Grid-based molecular modeling for pharmaceutical salt screening: Case example of...

Grid-Based Molecular Modeling for Pharmaceutical SaltScreening: Case Example of 3,4,6,7,8,9-Hexahydro-2H-pyrimido (1,2-a) Pyrimidinium Acetate

ROBERT B. HAMMOND, ROSE S. HASHIM, CAIYUN MA, KEVIN J. ROBERTS

Institute of Particle Science and Engineering, School of Process, Environmental and Materials Engineering,University of Leeds, Leeds LS2 9JT, United Kingdom

Received 21 November 2005; accepted 6 April 2006

Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.20657

ABSTRACT: The development of modeling capabilities for improving the efficiency withwhich solid-state pharmaceutical products can be developed is a key strategic goal for thepharmaceutical research and development sector. In this context, an important topic isthe salt-selection process associated with drug-product formulation development. In thisstudy, a systematic (grid-based) search method is used to predict the host/counter-ionbinding for a simple but representative organic salt (i.e., a type I organic acid salt formerhaving a single ionisable group): 3,4,6,7,8,9-hexahydro-2H-pyrimido (1,2-a) pyrimidi-nium acetate ([H2hpp][O2CCH3]). The relative disposition of the two ionicmoieties in theasymmetric unit and, from this, the crystal packing in this compound are also predictedusing the systematic grid-based searchmethod linked to the known crystallographic unitcell dimensions. The overall strategy adopted encompasses three main steps: molecularpair search; optimization and clustering; and crystal lattice search and optimization. Thepredicted results, using this method, reveal a good agreement between the calculatedcrystal structure and that obtained from the Cambridge Crystallographic StructureDatabase (CCSD), indicating that the approach offers considerable promise forapplication as part of an integrated strategy for pharmaceutical salt selection.� 2006Wiley-Liss, Inc. and theAmericanPharmacists Association J PharmSci 95:2361–2372, 2006

Keywords: solid dosage form; drug design; salt selection; co-crystals; molecularmodeling; systematic search; crystal structure; molecular simulations; crystallography;3,4,6,7,8,9-hexahydro-2H-pyrimido (1,2-a) pyrimidinium acetate

INTRODUCTION

With the advent of genomics, combinatorialchemistry, computational modeling, and proteo-mics, the unprecedented abundance of thera-peutic targets and associated potential drugcompounds has necessitated a fundamental shiftin the methods used for evaluating pharmaceu-tical properties and the impact of such propertieson drug development.1,2 As the formulation of

drug compounds in their solid dosage salt formpredominates in the market, an understandingof the influence of the physical properties of apharmaceutical salt’s crystalline form on thephysical and chemical properties downstreamfor the finished product can be most useful toensure the selection of the best salt-formingspecies. This choice influences various propertiessuch as melting point, hygroscopicity and pro-pensity to hydration, chemical and physicalstability, mechanical properties (morphology ofparticles, agglomeration, flowability) etc.

Salt formation is an acid–base reaction invol-ving either a proton-transfer or neutralizationreaction. A salt in solid form will dissolve in waterand immediately dissociate to the conjugate acid of

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 11, NOVEMBER 2006 2361

Correspondence to: Kevin J. Roberts (Telephone: þ44-0-1133432408; Fax: þ44-0-113 3432405;E-mail: [email protected])

Journal of Pharmaceutical Sciences, Vol. 95, 2361–2372 (2006)� 2006 Wiley-Liss, Inc. and the American Pharmacists Association

the weak base and the conjugate base of the acid,which are also in equilibrium with their corre-sponding base and acid, as determined by theirpKa and the pHof the solution. If the concentrationof the base exceeds its solubility, it can precipitateout until equilibrium is maintained. Hence, thestate of this equilibrium can be determined bya number of factors including: solubility of thesalt, ionization constants (pKa) of the acid andbase, pH of the solution, and solubility of theweak base (intrinsic solubility) with any changesin these parameters being likely to affect theequilibrium.

Several reviews have outlined general strate-gies and considerations for salt selection and drugdiscovery.3–11 One of the first aspects is counter-ion selection for salt formation. For stable saltformation, it is generally accepted that thereshould be a minimum difference of 2–3 pKa unitsbetween the drug and the counter-ion with thechoice of the counter-ion depending on whether adecrease in crystal lattice forces is essential forimproving solubility. Other key considerations forsalt selection include the physico/chemical proper-ties that would influence its physical and chemicalstability under storage and processing, processa-bility under manufacturing conditions, and tox-icological impact of the counter-ions.

