Why absorbed moisture splits some drug–PVP dispersions and not others

Background and motivation

Many drug candidates dissolve poorly in water, and one way to raise their apparent solubility is to deliver them in the amorphous form, often dispersed at the molecular level in a polymer (Hancock 2000, ref 1). The benefit lasts only as long as the drug does not crystallize, and that has to hold for the whole shelf life of the product. Absorbed moisture is one of the stresses known to speed crystallization of amorphous drugs, alone or in mixtures (Konno 2008, Rumondor 2009, Marsac 2008; refs 2–4) C4.

Moisture can do something less obvious before any crystal appears. In a binary dispersion in which drug and polymer were mixed at the molecular level, absorbed water can produce regions with different drug–polymer ratios: drug-rich and polymer-rich amorphous domains. The authors call this amorphous–amorphous phase separation (AAPS). Their group had shown it for felodipine–PVP (Marsac 2010, ref 5), while other dispersions stayed mixed under the same kind of stress (Rumondor 2010 and 2009, refs 6, 7) C5. The factors "thought to" decide the outcome were the strength of the drug–polymer interaction (refs 7, 8) and the hygroscopicity of the dispersion (ref 6), which itself depends on how hygroscopic each component is and on how strongly they interact (ref 9) C6. The phenomenon matters because drug-rich regions have been linked to faster crystallization, and a change in miscibility could also change dissolution (refs 3, 15, 16) C27.

Earlier work had also produced some rules of thumb. Dispersions made with less hygroscopic polymers such as HPMCAS were in general less susceptible to moisture-induced separation (ref 3), whereas dispersions with PVP, an extremely hygroscopic polymer, "appear to be highly vulnerable", but with a susceptibility that depends strongly on the drug C29. Very hydrophobic drugs that do not interact strongly with the polymer had been found to separate (ref 7) C34. What was missing was a view across many drug–polymer pairs of which properties make a dispersion resist.

The paper therefore does two things. It adds five model drugs chosen to differ in their ability to donate a hydrogen bond to PVP, so that the effect of interaction strength can be probed, and it combines their outcomes with 15 systems from earlier studies in a discriminant model to find the properties most associated with resistance C7·R15·P7.

Background concepts

Amorphous–amorphous phase separation. A one-phase amorphous dispersion is a glass in which drug and polymer molecules are mixed at the molecular level. If it splits into two amorphous phases of different composition, nothing has crystallized yet, but the drug-rich regions are no longer in close contact with the polymer. The paper does not spell out a molecular mechanism for the split; it treats the strength of the drug–polymer interaction and the amount of water taken up as the controlling factors C6·C1. It reads separation from changes in the infrared spectrum of films that were dried again after storage C16·P3.

Reading drug–polymer hydrogen bonds by infrared spectroscopy. PVP has carbonyl groups that accept hydrogen bonds and no donors. In pure PVP its carbonyl band sits at 1,682 cm⁻¹. When a drug donates a hydrogen bond to the carbonyl, part of that band moves to lower wavenumber (a "red shift"), so a dispersion shows a free band and a bonded band side by side R2·C11. A stronger donor gives a larger shift (refs 7, 20, 21), and the paper also uses the drug's pKa as a second measure of donor strength: a stronger acid (lower pKa) is "expected" to be a better hydrogen-bond donor, and donor strength depends on the electronegativity of the donor atom and on electron-withdrawing groups nearby C30. If moisture separates the dispersion, drug and PVP lose contact, the bonded band shrinks and the free band grows back, or the spectrum moves toward that of a simple mixture of the pure amorphous components C16.

Calculated physical-mixture spectra. To show that a dispersion is intimately mixed, the authors compare its spectrum with a "theoretical" spectrum of a physical mixture of the pure amorphous drug and PVP, calculated as in their earlier work (ref 7) P5. If the dispersion spectrum cannot be built from the pure spectra, the components interact; if a stored sample comes to look like the calculated mixture, the drug and polymer have separated C13·C25.

