How water during preparation and during storage splits evacetrapib amorphous solid dispersions

Background and motivation

About 90% of new chemical entities and a third of the drugs in the US Pharmacopeia are poorly water soluble, and amorphous solid dispersions (ASDs), in which the drug is dispersed in a polymer in the amorphous state, are a popular way to raise their apparent solubility. Unlike some other solubilisation strategies, an ASD can do this without lowering the apparent membrane permeability (Miller et al., ref 14). The authors start from the widely held view that drug–polymer miscibility, meaning drug and polymer mixed at the molecular level in one amorphous phase, is essential both for physical stability in storage and for reaching a supersaturated solution on dissolution. C2 A phase-separated ASD is more prone to crystallization, because the drug-rich phase holds less polymer, and it "would be expected" to reach a lower maximum concentration during dissolution (Rumondor et al., ref 18; Hancock and Parks, ref 19). C1

Detecting phase separation therefore matters early in development, and high-resolution imaging (atomic force microscopy, AFM; scanning and transmission electron microscopy, SEM and TEM) can resolve heterogeneity below 100 nm. Earlier TEM work had shown heterogeneous features in felodipine–poly(vinylpyrrolidone) films (Marsac et al., ref 24), and electron energy-loss spectroscopy had quantified drug concentrations in submicron regions (Ricarte et al., ref 25). Techniques that combine imaging with infrared, Raman, mechanical, thermal or elemental analysis add information on the composition of the phases. C3

Two gaps motivated the study. First, solvent evaporation, and spin coating in particular, is a standard way to make ASD films for miscibility studies, yet day-to-day variation in drug–polymer miscibility had been reported in such films by the same group (Li and Taylor, ref 26) without a known cause; solvent composition and the preparation method were also known to affect ASD quality (Paudel and Van den Mooter, ref 33; Janssens et al., ref 34). C4 Second, ASDs were known to undergo moisture-induced phase separation on storage at high humidity (Qi et al., ref 20; Marsac et al., ref 24; Rumondor et al., ref 35), but detailed information on the microstructure and composition of the separated phases was lacking. C4 The goal was a mechanistic understanding of how water affects the phase behaviour and microstructure of ASDs prepared by solvent evaporation, comparing water introduced during preparation with water taken up during storage. A1·A3·A6

The model drug is evacetrapib (ECP), an inhibitor of cholesteryl ester transfer protein with extremely low aqueous solubility (Cao et al., ref 36), combined with two polymers widely used in commercial ASDs: hydroxypropyl methylcellulose (HPMC, E3 grade) and polyvinylpyrrolidone/vinyl acetate (PVPVA, Kollidon VA 64) (Figure 1). M1·M2·M3 For imaging the authors combine AFM topography, AFM with Lorentz contact resonance (LCR), which the group had used before on a telaprevir ASD, and TEM with energy dispersive X-ray spectrometry (EDX). C5

Background concepts

Miscibility and amorphous phase separation. A drug and a polymer are called miscible when they form a single amorphous phase. If they are not, or if a third component such as water makes them incompatible, the mixture separates into a drug-rich and a polymer-rich amorphous phase. Which phase is continuous and which forms isolated domains matters for later behaviour, as the discussion of dissolution in this paper shows. In this study miscibility is judged in three ways: a featureless AFM or TEM image (at a lateral resolution of about 40 nm for the AFM settings used), a single glass transition temperature ($T_g$) in differential scanning calorimetry (DSC), and, for separated films, the chemical identity of the domains. C8·P8·P13

Amorphous solubility. When a poorly soluble drug is added to water beyond a certain concentration, it separates as a drug-rich liquid phase rather than crystallizing at once. The concentration at which this liquid–liquid phase separation (LLPS) begins is the "amorphous solubility", the upper limit of supersaturation an ASD can reach. For evacetrapib in pH 6.8 phosphate buffer the crystalline solubility was 33 ± 7 ng/mL and the LLPS onset 481 ± 98 ng/mL, so the drug is extremely hydrophobic and water is a strong antisolvent for it. M1·R1·C6

