A nanometre alumina shell on spray-dried ASD particles: stopping surface crystallization at high drug loading

Background and motivation

An amorphous solid dispersion (ASD) disperses a poorly water-soluble drug in a polymer so that it dissolves faster and to a higher concentration than the crystal. The price is the polymer itself. To keep the drug from crystallizing over the shelf life, a stable ASD often needs a large polymer fraction, and how large depends on how fast the drug crystallizes and how well the chosen polymer slows it down (Section 1). A large polymer fraction means a large dosage unit that is hard to swallow, or several units per dose, which the authors call the pill burden. Most stabilization work so far has aimed at crystallization in the bulk, usually with a polymer of high glass transition temperature ($T_g$) that interacts with the drug (Van Duong and Van den Mooter, 2016). C1 In the Discussion the authors add that crystallization tendency rises with drug loading (DL), and they name two other routes to a higher loading: combinations of polymers (Mudie et al., 2020) and amorphous salts of ionizable drugs (Van Duong et al., 2022). C50

The paper starts from a different observation. Crystal growth at the free surface of an amorphous drug has been reported to be faster than growth in the bulk by orders of magnitude (Wu and Yu, 2006; Zhu et al., 2008, 2010), because molecules at the surface have fewer neighbours and so more freedom of movement (Yu, 2016). The authors draw the consequence directly: surface crystallization is a failure mechanism for ASDs. C2 If the surface is the weak point, covering it should help, and earlier work supports this. Coatings of gold or polymers only a few nanometres thick dramatically inhibited crystallization of amorphous indomethacin, nifedipine, clofazimine and griseofulvin; gelatin or chitosan coatings stabilized amorphous films of indomethacin and nifedipine; and even milling amorphous acetaminophen with carnauba wax, or gently mixing amorphous indomethacin with Eudragit E PO, made the drug more stable, with interfacial drug–polymer interactions thought to lower the mobility of surface molecules. C3

The difficulty is practical. Many of these demonstrations used films or other idealized samples, whereas spray-dried ASD particles are small and hard to coat one by one. Three-fluid nozzle spray drying, rotating fluidized-bed coaters, dry powder coating and, most often, electrospraying have been tried, but these methods rely mostly on liquids and give little or no control of thickness, uniformity and conformality at the atomic or molecular scale. C4 The authors also argue that dip coating in an aqueous polymer solution, used for amorphous clofazimine particles (Gui et al., 2019), "may not be" viable for ASD particles, because the drug could dissolve or crystallize during dipping and drying. C51

Atomic layer coating (ALC), also called atomic layer deposition, is a gas-phase technique taken from semiconductors, solar cells, catalysts and medical devices. In pharmaceutics it had been used on crystalline drug particles and on one amorphous excipient (Section 1). Acetaminophen particles were coated without polymorphic change and released the drug more slowly (Kääriäinen et al., 2017). Indomethacin particles of about 6 µm with 30–35 nm aluminium oxide shells kept releasing drug in rats for more than 12 weeks after subcutaneous injection, while the uncoated drug was eliminated from plasma with a half-life of 15 h (Hellrup et al., 2019). Oxide films made budesonide powder highly hydrophilic and organic films made it superhydrophobic, and oxide coatings sustained its release and improved its dispersibility (La Zara et al., 2021; Zhang et al., 2017). Five TiO2 layers improved the flow of several common excipients (Hirschberg et al., 2019). The closest precedent is amorphous spray-dried lactose, which crystallizes within a few hours at high relative humidity: aluminium oxide nanoshells blocked its water uptake, and only about 15% crystallinity was found after six months at 40 °C/75% RH (Hellrup et al., 2017). C6 From these effects on wettability, dissolution, dispersibility, flow and stability, the authors infer that ALC "might be particularly valuable" for ASD particles, and they note that no ALC of an ASD had been reported. C7

The paper is a proof of concept for that idea. The authors spray-dried ezetimibe with hydroxypropyl methylcellulose acetate succinate (HPMCAS, MF grade) at 50% and 70% w/w drug, coated each powder with a thin and a thick aluminium oxide shell by ALC, characterized the six powders, and stored them in open dishes at 40 °C/75% RH for two years. A1·A2·A3·P2·P4·P16 Ezetimibe makes the test stringent: it is documented as a fast crystallizer both in the solid state (Correa-Soto et al., 2017) and during dissolution (Elkhabaz et al., 2018). M1·C25 The authors present the work as the first application of ALC to ASDs and as a route to higher drug-loading ASDs with a smaller pill burden. C8·C60

Background concepts

Surface crystallization and surface mobility. In an amorphous solid below its $T_g$, molecules in the bulk are caged by their neighbours and crystal growth is slow. Molecules at a free surface are less constrained. The Discussion cites reports that molecular mobility at the free surface exceeds bulk mobility by an order of magnitude or more, which accelerates crystallization at the surface (Zhu et al., 2008, 2010; Wu and Yu, 2006; Yu, 2016). C54 For a powder, with its very large surface, this means that crystals can start on the outside of the particles long before the bulk would crystallize. A coating of a low-mobility substance is expected to hinder the surface molecules and so the crystal growth (Wu et al., 2007; Li et al., 2019; Teerakapibal et al., 2018). C54 This paper tests the outcome of that idea, crystallization or no crystallization, but does not measure surface mobility itself. X7

Water and crystallization. The second idea the authors use is that water accelerates the crystallization of amorphous formulations (Rumondor et al., 2009; Mehta et al., 2016; Lehmkemper et al., 2017). A coating that lowers the rate and extent of water sorption could therefore slow crystallization by itself, as reported for alumina-coated lactose (Hellrup et al., 2017). C53 Water sorption is measured here by dynamic vapour sorption (DVS), which records the weight gain of a dried sample as the relative humidity (RH) is raised in steps at constant temperature. P9

