Mutational Analysis of Patients With Colorectal Cancer in CALGB/SWOG 80405 Identifies New Roles of Microsatellite Instability and Tumor Mutational Burden for Patient Outcome

Colorectal cancer (CRC) is the second-leading cause of cancer death in the United States.¹ Over the past two decades, 13 drugs, including antibodies targeting vascular endothelial growth factor (bevacizumab) and the epidermal growth factor receptor (EGFR; cetuximab and panitumumab), have been approved for the treatment of metastatic CRC (mCRC). However, the optimal combination and sequence of these drugs is likely dependent on many factors including the mutational status of the tumor cells.

Cancer and Leukemia Group B (CALGB)/SWOG 80405 was designed initially to test whether the addition of either cetuximab or bevacizumab or both to fluorouracil and leucovorin with either oxaliplatin (FOLFOX) or irinotecan leads to superior outcomes as first-line therapy in advanced CRC or mCRC. The primary end point of the trial was overall survival (OS). In 1,137 patients with KRAS wild-type (WT; codons 12 and 13), there was no statistically significant difference in OS between the bevacizumab arm and the cetuximab arm.² The regimens tested in CALGB/SWOG 80405 represent the current standard of care for first-line treatment in mCRC. CALGB is now part of the Alliance for Clinical Trials in Oncology (Alliance).

To individualize and optimize treatment, molecular analysis of tumors of patients, particularly those patients enrolled in clinical trials, is imperative. Outcome data and biospecimens collected from this relatively large study might provide an unprecedented opportunity for performing translational studies on molecular markers. We hypothesized that, in this context, novel somatic genetic alterations that drive tumor progression and/or resistance to therapy might affect patient outcome. Aside from our knowledge that the presence of any of a number of RAS mutations conveys resistance to EGFR inhibitors, and that microsatellite instability–high (MSI-H) tumors are more likely to respond to checkpoint inhibitors (which are under study in the first-line setting), there are no molecular markers to inform the selection of the most efficacious regimen for patients with mCRC.

This study aimed to determine the tumor mutational profile of patients with mCRC in CALGB/SWOG 80405, to evaluate the prognostic value of DNA mutations, and to determine the differential treatment effects of bevacizumab versus cetuximab in patients with a specific mutational profile.

The trial was designed to compare chemotherapies plus either cetuximab, bevacizumab, or cetuximab and bevacizumab as first-line treatment of advanced CRC and mCRC. Within 3 years, data suggesting that dual antibody combinations compromised outcomes and the discovery of the lack of efficacy of EGFR antibodies in KRAS mutant tumors led to a restriction of eligibility to patients with KRAS WT (codons 12 and 13; KRAS amendment) and to closure of the dual antibody group. A revised two-arm trial (chemotherapy plus either cetuximab or bevacizumab) became the primary cohort (Appendix Fig A1, online only), and the OS results have been reported.²

Eligible patients had pathology-documented, untreated, locally advanced CRC or mCRC. Patients were 18 years or older with an Eastern Cooperative Oncology Group performance status of 0 to 1 and normal hepatic, renal, and hematologic laboratory values. Random assignment was stratified by chemotherapy, prior adjuvant chemotherapy, and prior pelvic radiation.

The primary end point of OS was defined as time of random assignment until death. Patients without reported deaths were censored at their last known follow-up. Progression-free survival (PFS) was measured from time of random assignment until first documented progression or death. Patients alive without documented tumor progression were censored for PFS at the most recent disease assessment. Median follow-up is 66.5 months (95% CI, 64.3 to 69.8 months).

The numbers of patients enrolled in the trial and the specimens used for analyses are in Fig 1. Table 1 reports patient demographics and clinical characteristics. Institutional review board approval was required, and all participating patients provided written informed consent to the genetic analysis.

