Polymorph Selection: Challenges for the Future? - Crystal Growth

---

CRYSTAL GROWTH & DESIGN 2003 VOL. 3, NO. 6 873-885

Review Polymorph Selection: Challenges for the Future?† Nicholas Blagden* and Roger J. Davey Institute of Pharmaceutical Innovation, School of Pharmacy, University of Bradford, Bradford, BD7 1DP, UK, and Molecular Materials Centre, Department of Chemical Engineering, UMIST, M60 1QD, UK Received June 5, 2003

ABSTRACT: This contribution reviews the area of polymorph selection during solution crystallization using known and predicted structures. In particular, an overview of what has been learned about the relationship between structure and kinetics and of the interplay between thermodynamics and kinetics is presented. This review adds to the debate by discussing the future challenges that we envisage in realizing polymorph selection as a crystal engineering exercise by highlighting the role that a packing landscape approach may contribute in manipulating the selection of a polymorph. Introduction This contribution reviews work in the area of polymorph selection during crystallization from solution using known and predicted structures. In particular, a focus on what has been learned in the associated areas of structure and kinetics and the interplay between thermodynamics and kinetics with regards polymorph selection will be given. The future challenges that we envisage in realizing polymorph selection as a crystal engineering exercise are presented by developing the notion of a packing landscape using both known and predicted crystal packing. Polymorph Selection for Known Polymorphs Within a framework of structural and kinetic parameters, it appears that three major areas are important if we are to understand and control the crystallization of polymorphic systems from solution. These may be summarized as (1) An understanding of the structural similarities and differences between polymorphs. (2) An appreciation of the interplay between thermodynamic and kinetic factors. (3) A knowledge of the fundamental crystallization growth unit and its relationship to the structural synthon (crystal engineering term). Taking each of these themes in turn we review briefly some examples from our recent work. The reader is also † Based upon the presentation given at the ACS ProSpectives Polymorphism meeting, Tampa, FL, USA, 2003.

referred to the relevant cited papers for the specific experimental detail. Knowing the Similarities and Differences of the Molecular Packing between Polymorphs. An ability to differentiate between the molecular packing observed in polymorphs is essential to our understanding of the potential impact of solvents and impurities on the relative nucleation and growth kinetics of different forms. Following the work of Etter1 and Bernstein,2 we have found graph set analysis, based upon the hydrogen bond networks, a convenient way of achieving this. The inclusion of first and second level graph sets in RPLUTO3 has aided considerably the application of this methodology to our studies. For simplicity in this paper, the trends in hydrogen bond usage in a particular polymorph will be summarized using the key designators (D) and number atoms (minimum number to achieve the designator) involved (N), i.e., D(N), where D ) R for ring, I for intramolecular, and C for chain. For certain cases, such as the dimorphic system 2,6dihydroxybenzoic acid (DHB),4 visual inspection of the unit cells of the respective polymorph is sufficient to reveal the packing and molecular conformational differences between the two polymorphic forms. Thus, Figure 1a shows how polymorphism arises from the carboxylic acid syn conformation (form 1) or the carboxylic acid anti conformation (form 2), depending on the presence of one or two intramolecular hydrogen bonds. In form I-[R(8), I(6)] and form 2-[C(6), I(6)] are the critical packing motifs of the respective polymorphs. Gas phase conformational maps can be used to under-

10.1021/cg030025k CCC: $25.00 © 2003 American Chemical Society Published on Web 09/19/2003

874

Crystal Growth & Design, Vol. 3, No. 6, 2003

Figure 1. Polymorphism of 2,6-dihydroxybenzoic acid. (a) Form 1 is shown on LHS, and form 2 is shown on the RHS, and (b) the conformational map of the transformation from syn to trans conformations.

stand the low energy pathway between one conformer and the other. The situation for DHB is shown in Figure 1b. The generated map indicates a barrier of 12 kcal mol exists for the transition between the syn and anti conformations, with each conformation having similar internal energy. It is unusual to find the conformational minima in the solid state corresponding to the minima in the gas phase. Usually the points are displaced from

