The prognostic significance of BMI1 expression in invasive breast cancer is dependent on its molecular subtypes

The study cohort comprised 870 invasive BCs derived from the retrospective Nottingham Primary Breast Carcinoma Series of patients presenting to Nottingham City Hospital between 1986 and 1998. Patients’ clinical and pathological data including age at diagnosis, histological tumour type, tumour size, lymph node status, Nottingham Prognostic Index (NPI), lympho-vascular invasion (LVI) and adjuvant therapy were available and prospectively maintained. Data for oestrogen receptor (ER), progesterone receptor (PgR), HER2 status and Ki67 data were available [15, 16]. ER and PgR cut-off values were defined as ≥ 1% and HER2 status was defined as previously published [15, 17, 18]. Survival data were accessible and prospectively maintained including the following: (1) BC-specific survival (BCSS), defined as the time (in months) from the date of the primary surgical treatment to the time of death from breast cancer, and (2) distant metastasis free survival (DMFS), defined as the time (in months) from the surgery until the first event of distant metastasis [19]. The clinicopathological parameters for the study cohort are summarised in supplementary Table 1. BC intrinsic molecular subtypes were determined as previously descried [20]. Immunohistochemical detection of a panel of BCSC, including ALDH1A1, CD133, CD24, CD44, EPCAM, SOX9 and SOX10, had been previously performed [21‐23] and these were used in the current study to assess their relationship with BMI1 expression.

Full face BC tissue sections were stained using IHC to evaluate the pattern of immunohistochemical BMI1 expression prior to staining of TMAs. Kidney tissue was used as a positive control while the negative control was obtained by omitting the application of primary antibody in the IHC staining protocol.

Formalin-fixed paraffin-embedded BC tissue samples were arrayed as previously described [24]. Prior to IHC staining, the specificity of the anti-BMI1 antibody was validated using Western blotting in MCF7, MDA-MB-231, SKBR3, MDA-MB-468 BC cell lines’ lysates (American type culture collection, Rockville, MD, USA) and HeLa cells as control. This was performed using 1:5000 dilution of the primary antibody (EPR3745 (2), Abcam, UK), and 1:15,000 of the horseradish peroxidase-labelled secondary anti-rabbit antibody, with b-actin (1:5000) used as a loading control. A single band for BMI1 was observed at the predicted size (40 kDa), which confirmed the specificity of the antibody (Supplementary Fig. 1). IHC staining was performed on 4 μm TMA sections using Novolink polymer detection system (Leica, Newcastle, UK). In brief, the antigen retrieval was performed in citrate buffer (pH 6) in a microwave (Whirlpool JT359 Jet Chef 1000 W) for 20 min. The optimal dilution of BMI1 antibody in IHC was 1:100 and incubated for 1 h at room temperature. Stained TMA slides were scanned with high-resolution digital images (NanoZoomer; Hamamatsu Photonics, Welwyn Garden City, UK), at 20 × magnification and viewed by Xplore viewer (Philips, Belfast UK). The BMI1 staining TMA cores were evaluated on the basis of a semiquantitative scoring of core digital images using a modified histochemical score (H-score) [25]. All cases were scored by M. Althobiti, blinded to histopathological data and patients’ outcome. To validate the results and test for the inter-observers reproducibility of the scoring, 10% of the cases were randomly selected and rescored by another observer (M. Toss).

A cohort of 1980 BC patients was evaluated in terms of BMI1 gene copy number (CN) aberrations and mRNA expression using the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) [26, 27]. The cut-off point of BMI1 was determined using X-tile software (version 3.6.1, Yale University, USA), which was based on prediction of BCSS. The clinicopathological parameters of METABRIC series are summarised in supplementary Table 1. There was no difference in the distribution of clinicopathological parameters between Nottingham series and METABRIC series of patients (correlation coefficients = 0.733, all P < 0.0001) [28].

To validate the prognostic significance of BMI1 mRNA expression, another publically available database (Breast Cancer Gene Expression Miner v4.0 (Bc-GenExMiner v4.0), with the online dataset available at https://​bcgenex.​centregauducheau​.​fr), was used. This large dataset (n = 9616) allowed the evaluation of the prognostic role of BMI1 in BC cohorts, which include key prognostic parameters such as patients’ age, tumour grade, nodal status, NPI, ER and molecular subtypes. Univariate analyses for molecular BC subtypes were performed [29].

The IHC staining showed a homogenous staining pattern, with BMI1 expression localised in the nuclei of the invasive tumour cells. The staining intensities varied from negative (no stain) to strong intensity (Figs. 1, 2). Inter-observer agreement was determined, and the interclass correlation coefficient was 0.931, indicating an excellent concordance between the 2 scorers.

