Mar 7, 2016 - According to the FMO calculations, 3i (Figure 7B) and 3k (Figure 7C) display the same interactions as the parent 3g, but additional contacts are also evident. .... In case of 2câq (Table S2), the in-house kinase structure was replaced
Mar 7, 2016 - Inhibition of inducible T-cell kinase (ITK), a nonreceptor tyrosine kinase, may represent a novel treatment for allergic asthma. In our previous ...
Oct 11, 2012 - Performance analysis of open-source distributed file systems for practical large-scale molecular ab initio, density functional theory, and GW + BSE calculations. LoÃ¯c M. Roch , Tyanko Aleksiev , Riccardo Murri , Kim K. Baldridge. Inte
Coupled fragment molecular orbital method for interacting systems ... for Nondisjoint Monomers in Molecular Fragmentation Calculations of Covalent Molecules.
The fragment molecular orbital method (FMO) has been used with a large number of wave functions for single-point calculations, and its high accuracy in ...
Lubbock, Texas 79409. Received May 26, 1967. Three papers have recently appeared which describe. Diels-Alder reactions under mild conditions that yield.
Mar 16, 2007 - In particular, the Î±-helix, Î²-turn, and extended conformers of a ... with the rms deviation from ab initio of about 0.2 Ã , or 0.5Â° in terms of bond angles. ... The Journal of Physical Chemistry A 2016 120 (49), 9794-9804 .... Dimer
Mar 1, 2010 - method, for calculating nuclear magnetic resonance (NMR) chemical shifts ... density based on electrostatic potential obtained from FMO calculations, the NMR calculations ... The first are ab initio methods dealing with gauge-.
Theoretical Study of Activation of Oxirane by Bidentate Acids. Kiyoyuki Omoto and Hiroshi Fujimoto. The Journal of Organic Chemistry 2000 65 (8), 2464-2471.
Mar 7, 2016 - Graduate School of System Informatics, Kobe University, 1-1 Rokkodai-cho, Nada-ku, Kobe 657-8501, Japan. ABSTRACT: A subsystem ...
Subscriber access provided by MAHIDOL UNIVERSITY (UniNet)
Fragment Molecular Orbital Method Applied to Lead Optimization of Novel Interleukin-2 Inducible T-Cell Kinase (ITK) Inhibitors Alexander Heifetz, Giancarlo Trani, Matteo Aldeghi, Colin H MacKinnon, Paul A McEwan, Frederick A. Brookfield, Ewa I. Chudyk, Mike Bodkin, Zhonghua Pei, Jason D. Burch, and Daniel Fred Ortwine J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.6b00045 • Publication Date (Web): 07 Mar 2016 Downloaded from http://pubs.acs.org on March 8, 2016
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Fragment Molecular Orbital Method Applied to Lead Optimization of Novel Interleukin-2 Inducible T-Cell Kinase (ITK) Inhibitors
Alexander Heifetz1*, Giancarlo Trani1, Matteo Aldeghi2, Colin H. MacKinnon1, Paul A. McEwan1, Frederick A. Brookfield1, Ewa I. Chudyk1, Mike Bodkin1, Zhonghua Pei3, Jason D. Burch3 and Daniel F. Ortwine3*
ABSTRACT Inhibition of the non-receptor tyrosine kinase ITK may represent a novel treatment for allergic asthma.
In our previous reports we described the discovery of sulfonylpyridine
(SAP), benzothiazole (BZT), indazole (IND), and tetrahydroindazole (THI) series as novel ITK inhibitors, and how computational tools such as dihedral scans and docking were used to support this process. X-ray crystallography and modelling were applied to provide essential insight into ITK-ligand interactions. However, “visual inspection” traditionally used for the rationalization of protein-ligand affinity cannot always explain the full complexity of the molecular interactions. The fragment molecular orbital (FMO) quantum-mechanical (QM) method provides a complete list of the interactions formed between the ligand and protein that are often omitted from traditional structure-based descriptions. FMO methodology was successfully used as part of a rational structure-based drug design effort to improve the ITK potency of HTS hits, ultimately delivering ligands with potency in the sub-nanomolar range.
The understanding of binding interactions between any protein and a small molecule plays a key role in the rationalization of potency and selectivity1. The efficiency and costeffectiveness of the drug-discovery process can be accelerated by the availability of structural data regarding the target protein, and by the reliability of the computational tools for data explorations.2-4 However, even with the crystal structure in hand, “visual inspection” and force field-based molecular mechanics (MM) calculations often used for the rationalization of ligand-protein potency cannot always explain the full complexity of the molecular interactions.1 There is increasing evidence1, 5-7 that there are a number of nonintuitive interactions such as CH/π8, 9, halogen/π10, cation/π11, non-classical H-bonds12, and others, that play important roles in protein-ligand binding that are not properly parameterized in the most popular force fields (FF).6 The use of quantum mechanical (QM) methods has traditionally been employed to improve the reliability of the exploration of protein-ligand interactions13, 14. However, in spite of their many advantages classical QM methods have not been feasible until recently for large biological molecules such as kinases, due to their high computational cost.15 The fragment molecular orbital (FMO) method
9, 14, 16
offers substantial computational
time saving over traditional QM methods.17 This is achieved by dividing the system into smaller pieces called fragments. For example, in proteins, each residue can be represented by a fragment. Similarly, the ligand can be represented by single or multiple fragments as necessary. By performing QM calculations on fragments, one can achieve high efficiency in much shorter time, especially when parallel compute clusters are used to carry out the analysis.17 A typical FMO calculation on a kinase/inhibitor complex takes approximately 4h on a 36 CPU cores to complete, significantly faster than classical QM calculations.9 FMO takes as an input a protein-ligand complex and provides a list of interactions and their chemical nature.15 The strength of these interactions are quantified in terms of 3 ACS Paragon Plus Environment
energy stabilization of the complex (in kcal/mol).