Molecular descriptors. Table II describes each drug by molar mass, numbers and types of hydrogen-bond donors and acceptors, pKa (calculated), log P (estimated, a measure of hydrophobicity), log S (aqueous solubility), the glass transition temperature Tg and the heat-capacity step ΔCp of the amorphous drug, and the melting temperature Tm and heat of fusion ΔHf of the crystal; each system also has its PVP carbonyl shift and an outcome class, 1 for "no separation on moisture exposure" and 0 for "separation" R15·P8·P9.

OPLS-DA. Orthogonal projections to latent structures discriminant analysis is a supervised multivariate method. It looks for the combination of descriptors that best separates two classes (the predictive component) and puts the variation that does not help separate them into extra, orthogonal components (ref 10) P7. A loading tells how strongly a descriptor weighs on the predictive component; R²Y is the paper's "goodness of fit" and Q² its "goodness of prediction" R16. In a score plot of the orthogonal components, a Hotelling T2 ellipse marks the 95% boundary of the model's space; a point outside it is an outlier R18. The paper prints no equations; none are needed to follow it.

Methods in detail

The work has three parts: spectra of the dry dispersions to show that each drug mixes with PVP, spectra after humid storage to see which compositions separate, and a discriminant model over a 20-system data set. The flow below summarises them.

Part 1: five drugs chosen by their hydrogen-bond donor

Problem. If interaction strength decides whether a dispersion survives moisture, the test set has to span a range of interaction strengths with one polymer. What they did. The five drugs (Figure 1) cover the main donor types C7·M1·M2·M3·M4·M5. Indoprofen carries a carboxylic acid (calculated pKa 5.8, log P 2.84) R15.e. Clofoctol has one phenolic OH and probucol two, and both are very hydrophobic (pKa 10.28 and 10.59; log P 8.19 and 10.0) R15.f·R15.g. Nilutamide donates from an imide NH (pKa 15.01, log P 2.14) R15.s. Loratadine has no conventional donor; its carbamate carbonyl is an acceptor, and its crystal structure shows only weak C–H···O=C contacts (CSD code BEQGIN, ref 14) (log P 4.13) R15.t·R7. The polymer was PVP K29-32, dried over phosphorus pentoxide for at least a week before use M6.

Each drug and PVP were dissolved together in a common solvent (clofoctol in a 1:1 w/w dichloromethane–ethanol mixture, probucol in ethanol, indoprofen and loratadine in chloroform, nilutamide in isopropanol), checked visually for a single clear solution, spin-coated onto ZnS or KRS-5 windows at 500/2,500 rpm for 18 and 30 s, and kept for at least 1 min on a 90 °C hot plate to remove solvent P1. Compositions were 10, 30, 50, 70 and 90% PVP by weight, with pure PVP and pure amorphous drug films as references P1·P4. Transmission spectra were collected from 500 to 4,000 cm⁻¹ at 4 cm⁻¹ resolution with 128 scans, keeping the absorbance of the band of interest between 0.6 and 1.0 and purging with dry air below 10% RH P2. For indoprofen, nilutamide and loratadine the spectrum of a 50% physical mixture of amorphous drug and PVP was calculated for comparison P5.

Part 2: humid storage with re-drying

Problem. Sorbed water could mask changes in the interaction bands, and spectral changes from separation have to be told apart from crystallization. What they did. Films of each composition were stored in desiccators over saturated salt solutions giving 54, 75, 84 and 94% RH (Mg(NO₃)₂, NaCl, KCl, KNO₃); the title of Table I gives the temperature as 22 °C P3. At intervals the films were taken out and dried in the spectrometer with dry air until the OH region stopped changing, and only then measured, so that "absorbed moisture would not mask" changes in the interaction bands P3·C8. The spectra were compared with reference spectra of the amorphous and crystalline drugs, the latter measured by ATR on the drugs as received, to separate amorphous changes from crystallization P4·C17. As a second check, 50:50 drug–PVP powders made by rotary evaporation were measured by PXRD after preparation and after 18 and 42 h at 84% RH P6·R8. The storage time at each humidity is not given, and it is not stated whether one film went through the RH series or a fresh film was used for each humidity X7.

