Abstract Introduction: Anti-EGFR-based regimens are standard treatments for left-sided, RAS/BRAF wild-type metastatic colorectal cancer (mCRC), but current biomarkers, including RAS/BRAF mutation status and tumor sidedness, do not fully explain the variability in treatment response. Prior tissue-based studies suggest that genome-wide DNA methylation patterns define biologically distinct CRC subtypes. We aimed to develop and validate a circulating tumor DNA (ctDNA)-based methylation classifier to predict anti-EGFR efficacy in mCRC patients. Methods: Genome-wide methylation profiles from >3,000 CRC liquid biopsy samples were analyzed using Guardant360 Liquid (Guardant Health, Palo Alto, CA). After tumor fraction normalization, unsupervised clustering identified two reproducible subtypes: One cluster exhibited a globally high-methylation pattern consistent with the hypermethylated colorectal cancer (HMCC) state described in prior tissue-based studies, while the other represented a low-methylated colorectal cancer (LMCC) state. A random forest model trained on the top 50 principal components (PCs) of these methylation profiles was applied to a real-world cohort of patients receiving anti-EGFR or anti-VEGF therapies. Real-world survival, measured by time to treatment discontinuation (rwTTD), was assessed by treatment type, methylation status, and genotype (RAS/BRAF wild-type vs mutant). Results: Methylation-based clustering successfully stratified mCRC into HMCC and LMCC groups in samples with evaluable tumor fractions (≥0.5%, N=162). HMCC tumors were enriched for BRAF mutations and exhibited significantly shorter PFS on anti-EGFR therapy, particularly among right- or unknown-sided patients. LMCC patients demonstrated improved outcomes on anti-EGFR therapy (HR = 1.76, p = 0.024), including in patients with right- or unknown-sided, RAS/BRAF wild-type disease, with survival comparable to patients with left-sided disease (HR = 0.94, p = 0.85). Combing methylation status with RAS/BRAF genotype and tumor sidedness expanded the anti-EGFR-eligible population by 18.8% (from 101 to 120 patients) while maintaining survival comparable to standard eligible patients. Conversely, no significant survival difference was observed between HMCC and LMCC patients treated with anti-VEGF therapy (HR = 1.08, p = 0.64), confirming methylation status as a predictive biomarker for treatment selection rather than a purely prognostic factor. Conclusions: ctDNA methylation profiling enables expanded prediction of anti-EGFR efficacy in mCRC beyond current selection criteria. Methylation-integrated eligibility captures additional patients who might benefit from anti-EGFR therapy that are missed by current sidedness-based guidelines. These findings support ctDNA methylation as a feasible and scalable predictive biomarker for precision therapy in colorectal cancer. Citation Format: Xuwen Li, Mingyang Cai, Shile Zhang, Nicole Zhang, Reagan Barnett, Tingting Jiang, Kota Ouchi, Yoshiaki Nakamura, Kimberly Banks, Justin Odegaard, Darya Chudova. Methylation signatures from liquid biopsies predict anti-EGFR therapy response in patients with colorectal cancer [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 2584.

3070 Background: Short read lengths in next-generation sequencing and targeted panel probe design pose challenges to fusion detection by impairing the resolution of complex genomic rearrangements and the accurate mapping of intronic breakpoints. To address these limitations, we developed a cell-free DNA (cfDNA) methylation-based fusion epigenotyping method that leverages fusion-associated tumor epigenetic signatures to complement genomic-based methods and increase the fusion detection sensitivity of our liquid biopsy. We validated EML4 - ALK fusion detection by this algorithm in non-small cell lung cancer (NSCLC) using paired clinical tumor tissue and cfDNA samples. Methods: We trained a binary classifier to discriminate EML4-ALK fusion-positive from fusion-negative NSCLC using cfDNA methylation signal and genomic molecule support. To validate the epigenotyping classifier, an independent cohort of 577 clinical NSCLC samples was selected with paired tumor tissue and Guardant360 Liquid (Guardant Health, Palo Alto, CA) cfDNA for each patient (with epigenomic tumor fraction > 0.03%). The cohort included 94 tissue-confirmed fusion-positive samples (78 cfDNA genomic positives and 16 cfDNA genomic false negatives) with EML4-ALK detected in tumor tissue, and 483 tissue-confirmed fusion-negative samples ( ALK fusion negative in both tissue and paired cfDNA). The epigenotyping classifier predictions in cfDNA were evaluated against tissue-based orthogonal truth. Results: Among tissue-confirmed fusion-positive samples, tissue-liquid assessment showed a high positive percent agreement (PPA) / sensitivity of 89.36% (84/94), as well as 100% concordance with genomic caller positives (78/78). The rate of rescued fusions by the epigenotyping classifier from genomic false negative liquid cases was 38% (6/16). Among tissue-confirmed fusion-negative samples, tissue-liquid assessment showed a high negative percent agreement (NPA) / specificity of 99.38% (480/483). The false positive rate (FPR) of high confidence calls (probability exceeding a 99.7% specificity threshold or with partial genomic evidence) was 0.0% (0/483). Conclusions: cfDNA methylation-based fusion epigenotyping substantially increased detection of actionable ALK fusions while maintaining high specificity, as demonstrated by tissue-liquid concordance. The results of this approach showed the clinical value of giving NSCLC patients an increased likelihood to receive more effective, less toxic ALK inhibitor therapy, while the minimized FPR helps ensure appropriately matched treatment decisions.

