Determination of Configuration and Conformation of a Reserpine Derivative with Seven Stereogenic Centers Using Molecular Dynamics with RDC‐Derived Tensorial Constraints

Abstract NMR‐based determination of the configuration of complex molecules containing many stereocenters is often not possible using traditional NOE data and coupling patterns. Making use of residual dipolar couplings (RDCs), we were able to determine the relative configuration of a natural product containing seven stereocenters, including a chiral amine lacking direct RDC data. To identify the correct relative configuration out of 32 possible ones, experimental RDCs were used in three different approaches for data interpretation: by fitting experimental data based singular value decomposition (SVD) using a single alignment tensor and either (i) a single conformer or (ii) multiple conformers, or alternatively (iii) using molecular dynamics simulations with tensorial orientational constraints (MDOC). Even though in all three approaches one and the same configuration could be selected and clear discrimination between possible configurations was achieved, the experimental data was not fully satisfied by the methods based on single tensor approaches. While these two approaches are faster, only MDOC is able to fully reproduce experimental results, as the obtained conformational ensemble adequately covers the conformational space necessary to describe the molecule with inherent flexibility.


Introduction
Natural products playacritical role in drug discovery,a st hey provide access to new chemotypesa nd structural diversity. The substances are extracted and isolatedf rom differentl iving organismsa nd testedf or activity in assays against molecular targets of diseases.I no rder to take advantageo fn atural products as potential drug candidates, their exact chemical structure and configuration needs to be known.T his knowledge will allow to re-synthetize them in larger quantities andg ain accesst os tructural isomers. Structurea nd configuration determinationc an be achieved by X-ray crystallography,i fsufficient materiali sa vailable and diffracting crystals are obtained. Unfortunately,t his is at ime-consumingp rocess and in many cases it is not possible at all. Therefore, NMR spectroscopy is the more prevalent technique for determining the chemical structure of natural products. However, investigatingt he configuration of molecules containing many stereocenters is often not possible based on traditional NOE data and coupling patterns.I ns uch cases, complementary NMR data needst ob er ecordeda nd interpreted.I nr ecent years, residual dipolar couplingsh ave become an attractive source of such additional data, [1] which will be furtherexploitedi nt his manuscript.
The reserpine derivative RD-1 represents ac ase, where the assignment of the relative stereochemistry wasn ot possible based on traditional NMR analysis. Reserpine is aw ell-known natural product:i ti sa na lkaloid found in the roots of Rauwolfia serpentina and Rauwolfia vomitoria (for the structure of reserpinew er efer the reader to the supporting information) [2] Pseudoreserpic acid derivatives are useful as sedatives and antihypertensives. [3] The core of these molecules contains at ertiary amine, which is part of aq uinolizidine system, that is, two fused cyclohexane rings with an itrogen at the bridgehead position.
In total, the molecule contains seven stereocenters including the tertiary amine, which is treated herea sa na dditional chiral center.I ti sk nown that inversion of the nitrogenp yramid in amines can readily take place via ac hange in sp 3 hybridization of the nitrogen atom by passing throughaplanar sp 2 state [3b, 4] This inversion process in most amines is too fast to be detectable by NMR. However,i ns ome cases suitable substituents can sufficiently stabilize the two interchanging states and allow the separation of the two isomers.F or this molecule it was initially not obvious which of the cases applies, fast inversion or presence of just one conformation.
