Unraveling the Role of Linker Design in Proteolysis Targeting Chimeras

Abstract

A current bottleneck in the development of proteolysis targeting chimeras (PROTACs) is the empirical nature of linker length structure-activity relationships (SARs). A multidisciplinary approach to alleviate the bottleneck is detailed here. Firstly, we examine four published synthetic approaches that have been developed to increase synthetic throughput. We then discuss advances in structural biology and computational chemistry that have led to successful rational PROTAC design efforts and give promise to de novo linker design in silico. Lastly, we present a model generated from a curated list of linker SARs studies normalized to reflect how linear linker length affects the observed degradation potency (DC₅₀).

Graphical Abstract

INTRODUCTION

Since the first proof of concept in 2001,¹ small molecule–induced targeted protein degradation has become an exciting modality for pharmacological intervention. The two main strategies used to induce the degradation are the use of molecular glues which are small molecules that alter the substrate recognition domain of an E3 ligase thereby allowing the recruitment of neosubstrates for proteolysis, and the use of chimeric small molecules that consist of an E3 ligase binding moiety linked to a motif that binds desired target protein.² The later will be the focus of this perspective. Coined proteolysis targeting chimeras (PROTACs, also known as SNIPERs, uSMITEs or degraders), bifunctional small molecule degraders have become known for their catalytic activity (potency), ability to induce isozyme selectivity through interprotein interaction, and their deviation from the ‘rule–of–five’.^3–6 The utility of six unique chimeric degrader molecules in humans are currently being investigated in clinical trials.⁷ Of note, the indications of metastatic castration–resistant prostate cancer (ARV–110)⁸ and metastatic breast cancer (ARV–471),⁹ through the targeted degradation of the androgen receptors and estrogen receptors, respectively, and the non-oncogenic target IRAK4 degrader (KT-474) which could demonstrate their use in chronic dosing.⁷ Their advance into the clinic along with their unique ability to degrade challenging protein targets highlights the importance of understanding the salient features that guide successful PROTAC discovery. In this mini–perspective, we explore the strategies involved in enhancing throughput, examine successful rational design efforts, and provide a synopsis of the molecular trends within the current linker designs.

The unique metabolic activity of PROTACs comes from their ability to appropriate RING ubiquitin ligases from the ubiquitin–proteasome system (UPS), which regulates protein homeostasis by tagging proteins with poly–ubiquitin chains, thereby marking them for proteolysis.¹⁰ The UPS signal cascade (Figure 1a) begins with the E1 activating enzyme, which adenylates the C–terminus of ubiquitin, followed by a transfer of the ubiquitin to an active site cysteine. The E1 activating enzyme then engages an E2 conjugating enzyme and transfers the ubiquitin to the E2 active site cysteine through a trans–thioesterification. The ubiquitin–loaded E2 conjugating enzyme then binds a multisubunit E3 ligase complex, which transfers the ubiquitin to a peripheral lysine residue on the substrate protein, labeling the substrate for degradation by the proteasome.¹¹ With each step of the cascade (Figure 1a), the UPS confers specificity for its substrate protein. Efforts have been made at therapeutic intervention at each step in the UPS cascade.^12–14

Figure 1.

Metabolic activity of a PROTAC. a) Representative comparison of UPS pathway for cullin–RING E3 ligases in both the presence and absence of PROTAC molecules. Step 1: ubiquitin activation by E1. Step 2: Engagement of ubiquitin loaded E1 with E2 conjugating enzyme. Step 3: transfer of ubiquitin to E2 via trans–thioesterification and subsequent formation of E3 ligase complex. Introduction of PROTAC molecules can redirect the E3 ligase from its cognate substrate (sub) to a non–cognate protein of interest (POI). b) Representative example of a non–cognate ternary complex formed by the PROTAC dBET23, bromodomain BRD4, and E3 ligase substrate binding domain CRBN (PDB 6BN7). Note: E3 cartoon represents a multi-component complex in which the PROTAC is able to bind the substrate adapter protein in order to redirect its metabolic activity.

