Association of KRAS p.G13D Mutation With Outcome in Patients With Chemotherapy-Refractory Metastatic Colorectal Cancer Treated With Cetuximab FREE

Recent retrospective correlative analyses of metastatic colorectal cancer trials indicate that patients with KRAS-mutated tumors (NCBI Entrez Gene 3845) do not benefit from the anti–epidermal growth factor receptor (EGFR) monoclonal antibodies cetuximab and panitumumab.¹ These retrospective analyses were performed independently, and for each analysis, KRAS wild-type vs mutant were studied grouping codons 12 and 13 mutations together, without subgroup analysis. Health authorities in the United States and Europe have indicated that patients with KRAS codon 12– or KRAS codon 13–mutated tumors should not receive cetuximab or panitumumab.^2- 4

However, indications exist that not all KRAS mutations are equal in their biological characteristics. First, the pattern of KRAS mutations is tumor-type specific. Although colorectal tumors have both codon 12 and codon 13 mutations (79% and 17.6%, respectively), KRAS-mutated pancreatic tumors (75%-95%) almost invariably carry codon 12 mutations, and in non–small cell lung cancer, more than 90% of KRAS mutations are located in codon 12.⁵ Second, anecdotal reports indicate that a minority of patients (<10%) with KRAS-mutated tumors can respond to anti-EGFR therapy^6- 9 and that about 15% have long-term disease stabilization.¹⁰ In these patients' tumors, codon 13 mutations were overrepresented compared with the overall KRAS-mutated tumor population. Finally, KRAS codon 13 mutations exhibit weaker in vitro transforming activity than codon 12 mutations.¹¹

Based on these observations, we hypothesized that KRAS codon 13 mutations may be associated with a better outcome after cetuximab treatment than observed with other KRAS mutations. Because the glycine (G)-to-aspartate (D) transition mutation is the most frequent codon 13 mutation in colorectal cancer,⁵ we studied the association of this p.G13D mutation with outcome after cetuximab treatment in a pooled data set of 579 patients with chemotherapy-refractory metastatic colorectal cancer treated with cetuximab with or without chemotherapy.

All patients included had chemotherapy-refractory metastatic colorectal cancer. The National Cancer Institute of Canada Clinical Trials Group/Australasian Gastrointestinal Trials Group (NCIC CTG/AGITG) data set is from the CO.17 randomized phase 3 trial of cetuximab plus best supportive care vs best supportive care alone.¹²KRAS mutation analysis was performed by direct sequencing as described previously.¹³

All centers participating in the NCIC CTG CO.17 trial received approval from their local research ethics body for participation in the trial, including for collection of tissue for correlative biomarker analyses, and all NCIC CTG/AGITG patients provided written informed consent for the original clinical trial. The Leuven data set comprises patients from 4 clinical trials (EVEREST, BOND, SALVAGE, and BABEL).^14- 16KRAS mutation analysis was performed by allelic discrimination assay as described previously.¹⁰ The Italian data set comprises patients treated at the Ospedale Niguarda Ca’Granda in Milan, Italy, and the Ospedale San Giovanni in Bellinzona, Cantone Ticino, Switzerland. Patients were included in 3 clinical trials (BOND, MABEL, and EMR202600)^15,17- 18 or were considered suitable to receive a subsequent line of treatment off study (after becoming refractory to the standard chemotherapy lines). KRAS mutation analysis was performed by direct sequencing as described previously.^6- 8

All patients in the Leuven and Italian data sets provided written informed consent to the original clinical trial and to molecular analyses. The Italian data set had a significantly higher percentage of patients with KRAS wild-type tumors than the other data sets (P < .001) because of the early introduction of KRAS testing in Italy, leading to exclusion of patients with KRAS-mutated tumors from treatment with cetuximab. A detailed breakdown accounting for the number of participants with each KRAS mutation type and patient characteristics is presented for each data set in the eTable.

