A few nanometers of oxide: atomic layer coating to slow crystallization of amorphous solid dispersion powders

Background and motivation

An amorphous drug dissolves faster and reaches higher concentrations than its crystal, which is why amorphous solid dispersions (ASDs), in which the drug is mixed at the molecular level into a polymer, are a standard answer to poor solubility (Jermain et al. 2018, ref 1). The amorphous form, however, always has a driving force to crystallize (Williams et al. 2013, ref 2). The polymer slows crystallization, but only within limits set largely by drug loading (DL), and even trace crystallinity can lower the supersaturation reached on dissolution and with it bioavailability (refs 5–10, among them Moseson et al. 2020 and Purohit et al. 2019). Any step that keeps an ASD amorphous for longer, or at a higher DL, is therefore of direct practical value.

The usual levers on stability are drug–polymer interactions and molecular mobility, the latter indexed by the glass transition temperature ($T_g$). A rule of thumb (Hancock and Shamblin 2001, ref 13) holds that an ASD is kinetically stable when stored at least 50 °C below its $T_g$, although Theil et al. (refs 14, 15) reported 15–25 years of stability for samples stored only about 10 °C below $T_g$. Water complicates the picture. An amorphous solid takes water into its bulk, which lowers $T_g$ even at low levels (Mehta et al. 2016, ref 16), and water can reduce drug–polymer miscibility so that the system phase-separates and then crystallizes (Rumondor et al. 2009, ref 17). The manufacturing route changes crystallization rates as well (refs 18–21). Each factor caps the DL at which an ASD can be made.

One alternative targets the particle surface. Crystals grow at the free surface of an amorphous solid orders of magnitude faster than in its interior (Shi and Cai 2016, ref 29), so suppressing mobility at the surface alone could remove much of the problem. Thin films of poly(vinylpyrrolidone), dextran, alginate, gold, polyelectrolyte, gelatin and chitosan have delayed crystallization of amorphous films and particles (refs 30–36), and Yu and co-workers linked the effect to a thin, mobile surface layer that a coating can arrest (for example Wu et al. 2007, ref 34). These coatings, however, hold on through non-covalent forces and generally need relatively thick layers applied from solvent. Conventional pharmaceutical film coats are tens to hundreds of micrometers thick (ref 38) and suit tablets rather than single particles, and solvent-free alternatives need high temperatures or long curing (refs 43–46).

Atomic layer coating (ALC), also called atomic layer deposition, is a solvent-free gas-phase process that builds a metal-oxide film, commonly thinner than 10–100 nm, one atomic layer per cycle, with fine control of thickness and conformality (George 2010, ref 47; Parsons et al. 2013, ref 49). It is established in semiconductors, solar cells and medical devices and has improved the flow of organic drug powders (refs 25, 50–53). In the group's earlier study (Van Duong et al. 2022, ref 25), an aluminum oxide (Al2O3) coating kept a spray-dried ASD of 70% ezetimibe in hydroxypropyl methylcellulose acetate succinate (HPMCAS) amorphous at 40 °C and 75% relative humidity (RH) for about 2 years. There, coatings thinner than 20 nm and thicker than 30 nm differed in onset, and water sorption fell as the coating thickened. Jur et al. (2010, ref 54) had shown that substrate chemistry shapes the film: on a substrate with hydroxyl groups the coating forms an abrupt, well-defined interface, whereas without reactive groups the precursor and water diffuse below the surface, causing swelling, subsurface growth and a rough texture.

The paper asks whether the ezetimibe result is general, and why it works. The authors chose four drug–polymer systems that differ in whether drug and polymer carry hydroxyl groups, the presumed anchoring sites, and in polymer hygroscopicity. They made them by spray drying (SD) or hot melt extrusion (HME), which gives different particle sizes, and coated them with Al2O3 or zinc oxide (ZnO) at several thicknesses, hoping to find a minimum effective thickness. Both oxides were chosen as biocompatible, inert materials already used in cosmetics, food and medicines. Crystallization was followed by scanning electron microscopy (SEM) and X-ray powder diffraction (XRPD) for up to 48 weeks, with differential scanning calorimetry (DSC) and dynamic vapor sorption (DVS) as supporting methods.

