Following the huge success of the first event in 2021, the second BioSolveIT DrugSpace Symposium was held under the motto “It’s a small world”. With over 600 registrations from across the globe, the scientific community and fragment enthusiasts came together virtually to discuss the most recent advances in the field of fragment-based drug discovery (FBDD), small compound binders, and modifications of functional groups.
With the aim to expand the toolkit of medicinal and computational chemists, novelties in the field of chemical space exploration for FBDD were discussed. To profit from the possibilities lurking behind hundreds of billions of compounds, advancements for fragment growing, merging, and linking are required to efficiently mine from success-oriented sources.
This report summarizes the ten presented talks and potential trends of chemical space exploration utilizing FBDD methods.
University of Barcelona
“A Bottom-Up Approach to Screening Massive Virtual Collections”
Xavier Barril presented a case study of computational approaches to expand the druggable genome through exploration of vast chemical spaces. The exponentially growing make-on-demand chemical spaces bear the potential to transform drug discovery as they may allow to start with more advanced and more potent lead structures. In return, this would reduce the costs and the required time of modern and future drug discovery. The “bottom-up” approach bypasses the limitations of enumerated libraries through identification of the essential binding units, the fragments, and the subsequent selection of compounds containing those substructures from large chemical spaces. The previously undoable task to extract a matching subset of compounds from multibillion-sized chemical spaces was achieved using a new tool from Rarey’s lab, SpaceMACS, that Barril had early access to. To maximize the probability of finding actives, several increasingly rigorous computational methods were applied. In total, 85 compounds were purchased from Enamine from their REAL Space and delivered within 4-5 weeks with a delivery rate of 89%. Pharmacological assessment of the compounds confirmed several highly potent (KD 1-100 nM) and chemically diverse BRD4 binders with a hit rate of 7, 8, and 28% for fragments derived from drug candidates, experimental fragment hits, and computationally predicted fragment hits, respectively.
University of Dundee
“Amide-to-Ester Substitution as a Strategy for Optimising PROTAC Permeability and Cellular Activity”
Adam Bond delivered a talk on the pharmacological and -kinetic effects of amide-to-ester replacements at the pan-selective BET degrader MZ1. The bioisosteric replacement resulted in improved protein degradation through increased permeability of the ester-containing compounds — likely due to masking of solvent-exposed hydrogen bond donors. The ester derivative OMZ1 was 1.5-fold more potent at degrading BRD4 with 10-fold improved permeability by PAMPA than MZ1 but displayed 3-fold weaker binary and ternary affinities to VHL ± BRD4BD2. Furthermore, it formed a less cooperative, ternary complex between VHL and BRD4BD2. Similar trends were also observed for the amide/ester relatives ARV-771 and OARV-771, as well as AB2 and OAB2. In conclusion, degradation of the investigated PROTACs was driven by their ability to enter the cell as the compounds displayed excellent plasma stability.
“Zetta-Sized Chemical Space Navigation”
Gastreich showcased current BioSolveIT developments and innovations-in-the-making. The call for larger and accessible compound collections to explore the chemical space as a whole requires computational solutions beyond the limits of enumeration that currently range in the 1011 to 1012 region.
Starting with an introduction on the combinatorial generation of a chemical spaces based on chemical reaction rules and building blocks (synthons), results of a chemical space fragment growing campaign at PKA were presented for the first time. Within nine weeks time, a nanomolar binder was discovered after a single iteration cycle including the elucidation of its binding mode through X-ray crystallography.
Two new ReCore indices for 3D-scaffold-hopping and pharmacophore replacement were disclosed: the first one based on Peter Ertl’s Magic Rings featuring 15 million 3D-fragments, and the second one resulting from an ongoing collaboration with the Cambridge Crystallographic Data Center (CCDC), covering 49 million structures from high quality, experimental data.
The next addition to computational FBDD methods was announced: FastGrow, a tool to explore binding cavities starting from a ligand-target complex, will soon become a part of SeeSAR’s Inspirator Mode. FastGrow resulted from a collaboration between Servier in Paris, the Rarey Lab in Hamburg, and BioSolveIT.
