What happens to a drug's hydrogen bonds when its crystal lattice is lost: seven dihydropyridines compared by IR and Raman

Background and motivation

Much of what is known about how drug molecules associate comes from crystals. Single-crystal X-ray diffraction gives exact positions, and decades of structure analysis have produced packing rules: molecules pack so that free volume is minimised and density maximised (Kitagorodski's rule of close packing), the best hydrogen-bond donors tend to pair with the best acceptors, and electrostatic repulsions are minimised (Brock and Dunitz 1994, ref 1; Etter 1990, ref 2). C1 Amorphous solids have no long-range order, so diffraction says little about how their molecules sit relative to each other. Neutron scattering has been applied to amorphous glucose but is not widely available, while optical spectroscopy has long been used on glasses (refs 3, 4). C2

This matters for formulation because amorphous drugs, alone or in molecular mixtures such as lyophilized products and solid dispersions, are metastable and can crystallize. The authors expect intermolecular interactions to govern the properties of such phases, and earlier reports had already described drug–excipient interactions in binary amorphous systems (refs 5–9, among them the group's indomethacin–PVP and sugar–polymer studies, refs 6 and 7). C3 Their argument is that knowing how molecules associate in the amorphous phase, and how that differs from the crystal, "should provide" a better understanding of the tendency to crystallize and of how crystallization inhibitors work. C4 What was missing was a direct comparison of hydrogen bonding in single-component amorphous drugs with that in their own crystals.

The paper fills that gap with infrared (IR) and Raman spectroscopy on seven 1,4-dihydropyridine (DHP) calcium channel blockers: felodipine, nifedipine, nicardipine, nitrendipine, nimodipine, isradipine and nisoldipine (Figure 1). They share one core and differ in the aryl group and the ester side chains, which gives both chemical similarity and diversity; five of them have published single-crystal structures, so the crystal hydrogen bonds are known and can anchor the spectral reading. C5·P6·M1–M7

Background concepts

Reading a hydrogen bond from a vibrational band

A free X–H group (here N–H) gives a fairly sharp stretching band. When it donates a hydrogen bond, X–H···Y (Y is the acceptor atom), the X–H bond lengthens, the band moves to lower wavenumber and becomes much broader; a stronger bond lengthens X–H more and moves the band further (Lin-Vien et al. 1991, ref 11). C7 Crystal data show the same thing from the other side: a lower stretching frequency goes with a shorter X···Y distance (Nakamoto et al. 1955, ref 12). C7 The acceptor feels the bond too: a carbonyl that accepts a hydrogen bond moves to lower wavenumber. C7

The paper builds on two working assumptions. First, the NH band position reflects the average hydrogen-bond strength in the sample, and its width gives some indication of the distribution of hydrogen-bond energies (Skrovanek et al. 1985, ref 13). C8 Second, a carbonyl band that moves down when the NH band moves down marks that carbonyl as the acceptor. C11 The authors test the first assumption on their own crystals (Figure 2, below) and judge it "reasonably valid". C8·R4

Naming bond strengths

The authors use Jeffrey's classification (ref 14): a moderate hydrogen bond has an H···Y distance of about 1.5–2.2 Å and an X–H···Y distance of about 2.5–3.2 Å; a weak one has about 2.2–3.2 Å and about 3.2–4.0 Å. R5·C9 On this scale nifedipine's crystal contact (N···O 3.028 Å) is moderate, as the authors say. R6.d·C13 Note by the AI writer (not discussed in the paper): felodipine's second crystal contact, N···O 3.237 Å, falls in the weak range by the same classification. R6.b·X1

The seven compounds

Each compound has exactly one hydrogen-bond donor, the dihydropyridine NH, and several possible acceptors: the two ester carbonyls, ether-type oxygens, a nitro group in five of them, and the ring nitrogen. M1–M7 Table A lists what the structures in Figure 1 show and what the crystal data say about the hydrogen bond.

Table A. The seven dihydropyridines: structures from Figure 1, crystal hydrogen-bond data from Table I (published single-crystal structures, refs 15–19). Felodipine's angle is printed as "163°C" in Table I; the unit is degrees. M1–M7·R6·X3

Methods in detail

The study has three parts: prepare each drug as a crystal and as an amorphous solid, record the NH and carbonyl bands of both by IR and Raman, and read the bands against the crystal structures. The flow is summarised below.

