When humidity splits a drug–PVP glass: moisture-induced amorphous–amorphous phase separation before crystallization

Background and motivation

An amorphous drug dissolves faster and reaches a higher apparent solubility than its crystal, but it is physically unstable: it tends to crystallize, and water makes this worse. Amorphous solids absorb water into their bulk, water lowers the glass transition temperature ($T_g$) of both small molecules and polymers, and the extra mobility speeds crystallization (refs 8–10, 12–14 of the paper). The usual remedy is to mix the drug at the molecular level with a second component, often a hydrophilic polymer such as poly(vinylpyrrolidone) (PVP), to make an amorphous solid dispersion (ASD). A hydrophilic polymer, however, makes the dispersion more hygroscopic than the drug alone.

The authors' starting point is that water does more than plasticize. Sorbed water is a third component, so it changes the thermodynamics of the mixture (Marsac et al. 2008, ref 15). For a hydrophobic drug in a hydrophilic polymer, water may drive the system toward immiscibility: the drug and polymer, which were mixed, separate into amorphous drug-rich and polymer-rich domains. Earlier work had reported this for felodipine–PVP (refs 16, 17) and griseofulvin–PVP (ref 18) R27·C8. Amorphous–amorphous phase separation without water was also known, for indomethacin–citric acid (ref 22) and itraconazole–Eudragit E100 (ref 23), but how moisture affects drug–polymer miscibility had not been studied widely.

The paper therefore tests one hypothesis: certain drug–polymer ASDs undergo moisture-induced amorphous–amorphous phase separation before the drug crystallizes C10. The polymer is PVP throughout. The five model drugs, nifedipine, pimozide, droperidol, indomethacin and ketoprofen, were chosen to cover a range of chemistries and possible interactions with PVP; felodipine, the drug in which the effect was first seen, is carried along in the discussion from earlier work. Infrared (IR) spectroscopy is the main tool, backed by differential scanning calorimetry (DSC) and moisture sorption.

Background concepts

Amorphous–amorphous phase separation (AAPS). A one-phase amorphous dispersion can split into two amorphous phases of different composition, here a drug-rich and a polymer-rich phase, without any crystal forming. The paper also calls the moisture-driven version "moisture-induced immiscibility". It matters for stability because a polymer can only inhibit crystallization of drug molecules that it is mixed with.

IR as a probe of mixing. When a drug donates a hydrogen bond to the PVP carbonyl, the bands of the groups involved shift. PVP has one hydrogen-bond acceptor per repeat unit, a carbonyl at 1681 cm⁻¹ when free, and no donor. In the dispersions this carbonyl appears at about 1660 cm⁻¹ with nifedipine, 1661 with pimozide, 1657 with droperidol, 1639 with indomethacin and 1637 cm⁻¹ with ketoprofen (Table 1) R4. If the dispersion separates, drug molecules meet mostly other drug molecules, so drug–polymer bands weaken and drug–drug bands grow C9·C15. The authors compare each stored spectrum with two references: the crystalline drug (crystallization) and a calculated spectrum of a physical mixture of amorphous drug and PVP that do not interact (amorphous separation).

The calculated physical-mixture spectrum. The Appendix explains how it is built. The spectra of pure amorphous drug and pure PVP are added, each scaled so that a C–C band of the drug and one of PVP have the intensity ratio set by the number of C–C bonds at the chosen weight ratio E7·C58. For 70 g pimozide and 30 g PVP (PVP assumed to be 40,000 g/mol): 0.152 mol pimozide and $7.5\times {10}^{-4}$ mol PVP, that is $9.13\times {10}^{22}$ and $4.52\times {10}^{20}$ molecules; with 9 C–C bonds per pimozide and 1440 per PVP molecule, $8.22\times {10}^{23}$ and $6.5\times {10}^{23}$ bonds, a band-intensity ratio of about 1.26 (pimozide band at 1223 cm⁻¹, PVP band at 1286 cm⁻¹) R30. The method assumes that these C–C bands keep their position and relative intensity in the mixture.

