PROTAC development and characterization
Recently we reported the development of potent and selective DCAF1 WDR domain binders.16 Leveraging our extensive contribution to the development of potent, selective, and cell active WDR5 ligands21,22,24,28,30, we set out to investigate the viability of DCAF1 ligands as handles for WDR5 degradation. The disclosure of VHL31,32 and CRBN33 recruiting degraders of WDR5 further encouraged us to pursue the development of DCAF1-based WDR5 degrader strategy. Based on the crystal structure of OICR-8268 as our starting point (PDB ID: 8F8E)16, we modified the core pyrazole to allow for a substitution that would proceed along a vector to access the exposed solvent front. Unfortunately, the direct substitution on the pyrazole core abolished the DCAF1 potency, as previously discussed.16 However, switching from pyrazole to pyrrole allowed for such substitution while maintaining activity for DCAF1 (Scheme 1).
In order to pursue the design strategy, we opted to start with the WDR5 binders that we had previously developed and were used by Jin’s team32, and worked well within the VHL-based PROTAC. To link the two components of the PROTACs together, we started with simple alkyl and PEG based systems to vary the length from 2 to 14 atoms to confirm if recruiting DCAF1 could lead to WDR5 degradation (Scheme 2).
Following this approach, we developed PROTAC 1 (OICR-40333), PROTAC 2 (OICR-40407) and PROTAC 3 (OICR-40792) (Fig. 1). Binding of all three PROTACs to both WDR5 and DCAF1 were confirmed by surface plasmon resonance (SPR) with double digit nanomolar KD (Fig. 1 and Supplementary Fig. 1a-c). The kon and koff values for each PROTAC are presented in Supplementary Fig. 1, indicating reasonable residence time as expected.
The PROTAC molecules were orthogonally tested by differential scanning fluorimetry (DSF) assay as described in Supplementary Methods,34 to assess their binding to DCAF1 and WDR5 alone, and to both proteins together as a ternary complex (Fig. 2). In DSF, binding of a small molecule to the protein stabilizes the target and leads to an increase in melting temperature (Tm). In this experiment, all three PROTACs (1–3) stabilized both WDR5 and DCAF1 proteins, a confirmation of binding of these PROTACs to both proteins. Notably, the stabilization of WDR5 was more pronounced than that of DCAF1 (Fig. 2a, b, and d). PROTAC 1 and PROTAC 2 showed similar patterns and levels of stabilization of DCAF1 (ΔTm of 8.0 ± 0.1°C and 7.5 ± 0.1°C, respectively) and WDR5 (ΔTm of 13.4 ± 0.1°C and 10.6 ± 0.1°C, respectively). Interestingly, PROTAC 3 with the longest linker, showed less stabilization of both DCAF1 (ΔTm of 5.3 ± 0.1°C) and WDR5 (ΔTm of 5.9 ± 0.1°C).
To test ternary complex formation, we also elected to employ DSF. It has previously been reported that binding of small molecules to different domains of a protein will have differential stabilization effects on domain melting.35 Therefore, a mixture of proteins in solution is expected to behave in a similar manner. The melting curves of DCAF1 and WDR5 together in solution show double peaks representing the melting of DCAF1 first, followed by WDR5 denaturation (Fig. 2c). In the presence of the PROTACs, the DCAF1-WDR5 melting curves change shape and/or shift slightly rightward, supporting ternary complex formation (Fig. 2c).
Structural Studies of PROTAC ternary complexes
To further confirm ternary complex formation and better understand the impact of linkers in both enabling the DCAF1-WDR5 interaction and positioning WDR5 for DCAF1-mediated degradation, we solved the crystal structures of DCAF1-PROTACs (1–3)-WDR5 ternary complexes.
Structures of PROTAC ternary complexes show different orientations of WDR5 relative to DCAF1.
