Computational Fragment-Based Drug Design: Current Trends, Strategies, and Applications

Abstract

Fragment-based drug design (FBDD) has become an effective methodology for drug development for decades. Successful applications of this strategy brought both opportunities and challenges to the field of Pharmaceutical Science. Recent progress in the computational fragment-based drug design provide an additional approach for future research in a time- and labor-efficient manner. Combining multiple in silico methodologies, computational FBDD possesses flexibilities on fragment library selection, protein model generation, and fragments/compounds docking mode prediction. These characteristics provide computational FBDD superiority in designing novel and potential compounds for a certain target. The purpose of this review is to discuss the latest advances, ranging from commonly used strategies to novel concepts and technologies in computational fragment-based drug design. Particularly, in this review, specifications and advantages are compared between experimental and computational FBDD, and additionally, limitations and future prospective are discussed and emphasized.

INTRODUCTION

The drug discovery processes currently rely heavily on large-scale screening between compound libraries and different biological targets (1). The high-throughput screening (HTS) technology underwent dramatic improvements within recent years. Besides successful applications of HTS for drug development, limitations do exist with this approach. The HTS can suffer from (1) a low hit rate and (2) a limited coverage of drug-like chemical space. Experimental fragment-based drug design and fragment-based screening (FBS) become complementary choices for lead identification and drug development (2). It is generally believed that a chemical space of 10⁹ diversified molecules can be sampled with 10³ fragments (3). The coverage of chemical space can be improved with fragment-based drug design. Meanwhile, it is easier for conducting chemical modification with fragment hits. The selected fragments in a library usually follow the criteria of “rule of three” (4) (M.W. ≤300, hydrogen bond donor and acceptor ≤ 3, CLogP ≤3, NROT ≤3, and PSA ≤60). Questions commonly asked towards the application of rule of three include (1) whether we should use three-dimensional fragments instead of flat structures, (2) whether or not there are significant differences in fragment-binding sites among different protein-binding sites, (3) what are the specific criteria for determining the fragment hits given the low affinity of fragment-protein interaction, and (4) what are strategies and approaches to select and develop fragment hits fulfilled the rule of three into leads. Although discussions and argues do exist about the application of these rules (5,6), “rule of three” remains the preferred reference for fragments selection. Rule of three assists the selection of hits with the suitable size and physical-chemical properties and limits the complexity of molecules in the fragment libraries.

Experimental fragment-based drug design and FBS provided fruitful results in the past decades. Abbott, one example of a pharmaceutical company, and SGX Pharmaceuticals, one example of a biotechnology company, provided successful practices in developing fragments into clinical candidates (7). Other examples include fragment-based screening by X-ray crystallography to β secretase (8), development of FtsZ inhibitor with anti-staphylococcal activity (9), and the discovery of phosphodiester-ase inhibitors through crystallography and scaffold-based drug design (10). However, challenges and limitations for current experimental fragment-based drug design can inevitably restrict the application of this approach (3,11). First, large-scale pure, high-quality, and stable target proteins are required for the screening. The crystallization of certain kinds of proteins, like GPCRs, is challenging. The processes can be expensive and time-consuming. Also, the approach can be difficult when applied to targets without a known structure. Second, given the relatively small sizes of fragments, limited receptor-ligand interactions can be formed between an individual fragment and the surrounding residues and only a part of them is strong enough for detection. Only soluble fragments with relatively high solubility can be suitable for fragments screening. This requirement significantly narrows down appropriate fragments.

