Advancing targeted protein degradation for cancer therapy

Abstract

The human proteome contains approximately 20,000 proteins, and it is estimated that more than 600 of them are functionally important for various types of cancers, including nearly 400 non-enzyme proteins that are challenging to target by traditional occupancy-driven pharmacology. Recent advances in the development of small-molecule degraders, including molecular glues and heterobifunctional degraders such as proteolysis-targeting chimeras (PROTACs), have made it possible to target many proteins that were previously considered undruggable. In particular, PROTACs form a ternary complex with a hijacked E3 ubiguitin ligase and a target protein, leading to polyubiguitination and degradation of the target protein. The broad applicability of this approach is facilitated by the flexibility of individual E3 ligases to recognize different substrates. The vast majority of the approximately 600 human E3 ligases have not been explored, thus presenting enormous opportunities to develop degraders that target oncoproteins with tissue, tumour and subcellular selectivity. In this Review, we first discuss the molecular basis of targeted protein degradation. We then offer a comprehensive account of the most promising degraders in development as cancer therapies to date. Lastly, we provide an overview of opportunities and challenges in this exciting field.

Extensive efforts have been made to map cancer dependency by identifying genes that, when functionally inactivated, inhibit the growth of cancer cells and are thus potential therapeutic targets¹. These efforts have resulted in the identification of many protein-coding genes, both mutated and non-mutated in human cancers, which are potential cancer targets^1–4. The Sanger Institute identified 627 priority targets, of which 232 are enzymes (37%)⁴, and these have been the focus of developing small-molecule anticancer drugs. However, the 395 non-enzyme targets, which include proteins such as transcription factors, generally do not have an active site that is functionally critical and amendable to targeting by occupancy-driven small-molecule inhibitors. Thus, they are often referred to as ‘undruggable’, and the vast majority of them have not been therapeutically targeted thus far.

Targeted protein degradation (TPD), the process of eliminating a protein of interest (POI), holds great promise for the development of cancer therapeutics⁵. A key focus of TPD is the development of heterobifunctional small-molecule degraders, including proteolysis-targeting chimeras (PROTACs), which contain two linked moieties with one binding the POI and the other binding an E3 ligase. The formation of an E3–degrader–POI ternary complex results in polyubiquitination of the POI and its subsequent degradation by the 26S proteasome. Since the PROTAC concept was first published in 2001 (REF⁶), a number of critical discoveries have been made (FIG. 1). The pursuit of improved prototype peptide-derived PROTACs together with the discovery of immunomodulatory imide drugs (IMiDs) acting as molecular glues to connect neosubstrates to an E3 ligase led to the development of highly effective PROTACs in 2015 (REFS^7–12). The field of TPD has seen explosive growth over the past 5 years. Remarkably, the first two PROTACs entered clinical development as cancer therapies in 2019 (REFS^13,14).

Fig. 1 |. Key discoveries and developments in the targeted protein degradation field.

The figure shows a timeline beginning from the first publication of the proteolysis-targeting chimera (PROTAC) concept up to the most recent discoveries. AR, androgen receptor; CRBN, cereblon; DCAF15, DDB1- and CUL4-associated factor 15; ER, oestrogen receptor; EZH2, enhancer of Zeste homologue 2; IMiD, immunomodulatory imide drug; POI, protein of interest; STAT3, signal transducer and activator of transcription 3; VHL, von Elippel–Lindau tumour suppressor.

In this Review, we discuss the molecular basis of TPD by the ubiquitin–proteasome system, plasticity of E3 ligases in targeting neosubstrates and considerations for developing novel E3 ligands. We also review the advancement of heterobifunctional small-molecule degraders as cancer therapeutics, with emphasis on those that are first-in-class, in or soon to be in clinical development, or efficacious in in vivo preclinical studies. Lastly, we outline key considerations for developing heterobifunctional degraders as cancer therapeutics, including limitations and potential opportunities. For non-cancer-related heterobifunctional and other degrader-related chronology, we refer readers to previous reviews^5,15–17.

The ubiquitin–proteasome system

The ubiquitin–proteasome system regulates most, if not all, cellular processes¹⁸. It covalently attaches ubiquitin, a 76-amino-acid protein that is highly conserved and ubiquitously expressed in all eukaryotic species and tissues, to a lysine residue in substrate proteins via an enzyme cascade involving a ubiquitin-activating enzyme (E1), a ubiquitin-conjugating enzyme (E2) and a ubiquitin ligase (E3)^19,20 (FIG. 2a). Ubiquitin can be ligated either as a single moiety to one lysine residue or multiple lysine residues or as a polyubiquitin chain where successive ubiquitin molecules are connected²¹. One of the well-known consequences of polyubiquitination is the proteolytic degradation of substrate proteins by the 26S proteasome²². The ubiquitin–proteasome system manages immense complexity and, at the same time, achieves a high degree of specificity of protein ubiquitination through a large repertoire of more than 600 E3 ubiquitin ligases in human cells²³ (FIG. 2b). Recurrent alterations of many E3 ligases have been found in numerous human diseases, including cancers²⁴. However, only three E3-targeting small-molecule drugs, thalidomide, lenalidomide and pomalidomide, which bind a substrate receptor of the E3 ligase cereblon (CRBN)²⁵, have been approved by the US Food and Drug Administration. This is largely due to the lack of an active site in the RING-type E3 ligases, which account for more than 95% of E3 ligases (FIG. 2b). While challenges in targeting E3 ligases remain, PROTAC technology has now made it possible to exploit the vast repository of E3 ligases and the sophisticated ubiquitin–proteasome system to target oncoproteins.

Fig. 2 |. The ubiquitin–proteasome system and classification of E3 ubiquitin ligases.

a | Ubiquitination begins with the ATP-dependent activation of ubiquitin (Ub) by the E1 ubiquitin-activating enzyme (UBA), resulting in Ub~E1 thioester bond formation (the tilde denotes a high-energy thioester bond), followed by the transfer of ubiquitin to an E2 ubiquitin-conjugating enzyme. Finally, an E3 ubiquitin ligase brings the substrate and the E2~Ub conjugate together and promotes ubiquitin ligation to a lysine residue in the substrate protein. Ubiquitin can be ligated either as a single moiety to one lysine residue or multiple lysine residues or as a polyubiquitin chain where successive ubiquitin molecules are connected. The consequences of ubiquitination for the substrate protein depend not only on whether it is monoubiquitinated or polyubiquitinated but also on the topology of the polyubiquitin chain²¹. One of the well-known consequences of polyubiquitination is the proteolytic degradation of substrate proteins by the 26S proteasome²². b | The three types of E3 ubiquitin ligases. Traditionally, E3 ubiquitin ligases are separated into homologous to E6AP carboxy terminus (HECT) and RING types. Both types interact with E2~Ub and recognize substrate proteins, but differ in their catalytic mechanisms. While the classical HECT E3 ligases form a thioester intermediate with ubiquitin on an active-site cysteine before transferring it onto its substrates²¹⁰, RING finger proteins do not form an intermediate with ubiquitin²¹¹. Instead, they promote the direct transfer of ubiquitin from E2 to the substrate, in part by immobilizing the carboxy-terminal glycine of the ubiquitin in an otherwise highly flexible E2~Ub, for attack by the acceptor lysine^212–215. More recently, a third class of E3 ligases was identified that contains a tripartite RING1-in-between-RING2 (RBR) domain arrangement and catalyses ubiquitination through a RING–HECT hybrid catalytic mechanism. RING1 in the RBR domain forms a typical cross-brace RING structure to interact with E2~Ub. The ubiquitin is then transferred to, and forms a thioester intermediate with, an active-site cysteine in RING2, followed by the transfer of ubiquitin to a substrate. c | Classification of RING-type E3 ligases. RING type E3 ligases can be either a single subunit, which contain an intrinsic RING finger domain, or a multi-subunit complex assembled on the cullin scaffold that does not contain a RING finger domain but instead binds in trans a small RING finger protein, either RING box protein 1 (ROC1) or ROC2. See BOX 1 for details of the cullin–RING E3 ubiquitin ligase assembly. CRBN, cereblon; VHL, von Hippel–Lindau tumour suppressor.

