Differentiation of ERBB2-induced breast cancer according to different traits. (A) Experimental design. We generated a cohort of mice that were genetically heterogeneous and with different degrees of susceptibility and progression to breast cancer. Thus, we dissected the disease into different pathophenotypes. Then, we classified the mice according to prognosis. We considered different clinical pathophenotypes of ERBB2-positive tumors as the result of multiple molecular associations at different levels from a systems biology perspective to predict clinical outcomes. In sum, we show a global scenario of complex interactions at different levels that account for the heterogeneity in tumor behavior and susceptibility. (B) Comparison among different tumor pathophenotypes in FVB, F1 and F1BX mice. We show detailed comparisons of all the different pathophenotypes elsewhere (Table S1 in Additional file 1). (C) The network shows associations among different tumor pathophenotypes in the F1BX population (Cytoscape Software). Nodes represent breast cancer pathophenotypes: the yellow nodes are temporal stages of the disease and the blue nodes depict tumor traits. Edges indicate correlation coefficients, green for positive and red for negative correlations. Edge line width is directly proportional to the value of the r coefficient. We include all correlations with P < 0.05 in the figure. All r coefficient values are in Table S2 in Additional file 1. (D) Prognosis clusters identified by principal component analysis as shown by an HJ-biplot. A fifth cluster formed by the mice without tumors is not included in the figure. (E) Distribution of tumor pathophenotypes in each different cluster. The fifth cluster, which encompasses the mice that did not develop tumors, is not represented in the figure. The numerical values of this figure and post hoc tests are reported elsewhere (Table S3 in Additional file 1).