Significance of HER2 Low-Level Copy Gain in Barrett's Cancer: Implications for Fluorescence In situ Hybridization Testing in Tissues

Significant strides continue to be made in the treatment of patients with Barrett's cancer (Barrett's esophagus–associated adenocarcinoma), which is now the most common esophageal cancer and which incidence is increasing faster than that of any other malignancy in the United States (1–3). At this time, early detection and complete surgical resection are the surest ways to cure esophageal cancer (3). Thus, new therapeutic targets and better prognostic markers for patient stratification are needed.

Accurate determination of the status of HER2 in cancer provides significant insight into patient prognosis and may also inform selection of chemotherapeutic treatments. The effect of analyzing HER2 status by either fluorescence in situ hybridization (FISH) or immunohistochemistry has been investigated and also reviewed in extensive meta-analyses, supporting that FISH provides a more powerful prognostic indicator (4–7). This prompted us to investigate the genomic basis of HER2 expression and its prognostic implication in a cohort of Barrett's cancer patients. However, similar to immunohistochemistry methods, not all FISH methods are alike. To date, most FISH studies for standard HER2 testing of paraffin-embedded tissue have been done and evaluated on thin (4 μm) sections by visual signal counting according to established standard procedures (8, 9). Although this standard thin-section technique is relatively easy to analyze, there are limitations with this approach. Most importantly, incomplete nuclei as a result of truncation as well as overlapping cells and/or signals inherent in tissue sections interfere with accurate signal scoring of individual nuclei, thereby decreasing the accuracy of FISH in detection of the actually present signal number (10–14). Although this weakness of FISH is well known, thus far, less attention has been drawn on these effects for accurate evaluation of biomarkers by FISH and for HER2 FISH testing in tissue sections. Recent advantages in three-dimensional microscopic imaging technologies have helped to circumvent these problems leading to an increased sensitivity for signal scoring within thick-section (>10 μm) tissue specimen (10–16).

In this study, we compared standard laboratory FISH on thin (4 μm) sections to a three-dimensional image-based analysis on thick (16 μm) sections (three-dimensional FISH) for HER2 FISH testing. A tissue microarray (TMA) cohort of 124 Barrett's cancer patients was analyzed by standard FISH and three-dimensional FISH for HER2 and chromosome-17 as well as for expression of p185^HER2 by immunohistochemistry. Correlations with follow-up and histopathologic data were tested. Accordingly, paraffin-embedded cells of varying section thickness (2-20 μm, interval of 2 μm) were analyzed referring to the effect of nuclear truncation on the accuracy for HER2 testing. In doing so, we intended to find out which methods accurately measure HER2 status as validated by outcome of patients with Barrett's cancer.

All patients (n = 124) underwent primary surgical resection without chemotherapy or radiotherapy at the Department of Surgery, Klinikum Rechts der Isar, Technical University Munich. All cases had adenocarcinomas of the distal esophagus (Barrett's cancer) associated with histopathologically identified Barrett's esophagus (according to WHO 2000; ref. 17). Data were acquired with approval from the ethics committee of the Technical University Munich. Survival analyses were calculated from all 124 patients. The mean age at surgery was 65.61 years and maximum follow-up time was 164.0 months (median for disease-free survival, 31.2 months; median for overall survival, 33.1 months).

TMAs were constructed by sampling three tumor tissue cores (1.0 mm in diameter) from each paraffin-embedded tissue block using the technique pioneered by Kononen et al. (18). Serial sections were cut and transferred to adhesive slides using the “Paraffin Tape-Transfer System” (Instrumedics).

T47D cell line (American Type Culture Collection) was applied in a model experiment (a) to study the effect of nuclear truncation on the accuracy for HER2 FISH testing and (b) to determine the optimal section thickness for three-dimensional FISH in Barrett's cancer. This cell line is known as tetraploid with four gene copies for HER2 copy and four copies for chromosome-17 (19). The cells were maintained according to the recommendations of the manufacturer and a paraffin block was prepared as described by Crockett et al. (20). Sections of varying thickness (2-20 μm, interval of 2 μm) were cut and placed on slides.

