PIK3CA Mutations Contribute to Acquired Cetuximab Resistance in Patients with Metastatic Colorectal Cancer

This article is featured in Highlights of This Issue, p. 4525

Epidermal growth factor receptor (EGFR) signaling promotes cell proliferation and migration, and the constitutive activation of this pathway is associated with tumor progression and metastasis in colorectal cancer. Anti-EGFR antibodies, such as cetuximab, improve the survival of patients with metastatic colorectal cancer (mCRC). Cetuximab is currently recommended for patients with mCRC expressing wild-type RAS (1). However, in a large proportion of patients who respond, acquired resistance to cetuximab emerges despite the absence of detectable mutations in RAS.

It is now clear that embryologic origin of the colon contributes to the genomic profile of a colorectal cancer and has implications on prognosis and response to specific therapeutics. Right-sided colon cancers have embryologic origin in the midgut and have higher rates of microsatellite instability, gene promoter hypermethylation leading to gene silencing, and BRAF mutation, whereas left-sided tumors originate from the hindgut and have a genomic profile distinct from right-sided tumors. Recently, many studies (2–4) demonstrated that the predictive value of the location was mainly related to the response to biologic agents, cetuximab, and bevacizumab, of which, cell proliferation and developmental pathways, such as RAF/RAS/ERK/MEK and mTOR/PI3K/AKT pathways, could play vital roles in response to anti-EGFR antibody, cetuximab.

Recent studies have implicated mutations in several other genes in the EGFR signaling pathway, such as BRAF, PIK3CA, and PTEN, in the acquisition of resistance to cetuximab (5–7). However, it remains to be determined whether such resistance-causing mutations were present in the primary tumor or occurred during treatment. Diaz and colleagues have previously reported that KRAS mutations were responsible for acquired resistance to EGFR blockade in nine of 24 patients with mCRC (8). We hypothesized that in addition to KRAS, mutations in genes downstream of EGFR may also confer acquired resistance to cetuximab-based therapy.

Recent studies have used next-generation sequencing (NGS) to test for specific mutations in circulating tumor DNA (ctDNA) obtained from patient plasma. However, these studies monitored mutations identified in tumor tissue from the same patient or known “hot-spot” mutations (9–14); hence, any other mutations emerging during treatment that results in acquired resistance are likely to have been missed.

Here, we have evaluated a panel of genes that are thought to play pivotal roles in the activation of the EGFR signaling pathway or frequently mutated in colorectal cancer. We overcome these previous limitations by employing ultra-deep amplicon sequencing technology and a newly developed data analysis approach, and we describe the identification of emergent resistant mutations in ctDNA from longitudinal plasma samples obtained from patients with mCRC based on dynamic changes in allelic fractions. By analyzing structural and functional changes, as well as the clinical influence of these mutations, we extend our existing knowledge of mutations that contribute to acquired cetuximab resistance.

We performed a retrospective single-center study at the Affiliated Hospital, Academy of Military Medical Sciences, Beijing, China. Eligible patients (see Supplementary Materials and Methods) had pathologically confirmed mCRC harboring wild-type KRAS codons 12 and 13 and wild-type BRAF codon 600, as determined via Sanger sequencing of the tumor tissue DNA. NRAS was not recommended for routine testing at the onset of the study in 2011 (15). Patients received cetuximab treatment with or without chemotherapy. Prior treatments were permitted except for cetuximab. Serum carcinoembryonic antigen (CEA) and cancer antigen 19-9 (CA 19-9) levels were measured during each therapeutic cycle. Computed tomography (CT) scans were performed and reviewed every 6 to 8 weeks to evaluate clinical response using the Response Evaluation Criteria in Solid Tumors (RECIST), version 1.1 (16). Clinical data including response evaluation during the study were collected. Longitudinal blood samples (4 mL) from each patient were obtained at baseline and every 4 weeks until disease progression or until the last plasma sample was collected. Formalin-fixed paraffin-embedded (FFPE) tumor tissues and peripheral blood mononuclear cells were also collected before treatment. Blood specimens were frozen at −80°C and were linked to demographic, clinical, and genetic data stored in a secure research database. This study protocol was approved by the local ethics committee (KY-2011-8-3), and was conducted in accordance with International Ethical Guidelines for Biomedical Research Involving Human Subjects (CIOMS). Written informed consent was obtained from each patient.

