Main

Comprehensive genomic profiling is the standard of clinical care for advanced solid tumors in modern oncology1. Circulating tumor DNA (ctDNA)-based genomic profiling is widely used because of its minimally invasive nature and rapid turnaround time, both of which have been demonstrated to increase the identification of patients eligible for targeted therapy and improve survival2,3. Despite these advantages, ctDNA genotyping is still not universally adopted and is often used as an alternative to tissue-based testing4,5.

Another, less-studied advantage of ctDNA analysis with similar potential to improve the quality of precision oncology care is its ability to capture both temporal and geographic heterogeneity within tumors. In brief, ctDNA represents a composite of DNA released from all tumor subclones, some of which may comprise genomic alterations that differ from the original dominant tumor clone and that influence optimal therapy selection6,7,8. Moreover, the relative prevalence of clones bearing such nascent alterations (herein referred to as clonality) may influence their biological relevance, with rare clones potentially having less clinical relevance than more common ones. Despite this potential relevance, evidence regarding both methods for assessing intratumoral heterogeneity and its clinical impact is scant.

To address this knowledge gap and investigate ctDNA-guided treatment in general, we launched the GOZILA study, a large, nationwide, prospective observational study designed to profile ctDNA genomic alterations for patients with advanced solid tumors in which patients are treated based on identified biomarkers in affiliated clinical trials or clinical practice9,10. The GOZILA study previously showed the superiority of ctDNA genotyping in patient screening for clinical trials versus tissue genotyping11. Here, we evaluate the efficacy of targeted therapies matched to biomarkers detected by ctDNA in over 4,000 patients with advanced cancer and use this dataset to define the influence of biomarker clonality on outcomes in response to genomically targeted therapy. These data describe a rubric by which ctDNA may more informedly be integrated into clinical practice.

Results

Patient characteristics

Of the 5,280 patients enrolled in the GOZILA study from January 2018 to August 2022, 4,037 underwent subsequent systemic therapy after blood sampling and were included in this analysis (Extended Data Fig. 1). Median duration of follow-up was 10.8 months (range 0 to 56.1). The demographics and baseline characteristics of patients are shown in Extended Data Fig. 2. Median age was 64 years (range 20 to 91) with 39% of the patients being female. Major cancer types included colorectal (n = 1,820, 45%), pancreatic (n = 899, 22%), esophageal (n = 363, 9%), gastric (n = 327, 8%) and biliary tract cancer (n = 291, 7%). The most common metastatic sites were liver (n = 2,082, 52%), lymph node (n = 1,475, 37%), lung (n = 1,270, 31%) and peritoneum (n = 954, 24%). This large-scale cohort serves as a robust representation of patients commonly diagnosed with advanced solid tumors, with an emphasis on gastrointestinal cancers.

Biomarker-matched therapy guided by ctDNA genotyping

Similar to previous reports on such tumor types and published compendia, TP53 was the most frequently altered gene (33%), followed by APC (19%), KRAS (13%) and EGFR (11%) (Fig. 1a). Supplementary Table 1 presents the identified biomarkers. No variant was detected in 333 (8%) patients. As of the data cutoff date, 952 (24%) patients had been treated with at least one targeted therapy matching biomarkers identified by ctDNA genotyping (Fig. 1b). By contrast, 2,479 (61%) patients received no matched therapy despite having targetable biomarkers identified.

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

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.

Full size image

Among the more common cancer types with ≥50 patients, colorectal cancer (38%) was most likely to be treated with biomarker-matched therapy, whereas patients with pancreatic (4%), esophageal (8%) and neuroendocrine tumor or carcinoma (10%) received matched therapy less often. Biomarker-matched treatments primarily served as the first, second or third line of therapy (58%) (Fig. 1c). Of 1,106 biomarker-matched treatments, 362 (33%) were investigational, provided under clinical trials affiliated with GOZILA, with the remainder supplied as part of routine clinical practice (Fig. 1d). As expected from the targeted therapy approval landscape, the majority of patients with biliary tract cancer (76%), neuroendocrine tumor or carcinoma (63%), esophageal cancer (61%), gastric cancer (50%) and hepatocellular carcinoma (50%) received their matched therapy through clinical trials.