Given the restrictions of experimental fragment-based drug design, computational approaches can provide prospective solutions to the drug discovery processes in the future. With the dramatic progress in the field of computation technologies, calculation power has undergone an exponential increase during the last decades (12). Computational fragment-based drug design becomes an efficient and time- money-, and labor-saving approach to facilitate the traditional fragment-based drug design processes. Computational methods can easily establish an integrated fragment library either by filtering existing libraries with certain criteria or by breaking down a variety of compounds (13,14). With attention to the physical-chemical properties and chemical synthetic feasibility, only appropriate and desirable fragments can be included. The diversified fragment libraries, with an emphasis on both quantity and quality generated through computational methods, have significant advantages when compared with traditional fragment libraries. Computational fragment-based drug design can use molecular docking technology for performing the virtual screening. Molecular docking puts a small molecule into a binding pocket of a macromolecule, and the corresponding algorithm will evaluate the receptor-ligand affinities and interactions (15). There is a range of docking algorithms currently available for different types of works (15–18). Examples of current widely used docking programs and scoring functions are Glide (19), GOLD (20,21), Surflex-Dock (22) etc. The fragment-involved docking studies draw increasing attention since the continued developments in ligand efficiency matrices become available. The virtual screening can evaluate a massive amount of fragment-receptor interactions in a time-efficient manner, regardless of the solubility of an individual fragment. In case no crystal structure for the target is available, homology modeling can be used for the model generation (23). Homology modeling can build a target macromolecule based on the structure of the template protein. The template can be determined by the sequence and biological homology (24). Examples of commonly used programs are Modeler (25) and SWISS-MODEL (26). Moreover, computational fragment-based drug design can flexibly adopt different strategies (commonly used strategies will be discussed in the following sections) and introduce various linkers with self-defined criteria and rules, which is critical for achieving diversified results (27). As a fast-growing area of research, increasing attention is focused on the fragment-based drug design. There is a clear trend that the number of publications with this concept has increased dramatically during the last several decades when compared to the total number of publications on general drug design (Fig. 1). The number of publications on FBDD increased roughly 8 times from ~ 35 in 2001 to ~ 280 in 2017, while the total number of publications on general drug design increased around 2.3 times from ~ 3500 in 2001 to ~ 8000 in 2017.

Fig. 1.

The number of publications each year from 2001 to 2017 that include the concepts of “fragment-based drug design” (blue bars) and simply “drug design” (orange dots and line) through the literature search on SciFinder (scifinder.cas.org)

COMMON STRATEGIES IN FRAGMENT-BASED DRUG DESIGN

Fragment binding can reveal hot spots on proteins, which can result in high-affinity receptor-ligand interactions. The fragment-based approach is especially suitable for detecting the interaction spots among the protein-ligand binding, as fragments can interact with a certain region of the target protein. Once the identification of fragments inside the binding pocket is done, these fragments can be grown, linked, or merged to develop the potential ligands (Fig. 2).

Fig. 2.

Commonly used strategies for fragment-based drug design. a Growing. After identification of one suitable fragment inside the binding pocket, substitutions can be added to the identified fragment. The growing process will increase the lead likability of the original fragment to enhance the receptor-ligand interactions. b Linking. Multiple fragments can be identified for one binding pocket simultaneously targeting at different regions. Linkers will be introduced to connect separated fragments to increase the lead likability and to create a novel compound with the potential affinity towards the pocket. c Merging. A known lead can partially occupy the binding pocket. One or more fragment(s) can be identified to be suitable for the remaining space. One or more linker(s) can be introduced to connect the known lead and the fragment(s) to increase the strength of receptor-ligand interactions

Fragment Growing

With a given binding pocket, one suitable fragment can be identified to have preferable interactions with a certain sub-region inside the binding pocket. Multiple substitutions can be added to the identified fragment to grow it up. The combination of substitutions and the identified fragment usually follows the chemical combinatorial rules to guarantee the generation of synthetic feasible compounds. The determination of candidate substitutions can be adjusted according to (1) the shape, size, and physical-chemical properties of the remaining part of the binding pocket or (2) the pharmacophore features revealed from reported/known active compounds. The growing process will generally increase the druglikeness of the original fragment. The newly generated compounds are supposed to have enhanced receptor-ligand interactions.

Fragment Linking

Multiple fragments can be identified to occupy different sub-regions simultaneously inside a given binding pocket for a certain target. Different sub-regions may have distinctive structural features. The arrangements of several residues can create hydrophobic or hydrophilic environments among different sub-regions. Favored fragments among these sub-regions can have various physical-chemical properties. Multiple fragments can be identified to interact with the given binding pocket, with each fragment occupying a certain sub-region. Fragment linkers can be introduced to connect separate fragments to increase the druglikeness. The connection of multiple fragments usually follows the synthetic rules to generate synthetically reasonable novel compounds. The newly generated compounds are supposed to have both good affinity and selectivity to the binding pocket.

