Systematic Approach to Conformational Sampling for Assigning

Article pubs.acs.org/jmc

Systematic Approach to Conformational Sampling for Assigning Absolute Configuration Using Vibrational Circular Dichroism Edward C. Sherer,* Claire H. Lee, Joseph Shpungin, James F. Cuff, Chenxiao Da, Richard Ball, Richard Bach, Alejandro Crespo, Xiaoyi Gong, and Christopher J. Welch Merck Research Laboratories, Merck & Co., Inc., PO Box 2000, Rahway, New Jersey 07065, United States S Supporting Information *

ABSTRACT: Systematic methods that speed-up the assignment of absolute configuration using vibrational circular dichrosim (VCD) and simplify its usage will advance this technique into a robust platform technology. Applying VCD to pharmaceutically relevant compounds has been handled in an ad hoc fashion, relying on fragment analysis and technical shortcuts to reduce the computational time required. We leverage a large computational infrastructure to provide adequate conformational exploration which enables an accurate assignment of absolute configuration. We describe a systematic approach for rapid calculation of VCD/IR spectra and comparison with corresponding measured spectra and apply this approach to assign the correct stereochemistry of nine test cases. We suggest moving away from the fragment approach when making VCD assignments. In addition to enabling faster and more reliable VCD assignments of absolute configuration, the ability to rapidly explore conformational space and sample conformations of complex molecules will have applicability in other areas of drug discovery.

1. INTRODUCTION

molecule (compound 9 from He et al.) would be considered druglike, exhibiting significant conformational flexibility.8 While a fragment-based approach for assigning absolute configuration is suitable for some pharmaceuticals,17 this is by no means universal. A recent perspective on VCD in the pharmaceutical environment suggested that experimental spectra be measured at a lower temperature; this would allow the consideration of fewer conformations on the computational side, thus reducing computational cost.3 Minick et al. propose relying on a combination of the fragment approach and the comparative approach, which matches measured spectra of a series of optical isomers to the calculated VCD of a reference compound containing the same chiral core.17 Minick et al. describe the significant increase in absolute configuration assignments made per year as a function of the number of computer processors.17 The work presented below supports this assertion that parallelization is key to performing VCD calculations in an acceptable time frame. As available computation power continues to grow, and as modeling algorithms continue to evolve, the once daunting task of modeling conformationally complex pharmaceuticals is expected to become more straightforward. In current practice for modeling conformationally complex molecules, the process of selecting which conformers to include in the simulations (i.e., proper Boltzmann-weighted percentages) remains largely ad hoc, requiring the input of skilled and experienced experts. To

Chirality is a structural feature of many marketed drugs and an important focus of drug discovery teams in the pharmaceutical industry. Owing to the difficulty in assigning absolute configuration (AC), several methods are commonly employed, including single crystal X-ray diffraction, NMR combined with chiral derivatization, electronic circular dichroism (ECD), optical rotation, and structural proofs based on synthetic transformations.1−3 A relatively new technique for assigning absolute configuration,3−17 vibrational circular dichroism (VCD), has gained favor in recent years. Recent perspectives on the method emphasize that VCD can play a critical role for SAR development and provide insight into the dominant conformations of a molecule in solution.3,17 It has been emphasized that uptake of VCD into the pharmaceutical industry is still surprisingly slow.17 The technique in question relies on the comparison of measured and calculated VCD spectra. Accurate calculation of a VCD spectrum requires knowing the accessible conformations of a molecule, and early applications of the technique were largely confined to easily modeled, conformationally rigid molecules. In recent years, use of the VCD approach has been expanding to encompass systems of increasing conformational complexity.17−28 Nevertheless, a fragment or subunit approach is still often recommended for determination of absolute configuration of pharmaceuticals,6,17,26,29−31 which are often considered too difficult or computationally intensive to effectively model. Indeed, in a recent VCD review, only one © XXXX American Chemical Society

Received: October 15, 2013

A

dx.doi.org/10.1021/jm401600u | J. Med. Chem. XXXX, XXX, XXX−XXX

Journal of Medicinal Chemistry

Article

Figure 1. Structures of the compounds used with indication of known stereochemistry. instrument. Data was collected in blocks, where the instrument recorded ∼3000 scans over the course of 1 h and averaged those scans into one block. Typical runs involved averaging several blocks (at least three) for the sample and solvent. The solvent background average was then subtracted from the sample average. Collection times for sample and solvent ranged from 4 to 18 h, with the instrument optimized at 1400 cm−1. VCD spectra for camphor I and ibuprofen III were further corrected by using the respective enantiomer and racemate. Minick et al. provide a useful discussion of solvent considerations for measured VCD.17 2.3. VCD Calculations: Conformer Searching. An initial search of conformational space was performed with three conformer generation tools including the Merck in-house tools which are described in the literature but not publically available: ET32 and JG (an in-house version of DG33), and OpenEye’s OMEGA.34,35 The maximum number of conformers generated per method was set to n (n = 1500 by default), leading to a default combined possibility of 4500 initial conformers. In practice, the methodology of some tools does not nearly generate the limit; this was instead a global variable in the workflow. The ET tool generates conformers based on a set of rules which reflects chemical intuition. Flags were set for ET to allow rings to adopt boat, chair, and twist-boat conformations, a minimum distance between nonbonded heteroatoms was set to 2.0 Å, and the maximum number of output conformations was set as described above. The JG tool takes a more chemically naı̈ve approach. Atom positions are sampled in three dimensions based only on distance constraints. Flags for JG were kept as default, and the number of conformers was set as described. Conformers output from JG can necessarily have high energies so common workflows at Merck involve molecular mechanics minimization using MMFF9436 of the output conformers (convergence criteria 0.001 kJ/mol in total energy). However, we chose to keep this initial high energy “diversity” by branching the workflow to have both JG minimized and nonminimized conformers input into the output representative set. OpenEye’s OMEGA conformer generator (similar to ET in regards to chemically aware sampling) was run using default options with an RMS cutoff for output conformers set to 0.2 Å and the maximum number of conformers as described. Conformers output by ET and OMEGA are relatively similar, though the speed of OMEGA makes this generator appealing for larger

enable a broader use of the technique, a faster, less labor intensive, and more systematic approach to the mapping of conformational space is required. The current study presents VCD analysis on seven pharmaceutically relevant and complex molecules plus two classic test cases from the literature: camphor and pulegone. We show how, by leveraging the power of a large computational infrastructure, it is possible to carry out minimally redundant explorations of conformational phase space to identify the initial 50−200 molecular mechanical (MM) conformers that dominate the density functional theory (DFT) or quantum mechanical (QM) conformational landscape of typical molecules of interest in the pharmaceutical industry, thereby allowing a statistically accurate calculation of VCD spectra, which in turn can be used for robust assignment of absolute configuration. This approach aims to replace the easily biased, human-driven conformational sampling approach with a more systematic, machine-based approach.

2. EXPERIMENTAL SECTION 2.1. Molecules. Model compounds used in the study are shown in Figure 1. Compounds I−III were obtained from Sigma-Aldrich (Milwaukee, WI), and compounds IV−IX were obtained from the Merck compound collection. The absolute configuration of each compound was known prior to assignment. 2.2. VCD Measurements. (S)-(−)-Camphor I, (R)-(+)-pulegone II, (S)-(+)-ibuprofen III, efavirenz IV, and simvastatin IX samples were dissolved in CDCl3 (I: 160 mg/mL; II: 94 mg/mL; III: 74 mg/ mL; IV: 30 mg/mL; IX: 99 mg/mL). Aprepitant VII and ezetimibe VIII samples were dissolved in DMSO-d6 (i.e., DMSO) (VII: 40 mg/ mL; VIII: 28 mg/mL). Compound V and laropiprant VI samples were dissolved in both CDCl3 (V: 30 mg/mL; VI: 38 mg/mL) and DMSOd6 (V: 27 mg/mL; VI: 38 mg/mL). All experiments were performed using a 0.10-mm path length cell with BaF2 windows. The IR and VCD spectra were recorded using a ChiralIR VCD spectrometer equipped with the Dual PEM accessory (BioTools, Jupiter, FL), with 4 cm−1 resolution. A dry N2 purge was used to eliminate water from the B



Article

numbers of molecules. At the time of the initial studies, sampling of polar hydrogen dihedrals was not performed in OMEGA by design. Sampling of these rotamers is explicitly needed for identification of a complete Boltzmann population. Because a given hydroxyl rotamer will not minimize to a global minimum using DFT if it is already in a local minimum, the sampling of hydroxyl dihedrals needed to be handled using JG. 2.4. VCD Calculations: Conformational Clustering. After conformational searching, all output conformers were run through a renumbering scheme to standardize atomic labeling, and all isomers were checked for the correct output chiral configuration. The global list of conformers was then catenated to form one large conformers file. The resultant distribution was clustered based on 3D coordinates, and a subset of the total conformer pool was selected to represent the allowable range of conformations. Given that DFT/QM calculations are very expensive and the shared resource had many competing users, we needed to identify a strategy to reduce the number of DFT/QM jobs submitted while still ensuring a proper coverage of conformational space. We adopted a three-step approach. Clustering by energy is not feasible, as mechanics energies are not accurate enough and oftentimes similar conformers have identical energies. Instead, all output conformers were initially clustered, after MMFF9436 energy minimization, based on the RMS of all atoms at 0.6 Å. In other words, after superposition, if conformers differed by less than 0.6 Å, they were considered the same cluster. These clusters were then ordered according to the MMFF energies of the centroids. A set of three integers (p,q,r) controlled how this clustered file was handled, and the first integer p indicated the number of top-ranked cluster centroids to keep; for instance, 10 would keep the best 10 centroids ranked by MMFF energy. The second integer q controlled how many of the remaining clusters would be evenly sampled; for instance, 20 would evenly select 20 of the remaining higher energy clusters ordered by increasing strain energy. Owing to initial testing of the number of conformers needed, we determined that the best attempts at using MMFF to identify DFT minima was still suspect to occasional underperformance so we added in what was termed the “random factor”. This final integer r controlled a return to the original combined conformers file where the compounds were clustered without MMFF minimization of the energies. For example, an r of 45 would take the entire set of conformers and cluster them down by number until 45 centroids represented conformational diversity. This allowed conformers with very high MMFF “unminimized” energies to be included in the DFT submission. Testing of the convergence in VCD/IR matching, and sampling of conformations, was performed using a set of four parameter combinations providing total sampling of 20, 50, 100, or 200 conformers, respectively. Details of the parameters controlling conformer sampling/clustering are found in Table 1.

