When a few leftover crystals decide how an amorphous solid dispersion dissolves: bicalutamide in PVPVA

Background and motivation

An amorphous solid dispersion (ASD) holds a poorly soluble drug in the amorphous state inside a polymer, so that it can dissolve to concentrations well above the solubility of its crystals. Its value depends on keeping the drug amorphous, both in the solid and in solution. On contact with water an ASD can dissolve and create a supersaturated solution, or the solid itself can crystallize before it dissolves; the paper calls the second route matrix crystallization. Crystals can also nucleate and grow later from the supersaturated solution. Polymers can stabilize the hydrating solid, shape the release mechanism and slow crystallization from solution (Alonzo 2010, ref 17; refs 16, 18–21). C1

Matrix crystallization is the least understood of these routes. If the amorphous state is lost in the solid, the solubility advantage is never realized, and most evidence for this route comes from imaging or Raman spectroscopy of crystals forming in or on amorphous solids (Raina 2014, ref 18; Xie 2016, ref 19; Saboo 2020, ref 26; Savolainen 2009, ref 27). C2

Dissolution testing is the usual way to see whether an ASD has been damaged by processing or storage. Sink conditions, set historically for drugs whose solubility was not an issue, say little about supersaturating products; non-sink conditions capture both dissolution and crystallization, and in a series of tacrolimus studies they detected crystallization after accelerated storage and predicted in vivo performance (Sun 2016, ref 23; refs 17, 32–37). Dissolution can also be more sensitive to crystals than solid-state methods (refs 38, 39). Hermans et al. used non-sink dissolution to estimate the level of 60-µm bulk crystals but under-predicted crystallinity when the seeds were 2-µm jet-milled particles, thought to dissolve quickly (ref 40). C3

Crystals inside an ASD are generally considered harmful because they can seed further crystallization on storage or during dissolution, but acceptable thresholds of crystallinity are not known (refs 23, 40, 43, 44). Earlier work from the same group and others showed that the seeds themselves matter: paclitaxel seeds with different shapes and fast-growing faces changed desupersaturation rates (Que 2018, ref 45); in one in vivo study a fully crystallized ASD gave a higher exposure than reference crystals, suggesting defective crystals (ref 36), while another found the opposite (Wilson 2018, ref 15); an effective polymer could stop residual seeds from growing, and residual crystals from hot melt extrusion were more potent than bulk crystals (Moseson 2020, ref 24); an absorptive sink lessened the effect of seeds (Hate 2020, ref 37). C4

Hot melt extrusion (HME) dissolves crystalline drug into molten polymer. If temperature, time or input particle size are not right, part of the input crystals survives (Moseson 2018, ref 47; ref 48). C5 This paper uses that weakness on purpose. With one drug, bicalutamide (BCL), one polymer, the vinylpyrrolidone–vinyl acetate copolymer PVPVA, and one composition, it makes batches with different amounts of residual crystallinity, measures that crystallinity with five methods, tests dissolution at several doses with residual or added seeds, and follows which polymorph crystallizes, where and how fast. A1·A2·A3·A4

Background concepts

Solubility of the three solid forms. BCL is a Biopharmaceutics Classification System class 2 drug with two known polymorphs: the thermodynamically stable form 1 and the metastable form 2. In pH 6.8, 50 mM phosphate buffer at 37 °C their apparent solubilities are 4.3 and 15 µg/mL, and the amorphous form reaches 150 µg/mL, a maximum solubility advantage of about 35 times over form 1 (150/4.3). M1·R1·R2 The amorphous value is the amorphous solubility: the concentration at which a drug-rich liquid phase separates from the solution (liquid–liquid phase separation). The authors detected it with the fluorescent probe pyrene, which reports the polarity of its surroundings, and with UV extinction, which picks up light scattering by the drug-rich droplets; nanoparticle tracking and dynamic light scattering looked for the droplets directly. The measurement needed polymer in the buffer, because BCL crystallized too fast without it. P9·P10·P11·R27

Supersaturation, sink and non-sink. The doses in this paper are 3, 30, 60 and 150 µg/mL of BCL. The 3 µg/mL dose is below the form 1 solubility, so in principle everything can dissolve; the authors call it a sink condition while noting that it does not meet the traditional definition of sink conditions. It was the lowest dose they could weigh without adding solubilizers. C24·P13 The other three doses are non-sink: if the ASD released all its drug, the solution would be supersaturated with respect to both crystalline forms, and at 150 µg/mL it would sit right at the amorphous solubility.

