OUP user menu

HER2 Amplification in Gastroesophageal Adenocarcinoma
Correlation of Two Antibodies Using Gastric Cancer Scoring Criteria, H Score, and Digital Image Analysis With Fluorescence In Situ Hybridization

Oana M. Radu MD, Tyler Foxwell, Kathleen Cieply MSL, Sarah Navina MD, Sanja Dacic MD, Katie S. Nason MD, Jon M. Davison MD
DOI: http://dx.doi.org/10.1309/AJCPXQVS6YGHPDCY 583-594 First published online: 1 April 2012


We assessed 103 resected gastroesophageal adenocarcinomas for HER2 amplification by fluorescence in situ hybridization (FISH) and 2 commercial immunohistochemical assays. Of 103, 30 (29%) were FISH-amplified. Both immunohistochemical assays had greater than 95% concordance with FISH. However, as a screening test for FISH amplification, the Ventana Medical Systems (Tucson, AZ) 4B5 antibody demonstrated superior sensitivity (87%) compared with the DAKO (Carpinteria, CA) A0485 (70%). Of the cases, 28 were immunohistochemically 3+ or immunohistochemically 2+/FISH-amplified with the 4B5 assay compared with only 22 cases with the A0485 assay, representing a large potential difference in patient eligibility for anti-HER2 therapy. Cases with low-level FISH amplification (HER2/CEP17, 2.2–4.0) express lower levels of HER2 protein compared with cases with high-level amplification (HER2/CEP17, ≥4.0), raising the possibility of a differential response to anti-HER2 therapy. The H score and digital image analysis may have a limited role in improving HER2 test performance.

Key Words:
  • Esophagus
  • Stomach
  • Adenocarcinoma
  • HER2
  • Fluorescence in situ hybridization
  • FISH
  • Immunohistochemistry
  • Image analysis

Human epidermal growth factor 2 (HER2, ERBB2) is a membrane-associated receptor that dimerizes with HER family members and transduces extracellular signals to RAS-MAP kinase and PI3 kinase-AKT intracellular signaling networks.1 HER2 is overexpressed in several types of tumors, including breast2 and gastroesophageal (GE) adenocarcinoma.3,4 Protein overexpression is associated with genetic amplification of segments of chromosome 17 ranging from less than 1 megabase up to several megabases5 that may form tandem duplications on chromosome 17 or double-minute chromosomes.6 Clinically, it has been established as a predictive biomarker for HER2-targeted therapies.7

Trastuzumab (Herceptin) is a targeted anticancer therapy that functions by binding to the HER2 receptor, blocking receptor dimerization, facilitating immune recognition of tumor cells through antibody-dependent cell-mediated cytotoxicity, and causing endocytic degradation of cell surface–localized HER2 receptors.1 The recent demonstration that trastuzumab and conventional chemotherapy significantly improve survival for patients with advanced HER2+ gastric and GE junction adenocarcinoma over chemotherapy alone in the phase 3 ToGA trial8 has generated a need for assays that reliably identify HER2+ cases.

In the ToGA trial, almost all carcinomas were centrally evaluated by fluorescence in situ hybridization (FISH) and the HercepTest immunohistochemical assay (DAKO, Carpinteria, CA) for enrollment in the trial. A study conducted in advance of the clinical trial established unique criteria for scoring the immunohistochemical assay for gastric cancers on a 0+ to 3+ ordinal scale. These criteria were modified from breast cancer criteria to account for common, idiosyncratic staining characteristics of GE cancers (eg, basolateral HER2 localization and heterogeneous staining).9 Any patient with a tumor with HER2 amplification by FISH or 3+ immunohistochemical staining was eligible for enrollment in the trial.8

A post hoc subgroup analysis of the ToGA trial participants demonstrated that the greatest improvement in overall survival with trastuzumab was in the immunohistochemically 3+ or immunohistochemically 2+/FISH-amplified cohorts. No significant survival benefit from trastuzumab was observed in patients with immunohistochemically 0+ or 1+/FISH-amplified tumors.8 The US Food and Drug Administration has approved the use of trastuzumab for the treatment of metastatic GE carcinomas that meet the ToGA trial inclusion criteria,10 whereas the European Medicines Agency has approved trastuzumab for immunohistochemically 3+ or immunohistochemically 2+/FISH-amplified tumors only.11

