Do the hydrogen bonds a molecule forms in solution decide its polymorph? An IR and Raman test with indomethacin and anthranilic acid

Background and motivation

Many drugs can crystallize in more than one structure, and which polymorph appears first decides solubility, stability and processing. Before a crystal can form, solute molecules must come together in a supersaturated solution, and the authors start from the idea that this aggregation or clustering is a prime requirement for nucleation (Introduction). C1 Their motivation is simple to state: if the structure of these prenucleation aggregates could be identified and related to the structure of the final solid, the outcome of crystallization, including the polymorph, "might" be controlled. C2

The evidence for clusters in supersaturated solutions was largely indirect in 2006. Mullin and Leci, Garside and Myerson had inferred clusters, and the number of molecules in them, from concentration-dependent solution properties, while Narayanan found no change in solution properties on moving into supersaturation (refs 3–10). C1 Direct structural studies were few. Raman spectroscopy had been applied to supersaturated inorganic solutions (refs 11–15), but the authors knew of only one organic case in which vibrational spectroscopy showed a motif common to the solution and the crystal: dimers in some supersaturated solutions of tetrolic acid matched the dimers of the polymorph that then crystallized (Parveen et al. 2005, ref 16). For another molecule, concentration-dependent 1H NMR shifts were read as aggregation into motifs resembling the solid dimers (ref 17). C3

Views on what a cluster looks like also differed (ref 18). Clusters might hold much solvent and have no long range order; in polymorphic systems there might be clusters with short range order resembling each possible polymorph (Etter, ref 19); recent work suggested a two-step route in which the solute first forms a dense liquid-like phase and orders later (refs 20–22); and simulations of acetic acid suggested that this precursor liquid need not resemble the final crystal at all (Gavezzotti, ref 23). C4 Solvent adds a further layer: solvents chosen to disrupt or promote particular hydrogen-bonding networks had been used to engineer new crystal forms (refs 16, 24–26). C5

The paper therefore asks a focused question. Using IR and Raman spectroscopy on solutions of two polymorphic compounds, indomethacin and anthranilic acid, do any of the interactions seen in solution match those of the crystalline polymorphs, and in particular does a specific solute–solute interaction lead to a polymorph that contains the same interaction (Introduction)? A1 The authors set up explicit expectations: for indomethacin, solution spectra over a range of concentrations "might be expected" to show carbonyl-region differences connected to the final form; for anthranilic acid, since form II crystallizes first from most solvents, they "expect" solutions in a range of solvents to share spectral features of form II. C6·C8

Background concepts

The two polymorphic acids

Indomethacin (IMC) has many reported polymorphs, of which α and γ are the most common (ref 28). The γ form is the more stable; α turns into γ in a slurry. M1 Which one crystallizes depends on concentration: α comes from more concentrated solutions, γ from somewhat less concentrated ones, and in ethanol α was reported above a degree of supersaturation of 4 and γ below about 3.5 (Slavin et al., ref 29). R1 The two structures differ in exactly the feature a spectroscopist can see (Figure 1). γ consists only of discrete cyclic carboxylic acid dimers, with one molecular conformation. α has three molecules in its asymmetric unit, each in a different conformation: two form a carboxylic acid dimer, and the third forms a hydrogen bond from its carboxylic acid to an amide (benzoyl) carbonyl, linking the molecules into chains. M1

Anthranilic acid (2-aminobenzoic acid) has three known polymorphs (ref 31). Form I, stable at low temperature, has two molecules in its asymmetric unit, a zwitterion and a neutral molecule; forms II (metastable) and III (stable at high temperature) contain one neutral molecule each, joined into carboxylic acid dimers (Figure 3). M2 Form II crystallizes from most solvents below 50 °C and later transforms to form I; form III is obtained from the melt or from solutions above 50 °C (ref 33, Towler's thesis). R2

