How far does a dissolving amorphous solid dispersion take the drug? Danazol with PVP, HPMC and HPMCAS

Background and motivation

An amorphous solid dispersion (ASD) raises the effective solubility of a poorly soluble drug because the amorphous form has a higher energy than the crystal, and it often raises bioavailability too. C1 The polymer in an ASD is there to stop crystallization, both on storage and during dissolution; when crystallization is held off during dissolution, the drug can build up a supersaturated solution under non-sink conditions, that is, when more drug is offered than could stay dissolved at equilibrium. C2

Earlier work by the Taylor group and others had shown that such highly supersaturated solutions do not only crystallize. Above a boundary concentration, the drug stops being miscible with water and the solution splits into a water-rich phase and a dispersed, drug-rich amorphous phase: liquid-liquid phase separation, LLPS. This boundary had been found close to the predicted amorphous solubility of the drug saturated with water (Ilevbare and Taylor 2013; Almeida e Sousa 2015). C3 LLPS has a kinetic side: when supersaturation is generated quickly the observed LLPS concentration "may" lie above the amorphous solubility, rapidly crystallizing drugs "may never" show LLPS, and a polymer that inhibits crystallization can open up the LLPS region, as shown for nifedipine (Raina 2015). C4·C5

The ASD adds a twist. A slowly crystallizing amorphous drug alone dissolves up to, but not above, its amorphous solubility, so on its own it "should not" undergo LLPS. C6 Yet ASDs of the same drugs have been seen to produce LLPS on dissolution, which implies that the ASD briefly overshoots the amorphous solubility; colloidal species formed when a high-polymer ritonavir-PVP ASD dissolved were explained in this way (Ilevbare 2013). C7 Why does it matter? Because "precipitation" covers two very different events. A drug-rich amorphous phase leaves the solution supersaturated relative to the crystal and "has been suggested" to act as a reservoir that replenishes absorbed drug (Raina 2014), whereas crystallization removes the supersaturation. C8 What drives crystallization and membrane transport is the free, molecularly dissolved drug concentration, not the total. C26

Two questions were open. Does LLPS on ASD dissolution happen only with certain polymers? And does the drug:polymer ratio change the tendency to undergo LLPS? C7 The authors took danazol, a drug whose supersaturation and precipitation they had already studied, with methods for detecting LLPS and crystallization in danazol-polymer solutions already developed (Jackson 2014; Jackson thesis 2015). M1·R11 They prepared ASDs with three common ASD polymers, polyvinylpyrrolidone (PVP K29/32), hydroxypropylmethyl cellulose (HPMC, 606 grade) and HPMC acetate succinate (HPMCAS, MF grade), at three polymer:drug ratios, 90:10, 70:30 and 50:50 by weight, that is, 10, 30 and 50% drug loading. M2·M3·M4·P1 The aim was to characterise the phase behaviour of the solutions formed on dissolution, measure the supersaturation reached, and compare both with the expected amorphous solubility and with simple antisolvent screening experiments. A1·A2·A3

Background concepts

Crystalline solubility, amorphous solubility and supersaturation. For danazol the paper uses a crystalline solubility of about 1 µg/mL, citing earlier work, and 0.9 µg/mL in the results; the flux measurement on crystalline danazol in Table 2 gives 1.0 ± 0.1 µg/mL. M1·R9.a The amorphous solubility, which the paper treats as the LLPS onset concentration, is 8 µg/mL; it was determined earlier by UV extinction, fluorescence spectroscopy, ultracentrifugation and flux measurements (Jackson thesis 2015). M1 Dissolved polymer up to 1 mg/mL does not change either value (cited). R11 Supersaturation is defined through the activity of the drug:

