When an amorphous solid dispersion crystallizes before it dissolves: drug loading, milling and surfactants in bicalutamide–copovidone

Background and motivation

An amorphous solid dispersion (ASD) is meant to create and hold supersaturation: the amorphous drug, mixed at the molecular level with a polymer, dissolves to concentrations far above the solubility of its crystal. Both the amorphous solid and the supersaturated solution it produces are metastable, so crystallization can undo the advantage in two places. In solution, nucleation and crystal growth consume the supersaturation, and dissolved polymers are often good at slowing both. In the solid, the amorphous drug can convert to the crystal before it ever reaches the solution; the authors call this matrix crystallization, and it removes the thermodynamic driving force for dissolution at its source (Section 1). C1

The authors note that little is known about how matrix crystallization works. Most evidence comes from images or Raman spectra of crystals growing on amorphous solids: neat amorphous nifedipine crystallized directly in buffer but not when formulated as an ASD (Raina et al.); polymers in the medium did not change when amorphous celecoxib began to crystallize, though they slowed growth (Xie and Taylor); an indomethacin methyl ester–PVPVA ASD crystallized at the compact surface on hydration, "likely" because of its low glass transition temperature (T_g) and high crystallization tendency (Saboo et al.); and in situ Raman showed slow surface crystallization of indomethacin but immediate crystallization of carbamazepine (Savolainen et al.). C2

This paper continues the group's earlier study of BCL–PVPVA ASDs (Moseson et al., AAPS J 2021). There, a 30% drug-loading ASD made by hot melt extrusion began to nucleate at the solid–liquid interface after 20–30 min and then grew crystals of the metastable form 2 rapidly; matrix crystallization dominated how much supersaturation was reached. The authors had hypothesized that incongruent release, in which the polymer leaves faster than the drug, leaves the hydrated amorphous drug highly mobile and without the protection of interactions with the polymer. C3·R21·R23 They call matrix crystallization "the most detrimental failure mode during dissolution" and compare it with loss of the amorphous state on storage. C5 Two further facts frame the study: BCL is known to be mechanically activated by milling and compression (Nemet et al.; Szafraniec-Szczęsny et al.), so processing could matter, and PVPVA ASDs show two release regimes, congruent release below a limit of congruency and much slower, drug-controlled release above it. C4·C6 Surfactants are a known way to speed release but may also promote crystallization. C21

The paper therefore asks which of drug loading (DL), surface area, release rate and processing history put a BCL–PVPVA ASD at risk of matrix crystallization during hydration and dissolution, and which strategies prevent it (Section 1). C4·C15·C38

Background concepts

Matrix crystallization versus solution crystallization

In solution crystallization the drug first dissolves and then nucleates and grows from the supersaturated solution. In matrix crystallization the amorphous drug crystallizes inside or on the surface of the undissolved particle. The two leave the same signature in a dissolution curve (a plateau below the dose), so the study needs a way to tell them apart. The authors use a dose that gives at most 30 µg/mL BCL: in their earlier work this concentration did not nucleate in solution in the presence of PVPVA and showed near-zero solution crystal growth, so any shortfall from complete release is attributed to the solid. P10·R22·C7

Bicalutamide's solid forms

BCL is a Biopharmaceutics Classification System class 2 drug with high glass forming ability (class 3 in the Baird scheme). It has a stable form 1 and a metastable form 2; form 2 is the polymorph that appears on dissolution and matrix crystallization. The apparent solubilities in pH 6.8, 50 mM phosphate buffer, cited from the earlier study, set the reference lines used in every dissolution figure. M1·M10·R23

Table A. BCL solid forms (Section 2.1; values cited from the authors' earlier study). A dissolution curve that levels off near 50% of dose has reached about the form 2 solubility. M1·M10

Congruent and incongruent release, and the limit of congruency

For PVPVA ASDs, two release regimes are well documented (Saboo et al.; Hiew et al.; Que et al.; Indulkar et al.). At low DL, drug and polymer dissolve at the same normalized rate: release is congruent and controlled by the polymer, and drug concentration rises quickly. Above a threshold DL, the limit of congruency (LoC), the polymer dissolves faster than the drug: release is incongruent and controlled by the drug, often orders of magnitude slower, and sometimes it stops. C6 This paper's central idea is that the polymer-poor surface left behind by incongruent release is where matrix crystallization starts. C39

