Why some polymers slow crystal growth from amorphous felodipine more than others: hydrogen bonding, not Tg

Background and motivation

Many new drug candidates dissolve very poorly in water. Making them amorphous helps: the amorphous form is a high-energy state, and dissolving it can give a temporary rise in solution concentration and more drug absorbed in vivo (Law et al., ref 7). The catch is that the amorphous drug must not crystallize to a stable form during the shelf life of the product C1. One common way to prevent this is to add a polymer as a crystallization inhibitor. It is "generally accepted" that the polymer must be mixed with the drug at the molecular level, in one amorphous phase, which is what an amorphous solid dispersion (ASD) is; polymers that crystallize readily are therefore poor inhibitors C2.

Why a polymer slows crystallization is less clear. Several factors had been proposed: specific interactions between drug and polymer such as hydrogen bonds or ion–dipole contacts, the glass transition temperature (Tg) of the polymer, the Tg of the dispersion, and molecular relaxation times measured by calorimetry C29. Several authors had postulated a role for specific interactions, but, in the authors' words, there were "few systematic studies" linking the chemistry of the polymer with crystallization inhibition C3. In practice, polymers for amorphous formulations were chosen largely by trial C29.

The paper therefore asks a narrow, well-posed question. Take one drug, felodipine, and several polymers. Add so little polymer that Tg and the melting point of the drug do not change. Then measure how fast felodipine crystals grow with each polymer, and measure, separately, how strongly each polymer hydrogen-bonds to the drug. If the growth rate follows the hydrogen-bond strength while the bulk properties stay the same, specific interactions are the likely key factor C3·C4. The four polymers were chosen with this in mind: poly(vinylpyrrolidone) (PVP K-12), its copolymer with vinyl acetate (PVP/VA, Kollidon VA64, described as containing 40% acetate side chains), hydroxypropyl methylcellulose acetate succinate (HPMCAS) and poly(vinyl acetate) (PVAc). The first three have broadly similar Tg but carry different chemical groups, so they offer different possibilities for hydrogen bonding with felodipine C4·M2·M3·M4. PVAc carries only acetate groups and has a Tg below that of felodipine M5.

Two lines of earlier work frame the study. From pharmaceutics, Tang, Pikal and Taylor (ref 18) had shown how the NH group of felodipine hydrogen-bonds to carbonyl groups and how its infrared band reports the strength of that bond C18·R15. From polymer science, studies of crystallizable polymers blended with amorphous polymers had linked strong hydrogen bonding between the components with slower spherulite growth, but at additive levels high enough to change Tg and Tm as well (Talibuddin et al., ref 40; Kuo et al., ref 43) C39·R17·R18. This paper tests the same idea in a drug at additive levels so low that the bulk properties do not change C28·C41.

Background concepts

Crystal growth from the supercooled liquid. Above its Tg, amorphous felodipine is a supercooled liquid: it flows, and crystals that have nucleated in it can grow. Crystallization has two stages, nucleation (the birth of a crystal) and growth; only growth produces enough crystalline material to be detected C27. Here the crystals grow as spherulites, roughly round crystals made of fibres radiating from a centre, so their size can be followed under a microscope and the growth rate read as the slope of size against time R2·P4. The authors use the classical expression for growth from a supercooled liquid (eq 1, from Gutzow, ref 36) E1·C30:

Here $u$ is the crystal growth rate, $k$ a temperature-independent constant, $\eta$ the viscosity of the liquid, $\Delta G$ the free-energy difference between liquid and crystal, $R$ the gas constant and $T$ the temperature E1. In plain terms: growth speeds up when the driving force $\Delta G$ is larger (the term in brackets) and slows down when the liquid is more viscous (the factor $1/\eta$). On cooling, the driving force grows but the viscosity rises steeply, so the growth rate passes through a maximum at an intermediate temperature. The authors add that growth also depends on "molecular recognition events" at the crystal surface, which the equation does not contain C30.

How an additive can change growth, and how to tell. Equation 1 suggests two bulk routes by which a polymer could slow growth: by lowering the driving force or by lowering molecular mobility. Wu and co-workers (ref 37) used a simple test for polymer blends: normalize growth rates to the undercooling, $T_m-T$, which stands for the driving force, and to the distance above the glass transition, $T-T_g$, which stands for mass transport E2·C31. For polyoxyethylene with various polymeric additives, this collapsed growth rates that differed by about 3 orders of magnitude to within one order R16. A lower melting point $T_m$ means the crystallizing component has a lower chemical potential in the liquid, and so a smaller $\Delta G$; a changed $T_g$ means the additive has plasticized or antiplasticized the mixture C31. If neither $T_m$ nor $T_g$ changes, these two bulk routes cannot explain a change in growth rate C32·C36.