The aim of the study was the reliable determination of the relative stereochemistry of the reserpine derivativeR D-1 and the establishment of ap rotocol that could be appliedi nt he same wayt oo ther,s imilarm olecules. To achieve this, we also compared different computational methods for RDC-based configurational analysisu sing state-of-the-art approaches like SVD fitting as implemented in the program MSpin [5] and molecular dynamics (MD)s imulations with orientational RDC constraints( MDOC)a simplemented in COSMOS. [6] The latter program allows an effective coverageo fa ll possible conformations in an 80 ns long MD simulationt hat fulfils all experimental constraints averaged over the resulting structural and orientationale nsemblea sg ood as possible. [7] MSpin,i nc ontrast, only works on single or few pre-selectedf ixed 3D conformations of the molecules under study.S everald ifferenta pproaches are implemented in MSpin to fit or predict RDC data, of which three were tested here, namely: single conformer single alignmentt ensor fits and multiple conformers single alignment tensorf its, which are both based on singular value decomposition (SVD) [8] and additionally the TRAMITE prediction, which uses an approximation of RDCs from tensors of inertia. [9] From the three possible MSpin approaches as theg old standard in RDC-based structure determination of small molecules, the last had to be excluded at an early stage, as the implemented prediction did not lead to good agreements with single optimized conformers, indicating that more complex prediction methods like the one proposedb yF ranke tal. [10] or by Ibµnez de Opakua et al. [11] might be advisable.
In this manuscript, we show that the determination of the relative stereochemistry of the seven stereocentersr eserpine derivativeR D-1Àar elativelyc omplex molecule which exhibits ac ertain degree of flexibility-can be achieved by the applicable MSpina nd COSMOSR DC analysis methods;h owever,w ith significant differences concerning the fulfilmento fe xperimental constraints and the relatedt reatment of the inherent flexibility in RD-1. We address the potential inversion of the tertiary amine by 1 J CH coupling constants, its stereochemistry by 1 D CH couplings, and give ad etailed comparison of the static versus the molecular dynamics-based approaches as tools for the analysis of relative configurationa nd conformationo fc omplex molecules with inherentf lexibility.T he configurational assignment is verified with X-ray analysisw hich was performed independently.

Experimental RDC data
After the assignment of all NMR-active nuclei using standard procedures, residual dipolar coupling data were derivedf rom the measured one-bondc ouplingc onstants of the reserpine RD-1 recorded under isotropic ( 1 J CH )a nd anisotropic ( 1 T CH = 1 J CH + 1 D CH )c onditions, respectively.P artial alignment of the sample was achievedu sing ap olymer-based alignment medium. While al arge number of polymer gels [1c, 12] and liquid crystalline phases [13] have been reported in the literature, polymer gels in combination with ar ubber-based stretching device [14] in our hands provide the best flexibility in adjusting the alignment strength.A s[ D 6 ]DMSO was used as solventf or the isotropic measurements, we chose ap olyacrylonitrile (PAN) gel [12h, i] as the alignment medium for the anisotropic sample. Anisotropic data were recorded at ad euterium quadrupolar splitting of Dn Q = 6Hza sabalanced compromise between signalw idth and residual dipolarc oupling size. 1 D CH couplings were calculated from the difference betweent he scalar couplings 1 J CH of the isotropic sample and the total splitting of the anisotropic sample ( 1 T CH )a sm easured by CLIP-HSQC [15] and P. E.HSQC [16] experiments.Atotal of 21 1 J CH and corresponding 1 D CH couplings rangingf rom À25.6 to 32.1 Hz were thus accessible with errors ranging between0 .3 and5Hz as determined by the procedure for maximum error estimates as described by Kummerlçwe et al. [17] (see Ta ble S1). The experimental results for selected 1 J CH couplings in CH and CH 2 groups in proximity of the aminea re also summarized in Figure 2a nd Ta ble 1. As both methoxy groups showa veraged experimental values, corresponding RDCs wereo mitted in the single conformer fit and for better correspondence also in the multi conformer single alignment tensor fit.

Treatment of the tertiary amine
The determination of the configuration at the amine N8 has not been possible with standard NMR methods in isotropic solution such as NOE-derived distances and dihedrala ngles from 3 J-couplings. The aliphatic CH 2 and CH groups at positions 7, 9, and 23 provideasignificant stabilization of the tertiarya mine, but ap otential rapid inversion cannotb ee xcluded ap riori. A theoreticals tructurals earch for conformations with inverted amines showed little differences in energies of the two isoforms and it was necessary to find aw ay to experimentally give evidencef or the presence or absence of amine inversion.
