Circulating Tumor DNA Dynamics Reveal KRAS G12C Mutation Heterogeneity and Response to Treatment with the KRAS G12C Inhibitor Divarasib in Solid Tumors

Circulating tumor DNA (ctDNA) is an emerging tool to understand the mutational landscape of cancer with minimally invasive liquid biopsies (1) and has the potential to better capture tumor genomic heterogeneity that exists across all tumor lesions (primary and metastatic sites) in the body (2, 3). ctDNA can be utilized clinically in biomarker selection and risk stratification and can potentially serve as an early surrogate of tumor response and long-term outcomes (4–7). Although tissue-based molecular profiling remains the gold standard, liquid-based assays have become an alternative approach for tumor mutation detection and companion diagnostic purposes with multiple tests approved by the U.S. Food and Drug Administration (FDA) and/or European Medicines Agency (EMA; refs. 8–11). Most panel-based ctDNA assays have limit of detection (LOD) near 0.5% or 0.1%, while tumor-informed approaches can have sensitivity at 0.01% or below. Despite this, false negatives are still possible depending on the patient's disease, for example, if they have early stage disease or a non-shedding tumor. Therefore, current recommendations state that patients who test negative for select biomarkers with liquid biopsies should be reflexed to routine biopsy and testing with biopsy specimens if feasible (8–11). In addition to the clinical utility of companion diagnostics, there are collaborative initiatives to build an evidentiary roadmap to using ctDNA as an early endpoint for regulatory decision-making (12). However, key knowledge gaps remain, such as establishing the baseline ctDNA tumor fraction across tumor types, as well as the relationship between on-treatment ctDNA dynamics and patient outcomes with anticancer treatment.

Mutations in the Kirsten rat sarcoma viral oncogene homologue (KRAS) gene are among the most prevalent oncogenic driver mutations in human cancers (13–15). The KRAS G12C mutation occurs in approximately 12% to 14% of patients with non–small cell lung cancer (NSCLC), 4% of patients with colorectal cancer, and up to 4% of patients with other solid tumors (16). Divarasib (GDC-6036) is an oral, covalent KRAS G12C inhibitor that binds to the mutant cysteine residue and irreversibly locks the KRAS G12C protein in the inactive state, turning off its oncogenic signaling (17). Encouraging antitumor activity and durable responses with single-agent divarasib were shown in the phase 1 GO42144 study, with a confirmed response rate of 56% for NSCLC, 36% for colorectal cancer, as well as responses for other solid tumors at a dose of 400 mg once daily (18).

To evaluate the relationship between baseline ctDNA tumor fractions across tumor types and treatment response, we profiled more than 130 patients with solid tumors harboring a KRAS G12C mutation treated with single-agent divarasib in this phase 1 study. In addition, we performed serial profiling with a selected panel of personalized, tumor-informed genomic alterations to track the overall ctDNA tumor fractions and KRAS G12C variant allele frequency (VAF) simultaneously with LOD of 0.05%. These analyses provide insights into the association between ctDNA dynamics and response to treatment with divarasib, and reveal the extent of tumor heterogeneity of the KRAS G12C mutation.

The data we report here are from a previously described (18) ongoing phase I, open-label, multicenter, dose-escalation, and dose-expansion study of divarasib as a single agent in patients with advanced or metastatic solid tumors that harbor a KRAS G12C mutation. Adult patients with advanced or metastatic KRAS G12C-positive solid tumors were enrolled and received divarasib 50 to 400 mg orally once daily (QD) in 21-day cycles. Key inclusion criteria included the submission of a freshly collected pretreatment blood sample and documentation of KRAS G12C mutation by either local testing of tumor tissue or blood samples [using a validated molecular testing method performed at a Clinical Laboratory Improvement Amendments (CLIA) or equivalently certified laboratory], or central testing of blood samples (with the use of a clinical trial assay based on FoundationOne Liquid CDx). Key exclusion criteria were previous treatment with a KRAS G12C inhibitor and treatment with chemotherapy, immunotherapy, or biologic therapy within 3 weeks or five half-lives prior to initiation of study treatment, whichever is shorter. Overall, the age, sex, and Eastern Cooperative Oncology Group Performance Status score distribution in our study was similar to the overall distribution in the literature, though Black patients were underrepresented (Supplementary Table S1). The study protocol was approved by the Institutional Review Board (IRB) at each participating site prior to any patient receiving the study drug.

