Introduction

Gastric cancer is the second leading cause of cancer-related death worldwide. Annually, 989,600 new patients are diagnosed, and 738,000 people die of this disease [1]. Despite recent improvements in surgical approaches combined with chemotherapy and radiotherapy, the benefits of chemotherapy for advanced gastric cancer are still limited. Thus, there is an urgent need for the development of more effective drugs, particularly those targeting tumor progression mechanisms.

In gastric cancer, mutations in genes encoding components of the receptor tyrosine kinase (RTK)/RAS-signaling pathway are infrequent. In contrast, evidence indicates that amplification, overexpression [2–6], or epigenetic deregulation [7] may have an important role in tumor progression. The results of a recent ToGA trial revealed that the anti-HER2 antibody trastuzumab improves the survival of patients with HER2-positive advanced gastric cancer [2]. The results of other studies suggest the potential efficacy of targeted therapy against gene alterations in gastric cancers [8–10], and several other therapeutic drugs against these RTKs, including EGFR, are currently under investigation [11].

Deng et al. [5] detected amplification of genes encoding RTKs, including KRAS, HER2, EGFR, MET, and FGFR2, in approximately 37 % of gastric cancers, and suggested that amplification of these genes may act as an oncogenic driver often related to poor prognosis.

EGFR resides on chromosome 7p12 and encodes a 170-kDa RTK that contributes to cancer progression by mediating cellular proliferation, migration, invasion, and metastasis [12]. EGFR GCN alterations occur in a wide range of cancers and serve as a marker of poor prognosis. Inhibitors of EGFR activity include specific monoclonal antibodies and small molecule tyrosine kinase inhibitors that are used for standard therapy of lung and colorectal as well as head and neck cancers [13–16]. Therefore, EGFR is a strong candidate for molecular-targeted therapy in gastric cancer. EGFR overexpression and gene amplification serve as effective biomarkers for predicting the clinical benefit of anti-EGFR monoclonal antibodies for treating colorectal [17] and lung cancer [18]. The recent REAL3 (panitumumab) [19] and EXPAND (cetuximab) [20] randomized controlled phase III trials failed to demonstrate the efficacy of the anti-EGFR monoclonal antibodies for treating gastric cancer, presumably because precise patient inclusion criteria were not set for the trials. Several studies suggest that EGFR overexpression or gene amplification may serve as a potential biomarker for the efficacies of anti-EGFR therapies in gastric cancer [9, 21–23].

We reported that EGFR overexpression detected using immunohistochemistry (IHC) is associated with poor outcome of gastric cancer after curative resection [4]. However, the incidence of EGFR overexpression reported in previous studies varies from 2 % to 81 %, probably the result of differences in antibodies, scoring systems, and observers [4, 21, 24–32]. In contrast, the frequency of EGFR gene amplification in gastric cancers is consistently less variable (2–7 %), although most of these studies lacked a sufficiently large number of patients or a detailed EGFR GCN examination [3, 5, 22, 24, 26, 29, 30]. The correlation between overexpression and EGFR GCN has not been firmly established.

To address this issue, in the present study, we evaluated EGFR GCN in a large set of samples acquired from patients with gastric cancer and show that it correlates with protein expression and prognosis.

Materials and methods

Patients and pathological specimens

Formalin-fixed, paraffin-embedded (FFPE) gastric cancer tissues from consecutive patients who underwent surgical resection at the National Cancer Centre Hospital East in Japan from January 2003 to July 2007 were used. Construction of tissue microarrays (TMA) using these tumors is described elsewhere [33]. In brief, two separate tumor areas were selected based on depth and tumor histology, and a tissue core of 2 mm in diameter was obtained from each of the areas from each patient. Serial 4-µm sections of TMA blocks were analyzed using hematoxylin and eosin (H&E) staining, IHC, and dual in situ hybridization (DISH). Clinicopathological parameters including age, sex, tumor location, macroscopic type, histological type, lymphatic invasion, venous invasion, tumor depth (T), lymph node metastasis (N), HER2 status, resection margin, and adjuvant chemotherapy were acquired from medical records. Disease stage was classified using the TNM criteria of International Union Against Cancer (UICC), seventh edition. Overall survival was determined from the dates of surgical resection to death. The Institutional Review Board of the National Cancer Centre approved the study protocol.