The European Medicines Agency and the College of American Pathologists (CAP) recommend testing cases first with immunohistochemical studies and subsequent FISH testing for cases that are indeterminate (2+ immunohistochemical score).11,12 Cases that are immunohistochemically 3+ are considered positive, and cases that are immunohistochemically 0+ or 1+ are considered negative and, therefore, not eligible for treatment. This immunohistochemically based HER2 screening algorithm is a pragmatic strategy used by many laboratories owing to the relative labor-intensiveness of in situ hybridization, even though trastuzumab therapy is approved for all patients with FISH+ GE cancers (regardless of immunohistochemical score) in the United States. An immunohistochemistry-first screening protocol demands high concordance of the immunohistochemical and FISH assays. The optimal immunohistochemical screening assay should also demonstrate high sensitivity and low false-negative rates for FISH amplification to ensure that all patients who may benefit from anti-HER2 therapy are identified while limiting the number of patients who are needlessly exposed to potential side effects.

For this report, we tested 103 surgically resected GE adenocarcinomas for HER2 amplification by dual-color FISH with probes targeting the HER2 locus and the chromosome 17 centromere. To directly compare the ability of 2 commercial immunohistochemical assays to detect FISH-amplified tumors, we independently tested all cases with the Ventana 4B5 monoclonal antibody (Ventana Medical Systems, Tucson, AZ) and the DAKO A0485 polyclonal antibody (DAKO). We scored the cases using the gastric cancer criteria developed by Hofmann et al9 for the ToGA trial. Because immunohistochemical scoring and cutoffs can significantly influence its performance characteristics, we evaluated whether alternative scoring methods could improve the detection of FISH-amplified cases including an H score that varies with staining intensity and fraction of tumor cells with positive staining and digital image analysis. A secondary aim of this study was to explore the correlation of protein expression level with the level of HER2 FISH amplification.

Materials and Methods

Case Selection and Clinicopathologic Data Collection

We evaluated 103 cases of adenocarcinoma of the GE junction, distal esophagus, or stomach treated between 2000 and 2011, for HER2 status by FISH and immunohistochemical analysis. Of 103 patients, 99 underwent surgical resection of the primary tumor with lymph node dissection. The remainder of the patients underwent surgical exploration and resection of distant or regional metastatic disease. Only surgically resected tumors (n = 95) and surgically resected metastases (n = 8) were evaluated for HER2 status in the study; mucosal biopsies were not included. At least 1 H&E-stained section of the tumor was reviewed for each case to confirm the diagnosis of adenocarcinoma. Pathology reports and intensive chart and electronic medical record review were used to abstract pathologic stage, tumor location, and treatment history.

This study was approved by the university institutional review board (protocol PRO10040154).

Immunohistochemical Staining and Scoring

We used 4-μm sections of formalin-fixed, paraffin-embedded tumor for immunohistochemical detection of the HER2 protein. All tumors were immunostained with 2 different antibodies: the anti-HER2 rabbit monoclonal antibody (clone 4B5, Ventana) and the anti-HER2 rabbit polyclonal (catalog number A0485, DAKO). All staining was performed on the Ventana BenchMark XT automated staining platform with the iVIEW DAB (3,3′-diaminobenzidine tetrahydrochloride) detection kit (Ventana) according to manufacturer’s instructions. Slides stained with the Ventana 4B5 antibody were counterstained with Ventana hematoxylin II in preparation for image analysis with the Ventana Image Analysis System (VIAS; Ventana); slides stained with the A0485 polyclonal antibody were counterstained with Ventana hematoxylin I. An HER2-amplified breast carcinoma served as a positive control sample for 3+ staining, and a nonamplified carcinoma served as a negative control sample.

