Advances and challenges in drug repurposing in precision therapeutics of colorectal cancer

Abstract

Colorectal cancer (CRC) ranks as the third most common cancer globally and the second leading cause of cancer-related deaths, representing a significant health burden. Despite advancements in traditional treatments such as surgery, chemotherapy, targeted therapy, and immunotherapy, these approaches still face challenges, including high costs, limited efficacy, and drug resistance. Drug repurposing has emerged as a promising strategy for CRC treatment, offering advantages with reduced development timelines, lower costs, and improved drug accessibility. This review explores drug repurposing strategies for CRC, supported by multidisciplinary technologies, and discusses the current challenges in the field.

Key Words: Drug repurposing; Colorectal cancer; Treatment; Multidisciplinary collaboration; Precision therapeutics

Core Tip: Drug repurposing emerges as an essential strategy due to its advantages like shorter development timelines and lower costs. This review covers recent progress in repurposing traditional drugs and using high-throughput screening and artificial intelligence for colorectal cancers.

INTRODUCTION

Colorectal cancer (CRC) is one of the most common malignant tumors worldwide. It ranks third in global cancer incidence and second in cancer-related mortality. According to the 2022 global cancer statistics, over 1.9 million new CRC cases were reported, with more than 900000 CRC-related deaths, accounting for more than 10% of all cancer cases and deaths[1]. The incidence of CRC exhibits significant geographic disparities, with higher incidence and mortality rates observed in developed regions, such as North America, Europe, and Oceania[1]. However, in recent years, the incidence and mortality of CRC have been rapidly increasing in developing regions, such as Asia and Latin America, contributing to an escalating disease burden[2,3].

Tumor metastasis is the leading cause of patient mortality in CRC. Early-stage CRC is often asymptomatic, leading to delayed diagnosis. As a result, more than 30% of patients are diagnosed at an advanced or metastatic stage, significantly reducing treatment options and prognosis[4,5]. Once metastasis occurs, surgical resection becomes less effective, and chemotherapy and radiotherapy cannot control the disease. Metastatic CRC (mCRC) remains one of the most challenging cancers to manage and treat, highlighting the urgent need for novel therapeutic strategies and targeted interventions[6]. Despite recent advancements in surgery, chemotherapy, targeted therapy, and immunotherapy, mCRC still lacks effective treatment options, particularly targeted drugs for CRC metastasis[7-9]. Due to the high heterogeneity and resistance of mCRC, existing treatment strategies often fail to provide long-lasting efficacy.

Drug repurposing refers to the use of already approved or clinically investigated drugs for new indications. This includes repositioning Food and Drug Administration (FDA)-approved off-patent drugs, previously unsuccessful drugs, and patented drugs for novel clinical applications. By leveraging existing pharmacokinetic and safety data, drug repurposing offers a cost-effective and time-efficient strategy for identifying new therapeutic options, particularly for diseases with limited treatment choices[10]. Compared to patented drugs, off-patent and widely used generic drugs are easier to repurpose due to their extensive safety and efficacy data, lower cost, and reduced development risks. For example, aspirin was originally used for pain and fever relief, but it was later found to prevent cardiovascular diseases at low doses[11]. Nodial was initially used as an antihypertensive drug and later repurposed for the treatment of hair loss[12]. Failed drugs, such as thalidomide, which was withdrawn due to teratogenicity, have been repurposed for treating multiple myeloma and leprosy[13,14].

Significant progress has been made in drug repurposing research for the treatment of CRC. While the repurposing of traditional drugs is one strategy, others have focused on high-throughput screening and the reapplication of potential drugs using artificial intelligence (AI) methods. In the case of traditional anti-cancer drugs, aspirin and metformin have potential anti-CRC effects, likely by targeting the phosphatidylinositol 3-kinase (PI3K) signaling pathway, while the antihistamine desloratadine directly targets ARF1, inhibiting IQGAP1-ERK-Drp1-mediated mitochondrial fission, thereby suppressing colon tumorigenesis[15]. Despite these advancements, the widespread application of drug repurposing in CRC treatment still faces challenges, including the lack of large-scale clinical trial data, especially regarding effects in specific CRC subtypes or advanced cases, and the need for further research to elucidate their molecular mechanisms.

OVERVIEW OF DRUG REPURPOSING

Definition and principle

Drug repurposing, also known as drug repositioning and drug rediscovery, is a strategy that has recently gained significant attention. Its core definition involves the application of approved or investigational drugs to therapeutic areas distinct from their originally approved indications[16,17]. Drug repurposing is based on the following scientific foundations: (1) The same therapeutic targets and molecular characteristics; and (2) Drug pleiotropy, meaning that a drug may have effects beyond its originally intended purpose[18]. In addition, the multi-target properties of drugs or combination therapy strategies play a crucial role in treating complex diseases, including cancer and viral infections[19]. Studying the mechanisms of drug action is also closely related to drug repurposing research. A deeper understanding of the signaling pathways involved in drug action can provide theoretical support for their application in new indications[20,21]. At the same time, the established safety and pharmacokinetic profiles of repurposed drugs hold significant value in accelerating the development of combination therapy strategies[22].

In terms of implementation, drug repurposing strategies can be categorized into experimental approaches and computational approaches[23,24]. Drug repurposing research typically integrates both experimental and computational approaches to enhance efficiency and accuracy. Experimental approaches include in vitro high-throughput screening (e.g., drug-binding assays) and in vivo phenotypic screening, while computational approaches rely on the analysis of existing data, such as structural information, genomic data, multi-omics datasets, and real-world records[23,24]. With technological advancements, computational approaches have increasingly become essential tools in drug repurposing. Computational strategies, such as signature matching, structure- and ligand-based virtual screening, genome-wide association studies, and pathway mapping, have been widely applied in drug repurposing research[25,26]. Additionally, the application of knowledge graphs has introduced new perspectives for drug repurposing. By integrating diverse biomedical data, knowledge graphs enable more effective identification of potential drug applications[20,27,28].

Multidisciplinary support for drug repurposing

As an innovative drug development strategy, drug repurposing relies on the synergistic integration of multiple disciplines. This interdisciplinary approach enhances the efficiency and precision of drug repurposing and expedites the identification and clinical application of novel therapeutic indications (Figure 1).

Figure 1  Multidisciplinary approach to drug repurposing, integrating real-world data, molecular biology, structural chemistry, computer science, and information biology.

