What limits drug loading in amorphous solid dispersions: the drug's glass transition and the choice of polymer, not log P

Background and motivation

New drug candidates have drifted toward higher molecular weight and lipophilicity, which lowers their aqueous solubility and makes oral absorption hard to achieve (Williams 2013). An amorphous solid dispersion (ASD), in which the amorphous drug is mixed with a polymer at the molecular level (Huang 2014), can raise both the dissolution rate and the free drug concentration (Williams 2013; Almeida e Sousa 2015; Murdande 2010). On dissolution, many ASDs push the drug concentration above the amorphous solubility. The excess drug then separates by liquid–liquid phase separation (LLPS; Ilevbare 2013) or, when the drug-rich phase is glassy, by glass–liquid phase separation (GLPS; Mosquera-Giraldo 2015), giving roughly spherical, amorphous, drug-rich particles of about 50–1000 nm that the paper calls nanodroplets. Nanodroplets have been shown to help absorption in vitro (Indulkar 2016) and in vivo (Wilson 2018; Stewart 2017; Kesisoglou 2019), and smaller ones may outperform larger ones.

Drug loading (DL, % w/w of drug in the ASD) often decides whether an ASD releases its drug and forms nanodroplets (Jackson 2016; Mosquera-Giraldo 2018; Xie 2016). Work on polyvinylpyrrolidone/vinyl acetate (PVPVA) dispersions, much of it from the Taylor group, gave a working picture (Indulkar 2019; Chen 2016; Saboo 2019; Que 2019). At low DL the polymer controls dissolution, and drug and polymer leave at the same normalized rate; this congruent release comes with fast release and many nanodroplets. As DL rises, release drops abruptly over a small DL increase. Indulkar 2019 called this the "falling-off-the-cliff" effect and named the DL at which it occurs the limit of congruency (LoC); above it, release is incongruent and few or no nanodroplets form.

Two questions were left open. For the polymer, HPMC and hydroxypropyl methylcellulose acetate succinate (HPMCAS) had stabilized nanodroplets better than PVPVA for danazol (Jackson 2016) and telaprevir (Mosquera-Giraldo 2018), but how different polymers govern release across DLs was, in the authors' words, "not well studied". For the drug, Saboo 2019 found that cilnidipine, with the higher log P, dissolved better than nilvadipine from PVPVA and had a similar LoC, so log P and LoC did not correlate. Saboo 2020 found a lower PVPVA LoC for indomethacin (10% DL) than for its methyl ester (25% DL), an analogue with a lower glass transition temperature (T_g) and no H-bond donor. Each study compared one pair of drugs in one polymer and could not say which drug property is decisive.

This paper tests two drug properties, log P and the wet T_g, with two polymers. Four drugs fill a 2 × 2 grid (wet T_g above or below 37 °C; log P above or below 5.0), and two further high-T_g drugs are studied to a lesser extent. The polymers differ in hydrophilicity and H-bonding: PVPVA has a water contact angle of 30.6° against 58.6° for HPMCAS (Qian 2012) and one carbonyl H-bond acceptor per repeat unit, while HPMCAS has several acceptors and some donors (Kothari 2015). The authors "hypothesized" that a high log P drug would make the ASD more hydrophobic and harm release, and "further hypothesized" that a high-T_g drug would be detrimental to release and nanodroplet formation. They also stated a feature of the drug set that later limits the answer: all high-T_g drugs carry H-bond donors, while the two low-T_g drugs have none but contain chlorine and can halogen-bond with polymer carbonyl groups (Que 2020; Xu 2014).

Background concepts

Amorphous solubility and nanodroplets. The amorphous solubility is the highest concentration of dissolved drug that can coexist with the amorphous drug, several times the crystalline solubility. Above it, the solution splits into a drug-lean aqueous phase and dispersed nanodroplets. The paper takes the LLPS or GLPS onset concentration as the amorphous solubility (Ilevbare 2013) and measures it four ways: HPLC of the supernatant after ultracentrifugation; UV extinction at 430 nm, where no drug absorbs and the onset of scattering marks new particles; nanoparticle tracking analysis (NTA), where the particle count rises abruptly; and pyrene fluorescence, where the ratio of the first to the third emission peak, $I_1/I_3$ (371 and 382 nm), changes as pyrene enters the drug-rich phase. The fluorescence values were used for analysis. In the release tests, "total released drug" is dissolved drug plus drug in nanodroplets left after a short centrifugation, so a value above the amorphous solubility means nanodroplets are present.