Glass transition and plasticization. A miscible binary dispersion shows one $T_g$, between those of its components; two $T_g$ events mean two phases. Water lowers $T_g$. If $T_g$ falls below the storage temperature, molecules can diffuse on practical time scales, so separation and crystallization become possible.

Activity coefficients. The authors try two ways of expressing how badly each drug mixes with water. The first uses the water uptake of the amorphous drug (Eq. 1):

In plain terms: the water activity $a_{\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}$ in the solid equals the relative vapour pressure $P/P_0$ (RH/100), and dividing it by the mole fraction of water absorbed, $X_{\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}$, gives the activity coefficient of water ${\gamma }_{\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}$, which is 1 for ideal mixing and larger when water and drug dislike each other. The paper evaluates it from the uptake at 95% RH E1. The second way uses the drug's aqueous solubility (Eq. 2), with the free energy difference between supercooled liquid and crystal estimated by the Hoffman approach (Eq. 3):

In plain terms: $X_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ is the measured mole-fraction solubility, $\Delta G_c$ is the free energy by which the supercooled liquid lies above the crystal (also the driving force for crystallization), $\Delta H_f$ is the heat of fusion, $T_m$ the melting temperature and $T$ the temperature. Comparing the real solubility with the ideal solubility of the crystal gives the activity coefficient of the drug in water, ${\gamma }_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ E2·E3. Eq. 2 is printed without the usual division of $\Delta G_c$ by $RT$; this matters for the numbers and is taken up below.

Flory–Huggins interaction parameters and the ternary free energy. The authors' thermodynamic argument rests on the Flory–Huggins free energy of mixing written for water (W), drug (D) and polymer (P) (Eq. 4):

In plain terms: $R$ is the gas constant, $T$ the temperature, $n$ the amount and $\varphi$ the volume fraction of each component (the text calls $n$ a mole fraction). The first three terms are the combinatorial entropy of mixing, which always favours mixing. The last three are the interaction terms of the three pairs, each scaled by an interaction parameter $\chi$; the authors note that $\chi$ also absorbs any non-ideal entropy E4. A positive $\chi$ means unfavourable contacts. In a water–PVP–hydrophobic drug system, ${\chi }_{WD}$ is large and positive, ${\chi }_{WP}$ is favourable, and ${\chi }_{DP}$ depends on how well the drug bonds to PVP. The paper also uses drug–water $\chi$ values calculated from sorption at 95% RH by literature methods (refs 8, 38), without printing the formula E5.

Methods in detail

The study has one design: make each dispersion as a single phase, expose it to a fixed relative humidity (RH), and ask whether it separates before it crystallizes. Three techniques answer the question independently, and a set of drug properties is then compared with the outcome.

Contribution 1: IR of humidified films read against a calculated physical mixture

Problem. Both crystallization and amorphous separation destroy drug–polymer hydrogen bonds, so a loss of these bands alone cannot tell them apart; and water bands overlap the regions of interest.

What they did. Each drug and PVP were dissolved in a common solvent (nifedipine in a 1:1 w/w dichloromethane:ethanol mixture, droperidol and pimozide in chloroform, indomethacin and ketoprofen in ethanol), checked visually to be one phase, spin-coated on ZnS or KRS-5 at 500/2500 rpm for 18 and 30 s, and dried for at least 1 min on a 90 °C hot plate. Transmission spectra were collected from 500 to 4000 cm⁻¹ at 4 cm⁻¹ resolution with 128 scans, in dry air, keeping the absorbance of the region of interest between 0.6 and 1.0 P1·P2. Films were stored at room temperature in desiccators over saturated Mg(NO₃)₂, NaCl, KCl and KNO₃, giving 54, 75, 84 and 94% RH. At each time point a film was flushed with dry air (RH below 10%) until its OH region stopped changing, then measured P4. The compositions shown are 50% PVP (w/w) for nifedipine, 30% for pimozide, droperidol and ketoprofen, and 25% for indomethacin. One pimozide film with 50% PVP was stored at 4 °C and 75% RH P5. Amorphous references were spin-coated the same way, crystalline references measured by attenuated total reflectance P3, and the physical-mixture spectrum was calculated as described above P7.

