Can Drug Repurposing Accelerate Precision Oncology?

Cancer remains one of the leading causes of death worldwide, killing nearly 10,000,000 patients globally per year (1). In the past decades, our understanding of biology and treatment of cancer has improved significantly. Yet there is only a modest impact on mortality in the area of metastatic disease, exposing a clear and unmet need to improve systemic treatment outcomes (2). Traditional drug developers such as academia together with pharma and biotech generally follow conventional routes of drug development. This route, however, is a lengthy process with many hurdles and failures, including issues such as target specificity, poor pharmacodynamics and pharmacokinetics, and toxicity. These challenges are reflected in the low success rate of developing anticancer drugs, as approximately 7% of all drugs entering a phase I trial eventually reach market authorization (3). On average, development of novel cancer compounds costs ∼US$648 million and takes several years to more than a decade (4).

Therefore, there are clear benefits to the use of existing drugs for novel therapeutic strategies, known as drug repurposing, because safety, dosing, and pharmacokinetic properties are largely delineated. Classically, drugs in development for or approved in areas other than cancer are considered repurposing candidates, with aspirin and metformin as the most prominent examples in the last few decades. An abundance of clinical trials have been performed since the first clinical evidence suggesting these agents were possibly active anticancer treatments, respectively 33 and 16 years ago (5, 6). Despite these efforts, aspirin and metformin are not approved as effective treatments in patients with advanced cancer, although aspirin has been implemented in guidelines on prevention of colorectal cancer. Furthermore, the main advantages of drug repurposing are negated when drugs are used completely outside their intended indications and require different dosing or drug administration. For example, the dosing of methotrexate in autoimmune disorders (7.5–30 mg/week) is generally lower than when used as an anticancer drug (from 7.5 mg/m²/week up to 12 g/m²/week in high-dose regimens), with implications for safety and pharmacokinetic properties (7). Consequently, the road to clinical use of drug repurposing candidates as cancer drugs is generally not as easy as theoretically implied.

However, the definition of drug repurposing within precision oncology can be redirected to anticancer drugs that are used off-label and drugs that have been developed as cancer treatment but were discarded or deprioritized due to lack of efficacy or toxicity. With the expanding armamentarium of anticancer agents—more than 25% of new clinical trials involve these (8)—and use of molecularly guided drug discovery and development, optimizing the use of already available anticancer agents provides a potentially valuable approach to drug repurposing in precision oncology. Furthermore, cancer drugs are generally developed as single-agent treatments. With the insight that often multiple drivers cause cancer progression and the modest effect of single-agent treatment in patients with cancer, combination therapy has been regarded as a logical next step (9). Unfortunately, development of new combinatorial treatment approaches with a multitude of possible combinations has significant challenges. Drug repurposing may accelerate testing the concept of genomics-guided combination therapy. In this review, we therefore primarily focus on our definition of “intelligent” drug repurposing in precision oncology. We discuss off-label use of anticancer drugs, repositioning of deprioritized anticancer compounds, and combining repurposed agents in adaptive clinical trial platforms to accelerate access to promising oncologic drug-repurposing candidates.

Thalidomide, a drug that was initially approved as a sedative and used for morning sickness in pregnancy, was quickly withdrawn from market after recognition of its teratogenic effects (10). The drug made a reappearance on the market in the late 1990s as an effective treatment option for patients with multiple myelomas. In the same decade, bortezomib was identified as an anticancer drug in the first large-scale compound screening initiative. This initiative, set up by the NCI Developmental Therapeutics program, consisted of a screen of thousands of different chemical compounds ranging from natural products to FDA-approved drugs in a set of 60 human cancer cell lines (NCI-60; ref. 11). Bortezomib is now also approved for treatment in multiple myeloma.

These examples fueled the principle of using existing compounds, either drugs early in development or already approved therapies, as cancer treatment. Most prominently, aspirin and metformin were identified to have cancer preventive benefits based on (serendipitous) findings in large patient cohorts. Although aspirin use is now implemented in guidelines for prevention of colorectal cancer, lack of a clear mechanism of action and a well-defined target population have challenged clinical implementation of these agents (12, 13).