EGFR immunohistochemistry

All reagents and instruments for IHC were manufactured by Ventana Medical Systems (Tucson, AZ, USA). FFPE sections were stained with Ventana Benchmark XT using a primary antibody against EGFR (clone 5B7) and Ventana iVIEW DAB Universal Kit, according to the manufacturer’s protocol. Stained slides were examined under light microscopy and were interpreted by E.H. and T.K. Membranous staining was scored (0, 1+, 2+, or 3+) as follows: 0 = undetectable membrane staining or <20 cancer cells with any intensity; 1+ = at least 20 cancer cells with faint membranous staining; 2+ = at least 20 cancer cells with weak to moderate membrane staining; and 3+ = at least 20 cancer cells with strong membrane staining. If the two cores from the same tumors had different IHC scores, the higher score was employed as the final IHC result.

Dual in situ hybridization (DISH) detection of EGFR and CEN7

All regents and instruments for DISH were manufactured by Ventana. DISH analyses for EGFR were carried out with Ventana Benchmark XT, using DNA cocktail probes [EGFR and chromosome 7 centromere (CEN7)] according to the manufacturer’s protocol. The number of EGFR signals (silver ISH detection, black dots) and CEN7 signals (red ISH detection, red dots) of 20 representative cells were counted in each core, and the higher score of two cores from each sample was employed as the final result. The EGFR/CEN7 ratio was calculated by dividing the total number of EGFR GCN signals by that of the CEN7 signals. Amplification of EGFR was defined as EGFR/CEN7 ≥ 2.0. Chromosome 7 polysomy was defined as the average of CEN7 signals ≥3.0 per cell. Cancer tissue cores that had no more than 1 signal for either EGFR or CEN7 per cell were excluded from further analyses.

Statistical analyses

For the multiple subgroup analyses of overall survival, the hazard ratio (HR) and 95 % confidence interval (CI) within each subgroup were summarized and displayed in a forest plot using the Cox regression model. An analysis of overall survival between subgroups was performed using the Kaplan–Meier method, and differences were compared using the log-rank test. The clinical characteristics of the two groups were compared using the chi-square test, Fisher’s exact test, and Cramér’s V. Univariate and multivariate analyses were conducted using the Cox regression model. Variables for multivariate analysis were selected by a backward stepwise approach using a significance level of P < 0.05 for variables remaining in the model. All reported P values are two sided, and the level of significance was P < 0.05. All statistical analyses were performed using IBM SPSS statistics version 18 for Windows (IBM, Armonk, NY, USA).

Results

Characteristics of patients and TMA specimens

After examining 2,012 cores from 1,006 patients using IHC and DISH, we excluded 151 patients from further analysis because no gastric cancer cells or deficient DISH signals were observed, leaving 855 patients for further analyses. The median (range) follow-up period of the surviving patients was 61.2 (1–121) months. The profile of clinicopathological treatment-related characteristics of these patients is shown in Online Resource 1.

EGFR expression and prognosis

IHC scores for the 855 patients were as follows: 463 (54.1 %) were IHC 0, 251 (29.3 %) were 1+, 114 (13.3 %) were 2+, and 27 (3.2 %) were 3+. The hazard ratios of IHC 1+, 2+, and 3+ subgroups compared with those of IHC 0 are described in Online Resource 2a. IHC 2+ and IHC 3+ results were significant indicators of poor prognosis, and IHC 3+ was identified as the strongest prognostic value in overall survival curve estimates (Online Resource 2b).

EGFR GCN and prognosis

The median copy numbers of EGFR and CEN7 per cell were 1.9 and 2.3, respectively, and the median EGFR/CEN7 ratio was 0.90. Chromosome 7 polysomy was observed in 154 patients (Online Resource 1). We divided the patients into four grades according to EGFR GCN per cell: grade 1 (GCN < 1.5), grade 2 (1.5 ≤ GCN < 2.0), grade 3 (2.0 ≤ GCN < 2.5), and grade 4 (2.5 ≤ GCN). The hazard ratio of each EGFR GCN subgroup for grade 1 is described in Fig. 1a. There were stepwise increases in the hazard ratio with increases in the GCN grade, and GCN grade 4 (HR = 2.51, P < 0.001) was the most informative prognostic factor.