Review

each other, as the crystal field experienced by a molecule assembled into a crystal lattice exerts an influence on the conformation adopted by the molecule within the lattice. However, these searches can act as an indicator to the possibility of other conformational polymorphs. For other polymorphic systems, the situation is far more complex. Sulfathiazole,5 for example, is a pentamorphic system in which visual inspection of unit cells indicates that forms I and V are distinct while forms II, III, and IV are based upon similar packing motifs. Graph sets indicate that form I is based upon the dimeric growth unit, R(8) packed into chains, C(8), as shown in Figure 2a,b. Form V is based upon the tetrametric growth unit [R(4), R(10)] and a dimeric chain structure that uses the R(10) dimers, as shown in Figure 2c,d. Forms II, III, IV are based upon a dimeric growth unit, possessing a R(18) motif, shown in Figure 2e, assembled into sheets. The variation in assembly of the R(18) gives rise to the structural differences between forms II, III, and IV. Essentially, the usage of -SdO, -NH2, and thiazole >NH contacts differs in the three forms as highlighted in Figure 2f,g, respectively. Specifically, form II uses -SdO‚‚‚H2N- contacts (Figure 2f), form III uses both the -SdO‚‚‚H2N- contacts and >NH‚‚‚NH2 (Figure 2g), and form IV uses >NH‚‚‚NH2 contacts. The Interplay between Thermodynamic and Kinetic Factors. The interplay between kinetics and thermodynamics is essentially summarized in Ostwald’s Rule of Stages,6 which indicates that in a polymorphic system the crystallization processes may be complex starting with the appearance of the least stable form and finishing with the most stable. Thus, the primary nucleation stages are interspersed with polymorphic transformations that often involve a growth and dissolution process from a metastable phase to a more stable phase.7 The situation for a dimorphic system in which a solution-mediated transformation process is possible is shown schematically in Figure 3a. Initially, the nucleation and then growth of the metastable form occurs until its solubility is reached. The nucleation of the stable form occurs and the subsequent growth of stable form is driven by the dissolution of the metastable form, the overlap region of the profiles in Figure 3a. On complete dissolution of the metastable form, the growth of the stable form continues until the solubility of the stable form is reached. Such an understanding indicates that the nucleation or growth inhibition of one polymorph over that of another would lead to disruption of the kinetics of transformation to or from that polymorph to another and hence impact on the appearance and apparent stability of forms. The situation in which the growth and dissolution processes are manipulated is shown schematically in Figure 3b. An extended plateau region of the stable form profile and a delay of the subsequent transformation to the stable form typifies such a disruption. In certain cases, the onset of transformation may be only a half-hour, and in others the time scale may be years with the level of stabilization dependent on the system and the inhibitor used. Clearly, good experimental observation using microscopy, thermal methods, and spectroscopy and diffraction is essential in mapping out the phase behavior of such systems.

Review

Crystal Growth & Design, Vol. 3, No. 6, 2003 875

Figure 2. Polymorphism of sulfathiazole: (a) critical growth unit for form I. (b) Schematic of forms 1 packing, (c) critical growth unit in form V, (d) schematic of form V packing, (e) critical growth unit in forms II, III, and IV, (f) schematic of packing in present in forms II and III, and (g) schematic of packing in forms III and IV. Black ) carbon, grey ) hydrogen, red ) oxygen, blue ) nitrogen, and yellow ) sulphur. L-Glutamic acid and DHB are dimorphic systems that both follow Ostwald’s Rule of Stages as shown in Figure 3c-e. For glutamic acid, the transformation form R to β polymorphs at 35 °C takes approximately 200 min, and both a time sequence of micrographs (see Figure 3c) and time sequence of X-ray diffraction patterns (see Figure 3d) have been taken of this transformation. The micrographs clearly show the solution-mediated transformation of the metastable plates to the most stable needle form in solution. This transformation process is confirmed by the in-situ dispersive diffraction data

obtained from examining a cooling solution of glutamic acid on Station 16.4 at SRS Daresbury Laboratory, UK.8 For DHB, the same initial supersaturation was used in two different solvents, toluene and chloroform. By using both a desupersturation profile and optical micrographs in each solvent, the progress of the transformation between the two forms was monitored. The composite profiles for the transformation using these combined data sets are given in Figure 3e for toluene and Figure 3f for chloroform.9 It is clear from this type of monitoring that the transformation in both solvents

