Distribution of Ki-67 values within HER2 & ER/PgR expression variants of ductal breast cancers as a potential link between IHC features and breast cancer biology

In this study 1180 consecutive invasive ductal breast cancer patients (any stage) were included. All patients were diagnosed and treated in Osijek Clinical Hospital from the period January 2004 to December 2012. We have used a single institution set of breast cancer patients that has already been assembled as a part of a research project financed by the Croatian Ministry of Science (219–2,192,382-2426). Before grant submission to Croatian Ministry of Science and Education, collecting of breast cancer data was approved by the Ethical Committee of Osijek Medical Faculty, as compliant with the Helsinki Declaration. These same patients’ data were used for testing two other breast cancer models [8, 9] and the results of these testings were published elsewhere [10, 11].

All of the specimens were excisional biopsies or mastectomy specimens. Tumor grades were determined using the Bloom and Richardson scheme [12‐14].

All IHC slides were coded and independently evaluated by two pathologists, who are also the coauthors of this paper. They have used the ImageJ program tools (https://​imagej.​nih.​gov/​ij/​) when needed. Each immunostained slide was evaluated for the presence of ER and PgR expression, HER2 protein overexpression, and Ki–67 proliferation activity. Immunohistochemical staining was done by the standard avidin-biotin method (DAKO LSAB®2 System, HRP) using 4 μm sections from representative paraffin blocks. Nuclear staining with anti-ER, PgR and Ki-67 antibodies was also done and the percentage of positive cells per 500 tumor cells was calculated. Tumor cells were considered positive for HER2 protein over-expression when greater than 10% of the cells showed strong membrane staining (equivalent to a score of 3+ in the DakoCytomation HercepTest). An HER2 2+ result was considered overexpressed only if confirmed by chromogene in situ hybridization for gene amplification. Hormone receptors were reviewed and accepted as negative if 100% of cells lacked nuclear immunostaining.

From our previous pilot study, we have noticed for the Ki-67 values that the two independently estimated values were usually less than 7% apart, so in all cases when the difference was <6% we have used the arithmetic mean of these two estimates as the final value. In less than one fifth of patients, with the ki-67 gap >5%, two new independent estimations were done. The lowest and the highest value were discarded and the arithmetic mean of the remaining two values was used.

Collected data were organized in a spreadsheet by StatSoft, Inc. (2011) STATISTICA (data analysis software system), version 10. www.​statsoft.​com.

As shown in Table 1 the usual distribution of breast cancers was based on HER2 expression, low or high Ki67 values and combinations of ER and PgR presence, thus resulting in 16 subgroups (four HER2 variants (0+, 1+, 2+ and 3+) times four ER/PgR phenotypes).

Out of 16 subgroups in Table 1 ER-PgR+ subgroups had too few patients to be used in statistical tests (only 11 patients), so they were excluded from further statistic tests. Further more, out of the remaining 12 subgroups (four with ER + PgR+, four with ER + PgR- and four with ER-PgR- tumors), histograms of Ki-67 distributions were made in Fig. 1 only for HER2 subgroups 0+, 1+ and 3+. The three omitted HER2 2+ cancer subgroups were not suitable for histogram comparison, due to low number of patients.

Complex distributions shown in Fig. 1 suggested that more than one cluster of patients might be present in each subgroup. Possible existence of clusters within a single phenotypic subgroup was tested by applying the method of expectation maximization (EM) clustering [16] to the original Ki-67 data of ER + PgR+, ER + PgR- and ER-PgR- breast cancers (in total 12 subgroups). The v-fold cross-validation algorithm for automatically determining the number of clusters in the data (provided by the Statistica program) was applied during the clustering. The EM algorithm of clustering approximates the observed distributions of values by a mixture of distributions in different clusters.

We have done a two-stage EM clustering. The first stage is done within the described subgroups and it suggested that all subgroup clusters belong to only three clusters present in the whole set of patients. In the second stage, all data were pooled together to verify presence of these three overall clusters that were used in further analysis.