Published, validated guidelines were used by the pathologist (J.M.D.) to score HER2 staining in surgical specimens,9,13 and only invasive carcinoma was scored. A score of 0+ was assigned for cases with less than 10% discernible membranous staining; 1+ for weak and/or incomplete basolateral or membranous staining generally requiring at least ×200 magnification to confirm positive staining; 2+ for weak to moderate, continuous basolateral or membranous staining that was not readily apparent at low (×40) magnification; 3+ for cases with strong basolateral or membranous staining that was readily apparent at low magnification. Only membranous staining was scored. (Granular cytoplasmic staining, even when strong, was ignored.) For cases with heterogeneous staining patterns, the final score was based on the highest score comprising at least 10% of the tumor even if another pattern predominated. The same guidelines were used for the A0485 and 4B5 antibodies. Representative examples of each score are depicted in Image 1. Each antibody was scored without knowledge of the results obtained with the other antibody and the FISH results.

For cases stained with the Ventana 4B5 monoclonal antibody, we estimated the fraction of tumor cells in each case that had 0+, 1+, 2+, and 3+ intensity to allow the calculation of an H score. The H score was calculated based on the following formula: HScore=(%Tumor1+)+2(%Tumor2+)+3(%Tumor3+)

The H score can, therefore, range from 0 (cases with absent staining) to 300 (cases with 100% 3+ staining).

Quantification of HER2 Staining by Digital Image Analysis

The VIAS was used to quantify HER2 staining for each 4B5-stained case. Three representative (236.8 × 181.6-μm), nonoverlapping images were captured at ×20 magnification. Areas were selected for image acquisition by the pathologist after scoring the case on the 0 to 3+ scale. Areas that were most representative of the 0 to 3+ score were selected for digital image acquisition and analysis. For heterogeneous tumors, the region with the highest (0–3+) score representing at least 10% of the tumor was imaged. The default VIAS HER2/neu 4B5 image analysis algorithm settings were used to identify tumor cells and quantify the extent and intensity of membranous DAB staining. Each image was reviewed by the operator to ensure accurate identification of tumor epithelial cells in the field, exclude staining artifacts, and select epithelial cells that were not recognized by the image analysis algorithm. The continuous scaled score generated by the VIAS algorithm ranges from 0 to 3.5. This was recorded along with the ordinal 0 to 3+ score that is based on the algorithm’s default cutoffs (0+ for a score <0.5; 1+ for a score of 0.5–1.5; 2+ for a score of >1.5–2.5; 3+ for a score >2.5).

Fluorescence In Situ Hybridization

We used 5-μm sections of formalin-fixed, paraffin-embedded tumor for hybridization. The PathVysion HER2 dual-color probe, which combines a 190-kbp SpectrumOrange-labeled HER2 probe and SpectrumGreen-labeled chromosome 17 centromere probe (CEP17; Abbott Molecular, Des Plaines, IL), was used for gene and chromosome copy number detection. Nuclei were counter-stained with DAPI I (Abbott Molecular). Hybridization and analysis were performed as previously described.14 At least 60 tumor cells were analyzed for each case: counts for each probe (HER2 and CEP17) are recorded for each cell/nucleus, allowing for the calculation of average counts per cell and a precise calculation of the HER2/CEP17 ratio for each cell and for the population as a whole. The primary definition of amplification used in this study was an HER2/CEP17 ratio of 2.2 or more. However, other cutoffs were also examined, and we defined low-level amplification as an HER2/CEP17 of 2.2 to 4.0 and high-level amplification as an HER2/CEP17 ratio of 4.0 or more (examples shown in Image 1). FISH was scored without knowledge of the immunohistochemical result.

Statistical Analysis

Statistical analysis was performed with SPSS software, version 18 (IBM, Armonk, NY). A P value less than .05 was considered statistically significant. Statistical tests are specified in the tables and text.


Patient Characteristics

Table 1 compares HER2-amplified and nonamplified GE adenocarcinomas with respect to several clinicopathologic variables. Of 103 cases, 91 (88%) were located at the GE junction. The remainder were located in the stomach (Table 1). Neoadjuvant therapy was administered to 39 (38%) of 102 patients, but no significant difference was seen in the prevalence of FISH amplification between induction-naive and treated patients who underwent surgical resection. HER2-amplified cancers in our cohort had smaller mean tumor size and tended to be well or moderately differentiated (Table 1).

