Purpose: To investigate whether EGF receptor (EGFR) pathway mutations predicted response to monotherapy with panitumumab, an anti-EGFR monoclonal antibody, in a randomized phase III study of metastatic colorectal cancer.
Experimental Design: Using massively parallel multigene sequencing, we analyzed 320 samples for 9 genes, with multigene sequence data from 288 (90%) samples.
Results: Mutation rates were: KRAS (45%), NRAS (5%), BRAF (7%), PIK3CA (9%), PTEN (6%), TP53 (60%), EGFR (1%), AKT1 (<1%), and CTNNB1 (2%). In the randomized study and open-label extension, 22 of 138 (16%) wild-type KRAS (codons 12/13/61) patients versus 0 of 103 mutant KRAS (codons 12/13) patients had objective responses. Of 6 mutant KRAS (codon 61) patients, 1 with a Q61H mutation achieved partial response during the extension. Among wild-type KRAS (codons 12/13/61) patients, 0 of 9 patients with NRAS mutations, 0 of 13 with BRAF mutations, 2 of 10 with PIK3CA mutations, 1 of 9 with PTEN mutations, and 1 of 2 with CTNNB1 mutations responded to panitumumab. No patients responded to best supportive care alone. Panitumumab treatment was associated with longer progression-free survival (PFS) among wild-type KRAS (codons 12/13/61) patients [HR, 0.39; 95% confidence interval (CI), 0.28–0.56]. Among wild-type KRAS patients, a treatment effect for PFS favoring panitumumab occurred in patients with wild-type NRAS (HR, 0.39; 95% CI, 0.27–0.56) and wild-type BRAF (HR, 0.37; 95% CI, 0.24–0.55) but not mutant NRAS (HR, 1.94; 95% CI, 0.44–8.44).
Conclusions: These results show the feasibility and potential clinical use of next-generation sequencing for evaluating predictive biomarkers. Clin Cancer Res; 19(7); 1902–12. ©2012 AACR.
The anti–EGF receptor (EGFR) monoclonal antibodies panitumumab and cetuximab are recommended for the treatment of wild-type KRAS metastatic colorectal cancer (mCRC), but not all patients with wild-type KRAS mCRC benefit from their treatment. Other genes have been suggested as potential predictors of outcome in patients treated with anti-EGFR monoclonal antibodies for mCRC. Massively parallel multigene sequencing was used to analyze mutations in KRAS, NRAS, EGFR, BRAF, PTEN, PIK3CA, AKT, TP53, and CTNNB1 in tumor samples from patients who received panitumumab in a randomized phase III study or its open-label extension study. Among wild-type KRAS patients, a treatment effect for progression-free survival favoring panitumumab occurred in patients with wild-type NRAS and wild-type BRAF but not mutant NRAS. The results of this study, which was intended to generate hypotheses for future prospective studies or meta-analyses, show the feasibility and potential clinical use of next-generation sequencing for evaluating predictive biomarkers.
Improvements in progression-free survival (PFS) or objective response rate (ORR) following treatment with the anti–EGF receptor (EGFR) antibodies panitumumab and cetuximab are confined to patients with wild-type KRAS tumors as determined by allele-specific real-time PCR (1–6). Current treatment guidelines recommend anti-EGFR monoclonal antibodies for use only in patients with wild-type KRAS (codons 12 and 13) colorectal tumors, which are associated with lack of response to these agents (7). However, not all patients with wild-type KRAS metastatic colorectal cancer (mCRC) benefit from anti-EGFR antibodies. In addition to KRAS mutations, multiple other genes are known to be somatically mutated in mCRC and have been assessed as biomarkers of response to anti-EGFR therapy (8–13). Recent studies have suggested that mutations in BRAF (8, 9, 13), as well as mutations in PI3KCA and loss of PTEN expression (8, 9, 11–13), may be associated with poor outcome in patients with mCRC treated with agents targeting the EGFR. In a retrospective analysis of tumor samples from patients with mCRC treated with cetuximab plus chemotherapy, mutations in BRAF, NRAS, and PIK3CA exon 20 were associated with low response rates in patients with wild-type KRAS tumors (14). Conversely, mutations in the TP53 tumor suppressor gene have been associated with favorable disease control and time to progression among patients treated with cetuximab plus chemotherapy, including patients with wild-type KRAS tumors (10). On the basis of these observations, the strongest hypothesis was that somatic mutations in the RAS genes (KRAS and NRAS) beyond the KRAS codons 12 and 13 would be negatively predictive of response to panitumumab. In addition, based on the existing literature, we predicted that activating mutations in BRAF and PIK3CA and inactivating mutations in PTEN would be unfavorable in the context of anti-EGFR therapy, whereas TP53 mutations would result in a favorable outcome.
