Emerging Trends in Cancer Drug Discovery—From Drugging the “Undruggable” to Overcoming Resistance

The past decade has brought major advances in cancer biology, including breakthroughs in genomics technology that have yielded DNA sequences of thousands of tumors, transcriptomic profiling of tumors down to the single-cell level, functional characterization of genes through whole-genome CRISPR screens and identification of numerous drivers of tumorigenesis. Unfortunately, many of the candidate therapeutic targets that have emerged from these advances have been deemed “undruggable” according to conventional definitions (1), and are frequently characterized by large protein–protein interaction surfaces that lack obvious ligandable pockets. However, the drug discovery community has become increasingly engaged in a call to action on “drugging the undruggable,” as illustrated by several groundbreaking discoveries. A key insight that emerged is that druggability is a function of the technology status quo at a given time. Its boundaries are ever evolving, and we postulate that every candidate target may eventually become druggable. Although such a statement may seem provocative, we believe that there are sufficient historical and current data to support the case.

In this small molecule–centered Perspective article, we consider the example of drugging KRAS to understand the factors that drove the innovation required to tackle this notoriously undruggable target. We also discuss the discovery of other targets previously deemed intractable and hone in on the modality of chemically induced proximity, a platform integrating numerous distinct approaches, each with its own advantages and limitations. We then highlight the progress on novel screening strategies against poorly ligandable targets and provide an outlook on the potentially disruptive technology of targeting macromolecules at the RNA level. We end with a perspective on the challenging problem of drug resistance and promising developments that pave a path toward more durable efficacy and, eventually, cures.

KRAS, a membrane-bound guanine nucleotide-binding protein, is the most frequently mutated oncogene in human cancer, with mutations detected in ∼30% of all tumors. Consequently, substantial efforts across industry and academia have been devoted to targeting KRAS for several decades, but with little success (2). Initially, the focus was on the posttranslational modifications of KRAS and the targeting of the requisite enzymes (3). However, the broad substrate profile of these enzymes has challenged the ability to effectively deploy this strategy to selectively target RAS proteins. Among the various approaches that have been pursued, direct targeting of KRAS has generally been deemed the most attractive, but it has been challenged by the absence of an apparent binding pocket and the fact that nucleotide competition is not feasible due to the very strong binding affinity and high cytosolic concentration. Furthermore, selective targeting of KRAS over its homologs HRAS and NRAS and, particularly, mutant KRAS over the wild-type protein was believed to be necessary to achieve a sufficient therapeutic window, presenting another layer of complexity for drug discovery researchers.

The year 2013 marked a seismic shift in the field—when Kevan Shokat published his breakthrough results describing the mutant-selective covalent targeting of KRAS^G12C (4). Key to that success was a cysteine-based tethering strategy, a technology developed by Shokat's colleague at the University of California, San Francisco, Jim Wells, which was deployed against a bacterially expressed form of the KRAS^G12C protein in the GDP-bound form. Elaboration of the screening hits combined with X-ray crystallography confirmed modification of Cys-12 and revealed a breakthrough insight: the presence of a cryptic binding pocket in the switch II domain of KRAS that had not been observed previously (Fig. 1A). Mechanistic studies with further optimized compounds exposed an important feature of KRAS^G12C, intrinsic hydrolysis of its GTP-bound form to its GDP-bound form, offering an intervention point for therapeutic targeting. This discovery provided the proof-of-concept that KRAS^G12C GDP targeting might indeed be a therapeutically viable approach. Fast-forwarding to 2021, five different KRAS^G12C small-molecule covalent inhibitors are now in clinical trials specifically to treat patients with cancer whose tumors harbor KRAS^G12C mutations, with AMG-510 and MRTX849 currently being evaluated in Phase II monotherapy and combination clinical studies (5).

