Abstract Introduction: Genomic profiling of circulating tumor DNA (ctDNA) through liquid biopsies has become an important diagnostic method in clinical oncology. However, detection of variants related to clonal hematopoiesis (CH) is a major confounder that impairs the clinical utility of liquid biopsies. Strategies that reduce biological noise from CH in plasma NGS include deep sequencing of matched WBC DNA and/or tumor tissue sequencing. While these methods effectively distinguish most CH variants, the need for extra biospecimens and sequencing raises costs and limits feasibility. Methods: Using a training cohort of 426 variants identified in ctDNA NGS from 225 patients with stage I-IV solid tumors, we developed plasmaCHORD, a machine learning model (MLM) that includes fragmentomic, variant, and patient-level features to distinguish between tumor- and CH-origin for mutations detected by fixed gene panel hybrid capture NGS. Model performance was assessed by comparison to the reference origin of each plasma variant determined from matched WBC and tumor NGS. Following locking the model parameters, we applied plasmaCHORD to an independent validation cohort of 1,412 plasma variants detected in 114 patients with metastatic cancers, as well as to cfDNA NGS from patients enrolled in a prospective liquid biopsy-informed clinical trial (NCT05585684). Results: PlasmaCHORD predicted tumor versus CH-origin in the training set with high accuracy (cross-validated AUC=0.94), outperforming individual features such as variant allele frequency and canonical CH genes. Model performance remained robust when restricted to mutant DNA fragments supported by 3-5 mutant reads (AUC = 0.84). plasmaCHORD was locked for evaluation using a score of 0.5 as cutoff for distinguishing tumor- versus CH-origin variants. In the independent validation cohort, the locked model maintained similar overall accuracy (AUC=0.9) with a sensitivity of 82%, specificity 80.3% and accuracy of 80.2%. Our approach was shown to be highly reliable in classifying variant origin in clinically actionable genes not canonically associated with CH, including AKT1, ATM, BRCA1, BRCA2, and EGFR, as well as adjudicating cellular origin for TP53 mutations that are encountered in both solid and hematologic malignancies. Performance was consistent across cancer types, sequencing platforms, mutation classes, and a wide range of allele fractions. When applied to clinically challenging cases in the context of a precision oncology clinical trial, plasmaCHORD precisely determined variant origin, preventing mismatches with genotype-targeted therapies. Conclusions: plasmaCHORD, a multi-feature machine-learning classifier, can significantly enhance the ability to identify bona fide tumor variants in routine plasma-only NGS, addressing a critical need in implementing liquid biopsy-guided therapy by minimizing misinterpretation caused by CH. Citation Format: Daniel J. Rabizadeh, Jenna VanLiere Canzoniero, Ilias Ziakas, Jaime Wehr, Archana Balan, Amna Jamali, Blair V. Landon, Susan Combs Scott, Gavin Pereira, Vincent K. Lam, Christine L. Hann, Christine M. Lovly, Jessica Tao, Patrick M. Forde, Joseph C. Murray, Mark Sausen, Gerrit A. Meijer, Geraldine Vink, Remond J. A. Fijneman, Victor E. Velculescu, Jillian Ayn Phallen, Robert Scharpf, Valsamo Anagnostou. PlasmaCHORD- A machine learning method for identifying clonal hematopoiesis variants in liquid biopsies [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 95.

Abstract Background: Genomic profiling via liquid biopsies (LB) has advanced precision oncology decision-making; however, a major challenge is critically interpreting LB data to improve the selection and order of genotype-specific therapies. Methods: We present updated results from the first planned analysis of an observational biomarker trial, aimed at assessing the clinical utility of serial LB in patients with advanced or metastatic solid tumors (NCT05585684). Primary endpoints assessed feasibility, prevalence of actionable alterations, and fraction of patients receiving genotype-matched therapies. Secondary endpoints encompassed progression-free survival (PFS), overall survival (OS), and concordance between LB and tumor next-generation sequencing (NGS). Serial LBs at baseline, early (1-3 weeks) on therapy, and at progression, employed a clinically validated 33-gene panel NGS assay (Labcorp Plasma Focus). Matched white blood cell (WBC) NGS was used to identify clonal hematopoiesis (CH)-derived variants. Mutation actionability was evaluated using a multi-resource programmatic approach, and levels of evidence (1-4) were assigned, followed by review at the Johns Hopkins Molecular Tumor Board (JH MTB). Results: Among 50 patients, 45 baseline and 12 progression LBs were reviewed at MTB. JH MTB classified 72.5% (n=50) of baseline alterations as tumor-derived. In 19 patients with archival tissue NGS, 32 variants were detected in the baseline LB, of which 78.1% were also detected in tissue NGS. Furthermore, 23.2% (n=16) of LB alterations at baseline, 20.0% (n=6) early on-therapy, and 13.0% (n=3) at progression were CH derived. Upon MTB review, 29.2% variants at baseline, 21.9% at early on-therapy and 26.1% at progression were classified as actionable (58.8% level 1, 11.8% level 2, 11.8% level 3 and 17.6% level 4 evidence). Of 45 patients reviewed, 38 received therapeutic recommendations and 57.9% initiated recommended therapy. Patients treated with MTB recommended therapies had significantly longer median PFS and OS compared to those who received standard of care (p=0.027 and p=0.00062 respectively). MTB-recommended therapy selection was independently associated with PFS and OS, in multivariate analyses adjusting for clinical covariates including sex, smoking status, age, and prior lines of systemic therapy (p=0.043 and p=0.006, respectively). Subset of patients treated with genotype-matched MTB recommendations had significantly longer median PFS and OS compared to those who received standard of care (p=0.026 and p=0.0074 respectively). Conclusion: Our findings emphasize the importance of precision oncology interventions driven by programmatic workflows within a multidisciplinary MTB, supported by comprehensive LB molecular information to guide therapy selection and improve patient outcomes. Citation Format: Amna Jamali, Jaime Wehr, Jenna VanLiere Canzoniero, Maria Fatteh, Katerina Karaindrou, Michael Conroy, Ilias Ziakas, Mohamed Sherief, Timsy Wanchoo, Faith Too, Lily Scharpf, Ruth Moges, Dana Petry, Kala Visvanathan, Ellen Verner, Amy Greer, Kory Kreimeyer, Jonathan Spiker, Rachel Karchin, Christine L. Hann, Vincent K. Lam, Joseph Christopher Murray, Josephine Feliciano, Kristen Marrone, Julie R. Brahmer, Ming-Tseh Lin, Taxiarchis Botsis, Hao Wang, Mark Sausen, Christopher D. Gocke, Rena Xian, Jessica Tao, Valsamo (Elsa) K. Anagnostou. Liquid biopsy-informed precision oncology clinical trial to evaluate the utility of ctDNA genomic profiling for therapy optimization in patients with advanced or metastatic solid tumors [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 3908.