Like the NCI-60 compound screen, other high-throughput screening initiatives have been performed to systematically analyze antineoplastic properties of agents in molecularly defined cell line panels. As such, several potential candidates for drug repurposing as anticancer agents have been identified (Table 1; refs. 14–19). However, translation of these findings into clinical practice proved to be challenging on multiple levels. First, conducting clinical trials to confirm antineoplastic efficacy in patients with cancer is a lengthy and strenuous effort. Of the 10 compounds that have shown promising antineoplastic activity in large high-throughput screening efforts, seven have not (yet) been tested in clinical trials (Table 1). Second, toxicities may be limiting clinical use. For example, navitoclax, a BCL2 family antagonist initially developed for use in hematologic malignancies, was proposed as treatment for cancers with β-catenin mutations in a screen of 354 small molecules in 242 human cancer cell lines (15). Efficacy of navitoclax in CTNNB1-mutated tumors, however, has never been explored due to dose-limiting side effects (thrombocytopenia) of the drug (20). Third, clinical activity might be restricted by unfavorable drug pharmacokinetics. Disulfiram, a drug used to treat alcohol abuse, was identified as a possible agent against prostate cancer in a compound screen of 4,910 drug-like small molecules on prostate (cancer) cell lines (16), an effect that was confirmed in a later screen of 4,518 different drugs in 578 human cancer cell lines (19). Disulfiram has been tested in phase I and II trials in various tumor types, including prostate cancer. Although some clinical benefit has been suggested when used in combination with chemotherapy (21), no patients with clinical benefit in terms of RECIST 1.1 response or decline in PSA levels were reported (22). Unfavorable pharmacology, causing insufficient biological availability of disulfiram and its metabolite diethyldithiocarbamate, may have contributed to the lack of clinical responses. Finally, as is also the case for conventional drug development, drug repurposing is challenged by the complexity of translating drug activity of relatively homogeneous in vitro models to the heterogeneous patient setting. For example, PARP inhibitors were discovered to be active in Ewing sarcoma cell lines characterized by EWSR1–FLI1 translocations in a screen of 130 drugs in several hundred cancer cell lines (17). Consequently, olaparib was studied in 12 patients with Ewing sarcoma, but failed to show any radiologic response. As the median time to disease progression was only 5.7 weeks, the trial was discontinued (23). The discrepancy between preclinical activity and lack of clinical efficacy could potentially be explained by the overall complexity of translating in vitro and in vivo preclinical drug activity to the patient setting (24).

Drug repurposing of agents with known clinical activity in patients with cancer is considered an alternative, less risky strategy. Toxicity and pharmacologic profiles have already been established, and the agents have proven their efficacy in patients with cancer. Two main strategies can be applied. First, exploiting off-target activity—biological activity as a result of drug binding outside the intended target—can be used as a rationale for drug repurposing (Fig. 1). For example, imatinib was initially developed for chronic myeloid leukemia with a high affinity for the chimeric BCR–ABL fusion oncoprotein. However, after the observation that imatinib inhibits activated KIT, the association with gastrointestinal stromal tumor (GIST), a tumor type characterized by constitutive KIT activation, was rapidly made. Subsequently, and after rigorous testing in patients, imatinib is now approved for KIT-mutated GIST (25). Even selective inhibitors appear to affect at least one additional protein kinase and may also affect kinases without a close relation to the primary structure of the intended target (26). Platforms to explore interactions of kinase inhibitors with the human kinome are applied to systematically predict potential side effects and off-target therapy opportunities (27–30). Although these assays provide the opportunity to analyze drug interaction with the normal human kinome, another level of complexity is added by the fact that biological activity of kinases can be altered by specific mutations, and, consequently, different mutations within the same gene might require different treatment strategies. For example, BRAF mutations can be classified in three distinct classes with different therapeutic approaches due to differences in dependence on upstream RAS activation, kinase activation, and downstream kinase activity effects (31). To be able to scale up testing for off-target activity of kinase inhibitors in a mutation-specific setting, in silico binding assays—that is, computational approaches to predict drug–target interactions—have been developed and allow for testing a hypothetically endless number of potential combinations. For example, Keiser and colleagues were able to compare 901,590 different drug–target combinations by estimating ligand-based similarities across 3,665 compounds, identifying 23 new drug–target associations (32). With recent breakthroughs in protein structure predictions, mutation-specific binding assays might become a promising opportunity to identify drug-repurposing candidates (33).

The second strategy involves the use of therapies in mole­cularly similar tumors of different histologic subtypes (Fig. 1). Although this principle appears rather accessible, there is significant room to accelerate access to targeted agents for tumor types outside the initial label. Trastuzumab was long approved in HER2-amplified breast cancer, but it took more than 12 years for approval of trastuzumab in HER2-amplified gastric cancer. This illustrates an important lesson for the future: With an ongoing wave of novel targeted agents being approved, it is key to create broad access to molecular diagnostics regardless of tumor type to identify all patients who may potentially benefit from a targeted agent and simultaneously perform studies to demonstrate clinical benefit of these agents, even in rare cancers (34).