Two crystallization routes. In solution-mediated crystallization the drug first dissolves; crystals then nucleate or grow from the solution and the concentration drops, a pattern called desupersaturation. In matrix (solid-mediated) crystallization the undissolved amorphous solid turns into crystals in contact with the medium, so the drug never reaches solution. A dissolution curve that rises and then stays flat, without a drop, while the amount dissolved stays far below the dose, points to the second route; a rise followed by a clear fall points to the first. C19·C20

Polymorphs and Ostwald's step rule. When a supersaturated system crystallizes without seeds, the less stable form often appears first (Ostwald's step rule). That matters here: form 2 is about 3.5 times more soluble than form 1, so a solid that turns into form 2 can still hold the solution at a higher concentration than one that turns into form 1. R1·C62

Crystal growth inhibition by polymers. A polymer dissolved in the medium can adsorb on crystal surfaces and slow their growth. The authors express its effectiveness with the polymer effectiveness ratio (Eq. 1, after Schram et al., ref 54):

Here $R_{G,np}$ is the crystal growth rate without polymer and $R_{G,p}$ the rate with polymer. In plain terms: $E_G$ says how many times slower the crystals grow when the polymer is present; a value above 1 means the polymer slows growth. E1

Congruent and incongruent release. In an ASD the drug and the polymer can leave the solid together (congruent release) or at different rates. When the polymer leaves first, which is common at high drug loading, the remaining solid becomes enriched in drug and may form a drug-rich layer at the surface (Saboo 2019, ref 10; Saboo 2020, refs 11, 26). C34·C54·C58

Detecting crystals in an ASD. Each method sees crystals differently. X-ray powder diffraction (XRPD) measures crystalline diffraction but is diluted by the amorphous matrix; polarized light microscopy (PLM) sees birefringence but not crystals below the optical diffraction limit, and strained glass can also be birefringent; differential scanning calorimetry (DSC) can miss crystals that dissolve into the miscible polymer melt during heating; scanning electron microscopy (SEM) sees only surfaces; transmission electron microscopy (TEM) resolves nanometer crystals by contrast and electron diffraction, but samples only small fragments. C12·C13·C14·C36

Methods in detail

The study has four strands that feed one comparison: make extrudates with graded residual crystallinity and measure it; characterize BCL in solution; test dissolution at four doses; and watch the solid crystallize in buffer. The flow below shows how they connect.

Making extrudates with graded residual crystallinity

Problem. To study residual crystals one needs batches of the same composition that differ mainly in how many crystals survived processing. What they did. The authors prepared three lots of crystalline form 1: the as-received powder ("small", a highly cohesive micronized powder with a median size D50 of 3.5 µm and D10–D90 of 1.3–8.6 µm), a "medium" lot recrystallized from ethanol by cooling at 10 °C/min from 60 to 10 °C (D50 38 µm), and a "large" lot made by direct nucleation control, cooling at 0.1 °C/min with temperature cycles triggered by a focused beam reflectance probe (D50 93 µm). All lots were sieved below 250 µm and confirmed as form 1 by XRPD and DSC. M1·M2·M3 Amorphous BCL was made by melting at 210 °C for 10 min, and form 2 by holding an amorphous melt at 100 °C for 1 h. M4·M5

Extrudates of 30% BCL in PVPVA were made in a 5-mL co-rotating extruder with a recirculation loop. The product melt temperature was placed in the "dissolution regime" of the BCL–PVPVA temperature–composition phase diagram for this composition, between Tc = 145 °C and Tm = 195 °C, where crystals dissolve into the melt at a rate that depends on temperature and particle size; residence time and screw speed were held constant. The small lot was extruded at 180 and 155 °C (S/180 and S/155); the medium and large lots at 155 °C (M/155 and L/155). P1·M1 Flory–Huggins modelling of the phase diagram gave an interaction parameter χ = −1.57, which the authors read as favourable miscibility; together with literature on high-loading BCL dispersions in PVP, they expected acceptable physical stability. C10 Why it works. Larger input crystals and a lower temperature leave more undissolved crystals, so the same formulation spans from no detectable crystals to 30% crystallinity. The authors present the size range as the extremes of real-world input variation. C11

