Building drug-rich islands into a polymer matrix: how phase separation speeds release from high-loading lopinavir dispersions

Background and motivation

An amorphous solid dispersion (ASD) disperses a poorly soluble drug in a polymer so that the drug dissolves faster and reaches supersaturation, which raises both release rate and the driving force for absorption across the intestinal wall. The authors note that most earlier effort went into keeping ASDs physically stable (preventing crystallization, avoiding amorphous–amorphous phase separation, strengthening drug–polymer interactions), and much less into making them release well. C2 A second problem appears at high drug loading (DL): a miscible dispersion releases fast only at low loading, where the dissolving polymer carries the drug into solution; at high loading dissolution becomes controlled by the drug itself and is slow. Since low-potency drugs need high loadings to keep the dose small, ways to make high-loading dispersions release well are of practical interest. C48·C55

Amorphous–amorphous phase separation splits a dispersion into regions of different composition. Its effect on release was unclear when this paper was written. Crystallinity was known to impair dissolution, but reports on amorphous phase separation pointed both ways. C3 On one side, phase separation reduces domain size and changes local drug loading; Matteucci et al. linked the better performance of phase-separated itraconazole formulations to their high surface area. C4 On the other side, Chen et al. stored 40% DL spray-dried particles of compound BMS-817399 with polyvinylpyrrolidone (PVP) at 95% RH: the drug became enriched at the particle surface, the surface-normalized drug dissolution rate did not change, and polymer release slowed. Purohit and Taylor stored ritonavir dispersions (10–50% DL with PVP, copovidone or HPMCAS) at 97% RH and saw lower release for some of them, depending on polymer and loading. C5 (The reference list prints this ritonavir paper twice, as refs 19 and 29. X21)

Answering the question needs a way to measure composition on the scale of the domains, which range from a few nanometres to a few microns. Infrared and Raman microscopy probe about 1 µm deep and mix the signals of smaller domains; transmission electron microscopy averages through the sample thickness; AFM-based Lorentz contact resonance is qualitative or semi-quantitative. AFM-based nanoscale thermal analysis (nanoTA) reaches about 100 nm lateral resolution and about 50 nm depth sensitivity, and can estimate local composition when the two components have very different glass transition temperatures (Tg). C6 The authors chose lopinavir (LPV), a poorly soluble drug that crystallizes slowly, and hydroxypropylmethylcellulose (HPMC, Methocel E5), one of the highest-Tg polymers used in ASDs, so that the local softening temperature reads local composition over a wide window. C7·M1·M2

The study set out to answer three questions: how drug loading and the route and extent of phase separation change drug release; how several coexisting phases change the release mechanism; and which kinds of nanoparticles form when the dispersion disintegrates and dissolves. A1·A2·A3

Background concepts

Amorphous solubility and liquid–liquid phase separation. The amorphous "solubility" of a drug is the highest concentration of molecularly dissolved drug that can exist; above it, the solution separates into a drug-rich colloidal phase and a drug-lean aqueous phase. This is liquid–liquid phase separation (LLPS), and the concentration at which it starts is the onset LLPS concentration. For lopinavir at 25 °C the authors cite 16–18 µg/mL and use 17 µg/mL throughout. M1·C9 The crystalline solubility, used here only to express the degree of supersaturation (concentration divided by crystalline solubility), is not printed as a number; the plotted line in Figure 2 sits at about 3.1 µg/mL. E2·R1·X15 In the release tests, samples are filtered through 0.45 µm before analysis, so the measured concentration is close to the molecularly dissolved drug and cannot rise much above 17 µg/mL even when more drug is released as colloids. P5·C12·C16

Polymer-controlled and drug-controlled dissolution. For a miscible ASD with a water-soluble polymer, drug release generally speeds up as loading falls. Below a boundary the authors call the congruent release boundary, the polymer dissolves quickly and carries the drug with it, so release is paced by the polymer and can overshoot the amorphous solubility, after which the excess drug forms drug-rich droplets by LLPS. At higher loading the drug paces dissolution, release is slow and no droplets form. C48