Molecular biology

Binding assays play a pivotal role in drug development, encompassing a range of methodologies such as affinity quantification, dose-response analyses, and biochemical assays. Notably, multi-dose affinity assays and target engagement assays serve as key approaches, offering quantitative insights into compound-target interactions through multi-dose experimental designs. However, systematically evaluating the interactions between all approved drugs and the entire proteome remains a significant experimental challenge due to the high costs and time-intensive nature of such studies. Despite these limitations, binding assays provide critical data in the identification and characterization of drug-target interactions[29,30]. Dose-response biochemical assays represent another essential form of binding assays, widely employed in high-throughput screening to systematically validate compound-target interactions. By measuring compound activities across a range of concentrations, these assays enable a more precise assessment of binding affinity and potency.

Given the substantial variability in activity windows across different compounds and target classes, dose-response assays are recommended to enhance the reliability and reproducibility of experimental data[30]. Biochemical assays are often the first step in validating the ability of specific compounds to modulate a predicted target. These assays not only facilitate the assessment of compound selectivity and specificity but also provide a foundational framework for subsequent experimental validation, thereby enhancing the robustness of target engagement studies[30,31]. In mycobacterium tuberculosis research, binding assays have confirmed the inhibitory effects of two compounds on dihydrofolate reductase (DHFR), a clinically validated target in drug development. This approach underscores the significance of integrating biochemical and phenotypic assays to enhance the reliability of target identification and therapeutic potential[30]. By offering crucial insights into drug design and development, they serve as indispensable methodologies for drug repurposing efforts.

Phenotypic screening relies on cellular or in vivo models to directly observe functional or phenotypic changes induced by compounds, and it is widely used to identify candidate drugs eliciting desired biological effects. In the search for novel anti-tuberculosis agents, Mugumbate et al[32] integrated large-scale screening data of Mycobacterium tuberculosis with phenotypic effects to develop an ensemble model comprising similarity ensemble approach, naive bayes classifiers, and docking algorithms. Ultimately, this strategy identified two potent inhibitors targeting the Mycobacterium tuberculosis-specific gene DHFR, while their therapeutic potential was further validated using Mycobacterium bovis drug-resistant strains[30,32]. The primary advantage of phenotypic screening lies in its ability to identify bioactive compounds without prior knowledge of their molecular targets. Instead, this approach focuses on compounds that induce desirable phenotypic changes, with subsequent biochemical and genetic analyses employed to elucidate the underlying mechanisms. Cellular phenotypic assays enable direct observation of the selective cytotoxicity against cancer cells or virus-infected cells, thereby assessing its ability to achieve the desired therapeutic phenotype within a disease-relevant cellular context. Compared with conventional target-based screening, phenotypic screening has the distinct advantage of identifying compounds with multi-target activities, including those with unexpected pharmacological effects, thereby offering new opportunities for drug repurposing. By quantifying phenotypic changes, researchers can further leverage machine learning or deep learning algorithms to analyze compound effect patterns, optimizing screening workflows and expanding the potential for drug discovery[22,30]. However, a key limitation of this approach lies in the high cost and time-consuming nature of screening assays[23,24]. Overall, experimental phenotypic screening in drug repurposing offers a data-driven, experimentally validated approach, integrating high-throughput screening with computational techniques. This strategy not only enhances the efficiency of candidate drug identification but also provides new insights into the underlying mechanisms of complex diseases[33].

Structural chemistry

Quantitative structure-activity relationship (QSAR) analysis examines the relationship between chemical descriptors, such as molecular shape, charge distribution, and hydrophobicity, and biological activity. This approach provides a scientific basis for drug optimization and the development of new therapeutic indications. The goal of QSAR is to predict the binding affinity of drug molecules to targets or their biological effects, thereby enabling the screening of potential drug candidates. Recently, the introduction of machine learning techniques has significantly improved the prediction efficiency and accuracy of QSAR models, aiding classification tasks (such as distinguishing inhibitors from non-inhibitors), as well as predicting continuous variables (such as binding affinity or IC₅₀ values), greatly expanding the applicability of drug screening. The construction of QSAR models typically relies on extensive bioassay data sourced from online databases such as BindingDB, PubChem, and ChEMBL[34]. Deshmukh et al[35] developed a QSAR classification model for DNA repair enzyme FEN1 using support vector machine and random forest algorithms, based on bioassay data from PubChem. From over 50000 compounds, they identified five candidate molecules, one of which (JFD00950) exhibited significant cytotoxicity[34,35]. An extended form of QSAR, protein chemometrics (PCM), enhances predictive capabilities by integrating the features of both drug molecules and target proteins, such as amino acid sequences and chemical descriptors. With PCM, Van Westen et al[36] identified six high-affinity adenosine receptor ligands from over 100000 compounds.

Virtual screening combined with QSAR techniques has played a significant role in drug repurposing. Naive Bayes and recursive partitioning models were used to screen for indoleamine 2,3-dioxygenase 1 inhibitors, leading to the identification of three novel compounds with submicromolar concentrations. These inhibitors feature molecular scaffolds derived from the active components of the traditional Chinese medicine Danshen[34,37]. Molecular docking-based virtual screening of over 2.65 million compounds led to the identification of a novel adenosine A2A receptor inhibitor with nanomolar binding affinity[34,38]. Advances in QSAR models and computational methods have continuously improved the efficiency and accuracy of drug screening. These approaches not only facilitate the identification of potential drug-target interactions but also enable the discovery of novel drug candidates through large-scale compound library screening.

In addition to the chemical methods mentioned, computational tools for identifying new drug targets through chemical modification have advanced rapidly in recent years. Novel approaches based on molecular similarity prediction, using techniques like circular fingerprints and graph neural networks, enable detailed exploration of both chemical and target characteristics[19,34]. These tools have not only helped elucidate the fundamental biological mechanisms of certain compounds but also supported the validation of their potential indications across various diseases. For example, the Virtual Kinome Profiler, in combination with support vector machine algorithms, has improved the accuracy of predicting kinase-compound interactions. It has also validated experimental findings across multiple targets, significantly contributing to the repurposing of kinase-specific drugs[19,30].

