Supplementary Table S7. Tissue informed clinical management in EGFR-mutant NSCLC.

Impact of pCGP and ctDNA dynamics on patient management. A, Swimmer’s plot illustrating the survival of 12 patients with EGFRm NSCLC; these patients were chosen to provide a representative example of situations in which pCGP beyond the first diagnosis was informative for patient management. Examples of how pCGP informed management include the identification of T790M alterations (patients PCGP1, PCGP4, PCGP5, PCGP7, PCGP12); identification of a BRAF alteration (PCGP6); identification of MET amplification (PCGP8 and PCGP9); and identification of a C797S resistance alteration (PCGP2). B and C, The emergence of resistance on pCGP; fish plots illustrate a dynamic, longitudinal representation of molecular alterations identified on pCGP in patients with serial sampling. The Y axis represents the aggregate VAF of all alterations identified at each time point. Each vertical dashed line represents the time point of a pCGP assay, with the number beneath representing the VAF of the most prevalent alteration at that time. B, pCGP serial profiling for a patient with EGFR L858R–driven NSCLC who had five iterations of pCGP. They were on erlotinib up to the second pCGP, which was obtained at the time of radiographic progression. This identified EGFR T790M, and the patient was switched to osimertinib with subsequent suppression of this clone. The third pCGP was at the time of further radiographic progression, no new driver of resistance was identified but EGFR C797S was identified on liver biopsy. They were transitioned from osimertinib to a clinical trial at this point, with corresponding re-expansion of T790M and L858R variants after. Fourth pCGP, after carboplatin/pemetrexed/bevacizumab therapy, identified both C797S and L718Q, an uncommon mutation in EGFR and an additional resistance mechanism. Their final pCGP demonstrated a significant increase in the aggregate VAF of all alterations, and the patient died 2 months later. Separate regimens are separated by semicolons. C, pCGP serial profiling for a patient with ALK-fusion NSCLC that had four iterations of pCGP. Their disease had become resistant to alectinib at the time of the first pCGP, likely attributable to ALK E1210K and I1171T alterations. I1171T is associated with sensitivity to ceritinib, and the patient was switched to ceritinib at this time. After some clinical response with a corresponding contraction of the I1171T variant, expansion of the E1210K variant was seen in addition to an acquired oncogenic PIK3CA resistance variant. The third pCGP is at the time of progression on ceritinib. The patient transitioned to brigatinib initially but progressed after 6 months and transitioned to lorlatinib. Fourth pCGP after 22 months on lorlatinib identified an additional acquired on-target mutation in ALK G1269A. Compound mutations of this kind are common in patients on lorlatinib. Separate regimens are separated by semicolons.

Figure S4. Categorization of EGFR driver variants identified in the EGFRm NSCLC subset.

Supplementary Table S1: Training cohort description with plasmaCHORD feature values. Supplementary Table S2: Description of the serial cohort. Supplementary Table S3: Validation cohort description with plasamCHORD feature values. Supplementary Table S4: Description of plasmaCHORD features. Supplementary Table S5: Correlation of cfDNA fragmentomic feature statistics between baseline and follow-up time points in the serial cohort. Supplementary Table S6: Possible CH-origin variants identified on retrospective analysis of 45 patients reviewed at the JH-MTB.

Performance of individual features in predicting variant origin in the training cohort. Receiver-operator curves and area-under the curve (AUC) values for individual model features as predictors for CH versus tumor-origin. The top performing individual features are gene heme fraction, length cluster mutant, and length max delta S. However, no individual feature performs as well as the plasmaCHORD model combining all features.

Description of the training cohort. (A) Distribution of cancer types, stage, and patient age in the training cohort. The mean age for patients in this cohort was 65 years old (range 19-87), the most common cancer type was colorectal, and most patients had early stage disease (22 stage I, 101 stage II, 48 stage III, 53 stage IV, and 1 unknown stage). (B) Distribution of WBC only versus WBC and tumor matched sequencing and determined reference variant origin. The proportion of variants with tumor-derived reference origin is slightly higher from samples that had both WBC and tumor sequencing versus WBC sequencing alone. (C) Distribution of variants included in training cohort by gene and reference variant origin. After DNMT3A, TP53 was the second-most commonly mutated gene (98 variants) of which 67 (68%) are tumor-derived and 31 are WBC-derived.

Figure S3. Actionable alterations identified on first pCGP at first diagnosis (treatment-naive setting).

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.

Supplementary Table S10. Drivers of EGFR TKI resistance.

Supplementary Table S1. Context of first pCGP assays.

Characteristics of study population.

Supplementary Table S5. Summary of tissue informed clinical management in patients with concurrent pCGP and tissue biopsy.

