Building Up Co-Crystals: Structural Motif Consistencies Across Families of Co-Crystals

The creation of co-crystals as a route to create new pharmaceutical phases with modified or defined physicochemical properties is an area of intense research. Much of the current research has focused on creating new phases for numerous active pharmaceutical ingredients (APIs) to alter physical properties such as low solubilities, enhancing processability or stability. Such studies have identified suitable co-formers and common bonding motifs to aid with the design of new co-crystals but understanding how the changes in molecular structure of the components are reflected in the packing and resulting properties is still lacking. This lack of insight means that the design and growth of new co-crystals is still a largely empirical process with co-formers selected and then attempts to grow the different materials undertaken to evaluate the resulting properties. This work will report on the results of a combination of crystal structure database analysis with computational chemistry studies to identify what structural features are retained across a selection of families of co-crystals with common components. The competition between different potential hydrogen bonding motifs was evaluated using ab initio quantum mechanical calculations and this was related to the commonality in the packing motifs when observed. It is found while the stronger local bonding motifs are often retained within systems, the balance of weaker long-range packing forces gives rise to many subtle shifts in packing leading to greater challenges in the prediction of final crystal structures.


Introduction
Modifying the physicochemical properties of active pharmaceutical ingredients (APIs) through the creation of multi-component crystals (i.e., co-crystals and salts) is a rapidly developing field of study [1]. They have been primarily investigated to adjust solubility and so increase bioavailability [2], but other studies have also included increased stability under storage conditions, improved tableting performance or processability [3,4]. Such studies are often empirically driven with screening of standard co-formers under a range of experimental conditions used to locate new potential phases before undertaking evaluation of the physical properties [5]. Co-formers are frequently chosen by consideration of potential supramolecular synthons that could form between the components [6]. Many potential co-formers offer a number of potential supramolecular synthons and it is not always predictable which will from and how these will lead to desirable properties. One route to achieve this understanding is to undertake data mining exercises into the existing crystal structures from families of multi-component crystals with common structural features throughout the set and identify if there are relationships in the component choice and the resulting crystal structures [7][8][9].
Computational chemistry approaches offer key information to aid with these studies as the intermolecular interactions within known crystal structures, predicted crystal structures or supramolecular motifs can be readily evaluated. Studies into the formation thermodynamics of cocrystal and salts are limited, however the majority of multi-component crystals are dominated by the enthalpic component [10,11]. Analysis of lattice energy components indicates two or three close contacts dominate the energy. This suggests that an initial screening could be based on the interactions between the components [12]. For this to be effective suitable bonding motifs need to be identified or predicted for a given system. Isonicotinamide (INA) and nicotinamide (NA) are frequently used co-formers, both for modification of API's physicochemical properties [13,14] and as model systems for co-crystallization studies [15,16]. Both INA and NA offer one hydrogen bond donor/acceptor group (CONH2) and one hydrogen bond acceptor (Npyr) but in different orientations, this allows for studies into the influence on relative positioning on the final crystal packing. Thus, there are only three potential interactions with suitable hydrogen bond donor groups such as CO2H or OH groups (Error! Reference source not found.). Previous studies indicate a preference for the binding between co-formers and the pyridine nitrogen in most co-crystals [17], however it would be expected for carboxylic acids that the acid… amide motif should be lowered in energy as it comprises two hydrogen bonds. It has been shown for the acid amide interaction in benzoic acid-benzamide co-crystals that the strength of this interaction depends on the nature of the functional groups present [18] and so the changes in the balance between these interactions will be investigated to identify the interplay between the nature of the other functional groups present in the molecule.
Intermolecular hydrogen bonding motifs were identified in the various structures, extracted from the crystal structures and optimized at the DFT level (TPSS-D/def2-TZVP) [24][25][26][27] using the program orca. The interaction energy for these systems were then evaluated in orca [28] at the DFT (RI-PWPB95-D/ma-def2-QZVPP) [25,29]. The basis set superposition error was evaluated to be ~0.1 kJ mol −1 and so not included in calculations for tetramers and higher. Energy frameworks for selected crystal structure were evaluated in CrystalExplorer [30] using B3LYP/6-31G(d,p) DFT theory. Pairwise optimization of molecular components was carried out using the drift bias free differential evolution global optimization algorithm (Control parameters: K, F, Gmax, Np = 0.5, 0.98, 5000, 120) [31] using the AA-CLP force field [32] to evaluate the interaction energy between the molecules.

