Personalized Circulating Tumor DNA Analysis for Predicting Outcomes and Tracking Response to Radiotherapy and Pembrolizumab in Localized Sarcomas: Analysis of the SU2C-SARC032 Trial Authors: Ajay Subramanian, Serey CL Nouth, Neda Nemat-Gorgani, Shaghayegh Soudi, Karla V Ballman, Rachel S Heise, Claire Johns, Timothy J Sears, Siyer Roohani, Reinhardt Krcek, Angela M Hong, Kent J Weinhold, Matt van de Rijn, Brian E Brigman, Richard F Riedel, Yvonne M Mowery, David G Kirsch, Everett J Moding Published: 2026 ABSTRACT AbstractPURPOSECirculating tumor DNA (ctDNA) holds promise for prognostication and disease monitoring in solid tumors. However, rare and heterogenous tumors with low mutational burden like soft tissue sarcomas (STS) present a technical challenge for ctDNA analysis. We investigated the ability of personalized ctDNA analysis to predict outcomes and track response to radiotherapy and pembrolizumab in patients with localized STS.METHODSWe analyzed 313 plasma samples collected pretreatment, postradiotherapy/presurgery, and postsurgery from 106 patients treated with preoperative radiotherapy with or without perioperative pembrolizumab for high-risk, localized STS in the SU2C-SARC032 phase II randomized trial. Personalized ctDNA assays were designed to track a median of 46 somatic mutations per patient. Associations between ctDNA levels and outcomes were assessed using univariable and multivariable models adjusting for clinical risk factors.RESULTSPersonalized ctDNA profiling detected baseline ctDNA in 85% of patients. Baseline ctDNA levels correlated with tumor size, grade 3 disease, and expression of genes related to hypoxia and the cell cycle. ctDNA levels pretreatment, after radiotherapy with or without immunotherapy but before surgery, and postsurgery were strongly associated with disease-free and overall survival on univariable and multivariable analyses. Patients treated with pembrolizumab, who had significantly better disease-free survival, had a greater decrease in their ctDNA levels after neoadjuvant therapy. Finally, personalized ctDNA analysis enabled noninvasive subclone tracking and patients with multiple tumor clones detected by ctDNA analysis after surgery had inferior outcomes.CONCLUSIONTumor-informed, personalized profiling enables sensitive ctDNA detection and quantitative monitoring in localized STS, providing independent prognostic information across the perioperative course. These findings support prospective ctDNA-guided interventional trials in STS. INTRODUCTION Half of patients with high-risk, localized soft tissue sarcomas (STS) develop metastatic disease,1 but the role of neoadjuvant or adjuvant chemotherapy is controversial because of its toxicity and limited efficacy.2 Recently, the SU2C-SARC032 phase II randomized trial demonstrated that adding perioperative pembrolizumab to neoadjuvant radiotherapy and surgery significantly improved disease-free survival (DFS) in patients with stage III undifferentiated pleomorphic sarcoma, myxofibrosarcoma, or dedifferentiated/pleomorphic liposarcoma of the extremity or limb girdle.3 However, pembrolizumab increased grade 3 adverse events and approximately one third of patients developed recurrent disease despite receiving pembrolizumab. CONTEXT Key Objective Key Objective Can personalized circulating tumor DNA (ctDNA) analysis predict outcomes and monitor response to therapy in localized soft tissue sarcomas (STS)? Can personalized circulating tumor DNA (ctDNA) analysis predict outcomes and monitor response to therapy in localized soft tissue sarcomas (STS)? Knowledge Generated Knowledge Generated In patients with stage III STS treated with preoperative radiotherapy with or without perioperative pembrolizumab in the SU2C-SARC032 clinical trial, personalized ctDNA analysis detected baseline ctDNA in 85% of patients. Levels of ctDNA throughout the perioperative treatment course were independently associated with patient outcomes. In patients with stage III STS treated with preoperative radiotherapy with or without perioperative pembrolizumab in the SU2C-SARC032 clinical trial, personalized ctDNA analysis detected baseline ctDNA in 85% of patients. Levels of ctDNA throughout the perioperative treatment course were independently associated with patient outcomes. Relevance (R.G. Maki) Relevance (R.G. Maki) The strength of longitudinal ctDNA testing in a population of primary undifferentiated pleomorphic sarcoma (UPS) patients is highlighted in this work. These data provide a starting point for use of ctDNA as a biomarker to better personalize use of systemic therapy in sarcoma patients.* The strength of longitudinal ctDNA testing in a population of primary undifferentiated pleomorphic sarcoma (UPS) patients is highlighted in this work. These data provide a starting point for use of ctDNA as a biomarker to better personalize use of systemic therapy in sarcoma patients.