PROTACs are chimeric small molecules that are able to engage the substrate recognition domain of an E3 ligase and a protein of interest (POI) simultaneously, thereby inducing non–cognate ubiquitination and subsequent degradation of the POI (Figure 1a).¹⁵ A number of E3 ligases have been targeted for PROTAC development,¹⁶ however here we focus on PROTACs that hijack the substrate adapter domains cereblon (CRL4^CRBN)¹⁷ and Von–Hippel Lindau (CRL2^VHL)¹⁸ RING E3 ligases. Interestingly, PROTACs can exhibit an alluring ‘plug–and–play’ architecture,¹⁹ where in some cases the same POI can be degraded by recruiting different E3 ligases. For example, the bromodomain (BRD4) degraders MZ1 and dBET–23 (Figure 1b) recruit Von–Hippel Lindau (CRL2^VHL) and cereblon (CRL4^CRBN) E3 ligases, respectively.^20,21 In other cases swapping E3 ligase recruitment can show exquisite target selectivity through tertiary interaction as demonstrated by the development of tyrosine kinase (ABL/BCR-ABL) degraders developed with promiscuous kinase inhibitor warheads.¹⁹

The potency and isozyme selectivity of PROTACs can be optimized through structure–activity relationships (SARs) within the linker (Figure 1b). Here, the length and chemical composition has been shown to influence, among others, a PROTAC’s structural rigidity, hydrophobicity, and solubility.²² To date, linker length SAR studies are largely empirical and have proven to be time and labor intensive. While great strides have been made towards rational PROTAC design through structural biological and computational studies, linker design still presents a significant synthetic burden.

APPROACHES TO ENHANCE THROUGHPUT

The current approaches to streamline linker variant SAR studies involve increasing synthetic throughput via use of orthogonally protected bifunctional linkers²³ solid phase synthesis,²⁴ copper–catalyzed click chemistry,²⁵ activated esters,²⁶ and Staudinger ligation chemistry²⁷ (Figure 2). PROTAC synthesis involves an asymmetric three–part diversification between the two binding motifs and a linker region (Figure 2b). By exploiting simplified purification procedures and parallel synthetic strategies, these approaches decrease material lead-time without altering the empirical nature of the study.

Figure 2.

Published synthetic methods for accessing PROTACs. a) Synthesis of a BTK degrader using thalidomide preloaded resin (TPR). Secondary amine shown highlighted in COLOR is used as a linker diversification handle. b) Representative structure of a PROTAC bearing amide linker chemistry. c) Synthesis of bromodomain degraders using copper–catalyzed click chemistry. d) Synthesis of dBET1 utilizing pentafluorophenyl activated esters. e) Synthesis of dBET1 via Staudinger ligation chemistry. Reagents: a, iodoacetic acid, DIC, DCM, rt; b, despropenoyl ibrutinib, DIPEA, DMSO, rt then TFA/DCM (1:1), rt, 49% overall for a and b; c, CuSO₄ (20 mol%), sodium ascorbate (20 mol%), THF/H₂O, rt, 67–90%; d, N–Boc–1,4–butanediamine, DIPEA, DMF, rt, 81%; e, TFA/DCM (1:5), rt, 99%; f, DIPEA, DMF, rt, 81%; g, 4–azido–1–butanamine, HATU, DABCO, DMF, rt, 50%; h, DMF, 40 deg., 54%. Note: steps g and h can be accomplished in a one–pot fashion.

Krajcovicova et. al.²⁴ was able to prepare a suite of PROTACs consisting of five different kinase inhibitors using a thalidomide preloaded resin (TPR) (Figure 2a). The aminomethyl polystyrene–divinylbenzene resin (PS–DVB) was acylated with 4–(4–formyl–3–methoxyphenoxy)–butanoic acid yielding a terminal aldehyde, which after amidation and subsequent reduction to a secondary amine, provided a synthetic handle for linker diversification. The resulting resin was then treated with iodoacetic acid (with and without additional polyethylene glycol (PEG) spacer), producing electrophilic TPR.

Subsequent reactions of this electrophilic TPR with a family of nucleophilic kinase inhibitors yielded a family of PROTACs. The fully elaborated PROTACs were then cleaved from the PS–DVB resin with trifluoroacetic acid (TFA) yielding the crude PROTACs in 24–85% yield and 78–98% purity and were then further purified by reverse–phase high pressure liquid chromatography (RP–HPLC) on a 200 mg scale. Though in this study, the authors prepared a PROTAC library using an electrophilic TPR and nucleophilic kinase inhibitors, they noted that this methodology is generalizable and other synthetic strategies to append POI–targeting inhibitors to the TPR can be accommodated, including use of electrophilic inhibitors with a nucleophilic TPR. If TPR resin is stable upon storage, this strategy will undoubtedly buoy efforts to produce ‘user–friendly’ tool kits for the diversification of CRBN–recruiting PROTACs.