Phase separation during solvent evaporation. Polymer membrane science uses solvent evaporation deliberately to make porous films. In one approach the polymer is dissolved in a mixture of a volatile good solvent and a less volatile poor solvent; as the good solvent evaporates, the polymer separates. In vapor-induced phase separation (VIPS), a water-insoluble polymer is cast from a slowly evaporating, water-miscible solvent and placed in humid air: if water is absorbed faster than the solvent evaporates, the polymer separates (Figure 11, adapted from Li et al., ref 45). C35 The authors argue that an ASD solution in methanol behaves in a similar but more complex way, because it contains more components: at least drug, polymer, the organic solvents and water. C36·C37

How much water a solvent can pick up. The amount of water a solvent takes from the air depends on its affinity for water. Water is completely miscible with methanol, but dissolves in dichloromethane (DCM) only to 1.76 g/L at 25 °C. M4·M5 For methanol–water mixtures the authors cite near-ideal Raoult's law behaviour, which they state in words; written as an equation it reads

In plain terms: at equilibrium the mole fraction of water in the solvent, $x_w$, equals the water activity of the surrounding air, $a_w$, which is the relative humidity (RH) divided by 100. At 50% RH, $x_w=0.5$, so one mole (32 g) of methanol can take up almost one mole (18 g) of water. E1·R26·C39 The paper uses this only as an order-of-magnitude argument; how much water the spinning solution actually absorbs depends on the exposure time and was not measured. C39

Reading the images. Contact-mode AFM gives the height of the film surface; the average roughness $R_a$ summarises the height variation of an image, so a featureless film has $R_a$ of about 1 nm and a film with raised domains much more. E3·R5 In LCR mode the AFM cantilever is driven magnetically near its contact resonance frequency. That frequency depends on the stiffness of the material under the tip, so an image recorded at the resonance of one component lights up the regions where that component dominates. C5·P9 In TEM, darker contrast usually means more mass in the beam path; EDX measures the X-rays emitted by each element in a small spot. Evacetrapib carries two trifluoromethyl groups (Figure 1), while neither polymer contains fluorine, so the fluorine-to-nitrogen (F/N) ratio for PVPVA films, or fluorine-to-oxygen (F/O) ratio for HPMC films, rises with the drug content of a phase. The quantification uses an enhanced Cliff–Lorimer method with an absorption correction, which the paper names but does not print. E2·R22.a–b

Methods in detail

The study is built from thin films. Each experiment changes one way in which water can enter (humidity during spin coating, water added to the stock solution, humidity during storage) and then asks two questions of the film: is it split, and what are the domains made of? The overall flow is shown below.

Materials and the drug's solubility

Evacetrapib (purity above 95%) came from ChemShuttle, HPMC E3 from Dow and PVPVA (Kollidon VA 64) from BASF; methanol and DCM were from Mallinckrodt Baker, and Karl-Fischer titration gave 0.03 ± 0.00% water in the methanol. M1–M5·M4 The crystalline solubility was measured in triplicate by shaking excess drug in 10 mL of 50 mM pH 6.8 sodium phosphate buffer at 25 °C for 48 h, separating the solid by ultracentrifugation (35000 rpm, 30 min) and assaying by HPLC at 210 nm. P1·P3 The LLPS onset was found by adding 10 µL of a 10 mg/mL methanolic drug solution to 10 mL of buffer with or without 20 µg/mL HPMC or PVPVA, added to slow crystallization, and assaying the supernatant after ultracentrifugation. P2 The methods give ambient temperature (22 ± 2 °C) for this step, while the results and Table S1 state 25 °C. X11 The HPLC calibration range is printed as "20-100 ng/nL"; ng/mL is the plausible unit. X10

Spin coating under controlled humidity, and water spiking

Stock solutions contained ECP and polymer at 10–90% drug loading (DL, weight percent of drug in the solids) in 50:50 v/v methanol:DCM at 50 mg/mL total solids. ECP alone and ECP–PVPVA were also made in methanol alone at 5 mg/mL, to remove differences in evaporation rate between two solvents. P4 Films were spun from 15 µL of solution (6 s at 50 rpm, then 30 s at 2000 rpm) on silicon wafers or TEM grids, inside a glove box purged with nitrogen until the RH settled. The laboratory air was about 50–60% RH in summer and the lowest RH reachable with the purge was about 18%, so 18% was used as the low, 35% as the intermediate and 50% as the high condition, all at 22 ± 2 °C. The RH meter was specified to ±4% between 20 and 80% RH. P5 Fresh films were vacuum dried overnight. Film thickness, measured by scratching the film and scanning across the scratch, was 50–200 nm for ECP–PVPVA and 50 nm to 1 µm for ECP–HPMC from 10 to 90% DL; thinner HPMC films were made for TEM. P7