We used massively parallel multigene sequencing techniques (also known as next-generation sequencing) to analyze 3 types of alterations in parallel: additional RAS-activating mutations (KRAS codon 61; NRAS codons 12/13/61), other EGFR signaling pathway genes known to be mutated in mCRC (EGFR, BRAF, PTEN, PIK3CA, and AKT), and genes known to play a role in colorectal cancer (CRC) tumorigenesis and progression (TP53 and CTNNB1). Massively parallel sequencing technology enables rapid and sensitive sequencing, thus allowing the simultaneous identification of mutations from multiple patient samples in multiple genes potentially important for tumor development (15, 16). The assessment was conducted on banked tissue samples [formalin-fixed paraffin-embedded (FFPE)] that were previously assessed for KRAS codons 12 and 13 from patients with mCRC enrolled in a randomized phase III study (the 408 study; ref. 17) and an open-label extension study (the 194 study; ref. 18).
Patients and Methods
Samples were from patients with metastatic colorectal adenocarcinoma enrolled in the randomized multicenter phase III 408 study (Clinicaltrials.gov, NCT00113763; ref. 17). Patients in the best supportive care (BSC) arm of the randomized phase III study who had radiographically documented disease progression could enroll in the open-label extension 194 study (Clinicaltrials.gov, NCT00113776; ref. 18). The study protocol was approved by independent ethics committees at each participating center. Patients provided written informed consent.
Study design and treatment
Patients enrolled in the randomized phase III study (17) from January 2004 to June 2005 were randomly assigned 1:1 to receive 2-week cycles of intravenous panitumumab 6 mg/kg every 2 weeks plus BSC or BSC alone. Median follow-up time was 35 weeks. Randomization was stratified by Eastern Cooperative Oncology Group (ECOG) performance status (0/1 vs. 2) and region [Western Europe vs. Central/Eastern Europe vs. rest of the world (Canada, Australia, and New Zealand)]. The primary endpoint was PFS. Secondary endpoints included ORR per modified Response Evaluation Criteria in Solid Tumors (RECIST) version 1.0 (19) and overall survival (OS). Local and central reviews were conducted for all assessments.
Patients from the BSC arm of the randomized phase III study who had disease progression could enroll in the cross-over extension study (18) and receive intravenous panitumumab 6 mg/kg every 2 weeks plus BSC. In February 2006, median follow-up time was 61 weeks. The primary endpoint was safety. Exploratory endpoints included PFS, ORR per modified RECIST version 1.0 (19), time to response, duration of response, duration of stable disease, and OS, all per local review.
Archival CRC samples (all obtained from the patients before trial initiation) were previously analyzed for KRAS mutations (codons 12 and 13) using allele-specific PCR (DxS/Qiagen; ref. 1). For this analysis, tumor DNA was extracted from additional archival FFPE tissue sections using the QIAmp DNA FFPE Tissue Kit (Qiagen). One specimen per patient was analyzed. DNA isolates were quantified using the Quant-iT PicoGreen dsDNA reagent (Invitrogen). Slides representing the primary tumors of 324 patients were available for analysis. Two slides were available in most cases (n = 308), but only a single slide was available for 11 patients, and 3 or more slides were available for 5 patients. Sequencing libraries were generated and used in 454 GS FLX (Roche 454 Life Sciences) sequencing experiments. The PCR primers in this analysis targeted exons 2 and 3 of NRAS; exon 3 of KRAS; exon 2 of CTNNB1; exon 15 of BRAF; exon 3 of AKT1; exons 1, 2, 9, 10, and 20 of PIK3CA; exons 17 through 22 of EGFR; and all exons of PTEN and TP53 (primer pair sequences are presented in Supplementary Table S1).
