Targeted therapy guided by circulating tumor DNA analysis in advanced gastrointestinal tumors

Abstract

Although comprehensive genomic profiling has become standard in oncology for advanced solid tumors, the full potential of circulating tumor DNA (ctDNA)-based profiling in capturing tumor heterogeneity and guiding therapy selection remains underexploited, marked by a scarcity of evidence on its clinical impact and the assessment of intratumoral heterogeneity. The GOZILA study, a nationwide, prospective observational ctDNA profiling study, previously demonstrated higher clinical trial enrollment rates using liquid biopsy compared with tissue screening. This updated analysis of 4,037 patients further delineates the clinical utility of ctDNA profiling in advanced solid tumors, showcasing a significant enhancement in patient outcomes with a 24% match rate for targeted therapy. Patients treated with matched targeted therapy based on ctDNA profiling demonstrated significantly improved overall survival compared with those receiving unmatched therapy (hazard ratio, 0.54). Notably, biomarker clonality and adjusted plasma copy number were identified as predictors of therapeutic efficacy, reinforcing the value of ctDNA in reflecting tumor heterogeneity for precise treatment decisions. These new insights into the relationship between ctDNA characteristics and treatment outcomes advance our understanding beyond the initial enrollment benefits. Our findings advocate for the broader adoption of ctDNA-guided treatment, signifying an advancement in precision oncology and improving survival outcomes in advanced solid tumors.

Fig. 1. Overview of matched therapy based on ctDNA genotyping in GOZILA.

a, Pathogenic genomic alterations across cancer types. Genes identified in at least 10 patients are displayed. Bars indicate the number of patients harboring different classes of genomic alterations. The inset bar plot provides a zoomed-in view of the genes with lower frequencies of alterations, using a different scale on the x axis to enhance visibility. b, Proportion of patients receiving ctDNA-guided matched therapy by cancer types with at least 20 patients enrolled. The yellow bar indicates the percentage of patients receiving matched therapy within each cancer type. The gray bar represents the proportion of patients with no actionable biomarker. The total number of patients per cancer type is shown above the abbreviation for the cancer type. The red line represents the number of patients with each cancer type. c, Number of patients receiving matched therapies by treatment lines. d, Proportion of patients treated in clinical trials among those receiving matched therapy for tumor types with at least 20 patients enrolled. Abbreviations for cancer types are according to Extended Data Fig. 2.

Fig. 2. Efficacy of matched targeted therapy by biomarkers.

a, Number of patients receiving matched therapy by biomarkers. The inset bar plot provides a zoomed-in view of the genes with lower frequencies of alterations, using a different scale on the y axis to enhance visibility. b, Top 10 biomarkers with the highest percentage of patients receiving matched therapy among those with biomarkers identified in at least 10 patients, for which a targeted treatment was recommended. The yellow line represents the number of patients with biomarkers that had a recommendation. c, Percentage of each tumor response of matched therapy by biomarkers. d, Number of events and median PFS of matched therapy targeting each biomarker that led to treatment in at least 10 patients. e, Kaplan–Meier plots of PFS for patients treated with matched therapy by drug class. f, Comparison of ORR between GOZILA and previous studies. The x and y axes indicate ORR of targeted therapies in previous studies and GOZILA, respectively. CR, complete response; MSIH, MSI high; NA, not applicable; PD, progressive disease; PR, partial response; SD, stable disease; TMBH, TMB high .

Fig. 3. Clinical outcomes by biomarker-treatment matching status.

a, Tumor response by biomarker-treatment matching status. b, Kaplan–Meier plots of PFS by biomarker-treatment matching status. The PFS analysis is performed per treatment, with each treatment classified as matched, unmatched or no actionable biomarkers, based on the biomarker status. c, Kaplan–Meier plots of OS by biomarker-treatment matching status. The OS analysis is performed per patient, with each patient classified as having received matched therapy if at least one of their treatments was matched to their biomarker status, as having received unmatched therapy if none of their treatments was matched to identified actionable biomarkers, or as having no actionable biomarkers.

Fig. 4. Association of clonality with therapeutic efficacy.

a, Clonality of variants targeted in patients with clinical benefit and no benefit. The clonality was significantly higher in patients with clinical benefit (two-sided P = 0.014, Mann–Whitney U-test). Boxes represent the 25th–75th percentiles, center lines indicate the median, whiskers extend to the maximum and minimum values within 1.5× the interquartile range and dots indicate outliers. b, Correlation of PFS with clonality. Shaded area represents 95% CI. P values were calculated using the two-sided Wald test. c, Forest plots of HR for PFS with each threshold of clonality. HR (circles) and 95% CI (horizontal lines) are shown for each threshold of clonality. P values were calculated using the two-sided Wald test. The vertical dashed line indicates 1.0. d, Kaplan–Meier plots of PFS on therapies targeting variants with clonality ≥0.4 versus <0.4. 6E.

Fig. 5. Association of ApCN with therapeutic efficacy.

a, ApCN of gene amplifications targeted in patients with clinical benefit and no benefit. The ApCN was significantly higher in patients with clinical benefit (two-sided P < 0.001, Mann–Whitney U-test). Boxes represent 25th–75th percentiles, center lines indicate the median, whiskers extend to the maximum and minimum values within 1.5× the interquartile range and dots indicate outliers. b, Correlation of PFS with ApCN. Shaded area represents 95% CI. P values were calculated using the two-sided Wald test. c, Kaplan–Meier plots of PFS on therapies targeting amplifications with ApCN equal to or greater than the median (22.42) versus less than the median.