Two routes to amorphous phase separation. The paper makes phase-separated films in two ways. The first adds water to the casting solution (1:1 methanol:dichloromethane) before the solvent is evaporated; the amount is given as a "water-to-solvent" ratio in percent. The second stores an initially miscible film at 97% RH and 25 °C for 7 days and then dries it. P2·P3 The unit of the water-to-solvent ratio is given as w/w in the figure captions but as v/v in the spin-coating method, so water levels should be compared only within this paper. X4

nanoTA and the Gordon–Taylor calibration. In nanoTA a heated AFM tip is placed on one spot of the film and its temperature is ramped; when the material under the tip softens, the tip sinks in and the cantilever deflection drops (Figure 6A). The temperature of that event is reported as a local Tg. Because the measurement depends on the tip penetrating the film, it reads 20–60 °C above the calorimetric Tg (pure HPMC 209 ± 11 °C by nanoTA against 140 ± 5 °C by DSC; pure LPV 97 ± 4 °C against 69 ± 2 °C), but it rises steadily with polymer content, so it can be calibrated. C11·C25·M1·M2 The authors measured miscible films from 0 to 100% DL and fitted the Gordon–Taylor equation (eq 1):

In plain terms: the Tg of a miscible blend lies between the Tg of the drug ($T_{g1}$) and that of the polymer ($T_{g2}$), weighted by the drug weight fraction $w_1$, with the constant $K$ adjusting how strongly the polymer side counts. All temperatures are in kelvin. With the nanoTA values $T_{g1}$ = 367 K and $T_{g2}$ = 482 K and a fitted $K$ = 0.628, the equation is run backwards to turn each local Tg into a local LPV content. E1·R13 Recomputing from these printed constants reproduces the paper's conversions: 110 °C corresponds to about 79% LPV, 180 °C to about 17%, 150 °C to about 40% and 105 °C to about 85%. E1·R16

Particle sizing. Nanoparticle tracking analysis (NTA) follows individual particles under a microscope and gives a number-based size distribution; D10, D50 and D90 are the sizes below which 10, 50 and 90% of the counted particles fall. Dynamic light scattering (DLS), used here only for drug aggregates in methanol–water, gives an intensity-weighted mean size (the z-average). P12·P13

Methods in detail

The study has three strands that are later set side by side: release tests on films made in three ways, a map of local composition made with AFM and nanoTA, and a census of the particles that leave the films. The flow below shows how they connect.

Contribution 1: films that differ only in microstructure, and how fast they release

① Problem. To isolate the effect of microstructure on release, the films must have the same drug, polymer and loading and differ only in whether and how they are phase-separated. They must also be thin and reproducible enough for AFM.

② How. Stock solutions of 4 mg/mL total solids in 1:1 (v:v) methanol:dichloromethane were rotary-evaporated directly in scintillation vials, so the film formed on the vial bottom. Miscible films used the stock as is; phase-separated films were made by adding water to the stock first (4% for the loading series, 2–100% for the water series), or by storing miscible films at 97% RH (saturated potassium sulfate) and 25 °C for 7 days. All films were vacuum-dried for at least 48 h. P2·P3 For release, 40 mL of pH 6.5, 50 mM phosphate buffer was added to the vial at 25 ± 1 °C. Samples of 0.5 mL were filtered through a 0.45 µm PTFE filter (the first 0.25 mL discarded) and replaced with fresh buffer; LPV was assayed by HPLC at 210 nm. P5·P6 Three doses were used. At 30 µg/mL (1.8 times the amorphous solubility) sampling ran from 10 s to 10 min to catch the fast initial release; at 17 µg/mL, equal to the amorphous solubility, from 5 to 60 min; and at 200 µg/mL from 10 to 40 min. P5 The release rate is the first derivative of the mean concentration–time curve; recomputing the second plotted rate of the 15% DL miscible film from the concentrations at 10 s and 30 s gives about 25 µg/(mL·min), as plotted. E3·R2.a Agitation of the vials and the film area are not stated. P5

③ Why it works. Each film type is made at the same loading from the same stock, so differences in release can be traced to the preparation route. The 17 µg/mL dose tests whether a film can release all of its drug without exceeding the amorphous solubility; the 30 µg/mL dose tests how quickly it approaches the solubility limit when drug is in excess. P5·C12

④ Difference from earlier work. Earlier studies induced phase separation only by storage at high humidity and asked whether it harmed release. C5 Here phase separation is made on purpose during preparation, graded by the amount of water, and compared with humidity-induced separation at the same loadings. A2·A3

Contribution 2: reading local composition with nanoTA

① Problem. AFM height images show discrete domains and a continuous matrix, but not what they are made of. The question is whether the domains are drug-rich or polymer-rich, and how rich.