The copper–catalyzed click chemistry platform demonstrated by Wurz et. al.²⁵ (Figure 2c) relied upon the preparation of a library of alkyne–terminal PEG–linked CRBN ligands and VHL ligands (not shown), as well as an azide–linked derivative of known bromodomain inhibitor JQ1. The two ligands were then united through triazole formation using copper–catalyzed click chemistry.²⁸ A panel of ten bromodomain targeting PROTACs (five CRBN recruiting, five VHL recruiting) were prepared on 100 mg scale in 55–90% yields, and purity was demonstrated via liquid chromatography-mass spectrometry (LCMS). Purification was accomplished using reverse–phase chromatography (ISCO Combiflash). This is the only published synthetic platform that directly synthesized both VHL recruiting PROTACs in addition to CRBN recruiting PROTACs.

An approach developed out by Papatzimas et. al.²⁶ utilized pentafluorophenyl (Pfp) esters as synthons to access previously published PROTAC dBET1¹⁷ (Figure 2d) in 81% yield on 40 mg scale and could be readily purified on normal–phase silica gel. This strategy entailed isolating both the E3 ligase ligand and bromodomain inhibitor JQ1 as Pfp–esters. In a three–step sequence, a Boc–protected amine linker was then reacted with the thalidomide–Pfp–ester to yield the first amide bond formation, the Boc–protecting group was removed with TFA, and finally the JQ1 Pfp–ester was added to yield dBET1¹⁷ (Figure 2d).

We recently developed²⁷ a synthetic method that leverages the chemoselectivity of the Staudinger ligation in order to assemble PROTACs in an asymmetric, one–pot fashion (Figure 2e). We also targeted dBET1¹⁷ (BRD4 degrader) and two analogs as a model system to demonstrate that the entire PROTAC assembly could be choreographed in a single reaction flask. However, the resulting reaction mixture was quite complex and would require HPLC purification. We then chose to isolate azido–terminal linked thalidomide intermediates and showed that all of the linker variants could be synthesized in parallel in 39–85% yields from a stock solution of JQ1 borane–protected phosphine thioester. This only required silica microcolumn purification to obtain highly pure compound on 10 mg scale.

Taken together, there are a number of strategies that have been demonstrated to achieve modular synthesis of PROTAC linker variant suites for IMiD–based (CRBN recruiting) systems, although none of them appear to have widespread application in the field as of yet. Although it is conceivable that most of the strategies presented could be translatable to VHL or other E3 ligase recruiting systems, only the copper–catalyzed click platform²⁵ specifically demonstrated this, however introduction of the triazole moiety to the linker may not yield desirable physiochemical properties due to its high topological total polar surface area (TPSA).²⁹ All of these strategies would greatly benefit from the availability of libraries of functionalized (alkyne, Boc–protected amine terminal, or azide terminal) linked E3 ligase ligands prepared and aliquoted for posterity. Indeed, many are commercially available. Key considerations for choosing a platform include stability upon long term storage of the linked E3 ligase ligand libraries, purification capabilities of the laboratory, and desired linker chemistry (linear vs. triazole containing). Collectively, these strategies increase the synthetic throughput of linker length SAR studies, and we see them as a first step (potentially coupled with computational techniques) in PROTAC development to determine POI•E3 ligase compatibility. Once an optimal linker length hit is detected further linker SAR could be implemented to optimize rigidity, solubility, cell permeability, and pharmacological (PK/PD) profile to produce a lead degrader compound. This is commonly done by replacing PEG and linear aliphatic linkers with piperine and piperazine based linkers which reduce the degrees of freedom in the degrader and provide more favorable pharmacological properties, such as in the development of mutant BRAF kinase degrader SJF-0628³⁰ and recently released structures of ARV-110 and ARV-471, two of the degraders currently in the clinic. However, we view the linker rigidifying SAR as a next step in lead degrader development after the optimal linker length has been determined either by biological evaluation or computational prediction.

