3.1 Osterix protein expression in invasive breast cancer

Antibody specificity was confirmed using Western blotting prior to staining patient tissues (Fig. 1A). 1340 breast tumours were examined for both cytoplasmic and nuclear osterix expression; representative tissue staining is shown in Fig. 1B and C. Expression of osterix within the cytoplasm was significantly correlated with osterix expression in the nucleus (P < 0.001; R2 = 0.435).

Fig. 1

(A) Western blot assessment of osterix expression in breast cancer cell lines. Representative photomicrographs of high osterix immunohistochemical staining (B), and low staining (C), where photomicrographs are shown at 10× magnification with 20× magnification inset box

The median H-score for cytoplasmic expression of osterix was 140 (ranging between 50 and 240), and the median percentage score for nuclear osterix expression was 30 (ranging between 0 and 90). X-tile was used to generate cut points, with a cut point of 185 used for cytoplasmic osterix expression and a cut point of 25 used for nuclear osterix expression. Nuclear and cytoplasmic osterix expression were significantly correlated with one another (P < 0.001, R2 = 0.435) but not with strong biological relevance.

3.2 Osterix protein expression and association with clinicopathological parameters

Associations between cytoplasmic and nuclear expression of osterix protein expression and clinicopathological variables were assessed. Low levels of nuclear expression of osterix were found to be significantly associated with clinicopathological parameters characteristic of aggressive behavior using Pearson χ2 test of association, including larger tumour size (χ2 = 4.479, d.f.=1, P = 0.034), ER negative tumours (χ2 = 8.165, d.f.=1, P = 0.004), progesterone receptor (PR) negative tumours (χ2 = 7.876, d.f.=1, P = 0.005), triple receptor negative tumours (χ2 = 5.177, d.f.=1, P = 0.023), positive vascular invasion (χ2 = 6.506, d.f.=1, P = 0.011) and histological type (χ2 = 16.738, d.f.=4, P = 0.002), at the time of presentation; data is shown in Table 1. Linear by linear χ 2 tests were performed to identify linear trends; low levels of nuclear expression of osterix had a significant trend with increasing tumour grade (χ2 = 12.969, d.f.=2, P < 0.001), increasing pleomorphism variation score (χ2 = 7.120, d.f.=2, P = 0.008), and worse NPI prognostic group (χ2 = 10.831, d.f.=2, P < 0.001) at the time of presentation; data is shown in Table 1.

Table 1 Associations between the cytoplasmic and nuclear expression of Osterix determined using immunohistochemistry with clinicopathological variables. The P values are resultant from pearson Χ2 test of association, or linear by linear Χ2 test if an * is present, and significant values (P < 0.05) are highlighted in bold. ER is oestrogen receptor and PgR is progesterone receptor

Similarly, low levels of cytoplasmic osterix expression were significantly associated with larger tumour size (χ2 = 6.220, d.f.=1, P = 0.013) and histological type (χ2 = 25.863, d.f.=4, P < 0.001) at the time of presentation using Pearson χ 2 test of association; data is shown in Table 1. Linear by linear χ 2 tests were performed to identify linear trends; low levels of cytoplasmic expression of osterix had a significant trend with increasing tumour grade (χ2 = 12.969, d.f.=2, P < 0.001), increasing mitotic count score (χ2 = 6.458, d.f.=2, P = 0.011), and worse NPI prognostic group (χ2 = 6.412, d.f.=2, P = 0.011) ) at the time of presentation; data is shown in Table 1.

3.3 Osterix protein expression and patient outcome

Kaplan-Meier analysis was used to assess the impact of osterix protein expression on breast cancer-specific survival. Low nuclear osterix expression was significantly associated with adverse breast cancer specific survival (P = 0.016) (Fig. 2B) and shorter distant metastasis free survival (P = 0.003) (Fig. 2D). Cytoplasmic osterix expression was not associated with disease specific patient survival (P = 0.165) (Fig. 2A) or distant metastasis free survival (P = 0.126) (Fig. 2C).

