Why ritonavir–PVPVA dispersions stop releasing at higher drug loading: dispersed droplets versus a drug-rich surface layer

Background and motivation

An amorphous solid dispersion (ASD) molecularly disperses a poorly soluble drug in a polymer, which can raise both the apparent drug concentration in solution and the release rate (Section "Introduction", PDF p.3) C6. For some ASDs the drug and the polymer leave the solid at the same normalized rate. This is called congruent release, and under non-sink conditions it gives a solution above the amorphous solubility with a second, drug-rich phase of nano-sized droplets ("nanodroplets"), which earlier in vitro and in vivo studies suggested can improve absorption C6. The problem the paper starts from is that this good behaviour is usually lost as the drug loading (DL) rises. For dispersions in polyvinylpyrrolidone/vinyl acetate (PVPVA, Kollidon VA 64) a "falling-off-the-cliff" effect is "often observed": release is complete and congruent only at relatively low DL, and above a certain DL drug release drops sharply and becomes incongruent, with the polymer "usually" released faster than the drug C7. The DL at which congruency is lost is the limit of congruency (LoC); for ritonavir (RTV) in PVPVA, an earlier study by the same group reported an LoC of 25% DL (Indulkar et al. 2019) R15. Enteric polymers show a more gradual decline with DL C7.

What causes the cliff was not settled. Many studies linked poor release to water-induced amorphous–amorphous phase separation (AAPS), in which an initially miscible ASD separates into an insoluble amorphous drug-rich phase and a soluble polymer-rich phase once water enters, and whose kinetics and morphology depend on DL C8. Extensive AAPS at high DL was associated with poor release, and systems that resist AAPS, or dissolve faster than they phase separate, with better release C9. Yet Li and Taylor (2019) found that discrete drug-rich domains created by water-induced AAPS before dissolution gave higher release from lopinavir films than initially miscible films C9. The authors therefore describe the link between AAPS and release as "convoluted", with both kinetics and microstructure "likely crucial" C9.

The paper sets out to follow phase separation during hydration of RTV–PVPVA ASDs at several DLs and to correlate drug release with the evolving morphology at the compact–water interface C10. The working hypothesis is stated plainly: a continuous drug-rich barrier at the ASD surface is detrimental to release and appears at the DL where the cliff occurs ("We hypothesized") C10. The tools are surface normalized dissolution of compacts, confocal fluorescence imaging of compact surfaces and cross-sections after partial dissolution, a newly developed in situ (real-time) confocal method, and X-ray photoelectron spectroscopy (XPS) of the compact surface C5.

Background concepts

Congruent release and the normalized release rate. Congruency compares the release of drug and polymer after correcting for how much of each the ASD contains. The paper defines the normalized rate of release as (Equation 1, PDF p.6) E1:

Here $k$ is the slope of the linear regression of total concentration against time, $V$ the volume of dissolution medium, $A$ the exposed surface area of the compact and $x$ the weight fraction of the component of interest E1. In plain terms: $R$ is the mass of a component released per unit area per unit time, divided by that component's share of the ASD, so drug and polymer released at the same $R$ leave the compact in the same ratio as they were mixed. The paper defines $R$ but reports no values of $k$ or $R$; congruency is judged from the percentage-release curves of Figure 2 X12.

Amorphous–amorphous and liquid–liquid phase separation. In AAPS a drug–polymer glass that is a single phase when dry splits into two amorphous phases when water is added: one rich in the hydrophobic drug and one rich in the water-soluble polymer C8. Liquid–liquid phase separation (LLPS) is the corresponding split in a water-rich solution, where the drug concentration exceeds the amorphous solubility and drug-rich droplets form. A central argument of this paper is that the two are parts of one process described by a single drug–polymer–water phase diagram, occurring at different water contents, so that at the boundary between a hydrated gel and the solution the distinction blurs C51. Two boundaries of the phase diagram appear in the discussion: the binodal, the composition at which a mixture can first split into two phases, and the spinodal, beyond which it splits spontaneously C34·C57.

Gel layers in polymer dissolution. The authors use the established picture of how a soluble glassy polymer dissolves (refs 34, 36–46) C41. Water penetrates and plasticizes the polymer, converting the glassy matrix into a rubbery gel layer, with a sharp interface between the inner glassy core and the gel. Polymer chains then disentangle at the gel–solvent interface and diffuse into the medium. If disentanglement is slow relative to water penetration, the gel layer forms or grows and becomes a barrier to water; a steady state is reached when the two rates are equal, and steady-state gel layers of swellable matrices are a few hundred micrometres to a few millimetres thick C41·R20. In an ASD the gel is a mixture of drug, polymer and water whose water content increases from the glassy core towards the solution C44.

