Liquid drug-rich nanodroplets from dissolving ritonavir dispersions, and how additives control their size and crystallization

Background and motivation

Many drug candidates dissolve too poorly in intestinal fluid to be absorbed well; the paper quotes an estimate that about 40% of active substances in development fail because of low efficacy, often from poor bioavailability (refs 3–6). C1 One answer is to deliver the drug as very small particles. Nanosuspensions of nearly pure drug are already used for oral and parenteral delivery (refs 7–11), and they can be made top-down by wet milling, which tends to give crystals, or bottom-up by solvent shifting, in which water is added to an organic drug solution and the product can be crystalline or noncrystalline (refs 11–15).

Amorphous solid dispersions (ASDs), in which the drug is dispersed in a water-soluble polymer, offer a third route: when they dissolve, they too can generate colloidal drug species (refs 16–22). This was first reported in 1965, when Tachibana and Nakamura found colloidal aggregates in solutions of dissolved β-carotene/PVP dispersions. R22 Since then such colloids have been suggested to explain part of the bioavailability gain of ASDs (Frank et al., ref 18), and Frenkel et al. proposed that aggregates with hydrodynamic radii below 100 nm "may be responsible" for the enhanced bioavailability of non-nucleoside reverse transcriptase inhibitors (NNRTIs), by uptake through particle-recognizing microvilli cells in Peyer's patches (ref 23). C1·C2 Others suggested that the colloids help prevent crystallization (Friesen et al., ref 38). The authors note that there is "little direct evidence" for either supposition, and that little is known about how ASDs produce the colloids in the first place. C31

A colloid of noncrystalline drug has two enemies. The droplets can merge (coalesce) into larger ones, and, because they form only from highly supersaturated solutions, the drug can crystallize. Surfactants may prevent coalescence, but their effect on nucleation and crystal growth had not been widely explored; polymers were known to stabilize colloids and to act as nucleation and crystal-growth inhibitors. C3

The Taylor group had just described the phenomenon that links these threads. In earlier work, poorly water-soluble drugs supersaturated by a pH shift (Hsieh et al., ref 39) or by solvent shifting (Ilevbare and Taylor, ref 29) underwent liquid–liquid phase separation (LLPS): above a certain concentration, which is reproducible for a given medium and temperature, predictable, close to the "amorphous solubility" and independent of how supersaturation is generated, the solution splits into submicrometer drug-rich droplets and a drug-poor aqueous phase. C32·R18 Ritonavir, which is un-ionized at pH 6.8 (pKa 1.8 and 2.6) and very hydrophobic (log P 5.98), was known to form such colloids; its crystalline solubility at pH 6.8 and 37 °C is 1.3 µg/mL and its experimental amorphous solubility 19.8 µg/mL. M1 The group had also measured its crystal-nucleation induction time, about 120 min at about 20 µg/mL, and found the new cellulose ester CAAdP 0.33 to be a very effective nucleation inhibitor (Ilevbare et al., ref 26). R17·C28

The paper sets three goals: (1) to find how the aggregates form and what they are made of; (2) to see how additives change their initial size and size stability; and (3) to probe the crystallization of solutions that contain them. To widen the chemistry beyond commercial polymers, the authors included several newly synthesized cellulose ω-carboxyalkanoates and a carboxypullulan ether made by the Edgar group (refs 27 and 28). M12·M13·M14·M15·M16

Background concepts

Supersaturation and its two ceilings. A solution is supersaturated when it holds more dissolved drug than the crystalline solubility (here 1.3 µg/mL). M1 It can be pushed much higher, but not without limit. Once the concentration reaches the amorphous solubility, the solution is in equilibrium with the drug's own amorphous, liquid-like form, and adding more drug does not raise the dissolved concentration further: the excess appears as a separate drug-rich liquid phase. The concentration at which this happens in a turbidity experiment is the LLPS concentration, also called the C(L–L) boundary. For ritonavir in pH 6.8 medium at 37 °C it is 18.2 ± 0.35 µg/mL, close to the experimental amorphous solubility of 19.8 µg/mL. R10.a·M1 The paper calls the amorphous solubility "theoretical" when it is calculated, but does not print the calculated value.

The drug-rich phase is a liquid, not a solid. Above its glass transition temperature (Tg), an amorphous material is a supercooled liquid: viscous but mobile. Pure amorphous ritonavir has a Tg of 50 °C, so at 37 °C it is a glass. M1 Water lowers the Tg of amorphous materials (it acts as a plasticizer), so a ritonavir-rich phase that has taken up some water can have a Tg below 37 °C and behave as a liquid. This is why the paper measures the Tg of the isolated droplet phase.