Part 3: a 20-system data set and a discriminant model

Problem. The authors wanted the properties that matter most across all systems their group had studied, not only the five new ones. What they did. Table II combines the five new PVP systems with 15 systems from the group's earlier studies: indomethacin, quinidine, pimozide and felodipine each with PVP, PVPVA and HPMAS (the polymer the text calls HPMCAS), ketoprofen, droperidol and nifedipine with PVP R15. Each row carries molecular descriptors (molar mass from ref 24, hydrogen-bond counts from PubChem, pKa calculated with MarvinSketch unless marked, log P estimated with ChemBioDraw, log S from refs 26 and 4, Tg and ΔCp from DSC of the quench-cooled drug, Tm and ΔHf from refs 28 and 7), the PVP carbonyl shift where available, and the outcome class R15·P8·P9. Seven of the 20 rows are class 1 (no separation): indomethacin with all three polymers, ketoprofen–PVP, and quinidine, pimozide and felodipine with HPMAS R15.a·R15.b·R15.c·R15.d·R15.j·R15.n·R15.q.

OPLS-DA was run in SIMCA-P+ v.12.0.1 P7. The numbers of donors and acceptors, molar mass, log P, log S and Tg were dropped "since these variables made insignificant contributions", and donor and acceptor types entered as categories R16. The final model had one predictive and two orthogonal components R16. The paper does not say which of the 20 rows entered the model or how the N/A entries were treated; the score plot (Figure 12) shows 12 distinguishable markers, and the caption speaks of "three systems which remain miscible", although seven rows are class 1 X10·R18.

Experimental analysis

Main result 1: all five drugs mix with PVP when dry. All films were transparent, and the bulk powders gave a PXRD halo, which the authors read as amorphous ("suggesting") R1·C9. Table A lists the infrared evidence of mixing. For clofoctol a new band at 1,663 cm⁻¹ grows beside the free PVP band as the PVP content falls and is the only carbonyl band left at 10% PVP (Figure 2); the clofoctol OH band also moves from 3,365 to 3,263 cm⁻¹ and broadens R2·C11. Probucol gives the same 1,682 → 1,663 cm⁻¹ shift, "consistent with" hydrogen bonds reported earlier (Broman 2001, ref 12) R3·C12. Indoprofen develops a shoulder at 1,641 cm⁻¹, assigned to PVP carbonyl bonded to its COOH, and the authors state that the 50% dispersion spectrum "clearly" cannot be built from the pure amorphous spectra (Figure 3) R4·C13. Nilutamide develops a shoulder at 1,663 cm⁻¹, assigned to PVP bonded to its NH, while its own carbonyl bands at 1,787 and 1,729 cm⁻¹ do not move (Figure 4) R5·C14. Loratadine, with no donor, still gives a single merged band whose shape differs from the calculated physical mixture, which the authors take as intimate mixing (Figure 5) R6·C15.

Table A. Infrared evidence of drug–PVP mixing in the dry films (positions as stated in the text; Figures 2–5) R2·R3·R4·R5·R6.

Main result 2: humid storage splits the PVP-rich films of every drug. Table B reproduces Table I. At 70 and 90% PVP all five systems showed amorphous–amorphous phase separation after storage at increasing RH, and nilutamide and loratadine showed it at every composition R14. The signs differ by system. For nilutamide the bonded band at 1,663 cm⁻¹ fell and the free band at 1,682 cm⁻¹ grew gradually with RH at every composition, with no crystalline nilutamide (Figure 9) R12·C24. For loratadine the merged band at 1,686 cm⁻¹ (50% PVP) split into a peak at 1,692 cm⁻¹ and a shoulder at 1,675 cm⁻¹, and after 94% RH the spectrum resembled the calculated physical mixture; the authors say this "can be interpreted as evidence" of separation (Figure 10) R13·C25. For probucol at 50% PVP the 1,663 cm⁻¹ shoulder fell gradually from 54 to 94% RH, and the authors read the probucol results as "again" suggesting separation at 50% PVP or more (Figure 7) R10·C21.