RATIONAL PROTAC DESIGN

Alternatively, efforts have been made towards rational PROTAC design utilizing both X–ray crystallographic data and computational modeling.³¹ The first X-ray structure of a degrader (MZ1) in ternary complex was solved by Gadd et. al.³² and using these data they were able to rationally design a more selective bromodomain degrader (AT1). Farnaby and coworkers³³ were able to identify crucial stabilizing interactions between the PEG linker of PROTAC 1 (Figure 3a) and VHL from co–crystal structures of SMARCA2:PROTAC 1:VHL (PDB 6HAY). Based on these data, they introduced an additional T–shaped–stacking interaction and increased rigidity via insertion of a phenyl moiety in the linker region without sacrificing the key PEG interactions yielding ACBI1 (Figure 3a); an improved SMARCA2/4 degrader.

Figure 3:

Exemplary rational PROTAC development campaigns. a) Development of improved SMARCA2/4 degrader ACBI1 through identification of key interactions with the PEG linker. b) Design of isozyme selective bromodomain degrader ZXH-3–26 through limitation of the degrees of freedom via computationally-aided linker length reduction. c) Development of isozyme selective bromodomain degrader macroPROTAC-1 through limitation of the degrees of freedom via introduction of a macrocyclic linker.

Notably, Nowak et al.²¹ were able to develop a novel BRD4 selective degrader ZXH–3–26 based upon a known pan–bromodomain degrader, dBET6, (Figure 3b) by performing RosettaDock simulations that utilized the X–ray structures of a set of related, PROTAC-bound ternary complexes.³⁴ Stochastic sampling of low energy conformations using Rosetta³⁴ was able to recapitulate experimental structures. Structural predictions also suggested that minimization of PROTAC linker length would enhance degrader selectivity by reducing the number of favorable binding modes for the POI•PROTAC•ligase ternary complex.

Testa and coworkers³⁵ have also demonstrated the utility of computer-aided PROTAC design. They identified a potent and isoform selective (BRD4^BD2 selective) bromodomain degrader using molecular dynamics simulations of the complex of BRD4^BD2 and VHL bound with known bromodomain degrader MZ1 (PDB 5T35) (Figure 3c). Analysis of the simulations suggested that a macrocyclic linker could improve MZ1 effectiveness by reducing the degrees of freedom of the PROTAC, pre-organizing its POI and ligase moiety for ternary complex formation. Synthesis of a macrocyclic analog of MZ1 was realized using a ‘bespoke’ PEG-based linker. The resulting macrocyclic PROTAC demonstrated a lower binding affinity for both BRD4 and VHL in binary complex compared to MZ1, while maintaining a comparable degradation potency. Taken together, this suggests an increase in ternary complex formation efficiency.

More rigorous computational methods have been developed in efforts towards rationally designed de novo PROTAC development using both Molecular Operating Environment (MOE) and the open source Rosetta software suites. These methods have not only successfully reproduced, in silico, PROTAC binding modes identified from X-ray crystallography, but are also consistent with trends in potency and selectivity based upon prior biological evaluation. Herein, we describe select examples to illustrate the value of such efforts. Further reading on computer-aided PROTAC development may be found here.^36–41

Using the MOE software suite, Drummond et. al.^42,43 developed a series of protocols for generating and analyzing PROTAC ternary complexes (Figure 4). Of these protocols, the most effective sampled the conformational space of the degrader in the absence of both POI and ligase in order to identify degrader conformations that in a subsequent step would be subjected to docking simulations. These methods used co-crystal structures of the binding moieties of the PROTAC (no linker) bound to their respective targets. This team has also shown that computationally docked structures may also be used as input. Next global protein•protein docking simulations performed using MOE’s docking protocol in order to generate an ensemble of states and determine how the proteins might interact proximal to their ligated pockets.

Figure 4.

Recent advances in computational approaches to PROTAC development. MOE: General approach and highlights of the protocol used by Drummond et. al.³⁶ on the Molecular Operating Environment (MOE) platform. PRosettaC: General approach and highlights of the protocol used by Zaidman et al.⁴² on the Rosetta platform. Rosetta: General approach and highlights of the protocol used by Bai et al.⁴⁶ on the Rosetta platform.