Distribution of Ki-67 values among the 16 proposed phenotypic subgroups are shown in Table 1. Among them, 11 out of 1180 patients (0.93%) showed the rarest ER-PgR+ cancer phenotype, so in following tables these 11 patients were excluded, thus leaving 12 subgroups with 1169 patients.

Figure 1 shows discrepancies between distributions of Ki67 values among the remaining nine subgroups of patients regarding their ER/PgR phenotype and HER2 expression (0+, 1+ or 3+, HER2 2+ tumors were omitted due to low incidences). These data were validated by Kruskal-Wallis tests:

In short, if we compare cancers positive for ER and PgR, where HER2 expression increases Ki67 values, with the ER-PgR- cancers, were HER2 expression decreases otherwise very high KI-67 values, these unexpected differences obviously required further examinations. A plausible interpretation is that even in these narrow subgroups of breast cancers, unexpected distributions of Ki67 values might result from further subgroup divisions.

Table 2 shows results of the first stage EM clustering for the analyzed subgroups. Figure 2 shows distributions of EM clusters within nine subgroups analogous to the histogram setting in Fig. 1. Despite our expectations, the v-fold cross-validation algorithm detected only two clusters of patients in each subgroup:

The above results of EM clustering in various subgroups suggest that in all three analyzed ER/PgR phenotypes, some patients had breast tumors that do not overexpress HER2 and have similar intermediate mitotic activity independent of the presence of ER and PgR (all IMA clusters). On the other hand, clusters of low mitotic activity (LMA) were present only in ER positive cancers (both in PgR+ and PgR- cancers). Their share was slightly reduced in subgroups with HER2 expression (1+ to 3+), suggesting that these tumors of low mitotic activity were hormone driven and thus less EGFR/HER2 dependent. Among ER-PgR- tumors, cancers of high mitotic activity formed the HMA clusters, more common in variants poor in HER2 expression (HER2 0+ and 1+).

To test these observations, data of all patients were pooled together and Ki67 values were tested for the presence of three EM clusters (here defined as pooled LMA, IMA and HMA clusters), shown in Table 3 and Fig. 3 with mean Ki-67 values (LMA 1.17%, IMA 40.45% and HMA 77.79%).

Table 4 shows unexpected distribution of our patients according to their tumor type (Luminal A/B1, Luminal B2, triple-negative and pure HER2) and cluster participation. A very few patients with ER+ tumors have been classified as belonging to the overall HMA cluster (some of them were PgR+ and other PgR-). On the other hand, ER- patients were classified to belong to all three overall clusters (29.75% LMA tumors, 46.95% IMA and 23.30% HMA tumors), clearly suggesting that their distribution of Ki-67 values differs substantially from ER+ patients. Dark grey cells in Table 4. mark the fields in which observed frequencies were above the expected frequencies, while the light grey cells mark the opposite situation in which observed frequencies were below expectation.

Taken all together, these results suggest that breast cancers can be divided in three levels of mitotic activity, with different mechanisms behind ER positive and ER negative tumors. In the former HER2 overexpression increases number of IMA tumors on the expense of LMA tumors, while in the latter HER2 overexpression reduces number of HMA tumors. A possible interpretation is that ER + PgR+ and ER negative breast tumors are intrinsically so different that the HER2 overexpression reduces number of LMA ER + PgR+ tumors and HMA ER-PgR- tumors. This is supported by the observation that HER2 absent (0+) tumors show highest shares of LMA ER + PgR+ and of HMA ER-PgR- cancers.

To make these observations more clear, distributions of the pooled HER2 expression data within the three clusters and three ER/PgR phenotypes are shown in Fig. 4. as seven pie charts.

Speed of the primary tumor growth mainly depends on the mitotic rate (routinely estimated by the Ki67 value) and on the rate of cancer cell destruction by apoptosis and other mechanisms that threaten the survival of tumor cells.