Background concepts

Glass transition and the distance to it. Below its glass transition an amorphous solid is a glass in which molecules move slowly; above it, the material is a rubbery, supercooled liquid in which diffusion and crystal growth are much faster. The gap between $T_g$ and the storage temperature $T$ is therefore a rough index of mobility. The rule of thumb the authors cite (ref 13) reads:

In plain terms: an ASD stored at least 50 °C below its glass transition is expected to stay kinetically stable. Because water lowers $T_g$, what matters in humid storage is the $T_g$ of the powder after it has taken up water. The authors call this the "wet" $T_g$: each powder was held for 24 h at its storage condition, sealed (3–5 mg) in a hermetic DSC pan and scanned at 5 °C/min, in triplicate. They then compute, for each storage condition (Figure 10):

In plain terms: a positive value means the humidified powder was stored below its glass transition, a negative value that it was stored above it, in the rubbery state. Only these differences are printed, not the wet $T_g$ values.Card check: the text defines the quantity in words as the wet $T_g$ "subtracted from the storage temperature", literally the reverse difference; the sign convention the authors use (negative means stored above $T_g$) is followed here (card note X24).

Surface crystallization and surface mobility. Molecules in the outermost few nanometers of an organic glass are less constrained than those inside, and crystals can grow at the free surface far faster than in the bulk (ref 29); Sun et al. (2011, ref 64) showed that they grow upward, out of the surface. A coating that removes the free surface or binds to it could then slow the dominant route even with the bulk unchanged. The paper builds on this idea but does not measure surface mobility itself.

Absorption and adsorption of water. A crystalline solid holds water mainly on its surface (adsorption); an amorphous solid also takes it into the bulk (absorption), which is why its uptake is higher and why water plasticizes it. A DVS isotherm records only the total mass gained and cannot separate the two. This distinction carries one of the authors' two proposed mechanisms.

How ALC builds a film. Each cycle has two half-cycles (Figure 2). A metal precursor, trimethylaluminum for Al2O3 or diethylzinc for ZnO, is dosed and sorbs on the substrate, and the excess is purged with nitrogen. Water is then dosed as co-reactant; it forms the oxide and leaves new surface hydroxyls for the next cycle, and a second purge follows. A coating is conformal when it has the same thickness on every surface. The precursor needs reactive groups such as hydroxyls to anchor to; where they are lacking, it can diffuse into a polymer and react below the surface, which the paper calls infiltration (refs 49, 54).

Measuring a nanometer coating. In thermogravimetric analysis (TGA) the coated powder is burned in air to 800 °C; the residue is taken as the oxide content and, combined with the specific surface area (SSA, from nitrogen adsorption and Brunauer–Emmett–Teller (BET) theory), gives an average thickness. The paper describes this conversion in words only. Cross-sectional transmission electron microscopy (TEM) images the coating directly on a slice about 100 nm thick cut by a focused ion beam (FIB), but only at a few places on a few particles.

Powder flow and crystal detection. The FT4 shear cell gives the flow function (FF), classed by the authors as not flowing (below 1), very cohesive (1–2), cohesive (2–4), easy flowing (4–10) and free flowing (above 10); compressibility is the percent volume lost under 15 kPa. Higher FF and lower compressibility mean better flow. For crystals, SEM images individual particles and can show small surface crystals, while XRPD averages over the sample; the authors used SEM as the primary tool because microscopy is more sensitive (refs 9, 25, 59, 63). "Onset" below means the first storage time at which SEM showed crystals.

Methods in detail

The work splits into three contributions. The first builds a set of ASD powders that differ in hydroxyl groups, polymer, route and particle size, coats them, and measures the coatings. The second checks what the coating changes in the powder itself. The third stores coated and uncoated powders side by side under accelerated conditions, records when crystals appear, and interprets the failures through the wet $T_g$ and a TEM view of a crystallized coated particle.

Contribution 1: A set of powders built around anchoring groups, and the coatings on them

① Problem. The ezetimibe result came from one drug, one polymer and one process. Whether a coating forms well and protects could depend on hydroxyl groups that anchor the precursor, polymer hygroscopicity, manufacturing route, particle size, the oxide and its thickness. A test of generality needs systems that span these, and a measure of the coatings actually obtained.