On the basis of their catalytic mechanism, the E3 family of ubiquitin ligases can be divided into three types: the homologous to E6AP carboxy terminus (HECT) type, the RING type and the RING1-in-between-RING2 (RBR) type (FIG. 2b). The human genome encodes 305 RING finger proteins according to the HUGO Gene Nomenclature Committee (HGNC) database, most of which are believed to act as E3 ligases²⁶. Although not containing a RING finger domain themselves, members of the evolutionarily conserved cullin family of proteins bind to a small RING protein, either RING box protein 1 (RBX1; also known as ROC1) or RBX2 (also known as ROC2), to assemble as many as 300 distinct, multi-subunit cullin-RING E3 ligases (CRLs; BOX 1). All PROTACs reported to date hijack a RING-type E3 ligase, including both single-subunit and multi-subunit CRLs (FIG. 2c; TABLE 1). The single-subunit E3 ligases typically ubiquitinate very few substrates. For example, the single-subunit E3 ligase MDM2 primarily regulates p53, and BIRC2 (also known as cIAP1) controls CRABP2 and DDIT3 (also known as CHOP)^27,28. By contrast, the multi-subunit CRLs often target multiple substrates. Twenty-six substates have been reported for CRL1^SKP2 (the superscript denotes the substrate receptor within the complex), 28 have been reported for CRL1^β-TRCP (REF.²⁹), 27 have been reported for CRL3^SPOP (REF³⁰), 14 have been reported for CRL4^DTL (REF³¹) and 9 have been reported for CRL5^SOCS1 (REF³²). This high degree of plasticity of CRLs in targeting multiple substrates is achieved by the combination of the intrinsic flexibility of each individual subunit within the CRL complexes (BOX 1). Such plasticity is a requisite for TPD, which requires the E3 ligase to ubiquitinate the POI or neosubstrate. The two most commonly used E3 ligases for TPD are members of the CRL subfamily, namely CRL2^VHL and CRL4^CRBN (FIG. 2c).

Box 1 |. Assembly and plasticity of cullin-RING E3 ligases.

Cullins adopt an overall elongated crescent-shaped structure consisting of a stalk-like amino-terminal domain (NTD) that comprises three cullin repeats and a globular carboxy-terminal domain (CTD) that binds RING box protein 1 (ROC1) or ROC2 (REFS^216,217). The NTD in different cullins binds a different adaptor protein: CUL1 binds with SKP1 (REFS^218,219), CUL2 (REFS^220,221) and CUL5 (REF.²²²) bind with the elongin B–elongin C complex, and CUL4A and CUL4B bind with DDB1 (REFS^217,223). Each adaptor protein interacts with a different protein motif: SKP1 interacts with F-box^218,219, elongin B–elongin C interacts with SOCS box^220,224,225 and DDB1 interacts with WD40 repeat^226–229 (see the figure). Distinct from other cullins, CUL3 integrates the functions of an adaptor and a substrate receptor into one protein by using its NTD to bind a BTB domain^230–233. Human cells express an estimated 69 proteins containing an F-box²³⁴, 32 containing a SOCS box²³⁵, 188 containing a BTB domain²³⁶ and 18 containing a DDB1- and CUL4-associated factor (DCAF) (HUGO Gene Nomenclature Committee (HGNC) database)²³⁷. These proteins also contain additional domains for interacting with their substrate proteins and function as the substrate receptors, suggesting the potential assembly of more than 200 distinct cullin–RING E3 ligase (CRL) complexes²³⁸. The intrinsic flexibility of the cullin scaffold, adaptor, substrate receptor and ROC1 collectively contributes to the high degree of plasticity of CRL complexes. In CUL1, cullin repeat 3 can be twisted by 18° with respect to repeats 1 and 2 (REFS^216,239). A similarly flexible rotation within the CTD was also observed for CUL5 (REF,²⁴⁰). The adaptor proteins can be flexible as seen in DDB1, which contains three seven-bladed β-propellers, BPA, BPB and BPC. BPB binds CUL4 and is loosely attached to BPC, allowing it to rotate up to 150° before and after binding to its substrate receptor cereblon (CRBN) or SV5-V (see the figure)^{227,241,242,254}. ROC1 contains a flexible linker between the β-strand that binds the cullin CTD and the RING finger domain²¹⁶. The flexible linker allows the RING domain to reorient with respect to the CTD and form multiple potential catalytic geometries with E2 enzyme^243–245.

PDB, Protein Data Bank; ub, ubiquitin.

Table 1 |.

E3 ligands used for targeted protein degradation

E3 ligand	Binding E3 subunit	E3 ligase complexa	Neosubstrates of the E3 ligand	Toxicity of E3 ligand	Mechanism of action	Functional essentiality of the E3 subunit		Refs
						In cellsb	In mice
Nutlin 3, idasanutlin	MDM2	MDM2	Not reported	High	PROTAC	5.2%	Embryonic lethal	171–173
VHL ligands	VHL	CRL2VHL	Not reported	NA	PROTAC	0.5%	Embryonic lethal	12,33,89,128,174–176
IMiDs	CRBN	CRL4CRBN	ARID2, CK1α, GSPT1, IKZF1, IKZF3, p63, ZBTB16, SALL4, ZFP91, ZNF276, ZNF653, ZNF692, ZNF827	High	Molecular glue and PROTAC	0%	Viable	7,9,75,77,78,177–181
Bestatin, LCL161, MV-1	BIRC2	BIRC2	Not reported	Low	PROTAC	2.3%	Viable	182–189
Sulfonamides	DCAF15	CRL4DCAF1	RBM39, RBM23	Medium	Molecular glue and PROTAC	NA	NA	79–82,129,190–192
CCW16	RNF4	RNF4	Not reported	NA	PROTAC	2.3%	Embryonic lethal	132
KB02	DCAF16	CRL4DCAF1	Not reported	NA	PROTAC	NA	NA	131
Nimbolide	RNF114	RNF14	Not reported	None	PROTAC	NA	Viable	193
CR8, HQ461, dCeMM2, dCeMM3, dCeMM4	DDB1	CRL4CDK12	Cyclin K	NA	Molecular glue	67.2%	Embryonic lethal	83–85
Manumycin polyketides	UBR7	UBR7	p53	NA	Molecular glue	0%	NA	86

CRBN, cereblon; CRL, cullin–RING E3 ligase; DCAF, DDB1- and CUL4-associated factor; IMiD, immunomodulatory imide drug; NA, not available; PROTAC, proteolysis-targeting chimera; VHL, von Hippel–Lindau tumour suppressor.

^aThe superscript denotes the substrate receptor within the complex.

^bCalculated with use of the DRIVE (deep RNAi interrogation of viability effects in cancer) data portal³. A gene with a redundant small inferring RNA activity value of −3 or less is considered to be essential in that cell line. The total number of cell lines analysed is between 384 and 387, and the percentage represents the proportion of cell lines in which the gene is essential.

Heterobifunctional small-molecule degraders

To date, ten E3 ligases have been exploited for TPD in the form of either molecular glues or PROTACs (TABLE 1). Most reported PROTACs (Supplementary Table 1) were developed with use of two types of E3 small-molecule ligands, namely IMiDs that bind CRBN^7–9 and von Hippel–Lindau tumour suppressor (VHL) ligands that bind VHL^12,33,34 within the CRL complexes (TABLE 1; Supplementary Figs. 1,2). It is not yet clear why other E3 ligands have not been more widely used, and many E3 ligases are yet to be explored. However, the plasticity of the CRL complexes may greatly facilitate ternary complex formation and subsequent ubiquitination of the target protein. Key elements of the PROTAC and other TPD approaches are discussed in the following sections.

General principles of PROTAC design

The basic chemical moieties of PROTACs include (1) an E3 ligand that binds and recruits an E3 ligase, (2) a ligand that binds the POI and (3) a linker that connects the E3 and POI ligands (FIG. 3a). PROTACs are catalytic in nature, in contrast to traditional small-molecule inhibitors, which are non-catalytic and rely on occupancy-driven pharmacology^5,12. This important feature of PROTACs minimizes the need for a long residence time and continuous drug exposure. Thus, PROTACs have the potential to deliver robust therapeutic effects at low doses and in infrequent dosing regimens with fewer side effects³⁵.