investigate (vide infra) the effects of implicit solvent on the conformational landscapes only when problems arise in matching to in vacuo results. Output conformers were ranked according to DFT energy, and a clustering was performed to remove duplicates. Initial identification of duplicates was performed solely on an electronic energy basis where compounds were considered identical if the difference in Hartrees was less than 0.01. Rounding the differences led to inconsistencies in identification of duplicates. It became better to cluster the DFT minima by energy (threshold for similarity being 0.01 hartree) and then recluster each energy bucket by structure using an all-atom RMS of 0.6 Å. This faithfully removed only identical compounds. Two Boltzmann distributions were calculated based on electronic energy (E) and free energy (G). The Merck computational infrastructure is composed of a Linux cluster and a Cray supercomputer. All jobs were run in parallel during the evaluations using 12 processors for a quick turnaround. Through iterative testing of the maximal number of conformers sent for DFT, it was determined that for structures of the size and flexibility of compounds V, VI, and VIII (seven to eight rotamers), the number of conformers sampled should be on the order of 150. For smaller compounds such as I−IV (two to eight rotamers), the number could be set to around 60, where rigidity would at times act as a secondary limit of conformer diversity. Compounds such as VII (nine rotamers) and IX (twelve rotamers) could take as many as 200 conformers to identify all minima. We define rotamers as a rotatable bond which gives an energetically different structure, so free rotors oftentimes not included in rotatable bond count do in fact contribute under our definition. As will be discussed, this brute force approach often led to significant redundancy in the conformers sampled, and it became a good sign to see a “duplicates” file heavily populated. This leant support to a successful exploration of conformational space, and because our computational resources allowed this extensive job submission, we took advantage. 2.6. VCD Calculations: Extraction and Matching of Spectra. Generation of VCD spectra from the Gaussian output files was performed using the ViewVCD algorithm (BioTools). This program converts the discrete frequency intensities into a smooth spectrum using Lorentzian band broadening (this program was run with default settings). The spectra were matched using an in-house matching tool based off the publication of Shen et al. or the peak matching software tool CompareVOA distributed by BioTools (run using default settings).48,49 BioTools has written into their matching algorithm CompareVOA a method for outputting a confidence from 0% to 100%. The in-house method for comparing VCD and IR spectra is based on published methodology comparing the overlap integrals of the measured and calculated spectra after optimal scaling and shifting of individual peaks.48 We used the same formulas for calculating similarity for IR and VCD spectra of experimental and observed curves. On the basis of our experience in matching the curves by hand, we introduced the following modifications to the algorithm: we scale the spectra (0 to 1 for IR, −1 to 1 for VCD) before comparing them; we isolate each peak for movement rather than groups of peaks; we isolate peaks independently for IR and VCD spectra; when looking for the best match, we move the experimental peak only to higher frequencies with a maximum shift of 20 cm−1. The default algorithm only right shifts peaks, which was the direction of travel for peaks needing to be shifted during initial testing of the matching algorithm. This right shift was in part due to solvent shifting not accounted for by in vacuo calculations and inaccuracies in the calculated frequencies themselves. Motivation for shifting peaks in only one direction was in part due to frequency scaling factors being 2000 processors. In practice, a single user within our environment is allocated only a subset of the total available processors, but we still see full conformational sampling complete in ∼1 day.

The spectra were matched initially over a range of 1000−2000 cm−1, but this range was adjusted to account for lack of features, excessive noise, or artifactual peaks (Table 2). When a spectrum

Table 2. Details of Experimental Spectra and Decisions Made for Matching name

solvent

range (cm−1)

zeroing (cm−1)

I II III IV V

camphor pulegone ibuprofen efavirenz filorexant

1000−1600 1000−1800 1000−1600 1000−1650 1000−1900

none 1062:1551 1010:1494 none 1011:1885

VI

laropiprant

1000−2000

1081:1521

VII VIII IX

aprepitant ezetimibe simvastatin

CDCl3 CDCl3 CDCl3 CDCl3 botha, match on CDCl3 botha, match on CDCl3 DMSO DMSO CDCl3

1100−1800 1150−1800 1000−2000

none 1435:1772 1227:1868

compound

a

The experiment was run using both solvents; resolution was better for chloroform.

showed evidence of drift over this range, the spectrum was zeroed. Matching of the experimental IR and VCD spectra to the calculated spectra outputs similarity metrics on a scale of 0.0 to 1.0 for IR and −1.0 to 1.0 for VCD, where more positive values are better matches. A useful statistic output by CompareVOA is the enantiomeric similarity index, or ESI. This metric is the difference in the overall spectral overlap of one configuration/enantiomer over the other (e.g., if R = 0.52 and S = −0.52, then ESI = 1.04). Though the matching metric has a maximum at 1.0 for two identical curves, our experience shows us that good IR matches can be found in the 0.5 to 0.8 range and good VCD matches show up in the 0.2 to 0.5 range (of course with a requirement for a good ESI). We have also calculated confidence percentages using CompareVOA. Table 2 provides details about the solvent used for each compound, the range for spectra matching, and any zeroing of the measured spectra. Matches output from the inhouse matching tool will be discussed while those graphical images output from CompareVOA have not been included in the manuscript (because the qualitative matches are the same, we provide only CompareVOA statistics in the data tables). 2.7. Extension of Default Calculations. If initial matching of the experimental and calculated spectra was not conclusive, additional calculations were performed. The most common causes for poor matching were impurities in the submitted samples, incorrectly drawn chemical structures upon submission of a VCD request, dimerization of the sample, or direct interaction of the sample with solvent (hydrogen bonding). A script was used to submit a set of jobs based on of a threshold contribution percentage for the in vacuo Boltzmann distribution. This default percentage was set to compounds which contributed at least 5.0% to the population, but when intramolecular hydrogen bonding was present, the default needed to be lowered to 0.1% (i.e., extreme dominance of intramolecular hydrogen bond energies). Secondary jobs included implicit solvent, higher levels of theory (MP2), different DFT methods (PBE, M062X), and larger basis sets (6-311++G**).50−53 Implicit solvent calculations in DMSO or chloroform were performed using the Minnesota Solvation Model formalism, SCRF=(SMD, solvent=DMSO or chloroform).54 Explicit solvation (complexes) with DMSO molecules or molecular dimers were minimized using the same default basis set. While structural details are not reported in this manuscript, we have found great success in modeling compounds with iodine using a hybrid basis set of the low level default on all atoms except iodine which instead is modeled using the LANL2DZ basis set with ECP (effective core potential) on iodine.55 2.8. Moving Away from Fragments: Use of Complete Structures. With the existing computational architecture, we did not need to pursue truncation of compounds down to model

3. RESULTS AND DISCUSSION To be a practical platform tool for assigning absolute configuration in the pharmaceutical industry, the VCD approach must (1) be applicable to molecules of interest, (2) allow absolute configuration to be assigned with high confidence, and (3) generate results within the time constraints of an active drug discovery optimization cycle (assignment needing to be made in a week or less). It was clear from earlier work that the VCD approach has the potential to evolve into a platform tool, and that most perceived shortcomings are fundamentally a problem of computational speed and approach strategy. We decided to address this problem by adopting variations on known computational approaches, speeding up and simplifying certain aspects, and putting it all together into a potentially useful workflow. From the computational standpoint, there are two aspects to the approach which are considered high hurdles, conformational sampling, and matching of spectra; with the first being the historical limiter. As referenced above, several studies have relied on fragments to simplify the conformational search and speed up geometry optimizations. We initially worked with rigid fragments. However, ambiguity in matching over the entire spectral range soon led us to abandon this shortcut. Having the benefit of a large computational resource, we found much better success by modeling entire compounds. With computer power becoming more readily available (inexpensive cluster farms or cloud based systems) a head-on attack of this problem is now preferable. To demonstrate the utility of this workflow, we work through case studies of nine compounds originating either from the public domain or from the Merck compound collection (Figure 1). These compounds represent a diverse range of size, flexibility, and functionality and for these reasons make up a D



Article

importance of sampling hydroxyl rotamers was made apparent during the investigation of a proprietary structure (not disclosed here), but a general description of the problem should suffice: a structure was presented for calculation containing an aromatic ring with a hydroxyl rotamer linked via a short alkyl linker and poised over the aromatic surface. In this example, the hydroxyl ended up adopting a dihedral over the aromatic surface which was located with JG but not by ET or OMEGA. The other hydroxyl rotamers were local minima and as such were not able to jump into the true DFT global minimum. Use of all three conformer generators in concert has been demonstrated to perform well. 3.2. Convergence of Boltzmann Populations. The structures of the global DFT minima for each compound are depicted in Figure 2. Coordinates of these conformations and

valuable test set for benchmarking. The high degree of structural complexity of the molecules in the test set, with some molecules containing as many as seven stereocenters, represents a significant challenge for molecular modeling, one which requires a suitable stepwise approach to identifying relevant Boltzmann populations. 3.1. Conformer Sampling. After early disappointment with application of the fragment approach to calculating VCD spectra, we realized that we could minimize full druglike molecules in a timely fashion by leveraging parallel processing. Thus, we next attacked the notoriously difficult problem of conformer sampling. When sampling of conformations is performed by hand, the identification of minima can be rather arbitrary. Earlier studies have used conformer generators (usually only one), molecular mechanics searches, Monte Carlo, or molecular dynamics-based searches of conformational space to aid in the identification of conformational minima. We choose to adopt several conformer generators and package the output with mechanics minimizations. The approach detailed in Experimental Section uses a combination of three conformer generators followed by clustering and energy minimization to provide a diverse set of starting conformers for density functional minimizations. Initially, we sought a workflow which would identify 10−20 conformers which could be input into DFT for calculation of the VCD spectra in order to minimize the amount of calculations being run on the shared-resource computer clusters. While varying the clustering strategy and exploring the conformer generators in stand-alone mode, it was determined that the ability to locate the global minima, and the number of minima identified, was highly variable at low conformer sampling rates. Several rounds of clustering were performed, incrementing the number of MMFF minima sent for DFT from 10 to 200. It was not until these larger numbers were sent to DFT that a convergence in the number of minima identified and specificity of the global minima were observed consistently. Small sampling rates may have provided the same number of minima (though not the same minima), but these minima were fairly chaotic in convergence until larger sampling rates were used. Minick et al. indicate best practice is to use between 6 and 10 conformations per study.17 This final number is in-line with the number or Boltzmann-relevant conformations identified post conformer searching. As is commonly understood, DFT and MM energies are not always in agreement qualitatively or quantitatively, so a straight selection of compounds based on the best MMFF energies was found to be inadequate. The assumption of the opposite had been the backbone of our original clustering strategy. When larger numbers of MMFF conformers were sent for DFT, it was sometimes observed that a compound from the tail end of the clusters file (ordered by increasing energy) ended up minimizing to the global DFT minimum. In these cases, the conformers which had the lowest MMFF energy did not lead to the global minimum. To avoid any unforeseen miss of the lowest energy conformer, we added in a third tier of conformer selection which clustered all compounds by number prior to MMFF minimization. With this additional step, we did not see spurious behavior moving forward. As detailed in Experimental Section, OMEGA is a relatively fast conformer generator, and for this reason we initially investigated using this tool alone. At the time, OMEGA did not sample hydroxyl rotamers, so with further investigation we choose to complement this tool with ET and JG. The