Following DNA extraction and quantification, samples from plasma and tumor tissues were both subjected to amplicon deep sequencing (see Supplementary Materials and Methods) for the following target regions, which are thought to play pivotal roles in the activation of the EGFR signaling pathway or frequently mutated in colorectal cancer: AKT1 (exon 3), BRAF (exon 15), EGFR (exons 10 and 12), KRAS (exons 2, 3, and 4), NRAS (exons 2, 3, and 4), PIK3CA (exons 8 and 19), PTEN (exons 5, 7, and 8), and TP53 (exons 5, 6, and 7; refs. 6, 17, 18), as well as the splice site regions of these exons (i.e., intronic regions within 4 bp of an exon/intron boundary).

The amplicon libraries were subjected to deep sequencing using a Proton System (Life Technologies). The primer sequences are provided in Supplementary Materials and Methods. Targeted average sequencing depths were 1,000 × for FFPE tumor tissue or blood cell DNA samples and 10,000 × for cfDNA samples. All testing was conducted by an independent laboratory supplied with blinded samples. (QuestGenomics Biotechnology Co, Ltd.).

After routine quality control and data processing steps, we called single-nucleotide variants (SNVs) without filtering for any variant frequencies. Somatic mutations were identified by comparing SNVs from the genomic DNA extracted from peripheral blood mononuclear cells and FFPE samples. Using quartile analysis, mutations with significantly increased variant frequencies upon disease progression, or those that were present only during disease progression, were identified as candidate mutations associated with acquired resistance. Deep-sequencing data processing, SNV calling, and quartile analysis are described in the Supplementary Materials and Methods. Mutations were annotated using SNPnexus (19). All nonsynonymous mutations were further analyzed using PolyPhen-2 and SIFT to determine their potential effects on protein function (20, 21). Variants predicted to be “damaging” by both PolyPhen-2 and SIFT were selected for further analysis. The workflow was shown in Supplementary Fig. S1.

MODELLER (version 9v6; ref. 22) was used for homology modeling. Three-dimensional (3D) structures of wild-type and mutant PIK3CA binding to PIK3R1 were modeled using the 3D structure of 4JPS in the PDB (Protein Data Bank) database as a template (23). Hydrogen atoms were added using CHARMM (version c32b2; ref. 24). The protonation states of titratable residues were determined using an in-house CHARMM script (25). VMD (version 1.9.1; ref. 26) was used to view and analyze the modeled structures.

The PIK3CA point mutations identified in ctDNA and FFPE tumor tissue were introduced into the full-length PIK3CA coding sequence using site-directed mutagenesis and were inserted into an expression vector (RC213112, Origene). Sequences of the primers used to construct PIK3CA mutation expression vectors are provided in Supplementary Materials and Methods. A human colorectal cancer cell line (DiFi) was transfected with the PIK3CA mutation expression vectors and assessed by Western blot to identify changes in the phosphorylation levels of AKT and other downstream EGFR targets. The following antibodies were used: Flag (Sigma-Aldrich, Cat. # F3165), p-AKT (Ser473) (Cell Signal, Cat. # 4060), pan-AKT (Cell Signal, Cat. # 4685), p-ERK1/2 (Cell Signal, Cat. # 4367), and total ERK1/2 (Cell Signal, Cat. # 9107). Cell viability analyses were performed to detect sensitivity to cetuximab and/or to 5-fluorouracil (5-FU). The hotspot mutations in PIK3CA exons 9 and 20 (p.E542K, p.E545K, and p.H1047R) were included for comparison. Vector construction, Western blot, and cell viability analysis are described in the Supplementary Materials and Methods.

A validation cohort of patients was prospectively enrolled to verify the acquired resistance mutations detected only in ctDNA after the main data analyses were completed. These patients were enrolled on the same basis as the former patients, the exception being the inclusion of wild-type NRAS codons 12 and 13 in the eligibility criteria according to updated studies (1). Longitudinal plasma samples were collected every 4 to 8 weeks during the entire treatment period. Investigators blinded to clinical data conducted mutational analyses independently.