Samples: crystals and their amorphous forms

Felodipine came from AstraZeneca; nifedipine, nitrendipine, nimodipine and nicardipine hydrochloride were bought from Göteborgs Termometerfabrik; isradipine and nisoldipine were gifts from Novartis and Bayer. M1–M7 Nicardipine was only available as its hydrochloride salt. It was dissolved in water and treated with excess NaOH; the free base, which is very poorly water soluble, separated as a film, was washed, dissolved in ethanol and recovered by evaporation. This gave an amorphous solid that slowly turned partially crystalline, and a fully crystalline nicardipine sample could not be made, so its "crystalline" spectrum is that of a partly crystalline sample, recorded by IR only. P1·R3

Amorphous forms were made by heating each crystalline drug to just above its melting point, holding for 1 min and quenching the container in cold water or on a metal surface. P2 No degradation was seen; for felodipine HPLC gave a purity above 99.9% after melting. R1 The drugs are very poorly water soluble, the samples were exposed to room air (about 40% relative humidity) for less than 5 min, and moisture uptake was not a problem over that time. R1 Samples were measured immediately: for Raman the drug was melted directly inside an NMR tube that was then sealed, and for IR the sample went straight into the nitrogen-purged instrument. P2 Note by the AI writer (not discussed in the paper): the paper does not say whether the fully amorphous nicardipine sample came from the ethanol route or from melt quenching, and it gives no measurement temperature. X7·X11

Spectroscopy

FT-Raman spectra were collected with a 1064 nm Nd:YAG laser (typically 500 mW), back-scattering at 180°, 256 scans at 2 cm⁻¹ resolution, with the sealed tubes rotating and indene as the wavenumber standard. P3 FT-IR spectra were transmission spectra of KBr discs holding 0.5–1% drug, at 2 cm⁻¹ resolution and 64–256 scans, with the instrument purged by dry nitrogen. P4 Because pressing a KBr disc can change a sample, every sample was also measured by diamond ATR at 4 cm⁻¹; the ATR spectra matched the transmission spectra, which the authors take as showing that no transformation occurred. P5·R2·C6

The crystal reference and the calibration plot

For the five drugs with published crystal structures (refs 15–19) the authors took the donor–acceptor distance N···O, and where hydrogen positions were reported also the H···O distance and the N–H···O angle (Table I). P6·R6 These data identify the acceptor in each crystal and were used to assign the crystalline IR and Raman bands. P6 To check that the NH band position can serve as a strength gauge, the crystalline NH position was plotted against the N···O distance (Figure 2). The points fall near a straight line, and the text and caption give R² = 0.974. R4·E1·C8

Note by the AI writer (not discussed in the paper): two details of Figure 2 cannot be confirmed from the printed numbers. Felodipine has two crystal contacts; the figure plots it at about 3.24 Å, the methyl ester contact (3.237 Å), although the text calls the ethyl ester contact (3.155 Å) the shorter one. R4·X1 And recomputing the regression from the printed distances and the IR crystalline band positions gives R² of about 0.944 (0.949 with the Raman positions, 0.906 if felodipine is placed at 3.155 Å), not 0.974; the printed value equals the correlation coefficient R of the Raman data. E1·X2 The trend itself is clear: the five points rise monotonically with distance. R4

Acceptor basicity

To explain which acceptor the NH prefers, the authors calculated relative basicities of the carbonyl, nitro and ether groups of nifedipine with the HYBOT program. P8·R16 They also mention, without showing data, that self-association of the compounds in a nonpolar solvent preferentially involves the carbonyl. P9·R15

Experimental analysis

Felodipine: the worked example

Figure 3 compares the full IR and Raman spectra of crystalline and amorphous felodipine from 3600 to 1000 cm⁻¹. Bands broaden in the amorphous form, most visibly in the Raman spectrum, and shapes, intensities and positions change. R7 Figure 4 zooms in on the two regions that carry the argument. R8

In crystalline felodipine the NH bonds both ester carbonyls, and both contacts are longer than in the other crystals (N···O 3.155 and 3.237 Å). R6.a-b·C10 The NH band sits at 3372 cm⁻¹ (IR) and 3371 cm⁻¹ (Raman), and there are two carbonyl bands, 1699 and 1690 cm⁻¹ by IR. The authors assign the lower one to the ethyl ester carbonyl in the shorter bond and the higher one to the weakly bonded methyl ester carbonyl. R11.a-b·C10