For the spiking experiments, 10–100 µL of water was added to 1 mL of stock, mixed and spun immediately at 18% RH, "to minimize droplet coalescence and Oswald ripening" (the paper's spelling). P4·C57 The same idea was applied to the stock itself: nanoparticle tracking analysis (NTA) counted light-scattering particles in a 50:50 ECP–HPMC stock (5% w/v solids) after adding increasing amounts of water, injected into the flow cell straight after mixing. P11

Storage at high humidity

Films spun at 18% RH, whose miscibility the authors checked by AFM (data not shown), were stored for 7 days over saturated potassium sulfate (97.30 ± 0.45% RH at 25 °C) and then vacuum dried for at least 2 days; a further set of ECP–PVPVA films was stored at 75% RH for 10 days. Polarized light microscopy showed no birefringence, and ECP itself stayed amorphous over 7 days at 97% RH. P6·R2·R20 The paper also writes "98% RH" for the same chamber in some places. X3

Imaging and chemical identification

AFM topography (5 × 5 µm images, 256 × 256 points) and roughness were recorded on a nanoIR2 instrument in contact mode. P8 For LCR, at least five nanomechanical spectra were taken on each phase between 1 and 1000 kHz, and images were then recorded at the resonance of one component, with small shifts of the drive frequency to maximise contrast. P9 TEM was run at 200 kV; at least six images per sample were taken at different locations, and EDX spectra were collected with a beryllium holder so that iron signals from the microscope would not overlap with fluorine, using 10 spectra per phase for films split during preparation and for the reference ECP and PVPVA films, and 20 spectra per phase for the stored film. Copper from the grid bars stayed below 0.23 atomic%. P10

DSC of bulk dispersions

For DSC the authors made bulk ECP–PVPVA ASDs (200 mg of solids in about 2 mL of DCM:methanol, rotary evaporation, vacuum drying overnight). Fresh samples were heated at 5 °C/min with a modulation of 1 °C every 60 s, without pre-heating, to avoid mixing the phases by heat. Stored powder samples were held in DSC pans at 97% RH for 24 h, dried for 2 days and then scanned to 20 °C above their second $T_g$. One $T_g$ was read as miscible and two as phase separated. P12·P13·C59 Water uptake of the powders after storage is plotted in Figure 10B, but the procedure is not described. P14

Experimental analysis

Humidity during spin coating decides whether the film splits

Figure 2 is the starting observation. ECP–HPMC films spun at 18% RH showed no surface features at any drug loading from 10 to 90%, which the authors read as miscible films "at least at the 40 nm spatial resolution" of the AFM settings. R3.a·C8 At 35% RH, small round discrete domains appeared; the text places their onset at 30% DL, but in the images 30% DL shows only a faint texture and clear domains start at 40% DL, which the roughness values also suggest. R3.b–f·X17 At 50% RH discrete round domains were present at every loading, and their diameter and height grew with drug loading. R3.g–j At 35% RH the domain height rose from about 50 nm at 50% DL to about 120 nm at 90% DL. R3 Table A gives the roughness behind these images.

Table A. Surface roughness of ECP–HPMC films spun from 50:50 methanol:DCM at 50 mg/mL, read from Figure 3A (approximate). The figure legend labels the low-humidity series "0% RH"; the methods and text give 18% RH. R5·X1

Roughness generally rose with humidity and with drug loading, with "a few exceptions" at 70% DL and above, where, for example, the 50% RH series fell from about 46 nm at 60% DL to about 25 nm at 80% DL. The authors' explanation is hedged: this "may be" because the domains grow bigger and more crowded, leaving less continuous phase and shallower valleys between them. C10·C9·R5 ECP–PVPVA films behaved in the same way (Figure S3), with the 50% RH series reaching about 43–45 nm at 60 and 90% DL, and AFM showed round discrete domains in these films too. R6·R7 For the 90% ECP–HPMC film spun at 35% RH, most domains had areas below 0.1 µm² (Figure S1). R4