Fragment Merging

Usually, there are reported/known lead compounds for a given target. The known compounds may function as probes, which are sufficient for the fundamental studies with different purposes. The known lead may occupy a big portion of the binding pocket, and the remaining space can be suitable for fragment binding. Similar to fragment linking, linkers could be introduced to connect the known lead and the suitable fragment(s). The suitable fragments are selected with a consideration of (1) achieving preferable physical-chemical properties of the final product, (2) forming relatively strong receptor-ligand interactions, and (3) being synthetically reasonable for adding the linker(s). Fragment merging is not only used in novel compound design but also remains a tool for chemical modification and derivative generation.

Overview of Computational Fragment-Based Drug Design

The computational fragment-based drug design is a fast-growing area of research with new concepts and ideas coming through. The detailed discussion on the latest technologies and approaches will be focused on the following sections. The basic ideas remain the same. The flowchart for a typical computational fragment-based drug design involves five major steps (Fig. 3). First, the establishment of the diversified fragment library. Considering the advancement in computational power, a very large number of fragments can be screened in each experiment. This guarantees a good degree of fragment diversity and increases the statistical probability of finding suitable fragments for a given pocket. Second, virtual screening of the fragment library in order to find suitable fragments. Fragments are relatively small compounds, which will not likely give appealing binding energies or docking scores during the docking process. However, by analysis of the potential receptor-ligand interactions, like H-bond or hydrophobic interactions, suitable fragments can be identified. Third, design the lead compounds based on identified fragments. As mentioned above, fragment growing, fragment linking, and lead-fragments merging are strategies that can be used in this step. Fourth, the verification of generated novel leads with corresponding biological assays. Computational validation methods will be applied at the beginning to filter out false positives or false negatives. Corresponding biological assays will be followed to formally verify the effectiveness of these lead compounds. Fifth, the confirmation of binding. The last step mainly contributes to the mechanism understanding. After the growing, linking, or merging steps, fragments inside a compound may not recur at the identical places from the virtual screening. Confirming the binding mode of the compounds with X-ray crystallography or NMR will reveal receptor-ligand interactions and provide insight into the mechanism of receptor activation or inhibition. The recent advancement in FBDD approaches is reviewed below.

Fig. 3.

Flowchart of computational fragment-based drug design. Five major steps are considered: First, the establishment of the diversified fragment library; second, the virtual screening for the given target; third, the fragments processing for the lead generation; fourth, the hits validation with biological assays; and fifth, the confirmation of binding modes

Fragment-Based Molecular Evolutionary Approach

Kentaro Kawai, Naoya Nagata, and Yoshimasa Takahashi described a fragment-based similarity-driven molecular evolutionary approach for designing candidates of new drugs in 2013 (28). This approach emphasizes the generation of candidates with differences on both side chains and the scaffolds when compared with the reference molecule.

The approach can be generally summarized in six steps:

1. An active molecule will be inputted as a reference structure.

2. Seed fragments will be created through the fragmentation of the reference molecule.

3. Initial structures will be generated through the combination of seed fragments and fragments in the library.

4. Offspring for the next generation will be constructed by mutation and crossover.

5. Fitness will be evaluated through taminoto coefficient, and the survival will be selected through a tournament method.

6. Steps 4 and 5 will be repeated to allow the alternative of generation reaching a specified number.

All the fragments in the fragment library are classified into three groups: rings, linkers, and side chains. Connection rules are applied for the fragment combination. Ring-ring connections, ring-linker connections, and ring-side chain connections are three allowable connections. While linker-liner connections, side chain-side chain connections, and linker-side chain connections are three unallowable connections. In mutation, a random selection will be performed to choose a parent molecule. One of the three operations: (1) add a fragment, (2) remove a fragment, or (3) replace a fragment will be performed with a selected parent molecule. For crossover, two offspring molecules will be created through the exchange of the fragment sets from the parent molecules. The crossover point is randomly selected from each parent molecule. In this approach, chemical structures are described through the topological-fragment-descriptor (TFD). The TFD method describes each chemical structure as a multidimensional, numerical pattern vector. The Tanimoto coefficient will be used to determine the similarity between the results and the reference molecule as a fitness function. The tournament method will compare the fitness value between two individuals from the parent set and offspring set and will select the individual with the higher fitness value. Half of the population will be selected eventually.