876

Crystal Growth & Design, Vol. 3, No. 6, 2003

Review

Figure 3. The kinetics of dimorphic phase transformations: (a) as an uninhibited process, (b) as an inhibited process, (c) micrograph sequence of transformation of glutamic acid, (d) X-ray diffraction sequence of transformation of glutamic acid, (e) the ratio of transformation derived from the diffraction data in panel d, (f) micrograph sequence overlaid over supersaturation for DHB in toluene (5 mg/mL, 25 °C), and (g) micrograph sequence overlaid over supersaturation for DHB in chloroform (4.5 mg/mL, 25 °C) .

is from the metastable rhombus, form 1, to the stable needle, form 2. The difference in transformation time, 50 min in toluene and 5 min in chloroform, was associated with the influence of solvent on the nucleation kinetics of the stable form. The Ability to Tune the Fundamental Crystallization Unit. The ability to direct the nucleation event would represent a kind of “holy grail” in polymorph

crystallization. Previous studies10 have adopted the thesis that a supersaturated phase is inhabited by clusters of molecules having the packing of all the polymorphs and that the concepts of habit modification, templates, and supramolecular chemistry can be used to encourage the growth of a desired synthon. In such an approach, a structural based lock and key strategy is used to rationalize selectivity. Work on amino acids

Review

Figure 4. Lock and key scheme for polymorph selection: (a) routes to additive inhibition of crystal face, and (b) routes to polymorph stabilization.

and nonlinear optical compounds, for example, indicate the utility of this methodology as a route to polymorph selection.11 We have extended these ideas to the use of conformational selection as shown in Figure 4a, growth unit selection as shown in Figure 4b, and solution chemistry to attempt to encourage the formation of certain hydrogen bonded assemblies. To date, we have shown that additives, impurities, and solvents may all be used to direct the appearance of specific polymorphic forms for glutamic acid,12 sulfathiazole,9 and 2,6-dihydroxybenzoic acid.13 In the case of growth unit selection, the work on glutamic acid highlights how the growth units may be manipulated so that the metastable form is stabilized. For the L-glutamic acid system, trimesic acid was employed as a conformational mimic of glutamic acid in the stable β form; the overlay of additive and the conformation of glutamic acid in β form is given in Figure 5a. No such overlay exists for the additive and the conformation of glutamic acid in the R form. This conformational selectivity for β leads to the disruption of the growth along the principal growth axis of the stable β form, as shown in Figure 5b. The growth of the R form was unaffected by the additive, since no match between the additive in the fastest growing face of R form exists by virtue of the conformation. For the scenario of solution-mediated assembly and growth unit selection, the work undertaken on 2,6-DHB highlights this approach. In this study, the tentative link was made between molecular self-assembly in solution directed by the solvent. This was undertaken by analyzing the UV spectrum as concentration was varied for a selected solvent. Molecular aggregation was determined by analyzing the hypochromic effect, which is the variation in adsorption coefficient at λmax (the n-π* transitions at 325 nm for toluene and 320 nm for chloroform, as the DHB concentration was varied. (In Figure 6a,b, the following concentrations were used: toluene a ) 0.0032, b ) 0.0026, c ) 0.0016, d ) 0.008, e ) 0.004, f ) 0.003, g ) 0.002, and h ) 0.0001 molar; chloroform, a ) 0.0011, b ) 0.0084, c ) 0.0056, d )

Crystal Growth & Design, Vol. 3, No. 6, 2003 877

Figure 5. Additive stabilization of the metastable form of L-glutamic acid: (a) an overlay of additive with glutamic acid conformation in stable beta polymorph, and (b) the inclusion of the additive into the fastest growing face of the stable beta polymorph. Black ) carbon, grey ) hydrogen, red ) oxygen, blue ) nitrogen.