The unambiguous solution to the problem was found as a side product of the one-bond measurements:i fr apid inversion would occur,a ll structural parameters like chemical shifts and especially scalar couplings of the neighboring aliphatic groups should be averaged. The experimentally determined 1 J CH coupling constants of the axial protons of CH 2 7a nd 9( 128.4 and 128.7 Hz, respectively)a re much smaller than the 1 J CH couplings of the corresponding equatorial protons (137.0 and 136.4 Hz, respectively). [18] Equally,c hemical shifts are significantly different for the protons within the CH 2 9, 2.93 ppm for the equatorial proton compared to 2.39 ppm for the axial proton. It can therefore be concluded that one configuration strongly dominates at the amine and potential inversion to a good approximation can be neglected.

Generation of 32 possible diastereomers
RD-1 contains 7c hiral centers. Twoc hiral centers (2 and 4i n Figure 1) are within as terically hindered lactone ring, therefore only ar educed number of configurations is sterically accessible. In total, 32 relative configurations listed in Table 2a re possible, neglecting all enantiomers with S-configuration at C2. Static structuralm odels of the 32 different configurations have been generated with the program CORINA [19] and subsequently optimized using the Schrçdinger softwarep ackage Release 2014-2 (Maestro, Schrçdinger,L LC, New York, NY,2 016). Corresponding 3D structures with minimal energiesa sw ell as ensembles of energy-optimal structuresw ereu sed for the following MSpin calculations and as initial structures for the COSMOS MDOC runs.

Single conformer singlealignment tensor SVD approach
Each geometry-optimized configuration was fitted in MSpin to determine ag lobal alignment tensoru sing the SVD option. Resulting RDCs were back-calculated and compared with experimental data. The MSpin softwarep ackage uses the Cornilescu Q [20] as af ittingc riterion. This quality factor allows the relative evaluation of best fitting structuralm odelso ut of ap redefined set. However,n oa bsolute evaluation of the fulfillment of experimental constraints is possible. This can be much better achieved by the previously introduced quality factor (n/c 2 ), [6c] which is explained further in the experimentals ection and has been used throughout the manuscript. As tructural model that fully complies with experimental data within the experimental error mustr esult in n/c 2 > 1; consequently,avalue below 1i sa clear indication that at least one experimental value is outside its corresponding experimental error.I na ddition to the n/c 2 qualityf actor,w ea lso give the number of such outliers for every structuralm odel as another importantn umber for evaluation. In this study we provide all experimental errors as maximum error estimates using the procedure defined by Kummerlçwe et al. [17] As these error estimates correspond to av ery high confidence level of approximately three times the standard deviation (3s), as ingle outlieri si np rinciple sufficient to falsify as tructural model, as the model does not reproduce the experimental data within the error.
[a] Running numberi dentifyingt he configurations;[ b] R and S identify the configurationo ft he stereocenters in the following sequence: C2 C3 C4 C5 C6 N8 C23. Note that for each relative configurationamirror image (enantiomer) exists that is not listed and for whicht he sameanalysis can be applied.
In Figure 3, quality factorso btained from the MSpin calculations for the single best conformers of the 32 configurations are summarized. The RRSSSSR configurationh as noticeably the best n/c 2 value,c learly identifying this configuration as the best fitting one out of the set of 32 singlec onformers. For most practical applicationst his result might be sufficient. However,t he value for the best fitting configurationo fn / c 2 = 0.093 also leads to the conclusion that the model is by far not sufficient to represent the experimental results. The overall quality factor is severely below 1a nd more than 10 outliers disqualify the structural model.A st he overall fit of experimentalv ersus back-calculated RDCs correlates well (see Figure 4), the most probable reason for the insufficientf ulfillment of experimental data using the single conformer approach is the presence of flexibility in the RD-1 molecule. As dynamics cannot be included in as ingle conformer approach, the extension to am ultiple conformer analysiswas attempted in the next step.