BRAF mutations (12% [100 of 843]) were almost exclusively V600E (98%). Patients with BRAF mutated tumors had a median OS of 13.5 months (95% CI, 11.2 to 18.5 months), and patients with BRAF WT tumors had a median OS of 30.6 months (95% CI, 29.0 to 33.5 months; hazard ratio [HR], 2.01 [95% CI, 1.49 to 2.71]; P < .001; Fig 2 [top]; Table 2). To evaluate the contribution of tumor location to the effect of BRAF mutations, tumor location was removed from the statistical model, and the resulting HR was 2.26 (95% CI, 1.70 to 3.01; P < .001; Table 2). No difference in OS was observed between the two treatment arms (primary cohort) on the basis of BRAF status (interaction P = .293; Appendix Table A3, online only).

Patients with extended RAS mutant tumors (32% [266 of 838]) had a median OS of 25.0 months (95% CI, 22.6 to 28.1 months), and patients with RAS WT tumors had a median OS of 32.1 months (95% CI, 28.9 to 35.5 months; HR, 1.52 [95% CI, 1.26 to 1.84]; P < .001; Fig 2 [bottom]; Table 2). Removing tumor location from the model does not affect this result (Table 2). No difference in OS was observed between the two treatment arms (primary cohort) on the basis of the RAS status (interaction P = .602; Appendix Table A3).

Patients with triple-negative tumors (WT for NRAS/KRAS/BRAF) had a median OS of 35.9 months (95% CI, 33.0 to 38.8 months) versus 22.2 months in patients with at least one mutated gene in their tumors (95% CI, 19.6 to 24.4 months; P < .001). In the primary cohort, this difference is even more pronounced (36.4 v 19.4 months; P < .001). The presence or absence of a PIK3CA mutation did not change the results of quadruple-negative (WT for NRAS/KRAS/BRAF/PIK3CA) compared with triple-negative tumors (results not shown).

In patients whose tumors were MSI-H (6% [52 of 827]) versus MSS, the HR was 0.87 (95% CI, 0.60 to 1.28; P = .491). Removal of tumor location from the model does not affect this result (Table 2).

In patients with MSI-H tumors, longer OS was observed in the bevacizumab arm than in the cetuximab arm (HR, 0.13 [95% CI, 0.06 to 0.30]; P < .001; Table 2). In the bevacizumab arm, median OS was 30.0 months (95% CI, 23.6 months to NE) versus 11.9 (95% CI, 10.3 to 24.6 months) in the cetuximab arm (Fig 3). In patients with MSS tumors, no difference was observed between the two arms (P = .539; Table 2). There is a differential treatment effect across microsatellite instability status (interaction P < .001, primary cohort).

TMB was available from 536 patients. In patients with MSI-H tumors (n = 35), the median TMB was 52 mutations/Mb (range, 11 to 208 mutations/Mb), and in patients with MSS tumors (n = 475), the median TMB was six mutations/Mb (range, 0 to 361 mutations/Mb; Appendix Fig A2). Two patients in the MSS group had a high TMB (93 and 361 mutations/Mb) and were excluded from the analysis. In patients with MSS tumors, on the basis of an optimized TMB cutoff of eight mutations/Mb, there were 366 patients with eight or fewer mutations/Mb in their tumors who were classified as TMB low (77%), and the other 107 patients with more than eight mutations/Mb in their tumors were classified as TMB high (23%). No major differences in patient and tumor characteristics were noted between patients with TMB-high and TMB-low tumors (Appendix Table A4, online only). Patients with TMB-high tumors had a median OS of 33.8 months (95% CI, 30.1 to 43.1 months), and patients with TMB-low tumors had a median OS of 28.1 months (95% CI, 24.9 to 31.8 months; HR, 0.73 [95% CI, 0.57 to 0.95]; P = .020; Fig 4). Removing tumor location from the model does not affect this result (Table 2). No difference in OS was observed between the two treatment arms (primary cohort) on the basis of TMB status (interaction P = .848; Appendix Table A3). An inverse correlation was observed between TMB and HRs (Appendix Fig A3). The results of these associations with OS and PFS when including only patients in the primary cohort are reported in Appendix Table A3. Mutations in APC, PIK3CA, and TP53 (mutations in FBXW7 were not included because of their low prevalence) were not associated with outcome (P > .05, results can be provided on request).

