Combined Analysis of Concordance between Liquid and Tumor Tissue Biopsies for RAS Mutations in Colorectal Cancer with a Single Metastasis Site: The METABEAM Study

Colorectal cancer is the third most common cause of cancer-related deaths, and the fourth most diagnosed cancer worldwide (1). Substantial progress has been made in treating metastatic colorectal cancer (mCRC) through the development of anti-VEGF and anti-EGFR mAbs (2, 3). However, several studies have reported that patients with mCRC with tumors harboring RAS mutations are unlikely to benefit from anti-EGFR antibodies (2, 3); therefore, such therapies are not recommended (4).

Almost all patients with mCRC have low, but detectable levels of cell-free circulating tumor DNA (ctDNA), in which tumor mutations can be characterized (5). Liquid biopsy is highly sensitive for detecting mutations present at low frequencies.

The OncoBEAM™ RAS CRC Kit (Sysmex Corporation), which uses highly sensitive BEAMing digital PCR technology (6–8), is an in vitro diagnostic tool that has been CE-marked (conformity with health, safety, and environmental protection standards for products sold within the European Economic Area) and approved by the Pharmaceuticals and Medical Devices Agency in Japan for detecting plasma RAS mutations in patients with mCRC. This kit has been widely used for patients whose tissues are difficult to collect and who are planned to be treated with anti-EGFR antibody rechallenge. A comparison between the use of BEAMing technology to determine the RAS mutational status in plasma ctDNA and the reference method of tumor tissue DNA in previous clinical trials revealed concordance rates between 86% and 93% in all patients with mCRC regardless of the metastatic sites (9–13).

Previous studies of small numbers of patients suggested that the concordance rates in ctDNA and tissue RAS mutational status varied by metastatic site (9, 10). An adequate concordance between the plasma and tissue RAS mutational status was observed in patients with mCRC with liver metastases alone (9, 10). Although they were thought to be associated with a significant discordance in patients with only lung metastases, the baseline longest diameter and number of lesions in the lungs were significantly associated with concordance (9). Moreover, a range of concordance rates depending on the metastatic site was suggested in previous studies (9, 10).

We integrated individual patient data from cases evaluated in previous studies (9–11) wherein the RAS status in ctDNA using the OncoBEAM™ RAS CRC Kit was examined. Through a combined analysis, we investigated the clinical factors associated with concordance between the plasma and tissue RAS mutational status in patients with mCRC with liver, peritoneal, or lung metastases only to identify individuals eligible for liquid biopsy analysis.

Individual patient data from one Japanese (9) and two Spanish studies (refs, 10, 11; n = 662), which examined the concordance rates between plasma and tissue RAS mutational status using the OncoBEAM™ RAS CRC Kit and tissue reference method, respectively, were integrated. Patients who met the following criteria were selected (n = 261, CONSORT diagram; Supplementary Fig. S1): (i) patients with pathologically confirmed metastatic colorectal adenocarcinoma; (ii) patients who were chemo naïve or confirmed to have a progressive disease without having initiated subsequent treatment; (iii) patients with no prior treatment with anti-EGFR mAb or regorafenib; (iv) patients with only liver metastases, peritoneal metastases, or lung metastases; and (v) patients with no tumor removal before blood sample collection. The interval between plasma collection and CT was ≤3 months.

Tumor burden data for post-hoc analysis were obtained by investigating the baseline longest diameter of the largest lesion and number of tumor lesions on CT within 3 months before or after ctDNA sample collection. Scatterplots of concordant and discordant cases based on the tumor burden data were used to confirm the significance of the cutoffs. This combined study was approved by the ethics and review committees at each institution, and all patients provided written informed consent. All procedures related to the study were performed in accordance with the Helsinki Declaration and the Ethical Guidelines for Medical Health Research Involving Human Subjects in Japan and Spain.