Fig. 3 |. PROTAC degraders.

a | General schematic of proteolysis-targeting chimera (PROTAC) degraders. A protein of interest (POI) can be degraded by the ubiquitin-proteasome system mediated by a PROTAC, which consists of a POI ligand connected to an E3 ligand via a linker. PROTACs are catalytic. The tilde denotes a high-energy thioester bond. b | PROTACs induce new protein–protein interactions between the POI and the E3 ligase. The image shows an overlay of two SMARCA2 bromodomain–PROTAC–von Hippel–Lindau tumour suppressor (VHL)–elongin C–elongin B complex crystal structures (RCSB Protein Data Bank IDs 6HAX⁴⁸ and 6HAY⁴⁹) in a ribbon representation. PROTAC 1 and PROTAC 2 share the same VHL ligand and POI ligand but have different linkers, which are specified in the box. The colours within the chemical structure (PROTAC 1 and PROTAC 2) correspond to the colours of the compounds in the crystal (ribbon) structures, with the VHL ligand in blue, the linker in green and the SMARCA2 bromodomain ligand in magenta. c | Chemical structures of several key heterobifunctional degraders highlighted in this Review. ARD-69, SJF-0628 and DT2216 are VHL-based PROTAC degraders, SS47 and SD-36 are cereblon (CRBN)-based PROTAC degraders and MS1943 is a hydrophobic tag-based degrader. AR, androgen receptor; EZH2, enhancer of Zeste homologue 2; STAT3, signal transducer and activator of transcription 3; Ub, ubiquitin.

A unique feature of PROTACs as therapeutic modalities is that they rely on the expression of the corresponding E3 ligase in target cells. This raises the possibility for control of PROTAC activity in selected tissues, organs, tumour types or subcellular compartments. Expression levels and subcellular localization of E3 ligases in targeted cellular systems are therefore important considerations in PROTAC design (BOX 2).

Box 2 |. Selective expression of E3 ligase genes for targeted protein degradation.

Of ten E3 ligases that have been exploited for TPD, some localize predominantly in one specific subcellular compartment, such as MDM2 and UBR7 in the nucleus, and DDB1- and CUL4-associated factor 15 (DCAF15) in mitochondria, while others, such as cereblon (CRBN), von Hippel–Lindau tumour suppressor (VHL) and BIRC2, localize across the cell (UniProt Knowledgebase). DCAF16 is a nuclear protein and forms the substrate receptor subunit of cullin 4-RING ligase (CRL4). A proteolysis-targeting chimera (PROTAC) based on a DCAF16 covalent ligand was found to promote nuclear-restricted degradation of the protein of interest¹³¹, illustrating the potential for subcellular-specific degradation of a protein of interest in the context of selective localization of an E3 ligase. The expression of the ten E3 ligases currently used in the PROTAC field differs significantly in different tumours, based on data from the National Cancer Institute’s Genomic Data Commons²⁴⁶ (see the figure, top). One interesting possibility is to exploit oncogenic E3 ligases that are expressed at a very low level in normal tissues or only in a few normal cells but are amplified or overexpressed in tumours. For example, the MDM2 gene is frequently amplified in numerous cancers, resulting in p53 inactivation and contributing to tumorigenesis²⁴⁷. Currently available MDM2 antagonists are associated with high toxicity²⁴⁸; if the toxicity issue can be resolved, MDM2-recruiting PROTACs could be particularly efficacious in tumours with elevated MDM2 expression such as sarcoma. F-box proteins SKP2 and β-TRCP²⁹ are frequently overexpressed in many cancer cells, and small molecules binding to both proteins have been reported^249,250. SPOP, a BTB domain-containing protein and substrate receptor of CRL3 that is overexpressed in clear cell renal carcinoma³⁰, contains a well-defined substrate-binding groove²⁵¹, and small molecules binding to this groove have also been identified²⁵². The expression of the E3 ligases used in PROTACs also varies significantly in different tissues and organs, according to data from the Expression Atlas Database²⁵³ (see the figure, bottom).

BLCA, bladder urothelial carcinoma; BRCA, breast invasive carcinoma; COAD, colon adenocarcinoma; FPKM, fragments per kilobase of transcript per million mapped reads; HNSC, head and neck squamous cell carcinoma; iBAQ, intensity Based Absolute Quantification; KICH, kidney chromophobe; KIRC, kidney renal clear cell carcinoma; KIRP, kidney renal papillary cell carcinoma; LIHC, liver hepatocellular carcinoma; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; PRAD, prostate adenocarcinoma; READ, rectum adenocarcinoma; STAD, stomach adenocarcinoma; THCA, thyroid carcinoma; UCEC, uterine corpus endometrial carcinoma.

To design an effective PROTAC, it is critical to select an E3 ligand that has sufficient binding affinity for the desired E3 ligase and, importantly, recruits the functionally active E3 ligase complex without compromising its ubiquitination activity. The availability of such ligands is limited (TABLE 1). The POI ligand selection is also crucial for successful PROTAC design. Usually, a solvent-exposed region of the POI ligand is selected as a linker attachment point to minimize interference of the linker and E3 ligand with PROTAC–POI binding. Additionally, the linker type and length are critical, and it is challenging to predict computationally which linkers are optimal in the absence of ternary complex structures. In general, exploration of multiple E3 ligands, POI ligands, linker attachments, and linker types and lengths is needed to generate effective PROTACs.

Ternary complex assembly

For a PROTAC to effectively degrade the POI, a sufficiently stable POI–PROTAC–E3 ternary complex generally needs to be formed³⁶, although exceptions have been reported^37,38. Several technologies have been developed to study the ternary complex formation, intracellular target engagement and E3 ligase recruitment, which have been reviewed elsewhere^39,40. The linker type and length are critical elements for the formation of sufficiently stable ternary complexes. An optimal linker is one that contributes to stabilizing the ternary complex by enhancing protein-protein interactions between the POI and the E3 ligase. Multiple linker types can adequately facilitate ternary complex formation. PROTACs with an optimal linker can induce cooperative binding. For example, it was reported that for the VHL-recruiting BRD4-targeting PROTAC known as MZ1, there is positive binding cooperativity in the BRD4–MZ1–VHL ternary complex compared with binary complexes (BRD4–MZ1 and VHL–MZ1), leading to effective degradation of BRD4 in vitro⁴¹.

To date, several POI–PROTAC–VHL ternary complex^41–44 (RCSB Protein Data Bank (PDB) IDs 6SIS⁴⁵, 5T35 (REF⁴⁶), 6HR2 (REF⁴⁷), 6HAX⁴⁸, 6HAY⁴⁹ and 6ZHC⁵⁰), POI–PROTAC–CRBN ternary complex⁵¹ (PDB IDs 6BN7 (REF.⁵²), 6BN8 (REF.⁵³), 6BN9 (REF.⁵⁴), 6BNB⁵⁵ and 6BOY⁵⁶) and POI–PROTAC–BIRC2 ternary complex³⁸ (PDB IDs 6W7O⁵⁷ and 6W8I⁵⁸) crystal structures have been reported. Within these structures, new interactions between the POI and the E3 ligase have been identified (FIG. 3b). These PROTAC-induced protein–protein interactions provide a structural basis for the observed cooperative binding. The protein–protein interactions revealed by these ternary complex structures also enable structure-based design to generate more effective PROTACs, an opportunity barely explored in the PROTAC field to date.

An important advantage of PROTACs over traditional small-molecule inhibitors is that PROTACs can be selective for POI isoforms or subtypes even though the POI ligands used in the PROTACs are not. For example, PROTACs have been developed that are selective for HER2 (also known as ERBB2) over EGFR and vice versa⁵⁹, for CDK4 over CDK6 and vice versa⁶⁰, for signal transducer and activator of transcription 3 (STAT3) over other STAT isoforms⁶¹ and for BCL-XL over BCL-2 (REF.⁶²) even though the POI ligands used are not isoform selective. The ability of these PROTACs to induce isoform-selective degradation is likely due to a preference for ternary complex formation with one particular POI isoform over others. Additionally, PROTACs incorporating pan-kinase inhibitors such as foretinib as POI ligands demonstrated selective degradation of only a subset of the kinases that the POI ligands can bind⁶³. This selectivity could be due to the ability of the PROTACs to induce ternary complex formation for only a subset of the kinases, leading to polyubiquitination and degradation of these kinases only. Although ternary complex formation is necessary, this alone is not sufficient for PROTACs to induce POI degradation. For example, a VHL-recruiting PROTAC based on foretinib can form ternary complexes with two isoforms of the mitogen-activated protein kinase p38, but promotes degradation of only one of them⁶⁴. Other factors such as the half-life of the PROTAC, the synthesis rate of the POI and the abundance of the POI and the E3 ligase in cells are important for effective degradation of the POI.