e20581 Background: Immune checkpoint inhibitors (ICIs) have transformed care for advanced non-small cell lung cancer (aNSCLC), but predictive biomarkers remain imperfect. We previously developed a multimodal immunotherapy response score (MIRS), which integrated ctDNA epigenomic signatures with microsatellite instability (MSI) and tumor mutational burden (TMB) from a single blood draw, and validated its use as a predictive biomarker for ICI response. In this study, we assess the ability of MIRS to predict clinical outcomes in real-world patients treated with ICIs. Methods: MIRS was trained and validated using de-identified patient data from InfinityAI Data Library and expressed as percentile based on MIRS distribution with >20,000 aNSCLC samples. In this study, we validated the signature in an independent cohort of 32 aNSCLC patients by analyzing baseline plasma on Guardant360 Liquid (Guardant Health, Palo Alto, CA); 27/32 received PD-1/PD-L1 monotherapy and 7/32 had tumor PD-L1 TPS <1%. Patients with MIRS ≥50% were defined as MIRS-High. The primary endpoint was real-world progression-free survival (rwPFS). We fitted Cox proportional hazards models (adjusted for covariates such as sex, age, tissue PD-L1 expression, histologic subtype, ECOG, TNM stage and baseline methylation tumor fraction) and reported adjusted hazard ratios (aHR) with 95% CIs. Median rwPFS was estimated by Kaplan–Meier method. Model discrimination was summarized using concordance index (c-index). Results: MIRS-High patients (18/32) had significantly longer rwPFS (median 15.1 vs 7.2 months; aHR 0.24, 95% CI 0.07–0.82, p=0.02; c-index = 0.80). As a continuous variable, MIRS percentile was associated with improved rwPFS (p=0.04). Stratification by PD-L1 ≥50% (n=14) showed a consistent, but not statistically significant trend toward longer rwPFS (aHR 0.60, 95% CI 0.19–1.87, p=0.38; c-index = 0.75). When both biomarkers were combined, patients with PD-L1 ≥50% or MIRS-High (n=25) had significantly longer rwPFS (aHR 0.24, 95% CI 0.07–0.90, p=0.034). All complete responses (n=4) were MIRS-High (median MIRS score of 80th percentile in aNSCLC); patients with progressive disease (n=6) had a median MIRS of 27th percentile in aNSCLC (partial responders (n=15) had a median of 51th percentile in aNSCLC, and stable disease cases (n=7) had median of 55th percentile in aNSCLC). Conclusions: Patients with MIRS-High scores had a 76% lower adjusted hazard of progression or death even after accounting for key covariates. Additionally, all complete responses were MIRS-High and MIRS was associated with significantly improved PFS compared to known biomarkers such as tissue-based PDL1. These data indicate that the multimodal score taken via a single baseline blood sample strongly predicts ICI benefit and may be useful in ICI vs chemo combination decisions pending further ongoing validation.

Abstract Introduction. Immune checkpoint inhibitors (ICIs) have transformed cancer treatment, yet identifying patients most likely to benefit remains challenging. Established biomarkers such as tumor mutational burden (TMB) and microsatellite instability (MSI) have performance limitations. We hypothesized that combining MSI and TMB with epigenomic features of tumor-intrinsic immune regulation and the tumor microenvironment from pretreatment plasma would improve prediction. We developed and validated a multimodal immunotherapy response score (MIRS-Score) that integrates MSI, TMB, and epigenomic signatures (Guardant360 Liquid) to identify patients likely to respond to ICI alone or in combination with chemotherapy. Methods. From the de-identified GuardantINFORM database we identified 695 advanced NSCLC (aNSCLC) patients treated with first- or second-line ICI monotherapy or ICI+chemotherapy and randomly split them into training (n=483) and test (n=212) sets. A literature-curated, data-driven epigenomic signature associated with real-world time to treatment discontinuation (rwTTD) was combined with MSI and TMB to train the multimodal model. Patients ≥80th MIRS percentile were labeled MIRS-High. Cox proportional hazards models adjusted for sex, age, therapy type (mono vs combo), line of therapy, and baseline tumor fraction provided adjusted hazard ratios (aHR); median rwTTD was estimated by Kaplan-Meier. Results. In the independent test set (n=212), MIRS-High patients had longer rwTTD (median 8.7 vs 5.1 months; aHR 0.61, 95% CI 0.41-0.93, p=0.02) and improved overall survival (OS) (aHR 0.33, 95% CI 0.16-0.68, p<0.005). In the ICI monotherapy subgroup (n=69), MIRS-High showed median rwTTD 11.0 vs 4.9 months (aHR 0.31, 95% CI 0.13-0.75, p=0.01) and longer OS (aHR 0.18, 95% CI 0.04-0.79, p=0.023). MIRS-High was not associated with rwTTD in patients treated with chemotherapy alone (aHR 1.16, 95% CI 0.94-1.44, p=0.18). Conclusions. A pretreatment plasma-based score combining MSI, TMB, and epigenomic signatures identifies aNSCLC patients with superior outcomes on ICI and outperforms MSI or TMB alone. The ICI treatment-specific stratification and strong association with both rwTTD and OS support the clinical utility of this multimodal approach, warranting further evaluation to assess the potential for guiding ICI treatment decisions. Citation Format: Sean Gordon, Jing Wang, Shile Zhang, Marisa Juntilla, Tingting Jiang, Matthew Ellis, Vishnu Ramani, Reagan Barnett, Bernard Herrman, Justin Odegaard, Darya Chudova. Blood-based integration of epigenomic profiles, TMB, and MSI to predict immune checkpoint inhibitor response in advanced non-small cell lung cancer (aNSCLC) [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 100.