Research and development with advancedmolecular modeling methodologies12,13 provides apowerful tool to aid in the discovery of new saltforms, hence creating high value-added drugs andnew therapies. In addition, routine ab initiostructure prediction for crystalline molecularsolids is a significant long-term objective.14–17 Inthe context of the salts of pharmaceutical materi-als, it is valuable to examine the issue of counter-ion binding in materials where one or more of theionic moieties is/are molecular. In these cases,the binding energy associated with the molecularions alone can be expected to be a subtle balancebetween the coulombic and the nonspecific van derWaals interactions and, for example, specifichydrogen bonding interactions. Some of the moresubtle intermolecular interactionsmight affect thepredictions, and more sophisticated force fields,especially in the treatment of coulombic interac-tions, could have been adopted. However, thepurpose of this study was to assess the feasibilityof molecular modeling techniques testing thelimits of applicability using simple generic forcefield methods. Here, we examine the extent towhich the binding within an asymmetric unit,comprising the molecular ions within a salt dimer

structure, determines the packing within itsextended, three-dimensional crystal structureand, hence, serves as a potential methodology foruse in the structure prediction of co-crystals suchas, complexes, salts and solvates.

We takeas our startingpoint, for thepurposes ofthis investigation, the premise that the size of themolecular cluster, as a crystalline-form precursor,thatmust be considered to allow the assessment ofthe relative stability of any extended crystalstructure derived from that cluster is small. Thisbeing the case, and since a systematic approach forranking the stability of small clusters is tractable,such a procedure could also form the basis of aroute to full crystal structure prediction in duecourse. Attractively, this approach places calcula-tions for predicting crystal structures in thecontext of the crystal growth-environment since,for example, it is also possible to characterize therole of solvent molecules when examining andranking cluster stability for small clusters. Theapproach also reflects the fact that, intuitively, itseems clear that the molecular recognition pro-cesses involved in building the small molecularclusters, which underpin larger structures, mustbe key in the formation of a solid phase. To assessthe likelihood of the success of a method forstructure prediction for pharmaceutical saltsbased on an examination of counter-ion bindinggeometry and energies, the first stage adoptedwasto carry out a search of known crystal structuresfor these types of materials and, through this, toselect a suitable example to form the basis of a casestudy.

The crystal structure of a suitable organic saltwas selected from theCambridgeCrystallographicStructure Database18 (CCSD) for the purposes ofvalidating the overall approach adopted here. Theresults offer a promisingway forward towards newmodeling tools with applications for pharmaceu-tical design and drug product development.

SALT SELECTION FOR GRID-BASEDSEARCH METHODS

The CCSD18 was examined to select a suitablecandidate crystal structure for the application ofgrid-based search methods. The search was sub-divided into two groups, that is, being based oncounter-ion type (69 acids and 26 bases) and saltformer classes, that is, Class 1–3 using thecriteria of Pfaankuch et al.19 which are listedin Table 1, with the respective populations of

2362 HAMMOND ET AL.

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 11, NOVEMBER 2006 DOI 10.1002/jps

representative structures in the classes being30:27:12 for acids and 9:10:7 for bases.20

The survey showed that for the salt-formeracids in Class 1, acetic acid was, by a considerablemargin, the most common participant in 1244structures with other organic counter-ions. Over-all, the survey revealed that most organic salts inthe database have amolecular weight in the rangefrom 200 to 500. The breakdown of this survey oforganic salts within the CCSD is summarized inFigure 1.

Based on this search of the CCSD, the followingstrategy was adopted for the selection of a trialcandidate:

� Organic salt structure of type I with an acidiccounter-ion with strong hydrogen bondingbetween counter-ion and molecule.

� Counter-ion of the salt structure having onlyone ionisable group, that is, one equilibriumconstant (pKa) for the purpose of testing thesearch method.

Table 1. The Criteria of Salt Formers in Classes 1, 2 and 3 by Pfaankuch et al.19

Class of SaltFormers Criteria

Example

Salt FormerAcids

Salt FormerBases

Class 1 Salt formers that can be used without restriction becausethey can form physiologically ubiquitous ions

Acetic acid L-Arginine

They can occur as intermediate metabolites in biochemicalpathways

Glutamic acid Choline

Frequently used whilst in the past and presentClass 2 Salt formers that whilst not naturally occurring have,

through a number of applications, shown to exhibit lowtoxicity and good tolerability

Alginic acid Betaine

Oleic acid TromethamineClass 3 Salt-formers that are occasionally used, but mainly for the

purposes of achieving ion-pair formation and sometimessuitable to solve particular problems

Cinnamic acid Diethanolamine

Salicylic acid Piperazine

Figure 1. Flow chart showing the results of a survey of the CCSD for organic salts(numbers of compounds found from the CCSD are listed in the brackets).