Prevalence of HER2 FISH Amplification

By using the CAP definition of HER2 amplification for breast cancer (HER2/CEP17 ≥2.2), we found a 29% prevalence of amplification in our cohort of patients. Low-level amplification (HER2/CEP17, 2.2–4.0) was seen in 10 cases (10%) and high-level amplification (HER2/CEP17, ≥4.0) in 20 (19%). Borderline amplification (HER2/CEP17, 1.8–2.2) was seen in only 5 cases (5%). Including cases with borderline amplification, a total of 32 cases (31%) had an HER2/CEP17 ratio of 2.0 or more.

Image 1

Examples of HER2 fluorescence in situ hybridization (FISH) and immunohistochemical staining. Dual-color FISH with probes for chromosome 17 centromere (CEP17, green) and HER2 region (red) showing no amplification (A), low-level amplification (B), and high-level amplification (C). Immunohistochemical stains scored as 1+ (D), 2+ (E), and 3+ (F). G, A case that was scored 1+ in approximately 60% of the tumor with focal (5%) 2+ staining (H score = 70) was amplified by FISH (HER2/CEP17 = 3.1). H, A case that was scored 2+ in 10% (H score = 20) of the tumor cells was not amplified by FISH (HER2/CEP17 = 1.9; 120 cells scored). I, A case that was heterogeneous and scored 1+ in 40% and 2+ in 10% of cells (H score = 60) was amplified by FISH (HER2/CEP17 = 3.6) (AC, ×1,000; DF, H, and I, ×400; G, ×200).

HER2 FISH Amplification and Protein Expression: Ventana 4B5 vs DAKO A0485

The 2 antibodies had different performance characteristics. With the A0485 assay, 15 cases (15%) were scored 3+; another 9 cases (9%) were scored 2+ Table 2. With the 4B5 assay, 22 cases (21%) were scored 3+ and another 18 cases (17%) were scored 2+. More than 90% of 3+ cases with either antibody were amplified by FISH (Table 2). When comparing cases with 2+ immunohistochemical scores, a significantly higher proportion of A0485 2+ cases (7/9 [78%]) was amplified in comparison with 4B5 2+ (6/18 [33%]; Table 2; P = .046; Fisher exact test).

The concordance with FISH was more than 95% for both antibodies Table 3. However, when 3+ and 2+ immunohistochemical scores are considered positive, the 4B5 antibody is a more sensitive test for detecting FISH-amplified cases (Table 3).

Table 4 shows that there is only moderate agreement between A0485 and 4B5 immunohistochemical scores. Overall, 60 (58%) of 103 cases had identical scores when independently evaluated by 1 reviewer. Agreement was perfect for all 15 A0485 3+ cases, also scored as 3+ with the 4B5 antibody. However, only 22% of A0485 2+ cases (2/9) and 50% of A0485 1+ cases (12/24) had identical scores with the 4B5 antibody.

In general, use of the 4B5 antibody resulted in a higher immunohistochemical score for a given case. It is important to note that there were 16 cases of A0485 0+ or 1+ that were scored as 2+ with the 4B5 antibody. Of these, 5 (31%) were amplified by FISH. Of the 7 A0485 2+ cases that were scored as 3+ with the 4B5 antibody, 6 (86%) were amplified by FISH.

In our cohort, a larger percentage of cases would be deemed HER2+ and the patients eligible for trastuzumab therapy if tested with 4B5 immunohistochemical studies and FISH (28/103 [27%]) compared with 21 (20%) of 103 with a combination of A0485 immunohistochemical studies and FISH (Table 3). Of the 6 additional eligible cases identified by 4B5 staining, 3 had low-level amplification; 2 had high-level amplification, and 1 had no amplification of HER2 by FISH.