② How. Films for AFM and nanoTA were cast on 10 mm glass cover slides placed in the vials, with an amount equivalent to 30 µg/mL LPV in 10 mL, and then taped to a steel stud; the authors chose glass so that the film on the slide would match the film on the vial bottom used for dissolution. P4·C8 Topography was recorded in contact mode at 0.5 Hz and 256 × 256 points, and roughness Ra was computed with Gwyddion. P9 nanoTA used heated Thermalever cantilevers calibrated against three polymers of known softening point (polycaprolactone 55 °C, polyethylene 116 °C, poly(ethylene terephthalate) 235 °C). On each phase-separated film the same number of scans (5–30) was placed on discrete and on continuous regions, at least 10 per region for films made with up to 50% water. P10·M9 Each local Tg was converted to LPV content with the Gordon–Taylor calibration. For 33% DL films with 2–40% water, the authors then ignored morphology and grouped all local Tgs into three or four composition classes (drug-rich, intermediate, polymer-rich and polymer-richest), counting the share of scans in each class. P11·E1 A separate set of spin-coated films (50 mg/mL stock with 4% v/v water, spun at 3100 rpm) was dipped in 2 mL buffer for 3 s or 5 min, dried and imaged again to see which material dissolves first. P7·P8

③ Why it works. The nanoTA values of pure HPMC and pure LPV differ by more than 100 °C, so a local Tg places a region on the composition scale with useful resolution. Recomputing the conversion from the printed constants reproduces all of the paper's local compositions. E1·R13·R16

④ Difference from earlier work. Infrared and Raman methods mix the signals of sub-micron domains, and electron microscopy averages through the film; nanoTA measures the top ~50 nm of one spot about 100 nm across, which is close to the size of the domains seen here. C6 Note by the AI writer (not discussed in the paper): the "share" of a composition class is a count of scans, and equal numbers of scans were placed on discrete and continuous regions by design, so the shares are not phase amounts. X9 Also, release was measured on films formed in vials, nanoTA on films cast on cover slides, and the immersion series on spin-coated films; the link between microstructure and release assumes these films are equivalent. X10

Contribution 3: which particles leave a phase-separated film

① Problem. A miscible film at low loading produces drug-rich droplets by LLPS once dissolved drug exceeds the amorphous solubility. A phase-separated film might in addition shed its drug-rich domains intact. The two kinds of particle should differ in size.

② How. 33% DL films made with 2–100% water were dissolved to a theoretical 42.5 µg/mL LPV; after 5 min the unfiltered medium was analysed by NTA, using the measured viscosity of the filtered medium. A miscible 10% DL film dissolved to 34 µg/mL served as the LLPS control. P12 The same films were imaged by AFM, and domain sizes in the dry films were measured from at least two images (or at least 50 domains for sparse films). P9 To mimic what happens while a film dries, a 10 mg/mL methanolic LPV solution was titrated into water and the aggregates were sized by DLS immediately; dichloromethane was left out because it does not mix with water, and HPMC because it does not dissolve in methanol. P13

③ Why it works. The LLPS control produces only droplets formed from dissolved drug, so its size distribution is the reference. Particles larger or more varied than the control, and of similar size to the domains in the dry film, would point to domains released intact. C40·C47

④ Difference from earlier work. Earlier studies characterized the drug-rich colloids formed by LLPS from miscible dispersions; this study compares them with the particles released by a deliberately phase-separated film of the same drug and polymer. C48·C59

Experimental analysis

Release depends on loading and on the route of phase separation

Figure 2 is the core release evidence. For miscible films, release fell steeply with loading: after 10 min the 15% DL film had reached about 17 µg/mL, close to the amorphous solubility, while the 33% DL film had reached about 8 and the 100% DL film about 1.2 µg/mL. R2.a–f·C13 Phase-separated films released much more at intermediate loadings (Table A). R2·C14

Table A. Release of LPV from films at a 30 µg/mL dose in pH 6.5 buffer at 25 °C, read from Figure 2A–C and E–G (approximate). Concentrations are means of three films; rates are derivatives of the mean curve. R2