On the other hand, characterizing the chemical properties of drugs allows researchers to gain deeper insights into the relationship between molecular structure and target activity, forming a critical foundation for developing new indications. Data-driven computational techniques employ chemical fingerprints (descriptors) or molecular representations, combined with target activity data, to efficiently predict structure-activity relationships using machine learning. This approach not only reveals potential new targets or indications for existing approved drugs but also significantly accelerates the drug repurposing process[19,39]. By integrating the molecular chemical structure of drugs with computational methods from multi-omics technologies, a comprehensive analysis of their mechanisms of action (MoA) can be achieved. For example, deep learning models that integrate multilayer network data, such as gene regulatory data, protein-protein interactions, and chemical similarity, can efficiently identify key biological pathways affected by a drug. This approach facilitates the discovery of potential new indications and novel targets[20,27]. Compared to traditional experimental methods, these computational approaches not only significantly reduce time and cost but also provide a theoretical basis for predicting new therapeutic indications.

In the field of structural chemistry, advanced computational chemistry methods, combined with machine learning and deep learning algorithms, have significantly improved the efficiency of drug repurposing. Various machine learning models have deconstructed the interaction mechanisms between drugs and their targets based on molecular structures and target characteristics, providing systematic tools for predicting target activity. For example, Li et al[40] validated compound activity classifications against G protein-coupled receptors, enzymes, and nuclear receptors by integrating chemical substructure fingerprints with a rotation forest classifier. Additionally, the DGraphDTA model, based on graph neural networks, leverages deep molecular and protein representations to achieve more accurate predictions of target activity[19,41]. Emerging integrative models, such as iDrug, overcome the limitations of traditional methods by embedding the relationships among drugs, targets, and diseases into a unified framework. This facilitates efficient cross-network knowledge transfer, significantly enhancing the drug repurposing process[19,42]. Experimental validation of several candidate molecules further supports the model’s capacity to significantly accelerate the development of kinase-targeted drugs[43].

X-ray crystallography has been instrumental in providing high-resolution three-dimensional structural data, forming the foundation for structure-based drug design. Access to detailed target structures is crucial for virtual screening, molecular docking, and molecular dynamics (MD) simulations. In the absence of experimentally determined structures, computational models such as homology modeling can be employed to predict protein conformations. However, these alternative methods often yield lower accuracy compared to X-ray crystallography, potentially compromising the reliability of downstream drug discovery efforts[25,44,45]. Certain proteins, particularly membrane proteins and receptor molecules, pose significant challenges for crystallization, leading to a scarcity of X-ray crystallography data. In such cases, computational models may struggle to accurately reconstruct the complex dynamic behaviors and conformational changes of these proteins, thereby limiting the precision of structure-based drug design approaches[46]. Furthermore, X-ray crystallographic structures, being inherently static, do not capture dynamic biological processes such as protein conformational changes, ligand binding/unbinding pathways, or thermodynamic properties. These dynamic aspects necessitate complementary investigations using MD simulations, which provide crucial insights into the flexibility and functional mechanisms of biomolecular systems[47].

Virtual screening is a computational approach designed to rapidly identify potential drug candidates from chemical libraries. It primarily encompasses two strategies, structure-based virtual screening (SBVS) and ligand-based virtual screening. SBVS relies on the three-dimensional structural data of molecular targets and employs molecular docking to evaluate ligand binding affinities. In contrast, ligand-based virtual screening leverages topological features and similarity of known ligands to identify chemical compounds with analogous structures[28,41]. In drug repurposing, SBVS is typically applied to high-throughput screening of approved drugs or previously developed compound libraries, such as DrugBank and ZINC database, to identify novel therapeutic applications[33]. In cancer-related projects, virtual screening has been employed to identify kinase inhibitors, such as those targeting c-Src and BCR-ABL. Identified candidate compounds are further refined through molecular docking and MD simulations to optimize their binding affinity and stability[48]. However, virtual screening results often contain a high number of false positives, necessitating further refinement and validation. Integrating MD simulations can provide “target dynamic insights”, enabling more precise screening and validation of potential drug candidates[49,50]. Moreover, the influence of water molecules on binding stability is often underestimated, highlighting the need for dedicated studies to accurately account for their role in molecular interactions and drug-target affinity[51,52].

Molecular docking is widely employed in preliminary screening of drug candidates, serving as a computational approach to simulate interactions between ligands (small-molecule drugs) and target molecules (such as proteins). This method predicts the optimal binding conformation of a ligand and estimates its binding affinity. When integrated with virtual screening techniques, molecular docking enables the rapid identification of potential drug candidates from large chemical databases, including inhibitors for targets such as BRAF mutations and c-Src[33]. Through drug docking studies, it is possible to systematically explore drug-target interactions within the chemical space, providing reliable candidates for experimental validation. Key aspects of molecular docking include scoring functions, identification of binding regions, and subsequent validation. Docking scoring functions typically estimate binding free energy by considering van der Waals forces, electrostatic interactions, and solvation effects. However, these functions may lead to false positives or false negatives due to potential underestimation of solvation effects or failure to account for the complexity of macromolecular binding[48,49]. To reduce false positives in docking results, a consensus scoring strategy is commonly employed. This approach enhances reliability by combining scores from multiple scoring functions, thus providing a more robust prediction of binding affinity and improving the overall accuracy of the docking process[49,50,53]. Next, it is crucial to identify binding sites within the three-dimensional structure of the protein, such as the adenosine triphosphate binding site or allosteric sites. MD simulations can help identify potential binding regions that may not be evident from static docking analysis, like the allosteric site of c-Src kinase, thus complementing docking results[54]. After docking, it is essential to validate the accuracy and stability of the predicted docking results through MD simulations[55].

MD simulations can capture the transition of proteins from one conformational state to another, as well as the changes in interactions between key residues. Through conventional MD and unbiased MD (UMD) simulations, researchers can analyze the differences between active and inactive conformations and identify specific dynamic mechanisms associated with ligand binding. These conformational differences can be exploited to design highly selective, single-target inhibitors. The reconstruction of binding and unbinding mechanisms is crucial for studying the entire process of ligand entry into (binding) or departure from (unbinding) the protein’s binding pocket. Through UMD simulations, researchers can reconstruct the pathways and dynamics of protein-ligand interactions, unveiling intermediate states and binding dynamics/affinity[56-58]. Shan et al[55] used UMD simulations to elucidate the detailed mechanisms of binding and unbinding of dasatinib with c-Src kinase. They analyzed the relationship between dasatinib’s binding affinity and its conformational stability, providing deeper insights into the dynamic interactions between the ligand and target protein. Water molecules can mediate ligand binding by forming hydrogen-bond networks that either stabilize interactions within the binding pocket or, conversely, promote ligand dissociation by disrupting key contacts and altering the local solvation environment[25]. MD simulations further support target conformational analysis, binding pocket identification, and validation of screening results. After initial screening methods such as virtual screening and molecular docking, MD is often used to evaluate the stability of top-ranking compounds within protein-ligand complexes and to verify the reliability of predicted residue-ligand interactions[47,59,60].

