What are heterobifunctional molecules?

With the constant issues arising with small molecule drug discovery, such as enzymatic resistance mechanisms or high off-target toxicities, there has been a recent resurgence in alternative pharmaceutical approaches. One such approach is the use of proteolysis-targeting chimeras, or degraders. Instead of a single ligand binding to the active- or binding-site(s) of a protein or protein family, degraders typically consist of (1) a ligand binding to the target protein, (2) a ligand binding to an E3 ligase, and (3) a linker connecting the two ligands to allow for the target protein to be ubiquinated and therefore tagged for degradation. Figure 1 outlines the protein degradation cycle.¹

Figure 1. Proteolysis cycle, including recycling of the degrader and ubiquitination of the target protein of interest. This figure was partly generated using Servier Medical Art, provided by Servier, licensed under a CC BY 4.0 license (CC BY 4.0 Deed | Attribution 4.0 International | Creative Commons).

Degraders in drug discovery

Degraders do have several advantages over small molecules in terms of drug discovery approaches. For example, small molecules are known to lead to acquired resistance, particularly in cancer targets, whereas degraders have exhibited high activity in inhibitor-resistant cancer models. Small molecules also tend to have many off-target activities and therefore toxicities, while the complexity of these heterobifunctional molecules allow for more specific binding to the protein and ternary complex. Alongside their specificity, heterobifunctional degraders have demonstrated sustained binding, allowing for less frequent dosing than reversibly binding small molecules. With this sustained binding and complexity, degraders can also tolerate low affinity binding to the target protein, while small molecules tend to require high affinity to the target protein. However, because of their high complexity, degrader ADME properties can be much more difficult to optimize than their small molecule counterparts. Their complexity can also lead to more advanced resistance mechanisms, whereas the resistance mechanisms associated with small molecules tend to be single point mutations and therefore slightly easier to navigate.^1-2

Computational solutions to the challenges of degrader design

With the recent resurgence of targeted protein degradation as an approach in drug discovery, computational techniques provide a promising solution to the challenging task of designing heterobifunctional degraders, particularly in the design of degrader linkers. With Cresset’s proprietary Electrostatic Complementarity (EC) analysis in Flare™,³ we can assess the complementarity of a degrader structure to a protein target of interest, especially when the linker is involved in stabilization of the ternary complex. In this article, we will use Spark™ to run a bioisostere replacement experiment to modify the linker moiety of a known degrader, then use Flare’s EC feature to analyze the results. We will illustrate how slight changes in the electrostatics of a linker can affect the complementarity of the degrader to the protein binding sites.

The Electrostatic Complementarity of a known degrader

Flare’s EC feature uses the electrostatics of the ligand and an associated protein to determine how well they match.⁴ To illustrate the utility of this feature in designing degrader linkers, we will use compound MZ1 that targets bromodomain and extra-terminal domain (BET) proteins, namely the bromodomain-containing protein 4 bromodomain 2 (Brd4^BD2), for degradation. In particular, MZ1 exhibits tight binding to the Brd4^BD2 and the E3 ubiquitin ligase von Hippel-Lindau tumor suppressor protein (VHL) (K_d = 3.7 nM). MZ1 has been cocrystallized with Brd4^BD2 and VHL (pdb: 5T35),⁵ shown in Figure 2.

Figure 2. Degrader MZ1 bound to Brd4^BD2 and VHL (pdb: 5T35).⁵ The protein surface is colored according to the EC.

MZ1 consists of ligand JQ1 bound to Brd4^BD2 and ligand VH032 bound to pVHL joined together by a PEG linker. We can see in Figure 2 that the EC surface of the protein complex is consistent with the low dissociation constant: the surface is mostly green, indicating good EC. Additionally, two oxygen atoms of the PEG linker form hydrogen bonds with His437 on Brd4^BD2. However, a region of the pVHL protein that is interacting with the PEG linker is slightly red, indicating electrostatic clash, or bad EC. In this experiment, we will replace the PEG linker with a fragment that improves the EC to the binding site of the protein complex but also simultaneously yields a bioisosterically similar degrader.

Designing new degraders using Spark in Flare

Critical to our approach is the use of electrostatic and shape similarity to ensure that potential replacement linkers maintain favorable interactions with the target proteins. To do so, we utilized Spark’s Scaffold Hopping experiment in Flare,⁶ using pdb 5T35 as the ternary complex containing the reference degrader. The 2D structure of MZ1 is shown in Figure 3. Although PEG chains such as the linker in MZ1 can be advantageous in linker design, PEG moieties can be susceptible to peroxidation and subsequent degradation.⁷ The PEG chain was therefore identified as the ‘scaffold’ to be replaced, and Spark then searched for linkers which resulted in final degrader structures that (1) were bioisosterically similar to MZ1 and (2) don’t clash with the surrounding proteins. To take into account the size and flexibility often seen in linkers, some freedom of movement was permitted for the two ligands by allowing up to 15 rotatable bonds in the new linker, as well as generous distance and angle tolerances between the new linker and the existing binders of 1.5 Å and 30 deg, respectively. The search itself was conducted using the Enamine linker library for known and commercially available linkers, newly available in Spark and Spark in Flare.⁸ The top results were then selected for EC analysis.

