On the predictability of supramolecular interactions in molecular cocrystals – the view from the bench †

A series of theophylline cocrystals involving fluorobenzoic acids was prepared and structurally characterised. The cocrystals display compositions and hydrogen-bond patterns that could not be predicted based on extensive literature/database surveys and the use of other tools. The study demonstrates that – without the use of first-principles crystal structure prediction methods – it is still remarkably difficult to predict and understand the outcomes of cocrystallisation attempts involving small and rigid molecules.

difficult to empirically predict the formation of supramolecular synthons in cocrystals composed of molecules containing a broad range of functional groups. 4,8,27 In order to facilitate the design of cocrystals containing such complex entities, it will be necessary to considerably deepen our understanding of self-assembly processes in the solid state.
The current contribution suggests that such understanding can only be reached through well-planned crystallographic studies, and by embracing modern computational solid-state methods (such as crystal structure prediction 28,29 or CSP) as standard tools in crystal engineering. We present here a case study involving cocrystals of theophylline (thp) and fluorobenzoic acids (FBAs) (Scheme 1), i.e. cocrystals based on small and rigid molecules with multiple functional groups capable of hydrogen bonding. The study demonstrates the difficulty of empirically predicting compositions and synthon hierarchies in such cocrystals, and is complemented by information from crystallographic database surveys, as well as HBPC and MEPS calculations.
Thp is commonly used as a model compound in studies of pharmaceutical cocrystals, 8,[30][31][32][33] and as such has been primarily cocrystallised with carboxylic acids. 34 It is well known that thp engages in cocrystallisation with carboxylic acids through the formation of either an O-HĲcarboxyl)⋯NĲimidazole) hydrogen bond (marked as synthon A in Scheme 1) or a cyclic array of N-HĲimidazole)⋯OC(carboxyl) and O-HĲcarboxyl)⋯OC(amide) hydrogen bonds (marked as synthon B in Scheme 1). Our interest in (thp)·(FBA) cocrystals was sparked by our recent investigation of the (thp)·(BA) cocrystal (where BA = benzoic acid, Scheme 1). The latter is a solid reported in numerous prior studies, 35,36 but one for which no crystal structure had been reported in the literature at the time we commenced our studies. Initial Cambridge Structural Database 37 (CSD) surveys and HBPC suggested that interactions of type A and B are almost equally likely to occur in (thp)·(BA) (see ESI), while a subsequent single crystal X-ray diffraction study revealed that thp and BA interact through a type B synthon (Fig. 1a). 38 Another study of FBA cocrystals performed in our laboratories 39 then inspired us to investigate whether minimal structural changes in BA could potentially reorganise the selfassembly process to yield thp cocrystals based on type A synthons. The choice of FBAs as cocrystal formers appeared appropriate, as all mono, di-, tri-, tetra-and penta-substituted FBAs are similar in size and shape to BA due to the relatively small size difference between hydrogen and fluorine atoms (van der Waals radii: 120 pm vs. 147 pm, respectively). 40 Moreover, organic fluorine is not likely to act as a hydrogenbond acceptor, 41,42 while short intermolecular C-H⋯F contacts in molecular crystals only make a minor contribution to the cohesive energy of a molecular crystal. 43 It was therefore postulated that the presence of fluorine atoms in the (thp)·(FBA) cocrystals would only minimally interfere with the formation of O-H⋯N, N-H⋯O and O-H⋯O hydrogen bonds between the cocrystal components. This hypothesis was also in agreement with results of a recently published cocrystal study involving FBAs. 44 The cocrystallisation of thp with all nineteen FBAs was attempted by liquid-assisted grinding 45 (LAG) using equimolar amounts of thp and the relevant FBA (see ESI †). The formation of the cocrystals was recognised via powder X-ray diffraction (PXRD) in fifteen out of nineteen casesonly cocrystal screens involving 4FBA, 23diFBA, 24diFBA and 26diFBA did not result in cocrystal formation under the investigated crystallisation conditions. 46 The crystal structures of all produced solids were determined either directly from the LAG samples using laboratory PXRD data (by simulated annealing, 47 Rietveld refinement 48 and CASTEP 49 DFT geometry optimisation), or by single crystal X-ray diffraction (SCXRD) using crystals that were obtained through solutionmediated phase transformation 50 (SMPT) or by slow solvent evaporation (see ESI †).
A cocrystal of an unexpected composition was formed in a LAG experiment involving equimolar amounts of thp and 25diFBA where PXRD and SCXRD revealed the formation of a 1 : 2 cocrystal, (thp)·(25diFBA) 2 . The structural study also showed that thp and 25diFBA formed discrete three-component assemblies, which were sustained by synthons of both type A and B (Fig. 2a). Several attempts to produce a 1 : 1 thp : 25diFBA cocrystal either mechanochemically or through solution-based crystallisation methods have, so far, failed.
Our investigations also led to the discovery of a 1 : 1 cocrystal comprised of thp and 35diFBA, (thp)·(35diFBA). This material surprisingly exhibits discrete four-component assemblies featuring a O-HĲcarboxyl)⋯OC(urea) synthon (marked as type C in Scheme 1 and shown in Fig. 2b) that are statistically less likely to occur than synthons A and B, as determined by HBPC (see ESI †). More predictable synthons were, however, found in the 1 : 1 (thp)·(34diFBA) cocrystal wherein thp and the FBA interact via type A synthon (Fig. 2c).