② What they did. The drugs were erlotinib (ERL), naproxen (NAP) and lumefantrine (LUM), chosen for their high crystallization tendency (refs 24, 27, 57). Table 1 gives melting points and $T_g$ onsets of 155 and 54 °C (ERL), 156 and 6 °C (NAP, $T_g$ from ref 55 because amorphous NAP crystallizes on cooling) and 128 and 21 °C (LUM). The polymers were HPMCAS (M grade, $T_g$ 114 °C) and the vinylpyrrolidone–vinyl acetate copolymer PVPVA (Kollidon VA64, $T_g$ 103 °C). Four compositions were made: EH60 (60 wt% ERL in HPMCAS, SD), EP50 (50 wt% ERL in PVPVA, SD), NH10 (10 wt% NAP in HPMCAS, HME) and LP40 (40 wt% LUM in PVPVA, by SD and by HME). Figure 3 places them on a grid of drug against polymer hydroxyls: EH60 has hydroxyls in the polymer only, EP50 none, NH10 in both and LP40 in the drug only; hygroscopicity is ordered HPMCAS below PVPVA. Spray drying used a Buchi 290 with nitrogen and 10% solids, from methanol/dichloromethane (ERL) or ethyl acetate/dichloromethane (LUM) (Table S2). Extrusion used a Haake MiniCTW at 160 °C (NH10) or 150 °C (LP40) and 200 rpm (Table S3), followed by cryomilling and sieving to 0–250 µm (NH10), 0–106 µm (LP40-HME-A) or 106–250 µm (LP40-HME-B).Card check: Table 2 prints "LP40-HME-B" for both LP40 sieve fractions; all other tables and figures assign the 0–106 µm fraction to HME-A (card note X1).

Coating used a rotary reactor that agitates the powder, at 35 °C and 0.1 Torr base pressure, with precursor amounts set from the BET area and the powder mass; the number of cycles is not printed. EH60, EP50, LP40-SD and LP40-HME-A each received two Al2O3 thicknesses, NH10 and LP40-HME-B one, and LP40-SD also one ZnO coating, giving 11 coated powders. All were measured by TGA (25 to 800 °C at 10 °C/min in dry air) and six by FIB–TEM at 200 kV; energy-dispersive X-ray spectra (EDS) were acquired but not shown. The TGA thickness serves as the sample label.

③ Why it works. The hydroxyl grid turns the observation of Jur et al. into a prediction that can fail: if anchoring matters, the hydroxyl-free EP50 should carry the poorest coating and gain least. The LP40 composition, made by SD and as two extruded sieve fractions, is the one place where chemistry is fixed while route and particle size change. TGA and TEM together give both an average thickness and a direct view of the interface.

④ How it differs from earlier work. The ezetimibe study coated one spray-dried system; this one covers four compositions, two routes, three particle-size classes for one composition, two oxides and three humidities. Note by the AI writer (not discussed in the paper): the four systems also differ at once in drug, polymer, DL (10–60%), route, size and storage RH, and each hydroxyl class is one system (card note X20). The grid is a set of four case studies rather than a factorial design; only the LP40 series varies one factor at a time.

Contribution 2: What the coating changes in the powder

① Problem. A coating could protect a powder for reasons unrelated to its surface: by raising $T_g$, slowing water uptake or changing particle size. These must be checked before a surface mechanism can be argued, and the effect on handling matters in its own right.

② What they did. Particle size was measured by laser diffraction in triplicate (Figure S1). Flow function and compressibility at 15 kPa were measured on an FT4 rheometer for the LP40 powders only, with paired one-tailed t-tests against the uncoated powder; the replicate number is not printed (Figure 5). DSC in standard (non-hermetic) pans at 5 °C/min from 0 to 200 °C gave the dry $T_g$ (Table S1) and showed any recrystallization on heating (Figure 6). Water sorption and desorption were recorded at 40 °C from 0 to 95% RH, each step ending at less than 0.01% mass change in 5 min (Figure 7, Figure S2).

③ Why it works. This is an exclusion step. If water uptake and $T_g$ are the same with and without the coating, a delay cannot be traced to a moisture barrier or to a change in bulk plasticization, as far as these methods see. What remains must lie in something the bulk measurements miss, which is where the authors' surface argument starts.

④ How it differs from earlier work. In the ezetimibe study water sorption fell as the coating thickened, so a barrier effect could not be separated from a surface effect. The thinner coatings here, which the authors describe as 5–25 nm, let the question be asked directly: does protection persist when water uptake is unchanged?