Glass transition and wet T_g. Below T_g an amorphous material is a glass with slow molecular motion; above it, a supercooled liquid. Water plasticizes and lowers T_g. The wet T_g is the T_g of the drug saturated with water, and a drug is "high T_g" when its wet T_g is above 37 °C. As the paper applies it to loratadine, the wet value decides whether the water-saturated drug-rich phase is a liquid or a glass at the test temperature. DSC can measure the wet T_g only when it is not far below 0 °C (water freezes) or very high (water evaporates); only loratadine qualified. For the others the authors used a rule of thumb from Tong and Zografi (2004), written here as a formula:

Here $T_{g,\mathrm{d}\mathrm{r}\mathrm{y}}$ is the DSC T_g of the dry amorphous drug (°C) and $W_{100}$ its water content (% w/w) at 100% relative humidity, extrapolated from a 37 °C sorption isotherm measured from 5 to 95% RH. In plain terms: each 1% w/w of absorbed water is taken to lower T_g by 10 °C. The water contents are not printed. Note by the AI writer (not discussed in the paper): back-calculated from Table 1, the rule implies about 0.35% water for miconazole, 2.0% for danazol and 6.3% for elbasvir.

Congruent release and the normalized rate. To compare drug and polymer release from a compact of fixed area, the paper defines a normalized rate of dissolution (Eq. 1):

Here $R$ is the normalized rate (mg min⁻¹ cm⁻²), $k$ the slope of concentration against time, $V$ the medium volume (100 mL), $A$ the exposed compact area (8 mm diameter die) and $x$ the weight fraction of the component in the ASD. In plain terms: $R$ is how fast a component leaves each square centimetre of surface, scaled as if the ASD were made of that component alone. Equal values for drug and polymer mean they leave in the same ratio as in the solid, which is congruent release. The LoC is the highest DL at which the two rates do not differ significantly.

Amorphous–amorphous phase separation and the |Δχ| effect. When an ASD takes up water, a drug–polymer mixture that was miscible when dry can split into a drug-rich and a polymer-rich phase; this is amorphous–amorphous phase separation (AAPS). If AAPS at the dissolving surface is faster than dissolution, it "has been suggested" that a drug-rich layer forms and controls a much slower release (Que 2019). The polymer literature explains solvent-induced separation with the "|Δχ| effect" (Robard 1977), Eq. 2:

Here subscript 1 is the solvent (water), 2 and 3 are the blend components (drug and polymer), ${\chi }_{12}$ and ${\chi }_{13}$ are the Flory–Huggins interaction parameters between solvent and each component, and ${\chi }_{23}$ is the parameter between the two components alone. In plain terms: if water interacts very differently with the two components, adding water drives separation even when drug and polymer alone are miscible (${\chi }_{23}$ negative or small). Earlier work found much larger χ values with water for drugs than for polymers (Rumondor 2010), which makes $|\Delta \chi |$ large. No χ values were measured here; Eq. 2 serves only to argue why log P predicts poorly.

How a glassy polymer dissolves. Water must diffuse into a glassy polymer and turn it into a swollen gel layer; in a good solvent the chains then disentangle and leave, otherwise dissolution is limited by stress relaxation in the swollen layer. The larger the gap between T_g and the test temperature, the slower the water enters, so additives that change the matrix T_g change the dissolution rate; 20 wt% poly(ethylene oxide) doubled the dissolution rate of poly(methyl methacrylate) by plasticization (Cooper 1985). Additives can also slow dissolution by blocking the polymer's polar groups from water (Houlihan 2000). In an ASD, the drug is such an additive.

Nanodroplet size by DLS. Dynamic light scattering (DLS) gives the z-average diameter and the polydispersity index (PDI), from near 0 (narrow) to 1 (very broad). A z-average and PDI that rise over time indicate agglomeration, coalescence or crystallization.

Methods in detail

The work has four parts. First, the authors characterized the drugs and used the values to place four of them in a 2 × 2 grid. Second, they dissolved ASD powders over a range of DLs, reading total drug release and, for three DLs per system, polymer release from the same samples. Third, they dissolved loratadine compacts at 37 and 5 °C to change the drug's T_g relative to the test temperature without changing the molecule. Fourth, they followed nanodroplet size by DLS, a result that also set the conditions of the powder tests.

Contribution 1: Characterizing the drugs and building the 2 × 2 grid

① Problem. To ask whether log P or T_g governs release, the drug set must vary the two properties independently. Each drug also needs a threshold concentration above which nanodroplets must be present. The ASDs must be fully amorphous, so that poor release cannot be blamed on residual crystals.

