Chromatin regulatory targets for anti-cancer therapeutics

Abstract

Chromatin serves to organize and compact the genome but also functions as a signaling hub for the dynamic regulation of transcriptional programs that control cell type specification. The historical discovery that several pro-differentiation anti-cancer agents target chromatin regulatory enzymes buoyed early interest in developing drugs that modulate chromatin structure and function. Chromatin-based drug discovery has since flourished alongside major advances in discovery chemistry and target selection, producing a rich collection of chemical probes, drugs, and drug candidates targeting chromatin regulatory processes. The substantial growth and maturity of this field over the last several decades provides an opportunity to reflect on the successes and failures associated with translating chromatin regulatory targets into anti-cancer drugs. Taking a target-centric perspective, we discuss the motivation for pursuing specific chromatin regulatory proteins and review the chemistries that enabled small molecule discovery and development. In so doing, we hope to evaluate the strength of these targets, the agents that prosecute them, and the prospects for future efforts in this field.

1. Introduction

Chromatin is a DNA-histone complex that compacts the genome and controls the accessibility of DNA to regulatory factors that act in trans. It is therefore both the template for transcribing DNA into RNA and a signaling hub for gene regulation. The full appreciation that chromatin influences transcription required several decades of research, dating to initial observations that changes in chromatin structure coincide with changes in gene activity.¹ Over 30 years later, when histone acetyltransferase (HAT) and histone deacetylase (HDAC) enzymes were first isolated, cloned, and sequenced, they were found to share the identity of known transcriptional activators and repressors, providing the first direct link between chromatin modifiers and transcription.^2–7 The disciplines of synthetic chemistry, pharmacology, and chemical biology were featured prominently in this seminal work, wherein affinity purification was used to identify the cellular targets of the natural product, trapoxin, an agent known to induce hyperacetylation. The modern field of chemical biology is nicely reflected in this early study and through its foundational impact on the fields of chromatin and chemical biology, it has shaped, considerably, the present state of chromatin-directed drug discovery. Nearing another 30 years since this report, we will consider how chromatin regulation has matured as a target for small-molecule drugs, focusing on the development of anti-cancer therapeutics and the evolution of target selection. We will highlight the foundational discoveries that made specific target classes available to discovery chemistry and provide an assessment on the current outlook for translating these targets and agents into therapeutics. Our treatment of this subject will be mostly limited to compounds targeting chromatin regulatory factors wherein their potential anti-cancer effects are thought to be primarily and directly related to gene transcription. For example, we will not discuss the DNA repair protein PARP1, even though it is strongly associated with transcriptional biology, as the anti-cancer effects of PARP inhibitors in the clinic can be confidently ascribed to the modulation DNA repair [28676700]. Similarly, the inhibition of metabolic proteins that indirectly affect chromatin regulators, like IDH1/2 [cite], will not be discussed.

1.1. Chromatin biology background

Chromatin is a nucleoprotein polymer consisting of DNA wrapped around histones to form repeating nucleosome structures. Generally, a nucleosome is formed by ~1.6 turns of DNA (~147 base pairs) around a core histone octamer, forming a left-handed superhelix.⁸ This octamer is produced by two copies each of the histone proteins H2A, H2B, H3, and H4, but alternative assemblies that feature histone variants or other stoichiometries also exist.⁹ Linker histone H1 binds to DNA between and adjacent to nucleosomes, assisting with local and higher-order chromatin compaction [Cite]. Nucleosomes are post-translationally modified in abundance, most frequently on the unstructured amino (N)-terminal tails of histone subunits but also on their globular carboxy (C)-terminal cores (Figure 1a). These modifications can influence transcriptionally signaling by altering the accessibility of DNA sequences through chromatin compaction or by binding directly to regulatory factors that act in trans.^10–13

Figure 1.

Chromatin regulation and transcriptional signaling. (a) Chromatin regulators and chromatin-dependent transcriptional signaling. (b) Control of enhancer accessibility and activation.

Transcription begins with the recruitment of RNA polymerase (Pol) II and other general transcriptional machinery to gene promoters (located just upstream of the transcription start site), followed by an initial transcription of ~20–100 bp before Pol II reaches a regulated state of stable pausing.^14,15 Release from this promoter-proximal pause is controlled by P-TEFb (positive elongation factor b), a heterodimeric complex of CDK9 (cyclin dependent kinase 9) and cyclin T, which phosphorylates RNA Pol II and other negative regulatory factors, such as DSIF and NELF, allowing for productive transcriptional elongation (Figure 1a).¹⁶ During elongation, Pol II requires accessory factors to traverse through tightly bound nucleosome structures and maintain histone post translational modifications until termination sites are reached and the transcription cycle is completed.^17–19

The production of specialized cell types through organismal development is controlled by the spatiotemporally regulated transcription of specific subsets of genes encoded in the genome.²⁰ Gene-selective transcriptional processes are largely controlled by enhancer sequences located in cis.²¹ These cis-regulatory elements are densely populated with transcription factor (TF)-binding motifs and are most frequently located distal to the gene promoter, often encoded 10 to 100 kb away or further. Enhancer selection is thought to be controlled by a class of pioneering DNA-binding transcription factors (pioneer TFs) that possess a specialized ability to bind their motifs even when embedded in nucleosomes (Figure 1b).^22,23 After binding these occluded motifs within permissive regions of chromatin, pioneering TFs then function to enforce local regions of increased DNA accessibility (Figure 1b). Most TFs can only bind their cognate motifs in the context of free DNA, so this pioneering function allows for additional TFs to bind at enhancers and create an active regulatory element (Figure 1b). Despite being distally encoded, these enhancer elements become physically juxtaposed with the promoter of their target genes in three-dimensional space. This too is highly regulated by chromatin, as the signaling from enhancer-bound TFs to promoter-bound RNA Pol II involves the concerted activity of transcription co-regulators, many of which act directly by binding to chromatin or modifying its structure and function.

Two studies reported in 1996 played an outsized role in establishing that histone modifications could impact transcription.^2,3 Working to clone and sequence genes encoding histone acetyltransferases (HATs), Professor Allis and colleagues discovered that a HAT from Tetrahymena, a unicellular eukaryote, was highly conserved with the yeast transcriptional co-activator, Gcn5.² Contemporaneously, Professor Schreiber and colleagues reported the molecular target of the natural product, trapoxin, which was correctly presumed to be a direct-acting inhibitor of histone deacetylases (HDACs) based on structural similarity between its aliphatic epoxyketone and an acetylated lysine side chain.³ Once isolated through an affinity matrix purification, it was discovered that mammalian HDACs targeted by trapoxin were conserved with the yeast transcriptional co-repressor, Rpd3p. These two studies provided the first direct links between chromatin modifiers and transcriptional regulation, greatly accelerating interest in the modern field of chromatin biology.

Today, we know many distinct classes of chromatin modifiers that dynamically regulate the state of post-translational modifications (PTMs) on histones and other chromatin-bound proteins (Figure 1a).²⁴ Acetylation is joined by methylation, phosphorylation, ubiquitination, ADP ribosylation, and many other modifications of varying abundance. Site-specific histone modifications are often associated with discrete genomic features and transcriptional regulatory states. For example, nucleosomes within the promoter regions of actively transcribed genes are enriched with the trimethylation of lysine 4 of histone H3 (H3K4me3) [cite], active enhancers are enriched with H3K27ac [cite], and repressed enhancers are marked by H3K27me3 [cite]. Histone modifications can serve many different purposes but can often be understood to function through two main mechanisms: (i) altering chromatin structure to make DNA more or less accessible and (ii) serving as binding sites that are selectively recognized by chromatin reader domains (Figure 1a). In many cases, these mechanisms are mutually dependent, as PTMs can have intrinsic effects on protein-DNA interactions but can also serve to recruit regulatory factors that alter chromatin compaction and accessibility by acting in trans.^18,25

Chromatin reader domains are found in scaffolding proteins that recruit other regulatory factors to local regions of the genome and can also function as modules that allosterically regulate enzymatic activities encoded in the same protein or protein complex.²⁶ They are evolved to selectively bind chemical modifications on specific side chains of a protein (often histone tails), enabling combinatorial control in the recruitment of chromatin readers, depending on which modifications are present locally in the genome. For example, lysine acetylation is recognized by bromodomains, DPF (double PHD finger) domains, and YEATS (Yaf9, ENL, AF9, Taf14, Sas5) domains.^27–30 Lysine methylation is bound by chromodomains, WD40 domains, Tudor domains, and PWWP domains, among others.^31–39 Like chromatin modifiers, chromatin reader domains have often been the subject of targeted drug discovery efforts, resulting in many high-quality chemical probes and several investigational new drugs.⁴⁰

1.2. Genetic dependencies and common essential genes

Unlike many of the targets we will discuss in this review, pharmacological tools have been known for HDACs for decades.⁴¹ Chemical tools for HDACs are now widely available, representing an extraordinary range of selectivity and potency profiles, and have contributed enormously to the biological study of chromatin and transcription.⁴² Given their long history, extensive study, and chemical diversity, HDACs provide a particularly instructive example for the considerations that are relevant to targeting chromatin regulatory processes with small molecules. Much of the attention given to HDAC inhibitors can be attributed to the early observation that HDAC inhibition provokes terminal differentiation of cancer cells in tissue culture.^43–47 This work ultimately enabled the development of 4 FDA-approved drugs targeting HDACs for cancer therapy. However, HDAC inhibitors have failed in many clinical trials due to dose-limiting toxicity, which is equally instructive when considering the selection of targets for drug discovery. Outside of a few notable indications, the therapeutic utility of HDAC inhibitors has been limited by on-target toxicity, some of which can be attributed to the broad-spectrum inhibition of many HDAC isoforms by these agents. Another issue with HDAC inhibitors is that several HDAC isoforms are encoded by common essential genes – genes that are required for the survival of most or all cell types. Pharmacological inhibition is often not equivalent to genetic knockout, and there are important examples of common essential proteins that can be targeted with an acceptable therapeutic index, as recently highlighted [cite]. For example, the essential genes HMGCR and PTGS2 are the targets of statins and aspirin, respectively [cite]. However, in many instances, the intrinsic therapeutic window for common essential targets is narrowed by the effects of inhibiting these proteins in normal tissues.⁴⁸

The prevailing thought is that more effective therapeutics can be brought forth by targeting cancer-specific dependencies – genes that are selectively required for the survival of tumor cells versus normal tissues as a result of defined genetic alterations or other changes in cell state.⁴⁹ Thanks to consortia-led genetic screening efforts, which have used pooled CRISPR/Cas9 loss-of-function screens to catalog genetic dependencies in thousands of cancer cell lines (see the cancer dependency map, or DepMap, at depmap.org^50–52), it is now trivial to determine if a potential drug target is encoded by a common essential gene and would therefore likely be required by most normal tissues as well. DepMap uses the algorithm Chronos to transform CRISPR screening datasets into “gene effect” scores that report the impact of gene disruption on cell fitness [cite]. It affixes the median gene effect score of essential genes to −1 and non-essential genes to 0 such that the magnitude of effect can be rapidly interpreted. Since gene effect scores are calculated across thousands of cancer cell lines the distribution of scores can be used to quickly distinguish between common essential genes (those with a distribution centered on –1) and selective dependencies (those with a distribution centered near 0 with a fraction of cell lines with scores approaching –1). For selective dependencies, these score can be integrated with genetic and phenotypic features to identify the underlying cause of the dependency, further probe the biological pathway, and/or nominate biomarkers of drug response.

In chromatin biology, the promise of targeting context-specific cancer dependencies, as opposed to common essential genes, can be seen in the Food and Drug Administration (FDA) approval of EZH2 inhibitors for the treatment of EZH2-mutant lymphomas and SWI/SNF-mutant sarcomas, as EZH2 is not a common essential gene but is a selective dependency in these tumors (Figure 2a,b).⁵³ Likewise, the menin inhibitor revumenib is now approved for the treatment of acute leukemia caused by chromosomal rearrangements involving the MLL gene. Menin, which is not a common essential gene, is selectively required in these and other genetically defined acute leukemias.^54–60 In contrast to the favorable therapeutic window afforded by these targets, common essential gene products can be difficult to drug due to their importance in normal tissues. For example, BET bromodomain inhibitors and CDK9 inhibitors, which target the products of common essential genes (Figure 2c), have shown less promise in clinical trials. We will discuss these fundamental differences in target biology below, beginning our review, part of the thematic Special Issue on Drugging the Undruggable, with a discussion of HDACs, highlighting many of the strengths and challenges associated with targeting chromatin regulatory processes.

Figure 2.

Targeting context-specific cancer dependencies versus common essential genes. (a) Chemical structures of FDA-approved drugs targeting the chromatin regulatory enzymes, EZH2 and menin-MLL. (b) Histogram of DepMap gene effect scores for EZH2 and MEN1 (encodes menin protein) across 1150 cancer cell lines, as determined using pooled CRISPR/Cas9-based gene essentiality screens. Gene effect scores, which were downloaded from the DepMap CRISPR Public 24Q2 dataset hosted at DepMap.org, were calculated using Chronos, an algorithm that affixes the median score of essential genes to −1 and non-essential genes to 0. (c) Gene effect scores for the common essential genes BRD4 and CDK9.

2. Targeting essential proteins

2.1. Histone deacetylases

Zinc-dependent HDACs are categorized into four phylogenetic groups: Class I (HDAC1/2/3/8), Class IIa (HDAC4/5/7/9), Class IIb (HDAC6/10), and Class IV (HDAC11) HDACs [cite]. Class I HDACs are nuclear localized and principally associated with chromatin regulation, whereas class IIa/b enzymes can localize to the cytoplasm where they show diverse functions [cite]. HDAC inhibitors were used to discover the first HDAC enzymes and they have remained essential tools for studying HDAC biology [cite].^3,61 In the 1990s, the study of natural product HDAC inhibitors and the serendipitous development of synthetic HDAC inhibitors followed fascinatingly parallel paths that were eventually linked conceptually by their shared pro-differentiation effects. The first synthetic HDAC inhibitor, SAHA (suberoylanilide hydroxamic acid), originated from efforts to improve on the pro-differentiation activity of dimethyl sulfoxide (DMSO), first described to induce cellular differentiation in the 1970s.⁶² The unlikely path of medicinal chemistry that led from DMSO to SAHA was elegantly recounted in a past perspective by Marks and Breslow.⁴¹ In it, they highlighted the serendipity that contributed to the discovery of SAHA,^62–66 which relied on the iterative synthesis of polar compounds meant to mimic the properties of DMSO (Figure 3a). While both DMSO and SAHA inhibit HDAC enzymes, some intermediate compounds that led to the discovery of SAHA, like hexamethylene bisacetamide (HMBA), are not HDAC inhibitors. Moreover, only after the initial discovery of SAHA as an optimized pro-differentiation agent was it later realized that SAHA functions through the inhibition of HDACs. This realization was triggered by the structural similarity between SAHA and trichostatin A (TSA), a natural product that also harbors a hydroxamic acid and was known to inhibit HDACs (Figure 3a).⁴⁶ SAHA is now approved for the treatment of cutaneous T-cell lymphoma (CTCL) under the generic name, vorinostat.⁶⁷

Figure 3.

Common features of HDAC inhibitors. (a) Chemical structure of prototypical hydroxamic acid HDAC inhibitors, indicating cap, linker, and chelating functionalities on SAHA. (b-d) Chemical structures of prototypical macrocyclic (b), benzamide (c), and 4’-arylbenzamide HDAC inhibitors (d).

2.1.1. Overview of HDAC inhibitors

While the mechanism of HDAC inhibition by DMSO is not well understood, SAHA and TSA function by inserting a hydroxamic acid moiety deep into the HDAC active site where it chelates zinc (Figure 3a). These compounds share a cap-linker-chelate arrangement that combines a Zn-binding group, a surface recognition “cap”, and a linker that traverses through the HDAC active-site channel to join the two, which mimics the insertion of lysine side chains that HDACs have evolved to bind (Figure 3a). The majority of HDAC inhibitors share these features and fall into 3 major structural classes – hydroxamic acids, macrocyclic tetrapeptides, and ortho-aminoanilides (also known as benzamides) (Figure 3b–d). For example, the macrocycle, FK228, features an intramolecular disulfide that is reduced in the cellular environment to reveal a sulfhydryl that interacts with zinc in the HDAC active site.^68,69 The sulfydryl is connected to the cyclictetrapeptide “capping” structure through a hydrocarbon “linker”, conceptually analogous to the attachment of the hydroxamate of SAHA to its anilide capping group. Benzamides, like the prototypical CI-994 and MS-275, bind to zinc through their ortho amine groups, a structure that enabled later agents, the 4’-arylbenzamides, to exploit a deeper internal cavity located beyond the active site zinc (Figure 3c,d).^70–72 There are a few notable exceptions to the cap-linker-chelate logic, including HDAC inhibitors without a metal chelating group, which were discovered through efforts to systematically evaluate the structure-activity relationships (SAR) governing isoform-specific HDAC inhibition by macrocyclic tetrapeptide inhibitors,^73–75 as well as pharmacophores that enable covalent engagement.^76–79

HDACs stand out from other classes of chromatin regulatory proteins for the length of time that small-molecule inhibitors have been available to the research community. The convergence of several serendipitous HDAC inhibitor discoveries not only provided an early indication of the chemical tractability of these enzymes but also created an immediate wealth of structurally distinct chemical tools for biological research. In retrospect, the early discovery of HDAC inhibitors can likely be attributed to the intrinsic ligandability of the HDAC active site and the large effects that HDAC inhibition has on diverse cellular processes. These qualities of HDAC enzymes would, in principle, increase the chances that a pharmacopeia of bioactive small molecules might contain an HDAC inhibitor. Like SAHA, the ortho-aminoanilide class of HDAC inhibitors were also identified serendipitously. The first two compounds of this class, CI-994 and MS-275, were discovered contemporaneously in the 1990s as antiproliferative agents and then later recognized to inhibit HDACs. CI-994 was first used as an anticonvulsant before its antiproliferative activity was recognized in clinical observations and later studied in rodent models of leukemia.^80–82 MS-275 was developed out of a synthetic effort to discover agents that overcome cancer models of multidrug resistance.⁸³ Both were later discovered to inhibit HDACs, providing an orthogonal chelating group to target HDACs.

2.1.2. Common essential HDAC isoforms

In the years following the discovery of SAHA and its approval for CTCL, we have witnessed a broad effort to discover small-molecule modulators of chromatin regulatory proteins. Often, the motivation for developing these agents has been the expectation that they would make effective anti-cancer agents. Several striking examples of clinical successes have resulted from this work – for example, the approval of EZH2 and menin inhibitors as anti-cancer drugs. However, the field is now mature enough to offer several instructive counter examples that failed to yield the clinical advances expected from preclinical experimentation. HDAC inhibitors highlight this duality well. Vorinostat, for example, has been tested in hundreds of Phase 1 and Phase 2 clinical trials but only approved for CTCL.⁴⁸ This can, in part, be attributed to its broad-spectrum inhibition of Zn-dependent HDACs, of which there are 11 encoded in humans.⁴⁸ However, more selective HDAC inhibitors, such as trapoxin, have been known for as long as SAHA and have still failed to produce more widespread anti-cancer drug approvals, suggesting that improving isoform selectivity is not sufficient to eliminate on-target toxicity, the predominant limitation of these HDAC inhibitors. Consistent with the classification of HDACs by primary sequence similarity,⁸⁴ isoform selectivity correlates strongly with the HDAC classes.⁸⁵ However, many class I-selective compounds show less potent inhibition of Class I HDAC8 than Class IIb HDAC6,⁸⁵ which regulates tubulin acetylation.⁸⁶ FK228,^68,69 a natural product prodrug inhibitor of Class I HDACs is the only class-selective HDAC inhibitor to receive FDA approval – for CTCL and briefly for peripheral T-cell lymphoma (PTCL) before it was withdrawn for failing to meet specified endpoints in a confirmatory Phase 3 trial.^87–89 These approvals are joined by two additional broad-spectrum hydroxamate HDAC inhibitors, belinostat (approved for PTCL) and panobinostat (approved for relapsed and refractory multiple myeloma),^90,91 demonstrating that isoform-selective HDAC inhibitors have failed to yield major therapeutic breakthroughs beyond what has also been achieved with broad-spectrum HDAC inhibitors.

Focusing on Class I-selective HDAC inhibitors, much of the difficulty in developing these agents into therapeutics can be attributed to the general essentiality of their targets. Among Class I HDACs, HDAC3 is encoded by a common essential gene (Figure 4a). It is ubiquitously expressed across tissues, indispensable in mice for viability, and required for the fitness of all cancer cell lines.^{48,61,92–96} The latter is considered a strong indication that many untransformed cell types and normal tissues will also depend on the gene for survival. These targets present significant challenges for drug development as on-target toxicity to normal cells can create adverse effects that limit drug efficacy.⁴⁸ 4’-aryl ortho-aminoanilides were previously reported to achieve selectivity for HDAC1/2 over HDAC3 by occupying a unique internal cavity in HDAC1/2 located beyond the resident zinc.⁷¹ This structural feature was originally reported to prevent HDAC3 inhibition and preserve HDAC1/2 inhibition but was recently demonstrated to be an artifact of differences in assay design between HDAC1/2 and HDAC3. Most isoform selectivity profiling has used purified protein preparations, but newly developed Cora-Fluor TR-FRET probes⁹⁷ have enabled the development of ligand displacement assays (FITC-labelled SAHA) that are compatible with crude lysates.⁹⁸ These assays demonstrated that 4’-aryl benzamide inhibitors, like Cpd-60 (also known as Merck60) bind with high affinity to free, but not complex-bound, HDAC1/2/3. In cells, where HDAC1/2/3 are predominantly complex-bound, these compounds show no selectivity between HDAC1/2/3. HDAC1 and HDAC2 are subunits of several large multiprotein complexes, including CoREST, NuRD, SIN3A, and MiDAC. HDAC3 is a component of the NCoR complex. Biochemical assays using isolated recombinant protein have been possible to develop for HDAC1/2, but HDAC3 assays have required the addition of NCoR. In the past, comparison of in vitro assays performed on isolated HDAC1/2 versus NCoR-bound HDAC3 incorrectly suggested that 4’-aryl benzamides, like Cpd-60 (also known as Merck 60), are selective for HDAC1/2 over HDAC3. Furthermore, the kinetics of these HDAC inhibitors are often slow and frequently show state-dependent differences in activity, relying, for example, on protein partners and inositol phosphates.^99–102 Carefully tracking enzyme inhibition using continuous enzyme kinetics has similarly found reports of selectivity within HDAC1/2/3 – namely, the selective inhibition of HDAC3 over HDAC1/2 by RGFP966 – to be artifacts of standard end-point assays.^102,103

Figure 4.

Context-dependent inhibition of essential HDAC proteins. (a) Gene effect scores for the common essential gene, HDAC3. (b) Chemical structure of corin, a dual HDAC and LSD1 inhibitor designed to preferentially inhibit CoREST-bound HDAC enzymes. (c) Chemical structure of the molecular glue UM171, which selectively degrades CoREST by inducing proximity between KBTBD4 and HDAC1/2-CoREST.

Given the results of these studies, which challenge previous reports of selectivity within the HDAC1/2/3 subfamily, it is prudent to consider that many studies reporting the effects of HDAC1/2 inhibition on cancer cells are likely confounded by the inhibition of HDAC3, a common essential protein. Nevertheless, the combined activities of HDAC1 and HDAC2 are also broadly essential,^93,104–115 suggesting that even if HDAC1/2-selective inhibitors are possible to develop, they would likely be broadly toxic as well. HDAC1 and HDAC2 are highly homologous paralogs that incorporate into the same protein complexes (CoREST, NuRD, SIN3A, and MiDAC).⁴² Loss of one gene or protein is compensated by an increase in the other through post-translational changes in stability.^106,107,110 Combined loss of both paralogs is not viable for mouse development and conditional deletion experiments have shown that dual knockout is incompatible with the survival of many diverse cell types. In cancer cells, loss of one paralog is frequently tolerated, whereas loss of both is not. This redundancy creates a selective dependency on one paralog when the other is compromised by cancer-specific genetic deletions.⁹³ In Neuroblastoma, for example, HDAC1 is frequently compromised by deletions of chromosome 1p, creating a reliance on HDAC2 in these cells. Conversely, HDAC2 is deleted in multiple myeloma cells causing a requirement for HDAC1. This genetic relationship, where HDAC2 is only indispensable for cell fitness in the context of HDAC1 deletion (or vice versa), is defined as synthetic lethality – a term borrowed from classical genetics to describe 2 genetic alterations that are tolerated individually but lethal in combination.¹¹⁶ If HDAC1 or HDAC2 could be selectively targeted by small molecules, these genetic deletions could represent vulnerabilities for therapeutic development that would avoid the toxicity of combined HDAC1/2 inhibition to normal cells and potentially widen the therapeutic window of HDAC inhibition. However, pharmacological discrimination between these two paralogs is not currently possible and will likely rely on the development of new pharmacological approaches that go beyond active-site inhibition. For example, several studies have suggested the possibility of selectively degrading certain HDAC isoforms with small molecules that recruit E3 ubiquitin ligase machinery.^117–124 Later, we will discuss several other examples of synthetic lethal relationships between paralogs and emerging pharmacological approaches to exploit them, notably including similar methods that induce paralog-selective targeted protein degradation.

2.1.3. State-specific HDAC inhibitors

In retrospect, the growing appreciation that HDAC1/2/3 are indiscriminately important for cell fitness can likely rationalize much of the past difficulty in translating HDAC inhibitors into effective therapeutics. One interesting approach is to further improve specificity and reduce general toxicity by selectively targeting HDAC1/2 in specific multiprotein complexes. Rodin Therapeutics has disclosed benzamide inhibitors with selectivity for CoREST-bound HDAC1/2.¹²⁵ Here, selectivity was evaluated on substrate assays performed on immunoprecipiated complexes. Recent CoraFluor assays could be used to revisit these compounds, as dissociation of HDAC1/2 from the immunoprecipitated complex might create a pool of free HDAC1/2 that alters the apparent selectivity profile.⁹⁸ If it is too difficult to create orthosteric inhibitors with improved selectivity within Class I isoforms or across protein complexes, bidentate inhibitors have been suggested as a compelling path forward.^126,127 The CoREST inhibitor, Corin, is synthesized by tethering the Class I HDAC inhibitor, entinostat, to an inhibitor of the CoREST demethylase subunit, LSD1 (Figure 4b).¹²⁶ This bivalent compound showed improved activity against melanoma cell lines and decreased toxicity to normal melanocytes, potentially owing to its preferential inhibition of the intact CoREST complex versus other HDAC complexes.¹²⁶ Corin is also highly active against preclinical models of diffuse intrinsic pontine glioma, where a CRISPR screen found HDAC inhibition sensitizes these cancer cells to LSD1 inhibition.¹²⁸ The molecular glue UM171 causes a specific degradation of the CoREST complex by inducing interactions between HDAC1/2 and the ubiquitin ligase complex substrate receptor, KBTBD4 (Figure 4c),^129–133 suggesting that small-molecule degraders may also provide a method to antagonize specific HDAC1/2 complexes. Indeed, HDAC2 degradation in HDAC1-deficient cancer cells selectively destabilizes the NuRD complex compared to other HDAC1/2 complexes.⁹³

2.2. DNA methyltransferases

2.2.1. Nucleoside DNMT inhibitors

The nucleoside analog, azacitidine (5-azacytidine), was approved for the treatment of MDS in 2004.¹³⁴ Today, azacitidine and the related nucleoside analog, decitabine (5-aza-2′-deoxycytidine) (Figure 5a), are used in several additional indications, most notably in combination with the BCL2 inhibitor, venetoclax, for AML. These hypomethylating agents are DNA methyltransferase (DNMT) inhibitors that work through a fascinating mechanism. DNMTs catalyze the methylation of cytosine at the C5 position by first forming a covalent intermediate between an active site cysteine and C6.¹³⁵ The activated C5 position is then methylated by S-adenosylmethionine (SAM) followed by β-elimination to release the enzyme. Azacitidine was first discovered and described as an anti-cancer agent in 1964 and later found to cause the selective hypomethylation of DNA strands into which azacitidine is incorporated.^135,136 When these nucleoside analogs are incorporated into DNA, their 5-aza modifications prevent the release of DNMTs from DNA by β-elimination (Figure 5b), causing global DNA hypomethylation through the trapping of covalent DNMT-nucleic acid intermediates.¹³⁷ Reducing the levels of DNMT protein genetically causes a decrease in cellular sensitivity to azacytidine, demonstrating that the trapped species of DNMT represents a gain-of-function toxicity to cells that is independent of DNMT degradation and DNA hypomethylation.¹³⁷ Unlike decitabine, which is only incorporated into DNA, azacytidine is also incorporated into RNA, which causes additional cellular effects that are likely related to its improved efficacy in preclinical models.^138–141

Figure 5.

Progress in DNMT drug discovery. (a) Chemical structures of FDA-approved cytidine nucleoside analogs that inhibit DNMTs. (b) Mechanism of DNMT inhibition by nucleoside analog inhibitors. (c) Chemical structures of DNMT1-selective non-nucleoside inhibitors. (d) Co-crystal structure (PDB 6X9J) of GSK3830052 (blue) bound to the interface of DNA (beige) and DNMT1 (white). (e) Gene effect scores for the common essential gene DNMT1.

Azacitidine and decitabine inhibit all three enzymes responsible for DNA methylation – DNMT1, DNMT3A, and DNMT3B. DNMT3A/B are the de novo methyltransferases responsible for methylating new sites of DNA, whereas DNMT1 is responsible for propagating methylation marks onto hemi-methylated DNA after replication.^142–155 The therapeutic window of azacitidine and decitabine is likely limited, in part, by nucleoside-specific off-target effects at the high doses that are required to overcome the poor pharmacokinetic properties related to their nucleoside structures. Unfortunately, guadecitabine, a dinucleotide analog of decitabine and deoxyguanosine developed for improved stability,^156–158 failed in Phase 3 clinical trials for MDS and AML. Therefore, much effort has been invested in identifying non-nucleoside DNMT inhibitors that would reduce off-target nucleoside toxicities.

