Mechanistic In Situ and Ex Situ Studies of Phase Transformations in Molecular Co‐Crystals

Abstract Co‐crystallisation is widely explored as a route to improve the physical properties of pharmaceutical active ingredients, but little is known about the fundamental mechanisms of the process. Herein, we apply a hyphenated differential scanning calorimetry—X‐ray diffraction technique to mimic the commercial hot melt extrusion process, and explore the heat‐induced synthesis of a series of new co‐crystals containing isonicotinamide. These comprise a 1:1 co‐crystal with 4‐hydroxybenzoic acid, 2:1 and 1:2 systems with 4‐hydroxyphenylacetic acid and a 1:1 crystal with 3,4‐dihydroxyphenylactic acid. The formation of co‐crystals during heating is complex mechanistically. In addition to co‐crystallisation, conversions between polymorphs of the co‐former starting materials and co‐crystal products are also observed. A subsequent study exploring the use of inkjet printing and milling to generate co‐crystals revealed that the synthetic approach has a major effect on the co‐crystal species and polymorphs produced.


Introduction
Co-crystallisation is am ethod by whicht he physicalp roperties of am olecule can be altered withoutm aking or breaking any covalentb onds. [1] This can directly affect the propertieso fa n active pharmaceutical ingredient( API), and (for instance) may improvei ts solubility and/orr ate of dissolution. As ac onsequenceo ft he substance existingi nacrystalline form, co-crystals are likely to be more stable, reproducible and easier to purify than other solid forms of ad rug, and, therefore, are more desirable. [1] An umber of approaches exist to identify suitable co-formers for an API, [2][3][4] but at the momentt here is little understanding of the mechanism by whicht he crystals form.T oe xplore this, non-invasive probes are required. While these are easy to implement in the liquid or gas phase, non-invasivep robesfor the solid state are lacking.
Differentials canningc alorimetry (DSC) is the main method used in the pharmaceutical sciences to study how the physical state of materials changes as af unction of temperature. However,i td oes not allow definitive structural elucidation.P owder X-ray diffraction (XRD) is the "gold standard" analytical techniquef or the identification of crystalline forms, but standard powderd iffractometers are unable to heat the sample sufficiently rapidly to mirror DSC heatingr ates (ca. 10 8Cmin À1 ), and typically requireo ft he order of 10-30min to collecta high-quality diffraction pattern. Sequential analysis is possible, wherebyamaterial could be heatedi naD SC (or in an oven if largerm asses are required) and subsequently analysed by XRD to determinet he physical form. However,t he physicalf orm may change as the sample cools, particularlyi fm etastable materials are generated. To overcome this issue, we recently developed an ew hyphenated DSC-XRD analytical method. [5] This approachh as led to enhanced understanding of phase transitions in glutaricacidand sulfathiazole, [5] carbamazepine and dihydrocarbamazapine, [6] and paracetamol. [7] Most recently,w e employed the DSC-XRD platform with crystal structure prediction work and were able to identify and solvet he structure of an ew polymorph of olanzapine. [8] Herein, we extendt he DSC-XRD approach to explore the fabrication of co-crystals by thermalm ethods. Understanding these processes is important, becausef or industriala pplications co-crystals are most likely to be preparedu sing hot melt extrusion (HME), ac ontinuous manufacturinga pproach which appliesh eat energy to an intimate mixture of the co-formers to generateaco-crystal. [4,9,10] The DSC-XRD approach will permitu st ou nderstandt he transitions occurring during the thermals ynthesis of co-crystals, and as ar esult to designs uitable HME manufacturing protocols.
As proof-of-concept we have developed four new co-crystals based on isonicotinamide (INCT,F igure 1). INCT has been used extensively as ap harmaceutical co-former to improve the solubility of an API withoutc ompromising its efficacy or stability. [11][12][13][14][15][16][17][18][19][20][21][22] INCT is able to act as ah ydrogen bond acceptor through the pyridine group, andt he amide moiety is capable of engaging in aw ide range of differenth ydrogen-bonding motifs. Vishweshwar et al. generated a1 :1 co-crystal [23] between the antimicrobiala nd antioxidant [24] API 4-hydroxybenzoic acid (HBA, Figure 1) and INCT by crystallisation from hot methanol.Inthis work, we additionally explored 4-hydroxyphenylacetic acid (HPAA, Figure 1) and 3,4-dihydroxyphenylactic acid (DHPAA, Figure 1). Both are also antioxidants [25] and being closely structurally relatedt oH BA (see Figure 1) were expected to have highpotential for co-crystallisationwith INCT. [26] While HME is expected to be the most appropriate technique to preparec o-crystals for industrial use, traditionally solvent evaporation has been the mostc ommonly employed methodf or the preparation of pharmaceutical co-crystals. [27] Unfortunately,d ue to the use of organic solvents this process is ecologically unsound;i ti sa lso time consuming,w ith evaporation of the solvent occurring over days and weeks. Recently, thermali nkjet printing has proven to be ar apid alternative methodf or the preparation of pharmaceutical co-crystals. [28] Inkjet printing takes minutes to produce crystals of sufficient quantity and high enough quality for analysis. However,t he materials must first be in solution beforet hey can be printed and so the "green" problem remains. Another option is cogrinding stoichiometrica mounts of two dry powdered crystalline materials. This has been known to produce co-crystals since as early as 1893, although it has only gained prominence in academic laboratories since2 000. [29] The methodh as advantages over solvente vaporation and thermali nkjet printingi n that it is both fast (minutes timescale) and clean/green (requires no solvent), but it produces fine powders andn ot the single crystalsrequired forfull structure elucidation.
In this paper,w eu se hyphenated DSC-XRD to generate novel co-crystals of INCT with HBA, HPAA and DHPAA and investigate the formation mechanisms. We further explore sol-vent evaporation, thermali nkjet printing, and ball milling as alternative routes to co-crystallisation.