MOLECULAR MODELING FOR SALT SCREENING 2363

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 95, NO. 11, NOVEMBER 2006

� Organic molecules with minimal conforma-tional flexibility for easy identification ofglobal minima.

� Molecular weight of the structure (moleculeand the counter-ion) being less than 200 tomake the system simple for a trial case study.

Based on this strategy, a search of the CCSDidentified17potential crystal salt structures21 fromwhich 3,4,6,7,8,9-hexahydro-2H-pyrimido (1,2-a)pyrimidinium acetate22 ([H2hpp][O2CCH3]), here-inafter referred to as hpp acetate, was selected.The final selection was based on maximal mole-cular rigidity and the possession of a stronghydrogen-bonding network.

Table 2 and Figure 2 show, respectively, thecrystallographic data for the salt structure to-gether with molecular diagrams for the host andcounter-ionic species.

COMPUTATIONAL DETAILS

Identification of Dimer Pairs via SystematicSearch Methodology

The selected salt system, hpp acetate, with[H2hpp] and acetate molecules in the asymmetric

unit was used to demonstrate the potential appli-cation of a grid-based systematic search method23

for salt screening (Fig. 3). Note that the effect ofproton transfer had not been explicitly consideredin the present study. However, clearly, for abinitio work the effect of proton transfer will needconsideration, particularly in respect of the hostspecies.

The dimer search programme23 was used toidentify potential molecular pairs by rotating andtranslating one molecule with the other being at afixed location. In this study, the molecular pair of[H2hpp] and acetate molecules from the organicsalt, hpp acetate, was used to search for the best-calculated molecular pair structures. The dimersearch procedure is briefly reviewed here with adetailed description of the method being providedelsewhere.23 The molecular pair ([H2hpp] andacetate) were treated as two rigid bodies with[H2hpp] (shown as number 1 in Fig. 4) being treat-ed in the search procedure as the fixed moleculeand acetate (shown as number 2 in Fig. 4) as themobile molecule.

Six parameters, three translational (X, Y, andZ) and three rotational (yx, yy, and yz) degrees offreedom, were used to describe positions on a grid(Fig. 4). Cartesian spherical polar coordinates

Table 2. Unit Cell Parameters of hpp Acetate Crystal

Structure a (A) b (A) c (A) a (8) b (8) g (8)SpaceGroup

Unit CellVolume (A3)

AsymmetricUnit per Cell (Z)

Hpp acetate 7.888 11.182 11.902 90.00 91.91 90.00 P21/n 1049.216 4

Figure 2. Diagrams showing the molecular structures of Hpp (a) and acetate (b)molecules.

2364 HAMMOND ET AL.


were used together with translationmagnitude, l,and a unit vector defined by (y, f) with Dy and Dfbeing the angular intervals defining the spacing ofgrid points within the search. The new atomiccoordinates of the mobile molecule after transla-

tion and rotation were calculated using thefollowing equation:

x0iy0iz0i

0@

1A ¼ M

xiyizi

0@

1Aþ lR� ð1Þ

where xi, yi, zi are the atomic coordinates of themobile molecule at its starting location; x0i; y

0i; z

0i

are the coordinates upon transformation; M is arotation matrix with yx, yy, yz; l is a translationmagnitude that is minimized with respect tothe intermolecular-pair potential energy; R� isthe position vector of the center of coordinatesof the mobile molecule. For each translationdirection defined by the mobile molecule, theminimum separation distance between the cen-ters of the two molecules was determined by vanderWaals radii. The separation distance was thenused as the starting point for the one-dimensionalminimization process against the intermolecular-pair potential energy.

The atom–atom force field parameters for thepotential energy calculation were taken from

Figure 3. Application of agrid-based systematic searchmethod for screening salt:Hppacetate.

Figure 4. The Cartesian coordinate system forsearching molecular pairs of hpp located at origin (1)and acetate mobile molecule (2) with the detaileddescription given elsewhere.23 For clarity of presenta-tion, molecules shown without some hydrogen atoms.