98 Background: TMB is a biomarker for immune checkpoint inhibitors (ICIs). Analysis of circulating tumor DNA (ctDNA) has the potential to non-invasively identify pts likely to benefit from ICIs by assessing pTMB. We conducted a phase II basket trial as part of the SCRUM-Japan GOZILA umbrella/basket trial to evaluate the efficacy and safety of nivolumab monotherapy in pts with advanced GI cancers with high pTMB. Methods: Eligibility criteria included histologically confirmed unresectable or recurrent GI cancers; ECOG PS of 0 or 1; refractory or intolerant to the standard therapies; and high pTMB identified by Guardant360 CDx, a 74-gene ctDNA assay. Pts received intravenous nivolumab monotherapy of 360 mg every 3 weeks until progressive disease. The primary endpoint was objective response rate (ORR) per investigator assessment. The pTMB score was calculated by adjustment of mutation count by tumor fraction, and initial pTMB thresholds were determined based on previously reported ORR for ICI treatment for each GI cancer in the cohort. The cohort 2 proceeded only for esophageal cancer, and an ultra-high TMB cohort was created to include pts in the top 2% for pTMB scores. Results: Fifty-one pts with high pTMB in the cohort 1/2 (n=32) and the ultra-high TMB cohort (n=19) were enrolled. The median duration of follow-up was 7.5 months (range, 0.9–51.3). Deficient mismatch repair (dMMR) or high microsatellite instability (MSI-H) was observed in 1 pt in cohort 1/2 and 2 in ultra-high TMB cohort. The ORR was 12.5% and 21.1%, with median duration of response of 14.0 months and not reached, respectively. In an exploratory analysis of the overall population, pts with esophageal cancer had the highest ORR of 33.3% (4/12), followed by pancreatic (1/6, 16.7%) and colorectal cancer (3/27, 11.1%). Of 2 pts with proficient MMR/non-MSI-H colorectal cancer who had a response, 1 had a POLE mutation. Pts with response tended to have higher pTMB score (median, 26 vs. 8). dMMR/MSI-H (HR=0.16; 95% CI, 0.02–1.18), PD-L1 CPS ≥1 (HR=0.55; 95% CI, 0.29–1.01), and absence of liver metastasis (HR=0.36; 95% CI, 1.48–5.14) were significantly associated with longer progression-free survival. Treatment-related grade 3–4 adverse events occurred in 3 pts (9%) in the cohort 1/2 and 3 (16%) in the ultra-high TMB cohort, with no treatment death in either cohort. Conclusions: Nivolumab demonstrated antitumor activity in pretreated pts with advanced GI cancer and high pTMB. These results suggest that the use of pTMB potentially identify patients who may benefit from ICI treatment. Clinical trial information: UMIN000033182 .

Liquid biopsy (LB) provides valuable molecular insights in metastatic breast cancer (MBC). This study examines the relationship between the highest VAF (hVAF) in ctDNA and patient survival outcomes. In this retrospective analysis, the Guardant360 genomic assay, encompassing 74 genes and microsatellite instability status, was used on plasma samples prospectively collected at tumor progression from patients with MBC treated at Hospital Clinic of Barcelona. The hVAF for each patient was identified for analysis. Cox proportional hazards models assessed hVAF both as a continuous variable and in tertiles, focusing on its association with overall survival (OS). Multivariable analyses adjusted for menopausal status, performance status, bone-only disease, visceral disease, treatment line and cancer subtype was performed. For hormone receptor-positive(HR+)/HER2-negative(-) MBC treated after LB with endocrine therapy+CDK4/6 inhibitors (ET+CDK4/6i), hVAF's association with progression-free survival (PFS) was also explored. Between February 2021 and April 2022, 99 plasma samples from 99 patients were analyzed (median age: 59.7; 75.8% HR+/HER2-, 16.2% triple negative, 8.1% HER2+ MBC). Overall, ctDNA was detected in 98.0% of cases, with 94.0% showing pathogenic variants (PVs). The median number of PVs was 3 per patient (range: 1-16) and the most frequently mutated genes were TP53 (50;50.5%), ESR1 (38;38.4%) and PIK3CA (28;28.3%). Median follow-up was 27.2 months (0.56-37.45). OS was significantly different across hVAF tertiles (T3, T2 and T1 median OS [mOS] in months 13.1, 25.1 and NR; p<0.001). Higher hVAF significantly correlated with poorer OS both as continuous (hazard ratio [HR] 1.02, p<0.001) and categoric (HR T1 vs. T3: 3.76, p<0.001) variable. Multivariable analysis confirmed hVAF as an adverse prognostic factor (HR: 1.02, p=0.002). Continuous hVAF was negatively associated with PFS (HR: 1.05, p=0.007) in 24 patients under ET+CDK4/6i. This study reveals hVAF significantly impacts OS in MBC patients, suggesting its potential for refining prognostic stratification in clinical trials and guiding the selection of which patients might benefit from more aggressive therapies in subsequent lines.

