High-throughput nanoscale crystallization of dihydropyridine active pharmaceutical ingredients

The use of encapsulated nanodroplet crystallization, for high-throughput screening of a selection of dihydropyridine active pharmaceutical ingredients, resulted in access to single component crystalline forms for all examples as well as the discovery of two novel solvates.


Introduction
Understanding the solid-state landscape of a small molecule active pharmaceutical ingredient (API) is an important step in the development of a new drug, not least because the kinetic solubility of different forms can impact the bioavailability of the molecule.The discovery of crystalline forms of an API, and the growth of high-quality single crystals of those forms for diffraction analysis, is highly beneficial not only for validating connectivity and stereochemistry, but also to provide insight into the solid-state properties of the compound.The indepth examination of the experimental crystallization space of a small molecule API by classical methods is, however, highly labour-intensive, due to the large number of experimental variables that need to be explored (Jones, 1981;Van der Sluis et al., 1989;Spingler et al., 2012;Wen et al., 2019).Even with increasing automation, such studies are costly in both time and in the mass of sample required, and can still fail to identify important crystalline forms (e.g.polymorphs, solvates or hydrates) (Morissette et al., 2004).However, modern approaches to high-throughput crystallization of APIs for single-crystal X-ray diffraction (SCXRD) are becoming more widely available, including ion exchange and vapour diffusion methods, microbatch under-oil techniques and Encapsulated Nanodroplet Crystallization (ENaCt) (Metherall et al., 2023;Nievergelt et al., 2018;Babor et al., 2019;Tyler et al., 2020).ENaCt is a high-throughput technique for the crystallization of organic soluble small molecules, in which nanolitre scale droplets of organic solvent containing only a few micrograms of analyte are encapsulated in a larger oil droplet, and allowed to crystallize.Using liquid handling robotics, large numbers of ENaCt experiments can be set-up in parallel, using multiwell plate formats, allowing rapid screening of large volumes of the multi-dimensional experimental space.Successful ENaCt experiments provide single crystals suitable for SCXRD, with ENaCt being previously applied to the crystallization of Nheterocyclic carbenes, aromatic polyketides, a SARS-CoV-2 protease inhibitor and cannabidiol (Zhu et al., 2022;Al Subeh et al., 2022;Cooper et al., 2022;Straker et al., 2023).
In this work we have examined ENaCt for the rapid parallel crystallization of a set pharmaceutically relevant APIs.Using a standardized approach, we postulated that crystal forms of a set of APIs could be accessed with minimal sample usage, a few milligrams, within two weeks for most molecules.We chose to study the dihydropyridine calcium channel blockers, a widely used class of antihypertensive drugs, which form a chemically related series based around a core dihydropyridine structure.The six dihydropyridines investigated include felodipine ( 1

Materials
All compounds, oils, and solvents (Table S1, supporting information) were used as purchased without any further purification.
SWISSCI LCP glass plates with a 100 mm spacer and SWISSCI LCP cover glass slips were used as purchased.

Crystal growth by ENaCt
In order to develop a rapid, highly parallel, ENaCt screening method a range of four oils and 12 solvents were chosen, representing a variety of physical properties (Table S2).
The four different oils were selected to aid the encapsulation of the organic nanodroplets with variations in viscosity and miscibility with common organic solvents, including two fluorinated oils [Fomblin YR-1800 (FY) and Fluorinert FC-40 (FC-40)], a mineral oil (MO) and a silicone oil (PDMSO).
Stock solutions of each dihydropyridine API were freshly prepared for each set of crystallization experiments.Samples were weighed ($2 mg) into screw top vials and dissolved, through serial addition of solvent, to form a near-saturated solution (Table S3).This enabled preparation of samples near to the solubility limit on a small scale, without requiring knowledge of solubility information for each solute/solvent combination.Approximately 24 mg of dihydropyridine API was employed in each case to prepare stock solutions, with around 0.25-0.5 mg used in each individual crystallization experiment.This approach allowed large numbers of experiments to be set up simultaneously with minimal sample requirements, in comparison to classical crystallization methods.
A 200 nL droplet of each oil was first deposited into the wells of a 96-well glass plate using an SPT Labtech mosquito liquid-handling robot (aspirate 1.0 mm min À1 , dispense 1.0 mm min À1 ) (Fig. S1, supporting information).Each API stock solution (50 nL) was then injected into each oil droplet (aspirate 20 mm min À1 , dispense 20 mm min À1 ).Plates were then sealed with a glass cover slip, checked by optical microscopy, and left for 14 days in the dark at ambient temperature ($25 C).
For each dihydropyridine API, three 96-well glass plates were employed, equating to 288 individual crystallization experiments including all combinations of the 12 solvents and four oils selected, as well as no-oil crystallization conditions (Fig. S1).
Figure 1 The six dihydropyridine calcium channel blockers studied.