Covalent switch II pocket binders provide an elegant solution to drugging KRAS^G12C, and it is impressive that this work occurred within a relatively short time frame—five years from the first publication to the first Phase I compound. Notwithstanding the tremendous achievement, we can then ask why it took almost 40 years to therapeutically unlock mutant KRAS. Among the potential explanations are: (i) covalent screening (“tethering”) as a lead discovery modality was only recently adopted; (ii) the switch II binding site is a cryptic pocket and was previously not observed; and (iii) the relatively high hydrolysis rate of the GTP-bound form of KRAS^G12C was not widely known, discouraging researchers from considering the GDP-bound form as drug target. This experience has highlighted two critical elements of success in drugging poorly tractable targets—or perhaps in innovation in general: technology advancement and the courage to challenge dogma. As we will elaborate below, these two factors have played a consistent role in several other breakthrough discoveries.

Although the results achieved in drugging the G12C mutant have certainly been encouraging, the KRAS field as a whole still faces significant challenges. Preclinical and clinical studies with KRAS^G12C inhibitors show that not all tumor types are equally responsive, and targeting of colorectal cancer in particular, an indication that has been notoriously challenging for the development of MAPK pathway inhibitors, remains difficult. Durability of efficacy, even in more responsive tumor types, might be hampered by acquired drug resistance and will likely ultimately require combination treatments. Another unanswered question is whether the concept of switch II pocket binding, leveraged for G12C, can be applied to other KRAS mutants, particularly the highly prevalent G12D and G12V mutations. The corresponding mutant residues are not readily targetable via covalent modifications, and hydrolysis rates are slower, equating to smaller populations of potentially targetable GDP-bound forms.

Although successfully drugging KRAS^G12C represents notable progress against challenging targets, two major challenges remain as we pursue additional unconventional targets. First, a large number of the most highly recurrent oncoproteins and tumor suppressors have yet to be drugged. This group includes transcription factors such as MYC, β-catenin, androgen receptor V7, and tumor suppressors such as p53 and PTEN (6). Second, resistance to small-molecule drugs as monotherapies remains a major impediment to achieving durable efficacy in patients. As we will describe in the following sections, we believe that the research community has fully embraced these challenges and is well equipped to make major progress in addressing them with innovative solutions. A look back helps to instill optimism. We noted earlier that technology developments and the motivation to challenge dogma have been consistent factors in pushing the limits of druggability. Table 1 lists target classes pursued in cancer drug discovery that defied their original description as “intractable.”

Among the modalities listed in Table 1, the degradation of poorly ligandable zinc finger transcription factors (IKZF1/3) elicited by the phthalimide class of small-molecule ligands known as IMiDs warrants, in our opinion, particular attention, as it represents a specific case of target modulation via a proximity-inducing small molecule, a groundbreaking concept with broad potential. In 2010, cereblon (CRBN), a component of the DDB1–CRBN E3 ligase complex, was identified as the molecular target of the IMiDs, and further studies revealed the ubiquitination and subsequent CRBN-mediated degradation of the neosubstrates IKZF1 and IKZF3 in response to IMiD treatment. Interfacial binding of the IMiD ligand is essential to induce a protein–protein interaction between CRBN and the target proteins, providing a good example of what is now termed a “molecular glue” (7), exemplified in Fig. 1B (8). The findings described above motivated efforts to harness IMiDs as CRBN-recruiting ligands and conjugate them with ligands for various other targets to effect their degradation. Targeting the BRD4 oncoprotein using its previously established ligand JQ1 delivered one of the first examples of engineered heterobifunctional protein degraders, also called proteolysis-targeting chimeras (PROTAC; ref. 9). Pioneering work by Craig Crews, Jay Bradner, and others elevated the field from a scientific curiosity to a systematically pursued drug-discovery modality, with several E3 ligases co-opted (e.g., VHL, IAP, and others), and now exemplified with many reported examples, a prominent one being the VHL-based PROTAC MZ1, illustrated in Fig. 1C (10). Since 2015, PROTAC-related publications have continuously doubled every year, with 268 in 2020 and more than 650 in total, and as of December 2020, three heterobifunctional degraders have advanced into clinical trials, two degraders of the androgen receptor (ARV-110, CC-94676) and one of estrogen receptor (ARV-471). Degraders have the potential for differentiated therapeutic activity versus a small-molecule inhibitor, with an extended durability of response (target resynthesis is required to regain activity). Importantly, they expand the universe of druggable proteins and binding sites, for two key reasons:

1. Proteins can be degraded regardless of ligand function. Less than 20% of the proteome is associated with enzymatic activity, and having the ability to functionalize a mere binder offers unprecedented access to new target space.