What the bands report

A carbonyl (C=O) stretch moves to lower wavenumber when the carbonyl accepts a hydrogen bond, so its position tells whether the group is free, bonded to another solute molecule, or bonded to solvent. A carboxylic acid dimer, two COOH groups held by two hydrogen bonds in a ring, has an asymmetric C=O stretch that is IR active and a symmetric stretch that is Raman active. The paper relies on published solid-state assignments for indomethacin (Table 1, from Taylor and Zografi 1997, ref 30), summarised in Table A. R3

Table A. Vibrational assignments for solid indomethacin, transcribed from Table 1 of the paper (cited from ref 30). The printed table carries no form label on its first five rows; they belong to α by the caption order and by the text, which places the α band at 1650 cm^-1. R3·X4

The key marker is the α-only band near 1648–1650 cm^-1: it comes from the amide carbonyl that accepts a hydrogen bond from the carboxylic acid, the interaction that γ lacks. R3.e·C29 Figure 2 shows the solid-state carbonyl IR of the two forms; the caption explains the extra peaks of α as the result of its larger number of carbonyl interactions. R4·C7

For anthranilic acid the carbonyl region is crowded by intra- and intermolecular hydrogen bonds, in the solids and in many solutions, so the authors use the NH₂ symmetric and asymmetric stretches instead, which "proved most useful" to tell the forms apart (Figure 4). C12 Read from Figure 4 against its tick marks, form III shows its NH₂ doublet at about 3495 and 3381 cm^-1, form II at about 3474 and 3373 cm^-1, and form I, with its zwitterion, at about 3324 and 3238 cm^-1, with broader bands. R5

Supersaturation

The degree of supersaturation used for indomethacin is the ratio of the concentration used to the equilibrium solubility (Experimental Section):

In plain terms: c is the concentration of the solution and c^* the equilibrium solubility in the same solvent; a value of 1 is a saturated solution, values above 1 are supersaturated. E1 The solubilities given at 25 °C are 25 mg/mL in ethanol and 11 mg/mL in nitromethane, and solutions were made up to a supersaturation of 5.5. M1·P1 For anthranilic acid only limited solubility data were available, so no supersaturations are given. M2·P3

Methods in detail

The study has no models or simulations. It compares solution spectra with the spectra and assignments of the solid forms, and with the polymorph that crystallizes. The flow below summarises it.

Indomethacin: one solute, two kinds of solvent

Problem. If solution clusters carry the motif of the coming crystal, a series of increasingly concentrated solutions should show the motif growing, and a solvent that changes the available hydrogen bonds might change it. What they did. Indomethacin (99%, Sigma-Aldrich) was dissolved in ethanol and in nitromethane over a range of concentrations up to a supersaturation of 5.5. M1·P1 The two solvents differ in one property that matters here: ethanol both donates and accepts hydrogen bonds, nitromethane only accepts them. M3·M4·A2 Both were chosen because neither had formed indomethacin solvates in the authors' experience. C9 Figure 8 shows eight IR spectra in ethanol and Figure 9 six in nitromethane, but the concentrations of the individual spectra, and the temperature at which they were recorded, are not given. P1·X6 What is followed. The two carbonyl groups (acid and amide), the dimer bands near 1711–1714 cm^-1, the carboxyl OH deformation near 920 cm^-1, and, in Raman, the α marker near 1650 cm^-1. R11·R16·R18

Anthranilic acid: one solute, many solvents

Problem. If the polymorph that forms first (form II) is prefigured in solution, its features should be common to every solvent that gives it. What they did. Anthranilic acid (≥98%, Sigma-Aldrich) was dissolved in acetonitrile (150 mg/mL), DMSO (175 mg/mL), ethanol (200 mg/mL), toluene (25 mg/mL) and methanol (25 mg/mL), at levels that usually lead to form II; acetonitrile was also measured at 52 mg/mL, and toluene at further concentrations and temperatures that are not listed. M2·P2 Water was tried in search of the zwitterion of form I, but gave no solute signal, which the authors put down to the low solubility of anthranilic acid in water. R7·C11 Because solubility data were limited, the solutions were brought to saturation by a simple procedure: the solvent was heated to the crystallization temperature, solid was added with stirring until some remained undissolved, and the temperature was then raised to dissolve the excess, so the solutions were at least saturated at the crystallization temperature. P3·C10 The crystallization temperatures themselves are not stated. P3