Using heterologously expressed and purified proteins, we solved by molecular replacement the high-resolution crystal structures of DCAF1 and WDR5 in ternary complexes with PROTACs 1, 2, and 3 using previously determined structures of the WDR domains of DCAF1 (PDB id 8F8E) and WDR5 (PDB id 2GNQ) as search models (Fig. 3a).16,36 The structures were refined to a resolution of 1.92–2.06 Å with an R-work/R-free of 0.1724–0.2125/0.1836–0.2295, with greater than 99.5% of residues in favored and allowed Ramachandran regions (Table 1). Importantly, electron density for the PROTAC was seen in a simulated annealing composite omit map for each ternary complex (Supplementary Fig. 2a-c). Iterative structure refinement resulted in a clear 2Fo-Fc electron density map that allowed mapping of all atoms for each PROTAC and their interactions in ternary complex with DCAF1 and WDR5 (Fig. 3a, Supplementary Fig. 2e-f).
Both DCAF1 and WDR5 have a β-propeller domain topology consisting of seven blades, each formed by four antiparallel β-strands interconnected by loops of varying lengths (Supplementary Fig. 3). In each of these proteins, the first N-terminal β-strand is part of the seventh blade and connects to the first β-strand of the first blade. Since the PROTACs were built using DCAF1 and WDR5 anchors that bind to the top pocket of the protein’s WDR domains, as expected the ternary complexes show each PROTAC connecting the two proteins in a top-to-top orientation (Fig. 3a). In the PROTAC 1 ternary complex, WDR5 sits on top of DCAF1 in a tilted orientation due to bending of the WDR5 anchor relative to the straight, extended configuration of the rest of PROTAC 1. Interestingly, the PROTAC 2 and PROTAC 3 ternary complex structures show protein arrangements in which DCAF1 and WDR5 are in essentially the same orientation with respect to each other and superimpose with an r.m.s.d. of 0.20 Å. In both complexes, the longer PROTAC linkers turn inside the protein interface in a screw-like fashion, resulting in a placement of the WDR5 protein where its 7-fold pseudosymmetry axis is more aligned to that of DCAF1 than in the PROTAC 1 ternary complex (Fig. 3a).
Table 1
X-ray data collection and refinement statistics.
Structure | PROTAC 1 ternary complex | PROTAC 2 ternary complex | PROTAC 3 ternary complex |
PDB id | 9B9H | 9B9T | 9B9W |
Resolution Range | 42.57–2.06 (2.1–2.06) | 43.21–2.05 (2.09–2.05) | 43.29–1.92 (1.95–1.92) |
Space group | P 21 21 21 | P 21 21 21 | P 21 21 21 |
Unit cell dimensions a, b, c (Å) α, β, γ (°) | 84.21, 85.14, 111.64 90, 90, 90 | 79.86, 83.71, 130.13 90, 90, 90 | 79.90, 83.53, 130.83 90, 90, 90 |
Total reflections | 97,507 (4265) | 110,124 (5450) | 134,923 (5506) |
Unique reflections | 49,405 (2399) | 55,308 (2779) | 67,507 (2761) |
Multiplicity | 2.0 (1.8) | 2.0 (2.0) | 2.0 (2.0) |
Completeness (%) | 95.18 (68.18) | 99.07 (89.01) | 99.98 (100.00) |
Mean I/sig(I) | 25.11 (1.05) | 14.33 (1.22) | 15.04 (1.53) |
Wilson B-factor | 37.78 | 32.10 | 28.11 |
R-merge | 0.01625 (0.6408) | 0.03048 (0.5883) | 0.03192 (0.4724) |
R-meas | 0.02298 (0.9062) | 0.04311 (0.8319) | 0.04515 (0.6681) |
R-pim | 0.01625 (0.6408) | 0.03048 (0.5883) | 0.03192 (0.4724) |
CC1/2 | 1 (0.952) | 0.999 (0.897) | 0.999 (0.594) |
CC* | 1 (0.988) | 1 (0.972) | 1 (0.863) |
Reflections used in refinement | 47,892 (1991) | 54,931 (2544) | 67,500 (2761) |
Reflections used for R-free | 2,395 (106) | 2,650 (128) | 3,397 (139) |
R-work | 0.2108 (0.5275) | 0.1857 (0.4289) | 0.1719 (0.2646) |
R-free | 0.2291 (0.5826) | 0.1962 (0.4310) | 0.1830 (0.2593) |
No. of non-hydrogen atoms | 5,279 | 5,418 | 5,354 |
macromolecules | 4,882 | 4,787 | 4,786 |
ligands | 87 | 91 | 93 |
solvent | 310 | 540 | 475 |
Protein residues | 619 | 608 | 605 |
RMS (bonds) (Å) | 0.005 | 0.008 | 0.007 |
RMS (angles) (°) | 0.77 | 1.10 | 0.85 |
Ramachandran favored (%) | 94.47 | 95.67 | 95.49 |
Ramachandran allowed (%) | 5.37 | 3.83 | 4.34 |
Ramachandran outliers (%) | 0.16 | 0.50 | 0.17 |
Rotamer outliers (%) | 1.64 | 0.75 | 0.37 |
Clashscore | 8.34 | 7.53 | 6.06 |
Average B-factor | 46.56 | 36.04 | 31.07 |
macromolecules | 46.41 | 35.18 | 30.21 |
ligands | 41.07 | 35.10 | 33.38 |
solvent | 48.88 | 43.82 | 39.32 |
Statistics for the highest resolution shells are found in parentheses.