Finally, infiniSee 4 “Pandora” was introduced, now featuring four make-on-demand chemical spaces (GalaXi by WuXi LabNetwork, REAL Space by Enamine, CHEMriya by OTAVA, and Freedom Space by Chemspace) resulting in almost 50 billion purchasable compounds. The Tanimoto-like fingerprint exploration tool SpaceLight and the substructure searching tool SpaceMACS were announced as currently being implemented in infiniSee.
University of Strasbourg
“Analysis of Fragment Libraries”
Presenting a recent publication from her group, Esther Kellenberger elaborated on the potential behind FBDD through systematic analyses. Kellenberger discussed the challenges of designing an ideal fragment library as it requires a balancing act between low complexity and versatility, as well as chemical diversity and the number of entries. The key objectives of their study were (i) assessment of the commercial libraries for their type and size, (ii) quality of the fragments, (iii) frequency of scaffolds, and (iv) coverage of the chemical space. The aim was to visually represent the results in an easy-to-understand 2D chemical map. 81 different libraries from 14 different suppliers were investigated. Those libraries amounted to a total of 754,646 commercially available fragments with 512,284 individual molecules after removal of duplicates. The smaller libraries (<2000 fragments, around half of the investigated libraries) contained only a fraction of the whole set, namely about 33,000 entries. Around 90% of the 512,284 individual molecules had molecular weight in the range of 200 to 300Da, around 50% displayed weakly hydrophobic properties (logP between 0 and 2).
Topographic mapping of fragments was performed for seven types of libraries (general, chelatant, diverse, natural products, 3D, sp3-rich, miscellaneous) covering 433,433 entries and 59,270 scaffolds.27,28 61.7% (36,555) of scaffolds were unique among the fragments, 0.1% (46) were found in over 1000 entries, and 38.2% (22,669) were present in 2 to 999 fragments. The generated maps provided consistent grouping of similar chemical features allowing comparison of chemical space coverage of several libraries.
“Application of FastGrow to the Design and Synthesis of Drug Like Compounds”
Jeremy Edmunds introduced the concepts of medicinal chemistry in modern drug discovery campaign with a focus on appropriate specificity achieved by computational methods. Working simultaneously on several compound series, 3,046 compounds were assessed for the study of PKCθ inhibitors.
With the aim to answer the question of medicinal chemists, what to make next, and how to reduce the number of compounds to be synthesized, a computational workflow was established:
Starting from a seed compound in a structure-based approach to address unoccupied binding cavities, an early version of FastGrow by the Rarey Lab (see above) was used to systemtically explore all applicable exit vectors of a template molecule with building blocks originating from in-house knowledge to create a library of possible evolved compounds. Subsequently, after database generation, a chemical space of 3.6 million fragments was investigated with FastGrow within only 2.3 hours (limit 10,000 compounds) and 3.7 hours (limit 100,000 compounds) on Amazon Web Services (Intel® Xeon® Platinum 8275CL CPU @ 3.00GHz; Thread(s) per core: 2, Core(s) per socket: 8). 23% and 19% percent of the suggested compounds (for 10,000 and 100,000 results, respectively) displayed a predicted affinity <10nM with preferred torsions.
Further Edmunds pointed out that compounds should be synthesized based on what is needed instead of what is the simplest to make to avoid bias by medicinal chemistry experience.
In a current project FastGrow was successfully used to complete several weeks of standard CADD procedures within hours. Strikingly, all four compounds selected for synthesis suggested by FastGrow exhibited activity at the target with KD values of 0.002, 0.05, 0.2, and 7 µM, respectively.
“A Step toward NRF2‐DNA Interaction Inhibitors by Fragment‐Based NMR Methods”
Sven Brüschweiler presented their recent work in collaboration with Boehringer Ingelheim on the characterization of ligand-target interactions through utilization of NMR methods in early stages of drug discovery. After computational assessment of intrinsically disordered parts of the protein, 1,800 fragment molecules were screened against the Neh1-ΔLZIP domain of NRF2. 1H-15N HSQC was used to observe shifts of target amide groups, including those of asparagine and glutamine, identifying two hits based on chemical shift perturbations. infiniSee was used to search for analogs of the phenyl carboxylic acid scaffold in Enamine’s REAL Space in a “SAR-by-catalog” approach.