Supplementary Table S1. Summary of clinical information. Supplementary Table S2. Summary of sample characteristics. Supplementary Table S3. Summary of whole-genome sequencing and fragment analyses. Supplementary Table S4. Summary of protein analyses. Supplementary Table S5. Summary of machine learning models and scores. Supplementary Table S6. Summary of performance for ovarian cancer detection.

171 Background: Since 2020, guidelines recommend somatic next-generation sequencing (NGS) in patients with metastatic (M1) prostate cancer. Nationwide, genetic testing rates are low with known disparities in access. Whether prostate cancer disease characteristics and courses differ between those with and without testing are unknown. In this study, we determined the NGS testing rate in M1 prostate cancer at a single academic center and assessed social and clinical features by NGS testing status. In those without NGS testing, we reported the most likely reasons for lack of testing. Finally, we compared survival among those with and without testing. Methods: Retrospective chart review of M1 patients with prostate cancer seen as a new visit between 2020-2022 with at least one follow-up. Sociodemographics, comorbidities, prostate cancer clinical features, and vital status were obtained from the electronic medical record (EMR). Reasons for lack of NGS testing were assessed through chart review. A multivariable (MV) logistic regression assessed predictors of NGS testing (covariates: age, Gleason grade group, marital status, M1 diagnosis year). Survival analysis was conducted from the date of first visit with M1 disease to the date of death/last follow-up (covariates: age, Gleason grade group, M stage at initial diagnosis, NCI comorbidity index, days from M1 diagnosis date to first visit date). Results: 258 (59%) of 435 patients had NGS ordered. Patients who were older and not married were less likely to have NGS testing, which remained in MV analysis [age: odds ratio (95% confidence interval) = 0.96 (0.94-0.98); unmarried vs. married/partnered 0.62 (0.38-1.00)]. (Table 1). The most common likely underlying reason for no NGS testing ordered were perceived patient/disease factors (28%), no tissue available (16%), and shared care with another oncologist (18%). There was no clear reason in 22% of patients. Those with testing ordered had worse survival than those without [adjusted hazard ratio (95% confidence interval) = 1.43 (1.03-1.99)]. Conclusions: NGS testing remains underutilized in men with metastatic prostate cancer, particularly among older and unmarried individuals. Those without NGS testing had longer survival, contrary to what is seen in other diseases. Future research is needed to develop and assess interventions to improve NGS testing rates. Multivariable analysis of predictors of ordering next-generation sequencing (NGS) in metastatic (M1) prostate cancer (N=419). Variables Odds Ratio 95% confidence interval p-value Age at M1 diagnosis 0.96 0.94-0.98 <0.01 Gleason score at initial diagnosis 0.07 Group 1-3 1 ref Group 4 1.59 0.84-3.01 Group 5 1.78 1.09-2.91 Gleason score never performed 2.20 1.05-4.59 Marital Status 0.05 Married/partnered 1 ref Not married/partnered 0.62 0.38-1.00 Year of M1 diagnosis 0.07 2019 or earlier 1 ref 2020-2022 0.62 0.36-1.05