Detecting the crystals: five methods

Modulated DSC heated 3–5 mg of extrudate from 0 to 180 °C at 5 °C/min. XRPD used two settings: a survey scan from 5 to 40° 2θ at 2°/min, and a slow scan of 16–18° 2θ at 0.5°/min to quantify the area of the 17.2° peak. The calibration (0–65% crystalline content, R² = 0.9952) was made by mixing crystalline BCL into a fully amorphous extrudate, with a limit of detection of about 1% and a limit of quantification of about 3% crystalline content. P2·P3 PLM looked at particles in mineral oil under crossed polarizers (at least five images per sample), backscatter SEM at particle surfaces, and TEM at fragments wiped from the vial lid onto carbon grids, in bright field and electron diffraction, searching at least three grid squares for particles thin enough to image. P4·P5·P6 The basis of "% crystallinity" (share of the drug or of the whole extrudate) is not stated in the main text. X7

Solution behaviour of BCL

Apparent solubility was measured by adding 50 mg of each solid form to 100 mL of buffer at 37 °C, stirring at 100 rpm for about 20 h and following the concentration in line with second-derivative UV (298–308 nm). P8 Nucleation induction time was measured after creating supersaturation by adding a DMSO stock to 100 mL of buffer with or without polymer; the induction time was the time at which the concentration had fallen by 1%. Crystal growth was measured by adding form 1 seeds (small lot) to solutions at 30 and 150 µg/mL and following the initial fall in concentration. The polymers were PVPVA at 70 µg/mL, the same as the PVPVA dose that accompanies the 30 µg/mL BCL dose of the extrudate, and HPMCAS (hydroxypropyl methylcellulose acetate succinate) at 1 mg/mL. P12·A4 The growth rates themselves are in Figure S10b of the supporting information, so Eq. 1 cannot be recomputed here. E1

Dissolution and polymer release

Dissolution was followed for 4 h at 37 °C in pH 6.8, 50 mM phosphate buffer at doses of 3, 30, 60 and 150 µg/mL BCL, in triplicate. Besides the four extrudates, the authors tested a crystalline physical mixture (small crystals; 100% crystalline) and the crystal-free S/180 with form 1 seeds (small lot) added at 10% of the BCL dose. A variant with 1 mg/mL HPMCAS dissolved in the medium is described in the methods, but no dissolution data with HPMCAS are shown in the main text. Solids from the 150 µg/mL runs were filtered and identified by XRPD. P13 PVPVA release was measured at the 30 µg/mL dose (70 µg/mL PVPVA) for 1 h by withdrawing 2 mL samples, filtering them and quantifying the polymer by size-exclusion chromatography with UV detection at 205 nm. P14

Watching the solid crystallize

Two methods followed crystallization of the extrudate itself. In hot-stage PLM, particles were immersed in buffer on a glass slide at 37 °C and photographed under crossed polarizers at 0, 10, 20, 30 and 60 min. P15 In the Raman slurry test, 400 mg of extrudate was suspended in 12 mL of buffer (10 mg/mL BCL) at 37 °C and 200 rpm, and an immersion probe recorded a spectrum every 2 min for 2 h, starting at 1 min. Multivariate curve resolution with alternating least squares (MCR-ALS) split the spectra between 1551 and 1166 cm⁻¹ into a few components with non-negative, closed concentrations, starting from the spectra of the pure solid forms in buffer; the components matched the amorphous dispersion (component 1), form 2 (component 2) and form 1 (component 3). The relative growth rate is the slope of the linear part of each replicate's component curve, and groups were compared with paired one-tailed t tests. P16·P17 How this differs from earlier work. Earlier evidence for matrix crystallization came from imaging and Raman of neat amorphous drugs or compacts; here in-line Raman resolves the polymorph and its kinetics for ASD powders with known residual crystallinity, side by side with dissolution data. C2·C52

Note by the AI writer (not discussed in the paper): the slurry holds about 67 times more drug per volume than the amorphous solubility and 67 to 3,300 times more than the dissolution doses, and the hot-stage particles sit in a thin film of buffer. Onset times and growth rates from these tests rank the samples well but are not the times that apply in the dissolution vessel. X8