Three patterns stand out. First, at 33–85% DL both phase-separated films released more than the miscible film, in amount and in early rate; at 33% DL the film made with water reached about 13.3 µg/mL, a degree of supersaturation above 4, against about 8.1 for the miscible film. R2.b·R2.h·C14 The authors call this improvement significant, but no statistical test is described and the rate panels carry no error bars. X11 Second, at 15% and 100% DL the route made no clear difference: at 15% DL the miscible film already releases fast, and at 100% DL there is no polymer. R2·C15 Third, the authors write that the route "appears" to matter, with films made with water releasing faster than humid-stored films at 33–85% DL. The 10 min concentrations support this (33% DL: about 13.3 against 9.8 µg/mL), but the very first plotted rate at 33% DL is higher for the humid-stored film (about 18.6 against 14.5 µg/(mL·min)), and at 15% DL the humid-stored film has the highest peak rate (about 33). C15·X12 The ranking held at the two other doses. At 200 µg/mL, the 33% DL films reached about 12.8 (miscible) against 19.3 µg/mL (water) at 10 min; at 17 µg/mL, about 4.8 against 9.1 at 5 min. R4.b·R5.b·C16 The meaning of the 200 µg/mL dose is unclear: the methods and the Figure S1 caption call it "LPV + polymer", the results "200 µg/mL of drug". X19

At the dose equal to the amorphous solubility (Figure 2D and H), miscible films at 5–20% DL levelled off within 2 min at similar rates, while 25% and 33% DL released gradually (about 9.7 and 4.9 µg/mL at 5 min). None of these films reached the full 17 µg/mL; the 15% DL film plateaued at about 11.7 µg/mL. R3·C17 The Figure 2 caption assigns this dose to panels "D and G"; G is the humid-storage rate panel, so D and H are meant. X1

A little water in the solvent helps; a lot hurts

Figure 3A varies the amount of water added to the casting solution. The authors divide the axis into three regions (Table B). R6·C19

Table B. LPV (µg/mL) released at a 17 µg/mL dose versus water added to the solvent, read from Figure 3A (approximate; 6% and 10% water omitted here). R6

For 33% and 50% DL, release is "highly sensitive" to the first few per cent of water (Region I, 0–6%): at 33% DL it nearly doubles, from about 5.5 to 10.4 µg/mL. Between 6% and 40% water (Region II, which the authors call "optimal") it stays high, and above 40% (Region III) it falls again, to about 7.0 µg/mL at 100% water. The 15% DL film is insensitive except at 70% and 100% water. R6·C19 One explanation the authors consider and set aside is surface area. Roughness of 33% DL films jumps from about 0.8–2.5 nm without water to tens or hundreds of nanometres with 2–10% water and then levels off (Figure 3B); the added surface "may contribute to some extent" but "cannot completely explain" the gain. R7·C21 The spread at 2% water is large (Ra about 0.6 to 160 nm), and the SI images of the same film range from smooth to domain-covered, so this film is patchy. R7.b·R9

What the films look like, and what dissolves first

In AFM height images (Figure 4), miscible films are smooth, with height ranges of ±15–25 nm. Films made with 4% water carry round discrete domains at every loading; the domains grow with loading, from about 50 nm high at 15% DL to about 500 nm at 70% DL, and the 85% DL film was too rough to image. Humid-stored films show irregular discrete and semi-continuous domains from 100 nm to a few microns. R8·C20 The immersion series (Figure 5) shows how these structures behave in buffer. Initially miscible films changed their surface within 3 s of contact with water at 15–70% DL, forming fine features under 100 nm high; after 5 min parts of the film had gone, leaving a few discrete domains at 15% DL and more contiguous material at higher loadings. The authors read this as phase separation on contact with water followed by loss of the fast-dissolving material ("likely indicative of phase separation"). R10·C22 In films phase-separated beforehand, the domains kept their shape and lateral size but grew taller with time, most clearly at 50% DL (height range ±50, ±200 and ±300 nm after 0 s, 3 s and 5 min). The authors attribute this ("most likely") to the polymer-rich continuous phase dissolving and exposing the drug-rich domains, which, unlike the contiguous residue of a miscible film, "can readily be released". R11·C23·C24 Two caveats apply to these figures. The miscible images in Figure 4A and the 0 s images in Figure 5A are the same pictures, although the methods describe the two sets as differently prepared films. X3 The Figure 5 caption says the 70% DL films after 3 s were too rough to image, but the panel that is missing is the 70% DL film after 5 min. X2