3014 Background: HER2 ( ERBB2) gene amplification (amp) is a potential therapeutic target beyond breast and gastric cancers. HER2 amp can be detected in plasma cfDNA which may be an alternative to tissue biopsy. HERALD/EPOC1806 was a multicenter, investigator-initiated phase 2 trial of T-DXd for patients with HER2-amplified advanced solid tumors identified by cfDNA testing. Methods: We enrolled adults with advanced solid tumors harboring HER2 amp detected by next-generation sequencing of cfDNA (Guardant360) in the GOZILA study (UMIN000029315). We excluded breast or gastric cancer patients with HER2 overexpression (IHC 3+ or IHC 2+/ISH+). Patients received T-DXd 5.4 mg/kg every 3 weeks until disease progression or unacceptable toxicity. The primary endpoint was investigator-assessed objective response rate (ORR). Results: From December 2019 to January 2022, 4,734 patients were screened by cfDNA in the GOZILA study. Among 252 with HER2 amp, 62 with 16 cancer types were enrolled in HERALD. Median baseline plasma HER2 copy number (CN) was 8.55 (range, 2.4–73.9). All patients but 3 (all with salivary gland cancer) received prior anti-cancer therapy (median, 3 lines; range, 0–8). At a median follow-up of 8.9 months (data cut-off: July 17, 2022), the confirmed ORR was 56.5% (95% CI, 43.3–69.0%), statistically higher than the threshold value of 5%. Responses were observed for 13 cancer types, including KRAS-mutant colorectal (1/3), PIK3CA-mutant endometrial (5/6), and tissue HER2-negative gastric cancers (1/2). Plasma HER2 CN above vs. below the baseline median value did not impact response (ORR: 58.1% vs. 54.8%); however, the clearance vs. persistence of HER2 amp in cfDNA on Cycle 2 Day 1 corresponded to higher response (ORR: 88.0% vs. 22.7%). ORR by independent review was 58.1% (95% CI, 44.8–70.5%), and the disease control rate was 90.3% (95% CI, 80.1–96.4%). Median progression-free survival was 7.0 (95% CI, 4.9–9.7) months, and the median duration of response was 8.8 months (95% CI, 5.8–11.2). Most adverse events were mild to moderate. Interstitial lung diseases occurred in 16 patients (25.8%, G1/G2/G3; 14/1/1). Conclusions: T-DXd achieved a high ORR, durable response with a manageable safety profile in patients with advanced solid tumors and HER2 amp detected in cfDNA. Clinical trial information: JapicCTI-194707 .[Table: see text]

3102 Background: FGFR alterations are observed in approximately 7% of advanced solid malignancies and are associated with poor prognosis and resistance to traditional anti-cancer therapy. Despite this, optimal therapeutic strategies with FGFR inhibitors for most of FGFR-altered solid malignancies are yet to be defined. Circulating tumor DNA (ctDNA) analysis has the potential to accurately detect FGFR alterations by assessing spatial and temporal intratumoral heterogeneity. Methods: We conducted a multicenter, investigator-initiated, phase II basket-type trial, TiFFANY, to evaluate the efficacy and safety of futibatinib, a highly selective covalent pan-FGFR inhibitor, in patients (pts) with advanced solid malignancies with FGFR alterations identified by ctDNA analysis who were refractory or intolerant to standard-of-care treatment. ctDNA analysis was performed in the GOZILA study. Enrolled pts received futibatinib at a dose of 20 mg once daily in a 21 day-cycle. The primary endpoint was investigator-assessed objective response rate (ORR). We set the threshold ORR of 5% and expected one of 25%. Planned sample size was 26 with one-sided alpha of 2.5% and power of 80%. Blood and tissue samples collected before treatment (baseline), at week 3, and after disease progression were analyzed for biomarkers and resistance mechanisms. Results: Twenty-six pts with FGFR alterations (mutation, 9; amplification, 13; fusion, 4) in ctDNA were enrolled between August 2019 and March 2021. The primary endpoint was met with five (19.2%; 95% CI, 6.6–39.4%) achieving a confirmed partial response (PR) in various cancer types (biliary tract, gastric, urothelial, and urachal cancer). Median progression-free survival and overall survival were 2.6 and 8.9 months. The most common treatment-associated adverse events were hyperphosphatemia (100%), eye toxicity (46.2%) and diarrhea (42.3%), all of which were manageable. The median proportional change in ctDNA fraction from baseline to 3 weeks after treatment initiation in pts who achieved PR was significantly less than in pts with stable or progressive disease (0.11 vs. 1.0 vs. 1.6, respectively). Pts with no concurrent RTK/RAS/PI3K and cell cycle alterations in ctDNA were significantly more likely to respond to futibatinib than those with at least one of these alterations (ORR 50% vs. 0%; P=0.0038). Acquired gene alterations in ctDNA after progression tended to be more common in pts with FGFR amplification (83.3%) than in those with FGFR mutation or fusion (60% or 66.7%). Conclusions: Futibatinib demonstrated promising efficacy in refractory advanced solid malignancies with FGFR alterations in ctDNA with an acceptable toxicity profile. Oncogenic co-alterations detected by ctDNA genotyping may predict primary resistance. Clinical trial information: JapicCTI-194624 .