Mechanical activation

Mechanical energy from milling or compression can make a solid transform more readily, for example by creating defects, disorder or nuclei. Cryomilling is known to amorphize crystals by size reduction and defect formation, and milling amorphous materials is known to create nuclei that later cause surface crystallization on heating (Section 4.3). C53 For BCL itself, milling and compression change the polymorph (Nemet et al.). C4

Seeing drug enrichment and nanoscale mixing

Two techniques read the composition of the solid. Energy-dispersive X-ray analysis (EDX) in the scanning electron microscope (SEM) maps elements; because fluorine occurs only in BCL, the fluorine content of a region compared with that of an unexposed control measures how much drug the region has gained or lost. E1·C8

In plain terms: $x_F$ is the fluorine atomic percent measured by EDX; a ratio above 1 means the region holds more drug (and so less polymer) than the original solid, and a ratio of 1 means no change. The paper gives this definition in words (Section 2.2.5.2) and tests significance with a paired one-tailed t-test on four maps. E1·P17

Solid-state NMR (ssNMR) relaxation probes mixing at much smaller scales. The proton spin-lattice relaxation times in the laboratory frame (T1) and in the rotating frame (T1ρ) are averaged by spin diffusion across domains of a certain size. If the drug and polymer signals in an ASD relax with the same time constant, the two are mixed on that scale; the authors use about 150 nm for T1 and about 10 nm for T1ρ. P9·C35 Confocal fluorescence microscopy, finally, uses two dyes: Nile red, which partitions into disordered drug-rich regions (red), and Alexa Fluor 488, which partitions into polymer- and water-rich regions (green). C8·P15·P18

Methods in detail

The design holds the drug, the polymer, the medium and the dose fixed and changes one factor at a time around a 30% DL reference. Seventeen formulations were made (Table 1): binary BCL–PVPVA ASDs at 5, 10, 20, 30 and 50% DL, and 30% DL ternaries with 1.5 or 3 wt% sodium dodecyl sulfate (SDS) or vitamin E TPGS (TPGS), equal to 5 or 10% of the drug loading. Sample names read DL-process/milling: 30-SE/CM5m is a 30% DL ASD made by solvent evaporation and cryomilled for 5 min. P1·A1·A2·A3·A4·A5

Contribution 1: a dose that isolates crystallization of the solid

The problem. A dissolution plateau below the dose could come from crystallization in solution or in the solid. What they did. Every powder was dosed so that complete release gives 30 µg/mL BCL, whatever its DL, in 200 mL of pH 6.8, 50 mM phosphate buffer at 37 °C (the 30-HME/CM30s data come from the earlier study, run in 100 mL in a 150 mL vessel). BCL was followed every few minutes for 4 h with in-line UV probes, and PVPVA, for selected ASDs, by size-exclusion HPLC of filtered 2 mL samples over 2 h; each profile is the mean of three runs. P10·P11 Because 30 µg/mL does not nucleate in solution with PVPVA present, a plateau below 100% is read as matrix crystallization, and a plateau near 50% of dose means the solid has converted to form 2. R22·C7·M1 Release is plotted both as % of dose and as µg/mL. Measuring drug and polymer together shows directly whether release is congruent. C16·C17 Why it works. The approach turns an ordinary non-sink dissolution test into a test of the solid's stability in water, without sampling the solid.

Contribution 2: processing histories and formulations that change surface, energy and release rate

Processing. Solvent-evaporated (SE) powders were ground lightly in a mortar and pestle (MP) and sieved into 100–250 µm and <100 µm fractions (60 and 140 mesh; Table 1 names them 106–250 and <106 µm), or cryomilled (CM) at 10 Hz for 30 s, 2, 5 or 10 min in 1 min grinding and 1 min cooling cycles after a 5 min pre-cool. Hot melt extrudates (HME, 180 °C, 50 rpm, about 2 min residence) must be milled, and were cryomilled for 30 s. P2·P3·P4·P5·X2 This separates three things that are mixed in the original HME powder: preparation route, size reduction and particle size. A1·A2·A3

Formulation. DL was varied from 5 to 50% to move between the congruent and incongruent regimes, on the hypothesis that faster release could let the drug reach solution before nucleation; and SDS or TPGS was added to 30% DL ASDs as a second way to speed release. Medium controls dissolved the binary 30-SE/MP in buffer containing 3 µg/mL SDS or TPGS, to separate wetting from a change in the release mechanism. C15·A4·A5·A6·P10