The main efficacy end point was overall survival, defined as time from randomization in the NCIC CTG/AGITG data set, from start of cetuximab in the Leuven data set, and from start of cetuximab or date of randomization in the Italian data set to death due to any cause or to last known date alive. Secondary efficacy end points were response rate and progression-free survival. Tumor response was evaluated by radiologic imaging every 8 weeks in the NCIC CTG/AGITG data set and every 6 weeks until week 24, 30, or 36 and from then on every 12 weeks in the Leuven data set and for trial patients in the Italian data set. Response Evaluation Criteria in Solid Tumors were used to classify tumor response in all data sets. Progression-free survival was defined as the time from randomization or start of cetuximab to disease progression or death due to any cause. If a patient had not progressed or died at the time of data cutoff, progression-free survival was censored on the date of last disease assessment (NCIC CTG/AGITG data set) or last radiologic assessment (Leuven and Italian data sets).

Differences in response rates by KRAS status (p.G13D mutant, other KRAS mutant, or KRAS wild-type) were evaluated pairwise using the Fisher exact test. Median overall and progression-free survival were estimated by the Kaplan-Meier method and compared between groups using the log-rank test. Hazard ratios (HRs) between groups were estimated from Cox regression models including the following potential prognostic factors: age (<65 vs ≥65 years), sex (male vs female), performance status (Eastern Cooperative Oncology Group performance score, 0-1 vs 2), whether all 3 chemotherapy drugs (fluoropyrimidine [fluorouracil or capecitabine], irinotecan, and oxaliplatin) were previously received (yes vs no), site of primary malignancy (rectum only vs colon), data set (NCIC CTG/AGITG vs Leuven vs Italian), and an interaction term between KRAS status and treatment group as indicated.

For the predictive analysis (association between mutation status and outcome after cetuximab treatment), an additional analysis using the CO.17 database only was also undertaken. This was performed as a sensitivity analysis (avoiding the potential bias associated with cross-trial comparisons) rather than as the primary analysis because of the smaller number of patients (n = 195). All P values were 2-sided and statistical significance was assumed for P < .05. No adjustments were made for multiple comparisons. Analyses were carried out using SAS software, version 9 (SAS Institute Inc, Cary, North Carolina).

Details about the generation of KRAS-mutated isogenic cells, in vitro and in vivo treatment with cetuximab, proliferation assays, and biochemical profiling of signaling pathways are provided in the eAppendix.

Baseline patient characteristics in the 3 data sets by KRAS mutation status demonstrated no significant differences (Table 1). The mutation frequencies (40% KRAS-mutated, of which 14.5% were p.G13D-mutated) and distribution in this pooled data set (eTable) are similar to those reported for randomized trial populations (between 36% and 43% KRAS-mutated, of which 15.7% were p.G13D-mutated).^19- 22 The objective response rate in unselected patients was 10.2% in the monotherapy group and 22.2% in the cetuximab plus chemotherapy group. The median overall and progression-free survival were, respectively, 7.2 (95% confidence interval [CI], 6.2-8.3) months and 2.3 (95% CI, 1.9-2.9) months in the monotherapy group and 9.2 (95% CI, 8.7-10.1) months and 4.1 (95% CI, 3.9-4.8) months in the cetuximab plus chemotherapy group, comparable with outcomes from randomized trials.^12,15

In patients treated with best supportive care only in the NCIC CTG/AGITG CO.17 trial (n = 195), those with p.G13D-mutated tumors (n = 13) had significantly worse overall survival (median, 3.6 [95% CI, 2.2-4.8] months) than those with other KRAS-mutated tumors (median, 4.7 [95% CI, 3.6-6.7] months; HR, 1.90; 95% CI, 1.03-3.51; P = .04) or wild-type KRAS tumors (median, 5.0 [95% CI, 4.2-5.5] months; HR, 1.90; 95% CI, 1.05-3.41; P = .03) in univariate analysis (Table 2). In the multivariate analysis adjusting for potential prognostic factors, this difference became nonsignificant for the comparison between patients with p.G13D-mutated and other KRAS- mutated tumors (adjusted HR, 1.39; 95% CI, 0.73-2.64; P = .33) and between p.G13D-mutated and KRAS wild-type tumors (adjusted HR, 1.82; 95% CI, 0.99-3.34; P = .053). No significant difference in progression-free survival was seen for the 3 KRAS groups (Table 2). No analysis was performed for response rate because no patient responded under best supportive care.