View this table:
Table 1
View this table:
Table 2

Differences in Immunohistochemical Staining Between Low- and High-Level HER2 Amplification

Figure 1 depicts the relationship between gene amplification level and immunohistochemical score with both antibodies. As expected, these data show a clear positive association between amplification level and immunohistochemical score. As the HER2/CEP17 increases, the proportion of cases with 2+ or 3+ immunohistochemical scoring rises. Differences in immunohistochemical score with both antibodies across all HER2/CEP17 groups were highly significant (P < .001; Fisher exact test). The same held true when comparing cases with low-level vs high-level amplification (Figure 1). To illustrate this point, 85% (17/20) of high-level amplified cases scored 3+ with the 4B5 antibody compared with 30% (3/10) low-level amplified cases; the differences were even greater with the A0485 antibody (70% 3+ for high-level vs 0% 3+ for low-level amplification).

HER2/Nucleus Ratio and Chromosome 17 (CEP17) Hyperploidy

Breast cancers with an HER2/nucleus 6 or more can be considered amplified under the CAP guidelines for HER2 testing if there is no chromosome 17 copy number reference.15 In our series of 103 cases, all 24 tumors with HER2/nucleus of 6 or more by FISH were amplified by the HER2/CEP17 cutoff of 2.2. Furthermore, 22 of 24 had immunohistochemical scores of 2+ or 3+ with the 4B5 antibody (18 were 3+).

View this table:
Table 3

Evidence of low-level chromosome 17 copy number gain was fairly common: 30% of cases had a mean CEP17/nucleus (CNR) of more than 2.0; 13% had a mean CNR of 2.5 or more; but only 2% of cases had mean CNR of 3.0 or more. There was no correlation between levels of protein expression and chromosome 17 copy number (data not shown).

View this table:
Table 4
Figure 1

HER2/CEP17 Groups stratified by immunohistochemical score (4B5 and A0485). A and B, Cases are grouped by HER2/CEP17 ratio along the x axis with the number of cases in each group indicated below. The bars depict the percentage of cases within each group receiving a given immunohistochemical score (0+ to 3+). The difference in the immunohistochemical scores across all groups is significant (P < .001; Fisher exact test) with both antibodies. When comparing only low-level and high-level amplified cases, the differences in immunohistochemical score remained significant with both antibodies (P values in figure; Fisher exact test).

When we examined the relationship between HER2 protein expression level by immunohistochemical studies and HER2 copy number, HER2/CEP17 increased significantly with increasing immunohistochemical score. The mean ± SD HER2/CEP17 increased from 1.3 ± 0.4 in cases with an immunohistochemical score of 0+ to 22.5 ± 13.4 in 3+ cases when using the A0485 antibody (P < .001; Kruskal-Wallis test) and from 1.3 ± 0.3 to 17.1 ± 13.9 when using the 4B5 antibody (P < .001; Kruskal-Wallis test). In summary, our results confirm the belief that HER2 protein overexpression is associated with a relative increase in HER2 gene copy number rather than a gain of chromosome 17.

Alternative Scoring of HER2: Use of the H Score to Predict FISH Amplification and Quantify Heterogeneity of Protein Expression

We calculated an H score for cases stained with the Ventana 4B5 antibody to correlate this alternative HER2 protein expression score with HER2 gene amplification detected by FISH.

The H scores ranged from 0 to 300. As expected, the mean H score increased as the 0 to 3+ scaled score increased: the 0+ cases had a mean score of 1 ± 2, the 1+ cases had a mean score of 39 ± 26, the 2+ cases 86 ± 37, and the 3+ cases 257 ± 50. Only 36 (35%) of 103 of tumors exhibited a single pattern of HER2 staining, and all of these homogeneously stained cases had absent (0+) staining or were diffusely positive (3+). Cases classified as 2+ were the most variable: 89% (16/18) of 2+ cases had 3 or more intensity patterns (eg, 20% of the tumor 2+, 40% 1+, and 40% 0+).