Efficacy of biomarker-matched therapy guided by ctDNA

To understand the impact of biomarker-matched therapy on patient outcomes, treatments were curated according to distinct molecular biomarkers. The primary biomarkers guiding matched therapy included RAS wild-type (n = 473), ERBB2 amplification and mutations (n = 163), BRAF mutations and amplification (n = 113), high tumor mutational burden (TMB, n = 77) and microsatellite instability (MSI, n = 52) (Fig. 2a). Biomarkers leading to targeted therapy varied across cancer types, suggesting that the prevalence of actionable biomarkers and the availability of targeted therapies may differ among tumor types (Extended Data Fig. 3a). In targetable biomarkers for which targeted therapy could be recommended, MSI high (76%) most frequently led to actual treatment with matched agents, followed by ERBB2 (51%), KIT (48%), TMB high (28%), BRAF (22%) and MET (20%) (Fig. 2b). MSI or TMB high, and ERBB2 alterations consistently led to the administration of targeted therapy across various cancer types (Extended Data Fig. 3b). The drug targets for matched therapies provided were epidermal growth factor receptor (EGFR) (n = 480), human epidermal growth factor receptor 2 (HER2) (n = 150), BRAF (n = 101), programmed cell death 1 (PD-1) and programmed cell death ligand 1 (PD-L1) (n = 90) and MEK (n = 78) (Extended Data Fig. 4a).

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

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 .

Full size image

Classifying efficacy of targeting individual biomarkers when at least 10 patients were treated, therapies for ERBB2 amplification and mutations demonstrated the highest objective response rate (ORR) (50%), followed by MSI high (47%), RAS wild-type (36%), TMB high (34%), KRAS G12C (27%), FGFR1-3 alterations (26%), and BRAF alterations (24%) (Fig. 2c). Progression-free survival (PFS) was the most favorable with therapy for MSI high (median, 8.0 months), followed by RAS wild-type (median, 6.0 months), ERBB2 (median, 5.6 months), TMB high (median, 4.4 months), KRAS G12C (median, 4.1 months), BRAF (median, 2.9 months) and MET (median, 2.9 months) (Fig. 2d and Extended Data Fig. 4b). Targeted therapies directed against MSI or TMB high, or ERBB2 demonstrated consistently high efficacy across different cancer types (Extended Data Fig. 5a,b).

Recent advancements in cancer therapy have focused on the development of various drug classes for precise targeted treatment. We then assessed the classes of drugs used in the GOZILA study. These included monoclonal antibodies (54%), trailed by small-molecule inhibitors (29%), and antibody–drug conjugates (ADCs) (7%) (Extended Data Fig. 4c). When we examined the ORR for each class of drug in biomarker-matched therapy, ADCs showed the highest efficacy, with an ORR of 57%, compared with 36% for monoclonal antibodies (P < 0.001 versus ADCs) and 20% for small-molecule inhibitors (P < 0.001 versus ADCs) (Extended Data Fig. 4d). Median PFS was 6.0 months (95% confidence interval (CI), 4.8 to 10.1 months) with ADCs and 5.6 months (95% CI, 5.2 to 6.4 months) with monoclonal antibodies. Both of these drug classes significantly outperformed small-molecule inhibitors (ADCs versus small-molecule inhibitors, hazard ratio (HR) 0.46, 95% CI 0.34 to 0.65; monoclonal antibodies versus small-molecule inhibitors, HR 0.51, 95% CI 0.44 to 0.60), which showed median a PFS of 2.8 months (95% CI 2.6 to 3.0 months) (Fig. 2e).