Atomic layer coating. ALC builds a film one molecular layer at a time from two gases that never meet in the gas phase (Fig. 2). The substrate surface must carry reactive groups, such as hydroxyl or amine groups. A pulse of a metal precursor, here trimethylaluminum (TMA), reacts with these groups until all anchoring sites are used and the surface is saturated; the reaction then stops by itself. An inert-gas purge removes excess precursor and by-products. A pulse of an oxygen co-reactant, here water, converts the attached methyl groups back into hydroxyl groups, and a second purge clears the reactor. The cycle is repeated until the film reaches the desired thickness. C5·P4 Because each half-reaction is self-limiting, the literature describes the film as conformal, with thickness control at the ångström level, and as a complete monolayer that is pinhole-free after each cycle (Puurunen, 2005; Van Ommen and Goulas, 2019; Van Bui et al., 2017). C5 The caption of Fig. 2 gives the chemistry that follows the first reaction of TMA with the ASD surface:

In plain terms: in the first half cycle a surface hydroxyl group binds one TMA molecule, which gives off one methyl group as methane and leaves a dimethylaluminium group attached to the surface. E5

In plain terms: in the second half cycle water replaces each surface methyl group with a hydroxyl group and releases methane, which restores a hydroxylated surface for the next TMA pulse. E6

In plain terms: summed over the cycles, two TMA molecules and three water molecules leave one formula unit of aluminium oxide on the surface, and the only by-product is methane gas (Puurunen, 2005; George, 2010). E7 Two points matter for an ASD. The process runs at 35 °C without any solvent, so the drug is never dissolved, and the ASD surface itself must supply the first anchoring groups. P4 Note by the AI writer (not discussed in the paper): the structures in Fig. 1 show hydroxyl groups on ezetimibe and hydroxyl and carboxyl-bearing substituents on HPMCAS, which are plausible anchoring sites; the paper says only that the starting surface "ideally contains" a variety of functional groups. M1·M2·P4

Powder flow from tapped density. A cohesive powder settles a lot when it is tapped, because its particles first pack loosely around air. The authors measured the unsettled volume $V_o$ of a known mass $M$ of powder in a 100 mL cylinder and the tapped volume $V_F$ after cycles of 1,250 taps, repeated until the volume changed by less than 2 mL between cycles; bulk density is $M/V_o$ and tapped density $M/V_F$ (Section 2.9). P8·E1·E2 The two flow indices are:

Here CI is the compressibility index in percent and HR the Hausner ratio. E3·E4 In plain terms: the more a powder consolidates on tapping, the more loosely and unevenly its particles were packed and the worse it flows, so both indices rise together. Because both densities use the same mass, CI also equals 100 × (1 − bulk density / tapped density) and HR equals tapped density / bulk density, which lets a reader check Table 2: for 50% Thin coated, 100 × (1 − 0.34/0.51) gives 33.3% and 0.51/0.34 gives 1.50, as printed. E3·E4·R7.b The authors use the USP scale (Table S1), on which a CI of 21–25% and an HR of 1.26–1.34 count as "passable". R8

Supersaturation, amorphous solubility and non-sink release. An ASD dissolving in a limited volume can raise the dissolved drug concentration above the crystalline solubility. This supersaturation raises the rate of membrane transport and underlies the bioavailability gain of ASDs. C49 Its upper limit is the amorphous solubility, at which the solution separates into a drug-rich and a water-rich liquid phase. The authors measured it by adding a 10 mg/mL solution of ezetimibe in DMSO to pH 6.5 phosphate buffer at 37 °C until scattering at 350 nm, a wavelength where the drug does not absorb, showed the onset of liquid–liquid phase separation (Section 2.2). P1 The amorphous solubility of ezetimibe is 20.2 ± 0.4 µg/mL, against a crystalline solubility of 1.0 ± 0.1 µg/mL quoted from Elkhabaz et al. (2018). M1 The release test placed 10 mg of drug in 50 mL of buffer, so complete dissolution would give 200 µg/mL. P15 By the AI writer's arithmetic, that is about ten times the amorphous solubility: a strongly non-sink test in which the solution can crystallize as soon as it becomes supersaturated. Crystals already present in an ASD make this worse, because they deplete the supersaturation by secondary nucleation and crystal growth unless the polymer inhibits crystallization in solution, as shown for tacrolimus, bicalutamide and celecoxib ASDs. C49

Reading crystallization during storage from DSC. When an ASD that contains amorphous drug is heated, the drug may crystallize during the scan (an exotherm, called thermally induced recrystallization) and then melt (an endotherm). If some drug had already crystallized during storage, those crystals melt as well, so the melting enthalpy counts both kinds of crystal. The authors separate the two (Section 3.3, Fig. 15):

Here $\Delta H_m$ is the enthalpy of melting, $\Delta H_r$ the enthalpy of thermally induced recrystallization during the scan, and $\Delta H_c$ the enthalpy of the drug fraction that crystallized during storage, all in J per g of drug. E8 In plain terms: whatever melts but did not crystallize in the DSC pan must already have been crystalline before the scan began. A fully amorphous sample that crystallizes completely on heating has equal $\Delta H_r$ and $\Delta H_m$, and its $\Delta H_c$ is zero. E8·C21 The paper does not print the melting enthalpy of crystalline ezetimibe, so $\Delta H_c$ is not converted into percent crystallinity. R21

Three ways to detect crystals. Scanning electron microscopy (SEM) images the particle surface and can show a single crystal. Powder X-ray diffraction (PXRD) and differential scanning calorimetry (DSC) respond to the crystalline fraction of the whole sample and need a minimum amount of it, with reported detection limits of about 0.7% for PXRD and 0.3% for DSC (Correa-Soto et al., 2017; Pan et al., 2006). C55·R26 For quantification the authors followed the intensity of the strongest ezetimibe peak, near 19.6° 2θ, against calibration curves made separately for each drug loading, because at the same percent crystallinity the larger polymer fraction of the 50% ASD weakens the drug signal (Figs. S6 and S7). P13·C38

Methods in detail

The study has three parts: putting a shell on the particles, finding out what the shell changes in the fresh powder, and a two-year accelerated stability study read out by three solid-state methods and a release test. One drug and one polymer at two drug loadings, each uncoated, thin-coated and thick-coated, give six powders. A1·A2·A3 The flow below shows the order of the work.