We also found a significant positive association between increasing H score and prevalence of FISH amplification Table 5. By using receiver operator characteristic curve analysis of our cases, we identified cutoffs of 50 and 175 with optimal sensitivity and specificity characteristics. Only 3% of cases with an H score less than 50 had amplification by FISH, whereas 90% of cases with an H score of more than 175 had amplification. This equates to a 95% concordance with FISH results (Table 3). All 13 cases with an H score of more than 250 were amplified, and 12 (92%) of 13 had high-level amplification (HER2/CEP17, ≥4.0).

Because 13% of cases with 1+ immunohistochemical scores are amplified by FISH, we examined whether the H score could predict which 1+ cases were amplified. We found that 1+ cases with an H score of more than 50 trended toward a higher likelihood of amplification (Table 5; Image 1). Similarly, 6 (40%) of 15 immunohistochemically 2+ cases with an H score of more than 50 were amplified, but none of the 3 immunohistochemically 2+ cases with an H score less than 50 were amplified (Table 5; Image 1). Neither trend was statistically significant. Additional cases will be required to evaluate these apparent trends.

Alternative Scoring of HER2: Digital Image Analysis

The VIAS image analysis algorithms were developed to score HER2 immunohistochemical staining with the Ventana 4B5 antibody in breast cancer. We evaluated the agreement between VIAS score and pathologist score and found that there was substantial agreement between pathologist and image analysis results Table 6. Overall, 95 (92%) of 103 cases had exact agreement. Cases scored by the pathologist as 1+ were the least likely to agree with the VIAS score (24/30 [80%]; Table 6). None of the 5 cases that were scored 1+ by the pathologist but 2+ by VIAS had gene amplification by FISH. The concordance of the VIAS score with FISH amplification was identical to the pathologist 4B5 score (Table 3).

View this table:
Table 5


The results of the ToGA trial have demonstrated that it is possible to improve survival in patients with advanced HER2+ gastric and GE junction adenocarcinoma by combining conventional chemotherapeutic agents with the humanized monoclonal antibody trastuzumab. Moreover, exploratory analysis of the ToGA phase 3 clinical trial data suggests that higher levels of HER2 expression are associated with a greater survival benefit with trastuzumab.8

The primary aims of our study were to evaluate the relative performance of 2 Food and Drug Administration–approved antibodies and explore alternative or complementary immunohistochemical scoring criteria and cutoffs to identify cases with genetic amplification of the HER2 region that may benefit from trastuzumab therapy.

The total prevalence of HER2 gene amplification using the ToGA trial definition (HER2/CEP17, ≥2.0) in our study was 31%. This result is consistent with the prevalence of amplification reported in other studies that have a similarly high proportion of GE junction or esophageal adenocarcinomas.5,1621 In these studies, prevalence varies fairly widely, from 15% to 32%. Our study, which examined a cohort composed of more than 90% GE junction adenocarcinomas, agrees with the largest study that performed immunohistochemical studies and FISH and found 32% of GE junction adenocarcinomas positive for HER2 amplification.20

One important comparison we have made is between the Ventana 4B5 monoclonal antibody and the DAKO A0485 polyclonal antibody. It bears emphasis that we did not use the HercepTest protocol; rather, we adapted the A0485 polyclonal antibody to the automated staining equipment in our laboratory (see “Materials and Methods”). One pathologist scored cases using the validated 0+ to 3+ scoring system9,13 that has been endorsed by the CAP and was unaware of the FISH and immunohistochemical results of the other test. In general, cases with discrepancies between the 4B5 and the A0485 assays received a higher score in the 4B5 assay (Table 4). This finding is consistent with a prior comparative study of 27 gastric cancers13 and of 130 breast cancers.22

View this table:
Table 6

Although concordance with FISH was similar between the A0485 and 4B5 assays (Table 3), the 4B5 assay had more favorable properties as a screening test for amplification owing to its higher sensitivity and lower false-negative rate (Table 3). The end result of “screening” this population of 103 patients with the A0485 assay would be the identification of 22 immunohistochemically 3+ or immunohistochemically 2+/FISH+ tumors eligible for trastuzumab therapy. However, with the 4B5 assay, 27% more tumors (n = 28) would be scored as eligible for trastuzumab (Table 3). Of the 6 additional cases defined as eligible, 1 had no amplification and 3 had low-level amplification, raising the possibility that use of the more sensitive 4B5 antibody to screen tumors for HER2 amplification will result in the treatment of tumors that are less likely to respond. This possibility warrants further study.