Why it works. A film that has separated into amorphous phases should look like the non-interacting amorphous mixture; a film that has crystallized should show crystal bands. Drying removes the water bands. To check that drying itself does not disturb the drug–polymer bonds, nifedipine films were also measured wet after 75, 84 and 94% RH: the same changes were present, though harder to read (data not shown; ref 24) R1. The authors conclude that drying did not disrupt the interactions ("These results suggest") C11.

Contribution 2: DSC and water uptake as independent checks

DSC. Dry miscibility was checked by DSC only for droperidol–PVP and pimozide–PVP, because literature data existed for the other three. Solid dispersions made by rotary evaporation from the same solvents were heated at 20 °C/min to 10 °C above the drug's melting point, held about 12 s, cooled at more than 20 °C/min to −20 °C and reheated at 20 °C/min P8. To test the effect of humidity, rotary-evaporated dispersions in open pans were held at 25 °C and 94% RH in a gravimetric analyzer for 42–72 h, then either sealed wet or first dried in dry air (RH below 5%) to a weight plateau, and scanned at 20 °C/min; midpoint $T_g$ values are reported P9. Melting points and heats of fusion of the crystals came from 10 °C/min scans; droperidol showed two overlapping events at that rate, three at 0.5 °C/min (melting of one polymorph, recrystallization, melting of a second), and one at 200 °C/min, which was used P10·R2.

Water uptake. Uptake at 25 °C and 95 (±1)% RH was taken where the weight-gain curve first reached a plateau, for PVP, the five amorphous drugs and one dispersion of each P11. It was compared with an additive prediction, the weighted sum of the pure-component uptakes E6. The reasoning: drug–polymer interactions in a mixed dispersion change how much water it takes up, so a mixed dispersion should deviate from additivity, whereas separated phases with fewer drug–polymer contacts should sorb close to additively ("it would be anticipated") C25.

Contribution 3: testing drug properties as predictors

The authors collected the drugs' molecular weight, aqueous solubility, calculated log P, melting data, $\Delta G_c$ at 25 °C and density (Table 3) M1–M6, and calculated ${\gamma }_{\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}$, ${\gamma }_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ and the drug–water $\chi$ (Table 4) R24. They also derived, from Table 1, how far the PVP carbonyl band shifts when it accepts a hydrogen bond from each drug R29. The question is which of these, if any, separates the susceptible drugs from the others.

Experimental analysis

Main result 1: the dry dispersions start as one phase. In all five dry films the spectrum differed both from the calculated physical mixture and from the crystal, the signature of specific drug–polymer interactions R3. For nifedipine the NH band sits at 3342 cm⁻¹ in the amorphous drug and moves to 3288 cm⁻¹ when bonded to PVP; for all five the PVP carbonyl appears below its free position (Table 1) R4. DSC of droperidol–PVP and pimozide–PVP showed a single $T_g$ between those of drug and polymer at polymer contents up to 50% (dry weight); the pimozide–PVP values were 94 °C at 50% PVP and 81 °C at 30% PVP, against 60 °C for amorphous pimozide R5·R17·R18. For indomethacin–, nifedipine– and ketoprofen–PVP the single $T_g$ is taken from earlier reports (refs 29, 21, 30) R6. The authors conclude that all initially dry systems are miscible over the compositions studied C1·C14.

Main result 2: three dispersions separated before they crystallized. Table A collects the IR results (Figures 2–7).

Table A. IR results after storage at room temperature and drying (from the text and Figures 2–7) R7–R16.