Crystallization kinetics. 400 mg of a 30% DL ASD was slurried in 12 mL buffer at 37 °C (10 mg/mL drug) and Raman spectra were taken every 2 min for 2 h. Multivariate curve resolution–alternating least squares (MCR-ALS), started from the spectra of amorphous BCL and form 2, resolved the spectra into two components; the rise of component 2 tracks form 2. P12·P13

Contribution 3: looking at the surface and inside the solid

Compacts. 100 mg of powder (150 mg for confocal) was compressed at 1500 psi for 1 min in a Wood's intrinsic dissolution die, and one face was exposed to 100 mL of medium at 37 °C, 100 rpm, for 5, 30 or 60 min. The compacts were dried and either imaged by SEM, with EDX maps giving F/F₀, or, when made with the two dyes, embedded in paraffin, halved with a razor blade and imaged by confocal microscopy as z-stacks giving about 100 µm depth of field. Layer and compact thicknesses were measured ten times each. P14·P15·P16·P17·P18

Solid state. All powders were checked by X-ray powder diffraction (XRPD, 10–20° 2θ) and modulated differential scanning calorimetry (DSC). For three 30% DL ASDs with different histories (30-SE/MP, 30-HME/CM30s and 30-SE/CM5m), 13C cross-polarization magic angle spinning (CP-MAS) spectra looked for drug–polymer interactions, and 1H T1 and T1ρ, measured through the 13C signals, looked at mixing on about 150 nm and 10 nm scales. P6·P7·P8·P9

How it differs from earlier work. The earlier BCL study identified matrix crystallization in one extruded ASD; this one changes processing and formulation one at a time, measures polymer release alongside drug release, and adds imaging of the exposed surface and nanoscale mixing, so that the risk factors can be separated. C3·C38

Experimental analysis

All dissolution values below are read from the figures and are approximate; they are given as % of the 30 µg/mL dose, so 50% equals about 15 µg/mL, the form 2 solubility. All formulations were amorphous by XRPD (data not shown) and showed a single T_g by DSC (SI), so every difference below arises in water, not in the starting solid. R18

Processing route and particle size at 30% DL

Figure 1 compares three powders of the same composition. The solvent-evaporated powder in the 100–250 µm fraction rose to about 40% of dose at 30 min and about 86% (about 25.7 µg/mL; text ~26 µg/mL) at 4 h. The <100 µm fraction dissolved faster at first, about 46% at 30 min, but levelled off lower, at about 79% (text ~24 µg/mL), with a standard deviation of about ±11 points. The extruded and milled powder (30-HME/CM30s, <250 µm) levelled off at about 60%, near 18 µg/mL, just above the form 2 solubility. R1

Table B. BCL release of the 30% DL powders, % of dose (read from Figures 1 and 4, approximate; mean of n = 3). R1·R4

In the Raman slurries (Figure 2), form 2 appeared after about 20–25 min in all three powders (first non-zero points at about 21–29 min) and conversion was complete at about 57 min for the extruded powder, about 61 min for the small SE fraction and about 77 min for the large SE fraction. R2 The authors read the near-complete release of the large SE fraction as "reduced susceptibility to crystallization"; they say the larger surface area of the smaller particles "can contribute" to lower concentrations, and, because even the small SE fraction ended above the extruded powder, "additional factors must contribute". They find the kinetic differences "modest, and readily attributable to their corresponding surface area". C9·C10·C11

Note by the AI writer (not discussed in the paper): surface area was never measured; the ranking rests on SEM images (Figure 3). The Raman slurries contain 10 mg/mL drug, about 330 times the dissolution dose, so the onset times compare the powders with each other rather than predict the dissolution test. X9·X11

Cryomilling: the 30% DL plateau falls to about the form 2 solubility

Cryomilling produced agglomerated fines with rough, damaged surfaces, whereas mortar-and-pestle particles were flat-faced fragments (Figure 3). R3·C12 Cryomilling the SE powder for 30 s gave a final level of about 75% (text ~23 µg/mL) with a little more variability; 2 min brought it to about 50% (~15 µg/mL), and 5 min to about 49%, with no further change (Figure 4, Table B). The 2 and 5 min powders plateaued from about 100 min, at about the form 2 solubility and below the extruded powder. R4·C13 Polymer release did not follow (Figure 5): all 30% DL powders released about 38–51% of their PVPVA within 2.5 min, about 72–86% by 30 min and about 96–99% by 2 h, the extruded powder being slightly slowest at 30–60 min. R5·C14 So the polymer leaves the particles in all cases; what differs is how much of the drug crystallizes first.