The following gene regions were added to the former panel for Amplicon deep sequencing: APC (exons 14, 15, and 16), EGFR (exons 18 to 21), GNAS (exons 8), PIK3CA (exons 9 and 20), and TP53 (exons 4b and 8), as well as the splice site regions of these exons. All of these target regions were reported to be frequently mutated in colorectal cancer and are possibly implicated in drug resistance (27, 28). The primer sequences are provided in Supplementary Materials and Methods. The remaining conditions for the next-generation sequencing were the same as for the former experiments.

Kaplan–Meier methods were used to estimate progression-free survival (PFS). A univariate Cox regression analysis was performed for each of the variables of interest. The hazard ratio (HR), 95% confidence intervals (CIs), and Wald statistic P values are reported for each model. A P value of less than 0.05 was considered statistically significant, and an HR greater than 1.5 was considered clinically meaningful due to the small sample size. All statistical tests were performed using SPSS version 20.0 software (IBM).

In total, 59 patients from the initial development cohort were screened for eligibility between August 2011 and December 2013. Six patients were excluded from the study due to the lack of suitable tumor samples or of peripheral blood cell samples collected prior to therapy. The remaining 53 patients were enrolled in the study as the development cohort and received at least 4 weeks of cetuximab alone or in combination with chemotherapy. The patients were divided into two groups based on the response evaluation at week 12. We defined primary resistance as PFS < 12 weeks and acquired resistance as PFS ≥ 12 weeks (29).

The acquired-resistance group consisted of 20 patients with longitudinal plasma samples (cases 1–20) and 18 patients without longitudinal plasma samples (with only two to three plasma samples including baseline and progression points; cases 21–38). Of the 38 patients with acquired resistance, six patients without longitudinal plasma samples were excluded from further analysis due to the lack of plasma samples at progression. Therefore, a total of 32 patients with acquired resistance were included in the mutational analysis (Fig. 1). As a comparison group, we included 15 primary resistance patients (cases 39–53). Detailed clinical information is provided in Supplementary Table S1.

We sequenced 18 exons from eight genes involved in EGFR signaling, covering approximately 3,000 base pairs. The average sequencing depth (raw data) across all plasma samples reached 9,664 ×. We identified more than 20,000 SNVs among all samples processed. The vast majority of the SNVs were present in very low allelic fractions (<0.5%; Fig. 2A). We retained these variants in further analyses, as low-abundance mutations may be functionally relevant.

After screening for the allelic fractions of the SNVs by quartile analysis, we identified an average of 26 SNVs in each patient (Fig. 2B). Only those mutations with allelic fractions significantly higher than the majority were eventually selected as candidate mutations. Overall, we identified 39 nonsynonymous SNVs as candidates across all 32 patients with acquired resistance to cetuximab. Among these 39 nonsynonymous SNVs, 20 mutations were selected as acquired resistance mutations using PolyPhen-2 and SIFT (Supplementary Table S2). Some of these acquired resistance mutations, specifically p.W22G in AKT1; p.V600E in BRAF; p.K5N, p.G12D, p.G12V, p.G13D, and p.Q61H in KRAS; p.M134L and p.Q245* in PTEN, are described in the large intestine dataset of the COSMIC database (including in cecum, colon, and rectal cancers), whereas none of the mutations found in PIK3CA (p.K944N, p.F930S, p.V955G, p.V955I, and p.K966E) are described in COSMIC (Supplementary Table S2).

Eventually, in the 32 patients with acquired resistance, 14 mutations were present in 10 of the 20 patients with longitudinal plasma samples, and 12 mutations were present in eight of the 12 patients without longitudinal plasma samples. Twenty-five of the 26 mutations exhibited a highly significant difference (P value < 0.01) from the background nonreference allelic fractions (Table 1; Supplementary Table S3), whereas none of the candidate mutations were identified in the 15 patients with primary resistance.

The detected acquired mutations were observed most commonly in PIK3CA and KRAS, with a total of five PIK3CA mutations identified in seven of the 32 patients (22%), and five KRAS mutations in eight of the 32 patients (25%; three of these patients harbored a double mutation in KRAS). In addition, three point mutations in BRAF, including the hotspot mutation c.1799T>A (p.V600E), and sporadic mutations in AKT1, EGFR, TP53, and PTEN were identified. Five patients harbored mutations in more than one gene (Fig. 2C).