* *Relevance section written by JCO Associate Editor Robert G. Maki, MD, PhD, FACP, FASCO. *Relevance section written by JCO Associate Editor Robert G. Maki, MD, PhD, FACP, FASCO. Evidence continues to accumulate that circulating tumor DNA (ctDNA) can monitor response to systemic therapy and that detection of ctDNA minimal residual disease (MRD) after curative-intent therapy is highly prognostic for disease recurrence across multiple tumor types.4,5 Recent prospective clinical trials have demonstrated that ctDNA-guided adjuvant therapy is feasible6 and patients with detectable ctDNA after surgery have improved DFS and overall survival (OS) when treated with adjuvant immunotherapy.7 However, cancers like STS represent a technical challenge for ctDNA analysis because of their lower mutational burden with few recurrent single-nucleotide variants.8,9 Indeed, prior ctDNA studies in STS using fixed panels and/or low-pass whole-genome sequencing have achieved low detection rates,10 limiting the ability to monitor response to treatment or guide adjuvant therapy. Here, we investigated the ability of personalized ctDNA analysis using bespoke panels to improve ctDNA monitoring and outcome prediction in localized STS by analyzing tumor and blood samples from patients treated in the SU2C-SARC032 clinical trial.3 METHODS Patients and Study Design All human research in this study was conducted in accordance with the Declaration of Helsinki. The SU2C-SARC032 study protocol was reviewed and approved by SARC, Merck Sharp & Dohme, and the institutional review board at each institution participating in the trial. The correlative analyses were approved by the institutional review board at Stanford University. Written informed consent was obtained for all patient data and samples obtained and analyzed in this study. SU2C-SARC032 was an open-label, randomized, superiority phase II clinical trial that enrolled patients at 20 academic institutions in the United States, Canada, Australia, and Italy. SU2C-SARC032 was registered with ClinicalTrials.gov identifier: NCT03092323 on March 27, 2017. Patients with stage III (>5 cm and grade 2 or 3) undifferentiated pleomorphic sarcoma (including myxofibrosarcoma) or dedifferentiated/pleomorphic liposarcoma of the extremity were randomly assigned to receive neoadjuvant radiotherapy followed by surgery (control group) or 200 mg intravenous pembrolizumab once every three weeks for three doses before, during, and after neoadjuvant radiotherapy followed by surgery and up to 14 doses of adjuvant pembrolizumab (experimental group). Follow-up data were updated on February 24, 2025, for the correlative analyses. ctDNA Analysis Peripheral venous blood was collected in Cell-Free DNA Blood Collection Tubes (Streck, La Vista, NE) and centrifuged to isolate plasma and peripheral blood leukocytes pretreatment at the time of study enrollment, 1 week after completing radiotherapy before surgery, and at 3 and 12 months after surgery. Personalized hybrid capture panels were designed for ctDNA analysis based on whole-exome sequencing of each patient's tumor and matched normal DNA as described previously.11 Cell-free DNA was isolated from each plasma sample and prepared for next-generation sequencing using custom sequencing adapters with dual-index sample barcodes for demultiplexing and unique molecular identifiers.12 ctDNA analysis was performed using CAPP-Seq with error suppression incorporating the unique molecular identifiers and in silico background polishing.13 Subclonal mutational clusters were identified using PyClone-VI.14 The presence of tumor clones in the cell-free DNA was determined by monitoring for all variants assigned to each clone. Differential gene expression analysis and gene set enrichment analysis for patients with and without ctDNA detected pretreatment were performed using bulk RNA sequencing of pretreatment tumor samples described previously.15 Statistical Analysis SU2C-SARC032 was powered for the primary end point of DFS, and no power was assigned for secondary or exploratory analyses. Only evaluable patients were included in exploratory analyses, and patients with missing data were excluded. DFS and OS were calculated using landmark analysis to protect against guarantee-time bias, and patients with an event more than 7 days before the time of sample collection were excluded. No threshold for significance was prespecified because of the exploratory nature of the correlative analyses, and nominal P values were reported. Additional details are available in the Supplementary Methods (online only). RESULTS Association of Pretreatment ctDNA Levels With Sarcoma Biology and Prognosis In SU2C-SARC032, blood samples were collected for ctDNA analysis pretreatment, after radiotherapy with (experimental) or without (control) pembrolizumab but before surgery, and at 3 months and 12 months postsurgery (Figs 1 and 2A, Appendix Fig A1A [online only], Data Supplement, Table S1). Of 127 evaluable patients, 106 patients (83%) were included in the ctDNA analysis. The ctDNA cohort had similar demographics and outcomes to all patients enrolled (Data Supplement, Table S2, Appendix Figs A1B and A1C). The median follow-up for the ctDNA cohort was 50.7 months for DFS and 50.2 months for OS, and there were 51 DFS and 28 OS events observed. We performed whole-exome sequencing of tumor and germline DNA to design personalized hybrid capture panels to track all single-nucleotide variants (SNVs) from each patient using ctDNA analysis. Consistent with prior reports,8 TP53, ATRX, and RB1 were among the most commonly mutated genes (Fig 2B) and the distribution of mutations was similar between treatment arms (Data Supplement, Table S3). Although there were few recurrently mutated genes, we tracked a median of 46 SNVs from each patient (range, 2-2,372). Patient disposition. CONSORT diagram for the SU2C-SARC032 clinical trial. ctDNA, circulating tumor DNA; QC, quality