DCAF1 and WDR5 anchors of PROTACs are stabilized by various noncovalent interactions.
Similar to the binding of OICR-8268 to DCAF116, the shared DCAF1 anchor of PROTAC 1–3 is stabilized in the protein pocket by hydrogen bonds (H-bonds), electrostatic, hydrophobic, and van der Waals interactions with protein residues (Fig. 3b, Supplementary Fig. 2e-f). Briefly, an H-bond/electrostatic interaction occurs between a carbonyl group of the DCAF1 anchor and the guanidinium group of R1298 (blade 6). Together with R1225 (loop connecting blades 3 and 4, i.e. 3–4 loop), R1298 forms an “arginine clamp” that forms a cation-π interaction with the 2,4 di-halogenated aromatic ring of the DCAF1 anchor. Further stabilizing the DCAF1 anchor in the pocket are two H-bonds between its pyrrole group and the peptide carbonyl groups of T1097 (blade 1, water-mediated) and F1355 (blade 7, direct), and its backbone amino group and the carboxylate group of D1356 (blade 7). Another amino group forms H-bonds with the side chain carboxylate group of D1356 and the peptide carbonyl group of C1099 (blade 1, water-mediated). A carbonyl group of the DCAF1 anchor participates in an H-bond with the imidazole group of H1140 (blade 2). Several hydrophobic side chains line the pocket and form stabilizing interactions with the aromatic rings of the DCAF1 anchor. These interactions contribute to the tight binding of the PROTAC to DCAF1 (Figs. 1 and 3b, Supplementary Fig. 1, Supplementary Fig. 2e-f).
Contraposed to the DCAF1 anchor, the WDR5 anchor on the other end of each PROTAC binds the WDR5 pocket predominantly through hydrophobic interactions using two substituted heterocyclic rings and two substituted phenyl rings (Fig. 3c, Supplementary Fig. 2e-f). Two water molecules bridge an H-bond between a carbonyl group of the WDR5 anchor and the peptide amino group of C261 (5–6 loop), and an N atom of the piperazine ring of the WDR5 anchor and the peptide carbonyl group of C261. The piperazine ring is found in a stable chair conformation inside the binding pocket. Finally, two amino groups of the WDR5 anchor participate in H-bonding with the side chain hydroxyl group of S91 (blade 2) and carboxylate group of D107 (blade 2). PROTAC-induced proximity also enables DCAF1 to form part of the WDR5 anchor binding pocket using the indole group of W1156 (blade 2) (Fig. 3c, Supplementary Fig. 2e-f).
Structural differences between PROTAC ternary complexes.
a. The DCAF1-WDR5 interface changes with increasing PROTAC linker length.