Experimental analysis

BCL crystallizes fast from solution, and polymers slow it there

Without polymer, supersaturated BCL nucleated in under 30 min, and in under 5 min at the highest supersaturation tested; the authors call it "a fast crystallizing compound in a solution environment". With 70 µg/mL PVPVA the induction time shortened as supersaturation rose but stayed above 1 h at the highest concentration measured, and with 1 mg/mL HPMCAS no nucleation occurred during the 8 h of monitoring. R3·C6 Seeded form 1 crystals barely grew at 30 µg/mL, with or without polymer, but grew "extremely rapid[ly]" at 150 µg/mL; PVPVA slowed that growth 8.6-fold and HPMCAS 158-fold. R4·R5 The authors conclude that HPMCAS is the more effective inhibitor, that PVPVA is "moderately effective" with growth remaining "at a rapid, yet reduced rate", and that nucleation from solution is "not expected" during the dissolution tests because no concentration exceeded about 40 µg/mL. C7·C8 The supersaturation levels behind these statements are in the supporting information. Note by the AI writer: the PVPVA induction time of "above 1 h" is shorter than the 4 h test, and the polymer concentration in the higher-dose tests (up to 350 µg/mL if fully released) was not tested in the main text. X5

The four extrudates: only TEM separates the two X-ray amorphous batches

All four extrudates showed a single glass transition near 98 °C (inflection), the sign of a dispersion. R7 Table A sums up which method detected crystals.

Table A. Crystals detected in the four extrudates by each method (from Table I, with the processing conditions of the methods section). R6·P1

Each method failed in its own way. DSC showed crystals only for L/155, and not as a melting peak: a shallow, broad endothermic shift of the non-reversing heat flow between 120 and 170 °C, labelled a "crystal dissolution endotherm" in Figure 1a, because crystals dissolve into the miscible polymer melt as the sample is heated. R7·C12 The XRPD survey scan showed crystalline peaks only for L/155; the slow scan quantified M/155 at 3.5% and L/155 at 30% and left S/180 and S/155 below the detection limit. R8·R6 PLM showed bright birefringent crystals in M/155 and L/155, but strained glass in the other images was birefringent too (hazy orange or blue), which makes birefringence alone "somewhat equivocal". R9·C13 SEM saw smooth surfaces on three samples and a rough L/155 surface studded with crystals 340 ± 140 nm long (n = 30); the authors note that SEM sees little here because the crystals are mostly inside the particles. R10·C14 TEM, finally, showed a uniform S/180 fragment without diffraction spots, and heterogeneous fragments with embedded crystallites and ring-like diffraction for the other three; in an S/155 fragment the crystals measured 140 ± 10 nm (n = 10). R11

On this evidence the authors classify S/155 as containing below 1% crystallinity and S/180 as fully amorphous ("is considered"). C15 They read the nanometer size as typical of HME residual crystals, which shrink as they dissolve in the melt, and the uniform size as reflecting the narrow size distribution of the small input lot. C16 Two things should be kept in mind. "0%" means that TEM found no crystals in the fragments examined, not a measured zero. X6 And the authors' statement that crystals of 100–500 nm are undetectable by PLM sits awkwardly with PLM detecting crystals in M/155 and L/155; those detections are probably of larger crystals or clusters. C36·X15

The crystal-free ASD stops far below its amorphous solubility

Figure 2 holds the dissolution curves, and Figure 6 condenses them into the 4-h concentration and the fraction of the dose dissolved. At 30 µg/mL, S/180 rose steadily for about 2 h to about 18 µg/mL and stayed there; at 60 and 150 µg/mL it peaked at about 37 and 38 µg/mL after 1.5–2 h (the text says ~40 µg/mL), then declined slowly to about 34.5 and 35 µg/mL at 4 h. R12·R13·R14 Recovered solids after about 20 h contained both form 1 and form 2. R16 The fraction dissolved at 4 h was about 0.82, 0.59, 0.475 and 0.23 at 3, 30, 60 and 150 µg/mL. R17