How the droplets are detected. Three signals are used together. (i) Extinction at a wavelength where ritonavir does not absorb (280 nm, or 280–450 nm in the titrations): a clear solution gives almost none, and any second phase scatters light and raises it. (ii) Dynamic light scattering (DLS): the count rate (in kilocounts per second, kcps) rises when scattering particles appear, and the method also reports a mean size, here the Z-average, an intensity-weighted mean diameter. (iii) Pyrene, a fluorescent probe whose ratio of the first to the third emission peak (I1 at 373 nm, I3 at 383 nm) falls when its surroundings become less polar; pyrene can dissolve in a liquid drug-rich phase but cannot enter a crystal lattice. P3·P7 Scanning electron microscopy (SEM) of dried samples and differential scanning calorimetry (DSC) of the isolated phase complete the picture. P8·P9

Two ways to make droplets. Droplets can be made by dissolving an ASD, or by solvent shifting: pumping a small volume of concentrated ritonavir solution in methanol (4 mg/mL) into buffer. P4·P6 Solvent shifting lets the authors test an additive at a fixed ritonavir concentration "without the complication" of how fast an ASD dissolves. C23

Size stability versus phase stability. The paper separates two properties. Size stability means the droplets do not coalesce and grow; it is read from DLS size versus time. Phase stability means the droplets do not crystallize; it is read as the crystallization induction time, the time from making the supersaturated solution to the first rise in extinction at 280 nm caused by crystals. P6 A good additive for a long-lived colloid needs both.

Classical nucleation theory, in words. The paper uses this theory without printing its equations. It states that the rate at which crystal nuclei form in a phase increases with the absolute concentration of the solute in that phase and decreases with the interfacial tension between the nascent crystal and the surrounding solution. E1 This matters because, at equilibrium between droplets and the surrounding water, the supersaturation is the same in both phases, but the absolute ritonavir concentration is far higher in the droplets. C40

Methods in detail

The study has three parts: identify what forms when a ritonavir ASD overshoots the LLPS concentration; test whether polymers shift that concentration; and compare additives, first inside ASDs and then in a solvent-shift screen of fifteen additives. All solution work was done at 37 °C, mostly in 100 mM sodium phosphate buffer at pH 6.8 with stirring at 300 rpm. P4·P5·P6

Contribution 1: showing that the ASD colloid is an LLPS droplet phase

The problem. Colloids from dissolving ASDs had been seen many times, but it was not known whether they are crystals, amorphous particles or liquid droplets, or at what concentration they appear. C1

What they did. The ASDs were made by solvent evaporation: 50 mg ritonavir with polymer(s) to 500 mg total, dissolved in 100 mL methanol (or dichloromethane/methanol 1:1 for HPMC and HPMCAS), dried on a rotary evaporator at 65 °C and then under vacuum for 24 h. P1 The key experiment used an RTV/PVP 10:90 ASD (10 wt% ritonavir in PVP K29/32). After a 20 min blank, the ASD was added to 25 mL of buffer in portions of about 3 mg until the solution held about 20 µg/mL ritonavir, and after about 60 min it was diluted to 10 µg/mL. P4 The solution was pumped continuously (1 mL/min) through a DLS flow cell, read about every 5 min with backscatter detection at 173°, while a fiber-optic UV dip probe followed extinction. P3 The droplet phase was then characterized three ways. For DSC, an RTV/PVP ASD was dissolved in distilled water at 37 °C, the dispersed phase was spun down at 40,000 rpm (274356g) for 15 min, sealed in a pan with a pinhole lid and heated at 20 °C/min from −40 to 60 °C. P9 For fluorescence, pyrene was added at 0.5 µg/mL and excited at 334 nm. P7 For SEM, about 20 µL samples of dissolved RTV/PVP and RTV/PVP/Tween 80 ASDs in distilled water were dried on glass, sputter-coated with platinum and imaged at 5 kV. P8 Finally, droplets made from the ASD were compared by SEM with droplets made by solvent shifting in the presence of 180 µg/mL PVP (Figure S3). R11

Why it works. Crystals cannot redissolve while the solution is still supersaturated with respect to the crystal, but liquid droplets of the amorphous drug must redissolve once the concentration falls below the amorphous solubility. Diluting from 20 to 10 µg/mL, still about eight times the crystalline solubility, therefore separates the two possibilities. The DSC, pyrene and SEM results then test the liquid, drug-rich nature of the phase independently. C6·M1

Note by the AI writer (not discussed in the paper). The methods describe the pyrene run as 30 mg of RTV/PVP ASD dissolved in 50 mL of distilled water (a nominal 60 µg/mL ritonavir), while the results describe adding ASD to 100 mM phosphate buffer to an initial concentration of about 20 µg/mL; the medium and concentration of Figure 5 are therefore uncertain. X1

Contribution 2: do polymers shift the LLPS concentration?