Further analysis compared the ORR in our study’s cohort with those reported in previous tissue-based studies, stratified by biomarkers, treatments and cancer types12,13,14,15,16,17,18. Anti-PD-1 or anti-PD-L1 therapies for MSI or TMB high disease and HER2-targeted therapies based on ctDNA genotyping exhibited higher ORRs compared with those derived from tissue analysis, whereas the effectiveness was comparable in therapies targeting RAS wild-type colorectal cancer (CRC) and BRAF V600E and FGFR-altered solid tumors (Fig. 2f). Although there were differences in patient characteristics, particularly in the proportion of gastrointestinal cancers, the GOZILA study had a similar or higher number of previous treatment lines across all groups (Supplementary Tables 2–9). In summary, the study demonstrates that treatments aligned with biomarkers identified through ctDNA profiling can achieve therapeutic efficacy that is on par with, or in some cases superior to, that of traditional tissue-based genotyping. This finding strengthens the argument for the utilization of ctDNA genotyping as a viable and effective approach in precision medicine and targeted therapy.

Survival benefit of ctDNA-guided biomarker-matched therapy

Despite the advanced line of therapy of most of the enrollment and the investigational nature of many of the interventions, patients receiving ctDNA-matched therapy were more than twice as likely to respond (ORR 33%, 95% CI 30% to 36%) compared with those not harboring targetable biomarkers (ORR 12%, 95% CI 11% to 13%, P < 0.001) (Fig. 3a). Matched therapy was similarly associated with extended median PFS (4.7 months, 95% CI 4.2 to 5.3 months) relative to unmatched (2.8 months, 95% CI 2.8 to 3.0 months; HR 0.65, 95% CI 0.61 to 0.71) (Fig. 3b) as well as overall survival (OS) (18.6 months, 95% CI 17.4 to 20.5 months versus 9.9 months, 95% CI 9.5 to 10.4 months; HR 0.54, 95% CI 0.49 to 0.59) (Fig. 3c). Compared with the unmatched group, patients harboring no actionable biomarkers with recommended targeted therapies had similar ORR (10%, P = 0.08) and longer PFS (median 3.9 months; HR 0.73, 95% CI 0.68 to 0.79) and OS (median 16.8 months; HR 0.60, 95% CI 0.54 to 0.67). We further investigated the efficacy of matched therapy versus unmatched therapy for patients with APC, KRAS, KIT and BRCA1/2 alterations associated with low ORR but found no significant differences between the two groups (Supplementary Tables 10 and 11).

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

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.

Full size image

To address the heterogeneity of the study population and potential confounding factors, we performed multivariate analyses for ORR, PFS and OS, including cancer types, the type of therapy and line of therapy as variables (Extended Data Fig. 6). Despite the adjustment for these factors, matched therapy remained significantly associated with better treatment outcomes compared with unmatched therapy, strengthening our finding that ctDNA-guided matched therapy provides clinical benefit to patients with advanced solid tumors. The benefit from biomarker-matched therapy in OS was not different by sex (Extended Data Fig. 7a) and the greatest benefit was observed when used as a first-line treatment (Extended Data Fig. 7b). Statistically significant benefits for PFS and OS were observed in patients with colorectal, pancreatic, gastric and biliary tract cancers with the largest sample size (Extended Data Fig. 7c,d).

Therapeutic efficacy by ctDNA clonality

Using this dataset of patients receiving matched therapy, we next investigated whether the clonality of targeted biomarkers, calculated as the variant of interest’s allelic fraction (VAF) divided by the maximum VAF detected in that sample, influenced the observed therapeutic efficacy. We first plotted the observed efficacy as measured by both clinical benefit and PFS against VAF of genomic mutations of fusions targeted by matched therapy (Extended Data Fig. 8a,b), which demonstrated no consistent association and a trend towards an inverse correlation of benefit with VAF, potentially because of the confounding of clonal and subclonal variants and the confounding prognostic effect of high tumor fraction19. To correct for the differences in ctDNA shedding, we estimated the clonality of the targeted sequence variants as previously described11. Clinical benefit and PFS were significantly associated with clonality (Fig. 4a,b). Patients who experienced clinical benefits exhibited a significantly higher median clonality of targeted mutations or fusions (median, 92% versus 74%; P = 0.014) (Fig. 4a). PFS was also correlated with the clonality (Fig. 4b).