Note by the AI writer (not discussed in the paper): the methods list methanol as the fifth solvent, but the caption of Figure 5 names benzyl alcohol instead, and no methanol spectrum is shown; the benzyl alcohol concentration is not given. X2

Spectroscopy, and what is not reported

Raman spectra were recorded on a Kaiser RamanRxn1 with a 150 mW external cavity stabilized diode laser at 785 nm, integrating 5 s and summing 10 scans per spectrum. IR spectra were recorded on a Nicolet Nexus 670 FTIR with a multi-bounce horizontal ATR accessory (ZnSe crystal), 64 scans at 4 cm^-1 resolution. P4·P5 The paper does not say whether solvent spectra were subtracted, how the solid reference forms were prepared, or how the polymorph that crystallized from each solution was identified; the outcomes (form II; α) are stated in the Results without a described procedure. P6·P7·R19·X5

Experimental analysis

Anthranilic acid: five solvents, five spectra, one polymorph

The first observation is negative. The Raman spectra of anthranilic acid in the five solvents of Figure 5 (top to bottom: acetonitrile, benzyl alcohol, DMSO, ethanol, toluene) differ from one another in band positions and relative intensities, and the authors found no feature common to them. R6 Yet form II was the form crystallized from every solvent. R19 The authors conclude that solute in solution and the final solid share no structural features here, and that solute–solvent interactions "appear to dominate" the spectra. C13 Despite clearly different solute–solvent interactions, there was no solvent-directed polymorphism. C14

The authors then bring in crystal shape. Form II grows as flat plates from ethanol, and needles and prisms have also been reported (refs 32, 34). R20 For anthranilic acid, they write, solvent "seems to have a marked effect on crystal growth" but "little effect on nucleation". C15 This is an inference from cited morphology; no morphology or nucleation data were collected in this paper. C15

Note by the AI writer (not discussed in the paper): whether solvent bands were subtracted is not stated, and Figure 5 compares whole spectra between 1750 and 1310 cm^-1 of solutions in five different solvents, so part of the differences may come from the solvents themselves. X7

Toluene: a solution that looks like form III

Toluene, a nonpolar solvent, should favour neutral molecules (ref 35), and indeed no bands of the zwitterion found in form I appeared. R8·C16 The NH₂ region told a different story from the one expected. At every concentration and temperature measured that leads to form II, the NH₂ doublet sat at the higher-wavenumber position of form III. R9 Table B gives the positions read from Figures 4 and 6.

Table B. NH₂ stretching bands of anthranilic acid, read from Figures 4 and 6 against their tick marks (approximate; the tick labels in these figures are printed offset from the tick marks). R5·R9·X11

The toluene doublet coincides with form III to within a few cm^-1, while form II lies about 10 to 23 cm^-1 lower, more than the 4 cm^-1 resolution. R9·P5 The authors' reading is that IR "indicates a predominance of clusters in solution that do not correspond to the polymorph initially crystallized"; in the Conclusions they describe the toluene spectrum as the one "most closely resembling" form III, although form II crystallizes. C17

Note by the AI writer (not discussed in the paper): the doublet stayed at the same place at every concentration and temperature tried. A band position that does not change with concentration fits solvated single molecules, whose NH₂ environment may resemble that of form III, as well as clusters; no band positions against concentration are shown. X9

Acetonitrile: a dimer band that matches form II

Acetonitrile gave the one positive match. In Figure 7, a carbonyl band that the authors place at 1670 cm^-1 and assign to the carboxylic acid dimer is a shoulder in the 52 mg/mL solution and a distinct peak at 150 mg/mL, and it corresponds to a band of solid form II. R10·C18 On the plot, which has no tick marks, this peak reads about 1678 cm^-1 against the label centres, in both the 150 mg/mL solution and solid form II. R10·X11 Both solutions also carry a band near 1703 cm^-1 that the solid lacks, and all three traces share bands near 1625, 1600 and 1575 cm^-1, which the paper does not discuss. R10 In the Conclusions, this dimer is the solution feature that "reflects" the dimer of form II, the form that crystallizes. C18