Sixty-two somatic mutations were identified in 30 out of 53 patients. Thirty-nine of these mutations were identified in 22 out of 38 patients with acquired resistance (average two mutations per individual), which mainly clustered in KRAS and TP53 (30.8% each), whereas the remaining 23 mutations were identified in eight out of 15 patients with primary resistance (average three mutations per individual), which were uniformly distributed (17.4% each in KRAS, NRAS, EGFR, and PTEN). Among these 62 mutations, only 14 were detected in plasma samples before treatment, eight of which were in patients with acquired resistance (21% of 38 patients), with the remaining 6 mutations in patients with primary resistance (40% of 15 patients; Fig. 2D; Table 2).

Structural modeling revealed that all of the PIK3CA point mutations identified in ctDNA (p.K944N, p.F930S, p.V955G, p.V955I, and p.K966E) and in FFPE tumor tissue (p.V952A and p.L938*) affected the structural configuration of the PIK3CA protein and the kinase activity. p.F930 is located in the ATP-binding pocket, p. K944, p.V952, and p.V955 are in the activation-loop, and p.K966 is in the coil region close to the c-lobe, which is thought to stabilize the kinase conformation (Fig. 2E–H).

To confirm the role of PIK3CA mutations in cetuximab resistance, DiFi cells were transfected with PIK3CA mutations p.K944N, p.V955G, p.V955I, p.K966E p.F930S, and p.V952A, as well as the truncation mutation p.L938* were generated. We performed Western blot assays for pAKT (phosphorylated AKT) and pERK in DiFi cells transfected with these PIK3CA mutations. Compared with nontargeting control cells, PIK3CA mutations p.K944N, p.V955G, p.V955I, and p.K966E markedly increased the phosphorylation levels of AKT and MAPK3/MAPK1 (ERK1/ERK2), which were not affected by the addition of cetuximab alone or in combination with 5-FU. While the PIK3CA mutations p.F930S and p.V952A, as well as the truncation mutation p.L938*, had a less effect on AKT and MAPK3/MAPK1 phosphorylation, especially in the presence of cetuximab and/or 5-FU (Fig. 3I).

Cell viability analysis suggested that in the presence of cetuximab in combination with 10 μmol/L 5-FU, compared with mock transfected cells, DiFi cells overexpressing p.K944N, p.V955G, p.V955I, p.K966E, p.E542K, p.E545K, and p.H1047R exhibited a high degree of resistance. Wild-type PIK3CA, as well as its p.F930S and p.V952A mutants, resulted in moderate resistance to cetuximab, and the truncation mutation p.L938* did not exhibit resistance. In the presence of 10 μg/mL cetuximab, growth rates in cells with novel PIK3CA mutations (p.K944N, p.F930S, p.V955G, p.V955I, and p.K966E) and with wild-type PIK3CA were 59.18 ± 14.72% versus 31.37%, respectively (Fig. 2J; Supplementary Fig. S2). These results demonstrate that these candidate mutations are likely to be involved in conferring resistance because of their effects on the activation of the phosphorylation of AKT and MAPK3/MAPK1 (ERK1/ERK2).

Among the 20 patients with longitudinal plasma samples, 10 carried candidate mutations. We compared changes in the allelic fraction of the identified mutations in each patient with serum levels of corresponding tumor biomarkers and response ascertained using CT (as defined by RECIST; Fig. 3A–D and Supplementary Fig. S3). PIK3CA mutations were identified in four patients (Nos. 6, 7, 16, and 17), and KRAS mutations were identified in an additional four patients (Nos. 2, 12, 13, and 19). Only two of the 10 patients harbored mutations in other genes. Patient 14 harbored the BRAF p.V600E mutation, whereas patient 20 carried the PTEN p.Q245* mutation. These mutations exhibited an increasing allelic fraction during treatment (Table 1; Supplementary Table S3). Whereas, only five of the 10 patients had elevated CEA or CA19-9 levels that correlated well with the tumor response. The levels of candidate resistant mutations increased prior to clinical progression, as determined by CT scans. A median of 15.0 weeks of lead time (range, 0–35 weeks) was observed. In patients carried PIK3CA mutations, the average lead time was 10.8 weeks, whereas in patients with KRAS mutations, the average lead time was 21.8 weeks (Supplementary Table S1).