INCT-HBA
Of the four systems, only INCT-HBA has previously been shown to form co-crystals. In this work, we were able to grow from solution both the known structure (VAKTOR, termed form I here) [23] and as econd, previously unreported,p olymorph (form II). The key differenceb etween form Ia nd II of the INCT-HBA co-crystal is that form II is al ayered hydrogen-bonded structure, while form Ic ontains ah ydrogen-bonded network that extends in 3D ( Figure 2). Ad etailed discussion of the crystallographyc an be found in the Supporting Information (Section I.I, I.II, and I.III). DSC thermograms were recordedf or two different mixtures of INCT and HBA in a1 :1 molar ratio. The first was made by weighing and mixing the two as-supplied materials. The second was prepared by first grinding each of the two materials separately in ap estle and mortar, transferring the powders to ag lass vial, and mixingf or a3 0s on av ortex mixer.T he traces can be compared in Figure 3. Both mixtures produced similarp rofiles with an umber of events. The non-ground mixture has an almost imperceptible endotherm that peaks at 129.2 8C( onset:1 22.9 8C; enthalpy:3 .127 Jg À1 ), followed by another larger endotherm (onset: 143.2 8C, enthalpy:5 0.72 Jg À1 ) peakinga t1 45.6 8C. The ground mixture has as imilars et of events with as mall endotherm at 132.6 8C( peak:1 35.6 8C), immediately followed by as mall overlapping exotherm (peak:c a. 137 8C), which itselfi sf ollowed by an overlapping endotherm  (peak:1 44.0 8C). Due to the overlapping of these events accurate onset temperatures and enthalpies cannot be calculated. However,e stimates show the enthalpies of the two endotherms to be 4.8 Jg À1 and 63.7 Jg À1 respectively.
Both mixtures then go on to experience an endo--exo-endo event beginning at 173.9 8C( non-ground) and 173.4 8C (ground). Again, the onsetsa nd enthalpies of the succeeding exothermand endotherm in both datasets cannot be accurately analysed due to the overlapo ft he events. However,avisual comparison indicates the enthalpies are similara nd the peaks of the two events occur at 177.0 8Ca nd 185.0 8Cf or the mixture and at 176.9 8Ca nd 184.3 8Cf or the ground mixture. The similarity between all of the eventso bserved must be ar esult of the same phase transitions occurring in each of the two mixtures.
The events observed in the data for the mixtures all occur at temperatures at which there are no events observed in the thermograms fort he raw materials ( Figure 3). The highert emperaturee vents in both DSC traces must be caused by the cocrystallisation of the materials followed immediately by their melting. It appears that the initial co-crystallisation begins with am elt, characterised by the small endotherm immediately prior to the exotherm. It should be noted that some effects of decomposition are observed immediatelyf ollowing the endotherm at 145 8C. However,T GA data recordedf or samples of the same two mixtures( Figure SII.1) show that, although decomposition definitely begins prior to the second set of events, at 193 8C9 2.5 %o ft he material remains. The data strongly suggest the formation and subsequent meltingo fc ocrystals, despite this decomposition.
The explanation for the lower temperature group of events is less clear.D ata for the raw materials do not display such an event. To understand this in more detail, asample of the physical mixtureo ft he two materials as received was subjected to combined DSC-XRD analysis. The experiment was stopped at 170 8Ctop revent decomposition of the sample in the DSC cell. The diffraction data and DSC thermogram ( Figure 4) display two very obvious phase changes beginning at 123.4 8C The events observed in DSC-XRD are clearly the same as those observed by DSC, since they occur at similart emperatures and have similar associated enthalpies.
Rietveld refinement against the pattern for the physical mixture recorded at 40 8Ci nt he DSC can be seen in Figure SII.2. Startingm odels used for all refinementso nt his system are presented in Ta ble SII.1.R efinedu nit cell data can be seen in Ta ble SII.2. The initial sample was ap hysicalm ixture of INCT form Ia nd HBA ands ot he pattern recorded was expected to be ac ombination of the two and initial refinements were carried out using these materials. However,t here was as mall unidentified reflection at 2.618 and so other possible structures were introduced,i ncluding all known polymorphs of INCT,a nd the form I [23] and II co-crystals. The best fit was achieved when refiningt he structure for the form II INCT-HBA co-crystal. It seems that simply mixing the two materials resulted in as mall amount of co-crystallisation.
At 97 8Cc haracteristic reflections of the form II co-crystal are much more intense, and reflections from af ourth structure are present.T he resultso fR ietveld refinements carriedo ut on the pattern recorded at this temperature can be seen in Fig   veals the presenceo ff ive species:I NCT forms Ia nd II, HBA, and both co-crystals. Thus, rather than as imple conversion from the two initial materials to the known co-crystal there appear to be numeroust ransformationso ccurring. INCT Ia nd II are enantiotropically relatedw ith the I!II conversion reported by Li et al. [30] to occur at 131.7 8Ca taheating rate of 70 8Cmin À1 .A tl ower heating rates this can be expected to occur at al ower temperature and samples from the same study by Li were shownt oc onvert to form II at ca. 120 8C.
Thus the presence of form II at 122 8Ci sn ot unexpected.A t this temperature both formso ft he co-crystal are still present. The largestr eflection in the calculated pattern for this species( 3.48)i sp robably ar esult of the software compensating for the disproportionately high intensity reflection present at the same angle in the pattern of VAKTOR. The vast majority of the remaining crystalline materialc an be attributed to the structureo ft he form Ic o-crystal. However,t here remainss ome residual HBA that has not co-crystallised.I ts eems odd that there is pure HBA left but no INCT as the initial mixture was prepared at a1:1 molar ratio and the unit cell of the co-crystal formed contains one molecule of each. It could perhaps be the case that the initial mixture containeds lightly more HBA than INCT,o rt he two were not perfectly homogeneously mixed in the DSC pan.
Plotting the total integrated intensity of each phase present as af unction of time ( Figure 5) it can be seen that there is little change in the overall contento fH BA throughout the experiment until ca. 145 8C. There is some initial growth, presumably due to heat expansion causing more material to be lifted into the beam, followed by ag radual decline as the two cocrystalsg row.I NCT Ie xperiences ac oncomitant decline. At ca. 120 8Ct here is as harp drop in the content of INCT Ia nd form II INCT-HBA, accompanied by equally sharp increases in the content of INCT II and form II NCT-HBA. It seems that residual INCT Ii sc onverting to INCT II and form II of the co-crystal is converting to form I. At 132 8CI NCT Ia nd form II INCT-HBA have disappeared and the growth of form Is lows before, soon after,I NCT II and HBA undergo conversion to form II NCT-HBA. However,t he INCT disappears at am uch higherr ate than HBA.