Fig. 1
figure 1

Definition of the cutoff value of EGFR gene copy number (GCN) gain in patients with gastric cancer. a Hazard ratios and 95 % confidence intervals (CIs) for overall survival according to EGFR GCN. b Optimal cutoff values determined using a receiver operating characteristics curve

Full size image

Receiver operating characteristic (ROC) analysis was performed to define the optimal cutoff value that discriminates patients with poor prognosis (Fig. 1b). The ROC curve identified 2.35 signals of EGFR per cell as the optimal cutoff value discriminating between favorable and poor prognosis for 5-year survival. For further examination, we defined 2.5 EGFR signals per cell as the cutoff value for the definition of GCN gain, and 194 patients were assigned to this category.

Among the 194 patients with GCN gain, the EGFR/CEN7 ratio was <2.0 for 165 patients, whereas EGFR was amplified in only 29. Among the 29 patients with EGFR amplification, 20 patients possessed >20 copies of EGFR per cell and formed a subgroup distinguished from the others (Fig. 2). Further, a clustered pattern of EGFR signals was observed in the nucleus for 27 of 29 patients with EGFR amplification (Fig. 3a). In the remaining 165 GCN patients without detectable gene amplification, chromosome 7 polysomy (Fig. 3b) was observed in 103 (53.1 %). The number of EGFR ISH signals without amplification or polysomy of the remaining 62 patients was slightly increased (Fig. 3c). The prognoses of patients with EGFR GCN gain were significantly worse compared with those without this gain. EGFR amplification was the strongest prognostic marker in the EGFR GCN subgroups (Fig. 4a). The prognostic value of EGFR GCN gain without amplification was still significant to predict poor clinical outcome from the absence of EGFR GCN gain.

Fig. 2
figure 2

Distribution of EGFR GCN per cell and EGFR/chromosome 7 centromere (CEN7) ratios according to immunohistochemical (IHC) scores of 855 patients with gastric cancer

Full size image
Fig. 3
figure 3

Representative images of GCN gain with amplification (a), GCN gain without amplification derived from polysomy of CEN7 (b), and GCN gain without amplification showing slightly amplified EGFR signals and an EGFR/CEN7 ratio <2.0 (c)

Full size image
Fig. 4
figure 4

Kaplan–Meier analysis of overall survival. a Comparison between patients without GCN gain (Non-gain, n = 661), with GCN gain but without amplification [Gain (non-amp), n = 165] and with both GCN gain and amplification [Gain (amp), n = 29] among 855 total patients. b Comparison between patients without GCN gain (Non-gain, n = 615) and with GCN gain (Gain, n = 99) among 714 patients with IHC scores of 0/1+. c Comparison between patients without GCN gain (Non-gain, n = 46) and with GCN gain (Gain, n = 68) among 114 patients with IHC scores of 2+. d Comparison between patients without heterogeneous GCN gain [Hetero GCN (−), n = 95] and with heterogeneous GCN gain [Hetero GCN (+), n = 70] among 165 patients with GCN gain but without amplification. e Comparison between patients without heterogeneous amplification [Hetero Amp (−), n = 5] and with heterogeneous amplification [Hetero Amp (+), n = 24] among 29 patients with amplification. f Comparison between patients with EGFR amplification observed in <50 % (n = 19), 50–80 % (n = 6), and ≥80 % (n = 4) of all tumor cells among 29 patients with amplification

Full size image

Correlations between EGFR expression and EGFR GCN gain

EGFR was amplified in all 27 patients with IHC 3+, and GCN gain was detected in 59.6 % (68/114) of IHC 2+ patients (Table 1). GCN gain was a significant indicator of poor prognosis in the IHC 0/1+ (P = 0.014; Fig. 4b) and IHC 2+ subgroups (P = 0.042; Fig. 4c).