Contribution 1: Putting a nanometre shell on spray-dried ASD particles

① Problem. Spray-dried ASD particles are small: the size distributions in Fig. 4 have three populations, about 0.1–1 µm, 1–40 µm and above 40 µm. R4 A coating for them must cover each particle evenly at the nanometre scale, must not dissolve the drug or start its crystallization, and must leave the amorphous state intact. The liquid-based methods tried so far do not meet these needs. C4·C51

② How. Ezetimibe and HPMCAS MF were dissolved at about 10% solids in methanol:dichloromethane 50:50 (v/v) and spray-dried on a Buchi B-290 with an Inert Loop (nitrogen atomization 700 L/h, aspirator 35 m³/h, feed 4 mL/min, inlet 70–75 °C); the powder was kept under vacuum for 48 h. P2 The specific surface area of each powder was measured by five-point BET nitrogen sorption, because it sets how much precursor is needed. P3 The paper prints 1.27 for the 50% powder and 0.88 m²/g for the 70% powder; the first value carries the inverted unit "g/m²", most likely a misprint. R1·X1 About 100 g of powder was then loaded into the center tube of a rotary reactor (Fig. S1), whose rotation agitates the powder so that all surfaces are exposed. At 35 °C, TMA and deionized water were pulsed alternately into the reactor at 1–10 Torr with nitrogen purges in between; the amounts were calculated from the surface area, and a residual gas analyzer (Inficon Transpector CPM) monitored whether each pulse saturated the surface. P4 Two coating levels were made, called thin and thick. The number of cycles, the pulse and purge times and the rotation speed behind these labels are not printed, so the coatings are defined only by their measured thickness and residue. X2

Thickness was measured on cross sections cut by focused ion beam (FIB). Each sample was capped with an epoxy carbon layer and sputtered iridium, a lamella about 100 nm thick was lifted out, and it was imaged by transmission electron microscopy (TEM) at 200 kV in bright-field, high-resolution and high-angle annular dark-field modes; energy-dispersive X-ray (EDS) mapping was used to confirm the film composition. Thickness was read at six locations within one cutting area per sample (Table 1). P5·R2 To find the drug content of a coated powder, the authors first dispersed it in 1 M HCl, which dissolves the aluminium oxide, then added methanol and assayed ezetimibe by HPLC. P14·C9 Thermogravimetric analysis (TGA) to 850 °C gave the coating residue, on the reasoning that ezetimibe and HPMCAS degrade and volatilize between 170 and 700 °C, so that whatever remains is aluminium oxide. P10·C11

③ Why it works. The two half-reactions are self-limiting, so the dose only has to exceed what the surface can take up, and that capacity scales with the measured surface area; rotation keeps fresh surface in contact with each pulse, and the gas analyzer confirms saturation. P4·C5 The process is solvent-free and runs at 35 °C, which removes the risk of dissolving the drug during coating. P4·C51 Note by the AI writer (not discussed in the paper): 35 °C is also well below the glass transition events of the fresh powders, which lie at about 62–72 °C by DSC, so the particles should stay glassy during coating. R10

④ Difference from earlier work. Earlier ALC work in pharmaceutics coated crystalline drug particles and amorphous lactose. C6 Earlier coatings of amorphous drugs were made on films or by liquid or mechanical processes. C3·C4 This paper is, by the authors' account, the first to coat ASD particles by ALC. C8

Contribution 2: Finding out what the shell changes in the fresh powder

① Problem. A coating can change particle size, amorphous state, agglomeration, flow, water uptake, the drug content per gram of powder and the release behaviour. Each matters for whether a coated ASD can be processed and dosed. And to compare release fairly, the dose of drug must be the same, so the coating mass must be known.

② How. The fresh powders were characterized by laser diffraction (Malvern Mastersizer 3000 with Aero S dry disperser, feed rate 60%, 3 bar), PXRD (Rigaku SmartLab, Cu Kα at 40 kV and 44 mA, 4–40° 2θ at 4°/min, step 0.02°), SEM (Nova NanoSEM 200, platinum sputter for 60 s, 5 kV, 5 nm spot, 10 mm working distance), bulk and tapped density (n = 3), DVS at 37 °C (dried at 0% RH, then 5% to 95% RH in 5% steps of at most 3 h each, equilibrium below 0.001%/min over 5 min; n = 3) and DSC (TA Q2000, hermetic aluminium pans, 5 °C/min, nitrogen 50 mL/min). P6·P7·P8·P9·P11·P12 The release test (Section 2.15) used a Hanson Vision G2 apparatus with Pion Rainbow in situ fiber-optic UV probes of 5 mm path length. Powder below 250 µm containing 10 mg of ezetimibe, calculated from the measured drug content, was added to 50 mL of pH 6.5 phosphate buffer (ionic strength about 0.14 M) at 37.0 ± 0.5 °C, stirred at 200 rpm; spectra were taken every minute for 180 min, and the concentration was calculated from the second derivative of the spectrum over 295–305 nm, which removes the contribution of undissolved particles; n = 3. P15

③ Why it works. Correcting the powder mass for the measured drug content puts the same amount of drug into every vessel, so differences in release come from the powders, not from the dose. P15 The second-derivative method lets the UV probe read dissolved drug in a suspension that still contains undissolved, scattering particles. P15