Up to 24 patient samples were interrogated per sequencing experiment. Amplified sequencing libraries were evaluated by capillary electrophoresis using the Agilent 2100 Bioanalyzer and DNA1000 Kit (Agilent Technologies, Inc.) before sequencing. Libraries showing little or no amplification of target amplicons, and/or a molar excess of nonspecific PCR artifact relative to target amplicons (again using the Agilent 2100 Bioanalyzer and DNA1000 Kit), were not used for sequencing. Sequence data were base-called and filtered using default 454 GS FLX software settings and further filtered to reject sequences that did not contain 2 complete cognate locus-specific primer sequences (i.e., the forward and reverse primer sequences for 1 of the intended PCR amplicons). The latter filter prevented sequences arising from PCR artifacts being included in data used for mutation analysis.
Mutation analysis was carried out using the 454 Amplicon Variant Analysis (AVA) software version 2.0 (Roche 454 Life Sciences). Gene–subject combinations in which a mutation was observed were assigned mutant status, and wild-type status was assigned to gene–subject combinations in which (i) no mutations were observed, and (ii) the median number of sequence reads per amplicon for that gene in that subject was more than 40. If no mutation was observed but this second criterion was not met, the status of that gene–subject intersection was classified as “not determined.” Some FFPE DNA samples, when amplified by PCR, exhibit a sharply elevated frequency of artifactual transition substitutions. To avoid making false-positive mutation identifications in these cases, a metric of the severity of this phenomenon was derived by counting the number of nonreference bases present at each of 2,889 nucleotide positions within the panel in which a substitution would not alter the encoded polypeptide or disrupt a splice donor or acceptor sequence and for which there was no reported single-nucleotide polymorphism. Subsequently, a sample-specific cutoff for acceptance of putative missense or nonsense sequence variants was applied such that any variant present in a smaller fraction of sequence reads than the third most frequent “silent position” in a given sample was not included (i.e., variants with approximately >0.1% probability of being an artifact). The sample-specific cutoff was generally less than 5%.
The analysts conducting the assays were blinded to clinical outcome data. The multigene sequencing assays were extensively qualified to show the ability to confidently detect mutations at a level of 5% of sequence reads per amplicon using both cell line DNA and procured FFPE colorectal tumor samples. Because of the nature of this qualification, the demonstration of correlation between this sequencing methodology and bidirectional sequencing (20), the lack of commercially available validated assays for all of these mutations and the limiting amount of material available for these samples, confirmation of the mutation status using an additional method was not pursued.
The analysis included all intent-to-treat patients with available biomarker information. The statistical analysis plan was developed before analyses were conducted. As specified in the analysis plan, patient data were analyzed on the basis of randomized treatment assignment (regardless of actual treatment administered).
Assessment of gene frequency
Genotype frequency was calculated for each biomarker by treatment arm and by overall frequency.
Evaluation of treatment effect on PFS in the randomized study
In the randomized phase III study, the treatment effect on PFS (independent blinded radiologic review) of panitumumab plus BSC versus BSC for all evaluable patients and the wild-type KRAS subgroup was estimated by genotype using a Cox proportional hazards model adjusted for the randomization factors.
Evaluation of biomarker effect on PFS in the randomized study
The relative biomarker effect (mutant vs. wild-type) within each treatment arm (panitumumab plus BSC, BSC only, and total) was evaluated using a Cox proportional hazards model adjusted for the randomization factors.
Analysis of the randomized study and the extension studies combined
This analysis included patients enrolled in the panitumumab plus BSC arm of the randomized phase III study and in the open-label extension study. The analysis was conducted for each study and for both studies combined, for all evaluable patients, and within the wild-type KRAS subgroup. For the combined analyses, the patients enrolled in the extension study were considered equivalent to patients randomized to panitumumab in the phase III study. Because the extension study was reviewed locally, ORR and PFS based on local review were used for analyses of the randomized study and the extension study.
PFS was defined as time from the date of enrollment to disease progression or death. Patients who had not progressed or died during the study were censored at the last evaluable disease assessment date. For the panitumumab plus BSC arm in the randomized phase III study, the initial enrollment date was used. For patients in the extension study, enrollment date into the extension study was used. Because the key eligibility criteria for the extension study was disease progression in the randomized study, by definition, patients in the extension study already had one disease progression before entering the extension study.