An independent validation cohort of 32 patients (cases v1–v32) was prospectively enrolled between January 2014 and June 2016 to verify the acquired resistance mutations detected in the ctDNA of patients from the development cohort. Five patients (cases v1, v3, v9, v17, and v18) were placed in the primary resistance group, whereas the remaining 27 patients were classed as having acquired resistance (Table 3; Supplementary Table S4).

Thirty-seven mutations were identified in 14 of the 27 patients with acquired resistance, including one novel PIK3CA mutation (p.K944N) and the hotspot mutations p.V600E in BRAF; p.G12V in KRAS; p.G12V, p.G13V and p.Q61H in NRAS; p.H1047R and p.E545K in PIK3CA; and p.T790M in EGFR (Supplementary Table S5, Supplementary Fig. S4). Among these 14 patients, five patients carried multiple gene mutations, including two of them harbored both KRAS and PIK3CA hotspot mutations (case v5: p.H1047R in PIK3CA and p.M72V in KRAS; case v20: p.E545K, p.Q546K in PIK3CA, and p.G12V in KRAS). Other novel mutations in PIK3CA (p.F930S, p.V955G, p.V955I, and p.K966E) identified in the initial cohort were not detected in patients in this validation cohort. Although p.V600E in BRAF and p.G12V in KRAS were the only mutations verified in both cohorts due to the limited sample size and increased gene regions for sequencing in validation cohort, the identification of aforementioned hotspot mutations testified the reliability of the results in this validation cohort. No acquired resistance mutations were detected in the five patients with primary resistance in the validation cohort.

The relationship between tumor location and RAS or PIK3CA mutations were compared. We first analyzed RAS and PIK3CA mutations in tumor tissue before cetuximab treatment. In the 13 cases with right-sided colon, RAS mutation rate was 38.5% (5/13) and PIK3CA mutation rate was 7.7% (1/13); In the 38 cases with left-sided colon, RAS mutation rate was 31.6% (12/38) and PIK3CA mutation rate was 2.6% (1/38; Supplementary Table S6).

We also analyzed differences of acquired RAS or PIK3CA mutations in ctDNA between left and right colons. In 17 patients with right-sided colon, the RAS mutation rate was 23.5% (4/17) and the PIK3CA mutation rate was 11.8% (2/17); in 47 patients with left-sided colon, the RAS mutation rate was 19.1% (9/47), the PIK3CA mutation rate was 17% (8/47). No significant differences were observed between left and right colons both in the RAS and PIK3CA mutation rate (Supplementary Table S7).

Next, we examined progression-free survival (PFS) from the clinical data of 65 patients with acquired resistance, to determine the relative prognostic value of mutation profiles and other clinical features. These patients consisted of 38 patients from the development cohort and 27 patients from the validation cohort.

Patients from the development cohort who had preexisting mutations in FFPE samples that were also present in ctDNA prior to treatment showed significantly shorter PFS compared with those without preexisting mutations in ctDNA (mPFS, 14 weeks vs. 38 weeks; HR, 2.69; 95% CI, 1.12–6.49; P = 0.027; Fig. 3E), respectively. We did not perform this analysis in the validation cohort due to the unavailability of FFPE samples.

A trend toward a shorter PFS was noted in patients with a residual primary tumor compared with those with no residual primary tumor (mPFS, 20 weeks vs. 41 weeks; HR, 1.71; 95% CI, 0.94–3.13; P = 0.08; Fig. 3F).

When comparing to patients with no mutation (n = 27) detected in ctDNA, patients with PIK3CA (n = 7) or RAS (n = 10) mutations showed a pronounced mPFS decrease, although without significant difference due to very small sample size (40 weeks vs. 38 weeks vs. 29 weeks; HR, 1.26; 95% CI, 0.79–2.01; P = 0.34; Fig. 3G). Taken together, these results suggest that preexisting mutations present in ctDNA, residual primary tumor and RAS mutations will be potentially negative factors for cetuximab treatment.