Among patients who received any cetuximab-based treatment (cetuximab monotherapy or cetuximab plus chemotherapy) (n = 571), overall and progression-free survival were significantly longer in patients with p.G13D-mutated tumors (overall survival: n = 32; median, 7.6 [95% CI, 5.7-20.5] months; progression-free survival, n = 32; median, 4.0 [95% CI, 1.9-6.2] months) than in patients with other KRAS-mutated tumors (overall survival: median, 5.7 [95% CI, 4.9-6.8] months; progression-free survival: median, 1.9 [95% CI, 1.8-2.8] months) in both univariate analysis (overall survival: HR, 0.52; 95% CI, 0.33-0.80; P = .003; progression-free survival: HR, 0.54; 95% CI, 0.36-0.81; P = .02) and multivariate analysis adjusting for potential prognostic factors and data set (overall survival: HR, 0.50; 95% CI, 0.31-0.81; P = .005; progression-free survival: HR, 0.51; 95% CI, 0.32-0.81; P = .004) (Table 3). No significant difference in overall or progression-free survival was found between patients with p.G13D-mutated and KRAS wild-type tumors in either univariate analysis (P = .98 and P = .97, respectively) or multivariate analysis (P = .79 and P = .66, respectively) (Table 3). Response rate was not significantly different between patients with p.G13D-mutated and other KRAS-mutated tumors (2/32 [6.3%; 95% CI, 0%-14.6%] vs 3/188 [1.6%; 95% CI, 0%-3.3%], respectively; P = .15), but patients with KRAS wild-type tumors had a significantly higher response rate than patients with p.G13D-mutated tumors (91/345 [26.4%; 95% CI, 21.7%-31.0%] vs 2/32 [6.3%; 95% CI, 0%-14.6%], respectively; P = .02) (Table 4).

Subgroup analyses of outcome according to treatment (cetuximab monotherapy or cetuximab plus chemotherapy) are presented in Table 3. In the cetuximab plus chemotherapy subgroup, a statistically significant association between the p.G13D mutation and better outcome after cetuximab treatment was observed, unlike in patients with other KRAS-mutated tumors. Patients with p.G13D-mutated tumors (n = 22), compared with those with other KRAS-mutated tumors, had significantly longer overall survival (median, 10.6 [95% CI, 5.7-24.6] months vs 7.4 [95% CI, 5.5-9.0] months; adjusted HR, 0.46; 95% CI, 0.24-0.86; P = .02), longer progression-free survival (median, 4.1 [95% CI, 2.8-6.9] months vs 2.8 [95% CI, 2.5-3.7] months; adjusted HR, 0.49; 95% CI, 0.28-0.86; P = .01) (Table 3), and higher response rate (2/22 [9.1%; 95% CI, 0%-21.1%] vs 1/99 [1.0%; 95% CI, 0%-3.0%]; P = .08). No significant difference in either overall or progression-free survival was found between patients with p.G13D-mutated and KRAS wild-type tumors in either of these 2 subgroups (P = .41 and P > .99 for overall survival and P = .29 and P = .31 for progression-free survival in the cetuximab monotherapy and cetuximab plus chemotherapy groups, respectively) (Table 3).