Reading hydrogen bonds from the infrared spectrum. Felodipine (Figure 1a) has one N–H group, a hydrogen-bond donor, and two ester carbonyl groups, which are acceptors M1. In amorphous felodipine the NH group bonds to the carbonyl of a neighbouring felodipine molecule, and its stretching band appears at 3341 cm⁻¹; a free NH group would appear near 3420 cm⁻¹ R9.a·R15. The carbonyl stretch appears as two bands, at 1701 cm⁻¹ (not hydrogen-bonded) and 1682 cm⁻¹ (hydrogen-bonded) R10·R15. When a polymer offers its own acceptor groups, the NH can bond to the polymer instead. If the new bond is stronger than the felodipine–felodipine bond, the NH band moves to lower wavenumber; if weaker, to higher wavenumber C18. The polymers differ in their acceptors: PVP carries a pyrrolidone carbonyl on every repeat unit, PVAc an acetate carbonyl, PVP/VA both, and HPMCAS a cellulose backbone with acetate, succinate and hydroxypropyl substituents (Figure 1) M2·M3·M4·M5.

Polymorphs. Felodipine was supplied as crystalline form I, but the spherulites that grew in these experiments were the metastable form II, identified by Raman microscopy and melting point M1·R3. This matters later, because the melting points used to judge the driving force were measured on form I X7.

Methods in detail

The study has two independent measurement tracks that meet in a comparison: crystal growth rates of felodipine with and without polymer, measured by microscopy on a hot stage, and drug–polymer hydrogen bonding, measured by infrared spectroscopy on separate films with much more polymer. Differential scanning calorimetry (DSC) supplies Tg and melting points, which serve as controls for the bulk explanations.

Making samples with very little polymer

Problem. To show that bulk properties are not involved, the polymer level must be low, a few per cent; but a dispersion with so little polymer cannot be made directly by evaporating a solution, because felodipine would partly crystallize during evaporation and the mixture might not be uniform C5.

What they did. All polymers were dried under vacuum for 48 h. First a dispersion with 10% w/w polymer was made by dissolving drug and polymer in a 1:1 mixture of dichloroethane and ethanol, removing the solvent by rotary evaporation and drying the solid for 48 h in a vacuum oven P1. This dispersion was then mixed with pure crystalline felodipine in a cryogenic mill to give 3% w/w polymer, and in the same way 0.5, 1.5 and 4.5% w/w P1. For each growth experiment, 3–5 mg of this material was melted at 145–150 °C on a cleaned cover glass, covered with a second cover glass to give a transparent film about 30–50 µm thick, and cooled rapidly to about 50 K below Tg; the absence of birefringence under crossed polarizers confirmed that the film was amorphous P2. Films were then stored at 40 °C over phosphorus pentoxide (0% relative humidity) so that nuclei would form P3.

Why it works, and what it assumes. Melting the cryomilled blend at 145–150 °C turns the crystalline felodipine and the 10% dispersion into one liquid, so the polymer can spread through the whole film; the authors present the two-step route as a way to "achieve uniform drug–polymer mixing at the molecular level" C5. The uniformity of the final 3% films was not measured directly; the single Tg of the 3% dispersions, close to that of felodipine, is at least consistent with one amorphous phase R1. Note by the AI writer (not discussed in the paper): the Materials section lists dichloromethane, and the IR films were cast from ethanol and dichloromethane, so "dichloroethane" for the growth samples may be a misprint; the paper does not resolve it X8.

Measuring growth rates

What they did. Nuclei formed at 40 °C were grown on a hot stage (Linkam THMS 600) calibrated with melting-point standards, at 70, 80, 90, 95, 100, 105 and 110 °C P4. Time-lapse pictures recorded the diameter of each spherulite; the diameter rose linearly with time, and the slope of a linear regression was taken as the growth rate. Experiments were done in triplicate P4·C9. Two series were run: 3% w/w of each polymer across the temperature range, and 0 to 4.5% w/w of each polymer at 90 °C A2·A3.

Why this design. Separating nucleation (at 40 °C, just below the Tg of about 43 °C) from growth (at 70–110 °C) means that the measured quantity is the speed of an existing crystal front, not the time needed to form a crystal P3·P4·M1. The linear growth law makes the rate a single, well-defined number for each system and temperature C9.