Table B. Phase behaviour after storage at 54–94% RH and 22 °C, from Table I. NC: no change in the spectrum after storage at 94% RH; AAPS: evidence of amorphous–amorphous phase separation after storage at increasing RH R14.

For clofoctol at 50% PVP the text reports that the 1,663 cm⁻¹ band lost intensity relative to the 1,682 cm⁻¹ band after 54% RH and "kept decreasing" at 75, 84 and 94% RH, with overlap shifting the band toward 1,682 cm⁻¹ R9·C19. In Figure 6 the shoulder is lowest after 54% RH, intermediate after 75% RH and close to the initial height, with a broadened band, after 84 and 94% RH (read from the figure) R9. Note by the AI writer (not discussed in the paper): the separation after 54% RH is visible in the plot, a steady further decrease is not X9. No crystalline clofoctol (bands at 854 and 1,238 cm⁻¹) appeared R9.

The PXRD check gave amorphous patterns for the "selected" bulk dispersions after up to 42 h at 84% RH, which the authors call "in good agreement" with the infrared results R8·C18. The samples are not named, and the infrared data show indoprofen crystallizing at 84% RH or below at 10–50% PVP, so the PXRD result covers only part of Table B X8.

Main result 3: the two most hydrophobic drugs resist, but only at high drug loading. Clofoctol films with 30 and 10% PVP gave spectra after storage up to 94% RH that were identical to the initial ones; at 30% PVP the bonded band near 1,662 cm⁻¹ is the main band before and after (Figure 6, top) R9·C20. Probucol changed much less at 30% PVP than at 50% and not at all at 10% PVP R10·C21. The authors note that clofoctol and probucol (log P 8.19 and 10.0) form moderate hydrogen bonds with PVP, as phenols should ("it can be inferred"), and separate at some ratios, and they explain the resistance at high loading by water uptake: "it is thought that insufficient moisture is sorbed by the systems" because the drugs are so hydrophobic R15.f·R15.g·C35·C36. Water uptake was not measured in this paper X6. Nilutamide and loratadine, less hydrophobic and with weak or moderate interactions, separated at every composition "as expected" C37. The authors draw the general point that "not only drug and polymer properties, but also the relative proportions" of the components must be considered C38.

Main result 4: indoprofen separates slightly and then crystallizes. At 50% PVP the free PVP band grew slightly after 54 and 75% RH while the bonded band at 1,641 cm⁻¹ stayed visible, and no crystals were detected; after 84 and 94% RH the 1,641 cm⁻¹ band dropped abruptly and a sharp band of crystalline indoprofen appeared at 738 cm⁻¹ (Figure 8) R11·C22. With 30% PVP crystals appeared after 75% RH, and with 10% PVP after 54% RH, the lowest humidity used, so the authors could not say whether separation came first R11·C23. At 70 and 90% PVP the films separated without crystallizing up to 94% RH R11. Because some bonding survived, the authors conclude that indoprofen "seems to be" a carboxylic acid that can separate to a small extent C46.

Main result 5: across the data set, carboxylic acids with large shifts stay mixed in PVP. Table C lists the 12 PVP systems of Table II. The acids shift the PVP carbonyl band by 41–44 cm⁻¹; the OH and NH donors by 15–24 cm⁻¹; loratadine is listed at −17 cm⁻¹ R15. In PVP only indomethacin and ketoprofen resist; every OH, NH or no-donor drug separates, and so does indoprofen R15. The authors describe two groups, acids above 40 cm⁻¹ and OH or NH compounds "around 15–20 cm⁻¹", and an "excellent correlation" between shift and outcome with indoprofen as the exception C31·C32. They explain the exception "most likely" by indoprofen's higher pKa: 5.8 against 4.5 for indomethacin and 5.02 for ketoprofen C33·R15.a·R15.d·R15.e.