The problem. If a polymer dissolved in the drug-rich phase, or raised the crystalline solubility, it would change the LLPS concentration; the earlier work had shown that an additive changes the LLPS concentration only if it increases the equilibrium solubility of the crystalline form. C21

What they did. The 4 mg/mL methanol solution of ritonavir was pumped into pH 6.8 medium at 37 °C, stirred at 300 rpm, containing 180 µg/mL of PVP, PAA, HPMC, HPMCAS or CAAdP 0.85, and the LLPS concentration was taken as the point where scattered light began to rise, from extinction between 280 and 450 nm. P2 The polymer level, 180 µg/mL, is what a 10:90 ASD releases when it dissolves to 20 µg/mL ritonavir. P2 The addition rate, volume and number of replicates are not printed. P2

Contribution 3: additives in the ASD and in a solvent-shift screen

The problem. Which additives keep the droplets small, which keep them from crystallizing, and do the two go together?

In the ASD. Seven ASDs were compared: RTV/PVP 10:90; RTV/PVP/additive 10:80:10 with HPMC (606 grade), HPMCAS (AS-MF), CAAdP 0.33, CAAdP 0.85 or Tween 80; and RTV/PVP/Tween 80/CAAdP 0.33 10:70:10:10, a ratio given only in the legend of Figure 8. PVP was kept in every ASD for its high water solubility, so that the ASDs, whose other additives are mostly moderately hydrophobic, dissolve quickly. P1·C4·X20 About 20 mg of ASD, an excess, was added to 50 mL of medium (100 mM sodium phosphate buffer pH 6.8, 100 mM NaCl or distilled water), and DLS size was followed for 2–3 h in triplicate. P5 The captions of Figures 7 and 8 do not state the medium; the RTV/PVP series in them is the same as the buffer series of Figure 9. X10

In the solvent shift. Ritonavir from the methanol stock was pumped into 50 mL buffer to 20 µg/mL, just above the LLPS concentration, with an additive already dissolved at 5 µg/mL (2 µg/mL for butyl-6-CO2HPull). P6 The additives were three synthetic polymers (PVP, PVPVA, PAA), the surfactants SDS and Tween 80, the commercial cellulose derivatives HPMC, HPMCAS, PQ-10 (a cationic hydroxyethyl cellulose), HM-HEC and HM-HPMC (hydrophobically modified cellulose ethers), the new pullulan ether butyl-6-CO2HPull, and four new cellulose ω-carboxyalkanoates: CAAdP 0.33 and CAAdP 0.85 (cellulose acetate propionate adipates), CAAdB 0.81 (a cellulose acetate butyrate adipate) and CA Sub 0.90 (a cellulose acetate suberate). M2·M3·M4·M5·M6·M7·M8·M9·M10·M11·M12·M13·M14·M15·M16 The number after each new ester is its degree of substitution (DS) with the carboxylic acid-bearing side chain, from 0.33 to 0.90; the propionate esters also carry 2.09 propionate and 0.04 acetate per glucose unit (Table 2). M13·M14·M15·M16 The text counts fourteen additives, but Table 4 lists fifteen, including Tween 80. X4 Each run was done in triplicate; DLS gave the droplet size, and the extinction at 280 nm, scanned every 30 s, gave the induction time. P6 Selected runs were repeated at 17 µg/mL, below the LLPS concentration, without additive and with SDS or Tween 80. R16

How it differs from earlier work. Earlier studies of the group measured how polymers inhibit nucleation and crystal growth in clear supersaturated solutions (refs 24–26). Here the same kind of additives are tested in solutions that already contain a drug-rich droplet phase, and two read-outs, droplet size and induction time, are tracked in the same run. P6·C3

