Elevated MMP9 expression in breast cancer is a predictor of shorter patient survival

This study obtained ethics approval by the North West–Greater Manchester Central Research Ethics Committee under the title; Nottingham Health Science Biobank (NHSB), reference number 15/NW/0685. All samples from Nottingham used in this study were pseudo-anonymised and collected prior to 2006 and stored in compliance with the UK Human Tissue Act. MMP9 protein expression was evaluated using a well-characterised cohort of early-stage (operable) primary invasive BC (n = 675) treated in Nottingham University Hospital NHS Trust as previously described [15]. Clinical and pathological data of patients (including hormone receptor status, histological tumour type, tumour grade, tumour size, lymph node status, Nottingham Prognostic Index and lymphovascular invasion (LVI)) were available. Tumour types were revisited and coded according to the recent WHO blue book [16]. BC-specific survival (BCSS) was maintained on a prospective basis. The expression of a large panel of BC progression/metastasis-related biomarkers, including the Ki67 [17], EGFR [18], CDC42 [19], CD44 [20], PIK3CA [21] and basal marker (cytokeratin 5/6 and 17) [22], was also studied. BC molecular subtypes based on the IHC profile of Oestrogen Receptor (ER), Progesterone Receptor (PR) and Human Epidermal Growth Factor 2 (HER2) were defined as previously described [23]: Luminal A: ER+/HER2− Low Proliferation (Ki67 < 10%), Luminal B: ER+/HER2− High Proliferation (Ki67 ≥ 10%) or ER +/HER2 + , HER2-positive class: HER2 + regardless of ER status, Triple Negative Breast Cancer (TNBC): exhibiting negative expression of ER, PR, and HER2.

The clinicopathological significance of MMP9 mRNA expression, copy number alterations, differential gene expression analysis (DGE), and pathway analysis were assessed using the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) dataset (n = 1980 BC cases) [24]. External validation was performed using the Breast Cancer Gene-Expression Miner v4.0 (BC-GenExMiner v4.0) [15, 25] and The Cancer Genome Atlas (TCGA) [26] as previously described.

Specificity of MMP9 antibody was validated by western blotting prior to immunohistochemistry. Cell lysates blots of HEK293 and MCF7 cell lines which were used as positive and negative controls respectively (obtained from the American Type Culture Collection; Rockville, MD, USA) were incubated with anti-MMP9 antibody (Rabbit monoclonal [EP1254], Abcam) at 1:800 dilution for overnight (4 °C) and bands were detected using fluorescent secondary antibodies at (1:15,000) (IR Dye 800CW donkey anti-rabbit and 680RD donkey anti-mouse, LI-COR Biosciences, UK). Mouse β-Actin (A5441, Sigma-Aldrich; Clone AC-15; Sigma, UK) at 1:5000 was used as a house-keeping protein. Blocking and visualisation were done as previously documented [27]. The specificity of the antibody was validated with a single specific band at the predicted molecular weight (70 kDa, Supplementary Fig. 1a).

Tumour samples were arrayed onto tissue microarrays (TMAs) as previously described [28]. Full-face BC tissue sections and TMAs were immunoassayed using Novolink Max Polymer Detection system (Leica, Newcastle, UK). In brief, 4 µm BC tissue sections were deparaffinised with xylene and rehydrated through 100% ethanol. Heat-induced (pH6) citrate antigen retrieval was performed and MMP9 antibody (1:100) was incubated for overnight at 4 °C. 3-3′ Diaminobenzidine tetrahydrochloride (Novolink DAB substrate buffer plus) was used as the chromogen. Slides were counterstained with Novolink haematoxylin for 6 min, dehydrated and cover slipped. Normal kidney tissue was used as a positive tissue control, whereas no primary antibody was used as a negative control.

TMA slides were digitally scanned at 20X magnification and viewed using NanoZoomer NDP viewer (Hamamatsu Photonics, Welwyn Garden City, UK). Both the percentage of staining and staining intensity of MMP9 cytoplasmic expression in invasive tumour cells and stromal cells were individually assessed to calculate the final histochemical score (H-score). Staining was double scored blindly by two researchers (NO and IMM) for 25% cores to assess inter-observer concordance. Inter-observer agreement was determined, and the interclass correlation coefficient was 0.86, indicating an excellent concordance between scorers. Discordant cases were re-scored by both observers and a final score was agreed.