Table C. The 12 PVP systems of Table II, ordered by PVP carbonyl shift; * = measured in this paper. pKa calculated, log P estimated, as in the Table II footnotes R15.

Three details of Table II deserve a reader's attention. The text gives the PVP band moving from 1,682 to 1,663 cm⁻¹ for both clofoctol and probucol (19 cm⁻¹), while the table lists 18 and 15; the −17 cm⁻¹ for loratadine cannot be derived from the positions in the text X1. Quinidine–HPMAS, a dispersion without PVP, carries a "PVP carbonyl shift" of 19 cm⁻¹ and is class 1, so the stated correlation holds for the PVP rows only; droperidol at 24 cm⁻¹ also sits outside "15–20" X2·X12. And the ΔCp column is labelled J/molK although its values of 0.33–0.64 fit J/(g K) for these molar masses X4. Across polymers, all four HPMAS systems resist, and among PVPVA systems only indomethacin does R15.b·R15.c·R15.i·R15.j·R15.m·R15.n·R15.p·R15.q.

Main result 6: the model loads on the acid donor, the shift and pKa. The OPLS-DA model gave R²Y 0.98 and Q² 0.73, which the authors call "indicative of a good model" R16·C39. Table D gives the loadings of the predictive component read from Figure 11. The COOH donor (about 0.53) and the carbonyl shift (about 0.40) load positively, pKa (about −0.35) and the NH donor (about −0.28) negatively, and their error bars stay clear of zero R17. The authors read this as: a lower pKa, a larger red shift and a COOH donor favour a system that stays miscible; NH and OH donors favour separation; a lone CO acceptor favours stability C40. A lower ΔCp is "also a property common" to resistant systems, for reasons "at present, unknown", and fluorine "seems to" add to separation, possibly ("it can be speculated") through its effect on hydrophobicity C41·C42. The error bars of the CO, ΔCp and N, O, F loadings span or touch zero, and their meaning is not stated X11.

Table D. Loadings of the OPLS-DA predictive component, read from Figure 11 (approximate) R17.

The orthogonal components describe differences within each class: resistant systems lie along one line through the origin of the score plot and non-resistant ones along another, and two orthogonal components were needed to predict behaviour accurately C43·R18. Indoprofen–PVP sits at about (2.9, 2.85), on the edge of the 95% ellipse and away from the resistant systems (Figure 12) R18. The authors call it an outlier: it shares many features with resistant systems but also "key properties" of separating ones, it does not fit the resistant class "because of the presence of both CO and NH bond acceptors", and its missing ΔCp, hard to measure for a drug that crystallizes rapidly, "may be a reason" this variable matters C44·C45. Table II lists indoprofen's acceptors as "CO, N" and NH only as a donor type, so the wording in the text does not match the table X3.

Discussion

How the authors connect the findings. The argument runs as a balance between two things: how strongly the drug holds on to PVP, and how much water reaches the dispersion. The chain below follows the authors' reasoning C30·C32·C36·C37.

Indoprofen sits between the branches: a carboxylic acid with a 41 cm⁻¹ shift that still separated slightly and then crystallized, which the authors put down "most likely" to its higher pKa C33·C46. Their summary in the conclusions is that three characteristics "appear to result" in resistance: minimal hygroscopicity of the dispersion (itself a function of polymer, drug, their interaction and the RH), a less hydrophobic drug, and strong drug–polymer interactions C1. The abstract states the same more firmly ("In general, it was observed"), and says that OPLS-DA confirmed the importance of these properties C1·C2.

Where the conclusions apply. The infrared data cover five drugs in PVP K29-32, cast as thin spin-coated films, stored at 22 °C and 54–94% RH and measured after drying P1·P3. The model covers 20 systems with three polymers, 15 of them from earlier work R15. Nothing in the paper tests what the separated regions do afterwards; the link to faster crystallization and to dissolution is cited (refs 3, 15, 16) C27.