control. Overview of ctDNA analysis of patients treated in SU2C-SARC032. (A) Patients randomly assigned to the control arm were treated with neoadjuvant RT followed by surgery, and patients randomly assigned to the experimental arm were treated with three cycles of neoadjuvant pembrolizumab and radiotherapy followed by surgery and up to 14 cycles of adjuvant pembrolizumab. Blood samples were collected for ctDNA analysis pretreatment, post-RT/presurgery, 3 months postsurgery, and 12 months postsurgery. (B) Plot of patient characteristics and tumor variants tracked for each patient. The top heatmaps depict key patient characteristics, and the bottom heatmaps show mutations in genes most frequently altered by single-nucleotide variants, the total number of variants tracked using ctDNA analysis, and the pretreatment ctDNA AF. AF, allele fraction; ctDNA, circulating tumor DNA; DDLPS, dedifferentiated liposarcoma; MFS, myxofibrosarcoma; PLPS, pleomorphic liposarcoma; RT, radiotherapy; UPS, undifferentiated pleomorphic sarcoma. At baseline, we detected ctDNA in 85% of patients at a median allele fraction (AF) of 0.13% (range, 0.003%-15.6% in patients with ctDNA detected, Fig 3A, Data Supplement, Tables S4 and S5). Pretreatment ctDNA levels were positively correlated with primary tumor size measured by tumor volume on the radiation planning CT scan or clinical T stage (Fig 3B, Appendix Fig A2A). Pretreatment ctDNA levels were higher in sarcomas originating in the lower limb than the upper limb (Appendix Fig A2B), likely because of the larger volume of these tumors (Appendix Fig A2C). In a multiple linear regression model, tumor volume but not tumor location was independently associated with pretreatment ctDNA AF (Data Supplement, Table S6). Patients with grade 3 tumors had higher pretreatment ctDNA levels than patients with grade 2 tumors (Fig 3C). In addition, pretreatment ctDNA levels were higher in patients with pleomorphic liposarcomas than other histologies (Fig 3D). To further investigate biological features associated with ctDNA levels, we analyzed bulk RNA sequencing from the paired primary tumors to identify differentially expressed genes between patients with ctDNA detected and ctDNA not detected pretreatment. We observed enrichment for Hallmark gene sets associated with hypoxia and the cell cycle in patients with ctDNA detected (Fig 3E, Appendix Fig 2D, Data Supplement, Table S7). By contrast, tumors from patients with ctDNA not detected were enriched for the Hallmark myogenesis gene set, which includes genes involved in the development of skeletal muscle.16 These findings demonstrate the association of pretreatment ctDNA levels with sarcoma biology and clinical risk factors associated with increased risk of metastasis. Association of pretreatment ctDNA levels with patient characteristics, tumor biology, and outcomes. (A) Plot of pretreatment ctDNA AF. (B) Correlation of primary tumor volume measured on the radiation planning CT scan and pretreatment ctDNA AF. Spearman's correlation coefficient and two-sided P value are displayed in the graph. (C) Plot of pretreatment ctDNA AF based on primary tumor grade. P values were calculated using a two-sided Mann-Whitney test. (D) Plot of pretreatment ctDNA AF based on primary tumor histology (Kruskal-Wallis P = .03). Displayed P values were calculated using two-sided Dunn's post-tests corrected for multiple hypothesis testing using the Benjamini-Hochberg method. DDLPS (n = 5), PLPS (n = 5), UPS (n = 78), MFS (n = 10). (E) Bar plots showing Hallmark gene set enrichment analysis results comparing primary tumors from patients with ctDNA detected and not detected pretreatment. Gene sets with an adjusted P value of <.05 are displayed. (F) Table showing HRs and 95% CI for a multivariable Cox proportional hazards model of DFS including pretreatment ctDNA AF and patient characteristics. P values were calculated using two-sided Wald tests. Kaplan-Meier curves showing (G) DFS and (H) OS in patients with and without ctDNA detected pretreatment. P values were calculated using two-sided log-rank tests. AF, allele fraction; CT, computed tomography; ctDNA, circulating tumor DNA; DDLPS, dedifferentiated liposarcoma; DFS, disease-free survival; HR, hazard ratio; MFS, myxofibrosarcoma; ND, not detected; OS, overall survival; PLPS, pleomorphic liposarcoma; UPS, undifferentiated pleomorphic sarcoma. Log10-transformed pretreatment ctDNA AF % as a continuous variable was strongly associated with DFS on univariable analysis (hazard ratio [HR] 1.81 [95% CI, 1.38 to 2.38]; P = 1.96 × 10−5). On multivariable analysis including the treatment arm and clinical variables that have previously been validated to predict OS and distant metastasis in STS,17 log10-transformed pretreatment ctDNA AF % continued to be strongly associated with DFS (HR, 1.74 [95% CI, 1.25 to 2.42]; P = .001, Fig 3F, Data Supplement, Table S8), demonstrating that ctDNA analysis provides additional prognostic information not captured by previously described clinical risk factors. When stratifying patients by ctDNA detection, patients with ctDNA detected pretreatment had worse DFS and OS (Figs 3G and 3H). Among patients without ctDNA detected pretreatment (n = 15), only two patients experienced DFS events and there were no deaths in the follow-up period. There was no interaction between ctDNA detection and treatment arm (interaction P = .99, Appendix