A commercially available assay with fluorescence-labeled locus-specific DNA probes for HER2 and chromosome-17 centromeric α-satellite (Chrombios) was hybridized on the specimen according to a protocol described previously (21, 22). The hybridization specificity of the probes was tested in metaphase spreads and lymphocytes served as internal controls in tumor tissue samples.

Visual-based signal evaluation within 4 μm TMA sections. Signal evaluation within standard 4 μm TMA sections was done by visual counting using an epifluorescence microscope (Zeiss Axioplan 2, Carl Zeiss Microimaging GmbH) according to standard procedures as recommended in literature (8, 9). At least 50 invasive tumor cells per case were selected randomly, whereas only cells with a minimum of one signal for HER2 gene and centromere-17 were chosen, and the mean was calculated.

Image-based signal evaluation within 16-μm (thick) TMA sections and within T47D cell line (three-dimensional FISH). Copy numbers of HER2 gene and centromere-17 within 16 μm TMA sections and T47D cell line sections (2-20 μm) were assessed by optical sectioning and three-dimensional imaging (10, 12, 15). To obtain a three-dimensional data set, image stacks with an interval of 500 nm between two subsequent images were acquired using an Apotome, which was implemented into the setup of a Zeiss Axioplan 2 epifluorescence microscope. The microscope was equipped with a C-Apochromat 63×/NA 1.2 W objective and an AxioCam b/w charge-coupled device camera. Signal evaluation of HER2 and centromere-17 in three-dimensional data set was done by applying the AxioVision software (release 4.5) as described previously (12, 16). Signals were identified by three-dimensional visualization and scrolling through the optical sections. Signals were assessed only in obviously intact (i.e., not truncated cell nuclei with at least one signal for HER2 and centromere-17). A minimum of 50 tumor cell nuclei per case was counted.

FISH scoring criteria. FISH scoring for mean HER2 signal number was applied according to published criteria (6, 8, 9, 23) and four categories were determined: mean HER2 signal numbers <2.5, ≥2.5 to 4.0, ≥4.0 to 6.0, and ≥6.0. HER2 copies ≥6.0 were designated as high-level HER2 amplification.

The HER2 level was also determined as ratio of HER2-to-chromosome-17 within the following four scoring classes: mean ratio <1.1, ≥1.1 to 1.5, ≥1.5 to 2.0, and ≥2.0. HER2-to-chromosome-17 ratios ≥2.0 were designated as HER2 amplification.

The expression levels of HER2 oncoprotein were analyzed using the HercepTest kit according to the manufacturer's instructions (DAKO). Tumors were classified as 0, 1+, 2+, or 3+ as recommended (9). Freshly cut 5-μm TMA sections and the cell line control slides provided by the manufacturer were stained.

Associations between molecular variables and pT and G were examined by Spearman's rank correlation, whereas Mann-Whitney test was used for two-valued variables, such as pN, M, and R. Disease-free and overall survival rates were calculated according to the Kaplan-Meier method, including median and 95% confidence interval (95% CI), and tested with the log-rank χ² value. Multivariate survival analyses were determined using Cox proportional hazards regression for each of the HER2 variables together with the clinical variables pT, pN, and R, including risk ratios and 95% CI. Correlation analyses were done using Spearman's rank test. Calculations were done by use of the statistical data analysis system R (packages ‘stats’ and ‘survival’).⁸

Three-dimensional FISH within thick (16 μm) tissue sections: HER2 gene copy number/HER2-to-chromosome-17 ratios and clinicopathologic correlations.HER2 gene amplification (≥6.0 signals) was found in 10.5% of patients. The majority of the analyzed patients revealed low-level copy number changes for HER2 (≥2.5-4.0 signals; 59.7%). A mean signal number of HER2 <2.5 was observed in 20.2% of the analyzed patients (Table 1).