0.0028, e ) 0.0014, f ) 0.0007, and g ) 0.00035 molar.) The experimental outcome for toluene is shown in Figure 6a, and that for chloroform is shown in Figure 6b, with the respective figure showing the spectral series obtained and the variation in adsorption coefficient as concentration is varied illustrated in the respective insets. Such a nonlinear adsorption coefficient response is indicative of a differing mode of aggregation for the solution species present, and the mode of molecular assembly may be clarified and deduced from these data using nonlinear Beer’s Law analysis of the absorption coefficient profiles.14 The details of the regression analysis procedure used are beyond the scope of this review, and the reader is directed to relevant literature.7,13 Briefly, within this analysis a single-exponential profiles for adsorption coefficient with K2 ) 1.9 × 104 L mol-1 indicates a dimeric process and multiple profiles with K2 ) 1.6 × 103 L mol-1 and Kn 11.32 L mol-1 indicates a polymeric process, preceded by dimeric process. The relevant adsorption coefficient profile for each solvent is included as an inset with the relevant spectra as concentration was varied (see Figure 6a,b). This type of analysis suggests that toluene favors syn pairing, referred to as a dimer in Figure 6c, while chloroform favors an anti pairing referred to as a chain in Figure 6c. The role of solvent in this situation was complemented by undertaking both AMSOL15 and GRID16 calculations, to gauge the stability of the two modes of self-assembly, using isolated molecules and associated pairs taken from the respective crystal structures. For the GRID calculations Ar-H and R-Cl probes were employed for toluene and chloroform, respectively. These solvent probes were used to map an energy hypersurfaces at -3.5 kcal (viz energy of weak hydrogen bond) and the surfaces generated are given in Figure 6d. From such calculations, the increased binding of chloroform over toluene is evident from visual inspection with the toluene integrating more extensively

878

Crystal Growth & Design, Vol. 3, No. 6, 2003

Review

Figure 6. Solution chemistry of 2,6-dihydroxybenzoic acid: (a) hypochromic effect of DHB in toluene, (b) hypochromic effect of DHB in chloroform, (c) modes of assembly of syn and anti conformations, (d) GRID calculation hyposurfaces of DHB interaction with toluene and chloroform. Blue area in panel c represents solution spheres of molecule.

with the syn conformer, which generates dimers. A binding energy for the adduct in a solvent (∆Hsolv), using the difference between heats of formation in the gas and solution phases (see eq 1) was calculated using the heats of formation obtained using the AMSOL14 method contained in MOPAC7.17 gas ∆Hsolv ) (∆Hadduct - 2∆Hgas monomer) solvent - 2∆Hsolvent (∆Hadduct monomer) (1)

where heat of formation quantities for a monomer in solvent the gas phase, ∆Hgas monomer, in a solvent, ∆Hmonomer, the corresponding heats of formation for an adduct in the gas solvent gas phase ∆Hadduct , and in a solvent Hadduct were calculated using monomers and pairings from crystal

Table 1. Binding Energy Components in kcal/mol DHB solution species syn adduct, dimer anti adduct, chain

∆Hsolv (kcal mol-1) toluene chloroform -7.4 r -6.1

-7.8 -8.4 r

structures. It was found that toluene enhanced the stability of the dimer (row 2 in Table 1) while chloroform stabilized the chain (row 3 in Table 1), the relevant values are marked with an arrow. These values confirm the picture from the GRID calculations and solubility data that chloroform solvates DHB to a greater extent than toluene, and dimer formation in toluene occurs as the polar groups of the 2,6-DHB in syn conformation are less accessible to the nonpolar solvent. In contrast to chloroform, a polar

Review

Crystal Growth & Design, Vol. 3, No. 6, 2003 879 Scheme 1.

Polymorph Selection Strategy

hydrogen bond accepting solvent, the hydrogen bond donating groups in the polar regions of molecule in anti conformation are accessible. Polymorph Selection Strategy and Crystal Structure Prediction These findings led us to the important question of how we might apply such understanding of structure, thermodynamics, kinetics, and growth unit control within a polymorph isolation strategy, for a molecule of unknown crystal structures. Our starting point in this situation could not be a structural analysis and consequently we have explored the application of one of the major challenges of crystal engineering, namely, crystal structure prediction. For details of the crystal packing problem, and structure prediction methodology the authors refer the reader to a recent review by Price and co-workers.18 Important unresolved issues exist with these calculations in terms of the suitability of force fields to derive lattice energies, the algorithms used to generate trial structures, and whether the inclusion of conformation energy and entropy should be undertaken. It is also important to recognize that the objective of such solid-state calculations is to generate a packing landscape with the maximum diversity from a purely solid-state perspective. From a crystallization perspective, such computational approaches to structure pre-