The authors draw three conclusions from this pattern. The flat 30 µg/mL curve, with no desupersaturation, "indicates that crystallization occurred in the dissolving solid preventing complete release". C17 Raising the dose from 60 to 150 µg/mL "did not contribute to generating a greater level of supersaturation" before crystallization overtook dissolution. C18 And since every dose was at or below the amorphous solubility, the ceiling near 40 µg/mL shows that matrix crystallization dominates ("it is clear"); the falling fraction dissolved "provides support" that matrix crystallization is faster than dissolution. C41·C42 At 30 µg/mL the plateau lies just above the form 2 solubility of about 15 µg/mL, which the authors connect to the smaller solid surface left after matrix crystallization. C47 Figures 2b and 6a disagree on one value: the S/180 concentration at 60 µg/mL and 4 h is about 34.5 µg/mL in Figure 2b but about 28.5 µg/mL in Figure 6a (and a fraction of about 0.475 in Figure 6b); all other sample–dose pairs agree between the two figures. X2

Residual crystals, at any level, lower the supersaturation

Table B. Fraction of the dose dissolved at 4 h, with the 4-h concentration in µg/mL in brackets, and the peak concentration at the 150 µg/mL dose (all about, read from Figures 6a, 6b and 2a). The crystalline physical mixture stayed at its solubility of about 4.3 µg/mL at the non-sink doses. R17·R12·R13

Every sample with crystals, residual or added, ended below the crystal-free S/180 at every dose (Table B). R17·C38 At 30 and 60 µg/mL the three crystal-containing extrudates reached a plateau within 15–60 min and stayed nearly flat; the authors read the lack of desupersaturation as growth through the solid matrix being the dominant pathway ("is thus suggested"). R13·R14·C20 At 150 µg/mL the picture differs: S/155 followed S/180 for about 45 min, reaching about 32 µg/mL, and M/155 rose fast for 30 min to about 28 µg/mL, after which both fell, to about 24.5 and 16.2 µg/mL at 4 h. The authors' account is that the large amorphous fraction dominates early, and that the residual crystals then grow "extensively through the matrix" to a high crystalline surface area, at which point growth from solution outpaces dissolution. L/155, with far more crystals, levelled off at about 12 µg/mL (the text says ~10) and changed little. R12·C21·C22 Only form 1 was found in the solids recovered from the crystal-containing extrudates. R16

Adding 10% form 1 seeds to S/180 gave plateaus in the range of the residual-crystal extrudates at every dose, which the authors take to mean that crystals, whatever their origin or initial location, grew rapidly through solid-mediated crystallization once associated with the amorphous solid ("suggests"). R12·R13·R14·C23 The loss of supersaturation is not proportional to crystal content: at 30 µg/mL a trace below 1% already cut the fraction dissolved from about 0.59 to 0.36, while 30% took it only to about 0.22. The authors explain this, "presumably", by the fast growth of form 1 in the matrix whenever any seeds are present. R17·C44

Near sink, crystals make the ASD slower than crystalline drug

At 3 µg/mL, a dose below the form 1 solubility, S/180 dissolved fastest and reached about 2.45 µg/mL at 4 h in Figure 2d (the text gives a maximum of 2.8 µg/mL, which the figure does not show). The crystalline physical mixture started more slowly but reached about 2.42 µg/mL, consistent, the authors say, with the solubility of its solid form. R15·C25·X1 All crystal-containing samples stayed lower: about 1.92 µg/mL for S/155, 1.80 for S/180 with seeds, 1.52 for M/155 and 1.35 for L/155 at 4 h, slower even than the fully crystalline mixture. R15

The authors' explanation ("can be explained by") is that the hydrated amorphous solid around a residual crystal is locally highly supersaturated, whatever the bulk concentration, so the crystals grow quickly, consume the amorphous solid and end as relatively large or agglomerated crystals with low relative surface area, which dissolve more slowly than the small crystals of the physical mixture. C26·C67 They suggest that Wilson et al. "likely" saw the same effect when a crystalline enzalutamide physical mixture outperformed a crystallized ASD. C68·R30 The abstract describes these crystals as having "high relative surface area", contrary to the results and discussion; the discussion's wording is the one that fits the argument. X3 Neither the surface area nor the size of the grown crystals was measured. X10

Where and when the solid crystallizes

Under PLM (Figure 3), crystals appeared at the edges of S/180 particles around 10–20 min, and some amorphous solid remained at 60 min with crystals at the interfaces and throughout. In S/155 crystals appeared inside the particle after about 20 min, once they had grown large enough to see, and no amorphous solid was left at 60 min. L/155 showed crystals from the first image and looked fully crystalline within 20 min. Added seeds lay in the medium, some next to the particle, and the particle was consumed. R18·C64 The authors conclude that crystals grow by matrix crystallization within the particles and that the presence and level of residual crystals change the kinetics ("Clearly"). C27·C28