ORR and 95% confidence intervals (CI) were calculated for each biomarker by genotype and by study: the panitumumab plus BSC arm in the randomized study, the extension study, and both combined. Cox proportional hazard models adjusted for the randomization factors ECOG score and demographic region were used to estimate the effect of mutant versus wild-type of each biomarker on PFS under the panitumumab plus BSC treatment. For these exploratory analyses, the reported P values are presented for descriptive purposes only. P values near or less than 0.05 are believed to show potential for further investigation (i.e., for hypothesis generation). No adjustments for multiple comparisons were made with the reported P values.
As previously reported, 463 patients were randomized to panitumumab 6 mg/kg every 2 weeks plus BSC (n = 231) or BSC alone (n = 232) in the phase III study (study 20020408); 176 patients (76%) from the BSC arm with progressive disease enrolled in the optional extension study (study 20030194) of panitumumab monotherapy (Fig. 1; refs. 17, 18). In the randomized study, PFS was improved with panitumumab compared with BSC alone (HR, 0.54; 95% CI, 0.44–0.66; P < 0.0001; ref. 17). The treatment effect of panitumumab was greater (P < 0.0001) among patients with wild-type KRAS codons 12 and 13 (HR, 0.45; 95% CI, 0.34–0.59) than among those with KRAS mutations (HR, 0.99; 95% CI, 0.73–1.36; ref. 1). The ORR among patients with wild-type KRAS (codons 12/13) who were randomized to panitumumab was 17% (95% CI, 11%–25%); no responses occurred among patients with KRAS mutations (1). No significant difference in OS was observed between the treatment arms in the randomized study (HR, 1.00; 95% CI, 0.82–1.22; P = 0.81) and KRAS status was not predictive for OS (1, 17), perhaps because 76% of patients from the BSC arm received panitumumab in the extension study.
Gene mutation analysis
Overall, 320 archival FFPE tumor samples were available from the 463 patients in the 20020408 study, 288 (288 of 320; 90%) of which (panitumumab, n = 147; BSC, n = 141; Fig. 1) provided information for multiple genes. Demographic and clinical characteristics for patients with or without multigene information were generally similar; patients from Western Europe were more likely to have available multigene information (Supplementary Table S2). Among the 141 patients randomized to the BSC arm with samples, 110 were enrolled in the panitumumab extension study. Forty-nine primer pairs were used to amplify 41 exons, resulting in 5.7 × 106 DNA sequence reads, comprising 1.26 × 109 base pairs. Estimated total tissue volume [area × (number of slides) × 0.005] used for DNA isolation ranged from less than 0.1 to more than 9 mm3 (mean, 3.1 mm3), and the quantity of DNA recovered ranged from 0.1 to 5.37 μg (mean, 1.19 μg). Consistent with studies that have shown amplification of KRAS sequences from small quantities of FFPE tissue (21), there was only a weak relationship between amount of tissue processed and DNA yield (R2 = 0.16).
Mutation rates and data completeness for each of the genes analyzed (KRAS, NRAS, EGFR, BRAF, PTEN, PIK3CA, AKT1, TP53, and CTNNB1) are shown in Table 1 and mutations identified for individual patients are shown in Supplementary Table S3. Overall, 124 of 277 (45%) patients with available data had a mutation in KRAS, among whom 117 had mutations in KRAS codons 12 or 13, and 7 had mutations in KRAS codon 61. Five percent of patients had mutations in NRAS and 7% had mutations in BRAF. Sequencing identified 109 tumors with mutations in more than 1 gene and 20 tumors with more than 1 mutation in a single gene. Three tumors were identified with mutations in KRAS (codons 12/13/61) and NRAS, and 2 tumors were identified with mutations in both KRAS (codons 12/13/61) and BRAF (Table 2). Most patients with 2 or more mutant genes had a TP53 mutation and a mutation at another locus (or 2 other loci). Dual mutations in genes other than TP53 were observed in several patients.