The authors "suggest that mechanically activating the ASD powder leads to the formation of high surface energy particles and/or nuclei/phase separated domains, which drive the rate of matrix crystallization", and point out that it "remains unclear" whether extrusion itself or only the milling that must follow it produces this; 30 s of milling of the SE powder caused some crystallization, though less than in the extruded powder. C54·C55

Note by the AI writer (not discussed in the paper): Figure 4 compares the cryomilled <100 µm powders with the average curve of the larger 100–250 µm MP fraction. Against the size-matched <100 µm MP fraction of Figure 1 (about 79%), the 30 s effect (about 75%) lies within the particle-size effect and its scatter; the 2 and 5 min effects (about 49–50%) clearly do not. X10

Drug loading: congruent release protects the drug

Figure 6 measures drug and polymer together across DL (100–250 µm, SE/MP). At 5% and 10% DL more than 90% of both drug and polymer was released within 10 min, and BCL stayed at about 91% (5% DL) and about 96% (10% DL); the authors classify both as polymer-controlled release that leaves "little-to-no amorphous solid" to crystallize. R6·R7·C17 At 20% DL the polymer ran ahead: at 10 min about 82% of PVPVA but only about 33% of BCL was released, and BCL ended at about 88%. At 30% DL the values were about 68% and 18% at 10 min, and about 86% at the end. R6·R7·C18 At 50% DL BCL release started about as fast as at 30% DL (about 46% at 30 min) but levelled off at about 73% from about 120 min, while PVPVA release did not start until about 7 min and reached only about 60% at 1 h and 80% at 2 h. R6·R7

Table C. Drug and polymer release versus DL (read from Figure 6, approximate; mean of n = 3). R6·R7·C47

The plateau of the 50% DL powder is given as "~19 µg/mL" in the text but sits at about 73% of dose, about 22 µg/mL, on the right axis of Figure 6a; the difference does not change the conclusion that this powder crystallized. X6 The authors read its delayed polymer release as the polymer being "likely" entrapped inside particles that have formed a drug-rich amorphous or crystalline surface layer. C19 The results section and the discussion grade the 20–30% DL powders differently: the results say the 5–30% DL powders "appeared to have minimal (or no) propensity to crystallize", the discussion that every incongruent loading "experienced incomplete drug release, indicating some degree of crystallization". C18·C41·X14

The 10% DL composition was then pushed hard (Figure 7). After 10 min of cryomilling it started slower, about 67% at 5 min against about 95% for 10-SE/MP, which the authors attribute to wetting problems of agglomerated fines, but it reached about 91% by 15 min and about 93% at the end; the extruded 10% DL powder also reached about 93%. R8·C20 No LoC value is given; by these data it lies between 10% and 20% DL. X15

Surfactants: faster release, but not protection against milling

With 1.5 wt% SDS, release of BCL and PVPVA became congruent and BCL reached more than 80% within 30 min (about 82% for 106–250 µm and about 92% for <106 µm), ending at about 90–94% (Figure 8a). SDS in the medium (3 µg/mL) speeded the binary powder only slightly, about 46% against 42% at 30 min, so the authors conclude that SDS "does improve wettability" but that the fast release of the ternary ASDs "resulted from a change in release mechanism to polymer-controlled release". R9·R13·C22·C24 At 3 wt% SDS release was initially congruent, but one of three replicates of the large fraction lost much of its release and the small fraction was more variable still (error bars of about ±15–20 points), while the means ended at about 86–90% (Figure 8b). R10·C23 After 5 min of cryomilling, the 3 wt% SDS powder released fastest of the three 3 wt% SDS powders (about 80% at 30 min), peaked at about 84% at 45–60 min and then fell to about 78%: desupersaturation, which the authors read as "a greater quantity of crystal nuclei was formed in the matrix". R10·C26