In the whole BC cohort, the IHC expression of BMI1 ranged from 0 to 270 H-score. The data showed that high/positive BMI1 IHC expression (H-score > 130) was observed in 20% of cases (176/870), whereas 80% of cases (694/870) was considered low/negative. Interestingly, the immunoexpression of BMI1 in the luminal ER-positive (ER+) subtype was higher than in ER-negative (ER−) subtypes (P < 0.0001). In the METABRIC cohort, high expression of BMI1 mRNA was seen in 65% of cases (1288/1980). In the Nottingham cases of the METABRIC cohort (n = 340), there was a strong association between BMI1 mRNA expression and protein expression (P = 0.001). BMI1 CN gain was observed in 6% of cases (113/1980) whereas 1% of cases (25/1980) showed CN loss. Supplementary Table 2 summarises the mean, median and the range of expression of BMI1 in BC subtypes at both protein and mRNA levels.

In this cohort, 73% of ER+ subtype expressed high BMI1 while 60% of the ER-negative subtype expressed high BMI1.

High BMI1 protein expression showed an association with clinicopathological parameters characteristic of good prognosis including lower histological grade (P < 0.0001), more tubule formation (P = 0.004), lower mitotic count (P < 0.0001), lower nuclear pleomorphism (P < 0.0001), lower NPI scores (P = 0.024), special tumour type of good prognosis (P < 0.0001) and with tumours showing ER+ (P < 0.0001) and HER2− phenotypes (P = 0.002) (Table 1).

Similar findings were identified in the METABRIC cohort. High expression of BMI1 mRNA was positively associated with good prognostic factors, such as older age (P < 0.001), postmenopausal status (P < 0.001), lower grade (P < 0.0001), good NPI prognostic group (P < 0.0001), tumours of tubular subtype (P < 0.0001) and HER2− phenotypes (P = 0.001) as shown in Table 1.

These associations with good prognostic factors were also obtained in the BC Gene Expression Miner database; high expression of BMI1 was associated with older age, good prognostic factors, such as good NPI and luminal A subtype (P < 0.0001), as shown in supplementary Fig. 2.

Univariate survival analysis showed that high BMI1 protein expression was significantly associated with longer survival in terms of breast cancer-specific survival (BCSS) in the whole BC cohort (P = 0.02) (Fig. 2a). With regard to molecular subtypes, in the luminal ER+ tumours, high expression of BMI1 was significantly associated with BCSS (P = 0.04; Fig. 2b) and distant metastasis free survival (DMFS) (P = 0.04; supplementary Fig. 3). However, in the basal / triple negative subtype (TNBC), high expression of BMI1 was significantly associated with shorter BCSS (P = 0.04) (supplementary Fig. 3). In HER2-positive (HER2+) tumours, no significant association between BMI1 and outcome was identified.

The Cox regression model, including age at diagnosis, tumour size, tumour grade and nodal stage, showed that BMI1 as an independent predictor of good prognosis in the whole BC cohort and in the luminal ER+ patients (P = 0.04, HR 0.72; 95% Cl 0.51–0.99, P = 0.04, and HR 0.69; 95% Cl 0.48–0.99, respectively) (Tables 2, 3) but not in the TNBC subtype. When the multivariate analysis including BMI1 and other BCSC (ALDH1A1, CD133, CD24, and SOX9) markers, BMI1 was an independent predictor of good prognosis in the whole BC cohort (P = 0.017), supplementary Table 3.

With regard to the transcriptomic expression, high BMI1 mRNA expression in the METABRIC cohort was significantly associated with longer BCSS in the whole BC cohort (P < 0.001) as well as in the luminal ER+ tumours (P < 0.001) (see Fig. 3). However, no significant association between BMI1 and outcome was observed in ER− tumours or in HER2+ patients (P > 0.05). Moreover, CN gain of BMI1 was associated with longer BCSS (P = 0.031) in whole BC cohort.

Similar results were observed with the BC Gene Expression Miner dataset, where BMI1 mRNA was associated with longer survival in the ER+ tumour (P = 0.020), which confirms the previous findings in the METABRIC cohort as shown in Supplementary Fig. 4.

We further investigated the association of BMI1 expression and key BC stem cell (BCSC) biomarkers. At protein level, there was a significant negative correlation between the high expression of BMI1 and BCSC markers: ALDH1A1 (P = 0.017), CD133 (P = 0.006) and SOX9 (P = 0.004) in all BC cases (supplementary Table 4a). In the ER+ tumours, BMI1 maintained similar associations and showed negative correlation with ALDH1A1 (P = 0.005) and CD133 (P = 0.001). Interestingly, in the ER- tumours, BMI1 showed positive association with CD133 (P < 0.0001) and SOX10 (P = 0.01) expression (supplementary Table 4a).

Similarly at the mRNA level, high expression of BMI1 was negatively associated with BCSC including ALDH1A3 (P < 0.0001), CD133 (P = 0.0003), CD24 (P = 0.002) and CD44 (P = 0.0007) (Supplementary Table 4b) in the whole BC cohort. When the analysis was restricted to the luminal ER+ tumours, high BMI1 expression was associated with low expression of ALDH1A1 (P = 0.025), ALDH1A3 (P = 0.026), CD133 (P = 0.038), CD44 (P = 0.0001), CD24 (P = 0.004) and SOX10 (0.0006) (Supplementary Table 4b). In the basal TNBC subtype, positive association was observed between BMI1 and EPCAM and SOX10 (P < 0.001).