Association between gene mutations and PFS among patients with wild-type KRAS in the randomized study
The effect of panitumumab treatment on PFS (independent blinded radiologic review) for each genotype in patients with wild-type KRAS from the randomized study is shown in Fig. 2. Among patients with wild-type KRAS (codons 12/13/61) and wild-type NRAS (n = 138), treatment with panitumumab was associated with improved PFS (HR, 0.39; 95% CI, 0.27–0.56; P < 0.001), but in patients with wild-type KRAS (codons 12/13/61) and mutant NRAS (n = 11), treatment with panitumumab was not associated with longer PFS (HR, 1.94; 95% CI, 0.44–8.44; P = 0.379). Among wild-type KRAS patients, a Cox proportional hazards model with NRAS, treatment, and their interaction showed P = 0.076 for the interaction term, which was suggestive of an interaction. However, the presence of BRAF mutations among patients with wild-type KRAS did not seem to predict the effect of treatment with panitumumab on PFS. A favorable effect of panitumumab on PFS was observed in patients with wild-type BRAF (n = 115; HR, 0.37; 95% CI, 0.24–0.55; P < 0.001) and potentially in patients with BRAF mutations (n = 15; HR, 0.34; 95% CI, 0.09–1.24; P = 0.104). Similarly, a favorable effect of panitumumab on PFS was observed in patients with wild-type PTEN (n = 135; HR, 0.36; 95% CI, 0.25–0.52; P < 0.001) and in patients with wild-type PIK3CA (n = 128; HR, 0.39; 95% CI, 0.26–0.57; P < 0.001). Trends toward improved PFS among patients randomized to panitumumab were observed for patients with PTEN mutations (n = 9; HR, 0.11; 95% CI, 0.01–1.52; P = 0.100) and in patients with PIK3CA mutations (n = 10; HR, 0.15; 95% CI, 0.01–1.57; P = 0.114). Cox proportional hazard models similar to those described earlier for NRAS were conducted for BRAF, PTEN, and PIK3CA. In each case, the interaction terms did not suggest statistical significance (BRAF, P = 0.850; PTEN, P = 0.771; PIK3CA, P = 0.616).
Among patients with wild-type KRAS from the randomized study, the presence of BRAF mutations was associated with poor PFS regardless of treatment. Among patients with BRAF mutations, the HR for PFS for mutant versus wild-type status was 2.55 (95% CI, 1.04–6.24) in the BSC arm, 3.27 (95% CI, 1.52–7.01) in the panitumumab plus BSC arm, and 2.39 (95% CI, 1.36–4.21) overall (Fig. 3). No other gene mutation among patients with wild-type KRAS was negatively prognostic for PFS.
Association between gene mutations and objective response among panitumumab-treated patients with wild-type KRAS in the randomized and extension studies
The sample size was increased by examining the ORR in patients treated with panitumumab monotherapy in both the randomized and the extension studies. ORR was 16% for patients with wild-type KRAS (n = 138) versus 1% with mutant-type KRAS (codons 12/13/61; n = 109). One patient with a KRAS Q61H mutation had a partial response (local review) while receiving panitumumab in the extension study. Fifty-eight percent of this patient's KRAS codon 61 sequencing reads were Q61H. Furthermore, a mutation in TP53 was detected in this patient's sample; all other genes examined were wild-type.
ORRs for wild-type versus mutant genotypes among patients with wild-type KRAS (codons 12/13/61) who were randomized to panitumumab in the randomized study or the extension study are summarized in Table 3 and the Supplementary Figure. Among 126 patients with wild-type KRAS (codons 12/13/61) and NRAS, 22 achieved objective responses with an ORR of 17% (95% CI, 0.11–0.25). Among 98 patients with wild-type KRAS, NRAS, and BRAF, 18 achieved objective responses with an ORR of 18% (95% CI, 0.11–0.27). No responses to panitumumab occurred in patients with wild-type KRAS and mutations in NRAS or BRAF (0 of 22 patients; 95% CI, 0–0.15). No patients responded to BSC alone.
To our knowledge, this is the first use of next-generation sequencing to assess potential predictive biomarkers of response using tumor samples from a randomized phase III clinical trial. Previous reports described the use of massively parallel pyrosequencing techniques to investigate disease markers only in small patient populations (22–24).
In this retrospective analysis designed to explore effects of tumor genotype on outcomes, we were able to show trends and develop hypotheses; however, confidence of the observed trends was somewhat limited by the low prevalence of many of the gene mutations. For example, BRAF is mutated in approximately 10% of patients with mCRC. In this study, the treatment HR or effect size was calculated to be 0.34 (95% CI, 0.09–1.24), similar to that of wild-type KRAS. Using this HR, 46 specimens that are KRAS wild-type and BRAF mutant would be required to achieve 90% power with 2-sided type I error set to 0.05. For the KRAS wild-type/NRAS-mutant population, given an HR of 1.94 (95% CI, 0.44–8.44), we estimate that 129 patients with this genotype would be required to provide 90% power. Given these limitations, hypothesis testing of these low prevalence genes will likely require carefully constructed meta-analyses. The use of multiplexed, multigene analysis is a useful tool for hypothesis generation in translational studies.