Molecular evolution experiments on two different targets, hAA2A and r5HT1A, were conducted by Kawai and the group to verify the feasibility of the described approach (28). For ligand design on hAA2A (Fig. 4a), seed fragments were created from the fragmentation of reference molecule 1. One hundred molecules were designed in silico using the seed fragments following the evolutionary approach. Many of these compounds remained the same non-terminal vertex graph (a graph that has no terminal vertex and no isolated vertex) as 1, while molecule 5 did not. Notably, 5 had the same non-terminal vertex graph scaffold with a known hAA2A ligand 7. The same process was followed for ligand design on r5HT1A. There was a comparison between the reference molecule, designed molecule, and active molecule (Fig. 4b). Molecules 8 and 10 were designed when 2 was used as the reference. Known r5HT1A active compounds 9 and 11 were identified to have the same non-terminal vertex graph scaffolds as 8 and 10, respectively. Molecule 13 was designed using 12 as the reference. 13 exactly matched the known r5HT1A active compound 14. In all, new potential molecules can be designed and developed to have similar but different scaffolds towards the references.

Fig. 4.

Application of fragment-based molecular evolutionary approach on hAA2A and r5HT1A. a Molecule 5 was designed using 1 as the reference. 7 was identified as a known active hAA2A compound to share the same scaffold with 5. b Comparisons of scaffolds and structures between designed compounds, reference molecules, and identified known active molecules

SILCS Computational Functional Group Mapping Approach

Olgun Guvench reported a computational functional group mapping (cFGM) technology with all-atom explicit-solvent molecular dynamic simulation (MD) for drug discovery in 2016 (29). SILCS represents Site-Identification by Ligand Competitive Saturation and is included in cFGM involving all-atom explicit-solvent MD. This approach is supposed to provide the information about the selection of functional groups for affinity, specificity, and pharmacokinetic properties, which are critical for not only optimizing known candidate molecules but also for designing novel compounds.

Experimental FGM involves synthesis and testing of a series of small, simple fragments to the given binding pocket. Fragments can be screened with high-concentration biological assays and validated with nuclear magnetic resonance (NMR)-based confirmation of binding. The binding poses of multiple fragments can be functioned as a map for guiding the compound design for medicinal chemists. Fragments can be linked, merged, or grown in a synthetically reasonable manner to result in drug-like molecules.

The computational approaches benefit from (1) time-saving, (2) labor-saving, and (3) material-saving. Their major liabilities remain (1) the accuracy of the model and (2) sufficient conformational sampling. SILCS cFGM brings three scientific advantages to experimental methods. First, the MD trajectory could capture both long-lasting and fleeting interactions, which makes it possible for detecting low-affinity binding regions. Second, as an outcome from the first advantage, a functional group map can include all regions inside the entire target structure. Not only the target surface as revealed from the crystal structure is considered but also the inducible pockets resulting from the fragment binding are included. Third, the aggregation of hydrophobic fragments and the denaturation of the targets can be prevented. Meanwhile, SILCS cFGM can also have quantitative applications. 3D pharmacophore models can be prepared through cFGM. Models can function as screening tools for compound selection. Maps can also be used to construct scoring functions for molecular docking and pose refinement.

Deconstruction-Reconstruction Approach

Haijun Chen et al. gave a comprehensive summary on improvements in the deconstruction-reconstruction approach for recent decades in 2014 (30). Instead of using established fragment libraries for the screening towards the given binding pocket and selecting preferable fragments, the deconstruction-reconstruction approach uses known ligands as the source for fragment generation. A relatively small and specific fragment library can be generated through the deconstruction of known ligands. The following reconstruction step is supposed to combine fragments in different regions and create novel compounds with new scaffolds.

The process, which is constituted with two steps, for the deconstruction-reconstruction approach is straightforward. The first step is the deconstruction of known compounds. Fragments will be created. The virtual screening between the given target and the fragments will reveal the potential interactions. The awareness of suitable fragments for different subdomains will benefit the drug design using growing, linking, and merging in the reconstruction step. Usually, obstacles exist with the reconstruction procedure. The newly designed molecules can have distorted binding modes of their individual fragments. The deconstruction-reconstruction approach is an efficient methodology for compound design and novel scaffold generation.

