Introduction

Triple negative breast cancer (TNBC) is an aggressive form of breast cancer with limited treatment options in the form of systematic rather than targeted therapies. Because of this various groups have been searching for factors that could give additional prognostic information or explain the underlying biology of this disease. Elf5 is a member of the ETS family of transcription factors encoded at the 11p13–15 region of the genome and is found in the epithelial layer of a variety of tissues [1]. Its chromosomal location is significant as this region often undergoes the loss of heterozygosity in breast cancers, suggesting that the proteins coded in this region may be candidate tumour suppressor genes [1].

The roles of Elf5 in linage determination and cell fate in the normal breast have been studied extensively by Ormandy et al. Results of these studies revealed that Elf5 expression is essential in the differentiation of progenitor cells into differentiated luminal breast cells capable of the production of milk [2, 3]. This is partially achieved by the paracrine actions of ER/PR expressing hormone sensing mammary epithelial cells. These cells use RANK-ligand as a paracrine factor that induces Elf5 expression and subsequent epithelial differentiation. This process is potentiated by Elf5 mediated upregulation of rank [4] and possible actions on notch signalling [5].

Elf5 actions in the normal and pathological breast have been documented in detail in the mouse where the prolactin dependent expression of Elf5 is necessary for the development of normal alveolar lobules during pregnancy, and the development of ER− milk producing cells [3, 6]. Due to this connection with ERα expression Ormandy et al. also explored the expression and significance of Elf5 in breast cancer. He found that its expression associated with the basal such as subtype of breast cancer, and targets directly repressed by Elf5 included typical canonical elements of the luminal signature of breast cancer including FOXA1 and GATA3 [7].

The latter point, Elf5 directed regulation of FOXA1 and GATA3, is what caught our interest in examining its expression in triple negative breast cancers. We have previously examined the effects of androgen receptor (AR) expression in triple negative breast cancers [8,9,10] and a defining part of the actions of the AR are through its interactions with FOXA1 [11]. Thus, we speculated that ELF5 expression and interactions may underpin some of the biology of the differing actions of androgens and/or underlying subtypes of TNBC. In turn, the understanding that this generated could be used to aide in the creation of a unified biological picture of TNBC, rather than the current fragmentary classification scheme. To further evaluate this understanding, we also examined the interactions of Elf5 with other steroid receptors reported to be common in triple negative breast cancers such as the Estrogen Receptor beta (ERβ).

Methodology

Patient cohort

In the case of invasive ductal carcinoma (IDC) the samples consisted of 60 cases from Tohoku University School of Medicine. Ethical approval for this research was obtained from the Tohoku University Ethics committee. In the case of the TNBC ductal carcinoma in situ (DCIS) samples, a total of 51 samples were obtained from St Luke’s Hospital Tokyo and Tohoku Kosai hospital with ethical approval from the respective institutions. TNBC was defined by ERα and PR scores below 1% and HER2 scores of 3 + receptor expression by IHC or > 2.2 Her2 to CEP17 ratio by FISH. The rational of these cut-offs are given in Miyashita et al. [13].

Immunohistochemistry

Immunostaining was performed using an ELF5 primary antibody (#PAB6200, Abnova, Taipei City 114 Taiwan) with prostate tissue as positive control. Immunohistochemistry was performed as previously described [10]. Antigen retrieval was done using autoclave heat treatment in a pH6 (Citrate Buffer) and a 1–500 dilution of the primary antibody was used. The specificity of the staining was confirmed by antibody absorption test. Expression of Elf5 was quantified as either positive or negative on the basis of a 10% labelling index. The exception to this is in the case of survival analysis in the IDC samples, where an ROC analysis of values was used and gave a cut-off point of 50% as the optimal point as which survival outcomes differentiated. The samples used in this study are subsets from our previously published TNBC cohorts [9, 10, 12] and as such the correlations with AR and ERβ1 and CK5/6 and EGFR surrogates are based on the data contained within those papers. The antibodies used were; AR, AR44, DAKO, 1–50 dilution; ERβ1, Genetex, PPG5/10; 1:50); CK5/6, NCL-LL02, DAKO, 1–100; EGFR, K1492, DAKO, kit based; ERβ1. AR and ERb1 were scored via H-Score averaging five hot spots of at least 100 cells. EGFR and CK5/6 were considered positive if greater than 10% of tumour cells had membranous and cytoplasmic reactivity. Negative controls of antibody omissions were performed.