Experimental analysis

Droplets switch on above the LLPS concentration and off below it

Figure 4 is the central experiment. During the 20 min blank and at 10 µg/mL, the DLS count rate stayed at about 150–220 kcps and the extinction at 280 nm at about 0.00. When the dissolved ASD brought the solution to 20 µg/mL, the count rate jumped to about 4700 kcps and the extinction to about 0.09; both declined over the next 20 min (to about 3550 kcps and 0.06). After dilution to 10 µg/mL the count rate fell back to about 190–260 kcps, and the extinction to about 0.002–0.004 after a first reading of about 0.013. R1 The text adds that the solution became cloudy and that the droplets had a median diameter of about 350 nm when they first appeared; below about 18 µg/mL the count rate was too low for a size reading. R1

The authors place the onset at about 18 µg/mL, "approximately the same" as the LLPS concentration (18.3 µg/mL in the presence of PVP) and the experimental amorphous solubility (19.8 µg/mL). C5·R10.b·M1 Because crystals would not dissolve at 10 µg/mL, well above the crystalline solubility of 1.3 µg/mL, they conclude that the results "strongly support" liquid-like, drug-rich droplets that form once the LLPS concentration is exceeded. C6

Note by the AI writer (not discussed in the paper). Figure 4 shows only the plateaus at 0, 10 and 20 µg/mL, so the experiment itself places the onset between 10 and 20 µg/mL; the value of 18.3 µg/mL quoted with it is the separately titrated LLPS concentration with PVP (Table 3). The agreement is consistent with the LLPS reading but does not pin the onset to 18 µg/mL. X8

What the droplets are made of

Table A. Evidence on the nature of the droplet phase, RTV/PVP 10:90 ASD, 37 °C (from Figures 5, 6, S1 and S3 and the text; pyrene values read from Figure 5, approximate). R2·R3·R4·R11·M1·C7·C8·C9·C10·C11·C34

Taken together, the authors conclude that the droplets from the dissolving ASD form at a concentration similar to the LLPS concentration and have properties similar to the droplets made by antisolvent addition ("indicate"), and that fluorescence and SEM "confirms" their initially noncrystalline character. C12·C34 The water content of the droplet phase, which the low Tg is taken to reflect, was not measured. C8

Polymers barely move the LLPS concentration

Table B. LLPS concentration of ritonavir with and without polymer, pH 6.8, 37 °C (from Table 3; n not stated). R10

Four of the five polymers left the LLPS concentration within 0.5 µg/mL of the polymer-free value, and the authors read this as "suggesting" that the polymers are not substantially incorporated into the ritonavir-rich phase, or at least do not change its thermodynamic activity. R10·C22 CAAdP 0.85 raised the value to 21.3 ± 1.22 µg/mL, about 17% higher, which the paper does not discuss. X7 The titrations used 180 µg/mL polymer, while the solvent-shift screen below used 5 µg/mL, so these values cannot be carried over to the screen. X9

Inside the ASD, the additive sets droplet size and lifetime

Table C. DLS mean droplet size (nm) after ASD dissolution at 37 °C, n = 3 (read from Figures 7 and 8, approximate; medium not stated in the captions). R5·R6·C14·C15·C17

The ASD with PVP alone gave droplets that grew slowly over the first 45 min and then irregularly, reaching about 2530 nm at 100 min and about 5590 nm at 120 min; the authors read the gradual growth as coalescence and the steep rise as crystallization, at about 90 min in the text and at about 81 min by the arrow in Figure 8. R5·R6·C14 With HPMC, the droplets were already above 1000 nm at the start and grew to about 2000 nm ("unable to stabilize the colloidal aggregates against coalescence"), yet no crystallization was detected in the 2 h run. R5·C15 The more hydrophobic, negatively charged cellulose polymers stabilized the droplets: HPMCAS starting near 350 nm and reaching about 650 nm, CAAdP 0.33 and CAAdP 0.85 staying at about 290–400 nm for the whole run. R5·C16

Tween 80 gave the smallest droplets, about 210–240 nm, and stopped them from growing, but the size jumped at about 30 min, and SEM of the RTV/PVP/Tween 80 ASD showed radiating needle crystals among the droplets at 30 min and dense needle clusters at 45 min (Figure S2): Tween 80 "prevented droplet growth via coalescence" but "promoted crystallization". R6·R7·C17 Adding CAAdP 0.33 to the Tween 80 ASD delayed the step to about 90 min. R6·C17 The crystals grown with Tween 80 were far more needle-like, with a length-to-width ratio of about 22.4 against about 3.9 with PVP alone, and the authors caution that DLS sizes after crystallization in the Tween dispersions are "likely to be unreliable", since DLS reports a spherical-equivalent diameter. R8·C18·C19