2.2.2. Non-nucleoside DNMT inhibitors

Early examples of non-nucleoside inhibitors include the phthalimide, RG-108, which was discovered through a virtual screen for DNMT1 inhibitors;¹⁵⁹ the quinoline, SGI-1027,¹⁶⁰ which was discovered among a collection of molecules that can bind the minor groove of DNA; and nanaomycin, a quinone antibiotic, which was also identified through virtual screening.¹⁶¹ RG-108, which was originally characterized on bacterial DNMT, was later found to possess minimal activity against mammalian DNMTs.^162,163 The manuscript first reporting SGI-1027 was recently retracted, but other groups have validated the activity of SGI-1027 and its analog, MC3343, in purified DNMT assays.^162,163 Still, this class of quinolines are relatively weak inhibitors of DNMTs and were recently reported to have minimal effects on global DNA methylation levels in AML cells.¹⁶³ Nanaomycin is reported to inhibit DNMT3A selectively, sparing DNMT1 in biochemical assays, and showing decreased DNA methylation in cells.^161,164 However, its effects have not been thoroughly characterized in cells, it has not been tested in vivo, and its anthracycline core presents concerns for tolerability.

A series of dicyanopyridine compounds represent the most substantial advance in the development of non-nucleoside DNMT inhibitors (Figure 5c).¹⁶³ These were discovered in a high-throughput screening campaign to identify enzymatic inhibitors of DNMT1.¹⁶³ This screen identified GW623415X, a 6-amino-3,5-dicyano-4-ethylpyridine scaffold with low micromolar potency against DNMT1. A dimethylamino analog, GSK3484862, showed a nearly 10-fold improvement in activity and no inhibition of other DNMTs up to a concentration of 50 μM, while its enantiomer was inactive up to a concentration of 10 μM. This compound and related analogs, GSK3685032 and GSK3830052, inhibit DNMT1 reversibly and show impressive selectivity against DNMT3A/B. GSK3685032 is more than 2,500-fold selective for DNMT1 over DNMT3A/B. They induce hypomethylation in cells causing anti-proliferative responses in AML models in vitro and in vivo. Extensive characterization of their on-target effects, including genetic manipulations to rule out confounding off-target effects, established this series of compounds as the first potent and specific non-nucleoside DNMT1 inhibitors. Their binding to DNMT1 is context-dependent, showing dramatically improved affinity in the presence of hemimethylated DNA, and co-crystal structures revealed that the compound binds across the protein-DNA interface with the 3,5-dicyanopyridine portion intercalating into DNA (Figure 5d). This intercalation is dependent on the presence of DNMT1, as the compound does not bind to DNA nonspecifically. Not only is this intercalation important for compound binding, but it also displaces the active-site loop of DNMT1 from intercalating into DNA and forming a closed, active conformation of the enzyme. This fascinating mechanism of context-dependent DNA intercalation establishes that DNMT enzymes can be inhibited by non-nucleoside small molecules and may provide future advantages for drug development. However, DNMT1 is a common essential gene (Figure 5e), and it remains unclear what challenges non-nucleoside inhibitors might encounter in human clinical investigation related to on-target toxicity.

2.3. BRD4

2.3.1. Discovery of BET bromodomain inhibitors

BET (bromodomains and extraterminal domains) family proteins (BRD2, BRD3, BRD4, and BRDT) are chromatin readers that bind to acetylated lysine side chains through a tandem pair of N-terminal bromodomains. In 2010, two papers reported the discovery of JQ1 and I-BET (Figure 6a), triazolodiazepines that bind competitively with chromatin to the acetyllysine-binding pocket of BET bromodomains.^165,166 These reports provided the first demonstration that chromatin reader domains are pharmacologically tractable targets. The origins of both JQ1 and I-BET can be traced back to phenotypic screens for compounds with anti-inflammatory activity. Therefore, like HDAC and DNMT inhibitors, the biological activity of these compounds was known before their cellular targets were described. BET bromodomain inhibitors were quickly shown to possess extraordinary anti-cancer activity in preclinical models of diverse cancer types, leading to many clinical development programs and clinical trials. On-target toxicity has complicated therapeutic translation but BET bromodomain inhibitors remain under active clinical investigation.

Figure 6.

Prototypical BET bromodomain inhibitors. (a) Chemical structures of JQ1 and I-BET, the first reported BET bromodomain inhibitors. The (R)-JQ1 is an inactive stereoisomer that does not bind BET bromodomains. (b) Chemical structures of prototypical BET bromodomain inhibitors with alternative acetyl-lysine mimetic groups

The thienodiazepine BET bromodomain inhibitor, JQ1, was developed by Professors Bradner and Knapp and studied in midline carcinomas that are driven by oncogenic rearrangements of the BET protein, BRD4.¹⁶⁵ JQ1 was synthesized after patent filings from Mitsubishi Pharmaceuticals disclosed a connection between thienodiazepines (originally identified from anti-inflammatory phenotypic screens) and BET bromodomain proteins.¹⁶⁵ JQ1 is decorated with solvent-exposed t-butylester and 4-chlorophenyl pendant groups, in addition to a triazolyl moiety that inserts deep into the bromodomain pocket and mimics acetyllysine interactions with a key conserved asparagine. Biochemical profiling showed that this structure binds with 50 to 90 nM affinity to both bromodomain 1 (BD1) and BD2 on all BET isoforms (BRD2/3/4/T) and no significant binding is observed to non-BET bromodomains.

Further establishing the tractability of this class, Professor Tarakhovsky and colleagues at GSK contemporaneously reported the BET bromodomain inhibitor, I-BET.¹⁶⁶ This compound was discovered by GlaxoSmithKline through a phenotypic screen for ApoA1 reporter gene activation and studied in models of inflammatory gene signaling.^166,167 Here, a benzodiazepine compound, GW841819X, was discovered by phenotypic screening and an analog with similar activity, I-BET, was then found to target BET proteins through ligand affinity chromatography. Sharing the triazolodiazepine core and 4-chloro pendant group with JQ1, it featured a fused benzene instead of a thiophene and an aminoethyl amide instead of a t-butylester. It binds competitively with chromatin to the acetyllysine-binding pocket in the same mode as JQ1, contacting the conserved asparagine and exhibiting similar 50 to 60 nM affinity.

As shown in these index studies, both JQ1 and I-BET inhibit BET bromodomain binding to chromatin, which causes cell cycle arrest and differentiation in BRD4-NUT-driven NUT midline carcinoma (NMC) and anti-inflammatory effects in activated macrophages.^165,166 The anti-inflammatory effects of I-BET were attributed to the suppression of pro-inflammatory gene regulatory signaling. By chromatin immunoprecipitation and DNA sequencing (ChIP-seq) studies, I-BET could be shown to evict BET protein from chromatin, leading to selective downregulation of stimulus-responsive inflammatory genes. At these genes, transcriptional elongation by RNA Pol II was suppressed, consistent with prior studies connecting BRD4 to the activity of the elongation factor, P-TEFb.^168,169 These class-selective effects would later be shown many times over in diverse model systems with a major focus on the potential to induce anti-cancer effects.¹⁷⁰ In NMCs, JQ1 was shown to elicit a potent pro-differentiation effect and G1 cell growth arrest in tissue culture, ultimately prolonging survival of tumor-bearing mice.¹⁶⁵

2.3.2. Clinical evaluation of BET bromodomain inhibitors

Interactions between BRD4 and P-TEFb quickly inspired the study of BET bromodomain inhibitors in MYC-driven tumors, as previous studies had suggested an important role for P-TEFb in maintaining MYC expression.^168,169,171 BET bromodomain inhibition was accordingly found to possess anti-proliferative effects against several MYC-dependent hematological malignancies.^172–175 In the first report, BRD4 was identified in a pooled genetic knockdown screen that aimed to identify essential genes in acute myeloid leukemia (AML) cells from among ~250 chromatin regulatory proteins.¹⁷² Both genetic knockdown of BRD4 and BET bromodomain inhibition by JQ1 showed anti-leukemia activity in vivo as a result of induced myeloid differentiation and growth arrest,¹⁷² not unlike the pro-differentiation effects observed in NMC.¹⁶⁵ Transcriptional downregulation of MYC, together with eviction of BRD4 from the MYC enhancer were also observed in multiple lineage leukemia (MLL) where oncogenesis is driven by MLL fusion oncoproteins, in multiple myeloma (MM) cells where elevated MYC expression is driven by chromosomal translocations that connect the IgH enhancer to MYC promoter regulation, and in a variety of other leukemia and lymphoma cell lines.^172–178 Here, MYC was shown to be among the most quickly and severely downregulated genes in response to BET bromodomain inhibition.

BET bromodomain inhibition has since been studied in preclinical models of countless cancers, resulting in many clinical trials in oncology. Results of the initial trials evaluating the thienodiazepine OTX015 (birabresib) in leukemia, lymphoma, multiple myeloma, lung, prostate, and NMC have now been reported.^179–182 In general, thrombocytopenia and adverse gastrointestinal events lead to dose-limiting on-target toxicities. Drug holidays have been explored to address these toxicities, especially thrombocyteopenia, which is reversible upon drug cessation, but BET inhibition is now widely considered to be poorly tolerated. BET bromodomain inhibitors have advanced to a Phase 3 clinical trial in one instance, which evaluated CPI-0610 (pelabresib) in combination with the JAK inhibitor ruxolitinib for myelofibrosis (NCT04603495).¹⁸³ Preclinical studies have previously demonstrated that combined JAK and BET bromodomain inhibition is effective in animal models of myeloproliferative neoplasms (MPN), interestingly related to the anti-inflammatory effects of BET bromodomain inhibition in suppressing NF-kB transcriptional networks.¹⁸⁴ This recalls the original discovery of BET bromodomain inhibitors as anti-inflammatory agents and past characterizations of BET-dependent NF-kB signaling.^166,185 Conference presentations have reported that the Phase 3 trial of pelabresib and ruxolitinib met its primary endpoint, providing renewed hope for BET bromodomain inhibitors, but these agents have otherwise not shown promising therapeutic profiles in the clinic.

2.3.3. Chemical probe studies of BRD4

BET bromodomain inhibitors have made for exceptionally useful chemical probes of transcription and chromatin biology, which has assisted in illuminating the complications associated with translating BET bromodomain inhibitors into the clinic. When using pharmacological tools to study complex biological systems, assigning a phenotype to the on-target or off-target effects of a compound can be exceedingly difficult.¹⁸⁶ Several reviews and perspectives have previously crystalized the experimental strategies to definitively link biological phenotypes to on-target pharmacological activities.^186–190 First is to identify a drug-resistant mutant allele that blocks drug binding and suppresses the biological effects of the compound, second is to compare physiochemically matched active and inactive enantiomers, when possible, and third is to use multiple structurally distinct compounds that share the same intended target, as they are likely to have different off-target effects. Evaluating these three criteria for BET bromodomain inhibitors, resistance mutations that disrupt drug binding to BRD4 while also preserving BRD4 function have not yet been discovered.^191–194 However, the study of BET bromodomain proteins has benefited from the existence of stereochemistry-selective inhibitors (i.e. the existence of enantiomeric inactive controls) and several classes of chemical scaffolds that inhibit the same sites on BET proteins.⁴⁰ Indeed, an attractive feature of JQ1 that has contributed to its widespread use as a chemical probe for BET bromodomain proteins is the availability of an inactive enantiomeric control. The t-butyl ester substituent on JQ1 creates a chiral center on the diazepine core, resulting in a physiochemically matched inactive control (Figure 6a).¹⁶⁵ The active enantiomer, (S)-JQ1, and inactive control, (R)-JQ1, have been used extensively to help assign cellular phenotypes as being caused by the inhibition of BET bromodomains and not a cellular off-target. In addition to the diazepine BET bromodomain inhibitors, chemical probes with several distinct pharmacophores have been reported, as previously reviewed,¹⁷⁰ notably including early examples like the dimethylisoxazole (e.g. I-BET151) and dihydroquinazolinone (e.g. PFI-1) inhibitors (Figure 6b).^{173,195–197}

The rigorous use of structurally distinct chemical probes and paired enantiomeric controls has greatly assisted the study of BET proteins in cancer. Indeed, the preferential downregulation of MYC has been observed repeatedly in diverse cancer types,^{176–178,191,198–201} including in a pre-registered replication study,²⁰² and can be extended to the MYC paralog MYCN.²⁰³ However, this preferential downregulation of MYC is accompanied by a more general downregulation of lineage-specifying TFs, which has been rationalized by the disproportionate abundance of BRD4 at enhancers controlling the expression of these genes.^177,185 In addition to regulating the expression of lineage-driving TFs, BRD4 also binds to acetylated TFs, allowing for BRD4-dependent signaling between enhancer-bound acetylated TFs and RNA Pol II.^204,205 Altogether, the effects of BET bromodomain inhibition converge on the preferential suppression of transcriptional programs that are important for cell state and identity (Figure 7a), which is congruent with their pro-differentiation effects across cancer models.

Figure 7.

BRD4 is a global regulator of transcriptional activation. (a) BET bromodomain inhibition causes gene-selective inhibition of transcription. (b) Schematic depiction of PROTACs, heterobifunctional compounds that induce targeted protein degradation. (c) Chemical structure of the BET PROTAC, dBET6. (d) BET degradation causes global inhibition of transcriptional activity. (e) Selective degradation of BRD4 using the auxin-inducible degron (AID) system, which relies on auxin-induced proximity between a genetically fused AID tag and an ectopically expressed E3 substrate receptor, Tir1. BRD4 degradation causes global inhibition of transcriptional activity.

Despite these class-selective transcriptional consequences of BET bromodomain inhibition, it is now clear that BRD4 functions more broadly as an essential regulator of transcriptional elongation at most, if not all, actively transcribed genes.^206,207 This insight was made possible by the discovery of chemical tools that induce targeted protein degradation of BRD4 rather than inhibiting its isolated bromodomain functions. For over 2 decades, the concept of inducing targeted protein degradation has been pursued through the synthesis of PROTACs (proteolysis targeting chimeras) – bifunctional compounds that independently bind to an E3 ubiquitin ligase and a target protein of interest.²⁰⁸ These compounds can induce ubiquitin transfer onto non-native substrates through chemically enforced proximity to the E3 (Figure 7b). After the drug thalidomide was discovered to bind cereblon (CRBN),²⁰⁹ a cullin-RING E3 ligase complex substrate receptor, Bradner and colleagues developed the first CRBN-based PROTACs, selecting BET proteins to demonstrate proof-of-concept.²¹⁰ Attaching thalidomide to JQ1 through a polymethylene linker resulted in a bifunctional small molecule, dBET1, which elicits potent and selective degradation of BET proteins in living cells and animals.²¹⁰ An optimized molecule, dBET6 (Figure 7c), which featured a longer polymethylene linker was subsequently reported and used to investigate BET-dependent transcriptional regulation.²⁰⁶

Using kinetically resolved and unbiased genomic measurements of transcriptional activity following dBET6 treatment, BET degradation was found to universally halt the transcription of all genes by preventing the transition of RNA Pol II from promoter-proximal pausing to productive transcriptional elongation (Figure 7d).²⁰⁶ A later study, which used a genetically encoded system to enable selective degradation of BRD4 (sparing BRD2/3 from degradation), found that the global inhibition of transcriptional elongation by BET degraders could be specifically attributed to the activity of BRD4 (Figure 7e).²⁰⁷ Several direct-acting PROTACs have now also been reported that selectively degrade BRD4 over other BET proteins.^211,212 A notable early example, AT1, was developed from JQ1 and a small-molecule ligand for the von Hippel-Lindau E3 substrate receptor (VHL) using structure-guided optimization.²¹² Small-molecule ligands for VHL that mimic a key hydroxyproline degron were developed by iterative optimization of peptidomimetic ligands.^213–216 This afforded VHL-based small-molecule PROTACs capable of degrading diverse targets, including BRD4, which was reported contemporaneously with the first CRBN-based BRD4 degraders.^211,217 It is now well-established that multi-specific ligands can be fashioned into heterobifunctional PROTACs to degrade a subset of its targets.^{211,212,218–224} These compounds rely on the determinants of ternary complex formation rather than ligand affinity to achieve improved selectivity, as shown by BRD4-selective PROTACs that spare other BET proteins from degradation.^211,212,220

The discovery that BRD4 is universally required for transcription likely explains the finding that BRD4 is a common essential gene.⁴⁸ It might also explain why so many diverse preclinical models of cancer showed striking responses to BET bromodomain inhibition whereas clinical trials have not been nearly as promising. As discussed by Professor Sellers and colleagues, targeting common essential genes as an anti-cancer strategy presents major difficulties with on-target toxicity, limiting the therapeutic window in human patients.⁴⁸ Due to the likelihood that a common essential gene is important for an indispensable cell type, tissue, or function, systemic exposure to drugs targeting essential genes can create unacceptable toxicities. The global suppression of transcription by BRD4 degradation likely explains why the BET degraders that were once disclosed by industry groups have not been pursued as clinical programs.²²⁵ However, it is important to consider that BET bromodomain inhibitors specifically target the interaction between bromodomains and acetyllysine and do not necessarily disrupt all associated functions of BRD4. Indeed, BRD4 degradation elicits a much more severe effect on transcription and cell fitness than BET bromodomain inhibition,²⁰⁶ the former being much closer in nature, and in consequence, to the genetic loss-of-function screens that led to the annotation of BRD4 as a common essential gene. BET bromodomain inhibition, in contrast, elicits a class-selective suppression of a subset of genes.^177,206,207 Why BET bromodomain inhibition differs from BET degradation might be rationalized by the observation that bromodomain inhibition preferentially evicts BRD4 from enhancer regions where it is most densely localized.¹⁷⁷ Therefore, BET bromodomain inhibition preferentially impacts the genes targeted by these enhancers without eliciting the same global transcriptional suppression as BET degradation. Despite these differences, BET bromodomain inhibitors have still not led to the therapeutic breakthroughs that were originally signaled by early preclinical studies, further highlighting the challenges of targeting the protein products of common essential genes.⁴⁸

Chemically induced degradation has demonstrated that BRD4 functions as a global transcriptional regulator by controlling the kinase activity of CDK9,^206,207 the kinase component of P-TEFb. RNA Pol II is precisely phosphorylated at specific side chains within a heptapeptide repeat (the YSPTSPS sequence repeated 52 times in mammalian cells) on its C-terminal domain.²²⁶ CDK9 is responsible for phosphorylating RNA Pol II on Ser2 of this repeat, which—along with the CDK9-mediated phosphorylation of DSIF and NELF— is responsible for releasing RNA Pol II from promoter-proximal transcriptional pausing.¹⁶ BRD4 was previously thought to regulate transcription by recruiting CDK9 to chromatin,^168,169 but acute BRD4 degradation has shown that it is required for the activity, not recruitment, of CDK9.^206,207 Ultimately, CDK9 inhibition and BET degradation provoke substantively identical transcriptional responses.²⁰⁶

In general, cyclin dependent kinases that regulate transcription have proven to be fruitful targets for discovery chemistry (Figure 8a). Chemical probes and drug candidates targeting CDK7, CDK8, CDK9, CDK12, and CDK13 are also now well represented,^227–235 as was recently reviewed.^236,237 Inhibition of CDK9 by the ATP analog, DRB (5,6-dichloro-1-b-D-ribofuranosylbenzimidazole), was critical to the discovery of factors controlling pause release and transcriptional elongation.^238–251 The more cellularly active CDK inhibitor, flavopiridol, has been similarly useful for studying P-TEFb in cells, revealing a global requirement for P-TEFb across all transcribed genes (Figure 8b).^252–256 Most recently, compounds with improved selectivity for CDK9 over other CDKs, such as the bipyridine inhibitors, iCDK9 and NVP-2, or the pyrazolopyrimidine, KB-0742, have enabled unambiguous assignment of these effects to CDK9 (Figure 8c).^{206,221,257,258} Following on the clinical failures of non-selective CDK9 inhibitors, one CDK9-selective inhibitor, AZD4573, was developed for clinical use against hematological malignancies.²⁵⁹ Here, the logic was to target the short-lived transcript of the antiapoptotic gene, MCL1, by using selective CDK9 inhibition to transiently inhibit transcriptional activity globally, such that transcripts with short half-lives would be preferentially depleted. This strategy was developed with the goal of overcoming the challenging safety profiles seen with previous CDK9 inhibitors.²⁵⁹ However, after clinical trials reached as far as Phase 2, AstraZeneca suspended clinical development of AZD4573, citing strategic portfolio prioritization. Kronos, the developers of KB-0742, have terminated clinical investigation of their selective CDK9 inhibitor due to a lack of safety and futility (NCT04718675). Similar to BET bromodomain degraders, these developments present concerns that the global effects of CDK9 inhibition on transcription will prohibit therapeutic dosing in humans.

Figure 8.

CDK9 inhibitors and transcription elongation. (a) Early chemical tools to study transcription elongation and CDK9 function. (b) Inhibition of CDK9 inhibits transcriptional activity globally. (c) Chemical structures of selective CDK9 inhibitors.

2.3.4. Domain-selective BET bromodomain inhibitors

Each BET protein contains two bromodomains located in tandem near the N-terminus. Most BET bromodomain inhibitors (e.g. JQ1, I-BET, pelabresib) inhibit both bromodomain 1 (BD1) and BD2. In recent years, several groups in academia and industry have sought to mitigate the toxicity associated with BET bromodomain inhibition by pursuing selectivity between BD1 and BD2, which has resulted in compounds that selectively inhibit BD1 of BRD2/3/4 or BD2 of BRD2/3/4 (termed BD1-selective BET bromodomain inhibitors and BD2-selective BET bromodomain inhibitors, respectively).^260–286 BD1 and BD2 have been reported to mediate different biological activities, with BD1 generally suggested to be the primary mediator of binding to acetylated histones and BD2 thought to bind acetylated lysines on other proteins that are present on chromatin.^287–292 Each BD1 domain is more closely related to the BD1 domains from other proteins than they are to the BD2 domain in the same protein, which has enabled the development of compounds that inhibit all BET proteins but discriminate between BD1 and BD2. This was first indicated as a possibility by resveratrol analogs in 2013,²⁶¹ but more firmly established by a series of papers publishing high-quality BD1-selective and BD2-selective chemical probes in 2020.^268–272 ABBV-744, a BD2-selective chemical probe, contains a pyrrolopyridone core and was developed through structure-guided optimization to exploit side chain differences present on loops at the periphery of the ligand-binding site (Figure 9a).^268,269 In BRD4, for example, Asp144 and Ile146 in BD1 are replaced by His437 and Val439 in BD2 (Figure 9b). The team at AbbVie that developed ABBV-744 recognized modest selectivity (7-fold) for BD2 in one compound from a pyrrolopyridone series that had previously led to the development of the dual BD1/2 BET inhibitor, ABBV-075 (mivebresib) (Figure 9b). This compound features a primary amide group on the C2 position of the pyrrolopyridone core, which is unsubstituted in ABBV-075 but positioned in close proximity to the Asp144/His437 and Ile146/Val439 side chains of BRD4 BD1/BD2. Expansion of the amide to an ethyl amide further improved selectivity to 17-fold for BD2, rationalized in co-crystal structures by the inward projection of BD2-His437 into the periphery of the binding pocket, compared to BD1-Asp144 facing outward (Figure 9b). Optimization of the 2,4-difluorophenoxy group, which is positioned proximally to Ile146/Val439, led to >100-fold selectivity. Further optimization of pharmacokinetic properties led to ABBV-744, which maintained single-digit nanomolar potency on BD2 (K_i = 1.6 nM) and over 300-fold selectivity against BD1 (520 nM). Overall, this series of ligands showed far improved shape complementarity for BD2 than BD1, justifying their impressive selectivity.

Figure 9.

BD1-selective and BD2-selective BET bromodomain inhibitors. (a) Discovery of ABBV-774, a BD2-selective BET bromodomain inhibitor. (b) Co-crystal structures of BRD4 BD1 (PDB 6VIW) and BRD4 BD2 (PDB 6VIX) with a BD2-selective inhibitor from the ABBV-744 series. (c) Discovery of BD2-selective acetamide BET bromodomain inhibitor series. (d) Chemical structure of iBET-BD2, an optimized BD2-selective BET bromodomain inhibitor. (e) Discovery of iBET-BD1, a BD1-selective BET bromodomain inhibitor. (f) Effects of dual BD1/2 BET bromodomain inhibition versus BD2-selective BET bromodomain inhibition.

Contemporaneously, GSK reported both BD1-selective and BD2-selective BET bromodomain inhibitors, termed iBET-BD1 (GSK778) and iBET-BD2 (GSK046).^270–272 In comparison to the discovery of ABBV-744, which traced back to a pre-existing ligand series, GSK discovered the starting point for iBET-BD2 from a high-throughput screen of over 2 million compounds. Using a TR-FRET-based screen, two BRD4 constructs were employed, each containing both tandem bromodomains, but with either BD1 or BD2 inactivated by Tyr to Ala mutations (Figure 9c). An acetamide screening hit showed ~10-fold selectivity for BD2 over BD1 across all BET family members (Figure 9c). Co-crystal structures with BRD4 BD1 and BRD2 BD2 revealed the importance of His433 in BRD2 BD2 (analogous to His437 in BRD4 BD2), which showed conformational flexibility of rotamers alternately facing inward or outward. In contrast, Asp144 of BRD4 BD1 was conformationally restrained and pointed inward. Further structure-informed improvements led to iBET-BD2, which is moderately less potent than ABBV-744, but similarly selective (K_D,BD2 = 30 nM, >300-fold selective) (Figure 9d). iBET-BD1 was developed from a previously disclosed BD1/2 inhibitor – the dimethylisoxazole I-BET151 – using structure-guided efforts to capture interactions with Asp144 through a local water network that is not present in BD2, ultimately achieving 19 nM affinity for BD1 and >100-fold selectivity against BD2 (Figure 9e). While the BD1 selectivity of iBET-BD1 was less impressive compared to iBET-BD2, further selectivity for BD1 is possible. A structurally distinct compound with >1000-fold selectivity for BD1 over BD2, GSK789, was developed in a separate effort following observations that the ATAD2 bromodomain inhibitor, GSK8814, showed >50-fold selectivity for BRD4 BD1 over BD2.²⁹³

Comparison of the BD2-selective inhibitor, ABBV-744, to the dual BD1/2 inhibitor, ABBV-075, revealed a striking difference in living systems (Figure 9f).²⁶⁸ While ABBV-075 was shown to be broadly toxic across cell lines of diverse cancer types, ABBV-744 was selectively cytotoxic, preferentially impacting AML and prostate cancer cell lines. This selectivity was observed alongside a muted impact on global transcriptional control compared to dual inhibition and translated to a more tolerated profile in vivo that was associated with reduced thrombocytopenia and GI toxicity in mice. Promisingly, ABBV-744 has progressed into early-stage clinical trials, including a study in combination with ruxolitinib or navitoclax for myelofibrosis (NCT04454658), the indication where BET bromodomain inhibitors have performed best in the clinic.

In another study, iBET-BD2 was reported to be ineffective as an anti-cancer agent both in vitro and in vivo, whereas iBET-BD1 phenocopied dual inhibition. These data further establish the importance of BD1 to basal BRD4 function, but it is unclear why iBET-BD2 failed to elicit similar effects as ABBV-744 in preclinical cancer models. One possibility is that this reflects the weaker potency of iBET-BD2 compared to ABBV-744. The exceptionally quick maturation of BD1- and BD2-selective chemical probes and drug candidates will allow for unambiguous testing of domain-selective BET inhibitors in preclinical studies and clinical trials.

2.3.5. Repurposing BET bromodomain inhibitors to induce transcription

Beyond their utility as chemical probes of BET family proteins, JQ1 and other prototypical BET bromodomain inhibitors have also emerged as exceptionally useful tools for the development of heterobifunctional chemical inducers of proximity (CIPs). Several attributes have made JQ1 particularly attractive for proof-of-concept studies that seek to establish new forms of chemically induced proximity. First, it is readily available from commercial suppliers and is relatively inexpensive. Second, the tert-butylester group of JQ1 is solvent exposed when bound to bromodomains and therefore provides a conveniently accessible protected carboxylic acid that can be used to attach linkers without obstructing bromodomain engagement.²⁹⁴ Finally, the inactive enantiomer of JQ1 provides a valuable negative control to determine whether the cellular effects of a JQ1-based heterobifunctional CIP require on-target engagement of BET bromodomains. These qualities likely contributed to JQ1 being used in each of the first 3 proof-of-concept studies showing that CRBN ligands can be used to construct heterobifunctional PROTACs^210,211,225 and it has continued to be overrepresented in the development of new methods for inducing targeted protein degradation.^{120,211,212,295–315} Beyond targeted protein degradation, JQ1 has also helped to establish several other forms of chemically induced proximity, notably including chemically induced phosphorylation (PHICS),³¹⁶ O-GlcNAcylation (OGTAC),³¹⁷ nuclear import (NICE),³¹⁸ and regulated induced proximity targeting chimeras (RIPTACs).³¹⁹

A particularly inventive use of JQ1 is embodied in recent reports of chemical inducers of proximity that use JQ1 to recruit the powerful transcriptional co-activator functions of BRD4 to silenced genes, conditionally reactivating their expression (Figure 10).^318,320,321 The primary challenge in developing chemical inducers of proximity that selectively regulate gene transcription is the identification of small molecules that can localize to specific gene regulatory elements. This was first addressed using genetically encoded systems – for example, using chemically induced proximity to recruit a tagged transcriptional regulatory proteins to an engineered locus.^322–329 Later methods evoked pyrrole- and imidazole-based polyamides, which bind directly to DNA,³³⁰ to create heterobifunctional compounds that circumvent the need for locus engineering.^{320,331–335} While the limited specificity of polyamides has presented challenges for this approach, it has been particularly effective in models of Friedreich’s Ataxia,^320,336 where the pathophysiological silencing of FXN transcription is caused by the expansion of repressive GAA microsatellite repeats near the FXN promoter.^337,338 A heterobifunctional molecule attaching JQ1 to a GAA-targeted polyamide resulted in a remarkably selective activator of FXN transcription, called synthetic transcription elongation factor 1 (Syn-TEF1) (Figure 10a).³²⁰ By localizing to the GAA repeats near FXN, SynTEF1 can recruit BRD4 to the FXN promoter, which activates paused RNA Pol II for productive transcriptional elongation. Domain-selective BET bromodomain inhibitors have also been used in place of JQ1 to synthesize new FXN-activating SynTEFs, revealing a dramatic preference for BD2-based recruitment over BD1, which was suggested to reflect differences in the evolved functions of these domains.³³⁶

Figure 10.

Chemical inducers of proximity that control the assembly of transcriptional regulatory complexes. (a) SynTEF1 binds GAA repeats near the silenced FXN promoter through its polyamide DNA-binding moiety (blue) and recruits BRD4 to activate FXN transcription. (b) CEM87 binds FKBP12-tagged dCas9 through its FK506 moiety (blue), localizing to dCas9-bound regions of the genome and recruiting BRD4 to activate gene expression in a locus-specific manner. (c) TCIP1 binds to the transcriptional repressor BCL6 (blue) and recruits BRD4 to active BCL6-repressed gene targets. (d) CDK-TCIP1 recruits CDK9 (orange) to BCL6 (blue) through its kinase inhibitor moiety (orange), producing a locally high effective molarity of CDK9 near BCL6-bound promoters to active expression of BCL6-repressed gene targets.

