What is the role of heterobifunctional molecules in targeted protein degradation?
Heterobifunctional molecules, or degraders, are utilized widely in targeted protein degradation (TPD). These molecular systems normally consist of (1) a ligand targeting a ubiquitin ligase, (2) a ligand targeting a protein of interest, and (3) a linker to connect them. With this system, a protein of interest can be targeted for degradation and guided to the respective proteasomes. The heterobifunctional molecule is then recycled.1-2 Figure 1 depicts the protein degradation process with respect to the degrader.
Figure 1. Proteolysis cycle, including recycling of the heterobifunctional 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).
Advantages and challenges of heterobifunctional degraders
A recent resurgence in TPD research can be attributed to the several advantages over small molecule drug discovery. For one, many biological targets, particularly cancer targets, develop resistance to small molecule inhibitors or antagonists. The respective degraders, however, have exhibited potency against several inhibitor-resistant cancer strains.
Furthermore, due to the simplicity and size of small molecule drugs, they tend to frequently induce off-target toxicity, whether known or idiosyncratic. The advantage of heterobifunctional molecules is that their complexity increases their selectivity and limits their binding to known off-targets.
Small molecules also usually require high potency towards their biological target, and if they don’t contain an irreversibly binding warhead, can have low residence time in their target’s active site. Heterobifunctional degraders, however, can tolerate lower binding affinities and have exhibited sustained binding to their protein targets.
However, due to their size and complexity, their design can be quite challenging. For example, optimizing physical and chemical properties is a much more complex process than with small molecules. Another downside to their complexity is their associated resistance mechanisms. If a protein target develops resistance to a small molecule, it is usually a single-point mutation in the active site. Because these heterobifunctional molecules are much more complex, however, biological resistance mechanisms can be much more difficult to pinpoint. These challenges create a need for approaches that streamline the design and study of these heterobifunctional degraders.
Linker design: an illustrative example
To address the challenge of linker design, new databases of fragments have been created which can be used for bioisostere replacement or ligand joining experiments in Spark.3 These databases are derived from the PROTAC-DB library from the Tingjun Hou group4 and the ‘Stock’ and ‘MADE’ collections from Enamine Linkers for Linkerology.5 To illustrate their applications, this article will use the example of heterobifunctional molecule AK-2292 from Kaneshige, A. et al.6 This heterobifunctional molecule consists of a STAT5 binder, a CRBN binder, and small linker to join them (Figure 2).
Figure 2. Heterobifunctional molecule AK-2292 from Kaneshige, A. et al.6 The linker is highlighted in blue in both the 2D (left) and 3D (right) representations (PDB:7TVA).
In their study, the authors’ experiments tested a series of amide linkers of varying lengths and rigidity. Table 1 lists the linkers and their associated degraders’ half-maximal degradation concentration, DC50.
Table 1. Activity of the various tested linkers from Kaneshige, A. et al.6 The linker of AK-2292 (compound 30) is shown in blue.
| Linker | Compound ID | DC50 (STAT5, μM) SKNO1, 18 h | DC50 (STAT6, μM) SKNO1, 18 h | Rank in ‘Linkers’ databases | Rank in ChEMBL/eMolecules Screening databases |
 | 25 | > 5 | > 5 | 205 | Not found |
 | 26 | 0.23 | > 5 | 47 | Not found |
 | 27 | 0.50 | > 5 | 7 | Not found |
 | 28 | 1.1 | > 5 | 31 | Not found |
 | 29 | 1.4 | > 5 | 727 | Not found |
 | 30 | 0.19 | > 5 | 12 | Not found |
 | 31 | > 5 | N/A | Not found | 747 |
 | 32 | > 5 | N/A | Not found | 22 |
 | 33 | 4.2 | N/A | 11 | Not found |
 | 34 | > 5 | N/A | 227 | Not found |
The authors focused on linkers with a secondary or tertiary amide installed with a dehydrative coupling reaction on one end, and on the other, an sp-hybridized carbon to generate an internal alkyne, installed using a Sonogashira cross-coupling (one example consists of an amide/aniline combination). To keep the Spark experiments consistent with the study, we started with the heterobifunctional degrader molecule from the cocrystallized protein structure (PDB code 7TVA). Instead of a traditional scaffold hopping experiment, we treated the linker moiety as the scaffold to be replaced. In keeping with the preferences for chemistry, the atom type preferences were changed accordingly. Figure 3 highlights the experiment in Spark.