diction would require the incorporation of kinetic factors arising from solvation, molecular clustering, and the process of nucleation itself. These types of issues were identified during the course of our experimental work on developing a polymorph isolation strategy and recently from a theoretical viewpoint Gavezzotti highlighted this issue in detail.19 The overall resulting situation is that structural predictions yield many tens of potential structures with no account taken of the potential for kinetic-dynamic factors to dilute the packing landscape. As a consequence, the experimentalist is confronted with a daunting array of structures with no real idea of which might be the most likely to appear. To make such data experimentally accessible and useful we have then applied our structural analysis and growth unit selection procedures, the “packing landscape concept”, across the whole output from such predictions from which a crystal engineering approach was undertaken to direct experimental selection. In this way, it becomes possible to make the best use of current trends in theoretical and computational approaches by combining them with experimental experiences, with the aim of manipulation of the packing landscape. Packing Landscape Manipulation A possible route and a first step to being able to target specific assemblies from simulated packing landscapes

880

Crystal Growth & Design, Vol. 3, No. 6, 2003

Review

Figure 7. (a-h) Selected packing landscape of 2-amino-3-nitro phenol from the Cerius2 Polymorph Prediction Module.

was to use the approach of combining current practice in crystal structure simulation with experimental strategies to achieve polymorph selection. The isolation strategy employed is given in Scheme 1, and to date 2-amino-3-nitro phenol20 (ANP) and diflusinal21 have been examined in this way. The approach taken is to use a packing landscape approach to describe a set of critical building units for a family periodic structures, consequently ranking using energy or density are necessitated. The following procedures were adopted in this approach, and the reader should refer to Scheme 1 to see how the processes have been linked: (i) The similarities and differences in the molecular packing of simulated polymorph structures were examined using graph sets in the same manner as for experimental ones. From this, an interpretation of hydrogen bond usage over the packing landscape is undertaken. (ii) Solvent selection strategies for the stabilization of particular target motifs were identified and crystallization experiments performed. Please note crystallization experiments were undertaken in a limited set of solvents chosen on the basis of the intermolecular interaction required for a particular crystal packing. Crystals were generally grown at high, medium, and low supersaturation. High supersaturation was defined as the maximum amount material dissolved in a chosen solvent at 5 °C below its boiling

point, with crystallization achieved by cooling the solution to 20 °C and collecting crystals after 24 h. Medium supersaturation was taken as half this amount and low was taken as a quarter of this amount. The singlecrystal structure obtained from a selected crystal was used to simulate a powder pattern, which was then compared with the powder pattern obtained form the bulk sample obtained from a given solvent. The reader is also referred to the relevant cited paper for the specific experimental detail. Critical Building Unit Motifs. From the simulation studies, the critical building motifs were identified for ANP, shown in Figure 7, and for diflusinal, shown in Figure 8. For ANP, a variety of chain and ring structures were identified. The most frequently observed and typical chain and dimer-based structures are shown in Figure 7a-f and two ring-type structures are shown Figure 7g-h. All the structures make use of the possible hydrogen bond interactions between the functional groups contained on the ANP molecule (-NH2, -OH, and -NO2). A systematic analysis of molecular packing of the simulated structure families indicated that the packing landscape utilized building units for the structures composed of pairs of molecules based upon the hydrogen bonding between two specific functional groups to form open chain motifs, either in single C(8) motif (see Figure 7a,d) or a pair of chain motifs such as [C(8), C(7)] shown in Figure 7f,g. Examples of [C(8), C(6)] are

Review

Crystal Growth & Design, Vol. 3, No. 6, 2003 881

Figure 8. (a-e) Selected packing landscape of diflusinal from the Cerius2 Polymorph Prediction Module.

