Application of Mechanistic Multiparameter Optimization and Large-Scale In Vitro to In Vivo Pharmacokinetics Correlations to Small-Molecule Therapeutic Projects.

Computational chemistry and machine learning are used in drug discovery to predict the target-specific and pharmacokinetic properties of molecules. Multiparameter optimization (MPO) functions are used to summarize multiple properties into a single score, aiding compound prioritization. However, over-reliance on subjective MPO functions risks reinforcing human bias. Mechanistic modeling approaches based on physiological relevance can be adapted to meet different potential key objectives of the project (e.g., minimizing dose, maximizing safety margins, and/or minimizing drug-drug interaction risk) while retaining the same underlying model structure. The current work incorporates recent approaches to predict in vivo pharmacokinetic (PK) properties and validates in vitro to in vivo correlation analysis to support mechanistic PK MPO. Examples of use and impact in small-molecule drug discovery projects are provided. Overall, the mechanistic MPO identifies 83% of the compounds considered as short-listed for clinical experiments in the top second percentile, and 100% in the top 10th percentile, resulting in an area under the receiver operating characteristic curve (AUCROC) > 0.95. In addition, the MPO score successfully recapitulates the chronological progression of the optimization process across different scaffolds. Finally, the MPO scores for compounds characterized in pharmacokinetics experiments are markedly higher compared with the rest of the compounds being synthesized, highlighting the potential of this tool to reduce the reliance on in vivo testing for compound screening.

PepSeA: Peptide Sequence Alignment and Visualization Tools to Enable Lead Optimization.

Therapeutic peptides offer potential advantages over small molecules in terms of selectivity, affinity, and their ability to target "undruggable" proteins that are associated with a wide range of pathologies. Despite their importance, current molecular design capabilities that inform medicinal chemistry decisions on peptide programs are limited. More specifically, there are unmet needs for structure-activity relationship (SAR) analysis and visualization of linear, cyclic, and cross-linked peptides containing non-natural motifs, which are widely used in drug discovery. To bridge this gap, we developed PepSeA (Peptide Sequence Alignment and Visualization), an open-source, freely available package of sequence-based tools (https://github.com/Merck/PepSeA). PepSeA enables multiple sequence alignment of non-natural amino acids and enhanced visualization with the hierarchical editing language for macromolecules (HELM). Via stepwise SAR analysis of a ChEMBL peptide data set, we demonstrate the utility of PepSeA to accelerate decision making in lead optimization campaigns in pharmaceutical setting. PepSeA represents an initial attempt to expand cheminformatics capabilities for therapeutic peptides and to enable rapid and more efficient design-make-test cycles.

Enhancing Retrosynthetic Reaction Prediction with Deep Learning Using Multiscale Reaction Classification.

Chemical synthesis planning is a key aspect in many fields of chemistry, especially drug discovery. Recent implementations of machine learning and artificial intelligence techniques for retrosynthetic analysis have shown great potential to improve computational methods for synthesis planning. Herein, we present a multiscale, data-driven approach for retrosynthetic analysis with deep highway networks (DHN). We automatically extracted reaction rules (i.e., ways in which a molecule is produced) from a data set consisting of chemical reactions derived from U.S. patents. We performed the retrosynthetic reaction prediction task in two steps: first, we built a DHN model to predict which group of reactions (consisting of chemically similar reaction rules) was employed to produce a molecule. Once a reaction group was identified, a DHN trained on the subset of reactions within the identified reaction group, was employed to predict the transformation rule used to produce a molecule. To validate our approach, we predicted the first retrosynthetic reaction step for 40 approved drugs using our multiscale model and compared its predictive performance with a conventional model trained on all machine-extracted reaction rules employed as a control. Our multiscale approach showed a success rate of 82.9% at generating valid reactants from retrosynthetic reaction predictions. Comparatively, the control model trained on all machine-extracted reaction rules yielded a success rate of 58.5% on the validation set of 40 pharmaceutical molecules, indicating a significant statistical improvement with our approach to match known first synthetic reaction of the tested drugs in this study. While our multiscale approach was unable to outperform state-of-the-art rule-based systems curated by expert chemists, multiscale classification represents a marked enhancement in retrosynthetic analysis and can be easily adapted for use in a range of artificial intelligence strategies.

Drug-Drug Interactions between Atorvastatin and Dronedarone Mediated by Monomeric CYP3A4.