By summing up the pair interaction
energies (PIEs) calculated for individual fragments, one obtains an estimate of the overall binding energy gain that can be related to the resultant Ki of the compound as measured in a biological assay. If a significant correlation results, biological potencies for virtual compounds can be predicted. Pair interaction energy (PIE) between any two fragments calculated by FMO is a sum of four energy terms: electrostatics, exchange-repulsion, charge transfer, and dispersion, and is provided by a pair interaction energy decomposition analysis (PIEDA,18 Figure 1). The electrostatic and charge transfer terms are important in H-bonds (classical and non-classical1), polar interactions, and salt-bridges, while the dispersion term is more prominent in hydrophobic interactions. The exchange-repulsion term describes steric repulsion between electrons15 that prevents atoms from clashing with each other. These data are regularly used by medicinal chemists within modelling software packages to rationally engineer additional potency into subsequent target molecules. The role of hydrophobic interactions is vital for biomolecular recognition but there is still no reliably predictive method for its quantification.1 There are two significant benefits to using FMO.
Chemists can more readily relate
to QM results as the method quantifies chemically interesting interactions, in particular, nonpolar interactions which historically have been hard to quantify. Additionally, because only a small, predefined portion of the protein and ligand are studied, calculations become much faster due to a reduced computational cost. FMO has previously been used in a retrospective manner for prediction of cyclin-dependent kinase 2 inhibitor potency19, to correlate experimentally measured potencies and FMO calculated PIEs for FK506-protein and its ligands20, for analysis of the GPCR-ligand crystal structures21, and potency calculations on a novel Hsp90 fragment-linked inhibitor.22 As demonstrated herein, FMO methodology can provide a useful platform for examining detailed energetics of proteinligand interactions within defined subregions of the receptor, in proactive support of a lead optimization campaign.
ITK is a tyrosine kinase involved in T cell development, differentiation, and effector function.23,
It has been shown to play an important role in the development of T-cell
dependent late phase responses of allergic asthma. Studies using an ITK knock-out transgenic mouse model suggest that ITK kinase activity is required for the control of Th2 responses and the development of allergic asthma.25 The wealth of preclinical evidence supporting the role of ITK in allergic asthma and other inflammatory disorders has prompted substantial research effort from the pharmaceutical community toward the discovery of selective inhibitors of this kinase. Preclinical in vivo activity using ITK inhibitors has been reported26, resulting in the reduction of IL-2 production in one example and the reduction of IL-4 production and lung inflammation in another. In previous reports27-29 we have described the discovery of sulfonylpyridine27 (SAP – Supporting Information Table S1), benzothiazole29 (BZT – Supporting Information Table S2), indazole28 (IND – Supporting Information Table S3), and tetrahydroindazole (THI)30, 31 series as novel sub-nanomolar and selective ITK inhibitors. In these reports27-29,
demonstrated how X-ray crystallography and computational tools such as QM dihedral scan, docking, and physiochemical property calculations were applied to progress the hit-to-lead and lead-optimisation programs for these series. The FMO method was also used to analyze crystal structures and models of ITK-ligand complexes and to drive forward the design of subsequent generations of several of these series. The present report is a detailed description of how FMO was applied in the ITK project that wasn’t described in previous reports. FMO can be applied to any ligand-protein complex. The integration of FMO with X-ray crystallography and modelling can provide a powerful and efficient platform for supporting drug discovery programs, and therefore this illustrative example can be highly beneficial for medicinal and computational chemists.
A key challenge early-on in the ITK project was to improve in vitro potency of HTS hits while either maintaining or enhancing other attributes such as selectivity, solubility, and DMPK properties. This required detailed exploration of the interactions between ITK protein residues and ligands, aided by the availability of several crystal structures of protein/inhibitor complexes. Classical hydrogen bonds (H-bonds) with kinase active site hinge residues were readily identified by visual inspection; however, other key interactions were less straightforward to detect by eye. We decided to apply the FMO method to explore the full scope of interactions between ITK and inhibitors to gain a more detailed understanding of the nature and magnitude forces governing protein/ligand interactions at specific active site subsites, to better understand the emerging SAR and proactively apply the learnings to the design of more potent inhibitors.
Experimentally measured potency (pKi) vs. FMO calculated PIE
To assess how accurately the total pair interaction energy (PIE) described the interactions between ITK and our sub-series, we performed a correlation between total PIE energy calculated by FMO and the experimentally measured binding affinities (pKi). The correlation plots between measured pKi values (Supporting Information Tables S1-S3) and calculated PIEs are shown in Figure 2. Overall, in three series (SAP, BZT, and IND) we observed a significant correlation (r2 >0.83) between the experimental values and FMO calculated PIEs. This observation provided us with confidence that the PIE energy calculated by FMO quantitated the interactions between ITK and our compounds in these three series to a significant enough extent that it could be applied prospectively in an SBDD setting. It is important to emphasize that we did not expect to see a perfect correlation between pKi and PIE since the binding affinity is not only dependent on specific interactions calculated by the FMO technique but also on additional energy terms such as entropy, desolvation and strain energy present in the ligand’s bioactive conformation that are not accounted for by FMO. Also noteworthy is the fact that for the ITK protein-ligand inhibitor complexes studied, the 6 ACS Paragon Plus Environment
active site residues did not significantly change conformation, making this lock and key relationship well suited to a QM analysis where the protein is held rigid. Performance of the FMO method was also compared with binding energy values calculated using the MM/GBSA method (see Supporting Information Figure S1). The FMO method clearly outperformed the force field-based approach, demonstrating a high correlation (r2 >0.83) with experimentally measured ITK Ki values relative to that obtained by the MM/GBSA method (r2< 0.21).