Other approaches

Hydrophobic tag-mediated degradation.

While PROTACs have generated much of the excitement surrounding degrader-mediated therapies, other TPD approaches using heterobifunctional small molecules have been developed. One of these approaches is hydrophobic tagging, which links a POI ligand with a highly lipophilic moiety that activates the unfolded protein response, thereby inducing degradation of the POI⁶⁵. Fulvestrant is a selective oestrogen receptor (ER) degrader that first realized the concept of hydrophobic tagging⁶⁶. Fulvestrant stabilizes a conformation of ER that exposes the hydrophobic tag region, leading to its subsequent proteasomal degradation^66,67. This finding generated interest in creating compounds with a similar mechanism of action, such as selective androgen receptor (AR) degraders⁶⁸. While the mechanism of action has not been fully elucidated, it is postulated that hydrophobic tagging co-opts and triggers the function of chaperones. Typically, chaperone proteins assist in the refolding of the proteins that have been misfolded into a conformation that exposes hydrophobic, non-polar regions of the proteins. When the misfolded proteins cannot be ameliorated by chaperone proteins, they are marked for degradation by the proteasome⁶⁹.

Hydrophobic tag-mediated degradation starts in the endoplasmic reticulum and differs from PROTAC-mediated degradation⁷⁰. The interplay between hydrophobic tag-mediated degradation and the 26S proteasome was explored following the discovery that tert-butyl carbamate Boc₃-protected arginine (B₃A) acts as a hydrophobic tag to elicit degradation of its POI (a tagged fusion protein dihydrofolate reductase) in a B₃A-dependent, non-covalent manner (Boc3, three tert-butoxycarbonyl protecting groups)⁷¹. Additionally, haemagglutinin-tagged glutathione S-transferase A1 (GSTA1), a potential target for sensitizing chemoresistant cells to chemotherapy⁷², was degraded with an ethacrynic acid–B₃A compound in HEK293T cells⁷¹. Another hydrophobic tag, an adamantyl group, has also been used to induce TPD; the adamantyl-tagged compound HyT13 was shown to degrade a HaloTag-fused HRAS-G12V protein (a HaloTag is a self-labeling fusion-protein tag that can bind a ligand with a halogenated alkyl group), resulting in marked tumour reductions in a transformed mouse fibroblast NIH-3T3 HaloTag–HRAS–G12V mouse model⁷³. Another adamantyl-based degrader, derived from a covalent inhibitor of the pseudokinase HER3 (also known as ERBB3), was capable of degrading HER3 in vitro⁷⁴. In both cases, covalent POI ligands were used to generate the heterobifunctional degraders.

Molecular glues.

Molecular glues are not heterobifunctional molecules. Instead, they are simple small molecules that simultaneously bind an E3 ligase and a neosubstrate, leading to degradation of the neosubstrate. Of the ten E3 ligases that have been exploited for TPD, molecular glues have been identified for four of them (TABLE 1). IMiDs are the most well-studied molecular glues; they bind CRBN and create a new binding interface for as many as 13 reported neosubstrates^17,75. A glycine-containing β-hairpin motif in these otherwise diverse neosubstrates appears to be critical for their binding with CRBN^75–78. Sulfonamides such as indisulam and E7820 stabilize binding of the E3 ligase DDB1- and CUL4-associated factor 15 (DCAF15) with its neosubstrate, mRNA splicing factor RBM39, through non-polar interactions at a specific helical site found in RBM39 (REFS^79–82). Most recently, several CDK12 inhibitors (including CR8, HQ461, dCeMM2, dCeMM3 and dCeMM4) were identified as molecular glues of DDB1 to promote degradation of cyclin K^83–85. CR8 occupies the ATP-binding pocket of CDK12 and interacts with three residues in DDB1 through its hydrophobic phenylpyridine ring⁸³, effectively converting CDK12 into a substrate receptor of DDB1 and recruiting the CDK12 binding partner cyclin K as a neosubstrate for polyubiquitination and degradation⁸². In contrast to degradation, members of the manumycin family of polyketide metabolites act as molecular glues by covalently linking to a cysteine residue in the RING E3 ligase UBR7 and binding with the neosubstrate p53, resulting in non-proteolytic ubiquitination and functional activation of the p53 tumour suppressor in MDA-MB-231 breast cancer cells⁸⁶. A significant advantage of molecular glues is their low molecular weight, which more likely results in favourable pharmacokinetic properties such as oral bioavailability⁸⁷. However, a key issue is the lack of a rational design approach for discovering novel molecular glues. Nevertheless, molecular glues provide an attractive approach for advancing the TPD field.

First-in-class TPD cancer therapies

A large variety of heterobifunctional small-molecule degraders have been developed for many protein targets since 2015. While we do not discuss all of them here, we summarize those reported to date in the context of cancer in Supplementary Table 1. A subset of these degraders, which have established in vivo efficacy or sufficient in vivo pharmacokinetic properties, are summarized in TABLE 2. In the following section, we highlight key heterobifunctional degraders that are first-in-class and in or soon to be in clinical development or that are efficacious in vivo.

Table 2 |.

Cancer-related heterobifunctional degraders with in vivo efficacy or pharmacokinetics (PK) data

Compound	Disease	Target (POI)	MOA	Key bioassay data	Refs
MS4078	NSCLC, DLBCL	ALK	CRBN	In vivo PK and cellular efficacy	194
TD-004	NSCLC, DLBCL	ALK	VHL	In vivo efficacy	195
ARV-110	CRPC	AR	CRBN	Phase I/II clinical trial	14,93,94
ARD-69	CRPC	AR	VHL	In vivo efficacy	89
ARD-61	CRPC	AR	VHL	In vivo efficacy	90,91
DT2216	T-ALL, TCL	BCL-XL	VHL	In vivo efficacy	62
SIAIS178	CML	BCR–ABL	VHL	In vivo efficacy	196
BETd-246	TNBC	BET family	CRBN	In vivo efficacy	197
P4B	Melanoma	BRAF-V600E	CRBN	In vivo PD	122
SJF-0628	Melanoma	BRAF-V600E	VHL	In vivo efficacy	123
DP1	DLBCL	BRD4	DCAF15	In vivo efficacy	129
RC-1	AML	BTK	CRBN	In vivo efficacy	144
L18I	DLBCL	BTK	CRBN, VHL	In vivo efficacy	198
DD-03-171	MCL	BTK	CRBN	In vivo efficacy	199
CP5V	BC	CDC20	VHL	In vivo efficacy	200
MS140	MCL, CRC	CDK4, CDK6	CRBN	In vivo efficacy	201
MS39	NSCLC	EGFR	VHL	In vivo PK and cellular efficacy	202
Fulvestrant	BC	ER	Hydrophobic tag	FDA approved	66
ARV-471	BC	ER	CRBN	Phase I/II clinical trial	13,97,98
MS1943	TNBC	EZH2	Hydrophobic tag	In vivo efficacy	65
PROTAC FLT3	AML	FLT3	VHL	In vivo efficacy	203
SS47	MM	HPK1	CRBN	In vivo efficacy	127
MD-224	ALL	MDM2	CRBN	In vivo efficacy	204
MS432	MM, CRC	MEK1, MEK2	VHL	In vivo PK and cellular efficacy	205
MS934	MM, CRC	MEK1, MEK2	VHL	In vivo PK and cellular efficacy	206
MS4322	ER+ BC, NSCLC	PRMT5	VHL	In vivo PK and cellular efficacy	207
WL-40	MM	Rpn13 homologue	CRBN	In vivo efficacy	208
SD-36	AML, ALL	STAT3	CRBN	In vivo efficacy	61
CG428	CC	TRK	CRBN	In vivo PK and cellular efficacy	209

Entries are given in alphabetical order of the target. A comprehensive list of all cancer-related heterobifunctional degraders can be found in Supplementary Table 1. ALL, acute lymphoblastic leukaemia; AML, acute myeloid leukaemia; AR, androgen receptor; BC, breast cancer; BET, bromodomain and extra terminal; CC, colon cancer; CML, chronic myeloid leukaemia; CRBN, cereblon; CRC, colorectal cancer; CRPC, castration-resistant prostrate cancer; DCAF15, DDB1- and CUL4-associated factor 15; DLBCL, diffuse large B cell lymphoma; ER, oestrogen receptor; EZH2, enhancer of Zeste homologue 2; MCL, mantle cell lymphoma; MM, multiple myeloma; MOA, mechanism of action; NSCLC, non-small-cell lung cancer; PD, pharmacodynamics; POI, protein of interest; PROTAC, proteolysis-targeting chimera; Rpn13 homologue, 26S proteasome ubiguitin receptor (also known as ADRM1); STAT3, signal transducer and activator of transcription 3; T-ALL, T cell acute lymphoblastic leukaemia; TCL, T cell lymphoma; TNBC, triple-negative beast cancer; TRK, tropomyosin receptor family kinase; VHL, von Hippel–Lindau tumour suppressor.