Multiple conformer singlealignment tensor SVD approach
From the existing approaches that use ac ombination of conformations, only the multiple conformer single alignment tensor SVD-based fitting procedure is viable.T he single-tensor approacha llows the calculation of RDCs based on as ingle alignment tensor for an ensembleo fconformers of ag iven configuration. Clearly,f itting as ingle alignment tensor to a conformational ensemble is an approximation, which,however, successfully led to several configurationd eterminations in the past. [1a, 21] In our study,anumber of different conformers was generated from ac onformational search fore ach configuration with the softwareM acromodel. Using the defaults ettings of the Monte Carlo-based program with an OPLS 2005 force field, all conformations with an energy difference below 6.0 kcal mol À1 relative to the lowest energy structure were collected. In af irst approach, all of these low-energy conformers werec ombined to ensembles by weighing the population of each conformer according to its energy.T he theoretical RDCs were then calculated using the single tensor approach. Resulting RDCs did not improvec ompared to the single conformer single alignment tensor fit and data are not shown. We then used the input ensemble of conformers for af it where also the populations are optimized as the calculated energy-differences might not represent the actual energies.W ith this approach, configuration 11 still hasb yf ar the bestn / c 2 value (see Figure 5) and al ook at Figure 6s hows the qualitativelyg ood correlation of experimental and calculated data. However,t he number of outliers remains high:1 0c alculated RDCs are still outside the range of the maximum error estimates of the experimental RDCs. The result obtained for the best configuration1 1d oes practically not change compared to the single conformer approach. For this configuration, four different conformers are obtained with Maestro. These, however,o nly differ in the position of the methoxy groups that is, the core frame of the molecule is essentially the same for the different conformers andt herefore very similar D calc and consequently n/c 2 values are obtained. On the other hand, for some other configurations up to 16 different conformers are obtained by the Maestro software, partially with significant differences in the core frame.For example configuration 1h as 16 conformers and the SVD approach led to an improvement in n/c 2 from 0.0008 of as ingle conformer to approximately 0.002 for the multiple conformer fit. The multiple conformer single alignment tensor approach could therefore very slightly improvet he overall qualityo fs tructural models,b ut consistency with experimentalr esults could not be achieved.

Dynamicsimulation with orientational constraints (MDOC)
The program package COSMOS has as pecialized protocol for time averaged molecular dynamics (MD) simulations with orientational constraints and also includes am olecular mechanics force field. [6c, 22] The orientation encoded in tensorial constraints derived from experimental RDC values is used as af actor for the generation of pseudo forces, which constrain the orientation of the molecule within the MD (for details of the approach see Tzvetkovae tal. [6c] ). For each MDOC step af ull tensorial orientation is calculated and compared with the experimental data. Based on analytical solutions for the first and second derivativeo ft he tensor with respect to the laboratory frame axes x, y and z,t he MDOC run will rotate individual CÀHb ond vectors to improve the overall pseudoe nergy depending on the difference between each calculated and experimental RDC. For as uccessful MD run, different parameters have to be optimized,like the strength of pseudoforces (k)and the alignment scaling factor (s AM ), which reduces the calculation time by avoidingu nnecessary computation of isotropic tumbling of the molecule of interest. [6c] In the optimized runs, we set k to 5.5.10 À4 and s AM to 4.10 À3 to obtain calculated RDC values in the range of 1 D CH = À25.6 to 32.1 Hz (see Table 3f or the individualv alues). The MD runs were chosen to last for 80 ns in order to reach good convergence of the orientationalc onstraints.
During the MDOC runs, at discretet ime points as defined in the MDOC settings, the calculated 1 D CH RDCs are written into an output file (here every 20 ps). At the evaluation of the MDOC runs, these 1 D CH couplings are arithmeticallya veraged. The n/c 2 is then calculated from these averagedv alues neglecting the first nanosecond of the MDOC trajectory,w hich is neededf or initial system equilibration. In addition, every 40 ps ag eometry snapshot is saved as ac ontrol for the MD run.