Characterization of crystallization outcomes
2.3.1.Assessment of crystallization outcomes by optical microscopy.The crystallization plates were manually checked using optical microscopy for crystal growth.Observation of the experiment wells was carried out with a Nikon SMZ1000 microscope fitted with a linearly polarized light source and analyser.Digital images were captured with a GXCAM-U3-5 5.1 MP camera using the readily available ToupView (http:// www.touptek.com)software.The crystallization outcomes were categorized as F = experimental failure, 1 = sample remaining in solution, 2 = non-crystalline, amorphous or oily material, 3 = microcrystalline material and 4 = single crystal(s) suitable for SCXRD analysis (Fig. S2).

Single-crystal X-ray diffraction of crystals from
ENaCt experiments.Upon observation of suitable crystals (grade 4), the relevant wells were opened with the use of a tungsten carbide scriber to remove a small portion of the glass cover slide, and the crystal manipulated using MiTeGen Kapton microtools.Crystals were transferred onto a standard MiTeGen Kapton loop and mounted onto an in-house diffractometer, or stored in a liquid N 2 dry shipper for analysis on beamline I19 at Diamond Light Source, via remote access (Allan et al., 2017;Johnson et al., 2017).Unit-cell analysis of mounted crystals was undertaken, and full data collections were performed for all previously unknown dihydropyridine API crystal forms, along with representative examples of known dihydropyridine API crystal forms.In-house data were collected for 1 (form IV and form I), 2, 2Á1,4-dioxane, 4, 6 and 6ÁDMSO using either a Bruker D8 Venture (4 and 6ÁDMSO) with IS microfocus source (Cu K1, = 1.54178A ˚) and Photon II detector at 150 K, [data were reduced using APEX3 software (Bruker, 2015), incorporating SAINT (v.8.40B;Bruker, 2017), SADABS was used for absorption correction (Bruker, 2016), or a Rigaku Oxford Diffraction Synergy-S diffractometer [1 (form IV and form I), 2, 2Á1,4-dioxane and 6] (Cu K, = 1.54184A ˚) with a hybrid pixel array detector at 150 K [data were reduced using CrysAlis PRO with SCALE 3 ABSPACK correction implemented (Rigaku Oxford Diffraction, 2022)].All samples were cooled, and temperature maintained using an Oxford Cryosystems Cryostream (Cosier & Glazer, 1986).Data for 3 and 5 were collected at Diamond Light Source at 100 K, using synchrotron radiation ( = 1.54178A ˚and = 0.6889 A ˚, respectively) and the data were processed using APEX3 software.All structure solutions and Crystallographic data and corresponding refinement information for 1 (form IV), 2 ( form), 2Á1,4-dioxane) and 3.
H atoms treated by a mixture of independent and constrained refinement for all structures.
Starting models from structure solution were completed and corrected through iterative rounds of least-squares minimization and analysis of Fourier difference maps of electron density.Felodipine (1, form IV) showed disorder in both sidechain terminal methyl groups.NimodipineÁDMSO (6ÁDMSO) showed disorder in the dimethyl sulfoxide solvent molecule (Table 2).