This study's primary outcome was to investigate the clinical factors associated with concordance between the plasma and tissue RAS mutational status. The secondary outcome was to determine the optimal cutoffs for clinical factors to maintain a sufficient concordance greater than 90% at each metastatic site.

Continuous variables were described using medians and ranges, whereas categorical variables were presented as percentages. The concordance between the plasma and tissue RAS mutational status in the overall cohort, RAS-mutant concordance based on tissue RAS-mutant cohort, and RAS wild-type concordance based on tissue RAS wild-type cohort were calculated with 95% confidence interval (CI) and tested using Fisher's exact test.

Factors associated with concordance were calculated using univariate and multivariate logistic regression models. Multivariate analysis was performed for factors with P < 0.1 based on univariate analysis. Variables were entered directly if P < 0.1 in the univariate analysis. Statistical analyses were performed using R version 3.6.1.

Of the 261 patients with mCRC who were initially enrolled (CONSORT diagram; Supplementary Fig. S1), 40 were excluded because of missing measurements for the baseline longest diameter or number of lesions (n = 15) or due to an interval of more than 3 months between plasma collection and CT (n = 25). The full analysis set comprised of 221 patients who had liver metastases alone (n = 151), peritoneal metastases alone (n = 25), and lung metastases alone (n = 45). Patient characteristics categorized by metastatic sites are shown in Supplementary Table S1. The cutoffs for the baseline longest diameter and number of lesions were provisionally set to 20 mm and 10 lesions, respectively, based on a previous report (9).

The overall concordance rate in liver metastases alone was 91% (95% Cl, 85%–95%), whereas those of peritoneal and lung metastases alone were 88% (95% CI, 68%–97%) and 64% (95% CI, 49%–78%), respectively. Similar trends were observed in the European and Asian subsets (Fig. 1). The concordance rate of lung metastases alone was relatively low compared with that of liver or peritoneal metastases alone. The mean of mutation allele fraction (MAF) of liver, peritoneum, and lung was 6.8%, 7.2%, and 2.6%, respectively. The MAF was relatively low in lung metastases alone, as shown in Supplementary Fig. S2.

RAS mutations were detected in 39% and 47% of plasma and tissue samples, respectively. The overall concordance between plasma- and tissue-based analyses was 85% (188/221; 95% CI, 80%–89%), with 76% (79/104; 95% CI, 66%–84%) RAS-mutant concordance based on tissue RAS-mutant cohort and 93% (109/117; 95% CI, 87%–97%) RAS wild-type concordance based on tissue RAS wild-type cohort. The RAS-mutant concordance was significantly lower than the RAS wild-type concordance (P < 0.001). Almost no difference in the concordance rate for each codon of RAS was observed (data not shown).

Univariate/multivariate analysis of the clinical findings between concordant and discordant cases was conducted. The analysis for identifying factors associated with discordance is described in Supplementary Table S2. On multivariate analysis, the most significant factors associated with concordance were the baseline longest diameter of the lesion and the number of lesions (P = 0.004; Table 1). Both factors are related to tumor burden and are evaluable factors in clinical practice, thus, these values were adopted as cutoffs to select concordant cases in this study.

Furthermore, univariate and multivariate analyses of the clinical findings by tissue RAS mutational status between concordant and discordant cases were conducted (Supplementary Tables S3 and S4). In the tissue RAS-mutant cohort, the analysis for identifying factors associated with discordance is described in Supplementary Table S3. In multivariate analysis of the tissue RAS-mutant cohort, the significant factors associated with the RAS-mutant concordance were collection interval (P = 0.004), baseline longest diameter, and number of lesions (P = 0.005), as it was with the overall cohort. On the other hand, there was no significant factor in univariate analysis of the tissue RAS wild-type cohort (Supplementary Table S4).