TEM shows that the domains are not only on the surface (Figure 3B–C). An 80% ECP–HPMC film spun at 18% RH had no distinct features, whereas the one spun at 50% RH contained dark spheres that overlap in projection, so they sit inside the film as well as at its surface; some droplets were bridged together, and a stereo pair (Figure S6) showed that other overlapping spheres stayed separate. R8·C11 The text calls these panels "Figure 3C and D", but the figure has panels B and C. X2 Heating a split film to 170 °C for 1 min made it featureless again by AFM, which the authors read as heat mixing ECP with the polymer (data not shown). R10·C12 They conclude that environmental humidity "played a critical role" and that higher drug loadings are more prone to splitting when the film forms at 50% RH. C13

The water acts during drying, not in the solution

If humid air works by adding water to the solution, adding water directly should have the same effect. It did. All spiked stock solutions stayed transparent with no visible precipitate, yet films spun from them at 18% RH split once 20 µL of water had been added to 1 mL of the 50% ECP–HPMC stock (Figure 4). Domain size increased with the amount of water and, at a fixed amount of water, with drug loading: the 10% DL films showed only a faint texture at 20 µL and fine domains from 30 µL, all within a ±25 nm height scale, while the 50% DL films reached a ±250 nm scale from 30 µL. R12·C15 By comparing domain sizes and roughness with the films spun at 50% RH, the authors estimate that the binary solvent picks up between 10 and 30 µL of water per mL during spinning at 50% RH. R13·C16 The abstract turns the spiking threshold into "water contents of as low as 2%"; this is 20 µL of added water per mL of stock, not a measured water content of the solvent. C55·X14

NTA then asked when the stock solution itself separates (Figure 5). Table B places the two thresholds side by side.

Table B. Film and solution thresholds for the 50% ECP–HPMC system. Film observations from Figure 4; Ra and NTA counts read from Figure 5 (approximate). R12·R14·R15

The particle count stayed low up to about 170 µL of water per mL, which the authors read as a homogeneous solution with drug and polymer molecularly dissolved, and then rose steeply. R14·C17 Films therefore split at far less water than the solution: 20 against about 170 µL per mL. The authors conclude that the stock solutions, even with up to 15% water, were single phase and that the separation occurred during film formation; the cause is "presumably" that water evaporates much more slowly than methanol and DCM, so the fast loss of the volatile solvents concentrates drug, polymer and water in the drying film. C17·C18 The composition of the drying film was not measured. C18

Methanol carries the water; the polymer holds the drug

Two controls point to methanol as the route by which humid air brings water in. ECP–HPMC films spun from pure DCM, which dissolves little water, did not split at 50% RH (data not shown). R11·M5·X13 And ECP alone, spun from methanol at 5 mg/mL, gave a flat film at 18% RH but small regular round droplets at 50% RH (Figure 6A–B, height scale ±400 nm). R16 The authors write that these drug droplets "most likely" formed in the stock solution during the first stages of methanol evaporation after it had absorbed water, and were deposited as small spherical amorphous particles once the methanol was gone. C19 They conclude that the phase separation during spin coating is "likely" caused by absorption of water by methanol. C14

The third panel of Figure 6 makes a comparison with the polymer present. ECP alone in the binary solvent (25 mg/mL) with 10 µL of water per mL, spun at 18% RH, formed crowded round domains, whereas the 50% ECP–HPMC film made with the same 10 µL of water was featureless (Figure 4). The authors read this as the polymer "appears to decrease the tendency of the drug to phase separate from the solvent in the presence of water". R16·R12.f·C40 Note by the AI writer (not discussed in the paper): the 50% ECP–HPMC stock at 50 mg/mL contains the same 25 mg/mL of drug as the drug-only stock, so the comparison is at equal drug concentration; it rests on one water level.

