Fc-γ Receptor Polymorphisms, Cetuximab Therapy, and Survival in the NCIC CTG CO.17 Trial of Colorectal Cancer

Only a proportion of patients with colorectal cancer receiving cetuximab derive benefit from the drug (1). However, other than RAS and BRAF mutations, no biomarkers have been identified as clinically useful predictors of response to cetuximab (2, 3). As a therapeutic mAb, cetuximab contains an antigen-binding fragment (Fab) that binds to the extracellular domain of the transmembrane EGFR, which is highly expressed in patients with colorectal cancer. Cetuximab also contains the crystalline fragment (Fc) that binds the Fc-γ receptor (FCGR) on a host effector cell, typically a natural killer cell, macrophage, or monocyte, and thereby may induce tumor cell killing via antibody-dependent cellular cytotoxicity (ADCC). This is initiated when the Fab binds to EGFR while the Fc interacts with the effector cell's FCGR, resulting in target cell recognition and a lytic attack (4–6). Other potential mechanisms of action include FCGR-mediated endocytosis and phagocytosis of antibody-coated tumor cells that utilize tumor-directed T-cell immunity (7). Three classes of FCGR exist, encoded by related genes on the long arm of chromosome 1; FCGR1–CD64; FCGR2–CD32; and FCGR3–CD16 (8).

Two polymorphisms in the coding regions of the FCGR2A and FCGR3A genes, respectively, appear to have clinical significance, correlating with responses to cetuximab. A coding polymorphism in the extracellular domain of FCGR2A (rs1801274) changes the amino acid from histidine (H) to arginine (R; ref. 9). Although the FCGR2A receptor has highest affinity for human IgG1 and IgG3 (10), the H-to-R substitution is known to affect binding to IgG2 (and not IgG1) in humans, with the R form showing poor affinity (9). The rs396991 polymorphism of FCGR3A is also found in the extracellular domain, leading either to a phenylalanine (F) or valine (V); this FCGR3A polymorphism interacts with the lower hinge region of IgG1 (11, 12).

Prior studies evaluating the role of FCGR polymorphisms and the efficacy of cetuximab in colorectal cancer (13–19) have identified associations for both polymorphisms that now require validation. Furthermore, many prior studies had various limitations, which included (i) small study sizes; (ii) the use of convenient case series, rather than prospective, systematically collected samples; (iii) lack of consideration of KRAS mutation status in the analysis; (iv) various concurrent therapies with cetuximab; and (v) technical issues with genotyping. None has involved the evaluation of a large, randomized controlled trial of cetuximab monotherapy with a no-active-treatment control arm. Not surprisingly, results from these prior studies yielded conflicting results. CO.17 is a unique validation dataset because it addresses a number of these limitations that allows us to analyze the potential predictive effects of these polymorphisms on treatment outcome.

Our primary objective was to determine whether survival is affected by a significant interaction between FCGR polymorphism and cetuximab treatment after adjusting for other potential prognostic factors. To address the known importance of KRAS mutations on cetuximab efficacy, we performed separate analyses by KRAS mutation status, as well as in all patients (adjusting for KRAS status). A secondary, exploratory objective was to determine whether these same polymorphisms were associated with general prognosis in patients untreated with cetuximab (i.e., those treated with best supportive care only), after adjusting for other potential prognostic factors.

This retrospective, secondary analysis of two germline polymorphisms, FCGR2A:H→R,(rs1801274) and FCGR3A:F→V,(rs396991), used available DNA samples from a multicenter, open-label, randomized (1:1), phase III trial of cetuximab monotherapy versus best supportive care (NCIC-CTG CO.17; BMS-CA225-025). Polymorphism–treatment arm interactions for each of the two polymorphisms on overall survival (OS) were evaluated as the primary endpoint. Secondary endpoints included treatment arm–specific subset analyses for each polymorphism. REMARK guidelines were followed (20).

This trial was led by NCIC Clinical Trials Group (NCIC-CTG) in collaboration with the Australasian Gastro-Intestinal Trials Group (1). All patients received best supportive care, defined as those measures designed to provide palliation of symptoms and improve quality of life as much as possible. Patients had previously treated, metastatic, EGFR immune-positive colorectal cancer. The primary trial endpoint was survival. The final analysis for this study was performed in 2006, demonstrating a significant survival advantage in the cetuximab arm with a median OS of 6.1 versus 4.6 months in the best supportive care group (HR, 0.77; P < 0.005). An analysis by somatic KRAS status found that the benefits of cetuximab were restricted to patients with KRAS wild-type versus KRAS-mutated tumors (median OS 9.5 vs. 4.8 months; HR, 0.55; P < 0.001; ref. 2).