BackgroundComprehensive biomarker testing is essential in selecting optimal treatment for patients with metastatic colorectal cancer (mCRC); however, incomplete genotyping is widespread, with most patients not receiving testing for all guideline-recommended biomarkers, in part due to reliance on burdensome sequential tissue-based single-biomarker tests with long waiting times or availability of only archival tissue samples. We aimed to demonstrate that liquid biopsy, associated with rapid turnaround time (TAT) and lower patient burden, effectively identifies guideline-recommended biomarkers in mCRC relative to standard of care (SOC) tissue testing.Patients and methodsProspectively enrolled patients with previously untreated mCRC undergoing physician discretion SOC tissue genotyping submitted pretreatment blood samples for comprehensive circulating tumor DNA (ctDNA) analysis with Guardant360 and targeted RAS and BRAF analysis with OncoBEAM.ResultsAmong 155 patients, physician discretion SOC tissue genotyping identified a guideline-recommended biomarker in 82 patients, versus 88 identified with comprehensive ctDNA (52.9% versus 56.8%, noninferiority demonstrated down to α = 0.005) and 69 identified with targeted PCR ctDNA analysis (52.9% versus 44.5%, noninferiority rejected at α = 0.05). Utilizing ctDNA in addition to tissue increased patient identification for a guideline-recommended biomarker by 19.5% by rescuing those without tissue results either due to tissue insufficiency, test failure, or false negatives. ctDNA median TAT was significantly faster than tissue testing when the complete process from sample acquisition to results was considered (median 10 versus 27 days, P < 0.0001), resulting in accelerated biomarker discovery, with 52.0% biomarker-positive patients identified by ctDNA versus 10.2% by SOC tissue 10 days after sample collection (P < 0.0001).ConclusionsComprehensive ctDNA genotyping accurately identifies guideline-recommended biomarkers in patients with mCRC at a rate at least as high as SOC tissue genotyping, in a much shorter time. Based on these findings, the addition of ctDNA genotyping to clinical practice has significant potential to improve the care of patients with mCRC.

A possible association between changes in tumor variant allele fraction (VAF) from baseline to cycle 2/ day 1 (C2D1) using the 74-gene Guardant360 assay and progression free survival (PFS) was observed in 45 patients (pts) with ABC treated with a CDK4/6i + ET. Here, we tested the same methodology in an independent dataset and explored the association between overall survival (OS), intrinsic subtype (IS) and ctDNA dynamics. Baseline and C2D1 plasma samples were obtained from 113 pts with HR+/HER2-negative ABC treated with CDK4/6i + ET in an international, multicentric observational study. The first 45 pts were used to derive the mean VAF ratio (mVAFR: mean of the log VAFR for all mutations) methodology associated with PFS (NPJ Breast Cancer 2021). mVAFR was calculated for each alteration with a VAF ≥0.4% at any time point. Pts with VAFs <0.4% at all timepoints were considered to have ctDNA-low disease. Molecular response was defined by the mVAFR. The main objective was to assess the association of mVAFR with PFS in the 68-pt independent validation cohort. We also explored the association of mVAFR with OS (n=113) and with IS (n=54). PFS and OS uni- and multivariable cox models were performed adjusting for clinical variables. With a median follow up (mFUP) of 16.6 months (m), pts with medium or high mVAFR had shorter median PFS than pts with low mVAFR and ctDNA-low disease (11.2 vs 25.0m; adjusted hazard ratio [aHR]=2.85, 95% confidence interval [CI] 1.2-6.7, p=0.017). mVAFR as a continuous variable was also found associated with PFS (HR=2.0, 95% CI 1.0–4.1, p=0.047). In the 113-pt combined cohort, with a mFUP of 19.4m, pts with medium or high mVAFR had shorter OS than pts with either low mVAFR or ctDNA-low disease (31.3m vs not reached; aHR=10.9, 95% CI 3.7-31.6, p<0.001). mVAFR as a continuous variable was also found associated with OS (aHR=4.1, 95% CI 1.7-9.5, p<0.001). Finally, mVAFR was associated with IS (p=0.022). Early ctDNA dynamics after CDK4/6i + ET can identify different pt populations. The absence of molecular response identifies pts with poor outcome, and future studies should focus on improving treatment options for this population.