Contribution 3: Accelerated storage, the wet glass transition and a failed particle

① Problem. Onset must be detected sensitively and compared fairly between coated and uncoated powders, and the cases where coated powders still crystallize need an explanation.

② What they did. Powders were stored in open dishes inside sealed containers at 40 °C over saturated salt solutions giving 31, 43 or 75% RH: EH60 at 43 and 75%, EP50 at 43%, NH10 at 75%, LP40-SD at 31 and 43%, and both LP40-HME fractions at 31%. This gives eight system/humidity sets, each with one uncoated control and one to three coated powders, 13 coated arms in all. SEM (platinum-coated particles, 5 kV) and XRPD (Cu Kα, 5–35° 2θ) were run periodically for up to 48 weeks; the XRPD series for NH10, EP50 and EH60 at 75% RH end at 24 weeks. Sample numbers per arm and SEM sampling times are not printed. Wet $T_g$ was converted into $T_g-T$ (Figure 10), and a coated EH60 particle that had crystallized after 12 weeks at 40 °C/75% RH was cut and imaged by TEM (Figure 11).

③ Why it works. Each coated powder is stored beside its uncoated parent, so a difference in onset can be credited to the coating rather than to storage. $T_g-T$ places arms of different chemistry and humidity on one mobility scale. A cross-section of a failed particle can show whether crystals form beneath the coating, through a gap in it, or above it.

④ How it differs from earlier work. The ezetimibe study used one storage condition. Here three humidities are used, the wet $T_g$ is measured at each, and the failure of a coated particle is examined directly rather than only scored.

Experimental analysis

Experimental setup. Six uncoated powders (four compositions, with LP40 made three ways) and eleven coated powders were characterized, and thirteen coated arms were stored with their eight uncoated controls at 40 °C. Table A summarizes the starting powders and their coatings. D50 spans a factor of about 20, from 7.43 µm (EH60) to 151 µm (LP40-HME-B).

Table A. Uncoated ASD powders and their coatings (from Table 2 and Table 3). D50 by laser diffraction, n = 3; D50 and BET area are for the uncoated powders.

Result 1: the coatings obtained. By TGA the coatings range from 7.4 to 24 nm, by TEM from 4.4 to 37.7 nm. The authors call the two "generally consistent", but the TEM/TGA ratio runs from about 0.6–0.7 (EP50 7.4 nm) to about 1.5–2.0 (EH60 13 nm; LP40-HME-A 19 nm), so thicknesses compare safely only within one method (card note X10). Two LP40-SD coatings labelled 8.0 and 11 nm share the same oxide content, 3.47% (card note X2). Note by the AI writer (not discussed in the paper): every other TGA thickness in Table A is reproduced within a few percent by dividing the oxide content by the BET area and assuming an Al2O3 density of about 3.0 g/cm³, a back-calculated value. An 11 nm coating on LP40-SD would then need about 4.8% oxide, so one of the two printed values is probably a transcription error. The same arithmetic shows that the TGA label assumes a uniform shell on the outer surface; infiltration, uneven coverage or uncoated inner surfaces would be hidden inside it.

The TEM images (Figure 4) show three kinds of coating: on EH60 an Al2O3 layer (marked 20.3 nm) with a sharp boundary; on the hydroxyl-free EP50 a layer marked 9.4 nm over a lighter band labelled "infiltration"; on LP40-SD a ZnO layer (marked 8.6 and 9.6 nm) with mottled contrast. The authors state that most systems carried a conformal coating, that infiltration on EP50 "is suggested by the TEM imaging", and that ZnO "appears to be less conformal" owing to its microcrystalline character (ref 61). This matches the Jur et al. picture but rests on one image per system, and the EDS data that could show aluminum below the EP50 surface were not shown.

Result 2: what the coating did and did not change in the bulk. Table B sets the dry $T_g$ and the water uptake of each powder against its coated versions.

Table B. Dry glass transition onset (from Table S1; DSC at 5 °C/min, standard pans) and water uptake at 40 °C (read from Figure 7 and Figure S2, approximate). For NH10 and the LP40 powders, coated and uncoated sorption curves agree within about 1% up to 75% RH.