We evaluated the H score as a complementary method for scoring HER2 immunohistochemical studies and found that by using H score cutoffs of 50 for negative staining and 175 for positive staining, we could modestly improve performance characteristics of the 4B5 assay for FISH amplification (Table 3). The H score accounts for intensity and extent of staining. The simpler 0+ to 3+ scoring method has the likely advantage of greater interobserver reproducibility13 but is affected by substantial loss of information encoded in the extent of staining that may help predict which cases are more likely to be amplified. Based on our data, cases scoring between 50 and 175 could be considered indeterminate and undergo confirmatory FISH, regardless of the conventional 1+ to 3+ score. If this approach were taken with our cohort of cases, 23 would undergo confirmatory FISH after initial 4B5 immunohistochemical studies (a 28% increase over the number of 2+ cases), resulting in the identification of 2 additional treatment-eligible tumors (a 7% increase, Table 3). These H score cutoffs require validation in an independent cohort to truly assess their performance relative to the conventional system.

When scoring a surgically resected tumor (rather than a biopsy sample), the CAP has adopted a 10% cutoff for positive staining following the recommendations of 2 often-cited validation studies.9,13 This 10% cutoff may be specific to the HercepTest assay, as our data suggest that, with the Ventana 4B5 assay, cases with an H score less than 50 are very unlikely to have HER2 amplification by FISH (Table 5). The possible exception to this are cases with focal 3+ staining that have been reported to show clones of cells with genetic amplification.9

The H score data we present give a unique quantitative picture of the heterogeneity of HER2 expression in GE adenocarcinomas with the 4B5 assay that has been previously observed.19 Only a third of the cases in our study had uniform staining across the entire tumor section—all were scored 0+ or 3+. The cases we scored 2+ had considerable heterogeneity with almost 90% showing areas with 0+ and 1+, as well as rare cases with focal 3+ staining (<10% of cells). We also observed considerable staining heterogeneity with the A0485 assay, but we did not quantify it. We did not specifically address the question of molecular cytogenetic heterogeneity in our cases, although this has been reported as well.23

We evaluated a digital image analysis system in this study as another alternative scoring method that may help identify cases with HER2 amplification. The exact agreement on score seen in more than 92% of cases suggests that the proprietary scoring algorithms that were designed for scoring HER2 staining of breast cancer function reasonably well with the scoring of GE adenocarcinoma. Despite the agreement we observed, it may be possible to further optimize the scoring algorithm for assigning 0+ to 3+ ordinal scores.

The most striking difference between VIAS and pathologist scores was a 22% increase in the number of 2+ scores that would require FISH in the setting of immunohistochemical screening. In our experience, a challenging aspect of scoring HER2 immunohistochemical studies is differentiating weak 2+ from 1+ staining. This is an important distinction because a score of 1+ may not undergo FISH and would not qualify a patient for a trial of trastuzumab. A well-validated digital image analysis algorithm could have a very useful role in this context, as the distinction between 1+ and 2+ is based on a judgment of the intensity and completeness of membrane staining—a judgment that is theoretically well-suited to digital image analysis. Based on our results, the anticipated clinical impact may be rather insignificant: none of the 5 cases in our cohort that were up-scored from 1+ (pathologist) to 2+ (image analysis) had genetic amplification.

In a traditional workflow involving the scoring of glass slides (rather than whole slide images), we see little potential benefit to digital image analysis for cases with immunohistochemical scores of 0+ or 3+. It would be worth evaluating in future studies whether digital image analysis can improve interobserver agreement in HER2 scoring or function in the training of persons who perform clinical scoring of cases. The H score data we present also suggest that whole slide image analysis—rather than analysis of representative images—has the potential to improve the sensitivity of an HER2 immunohistochemical assay for detecting cases that would benefit from trastuzumab therapy.

