1. Introduction

Due to changes in diet, lifestyle, and a gradually aging population, cancer became more common throughout the world [1]Aging, alcoholism, hormone therapy, obesity, and genetic inheritance constituted the top risk factors for cancers [2]. Breast cancer, especially, is the most frequently diagnosed cancer in women and ranked second among cancer-related causes of death in women. It affected approximately one in eight women in high-income countries by the age of 85 years [3,4]Although many studies proved that breast cancer is treatable, more than 2 million females today globally are diagnosed with breast cancer, with nearly 700,000 death cases [5]. Current treatments for breast cancer included chemotherapy, hormone therapy, immunotherapy, radiotherapy, and surgery. When undergoing these therapies, drugs were often used as adjuvant therapies. For example, modern adjuvant chemotherapy, where drugs were discovered and combined with chemotherapy, was considered effective in treating cases of thousands of patients with hormone unresponsiveness or other complications [6]. Therefore, to explore the maximal feasibility of adjuvant therapies with drugs, drug repurposing could be used as an effective approach.

Drug repurposing is a strategy that reused approved or well-established clinical drugs for other diseases. It had been widely and successfully used in drug discovery [7]. Compared with the traditional drug development process, drug repurposing significantly improved the chance of success and reduced the cost for drug discovery in medical research. For instance, by using the drug repurposing method in the treatment of colorectal cancer, compound 19 inhibited STAT3 binding toward the hTERT promoter, indirectly inhibiting telomerase activity, and presented a specific arrest in the colorectal cancer cell cycle [8,9]. The drug repurposing process comprised four stages: the collection of candidate drugs and targets, feature representation of molecules, computational modeling and docking for the molecules, and real-world application [10].

Generating suitable drug repurposing candidates and targets was a pivotal step among the drug repurposing pipelines. As drug repurposing provided the foundational method of repositioning the drugs, recent advances in genomics, transcriptomics, proteomics, and metabolomics provided vast and deep knowledge about the molecular and metabolic alterations that occurred in cancers [11]. Specifically, single-cell RNA sequencing (scRNA-seq) technology allowed massively parallel characterization of thousands of cells at the transcriptome level and detected the heterogeneity of cell types in the tumor microenvironment [12,13]. Recently, scRNA-seq-induced drug repurposing improved accuracy and fidelity by using differentially expressed genes across different cell types [12]. For instance, researchers successfully utilized single-cell RNA sequencing to identify 19 different cell types in Bladder carcinoma (BC) environment and later highlighted the role of inflammatory cancer-associated fibroblasts (iCAFs) in tumor progression in BC, additionally providing essential data for future drug development [14]. For breast cancer, scRNA-seq was employed to separate and sequence breast cancer cells across three subtypes: Luminal (ER+, PR+/-), HER2+, and triple-negative (TNBC; ER-, PR-, HER2-). Researchers analyzed the data and, using scRNA-seq signatures, identified nine distinct "ecotypes" or clusters for breast cancer cells, each presenting unique cellular architectures [15]. Hence, by further applying the scRNA-seq technique, it was possible to predict the clinical outcomes of each ecotype cells based on its sequencing and structure, and to explore interactions with repurposed drugs. Thus, scRNA-seq permitted the detection of novel therapeutic targets, promoting drug repurposing studies for breast cancer [16].

This study conducts drug repurposing discovery for breast cancer based on scRNA-seq data. By analyzing the differentially expressed genes in different cell types of breast cancer, the study maps the genes to potential protein targets, followed by drug-target interaction analysis. This study provides a computational framework for the future design of combination treatment strategies for breast cancer therapies.

2. Methods and materials

2.1. Dataset

The scRNA-seq dataset for breast cancer was extracted from GEO database with accession ID: GSE158399. The original study was to obtain information about the lymph node metastasis of breast cancer cells, they selected the matched primary breast cancer (PC), positive lymph nodes (PL), and negative lymph nodes (NL) of the same patient to perform integrated analysis. The matched primary breast cancer (PC), positive lymph nodes (PL), and negative lymph nodes (NL) samples were analyzed with single-cell RNA sequencing.