Phase morphology and phase inversion. When two phases form, the one present in smaller volume tends to become the dispersed (discrete) phase and the larger one the continuous phase; as the share of the minority phase grows, the continuous phase can switch, which is called phase inversion C46. The authors draw on the literature of polymer blends and membrane formation, where the morphology depends strongly on the starting composition and the path into the two-phase region, and the low-composition component is "more likely" to be the discrete phase (refs 53–61) C59.

Reading the fluorescence images. Two trace dyes were added to the ASDs: Nile red at 0.02% w/w and Alexa Fluor 488 at 0.001% w/w P1. With a clogP of 4.62, Nile red "is expected" to follow the drug-rich phase, and Alexa Fluor 488, with a clogP of −8.74, the polymer/water-rich phase C11·M3·M4. For RTV–PVPVA, an earlier study by the group showed that Nile red partitions preferentially into drug-rich domains (Yang et al. 2022) R17. In two-channel images, mixed (unseparated) material appears yellow/orange, drug-rich regions red and polymer/water-rich regions green C25. Water also quenches Nile red fluorescence (ref 35), so a hydrated, unseparated gel looks green in the real-time images R18·C33.

Methods in detail

The study has one drug–polymer pair and one main variable, the drug loading, which runs across the reported LoC: 10, 20, 25, 30, 40 and 50% DL, with neat amorphous RTV and neat PVPVA as controls A1·A2·A3·A4. Every experiment starts from the same kind of compact, so that release, morphology and surface chemistry can be compared loading by loading. The workflow is shown below.

Making the dispersions and compacts

RTV (ChemShuttle) and PVPVA (Kollidon VA 64, BASF) were dissolved in methanol at 10 wt% solids and dried in a rotary evaporator with a 40 °C water bath, then kept under vacuum at room temperature overnight, milled and sieved to the 60–100 mesh fraction (149–250 µm) P1·M1·M2. ASDs for imaging also contained the two dyes, added to the stock solution P1. Polarized light microscopy showed no crystallinity in any formulation before the experiments R3. For every test, 100 mg of powder was compressed in a circular die of 8 mm diameter at 1500 psi for 60 s P2.

Release test

The die holding the compact was mounted on an overhead stirrer and rotated at 100 rpm in 100 mL of pH 6.8, 50 mM phosphate buffer at 37 °C, so only one face of the compact was exposed P2. At each time point 1 mL was withdrawn and replaced, and the samples were deliberately not filtered, so that drug present as nanodroplets is counted as released together with dissolved drug P2. RTV was assayed by reversed-phase HPLC (C18 column, 45% acetonitrile) and PVPVA by aqueous size-exclusion chromatography with UV detection at 205 nm; each measurement was done in triplicate P3·P4·P2. Release was tested at 20, 30, 40 and 50% DL and for the two neat components A1.

Imaging after partial dissolution

Dye-containing compacts were dissolved in the same way and removed at set times between 5 and 30 min P5. Each compact was ejected from the die; its surface was dried with a gentle stream of dry air and imaged in the Nile red channel, or the compact was cut in half with a razor blade and the cross-section imaged in both channels P5. A Nikon A1Rsi confocal microscope excited Nile red at 561 nm (emission 570–620 nm) and Alexa Fluor 488 at 488 nm (emission 500–550 nm) P6. On the cross-sections the authors also measured how far the boundary between the glassy core and the gel layer had moved: the depth penetrated at a given time is the initial compact thickness minus the thickness of the remaining glassy core, expressed as a percentage of the initial thickness to correct for different compact thicknesses; n = 3 P10. The authors use the movement of this gel–glass interface as their measure of how fast water penetrates C28.

Real-time imaging during hydration

The new element of the paper is an in situ confocal set-up C5. A dye-containing compact was ejected after compression and fused to the cover-glass bottom of a 35 mm dish by heating at 180 °C for 30 s; its top was sealed with vacuum grease so that water could enter only from the edge P7. After 8 mL of pH 6.8 buffer was added in a chamber at 37 °C, images were taken every 30 s, looking at the edge of the compact in cross-section as the water front moved in P7. The authors are explicit about the limits of this set-up. Agitation was supplied by hand, by gently removing and re-adding medium with a pipette, so the hydrodynamics were neither controlled between samples nor comparable with the release test; and if the fused contact with the glass was imperfect, water could also reach the compact from below C31. Rates of hydration and dissolution in these runs therefore differ from the release test and "cannot be directly compared" between samples; what the authors interpret is the sequence of phase changes at each loading C31.