Table C. Crystallization in the Raman slurries (10 mg/mL BCL, 37 °C). Onsets as stated in the text; half and full conversion and relative growth rates (no unit printed) about, read from Figure 4; n = 3. R19·R20·R21·R22·X14

In-line Raman confirmed and timed this (Table C). S/180 turned into form 2 after a lag, starting at 20–30 min and complete within about 1 h. Every sample containing form 1, residual or added, grew form 1 only, and the lag shrank as the crystal content rose; seeded S/180 needed no lag at all. R19·R21 XRPD confirmed the polymorphs, and SEM of the solids harvested after 2 h showed needle-shaped form 2 crystals growing in circular clusters and irregular plates of form 1, both unlike the input crystals, which the authors attribute to growth under different supersaturation or with additives. R16·R23·C33 The authors read the switch from form 2 to form 1 as Ostwald's step rule at work when no seeds are present, overridden by growth of the stable form when seeds are present; since the two forms differ about 3.5-fold in solubility, the polymorph sets the supersaturation that remains. C62·C49

HPMCAS in the slurry medium delayed the S/180 lag by about 10 min but left the form 2 growth rate unchanged (about 0.059 against 0.056, not significant), and form 2 still formed. The authors take this as further confirmation that crystallization runs through the solid ("again confirms"). R19·R20·C29 For S/155 the same additive did slow form 1 growth, from about 0.071 to 0.054 (p < 0.01). R22 M/155 crystallized faster than S/155 (0.088 against 0.071, p < 0.05), which the authors relate to a larger overall crystal surface area. R22·C30 The surprising result is the seeded sample: with 10% added crystals, far more by weight than 0–3.5%, its form 1 grew at only about 0.044, slower than both residual-crystal extrudates (p < 0.01). The authors conclude that the external seeds offer less overall surface area for growth. R22·C31

The polymer leaves first

At the 30 µg/mL dose, PVPVA from M/155 was already at about 55 µg/mL after 2.5 min and reached about 63 µg/mL, some 90% of the 70 µg/mL polymer dose, within 20 min, then levelled off. From S/180 it started at about 36 µg/mL, slowed after 15 min and reached about 59 µg/mL by 60 min (Figure 5). Over 4 h only about 30–60% of the BCL dose dissolved, so the polymer clearly leaves faster than the drug; the authors call this "indicative of incongruent release". R25·R17·C34 From these data they compute that the undissolved solid went from 7:3 PVPVA:BCL to 1:1 within 5 min for S/180, and to 1:2 for the extrudate with residual crystals. R26 SEM after 20 min in buffer (Figure S18) showed rough, pitted surfaces like those of incongruently releasing ASDs; with residual crystals, the surface had two distinct regions, which the authors read as partly crystallized and partly amorphous, "presumably drug-rich". R24·C32

Discussion

The authors' account of the two pathways

The authors assemble the results into one sequence (Figure 7). PVPVA dissolves preferentially from the solid, "as would be expected at such a high drug loading", and the solid gains "significant molecular mobility upon hydration enabling it to crystallize". C54 Having lost most of its polymer within minutes, the hydrated drug-rich solid "likely has a Tg that is depressed below" 37 °C and behaves as a supercooled liquid; the dry Tg of BCL is 53 °C and several percent of water "would be expected". The wet Tg could not be measured because neat amorphous BCL crystallized as soon as it met moisture. C55·R27 With the polymer gone, drug–polymer interactions that may have stabilized the dry dispersion are disrupted, and drug molecules can reorganize into crystals or join crystals already present. C56 From there the path splits.