Figure 2. Structures of the global gas-phase minima of the compounds at B3LYP/6-31G**.

any dimers associated with the monomers are provided in Supporting Information, along with tables of the Boltzmannpopulation analysis in vacuo using both electronic energy and free energy. Because we observed little variation in statistical or visual matching when using either E or G, we report only analysis of the electronic energy distributions. Results of the conformational sampling over the four main target pools of 20, 50, 100, and 200 conformers (details in Table 1) can be found in Tables 3−11. Each results table presents an analysis of convergence in the identified conformational sampling as judged by the number of conformers contributing >5% to the Boltzmann distribution. The original and final IR and VCD matching statistics are presented along with the ESI, identified absolute configuration, and CompareVOA output statistics. High target conformer values, e.g., 200, were satisfied only in very flexible molecules, where small more rigid molecules obviously have far fewer initial conformers being output by the conformer generators. E



Article

nonenergy minimized diversity sampling, or fewer numbers of conformers sampled, a nonoptimal workflow did lead to some fluctuations in assigned absolute configuration (data not provided). 3.3. Matching of Spectra. Extensive work in the field of calculated VCD spectra indicates that relatively moderate basis sets combined with hybrid DFT methods are of sufficient accuracy for assigning absolute configuration.4−12 Many of the cited studies involve complex investigations which provide close matching of sign and intensity for all spectra peaks. While this detailed approach is understandable given the recent emergence of the VCD field, a less exhaustive but more practical approach is often sufficient for routine problem solving. Assignment can often be made with confidence given an acceptable level of agreement between the calculated and experimental spectra. We note that this acceptable level is still quite difficult to define quantitatively, and that simple qualitative visual examination often provides the best assessment of satisfactory matching of spectra. Combining a visual interpretation with quantitative statistical output parameters from the spectra matching tools has proven successful. In addition to statistics output from our in-house algorithm, we make use of the BioTools metric when making assignments.49 We have found that pairing the eye of a spectroscopist with that of a computational chemist provides a good system of checks and balances during the curve matching and absolute configuration assignment process. Advocating more scientific rigor on the front-end (conformational sampling) and less on the back-end (“by-eye” curve matching) does indicate that there is more room for

Table 3. Convergence in Boltzmann Sampling for Compound Ia total possible conformers 20

50

100

200

initial generatedb Boltzmann, >5% absol config ESI original IR final IR original VCD final VCD original enantiomer final enantiomer

3 1 SS 0.77 0.52 0.59 0.46 0.57 −0.46 −0.19

3 1 SS 0.77 0.52 0.59 0.46 0.57 −0.46 −0.19

3 1 SS 0.77 0.52 0.59 0.46 0.57 −0.46 −0.19

3 1 SS 0.77 0.52 0.59 0.46 0.57 −0.46 −0.19

CompareVOA ESI-CompareVOA confidence

SS 80 100

a

Gas-phase monomer unless otherwise noted. bInitial generated is the total number in the clustered file prior to Gaussian minimization.

When molecules are small and rigid, we see little variation in the ESI and assigned absolute configuration. The larger more flexible molecules exhibit increased resolution of the ESI but rarely do the assigned absolute configurations change owing to better conformer sampling. It is important to note that while none of the cases presented here exhibited such inversions, in trial versions of the script where sampling involved only one of the three conformer generators, nonoptimal clustering, lack of

Figure 3. Raw overlay of calculated (red) and measured (black) VCD spectra (A) for I, overlay of VCD and IR after peak shifting and scaling (B) for I, raw overlay of calculated (red) and measured (black) VCD spectra (C) for II, and overlay of VCD and IR after peak shifting and scaling (D) for II. Intensities scaled from −1.0 to 1.0 (0.0 to 1.0 for IR). F



Article

improvement in this process. While the evolution of such peak matching algorithms over time is expected to improve, at present the human eye is much better and much faster at making such an assignment. Therefore, we use combination of different algorithms for optimal curve alignment but require a visual match for final assignment. As detailed in the following case studies, confidence in the assignment of absolute configuration can be made in a timely fashion when balancing the level of rigor applied to conformational sampling, solvent modeling, dimerization, and curve matching. 3.4. Compound I: Camphor. Camphor (Table 3) is a drug from the public domain, and the VCD analysis has been previously published.4,56−59 Devlin et al. provide a thorough peak by peak analysis of which our results are in full agreement. We selected this compound as an initial test case because we had experimental data in the literature to compare our own experimental spectra. Additionally, the structure is small, has few rotatable bonds (none by conventional definition), and hence is very rigid. The absolute configuration of camphor is known, and both (R,R) and (S,S) enantiomers are commercially available. Both configurations were purchased and measured experimentally in CDCl3. The structure presents no confounding functionality as seen in later compounds. There is only one conformation. The four possible target conformer pools each provided three conformers for input into DFT. Each time the conformers converged to just one DFT minimum. Matching was performed over the range 1000−1600 cm−1. Details for each match (compounds I−IX) are provided in Table 2, which indicates the solvent used, whether there was baseline correction, and the range of the match. The IR match was 0.52 with a VCD ESI of 0.77 (0.57 for correct AC VCD), indicating a very good statistical match and a good match by eye (Figure 3A,B). The match done by CompareVOA agrees, assigning an ESI of 80 with a confidence of 100%. A related structure to I is (−)-camphanic acid, which has an additional carboxylic acid moiety. Buffeteau and co-workers found camphanic acid to exist as a monomer in CDCl3 at low concentrations (0.005 M) and as a dimer at higher concentrations (0.2 M). Their modeling of the monomer and dimer at a level of theory similar to that used in this report was successfully able to reproduce the two experimental results.60 3.5. Compound II: Pulegone. The AC of pulegone (Table 4) is known, and both enantiomers are commercially available. Experimental measurements were performed on (R)-pulegone in CDCl3. The structure is nearly as rigid as camphor, so little difficulty was encountered in sampling conformations. Experimental VCD determination for this molecule has been previously published in the literature.61−64 The calculated IR and VCD spectrum of the in vacuo monomer agree, to a large extent, with the experimental curves; however, the ketone stretch at approximately 1675 cm−1 does not agree in sign. Debie et al. provided a detailed analysis, indicating the need to model a solvent complex between pulegone and CDCl3 which, when modeled in implicit solvent, maintained all peak matches including the desired sign inversion of the ketone VCD stretch. It was shown that outside of the normal fingerprint region used for VCD assignment (1000−2000 cm−1) an additional peak was observed experimentally which corresponded directly to the solute−solvent complex (peak at 2250 cm−1). We followed the recommendations of Debie et al. in modeling II and observed equivalent behavior for the modeled systems. The calculated VCD peak for the ketone in moving

Table 4. Convergence in Boltzmann Sampling for Compound IIa total possible conformers b

initial generated Boltzmann, >5% absol config ESI original IR final IR original VCD final VCD original enantiomer final enantiomer CompareVOA ESICompareVOA confidence

complexc implicitd

20

50

100

200

11 2 R 0.85 0.24 0.28 0.41 0.44 −0.41

22 2 R 0.86 0.24 0.28 0.41 0.45 −0.41

40 2 R 0.87 0.24 0.29 0.42 0.45 −0.42

71 2 R 0.87 0.24 0.29 0.42 0.45 −0.42

1 1 R 0.90 0.30 0.33 0.46 0.45 −0.46

1 1 R 0.66 0.45 0.46 0.50 0.47 −0.50

−0.41

−0.42

−0.42

−0.42

−0.45

−0.19

R 62.7 100

a

Gas-phase monomer unless otherwise noted. bInitial generated is the total number in the clustered file prior to Gaussian minimization. c Complex with explicit CDCl3 molecule. dExplicit complex with implicit solvation.

from the monomer to CDCl3 complex in implicit solvent shifted from 1725 to 1700 to 1675 cm−1 with normalized intensities of approximately 0.5 to 0.25 to −0.1. This is a striking example visually, but as the figures have been published previously, we refer the reader to the cited work.63 Our experience with other ketone-containing molecules indicates that this solvent effect is not always as dramatic and that in vacuo modeling at times suffices. It is clear from the cited work that additional computational work beyond the in vacuo treatment allowed for a more satisfying match, but the assigned absolute configuration was not altered after this additional computational expense. Modeling the monomer form of II in vacuo allows the absolute configuration to be correctly assigned with good confidence. The final match of the IR and correct enantiomer VCD is 0.29 and 0.45, respectively, with an ESI of 0.87 (Figure 3C,D). CompareVOA assigns an ESI of 62.7 and a confidence of 100%. More rigorous modeling of the system with solvent complexation and implicit solvent increases the IR and VCD match to 0.46 and 0.50, respectively, with an ESI of 0.66. Interestingly, the statistics of the individual IR and VCD spectra indicate better matches with the more complete system; however, a more discerning ESI is obtained with monomer modeling in vacuo. The final in vacuo monomer IR match is rather low compared to where we would want that statistic to normally track, and it is apparent that this is due to a lack of overlap of the two peaks relating to the ketone (∼1675 cm−1) and alkene (∼1600 cm−1). Noting that this is likely a solvent effect on the peak position, we would expect the overlap of these two peaks to be much larger given more rigorous modeling including implicit and explicit solvation (as is the case, data not presented). The visual IR match is much better than calculated statistically. 3.6. Compound III: Ibuprofen. Analysis of ibuprofen (Table 5) is more difficult than expected. The compound has a single chiral center and few rotatable bonds, so similarly to G