In plain terms: the thermodynamic activity $a$ of dissolved drug is estimated as the free drug concentration $C_i$ divided by the crystalline equilibrium solubility $C_S$, and this ratio is the supersaturation $S_i$ (Equation 2). A solution with $S_i$ below 1 cannot crystallize; above 1 it has a driving force to do so. E2 For danazol, the amorphous solubility therefore corresponds to a supersaturation of about 8 to 9, depending on whether 1.0 or 0.9 µg/mL is taken for $C_S$. M1·X15

Why the free drug concentration, and why UV is not enough. An in situ UV probe measures the apparent concentration: all drug in the light path, dissolved or not. Once submicron drug-rich droplets form, they both absorb and scatter UV light, so UV readings above the amorphous solubility overestimate the molecularly dissolved drug; the authors call them "not reliable" there (Van Eerdenbrugh 2011). Filtering the sample is no cure, because droplets may pass the filter and filtration may itself induce crystallization (Hsieh 2014). C12

The pyrene probe. Pyrene is an environment-sensitive fluorescent probe. The ratio I1/I3 of the intensities of the first and third peaks of its emission spectrum reports the polarity around the probe: about 1.9 in water, lower when pyrene sits in a hydrophobic environment such as a drug-rich phase (Kalyanasundaram and Thomas 1976). R3·C13 The paper gives the peak positions as 273 and 283 nm, which lie outside the 350-420 nm range that was recorded, so they cannot be the measured peaks; the AI writer reads them as misprints. X1

Nanoparticle tracking analysis. NTA follows individual scattering particles in a laser-illuminated volume and reports their size and number concentration; here it was used to see whether a submicron phase exists at the concentration maximum. P5

Flux across a membrane. In a side-by-side diffusion cell, dissolved drug crosses a membrane from a donor into a receiver compartment. The flux $J$ is the mass $M$ transported per unit time $t$ and per unit membrane area $S$ (Equation 1):

In plain terms: the flux depends on the diffusion coefficient $D$ of the solute, its thermodynamic activity $a$ in the donor, the thickness $h$ of the membrane and the activity coefficient ${\gamma }_m$ of the solute in the membrane. E1 Substituting Equation 2 into Equation 1 gives Equation 3:

In plain terms: with $D$, $h$ and ${\gamma }_m$ fixed by the setup, flux is directly proportional to the supersaturation of the free drug. The authors stress that flux follows the activity of the dissolved drug "rather than total concentration", so a flux measurement reads the free drug concentration even when UV cannot; the regenerated cellulose membrane has pores smaller than 2 nm. E3·C11·P4 This is why the paper expects that all ASDs which undergo LLPS "should" give the same flux. C27

Methods in detail

The paper applies three read-outs, previously used on supersaturated solutions made by antisolvent addition, to solutions made by dissolving ASDs, and compares the results with the earlier antisolvent data. A3·P7

Making the films

Danazol and polymer, 1 g in total, were dissolved in 10 mL of ethanol:dichloromethane (1:1 w/w); the solutions were checked by eye to be single-phase. A few drops were spin-coated onto a 22 mm × 22 mm glass cover slip (500 rpm for 10 s, then 1000 rpm for 30 s), and the step was repeated on the same slip to build up layers to the weight needed. P1 The drug amount was chosen so that complete dissolution would give 20-30 µg/mL, well above the 8 µg/mL expected LLPS onset, which makes the experiments non-sink for both the crystalline and the amorphous solubility. P1·C10 Polarized light microscopy found no crystals in any 90:10 or 70:30 film, but the 50:50 films contained crystalline domains over about 10% of their surface area. R1 The film masses are not printed.