Next, the ligands were then constrained to their bound conformation used in the global protein•protein docking operation, and a series of PROTAC conformers were generated using LowModeMD.⁴⁴ The resulting PROTAC conformers were then superimposed onto the docked POI•ligase structures, multiple energy minimizations were performed on the PROTAC, and the coordinates of the bound relaxed PROTACs were compared to the unlinked ligand-protein structures. Finally, a restrained minimization protocol was performed on the ternary complex as a whole to remove steric clashes between the proteins and the PROTAC. The method described was able to recapitulate degrader binding modes and potency trends for both VHL-recruiting and CRBN-recruiting degraders,^21,32,33 rationalize the degradation potency trends of macrocyclic PROTAC-1 (Figure 3c),³⁵ a series of CRBN-recruiting Bruton’s tyrosine kinase (BTK) degraders,⁴⁵ PROTACs bearing a more rigid linker design,⁴⁶ and predict with reasonable accuracy the degradation selectivity among kinases of a degrader bearing a pan-kinase inhibitor.⁴⁷

Alternatively, Zaidman et. al.⁴⁸ used the open source Rosetta software suite in order to develop PRosettaC (Figure 4). PRosettaC is similar to that of Drummond and coworkers^42,43 insofar as that the structures of the binary complexes of both ligands with their respective targets are requisite inputs. The PRosettaC protocol requires the definition of two anchor points corresponding to the binding epitopes on the degrader. The anchor points are then used to perform a rough sampling of the distances and ligand positions between the two anchor points that could accommodate a full length PROTAC, yielding an ensemble of degrader conformations that are binned based on the distance between the anchor atoms. These simulations provide information about the distance constraints for global protein•protein docking simulations using PatchDock.⁴⁹ This approach facilitates the rapid sampling of protein•protein interaction space. Restraint-free local protein•protein docking using RosettaDock⁵⁰ are subsequently performed to refine the structures of modeled POI•E3 ligase complexes.

Lastly, each of the solutions had their ligand positions fixed and were superposed with 100 full PROTAC conformations generated using RDkit, and the optimal conformation was chosen using Rosetta Packer.⁵¹ The resulting ternary complexes were then filtered via Rosetta energy score, and clustered with the assumption that the near native solutions would be sampled many times. This method was able to recapitulate experimentally determined ternary complexes of CRBN-recruiting and VHL-recruiting degraders,^21,32,33,35 as well as rationalize the degradation potencies of a series of CRBN-recruiting BTK degraders.⁴⁵

Bai et. al.⁵² described an alternative Rosetta-based protocol (Figure 4). The protocol was still reliant on ligand-bound binary X-ray crystal structures as initial inputs, however it used a global docking protocol initially developed for antibody-antigen docking⁵³ to generate the initial ensemble of diverse poses of the POI and ligase about a fixed ligand position. Poses were evaluated based upon the stabilizing effect of protein•protein interactions formed between the POI and the ligase with the most highly scored decile of poses chosen for further refinement. Independently, the OMEGA software was used to generate a series of low energy linker conformations that were assembled from various linkers examined attached to small chemical moieties representing the functional group used at the binding moiety attachment site.⁵⁴ Up to 1,000 conformers were generated for each linker and aligned with the ligand bound global docking solutions to determine if the linker adequately bridged the gap between proteins. The selected linkers were then merged with the bound ligands to create the full PROTAC and an energy minimization was performed to eliminate steric clash and bond angle distortions. The resulting ternary complexes were then filtered based on the median interaction energy of the initial protein•protein interaction and normalized by the number of initially docked models and linker conformations. The protocol was able to recapitulate degrader potency data for CRBN-recruiting BRD4 degraders,^21,46 as well as rationalize kinase selectivity between both CRBN-recruiting and VHL-recruiting bearing the same kinase inhibitor.³⁷

Both MOE and Rosetta were able to successfully recapitulate ternary complex binding modes elucidated from prior X-ray crystallographic studies, although these binding modes could represent thermodynamic artifacts from the process of crystallization.⁵² For this reason, we find the ability of the protocols to rationalize biological trends in potency and target selectivity is, perhaps, a more apt benchmark of success as it pertains to de novo PROTAC development.

Due to the non-cognate nature of the induced protein•protein interactions, a common theme between global docking strategies was the biasing towards hydrophobic interactions, which is consistent with the plastic protein•protein interface model²¹ where multiple binding modes may be populated with direction-dependent polar interactions playing a secondary role in binding. A hurdle for the field to overcome will be the reliance on input of X-ray crystallographic data of ligand-bound POI targets, however Drummond et. al.^42,43 have demonstrated moderate success with computationally docked structures, attributing the protocols shortcomings to the deviation of the position of the ligand anchor point atom in the docked structures from that of the X-ray crystal structure. Improvement in ligand docking protocols and strategies with reduced reliance on anchor points in global docking could overcome this issue.