Article

compound contains a trifluoromethyl group which is significant in that the presence of halogen atoms is thought to at times confound the ability to match VCD spectra. The suspected low conformational flexibility is misleading, and it was observed that larger initial conformer sampling converged to a Boltzmann population of 10 structures contributing >5%. This distribution was apparent for sample sizes of 100 or 200 conformers sent for DFT; however, lower conformer pools of 20 and 50 provided only 2 and 5 Boltzmann-relevant conformers, respectively. The IR matching was fairly stable for compound IV at 0.55, as was the near equality of enantiomer matches leading to an ESI of at most 0.07 (Figure 4E,F). Use of CompareVOA provided a much more optimistic view of ESI at 60.1 with a percent confidence in assignment of the correct stereochemistry of 100%. Visual inspection of the overlays indicates that the very small VCD overlap score is driven by the broad experimental peak at about 1350 cm−1 which only partially overlaps onto the calculated peak in the same region. Much of the two spectra qualitatively agree but algorithmic matching leads to an artificially low score. As described, the default matching algorithm only right shifts peaks. Though the two enantiomers end up with nearly identical scores for matching of VCD curves, the correct assignment is clearly S. CompareVOA which does not contain a separate, independent, peak shifting step does a better job at matching the spectra, with a much more discerning ESI. 3.8. Compound V: Filorexant. Filorexant75−77 (Table 7) represents a substantially more complex molecule compared to molecules I−IV. The relatively large number of rotatable bonds is further complicated by the need to sample ring puckering of the central piperidine linker. Conformer sampling using low initial sampling sizes showed lack of coverage of all local minima. Larger input populations led to a stable number of minima identified at either seven or eight with the greatest ESI of 0.42 (best IR match of 0.64), indicating an assignment of RR, agreeing with the known initial configuration (Figure 5A,B). There are no confounding functional groups in this molecule, and little ability for the molecule to interact with solvent or form higher order structures. Comparison of the calculated and experimental curves measured in CDCl3 (DMSO experimental work provided the same assignment) shown in Figure 5 indicates a complicated VCD signal, but one which was matched very well by calculation. The largest disconnect between the curves is the experimental peak at 1600 cm−1 corresponding to the ketone which can be attributed to solvent effects. The peak is not present in the calculated VCD and is shifted to 1700 cm−1 in the calculated IR. Modeling of V in implicit solvent would be expected (but as noted, the correct assignment can be made without this more rigorous evaluation) to selectively shift this peak to a region more in-line with experiment both cleaning up the overall match and increasing the statistical IR match above 0.64. CompareVOA assigns an ESI of 49.8 with a confidence of 89% for this match. 3.9. Compound VI: Laropiprant. Molecule VI78 (Table 8) has a carboxylic acid which complicates the calculations. Initial work with VI was performed with the monomer only, and the global minimum was not identified with either the submission of 20 or 50 conformers. Only at 100 conformers was the true global minima identified which shifted the Boltzmann distribution from 7 low energy conformers identified using 50 submitted conformers to only 4 at either 100 or 200

Table 5. Convergence in Boltzmann Sampling for Compound IIIa total possible conformers 20

50

100

200

dimerc

initial generated Boltzmann, >5% absol config ESI original IR final IR original VCD final VCD original enantiomer final enantiomer

20 4 S 0.30 0.12 0.13 0.18 0.31 −0.18 0.01

37 4 S 0.30 0.12 0.13 0.18 0.31 −0.18 0.01

58 4 S 0.30 0.13 0.13 0.19 0.32 −0.19 0.02

88 4 S 0.30 0.13 0.13 0.19 0.32 −0.19 0.02

10 7 SSc 0.58 0.35 0.53 −0.01 0.59 0.01 0.01


S 55.5 87

b

a

Gas-phase monomer unless otherwise noted. bInitial generated is the total number in the clustered file prior to Gaussian minimization. cAcid dimer.

compounds I and II, there should be little trouble in generating a calculated spectrum and matching it to a measured spectrum. Prior VCD analysis is available.65,66 Initial conformer searching indicated fairly quick convergence of a Boltzmann distribution composed of four conformers. Use of CompareVOA to match the experimental spectra in CDCl3 against the calculated spectra of the monomer provided a confidence of 100% with an ESI of 55.5 (Figure 4A,B). As this compound was run during the initial stages of the benchmarking, we knew two facts: (1) the compound had a known absolute configuration of R, and (2) the matching algorithm was telling us that the match was 100%, indicating the correct assignment. What we learned was that hidden in the correct assignment of the monomer was a more properly modeled dimer. Our initial benchmarking was performed solely using CompareVOA, and what we began to observe in some cases was rather high confidence values for matches which looked less than visually perfect. This in turn led us to pursue the coding of an in-house matching algorithm which, when used for the monomer match, assigned R as expected with a similar ESI. However, the IR matched at only 0.13, indicating a very poor overlay, which by-eye was obviously better than statistics would indicate. Knowing the answer in advance for this compound led us to initially become less analytical during interpretation of the IR. Careful analysis of the peak by peak vibrational contributions to the IR and VCD calculated spectra led us to consider a dimerization possibility for carboxylic acids, a finding which had been published several times in the literature.31,60,66−73 Compound III was studied previously as a monomer and dimer, and it was shown that the dimer was forming in solution during the experiment.66,67 When the dimer is compared to the experimental spectra, the match statistics improve to 0.53 for the IR and 0.59 for the correct enantiomer VCD with an ESI of 0.58 (Figure 4C,D). A single dimer was used for the demonstrated improvement in VCD assignment, but we are pursuing a more extensive analysis of this compound to help refine our approach. 3.7. Compound IV: Efavirenz. Efavirenz74 (Table 6) is a semirigid small molecule with a known configuration of S. This H



Article

Figure 4. Raw overlay of calculated (red) and measured (black) VCD spectra (A) for III monomer, overlay of VCD and IR after peak shifting and scaling (B) for III monomer, raw overlay of calculated (red) and measured (black) VCD spectra (C) for III dimer, overlay of VCD and IR after peak shifting and scaling (D) for III dimer, raw overlay of calculated (red) and measured (black) VCD spectra (E) for IV, and overlay of VCD and IR after peak shifting and scaling (F) for IV. Intensities scaled from −1.0 to 1.0 (0.0 to 1.0 for IR).

Owing to the carboxylic acid functionality being implicated in the largest discrepancies in the spectral matching, we assumed acid dimerization. Preliminary findings suggest this assumption is valid for chloroform but would be less valid when moving to DMSO where the DMSO complex would be expected to dominate. When modeling the dimer in the gas phase, the ESI jumps from 0.03 to 0.21 while maintaining a poorer IR match than desirable. From the matches depicted in Figure 5C−F, the unshifted match (Figure 5C) for the monomer indicates a significant mismatch around 1150 cm−1. The shifting of the spectra (Figure 5D) for the monomer actually inverts the initial match of R and instead selects the S enantiomer as best, owing to the

submissions. The compound is fairly rigid with relevant conformers sampling carboxylic acid rotamers, phenyl rotamers, and positioning of the methylsulfone above and below the aromatic plane. The configuration of VI is known to be R, and with initial matching of the monomer spectra in the gas phase it was noted that no assignment could be made with the best ESI calculated to be 0.08 and the best IR match below 0.50 (Figure 5C,D). Assignment was indicating S, the incorrect enantiomer, though the statistics and visual match were so poor as to invalidate any assignment. Both DMSO and chloroform solvents were used to gather experimental spectra with the later having better resolution for matching. I



Article

rotamers of only one of the trifluoromethyl groups. In the minima, one CF3 group packs against the fluorophenyl, leaving the opposite ortho-substituted group free to rotate. The spectra were gathered only in DMSO, and initial matching of the gas-phase monomer to the experimental curves indicated a successful assignment of known configuration SRR (Figure 6A,B). Observation of the unshifted spectra shows two dominant peaks in the experimental measurement at approximately 1700 and 1280 cm−1. While these peaks are likely not artifacts, owing to their presence in the calculated spectra as well, the peak intensities from the experimental spectra swamped out all other spectral features. Cutting these peaks out altogether (data not shown) indicates a very good overlay of the correct enantiomer over the region from 1100 to 1400 cm−1, but we choose to include those peaks because they are present in the calculated spectrum as well. Rather than removing/ignoring the peaks (a methodological choice also possible in CompareVOA), we decided to damp down the peaks by artificially weighting the two dominant peaks to still maintain dominance in the experimental spectrum but at the same time allow other features of the curve to become more signal than noise. Sampling of conformer space was again only exhaustive once either 100 or 200 conformers were submitted initially. The Boltzmann distribution was composed of four conformers, with three contributing approximately 30% each. The ESI peaked at 0.29 with an IR match of 0.35, low owing to the large disconnect in positioning of the ketone peak. Agreeing with the best in-house match, CompareVOA assigned the correct absolute configuration of SRR with an ESI of 48.6 and a confidence of 76%. Modeling of VII in implicit solvent did not improve the match significantly, though as expected, the ketone peak observed at 1700 cm−1 did shift downfield in the calculation for a closer alignment. Further modeling of explicit solvent complexes would likely improve the match, though this work was beyond the scope of our initial assignment for this compound. The second large peak which was dampened during the matching was the peak at 1280 cm−1 which in the calculated spectrum looks to be associated with a vibration of the phenyl ring holding the two trifluoromethyl groups such that the two CF3 play a dominant role. This vibration appears to be magnified in the experiment, although it is not entirely clear how the solvent would directly impact this specific vibration, as opposed to the more straightforward directed interaction with the NH group of the triazolone. 3.11. Compound VIII: Ezetimibe. Compound VIII81 (Table 10) has a large number of rotatable bonds, several aromatic rings, and polar functionality which could interact with solvent molecules. The absolute configuration is known to be 1S, 2R, 3S (see Figure 1). The sample was measured only in DMSO, and a large intensity peak was observed in the experiment at ∼1510 cm−1. Sampling conformers did not converge at 20, 50, or 100 conformers with the full Boltzmann set of four minima only being identified after 200 conformers were submitted. While the IR match is clearly correct for this compound, subtle shifting of the ketone peak lowers the apparent statistical match to only 0.47 where it would be higher given modeling in implicit solvent (Figure 6C−E). The peak at ∼1510 cm−1 has been edited out in Figure 6E to clarify that the rest of the VCD curves are overlaying. It is evident that the calculated curve over the region 1100−1400 cm−1 is fairly well representative of the

Table 6. Convergence in Boltzmann Sampling for Compound IVa total possible conformers 20

50

100

200


9 2 S 0.00 0.53 0.55 0.08 0.10 −0.08 0.10

20 5 S 0.05 0.53 0.55 0.07 0.15 −0.07 0.10

70 10 S 0.07 0.53 0.55 0.07 0.16 −0.07 0.10

70 10 S 0.05 0.53 0.55 0.07 0.15 −0.07 0.10


S 60.1 100

a


Table 7. Convergence in Boltzmann Sampling for Compound Va total possible conformers 20

50

100

200


20 5 RR 0.21 0.48 0.65 0.26 0.31 −0.26 0.10

49 7 RR 0.37 0.48 0.65 0.32 0.44 −0.32 0.07

100 8 RR 0.40 0.48 0.64 0.30 0.45 −0.30 0.05

199 7 RR 0.42 0.47 0.64 0.33 0.48 −0.33 0.06


RR 49.8 89

a


large overlap of the two significant peak areas at 1150 cm−1. Comparison of the monomer to dimer matches (Figure 5, parts C,D to E,F) suggests that there are aspects of both the monomer and the dimer apparent in the experimentally determined spectra. Testing this hypothesis by mixing the calculated spectra did not improve the match statistics. While the dimer does allow for a statistically meaningful match to the correct configuration (Figure 5E), the overall satisfaction in this match was lower than desired. For the dimer match, CompareVOA agrees with the assignment of R given an ESI of 40.9 but does output a low confidence of only 56%. For these reasons, we are undertaking more extensive work to characterize this compound. 3.10. Compound VII: Aprepitant. Compound VII79,80 (Table 9) would seem to be very flexible based on visual inspection of the structure but turns out to be a rather rigid molecule after conformer searching with a Boltzmann distribution from the largest sampling size of only four conformers. The conformers are identical except for the J



