Recently, it has been increasingly recognized that ER positive and ER negative breast cancer are distinct types of breast cancer. To date, few studies have rigorously assessed drug response genes in ER negative and ER positive breast cancer. In this analysis, genes related to multidrug response in ER positive and ER negative breast cell lines were comprehensively identified. The results show that the genes related to multidrug response in ER positive cell lines are distinct from those in ER negative cell lines. Among 188 genes identified in ER positive cell lines (123 positively related, 65 negatively related) and 32 identified in ER negative cell lines (14 positively related, 18 negatively related), two genes (TOP2A and DBI) have similar association in both cell types, and one gene (PMVK) associated in opposing directions in each cell type. By strictly controlling variables, including using gene expression profiles measured by the same platform and the same set of cells, testing the same panel of drugs, and using the same well-established chemoresponse assay, our results strongly indicate that the limited gene overlap is related to differences inherent in ER status.
Functional analysis also indicates that different biological processes are related to drug response in ER positive versus ER negative breast cells. Most of the enriched pathways in ER positive cells are associated with various types of cellular signaling, including cell cycle regulation, apoptosis, cellular stress and injury, cytokine signaling, and growth factor signaling. Noticeably, a number of signaling pathways that are uniquely enriched in the ER positive cell lines have complex cross talk with ER signaling at the receptor level (HMMR and ITGA3), as well as downstream of the receptor level, such as signaling adaptor proteins (e.g. SHC1), kinases (e.g. PI3K, PRKC1, CRKL and ERK), phosphatases (e.g. PPP2R2A, PPP2R5E), enzymes (NQO1), and transcription regulators (e.g. CITED2, CNOT7 and STAT6).
In ER negative cells, all five enriched pathways are related to metabolic functions, including Glycolysis/Gluconeogenesis, Phenylalanine Metabolism, Methane Metabolism, Stilbene, Coumarine and Lignin Biosynthesis, Biosynthesis of Steroids. This is biologically reasonable since these pathways have been shown to play important roles in cell adhesion and modulate signaling, which may affect drug response. Particularly, previous studies showed that cells with high glycolytic activity tend to have a decreased sensitivity to various anticancer agents and inhibition of glycolysis may be a promising therapeutic strategy 
In addition to mechanisms that were unique to each molecular tumor type, functional analysis indicates that for both ER positive and negative cell lines, the efficacy of anticancer treatment was related to cell cycle and cell death (Table S3
). Many chemotherapeutic agents cause DNA damage or interfere with the ability of cells to replicate DNA correctly. Cells that cannot replicate DNA will often die by apoptosis. As such, regardless of distinct mechanisms of action, the efficacy of anticancer treatments depends on cell cycle. This observation that the efficacy of anticancer treatment was related to cell cycle and cell death, especially in ER positive cells is consistent with other studies. Previous studies also show that although several genomic signatures which have predictive value of drug response demonstrate limited overlap among them, they all include genes that are related to cell proliferation 
. The fact that cell cycle and cell death related genes have been identified from this cell based study, which is consistent with results from patient studies, support the feasibility of using cell line model to study drug response.
In this analysis, cell lines were used to identify genes related to drug response. Compared to patient-based studies, cell lines afford experimental advantages of controlling experimental variables, as well as measuring the effect of multiple drugs simultaneously, which cannot be done in patient studies. However, cell lines are not identical to cells from patient samples, and the use of cell lines ignores the influence of the tumor microenvironment on drug response. Although breast cancer cell lines mirror many of the biological and genomic properties of in vivo tumors, cell lines also have characteristics that differ. For example, patient-based studies have shown that, based on gene expression profiling, breast cancers can be primarily classified as luminal-A, luminal-B, HER2-enriched, and basal-like, as well as several other subtypes 
. In contrast, for breast cancer cell lines, there is no obvious distinction between luminal-A and luminal-B subtypes, and HER2-enriched cells do not form a separate subtype. Moreover, basal-like cell lines form two clusters, with basal-B generally being more responsive than basal-A 
. Of the 27 cell lines that were used in this study, all 11 ER positive cell lines are classified as luminal (without A or B distinction), and, for the 16 ER negative cell lines, 6 were grouped into basal-A, 6 as basal-B and 4 as luminal. Since breast cells lines of differing intrinsic subtype and ER status display distinct response patterns (data not shown), it would be informative to identify the genes related to multidrug response when further stratifying the cell lines by both intrinsic subtype and ER status in this current study. However, the limited number of cell lines in each subgroup in this study has prevented this type of additional analysis. In the future, as more cell lines become available for research, drug response genes may be identified by stratifying by both intrinsic subtype and ER status. In addition, further investigations may be implemented towards understanding the possible discrepancy between cell lines and patient tumors with respect to intrinsic subtype, as well as elucidating a mechanism for translating cell line-based findings to patient tumors.
Finally, an additional strength of this study was the use of r-th meta-analysis rather than the more commonly used two-step approach 
. The r-th meta-analysis employs a permutation test for statistical inference and controls the false discovery rate. This unified method is more powerful than the simpler methods. Moreover, by controlling r-th, we were able to identify the genes with biological interest. For example, in this analysis, we set r-th at 5 out of 7, which allowed us to identify the genes related to the majority of drugs (5 out of 7). r-th can be set at a maximun to identify genes that are important for all drugs tested or at a minimum to identify those important for only a single drug. Analysis with r-th setting at a maximum or minimum shows similar trend that distinct genes are related to drug response in ER positive vs ER negative cells.
In summary, by taking advantage of the established gene expressions profiles of well-characterized breast cancer cell lines, applying a more powerful analytical method, and examining ER positive and ER negative cell lines separately, we have identified a number of genes related to multidrug response in these cells. Further, we have found that they are predominantly distinct in the 2 cell types and related to distinct cellular processes. These findings provide a basis for further research into the biological mechanisms of drug resistance. Such information may ultimately lead to the identification of biomarkers for potential therapeutic options.