Surface chemistry and crystallinity

XPS measured the surface composition of 40% DL compacts before and after 30 min of dissolution (monochromatic Al Kα, 1486.6 eV) P9. The C 1s peak was fitted with the reference C 1s spectra of pure RTV and pure PVPVA, the atomic percentages were converted to a weight ratio, and the error bars are standard deviations of five measurements on a single compact P9·R14. To rule out crystallization as a cause of poor release, gel layers were scraped off compacts after 30 min (20% DL) or 60 min (30–50% DL) of dissolution and examined by polarized light microscopy P8.

Experimental analysis

Release: the cliff lies between 20 and 30% drug loading

Figure 2 gives the release profiles. At 20% DL the drug and polymer curves overlap throughout: about 18% of each after 5 min, about 64% after 30 min and about 97% after 60 min, so the compact released completely and congruently within about an hour R2.a–b·C12. At 30% DL the polymer led from the start; at 90 min about 69% of the PVPVA but only about 35% of the RTV had been released R2.c–d·C13. The text gives the drug figure as "only 30% of the total dose", whereas the plotted mean reads about 35% (SD about 6) R1·X1. At 40 and 50% DL almost nothing came out: at most about 3% of the drug and about 14% (40% DL) or 10.5% (50% DL) of the polymer by 90 min, which the text describes as "no drug release and very minimal polymer release" R2.e–h·C14·X2. Neat amorphous RTV released at most about 0.5%, which the authors expect from its very low amorphous solubility, and neat PVPVA dissolved quickly, about 89% in 30 min R2.i–j·C16. No crystals were found in gel layers scraped off after up to 60 min, so crystallization does not explain the drop R3.

Table A. Release at 90 min, read from Figure 2 (approximate; mean of n = 3; neat PVPVA plotted to 60 min only) R2·R1.

The authors say these results agree with the earlier LoC of 25% DL for this system C15·R15. Two points qualify that agreement. First, release was not measured here at 10 or 25% DL; the 10% DL result (congruent, rapid) is cited from the earlier study, so in this paper's own data the switch lies between 20% and 30% DL R16·X7. The abstract's phrase "complete and rapid drug release below 30% DL" therefore rests on one measured loading plus cited data C2·X7. Second, the discussion supports the same agreement with reference 2, a general review of solid dispersions, whereas the results section cites reference 8; this looks like a misnumbered citation X3.

The dissolving surface: from fine droplets to a continuous drug-rich cover

Figure 3 shows Nile red images of compact surfaces after partial dissolution. The neat RTV and neat PVPVA controls were uniformly red at 5 and 20 min, which the authors take to mean that the dye does not segregate by itself, so red contrast in the ASDs is "consistent with" phase separation R4.m–p·C19. At 10, 20 and 25% DL the surfaces showed small discrete drug-rich domains at both time points, giving a diffuse appearance R4.a–f. At 30% DL the surface was heterogeneous at both 5 and 30 min, with patches of continuous drug-rich layer (solid red) next to regions of small red dots R4.g–h. At 40 and 50% DL an interconnected, porous drug-rich phase covered the surface after 5 min, and after 30 min the whole surface was a drug-rich layer R4.i–l. The authors set these patterns against the release behaviour: discrete domains below the LoC, a mixed surface where release is incongruent and incomplete, a continuous cover where no drug is released C20. The text and caption disagree on one detail: the text says the 10–25% DL surfaces were imaged at 5 and 20 min, but the caption gives 30 min for the second 25% DL image; 25% DL, equal to the cited LoC, is also grouped "below" it X4·X14.

Cross-sections: core, gel and a drug-rich crust

Figure 4, a Nile red image of a 30% DL compact after 5 min, shows the structure that frames the rest of the paper: a glassy core with particle boundaries and cracks; above it a homogeneous gel layer without particle boundaries, containing circular voids that are "likely" trapped air; and on top a thin drug-rich layer, 5–10 µm thick, continuous on the water side and more porous just below R5·C21·C22. The gel layer, in the authors' reading, forms by plasticization of the polymer by water, which lowers the glass transition temperature of the matrix C23. At all loadings from 10 to 40% DL the gel layer had a well-defined boundary to the glassy core and no grain boundaries, which the authors take to show that the gel is above its glass transition, while the core may contain some water but not enough to become rubbery R7·C42.