Preliminary antitumor activity was determined by the investigator according to RECIST, version 1.1, and included objective response rate (ORR) and progression-free survival (PFS). Confirmed objective response in patients with measurable disease was defined as complete response (CR) or partial response (PR) on two consecutive tumor assessments at least 4 weeks apart, whereas best response did not require a confirmatory assessment. PFS was defined as the time from first treatment to the first occurrence of disease progression or death from any cause during the study (whichever occurred first).

This study was conducted in full conformance with the International Council for Harmonisation E6 guidelines for Good Clinical Practice and the principles of the Declaration of Helsinki or the applicable laws and regulations of the country in which the research was conducted, whichever afforded greater protection of the patient. All patients provided written informed consent before enrollment.

Plasma for ctDNA analysis was collected at baseline [cycle 1 day 1 (C1D1), prior to divarasib treatment administration] and early on-treatment [cycle 1 day 15 (C1D15) and cycle 3 day 1 (C3D1)] time points (Fig. 1A). The C3D1 plasma collection time point also approximately matches the first RECIST response assessment. Blood (10 mL) was collected in a K2 EDTA blood collection tube, centrifuged at 2,000 × g for 15 minutes, followed by a second centrifugation of the plasma supernatant at 3,000 × g for 15 minutes. Cleared plasma was stored at −80°C. Circulating cell-free DNA (cfDNA) was extracted from plasma samples using the QIAamp circulating nucleic acid kit. Quantity and quality of the purified cfDNA were checked using a Qubit fluorimeter and Bioanalyzer 2100. Genomic DNA was extracted from matched peripheral blood mononuclear cells to control for clonal hematopoiesis of intermediate potential and germline mutations.

PredicineWES+ assay is a boosted whole exome sequencing (WES) assay with 20,000× sequencing depth in 600 cancer-related genes (LOD: 0.25%) and 2,500× depth at the rest of exome regions (LOD: 1%). PredicineBEACON is a personalized tumor-agnostic minimal residual disease (MRD) assay. The personalized MRD panel was designed based on baseline (C1D1) plasma samples using the PredicineWES+ assay, and up to 50 somatic mutations were then selected for personalized panel design. The personalized MRD panel plus a fixed panel covering 500 hotspots and actionable mutations (including the KRAS G12C mutation) were used for ctDNA MRD detection with 100,000× sequencing depth. Analytical validation results showed 0.05% LOD for individual mutations and 0.005% sample-level LOD with the PredicineBEACON MRD assay. On-treatment (C1D15 and C3D1) plasma samples were evaluated by the PredicineBEACON MRD assay (Fig. 1A). In addition, the PredicineSCORE copy number burden (CNB) assay with a ∼3× low-pass whole genome sequencing (LP-WGS) was performed for both baseline and on-treatment plasma samples. The concordance between PredicineWES+ and PredicineBEACON MRD assays in the assessment of overall ctDNA tumor fraction and KRAS G12C VAF was validated (Supplementary Fig. S1). The Predicine assays were performed in a CLIA-certified and the College of American Pathologists–accredited laboratory (Predicine Inc., Hayward, CA).

For the estimation of ctDNA tumor fraction at baseline, the mutation-derived ctDNA tumor fraction was first estimated from the PredicineWES+ results based on the allele frequencies of autosomal somatic mutations as described previously (19). Briefly, the mutant allele fraction (MAF) and ctDNA tumor fractions were related as MAF = (ctDNA × 1)/[(1 − ctDNA) × 2 + ctDNA ×1], and so ctDNA = 2/[(1/MAF) + 1]. Somatic mutations in genes with a detectable copy number change were omitted from ctDNA tumor fraction estimation. Then, using PredicineSCORE LP-WGS data, the ichorCNA algorithm (20) was applied to guanine-cytosine content and mappability-normalized reads to estimate plasma copy number variations and tumor fractions using the hidden Markov model. Using LP-WGS to derive ctDNA tumor fractions is more robust when ctDNA tumor fraction is high, while mutation-derived ctDNA tumor fraction is more sensitive when ctDNA tumor fraction is low. Therefore, a composite ctDNA tumor fraction was calculated which equaled LP-WGS-derived ctDNA tumor fraction when LP-WGS-derived ctDNA tumor fraction was higher than 10%; otherwise, it equaled the mutation-derived ctDNA tumor fraction.

