A subset of gastric cancers with EGFR amplification and overexpression respond to cetuximab therapy

A preclinical trial identified 4 of 20 (20%) gastric cancer (GC) patient-derived xenografts responded to cetuximab. Genome-wide profiling and additional investigations revealed that high EGFR mRNA expression and immunohistochemistry score (3+) are associated with tumor growth inhibition. Furthermore, EGFR amplification were observed in 2/4 (50%) responders with average copy number 5.8 and >15 respectively. Our data suggest that a GC subtype with EGFR amplification and overexpression benefit from cetuximab treatment.

inoculation. Then we set out to test a cohort of randomly selected 20 GC-PDXs in a clinical trial-like study to assess cetuximab activities by subjecting them to the drug treatment (50 mg/kg, intraperitoneally, IP) once weekly for 2 weeks. The original patient clinicopathological features, along with the model pathology confirmation, are summarized in Supplementary Table 1.
The tumor response to cetuximab is quantified by DT / DC 15 and summarized in Table 1. The tested GC-PDXs fall into two distinct categories according to the drug activities: 4 of 20 (20%) responded with nearly complete response ( DT / DC , 0) to cetuximab treatment; 16 of 20 (80%) did not, with partial or complete resistance (DT/DC . 30%). The representative tumor response curves are shown in the left column of Figure 1B. GA0152 and GA0075 are examples of cetuximab sensitive models, while GA0119 and GA0139 are resistant models. Our data clearly suggest that a subset of GC tumors can potentially benefit from cetuximab treatment.
About 50% responders display EGFR gene amplification. In order to discover potential predicting markers of cetuximab response, therefore, we performed molecular characterization of these models, including genome-wide copy number variation and transcriptome profiling, First, we interrogated copy number variation of GC-PDXs using Affymetrix genome-wide human SNP6.0 array and PICNIC (Predicting Integral Copy Numbers In Cancer) algorithm 15 . We found that EGFR copy numbers of all four responders are higher than most of those non-responders (Table 1, P 5 0.002). To further confirm this finding, we assessed EGFR gene copy number by realtime quantitative PCR (q-PCR) and found that all responders have copy number $4, while only 2 of 16 (12.5%) non-responders have copy number $4. The difference between these two group is significant (P 5 0.008). The highest value, 15 by SNP6 1 PICNIC analysis and 1040.9 by q-PCR, is from GA0152, which is also the best responder.
To further confirm the EGFR gene amplification, we further performed fluorescence in situ hybridization (FISH), a more accurate assay used to determine HER2 gene amplification for guiding anti-HER2 treatment for advanced GC in the clinical practice. At least 100 non-overlapping interphase nuclei were observed for the number of copies of EGFR. EGFR status was scored as the number of EGFR signals per nucleus. Our result demonstrated EGFR amplification in 2/4 (50%) responders with average copy number 5.8 (GA0075) and .15 (GA0152), respectively (Fig. 1A, Table 1). GA0152 was also with EGFR/CEP7 ratio .15. Thus, 2/4 (50%) responders could be predicted by EGFR amplification.
All responders display higher EGFR mRNA expression level. On the other hand, transcriptome profiling using Affymetrix HG-U219 GeneChip, revealed that all of the four responders expressed higher levels of EGFR mRNA expression than all 16 non-responders did (P 5 0.003) ( Table 1). EGFR gene expression was further quantified by q-RT-PCR (quantitative reverse transcription-PCR) against housekeeping gene GAPDH. Among the samples tested, 4 samples exhibited high EGFR mRNA levels (relative intensity $0.5, arbitrarily defined) were all responders, in contrast to the remaining models showing medium to low EGFR mRNA levels (relative intensity # 0.1) ( Table 1, Fig. 1A). The difference is significant (P 5 0.002). In particular, the highest value is from GA0152, with 10.5 by GeneChip analysis and 13 by q-RT-PCR, which can be attributed to the EGFR amplification mentioned above.
All responders display higher EGFR immunohistochemistry score. Then we performed EGFR immunohistochemistry (IHC), a clinically practical assay to determine HER2 expression for anti-HER2 treatment for GC. IHC demonstrated positive EGFR immunostaining in 12/20 (60%) models. Among them, 6/12 had staining intensity score of 11, 3/12 of 21, and 3/12 of 31. All responders were found EGFR IHC 31, while the non-responders displayed lower EGFR IHC score 0-21 (P 5 0.002) ( Table 1, Fig. 1A). The typical EGFR strong immunostaining (GA0152 and GA0075) is showed in Figure 1B. These results demonstrated that the EGFR high expression (in both mRNA and protein level) is correlated to the response to cetuximab.
Interestingly, few of the tested models, regardless responders or nonresponders, showed any aberrations with exception of GA0139 containing G13D KRAS mutation, GA0044 containing 327-329 deletion in PIK3CA, and GA0098 containing G545Y PIK3CA mutation (Table 1). Therefore, the non-response of GC xenografts to cetuximab apparently cannot be simply attributed to these oncogene mutations.