2.2. KEGG

KEGG (Kyoto Encyclopedia of Genes and Genomes) is a knowledge database resource for understanding high-level functions and utilities of the biological system from molecular-level information, especially large-scale molecular datasets generated by genome sequencing and other high-throughput experimental technologies. The dataset is known to computerize data and knowledge on protein interaction networks (PATHWAY database) and to reconstruct protein interaction networks for all organisms whose genomes are completely sequenced (GENEsS and SSDB databases) [17]. This study merged KEGG dataset with the separated breast cancer single cell sets for further scRNA seq DEG analysis.

2.3. DrugBank

In this study, we used DrugBank datasets to obtain cancer-related drug candidates. DrugBank is a comprehensive, free-to-access, online database containing information on drugs and drug targets which combined detailed drug (chemical, pharmacological and pharmaceutical) data with comprehensive drug target (sequence, structure, and pathway) information [18]. repoDB, which is a database further developed for drug repurposing use, contains approved and failed drugs and their indications. It is the combination of datasets from both AACT and DrugCentral [19,20].

2.4. scRNA seq differentially expressed gene analysis

Single cell RNA sequenced differentially expressed gene analysis refers to the computation of the RNA molecules within each cell of a given sample and further analysis of that base on different gene expression [21]. Contrasting to the conventional bulk-RNA seq technique, scRNA-seq provides gene measurements for a genome wide range of individual cells and labels them based on the sorted information [22]. Thus, differentially expressed gene analysis under multiple conditions can be done with separated cell types, which allows direct comparison and biological interpretation for needs. One of the developed tools we used in this study is through pseudobulkDGE in scran [23]. It is a wrapper function around edgeR’s quasi-likelihood methods to conveniently perform differential expression analyses on pseudo-bulk profiles, allowing detection of cell type-specific changes between conditions in replicated studies.

2.5. Drug candidate generation

The generation and selection of drug candidates were crucial steps in this study. We needed to filter and identify cancer-related drugs that could effectively bind with protein targets for DTI docking. We utilized the Drug2cell tool, which identifies specific cellular targets of bioactive molecules with drug properties based on single-cell RNA-seq data. Drug2cell allows filtering of drugs and target molecules based on quantitative bioactivity metrics, drug categories (ATC classification), clinical trial phases, and classes of molecular targets. This approach reveals hidden mechanisms of action and predicts the impact of medicines on specific cell types [24].

2.6. Molecular featurization

We applied the molecular featurization on the drug candidates and proteins in order to the computational software to read and analyze the dataset. It plays a critical role in transforming the chemical structures of the drugs and gene sequences of proteins targets to feature vectors or numbers that can be passed down to learning algorithms since we required computational support to analyze the extensive dataset [25]. The developed tool we used in this project is RDKit, which is an open-source cheminformatics and machine learning machine [26]. The RDKit supports several different models and allows us to define our own by providing a function that assigns aromaticity.

2.7. Molecular docking simulation

We used the molecular docking methodology to explore the behavior of small molecules (drugs) in the binding site of a target protein. Because it is crucial in predicting the orientation of the ligand/drug when it is bound to a protein receptor using shape and electrostatic interactions to quantify it in modern drug discovery [27]. The docking process involves two steps: prediction of the ligand conformation, position, and orientation within these sites (usually referred to as pose) and assessment of the binding affinity. Autodock vina, a docking engine we used in the project, only requires the structures of the molecules being docked and the specification of the search space including the binding site [28].

3. Results

Figure 1. Framework of drug repurposing by single cell RNA-seq for breast cancer. Panel A shows the protein targets preparation for breast cancer cells through scRNA-seq analysis following by the mapping of DEGs into correlated protein targets through Uniprot database. Panel B shows the molecular featurization of cancer related drug candidates from drug bank based on its chemical structures and prior clinical trial results. Panel C shows drug-target interaction prediction by molecular docking (Autodock Vina), which first predict the protein-ligand conformation, and secondly assess the binding affinity. Panel D shows the DTI ranking followed by the integration of all the computational pipelines and dataset to establish an efficient drug repurposing protocol for breast cancer.