Supplementary Figure S1. Evaluation of screening model DELFI-Pro scores and comorbidities in individuals without cancer. Supplementary Figure S2. DELFI-Pro score evaluation in available clinical characteristics of patients with cancer. Supplementary Figure S3. Stability analysis across fold repeats and collection source. Supplementary Figure S4. Detection of ovarian cancer subtypes using DELFI-Pro screening model. Supplementary Figure S5. ROC analyses of asymptomatic individuals in the using screening or diagnostic models in the Discovery Cohort. Supplementary Figure S6. Performance of ichorCNA and median cfDNA fragment lengths in the Discovery Cohorts. Supplementary Figure S7. Detection of ovarian cancer subtypes using DELFI-Pro at high specificity. Supplementary Figure S8. Genome-wide fragmentation profiles are altered in patients in the Validation Cohort with ovarian cancer. Supplementary Figure S9. Analyses of chromosomal changes in Discovery and Validation cohorts. Supplementary Figure S10. Performance of DELFI-Pro for detection of ovarian cancer in the Validation Cohort. Supplementary Figure S11. Correlation of rank ordered DELFI-Pro scores for the Screening and Diagnostic models. Supplementary Figure S12. Assessment of DELFI-Pro in women with benign lesions. Supplementary Figure S13. Performance of DELFI-Pro for distinguishing ovarian cancer from benign masses. Supplementary Figure S14. Performance of DELFI-Pro for distinguishing ovarian cancer subtypes from benign masses. Supplementary Figure S15. Performance of DELFI-Pro for distinguishing ovarian cancer subtypes from benign masses in Validation Cohort. Supplementary Figure S16. Evaluation of overall tumor burden using the sum of reported lesion diameters. Supplementary Figure S17. Evaluation of CA1-25 and HE4 blood concentrations measured at different centers.

e13672 Background: The growing compendium of genomic alterations linked to FDA-approved therapies or drugs in development, requires the implementation of informatics solutions that tailor next-generation sequencing and tumor molecular profiling to evidence based ranked molecularly guided targeted therapies in an automated, scalable and generalizable manner. Nevertheless, such comprehensive informatics solutions are currently lacking, highlighting a significant gap in precision oncology. Methods: We developed the Open Navigator through Precision Oncology INformatics Technology (OnPOINT) platform that utilizes open-source tools, retrieves data from public repositories via APIs, processes multi-source sequencing files, synthesizes genomic with clinical data and matches actionable molecular findings with clinical trials (CTs). Clinical and genomics data were standardized using the Precision Oncology Core Data Model (Precision-DM). Public APIs were utilized to retrieve information from variant registries, knowledgebases and clinicaltrials.gov. Variant oncogenicity was characterized by an ensemble approach through the meta-annotator OpenCravat. The clinical utility of OnPOINT was tested at the Johns Hopkins Molecular Tumor Board (JH MTB) and in the community setting. Results: OnPOINT operates in an IRB-approved web environment, hosting standardized data for > 1,200 cancer patients reviewed at the JH MTB. Following programmatic mutation characterization by oncogenicity and actionability, biomedical literature is annotated with the National Library of Medicine PubTator tool to retrieve gene, variant and drug entities and their relationships. In tandem, OnPOINT automatically identifies CTs tailored to a patient’s clinical-genomic profile. All source data, external knowledge, and generated information is displayed in an interactive dashboard and summarized in an auto-populated report. To evaluate the platform’s effectiveness, we performed a pilot benchmark using 15 cases manually reviewed by the JH MTB. OnPOINT automatically captured 88% of genotype-matched clinical trials recommended by the MTB experts; missed trials were mainly those with status changes from active to non-recruiting between the MTB review and the benchmark analysis. We then evaluated OnPOINT using a set of 12 patients receiving care at a community hospital; of 116 identified mutations, 23.3% were deemed actionable and matched to 40 genotype-targeted trials. Notably, none of these patients received genotype-targeted therapies, highlighting underscoring the utility of our approach in matching patients with clinical trials. Conclusions: Integrative informatics approaches open a window of opportunity for scaling MTB operations and maximizing genotype-matched clinical trial visibility in the community setting.

Abstract Ovarian cancer is a leading cause of death for women worldwide, in part due to ineffective screening methods. In this study, we used whole-genome cell-free DNA (cfDNA) fragmentome and protein biomarker [cancer antigen 125 (CA-125) and human epididymis protein 4 (HE4)] analyses to evaluate 591 women with ovarian cancer, with benign adnexal masses, or without ovarian lesions. Using a machine learning model with the combined features, we detected ovarian cancer with specificity >99% and sensitivities of 72%, 69%, 87%, and 100% for stages I to IV, respectively. At the same specificity, CA-125 alone detected 34%, 62%, 63%, and 100%, and HE4 alone detected 28%, 27%, 67%, and 100% of ovarian cancers for stages I to IV, respectively. Our approach differentiated benign masses from ovarian cancers with high accuracy (AUC = 0.88, 95% confidence interval, 0.83–0.92). These results were validated in an independent population. These findings show that integrated cfDNA fragmentome and protein analyses detect ovarian cancers with high performance, enabling a new accessible approach for noninvasive ovarian cancer screening and diagnostic evaluation. Significance: There is an unmet need for effective ovarian cancer screening and diagnostic approaches that enable earlier-stage cancer detection and increased overall survival. We have developed a high-performing accessible approach that evaluates cfDNA fragmentomes and protein biomarkers to detect ovarian cancer.