Co-former and Compositional Analysis
A total of 215 isonicotinamide and 158 nicotinamide multi-component crystals were located and grouped based on the functional group hydrogen bonding to the pyridine nitrogen), the numbers of different compositions were also identified. The results show that the distribution is similar for either co-former. NA systems have a higher proportion of compositions compared to INA, which may be just to the relative bent geometry of NA functional compared to the linear location in INA. There may be a preference for (1:1) compositions, but this may reflect the screening approaches used and the higher composition systems may be harder to grow compared to the (1:1) composition. It has been reported that the (2:1) benzoic acid-INA co-crystal is metastable compared to the (1:1) and only forms under limited experimental conditions [15,33].

Crystal Packing Analysis
(1:1) systems were selected for further analysis to identify repeating patterns in the bonding. Polymorphic systems were counted as separate systems as the hydrogen bonding motifs could be different in the different phases. First level graph set patterns were generated for each system (Table 1) to evaluate the differences between the phases.  For INA, the dominant structure is a tetramer motif in 72% of the INA co-crystals (Error! Reference source not found.a). The remaining systems form a chain motif through two or more functional groups present in the co-former. In contrast, the opposite behaviour is observed in NA systems with only 36% of systems forming a tetramer motif (Error! Reference source not found.b). While the INA examples form planar structures, the motif in NA is twisted (the angle between benzoic acid and INA/NA molecular planes are 6° and 127° respectively). The carboxylic acid-INA systems show shifts in the form of this tetramer with 28% featuring a cis geometry between the carbonyl groups (Error! Reference source not found.). Polymorphic 2,4,6-trifluorobenzoic acidisonicotinamide (PIRNOV) display both geometries in different phases. DFT energy calculations, show the trans geometry is lower by 2.15 kJ mol −1 . All bar one of the cis based tetramers form a consistent 3rd level hydrogen bonding motif forming a 1-D ladder structure built from a (26) motif (Error! Reference source not found.). The exception is form I of 5-chlorosalicylic acid (JIPJEQ) which forms a (11) motif instead. However, none of the 3-D structures display any further packing similarities. Thus, the strongest interactions in the structure may be predictable, but the range of weaker packing forces can lead to a number of different crystal structures.   The majority of trans systems can be classified into three sets: M2 has a 1-D ladder linked through a (22) motif (24/56 systems, Figure 1a), M3 an interlocked 3-D structure with (11) motifs (23/56 systems, Figure 1b), M4 links into a 1-D ladder through amide…amide ring with a (8) motif (4/56 systems, Figure 1c). The M2 set has three isostructural sets of structures, while 15 M3 structures pack in an isostructural set of C2/c structures, typified by benzoic acid (Figure 1b). The remaining groups have flatter angles. The M4 set splits into two isostructural subsets. This again confirms consistent local bonding but differences in longer packing is different.

Binding Interactions
The energy of intermolecular interactions within the (1:2) dicarboxylic acid-INA co-crystals were evaluated within CrystalExplorer. Summing the three strongest interactions in the structure (INA … INA, N-H … O=C, CO2H … Npyr)) shows a general decrease with increasing chain length, however a slight odd-even alternation is noted (Error! Reference source not found.). Comparing this to the reported melting point shows a complementary pattern. However, the trend is not perfectly matched; there are two reported values for the melting point of the succinic acid co-crystal (135 °C [34] and 206 °C [35]) which affects the curves. Optimization of the position and orientation of molecules of oxalic acid with INA links the molecules through an acid … amide bond. This motif is not observed in any of the oxalic co-crystal structures but is lower in energy than the observed motif ( Table 2). Evaluation of formation energy for the tetramer over two acid … amide dimers for a model system shows that the tetramer is favoured energetically by −26.74 kJ mol −1 . Thus, a set of slightly weaker interactions can give overall increased energy compared to the utilization of strongest possible interaction at each point. Similar evaluations for NA systems show similar energy values to INA ones and so the differences in packing appears to be dominated by the geometry and packing forces rather than preferential energetics.

Conclusions
INA co-crystals and salts show consistent bonding features in the resulting crystal structures, with a common tetramer forming 72% of cases, in contrast NA only forms this motif in 36% of structures. Evaluation of the energy of the hydrogen bonds shows that in INA, while the stronger individual interactions can be predicted the combination of motifs in the tetramer gives a lower overall energy. This highlights that the balance between numerous factors dominates the cocrystallization processes. Comparing INA and NA system have similar energy values and so the differences in packing appear to be driven by the ability to pack efficiently in the crystal form rather than energetically. This indicates that to develop predictive methods based on pointwise contact location of the range of possibilities is more important compared to location of the global minimum interaction and evaluation of potential higher order building blocks will give insights into the resulting properties of the materials.

Conflicts of Interest:
The authors declare no conflict of interest.