Tag-specific heterobifunctional small molecules have been used to enable flexible targeting of BRD4 to specific genomic locations bound by Cas9. This method requires expression of a nuclease “dead” Cas9 mutant (dCas9) fused to FKBP12 and a locus-specific small guide RNA (sgRNA) that navigates dCas9 to a desired region of the genome. A heterobifunctional fusion of JQ1 and FK506 can then be used to recruit BRD4 to dCas9-bound locations of the genome through its FKBP12 tag, allowing for locus-specific and ligand-dependent activation of gene expression (Figure 10b). JQ1 has even been conjugated directly onto dCas9 using split intein chemistry, allowing for the programmable localization of JQ1 on the genome by cellular delivery of semisynthetic dCas9-JQ1 conjugates.³³⁹ While the use of dCas9 to direct small molecules to specific genomic locations has enabled elegant proof-of-concept studies in cells, the challenge of delivering the large dCas9 construct in vivo present a substantial hurdle for clinical translation. More recent designs have used recombinant zinc finger (ZF) DNA-binding domains in place of dCas9 such that delivery is possible by adeno-associated virus (AAV), yet these have only been applied in tissue culture to date [cite].^340,341

The repressive transcription factor BCL6, has been coopted to enable locus-specific delivery of JQ1 without the need for any genetic engineering or protein delivery.³²¹ In this elegant approach, BCL6 inhibitors were repurposed to synthesize heterobifunctional transcriptional chemical inducers of proximity (TCIP) compounds that can recruit BRD4 to BCL6, resulting in small molecules that induce remarkably potent anti-lymphoma activity by activating genes that are ordinarily repressed by BCL6.³²¹ The BCL6 transcription factor is an essential driver of diffuse large B-cell lymphoma (DLBCL) and functions by interacting with transcriptional co-repressor complexes to suppress BCL6 target genes, as previously reviewed.³⁴² Small molecules that bind to the BCL6 BTB domain and inhibit its interactions with co-repressors have previously been discovered using a high-throughput fluorescence polarization screen of nearly 2 million compounds, which measured the displacement of a co-repressor-derived peptide from the BTB domain.³⁴³ From an initial 4-amino-5-chloropyrimidine screening hit, which showed 20 μM affinity for the BTB domain, structure-guided optimization provided dramatic improvements in binding affinity, affording two notable chemical probes, BI-3802 and BI-3812. Both compounds inhibit the interaction between the BTB domain and co-repressor peptide with single-digit nanomolar potency and possess only slight differences in their overall chemical structures. However, different substituents on a terminal piperidine group afford BI-3802 the ability to induce potent and selective BCL6 degradation.³⁴³ Later studies determined that BCL6 degradation results from chemically induced polymerization of BCL6, a proteoform that is recognized and ubiquitinated by the SIAH1 E3 ubiquitin ligase.³⁴⁴

Compared to BCL6 inhibition by BI-3812, BCL6 degradation by BI-3802 produces more substantial de-repression of BCL6 target genes and causes anti-proliferative responses in DLBCL cell lines that do not respond to the inhibitor. However, even stronger activation of BCL6 target genes can be elicited by compounds that recruit BRD4 to BCL6. TCIP1, a chimera of BI-3812 and JQ1, recruits BRD4 to BCL6, activating the expression of BCL6 target genes beyond the simple de-repression that is achieved by BCL6 degraders.³²¹ This heterobifunctional compound induces cooperative BRD4-BCL6 interactions, recruiting BRD4 to BCL6 target genes without affecting the normal bulk distribution of BRD4 elsewhere in the genome (Figure 10c). The recruitment of BRD4 causes an immediate increase in transcriptional elongation by RNA Pol II at BCL6 target genes, including important pro-apoptotic genes, ultimately causing the death of DLBCL cell lines. These important proof-of-concept studies could provide a renewed purpose for the large pharmacopeia of BET bromodomain inhibitors.

Like JQ1, CDK9 inhibitors have also been used to construct BCL6 TCIPs.³⁴⁵ CDK-TCIP1, a chimera of BI-3812 and the CDK inhibitor SNS-032 can induce BCL6-CDK9 proximity and rapidly activate BCL6 target genes by inducing CDK9 to phosphorylate RNA Pol II at BCL6-bound promoters (Figure 10d). CDK-TCIP2, which harbors a more rigid and structured spirocyclic linker, showed more potent activity and improved pharmacokinetic behavior, allowing its use in animal models where it potently inhibits the proliferation of B cells in the germinal center. An interesting distinction from BCL6-BRD4 TCIPs is that BCL6-CDK9 TCIPs are constructed from active-site inhibitors of CDK9 enzymatic activity. Recruitment of mutant CDK9 isoforms lacking catalytic activity failed to activate BCL6 target genes, indicating that the catalytic activity of CDK9, and not its scaffolding functions, is strictly required for TCIP activity. This discovery indicates that an inhibited form of CDK9 can be recruited to BCL6 and then released to form a locally high effective molarity of catalytically competent CDK9 molecules that then mediate locus-specific transcriptional activation. This is consistent with a previous study that successfully repurposed HDAC inhibitors to recruit HDACs to a specific genomic locus for chemically induced transcriptional silencing.³²⁸ Similar to BET bromodomain inhibitors, TCIP pharmacology may provide a renewed interest in CDK9 inhibitors, which have otherwise proven challenging to translate into the clinic.

Expansion of TCIP pharmacology beyond BCL6 will require further advances in TF ligand discovery. However, with few exceptions, such as nuclear hormone receptors,³⁴⁶ TFs typically lack the types of naturally evolved ligand-binding domains that are preferred for chemical probe and drug discovery.^347,348 Chemoproteomic-based methods for ligand discovery, which have matured at a remarkable pace in recent years,^349–351 show great promise for the future of TF-directed drug discovery.^352–356 Cysteine-reactive ligands are now reported for MYC, β-catenin, SOX10, NF-κB1, and FOXA1.^{352,353,355,356} These early-stage compounds have produced a great diversity of functional effects, for example, inducing degradation of β-catenin, stabilizing SOX10 dimerization, and altering the sequence specificity of FOXA1-DNA interactions. Most relevant to the chromatin-specific focus of this review, covalent ligands for the pioneer factor FOXA1 were recently discovered by chemoproteomic-based ligand discovery.³⁵² The tryptoline acrylamide, WX-02–23, stereoselectively engages a cysteine (C258) within the FOXA1 DNA-binding domain, improving its affinity for weak motifs without affecting strong ones, thus relaxing the basal sequence preferences of FOXA1. In cells, this results in many thousands of gained and lost FOXA1 binding sites, with these sites showing increases and decreases in chromatin accessibility, respectively.³⁵² This effect on chromatin accessibility is consistent with the description of FOXA1 as a pioneer TF,^357,358 and, interestingly, the mechanism by which WX-02–23 affects FOXA1 after engaging C258 resembles a recurrent cancer-associated mutation (R261G) located nearby.^{352,359–361} This cellularly active small-molecule modulator of FOXA1 highlights the potential for chemoproteomic-based ligand discovery to address historically challenging targets, like transcription factors.

2.4. Context-specific inhibition of pan-essential gene products

While the products of common essential genes typically make for challenging drug targets, the complex genetic alterations observed in human cancers can present opportunities to selectively modulate an essential protein in tumor cells versus a normal genetic background. This principle has led to the clinical development of PRMT5 (protein arginine methyltransferase) inhibitors that target cancers harboring homozygous deletions of methylthioadenosine phosphorylase (MTAP). Homozygous deletions within the chromosome 9p21 region, which inactivate the tumor suppressor gene CDKN2A, are among the most commonly observed genetic alterations in cancer.³⁶² MTAP, which is in close genomic proximity to CDKN2A, is often co-deleted, causing a dramatic sensitization to knockdown of PRMT5, WDR77 and RIOK1 – genes encoding subunits of the PRMT5 methylosome.^363–365 In the methionine salvage pathway, the conversion of methylthioadenosine (MTA) to adenine and methionine is catalyzed by MTAP (Figure 11a). Tumor cells harboring MTAP deletions therefore accumulate high concentrations of MTA, which selectively inhibits PRMT5 by binding competitively with S-adenosylmethionine (SAM), the cofactor required for methylation by PRMT5 (Figure 11b).^363–365 This partial inhibition of PRMT5 by MTA in MTAP-deficient tumors has enabled the development of compounds that target the cell-essential functions of PRMT5 selectively in tumor cells.^366–369

Figure 11.

Development of MTA-cooperative PRMT5 inhibitors to selectively target the common essential protein PRMT5 in MTAP-deleted tumors. (a) Metabolism of MTA by MTAP. (b) MTAP deletion elevates cellular MTA concentrations, partially inhibiting PRMT5-mediated deposition of symmetric dimethyl arginine (SDMA) modifications. (c) PRMT5 inhibitors that do not bind cooperatively with MTA inhibit PRMT5 in “normal” cells and MTAP-deleted tumor cells, diminishing their therapeutic window. (d) MTA-cooperative inhibitors selectively inhibit PRMT5 in MTAP-deleted tumor cells. (e) Chemical structures of MTA-cooperative PRMT5 inhibitors.

Knockout of Prmt5 causes embryonic lethality in mice and DepMap demonstrates PRMT5 is encoded by a common essential gene (Figure 11c).³⁷⁰ It catalyzes the post-translational modification of arginine to symmetric dimethylarginine (SDMA), primarily regulating histone proteins, RNA splicing factors, and ribosomal proteins.^371–379 While RNAi knockdown of PRMT5 is tolerated by wild-type cells, the hypomorphic state of PRMT5 caused by the accumulation of MTA makes MTAP-deleted cells sensitive to further PRMT5 loss-of-function,^363–365 but a complete knockout impairs the viability of both wild-type and MTAP-deleted cancer cells.³⁶⁴ This proved to be a critically important distinction in developing PRMT5 agents that can exploit MTAP synthetic lethality. EPZ015666, a selective PRMT5 methyltransferase inhibitor predated the report of PRMT5-MTAP synthetic lethality.^380,381 Originally developed to target PRMT5 in mantle cell lymphoma and discovered by high-throughput screening, EPZ015666 binds cooperatively with SAM and competitively with PRMT5 substrates.^380,381 This mechanism of inhibition is indifferent to MTA concentrations, causing antiproliferative effects indiscriminate to MTAP status (Figure 11d).^363–365 This inspired the need to develop inhibitors that could bind cooperatively with MTA to the PRMT5-MTA complex an effectively exploit the synthetic lethal relationship.^363–365

Several MTA-cooperative PRMT5 inhibitors have now been reported, including MRTX1719, TNG908, AZ-PRMT5i-1, and AMG 193 (Figure 11e,f).^{366–369,382} MRTX1719, the first reported MTA-cooperative inhibitor, was developed using structure-based optimization of 4-(aminomethyl)phthalazin-1(2H)-one, a hit discovered by surface plasmon resonance (SPR)-based fragment screening.^366,367 The fragment showed 5-fold preference in affinity for PRMT5-MTA over PRMT5-SAM (K_D = 10 μM versus 51 μM) and was co-crystalized with both complexes to understand the structural basis of its selectivity. It was further optimized using consecutive rounds of structure-based design, with analogs being tested in biochemical enzyme assays with and without MTA present to optimize for MTA cooperativity. MRTX1719 inhibits PRMT5-mediated SDMA marks with a potency of 8 nM in MTAP-deleted cells and inhibits growth with a similar potency of 12 nM. In cells with wild-type MTAP, inhibition of SDMA is shifted to 653 nM and cellular growth inhibition drops to 890 nM. This is consistent with SPR experiments showing that MRTX1719 has a 67-fold greater preference for binding PRMT5-MTA over PRMT5-SAM (K_D= 0.140 pM versus 9.4 pM). An atropisomer of MRTX1719 was nearly 200-fold less active in MTAP-deleted cells, further confirming its on-target activity. In mouse models, this leads to potent anti-tumor activity against MTAP-deleted, but not wild-type xenografts, validating their sensitization to MTA-cooperative PRMT5 inhibition in vivo.^366,367 Clinical reports from patients treated with MRTX1719 in a Phase 1/2 study have also confirmed SDMA inhibition in MTAP-deleted tumors and several objective responses (RECIST 1.1 criteria) have been observed with tumor size decreasing over time, providing an impressive clinical proof-of-concept (NCT05245500).³⁶⁷ Another clinical agent, AMG 193 was optimized from an aminoquinoline scaffold discovered in a DNA-encoded library (DEL) screen that was run in the presence or absence of MTA.³⁶⁹ Selectivity for MTAP-deleted cancer cells was improved from 3-fold to 40-fold through structure-based optimization, with AMG 193 showing remarkable selectivity for MTAP-deleted cells and a strong correlation to PRMT5 knockdown across nearly 1000 cancer cell lines. Like MRTX1719, AMG 193 shows selectivity for MTAP-deleted xenografts and early clinical reports show activity against MTAP-deleted cancers (NCT05094336).³⁶⁹ These initial clinical reports ease concerns about a prior study that suggested the synthetic lethality between MTAP and PRMT5 would be muted in vivo by stroma-mediated metabolism of excess MTA.³⁸³

3. Targeting genetically restricted vulnerabilities

To overcome the challenges related to on-target toxicity that can be presented by targeting common essential genes, there have been broad efforts to identify genes that are required for the survival of cancer cells harboring specific genetic alterations. In principle, drugs targeting these genes can leverage synthetic lethality to selectively target cancer cells and spare normal tissues and physiological functions. These genetically restricted cancer vulnerabilities can take many forms and are richly associated with chromatin biology. For example, we discuss dependencies that arise from the unnatural behavior of chimeric oncoproteins, like MLL fusions, or hypermorphic mutant enzymes, like EZH2. In a more specialized case—paralog synthetic lethality—genetic loss of function affecting one of two closely related paralogs can create a highly selective vulnerability on the other. This most often occurs when the overlapping function of these paralogs is required, in general, for cell survival. Thus, when one paralog is missing in cancer cells but not normal cells, the other paralog can present a highly attractive drug target. While chromatin biology harbors many examples of this synthetic lethality, paralogs also tend to have highly similar structures and ligand-binding pocket. This presents substantial challenges for discovery chemistry, as a chemical agent must selectively target the remaining paralog to effectively leverage the synthetic lethal relationship. As we will discuss, this challenge remains a major area of ongoing effort and innovation for chromatin drug discovery.

3.1. MLL-fusion dependencies

MLL (mixed lineage leukemia), also known as KMT2A, is a histone methyltransferase critical for the transcriptional regulation of hematopoiesis and cellular differentiation.³⁸⁴ MLL controls the expression of developmental programs, most notably by activating the transcription of HOX cluster genes.^385,386 The MLL gene is frequently rearranged by chromosome 11q23 translocations, leading to the production of more than 80 different MLL fusion oncoproteins that drive the development of acute leukemia.^387–391 These fusion proteins combine the N-terminal chromatin-binding domains of MLL with the C-terminal region of a partner protein, most often a transcriptional coactivator.^392–394 The resulting oncoprotein localizes to MLL target sites on chromatin where it aberrantly activates transcription via the C-terminal fusion partner (Figure 12a).³⁹⁵ In this way, MLL fusions dysregulate the developmental control of MLL target genes, promoting the ectopic expression of leukemic proto-oncogenes, such as HOXA9 and MEIS1.^396,397 In preclinical models, leukemias driven by this mechanism are preferentially vulnerable to the disruption of many transcriptional co-activators and chromatin regulators that support the function of MLL and MLL-fusion oncoproteins.³⁹⁸ This has motivated the discovery of chemical tools and clinical development programs for several chromatin regulatory proteins such as ENL, DOT1L, WDR5, and menin, which are discussed throughout this chapter. Notably, the success of targeting menin with a small molecule menin-MLL inhibitor, recently approved by the FDA for the treatment of MLL-fusion leukemia, provides strong validation for this therapeutic approach in a clinical setting.

Figure 12.

Oncogenic MLL fusions create cancer-specific vulnerabilities. (a) MLL fusion proteins form pathogenic transcriptional regulator complexes and vulnerabilities to inhibition of specific transcriptional regulators, such as DOT1L. (b) DOT1L is a selective dependency in MLL-rearranged leukemia. Boxplot of CRISPR gene effect scores for DOT1L across 1150 cancer cell lines (DepMap) grouped by the presence of an MLL-rearrangement (N = 12). Boxes represent the median and IQR 25–75%. Whiskers extend 5–95%.

3.1.1. DOT1L

Fusion proteins resulting from MLL chromosomal translocations retain the N-terminus of MLL fused with the C-terminus of another protein. While dozens of MLL fusion partners have been described, the most prevalent are proteins involved in transcriptional activation, including AF4, AF9, ENL, and AF10. Many of these fusion partners have been shown to interact with the histone methyltransferase, DOT1L, either directly or via higher-order multiprotein complexes.^{395,399–408} DOT1L (disruptor of telomeric silencing 1-like) was originally discovered in yeast as a regulator of telomere silencing and has since been identified in mammals as the only known histone H3 lysine 79 (H3K79) methyltransferase.^409–412 H3K79me2/3 modifications are present in the body of actively transcribed genes and elevated levels of H3K79me2/3 are positively correlated with transcriptional output.^413–415 DOT1L activity is required during development for proper erythropoiesis and has important roles in adult hematopoiesis.^416,417 However, MLL-fusion leukemia are hypersensitive to DOT1L loss-of-function, with MLL-fusion oncoproteins shown to colocalize with a disproportionately high amount of H3K79me2 marks at MLL-specific target genes (Figure 12b).^{401,405,418–420} The interaction between MLL-fusion proteins and DOT1L triggers the recruitment of DOT1L to ectopic sites, resulting in the hypermethylation of H3K79 and the dysregulated expression of genes such as HOXA9 and MEIS1, which are crucial for initiating and sustaining leukemogenic proliferation. Genetic knock-out and mutagenesis experiments have shown that DOT1L, and more specifically its histone methyltransferase (HMT) activity, are indispensable for the growth of MLL-fusion leukemia.^{419,421–426}

Unlike other lysine methyltransferase proteins, DOT1L does not contain a canonical SET (su(var)3–9, enhancer-of-zeste, and trithorax) methyltransferase domain. Instead, it contains a class 1 methyltransferase domain that is structurally similar to arginine methyltransferases and DNA methyltransferases.^427,428 Initial efforts to discover DOT1L methyltransferase inhibitors used libraries of nucleoside analogs based on the native methyl donor, SAM (Figure 13a).^{420,429–432} The SAM cofactor binds a narrow channel on DOT1L with the methionine group of SAM projecting into the center of the protein (Figure 13a). A loop at the entrance to the tunnel is unique to DOT1L and likely contributes to its specificity for H3K79.⁴²⁷ All histone methyltransferases use SAM as a methyl group donor; however, likely due to its distinct non-SET domain, DOT1L inhibitors show excellent selectivity against other methyltransferases, as witnessed by the first prototypical DOT1L inhibitor, EPZ004777.^420,432 EPZ004777 retains the adenosyl core of SAM but replaces the sulfur atom with a tertiary amine that connects to an isopropyl group and a tert-butyl phenyl urea (Figure 13b). EPZ004777 binds competitively to the SAM-binding pocket and causes the opening of a previously inaccessible hydrophobic pocket (Figure 13b). EPZ004777 exhibits a similar binding pose as SAM, maintaining interactions with key amino acid residues within the active site, as well as a novel interaction with the carbonyl oxygen of Gly163.⁴³² Treatment with EPZ004777 in vitro causes a global decrease in H3K79me2 levels, but several days are required to observe this effect.⁴²⁰ There is no cellular demethylase known for the removal of H3K79 methylation, so the global depletion of this mark appears to rely on histone H3 turnover, resulting in a delay in the onset of effects from DOT1L inhibition.^433–435 EPZ004777 treatment produces a dose and time dependent decrease in HOXA9 and MEIS1 transcripts, key drivers of MLL-fusion leukemia, and selectively inhibits the proliferation of MLL-rearranged leukemia cells, inducing differentiation and apoptosis.⁴²⁰ Evaluating the anti-leukemia activity of EPZ004777 in mice required administration via subcutaneous pumps, where it reduced H3K79me2 levels and significantly increased the survival of mice bearing xenografts of the MV4;11 cell line, a model of MLL-AF4-rearranged AML. This compound validated the therapeutic strategy for targeting DOT1L in MLL-rearranged leukemia, but it had insufficient pharmacokinetic properties for clinical development.⁴²⁰

Figure 13.

DOT1L methyltransferase inhibitors (a) DOT1L cofactor, SAM (blue), in complex with DOT1L (PDB: 3QOW). (b) EPZ004777 bound to DOT1L (PDB: 4EKI). (c) EPZ-5676 bound to DOT1L (PDB: 4HRA). The DOT1L loop at the entrance to the tunnel is shown in pink in each structure.

Many groups underwent efforts to optimize the pharmacokinetic properties of adenosyl DOT1L inhibitors.^{430–432,436} The addition of a halogen atom (bromine) to S-adenosyl homocysteine (SAH) resulted in an almost 8-fold greater affinity for DOT1L,⁴³¹ and application of this finding to EPZ004777 yielded the brominated chemical probe, SGC0946.⁴³⁷ The cryptic hydrophobic pocket revealed by EPZ004777 suggested that DOT1L could accommodate larger hydrophobic substituents.^432,437 Structure-guided optimization of EPZ004777 led to the development of the first clinical candidate for DOT1L, EPZ-5676 (Figure 13c). EPZ-5767 (pinometostat), replaced the tert-butyl phenylurea from EPZ004777 with a tert-butyl benzimidazole, connecting it with a cyclobutyl group, rather than a polymethylene chain, introducing more rigidity to the molecule.⁴³⁶ X-ray crystallography of pinometostat and DOT1L revealed that, similar to EPZ004777, EPZ5676 induces a conformational change that widens the cryptic hydrophobic binding pocket and increases favorable interactions with the compound (Figure 13c).^432,436 Pinometostat shows greater HMT inhibition, potently decreasing expression of HOXA9 and MEIS1 and robustly inhibiting the proliferation of MLL rearranged leukemia cells. Due to its more suitable pharmacokinetic properties, it was administered by continuous IV infusion to MV4;11 xenograft mice, where it decreased tumor volume, oncogenic gene expression, and H3K79 methylation.^436,438,439 In the clinic, pinometostat was well tolerated at levels sufficient to inhibit DOT1L HMT activity, but therapeutic responses were minimal.⁴⁴⁰ In contrast to the anti-tumor effects seen in preclinical models, clinical responses were far more varied, with 2 of 51 patients achieving complete remission. In pediatric studies, pinometostat treatment caused a brief reduction in peripheral or bone marrow blasts but did not provoke sustained responses.^441,442 This showed that the effects of disrupting DOT1L methyltransferase activity are likely not sufficient for single-agent anti-cancer activity.⁴⁴³ Promising in vitro results have pointed toward the potential for combination therapies, but no such studies have been initiated in the clinic.^444,445

DOT1L inhibitors have been explored as possible therapies for other cancer types, including glioblastoma, breast cancer, NPM1-mutant leukemia, and others.^446–455, Given the unfavorable pharmacokinetics of adenosine nucleoside inhibitors, many research groups have sought to develop compounds with novel chemical scaffolds to address the rapid degradation of previous inhibitors and improve bioavailability.^456–458 A variety of screening methods have led to the discovery of non-nucleoside DOT1L inhibitors.^459–461 Using SAH production to identify DOT1L inhibitors in a high-throughput chemical screen, a team at Novartis identified a non-nucleoside hit that was optimized to sub-nanomolar potency through structure-based design, ultimately yielding compounds with suitable pharmacokinetic properties for use in vivo (Figure 14a).⁴⁶² Interestingly, these compounds bind competitively with SAM despite not entering the SAM-binding pocket. They form contacts with Phe243 and Pro130, which are positioned on opposite sides of the entrance to the SAM-binding pocket (Figure 14b).⁴⁶² Two optimized compounds (Compounds 10 and 11) have been shown to inhibit H3K79me2 deposition, cell proliferation, and target gene expression in DOT1L sensitive cell lines, consistent with EPZ5676 treatment. In MLL-AF9 and MLL-AF6 patient derived xenograft (PDX) mouse models, oral delivery of either compound reduces leukemia burden in bone marrow, spleen and the peripheral blood.⁴⁶³

Figure 14.

Chemical structures of bioavailable non-nucleoside DOT1L inhibitors. (a) Chemical structure of high throughput screening (HTS) hit compound and optimized compound 10. (b) Crystal structure of Compound 10 bound to DOT1L (PDB: 6TEL) showing interactions with key amino acids in green.

The use of such non-nucleoside inhibitors could help advance preclinical assessment of DOT1L target biology in vivo; however, none are currently suitable for clinical investigation due to limited efficacy.^462–464 In addition to these methyltransferase inhibitors, there have also been efforts to directly perturb DOT1L through alternative mechanisms, notably by inhibiting its interactions with MLL fusion proteins.^{403,465–470} DOT1L interacts with the ANC-1 homology domain (AHD) of the common MLL fusion partners, AF9 and ENL, which is required for SEC recruitment and proliferation in MLL-r leukemias. Peptidomimetics based on the DOT1L peptide that binds the AHD of ENL/AF9 validated that disrupting DOT1L protein-protein interactions is sufficient to inhibit proliferation in dependent leukemias.^403,465 Small-molecule inhibitors have been designed for the AHD of ENL/AF9 and inhibit proliferation in the low micromolar range, validating the tractability and utility of targeting DOT1L protein-protein interactions via the AHD domain.^{400,401,428,468,469,471} Despite the structural diversity of DOT1L agents, clinical development appears to have stalled with the termination of EPZ5676 clinical trials. DOT1L has all the makings of a profoundly compelling anti-leukemia target. It is a highly selective dependency in MLL-rearranged leukemia and DOT1L inhibitors show promising anti-cancer effects in preclinical studies. It is therefore surprising that pinometostat did not show more success in the clinic. The fact that it elicited positive pharmacodynamic responses but not therapeutic responses, together with the lack of development for non-nucleoside inhibitors, indicates that DOT1L dependency may simply not be strong enough to elicit sufficient therapeutic responses in patients.

3.1.2. Menin

MEN1, which encodes for the protein menin, was initially discovered through efforts to map the gene on chromosome 11q13 that is responsible for the autosomal dominant familial cancer syndrome, multiple endocrine neoplasia-type 1 (MEN1).⁴⁷² Germline loss-of-function mutations in this gene lead to tumorigenesis in multiple endocrine organs, suggesting its role as a tumor suppressor.⁴⁷² Later, it was discovered to form stable interactions with the N-terminal region of the methyltransferase, MLL (Figure 15a).^473–475 This N-terminal region is retained in MLL-fusion oncoproteins and menin is required for the pathogenic function of MLL-fusion oncoproteins. Genetic loss of menin or deletion of the menin-MLL interaction site results in decreased gene expression of HOX genes and causes cell differentiation.⁴⁷⁴ Acute leukemia driven by MLL-fusion oncoproteins are dependent upon menin to maintain stem-like gene expression and evade differentiation, but menin is dispensable in many healthy tissues.^474,476 These studies revealed the importance of menin in MLL-rearranged leukemias and motivated the menin-MLL interaction as a possible target for small-molecule inhibition.

Figure 15.

Small molecule inhibitors of menin-MLL. (a) Depiction of minimal MLL peptide interacting with menin (PDB: 3U88). (b) structure of early lead compound MI-2 and its interactions with menin (PDB: 4GQ3). Key amino acid residues are highlighted in green with the novel interactions compared to MLL in maroon. (c) Crystal structure of menin interacting with VTP-50469 (PDB: 6PKC). A novel distal interaction with tryptophan 341 is shown in purple.

The first small-molecule inhibitor of the menin-MLL protein-protein interaction, reported in 2012, was discovered using a fluorescence polarization (FP) assay to screen for compounds that disrupt the interaction of menin with a synthetic and fluorescently-labeled MLL peptide.⁵⁴ Optimization of a thienopyrimidine screening hit (MI-1) resulted in the discovery of MI-2 and MI-3, which bind to menin with sub-micromolar affinity (Figure 15b). These compounds recapitulate genetic knockout of MEN1 in leukemia cell lines, causing a decrease in HOX and MEIS1 gene expression, as well as inducing G1/G0 cell cycle arrest in MLL-rearranged (MLLr) leukemia cells. A high-resolution crystal structure of MI-2 bound to menin demonstrated that these inhibitors bind the same cavity as the MLL peptide and engage in many of the same interactions (Figure 15b).⁴⁷⁷ The nitrogen atoms in the thienopyrimidine ring of MI-2 form hydrogen bonds with menin amino acid residues Tyr276 and Asn282 and a gem-dimethylthiazole moiety fits into a cavity formed by Tyr319, Tyr323 and Met322, mimicking interactions that are crucial for MLL binding. This structure revealed the relatively shallow nature of the menin binding pocket, highlighting the impressive ability of these compounds to inhibit a protein-protein interaction (PPI). Structure-guided optimization resulted in the discovery of MI-2–2, which replaced a propyl substituent on the thienopyrimidine core with a trifluoroethyl group and improved potency by approximately 10-fold compared to MI-2.⁴⁷⁷

Further improvements primarily focused on retaining the thienopyrimidine core^477–480 and optimizing pendant groups on the other end of the molecule to improve affinity and enable in vivo study, which was not possible with MI-2–2 due to modest activity and poor metabolic stability.^479,481 Efforts to develop the MI-2–2 scaffold for in vivo assessment led to MI-136, which contains a cyanoindole-ring connected to the thienopyrimidine core through a piperidine linker. A further optimized ligand, MI-503, binds menin with single-digit nanomolar affinity and is able to inhibit leukemic gene expression in MLL-rearranged cell lines. It shows suitable pharmacokinetic properties and when tested in MV4–11 xenograft mice,^479,481 improves overall survival, and decreases HOX/MEIS1 gene expression. It causes differentiation of leukemia blasts without substantially impairing normal hematopoiesis, providing the first in vivo demonstration that menin inhibition could provide a well-tolerated therapeutic approach in MLL rearranged. Additional improvements of this scaffold have led to orally bioavailable thienopyrimidine menin inhibitors (MI-3454), which induce regression of established tumors, and ziftomenib (KO-539), which is in clinical trials to evaluate its use as a single agent or combinatorial therapeutic in patients with relapsed or refractory AML.^{56,479,481–485}

VTP-50469 was designed from the piperazinyl pyrimidine fragment of MI-2 using an iterative structure-based approach.⁵⁷ VTP-50469 lacks the fused thiophene moiety found in the thienopyrimidine structures but maintains the key pyrimidine interactions with Tyr276. The cyclohexyl sulfonamide group extends beyond the canonical binding site to form a distant hydrogen bond with Trp341 (Figure 15c). Replacing the piperazine with a diamino spirocyclic moiety kept key cation interactions with Tyr319 and Tyr323 and caused a ~30-fold increase in potency (K_i = 104 pM). VTP-50469 (also SNDX-50469) showed greater cell growth inhibition than previous menin inhibitors in vitro, suppressing the growth of MLL-rearranged PDX models so substantially that a subset of mice remained disease-free and on treatment for over a year.⁵⁷

VTP-50469 and closely related analogs have consistently shown profound activity as a single agent and combination therapeutic in mouse models of AML.^58,483,486 Most notably, this includes not only MLL-r leukemia, but also AML harboring nucleophosmin (NPM1) mutations. NPM1 is among the most frequently mutated genes in AML (>25%), encoding a mutant protein that is mis-localized from the nucleolus to the cytoplasm (NPM1c).^487,488 Expression of genes within both the HOXA and HOXB cluster is characteristic of NPM1c-positive AML and is a direct and pathogenic consequence of the mis-localization of the protein.^489–491 The activity of menin was first studied in NPM1-mutant AML based on the hypothesis that MLL would regulate HOX expression in diverse leukemia, not only MLL-r.⁴⁵¹ Indeed, MI-503 was shown to be highly active against the NPM1c AML cell line, OCI/AML-3, in vitro and to cause a significant, albeit modest, extension of lifespan for mice bearing OCI/AML-3 xenografts. Later studies, using MI-3454 and VTP-50469 to treat PDX models of NPM1c AML showed remarkable activity, with treatment causing disease regression rather than simply slowing its progression.^56,58 VTP-50469 treatment was tolerated for over 100 days with overall survival being prolonged by hundreds of days after withdrawal of treatment, highlighting the exceptionally impressive activity of menin inhibition in these aggressive PDX models. It is perhaps instructive to compare the preclinical effects of menin-MLL inhibitors with DOT1L inhibitors, which never showed the same degree of profound and durable responses in mouse models.