Sulfonylpyridine (SAP) ITK inhibitors
In our previous report27 we described in detail how computational tools such as docking and QM-based dihedral scan were applied to progress an initial HTS hit into the potent and selective lead compound 1j (Table 1). FMO analysis was applied for exploration of ITK-1j interactions.
Biological data has been published.27 19 A calculated difference of 5 kcal/mol in PIE values is considered significant
Ten significant interactions were identified by FMO as shown in Figure 3A. The aminopyrazole moiety of 1j forms three H-bonds with the hinge residues, two to Met438 and one to Glu436. An additional H-bond is formed between the nitrogen linker of 1j and the backbone carbonyl of P-loop residue Ile369. Compound 1j is sandwiched in a narrow lipophilic cleft, formed by the N and C lobes of ITK, and forms three significant hydrophobic interactions with Ile369, Cys442, and Leu489. The sulfone linker facilitates an edge–to-face, π-stacking interaction between the phenyl ring and Phe437, and the same phenyl ring on the ligand forms a non-classical H-bond12 with the backbone carbonyl of Glu439. The ligand cyclohexanol group forms a hydrophobic interaction with Cys442 and an H-bond to a water molecule (HOH2) that mediates the interaction of 1j with Ser499 and Asp500.
In order to explore whether the binding affinity and selectivity profile of 1j could be further improved, we prepared a range of heterocycles (Supporting Information. Table S1). In designing these analogs we took into account the fact that classical H-bonds can be partially replaced by non-classical ones12. Thiazole compound 1q demonstrated improved subnanomolar affinity for ITK, (Ki = 0.27 nM, a 14-fold increase over 1j). The FMO analysis of ITK-1q complex (Figure 3B) in comparison to parent 1j (Figure 3C) revealed a similar set of 10 interactions. However, despite the high similarity between these compounds, FMO analysis highlighted several significant differences that might otherwise be missed by visual inspection. Specifically, the non-classical H-bond formed by the thiazole ring of 1q with Glu436 was found to be weaker than the classical H-bond formed by the pyrazole of 1j. However this weaker interaction was overcome by stronger interactions elsewhere in the inhibitor with Ile369, Phe437 and HOH2 (Figure 4). As can be observed from this PIEDA analysis, the stronger interactions formed by 1q with these three residues and the water molecule was mainly a result of increased electrostatic and reduced exchange repulsion energy terms of PIE that could be related to differences in the polarization effect of the 8 ACS Paragon Plus Environment
pyrazole compared to the thiazole. According to overall PIE calculations (Table1) 1q was predicted to be more potent than 1j. Our hunt for more potent and selective ITK inhibitors culminated in the design of compound 1r. As indicated by FMO analyis (Figure 3C), increasing the size of the pyrazole substituent from methyl (1j) to cyclopentyl (1r) led to additional favorable interactions with Phe435 and Asp500, which translated into a 23-fold improvement in ITK potency over 1j. The electrostatic attraction between the cyclopentane group of 1r and Asp500 (as seen from the PIEDA plot, Figure 3C) could be rationalized by the location of a specific portion of the positive electrostatic potential projecting into a deep and weakly desolvated cyclopentane area of the ITK binding site, close to the gatekeeper and adjacent to the negatively charged Asp500 (Figure 5). Selectivity against one of the major off-target kinases (Lck) was also greatly improved due to the lost hydrophobic interaction with the gatekeeper residue between ITK (Phe gatekeeper) and Lck (Thr gatekeeper).In summary, the application of FMO was useful for revealing several key and non-obvious interactions formed by parent 1j and was thus helpful in the design of 1r.
Benzothiazole (BZT) ITK inhibitors As previously reported29 the benzothiazole HTS hit (compound 2d, Table 2) had a moderate ITK Ki of 460 nM.
Biological data has been published29 Unpublished data c 19 A calculated difference of 5 kcal/mol in PIE values is considered significant
FMO analysis of ITK-2d interactions (Figure 6A) revealed 6 key interactions with ITK. Two H-bonds are formed between hinge residue Met438 and the two nitrogen atoms of the benzothiazole amide of 2d. A non-classical H-bond is formed by the backbone carbonyl of Glu436 and the hydrogen atom on C4 of the benzothiazole ring. Similar to other ITK ATPsite binders, the inhibitor is positioned in a narrow lipophilic cleft between the N and C lobes of ITK and forms a CH-π interaction with Ile369 and a hydrophobic interaction with Leu489. Finally, the cyclopropyl moiety directs the attached benzene ring to form a face-to-edge, πstacking interaction with Phe437. Analogs where this benzene ring was replaced with a cyclohexyl (2a, Supporting Information Table S2) or 2-pyridyl (2b) group displayed decreased potency, which was consistent with observations from the FMO analysis.