TPGS did not change the release mechanism: release of BCL and PVPVA stayed incongruent. With 1.5 wt% TPGS, the large fraction reached about 54% at 30 min against about 42% for the binary powder and about 50% with TPGS in the medium, and ended at about 93% (Figure 8c). The authors conclude that this faster release "resulted in the elimination of significant matrix crystallization". R11·R13·C25 At 3 wt% TPGS the MP powders ended at about 88–91% (Figure 8d). After 5 min of cryomilling the TPGS powder released slowly (about 37% at 30 min) and stopped at about 60%, about 18 µg/mL: "extensive matrix crystallization" in the authors' words, and "additional evidence" that more than surface area is involved. R12·C27

Table D. Surfactant ASDs and controls (read from Figures 1, 4 and 8, approximate; n = 3; release type from the text, with polymer data in Figures S5 and S6, not available here). R1·R4·R9·R10·R11·R12·R13

The sample labels in this part of the paper conflict, and the reader has to choose. The Figure 8 legends name samples by surfactant as a percentage of the drug loading (SDS5, SDS10, TPGS5, TPGS10), and all three entries of panel (d) read "TPGS 1.5%", while the caption, Table 1 and the methods put 3 wt% TPGS in panel (d); the text assigns the milled result to "30-TPGS1.5", but Table 1 lists only 30-TPGS3/CM5m as a milled TPGS sample. The text describing the failed 3% SDS replicate also cites "30-SDS1.5/MP". This article follows Table 1. X3·X4·X5

Note by the AI writer (not discussed in the paper): the medium controls contain 3 µg/mL surfactant, which is what the 3 wt% ASDs deliver at this dose (100 µg/mL of ASD), but twice what the 1.5 wt% ASDs deliver. For TPGS, the ASD and the medium control are close (about 54% against 50% at 30 min), so the discussion's statement that both surfactants raised release "beyond that which could be attributed to improved wetting alone" is supported for SDS but not clearly for TPGS. And the milled surfactant powders, although worse than their unmilled versions, still ended above the milled binary powder (about 60% and 78% against 49%). X12·X13·X21

What the surface looks like

SEM of compacts after 30 min (Figure 9) separated the formulations in the same order as dissolution. The 10-SE/MP surface stayed smooth. The 30-SE/MP surface was uneven and drug-enriched, F/F₀ = 1.23 ± 0.07 (p < 0.01), an increase of about 20%, which the authors take as "confirming the depletion of polymer from the surface". The cryomilled 30-SE/CM5m surface had pores and a broad, striated crystal; the extruded 30-HME/CM30s surface showed needle-shaped form 2 crystals beginning to form; the 50-SE/MP surface carried needle-shaped crystal clusters on a beehive-like layer of undissolved amorphous drug-rich domains. After 1 h, crystal clusters lay both under and on the surfaces. R14·C28 The authors suggest that the beehive layer "may" explain the delayed polymer release at 50% DL. C29

Cross-sections after 1 h (Figure 10) show the layering. The 30-SE/MP compact had a dense boundary layer 94 ± 8 µm thick containing crystals, with F/F₀ = 1.41 ± 0.10 (about 40% drug enrichment, p < 0.01), above a porous gel layer whose composition had not changed (F/F₀ = 0.93 ± 0.06, not significant). The 50-SE/MP compact had a crust of form 2 crystals only 5–10 µm thick, with a much thinner gel layer below it (F/F₀ = 0.96 ± 0.05, not significant). R15·C30

Table E. Layers in exposed compacts (Figures 10 and 11 and Section 3.4; values as printed, n = 10 for thicknesses, n = 4 for F/F₀). R15·R16·R17

Confocal imaging (Figure 11) added the time course. In 30% DL compacts the red drug-rich layer doubled from 5 to 30 min while water penetrated further and the compact swelled; its surface became irregular by 60 min, "likely due to crystallization at the surface". R16·C31 In 50% DL compacts the red domains at 5 min resembled the crystals seen by SEM, thinned to a 1.9 µm continuous layer at 30 min and were gone at 60 min; the authors read this as surface crystallization expelling the dye, and the smaller swelling as the drug-rich layer having "potentially inhibited water ingress". R17·C32·C33

Note by the AI writer (not discussed in the paper): the crystals on the compacts were identified as form 2 by their needle shape; no diffraction or Raman of the compact surfaces is reported. The layer dimensions come from compacts rotating at 100 rpm, not from powder particles, which lose polymer faster. X18·X20