3555 Background: HER2 amp occurs in 1-4% of mCRC pts. Two single arm phase 2 studies, HERACLES and MyPathway, showed efficacy for dual HER2-targeted therapy in pts with RAS wild type ( RAS wt) mCRC with HER2 amp detected in tumor tissue; however, efficacy for pts prospectively enrolled with HER2 amp identified in ctDNA is unknown. Furthermore, the efficacy of real-world non-HER2-targeted SOC for HER2 amplified RASwt mCRC pts is not clear. Methods: We conducted a phase 2 trial to evaluate the efficacy of pertuzumab (P) plus trastuzumab (T) in RASwt mCRC pts with HER2 amp centrally confirmed by tissue (IHC and/or FISH) and/or ctDNA (Guardant360) who had progressed on SOC including EGFR blockade. Pts received intravenous P (840 mg loading dose followed by 420 mg) and T (8 mg/kg loading dose followed by 6 mg/kg) every 3 weeks. The primary endpoint was confirmed objective response rate (ORR) by investigator assessment, analyzed for two primary populations: pts with HER2 amp in tissue (tissue+) or in ctDNA (ctDNA+). Efficacy of real-world non-HER2-targeted SOC for HER2 amplified RASwt mCRC pts was prospectively assessed in a concurrent registry: the SCRUM-Japan registry. Results: Among 75 pts screened, concordance of HER2 amp between tissue and ctDNA was 83%. The primary endpoint was met in each cohort of TRIUMPH, with confirmed ORR of 30% (95% CI 14-50%) in 27 tissue+ pts and 28% (12-49%) in 25 ctDNA+ pts. In contrast, ORR in first salvage SOC after EGFR blockade was 0% (0.0-24.7%) in the real-world cohort. Median progression free and overall survival were 4.0 months (1.4-5.6) and 10.1 months (4.5-16.5) in the tissue+ pts and 3.1 months (1.4-5.6) and 8.8 months (4.3-12.9) in the ctDNA+ pts. One pt withdrew due to an adverse event (grade 3 decreased ejection fraction), but no treatment related deaths occurred. In exploratory analyses, pts without ctDNA mutations of RAS/ BRAFV600/ PIK3CA/ HER2 were more likely to respond to P+T than those with a ctDNA mutation in at least one of these genes (ORR 44% vs. 0% in tissue+ and 37% vs. 0% in ctDNA+). Decreased ctDNA fraction and HER2 plasma copy number at 3 weeks after treatment initiation corresponded to P+T response. At least one actionable alteration emerged after progression in 16 (62%) of 26 pts with ctDNA results at both baseline and progression. Among 5 pts who achieved response and had ctDNA results at both time points, 4 pts acquired actionable alteration at progression. Conclusions: We demonstrate promising efficacy and safety of P+T for RASwt mCRC pts with HER2 amp in either tumor tissue or ctDNA. Our results show that complete ctDNA genotyping identifies pts most likely to benefit from dual HER2 blockade and can be used to monitor response and detect actionable resistance biomarkers. Clinical trial information: UMIN000027887 and UMIN000028058.

9027 Background: Somatic genomic testing is recommended by numerous expert guidelines to inform targeted therapy treatment for patients with advanced lung adenocarcinoma (aLUAD). The NILE study was a prospective observational study that demonstrated non-inferiority of cell-free circulating tumor DNA (cfDNA)-based tumor genotyping compared to tissue-based genotyping to find targetable genomic alterations in patients with newly diagnosed aLUAD. As the cohort has matured, clinical outcomes data can now be reported. Methods: This prospective, multicenter North American study (NCT03615443) enrolled patients with previously untreated aLUAD who had standard of care (SOC) tissue genotyping performed and concurrent comprehensive cfDNA analysis using the commercially available Guardant360 assay (Guardant Health, Redwood City, CA). After 12 months of study enrollment, objective response rates, disease control rate, and time to treatment data were collected for patients with targetable genomic alterations, as defined by NCCN guidelines, who were treated with physician’s choice of therapy. Results: Among 282 patients on the study, 89 (31.6%) had an actionable biomarker detected by tissue (21.3%) and/or cfDNA (27.3%) analysis. Sixty-one (68.5%) of these patients were treated with an FDA-approved targeted therapy guided by somatic genotyping results ( EGFR, ALK, ROS1). Thirty-three patients were eligible for clinical response evaluation and demonstrated an objective response rate of 58% and disease control rate of 94%. Twenty-five (76%) achieved a durable response > 6 months; 17 (52%) achieved a durable response > 12 months. Patients responded to targeted therapy regardless of the variant allele frequency of the target alteration. The time to treatment (TtT) was significantly faster for cfDNA-informed biomarker detection as compared to tissue genotyping (median 18 vs 31 days, respectively; p = 0.0008). Conclusions: This is the first prospective community-based study to find that cfDNA detects guideline-recommended biomarkers at a rate similar to tissue genotyping, and therapeutic outcomes based on plasma-based comprehensive genomic profiling are comparable to published tissue-based targeted therapy clinical outcomes. The NILE study complements and confirms findings in the prospective FLAURA and SLLIP studies, which exclusively enrolled at academic sites. Clinical trial information: NCT03615443.