In our attempt to improve the potency of the BZT series we envisaged that an aromatic substituent off the 6-position of the benzothiazole core might form an interaction with the gatekeeper residue Phe435 and facilitate additional interactions with residues in the DFG/αC helix area of ITK. A rational SBDD process resulted in the design of 2p and 2q (Table 2). FMO analyses of 2p (Figure 6B) and 2q (Figure 6C) revealed many new interactions beyond the original 6 observed for the parent 2d (Figure 6A). Compound 2q displayed 10 interactions with ITK compared to 9 in the case of 2p. A face-to-face π-stacking interaction with Phe435 and a non-classical H-bond between a cyclopropyl hydrogen and the carbonyl backbone oxygen of Glu439 were observed for both 2p and 2q. Interestingly, a strong H-bond interaction with Glu406 (αC helix residue) was present solely with ligand 2p 10 ACS Paragon Plus Environment
while two significant interactions with Ser499 and DFG Asp500 were predicted solely in the case of 2q. According to the total PIE calculation, 2q was predicted to be more potent than 2p, and indeed, the experimental binding affinities bore this prediction out (Table 2). Compounds 2p and 2q demonstrated 112- and 657- fold increases in ITK potency, respectively, over the original HTS hit (2d).29 The existence of 10 interactions in 2q has been validated by the crystal structure of ITK-2q (PDB entry 4MF129). FMO methodology was useful for exploring the interactions of 2d with ITK, and for identifying potential interactions that this compound lacked with ITK. This subsequently enabled the use of FMO in the design process of 2p and 2q; two compounds that made several additional interactions with ITK and benefitted from the increased potency derived from these interactions.
Indazole (IND) ITK inhibitors
As we previously reported28 compound 3g was identified from an HTS campaign and was used as a starting point for chemistry optimization.
Biological data has been published28 PIE calculated difference of 5 kcal/mol is considered significant19
A high resolution co-crystal structure of ITK and 3g (resolution 2.6Å, PDB entry 4PP928) was subjected to FMO analysis that detected 12 significant interactions (Figure 7A). Hinge residues Met438 and Glu436 form three H-bonds with the amide indazole core of 3g, and an additional non-classical H-bond is formed between Met438 and the pyrazole moiety. A non-classical H-bond is also formed between the carbon linker of the ligand and carbonyl backbone of Glu439. The average PIE for a non-classical H-bond interaction was reported to be -6.5 kcal/mol33, however in the case of Glu439, we observed a contribution twice the normal size (-12 kcal/mol, see plot in Figure 7A), which indicated that this residue likely forms a second non-classical H-bond with an aromatic hydrogen of the cyanobenzyl phenyl ring. This ring also forms a strong face-to-edge π-stacking interaction with Phe437, and the cyano substituent H-bonds with Lys387. As with the other series, the IND series is bound in a narrow lipophilic cleft, formed by the N and C lobes of ITK, and forms hydrophobic interactions with Ile369, Val377, Leu489, and gatekeeper residue Phe435.
Based on the ITK-3g crystal structure and FMO analyses, exploration of substitution at the indazole C6 position was undertaken (Supporting Information Table S3) targeting the gatekeeper, DFG, and αC Helix residues. This process culminated in the design of compounds 3i and 3k (Table 3). Overall PIE energy calculated by the FMO (Table 3) suggested that the 4-pyrazole substitution of 3k was preferable in terms of ITK potency compared to the 3-pyrazole of 3i. According to FMO calculations, 3i (Figure 7B) and 3k (Figure 7C) display the same interactions as parent 3g but additional contacts are also evident. Specifically, a π−π stacking interaction is formed with gatekeeper Phe435, an H12 ACS Paragon Plus Environment
bond is evident with Lys391 (according to PIEDA, this interaction in 3k is 3-fold stronger than in 3i), and interactions with Ser499 are seen. In the case of 3i, the 3-pyrazole forms an additional interaction with DFG residue Asp500, while the 4-pyrazole of 3k forms an interaction with αC Helix residue Met410. We rationalized the stronger interaction with Lys391 in the case of 3k is decisive in making this compound more potent than 3i. When tested in the ITK binding assay, compound 3k showed a 54-fold increase in ITK potency over the HTS hit (3g) compared to just 10-fold for 3i over 3g.
Additionally, FMO analyses of the pyrazole linker moiety compared to other linker heterocycles as reflected in compounds: 3a, 3b, 3c and 3d (Supporting Information, Table S3) indicated a significant decrease in their overall PIE energy compared to the parent 3g. This was consistent with observed differences in potency among these analogs. A detailed analysis of FMO calculations between 3g and 3a is shown in Supporting Information Figure S2. In summary, FMO analysis of the 3g complex detected a large number of nonobvious interactions that were important for ITK binding. This information was incorporated in the design of highly potent IND analogs, leading ultimately to the discovery of 3k. Compound 3k was used as starting point for the design of second generation tetrahydroindazole (THI) series that are beyond the scope of this manuscript. 30, 31, 16
In this work we described how FMO was successfully applied to rational SBDD of novel and selective ITK inhibitors. A key advantage of this method was that it provided detailed information on the individual contribution and chemical nature of each residue and water molecule to the ligand binding that normally would be hard to detect without QM. This information was useful for the design of subsequent generations of ITK analogs. It helped to 13 ACS Paragon Plus Environment
rationalize differences in ITK potencies when compounds appeared to be very similar on visual inspection. The correlation between experimentally measured affinity and FMO calculated pairwise interaction energy gave us confidence to use it as a filter of new targets for synthesis.
FMO has been demonstrated to be an efficient tool for rational SBDD, as it provides an accurate and comprehensive list of strong, weak, or repulsive interactions between the ligand and its surrounding protein residues. Even small energetic contributions, when summed, may give rise to a large total value. Such insights can be useful to guide substitution, modification, linking, or extension of ligands to form stronger or new interactions with the protein or even to decrease undesired repulsions. The chemical nature of such interactions computed by PIEDA can be important for the design of the next round of compounds by highlighting the most appropriate type of modification required to improve binding whether it involves modifying dispersive, electrostatic, or charge-charge interactions. As demonstrated in this paper, newly designed molecules can also be prioritized ahead of synthesis using FMO calculations. In appropriately defined cases, energy terms calculated by FMO can also be used to predict affinity.