The qualityf actors of the data obtained from COSMOS are summarized in Figure 7. While three configurations (4, 5, and 11)h ave overall quality factors n/c 2 above 1a nd therefore potentially comply with experimental data, ad etailed analysis re-  Table 2. The color of the bar encodes the numbero fo utliers from the measured RDCs values,a ss hownint he colors cale on the upper right. Event he best static ensemblehas 10 RDCs thatd onot comply with data within experimental errors. Figure 6. Plot of back-calculated RDCs obtained with multiple conformer single tensor fit, against experimental RDCs for the best configuration1 1 (RRSSSSR). Although the overall fit of experimental versus calculated data shows at first sight ag ood correlation, 10 out of the 19 RDCs are not within the range of maximum error estimates of the experimental data (red diamonds). The diagonal line represents the case of full correlation of experimentala nd calculated data. Only 19 out of 21 experimental RDCsa re used in MSpin, because the two flexible methoxy groups (26 and 28 in Figure 1) have been excluded for the analysiswith this software. veals that only configuration 11 (RRSSSSR)f ulfills all experimental constraints within the experimental errors while the other two matching structural models have two and five outliers, respectively.T he comparison of measured and calculated RDCs, as shown in Figure 8, furthers upports the stereochemistry of configuration 11 (RRSSSSR)a st he correct one. Following the evaluation of 1 J CH coupling constants as described above, amine inversion was prevented during MDOC runs by fixed distances in the amine surrounding as described in the SI. Without the additional distance constraints amine inversion occurs andc onfigurations differing only by an inversion of N8 are virtually indistinguishable (see SI), leading, how-ever,t ov ery similar resultsw ith justs lightly reduced quality factors.

X-ray data
In parallel to the NMR efforts, we were able to determine the crystal structure of RD-1. The structurea nalysis allowed determiningt he configurationo fR D-1 as configuration1 1f rom NMR-based analysis, which is in full agreement with the RDCbased NMR analysiso fa ll seven stereocenters. Based on the presence of anomalous scatterers oxygen and nitrogen, the absolute configuration of RD-1 could be unambiguously assigned as C2R, C3R, C4S, C5S, C6S, N8S, C23R (numbersr efer to X-ray structure labellings cheme shown in Figure 9). The result is supported by aF lack x parameter of 0.02 (13). [23] Comparisono fN MR-based approaches Out of previously reported direct fitting approaches for the determination of conformational ensembles based on residual dipolar couplings, to the knowledge of the authors only two approaches are practically feasible for medium-sized organic moleculesl ike RD-1 with al imited set of RDCs.A ll other approaches are either too complex to be applied to this class of molecules, as for example, methods based on the mean field additive potentialp rinciple, [24] or too few RDCs are accessible for areliable fit, as is the case for the multiple conformer multiple alignment tensorf it. The two remaininga pproaches used here are both based on fitting as ingle alignment tensor either to as ingle rigid conformer or to as et of selected conformers using singular value decomposition (SVD). In both cases an ensemble of possible conformers, usually representing lowestenergy structures,i sp reselected typically based on ab initio calculations or other computer aided structure elucidation approaches.