④ Difference from earlier work. Changes in flow, wetting and release after ALC had been reported for crystalline drugs and excipients. C6 Here they are measured for an ASD, where the release test also has to compete with crystallization of a supersaturated solution. C25

Contribution 3: Two years of accelerated storage, read out at the surface and in the bulk

① Problem. The claim to be tested is that the coating slows crystallization under high-stress storage. Because crystals are expected to start at the surface, the readout must be able to see the first crystals on a particle surface, and must also follow how much of the drug crystallizes over time. C8·C2

② How. The six powders were stored in open dishes at 40 °C/75% RH. At each time point a sample was dried in a vacuum oven for 24 h to remove residual moisture, then analysed. P16 SEM images are shown for the initial powders and after 12 and 19 days, 1, 6, 7 and 24 months (images at 3, 9, 12 and 16 months are in Figs. S4 and S5). PXRD was recorded up to 24 months, and crystallinity was quantified from the 19.6° 2θ peak from 12 to 270 days. DSC was recorded at 0, 1, 3, 6, 9, 12 and 24 months, with monthly enthalpies of the uncoated powders up to 12 months. Release was tested after 1, 3, 6 and 12 months (Fig. 16), and Fig. 17 adds a 9-month point for Cmax and Tmax. Drug content was assayed again after 6 months (Table S2). P16·P13

③ Why it works. SEM catches surface crystals early, while PXRD and DSC measure the crystalline fraction of the whole sample; agreement between them makes the onset times and the ranking of the powders robust. C55 The DSC enthalpy balance, $\Delta H_c=\Delta H_m-\Delta H_r$, separates crystals formed during storage from those formed during the scan. E8

④ Difference from earlier work. The coating studies the authors cite tested idealized samples such as films, and the lactose study with ALC shells reported six months at 40 °C/75% RH. C50·C6 Here particles of a fast-crystallizing drug at 50–70% loading are followed for two years, with crystallinity quantified for nine months. R19·X14

Experimental analysis

Experimental setup. All results come from one drug–polymer pair, ezetimibe in HPMCAS MF, at nominal drug loadings of 50 and 70% w/w, each uncoated, thin-coated and thick-coated; the six powders are named 50% Uncoated, 50% Thin coated, 50% Thick coated, 70% Uncoated, 70% Thin coated and 70% Thick coated. A1·A2·P2 Stability was tested under one condition, open dishes at 40 °C/75% RH. P16 Most measurements are triplicates (n = 3), while shell thickness comes from six readings within one FIB cut per sample. R7·R9·R19·R23·R2 Table A collects the properties of the six fresh powders.

Table A. The six fresh powders: shell thickness by TEM (Table 1, range of six readings in one cutting area), drug content by HPLC and TGA residue (Table 3, n = 3), density and flow indices (Table 2, n = 3; density SDs 0.01–0.02 g/mL omitted). R2·R12·R7

Main result 1: the shell is 12–40 nm thick and leaves the particles as they were. The thin shells measured 17.8–18.0 nm at 50% loading and 11.7–15.0 nm at 70% loading, the thick shells 36.2–38.8 and 31.8–39.7 nm. R2 The labels therefore do not mean the same thing at the two loadings: the 50% thin shell is thicker than the 70% thin shell. X2 The TEM cross section in Fig. 3 shows a continuous AlOx band between the particle and the carbon and iridium capping layers, with thickness labels of 11.7 and 12.8 nm in the high-magnification frame, and the text describes the coverage as "homogeneous and uniform". R3·C12 The figure does not say which sample the particle comes from (its labels fall in the 70% Thin coated range), each thickness range comes from one particle, and the EDS maps are not shown. X3 Coating hardly changed the particles otherwise. The three size populations of Fig. 4 stayed in place; the medium mode near 6–7 µm rose a little with coating (from about 3.25 to about 3.62% volume density at 50%, from about 3.24 to about 3.79% at 70%), which the authors read as minimal impact because the shell is nanoscale. R4·C13 Every fresh powder gave only an amorphous halo near 20° 2θ by PXRD and showed no crystals by SEM (Figs. 5 and 6). R5·R6·C14

The coated powders assayed lower in drug: 50.2 and 43.6% against 54.6% at the 50% loading, and 66.2 and 63.4% against 70.0% at the 70% loading. Their TGA residues were 2.5, 9.2, 1.7 and 4.5%. R12 The authors attribute the lower drug content to the mass added by the coating ("can be attributed") and find the residues in "reasonable agreement" with the "% Reduction" column of Table 3. C10·C23 That column, however, lists percentage-point drops (54.6 − 50.2 = 4.4, printed as 4.5; 11.0; 3.8; 6.6). Relative to the uncoated drug content the drops are 8.1, 20.1, 5.4 and 9.4%, about twice the TGA residues, so the true coating mass fraction lies somewhere between the residue and the relative loss of drug content. X5 Note also that the uncoated 50% powder assayed at 54.6% drug, above its nominal loading. R12.a·X5

Main result 2: coated powders pack denser and flow better. Coating raised bulk density from 0.24 to 0.34 and 0.44 g/mL at the 50% loading and from 0.26 to 0.42 and 0.48 g/mL at the 70% loading; tapped density rose from 0.41 to 0.51 and 0.62 g/mL and from 0.41 to 0.62 and 0.63 g/mL. R7 The compressibility index fell from 42.6 to 33.3 and 29.5% and from 36.4 to 32.1 and 25.0%, and the Hausner ratio from 1.7 to 1.5 and 1.4 and from 1.6 to 1.5 and 1.3. In each case the thick coat did more than the thin. R7·C15 With a compressibility index of 25.0% and a Hausner ratio of 1.3, only 70% Thick coated reached the USP "passable" class, whereas the uncoated powders rated "very, very poor" to "very poor". R7.f·R8·C16 SEM of the fresh powders agrees (Fig. 6): the 70% Uncoated image is dominated by one large cluster of fused spheres, while the coated samples show many separate particles, and the authors read the density data as support for less agglomeration and better flow. R6·C15 They note the same effect of ALC on the flow of excipients (Hirschberg et al., 2019). C6