Figs A2E and A2F). These results suggest that pretreatment ctDNA levels can be used to identify patients with localized STS with a poor prognosis. Monitoring Response to Neoadjuvant Therapy With ctDNA Analysis A total of 84 patients had evaluable postradiotherapy/presurgery ctDNA samples, and two patients had disease progression before their blood draw and were excluded from DFS analysis. In patients with ctDNA detected pretreatment, ctDNA decreased at the postradiotherapy/presurgery time point in most patients (Fig 4A). We observed a larger decrease in ctDNA levels postradiotherapy for the experimental arm (Fig 4B), suggesting that ctDNA dynamics during neoadjuvant therapy could provide an early readout of neoadjuvant treatment efficacy. Change in ctDNA levels after neoadjuvant therapy. (A) Comparison of pretreatment and post-RT/presurgery ctDNA AF for patients with ctDNA detected pretreatment. P values were calculated using a two-sided Wilcoxon matched-pairs signed-rank test. (B) Log10 fold change in ctDNA AF from pretreatment to post-RT/presurgery for patients treated in the control and experimental arms. P values were calculated using a two-sided Mann-Whitney test. (C) HRs and 95% CIs for univariable Cox proportional hazards models of DFS based on log10 absolute ctDNA AF post-RT/presurgery, log10 fold change in ctDNA AF from pretreatment to post-RT/presurgery in all patients, log10 fold change in ctDNA AF from pretreatment to post-RT/presurgery in patients with <90% necrosis in their surgically resected sarcoma, pathologic percent necrosis as a continuous variable, and ≥95% pathologic percent necrosis as a categorical variable. P values were calculated using two-sided Wald tests. For analyses of ctDNA log10 fold change from pretreatment to post-RT/presurgery and pathologic percent necrosis, only patients with ctDNA detected pretreatment were included. (D) Table showing HRs and 95% CI for a multivariable Cox proportional hazards model of DFS including post-RT/presurgery ctDNA AF and patient characteristics. P values were calculated using two-sided Wald tests. (E) Stacked bar plot comparing the percent of patients with ctDNA detected post-RT/presurgery in the control and experimental arms. P value was calculated using a two-sided Fisher's exact test. (F) Kaplan-Meier curves showing DFS in patients with and without ctDNA detected post-RT/presurgery. P value was calculated using a two-sided log-rank test. AF, allele fraction; ctDNA, circulating tumor DNA; DDLPS, dedifferentiated liposarcoma; DFS, disease-free survival; HR, hazard ratio; MFS, myxofibrosarcoma; ND, not detected; OS, overall survival; PLPS, pleomorphic liposarcoma; RT, radiotherapy; UPS, undifferentiated pleomorphic sarcoma. On landmark analysis from the postradiotherapy/presurgery blood draw in patients who remained event-free at the landmark, log10-transformed postradiotherapy/presurgery ctDNA AF % (HR, 1.74 [95% CI, 1.35 to 2.24]; P = .00002) and log10-transformed ctDNA fold change from pretreatment (HR, 1.39 [95% CI, 1.03 to 1.89]; P = .03) were associated with DFS as continuous variables (Fig 4C). We hypothesized that response to local therapy could confound the ctDNA fold change because some patients might have a large decrease in ctDNA because of a favorable response to radiotherapy despite having micrometastatic disease. In support of this possibility, we observed a stronger association between log10-transformed ctDNA fold change and DFS when excluding patients with ≥90% necrosis in their primary tumor after radiotherapy based on surgical pathology (HR, 2.01 [95% CI, 1.24 to 3.26]; P = .005). Notably, pathologic percent necrosis was not associated with DFS as a continuous variable (HR, 0.99 [95% CI, 0.98 to 1.01]; P = .63) or in patients with ≥95% necrosis (HR, 0.66 [95% CI, 0.31 to 1.40]; P = .28). On multivariable analysis, the log10-transformed absolute ctDNA AF % postradiotherapy/presurgery (HR, 1.75 [95% CI, 1.27 to 2.41]; P = .0006, Fig 4D, Data Supplement, Table S9) but not the log10-transformed ctDNA fold change was associated with DFS (HR, 1.29 [95% CI, 0.93 to 1.80]; P = .13, Appendix Fig A3A, Data Supplement, Table S10). In patients with ctDNA detected pretreatment, a numerically higher proportion of patients converted to ctDNA not detected after radiotherapy in the experimental arm than the control arm (71% v 53%, P = .16, Fig 4E). Finally, ctDNA detection postradiotherapy/presurgery was associated with worse DFS (Fig 3F), but there was a weaker association with OS (Appendix Fig A3B). These results suggest that ctDNA dynamics and levels after neoadjuvant therapy are strongly associated with outcomes in STS. Detection of Minimal Residual Disease With ctDNA Analysis A total of 81 patients had evaluable ctDNA samples collected 3 months postsurgery, and one patient had disease progression before their blood draw and was excluded from DFS analysis. On landmark analysis from the 3-month postsurgery blood draw in patients who remained event-free at the landmark, log10-transformed ctDNA AF % as a continuous variable was associated with DFS on multivariable analysis (HR, 1.40 [95% CI, 1.08 to 1.80]; P = .01, Fig 5A, Data Supplement, Table S11). Detection of ctDNA at 3 months postsurgery was numerically lower for patients treated with pembrolizumab in the experimental arm than the control arm (18% v 27%, P = .54, Appendix Fig A3C). Detection of ctDNA at 3 months