Ap hysicalm ixture of INCT and HPAA at am olar ratio of 2:1 was explored by DSC-XRD. Initial DSC analysis can be seen in Figure SII.7. Thet hermogramr ecorded for the mixture has three endothermic events, all of which are absent from the thermograms of the two raw materials. The first is very broad and has an onseta t9 4.4 8Ca nd peaks at 97.9 8C. The enthalpy associated with this event is ca. 54.9 Jg À1 but there is av isible difference in the baseline before and after the peak due to the  extendedl ead in to the following endotherm, and thus accurate measurement is impossible. The following events are overlapping, but the onset of the first occurs at ca. 123.6 8C, with a peak at 125.2 8C, and the second has an onset of ca. 126.3 8C and peaks at 127.7 8C. The associatede nthalpies cannotb ea ssessed. The broad nature of the first endotherm may conceal a number of events, whereas the second and third are much sharper and may indicate melting. This cannotb ea ttributed to the melting of either of the raw materials in their most stable forms, as the temperature is too low.T GA of the mixture and the raw materials (data not shown)s hows that mixing the two has as tabilising effect, with the raw materials experiencing 10 %m assl oss at 193 8C( INCT) and 195 8C( HPAA), and the mixture at 205 8C. This suggestst hat intermoleculari nteractions between the components may exist.
The same physicalm ixture was subjected to combined DSC-XRD (Figure 7). The diffraction datas how two major phase transitions, occurring at ca. 95 8Ca nd ca. 125 8C. Each of these has ac orresponding endotherm in the DSC trace with onsets at 94.6 8Ca nd 127.0 8Cr espectively.T he thermogramh as a very similar form to those discussed above.T he first endotherm is smallert han the second but has as mall shoulder on either side, indicating that the event occurring at this temperature either occurs in multiple stageso ri si nf act multiple events. The second, larger,e ndotherm also hasasmall event occurring justb efore it. Again, this suggestsatwo stage process or multiple events. The total loss of Bragg reflections fol-lowing the endotherm at 127.0 8Cc onfirms the event as a melt.
Results of Rietveld refinement against the initial pattern recordedf or the mixture can be seen in FigureSII.8 and Table  SII.7. The calculated patterns show the initial sample to be made up of mostly INCT Ia nd HPAA as expected, buts urprisingly there was also anothers pecies present. Structures of each of the known polymorphs of INCT were included in the refinement but none improved the fit. There are no other knownp olymorphs of HPAA, but inclusion of the 2:1c o-crystal showed it to be ac lose match to the unknown species. The fit of the calculated pattern to the data is not as good as for previous systems, with a R wp of 0.1431. This can be partially explained by thes ignificant difference in intensity between the data and the refinements of the reflections at 2.718 and 3.178, characteristic of INCT.T his can be attributed to the presence of large grains of material, which resultedi ne ffects similar to preferred orientation.T here are also three low intensity observed reflectionsa t0 .738,2 .628,a nd 2.928 that are absentf rom the calculated pattern. The cause of these reflections is unclear.A ll three reflections are present from the first recordedp attern and fade from the data at the same temperature as the HPAA reflections, suggesting they are related. In contrast, the INCT reflectionsf ade at ah ighert emperature and the co-crystalr eflections are present throughout the experiment. The three unexplained reflectionsmay be the result of some impurity in the sample.
Refinement against the pattern recorded at 113 8C ( Figure SII.9), between the two endotherms, has shown the sample to consisto fj ust INCT and the co-crystal following the first phase transition, with the majority of the material being the co-crystal. The overall R wp and the standard uncertainties (Table  SII.8) for the co-crystal are much lower than those calculated for the same speciesi nt he 40 8Cp attern. The major reflections in the co-crystal pattern occur at similar angles to those in the lower temperature pattern, and the three unidentified reflections are no longer present.T his strengthens the argument that the co-crystal was present in the 40 8Cp attern. It appears that, like the INCT-HBA mixture, just mixing INCT and HPAA causeds ome co-crystallisation to occur.
The originals tock mixture was made up at an INCT:HPAA molar ratio of 2:1, so it seems odd that at 113 8C, following the exhaustion of the HPAA, there is still as ignificant amount of INCT left in the sample. The integrateda rea under the curve for the pattern of each of the reactants at 40 8Cw as used to calculatet he relative amountso fe ach materiali nt he beam, and it seems there wasa ne xcess of INCT of around 22 %o f the total sample in the beam. The remaining 12 %c an be accounted for by the meltingo fs ome of the HPAA. This is visible in the diffraction patterns above ca. 100 8Ca sa ni ncrease in the background intensity (Figure 7), characteristico faliquid or amorphous material.
The evolution of the systemsp resent in the sample can be visualised in Figure 8. Initially the materialc onsisted of mostly INCT,w ith less HPAA andalittle of the co-crystal. The content of all three remained relatively stable until around8 0-90 8C when both INCT and HPAA began to decrease rapidly,w hilst the co-crystal content increases. This is the result of co-crystallisation and coincides with the first endotherm in the DSC trace. This endotherm arises from multiple events, with both co-crystallisation and melting occurring simultaneously.A s there are three peaks it is probable that both INCT and HPAA undergo al ocal melt, and then recrystallise as the 2:1c o-crystal. The combination of the two endothermic eventsp resumably masks the exothermic crystallisationp eak. Abovec a. 100 8C the HPAA hasb een exhausted and the remainingI NCT continues to melt while co-crystal formation slows and eventually stops. Finally,t he co-crystal melts. The melting of INCT and the co-crystal coincide with the second endothermi nt he DSC trace, and the occurrence of these two melts explains the presence of the small shouldero nt his event.T he presence of the co-crystal appearst od estabilise INCT so that it melts at a much lower temperature than the pure crystalline powder.