The (thp)·(FBA) cocrystals described above clearly demonstrate the unpredictability of synthon hierarchies in twocomponent cocrystals with several hydrogen-bonding functional groups. The observed supramolecular interactions could neither be predicted empirically nor with the use of HBPC and MEPS calculations. In particular, HBPC has been performed on all structurally characterised solids using Mercury 3.7 and the CSD 5.37 (see ESI †). Although the calculations encouragingly predicted the formation of the observed hydrogen bonds between thp and the FBA in each case, the synthon hierarchies could not be foreseen (see ESI †). In addition, DFT calculations of MEPS were also performed to predict optimal hydrogen-bond donor-acceptor pairs (see ESI †). The values of the electrostatic potentials on the hydrogenbond donors and acceptors could, unfortunately, not be correlated to the formation of the observed synthons in the cocrystal series (see ESI †). We observed, however, that under the initially studied LAG conditions, almost all tri-, tetra-and penta-FBAs form cocrystals based on type A synthons, while the mono-and di-FBAs participated in hydrogen bonding through type A, B and C interactions (see ESI †).
The difficulty of predicting supramolecular synthons in cocrystals composed of thp and FBA is further highlighted by the discovery of polymorphs of four (thp)·(FBA) cocrystals, namely (thp)·(34diFBA), (thp)·(245triFBA), (thp)·(246triFBA) and (thp)·(23456pFBA). These polymorphs were observed during attempts to grow single crystals of the mechanochemically obtained crystal forms by crystallisation through slow solvent evaporation (see ESI †). Structural analyses revealed that the four solids exhibited both packing and synthon polymorphism. 51 Packing polymorphism was observed in the case of (thp)·(34diFBA), whereas (thp) ·(245triFBA), (thp)·(246triFBA) and (thp)·(23456pFBA) displayed synthon polymorphism. In particular, the synthon polymorphism of (thp)·(245triFBA) was realised through the formation of the type A synthon (Fig. 4a). On the other hand, synthon polymorphism in the cases of (thp)·(246triFBA) and  (thp)·(23456pFBA) was realised through the occurrence of N-HĲimidazole)⋯OC(urea) hydrogen bonds between two thp molecules, rather than the generally observed cyclic arrays of N-HĲimidazole)⋯OC(amide) hydrogen bonds (Fig. 4b). The serendipitous discovery of the four polymorphs is not particularly surprising, considering the rapidly increasing number of studies reporting cocrystal polymorphs, 51 while a recent study recognised that cocrystals exhibit polymorphism as readily as single-component organic crystals. 52 In light of these findings, it appears that a better understanding of synthon hierarchies in cocrystals could be achieved by surveys of their structural landscapes 29,53 through polymorph screens, which are not regularly performed at the present time. 54 So, what is the significance of the intriguing synthon diversity in the presented (thp)·(FBA) cocrystal series? The unpredictability of supramolecular interactions in the (thp)·(FBA) cocrystals, and lack of knowledge about the structural and electronic effects (on synthon hierarchies and cocrystal compositions) of fluorine substitutions in the cocrystal former, suggest that cocrystals are not as readily designed and predictably constructed as is so often portrayed in the literature. Supramolecular synthons are an incredibly useful synthetic tool that has greatly contributed to the development of advanced functional materials, 55 and to the preparation of strikingly complex solidstate structures. 56 The current state of the art in cocrystal design, however, does not support high-yielding supramolecular syntheses 57 of cocrystals for cases entailing molecules with more than two functional groups. To improve our ability to better design and more accurately construct cocrystals comprised of complex multifunctional molecules (such as pharmaceuticals), more targeted and systematic crystallographic studies need to be pursued and published. 11 But to fully un-derstand cocrystalstheir structural landscapes, synthon hierarchies and thermodynamic propertiescomputational methods (such as CSP 29,58 or lattice energy calculations 59 alone) need to be regularly used in crystal engineering. Comprehensive computational solid-state studies are very intensive (requiring a great deal of CPU time and time to process the generated data) and it can be argued that their use is, nowadays, only justified when dealing with specialty chemicals of particular interest (e.g. drug candidates in development for marketing). 60 Recent reports, however, have shown that even less exhaustive predictions can contribute substantially to experimental cocrystal research. 61 The maturity of contemporary first principles computational solid-state methods [62][63][64][65] and their relevance to crystal engineering is now widely recognised, 66,67 and it is time that they are equally widely embraced by the experimental community.
MKC and DKB gratefully acknowledge financial support from the UCL Faculty of Mathematical and Physical Sciences. DKB and WJ thank the Royal Society for a Newton International Fellowship and the Isaac Newton Trust (Trinity College, University of Cambridge) for funding. MA thanks the EPSRC for a studentship, while SAS acknowledges funding through the EPSRC CASE scheme with Pfizer. We are grateful for computational support from the UK national high performance computing service, ARCHER, for which access was obtained via the UKCP consortium and funded by EPSRC grant (EP/K013564/1). Lastly, we thank Prof. Sarah L.