The surface representations in Fig. 4a show different overall shapes of the PROTAC ternary complexes. This difference is due to the changing juxtaposition of WDR5 and DCAF1 in the complexes, arising from the rotation of the WDR5 anchor relative to the DCAF1 anchor. Superposition of the DCAF1 component of the three complexes shows that WDR5 in both PROTAC 2 and PROTAC 3 ternary complexes is tilted vertically by 8º and rotated clockwise by 152º from its position in the PROTAC 1 complex (Supplementary Fig. 4). The ternary complexes also exhibit diminishing contacts between DCAF1 and WDR5 with increasing linker length of the bound PROTAC, despite the PROTACs having similar effective lengths inside the complex (Fig. 4a). This arises from the changing blade interactions engendered by WDR5 tilting and rotation relative to DCAF1 (Fig. 4b-d, Supplementary Fig. 4). Among the three, the PROTAC 1 ternary complex has the largest buried surface area37 at the protein interface (1100 Å2), which is stabilized by H-bonding and electrostatic interactions between DCAF1 and WDR5 loop residues. Briefly, these H-bond/electrostatic interactions occur between the side chains of N1132 (DCAF1 blade 1) and K67 (WDR5 blade 1), the side chain of N1135 (1–2 loop) and the peptide carbonyl of A65 (WDR5 blade 1), the side chains of E1091 (DCAF1 blade 7 − 1 loop) and K87 (WDR5 blade 1–2 loop), the side chain of K1327 (DCAF1 blade 6) and the peptide bond carbonyl of S129 (WDR5, blade 2–3 loop), and the side chains of R1325 (DCAF1 blade 6) and E151 (WDR5 blade 3) (Fig. 4b). DCAF1’s R1325 and K1327 residues are found in the protein’s longest loop of blade 6 spanning residues 1313–1333 (Supplementary Fig. 3). This flexible loop is fully resolved in the PROTAC 1 ternary complex but not in the PROTAC 2 and 3 ternary complexes, and its conformation varies from one complex to another.
In the PROTAC 2 ternary complex, the protein interaction’s buried surface area is decreased (820 Å2), with DCAF1 and WDR5 brought together by PROTAC 2 and further held together by three electrostatic/H-bonding interactions involving the side chain of R1325 (DCAF1 blade 6) and the side chains of Y260 (WDR5, blade 5–6 loop) and D302 (WDR5, blade 6–7 loop) (Fig. 4a,c). In the PROTAC 3 ternary complex, H-bonding and electrostatic interactions are absent, and only three pairs of amino acid residues are close enough to each other to participate in hydrophobic and van der Waals interactions. These pairs of opposing residues include Q1158 (DCAF1 blade 2) and E151(WDR5 blade 3), W1156 (DCAF1 blade 2) and Y191 (WDR5 blade 4), and L1378 (DCAF1 blade 7) and N323 (WDR5 blade 7) (Fig. 4d). Like DCAF1’s R1325 and K1327 residues that comprise blade 6’s long flexible loop involved in WDR5 binding, the L1378 residue is part of a nearby flexible loop in blade 7, spanning residues 1372–1381, that has different conformations across the ternary complexes (Fig. 4b-d). These flexible loops appear to be important in binding each WDR5-PROTAC pair to form the ternary complex. Nevertheless, the paucity of stabilizing non-covalent forces at the DCAF1-WDR5 interface highlights the dominant role of the PROTACs in bringing the two proteins together to form the ternary complex. PROTACs with longer linkers may lead to more separated DCAF1 and WDR5 components.
b. PROTAC linkers assume similar effective lengths in ternary complexes.