The concordance rate in the liver metastases alone was more than 90%, with any cutoff value used (Fig. 2). Using the baseline longest diameter of lesion ≥20 mm as a cutoff, the concordance rate of peritoneal metastases alone was 94%, providing a sufficient concordance and indication for OncoBEAM™ testing, with a coverage of 68%, that is, this cutoff covered 68% of patients with liver metastases. The reliability of using the cutoff of a baseline longest diameter ≥20 mm for patients with mCRC with peritoneal metastases alone was reflected by a high RAS-mutant concordance of 86% (95% CI, 42%–99%), along with 100% (95% CI, 66%–100%) RAS wild-type concordance. Using 20 mm as the cutoff value for the baseline longest diameter in peritoneal metastases alone, 67% (2/3) of discordant cases were below the cutoff value, and 73% (16/22) of concordant cases were above the cutoff (Supplementary Fig. S3).

The concordance rate for lung metastases alone using the cutoff of the baseline longest diameter ≥20 mm alone was 90%, but the coverage was poor (22%; Fig. 2). Similarly, when the number of lesions ≥10 alone was used as a cutoff, the concordance rate for lung metastases alone was 89%, with poor coverage of 20% (Supplementary Fig. S4). However, the concordance rate of lung metastases alone increased to 94%, with improved coverage of 38% when using the cutoff of a baseline longest diameter ≥20 mm and ≥10 lesions (Fig. 3), with 88% RAS-mutant concordance. Most discordant cases (94%, 15/16) occurred in the baseline longest diameter of lung lesions <20 mm and <10 lesions (Fig. 4, as shown in the black square). Concordance within and without the cutoffs were 46% (95% CI, 28%–66%) and 94% (95% CI, 69%–100%), respectively, showing a significant difference (P < 0.001).

This is the first combined analysis of individual patient data from Europe and Asia on the concordance of RAS mutational status by metastatic site between plasma ctDNA and tumor tissue DNA testing (14). There were no differences between the European and Asian data. We found that the concordance rates differed by the metastatic site. Although there is currently no clear explanation for this result, the effects may have been organ specific. Therefore, it is necessary to investigate the clinical factors associated with high concordance in each metastatic site so that patients with these clinical factors can be selected for OncoBEAM™-based detection of their RAS mutational status.

In our multivariate statistical model, the most significant concordance factors were the baseline longest diameter and the number of lesions (P = 0.004), followed by the sample collection interval (P = 0.036). Tumor burden (i.e., baseline longest diameter and number of lesions) impacts the amount of ctDNA released into plasma (15). Thus, an increasing amount of ctDNA results in increased concordance rates. The baseline longest diameter and number of lesions can be easily measured and implemented in clinical practice. In addition, the sample collection interval has been associated with discordance, in which longer intervals between plasma and tissue collections may contribute to clonal evolution (9, 16).

For patients with liver metastases alone, the concordance rate was high in Europe and Asia. As such, the lesion diameter and number are not necessary to select patients for testing using the OncoBEAM^TM RAS CRC Kit. The median MAF was higher in liver metastases than in peritoneal and lung metastases (9, 10, 12, 17). Bachet and colleagues reported that primary tumor resection and the presence of liver metastases were significantly associated with the presence of ctDNA, although liver metastases appeared to be the key factor (18). Elez and colleagues observed a high MAF burden in liver metastases, which was not related to the volume and number of lesions in liver metastases (17). ctDNA is thought to be easily released into the circulation regardless of the tumor size or lesion number in the liver.

In previous studies, patients with peritoneal metastases alone showed a low MAF burden (12, 17). MAF from patients with only peritoneal metastases was lower than in patients with peritoneal metastases and at least one other metastatic site (12). The discordance was associated with the presence of peritoneal metastases (18). However, our combined study revealed considerable high concordance (88%) without setting a cutoff value. This may be because we enrolled cases in which lesions could be detected by CT; a total of 68% (17/25) of peritoneal metastases with the longest lesion diameter ≥20 mm were included. In addition, sufficient concordance and good coverage were observed when we selected cases with the RAS wild-type concordance longest diameter ≥20 mm, with high RAS-mutant concordance, and RAS wild-type concordance. In peritoneal metastases alone, the maximum size may be related to the amount of ctDNA release.