FMO methodology should always be applied to well modeled or crystal structures. In this work we have demonstrated that FMO can be useful in cases where at least one crystal structure per series is available, provided that binding poses can be generated for the remaining ligands by overlaying them onto the appropriate parent compounds. The FMO version used in this project performed its calculations in vacuum, without taking into account the desolvation effect. The most recent version of FMO now includes a polarizable 14 ACS Paragon Plus Environment
continuum solvation model (PCM)34 that allows the inclusion of a desolvation term into the FMO calculation.
One can argue that a competent medicinal chemist can visualize many if not most of the interactions described in this work. An advantage of the FMO method is a quantitation of the effect if carefully applied to specific series in a lead optimization project where the protein is reasonably rigid.
Another benefit is the detection of more subtle non-traditional H-
bonding/electrostatic stabilization that can contribute to potency or selectivity differences. Such subtle interactions can be exploited in the examination of multiple candidate analogs, since the technique is computationally efficient and can be rapidly run in a time frame soon enough to impact fast-moving projects in a lead optimization stage. Like many other SBDD computational approaches such as docking and de novo design, it is important to know the conformational preferences of the complex under study before undertaking these calculations, as significant protein motion will undermine the results. Nevertheless, in the case of ITK compound design, these calculations proved useful in revealing subtle interactions that rationalized potency differences and provided suggestions for synthesis prioritization. The in-silico methods that were applied in this project, including FMO, complemented each other and provided a powerful, efficient, and cost-effective platform for successful SBDD against this target.
Structure preparation The structures of ITK-ligand complexes were determined by crystallography and modelling27-29. Hydrogen atoms were added to the ITK-ligand complexes at physiological pH 15 ACS Paragon Plus Environment
(7.0) with the Protonate 3D35 tool implemented in version 2010.10 MOE (Chemical Computing Group). Default Protonate 3D settings were used. For electrostatic interactions the Generalized Born / Volume Integral methodology with a 15 Å cutoff solvent dielectric constant 80 was applied. The complexes of all SAP analogues with ITK were initially modelled by docking into a published ITK structure36. As the project progressed, this protein structure was replaced by one containing 1l (PDB entry 4QD627, resolution 2.45Å). For FMO calculations, we generated ITK-SAP complexes by using the OMEGA-ROCS tool as implemented in the OpenEye software package (version 3.0.0) to overlay SAP compounds onto the binding pose of 1l (extracted from the crystal structure) assuming that similar ligands in the same protein would bind in the same mode. The crystal and modelled structures were subjected to a restrained minimization procedure with the MMFF94x force-field37 implemented in MOE (Molecular Operating Environment of Computational Chemical Computing Group, version 2010.10), wherein each atom was allowed to deviate by up to 1.0 Å from its original position in the crystal structure or model. Because small errors in the positions of atoms can translate into large deviations in energy terms, it was important to optimize the structures before applying any type of calculations to them4. In case of modelled complexes the minimization was required to reduce potential clashes and optimize interaction geometries. For the BZT series, co-crystal structures were solved for 2e (unpublished), 2m (PDB entry 4MF029), and 2q (PDB entry 4MF129). Unfortunately for 2e, the Xray was incompletely resolved. Therefore we relied on a model of 2e built from closely related structures of 2m and 2q. For the FMO calculations, we used the model of 2e and the Xray of 2m to model other ITK-BZT structures in the same way as we did for the SAP series. In case of 2c-2q (Supporting Information Table S2) the in-house kinase structure was replaced by an ITK crystal structure taken from the PDB (code 3QGY) due to a poorly resolved αC helix region
in our structure.38 We wanted to explore potential interactions of designed compounds with this region, particularly with Glu406. All crystal structures and models were subject for minimization as described for SAP. In the case of the IND series, crystal structures were solved for HTS hit 3g (PDB entry 4PP928) and for 3k (4PPA28). For the FMO calculation we used the 3g crystal structure and ROCS to generate the ITK-ligand complexes. In the case of compounds 3f, 3h, 3j, 3i, and 3k (before crystal structure of 3k was solved) the in-house crystal structure of the protein was replaced by an ITK crystal structure taken from the PDB (code 3QGY38). This was carried out in order to explore the interaction of these compounds with αC Helix residues Glu406 and Met410. A similar minimization procedure was performed for crystal structures and modelled ITK-IND complexes as in the SAP series.
FMO method and protocol The FMO method is a general quantum mechanical method that can be applied to any set of atoms, irrespective of whether they are a part of a soluble or a membrane-bound protein. Here, the FMO method9 was applied to kinase-ligand complex using FMO code17 version 5.0 as embedded in General Atomic and Molecular Electronic Structure System (GAMESS)39,
which is a general ab initio quantum chemistry package. The major
differences between FMO and MM methods were discussed in previous report21 and originates mainly from the fact that FMO takes into consideration polarization (in the selfconsistent mutual polarization of fragments) and charge transfer (whereby charge is allowed to flow between fragments). 9, 33 In FMO calculations, a large biological system is divided into smaller parts called fragments (Figure 1)9,
. In this study, we adjusted the standard FMO practice to use
traditional amino acids (including the NH and CO linkers) as fragments. Thus, each residue was characterised as a fragment, and the interaction energies reported herein correspond to 17 ACS Paragon Plus Environment
actual amino acid residues as opposed to residue fragments9.