In amorphous felodipine the NH band is broader and lies about 33 cm⁻¹ lower (3339 cm⁻¹ by IR, 3335 cm⁻¹ by Raman). R9.a·R11.a-b At the same time one carbonyl band moves down by 8 cm⁻¹ (1690 to 1682 cm⁻¹, IR) while the other moves up by 2 cm⁻¹ (1699 to 1701 cm⁻¹). R9.a·R11.a The authors read this as follows: the downward NH shift "suggests" that the average hydrogen bond is stronger in the amorphous form than in the crystal; the concomitant downward carbonyl shift supports this and identifies that carbonyl as the acceptor; and because the other carbonyl moves up, only one carbonyl is hydrogen bonded in the amorphous form while the other is unassociated or very weakly bonded. C11 A small band near 3420 cm⁻¹ appears in both amorphous spectra; its position lies in the range reported for non-bonded NH, and the authors think it "is thought to arise" from a small fraction of NH groups without a partner, as seen in other amorphous materials. R10·C12

Compound by compound

Table B collects the NH and carbonyl band positions of all seven drugs from Table II. R11

Table B. NH and carbonyl stretching band positions (cm⁻¹) of the crystalline and amorphous forms (from Table II, footnotes paraphrased). R11

Nifedipine. In the crystal the NH forms a medium-strength bond to one methyl ester carbonyl; its N···O (3.028 Å) is shorter than felodipine's, which suggests a stronger bond if the geometry is similar, and the NH band is correspondingly lower (3328 cm⁻¹ by IR). R6.d·R11.c·C13 Two carbonyl bands fit one bonded and one free carbonyl. C13 On amorphisation the NH band moves up by 14 cm⁻¹ by IR, "indicating" that, unlike felodipine, the hydrogen bond weakens; both carbonyl bands move up, which the authors take as support for this "conjecture". R9.b·R11.c·C14 The amorphous NH and carbonyl positions are very close to those of amorphous felodipine, and there was "no evidence" that other acceptor groups take part (data not shown). C14·C15·R12 Note by the AI writer: by Raman the nifedipine NH band moves only 2 cm⁻¹ (3330 to 3332 cm⁻¹), which is the size of the Raman resolution, so the weakening rests on the IR spectrum. R11.d·X5

Nisoldipine and nitrendipine. Both crystals bond the NH to a carbonyl, like nifedipine, but with shorter contacts (N···O 2.894 and 2.892 Å) and correspondingly lower NH bands (3321 and 3318 cm⁻¹ by IR). R6.j·R6.l·R11.i·R11.k·C16 On amorphisation the NH band rises in both, "suggesting" weaker bonding, and the amorphous carbonyl positions match those of the other amorphous drugs, "implying" the same pattern and strength. R9.c·C16 The text gives no number for these rises; by Table II they are 20 cm⁻¹ by IR for both, and 14 (nisoldipine) and 8 cm⁻¹ (nitrendipine) by Raman. R11.i-l·X5

Nimodipine. This crystal is the odd one: its NH bonds the ether oxygen of the methoxyethyl ester, with the shortest N···O of all (2.835 Å), which suggests the bond "may be stronger" than in the other crystals. R6.h·C17 The NH band is indeed the lowest of the set, 3299 cm⁻¹ by both methods, and the carbonyl bands sit high (1700 and 1693 cm⁻¹ by IR), as expected when neither carbonyl is bonded. R11.g-h·C17 In the amorphous form the NH band rises by "around 35" cm⁻¹ while one carbonyl band drops by 14 cm⁻¹, and the C–O stretching band moves up by 12 cm⁻¹. R9.d "These results indicate" that the bond weakens and that the acceptor changes from the ether oxygen to a carbonyl. C18 Note by the AI writer: the Table II differences are +39 (IR) and +29 cm⁻¹ (Raman) for the NH band and −16 (IR) and −13 cm⁻¹ (Raman) for the carbonyl, so the text values are rounded or averaged. X6

Isradipine. No crystal structure was available, so the crystal pattern is inferred. Crystalline isradipine shows a single carbonyl band near 1700 cm⁻¹ (1703 cm⁻¹ by IR, 1698 by Raman), thought to contain both carbonyls; its high position means neither is hydrogen bonded. R11.e-f·C19 The authors therefore "speculate" that the NH bonds another group, "most likely" the benzoxadiazole ring, and that this bond is relatively weak because the NH band is high (3349 cm⁻¹). C19 In the amorphous form the NH band moves down (to 3341 cm⁻¹ by IR, 3335 by Raman), "implying" a stronger bond, and two carbonyl bands appear (1699 and 1680 cm⁻¹ by IR); the authors "propose" that one carbonyl becomes the acceptor, as in the other amorphous drugs. R9.e·R11.e-f·C20