The impressive preclinical effects of menin-MLL inhibitors resulted in clinical trials to evaluate their effects in both MLL-r and NPM1c acute leukemia. Revumenib (SNDX-5613), a close analog of VTP-50469, was recently reported to be safe and efficacious in a Phase 1/2 clinical trial evaluating its use in relapsed or refractory disease (NCT04065399, AUGMENT-101) (Figure 16a).^59,60 Complete remission or complete remission with partial hematological recovery was observed in 30% of 60 patients.⁶⁰ The overall response rate to revumenib treatment was 59% in MLL-r leukemia and 36% in NPM1c leukemia, with patients showing on-target decreases in leukemogenic MLL target genes, such as HOXA9/MEIS1, and an increase in the differentiation marker, CD11b.⁶⁰ No responses were observed in the 8 patients who lacked MLL or NPM1 mutations, supporting the conclusion that the efficacy of revumenib is due to on-target inhibition of menin-MLL. This was definitively established by sequencing of patient samples, which revealed acquired menin mutations as a primary cause of therapeutic resistance.⁵⁹ Recently, the FDA approved revumenib for relapsed and refractory acute leukemia with MLL rearrangements, marking a major breakthrough for both patients and the field.

Figure 16.

FDA-approved menin inhibitor, revumenib. (a) Chemical structure of revumenib. (b) Crystal structure showing revumenib bound to menin (PDB: 7UJ4). Amino acids involved in acquired resistance (M322 and T344) are shown in green. (c) Schemic depiction of revumenib (hexagon) mechanism of action and resistance mechanisms. Resistance mutations preventing binding of menin to revumenib but allow menin-MLL reassociation and MLL target gene expression.

In four patients who initially responded to revumenib but then relapsed or progressed while on treatment, each harbored at least one mutation in MEN1.⁵⁹ This motivated a broader search for MEN1 mutations, examining 38 patients before and after nearly two months of treatment. While no mutations were detected before treatment, 12 of 38 patients carried one or more MEN1 mutations after treatment. These mutations, such as M322I and T344M, are closely clustered to a distal region of the ligand-binding pocket that contacts the cyclohexyl sulfonamide moiety, opposite the pyrimidine core (Figure 16b). They selectively disrupt the binding of revumenib to menin without substantially affecting the menin-MLL interaction (Figure 16c). Further study in preclinical models demonstrated that these mutations were sufficient to confer evasive resistance to revumenib and similar mutations within the MLL binding pocket could be identified by unbiased forward genetic approaches, both in tissue culture and in mouse models of acute leukemia. Altogether, these clinical results (i) validated the on-target activity of revumenib, (ii) confirmed the long-standing preclinical hypothesis that menin is a viable target for safe and effective leukemia therapeutics, and (iii) highlighted several opportunities to optimize this class of therapeutics. Interesting, while M327I affects revumenib and MI3454 equivalently, T349M more severely disrupted revumenib binding compared to MI3454, suggesting different scaffolds could possess non-overlapping liabilities to mutations in menin, which might allow for sequential therapy with distinct menin inhibitors to overcome resistance. The concentration of mutations within a region of menin that is dispensable for MLL binding suggests the need to avoid drug contacts with these side chains. However, therapeutic combinations may also minimize the opportunities for resistance to emerge through MEN1 mutations and clinical trials are currently underway to assess revumenib as part of combination regimens.^{485,492–494}

In addition to the thienopyrimidine class of menin inhibitors, several other scaffolds have been reported.^478,480,495 Structural analysis of the menin-MLL peptide complex lead to the development of the macrocyclic peptidomimetic inhibitor, MCP-1.⁴⁹⁵ MCP-1 binds with high-affinity to menin but requires further optimization due to poor cell permeability and its large molecular weight. A hydroxy- and amino-methylpiperidine class, which binds the same MLL pocket, was identified from a high-throughput screen shortly after the initial report of thienopyrimidine inhibitors.⁴⁸⁰ Since prolonged menin inhibition is required for optimal anti-leukemia effects, the aminomethylpiperidine compound, MIV-6, was utilized to explore the effects of irreversible covalent inhibition (Figure 17a). Structure-based optimization of this scaffold was followed by installation of an electrophilic acrylamide to capture covalent interaction with Cys329. This resulted in M-525, a potent covalent menin inhibitor that shows long-lasting anti-tumor activity in vivo (Figure 17b).⁴⁹⁶ Extensive efforts to improve this scaffold have resulted in compounds with greater potency against menin and dramatically improved pharmacokinetic properties, including an orally active covalent inhibitor, M-1121.^497,498 A covalent inhibitor, BMF-219, is currently in clinical trials for the treatment of patients with relapsed and refractory AML.⁴⁹⁹

Figure 17.

Development of covalent inhibitors for MENIN. a) Chemical structure of MIV-6 and crystal structure of MIV-6 bound to MENIN (PDB:4OG8). b) Chemical structure of covalent inhibitor, M-525, bound to MENIN (PDB:6B41). Key amino acid resides for the MENIN-MLL interaction are shown in green and the covalently bound cysteine 329 is shown in purple.

The past decade of research has validated both the dependency and pharmacological tractability of the menin-MLL interaction in MLL- and NPM1-mutant cancers. Promising clinical trial results in leukemia patients have shown that these compounds are safe and efficacious as single agents and could be promising in combinatorial therapies.⁵⁰⁰ The continued exploration and development of more effective menin inhibitors alone, or in combination with other clinically relevant anticancer therapies, aims to address resistance mechanisms observed in vitro, that potentially await clinical validation.⁵⁰¹ Interestingly, over the last few years, menin inhibitors have shown promising results in other indications, demonstrating anti-cancer effects across a wide breadth of cancer cell types.^55,502–505

3.1.3. WDR5

MLL fusion proteins have been shown to activate oncogenic gene expression in cooperation with the wild-type MLL protein.^506,507 MLL fusion proteins typically lack the C-terminal region of MLL, which contains both the SET domain and an arginine-containing sequence that is recognized by a binding site on WDR5. This sequence, called the WIN motif, is conserved across other SET-domain-containing proteins and the conserved arginine residue is required for recognition by WDR5.^508,509 The extent of MLL methyltransferase activity depends on the stability of the MLL-WRAD complex, which consists of MLL, WDR5, RbBP5, ASH2L, and two copies of DPY-30.⁵⁰⁹ Mutations in the MLL WIN motif or the WDR5 WIN binding site (WIN site) can disrupt the MLL-WDR5 interaction, resulting in a loss of MLL methyltransferase activity.⁵⁰⁹ Since MLL is an activator of oncogenic HOX gene expression, targeting MLL catalytic activity through its interaction with WDR5 could offer a novel therapeutic strategy.

WDR5 is a member of the WD40 repeat domain-containing family and is a core scaffolding component of MLL/SET methyltransferase complexes. WD40 repeat (WDR) domains are comprised of a conserved 40 amino acid sequence ending with tryptophan and aspartic acid residues. Most WDR proteins, like WDR5, contain 7 repeats and forms a seven-bladed propeller conformation (Figure 18a). There are two binding sites on WDR5 through which its protein-protein interactions occur, the WBM (WDR5 Binding Motif) site and the WIN (WDR5 Interaction) site (Figure 18b).⁵¹⁰ The WIN site is an arginine binding cavity on the surface of WDR5 that interacts with the MLL WIN motif as well as functions to read H3K4me marks.^38,511,512 Genetic loss of WDR5 causes global decreases in H3K4 methylation and, ultimately, leads to inactivation of many essential genes.^38,511 MLL chromatin localization, however, is unchanged after WDR5 loss, suggesting that WDR5 regulates the methyltransferase activity of MLL and not its localization.³⁸

Figure 18.

WDR5 WIN site ligandability. (a-b) WDR5 structure showing the WIN site (pink) and the WBM site (blue) located on opposing faces of the seven-bladed barrel structure (PDB: 3EG6). (c) Development of WDR5 WIN site inhibitors based on the MLL WIN peptide sequence. The key arginine motif is highlighted in orange.

The arginine residue on the MLL WIN motif (R3765) is required for binding to the central cavity of WDR5 and inspired the development of peptidomimetic inhibitors containing a common arginine residue.^513–516 Motivated by the promising effects of peptidomimetics, multiple groups sought small-molecule inhibitors of the MLL-WDR5 interaction.^517–521 One group used a fluorescent polarization (FP)-based high-throughput screen to identify a single hit compound, the benzamide WDR5–0101.⁵²² A follow-up screen with structurally similar benzamide compounds identified WDR5–47 (Figure 19a), which was structurally determined to occupy the central arginine binding cavity on WDR5. It binds competitively with and displaces the binding of WIN motif containing proteins, such as MLL.⁵²³ These discoveries inspired further compound optimization that, ultimately, identified numerous selective small-molecule ligands for WDR5, like OICR-9429 (Figure 19a).^523–527 OICR-9429 effectively inhibits the methyltransferase activity of MLL in vitro and decreases the proliferation of MLL-rearranged cells.

Figure 19.

Small-molecule inhibitors of the WDR5 WIN domain. (a) Development of the WIN site inhibitor OICR-9429. (b) Chemical structure of bioavailable WIN site inhibitors DDO-2093 and DDO-2213. (c) Gene effect scores for the common essential gene, WDR5 (DepMap).

Additional work on this scaffold led to the discovery of compounds with suitable pharmacokinetic properties for in vivo use.^528,529 These compounds retained the methylpiperazine and central phenyl groups of OICR-9429 but included modifications to the C5 position of the phenyl ring (Figure 19b). One compound, DDO-2213, was discovered using a scaffold hopping approach, resulting in the replacement of the benzamide with a pyrimidine ring.⁵²⁸ DDO-2213 binds to WDR5 with high affinity and selectively reduces the MLL methyltransferase activity over other HMT proteins.⁵²⁸ Another optimized compound, DDO-2093, retained the core scaffold of OICR-9429 but replaced the phenyl morpholine ring of OICR-9429 with a phenyltriazole moiety (Figure 19b). Both compounds have been shown to possess suitable properties for assessment in mouse xenograft models of leukemia, each causing a decrease in tumor mass, tumor weight, and HOXA1/MEIS1 expression without obvious toxicities. Promisingly, DDO-2213 has suitable pharmacokinetic properties to be administered orally. These compounds are the first inhibitors of the WIN site to show proof-of-concept in vivo. However, WDR5 is a common essential protein (Figure 19c), and it remains unclear whether a safe therapeutic window can be achieved with WDR5 inhibition.

The WDR5 inhibitor OICR-9429 has been used to fashion WDR5 PROTACs.^530–533 The resulting degraders showed greater efficacy in cells than OICR-9429 and were highly selective for WDR5.^530–532 One CRBN-based PROTAC showed degradation of WDR5 and off-target degradation of the CRBN neo-substrates, IKZF1 and IKZF3. This degrader elicits greater inhibition of WRAD complex target gene expression, attributed to decreased MLL occupancy on chromatin, and more effectively suppresses the growth of MV4;11 and EOL-1 AML cell lines compared to a WDR5-selective degrader.⁵³³ Interestingly, despite WDR5 being considered common essential, the WDR5 and IKZF1/3 degrader lacks general toxicity, suggesting that WDR5 may still present a viable target for MLLr leukemias.⁵³³

WIN site inhibitors and PROTACs are joined by WBM site inhibitors, where the oncogenic transcription factor, MYC, binds to WDR5.⁵³⁴ Multiple groups have shown that disruption of the WBM site prevents MYC from associating with a subset of its target genes,^535,536 providing support for targeting MYC through its interaction with WDR5. A high-throughput screen for small-molecule inhibitors of this interaction revealed a biarylsulfonamide hit that binds the WBM site.⁵³⁷ Structure-guided optimization afforded compounds that inhibit the WDR5 interaction with MYC in cells, along with structurally related negative control compounds that do not bind WDR5.⁵³⁷ Hits from an NMR-based fragment screen combined with these biarylsulfonamide inhibitors allowed for structure-guided scaffold “merging” to design new hits with improved drug-like properties.⁵³⁸ The compound with the highest affinity, compound 12 (K_D = 0.10 nM), causes a fourfold decrease in WDR5-MYC association in cells and decreases MYC binding to WDR5-dependent chromatin targets.⁵³⁸

Both WIN site inhibition and WBM disruption appears to cause a reduction of WDR5 on select chromatin targets, specifically those encoding ribosomal proteins.⁵¹⁷ WDR5 functions at these targets to recruit MYC through its WBM. Therefore, it is possible that disrupting either the WIN site or the WBM causes the same result – reduced MYC recruitment to genes that encode ribosomal biogenesis proteins.⁵³⁴ While its importance in MLL-rearranged leukemia is supported by both chemical and genetic studies,⁵³⁹ a role for WDR5 in ribosomal biogenesis may be consistent with its annotation as a common essential protein. Site-specific WDR5 inhibitors, which affect a subset of WDR5 functions,^540,541 may therefore allow for selective targeting of non-essential chromatin related functions of WDR5 while avoiding toxicity predicted by genetic loss of function. Contrary to earlier reports, the effects of WIN inhibition are caused by disruption of WDR5 binding to chromatin and cannot be attributed directly to changes in histone methylation, which only occur subsequently,^517,541 indicating a less privileged link to MLL fusion tumors than previously hypothesized.

CRISPR/Cas9-based loss-of-function genetic modifier screens have shown that the cellular response to WIN inhibitors relies on decreased ribosomal gene transcription and activation of p53 response pathways, with p53-deficient cancer cell lines showing a lack of sensitivity to WIN inhibition.⁵⁴² These studies afford more comprehensive understanding of the mechanisms underlying the sensitivity to WDR5 WIN inhibitors, which may help to identify biomarkers that predict responsiveness to WDR5 antagonists in the diverse set of cancers that have been shown sensitive to WDR5 loss-of-function.^543–548 Overall, WIN inhibitors appear to target an oncogenic role of WDR5 that partially avoids its common essential functions, implying that there may be opportunities to develop selective anti-cancer therapeutics targeting WDR5.

3.1.4. ENL

ENL was first identified as a potential acute leukemia target in two studies reported together in 2017.^549,550 In one, a genome-scale CRISPR/Cas9-based loss-of-function screen identified ENL as an essential gene in leukemia driven by MLL fusion oncoproteins.⁵⁴⁹ In another study, the prevalence of ENL and AF9 in MLL fusion oncogenes motivated a reverse genetic approach to study the function of wild-type ENL and AF9 alleles in MLL-rearranged leukemias.⁵⁵⁰ This revealed ENL, but not AF9, as being required for the fitness of MLL-rearranged acute leukemia cells. Both reports demonstrated normal hematopoietic stem and progenitor cells are insensitive to ENL loss-of-function, suggesting ENL-targeted therapeutics might possess limited on-target toxicity to normal tissues, a finding that is further supported by its profile in the DepMap dataset (Figure 20a).

Figure 20.

Exploiting ENL dependencies in MLL-rearranged leukemias. (a) Gene effect scores for ENL in cell lines grouped by MLL-rearrangement (DepMap). (b) H3K27ac (blue) occupying the YEATS domain of ENL (PDB:5J9S). Tyrosine 78 (Y78) and Phenylalanine 59 (F59) on loops 6 and 4, respectively, are shown in green. (c) JalView sequence alignment of all 4 human YEATS domains from ENL, AF9, YEATS2 and YEATS4. Darker red corresponding to higher conservation. Conserved secondary structure is depicted below alignment L corresponds to loops and arrows correspond to beta sheets. (d) Chemical structure of ENL/AF9 YEATS domain inhibitor, SGC-iMLLT1 (top), and the X-ray crystal structure showing SGC-iMLLT1 (blue) bound to ENL (bottom) (PDB: 6HT1). At top, SGC-iMLLT is colored by its orientation of binding to the ENL YEATS domain relative to acetyllysine. Green depicts binding to the entrance of the tunnel into which acetyllysine binds. Pink shows the central amide that mimics the binding of the acetyllysine amide and blue depicts binding to the exit of tunnel. (e) Chemical structures of other YEATS domain inhibitors, colored the same as SGC-iMLLT1 including coloring for their E3 ligase. (f) Chemical structures of ENL PROTACs and a molecular glue degrader.

ENL functions as a chromatin reader protein through its conserved YEATS domain, first reported in 2014 to bind acetylated lysine side chains (Figure 20b).²⁸ Four YEATS domains (named for the founding members Yaf9, ENL, AF9, Taf14, and Sas5)⁵⁵¹ are encoded in the human genome (ENL, AF9, GAS41/YEATS4 and YEATS2) and each is part of chromatin regulatory complexes.^552,553 This conserved domain (Figure 20c) is organized into an immunoglobulin fold and forms an open-ended channel between loops 4 and 6 to accommodate acyllysine side chains (Figure 20b).^28,554–557 Early structural and mutagenesis studies of the YEATS domain identified key amino acids involved in binding acetyllysine, which makes π-π-π stacking interactions with Tyr78 and Phe59.^{549,550,554,558} Electrostatic interactions play key roles in acyllysine recognition, including hydrogen bond interactions with the backbone amide of Tyr78 and the Ser58 side chain,⁵⁵⁹ CH-π interactions with Phe28 and the tip of the acetyllysine, and amide-π interactions with Tyr78 and Phe59 (Figure 20b).^28,560,561

Based on the structural analyses of ENL/AF9 interactions with acetyllysine, point mutations have been discovered that decrease acetyllysine affinity in vitro and prevent ENL and AF9 localization to chromatin in cells.^28,549,550 In acute leukemia cells that depend on ENL for survival, these mutant alleles were unable to complement the loss of wild-type ENL, suggesting that interactions between ENL YEATS and acetyllysine are essential for its pro-leukemic function.^549,550 Based on this discovery, there have been widespread efforts to identify small molecules that competitively inhibit the ENL YEATS domain. The first reported inhibitor, SGC-iMLLT, originated from the Structural Genomics Consortium (SGC) in a screen of 40,000 small molecules that measured the displacement of a synthetic histone peptide from the ENL YEATS domain.⁵⁵⁹ Optimized from a benzimidazole hit, SGC-iMLLT1 contains a central amide bond that mimics the acetyllysine substrate by engaging in key hydrogen bond interactions with Ser58, Tyr78, and Ala79 (Figure 20d). The flanking indazole and benzamidazole moieties make contacts with the entrance and exit of the binding tunnel, respectively (the entrance and exit are defined relative to the insertion of acetyllysine). Binding of the native acetyllysine substrate and SGC-iMLLT1 is facilitated by conformational flexibility of Tyr78, which is a more constricted tryptophan residue in YEATS2/4, potentially contributing to its selectivity for ENL/AF9 (Figure 20d). Many ENL/AF9 YEATS domain inhibitors have since been reported, which has revealed recurrent structural themes of YEATS domain inhibitors (Figure 20e). These contain, for example, central amide or urea groups flanked by piperazine, imidazopyridine, and isoxazole groups that contact the exit tunnel and indazole, cyclobutyl, and indane groups that contact its entrance.^{559,562–571}

The imidazopyridine inhibitor, SR-0813, was optimized from a FRET-based HTS hit using a sulfur(VI) fluoride exchange (SuFEx) based high-throughput medicinal chemistry approach (Figure 20e).⁵⁶² SR-0813 is potent and selective for ENL/AF9 YEATS domains, (ENL YEATS K_D = 30 nM) with minimal effects on YEATS2 at high concentrations (10 – 50 μM) and virtually no binding to YEATS4 by SPR. SR-0813 inhibits the proliferation of ENL-dependent, MLL-rearranged leukemias by evicting ENL from chromatin through competitive binding with acetyllysine, resulting in a concomitant downregulation of ENL target gene expression. Promisingly, SR-0813 treatment showed no detectable anti-proliferative effects in cell lines insensitive to ENL loss, indicating that the observed effects in ENL-dependent cell lines are likely due to its on-target activity. This research recapitulated the effects of ENL degradation and validated that the pharmacological inhibition of the ENL YEATS domain represents an effective way to exploit ENL dependencies in MLL rearranged leukemias.⁵⁶² Despite the in vitro potency of SR-0813, however, its poor metabolic stability precluded it from further assessment in vivo.

Modification of SGC-iMLLT produced the first inhibitor capable of probing ENL/AF9 YEATS domain in vivo. This compound, TDI-11055, replaced the benzimidazole of SGC-iMLLT with a pyrrolopyridine group to yield much improved pharmacokinetic properties, including acceptable oral bioavailability (Figure 20e).^565,567 TDI-11055 selectively binds the ENL/AF9 YEATS domains (K_D = 119 nM), prevents the association of ENL with chromatin, and reduces the expression of ENL target genes in MLL-rearranged cell lines. Encouragingly, TDI-11055 is well tolerated in vivo and produces anti-cancer effects in xenograft models of MLL-rearranged leukemias. Furthermore, TDI-11055 was used to uncover a dependency on ENL in NPM1-mutant leukemia and it showed strong anti-cancer effects in animal models of the disease.⁵⁶⁷ This aspect of ENL target biology is notably similar to menin, which is a targetable dependency in both MLL-rearranged and NPM1-mutant leukemia. A CRISPR/Cas9-based ENL mutagenesis strategy was able to identify an in-frame deletion that confers resistance to TDI-11055, validating its anti-leukemic effects as being on-target. Interestingly, this deletion overlaps a hotspot region of recurrent gain-of-function insertion-deletion mutations observed in Favorable Histology Wilms tumors that cause increased biomolecular condensate formation.^572–578 Another reported orally bioavailable inhibitor, ⁵⁶³SR-C-107, is based on an amido-triazolopyridine screening hit,⁵⁶⁸ with the addition of the pyrrolopyridine group from TDI-111055 to improve pharmacokinetic behavior..^564,565,567

SR-0813, TDI-11055, SR-C-107, and other chemical probes for the ENL/AF9 YEATS domains elicit anti-proliferative effects that are highly selective for ENL-dependent leukemias, and their suppression of gene transcription is similarly selective for ENL target genes.^562,564,567 However, their anti-proliferative effects are delayed, requiring several cell divisions and more than a week of drug exposure to cause an effect in vitro. The strengths of their effects are also weaker than compounds inhibiting similar targets, like menin, with high doses required to inhibit ENL-dependent leukemia growth in vitro and in vivo (both TDI-11055 and SR-C-107 are used at 200 mg/kg doses in mice). While these compounds leave significant room to improve affinity, it is likely that YEATS domain inhibition elicits a sub-maximal effect compared to degradation or genetic knockout.^{549,550,563,579,580}

YEATS domain ligands have been used to design several ENL/AF9 PROTACs. A precursor to SR-0813 was used to discover the first ENL/AF9 degrader, the CRBN-based PROTAC, SR-1114.⁵⁶² However, SR-1114 also induces degradation of the transcription factor IKZF1, a common neosubstrate target of CRBN-based degraders, limiting its deployment as a chemical probe. Compound 1, which used SGC-iMLLT to construct a more potent ENL degrader (3-fold over SR-1114) also elicits anti-leukemic effects against ENL-dependent cell lines, but its selectivity for ENL has not been evaluated at proteome scale.⁵⁸⁰ MS41, which was designed by linking the ENL/AF9 YEATS ligand, PFI-6,⁵⁶⁶ to a VHL ligand, is a potent (DC₅₀ = 3.5 nM) and selective degrader of ENL.⁵⁷⁹ In vitro, it selectively inhibits the growth of ENL-dependent leukemia cell lines and in vivo, it suppresses the growth of an MV4;11 xenograft model of acute leukemia without causing toxicity to normal hematopoiesis. Joining these heterobifunctional PROTACs is dHTC1, a molecular glue degrader based on the SR-0813 scaffold (Figure 20f).⁵⁶³ dHTC1 was developed using a new, prospective approach for molecular glue discovery,⁵⁶³ which relies on sulfur(VI) fluoride exchange (SuFEx)-based transformations for parallel ligand diversification.^{562,581–583} This approach allows for the discovery of rare chemical modifications that can transform pre-existing ligands into molecular glues, affording dHTC1, a potent and selective ENL degrader.⁵⁶³ It binds to CRBN with high affinity only after pre-forming a bimolecular complex with ENL, inducing ENL degradation and eliciting anti-leukemic effects in vitro and in vivo.⁵⁶³ While further comparative analysis of YEATS domain inhibitors and ENL degraders is needed, and additional improvement in YEATS domain affinity are likely possible, these early-stage degraders strongly suggest that degradation would be the preferred method to perturb ENL therapeutically. These degraders are excellent tool compounds for assessing ENL degradation as potential therapeutics, but no ENL-targeted agents have advanced into clinical investigation.

YEATS domain inhibitors have also been reported for YEATS4, which was originally identified as an overexpressed oncoprotein in glioma, for which it was named GAS41 (glioma amplified sequence 41).⁵⁸⁴ Like other YEATS-domain-containing proteins, YEATS4 regulates transcription by recognizing post translational modifications on histone tails and recruiting multiprotein complexes to influence gene expression.⁵⁸⁵ YEATS4 amplifications have been found in many indications including sarcomas, non-small cell lung cancers (NSCLC),⁵⁸⁶ colorectal, and gastric cancers.^587–591 Genetic knock out of YEATS4 has been shown to impair proliferation in NCSLC with YEATS4 amplifications.⁵⁹² The prevalence of YEATS4 amplifications in various cancer types and its role in transcriptional regulation suggest that it could be tractable target for small-molecule inhibitors, but it is also annotated as a common essential gene by DepMap (Figure 21a).

Figure 21.

YEATS domain of YEATS4 is chemically tractable. (a) Histogram of gene effect scores for common essential gene YEATS4 (DepMap). (b) YEATS4 YEATS domain inhibitors. Chemical structures are colored based on their orientation of binding to the acetyllysine channel, as described in Figure 19 (blueD= exit, pink = central amide, and green = entrance).

The first reported YEATS4 YEATS domain inhibitor was developed from fragment library screening hit, a thiophene amide scaffold that was later optimized to contain a benzothiazolyl substituent (compound 11).⁵⁹³ Based on findings that YEATS4 binds to histones more strongly as a dimer, a bivalent analog of compound 11, connected by a short polymethylene linker, was developed (Compound 9) (Figure 21b).^593,594 This compound decreases the expression of YEATS4 target genes and showed moderate antiproliferative effects in cancers with YEATS4 amplifications. A second reported scaffold, a pyrazolopyridine series of potent and selective YEATS4 inhibitors with suitable pharmacokinetic properties for use in vivo, was developed from a virtual screening hit.⁵⁹⁵ Interestingly, although YEATS4 is a common essential protein, these compounds show minimal cellular toxicity, which suggests that the YEATS domain of YEATS4 may not be required for its cell-essential functions.⁵⁹²

3.1.5. ASH1L

ASH1L (absent, small, or homeotic-like 1) is a large, 333 kDa protein that contains a SET domain, bromodomain, PHD, and BAH domain, as well as a long, unstructured N-terminal region. The SAM-dependent SET domain catalyzes H3K36 mono- and di-methylation of histone 3 lysine 36 (H3K36) while the other domains bind to posttranslational modifications.⁵⁹⁶ It is expressed in hematopoietic stem and progenitor cells (HSPCs), where is regulates HOX gene expression and supports self-renewal.⁵⁹⁷ This activity is reminiscent of MLL, with the two proteins sharing a conserved functional relationship.^598–600 Indeed, ASH1L regulates an essential writer-reader-eraser axis in MLL-rearranged leukemias, where it catalyzes the deposition of H3K36me2 marks that are recognized by the MLL interactor, LEDGF, and colocalizes with MLL to activate a leukemogenic transcriptional program.⁶⁰¹ Loss of ASH1L mitigates leukemogenesis in MLL-transformed cell lines and xenografts, altogether supporting a model in which ASH1L is critical in MLL-rearranged leukemias.⁶⁰¹ Furthermore, in a mouse model harboring an in-frame deletion of the Ash1l SET domain, bone marrow progenitors are resistant to transformation by MLL-fusion oncogenes but not by other control oncogenes, leading to the pursuit of ASH1L methyltransferase inhibitors.⁶⁰²

The first ASH1L inhibitors originated from an NMR-based screen of approximately 1,600 compounds. The chemical shifts from a thioamide hit mapped to a region of the protein near an autoinhibitory loop, which was confirmed via point mutation. Medicinal chemistry optimization of this compound yielded a compound, AS-5, that binds to a buried site and is conformationally restrained by the concomitant binding of the endogenous SAM cofactor (Figure 22). This compound features an indole scaffold with hydroxyethyl and aminomethyl substituents that engage in stabilizing hydrogen-bonding interactions within the pocket. These were replaced with 1-trifuloromethylsulfonylpiperidine and methyl azetidine moieties, respectively, to yield the lead compound AS-99, which binds in a similar fashion. The thioamide engages in a chalcogen-bonding interaction with the backbone carbonyl of H2193 and replacement of the sulfur with oxygen abolishes activity, providing an excellent negative control compound (Figure 22). AS-99 binds the SET domain with a K_D of 0.89 μM and inhibits ASH1L with an IC₅₀ of 0.79 μM. It is selective against 20 other HMTs including NSD1, NSD2 and NSD3, as well as SETD2, likely due to the lack of structural conservation in their autoinhibitory loops. AS-99 is antiproliferative in cell lines harboring MLL fusions and recapitulates the pro-differentiation phenotype seen with genetic ASH1L knockdown and SET domain deletion. A decrease in H3K36me2 abundance is seen after drug treatment, which produces a dose-dependent decrease in the transcription of gene signatures associated with MLL fusions. The compound also decreases leukemic burden in mice. Altogether, this report provides a compelling proof-of-concept for selectively inhibiting ASH1L, further demonstrating its potential as a therapeutic target.⁶⁰²

Figure 22.