Note by the AI writer (not discussed in the paper): the text calls Figure 3 the change in radius, while its caption and axis say diameter. Slopes calculated from the Figure 3 points at 80 °C are about twice the 80 °C rates plotted in Figure 4 for all five systems, which would fit Figure 4 reporting the radial rate (half the diameter slope). Ratios and rankings are not affected, but absolute rates should be compared with other papers with this in mind X3.

Measuring hydrogen bonding

Problem. At 3% polymer, the felodipine–polymer contacts are too few for the infrared spectrum to show clear differences C6.

What they did. Films with 0, 20, 30, 40 and 50% w/w polymer were spin-coated from ethanol/dichloromethane solutions onto a substrate spinning at 500–2000 rpm, dried at 90 °C for several minutes in a nitrogen-purged glove box, and measured by transmission infrared spectroscopy (128 scans, 4000–400 cm⁻¹) P6·P7. The authors followed the felodipine NH band (3150–3450 cm⁻¹) and the carbonyl region (1550–1800 cm⁻¹) as the polymer content increased (Figures 7–10) R11·R12·R13·R14.

What it assumes. The ranking of hydrogen-bond strength found at 20–50% polymer, at the (unstated) measurement temperature, is taken to hold for the 3% films at 70–110 °C C6. The paper does not test this transfer X5.

Thermal controls

Onset Tg was measured by DSC at 10 °C/min after erasing the thermal history, for the pure materials and for the 3% dispersions P8. The melting point (as the extrapolated offset of the melting peak) was measured at 1 °C/min for crystalline felodipine and for its 3% physical mixtures with each polymer, prepared by cryomilling P9. Raman microscopy (783 nm laser) and the melting point identified the polymorph of the grown crystals P5·R3.

Experimental analysis

Result 1: low polymer levels slow growth, most at low temperature

All crystals grew as spherulites with a smooth front, with or without polymer, so the polymers changed how fast the crystals grew, not how they looked (Figure 2) R2. At 80 °C, a spherulite without polymer grew from about 106 to 182 µm in 1500 s, while one with 3% PVP grew only from about 90 to 130 µm in 8100 s (Figure 3, read from the plot) R4.a·R4.e. The slopes ranked felodipine alone > PVAc > HPMCAS > PVP/VA > PVP C9.

Figure 4 shows the growth rate across temperature. Without polymer, growth is fastest at about 105 °C, about 1.8 × 10⁻⁷ m/s, and slows on either side R5.a·C10. On cooling to 70 °C it falls to about 3.5 × 10⁻⁹ m/s R5.a. The text describes this as a decrease of 3 orders of magnitude; the plotted values give about 51-fold, or 1.7 orders of magnitude C10·X1. With 3% polymer the maximum "appears to have shifted slightly" to 110 °C, except with PVAc C11. Table A lists the values read from the figure.

Table A. Crystal growth rate of felodipine (m/s), all values about, read from the log axis of Figure 4 (95 °C omitted here) R5.

Figure 5 turns these into a ratio: the growth rate without polymer divided by the rate with polymer. Table B gives the ratios read from the figure. At 70 °C the text gives about 4-fold for HPMCAS, 7-fold for PVP/VA and 10-fold for PVP, and calls PVP K-12 "by far the best crystal growth inhibitor" R6·C14. For PVAc the mean growth rate was slightly lower than without polymer, but the difference was not statistically significant at any temperature R7·C13. For the three other polymers the ratio falls almost linearly as the temperature rises; the authors summarize that "polymer effectiveness increases with decreasing temperature", as Ishida and co-workers had seen for another system (ref 20) C12·R19.

Table B. Ratio of growth rates (felodipine alone / with 3% polymer), all values about, read from Figure 5 R6.

The text states that at 110 °C "the polymers appear to have no effect" on growth C11. Table B supports this for PVAc, HPMCAS and PVP/VA, whose ratios are about 1.1, but not for PVP, whose ratio is still about 2.9 R6.a·X2. Figure 4 shows the same: at 110 °C the PVP curve lies at about 5.3 × 10⁻⁸ m/s, against about 1.55 × 10⁻⁷ m/s without polymer R5.a·R5.e.

Result 2: the order holds at every polymer level

At 90 °C the logarithm of the growth rate falls linearly with polymer content between 0 and 4.5% w/w for each polymer (Figure 6) R8·C15. Table C gives the values read from the figure.