Note by the AI writer (not discussed in the paper): Figure 3 shows the carboxylic acid dimer in form III as well as in form II, and no carbonyl-region spectrum of form III is shown, so this band identifies dimers rather than form II specifically. X8

Indomethacin in ethanol: from solvent-bonded carbonyls to dimers

In dilute ethanol, IR and Raman show a single broad band near 1660 cm^-1, at the same position in both techniques, which the authors assign to both carbonyl groups hydrogen bonded to solvent; its breadth resembles the band of amorphous indomethacin. R12·C19 As concentration rises (Figure 8), two changes appear: a shoulder grows at 1711 cm^-1, the asymmetric stretch of a carboxylic acid dimer, and the main band moves from 1660 to 1680 cm^-1. R13·C20 The symmetric dimer stretch is not seen in the Raman spectrum, as is also true of γ. R14·C20 Read from Figure 8, the maximum lies near 1660 cm^-1 in the lowest and near 1684 cm^-1 in the highest spectrum, and bands near 1612 and 1595 cm^-1 also grow with concentration without comment in the text. R13

The authors explain the shift as carbonyls that were hydrogen bonded to solvent becoming free. "We can link this", they write, to indomethacin molecules associating into solute-rich clusters that leave less room for solvent and exclude it. C21 In general terms, the spectra and therefore the clusters "appear" to be changing to a more γ-like nature, built on dimers; despite this, α is the polymorph initially crystallized. C22

Indomethacin in nitromethane: free carbonyls, then dimers

Nitromethane can only accept hydrogen bonds, and the spectra differ accordingly. At low concentration two carbonyl bands appear, at 1685 cm^-1 (benzoyl, that is the amide carbonyl) and 1740 cm^-1 (acid), neither hydrogen bonded; as concentration rises, a band appears at 1713 cm^-1, assigned to carboxylic acid dimers (Figure 9). R15·C23 Table C lists the solution assignments of Table 2 next to the solid-state bands.

Table C. IR bands of indomethacin solutions (Table 2 of the paper) with the nearest solid-state bands from Table 1; the last column is an AI comparison of printed values, not a table in the paper. The text gives 1711, about 1660, 1740, 1713 and 1685 cm^-1 for the same bands, within the 4 cm^-1 resolution. R3·R11·X3

In both solvents a band near 920 cm^-1, the OH deformation of the carboxyl group and typical of dimers (ref 36), grows with concentration; these spectra are described but not shown. R16·C24 The authors summarise that below a certain concentration the solute "appear[s]" to interact mainly with the solvent, and that solute–solute interactions and dimers increase with concentration. Dimer evidence is present in undersaturated as well as supersaturated solutions, which they expect, since the dimer is the stable configuration of carboxylic acids with bulky substituents (ref 37) and exists even in dilute solution (ref 36). R17·C25 The dimer is the motif found in both the α and γ structures. C26

Note by the AI writer (not discussed in the paper): the solution dimer bands at 1711–1714 cm^-1 sit closer to the amorphous dimer band (1710) than to γ (1718), and the high-concentration amide band at 1680 matches α (1680) and amorphous (1684) better than γ (1691). "γ-like" is therefore best read as "dimer-based, without the α acid–amide band" rather than as a match to γ band positions. X12

The missing α band

Figure 10 is the paper's most direct test. It sets the carbonyl-region Raman spectra of indomethacin in ethanol and in nitromethane, at supersaturations where α crystallizes, against solid α. R18 Solid α shows bands at about 1688, 1679 and 1648 cm^-1, matching Table 1 (1689, 1679, 1649). R18·R3 The ethanol solution has one broad maximum near 1681 cm^-1 that tails toward lower wavenumber, and the nitromethane solution a maximum near 1682 cm^-1; neither has a band at 1650 cm^-1. R18 "The inference is, therefore", the authors write, that an interaction present in the solid form crystallized at these supersaturations is not present in the solution before crystallization; the Conclusions repeat that no evidence was observed for the acid–amide interaction even at concentrations from which α crystallizes. C29 The supersaturations of these two solutions are not given. X6