Summary of medicinal chemistry results from hit (AS-5) to lead (AS-99) inhibitor of ASH1L., as well as negative chemical control (AS-nc).

3.2. PRC2

Polycomb repressive complex 2 (PRC2) is a multi-subunit chromatin regulatory complex best known for its role in transcriptional repression. Mechanistically, PRC2 functions through its enzymatic subunit, enhancer of zeste homologue 2 (EZH2), which catalyzes the methylation of H3K27, a canonically repressive histone PTMs (notably, PRC2 is the exclusive writer of H3K27me3). EZH2 exists in the PRC2 complex alongside suppressor of zeste 12 homolog (SUZ12) and embryonic ectoderm development (EED), a scaffolding and chromatin reader protein, respectively, which are indispensable for the histone methyltransferase activity of PRC2 (Figure 23a).^326,603,604 As first discovered in Drosophila and later shown to be conserved in humans, PRC2 activity is counteracted by that of SWItch/Sucrose Non-Fermentable (SWI/SNF).^605,606 While SWI/SNF remodels chromatin to a more open and active regulatory state through ATP-dependent remodeling of nucleosomes—allowing transcription factors and transcriptional machinery to access DNA within cis-regulatory elements—PRC2 mediates gene-silencing through its methyltransferase activity. This oppositional relationship is also directly competitive. Genetically engineered systems that allow for rapamycin-dependent recruitment of the BAF complex to a specific genomic locus have shown that the most immediate effect of BAF recruitment, even preceding chromatin decondensation, is the eviction of PRC2, as evidenced by anti-Ezh2 ChIP and proxied by anti-H3K27me3 ChIP.³²⁶

Figure 23.

PRC complexes mediate repressive transcriptional markers and play a role in malignant transformation. (a) PRC1 and PRC2 complexes. (b) PRC2 exists in precise regulatory balance with SWI/SNF. Genetic aberrations that disrupt this balance play pivotal roles in oncogenesis.

Genetic alterations that disrupt the balance between BAF and PRC2 can lead to tumor formation (Figure 23b). Hyperactivation of PRC2 in cancer can be caused by mutations^607–609 and amplifications^610–615 of EZH2, as well as by the inactivation of certain BAF subunits.^616–621 Activating point mutations at tyrosine 641 (Y641) (which include Y641F, Y641N, Y641S, and Y641H) are present on a single allele of EZH2 in a large percentage of non-Hodgkin lymphomas—a diverse group of cancers that arise from malignant proliferation of B-, T-, or NK-lymphocytes outside the classic Hodgkin’s lymphoma subtype—creating an oncogenic hypermorph that works cooperatively with the wild-type protein to drive a chromatin architecture with elevated levels of H3K27me3 (Figure 23b).^607,622

Wild-type EZH2 is most active as a mono-methyltransferase, with substantially reduced activity for catalyzing the second and third methylation. In contrast, mutated EZH2 has highest activity for the final methylation and little activity for catalyzing mono-methlyation. Hyperaccumulation of trimethylated substrates occurs in cells expressing both the wild-type and mutant enzymes, which cooperate to catalyze the early and late methylations.^607,622 EZH2 overexpression can also result from loss-of-function mutations within the SWI/SNF complex. For example, malignant rhabdoid tumors (MRT)—a rare and highly aggressive pediatric cancer most commonly originating in the kidney or CNS—are nearly uniform in harboring biallelic inactivation of SMARCB1,⁶²³ which leads to EZH2 overexpression and hypersensitivity to EZH2 loss-of-function.⁶²⁴ These genetically restricted sensitivities to EZH2 loss-of-function motivated the development of PRC2 inhibitors. Because EZH2 is the enzymatic subunit, early drug discovery efforts focused heavily on orthosteric inhibitors of its SET catalytic domain. While this has yielded clinically approved agents, to be discussed in the following subsections, resistance mutations in EZH2 highlight an important limitation of targeting only the catalytic site. This has led to growing interest in other PRC2 subunits—particularly EED, whose interaction with EZH2 is required for enzymatic activity—as alternative therapeutic entry points. For example, the allosteric EED inhibitor EED226, also to be discussed, can suppress PRC2 activity even in the setting of EZH2 mutations, establishing a compelling rationale for exploring non-catalytic PRC2 subunits as drug targets.

3.2.1. EZH2

The molecule 3-deazaneplanocin A (DZNep), an S-adenosylhomocysteine hydrolase inhibitor, elicits downstream by-product inhibition of SAM dependent methyltransferases, a large superfamily of enzymes, consisting of over 200 enzymes in humans alone to which EZH2 belongs.^625–627 While the discovery of this molecule contributed to initial investigations of protein function, it was found to be insufficiently specific, resulting in changes to global chromatin architecture that could not be reliably sourced to EZH2 inhibition. This called for direct-acting and selective small molecules that inhibit EZH2/PRC2 activity.

Groups at Epizyme, GSK, and Novartis ushered in the first of these molecules with the disclosure of SAM-competitive EZH2 inhibitors in 2012.^628–630 Epizyme first reported EPZ005687, which was discovered through the screening of 175,000 compounds against recombinant PRC2, followed by an additional 5,000 compounds representing 25 standout structural clusters (Figure 24a). The resulting hit was a pyridone-containing compound, which was further optimized to improve solubility and potency, yielding EPZ005687.⁶²⁸ This first report highlighted what would continue to be a common hurdle in PRC2 inhibitor discovery – that it is difficult to generate both apo- and co-crystal structures of ligand bound to full PRC2 complex, making structure-based drug design and optimization difficult (although recent efforts toward structural elucidation may be changing this for future work).^631–635 Nevertheless, the reported compound demonstrated competitive antagonism with the native SAM substrate, while remaining noncompetitive with oligonucleosomes. EPZ005687 selectively inhibits EZH2 (K_i = 24 nM) over EZH1 and other methyltransferases (50 and >500 - fold selectivity, respectively) resulting in the decrease of H3K27me3 marks in diverse cell types while sparing other post-translational modifications. Consistent with genetic studies, the antiproliferative activity of EPZ005687 is selective for EZH2-mutant cancer cell lines over EZH2-wildtype cancers.⁶²⁸

Figure 24.

Disruption of PRC2 by small-molecule inhibitors of EZH2 methyltransferase. (a) Prototypical chemical probes targeting EZH2. (b) Clinically evaluated small-molecule inhibitors of EZH1/EZH2.

Contemporaneously, GSK disclosed a similar pyridone EZH2 inhibitor, which had resulted from the screening of an internal compound collection.⁶²⁹ Like EPZ005687, the key pharmacophore consists of a pyridone, 4,6-disubstitution pattern, a linking amide, and a branched alkyl group at the 1-azaindazole. Medicinal chemistry optimization led to GSK126, an EZH2-selective, SAM-competitive, and in vivo-active inhibitor that elicits a transcriptional activation profile correlated to antiproliferative activity in EZH2-mutant cancers over wild-type cancers (Figure 24a).^629,636 EI1, another pyridone EZH2 methyltransferase inhibitor, was also disclosed in 2012 by Novartis (Figure 24a). EI1 invokes cell cycle arrest and apoptosis in EZH2-mutant DLBCL cancer cells and reactivates expression of PRC2 target genes.⁶³⁰ A series of other compounds, UNC1999, CPI-169 and CPI-360, were disclosed shortly after and used to validate similar biological activity, namely, an antiproliferative phenotype in EZH2-mutant DLBCL cells that correlates with a global reduction of H3K27me3 marks.^630,637,638

The first, and as-yet only, EZH2 inhibitor to be approved by the FDA is EPZ-6438, or tazemetostat (Figure 24b).^624,639 Tazemetostat featured remarkable improvements in potency, selectivity, and pharmacokinetic properties compared to the previous tool compound, EPZ005687. In lymphoma cell lines bearing EZH2 point mutations, treatment with tazemetostat leads to G1 arrest and induction of apoptosis, with cell death correlating to a commensurate reduction in H3K27me3 abundance. PRC2/EZH2 target genes are upregulated upon compound treatment, indicating a beneficial transcriptional response elicited by a change in the chromatin landscape. In mice and rats, the compound demonstrates low clearance with steady volumes of distribution, oral bioavailability, and rapid absorption, among other beneficial drug-like properties. In mouse xenograft studies, doses as low as 34.2 mg/kg are sufficient to inhibit tumor growth while higher doses, above 114 mg/kg, are sufficient to eradicate xenograft tumors, with no tumor regrowth after drug cessation. These promising results led to the clinical exploration of tazemetostat and its eventual FDA approval for follicular lymphoma in 2020.^640,641 In a phase ll clinical trial (NCT01897571) of 42 patients with EZH2 mutant follicular lymphoma, there was an overall response rate of 69% (12% complete and 57% partial responses); these numbers were lower in patients with EZH2 wild-type follicular lymphoma, concordant with preclinical mechanistic results.⁶⁴²

Epizyme further optimized the tazemetostat scaffold, seeking to mitigate oxidative metabolic liabilities associated with the pyridone-benzamide core. Modification of the pyran of tazemetostat, addition of a methoxyethyl group to the cyclohexaylamine, and replacement of the second benzene ring with an acetylene linker produced EPZ011989.⁶⁴³ While EPZ011989 shows excellent oral bioavailability and improved metabolic stability, it did not supplant the development of tazemetostat. Others have explored the degree of selectivity between EZH2 and its paralog, EZH1. Tazemeteostat and EPZ011989 are 35-fold and 15-fold selective for EZH2 over EZH1, respectively. Compounds that inhibit EZH1 more effectively than tazemetostat, such as the dual inhibitor valemetostat, may be able to address some cancers where EZH1 can compensate for EZH2, although dual inhibition carries potential limitations, such as increased toxicity (Figure 24b).⁶⁴⁴ While valemetostat does not carry the FDA approval that tazemetostat does, it is being evaluated in trials for peripheral T-cell lymphoma (PTCL) and adult T-cell leukemia/lymphoma (ATLL).⁶⁴⁵

Tazemetostat is also highly efficacious against MRT. The biallelic inactivation of the SWI/SNF subunit, SMARCB1, which is pathognomonic for MRT, sensitizes cells to inhibition of EZH2.⁶⁰⁶ In a 2013 report, EZH2 inhibition was shown to inhibit the growth of SMARCB1-deficient MRT cells in vitro and cause tumor regression in xenograft models of MRT in vivo.⁶²⁴ Based on positive clinical studies,⁶⁴⁰ tazemetostat has now been approved for the treatment of SMARCB1-mutant sarcoma, but response rates seen in sarcoma are well below those seen in EZH2-mutant lymphoma.

3.2.2. EED

EZH2 functions within the PRC2 complex together with EED and SUZ12, and disrupting these interactions can also block PRC2 activity. A hydrocarbon-stapled peptide called SAH-EZH2 was designed to mimic a short helical region of EZH2 that normally binds EED. By occupying this interface, SAH-EZH2 prevents EZH2 from associating with EED, leading to EZH2 destabilization. In leukemia models, this approach halted cell growth and promoted differentiation, illustrating that PRC2 can be inhibited by breaking apart the complex rather than directly targeting its catalytic site.^646,647

EZH2 catalytic activity can also be modulated through inhibition of EED, which allosterically activates EZH2 by binding H3K27me3 and other methylated lysines.^{604,647–649} EED is a WD40-repeat protein, folding into a seven-bladed β-propeller structure that recognizes trimethyllysine within an aromatic cage.^604,650 In 2017, groups at Novartis and Abbvie reported the potent and selective allosteric PRC2 inhibitors, EED226 and A-395, which function by disrupting the interaction of EED with methylated lysine (Figure 25a).^651,652 The EED226 scaffold was identified in a high throughput screen using homogenous time-resolved fluorescence (HTRF) to detect the methylation of H3K27 by recombinant PRC2,^651,653,654 a screen that had previously yielded SAM-competitive EZH2 inhibitors.⁶³⁰ Characterization of non-competitive inhibitors led to the discovery of a triazolopyridine scaffold that binds competitively with H3K27me3 to the WD40 domain of EED.^651,654 Treatment of G401 cells with EED226 leads to an inhibition of H3K27 methylation, which is concomitant with a transcriptional response that mirrors orthosteric EZH2 inhibition.⁶⁵¹ EED226 is effective in vivo in DLBCL xenograft models and represents an opportunity to overcome resistance to orthosteric EZH2 inhibitors. In preclinical models, forward genetic approaches have identified secondary EZH2 mutations causing acquired resistance to SAM-competitive EZH2 inhibitors.^655,656 In cellular models, EED226 was able to overcome this resistance, suggesting that combination or sequential therapy with EZH2 and EED inhibitors could prevent or overcome resistance.⁶⁵¹

Figure 25.

EED inhibitors and other methods to disrupt PRC2. (a) First-in-class EED chemical probes. (b) Other representative PRC2 inhibitors. (c) Summary of improvement of EED degrader UNC6852 to UNC7700.

A-395, a structurally divergent EED inhibitor that also binds the H3K27me3 recognition pocket of EED, was reported contemporaneously with EED226 (Figure 25a). It effectively reduces H3K27 methylation, inhibits cancer cell proliferation, and demonstrates potent anti-tumor efficacy in preclinical xenograft models.⁶⁵² Importantly, A-395 retained activity in cell lines resistant to EZH2 inhibitors, suggesting its potential to overcome therapeutic resistance. Its PK properties, however, limit its clinical application, although it, alongside EED226, represented a valuable chemical tool for probing PRC2 biology.

In 2022, Novartis disclosed significant improvements to the tool compound EED226, fashioning a more drug-like molecule MAK683 (Figure 25b). EED226 suffers from undesirable solubility and permeability, and the furan moiety is prone to oxidation and reactive metabolite formation, making it a poor clinical candidate. To this end, Novartis focused on three main moieties: (1) the furan, which resides in a deep pocket that prefers electron rich aromatic systems; (2) the triazolopyrimidine core, which resides in the aromatic cage that prefers electron deficient systems; and (3) the solvent-exposed sulfone, which is inherently permissive to change and a desirable location for property modulation. Studies illustrated that CYP3A4 and CYP11A1 act on EED226 to form reactive species, particularly through modification of the furan.^657,658 Replacement of the furan with a dihydrobenzo furan yielded significant improvements and fluorination of the ring helped to improve off-target liabilities. Additional modifications to shield the compound from heme ligation and decrease its metabolic reactivity produced MAK683, which is effective against Karpas422 xenograft models and demonstrates favorable cross-species ADME properties, including little to no CYP inhibition.⁶⁵⁷ It was eventually progressed into clinical evaluation for the treatment of DLBCL, but a full report of the trial results has not been disclosed, and further development was halted before Phase 2.⁶⁵⁹

Non-enzymatic scaffolding functions of EZH2 have been reported to be important for SWI/SNF-mutant tumors, providing a rationale to develop targeted protein degradation approaches for EZH2.^53,660,661 Small-molecule inhibitors of EZH2 and EED have enabled the development of PROTACs and other targeted degradation approaches to inactivate PRC2 function. The first example, MS-1943, which degrades EZH2 by hydrophobic tagging, was synthesized by appending an adamantane group to the EZH2 ligand C24 (Figure 25b).⁶⁶² This molecule degrades EZH2, causing collateral degradation of SUZ12 but not EED, as EED is a core scaffold component that can exist stably independently and is not dependent on EZH2. The degrader suppresses the H3K27me3 mark, albeit less effectively than the parent compound C24. MS-1943 is, however, more cytotoxic in EZH2 dependent cell lines than C24, due to chronic stimulation of the unfolded protein response caused by compound-induced endoplasmic reticulum (ER) stress. This may be an off-target activity of the adamantane tag, as this effect has not been reported for PRC2-targeted PROTACs. Such PROTACs include UNC6852, a bivalent degrader designed by connecting a EED226 to a ligand for the E3 ubiquitin ligase VHL (Figure 25c).⁶⁶³ UNC6852 degrades EED and causes the collateral degradation of the PRC2 components EZH2 and SUZ12, which depend on EED for proper folding and stabilization. Disruption of the PRC2 complex results in reduced H3K27me3 levels in EZH2-Y641N-mutant cells and is antiproliferative in DLBCL. Other PROTACs have been built on both EED ligands and EZH2 inhibitors.^664,665 UNC7700, an analog of UNC6852 that implements a cis-cyclobutane linker in place of the more flexible polymethylene linker, shows greatly improved potency of degradation (Figure 25c).⁶⁶⁶

A recent study also suggests that covalent engagement of EZH2 is sufficient to result in the depletion of EZH2 protein levels.⁶⁶⁷ IHMT-337, which covalently binds EZH2 at Cys663 was discovered from a collection of 200 compounds designed to irreversibly inhibit EZH2 (Figure 25b). EZH2 is subsequently degraded via the CHIP-mediated ubiquitination pathway. Interestingly, this molecule was useful in elucidating a non-enzymatic, PRC2-independent function of EZH2 as a transcriptional cofactor for CDK4. The compound is not only effective in EZH2-mutant DLBCL cells, but also in triple negative breast cancer (TNBC), which are not dependent on EZH2 enzymatic activity. Several other degraders of EZH2 and EED have been reported, combining various PRC2-ligands with various E3-ligase ligands to fashion heterobifunctional molecules capable of depleting protein levels in vitro and in vivo.^{663–665,668,669}

While allosteric approaches to target PRC2 show preclinical activity against EZH2-mutant lymphoma similar to orthosteric inhibition, their ultimate success in the clinic remains uncertain. They might enable major therapeutic advances if able to overcome mutations causing acquired resistance to SAM-competitive EZH2 inhibitors. Indeed, there is preliminary evidence that acquired EZH2 mutations identified to promote resistance in cell culture are also observed in patients treated with tazemetostat.⁶⁷⁰ Ongoing clinical trials will determine if EED inhibitors, such as ORIC-944, which is being tested in prostate cancer (NCT05413421), can provide therapeutic benefit as single agents or combination therapeutics.⁶⁷¹

3.2.3. PRC1

PRC1 is a multiprotein complex that functions together with PRC2 to enable gene silencing. It possesses a core E3 ubiquitin ligase, RING1A/RING1B, a chromobox (CBX) reader protein, and one of 6 polycomb group ring finger (PCGF) paralog proteins (Figure 26a).⁶⁷² Of the 6 PCGF proteins, PCGF4, also called BMI1, is a member of the canonical PRC1 complex. CBX proteins recognize repressive H3K27me3 marks deposited by PRC2 and the PRC1 complex mono-ubiquitinates H2AK118 and H2AK119 via RING1A/B.⁶⁷³ Together, PRC1 and PRC2 silence gene transcription to promote stemness during development and cancer.⁶⁷⁴ Interestingly, continual BMI1 expression has been shown to be selectively required for the proliferation of leukemia stem cells and it has therefore has been proposed as an attractive target for small molecule inhibition.⁶⁷⁵ It has also been suggested as a target in multiple other indications beyond AML.^676–678 E3 ligase activity of RING1B/A is essential for the maintenance of leukemia stem cells and leukemogenesis, which has inspired several attempts to inhibit H2A ubiquitination (H2Aub) levels using small molecules.^679–684

Figure 26.

PRC1 inhibition. (a) Chromatin regulation by PRC1. (b) chemical structures of RING1B inhibitors RB-2 and RB-3. (c) Chemical structures of RING1B/BMI1 PROTACs constructed with ligands binding EED (purple) and the E3 ligase CRBN (yellow) or VHL (grey).

Initial efforts to discover small molecule inhibitors of RING1B and BMI1 focused on inhibiting the ligase activity of PRC1. Small molecule chemical screens that used H2Aub signal as the output led to the discovery of two compounds, GW-519 and PRT4165.^685,686 Both compounds inhibit H2Aub and decreased proliferation, but their mechanisms of action have not been elucidated. Screening for compounds that decreased BMI1 expression has led to the discovery of PTC-209 which decreases BMI1 expression in colorectal cancer.⁶⁸⁰ The mechanism of PTC-209 remains unknown, but its proposed effect is on BMI1 translation leading to the observed reduction in BMI1 protein expression. While targeting the PRC1 through disruption of BMI1 scaffolding function remains an attractive strategy, the mechanism of PTC-209 has not been the focus of active investigation. Instead, recent efforts have shifted toward the development of next-generation compounds with improved pharmacological properties and less convoluted mechanisms of action.^687,688

The first direct-acting and mechanistically characterized inhibitor to be reported was discovered in a fragment screen using ¹H-¹⁵N heteronuclear single quantum coherence (HSQC) experiments and a ¹⁵N-labeled RING1B-BMI1 fusion protein.⁶⁷⁹ A benzene- and thiophene-containing fragment hit was found to bind the RING1B-BMI1 fusion protein in the mM range. Subsequent medicinal chemistry, including replacement of the thiophene ring with a pyrrole to increase solubility, yielded RB-2, which binds with 11.5 μM affinity to the RING domain of RING1B (Figure 26b). Replacement of the RB-2 ethyl with an isopropyl and the chloro-phenyl with a chloro-indole led to increased hydrophobic interactions with RING1B and improved affinity to 2.8 μM. The resulting compound, RB-3, inhibits ubiquitination in vitro and in cells and causes a decrease in PRC1 localization to its target genes.⁶⁷⁹ Further optimization has been employed on this scaffold,⁶⁸⁹ but improvements in potency and in vivo activity are crucial to more fully explore RING1B as a potential therapeutic target.

PRC1 can also be targeted by inducing collateral degradation of BMI1 and RING1B through EED-based PROTACs.⁶⁹⁰ This discovery was motivated by the previous observation that an EED PROTACs can induce the degradation of additional PRC2 subunits.^663,664 EED has been shown to incorporate into PRC1 with RING1B and BMI1,^691,692 suggesting that EED PROTACS may also be able to induce the degradation of PRC1 components.⁶⁹⁰ MS147, a heterobifunctional construction of ligands for EED and VHL, preferentially degrades RING1B and BMI1 over PRC2 components, including EED, causing a decrease in H2AK119ub without altering H3K27me3.⁶⁹⁰ Interestingly, MS147 shares a similar VHL and EED ligand to the previously reported EED/PRC2 degrader, PROTAC 2 (Figure 26c).⁶⁶⁴ The main difference between these compounds is linker length, which contributes to MS147 inducing degradation of both RING1B and BMI1, but only degrading PRC2 subunits at high concentrations. Similar results were observed with MS181, a CRBN-based EED degrader that induces PRC1 subunit degradation (Figure 26c).⁶⁹³ However, proteome-scale measures of selectivity have not been reported for MS147 and MS181.

3.3. SWI/SNF

The mammalian SWI/SNF (switch/sucrose non-fermentable) complex, also known as the BAF (BRG-/BRM-associated factor) complex, is an ATP-dependent chromatin remodeler complex and transcriptional co-activator.^694,695 A diverse collection of protein products from 29 genes are incorporated into three distinct assemblies of the BAF complex: canonical BAF (cBAF), polybromo-associated BAF complex (PBAF), and the recently described non-canonical BAF (also reported as GLTSCR1 BAF, or GBAF, hereafter referred to as ncBAF) (Figure 27).^696–699 While all three complexes share a core set of subunits, each distinct complex specifically incorporates certain subunits not shared by the others. ARID1A/ARID1B and DPF2 are incorporated into cBAF; PBRM1, ARID2, and BRD7 are incorporated into PBAF; and GLTSCR1/GLTSCR1L and BRD9 are incorporated into ncBAF (Figure 27). Both SMARCA2 or SMARCA4, the paralogous ATPase subunits of the complex, can be incorporated into each of the three complexes. They couple the hydrolysis of ATP to the disruption of histone-DNA contacts to open chromatin structures within cis-regulatory elements of the genome and allow transcriptional activation.⁷⁰⁰ Chromatin immunoprecipitation using antibodies specific to each complex (e.g., anti-SMARCA2/SMARCA4 for pan-SWI/SNF recognition, anti-BRD9/GLTSCR1 for ncBAF specific recognition, anti-DPF2 for cBAF specific recognition, and anti-BRD7 for PBAF specific recognition) have shown that ncBAF, cBAF and PBAF are differentially localized to promoters, distal sites, and gene bodies, respectively (Figure 23). Furthermore, like SMARCA2/4, many BAF subunits can be alternately fulfilled by close paralogs, leading to important functional diversifications of BAF complexes in specific cell types.⁷⁰¹ The hyper-diversity of these complexes has been a rich source of biological investigation and underlies a number of cancer-specific vulnerabilities, whereby an imbalance in BAF activity makes other chromatin regulatory proteins indispensable for cell fitness.

Figure 27.

Graphical representation of BAF complexes, with subunits uniquely incorporated into each complex labeled. Complexes are centered over their asymmetric genomic destinations.

The SWI/SNF complexes are large, multi-subunit transcriptional regulators that have specialized functions across organismal development and are recurrently mutated in diverse human malignancies.⁷⁰² Synovial sarcoma (SS) and malignant rhabdoid tumor (MRT) are among the clearest examples of SWI/SNF-driven tumors, where nearly all cases are driven by a specific SWI/SNF alteration. In SS, the t(X;18)(p11.2;q11.2) translocation is found in virtually all tumors and considered pathognomonic for disease.^703–706 The result is an onco-fusion protein SS18-SSX, which is defined by the replacement of eight C-terminal amino acids of SS18 with the 78 C-terminal amino acids of one SSX protein paralog.^704–706 This fusion product is the sole genetic anomaly in a cancer characterized by an otherwise remarkably low mutational burden, and is causal of disease.⁷⁰⁷ SS18 is a subunit of the SWI/SNF complex and, when fused to SSX, its incorporation leads to eviction of SMARCB1 (also known as BAF47/SNF5/INI1) from the complex (Figure 28a).^708,709 This causes SWI/SNF to be redistributed on the genome, dysregulating its opposition to PRC-mediated gene repression and activating a synovial sarcoma gene signature.^710,711

Figure 28.

BRD9 is an induced sensitivity in SWI/SNF-mutant tumors. (a) Schematic of SMARCB1 mutations in synovial sarcoma and malignant rhabdoid tumor. (b) Representative small-molecules chemical probes that inhibit the BRD9 bromodomain. (c) Development of BRD9 PROTACs based on BI9564.

3.3.1. BRD9 synthetic lethality

In MRT, SMARCB1 is biallelically inactivated in up to 95% of all tumors (Figure 28a).^712,713 This alteration evokes a synthetic lethal dependency on BRD9 and the ncBAF complex, which was discovered in an unbiased genetic screen and shown further with chemical degraders of BRD9.⁷⁰¹ It has also been validated by multiple independent groups.⁷¹⁴ A similar dependency on BRD9 is seen in synovial sarcoma, where the degradation of BRD9 disrupts SS18-SSX complexes, downregulates oncogenic genes, and inhibits tumor growth.⁷¹⁵ BRD9 is essential for the incorporation of GLTSCR1, a defining subunit of the non-canonical BAF (ncBAF) complex. GLTSCR1 functions as the signature scaffolding subunit that distinguishes ncBAF from canonical and PBAF complexes, enabling ncBAF to localize to specific chromatin sites and regulate a distinct set of genes. Without GLTSCR1, ncBAF cannot assemble into a functional complex. In the setting of SMARCB1 loss, BRD9 incorporation is increased, driving ncBAF formation and creating an acquired vulnerability to BRD9 inhibition (Figure 28a).

Given the discovery of BRD9 as a genetically defined dependency in SWI/SNF-mutant tumors, the protein has become an attractive target for drug discovery. However, BRD9 chemical probes preceded the identification of BRD9 as a potential drug target and were critical to the discovery and validation of BRD9 as a cancer-specific dependency. Like many other bromodomain proteins, BRD9 bromodomain inhibitors were widely pursued after the first generation of BET bromodomain inhibitors established the ligandability of this class. LP99, disclosed in 2015, is a sub-micromolar, selective inhibitor of BRD7/9 bromodomains, discovered using complexity-building cascade chemistry to assist with structure-guided optimization of a quinolone fragment hit (Figure 28b).⁷¹⁶ Other early compounds include triazolophthalazine⁷¹⁷ and 9H-purine⁷¹⁸ scaffolds that compete with acetyllysine to bind the bromodomain of BRD9. However, these early compounds did not possess sufficient activity and selectivity to faithfully study BRD9 biology.