In this study, our purpose was to investigate the potentially acquired resistance mutations in eight targeted genes after clinical benefit from cetuximab-based regimens. The prerequisite concerning the acquired resistance during the cetuximab therapy is the PFS ≥ 12 weeks. For this reason, we enrolled all mCRC patients with no previous anti-EGFR antibody treatment prior to study participation who would be treated with cetuximab-based regimen, no matter how many previous treatment lines patients experienced.

We sequenced targeted gene areas with deep coverage to identify emergent mutations in longitudinal plasma samples collected from patients with mCRC. Our quartile-based selection strategy helped us to identify mutations that started at extremely low allelic fractions but increased in frequency during treatment, independently of the identification of existing somatic mutations in tumor tissues or in known “hot-spots.” We were therefore not only able to identify novel mutations associated with acquired resistance, but also to build an understanding of resistance in each patient, even those with diverse mutation profiles.

Previous studies have demonstrated that mutations in the EGFR signal transduction pathway, such as RAS, contribute resistance to cetuximab (30). However, whether the PIK3CA mutations at exon 9 and 20 that have been identified in tumor tissue are correlated with drug resistance remains controversial (5, 17, 31, 32). As in the previous study, Bettegowda and colleagues analyzed known “hot-spot” mutations in NRAS, BRAF, EGFR, and PIK3CA and concluded that KRAS mutation is the most important factor for acquiring resistance to anti-EGFR therapy. However, they did not identify any known PIK3CA mutations associated with the acquisition of resistance (13, 14).

For PIK3CA, we examined exons 8 and 19 instead of previously reported hotspot mutations in exon 9 and 20, because the hotspot mutations in exons 9 and 20 were not identified in previous reports of acquired cetuximab resistance (13, 14). Bettegowda and colleagues had examined the acquired mutations of key genes in EGFR signaling pathway including KRAS, NRAS, BRAF, EGFR, and PIK3CA (exons 9 and 20) from ctDNA of 24 patients with mCRC during EGFR blockade. They observed 70 acquired mutations including KRAS, NRAS, BRAF, and EGFR (13). Siravegna and colleagues had monitored the acquired mutations of 226 genes, including PIK3CA (exon 20), in 16 patients with mCRC during anti-EGFR antibody therapy. They identified only KRAS and EGFR mutations. These two studies did not identify treatment-related mutations in exons 9 and 20 of the PIK3CA gene (14). Exon 8 encodes the region responsible for plasma membrane binding and exon 19 encodes part of the catalytic domain, including the ATP-binding pocket (33, 34). We reasoned that mutations in these PIK3CA sites might cause the constitutive activation of the kinase, similar to previously identified hotspot mutations (35, 36), although no any mutations in exons 8 and 19 had been reported in previous studies.

In this study, by analyzing different exons, we discovered five novel mutations in PIK3CA exon 19 in the development cohort and one novel mutation in exon 19, two “hot-spot” mutations in exons 9 and 20 in the validation cohort are closely associated with the development of resistance. Protein structure and functional assays revealed that all of these mutations would affect the structural configuration of PIK3CA, resulting in functional changes of protein. In vitro studies also demonstrated that novel PIK3CA mutations (p.K944N, p.V955G, p.V955I, and p.K966E) exhibited a high degree of resistance to cetuximab because of their effects on the activation of the phosphorylation of AKT and MAPK3/MAPK1 (ERK1/ERK2).

Analyses of molecular pathology and its interactions with environment are increasingly important in cancer research (37–42). As for the comparison of RAS or PIK3CA mutation rate of left- and right-sided colons, both RAS and PIK3CA mutation rates in tumor tissue were higher in the right-sided colon, which is consistent with findings in the literature reports (43–48), although there were no significant differences due to small sample size. Whereas acquired RAS mutation is more frequent in the right colon, acquired PIK3CA mutation is more frequent in the left colon. Also, there were no significant differences due to small sample size.

We further compared the dynamics of acquired RAS and PIK3CA mutations in ctDNA and their influence on PFS. As compared with those with KRAS mutations, patients with PIK3CA mutations had a shorter lead time of presence of acquired mutations in ctDNA to clinical progression (21.8 vs.10.8 weeks). Patients harboring PIK3CA mutations exhibited increased median PFS, although the difference was not statistically significant due to the limited sample size. Because RAS is the upstream activator of the PI3K pathway (18, 49), we speculate that PIK3CA mutations in sites other than the RAS-binding domain are late-onset molecular events, relative to RAS mutations, that occur after exposure to cetuximab. These mutations might potentially exert a complementary effect in tumor progression and acquired resistance to cetuximab.