What the domains are made of after preparation

Height images alone cannot say which phase is the drug. LCR gives a first answer (Figure 7). In a split film of ECP alone on silicon, the discrete domains responded strongly at 117 kHz, near the drug's contact resonance, and the continuous phase at 120 kHz, the silicon substrate. In a split ECP–HPMC film, the domains responded at 123.5 kHz (drug) and the continuous phase at 121 kHz (HPMC). R17·C21 The reference spectra peak at about 119.5 kHz (ECP) and 120.1 kHz (silicon), and at about 126.1 kHz (ECP) and 122.7 kHz (HPMC); the authors explain the shift of the imaging frequencies by a torque on the moving cantilever, and note that partial miscibility "may alter" the peak positions of each phase. R17·C22 From the order of the peaks they conclude that silicon is stiffer than ECP and ECP stiffer than HPMC. C23 Note by the AI writer: the ECP and silicon peaks differ by less than 1 kHz, and the paper does not say how the two imaged films were prepared. X12

TEM with EDX then examines ECP–PVPVA films spun at 50% RH from methanol (Figure 8A). Spherical dark domains grew with drug loading; at 80 and 90% DL the images "hinted at the possibility" of three phases: dark spheres, a dark net-like shading in the background and light holes in the background. R18·C25 Table C lists the EDX ratios.

Table C. EDX atomic ratios from the text (manuscript p.27) and Figure 9. The Figure 9C label for the dark particles reads 0.14. R22·X4

The dark particles in the 90% film "appear to be ECP-rich" and the background ECP-lean; the shaded and light background areas could not be told apart, which "could be" because their difference is below the detection limit of TEM-EDX or because the contrast reflects thickness differences. C26·C27 Note by the AI writer: every phase of a film that is 90% drug reads far below the pure-drug ratio of 1.0, so these ratios rank the phases within one film rather than measure their drug content; the paper does not convert them. X6

Humid storage gives the opposite microstructure

Films that started miscible behaved differently when water came in during storage (Figure 8B). After 7 days at 97% RH, ECP–PVPVA films at 10–80% DL showed net-like structures and the 90% film stayed homogeneous. At 10–30% DL the domains were irregular; at 40–80% DL regular round light features were spread evenly through a darker continuous phase, and they became smaller as drug loading rose. R19 The panels at 60–90% DL are at higher magnification than those at 10–50%, so the size comparison has to be read against the scale bars. X18 EDX inverted the picture of the preparation route: in the 50% film the continuous phase gave F/N 0.10 ± 0.02 and the dispersed phase 0.06 ± 0.01, and in a 40% film 0.09 ± 0.01 against 0.01 ± 0.01, with F/O in the same order (Table C). The dark continuous phase "appears to be ECP-rich" and the holes polymer-rich, which fits the holes shrinking as drug loading rises. R22.f–i·C28

The same inversion appeared with HPMC. Stored ECP–HPMC films (Figure S5B) showed light domains in a darker continuous phase at 40–80% DL, and EDX gave higher F/O in the dark than in the bright domains at every loading tested: 0.14 ± 0.01 against 0.07 ± 0.02 at 30% DL, 0.21 ± 0.06 against 0.09 ± 0.02 at 40%, 0.46 ± 0.09 against 0.29 ± 0.04 at 50%, and 0.90 ± 0.17 against 0.31 ± 0.29 at 80%. The authors conclude that the dark areas are drug-rich and the light regions polymer-rich. R21·R23·C29 By AFM, stored ECP–PVPVA films showed pores or textured surfaces at low and intermediate loadings and smooth surfaces at the highest loadings, and films stored at 75% RH for 10 days showed similar net-like features at 10–50% DL (Figures S2B, S4B, S7). R20 The authors conclude that preparation and storage are two distinct routes of phase separation. C24·C32

Bulk dispersions: DSC and water uptake

DSC of bulk ECP–PVPVA ASDs gives a second view of the storage route (Figure 10). Fresh dispersions showed a single $T_g$ at every loading, falling with drug loading because ECP ($T_g$ 72 ± 1 °C) has a lower $T_g$ than PVPVA; the authors read this as miscibility from 10 to 90% DL. R24·M1·C30 After 24 h at 97% RH and drying, two $T_g$ values appeared at 10–50% DL and one at 60–90% DL. R24

Table D. Glass transitions and water uptake of bulk ECP–PVPVA ASDs, read from Figure 10A and 10B (approximate); error bars not given here. The Figure 10B caption gives 98% RH for the same storage. R24·R25·X3