The thienopyridone, I-BRD9, was disclosed in 2016 and reported to possess >700-fold selectivity against the BET bromodomain family and 200-fold selectivity against its paralog, BRD7, representing a major advance for BRD9 tool development (Figure 28b).⁷¹⁹ This selectivity is driven by a charged amidine favoring the less hydrophobic environment of BRD9 relative to other bromodomains. A structurally divergent pyridone series with suitable pharmacokinetic and pharmacodynamic (PK/PD) properties to interrogate BRD9 biology in vivo was reported shortly after I-BRD9.⁷²⁰ These compounds, exemplified by the chemical probes, BI-7273 and BI-9564 (Figure 28b,c), originated from a fragment library screen and virtual ligand screening, which converged on pyrimidinone and dimethylpyridinone cores as starting points for structure-based medicinal chemistry. Elaboration of the fragments yielded BI-7273, which binds with 15 nM affinity to BRD9 and does not bind BET bromodomains up to 100 μM (Figure 28b). This improved selectivity, compared to LP99 and I-BRD9, proved exceptionally useful for studying BRD9 biology. Based on prior studies demonstrating a role for the SWI/SNF ATPase, SMARCA4, in acute leukemia,⁷²¹ BI-7273 was evaluated in acute leukemia cells and compared to genetic BRD9 loss-of-function perturbations.⁷²² Both chemical and genetic perturbations demonstrated cell-type-selective anti-leukemic activity, which could be positively attributed to BRD9 through genetic domain-swap experiments that made drug-resistant alleles of BRD9. BI-9564, a more selective inhibitor from this series, also inhibits cancer cell proliferation in vitro, demonstrates an acceptable pharmacokinetic profile, and is efficacious in a murine model of AML (Figure 28c).⁷²⁰ These chemical probes are joined by several more examples of structurally diverse BRD9 bromodomain ligands.^723–726

In contrast to acute leukemia cells, which are sensitive to BRD9 bromodomain inhibition, multiple independent reports have demonstrated that SS and MRT are dependent on BRD9 for survival but insensitive to BRD9 bromodomain inhibition, suggesting an essential scaffolding role rather than a dependence on its acetyl-lysine-binding bromodomain.^701,714,715 This suggested that chemically induced degradation of BRD9 would be required to effectively exploit BRD9 vulnerabilities in SWI/SNF-mutant tumors. A series of CRBN-based BRD9 PROTACs had previously been constructed from I-BRD9 or BI-7273, including dBRD9, a BI-7273-pomalidomide heterobifunctional that induces potent and selective BRD9 degradation (Figure 26c).²¹⁸ Mass spectrometry (MS)-based proteomics has shown that dBRD9 induces the degradation of BRD9 exclusively. No other proteins were found to be degraded by dBRD9, including BRD7, the paralogous target of BI-7273. This was a particularly interesting discovery, as it was one of the first examples showing that a multi-targeted ligand can be used to construct a more selective degrader, now a well-established feature of PROTAC pharmacology.²²²

The PROTAC, dBRD9, and an optimized analog, dBRD9-A, were subsequently used to demonstrate that synovial sarcoma and malignant rhabdoid tumors are hypersensitive to BRD9 degradation (Figure 28c).^701,715 The BRD9-specific BAF complex, ncBAF, is a critical dependency in SS and MRT. This was discovered using CRISPR-Cas9-based gene essentiality screens, in which a pooled library of single guide RNAs is introduced into a population of cells, where each guide directs Cas9 to disrupt a specific gene. By tracking which guides become depleted or enriched over time, one can identify genes that are essential for cell survival or confer selective advantages under defined conditions. This experiment demonstrated that both SS and MRT cell lines are selectively sensitive to loss of BAF subunits that are specific to ncBAF.⁷⁰¹ Independently, a domain-focused CRISPR screen, which can be used to identify domain-specific dependencies within protein targets,⁷²⁷ demonstrated that the bromodomain of BRD9 is important to the fitness of SS cells but dispensable for other sarcomas.⁷¹⁵ These results were corroborated with rescue experiments showing that wild-type, but not bromodomain-inactive, BRD9 can support SS fitness. Interestingly, an additional domain, containing amino acids 311–345, was identified as being necessary for association into the ncBAF complex.⁷¹⁵ These amino acids are within the DUF3512 domain (amino acids 274–505), which was similarly reported by another group to be necessary and sufficient for BRD9 incorporation into ncBAF.⁷⁰¹ This was hypothesized to explain the observation that BRD9 degradation is more effective at inhibiting the proliferation of SS cells than BRD7/9 bromodomain inhibition.⁷¹⁵ Ultimately, these findings motivated the clinical development of BRD9 PROTACs, including CFT8634 (C4 Therapeutics, NCT05355753) and FHD-609 (Foghorn Therapeutics, NCT04965753), for the treatment of SS and MRT. However, both trials were terminated due to FHD-609 causing dose-limiting cardiotoxicity and CFT8634 showing a lack of therapeutic efficacy despite efficient BRD9 degradation, according to company press releases.

3.3.2. SMARCA2/4 paralog synthetic lethality

SWI/SNF complexes can incorporate one of two paralogous ATPase subunits, SMARCA2 (SWI/SNF Related, Matrix Associated, Actin Dependent Regulator of Chromatin, Subfamily A, Member 2; also known as BRM) and SMARCA4 (also known as BRG1), which share 75% sequence identity, 93% conserved ATPase domain, and functional redundancy (Figure 29a). These subunits are mutually exclusive with each other and highly homologous, possessing catalytic ATPase domains and acetyl-lysine reader bromodomains.⁷²⁸ Recurrent homozygous mutations of SMARCA4 are found in 10–35% of NSCLC,^{619,729–732} 15% of Burkett Lymphoma,^617,733 5–10% of childhood medulloblastoma,^734–736 as well as in pancreatic adenocarcinoma,⁷³⁷ melanoma,⁷³⁸ and other cancers, while SMARCA2 is often epigenetically silenced in cancer.⁶²⁰ The structural and functional redundancy of SMARCA2 and SMARCA4 create a potentially targetable synthetic lethality whereby SMARCA2 is essential for the survival of SMARCA4-deficient tumors (Figure 29a). We have also included DepMap data for these genes (Figure 29b).

Figure 29.

SMARCA2 is a synthetic lethality in SMARCA4-deficient cancers. (a) Schematic of SMARCA2 synthetic lethality. (b) Histogram of gene effect scores for the genes SMARCA2 and SMARCA4 (DepMap). (c) Chemical probe for the SMARCA2/4 bromodomain. (d) Chemical probes that inhibit SMARCA2/4 ATPase activity. (e) Degraders of SMARCA2/4. (f) Development of optimized PROTACs, such as SMD-1087, with SMI-1074, a recently reported SMARCA2/4 bromodomain ligand.

This synthetic lethal relationship was first discovered in a hypothesis-driven study. Here, RNAi-based silencing of SMARCA2 expression was shown to suppress the growth of SMARCA4-deficient tumors through the p21/CDKN1A axis, whereby p53-induced expression of the cyclin-dependent kinase inhibitor p21 enforces cell-cycle arrest, indicative of a senescent phenotype.⁷³⁹ However, SMARCA2 knockdown had no effect on non-transformed fibroblasts that express both SMARCA2 and SMARCA4, highlighting a clear synthetic lethal relationship. This senescent phenotype is rescued through ectopic expression of wild-type SMARCA4 but not mutant SMARCA4 alleles that inactivate the ATPase domain, nominating its catalytic domain as a potential target for drug discovery. However, as noted by the authors of this study, some cancers show genetic alterations affecting both SMARCA2 and SMARCA4, potentially pointing to a limitation of exploiting SMARCA2/4 synthetic lethality.⁷³⁹

Shortly after this publication, two concurrent manuscripts reported the discovery of SMARCA2/4 synthetic lethality through unbiased, large-scale genetic screens.^621,740 In the first, a collection of pooled RNAi essentiality screens performed across hundreds of cell lines (Project Achilles), identified SMARCA2 as the most differentially essential gene in cancers with inactivating mutations in SMARCA4. After validating the context-specific essentiality of SMARCA2, the authors demonstrated that SMARCA4 mutation leads to the upregulation of SMARCA2 and increases its incorporation into the SWI/SNF complex. Contemporaneously, another large-scale RNAi-based screen for essential genes, which used a chromatin-focused library to screen dozens of cancer cell lines, similarly identified SMARCA2 as a strong and selective dependency in SMARCA4-deficient cancer.⁶²¹ The integrity of the SWI/SNF complex was found to be maintained upon SMARCA2 knockdown in SMARCA4-deficient cancers, suggesting that the antiproliferative effect of SMARCA2 loss-of-function is caused by the redundancy of SMARCA2/4 ATPase activity and not due to destabilization of the entire complex (Figure 29a).

These genetic findings galvanized interest in the pursuit of small molecule ligands that target SMARCA2/4, both to further probe the biology of this system and to potentially exploit the synthetic lethal relationship for therapeutic discovery. Bromodomain ligands were the first small molecules to be reported for SMARCA2/4. The earliest example, PFI-3, was collaboratively reported by the Structural Genomics Consortium and Pfizer and binds SMARCA2/4 bromodomains with an affinity of 89 nM (Figure 29b).^616,741,742 Interesting, the selectivity of PFI-3 against other bromodomain families is attributed to a unique binding mode in which the phenolic head group of the molecule displaces a series of water molecules that are traditionally conserved when liganded by other bromodomain inhibitors. While the compound can elicit a dose-dependent displacement of the isolated SMARCA2 bromodomain from chromatin, it does not show antiproliferative activity in SMARCA4-deficient cancer cells. Subsequent genetic rescue experiments using mutant alleles of SMARCA2 demonstrated that its ATPase activity is required for cell survival but its bromodomain is dispensable.⁶¹⁶ This demonstrated that the ATPase domain is the relevant target for drug discovery in oncology, which was further supported by the discovery that PFI-3 is unable to displace full-length SMARCA2 (as opposed to the isolated bromodomain) from chromatin.⁷⁴² Interestingly the dispensability of the bromodomain is consistent with Drosophila experiments reported in 2002, which demonstrated that the bromodomain of the fly ortholog is not essential for its localization to chromatin.⁷⁴³ It is important to note that even if the bromodomain had been relevant to SMARCA2 function, its close homology with the SMARCA4 bromodomain would have presented an enormous challenge for the goal of discovering ligands that selectively inhibit SMARCA2 over SMARCA4, which is ultimately needed to exploit this synthetic lethality.

The functionally relevant ATPase domain of SMARCA2/4 represented a more difficult target for ligand discovery than the bromodomain. Toward the goal of developing ATPase inhibitors of SMARCA2/4, a team at Novartis designed an ATP turnover assay using a truncated version of SMARCA2 (ATPase and SnAC domains) for high-throughput screening, which was further supported by multiple downstream biophysical assays to assess hit quality.⁷⁴⁴ A diheteroaryl-urea hit emerged from a screen of 72,000 compounds and was moved forward for optimization, leading to the discovery of tool compounds, BRM011 and BRM014 (Figure 29c). Crystal structures demonstrated hydrogen bonding interactions between the urea nitrogen and the catalytic Glu852 side chain, which traps it in an inactive conformation. Interestingly, these compounds were found to bind in an allosteric pocket near the active site, a potential way to circumvent challenges with liganding the associated orthosteric cleft. BRM014 has been used effectively as a chemical tool, for example, demonstrating the widespread importance of SMARCA2/4 for maintaining chromatin accessibility.^745–747 While BRM014 showed potent activity in cell line models and in vivo, it elicited tolerability issues likely owing to the dual inhibition of both paralogs. Similar to the SMARCA2/4 bromodomains, selectively inhibiting one of the two ATPase domains will likely present a major challenge for drug discovery, due to their highly homologous structures.^748,749 While a SMARCA2/4 ATPase inhibitor, FHD-286, is being evaluated in a Phase 1 clinical trials, it suffered a partial clinical hold in relapsed/refractory AML and myelodysplastic syndrome and a full suspension of its clinical exploration in metastatic uveal melanoma.⁷⁵⁰

While SMARCA2/4 bromodomain inhibitors are incapable of phenocopying genetic loss-of-function, they have found a renewed purpose in the development of PROTACs that can exploit SMARCA2/4 synthetic lethality. Here, small molecules targeting the bromodomain have been used to construct PROTACs that selectively degrade SMARCA2 over SMARCA4. While developing bromodomain inhibitors with high isoform selectivity has proven difficult due to the strong structural conservation of bromodomain binding pockets, degraders can achieve apparent selectivity through a different principle. Degraders rely not only on bromodomain binding but also on the formation of a productive ternary complex with an E3 ligase, which introduces an additional layer of specificity based on surface complementarity and complex stability. Thus, bromodomain selectivity is rarely achieved at the inhibitor level but can emerge during targeted protein degradation because the degrader exploits unique degron–ligase–target interactions. The first SMARCA2/4 PROTAC, ACBI1, which degrades both paralogs, was discovered using a rational design strategy that emphasized the formation of a cooperative ternary complex between SMARCA2/4 and VHL (Figure 29d).⁷⁵¹ It was constructed from an aminopyridazine bromodomain ligand and the VHL ligand, VH101, and induces degradation of both SMARCA2 and SMARCA4 at sub-micromolar concentrations. It also degrades PBRM1, a high-affinity bromodomain target of the aminopyridazine ligand. These effects result in potent anti-proliferative activity, recapitulating SMARCA2/4 genetic loss-of-function and SMARCA2/4 ATPase inhibition. The SMARCA2/4 PROTAC, AU-15330, which is constructed from the same bromodomain and VHL ligands but uses a shorter linker and different linker attachments, also degrades both paralogs (Figure 29d).⁷⁵² Despite degrading SMARCA2/4 and PRBM1, prostate cancer cells were reported to be preferentially sensitive to AU-15330.

A next-generation degrader, ACBI2, offers an improved selectivity window (Figure 29d). This VHL-based PROTAC of SMARCA2 was redesigned using a previously reported quinazolinone PRBM1 bromodomain ligand that was further optimized for use in PROTACs by considering rigidity and minimization of hydrogen bond donors.⁷⁵³ Multiple exit vectors were considered on the VHL ligand, arriving at a benzylic attachment site that was evaluated with a variety of linker compositions. Iteratively improving on moderate initial selectivity for SMARCA2 over SMARCA4 with this series of compounds, ACBI2 was found to elicit degradation of SMARCA2 at single-digit nanomolar concentrations and is 30-fold selective against SMARCA4. It also demonstrates PK/PD properties that allow for oral delivery in vivo, eliciting tumor stasis in mouse xenograft models of SMARCA4-deficient lung cancer. A series of selective SMARCA2 degraders have since been disclosed, including A947, SMD-3040 and a collection of diazabicyclooctane molecules.^754–756 Comparative analyses have demonstrated the most improved selectivity with SMD-3040.⁷⁵⁵ Whereas other published PROTACs show relatively strong maximal degradation (D_max) of SMARCA4 at higher concentrations,^751,753,754 SMD-3040, which is constructed by tethering a pyrazolo-aminopyridine bromodomain ligand to a VHL ligand through a rigid spirocyclic linker, is ~100-fold selective for SMARCA2 and shows a diminished D_max for SMARCA4 (Figure 29d).⁷⁵⁵ This compound is useful for in vivo experimentation and demonstrates effective tumor inhibition in SMARCA4-deficient xenograft mouse models with a well-tolerated toxicity profile.⁷⁵⁵ Another redesigned SMARCA2/4 ligand has contributed to the discovery of PROTACs, such as SMD-1087 and SMD-3236, with further improved profiles in vitro and in vivo, which could enable rapid clinical translation (Figure 29e).^757,758 Indeed, as a class, SMARCA2-selective PROTACs are already in clinical-stage investigation, with Phase 1 trials ongoing for the reportedly SMARCA2-selective degraders, PRT3789 (NCT05639751) and PRT7732 (NCT06560645).

3.3.3. ARID1A/ARID1B synthetic lethality

The AT-rich interaction domain 1A (ARID1A) is a SWI/SNF complex subunit that has physiologically relevant roles in transcriptional regulation and chromatin remodeling.⁷⁵⁹ It also serves as a tumor suppressor gene,^760,761 and, in some stage- and tissue-specific cases, as an oncogene.⁷⁶² ARID1A is mutated in almost 10% of human cancers, making it the most commonly mutated SWI/SNF subunit.⁷⁶³ ARID1B, a mutually exclusive and functionally unique counterpart, shares approximately 60% identity with ARID1A. As first revealed by examination of the Project Achilles dataset, ARID1A-mutant tumors are hyper-dependent on ARID1B for survival,⁷⁶⁴ as it incorporates into residual SWI/SNF complexes necessary for cell survival.^618,764,765 No small-molecule ligands for ARID1A/B have been reported, but these genetic data make a compelling argument for developing pharmacologic tools to study this synthetic lethality. Small-molecule tools have been utilized to explore other vulnerabilities of ARID1A-mutant cancers, such as molecules targeting DNA damage repair, cell cycle, and immune checkpoint pathways.^766–768

3.4. CBP/p300

The paralogs, CBP (encoded by CREB-binding protein, CREBBP) and p300 (encoded by E1A binding protein p300, EP300), are transcriptional coactivators and lysine acetyltransferases with diverse cellular functions, most notably including redundant and non-redundant roles in transcriptional regulation. They share nine structurally conserved functional domains involved in protein-protein interactions, recognition of post-translational modifications, and catalysis of lysine acetylation.^769,770 Two of these are recognized as promising targets for small-molecule therapeutics: the bromodomain and the HAT domain. The bromodomain binds to acetylated lysines and regulates CBP/p300 HAT activity, which is responsible for acetylating thousands of lysines on histone and non-histone substrates, like transcription factors and transcriptional co-regulators.⁷⁷¹ CBP/p300 regulate enhancer-mediated transcriptional activation, highlighting their critical roles in both normal physiological and pathophysiological processes.⁷⁷² These proteins play a critical role in the development and maintenance of solid and hematological tumors alike, most notably as oncogenic fusions and non-oncogene dependencies, which has motivated broad efforts in chemical probe and drug discovery.⁷⁷²

CBP/p300 are critical to a variety of essential cellular functions, including cell cycle regulation, DNA synthesis and repair, and transcriptional control. Heterozygous loss-of-function mutations are known to cause Rubinstein-Taybi syndrome (RTS), a congenital disease characterized by growth deficiency, mental deficit, developmental delays and abnormalities, and a characteristic dysmorphology.⁷⁷³ Another critical feature of RTS is the propensity to develop cancer, potentially due to somatic mutation of the remaining functional allele. In addition to RTS tumors, loss-of-function alterations to EP300 and CREBBP – including loss-of-heterozygosity, truncations, and missense mutations – have been known to occur in colorectal, breast, ovarian, lung, and gastric cancers for several decades.^774–780

Inactivating mutations are also common in some hematological malignancies, including acute lymphoblastic leukemia and lymphoma.^781,782 In healthy lymphoid tissues, CBP/p300 maintains the acetylation pattern required for the terminal differentiation of mature B cells—that is, the final differential step in which B cells irreversibly commit to a cellular identity of plasma cell or memory B cell.⁷⁸³ In normal immunophysiology, a subset of B cells will endure mutagenic germinal center reactions, in which they deliberately acquire somatic hypermutations and class-switch recombination to diversify the antibody repertoire, in response to encountering immunogenic antigens. Loss-of-function mutations of CREBBP, however, impair the reacetylation of enhancers that are necessary to facilitate germinal center exit, ultimately promoting lymphomagenesis.⁷⁸¹ Follicular lymphoma (FL) and diffuse large B-cell lymphoma (DLBCL), for example, exhibit frequent inactivating structural mutations in the HAT domain of CBP and, less frequently, p300.⁷⁸¹ These mutations were discovered by whole-exome sequencing and predicted to cause the elimination or truncation of the HAT domain in CBP, reducing its affinity for acetyl-CoA binding and preventing acetylation of its substrates.⁷⁸¹ The lack of BCL6 acetylation by mutant CBP has been shown to result in unopposed activity of BCL6/SMRT/HDAC3 complexes at B-cell signaling enhancers, contributing to lymphomagenesis.⁷⁸⁴

Cancers with CREBBP loss-of-function alterations are hyper-dependent on p300 function, creating a synthetic lethal target for drug discovery and development. This was originally discovered in an unbiased forward genetic screen using a library of siRNAs to knock down the expression of chromatin regulatory proteins.⁷⁸⁵ It has been further reinforced in vivo using mouse models for conditional deletion of Crebbp or Ep300 in the germinal center (GC), altogether showing the combined essentiality of CBP/p300.⁷⁸⁶ The loss of both paralogs in B cells compromises GC formation, creating a novel vulnerability to EP300 deletion in DLBCL cell lines that harbor CREBBP mutations.⁷⁸⁶

Instability of the genome near the 5’ end of the CREBBP gene, alongside potential breakpoints in the second intron, predispose CREBBP to chromosomal translocations yielding oncogenic CBP fusion proteins.^787,788 This phenomenon is more common with CBP than p300, and these translocations are frequently implicated in the transformation of hematological malignancies. These include CBP fusions with monocytic leukemia zing-finger (MOZ, also known as KAT6A),^789,790 MOZ-related factor (MORF, or KAT6B),⁷⁹¹ and MLL.^792,793 Rarely, fusions can also be seen with p300, as evidenced by MOZ-p300 fusion oncoprotein.⁷⁹⁰ These leukemogenic fusion proteins form aberrant chromatin acetylation and transcriptional co-activation patterns through the mis-localization of their enzymatic activity on the genome.

The first pharmacological perturbations of HAT proteins were enabled by the design and synthesis of bisubstrate inhibitors that engage the binding sites for both histone substrates and the acetyl-CoA cofactor. This is best exemplified by Lys-CoA, a potent CBP/p300 HAT inhibitor with ~500 nM potency.⁷⁹⁴ While bisubstrate inhibitors were exceptionally useful for in vitro biochemistry and structural biology studies,⁷⁹⁵ their lack of drug-like physiochemical properties prohibited cell biological studies. The natural products, garcinol and anacardic acid, were later reported as CBP/p300 inhibitors,^796,797 along with a synthetic small-molecule, C646,⁷⁹⁸ but these tools were later found to be complicated by non-specific off-target effects.⁷⁹⁹ C646, which found widespread use as a p300 inhibitor, shows off-target reactivity with thiols in cells and was recently reported to function through XPO1, with p300 being perturbed indirectly by C646-induced degradation of XPO1.^800,801 In more recent years, these tools have been succeeded by highly potent and selective chemical probes for both the bromodomains and HAT domains of CBP/p300, providing a structurally and functionally diverse collection of chemical tools to study CBP/p300.

3.4.1. CBP/p300 bromodomain inhibitors

CBP/p300 bromodomain inhibitors were among the first non-BET bromodomain inhibitors to be reported after the discovery of JQ1 and I-BET. Preceded by the discovery of fragments and macrocyclic peptide with relatively weak potency,^802–804 the first potent and selective chemical probe for CBP/p300 bromodomains, SGC-CBP30, was reported by the Structural Genomics Consortium in 2014.⁸⁰⁵ This discovery was enabled by optimizing a 3,5-dimethylisoxazole BET bromodomain inhibitor that had previously been shown to possess off-target activity against CBP.⁸⁰⁶ A collection of 83 analogs – synthesized by parallel Suzuki couplings of commercially available pinacol esters and heteroaryl bromides – were screened by differential scanning fluorimetry (DSF) for selective engagement of CBP over BRD4.⁸⁰⁷ Further modifications of a hit from this screen, together with structure-guided design, afforded the chemical probe SGC-CBP30 (Compound 59 in the index manuscript), which binds with 20–30 nM affinity to CBP/p300, but is only 40-fold selective against BRD4 (Figure 30a).^{807 808} Based on this scaffold, additional structure-based drug design afforded the chemical probe PF-CBP1, an inhibitor with slightly reduced potency for CBP/p300 (IC₅₀ = 130 nM), but dramatically improved selectivity against BRD4 (IC₅₀ = 18 μM) (Figure 30a).⁸⁰⁹ The same isoxazolylbenzimidazole scaffold is shared by CCS1477, the first CBP/p300 bromodomain inhibitor to advance into human clinical investigation (NCT03568656 and NCT04068597) (Figure 30a).⁸¹⁰ It is highly potent against CBP/p300, showing 1.3 nM affinity for p300, and is more than 100-fold selective against BRD4 (222 nM affinity). Clinical trials for CCS1477 are focused on prostate cancer, where CBP/p300 function as co-activators for androgen receptor (AR). CCS1477 is orally bioavailable and preclinical experiments have shown it can delay tumor growth in patient-derived xenograft models of prostate cancer.⁸¹⁰ which is consistent with prior reports showing that prostate cancer models are hypersensitive to CBP/p300 bromodomain and HAT domains inhibitors.^811,812

Figure 30.

Small-molecules targeting CBP/p300. (a) Benzimidazole-based bromodomain inhibitors. (b) Piperidine-based bromodomain inhibitors. (c) Benzodiazepinone-based bromodomain inhibitor. (d) Benzoxazepine based bromodomain inhibitor. (e) Structurally distinct CBP/p300 HAT inhibitors. (f) CBP/p300 PROTACs, which share piperidine-based bromodomain ligands to target CBP/p300 for degradation.

A pyrazolopiperidine core has also afforded CBP/p300 bromodomain inhibitors with potent activity in vivo and excellent selectivity against BET bromodomains.⁸⁰⁶ This scaffold was first discovered as a screening hit from a thermal shift assay for CBP.⁸¹³ Structure-based drug design motivated three areas for hit optimization: an N-acetylpiperidine group was preferred over alternative acetyllysine mimics, a pyrazole substituent was found to make favorable interactions with the WPF shelf, and a tetrahydrofuran moiety occupies the ZA loop. An analog that used an N-methylpyrazole moiety to contact the WPF shelf showed good potency and physicochemical properties, yet suffered from metabolic instability. This was improved by an analog that moved the N-methylpyrazole from the 3 position to the 4 position of a phenyl ring that was also fluorinated at 2 position to decrease glucoronidation. The resulting chemical probe, GNE-272, shows low clearance, good oral bioavailability, and excellent selectivity for CBP/p300 over other off targets (Figure 30b). In a leukemia xenograft model (using the MOLM-16 AML cell line), it was shown to decrease MYC expression and repress leukemia burden at doses that pharmacokinetic studies predicted would inhibit CBP/p300 but not BET bromodomains.⁸¹³

Additional improvements on GNE-272 were subsequently reported with contemporaneous disclosures of GNE-049, GNE-781, and GNE-207 (Figure 30b).^811,814,815 Concerns that the aniline in GNE-272 could be a possible toxicophore motivated the direct attachment of the phenyl group to the pyrazolopiperidine core, which maintained activity.⁸¹⁶ Successive replacements of the phenyl group with indole and isoquinolone groups, together with exchanging the tetrahydrofuran for a tetrahydropyran, delivered progressive improvements in potency. Further substitution of the isoquinolone with an amidopyridine yielded GNE-207, a highly potent inhibitor of the CBP bromodomain (IC₅₀ = 1 nM) with >2000-fold selectivity over BRD4 (Figure 30b). While this compound demonstrated acceptable PK for in vivo study,⁸¹⁶ more substantial improvements were obtained with the contemporaneous report of GNE-781 (Figure 30b).⁸¹⁷ Here, it was found that constraining the aniline nitrogen of GNE-272 into a tetrahydroquinoline group prevented the molecule from adopting a planar conformation when bound to BRD4, introducing a steric clash and improving selectivity. Introduction of a difluoromethyl substituent to position 7 of the tetrahydroquinoline further improved CBP potency and selectivity over BRD4 and replacement of the tetrahydrofuran with a tetrahydropyran (as in GNE-207) improved metabolic stability. A final replacement of the acetamide moiety, which mimics the binding of acetyllysine to the conserved bromodomain asparagine, with a methylurea group addressed adverse effects on the central nervous system, ultimately yielding a compound with subnanomolar potency in biochemical assays, <10 nM potency in cells, >5,000-fold selectivity over BET bromodomains, and suitable oral bioavailability. This compound, GNE-781, is highly selective for CBP/p300 over a large panel of other bromodomains, kinases, and cytochromes p450s, providing an excellent chemical probe for CBP/p300 both in vitro and in vivo, while other compounds were prioritized for clinical development (Figure 30b).⁸¹⁷

Additional scaffolds to probe CBP/p300 are well represented by CPI-637 and I-CBP112. The benzodiazepinone CPI-637 is a potent and selective chemical probe for CBP/p300 that arose from the optimization of a fragment screening hit for CBP (Figure 30c).⁸¹⁸ A co-crystal structure demonstrated that CBP Asn1168, the conserved asparagine that contacts acetyllysine side chains in all bromodomains, makes a critical hydrogen-bonding interaction with the benzodiazepinone carbonyl group. It also showed electron density only for the R enantiomer, leading to the discovery that this scaffold stereoselectively inhibits CBP/p300. Structure-based drug design ultimately yielded CPI-637, a potent and stereoselective CBP/p300 bromodomain inhibitor (IC₅₀ = 30 nM for CBP) with favorable selectivity against BRD4 bromodomains (IC₅₀ = 11 μM).⁸¹⁸ The benzodiazipinone scaffold of I-CBP112 was discovered by screening benzoxazepine molecules similar to the benzodiazepine class of BET bromodomain inhibitors (Figure 30d).⁸⁰⁸ Compared to more recent inhibitors, I-CBP112 only shows moderate affinity for CBP/p300 (K_D = 151 nM) and modest selectivity against BRD4 bromodomains (K_D = 5.6 μM).