② What they did. Crystalline solubility: excess drug stirred in 10 mL of buffer at 300 rpm and 37 °C for 48 h, ultracentrifuged (35,000 rpm, 30 min) and assayed by HPLC at 210 nm. The buffer pH was set at least two units above the drug pKa to keep the drug unionized: 100 mM phosphate at pH 7.5 for loratadine, danazol and elbasvir, and 100 mM glycine–NaOH at pH 9.0 for miconazole, the same buffers being used for dissolution. Amorphous solubility was measured by the four methods above. Dry T_g was measured by DSC (about 5 mg, 3 h at 20 °C under nitrogen, heat–cool–reheat at 10 °C/min; midpoint, n = 3). Wet T_g was measured by DSC for loratadine only, on amorphous drug precipitated in buffer at 37 °C, pelleted and dried to constant weight; for miconazole, danazol and elbasvir it was estimated with the 10 °C per 1% water rule. Paclitaxel and ivacaftor values were taken from the literature (dry T_g 152 and 175 °C; wet T_g not determined). ASDs were made from 1 g of drug plus polymer in 20 mL of ethanol (PVPVA) or 1:1 methanol/dichloromethane (HPMCAS) by rotary evaporation, dried overnight under vacuum, cryo-milled and sieved to 149–250 μm; all were amorphous by powder X-ray diffraction (PXRD), patterns not shown.

③ Why it works. With wet T_g crossed against log P, a log P effect would appear as a difference within a T_g class, and a T_g effect as a difference within a log P class. The classes were loratadine (low T_g, low log P), miconazole (low T_g, high log P), danazol (high T_g, low log P) and elbasvir (high T_g, high log P). The amorphous solubility gives each drug a threshold that does not depend on the ASD, and four methods show how much it depends on the technique.

④ How it differs from earlier work. Saboo 2019 and Saboo 2020 each compared one pair of drugs in PVPVA. Here four drugs are chosen to separate two properties, two more widen the high-T_g class, and every drug is tested with two polymers. The built-in weakness, which the authors name, is that the T_g classes coincide with the presence or absence of H-bond donors.

Contribution 2: Release against drug loading from powders, with drug and polymer measured together

① Problem. The release-versus-DL curve had to be mapped for twelve drug–polymer pairs. Each sample must count drug in nanodroplets as released while excluding undissolved ASD particles. To tell whether loss of release comes with loss of congruency, polymer must be measured in the same samples.

② What they did. ASD powder was dissolved in 50 mL of buffer at 37 °C, stirred at 300 rpm, using the mass that gives 100 μg/mL total drug at complete release (50 mg for a 10% DL ASD). Samples of 1 mL were withdrawn and replaced, centrifuged for 1 min, and the supernatant assayed for drug by HPLC, for PVPVA by size-exclusion HPLC at 205 nm and for HPMCAS by a phenol–sulfuric acid colorimetric method. The centrifugation speed was set in two steps: nanodroplets made by solvent shift were spun at a series of speeds to find the highest that removed none, and then 70% DL powders, which stay largely undissolved, were dissolved and the supernatant checked by cross-polarized light microscopy, raising the speed until no ASD particles remained. The final speeds were 1000 rpm (loratadine, miconazole), 3000 rpm (danazol) and 5000 rpm (elbasvir); those for paclitaxel and ivacaftor, and their buffers, are not stated. All PVPVA ASDs were dissolved in buffer already containing 400 μg/mL of dissolved HPMCAS, to prevent agglomeration or crystallization of nanodroplets; HPMCAS ASDs were dissolved in plain buffer. Total released drug was read at 60 min, except for danazol at 15 min because it crystallized later (Figure 3). For each core drug with each polymer, three DLs giving complete, partial and low release were followed for drug and polymer (Figure 5; Figures S2 and S3). All tests were in triplicate.

③ Why it works. Fixing the total drug at 100 μg/mL makes the y-axis of Figures 3 and 4 a direct percent release on one scale for all drugs. The amorphous solubility drawn on the same plot shows whether nanodroplets must be present. Polymer measured in the same aliquot gives congruency without a second experiment.

④ How it differs from earlier work. The cliff effect had been reported with surface-normalized compacts (Indulkar 2019; Saboo 2019; Que 2019); here it is tested with powders, whose area changes as they dissolve, and with HPMCAS as well as PVPVA. The HPMCAS added to the PVPVA media is new: it makes the measurement more robust, but the PVPVA powders were never tested in plain buffer.

Contribution 3: Loratadine compacts at two temperatures as a T_g switch

① Problem. In the drug set, high T_g and H-bond donors go together, so the powder data cannot say which lowers the loading limit. A test was needed that changes the drug's T_g relative to the test temperature while leaving its chemistry unchanged, with a constant surface area so that rates are comparable.