AR degraders

The first report of a heterobifunctional degrader as a potential cancer therapeutic was the AR-targeting Protac-3 (REF.⁸⁸). Protac-3 was constructed by linking an IκBα (also known as NFKBIA) phosphopeptide that binds to β-TRCP (also known as BRTC), the substrate receptor of the CRL1^β-TRCP complex, with an AR ligand, dihydrotestosterone; the compound was shown to rapidly reduce the GFP signal in cells expressing a GFP–AR fusion protein in vitro⁸⁸. More recently, potent AR PROTACs such as ARD-69 (FIG. 3c; TABLE 2) have been developed. ARD-69, a VHL-recruiting PROTAC, achieved complete degradation of AR in AR⁺ prostate cancer cell lines and effectively reduced the levels of both AR and prostate-specific antigen for 48 hours in xenograft models⁸⁹. ARD-61 (TABLE 2), another VHL-recruiting PROTAC and a close analogue of ARD-69, was subsequently developed⁹⁰. ARD-61 is the most potent AR PROTAC degrader to date, and was efficacious at inhibiting tumour growth in vivo in both castration-resistant prostate cancer and AR⁺ breast cancer xenograft models^90,91. These preclinical findings suggest that AR degradation may provide a potential therapeutic strategy for overcoming the AR amplification or overexpression seen in castration-resistant prostate cancer that small-molecule antagonists such as enzalutamide cannot overcome.

Importantly, the AR-targeting PROTAC ARV-110 (TABLE 2), which is unrelated to ARD-69 and ARD-61, has entered clinical development and has shown promising preliminary clinical results^92–94. This orally bioavailable compound showed AR degradation in a variety of AR⁺ tumour cells with various point mutations in preclinical studies, and acceptable safety profiles in a phase I/II trial^92,93. The phase I/II dose escalation clinical trial (NCT03888612) included 18 patients, and only one patient experienced adverse events (grade 1) with treatment. Additionally, ARV-110 dosing at 420 mg orally showed evidence of antitumour activity in patients with T878A or H875Y mutations in AR. Notably, one patient demonstrated a confirmed partial response by RECIST (Response Evaluation Criteria in Solid Tumors), with a 97% decline in prostate-specific antigen level and 80% reduction in tumour size⁹⁴. Although preliminary, the results from this phase I/II trial suggest that ARV-110 has an acceptable safety profile and shows promise as a treatment for enzalutamide-resistant prostate cancer. This advancement into clinical development marks a significant milestone in the field of PROTAC therapeutics, and additional results from this ongoing clinical trial are keenly awaited.

ER degraders

ER acts as a master regulator of gene expression and is involved in the pathogenesis of breast cancer⁹⁵. The discovery that fulvestrant acts as a hydrophobic tag to mediate degradation of ER (see the section Hydrophobic tag-mediated degradation) has prompted further research into therapeutic ER degradation. Recently, ERD-308 (Supplementary Table 1) was developed as a VHL-recruiting ER-targeting PROTAC through successive rounds of structure–activity relationship studies and optimizations⁹⁶, and it was found to completely degrade ER in ER⁺ breast cancer cells⁹⁶. Importantly, ARV-471 (TABLE 2), an orally bioavailable ER-targeting PROTAC with potent in vivo activity, is now in clinical development¹³. ARV-471 is only the second PROTAC to enter clinical development (AR-targeting ARV-110 being the first). On the basis of a preliminary report, ARV-471 displayed potent ER degradation in cells and significant reduction of tumour burden in xenograft models⁹⁷. A recent congress abstract describing a phase I/II clinical trial in 21 patients with treatment-refractory, advanced ER⁺ and HER2⁻ breast cancer reported no dose-limiting toxicities and no grade 3 or grade 4 treatment-related adverse events in patients who received ARV-471 up to a 360-mg dose^13,98. Plasma exposures of ARV-471 were dose proportional up to a once-daily oral dose of 360 mg, and the half-life was 28 hours. Notably, one patient demonstrated a confirmed partial response (51% reduction in lesion size) per RECIST⁹⁸. These preliminary clinical results suggest that ARV-471 has a tolerable safety profile and promising antitumour activity.

STAT3 degraders

A significant breakthrough in therapeutic targeting of STAT3 was recently achieved via development of SD-36 (FIG. 3c; TABLE 2), a potent and selective STAT3 PROTAC⁶¹. SD-36 incorporates an analogue of the CRBN ligand lenalidomide and a STAT3 inhibitor (SI-109)^61,99, and was found to potently and rapidly degrade STAT3 in leukaemia and lymphoma cell lines⁶¹. Interestingly, SD-36 was highly selective for STAT3 degradation over other STAT isoforms even though SI-109 itself is not isoform selective. SD-36 showed a robust and long-lasting STAT3 degradation response in vivo, and treatment with SD-36 resulted in complete tumour regression in several xenograft mouse models, which lasted for 2 weeks after treatment termination⁶¹. SD-36 was also well tolerated in immunocompetent mice, with no toxic effects even though it induced similarly sustained degradation of STAT3 (REF.⁶¹). Collectively, these results suggest that pharmacological degradation of STAT3 is a promising therapeutic strategy for treatment of STAT3-driven malignancies. These encouraging results also highlight the potential of using degraders to target transcription factors that where once deemed ‘undruggable’.

BCL-XL degraders

BCL-XL, a member of the BCL-2 protein family, protects cells from undergoing programmed cell death¹⁰⁰ and underlies the ability of many malignancies to escape normal cell cycle regulation¹⁰¹. Many solid and haematological tumours that display BCL-2-independent survival show increased BCL-XL expression, suggesting a hierarchy of anti-apoptotic proteins necessary for survival^102,103. BCL-XL, as a key regulator of cell death, is therefore a promising therapeutic target for BCL-2-independent cancers. Small-molecule inhibitors have been developed to target the BCL-2 class of proteins, such as ABT263 (BCL-2 and BCL-XL dual inhibitor) and ABT199 (BCL-2-selective inhibitor)^62,104,105. However, significant side effects, including severe thrombocytopenia, have limited the utility of these compounds^106,107. Recently, DT2216 (FIG. 3c; TABLE 2), generated by linking ABT263 with a VHL ligand, was developed as a potent BCL-XL PROTAC with much reduced thrombocytopenic side effects in vivo relative to ABT263 due to the very low expression of VHL in platelets⁶². DT2216 displayed potent BCL-XL degradation in tumour cells, while inducing little BCL-XL degradation in platelets. Interestingly, DT2216 did not induce BCL-2 degradation even through it binds BCL-XL and BCL-2 with similar affinities. Furthermore, DT2216 completely inhibited MOLT-4 T cell acute lymphoblastic leukaemia xenograft growth in vivo, with only mild reductions in platelet count and no change in body weight⁶², in contrast to ABT263 treatment, which resulted in significant tumour burden reductions but also severe thrombocytopenia⁶². This elegant study provides support for advancing BCL-XL PROTACs to the clinic. It also exploited the differential tissue expression of the E3 ligase VHL; as additional E3 ligands are being developed, it is expected that PROTACs will increasingly exploit this selective expression of E3 ligases.