In CALGB/SWOG 80405, the presence or absence of MSI-H, high-TMB, BRAF, and extended RAS mutations define patient subgroups with different outcomes. To our knowledge, this study provides the first indication that TMB correlates with the prognosis of patients with CRC with MSS tumors. TMB, the cumulative number of somatic DNA point mutations, describes the level of genomic instability in each patient’s tumor. Patients with nonhypermutated MSS tumors (93% of patients in our study) had different OS on the basis of whether their TMB was classified as high or low. Patients with a high TMB had better OS than did patients with a low TMB (Fig 4). A high TMB probably reflects the presence of mutation-associated neoantigens, with consequent increased lymphocyte infiltration in the tumor microenvironment. This phenomenon has been observed even in MSS tumors.⁵ A subset of MSS tumors are immunogenic, and patients with that subgroup of tumors have a different risk of relapse. In stage I to III CRCs, up to 50% of MSS tumors have a high immunoscore.⁶ An elevated neoantigen load and/or high immunoscore was associated with better survival.^5,6 It has also been demonstrated that 16% of nonhypermutated MSS tumors have an immunophenotype similar to that of MSI-H tumors.⁷ In our study, detecting a high TMB in patients with stage IV MSS might identify tumors with a favorable immune microenvironment.

In patients with MSS mCRC in this study, the effect of TMB on OS is not as strong as other markers (eg, BRAF), and its relevance to direct therapy is still to be determined. Because the relevance of TMB in this setting is more biologic than clinical, additional analyses should evaluate which TMB-high MSS tumors are immunogenic and which mutations give rise to neoantigens that elicit an antitumoral, adaptive response.

With the recent US Food and Drug Administration approval of checkpoint inhibitors for the treatment of patients with MSI-H mCRC refractory to standard chemotherapy, MSI-H testing is now requested routinely. In our study, MSI-H status did not confer a prognostic effect after multivariable adjustment. It should be noted that patients with MSI-H tumors, when not adjusted by all the other covariables in the model, exhibited a trend to a worse outcome, suggesting that it may be a negative prognostic marker (median OS, 21.5 months [95% CI, 14.8 to 26.9 months] v 29.5 months [95% CI, 27.2 to 31.7 months] in patients with MSS tumors; P = .087). When patients with MSI-H tumors received chemotherapy and bevacizumab, this treatment conferred a survival advantage over chemotherapy and cetuximab (Fig 3). In the chemotherapy and bevacizumab arm, median OS was 30.0 months versus 11.9 months in the chemotherapy and cetuximab arm, almost a threefold difference. No difference between the two arms has been observed in the MSS group. The effect of MSI-H is observed in a small subset but is corroborated by several lines of evidence. A beneficial effect of FOLFOX and bevacizumab compared with FOLFOX and placebo was reported in patients with MSI-H tumors (and, again, not in patients with MSS tumors) with stage II to III disease.⁸ This observation is important because the effect of bevacizumab is against placebo, pointing toward a predictive role of MSI-H. Patients with CMS1 tumors (enriched with MSI-H) show increased OS in the chemotherapy and bevacizumab arm over the chemotherapy and cetuximab arm.⁹ The contribution of resistance to EGFR inhibitors in patients with MSI-H tumors cannot be excluded, as suggested by a reduced efficacy of cetuximab in tumors originating in the right or transverse colon (also enriched with MSI-H) compared with those tumors originating in the left colon.^10,11 It is also interesting to observe that in CALGB/SWOG 80405, the effect of the chemotherapy and bevacizumab and cetuximab arm on OS is intermediate between the arms with the single biologics (Fig 3).

