The development of macrocycles has received increasing interest in drug discovery as a rational approach for restricting the conformation of small molecules, which results in more energetically favoured binding to the target (reduced entropy and increased binding free energy). When compared to their acyclic counterparts, macrocycles could offer significant improvements in potency, selectivity, as well as pharmacokinetic and pharmacodynamic properties. Despite their drug likeness benefits, the synthetic challenges associated with macrocycles design and synthesis make them higher risk strategies.1 To mitigate the risk associated with high investment in synthetic resources, higher scrutiny at the planning phase of the process is essential to prioritize ligand designs with higher chance of succeeding.2
The recent FDA approvals of macrocycles inhibitors such as lorlatinib (LORBRENA®) and pacritinib (VONJO®) validated the macrocyclization as a drug discovery strategy against kinases. The latter is an orally bioavailable macrocyclic inhibitor of Janus Kinase 2 (JAK2) and has shown efficacy for the treatment of myelofibrosis.3-4
To overcome the synthetic strategy drawbacks and assess scaffold diversification at an earlier stage in the macrocyclization process, in a recent paper, we present a prioritization workflow using computational aided drug design (CADD) approaches for the selection of macrocyclic designs for synthesis.5 We applied Electrostatic Complementarity (EC),6 3D-Field QSAR, and Free Energy Perturbation (FEP) methods,7 all available in drug design solution Flare™,8 for the profiling of a dataset of macrocyclic JAK2 inhibitor analogues of pacritinib.
The high structural resemblance of this series of compounds suggests a similar binding mode to the one recently determined for pacritinib bound to JAK2 (PDB: 8BPV)4. Because of that, all ligands were aligned using a maximum common substructure algorithm to the bioactive conformation of pacritinib, with the JAK2 as an excluded volume. The experimentally measured pIC50 values were then compared with the calculated EC scores which quantifies how well the electrostatic of the ligand fits the protein binding site (as exemplified in Figure 1). The statistic correlations showed EC as a good method to determine if a compound was unlikely to be active since compounds with an EC score < 0.29 also have a pIC50 < 7.0. The method’s ability to quickly deprioritize ligands with low scores can dramatically reduce synthetic efforts that would otherwise be associated with preparation of challenging compounds.
Furthermore, a 3D-Field QSAR model and relative Free Energy Perturbation calculations were performed to generate predictive pIC50 values to rank a ligand’s potency. Sixty-nine pacritinib analogues, partitioned into training set (49 ligands) and test set (20 ligands), were used to build a highly confident 3D-Field QSAR model (q2 cross-validated training set = 0.47 and r2 test set = 0.52). Despite many predicted activities being within 0.5 log units of the experimental value, there were cases at the extremities of the activity range where predictions differ as high as 1 log unit. In addition, the coefficients generated by the Field model (refer to Figure 2) provide key insights about the regions one might optimize through modeling to improve potency.
To gain further insights, a Free Energy Perturbation benchmark study, using the test set 20 compounds, was carried out. As seen in Figure 3, most predicted activities are within 0.5 log units of the experimental value, and the Kendall’s Tau and Pearson’s correlation values highlights the robustness and high accuracy of the method.
Interestingly, the use of a similar binding mode to the bioactive conformation of pacritinib significantly underpredicts the most active compound (23). Hence, various low-energy conformations generated by quantum mechanics calculations were tested to identify the one that accurately predicts the potency of compound 23. This showcases the importance of understanding the 3D conformation and flexibility of a ligand with these effects having a significant impact on the accuracy of predictions (refer to Figure 4).
Table 1. Experimental and computationally predicted pIC50 values of the JAK2 macrocyclic inhibitors pacritinib 4 and 23
| Compound | Exp. pIC50 | 3D-QSAR pIC50 | FEP pIC50 |
|---|---|---|---|
| 4 | 7.6 | 7.6 | 7.2 |
| 23 | 8.2 | 7.5 | 8.1 |
We demonstrate that these three CADD approaches provide valuable insights: while the computationally inexpensive methods Electrostatic Complementarity and 3D-Field QSAR can make predictions that quickly aid in prioritization of ligands which might lower the synthetic effort; the greatest accuracy in the prediction of binding affinity is only attained by Free Energy Perturbation calculations. Of note, these methods are sensitive to the preparation of the protein system and of the ligands being assessed.8 In summary, this flexible streamlined workflow guides and prioritizes the design of novel therapeutic molecules, helps organizations de-risk projects, shortens development timelines, and increases the probability of success.
For more details of the methods used, please refer to our original publication5:
ACS Med. Chem. Lett. 2025, 16, 6, 1066–1072.
References
- Amrhein, J. A.; Knapp, S.; Hanke, T. Synthetic Opportunities and Challenges for Macrocyclic Kinase Inhibitors. J. Med. Chem. 2021, 64, 7991-8009.
- Diao, Y.; Liu, D.; Ge, H.; Zhang, R.; Jiang, K.; Bao, R.; Zhu, X.; Bi, H.; Liao, W.; Chen.; Z.; Zhang, K.; Wang, R.; Zhu, L.; Zhao, Z.; Hu, Q.; Li, H. Macrocyclization of Linear Molecules by Deep Learning to Facilitate Macrocyclic Drug Candidates Discovery. Nat. Commun. 2023, 14, 4552.
- Lamb, Y. N. Pacritinib: First Approval. Drugs 2022, 82, 831-838.
- Miao, Y.; Virtanen, A.; Zmajkovic, J.; Hilpert, M.; Skoda, R. C.; Silvennoinen, O.; Haikarainen, T. Functional and Structural Characterization of Clinical-stage Janus Kinase 2 Inhbitiors Identifies Determinants for Drug Selectivity. J. Med. Chem., 2024, 67, 10012-10024.
- Braz, N.F.; Slater, M.J.; Lang, S. Application of Free Energy Perturbation (FEP) Methodology for Predicting the Binding Affinity of Macrocyclic JAK2 Inhibitor Analogues of Pacritinib. ACS Med. Chem. Lett. 2025, 16, 1066-1072.
- Bauer, M. R.; Mackey, M. D. Electrostatic Complementarity as a Fast and Effective Tool to Optimize Binding and Selectivity of Protein-Ligand Complexes. J. Med. Chem. 2019, 62, 3036-3050.
- Kuhn, M.; Firth-Clark, S.; Tosco, P.; Mey, A. S. J. S.; Mackey, M.; Michel, J. Assessment of Binding Affinity via Alchemical Free-Energy Calculations. J. Chem. Inf. Model. 2020, 60, 3120-2130.
- Flare™, version 10, Cresset®, Litlington, Cambridgeshire, UK; https://cresset-group.com/flare/; Cheeseright T.; Mackey, M.; Rose, S.; Vinter, A. Molecular Field Extrema as Descriptors of Biological Activity: Definition and Validation. J. Chem. Inf. Model. 2006, 46, 665-676.