For the current analysis, this underlying trial population was used to generate the following defined datasets: (i) a biomarker dataset, henceforth described as “all patients”, which were all subjects who were randomized and had successful genotyping for one or both of the two FCGR polymorphisms; (ii) a KRAS wild-type biomarker dataset, which included all randomized subjects with successful genotyping and a wild-type KRAS status; and (iii) a KRAS-mutant biomarker dataset, which included all randomized subjects with successful genotyping and KRAS-mutant status.

The NCIC-CTG trial database was used for all analyses. Clinical outcomes were defined as per RECIST 1.0 and included OS and progression-free survival (PFS). OS was defined as the time from study entry until death or censored at last follow-up, if alive. PFS was defined as the time from study entry until first progression of disease, death, or censored at last follow-up. There were too few objective clinical responses for interaction analyses; hence, best clinical response was not evaluated as an outcome.

Formalin-fixed and paraffin-embedded (FFPE) normal and tumor tissue samples from local sites were archived at the NCIC-CTG central tumor bank (Queen's University, Kingston, Canada), where a single 1.2-mm diameter core was taken from each block. Extracting DNA from glass slides (some over a decade old) yielded highly variable success; thus, only DNAs obtained from cores were analyzed. DNA was extracted using the Qiagen FFPE DNA Kit. DNA quantity (spectrophotometry) and quality (polymerase chain reactions) were checked. DNA was genotyped blindly by Transgenomic, Inc. using Sanger sequencing. Replicate sequencing was performed in the laboratories of G. Liu (Sanger sequencing) and A. Dobrovic (pyrosequencing). All samples were tested in at least two different laboratories, and any discrepant results were genotyped in the third laboratory. Primer sequences are presented in Supplementary Table S1. FCGR3A genotyping is especially challenging because of the presence of a pseudogene. We utilized two different sets of unique primers across two or more laboratories to ensure accuracy and reproducibility and further checked results using Hardy–Weinberg equilibrium testing (21). Germline DNA extracted from lymphocytes of individuals with known FCGR genotypes was used as controls. KRAS mutations were identified previously (2).

The characteristics of all patients in the original trial and in the subset of patients with genotyping results were tabulated and compared. OS and PFS were assessed using Kaplan–Meier curves, log-rank tests (univariable analyses), and Cox proportional hazard models in multivariable analyses, adjusting for clinically relevant factors identified in the original trial analysis (1). The primary analyses utilized two approaches to genetic inheritance: (i) additive genetic models used for screening analyses; and (ii) additional analyses using other models of inheritance, particularly dominant, codominant, and recessive models. It should be noted that a priori data suggested a dominant model for both FCGR2A and FCGR3A polymorphisms, and thus, the dominant model was adopted as the main secondary analysis. Tests of assumption of proportional hazards were performed. SAS software v.9.1 was used for all analyses.

The consort diagram of samples is shown in Fig. 1; demographic and disease variables are summarized in Table 1. There was no difference in the distribution of these variables between all randomized patients and the subset of patients with genotyping data. In addition, neither polymorphism was associated with the following factors (P > 0.10 for each comparison): age, sex, performance status, disease site (colon and rectum), prior therapies, site and number of metastatic sites, and assignment of treatment arm. Hardy–Weinberg equilibrium P values were P = 0.59 (FCGR2A) and P = 0.65 (FCGR3A). 98.6% (FCGR2A) and 99.3% (FCGR3A) of results were concordant across laboratories; although consensus through a third genotyping attempt resolved conflicting results in most of the remaining samples, a decision was made to exclude these small number of discrepant samples from further analyses.