2. Selectivity can be engendered from nonselective target ligands. Achieving degradation of the protein of interest (POI) requires bringing E3 ligases into proximity with the POI and ubiquitination of distinct lysine residues in the POI. These required events, POI–E3 ligase complex formation and specific ubiquitination, provide selectivity handles that are separate from the selectivity of ligand–POI binding.

Another advantage is the potential for obtaining improved safety profiles by virtue of tissue-restricted ligase expression/activity. However, heterobifunctional degraders come with practical limitations that are important to consider. Due to their size and physicochemical properties, achieving cellular penetration and in vivo exposure, especially following oral dosing, can be extremely challenging, often necessitating extensive optimization, formulation, or conjugation efforts. Furthermore, built into the design concept of PROTACs, a target ligand with adequate affinity needs to be available to enable a PROTAC discovery campaign, which is often the central problem faced when pursuing poorly tractable targets.

The drug property downside of PROTACs is elegantly addressed with the above-mentioned glue molecules, compounds that are small and bind at the interface between two proteins. Discovering such molecules, however, is still largely empirical. Celgene scientists aimed to address this downside by creating large IMiD ligand–based chemical libraries and prospected new degrader targets using phenotypic screening. This work, together with structural studies, revealed that the discovered neosubstrates all possess a common degron, a beta-hairpin with a characteristic glycine residue, providing useful information on the scope of this class of compounds (11). Glue molecule–mediated degradation is not limited to CRBN recruiting systems. Recent work identified arylsulfonamides, such as indisulam, to promote ternary complex formation between a DCAF15-containing E3 ligase complex and the RNA splicing factor RBM39, promoting its degradation (12). Furthermore, a research team led by Jay Bradner established that VHL ligands too can be harnessed as glue molecules, and discovered and structurally characterized an optimized compound that binds at the interface of VHL and the nonheme iron enzyme CDO1 (13).

The approaches described above represent ever-expanding “fishing expeditions” to find neosubstrates of potential therapeutic value. The opportunities are vast, but a major element of serendipity is needed to properly engage the targets that are of highest interest. The examples of fulvestrant and other selective estrogen receptor degraders, the BCL6 degrader BI-3802, and the selective BRD4 degrader GNE-0011 illustrate the exciting possibility of engendering target ligands with molecular features that can impart degradation (14–16). The degradation mechanisms utilized by such target ligand–based degraders are not known in all cases and could be diverse, distinguishing these compounds and this discovery approach from PROTACs and E3-ligase ligand–based glue molecules (13).

Recently, two papers published by Ebert and colleagues shed light on the mechanistic basis for two classes of monovalent degrader molecules. BI-3802 was identified in a screen for BCL6 inhibitors and found to unexpectedly degrade its target (16). Elegant biochemical and cellular studies showed that the ligand induces reversible polymerization of BCL6, followed by its sequestration into cellular foci and subsequent degradation. Cryo-electron microscopy revealed how the solvent-exposed moiety of BI-3802 contributes to a composite ligand–protein surface that engages BCL6 homodimers to form a supramolecular structure (Fig. 1D), followed by ubiquitination by the SIAH1 E3 ubiquitin ligase. Importantly, SIAH1 functioned as an E3 ligase for BCL6 even in the absence of BI-3802, and drug-induced formation of BCL6 filaments was shown to strengthen the interaction with S1AH1 and thereby accelerate degradation.