postsurgery was strongly associated with both DFS and OS (Figs 5B and 5C). Notably, the two patients with ctDNA MRD detected who had the longest follow-up without disease progression were treated in the experimental arm of the trial. In both patients, ctDNA became undetectable at 12 months postsurgery consistent with a response to adjuvant pembrolizumab. There was no interaction between ctDNA status at 3 months postsurgery and the treatment arm (interaction P = .56, Appendix Figs A3D and A3E). DFS from the last blood draw collected 3 or 12 months postsurgery was worse in patients with ctDNA detected (Appendix Fig A3F). These findings demonstrate the potential for ctDNA MRD to guide adjuvant therapy in patients with STS. Detection of ctDNA minimal residual disease after surgery. (A) Table showing HR and 95% CI for a multivariable Cox proportional hazards model of DFS including 3-month postsurgery ctDNA AF and patient characteristics. P values were calculated using two-sided Wald tests. Kaplan-Meier curves showing (B) DFS and (C) OS in patients with and without ctDNA detected 3 months postsurgery. P values were calculated using two-sided log-rank tests. AF, allele fraction; ctDNA, circulating tumor DNA; DDLPS, dedifferentiated liposarcoma; DFS, disease-free survival; HR, hazard ratio; MFS, myxofibrosarcoma; ND, not detected; OS, overall survival; PLPS, pleomorphic liposarcoma; RT, radiotherapy; UPS, undifferentiated pleomorphic sarcoma. Longitudinal Subclone Monitoring With ctDNA Analysis Subclonal tumor populations that arise later in tumor development can drive treatment resistance, and prior studies have suggested that polyclonal metastatic dissemination is associated with worse outcomes.18,19 Because our bespoke panels monitor all clonal and subclonal mutations present in the primary tumor, they enable tracking subclonal tumor populations in the plasma.20 Therefore, we investigated whether ctDNA clonal architecture was associated with patient outcomes. Based on tumor sequencing, we identified multiple tumor clones in most patients (Fig 6A, Data Supplement, Tables S12 and S13) and multiple clones were also detected in the plasma pretreatment for most patients (Fig 6B, Data Supplement, Table S14). Clones that were detected in the plasma pretreatment had a higher cellular prevalence within the primary tumor than clones that were not detected in the plasma (Appendix Figs A4A and A4B). The number of clones detected decreased over the course of treatment (Figs 6C and 6D, Appendix Fig A4C). Among patients with ctDNA detected, DFS by landmark analysis was worse for patients with multiple clones detected, particularly at the 3-month postsurgery blood draw (Figs 6E-6G). Notably, there was no difference in ctDNA AF at 3 months postsurgery in patients with one versus multiple tumor clones detected (Appendix Fig A4D). Together, these results suggest that ctDNA subclonal architecture could provide additional prognostic information beyond ctDNA levels alone. Longitudinal subclone monitoring with ctDNA analysis. Donut charts displaying the number of different tumor clones (A) identified in the primary tumor based on whole-exome sequencing and detected by ctDNA analysis of (B) pretreatment plasma samples, (C) postradiotherapy/presurgery plasma samples, and (D) 3-month postsurgery plasma samples. Kaplan-Meier curves showing DFS based on the number of clones detected by ctDNA analysis: (E) pretreatment, (F) postradiotherapy/presurgery, and (G) 3 months postsurgery. P values were calculated using two-sided log-rank tests. Only patients with at least two different clones identified in the primary tumor are included in (E-G). ctDNA, circulating tumor DNA; DFS, disease-free survival; HR, hazard ratio. DISCUSSION Using personalized, tumor-informed ctDNA profiling in samples from the SU2C-SARC032 clinical trial, we demonstrate that ctDNA analysis is feasible in STS and provides clinically meaningful information across the perioperative treatment course. Levels of ctDNA before therapy, after neoadjuvant therapy but before surgery, and after surgery were strongly associated with outcomes, even after adjusting for established clinical risk factors, supporting ctDNA as an independent prognostic biomarker. Together, these findings suggest that ctDNA may enable biologically informed monitoring in rare and heterogenous cancers to guide personalized treatment decisions. A key finding of this work is robust ctDNA detection in localized sarcomas using a bespoke approach. Prior studies in sarcomas have reported modest detection rates with fixed gene panels designed to track recurrent SNVs and structural variants or low-pass whole-genome approaches aimed at identifying copy number alterations.10 Bespoke, tumor-informed strategies have improved sensitivity for ctDNA detection in more common solid tumors,4,12,19,21,22 and our findings add to recent studies that have demonstrated the feasibility of this approach in smaller cohorts of patients with STS.23,24 Baseline ctDNA levels correlated with primary tumor size, grade, and histology. Tumor size and grade are known predictors of distant metastasis in STS,17,25 and pleomorphic liposarcomas, which had the highest baseline ctDNA levels, have been reported to metastasize at a higher rate than other liposarcoma subtypes.26 By analyzing gene expression in matched primary tumors, we identified upregulation of pathways involved in the