Contribution 1: two in situ read-outs during dissolution

Problem. UV alone cannot tell whether a supersaturated solution has formed a drug-rich phase, and above the amorphous solubility it overestimates dissolved drug. C12

What they did. Each film was placed in 50 mL of 10 mM pH 6.8 phosphate buffer containing 2 µM pyrene, stirred at 300 rpm and held at 25 °C. A fiber optic UV dip probe (1 cm path) recorded spectra every 60 s, processed at 288 nm (the danazol maximum) and 350 nm (no absorption), against methanol standards of 1-100 µg/mL (linear, r² > 0.999). At chosen time points 2 mL was withdrawn for a pyrene emission spectrum (excitation 332 nm, emission 350-420 nm) and replaced with fresh buffer. P2 Two read-outs follow. The LLPS onset concentration is the UV concentration at the time point where I1/I3 drops sharply; crystallization shows as a fall in UV concentration with a concurrent rise in I1/I3, and the induction time is when it starts. P3 One example run per system is shown (Figures 2 and 3); the number of replicates behind the summary in Figure 4 is not stated. X11

Why it works. The two signals respond to different things: UV to the total drug in the light path, pyrene to the appearance of a hydrophobic environment. A drop in I1/I3 while UV is still rising marks a new phase; a rise in I1/I3 while UV falls marks its loss. C13·C14

Contribution 2: sizing the drug-rich phase

One mL samples of the 90:10 ASD solutions were taken at the maximum of the apparent concentration and analysed by NTA (NanoSight LM10, 532 nm laser, 25 °C, 30 s, fixed camera settings). P5 The authors take these samples as a snapshot of the size and amount of the drug-rich phase before crystallization begins. C24 Antisolvent solutions from the earlier study (15 µg/mL danazol added, with 100 µg/mL polymer pre-dissolved) are the comparison. R6·P7

Contribution 3: measuring the free drug by flux

Problem. If LLPS caps the free drug concentration, the ASDs that undergo LLPS should all reach the same free drug concentration, even if their apparent UV concentrations differ; this needs a measurement that sees only dissolved drug. C27

What they did. A side-by-side diffusion cell (Figure 1) had donor and receiver compartments of 200 mL each, filled with buffer and separated by a regenerated cellulose membrane (MWCO 6-8 kDa, pores below 2 nm) over a circular opening of 30 mm diameter, at 25 °C. The ASD sat in a biopsy cassette with 1 mm pores in the donor, and UV dip probes (2 cm path in the receiver, 1 cm in the donor) read both sides every 15 s. P4 The flux was taken from the slope of the receiver concentration once the donor concentration had peaked and before crystallization (Figure 6). A standard curve of flux against solutions of known supersaturation, spanning the crystalline and amorphous solubilities, converted flux into a free drug concentration (linear, r² > 0.997); each value is the mean of three runs. P4·R9

An example. For the 90:10 HPMC ASD (Figure 6), the donor UV concentration rose to about 9.4 µg/mL at about 120 min and then drifted down to about 8.3 µg/mL at 250 min and 6.4 µg/mL at 480 min, while the receiver concentration rose almost linearly to about 1.8 µg/mL; the flux window ran from about 123 to 250 min. The caption gives a slope of 0.00454 µg/mL/min (R² = 0.999) and a flux of 0.908 µg/min. R8

Note by the AI writer (not discussed in the paper). Two numbers in this method do not fit together. The 0.908 µg/min of the caption is the slope times the 200 mL receiver volume; divided by the 30 mm opening (706.9 mm²) it gives about 128 × 10⁻⁵ µg/(min·mm²), about 7 times the 17.5 × 10⁻⁵ µg/(min·mm²) listed in Table 2 for the same system. E1·X4 In addition, the methods say the probes recorded "for 60 minutes", whereas Figure 6 runs to 480 min. X6 Neither point changes the comparison between formulations, which uses one consistent flux scale, but the absolute flux values should not be reused. X4

Experimental analysis

Main result 1: most ASDs overshoot 8 µg/mL and the probe sees a new phase. The 90:10 dispersions of all three polymers dissolved quickly: the apparent concentration passed the crystalline solubility within minutes and then the 8 µg/mL amorphous solubility (Figure 2). R3 Table A collects the key points of the example runs.