Overall, these studies demonstrate that PROTAC development has benefitted from computer-aided optimization (Figure 3). Recent advances in computational protocols for small molecule design presage an era of rational de novo PROTAC design (Figure 4) as recently demonstrated in the design of the enhancer lysine acetyltransferase (CBP/p300) degrader dCBP-1 where both the ligand and linkage point were successfully determined prior to synthesis using Rosetta.⁵⁵ With available tools, we envision a workflow that entails the computational protocols first being implemented as a method to determine POI•E3 ligase compatibility through examination of the energy landscapes of the protein•protein interfaces. Potent degrader hits could then be accessed through a mixture of in silico linker length prediction combined with the synthetic strategies outlined previously (Figure 2). Translating degrader leads that are selective between highly conserved protein targets would be carried out more rigorously, such that degrader potency is maintained, while degrees of freedom of the ternary complex are minimized, as previously shown (Figure 3). Ultimately, the goal of in silico PROTAC development aims to increase throughput by minimizing the number of synthesized and biologically evaluated molecules in order to satisfactorily degrade a target protein. To this end, these protocols have already achieved their goal, and future advancements bode well for this hybrid approach.

TABULATING THE EMPERICAL SAR OF LINKER LENGTH

One of the most critical features observed early in PROTAC discovery, was the role of linker length.²² Early studies^21,45 showed that there is complex interplay between having a too short or too long of a linker. In an effort to further reduce the synthetic burden of linker length SAR studies, we explored the PROTAC literature and systematically tabulated potency data as a function of linker length in order to search for trends. Our goal was to explore a possible model with predictive properties that would be agnostic to the PROTAC system under study and could suggest optimal linker length from minimal empirical inputs. This strategy would not eliminate the empirical nature of an SAR study but could potentially reduce the number of molecules needed to determine the linker length that produces the most potent degrader.

PROTACs are primarily characterized through binary and ternary complex formation dissociation constants (K_d), half maximal inhibitory concentrations (IC₅₀ values), maximal degradation percentage (D_max), and half maximal degradation concentration (DC₅₀ values).⁵⁶ For this study we chose to examine the reported DC₅₀ values, defined as the concentration at which half-maximal degradation is observed, of the PROTACs in comparison to their relative linker lengths based on the fact that they encapsulate in a single measured value the abundance of metabolic processes at play. Using DC₅₀ values one can assume that the effects of target protein degradation and resynthesis kinetics, cellular permeability and efflux, native protein expression levels, rate of ubiquitination and subsequent degradation, and other metabolic considerations between different protein targets and cell lines would be included within the dataset.

Similar efforts have been made to correlate heterogeneous IC₅₀ values with chemical structure, and it has been shown⁵⁷ that this strategy does not work due to variability in the assays and conditions used to measure the values, as well as errors in data tabulation on large databases. To avoid errors in data tabulation we chose to normalize the degradation potency data by hand, with careful consideration to compare PROTACs across different studies. Maple et. al.²⁹ developed a degrader scoring measurement and compared 422 degraders from 73 different articles and were able to show correlations between degrader efficacy and increasing clogP, decreasing total polar surface area, and decreasing hydrogen bond donor count. Here we take a narrow view and attempt to add to these findings as it pertains to linker length specifically.

In terms of PROTAC structure, comparison of degradation values were only made when they utilized the same E3 ligase ligand, POI ligand, linkage point on each ligand, linkage functionality (such as amide bond or analine linkages) and linker chemical composition (exemplified by aliphatics or PEGs). Also, only linear linkers were considered for the study due to a lack of extensive published SAR series for more rigid (piperazine-type) linkers,⁵⁸ as it pertains directly to linear length versus DC₅₀ values. We also excluded studies of covalent degraders and homo-degraders due to their implicit differences in mechanism. In terms of biological evaluation, comparison was only made when they shared the same target, were evaluated in the same cell line, DC₅₀ values were collected at the same time points via the same methodology. These normalization restrictions effectively allow only for direct comparison of linker variants across an SAR series within the same study; however, it also allows comparison of all the various compound series for which there is a sufficiently large (> 2 compounds) data set.