cell cycle and hypoxia in patients with ctDNA detected and upregulation of myogenesis in patients with ctDNA not detected pretreatment. Similar associations between the cell cycle and myogenesis with ctDNA detection were also previously identified in patients with urothelial carcinoma,7 and hypoxia has been shown to promote metastasis in STS.27,28 Importantly, baseline ctDNA was strongly associated with both DFS and OS and remained prognostic after accounting for clinical covariables, suggesting that ctDNA captures additional aspects of risk. Clinically, our data suggest that patients with high-risk, localized STS without ctDNA detected at baseline have excellent outcomes and may be candidates for local therapy alone. By contrast, patients with high-risk, localized STS with ctDNA detected at baseline have worse outcomes and may benefit from systemic therapy or enhanced surveillance. In contrast to our prior study in metastatic sarcomas, where we observed a temporary increase in ctDNA levels after local cryotherapy,11 most patients exhibited decreased ctDNA levels after radiotherapy. This discrepancy may reflect the timing of blood collection, different modes of tumor cell death, or the more prolonged 5-week course of radiotherapy that patients received in SU2C-SARC032. Notably, patients receiving neoadjuvant pembrolizumab demonstrated a larger decrease in their ctDNA levels, paralleling the improved DFS observed in the experimental arm of SU2C-SARC032.3 In addition, patients with a larger decrease in their ctDNA levels after neoadjuvant therapy had improved clinical outcomes, consistent with previous studies in other solid cancers.29-32 In contrast to prior studies,33,34 we did not observe an association between pathologic response and DFS, but collection of pathologic response data was not standardized in the trial protocol and was collected by the treating institution rather than by central pathology review. These findings demonstrate the potential for ctDNA to monitor response to neoadjuvant therapy in patients with STS. Detection of ctDNA 3 months after surgery was strongly associated with DFS and OS, aligning with a growing body of evidence across solid tumors showing that MRD identifies patients at high risk of relapse.4 We also found that persistence of multiple clones by ctDNA analysis after surgery was associated with inferior outcomes even when ctDNA levels were similar, mirroring the observation that patients with non–small cell lung cancer who had polyclonal metastatic dissemination detected by ctDNA analysis had inferior outcomes compared with patients with monoclonal metastatic dissemination.19 Prospective studies in colorectal cancer and bladder cancer have demonstrated that ctDNA-guided adjuvant treatment is feasible and can improve patient outcomes,6,7 and meta-analyses have suggested a small OS benefit for adjuvant chemotherapy in patients with localized, resectable STS.35,36 Given the substantial toxicity of standard chemotherapies for sarcoma, incorporating MRD analysis after surgery may help guide patients with sarcoma and their oncologists as they consider adjuvant chemotherapy. Despite SU2C-SARC032 being one of the largest clinical trials reported to date in patients with a defined subset of STS, the rarity of sarcomas limited the number of patients who could be enrolled in the trial. The study was not powered for correlative studies or to show the benefit of pembrolizumab in subsets of patients (eg, patients with ctDNA detected pretreatment), limiting definitive conclusions regarding the predictive ability of ctDNA analysis. A limitation of our personalized, tumor-informed ctDNA analysis approach is the requirement for paired tumor and germline sequencing, which may not be available for all patients. Finally, while we adjusted for key clinical covariables, residual confounding is possible. Because of the exploratory nature of these correlative studies, prospective validation will be necessary to confirm our findings. In conclusion, personalized ctDNA profiling enabled reliable ctDNA detection and treatment response monitoring in localized STS. Levels of ctDNA before treatment, after neoadjuvant radiation with or without immunotherapy, and postoperatively added to standard clinical risk factors and tracked response to therapy. These results support ctDNA as a promising biomarker for personalizing treatment decisions in STS. Prospective ctDNA-guided interventional trials will be necessary to demonstrate the benefit of personalizing therapy based on ctDNA analysis. ACKNOWLEDGMENT We thank the patients and caregivers who participated in this study. We thank Denise Reinke, Steven Young, Scott Okuno, and the research coordinators at SARC for supporting this trial by collecting patient data and samples. In particular, we thank Erin Kozlowski and Meghan Blalock who provided project management at SARC. We also thank the participating clinicians and clinical research teams at the participating sites for enrolling patients to make this international trial and correlative studies possible. The schematic in Fig 2A was created using BioRender.com. APPENDIX Sample inclusion and exclusion. (A) Stacked bar plot displaying the percent of plasma samples from evaluable patients included at each time point and reasons for sample exclusion. Kaplan-Meier curves showing (B) DFS and (C) OS in patients from the ctDNA cohort and all enrolled