Three points follow. First, up to 75% RH, which covers every storage condition, coated and uncoated powders take up the same water within about 1%; differences appear only at 85–95% RH and go both ways. The authors conclude that 5–25 nm coatings leave sorption unchanged, so they are not a moisture barrier at these thicknesses. Second, Al2O3 shifts the dry $T_g$ by only −1.8 to +2.6 °C, but ZnO lowers that of LP40-SD from 65.5 to 57.3 °C, so the statement that the coating "does not alter" $T_g$ holds for Al2O3 only (card note X12). Third, the dry $T_g$ values are themselves puzzling: EH60 and EP50 (43.9–44.6 °C) lie below the $T_g$ of every component (ERL 54, HPMCAS 114, PVPVA 103 °C), and one LP40 composition differs by about 12–13 °C between SD and HME. Water content before DSC is not reported and the pans were not sealed (card note X13), so these values serve only to compare coated with uncoated versions of one powder.

Uncoated EH60 has a sorption feature of its own (Figure 7a): its mass falls between 85 and 95% RH and its desorption branch lies below the sorption branch. The authors read this as crystallization during the run, prevented by the coating; no solid-state check after DVS is shown (card note X22). On DSC heating (Figure 6), uncoated EH60 shows a sharp recrystallization exotherm near 105 °C and melting near 150 °C, coated EH60 only a small, broad exotherm between about 105 and 125 °C; uncoated EP50 shows an exotherm near 122 °C, coated EP50 only weak exotherms near 130 °C (read from Figure 6, approximate). The authors add that the effect grows with thickness, but no enthalpies are printed and the 13 and 14 nm EH60 traces look alike (card note X23). Note by the AI writer (not discussed in the paper): this is the only test in which the coating acts within minutes and without humidity, which fits crystallization on heating also starting at particle surfaces.

Result 3: particle size and flow. The authors state that ALC does not alter particle size, and for NH10, EH60 and LP40-HME-B the distributions nearly superimpose (Figure S1). For LP40-HME-A, however, the main mode moves from about 40 µm uncoated to about 27 µm (19 nm) and 16 µm (11 nm) (card note X11). Flow was measured on LP40 only (Table C).

Table C. FT4 flow function and compressibility of the LP40 powders (read from Figure 5, approximate). p values from paired, one-tailed t-tests against the uncoated powder, as marked in Figure 5; flow classes as defined in the paper.

Every coated LP40 powder moved up at least one flow class, the 0–106 µm extruded fraction from very cohesive to free flowing, and compressibility fell in all four pairs, significantly in three. The authors call the flow gain consistent "regardless of particle size", in line with earlier ALC work (refs 25, 52, 62). Stored powders agree (Figure S10): after 2 weeks at 40 °C/75% RH (EH60) or 3 weeks at 40 °C/43% RH (EP50), coated powders were loose while uncoated ones had clumped. The caveats are that only LP40 was tested, replicates are not printed, and coated LP40-HME-A was finer than its parent. Note by the AI writer (not discussed in the paper): finer powders are normally more cohesive, so the size shift does not explain the gain, but the pair is not size-matched.

Result 4: crystallization in storage. Table D, the core of the paper, brings together the SEM onset, the wet $T_g-T$ and the XRPD patterns for all eight system/humidity sets.

Table D. SEM onset at 40 °C (read from Figure 9, approximate, cross-checked with the table in Figure 10), wet $T_g-T$ (read from Figure 10, approximate; the two values with errors are printed in the text) and XRPD findings (card reading of the plotted patterns; the authors give no peak list).

Every coated arm crystallized later than its uncoated control stored beside it. Uncoated powders showed crystals in about 1–2 weeks in all eight sets; coated powders at about 3 weeks (EH60, 75% RH), about 6 weeks (EP50), about 24 weeks (NH10; LP40-SD, 43% RH), or not within 48 weeks in 7 of 13 arms. In SEM (Figure 8), uncoated EH60 and EP50 particles fused into agglomerates bearing faceted plates and uncoated NH10 and LP40-HME carried needles or rods, while coated particles stayed discrete and smooth. The authors' wording is "effectively delayed or completely inhibited".Card check: uncoated NH10 and LP40-SD (43% RH) onsets are about 1 week in the Figure 9 bars but 2 weeks in the Figure 10 table, hence "1–2 weeks" (X5); the results text gives "1−4 weeks" for uncoated powders, the abstract and both figures 1–2 weeks (X6); the NH10 coating is labelled 8 nm in Figure 9 but 9.8 nm in Table 3, and the thicker EP50 coating 14 nm in Table 3 but 14.5 nm in the figures (X3).