Nanoscale mixing and interactions

In 13C CP-MAS spectra (Figure 12) most BCL peaks were unchanged in the three 30% DL ASDs, but the peak at 144.4 ppm, "tentatively assigned" to the aromatic carbon next to the N-H group, shifted by about 1.5 ppm (read from Figure 12B: about 142 ppm in amorphous BCL and about 144 ppm in the ASDs). The authors read this as a "possible" N-H…O=C hydrogen bond between BCL and PVPVA, in line with earlier FTIR and Raman evidence for BCL with PVP. R19·R25·C34

Table F. 1H relaxation times and the authors' homogeneity verdict (Table 2, as printed). R20

By T1, all three ASDs were mixed on a scale of about 150 nm, consistent with their single T_g. By T1ρ, only 30-SE/MP was homogeneous on about 10 nm; in both milled ASDs the BCL signal relaxed at 33.9 ms, close to amorphous BCL, while the PVPVA signal did not. R20·C35 The authors take this as support that milling creates phase-separated domains, which "would be anticipated to promote crystallization". C45·C54 Note by the AI writer (not discussed in the paper): the number of measurements and the criterion for "homogeneous" are not stated, the extruded powder was also cryomilled, and the 30 s milled SE powder and the surfactant ASDs were not measured, so Table 2 cannot separate extrusion from milling or link the heterogeneity to the crystallization rate. X17

Discussion

The authors' picture: a race between release and crystallization at the surface

The discussion (Section 4) assembles the results into one chain, drawn in Figure 13. In the incongruent regime the polymer releases faster than the drug and leaves a drug-rich layer on the particle surface; a drug-enriched layer "is highly susceptible to crystallization, since it contains a much lower concentration of polymer", and the hydrated, highly mobile drug crystallizes in it, lowering the dissolved concentration and the driving force for further release. C39·C46 The boundary layer seen in the 30% DL compact "could prevent release, promote surface crystallization, and prevent or slow continued hydration". C43 Incongruent release itself, they argue, "occurs independently of processing technique", because the polymer left all 30% DL powders at much the same rate. C40

Congruent release, in this reading, protects the drug in two ways: drug and polymer remain in the solid "at the formulated ratios, where drug-polymer interactions help to inhibit crystallization", and release finishes before the nucleation induction time. Hence "congruent release of drug and polymer may be a strategy to avoid matrix crystallization". C48 The authors hypothesize that keeping the matrix homogeneous until release shields the drug from preferential polymer dissolution and from the mobility gained when water lowers the T_g. C60

Within the incongruent regime, the rate of boundary-layer formation depends on how fast the polymer leaves, and the rate at which the layer crystallizes depends on the processing history; the authors "hypothesize" that this is why 30-SE/MP and 30-SE/CM5m dissolve differently although their crystallization kinetics are similar. C44 The 50% DL powder is the extreme: at greatest risk "due to greater density of metastable amorphous drug in the drug-rich layer", it forms a thin crystal crust that blocks hydration and release, and its "slow, seemingly congruent" release reflects repeated dissolution of form 2 crystals that exposes polymer while new crystals form. C42·C47 Note by the AI writer (not discussed in the paper): the crystallization kinetics of 30-SE/CM5m were not shown (Figure 2 covers only the SE/MP and HME powders), and the polymer content of the 50% DL crust is inferred from the dye, not measured. X8·X16

Why milling matters

For mechanical activation the authors give two reasons why a milled sample "is likely to undergo surface disruption at the molecular level at faster rates": it is less homogeneous at the nanometre scale (Table 2), and fractures in mechanically treated amorphous solids are known to favour heterogeneous nucleation (Su et al.; Shi et al.). C45 For congruently releasing powders the effect was "diminished": the fast dissolution of the 10% DL ASD "may have been sufficient to overcome the crystallization tendency induced by cryomilling". C56 For surfactants, they point to reports of surfactant enrichment at the surface of amorphous drugs (Yu et al.), which "may provide for high local mobility", as "a potential underlying mechanism" for the extra susceptibility after milling, and conclude that "use of a surfactant may be a high-risk formulation strategy if milling is required". C58·C57 They also report that, to their knowledge, this is the first account of mechanical activation in an ASD with high polymer content. C64