For nifedipine–PVP, the authors read the loss of the bonded band and the growth of the band of amorphous nifedipine as a shift from drug–polymer to drug–drug interactions, "consistent with" a drug-rich amorphous phase; the missing crystal band shows that crystallization did not cause it, and crystal bands appeared only after 48 h R7·C17. In Figure 2 the 24 h and 4 day spectra both peak at about 3343 cm⁻¹, on top of the physical-mixture and amorphous-drug traces. Note by the AI writer (not discussed in the paper): Figure 2 contains no 1 h spectrum, and no separate crystal band near 3317 cm⁻¹ is visible even at 4 days, so both the 1 h change and the 48 h crystallization rest on spectra that are not shown X8.

For pimozide–PVP, the free drug carbonyl and the bonded PVP carbonyl lost intensity; the authors read this as more drug–drug hydrogen bonding and separation into drug-rich and polymer-rich amorphous regions R9·C18. Figure 4 shows the RH dependence: with increasing storage RH the spectrum moves progressively toward the physical mixture, and a change is already present after 3 days at 54% RH R10. The 50% PVP film at 4 °C and 75% RH changed in the same way, but "at a much slower rate" than at 22 °C R11. Note by the AI writer: in Figure 4 the storage time differs at each RH (3 days at 54%, 2 days at 75 and 84%, 1 h at 94%), so RH and time change together; the 54% RH spectrum still has the two-band shape of the fresh film, with a weaker shoulder near 1660 cm⁻¹ X9.

For droperidol–PVP, the free drug carbonyl at 1717 cm⁻¹ became a shoulder within 1 h; the band at 1685 cm⁻¹ then moved to 1683 cm⁻¹ (crystal 1682 cm⁻¹) after 22 h, more so after 6 days R12·C19. Similar changes were seen at 75% RH (data not shown) R13.

Main result 3: indomethacin and ketoprofen stayed one phase. The bonded bands of indomethacin–PVP kept their position and relative intensity for 40 days at 94% RH, and those of ketoprofen–PVP for 68 days, "indicating no detectable molecular level changes" R14·R16·C20. Indomethacin then crystallized: IR changes and visible crystals after 46 days, assigned to the δ-polymorph from a band at 1357 cm⁻¹; by 86 days a new band near 1734 cm⁻¹ is visible in Figure 6 R14·C21. Ketoprofen–PVP developed a shoulder near 1688 cm⁻¹ and a small band at 1651 cm⁻¹ at 125 days, together with visible crystals R16. Dispersions in the lower molecular weight PVP K12 gave the same IR result for indomethacin (data not shown) R15.

Main result 4: DSC sees two amorphous phases in wet pimozide–PVP and one in indomethacin–PVP. Table B lists the $T_g$ values.

Table B. Midpoint $T_g$ values from Figures 8 and 9 and the text R17–R21·R25.

Pimozide–PVP has two $T_g$ events after humidity exposure, and they persist after drying; the lower dried value (61 °C at 50:50, 59 °C at 70:30) sits at the $T_g$ of pure amorphous pimozide, which the authors read as a drug-rich amorphous phase, "supporting the IR results" R17·R18·C22. Indomethacin–PVP K29-32 kept one $T_g$, but its wet $T_g$ reportedly stays above room temperature (ref 29) R20, so its mobility could be the limit. The authors therefore tested a more plasticized system, indomethacin–PVP K12: its $T_g$ fell from 68 °C to 11 °C and it still showed one event R21·C23. Note by the AI writer: the caption of Figure 9b mentions only the dry and humidified samples, but the panel also shows a third, re-dried trace with a step and endotherm near 70 °C that the text does not discuss X15.

Main result 5: water uptake is additive only where the phases separated.

Table C. Water sorbed at 25 °C and 95% RH (from Table 2) R22.

The two dispersions that stayed mixed took up far less water than the additive prediction, as had been reported for other PVP dispersions (ref 29) R23; the three that separated were within 7.4%. The authors call this "consistent with" phase separation C24. The prediction reproduces from the pure-component values (for example $0.75\times 3.0+0.25\times 77.7=21.7$), and the "% difference" column follows (predicted − measured)/measured within rounding, a relation the paper does not print E6.