Local composition: drug-rich islands in a polymer-rich matrix

The nanoTA calibration on miscible films fell from about 209 °C (pure HPMC) to about 94 °C (pure LPV), with a scatter of roughly ±10–25 °C at each loading (Figure 6B). R13 Figure 6C compares the two routes at 33% DL. In the film made with 4% water, discrete domains softened at 141 ± 34 °C and the continuous matrix at 191 ± 20 °C; in the humid-stored film, at 83 ± 16 and 173 ± 13 °C. The miscible 33% DL film lies at about 144–153 °C. R14 The humid-stored matrix is therefore drug-richer than the matrix of the film made with water. The authors connect this to the slower release of humid-stored films: regions of lower Tg contain more drug and are "expected" to dissolve more slowly. C26

Table C. Local Tg of 33% DL films versus water in the solvent: box medians read from Figure 6D (approximate) and the values printed in the text; LPV contents as converted by the authors with the Gordon–Taylor calibration. R15·R16

Across the water series (Table C), discrete domains stayed below 110 °C, which the authors convert to about 80% LPV or more, and up to 40% water the continuous phase stayed above 180 °C, about 17% LPV or less: drug-rich islands in a polymer-rich matrix. R15·R16·C27 With 50, 70 and 100% water the continuous phase fell to 142 ± 11, 116 ± 16 and 124 ± 7 °C, that is 37–57, 56–92 and 57–71% LPV. In these films the two regions no longer differed within the film plane, and the authors state that the mass balance "cannot be achieved"; they infer drug enrichment at the surface ("suggesting heterogeneity in the z-direction"), which matches the lower release above 40% water. R15·R17·C28·C30 The mass-balance calculation is not shown and no depth profile was measured. X18

Note by the AI writer (not discussed in the paper): two features of the calibration limit how precisely these contents can be read. Several drug-rich values lie below the nanoTA value of pure LPV used in the equation (367 K, about 94 °C), for example the humid-stored discrete domains at 83 ± 16 °C, so the equation gives no composition for them; they are best read as nearly pure LPV within the scatter of the pure-drug calibration points (about 86–102 °C). X16 And the miscible 33% DL films measured at about 145–154 °C sit below the fitted curve at 33% DL (158 °C), so a region at the miscible value would be read as about 40% LPV rather than 33%. Local contents therefore carry roughly ±10% DL uncertainty. X17

The share of polymer-rich phase follows release

Grouping the local Tgs of 33% DL films by composition, regardless of morphology, gave four classes at 2–8% water and three at 10–40% water (Figure 7). The authors conclude that topography alone "appears to be insufficient" to describe the films, because composition varies within both the discrete and the continuous regions. R18·C29·C31 Table D lists the shares of scans in each class.

Table D. Share of nanoTA scans (%) in each composition class of 33% DL films, read from Figure 7B (approximate), next to release from Figure 3A. R19·R6

The authors read Table D as follows. At 10–40% water the fast-dissolving polymer-rich class takes about 40–60% of scans, in line with the largely similar release over that range. C32 Compared with 10% water, the film with 2% water has fewer polymer-rich scans (green and purple together about 26% against 54%) and many more intermediate scans (about 39% against 8%), in line with its lower release; the authors also write that it has more drug-rich phase, which the drug-rich shares (about 35% against 38.5%) do not show. C33·X8 Between 2% and 8% water, the polymer-richest share rises from about 17% to 32% and the intermediate share falls from about 39% to 14%, with local LPV of the intermediate class at about 50–80%; release rises from about 6.5 to 11.4 µg/mL, and the authors conclude that "the proportion of the polymer-rich phase is also critical". C39·R16 The colour classes are rankings within each sample rather than fixed Tg bands: the intermediate class at 40% water is a single value of about 164 °C, above the 150 °C upper limit the text gives for it. X14

The same analysis at three loadings with 4% water (Figure 8 and Figure S4) found four classes with roughly the same Tgs at every loading (about 91–103, 130–151, 178–187 and 192–207 °C). R20·R21·C35 At 15% and 33% DL the class shares were similar (drug-rich about 17–18%, polymer-richest about 11–13%), which the authors match to the similar release of these two films. At 50% DL the drug-richest class rose to about 43% of scans, and the authors write that this much drug-rich phase "is expected to limit" the improvement, as seen in Figure 2B. R20·C36·C37 Their summary of the composition data is that polymer-rich phases with high Tg were present in all the faster-releasing films (8–40% water) and absent at the surface of the slower films (50–100% water), which "helps explain, at least in part" the faster release of phase-separated high-loading films. C38