Sheraga force field24 with the atomic pointcharges of each of the atoms being computedusing the MOPAC25 method using the ModifiedNeglect of Diatomic Overlap (MNDO) methodfor molecular geometry optimization.26 Althoughthe semiempirical approach used here for derivinga set of atom centered charges proved to beadequate for the purpose of the present case study,the use of full ab initio quantum mechanicalmethods for the calculation of charges is clearlypreferable as part of the future developmentof this approach. The intermolecular potentialenergy was then calculated by using the followingequation:

E ¼XMi

i¼1

XMj

j¼1

�Aij

r6ijþ Bij

r12ij

!þ �Cij

r10ijþDij

r12ij

!þ gigjDrij

" #

ð2Þ

where Aij, Bij, Cij, and Dij are atom–atom forcefield parameters for atoms i and j in the first andsecond molecules; gi and gj are atomic pointcharges; D is the dielectric parameter, and rij isthe central distance between atoms i and j. In thisformulation12 the 6–12 potential termwas used tomodel the isotropic van der Waals component forall atoms except those involved in hydrogen bond-ing where the more appropriate 10–12 formalismwas employed. Note that, generally speaking, inthe case of host counter-ion linkage involvinghydrogen bonding, one would intuitively feel thatthe intradimer bond would always dominate.Although the focus here is on a structure predic-tion capability for host and counter-ion salts, theoverall shape of the dimer packing for molecular/counter ion is bound to influence the resulting saltstructure, postnucleation, which offers anotherroute for potential development with regards topolymorph prediction methods.14–17

Optimizing and Clustering of Molecular Pairs

The grid-based dimer search results in the gen-eration of a very large number of molecular pairstructures which, if analyzed individually usingthe crystal lattice systematic search, would re-quire considerable CPU time resource to fullyassess. Therefore, the molecular, ionic pairstructures identified were first optimized, andthen clustered into a smaller number ofputative pairs before being subjected to furtheranalysis.

In the optimization process, the molecular pairstructures generated within a specified energy

window (e.g., in this case having an energy lessthan �40 kcal/mol) were minimized using Cerius2

molecularmodeling package27 with the same forcefield parameters that were used for dimer search-ing. The optimized pairs were then clustered intodifferent groups using a test of their structuralsimilarity to reduce the number of pairs byeliminating any repeated or symmetry-relatedstructures. For the purpose of comparing differentdimers in a pairwise manner, the two H2hppmoieties were first superimposed and the sum ofthe squared distances between correspondingatoms of the acetate moieties were calculated.The root of the mean of the squared distances(rms distance) was then calculated. Any two pairstructureswitha rmsdistance of zero or less thanarms tolerance (e.g., 0.5) were considered as similarpairs and hence grouped within a single cluster.The rms tolerance plays an important role indetermining the number of clusters generated aswell as the average number of pair structures inone cluster. It is worth noting that the energywindow and rms deviation between configurationsare case specific parameters that can be optimized,for example, by performing a preliminary coarsesearch prior to amore intensive grid-based search.The representatives from each cluster, togetherwith the nonclustered molecular pairs, were thenranked by their optimized pair energy to form thefinal list of molecular pair structures. Subse-quently, the molecular pair with the lowestpotential energy (that resembles that provided bythe CCSD in terms of configuration and position)was analyzed and examined in the crystal latticebased systematic search procedure.

In this instance, the search for the optimalpacking arrangement of the dimer asymmetricunit was confined solely to the most energeticallyfavorable molecular pair configuration as theintended purpose of this preliminary work hasbeen directed towards proof of concept. Given thatthe most favorable intermolecular (salt pair)packing arrangement from the grid-based latticesearch was found to correspond to the knowncrystal structure, this purpose was thereforeserved. It is, however, fully recognized that forthe general case it would be clearly essential totreat a range of the molecular pair configurations,including preferred and less energetically stablepairs as identified via clustering in the full crystallattice systematic search method. Note that whenapplying this method to an unknown salt crystalstructure, the above approach would clearly beessential.