Fig. 4: Association of clonality with therapeutic efficacy.
figure 4

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.

Full size image

Because threshold selection is critical for therapeutic decision-making, we investigated the sensitivity of the association between clonality and patient benefit from biomarker-targeted therapies to variation in clonality threshold. As shown in Fig. 4c, HR values for PFS were consistently below 1.0 with any threshold level; however, because the clonality distribution was skewed to the extremes, the enrichment for clinical benefit was relatively robust to changes in threshold above 0.4.

At the optimal clonality threshold of 0.4, patients whose therapy was selected based on clonal biomarkers had a markedly increased rate of clinical benefit as measured by both radiographic response (26% versus 7%, P = 0.055) and PFS (median 3.0 months versus 1.4 months; HR 0.56, 95% CI 0.36 to 0.85) (Fig. 4d).

Therapeutic efficacy by plasma copy number

Similar to the effect observed for sequence variant VAF, absolute observed plasma copy number (pCN) had minimal association with either clinical benefit or PFS, with a trend toward reduced benefit in higher pCN, compatible with a confounding effect from tumor fraction (Extended Data Fig. 8c,d). However, when corrected for tumor fraction as previously described20,21, the adjusted pCN (ApCN) was significantly associated with clinical benefit and PFS (Fig. 5a,b). Patients with clinical benefit had higher ApCN (36 versus 13, P < 0.001) (Fig. 5a) and PFS was significantly correlated with ApCN (Fig. 5b). Both ORR (54% versus 31%, P = 0.003) and PFS (median, 5.9 versus 3.7 months; HR 0.64, 95% CI 0.47 to 0.89) were significantly improved in patients with high ApCN versus low when stratified by the median value of 22.42 (Fig. 5c). A consistent trend with higher ApCN in patients who experienced clinical benefit was observed across the top five cancer types (colorectal, pancreatic, esophageal, gastric and biliary tract cancers) (Extended Data Fig. 9).

Fig. 5: Association of ApCN with therapeutic efficacy.
figure 5

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.

Full size image

To further investigate the effect of ERBB2 amplification, we conducted separate analyses for ERBB2 and other genes (Extended Data Fig. 10a,b). In the ERBB2 group (n = 127), patients with clinical benefit had significantly higher ApCN compared with those without clinical benefit (median, 48 versus 18; P < 0.001). Although not statistically significant, a similar trend was observed in the other genes group, which included therapies targeting fibroblast growth factor receptor (FGFR) (n = 24), multi-tyrosine kinase (n = 22), MEK (n = 6), EGFR (n = 2) and MET (n = 2) (median, 13 versus 8; P = 0.23). Separate correlation analyses between ApCN and PFS for ERBB2 and other genes revealed positive trends in both groups, although the correlations were no longer statistically significant (Extended Data Fig. 10c,d).

These findings highlight the importance of considering both clonality and ApCN when interpreting ctDNA results and making treatment decisions based on targeted biomarkers.

Discussion

The GOZILA study has established a new benchmark in the field of precision oncology by demonstrating that patients treated with matched targeted therapy based on biomarkers identified by ctDNA genotyping can significantly extend OS compared with those receiving unmatched therapy. Although previous research has indicated a survival advantage with therapies aligned to tissue genotyping22,23,24,25,26,27,28,29, we illustrated the survival benefits of ctDNA-based treatment matching across cancer types. Furthermore, patients with clonal sequence variants and high ApCN amplifications were shown to enjoy improved efficacy, suggesting a clinical decision-making framework critical for proper deployment of ctDNA in patient care. These findings underscore the value of ctDNA genotyping as an important tool for biomarker identification in precision oncology.