The ligand can also be
represented as one fragment or can be fragmented; some ligands can be very large and dividing a ligand into several fragments has the benefit of both reducing the computational cost and providing a more detailed analysis. The detailed description of the FMO strategy and methodology can be found in the published reviews9,
, including a detailed
mathematical formulation that are beyond the scope of this manuscript. In this work we used a well-established FMO protocol that was previously described and tested. 8, 9, 19, 40, 41 The FMO calculation consists of the following steps: (a) Fragmentation (i.e., assigning atoms in a system to a fragment); (b) Fragment self-consistent field (SCF) calculations in the embedding polarizable potential, so that fragments mutually polarize each other in a self-consistent fashion whereby intra-fragment charge transfer and other quantum effects are accounted for; (c) Fragment pair SCF calculations, bringing in inter-fragment charge transfer; (d) Total property (energy, gradient, etc.) evaluation. By performing QM computations on fragments one can achieve high efficiency, often resulting from linear scaling and computational speed. parallelized for PC clusters
The FMO method has been
. In this work we used the MP2 method (2nd order Møller-
Plesset perturbation theory42) with the 6-31G* basis set. This basis set is most commonly used and is often considered the best compromise between speed and accuracy.19 Residues and water molecules within a radius of ≤ 4.5Å around the ligand atoms were included in the FMO calculations, since previous work demonstrated21 that including these atoms significantly increases the speed of the calculation without compromising the results. The FMO calculations in this work were performed in vacuum, which can lead to overestimation of charge-charge interactions formed by normally charged residues. To avoid this situation in this work, the charged residues Asp, Glu, Arg, and Lys were neutralized. The basic assumption was that most of these residues (Lys387, Lys391, Glu406, Glu439 and Asp500) are located in the water exposed ITK binding site and due to the de-solvation effect their contribution to the charge-charge interaction with the ligand would be low. In modern
FMO versions there is no need to neutralize the charged residues since they can be treated with a polarizable continuum solvation model (PCM)34. int
PIE – interaction energy ( ∆E ) between fragments i and j is a sum of 4 PIE terms: es
electrostatics ( ∆E ), exchange-repulsion ( ∆E ), charge transfer ( ∆E ) and dispersion (
∆E di ) (Equation 1). The chemical definition of each term is described in Figure 1.
∆Eijint = ∆Eijes + ∆Eijex + ∆Eijct + ∆Eijdi
The PIE is not a difference between energies of the protein-ligand complex and the sum of the ‘free’ protein and ligand, but rather represents the ‘strength’ of the interaction between the ligand and protein residues in the complex. The ∆ signs refer to the differences in the total QM energy of a fragment pair ij and two individual fragments i and j, both computed in the receptor-ligand complex. In the present work, fragment i was the ligand and the other n fragments were receptor residues and water molecules. In the equations below there is no self-interaction (the sums exclude j=i). The total PIE calculated by the FMO method describes the stability of the receptor-ligand complex. This stability correlates to, but is not the same as, the binding energy19. The difference lies in the polarization factors - the ligand is polarized by the protein and vice versa17. Based on previous reports9, we consider any interaction with an absolute PIE greater than or equal to 3.0 kcal/mol to be significant. int
The differences between PIElig1 and PIElig2 ( ∆∆Elig1,lig 2 ), and the corresponding component energy terms, can be calculated using Equation 2.
int es ex ct di ∆∆Elig 1,lig 2 = ∆∆Elig1,lig 2 + ∆∆Elig1,lig 2 + ∆∆Elig1,lig 2 + ∆∆Elig1,lig 2
Kaur, M.; Bahia, M. S.; Silakari, O., Inhibitors of Interleukin-2 Inducible T-Cell Kinase
as Potential Therapeutic Candidates for the Treatment of Various Inflammatory Disease Conditions. Eur. J. Pharm. Sci. 2012, 47, 574-588. 27.
Trani, G.; Barker, J. J.; Bromidge, S. M.; Brookfield, F. A.; Burch, J. D.; Chen, Y.;
Eigenbrot, C.; Heifetz, A.; Ismaili, M. H.; Johnson, A.; Krulle, T. M.; MacKinnon, C. H.; Maghames, R.; McEwan, P. A.; Montalbetti, C. A.; Ortwine, D. F.; Perez-Fuertes, Y.; Vaidya, D. G.; Wang, X.; Zarrin, A. A.; Pei, Z., Design, Synthesis and Structure-Activity Relationships of a Novel Class of Sulfonylpyridine Inhibitors of Interleukin-2 Inducible T-Cell Kinase (Itk). Bioorg. Med. Chem. Lett. 2014, 24, 5818-5823. 28.
Pastor, R. M.; Burch, J. D.; Magnuson, S.; Ortwine, D. F.; Chen, Y.; De La Torre, K.;
Ding, X.; Eigenbrot, C.; Johnson, A.; Liimatta, M.; Liu, Y.; Shia, S.; Wang, X.; Wu, L. C.; Pei, Z., Discovery and Optimization of Indazoles as Potent and Selective Interleukin-2 Inducible T Cell Kinase (ITK) Inhibitors. Bioorg. Med. Chem. Lett. 2014, 24, 2448-2452. 29.