The similar effective lengths of PROTACs in the ternary complexes, measured as the distance between the para fluorine atom of the DCAF1 anchor and the N-methyl carbon atom of the WDR5 anchor, was an unexpected finding (Fig. 4a). In the PROTAC 1 ternary complex, this length is 39.1 Å for the extended form of PROTAC 1. In contrast, in the ternary complexes, PROTAC 2 and 3 have an effective length of 38.2 Å, despite having longer linkers than PROTAC 1. To fit inside the binding space between DCAF1 and WDR5, the linkers of PROTAC 2 and 3 compact by making a turn beginning at C5 of the pyrrole ring of the DCAF1 anchor. The PROTAC 3 linker makes a wider turn and has a smaller pitch than that of PROTAC 2 to fit inside the binding pocket (Supplementary Fig. 5). Their linkers then merge at the carbonyl carbon of the WDR5 anchor, arriving at the same position for their WDR5 anchors that is yet distinct to that in the PROTAC1 ternary complex. Overall, this leads to two different orientations for WDR5 in the ternary complexes (Supplementary Fig. 4, 5). It appears that the effective PROTAC lengths observed in crystal structures is determined by the available binding space between DCAF1 and WDR5 in the most stable configuration(s) of the ternary complex. As a corollary to this, a linker length limit exists at which the PROTAC can no longer fit inside the binding space and DCAF1 and WDR5 can no longer form stabilizing contacts, leading to ternary complexes with non-interacting DCAF1 and WDR5 components. Conversely, PROTACs with shorter linkers than PROTAC 1 may be unable to form stable ternary complexes due to insufficient separation between DCAF1 and WDR5, forcing an energetically unfavorable collision between them. Indeed, we solved the crystal structure of a DCAF1-WDR5 binary complex from a crystal grown in the presence of PROTAC 4 (Supplementary Fig. 6, Supplementary Table 1), whose linker is shorter than that of PROTAC 1. As with PROTAC 1–3, PROTAC 4 also binds to DCAF1 (SPR KD of 15.2 ± 2 nM) and WDR5 (SPR KD of 26.4 ± 4 nM) (Supplementary Fig. 7), which were orthogonally confirmed by DSF (Supplementary Fig. 8). PROTAC 4 even stabilized both DCAF1 and WDR5 in a mixture when tested at equimolar concentrations, indicating its ability to bind to both proteins at the same time. However, the crystal structure is devoid of bound PROTAC 4, indicating the ternary complex probably is of low stability. A few non-covalent interactions, one of which involves the E1324 residue in the flexible loop of DCAF1’s blade 6, stabilize the observed (DCAF1) top-to-(WDR5) bottom configuration that is unlike the top-to-top configuration expected of an active PROTAC (Supplementary Fig. 9a-d). Although no electron density was observed for PROTAC 4 in this structure, it appears to bring DCAF1 and WDR5 together stably enough to co-crystallize. DCAF1 and WDR5 did not co-crystallize in the absence of the PROTAC (control experiment), further supporting a role for PROTAC 4 in the binary complex formation.
c. PROTAC linker curving engenders different interaction patterns at the DCAF1-WDR5 interface.
Different interaction patterns involving protein residues and ordered water molecules help in stabilizing PROTAC linker configurations observed in crystal structures (Fig. 4e-g). In the PROTAC 1 ternary complex, the extended linker is stabilized by various hydrophobic and van der Waals interactions with protein residues lining the pocket, mainly from nearby β-strands and loops of blades 1 and 2 (Fig. 4e). The imidazole group of W1156 (blade 2) of DCAF1 forms water-bridged H bonds with an amide carbonyl group proximal to the WDR5 anchor, while also providing a hydrophobic binding face for the aliphatic parts of the linker as it is wedged in between the linker and three aromatic rings. In addition, DCAF1 residue T1097 (blade 1) uses a peptide carbonyl group to form water-mediated H-bonds with the amide carbonyl proximal to the DCAF1 anchor, while R1225 (3–4 loop) forms a direct H-bond with the amide N-H of the linker. The DCAF1, blade 6 flexible loop (residues 1313–1333) is fully ordered but does not interact with the PROTAC 1 linker, leaving a solvent-filled cavity. However, through the R1325 (DCAF1)-E151 (WDR5) salt bridge, the flexible DCAF1 loop extends towards the surface of WDR5, thereby forming a significant part of the wall of the linker’s binding pocket (Fig. 4b,e). In WDR5, D172 (3–4 loop) forms water-mediated H-bonds with the amide N-H group and the oxygen atom of the of the PEG subunit proximal to the WDR5 anchor.