given Figures 7e,h, and those involving [C(7), C(6)] are shown Figure 7b,c. Any residual hydrogen bond contacts are used to interconnect these chains into open rings or open dimeric sheets based on a R(10), (see Figure 7f) or based upon an extended ring using a R(14) motif (as shown in Figure 7a-c), and it was noted extensive open ring based upon R(28) assemblies were possible (see Figure 7g-h). The typical assemblies of diflusinal generated by the simulation are given in Figure 8. A number of packing motif trends was observed within the packing landscape generated. For diflusinal, it must be recognized that the strongest hydrogen bond interactions are between a hydroxyl group and carboxylic carbonyl to form an I(6) intermolecular hydrogen ring (consequently, this hydrogen bond was included in the initial conformation of diflusinal for the packing simulations), and the hydrogen bonding between two carboxylic acids, generating the R(8) dimer motif (see Figure 8a,c,d). The residual hydrogen bond sites are then used to interconnect R(8) dimers to form open chain networks. These interconnections involve a C(11) motif (see Figure 8c), a C(6) motif (see Figures 8d), or a R(4) dimer contact if the hydroxy groups are involved (see Figure 8b). If the dimeric contacts are absent then a open chain packing based upon either a C(8) motif (see Figure 8e) or C(11) motif are utilized (see Figure 8b). The difference in packing of structures using these classifications of contacts arises when this interconnection is formed using residual hydrogen bond contacts on the carboxyl end of the molecule. Interactions between the fluorine

groups, and between fluorine and aromatic hydrogens were suggested by packing outcomes from the simulation; however, these were not formally included in the graph set analysis. The Solvent Lock Key Strategies and the Experimental Outcome. For ANP, the basis of the solvent strategy was to use a solvent to promote a specific pairing, by targeting the remaining groups with a specific solvent interaction; this is illustrated in Figure 9. In this approach, the aim was to envisage a solvent as a template for a specific motif. In such a mode, to stabilize a chain motif, which leaves amino groups exposed, the use of nitro solvents was envisaged as a template (see Figure 9a). For chain motifs that leave nitro groups exposed, a solvent containing amino groups was envisaged as the template (see Figure 9b). Finally, for chains that leave hydroxyl groups exposed, hydroxyl-containing solvents were envisaged as template (see Figure 9c). For the open ring motifs, aromatic solvents (e.g., toluene) were envisaged as template (see Figure 9d,e). The outcome from a crystallization using a nitromethane was a crystal containing a [C(7), R(14), R(16)] motif system (see Figure 10a). This type of packing arrangement contains the desired chain templates shown in Figure 9a as a C(7) motif, but also contained a complex dimeric chain as a [R(14), C(7)] motif; this is shown in Figure 10b. The crystallization from toluene resulted in a crystal containing the desired [C(7), C(8), R(14), R(28)] motif (see Figure 10c). The detail of the extended ring motif for the crystal grown form toluene

882

Crystal Growth & Design, Vol. 3, No. 6, 2003

Review

Figure 9. Envisaged solvent templates for 2-amino-3-nitro phenol: (a-c) chain template routes and (d-e) ring template routes.

is shown in Figure 10d. Crystallization from methanol yielded a solvate crystal, containing a [C(7), C(8), R(14)] motif (see Figure 10e); this packing exhibits the desired chain template shown in Figure 9c, as a [C(7), C(8)] chain within a helix arrangement generated from interlocking these chains with [R(14)] dimers; this is shown in Figure 10f, and this outcome is structural isomorphous with the simulation given in Figure 7f. To date, crystals from ethanol have been highly disordered and a suitable a single-crystal structure has yet to be obtained. For diflusinal, a similar approach was taken to template-specific motifs based upon solvent choice. Growth from acetic acid was undertaken in an attempt to disrupt the formation of the carboxylic dimer and thus push the system to an assembly that does not employ the R(8) ring, but could include the [C(11), I(6)] motif (see Figure 8b) or the [C(8), I(6)] motif (see Figure 8e). Such motifs may involve packing utilizing hydroxylcarbonyl chains. Growth from toluene was envisaged to maximize all uses of hydrogen bond donors and acceptors, thus promoting motifs [R(8), C(11), I(6)] as shown in Figure 8c or the [R(8), R(4), C(6), I(6)] motif shown Figure 8d. Ethanol and acetone were employed to disrupt both hydroxyl and fluorine hydrogen-bonding contacts, to promote the [R(8), I(6)] motif (see Figure 8a) and if only hydroxyl hydrogen bonding is disrupted only a [R(8), C(11), I(6)] is promoted (see Figure 8c). Chloroform was chosen to disrupt fluorine and contacts