To interrogate patterns of ctDNA shedding in different tumor types exhibiting the KRAS G12C mutation, we leveraged an independent real-world pan-cancer cohort comprising patients who received liquid biopsy-based comprehensive genomic profiling (CGP) using FoundationOne Liquid CDx, which targets 324 genes, as part of routine clinical care (21). For this cohort, approval, including a waiver of informed consent and a HIPAA waiver of authorization, was obtained from the WCG Institutional Review Board (IRB; Protocol No. 20152817). Specifically, samples from patients with physician-reported stage IV NSCLC, colorectal cancer, and other solid tumors were examined. The ctDNA tumor fraction was estimated using two approaches: an aneuploidy-based measure when feasible or the maximum somatic allele frequency calculated from the allele fraction of somatic coding alterations, as described previously (22).

The Flatiron Health–Foundation Medicine NSCLC clinicogenomic database (FH-FMI CGDB) was used to assess the prognostic impact of ctDNA levels in patients with KRAS G12C-positive tumors (23). This nationwide (US-based) database contained data collected between January 2011 and September 2023 that originated from approximately 280 US cancer clinics (∼800 sites of care). Only patients who received a FoundationOne Liquid CDx test result within 60 days of beginning first-line advanced therapy that detected a KRAS G12C alteration were included.

ctDNA tumor fraction was estimated using multiple approaches. For samples with significant aneuploidy, the copy number modeling–based purity assessment was used to determine the ctDNA tumor fraction estimate (22). However, if significant aneuploidy was not detected, the VAFs of the detected somatic short variants and rearrangements in a sample were used to estimate ctDNA tumor fraction. For the latter approach, somatic status determination for short variants and rearrangements was made using the presence in a whitelist of specific variants and categories of variants that are highly biased toward being somatic (and not due to clonal hematopoiesis) or evidence of statistically significant differences in quantitative metrics from fragments harboring the mutated allele. While these approaches may not identify all true somatic variants found for a given sample, a substantial majority of samples have at least one or a few variants that can be used. The ctDNA tumor fraction estimate may then include the highest VAF for a short variant or rearrangement deemed to be somatic.

Patients were categorized as being ctDNA positive if found to have a ctDNA tumor fraction ≥1% and ctDNA negative if found to have a ctDNA tumor fraction <1%. All data in the FH-FMI CGDB were de-identified to protect patient confidentiality and clinical data were linked to genomic data by de-identified, deterministic matching (23). Retrospective longitudinal clinical data were derived from electronic health records and composed of both patient-level and unstructured data, while treatment information was abstracted from oncologist-defined, rule-based lines of therapy. IRB approval of the study protocol was obtained prior to study conduct and included a waiver of informed consent (WCG IRB, Protocol No. 1342690).

Real-world OS (rwOS) was assessed from start of first-line treatment until death. Patients without a date of death were censored at date of last clinic visit or structured activity. To account for immortal time, results were left-truncated and patients were treated as at risk of death only after the later of their first Foundation Medicine report date and their second visit in the Flatiron Health network, as both are requirements for inclusion in the database. Survival rates were identified by a Kaplan–Meier estimate and survival curves were compared using a log-rank test. Univariate analyses were performed using a Cox proportional hazard regression model to determine hazard ratios and 95% confidence intervals (CI).

Descriptive statistics were used to summarize baseline disease and clinical characteristics and ctDNA tumor fractions, including mean, median and range for continuous variables, and frequency and proportion for categorical variables. Associations of ctDNA tumor fractions and baseline disease and clinical characteristics were evaluated using Wilcoxon rank sum test. Correlation between ctDNA tumor fractions and continuous variables was described using Pearson’s correlation.