In the second report, Ebert and his team identified a degrader through the systematic mining of databases for correlations between the cytotoxicity of a large set of clinical and preclinical small molecules and the expression levels of E3 ligase components across hundreds of human cancer cell lines (17). Their studies led them to compound CR8, a cyclin-dependent kinase (CDK) inhibitor that appeared to function as a molecular glue degrader. The CDK-bound form of CR8 has a solvent-exposed pyridyl moiety that induces the formation of a complex between CDK12–cyclin K and the CUL4 adaptor protein DDB1 (Fig. 1E), bypassing the requirement for a substrate receptor and presenting cyclin K for ubiquitination and degradation. Of note, a weak interaction between CDK12–cyclin K and DDB1 was observed even in the absence of drug and in the presence of CR8, and this interaction was strengthened by 500 to 1,000-fold, driving degradation of cyclin K. Both of these examples underscore one key feature of molecular glue molecules, namely, that they may often function by strengthening a preexisting low-affinity protein–protein interface rather than inducing a de novo protein–protein interaction. Technologies to identify these low-affinity protein–protein interactions may enable more hypothesis-driven approaches to molecular glue-like molecules. Furthermore, in both of these examples it was the solvent-exposed regions of the small-molecule inhibitor that appeared to be responsible for these strengthened interactions, suggesting that traditional approaches to optimizing binding of small molecules to active sites using structure-based drug design may not be applicable to molecular glue-like molecules. Rather, these low-affinity interactions first need to be identified in the native context (i.e., in cells), followed by optimizing the interaction using functional assays evaluating degradation of the target protein.

Another important type of a proximity-inducing drug, seen by some as the prototypical molecular glue, is cyclosporin, a natural product drug used to prevent organ rejection in transplant patients. Its mechanism of action was uncovered by Stuart Schreiber and others eight years after its approval, with cyclosporin acting as molecular glue between the ubiquitous protein cyclophilin and its target protein calcineurin, leading to inhibition of its phosphatase function and IL2 production (7). Other molecules with similar mechanisms of action followed suit, including the immunosuppressive drug rapamycin (Fig. 1F; ref. 18). In 2012, WarpDrive Bio launched to build a platform around the same concept, harnessing ubiquitous cellular proteins (such as cyclophilin) as presenter proteins that, in complex with drug molecules, form a composite surface that binds featureless, difficult-to-ligand protein targets to block their biological function. Revolution Medicines, which acquired WarpDrive, carried on the research and recently presented promising data on KRAS^G12C and pan-KRAS inhibitors that are now entering clinical development. Associated patent applications suggest that the target ligands are macrocycles containing cyclophilin binding elements linked to a linear portion optimized for target binding.

Further advancement of the proximity-based platforms described above will be highly impactful for cancer drug discovery. However, the requirement for either ligands that were previously identified as starting points, or serendipity, represents a significant limitation when it comes to drugging historically intractable targets such as transcription factors with intrinsically disordered domains. Inspired by and related to Shokat's work on KRAS^G12C, covalent ligand screening using chemoproteomic approaches such as activity-based protein profiling has arisen as a powerful strategy for discovery of covalent ligands and/or ligandable hotspots for proteins that defy conventional drug-discovery efforts (19). Recently, Nomura and colleagues conducted a cysteine-reactive covalent ligand screen to identify compounds that could disrupt the binding of MYC to its DNA consensus sequence (20). From this screen they identified EN4, a covalent ligand that targets Cys-171 of MYC within a predicted intrinsically disordered region, and further optimized this molecule, establishing some early structure–activity relationship. We anticipate that covalent screening using mass spectrometry approaches ideally in the native cellular environment may lead to the identification of ligandable pockets in a multitude of classic “undruggable” targets and could open the door to targeting intrinsically disordered regions of proteins that are frequently observed in transcription factors.

Growing interest and promising data are driving the extension of small-molecule therapeutics beyond targeting proteins to modulating RNA. Venturing into this space is anticipated to expand the landscape of targetable macromolecules by more than an order of magnitude, providing appealing opportunities particularly for highly sought-after cancer targets with profound druggability challenges at the protein level. Although not in the field of oncology, the example of proprotein convertase subtilisin/kexin type 9 (PCSK9), a protein that downregulates low-density lipoprotein receptor and an important target for the treatment of hypercholesterolemia, is illuminating. Costs and convenience of administration spurred interest in identifying PCSK9 small-molecule drugs to complement the existing biological drugs. However, the PCSK9 protein lacks ligandable binding pockets. A phenotypic screen succeeded in identifying a small-molecule inhibitor of PCSK9 secretion, R-IMPP, and systematic investigation of the mechanism of action revealed that R-IMPP binds to ribosomes, causing inhibition of PCSK9 translation (21).