Table 1 Correlation between immunohistochemistry (IHC) score and EGFR gene copy number (GCN) status
Full size table

Correlation between clinicopathological characteristics and EGFR GCN gain

The correlations between clinicopathological characteristics and EGFR GCN gain are summarized in Table 2. EGFR GCN gain was more prevalent for patients with characteristics as follow: male, age >65 years, macroscopic type 2 or 3, differentiated type, presence of lymphatic or venous invasion, advanced tumor depth, lymph node metastasis, advanced stage, HER2-positive, R2 of resection margin, and adjuvant chemotherapy.

Table 2 Comparison of clinicopathological characteristics according to EGFR GCN status in 855 patients with gastric cancer
Full size table

Multivariate analysis of prognostic factors for survival

Clinicopathological factors of 855 patients were analyzed using Cox regression to determine their prognostic significance. Thirteen factors were significant for overall survival as follows: age >65 years, macroscopic type 4, undifferentiated type, presence of lymphatic invasion, presence of venous invasion, T4 tumor depth, lymph node metastasis, TNM stage III/IV, R2 resection margin, IHC 2+, IHC 3+, EGFR GCN gain without amplification, and EGFR GCN gain with amplification (Table 3). The eight factors, which were identified using multivariate analysis of overall survival, correlated independently with worse prognosis as follows: age >65 years, macroscopic type 4, undifferentiated type, venous invasion, TNM stage III/IV, R2 of resection margin, EGFR GCN gain without amplification (HR = 1.53; 95 % CI = 1.11–2.10, P = 0.009), and EGFR GCN gain with amplification (HR = 2.46; 95 % CI = 1.44–4.21, P < 0.001) (Table 3). Note that IHC 2+ was not demonstrated as an independent indicator of poor prognosis. Further, IHC 3+ was excluded by multivariable analyses because of confounding GCN gain results with amplification. Even when survival analyses were adjusted for stage and resection margin, EGFR GCN gain was a significantly worse prognostic indicator in patients compared with the absence of GCN gain (Online Resource 3a/b).

Table 3 Hazards analysis of prognostic factors in 855 patients with gastric cancer
Full size table

Intratumoral EGFR GCN alteration heterogeneity

In the present study, a tumor was defined as heterogeneous if different EGFR GCN gain or amplification results were obtained from two separate TMA cores from the same tumor. Heterogeneity status of the EGFR GCN alteration was detected for 70 (42.4 %) of 165 patients with EGFR GCN gain and for 24 (82.8 %) of 29 patients with amplification. As shown in Fig. 4d, e, a heterogeneous EGFR GCN status tended to have better prognostic impact upon EGFR gene amplification as well as GCN gain tumors (P = 0.01 and P = 0.082, respectively). Among the 29 patients with amplification, the proportion of tumor cells exhibiting EGFR amplification in two TMA cores was <50 % in 19 patients (65.5 %), 50–80 % in 6 patients (20.7 %), and ≥80 % in 4 patients (13.8 %). The patients in whom ≥50 % of tumor cells exhibited EGFR amplification had a significantly poor prognosis (P = 0.006; Fig. 4f), and an increased proportion of cells exhibiting amplification was suggested to be associated with a poor prognosis.

Discussion

EGFR gene amplification results from an aberrant DNA replication and leads to a large number of gene copies integrated into either intrachromosomal tandem repeats or extrachromosomal double minutes and is a key mechanism of EGFR signaling activation [34]. In contrast, EGFR GCN gain is defined as GCN ≥ 2.5 per cell and results from various causes such as segmental chromosomal duplications or polysomy. EGFR GCN gain may accelerate EGFR-associated signaling, although the biological significance of EGFR GCN gain may differ from EGFR gene amplification in EGFR signaling activation.

We showed that EGFR amplification was a rare event (3.4 %) and indicated the worst prognostic impact. Although the criteria for EGFR amplification are not formally established and a wide range of EGFR/CEN7 ratios (1.7–6.0) were used in previous studies [3, 5, 22, 24, 26, 29, 30], the frequency of EGFR amplification did not vary among these studies, which may be attributed to the fact that the EGFR/CEN7 ratio is >10.0 in most patients with EGFR amplification, and a defined cutoff value would not influence the results (Fig. 2). For example, in the present study, 27 of 29 patients with EGFR amplification exhibited large clusters of signals in the nucleus.