3.4.2. CBP/p300 HAT inhibitors

A-485, the first potent, selective, and cellularly active chemical probe for a HAT domain—that of CBP/p300—was reported in 2017 (Figure 30e).⁸¹² Compounds that resulted from earlier efforts to inhibit HAT domains were typically limited by non-specific assay interference and off-target activities.⁷⁹⁹ This major breakthrough in HAT pharmacology was enabled by a virtual screen of 800,000 commercially available compounds.⁸¹⁹ Hypothesizing that p300 may undergo conformational changes known to take place when other proteins bind acetyl-CoA, open conformations of the enzyme were computationally predicted and used for virtual ligand screening (VLS), which deviated from prior VLS efforts that had used published X-ray crystal structures. Among 1,300 VLS hits that were experimentally tested in a radioactivity-based p300 HAT enzyme assay, a hydantoin compound was prioritized for its low micromolar potency, lack of potentially reactive groups, and synthetic tractability. Iterative medicinal chemistry optimization arrived at a spiro-indane hydantoin core substituted with a methylurea moiety and then a spiro oxazolidinedione core with improved permeability, clearance, and oral bioavailability. Interestingly, the stereogenic spiro center showed opposite enantioselectivity in the spirooxazolidinedione series compared to the spirohydantoin (Figure 30e). Additional modifications to improve PK properties resulted in A-485, a potent, selective, and in vivo active chemical probe that inhibits the acetylation of CBP/p300 substrates (e.g. H3K27Ac and H3K18ac) without affecting other acetylation sites (e.g. H3K9ac).^812,819 A co-crystal structure of A-485 bound to the HAT domain of p300 showed direct engagement of the acetyl-CoA binding pocket and biochemical analyses determined that it binds competitively with this cofactor.^812,819 Notable interactions are made with a loop (L1) that is present in CBP/p300 but absent in other HATs, which was hypothesized to explain both its selectivity against other acetyltransferases and the lack of activity for the inactive control compound, A-486, a urea regioisomer of A485.⁸¹²

Additional analogs of the oxazolidinedione and hydantoin scaffolds have been reported since the initial discovery of A-485.^820–822 This includes the highly optimized spirohydantoin, B026, a CBP/p300 HAT inhibitor 30-fold more potent than A-485 (Figure 30e).⁸²³ B026 and other A-485 analogs have independently converged on replacing the methylurea with a methylpyrazole group, removing hydrogen bond donors to improve cellular permeability and oral bioavailability.^820,821 A structural analysis of an optimized hydantoin analog has provided a possible explanation for the switch of stereochemical preference between the spirohydantoin and spirooxazolidinedione scaffolds, with the (S) spirocenter in the hydantoin series allowing for an additional hydrogen bonding interaction between the hydantoin core and p300. Structural analyses also allowed for the design of a covalent CBP/p300 inhibitor by attaching a phenylacrylamide to the urea of A485, which is located just 7 Å from Cys1450 of p300.⁸²⁴

A structurally distinct 2-aminopyridine chemical probe, CPI-1612, complements the spirohydantoin and spiroxazolidinedione scaffolds for CBP/p300.⁸²⁵ CPI-1612 originated from a high-throughput screen of 200,000 small molecules, which used a FRET-based assay to monitor p300-mediated acetylation of a histone peptide substrate.⁸²⁶ From ~2,500 hits, a total of 18 compounds passed a series of counter-screen assays, including both acetyl-CoA-competitive and histone peptide-competitive inhibitors.⁸²⁶ Medicinal chemistry optimization of an acetyl-CoA-competitive indole compound led to a redesigned amidopyridine core and, ultimately, the discovery of CPI-1612.⁸²⁵ Interestingly, these optimization efforts identified the same methylpyrazole group that is found in the optimized spirohydantoin compounds. Despite their unique chemical scaffolds, both classes of compounds position the methylpyrazole group in the same site on p300 and with the same orientation.^821,825 CPI-1612 shows impressive sub-nanomolar potency against p300 and reduces H3K27Ac and tumor growth in vivo with just twice daily oral dosing of 0.5 mg/kg.⁸²⁵

3.4.3. CBP/p300 PROTACs

Both bromodomain and HAT domain inhibitors have been used to design CBP/p300 PROTACs. The first example, dCBP1, is a highly potent and selective CRBN-based PROTAC designed from the bromodomain ligand, GNE-781.⁸²⁷ Using an in silico docking analysis to nominate potential linker attachment sites for GNE-781 and A-485, the tetrahydropyran of GNE-781 was prioritized for PROTAC construction. dCBP1 was synthesized by substituting the tetrahydropyran of GNE-781 for a conjugatable piperidine and attaching 5’-aminothalidomide through a polyethylene glycol (PRG)-4 linker (Figure 29f). It induces potent, selective, and rapid degradation of both CBP and p300 in cell culture.⁸²⁷ The rapid disruption of CBP/p300-mediated acetylation and enhancer-dependent transcription by dCBP1 offers a well-characterized chemical probe for cellular study that complements bromodomain and HAT domain inhibitors.

Following the discovery of dCBP1, more potent degraders have been reported, including compounds that can induce degradation at low picomolar concentrations. JET-209, a CRBN-based PROTAC, is ~10-fold more potent than dCBP1 (Figure 30f).⁸²⁸ It is constructed from GNE-207 with a short and rigid piperidine linker using an alternative linker attachment site compared to dCBP1. It is highly active in vivo, degrading CBP/p300 within 3 hours of a 1 mg/kg dose.⁸²⁸ The GNE-049 scaffold was used in the successful pursuit of orally bioavailable degraders, as exemplified by the contemporaneously reported PROTACs, CBPD-409 and CBPD-268 (Figure 30f).^829,830 CBPD-268 features a redesigned CRBN ligand, TX-16, which had previously afforded potent and orally available nuclear hormone receptor PROTACs.^831,832 To minimize hydrogen bond donors, TX-16 was attached directly to a GNE-049 analog that replaced the tetrahydropyran for a cyclohexane, resulting in a compound, CBPD-268, with low picomolar potency and improved pharmacokinetic properties (Figure 30f).⁸³³ Impressively, it is capable of inducing tumor regression in a xenograft model of prostate cancer with 1 mg/kg oral dosing given just 3 times per week.

These and other bromodomain-based CBP/p300 degraders, such as the CCS1477-based PROTAC, QC-182,⁸³⁴ induce degradation of both paralogs equivalently.⁸³⁵ Interestingly, early progress in the development of paralog-selective degraders, which will be needed if the synthetic lethality observed between CBP and p300 is to be exploited therapeutically, has come in the form of HAT-based PROTACs. The first example, JQAD1, is a CRBN-based PROTAC constructed from the CBP/p300 HAT inhibitor, A485.⁸³⁶ While the degradation induced by JQAD1 is relatively slow for both CBP and p300, there is a kinetic window in which p300 degradation is observed without CBP degradation.⁸³⁶ In vivo, this window enables the selective degradation of p300 over CBP, establishing an important proof-of-concept for paralog-selective degradation.⁸³⁶ The HAT inhibitor CPI-1612 has also been used to develop CBP/p300 PROTACs,⁸³⁷ with a recently reported VHL-based compound, MC1, selectively degrading p300 over CBP at 1 μM.⁸³⁸ MC1 is more broadly active across cell lines compared to JQAD1 and appears to show improved kinetics of degradation. These studies highlight the possibility of progressing p300-selective PROTACs as drug candidates for CREBBP-mutant cancers.

3.4.4. Clinical translation of CBP/p300 antagonists

While the synthetic lethality between CBP and p300 provides a potentially powerful mechanism to selectively target several cancer types, indiscriminate inhibition of both CBP and p300 presents the possibility for on-target toxicity to normal tissues. Nevertheless, some cancer subtypes show hypersensitivity to CBP/p300 inhibition, suggesting that pharmacological inhibition may have nuanced effects across different lineages. Profiling of the orthosteric CBP/p300 HAT inhibitor, A-485, across diverse cancer cell lines has shown it is preferentially cytotoxic to multiple myeloma and prostate cancer cell lines.⁸³⁹ Even within a single lineage, sensitivity to A-485 varies widely, a phenomenon that cannot be attributed to differences in the extent of CBP/p300 inhibition.^812,840,841 Therefore, it can be inferred that there are intrinsic differences in the cellular requirement for CBP/p300 activity, both within and across cancer cell lineages. This might reasonably be expected to hold true across normal tissues as well, which could make dual inhibition or degradation better tolerated than perhaps initially expected. Multiple myeloma, which is among the most acutely sensitive lineages to dual inhibitors and degraders, are also the most sensitive to genetic deletion of CREBBP/EP300,^812,840,842 supporting the idea that CBP/p300 activity might be a targetable dependency with a reasonable therapeutic window for some cancer lineages.

If dual degraders or orthosteric HAT inhibitors prove to be too toxic, bromodomain inhibitors may prove a better tolerated approach to exploit lineage-restricted CBP/p300 dependencies. Unlike BET bromodomain inhibitors, which evict their targets from chromatin by competitively blocking their association with acetyllysine side chains, CBP/p300 bromodomain inhibitors do not inhibit chromatin localization.⁸⁴³ Both genetic and chemical studies show that the bromodomain-acetyllysine interaction is not required for CBP/p300 localization to chromatin.⁸⁴³ Instead, the bromodomain is part of the catalytic core, with CBP/p300 bromdomain inhibitors inhibiting catalytic activity rather than chromatin localization.^843,844 Interestingly, however, the inhibition of CBP/p300 catalytic activity by bromodomain inhibitors is substrate-selective.⁷⁷¹ Whereas CBP/p300 HAT inhibition provokes the loss of acetylation across thousands of lysines, bromodomain inhibition affects a much smaller subset of CBP/p300 substrates.⁷⁷¹ This is very clearly observed by H3K27ac, which is depleted by both HAT and bromodomain inhibitors, and H3K18ac, which is only depleted by HAT inhibitors.⁸⁴³

Mechanistically, this milder inhibition of CBP/p300 function by bromodomain inhibitors might lead to improved tolerability. The lineage-specific dependency on CBP/p300 in prostate cancer has also been observed with CBP/p300 bromodomain inhibitors, motivating ongoing clinical trials for CCS1477 in castration-resistant prostate cancer (CRPC).^811,845 Unfortunately, CCS1477 and the structurally distinct CBP/p300 bromodomain inhibitor, GNE-781, both induce dose-limiting thrombocytopenia in mice, presenting a possible on-target toxicity concern for CBP/p300 inhibition that has led to the implementation of intermittent dosing in humans.^846,847

In the coming years, the results of clinical trials for CBP/p300 bromodomain inhibitors will answer many questions about CBP/p300 tolerability and lineage-restricted dependencies. The development of CBP/p300 HAT inhibitors has not yet matured to the point of clinical approval, which may, in part, be due to the toxicity of such compounds, but any new trials would be similarly informative. Much excitement would remain for paralog-selective degraders if this pharmacology can continue to progress. Other strategies to target CBP/p300 continue to be reported, such as the recent description of a photorelease strategy that aims to improve tolerability to this class of drugs.⁸⁴⁸ CBP/p300 bromodomain inhibitors have also found use in bifunctional molecules that recruit CBP/p300 for inducing targeted lysine acetylation in cells,^{329,849–851} suggesting new strategies to target or harness CBP/p300 for anti-cancer therapeutics may yet be awaited.

3.5. GCN5/PCAF

The proteins KAT2A and KAT2B, also known as GCN5 and PCAF, respectively, are two closely related lysine acetyltransferases that play critical roles in the regulation of gene expression. PCAF (P300/CBP-associated factor) and GCN5 (general control non-derepressible 5) are members of the GCN5-related N-acetyltransferase (GNAT) family and are key components of multi-protein transcriptional co-regulatory complexes, importantly the Spt-Ada-Gcn5-acetyltransferase (SAGA) complex. Several recent studies have reported SAGA complex subunits as potentially targetable cancer dependencies. An analysis of DepMap recently revealed that the core subunits TADA2B, SUPT20H, TAA1, and TAF5L are critical neuroblastoma-specific dependencies.⁸⁵² This was orthogonally validated with CRISPR/Cas9- and degron-based loss-of-function perturbations and led to the further discovery that the combined activity of GCN5/PCAF is essential for MYCN-amplified neuroblastoma.⁸⁵² Two contemporaneous reports found the SAGA subunit, SGF29, to be a leukemia-specific vulnerability.^853,854 In particular, domain-focused CRISPR-Cas9 screening experiments in these reports found its chromatin reader Tudor domain to be necessary for SGF29-dependent leukemia survival, nominating a potential new target for drug discovery.^853,854

Fragments that bind the bromodomain of PCAF and block acetyllysine recognition have been known since 2005 but generally featured limited affinity.^855–858 In 2008, a phenotypic screening hit, CPTH2, was reported to inhibit the yeast Gcn5p,⁸⁵⁹ but it has since been described as a promiscuous electrophile with likely limited value as a chemical probe.⁷⁹⁹ Discovery of the first potent and cell-active PCAF/GCN5 chemical probes arrived with multiple reports published in 2016. The PCAF/GCN5 bromodomain inhibitor, GSK4027, originated from a screen of 30,000 molecules with known and putative acetyllysine mimetic groups using a fluorescence polarization (FP) assay that measured the displacement of a promiscuous bromodomain ligand (Figure 31a).⁸⁶⁰ A prioritized pyridazinone hit was used for structure-based optimization, which focused on improving PCAF/GCN5 potency while also improving selectivity against BRD4, ultimately leading to the discovery of GSK4027 and an enantiomeric negative control.⁸⁶⁰ The chemical probe, L-45 (also known as L-Moses), was reported contemporaneously and features a structurally distinct triazolopthalazine scaffold (Figure 31b). Using promiscuous bromodomain inhibitors as starting points, structure-based drug design, virtual screening, and iterative medicinal chemistry optimization delivered the chemical probe L-45 and an inactive enantiomer.⁸⁶¹ Both GSK4027 (K_D,PCAF = 1.4 nM) and L-45 (K_D,PCAF = 126 nM) exhibit excellent selectivity against BRD4 bromodomains, with GSK4027 being more potent.⁸⁶¹ Several additional compounds have been reported for GCN5/PCAF,^862,863 notably including compounds related to GSK4027 that replace the pyridazinone warhead with a pyrrolopyrimidinone system.⁸⁶⁴ This new scaffold was discovered by virtual screening and optimized with 2-position substitutions to improve its potency (Figure 31c).⁸⁶⁴

Figure 31.

PCAF/GCN5 drug discovery. (a) Summary of medicinal chemistry optimization leading to GSK4027. (b) Summary of medicinal chemistry campaign leading to L-45. (c) Summary of hit to lead [(R,R)-36n] medicinal chemistry optimization of PCAF/GCN5 chemical probe. (d) Degrader of PCAF/GCN5.

In an immunological context, PCAF and GCN5 are reported to play a role in the production of inflammatory cytokines, as evidenced by the reduced levels of interleukin-6 and tumor necrosis factor in LPS challenged mice that have had PCAF genetically ablated.^{865,866 865,866}Bromodomain inhibition of PCAF/GCN5 by GSK4027 are unable to phenocopy the immunomodulatory effect of PCAF knockout, which inspired the development of PROTAC degraders.⁸⁶⁷ Examination of the crystal structure of GSK4027 bound to PCAF illuminated an appropriate exit vector to which an E3 ligase ligand could be appended, resulting in the CRBN-based PROTAC, GSK699, and an inactive enantiomer, GSK702 (Figure 31d).⁸⁶⁷ GSK699 was able to deplete PCAF and GCN5 protein levels with 1.5–3 nM potency and induce a marked reduction of inflammatory markers. In addition to serving as a prominent proof of concept for inducing PCAF and GCN5 degradation, this study convincingly established the need to degrade GCN5/PCAF rather than inhibit their bromodomains. While GCN5/PCAF bromodomain inhibitors were not studied in the recent report of SAGA dependencies in neuroblastoma, treatment with GSK699 was found to inhibit the growth of neuroblastoma cells in vitro and in vivo.⁸⁵² In support of this being an on-target effect, the inactive enantiomer of GSK699 could not suppress neuroblastoma growth, establishing SAGA dependencies as potential new targets for drug development.

3.6. MYST HAT dependencies

The first orthosteric HAT inhibitor to enter human clinical investigation targets the paralogous acetyltransferases, KAT6A (also known as MOZ) and KAT6B (MORF).⁸⁶⁸ KAT6A/B contain a MYST family (named for MOZ, Ybf2, Sas2, Tip60) HAT domain consisting of an acetyl-CoA-binding motif and a zinc finger fold. They function as transcriptional co-regulators by catalyzing lysine acetylation, with an emerging consensus on their importance for H3K23 acetylation.^869–872 KAT6 can impact diverse physiological processes, including development,^873–878 cellular senescence,^879–881 and hematopoiesis.^882–886 In cancer, KAT6A was originally discovered as a partner of CBP-fusion oncogenes and has since been described in several other oncogenic fusions.^{789,791,887–900} In addition to its presence in fusion oncogenes, multiple studies have nominated wild-type KAT6A as a genetic dependency of hematological malignancies, namely lymphoma and AML.^900,901 However, KAT6A acetyltransferase activity is also essential for the maintenance of normal hematopoietic stem cells and its roles in normal and malignant hematopoiesis are closely linked.^883,884,886 For example, its importance in lymphoma is closely related to its effect on the proliferation of normal B-cell progenitors⁹⁰¹ and, similarly, the discovery of KAT6A dependency in AML originated in a forward genetic screen for chromatin regulators that block terminal differentiation.⁹⁰⁰

The first MYST acetyltransferase inhibitors were discovered in a high-throughput screen using a biochemical assay that measures the acetylation of a synthetic histone peptide.⁹⁰² Aided by efficient biophysical assays for hit triage, a naphthylsulfonohydrazide screening hit with low micromolar potency, CTx-0123143, was identified as the only true-positive hit from a screen of 243,000 compounds.⁹⁰² Using this scaffold for extensive medicinal chemistry optimization resulted in the discovery of WM-8014 and WM-1119, two potent and selective KAT6A/B acetyltransferase inhibitors (Figure 32a,b).^881,903,904 These acetyl-CoA-competitive inhibitors directly engage the MYST catalytic pocket, positioning the acylsulfonylhdrazide to mimic the contacts of the acetyl-CoA diphosphate group.⁹⁰⁴ Both WM-8014 and WM-1119 bind KAT6A/B with high affinity (K_D < 10 nM), but WM-1119 features improvements in selectivity (250-fold selective over KAT7) and pharmacokinetic properties.⁸⁸¹ These improvements enabled limited studies of KAT6A/B inhibition in vivo, with WM-1119 provoking an anti-lymphoma effect in mice with 3–4 times daily dosing (50 mg/kg per dose ).⁸⁸¹

Figure 32.

Small molecules targeting MYST family lysine acetyltransferases. (a-c) Chemical probes that inhibit KAT6A and KAT6B acetyltransferase activity. (d) Small molecule inhibitor of KAT7/HBO1 acetyltransferase activity.

The same high-throughput screening cascade that was used to discover the acylsulfonylhdrazide series was later used to discover a benzisoxazole sulfonamide scaffold with improved pharmacological properties.⁸⁷⁰ An initial screen of 250,000 compounds led to the discovery of 3 biophysically validated hits, including the benzisoxazole CTx-533. Medicinal chemistry was supported by biochemical, biophysical, and cellular assays to measure potency and selectivity for KAT6A, ultimately resulting in the discovery of PF-9363 (Figure 32c).⁸⁷⁰ A co-crystal structure with KAT6A demonstrated that the sulfonamide and isoxazole form the functional acetyl-CoA diphosphate mimic, similar to the acylsulfonylhydrazide moiety in WM-1119. PF-9363 is exceptionally potent, inhibiting KAT6A and KAT6B with potencies of 270 pM and 2.4 nM, respectively, and it possesses excellent drug-like physicochemical properties for study in animal model systems.⁸⁷⁰ Among the other 3 MYST acetyltransferases encoded in the human genome, PF-9363 is >250-fold less potent at inhibiting KAT7 and >1,000-fold less potent for KAT5/8, altogether affording an excellent chemical probe to study KAT6A/B and, at higher concentrations, KAT7.⁸⁷⁰

In AML, KAT6A dependency was first proposed to be mediated through the chromatin reader protein ENL, another AML dependency.⁹⁰⁰ Specifically, it was suggested that ENL binds to chromatin by recognizing H3K9ac marks deposited by KAT6A. While PF-9363 was later used to demonstrate that KAT6A/B are not likely involved in the acetylation of H3K9 (at least not globally),⁸⁷⁰ a functional connection between ENL and KAT6A is supported by a physical interaction whereby acetylated lysine side chains from KAT6A bind into the YEATS domains of AF9 and ENL.⁹⁰⁵ More recent study has shown that KAT6A is a dependency across several models of acute leukemia, but combining KAT6A/B with KAT7 inhibition provides a stronger anti-leukemic effect.^906,907 Prior to this report, KAT7 acetyltransferase activity had been discovered as a critical requirement for the maintenance of leukemia stem cells populations,⁹⁰⁸ which has been attributed to a functional and physical interaction between KAT7 and menin-MLL complexes.^908–910 Consistent with this link, both genetic knockout and pharmacological inhibition of KAT6A have been found to sensitize MLL-fusion leukemia to menin-MLL inhibition in vitro.⁹¹¹ Indeed, combining PF-9363 with the menin-MLL inhibitor, VTP-50469, elicits profound anti-leukemia effects in vivo, overcoming primary and acquired resistance to menin-MLL inhibitors.^906,907 A related effect has been observed in gastrointestinal stromal tumors (GIST), where a correlated dependency on KAT6A and MLL has recently been described.⁹¹² Interestingly, combined treatment with WM-1119 and the menin-MLL inhibitor, VTP-50469, inhibits GIST growth in vivo, despite both compounds failing to show single-agent activity in this disease.⁹¹²

In the future, KAT6/7 acetyltransferase inhibitors may have considerable potential as acute leukemia therapeutics, especially in combination with the now FDA-approved class of menin-MLL inhibitors. Currently, however, both clinical-stage KAT6A/B inhibitors, PF-07248144 and MEN-2312, are being evaluated in breast, prostate, and lung cancer (NCT04606446 and NCT06638307). Based on prior observations that KAT6A is frequently amplified in breast cancer and that KAT6A knockdown reduces estrogen receptor (ER) levels,^913,914 PF-9363 was first studied in ER-positive breast cancers with KAT6A overexpression.⁸⁷⁰ In breast cancer cell lines, anti-proliferative responses to PF-9363 are most closely associated with KAT6A overexpression and ER-positive luminal subtypes, which is consistent with the observed ability of PF-9363 to suppress ER-dependent transcriptional programs.⁸⁷⁰ PF-9363 elicits on-target anti-tumor activity in mouse xenograft models of breast cancer, including some models that have shown drug-induced tumor regressions.⁸⁷⁰ In a first clinical report from the Phase 1 clinical trial, NCT04606446, the KAT6A/B inhibitor, PF-07248144, was found to be sufficiently well tolerated to be given at doses that cause a pharmacodynamic response (loss of H3K23ac) in peripheral blood mononuclear cells and tumors.⁸⁶⁸ This preliminary report showed PF-07248144 possesses antitumor activity both as monotherapy and in combination with the selective estrogen receptor degrader, fulvestrant, establishing KAT6A/B inhibition as a compelling therapeutic strategy for metastatic breast cancer.

3.7. NSD2 oncogene dependency

H3K36 dimethylation (H3K36me2) is catalyzed by the nuclear SET domain (NSD) family of histone methyltransferases – NSD1, NSD2, and NSD3 – and is associated with active gene transcription, mRNA processing, and DNA damage repair.⁹¹⁵ NSD2 is also known as MMSET (multiple myeloma SET) and WHSC1 (Wolf Hirschhorn syndrome candidate 1), names that originate from the association of NSD2 alterations with these diseases. In multiple myeloma (MM), recurrent NSD2 mutations or increases in NSD2 expression create an increase in H3K36 methylation,^916,917 with consequences on cellular adhesion,⁹¹⁸ proliferation,⁹¹⁹ and transformation.⁹²⁰ NSD2 also plays consequential roles in a variety of other cancer types, including lung cancer,⁹²¹ prostate cancer,⁹²² and pediatric malignancies,⁹²³ among others.^924–929 In such cases, NSD2 can either be overexpressed, commonly due to chromosomal translocations, or feature point mutations. For example, oncogenic t(4;14) translocations, which are observed in 15–20% of MM patients, place the NSD2 gene under control of the Ig heavy chain locus, dramatically increasing its tissue-specific expression.⁹³⁰ E1099K, D1125N and T1150A mutations enhance its interaction with the nucleosome by destabilizing the autoinhibitory loop, keeping it in an open hypermorphic state and allowing for hypermethylation.^931–935 E1099K and T1150A mutations are oncogenic and poor prognostic indicators for B-ALL and are associated with relapse in patients with mantle cell lymphoma.^936–939 These findings have nominated NSD2 as a potential therapeutic target, motivating the development of chemical tools and drug candidates.

3.7.1. NSD SET domain inhibitors

NSD2 harbors a SET methyltransferase domain and multiple protein folds that mediate chromatin interactions, including two PWWP (proline-tryptophan-tryptophan-proline) domains, multiple PHD (plant homeodomain) fingers, and an HMG (high mobility group) box (Figure 33a). While potent inhibitors have been achieved through the development of peptidomimetic inhibitors like PTD2, a pentapeptide that replaces the substrate lysine with norleucine, these compounds are not suitable for cellular studies.⁹⁴⁰ Many efforts to develop small-molecule NSD2 inhibitors have focused on repurposing quinazoline inhibitors of the G9a methyltransferase, like BIX-01294 (Figure 33b). These quinazoline inhibitors, which bind to the histone substrate-binding site of G9a,^941–946 can also possess weak off-target activity against NSD2 in vitro.^947,948 After initial difficulties with structure-guided optimization, extensive structure-activity relationship studies have recently discovered analogs that potently inhibit NSD2 and are selective against G9a and several other non-NSD methyltransferases.^949,950 While still early, preliminary biological studies with these compounds suggest they can preferentially inhibit the growth of cell lines with oncogenic NSD2 alterations and suppress tumor growth in mouse xenograft models.⁹⁴⁹

Figure 33.

NSD2 drug discovery. (a) Summary of NSD2 domain structure for drug discovery. (b) Small-molecule inhibitors of the NSD SET domain. (c) Discovery and optimization of NSD2-PWWP1 ligands and structure of UNC6852 bound to NSD2-PWWP1 (PDB: 6XCG). (d) Small-molecule degraders of NSD2 based on UNC6834 and a summary of their mechanism of action.

The NSD2 SET domain has shown signs of ligandability with other scaffolds as well, but none have yielded potent and selective inhibitors.^951–953 For example, the natural product sinefungin, which is an analog of the SAM cofactor used by the NSD2 SET domain, is known to inhibit NSD2 but is weakly potent and has yet to afford highly active analogs.^948,954 A recently developed high-throughput screening assay for NSD2 may afford additional scaffolds, but the hits identified in the first reported screen possess notable liabilities.⁹⁵¹ The assay for this screen used native nucleosomes purified from HeLa cells as the NSD2 substrate, since NSD2 is not catalytically active on synthetic histone peptides,⁹⁵⁵ ultimately enabling high-throughput screens.⁹⁵¹ However, in a preliminary screen of approximately 16,000 compounds, most confirmed hits show notable liabilities. For example, the screening hit, DA3003–1, a previously reported Cdc25 phosphatase inhibitor, was validated to bind the SET domain and inhibit NSD2 activity, but it was also found to inhibit dozens of other methyltransferases.⁹⁵¹ Some improvements in selectivity have been reported for structural analogs of DA3003–1,⁹⁵⁶ but DA3003–1 is known to participate in redox cycling and was originally proposed to engage NSD2 through nonspecific interactions.⁹⁵¹ While NSD2 SET has proven a challenging target for discovery chemistry, further improvements are likely possible. Indeed, a Phase 1 clinical trial is currently underway to evaluate KTX-1001, described as a selective NSD2 inhibitor, for the treatment of multiple myeloma (NCT05651932).⁹⁵⁷ Recent work described the discovery of potent, selective NSD2 catalytic inhibitors related to KTX-1001, IACS-17596 and IACS-17817) that deplete H3k36me2, reprogram chromatin and suppress oncogenic transcriptional programs in KRAS-driven pancreatic and lung cancer models.⁹⁵⁸ These compounds act through a SAM-competitve mechanism coupled with catalytic channel obstruction, leading to durable tumor regression and prolonged survival in preclinical systems.⁹⁵⁸ Together, these findings highlight the therapeutic potential of directly targeting NSD2’s enzymatic activity, complementing ongoing clinical evaluation of KTX-1001 in multiple myeloma.

Selectivity within the NSD family has recently been accomplished with the discovery of covalent NSD1 inhibitors.⁹⁵⁹ NSD1 is involved in recurrent leukemogenic chromosomal translocations that fuse it with NUP98, making it an attractive target for drug discovery.^960–962 A series of covalent NSD1 inhibitors were recently developed from a benzothiazole scaffold identified in an NMR-based fragment screen of 1,600 compounds.⁹⁵⁹ The initial benzothiazole fragment hit, BT1, was fashioned with a thiocyanate handle after NMR studies showed compound-induced perturbations to the chemical shift of Cys2062.⁹⁵⁹ A co-crystal structure with the resulting compound, BT3, showed formation of a disulfide bond and illustrated the emergence of a channel-like pocket that is absent in the apo structure.⁹⁵⁹ Replacing the thiocyanate with a methyl aziridine yielded BT5, which irreversibly engages NSD1 SET.⁹⁵⁹ The compound binds NSD1-SET in cells, inhibits HMT activity, and selectively inhibits the growth of murine bone marrow progenitor cells transformed with the oncogenic NSD1-NUP98 fusion over other oncogenes.

3.7.2. PWWP domains and the NSD family

The NSD2 PWWP1 domain, and PWWP domains more generally, represent tractable targets for ligand discovery. The PWWP1 domain of NSD2 binds to H3K36me2-modified nucleosomes through a conserved aromatic cage, making cation-π interactions with the methyl lysine ammonium.^963–965 BI-9321, a chemical probe for NSD3, was the first small-molecule ligand to be reported for a PWWP domain.⁹⁶⁶ NSD3 has been proposed to mediate the binding of the chromatin remodeler protein, CHD8, to BRD4, providing a function that is indispensable for acute leukemia survival.⁹⁶⁷ This function is specific to the short isoform of NSD3 which includes its first, but not second, PWWP domain. Fragment-based ligand discovery for the NSD3 PWWP1 domain previously identified a methylimidazole core that makes CH-π interactions with the conserved PWWP aromatic cage that binds methyl lysine.⁹⁶⁶ Structure-based design afforded a ligand, BI-9321, that possesses mid-nanomolar affinity for NSD3 and is selective against 14 other PWWP domains tested when assayed by DSF.⁹⁶⁶ High concentrations of the ligand were required to show anti-leukemia activity in tissue culture, suggesting that additional improvements in ligand affinity would be required to more fully validate NSD3 as a target for drug discovery.⁹⁶⁶ However, BI-9321 has been used to construct NSD3 PROTACs, such as MS9715, and additional scaffolds have recently been reported to support targeted protein degradation.^968–970 MS9715 shows weakly potent but selective degradation of NSD3 and modest anti-leukemia activity in tissue culture.

In contrast to NSD3, the PWWP domain of LEDGF (also known as PSIP1) has proven more challenging for ligand discovery. LEDGF is selectively required for transformation and oncogenic gene expression in MLL-rearranged leukemias.⁹⁷¹ Its PWWP domain binds to nucleosomal H3K36me3 and is important for facilitating the binding of MLL to its chromatin targets.^972–974 While a recently reported DNA encoded library (DEL) screen failed to identify any starting points for LEDGF PWWP ligand optimization, a chemoproteomics-based method for mapping ligand-binding sites in cells suggests that the PWWP proteins, HDGF and LEDGF, can be stereoselectively bound by small-molecule ligands.⁹⁷⁵

Small-molecule ligands of the NSD2 PWWP1 domain were first identified from a virtual screen for the PWWP domain of ZMYND11.⁹⁷⁶ While none of the hits from this virtual screen were found to bind ZMYND11, additional testing against other PWWP domains led to the discovery of a fragment-like ligand with mid-micromolar affinity for NSD2 PWWP1. A scaffold-hopping approach was then used to identify a ligand with a greater number of commercially available analogs that could be used to evaluate SAR. This led to the discovery of a tertiary N-cyclopropyl amide compound with single-digit micromolar affinity. A co-crystal structure was solved, which confirmed engagement of the methyllysine-binding site of NSD2 and showed that the isopropyl group, which was critical for activity in SAR studies, fills a narrow hydrophobic pocket at the bottom of the aromatic cage. This molecule, MR837, demonstrated that the NSD2 PWWP1 domain is a ligandable target and, furthermore, served as a crucial starting point for the optimization of potent and selective chemical probes (Figure 33c).