IBM SPSS 24.0 (SPSS IBM Corp, Chicago, IL, USA) software was used for statistical analysis and reported in line with REMARK guidelines [29]. The MMP9 H-score was dichotomised into high and negative/low expression using the median cut-point value. Chi-squared test was used to evaluate the association between MMP9 expression and the clinicopathological parameters.

Gene expression was analysed in the subset of METABRIC patients for which MMP9 expression was available. DGE between A) low Vs high MMP9 cytoplasmic expression, and B) low Vs high MMP9 stromal expression were calculated using the Robina implementation of Edge-R statistical tool [30] and DGE with > twofold-change, and adjusted p values < 0.05 were considered significant. The DGE were examined using the online WebGestalt platform and adjusted p < 0.01 considered statistically significant [31, 32] and significantly enriched gene ontologies common for both cytoplasmic and stromal MMP9 protein expression. Furthermore, Venny 2.0 [33] was used to identify the overlapping DGEs common to both the cytoplasmic and stromal MMP9 protein expression.

Full-face BC tissue (n = 10) sections were used to evaluate the pattern of MMP9 protein expression prior to staining of TMAs. This showed uniformly weak MMP9 expression in normal glandular epithelium (Fig. 1a) and ductal carcinoma in situ (DCIS; Fig. 1b). There was slightly increased immunoreactivity of MMP9 observed in the co-existing invasive BC cells (Fig. 1c), in the intravascular tumour cell emboli (Fig. 1d), stromal expression (Fig. 1e) and Fig. 1f showing No Primary Antibody Control. MMP9 cytoplasmic and stromal protein expression (C+/S+) showed a positive correlation with MMP9 mRNA (Spearman’s coefficient 0.218; p = 0.027), this association was confirmed using TCGA data [34, 35] (Supplementary Fig. 1b).

On BC TMAs, a variable degree of MMP9 protein expression was observed (Fig. 1g–k). The H-scores of both cytoplasmic and stromal expressions did not follow a normal distribution. Therefore, for dichotomisation into negative/low and high expression, the median H-scores (0 and 50, respectively) were used. Out of 675 informative TMA cores, 71% showed negative/low expression in the cytoplasm (Fig. 1g) while 29% showed high expression (Fig. 1h). In the stroma 53% showed negative/low expression (Fig. 1i), while 47% showed high expression (Fig. 1j). Positive immunoreactivity was observed in human kidney sections (Fig. 1k). H-scores for cytoplasmic and stromal staining showed a positive correlation (Spearman’s coefficient 0.1; p = 0.020).

Elevated MMP9 cytoplasmic and stromal staining were associated with high tumour grade (p < 0.0001), poor Nottingham Prognostic Index (NPI) (p < 0.0001; Table 1), hormone receptor negativity (p < 0.01), among IHC subtypes associated with TNBC and HER2 + tumours (p = 0.002). High stromal MMP9 expression additionally showed association with LVI positivity (p = 0.046; Table 1). High MMP9 expression positively associated with proliferation marker Ki67 (p = 0.004), signalling pathway associated markers; EGFR (p = 0.024) and PIK3CA (p = 0.039), cytoskeleton remodelling markers; CDC42 and CD44 (p < 0.003 and p = 0.025, respectively). High cytoplasmic MMP9 showed positive association with basal cytokeratin CK17 (p = 0.001; Table 2) and revealed a low expression in lobular and special type tumours (p = 0.003).