Amivantamab, an epidermal growth factor receptor (EGFR)-MET bispecific antibody with immune cell-directing activity, targets activating/resistance EGFR mutations and MET mutations/amplifications. In the ongoing CHRYSALIS study (NCT02609776), amivantamab demonstrated antitumor activity in patients with EGFR exon 20 insertion (Exon20ins) disease. To identify patients likely to benefit from amivantamab therapy, we clinically validated 2 novel candidate companion diagnostics (CDx) for detecting Exon20ins variants in tumor tissue and plasma, with combined coverage of >100 variants.

10050 Background: ctDNA detected by ddPCR predicts ICI response in MM, although its utility is limited to pts with known recurring mutations eg. BRAF, NRAS, KIT. We sought to overcome this limitation by using a next generation sequencing approach in BRAF V600 wild type (WT) MM pts. Methods: Plasma was collected at baseline and Week (wk) 6 in 35 BRAF V600 WT MM pts treated with ICI. Cell free (cf)DNA was analyzed using Guardant360 and only somatic non-synonymous and promoter variants were considered. Pts who failed cfDNA extraction at baseline were excluded (n = 3). Favorable ctDNA was defined as undetectable ctDNA at wk 6 and unfavorable ctDNA defined as detectable ctDNA at wk6. Response was according to RECIST at first restaging. Results: Of the evaluable 32 pts (64 plasma samples), median baseline cfDNA quantity was 33ng (range 4-657ng) and ctDNA was detected in 29/32 pts (91%). All 3 pts with undetectable baseline ctDNA had less than 10ng cfDNA compared to only 1/29 pts with detectable baseline ctDNA. Number of mutations identified in the 29 ctDNA-positive pts was 4 per pt (range 1-22). Response assessment was performed on 30 evaluable pts. Candidate driver mutation(s) in BRAF, NF1, or N/K/HRAS were identified in 26/30 pts. These mutations were often detected with other established mutations involved in tumorigenesis (eg. TERT promoter), or passenger mutations (eg. clonal hematopoiesis). Analysis of driver mutations revealed a sensitivity and specificity in predicting treatment failure of 92% and 93%, respectively (table). When all mutations identified were evaluated for treatment response, 9/18 responding pts retained some ctDNA at wk 6, although this never included TERT variants. The resulting sensitivity and specificity in predicting treatment failure changed to 100% and 50%, respectively, when all cfDNA variants were included. Conclusions: The extensive coverage of Guardant360 improves ctDNA detection in BRAF V600 WT MM pts, allowing non-invasive, rapid, and longitudinal assessment of response in a broader population. The expanded coverage also identifies passenger variants of potential non-MM origin, eg. clonal hematopoiesis, and with significant overlap with ctDNA, it is not possible to distinguish between the two in the circulation. We therefore recommend identification and monitoring of known cancer driver mutations only. [Table: see text]

Changes in ctDNA levels may predict response to a variety of drugs, including CDK4/6i; however, the best assay and method are still to be defined. This is a prospective single-center study in hormone receptor-positive/HER2-negative advanced breast cancer pts treated with CDK4/6i and endocrine therapy (ET). Paired plasma samples were collected at cycle 1 day 1 (C1) and cycle 2 day 1 (C2). Somatic alterations and variant allele fraction (VAF) were assessed using the 74-gene Guardant360 assay (Guardant Health). A VAF ratio (VAFR) was calculated for each alteration with a VAF of ≥ 0.4% at C1 or C2. Molecular response was defined for each patient as the mean of all VAFRs (mVAFR). Pts with VAFs < 0.4% at C1 and C2 were considered to have low-shedding tumors. Progression free survival (PFS) hazard ratios (HR) were calculated using a univariate Cox model. PAM50 subtypes and tumor infiltrating lymphocytes (TILs) were determined in a subset. 48 pts treated with ET and palbociclib (89%) or ribociclib (11%) were analyzed with a median follow-up of 12.0 months (IQR 6.7-14.6). Clinical characteristics: 65% had visceral disease, 48% were treated as 1st-line, 35% as 2nd-line, 57% used fulvestrant and 33% an aromatase inhibitor. PAM50 subtype distribution (n=27): Luminal A (n=9), Luminal B (n=10), HER2-enriched (n=4), Normal (n=3) and Basal-like (n=1). ctDNA was detected in 96% of pts. mVAFR < 0.3 (high-ctDNA responders) (n=12) and low-shedding tumors (n=13) correlated with significantly improved PFS (HR=0.39, p=0.025), especially when compared to pts with ctDNA mVAFR > 1 (HR=0.27, p=0.010, n=12). Within PAM50 tested tumors, non-Luminal (n=5) were low-ctDNA responders (mVAFR > 0.3) (n=3) or low-shedding (n=2); Luminal A or B were high-ctDNA responders (n=8), low-ctDNA responders (n=7) and low-shedding (n=4). TILs were increased in low relative to high-ctDNA responders (mean 3.3% vs 1.8%). ctDNA dynamics are an early surrogate of CDK4/6i + ET efficacy. The clinical utility of this biomarker should be tested in prospective clinical trials in which pts with unfavorable ctDNA responses are randomized to alternative treatment strategies.