Table B. Water sorption (% weight gain) of the fresh powders at 37 °C, read from Fig. 7 (approximate; n = 3). R9

Main result 3: only the thick shells cut water uptake. The isotherms in Fig. 7 rise smoothly from 5 to 95% RH, without a step. R9 The thick shells lowered water uptake at every RH: at 95% RH, 50% Thick coated sorbed about 3.17% water against about 5.65% for 50% Uncoated, and 70% Thick coated about 4.35% against about 4.88%. R9 The thin shells did not: 50% Thin coated sorbed about as much as 50% Uncoated and slightly more at 90–95% RH (about 5.83 against 5.65% at 95% RH), and 70% Thin coated sorbed more than 70% Uncoated above about 75% RH (about 5.21 against 4.88% at 95% RH). R9 The authors describe this accurately in the results, the thick coat giving "notably less" sorption and the thin coat being "less able" to reduce it, and they note the agreement with alumina-coated lactose. C17 The abstract and conclusions, however, state lower hygroscopicity for coated powders in general. C59·X6 One cross-comparison caught the authors' eye: 50% Uncoated sorbed slightly more water than 70% Uncoated, especially above 70% RH, yet 50% Thick coated sorbed less than 70% Thick coated. They suggest the coating thickness ("possibly") and the larger coating amount of 50% Thick coated, which had the highest TGA residue, 9.2% ("likely"). C18·C24·R12.c Note by the AI writer (not discussed in the paper): the two thick shells overlap in thickness (36.2–38.8 against 31.8–39.7 nm), whereas their residues differ twofold (9.2 against 4.5%), so the coating amount is the more discriminating of the two explanations. R2·R12

Main result 4: coating shifts the DSC events of the fresh powders. In Fig. 8 the event labelled glass transition lies at about 68, 70 and 72 °C for the uncoated, thin- and thick-coated 50% powders, and at about 62, 64 and 65 °C for the 70% powders. R10 The authors write that coating increased the "apparent" $T_g$ by several degrees, thick more than thin. C19 The 50% powders showed no crystallization on heating. All three 70% powders crystallized and then melted near 157–159 °C; the uncoated one gave a large exotherm with its onset near 113 °C and its peak near 124 °C, the coated ones only small, broad exotherms between about 125 and 150 °C, and coating raised the apparent onset temperatures of both events. R10·C20 Because the crystallization and melting enthalpies were similar within each 70% sample, the authors conclude that this crystallization occurred entirely during heating ("suggesting"), and that the smaller events of the coated samples suggest better physical stability. R11·C21·C22 Note by the AI writer (not discussed in the paper): a shell a few tens of nanometres thick is a small fraction of a micrometre-sized particle, so a shift of the bulk glass transition is not expected from the shell alone; the authors' word "apparent" leaves the cause open. A smaller heat-induced crystallization of the coated 70% powders would, however, fit crystallization that starts at the particle surface even in the DSC pan. A small step near about 135 °C in all three 50% traces is not discussed in the text. R10

Table C. Release from the fresh powders (µg/mL; 10 mg drug in 50 mL pH 6.5 buffer, 37 °C), read from Fig. 9 and Fig. 17B (approximate; n = 3). Amorphous solubility 20.2 µg/mL; crystalline solubility 1.0 µg/mL; complete dissolution would give 200 µg/mL. R13·R24·M1

Main result 5: in fresh powders the coating lowers release at 50% loading and raises it at 70%. No sample dissolved more than 7% of its drug before the concentration started to fall, which the authors attribute to crystallization of ezetimibe. R13·C25 Every maximum lay above the crystalline solubility and below the amorphous solubility, so all powders produced supersaturated solutions without reaching the amorphous solubility. C26·M1 At the 50% loading the uncoated powder rose fastest, to about 12.4 µg/mL at 5 min, held about 13–14 µg/mL with a second maximum of about 13.7 µg/mL near 85 min, and then fell steeply between about 90 and 120 min. 50% Thin coated peaked at about 12.5 µg/mL and 50% Thick coated at about 10.1 µg/mL, the thick coat rising slowly (about 5 µg/mL at 5 min). R13.a·R13.b·R13.c The authors say coating "seemed to delay" release, the coated samples taking more time to reach their maximum. C27 They attribute this to the time needed for the coating to rupture or erode, a prerequisite for release, combined with the fast crystallization of ezetimibe: "it is likely" that more drug crystallized in the coated matrices before the coating was removed. C28 The plotted values show the delay in the first minutes rather than in the time of the maximum: the initial Tmax of the coated 50% powders (about 16.5 and 24 min) is shorter than that of the uncoated powder (about 29 min), whose plateau makes its own Tmax ill-defined. R24·X8

At the 70% loading the order reversed. 70% Uncoated rose slowly to about 7.6 µg/mL at about 25 min, while the thin- and thick-coated powders reached about 11.9 and 10.9 µg/mL within 10–15 min. R13.d·R13.e·R13.f·C29 The authors give wetting as the likely reason ("can be likely attributed"): the uncoated 50% powder wetted and dispersed readily, the uncoated 70% powder was poorly wetted and floated on the medium (Fig. S2), and coating improved the wetting of the 70% powders, as reported for ceramic-coated budesonide. C30·R14 The wetting evidence consists of images in the SI, described in the text for the 70% powders only, without a contact angle or wetting time. X17 At both loadings the thick coat released less than the thin one. C31 All six profiles ended at about 3–5 µg/mL after 180 min. R13 One small inconsistency: the dotted equilibrium-solubility line in Fig. 9 is drawn at about 0.8 µg/mL, whereas the text gives 1.0 ± 0.1 µg/mL. X16