One of the main hurdles in this aspect is the unclear influence of tissue context. After the enthusiasm of the efficacy of BRAF-directed treatment in melanomas, the lack of activity in BRAF-mutated colorectal cancers was disappointing (35). Likewise, the specific KRAS^G12C inhibitor sotorasib showed limited clinical activity in patients with colorectal cancer compared with patients with non–small cell lung cancer (36). Tissue-specific responses to targeted therapies have also been reported for pan-HER kinase inhibitors in ERBB2-mutated tumors (37). Even with similar alterations [e.g., ERBB2 hotspot extracellular domain mutations (S310)], responses appear to be tissue type–dependent, as clinical efficacy was observed in breast, cervical, and biliary tumors, but not in bladder cancer.

On the contrary, tissue context seems to be less relevant for other drugs. Pembrolizumab has been approved for tumor-agnostic treatment of mismatch repair–deficient and tumor mutational burden–high solid tumors, and the TRK inhibitors larotrectinib and entrectinib have been approved for all solid tumors with an NTRK gene fusion (38, 39). These findings have accelerated the efforts to develop tumor-agnostic therapies (40). Nevertheless, although tumor-type agnostic therapy approaches are becoming more common, tissue context and presence of a specific molecular target remain factors to be carefully considered in drug repurposing (41). In this light, tissue-dependent clinical responses might prove to be less prominent than initially thought. Whereas BRAF^V600E mutations were seen as a prime example of tumor lineage context–dependent oncogenic effects, single-agent BRAF inhibitor therapy showed responses in 13 nonmelanoma cancer types with BRAF^V600E mutations (42). High-throughput screens provide the possibility to not only uncover genetic correlates of targeted therapy efficacy but also incorporate lineage dependencies of specific tumor types (18). For example, unresponsiveness of colorectal cancer cell lines to vemurafenib through activation of CRAF was confirmed in the Cancer Therapeutics Response Portal (15). These preclinical models could therefore function as a method to determine upfront whether drug-repurposing candidates should be applied in a tumor-agnostic or tumor-specific manner in clinical trials.

Adaptive clinical trials provide the opportunity to look for early signs of activity of drugs in molecularly defined patient populations while simultaneously safeguarding patient safety. For compounds with presumed activity in settings other than the original label, such trials can facilitate the validation of concept. This type of trial design, specifically focusing on signal-finding, has already proven its worth in off-label repurposing of available anticancer drugs (43).

Off-label use of anticancer drugs can be a fast and (relatively) low-cost method to identify novel treatment opportunities for cancer patients with unmet medical needs. Here, approved targeted agents or immunotherapies are matched with patients having a suitable molecular profile, but not the histologic tumor type for which the drug was originally approved. This includes use of targeted therapy in molecularly similar tumors of different histologic subtype (e.g., HER2-directed therapy in HER2-amplified carcinomas other than breast or gastric cancer; ref. 44) and off-target effects of kinase inhibitors (e.g., afatinib for tumors with an NRG1 fusion; ref. 45). Drugs are generally selected based on the presence of a molecular target in a tumor type incongruent with the approved drug label. In daily practice, off-label prescription is generally an opportunistic approach to help patients, but the absence of prospective data collection prevents evidence-based decision-making on further drug development or approval. Fortunately, this problem has been recognized, and off-label drug efficacy is now increasingly evaluated in novel trial designs as a means to address these issues.

One of the first trials studying the effect of molecularly targeted off-label treatment reported discouraging results: In the SHIVA trial, molecular profile–driven treatment was compared with physicians’ choice therapy in patients with treatment-refractory cancer (46). No significant improvement of progression-free survival was seen, although insufficient power has been suggested as an explanation for this finding. Furthermore, the study made use of only 11 targeted agents and was therefore limited in its scope. The trial, however, proved pivotal: Studies investigating the value of oncologic agents outside their registered label have since been broadly conducted. Commonly used trial designs include umbrella trials, in which molecularly targeted treatments are focused within one histologic tumor type; basket trials, in which off-label efficacy of targeted therapies in tumor-agnostic cohorts with a specific molecular alteration is investigated; or combinations thereof. Large initiatives include the U.S. NCI-MATCH (NCT02465060) and I-PREDICT (NCT02534675) trials and the French MOSCATO 01 study (NCT01566019) that, despite minor differences in design, combine molecular screening programs with multiple parallel cohorts that include patients based on the presence of predefined aberrations. These trials have shown the feasibility of exploring efficacy of multiple treatment modalities, including chemotherapies and targeted therapies, within one master protocol in both off-label and experimental settings and paved the way for future studies.