Associations of ctDNA tumor fractions with clinical outcomes, including confirmed ORR, best overall response, and PFS, were evaluated by tumor type. For subgroups defined by ctDNA tumor fractions, the ORR was reported for patients with measurable disease at baseline and summarized with 95% CI using the Clopper–Pearson method and compared using Fisher’s exact test. The PFS was compared between subgroups defined by ctDNA tumor fractions using a univariate Cox proportional hazards model to estimate hazard ratio (HR) and a log-rank test to report the P values. All P values were descriptive and not adjusted for multiple hypothesis testing. All statistical analyses were performed in R version 4.1.1.

Associated data will be made available to qualified researchers at the European Genome-Phenome Archive under accession number EGAS50000000315. To request access to such data, researchers can contact devsci-dac-d@gene.com. The data will be released to such requesters with necessary agreements to enforce terms such as security, patient privacy, and consent of specified data use, consistent with evolving, applicable data protection laws.

The FH-FMI CGDB data that support the findings of this study were originated by Flatiron Health, Inc., and Foundation Medicine, Inc. Requests for data sharing by license or by permission for the specific purpose of replicating results in this manuscript can be submitted to PublicationsDataaccess@flatiron.com and cgdb-fmi@flatiron.com.

Among a total of 137 patients (60 with NSCLC, 55 with colorectal cancer, and 22 with other solid tumors) who received at least one dose of single-agent divarasib in the phase 1 study, 106 (77.4%) patients were enrolled with local testing of tissue samples, 26 (19.0%) patients were enrolled with local or central testing of blood samples, and five (3.6%) patients were enrolled with a local test with unknown assay type. Pretreatment C1D1 plasma samples were evaluable in 134 patients by the PredicineWES+ ctDNA assay. At least one on-treatment plasma sample was evaluated in 95 patients; among these 95 patients, 74 patients had both C1D15 and C3D1 plasma samples, 16 patients had C1D15 only, and five patients had C3D1 plasma samples only that were evaluated by the PredicineBEACON MRD ctDNA assay (Fig. 1B and C).

At C1D1, the KRAS G12C mutation was detected in plasma samples in 72.9% (43/59) of patients with NSCLC, which is within the range of detectability of the KRAS G12C mutation in NSCLC with blood-based assays reported by others (24–26). In contrast, the KRAS G12C mutation was detected in plasma samples in 92.6% (50/54) of patients with colorectal cancer. The different KRAS G12C mutation detectability in plasma is likely due to different levels of ctDNA shedding, as the median ctDNA tumor fraction was 4.5% and 17.1% among patients with NSCLC and colorectal cancer, respectively (Fig. 2A). Consistently, median ctDNA tumor fractions tended to be higher in patients where the KRAS G12C mutation was detectable in plasma samples versus those not detectable in plasma samples (P = 0.08, 0.08, and 0.05 for NSCLC, colorectal cancer, and other solid tumors, respectively; Fig. 2B). In solid tumors other than NSCLC and colorectal cancer, the KRAS G12C mutation was detected in plasma samples in 81.0% (17/21) of patients and the median ctDNA tumor fraction was 3.7% (Fig. 2A). Due to the small number of patients with each tumor type in the other solid tumor cohort, we focused further analyses on the NSCLC and colorectal cancer cohorts.

To further validate the finding that different tumor types bearing the same KRAS G12C mutation may have distinct levels of ctDNA shedding, we utilized a pan-cancer real-world genomic database comprising patients who received liquid biopsy-based CGP at Foundation Medicine Inc., as part of routine clinical care. Estimated using the FoundationOne Liquid CDx assay, the median ctDNA tumor fraction was 3.5%, 13.2%, and 4.6% among patients with stage IV KRAS G12C-positive NSCLC (n = 507), colorectal cancer (n = 83), and other solid tumors (n = 271), respectively (Supplementary Fig. S2).

Baseline ctDNA tumor fraction was positively associated with tumor volume [sum of longest diameter (SLD)] with Pearson’s r = 0.412 and r = 0.420 in NSCLC and colorectal cancer, respectively (Fig. 2C), consistent with previous studies (27). There was also a trend of positive association between baseline ctDNA tumor fraction and the number of metastatic sites (Fig. 2D). Among patients with NSCLC, the presence of liver, bone, lymph node, or adrenal gland metastasis showed a mild trend of positive association with higher ctDNA tumor fractions (Fig. 2E). On the other hand, the presence of brain metastasis showed no association with higher ctDNA tumor fractions (Fig. 2E). Among patients with colorectal cancer, the presence of liver metastasis was associated with higher levels of ctDNA tumor fraction regardless of the presence or absence of lung metastasis (Fig. 2F and G). In contrast, no association was found between baseline ctDNA tumor fraction and the number of prior lines of systemic therapy (Fig. 2H).