Heterotropic interactions between atorvastatin (ARVS) and dronedarone (DND) have been deciphered using global analysis of the results of binding and turnover experiments for pure drugs and their mixtures. The in vivo presence of atorvastatin lactone (ARVL) was explicitly taken into account by using pure ARVL in analogous experiments. Both ARVL and ARVS inhibit DND binding and metabolism, while a significantly higher affinity of CYP3A4 for ARVL makes the latter the main modulator of activity (effector) in this system. Molecular dynamics simulations reveal significantly different modes of interactions of DND and ARVL with the substrate binding pocket and with a peripheral allosteric site. Interactions of both substrates with residues F213 and F219 at the allosteric site play a critical role in the communication of conformational changes induced by effector binding to productive binding of the substrate at the catalytic site.

Coupling X-Ray Reflectivity and In Silico Binding to Yield Dynamics of Membrane Recognition by Tim1.

The dynamic nature of lipid membranes presents significant challenges with respect to understanding the molecular basis of protein/membrane interactions. Consequently, there is relatively little known about the structural mechanisms by which membrane-binding proteins might distinguish subtle variations in lipid membrane composition and/or structure. We have previously developed a multidisciplinary approach that combines molecular dynamics simulation with interfacial x-ray scattering experiments to produce an atomistic model for phosphatidylserine recognition by the immune receptor Tim4. However, this approach requires a previously determined protein crystal structure in a membrane-bound conformation. Tim1, a Tim4 homolog with distinct differences in both immunological function and sensitivity to membrane composition, was crystalized in a closed-loop conformation that is unlikely to support membrane binding. Here we have used a previously described highly mobile membrane mimetic membrane in combination with a conventional lipid bilayer model to generate a membrane-bound configuration of Tim1 in silico. This refined structure provided a significantly improved fit of experimental x-ray reflectivity data. Moreover, the coupling of the x-ray reflectivity analysis with both highly mobile membrane mimetic membranes and conventional lipid bilayer molecular dynamics simulations yielded a dynamic model of phosphatidylserine membrane recognition by Tim1 with atomic-level detail. In addition to providing, to our knowledge, new insights into the molecular mechanisms that distinguish the various Tim receptors, these results demonstrate that in silico membrane-binding simulations can remove the requirement that the existing crystal structure be in the membrane-bound conformation for effective x-ray reflectivity analysis. Consequently, this refined methodology has the potential for much broader applicability with respect to defining the atomistic details of membrane-binding proteins.

Arachidonic Acid Metabolism by Human Cardiovascular CYP2J2 Is Modulated by Doxorubicin.

Doxorubicin (DOX) is a chemotherapeutic that is used in the treatment of a wide variety of cancers. However, it causes cardiotoxicity partly because of the formation of reactive oxygen species. CYP2J2 is a human cytochrome P450 that is strongly expressed in cardiomyocytes. It converts arachidonic acid (AA) into four different regioisomers of epoxyeicosatrienoic acids (EETs). Using kinetic analyses, we show that AA metabolism by CYP2J2 is modulated by DOX. We show that cytochrome P450 reductase, the redox partner of CYP2J2, metabolizes DOX to 7-deoxydoxorubicin aglycone (7-de-aDOX). This metabolite then binds to CYP2J2 and inhibits and alters the preferred site of metabolism of AA, leading to a change in the ratio of the EET regioisomers. Furthermore, molecular dynamics simulations indicate that 7-de-aDOX and AA can concurrently bind to the CYP2J2 active site to produce these changes in the site of AA metabolism. To determine if these observations are unique to DOX/7-de-aDOX, we use noncardiotoxic DOX analogues, zorubicin (ZRN) and 5-iminodaunorubicin (5-IDN). ZRN and 5-IDN inhibit CYP2J2-mediated AA metabolism but do not change the ratio of EET regioisomers. Altogether, we demonstrate that DOX and 7-de-aDOX inhibit CYP2J2-mediated AA metabolism and 7-de-aDOX binds close to the active site to alter the ratio of cardioprotective EETs. These mechanistic studies of CYP2J2 can aid in the design of new alternative DOX derivatives.

Atomic-level description of protein-lipid interactions using an accelerated membrane model.