Table C. Growth rate of felodipine at 90 °C (m/s) against polymer content (w/w), all values about, read from the log axis of Figure 6; the 0% point is shared R8.

The authors read Figure 6 as follows: more PVAc "appears to have little effect" (a very small slope); HPMCAS needs at least 3% to give any measurable decrease; 1.5% PVP/VA or PVP already gives a noticeable reduction; and at every concentration the order is PVP > PVP/VA > HPMCAS > PVAc C15·C16·C17. Note by the AI writer (not discussed in the paper): the plotted points show some decrease already at 1.5% HPMCAS (about 4.3 × 10⁻⁸ against 6.7 × 10⁻⁸ m/s) and with 4.5% PVAc (about 3.8 × 10⁻⁸ m/s, roughly 1.8-fold below the drug alone), so these two statements are the authors' judgement of what counts as measurable X4.

Result 3: Tg, melting point, polymer Tg and molecular weight do not explain the order

Table D. Glass transition and melting temperatures from Table 1 of the paper (standard deviations within ±1 °C; Tm of crystalline felodipine and of its 3% physical mixtures), with the polymer molecular weights given in the Materials section R1·M1·M2·M3·M4·M5.

The Tg of the 3% dispersions lay between 42.7 and 44.1 °C, within about 1 °C of felodipine (43.3 °C), even though the pure polymers range from 38.7 °C (PVAc) to 115.5 °C (HPMCAS) R1·M1·C7. The melting point of felodipine was not depressed by 3% of any polymer R1·C8. Applying the Wu test, the authors conclude that "increases in Tg cannot account for the observed decrease" in growth rate, and that it "does not appear" that the polymers change the driving force at these levels C32·C36. They place this alongside earlier reports of polymers inhibiting crystallization without raising the dispersion Tg, or even while lowering it, and call it part of "a mounting body of evidence" that Tg is not a good predictor of crystallization tendency in multicomponent amorphous systems C33·R20.

Two other polymer properties fail as well. PVP, PVP/VA and HPMCAS have similar Tg (about 103, 103 and 115 °C), yet PVP and PVP/VA inhibit far better than HPMCAS, so the Tg of the pure polymer does not set the order C34·M2·M3·M4. And PVP K-12, the best inhibitor, has the lowest molecular weight of the four (order PVP K-12 < HPMCAS < PVP/VA < PVAc), so there is no correlation with molecular weight either C35·M2·M3·M4·M5. Note by the AI writer (not discussed in the paper): the melting points in Table 1 are those of form I in physical mixtures, while the crystals that grow are form II; the argument that the driving force is unchanged therefore rests on form I melting X7.

Result 4: PVP and PVP/VA form stronger hydrogen bonds with felodipine than felodipine forms with itself

Table E. Felodipine NH band position from Table 2 of the paper, with the authors' reading of the direction R9·C19·C21·C23.

With PVAc and HPMCAS the NH band moved slightly upward: to 3345 and 3343 cm⁻¹ at 20% polymer, and to 3354 and 3348 cm⁻¹ at 50% R11·R12·R9.b·R9.c. At the same time, the felodipine carbonyl band at 1701 cm⁻¹ (not hydrogen-bonded) grew relative to the band at 1682 cm⁻¹ (hydrogen-bonded) R11·R12. The authors read this as felodipine forming contacts with these polymers that are slightly weaker than its contacts with itself, "at the expense of drug–drug interactions" C19·C20.

With PVP/VA, the NH band stayed at 3341 cm⁻¹ at 20% polymer but broadened on its low-wavenumber side; with more polymer it moved down, to 3298 cm⁻¹ at 50%, and a shoulder at 3290 cm⁻¹ appeared and grew R13·R9.d. PVP gave the largest change of all: a downward shift and broadening that grew with PVP content, reaching 3294 cm⁻¹ at 50% R14·R9.e. Both patterns indicate a population of drug–polymer hydrogen bonds stronger than the drug–drug bonds C21·C23. In the carbonyl region the strong polymer carbonyl bands overlap those of felodipine, so no conclusion could be drawn there for PVP/VA and PVP C22. Note by the AI writer (not discussed in the paper): in Figures 9a and 10a a second band near 3225–3230 cm⁻¹ also grows with PVP/VA or PVP content (read from the figures) R13·R14.