Clinical proof-of-concept for RNA-targeted drugs has already been achieved not only by antisense oligonucleotides but also by small molecules (e.g., linezolid, ribocil, and risdiplam). Together with rapidly growing structural understanding, this suggests that disease-causing RNAs can display appropriate ligand-binding pockets with sufficient information content and uniqueness to enable potent and selective drug binding. Furthermore, various mechanisms of action are theoretically exploitable, including splicing modulation and translational stalling. Although the currently approved RNA-targeting small-molecule drugs were discovered through phenotypic screening approaches, and the role of RNA modulation was brought to light post-discovery, the field is now shifting toward rational design (22). Clinical impact and learnings from these efforts are awaited with great anticipation.

An important mandate in the successful development of targeted therapies is to address the inevitable emergence of treatment resistance. Acquired resistance to targeted single-agent therapeutics is observed in nearly all patients with cancer and is associated with both genetic and nongenetic mechanisms—many of which have been elucidated through preclinical modeling, as well as the molecular analysis of posttreatment biopsies.

The best-understood resistance mechanisms are those associated with preexisting or acquired mutations in the gene encoding the drug target, which confer biophysical resistance to the drug through changes that preserve oncogenic function while disrupting drug action. This has been well documented for a variety of oncogenic targets, including the estrogen and androgen receptors and the ABL, EGFR, BRAF, ALK, MET, and BTK kinases, among others (23, 24). In these cases, second-generation, or even third- or higher-generation, drugs with potency against such resistant mutations were developed for several targets (25); the lineage of EGFR inhibitor generations provides a particularly instructive example. In addition to the development of these next-generation molecules, an intriguing approach to mitigating the effects of resistance mutations in targets is the use of a combination of inhibitors that target the same protein at two different sites, such as with ATP-competitive and allosteric inhibitors of kinases (26). This approach has been demonstrated to be effective preclinically in the context of BCR–ABL, where a combination of an ATP-competitive and allosteric ABL inhibitor dosed concurrently was shown to extend the durability of response in xenograft models of chronic myeloid leukemia and is now being tested in patients in the clinic (27). Similarly, Rodrik-Outmezguine and colleagues exploited the unique juxtaposition of two drug-binding pockets within mTOR to design a bivalent molecule (Rapalink1) that inhibits both wild-type and resistant mutants (28).

Although the targeted protein degraders described above (both PROTAC-type degraders and molecular glues) provide a promising new approach to drugging proteins that have previously been considered intractable, the emergence of mutationally driven resistance to such degraders is not only similarly anticipated, but may in fact develop even more readily. Thus, as in the cases of resistance to more conventional function-blocking drugs, where mutations can emerge that disrupt drug binding, it is similarly expected that mutations in degrader targets that affect ligand binding would also contribute to drug resistance. However, in addition to such mutations of the degrader target itself, loss-of-function mutations in the relevant E3 ligase, or other necessary components of the proteasome machinery, may present additional potential paths to resistance (29). As these degraders are advanced in clinical development, it should become clear whether alternative strategies to mitigate acquired drug resistance associated with such mechanisms will be needed.

An important role for nonmutational mechanisms of drug resistance is also now becoming increasingly recognized. One well-established form of such resistance is driven by “adaptive reprogramming.” A good example of such a mechanism emerged from preclinical studies that demonstrated that the lack of activity of BRAF^V600E inhibitors in colorectal tumors was due to reactivation of MAPK signaling through EGFR activation (30). In this case, the resistance is “innate,” as opposed to being acquired over time following treatment, due to the relatively rapid signal compensation that cancer cells can undergo following acute pathway disruption. Importantly, this preclinical observation prompted a clinical study in which the BRAF^V600E inhibitor encorafenib was combined with the anti-EGFR antibody cetuximab to overcome this innate resistance to BRAF monotherapy. Clinical data from the BEACON study published in 2019 (31) demonstrated that indeed this combination exhibited clinical activity in patients with BRAF^V600E colorectal cancer despite the lack of significant clinical activity for either agent as a monotherapy. This clinical finding has generated newfound enthusiasm for developing rational drug combinations to overcome drug resistance that arises due to adaptive reprogramming.