Perhaps the greatest challenge in precision oncology is the timely identification of eligible patients. In GOZILA, 24% of patients were treated with therapies matched to ctDNA-based biomarkers, much higher than in other published studies examining the utility of tissue-based genomic profiling30,31,32,33,34,35,36,37 and congruent with the published literature regarding the advantages of ctDNA genotyping relative to tissue-based testing. Indeed, the initial report of the GOZILA enrollment11 specifically demonstrated that the increase in matched therapy with ctDNA screening was largely a because of ctDNA’s faster turnaround time resulting in more patients having genotyping results at the time of therapy selection relative to tissue testing. In addition, the recent PRISM study underscores the importance of access to advanced biomolecular technologies for the successful implementation of precision oncology. The proportion of patients receiving matched therapy based on identified biomarkers varied, potentially influenced by the availability of relevant clinical trials. Notably, biomarkers more frequently associated with the administration of matched treatments included ERBB2, BRAF, RAS wild-type, MET, FGFR, TMB and BRCA1/2, which were targeted in GOZILA-affiliated trials10.

In addition to increasing the proportion of patients receiving targeted therapy, ctDNA-directed therapy in GOZILA also demonstrated widespread efficacy, with patients more than twice as likely to respond to matched therapy and a concomitant reduction in their risk of death by more than 40% regardless of confounding baseline characteristics. As expected from their own literature, survival effects did vary between specific therapies, reflecting the heterogeneity of the treatments and patient populations included in the study. This heterogeneity is also possibly associated with a favorable prognosis in the no actionable biomarker group. The absence of actionable biomarkers is associated with better prognosis in colorectal cancer, possibly because of the lack of genomic alterations related to the resistance against anti-EGFR therapy (for example, RAS, BRAF V600E and ERBB2 amplification)38,39. By contrast, in lung and pancreatic cancers, patients without actionable biomarkers have been reported to have similar outcomes as those with actionable biomarkers who do not receive matched therapy23,28. Moreover, the favorable prognosis of patients with no actionable biomarkers may be attributed to patients with the absence of detectable variants in ctDNA, which is known to be associated with significantly better prognosis19,40. Despite the favorable prognosis of patients with no actionable biomarkers, our study still demonstrates a trend towards better outcomes for patients with actionable biomarkers who received matched therapy compared with those with no actionable biomarkers. This finding further highlights the importance of matched therapy in precision oncology.

One observation of particular importance in this report is that the survival advantage associated with ctDNA-guided therapy diminishes after first-line treatment. This is compatible with commonly held theories of cancer evolution under therapeutic selective pressure, which have progressively shifted targeted therapy application to earlier lines of treatment to improve survival41,42. However, realization of this survival benefit has been hampered by access to and use of comprehensive genomic profiling to guide first-line treatment. In part, this is due to reliance on tissue-only testing practices; tissue testing’s lengthy processing time often leads to the initiation of empirical systemic therapy before the availability of genomic results, whereas ctDNA genomic profiles are typically returned quickly enough to inform first-line treatment decision-making. Moreover, in other healthcare systems, access to comprehensive genomic profiling is limited to later lines of therapy or even after the exhaustion of all other therapeutic options. The data in this report and others3 demonstrate that such practices reduce patient survival.

To fully access the survival benefits of ctDNA-based testing, clinical guidelines are necessary to ensure evidence-driven implementation, particularly in diagnostic scenarios unique to ctDNA testing such as decision-making regarding subclonal biomarkers. The initial publication of GOZILA11 highlighted the high prevalence of subclonal biomarkers, and the current work underscores the clinical relevance of treating them appropriately. Specifically, absolute VAF and pCN did not accurately predict benefit from targeted therapy; however, variant clonality and ApCN did, suggesting these are more appropriate metrics for clinical decision-making. This finding is compatible with previous research regarding the effectiveness of HER2-targeted therapies in relation to ApCN in patients with HER2-positive colorectal cancer20,21.