Structure-guided optimization of MR837, via the key intermediate, MRT866, led to the discovery of UNC6934, a potent and selective chemical probe for the NSD2 PWWP1 domain (Figure 33c).⁹⁷⁷ It binds NSD2 PWWP1 with an affinity of 91 nM and is selective against all PWWP domains tested. Fascinatingly, liberation of the cyclopropyl ring to the corresponding isopropyl was sufficient to entirely abrogate compound binding, providing a useful inactive negative control, UNC7145. UNC6934 effectively disrupts the interaction between purified H3K36me2 nucleosomes and the isolated PWWP1 domain of NSD2, but cannot do the same to full-length NSD2 protein, likely because it makes additional contacts with nucleosomal DNA outside of the methyllysine-binding pocket. Crystallographic studies proved informative, highlighting that UNC6934 binds adjacent to the basic DNA binding surface in the canonical methyllysine binding pocket of the PWWP1 domain, but does not compete with DNA. This structure again highlighted the importance of the constrained geometry of the cyclopropyl group, which inserts into the aromatic cage with exceptional shape complementarity. It also allowed for the rational design of a biotin-conjugated affinity probe, which was used to demonstrate the exquisite selectivity of UNC6934 for NSD2. In cells, inhibition of the NSD2 PWWP1 domain by UNC6934 modestly increases nucleolar accumulation of NSD2, but it does not displace NSD2 from chromatin globally, reduce H3K36me2 levels, or impair the growth of multiple myeloma cells. This highlights a broader challenge with domain-selective probes: while they can be useful for dissecting local structural functions, targeting a single domain of a large, multidomain chromatin regulator may not phenocopy inhibition or loss of the full-length protein, limiting their translational potential.

Due to PWWP1 engagement eliciting minimal phenotypic effects, UNC6934 was quickly repurposed for targeted protein degradation. However, rather than conjugating UNC6934 to the CRBN or VHL ligands that are most frequently used for PROTAC discovery, the first reported NSD2 degraders were designed by decorating UNC6934 with basic amino acid side chains.⁹⁷⁸ Inspired by a previous report,⁹⁷⁹ this approach was meant to recruit UBR-box E3 ligases that recognize N-degons that are rich in arginine, lysine, and histidine. Using the same attachment site as previously used to synthesize biotinylated analogs, simple heterobifunctional constructions of UNC6934 and arginine were found to cause modest NSD2 degradation.⁹⁷⁸ An optimized molecule, UNC8153, which features a simple polymethylene-linked primary amine, was found to elicit potent (sub-micromolar) and exclusive degradation of NSD2 (Figure 33d). However, while a free N-terminal alpha amino group is necessary for degron recognition by UBR-box domain E3 ligases,^980–982 eliminating this element still afforded active degraders and knockdown of UBR proteins could not rescue NSD2 degradation.⁹⁷⁸ A later study found that the primary amine of UNC8732 is converted to an aldehyde, which forms a reversible covalent bond with the E3 substrate receptor, P22, to mediate targeted protein degradation (Figure 33d).⁹⁸³ These findings were reported contemporaneously with an independent discovery of FKBP12 degraders that proceed through the same mechanism. Both studies found that a previously reported XIAP degrader proceeds through this mechanism,^984–986 highlighting it as a generalizable approach to convert ligands into degraders.⁹⁸⁵ (Another contemporaneous report, which used a CRISPR activation screen to uncover an alternative mechanism of ligand-induced degradation through FBXO22, further established the ability to FBXO22 to support targeted protein degradation.³¹²)

UNC8153-induces rapid degradation of NSD2, which causes a delayed reduction in H3K36me2 levels after nearly a week of treatment, likely demonstrating that H3K36me2 modifications must be lost by dilution through cell divisions.⁹⁷⁸ While UNC8153 was shown to reduce the growth of multiple myeloma cells expressing NSD2-E1099K, these effects are modest and only observed at high concentrations of the drug.⁹⁷⁸ Furthermore, UNC8153 is completely ineffective against t(4;14)-positive multiple myeloma cells in vitro. This is potentially due to the sub-maximal degradation elicited by UNC8153, as a more effective analog, UNC8732, showed more potent anti-proliferative activity.⁹⁸⁴ UNC8732, which features a constrained linker attachment site but preserves the primary amine modification, is 10-fold more potent and elicits more complete degradation of NSD2 than to UNC8153 (Figure 31c). It also suppresses the growth of ALL cells expressing NSD2 E1099K in tissue culture, although multiple weeks of drug exposure are required to observe this effect.⁹⁸⁴

Multiple CRBN-based NSD2 PROTACs have now been reported. MS159, which is constructed from the UNC6934 scaffold, degrades NSD2 through CRBN but elicits off-target degradation of the CRBN neo-substrates, IKZF1 and IKZF3.⁹⁸⁷ Since IKZF1/3 degradation is responsible for the therapeutic effect of thalidomide analogs in multiple myeloma,^988,989 the degree to which NSD2 degradation is responsible for MS159 anti-proliferative effects is unclear. The PROTAC LLC0424, which boasts a 5’ thalidomide moiety linked to a UNC6934 derivative through a conformationally rigid amido piperidine linker, induces potent and exclusive degradation of NSD2 and is active both in vitro and in vivo.⁹⁹⁰ The steadily improving activity of NSD2 degraders suggest that they may have substantial potential for clinical development alongside the undisclosed SET domain inhibitor, KTX-1001.

3.7.3. SETD2

The histone methyltransferase SETD2 is the exclusive writer of the trimethylation mark on lysine 36 of histone H3 (H3K36),⁹⁹¹ a chromatin modification linked to transcriptional fidelity, RNA splicing, DNA repair, and genomic stability.^992–994 While SETD2 has been shown to function as a classic tumor suppressor in many solid malignancies—where it is often biallelically inactivated^995,996—emerging evidence supports a different and targetable role in hematologic malignancies. SETD2 mutations are commonly heterozygous in leukemias, lymphomas, and multiple myeloma, suggesting a haploinsufficiency as opposed to complete loss of function.⁹⁹⁷ This partial loss sensitizes these cells to SETD2 inhibition, creating a unique therapeutic opportunity, especially in genetically or epigenetically primed contexts. Some flagship examples include that of t(4;14) multiple myeloma, as well as in MLL-rearranged leukemias. In t(4;14) MM, NSD2 overexpression leads to elevated H3k36me2 levels, driving cells toward a reliance on SETD2 to mediate the balance between di- and tri-methylated states.⁹⁹⁸ In MLL-rearranged leukemia, SETD2 disruption alters transcriptional programs and DNA repair pathways; selective inhibition of SETD2 catalytic activity may phenocopy or exacerbate these states, leading to synthetic lethality while sparing healthy tissue.⁹⁹⁹ Furthermore, recent pan-cancer analyses demonstrate that SETD2 mutation or reduced expression is associated with widespread dysregulation of DNA methylation across diverse cancer types—including hematologic and solid tumors—suggesting that SETD2 loss may drive tumorigenesis via epigenetic reprogramming beyond histone methylation.¹⁰⁰⁰ Altogether, these biological observations have motivated chemical efforts to fashion potent and selective SETD2 inhibitors to continue to study SETD2 biology.

The first selective chemical probes for SETD2 inhibition were derived from the cofactor analog sinefungin by introducing simple alkyl substituents, yielding compounds such as N-propyl sinefungin (Pr-SNF) and N-benzyl sinefungin.¹⁰⁰¹ These analogs showed sub-micromolar potency (IC₅₀ = 0.48–0.80 μM) and more than ten-fold selectivity for SETD2 over related methyltransferases, providing the first demonstration that selective inhibition of SETD2 is feasible.¹⁰⁰¹ Beyond serving as early tool compounds, they highlighted the tractability of SETD2 as a drug target and set the stage for subsequent inhibitor development by establishing important structural foundations.

Building on these insights, Epizyme undertook a structure-based drug discovery campaign, screening a methyltransferase-biased library and identifying a 2-amidoindole scaffold with activity against SETD2.¹⁰⁰² Medicinal chemistry optimization yielded EPZ-719, a potent, reversible SETD2 inhibitor with nanomolar biochemical potency (IC₅₀ = 8 nM) and low-nanomolar activity in cellular assays monitoring H3K36me3 levels. EPZ-719 demonstrated sufficient metabolic stability and selectivity to serve as an in vivo tool compound for target validation studies.¹⁰⁰² However, it lacked adequate solubility and oral bioavailability to advance into clinical development.

To address these limitations, a conformational design strategy was used, incorporating a rigid cis-(1R,3S)-diaminocyclohexyl linker to enhance metabolic stability and oral pharmacokinetics.¹⁰⁰³ This approach yielded EZM0414, a highly selective and orally bioavailable SETD2 inhibitor with potent biochemical and cellular activity (IC₅₀ = 18 nM and 31 nM, respectively) and robust antiproliferative effects in multiple myeloma cell lines bearing t(4;14) translocations.¹⁰⁰³ Crystallographic studies confirmed that EZM0414 binds deep within the lysine channel of SETD2, displacing autoinhibitory residues and stabilizing the enzyme in an open conformation.¹⁰⁰³ EZM0414 exhibited favorable ADME properties and demonstrated in vivo activity in xenograft models, ultimately entering clinical evaluation for the treatment of multiple myeloma and B-cell malignancies.¹⁰⁰³

The evolution from sinefungin-based probes to EPZ-719 and finally to the clinical candidate EZM0414 exemplifies the power of rational, structure-guided drug discovery in the epigenetic landscape. In hematologic cancers, where SETD2 functions as a haploinsufficient dependency rather than a tumor suppressor, these chemical probes provide critical tools for mechanistic dissection and therapeutic intervention.

4. State- and lineage-specific dependencies

In addition to genetically restricted vulnerabilities, some targets show cancer-specific activity that is independent of a specific genetic alteration. These include lineage-specific dependencies, in which a target or a drug-like chemical agent shows activity in specific cancer lineages, sometimes without a clear understanding of the underlying basis for its cancer-specific effect. State-specific dependencies—factors that are important only under certain conditions or cell states—can also make for compelling anti-cancer drug targets.

4.1. LSD1

LSD1 (lysine-specific histone demethylase 1, also known as KDM1A) is a flavin adenine dinucleotide (FAD)-dependent lysine demethylase and transcriptional co-regulator. Before being reported as the first known example of a histone demethylase,¹⁰⁰⁴ LSD1 had initially been identified in transcriptional co-repressor complexes (e.g. CoREST and NuRD) and found to possess homology with amine oxidases.^1005–1008 LSD1 regulates gene expression by removing H3K4 and H3K9 mono- and di-methylation marks from gene enhancers and also through non-enzymatic scaffolding functions.^{1004,1009–1015} Its FAD-dependent mechanism of catalysis, which requires a protonated amine as a substrate, makes it unable to demethylate the trimethylated H3K4 modification that is primarily present at promoters.¹⁰⁰⁴ Current interest in LSD1 drug development is largely motivated by experiments showing that LSD1 is required for AML and lung cancer, in addition to other malignancies.^{1011,1013,1016–1025}

The earliest efforts to inhibit LSD1 used mechanisms similar to inhibitors of related monoamine oxidase enzymes, which covalently modify the FAD cofactor within the enzyme active site.¹⁰²⁶ The first, a peptidic inhibitor based on the H3K4 substrate, used a propargylamine analog of Lys4 to form a covalent adduct with FAD, making it a suicide inhibitor of LSD1 enzyme function.^1027–1030 In a contemporaneous report, the FDA-approved antidepressant monoamine oxidase inhibitor, trans-2-phenylcyclopropylamine (tranylcypromine or TCP), was found to be the first small-molecule and cell-active inhibitor of LSD1.^1031,1032 Tranylcypromine reacts to form a covalent linkage to FAD within a hydrophobic pocket of the LSD1 active site (Figure 34a,b).^1031–1034 Many groups have discovered or designed tranylcypromine-based LSD1 inhibitors with substitutions on both the phenyl ring and the primary amino group,^{1016,1033,1035–1037} which has afforded several clinical-stage tranylcypromine-based LSD1 inhibitors.^{1031,1033,1035,1038–1040} These include GSK2879552, ORY-1001 (iadademstat), bomedemstat, INCB059872, and JBI-801 (LSD1/HDAC dual inhibitor) (Figure 34c).^{1016,1021,1041–1045} In addition to these irreversible inhibitors, two structurally differentiated reversible inhibitors, pulrodemstat (CC-90011) and seclidemstat (sp-2577), have also advanced to clinical investigation (Figure 34c).^1046–1048

Figure 34.

LSD1 inhibitors. (a) Formation of the TCP-FAD tetracyclic adduct by LSD1 inhibitors. (b) Crystal structure of (+)-tPCPA in complex with LSD1 (PDB: 2XAH). Important interactions of TCP-FAD with LSD1 amino acids shown in green. (c) Chemical structures of additional irreversible inhibitors (TCP moiety highlighted in pink). (d) Structure of reversible LSD1 inhibitors.

ORY-1001 was discovered using computational modeling of LSD1 crystal structures to assist in the design of more than 800 tranylcypromine analogs.¹⁰²¹ It is a potent (IC₅₀ = 18 nM) and selective inhibitor of LSD1, binding rapidly and irreversibly to FAD. It elicits antiproliferative effects at low nanomolar concentrations in AML cell lines, causing an increase in terminal differentiation, and is active against AML and T-ALL xenografts in vivo. As a clinical agent, ORY-1001 has shown promising signals for tolerability and efficacy in a Phase 1 report of relapsed and refractory acute myeloid leukemia (R/R AML),¹⁰⁴⁵ further encouraging LSD1 as a target for drug development. Due to promising pre-clinical results, iadademstat was explored as a combinatorial therapy, recently progressing to a Phase 2 clinical trial in combination with the azacitidine.¹⁰⁴⁹ In patients with newly diagnosed AML, the therapeutic combination displayed a manageable safety profile and showed an encouraging 52% of patients achieving complete remission.¹⁰⁴⁹ Additional clinical trials exploring other combinatorial therapies with ORY-1001 in myeloid malignancies and solid tumors are ongoing (NCT05546580, NCT05420636, NCT06357182, and NCT06287775).

Several years before ORY-1001 was reported, structurally related compounds were discovered in a screen of 2.5 million compounds using a biochemical assay of LSD1 enzyme activity.¹⁰¹⁶ Optimization efforts led to the discovery of the selective LSD1 inhibitors, GSK-LSD1 and GSK2699537, as well as the clinical agent, GSK2879552 (Figure 34c).¹⁰¹⁶ GSK-LSD1 and GSK2879552 both contain a tranylcypromine core and a piperidine moiety, while GSK2879552 includes a benzoic acid extending from the piperidine nitrogen. In a proliferation screen, GSK2879552 inhibited the growth of AML and some small cell lung carcinoma SCLC cell lines. This translated to in vivo antitumor activity in SCLC xenograft mouse models,¹⁰¹⁶ which has motivated the clinical translation of GSK2879552 and other tranylcypromine-based inhibitors, such as INCB059872,¹⁰⁵⁰ for SCLC. However, despite its preclinical promise, clinical trials using GSK2879552 were suspended due to severe adverse effects and insubstantial efficacy.¹⁰⁵¹ In mouse models, GSK2879552 was reported to cause anemia, but the most common clinical toxicity in humans was thrombocytopenia and the most serious adverse event was encephalopathy, which caused several patient deaths.¹⁰⁵¹ Fortunately, encephalopathy has not been observed in other clinical trials, suggesting it is not likely a class-wide effect of on-target LSD1 inhibition. According to clinical trial records (NCT02712905), a strategic business decision was made to terminate clinical investigation of INCB059872, leaving only ORY-1001 and bomedemstat (IMG-7289) in clinical development. IMG-7289, which has advanced to Phase 2b clinical trials to assess pharmacokinetic and safety in patients with myelofibrosis,¹⁰⁵² is effective in preclinical models of myeloproliferative neoplasms (MPN), castration resistant prostate cancer, and AML, among other cancers.^{1041,1053–1055}

Despite being designed and demonstrated to inhibit the enzymatic activity of LSD1, it is now clear that the effects of LSD1 inhibitors are primarily mediated by disrupting a protein-protein interaction with GFI1.^{1011,1013,1014} This was first uncovered in characterization of the tranylcypromine-based inhibitor, T-3775440, which was found to inhibit LSD1 interactions with the transcriptional repressor, GFI1, while leaving the CoREST complex intact.¹⁰¹¹ A study published shortly after this initial report found that the tranylcypromine-based inhibitor, OG86, alters gene transcription much earlier than it impacts histone lysine methylation levels, which led to the discovery that the demethylase activity of LSD1 is not required for LSD1-dependent AML growth and survival.¹⁰¹³ Instead, OG68 also disrupts the LSD1-GFI1 interaction, leading to loss of LSD1 from chromatin and a decrease in GFI1-mediated transcriptional repression.¹⁰¹³ A CRISPR-based method for high-density mutagenesis that was later used to uncover mechanisms of resistance to LSD1 inhibitors discovered that several mutations conferring drug resistance to LSD1 inhibitors inactivated enzyme function, unambiguously demonstrating that demethylase inhibition is not necessary for the phenotypic effects of LSD1 inhibitors.¹⁰¹⁴

The interaction between LSD1 and GFI1 has also been highlighted in a study showing that LSD1 inhibitors are effective against GFI1-driven medulloblastoma, both in vitro and in vivo.¹⁰⁵⁶ Interestingly, it was recently discovered that UM171,¹⁰⁵⁷ a small molecule that induces the degradation of LSD1/CoREST through the E3 substrate receptor KBTBD4,^133,1058 mimics the effects of medulloblastoma mutations in KBTBD4 that also lead to neomorphic degradation of LSD1/CoREST.¹³¹ Mechanistically, UM171 induces CoREST degradation by binding into the active site of HDAC1/2 and stabilizing their interaction with KBTBD4, ultimately making contacts with HDAC1/2 that are analogous to those made by mutant forms of KBTBD4 in medulloblastoma, both of which lead to LSD1/CoREST degradation.^129,130 This pro-tumorigenic effect of LSD1/CoREST degradation stands in contrast to LSD1 inhibitors and may provide an important phenotype to monitor in future preclinical and clinical investigation.

A reversible LSD1 inhibitor, puldrodemstat (CC-90011) has also advanced to clinical trials for AML and SCLC (Figure 34d).^{1046,1047,1059} CC-90011 was developed from a hit identified in a high-throughput screen of more than 300,000 compounds and has an optimized pyrimidinone core (Figure 34d).¹⁰⁵⁹ It binds reversibly to LSD1, contacting the catalytic lysine and extending into the FAD-binding pocket, selectively inhibiting LSD1 activity with sub-nanomolar potency and showing anti-AML effects at single-digit nanomolar concentrations. Clinical trial results will be required to evaluate the benefits of using covalent tranylcypromine-based inhibitors versus this new class of reversible LSD1 inhibitors, but a recent comparative study of LSD1 inhibitors found that while CC-90011 and ORY-1001 are equivalently potent enzymatic inhibitors (the most potent among all clinical stage inhibitors of LSD1), CC-90011 is a weaker inhibitor of both LSD1-GFI1 interaction and AML cell growth.¹⁰⁶⁰

4.2. KDM6B

The Jumonji C (JmjC) family of histone demethylase enzymes differ from those of the LSD family through their demethylase mechanisms. Unlike the LSD enzymes that utilize an amine oxidase reaction in the removal of methylation marks, the Jumonji domain family are iron-containing demethylase enzymes that use α-ketoglutarate as a cofactor to catalyze a hydroxylation reaction that can also operate on trimethyllysine substrates.^1061–1068 KDM6A (UTX), and KDM6B (JMJD3) selectively demethylate H3K27me2/3,^1069–1071 reversing transcriptionally repressive chromatin states mediated by PRC1 and PRC2.⁶⁰⁰ A loss of these marks is associated with stem-like gene expression during development and KDM6A/B are essential for normal transcriptional programs in development and for oncogenic programs in cancers such as T-cell acute lymphoblastic leukemia (T-ALL) and glioma.^1072–1074

The first KDM6 inhibitor, GSK-J1, was developed using structure-guided optimization of a weak hit identified in a screen of 2 million small molecules (Figure 35a).¹⁰⁷⁵ GSK-J1 binds competitively with a-ketoglutarate but not with the histone substrate. Its propanoic acid mimics the a-ketoglutarate cofactor binding, the tetrahydrobenzazepine fits into a narrow cleft to mimic a proline residue on the histone substrate (Pro30) and the pyridylpyrimidine interacts with the Fe(II) cofactor, which is critical for inhibition of JMJD3 (Figure 35b).¹⁰⁷⁵ The polar carboxylate of GSK-J1 limits cell permeability, so GSK-J4, an ethyl ester prodrug of GSK-J1 is used for cell-based studies (Figure 35a).¹⁰⁷⁵ GSK-J4 significantly reduces the proliferation of T-ALL, prostate, glioma, and other cancers.^{1072,1074,1076} Unlike LSD1, these effects appear to be due to the catalytic activity of JMJD3, since introduction of catalytically inactive JMJD3 could not rescue the effects of GSK-J4 inhibition.¹⁰⁷² More advanced compounds, including those with activity in animal model systems, are awaited for further evaluation of KDM6B as a potential cancer target.

Figure 35.

Jumonji domain inhibitors. (a) chemical structures of JMJD3 inhibitor GSK-J1 and prodrug GSK-J4. (b) Crystal structure depicting the essential interactions of GSK-J1 with the iron atom (pink) in the JMJD3 active site (PDB: 4ASK). (c) Chemical structure of pan-KDM5 family inhibitor CPI-455.

4.3. KDM5/6 and drug-tolerant persistence

The KDM5 family of lysine demethylases, part of the larger JmjC-domain-containing class, play critical roles in gene regulation and chromatin remodeling. These enzymes, which demethylate H3K4, are associated with gene repression.¹⁰⁷⁷ Overexpression and amplification of KDM5 proteins, particularly KDM5A and KDM5B, have been observed in a wide variety of cancers where they regulate the expression of both tumor suppressor genes and oncogenes.^1078–1084 KDM5 and KDM6 family members are also implicated in maintaining drug-tolerant cell states.^1085–1089 Drug-tolerant persister (DTP) cells represent a population of cancer cells that emerge following drug treatment and are characterized by a slow cell cycle and a reversible state of drug insensitivity. Unlike drug resistance driven by genetic mutations, DTP cells evade treatment through non-genetic mechanisms, such as chromatin rewiring and transcriptional changes. Notably, these cells often regain drug sensitivity and proliferative growth after treatment is removed, making them a key target for therapeutic intervention. Gene expression profiling of DTP cells has shown that the expression of KDM5A/B is elevated in these populations and that DTPs are dependent on KDM5 enzymes, specifically their enzymatic activity for maintaining the DTP state.¹⁰⁸⁸ CPI-455, a selective pan-KDM5 family inhibitor that binds competitively with α-ketoglutarate, has shown the ability to prevent the survival of DTP cells (Figure 35c).¹⁰⁹⁰ While more potent KDM5 inhibitors show activity in a variety of cancers, none of these have progressed to the clinic.^1091–1096

4.4. GLP/G9a

G9a is a SET domain containing methyltransferase protein which specifically catalyzes the mono- and di-methylation of H3K9 (H3K9me1/2), modifications associated with transcriptional repression.¹⁰⁹⁷ Aberrant H3K9 methylation has been observed in pathogenesis where it contributes to the silencing of tumor suppressors, such as p53.^1098–1103 G9a and its paralog, GLP (G9a-like protein), are primarily responsible for the addition of this modification and G9a overexpression has been observed in many cancers and is correlated with a poor prognosis.¹¹⁰⁴ Knockdown of G9a reduced H3K9me2 at silenced loci leads to the re-expression of tumor suppressor genes, inhibition of cell proliferation and invasion, and supports G9a as a therapeutic target.^1105,1106

There are two main classes of G9a/GLP inhibitors: SAM-competitive and substrate-competitive inhibitors. The most successful and specific compounds are those that fall into the substrate-competitive category. The first substrate inhibitor for GLP/G9a emerged from a high-throughput screen for inhibition of enzyme activity. This compound, BIX-01294, specifically inhibited the methyltransferase activity of GLP/G9a (IC₅₀=1.7 μM and IC₅₀=38 μM respectively) without activity on other methyltransferases (up to 45 μM).^941,942 However, BIX-01294 is cytotoxic at concentrations > 4.1 μM limiting its use. SAR of BIX-01294 lead to the discovery of several compounds and the report of G9a-inhibitor co-crystal structures.^{943–946,1107} The first structure of G9a-UNC0224 validated compound binding and facilitated the discovery of subsequent structure-based inhibitors with greater potency, permeability, and selectivity, many of which possessed the quinazoline scaffold of BIX-01294.⁹⁴⁴ Further optimization of the substituents off the quinazoline scaffold lead to the development of UNC0642, an in vivo probe suitable for animal studies. UNC0642 had greater in vivo activity, potency (IC₅₀ < 2.5nM), and selectivity (>2,000 fold) over other histone methyltransferases.¹¹⁰⁸ Additionally, expanding the quinazoline ring to a benzodiazepine led to a novel chemotype of compounds with similar activity as the previous scaffold.¹¹⁰⁹

Recently, molecular glue degraders for the transcription factor WIZ (dWIZ-1/2) were disclosed.¹¹¹⁰ WIZ was revealed as a repressor of fetal hemoglobin via association with repressive complexes and GLP/G9a to facilitate methylation and gene repression.^1111,1112 Degradation of WIZ reduced H3K9me2 and increased the expression of GLP/G9a gene targets. dWIZ-2 toxicity was assessed in >300 cell lines and was revealed to be significantly less cytotoxic than G9a/GLP inhibition with UNC0642.¹¹¹⁰ This finding presents a novel approach for targeting proteins via disrupting or degrading G9a/GLP interactors.

5. Conclusion

By focusing on the potential for chromatin regulation to support the discovery of anti-cancer therapeutics, and by reviewing chromatin targets from the perspective of the genetic evidence that might influence their chances for therapeutic success, we have attempted to make for a manageable perspective on the field as it currently stands. Several reviews of exceptional quality have been published previously on related subjects in chromatin and drug discovery,^1113,1114 including one published while this work was in development that is similarly organized around genetic rationale for target selection.¹¹¹⁵ We sought a unique perspective in highlighting the evolution of chemistries that have enabled target prosecution and therapeutic translation. Still, we must sincerely apologize for any references missed within this work, especially from among the many biological studies that follow the discovery of a new chemical agent.

Assisted tremendously by the growing sophistication of chemical biology as a discipline, the study of chromatin biology has been advanced immeasurably by the pursuit of chemical tools. The discovery of HDACs was made possible by the study HDAC inhibitors, which provided a first concrete example of transcriptional control as a chromatin-dependent regulatory process. The first therapeutic hypotheses originated from these early studies, propelling HDAC inhibitors into the clinic where they continue to find use today. Buoyed by human genetics, functional genomics, and chemical biology, far more targets are now at our disposal, which will undoubtedly animate many more years of groundbreaking discoveries in chromatin biology, cancer pharmacology, and, hopefully, clinical development.

These new targets can take valuable lessons from the historical evidence supporting targets like HDACs and DNMTs, which revealed the therapeutic potential for targeting essential genes while also highlighting the challenges associated with their therapeutic windows. Target-based drug discovery has dominated chromatin drug discovery for many years now, but resources like DepMap have only become widely available in the last 10 years. While EZH2 and menin predate the existence of large-scale cancer dependency maps, the clinical successes for these cancer-specific vulnerabilities stand out as particularly hopeful signs for the richly informed targets of the future. We expect that ongoing clinical trials for targets like KAT6/7, LSD1, CBP/p300, and SWI/SNF, among others, will prove similarly instructive, giving new opportunities for us to revisit this document and evaluate—or reevaluate—the qualities that make for successful target selection.

As our synthesis of chromatin regulation and tumor biology continues to improve, we can expect continued challenges for discovery chemistry. Attractive targets like those formed by the paralog synthetic lethality between CBP/p300 and SMARCA2/4 have already provided formidable challenges for drug discovery, but new innovations have provided inroads to these otherwise intractable targets. Targeted protein degradation, which has great potential for achieving the selectivity needed to exploit paralog synthetic lethality, has only become a widespread field of study in the last 10 years. Importantly, there was no thought of being able to address paralog synthetic lethality with target protein degradation at its earliest inception,¹¹¹⁶ highlighting the need to continue innovating new approaches in drug discovery and chemical biology. The emerging ability to rewire chromatin regulatory machinery is among the most exciting new possibilities, as it enables gain-of-function outcomes that are not possible with conventional inhibitors. Importantly, TCIPs, synTEFs, and related proximity agents enable small molecules that target common essential proteins to be repurposed as recruitment handles that redirect an essential protein for therapeutic benefit. This could allow for unprecedented precision in chromatin modulation and a new generation of therapeutic agents with unique considerations for target selection. We await these and other advances with great excitement.

Biographies

Paige Barta earned her B.A. in Biochemistry and Molecular Biology from Lewis & Clark College in 2020. After graduation, she worked as a research technician in Jared Rutter’s laboratory at the University of Utah before beginning her graduate studies at The Scripps Research Institute. She is currently pursuing her Ph.D. in the Erb Group, where she couples chemical probe perturbations with transcriptional genomics and forward genetics to study transcriptional control in cancer.

Trever Carter obtained his B.S. in Biochemistry from the University of Notre Dame in 2019. After a year of working with Prof. Craig Lindsley as a research assistant, he joined the Medical Scientist Training Program at the University of California – San Diego in 2020. Currently, he is pursuing his PhD in the Erb Group at The Scripps Research Institute, where he utilizes high-throughput chemistry toward the prospective discovery of molecular glues.

Michael Erb is an Associate Professor in the Department of Chemistry at The Scripps Research Institute. Before starting his independent career as a Scripps Fellow, he received his B.A. in Biochemistry from Claremont McKenna College (2014) and his PhD from Harvard University (2017) under the mentorship of Prof. James Bradner. The Erb laboratory combines expertise in transcriptional biology and discovery chemistry to drug cancer-associated transcriptional regulatory pathways. His research has been recognized and supported by an NIH Director’s Early Independence Award and the Ono Pharma Foundation Breakthrough Science Initiative Award.