35 compounds were discovered; 27 came from the Boehringer Ingelheim and Sigma Aldrich libraries, and eight were purchased from Enamine. The set allowed SAR elucidation of the compound series.
1H-15N HSQC was subsequently used to identify the ligand binding site and a crystal structure was resolved. Second step data-driven docking calculations resulted in three docking clusters complemented by chemical shift changes for the complex structures indicating exchanging conformations due to weak binding.
University of Toulouse
“Crystallographic Screening in FBDD: Application to the Phosphopantetheinyl Transferase of Mycobacterium Abscessus”
Laurent Maveyraud gave an update on the screening campaign of the so-called Plateforme Intégrée de Criblage de Toulouse (Integrated screening plateform of Toulouse, PICT) to discover possible novel inhibitor scaffolds for the bacterial PPTase of Mycobacterium abcessus — a pathogen that often affects humans with cystic fibrosis. The low success rate of antibacterial intervention gave the bacterium the nickname “antibiotic nightmare”.
11,390 drug-like molecules and 939 fragments with a comparatively low molecular weight of 190Da. Initial results of differential scanning fluorimetry were additionally assessed by high throughput co-crystallization with a 91% success rate. 855 crystals were processed and finally, 684 structures were obtained. Assessment of structures resulted in 71 positive hits at two distant binding sites of the PPTase; one close to the substrate CoA binding site and the second denoted as a surface binding site at the interface between two neighboring protein in the crystal therefore unlikely to exist in vivo. Properties of the substrate binding sites were characterized for subsequent fragment evolution with emphasis of mechanism of action as no measurement of enzymatic inhibition or binding affinity was possible. Interestingly, the binding site rotamers were conserved among the elucidated structures including highly flexible side chains while in some exceptions His90 and Glu153 interacted with the co-crystallized fragments. Electron density of the endogenous ligand was decreased in those cases indicating a displacement of the ligand and the subsequent inhibition of the target.
“Chemical Space Docking: Large-Scale Structure-Based Virtual Screening for ROCK1 Inhibitors”
Paul Beroza outlined Chemical Space Docking (CSD) as a way to access large chemical spaces using structural information from a target. Beroza stated that the game- changing influence of Enamine on drug discovery through reliable reaction schemes and compatible building blocks allowed mining from a combinatorially exploding set of billions of molecules. This, alongside other bigger spaces beared the potential for more and better molecules as the hit rate improved with higher ranking docking scores. Yet, the brute force handling of extremely large chemical spaces through enumeration is coupled to high computational data storage requirements: The enumeration of the GSK XXL, the largest proprietary chemical space to date that had been created by GlaxoSmithKline with BioSolveIT’s CoLibri suite, exhibiting a breathtaking number of 1026 entries, would require 4 x 1017GB of data for storage. Therefore, the only way out is the directed, reaction-based generation of molecules during the search.
Setting it up that any member of the space may emerge as a hit a priori, CSD picks up on this approach and includes protein structural information to increase the chemical diversity and to include data from the protein binding site to discover compatible compounds. Starting from docking of single fragments, products are sequentially grown40 with building blocks and defined chemical reactions from Enamine’s REAL Space. Additionally, the synthon approach allows docking of products from best ranking synthons only, thus allowing a more efficient chemical space exploration.
In a case study, ROCK1 was selected as the target due to the known hinge interaction motif of ligands. Following a computational workflow including the CSD, 69 compounds were purchased and pharmacologically assessed. Among the compound classes (pyrazoles, lactams, pyridines, azaindoles, indazoles), 7 molecules displayed inhibitory activity in the range of 10 – 100 nM, with an overall hit rate of 40%. The structure of the most potent compound (Ki = 38 nM) was resolved (PDB-ID: 7S25) which was pretty similar to a compound class discovered in parallel by another group.