The single $T_g$ at 60–90% DL is "probably" because the domains are below the DSC detection limit and/or because the phases mix during heating, according to the authors. C31 The water uptake in Figure 10B is not discussed in the text, although it is the measurement behind their later explanation of why high drug loadings resist storage-induced separation. R25·C52 Note by the AI writer: some $T_g$ values lie outside the range set by the pure components (about 65 °C at 10% DL, below pure ECP; about 107 °C at 20% DL, above fresh PVPVA), without comment in the paper. X7

Discussion

Two routes on two phase diagrams

The authors summarise their results as two routes by which water, an antisolvent for evacetrapib, splits an ASD: solution-state phase separation during preparation and solid-state phase separation during storage at high RH; they write that the results "clearly indicate" how susceptible these systems are to water. C32 They explain both routes with schematic phase diagrams (Figure 12), which show phase boundaries only and leave kinetics out. C38

For the preparation route the simplest case is ECP alone in methanol, a drug–methanol–water triangle (Figure 12A). A dry solution at point A is one phase. Trace water, added or absorbed from the air, moves it to B, still one phase. During spinning the organic solvent evaporates faster than water, so the composition moves along B toward D, the pure drug corner, and crosses the binodal at C, where the drug separates. Adding more and more water to the stock moves the solution along A–B–E instead, and it separates only at E. Less water is needed along BCD than along BE, which is the same contrast as the film and solution thresholds of Table B. When the film reaches D all solvent and water have gone and the composition is back in the one-phase region, but "the drug droplets are kinetically trapped due to lack of mobility in the absence of solvents". C38·R14·R15·R16 With polymer present the diagram becomes quaternary and "much more complex", and the composition path also shapes the domain size, shape and porosity, as in membrane formation (refs 46, 48). C36·C41

For storage (Figure 12B, a drug–water–polymer triangle) a high-drug film absorbs little water and stays in the one-phase region (line AB), whereas a low-drug film absorbs more water because it holds more polymer and crosses the binodal (line CD). C42 In the printed Figure 12B the points A–D and the triangle sides are not visible, although the letters exist in the PDF text layer, so the text is the only description of this panel. X9

Why do the two routes leave opposite microstructures? The authors say this "is to be expected". In solution, highly mobile solvated drug and polymer molecules sit in a solvent–water mixture whose composition changes during drying, and "the high mobility and antisolvent effect appear to lead to the formation of spherical drug-rich particles" embedded in a polymer-rich matrix. In the humid film only water is present and mobility is lower, which "perhaps explains" the different, net-like structure in which the drug-rich phase is continuous. C34 The flow below gathers their argument.

The authors also consider an alternative: the "miscibility" of the dry films might be "a kinetically trapped thermodynamically metastable or unstable state" from rapid evaporation or cooling, with water only raising mobility so that a thermodynamically favoured separation can proceed. They think this "less likely", citing earlier work on PVP dispersions in which indomethacin–PVP and ketoconazole–PVP did not separate on exposure to moisture, while pimozide–PVP separated at high RH even at 4 °C, only more slowly than at 22 °C; from that work they take thermodynamics as deciding whether moisture-induced separation happens and mobility as setting its rate. C43·C44·R27

What it means for preparation

The authors draw a practical lesson for laboratory work: humidity control is "critical" in lab-scale studies, since spin coating and solvent casting are widely used, and they report that moisture-induced phase separation during spin coating, described before (ref 26), "is something that we have observed widely in our laboratory for several drug-polymer systems". Hygroscopic solvents such as methanol can hold significant water, so "caution should be used during solvent transfer and handling" when miscibility is assessed. C45·R30 For spray drying, one of the two main commercial ASD processes, the nitrogen purge makes moisture uptake during atomisation less likely, but they list ways water could still enter: water added to the solvent to dissolve a polymer or surfactant, solvent exposed to the air, faster evaporation of the organic solvent than of water, and residual solvent between primary and secondary drying. The added water "could potentially cause phase separation during spray drying", the solvent system should be optimised with this in mind, and these are "areas that need further investigation". C46 The abstract states that the findings have "profound implications on lab-scale ASD preparation and potentially also for industrial production". C54