Fig. 2

Kaplan–Meier analysis of breast cancer specific survival showing the impact of low (grey line) vs. high (black line) protein expression: (A) osterix cytoplasmic expression; (B) osterix nuclear expression of low (grey line) and high (black line) protein expression: (C) osterix cytoplasmic expression; (D) osterix nuclear expression

Multivariate survival analysis was performed using Cox’s proportional hazard method, and included tumour size, tumour grade, and stage, low nuclear osterix expression was independently associated with distant metastasis free survival (hazard ratio (HR) = 0.777, 95% confidence interval (CI) = 0.614–0.982, P = 0.034), but not disease specific survival (HR = 0.826, 95% CI = 0.644–1.058, P = 0.130).

The effect of osterix expression on clinical outcome in patient subgroups was explored. Low nuclear expression of osterix was significantly associated with both disease free survival and metastasis free survival in patients with HER2 positive tumours (P = 0.027 and P = 0.008 respectively), but not in patients with HER2 negative tumours (Fig. 3).

Fig. 3

Kaplan–Meier analysis of breast cancer specific survival showing the impact of low (grey line) vs. high (black line) nuclear osterix expression: (A) HER2 positive tumours; (B) HER2 negative tumours; (C) HER2 positive tumours; (D) HER2 negative tumours

3.4 SP7 mRNA expression in breast cancer

SP7 mRNA expression was analysed in the METABRIC patient cohort, which has limited overlap of patient tumours assessed within the tissue microarray. Data analysis was performed on mRNA expression z-scores relative to all samples with a median expression of -0.485 ranging from − 3.20 to 14.54. X-tile was used to generate a cut point of -0.20 to dichotemise data into low and high expression, with 847 cases having low SP7 mRNA expression and 1134 cases having high SP7 expression. Expression of SP7 was not associated with survival in the total patient cohort (Fig. 4A). Associations between SP7 expression and clinicopathological variables were assessed, but no significant associations were observed (Table 2).

Fig. 4

(A) Kaplan–Meier analysis of breast cancer specific survival showing the impact of low (grey line) vs. high (black line) SP7 mRNA expression; (B) correlation matrix demonstrating Pearson correlation between SP7 mRNA and network partners in the METABRIC cohort; (C) gene enrichment plot of hallmark geneset significantly enriched in high SP7 mRNA group

Table 2 Associations between mRNA expression of SP7 and clinicopathological variables in the METABRIC patient cohort. The P values are resultant from pearson Χ2 test of association and significant values (P < 0.05) are highlighted in bold. ER is oestrogen receptor and PgR is progesterone receptor3.5 SP7 mRNA expression with known network partners

Gene expression data from network partners identified in String (v12.0), or through publication (VEGF, MMP9, S100A4) were assessed for correlation with SP7 (Fig. 4B). SP7 mRNA expression was not significantly associated with expression of RIOX1, RUNX2, NFATC1, CTNNB1, DLX5, IBSP, BMP2, COL1A1, MMP9, VEGFA, VEGFB, and S100A4; but expression was significantly, but weakly negatively correlated with SOST (R2=-0.067, P = 0.003), and weakly positively correlated with VEGFC (R2 = 0.072, P = 0.001) expression.

3.6 Gene enrichment analysis

GSEA was used to explore the METABRIC microarray data for enrichment of genes in the curated hallmarks of cancer gene sets in high and low SP7 expressing tumours. Normalised enrichment scores (NES) with a false discovery rate of less than 1% identified 1/50 enriched gene sets in tumours with high expression of SP7, but none in tumours with low expression of SP7. The gene set enriched in high SP7 expressing tumours was the geneset that represents genes down-regulated by K-Ras activation (HALLMARK_KRAS_SIGNALLING_DN), with an NES of 1.58 (P = 0.0058) (Fig. 4C).