In all Cox models, the assumption of proportional hazards was never violated. Table 2 presents the interaction analysis between treatment arm and genotype of the FCGR2A polymorphism on survival, which was performed under a screening additive genetic model. Three sets of comparisons were performed: in the KRAS wild-type subset of patients (primary analysis); for comparison purposes, data were also reported for KRAS-mutant patients; and in all patients (irrespective of KRAS mutation status). In all three comparisons, the same association was observed: patients carrying the H/H genotype (31% of KRAS wild-type; 28% of all patients irrespective of KRAS status) benefited greatly from cetuximab, whereas those carrying H/R or R/R benefited much less. For OS (Table 2, top), cetuximab had an approximate 3-fold benefit in both KRAS wild-type and KRAS-mutant patients (together or separately) carrying the H/H genotype. H/H patients also had improved PFS when compared with R allele carriers. For PFS (Table 2, bottom), an association was observed most prominently in the KRAS wild-type patients, where carrying the H/H genotype resulted in cetuximab yielding a 5-fold survival benefit with an adjusted HR (aHR) = 0.19, comparing cetuximab-treated patients with patients on no active treatment. In contrast, carrying the R allele meant that receiving cetuximab approximately doubled PFS. This association was not observed in KRAS-mutant patients (aHRs comparing treated vs. untreated of 0.65 for H/H and 0.70 for R/R genotypes). In a separate analysis of the entire population, which adjusted for clinically relevant covariates including KRAS, the interaction between genotype and benefit from cetuximab therapy had P values of 0.005 (OS) and 0.06 (PFS). Kaplan–Meier curves are shown in Fig. 2 (OS) and Fig. 3 (PFS).

In our study, alternative genetic inheritance assumptions were prespecified and integral. No recessive or codominant inheritance models demonstrated significant (P > 0.05) interactions. However, significant interactions existed between cetuximab and FCGR2A for the dominant genetic models (Table 2, top, right-indented data). Although for patients carrying the H/H genotype, treatment with cetuximab conferred a 3-fold improved OS, the benefit of receiving cetuximab was diminished for patients carrying at least one R allele with aHRs of 0.78 (KRAS wild type; P > 0.05) and 1.14 (KRAS mutant; P > 0.05). Results were similar when considering PFS as the outcome (Table 2, bottom); even though there was benefit in all genotype groups, the strongest benefit was found in the patients carrying the H/H genotype.

There were no significant interactive relationships between carrying FCGR3A and cetuximab therapy on OS or PFS. We also evaluated the joint effect of combinations of both FCGR polymorphisms, using categories found in prior cetuximab studies that reported significance with combination definitions (13, 16, 17). None of the interaction terms in these analyses was significant in our data (P > 0.10 for each comparison). Thus, the addition of the FCGR3A polymorphism data to the FCGR2A analysis did not improve the significance or magnitude of associations, regardless of how the two polymorphisms were combined (dominant–dominant, addition of number of variant genotypes, or combinations based on prior publications).

Among patients in the best supportive care arm, OS was significantly longer in patients carrying at least one R allele of FCGR2A, compared with patients carrying the FCGR2A H/H genotype (Table 3; adjusted P < 0.0001 for the dominant model). This association is seen only with OS, and not with PFS. There were no prognostic relationships observed with the FCGR3A polymorphism.

In metastatic colorectal cancer, cetuximab is currently indicated for patients with KRAS wild-type tumors. In this study, the relative survival benefit of receiving cetuximab as defined by FCGR genotypes was measured in patient subsets based on tumor KRAS mutation status. Individuals with KRAS wild-type tumors carrying the FCGR2A H/H polymorphic variant (representing almost a third of KRAS wild-type patients) had a 3-fold increase in survival when receiving cetuximab (aHR = 0.36), corresponding to a 5.5-month absolute OS benefit (P = 0.003) and a 3.7-month PFS benefit (P = 0.02). In contrast, the two thirds of KRAS wild-type patients carrying any other FCGR2A genotype (R/R or H/R, i.e., one or more R alleles) derived significantly less benefit from cetuximab (aHR = 0.78), with a nonsignificant 2.8-month OS benefit (P = 0.27) and a significant, albeit small, 1.6-month PFS benefit (P = 0.003). In the additive screening model, the greater number of arginine alleles carried by a patient led to a lower benefit of receiving cetuximab. These relationships correspond to significant interactions between cetuximab therapy and the FCGR2A genotype for survival in metastatic colorectal cancer patients and specifically in the KRAS wild-type subgroup of patients (Table 2).