In the PROTAC 2 ternary complex, the PROTAC linker is curved and stabilized predominantly by H-bonds with DCAF1 residues, most of which are mediated by water molecules (Fig. 4f). Importantly, the blade 6 flexible loop (residues 1313–1333) assumes a different, “close lid” conformation whose shape complements the curvature of the linker. Although residues 1317–1320 of this loop could not be modeled, the conformation of the rest of the loop could be confidently assigned and is seen to form important interactions with the PROTAC 2 linker. For instance, the side chains of M1322 and M1326 straddles the linker, forming a “methionine clamp” around the midsection. The peptide backbone groups of E1324, K1327, and M1332 form water-mediated H-bonds with the linker, while E1323 uses its side chain carboxylate group to form water-mediated H-bonds. The missing residues 1317–1320 of the flexible loop (DEDD sequence), although not modeled due to disorder, likely interacts with both polar and nonpolar atoms of the linker through their side chains. Other solvent-bridged H-bonds with the linker involve the peptide carbonyl groups of I1138 and T1139 (blade 2), and the side chain hydroxyl group of T1155 (blade 2). The indole group of W1156 (blade 2) participates in a direct H-bond with the amide carbonyl proximal to the WDR5 anchor. The guanidinium group of R1225 (blade 3–4 loop) also forms both direct and water-mediated H-bonds with the linker, while at the same time helping maintain loop conformation through water-mediated H-bonds with the peptide amino group of Q1314 (blade 6) and the peptide carbonyl group of P1329 (blade 6). An extensive network of hydrophobic and van der Waals interactions exists inside the binding pocket between the PROTAC 2 linker and DCAF1 residues, whose side chains collectively, determine the overall shape of the pocket. Notably, the indole group of W1156 (blade 2) undergoes a rotation from that seen in the PROTAC 1 ternary complex, presumably to adjust to the shape of and form a better binding pocket for the linker. Finally, even though no WDR5 residue interacts with the PROTAC 2 linker via H bond, the side chains of WDR5 residues K259 (blade 5–6 loop) and Y260 (blade 5–6 loop) form a binding surface for a portion of the linker.
The structure of the PROTAC 3 ternary complex reveals a more compressed PROTAC 3 linker constrained by the available space between DCAF1 and WDR5 (Fig. 4g, Supplementary Fig. 5). This configuration is sustained by several H-bond interactions with DCAF1 residues, most of which are bridged by water molecules. Interestingly, this configuration is stabilized by water-mediated H-bonds between polar atoms of PROTAC 3 itself. As in the PROTAC 2 ternary complex, W1156 (blade 2) of DCAF1 provides a hydrophobic face to bind the WDR5 anchor and part of the linker that follows, yet is no longer wedged between the linker and aromatic rings as seen in the PROTAC 1 complex due to rotation of the WDR5 anchor (Fig. 4e,g). A direct H-bond with the amide carbonyl group of the linker near the DCAF1 anchor occurs using the guanidinium group of R1225 (blade 3–4 loop), which, like in the PROTAC 2 ternary complex, helps position the flexible loop (residues 1313–1333) to form the pocket through water-mediated H-bonds with the peptide carbonyl groups of Q1314, K1327, S1328, and P1329 of the flexible loop. This loop-stabilizing role of R1225 observed in the PROTAC 2 and 3 complex structures is not seen in that of the PROTAC 1 ternary complex, where no bridging water molecules are observed in the gap between residue and loop (Fig. 4e). Other water-mediated H-bonds with the linker are formed by DCAF1 using the hydroxyl group of T1155 (blade 2, two rotamers modeled), and the peptide carbonyl groups of I1138 (blade 2) and T1139 (blade 2). No direct H-bonds occur between the PROTAC 3 linker and any WDR5 residues. However, as in the PROTAC 2 ternary complex, the side chains of K259 (blade 5–6 loop) and Y260 (blade 5–6 loop) of WDR5 interact with part of the linker through van der Waals and hydrophobic interactions.
To accommodate increased linker curvature, the blade 6 flexible loop (residues 1313–1333) of DCAF1 has moved away in the crystal, as indicated by broken electron density for part of the loop (Fig. 4g). Residues 1317–1326 (DEDDLMEERM) of the loop were not modeled since their side chains could not be resolved, although electron density for backbone atoms for residues 1324–1326 (ERM) could be seen (Supplementary Fig. 10a). This indicates that this loop remains part of the linker’s binding surface while some of its constituting residues change side chain orientations, possibly using polar and/or negatively charged groups to interact with polar O atoms of the linker. A nearby flexible loop in blade 6 (residues 1372–1381) also forms part of the linker binding pocket, providing a van der Waals interactions surface. Several other residues of DCAF1 participate in van der Waals interactions with the PROTAC 3 linker, while fewer are made by those of WDR5. Due to the increased separation between DCAF1 and WDR5 in the complex, the binding interface is more porous, leaving the linker more deprotected from solvent than those of PROTAC 1 and 2 (Fig. 4a,g). Overall, DCAF1 loops play significant roles in providing plasticity in interacting with different PROTAC-WDR5 complexes, perhaps a reflection of DCAF1’s ability to function as a substrate receptor for E3 ligases with diverse set of substrates.