Comparison with the only other method to date that pursues a similar approach clearly highlights that CSD is an efficient method of compound screening. Beroza corroborated this thesis with very high hit rates and docking score distributions that clearly demonstrated that CSD-based docking scores have a pronounced, higher count at better scores, whereas classical dockings are distributed across all score ranges (Gaussian-shaped): A subset of molecules was fully enumerated, docked, and the results were compared to CSD. In his quantification example, over 95% of the classical, enumerated results displayed very poor poses while the CSD selection accumulated at higher docking scores.
“Magic Rings: Navigation in the Ring Chemical Space Guided by the Bioactive Rings”
Rings and scaffolds are likely the most important part of bioactive molecules since they define the shape of the ligand and orient the substituents into the right positions for interactions with the target. 98% of compounds from the ChEMBL database contain at least one ring system. Peter Ertl used this as a motivation for his study to identify those and analyze them for their frequency and target structures. Using a set of 42 simple topological descriptors, data from the ChEMBL and ZINC database was analyzed revealing minor differences between the distribution of bioactive and common synthetic rings. For bioactive systems fused aromatic rings often containing a nitrogen were reported, while synthetic systems on the other hand featured aliphatic rings and spiro centers. The bioactive rings were classified into eight different categories depending on their preferred target: GPCRs, kinases, proteases, other enzymes, nuclear receptors, ion channels, epigenetic, multiple, and other, totaling in fourty thousand individual rings. An interactive, visual map was designed, capturing the size and complexity of the investigated ring systems. Clusters of bioactive regions within the chemical space were observed.
The data set can further be used for various approaches including diversification of molecule collections, virtual screenings, library designs, and many more. A ReCore index of the bioactive ring systems featuring 15 million entries with two exit vectors each is available for SeeSAR.
The bioactive rings application can be accessed through Peter Ertl’s website at no cost.
Helmholtz Institute for Pharmaceutical Research Saarland (HIPS)
“Exploiting Fragment-Based Design in Anti-Infective Drug Discovery”
Anna Hirsch reported three case studies featuring fragment-based and computer-aided drug design for novel antibiotics to combat the antimicrobial resistance crisis
For the exploration of binding possibilities at a target of interest, two methods were presented: First, the dynamic combinatorial chemistry (DCC) approach where the reaction between initial building blocks is reversible and the best binders are selected for amplification. Second, the kinetic target-guided synthesis (KTGS) where reaction partners bind to adjacent pockets and are irreversibly connected in their proper orientation.
Applying multiple hit identification strategies, a virtual screening campaign at an allosteric pocket of ECF-T was performed. 1.3 million compounds from the Express Collection of Princeton BioMolecular Research were docked, 12 compounds were purchased of which six were sufficiently soluble. Two decent fragment hits were discovered. Subsequently the in-house HIPS small synthetic molecules library was screened resulting in five additional hits. SARs were established for two scaffolds with good oral bioavailability. Interestingly, an impurity in another scaffold was found to be the active compound in the screening assay.
The second case report on DnaN featured DCC- and KTGS-guided discovery of peptide-like structures. In another campaign on DnaN, five hits were retrieved by a modified three-step biophysical screening cascade of a fragment library. SARs of the most potent hit were further investigated by fragment growing and lead optimization resulting in improved binding affinity and antimycobacterial activity.
The third and last report was focused on LasB as target. FRET-based screening of 1192 compounds provided a fragment hit compound which was optimized resulting in a free thiol that interacts with the zinc cation in the active site. The following medicinal chemistry optimization led to the class of 3-mercaptosuccinimides to rigidify the scaffold. The merging/linking with a second fragment finally resulted in a subnanomolar class of phosphate-containing inhibitors with high selectivity.
Several decades ago, fragments were not perceived as powerful starting points by the scientific community. With six drugs approved by the FDA originating from FBDD, it has nowadays reached a leading status as a successful method with great potential to keep on delivering.
The methods and case studies presented in the ten talks of the DrugSpace Symposium did highlight that FBDD can be taken to another level by combining basic methods with the possibility to mine from vast chemical spaces to efficiently obtain potent drug candidates by guiding the process into the right directions and focusing on important parts of the subspaces.
Larger chemical spaces bear uncounted possibilities which can be exploited using fragments and small molecules as a starting point.
The report is also available as a PDF version.