In contrast to these direct fitting methods, constrained molecular dynamics (MD) simulations are used to optimize a single structure (e.g. via simulated annealing) or an ensemble of structures that best fit experimental results. Several implementationsh ave been reported based on an approximated  Table 2. The color of the bar encodes the number of outliersoft he measured RDC values (i.e. 1/c 2 < 1). For comparison, the result of the multiple conformer single tensor fit is givenw ith the same scale (samed ata as in Figure 5) emphasizing the much improved agreement with the data provided by MDOC (B).  alignment tensor as initial input. [25] An alignment tensor for a flexible molecule, however,i sper se ill-defined and the recently published MDOCa pproachs eems far more appropriate in this case. [6c, 7a] Out of the three approaches applied to RD-1, both alignment tensor approaches led to basically identicalr esults:f rom the preselected conformers of all possible 32 relative configurations clearly the correct one, which was simultaneously identified by X-ray diffraction, fitted the data best, demonstrating again the enormousp otentialo fR DCs for structure and in particular configuration determination. The molecule with seven stereogenic centers represents ac omplexs tructure that at first sight might be considered relatively rigid. The configuration and dynamics at the amine N8 could not be determined by using NOE and 3 J-coupling data alone. Based on the measurement of 1 J CH constants (see Table 1) and using the Perlin effect, fast inversiona tN 8c an be excludedt oal arge extend. However,the preselected conformers did not allow producing astructural ensemble that fully agreedw ith experimental data using the SVD fitting approach. Even in the best case 10 out of 19 RDCs were outside the error margins of the experiment.A pparently,a dditional conformers would have to be taken into account that were not part of as tandard conformational search procedure, and potentiallym ultiple individual alignment tensors would have been needed. Ac ertain deviation of RDCs is expectedf rom vibrationalc orrection, [26] but the large differences cannotfully be explainedt his way.
The MDOCa pproach, in contrast, has no pre-assumptions concerning the conformationals pace andt he correct structure results in an orientational and conformational ensemblet hat fully reproduces the experimental data. The resulting structural ensemble reveals the large dynamics of RD-1 (Figure 10), partly caused by vibrational motions, but also by distinct conformational changes, as visualized by the time-dependence and distributiono ft he example dihedrala ngle H 10A -C 10 -C 9 -H 9A .A ccording to the dihedrald istribution of the MDOC ensembleu pt o 10 %p opulation can be attributed to at least one minor conformation which is not per se obvious in the NMR experimental data. The few accessible scalarc oupling constants did not allow the detection of such am inor conformer,b ut line broadening in the C-ring has been observed, corroborating qualitatively the determined flexibilitya nd the existence of am inor conformer. As the corresponding conformer is missing in the preselection of the alignment tensorf itting approaches, it explains the inability of both single andm ultiple conformer fits to reproduce the experimental RDCs.
As the full accessible conformational space is only restricted by RDC constraints in the MDOC approach, the ability to discriminate the different relative configurations must necessarily be reduced. Still, if both n/c 2 values and the number of outliers are taken into account, the correct configuration1 1i su nambiguously determined.

Conclusions
In summary,w es how with the example of ah ighly complex molecule that RDCs represent av aluable tool for the determi-nation of the conformation and configuration of even very complex organic molecules and demonstrate the abilities of state-of-the-art data interpretation approaches for the distinction of relative configurations and the determination of a structurale nsemble that fulfills all experimental constraints within error margins. The configuration of seven stereogenic centersi napartially flexible molecule (RD-1)c ould be determined,i ncluding the stereochemistry of an amine. Thisw as not possible using only standard NMR data (NOEs and scalar couplings). For ap artially flexible molecule as shown here, the fast SVD approachw ith preselected conformers fore ach configuration led to ac lear discrimination of the correct versus wrong configurations of RD-1, corroborating the CASE-3Da nalysis approach. [27] With al imited set of RDCs this approachi s certainly the way to go for as imple configurational analysis. However,e xperimental data could not be reproduced within their error ranges, indicating ac lear lack in the coverage of conformational space. The latter was straightforwardly achieved by the so-called MDOC (molecular dynamics with orien- tationalc onstraints) approach, which resulted in av alid conformationale nsemble and also provided ac lear distinction of diastereomers unbiased by al owest-energy preselection of conformers. In essence, all three RDC-based approaches provide a clear and correct determination of configurationo ft he complex molecule RD-1, provenf or six out of the seven stereocenters by X-ray crystallography (except the amine), and the recently introduced MDOC approach in addition leads to av alid orientational and conformational ensemble, demonstrating the unique power of anisotropic NMR parameters and the high level of state-of-the-art data interpretation.