The frequency of gene mutations observed in this study with 5% sensitivity is generally consistent with mutation rates for adenocarcinoma of the colon or rectum published in the Catalogue of Somatic Mutations in Cancer database (25) and with those reported by Vaughn and colleagues (26). The consistency of our results with previously reported studies suggests that the data are reliable.
Consistent with other reports of anti-EGFR antibodies in mCRC (1–3, 6, 14, 27), objective responses to panitumumab were confined to patients with wild-type KRAS tumors. Interestingly, 1 patient with a codon 61 mutation (Q61H) had a partial response per local review. Prior analyses of KRAS mutational status in anti-EGFR monoclonal antibody therapy have primarily assessed mutations only in codons 12 and 13 (1, 3, 6, 14, 27); however, 1 study reported that no patients with mutations in codon 61 (n = 8) responded to treatment with cetuximab plus irinotecan (28). We did not assess the predictive or prognostic value of individual codon 12/13 mutations but a recent pooled analysis of 3 randomized phase III studies of panitumumab in mCRC (which included patients from the randomized phase III 408 study), found that no individual mutant KRAS allele was consistently associated with PFS or OS outcomes (29). Conversely, pooled analyses of clinical trials in which cetuximab was administered as a component of first-line chemotherapy (30) or in chemotherapy-refractory disease (31) have suggested that patients with colorectal tumors bearing KRAS G13D mutations have favorable outcomes versus patients with other KRAS mutations.
Our data are consistent with the hypothesis that NRAS mutations, which occur infrequently (<5%) in mCRC (14, 25, 26, 32, 33), may limit the efficacy of panitumumab. In patients with wild-type KRAS and mutant NRAS tumors assigned to panitumumab therapy in either the randomized or extension study (n = 9), there was a lack of response (Table 3) and a lack of improved PFS (HR for mutant vs. wild-type, 1.71; 95% CI, 0.81–3.62). As described earlier, because mutations in NRAS are low-prevalence mutations, their true predictive or prognostic value must be confirmed in larger studies.
The negative prognostic effect of BRAF is clearly seen for PFS (Fig. 3). Because no patients responded in the BSC arm of the randomized study, evaluation of its prognostic potential for ORR is difficult. However, the negative effect of BRAF mutation was clearly seen in the treatment arm within the KRAS wild-type population, in which the response rate was 17% with wild-type BRAF compared with no response with mutant BRAF. In addition, there is ample evidence that BRAF mutations may have use as a prognostic marker in mCRC (14, 34–38). In this analysis, it is unknown whether the observed dual mutations in KRAS and BRAF arose from tumor heterogeneity (i.e., different cell populations within the tumor) or the presence of both mutations within individual tumor cells. The use of techniques capable of discrimination at the single-cell level will be required to address this question.
Our results do not support the hypothesis that PIK3CA or PTEN mutations are negative predictive markers as has been suggested by the work of Sartore-Bianchi and colleagues (39). In the randomized and extension studies, 18 patients with wild-type KRAS (codons 12/13/61) treated with panitumumab monotherapy had PIK3CA and/or PTEN mutations, and of these, 3 patients (PTEN mutation, n = 1; PIK3CA mutation, n = 2) achieved a partial response. In addition, 1 of these 3 patients had an exon 20 PIK3CA mutation. Our observations are not consistent with the hypothesis that exon 20 PIK3CA mutations (but not exon 9 PIK3CA mutations) predict response to treatment with anti-EGFR monoclonal antibodies (14). De Roock and colleagues previously reported that although PIK3CA exon 9 mutations did not influence outcomes among patients receiving cetuximab-based chemotherapy for mCRC, exon 20 mutations were associated with a lower response rate (0% vs. 37% for wild-type PIK3CA; P = 0.029) and disease control rate (33% vs. 76%; P = 0.0078), as well as shorter median PFS (11.5 vs. 24 weeks; P = 0.013) and OS (34 vs. 51 weeks; P = 0.0057). However, it is possible that our results may have been affected by the smaller number of patients with PIK3CA data evaluable for response in the phase III study (n = 79) and extension study (n = 48) compared with the De Roock and colleagues analysis (n = 339). Although, we found favorable effects of panitumumab on PFS in patients with wild-type and mutant PTEN and PIK3CA, our results should be interpreted with caution given the small number of patients with these mutations. This exploratory analysis suggests that other mutated genes within the EGFR signaling pathway do not seem to confer the same strictly negative predictive value for response to anti-EGFR antibody therapy that specific mutations in the RAS family genes (KRAS and NRAS) seem to confer.