INCT-DHPAA
The INCT-DHPAA co-crystal which forms from solvent evaporation is made up of staggered chains (Figure 9) with alternate INCT and DHPAA molecules linked by phenol···acid and phe-nol···pyridine hydrogen bonds (see also the Supporting Information, SectionI.VI).
Diffraction patterns collected during simultaneousD SC-XRD experiments on ab inary mixture of INCT and DHPAA ( Figure 10) appear to show the occurrence of two phase transitions. Reflections of the phases present before and after the first transition overlap in ab road range of temperatures (ca. 95-115 8C). From thesed ata alone it would appear that a single event is occurring. However,t he DSC thermogram shows that there are in fact at least three endothermic events taking place between 90 8Ca nd 120 8C. The final transition is much clearer and is represented by the complete loss of Bragg reflectionsa nd as harp endotherm, and is the final melting of the sample.
Closer examination of the diffractiond ata between 80 8C and 120 8Cd oes not offer any explanation of the multiple events in the thermogram. The Bragg reflections presenta t the beginning of the experiment are continuous until their dis-appearanceatc a. 120 8Cand the same is true of the reflections representing the second crystalline phase from their appearance at ca. 90 8Cu ntil their disappearance at 126 8C. This then suggestst hat the multiple endotherms are probably causedb y different stages of the same process as the crystal structures of the two materials are first disruptedb efore realigning into the structure of the co-crystal. Figure SII.10 and Table SII.9 show refinementd ata for the initial pattern recorded for the sample at 40 8C. Using the structures of the two raw materials achieved a very good fit with an overall R wp of 0.0556.
The pattern recorded at 122 8C( Figure SII.11), after the first phase transition, could not be fitted by the structures of any of the known polymorphs of the two raw materials or that of the INCT-DHPAA co-crystal grownb ys olvente vaporation. Many of the reflectionsi nt he pattern collected by DSC-XRD occur at similar angles to reflectionsi nt he predicted pattern (Figure SII.11), butt here are significant absences (in particulart he    028 and 3.768). The new structure must be either the result of co-crystal formation or the crystallisation of an ew polymorph of one of the two raw materials. There is also ah igh backgroundt hat emerges at the same time as the third endotherm in DSC ( Figure 11). This is not present in the lower temperature patterns, and indicates the presenceo fs ome amorphous or meltedm aterial.
Plottingt he integrated area under the calculated pattern of each species as af unctiono ft emperature ( Figure 11)i tc an be seen that the content of both of the raw materials remains relatively constant until ca. 95 8C, at which point both begin to decline.T his coincidesw ith the onset of the first endotherm in the thermogram( 94.2 8C). However,t he total disappearance of DHPAA and INCT does not occur until 108.8 8Ca nd 113.9 8Cr espectively.T hese endpoints match closely the minima of the second (ca. 108 8C) and third (111.05 8C) endotherms. The overlappingn ature of thesee ventsa nd the likelihood of an invisible exothermic event relating to crystallisation meanst hat a confident assignment is not possible. Thats aid, it can be deduced that the disappearance of both structuresf rom the sample is not due simply to melting, as the raw materials have meltingp oints at temperatures in excess of 120 8C. It is, therefore, likely to be the result of ac onversion from one solid form to another.I tc annot be ascertained from these data whether that form is ac o-crystal or an ew polymorpho fo ne of the two individual components.