② What they did. 100 mg of loratadine ASD was compressed in an 8 mm die (1500 psi, 60 s) and rotated at 100 rpm in 100 mL of medium in a Wood's intrinsic dissolution apparatus at 37 or 5 °C; neat PVPVA and HPMCAS compacts were run the same way, all in triplicate. The DLs were 5–50% (HPMCAS, 37 °C), 2–10% (HPMCAS, 5 °C), 20–40% (PVPVA, 37 °C) and 2–20% (PVPVA, 5 °C). Rates came from Eq. 1 using the linear part of each profile, or the first 30 min when it was not linear. At each DL the drug and polymer rates were compared by unpaired Student's t-test; the LoC is the highest DL with no significant difference (Figures 8 and 11). Whether the PVPVA compact media contained dissolved HPMCAS is not stated.

③ Why it works. Loratadine's wet T_g is 13.8 °C (Table 1; the Results text rounds it to 13 °C). At 37 °C its drug-rich phase should be a supercooled liquid; at 5 °C, in the authors' words, loratadine "becomes a high Tg compound" and its nanodroplets would be glassy. Its H-bonding capacity is the same at both temperatures. The neat-polymer compacts test whether cooling alone slows the polymer enough to explain a lower LoC.

④ How it differs from earlier work. Saboo 2020 compared indomethacin with its methyl ester, a change that removes the H-bond donor and lowers T_g at once. Changing temperature keeps the molecule fixed. The price is that temperature also changes rates, water uptake and the phase behaviour of the wet ASD, as the authors note for the last.

Contribution 4: Nanodroplet size and stability by DLS

① Problem. Nanodroplets that grow or crystallize can be spun down during sampling and counted as unreleased, and larger species may help absorption less. The polymer's effect on size had to be known to set the test conditions and to read the results.

② What they did. Samples taken over 60 min from dissolving ASDs were centrifuged as in the powder tests and measured by DLS (173° backscatter, n = 3). For elbasvir at 2.5% DL (Figure 2), three series were run: the PVPVA ASD in buffer, the HPMCAS ASD in buffer, and the PVPVA ASD in buffer containing 400 μg/mL dissolved HPMCAS. The same design was applied to danazol (2.5% DL), miconazole (10%) and loratadine (10%) in Figure S1.

③ Why it works. The third series separates polymer in the solid from polymer in solution: if the PVPVA ASD gives small, stable nanodroplets once HPMCAS is in the medium, dissolved HPMCAS is the stabilizer.

④ How it differs from earlier work. Jackson 2016 and Mosquera-Giraldo 2018 had shown better stabilization by HPMC or HPMCAS for danazol and telaprevir. This paper extends the comparison to four drugs and uses it to design the release tests, which shapes how the polymer comparison can be read.

Experimental analysis

Setup in brief. Six drugs were formulated with PVPVA and with HPMCAS-MF at DLs from about 2% to 70%, depending on the system. Powder dissolution at 37 °C gave total released drug against DL for all twelve pairs, and drug and polymer profiles at three DLs for the eight pairs of the four core drugs. Loratadine was also tested as compacts at 37 and 5 °C, and DLS covered the four core drugs. Table A collects the drug properties on which the analysis rests.

Table A. Properties of the six drugs (from Table 1 and Table 2; measured T_g values are mean ± SD, n = 3; amorphous solubility is the LLPS/GLPS onset, fluorescence value where available). Wet T_g of miconazole, danazol and elbasvir is estimated from water sorption.

What Table A shows. The amorphous solubility is about 5 to 9 times the crystalline value for the three drugs where both were measured (loratadine 11.0 against 1.8 μg/mL). The four methods differ by up to a factor of about two (loratadine 10.0–18.0 μg/mL), and elbasvir's value rests on NTA alone because its drug-rich species could not be pelleted and the optical methods failed. Wet T_g and log P share no order: the lowest-T_g drug (miconazole) has the highest log P, and the low-log P drugs include one low-T_g and one high-T_g compound. The grid therefore separates T_g from log P, but not from H-bond donors, which follow T_g exactly.Card check: miconazole log P is 5.69 in Table 1 but 5.22 in the Discussion; both exceed the 5.0 cutoff, so its class does not change (X1).

Main result 1: PVPVA dispersions fall off a cliff; HPMCAS dispersions decline gradually. Every ASD released all its drug at a low enough DL, and release fell as DL rose, at a DL that depended on drug and polymer (Table B).

Table B. Total released drug (dissolved plus nanodroplets, μg/mL; 100 = complete release) at selected DLs, 60 min except danazol at 15 min (read from Figure 3 and Figure 4, approximate; n = 3). PVPVA ASDs were dissolved in buffer containing 400 μg/mL HPMCAS.