3052 Background: Genomic profiling through liquid biopsies (LB) has enabled precision oncology decision making, however a key challenge lies in critically interpreting LB data to optimize patient care. Methods: We report results from the first planned interim analysis of an observational biomarker trial, designed to evaluate the clinical utility of serial LB in patients with advanced/metastatic solid tumors (NCT05585684). Primary endpoints were to determine feasibility, prevalence of actionable alterations in LB and the fraction of patients with enacted genotype-matched therapies. Secondary endpoints included progression-free (PFS) and overall survival (OS), time to subsequent therapy and concordance between LB and tumor next generation sequencing (NGS). Exploratory endpoints included correlation of ctDNA dynamics with survival. Serial LBs were obtained at baseline, 1-3 weeks on therapy and at progression using a CAP/CLIA validated NGS panel (Labcorp, MD). Patient-matched white blood cell (WBC) NGS was utilized to identify clonal hematopoiesis (CH)-derived variants. Actionability of genomic alterations was assessed by an ensemble multi-resource programmatic approach; results were reviewed at the Johns Hopkins Molecular Tumor Board (JH MTB). Results: Between March 2023 and July 2024, 51 patients with NSCLC, SCLC and esophageal cancer were enrolled, with 45 evaluable baseline and 12 progression LBs reviewed at JH MTB. Median turnaround time from baseline and progression LB to MTB recommendation was 14 and 13 days respectively. Patient-matched analyses of baseline WBC samples revealed 30.1% (n = 22) CH-derived alterations. The frequency of actionable variants was 28.8% (n = 21) at baseline, 23.3% (n = 7) on therapy and 21.7% (n = 5) at progression. Of the 45 patients reviewed at baseline, 33 received a recommendation for genotype-matched therapies; 48.5% (n = 16) based on tumor molecular profiling, 15.2% (n = 5) based on LB alone and 36.3% (n = 12) based on LB and tissue NGS. Thirteen patients were treated according to MTB recommendations. Patients who were treated with genotype-matched MTB recommended therapies had longer OS and PFS compared to those who received alternate therapies (not reached-NR vs. 14.8 months, log-rank p = 0.028 and NR vs 6.2 months, log-rank p = 0.21 respectively). Among the 12 patients reviewed at progression, 5 received an MTB recommendation for genotype-tailored therapies based on LB alone (n = 3) or in combination with tissue NGS (n = 2). Early on-therapy ctDNA clearance was associated with longer PFS and OS (log rank p = 0.02 and p = 0.06). Conclusions: Our findings highlight the value of a multidisciplinary MTB when supported by comprehensive liquid biopsy molecular information to inform therapy selection and improve patient outcomes. Clinical trial information: NCT05585684 .

Sacituzumab is FDA approved for unresectable locally advanced or metastatic triple-negative or HR-positive/HER2-negative breast cancer in the metastatic setting. Sacituzumab’s ability to prevent or control CNS disease is not yet known. Thus, we investigated the patterns of CNS involvement in conjunction with Sacituzumab use in metastatic breast cancer. The Johns Hopkins Electronic Medical records from 4/2020-4/2024 were queried utilizing the following criteria: 1) Age ≥18 years; 2) Diagnosis and/or problem list including breast cancer; 3) ≥2 months of documented sacituzumab therapy. Patterns of disease progression, median progression-free (mPFS), and overall survival (mOS) was evaluated. Forty-three female patients (49% White, 35% Black, 12% Asian, 2% Hispanic) were identified. Median age was 59 years (range 33-85). Sacituzumab was the median third-line therapy (range 1-10). Median time on sacituzumab was 4.4 months (range 2.1-15.6 [IQR 2.9-6.2]). Seven patients (16%) had preexisting CNS involvement: 5 brain/spine parenchymal, 1 leptomeningeal, and 1 with both. mPFS was 6.6 months (range 3.4-14.5) in the preexisting CNS disease group versus 4.0 months (range 1.3-16.4) in the group without prior CNS involvement (log-rank p=0.13). Visceral progressions were the most common (30; 70%), followed by bone (18; 42%), whereas CNS progression was only observed in 5 (12%). 2 (29%) patients with preexisting CNS disease developed further CNS progression (4.4 and 4.9 months after Sacituzumab commenced). mOS was 13.1 months (range 6.9-23.2) in the preexisting CNS disease group, versus 10.9 months (3.0-26.5) among patients without prior CNS involvement (log-rank p = 0.17). Five (14%) of known causes of death were neurologic. This is a work-in-progress studying CNS metastatic patterns in metastatic breast cancer treated with Sacituzumab. Relatively low CNS progression rates were observed. Those with CNS involvement did not have shorter survival than those without. A detailed radiographic review of the metastasis cases is pending.