Table A. Key points of the example dissolution runs, read from Figures 2 and 3 (approximate); I1/I3 starts at about 1.9-1.96 in every run. R2

In every 90:10 run the pyrene ratio fell from about 1.9 as the concentration rose, while the probe was in a "more hydrophobic average environment"; the authors read this as "consistent with" the probe partitioning into a newly formed disordered drug-rich phase. R2.a-c·C13 When the UV concentration later fell, which the authors read as crystallization starting ("indicating"), I1/I3 rose back, to about 1.78 for PVP and to about 1.72 and 1.65 at 1440 min for HPMC and HPMCAS. R2.a-c·C14 The text places these events at about 75 min for PVP and at 300 and 360 min for HPMC and HPMCAS in the example runs. R3 All 70:30 dispersions (not shown) and the 50:50 PVP dispersion behaved the same way. R4·C16 Note how high the HPMCAS 90:10 trace went by UV, about 16 µg/mL, twice the amorphous solubility; the authors warn that UV values above 8 µg/mL are not reliable once droplets are present. R2.c·C12

Figure 4 (right) summarises the LLPS onset concentrations (Table B). The text puts them at 6-10 µg/mL for all dispersions except 50:50 HPMCAS, "in good agreement" with the 8 µg/mL amorphous solubility and with the antisolvent solutions, with the caveat that the fluorescence time points were sampled by hand and are coarser than the UV data. R5·C21

How long the supersaturation lasts

Table B. Crystallization induction times and LLPS onset concentrations, read from Figure 4 (approximate; error-bar type and replicates not stated). Antisolvent values are cited from the earlier study (ref 30). R5

All PVP systems had the shortest induction times and the HPMCAS systems the longest, and the antisolvent solutions with the same polymer pre-dissolved ranked them the same way; the authors conclude that this simple screening predicts the relative stability of the ASDs to crystallization during dissolution "quite well". C19·C20 The 50:50 HPMC and HPMCAS dispersions did not desupersaturate within the experiments. R5.g·R5.k·C19 The duration of supersaturation therefore depends on the polymer and on the drug loading. C15 The rank order is the shared part: the PVP ASDs crystallized after about 56-70 min against about 111 min in the antisolvent solution, roughly half. X8

The drug-rich phase seen by NTA

Table C. NTA mean particle size and particle concentration (from Table 1). Antisolvent: 15 µg/mL danazol added, polymer 100 µg/mL (cited, ref 30). R6

The ASD solutions held submicron particles of about the same size as the antisolvent ones, 246-284 nm against 205-267 nm, and somewhat more of them, which the authors say "can be readily accounted for" by the larger amount of drug added, 20-30 against 15 µg/mL. R6·C22·C23 Single video frames (Figure 5) show discrete scatterers in both solutions, more in the ASD one. R7 The authors take this as "additional support" for their conjecture that these solutions undergo amorphous precipitation without losing the amorphous solubility advantage. C25 NTA was run on the 90:10 dispersions only, and by itself it does not distinguish an amorphous droplet from a small crystal. X13

Main result 2: the free drug stops at the amorphous solubility

Table D. Flux and free drug concentration, mean ± SD, n = 3 (from Table 2); antisolvent rows cited from ref 30. R9

Every ASD that showed LLPS gave a free drug concentration between 7.1 and 9.1 µg/mL, and so did the antisolvent solutions (7.2-8.2 µg/mL with polymer, 8.1 µg/mL without). R9 As supersaturation (Figure 7) the ASD values sit at about 7.8 to 9.8, near the line the authors draw for the amorphous solubility. R10 The authors conclude that all ASDs except 50:50 HPMCAS reached "the same measurable maximum supersaturation", equal "within experimental error" to the amorphous solubility, even though the donor UV readings and the amounts added were higher, and that what had been shown for antisolvent solutions "is confirmed here" for ASD dissolution. C28·C30 The contrast with UV is plain for 90:10 HPMCAS: about 16 µg/mL apparent, 8.2 µg/mL free. R2.c·R9.i

