MET and PTEN gene copy numbers and Ki-67 protein expression associate with pathologic complete response in ERBB2-positive breast carcinoma patients treated with neoadjuvant trastuzumab-based therapy

This study was approved by the Cleveland Clinic Institutional Review Board. All patients who received trastuzumab at the Cleveland Clinic from January 2008 to December 2010 were reviewed for study inclusion (234 patients). Of the 234 cases, 29 satisfied inclusion criteria which included a diagnosis of primary invasive breast cancer, neoadjuvant trastuzumab therapy, and a pre-treatment biopsy performed at the Cleveland Clinic. Pathology data was obtained from the Anatomic Pathology information system CoPath Plus (Cerner Corporation, Kansas City, MO). Clinical data was obtained from the electronic medical record Epic (Epic Systems Corporation, Verona, WI). The age, tumor size, pre-treatment clinical stage, hormone receptor status by IHC, ERBB2 status by ISH, post-treatment pathologic stage, and presence or absence of a pCR were recorded for all patients (Table 1).

A total of 29 patients meeting inclusion criteria with tissue available for IHC and ISH studies were identified (Table 1). Of these patients, 27 had Ki-67 expression data, 24 had MET and CEN7 ISH counts, 25 had PTEN and CEN10 ISH counts, and 24 had both Ki-67 expression data and MET and CEN7 ISH counts. The mean and median age at presentation was 53 and 52 years, respectively. The mean and median pre-treatment tumor size was 49 and 41 mm, respectively (range = 12−86). Patients who presented with Stage IV disease and who underwent breast surgery were excluded. Overall, 15 of 29 (52 %) patients had a pCR, defined as ypT0/ypTis N0. Among the 14 patients who did not have a pCR, the residual invasive tumor measured less than 1 mm in 1 patient and 1 patient had residual ductal carcinoma in situ (DCIS) and a positive lymph node (ypTisN1).

Gene copy numbers for TOP2A, MET, PTEN, and PIK3CA, and their corresponding chromosome copy numbers were measured by ISH, and Ki-67 protein expression was measured by IHC to identify associations with pCR individually and in combination. Figure 1, parts A, B, and C, show representative Ki-67 IHC, MET (black) + CEN7 (red) ISH, and PTEN (black) + CEN10 (red) ISH staining, respectively, on selected FFPE sections of biopsy specimens from the neoadjuvant breast cancer cohort. Figure 1a shows Case #2 (see Table 1 for characteristics of case #2) stained by IHC for Ki-67 and was determined to express Ki-67 in greater than 90 % of the cells. Figure 1b shows Case #2 stained for MET (black) and CEN7 (red) by ISH and shows cells with a lesser number of MET signals than CEN7 signals (48 % of cells showed relative MET loss). The tumor had other areas with a greater number of MET signals than CEN7 signals (34 % of cells showed relative MET gain) as well. Figure 1c shows Case #3 stained for PTEN (black) and CEN10 (red) with increased copy numbers for both loci (86 % of cells are near-tetrasomy).

ISH signals were interpreted using several different parameters to express gene and chromosome copy numbers in order to determine which parameters best associated with pCR. Figure 2 shows plots of 5 ROC curves, each representing a different parameter describing MET gene copy number relative to chromosome 7 copy number (see Methods section for a definition of each parameter). AUC values for each curve are listed in Table 2, along with the optimal cutoff values, whether the high parameter values (equal to or above the cutoff) or low parameter values (below the cutoff) were associated with pCR to generate the ROC curve and contingency analysis, the sensitivity and specificity obtained for those optimal cutoff values, and p-values using Fischer’s Exact test for the 2x2 contingency tables generated at those cutoffs. These ROC curves show a large difference between the association of the various MET/CEN7 parameters and pCR. The two ROC curves representing average MET/CEN7 counts, one associating ratios greater than the cutoff with pCR and the other associating ratios lower than the cutoff with pCR, show little if any association (AUC values near 0.5). The parameters representing the percentages of cells with more copies of MET than CEN7 (MET/CEN7 gain) and the percentages of cells with less copies of MET than CEN7 (MET/CEN7 loss) provided larger AUC values (0.613 and 0.651, respectively), while the parameter based on the sum of the percentage of cells with more copies and less copies of MET than CEN7 (MET/CEN7 gain or loss) provided the largest AUC value (0.791).