How, then, does α form? The authors offer two suggestions without testing them. In nitromethane many amide carbonyl groups stay free, which "might facilitate the nucleation of the α polymorph"; in ethanol the remaining carbonyls appear hydrogen bonded to solvent. C27 In ethanol, the dimer-based clusters form at the exclusion of solvent and free some carbonyls, and "it seems" that these few free carbonyls are sufficient to provide a pathway to α. C28 Finally, because α crystallized from both solvents although their hydrogen-bonding possibilities are very different, they state that the polymorphic outcome of indomethacin is not influenced by the solvent, matching the observation of Slavin et al. that the morphology of γ indomethacin is independent of the crystallization solvent (ref 29). C30·R21

Table D. Each comparison the paper makes, with the authors' reading. The crystallized forms are stated in the paper without a described identification method. R6·R9·R10·R18·R19·C13·C14·C17·C18·C22·C27·C29·X5

Discussion

The authors' argument can be laid out as a chain that splits at the end. Clustering is thought to come before nucleation, and if a cluster motif matched one polymorph it could steer the outcome. C1·C2 At low concentration the solute bonds mainly to the solvent; as concentration rises, solute–solute dimers form and push solvent out of the clusters. C19·C21·C25 Whether the motif seen in solution then matches the polymorph that crystallizes differs from case to case.

In toluene and in both indomethacin solvents the solution motif does not match the crystal, in acetonitrile it does, and the authors draw a cautious general conclusion. C17·C18·C22·C29 Vibrational spectroscopy is "potentially useful" for identifying molecular aggregates in solution, but solvent–solute interactions "may dominate" the spectra and complicate the characterisation of solute clusters. C31·C32 Where solute–solute interactions can be identified, the motifs in solution "are not necessarily indicative" of the polymorph crystallized first, which they note agrees with other recent work (Davey et al. 2006, ref 38). C33 The abstract puts it for the systems studied: solute–solute interactions could be detected, but there was "generally a limited relationship" with the solid forms, so these assemblies were not predictive of the polymorph crystallized first. C35 The results, they add, "highlight the complex nature of the molecular recognition events" and time scales involved in nucleation. C34

Two strands of the discussion deserve separate attention. First, solvent. For anthranilic acid the solvent visibly changes the solution spectra and, from the literature, the crystal habit of form II, but not the polymorph; the authors conclude that solvent "seems" to act on growth rather than on nucleation. C14·C15·R20 For indomethacin the polymorph was α in both a donor–acceptor and an acceptor-only solvent, and they state that the outcome is not influenced by the type of interaction the solvent allows. C30 Second, mechanism. The authors suggest how α can still form from solutions without its marker interaction: the few carbonyls freed inside dimer-based clusters in ethanol "seem" sufficient, and the free amide carbonyls in nitromethane "might facilitate" α nucleation. C28·C27 Neither suggestion is tested; no measure of the free carbonyl population or of nucleation rate is reported. C27·C28

Note by the AI writer (not discussed in the paper): for a formulation scientist the paper's practical message is a caution. In this work the polymorph that formed followed rules already known from the literature, α above a supersaturation of 4 in ethanol and form II below 50 °C, rather than the motifs seen in solution. R1·R2·R19 Indomethacin, a common model drug for amorphous formulations, forms carboxylic acid dimers in solution even below saturation, so dimer bands in a supersaturated indomethacin solution do not by themselves show that γ-like order is building, and the absence of the α band from a bulk spectrum does not rule α out. R17·R18·X10·X12 Solution IR or Raman is therefore context for a crystallization or solvent screen, not a predictor of the form.