With PVPVA, release dropped from near-complete to low within about one 5–10% DL step in every system: loratadine from about 85 to 8 μg/mL between 20 and 30% DL, miconazole from about 93 to 3 between 40 and 50% DL. With HPMCAS the fall was spread out; loratadine still gave about 86 μg/mL at 40% DL, 46 at 50% and 12 at 70%. The authors saw "No sudden decline" for HPMCAS and call a negative "slope" "a more appropriate description". Elbasvir was the exception, with similar curves for both polymers. Because earlier cliff reports used compacts, the authors read the cliff in powders as showing that PVPVA behaviour does not depend on the test type ("This suggests").Card check: the stated cliff DLs mix two conventions. Loratadine's 25% is the first DL with reduced release (about 41); miconazole's 40%, danazol's 5% and elbasvir's 10% are the last DL before the drop. Each is accurate to one tested DL (X26).

At the same DL, HPMCAS ASDs released more mainly at high loading: loratadine at 40% DL about 86 against 4 μg/mL, danazol at 10% about 75 against 9.5, paclitaxel at 10% about 82 against 19, ivacaftor at 10% about 99 against 21. This did not hold everywhere: miconazole gave about 93–103 μg/mL from PVPVA against 83–97 from HPMCAS at 10–40% DL, elbasvir at 10% DL released more from PVPVA (about 77 against 57), and ivacaftor at 20% DL gave about 7 from HPMCAS against 9 from PVPVA. The authors' wording is qualified accordingly: HPMCAS dispersions "generally performed better", and at medium to high DL "The general trend" is more nanodroplets from HPMCAS.

Main result 2: wet T_g, not log P, ordered the drugs. With PVPVA, the two low-T_g drugs kept good release up to 20% (loratadine) and 40% DL (miconazole), while the four high-T_g drugs failed at 10% DL or lower (Table B). Paclitaxel and ivacaftor, added "To confirm" the high-T_g effect, agreed. With HPMCAS, both low-T_g drugs still exceeded their amorphous solubility at 50% DL (loratadine about 46 against 11 μg/mL; miconazole about 44 against 5), and above 50% DL only they gave "a reasonable amount of release".Card check: the Results say high-T_g HPMCAS ASDs at 50% DL or more formed few to no nanodroplets with the amorphous solubility not exceeded, but the plotted values lie above it (danazol about 14 against 8; elbasvir about 11 against 0.6–1 μg/mL), and among high-T_g drugs only danazol was tested above 50% DL (X13, X14).

Log P did not track any of this. Miconazole, elbasvir and ivacaftor have similar log P (5.69, 5.33 and 5.68), yet with PVPVA miconazole dropped off at 40% DL and the other two at 10% DL or lower. Danazol and elbasvir differ widely in log P (3.8 and 5.33) but dropped off at similar low DLs, and the HPMCAS dispersions behaved in parallel. The authors call this result "somewhat counterintuitively" observed; no correlation statistic was computed. In the Conclusions the contrast is framed as low-T_g drugs without H-bond donors against high-T_g drugs with donors.

Main result 3: release failed when drug and polymer stopped dissolving together. Table C shows drug and polymer release from the powders.

Table C. Drug and polymer released from ASD powders (% of the amount present) at 60 min unless stated (read from Figure 5, Figure S2 and Figure S3, approximate; n = 3). The danazol–HPMCAS low-DL panel is titled 5:95, although the caption says 2%.

Complete drug release came with congruent release. At higher DLs the polymer ran ahead of the drug, and the two T_g classes split. For low-T_g miconazole in PVPVA at 50% DL, about 97% of the polymer dissolved but only about 2% of the drug. For high-T_g elbasvir in PVPVA at 50% DL, the polymer itself was held back (about 34%), and danazol–PVPVA (37%) and elbasvir–HPMCAS (23%) at 50% DL followed suit; low-T_g loratadine–PVPVA at 50% DL sat between (polymer about 72%). The authors state that these patterns "were true for all binary systems" and read two kinds of failure: fast polymer release from miconazole–PVPVA "most likely" reflects water-induced AAPS, with the polymer-rich phase dissolving past a mobile, liquid drug-rich phase, while polymer dissolution "is impeded" at high loadings of a high-T_g drug. AAPS was not measured here; the miconazole–PVPVA evidence is cited (Saboo 2017). The danazol–PVPVA rows add a separate complication: at 2% DL drug release fell from about 91% at 15 min to about 17% at 60 min while the polymer reached 94%, because danazol crystallized in solution.Card check: the Discussion cites Figure 5a for rapid polymer release at very low drug release, but Figure 5a is the 10% DL panel with complete release; the relevant profiles are Figure 5b,c (X10).

Main result 4: cooling loratadine below its wet T_g cut the LoC. Table D gives the normalized rates of the compacts.