Hydroxyl groups and polymer. The anchoring hypothesis predicted that hydroxyl-free EP50 would gain least, yet its onset still moved from about 2 to about 6 weeks despite the partial infiltration seen by TEM. Every hydroxyl class gained, and the authors note that ALC also works on substrates with amines (ref 66). The erlotinib pair is their evidence that the polymer matters more: at 43% RH, EH60 and EP50 sat at similar $T_g-T$ (−2.3 ± 1.1 and −0.5 ± 1.2 °C), yet coated EP50 crystallized at about 6 weeks while coated EH60 held for 48 weeks. Citing also the higher ERL loading in HPMCAS (60 vs 50%) and Mudie et al. (ref 27), they write that "the study results suggest" HPMCAS suits ERL better, with acid–base interactions "likely the reason" (ref 24). Each loading was made once, so this describes two formulations, not a measured limit (card note X19).

Thickness, oxide, route and size. Paired thicknesses behaved alike: EH60 with 13 and 14 nm both failed at about 3 weeks, EP50 with 7.4 and 14 nm both at about 6 weeks, and LP40-SD (8, 11 nm) and LP40-HME-A (11, 19 nm) all held to 48 weeks. The authors conclude that "No minimum coating thickness was required", with 7–8 nm sufficient and "limited differentiation" from coatings above 15 nm. ZnO on LP40-SD held to 48 weeks at 31% RH, as Al2O3 did, despite its less conformal look. LP40 crystallized in about 1–2 weeks uncoated whether spray dried (D50 13.6 µm) or extruded (28.5 or 151 µm), and held to 48 weeks coated in all three forms.

Humidity. Humidity is the one factor that changed the outcome for a given coated powder: coated EH60 (13 nm) held at 43% RH but failed at about 3 weeks at 75% RH, and coated LP40-SD (8 nm Al2O3) held at 31% RH but failed at about 24 weeks at 43% RH. The authors attribute this to moisture lowering the wet $T_g$ (refs 67, 68). Uncoated onsets did not shorten with humidity (EH60 about 2 weeks at both; LP40-SD about 2 and 1–2 weeks), so the effect is seen only in coated arms (card note X16). Note by the AI writer (not discussed in the paper): with uncoated onsets of 1–2 weeks and no printed sampling schedule, the uncoated arms may have had no room to show a difference.

Result 5: onset and the wet glass transition. The authors write that onset "can be linked" to $T_g-T$: the least stable arms had negative values, longer-lived arms "generally" positive ones, and "This confirms that the mobility of the sample was the primary mechanism" by which coated systems still crystallized. Their causal chain is summarized below.

The data fit this chain only in part. $T_g-T$ ran from about −8.8 to about +14.3 °C (text: −10 to +15 °C), and the two coated arms that failed soonest were stored above their wet $T_g$. But coated EH60 at 43% RH, also above it (−2.3 °C), never crystallized, while coated NH10 and LP40-SD at 43% RH, about 11 °C below it, crystallized at about 24 weeks; and uncoated onsets were 1–2 weeks whatever the $T_g-T$. Whether wet $T_g$ was measured on coated or uncoated powder is not stated (card note X15), and the text places EP50 at 75% RH although it was stored only at 43% RH (card note X7). $T_g-T$ is one factor among several here, and "confirms" is stronger than the data allow.

Note by the AI writer (not discussed in the paper): adding 40 °C to the plotted differences gives implied wet $T_g$ values of about 51 °C for both LP40-HME fractions at 31% RH, only 1–2 °C below their dry $T_g$, but about 54 °C for LP40-SD, about 11 °C below its dry 65.5 °C. As all LP40 forms sorbed water alike, this mismatch is another sign that the Table S1 samples were not equally dry (card note X13).