Discussion
Our data point to a positive correlation between cetuximab response in GC and the EGFR high expression at both mRNA and protein level, as well as EGFR gene amplification. This correlation is exemplified by GA0152 that has the highest EGFR mRNA expression, IHC score and gene amplification. The data seem to suggest the higher activity of EGFR via higher expression drives the oncogenic transformation in these tumors, and therefore its inactivation by cetuximab thus inhibits tumor growth. Overexpression of EGFR could be attributes to the gene amplification in two cases, however, the exact mechanism of EGFR high expression in the other two cases has yet to be investigated.
A recent phase II trial 8 , with cetuximab combined therapy for GC (European Clinical Trials Database number 2004-004024-12) showed association between higher EGFR copy number (defined as $4, 8 of 36 cases, 22.2%; including 1 amplification case $6 and FISH positive) and better overall survival 20 . Their clinical data seem to be consistent with our data in this mouse clinical trial that all responders display higher EGFR gene copy number $4 while only two (50%) are FISH positive. Our data also demonstrated the EGFR high expression in both mRNA and protein level is correlated to the response. However, since the mRNA expression of EGFR genes is not routinely assayed in the clinical samples, and IHC can be of controversy due to biological and technical factors, we recommend that the combination of FISH and IHC tests are suitable for predicting cetuximab efficacy as routine clinical practice, similar to the clinical practice of anti-HER2 treatment.
In summary, our study suggests that a GC subtype with high EGFR mRNA expression and IHC score 31 may benefit from cetuximab treatment, and the EGFR gene amplification by FISH can also accurately predict the responders with positive predictive value around 50%. These markers can be helpful for guiding future a potentially successful clinical trial and eventually as a patient stratification guide for clinical treatment.

Methods
Patient tumor samples and engraftment in immunocompromised mice. Freshly and surgically removed tumor tissues were obtained from the patients diagnosed as GC in Peking University Cancer Hospital through approval by the Institutional Review Boards of the hospital and the informed consents from all patients. The engraftment of patient tumor fragments into immunocompromised mice subcutaneously was previously described 15 . Briefly, the tumors were sliced into 3 3 3 3 3 mm 3 fragments and inoculated subcutaneously on the flank of mice (BALB/c nude, 6-to 8-weeks old female mice, Beijing HFK Bioscience Co., Beijing, China). The tumor growth was monitored twice weekly using a caliper. The established tumor models, called passage 0 or P0, were serially re-engrafted to maintain tumors in vivo. These subsequent passages were called P1, 2, 3… (,10). When tumors sizes reach 500-700 mm 3 (1/2 length 3 width 2 ), they were harvested for the next round of engraftment for serial passage or conducting studies of pharmacology, histopathology, immunohistology, cellular and molecular analysis. All procedures were under sterile conditions at Crown Bioscience SPF facility and conducted in strict    Evaluation of antitumor activity. When tumor volume reaches 100-150 mm 3 , the mice were randomly grouped into two groups of five mice with similar average tumor volume. Immediately after grouping, the control group was treated with vehicle (PBS, weekly intraperitoneal injection or IP for 2 weeks), and the treatment groups were injected with cetuximab (weekly IP injection for 2 weeks, 50 mg/kg, Merck KGaA). The tumor growth was monitored twice weekly, and DT / DC value was calculated for assessing tumor response to the treatment (DT 5 tumor volume change in the treatment group and DC 5 tumor volume change in the control group). The total number of the mice for xenograft is 200 (10 mice/model for 20 PDX models).
EGFR IHC analysis of GC tumors. Standard immunohistochemistry (IHC) was used to analyze tumor tissues from the PDX xenograft models. Briefly, the tissues were fixed in 10% neutral buffered formalin and embedded in paraffin per standard histological procedures. After deparaffinization and rehydratation, 3-mm thick tissue sections were pretreated in 0.01 M sodium citrate, pH 6.0 solution at 95uC for 30 min, followed by staining with rabbit anti-human EGFR antibody (Cell Signaling, Boston, USA) at final dilution 15200. Positive staining was detected using Detection System HRP Polymer Kit (Lab Vision, Fremont, USA). DAB was used as the chromogenic substrate, and sections were counterstained with Gill's hematoxylin (Fisher Scientific, Fair Lawn, NJ). The test specimens were then scored independently by three investigators in a blinded fashion per following criteria recommended by Shia et al in 2005 21 : Score 0 is when there was no specific membrane staining within the tumor, and positive when there was any staining of tumor cell membrane above background level. The positive cases were further classified into 11, 21 and 31 based on the staining intensity of the membrane.
Areas of most intensity were identified by scanning tumor sections at low power (1003), and then images were photographed at high magnification (4003) using Olympus BX51 microscopy system with DP71 digital camera (Olympus, Melville, NY).