EZH2 degraders

Enhancer of Zeste homologue 2 (EZH2) is the main catalytic subunit of Polycomb repressive complex 2 (PRC2), which catalyses histone H3 lysine 27 methylation, and elevated EZH2 activity is a feature of many cancers¹⁰⁸. As one of the most prominent epigenetic drug targets, EZH2 has been intensively targeted, and multiple EZH2 small-molecule inhibitors have entered clinical developoment¹⁰⁹. Among them, tazemetostat has been approved for clinical use for the treatment of epithelioid sarcoma and follicular lymphoma¹¹⁰. However, EZH2 also has functions independent of its catalytic activity in lung, prostate and breast cancers^111–114, which cannot be effectively targeted by traditional inhibitors. MS1943 (FIG. 3c; TABLE 2) is the first EZH2-selective degrader and uses a hydrophobic tagging approach similar to the mechanism of action of fulvestrant⁶⁶. MS1943, derived from an EZH2 non-covalent inhibitor linked to an adamantyl group, effectively degraded EZH2 and suppressed proliferation in multiple EZH2-dependent triple-negative breast cancer and other cancer cell lines, while EZH2 inhibitors, including tazemetostat, did not⁶⁵. MS1943 was orally bioavailable in mice and completely inhibited tumour growth, with no weight loss or other toxicities in xenograft models⁶⁵. Additionally, the development of a CRBN-recruiting EZH2 PROTAC and the development of two PROTACs that target Polycomb protein EED (Supplementary Table 1), a core component of PRC2, was reported recently^115–117. Overall, these results provide evidence that pharmacological degradation of PRC2 components may be a promising therapeutic strategy, and EZH2 degradation may offer advantages over EZH2 inhibition for treatment of EZH2-dependent tumours.

BRAF degraders

BRAF, a component of the RAS–RAF–MEK–ERK signalling pathway, is frequently mutated in a variety of cancers, making it a desirable therapeutic target¹¹⁸. The BRAF V600E mutation accounts for more than 80% of the activation mutations in BRAF. Vemurafenib, a BRAF small-molecule inhibitor that inhibits both wild-type (WT) BRAF and BRAF-V600E, is used in the clinic for treatment of advanced melanoma as it was shown to prolong progression-free survival in patients with advanced melanoma¹¹⁹. However, high rates of resistance to vemurafenib and other BRAF inhibitors have resulted in the development of new therapeutic strageies¹²⁰. The first BRAF-V600E-selective PROTACs were reported recently¹²¹. CRBN-recruiting degraders 12 and 23 (Supplementary Table 1), which are based on the BRAF inhibitors vemurafenib and BI882370, respectively, potently induced degradation of BRAF-V600E in melanoma cells¹²¹. Notably, the degradation was selective for BRAF-V600E over WT BRAF, even though both degraders bind WT BRAF and BRAF-V600E equally well¹²¹.

Plausible mechanisms for this BRAF mutant selectivity were subsequently explored in two subsequent studies by Posternak et al. and Alabi et al.^122,123 Another BRAF PROTAC, P4B (TABLE 2), was reported by Posternak et al.¹²². P4B, generated by linking the BRAF inhibitor BI882370 to pomalidomide, an IMiD-based CRBN ligand, displayed potent degradation of BRAF-V600E but not WT BRAF. Interestingly, P4B was capable of degrading other BRAF mutants besides BRAF-V600E¹²². This was posited to result from the fact that WT BRAF exists in an inactive state that resisted degradation due to blockade of the ternary complex formation, while BRAF mutants exist in an active form that more easily undergoes ternary complex formation, thereby facilitating degradation. Additionally, while P4B displayed poor in vivo exposure, it degraded BRAF-V600E in malignant melanoma xenograft models¹²².

More recently, Alabi et al. reported that the PROTAC SJF-0628 (FiG. 3c; TABLE 2) induced degradation of mutant BRAF while sparing WT BRAF, and displayed antitumour activity in an A375 melanoma xenograft model¹²³. SJF-0628 was generated by linking vemurafenib with a VHL ligand. It potently degraded mutated BRAF proteins and effectively inhibited growth in cells with mutant BRAF but not in WT BRAF cells. It also induced interactions of all three BRAF mutant classes, but not WT BRAF, with the CUL2 E3 ligase complex. Interestingly, WT BRAF degradation could be triggered by promoting it to an active state that more closely resembles mutated BRAF. For example, activation of WT BRAF by pretreatment with the MEK inhibitor trametinib made it susceptible to SJF-0628-mediated degradation. Notably, SJF-0628 induced marked degradation of BRAF-V600E and inhibited tumour growth in vivo in an A375 melanoma xenograft model, with no significant body weight loss¹²³.

In summary, the ability of PROTAC technology to selectively target mutant BRAF over WT BRAF highlights another advantage of PROTACs over traditional enzymatic inhibitors. Mechanistic understanding of this selectivity for mutant proteins will aid in future development of mutant-selective degraders for other notoriously challenging targets.

HPK1 degraders

Modulation of the proteins that control T cell function holds great promise as a potential strategy for enhancing the antitumour effect of T cell-mediated therapy; one such regulator is the serine/threonine STE20-related protein kinase HPK1 (also known as MAP4K1)¹²⁴. HPK1 is an intracellular negative regulator of T cells that phosphorylates lymphocyte cytosolic protein 2, which in turn destabilizes the T cell receptor complex, leading to inhibition of T cell proliferation and activation^125,126. Recently, the first HPK1-targeting PROTAC, SS47 (FIG. 3c; TABLE 2), was generated by linking the HPK1 inhibitor ZYF0033 to the CRBN ligand thalidomide¹²⁷. SS47 induced degradation of HPK1 in cells and in an ex vivo model. Treatment of bone marrow-derived dendritic cells with SS47 resulted in significantly increased proliferation of CD4⁺ and CD8⁺ T cells upon antigen presentation. In addition, treatment with SS47 in a 4T-1 breast cancer mouse model significantly reduced tumour growth and synergized with an anti-PD1 antibody¹²⁷. While these results are encouraging, it is worth noting that selectivity of SS47 was not assessed. In particular, it is unknown whether this CRBN-recruiting PROTAC degrades CRBN neosubstrates, including GSPT1.

Design and development considerations

Developing novel and functional E3 ligands

Identifying a functional E3 ligand that can be used for the development of degraders is challenging, mainly because only a limited number of binding sites in an E3 ligase are suitable for productive polyubiquitination. As a result, binding of a protein substrate to a random site in an E3 ligase is unlikely to lead to productive ubiquitination. Not only does the ‘right binding site’ need to accommodate interactions with a protein substrate but it also needs to have sufficient space to allow processive ubiquitin chain elongation. The following approaches should be considered when one is attempting to identify functional E3 ligands.

First, converting substrate peptides into drug-like small-molecule E3 ligands is an approach that was successfully used in the development of VHL ligands¹²⁸. A major advantage of this strategy is that increasing numbers of E3 ligase–substrate crystal structures are available to enable structure-based design. However, this approach is challenging and laborious, as protein–protein interactions often involve large, flat and hydrophobic binding interfaces, lacking deep binding pockets necessary for small molecules to achieve sufficient affinity.

Another approach involves designing E3 ligands from molecular glues. The major advantage of using molecular glues as starting points is that they occupy the ‘right binding site’ where the neosubstrate is recruited and ubiquitinated. The feasibility of developing PROTACs based on molecule glues is expected to vary on the basis of each molecular glue’s affinity for the E3 ligase in the absence of a neosubstrate. In addition to the IMiD class of molecular glues, a PROTAC based on a DCAF15 molecular glue was recently developed¹²⁹ (TABLE 1). Generally, molecular glues with high E3 ligase affinity are preferred for developing PROTACs.

A third approach involves identification of covalent ligands. A recently developed chemoproteomics approach has enabled discovery of cysteine-reactive covalent ligands for undruggable proteins¹³⁰, and several covalent E3 ligands have been identified and can be developed into PROTACs to degrade a POI^86,131,132. Many E3 ligases are postulated to contain at least one reactive cysteine residue that could be modified by a covalent ligand¹³¹. Identification of covalent E3 ligands that have sufficient selectivity and that can efficiently hijack the ubiquitin–proteasome system would be highly valuable.

Finally, there are lessons to be learned from viral hijacking. Many viruses have evolved a common mechanism to counter host defence by encoding proteins that hijack host E3 ligases, especially the CRLs, to degrade antiviral factors that are otherwise not degraded in uninfected cells¹³³. Human papillomavirus protein E6 hijacks human E6-associated protein (E6AP; also known as ubiquitin-protein ligase E3A), a member of the HECT E3 ligases, to degrade p53 (REF.¹³⁴). Human and woodchuck hepatitis virus protein X (HBx and WHx, respectively) and paramyxovirus SV5-V protein bind to DDB1 to promote degradation of different host proteins, using a shared α-helical motif, H-box, that was also found in multiple cellular DCAF proteins, and can bind a deep pocket in DDB1 (REF.¹³⁵). Primate lentiviruses produce small proteins that hijack three different host E3 ligases, all CRLs, to degrade multiple host antiviral restriction factors¹³⁶. These small viral proteins can be viewed as prototypical PROTACs as they contain two functional domains, one for binding the E3 ligase and the other for binding the neosubstrate antiviral factor. These viruses identify the ‘right site’ in the E3 ligases that can interact with and support the polyubiquitination of neosubstrate antiviral factors. Importantly, the same E3 ligase binding site can be hijacked by different viral proteins¹³³, suggesting that this binding site may be targeted by distinct E3 ligand scaffolds. Thus, targeting these viral binding sites to discover novel E3 ligands offers a unique approach that is worth consideration.