Table D. Normalized dissolution rates (Eq. 1, mg min⁻¹ cm⁻²) of loratadine ASD compacts and neat polymers (read from Figure 8 and Figure 11, approximate, except the printed neat HPMCAS rates; n = 3; ns = not significant). The LoC is the highest DL with no significant drug–polymer difference.

Cooling from 37 to 5 °C lowered the LoC from 30 to 2% DL with PVPVA and from 20 to 5% DL with HPMCAS. At 37 °C, PVPVA compacts dissolved much faster than HPMCAS compacts (about 3.3 against 0.4 mg min⁻¹ cm⁻² at 20% DL), and their collapse above the LoC was sharp: between 30 and 35% DL the PVPVA rate fell from about 2.70 to 0.36. HPMCAS compacts lost congruency gradually, the drug rate falling from about 0.43 at 10% DL to 0.14 at 50%. At 5% DL and 37 °C, HPMCAS dissolved faster from the ASD than as neat polymer (about 0.61 against 0.47), which the authors find "feasible" to explain by plasticization by a little low-T_g drug; the ASD T_g was not measured.

The release profiles add the extent of release. Each compact holds 100 mg of ASD in 100 mL, so the drug at complete release rises with DL and the amorphous-solubility line, as percent released, falls (for HPMCAS at 37 °C about 11, 5.5, 3.7 and 2.8% for 10–40% DL; Figure 6). At 20% DL, PVPVA compacts released all of the loratadine by 60 min (about 105% drug, 104% polymer), while HPMCAS compacts had released about 19% by 90 min. Above the LoC, nanodroplets could still form: loratadine–PVPVA at 35% DL released about 23% of its drug by 90 min against a line of about 3.2%, while at 40% DL it released about 0.2% and formed none (Figure 9).Card check: the Results say PVPVA compacts at 30% DL and below released fully within 1 h, but Figure 9b shows drug at about 80% at 60 min and 85% at 90 min for 30% DL, congruent by rate but not complete (X11). The text also calls 2% DL at 5 °C complete, against about 78% drug by 180 min (X12).

The control is the neat polymer. Neat HPMCAS dissolved at 0.472 and 0.469 mg/min cm² at 37 and 5 °C, so the authors conclude that a slower polymer alone "cannot explain" the LoC drop. Neat PVPVA, however, slowed from about 3.45 to 1.68 mg min⁻¹ cm⁻².Card check: the polymer-rate argument is supported by the neat HPMCAS data, not by the neat PVPVA data (X16). Note by the AI writer (not discussed in the paper): the authors' other argument, that cooling slows drug and polymer alike and so should not by itself break congruency, still applies to PVPVA; what the PVPVA control cannot show is that the polymer's own slowing plays no part.

Main result 5: HPMCAS kept elbasvir and loratadine nanodroplets small; PVPVA did not. Table E summarizes the DLS data.

Table E. Nanodroplet z-average diameter from first to last time point (5 to 60 min unless stated) and PDI by DLS (read from Figure 2 and Figure S1, approximate; n = 3).

Elbasvir nanodroplets from the PVPVA ASD in plain buffer grew from about 670 to 1190 nm in 60 min, broadened and crystallized after 30 min, while those from the HPMCAS ASD, or from the PVPVA ASD with dissolved HPMCAS, stayed at about 250–390 nm, so dissolved HPMCAS did the stabilizing. Loratadine showed the same contrast between plain PVPVA and HPMCAS (about 2120 against 570 nm at 60 min). The authors state that "Similar trends" held for loratadine, danazol and miconazole, but the SI data are less uniform: danazol nanodroplets from the PVPVA ASD with dissolved HPMCAS grew to about 2780 nm, and for miconazole all three series grew alike.Card check: the text gives "around 400 nm" for elbasvir nanodroplets from the PVPVA ASD with dissolved HPMCAS; Figure 2a shows about 250–300 nm (X8). The Figure S1c–f legends name the third series as HPMCAS-based, so its identity is unresolved (X7). On the strength of the elbasvir data, every PVPVA powder test was run with 400 μg/mL dissolved HPMCAS.

Comparisons with the literature and what the controls rule out. The authors set their results beside earlier PVPVA data: indomethacin (wet T_g 29 °C, H-bonds with PVPVA) has an LoC of 10% DL and its methyl ester (wet T_g below 1.3 °C, no donor) 25% (Saboo 2020); ledipasvir (T_g 160 °C) has 5% (Que 2019) and ritonavir (wet T_g about 0 °C) 25% (Indulkar 2019); ritonavir–HPMCAS gave nanosized colloids up to 50% DL (Purohit 2017). Together, these "support the hypothesis" that high-T_g compounds "tend to" release worse, while indomethacin, low-T_g by the 37 °C criterion yet with a low LoC, points to H-bonding. Among the controls, PXRD rules out residual crystals, the centrifugation screen limits miscounting of undissolved ASD and nanodroplets, the neat HPMCAS compacts rule out slower polymer dissolution as the cause of the 5 °C drop for HPMCAS, and the polymer assay shows that in the low-T_g systems most of the polymer dissolved even at the highest DLs (about 72–97%). No control addresses the dissolved HPMCAS in the PVPVA powder media.

