Characterization of extracellular matrix composition in breast carcinoma

Background

A different view of the tumour as a functional tissue interconnected with the microenvironment has recently been described [1]. Numerous reports in the past years indicate that the growth and progression of breast cancer cells, as well as other tumour cells, depend not only on their malignant potential, but also on stroma components present in the surrounding microenvironment [2]. Recent gene expression profiling studies on breast cancer showed that molecular classification of tumours based on the gene expression patterns can identify clinically different subtypes of cancer with different prognosis or disease outcome [3, 4]. However, since tumours are functional tissues dynamically interconnected with the micro-environment, this approach could give even more information if the tumor phenotype profile is related to the tumor-surrounding stroma characteristics.

Materials and methods

Immunohistochemical staining was carried out for several extracellular matrix (ECM) molecules such as fibronectin, fibulin 1, and laminin in a cohort of 29 formalin-fixed, paraffin-embedded primary breast tumours. Furthermore, evaluation of haematoxylin and eosin sections was performed to classify the surrounding stroma in categories of loose, dense or mixed, respectively.

Gene expression analysis was performed using 22K 60-mer Human 1A Oligonucleotide (G4110A) provided by Agilent. A specific extracellular matrix ECM gene list of 282 unique ECM-related genes was created by a basic search in the Human 1A(V2) platform on the Agilent website [5] and was used to interrogate the 29 breast cancer transcriptional profiles. To validate and test the robustness of the obtained results, a new dataset of 123 primary breast carcinomas was queried with the ECM gene list.

Results

We defined a set of 282 ECM-related genes whose expression separated the tumours in three main groups. We compared the ECM groups, defined by gene expression profile, with the IHC staining results and with stroma categories. Significant correlation between stroma categories and the ECM gene expression profile was retrieved. In contrast, the IHC results were not significantly correlated to the results from with the other methodologies we applied.

Samples were also classified based on breast cancer subtypes published by Sørlie and colleagues [6] using a selected list of 534 genes, known to discriminate tumours in subclasses with clinical implication. In our study, basal-like tumours showed a strong steadiness in clustering in a specific group (ECM1), while the other subtypes were clustering across the three ECM groups.

To validate our data and to correlate the ECM groups to clinical outcome, we utilized the ECM gene list in a new dataset of 123 samples, where long-term follow-up information was available. The new analysis allowed us to identify the same three clusters, indicating the robustness of the ECM classification. Survival analyses showed significantly different outcomes for the patients belonging to the three ECM groups. The ECM1 group was associated with poor overall survival and this was not only related to the high frequency of basal tumors in this group, which are known to display a poor prognosis, since luminal-like tumors were also classified in this group.

Conclusion

Gene expression profiling of breast carcinomas allows the identification of three subgroups of tumors according to ECM-associated gene expression. This classification provides new information on breast carcinoma biology and new parameters that may impact both prognosis and prediction of response to therapy.