Computer science

AI has recently emerged as a powerful tool in drug repurposing research, owing to its advanced data processing and analytical capabilities. Techniques such as deep learning, graph neural networks, and reinforcement learning have shown considerable promise in extracting and interpreting large-scale, multidimensional biological data, thereby accelerating the discovery of novel therapeutic applications for existing drugs[20,55,61]. Among data-driven AI models, deep learning has been widely used to predict drug-disease associations and elucidate MoA. Since the development of the Library of Integrated Network-Based Cellular Signatures and the Connectivity Map, large-scale drug-induced gene expression profiles have become publicly available, providing crucial resources for AI-driven drug repurposing. Leveraging these datasets, deep learning approaches such as DeepDR, which leverage deep neural networks to integrate drug-disease and drug-target networks, have demonstrated high accuracy in predicting potential therapeutic indications[20,34,62]. Furthermore, computational pipelines such as PREDICT, which integrate drug and disease similarity networks, have successfully identified novel therapeutic indications[34]. Although Chat Generative Pre-Trained Transformer (GPT) is not specifically trained for drug design, as a large language model (LLM), it can provide multifaceted support for drug repurposing in CRC through its capabilities in natural language understanding, knowledge integration and generation, and data-driven analytical assistance[63]. By leveraging natural language processing, GPT-4 can synthesize vast amounts of literature in cancer biology, pharmacology, and clinical research, rapidly identifying potential associations between drugs, signaling pathways, phenotypes, and gene mutations. Furthermore, with the aid of structured databases, it can assist bioinformatics teams in constructing drug-gene-phenotype networks[64], which facilitate the prioritization of candidate therapeutics (Table 1).

Table 1 Summary of databases related to drug repurposing.

Database	Description	Applications
ChEMBL	Bioactivity database	Contains bioactivity data of small molecule compounds, associated drug targets, biological data, etc., for drug development
Drug target commons	Drug target database	Provides drug and target interaction data, including target selectivity, mechanism of action and experimental data, for drug target selection and development
Binding database	Binding affinity database	Includes binding affinity data between small molecule compounds and proteins, such as inhibitory concentration (IC50), binding constant (Kd), for drug screening
Protein data bank	Structured database	Stores three-dimensional structural information of biological macromolecules (e.g. proteins, nucleic acids, and complexes) for use in molecular modeling and structural analysis
Sider	Drug side effects database	Provides information on drug related side effects, showing the adverse reactions and mechanisms of action of known compounds, for safety assessment
OMIM	Genetic diseases and mutations database	Contains information related to human genes and genetic diseases, including disease phenotypes, gene mutations, and genetic relationships
DrugBank	Comprehensive drug database	Integrates information on pharmaceutical chemistry, biological targets, metabolic pathways, and mechanisms of action, it is an important tool for pharmacology and medicinal chemistry
BioVU DNA Databank	Biological sample database	Contains information about a patient’s DNA, linked to clinical data, for genetics and precision medicine research
United Kingdom Biobank	Cohort and biobank	A national-level repository of genomic data, health records, and lifestyle data for epidemiological and disease association studies
Drug Repurposing Hub	Drug repurposing database	Provides data on drugs that have been approved or failed, as well as information on their potential in new indications
Repurpose database	Drug indications repurpose database	Covers data on drugs or indications that have been reported and focuses on revealing new uses of drugs through data mining
Pathway commons	Path integration database	Collects and integrates relational networks of multiple biological pathways, such as metabolic, signaling, and gene regulatory networks, for systems biology and network analysis
KEGG pathways	Pathway database	Provides data on the biological pathways of genomes, metabolism, diseases and drugs, and is an important reference resource for biology and drug development
Pharmaco database	Drug sensitivity database	Contains sensitivity data on the response of cancer cell lines to drugs for anticancer drug screening and personalized treatment research
Cell model passports	Cell model database	Provides cell model information for cancer and other disease research, including genomic data, omics data, and cell origin information
Xeva	Drug response modeling tools	Tools for modeling xenograft data, such as PDX models, to provide drug efficacy predictions and inter-model comparisons
FAERS	Adverse drug event database	An adverse drug event reporting system administered by the FDA, containing adverse event reports and drug safety monitoring information
Sentinel system	Drug safety database	A signal monitoring system developed by the United States FDA to monitor the safety of drugs, vaccines and medical devices on the market
OncoPDSS	Decision support system for personalized oncology chemotherapy	It is used to assist in the selection of appropriate chemotherapy drugs and treatment regimens based on patient genomic information to enhance precise tumor therapy

FDA: Food and Drug Administration; KEGG: Kyoto Encyclopedia of Genes and Genomes.

Meanwhile, transcriptomic data have been leveraged to develop mechanistic drug analysis tools such as GPAR. By integrating deep learning with pathway enrichment analysis, GPAR overcomes the limitations of traditional gene set enrichment analysis and significantly enhances predictive accuracy[65,66]. In molecular generation, generative adversarial networks and reinforcement learning have been widely adopted[34]. Moreover, owing to their superior ability to model chemical structures, graph convolutional networks (GCNs) have been extensively employed in molecular generation. A key advantage of GCNs is their ability to directly process molecular graphs without requiring conversion into two-dimensional representations[22]. With the widespread adoption of electronic health records (EHR) and multi-omics data, AI has significantly advanced data integration technologies. This has facilitated the rapid identification of potential drug repurposing candidates based on patients’ medication histories[33,67]. Furthermore, graph algorithms have increasingly become a crucial analytical tool, offering novel data mining approaches for drug repurposing by integrating multi-layered relationship networks of drugs, diseases, and targets[34]. However, the performance of these models is highly dependent on high-quality and consistent datasets. Many data sources exhibit unavoidable experimental variability, necessitating adherence to principles (findable, accessible, interoperable, reusable) to enhance data usability and reproducibility[68-71]. Moreover, the “black-box” nature of AI models presents a major challenge in interpreting predictive outcomes, particularly in elucidating drug MoA and complex disease pathophysiology. DrugCell employs a visible neural network to map predictive nodes to relevant gene data, enhancing the transparency of drug response predictions. However, challenges such as high computational costs remain a significant hurdle[20]. Additionally, the limited clinical validation of AI-generated predictions remains a significant barrier to their real-world application[34].