Abstract Plasma comprehensive genomic profiling (pCGP) is implemented in the clinical care of non–small cell lung cancer (NSCLC). Although less well documented, serial pCGP may also guide the management of oncogene-driven NSCLC after first progression. In this study, we assessed the clinical value of serial pCGP, focusing on EGFR-mutant NSCLC. We conducted a retrospective study of 718 patients with NSCLC who underwent pCGP between 2015 and 2022. Clinical genomic data were programmatically extracted from the data workflows of the Johns Hopkins Lung Cancer Precision Medicine Center of Excellence. After variant annotation and actionability characterization, we examined the prevalence and evolving comutation patterns across serial pCGP, focusing on genomic mechanisms of tyrosine kinase inhibitor (TKI)–acquired resistance in EGFR-mutant NSCLC. A total of 718 patients had 818 instances of pCGP, with 79 patients having longitudinal pCGP (range, 2–5). pCGP uniquely informed management in 13% of patients (n = 92), both at initial diagnosis and on serial genotyping. This occurred predominantly through the identification of actionable mutations when tissue testing was unavailable. Among 214 patients with EGFR-mutant NSCLC, pCGP identified PI3K pathway alterations in 11% after first-line therapy. BRAF V600E (3%) and MET exon 14 skipping mutations (3%) emerged after third-generation TKI therapy. After TKI progression, 31 patients (22%) with EGFR-mutant disease had actionable pCGP findings, of whom 18 (58%) were matched to targeted therapy. Serial pCGP can inform treatment decisions in patients with NSCLC. In those with EGFR-mutant disease, pCGP at progression identifies actionable drivers of therapy resistance, enabling therapeutic intervention. Significance: Our study provides critical insights into the routine implementation of serial pCGP within a thoracic oncology program, supported by a precision oncology informatics framework, in a tertiary healthcare institution. We show that pCGP enables genotyping when tissue testing is not feasible and identifies actionable mutations at resistance. The clinical implementation of pCGP can drive improved clinical outcomes by matching patients with effective interventions in a timely and minimally invasive manner.

Therapy resistance remains a leading cause of treatment failure and mortality in thoracic malignancies, despite major advances in targeted therapies and immunotherapies. Resistance evolves through heterogeneous, patient-specific mechanisms, including genetic alterations, epigenetic reprogramming, lineage plasticity, and tumor-microenvironment interactions, often emerging before radiographic progression or clinical relapse. Conventional tissue biopsies and imaging, while essential for diagnosis and treatment selection, are limited in their ability to capture spatial and temporal tumor heterogeneity or to support dynamic monitoring of tumor evolution. Liquid biopsy has therefore emerged as a minimally invasive approach for real-time assessment of tumor-derived biomarkers in circulation, enabling longitudinal tracking of resistance biology across the cancer care continuum. In this review, we highlight recent advances in liquid biopsy applications for thoracic malignancies, focusing on circulating tumor DNA (ctDNA) and circulating tumor cells (CTCs) as complementary analytes for baseline molecular profiling, detection of primary and acquired resistance to therapy, and identification of minimal residual disease and molecular relapse. Beyond mutation-based approaches, we highlight emerging non-genomic ctDNA features, including epigenomic and fragmentomic signatures, that capture treatment-induced adaptation and lineage plasticity not detectable by conventional plasma genomic profiling, and discuss advances in CTC technologies that preserve cellular and phenotypic context relevant to resistance and metastatic potential. Finally, we examine multimodal liquid biopsy strategies that integrate multiple circulating analytes with artificial intelligence-assisted methods to enhance sensitivity, provide a more comprehensive view of tumor biology, and inform adaptive therapy strategies. We also outline key analytical and clinical challenges that must be addressed through standardized, prospective trials to translate liquid biopsy-guided surveillance and early interception of resistance into improved patient outcomes.

e16056 Background: Advanced esophageal and gastroesophageal junction (E/GEJ) tumors carry a poor prognosis and reliable early biomarkers for therapeutic efficacy are lacking. The utilization of ctDNA monitoring longitudinally is not standard of care at present. We hypothesized that a novel, tumor-naive, methylation-based pan-cancer assay could measure disease burden and provide a signal of early molecular response in E/GEJ patients. Methods: A Phase I/II trial (NCT04921904) evaluated the safety and efficacy of abemaciclib combined with ramucirumab in metastatic/recurrent E/GEJ adenocarcinoma patients who had progressed after frontline therapy. Primary endpoints focused on safety, with secondary endpoints including response, survival, and molecular profiling. Peripheral blood samples were collected from 20 enrolled patients pre-treatment and within 30 days on-treatment. These were analyzed using a methylation-based ctDNA assay that quantifies longitudinal changes across > 500 uniquely methylated loci to yield a Tumor Methylation score and the association of this score with therapeutic response, to assess changes in disease burden. Results: Safety and efficacy have previously been reported with a disease control rate of 40% and a duration of clinical response of 6.4 months in heavily pretreated (70% of patients being third line therapy or higher) advanced metastatic E/GEJ. 13 of 20 patients provided samples to assess ctDNA. We found that early molecular progressive disease (mPD), was significantly associated with a shorter progression-free survival (PFS) measured using RECIST v1.1 (Hazard Ratio = 4.9, 95% Confidence Interval 1.1-22). Notably, this association trended toward improved OS, yet not significant. Further analysis comparing patients with rapid progression (PFS ≤ 5.2 months) versus non-rapid progression revealed significant differences in baseline methylation patterns at 128 specific loci, including CCNA1 and CHL1. Conclusions: This exploratory study demonstrates that a tumor-naive, methylation-based ctDNA approach may be a viable tool for monitoring CDK4/6 pathway response in advanced E/GEJ cancers. Early mPD detection at one month by ctDNA kinetics strongly correlates with shorter PFS, offering a rapid signal of efficacy. This provides a substantial advantage over traditional radiographic evaluations (e.g., RECIST) and expands ctDNA's utility for response evaluation. Distinct baseline methylation patterns in rapid progressors highlight the potential of upfront epigenetic profiling to risk-stratify patients pre-treatment, guiding treatment selection and avoiding unnecessary toxicities. These findings suggest broad applicability for the tumor-agnostic approach targeting CDK4/6 specific alterations across various heavily pretreated solid tumors. Clinical trial information: NCT04921904 .