Peripheral membrane proteins are structurally diverse proteins that are involved in fundamental cellular processes. Their activity of these proteins is frequently modulated through their interaction with cellular membranes, and as a result techniques to study the interfacial interaction between peripheral proteins and the membrane are in high demand. Due to the fluid nature of the membrane and the reversibility of protein-membrane interactions, the experimental study of these systems remains a challenging task. Molecular dynamics simulations offer a suitable approach to study protein-lipid interactions; however, the slow dynamics of the lipids often prevents sufficient sampling of specific membrane-protein interactions in atomistic simulations. To increase lipid dynamics while preserving the atomistic detail of protein-lipid interactions, in the highly mobile membrane-mimetic (HMMM) model the membrane core is replaced by an organic solvent, while short-tailed lipids provide a nearly complete representation of natural lipids at the organic solvent/water interface. Here, we present a brief introduction and a summary of recent applications of the HMMM to study different membrane proteins, complementing the experimental characterization of the presented systems, and we offer a perspective of future applications of the HMMM to study other classes of membrane proteins. This article is part of a Special Issue entitled: Membrane proteins edited by J.C. Gumbart and Sergei Noskov.

The cellular membrane as a mediator for small molecule interaction with membrane proteins.

The cellular membrane constitutes the first element that encounters a wide variety of molecular species to which a cell might be exposed. Hosting a large number of structurally and functionally diverse proteins associated with this key metabolic compartment, the membrane not only directly controls the traffic of various molecules in and out of the cell, it also participates in such diverse and important processes as signal transduction and chemical processing of incoming molecular species. In this article, we present a number of cases where details of interaction of small molecular species such as drugs with the membrane, which are often experimentally inaccessible, have been studied using advanced molecular simulation techniques. We have selected systems in which partitioning of the small molecule with the membrane constitutes a key step for its final biological function, often binding to and interacting with a protein associated with the membrane. These examples demonstrate that membrane partitioning is not only important for the overall distribution of drugs and other small molecules into different compartments of the body, it may also play a key role in determining the efficiency and the mode of interaction of the drug with its target protein. This article is part of a Special Issue entitled: Biosimulations edited by Ilpo Vattulainen and Tomasz Róg.

Asymmetric Binding and Metabolism of Polyunsaturated Fatty Acids (PUFAs) by CYP2J2 Epoxygenase.

Cytochrome P450 (CYP) 2J2 is the primary epoxygenase in the heart and is responsible for the epoxidation of arachidonic acid (AA), an ω-6 polyunsaturated fatty acid (PUFA), into anti-inflammatory epoxide metabolites. It also epoxidizes other PUFAs such as docosahexaenoic acid (DHA), linoleic acid (LA), and eicosapentaenoic acid (EPA). Herein, we have performed detailed thermodynamic and kinetic analyses to determine how DHA, LA, and EPA modulate the metabolism of AA by CYP2J2. We use the Nanodisc system to stabilize CYP2J2 and its redox partner, CYP reductase (CPR). We observe that DHA strongly inhibits CYP2J2-mediated AA metabolism, LA only moderately inhibits AA metabolism, and EPA exhibits insignificant inhibition. We also characterized the binding of these molecules using ebastine competitive binding assays and show that DHA binds significantly tighter to CYP2J2 than AA, EPA, or LA. Furthermore, we utilize a combined approach of molecular dynamics (MD) simulations and docking to predict key residues mediating the tight binding of DHA. We show that although all the tested fatty acids form similar contacts to the active site residues, the affinity of DHA for CYP2J2 is tighter because of the interaction of DHA with residues Arg-321, Thr-318, and Ser-493. To demonstrate the importance of these residues in binding, we mutated these residues to make two mutant variants, CYP2J2-T318A and CYP2J2-T318V/S493A. Both mutant variants showed weaker binding than the wild type (WT) to DHA and AA; DHA inhibition of AA was also mitigated in the mutants compared to the WT. Therefore, using a combined experimental and MD simulation approach, we establish that CYP2J2 inhibition of AA metabolism by DHA, EPA, and LA is asymmetric because of tighter binding of DHA to select residues in the active site.

Incorporation of charged residues in the CYP2J2 F-G loop disrupts CYP2J2-lipid bilayer interactions.