Analysis of crystallization outcomes
The first crystals suitable for analysis by SCXRD were observed one hour following crystallization plate set-up, with further crystallization occurring and crystal growth over time.
After 14 days no further significant change in the total number and size of crystals was observed and end-of-experiment plate readouts were performed (Section S5.2, supporting information) and crystallization outcomes recorded (Fig. 2).
For each dihydropyridine API, the most successful solvent/ oil combinations to produce crystals suitable for SCXRD analysis [grade 4 (Fig. S2)] were examined.Some molecules only provided crystalline material from a very limited set of conditions, such as a single solvent or oil, while others were more versatile, crystallizing from wide variety of conditions.Across the dihydropyridines examined (1-6), in all cases the use of an encapsulating oil was beneficial in comparison to the 'no oil' controls (Fig. 2).
The crystallization of felodipine ( 1) is clearly favoured in the presence of mineral oil, with every solvent in combination with mineral oil resulting in crystals suitable for SCXRD analysis.In this case the choice of oil had the biggest influence on the crystallization outcomes, with the solvent being less --Absolute structure parameter -À0.6 (7) --important, making felodipine (1) a partial outlier in the series, although MO also proved useful in the crystallization of nifedipine (2) and nimodipine (6).Interestingly, two polymorphs of felodipine (1) were identified, with form IV being obtained from DMSO, MeOH, toluene, DCE and MeNO 2 , whilst form I being obtained from DMF, 2,2,2-TFE, 2-MeTHF, 1,4-dioxane, EtOAc, MeCN and MIBK.The two crystals selected for structural determination were obtained from toluene/MO (form IV) (Fig. S11) and EtOAc/MO (form I) and correspond to the previously reported structures [CSD refcodes: DONTIJ03 (Surov et al., 2012) and DONTIJ (Fossheim, 1986)].Nifedipine (2), also gave a number of crystals of a suitable size and quality for analysis by SCXRD, again with mineral oil and FC-40 providing the most promising results, albeit with fewer overall hits than in the case of felodipine (1).The crystallization of nifedipine (2) appears to benefit from the use of more polar solvents including 2,2,2-TFE, MeNO 2 , DMF and DMSO, but also 1,4-dioxane.The crystal initially selected for SCXRD analysis came from 1,4-dioxane/MO (Fig. S11) and a nifedipineÁ1,4-dioxane solvate (2Á1,4-dioxane) was obtained, which proved to be a new polymorph of the previously known nifedipineÁ1,4-dioxane solvate [CSD refcode: ASATOD (Caira et al., 2003)].A similarly high crystallization preference came from the use of 2,2,2-TFE as the solvent and, therefore, a crystal was also selected from 2,2,2-TFE/MO (Fig. S11).SCXRD analysis showed that the nifedipine (2) crystal matched the previously reported form [CSD refcode: BICCIZ06 (Ou et al., 2020)].It should also be noted that unitcell analysis also revealed the presence of the form, although a full data collection was not possible from the crystals obtained.In the case of nifedipine (2), the form is the more favoured polymorph, being obtained from DMSO, 2,2,2-TFE, 2Me-THF, MeCN, MIBK and MeNO 2 , whilst the form was only observed from DMF and DCE.
FY proving successful and with no observable trends based on the appearance of grade 4 crystals (Fig. 2).However, if microcrystalline (grade 3) results are also taken into account, then a preference for toluene and MIBK as solvent can be observed (Fig. S9).The crystal selected and analysed using SCXRD came from MIBK/FY oil (Fig. S11) and corresponded to the previously reported structure [CSD refcode: FULPAD (Fossheim et al., 1988)].
Nitrendipine (4) proved to be highly indiscriminate, providing crystals suitable for SCXRD across a wide range of different solvent and oil combinations, with DMSO and DMF being the most successful solvents.Even 'no oil' conditions gave good quality crystals from DMSO and DMF.The crystal selected and analysed by SCXRD came from DMF/MO (Fig. S11) and corresponded to the previously reported structure [CSD refcode: JEXKUS (Langs et al., 1990)].
The crystallization of cilnidipine ( 5) by ENaCt proved more challenging, with toluene being the preferred solvent in combination with the fluorinated oils FC-40 and FY.Additionally, a small number of crystals were obtained from DCE and MeNO 2 , both in combination with mineral oil.The conditions used to grow the crystal analysed were toluene/FY (Fig. S11) and corresponded to the previously reported structure [CSD refcode: VELZUI (Hu et al., 2006)].
Crystals of nimodipine ( 6) suitable for analysis were observed across multiple experimental conditions, showing a slight preference for the use of mineral oil for droplet encapsulation.The crystal initially selected was from MeNO 2 / MO (Fig. S11) and corresponded to the previously reported structure [CSD refcode: VAWWEW (Langs et al., 1990)].It was noteworthy that nimodipine (6) gave high-quality crystals from DMSO in combination with all four oils and even the nooil conditions (Fig. 2).Since this strong solvent dependence was unusual, a crystal was selected from DMSO/FY (Fig. S11), resulting in the solution using SCXRD of a novel nimodipi-neÁDMSO solvate (6ÁDMSO).