Similar aHRs for cetuximab were observed in both KRAS wild-type and KRAS-mutated patients with metastatic colorectal cancer, suggesting that the association between the FCGR2A polymorphism and cetuximab benefit may be independent of KRAS mutation status. This was expected, given that KRAS and FCGR are involved in distinct molecular mechanisms important to the effectiveness of the drug. However, the baseline absolute benefit of cetuximab was significantly smaller in KRAS-mutated patients, and survival was generally poorer at baseline, resulting in an OS improvement of only 1.8 months (P = 0.02) and no significant PFS improvement. Thus, the clinical impact of the KRAS mutation continues to overshadow any modest benefit derived from carrying the FCGR H/H genotype in patients carrying the KRAS mutation.

Although our findings validate the results of several prior publications (10, 13, 17, 20, 23), in some of these studies, the identified relationship involved a combination of FCGR2A and FCGR3A polymorphisms, rather than with the FCGR2A polymorphism alone. A recent consortium analysis of over a thousand cases, however, did not find any FCGR2A relationships (24). What sets our study apart is the confirmation that the FCGR2A relationships are predictive of cetuximab outcome in a pure randomized cetuximab versus no cetuximab setting, which is a key critical point in determining the clinical impact of a biomarker. The idea that the FCGR2A polymorphism status of a patient might modify the effect of cetuximab on survival is biologically plausible. Musolino and colleagues (10) studied trastuzumab (another therapeutic monoclonal IgG1 antibody) and identified greater ADCC-related cytotoxicity in patients with the H/H genotype, when compared with other patients. Importantly, in a study of patients treated with trastuzumab, Tamura and colleagues (2011) also reported that the FCGR2A H/H genotype had favorable outcomes of pathologic complete response to trastuzumab in a neoadjuvant setting and PFS in a metastatic setting (22).

In our original analysis plan, we prespecified a standard screening genetic additive model. In retrospect, based on functional testing (10) and publications (10, 13, 17, 20), the bulk of evidence now points to dominant genetic model of FCGR2A. It is also under this dominant genetic model that CO.17 demonstrates the strongest interaction and greatest significance for OS/PFS (by interaction P value). The results of this analysis are of significant interest, as other trials are unlikely to be used for a similar interaction analysis, either because there is no comparable non-cetuximab arm, or the arms will be confounded by concurrent therapy with other antineoplastic agents.

The interpretation of the FCGR3A data is more complex. As reviewed by Mellor and colleagues (21), there have been equal numbers of studies that found that the V allele was beneficial as studies that found that the F allele was beneficial. For example, although Zhang and colleagues (18) demonstrated a favorable outcome association with the F allele of FCGR3A, Musolino and colleagues (10) reported a favorable association of the opposite FCGR3A V allele. In contrast, both studies reported that the FCGR2A H/H genotype had favorable outcome characteristics, consistent with the results of this study. It is therefore not surprising that in our analyses, we found no relationship between the FCGR3A polymorphism and outcome in either the cetuximab with best supportive care or the best supportive care alone arms; furthermore, we found no evidence of a polymorphism–treatment arm interaction.

A major concern has been the appropriate genotyping of the FCGR3A polymorphism, which has high homology with a known pseudogene that contains a specific sequence that affects the detectability of this polymorphism (21). By utilizing three different blinded laboratories, and three different sets of primers/techniques, the risk of incorrectly genotyping this polymorphism was minimized. Furthermore, unlike other studies (13, 19), the FCGR3A polymorphism in our study was in Hardy–Weinberg equilibrium (P > 0.20), consistent with accurate genotyping.

In an exploratory analysis, carrying the R/R or H/R genotypes of the FCGR2A polymorphism was observed to have improved OS (i.e., a prognostic relationship, unrelated to cetuximab administration), but not PFS, after adjustment for important covariates. We report this prognostic association as being hypothesis generating and to better understand the underlying relationships between FCGR polymorphisms and outcome in this particular trial population. This prognostic observation should be replicated in other colorectal cancer populations and possibly in other cancers, such as breast cancer or lymphoma, before being generalized into other settings.