CYP2J2 epoxygenase is an extrahepatic, membrane bound cytochrome P450 (CYP) that is primarily found in the heart and mediates endogenous fatty acid metabolism. CYP2J2 interacts with membranes through an N-terminal anchor and various non-contiguous hydrophobic residues. The molecular details of the motifs that mediate membrane interactions are complex and not fully understood. To gain better insights of these complex protein-lipid interactions, we employed molecular dynamics (MD) simulations using a highly mobile membrane mimetic (HMMM) model that enabled multiple independent spontaneous membrane binding events to be captured. Simulations revealed that CYP2J2 engages with the membrane at the F-G loop through hydrophobic residues Trp-235, Ille-236, and Phe-239. To explore the role of these residues, three F-G loop mutants were modeled from the truncated CYP2J2 construct (Δ34) which included Δ34-I236D, Δ34-F239H and Δ34-I236D/F239H. Using the HMMM coordinates of CYP2J2, the simulations were extended to a full POPC membrane which showed a significant decrease in the depth of insertion for each of the F-G loop mutants. The CYP2J2 F-G loop mutants were expressed in E. coli and were shown to be localized to the cytosolic fraction at a greater percentage relative to construct Δ34. Notably, the functional data demonstrated that the double mutant, Δ34-I236D/F239H, maintained native-like enzymatic activity. The membrane insertion characteristics were examined by monitoring CYP2J2 Trp-quenching fluorescence spectroscopy upon binding nanodiscs containing pyrene phospholipids. Relative to the Δ34 construct, the F-G loop mutants exhibited lower Trp quenching and membrane insertion. Taken together, the results suggest that the mutants exhibit a different membrane topology in agreement with the MD simulations and provide important evidence towards the involvement of key residues in the F-G loop of CYP2J2.

Mechanism of drug-drug interactions mediated by human cytochrome P450 CYP3A4 monomer.

Using Nanodiscs, we quantitate the heterotropic interaction between two different drugs mediated by monomeric CYP3A4 incorporated into a nativelike membrane environment. The mechanism of this interaction is deciphered by global analysis of multiple-turnover experiments performed under identical conditions using the pure substrates progesterone (PGS) and carbamazepine (CBZ) and their mixtures. Activation of CBZ epoxidation and simultaneous inhibition of PGS hydroxylation are measured and quantitated through differences in their respective affinities for both a remote allosteric site and the productive catalytic site near the heme iron. Preferred binding of PGS at the allosteric site and a stronger preference for CBZ binding at the productive site give rise to a nontrivial drug-drug interaction. Molecular dynamics simulations indicate functionally important conformational changes caused by PGS binding at the allosteric site and by two CBZ molecules positioned inside the substrate binding pocket. Structural changes involving Phe-213, Phe-219, and Phe-241 are thought to be responsible for the observed synergetic effects and positive allosteric interactions between these two substrates. Such a mechanism is likely of general relevance to the mutual heterotropic effects caused by biologically active compounds that exhibit different patterns of interaction with the distinct allosteric and productive sites of CYP3A4, as well as other xenobiotic metabolizing cytochromes P450 that are also involved in drug-drug interactions. Importantly, this work demonstrates that a monomeric CYP3A4 can display the full spectrum of activation and cooperative effects that are observed in hepatic membranes.

Efficient Exploration of Membrane-Associated Phenomena at Atomic Resolution.

Biological membranes constitute a critical component in all living cells. In addition to providing a conducive environment to a wide range of cellular processes, including transport and signaling, mounting evidence has established active participation of specific lipids in modulating membrane protein function through various mechanisms. Understanding lipid-protein interactions underlying these mechanisms at a sufficiently high resolution has proven extremely challenging, partly due to the semi-fluid nature of the membrane. In order to address this challenge computationally, multiple methods have been developed, including an alternative membrane representation termed highly mobile membrane mimetic (HMMM) in which lateral lipid diffusion has been significantly enhanced without compromising atomic details. The model allows for efficient sampling of lipid-protein interactions at atomic resolution, thereby significantly enhancing the effectiveness of molecular dynamics simulations in capturing membrane-associated phenomena. In this review, after providing an overview of HMMM model development, we will describe briefly successful application of the model to study a variety of membrane processes, including lipid-dependent binding and insertion of peripheral proteins, the mechanism of phospholipid insertion into lipid bilayers, and characterization of optimal tilt angle of transmembrane helices. We conclude with practical recommendations for proper usage of the model in simulation studies of membrane processes.