Article

Figure 5. Raw overlay of calculated (red) and measured (black) VCD spectra (A) for V, overlay of VCD and IR after peak shifting and scaling (B) for V, raw overlay of calculated (red) and measured (black) VCD spectra (C) for VI monomer, overlay of VCD and IR after peak shifting and scaling (D) for VI monomer, raw overlay of calculated (red) and measured (black) VCD spectra (E) for VI dimer, and overlay of VCD and IR after peak shifting and scaling (F) for VI dimer. Intensities scaled from −1.0 to 1.0 (0.0 to 1.0 for IR).

measured spectrum. The peak at 1500 cm−1 is present in the calculated IR and VCD and so should not be removed altogether. This peak is composed of three overlapping peaks related to molecular vibrations of each of the three aromatic rings. As one contains an OH group, it appears that interaction of the OH with DMSO increases the intensity of this vibration, leading to the large measured peak. Assignment of SRS by CompareVOA was made with an ESI of 39.6 and a moderate confidence of 54%. 3.12. Compound IX: Simvastatin. Compound IX82,83 (Table 11) represents the molecule with the highest degree of complexity originally selected to benchmark the VCD workflow. The molecule contains seven chiral centers, and it must be

stated that no attempt would be made at assignment of all centers from scratch, as the number of possible diastereomers is too large and the ability of statistical matching to the known configuration being able to differentiate all other diastereomers would be unlikely. Given that caveat, it was of interest to determine whether the correct enantiomer of simvastatin could be modeled well enough to make an assignment. Lack of convergence in the Boltzmann distribution was seen with small numbers of input conformers and with a sampling of 200 input structures the distribution contained six dominant conformations. Experimental measurements were run in chloroform, and IX was modeled as a monomer in the gas phase. Two sets of K



Article

shows agreement, calculating an ESI of 68.2 and a confidence of 100%. It is interesting to note that a better ESI is calculated for the smaller sampling size calculations where the main difference is that the intramolecular hydrogen bonded species was not identified in the conformer search until the sampling size reached 100 or 200 input conformers. The intramolecular hydrogen bonded conformer may be less relevant experimentally in chloroform which is why lack of this conformer’s contribution to the Boltzmann match at times indicates a better statistical assignment of the correct configuration. 3.13. Efficiency Considerations and Methodology Enhancements. Throughout these studies it was clear that the workflow might be sped up using a series of smaller basis sets to prefilter the conformer space to decrease the load on our computer resources. During test cases of this possibility, it was observed that the same conformational distribution was not always achieved because smaller basis sets did not converge to the same Boltzmann population compare with the default basis set. We have found that mismatches in spectra were seldom resolved by moving to higher levels of theory (MP2), different DFT methods (PBE, M062X), or larger basis sets (6-311+ +G**). One benefit of the fragment approach is that higher levels of theory can be used,17 but in practical application we find that variation in the fragment to whole molecule calculated peaks is more disruptive to the assignments than more rigorous calculated intensities. While a satisfactory match can be made using the in vacuo monomer, we understand a more extensive study of modeling complexes with CDCl3 and inclusion of implicit solvent could be undertaken for each compound (for instance compound II). However, within the pharmaceutical setting and with a backlog of samples waiting for absolute configuration assignment, it is not always possible or needed to perform a more exhaustive analysis which considers implicit solvent, explicit solvent, or higher order complexes. We have seen evidence of improved qualitative and quantitative matches when moving to larger basis sets because peak intensities, at times, match experimental intensities better. However, the assignment of absolute configuration was not altered given this additional time expense. Additionally, while the use of robust modes61,84 (matching only those modes which are conserved across several computational methods) has been shown to be an effective methodological advancement, we did not see evidence that this approach altered the assignments made. We emphasize the difference between being able to accurately and efficiently assign absolute configuration as a “service” and the more exhaustive analysis which would correctly match each and every peak. 3.14. Prospective Use in the Pharmaceutical Setting. After the initial benchmarking studies where 14 out of 14 compounds (the subset discussed here) were assigned the correct absolute configuration based on this workflow, the methodology was put into production. A stereochemical assignment group was formed at Merck, taking on the task of absolute configuration assignment where configuration determination often relies on cross-functional collaborations among VCD, IR, NMR, medicinal chemistry, crystallography, and computational chemistry. For platform assignments of absolute configuration at Merck, we have adopted a system of three levels of combined certainty in the assignment: no assignment (owing to lack of signal, low signal-to-noise, or difficulty in modeling), low confidence (assignment made but suggestion to

Table 8. Convergence in Boltzmann Sampling for Compound VIa total possible conformers 20

50

100

200

dimerc


20 4 S 0.05 0.38 0.43 0.00 0.08 0.00 0.12

49 7 S 0.08 0.39 0.42 −0.01 0.09 0.01 0.17

100 4 S 0.04 0.38 0.44 −0.01 0.03 0.01 0.07

200 4 S 0.03 0.38 0.43 0.01 0.09 −0.01 0.12

1 1 RR3 0.21 0.35 0.42 0.08 0.26 −0.08 0.05


R 40.9 56

b

a

Gas-phase monomer unless otherwise noted. bInitial generated is the total number in the clustered file prior to Gaussian minimization. cAcid dimer.

Table 9. Convergence in Boltzmann Sampling for Compound VIIa total possible conformers 20

50

100

200

initial generatedb Boltzmann, >5% absol configc ESI original IR final IR original VCD final VCD original enantiomer final enantiomer

20 1 SRR 0.17 0.26 0.34 0.16 0.21 −0.16 0.04

49 2 SRR 0.21 0.26 0.34 0.18 0.25 −0.18 0.04

100 4 SRR 0.28 0.26 0.35 0.18 0.28 −0.18 0.01

199 4 SRR 0.29 0.26 0.35 0.19 0.29 −0.19 0.01


SRR 48.6 76

a

Gas-phase monomer unless otherwise noted. bInitial generated is the total number in the clustered file prior to Gaussian minimization. cFor a correct match to be made, the experimental peak at ∼1290 cm−1 needed to be artificially scaled down or removed; herein we report the scaled result.

comparisons were made, over the range 1000−2000 cm−1 and the range 1000−1550 cm−1, to remove the ketone-related peaks at 1700 cm−1 (Figure 7A−D). In the area of the large ketone peak, the calculated spectrum contains three peaks corresponding to the two ketones, and an intramolecular hydrogen bond formed between the hydroxyl and open chain ester carbonyl. These peaks coalesce into one broad peak in the experiment. Irrespective of number of conformers contributing to the Boltzmann, the statistical match to the correct enantiomer RSRSSRR is supported with ESIs in the range of 0.30−0.41 for comparison over the full spectral range. Cutting the window down to the smaller comparison range leads to a significant increase in the ESI and IR match. Over both ranges, the assignment of the correct absolute configuration is indicated (Figure 7C,D). Matching of the spectra using CompareVOA L



Article

Figure 6. Raw overlay of calculated (red) and measured (black) VCD spectra (A) for VII, overlay of VCD and IR after peak shifting and scaling (B) for VII, raw overlay of calculated (red) and measured (black) VCD spectra (C) for VIII, overlay of VCD and IR after peak shifting and scaling (D) for VIII, and raw overlay of calculated (red) and measured (black) VCD spectra (E) for VIII with one experimental peak ignored (∼1510 cm−1). Intensities scaled from −1.0 to 1.0 (0.0 to 1.0 for IR).

all molecules which have been assigned by both VCD and other complementary methods (e.g., small molecule crystallography) have indicated the same absolute configuration. Making assignments by modeling the full molecule and not relying on fragments or comparative studies may offer clear advantages if a higher percentage of correct assignments are made. The fragment approach requires the loss of some functional group contributions to the calculated IR and VCD.17 While qualitative matching can imply configurational assignment, we have departed from this approach completely. Importantly, a high quality match of the IR has been a significant filter for our matches such that poor IR agreement implies no assignment.

pursue additional structural characterization is made when results are reported), and high confidence. Full DFT characterization of a druglike molecule can be obtained on the order of time it takes to run the experiment, i.e., one day. This represents a significant increase in productivity compared to the estimates of days-to-weeks made as recently as 2007.17 More than 100 submissions for assignment of absolute configuration have been made with VCD leading to confident assignments in all but ∼10 cases (impact rate of ∼90%). One example of prospective application of the workflow has been published.24 Unlike Minick et al.,17 we have not seen any evidence for incorrect assignments, and M



Article

Table 10. Convergence in Boltzmann Sampling for Compound VIIIa total possible conformers 20

50

100

200

20c

50c

100c

200c


20 2 SRS 0.30 0.20 0.46 0.10 0.25 −0.10 −0.06

49 2 SRS 0.22 0.20 0.47 0.09 0.18 −0.09 −0.04

100 2 SRS 0.22 0.20 0.47 0.09 0.18 −0.09 −0.04

200 4 SRS 0.33 0.20 0.47 0.10 0.26 −0.10 −0.07

20 2 SRS 0.31 0.20 0.46 0.18 0.21 −0.18 −0.10

49 2 SRS 0.24 0.20 0.47 0.15 0.16 −0.15 −0.08

100 2 SRS 0.24 0.20 0.47 0.15 0.16 −0.15 −0.08

200 4 SRS 0.32 0.20 0.47 0.17 0.21 −0.17 −0.11


SRS 39.6 54

Gas-phase monomer unless otherwise noted. bInitial generated is the total number in the clustered file prior to Gaussian minimization. cStatistics for the second block of matches are when the experimental peak at ∼1510 cm−1 was removed.

a

Table 11. Convergence in Boltzmann Sampling for Compound IXa total possible conformers 20

50

100

200

20c

50c

100c

200c


20 4 RSRSSRR 0.39 0.43 0.56 0.15 0.26 −0.15 −0.13

50 3 RSRSSRR 0.30 0.45 0.57 0.14 0.20 −0.14 −0.10

99 4 RSRSSRR 0.30 0.44 0.55 0.12 0.20 −0.12 −0.09

200 6 RSRSSRR 0.41 0.49 0.48 0.14 0.33 −0.14 −0.09

20 4 RSRSSRR 0.95 0.78 0.80 0.43 0.60 −0.43 −0.35

50 3 RSRSSRR 0.80 0.80 0.81 0.42 0.53 −0.42 −0.27

99 4 RSRSSRR 0.76 0.82 0.82 0.38 0.49 −0.38 −0.27

200 6 RSRSSRR 0.61 0.80 0.85 0.18 0.53 −0.18 −0.09


RSRSSRR 68.2 100

b

Gas-phase monomer unless otherwise noted. bInitial generated is the total number in the clustered file prior to Gaussian minimization. cStatistics for the second block of matches are over the reduced range 1000−1550 cm−1 compared to the full range 1000−2000 cm−1.

a

4. CONCLUSIONS Assignment of absolute configuration using a combination of VCD and IR spectroscopy has moved from being a detailoriented single molecule research project into a streamlined and semiroutine platform technology. We describe a step-bystep process for taking a 2D structure and producing a statistically significant in vacuo Boltzmann-distribution of conformers which can be used to assign absolute configuration after fitting to experimentally determined spectra. Assignments were made for nine molecules, seven of which represent complex, flexible compounds. This sampling of test cases expands the field’s understanding of the application of VCD to druglike molecules. Importantly, our experiences indicate that a timely and correct assignment of absolute configuration can be made with sufficient confidence primarily using in vacuo conformational sampling, and only in select cases is there a need to move to more sophisticated modeling to include solvent and complexation. We show that accurate assignments of complex molecules can be rapidly made without needing to rely on assumptions related to the fragment approach.

