Basal BCs are poor-prognosis tumors, which require both improvement of our ability to predict the clinical outcome for better tailoring treatment and identification of new therapeutic targets. Their prognosis is heterogeneous, and it is currently impossible to predict which patients will or will not relapse using classical histoclinical factors or the recently reported prognostic GES, notably those currently tested in clinical trials . In the same way, the HDDP classifier  identified using ERBB2+ tumors, failed to classify basal samples. Prognostic analyses should be done per subtype .
Analysis of kinase and kinase-related genes might help develop new targeted therapies. We report a kinase-based model that divides basal BCs into two subgroups with balanced histoclinical factors but different survival (25% difference for 5-year DFS). This model is based on the expression of an immune 28-gene metagene. Identified in a learning set, its prognostic value was confirmed in an independent data set of 518 cases. The model outperformed the individual current prognostic factors on multivariate analysis, both in the learning and validation sets. Patients with high expression of the immune metagene had a better DFS than other patients. This prognostic value remained when applied to patients treated without any adjuvant chemotherapy, suggesting a link with the metastatic potential. An additional link with chemosensitivity cannot be excluded as "Immune-High" patients experienced a higher, but non significant, pCR rate than "Immune-Low" patients.
The favorable prognostic impact of the immune response, particularly the T-cell response, has been reported in ER-negative [8, 13, 14, 26, 41–43] or ERBB2+ BCs [8, 28, 31, 44]. Similar finding was reported in 97 triple-negative BCs  in which increased expression of interferon-related genes tended to confer better prognosis. In our previous study  and the present one, we focused on basal BC only, since this subtype is even more homogeneous than the triple-negative group . In our previous study, we defined a 368-gene prognosticator, which confirmed the positive influence of TH1 cells and high cytotoxic activity. This model outperformed two immune signatures in multivariate analysis of DFS [28, 42]. We showed here that both the immune kinase model and this previous model maintain their prognostic value in multivariate analysis, suggesting their independence. It is of note that our "immune-metagene" model presented a prognostic value in luminal B (p = 0.03, Wald test) and ERBB2-overexpressing cases (p = 0.02, Wald test), but not in luminal A and normal-like samples (p = 0.58 and 0.98, respectively, Wald test). Moreover, it is worth noting that previously published signatures (Genomic grade index, 70-gene signature, and 76-gene signature), mainly based on proliferation, failed to separate good from poor prognosis basal breast cancers.
Ingenuity analysis of both the 28 genes and the genes differentially expressed between the two subgroups defined by our kinase immune metagene confirmed that the differences between these histoclinically similar subgroups are in immune genes. Upregulated kinome-genes suggest the presence of an activated lymphocyte infiltrate in "Immune-High" patients. This lymphocyte-activated status is due to stimulations by cytokines (JAK3, STAT1, STAT4, TBX21 and TH1 cytokine receptors), by T-cell receptor (T-cell receptor chains [alpha, beta and gamma], CD3E, CD3D, CD247/CD3Z, CD28, CD27, CD2, CD8A, CD4, LAG3, MAL, LAT2, PIM2), by B-cell receptor (CD19, CD79b, CD27, CD40, IGJ, IGK@, IGH@, BTK, BLNK, BANK1), and by anti-tumor receptors (KLRK1, KLRB1, GAB3, SLAMF1, SLAMF6-8). The lymphocyte infiltrate is strictly TH1-biased with the overexpression of IL2RG, IL23RB and IL7R involved in lymphocyte survival, of IL12RB1, IL15RA, IL18BP, and IL21R TH1-biased receptors, of STAT1, STAT4, and TBX21 TH1 transcription factors, and of several interferon-inducible molecules (GVIN1, ISG20, GBP2, IRF1, IRF4, IRF7, and IRF8). This agrees with increased levels of cytotoxic granules and pore-forming molecules (VAMP1, GZMA, GZMB, GZMH, GZMK, GNL, PRF1, CFLAR, CASP1, and CASP10). Interestingly, there are also several genes encoding activated memory lymphocyte recruitment such as IL16, XCL1, CCL5, CCR5, CXCL9, CXCR3, CCL19, CCR7, and CXCR6 (mostly helper and cytotoxic T-cells), and CXCL13, CXCR5 (activated B-cells), among which some are strictly produced by activated T-cells, such as CCL4 and CCL5. Finally, we also found transcripts involved in lymphocyte migration and/or activation (ITGAL and ITGB2 heterodimers, ITGA4, ITGAX, ITGB7, SELL, SELP, SELPL, and CD69).
Thus, we show that the immune response, and notably the adaptive cytotoxic TH1-cell response , influence survival of basal BC patients. Despite the small size of the independent population with lymphocyte infiltrate data available, which does not allow to really conclude about the impact of the quantity of lymphocyte infiltrate on the expression of immune response-related genes, the absence of correlation between the immune metagene and lymphocyte infiltration in our cohort and in two independent data sets [5, 8] as well as the function of genes, suggest that this influence does not depend on the degree of lymphocyte infiltrate, but on the efficiency of its cytotoxic activation status. The differential expression of these "immune genes" is probably also due to a variable expression of epithelial-derived molecules [13, 42, 48], which activate (in "Immune-High" cases) or repress (in "Immune-Low" cases) the local immune response to the tumor. These hypotheses deserve further investigation to understand the respective role of tumor-infiltrative lymphocytes and cancer cells on cancer history.