Balanced Permeability Index: A Multiparameter Index for Improved In Vitro Permeability.

The optimization of passive permeability is a key objective for orally available small molecule drug candidates. For drugs targeting the central nervous system (CNS), minimizing P-gp-mediated efflux is an additional important target for optimization. The physicochemical properties most strongly associated with high passive permeability and lower P-gp efflux are size, polarity, and lipophilicity. In this study, a new metric called the Balanced Permeability Index (BPI) was developed that combines these three properties. The BPI was found to be more effective than any single property in classifying molecules based on their permeability and efflux across a diverse range of chemicals and assays. BPI is easy to understand, allowing researchers to make decisions about which properties to prioritize during the drug development process.

Characterizing the membrane-bound state of cytochrome P450 3A4: structure, depth of insertion, and orientation.

Cytochrome P450 3A4 (CYP3A4) is the most abundant membrane-associated isoform of the P450 family in humans and is responsible for biotransformation of more than 50% of drugs metabolized in the body. Despite the large number of crystallographic structures available for CYP3A4, no structural information for its membrane-bound state at an atomic level is available. In order to characterize binding, depth of insertion, membrane orientation, and lipid interactions of CYP3A4, we have employed a combined experimental and simulation approach in this study. Taking advantage of a novel membrane representation, highly mobile membrane mimetic (HMMM), with enhanced lipid mobility and dynamics, we have been able to capture spontaneous binding and insertion of the globular domain of the enzyme into the membrane in multiple independent, unbiased simulations. Despite different initial orientations and positions of the protein in solution, all the simulations converged into the same membrane-bound configuration with regard to both the depth of membrane insertion and the orientation of the enzyme on the surface of the membrane. In tandem, linear dichroism measurements performed on CYP3A4 bound to Nanodisc membranes were used to characterize the orientation of the enzyme in its membrane-bound form experimentally. The heme tilt angles measured experimentally are in close agreement with those calculated for the membrane-bound structures resulted from the simulations, thereby verifying the validity of the developed model. Membrane binding of the globular domain in CYP3A4, which appears to be independent of the presence of the transmembrane helix of the full-length enzyme, significantly reshapes the protein at the membrane interface, causing conformational changes relevant to access tunnels leading to the active site of the enzyme.

Anti-inflammatory dopamine- and serotonin-based endocannabinoid epoxides reciprocally regulate cannabinoid receptors and the TRPV1 channel.

The endocannabinoid system is a promising target to mitigate pain as the endocannabinoids are endogenous ligands of the pain-mediating receptors-cannabinoid receptors 1 and 2 (CB1 and CB2) and TRPV1. Herein, we report on a class of lipids formed by the epoxidation of N-arachidonoyl-dopamine (NADA) and N-arachidonoyl-serotonin (NA5HT) by epoxygenases. EpoNADA and epoNA5HT are dual-functional rheostat modulators of the endocannabinoid-TRPV1 axis. EpoNADA and epoNA5HT are stronger modulators of TRPV1 than either NADA or NA5HT, and epoNA5HT displays a significantly stronger inhibition on TRPV1-mediated responses in primary afferent neurons. Moreover, epoNA5HT is a full CB1 agonist. These epoxides reduce the pro-inflammatory biomarkers IL-6, IL-1ß, TNF-α and nitrous oxide and raise anti-inflammatory IL-10 cytokine in activated microglial cells. The epoxides are spontaneously generated by activated microglia cells and their formation is potentiated in the presence of anandamide. Detailed kinetics and molecular dynamics simulation studies provide evidence for this potentiation using the epoxygenase human CYP2J2. Taken together, inflammation leads to an increase in the metabolism of NADA, NA5HT and other eCBs by epoxygenases to form the corresponding epoxides. The epoxide metabolites are bioactive lipids that are potent, multi-faceted molecules, capable of influencing the activity of CB1, CB2 and TRPV1 receptors.

Endothelial ERK1/2 signaling maintains integrity of the quiescent endothelium.