patients. P values were calculated using two-sided log-rank tests. DFS, disease-free survival; HR, hazard ratio; OS, overall survival; QC, quality control. Association of pretreatment ctDNA with primary tumor characteristics and outcomes based on the treatment arm. Plots of pretreatment ctDNA AF based on primary tumor (A) T stage and (B) location. (C) Plot of primary tumor volume measured on the radiation planning CT scan based on the location of the primary tumor. P values were calculated using a two-sided Mann-Whitney test. (D) Volcano plot of differential gene expression analysis in primary tumors based on pretreatment ctDNA detection. Genes associated with hypoxia (hypoxia and angiogenesis Hallmark gene sets), cell cycle (E2F targets and G2M checkpoint Hallmark gene sets), and myogenesis (myogenesis Hallmark gene set) are colored in the plot. Kaplan-Meier curves showing DFS in the experimental versus control arm for patients with (E) ctDNA not detected and (F) ctDNA detected pretreatment. P values were calculated using two-sided log-rank tests. AF, allele fraction; CT, computed tomography; ctDNA, circulating tumor DNA; DFS, disease-free survival; HR, hazard ratio; ND, not detected. Association of postradiotherapy and postsurgery ctDNA with outcomes. (A) Table showing HRs and 95% CI for a multivariable Cox proportional hazards model of DFS including the log10 fold change in ctDNA AF from pretreatment to post-RT/presurgery and patient characteristics. Only patients with ctDNA detected pretreatment were included. P values were calculated using two-sided Wald tests. (B) Kaplan-Meier curves showing OS in patients with and without ctDNA detected post-RT/presurgery. P value was calculated using a two-sided log-rank test. (C) Stacked bar plot comparing the percent of patients with ctDNA detected 3 months postsurgery in the control and experimental arms. Only patients with ctDNA detected pretreatment were included. P value was calculated using a two-sided Fisher's exact test. Kaplan-Meier curves showing DFS in the experimental versus control arm for patients with (D) ctDNA not detected and (E) ctDNA detected 3 months postsurgery. P values were calculated using two-sided log-rank tests. (F) Kaplan-Meier curves showing DFS in patients with and without ctDNA detected at the last blood draw collected 3 or 12 months postsurgery. P values were calculated using a two-sided log-rank test. AF, allele fraction; ctDNA, circulating tumor DNA; DFS, disease-free survival; HR, hazard ratio; OS, overall survival; RT, radiotherapy. Tumor prevalence of clones detected with ctDNA analysis. (A) Plot of clone cellular prevalence (fraction of total tumor cells assigned to each clone) in the primary tumor based on whether the clone was detected in the pretreatment plasma using ctDNA analysis. P value was calculated using a two-sided Mann-Whitney test. (B) Representative phylogenetic trees based on primary tumor sequencing with clones colored based on detection in the pretreatment plasma using ctDNA analysis. (C) Representative plots of tumor clone ctDNA AFs across longitudinal time points in individual patients. (D) Plot of ctDNA AF 3 months postsurgery in patients with one or at least two different clones detected by ctDNA analysis. P value was calculated using a two-sided Mann-Whitney test. AF, allele fraction; ctDNA, circulating tumor DNA; ND, not detected; RT, radiotherapy. Ajay Subramanian Patents, Royalties, Other Intellectual Property: Patent for a different project about cardiovascular disease Karla V. Ballman This author is a member of the Journal of Clinical Oncology Editorial Board. Journal policy recused the author from having any role in the peer review of this manuscript. Consulting or Advisory Role: GlaxoSmithKline Expert Testimony: Janssen Rachel S. Heise Employment: CSL Behring (I) Claire Johns Employment: Stanford University, University of California, San Francisco (I) Travel, Accommodations, Expenses: Stanford University Angela M. Hong Consulting or Advisory Role: Telix Pharmaceuticals Travel, Accommodations, Expenses: Roche (I) Brian E. Brigman Consulting or Advisory Role: Deciphera, Daiichi Sankyo/UCB Japan, Musculoskeletal Transplant Foundation Research Funding: Musculoskeletal Transplant Foundation Richard F. Riedel Employment: Limbguard (I) Leadership: Limbguard (I) Stock and Other Ownership Interests: Limbguard (I) Consulting or Advisory Role: Daiichi Sankyo, Deciphera, SpringWorks Therapeutics, AADi, GlaxoSmithKline, Adaptimmune, Boehringer Ingelheim, Recordati, EMD Serono Research Funding: TRACON Pharma (Inst), AADi (Inst), Arog (Inst), Daiichi Sankyo (Inst), GlaxoSmithKline (Inst), SpringWorks Therapeutics (Inst), Blueprint Medicines (Inst), Epizyme (Inst), Ayala Pharmaceuticals (Inst), Deciphera (Inst), PTC Therapeutics (Inst), Cogent Biosciences (Inst), BioAtla (Inst), InhibRx (Inst), SynOX (Inst), Intensity Therapeutics (Inst), Servier (Inst) Patents, Royalties, Other Intellectual Property: PandoNet—Limbguard (I) Travel, Accommodations, Expenses: Daiichi Sankyo, NanoCarrier, SpringWorks Therapeutics, Deciphera, Recordati Yvonne M. Mowery Employment: Duke Health Private Diagnostic Clinic, UPMC Hillman Cancer Center/University of Pittsburgh Honoraria: UpToDate, Binaytara Consulting or Advisory Role: Catalys Pacific Research Funding: Merck (Inst) David G. Kirsch Stock and Other Ownership Interests: Lumicell, X-RAD Therapeutics Research Funding: Merck Patents, Royalties, Other Intellectual Property: Patent on an imaging device licensed to Lumicell, Royalty on an imaging agent being commercialized by Lumicell, Patent on small-molecule radiosensitizers by XRAD Therapeutics Everett J. Moding Consulting or Advisory Role: Guidepoint Pharmacy, Gerson Lehrman Group, Invoke Bio Research Funding: Merck (Inst) No other potential conflicts of interest were reported. SUPPORT Supported by a Stand Up To Cancer Catalyst Research Team grant with support from Merck (E.J.M., D.G.K., Y.M.M., and K.J.W.), a research grant from the Investigator-Initiated Studies Program of Merck Sharp & Dohme Corp (D.G.K.), the National Cancer Institute (E.J.M.: K08CA25542501, D.G.K.: R35CA197616), the Department of Defense (E.J.M.: W81XWH-22-1-0161), the My Blue Dots organization (E.J.M.), the Tad and Diane Taube Family Foundation (M.v.R. and E.J.M.), and the GPA Andrew Ursini Charitable Fund (A.M.H.). The opinions expressed in this study are those of the authors and do not necessarily represent those of Merck Sharp & Dohme Corp. CLINICAL TRIAL INFORMATION NCT03092323 (SU2C-SARC032) A.S., S.C.L.N., and N.N.-G. contributed equally to this work. DATA SHARING STATEMENT A data sharing statement provided by the authors is available with this article at DOI https://doi.org/10.1200/JCO-26-00769. AUTHOR CONTRIBUTIONS Conception and design: Karla V. Ballman, Siyer Roohani, Brian E. Brigman, Yvonne M. Mowery, David G. Kirsch, Everett J. Moding Financial support: Angela M. Hong, Kent J. Weinhold, Matt van de Rijn, Yvonne M. Mowery, David G. Kirsch, Everett J. Moding Administrative support: Kent J. Weinhold, Yvonne M. Mowery, David G. Kirsch, Everett J. Moding Provision of study materials or patients: Angela M. Hong, Kent J. Weinhold, Richard F. Riedel, David G. Kirsch Collection and assembly of data: Ajay Subramanian, Serey C.L. Nouth, Neda Nemat-Gorgani, Claire Johns, Timothy J. Sears, Siyer Roohani, Reinhardt Krcek, Angela M. Hong, Kent J. Weinhold, Brian E. Brigman, Yvonne M. Mowery, Everett J. Moding Data analysis and interpretation: Ajay Subramanian, Neda Nemat-Gorgani, Shaghayegh Soudi, Karla V. Ballman, Rachel S. Heise, Timothy J. Sears, Siyer Roohani, Matt van de Rijn, Brian E. Brigman, Richard F. Riedel, Yvonne M. Mowery, David G. Kirsch, Everett J. Moding Manuscript writing: All authors Final approval of manuscript: All authors Accountable for all aspects of the work: All authors AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST Personalized Circulating Tumor DNA Analysis for Predicting Outcomes and Tracking Response to Radiotherapy and Pembrolizumab in Localized Sarcomas: Analysis of the SU2C-SARC032 Trial The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center. Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments). Ajay Subramanian Patents, Royalties, Other Intellectual Property: Patent for a different project about cardiovascular disease Karla V. Ballman This author is a member of the Journal of Clinical Oncology Editorial Board. Journal policy recused the author from having any role in the peer review of this manuscript. Consulting or Advisory Role: GlaxoSmithKline Expert Testimony: Janssen Rachel S. Heise Employment: CSL Behring (I) Claire Johns Employment: Stanford University, University of California, San Francisco (I) Travel, Accommodations, Expenses: Stanford University Angela M. Hong Consulting or Advisory Role: Telix Pharmaceuticals Travel, Accommodations, Expenses: Roche (I) Brian E. Brigman Consulting or Advisory Role: Deciphera, Daiichi Sankyo/UCB Japan, Musculoskeletal Transplant Foundation Research Funding: Musculoskeletal Transplant Foundation Richard F. Riedel Employment: Limbguard (I) Leadership: Limbguard (I) Stock and Other Ownership Interests: Limbguard (I) Consulting or Advisory Role: Daiichi Sankyo, Deciphera, SpringWorks Therapeutics, AADi, GlaxoSmithKline, Adaptimmune, Boehringer Ingelheim, Recordati, EMD Serono Research Funding: TRACON Pharma (Inst), AADi (Inst), Arog (Inst), Daiichi Sankyo (Inst), GlaxoSmithKline (Inst), SpringWorks Therapeutics (Inst), Blueprint Medicines (Inst), Epizyme (Inst), Ayala Pharmaceuticals (Inst), Deciphera (Inst), PTC Therapeutics (Inst), Cogent Biosciences (Inst), BioAtla (Inst), InhibRx (Inst), SynOX (Inst), Intensity Therapeutics (Inst), Servier (Inst) Patents, Royalties, Other Intellectual Property: PandoNet—Limbguard (I) Travel, Accommodations, Expenses: Daiichi Sankyo, NanoCarrier, SpringWorks Therapeutics, Deciphera, Recordati Yvonne M. Mowery Employment: Duke Health Private Diagnostic Clinic, UPMC Hillman Cancer Center/University of Pittsburgh Honoraria: UpToDate, Binaytara Consulting or Advisory Role: Catalys Pacific Research Funding: Merck (Inst) David G. Kirsch Stock and Other Ownership Interests: Lumicell, X-RAD Therapeutics Research Funding: Merck Patents, Royalties, Other Intellectual Property: Patent on an imaging device licensed to Lumicell, Royalty on an imaging agent being commercialized by Lumicell, Patent on small-molecule radiosensitizers by XRAD Therapeutics Everett J. Moding Consulting or Advisory Role: Guidepoint Pharmacy, Gerson Lehrman Group, Invoke Bio Research Funding: Merck (Inst) No other potential conflicts of interest were reported. REFERENCES Associated Data Data Availability Statement A data sharing statement provided by the authors is available with this article at DOI https://doi.org/10.1200/JCO-26-00769.