Experimental Section
Sample preparation:T he isotropic sample of RD-1 was prepared by dissolving 2.8 mg in 0.5 mL of [D 6 ]DMSO leading to af inal concentration of 14.7 m.The partially aligned sample in PAN/[D 6 ]DMSO contained ad ry polymer stick of cross-linked PANp laced inside the Kalrez tubing of the stretching apparatus with 300 mL[ D 6 ]DMSO and 10 mg of compound leading to an approximate concentration of 87.2 mm.Ad ry PANp olymer stick of 3mmd iameter irradiated with accelerated electrons (200 kGy) was used. [12i] NMR spectra for the assignment in isotropic phase:A ll NMR measurements ( 1 H-1D, HSQC, COSY,R OESY and HMBC) for the assignment of RD-1 in [D 6 ]DMSO were recorded at 26 8Co naBruker 500 MHz AvanceI II spectrometer (500.09 MHz for 1 Ha nd 125.75 MHz for 13 C) equipped with a5mm BBFO probe head with actively shielded z-gradients. The 1 HNMR spectrum was acquired by using 64k data points at as pectral width of 12 kHz, and a1 .5 s repetition delay.2 D 1 H, 13 C-correlation spectra were recorded with 2k data points in the 1 Hd imension and 128 points in the 13 Cd imension. 2D 1 H, 1 H-correlation spectra were recorded with 2k data points and 128 points in the indirect dimension. The repetition delay for the 2D experiments was 1s.
NMR spectra for the RDC measurements:T he NMR spectra were recorded on Bruker 800 MHz Avance III HD spectrometer equipped with a5mm CPTCI inversely detected 1 H, 13 C, 15 Nt riple resonance cryogenically cooled probe with actively shielded z-gradients and frequencies of 800. 16 MHz for proton,201.20 MHz for carbon and 122.83 MHz for 2 H. The temperature was controlled with aB ruker SmartVT-unit to be 26 8Ct hroughout all experiments. To assess the introduced alignment using the stretching device, ad euterium spectrum was recorded leading to aq uadrupolar splitting of Dn Q = 6.3 Hz. The homogeneity of the alignment media was controlled by ad euterium imaging experiment. [28] The residual dipolar coupling values were obtained from CLIP-HSQC [15] and P. E.HSQC [16] experiments under both isotropic and anisotropic conditions. A P. E.HSQC spectrum was measured in addition in order to compare and confirm the residual dipolar couplings. All 2D spectra were acquired with 32k( 1 H)*512( 13 C) data points, unless stated otherwise. The repetition delay was set to 1s.T he 1 H, 13 C-CLIP-HSQC spectrum in isotropic condition was acquired with a6 4k( 1 H)*512( 13 C) data matrix, while the, P. E.HSQC in the isotropic case had 1.5k points in the indirect dimension. All spectra were processed using the software To pspin 3.2 and were apodized by a9 0 8 shifted sine squared window function for 13 Ca nd for 1 H, with prediction of 512 points and zero filling up to 2k points in the directly acquired dimension. The couplings were measured by superimposing the left side of the split signals with the right side of the same signal from a second copy of the same row of the 2D experiment. The experi-mental errors were determined as maximum error estimates following the procedure described in [17].
Generation of chemical structures for all configurations:T he initial three-dimensional structure of each possible configuration was built with CORINA. [19] All trial structures were energy minimized using Schrçdinger Release 2014-2 (Maestro, Schrçdinger,L LC, New York, NY,2 016). The conformational search for each configuration was realized with MacroModel (Schrçdinger Release 2014-2 Macromodel, Schrçdinger,L LC, New York, NY,2 016). OPLS 2005 was used as force field and the conformational research was done in Low-Frequency-Mode with default parameters in water.
MSpin:F itting experimental RDCs using singular value decomposition:T he fitting procedure of the experimental RDC data was performed by using the MSpin program. [5] 19 experimentally determined 1 D CH couplings from RD-1 and the coordinate files of 32 possible configurations were given as input data. The alignment tensors were determined using singular value decomposition (SVD). The fit between experimental and back-calculated RDCs is given by default with the Cornilescu quality factor Q [20] [Eq. (1)]: For our study the fit between experimental and back-calculated RDCs was expressed with the value n/c 2 in order to have au niform quality criterion with the MDOC results [6c] where ni st he number of experimentally obtained RDC values and c 2 is defined as [Eq.