Note by the AI writer (not discussed in the paper). "The same" covers a spread: the PVP ASDs sit at 7.1-7.3 µg/mL and the HPMC ASDs at 8.7-9.1 µg/mL, with standard deviations of 1.0-1.6 µg/mL and no statistical test. X7 Two bars in Figure 7 do not follow Table 2: 90:10 HPMCAS is drawn at about 9.8 where 8.2/0.9 gives 9.1, and 50:50 HPMC at about 8.2 where 9.1/0.9 gives 10.1; the other eight bars match Table 2 divided by 0.9 µg/mL. X3 And the two columns of Table 2 are not in one ratio, as Equation 3 would require: free drug/flux is 1.25 for crystalline danazol but about 0.5 for every supersaturated row, as from a standard curve with an intercept. X5 Table D, not Figure 7, is the value to use.

Main result 3: at 50% loading the cellulose polymers slow release

At 50:50 the picture changes for HPMC and HPMCAS (Figure 3). Both dissolved much more slowly at first and showed no drop in apparent concentration. R2.e-f·C17 The 50:50 HPMC dispersion passed about 8 µg/mL only after about 270 min, kept rising to about 9.8 µg/mL at about 1200 min, and its I1/I3 fell from about 1.92 to about 1.585 after the threshold was crossed: LLPS came late. Its flux, 18.3 ± 2.3 × 10⁻⁵ µg/(min·mm²) or 9.1 ± 1.6 µg/mL, was the highest measured. R2.e·R4·R9.g The 50:50 HPMCAS dispersion levelled at about 6 µg/mL by about 600 min and never reached the amorphous solubility; its I1/I3 drifted only to about 1.82-1.89 and, at about 890 min, about 1.745, a small change the authors "probably" attribute to the probe meeting undissolved film. R2.f·R13·C18 Its flux gave 5.4 ± 0.8 µg/mL free drug, a supersaturation of about 5.9, "consistent with a lower overall supersaturation". R9.j·R10.j·C33 The 50:50 PVP dispersion, by contrast, dissolved as fast as the 90:10 one, peaked at about 9.3 µg/mL at about 62 min and gave 7.1 ± 1.1 µg/mL. R2.d·R9.d·X9

Discussion

How the authors connect the results. The argument runs as a chain that splits three ways.

In the authors' words, ASDs other than 50:50 HPMCAS dissolve to a concentration above the amorphous solubility, "but the increase in concentration is short-lived", because a drug-rich phase in metastable equilibrium with the water-rich phase forms rapidly. C31 The consequence they draw is a limit: an ASD may in principle offer more solubility than the amorphous drug alone, but "this solubility advantage cannot be realized" once LLPS sets in, and "hence when ASDs undergo LLPS, the maximum in membrane transport rate is achieved". C32 Their closing statement generalises this: the type of precipitate, amorphous or crystalline, "dictates" the extent of supersaturation, and the maximum achievable supersaturation upon ASD dissolution "is dictated by the LLPS concentration". C41·C38 The overall supersaturation profile then follows from "a complex interplay between dissolution rate, polymer type, drug loading, and the kinetics of crystallization". C42

The 50:50 HPMCAS case. This is the one dispersion that never reached LLPS, and the authors are explicit that they cannot say why. The aqueous solubility of a drug formulated as an ASD "cannot be determined or predicted", so they cannot separate a lowered solubility of danazol in this formulation from a much slower release rate or from crystallization running alongside dissolution. C34 One fact from the methods bears on the third option: the 50:50 films already contained crystalline domains over about 10% of their surface before dissolution, which the results section does not mention. R1·X10

What it means for formulation, as the authors state it. The results "clearly demonstrate the potential benefits" of designing an ASD so that LLPS occurs on dissolution, because that gives a maximally supersaturated solution and the maximum transport rate. C35 For danazol, quite high loadings still give LLPS: up to 30% with all three polymers. C36 At 50% loading the dispersions dissolve much more slowly and reach the LLPS threshold, if at all, only late, so they "might not be expected to perform comparably in terms of membrane transport in vivo". C37 And the polymer ranking for crystallization during ASD dissolution can be predicted with a simple antisolvent screen with the polymer pre-dissolved. C20