Particles: broader than LLPS droplets

The 10% DL miscible control dissolved completely; of its 34 µg/mL, half stayed dissolved at the amorphous solubility and half formed drug-rich droplets, with a narrow distribution peaking near 170 nm (D50 197 nm). R27·R22·C40 The miscible 33% DL film never exceeded the amorphous solubility and released no particles. R28·C41 The phase-separated 33% DL films released particles in every case (Table E). R23

Table E. Particles released after 5 min by 33% DL films made with water in the solvent (NTA, Table S1) and domains on the dry films (AFM, Table S2). R23·R25

Compared with the control, the phase-separated films released particles of greater size diversity: D90 was larger in every case (375–597 against 301 nm) and D10 smaller in all but the 100% water film. R23·C42 Particle numbers were lowest with 2% and 4% water and highest with 20% water; the authors write that 8–40% water gave the most particles, but the 70% water film (101 × 10⁷/mL) exceeds the 8% and 40% films. R23·C43·X6 Figure S5 gives somewhat different D50 values for the same samples (for example about 232 nm at 8% water against 187 in Table S1) and a higher particle count at 8% water (about 10.9 × 10⁸/mL against 6.85 × 10⁸), and the paper does not say how the two were computed. R24·X5 Domains in the dry films had D50 of 340–460 nm, larger than the D50 of the LLPS droplets. R25·C44

The titration of methanolic LPV into water (Figure 10) gave aggregates at 150–5000% water-to-methanol: about 510–1160 nm at 150–300% water, about 310 nm at 1000%, and about 220–260 nm at 2000–5000%. R29 The authors argue that during drying dichloromethane, the more volatile solvent, leaves first, so the drug precipitates from the remaining methanol and water as drug-rich aggregates ("likely"), and that aggregates formed below 1000% water are larger than the LLPS droplets, as are the domains in the films. C45·C46 The comparison mixes a DLS z-average with an NTA D50, so it holds only qualitatively. X20 The authors also state that above 6000% water the drug was below its aqueous amorphous solubility and fully dissolved; from the stated 10 mg/mL stock, however, 6000% water still gives about 164 µg/mL LPV, roughly ten times 17 µg/mL, so this statement does not follow from the printed conditions. X7

From these data the authors propose ("it is likely") that the particles from phase-separated films have two origins: polymer-rich regions dissolve, push LPV above the amorphous solubility and form LLPS droplets, and drug-rich domains are released intact as the matrix dissolves, which would explain the larger and smaller species. C47 They add that the first route works only if the polymer-rich share is large enough, above about 40%, because 17 of the 42.5 µg/mL must dissolve before any droplet can form. C53 The composition of the particles was not measured, so the second origin rests on sizes alone. C59

Discussion

The authors' picture of release. The discussion pulls the results into one mechanism. A phase-separated dispersion releases like a sum of small dispersions with different drug-to-polymer ratios ("can be viewed as"), and the size of some regions "may also play a role". C49 In the simplest case of two phases (Figure 11), the polymer-rich phase acts like "glue" holding drug-rich "particles" together. On contact with water the glue dissolves quickly; the drug-rich phase then disintegrates if its micro-regions are not interconnected. The freed domains dissolve slowly, because their dissolution is controlled by the drug, but they "may" dissolve faster than a bulk drug-rich phase because they are small. C50 The immersion images "broadly confirm" this pattern, and the particle sizes are read as showing that drug-rich domains leave the film. C51 The chain below summarizes how the authors connect loading, water and release.

The authors add that real films are more complex than two phases: several composition classes coexist, the surface may differ from the bulk, surface drug enrichment is known to slow release, and the size and connectivity of drug-rich regions can vary greatly; all of these "are likely to impact drug release". C52 Their overall conclusion is that microstructure formation affected release and that the faster release of phase-separated films "was attributed to the creation of fast-dissolving polymer-rich phases", with the composition and amount of each phase and the size and location of the drug-rich phases as the factors that matter. C58·C54

