SUPERTAM_HGU133A Gene Set Enrichment Analysis (GSEA) in term of lymph node and distant metastasis
Description
Background Lymph node status is generally considered as one of the best prognostic factors in breast cancer. However, association between nodal and distant metastasis is not straightforward. Here we analyze differences between molecular mechanisms responsible for nodal and distant metastasis, and the impact of molecular subtype on this association. Methods We use SUPERTAM_HGU133A cohort of 836 patients with full microarray data available. Logistic regression evaluates distant metastasis risk, and Gene Set Enrichment Analysis (GSEA) is used to identify molecular mechanisms targetable in the key clinical scenarios. Results GSEA shows different molecular background for lymph node and distant metastasis and unique patterns of deregulated molecular processes in tumors of four breast cancer subtypes. Risk of distant metastasis in node positive luminal A patients is strong in SUPERTAM_HGU133A (OR: 2.401, CI: 1.316-4.380) dataset. For luminal A tumors, nodal positivity is associated with enrichment of NF-κB and Src, while distant metastasis is associated with strong mechanisms of cell cycle regulation, thrombolysis, DNA-repair, and immune response. Based on GSEA results, we select panels of promising inhibitors applicable for the key clinical scenarios depending on lymph node status that are currently being tested in vitro, in vivo, or in clinical trials. Conclusions Potential molecular targets are different for nodal and distant metastasis in breast cancer and are highly variable among different molecular subtypes. Panels of inhibitors have potential to improve the outcome of luminal A breast cancer patients based on lymph node status. We hope that further clinical trials have potential to translate the current knowledge from the laboratory to an improved treatment of breast cancer patients.
Files
Steps to reproduce
Publicly available gene expression dataset SUPERTAM_HGU133A consisting of data from MDA5, TAM, VDX and VDX3 datasets (all platform Affymetrix Human Genome U133A, 856 samples in total) was downloaded from Gene Expression Omnibus (GEO) database under IDs: GEO: GSE17705 (MDA5), GEO: GSE6532/GSE9195 (TAM), GEO: GSE2034/GSE5327 (VDX) and GEO: GSE12093 (VDX3) in a log2 normalized form. The most variable probeset per gene based on interquantile range (IQR) was selected. Samples with available lymph node status and documented distant relapse (836 samples in total) were classified into four molecular breast cancer subtypes using a SCMOD2 classification model, as it showed higher robustness than commonly used PAM50. This resulted in 341 luminal A, 281 luminal B, 71 Her2+ and 143 basal patient samples. All calculations were performed in R 3.4.1 using limma 3.32.2 package from Bioconductor. Association between gene expression and local or distant metastases was assessed by moderated t-statistics (method implemented in the Limma package version 3.32.2 in R version 3.4.1) on the set of 13 091 genes. The most variable probesets per gene was selected from the original 22,283 probesets. P-values were adjusted for multiple hypothesis testing by Benjamini-Hochberg FDR correction. To find the most involved pathways in the metastasis associated processes, we used javaGSEA 4.0.3 desktop application. Student t test was used for ranking the genes, minimal size of small sets for exclusion was set to 10, 1000 permutations were used, and default settings were used for other parameters. Enrichment analysis applied pathways information from BIOCARTA database. Enrichment score (ES) was calculated for each gene set. Pathways were considered significant (i) if nominal p-value was below 0.05 and (ii) the pathways were enriched in lymph node positive or distant metastasis positive phenotype.