Alpelisib is FDA-approved for patients with HR-positive, HER-2 negative, PIK3CA-mutated metastatic breast cancer with progression on/after receiving endocrine therapy. We investigated the CNS progression patterns of patients on alpelisib. We queried the Johns Hopkins Electronic Medical records from 5/2019-4/2024 using the following criteria: 1) Age ≥18 years; 2) Diagnosis and/or problem list including breast cancer; 3) ≥2 months of documented alpelisib therapy. Patients were subsequently dichotomized based on the ICD10 codes for CNS metastases. Forty-three female patients were identified (74% White, 19% Black, 7% Asian, 2% Hispanic). Median age was 60 years (range 35-87). Alpelisib was the median third line therapy (total lines 2-13). Median time on alpelisib was 7.1 months (2.2-23.9 [IQR 3.6-10.3]). Six patients (14%) had preexisting CNS disease before alpelisib, and 37 (86%) had no previous CNS involvement. The non-CNS involvement patients were on drug for 7.9 months on average (range 2.2-23.9 [IQR 3.6-10.3]), whereas those with preexisting CNS disease were on drug 4.7 months (range 2.8-11.0 months). While on alpelisib, six patients (16%) developed new CNS disease (mean time of 6.1 months [range 3.5-9.5]). In the preexisting CNS disease group, only one patient had confirmed CNS progression in the leptomeninges after 3.8 months. A total of 7 (16%) patients suffered CNS progression; 4 involved dural/leptomeningeal, and 2 brain/spinal parenchymal and 1 progression in both. Overall survival from initiation was 12.2±11.2 months in the pre-alpelisib brain metastases group versus 18.9±12.3 months (log-rank p=0.03) in the non-CNS group. Two (16%) known causes of death were neurologic; both developed metastases while on alpelisib (5.8 and 9.0 months after initiation). This is a work-in-progress of CNS metastatic patterns in patients who received alpelisib, showing a leptomeningeal predominance in CNS progression. A detailed radiographic review of the metastasis cases is pending.

1009 Background: Patients with Stage II/III breast cancer that overexpresses the human epidermal growth factor-2 (HER2+) or is triple-negative (TNBC) generally receive upfront neoadjuvant therapy (NAT) before definitive surgery. Pathologic complete response (pCR) after NAT is associated with improved survival but a small proportion of patients remain at risk for recurrence. Circulating tumor DNA (ctDNA) in patients whose primary tumors have detectable mutations could improve the identification of patients who remain at risk after NAT. Methods: The Pathologic Response Evaluation and Detection In Circulating Tumor-DNA (PREDICT-DNA) trial was a prospective, multi-center study aimed at validating ctDNA as a biomarker for treatment response in Stage II/III HER2+ or TNBC. The primary aim was to determine the negative predictive value (NPV) of ctDNA for residual disease following NAT; secondary aims included five-year invasive disease-free survival (IDFS), which were fit to a Cox proportional hazards model. Mutations were identified from tumor tissue; ctDNA was then analyzed in pre- and post-NAT blood and compared with surgical pathology. Proposed sample size was 229 patients based on simulation to control expected half-width of a confidence interval on NPV to be ≤15% when NPV=90%. The Personalis NeXT Personal ctDNA assay was centrally performed. Results: 228 participants were enrolled in 24 sites between 2016 and 2018. 53% had TNBC, and 47% had HER2+ disease. 92.2% (n=166/180) had detectable ctDNA at baseline, and 46% of patients had pCR (42% TNBC, 50% HER2+). 54% of all post-NAT ctDNA detections were in the ultrasensitive range below 100 PPM. Among 112 subjects with undetectable ctDNA prior to surgery, 45 were found to have residual disease resulting in an NPV of 60% (CI 0.51-0.69). Patients with TNBC and detectable ctDNA prior to surgery were approximately 12 times more likely to experience a recurrence regardless of pCR (HR 12.8 [95% CI: 2.3-71.5]). See Table describing landmark IDFS analyses after surgery. Conclusions: While lack of ctDNA detection after NAT and before surgery did not predict pCR, initial analysis of predefined secondary objectives suggest that ctDNA-negative patients before surgery have excellent prognosis regardless of pCR, particularly if TNBC. This suggests that ctDNA may be a better biomarker for long term clinical outcomes than pCR. Further correlations and interactions will be presented. Clinical trial information: NCT02743910 . Invasive disease-free survival (IDFS) by breast cancer subtype, according to ctDNA after NAT and pathologic response. TNBC (total n=64) 3y IDFS(n=40) 4y IDFS(n=32) 5y IDFS(n=18) ctDNA- & pCR (n=20) 94.1% 94.1% 94.1% ctDNA- & RD (n=25) 95.8% 89.8% 89.8% ctDNA+ & RD (n=19) 48.9% 48.9% 48.9% HER2+ (total n=58) 3y IDFS(n=41) 4y IDFS(n=31) 5y IDFS(n=17) ctDNA- & pCR (n=19) 94.1% 94.1% 94.1% ctDNA- & RD (n=31) 92.6% 87.5% 87.5% ctDNA+ & RD (n=8) 60.0% 60.0% 60.0%