Figure 5 adds the second dye. At 10% DL, very small red droplets of 500 nm to 2 µm sat in an otherwise homogeneous gel at the compact surface after 5 and 10 min (enlarged in the insert of Figure 5a), and by 20 min the compact was almost dissolved R6.a–c. At 20% DL the picture was the same, with some larger red and green regions only in the last remnant at 20 min R6.d–f. At 30% DL a continuous drug-rich layer was already present at 5 min; by 20 and 30 min it was thicker but no longer continuous, with green polymer/water-rich regions carrying red droplets next to regions under a thick red layer R6.g–i. The authors expect the green regions to release polymer and the embedded nanodroplets, and relate the morphology to the faster release of polymer than drug, "since majority of the drug is stuck in the thick drug-rich layer" C26. At 40% DL a continuous orange-red surface layer was present at all three times and was about 12 µm thick after 30 min, with green polymer/water-rich regions beneath it R6.j–l·C27. Neat PVPVA cross-sections stayed uniformly orange, so colour segregation in the ASDs is attributed to phase separation R6.m–o·C24. From these images the authors conclude that, at low DL, nanodroplets form only at the surface of the compact and not deeper in the hydrated gel ("it is clear") C49.

Water penetration slows where the crust forms

Table B. Depth reached by the gel–glass interface, % of the initial compact thickness, read from Figure 6 (approximate; mean of n = 3; "–" = no point plotted) R8.

For neat PVPVA and the 10 and 20% DL ASDs the gel–glass front advanced at the same, roughly linear rate: about half of the compact was penetrated after 10 min and all of it after 20 min R8.a–c. The authors read this as evidence that, below the LoC, the drug and any drug-rich domains in the gel do not slow water entry ("suggests") C29. At 30 and 40% DL penetration was slower and stalled after 20 min, at about 25–29% and about 17% of the compact thickness R8.d–e. The authors attribute the plateau to the hydrophobic drug-rich layer on the surface, given the very low affinity of RTV for water compared with PVPVA, and conclude that the layer "effectively acts as a barrier" to further water ingress C30. Water content itself was not measured; the front position is a proxy C28.

Real-time imaging: where the droplets form and where they stop

Neat PVPVA served as the control (Figure 7). In the first minute the compact grew longer, which the authors explain by swelling before the chains could disentangle; it later shrank as the polymer dissolved R9·C32. A sharp orange–green boundary marked the water front; behind it the dry compact was unchanged, in front of it the gel looked green because water quenches Nile red, and the green faded towards the solution as the dye left with the dissolving polymer R9·C33. The gel stayed homogeneous R9.

At 10% DL (Figure 8) small red droplets appeared close to the gel–solution interface from the first minute and kept forming there as the water front moved into the compact; they entered the solution together with the dissolving polymer R10. The authors identify them as RTV-rich nanodroplets formed by AAPS once the drug–polymer–water system crosses its binodal composition, and link them to the congruent, rapid release of 10% DL ASDs reported earlier C34·R16. That the droplets formed mainly near the gel–solution interface, not throughout the gel, "can be rationalized" by a minimum water content needed to cross the binodal C35. At 20% DL (Figure 9) the same process occurred, but droplets formed deeper in the gel and the clear zone between the water front and the first droplets was shorter, from which the authors infer that a higher water content is needed to cross the binodal at 10% DL than at 20% DL R11·C36. The text points to "Figure 7g and Figure 8g" at 20 min for this comparison; the enlargements are in fact Figures 8g and 9g, both of the 15 min frame X6.

At 30% DL (Figure 10) a continuous but porous RTV-rich layer formed at once at the gel–solution interface; it was not rigid, since air bubbles escaped through it at several times, and green colour appeared beyond it R12. The authors infer that polymer still left through the pores while RTV trapped in the continuous layer did not, which "explains why only partial release of drug is observed" C37. They also state that water ingress was much slower than at lower loadings, a comparison between real-time runs that, by their own caveat, cannot be compared directly; the cross-section measurement of Figure 6 is the controlled comparison C37·C31·X9. The same paragraph cites Figures 3e–f and 5e–f, which show 25% and 20% DL; the 30% DL panels are Figures 3g–h and 5g–i X5.