Figure 3. The structure of MZ1 with the PEG fragment to be replaced highlighted in orange.

Calculating the EC of the new Spark designs

The Spark ligand-based search results gave some interesting analogs of the PEG chain in MZ1. Two of the most notable were results 3 and 12 (Figure 4), containing more diverse hydrogen bond acceptors and donors than seen in the original PEG chain.

Figure 4. Two of the top results from the Spark scaffold hopping experiment with their respective scores. Linkers are highlighted in orange.

The EC surfaces (Figure 5) of the proteins were then calculated for each selected degrader result using Flare. Using the Contacts feature, we observe that the pyrimidine 12 potentially makes the same hydrogen bonding interactions as the PEG chain to His437 on Brd4^BD2, with an additional improved EC compared to MZ1 (See Figure 2). This suggests that the new linker, with its different electrostatics, would be an improvement over the PEG linker. The sulfone chain 3 on the other hand, despite the higher similarity score to the original PEG chain, does not hydrogen bond to Brd4^BD2 as does MZ1, and there is an additional electrostatic clash with Brd4BD2 when compared to the EC of MZ1 (See Figure 2).

Figure 5. Two of the top results visualized in Flare and overlayed with MZ1 (gray). The protein surface is colored according to the EC of each ligand to the Brd4^BD2-VHL complex.

If we apply the electrostatic potential (ESP) and EC surfaces instead to the degraders, we can more clearly see the similarities and differences among the linkers (Figure 6). While we see less subtle differences when viewing the ESP positive and negative surfaces, the differences are more-clearly identified with the EC surfaces. The PEG chain in MZ1 has minor electrostatic clashes with the protein. These clashes seem to be resolved with the pyrimidine 12; there is a larger surface area of the linker which is coloured greener than that of MZ1. The sulfone 3, on the other hand, has significant clashing with the proteins. When we calculate the EC score for each degrader, or the average complementarity over the whole surface, the scores match the observed EC surfaces: MZ1 and 12 are similar, with 12 being slightly better, and 3 is the poorest with the most clash.

Figure 6. The ESP (red/blue) and EC surfaces (green/red) and scores of the co-crystallized MZ1 (left) with the two selected Spark results, results 3 (middle) and 12 (right). Linkers are circled.

From these results, we can conclude that although MZ1 is a tight binder of the binary protein complex, the pyrimidine 12 with its more complementary electrostatics may lead to improved binding. With only ligand-based results, or the Spark similarity scores, we might have thought that result 3 was a better match. However, when combining the ligand-based Spark results with the structure-based EC results in Flare, we can predict that result 12, even with the lower bioisostere factor (5% vs 18%), may in fact be more active in vivo.

EC can help prioritize new degrader linker designs

In this experiment, we have computationally evaluated new Spark linker designs with the EC feature in Flare. Using Spark and Flare synergistically, we have found a linker that has potentially better binding to the Brd4^BD2-VHL complex. While these results are not biologically evaluated, they do offer insight into the potential diversity of linkers that complement the binding site of the protein complex.

EC can also be used to study protein-protein interactions

EC can also be used to study the protein-protein interactions at the binding site. Figure 7 shows the EC surfaces of Brd4^BD2 and VHL.

Figure 7. The EC surfaces of the two proteins. (Left) EC surface of Brd4^BD2, VHL shown as sticks. (Right) EC surface of VHL, Brd4^BD2 shown as sticks. The ‘active site’ was clipped to 6 Å.

These surfaces are indicating areas of good EC and areas of electrostatic clashes between the two proteins. We can see that overall, the protein surfaces are quite complementary with only minor clashing. To obtain these surfaces, each chain (corresponding to either the Brd4^BD2 or VHL protein) was treated as a ligand (moved to the Ligands table in Flare), and the EC surface was applied to the remaining protein chain.

References

1. Li, F. et al. Nat Commun 2022, 13, 7133
2. Xie, H. et al. Explor Target Antitumor Ther. 2021, 2(6), 511–521
3. https://cresset-group.com/software/flare/
4. Bauer, M. R. and Mackey, M. D. J Med Chem 2019, 62, 3036 -3050
5. Gadd, M. S. et al. Nat Chem Biol 2017, 13, 514-524
6. https://cresset-group.com/software/spark/
7. Shi, D. et al. Adv Drug Deliv Rev 2022, 180, 114079
8. https://cresset-group.com/about/news/cresset-collaborates-global-leaders-enamine-enable/