2366 HAMMOND ET AL.


Crystal Structure Packing and Optimization viaLattice Based Systematic Search

The crystal lattice based systematic search pro-cedure acts as a very fast and efficient filter forselecting potential trial structures. This methodtakes into account both in-built chemical senseand the crystal lattice energy which are keypoints in determining the suitability and rankingof the potential arrangement of the pairs. Thelattice energy is calculated by summing all theinteractions between a central molecule and allthe surrounding molecules. The intermolecularinteractions are considered to be the sum of theconstituent atom–atom interactions28,29 with thelattice energy, Elatt, then being calculated from:

Elatt ¼1

2

XNk¼1

Xni¼1

Xn0

j¼1

Vkij ð3Þ

where the factor ½ reflects the pairwise nature ofintermolecular forces; Vkij is the interactionpotential between atom i in the central moleculeand atom j in the surrounding molecule k ascalculated using Eq. 2. Note that n¼n0 for singlecomponent organic structures, whereas in thecase of co-crystal molecular complexes, such assalts and hydrates, they will be different. Basedon a typical lattice energy convergence for amolecular crystal, an intermolecular radial cut-off distance and short atom–atom distance cut-offs for typical van de Waals and hydrogen-bonding distance interactions were adopted30

(Fig. 2). Generally speaking, for molecular crys-tals (including salts) the electrostatic componentof the lattice energy represents a small part oftotal lattice energy, and hence a real space sum-mation is usually just as effective as using Ewaldmethod. All the trial crystal structures obtainedwith the subset of pair structures obtained fromthe dimer search were then subjected to latticeenergy minimization using the same force fieldparameters that were employed in the lattice-based systematic search. In this work, the topranked structures, before and after lattice energyminimization, were compared with the knowncrystal structure as obtained from the CCSD.

RESULTS AND DISCUSSION

Search of Molecular Pair Structures

For the dimer search the polar angle, y, step sizewas set at 108, whilst the corresponding polar

angle, f, step size was determined by using therelation Dy/(p siny) with 08< y< 1808, whichcreated grid cells on a spherical surface ensuringthat all the grid cells were generated with thesame size.23 The step sizes for the three rotationangles were set at 208 with the range of variationfrom 08 to 3608. For an energy cut off value of�40 kcal/mol, some 23300 molecular pair struc-tures were generated. The structures exhibitingthe highest and lowest potential energy valueswere �41.70 and �46.69 kcal/mol, respectively,associated with corresponding polar angles (y, f)of (1008, 2528) and (908, 2558). The energydistribution of these identified pair structures isshown in Figure 5.

As can be seen in Figure 5, about 4556 dimerpairs fall in the energy range from�41 to�42kcal/mol. Given the relatively coarse grid size adopted,it was found that the dimer pairs derived from thesearch process did not provide an exact match tothemolecular pair configuration obtained from theCCSD. Whilst a finer grid size would obviouslyprovide a closer match, this would substantiallyincrease the CPU time. For example, halvingthe step size for each of the search parameterswould increase the CPU time required to at least11.5 days, which is clearly not a realistic option fora proof of concept study. Therefore, potentialenergy minimization was used to optimize all thepair structures to generate more realistic bindingconfigurations.

Optimization and Clustering of the MolecularPair Structures

The objective of the optimization was to find thelowest local potential energy for each of the pair

Figure 5. Energy distribution of pair structuresbefore (solid line and grey bars) and after minimization(dashed line and striped bars).



structures by translating and rotating the mobile,acetate, molecule locally with much finer stepsto minimize the interaction energy.31 All 23300dimers were optimized and the distribution of theminimized potential energy (Fig. 5) was found todecrease with the number of pair structures atlower (more stable) energies increasing. As aresult, optimization of the molecular pairs suc-cessfully produced pair structures with the lowestpossible potential energy locally due to the relaxa-tion of constraints on grid cell size and orientationangle step.

As expected, more clusters were generated withsmaller values of rms tolerance (Tab. 3). Theenergy distribution at different rms tolerances isshown in Figure 6.

It is unrealistic to run the systematic search forcrystal structure packing with a large number ofclusters, that is, very small value of rms tolerance.Therefore, the rms tolerance was set at 0.5 for thisstudy, which produced a more sensible set of 72clusters which included 13 single pair structureclusters, that is, those which could not be assignedinto any of the clusters. The 72 clustered struc-

tures that were obtained from the dimer searchtogether with the pair structure taken from thecrystal structure given in the CCSD are shown inFigure 7. It can be seen that the locations of some ofthe acetate molecules are quite close to thatobtained from the CCSD crystal structure. Theclusters were ranked according to their minimizedintermolecular potential energy. In this study,the pairwith the closest relative position of acetatecounter-ion species as compared to that in theCCSD structure was found to be the top-rankedcluster with the optimized pair potential energybeing �46.79 kcal/mol. This structure was thenused as the basis of a full crystal lattice basedsystematic search with the remaining 71 clustersbeing considered as a reference.