and promote motifs involving the [R(8), I(6)] motif as shown in Figure 8a, the [R(8), R(4), C(6), I(6)] motif shown Figure 8d, and the [C(8), I(6)] motif of Figure 8e. The resulting crystal structures from this proposed solvent selection indicated that acetic acid was unable to disrupt the carboxyl dimer formation. A projection of the crystal structure obtained from acetic acid is shown in Figure 11a, and possesses similarities to simulated packing shown in Figure 8c by utilizing the [R(8), C(11), I(6)] motif. The packing of crystal grown chloroform are given in Figure 11b, and utilize the [R(8), I(6)] motif of Figure 8a. The known form grown from acetone is shown in Figure 11c and also utilizes the [R(8), I(6)] motif of Figure 8a. From toluene, the crystal structure is shown in Figure 11d, which is identical to envisaged packing shown in Figure 8d, based upon a [R(8), R(4), C(6), I(6)] motif. Finally, for ethanol the packing of the crystal obtained is given in Figure 11e, which also utilizes the [R(8), I(6)] motif of Figure 8a. Discussion and Conclusions Polymorph selection attempts to combine aspects of structure, thermodynamics, kinetics, and molecular assembly to develop a strategy for the exploration and control of polymorphic crystal systems. Up to now, this

Review

Crystal Growth & Design, Vol. 3, No. 6, 2003 883

Figure 10. Projections of the experimental structure obtained for 2-amino-3-nitro phenol: (a,b) nitromethane, (c,d) toluene, and (e,f) methanol.

approach has met with a mixed level of success. Starting with systems with well-documented crystal structures and phase behavior, the results show that a high level of manipulation and control is possible; however, as this work has moved into the area of molecules for which with little or no previous data existed the outcome is less clear. This required the examination of how structure prediction might be included into the polymorph selection process developed during the course of our experimental work using known crystal structures. These studies in particular were undertaken to explore the concept of directing a specific assembly, and to gain an insight into the possibilities and limitations of this type of approach. A certain of level of control has been achieved to the extent that patterns of assembly in the crystal packing identified during the simulation and selection process were observed in the experimental crystal structures obtained with crystals grown in selected solvents. The

failings encountered indicate that some additional critical steps are required when undertaking crystal prediction. These would include principle growth unit identification and subsequent use of mixed growth units, the recognition of the role solvent makes to viable growth units, and the role kinetics has on understanding the Aufbau principle22 of how molecules pack into crystals. Key challenges remain: we need to better understand how to engineer selection by eliminating areas of packing space by improving the identification of key solute-solvent parameters, which influence the population of building units and thus the kinetics of the resulting nucleation and crystal growth process. This in turn would hopefully further reduce the landscape of possible polymorphs or alter the ranking of polymorphs. In this way, we may be able to selectively engineer a structure. The current understanding we have may make it possible to reduce the predicted packing landscape, as

884

Crystal Growth & Design, Vol. 3, No. 6, 2003

Review

Figure 11. Projection of the experimental structure obtained for diflusinal (a) acetic acid, (b) chloroform, (c) acetone, (d) toluene, and (e) ethanol.

simulations undertaken may only be concerned with what is identified as a viable molecular aggregate in selected solvent. This may be achieved by going through a number of cycles of the strategy we have outlined, initially undertake a lock and key approach to generate schemes of pairwise assemblies, followed by lock key schemes of solvent selection, which are then followed up by subsequent calculations on pairing of molecules in selected solvents. These pairings could then be used to undertake packing simulations either by using the conformation of one molecule in pairing, or the use the pair of molecules. We would also anticipate that the concepts reviewed in this paper would also find use in polymorph screening initiatives, particularly when high throughput identification of polymorphs utilizes simulated structures in ways described in this paper to design the overall experimental protocol employed. More Questions Than Answers. This of course leaves us with more intriguing questions to be answered, with even bigger challenges for researchers in the field. One present challenge in this area is both a computational one and an experimental one, and relates to what simulation or real-time monitoring of the crystallization process can be undertaken from prenucleation, nucleation, and crystal growth. These issues were raised by Dunitz,23 when considering “Are crystal structures predictable?”, with the opening line response, “The one word answer to the title question is still ‘No’, although at certain levels of discussion a ‘Maybe’ or even a conditional ‘Yes’ may be entertained as possible responses”. Within the work, we have discussed in this