Result 6: how a coated particle fails. A coated EH60 particle after 12 weeks at 40 °C/75% RH carried surface crystals (Figure 11a), and TEM cross-sections showed crystals lying above a continuous ALC layer marked 8–9 nm and 7–8 nm (Figure 11b, d), with the film present even beneath one outgrowth (Figure 11c). The authors read this as crystals growing out of the particle, as is common for amorphous materials (ref 64), with larger ones breaking off and lying flat on the coating; the film under the outgrowth "more likely suggests" that the small failure site was missed by the FIB cut. The coating next to the crystals (7–9 nm) was thinner than in "initial measurements by TEM (∼14 nm)", and "It cannot be determined" whether storage or particle-to-particle variation is the cause.Card check: Table 3 gives a TEM thickness of 20–21 nm for the EH60 13 nm sample and none for the 14 nm sample, so "∼14 nm" matches a TGA value, and Figure 11 does not say which coating was imaged; the thinning comparison is unresolved (card note X8).

Result 7: SEM against XRPD. The authors state that XRPD saw crystals "at a slightly longer time point" than SEM. Table D shows a wider gap: uncoated LP40-HME-A and LP40-HME-B show no Bragg peaks through 48 weeks and uncoated LP40-SD at 31% RH only a faint feature at 48 weeks, against SEM onsets of 1–2 weeks, and coated EP50, NH10 and LP40-SD (43% RH), scored as failed at 6–24 weeks, show no or faint peaks (card note X14). This fits the premise that microscopy is more sensitive, but these onsets rest on SEM alone, without replicates.

Controls and what they rule out. The strong control is the matched uncoated powder, and that comparison favors the coating in all 13 arms. The sorption data rule out a bulk moisture barrier at these thicknesses, the dry $T_g$ data a bulk effect of Al2O3, and the LP40 series a large effect of route and size at 31% RH. Note by the AI writer (not discussed in the paper): the controls cannot separate the authors' two proposed routes, since neither surface water adsorption nor surface mobility was measured, nor can they exclude a purely mechanical effect in which the shell hinders crystals growing upward out of the surface. All three readings predict a delay with unchanged water uptake and $T_g$.

Discussion

Where the result applies. The evidence covers four ASDs (10–60% DL) of three crystallization-prone drugs in HPMCAS or PVPVA, made by SD or HME, coated at 35 °C with 7–24 nm (TGA) of Al2O3 or 8.7 nm of ZnO, and stored open at 40 °C and 31–75% RH. Within that frame the delay is consistent: every coated arm outlasted its control. Its size depended on the condition, from about one extra week for EH60 at 75% RH to more than 46 extra weeks in the arms that never crystallized. The authors describe a "clear benefit for systems at high risk to crystallize", stabilizing ASDs at $T_g-T$ of about 0–15 °C, far below the 50 °C rule; that range leaves out coated EH60 at 43% RH (−2.3 °C), which also held.

What remains unresolved. The mechanism is the main open question. The authors propose two routes. In the first, the coating "would block water adsorption to the particle surface", so that an uncoated particle has a more mobile surface even at the same total uptake, a difference that "would be too small to detect". In the second, the coating "may directly inhibit mobility at the surface" by replacing the surface–air interface with an ASD–oxide interface, bonded where hydroxyls exist and infiltrated where they do not. The abstract settles on "decreased surface molecular mobility" as the primary cause, yet surface mobility, surface water and coating–substrate bonding were not measured; the mechanism is an inference from the exclusion of a bulk barrier and from cited work (refs 33–36). Thickness is also open: by analogy with the ezetimibe study, the authors suggest failure "may be based on coating thickness, potentially due to coating conformality", with all coatings here in the thinner category.

What it suggests for further work. The authors name the release behavior of coated particles as the next question, since coatings have both sped and slowed release in other work, and they flag that spray-dried particles may be hollow, with inner surfaces the coating cannot reach. Note by the AI writer (not discussed in the paper): three experiments would sharpen the conclusions. Surface crystal growth rates on coated and uncoated amorphous films of one drug would test the surface-mobility route directly. Coatings above about 20–30 nm should be compared where thin ones fail, for instance EH60 at 75% RH. A drug-loading series, coated and uncoated, would test whether ALC raises the usable DL.