In a second example, preclinical studies with KRAS^G12C inhibitors predicted that adaptive reprogramming could occur through reactivation of MAPK signaling via upstream RTK signals converging on reactivation of RAS (32). In this context, combination treatment with inhibitors of SHP2, a phosphatase that regulates RAS nucleotide exchange, was shown to exhibit synergistic efficacy with KRAS^G12C inhibitors. These preclinical studies have provided a mechanistic rationale for combining KRAS^G12C inhibitors with SHP2 inhibitors, a combination that has already been reported to show early signs of clinical activity (33).

In addition to pathway reactivation and mutations of the drug target, other nongenetic mechanisms of resistance to targeted therapeutics have been described. For example, mechanisms have been identified in which changes in cell state or lineage yield a substantial phenotypic change in a cancer cell such that it no longer exhibits its original oncogene dependencies (34). The first clinical evidence pointing to the role of such cell plasticity as a mechanism of resistance to targeted therapies came from non–small cell lung cancer transformation into small cell lung cancer upon treatment with EGFR kinase inhibitors (35). Similarly, some metastatic prostate cancers that progress on treatment with antiandrogen therapies appear to have undergone differentiation to a neuroendocrine phenotype that is no longer dependent on androgen signaling (36). More broadly, evidence is accumulating that many epithelial cancers that become resistant to targeted drugs display features of mesenchymal differentiation. Such plasticity-associated resistance mechanisms are likely to involve epigenetic changes, and efforts to target a variety of chromatin regulatory proteins provide a promising avenue toward novel therapeutic strategies to overcome or even prevent such resistance mechanisms. Plasticity-mediated drug resistance is also likely to reflect the action of master transcription factors that drive cell lineage and fate decisions. Although transcription factors remain among the most challenging drug targets, we anticipate that the next frontier of cancer therapeutics will likely include small molecules of this important target class as well, thereby yielding yet additional opportunities to combat drug resistance.

The emergence of novel therapeutic modalities has expanded the scope of new cancer drug approvals in the past decade (Fig. 2). This trend underscores the important role that new technologies as well as fundamental biological insights make in breakthrough therapies for patients with cancer. In this targeted therapy–centric article, we provide our perspective on groundbreaking recent developments in therapeutically targeting challenging cancer targets of major importance. Advancements in technology and the courage to challenge current dogma, coupled with ever-evolving biological insight, have been consistent key elements in enabling this progress, and we are confident that currently unresolved druggability challenges will eventually succumb to our combined creativity and imagination. Among the recently explored drug modalities, chemically induced proximity offers exciting therapeutic potential, and it is a matter of time until existing caveats of the various tactics will be overcome. Targeting RNA offers an entirely new level of disease intervention that we predict will establish itself as an eminent complement to therapeutic targeting at the protein level. Novel drug modalities and drug combinations as well as the ability to target transcription factors will be required to overcome the inevitable and ubiquitous resistance to therapy, enhancing our ability to deliver more durable efficacy to patients. We expect that these approaches, combined with immunotherapy regimens including checkpoint blockade, T cell–dependent bispecific antibodies, and cancer vaccines, will ultimately aid us in achieving transformative efficacy and potentially cures for patients with cancer.

J. Rudolph is an employee/shareholder in Genentech/Roche outside the submitted work. J. Settleman reports that he is an employee and shareholder of Pfizer, Inc. S. Malek is an employee/shareholder in Genentech/Roche outside the submitted work.

We thank Dr. Scott Rosenberg for generating the structure figures for the manuscript and Dr. Kevan Shokat for helpful discussions.