In an effort to tackle issues concerning small patient numbers and accelerate access to approved drugs outside their label, international consortia for off-label oncology have been developed. The first innovative approach in this direction came from two similar basket and umbrella platform studies that were independently developed: DRUP (the Netherlands, NCT02925234) and TAPUR (the United States, NCT02693535). These studies, both initiated in 2016, have demonstrated the feasibility of international collaboration between separate yet similar trials. Conceptually, the studies include European Medicines Agency (EMA)– and/or FDA-approved, molecularly targeted agents, which are subsequently tested for efficacy and safety in patient subgroups harboring a specific molecular variant but have a tumor type that does not correspond to label criteria. In small cohorts, early signs of activity are studied, and only further explored in larger patient groups if a minimum threshold of activity has been observed. Adaptive trial protocols allow rapid initiation of new cohorts either by incorporating new drugs in the protocol or by defining new target populations for already included drugs. In response to these initiatives, multiple similar trials have recently been initiated, including CAPTUR in Canada (NCT03297606) and more recently ProTarget in Denmark (NCT04341181), IMPRESS in Norway (NCT04817956), and MEGALiT in Sweden (NCT04185831). Data sharing between these similarly structured studies is a key part of the study objectives.

In the past years, however, it has become clear that these innovative studies face significant challenges. The first main challenge lies on a diagnostic level: Current molecular diagnostics are generally aimed at the detection of aberrations in genes that are relevant for standard-of-care decisions (e.g., HER2 in breast cancer and EGFR in non–small cell lung cancer) but fail to include a wider range of potential actionable targets. Large sequencing studies have revealed the potential of comprehensive molecular profiling to detect therapy options for patients, detecting actionable targets in 70% to 90% of patients (47–49). Broad molecular profiling holds the advantage to optimize allocation of patients to clinical trials and not only across available cohorts within one clinical trial. For example, in the NCI-MATCH trial (NCT02465060), in which molecular profiling was embedded in the trial protocol, only 17.8% of participants could be allocated to a cohort within the trial. This select number of patients who can be matched to personalized treatment in precision oncology trials has been a recurring point of criticism (50). Fortunately, many comprehensive cancer centers have started to incorporate whole-genome, whole-exome, or large panel sequencing approaches in their diagnostic work-up of patients with cancer instead of in study context, as has been recommended in international guidelines for cancer molecular diagnostics (51).

Second, a major challenge lies in accrual of sufficient patient numbers to definitively conclude the value of a treatment. Molecular profiling of tumors has led to further subclassification of tumor types with consequential low patient numbers per cohort, and the small numbers make it challenging to perform large-scale randomized clinical trials in a financially feasible manner and acceptable time frame. Given that standard evidence appraisal methodologies, relying on multiple trial phases and comparisons to control arms, are unfit for assessment of single-arm studies, delays in drug evaluation and subsequent approval are inevitable. Aforementioned basket trials, in which potency of a molecularly targeted treatment is validated in a tumor-agnostic fashion rather than traditional histology-based classification in umbrella trials, have been posed as a strategy to overcome insufficient patient numbers of trials that focus on a single tumor type, but this trial design also has its own shortcoming: Negation of histology assumes tissue context plays no role in drug efficacy. Results of basket trials should therefore be considered with caution but have nevertheless recently led to the first tumor-agnostic regulatory approvals in oncology (38, 39). Alternatively, these trials could serve to generate evidence to refrain from further treatment exploration if 0 patients respond in a small, defined cohort of 8 to 12 patients with cancer. Ideally, preclinical models that incorporate genetic and lineage dependencies could be used to determine optimal clinical trial strategies for each compound upfront, and clinical data can subsequently be used to improve these algorithms.

Taken together, dynamic trial designs may pose a durable solution to the continuing need for new treatment options in cancer, by enabling low-threshold efficacy exploration of anticancer drugs in the context of drug repurposing, in molecularly defined cohorts (Fig. 1).

Despite previous successes, precision oncology has been the subject to criticism in terms of effectiveness. Confined response rates and number of exceptional responders not outweighing number needed to treat have been reported as limitations of the paradigm (50). Given the complexity of cancer, broad consensus exists that single-agent treatment may only have a modest impact on survival of patients with cancer due to resistance mechanisms and the presence of multiple genomic drivers in a tumor (52). Optimization of precision oncology efficacy may therefore lie in combining targeted agents. Screens that focus on additive, synergistic, or synthetic lethal combinations of drugs have been successfully performed. Also, in recent precision oncology trial designs, experimental combinatorial regimens have already been introduced. Rodon and colleagues reported correlation between high matching scores [defined as the number of molecular alterations that could be matched to a drug (or drugs), divided by the total number of alterations present within a patient] within the French WINTHER trial (53). Similar findings in a comparable patient group were later reported by Sicklick and colleagues in the I-PREDICT study, where high matching scores were significantly correlated with improved disease control rate and survival outcomes (54). Although no comparisons were made between monotherapy and combinatorial treatment, the results emphasize the need for an innovative approach to precision oncology, wherein combinations of available drugs can possibly enhance efficacy.