What it means for formulation. The authors see ALC as having "the potential to be a scalable strategy" for higher DL and robustness to temperature or humidity excursions, and point to better flow and less agglomeration, which matter most for cohesive particles below about 30 µm. Note by the AI writer (not discussed in the paper): the coating acts on the surface, not the bulk. Where humidity made the bulk clearly mobile (EH60 at 75% RH, about 9 °C above its wet $T_g$) it bought a week; where the powder sat below its wet $T_g$ it bought months. It adds to, rather than replaces, a sound choice of polymer, DL and moisture protection, and at these thicknesses it is no substitute for protective packaging. The mild, solvent-free process suits low-$T_g$ ASDs, and the flow gain may be worth having on its own, but dissolution from coated particles must be checked first.

Limitations

Limitations stated by the authors. The thin coatings did not reduce water sorption, "indicating that the water permeation rate was not sufficiently reduced", and differences in surface water adsorption "would be too small to detect". At the failed particle, finding the exact failure site by FIB cutting "is statistically difficult", and whether the thinner coating next to the crystals reflects storage or particle-to-particle variation "cannot be determined", because the TEM work was done for morphology, not statistics. Hollow spray-dried particles may leave inner surfaces uncoated and so be "more susceptible to crystallization". The mechanisms of release from coated particles "warrant future investigation".

The AI writer's own judgement. The card rates the evidence as moderate. The delay itself is well supported: it appears in all eight system/humidity sets against a matched control, and SEM images, powder photographs and DSC agree. Several things weaken the wider conclusions.

Onset precision. Onset times carry no replicates or uncertainty, and the SEM sampling schedule is not printed, so the resolution of "about 3 weeks" or "about 6 weeks" is unknown (card note X21). XRPD does not confirm several arms: it shows no crystals through 48 weeks in uncoated LP40-HME, where SEM saw them at 1 week, and none or faint peaks in coated arms scored as failed (X14).

Generality. Each hydroxyl class is one system, and drug, polymer, DL, route, size and RH change together (X20). ZnO was tested only on LP40-SD at 31% RH, where all Al2O3 coatings also held, and NH10 only at 75% RH. The statement that "all systems were successfully prevented from crystallization" overstates arms that failed at 3–24 weeks, for which "delayed" is the accurate word (X18).

Thickness and mobility. Thickness independence was never tested above about 15 nm under conditions where coated powders failed, and TEM puts the LP40-HME-A 19 nm and LP40-HME-B 24 nm coatings above 20 nm, outside the "thinner category" the authors assign (X17, X10). Wet $T_g-T$ only partly orders the failures, and the humidity effect is seen only in coated arms (X15, X16), so "This confirms" overstates the role of bulk mobility.

Claims beyond the data. ZnO lowered the dry $T_g$ by about 8 °C although $T_g$ is said to be unaltered (X12); coated LP40-HME-A was finer than its parent although size is said to be unaltered (X11); and the statement that ALC enables a higher DL rests on one DL per system (X19). The mechanism rests on argument alone.

Reporting. Several labels and values conflict between tables and figures: "LP40-HME-B" printed twice in Table 2 (X1), the NH10 coating as 8 or 9.8 nm (X3), coating labels reversed in Figure S1b (X4), uncoated onsets of 1 or 2 weeks (X5), and cross-references to the wrong figures (X9). None changes the main result, but individual numbers need care when reused. The study was co-designed and funded by Applied Materials, which employs four of the authors and whose staff filed a patent application on the coating process (ref 56).

Conclusion

The paper turns a single earlier observation into a broader test. Al2O3 or ZnO coatings of 7–24 nm by TGA, applied at 35 °C without solvent to spray-dried and extruded ASD particles, delayed SEM-detected crystallization at 40 °C in all four drug–polymer systems, with or without hydroxyl anchoring groups. Uncoated powders crystallized in about 1–2 weeks; coated powders at 3–24 weeks, or not within 48 weeks in 7 of 13 arms. The coatings left water uptake up to 75% RH and, for Al2O3, the dry $T_g$ unchanged, and they improved the flow of the cohesive LP40 powders. In the authors' words, the effect "is attributed primarily to decreased surface molecular mobility".

The data firmly show the delay and the absence of a bulk moisture barrier. They do not show the mechanism, independence from thickness under failing conditions, or a gain in usable drug loading, and humidity, through the wet glass transition, still decides how much time the coating buys. For formulation, ALC is a candidate add-on for crystallization-prone ASD powders, provided its effect on dissolution is confirmed.