Multitasking Computational Model Approach

Alejandro Speck-Planche and M. Natalia D. S. Cordeiro reported the first multitasking (mtk) computational model for in silico fragment-based compound design of multiple-target inhibitors (31). One case study of designing molecules with good inhibition activities on multiple breast cancer-related proteins was presented in their paper to illustrate the effectiveness of the proposed approach. Instead of using traditional high-throughput screening or virtual screening for selecting active candidates through a huge database, the multitasking computational model likes to utilize the accumulated high-quality experimental data. Quantitative contributions of inhibition activities for each fragment will be calculated. And the physicochemical interpretations of descriptors in the mtk computational model will be considered. Both the fragment quantitative contribution and the descriptor physicochemical interpretation will provide the guidance for the multi-target compound design.

The establishment of the multitasking computational model involves three steps. The first step is the extraction of biological data for molecules through databases like CHEMBL. Restrictions can be applied towards the selection of molecules. Examples of restrictions are the number of reported measurements on activities, the type of measurements, and the accuracy and reproducibility of the assays. The second step is the calculation of molecular descriptors for selected molecules. Topological descriptors can be calculated by QUBILs-MIDAS (32), which is based on multilinear algebraic maps and discrete mathematics. The third step is the generation of the mtk computational model. Statistical cases will be split into training and test sets. The optimized model will be searched and determined by the training set. The test set will be used to show that the model is equipped with desirable prediction power. Several statistic indices will be utilized for the assessment of the model quality.

The established multitasking computational model will be further used to determine the quantitative contribution of each fragment. Fragments with positive contributions to the biological activities will be selected to connect with each other to generate new molecules. The druglikeness of the newly generated molecules will be analyzed, and the virtual screening between the target(s) and the newly generated molecule(s) will be followed to verify their effectiveness (Table I).

Table I.

Summary of the Latest and Widely Used Computational Programs and Algorithms for Fragment-Based Drug Design

Program name	Features	Universal resource locator
ACFIS server (33)	Core fragment can be generated from active molecules. Fragments can be grown to the junction site. Energy calculation will be used to determine the fragment fit.	http://chemyang.ccnu.edu.cn/ccb/server/ACFIS/
AutoT&T v.2 (34)	The multi-round optimization jobs are improved to give a higher speed. Structural crossover can be performed between several leads. Online server is available.	http://www.sioc-ccbg.ac.cn/software/att2/
CHARMMing Web User (35)	Fragment-based docking protocol is included. Commonly used computational chemistry methodologies are incorporated. Virtual screening can be directly performed through the online server.	https://www.charmming.org/charmming/
e-LEA3D (36)	Fragments or combination of fragments can be selected to fit a QSAR model or the binding pocket of a given protein. Virtual screening can be performed with user-specified criteria. Online server is available.	http://chemoinfo.ipmc.cnrs.fr
eMolFrag (37)	Non-redundant libraries of fragments could be prepared from a given set of molecules. The algorithm is available as both stand-alone software and a web server.	www.brylinski.org/emolfrag
eSynth (38)	Chemically feasible molecules could be reconstructed from molecular fragments. Diverse collections of molecules with the desired activity profiles could be generated.	www.brylinski.org/content/molecular-synthesis
FluX(39)	Druglike molecules will be deconstructed with pseudo-retrosynthesis rules. New molecules will be generated with stochastic search algorithm and ligand-based similarity scoring.	http://pubs.acs.org/doi/abs/10.1021/ci0503560
FTMAP (40)	Billions of positions of small organic molecules are sampled as probes. Probe poses are scored using a detailed energy expression. Binding hot spots are identified by clusters of multiple probe types.	http://ftmap.bu.edu/publications.php
iScreen (41)	Cloud computing web server. Protein preparation tool is implemented for estimating the size of the ligand binding pocket. The included docking modes are standard, in-water, pH environment, and flexible.	http://iscreen.cmu.edu.tw/intro.php
LigBuilder 2 (42)	Designed compounds will be analyzed for synthesis accessibility. The drugability of binding pockets on the protein surface will be quantitatively assessed. Fragments meeting certain drug-like criteria are selected for constructing novel compounds.	http://pubs.acs.org/doi/abs/10.1021/ci100350u The package can be acquired by contacting the authors.
Novoflap (43)	The building blocks are 1300 fragments in a fragment library. An evolutionary algorithm and a ligand-based scoring function were combined for this approach.	http://pubs.acs.org/doi/abs/10.1021/ci100080r
MED-SuMo (44)	The MED-SuMo fragment database can be used for populating the binding sites. MED-Portion chemical moieties hopping is the major strategy for this approach.	http://medit-pharma.com/index.php?page=fragment-based-approach
molBLOCKS (45)	molBLOCKS is a suite of programs to (1) break down sets of molecules, (2) cluster the resulting fragments, and (3) uncover enriched objectives.	http://compbio.cs.princeton.edu/molblocks
S4MPLE(46)	Several entities can be docked simultaneously. Sampling is based on the hybrid genetic algorithm. AMBER force field was used for energy calculation.	http://pubs.acs.org/doi/abs/10.1021/ci300495r
SHAFTS (47)	3D molecular similarity calculation uses a hybrid similarity metric and labeled chemistry groups. Experimentally determined conformations are not required for templates in ligand-based virtual screening.	http://pubs.acs.org/doi/abs/10.1021/ci200060s