Nicardipine. The partly crystalline free base has its NH band 15 cm⁻¹ above that of the amorphous sample (3354 against 3339 cm⁻¹ by IR), which the authors present as a further case where hydrogen bonding "appears to increase" on amorphisation; the amorphous NH and carbonyl positions again match the other amorphous drugs. R3·R9.f·R11.m·C21 The patterns for both forms are only suggested, from the spectra. R6.n-o·C21

The group picture

Figure 5 puts the IR NH positions of all seven drugs side by side, ordered by the crystalline value. The crystalline points climb from 3299 cm⁻¹ (nimodipine) to 3372 cm⁻¹ (felodipine), a spread the text gives as 70 cm⁻¹; even within the drugs whose crystals bond a carbonyl, the spread is about 50 cm⁻¹. R13·R11 The amorphous points form an almost flat row between 3338 and 3342 cm⁻¹, less than 5 cm⁻¹ apart, which the authors describe as a "very uniform" average bond strength. R13·C26 Note by the AI writer: Figure 5 and the "less than 5 cm⁻¹" statement use the IR values; the Raman amorphous values in Table II span 3326–3336 cm⁻¹, a 10 cm⁻¹ spread, still far narrower than the crystalline spread of 72 cm⁻¹ by Raman. R11·X4

The carbonyl region tells the same story. Every amorphous form shows two carbonyl bands, one at 1696–1703 cm⁻¹ and one at 1677–1684 cm⁻¹ (IR), and the authors read the lower one as the single hydrogen-bonded carbonyl and the higher one as free. R11·C27 In every case the amorphous NH band is also broader than the crystalline one, which they take as consistent with a broader distribution of hydrogen-bond lengths in the disordered phase; the widths were not measured as numbers. R14·C29

The direction of change depends on the drug. The NH band falls on amorphisation for felodipine, isradipine and nicardipine, read as a stronger average bond, and rises for nifedipine, nisoldipine, nitrendipine and nimodipine, read as a weaker one. R11·C32 The authors call the strengthening "perhaps somewhat surprising"; in the abstract the weakening of the other four is "as anticipated". C32 Seen in Figure 5, the dividing line is simply where each crystal sits relative to the common amorphous value near 3340 cm⁻¹: the three crystals above it gain strength, the four below it lose strength. R13

Table C. Crystal against amorphous for each drug: acceptors from Table I (amorphous acceptors deduced from the spectra), NH positions from Table II (cm⁻¹), and the authors' reading of the change. R6·R11·C18·C20·C32

Discussion

Why do the crystals differ so much?

The authors start from the observation that the chemistry of the donor and acceptor groups, which they call "undoubtedly" the largest influence on hydrogen-bond strength, is nearly the same across the seven drugs and therefore cannot explain a 70 cm⁻¹ spread among the crystals. C25·R13 They cite work showing that most variation of hydrogen-bond distances in organic crystals comes from crystal packing (R. Taylor et al. 1984, ref 22), and the view that maximising density is the primary packing rule (Brock and Dunitz 1994, ref 1). From this, "we can thus speculate" that in some crystal structures the chance to form the best hydrogen bond has been sacrificed for other packing factors. C25 The summary words the same point more firmly ("it was concluded"). C25 The amorphous data are their supporting evidence: once packing constraints are removed, the same chemistry gives very similar patterns and strengths in all seven drugs ("These observations suggest"). C28

Why the carbonyl?

In the amorphous forms the authors found evidence only for carbonyl acceptors; the nitro, ether and ring-nitrogen acceptors "appear to be largely ignored", and self-association in a nonpolar solvent also favoured the carbonyl (data not shown). C30·R15 They rationalise the choice with the HYBOT basicities calculated for nifedipine: 1.2 for the carbonyl, 0.48 for the nitro group and 0.12 for the ether group, so the carbonyl is more than twice as basic as the nitro group, and "this suggests" that the NH reaches its lowest energy state by bonding a carbonyl. R16·C31 Nimodipine's crystal, where the NH bonds an ether, is called unusual for the same reason: carbonyls are usually preferred over ethers (Murray-Rust and Glusker 1983, ref 21). C24

Note by the AI writer (not discussed in the paper): nifedipine has no ether oxygen other than the single-bonded oxygens of its esters, so its "ether" basicity does not describe the methoxyethyl ether of nimodipine, and no value is given for isradipine's benzoxadiazole. The basicity argument is therefore a one-compound calculation applied to all seven. X9

Stronger or weaker on amorphisation

The authors' chain of reasoning can be drawn as follows; the two outcomes are different cases, not better and worse.