Table 3. Number of Pair Structures Before and After Clustering According tothe rms Tolerance

Before Clustering

After Clustering (rms tolerance)

0.02 0.05 0.1 0.2 0.5

Number of pairs 23300 8925 3676 1510 488 72Average number of

pairs in one clusters1 �3 �6 �15 �477 �324

Figure 6. The effect of rms tolerance on the distribu-tion of [H2hpp] and acetate pairs after clustering.

Figure 7. Orientation of 72 clusters derived fromthe clustering process ([H2hpp] acetate dimer derivedfrom the CCSD, carbon coloured green; 1st, 2nd, 3rd,4th and 5th ranked dimers from search–carboncoloured magenta, yellow, orange, brown and purple,respectively).

2368 HAMMOND ET AL.


During clustering, the distribution of potentialenergy was found to change dramaticallycompared to that obtained before and after mini-mization (Fig. 8). Most of the pairs clustered werefound to have a potential energy in the range of�42 to �43 kcal/mol with the total number of pairstructures being found to decrease from 23300 to72. This indicates that the clustering process hassuccessfully produced a realistic number of pairswhich can be evaluated on a reasonable time scaleusing the lattice based systematic search.

Trial Structure Generation and Optimization usingthe Lattice Based Systematic Search

The lattice based systematic search method wasused to search the best packing structure of hppacetate salt treating the predicted dimer pairconfiguration with lowest energy. The unit cellparameters (Tab. 2) for hpp acetate as obtainedfrom the published crystal were used for thesystematic search.

For the trial structure generation calculations,the increment for each rotation angle was fixed at108with thenumberof rotational stepsabout thex,y, and z-axis being specified as 36, 36, and 36,respectively. For each rotation, 1000 translationalsteps were performed. Therefore a total of 46.656million steps were calculated in the search proce-dure. A lattice energy window was set up with anenergy cut-off value of �10 kcal/mol, whichproduced about 300 crystal structures. Analysisof this data revealed that the most energeticallystable structure, that is, that with the lowestcalculated lattice energy was very similar to thatprovided in the CCSD. Figures 9 and 10 show thecomparisons between the original crystal struc-ture from the CCSD and the best calculatedstructure from lattice based systematic searchfollowing optimization before and after opti-mization step, respectively, together with thecorresponding comparisons of their simulatedX-ray diffraction patterns based on these trialstructures.

Figure 8. The comparison of potential energy distri-butions after clustering, before and after minimisation(~, beforeminimisation;&, afterminimization; , afterclustering (rms¼ 0.50)).

Figure 9. Comparisons of (a) crystal structures between the best calculated structurefrom systematic search (carbon coloredMagenta) and hpp acetate crystal structure fromthe CCSD (carbon colored Green) and (b) the corresponding X-ray diffraction patterns(yellow, systematic search; red cross, CCSD; blue, difference).



Following this optimization, the best trialstructure was found to be in excellent agreementwith that taken from the CCSD. Note, prior tooptimization of the crystal packing, this specificstructure was ranked 39th (Fig. 9) from 277becoming the top ranked structure (Fig. 10) withan optimized lattice energy of �29.18 kcal/mol.These calculations indicate that with the smalldifference between the theoretically determinedstructure and the one from the CCSD, thesystematic search approach has successfully gen-erated trial structures which could be used withconfidence to determine structure of the hppacetate co-crystal using the Rietveld refinementmethod.32

Although preliminary, the results from thisstudy indicate that systematic-search basedmolecular modeling has the potential to be imple-mented in pharmaceutical formulation design asa tool to model and understand salt formingcharacteristics.

CONCLUSIONS

A systematic search method for dealing with morethan one molecule in an asymmetric unit18 hasbeen further developed and applied to validate theapplicability of a method for salt screeningthrough detailed studies on an organic salt:

3,4,6,7,8,9-hexahydro-2H-pyrimido (1,2-a) pyri-midinium acetate crystal. The overall modelingapproach adopted has included: pair structuresearching, optimization and clustering, andlattice-based searching. The study has success-fully identified a very small number of pairstructures representing all differentiated pairgroups from a large data set which has beensubsequently refined to elucidate the bulk crystalstructure in good agreement with the CCSD.