What the paper suggests for formulation. The authors recommend checking drug–polymer mixing after moisture exposure and not only after manufacture, and considering "the interplay" between interaction strength, hygroscopicity and drug hydrophobicity when designing dispersions and accelerated stability tests C26·C3. Note by the AI writer (not discussed in the paper): in PVP, the type of hydrogen bond a drug can donate decides most of the outcome in this data set, so a dry-film infrared measurement of the PVP carbonyl shift is a cheap first screen. A shift near 15–24 cm⁻¹ from an OH or NH donor went with separation in every PVP system listed, and a shift above 40 cm⁻¹ from an acid went with resistance in two of three R15. The resistance of the very hydrophobic drugs is tied to high drug loading and, in the authors' reading, to low water uptake; a lower loading or added hygroscopic excipients could remove it R14·C36.

Limitations

Limitations stated by the authors. For indoprofen at 10% PVP, crystals appeared at the lowest humidity used, so "definite conclusion" on whether separation came first "cannot be made" C23. The basis of the link between a lower ΔCp and resistance is "at present, unknown" C41. The fluorine effect is offered only as something that "can be speculated" C42. The missing ΔCp of indoprofen "may be a reason" for the significance of that variable C45. The PVP-rich dispersions are described as "highly vulnerable", but the authors also stress that the susceptibility "is highly dependent on the specific API used" C29.

The AI writer's own judgement (1): only one of the three factors is tested here. The abstract says OPLS-DA confirmed the importance of interaction strength, hygroscopicity and hydrophobicity. Table II has no hygroscopicity or water-uptake variable, and log P was dropped from the model as insignificant X5·R16. In this paper's own data the direction of the hydrophobicity factor is also not supported: the only compositions without change belong to the two most hydrophobic drugs, clofoctol (log P 8.19) and probucol (log P 10.0), while the less hydrophobic nilutamide (2.14) and loratadine (4.13) separated at every composition X14·R14·R15. The authors' explanation for the hydrophobic drugs, that too little water is sorbed, is reasonable but untested, because no water uptake was measured X6. Read strictly, the paper supports the interaction factor; the other two rest on the group's earlier work.

The AI writer's own judgement (2): the model and its inputs. The rows that entered OPLS-DA and the handling of N/A entries are not stated; the score plot shows 12 markers for a 20-row table, so R²Y 0.98 and Q² 0.73 cannot be reproduced from the paper X10. The class labels summarise each system with one value, although for clofoctol, probucol and indoprofen the outcome depended on composition X13. Some inputs are inconsistent: the probucol shift is 15 cm⁻¹ in the table but 19 cm⁻¹ by the text, quinidine–HPMAS carries a PVP shift, and the ΔCp unit is doubtful X1·X2·X4. Several of the loadings the authors interpret (the lone CO acceptor, ΔCp, fluorine) have error bars that reach zero X11.

The AI writer's own judgement (3): the experimental record. Storage times at each humidity are not given, and it is unclear whether one film was moved through the series or fresh films were used, so the RH comparisons may mix humidity and time X7. The PXRD "agreement" covers unnamed bulk samples after 42 h at most, while the infrared data show indoprofen crystallizing within the storage series X8. For clofoctol at 50% PVP the spectra after 84 and 94% RH do not show the further decrease described in the text X9. The infrared evidence itself is direct and well controlled with calculated physical mixtures and crystal reference bands, which is why the five-system result (Table I) is the strongest part of the paper R14·C17.

Conclusion

The paper shows, for five drugs in PVP, that absorbed moisture split every system into drug-rich and polymer-rich amorphous regions at 70 and 90% PVP, and nilutamide and loratadine at every composition R14. Only the very hydrophobic clofoctol and probucol resisted, and only at high drug loading R14. Across 20 systems, carboxylic acids that shift the PVP carbonyl band by more than 40 cm⁻¹ were the ones that stayed mixed in PVP, with indoprofen as a partial exception, and a discriminant model put the weight on the acid donor, the shift and pKa R15·R17·C40. The authors' wider conclusion, that low hygroscopicity, low hydrophobicity and strong interactions "appear to result" in resistance, is supported by this paper mainly for the interaction part C1·X5·X14.