Due to the conservative normalization strategy pursued, the number of eligible SAR campaigns was greatly narrowed. An exhaustive review of the literature (as of July 12, 2020) revealed 26 series of compounds across 12 SAR studies^{25,45,59–68} that were normalized and compared for this analysis. There are likely more unpublished PROTAC DC₅₀ values that were omitted from publication due to lackluster performance, and we urge the future publication of such negative data. Even though some compounds may not make useful preclinical candidates, there is great value in sharing all data sets for the gestalt of PROTAC development. To this end, we have provided the raw data from this analysis for others to access and elaborate upon the model (see Supporting Information).

To begin our analysis we normalized our entire data set to the single most potent degrader across all 26 compound series and compared how linker length affected the degradation potency (ΔDC₅₀ value). As might be expected, no coherent correlation was found when directly comparing the eligible literature degradation data as a whole (Figure 5a). In order to overcome the issues ingrained with direct comparison, we applied the normalization strategy described above. The most optimal degrader from each SAR compound series, as chosen by the authors, had both coordinates set to zero (ΔDC₅₀ value) and zero linker length in order to observe how a change in the number of linear atoms affects potency. Other PROTACs from within the same series would then be normalized in reference to the most potent compound for linker length (number of linear atoms)(Figure 5b). A final consideration had to be made pertaining to compounds reported as having a DC₅₀ > 3000 nM.^25,60 We have reported these values as the maximum (3000 nM) for graphical representation only, however, the overall trend is apparent without their inclusion. We also expect evaluation of the less potent degraders to be complicated by the hook-effect,⁶⁹ as increasing concentrations of PROTAC are required.

Figure 5.

Comparison of aggregate and filtered linker length SAR studies after normalization. a) The selected SAR studies normalized to the most potent reported compound as an aggregate. b) The selected SAR studies normalized to other compounds that meet the more rigorous filtering criteria described.

The results of normalization (Figure 6) show an apparent ‘boot-shaped’ pattern across all of the compound series examined, and although the data is not coherent enough to provide useful quantitative (non-linear regression based) interpretation, useful qualitative inferences can be made. We qualitatively break this data into three sections relative to linker length. The section left of Δ linker length = 0 yields a steep drop-off in PROTAC potency, which we attribute to increased contributions of steric clash due to shorter linker lengths, which likely diminishes ternary complex formation efficiency (Figure 6). The section centered around 0 yields the most potent PROTACs (due to normalization procedures), but when compared to the section on the left suggests this zone is likely where protein•protein interactions are most favorable for the given system, regardless of whether they are cooperative,^32,33,70 non-cooperative,⁴⁵ or negatively cooperative.^21,45 The zone to the right of +6 yields effective PROTACs with diminishing potency, which we attribute to entropic effects intrinsic to longer linker lengths and more degrees of freedom in the ternary complex. Ongoing debate has focused on the importance of protein•protein interaction between the E3 ligase and POI as it pertains to PROTAC development, and although leveraging this interaction is likely paramount to considerations of selectivity, this analysis suggests the proximity model could also coexist with regard to degradation potency.

Figure 6.

Representation of the qualitative inferences made. Qualitatively we describe this trend in three zones. Left of center represents a region where steric clash can hinder successful ternary complex formation leading to a sharp drop off in degradation potency. Center represents the most potent degraders. Right of center represents the diminishing degradation potency due to increased entropy associated with longer linkers.

This analysis suggests that an empirical linker length SAR study might be best initiated with longer linker lengths as a means to test for compatibility between the E3 ligase and POI. However, the data does not necessarily suggest that any linker length larger than the optimum will produce a successful degrader; only that, in general, the chances of finding a degrader are increased with longer linkers due to the absence of steric clash. If a compatible E3 ligase•POI system is identified, systematically shortening the length of the linker will identify the sharp cutoff point in potency and locate the linker length that induces potency abolishing steric clash. In general, this analysis suggests that the most potent PROTACs were found within a few atoms of that linker length. In addition, this region may provide the best chances of selectivity by minimizing degrees of freedom in the ternary complex while maintaining potency.