Limitations

The authors state several limits themselves. Solvent–solute interactions "may dominate" the solution spectra and complicate the characterisation of solute clusters. C32 For anthranilic acid, solubility data were limited, so supersaturations "could not be accurately calculated" and the solutions are described only as at least saturated at the crystallization temperature. P3·C10 In water no solute signal could be recorded because of the low solubility. C11 And the motifs seen in solution are "not necessarily indicative" of the polymorph, because nucleation involves complex recognition events and time scales that these measurements do not resolve. C33·C34

The AI writer's own judgement (1): what was and was not documented. The polymorph that crystallized, form II from every anthranilic acid solution and α from both indomethacin solvents, is stated in the Results without any crystallization procedure or solid-form identification, and it is the outcome the cited literature predicts for these conditions. X5·R1·R2 The concentrations, supersaturations and measurement temperature of the individual spectra in Figures 8–10 are not given, nor are the further toluene conditions, so the statements about undersaturated solutions and about "supersaturations where α crystallizes" cannot be checked against numbers. X6 Solvent subtraction is not stated, which matters most for Figure 5, where whole spectra in five different solvents are compared. X7 The comparison set is small: two compounds, and for indomethacin only conditions that give the metastable α form, so the paper cannot show whether solutions that give γ would look different. X5

The AI writer's own judgement (2): how far each comparison reaches. The absence of the α band in Figure 10 is the clearest result, but a bulk Raman spectrum reports the average environment of all solute molecules; a small population of α-like clusters below detection is not excluded, so the inference is best read as "not detected in the bulk solution". X10 The toluene NH₂ doublet did not move with concentration, which fits solvated single molecules as well as clusters. X9 The acetonitrile match does not distinguish form II from form III, since both contain the acid dimer and no form III carbonyl spectrum is shown. X8 The "γ-like" reading of the ethanol spectra is loose, because the solution dimer band (1711–1712 cm^-1) lies nearer the amorphous (1710) than the γ (1718) dimer position. X12

The AI writer's own judgement (3): small inconsistencies in the source. In Figure 2, the trace drawn as the upper dashed line and captioned α has two carbonyl maxima at about 1717 and 1690 cm^-1, the γ pattern of Table 1, while the lower solid trace captioned γ has the four α bands at about 1734, 1691, 1679 and 1649 cm^-1; the caption's own remark that α shows extra peaks also fits the solid trace, so the line labels appear swapped. X1·R4 The methods list methanol as a solvent but Figure 5 names benzyl alcohol. X2 Table 2 and the text differ by up to about 4 cm^-1 for the same bands, within the stated resolution. X3 Figures 2, 4 and 6 print their tick labels offset from the tick marks and Figures 5, 7, 8 and 9 have no tick marks, so plot readings are uncertain by several cm^-1; read against its label centres, the Figure 7 dimer peak lies at about 1678 rather than the 1670 cm^-1 of the text. X11 None of these changes the paper's conclusions.

Conclusion

This short paper tested a natural idea: that the hydrogen-bonding motif a molecule adopts in a concentrated solution foreshadows the polymorph it will crystallize as. With IR and Raman spectroscopy the authors could see solute–solute interactions directly: indomethacin forms carboxylic acid dimers in ethanol and nitromethane as its concentration rises, even below saturation, and anthranilic acid in acetonitrile shows a dimer band that grows with concentration and matches form II. R13·R15·R16·R17·R10·C25·C18 But the motifs did not predict the outcome. Toluene solutions of anthranilic acid carried the NH₂ signature of form III while form II crystallized; five solvents gave five different Raman spectra and the same polymorph; and indomethacin solutions at α-producing supersaturations lacked the acid–amide hydrogen bond that defines α. R9·R6·R18·R19·C17·C13·C29 The authors conclude that solvent–solute interactions may dominate solution spectra and that solution motifs are "not necessarily indicative" of the polymorph crystallized first. C32·C33 Because the crystallized forms and the solution concentrations are not documented in detail, and only two compounds were examined, the result is best taken as a well-argued counterexample to a simple structural link rather than as a general rule. X5·X6