MacKinnon, C. H.; Lau, K.; Burch, J. D.; Chen, Y.; Dines, J.; Ding, X.; Eigenbrot, C.;
Heifetz, A.; Jaochico, A.; Johnson, A.; Kraemer, J.; Kruger, S.; Krulle, T. M.; Liimatta, M.; Ly, J.; Maghames, R.; Montalbetti, C. A.; Ortwine, D. F.; Perez-Fuertes, Y.; Shia, S.; Stein, D. B.; Trani, G.; Vaidya, D. G.; Wang, X.; Bromidge, S. M.; Wu, L. C.; Pei, Z., Structure-Based Design and Synthesis of Potent Benzothiazole Inhibitors of Interleukin-2 Inducible T Cell Kinase (ITK). Bioorg. Med. Chem. Lett. 2013, 23, 6331-6335. 30.
M. H.; Lau, K.; Lin, Z.; Ortwine, D. F.; Zarrin, A. A.; McEwan, P. A.; Barker, J. J.; Ellebrandt, C.; Kordt, D.; Stein, D. B.; Wang, X.; Chen, Y.; Hu, B.; Xu, X.; Yuen, P. W.; Zhang, Y.; Pei, Z., Tetrahydroindazoles as Interleukin-2 Inducible T-Cell Kinase Inhibitors. Part II. SecondGeneration Analogues with Enhanced Potency, Selectivity, and Pharmacodynamic Modulation in Vivo. J. Med. Chem. 2015, 58, 3806-3816. 31.
Burch, J. D.; Lau, K.; Barker, J. J.; Brookfield, F.; Chen, Y.; Chen, Y.; Eigenbrot, C.;
Ellebrandt, C.; Ismaili, M. H.; Johnson, A.; Kordt, D.; MacKinnon, C. H.; McEwan, P. A.; Ortwine, D. F.; Stein, D. B.; Wang, X.; Winkler, D.; Yuen, P. W.; Zhang, Y.; Zarrin, A. A.; Pei, Z., Property- and Structure-Guided Discovery of a Tetrahydroindazole Series of Interleukin-2 Inducible T-Cell Kinase Inhibitors. J. Med. Chem. 2014, 57, 5714-5727. 32.
Feng, J. A.; Aliagas, I.; Bergeron, P.; Blaney, J. M.; Bradley, E. K.; Koehler, M. F.;
Lee, M. L.; Ortwine, D. F.; Tsui, V.; Wu, J.; Gobbi, A., An Integrated Suite of Modeling Tools That Empower Scientists in Structure- and Property-Based Drug Design. J. Comput.-Aided Mol. Des. 2015, 29, 511-523. 33.
Fedorov, D. G.; Kitaura, K., Energy Decomposition Analysis in Solution Based on the
Fragment Molecular Orbital Method. J. Phys. Chem. A 2012, 116, 704-719.
Li, H.; Fedorov, D. G.; Nagata, T.; Kitaura, K.; Jensen, J. H.; Gordon, M. S., Energy
Gradients in Combined Fragment Molecular Orbital and Polarizable Continuum Model (FMO/PCM) Calculation. J. Comput. Chem. 2010, 31, 778-790. 35.
Labute, P., Protonate3D: Assignment of Ionization States and Hydrogen Coordinates
to Macromolecular Structures. Proteins 2009, 75, 187-205. 36.
Kutach, A. K.; Villaseñor, A. G.; Lam, D.; Belunis, C.; Janson, C.; Lok, S.; Hong, L.-
N.; Liu, C.-M.; Deval, J.; Novak, T. J.; Barnett, J. W.; Chu, W.; Shaw, D.; Kuglstatter, A., Crystal Structures of Il-2-Inducible T Cell Kinase Complexed with Inhibitors: Insights into Rational Drug Design and Activity Regulation. Chem. Biol. & Drug Des. 2010, 76, 154-163. 37.
Halgren, T. A., Merck Molecular Force Field. I. Basis, Form, Scope,
Parameterization, and Performance of MMFF94. J. Comput. Chem. 1996, 17, 490-519. 38.
Charrier, J. D.; Miller, A.; Kay, D. P.; Brenchley, G.; Twin, H. C.; Collier, P. N.;
Ramaya, S.; Keily, S. B.; Durrant, S. J.; Knegtel, R. M.; Tanner, A. J.; Brown, K.; Curnock, A. P.; Jimenez, J. M., Discovery and Structure-Activity Relationship of 3-Aminopyrid-2-Ones as Potent and Selective Interleukin-2 Inducible T-Cell Kinase (ITK) Inhibitors. J. Med. Chem. 2011, 54, 2341-2350. 39.
Schmidt, M. W.; Baldridge, K. K.; Boatz, J. A.; Elbert, S. T.; Gordon, M. S.; Jensen, J.
H.; Koseki, S.; Matsunaga, N.; Nguyen, K. A.; Su, S.; Windus, T. L.; Dupuis, M.; Montgomery, J. A., General Atomic and Molecular Electronic Structure System. J. Comput. Chem. 1993, 14, 1347-1363. 40.
Yoshino, R.; Yasuo, N.; Inaoka, D. K.; Hagiwara, Y.; Ohno, K.; Orita, M.; Inoue, M.;
Shiba, T.; Harada, S.; Honma, T.; Balogun, E. O.; da Rocha, J. R.; Montanari, C. A.; Kita, K.; Sekijima, M., Pharmacophore Modeling for Anti-Chagas Drug Design Using the Fragment Molecular Orbital Method. PLoS One 2015, 10, e0125829. 41.
Hitaoka, S.; Chuman, H.; Yoshizawa, K., A Qsar Study on the Inhibition Mechanism
of Matrix Metalloproteinase-12 by Arylsulfone Analogs Based on Molecular Orbital Calculations. Org. Biomol. Chem. 2015, 13, 793-806. 42.