Note by the AI writer (not discussed in the paper). Three practical readings follow from the data, none of them stated by the authors. First, above the amorphous solubility a higher UV peak is not more free drug: 90:10 HPMCAS read about 16 µg/mL by UV but gave 8.2 µg/mL by flux, so formulations should be ranked there by a free-drug measurement such as flux. R2.c·R9.i Second, for danazol the polymer changed how long supersaturation lasted (about 1 h with PVP, 3 h with HPMC, 5 h with HPMCAS), not how high it went. R5·R9 Third, the HPMCAS result shows a trade-off at high loading: the polymer that held 90:10 and 70:30 solutions supersaturated longest gave the lowest supersaturation at 50:50. R5·R9.j The 50% statement of the authors also needs a qualifier, since the 50:50 PVP dispersion dissolved as fast as the 90:10 one and the 50:50 HPMC dispersion gave the highest flux of all. X9·R9.g

Limitations

Limits the authors state. Above the amorphous solubility, UV concentrations are "not reliable" because droplets absorb and scatter light. C12 The fluorescence time points were sampled by hand, so their time resolution is lower than that of UV; the onset concentrations are given "with this caveat in mind". C21 For 50:50 HPMCAS, "it is not possible to determine" whether solubility, release rate or concurrent crystallization explains the missing LLPS. C34

The AI writer's own judgement.

• Figure and table values disagree in places. Two Figure 7 bars and the amorphous-solubility line do not follow Table 2; the flux in Figure 6 does not reproduce the Table 2 unit; the Table 2 columns are not proportional as Equation 3 requires. X3·X4·X5
• The comparator is cited and partly unclear. All antisolvent values come from a PhD thesis (ref 30), and the pre-dissolved polymer concentration is 100 µg/mL in the table footnotes but 10 µg/mL in the Figure 4 legend. X16·X2
• Single runs and missing data. The dissolution profiles are one example each, the replicates behind Figure 4 are not stated, the 70:30 profiles are not shown, no flux was measured at 70:30, and NTA covers only 90:10. X11·R4·X13
• The overshoot is inferred. The concentration above 8 µg/mL that is said to precede LLPS was seen only by UV, the method the authors themselves call unreliable there; the free drug was measured only on the plateau. X12
• Partly crystalline 50:50 films. All 50:50 results start from films with crystalline domains over about 10% of their surface. X10
• Smaller points. The printed pyrene peak wavelengths lie outside the recorded spectrum, the flux recording time stated in the methods conflicts with Figure 6, and whether the UV data were corrected for the 2 mL aliquots replaced with buffer is not stated. X1·X6·X14
• Scope. One drug, one buffer without bile salts or surfactants, 25 °C and spin-coated films; transport was measured across a dialysis-type membrane, not in vivo. P2·P4

Conclusion

The paper shows, for danazol, that dissolving an ASD under non-sink conditions usually carries the solution past the amorphous solubility into liquid-liquid phase separation, seen as a fall in the pyrene I1/I3 ratio and as submicron particles by NTA, and that the free drug concentration measured by flux then sits at about the 8 µg/mL amorphous solubility for every ASD that showed LLPS. R2·R6·R9·C29·C28 The polymer set the duration of supersaturation, PVP shortest and HPMCAS longest, in the same order as an antisolvent screen, and drug loading mattered at 50%, where HPMC and HPMCAS released slowly and HPMCAS never reached LLPS. R5·C20·C17·C33 The authors' central claim, that the maximum achievable supersaturation from an ASD "is dictated by the LLPS concentration", rests on the flux data; the transient overshoot that precedes LLPS and any in vivo consequence remain to be measured. C38·X12·C37