There was a statistically significant correlation between HER2 and pT (P = 0.041) but none for pN, M, G, and R (Table 2).

Mean HER2 gene copy number was significantly associated with disease-free survival (P = 0.017) but not with overall survival (P = 0.052). In survival analysis, the subgroup with low-level copy number gain (≥2.5-4.0 signals) was as aggressive as the group with HER2 gene amplification (≥6.0 signals; Fig. 1A).

Among the most important established prognostic factors (pT, pN, and R), multivariate Cox regression analysis revealed HER2 as an independent prognostic factor for disease-free survival (relative risk, 2.1; 95% CI, 1.06-4.26; P = 0.033) but not for overall survival (Table 2).

HER2 gene amplification, as defined by the ratio of HER2-to-centromere-17 ≥2.0, was found in 16.9% of patients; 61.2% showed a ratio ≥1.1 to 2.0 and 21.0% displayed a ratio <1.1 (Table 1).

HER2-to-chromosome-17 ratio was also statistically associated with pT (P = 0.0038) and disease-free survival (P = 0.044), but there was no correlation for pN, M, G, and R and overall survival (P = 0.053; Table 2). In survival analysis, the subgroup with low-level copy number gain (ratio, ≥1.1-2.0) was as aggressive as the group with HER2 gene amplification (ratio, ≥2.0; Fig. 1B).

Centromere-17 copy number on its own was not related to disease-free survival (P = 0.59) and overall survival (P = 0.75), indicating that aneusomy of chromosome-17 did not show prognostic significance. Similarly, multivariate Cox regression analysis did not display HER2-to-chromosome-17 ratio as a significant predictor, neither for disease-free survival nor for overall survival (Table 2).

Standard FISH within thin (4 μm) tissue sections: HER2 gene copy number/HER2-to-chromosome-17 ratios by visual counting and clinicopathologic correlations.HER2 gene amplification (≥6.0 signals) was found in 9.3% of patients. Low copy number changes (≥2.5-4.0 signals) were present for 5.6% patients. The vast majority of cases (80.4%) displayed an average signal number of <2.5 signals (Table 1).

Neither HER2 mean copy number nor HER2-to-centromere-17 ratio was associated with pT, pN, M, G, and R, as well there was no statistically significant correlation with patients' disease-free survival (Fig. 1C and D) and overall survival (Table 2). Similarly, centromere-17 copy number was not related with patients' clinical outcome (disease-free survival, P = 0.36; overall survival, P = 0.23).

As there was no significance in univariate analysis, multivariate Cox regression analysis did not show HER2 to be an independent prognostic marker (Table 2).

Sectioning effects bias HER2 signal evaluation in the low-level copy interval. In 4 μm T47D sections, obviously all nuclei were truncated, whereas, in 16 μm sections, the majority of nuclei (80%) were intact (Fig. 2A). The mean HER2 gene copy number within 4 μm and 16 μm T47D sections was calculated with 2.1 ± 0.12 and 3.98 ± 0.03, respectively (Fig. 2B). The detection of the true HER2 copy number of 4 depended on section thickness >12 μm, whereas, in 4-μm sections, the true signal number was underestimated (mean HER2 copy number, 2.1 ± 0.12). Accordingly, only 6% of analyzed nuclei within 4-μm sections exhibited the total number of four HER2 signals, whereas within 16 μm sections, almost all (i.e., 96%) nuclei were detectable bearing four HER2 copies (Fig. 2C). Within sections of 2 to 8 μm, a mixture of 1, 2, 3, or 4 signals could be evaluated in individual nuclei. In contrast, within sections ≥12 μm, predominantly, nuclei with three and four HER2 copies were present, whereas the maximum of 4 signals remained constant in sections ≥14 μm (Fig. 2C).

Figure 3 shows sectioning effects on FISH signal evaluation in tissue sections of Barrett's cancer.