Few studies have revealed a correlation between clinical outcome and EGFR GCN gain without amplification in gastric cancer. In the present study, we found that prognosis worsened as EGFR GCN increased. We defined the cutoff value of GCN gain as 2.5 EGFR signals per cell in the present study, and this value effectively delineated between poor and favorable prognoses. Thus, EGFR GCN gain without amplification was also a poor prognostic factor. The pathological characteristics of patients with EGFR GCN gain, including lymphatic invasion, venous invasion, tumor depth, and lymph node metastasis, were more severe and frequent in advanced stages, suggesting that EGFR GCN gain increases EGFR RTK signaling and contributes to tumor progression. Therefore, EGFR GCN gain may serve as a potential biomarker for guiding anti-EGFR therapies.

Evidence indicates that polysomy of chromosome 17 in patients with breast cancer may represent a clinical response to trastuzumab, even in patients without amplification of HER2 [35]. Polysomy of chromosome 17 indicates some degree of clinical benefit for the use of trastuzumab in gastric cancer as well [36]. Our data raise the possibility that chromosome 7 polysomy may represent a potential biomarker for EGFR therapy, although further studies are needed for corroboration.

No published data have been available to show a correlation between EGFR GCN and EGFR overexpression in patients with gastric cancer. Although there are no acknowledged consensus criteria for the histological determination of EGFR status in gastric cancer, some studies show that strong IHC staining of the tumor cell membrane (corresponding to IHC 3+ or IHC 2+/3+ in the present study) indicates poor prognosis [4, 26]. In the present study, we used a monoclonal anti-EGFR antibody (clone 5B7) raised against the functionally active intracellular domain of EGFR. When used for studies on colon cancer, this antibody exhibited higher sensitivity and specificity for EGFR overexpression than other anti-EGFR antibodies that bound to the external domain of the EGFR [37]. As expected, in the present study, EGFR amplification was detected in all IHC 3+ samples. The sensitivity and specificity of EGFR amplification corresponding to EGFR IHC 3+ were 100 % and 99.8 %, respectively. Moreover, we detected GCN gain in more than half the samples with IHC 2+, and this gain was a significant prognostic factor for the IHC 0/1+ and IHC 2+ subgroups. These results suggest that the evaluation of GCN gain is a more effective biomarker than the IHC score for identifying poor prognosis.

In the present study, we used the DISH assay to evaluate the EGFR and CEN7 copy numbers. Notably, in some patients, the number of EGFR signals per cell was less than 1, possibly because the sizes of the dots representing EGFR were occasionally smaller than those representing CEN7 and were thus more difficult to identify. Accordingly, the frequency of GCN gain observed in the present study was lower than the frequencies in previous reports that incorporated the fluorescence in situ hybridization (FISH) method [22]. Therefore, we employed a cutoff value of 2.5 EGFR signals per cell. Nevertheless, we assumed that evaluation of EGFR GCN status using DISH must have an advantage over evaluation using FISH, primarily because DISH particles can be evaluated via conventional light microscopy, facilitating the identification of tumor cells. In the present study, we showed that cells exhibiting EGFR amplification were distributed heterogeneously throughout tumors (82.8 %) and that an increased proportion of EGFR-amplified cells was associated with a poor prognosis. Therefore, it is important to examine the EGFR GCN status throughout the tumor area and to evaluate the proportion of tumor cells exhibiting EGFR amplification. DISH is recommended for HER2 testing by the ASCO/CAP 2013 guidelines for breast cancer [38] and proves advantageous because it enables examination of tumors exhibiting more heterogeneity, such as gastric cancer.

In conclusion, we showed here that EGFR amplification was relatively rare in gastric cancer, although EGFR GCN gain was observed in 22.7 % of the patients. EGFR GCN gain correlated significantly with the IHC results and may serve as a biomarker with increased sensitivity and specificity for predicting prognosis. EGFR GCN gain, especially if combined with amplification status and IHC score, may provide a superior method for distinguishing among patients who may benefit from therapies that target EGFR.