Table D. Crystallization during open storage at 40 °C/75% RH: onset by SEM (Figs. 10, 11) and PXRD (Fig. 12, text), crystallinity read from Fig. 13 (approximate; n = 3), DSC traces (Fig. 14, text). R16·R17·R18·R19·R20

Main result 6: the uncoated powders crystallize within weeks; three coated powders never do. Under storage the uncoated powders caked within 12 days, while the coated powders stayed largely free of agglomerates (Fig. S3); SEM showed large clumps of uncoated particles and much less agglomeration for coated ones. R15·C32 On the uncoated particles crystals followed quickly. Plate-like crystals covered 70% Uncoated particles at 12 days, and needle-like crystals appeared on 50% Uncoated agglomerates at 19 days; at 24 months both samples were masses of crystals (Figs. 10 and 11). R16·R17·C33 The coated 50% particles and the 70% Thick coated particles stayed smooth for 24 months. 70% Thin coated stayed smooth at 6 months, showed one circled crystal cluster on a particle at 7 months and many elongated crystals at 24 months. R16·R17·C34 The authors summarize that "Overall" the drug crystallized more slowly on the surface of particles with the lower drug loading and with the thicker shell. C35

PXRD agreed with SEM. Bragg peaks, in particular the one near 19.6° 2θ, appeared after 12 days for 70% Uncoated and after 7 months for 70% Thin coated. For 50% Uncoated they rose above the noise only at 3 months, although SEM had shown crystals at 19 days, which the authors take as evidence that SEM is the more sensitive method. R18·C36 The crystals matched form A, the starting polymorph, by comparison with the crystalline drug in Fig. 12D. R18·C37 The other three coated powders kept an amorphous halo for two years. R18 The calibrated crystallinity in Fig. 13 puts numbers on the difference: 70% Uncoated rose from about 2% at 12 days to about 17% at 61 days and about 50% at 270 days; 50% Uncoated stayed near zero up to about 40 days, reached about 1.3% at 90 days, about 15.7% at 180 days and about 26% at 270 days; 70% Thin coated read about 0.3% at 240 days and about 1.5% at 270 days; and the other coated powders stayed at about zero. R19·C39

Table E. DSC enthalpies of the uncoated powders during storage at 40 °C/75% RH (J per g drug), read from Fig. 15 (approximate; n = 3). The coated powders were not analysed this way. R21

DSC told the same story by a different route (Fig. 14). The coated 50% powders showed no recrystallization exotherm or melting endotherm even after two years, and 70% Thick coated showed no notable change; 70% Thin coated had larger melting endotherms at 9, 12 and 24 months, consistent with SEM and PXRD. R20·C40 In the text the unchanged traces of 70% Thick coated are attributed to "Fig. 12F", which is the PXRD panel; the DSC traces are in Fig. 14F. X12 The two uncoated powders crystallized in different ways (Table E). In 50% Uncoated, a melting peak near 152 °C appeared after one month and kept growing while the small recrystallization exotherm stayed about constant, so more and more drug had crystallized during storage. In 70% Uncoated the melting endotherm stayed at about 46 J/g drug while the recrystallization exotherm shrank, so less amorphous drug remained to crystallize on heating. R20·R21·C41 The enthalpy of storage crystallization, $\Delta H_c$, reached about 13.2 J/g drug for 50% Uncoated and about 36.0 J/g drug for 70% Uncoated after 12 months, which the authors read as faster crystallization at the higher drug content. R21·C42 The authors' summary of the stability study is that ALC prolonged the amorphous state of ezetimibe "from about two weeks to at least two years". C52

Two caveats on timing belong here. First, the quantitative series end before the two-year claim: crystallinity at 270 days, DSC enthalpies at 12 months and only for the uncoated powders, release at 12 months; at 16 and 24 months the evidence is SEM images and PXRD and DSC traces. X14 Second, the date stamps on the SEM images (initial 2/25/2020; 12 days 3/11 and 3/12/2020; 19 days 3/24/2020; 1 month 4/9/2020; 6 months 9/15/2020; 7 months 10/8/2020; 24 months 3/24/2022) do not map onto a single storage start date, and the delay between sampling, 24 h of vacuum drying and imaging is not stated, so the SEM onset times are storage labels with an uncertainty of several days. R16·R17·X13

Table F. Peak concentration (Cmax, µg/mL) in the release test after storage at 40 °C/75% RH, read from Fig. 17A (approximate; n = 3). The initial values differ by up to 0.1 µg/mL from Table C because they are read from a different figure. R23·R13

Main result 7: after storage the coated powders keep the higher release. For the uncoated powders, crystallization during storage cut release (Fig. 16A, D). Cmax of 50% Uncoated fell from about 14.0 µg/mL to about 8.3 after one month and about 5.6 after 12 months; Cmax of 70% Uncoated fell from about 7.5 to about 6.1 after one month and about 2.6 after 12 months, when its profile was flat at about 2.7–2.8 µg/mL. R23·R22 The authors call the decrease "dramatic", "even after one month of storage", and the change for the coated powders "much less". C43 From one month on, every coated powder gave a higher Cmax than its uncoated counterpart at every time point, which the authors summarize as "always considerably higher". R23·C44 Read against the plotted values, the "dramatic" one-month drop applies to 50% Uncoated; 70% Uncoated declined gradually. X10 The coated powders split by loading. The 50% coated powders kept Cmax at or above its initial value (about 12.1–13.1 µg/mL for thin, about 10.2–12.2 µg/mL for thick) but reached it later, with Tmax up to about 46 and about 85 min; the 70% coated powders lost Cmax gradually, to about 8.3 and 7.5 µg/mL at 12 months. R23·R24 The authors attribute the "slight decrease" of release from coated powders to agglomeration under humid storage (Fig. S8) and the longer Tmax of all samples after one month to the same agglomeration ("likely"). C45·C46 The decrease they describe is therefore seen only at the 70% loading. X9 Fig. 17 also includes 9-month points for which Fig. 16 shows no profile. X15 Note by the AI writer (not discussed in the paper): after storage the 50% Thick coated powder held a higher concentration at 180 min (about 10.3 µg/mL at 1 month, about 8.1 at 12 months) than when fresh (about 5.4), so for this powder the slower release after storage also meant less loss of supersaturation by the end of the test. R22.c