SUCCESSFUL APPLICATIONS OF COMPUTATIONAL FRAGMENT-BASED DRUG DESIGN AND DISCOVERY

Case Study 1: Discovery of Small-Molecule STAT3 Inhibitor

Li, Yu, and their groups reported the application of in silico site-directed FBDD for designing the novel inhibitors for STAT3 (48). STAT3 is an attractive target for therapeutic uses among different carcinomas, including breast cancer and sarcomas. The phosphorylation and dimerization of STAT3 are closely related to the oncology of the above-mentioned carcinomas (49,50). STAT3 inhibitors should bind to the SH2 domain on the protein in order to block both the phosphor-ylation and dimerization processes through competing with the native phosphotyrosine loop (48). A flowchart (Fig. 5) was summarized to demonstrate the workflow. In their study, they first did the molecular docking studies between the SH2 domain and the known STAT3 dimerization inhibitors. The fragment library was established based on the known inhibitors, and the sublibraries, which include fragments belonging to the site pTyr705 and the side pocket, were determined according to the binding modes revealed through the docking studies. Then, selected fragments from one sublibrary were linked with fragments in the other sublibrary to build a new library for novel lead compounds. Finally, the molecular docking studies between the novel leads and the SH2 domain were performed to determine the final leads for chemistry synthesis and biological evaluation. Four out of five synthesized compounds have their IC₅₀ less than 5 micromolar against the cancer cell line U2OS. Especially, compound 8 has a great potency among a variety of cancer cell lines. The fluorescence polarization assay verified the binding of 8 with the STAT3 SH2 domain.

Fig. 5.

An illustrated flowchart of small-molecule inhibitor design on STAT3 using “fragment linking.” Fragments belonging to different sites were colored in blue and orange. The structure of compound 8 was specified. The two colors in the compound indicated the interaction sites

Case Study 2: In Silico Design of Allosteric Modulators on mGluR5

Xie and his group reported their practice of using computational FBDD for finding novel allosteric modulators on metabotropic glutamate receptor 5 (51). mGluR5 is an appealing target for drug development since it is associated with multiple diseases in the central nervous system. Furthermore, designing GPCR allosteric modulators is a promising strategy for drug development, given that allosteric binding pockets allow additional receptor-ligand interactions (51,52). The process of their application on compound design was summarized into a flowchart (Fig. 6). In the study, 27,262 known GPCR allosteric modulators were broken into fragments using RECAP analysis first. Rule of three was used here for guiding the selection of preferred fragments. The molecular docking studies between the fragment library and the mGluR5 allosteric binding pocket were followed. The upper and bottom regions inside the pocket were revealed through the docking studies. Then, the fragments were categorized into two groups according to the interactions and affinities between the fragments and the residues of a certain region. RECAP synthesis was adopted for generating novel leads by connecting fragments in these two regions. The molecular docking studies between newly generated leads and the mGluR5 allosteric binding pocket were conducted for selecting the potential modulators. Multiple docking algorithms were used for evaluation. The enrichment test with challenging decoys and the QSAR simulation were continued for validation. Among the top 20 in silico hits, structural diversified compounds with reported or unreported scaffolds could be observed. A similarity search revealed that a series of patented scaffolds with reported mGluR allosteric activities could recur through the in silico design.

Fig. 6.