On many occasions we have been able to track down the cause of poor IR agreement to chiral and achiral impurities or incorrect compound submission. In one case the 2D chemical structure was drawn incorrectly upon submission (only a single carbon to oxygen swap) and poor IR matching was used to identify this problem. Relying on the fragment approach would have prevented these spot-checks. Our assignments have supported >40 medicinal chemistry programs and had impact including time savings, cost savings, supporting documentation for clinical candidate nomination, publication, and confirmation or lack thereof of the configuration of purchased reagents/compounds. We have had success in assignment of absolute configuration almost exclusively for pairs of enantiomers and not diastereomers. Any work on diastereomers (or relative configuration) relies heavily on collaborations with Merck’s NMR group. Importantly, the procedures described here have been used cross-platform to support all modeling efforts which require the identification of Boltzmann populations, conformational energetics, or just global minimum conformations. N



Article

Figure 7. Raw overlay of calculated (red) and measured (black) VCD spectra (A) for IX, overlay of VCD and IR after peak shifting and scaling (B) for IX, raw overlay of calculated (red) and measured (black) VCD spectra (C) for IX over a reduced range, and overlay of VCD and IR after peak shifting and scaling (D) for IX over a reduced range. Intensities scaled from −1.0 to 1.0 (0.0 to 1.0 for IR).

■

ASSOCIATED CONTENT

Yanan He from BioTools and Jim Cheeseman at Gaussian for helpful discussions.

S Supporting Information *

■

Gas-phase B3LYP/6-31G** Boltzmann distributions and the coordinates for the global minima for each compound. This material is available free of charge via the Internet at http:// pubs.acs.org.

■

ABBREVIATIONS USED AC, absolute configuration; DFT, density functional theory; ESI, enantiomeric similarity index; IR, infrared; MM, molecular mechanics; MMFF, Merck molecular forcefield; NMR, nuclear magnetic resonance; QM, quantum mechanics; RMS, rootmean-square; VCD, vibrational circular dichroism; nonabbreviated computer programs, ET, JG, DG

AUTHOR INFORMATION

Corresponding Author

■

*Phone: +1 732 594-2883. E-mail: edward_sherer@merck. com. Notes

REFERENCES

(1) Allenmark, S.; Gawronski, J. Determination of absolute configuration–an overview related to this special issue. Chirality 2008, 20, 606−608. (2) Harada, N. Determination of absolute configurations by X-ray crystallography and 1H NMR anisotropy. Chirality 2008, 20, 691−723. (3) Chavali, B.; Krishnamurthy, K.; Dage, J. Mid IR CD Spectroscopy for Medicinal Chemistry: A Pharmaceutical Perspective. Am. Pharm. Rev. 2007, 10, 94−98. (4) Devlin, F. J.; Finley, J. W.; Stephens, P. J.; Frisch, M. J. Ab Initio Calculation of Vibrational Absorption and Circular Dichroism Spectra Using Density Functional Force Fields: A Comparison of Local, Nonlocal, and Hybrid Density Functionals. J. Phys. Chem. 1995, 99, 16883−16902. (5) Barron, L. D.; Buckingham, A. D. Vibrational optical activity. Chem. Phys. Lett. 2010, 492, 199−213. (6) Dunmire, D.; Freedman, T. B.; Nafie, L. A.; Aeschlimann, C.; Gerber, J. G.; Gal, J. Determination of the absolute configuration and solution conformation of the antifungal agents ketoconazole,

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We thank Merck’s New Technologies Review and Licensing Committee (NT-RLC) for funding of the VCD instrument. We thank Merck Process and Discovery Chemistry for synthesis of the compounds used in this study along with Wendy Cornell, Chris Culberson, Brad Feuston, Gene Fluder, Simon Kearsley, Viktor Hornak, Daniel McMasters, Dennis Schuchman, Brad Sherborne, and Johannes Voigt either for generation of scripts used in this work or for guidance during the evaluation stage. Calculations were run with modeling software available within the Merck MIX modeling environment and relied heavily upon the resources of Merck’s Scientific Computing group and methods developers. We thank Larry Nafie, Rina Dukor, and O



Article

itraconazole, and miconazole with vibrational circular dichroism. Chirality 2005, 17 (Suppl), S101−108. (7) Freedman, T. B.; Cao, X.; Dukor, R. K.; Nafie, L. A. Absolute configuration determination of chiral molecules in the solution state using vibrational circular dichroism. Chirality 2003, 15, 743−758. (8) He, Y. A.; Wang, B.; Dukor, R. K.; Nafie, L. A. Determination of Absolute Configuration of Chiral Molecules Using Vibrational Optical Activity: A Review. Appl. Spectrosc. 2011, 65, 699−723. (9) Mcconnell, O.; Bach, A.; Balibar, C.; Byrne, N.; Cai, Y. X.; Carter, G.; Chlenov, M.; Di, L.; Fan, K.; Goljer, I.; He, Y. N.; Herold, D.; Kagan, M.; Kerns, E.; Koehn, F.; Kraml, C.; Marathias, V.; Marquez, B.; McDonald, L.; Nogle, L.; Petucci, C.; Schlingmann, G.; Tawa, G.; Tischler, M.; Williamson, R. T.; Sutherland, A.; Watts, W.; Young, M.; Zhang, M. Y.; Zhang, Y. R.; Zhou, D. H.; Ho, D. Enantiomeric separation and determination of absolute stereochemistry of asymmetric molecules in drug discovery - Building chiral technology toolboxes. Chirality 2007, 19, 658−682. (10) McConnell, O.; He, Y. A.; Nogle, L.; Sarkahian, A. Application of chiral technology in a pharmaceutical company. Enantiomeric separation and spectroscopic studies of key asymmetric intermediates using a combination of techniques. Phenylglycidols. Chirality 2007, 19, 716−730. (11) Stephens, P. J.; Devlin, F. J.; Pan, J. J. The determination of the absolute configurations of chiral molecules using vibrational circular dichroism (VCD) spectroscopy. Chirality 2008, 20, 643−663. (12) Freedman, T. B.; Nafie, L. A. Stereochemical Aspects of Vibrational Optical Activity. In Topics in Sterochemistry; Wilen, S. H., Eliel, E. L., Eds.; John Wiley & Sons: New York, 1987; Vol. 17, pp 113−206. (13) Castiglioni, E.; Di Fabio, R.; Togninelli, A.; Brough, S.; Brown, F.; Dal Cin, M.; Gianotti, M.; Marchioro, C.; Merlo, G.; Spinosa, R.; Wigglesworth, M. J.; Botta, M. Towards the Discovery of New Hypnotic Agents: Synthesis and Preliminary Pharmacological Evaluation of a Novel Class of Dibenzo[a,d]cycloheptene Derivatives. ChemMedChem 2010, 5, 1843−1846. (14) Micheli, F.; Cavanni, P.; Arban, R.; Benedetti, R.; Bertani, B.; Bettati, M.; Bettelini, L.; Bonanomi, G.; Braggio, S.; Checchia, A.; Davalli, S.; Di Fabio, R.; Fazzolari, E.; Fontana, S.; Marchioro, C.; Minick, D.; Negri, M.; Oliosi, B.; Read, K. D.; Sartori, I.; Tedesco, G.; Tarsi, L.; Terreni, S.; Visentini, F.; Zocchi, A.; Zonzini, L. 1-(Aryl)-6[alkoxyalkyl]-3-azabicyclo[3.1.0]hexanes and 6-(Aryl)-6-[alkoxyalkyl]3-azabicyclo[3.1.0]hexanes: A New Series of Potent and Selective Triple Reuptake Inhibitors. J. Med. Chem. 2010, 53, 2534−2551. (15) Minick, D. J.; Copley, R. C.; Szewczyk, J. R.; Rutkowske, R. D.; Miller, L. A. An investigation of the absolute configuration of the potent histamine H3 receptor antagonist GT-2331 using vibrational circular dichroism. Chirality 2007, 19, 731−740. (16) Polavarapu, P. L.; He, J. T. Chiral analysis using Mid-IR vibrational CD spectroscopy. Anal. Chem. 2004, 76, 61a−67a. (17) Minick, D. J.; Rutkowske, R. D.; Miller, L. A. D. Strategies for Successfully Applying Vibrational Circular Dichroism in a Pharmaceutical Research Environment. Am. Pharm. Rev. 2007, 10, 118. (18) Crombie, A. L.; Antrilli, T. M.; Campbell, B. A.; Crandall, D. L.; Failli, A. A.; He, Y. N.; Kern, J. C.; Moore, W. J.; Nogle, L. M.; Trybulski, E. J. Synthesis and evaluation of azabicyclo[3.2.1]octane derivatives as potent mixed vasopressin antagonists. Bioorg. Med. Chem. Lett. 2010, 20, 3742−3745. (19) Yang, G. C.; Tran, H.; Fan, E.; Shi, W.; Lowary, T. L.; Xu, Y. J. Determination of the Absolute Configurations of Synthetic Daunorubicin Analogues Using Vibrational Circular Dichroism Spectroscopy and Density Functional Theory. Chirality 2010, 22, 734−743. (20) Reina, M.; Burgueno-Tapia, E.; Bucio, M. A.; Joseph-Nathan, P. Absolute configuration of tropane alkaloids from Schizanthus species by vibrational circular dichroism. Phytochemistry 2010, 71, 810−815. (21) Urbanova, M.; Setnicka, V.; Bour, P.; Navratilova, H.; Volka, K. Vibrational circular dichroism spectroscopy study of paroxetine and femoxetine precursors. Biopolymers 2002, 67, 298−301. (22) Bour, P.; Kim, J.; Kapitan, J.; Hammer, R. P.; Huang, R.; Wu, L.; Keiderling, T. A. Vibrational circular dichroism and IR spectral analysis