(2), Eq. (3)]: where ir uns over all measured RDC data, D calc and D exp are the back-calculated and the experimental values, respectively,a nd DD i exp are the experimental errors of each D exp given as maximum error estimates. [17] Fitting experimental RDCs using single tensor fit with multiple conformers:F or this approach the MSpin program has been used. Ac ommon coordinate system for all conformations is determined by the superposition of the different geometries. The populations of conformers generated for ag iven configuration have been both weighted with the energies provided by Maestro (Maestro, Schrçdinger,L LC, New York, NY,2 016) using the molecular force-field OPLS 2005 in Macromodel (Schrçdinger Release 2014-2:M acromodel, Schrçdinger,L LC, New York, NY,2 016) or directly optimized within the fitting procedure. Data have been fitted using the single-tensor procedure and back-calculated RDCs (D calc )h ave been compared to the experimental data (D exp )a llowing the determination of the quality factor n/c 2 .
MDOC:T he MDOC simulations were performed using COSMOS 6.0 with the COSMOS force field. [6b] Each MDOC simulation was run for 80 ns with 160 million steps. Snapshot coordinates were saved every 40 ps resulting in 2000 snapshots for the flexibility analysis in MSpin. Different distance types were fixed during the MD simulations:o ne bond CÀHd istances and the distances between the carbons around the amine in order to avoid unphysical inversion of the amine. The experimental RDC data were used to determine the relative configuration of RD-1 as orientational constraints. For the MDOC run the dipolar couplings were scaled with as caling factor 4.0 10 À3 as previously described 6c .T he optimal value for the pseudo force constant for RD-1 was optimized to 5.5 10 À4 following the procedure described in [6c, 7a].T he temperature is monitored via at hermal bath in order to control the behavior of the molecule during the MD;i ti ss et to 300 K. The COSMOS output files with the back-calculated RDC values were evaluated via home-written scripts in Matlab (The MathWorks, Inc.). The first nanosecond of the MD runs were excluded to avoid equilibration artefacts and all further sampled values were used to generate arithmetic averaging of RDCs. Subsequently,a veraged RDC values obtained from 4000 snapshots were used to determine quality factors n/c 2 for all 32 relative configurations as shown in Figure 7.
X-ray diffraction:C rystals of the reserpine derivative RD-1 were obtained from as olution of RD-1 in tetrahydrofurane by slow evaporation of the solvent at room temperature. Diffraction data were collected at 100 Ko naB ruker AXS MicroStar diffractometer using aS MART 6000 CCD detector on at hree-circle platform goniometer with Cu(K a )r adiation (l = 1.54178 )f rom am icrofocus rotating anode generator equipped with Incoatec multilayer optics. 16 w-scans at different f-positions were performed to ensure appropriate data redundancy (5.9, Friedel pairs not merged). The crystal structure was solved by dual space-recycling methods and refined based on full-matrix least-squares on F2 using the SHELXTL program suite (Sheldrick GM (2001)). Anisotropic displacement parameters were used for all non-hydrogen atoms. Hydrogen atoms were located in aD Fm ap and refined in idealized positions using ar iding model. The absolute structure was determined based on the anomalous scatterers present (O and N). For the C2R, C3R, C4S, C5S, C6S, N8S, C23R diastereomer the Flack x parameter refined to 0.02 (13). In the crystal structure the amine is only present in one configuration (S). This prevalence is, however,i .e. based on data collected on one crystal only and might be solid-state-driven, it is not necessarily reflecting the distribution of the N8R and N8S diastereomers in solution.
Deposition number 19997972 contains the supplementary crystallographic data for this paper.T hese data are provided free of charge by the joint Cambridge Crystallographic Data Centre and Fachinformationszentrum Karlsruhe Access Structures service.