Gastroesophageal cancer dynamics and drivers of clinical responses with immune checkpoint inhibitors (ICI) remain poorly understood. Potential synergistic activity of dual programmed cell death protein 1 (PD-1) and lymphocyte-activation gene 3 (LAG-3) inhibition may help improve immunotherapy responses for these tumors. We report a phase Ib trial that evaluated neoadjuvant nivolumab (Arm A, n = 16) or nivolumab–relatlimab (Arm B, n = 16) in combination with chemoradiotherapy in 32 patients with resectable stage II/stage III gastroesophageal cancer together with an in-depth evaluation of pathological, molecular and functional immune responses. Primary endpoint was safety; the secondary endpoint was feasibility; exploratory endpoints included pathological complete (pCR) and major pathological response (MPR), recurrence-free survival (RFS) and overall survival (OS). The study met its primary safety endpoint in Arm A, although Arm B required modification to mitigate toxicity. pCR and MPR rates were 40% and 53.5% for Arm A and 21.4% and 57.1% for Arm B. Most common adverse events were fatigue, nausea, thrombocytopenia and dermatitis. Overall, 2-year RFS and OS rates were 72.5% and 82.6%, respectively. Higher baseline programmed cell death ligand 1 (PD-L1) and LAG-3 expression were associated with deeper pathological responses. Exploratory analyses of circulating tumor DNA (ctDNA) showed that patients with undetectable ctDNA post-ICI induction, preoperatively and postoperatively had a significantly longer RFS and OS; ctDNA clearance was reflective of neoantigen-specific T cell responses. Our findings provide insights into the safety profile of combined PD-1 and LAG-3 blockade in gastroesophageal cancer and highlight the potential of ctDNA analysis to dynamically assess systemic tumor burden during neoadjuvant ICI that may open a therapeutic window for future intervention. ClinicalTrials.gov registration: NCT03044613 .

1042 Background: Circulating tumor DNA (ctDNA) can overcome limitations of tissue biopsy (tbx) in patients with metastatic breast cancer (mBC). While tbx provides a snapshot of a tumor’s molecular profile from one site and time, ctDNA provides a more comprehensive landscape of the tumor genome. Serial liquid biopsies (lbx) can be easily obtained for monitoring clinical response and resistance over time. The Individualized Molecular Analyses Guide Efforts in Breast Cancer (IMAGE) II study evaluates tbx and serial lbx methods for tumor genomic profiling in mBC. Methods: IMAGE II (NCT02965755) is a prospective, multi-center trial to evaluate the clinical utility of lbx versus tbx. Eligible individuals had mBC of any subtype, with progression requiring therapy change. Tbx sequencing per standard of care was required unless biopsy was not technically feasible. Upon enrollment, we collected baseline plasma for tissue-agnostic comprehensive genomic profiling (CGP) by Foundation Medicine. Both lbx and tbx sequencing results were reviewed in real time by the Johns Hopkins Molecular Tumor Board to provide treatment recommendations. Clinical data was recorded. We collected serial lbx samples 1-2 weeks post initiation of next line of therapy, at first restaging, and at progression. Here we present the analysis of data collected at enrollment, and serial lbx results will be reported at a later date. Results: Total enrollment was 199 pts, of whom 194 were women (median age = 57 years, range 27-86 years). There were 140 HR+HER2-, 4 HR-HER2+, 20 HR+HER2+ and 35 triple negative (TNBC) cases. Metastases were found in bone only in 19 pts, or included lung in 99 pts, liver in 97 pts, and brain in 14 pts. Median prior lines of metastatic therapy was 2, with 44 pts (22.1%) having been on 4 or more, including antibody-drug conjugates (14 pts), immunotherapy (2), cytotoxic chemotherapy (74), endocrine (88), and targeted therapy (12). 9 pts were newly diagnosed with mBC at the time of enrollment. 110 pts (55%) had tbx sequencing. In lbx-based CGP results from 190 patients, the genes most frequently altered were TP53(90 patients), PIK3CA (63), ESR1 (54). Copy number amplifications were found in FGFR1 (21 patients) and FGFR3 (12), and copy number losses were found in PTEN in 3 patients. Of the 87 patients whose baseline lbx was profiled using FoundationOne Liquid CDx which allows calculation of ctDNA tumor fraction (TF), 69 (79.3%) pts had detectable ctDNA TF. 60 patients (69.0%) had ctDNA TF at least 1%. We will present the frequency of clinically actionable mutations detected via lbx vs tbx genomic profiling, accounting for sites of metastases, receptor subtype and prior therapy exposure, as well as the association between ctDNA TF and sites of disease. Conclusions: A high rate of ctDNA TF detection in pts with mBC highlights the value of lbx for CGP in this setting for detecting multiple alteration classes across mBC subtypes and metastatic sites. Clinical trial information: NCT02965755 .

Modern generative artificial intelligence techniques like retrieval-augmented generation (RAG) may be applied in support of precision oncology treatment discussions. Experts routinely review published literature for evidence and recommendations of treatments in a labor-intensive process. A RAG pipeline may help reduce this effort by providing chunks of text from these publications to an off-the-shelf large language model (LLM), allowing it to answer related questions without any fine-tuning. This potential application is demonstrated by retrieving treatment relationships from a trusted data source (OncoKB) and reproducing over 80% of them by asking simple questions to an untrained Llama 2 model with access to relevant abstracts.