Where the approach applies. The gain appeared only in a window. At 15% DL the miscible film already releases fast and phase separation added little; at 100% DL there is no polymer to dissolve. R2·C34 Within 33–50% DL, release depended on the amount of water: the first few per cent of water gave most of the gain, and more than 40% water reduced it. R6·C19 The drug was chosen for its slow crystallization, and all tests lasted at most an hour. C7·P5 The films were cast at laboratory scale; the authors note that the same idea "would be amenable" to spray drying, where water in the spray solvent is known to induce phase separation, and call this an area "worth further investigation". C56

What it suggests for formulation. The authors present creating microstructure in dispersions of slowly crystallizing drugs as "a potential strategy" to improve release at high loading, and point out that adding water to the solvent is a simple processing step. C55·C56 They also draw a quality-control lesson: humid storage changed release even though no crystals formed, so storage humidity and accelerated-stability conditions need to be controlled and understood. C57 Note by the AI writer (not discussed in the paper): in this study humid storage improved release relative to the miscible film, but it left a drug-richer matrix and a smaller gain than water in the solvent; whether storage-induced phase separation helps or hurts seems to depend on the composition it produces, so the result here does not make humidity exposure safe in general. R2·R14

Limitations

Limitations stated by the authors. The authors are cautious in several places. Roughness "may contribute to some extent" but "cannot completely explain" the improved release. C21 The topography of miscible films after contact with water is "likely indicative" of phase separation, and the height increase of pre-phase-separated domains is "most likely" due to exposure. C22·C23 The nanoTA softening temperatures differ from calorimetric Tgs because of how the measurement works. C11 The two-phase schematic is an assumption, and real systems are "more complex", with surface and bulk possibly differing. C50·C52 The origins of the particles are given as "likely", and the antisolvent origin of the domains as "likely". C47·C45

The AI writer's own judgement (1): what is measured and what is inferred. The release data are direct and repeat at three doses, so the main claim, that phase-separated films released more at 33–85% DL, is well supported for this drug, polymer and set-up. R2·R4·R5 The explanation through polymer-rich phases is a correlation: no phase was dissolved on its own, the particles were sized but not analysed for composition, and surface drug enrichment at high water rests on a mass balance that is not shown. C38·C59·X18 The composition shares count nanoTA scans placed in equal numbers on discrete and continuous regions, so they describe the sampled spots rather than the amount of each phase, and one stated comparison (more drug-rich phase at 2% than at 10% water) is not borne out by the plotted shares. X9·X8 The local compositions themselves carry roughly ±10% DL uncertainty from the calibration. X17·X16 The composition series covers only 33% DL, and release, nanoTA and immersion were measured on films made in three different ways (vial, cover slide, spin-coating). X10 "Significant" differences are stated without a statistical test. X11

The AI writer's own judgement (2): inconsistencies in the source. Several items in the figures and SI do not agree with each other or with the text. The Figure 4A images reappear as the 0 s images of Figure 5A for films described as differently prepared. X3 Table S1 and Figure S5 give different D50 values and, at 8% water, different particle numbers for the same samples. X5 The "highest particle numbers at 8–40% water" holds clearly only for 20% water. X6 The statement that LPV was fully dissolved above 6000% water-to-methanol does not follow from the 10 mg/mL stock. X7 The abstract places the improvement "above 33% w/w", while the data include 33% DL. X13 Captions misname panels (Figure 2 "D and G"; Figure 5 "70% LPV 3s"), the unit of the added water switches between w/w and v/v, and the 200 µg/mL dose is described both as drug and as drug plus polymer. X1·X2·X4·X19 None of these changes the direction of the main release result, but they matter for anyone reusing the particle and composition numbers.

Conclusion

This paper shows, for lopinavir–HPMC films, that amorphous phase separation is not simply a defect. Made on purpose by adding a little water to the casting solvent, or arising from storage at 97% RH, it raised release at 33–85% drug loading above that of miscible films of the same loading, while having no effect at 15% and 100% loading. R2·C14 nanoTA located the drug in discrete islands of about 80% LPV or more inside a polymer-rich matrix of about 17% LPV or less, and the authors attribute the faster release to that matrix dissolving first and freeing the islands; too much water made the matrix itself drug-rich, and release fell. R15·R16·C58·C28 The particles released were broader in size than LLPS droplets and are proposed to include intact drug-rich domains. R23·C47 The authors present microstructure creation as a potential strategy for high-loading dispersions of slowly crystallizing drugs, and humidity control as a quality issue; the mechanism behind the gain, the phase shares and the particle origins remain inferences that the next experiments would need to test directly. C55·C57·X9