3.1. Framework of drug repurposing by single cell RNA-seq for breast cancer

Figure 1 shows the framework of drug repurposing for breast cancer. First, we collected and clustered various types of breast cancer cells based on their genomic characteristics. We then performed scRNA-seq statistical analysis to generate a list of differentially expressed genes (DEGs) for each cell type. These DEGs were analyzed through univariate and multivariate methods and mapped to corresponding protein targets using the UniProt database. Next, we filtered and selected candidate cancer-related drugs after identifying the DEGs and their associated protein targets. Using molecular docking simulations with AutoDock Vina, we examined the behavior of small molecules (drugs) at the binding sites of target proteins. The best-fit candidate drugs for each cell-type protein target were ranked based on binding scores, resulting in a list of the top repurposed drugs. Finally, we integrated all computational pipelines and datasets to establish an efficient drug repurposing web server for breast cancer.

3.2. Single-cell RNA sequencing data analysis

After processing the breast cancer scRNA-seq data, we annotated ten cell types: Pericytes, Myeloid cells, CD8 Effector cells, B cells, Myofibroblasts, Naive T cells, CXCL14 cancer cells, Plasma cells, Macrophages, and Cancer stem cells (Figure 2). Among these, Pericytes and Myeloid cells had the most drug candidates applied to them. Differentially expressed genes (DEGs) were identified through pseudobulk DEG analysis in Scran, considering cell fractions and mean expression levels within groups. Key DEGs included CACNA1I, MS4A1, COL5A2, COL6A3, COL3A1, IL12B, PDE7B, COL4A1, and IL5RA. By using these top ten DEGs, we identified over a hundred protein targets. For example, Q01344 and P11836 are cell membrane proteins that exhibit specific linkage with HER2+ breast cancer subtype,[29]. and Q9P0X4 is associated with the TNBC subtype.

Figure 2. Leiden clustering and cell type annotation of scRNA-seq data in three subtypes of breast cancer. Panel A shows the Leiden clustering of scRNA-seq data. Panel B shows the cell type annotation of scRNA-seq data in breast cancer.

Figure 3. Differentially expressed gene analysis in scRNA-seq data and examples of drug candidates. Panel A shows the differentially expressed gene analysis in scRNA-seq data. Panel B shows the examples of drug candidates across different cell types.

Figure 4. The number of drug candidates applied on each target. Panel A shows the numbers of drug candidates found per cell type. Panel B shows the numbers of drug candidates found per gene. This showcases the cell types and genes that are most suitable for breast cancer and repurposed drugs targeting.

3.3. Drug repurposing by drug2cell and molecular docking simulations

Based on the DEGs, drug candidates were selected using the Drug2cell (Figure 3). These candidates include Doxorubicin, Cyclophosphamide, Everolimus, Tamoxifen, Anastrozole, Paclitaxel, Aspirin, Trastuzumab deruxtecan, Capivasertib, Gemcitabine, Methotrexate, Itraconazole, Simvastatin, and Metformin (Table 1). For all candidates, we also show the number of drug candidates applied on each target in Figure 4. The proteins and drugs were processed through RDKit and AutoDock Vina for sorting and scoring their drug-target interactions (Figure 5). Among the candidates, fourteen drugs were particularly notable, including Doxorubicin, Cyclophosphamide, and Everolimus for the HER2+ cancer subtype, and Paclitaxel and Gemcitabine for TNBC subtypes. These drugs were repurposed with clinical evidence supporting their effectiveness in breast cancer treatment.

Figure 5. Flow chart showing the processes of drug-target interaction (DTI).

Table 1. The top candidates for feasible repurposed drugs with the related chemical structures, biomarkers, responsible breast cancer types for treatments, and reported clinical outcomes.

4. Conclusions and future directions

Although the fourteen drugs exhibited high efficiency in treating breast cancer, some clinical trials revealed some adverse effects on other body tissues, such as over-toxicity and apoptosis of healthy cells [30]. Despite these challenges, the drug-repurposing strategy combined with scRNA-seq data has the potential to enhance therapeutic target discovery and improve treatment success rates for breast cancer.