What it might mean for dissolution

The paper contains no dissolution data, but the authors sketch a link. Phase separation also happens when an ASD hydrates during dissolution; probucol and ritonavir ASDs with PVPVA at 50% DL separated immediately on contact with water, faster than they dissolved (ref 42). R28 If separation occurs during storage at high RH, "albeit with slower kinetics and no material transfer", this "may be predictive" of the behaviour during dissolution. C47 When an ASD separates on hydration, the polymer-rich regions are "likely" to leave and dissolve, leaving the drug-rich phase behind. If the drug-rich phase forms small discrete domains, nanosized drug-rich particles could be released that "may dissolve faster" than larger ones, as proposed for anacetrapib from PVPVA dispersions (ref 52); if the continuous phase is drug-rich, like the net-like storage structure, "it is likely to become a porous scaffold upon polymer exit", with dissolution controlled by the drug (Simonelli, Mehta and Higuchi, ref 53). C48·R29 They therefore call an understanding of how ASD microstructure changes in the presence of water "likely essential" for a mechanistic understanding of dissolution. C49

Note by the AI writer (not discussed in the paper): taken together, the two routes put the drug in opposite places for the same composition. In the authors' dissolution picture a film split during preparation (drug-rich droplets in a polymer matrix) and a film split in storage (a drug-rich network) would release drug differently, and drug loading pulls the two risks in opposite directions: high loading separates most readily when water is in the solvent, but resists humid storage. None of this was tested on dissolution here.

Limitations

Limits the authors state. The quaternary drug–polymer–solvent–water system "may be much more complex" than the ternary polymer systems used as analogies, and no phase diagram was measured. C36 The single $T_g$ at 60–90% DL after storage is attributed "probably" to the detection limit of DSC or to mixing on heating. C31 Partial miscibility of drug and polymer "may alter" the LCR resonance of each phase, and EDX could not distinguish the two background features of the 90% film. C22·C27 Different locations on one spin-coated film vary, so the images shown are representative of general trends. C58 The spray-drying and dissolution consequences are presented as needing further investigation. C46·C47

The AI writer's own judgement. The central observation is well supported: a film that is featureless when spun in dry air splits when water is present during spinning, and the same trend appears for both polymers, in images and in roughness, and in a direct comparison of film and solution thresholds. R3·R5·R6·R12·R14 The weaker parts are the following.

• The EDX ratios rank phases within a film but do not measure compositions: all phases of the 90% film read far below the pure-drug ratio. X6
• Several controls are reported as "data not shown": the pure-DCM films, the remixing on heating and the initial miscibility of the stored films. X13
• The explanation that high drug loadings resist storage because they absorb less water rests on powder uptake after 24 h, not on films after 7 days; uptake at 20–50% and 60–80% DL differs by only a few percent, yet the films separated up to 80% DL. DSC (24 h, powders) and TEM (7 days, films) also differ in exposure. X8·X15
• The LCR identification relies on frequency differences of about 0.6–2.6 kHz on two films whose preparation is not given. X12
• Labels disagree in places: the "0% RH" legend for 18% RH films, 97 and 98% RH for the same chamber, the Figure 3 panel letters, two SI figures with the same caption but different films (Figures S2A and S4A), the incomplete Figure 12B, and the 22 versus 25 °C temperature of the LLPS measurement. X1·X3·X2·X5·X9·X11
• Some $T_g$ values fall outside the range of the pure components without comment. X7
• The study uses one drug, thin spin-coated films and an RH meter specified to ±4% RH; the water thresholds belong to 50 mg/mL stocks. P5·P7·R12

Conclusion

The paper shows that evacetrapib dispersions with HPMC and PVPVA are, in the authors' words, "extremely sensitive to the presence of trace amounts of water present in the solvent during film formation". C50 Spun in dry air they are miscible from 10 to 90% drug; with water brought in by humid air through methanol, or added at 20 µL or more per mL of stock, nanosized drug-rich spheres form during drying and become embedded in a polymer-rich continuous phase, while the stock solution itself stays single phase up to about 170 µL per mL. C51·R12·R14·C17 Water taken up during storage at high RH splits initially miscible ECP–PVPVA films the other way round, with a drug-rich continuous phase and polymer-rich discrete regions at 10–80% DL, and higher drug loadings resist this route, "presumably because moisture sorption is suppressed". C28·C33·C52 The practical message is a laboratory one: a spin-coated miscibility result depends on the water in the solvent and the air, and the paper's imaging approach, AFM with LCR and TEM with EDX, makes the resulting microstructures visible. C45·C53