Future meta-analyses would ideally use the same well-established methodology to ensure confidence and consistency in the wild-type and mutation assignments. This is especially important due to the varying tumor content of each tissue section, as well as the potential for heterogeneity within tumor cells of the individual tissue section. It should be noted that when more than 1 mutation is present in a tissue section extract, the percentage of mutant sequence reads is not always the same across the different amplimers, thus reflecting either aneuploidy of the different gene segments or tumor cell heterogeneity (data not shown). It should also be noted that we did not confirm mutations using a second analytic technique. However, in a separate comparability study, we found assessment of KRAS status using Roche 454 pyrosequencing to have a high level of agreement (κ = 0.94) with direct sequencing (20).
In summary, although only KRAS mutational status predicted response to treatment with panitumumab, among patients with wild-type KRAS, objective responses did not occur in patients with mutations in NRAS or BRAF. Because of the relatively limited frequencies of mutations in KRAS codon 61 and NRAS codons 12/13/61, a large prospective analysis of these biomarkers will be difficult to conduct. However, because these results show the use of retrospective analysis of tumor specimens using massively parallel multigene sequencing, a pooled analysis of gene mutations in the RAS family is warranted.
Disclosure of Potential Conflicts of Interest
M. Peeters and Y. Humblet have honoraria from Speakers Bureau and are consultant/advisory board members of Amgen, Inc. K.S. Oliner, J. Huang, J. Wiezorek, D. Reese, and S.D. Patterson are employees of Amgen, Inc. and have ownership interest (including patents) in the same. A. Parker was an employee of Amgen, Inc. at the time the study was conducted. T. Andre has honoraria from Speakers Bureau of Amgen, Inc. and Merck Serono and is a consultant/advisory board member of Roche. No potential conflicts of interest were disclosed by the other authors.
Conception and design: M. Peeters, K.S. Oliner, A. Parker, S. Siena, E. Van Cutsem, J. Wiezorek, D. Reese, S.D. Patterson
Development of methodology: K.S. Oliner, A. Parker, E. Van Cutsem, J. Wiezorek, S.D. Patterson
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Peeters, K.S. Oliner, A. Parker, S. Siena, E. Van Cutsem, Y. Humblet, J.-L. Van Laethem, T. André, S.D. Patterson
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Peeters, K.S. Oliner, A. Parker, S. Siena, E. Van Cutsem, J. Huang, Y. Humblet, J.-L. Van Laethem, T. André, J. Wiezorek, D. Reese, S.D. Patterson
Writing, review, and/or revision of the manuscript: M. Peeters, K.S. Oliner, A. Parker, S. Siena, E. Van Cutsem, J. Huang, Y. Humblet, J.-L. Van Laethem, T. André, J. Wiezorek, D. Reese, S.D. Patterson
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.S. Oliner
Study supervision: M. Peeters, E. Van Cutsem, D. Reese
This study was funded by Amgen Inc.
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.
The authors thank the extended Clinical Study Team that conducted the extensive measures that allowed sample analysis (Simon Collins, Jennifer Hawkins, Mark Ekdahl, Feng Hong, and Carol Johnson) as well as the Amgen Massachusetts Molecular Sciences Laboratory, particularly Sunita Badola, Kim Tsui, and Lauren Young for generating the raw sequencing data. The authors also thank Ying Cheng and Lixia Zhu (i3 Statprobe) for their extensive programming efforts, and Ali Hassan and Benjamin Scott (Complete Healthcare Communications, Inc., Chadds Ford, PA) for assistance in the preparation of this article.
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).
Prior Presentation: These results were presented in part at the 2011 American Society of Clinical Oncology Annual Meeting, June 3–7, Chicago, IL.
- Received June 25, 2012.
- Revision received December 5, 2012.
- Accepted December 19, 2012.
- ©2012 American Association for Cancer Research.