The authors' proposed pathways, after Figure 7. Step 1 and the outcomes are measured; the mobility of step 2 and the drug-rich layer are the authors' interpretation. R26·C55·C45·C49·C57·C58·C67

Without residual crystals, polymer release proceeds until a drug-rich layer forms and slows it; after 20–30 min form 2 nucleates at the interfaces, and dissolution and form 2 growth compete until all amorphous solid is consumed and the remaining polymer is expelled. The authors call this "a surface-driven reaction, triggered by both hydration and rapid depletion of polymer". C45·C58·C59 The slight desupersaturation at 150 µg/mL "can be attributed to" the larger amount of solid and the higher supersaturation, at which PVPVA slows growth less. C46 With residual form 1 crystals, drug integrates into them from the first contact with water, which "likely" expels polymer faster and needs no induction time; even at below 1%, form 1 growth outpaces form 2 nucleation, and all the amorphous solid is consumed within an hour. C49·C57 Because this growth is driven by the local supersaturation inside the hydrated solid, the authors describe the pathway as "largely insensitive to the solution environment", the reason it persists under sink conditions. C50

The authors list why BCL is vulnerable: its physicochemical properties, including a low Tg, a high crystallization tendency, the high drug loading and the rapid loss of polymer, the crystallization inhibitor. They add that BCL is known to be mechanically activated by milling and compression, but that the relevance to this study is "unknown", since all extrudates were made by the same basic process. C60·C61 More generally, they suggest that the crystallization tendency of the hydrated amorphous drug and the lack of persistent drug–polymer interactions "likely play a role" in making matrix crystallization dominant, and they place their results next to earlier reports on nifedipine, celecoxib, indomethacin methyl ester, indomethacin and carbamazepine. C53·C52

Why residual crystals are such potent seeds

External seeds are often added in vitro to predict what residual crystals will do, but "all seeds are not created equal" (refs 45, 77–79). C63 Here the 10% external seeds reproduced the dissolution behaviour of the residual-crystal extrudates and suppressed form 2, but grew more slowly than residual crystals at <1% or 3.5%. C64·R22 The authors rationalize this ("can be rationalized") by surface area rather than mass, and by location and history: the residual crystals are nanometer sized (140–340 nm against a seed D50 of 3.5 µm), at least an order of magnitude smaller than the external seeds; they sit inside the amorphous solid, next to drug at high local supersaturation; and residual crystals from HME have been found to be highly defective, which "may also contribute". On this basis they rank the effective surface area for growth as seeded S/180 < S/155 < M/155. C65·C66·R10·R11 They draw a broader point: small particles below 1 µm, traditionally seen as low risk because they dissolve quickly, are "clearly not necessarily" low risk on a matrix crystallization route, where confined nanometer crystals grow into an agglomerated mass of low surface area. C69

Risk assessment

An earlier paper by the group (ref 24) described two scenarios: a low-risk one, as for indomethacin/PVPVA, where polymer adsorption keeps crystal growth near zero so residual crystals do little; and a high-risk one, where crystals lower the rate or extent of dissolution. R29 BCL/PVPVA falls in the high-risk scenario, "even for crystallinity levels that were undetectable by XRPD"; the two X-ray amorphous batches diverged because S/180 nucleated the more soluble form 2 while trace form 1 crystals in S/155 grew through the solid. C70·C71 The authors condense the study into five risk factors: C72·C73·C74·C75·C76

• Matrix crystallization. "Of all crystallization routes, this pathway is the most critically detrimental", because the solubility advantage is never, or only moderately, realized; polymers in the medium were "largely ineffective", and residual crystals remove the induction time. C72
• Solution-mediated crystallization. Polymers that inhibit growth from solution reduce risk, but here the matrix route "limited the impact" of that inhibition. C73
• Seed properties. Residual seeds, mechanically damaged, small and embedded, are "commonly more potent than bulk seeds". C74
• Polymorphic form. Crystallizing to the metastable form is preferable, so seeds of the stable form are "highly detrimental". C75
• Supersaturation conditions. Higher doses give more supersaturation and faster growth from solution, so seed potency "would be greatest" in a low-fluid-volume scenario; matrix crystallization occurred at every dose. C76

For detection, the authors argue that XRPD, PLM and DSC each have a blind spot for low levels of small crystals, that higher-resolution methods "must be employed", and that the method used for a critical quality attribute "should be fit for purpose": here only TEM separated two batches that performed differently. C36·C37·C38 They also recall that, ideally, specifications for input materials and a process design space would prevent residual crystallinity in the first place. C35

Note by the AI writer (not discussed in the paper): for formulation work the practical message is narrower than "crystals are bad". In this system a solution precipitation inhibitor (HPMCAS) did not rescue the solid, and a batch that passed XRPD failed in dissolution. Remedies would have to act in the solid or the process, for example complete amorphization during extrusion (180 °C rather than 155 °C with the same input crystals), or a composition in which the polymer does not leave the drug behind; the paper tests none of these alternatives beyond the two extrusion temperatures.