Why BCL in particular

The authors note that not all ASDs show matrix crystallization and offer reasons that "can be speculated upon" for BCL: its hydrogen bonds to PVP-type polymers (also seen here by ssNMR) are, by analogy with other systems, rapidly disrupted on hydration; amorphous drugs above their "wet" T_g are more mobile; amorphous BCL crystallizes rapidly above its T_g despite high glass forming ability; and amorphous BCL can be mechanically activated. C63·R25·R26 None of these was measured in the hydrated ASDs. C63

What it suggests for formulation

The authors' recommendations are direct: "the best-case scenario when formulating an ASD is to select a DL at which the drug and polymer release congruently, i.e., a DL at or below the LoC"; surfactants can raise the release rate and allow a higher DL, but their presence "was a risk factor for matrix crystallization", particularly if the powder must be milled. C61·C62·C57 They stress that once matrix crystallization occurs, no further solubility advantage is gained beyond the drug already released, and that the new crystals "may act as seeds" for growth from solution. C68 In the abstract this becomes a simple risk list: high DL, mechanical activation and surfactant raise the risk; rapid, congruent release lowers it. C67

Note by the AI writer (not discussed in the paper): for products that must be milled, such as extrudates, the data favour a congruently releasing loading; above the LoC, the choice of size-reduction method mattered at least as much as the choice between solvent evaporation and extrusion, since 2–5 min of cryomilling took a 30% DL SE powder below the extruded one. A non-sink test at the intended dose and particle size that measures drug and polymer together shows both the release regime and the risk; a plateau near the solubility of the metastable form is the warning sign. R1·R4·M1

Limitations

Stated by the authors

• It "remains unclear" whether HME itself produces high-energy particles, nuclei or phase-separated domains, or whether only the milling that HME requires does. C55
• Other factors not studied "may" govern susceptibility: the T_g of the drug and/or ASD, the persistence of drug–polymer interactions, and the kinetics of amorphous–amorphous phase separation. C70
• How and why surfactants change release behaviour is "an active area of research". C51

The AI writer's own judgement

The core dissolution results are solid: they are replicated (n = 3), the effects are large (for example about 79% versus about 49–50% after 2–5 min of milling at the same sieve size, with polymer release unchanged), and the surface imaging agrees with them. R1·R4·R5·R14·R15 The mechanistic layer is weaker:

• Surface area was never measured, so the separation of surface-area effects from mechanical activation rests on SEM images and on the size-matched comparison, which the paper's own Figure 4 does not make. X9·X10
• Mechanical activation is inferred from dissolution behaviour and one T1ρ comparison without stated replicates; the milled powders were amorphous by XRPD and no nuclei or surface energy were measured. X17·X19
• Crystallization kinetics were shown only for the SE/MP and HME powders, at 10 mg/mL, far above the dissolution dose. X8·X11
• Several sample identities conflict (Table 1 composition of 30-SDS1.5 sums to 98.5 wt%; Figure 3 and Figure 8 labels; the milled TPGS sample), the 50% DL plateau differs between text and figure, and the results and discussion disagree on whether 20–30% DL crystallized. X1·X3·X4·X5·X6·X14
• The surfactant controls carry twice the surfactant of the 1.5 wt% ASDs, and a TPGS effect beyond wetting is not shown; the milled surfactant powders were worse than their unmilled versions but not worse than the milled binary powder. X12·X13·X21
• Surface layers were measured on compacts, and the crystals there were identified by shape; one drug, one polymer and one buffer were studied. X20·X18

Conclusion

This study turns a known failure, crystallization of an ASD while it dissolves, into a set of measured risk factors for one fast-crystallizing drug. With a dose that excludes crystallization from solution, BCL–PVPVA dispersions lost release in proportion to how quickly a polymer-poor, drug-rich layer formed at their surface and how readily that layer crystallized: incongruent release above the LoC, cryomilling, and surfactants combined with milling all made the loss worse, down to about the solubility of form 2. R1·R4·R6·R12·R15·C65 Drug loadings of 5–10%, where drug and polymer dissolve together, released completely regardless of milling or extrusion, which supports the authors' advice to formulate at or below the LoC when matrix crystallization is a risk. R6·R8·C61·C66 The explanations offered for why milling and surfactants raise the risk, nanoscale heterogeneity, nuclei, surface energy and surfactant mobility, remain to be measured directly. C54·C58·X19