In contrast, the concordance in patients with lung metastases alone was extremely low in Europe and Asia (9, 10), likely because of the low MAF (9, 10, 12, 17). Like previous studies, this study's results also showed that the MAF with lung metastases tended to be lower than for other metastatic sites (Supplementary Fig. S2). The release of ctDNA from lung metastases is thought to be extremely low, thus, tumor burden plays a more important role in lung metastases. As such, the lesion size and number must be determined to select appropriate patients with lung metastases alone for OncoBEAM™ testing, that is, a closer evaluation of the tumor burden is required. These two factors were also selected as cutoffs for patients with lung metastases alone in the Japanese study (9). Under the cutoff value, a liquid biopsy may not detect RAS mutations in lung metastases. In such cases, the tissue-based analysis would be the preferred choice.

In a post-hoc analysis, we reviewed the size and number of metastatic tumors of baseline CT in each case. This is the first international collaborative study to evaluate the relationship between the tumor burden and concordance by metastatic site between plasma- and tissue-based RAS mutational analysis. This study reveals the impact of the tumor burden and organ specificity on concordance rates.

In our study, the presence of previous chemotherapy was significantly associated with a higher discordance than its absence for the overall cohort (P = 0.045; Supplementary Table S2). In univariate analysis of clinical findings between concordance and discordance in overall and tissue RAS-mutant cohorts, previous chemotherapy was indicated to be one of the significant factors, as described in Table 1 and Supplementary Table S3 (P = 0.032 and P = 0.019, respectively); however, it was not a significant factor in multivariate analysis for overall and RAS-mutant cohorts (P = 0.500 and P = 0.214, respectively). This study included 50 patients who received chemotherapy between tissue and plasma collection. The mutational state was changed in 12 cases, including 10 cases (one case in the liver, three in the peritoneum, and six in the lung) that changed from mutant to wild-type, suggesting below limit-of-detection level (called NeoRAS; ref. 19). All of the 10 cases had tumors with sizes <20 mm, and each case had <10 lesions, indicating a low tumor burden. The collection interval between tissue and plasma sampling is also a significant factor in the multivariate analysis for the overall and tissue RAS-mutant cohorts, as shown in Table 1 and the Supplementary Table S3 (P = 0.036 and P = 0.004, respectively), indicating that the tumor volume reduction as a result of chemotherapy may affect the discordance of the tissue RAS mutant. Only two cases (one case each in the lung and the liver) changed from wild-type to mutant during chemotherapy. These might be due to clonal evolution or tumor heterogeneity, even in cases that did not receive targeting agents, such as an anti-EGFR antibody and/or regorafenib.

In conclusion, the OncoBEAM™ RAS CRC Kit's utility in patients with mCRC may depend on the metastatic site. Although no further selection is required for patients with liver metastases alone, careful attention should be given to patients with peritoneal metastases alone with lesion size <20 mm, as well as those with lung metastases alone with lesion size <20 mm and <10 metastatic lesions when determining the RAS mutational status using the ctDNA-based assay.