BCSS of patients with tumours expressing high cytoplasmic MMP9 was significantly shorter than that of the negative/low expression subgroup (p = 0.013; HR = 1.5; 95%CI 1.1–2.0; Fig. 2a). Stromal MMP9 also showed a similar trend but did not reach significance (p = 0.058; HR = 1.3; 95%CI 0.9–1.8; Fig. 2b). Combined MMP9 cytoplasmic and stromal (CS) survival analysis demonstrated that tumours with high cytoplasmic and stromal expression were associated with poor prognosis (p = 0.008; HR = 1.2; 95%CI 1.0–1.4; Fig. 2c). In multivariate Cox regression analysis, cytoplasmic MMP9 expression on its own was an independent predictor of BCSS (p = 0.026; HR = 1.6; 95% CI 1.1–2.3; Table 3). There was no association between MMP9 protein and outcome in Luminal A and B, TNBC or HER2 + subgroups.

Consistent with the results obtained for MMP9 protein expression, in the METABRIC and TCGA datasets, MMP9 copy number gain (14.0%) and high mRNA expression (50.1%) was associated with negative ER and PR status, high histological grade, and poor NPI (all; p < 0.01; Table 4). MMP9 copy number gain and high MMP9 mRNA expression was associated with poor prognostic METABRIC Integrative clusters [24] such as 1, 5 and 9 (p < 0.0001). Associations between MMP9 copy number alterations, mRNA expression and clinicopathological variables are summarised in Table 4. External validation of the pooled data using BC-GenExMiner v4.0 was in agreement where high MMP9 mRNA expression was associated with ER, and PR negativity, high histological grade (all; p < 0.001) and poor NPI (p < 0.01). In PAM50 subtypes, high MMP9 expression was associated with basal-like and HER2 + classes (p < 0.0001).

High MMP9 mRNA showed positive association with other MMPs (MMP1, MMP2, MMP7, MMP11, MMP14 and MMP15; all p = 0.0001), collagens [COL27A1; (p = 0.008), COL23A1 (p = 0.003) and COL11A2; (p = 0.045)], TGFβ1 (p < 0.0001) and cytoskeletal remodelling biomarkers CDC42 (p = 0.004; Table 5). These associations were confirmed using BC-GenExMiner v4.0 (Supplementary Fig. 2) via gene targeted analysis.

In the METABRIC cohort, MMP9 copy number gain was associated with significantly shorter BCSS (p = 0.016; HR 1.3; 95%CI 1.1–1.7; Fig. 2d). Tumours expressing high MMP9 mRNA expression showed significantly shorter BCSS than the low expression subgroup (p < 0.0001; HR = 1.5; 95%CI 1.2–1.8; Fig. 2e). Pooled MMP9 gene expression data (n = 2071) in BC-GenExMiner v4.0 confirmed the association of high MMP9 expression with poorer outcome (p = 0.007; HR = 1.3; 95%CI 1.1–1.5; Fig. 2f) and in agreement with protein expression results. In multivariate Cox regression analysis, MMP9 mRNA was an independent predictor of BCSS (p = 0.048; HR = 1.3; 95% CI 1.0–1.6) independent of the standard prognostic parameters of BC including tumour size, histological grade, nodal stage, and proliferative fraction as assessed by Ki67.

To understand the molecular biology of MMP9 protein expression as an end point, the subset of the Nottingham series that was included in the METABRIC dataset (n = 113) were used for DGE analysis. The dichotomisation of cases into negative/low versus high groups was based on the dichotomisation of MMP9 protein expression. Cytoplasmic MMP9 expression displayed high expression in 36/113 cases (32%), while, stromal MMP9 showed high expression in 53/113 cases (47%). DGE analysis identified 1630 significantly differentially expressed genes associated with cytoplasmic MMP9 expression, with decreased cytoplasmic MMP9 expression displayed 720 upregulated and 910 downregulated genes, respectively. Stromal MMP9 showed 1480 differentially regulated genes, reduced stromal MMP9 expression was associated with 667 upregulated and 813 downregulated differentially expressed genes (Fig. 3). The overlapping DGEs between (A) low Vs high MMP9 cytoplasmic expression, and (B) low Vs high MMP9 stromal expression revealed 277 upregulated differentially expressed genes and 276 downregulated differentially expressed genes (Fig. 4). The common differentially expressed genes (n = 553) which were significantly associated with ECM related gene ontologies (Fig. 5, Table 6).