Figure S6. Plasma-only variants among patients with simultaneous plasma and tissue NGS.

Abstract Introduction: Overcoming immunotherapy (IO) failure in advanced NSCLC requires distinguishing molecular features of acquired vs primary resistance. Here we used comprehensive multi-omic analyses to define resistance mechanisms. Methods: Analyses were performed on the largest-to-date IO resistance cohort: 1819 biospecimens from 892 patients with NSCLC (371 primary, 521 acquired) from the Phase 2 HUDSON study (NCT03334617) of combination regimens after progression on anti-PD-(L)1/chemotherapy. Targeted next-generation sequencing of unpaired tumor biopsies pre- (n=226) and post-IO (n=497) was used to profile mutation signatures, clonality, aneuploidy and genomic instability. Post-IO tumor burden was assessed by plasma ctDNA (n=445); peripheral T-cell repertoires were analyzed (n=335). Bulk (n=205) and single-cell (sc, n=75) RNA sequencing (RNAseq) of unpaired post-IO tumor biopsies enabled gene set enrichment analysis (GSEA) and high-resolution cell type annotation. Results: Comprehensive analyses revealed acquisition of sub-clonal genomic alterations at the time of acquired resistance. Despite lower systemic ctDNA burden at progression, activating FGF10 and RICTOR and inactivating RBM10 and MSH6 mutations were enriched in acquired vs primary resistance (false discovery rate [FDR] p<0.05). Post-IO tumors harbored more CDK4, CDK6, and CD22 activating mutations, and inactivating mutations in KDM6A, SMARCA4, and CDKN2A (FDR p<0.05). Genomic instability (increased homologous recombination deficiency signatures, elevated large-scale transitions, telomeric allelic imbalance) was noted in both primary and acquired resistant tumors. Bulk RNAseq GSEA detected upregulation of epithelial-to-mesenchymal transition (EMT), IFNγ response, and inflammatory pathways (FDR p<0.05) in acquired vs primary resistant tumors. Single-cell transcriptomics showed enrichment of tumor-reactive, tissue-resident memory CD8+ T-cell clusters in acquired resistant tumors. Notably, a naïve/stem-like CD8+ T-cell cluster was also enriched in acquired resistant tumors. GSEA in early, central, and tissue-resident memory CD4+ T-cell clusters revealed an upregulation of naïve/stem-like gene sets and a downregulation of antigen processing/presentation gene sets in acquired resistant tumors. scRNAseq and differential expression analysis of epithelial populations highlighted pronounced EMT activation, increased lineage plasticity, and neuroendocrine differentiation gene signatures, implicating cellular reprogramming and phenotypic plasticity as potential contributors to acquired IO resistance. Conclusion: Acquired IO resistance in NSCLC involves dynamic and unique genomic and transcriptomic remodeling, encompassing EMT, lineage plasticity, stem-like programs, and immune reprogramming—highlighting potential avenues for therapeutic intervention. Citation Format: Archana Balan, Sonia Iyer, James Conway, Christopher Cherry, Noushin Niknafs, Mohamed Reda Keddar, Avinash Reddy, Robert McEwen, James White, Grace Kim, Anissa Dallmann, Nima Boluriaan, Sreeharsha Gunda, Mark Awad, Glenwood Goss, Se-Hoon Lee, Keunchil Park, Martin Reck, Michael Thomas, Rachel Karchin, Jane Peters, John F. Kurland, Giuseppe Galletti, Simon T. Barry, Jan Cosaert, J. Carl Barrett, Benjamin Besse, John V. Heymach, Patrick M. Forde, Valsamo Anagnostou. Multi-modal multi-omic analyses reveal mechanisms of immunotherapy resistance in non-small cell lung cancer (NSCLC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(8_Suppl):Abstract nr CT233.

Supplementary Table S8. On-target acquired resistance in EGFR-mutant NSCLC.

Figure S5. Mechanisms of resistance to EGFR TKI according to first line of therapy received.

Supplementary Table S9. Putative uncommon on-target drivers of EGFR TKI resistance.