9553 Background: Genotyping is required to identify cancer patients (pts) eligible for targeted therapy; however, many do not receive biomarker testing, in part due to limitations associated with tissue-only genotyping practices and the growing list of biomarkers recommended to be tested. Liquid biopsy overcomes many of these limitations but is not yet fully adopted. We report here the clinical performance of a comprehensive liquid biopsy test based on next generation sequencing (NGS) of circulating tumor DNA (ctDNA) for the identification of NSCLC patients with EGFR exon 19 deletions (ex19del) or L858R mutations ( EGFRm) or EGFR T790M, eligible for treatment with osimertinib. Methods: 441 (79%) of 556 pts randomized in FLAURA (NCT02296125; first-line osimertinib vs comparator EGFR TKI in EGFRm NSCLC) and 300 (72%) of 419 pts from AURA3 (NCT012151981; osimertinib vs chemotherapy in NSCLC pts with T790M at progression on EGFR TKI) were retrospectively tested with Guardant360 (G360), a 74-gene ctDNA NGS assay assessing single nucleotide variants, insertion-deletions, amplifications, and fusions in genes relevant to targeted therapy selection as well as microsatellite instability. Progression-free survival (PFS) of pts with EGFRm or T790M detected by G360 was compared to pts detected by the cobas EGFR Mutation Test (cobas) using tissue or plasma with an unadjusted cox model. Results: Treatment with osimertinib was associated with a significant PFS benefit relative to control therapy in NSCLC pts with EGFRm (FLAURA) and T790M (AURA3) detected using G360 (Table). Observed clinical benefit for pts with EGFRm or T790M detected by G360 was similar to that for pts with EGFRm or T790M identified by cobas using tissue or plasma specimens. Conclusions: This analysis demonstrates that G360 accurately identifies pts for osimertinib therapy while simultaneously providing comprehensive genotyping for other therapeutic molecular targets. The application of NGS liquid biopsy has the potential to increase rates of pts genotyped and access to precision medicine. [Table: see text]

ctDNA is a blood-based biomarker with promising potential in lung cancer for minimal residual disease (MRD) assessment and early detection of recurrence. However, data regarding feasibility are limited, especially for stage I-II disease. We performed longitudinal plasma ctDNA profiling of early-stage lung cancer patients (pts) that underwent resection at MD Anderson Cancer Center from Apr 2016 to Jan 2017. Plasma ctDNA was analyzed from pre-operative and multiple post-operative time points until disease recurrence. ctDNA profiling was performed using a 30kb Digital Sequencing panel (Guardant Health) covering SNVs in 21 genes and indels in 9 genes that are commonly present in lung cancer. ctDNA profiles from ∼30,000 lung cancer pts were used to train a classifier to exclude non-tumor related mutations. A total of 40 pts were included in this analysis, comprised of the first 17 pts with recurrence in the longitudinal study and 23 consecutive pts without recurrence. This cohort was primarily stage I and II (15 [38%], 16 [40%]). Histology included adenocarcinoma (29 [73%]), SCC (6 [15%]), and SCLC (2 [5%]). 58% had adjuvant therapy. Median follow-up was 17.7 (3.4 – 24.5) months and median time to recurrence was 7.1 (3.4 – 16.5) months in this selected cohort. At least one ctDNA alteration was detected in 55% (21/38) of pts with evaluable pre-op samples and in 22% (8/37) of pts at 4 weeks post-op. Presence of ctDNA at 4 weeks post-op heralded eventual recurrence with 43% sensitivity and 91% specificity (75% PPV, 73% NPV) and was significantly associated with worse recurrence free survival (p=0.022, HR 6.52; 95% CI 1.3 – 32.6), while also accounting for stage. In the absence of the variant classifier, an additional 7/37 pts had non-tumor alterations detected at 4 weeks post-op with a recurrence sensitivity and specificity of 57.1% and 69.6%. ctDNA was identified in 76% (13/17) of pts prior to or at the time of recurrence. The median interval between ctDNA detection and radiographic recurrence was 91 days. Detection of post-op ctDNA, as early as 4 weeks after resection of early-stage lung cancer, is associated with significantly increased risk of recurrence. Accurate detection of ctDNA in this MRD setting is enabled by a highly sensitive sequencing platform that incorporates a novel variant classifier to enhance clinical specificity.