Other parameters showing associations with pCR included Ki-67 expression (AUC = 0.726) and the percentage of cells with greater than the normal 2 PTEN copies, PTEN gain (AUC = 0.674), the ROC curves of which are plotted in Fig. 3, and whose AUC values, optimal cutoffs, related sensitivities and specificities, and p-values are included in Table 2. The parameter MET/CEN7 gain or loss was further analyzed in combination with Ki-67 expression, the 2 parameters providing the largest AUC values alone. This was done by generating an ROC curve [44] in which the cutoff for MET/CEN7 gain or loss was held at its optimal value of 50 % cells while varying the cutoffs for Ki-67 expression (plotted in Fig. 3), requiring both parameters in the combination to be equal to or greater than their respective cutoffs for a positive designation. The additional AUC provided by the combination with Ki-67 was 0.056 over that of MET/CEN7 gain or loss alone. The optimal cutoff for Ki-67 expression in combination with MET/CEN7 gain or loss was 8 % cells, maintaining the sensitivity at 92 % while increasing the specificity from 67 to 83 %, and improving the p-value to 0.0006 (see Table 2). PTEN copy number was also evaluated in combination with Ki-67 expression but improvement in p-value and combined sensitivity and specificity over the individual parameters was less (data not shown).

Clinical and pathologic characteristics listed in Table 1 for each patient were compared to the ISH and IHC parameters in Table 2 that showed statistically significant associations with pCR, as well as compared to pCR, using contingency tables (stage, ER IHC, PR IHC) and t-tests (age, tumor size, ERBB2/cell, and ERBB2/CEN17). Results are listed in Table 3 and show few statistically significant associations. MET/CEN7 gain or loss and MET/CEN7 gain or loss combined with Ki-67 expression were significantly associated with age (trend with tumor size), Ki-67 expression was strongly associated with both ERBB2/cell and ERBB2/CEN17, and pCR was significantly associated with only PR expression (trends with ER expression, ERBB2/cell, and age).

In an effort to better identify patients more likely to achieve pCR in patients with ERBB2-positive breast cancer, we have evaluated a series of additional gene and centromere probes as well as Ki-67 expression. In our cohort pCR was achieved in 52 % of patients. As part of the analysis of ISH results we have evaluated different parameters for describing abnormal gene and chromosome copy numbers. This is because there is no single parameter that best describes copy number for all genes and chromosomes for all tumors. For example, gene amplification is often defined as the presence of two or more copies of a gene per copy of the chromosome on which the gene normally resides. This definition of gene amplification was found to have strong clinical relevance with respect to prognosis [45] in breast cancers but has been applied widely to other genes and other cancers with little or no justification. In the present study we have also evaluated the ratio of various gene copy numbers to their respective chromosome copy numbers (as represented by the centromere copy number) and evaluated a wide range of cutoff values. As another measure of relative gene copy number, the percentage of cells with more gene than chromosome copies (gene/centromere gain), or less gene than chromosome copies (gene/centromere loss), or either more or less gene copies relative to their respective chromosomes (gene/centromere gain or loss) were evaluated. Additionally, we have looked at the percentages of cells with greater than two gene or chromosome copies (gene or chromosome gain), or less than two copies (gene or chromosome loss), or the sum of these abnormal copy numbers (gene or chromosome gain or loss). This series of parameters is similar to those used in previous studies that compared different copy number parameters for associations with patient diagnoses and/or outcomes in melanoma [46], esophageal cancer [47], lung cancer [48], and cervical cancer [49]. This is the first application of these parameters in cases with neoadjuvant treatment of ERBB2-positive breast cancer.