According to guidelines, breast cancer patients are after surgery treated according to their cancer type. Within Luminal tumors, Ki67 values define two cancer types, Luminal B1 and Luminal B2. This means that immunohistochemical phenotype of tumor tissue somehow influences the course of disease and effects of various treatments including targeted drugs. Here reported disparities in Ki-67 values between tumors with normally expressed HER2 (subset of patients with cancers expressing HER2 from 0+ to 2+) suggest that five common types of breast cancer are not as homogeneous as it can be expected.

Table 5 shows an attempt to interpret here presented relations between the phenotype variants and breast cancer biology among our patients. Here proposed explanation is that subgroups of breast cancer phenotypes differ in their Ki-67 distributions due to separate mechanisms that also include Ki-67 dependency on HER2 expression:

A study by Wang XZ et al. [17] can be used as an illustration that less recognized tumor growth mechanisms have been proposed in triple-negative breast cancer patients. They have analyzed 264 patients with breast cancer divided into four molecular types plus the expression of p53 and EGFR. Triple-negative and HER2 overexpressed cancers were found to be larger and with higher Ki-67 as compared with the Luminal types. Beside that, triple-negative tumors showed less positive lymph nodes and higher CK5/6 and EGFR expression than the other three types, while p53 expression positively correlated with the EGFR expression only among triple-negative tumors, suggesting that tumor growth mechanism in triple-negative might differ from other breast cancers [17].

In triple-negative tumors promitotic mechanisms can include various mediators that do not interact with ER and PgR. Beside androgen receptor, EGFR ligands, activin/inhibin interactions also seem plausible [18‐20].

Based on these observations, Table 5 also addresses few open questions regarding Immunohistochemical phenotypes of tumors of the three clusters based on their mitotic activity:

The prevailing interpretation of the breast cancer occurrence is that increased or prolonged estrogen exposure leads to an increased risk for the development of breast cancer [21]. Estrogen via ER molecules stimulates proliferation of breast cancer cells and regulates the expression of other proteins in the tumor cells, including the progesterone receptor [22]. The presence of ER or PgR on breast cancer cells typically suggests slower-growing tumors, amenable to hormonal manipulation [23].

Here presented results suggest that the PgR expression on breast cancer cells is related to the Ki-67 value, here used as marker of tumor biology. It is important to note that the role of PgR expression in breast cancer cells remains not fully elucidated, since PgR expression is influenced by the estrogen milieu [7] and it has been reported that the lack of PgR in ER+ tumors is associated with worse survival [6]. A research study involving 327 ER+ breast cancer patients as shown that the Luminal B patients with PgR- tumors had a relatively higher pathological complete response rate than patients with PgR+ tumors (29.5% versus 4.7% pCR, P < 0.001), but in Luminal B patients with a residual tumor after neoadjuvant chemotherapy, PgR absence was independently correlated with poor relapse-free survival (P = 0.017) and overall survival (P = 0.013) [24]. These authors have concluded that the lack of PgR expression might be an important determinant of tumor biology in Luminal types of breast cancers.

Among 4115 patients with ER or PgR positive and not HER2 overexpressed breast cancers, reduced cancer-free intervals were noted in patients whose tumors had lower PgR expression and higher Ki-67 value [25]. This is possibly related to the second report that among 398 patients early relapses in patients with Luminal B and HER2-negative breast cancers were related to PgR negativity [26].

It remains unsettled whether the PgR expression threshold should be as low as 1% or higher. Among 1522 patients with primary breast cancer ER+/PgR−/HER2- tumors showed poorer clinicopathologic characteristics compared with ER+/PgR+/HER2- tumors using a PgR threshold of 20% instead of 1% [27]. In another report of 327 surgically removed ER positive and HER2 not-overexpressed breast cancers, only among postmenopausal patients it was reported that high Ki-67 value and low PgR expression (<20%) were significant independent factors for worse distant relapse-free survival [7].