as a test of theoretical conformational modeling for a cyclic hexapeptide. Chirality 2008, 20, 1104−1119. (23) Molina-Salinas, G. M.; Rivas-Galindo, V. M.; Said-Fernandez, S.; Lankin, D. C.; Munoz, M. A.; Joseph-Nathan, P.; Pauli, G. F.; Waksman, N. Stereochemical Analysis of Leubethanol, an Anti-TBActive Serrulatane, from Leucophyllum f rutescens. J. Nat. Prod. 2011, 74, 1842−1850. (24) Liu, Z. Q.; Shultz, C. S.; Sherwood, C. A.; Krska, S.; Dormer, P. G.; Desmond, R.; Lee, C.; Sherer, E. C.; Shpungin, J.; Cuff, J.; Xu, F. Highly enantioselective synthesis of anti aryl beta-hydroxy alpha-amino esters via DKR transfer hydrogenation. Tetrahedron Lett. 2011, 52, 1685−1688. (25) Velazquez-Jimenez, R.; Torres-Valencia, J. M.; Cerda-GarciaRojas, C. M.; Hernandez-Hernandez, J. D.; Roman-Marin, L. U.; Manriquez-Torres, J. J.; Gomez-Hurtado, M. A.; Valdez-Calderon, A.; Motilva, V.; Garcia-Maurino, S.; Talero, E.; Avila, J.; Joseph-Nathan, P. Absolute configuration of podophyllotoxin related lignans from Bursera fagaroides using vibrational circular dichroism. Phytochemistry 2011, 72, 2237−2243. (26) Batista, J. M.; Batista, A. N. L.; Rinaldo, D.; Vilegas, W.; Ambrosio, D. L.; Cicarelli, R. M. B.; Bolzani, V. S.; Kato, M. J.; Nafie, L. A.; Lopez, S. N.; Furlan, M. Absolute Configuration and Selective Trypanocidal Activity of Gaudichaudianic Acid Enantiomers. J. Nat. Prod. 2011, 74, 1154−1160. (27) Catalano, J. G.; Gudmundsson, K. S.; Svolto, A.; Boggs, S. D.; Miller, J. F.; Spaltenstein, A.; Thomson, M.; Wheelan, P.; Minick, D. J.; Phelps, D. P.; Jenkinson, S. Synthesis of a novel tricyclic 1,2,3,4,4a,5,6,10b-octahydro-1,10-phenanthroline ring system and CXCR4 antagonists with potent activity against HIV-1. Bioorg. Med. Chem. Lett. 2010, 20, 2186−2190. (28) Johns, B. A.; Kawasuji, T.; Weatherhead, J. G.; Taishi, T.; Temelkoff, D. P.; Yoshida, H.; Akiyama, T.; Taoda, Y.; Murai, H.; Kiyama, R.; Fuji, M.; Tanimoto, N.; Jeffrey, J.; Foster, S. A.; Yoshinaga, T.; Seki, T.; Kobayashi, M.; Sato, A.; Johnson, M. N.; Garvey, E. P.; Fujiwara, T. Carbamoyl Pyridone HIV-1 Integrase Inhibitors 3. A Diastereomeric Approach to Chiral Nonracemic Tricyclic Ring Systems and the Discovery of Dolutegravir (S/GSK1349572) and (S/GSK1265744). J. Med. Chem. 2013, 56, 5901−5916. (29) Zuber, G.; Goldsmith, M. R.; Hopkins, T. D.; Beratan, D. N.; Wipf, P. Systematic assignment of the configuration of flexible natural products by spectroscopic and computational methods: The bistramide C analysis. Org. Lett. 2005, 7, 5269−5272. (30) Munoz, M. A.; Areche, C.; Rovirosa, J.; San-Martin, A.; JosephNathan, P. Absolute Configuration of Sargaol Acetate Using DFT Calculations and Vibrational Circular Dichroism. Heterocycles 2010, 81, 625−635. (31) Freedman, T. B.; Cao, X. L.; Phillips, L. M.; Cheng, P. T. W.; Dalterio, R.; Shu, Y. Z.; Zhang, H.; Zhao, N.; Shukla, R. B.; Tymiak, A.; Gozo, S. K.; Nafie, L. A.; Gougoutas, J. Z. Determination of the absolute configuration and solution conformation of a novel disubstituted pyrrolidine acid A by vibrational circular dichroism. Chirality 2006, 18, 746−753. (32) Feuston, B. P.; Miller, M. D.; Culberson, J. C.; Nachbar, R. B.; Kearsley, S. K. Comparison of knowledge-based and distance geometry approaches for generation of molecular conformations. J. Chem. Inf. Comput. Sci. 2001, 41, 754−763. (33) Blaney, J. M.; Crippen, G. M.; Dearing, A.; Dixon, J. S. DGEOM; Quantum Chemistry Program Exchange; Indiana University, Bloomington, 1990. (34) Hawkins, P. C.; Nicholls, A. Conformer generation with OMEGA: learning from the data set and the analysis of failures. J. Chem. Inf. Model. 52, 2919-2936. (35) Hawkins, P. C.; Skillman, A. G.; Warren, G. L.; Ellingson, B. A.; Stahl, M. T. Conformer generation with OMEGA: algorithm and validation using high quality structures from the Protein Databank and Cambridge Structural Database. J. Chem. Inf. Model. 50, 572-584. (36) Halgren, T. A. Merck molecular force field 0.1. Basis, form, scope, parameterization, and performance of MMFF94. J. Comput. Chem. 1996, 17, 490−519. P



Article

(37) Petersson, G. A.; Allaham, M. A. A Complete Basis Set Model Chemistry. 2. Open-Shell Systems and the Total Energies of the 1stRow Atoms. J. Chem. Phys. 1991, 94, 6081−6090. (38) Petersson, G. A.; Bennett, A.; Tensfeldt, T. G.; Allaham, M. A.; Shirley, W. A.; Mantzaris, J. A Complete Basis Set Model Chemistry. 1. The Total Energies of Closed-Shell Atoms and Hydrides of the 1stRow Elements. J. Chem. Phys. 1988, 89, 2193−2218. (39) Rassolov, V. A.; Pople, J. A.; Ratner, M. A.; Windus, T. L. 631G* basis set for atoms K through Zn. J. Chem. Phys. 1998, 109, 1223−1229. (40) Rassolov, V. A.; Ratner, M. A.; Pople, J. A.; Redfern, P. C.; Curtiss, L. A. 6-31G*basis set for third-row atoms. J. Comput. Chem. 2001, 22, 976−984. (41) Francl, M. M.; Pietro, W. J.; Hehre, W. J.; Binkley, J. S.; Gordon, M. S.; Defrees, D. J.; Pople, J. A. Self-Consistent Molecular-Orbital Methods. 23. A Polarization-Type Basis Set for 2nd-Row Elements. J. Chem. Phys. 1982, 77, 3654−3665. (42) Hehre, W. J.; Ditchfield, R.; Pople, J. A. Self-Consistent Molecular Orbital Methods. 12. Further extensions of Gaussian-type basis sets for use in molecular-orbital studies of organic-molecules. J. Chem. Phys. 1972, 56, 2257. (43) Becke, A. D. Density-functional thermochemistry. III. The Role of Exact Exchange. J. Chem. Phys. 1993, 98, 5648−5652. (44) Lee, C. T.; Yang, W. T.; Parr, R. G. Development of the ColleSalvetti Correlation-Energy Formula into a Functional of the ElectronDensity. Phys. Rev. B 1988, 37, 785−789. (45) Miehlich, B.; Savin, A.; Stoll, H.; Preuss, H. Results Obtained with the Correlation-Energy Density Functionals of Becke and Lee, Yang and Parr. Chem. Phys. Lett. 1989, 157, 200−206. (46) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Scalmani, G.; Barone, V.; Mennucci, B.; Petersson, G. A.; Nakatsuji, H.; Caricato, M.; Li, X.; Hratchian, H. P.; Izmaylov, A. F.; Bloino, J.; Zheng, G.; Sonnenberg, J. L.; Hada, M.; Ehara, M.; Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.; Honda, Y.; Kitao, O.; Nakai, H.; Vreven, T.; Montgomery, J., J., A.; Peralta, J. E.; Ogliaro, F.; Bearpark, M.; Heyd, J. J.; Brothers, E.; Kudin, K. N.; Staroverov, V. N.; Kobayashi, R.; Normand, J.; Raghavachari, K.; Rendell, A.; Burant, J. C.; Iyengar, S. S.; Tomasi, J.; Cossi, M.; Rega, N.; Millam, J. M.; Klene, M.; Knox, J. E.; Cross, J. B.; Bakken, V.; Adamo, C.; Jaramillo, J.; Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi, R.; Pomelli, C.; Ochterski, J. W.; Martin, R. L.; Morokuma, K.; Zakrzewski, V. G.; Voth, G. A.; Salvador, P.; Dannenberg, J. J.; Dapprich, S.; Daniels, A. D.; Farkas, Ö .; Foresman, J. B.; Ortiz, J. V.; Cioslowski, J.; Fox, D. J. Gaussian 09, Revision A.02, Gaussian, Inc., Wallingford, CT, 2009. (47) Cheeseman, J. R.; Frisch, M. J.; Devlin, F. J.; Stephens, P. J. Ab initio calculation of atomic axial tensors and vibrational rotational strengths using density functional theory. Chem. Phys. Lett. 1996, 252, 211−220. (48) Shen, J. A.; Zhu, C. Y.; Reiling, S.; Vaz, R. A novel computational method for comparing vibrational circular dichroism spectra. Spectrochim. Acta, Part A 2010, 76, 418−422. (49) Debie, E.; De Gussem, E.; Dukor, R. K.; Herrebout, W.; Nafie, L. A.; Bultinck, P. A Confidence Level Algorithm for the Determination of Absolute Configuration Using Vibrational Circular Dichroism or Raman Optical Activity. ChemPhysChem 2011, 12, 1542−1549. (50) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 1996, 77, 3865−3868. (51) Zhao, Y.; Truhlar, D. G. The M06 suite of density functionals for main group thermochemistry, thermochemical kinetics, noncovalent interactions, excited states, and transition elements: two new functionals and systematic testing of four M06-class functionals and 12 other functionals. Theor. Chem. Acc. 2008, 120, 215−241. (52) Frisch, M. J.; Headgordon, M.; Pople, J. A. A Direct Mp2 Gradient-Method. Chem. Phys. Lett. 1990, 166, 275−280. (53) Headgordon, M.; Pople, J. A.; Frisch, M. J. Mp2 Energy Evaluation by Direct Methods. Chem. Phys. Lett. 1988, 153, 503−506.