Comparison between three-dimensional FISH on thick (16 μm) sections and standard FISH analyses on thin (4 μm) sections. Both FISH methods identified almost the same proportion of Barrett's cancer patients with HER2 gene amplification (i.e., HER2, ≥6.0 signals; HER2-to-chromosome-17 ratio, ≥2.0) as detected by three-dimensional FISH (10.5% and 16.9%) or by standard FISH (9.3% and 13.1%).

However, there was a difference within the other scoring classes. In detail, 80.4% of cases analyzed within thin (4 μm) sections showed HER2 gene copies <2.5, whereas this group consisted of only 20.2% in thick (16 μm) sections when analyzed by three-dimensional FISH. Accordingly, the group showing low-level copy number gain (≥2.5-4.0 signals) as detected by standard FISH accounted to 5.6%, whereas this low-level copy group increased to 59.7% as detected by three-dimensional FISH.

There was a correlation of mean HER2 signal number evaluated within thin (4 μm) and thick (16 μm) sections (Spearman rank correlation r_s = 0.43; P < 0.0001). However, particularly in the interval of ≥2.5 to 6.0 signals, both methods did not correlate (r_s = 0.20; P = 0.095) and three-dimensional FISH detected 1.9-fold higher values than those that had been assessed by standard FISH (Fig. 4).

HER2 oncoprotein expression and clinicopathologic correlations. Overexpression of HER2 protein indicated by a score of 3+ was observed in 9.8% of patients. A score of 2+ was observed in 7.3% of patients, whereas a score of 1+ or 0 was observed in 13.8% and 69.1% of patients, respectively.

We observed a statistically significant inverse correlation between HER2 protein expression and tumor grading (P = 0.013), whereas differentiated (G₁ and G₂) tumors predominantly showed overexpression of HER2 protein. No significant association with other clinicopathologic features (pT, pN, M, and R) was found as well there was no prognostic effect on disease-free and overall survival (Table 2).

By careful quantification of the exact copy numbers of HER2 and chromosome-17, comparing standard FISH on thin (4 μm) sections and image-based three-dimensional approach on thick (16 μm) sections (three-dimensional FISH), we have shown thus far unrecognized weakness in FISH, a current standard for evaluation of HER2 status. Specifically, we found in our patient cohort that low-level copy number gain (≥2.5-4.0 signals) of HER2 is not reliably detectable by standard FISH analysis in thin (4 μm) tissue sections. Importantly, such subtle copy number changes of HER2 occur frequently and define a biologically distinct subpopulation as unfavorable for patient outcome as HER2 gene amplification in Barrett's cancer (Fig. 1).

Albeit FISH is a sensitive quantitative method appropriate for assessment of the full dynamic range of HER2 gene copy number changes, most studies use this technique rather than a binary approach for the detection of the extreme top signal number (i.e., the presence or absence of gene amplification). Hence, numerous reports have been shown the reliability of HER2 gene amplification detection by the well-established thin-section (4 μm) FISH technique (8, 9). However, thus far, less attention has been drawn on effects of tissue sectioning for standard HER2 FISH testing. Our data indicate that high-level HER2 amplification (HER2 signal number, ≥6.0; HER2-to-chromosome-17 ratio, ≥2.0) is reliably detectable by both approaches (thin-section versus thick-section FISH technique; Table 1). Unlike, it has not been shown thus far, whether the standard thin-section FISH technique reliably detects low-level copy number changes and the prognostic and therapeutic significance of low-level copy number gains is ambiguous. Such uncertainty might also contribute to the recently published considerable variation in interpretation of results for breast cancer with low-level HER2 amplification (24). In contrast to detection of HER2 amplification, we observed a striking variation for the detection of low-level copy number gain (HER2 signal number, ≥2.5-4.0) in 4-μm sections (5.6%) compared with 16 μm sections (59.7%; Table 1). In other words, a large proportion of tumors with slight copy number increase in the low-level group detected by thick-section technique (three-dimensional FISH) would not have been recognized in standard HER2 FISH testing and hence would have been classified as “normal” (Fig. 3C and D). Further evidence comes from our experiments investigating the influence of section thickness on FISH signal evaluation in a paraffin-embedded cell line with four gene copies for HER2 (19). In standard thin-section FISH technique, obviously, all tumor cell nuclei are truncated due to tissue sectioning (Figs. 2 and 3). This effect is self-explanatory as the mean diameter of tumor cell nuclei (≈10 μm) is more than twice of the standard section thickness (4 μm), resulting in a loss of at least 40% of the nuclear volume. Hence, nuclear truncation decreases the accuracy of FISH in detection of the actually present signal number as published repeatedly (10–14). Correspondingly, our data indicate a nonlinear correlation between nuclear truncation and mean signal number, leading to a bias particularly affecting the low-level copy interval in thin (4 μm) sections (Fig. 2).