Main result 6: the separating systems crystallized sooner. Crystal bands appeared after 22 h (droperidol), 48 h (nifedipine) and 6 days (pimozide) at 94% RH, against 46 days (indomethacin) and 125 days (ketoprofen) R7·R9·R12·R14·R16. The abstract summarizes this as "<6 days" against "≥46 days" C3. Note by the AI writer: pimozide's first sign came "after 6 days", which does not fit "<6 days" exactly, and the times are sampling points at which crystals were first reported rather than measured onset times X6.

Main result 7: what sorts the drugs. Table D puts the candidate predictors side by side.

Table D. Drug properties (Table 3), activity coefficients and drug–water χ at 25 °C and 95% RH (Table 4), and the PVP carbonyl red shift calculated from Table 1 M1–M6·R24·R29.

Log P does not predict the outcome: droperidol, felodipine and nifedipine are less hydrophobic than indomethacin and ketoprofen, yet they separate; it does not predict the uptake of the pure amorphous drug either ("It is clear") C27. Melting point, heat of fusion and $\Delta G_c$ show no correlation C32. All activity coefficients are well above 1, so water and each drug mix non-ideally with unfavourable contacts, but neither $\gamma$ shows a trend with separation C28·C29. The drug–water $\chi$, however, does split the groups: the four drugs with the highest values separated (pimozide 3.58, felodipine 3.18, droperidol 2.85, nifedipine 2.84) and the two with the lowest did not (ketoprofen 2.71, indomethacin 2.50), while PVP–water interactions are reported to be much more favourable R24·R28·C45·C46. So does the donor group: the four susceptible drugs bond to PVP through an NH group, the two others through a carboxylic acid, and the PVP carbonyl shifts 42–44 cm⁻¹ with the acids against 20–24 cm⁻¹ with the NH drugs R29·C47·C48.

Note by the AI writer (from the card's extractor checks): three points about Table 4 need care. First, the ${\gamma }_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ values cannot come from Eq. 2 as printed. With $\Delta G_c/RT$ (8.23 kJ/mol divided by $RT$ at 298.15 K, about 3.32) and the mole-fraction solubility of nifedipine ($2.03\times {10}^{-6}$), Eq. 2 gives about $1.36\times {10}^7$, not $4.95\times {10}^5$; the printed values match $1/X_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ within 0.5% for nifedipine, droperidol, pimozide and ketoprofen, as a kJ-versus-J slip would produce. The ${\gamma }_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ column therefore ranks the drugs by solubility alone X1. Second, the text says indomethacin and droperidol have the smallest ${\gamma }_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$, but the table gives the smallest values to ketoprofen and nifedipine X2. Third, the text groups droperidol with felodipine and pimozide as having the least favourable water interactions "based on activity coefficients calculated from water vapor sorption data", yet droperidol has the second-lowest ${\gamma }_{\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}$ (3.00); that grouping matches the $\chi$ order instead X3. The ${\gamma }_{\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}$, $\Delta G_c$ and $\chi$ values themselves reproduce from the printed inputs: for nifedipine, 2.2% uptake gives ${\gamma }_{\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}\approx 3.20$ (printed 3.19), and Eq. 3 gives $\Delta G_c=8.23$ kJ/mol (printed 8.2) E1·E3·E5.

Discussion

Thermodynamics or kinetics? For moisture-induced separation to happen, the separated state must have a lower free energy than the one-phase ternary mixture, and the molecules must be mobile enough to diffuse on the time scale of observation; water can affect both C34. Figure 10 sets two limiting pictures side by side. In a purely kinetic route, the dry drug–PVP glass is a one-phase mixture trapped above a free energy maximum, already wanting to separate; water only plasticizes it and lets it go, which would mean the dry binaries were never thermodynamically miscible C36. In a purely thermodynamic route, the dry mixture sits at a free energy minimum; sorbed water interacts favourably with the polymer and makes the matrix more polar, the hydrophobic drug interacts unfavourably with this polymer–water "cosolvent", and splitting into drug-rich and polymer-rich regions lowers the free energy C37. The figure caption adds that in reality water "likely" affects both.