Despite the physical changes, no powder degraded chemically: the drug content after 6 months (Table S2) matched the initial values. R25·C48 The authors close the results with a caution that the release profile is a "complex interplay" of the crystallization rate, the erosion rate of the coating and the wetting and dispersion of the particles. C47

Role of the controls and what the comparisons can show. Each coated powder has an uncoated counterpart at the same loading, characterized and stored side by side, so the coated–uncoated contrast at each loading is direct. A2·A3 The two loadings show that the uncoated 70% powder crystallizes earlier and further than the uncoated 50% powder. R19·R21 The effect of shell thickness can be read cleanly only within a loading. Between loadings, thickness and loading change together: the 50% thin shell (about 18 nm) is thicker than the 70% thin shell (11.7–15.0 nm), so the authors' statement that lower loading and thicker shells both slow crystallization rests partly on confounded comparisons. C35·X2 The study has a single storage condition, no second drug or polymer, and no measurement of surface mobility or of water content during storage, so it shows that the shell works under these conditions but not why. P16·A1·X7

Discussion

How the authors explain the stability. The Discussion (Section 4) calls the stability gain "striking" and offers "two possible mechanisms". C52·C53 First, the coating "may serve as a barrier" that lowers the rate and extent of water sorption, water being known to accelerate the crystallization of amorphous formulations. C53 Second, "and perhaps more importantly", the coating slows the mobility of drug molecules at the particle surface; because surface mobility exceeds bulk mobility by an order of magnitude or more, a layer of a low-mobility substance hinders the fastest route to crystallization, and the authors call reduced surface mobility "the likely explanation" for the result. C54 The Conclusions combine the two: the stability "can be attributed to the moisture barrier effect and/or decreased surface molecular mobility". C53·C54 Neither mechanism was measured. X7 The chain the authors propose, with the two outcomes their data show, is:

Causal chain as proposed by the authors (steps 1–2 and the mechanism in the branches are their interpretation; the outcomes are measured). C2·C53·C54·R16·R17·R19·R23

Which mechanism do the data favour? The paper does not separate the two, but its own data bear on the question. The authors give the moisture barrier as the first possibility and the reduced surface mobility as "perhaps more" important. C53·C54 Note by the AI writer (not discussed in the paper as such): two powders argue against a moisture barrier as the main cause. 50% Thin coated sorbed about as much water as 50% Uncoated at 37 °C (about 3.3 against 3.56% at 75% RH, about 5.83 against 5.65% at 95% RH), yet stayed free of crystals for 24 months while 50% Uncoated crystallized within 19 days. R9·R16·R19·X7 70% Thin coated sorbed slightly more water than 70% Uncoated above 75% RH, yet its first crystals appeared after 7 months rather than 12 days. R9·R17 If lower water uptake were the main cause, these thin shells should not have helped. The surface-mobility explanation fits these results, as does the smaller heat-induced crystallization of the coated 70% powders in the DSC pan. R10·C22 It remains an inference: surface diffusion or surface crystal growth was not measured with and without the shell, and the sorption isotherms were recorded at 37 °C on fresh powders, not during storage at 40 °C/75% RH. X7

Why the release results are mixed. The authors are frank that the effect of ALC on release is "complicated". In their reading three factors interact: wetting and dispersion, which ALC improves; coating thickness, where "a too thick coating can delay release"; and the crystallization tendency of the drug, since a delay may let it crystallize in the hydrated matrix. C57 The fresh-powder data show each factor once. At 50% loading, where the uncoated powder already wetted well, the shell mostly cost speed and peak concentration. At 70% loading, where the uncoated powder floated, the shell's better wetting dominated. R13·R14·C28·C30 After storage the crystallization of the uncoated powders outweighed everything else, so the coated powders "always" released more. C44·R23 The authors conclude that these issues "need to be further explored". C57

Where the approach applies. The evidence covers one fast-crystallizing model drug in one polymer at 50 and 70% w/w, stored open at 40 °C/75% RH, with aluminium oxide shells of about 12–40 nm. A1·P16·R2 Within that window, the stability gain is large and consistent across three detection methods. R16·R18·R19·R20 Whether the same shell thickness is enough for drugs with a different crystallization tendency, for other polymers, at lower humidity or in closed packaging, and whether the coated powder behaves the same after tableting, was not tested. The authors themselves point to downstream processing and to the choice of coating material as the next questions. C56·C61

What it suggests for formulation, in the authors' words. The authors see ALC as a route to higher drug-loading ASDs without loss of physical stability, which would reduce the pill burden; in their phrase it "paves the way". C60 They add that better flow and compressibility could make capsule filling or tableting of spray-dried intermediates easier, a point they say should be investigated. C56 They also note that dip coating in aqueous polymer solutions is probably not an option for ASD particles, whereas ALC is solvent-free and works on powders. C51