The importance of evaluating different parameters can be seen in Table 2 and Fig. 2 for the series of different parameters describing MET copy number relative to chromosome 7 copy number. In this series of parameters both gain of MET relative to chromosome 7 (AUC = 0.613, sensitivity = 0.58 and specificity = 0.67 at the optimal cutoff of 32 % cells with relative gain) and loss of MET relative to chromosome 7 (AUC = 0.651, sensitivity = 0.50 and specificity = 0.83 at the optimal cutoff of 45 % cells with relative loss) are associated weakly with pCR (p = 0.41 and 0.19, respectively at the optimal cutoffs). However, the sum of cells with relative gain and loss (MET/CEN7 gain or loss) is highly associated with pCR (AUC = 0.791, sensitivity = 0.92 and specificity = 0.67 at the optimal cutoff of 50 % cells with abnormal relative copy number with p = 0.0032). Furthermore, the average ratio of MET/CEN7 has no association with pCR, either high ratios or low ratios (p = 1.0 for either relationship at the optimal cutoffs), which is understandable since both MET/CEN7 gain (equating to higher ratio) and MET/CEN7 loss (equating to lower ratio) are associated with pCR and would tend to cancel each other in a ratio calculation. Therefore, proper parameter selection is very important since evaluation of these neoadjuvant specimens using a MET-to-CEN7 ratio, the most common measure of relative copy number, would have erroneously indicated a lack of prognostic value while use of the relative gene-to-chromosome imbalance parameter MET/CEN7 gain or loss provides a high statistical association. Similar to MET and chromosome 7, ISH data for the other genes and corresponding chromosomes were evaluated in terms of the various parameters defined in the Methods section of this paper. Of the other 3 gene probes only PTEN reached statistical significance and this was using PTEN gain in which patient tumors having higher numbers of cells with more than 2 copies of PTEN were associated with pCR (AUC = 0.674, sensitivity = 0.38 and specificity = 1.00 at the optimal cutoff of 58 % cells with PTEN gain, p = 0.039). Examining the actual ISH signals per cell and the average PTEN/CEN10 ratios, gain of PTEN was likely a result of chromosome 10 polysomy, with CEN10/cell ranging between 1.5 and 4.2 and PTEN/CEN10 not exceeding a ratio of 1.4 for any one specimen. Several specimens did exhibit PTEN deletion (PTEN/CEN10 < 0.7) but these appeared in both the pCR and non-pCR groups and no association was found for deletion using the parameters of PTEN loss, PTEN/CEN10 ratio, or PTEN/CEN10 loss.

Ki-67 expression, as measured by IHC, was also highly associated with pCR (AUC = 0.726, sensitivity = 1.00 and specificity = 0.42 at the optimal cutoff of 8 % cells expressing Ki-67, p = 0.0098). The only Ki-67 parameter evaluated was the percentage of cells expressing Ki-67, since this is a standard parameter used by pathologists in the evaluation of breast tissues. Since both MET/CEN7 gain or loss and Ki-67 parameters were highly associated with pCR, the two were analyzed in combination by holding MET/CEN7 gain or loss at its optimal cutoff of 50 % and varying the Ki-67 cutoff over a wide range (Fig. 2 and Table 2). This improved the AUC, and the optimal Ki-67 cutoff (8 % cells) provided sensitivity and specificity of 92 % and 83 %, respectively, and reduced the p-value below either single parameter (p = 0.0006).

The association between pCR and MET/CEN7 gain or loss may be an important finding since common clinical and pathologic characteristics were not found to have statistically significant associations with pCR (Table 3) in the neoadjuvant setting tested here. Of interest, MET/CEN7 gain or loss had little association with ERBB2 gene status but was associated with patient age and trended with tumor size and PR expression.

Our observations do not provide a mechanistic understanding of the role of MET copy number alterations in predicting pCR in HER2-positive breast cancer. And, these findings need to be validated in a subsequent, larger study of MET copy number alterations in a similar patient population. In an updated expansion cohort including patients treated with pertuzumab, additional studies with immunohistochemistry for MET and other markers to interrogate related signaling pathways (e.g., PI3K/AKT pathway) could be informative. One hypothesis is that any alteration of the MET signaling pathway may have some relationship to developing a pCR in the two treatment groups (ACTH and TCH) analyzed in this study. Increased or decreased MET expression potentially resulting from MET gene copy number change might create a fragile condition sensitive to perturbation by therapeutics.