Another negative prognostic factor revealed in our study is the presence of residual primary tumor, which few studies in cetuximab treatment have mentioned. It is possible that when clones with wild-type genes are eliminated under the selective pressure of cetuximab, the remaining clones with preexisting and other emergent mutations in primary tumor would expand and display resistance over a relatively short time period, resulting in a shorter PFS. In the COIN clinical trial (50), which failed to demonstrate the survival benefit with the addition of cetuximab to chemotherapy, one underlying factor affect the survival benefit could be a substantial proportion of patients (>40%) with advanced CRC enrolled had unresected primary tumors. This independently supports our finding.

One limitation of this study is that the sample size in both cohorts was small. Larger sample size might help to discover more novel mutations, especially to distinguish the impact of clinical characters, such as tumor location (45, 48) and preexisting mutations on PFS between patients with and without acquired resistance mutations. Second, our present study focused only on the genomic variations analysis. Future investigation will be warranted to explore the potential mechanism beyond EGFR signaling pathway.

In summary, our data indicate that the PIK3CA mutations contribute to acquired cetuximab resistance in patients with mCRC. Compared with RAS mutations, which were reported to present before cetuximab treatment (8, 13, 14), PIK3CA mutations probably occur after exposure to cetuximab, and may potentially exert a complementary effect in acquired resistance to cetuximab.

No potential conflicts of interest were disclosed.

We use HUGO (Human Genome Organisation)-approved symbols for genes and gene products, including AKT1, APC, BRAF, EGFR, GNAS, KRAS, MAPK1, MAPK3, NRAS, PIK3CA, PIK3R1, PTEN, and TP53, all of which are described at www.genenames.org. Gene names are italicized, and gene product names are nonitalicized.

Conception and design: J.-M. Xu, Y.-L. Wang, M. Ni, Z.-L. Tan

Development of methodology: Y.-L. Wang, Y. Wang, M. Ni

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Wang, Y.-L. Wang, T. Liu, L. Lin, F.-J. Ge, J.-Y. Gu, R. Jia, Y.L. Chen, R.-R. Liu, C.-H. Zhao, Z.-L. Tan, Y. Jin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis):J.-M. Xu, Y. Wang, Y.-L. Wang, Y. Wang, M. Ni, M.-S. Li, C. Gong, Z.-L. Tan, Y. Zhu, S. Ogino, Z.R. Qian

Writing, review, and/or revision of the manuscript: J.-M. Xu, Y. Wang, Y.-L. Wang, R. Jia, S. Ogino, Z.R. Qian

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Wang, Y.-L. Wang, H.-F. Wang

Study supervision: J.-M. Xu, Y. Wang, Z.R. Qian

We thank all medical and ancillary staff at the Cancer Center and the patients for consenting to participate; Bo Zhang of the Department of Pathology for assistance with the preparation and assessment of formalin-fixed paraffin-embedded tumor samples; Xiang-Hua Zuo of the Clinical Laboratory for assistance with CEA and CA 19-9 analysis; Gong-Jie Li of the Department of Radiology for assistance with radiological response evaluation; and Su-Ping Lang and Chang-Hua Chen of GCP ClinPlus Co., Ltd., for their contribution to the statistical analysis; Peter Gorsuch, the scientific editor of Nature Research Editing Service, for their assistant in manuscript preparation.

This work was supported by the National Natural Science Foundation of China (to J. Xu, Y. Wang, F. Ge, and R. Jia; Project No. 81573458), the Translational Medicine Fund of Academy of Military Medical Sciences (to Y. Wang, J. Xu, Y. Wang, and R. Jia; Project No. 2013ZHYX006), the Chinese Society of Clinical Oncology Merck Serono Oncology Research Fund (to Y. Wang, J. Xu, Y. Wang, F. Ge, and R. Jia; Project No. Y-MX2014-052) the U.S. National Institutes of Health (NIH) grant (to S. Ogino; R35 CA197735), and the Medical Oncology Translational Grant Program of Dana-Farber Cancer Institute(to Z.R. Qian).

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.