Among patients with lower baseline ctDNA tumor fractions, a numerically higher confirmed ORR was observed compared to patients with higher baseline ctDNA tumor fraction (Fig. 3A). Furthermore, baseline ctDNA tumor fraction was associated with PFS in patients with colorectal cancer [ctDNA ≥ median (17.1%) versus ctDNA < median: HR, 3.75 (95% CI, 1.90–7.39); P < 0.001] (Fig. 3B), indicating that patients with lower baseline ctDNA tumor fraction had a longer PFS compared to patients with higher baseline ctDNA tumor fraction. Notably, baseline ctDNA tumor fraction remained as a significant prognostic factor in the multivariable analysis including key baseline demographic and clinical characteristics (Supplementary Fig. S3). In patients with NSCLC, an overall similar trend was observed [ctDNA ≥ median (4.5%) versus ctDNA < median: HR, 1.56 (95% CI, 0.71–3.46); P = 0.3] (Fig. 3B; Supplementary Fig. S3). This is consistent with the prognostic value of ctDNA tumor fraction across cancer types reported previously (28–31). To further validate the prognostic impact of ctDNA tumor fraction in patients with KRAS G12C-positive tumors, we surveyed the Flatiron Health–Foundation Medicine NSCLC clinic-genomic database. The rwOS was assessed from start of first-line treatment until death. Consistently, higher baseline ctDNA tumor fraction was associated with shorter median rwOS [ctDNA ≥ 1% (ctDNA positive: n = 80) vs. ctDNA <1% (ctDNA negative: n = 47): HR, 1.9 (95% CI, 1.1–3.2); P = 0.02; Supplementary Fig. S4].

Twenty-five patients with NSCLC and 37 patients with colorectal cancer had longitudinal ctDNA profiling at all three time points: C1D1, C1D15, and C3D1. In the majority of these patients, a reduction in ctDNA tumor fraction was observed as early as C1D15. At C3D1, all but two patients with a PR had ctDNA tumor fractions less than 3% (Fig. 4A and B). Overall, the patterns of ctDNA dynamics were consistent with the changes in the KRAS G12C VAF as reported in Sacher and colleagues (18).

Next, we assessed the association between changes in ctDNA tumor fraction and best overall response with divarasib. Comparing on-treatment ctDNA fraction to the baseline level (C1D15 vs. C1D1 or C3D1 vs. C1D1), we observed that patients achieving a best response of CR or PR had a deeper decline in ctDNA fraction than patients with stable disease (SD) or progressive disease (PD) as best response (Fig. 4C and D). A similar trend was observed for changes in the KRAS G12C VAF in association with best overall response (Supplementary Fig. S5).

Lastly, we explored the dynamics in ctDNA tumor fraction in association with PFS. Comparing ctDNA tumor fractions at C1D15 to C1D1, patients with NSCLC who achieved a deeper ctDNA reduction had a longer PFS than those with less ctDNA reduction [ctDNA reduction ≥ median (−97.8%) versus ctDNA reduction < median: HR, 0.47 (95% CI, 0.17–1.26), P = 0.1]. A similar observation was made comparing C3D1 to C1D1 [HR, 0.44 (95% CI, 0.14–1.36); P = 0.2, median ctDNA reduction: −98.9%] (Fig. 4E and F). Similarly, patients with colorectal cancer who had a deeper ctDNA reduction had a longer PFS [ctDNA reduction ≥ median vs. ctDNA reduction < median: C1D15 HR, 0.34 (95% CI, 0.16–0.69); P = 0.003; C3D1 HR, 0.31 (95% CI, 0.15–0.66); P = 0.002. Median ctDNA reduction for C1D15 and C3D1, respectively, are −79.8% and −90.5%; (Fig. 4E and G)]. In addition, we analyzed the dynamics in the KRAS G12C VAF in association with PFS. Overall, a similar trend as reported in Fig. 4 was observed (Supplementary Fig. S5).