Pharmacokinetics and oral bioavailability

The catalytic nature of heterobifunctional degraders is a key advantage of these compounds. Because these degraders do not rely on occupancy-driven pharmacology, the need for a long residence time and continuous drug exposure is significantly reduced. However, the high molecular weight, high lipophilicity and more rotatable bonds of most degraders present significant challenges for achieving adequate cell permeation ability, aqueous solubility and oral bioavailability^137–139. Despite these obstacles, effective protein knockdown by heterobifunctional degraders has been achieved in many preclinical species. For example, RC32, an FKBP12 degrader, showed essentially complete degradation of FKBP12 across heart, liver, kidney, skin and adipose tissues in mice, Bama pigs and rhesus monkeys¹⁴⁰. Furthermore, orally bioavailable and in vivo efficacious heterobifunctional degraders such as ARV-110, ARV-471 and MS1943 have been generated^65,94,98.

An extensive analysis of 38 PROTACs was conducted to predict which would have sufficient oral bioavailability using physicochemical properties, including the calculated distribution coefficient (log D), the molecular weight, the number of hydrogen bond donors and hydrogen bond acceptors and the number of rotatable bonds¹⁴¹. MDM2-recruiting and BIRC2-recruiting PROTACs exhibit high molecular weights, lipophilicities and numbers of rotatable bonds, suggesting that it would be very challenging to achieve sufficient oral bioavailability for these types of degraders. VHL-recruiting PROTACs fared better than MDM2-recruiting and BIRC2-recruiting PROTACs in the analysis. However, they still have high molecular weights and numbers of rotatable bonds, suggesting a big hurdle for achieving good oral bioavailability. On the other hand, CRBN-recruiting PROTACs have the lowest molecular weights, moderate lipophilicities and reduced hydrogen bond donor/rotatable bond counts, and are predicted to be more likely to be orally bioavailable¹⁴¹. Analyses of a diverse set of PROTACs generated by AstraZeneca’s internal programmes revealed that acceptable oral bioavailability (more than 30%) can be regularly achieved for CRBN-recruiting PROTACs but not for VHL-recruiting PROTACs¹⁴². While there will be exceptions to these generalizations, it is important to pay keen attention to minimizing the molecular weight, lipophilicity, number of hydrogen bond donors and number of rotatable bonds, and to using rigid linkers when one is optimizing oral bioavailability of heterobifunctional degraders.

Consideration of drug metabolism properties of heterobifunctional degraders is also worth noting¹⁴³. For example, while degraders generally display acceptable metabolic profiles, issues surrounding potentially active metabolites exist. A recent study reported the observation of potentially active metabolites, including both POI and E3 ligand derivatives¹⁴². Therefore, careful consideration should be given when one is evaluating in vivo stability and toxicity of heterobifunctional degraders.

Covalent reversible and irreversible POI binders

Three major classes of POI binders have been used in the design of PROTACs: non-covalent, covalent reversible and covalent irreversible POI binders. The vast majority of reported PROTACs use non-covalent POI binders, however, PROTACs using covalent reversible or irreversible POI binders were reported recently^144,145. While PROTACs bearing covalent irreversible POI binders are not catalytic, those based on covalent reversible POI binders can maintain the catalytic nature¹⁴⁴. In an elegant study, three BTK degraders using a reversible covalent POI binder (RC-1), a reversible non-covalent POI binder (RNC-1) and an irreversible covalent POI binder (IRC-1) were designed and compared¹⁴⁴. RC-1 induced potent degradation of BTK, while RNC-1 was less potent, and not surprisingly, IRC-1 was least effective and showed no degradation even at high concentrations. RC-1 also induced greater inhibition of cell proliferation than RNC-1 in vitro. These results suggest that reversible covalent POI binders offer an advantage over reversible non-covalent POI binders in PROTAC design. Additional studies to further validate these findings would be valuable.

Limitations of current TPD technology

Limitations of current E3 ligands

In addition to the paucity of E3 ligands, several aspects of current E3 ligands likely limit their broad applicability for therapeutic development, such as toxicity and acquired resistance resulting from functional non-essentiality of the E3 gene. The dependency of degraders on E3 ligases provides a new avenue for tumour cells to develop resistance through impairing the function of E3 ligase components. Indeed, it was demonstrated that tumour cells acquired resistance to both VHL-recruiting and CRBN-recruiting PROTACs following prolonged treatment^146,147. Acquired resistance to the PROTACs was primarily caused by genomic alterations or significantly reduced expression of the CRBN gene or the CUL2 gene instead of point mutations targeting the residues involved in binding the POI and E3 ligase^146,147. Genomic deletion of CRBN was also found as the primary cause of resistance to IMiDs in myeloma cells¹⁴⁸. Consistently, genome-wide CRISPR–Cas9 knockout screens in the leukaemia cell line KBM7 treated with five degraders based on three different E3 ligases, CRL2^VHL, CRL4^CRBN and CRL4^DCAF15, showed that deletion of CRBN, VHL and DCAF15 conferred resistance to these degraders¹⁴⁹. These findings highlight a new and more likely route for tumour cells to develop resistance to degraders based on functionally non-essential E3 ligases. Of the eight E3 ligases that have been exploited for PROTAC development, data from deep RNAi interrogation of viability effects in cancer are available for five of them, all of which are either functionally non-essential or essential in only a small fraction of cancer cells³ (TABLE 1). Interestingly, there was no overlap in genes whose alteration conferred resistance to VHL-recruiting or CRBN-recruiting PROTACs after prolonged treatment in vitro^146,147. This observation suggests that alternating use of degraders recruiting different E3 ligases may provide a viable strategy to achieve durable responses.

Drug toxicity is a major factor limiting the usefulness of therapeutic agents. In the TPD field, the toxicity of IMiDs is an issue worthy of attention. Thalidomide was approved in the late 1950s but was banned after the discovery that it caused widespread (more than 10,000) severe birth defects¹⁵⁰. It was repurposed for treating multiple myeloma¹⁵¹. Toxic effects including hypersensitivity rash, neutropenia and cardiac dysrhythmia were observed for IMiDs¹⁵². Some of these toxic effects are believed to be caused by IMiD-mediated degradation of CRBN neosubstrates. MDM2 ligands such as nutlin 3 were originally discovered as MDM2 antagonists and p53 activators by inhibiting MDM2–p53 interactions¹⁵³. While preclinical studies and early phase clinical trials showed the efficacy of nutlin 3 for treating cancers, its dose-limiting on-target toxicities resulting from p53 activation in normal cells remain a significant challenge to advance nutlin 3 as a monotherapy agent^154,155 and will likely limit its use in TPD. Sulfonamides, which are DCAF15 ligands, induce degradation of RNA splicing factors such as RBM39, which is functionally essential in many cell types; in line with this, dose-limiting toxicities have been observed in clinical trials¹⁵⁶. However, the BIRC2 ligand bestatin¹⁵⁷ and the RNF114 ligand nimbolide¹⁵⁸ both appear to have low toxicity and relatively positive safety profiles in clinical use. Given the paucity of current E3 ligands and their limitations, the TPD field has an urgent need and plentiful opportunities to develop novel E3 ligands with low toxicity.