INCT-HBA
The diffractionp atterns recorded for crystalso btained from printing have some significant differences from those of the raw materials (FigureSII.12). When compared to the predicted powderp attern of the structure of the form Ic o-crystal (VAKTOR) [23] andt hat of form II it is clear that the pattern of the printed crystals corresponds to the form Is ystem ( Figure 12). The data recorded here are of relatively lowr esolution due to the small crystallite and sample size, but all of the reflectionso ccur at the expected 2q anglesa nd the intensity ratios are very similar to those of form I. The slight discrepancy between the calculated pattern for form Ia nd that recorded for the printed sample is attributed to the two datasets being recorded at À196 8Ca nd room temperature, respectively.T he formation of this co-crystal is confirmed DSC, TGA, and IR spectroscopy ( Figure SII.13).

INCT-HPAA
Crystals couldb ep rinted from solutions of INCT and HPAA (molarr atio:2 :1) in either ethanol or ethanol and water (Figure SII.14). All the solutionse xplored yielded crystals with the same structure. The patterns obtained clearly do not match those of the individual as-supplied co-formers ( Figure SII.14), nor of any known polymorphs of the co-formers (data not shown). The printed crystals' pattern agreesw ell with the predicted pattern of the 2:1I NCT-HPAA co-crystal structure (Figure SII.15). Bragg reflectionsf or the printed crystalso ccur at slightly lower angles than for the single crystal data, but this is simplyaresult of the temperature difference between the measurements. The formation of the co-crystal was verified by DSC, TGA, and IR spectroscopy ( Figure SII.16).