An important caveat in this study is the heterogeneity of the cancer types and therapeutic regimens, which includes subsets with relatively few patients and investigational therapies with variable efficacy. Although ctDNA-guided therapy demonstrated significant therapeutic benefit for the overall cohort and for cancer types with adequate sample sizes, additional investigation is necessary for cancers represented here by smaller numbers of patients. Furthermore, the majority of cancers included in this study are of gastrointestinal, hepatobiliary and pancreatic origin. The high mutational heterogeneity of these cancers raises concerns about the generalizability of our results to other tumor types with different genetic landscapes. Future studies involving a broader spectrum of cancer types are needed to validate the efficacy of ctDNA-guided therapy across diverse oncological conditions. The analysis by therapeutic classes also has limitations, such as the heterogeneity of targets and therapies, pharmacokinetic and pharmacodynamic variability, tumor tissue of origin, and other potential imbalances between the groups. Another limitation is the lack of tissue profiling data as a comparator. Without this comparison, it remains uncertain how much incremental clinical value ctDNA analysis provides over traditional tissue-based methods, including the extent to which low ctDNA shedding may affect the detection of biomarkers compared with tissue testing. This gap highlights the need for integrated studies that compare the outcomes of ctDNA and tissue profiling side by side. Nevertheless, the efficacy of ctDNA-guided therapy was comparable to or better than one reported in pivotal clinical trials, although such comparisons must be interpreted with caution because of inherent biases related to different patient populations and therapies.

Together, these data describe the relevance and survival benefit of comprehensive genomic profiling using ctDNA across solid tumors and therapeutic interventions. Furthermore, the predictive ability of clonality describes an effective, data-driven rubric by which clinical decisions may be made in the common scenario of subclonal biomarkers with therapeutic relevance. ctDNA is already broadly used in genotyping patients with advanced cancer to select therapies, but guidelines have not provided comprehensive recommendations. For example, the European Society for Medical Oncology recommendation refers to the use of ctDNA for limited variants, such as PIK3CA, ESR1 and BRCA1/2 in breast cancer, ERBB2 in gastric cancer, IDH1 and FGFR2 in cholangiocarcinoma, RAS, BRAF V600E, EGFR and MSI in colorectal cancer, and DNA damage repair genes in prostate cancer5. In our study, we leverage data from over 4,000 patients with advanced cancers representing multiple tumor types. We provide both survival-based evidence supporting the use of ctDNA and, critically, outcome-based data clarifying the interpretation of VAF and variant clonality in clinical practice. These findings warrant the adoption of comprehensive genomic profiling using ctDNA as a standard-of-care approach to first-line patients with advanced solid tumors.

Methods

Study design and patients

SCRUM-Japan GOZILA is a nationwide plasma genomic profiling study involving 31 core cancer institutions in Japan (National Cancer Center Hospital East, Aichi Cancer Center, National Cancer Center Hospital, NHO Kyushu Cancer Center, Hokkaido University Hospital, Saitama Cancer Center, Kanagawa Cancer Center, Kansai Rosai Hospital, NHO Shikoku Cancer Center, NHO Osaka National Hospital, University of Tsukuba Hospital, Chiba Cancer Center, Kyorin University Hospital, Kindai University Hospital, Kyushu University, St. Marianna University School of Medicine, Osaka University, Cancer Institute Hospital of Japanese Foundation for Cancer Research, Kobe City Medical Center General Hospital, Osaka Medical and Pharmaceutical University Hospital, Gifu University, Kanazawa University, Shizuoka Cancer Center, Kagawa University Hospital, Keio University Hospital, Saitama Medical University International Medical Center, Shimane Prefectural Central Hospital, Kansai Medical University Hospital, Kyoto Katsura Hospital, Osaka International Cancer Institute, Osaka General Medical Center). The key eligibility criteria include: (1) histopathologically confirmed unresectable or metastatic gastrointestinal cancers; (2) patient aged ≥20 years; and (3) a life expectancy of at least 12 weeks. Although the study initially focused on gastrointestinal cancers, the eligibility expanded to include all solid tumors as the study progressed. To avoid the suppression of ctDNA shedding due to chemotherapy, patients are required to have received no previous systemic therapy or to have disease progression during systemic therapy and should not have started subsequent therapy at the time of blood sampling. Eligible patients provide written informed consent, and ctDNA genotyping was performed using Guardant360 (Guardant Health).