However, patients with KRAS wild-type tumors receiving cetuximab with chemotherapy had a significantly higher response rate than patients with p.G13D-mutated tumors (68/199 [34.2%; 95% CI, 27.6%-40.8%] vs 2/22 [9.1%; 95% CI, 0%-21.1%], respectively; P = .03). Figure 1 and Figure 2 show predictive effects by KRAS status. Among patients who received any cetuximab-based treatment, those with p.G13D-mutated tumors had a longer median overall survival (7.6 [95% CI, 5.7-20.5] months) and progression-free survival (4.0 [95% CI, 1.9-6.2] months) than those receiving best supportive care alone (3.6 [95% CI, 2.2-4.8] months and 1.7 [95% CI, 1.5-1.7] months, respectively), which were significant in univariate analysis (overall survival: HR, 0.24; 95% CI, 0.12-0.51; P < .001; progression-free survival: HR, 0.39; 95% CI, 0.20-0.78; P = .006) but became nonsignificant in multivariate analysis (overall survival: adjusted HR, 0.40; 95% CI, 0.13-1.28; P = .12; progression-free survival: HR, 0.53; 95% CI, 0.16-1.73; P = .29).

For patients with other KRAS mutations, no significant difference in overall survival was found between any and no cetuximab treatment (median, 5.7 [95% CI, 4.9-6.8] months vs 4.7 [95% CI, 3.6-6.7] months; adjusted HR, 1.07; 95% CI, 0.74-1.60; P = .71), while a significant difference was found in progression-free survival (median, 1.9 [95% CI, 1.8-2.8] months vs 1.8 [95% CI, 1.7-1.9] months) in univariate analysis (HR, 0.69; 95% CI, 0.52-0.92; P = .01) but not in multivariate analysis (adjusted HR, 0.93; 95% CI, 0.71-1.39; P = .96). As expected, patients with KRAS wild-type tumors receiving any cetuximab-based treatment, compared with patients receiving best supportive care alone, had significantly longer median overall survival (10.1 [95% CI, 9.4-11.3] months vs 5.0 [95% CI, 4.2-5.5] months, respectively; adjusted HR, 0.60; 95% CI, 0.44-0.81; P < .001) and progression-free survival (4.2 [95% CI, 3.9-5.4] months vs 1.9 [95% CI, 1.8-2.0] months; adjusted HR, 0.42; 95% CI, 0.32-0.56; P < .001).

We performed a test for interaction between any vs no cetuximab treatment and p.G13D vs other KRAS mutations. The adjusted P value for the interaction was P = .003 (HR, 0.30; 95% CI, 0.14-0.67) for overall survival and P = .05 (HR, 0.47; 95% CI, 0.22-1.00) for progression-free survival. There was no interaction between p.G13D mutation vs wild-type KRAS status and overall survival benefit from cetuximab-based treatment (any vs none) (HR, 0.49; 95% CI, 0.23-1.04; P = .06).

In univariate analysis, patients with p.G13D-mutated tumors had a significantly longer overall survival compared with patients receiving best supportive care (any cetuximab: HR, 0.24; 95% CI, 0.11-0.50; P < .001; cetuximab monotherapy: HR, 0.34; 95% CI, 0.13-0.87; P = .02). Similarly, progression-free survival was superior in p.G13D patients receiving any cetuximab (HR, 0.39; 95% CI, 0.19-0.78; P = .006) and cetuximab monotherapy (HR, 0.38; 95% CI, 0.15-0.98; P = .04) compared with patients receiving best supportive care alone. Although the differences were no longer statistically significant in multivariate analysis, the significant P value for interaction confirms that in this data set, the p.G13D mutation was associated with significantly greater overall survival benefit than tumors expressing other KRAS mutations (Figure 2 and Figure 3).