An illustrated flowchart of allosteric modulator design on mGluR5 using fragment linking. Upper and bottom regions, as well as the corresponding fragments, were colored in blue and orange

Case Study 3: Fragment-Based Inhibitor Discovery on CTX-M β-Lactamase

Chen and Shoichet reported their research on using the in silico fragment-based drug design to facilitate the compounds development processes. Their efforts resulted in the first micromolar-range class A β-lactamase, CTX-M, non-covalent inhibitor (11). β-lactamase CTX-M turns out to be one of the most important mechanisms for the bacteria resistance towards third-generation cephalosporins (53). With a large and open active site, CTX-M has been considered a difficult target for developing non-covalent inhibitors. In their study, molecular docking studies between the active site and a fragment subset (67,489 compounds) or a lead-like subset (1,147,326 compounds) were performed. Compounds were selected among the top-ranking list. Ten out of 69 investigated fragment inhibitors had their IC₅₀ values in millimolar level. Given the complementarity between the active site and the tetrazole group, which was discovered through the fragment docking study, they progressed to look for larger analogs. Compound 12 was subsequently identified with Ki value of 21 micromolar. 12 was ranked 982 among the docking screening for lead-like compounds, which is low enough for researchers to ignore it. Co-crystallization of the protein and inhibitors was performed for confirming the binding and comparing the docking modes. Competitive binding assays on CTX-M and AmpC were continued to evaluate the specificity of inhibitors’ binding. The key points of their approach were summarized and a flowchart was created to illustrate the workflow (Fig. 7). Dr. Yu Chen and Dr. Brian K Shoichet discovered the first micromolar-range CTX-M non-covalent inhibitor with computational fragment-based drug design. Also, they stated that “the high hit rates among fragments partly arise from their initial low specificity, and that specificity can be designed into these molecules as they are advanced”.

Fig. 7.

An illustrated flowchart of inhibitor discovery on CTX-M β-lactamase using fragment growing. Molecular docking studies using lead-like subset were followed after identifying critical fragment hit through fragment-based screening. The structure of compound 12 was specified. The tetrazole group was colored in blue

LIMITATIONS AND FUTURE PROSPECTIVE

Fragments are relatively smaller molecules compared with lead-like and drug-like compounds, which makes the molecular docking processes much more difficult (54). The difficulty mainly occurs because of two aspects. (1) Fragments usually cannot form sufficient interactions with the surrounding residues. The limited amount of functional groups results in the low-affinity binding. Investigators may miss valuable and potential fragment hits because these fragments have a tiny size and weak interactions towards the protein. (2) Due to the relatively small size, the docking modes for fragments can be promiscuous. Fragments can be favored by multiple sub-regions, in case they share similar chemical-physical properties, inside a binding pocket. The prediction is hard to perform, and the subsequent analysis processes can be time-consuming.

Molecular docking technology underwent progress in the last several decades. Researchers can choose from systematic or stochastic methods for conformational search and force-field-based, empirical, or knowledge-based functions for scoring (55). However, the accuracy for predicting the receptor-ligand binding modes still cannot be guaranteed. Considering that parameterized lead-like or drug-like molecules can often have the inaccurate prediction, it is even more challenging to give reliable predictions for small fragments.

Reliable protein models and precise sites of the binding pockets are required for computational fragment-based drug design. The majority of the proteins, especially for membrane proteins like GPCRs, currently do not have the crystallized structure information, which challenges the application of this drug development strategy. Homology modeling relieves this problem to some extent if the target protein shares high sequence similarity with the crystallized template. But the overall situation is far from optimistic. Also, static models may not be able to represent the proteins in the dynamic biological environments. The rotation of one or more critical residues may cause the dramatic conformational change to the shape of the binding sites, or even the whole protein structure. The binding modes revealed through static models may not recur the receptor-ligand interactions in reality.

The development and improvement of novel and current methodologies will be the focus in the future in order to consummate the computational FBDD. New methodologies should improve the accuracy of the fragment docking and make the docking poses predictable. The combination of experimental and computational approaches should also be emphasized. In silico approaches will reduce the workload, save resources, and provide guidance for chemical synthesis and biological tests to some extent. With more practice and continuing improvements in computational FBDD, challenging barriers can be overcome, and future drug development processes can benefit.