LLMs have demonstrated substantial potential in drug repurposing and discovery, emerging as an efficient and cost-effective solution for accelerating pharmaceutical research and development[30,72,73]. BioMed-RoBERTa, a pre-trained model trained on 2.68 million scientific articles and 7.55 billion tokens, has demonstrated superior performance in biomedical natural language processing tasks, surpassing many general-purpose language models[74]. Similarly, LLMs such as BERT, SciBERT, BioBERT, and BlueBERT have been widely applied to identify drug-target interactions, predict protein-peptide interactions, and facilitate the translation between drug molecules and therapeutic indications. These models have significantly enhanced the efficiency and accuracy of biomedical text mining, paving the way for data-driven drug discovery and repurposing[75-77]. Moreover, GPT-based architectures have been employed in a range of tasks, including de novo drug design and biological query resolution, further expanding the applications of LLMs in drug discovery[78,79]. Beyond text mining, LLMs have been integrated into deep learning frameworks for predicting drug-target interactions[30,31]. All in all, LLMs, with their powerful capabilities in natural language processing, biomolecular representation, and drug-target prediction, are driving innovation and progress in the field of drug repurposing and discovery. However, despite the impressive performance of these models, the reliability of their predictions still depends on experimental validation to ensure their effectiveness and accuracy in real-world applications[30].

Information biology

The rapid accumulation and integration of multi-omics data have provided a powerful foundation for drug repurposing. The richness of bioinformatics resources, including gene expression profiles, compound structures, protein targets, pathway mapping, and phenotypic data, enables researchers to systematically explore drug MoA and identify potential indications from a multidimensional perspective[20,30,80]. These data not only provide the foundation for drug repurposing but also offer insights into the complex interactions between drugs and biological systems. In drug repurposing, the ability to integrate biological information is particularly crucial. For example, multi-layered analysis of genomics, transcriptomics, proteomics, and epigenomics data can reveal the modes of action of drugs within different biological systems. The complementarity of multi-omics data enables researchers to comprehensively analyze drug targets and their associated pathways[81,82]. Moreover, integrating drug-target interaction data with drug chemical structures and biological information of target proteins can further enhance the precision and applicability of drug screening[19].

However, the quality and accessibility of biological data remain significant challenges[19]. Patient-derived cell model data, such as drug sensitivity data, have not been made fully available, which restricts the depth and breadth of data integration. Nevertheless, the open sharing of public databases, such as DrugBank and The Cancer Genome Atlas, provides valuable resources, making the integration and analysis of multi-omics data possible[55]. The integration of biological data has made notable progress in drug repurposing. For instance, the OpenMoA tool, which combines multi-omics data with network pathway analysis, successfully predicted drug MoA and experimentally validated these predictions[83]. Furthermore, the combination of EHR data with pharmacogenomic data has revealed new drug-disease associations and provided strong support for clinical translation. These biologically driven research strategies are advancing drug repurposing from the laboratory to clinical practice[84,85].

Real-world data

Real-world data (RWD) and its core component, EHR, have increasingly become pivotal in the field of drug repurposing, serving as key drivers for accelerating drug development. RWD encompasses a variety of sources, including EHR, administrative healthcare data, disease registries, and pharmacovigilance systems. The real-world evidence (RWE) generated through retrospective analyses of these data can provide robust support for the study of drug safety and efficacy. Compared to traditional randomized controlled trials, RWD research not only offers faster and more cost-effective solutions but also boasts significant advantages such as large sample sizes, extended follow-up periods, and the inclusion of more representative populations, positioning it as an irreplaceable asset in drug development[33].

With its wealth of longitudinal clinical information and patient pathophysiological data, EHR data has become the cornerstone of RWD. EHRs not only capture a patient’s long-term health status within real-world healthcare settings but also integrate unstructured data, such as clinical notes and online forum posts. By leveraging natural language processing, machine learning, and deep learning techniques, key clinical insights can be extracted, providing a comprehensive view of patient health and treatment outcomes. For instance, Li et al[86] applied deep learning models to EHR notes, significantly improving the accuracy of adverse drug event detection, outperforming traditional systems. Moreover, a unique advantage of EHR lies in its ability to support large-scale parallel testing of drug hypotheses, enabling the rapid generation and validation of drug repurposing results, thereby significantly reducing both time and economic costs[33].

The application of RWD has garnered significant attention from international regulatory agencies. Since 2016, the FDA has issued a series of frameworks and guidance documents, outlining the standards for using RWD in clinical research and drug approval processes. In 2018, the FDA launched the “Real-World Evidence Program” to further promote the integration of RWE into drug approval. Notably, during the coronavirus disease 2019 pandemic, the FDA leveraged RWD-based protocols to assess the safety and efficacy of existing drugs, providing critical support for pandemic response efforts[24]. At the same time, the European Medicines Agency (EMA) has outlined plans for establishing an RWD research framework in its Regulatory Science Strategy to 2025. Through the DARWIN EU project, EMA is coordinating data partnerships to enhance real-world research capabilities, facilitating the integration of RWD into regulatory decision-making and drug development processes[24].

In summary, RWD offers a clinically relevant perspective for drug repurposing while significantly enhancing the efficiency of hypothesis generation and validation, establishing itself as an indispensable resource in drug development. With continuous advancements in AI and machine learning, alongside the refinement of global regulatory frameworks, RWD-driven drug repurposing is poised to unlock greater potential in the future, driving progress in precision medicine and public health.

RECENT ADVANCES IN DRUG REPURPOSING FOR CRC TREATMENT

As research on drug repurposing continues to advance, its critical role in precision oncology has become increasingly evident, particularly in the treatment of CRC, where it holds substantial therapeutic potential[87-89]. As one of the most prevalent and lethal malignancies worldwide, the molecular mechanisms of CRC involve a variety of pathways and oncogenic factors. Given its persistently high mortality rate, the identification of novel therapeutic targets has become an urgent priority[90]. The following section provides an overview of the latest research progress in drug repurposing for CRC treatment, focusing on precision medicine perspectives, involving signaling pathway, metabolic reprogramming, and epigenetic regulation (Table 2).