For patients with other KRAS-mutated tumors, progression-free survival was significantly longer when receiving cetuximab with chemotherapy than when receiving best supportive care alone (median, 2.8 [95% CI, 2.5-3.7] months vs 1.8 [95% CI, 1.7-1.9] months; adjusted HR, 0.53; 95% CI, 0.36-0.79; P = .002) but not when receiving cetuximab as monotherapy (Figure 2), suggesting some retained chemosensitivity in these patients. Regardless of whether cetuximab was received with or without chemotherapy, patients with KRAS wild-type tumors, compared with those receiving best supportive care alone, had significantly longer median survival (cetuximab plus chemotherapy: median, 11.3 [95% CI, 9.9-13.6] months vs 5.0 [95% CI, 4.2-5.5] months; adjusted HR, 0.44; 95% CI, 0.32-0.62; P < .001; cetuximab monotherapy: median, 9.4 [95% CI, 7.7-10.3] months vs 5.0 [95% CI, 4.2-5.5] months; adjusted HR, 0.58; 95% CI, 0.42-0.78; P < .001) and progression-free survival (cetuximab plus chemotherapy: median, 5.5 [95% CI, 4.2-5.5] months vs 1.9 [95% CI, 1.8-2.0] months; adjusted HR, 0.22; 95% CI, 0.16-0.31; P < .001; cetuximab monotherapy: median, 3.7 [95% CI, 2.8-4.1] months vs 1.9 [95% CI, 1.8-2.0] months; adjusted HR, 0.39; 95% CI, 0.29-0.53; P < .001) (Figure 2).

A separate analysis of the CO.17 trial (n = 195), containing the only randomized patients (cetuximab plus best supportive care vs best supportive care alone) in the pooled data set, was performed as a sensitivity analysis to avoid the potential bias associated with cross-trial comparisons and to allow the purest assessment of the impact of KRAS mutations on the effect of cetuximab.

Within the p.G13D mutation subset (n = 13), the adjusted HR for overall survival for cetuximab therapy compared with best supportive care alone was in the same direction as in the cross-trial comparison, although the P value was not significant (adjusted HR, 0.61; 95% CI, 0.17-2.19; P = .45). There was no benefit for cetuximab therapy in the other KRAS mutations subset (adjusted HR, 0.99; 95% CI, 0.67-1.44; P = .94). There was a significant interaction between p.G13D mutation status (p.G13D vs other KRAS mutations) and overall survival benefit from therapy (cetuximab vs best supportive care) in the univariate analysis (HR, 0.33; 95% CI, 0.11-1.00; P = .05), which was not significant in the multivariate analysis (adjusted HR, 0.43; 95% CI, 0.14-1.34; P = .15) (Figure 4).

We introduced p.G12V and p.G13D alleles in the genome of human colorectal SW48 cells by targeted homologous recombination (eAppendix).²³ While the proliferation of p.G12V-mutated SW48 cells was unaffected by cetuximab, the isogenic p.G13D-mutated cells displayed a drug response similar to their wild-type counterpart (eFigure 1, A). Importantly, the proliferative capabilities of parental and KRAS-mutated cells were undistinguishable (eFigure 1, B). Cetuximab administration prominently inhibited the growth of tumors formed by wild-type or KRAS p.G13D mutant cells grown as xenografts in immunocompromised mice (eFigure 1, C). In contrast, the growth of tumors formed by the KRAS p.G12V cells was not significantly affected by cetuximab treatment (eFigure 1, C). We then measured the level of activation (phosphorylation status) of EGFR and its downstream effectors (mitogen-activated protein kinase/extracellular signal-regulated kinase and v-akt murine thymoma viral oncogene homolog) in KRAS wild-type and mutant SW48 cells. In the presence of cetuximab, the p.G12V-mutated cells seemingly could still activate the extracellular signal-regulated kinase pathway but the p.G13D-mutated cells could not (eFigure 2). Of note, the levels of activated KRAS (guanosine-5′-triphosphate-bound) were similar in p.G12V- and p.G13D-mutated cells (eFigure 3). Overall, these results indicate that the KRAS p.G12V and p.G13D mutations differently affect response to cetuximab in preclinical models.

In a large, retrospective pooled exploratory analysis of patients with chemotherapy-refractory colorectal cancer, we show for the first time that there is a positive association between KRAS p.G13D mutations and cetuximab treatment in regard to better overall and progression-free survival.