Analysis of the crystal structures
The crystal structures of dihydropyridines 1-6, including the novel solvates 2Á1,4-dioxane and 6ÁDMSO, are now discussed in detail (Fig. 3).

Supramolecular features
In this section we compare the packing across the series of dihydropyridine crystal forms, including a detailed analysis of the new solvates nifedipineÁ1,4-dioxane (2Á1,4-dioxane) and nimodipineÁDMSO (6ÁDMSO).The packing of all the nonsolvated crystal structures (1-6) is dominated by one-dimen-sional N-HÁ Á ÁO hydrogen-bond chains.All compounds besides nifedipine (2) have chiral centres, and all compounds besides cilnidipine (5) crystallize in centrosymmetric space groups, with both R and S enantiomers present for the chiral molecules.Cilnidipine (5) crystallized in the non-centrosymmetric space group Fdd2.For crystal structures 1, 3 and 4 a linear supramolecular chain is observed, involving the dihydropyridine N-H and the carbonyl oxygen in the side chain of the adjacent molecule.In contrast to this, crystal structures 2, 5 and 6 show zigzag supramolecular chains involving the dihydropyridine N-H and either a carbonyl oxygen (2) or an ether oxygen (5 and 6) in the side chain of the adjacent molecule (Fig. 4).
On closer inspection, the linear chain arrangement exists when a carbonyl oxygen in the 3,5-position (that is involved in the hydrogen-bond network) faces in the opposite (anti) direction to the dihydropyridine N-H group (anti-carbonyl).In contrast to this, the zigzag arrangement exists when the carbonyl oxygen points in the same (syn) direction as the dihydropyridine N-H (syn-carbonyl).
The solvated crystal structure nifedipineÁ1,4-dioxane (2Á1,4dioxane) shows a linear supramolecular chain involving an anti configuration of the dihydropyridine N-H and the carbonyl oxygen in the side chain of the adjacent molecule (Fig. 5).This is in contrast with the non-solvated crystal structure of nifedipine (2), in which a zigzag arrangement is observed.The change in the packing of the solvate is due to the inclusion of the 1,4-dioxane solvent molecule.The 1,4-dioxane molecule occupies channels nearest to the syn-carbonyl oxygen, previously used in the extended hydrogen-bond network, thus only the anti-carbonyl oxygen is available for inclusion in the supramolecular chain.
In the case of the nimodipineÁDMSO solvate (6ÁDMSO), the inclusion of DMSO disrupts the common hydrogen-bond networks observed across the related dihydropyridine struc-tures, in this case disrupting the dihydropyridine N-HÁ Á ÁO (ether) hydrogen bond.Thus, the nimodipineÁDMSO solvate (6ÁDMSO) crystallizes as a finite hydrogen-bonded unit, in which the DMSO hydrogen bonds to nimodipine via the dihydropyridine N-H bond.The disruption of the hydrogenbond network can also be seen through overlay of the molecules within the crystal structures of 6 and 6ÁDMSO, which shows that the ether-containing side chain has reoriented from the non-solvated to the solvated structure (Fig. 6).