Table 2 Recent advances and drugs in drug repurposing for colorectal cancer treatment.

Pathways	Drugs	Purposes
Wnt/β-catenin signaling pathway	Pyrotinib	TKI inhibitor to degrade β-catenin
	Risedronate	ADC to target therapeutic sites
EGFR/RAS/MAPK signaling pathway	Aspirin	NSAID to inhibit EGFR activity
	Simvastatin	Lipid-lowering drug to inhibit MAPK activity
PI3K/AKT/mTOR signaling pathway	MET	Antidiabetic drug to activate AMPK/NF-κB signaling pathway
Tumor inflammation and immune regulation	Tocilizumab	MRA to block IL-6 signaling
	Aspirin and pembrolizumab	Combinations to modulate tumor microenvironment
Metabolic reprogramming	2-DG	Glycolysis inhibitor to reprogram glucose metabolism
	Metformin	Antidiabetic drug to reduce lactate production
Epigenetic regulation	Vorinostat	HDAC inhibitor to induce histone hyperacetylation
Anti-angiogenesis	Bevacizumab	Monoclonal antibody to target VEGF
	Thalidomide	Treatment of nausea in pregnant women to inhibit VEGF pathway

EGFR: Epidermal growth factor receptor; MAPK: Mitogen-activated protein kinase; PI3K: Phosphatidylinositol 3-kinase; AKT: Protein kinase B; mTOR: Mammalian target of rapamycin; MET: Metformin hydrochloride; MRA: Tocilizumab; 2-DG: 2-Deoxy-D-glucose; TKI: Tyrosine kinase inhibitor; ADC: Antibody-drug conjugates; NSAID: Nonsteroidal anti-inflammatory drugs; AMPK: Adenosine 5’-monophosphate-activated protein kinase; NF-κB: Nuclear factor kappa-B; IL: Interleukin; HDAC: Histone deacetylase; VEGF: Vascular endothelial growth factor.

Wnt/β-catenin signaling pathway

The Wnt/β-catenin signaling pathway is an evolutionarily conserved mechanism that plays a pivotal role in regulating cell fate determination, organogenesis, tissue homeostasis, and various pathological conditions. Its dysregulation is also critically implicated in cancer, where it contributes to tumor initiation, progression, and therapy resistance[91]. Wnt/β-catenin signaling serves as a critical driver of CRC pathogenesis and progression, with approximately 80% to 90% of patients harboring mutations that lead to pathway activation[92-94].

The post-translational modifications of Wnt ligands primarily include glycosylation, palmitoylation, and acylation. Notably, acylation is essential for extracellular transport as well as receptor/co-receptor recognition and binding[95,96]. Pyrotinib is a tyrosine kinase inhibitor (TKI) that primarily targets epidermal growth factor receptor (EGFR) and human epidermal growth factor receptor 2 (HER2). It was initially developed for the treatment of HER2-positive breast cancer[97]. It was suggested that TKIs promote β-catenin phosphorylation and degradation in CRC. β-catenin degradation attenuates the activation of its downstream oncogenic genes, thereby inhibiting CRC cell proliferation, reducing stem-like properties, and enhancing treatment sensitivity[94].

Besides pyrotinib, non-oncologic drugs have also been identified as potential modulators of the Wnt/β-catenin pathway, suggesting their potential therapeutic role in CRC treatment. Risedronate, a bisphosphonate primarily used for the treatment and prevention of osteoporosis and Paget’s disease, functions by inhibiting osteoclast activity[98]. Further investigation has extended its effects beyond bone metabolism, as emerging evidence suggests it can modulate the tumor microenvironment. This mechanism is likely mediated through Wnt signaling inhibition, which may alter the tumor niche, thereby influencing cancer stem cell differentiation and proliferation[99].

EGFR/RAS/mitogen-activated protein kinase signaling pathway

EGFR can activate its downstream intracellular signaling cascades, including the mitogen-activated protein kinase (MAPK) and PI3K pathways, which serve as central regulators of cell proliferation, differentiation, survival, and migration. Aberrant activation of these pathways has been implicated in the progression and therapeutic resistance of various cancers, including CRC, lung cancer, and melanoma[100,101]. Importantly, the EGFR/RAS/MAPK signaling pathway is regulated by multiple genetic factors. In CRC, mutations in KRAS, NRAS, and BRAF lead to its constitutive activation, driving uncontrolled tumor growth and metastasis[102]. This genetic landscape also renders patients resistant to anti-EGFR monoclonal antibodies, such as cetuximab and panitumumab. Consequently, the urgent need to develop novel therapeutic strategies targeting the EGFR/RAS/MAPK pathway has become increasingly evident.

Aspirin can exert anticancer effects by inhibiting EGFR activity and restoring its normal expression. In CRC prevention, aspirin plays a particularly significant role in high-risk populations. By modulating inflammatory pathways and directly targeting EGFR, aspirin has a significant anti-cancer effect in reducing CRC incidence and CRC-related mortality, especially in early-stage patients[103-105]. In addition to aspirin, the lipid-lowering drug Simvastatin has demonstrated potential in inhibiting MAPK signaling activity in RAS-mutant CRC. However, the precise molecular mechanisms underlying its effects on tumor progression and therapy response remain to be fully elucidated, warranting further investigation[106,107]. The repurposing of these drugs in different genetic backgrounds provides a foundation for personalized treatment strategies, enabling tailored therapeutic approaches based on the molecular characteristics of individual patients.

PI3K/protein kinase B/mammalian target of rapamycin signaling pathway

As one of the most frequently activated pathways in human cancers, abnormalities in the expression or mutations of various components within the PI3K/protein kinase B (AKT)/mammalian target of rapamycin (mTOR) signaling pathway are closely associated with tumorigenesis. By promoting cell survival, growth, and cell cycle progression, the PI3K/AKT/mTOR signaling pathway significantly influences epithelial-mesenchymal transition (EMT) and metastasis. Therefore, the identification of drugs targeting the PI3K/AKT/mTOR pathway is a critical focus in the development of molecular targeted therapies for CRC[108,109].