Discussion

The authors' mechanism. The Discussion builds an explanation in hedged steps. The expected route was that water-induced AAPS at the dissolving surface creates a drug-rich layer that slows release, and that a more lipophilic drug would separate more readily. Log P failed, and the authors argue that AAPS depends on water–drug, water–polymer and drug–polymer interactions and on hygroscopicity, so a single descriptor cannot capture it. They then propose that "There appears to be" two key determinants of release, the polymer dissolution rate and congruency, and that drug T_g and drug–polymer H-bonding "can potentially" affect either. The chain below summarizes the picture; none of its middle steps was measured in this paper.

Two routes feed the bad branch. By the H-bonding route, inferred from the indomethacin comparison, strong drug–polymer interactions "are likely to" lower the polymer release rate and with it the drug release rate. By the T_g route, a high-T_g drug or a glassy drug-rich phase slows water entry and chain disentanglement, while a low-T_g drug can plasticize. For any drug, more low-polarity drug "would be anticipated" to reduce water entry and so slow the polymer. The 5 °C experiment is the authors' main argument for the T_g route ("This appears to be a Tg effect"), but they also give "an alternative explanation": because the ternary drug–polymer–water phase diagram changes with temperature (Manjkow 1987), drug–polymer miscibility in water may fall at 5 °C, so a drug-rich phase forms at a much lower DL. They conclude that both T_g and H-bonding matter, high T_g acting on kinetics and H-bonding on thermodynamics.

The polymer comparison. For nanodroplet stability, the authors point to adsorption of HPMCAS on nanodroplet surfaces, which "has been hypothesized" before. HPMCAS is negatively charged at the study pH, so adsorbed chains can add electrostatic to steric hindrance, which neutral PVPVA cannot. Cited evidence supports adsorption: HPMCAS lowers the zeta potential of danazol nanodroplets (Jackson 2016) and adsorbs more strongly than PVPVA on loratadine nanoparticles (Zhang 2017). The exact mechanism "clearly requires more investigation". For release, the gentler decline of HPMCAS ASDs means some nanodroplets form over a wider DL range even when release is incongruent.

Where the findings apply. The evidence covers binary ASDs of six drugs, in buffers chosen to keep them unionized, with PVPVA and one HPMCAS grade (MF), made by rotary evaporation and tested as 149–250 μm powders at 37 °C, plus loratadine compacts at 37 and 5 °C. The PVPVA powder results hold for media containing 400 μg/mL dissolved HPMCAS. Release limits are read at one time point from a non-sink test with 100 μg/mL total drug, with no surfactant, second polymer, biorelevant medium, membrane flux or in vivo data. Within these conditions the ordering by wet T_g holds across all systems, and the PVPVA–HPMCAS contrast across all drugs except elbasvir; each loading limit is accurate to about one tested DL step.

What remains unresolved. The central open question is the one the authors name: whether the loading limit is set by T_g, by H-bonding or by both. No low-T_g drug with donors (other than the cited indomethacin) and no high-T_g drug without donors was tested, and H-bonds or halogen bonds in these ASDs were not measured. The 5 °C test changes T_g without changing chemistry but also changes phase behaviour, so it supports the T_g route without excluding the miscibility route. The role of AAPS "remains unclear": some studies link it to poor performance, others show phase-separated domains that improve release or nanodroplets that originate from AAPS (Li 2019; Harmon 2016). Whether the nanodroplets formed at 5 °C were glassy was not measured.

Implications for further work and for formulation. The authors draw two practical conclusions. Log P "shows little indication" of ASD performance, so the ASD formulation space "is not limited by" log P, even above the value of 5 that Lipinski's rule treats as unfavourable. And T_g "should be considered as a potentially determining factor for API drug loading"; since T_g tends to rise with molecular weight, "More nuanced ASD strategies may be necessary". Note by the AI writer (not discussed in the paper): with PVPVA the switch from good to poor release spans about one 5–10% DL step, so a chosen loading needs a margin below the LoC. For high-T_g drugs with donors, PVPVA allowed good release only to 10% DL or less here, whereas HPMCAS kept partial release over a wider range, though not for elbasvir. Tests below 37 °C can understate the loading limit of a drug whose wet T_g lies between the test temperature and 37 °C. A natural next step is a drug pair that separates T_g from H-bonding, with direct measurement of AAPS in the hydrated ASDs.