What it suggests for formulation, in the AI writer's reading (not stated in the paper). First, shell thickness is a design variable, not a yes/no choice: at 70% loading an 11.7–15.0 nm shell delayed crystallization by months, while a 31.8–39.7 nm shell held for two years. R2·R17·R19 Second, a water-sorption screen alone would underrate a thin coating, because the thin 50% shell protected without lowering water uptake. R9·R19 Third, a coated ASD needs its own assay and dose correction: the 50% thick coat lowered the drug content from 54.6 to 43.6%, and the coating mass is uncertain by about a factor of two. R12·X5 Fourth, for a drug that crystallizes fast in solution, a slower start costs supersaturation, so the shell should be no thicker than stability needs: in fresh powders the thick shell lowered the 50% peak from about 14 to about 10 µg/mL. R13 Finally, the alumina shell dissolves in 1 M HCl, but whether it affects dissolution in gastric conditions or in vivo was not examined. C9

Limitations

Limitations stated by the authors. The authors state few limits explicitly. In the results and Discussion they report that one coated powder, 70% Thin coated, did crystallize, with a few surface crystals after 7 months and more after 24 months. C34·C52 They describe the release behaviour as a complex interplay that "need[s] to be further explored", including what a poorly soluble inorganic coating does to an ASD, and they note that a too thick coating can delay release and let the drug crystallize in the hydrated matrix. C57 They attribute the slight loss of release from coated powders after storage to humidity-driven agglomeration. C45 They point out that PXRD and DSC need a minimum amount of crystalline drug and are less sensitive than SEM. C55·C36 And they leave downstream processing and the choice of the best coating material to future work. C56·C61

The AI writer's own judgement (1): what is solid. The central result is direct and robust within its system. Crystallization of the two uncoated powders and its absence in three coated powders are seen by three independent methods, SEM, PXRD and DSC, with calibrated PXRD crystallinity in triplicate up to 270 days. R16·R17·R18·R19·R20 The ranking is consistent across methods: 70% Uncoated crystallizes first and most, then 50% Uncoated, then 70% Thin coated, while the thick shells and the 50% thin shell do not crystallize. R19·C35 The flow improvement is measured in triplicate and follows shell thickness at both loadings, and after storage every coated powder released more than its uncoated counterpart at every time point. R7·R23 Chemical stability was checked at 6 months. R25 What the data cannot carry is generality: one drug, one polymer, two loadings and one open-dish storage condition. A1·P16

The AI writer's own judgement (2): the abstract and conclusions say more than the results. The abstract states that the coated samples "showed no evidence of physical instability for two years" and the conclusions that crystallization was prevented for two years, without the exception of 70% Thin coated given in the results; the abstract also places the first crystals of the uncoated powders "after only a few days", whereas they appeared at 12 days (70%) and 19 days (50%). C52·X11 Lower hygroscopicity is stated for coated powders in general, although only the thick shells lowered water uptake. C59·X6 Improved wettability rests on images described for the 70% powders only. X17 The pinhole-free, conformal coverage "irrespective of particle size and shape" is a property claimed for the technique; here it rests on the ALC literature and one TEM cross section. C58·X4

The AI writer's own judgement (3): the coating is under-described. The number of ALC cycles, the doses, pulse and purge times and the rotation speed behind "thin" and "thick" are not printed, so another laboratory could not reproduce a given shell from the paper alone. X2 Thickness comes from one cutting area, that is one particle, per sample, and the particle in Fig. 3 is not identified. X3 The coating mass is uncertain by about a factor of two, depending on whether it is read from the TGA residue or from the relative loss of drug content. X5 The surface area of the 50% powder is printed with an inverted unit. X1

The AI writer's own judgement (4): some statements do not match the plotted values. The coated 50% powders did not reach their maximum later than the uncoated one in the fresh release test; they rose more slowly at first. X8 The "slight decrease" of release with storage applies to the 70% coated powders; the 50% coated powders kept their Cmax. X9 The "dramatic" decrease after one month applies to 50% Uncoated, whereas 70% Uncoated declined gradually. X10 Fig. 17 contains 9-month points for which no profile is shown, the dotted solubility line of Fig. 9 sits at about 0.8 rather than 1.0 µg/mL, and the text cites Fig. 12F for DSC traces that are in Fig. 14F. X15·X16·X12 None of these changes the main conclusion, but they matter for anyone using the numbers.

The AI writer's own judgement (5): the two-year claim and the mechanism. Quantitative crystallinity ends at 270 days and the release data at 12 months; beyond that the two-year claim rests on SEM images and PXRD and DSC traces, which is adequate for "no crystals detected" but gives no numbers. X14 The SEM time labels carry an uncertainty of several days because the date stamps do not fit one start date. X13 The mechanism, finally, is an interpretation: the authors' own preference for reduced surface mobility is plausible and fits the thin-shell results better than a moisture barrier does, but neither was measured. C54·C53·X7

Conclusion

The paper shows, for the first time by the authors' account, that atomic layer coating can put an aluminium oxide shell a few tens of nanometres thick on spray-dried ASD particles without changing their size distribution or amorphous state. C8·R2·R4·R5 For ezetimibe–HPMCAS at 50 and 70% drug loading stored open at 40 °C/75% RH, the shell turned powders that crystallized within 12–19 days into powders that showed no detectable crystals for two years, with one exception: the thinnest shell at the highest loading, which held for 7 months. R16·R17·R18·R19·C34 The coating also made the powders denser and better flowing, the thick shells lowered water uptake, and after storage every coated powder kept a higher release than its uncoated counterpart, although in fresh powders the shell slowed release at 50% loading. R7·R9·R13·R23 The authors attribute the stability to lower molecular mobility at the particle surface, "perhaps" together with a moisture barrier; the thin-shell results favour the surface explanation, but neither was measured. C53·C54·X7

For formulation, the result reframes the stability limit of a high-loading ASD as partly a surface problem that can be engineered after spray drying. C2·C60 What a formulator would still need before relying on it is the information the paper does not give: the coating settings that make a given shell, the minimum thickness for a given drug and loading, behaviour under other storage conditions and after tableting, and a direct test of why the shell works. X2·C56·C61·X7