The conventional antidiabetic drug Metformin hydrochloride (MET) exerts anticancer effects by activating the adenosine 5’-monophosphate-activated protein kinase/nuclear factor kappa-B (NF-κB) signaling pathway, leading to mTOR phosphorylation and subsequent inhibition of the mTOR complex 1 complex. This blockade of the PI3K/AKT/mTOR signaling pathway suppresses protein synthesis and expression of proliferation-associated genes, thereby reducing CRC cell proliferation. Additionally, MET has been shown to inhibit AKT activity and reprogram the tumor metabolic microenvironment, thereby enhancing immune responses. These multifaceted effects further highlight its potential as a promising therapeutic agent in cancer treatment[110]. It is likely that other non-cancer clinical drugs are capable of inhibiting the PI3K/AKT/mTOR signaling pathway. However, the identification of additional candidates requires further investigation and continued efforts from researchers.

Tumor inflammation and immune regulation

Chronic inflammation and immune evasion play pivotal roles in CRC pathogenesis, particularly in inflammatory bowel disease (IBD)-associated CRC and microsatellite instability-high (MSI-H) CRC. In IBD, persistent inflammation sustains epithelial damage and modulates immune responses, fostering a tumor-promoting microenvironment[111]. Moreover, MSI-H CRC is characterized by defects in the mismatch repair system. Although these tumors exhibit high immunogenicity, their ability to evade immune surveillance further contributes to tumor progression[111,112].

On the other hand, tumor-associated macrophages (TAMs) represent one of the most abundant immune cell populations within the tumor microenvironment and play a pivotal role in tumor progression. Activated TAMs secrete various pro-inflammatory cytokines, such as interleukin (IL)-6, as critical mediators of immune regulation, inflammatory responses, and EMT in tumor cells. Accordingly, TAMs contribute to the promotion of tumor cell cycle progression and the inhibition of apoptosis, thereby facilitating cancer progression[113-116].

Tocilizumab is a humanized monoclonal antibody targeting the IL-6 receptor, specifically blocking IL-6-mediated intercellular signaling. It has significant efficacy in the treatment of inflammatory diseases, including Castleman disease, rheumatoid arthritis, juvenile idiopathic arthritis, and Crohn’s disease[117]. Given its success in these conditions, tocilizumab holds potential as a therapeutic agent for IL-6-driven cancers, including CRC, where IL-6 signaling plays a critical role in tumor progression and immune evasion[118]. Moreover, the combination of aspirin with immune checkpoint inhibitors such as pembrolizumab holds promise for enhancing therapeutic responses in MSI-H CRC. Given its anti-inflammatory properties, aspirin may contribute to modulating the tumor immune microenvironment, thereby augmenting the efficacy of immune checkpoint blockade[119].

Metabolic reprogramming

Metabolic reprogramming, first identified as a hallmark of malignancies nearly a century ago, refers to the acquisition of metabolic traits that support cell survival, immune evasion, and proliferative growth in neoplastic cells[120]. Driven by oncogenic mutations, these metabolic alterations not only fuel tumor progression but also serve as potential biomarkers for cancer diagnosis, monitoring, and therapeutic targeting. CRC cells undergo metabolic reprogramming to sustain their rapid proliferation, presenting opportunities for therapeutic intervention[121]. The glycolysis inhibitor 2-Deoxy-D-glucose has demonstrated significant efficacy against glucose metabolism-dependent CRC[122]. Additionally, by reducing lactate production, metformin modulates the tumor metabolic microenvironment and suppresses tumor growth, making it a focal point of interest in this field[123].

Epigenetic regulation

Epigenetics refers to heritable changes in gene expression that do not involve alterations in DNA sequence. DNA methylation, histone modifications, and non-coding RNAs function as key epigenetic regulators in CRC development. These mechanisms orchestrate complex gene expression networks, influencing tumor initiation, progression, and therapeutic response, thereby presenting promising targets for precision medicine[124,125].

Histone deacetylases (HDACs) mediate the removal of acetyl groups from target proteins, including histones, transcription factors, and other cellular proteins, thereby modulating their function. HDAC inhibitor (HDACi) exerts its effects by inducing hyperacetylation of histones and other proteins, leading to the regulation of multiple gene expression programs. This includes genes involved in cell cycle arrest, cancer cell death, and apoptosis, highlighting HDAC inhibition as a promising therapeutic strategy in CRC and other malignancies[126]. The HDACi vorinostat exerts its effects by inducing histone hyperacetylation in CRC, thereby modulating epigenetic mechanisms that prevent the cancer progression. By altering chromatin accessibility and regulating key oncogenic and tumor-suppressive pathways, vorinostat represents a promising strategy for targeting epigenetic dysregulation in CRC[127,128]. Aberrant DNA methylation is a hallmark of cancer, characterized by global hypomethylation alongside hypermethylation of oncogenes and transcriptional silencing of tumor suppressor genes. By targeting this epigenetic dysregulation, previous studies have demonstrated that azacitidine can significantly reduce genome-wide DNA methylation, thereby restoring normal gene expression. Moreover, it exerts gene-specific demethylation effects, further enhancing its therapeutic potential in CRC[129,130].

Anti-angiogenesis

In contrast to normal cells, cancer cells can often induce neo-angiogenesis[131]. For patients with high vascular endothelial growth factor (VEGF) expression, anti-angiogenic strategies play a crucial role in disease management. Bevacizumab, a monoclonal antibody targeting VEGF, has been widely utilized in CRC treatment, demonstrating remarkable clinical efficacy. By inhibiting VEGF-mediated angiogenesis, bevacizumab disrupts tumor vascularization, thereby limiting nutrient and oxygen supply to cancer cells and enhancing the effectiveness of conventional therapies[132]. Thalidomide, originally synthesized and marketed decades ago as a treatment for respiratory infections, was later prescribed to alleviate morning sickness in pregnant women. However, due to its teratogenic effects, its clinical use was discontinued. In recent years, clinical studies have revealed that thalidomide and its analogs can suppress tumor proliferation and angiogenesis by inhibiting VEGF, PI3K/AKT, and NF-κB signaling pathways, thereby restraining tumor invasion and metastasis. Beyond its anti-angiogenic properties, thalidomide appears to possess additional advantages as an immunomodulator and anti-inflammatory agent, potentially counteracting tumor immune evasion and further enhancing its therapeutic potential in cancer treatment[133].