What is the biologic underpinning of the effect of MSI-H on the activity of biologics? We hypothesize a concurrent effect on both cetuximab (negative) and bevacizumab (positive). For the effect on cetuximab, it is well established that hypermethylation typical of MSI-H tumors results in lower expression of EGFR ligands,¹² reducing the efficacy of EGFR inhibitors.^13,14 For the effect of bevacizumab, vessel normalization induced by bevacizumab correlated with immunostimulatory pathways, especially Th1 lymphocyte infiltration and activity,¹⁵ possibly potentiating the antitumor effects of an already T-cell–infiltrated and activated microenvironment, such as that of MSI-H tumors.

This study represents one of the largest series on the negative prognostic effect of BRAF V600E in a single clinical trial in the first-line setting of mCRC. Although BRAF V600E is more prevalent in right and transverse (81%) versus left (19%) tumors, removal of tumor location as an adjusting covariable does not modify the effect of BRAF V600E HR by a great extent (HR from 2.01 to 2.26, Table 2), suggesting that BRAF V600E has an effect on survival that is independent of tumor location. By looking at the effect of treatment, despite a weak signal of improved OS in the chemotherapy and bevacizumab arm compared with the chemotherapy and cetuximab arm in patients with BRAF V600E tumors (HR, 0.67 [95% CI, 0.37 to 1.20]; P = .176; Appendix Fig A4; Table 2), median OS remains poor in both arms (15.0 and 11.7 months, respectively), as also reported in FIRE-3.¹⁶

The analysis of extended RAS clearly supports a negative prognostic effect. Patients with RAS mutant tumors have worse OS than do patients with RAS WT tumors, irrespective of treatment (Fig 2 [bottom]; Table 2). A detrimental effect of RAS mutations in patients enrolled in the chemotherapy and cetuximab arm compared with the chemotherapy and bevacizumab arm is more evident in the primary cohort for PFS (median, 9.2 v 11.4 months; P = .006) than for OS (median, 22.9 v 26.2 months; P = .855), probably because of the compensatory effects of postprogression therapies.²

One potential limitation of this article is that the mutational analysis was not available from all patients enrolled in the trial, particularly for TMB. This is a limitation for studies such as this one that use material from primary tumors retrospectively, because specimen collection might introduce bias.

In conclusion, molecular DNA analysis of tumor alterations in patients with mCRC can provide new tools to predict patient outcome and improve therapeutic decision making. Molecularly driven, novel immunotherapy-based combinations are urgently needed in patients with first-line MSS mCRC. In patients with MSI-H tumors, if the benefit of bevacizumab combined with chemotherapy is further confirmed, this regimen could be a preferable therapeutic option in patients with MSI-H, either as front-line therapy or in patients who are not responsive and/or suffer from clinical and financial toxicity of checkpoint inhibitors. Additional evaluation of the neoantigen-specific T-cell responses associated with TMB is needed to identify new targets and pathways and to guide the testing of targeted interventions. TMB can be obtained readily using a reliable and simple US Food and Drug Administration–approved diagnostic test that is often used in the clinic.

The proportional hazard assumption was examined by the Kolmogorov-type supremum test and graphical method using weighted residuals (Grambsch PM, Therneau TM: Biometrika 81:515-526, 1994). On the basis of the Kolmogorov-type supremum test, microsatellite instability (MSI) status and BRAF status violated the proportional hazard assumption. After closer examination of the weighted residual figures (Fig A5), it seems that the nonproportionality for MSI status is probably inconsequential because the confidence band for the time-varying coefficient covers 0 at all time and we reported a nonsignificant prognostic effect (Fig A5A). For BRAF, it does seem that there is a slightly diminished effect over time (Fig A5A and B). The negative prognostic effect is strong for the first 2 years after registration but then diminishes toward the null after 2 years. Given that the hazard ratio (HR) is still the most widely accepted method of reporting time-to-event end points and the fact that the effect never crosses 0 (ie, maintaining a negative prognostic effect), we reported the HR for BRAF in this article. When nonproportionality exists, the HR is simply an average of the time-varying effect over time; therefore, the HR may be pulled toward the null. Even so, because BRAF is a known negative prognostic factor, the HR has still a big effect size. Therefore, we decided to report the HR because it is easier to understand than a time-varying coefficient.