(54) Marenich, A. V.; Cramer, C. J.; Truhlar, D. G. Universal Solvation Model Based on Solute Electron Density and on a Continuum Model of the Solvent Defined by the Bulk Dielectric Constant and Atomic Surface Tensions. J. Phys. Chem. B 2009, 113, 6378−6396. (55) Hay, P. J.; Wadt, W. R. Ab initio effective core potentials for molecular calculations - potentials for K to Au including the outermost core orbitals. J. Chem. Phys. 1985, 82, 299−310. (56) Devlin, F. J.; Stephens, P. J. Ab-Initio Calculation of Vibrational Circular-Dichroism Spectra of Chiral Natural-Products Using Mp2 Force-Fields - Camphor. J. Am. Chem. Soc. 1994, 116, 5003−5004. (57) Debie, E.; Jaspers, L.; Bultinck, P.; Herrebout, W.; Van der Veken, B. Induced solvent chirality: A VCD study of camphor in CDCl3. Chem. Phys. Lett. 2008, 450, 426−430. (58) Abbate, S.; Burgi, L. F.; Gangemi, F.; Gangemi, R.; Lebon, F.; Longhi, G.; Pultz, V. M.; Lightner, D. A. Comparative analysis of IR and vibrational circular dichroism spectra for a series of camphorrelated molecules. J. Phys. Chem. A 2009, 113, 11390−405. (59) Morita, H. E.; Kodama, T. S.; Tanaka, T. Chirality of camphor derivatives by density functional theory. Chirality 2006, 18, 783−789. (60) Buffeteau, T.; Cavagnat, D.; Bouchet, A.; Brotin, T. Vibrational absorption and circular dichroism studies of (−)-camphanic acid. J. Phys. Chem. A 2007, 111, 1045−1051. (61) Nicu, V. P.; Debie, E.; Herrebout, W.; Van der Veken, B.; Bultinck, P.; Baerends, E. J. A VCD robust mode analysis of induced chirality: the case of pulegone in chloroform. Chirality 2009, 21 (Suppl1), E287−E297. (62) Aviles-Moreno, J. R.; Horno, E. U.; Urena, F. P.; Gonzalez, J. J. L. IR-Raman-VCD study of R-(+)-Pulegone: Influence of the solvent. Spectrochim. Acta, Part A 2011, 79, 767−776. (63) Debie, E.; Bultinck, P.; Herrebout, W.; van der Veken, B. Solvent effects on IR and VCD spectra of natural products: An experimental and theoretical VCD study of pulegone. Phys. Chem. Chem. Phys. 2008, 10, 3498−3508. (64) Abbate, S.; Longhi, G.; Boiadjiev, S.; Lightner, D. A.; Bertucci, C.; Salvadori, P. Analysis of vibrational circular dichroism data in the near infrared and visible range. Enantiomer 1998, 3, 337−347. (65) Freedman, T. B.; Long, F.; Citra, M.; Nafie, L. A. Hydrogenstretching vibrational circular dichroism spectroscopy: Absolute configuration and solution conformation of selected pharmaceutical molecules. Enantiomer 1999, 4, 103−119. (66) Izumi, H.; Ogata, A.; Nafie, L. A.; Dukor, R. K. A revised conformational code for the exhaustive analysis of conformers with one-to-one correspondence between conformation and code: Application to the VCD analysis of (S)-ibuprofen. J. Org. Chem. 2009, 74, 1231−1236. (67) Freedman, T. B.; Cao, X. L.; Dukor, R. K.; Nafie, L. A. Absolute configuration determination of chiral molecules in the solution state using vibrational circular dichroism. Chirality 2003, 15, 743−758. (68) Kuppens, T.; Herrebout, W.; van der Veken, B.; Bultinck, P. Intermolecular association of tetrahydrofuran-2-carboxylic acid in solution: A vibrational circular dichroism study. J. Phys. Chem. A 2006, 110, 10191−10200. (69) He, J. T.; Polavarapu, P. L. Determination of intermolecular hydrogen bonded conformers of alpha-aryloxypropanoic acids using density functional theory predictions of vibrational absorption and vibrational circular dichroism spectra. J. Chem. Theory Comput. 2005, 1, 506−514. (70) Gobi, S.; Vass, E.; Magyarfalvi, G.; Tarczay, G. Effects of strong and weak hydrogen bond formation on VCD spectra: a case study of 2-chloropropionic acid. Phys. Chem. Chem. Phys. 2011, 13, 13972− 13984. (71) Nicu, V. P.; Neugebauer, J.; Baerends, E. J. Effects of complex formation on vibrational circular dichroism spectra. J. Phys. Chem. A 2008, 112, 6978−6991. (72) Losada, M.; Tran, H.; Xu, Y. Lactic acid in solution: investigations of lactic acid self-aggregation and hydrogen bonding interactions with water and methanol using vibrational absorption and Q



Article

vibrational circular dichroism spectroscopies. J. Chem. Phys. 2008, 128, 014508:1−014508:11. (73) Polavarapu, P. L.; Donahue, E. A.; Hammer, K. C.; Raghavan, V.; Shanmugam, G.; Ibnusaud, I.; Nair, D. S.; Gopinath, C.; Habel, D. Chiroptical Spectroscopy of Natural Products: Avoiding the Aggregation Effects of Chiral Carboxylic Acids. J. Nat. Prod. 2012, 75, 1441−1450. (74) Pierce, M. E.; Parsons, R. L.; Radesca, L. A.; Lo, Y. S.; Silverman, S.; Moore, J. R.; Islam, Q.; Choudhury, A.; Fortunak, J. M. D.; Nguyen, D.; Luo, C.; Morgan, S. J.; Davis, W. P.; Confalone, P. N.; Chen, C. Y.; Tillyer, R. D.; Frey, L.; Tan, L. S.; Xu, F.; Zhao, D. L.; Thompson, A. S.; Corley, E. G.; Grabowski, E. J. J.; Reamer, R.; Reider, P. J. Practical asymmetric synthesis of efavirenz (DMP 266), an HIV-1 reverse transcriptase inhibitor. J. Org. Chem. 1998, 63, 8536−8543. (75) Winrow, C. J.; Gotter, A. L.; Cox, C. D.; Tannenbaum, P. L.; Garson, S. L.; Doran, S. M.; Breslin, M. J.; Schreier, J. D.; Fox, S. V.; Harrell, C. M.; Stevens, J.; Reiss, D. R.; Cui, D.; Coleman, P. J.; Renger, J. J. Pharmacological characterization of MK-6096 - a dual orexin receptor antagonist for insomnia. Neuropharmacology 2012, 62, 978−987. (76) Coleman, P. J.; Schreier, J. D.; Cox, C. D.; Breslin, M. J.; Whitman, D. B.; Bogusky, M. J.; McGaughey, G. B.; Bednar, R. A.; Lemaire, W.; Doran, S. M.; Fox, S. V.; Garson, S. L.; Gotter, A. L.; Harrell, C. M.; Reiss, D. R.; Cabalu, T. D.; Cui, D.; Prueksaritanont, T.; Stevens, J.; Tannenbaum, P. L.; Ball, R. G.; Stellabott, J.; Young, S. D.; Hartman, G. D.; Winrow, C. J.; Renger, J. J. Discovery of [(2R,5R)-5-{[(5-fluoropyridin-2-yl)oxy]methyl}-2-methylpiperidin-1yl][5-methyl-2 -(pyrimidin-2-yl)phenyl]methanone (MK-6096): A dual orexin receptor antagonist with potent sleep-promoting properties. ChemMedChem 2012, 7, 415−424. (77) Girardin, M.; Ouellet, S. G.; Gauvreau, D.; Moore, J. C.; Hughes, G.; Devine, P. N.; O’shea, P. D.; Campeau, L. C. Convergent Kilogram-Scale Synthesis of Dual Orexin Receptor Antagonist. Org. Process Res. Dev. 2013, 17, 61−68. (78) Sturino, C. F.; O’Neill, G.; Lachance, N.; Boyd, M.; Berthelette, C.; Labelle, M.; Li, L. H.; Roy, B.; Scheigetz, J.; Tsou, N.; Aubin, Y.; Bateman, K. P.; Chauret, N.; Day, S. H.; Levesque, J. F.; Seto, C.; Silva, J. H.; Trimble, L. A.; Carriere, M. C.; Denis, D.; Greig, G.; Kargman, S.; Lamontagne, S.; Mathieu, M. C.; Sawyer, N.; Slipetz, D.; Abraham, W. M.; Jones, T.; McAuliffe, M.; Piechuta, H.; Nicoll-Griffith, D. A.; Wang, Z. Y.; Zamboni, R.; Young, R. N.; Metters, K. M. Discovery of a potent and selective prostaglandin D-2 receptor antagonist, [(3R)-4(4-chlorobenzyl)-7-fluoro-5-(methylsulfonyl)-1,2,3,4tetrahydrocyclopenta[b]indol-3-yl]acetic acid (MK-0524). J. Med. Chem. 2007, 50, 794−806. (79) Zhao, M. M.; McNamara, J. M.; Ho, G. J.; Emerson, K. M.; Song, Z. J.; Tschaen, D. M.; Brands, K. M.; Dolling, U. H.; Grabowski, E. J.; Reider, P. J.; Cottrell, I. F.; Ashwood, M. S.; Bishop, B. C. Practical asymmetric synthesis of aprepitant, a potent human NK-1 receptor antagonist, via a stereoselective Lewis acid-catalyzed trans acetalization reaction. J. Org. Chem. 2002, 67, 6743−6747. (80) Brands, K. M.; Payack, J. F.; Rosen, J. D.; Nelson, T. D.; Candelario, A.; Huffman, M. A.; Zhao, M. M.; Li, J.; Craig, B.; Song, Z. J.; Tschaen, D. M.; Hansen, K.; Devine, P. N.; Pye, P. J.; Rossen, K.; Dormer, P. G.; Reamer, R. A.; Welch, C. J.; Mathre, D. J.; Tsou, N. N.; McNamara, J. M.; Reider, P. J. Efficient synthesis of NK(1) receptor antagonist aprepitant using a crystallization-induced diastereoselective transformation. J. Am. Chem. Soc. 2003, 125, 2129−35. (81) Fu, X. Y.; McAllister, T. L.; Thiruvengadam, T. K.; Tann, C. H.; Su, D. Process for preparing ezetimibe intermediate by an acid enhanced chemo- and enantioselective CBS catalyzed ketone reduction. Tetrahedron Lett. 2003, 44, 801−804. (82) Hartman, G. D.; Halczenko, W.; Duggan, M. E.; Imagire, J. S.; Smith, R. L.; Pitzenberger, S. M.; Fitzpatrick, S. L.; Alberts, A. W.; Bostedor, R.; Chao, Y. S.; Germershausen, J. I.; Gilfillan, J. L.; Hunt, V. 3-Hydroxy-3-methylglutaryl-coenzyme A Reductase Inhibitors. 9. The Synthesis and Biological Evaluation of Novel Simvastatin Analogs. J. Med. Chem. 1992, 35, 3813−3821.

(83) Lee, T. J.; Holtz, W. J.; Smith, R. L. Structural Modification of Mevinolin. J. Org. Chem. 1982, 47, 4750−4757. (84) Nicu, V. P.; Baerends, E. J. Robust normal modes in vibrational circular dichroism spectra. Phys. Chem. Chem. Phys. 2009, 11, 6107− 6118.

R


Systematic Approach to Conformational Sampling for Assigning

Recommend Documents