CHALLENGES FOR DRUG REPURPOSING

Insufficient investigation of biological mechanism

In drug development and repurposing, insufficient investigation of underlying biological mechanisms often hinders therapeutic optimization and precision medicine. While computational modeling, structural predictions, and clinical observations can facilitate the rapid identification of potential drug targets or indications, the absence of in-depth mechanistic studies may compromise the accurate prediction of therapeutic efficacy and impede our comprehensive understanding. In the development of anti-cancer therapies for CRC, some targeted therapies have demonstrated promising efficacy in preclinical studies[134]. However, the incomplete elucidation of their MoA may lead to unforeseen adverse effects or resistance issues during clinical trials. Moreover, the limited understanding of key signaling pathways and molecular regulatory mechanisms underlying disease progression can restrict the rational combination of therapeutics and narrow their clinical applicability, ultimately preventing these treatments from achieving their full therapeutic potential[135].

With the continuous advancement of precision oncology, the identification of therapeutic targets through sequencing and other genomic approaches has become an essential strategy for guiding clinical decision-making. Genomic sequencing of CRC has revealed its intrinsic heterogeneity, challenging the notion of CRC as a uniform disease. As tumors evolve, new mutations accumulate within individual cells and are inherited by subsequent generations of proliferating cells. This dynamic process gives rise to intratumoral heterogeneity, a phenomenon characterized by genetic divergence among different tumor cell populations. The genetic diversity of mutated genes in CRC gives rise to distinct molecular subtypes, each characterized by unique oncogenic drivers[136]. For instance, MSI-H subtype CRC often exhibits poor responsiveness to 5-fluorouracil[137], while microsatellite-stable CRC shows resistance to both conventional chemotherapy and monotherapy with targeted agents[138]. Similarly, KRAS-mutant breast cancer is generally unresponsive to anti-VEGF therapies[139]. This heterogeneity poses significant challenges for the effective deployment of molecularly targeted therapies, as the therapeutic targets vary across different CRC subtypes[140]. However, emerging studies have demonstrated that drug repurposing strategies, such as screening small-molecule targeted agents[141] or incorporating agents like aspirin that modulate the tumor immune microenvironment, may enhance tumoricidal effects in these otherwise treatment-resistant CRC subtypes.

Limitations of clinical trials

In the exploration of treatment strategies for CRC, drug repurposing has emerged as a promising area of research, with several agents originally developed for other indications being evaluated in CRC clinical trials. For example, the KRAS G12C inhibitor JNJ-74699157 (No. NCT04006301) has completed phase I trials and demonstrated good tolerability and preliminary antitumor activity in CRC patients harboring KRAS G12C mutations[142]. GO-203-2c (No. NCT01279603), which targets intracellular antioxidant systems, also completed a Phase I study, showing acceptable safety and potential efficacy in patients who had failed standard therapies[143].

Despite the pivotal role of clinical trials in evaluating the efficacy and safety of novel therapies, several inherent limitations must be acknowledged, which may impact treatment effectiveness and applicability. The enrollment of patients with specific molecular subtypes or rare cases of CRC remains a major burden. The stringent inclusion criteria in clinical trials often lead to a limited patient pool, reducing the statistical power and generalizability of the findings[144].

Besides, the survival benefits of repurposed non-cancer drugs are unclear. For instance, although mTOR inhibitors have demonstrated potential in suppressing tumor cell proliferation and angiogenesis, their clinical benefit in terms of overall survival remains limited[145]. This limitation may stem from the intricate regulatory network governing the mTOR pathway, the emergence of resistance mechanisms, and the inherent heterogeneity of tumors. Therefore, further investigation into the MoA of non-cancer therapeutic agents, combined with their strategic use alongside targeted therapies, may enhance their efficacy and broaden their clinical applicability.

More critically, therapeutic options for patients with advanced or drug-resistant CRC remain extremely limited. The inherently poor prognosis of these patients complicates the implementation of clinical trials, further restricting research efforts aimed at overcoming complex resistance mechanisms[146]. This paucity of research, in turn, hinders the development of effective treatment strategies, creating a cycle where limited therapeutic advancements perpetuate poor clinical outcomes and further impede trial feasibility.

Advancements in genomic analysis and AI-assisted drug development have substantially accelerated the process of drug repurposing. However, significant technical and data-related limitations continue to hinder its broader clinical application. A major constraint lies in the limited number of clinical trials dedicated to drug repurposing, which hampers comprehensive evaluation of both therapeutic efficacy and safety[147]. Furthermore, the lack of standardization in clinical and RWD pose considerable challenges to evidence integration and clinical translation. Addressing these challenges requires a concerted effort to systematically collect existing clinical data, refine analytical methodologies, and advance repurposing strategies for CRC, thereby enhancing the likelihood of successful clinical implementation.

CONCLUSION

Drug repurposing has emerged as a transformative strategy in the field of CRC treatment, offering a promising avenue to address the persistent challenges posed by this highly prevalent and deadly malignancy. The traditional drug development process, fraught with high costs, lengthy timelines, and significant risks of failure, has long been a bottleneck in delivering effective therapies to patients. In contrast, drug repurposing leverages the safety profiles and pharmacokinetic data of existing drugs, significantly accelerating the discovery process and reducing financial and temporal burdens. This approach has gained substantial traction in recent years, supported by multidisciplinary technologies and collaborative efforts across various fields. It has facilitated remarkable progress in drug repurposing for CRC, with several traditional non-cancer drugs demonstrating potential anticancer effects. The application of advanced technologies, such as high-throughput screening and AI, has further accelerated the drug discovery process. Computational strategies like three dimensional-QSAR and molecular docking have enabled the identification of new potential HDAC2 inhibitors, validated through in vitro experiments. The success of drug repurposing in CRC research is intrinsically linked to the synergistic integration of multiple disciplines. Molecular biology techniques, such as binding assays and phenotypic screening, provide critical insights into drug-target interactions and biological effects. Structural chemistry methods, including QSAR analysis and virtual screening, enhance the prediction and optimization of drug candidates. Computer science, particularly AI and machine learning, have revolutionized data analysis and prediction capabilities, enabling the extraction of valuable information from vast datasets. Information biology and real-world data integration further enrich the research landscape, offering comprehensive perspectives on drug mechanisms and patient outcomes. In conclusion, drug repurposing represents a beacon of hope in the ongoing battle against CRC. By building on the achievements of multidisciplinary research and embracing emerging technologies, we can unlock the full potential of existing drugs and deliver more effective, personalized, and accessible treatments to patients. The future of CRC therapy lies not just in developing new drugs but in reimagining the possibilities of those we already have, ultimately transforming the landscape of cancer care and improving outcomes for millions worldwide.