For the general case they point to the only comparable study they knew of, a neutron scattering study in which the average hydrogen-bond length of amorphous glucose was longer than in the crystal (Tromp et al. 1997, ref 3), and "speculate" that weakening on disordering "is probably more typical than the converse". C33·R17

What it could mean for crystallization and stabilisation

The authors then extend the argument, clearly marked as speculation. "It can be speculated" that different energy barriers to rearrangement exist in the amorphous phase before crystallization, depending on how different the association is in the two phases, although "clearly other factors such as molecular mobility" must be considered. C34 Amorphous felodipine must disrupt or weaken some hydrogen bonds to crystallize; amorphous nimodipine must break its carbonyl bonds and re-form them with the ether; the others need little rearrangement of hydrogen bonding. C34 Because crystallization needs a particular association, knowing how to disrupt the amorphous hydrogen-bond pattern "would presumably be useful" for physical stabilisation, for example by targeting the drug's donor or acceptor groups with polymeric crystallization inhibitors. C35 They see studies of molecular association as complementary to studies of molecular mobility. C36

Note by the AI writer (not discussed in the paper): three practical points follow for anyone using IR to look for drug–polymer hydrogen bonds with these drugs. First, the reference spectrum should be the amorphous drug, not the crystal: for felodipine the crystalline NH band lies 33 cm⁻¹ above the amorphous one, so a shift measured from the crystal would overstate any polymer effect. R11.a Second, the crystal structure does not tell you what the amorphous NH binds, as nimodipine shows. R6.h-i Third, in these amorphous drugs the NH is already bonded to a drug carbonyl, so a polymer acceptor must compete with that drug–drug bond rather than bind a free NH. R6·C27

Limitations

Limits the authors state or imply

The strength reading rests on an explicit assumption: that the NH band position reflects the average hydrogen-bond strength, which the crystal correlation makes "reasonably valid". C8 The crystal acceptor of isradipine is speculation ("we can thus speculate", "most likely") and the patterns of nicardipine are only "suggested", since neither has a crystal structure. C19·C21 The links to crystallization are offered as speculation, with molecular mobility named as another factor that "clearly" has to be considered. C34 Several supporting observations are described as "data not shown": the absence of other acceptors in amorphous nifedipine and the carbonyl preference in nonpolar-solvent self-association. R12·R15

The AI writer's own judgement

The core observation is solid: every band position is tabulated for both forms, the IR and Raman spectra move the NH band in the same direction for each of the six drugs measured by both, and the amorphous NH bands of all seven drugs lie within 4 cm⁻¹ by IR, against a crystalline range of more than 70 cm⁻¹. R11·R13 The interpretation is weaker in the following places.

• The stated R² of 0.974 for the calibration plot recomputes to about 0.94–0.95 from the printed values, and felodipine is plotted at its longer contact without comment; the trend holds, the number does not. X2·X1
• "Very uniform" is an IR statement: by Raman the amorphous NH bands spread 10 cm⁻¹. X4
• Nifedipine's weakening is +14 cm⁻¹ by IR but +2 cm⁻¹ by Raman, at the resolution of the Raman spectrum. X5
• "Stronger" and "weaker" are inferences from band position, calibrated on five crystals; for isradipine and nicardipine there is no crystal structure, and nicardipine's reference sample is only partly crystalline. X8·X7
• The carbonyl-only pattern of the amorphous forms is inferred from carbonyl bands; no band of another acceptor is shown for any compound. X10
• The basicity argument uses one calculation, for nifedipine. X9
• The measurement temperature is not given, and a cooling experiment mentioned in a footnote to Table II is not described. X11
• No crystallization or drug–polymer data were collected, so the formulation consequences remain untested here. C34·C35

Conclusion

For seven dihydropyridine calcium channel blockers, IR and Raman spectra show that the crystals differ widely in which group the single NH binds and in how strongly, while the melt-quenched amorphous forms converge on one pattern, NH to an ester carbonyl, with nearly the same average band position. R11·R13·C37 The acceptor in an amorphous drug is "not necessarily the same" as in its crystal, as nimodipine shows by switching from an ether to a carbonyl, and amorphisation can strengthen the average hydrogen bond (felodipine, isradipine, nicardipine) as well as weaken it (the other four). C37·C18·C32 The authors attribute the crystalline diversity to packing and the amorphous uniformity to shared chemistry and acceptor basicity, and they suggest, without testing it, that the rearrangement needed to go from the amorphous to the crystalline pattern sets different barriers to crystallization. C25·C28·C31·C34 The AI writer's reading for formulation (not stated in the paper): the hydrogen bonds of an amorphous drug have to be measured in the amorphous drug itself, because its crystal is not a reliable guide. R6.h-i·R11