Strengths of this study include (i) the presence of a randomized nondrug comparison arm, enabling interaction analysis; (ii) the relatively large sample size compared with prior studies; and (iii) having uniformly defined patient management and clinically relevant outcomes. However, there are a number of limitations related to genotyping. First, only 51% of the participants in the original trial provided adequate tissue for genotyping. Although the genotyped group was similar in all key clinical and demographic characteristics when compared with the entire trial population (Table 1), some undetected bias may remain. Second, DNA in a minority (7%) of cases was derived from tumor tissue; however, a sensitivity analyses found no change in conclusions when excluding DNA derived from tumor tissue. Third, because of sample size, there was insufficient power to exclude subtle interactive effects between cetuximab therapy and the FCGR3A polymorphism. Fourth, the KRAS wild-type subset was defined based on the assessment of lack of mutations seen in exon 2; the effect of other, significantly rarer RAS and BRAF mutations on this polymorphism–outcome association is unknown.

In summary, through Cox proportional hazards models of patient survival, we have validated a biologically plausible relationship between a dominant genetic model of the FCGR2A polymorphism and monotherapy with cetuximab. In NCIC-CTG CO.17 patients, cetuximab was associated with a greater survival benefit in FCGR2A H/H patients with KRAS wild-type colorectal cancers. Significantly lower benefit was observed in patients carrying FCGR2A R alleles. The criteria for withholding a potentially beneficial therapy in colorectal cancer must be quite stringent; because all genotypic groups benefited, but to different extents, we cannot recommend withholding cetuximab therapy solely on the basis of FCGR2A genotyping results; standard recommendations based on RAS/BRAF mutations should remain. Nonetheless, our findings in CO.17, in addition to those from Kjerson et al (23), provide compelling evidence that warrants prospective studies evaluating the utility of the FCGR2A polymorphism as a marker in clinical management. That the Geva and collegues' (24) retrospective analysis of observational cohorts is negative further raises the need for a prospective analysis. Such studies should also evaluate the role of rarer RAS and BRAF mutations in this polymorphism–outcome relationship, where the clinical comparison would be to offer KRAS wild-type patients carrying the R allele cetuximab versus alternative therapy, possibly in the context of a multiarm molecular targeting trial.

T.J. Price is a consultant/advisory board member for AMGEN and Merck, N.C. Tebbutt and C.S. Karapetis are consultant/advisory board members for Merck Serono, and J.R. Zalcberg reports receiving commercial research grants from Merck Serono. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G. Liu, D. Tu, M. Lewis, J. Simes, D.J. Jonker, J.R. Zalcberg, C.S. Karapetis, A. Dobrovic

Development of methodology: G. Liu, M. Lewis, D. Cheng, L.A. Sullivan, J.D. Mellor, T. Mikeska, D.J. Jonker, C.J. O'Callaghan, C.S. Karapetis, A. Dobrovic

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Liu, T.J. Price, N.C. Tebbutt, J.D. Shapiro, G.M. Jeffery, J.D. Mellor, T. Mikeska, S. Virk, L.E. Shepherd, D.J. Jonker, C.J. O'Callaghan, C.S. Karapetis, A. Dobrovic

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Liu, D. Tu, E. Morgen, J. Simes, T.J. Price, N.C. Tebbutt, J.D. Shapiro, J.D. Mellor, T. Mikeska, L.E. Shepherd, C.J. O'Callaghan, J.R. Zalcberg, C.S. Karapetis

Writing, review, and/or revision of the manuscript: G. Liu, D. Tu, E. Morgen, J. Simes, T.J. Price, N.C. Tebbutt, J.D. Shapiro, G.M. Jeffery, J.D. Mellor, T. Mikeska, S. Virk, L.E. Shepherd, D.J. Jonker, C.J. O'Callaghan, J.R. Zalcberg, C.S. Karapetis, A. Dobrovic

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Liu, Z. Chen, E. Morgen, J.D. Mellor, S. Virk, C.J. O'Callaghan

Study supervision: T.J. Price, G.M. Jeffery, C.J. O'Callaghan, J.R. Zalcberg, C.S. Karapetis, A. Dobrovic

This research was supported by an OICR High Impact Clinical Trial Small Project Grant and by an unrestricted grant from Transgenomic, Inc. G. Liu was supported by the Alan B. Brown Chair in Molecular Genomics and the Cancer Care Ontario Chair in Experimental Therapeutics and Population Studies, E. Morgen was supported by a CIHR Banting and Best Canada Graduate Scholarship, and J.D. Mellor was supported by a grant from the Peter MacCallum Cancer Centre Research Foundation. This work was also supported by the NCIC Clinical Trials Group Tumour Tissue Data Repository (TTDR). The NCIC CTG TTDR is a member of the Canadian Tumour Repository Network.

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.