Limitations

Limits the authors state. The 3 µg/mL dose does not meet the traditional definition of sink conditions. C24 Birefringence alone is "somewhat equivocal", SEM sees little because the crystals are inside the particles, and crystals in the 100–500 nm range escape PLM. C13·C14·C36 The wet Tg of amorphous BCL could not be measured. R27 The relevance of mechanical activation of BCL is "unknown". C61 The underlying mechanisms of matrix crystallization "must be understood in order to mitigate this failure mode", which this paper does not yet do. C51

The AI writer's own judgement (1): what is measured and what is inferred. The core observations are solid. The order 0% > <1% > 3.5% and seeded > 30% holds at every dose in triplicate; two independent methods, PLM and in-line Raman, see the crystals form inside the particles; XRPD identifies the polymorph; and the external-seed control, which reproduces the residual-crystal behaviour starting from the crystal-free batch, ties the effect to crystals rather than to processing. R17·R18·R19·R21·R16 The mechanism is less secure. The glass transition of the hydrated drug-rich solid was not and could not be measured, and the polymer-depletion ratios come from solution data, not from analysis of the solid. X11 The agglomerated, low-surface-area crystal mass and the surface-area ranking of seeds are inferred from rates; no surface area or size of the grown crystals was measured. X10

The AI writer's own judgement (2): conditions and comparisons. The crystallization kinetics come from slurries at 10 mg/mL BCL and from particles on a hot stage, far from the dissolution vessel; the onset of S/180 crystallization is 10–20 min by PLM and 20–30 min by Raman, and the discussion applies the latter to dissolution. X8 S/180 and S/155 differ in extrusion temperature as well as in crystal content, and M/155 and L/155 in input crystal size; the seeded control makes the crystal explanation the more likely one. X9 The basis of "% crystallinity" is not stated, and Figure 6c plots the seeded sample at 10% crystallinity although the seeds were 10% of the dose, so seed level and residual crystallinity compare only qualitatively. X7 The statement that solution crystallization is minor because growth from solution is slow with PVPVA (abstract and conclusion) does not hold at the 150 µg/mL dose, where the authors themselves attribute the fall of S/155 and M/155 to growth from solution, and the nucleation data with PVPVA cover a shorter time than the 4 h test. X4·X5

The AI writer's own judgement (3): inconsistencies in the printed paper. The text gives 2.8 µg/mL as the S/180 maximum at 3 µg/mL, but Figures 2d and 6b show about 2.45. X1 The S/180 value at 60 µg/mL and 4 h is about 34.5 µg/mL in Figure 2b and about 28.5 µg/mL in Figure 6a. X2 The abstract says the agglomerated crystals have "high relative surface area"; the results and discussion say "low", which is what the argument needs. X3 The surface-area reductions of the input lots are printed as 12- and 7-fold in the order medium, large, which does not follow particle size; the BET data are in the supporting information. X13 The SEM images in Figure 4 are described as solids "harvested after 2 h of dissolution" but, by the methods, come from the Raman slurries. X12 None of these changes the direction of the conclusions; the supporting information, which holds the solubility, nucleation and growth curves, the extrusion conditions and the dissolution design, was not available for this reading.

Conclusion

The paper shows, for one drug–polymer pair at one drug loading, how residual crystallinity damages an amorphous solid dispersion. BCL in PVPVA is "relatively robust" against nucleation and growth from solution, yet it never approaches its amorphous solubility, because the hydrated, polymer-depleted solid crystallizes before it can dissolve. C77·C41 When crystals of the stable form are present, whether left by extrusion at levels only TEM can detect or added as seeds, that crystallization starts at once, follows the stable polymorph and cuts supersaturation at every dose; the authors attribute this to "fast amorphous-to-crystalline transformation of the amorphous solid". C78·C38 Their conclusion is that matrix crystallization is the most detrimental crystallization route for an amorphous formulation and that residual crystallinity, "even trace levels", is "of high risk" to the performance of a BCL/PVPVA dispersion. C72·C79

The study leaves the mechanism of matrix crystallization, the role of the polymer leaving the solid, and the generality of the effect across drugs, polymers and loadings to further work; the authors' final sentence claims only "implications for dissolution performance outcomes of ASDs containing crystallinity". C79·C51