Abstract Background: Circulating Tumor Cells (CTC) can be isolated in 40-80% of patients with metastatic breast cancer (MBC). High levels of CTC, defined as ≥5/7.5 ml of blood are associated with shorter progression-free survival and overall survival. CTC can be tested for clinically relevant biomarkers, such as ER, and HER2 status, and have the potential to guide therapy as well as help with disease monitoring. We investigated CTC detection rate, its correlation with clinical characteristics, pathological characteristics, and response to treatment in patients with any type of metastatic breast cancer who progressed on at least one line of therapy. Methods: Eligible patients with MBC were enrolled in the IMAGE-II study (Individualized Molecular Analyses Guide Efforts in Breast Cancer) NCT02965755, in which we obtained archival tissue as part of the standard of care, and prospectively collected serial blood samples for CTC analysis. We compared CTC levels and biomarker status with clinical factors, tissue-based pathology, and molecular biomarkers. We assessed whether changes in CTC count correlated with treatment response. Blood samples were collected at baseline (Day 1), after 1-2 weeks of therapy, after 8-12 weeks, and at subsequent restaging (every 8-12 weeks). Medical records were reviewed every 3 months for ongoing treatment response, and death. Samples were processed at the Biocept CLIA-certified Laboratory. CTC were enumerated by the presence of CD45- and DAPI+ cells. Further CTC characterization was performed by antibody staining of specific protein biomarkers (ER, HER2). Fisher’s exact test was used to evaluate the association between baseline CTC &lt; 5 and ≥5/7.5ml with patient characteristics. McNemar’s test was used to assess discordance between tissue-based and CTC-defined markers. Results: Between 1/26/2018 and 12/31/22 baseline samples were collected from 70 women with a median age of 56 (36-82) who were enrolled at four study sites. CTC was detected in 59 (84%) of participants. Baseline CTC counts were &lt; 5/7.5 ml in 37/59 (63%) and ≥5/7.5 ml in 22/59 (37%) of participants. Differences between breast cancer subtype and CTC-receptor status were observed (Table 1). Among the 41 patients who were ER-positive, 31 (76%) were CTC ER-negative(p&lt; 0.001). Among 29 patients who were ER-negative, 4 (14%) were CTC ER-Positive (P&lt; 0.001). Among the 7 patients who were HER2 positive, 2(29%) were HER2 positive on CTC. Among 63 patients who were HER2 negative, 7(11%) had HER2 expression on their CTC. The discordance in the classification of HER2 wasn't statistically significant (p=0.77). Elevated CTC at baseline was more frequently detected in younger participants (&lt; 50 years old) (55% vs 27%; P=0.22), in Black women compared to White (60 % vs 29 %, P=0.27), and in participants with visceral vs non-visceral metastasis (52% vs 28 %, P=0.40). Patients with CTC ≥5 vs &lt; 5/7.5mL at baseline had a shorter duration of anti-HER2-based therapy (61.5 vs 836 days, p=0.04). There were no statistically significant differences among participants who received chemotherapy agents (316.5 vs 365.5 days, P=0.88), endocrine therapy (376 vs 490 days, P=0.89), or overall therapy (272 vs 390 days, P=0.21). Conclusions: We observed significant differences in the expression of ER between tumor tissue and CTC, which can be partly due to tumor evolution over time. Additionally, participants who are young, Black, and those who have visceral metastasis may have higher CTC counts. Higher CTC counts were associated with a shorter duration of anti-HER2 therapy. Although it didn’t meet statistical significance, a similar trend was observed in patients who received chemotherapy and endocrine therapy. Table 1: MBC Tissue-Based Subtype and CTC-receptor status at baseline Citation Format: Asnakech Bayable, Tingchang Wang, Amanda Blackford, Jessica Tao, Jenna Canzoniero, Seoho Lee, Faith Too, Barbara Blouw, Mary Wilkinson, Rima Couzi, Antonio Wolff, Christie Hiton, Ben Park, Vered Stearns, Cesar Santa-Maria. Correlation of Circulating Tumor Cells (CTC) with Clinical Characteristics, Pathological Factors, and Treatment Response in Patients with Metastatic Breast Cancer (MBC) [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO1-16-02.

There is a critical need for a streamlined process to identify genotype-matched individuals eligible for enrollment into clinical trials and/or targeted therapies, as current methodologies face challenges in integrating diverse molecular data sources. We have developed a precision oncology platform to assist molecular tumor boards and community oncologists in reviewing patients' phenotypes, evaluating related knowledge, and identifying genotype-matched therapies.