Figure 3. (Left) The scaffold hopping experiment window with AK-2292. The linker moiety is haloed, indicating that it is the moiety to be replaced with bioisosteric linkers. The selected atom types are highlighted in the top left corner of the window. Attachment point 1 (the amide) was set to be any nitrogen, while Attachment point 2 (the alkyne carbon) was set to be any nitrogen or an sp-hybridized carbon. (Right) The starter molecule AK-2292 in 3D with Cresset field points.7 The linker moiety is haloed, indicating that it is the moiety to be replaced with bioisosteric linkers.
Because this heterobifunctional molecule is much larger than the small molecules typically used in a Spark scaffold hopping experiment, the advanced settings for distance and angle allowance were also adjusted to allow for some movement of the unchanged portion. The RMSD Distance setting was changed to 2 Å (from 0.75 Å) and 45 degrees (from 15 degrees). The number of allowed rotatable bonds was also increased to 10 to include more flexible linkers.
The scaffold hopping experiment was first run using the newly created ‘Linkers’ databases. With these libraries, the top Spark bioisostere replacements included eight out of the ten reported linkers from the study, with seven being in the top 250 results (see Table 1). Even more promising is that the third most potent degrader, Compound ID 27, was ranked #7 out of the top 1000 results. These highlighted results are shown in Figure 4.
Figure 4. The reported linkers in the results of the Spark experiment using the new ‘Linkers’ databases. The Spark Tile Results dock shows the bioisostere replacement fragment and the result similarity with respect to the starter linker AK-2292. The activity of AK-2292 is in blue.
In fact, the only two linkers that were not found by Spark were the cyclic amides 31 and 32, both of which showed very low degradation activity. The three submicromolar results were found in the top 50 Spark results, indicating that not only can Spark find promising bioisosteres, but that it prioritizes linkers with desirable activity. Linker 29 has a negative BIF%, or Bioisostere Factor: positive BIF% values indicate favorable bioisosteres, while negative values reflect a replacement that can reproduce the geometry of the original molecule but is a poor mimic of the deleted moiety. This result is expectedly also one of the less potent linkers. The top 10 results from the Spark experiment are shown in Figure 5. These include linkers which are structurally very similar to those reported in Table 1, with varying chain lengths and rigidity of the amide. However, they also contain some rigid cyclic amides, which may decrease the potency of the overall molecule, although this remains to be experimentally confirmed.
Figure 5. The top 10 results from the Spark experiment, including one reported (27, Rank 7). The activity is shown.
These experimental results are very promising for linker design, as they maintain the desired shape and electrostatics of the starting linker, while also including small changes that could be promising (e.g., an ether or stereocenter). Figure 6 shows the top 5 results from the Spark experiment alongside the AK-2292 linker, with their associated positive and negative molecular electrostatic interaction potentials.7 We can see that despite some structural changes, the overall electrostatic potentials are very similar to those of the original linker. In other words, they are promising bioisosteres.
With this new information, the ‘Linkers’ databases could additionally be used in a Spark Ligand Joining experiment to explore empty space between the ligand targeting the protein of interest and the ligand targeting the E3 ligase.
Figure 6. AK-2292 and the top 5 results with their associated molecular electrostatic interaction potentials.7 The linker area is highlighted in orange.
To further confirm the usefulness of these new linker-focussed databases, we repeated the experiment using the already available Spark ‘VeryCommon’ and ‘Common’ databases (containing fragments derived from commercial compounds), as well as the ‘ChEMBL_common’ database (fragments derived from molecules in literature reports). The results from this search are far less promising, as only two of the ten reported linkers were found, even though more than 161K fragments were searched in the Commercial and ChEMBL databases, while only 9,672 fragments are available in the ‘Linkers’ databases. These results are also shown in Table 1.
The results of this second experiment confirm that there are fewer known degrader linkers in the existing Spark databases, as these typically include much smaller, rigid fragments unsuitable for linker design. On the contrary, the new ‘Linkers’ databases, specific for linkers of heterobifunctional molecule, are a very promising tool in the computational approach to linker design.
Conclusions
In this article, we have explored the application of the Spark scaffold hopping experiment to a heterobifunctional molecule. We have seen that the new linker libraries comprehensively include known and commercial degrader linkers as supported by experimental data and are therefore a promising tool for approaching linker design for heterobifunctional degraders.
References
- Li, F. et al. Nat Commun 2022, 13, 7133
- Xie, H. et al. Explor Target Antitumor Ther. 2021, 2(6), 511–521
- New Spark™ ‘Linkers’ databases to address the challenge of linker design
- http://cadd.zju.edu.cn/protacdb/
- https://enamine.net/building-blocks-mob/linkers-for-linkerology
- Kaneshige, A. et al. J Med Chem 2023, 66, 4, 2717–2743
- https://cresset-group.com/science/overview/