Degradation of E3 ligase neosubstrates by PROTACs

While no neosubstrates have been reported for VHL, numerous neosubstrates of the IMiD–CRBN complex have been reported (TABLE 1), and many of them contain a C2H2 zinc-finger degron motif^77,159. Among these, GSPT1, a translation termination factor, is probably the most concerning neosubstrate, which may be degraded by CRBN-recruiting PROTACs because of the critical role of GSPT1 in most cells and potential toxic effects resulting from its degradation. It was first reported that GSPT1 was unintentionally degraded by MI-389, a CRBN-recruiting PROTAC that uses the pan-kinase inhibitor sunitinib as the POI binder, while the intended target, KIT, was not degraded¹⁶⁰. However, degradation of GSPT1 by a CRBN-recruiting PROTAC may lead to degradation of the POI as a secondary effect due to the critical role of GSPT1 in translation^77,161. Degradation of other CRBN neosubstrates by CRBN-recruiting PROTACs has been reported³⁶. For example, CDK4 and CDK6 PROTACs were found to degrade the lymphoid transcription factors IKZF1 and IKZF3, two neosubstrates of CRBN-recruiting IMiDs, with notably higher efficacies in killing mantle cell lymphoma cells over CDK4 and CDK6 degradation alone⁶⁰. Therefore, it is critical to determine neosubstrate degradation profiles of CRBN-recruiting PROTACs to rule out ‘secondary’ pharmacology, which may contribute to the observed phenotypes. Furthermore, both RBM39 and cyclin K, the neosubstrates of DCAF15 and DDB1-CDK12, respectively, are functionally essential in many cell types⁸⁴; therefore, it will be important to determine neosubstrate degradation profiles of DCAF15-recruiting and DDB1–CDK12-recruiting PROTACs.

Recent advances and future directions

To gain temporal control and avoid side effects, optically controlled small-molecule degraders have recently been developed^162–166. For example, pomalidomide, and BET and ALK PROTACs were caged using a photolabile group, which can be activated (uncaged) with short ultraviolet light irradiation, leading to spatio-temporally controlled degradation of their target proteins¹⁶⁶.

Another advancement in the field has been the creation of the dTAG system, which allows degradation of protein targets in cells and mice, and serves as an excellent research tool¹⁶⁷. This technology uses a fusion protein (POI fused with FKBP12-F36V) paired with a PROTAC derived by linking an FKBP12-F36V binder to a CRBN ligand, enabling rapid and effective degradation of the POI fusion protein by the PROTAC¹⁶⁷. The dTAG system proved fundamental for validating that transmembrane solute carrier protein can be degraded using the PROTAC approach, leading to development of the first solute carrier protein PROTAC that effectively degraded solute carrier protein transporters¹⁶⁸.

Lastly, LYTAC (lysosome-targeting chimera) technology was recently developed to degrade extracellular and membrane-associated proteins¹⁶⁹. The LYTAC system uses a small-molecule or antibody binder of the POI that is linked to a glycopeptide ligand, which binds to the mannose 6-phosphate receptor, a cell-surface lysosome-shuttling receptor, resulting in lysosomal degradation of the POI. Several extracellular and membrane-associated proteins were successfully degraded with use of this technology¹⁶⁹. Autophagy-mediated degradation technologies, such as autophagy-targeting chimeras (AUTACs) and autophagosome-tethering compounds (ATTECs), and chemical genetic degradation systems such as HaloPROTAC have also been developed¹⁷⁰. Collectively, these latest developments have laid a solid foundation for further advancing the heterobifunctional molecule field.

Conclusions

Following the development of the first small-molecule CRBN-recruiting and VHL-recruiting PROTACs in 2015, the pace of discovery has accelerated exponentially. Numerous first-in-class heterobifunctional degraders with robust in vivo efficacy have been developed. Additional neosubstrates of CRBN and new molecular glues and E3 ligands have been identified. Remarkably, two PROTACs have entered clinical development as cancer therapies^13,14.

Tremendous opportunities exist for further advancing the TPD field. First, ‘undruggable’ targets such as transcription factors that lack an enzymatic active site can be targeted. Transforming low-affinity, functionally inactive binders into effective PROTACs is a significant opportunity. Second, enormous opportunities exist for the development of novel and functional small-molecule ligands for E3 ligases. Third, several isoform-selective PROTACs based on isoform non-selective POI ligands have been generated. The field should further exploit this unique dimension to achieve isoform selectivity for suitable targets. Fourth, advancements towards generating mutant-selective degraders for oncoproteins are a unique opportunity. Fifth, exploiting differential E3 ligase expression levels in tumour and normal cells to rationally develop PROTACs with reduced side effects and an improved therapeutic window is another significant opportunity. Sixth, hydrophobic tag-based degraders are currently underexplored, and opportunities exist in this area as in vivo efficacious, orally bioavailable degraders can be developed⁶⁵. Lastly, structure-based rational design of effective PROTACs is another aspiration for the field as more POI–PROTAC–E3 ternary complex structures are solved.

Overall, the TPD field has seen exciting advancements. While challenges exist, we expect that the breath-taking pace of discovery in recent years will continue to accelerate and more heterobifunctional degraders will enter clinical development, some of which will likely become effective cancer therapeutics for patients.

Glossary

26S proteasome

A 2.5-MDa proteolytic protein complex that controls protein homeostasis and specific cellular process in all eukaryotes.

Immunomodulatory imide drugs (IMiDs)

A class of small molecules that have been used in the clinic to modulate the immune system via binding the E3 ligase cereblon (CRBN). IMiDs recruit neosubstrates to CRBN for ubiquitination and subsequent degradation and have often been used as E3 ligands in proteolysis-targeting chimeras (PROTACs).

Molecular glues

Small molecules that act like adhesives to induce or stabilize protein–protein interactions between an E3 ligase and a neosubstrate, leading to degradation of the neosubstrate.

Neosubstrates

Substrates of an E3 ligase that are not recognized by the E3 ligase under physiological conditions but interact with the E3 ligase in the presence of a molecular glue.

HECT

The HECT domain, which is approximately 350 amino acids long and homologous to the E6AP carboxy terminus (HECT), contains an evolutionarily conserved cysteine residue that forms a thioester linkage with ubiquitin.

RBR

A tripartite domain of approximately 140 amino acid in length, consisting of three zinc-binding domains, RING1–IBR–RICN2. RING1-in-between-RiGN2 (RBR) ligases combine mechanistic features of RING-type and homologous to the E6AP carboxy terminus (HECT)-type ligases by using RING1 to recognize the E2 ~ ubiquitin complex (the tilde denotes a high-energy thioester bond) and RING2 to form the thioester intermediate with ubiquitin.

RING finger

First identified as a novel cysteine-rich sequence motif present in the ’really interesting new gene’ (RING 1). RING fingers promote the transfer of ubiquitin directly from E2 to the substrate by locking the E2 ~ ubiqution conjugate (the tilde denotes a high-energy thioester bond) in a closed conformation.

SKP1

S-phase kinase associated protein 1 (SKP1) binds CUL1 and functions as the adaptor protein for CRL1 complexes.

Elongin B–elongin C complex

A heterodimer that functions as the adaptor protein complex for both CRL2 and CRL5.

F-box

A domain, first identified in cyclin F, approximately 40 amino acids long that binds the adaptor protein SKP1 and functions as the substrate receptors for CRL1 complexes.

SOCS box

An approximately 40 amino acid region originally identified in members of suppressors of cytokine signalling proteins that consists of two separate sequences, one for binding the elongin B–elongin C heterodimer (BC box) and one for binding CUL5 (CUL5 box).

WD40 repeat

A domain defined at the primary sequence level by a Gly-His dipeptide and a Trp-Asp (WD) dipeptide separated by 20–30 residues that is commonly found in many proteins of diverse function and typically forms β-propeller structures.

BTB domain

Also known as the POZ domain, a conserved domain of 115–130 residues that consists of five α-helices and binds the amino-terminal domain of CUL3 and functions as the substrate receptor for cullin 3-RING ligase (CRL3) complexes.

Cooperative binding

The enhanced binding of a proteolysis-targeting chimera (PROTAC) to both the protein of interest and the E3 ligase compared with the binding of the PROTAC to the protein of interest or the E3 ligase alone.

Unfolded protein response

A cellular stress response that is activated by high levels of misfolded or unfolded proteins in the endoplasmic reticulum. The unfolded protein response aims to decrease the amount of unfolded proteins to maintain cellular function or induce apoptosis when this cannot be achieved.

DDB1- and CUL4-associated factor (DCAF)

A member of a family of proteins also known as DDB1-binding WD40 (DWD) proteins that bind DDB1 and function as substrate receptors for CRL4 complexes.

DDB1

Damaged DNA-binding protein complex subunit 1 (DDB1) binds CUL4 and functions as an adaptor protein for cullin 4-RING ligase (CRL4) complexes.

Degron motif

A specific molecular feature that is recognized by E3 ligases.