The authors favour the thermodynamic route: "It is our opinion that moisture induced amorphous-amorphous phase separation is a thermodynamically driven event" that needs some enhanced mobility C35. Their evidence comes from the two systems that did not separate. Ketoprofen–PVP 70:30 had a midpoint $T_g$ of −23 °C after storage at 94% RH, so it was well above $T_g$ at room temperature, mobile enough to separate, yet it stayed one phase R25·C38; indomethacin–PVP K12, with a wet $T_g$ of 11 °C, also stayed one phase R21·C39. Being above $T_g$ was therefore not enough. In addition, earlier thermodynamic analysis found that mixing nifedipine or indomethacin with PVP is favourable without water, which contradicts the premise of the kinetic picture (ref 33) R26·C40. Mobility still sets the rate: pimozide–PVP separated at 4 °C, but much more slowly than at 22 °C, as felodipine–PVP had done (ref 27) R11·C41.

Why some drugs and not others. The authors reason from Eq. 4. Adding water forms a one-phase ternary mixture; since the entropy of mixing favours mixing, separation "must" come from a positive mixing enthalpy, that is from the balance of the three pair interactions C42. Complete miscibility is expected when the water–drug, water–polymer and drug–polymer interactions balance so that $\Delta H_{\mathrm{m}\mathrm{i}\mathrm{x}}$ is negative or only slightly positive and $\Delta G_{\mathrm{m}\mathrm{i}\mathrm{x}}$ has one minimum C43. When the pair interactions differ widely, as for water, PVP and a hydrophobic drug, sorbed water can make the environment unfavourable: water and PVP form a cosolvent whose ability to dissolve the drug falls as it takes up more water, an effect documented for ternary solvent systems (ref 37) and for felodipine–PVP–water (refs 17, 27) C37·C44. Two of the pair terms then decide the outcome. Drug–water: all ${\chi }_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}-\mathrm{w}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{r}}$ values are large and positive, and the four drugs with the highest values separated C45·C46. Drug–polymer: the NH-donor drugs bond less strongly to the PVP carbonyl than the acid drugs, because the hydrogen of a COOH group is the better donor, as the larger red shift shows C47·C48. A stronger bond should make the drug–polymer interaction more favourable; the reported $\chi$ is −0.8 for indomethacin–PVP against 0.0 for nifedipine–PVP, and this "might lead to a greater resistance to moisture induced immiscibility" R26·C49. The authors close the argument with a caveat: "Clearly more work needs to be done" on how the effect depends on drug and polymer chemistry C50.

The chain the authors propose, from water uptake to the crystallization route, can be drawn as follows.

Two routes to crystals. Figure 11 turns this into the paper's main conclusion: when a molecular-level dispersion of a hydrophobic drug and a hydrophilic polymer takes up water, the drug can crystallize either directly from the plasticized one-phase dispersion or from a plasticized drug-rich amorphous phase after separation C5. In the first case the polymer is still mixed with the drug and "can inhibit crystallization to a greater extent"; after separation the polymer concentration in the drug phase is reduced. "It would be anticipated that drug crystallization would be much faster following amorphous-amorphous phase separation", and the authors find "this supposition is supported" by the observation that nifedipine– and pimozide–PVP crystallized after a few days while indomethacin– and ketoprofen–PVP were stable for weeks C6·C3.

What it suggests for formulation. The authors draw three implications. The observations "underscore the importance of drug-polymer miscibility" for physical stability C51. Separation at RH as low as 75 or 54% shows a problem with accelerated stability testing: for susceptible systems "the route of crystallization may be altered at high RH", so behaviour under accelerated conditions may not represent lower RH C52. And during dissolution, a hydrating dispersion reaches very high water activity, so separation before dissolution "can be postulated", possibly promoting crystallization in the hydrated matrix, which would matter most for controlled-release products C53.