Fedorov, D. G.; Kitaura, K., Second Order Moller-Plesset Perturbation Theory Based
Upon the Fragment Molecular Orbital Method. J. Chem. Phys. 2004, 121, 2483-2490. 43.
Davenport, A. J.; Moller, C.; Heifetz, A.; Mazanetz, M. P.; Law, R. J.; Ebneth, A.;
Gemkow, M. J., Using Electrophysiology and in Silico Three-Dimensional Modeling to Reduce Human Ether-a-Go-Go Related Gene K(+) Channel Inhibition in a Histamine H3 Receptor Antagonist Program. Assay Drug Dev. Technol. 2010, 8, 781-789. 44.
Xu, L.; Sun, H.; Li, Y.; Wang, J.; Hou, T., Assessing the Performance of MM/PBSA
and Mm/GBSA Methods. 3. The Impact of Force Fields and Ligand Charge Models. J. Phys. Chem. B 2013, 117, 8408-8421.
Figure 1. Workflow for PIEDA calculations and details on each of the PIE terms that are computed.15 The electrostatic component arises from the Coulomb interaction between polarized charge distributions of fragments. The dispersion term arises from the interaction between instantaneous dipole moments of two fragments. It is hydrophobic (non-polar) in nature and is obtained in PIEDA from the correlation energy of electrons. The charge transfer term arises from the interaction between occupied orbitals of a donor and unoccupied orbitals of an acceptor. The exchange repulsion term is derived from the interaction between fragments situated in close proximity and is always repulsive; it is due to the Pauli repulsion and is related to the overlap of two occupied orbitals.
Figure 2. Correlation between experimentally measured affinity (ITK pKi) and FMO calculated PIE (total ∆Eint) for 3 series: SAP (A), BZT (B) and IND (C). Experimental data taken from Supporting Information Tables S1-S3.
Figure 3. FMO results for SAP: Compounds 1j (A), 1q (B) and 1r (C). The carbon atoms of the ligand are shown in light orange and those of the receptor according to PIE values calculated by FMO (the scale is from dark green for the strongest interactions to white for the weakest ones). The key interactions according to FMO calculations are marked as yellow dashed lines. The left-hand bar plots describe the PIE of the most significant residues, and the right-hand plots describe the PIEDA of these key interactions. PIE terms: electrostatics, dispersion, charge-transfer, and exchange-repulsion are color-coded in yellow, blue, red, and green, respectively. We consider any interaction with an absolute PIE greater than or equal to 3.0 kcal/mol to be significant; these are shown in the plots. The plots that contain sorted PIE and PIEDA contributions of all binding site residues can be found at Supporting Information Figure S3. To be consistent across plots, the interaction of -2.7 kcal/mol with C442 was included in the plot for 1j.
Figure 5. Binding mode of 1r within the ITK ATP site, which is shown as pink surface. The ITK ribbon is color coded as follows. The hinge region is yellow; P-loop, green; DFG motif, cyan; and activation loop, red. The carbon atoms of 1r are orange and those of ITK are grey. Asp500 and Phe435 are shown as thick grey lines. The distance between the CO2 oxygen of Asp500 and the closest cyclopentyl carbon within 1r (4.32 Å) is shown as a dashed green line. The portion of the electrostatic potential (ESP) induced in the vicinity of the cyclopentane is shown as red (negative) and blue (positive) surfaces. The ESP was calculated for the whole system by solving the non-linear Poisson-Boltzmann equation43 using AM1-BCC partial charges44.
Figure 6. FMO results for the BZT series: compounds 2d (A), 2p (B) and 2q (C). The carbon atoms of the ligand are shown in light orange and those of the receptor according to PIE values calculated by FMO (the scale is from dark green for the strongest interactions to white for the weakest ones). Key interactions as calculated by FMO are marked as yellow dashed lines. The left-hand bar plots describe the PIE of the most significant residues, and the right-hand plots describe the PIEDA of these key interactions. PIE terms: electrostatics, dispersion, charge-transfer, and exchange-repulsion are color-coded in yellow, blue, red, and green, respectively. We consider any interaction with an absolute PIE greater than or equal to 3.0 kcal/mol to be significant; these are shown in the plots. The plots that contain sorted PIE and PIEDA contributions of all binding site residues can be found at Supporting Information Figure S4.
Figure 7. FMO results for IND: Compounds 3g (A), 3i (B) and 3k (C). The carbon atoms of the ligand are shown in light orange and those of the receptor according to PIE values calculated by FMO. Key interactions as calculated by FMO are marked as yellow dashed lines. The left-hand bar plots describe the PIE of the most significant residues, and the righthand plots describe the PIEDA of these key interactions. PIE terms: electrostatics, dispersion, charge-transfer, and exchange-repulsion are color-coded in yellow, blue, red, and green, respectively. We consider any interaction with an absolute PIE greater than or equal to 3.0 kcal/mol to be significant; these are shown in the plots. The plots that contain sorted PIE and PIEDA contributions of all binding site residues can be found at Supporting Information Figure S5.
33 ACS Paragon Plus Environment
Journal Ki = 460 nM of Medicinal Chemistry 1 2 3 4 5 6 7 8 9
Electrostatic (ΔEes) Forces between point charges, permanent and induced.
Dispersion(ΔEdi) Interaction forces due to instantaneous polarization multipoles caused by the movement of electrons in nearby molecules. Charge transfer (ΔEct) Interactions between an occupied orbital of a donor and an unoccupied orbital of an acceptor. Orbital energy gap and overlap are important factors.
Exhange repulsion (ΔEex) Repulsive forces between molecules that are close together, mainly due to the overlap of occupied orbitals.