Y. Kagawa reports personal fees from Bayer, Chugai, Eli Lilly, Merck, Ono, Taiho, and Yakult Honsha, and grants and personal fees from Takeda outside the submitted work. E. Elez reports personal fees from AbbVie, Amgen, Array BioPharma, Boehringer Ingelheim, Bristol-Myers Squibb, GlaxoSmithKline, Roche, Merck Serono, MSD, Pierre-Fabre, Sanofi-Aventis, and Servier outside the submitted work. H. Bando reports grants from AstraZeneca and Sysmex and personal fees from Taiho Pharmaceutical and Eli Lilly outside the submitted work. H. Taniguchi reports grants from Sysnex, Takeda, Taiho, and Daiichi Sankyo, and personal fees from Chugai, Lilly, Takeda, and Taiho outside the submitted work. A. Vivancos reports consultancy for Sysmex Inostics. T. Denda reports personal fees from Sysmex and SAWAI Pharmaceutical Co and grants from MSD and ONO Pharmaceutical outside the submitted work. T. Nishina reports grants and personal fees from ONO Pharmaceutical, Taiho, Chugai, Merck Serono, and Lilly, and grants from Takeda, Daiichi Sankyo, and MSD during the conduct of the study. Y. Komatsu reports grants and personal fees from Takeda and Merck during the conduct of the study. Y. Komatsu also reports grants and personal fees from Daiichi Sankyo, ONO Pharmaceutical, Chugai Pharmaceutical, Taiho Pharmaceutical, Nipro, and Eli Lilly; grants from MSD; and personal fees from Yakult outside the submitted work. D. Ciardiello reports other from Sanofi during the conduct of the study. E. Oki reports personal fees from Taiho Pharmaceutical, Eli Lilly, Bayer, Takeda Pharmaceutical, Chugai Pharmaceutical, and ONO Pharmaceutical outside the submitted work. T. Yamanaka reports grants and personal fees from Chugai Pharmaceuticals and Bayer, personal fees from Takeda, and grants from Taiho during the conduct of the study. J. Tabernero reports personal fees from Array Biopharma, AstraZeneca, Bayer, BeiGene, Biocartis, Boehringer Ingelheim, Chugai Pharmaceuticals, Foundation Medicine, Roche, Genentech, Genmab, HalioDX SAS, Halozyme, Imugene, Inflection Biosciences, Ipsen, Kura, Eli Lilly, Menarini, Merck Serono, Merrimack, Merus, Molecular Partners, MSD, Novartis, Peptomyc, Pfizer, Pharmacyclics, ProteoDesign SL, Rafael Pharmaceuticals, Roche Diagnostics, Sanofi, Seagen, Seattle Genetics, Servier, Symphogen, Taiho Pharmaceuticals, and VCN Biosciences outside the submitted work. T. Yoshino reports grants from Taiho Pharmaceutical, Sumitomo Dainippon Pharmaceutical, ONO Pharmaceutical, Chugai Pharmaceutical, Amgen, Parexel International, MSD, Daiichi Sankyo, and Sanofi outside the submitted work. No disclosures were reported by the other authors.

Y. Kagawa: Writing–original draft, project administration, writing–review and editing. E. Elez: Writing–review and editing. J. García-Foncillas: Writing–review and editing. H. Bando: Project administration, writing–review and editing. H. Taniguchi: Writing–review and editing. A. Vivancos: Writing–review and editing. K. Akagi: Writing–review and editing. A. García: Writing–review and editing. T. Denda: Writing–review and editing. J. Ros: Writing–review and editing. T. Nishina: Writing–review and editing. I. Baraibar: Writing–review and editing. Y. Komatsu: Writing–review and editing. D. Ciardiello: Writing–review and editing. E. Oki: Writing–review and editing. T. Kudo: Writing–review and editing. T. Kato: Writing–review and editing. T. Yamanaka: Writing–review and editing. J. Tabernero: Writing–review and editing. T. Yoshino: Project administration, writing–review and editing.

We would like to express our special gratitude to all participating patients, their families, and all participating investigators. The authors thank P4 Statistics Co. Ltd. for providing support during the statistical analyses, and Celestina Chin and Yen May Ong of MIMS Pte Ltd. for providing medical writing support/editorial support. We also thank Editage (www.editage.com) for English language editing, which was funded by Sysmex Corporation, in accordance with Good Publication Practice (GPP3) guidelines (http://www.ismpp.org/gpp3). This work was funded and sponsored by Sysmex Corporation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.