Note by the AI writer (not discussed in the paper): two practical screens follow from the data. A dry dispersion with a single $T_g$ can still separate in humidity, so a film stored at 75% RH and dried, with its IR spectrum compared against a calculated physical mixture, is a cheap test. And water uptake close to the weighted sum of drug and polymer uptake, as for the three separating systems here, is a quick warning sign, whereas strongly sub-additive uptake went with the dispersions that stayed mixed.

Limitations

Limitations stated by the authors. The authors present the drug–polymer interaction explanation as a speculation ("It is speculated that the strength of the drug-polymer interactions ... is important"), name drug–water interaction strength and the total amount of water sorbed as other factors "of likely importance", and say that more work is needed on the dependence on drug and polymer chemistry C7·C56·C50. They note that the water bands complicated the wet spectra, which is why films were dried before IR, and that PVP K29-32 does not plasticize indomethacin dispersions below room temperature, which is why they added PVP K12 R1·R20.

The AI writer's own judgement (1): how far the evidence for separation goes. The cleanest evidence is for pimozide–PVP: IR, two $T_g$ events wet and re-dried at two compositions, and additive water uptake all agree R9·R17·R18·R22. For nifedipine and droperidol, separation is called from IR of re-dried films and from water uptake only; the check that drying does not alter the spectra was made for nifedipine and is not shown, and no DSC after humidity is reported for either X10. The IR evidence is qualitative throughout: band intensities are compared by eye, without ratios or curve fitting, and several key spectra (nifedipine at 1 h and 48 h, all 75% RH data, the 4 °C series, PVP K12) are described but not shown X8. The samples also differ between techniques: spin-coated films at room temperature for IR, rotary-evaporated solids at 25 °C for DSC and sorption, and the dry miscibility $T_g$ measured after melting and quenching X11. Single-$T_g$ evidence of dry miscibility was produced here only for droperidol and pimozide X12.

The AI writer's own judgement (2): what the comparisons can and cannot tell. The faster crystallization of the separating systems compares different drugs; no drug was crystallized both with and without separation, and other differences point the same way, for example the $T_g$ of indomethacin–PVP K29-32 reportedly stays above room temperature even at high RH, so that dispersion is less mobile, and ketoprofen has the smallest $\Delta G_c$ X7. The explanation of susceptibility is confounded: donor type (NH or COOH), PVP carbonyl red shift and drug–water $\chi$ split the six drugs into exactly the same two groups, and the paper cannot say which factor governs. The $\chi$ gap between the groups is only 0.13 (nifedipine 2.84 against ketoprofen 2.71), the $\chi$ values depend on densities of which three are approximated as 0.9 times the crystal density, felodipine's sorption input is not reported, and its separation is cited rather than measured X13·X4·X5. The step from "entropy favours mixing" to "separation must come from a positive mixing enthalpy" sits uneasily with the paper's own statement that $\chi$ also contains non-ideal entropy X14. Finally, the ${\gamma }_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ column and two text rankings of the activity coefficients are inconsistent with Table 4 and Eq. 2, as described above, although the paper's conclusion that the activity coefficients show no trend does not depend on them X1·X2·X3.

Conclusion

The paper shows that moisture-induced amorphous–amorphous phase separation is not limited to felodipine. Of five drugs in PVP, nifedipine, pimozide and droperidol separated before crystallizing, pimozide from 54% RH, while indomethacin and ketoprofen stayed one phase until they crystallized much later C2·C3·C4. IR against a calculated physical mixture, two $T_g$ events in wet pimozide–PVP and near-additive water uptake give three independent views of the same split C16·C22·C24. Mobility alone did not cause separation, which the authors take as evidence that it is thermodynamically driven C35·C38·C39. Which drugs are prone is linked to weaker NH-donor bonds to PVP and more unfavourable drug–water interactions, an explanation the authors themselves call speculative C7·C46·C47. For formulation, the lesson is that dry miscibility is not enough: a dispersion must stay mixed in the presence of water, and high-RH stability tests can switch a susceptible system to the faster crystallization route C51·C52.