The model presented was comprised of limited published data and can only be used to make qualitative inferences about how linker length can affect degrader potency. We would expect that no such model could be generated for degrader selectivity, as each system will have its own unique interprotein contacts, and normalization for direct comparison would be difficult between systems. However, we can imagine that, were enough data reported and added to the normalization, a quantitative description of the trend may become relevant through non-linear regression. This would likely take the form of a weakly fitted curve due to the intrinsic biological and pharmacological complexities inherent in DC₅₀ measurements, yet it could wield suggestive power by fitting a novel limited linker SAR dataset to the determined curve to examine which zones (Figure 6) have been identified for a compound series in a novel system. For this reason, we intend to update and evaluate the publicly available dataset into the future. This could offer another tool to reduce the synthetic bottleneck caused by linker length SAR studies.

SUMMARY

In conclusion, the empirical nature of PROTAC development campaigns is being eroded by the harmonious development in synthetic methodologies of chimeric small molecules, structural characterization of ternary complexes, advances in computational protocols, and systematic analysis of prior studies. PROTAC development presents a dramatic detour for drug design, in that the focus has shifted to small molecules capable of stabilizing non-cognate protein interactions long enough for poly-ubiquitination. This event–driven pharmacology allows for repurposing of previously abandoned ligands that had been shelved due to lack of potency or lack of selectivity. It also offers a method to effectively target non–catalytic protein targets, a significant expansion of possible leads. Through continued cooperation between the fields of expertise outlined here, we expect the coupled reduction of lead-time for development of PROTACs through the development of a solution to the empirical nature of the linker domain. We await further validation of this exciting pharmacological modality in the clinic.

ABBREVIATIONS

Boc

tert-butoxycarbonyl

BRD4

bromodomain-containing protein 4

BTK

Bruton’s tyrosine kinase

CRBN

cereblon

CRL2^VHL

E3 ligase Cul2 with Von Hippel-Lindau substrate binding domain

CRL4^CRBN

E3 ligase Cul4 with cereblon substrate binding domain

DABCO

1,4–diazabicyclo[2.2.2]octane

DC₅₀ value

half-maximal degradation concentration

DIC

N,N’–diisopropylcarbodiimide

DIPEA

diisopropylethylamine

D_max

maximal degradation percentage

HATU

O–(7–azabenzotriazol–l–yl)–N,N,N’,N’–tetramethyluronium hexafluorophosphate

HPLC

high pressure liquid chromatography

IC₅₀ value

half-maximal inhibitory concentration

IMiD

immunomodulatory imide drugs

K_d

dissociation constant

LCMS

liquid chromatography-mass spectrometry

MOE

Molecular Operating Environment

PD

pharmacodynamic

PEG

polyethylene glycol

Pfp

pentafluorophenyl

PK

pharmacokinetic

POI

protein of interest

PROTAC

proteolysis targeting chimera

PS-DVB

polystyrene-divinyl benzene

RP-HPLC

reverse phase high pressure liquid chromatography

SAR

structure activity relationship

SMARCA2

probable global transcription activator SNF2L2

SMARCA4

transcription activator BRG1

TFA

trifluoroacetic acid

TPR

thalidomide preloaded resin

UPS

ubiquitin-proteasome system

VHL

Von Hippel-Lindau

Biographies

Biographies

Troy A. Bemis received his BSc in Chemistry from the University of Wisconsin – Madison where he worked with Prof. Sandro Mecozzi developing theranostic nanoparticles. In 2015 he began his Ph.D studies at the University of California – San Diego with Prof. Michael D. Burkart where he is currently a Ph.D candidate with a focus on the development of chimeric small molecules for the purposes of drug development and characterization of natural product biosynthetic pathways.

James J. La Clair received his Ph.D. degree in Chemistry from SUNY Buffalo under the tutelage of Prof. Peter T. Lansbury. His studies continued through a postdoctoral fellowship with Prof. Gilbert Stork at Columbia University. He then undertook a position at the Scripps Research Institute. In 1999, he shifted his focus to work on a project–by–project basis. This has included research positions at UC San Diego and the Salk Institute, among other collaborative endeavors

Michael D. Burkart is the Teddy Traylor Faculty Scholar of Chemistry & Biochemistry at the University of California, San Diego. He received a BS in Chemistry from Rice University and a Ph.D. in Chemistry from The Scripps Research Institute, followed by a post-doctoral fellowship at Harvard University. The Burkart Laboratory pursues interdisciplinary research in the fields of organic and biological chemistry, with a focus on natural product synthesis and biosynthesis.