDNA-positive patients without a recurrence diagnosis, no follow-up was available after the positive blood test. The observed lead time in recurrence detection provides a window of opportunity for earlier initiation of therapy, which could improve treatment efficacy and thereby survival (Data Supplement). 30 Similar findings have been observed in other cancer types but with great variability in the observed lead times. 31 This might reflect a bias in the frequency of plasma sampling compared with imaging. In our study, however, we observed a similar lead time in recurrence detection when restricting our analyses to time points with simultaneous plasma sampling and imaging.
Earlier work has demonstrated ERCC2 mutations 32 and gene expression-based subtypes 33,34 to be predictors of chemotherapy response. Here, we observed limited predictive power associated with the tumor-centric biomarkers, and the ERCC2-related mutational signature was the strongest predictor of response in this work.
A low number of ctDNA molecules were detected in many samples (down to two molecules), which makes the case for highly sensitive and specific NGS-based methods. The selection of clonal mutations on the basis of WES of the primary tumor makes it possible to perform ultra-deep sequencing of the patient-specific mutations in plasma ctDNA; the disadvantage of not being able to detect novel mutations that arise during tumor evolution and disease dissemination exists. We performed deep WES of cfDNA from plasma for a subset of patients and observed heterogeneity between primary tumors and plasma at the time of metastatic relapse, but of note, all clonal mutations selected from the primary tumors were also detected in the plasma sample at relapse. Earlier work has shown high levels of genetic heterogeneity between primary tumors and metastases [35][36][37] ; however, our data document that genetic heterogeneity is not affecting assay performance when clonal mutations are selected.
In conclusion, we have found ctDNA testing in patients with bladder cancer who undergo chemotherapy and cystectomy to be highly sensitive and specific for early risk stratification of patients, prediction of treatment response, and early detection of metastatic relapse. ctDNA biomarkers are superior to tumor-centric biomarkers (mutations and subtypes) for predicting treatment efficacy, and novel randomized clinical trials should be initiated to determine the clinical impact of ctDNA-stratified therapeutic approaches.