Statistical analysis

All pairwise comparisons were tested by Wilcoxon rank sum test. Due to the problem of ties, confidential intervals were not calculated. Categorical relationships were tested by χ2 test. Survival analysis was tested with the Peto & Peto modification of the Gehan–Wilcoxon test. Multivariate analysis of survival curves was modeled using the Cox-hazard model. All tests were 2 sided and a significance level of p value less than 0.05 were considered significant. Variable selection was conducted using the stepwise Akaike information criterion (AIC). All analysis was done by using R version 3.4.0. survival and MASS packages.

Results

Elf5 immunoreactivity in carcinoma samples

Elf5 immunoreactivity was detected with varying degrees of intensity in histologically adjacent normal, DCIS and IDC epithelia. In histologically normal ducts Elf5 was expressed in a large number of luminal epithelial cells, whereas in carcinoma the patterns of its immunolocalization ranged from complete expression throughout the entire carcinoma to heterogeneous patchy expression (Fig. 1). Notably positive DCIS cases also exhibited marked perinuclear staining in 24% of samples, a pattern not detected in IDC samples. It is worth noting that the relative immunointensity in normal epithelial cells almost always exceeded that of carcinoma cell in the same tissue sections. While the average level of Elf5 immunoreactivity was lower in DCIS than in IDC, this difference was not statistically significant (DCIS: 31% LI ± 11% vs IDC 46 ± 9% LI, χ2 test: p = 0.1479).

Fig. 1
figure 1

Elf5 expression in histologically normal, DCIS and IDC samples. a–f show representative images of Elf5 staining in TNBC breast cancer specimens. a Shows positive Elf5 staining in histologically normal glands adjacent to IDC. b Shows a representative example of negative carcinoma cells in ductal in situ. c, d Show representative image of positive DCIS staining, both nuclear and perinuclear staining could be observed in these samples. e, f Show representative images of positive and negative staining for Elf5 in IDC samples. One characteristic observed across positive and negative DCIS and IDC samples were the presence of strongly positive cells in the stroma, however, these were not the primary focus of this manuscript and were not considered in our analysis

Full size image

Correlation of Elf5 with clinicopathological factors

In IDC the status of Elf5 was not significantly associated with any of the clinicopathological factors previously shown to be prognostic in TNBC. The one factor showing a significant association was age, with a higher proportion of younger patients exhibiting positive Elf5 expression (Table 1).

Table 1 Association between Elf5 and clinicopathological factors in IDC
Full size table

Correlation of Elf5 with regulators of steroid signalling

To elucidate if Elf5 was associated with steroid signalling pathways in TNBC we examined the correlation of Elf5 status with the two steroid nuclear receptors, AR and ERβ1 (Fig. 2). In DCIS, this analysis revealed that Elf5 status was significantly associated with higher levels of ERβ expression (p = 0.025) as well as a trend towards decreased levels of AR expression (p = 0.12). In IDC, the same interaction with ERβ was not detected.

Fig. 2
figure 2

Elf5 association with AR and ERβ in DCIS and IDC samples. In the graphs above you can see the association of ELF5 expression (10% LI cut-off) and AR (a, b) and ERβ1 (c, d) H score. In DCIS Elf5 positivity was significantly associated with ERβ1 expression (c) and showed a trend towards interaction with AR expression (a). The same patterns were not seen in IDC samples (b, d)

Full size image

Survival analysis

As Elf5 is hypothesised to have effects on stemness, which in turn may be expected to influence survival of the patients, we examined the associations between ELF5 status and survival in both univariate and multivariate analysis in IDC cases. Results of univariate survival analysis, using a 10% labelling index (LI) showed that Elf5 positive cases were associated with a worse disease free and overall survival than Elf5 negative cancers, however, this trend did not reach statistical significance (Fig. 3a, b). Subsequently, we used an ROC analysis to identify if there was an optimal LI in determining survival and found that 50% immunoreactivity was optimal. If divided on this basis, survival curves were significantly different between ELF5 low expression and ELF5 high expression samples (Fig. 3c, d).