Further application of this complete methodol-ogy for structure determination of organic co-crystals such as salts and solvates is currentlybeing developed with the aim of providing a wayforward to understand the crystal science asso-ciated with (pseudo-) polymorphic and relatedeffects. In addition, further investigations arebeing carried out to screen more complex saltstructures including base-forming salts andcounter-ions from salt forming classes 2 and 3.Through these investigations the sensitivity of theapproach to parameters such as the energywindowand the criteria used for clustering similarstructures will be fully evaluated. If the approachcontinues to prove successful then future chal-lenges in this area will include studies of hostcompounds with more than one ionisable group aswell as systems with higher molecular weightsand associated degrees of molecular flexibilitywhich will hopefully address some issues such as

Figure 10. Comparisons of (a) crystal structures between the optimised bestcalculated one from systematic search (carbon colored Magenta) and hpp acetate crystalstructure from the CCSD (carbon colored Green) and (b) the corresponding X-raydiffraction patterns (yellow, systematic search; red cross, CCSD; blue, difference).

2370 HAMMOND ET AL.


scoring/ranking of trial structures and the effect ofinput structures on crystal structure prediction.

ACKNOWLEDGMENTS

The motivation for this study follows discussionwith Professor Robert Docherty (Pfizer Ltd.) towhom we are most grateful. We are also grate-ful for the financial support from UK EPSRC(GR/R14491 and GR/N/06670) which forms part ofcollaborations with Dr. Peter Halfpenny at theUniversity of Strathclyde and Professor RobinHarris at the University of Durham. This workforms part of the MSc programme of one of us(RSH).21We are also grateful to one of this paper’sreviewers who made a number of perceptivecomments particular concerning the issue relatedto ranking of trial structures that have beenincluded in the revised version presented here.

REFERENCES

1. Lipinski CA, Lombardo F, Dominy BW, Feeney PJ.1997. Experimental and computational approachesto estimate solubility and permeability in drugdiscovery and development settings. Adv Drug DelRev 23:3–25.

2. Venkatesh S, Lipper RA. 2000. Role of the devel-opment scientist in compound lead selection andoptimization. J Pharm Sci 89:145–154.

3. Bastin RJ, Bowker MJ, Slater BJ. 2000. Saltselection and optimisation procedures for pharma-ceutical new chemical entities. Org Process Res Dev4:427–435.

4. Gould PL. 1986. Salt selection for basic drugs. Int JPharm 33:201–217.

5. Morris KR, Fakes MG, Thakur AB, Newman AW,Singh AK, Venit JJ, Spagnuolo CJ, SerajuddinATM. 1994. An integrated approach to the selectionof optimal salt form for a new drug candidate. Int JPharm 105:209–217.

6. Datta S, Grant DJ. 2004. Crystal structures ofdrugs: Advances in determination, prediction andengineering. Nat Rev Drug Discov 3:42–57.

7. Morissette SL, Almarsson O, Peterson M, RemenarJ, Read M, Lemmo A, Ellis S, Cima MJ, GardnerCR. 2004. High-throughput crystallization: Poly-morphs, salts, co-crystals and solvates of pharma-ceutical solids. Adv Drug Deliv Rev 56:275–300.

8. Desrosiers PJ. 2004. The potential of preform. ModDrug Discov 1:40–43.

9. Storey RA, Docherty R, Higginson PD. 2003.Integration of high-throughput screening method-

ologies and manual processes for solid form selec-tion. Am Pharm Rev 6:100–105.

10. Remenar JF, MacPhee JM, Larson BK, Tyagi VA,Ho J, McIlroy DA, Hickey MB, Shaw PB, Almars-son O. 2003. Salt selection and simultaneouspolymorphism assessment via high-throughputcrystallization: The case of sertraline. Org ProcessRes Dev 7:990–996.

11. Ware EC, Lu DR. 2004. An automated approachto salt selection for new unique trazodone salts.Pharm Res 21:177–184.

12. Clydesdale G, Roberts KJ, Docherty R. 1997.Computational studies of the morphology of molec-ular crystals through solid-state intermolecularforce calculations using the atom-atom method.Theoretical aspects and computer modelling, editedby A. Gavezzotti.

13. Clydesdale G, Roberts KJ, Walker EDL. 1997. Thecrystal habit of molecular materials: A structuralperspective. Theoretical aspects and computermodelling, edited by A. Gavezzotti.

14. Lommerse JPM, Motherwell WDS, Ammon HL,Dunitz JD, Gavezzotti A, Hofmann DWM, LeusenFJJ, Mooij WTM, Price SL, Schweizer B, SchmidtMU, van Eijck BP, Verwer P, Williams DE. 2000. Atest of crystal structure prediction of small organicmolecules. Acta Cryst B56:697–714.