A secondary aim of the study was to examine in greater detail the relationship between protein expression and genetic amplification of HER2—as gene amplification level is a variable that has yet to be evaluated in relationship to treatment response.

We found that a third (10/30) of amplified cases have low-level amplification (defined as HER2/CEP17, 2.2–4.0). These tumors are distinct from cases with high-level amplification (HER2/CEP17, ≥4.0) based on differential HER2 protein expression (Figure 1). Our findings reflect those of a large, retrospective analysis of breast tumors that observed that as the HER2/CEP17 ratio increases beyond 2.0, the proportion of cases scored as 3+ continues to increase and the proportion of cases with immunohistochemical scores of 0+ and 1+ decreases.24 Our findings are at odds, however, with those of a study of esophageal/GE junction cancers that only observed low-level HER2 amplification (defined as HER2/CEP17, 2.0–3.0) in squamous cell carcinoma but not adenocarcinoma.17 This study was conducted using a tissue microarray, and the report did not describe in detail the methods used for scoring, making meaningful comparison difficult.

We chose to separately analyze the group of cases with an HER2/CEP17 of 1.8 to 2.2 because this range is considered indeterminate in breast cancers. We had only 5 cases that fell into this range, and none of them had an immunohistochemical score of 3+, suggesting that cases with this borderline level of HER2 amplification should be carefully evaluated in a manner similar to breast cancers by counting additional nuclei and surveying the tumor for cytogenetic heterogeneity in an effort to accurately score the HER2/CEP17. A much larger study population will likely be required to fully evaluate this issue.

An HER2/nucleus ratio of more than 6.0 may be used as a cutoff for defining HER2 amplification in breast cancers with no chromosome 17 reference probe according to CAP guidelines.15 We found that all GE adenocarcinomas in our series with an HER2/nucleus ratio of more than 6.0 were amplified by an HER2/CEP17 cutoff of 2.2 or more, suggesting that the alternative FISH amplification criterion of an HER2/nucleus ratio of more than 6.0 may be applicable to GE carcinomas as well as breast carcinomas.

We document that low-level chromosome 17 hyperploidy is present in a third of cases if we use mean CEP17 copy number per nucleus of more than 2.0 as a cutoff, a finding that echoes the observations of a detailed 3-dimensional image analysis of FISH performed on thick sections to facilitate highly accurate scoring of the number of FISH signals per whole nucleus.25 Our data substantiate the prevailing belief that HER2 protein overexpression is nearly always associated with amplification of the HER2 region of chromosome 17 rather than hyperploidy of chromosome 17 in GE adenocarcinoma.

We report a large cohort of tumors that have been evaluated by FISH for genetic HER2 amplification and stained using 2 commercially available immunohistochemical assays to assess the correlation of immunohistochemical studies with FISH. Although there is high concordance with FISH with both immunohistochemical assays, we have shown that the 4B5 assay has superior performance characteristics as a screening test for FISH amplification when compared with the A0485 assay. In laboratories where immunohistochemical analysis is used as a screen and only indeterminate (2+) immunohistochemical results undergo FISH testing, we have shown that it may be possible to modestly improve the detection of FISH-amplified cases by determining an H score that accounts for the extent and intensity of positive staining. Our H score cutoffs should be validated in an independent cohort to determine whether they can predict which cases with 1+ and/or 2+ immunohistochemical scores are likely to be amplified. The use of digital image analysis on representative images of immunohistochemical staining did not significantly improve the identification of FISH-amplified cases. However, our H score results suggest that a whole slide image analysis algorithm that simultaneously quantifies the extent and intensity of positive staining may facilitate and improve HER2 immunohistochemical scoring. Last, we have shown that cases with low-level HER2 FISH amplification differ from those with high-level amplification with respect to the intensity of protein expression. This subset of cases with low-level FISH amplification merits further study with respect to response to anti-HER2 treatment.


We thank Samuel Yousem, MD, for critical reading of the manuscript and Rohit Bhargava, MD, for several helpful discussions.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
View Abstract