To define the role of ERK1/2 signaling in the quiescent endothelium, we induced endothelial Erk2 knockout in adult Erk1-/- mice. This resulted in a rapid onset of hypertension, a decrease in eNOS expression, and an increase in endothelin-1 plasma levels, with all mice dying within 5 wk. Immunostaining and endothelial fate mapping showed a robust increase in TGFß signaling leading to widespread endothelial-to-mesenchymal transition (EndMT). Fibrosis affecting the cardiac conduction system was responsible for the universal lethality in these mice. Other findings included renal endotheliosis, loss of fenestrated endothelia in endocrine organs, and hemorrhages. An ensemble computational intelligence strategy, comprising deep learning and probabilistic programing of RNA-seq data, causally linked the loss of ERK1/2 in HUVECs in vitro to activation of TGFß signaling, EndMT, suppression of eNOS, and induction of endothelin-1 expression. All in silico predictions were verified in vitro and in vivo. In summary, these data establish the key role played by ERK1/2 signaling in the maintenance of vascular normalcy.

Dimeric structure of the uracil:proton symporter UraA provides mechanistic insights into the SLC4/23/26 transporters.

The Escherichia coli uracil:proton symporter UraA is a prototypical member of the nucleobase/ascorbate transporter (NAT) or nucleobase/cation symporter 2 (NCS2) family, which corresponds to the human solute carrier family SLC23. UraA consists of 14 transmembrane segments (TMs) that are organized into two distinct domains, the core domain and the gate domain, a structural fold that is also shared by the SLC4 and SLC26 transporters. Here we present the crystal structure of UraA bound to uracil in an occluded state at 2.5 Å resolution. Structural comparison with the previously reported inward-open UraA reveals pronounced relative motions between the core domain and the gate domain as well as intra-domain rearrangement of the gate domain. The occluded UraA forms a dimer in the structure wherein the gate domains are sandwiched by two core domains. In vitro and in vivo biochemical characterizations show that UraA is at equilibrium between dimer and monomer in all tested detergent micelles, while dimer formation is necessary for the transport activity. Structural comparison between the dimeric UraA and the recently reported inward-facing dimeric UapA provides important insight into the transport mechanism of SLC23 transporters.

Capturing Spontaneous Membrane Insertion of the Influenza Virus Hemagglutinin Fusion Peptide.

Hemagglutinin (HA) is a protein located on the surface of the influenza virus that mediates viral fusion to the host cellular membrane. During the fusion process the HA fusion peptide (HAfp), formed by the first 23 N-terminal residues of HA and structurally characterized by two alpha helices (Helix A and Helix B) tightly packed in a hairpin-like arrangement, is the only part of the virus in direct contact with the host membrane. After encountering the host cell, HAfp is believed to insert into the membrane, thereby initiating the fusion of the viral and host membranes. Detailed characterization of the interactions between the HAfp and cellular membrane is therefore of high relevance to the mechanism of viral entry into the host cell. Employing HMMM membrane representation with enhanced lipid mobility, we have performed a large set of independent simulations of unbiased membrane binding of HAfp. We have been able to capture spontaneous binding and insertion of HAfp consistently in nearly all the simulations. A reproducible membrane-bound configuration emerges from these simulations, despite employing a diverse set of initial configurations. Extension of several of the simulations into full membrane systems confirms the stability of the membrane-bound form obtained from HMMM binding simulations. The resulting model allows for the characterization of important interactions between the peptide and the membrane and the details of the binding process of the peptide for the first time. Upon membrane binding, Helix A inserts much deeper into the membrane than Helix B, suggesting that the former is responsible for hydrophobic anchoring of the peptide into the membrane. Helix B, in contrast, is found to establish major amphipathic interactions at the interfacial region thereby contributing to binding strength of HAfp.

Changing the peptide specificity of a human T-cell receptor by directed evolution.

Binding of a T-cell receptor (TCR) to a peptide/major histocompatibility complex is the key interaction involved in antigen specificity of T cells. The recognition involves up to six complementarity determining regions (CDR) of the TCR. Efforts to examine the structural basis of these interactions and to exploit them in adoptive T-cell therapies has required the isolation of specific T-cell clones and their clonotypic TCRs. Here we describe a strategy using in vitro-directed evolution of a single TCR to change its peptide specificity, thereby avoiding the need to isolate T-cell clones. The human TCR A6, which recognizes the viral peptide Tax/HLA-A2, was converted to TCR variants that recognized the cancer peptide MART1/HLA-A2. Mutational studies and molecular dynamics simulations identified CDR residues that were predicted to be important in the specificity switch. Thus, in vitro engineering strategies alone can be used to discover TCRs with desired specificities.