Status of De Novo Drug Design (I)
When it comes to R&D of new drug, people would think about high investment, long term, and high risk. Although drug design and screening technology has been improving, the disease is complex. Even if the drug was successfully marketed, it will still face many unknown factors such as adverse reactions, market sales, and national policies. In fact, it is not impossible to solve these problems, which is to design a few good enough drug candidates.
Does it easy to find preferable candidates?
With the rapid development of X-ray diffraction, NMR, electron microscopy and other technologies, the structural parameters of the target protein are gradually clear. On the other hand, the significant increase in computing performance and the reduction in hardware prices have made computer-aided drug design tools an indispensable helper for scientists. Various computer algorithms can now perform virtual screening or directly give new molecular structures, and help scientists optimize the medicinal properties and physical and chemical properties of seedling compounds. So, can a computer now automatically design a drug molecule from scratch by giving it a chemical structure like artificial intelligence?
In the field of drug design, computer programs have assisted researchers in developing some drugs in the last century. The most common method is to use a virtual screening technique such as molecular docking to target a protein with a known structure in a massive chemical structure to find the hit compound. In addition, there are some drug design programs based on the chemical structure of small molecules, such as de novo design programs. Recently, Dr. Gisbert Schneider of the Federal Institute of Technology in Zurich (ETH Zürich) and others summarized the literature of automatic de novo drug design in recent years and detailed 10 drug target cases so that readers can get an overview of the overall situation or understand the frontier direction. Related papers were published in Angew. Chem. Int. Ed.
- β secretase inhibitor
BACE-1 is one of the targets of neurodegenerative diseases, but its patentability is quite challenging. Fishwick et al. reported a de novo drug design program SPROUT based on protein structure, and designed a non-polypeptide structure β-secretase BACE-1 inhibitor (J. Med. Chem., 2013, 56, 1843). Through the X-ray structure of the complex of BACE-1 and peptide-like ligands, SPROUT resolved key active sites and designed the original compound 1. After synthesizing Compound 1 and testing its biological activity, it was found that its activity is weak (IC50 = 323 μM). After structural optimization, compound 2 was obtained with moderate activity (IC50 = 27 μM). This case confirmed the feasibility of synthesizable compounds acting on targets based on automatic structural design.
- Aurora A kinase A inhibitor
VX-680 is a pan-Aurora kinase inhibitor with high anti-Aurora activity (Ki = 0.6 nM) and was terminated in clinical trials due to heart problems with prolonged QT intervals. Rodrigues et al. hoped to design a brand new chemical framework (Chem. Sci., 2013, 4, 1229) with the ligand-based de novo drug design program DOGS (Design Of Genuine Structures), which has similar activities as VX-680, but avoid heart problems while breaking through the intellectual property restrictions surrounding this structure. The DOGS program is based on 88 skeletons and produces 172 chemical structures. Since the aromatic sulfonamide structural fragments are often scored high in the scoring system, and the chemical synthesis is highly feasible, the corresponding compounds 3 and 4 were selected for synthesis. Another advantage of DOGS is that it can design synthetic routes, and design algorithms are derived from chemical transformation rules. In vitro experiments found that 100 μM 3 only inhibited 5% of Aurora A. The activity of 4 is IC50 = 10 μM, which is suitable for optimization from seedling compounds to lead compounds. The huge difference in activity between 3 and 4 is likely to come from imidazole structural fragments. Under this guidance, they obtained compound 5 by optimization (IC50 = 2.7 μM).
From this case, the following three points can be seen: (1) The structure of compound 5 is completely different from the original VX-680, and it is a completely new structure. (2) The program only needs one inhibitor structure to start the program, not even the protein structure, so this method has a wide range of applications. (3) Most of the designed compounds can be synthesized and used to test the activity.
Starting from the crystal structure of the complex of Aurora A and inhibitors, Park et al. used a structure-based virtual screening to find a key fragment structure MPPA in the fragment database. Their secret is to automatically dock and modify the scoring function (J. Chem. Inf. Model., 2018, 58, 700). Using the hinge-binding site as the starting point, the team used LigBuilder to focus on the fragment structure database of Aurora A. Designed new compounds using de novo drug design methods, followed the 5 rules of drug-like drugs in drug design, and finally selected 35 compounds and synthesized 17, the most active compound is 6 (IC50 = 0.012 nM).
- COX-2 and LTA4H dual inhibitor
Multiple pharmacology has also received attention in recent years, that is, a compound achieves a therapeutic effect by acting on more than one target. Shang et al. used a de novo drug design program LigBuilder to design compounds that act on cyclooxygenase-2 (COX-2)/5-lipid oxidase (5-LOX)/leukotriene A4 hydrolase (LTA4H) in order to achieve more Good anti-inflammatory effect (J. Chem. Inf. Model., 2014, 54, 1235). The authors extracted the fragment structure from the known inhibitors of these three targets and obtained 21 compounds, and then tested the activity at a high concentration (1 mM) to obtain 9 seedling compounds. The LigBuilder program performs a variety of binding mode calculations on these structural fragments, designed 1.1 million compounds, selected the first 1,000 of them to continue research, and finally synthesized 6 compounds, of which compound 7 has the best activity, and COX at a concentration of 100 μM The inhibition rate of -1 and COX-2 was 90%, and the inhibition rate of LTA4H was 43%, thus entering the second round of de novo drug design to obtain compound 8, and the third round to obtain compound 9. Compared with COX1, compound 9 has a very good selectivity for COX2 (about 11 times).
- Helicobacter pylori HtrA inhibitor
Helicobacter pylori can cause stomach/duodenal ulcers and even gastric cancer. The serine protease pathogenic factor HtrA (high temperature reguirement A) is the target of Helicobacter pylori and other pathogenic bacteria. The compound currently available is rhodanine derivative 10, but it has weak activity, poor solubility, and the rhodanine skeleton problem.
Perna et al. used DOGS software to start at 10 (Angew. Chem. Int. Ed., 2015, 54, 10244), designed 1707 compounds, screened by solubility combined with three-dimensional structure, and selected 65 of them for synthesis , Compound 11 best meets the following 4 requirements: solubility and fit rank top; no Michael addition acceptor structure; very different from the structure of Compound 10; a completely new type of chemical structure.
11 can be synthesized through a two-step reaction. Its affinity is similar to compound 10 (KD = 37 μM), but the ligand efficiency LE (ligand efficiency) is better. In the configuration relationship study, its analog 12 had the best activity (KD = 37 μM, LE = 0.3), and all had anti-Helicobacter pylori activity in vitro.
To be continued in Part Two…