At 40% DL (Figure 11) red and green regions appeared at the compact–solution interface immediately after the medium was added, and a thin continuous red layer then covered the whole interface with all green regions beneath it R13. The layer swelled and thickened with time and looked more rigid and non-porous than at 30% DL, and no green appeared beyond it; the authors conclude that no Alexa Fluor 488, "and by extension, polymer", could leave the matrix R13·C38. The phase-separated region reached 50–100 µm into the matrix R13·C52. In the release test, however, a 40% DL compact still released about 10% of its polymer by 45 min and about 14% by 90 min, so the dye-based inference should be read as strongly reduced polymer release rather than none R2.f·X8.

Table C. Release and morphology at each drug loading across the four methods R2·R4·R6·R10·R11·R12·R13·R16.

The surface becomes drug-rich

XPS on a 40% DL compact (Figure 12) gave 35 wt% RTV at the surface before dissolution, close to the composition of a homogeneous ASD, and 72 wt% after 30 min in buffer (SD about 10) R14. The authors read this as surface enrichment of drug during incongruent release, as reported for other ASD compacts, caused by preferential release of polymer and consistent with the drug-rich layer seen by confocal microscopy at this loading ("suggests"; "is due to") C39·C40. The values come from five measurements on one compact per condition X13.

Discussion

The authors' mechanism

The discussion assembles the observations into one chain. Water enters the compact faster than polymer chains can disentangle, so a rubbery gel layer forms over the glassy core; for RTV–PVPVA, with a low-Tg drug, polymer dissolution goes through this gel route C23·C42·C43. Because the gel replaces a sharp solid–liquid boundary with a gradient of solids, and the water content is highest next to the solution, water-induced AAPS is "favored" at the gel–solution interface C44. What happens next depends on how much drug-rich phase forms, which the authors relate to the drug loading C46.

At low DL the drug is the minority, so it forms the discrete phase; the continuous polymer-rich phase is highly soluble in water, and as it dissolves the embedded drug-rich domains are released as nanodroplets, giving rapid congruent release C47. The authors present this as a rationale that "potentially" explains the release profiles C47. At high DL a drug-rich network forms at the surface within 5 minutes and deepens as water penetrates; because mobility is high at high water content and the wet glass transition of RTV is low (about 14 °C, cited), the drug-rich domains "can further agglomerate and coalesce" into a contiguous coating C52·C54·M1. This coating "cannot dissolve" because the bulk solution is saturated at the amorphous solubility, so it persists unless removed mechanically by shear or rapid polymer swelling C53. It impedes polymer dissolution by limiting the water that dissolution needs, and it "may serve" as a physical barrier to polymer release even where water gets through, which in the authors' view explains why poor drug release is often accompanied by poor polymer release C55.

The 30% DL compact sits between the two branches. Discrete drug-rich regions are released early and extensive drug-rich domains dominate later ("there appears to be a transition"), and the layer is porous enough for the polymer-rich phase and its embedded droplets to escape C56. The authors offer dynamic phase separation as a possible cause: the first polymer-rich phase still contains some drug and, as it takes up more water, its composition passes the spinodal and it throws out further drug-rich droplets ("can be caused") C57.

What sets the switch, according to the authors, is the relative volume of the two phases: the lower-volume phase is "anticipated" to be discrete and the higher-volume phase continuous, so phase inversion is "anticipated" as DL rises, with coalescence of the low-Tg drug-rich phase also contributing C46. They support this with the general rule from polymer blends and membrane formation that the low-composition component is "more likely" to form the discrete phase C59. Their overall conclusion is that correlating morphology with release "indicates" that a continuous drug-rich phase is "catastrophic" for release, and that the transition from congruent to incongruent release is "likely dependent" on the morphology after AAPS, in agreement with Han et al. (2019) C58. They regard the imaging, especially the real-time imaging, as "strong support" for this link in RTV–PVPVA C60.

How the picture fits earlier mechanisms

The authors place their result against three earlier accounts of nanodroplet formation. Harmon et al. (2016) proposed that drug-rich domains form inside ASD particles by preferential dissolution of the polymer and are released when a surfactant prevents their aggregation; here no surfactant was present, and the authors "propose" instead that drug-rich and polymer-rich domains form together by AAPS, only where the water content brings the system into the two-phase region C48. Li and Taylor (2019) described a "particle and glue" mechanism in which drug-rich domains are released when the polymer holding them dissolves; the cross-sections here show nanodroplets forming only at the compact surface, not deep in the gel C49. Indulkar et al. (2017) concluded from an isotope scrambling study that nanodroplets form in solution by LLPS once the amorphous solubility is reached; the authors argue that AAPS at the gel–solution interface, where the material is in contact with the bulk solution, "would presumably" also meet the conditions for isotope scrambling C50. This is where their view that AAPS and LLPS are one process at different water contents comes in C51.