Why does PVP/VA rank below PVP? The authors explain it by composition: 40% of PVP/VA is acetate side chains, which "appear to interact only weakly" with felodipine, as the PVAc films show; so a given weight of PVP/VA offers fewer pyrrolidone groups for hydrogen bonds than the same weight of PVP C24. Earlier work had also found that the acetate group of PVP/VA interacts less with a hydrogen-bond donor than the pyrrolidone group does (ref 31) C24. The authors summarize the hydrogen-bond strength or extent as PVP > PVP/VA > HPMCAS ≥ PVAc, and note that the differences are "expected" from the chemistry and number of acceptor groups of the polymers (Figure 1) C26·C25.

Result 5: the two rankings coincide

Put side by side, the two orders match: growth inhibition PVP > PVP/VA > HPMCAS > PVAc, hydrogen bonding PVP > PVP/VA > HPMCAS ≥ PVAc C17·C26. The polymers with the strongest drug–polymer interactions, PVP and PVP/VA, were the best growth inhibitors, and PVAc, the weakest, had essentially no effect; the authors write that the IR order "corresponds well" with the order of growth inhibition and that the results "strongly suggest" a key role for specific interactions C38·C40. The PVP–PVP/VA pair is the most direct comparison in the paper: the two polymers have almost the same Tg (102.9 and 102.6 °C) but differ in acceptor groups, and they differ in inhibition in the same direction as in hydrogen bonding M2·M3·R6·R9·C38.

Discussion

The authors' mechanism: demixing at the growth front

Having ruled out the bulk routes, the authors state that "the exact mechanism of retardation is not readily apparent", since the polymer levels used cause no measurable change in key bulk kinetic and thermodynamic properties C41. They then offer a molecular picture, which they themselves call "speculation" C42. For a crystal to grow, drug molecules must leave the polymer, that is, drug and polymer must demix, so that drug molecules can be incorporated at the growth front. If drug–polymer mixing is thermodynamically favourable, as in a miscible system, demixing is an activated step that requires breaking drug–polymer interactions; the more favourable these interactions, the less favourable the demixing, and the slower the addition of molecules to the crystal C42.

The authors' proposed chain, as described in the Discussion of the paper; steps 2 onwards are not measured C32·C36·C42·R6·X6.

The temperature data are the authors' support for this picture. At high temperature, thermal energy would be expected to weaken and disrupt drug–polymer hydrogen bonds, especially weak ones, so demixing becomes easier and the polymers are poor inhibitors near 110 °C; as the temperature falls, hydrogen bonding becomes more important, the polymers become better inhibitors, and they differ more from one another C43·R6. The authors conclude that the greater effectiveness at lower temperature is "consistent with our supposition" that the strength or extent of drug–polymer hydrogen bonding is "an important contributor" to the slower growth C44.

Note by the AI writer (not discussed in the paper): the temperature argument fits Figure 5, but it is not a test. Hydrogen bonding was measured only at one temperature, in films with 20–50% polymer, and no measurement shows demixing, or polymer piling up, at the growth front X5·X6. The same temperature trend could in principle arise if the polymer's effect on mobility near the crystal surface grew on cooling; the paper offers no data that separate the two.

The wider context the authors draw on

The authors set their result beside work on polymer blends, where spherulite growth of a crystallizable polymer was slowed much more by strongly hydrogen-bonding diluent polymers than by weakly interacting ones of similar Tg (Talibuddin et al., ref 40), and where, for poly(caprolactone), the order of growth inhibition by three polymers matched the order of hydrogen-bond strength (Kuo et al., ref 43) R17·R18. In those blends the additive was present at much higher levels, where Tg and Tm effects also had to be considered; only in weakly interacting systems did Tg become the more important factor (refs 32, 43) C39·R18. The present study extends the picture to a drug at 0.5–4.5% w/w polymer, where Tg and Tm do not change A3·R1.

What it suggests for formulation

The authors' own conclusion is modest: their observations "provide a potential framework" for selecting the best crystallization inhibitor for a given drug by considering the chemistry of drug and polymer and their potential for intermolecular interactions C46. They also stress that very low polymer levels (3% w/w) can substantially reduce crystal growth from the supercooled liquid, with large differences between polymers C28.