Lastly, we explored the clonality of the KRAS G12C mutation by directly comparing the dynamics of KRAS G12C VAF versus dynamics of overall ctDNA tumor fraction. To ensure robustness of this analysis, we excluded cases that had low KRAS G12C VAF (<1%) and low ctDNA tumor fraction (<2%) at baseline. A strong correlation was observed between changes in the KRAS G12C VAF upon treatment and changes in the overall ctDNA tumor fraction, with the exception of a few outliers (Fig. 5A and B). Based on the standardized residuals of the linear regression between changes in KRAS G12C VAF versus changes in ctDNA tumor fraction, five outliers were identified in total for the comparison of C1D15 or C3D1 to C1D1. Out of the five outliers, RAS/RAF oncogenic driver mutations other than KRAS G12C were identified in three patients. In these three patients, a rapid increase in the VAF of non-KRAS G12C mutations was observed concurrently with a decrease in KRAS G12C VAF upon treatment with divarasib (Fig. 5C–E). This suggests that in these rare cases, KRAS G12C represented a subclonal mutation that was present only in a subset of tumor cells, and the presence of RAS/RAF oncogenic driver mutations other than KRAS G12C was associated with intrinsic resistance to divarasib treatment.

Oncogenic driver mutations in KRAS have been shown to be a frequent initiating event that usually occurs early during tumorigenesis (32, 33). However, subclonal KRAS driver mutations also exist and were reported to be more common in certain tumor types (e.g., colorectal cancer) than others (34). Here, we tracked KRAS G12C and overall ctDNA tumor fractions simultaneously at baseline and upon treatment with divarasib using serial plasma profiling. Except for a few outliers with a subclonal KRAS G12C mutation, dynamics of KRAS G12C VAF upon divarasib treatment mirrored that of the overall ctDNA tumor fraction. These data from a clinical study with a KRAS G12C inhibitor represent strong evidence that the KRAS G12C mutation is likely truncal in the vast majority of patients across tumor types.

In this study, we observed that patients with metastatic NSCLC, colorectal cancer, and other solid tumors bearing the KRAS G12C mutation have different mutation detectability in plasma samples associated with distinct levels of ctDNA shedding, and these results were further validated with an independent real-world pan-cancer liquid biopsy cohort. In general, patients with metastatic colorectal cancer tend to have a higher ctDNA tumor fraction and thus higher detectability of the KRAS G12C mutation in plasma samples than patients with metastatic NSCLC. This may impact the utility of a blood-based diagnostic assay for KRAS G12C inhibitors, potentially resulting in a higher rate of reflex testing with a tissue-based assay in NSCLC. In addition to tumor type, we found that ctDNA tumor fractions were associated with the extent of disease burden and sites of metastatic disease. ctDNA tumor fractions were significantly correlated with extent of disease as measured by SLD per RECIST v1.1, supporting the utility of ctDNA as a proxy of tumor burden. We also observed that patients with colorectal cancer and liver metastasis tend to have a higher ctDNA tumor fraction compared to those patients with colorectal cancer without liver metastasis. This observation is consistent with previous reports (35, 36) and may partially explain the poor prognosis associated with liver metastasis in patients with colorectal cancer.

Consistent with previous reports (24, 26), we observed that high baseline ctDNA levels tended to be associated with shorter PFS. Additionally, patients with deeper declines in ctDNA tumor fraction had a better response to treatment with divarasib and longer PFS compared to patients with less decline in ctDNA tumor fraction. These results, together with previous studies (4, 37, 38), reinforce the value of ctDNA as a potential early marker of treatment response and patient outcomes.

Our study has limitations. Despite the overall study size of more than 130 patients, there is a smaller sample size for each tumor type that limits the statistical power of this study. Additionally, we utilized two different ctDNA assays, PredicineWES+ and PredicineBEACON MRD assay, for the profiling of baseline and on-treatment plasma samples, respectively. The PredicineBEACON MRD assay was deployed for its enhanced sensitivity for on-treatment samples where ctDNA tumor fractions were generally very low. Despite this difference in the assay type resulting in the distinct algorithms of calculating ctDNA tumor fractions, the study of ctDNA dynamics comparing on-treatment versus baseline levels is believed to be robust with the high concordance between these two assays (Supplementary Fig. S1). As an independent validation, we also conducted analyses focusing on the KRAS G12C mutation where the calculation of VAF is minimally impacted by the ctDNA assay type. Overall, observations made using KRAS G12C VAF were highly consistent with those using overall ctDNA tumor fraction. Lastly, these analyses are retrospective and exploratory in nature, and, therefore, future independent validation is warranted.