Deprioritized anticancer drugs might prove to be suitable candidates for these combinatorial approaches. As mentioned previously, targeted therapy with BRAF + MEK inhibitors failed to show clinical efficacy in BRAF^V600E-mutated colorectal cancer, in contrast to BRAF-mutated melanoma. After preclinical models suggested feedback activation through EGFR as a cause of inadequate pathway signaling inhibition with BRAF inhibitor monotherapy, these findings were translated to clinical trials evaluating combination therapy with BRAF, MEK, and EGFR inhibitors (35). Rather than deprioritizing BRAF + MEK inhibitors for patients with colorectal cancer due to inefficacy, these clinical trials eventually led to registration of this treatment combination (55).

In addition to disappointing clinical single-agent activity, safety issues may also be a reason to refrain from further drug development. But even in that situation, opportunities for use may lie in different contexts; navitoclax failed as monotherapy in lymphoid malignancies due to dose-limiting neutro- and thrombocytopenia but is now identified as an effective agent in different (pre)clinical models for combinatorial regimens in two distinct approaches: It proved effective in a synthetic lethal interaction with PARP inhibition in preclinical ovarian and breast cancer models (56), and in combination with venetoclax and chemotherapy in patients with acute lymphoblastic leukemia and lymphoblastic lymphoma (57). Several other examples of unsuccessful monotherapies that succeeded in combinatorial strategies such as ERN1 kinase inhibitors and selective KRAS^G12C inhibitors have been reported (58, 59). Strategies to uncover rational combinations, in terms of compounds and treatment schedules, have recently been comprehensively reviewed and could facilitate future research in this field and unlock the therapeutic potential of depriotized compounds (9).

Although there is a good rationale for combinatorial approaches and candidate drugs are readily available for proof-of-concept testing, there is also a significant level of complexity with regard to safety and pharmacokinetic interactions to be considered. For experimental agents, with limited knowledge of side effects and pharmacokinetics, multidrug combination phase I studies need to be carefully conducted. Alternatively, combining compounds with well-established toxicity profiles and pharmacokinetics has the advantage that side effects can be better anticipated, and overlapping toxicities can be avoided or reduced. Interestingly, reduced toxicity of combination regimens has been observed clinically (60) and could be expected when individual agents are combined in lower doses (61). Moreover, several studies demonstrated the feasibility of de novo combinations of two and three targeted and/or cytotoxic agents when drug doses (either an FDA-approved, recommended phase II dose or maximum tolerated dose) were already established (62).

In the era of precision medicine, increasingly refined molecular profiles require tailor-made treatment approaches. Drug repurposing is a promising strategy to accelerate access to novel treatment opportunities for patients with cancer. Although the concept of drug repurposing was originally proposed for nononcologic drugs, drug repurposing of anticancer drugs that were deprioritized or used only for specific labeled indications has now shown significant clinical value. However, several factors are essential for successful drug repurposing. First, a clear molecular target should be identified to ensure adequate patient selection for clinical trials. Second, repurposing candidates could benefit from preclinical identification of genetic targets and lineage dependencies to determine optimal clinical trial strategies upfront. Third, the presence of an established safety and pharmacokinetic profile is very valuable to determine and accelerate combinatorial treatment approaches.

The large number of candidates for drug repurposing and experimental drugs, together with novel technologies such as machine learning, bioinformatical pipelines, in silico analyses, and innovative trial designs, provides a unique opportunity to accelerate precision medicine (Fig. 1).

L.J. Zeverijn reports grants and nonfinancial support from Amgen, AstraZeneca, Bayer, Bristol Myers Squibb, Clovis Oncology, Eisai, Lilly, Merck Sharp & Dohme, Novartis, Pfizer, and Roche, and grants from the Barcode for Life Foundation and The Dutch Cancer Society outside the submitted work. M.J. Garnett reports grants from Open Targets, AstraZeneca, GlaxoSmithKline, and Astex Therapeutics and other support from Mosaic Therapeutics outside the submitted work. No disclosures were reported by the other authors.