INCT-DHPAA
Equimolars olutionso fI NCT and DHPAA in either ethanol or ethanol/water mixtures all produced crystals with the same structure ( Figure SII.17). Ac omparison of patterns recorded for the printed crystals, the raw materials separately,a nd ap hysical mixture of the two can be seen in Figure SII.18. The pattern for the printed crystalsi sc learly very different to those of the raw materials,b ut matches closely with the INCT-DHPAA cocrystal described above ( Figure SII.19). Successful formation of ac o-crystal was confirmed by DSC, TGA, and IR spectroscopy ( Figure SII.20).

INCT-HBA
The powder produced by ballm illinga ne quimolar mixture of INCT and HBA at 20 Hz for 15 min was analysed by DSC and  XRD. The DSC data ( Figure SII.21) show one clear endothermic event with an onset at 183.3 8C, after whicht he material begins to degrade. This endotherm occurs at the same temperaturea so bserved for the printed crystals (181.6 8C), and so again it appears that ac o-crystal has formed. XRD analysis ( Figure 13) resulted in ap attern very similar to that of the previously reported form Ic o-crystal, [23] albeit with an overall shift in reflection positions to lower angles. The intensity ratios are very similar, as are the relative peak positions. The similarity of the two patterns suggests that the structures of the two samples are the same. The shift in angles of diffraction can be attributed to the difference in temperature of the two samples. The FTIR spectrum recorded for the milled sample (Figure SII.22) is almost identicalt ot he spectrum recordedf or the printed crystals, which again supports the conclusion that a co-crystal has formed upon milling.