Abstract Background: The expanding role of next-generation sequencing (NGS) in precision oncology has led to a paradigm shift towards genotype-targeted therapies, with liquid biopsies (LBs) offering a minimally invasive means of comprehensive genomic profiling. Methods: This is an observational LB-informed precision oncology clinical trial with primary endpoints to assess feasibility and determine the prevalence of actionable mutations in LB informing genotype-targeted therapeutic recommendations by the Johns Hopkins Molecular Tumor Board (JH MTB; NCT05585684). Secondary endpoints are progression-free and overall survival, ctDNA dynamics assessment, and concordance between plasma and tumor NGS. Target enrollment is 150 patients with advanced solid tumors who have progressed on at least one line of systemic therapy. Serial LB-derived genotypes are obtained at baseline, early on therapy, and upon progression using a CLIA ultra-sensitive 33-gene panel NGS assay (Elio Plasma Resolve, Personal Genome Diagnostics/Labcorp). Variant actionability is assessed by an ensemble knowledgebase, registry, and computational characterization approach, and results are reviewed by the JH MTB, followed by treatment recommendations. Results: The trial’s pre-specified interim analysis included 30 enrolled patients with evaluable LBs; 30 baseline, 13 on-therapy, and 6 progression LBs were analyzed. Twenty-five patients had non-small cell lung cancer among which 9 were EGFR-mutated, 1 patient had small cell lung cancer, 3 patients had gastro-esophageal cancer, and 1 patient had a neuroendocrine tumor. In total, 74 variants were identified across all time points; these were cross-referenced in somatic and germline registries, variant knowledgebases, and the biomedical literature, together with computational characterization of functional consequence to assign mutation actionability. Tumor-confirmed variants comprised 36% of baseline and 42% of both on-therapy and progression variants. At baseline, 20% of the variants were putatively clonal hematopoiesis-derived with similar findings at the on-therapy timepoint. Thirty percent of baseline variants and 33% of variants at progression were deemed actionable. Upon JH MTB review, 29% of patients received a genotype-matched therapeutic recommendation based on LB results, with MTB recommendations followed in 38% of the patients. In analyzing ctDNA dynamics, of the 13 patients with on-therapy samples, 23% showed ctDNA clearance, 31% showed persistence, while 46% had undetectable ctDNA at both time points. Among patients with LBs at the time of progression, a third showed ctDNA emergence, a third had ctDNA persistence, and the remainder had undetectable ctDNA throughout the follow-up period. Conclusion: Our findings highlight the clinical relevance of liquid biopsies in guiding treatment selection and monitoring. Citation Format: Maria Fatteh, Michael Conroy, Timsy Wanchoo, Selina Shiqing Teh, Jaime Wehr, Archana Balan, Rachel Karchin, Ellen Verner, Katerina Karaindrou, Rena Xian, Christopher Gocke, Ming-Tseh Lin, Peggy Fitzpatrick, Christine L. Hann, Joseph C. Murray, Vincent K. Lam, Josephine L. Feliciano, Julie R. Brahmer, Kristen A. Marrone, Patrick M. Forde, Katie Fiallos, Dana Petry, Taxiarchis Botsis, Mark Sausen, the Johns Hopkins Molecular Tumor Board Investigators, Jenna Canzoniero, Jessica Tao, Valsamo Anagnostou. Liquid biopsy-informed precision oncology clinical trial to evaluate the utility of ctDNA comprehensive genomic profiling [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 970.

Polyclonal convergent evolution to PARPi resistance in a patient with metastatic breast cancer with gPALB2.

TPS624 Background: In patients with triple negative breast cancer (TNBC) the addition of pembrolizumab to neoadjuvant chemotherapy has improved pathologic complete response rates (pCR) and 5-year event free survival (EFS). Patients with TNBC who achieve a pCR have a significantly lower risk of recurrence compared to those with residual disease (RD). However, pCR can only be assessed at the time of surgery after completion of 6 months of intensive chemo-immunotherapy, associated with up to 77% high-grade treatment-related toxicity. Identifying pCR early on is imperative in developing optimized and individualized treatments, however, there are no predictive biomarkers. Studies in HER2-positive breast cancer have shown that early fluorodeoxyglucose positron emission tomography (FDG-PET) changes soon after initiation of neoadjuvant treatment can help predict the pathological outcomes at the time of surgery and also the risk of recurrence. Emerging data also support that clearance of circulating tumor DNA (ctDNA) during neoadjuvant treatment may be associated with higher rates of pCR. We thus designed a study to determine the association of early FDG-PET changes and ctDNA clearance with pCR in early-stage TNBC. Methods: NeoADAPT is a phase 2, open label, single center clinical trial including adult patients with previously untreated clinical stage 2 or 3 TNBC. Participants will receive 4 cycles of neoadjuvant paclitaxel/carboplatin with pembrolizumab (TC/pembro), per standard-of-care (SOC). An FDG-PET will be performed at baseline and after 1 cycle. A breast MRI will be performed at baseline and after completion of 4 cycles of treatment. Patients with a complete clinical response (cCR) will proceed to surgery and will forgo more neoadjuvant treatment. Patients with clinical RD on MRI will complete SOC neoadjuvant therapy with 4 cycles of doxorubicin/cyclophosphamide (AC) with pembro prior to surgery. If RD is identified after surgery, adjuvant therapy will be determined by the treating oncologist, which may include pembro with AC (if not given prior), capecitabine, etc. The primary objective is to evaluate if lack of decrease in maximum FDG-PET standard uptake value of lean body mass (SULmax) by less than 40% after 1 cycle of neoadjuvant treatment correlates with RD at the time of surgery (i.e. determining the negative predictive value of FDG-PET). Additional key endpoints include ctDNA clearance, evaluation of pCR and cCR rates in the treatment population and EFS. Distributions of percent reduction in SULmax will be compared between pCR and no-pCR groups using the Wilcoxon rank sum test. Receiver operating characteristic (ROC) curve analysis will be performed to estimate the ability of FDG-PET SULmax reduction and cCR by MRI in predicting pCR at surgery. Clinical trial information: NCT06245889 .