In summary, longitudinal ctDNA profiling in this study establishes the baseline ctDNA tumor fractions across KRAS G12C-positive solid tumor types and provides valuable insights into the relationship between ctDNA dynamics and clinical outcomes with divarasib treatment.

Y. Choi reports personal fees from Genentech and F. Hoffmann La Roche outside the submitted work. N.V. Dharia reports other support from Genentech Inc., a member of the Roche Group, during the conduct of the study. T. Jun reports other support from Genentech during the conduct of the study and from Sema4 outside the submitted work. J. Chang reports other support from Genentech outside the submitted work. S. Royer-Joo reports other support from Genentech/Roche during the conduct of the study. K.K. Yau reports being an employee (and shareholder) of Hoffmann-La Roche Ltd. as noted in the author affiliations. Z.J. Assaf reports other support from Roche/Genentech during the conduct of the study and from Roche/Genentech outside the submitted work. J. Aimi reports other support from Roche/Genentech during the conduct of the study. S. Sivakumar reports personal fees from Foundation Medicine and other support from Roche during the conduct of the study. M. Montesion reports personal fees from Foundation Medicine and Roche Holding AG outside the submitted work. A. Sacher reports being a member of the Consulting and Advisory Board (no personal fees) for AstraZeneca, Genentech-Roche, Merck, and Amgen as well as being an Institutional Research and Clinical Trial PI for AstraZeneca, Amgen, Genentech, Merck, Lilly, Pfizer, BMS, Spectrum, GSK, Iovance, CRISPR Therapeutics, BridgeBio, HotSpot Therapeutics, and AdaptImmune. P. LoRusso reports other support from AbbVie, Roche-Genentech, Takeda, SOTIO, Agenus, IQVIA, Pfizer, GlaxoSmithKline, QED Therapeutics, AstraZeneca, EMD Serono, Kyowa Kirin Pharmaceutical Development, Kineta, Zentalis Pharmaceuticals, Molecular Templates, ABL Bio, STCube Pharmaceuticals, I-Mab, Seagen, imCheck, Relay Therapeutics, Stemline, Compass BADX, Mekanistic, Mersana Therapeutics, BAKX Therapeutics, Scenic Biotech, Qualigen, Roivant Sciences, NeuroTrials, Actuate Therapeutics, Atreca Development, Amgen CodeBreak 202, Cullinan, DrenBio, Quanta Therapeutics, Schrodinger, and Boehringer Ingelheim outside the submitted work. J. Desai reports grants, personal fees, and non-financial support from BeiGene, Amgen, and from Roche/Genentech during the conduct of the study; personal fees from Pierre-Fabre, Bayer, Boehringer Ingelheim, Daiichi Sankyo Europe GmbH, Pfizer, Ellipses Pharma, and from Incyte outside the submitted work; grants and non-financial support from GlaxoSmithKline and AstraZeneca; and non-financial support from Bristol Myers Squibb. J.L. Schutzman reports other support from Genentech during the conduct of the study. Z. Shi reports other support from Roche/Genentech during the conduct of the study.

Y. Choi: Formal analysis, writing–review and editing. N.V. Dharia: Writing–review and editing. T. Jun: Writing–review and editing. J. Chang: Writing–review and editing. S. Royer-Joo: Writing–review and editing. K.K. Yau: Writing–review and editing. Z.J. Assaf: Writing–review and editing. J. Aimi: Writing–review and editing. S. Sivakumar: Writing–review and editing. M. Montesion: Writing–review and editing. A. Sacher: Investigation, writing–review and editing. P. LoRusso: Investigation, writing–review and editing. J. Desai: Investigation, writing–review and editing. J.L. Schutzman: Writing–review and editing. Z. Shi: Conceptualization, writing–original draft, writing-review and editing.