Gene expression profiling and gene copy number analysis of GC-PDX. Fresh GC-PDX tumor tissues were collected from the tumor-bearing mice, snap-frozen and stored at 280uC before being used for genetic and genomic analysis. For gene profiling analysis, the total RNA was isolated from the frozen tissues using Trizol (Invitrogen, Carlsbad, CA) per the manufacturer's instructions, and purified using RNeasy mini columns (Qiagen). RNA quality was assessed on a Bioanalyzer (Agilent). Only RNA samples with high quality (RIN . 8) were used for expression profiling assays on Affymetrix HG-U219 array plates following standard protocol (http://media.affymetrix.com/support/downloads/manuals/3_ivt_express_kit_ manual.pdf). Raw CEL data sets of all samples were normalized by RMA algorithm. Probe set intensity was expressed as log(2) transformed values. For CNV assay using Affymetrix SNP6.0 chips, genomic DNA was isolated and purified using Genomic DNA Tissue and Blood Isolation Kit (Qiagen) following manufacturer's instruction. DNA processing and chip hybridization were performed following standard Affymetrix protocol (http://media.affymetrix.com/support/downloads/manuals/ genomewidesnp6_manual.pdf). Raw CEL data were QC-ed and filtered to remove low call-rate samples, and gene copy number analysis were performed by PICNIC and/or PennCNV methods.
For all of the samples, the relative EGFR gene expression level was determined by quantitative RT-PCR. Extracted mRNA was subjected to amplification using human EGFR specific primers by TaqMan q-PCR. The human GAPDH gene was used as a reference. TaqMan probes and primers for EGFR (assay ID: Hs01076078_m1), GAPDH (Assay ID: Hs99999905_m1) were obtained from Applied Biosystems. The raw data generated by the system were processed using the DCT relative quantification. DCT 5 (CT value of target gene) -(CT value of reference gene). DCT values were then converted into intensity value (relative mRNA level 5 2 ' (2DCT).
Also, EGFR gene copy numbers were determined by quantitative PCR. Briefly, the same genomic DNAs were subjected to amplification by TaqMan q-PCR. The primers for EGFR (assay ID: Hs04960197_cn) and RNase P as endogenous reference (part number 4401631) were purchased from Applied Biosystems. The raw data was transferred to CopyCaller software and analyzed.
FISH. Three micrometer thick tissue sections were treated with the procedure provided by fluorescence in-situ hybridization (FISH) detection kit (DakoCytomation, Glostrup, Denmark). Samples were placed in pretreatment solution for 30 min at 96uC, and digested with pepsin solution for 30 min at room temperature. Dual-color, dual-target FISH assays were done with the EGFR Spectrum Orange/CEP7 Spectrum Green Probe (Vysis, USA). Tissue sections, covered with 10-mL probe solution, were incubated at 75uC for 5 min to co-denature the EGFR and CEP7 (chromosome seven a-centromeric) probes and allowed to hybridize overnight at 37uC. Co-denaturation and hybridization were done sequentially. Posthybridization stringency wash was done in a water bath at 65uC for 10 min. After washing twice and drying at room temperature for 15 min, tissue sections were covered with 496-diamidino-2-phenylindole (DAPI II, Vysis, USA) for chromatin counterstaining.
Analyses were done with a fluorescence microscope (Zeiss Axiophot, Germany) equipped with a Metachrome II cooled-charged device camera (Zeiss, Germany).
EGFR was visualized as a red signal with a standard TRITC (tetramethyl rhodamine isothiocyanate) filter, CEP7 as a green signal with a FITC (fluorescein isothiocyanate) filter, and nuclei as a blue signal with a DAPI filter. Representative images of samples were acquired and then analyzed.
Two independent observers scored at least 100 non-overlapping interphase nuclei for the number of copies of EGFR and CEP7 by use of predefined scoring guidelines. EGFR status was scored as the number of EGFR signals per nucleus and as the ratio of EGFR signals to CEP7 signals. Negative controls consisted of a cultured retinal pigment epithelial (RPE) cell line; the control for amplified EGFR was the A431 cell line derived from human epidermoid carcinoma. Amplification was defined as the presence of 5 or more signals per nucleus, i.e., EGFR copy number $5.
Polymerase chain reaction was performed in 50 mL reaction mixtures containing: 100 ng of genomic DNA, 5 mL 103 PCR buffer, 0.2 mM each of primers, 0.2 mM 43 dNTPs and 1 mL TaqE. Reaction was carried out for 40 amplification cycles. The amplified PCR products were gel purified and sequenced by Sanger Automated Sequencer (ABI). The specificity of the primers to human genes had been assured by BLAST search. Sequencing data alignment analysis and mutation identification was performed using BioEdit software.
Statistical analysis. The nonparametric Mann-Whitney U test was applied for comparing of the profiling data of two groups, i.e., non-responders and responders. In all analyses, p , 0.05 was considered statistically significant. The statistical analysis was carried out with SPSS V13.0 software (SPSS Inc., Chicago, IL, USA).