Fig. 3
figure 3

Elf5 is association with survival in triple negative breast cancers. Elf5 expression showed a tendency to worse disease free and overall survival when dichotomised at a LI of 10% (a, b). After using ROC analysis to find the optimal point to dichotomise, the effect on survival was statistically significant (c, d). Interestingly, and not dependent of cut off point of LI, Elf5 positive non-basal cancers routinely had worse survival outcomes (e–h). In multivariate analysis Elf5 at a cut-off of 50% LI was not independently significant in DFS (i) and overall survival (j)

Full size image

Interestingly, when split into the EGFR+ and/or CK5/6+ (basal surrogate) and EGFR− and CK5/6− (non-basal surrogate), we found that the majority of adverse survival associated with Elf5 expression is attributable to its effects in the EGFR− and CK5/6− samples. These differences in survival were significant regardless of LI cut-off point used (Fig. 3e, f). In multivariate analysis, we used Age, Ki-67 LI, tumour size, CD31, pathologic node status, basal status, ELF5 status of LI > 50 as surrogate variables determining survival outcomes described in previous literature [13]. While Elf5 did not reach significance at any of the multivariate analysis (Fig. 3i, j), it was retained during variable selection by stepwise AIC suggesting that its inclusion contributed to the overall ability of the multivariate model to predict survival.

Discussion

In this study, we showed that the differentiation factors Elf5 is expressed in triple negative breast cancers and is a significant factor in the univariate analysis of survival. Additionally, we showed its presence in pre-invasive cancers and a potential association with steroid receptors. Our findings partially confirmed those previously published insofar as they suggest that Elf5 expression in carcinoma cells may suggest a worse prognosis carcinoma.

Interestingly, in our present study, we see the most dramatic effects in the EGFR negative, CK5/6 negative (non-basal surrogate) triple negative breast cancers. In normal luminal cells, Elf5 stimulates the secretion of Rank as a paracrine mediator of differentiation [4] and possibly affects notch signalling [5]. As both these pathways have been implicated as adverse actors in triple negative breast cancer [14, 15] this may partially explain the results we observed in our study. There are also other suggestions of ELF5 expression being related to proteins that relate to EMT processes [16]. If this is true in breast cancer samples it may also explain our observations of poor prognosis associated with high levels of ELF5 expression. Unlike previous studies [7] we did not see a selective presence of Elf5 in the basal subtype. While this may be explained by our comparatively small sample size, or potentially variability in Elf5 isoform expression [17] further studies are needed to confirm or disprove these ideas.

Despite our initial hypothesis, we did not see strong evidence of Elf5 actions on steroid regulation of TNBC biology in IDC samples, or in IDC cell lines (data not shown). The associations observed in DCIS between Elf5 positivity and hormone receptor positivity are curious and potentially suggest some sort of dysregulation of cell differentiation in the DCIS stage of triple negative breast cancers. Sex hormone receptors and Elf5 do appear to converge on a number of pathways including interactions with Notch signalling, FOXA1 transcription factors and cancer differentiation [18,19,20]. This suggests multiple possibilities for their interaction. While our study suggests that this interaction is not direct in invasive carcinoma, it remains a worthwhile project to continue disentangling the biology underlying the multiple pathways of triple negative cancers. Hopefully, this biological understanding will lead to more effective treatment and management options in triple negative breast cancers.

Conclusion

In our present study, we demonstrate expression and localisation of Elf5 in invasive and non-invasive triple negative breast cancers. We also demonstrate that, in our cohort, Elf5 is an adverse prognostic factor, independent of other clinicopathological classifications including basalness and that this is particularly pronounced in EGFR-/and CK5/6- (non-basal surrogate) triple negative breast cancer specimens.