INCT-HPAA
The results obtained after millingm ixtures of INCT and HPAA in a2:1 molar ratio are consistent with the production of ac ocrystal.T he DSC data (Figure SII.23) display as ingle endotherm with onset at 126.3 8C, the same temperature as the printed crystals( 125.9 8C). The XRD pattern of the product of milling matches closely with that predicted for the 2:1c o-crystal (see Figure SII.24). TheF TIR spectra (Figure SII.25) are also very similar for the milled and printed samples, thus confirming that the same co-crystal has been generated in both cases.

INCT-DHPAA
Contrary to the previous examples, the data for the milled mixture of INCT and DHPAA are not identical to those from printing. There are distinct differences noted in the melting points in DSC (Figure SII.26) and the reflection positions and intensities in XRD ( Figure SII.27). The latter do not suggest as imple physicalm ixture of the two components, but rather that a second co-crystal system has been generated.Acomparison of the XRD data from the milled system and the phase observed at 122 8Ci nt he DSC-XRD experiment suggest that the two materials are the same, however ( Figure SII.28). The full details of this are not yet known, and to date we have been unable to produce single crystalso ft his material. Investigations are ongoing.

Discussion
This studyr eportsasystematic exploration of the formationo f co-crystals containing INCT and as eries of APIs. It is clear that the formation of co-crystals is ac omplicated and multi-faceted process. In the case of INCT-HBA, two polymorphic co-crystals can be obtained from solvente vaporation experiments. Both form Ia nd form II are observedd uring thermals ynthesis, with form II forming first and converting to form Iu pon continued heating. Ink jet printing and ball milling yieldedo nly form Io f the INCT-HBA co-crystal. In the case of INCT-HPAA, co-crystals with two different stoichiometries (2:1 and 1:2) formed after solvente vaporation. Inkjet printing and ball milling the two co-formers at a2:1 molar ratio resulted in the 2:1c o-crystal, as did at hermalt reatment. A1 :1 INCT-DHPAA co-crystal forms upon solvent evaporation and from ink jet printing, but milling and thermals ynthesis result in ad ifferent material, the structure of whichcould not be determined.
What is stark from these resultsi st hat different routes of cocrystallisation yield different products, andt hat this is highly system-specific. While it is possible to generate co-crystals via ink jet printing and milling approaches, and hence it can be concluded that these routesd oh ave the potential to be used for pharmaceutical manufacturing, the complex polymorphism which arises in co-crystallisationi sacomplicating factor,a nd great care will be required to ensure that the desired material is generated. Further,w hen developing medicalp roducts in the form of co-crystals, it will be necessary for pharmaceutical companies to employarange of synthetic processes during preformulation studies in order to ensure that the polymorphic landscapei sw ellu nderstood.C rucially,t he standard solvent screening method alone does not allow identification of all the phases possible for ac o-former system. DSC-XRD experiments reveal that ar ange of simultaneous processes occur during the thermals ynthesis of co-crystals. Rathert han simple solid-solid or melt-crystallisation transformations, we observe the different co-formers to undergo polymorphict ransitions themselves, and to melt at different points. It is also notable that simple mixing appears sufficient to cause as malla mount of co-crystal formation. These findings will be important to pharmaceutical manufacturers, given that the most likely route to generate co-crystals in industry is HME, a thermally-mediatedf abrication method.

Chemistry-A European Journal
Full Paper doi.org/10.1002/chem.202002267 duced co-crystals, but there were differences in the phases obtained from the differentm ethods. In the case of INCT-HBA, both co-crystal polymorphs wereo bserved during heating, but only the previously reported form Ir esulted from inkjet printing and milling. The 2:1c o-crystal of INCT-HPAA was observed from all synthetic routes, but with the INCT-DHPAAs ystem while the same co-crystal was seen from solvent-based approaches (evaporation and printing) ad ifferent materialw as obtainedt hermally and from milling. Overall, it is clear that a wide range of approaches need to be implemented in the development of pharmaceutical co-crystals to ensure that the polymorphic landscapei sf ully understood.

Experimental Section
Full detailso ft he experimental proceduresu sed in this work are given in the Supporting Information (Section III).