Limitations

Limitations stated by the authors. The authors acknowledge that their drug set confounds two factors: poor release at low DL "may be" an H-bonding effect, a T_g effect "or a combination of both factors". They write that "Clearly, further studies need" to be performed to separate the kinetic effect of high T_g from the thermodynamic effect of H-bonding, and they give "an alternative explanation" for the 5 °C result. They describe the link between AAPS and release as unclear, the nanodroplet stabilization mechanism as needing more investigation, and the role of T_g in drug loading as something that "needs more investigation". On method, wet T_g could be measured by DSC only for loratadine, danazol had to be read at 15 min because it crystallized, and elbasvir's amorphous solubility could be obtained only by NTA.

The AI writer's own judgement (1): what the design can show. The extractor rates the evidence as moderate overall. The cliff-versus-slope contrast and the T_g ordering hold across six drugs and two polymers in triplicate, and the cooled-loratadine result agrees. But T_g and H-bond donors are fully confounded, so the data support "wet T_g and/or H-bond donors", not T_g alone. The log P conclusion rests on six drugs without a statistic. Paclitaxel and ivacaftor are classed high-T_g on cited dry T_g only, and their test conditions are not stated. None of the proposed mechanisms (AAPS, χ, water uptake, ASD T_g, adsorption, zeta potential, phase behaviour at 5 °C) was measured, and the compact LoC exists for loratadine only.

The AI writer's own judgement (2): the polymer comparison. The PVPVA and HPMCAS powder data are not like for like: every PVPVA powder dissolved in a medium containing 400 μg/mL dissolved HPMCAS, while HPMCAS powders dissolved in plain buffer. The PVPVA results therefore describe a PVPVA ASD helped by dissolved HPMCAS, which limits comparison with other PVPVA studies. The HPMCAS release advantage holds at high DL, not at every DL, and its nanodroplet advantage is clearly shown for elbasvir and loratadine only; with dissolved HPMCAS, danazol nanodroplets still grew to about 2780 nm.

The AI writer's own judgement (3): statements stronger than the plots, and uncertain inputs. Several results are stated more strongly than the figures show: 30% DL loratadine–PVPVA compacts are congruent but about 85% released by 90 min; the 2% DL compact at 5 °C about 78% by 180 min; high-T_g HPMCAS ASDs at 50% DL lie above the amorphous solubility although called nanodroplet-poor; and the polymer-rate argument at 5 °C rests on HPMCAS only. Nanodroplet calls at 5 °C use the 37 °C amorphous solubility, since it was not measured at 5 °C. Elbasvir's amorphous solubility rests on one NTA value, and its higher pKa (5.88) is only 1.62 units below the pH 7.5 buffer, so partial ionization is not excluded by the paper's own rule. The Table 1 molecular weight of miconazole (479.1 g/mol) matches a nitrate salt, so the basis of its % w/w DL is unresolved. Labelling errors (the Figure 1a loratadine structure, drug legends in Figure 5a–f, the "35:75" title of Figure 9c, the 2% DL caption in Figure S3) and small inconsistencies (danazol–HPMCAS time point in Figure 3c, miconazole–HPMCAS at 65% DL, three or six replicates for crystalline solubility) do not change the main conclusions, but individual plotted values should be read with care.

Conclusion

The paper asked whether drug log P or drug wet T_g decides how release and nanodroplet formation from PVPVA and HPMCAS dispersions change with drug loading. Log P did not: drugs with similar log P failed at very different loadings, and drugs with very different log P at similar ones. The two low wet-T_g drugs, which also lack H-bond donors, released well to much higher loadings than the four high-T_g drugs with donors, and cooling loratadine below its wet T_g cut its compact LoC from 30 to 2% DL with PVPVA and from 20 to 5% DL with HPMCAS. Loss of drug release went with loss of congruency, and for high-T_g drugs at high loading the polymer itself was held back. PVPVA dispersions fell off a cliff within about one DL step, HPMCAS dispersions declined gradually, and HPMCAS kept elbasvir and loratadine nanodroplets small.

The authors' summary is that the findings "point to the potential importance" of drug T_g and/or drug–polymer interactions, leaving their separation to further work. For formulation, the paper removes log P as a gatekeeper for ASD candidates and adds wet T_g and H-bond donors as first checks on how much drug a dispersion can carry. It also gives a reason to consider HPMCAS for high-T_g drugs at medium to high loading, to be confirmed without dissolved HPMCAS in the PVPVA media.