ABSTRACT
Intro: In vitro cell line models provide a valuable resource to investigate compounds useful in the systemic chemotherapy of cancer. However, the due to the dispersal of the data into several different databases, the utilization of these resources is limited. Here, our aim was to establish a platform enabling the validation of chemoresistance-associated genes and the ranking of available cell line models. Methods: We processed four independent databases, DepMap, GDSC1, GDSC2, and CTRP. The gene expression data was quantile normalized and HUGO gene names were assigned to have unambiguous identification of the genes. Resistance values were exported for all agents. The correlation between gene expression and therapy resistance is computed using ROC test. Results: We combined four datasets with chemosensitivity data of 1562 agents and transcriptome-level gene expression of 1250 cancer cell lines. We have set up an online tool utilizing this database to correlate available cell line sensitivity data and treatment response in a uniform analysis pipeline (www.rocplot.com/cells). We employed the established pipeline to by rank genes related to resistance against afatinib and lapatinib, two inhibitors of the tyrosine-kinase domain of ERBB2. Discussion: The computational tool is useful 1) to correlate gene expression with resistance, 2) to identify and rank resistant and sensitive cell lines, and 3) to rank resistance associated genes, cancer hallmarks, and gene ontology pathways. The platform will be an invaluable support to speed up cancer research by validating gene-resistance correlations and by selecting the best cell line models for new experiments.
ABSTRACT
Limited therapeutic options exist for the treatment of patients with triple negative breast cancer (TNBC). Neoadjuvant chemotherapy is currently the standard of care treatment in the early stages of the disease, although reliable biomarkers of response have been scarcely described. In our study we explored whether immunologic signatures associated with inflamed tumors or hot tumors could predict the outcome to neoadjuvant chemotherapy. Publicly available transcriptomic data of more than 2,000 patients were evaluated. ROC plots were generated to assess the response to therapy. Cox proportional hazards regression was computed. Kaplan-Meier plots were drawn to visualize the survival differences. Higher expression of IDO1, CXCL9, CXCL10, HLA-DRA, HLA-E, STAT1, and GZMB were associated with a higher proportion without relapse in the first 5 y after chemotherapy in TNBC. The expression of these genes was associated with a high presence of CD8 T cells in responder patients using the EPIC bioinformatic tool. The strongest effect was observed for STAT1 (p = 1.8e-05 and AUC 0.69, p = 2.7e-06). The best gene-set signature to predict favorable RFS was the combination of IDO1, LAG3, STAT1, and GZMB (HR = 0.28, CI = 0.17-0.46, p = 9.8 E-08, FDR = 1%). However, no influence on pathological complete response (pCR) was observed. Similarly, no benefit was identified in any other tumor subtype: HER2 or estrogen receptor positive. In conclusion, we describe a set of immunologic genes that predict the outcome to neoadjuvant chemotherapy in TNBC, but not pCR, suggesting that this non-time to event endpoint is not a good surrogate marker to detect the long term outcome for immune activated tumors.