What remains open

The authors name one question themselves: it is "unclear" why the LoC varies so much between drugs formulated with PVPVA, with values as low as 5% for some drugs C61·R19. If the drug-rich phase were mostly drug with little polymer, a 5% DL ASD could not form enough of it to cover the surface, so their phase-volume explanation does not obviously account for such low LoCs C61. They suggest that the imaging methods developed here "may provide insight" into other mechanisms of poor release for other drugs, and that dissolution of ASDs with high-Tg drugs or polymers "may potentially" follow other mechanisms, such as crack formation C62·C43. For modelling, they expect the insight to underpin first-principle release models built on two regimes: loadings where drug release is coupled to polymer release, and loadings where a drug-rich barrier layer controls release C63.

Note by the AI writer (not discussed in the paper). For formulation work the paper suggests a simple diagnostic: measure drug and polymer release together, because the polymer running ahead of the drug, as at 30% DL here, is the first sign that a drug-rich surface layer is forming R2.c–d·C13. For RTV–PVPVA the usable range is narrow, congruent at 20% DL and incongruent at 30% DL, and the loadings in between were not tested for release in this paper R2·X7.

Limitations

Limitations stated by the authors. The authors stress that the real-time experiments use manual agitation, uncontrolled hydrodynamics and possible wetting from below, so that their hydration and dissolution rates differ from the release test and cannot be compared between samples; release or morphology "may" differ between the two set-ups C31. They state that the reason for the large variation of the LoC between drugs is unclear C61 and that ASDs with high-Tg components may dissolve by other mechanisms C43.

The AI writer's own judgement (1): what the data show and what they do not. The release data and the three imaging methods agree well on where the drug-rich phase turns continuous, and the neat-component controls rule out dye artefacts R2·R4·R6·R10·R11·R12·R13. But the link between morphology and release is a correlation within one drug–polymer pair; no experiment changes the morphology at a fixed loading to show that the layer causes the drop C3. The quantities on which the explanations rest were not measured: the volumes and compositions of the coexisting phases, behind the phase-volume argument, and the drug concentration in the medium relative to the amorphous solubility, behind the claim that the layer "cannot dissolve" X10·X11. Release was measured at 20, 30, 40 and 50% DL only, so the switch is located between 20 and 30% DL in this paper's data, with the 25% LoC taken from earlier work X7.

The AI writer's own judgement (2): readings and wording to treat with care. Several statements in the text are stronger or differ from the plotted values: 30% DL drug release is given as 30% where the figure reads about 35%; 40 and 50% DL are described as "no release" where up to about 3% of the drug and 10–14% of the polymer are released; and polymer retention at 40% DL is inferred from the absence of dye signal, while the release test shows some polymer release X1·X2·X8. Some figure and reference citations are wrong (Figures 3e–f and 5e–f for 30% DL, Figures 7g and 8g for the 10% versus 20% DL comparison, reference 2 for the LoC agreement), and one time point differs between text and caption X3·X4·X5·X6. The XPS result rests on one compact per condition X13. Layer thicknesses from the release-test cross-sections (5–10 µm and about 12 µm) and from the real-time set-up (a phase-separated region of 50–100 µm) come from experiments with different hydration conditions and should not be compared directly X15.

Conclusion

The paper links the release cliff of ritonavir–PVPVA dispersions to what water does at the dissolving surface. Below the cliff (20% DL, and 10% DL from earlier work) the drug-rich phase formed at the gel–solution interface stays as droplets of about 0.5–2 µm that leave with the dissolving polymer, and drug and polymer are released together R2.a–b·R6.a·R10·R16. Above it (30% DL partly, 40 and 50% DL fully) the drug-rich phase becomes a continuous surface layer, water entry stalls, the surface becomes drug-rich and release stops R2.c–h·R4·R8·R14. The authors' interpretation is that release "can be rationalized" by this morphology and that the congruent-to-incongruent transition is "likely dependent" on it; the phase volumes and solution saturation that would explain why the switch falls where it does were not measured C3·C58·X10·X11. The new real-time imaging method is the paper's main tool contribution C5.