The study is conducted in accordance with the Declaration of Helsinki and the Japanese Ethical Guidelines for Medical and Biological Research Involving Human Subjects. The study protocol was approved by the institutional review board of each participating institution and registered at the University Hospital Medical Information Network Clinical Trials Registry (UMIN000029315), and the study was launched in January 2018. All patients provided written informed consent before participating in the study.

ctDNA genotyping

Next-generation sequencing analysis of ctDNA was performed using Guardant360 (Guardant Health), an assay approved by the US Food and Drug Administration and Japan Pharmaceuticals and Medical Devices Agency and performed at a Clinical Laboratory Improvement Amendments-certified, College of American Pathologists-accredited, laboratory approved by New York State Department of Health, as previously described43. Guardant360 interrogates 74 genes and detects single nucleotide variants, indels, fusions and copy number alterations as well as MSI. For the GOZILA study, 2 × 10 ml of whole blood was collected in a cell-free DNA blood collection tube (Streck, Inc.) from enrolled patients and sent to Guardant Health. Then, 5–30 ng of cell-free DNA isolated from plasma was labeled with nonredundant oligonucleotides (‘molecular barcoding’), enriched using targeted hybridization capture, and sequenced on the NextSeq 550 platform (Illumina). Base call files generated using RTA software v.2.12 (Illumina) were demultiplexed using bcl2fastq (v.2.19) and processed using a custom pipeline for molecular barcode detection, sequencing adapter trimming and base quality trimming. Processed reads were aligned to hg19 using the Burrows–Wheeler aligner-MEM algorithm44. Somatic ctDNA alterations were detected by comparing read and consensus molecule characteristics to sequencing platform- and position-specific reference error noise profiles determined independently for each position in the panel43. ctDNA fraction was measured by the maximum VAF. To estimate the cell-free DNA clonality for somatic single nucleotide variants, indels and fusions, relative clonality was initially defined as alteration VAF divided by maximum somatic VAF in the sample. The ApCN was calculated as follows: ApCN = (observed pCN − 2 × [1 − T%])/T%, where T% = 2 × maximum VAF/100. TMB score was calculated by adjustment of somatic mutation count by tumor fraction and TMB score of ≥10 was defined as TMB high.

Therapeutic selection and clinical data collection

Patients were treated according to the guidelines for their cancer type, with enrollment in clinical trials of genotype-matched targeted agents based on biomarker results identified by ctDNA genotyping. The affiliated trials included company-sponsored and investigator-initiated trials. If the biomarker identified in the patient could be a target of the drug used to treat the patient, then the patient was considered to have received matched therapy. Patients with no biomarker that can be targeted by available therapies were grouped as having no actionable biomarker.

Clinicopathological information and efficacy data for systemic therapy were collected using an electronic data capture system, TrialMaster v.5.0 (Anju Software). The clinical data were updated annually. All data were finalized by combining auto-generated and manually added queries from the SCRUM-Japan Data Center. Clinical data and biomarker results were stored in a clinical grade database and used for integrated clinicogenomic analysis.

Statistical analysis

Tumor response to systemic therapy was assessed according to the Response Evaluation Criteria in Solid Tumors v.1.1. Clinical benefit was defined as tumor response or stable disease for four or more months. PFS was estimated as the time from the date of treatment initiation to the date of disease progression according to the investigator’s assessment or the date of death from any cause. OS was estimated from the date of enrollment to the date of death from any cause.

ORR, PFS and OS were compared among patients who received and did not receive matched therapy and those with no actionable biomarker leading to targeted therapy. The Kaplan–Meier method was used to estimate survival rates, and treatment groups were compared using the log-rank test. HR values and 95% CI were calculated using the Cox proportional hazards model. Statistical comparisons of ORR were performed using the chi-squared test or Fisher’s exact test. Clonality and pCN were compared using the Mann–Whitney test between patients with and without clinical benefit. All statistical tests were two-sided without adjustment for multiple comparisons. Statistical analyses were performed using R v.4.2.1. P values <0.05 were considered statistically significant. The data cutoff date for the analyses was 31 March 2023.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.