Hirshfeld surface analysis
Key features of intermolecular interactions in the crystal structures of the dihydropyridine APIs can be more easily visualized with the aid of Hirshfeld surfaces.
The differences between the zigzag arrangements (2, 5 and 6) and linear arrangements (1, 3 and 4) can be readily seen in the Hirshfeld surfaces, particularly in the case of nifedipine (2) and nifedipineÁ1,4-dioxane solvate (2Á1,4-dioxane).In the structure of nifedipine (2), the zigzag hydrogen bonding results in an angle of 121.8 between the carbonyl oxygen on one molecule, the dihydropyridine nitrogen atom and the carbonyl oxygen on the adjacent molecule.In comparison, the linear hydrogen-bond network in nifedipineÁ1,4-dioxane  solvate (2Á1,4-dioxane) results in an angle of 150.5 between the carbonyl oxygen on one molecule, the dihydropyridine nitrogen atom and the carbonyl oxygen on the adjacent molecule (Fig. 8).
Across the series of the dihydropyridine crystal structures studied, in almost all cases 'hot spots' corresponding to the hydrogen bonding interactions with either a carbonyl or an ether oxygen are present.The Hirsheld surface of nimodipi-neÁDMSO solvate (6ÁDMSO) shows that this is the exception.The inclusion of DMSO into the structure interrupts the hydrogen bonding interactions with adjacent nimodipine molecules, such as seen in the Hirsheld surface of 6.As a result, the 'hot spot' around the ether oxygen is no longer present in the Hirsheld surface of 6ÁDMSO, the major interaction being the hydrogen bond from the dihydropyridine N-H to the DMSO oxygen.The Hirsheld surface also highlights a weak interaction with the DMSO hydrogen and the aromatic ring of nimodipine (Fig. 7).

Conclusions
Six dihydropyridine APIs were successfully crystallized using the high-throughput parallel ENaCt method.A total of 1728 individual crystallization experiments were performed, 288 per compound covering 60 different experimental conditions (solvent/oil combinations), with minimal experimental set-up time.In all cases, single crystals suitable for SCXRD were obtained within two weeks.Analysis of the successful crystallization 'hot spots' from ENaCt allowed the detection of two previously unreported crystalline forms, a new polymorph of the nifedipineÁ1,4-dioxane solvate (2Á1,4-dioxane) and the first known solvate of nimodipine, nimodipineÁDMSO (6ÁDMSO).In addition, two polymorphs of both felodipine (1) and nifedipine (2) were observed.This rapid access to crystalline forms for a series of APIs, along with the discovery of the novel solvates, demonstrates the potential of highthroughput ENaCt screening methods for application to the discovery of API crystalline forms.It is also interesting to note, for some dihydropyridines, the choice of solvent appeared to be the major factor in determining successful crystallization [e.g.cilnidipine (5)/toluene] whilst for others the oil employed was the major driving factor [e.g.felodipine (1)/MO].Further research to better understand the role of solvents and oils in ENaCt for APIs is the subject of ongoing work.

Figure 4
Figure 4 Supramolecular features of non-solvated dihydropyridines 1-6 showing linear and zigzag chain arrangements.Extension of the hydrogen-bond network of 1 and 3 in the crystallographic [101] direction, 4 in the [001] direction, 2 in the [010] direction, 5 in the [001] direction and 6 in the [010] direction.Hydrogen atoms not involved in hydrogen bonding are omitted for clarity.Grey: carbon, blue: nitrogen, red: oxygen and green: chlorine.

Figure 5
Figure 5Top: overlay of nifedipine molecules from non-solvated (grey, 2) and solvated (green, 2Á1,4-dioxane) crystal structures showing conformational similarity, anti and syn arrangements highlighted.Bottom: extension of the hydrogen-bond network viewed in the crystallographic (101) plane.Hydrogen atoms not involved in hydrogen bonding are omittedfor clarity.

Table 2
Crystallographic data and corresponding refinement information for 4, 5, 6 and 6ÁDMSO.H atoms treated by a mixture of independent and constrained refinement for all structures.