Extended Data Fig. 1. Patient flow.

Of the 5,280 patients enrolled in the GOZILA study from January 2018 to August 2022, 5,178 had available ctDNA data. Among them, 1,141 patients with no systemic therapy initiation after ctDNA test were excluded. Consequently, 4,037 were included in the analysis.

Extended Data Fig. 2. Patient baseline characteristics.

Summary of baseline characteristics of patients included in the analysis. Data are n (%) or median (range).

Extended Data Fig. 3. Number of patients receiving matched therapy by drug targets.

a. Number of patients receiving matched therapy by biomarkers in each cancer type with at least 20 patients enrolled. b. Top 10 biomarkers with the highest percentage of patients receiving matched therapy, for which a targeted treatment was recommended. Yellow line represents the number of patients with biomarkers that had a recommendation. Data are shown by each cancer type with at least 20 patients enrolled.

Extended Data Fig. 4. Efficacy of matched targeted therapy by biomarkers and drug class.

a. Number of patients receiving matched therapy by drug targets. b. Kaplan-Meier plots of progression-free survival (PFS) of matched therapy by biomarkers. c. Number of patients receiving matched therapy by drug class. d. Tumor response to matched therapy by drug class.

Extended Data Fig. 5. Objective response rate (ORR) and progression-free survival (PFS) of matched targeted therapy by biomarkers in each cancer type with at least 20 patients enrolled.

a. Percentage of each tumor response of matched therapy by biomarkers. b. Median PFS (mPFS) of matched therapy targeting each biomarker. NA, not available.

Extended Data Fig. 6. Multivariate analysis for efficacy endpoints.

The odds ratio (boxes) for ORR (a) and hazard ratios (boxes) for PFS (b) and OS (c) and 95% confidence intervals (horizontal lines) are shown for each subgroup. P values were calculated using the two-sided Wald test.

Extended Data Fig. 7. Benefit of matched therapy by subgroup.

a. Forest plots of hazard ratio (HR) for overall survival (OS) of matched therapy by sex. The HR (circles) and 95% confidence intervals (CIs, horizontal lines) are shown for each subgroup. P values were calculated using the two-sided Wald test. b. Forest plots of HR for OS of matched therapy by treatment line. The HR (circles) and 95% CIs (horizontal lines) are shown for each subgroup. P values were calculated using the two-sided Wald test. c. Forest plots of HR for progression-free survival (PFS) in each cancer type including at least 40 patients. The HR (circles) and 95% CIs (horizontal lines) are shown for each subgroup. P values were calculated using the two-sided Wald test. d. Forest plots of HR for OS in each cancer type including at least 40 patients. The HR (circles) and 95% CIs (horizontal lines) are shown for each subgroup. P values were calculated using the two-sided Wald test.

Extended Data Fig. 8. Association of variant allelic fraction (VAF) or plasma copy number (pCN) with therapeutic efficacy.

a. VAF of variants targeted in patients with clinical benefit and no benefit. The clonality was significantly higher in patients with clinical benefit (two-sided p = 0.001, Mann–Whitney U-test). The boxes represent 25th–75th percentiles; center lines indicate the median; whiskers extend to the maximum and minimum values within 1.5× of the interquartile range; and dots indicate outliers. b. Correlation of progression-free survival (PFS) with VAF. Shaded area represents 95% confidence interval. P values were calculated using the two-sided Wald test. c. pCN of gene amplification targeted in patients with clinical benefit and no benefit. The pCN was significantly higher in patients with clinical benefit (two-sided p = 0.077, Mann–Whitney U-test). The boxes represent 25th–75th percentiles; center lines indicate the median; whiskers extend to the maximum and minimum values within 1.5× of the interquartile range; and dots indicate outliers. d. Correlation of PFS with pCN. Shaded area represents 95% confidence interval. P values were calculated using the two-sided Wald test.

Extended Data Fig. 9. Association of adjusted plasma copy number (ApCN) with therapeutic efficacy by cancer types.

ApCN of gene amplifications targeted in patients with clinical benefit and no benefit in top 5 cancer types. The boxes represent 25th–75th percentiles; center lines indicate the median; whiskers extend to the maximum and minimum values within 1.5× of the interquartile range; and dots indicate outliers.

Extended Data Fig. 10. Association of adjusted plasma copy number (ApCN) with therapeutic efficacy in ERBB2 and non-ERBB2 genes.

a. ApCN of ERBB2 gene amplifications targeted in patients with clinical benefit and no benefit. The ApCN of ERBB2 was significantly higher in patients with clinical benefit (two-sided p < 0.001, Mann–Whitney U-test). The boxes represent 25th–75th percentiles; center lines indicate the median; whiskers extend to the maximum and minimum values within 1.5× of the interquartile range; and dots indicate outliers. b. ApCN of non-ERBB2 gene amplifications targeted in patients with clinical benefit and no benefit. The ApCN of non-ERBB2 tended to be higher in patients with clinical benefit (two-sided p = 0.23, Mann–Whitney U-test). The boxes represent 25th–75th percentiles; center lines indicate the median; whiskers extend to the maximum and minimum values within 1.5× of the interquartile range; and dots indicate outliers. c. Correlation of progression-free survival (PFS) with ApCN of ERBB2. Shaded area represents 95% confidence interval. P values were calculated using the two-sided Wald test. d. Correlation of progression-free survival (PFS) with ApCN of non-ERBB2 genes. Shaded area represents 95% confidence interval. P values were calculated using the two-sided Wald test.