The improved survival observed in patients with p.G13D-mutated tumors in the cetuximab monotherapy group suggests that p.G13D-mutated tumors may be sensitive to cetuximab and precludes a chemotherapy-driven effect. Patients with p.G13D-mutated tumors treated with combination regimens also have significantly better overall survival than do those with other KRAS-mutated tumors, which mirrors the observation in monotherapy-treated patients, suggesting cetuximab-dependent effects, although it cannot be excluded that chemotherapy is a confounding factor in patients treated with cetuximab plus chemotherapy.

In the monotherapy group, the difference in overall and progression-free survival between patients with p.G13D-mutated tumors and those with other KRAS-mutated tumors was not statistically significant. However, the magnitude of the effect was comparable and the direction was the same as in the patients treated with cetuximab plus chemotherapy.

Because this is a pooled analysis, to reduce the risk of biases implicit in this kind of study, we adjusted for type of previous treatment (whether all 3 chemotherapy drugs [fluoropyrimidine, irinotecan, and oxaliplatin] were previously received) and data set in the multivariate analyses. However, in the absence of randomization, there may be inadequate controlling for unknown confounders. For the predictive analysis and the estimation of treatment effect of cetuximab over no cetuximab, the comparator group for all patients from the pooled data set was the best supportive care group from the CO.17 trial. The sensitivity analysis from the CO.17 trial provided an unbiased estimate that was consistent with the finding of the pooled analysis, although it was not statistically significant.

To study the association of the p.G13D mutation with outcome in metastatic colorectal cancer, we compared overall and progression-free survival between the different KRAS mutation groups in the 195 patients in the CO.17 trial randomized to best supportive care alone. In this subset, the 13 patients with p.G13D-mutated tumors had a worse overall survival than those with KRAS wild-type tumors and those with tumors bearing other KRAS mutations, in univariate but not in multivariate analysis. Of particular relevance, patients with p.G13D-mutated tumors in our series also seemed to benefit more from cetuximab treatment than those with KRAS wild-type tumors, suggesting that the poor prognosis of a p.G13D mutation is mitigated by cetuximab treatment. Given the relatively small number of patients with p.G13D-mutated tumors, caution in drawing conclusions is warranted. Comparing overall survival between patients with p.G13D-mutated vs other KRAS-mutated and KRAS wild-type colorectal cancers in the control groups of randomized trials will contribute to determining whether this is a true association.

Although p.G13D-mutated tumors do not behave like other KRAS-mutated tumors, they appear to behave somewhat differently than KRAS wild-type tumors. Our results indicate that patients with p.G13D-mutated tumors respond to cetuximab therapy, albeit with a lower response rate than those with KRAS wild-type tumors.

The prolonged progression-free and overall survival of patients with p.G13D-mutated tumors in comparison with those with other KRAS-mutated tumors may not be due to a real reduction in tumor burden but to a delay in progression. A possible explanation of this clinical observation is that p.G13D mutant tumors do not undergo apoptosis (cytotoxic effect) on EGFR inhibition, but proliferation is inhibited (cytostatic effect).

When assessing the effect of cetuximab treatment on cellular proliferation in SW48 isogenic clones carrying p.G12V or p.G13D mutations, we found that while p.G12V-mutated cells were insensitive to cetuximab, p.G13-mutated cells were nearly as responsive to cetuximab as wild-type cells. These results provide a cell-based molecular explanation to our clinical observation that patients with p.G13D-mutated tumors benefit from cetuximab treatment, while those with other KRAS-mutated tumors do not.

Our study is limited because it is a retrospective observational study that relies largely on nonrandomized or cross-trial comparisons for discussion of possible treatment effects. It can therefore only suggest an association between p.G13D mutation status and survival benefit after cetuximab-based treatment. The results from the patient sample analyses are supported by isogenic cellular models, wherein the different effects of the G12 and G13 KRAS alleles on response to cetuximab are evident.

In conclusion, our study retrospectively observed an association between the presence of a p.G13D mutation and survival benefit in chemotherapy-refractory metastatic colorectal cancer treated with cetuximab. Prospective randomized trials are needed before conclusions about potential beneficial effects of cetuximab in p.G13D-mutated chemotherapy-refractory metastatic colorectal cancer should be inferred.