15. Giovannini J, PerrinMA, LouerD, Leveiller F. 2001.Ab initio crystal structure determination of threepharmaceutical compounds from X-ray powderdiffraction data. Mater Sci Forum 2:582–587.

16. Bond AD, Feeder N, Teat SJ, Jones W. 2000.The solid-state structure of 3-hydroxy-4-methyl-2(3H)-thiazolethione: Prediction and measure-ment. Tetrahedron 56:6617–6624.

17. Motherwell WDS, Ammon HL, Dunitz JD,Dzyabchenko A, Erk P, Gavezzotti A, HofmannDWM, Leusen FJJ, Lommerse JPM, Mooij WTM,Price SL, Scheraga H, Schweizer B, Schmidt MU,van Eijck BP, Verwer P, Williams DE. 2002.Crystal structure prediction of small molecules: Asecond blind test. Acta Cryst B58:647–661.

18. Cambridge Structure Database, version 5.25. 2003.http://www.ccdc.cam.ac.uk/products /csd

19. Pfaankuch F, Rettig H, Stahl PH. 2002. Biologicaleffects of the drug salt form. In: Stahl PH,WermuthCG, editors. Handbook of pharmaceutical salts.Zurich: Wiley-VCH.

20. Stahl PH. 2002. Handbook of pharmaceutical salts(Appendix). Zurich: Wiley-VCH.

21. Hashim RS. 2004. Application of molecular model-ling techniques in predicting the salt formingcharacteristics used in pharmaceutical formulationdesign. MSc Thesis, University of Leeds.

22. Cotton FA, Timmons DJ. 1998. New multiple-bonded dimetal compounds containing bridging1,3,4,6,7,8-hexahydro-2H-pyrimido [1,2-a] pyrimi-dino groups–I. The V2

4þ, Cr24þ and Mo2

4þ compounds



and some salts of the protonated ligand. Poly-hedron 17:179–184.

23. Hammond RB, Ma CY, Roberts KJ, Ghi PY,Harris RK. 2003. Application of systematic searchmethods to studies of the structures of urea withdihydroxy benzene co-crystals. J Phys Chem B107:11820–11826.

24. Nemethy G, Pottle MS, Scheraga HA. 1983. Energyparameters in polypeptides. Updating of geome-trical parameters, nonbonded interactions, andhydrogen bonds interactions for the naturallyoccuring amino acids. J Phys Chem 87:1883–1887.

25. MOPAC v. 6.0. 1993. QuantumChemistry ProgramExchange Program No. 455. Creative Arts Building181, Indiana University, Bloomington, IN 47405.

26. Dewar MJS, Thiel W. 1977. Ground State ofMolecule. The MNDO Method. Approximationsand parameters. J Am Chem Soc 99:4899–4907.

27. CERIUS2 v. 4.2. 2001. Molecular Modelling Soft-ware for Materials Research, Accelrys, Inc., SanDiego, CA, Cambridge, UK.

28. Hammond RB, Roberts KJ, Docherty R, EdmonsonM. 1997. Computationally assisted structure deter-

mination for molecular materials from x-raypowder diffraction data. J Phys Chem B 101:6532–6536.

29. Smith EDL, Hammond RB, Jones MJ, Roberts KJ,Mitchell JBO, Price SL, Harris RK, Apperley DC,Cherryman JC, Docherty R. 2001. The deter-mination of the crystal structure of AnhydrousTheophylline by X-ray powder diffraction with asystematic search algorithm, lattice energy calcu-lations, and 13C and 15N solid-state NMR: Aquestion of polymorphism in a given unit cell.J Phys Chem B 105:5818–5826.

30. Hammond RB, Roberts KJ, Docherty R, Edmond-son M, Gairns R. 1996. X-form metal free phthalo-cyanine: Crystal structure determination using acombination of high resolution X-ray diffractionand molecular modelling techniques. J Chem SocPerkin Trans 2 8:1527–1528.

31. Moore E. 2002. Molecular and bonding. MiltonKeynes: The Open University.

32. Rietveld HM. 1969. A profile refinement method fornuclear and magnetic structures. J Appl Crystallgr2:65–71.

2372 HAMMOND ET AL.


Date post:	13-Nov-2023
Category:	Documents
Upload:	independent
View:	0 times
Download:	0 times

Grid-based molecular modeling for pharmaceutical salt screening: Case example of...

Documents