paper relating to “engineering polymorph selection from solution”, and the current outcomes add further to the questions relating to the contribution and role of nucleation. This was also highlighted recently by analyzing the current understanding we have of nucleation control of isolating polymorphs24 and the outcome of kinetics on the future of crystal prediction.19 It seems that we always return to this question when trying to rationalize the data generated during the course of such studies. With this in mind, we can envisage that a considerable amount of future effort will be put into problem of understanding nucleation, clustering, and prenucleation events in solution to identify the links between the solution chemistry and crystallography. References (1) Etter, M. C. Acc Chem. Res. 1990, 23, 120-126. (2) Bernstein, J.; Etter, M. C.; MacDonald, J. C. J. Chem. Soc., Perkin Trans. 1990, 2, 695-698. (3) Martinez-Oharriz, M. C.; Martin, C.; Goni, M. M.; Rodriguez Espinosa, C.; Tros De Ilarduya-Aapaolaza, M. C.; Sanchez, M. J. J. Pharm. Sci. 1994, 83, 174-177. (4) Blagden, N.; Davey, R. J.; Alison, H.; Fuller, S. Cryst. Growth Des. 2001, 1, 59-65. (5) Blagden, N.; Davey, R. J.; Lieberman, H. F.; Williams, L.; Payne, R.; Roberts, R.; Rowe, R.; Docherty, R. J. Chem. Soc. Faraday Trans. 1998, 94, 1035-1044. (6) Ostwald, W. Z. Phys. Chem. 1897, 22, 289-330 (7) Cardew, P. T.; Davey, R. J. Faraday Discuss. 1993, 95, 160162.

Review (8) Blagden, N.; Davey, R. J.; Song, M. JCGD 2003, 3, 197202. (9) Davey, R. J.; Blagden, N.; Righini, S.; Alison, H.; Ferrari, E. S. J. Phys. Chem. B 2002, 106, 1954-1959. (10) Weissbuch, I.; Popoviz-Biro, R.; Leiswerowitz, L.; Lahav, M. In Lock Key Principle; Behr, J. P., Ed.; Wiley: New York, 1994; pp 34-35. (11) Weissbuch, I.; Leiserowitz, L.; Lahav, M. Adv. Mater. 1990, 2, 40-43. (12) Davey, R. J.; Blagden, N.; Potts, G. D.; Docherty, R. J. Am. Chem. Soc, 1997, 119, 1767-1772. (13) Davey, R. J.; Blagden, N.; Righini, S.; Alison, H.; Ferrari, E. S. J. Phys. Chem. B 2002, 106, 1954-1959. (14) Morcillo, J.; Galego, E.; Peral, F. J. Mol Struct. 1987, 157, 353-369. (15) Giessen, D. J.; Hawkins, G. D.; Liotard, D. A.; Cramer, C. J.; Truhlar, D. G. Theor. Chem. Acc. 1997, 98, 85-109. (16) Goodford, P. J. J. Med Chem. 1985, 28, 849-857.

Crystal Growth & Design, Vol. 3, No. 6, 2003 885 (17) Stewart, J. J. P. J. Comput.-Aided Mol. Des. 1990, 98, 85109. (18) Beyer, T.; Lewis, T.; Price S. L. CrystEngCommun 2001, 44, 1-35. (19) Gavezzotti, A. CrystEngCommun 2002, 4, 343-347. (20) Blagden, N.; Cross, W. L.; Davey, R. J.; Broderick, M.; Pritchard, G.; Roberts, R. J.; Rowe, R. C. Phys. Chem. Chem. Phys. 2001, 3, 3819-3825. (21) Cross, W.; Davey, R. J.; Blagden, N.; Pritchard, R. G.; Neuman, M. A.; Roberts, R. J.; Rowe, R. C. JCGD 2003, 3, 151-158. (22) Perlstein, J. J. Am. Chem. Soc. 1994, 116, 11420-11432. (23) Dunitz, J. Chem. Commun. 2003, 5, 545-548. (24) Davey, R. J.; Allen, K.; Blagden, N.; Cross, W. I.; Lieberman, H. F.; Quayle, M. J.; Righni, S.; Seton, L.; Tiddy G. J. T. CrystEngCommun 2002, 4, 257-264.

CG030025K