Beyond these technical considerations, our data may also have a biological aspect. Thus far, such adverse effects of low-level gain on patient outcome as observed in our series has not been shown yet for HER2 gene copy number but for low expression of HER2 protein in breast cancer (25–28). Corresponding to our observations, these reports have shown that more sensitive assays (i.e., AQUA method or radioimmunohistochemistry), which increased the dynamic range of HER2 protein detection, could identify patient subgroups with low or even normal expression of HER2 protein being as unfavorable for survival as strong overexpression (25–28). However, it is unclear, thus far, if low-level copy number gain actually leads to elevated expression of HER2 protein. In our series, the patient subgroup with low-level copy number gain was not definable by standard immunohistochemistry. This may be explained by the low sensitivity of this standard approach, which is unable to detect slightly elevated expression levels if actually present (29).

Previous studies examining HER2 status in Barrett's cancer observed a prevalence of HER2 protein overexpression or gene amplification in Barrett's cancer ranging from 0% to 83% (21, 30–40). Lack of agreement among these studies may be related to the differing sensitivities and specificities of the assay methods used to assess HER2 as a prognostic marker. Several authors suggested that the accurate evaluation of HER2 may be useful as a biomarker of progression from intestinal metaplasia to intraepithelial neoplasia in carcinogenesis of Barrett's esophagus (30, 31). Because most moleculargenetic changes are rather subtle in precursor lesions, this research field might particularly benefit from sensitive methods such three-dimensional FISH to detect subtle signal changes.

Importantly, we did not find prognostic significance for chromosome-17 aneusomy in our Barrett's cancer cohort. Clinically, it is uncertain, thus far, whether an increase in HER2 copy number alone or an increase of HER2 copies relative to copies of chromosome-17 is more useful in predicting outcome. According to our findings, the results of a recent report in breast cancer suggest that increased HER2 gene dosage may be the most important determinant of HER2 gene expression, whereas that resulting from polysomy 17 alone is rather unlikely to significantly contribute to HER2 overexpression (41).

From a clinical perspective, response to Herceptin has largely been seen in HER2 high expressers or HER2-amplified cases. Although Barrett's cancer patients with low-level gene copy gain of HER2 are unlikely to respond to Herceptin, they may benefit from more aggressive traditional chemotherapy. HER2 status might be incorporated into a clinical decision, along with other prognostic factors, about whether to give any adjuvant systemic therapy. HER2 status may also to be predictive for either resistance or sensitivity to different types of chemotherapeutic agents. Thus, the ability to accurately distinguish between low-level gene copy gains and high-level gene amplification of HER2 by three-dimensional FISH does not only have prognostic value but may also help in the development and evaluation of new therapeutics targeted to treat this patient subpopulation.

In summary, this study highlights a thus far unrecognized weakness in standard HER2 FISH testing, specifically referring to the low-level copy interval, which could not reliably be detected in standard thin (4 μm) tissue sections in our study. Importantly, these subtle copy number changes of HER2 defined a considerable and biologically distinct subpopulation in Barrett's cancer. These results should be taken into account for accurate evaluation of biomarkers by FISH and for HER2 FISH testing in tissue sections.

We thank Lieselotte Bokla, Ulrike Buchholz, Eleonore Samson, and Andreas Voss for excellent technical assistance.