Note by the AI writer (not discussed in the paper): three practical points follow from the data. First, for a drug with a single NH donor like felodipine, a polymer rich in strong carbonyl acceptors (the pyrrolidone of PVP) slowed growth far more at the same weight than a polymer with fewer such groups (PVP/VA) or only acetate groups (PVAc), with no change in Tg R6·R9·R1. Second, the polymer with the highest Tg, HPMCAS, was not the best growth inhibitor here, so polymer Tg alone is a poor guide at these levels C34·M4. Third, because the polymers lose most of their effect near 110 °C, a growth test at high temperature would hide most of the differences seen at 70 °C; rankings are clearest at the lowest temperature tested R6. All three points come from one drug and four polymers, measured for growth in the dry supercooled melt; they say nothing about nucleation, humid storage or crystallization from solution X10.

Limitations

Limits the authors state. The authors say plainly that the exact mechanism of retardation "is not readily apparent", and they present the demixing picture as "speculation" C41·C42. In the infrared spectra, the carbonyl region could not be interpreted for PVP/VA and PVP because the polymer carbonyl bands overlap those of felodipine C22. Infrared spectroscopy needed 20–50% polymer to show differences, far more than the 3% used for growth C6. They close by saying that the link between drug–polymer interactions and growth inhibition "should be further investigated" with drugs whose chemistry differs from felodipine and with strongly and weakly interacting polymers C45.

The AI writer's own judgement (1): what the evidence supports. The growth data are the strongest part of the paper. They were measured in triplicate, and the same order, PVP > PVP/VA > HPMCAS > PVAc, appears at every temperature (Figure 4) and at every polymer level (Figure 6) P4·R5·R6·R8. The Tg and melting point measurements make the bulk explanations unlikely R1. The link to hydrogen bonding, however, is a rank correlation over four polymers; no quantitative relation between band shift and growth rate is tested, and the hydrogen bonds were measured at other compositions and at one temperature C38·X5. The demixing mechanism and the weakening of hydrogen bonds on heating were not measured X6. Only growth was measured; nucleation with the polymers was not, although the abstract and conclusions speak of "crystallization inhibitors" X10. An overall rating of moderate seems fair: the main ranking is solid, the explanation is plausible but untested.

The AI writer's own judgement (2): where the text and the figures differ. Several statements in the text are stronger or different from what the figures show, and a reader using the numbers should prefer the figures:

• The fall of the growth rate of felodipine alone from 105 to 70 °C is described as 3 orders of magnitude; Figure 4 shows about 51-fold X1.
• At 110 °C the polymers are said to "appear to have no effect", but PVP still slows growth about 3-fold (Figure 5) X2.
• Slopes from the Figure 3 points are about twice the Figure 4 rates at 80 °C, and Figure 3 is called both a change in radius (text) and in diameter (caption and axis); Figure 4 may report the radial growth rate X3.
• HPMCAS is said to need 3% for any measurable decrease, but the 1.5% point in Figure 6 already lies below the drug alone; PVAc is said to have little effect, but 4.5% PVAc lowers the plotted rate about 1.8-fold X4.
• The melting points used to argue that the driving force is unchanged are those of form I in physical mixtures, while form II grows X7.
• The solvent for the growth samples is printed as dichloroethane, while the Materials section lists dichloromethane X8; the meaning of the error bars and the statistical test for PVAc are not given X9; and whether "40% acetate" in PVP/VA refers to mass or moles is not stated X11.

None of these changes the ranking of the polymers, which is the paper's main result.

Conclusion

This short paper makes one clean comparison. With one drug, felodipine, and four polymers added at only 3% w/w, the growth rate of crystals from the supercooled liquid fell by up to about ten-fold at 70 °C, in the order PVP > PVP/VA > HPMCAS > PVAc, and the differences shrank as the temperature approached 110 °C R6·C17·C12. Neither the Tg of the dispersions, nor the melting point, nor the Tg or molecular weight of the pure polymers followed this order C32·C34·C35·C36. The infrared order of drug–polymer hydrogen bonding did: the pyrrolidone polymers formed bonds with the felodipine NH stronger than those felodipine forms with itself, while HPMCAS and PVAc formed weaker ones C26·C38. The authors conclude that polymers forming stronger or more extensive hydrogen bonds with the drug "appear to be better crystallization inhibitors", and they propose, as speculation, that strong bonds make it harder for the drug to demix from the polymer at the growth front C40·C42.

For formulation work the lesson is that, at low loadings, the chemistry of the polymer can matter far more than its glass transition: a polymer that offers the drug strong hydrogen-bond partners can slow crystal growth substantially even when it leaves Tg unchanged C33·C46. Testing the idea on other drugs, measuring hydrogen bonding under the growth conditions and measuring nucleation as well as growth are the obvious next steps C45·X5·X10.