Droplets are smaller and grow more slowly in water

The same RTV/PVP ASD gave smaller droplets in distilled water than in 100 mM phosphate buffer (Figure 9). The first readings are about 240 nm in water and about 360 nm in buffer; in water the droplets stayed below about 400 nm for the first hour and grew slowly to about 1420 nm at 185 min, while in buffer they grew to about 5590 nm by 120 min. R9·R5 Droplet sizes in phosphate buffer and in NaCl of the same ionic strength were comparable (data not shown). R9·C20 The text gives initial sizes of about 237 nm (water) and 500 nm (buffer) and calls the buffer droplets "approximately 50% larger"; 500 against 237 nm is about a twofold difference, whereas the plotted first points differ by about 1.5-fold, so the size of the effect depends on which value is used. C20·X2

The solvent-shift screen: four classes of additive

Solvent shifting to 20 µg/mL gave a cloudy, slightly bluish solution, "characteristic of" a second scattering phase. R15·C24 Without additive, the droplets grew from about 250 nm to about 1200 nm by 175 min and about 1800 nm at 180 min, and the induction time was 119.3 ± 8.14 min. R13·R12.a Table D lists the size range before crystallization and the induction time for every additive, grouped as in Table 5 of the paper.

Table D. Solvent shift to 20 µg/mL ritonavir in 100 mM phosphate buffer pH 6.8, 37 °C, additive at 5 µg/mL unless stated; induction time mean ± SD of triplicates (from Tables 4 and 5; "promoter" marks the additives the paper flags as promoting crystallization, plus SDS). R12·R14

Group I. SDS, PVP, PAA and HPMC (the last in group III) did not stop the droplets from growing; their size curves resemble the no-additive control, and the gradual rise is "indicative of droplet coalescence". R13·C25 The abrupt size changes with PAA (about 100 min), PVP (about 130 min) and SDS (about 40 min) match their induction times, so PAA, PVP and SDS are ineffective crystallization inhibitors, and PAA and "particularly SDS" promoted crystallization. C26·R12.b–c·R12.f PVPVA behaved like PVP. C29

Group II. The hydrophilic positively and negatively charged polymers PQ-10 and butyl-6-CO2HPull held the droplets at about 180–460 nm for up to 3 h but cut the induction time to 34.5 and 28.0 min; HM-HEC and HM-HPMC behaved in the same way (53.3 and 59.5 min), and so did Tween 80 (23.6 min). R13·R12.e·R12.g·R12.j–l·C27·C29

Group III. HPMC let the droplets grow (to about 700 nm by 140–175 min) but lengthened the induction time to 134.0 min. R13·R12.h

Group IV. HPMCAS, CAAdP 0.85, CA Sub 0.90, CAAdP 0.33 and CAAdB 0.81 kept the droplets small (203–352 nm before crystallization) and lengthened the induction time to 146.0–242.8 min. R12.i·R12.m–p The authors describe CAAdP 0.33, the most hydrophobic cellulose polymer tested, as the most effective at inhibiting both crystallization and droplet growth, and note that all the more hydrophobic cellulose derivatives had some effect, the best raising induction times from about 2 to 4 h. C28·C44 CAAdB 0.81, which entered the paper only through Tables 4 and 5, was at least as effective (242.8 ± 7.66 min, 204–210 nm), and the authors state that it behaved like CAAdP 0.33. X17·C29

The authors summarize the pattern in three statements: the more hydrophilic uncharged additives were ineffective or less effective against crystal formation and/or coalescence, while the more hydrophobic negatively charged cellulose-based polymers inhibited both; with the exception of HPMC, every effective crystallization inhibitor was also an effective colloid stabilizer, but not the reverse; and, "within our limited data set", cellulose esters "tend to" inhibit crystallization more effectively than polysaccharide ethers. C30·C51·C46

Below the LLPS concentration, crystallization slows and surfactants lose their effect

Table E. Crystallization induction time of ritonavir just above and just below the LLPS concentration (20 µg/mL: Table 4, triplicates; 17 µg/mL: single values from the Discussion, no SD; the no-additive value is printed as ">280 min" in one paragraph and "∼284 min" in the next). R12.a·R12.f–g·R16·X18

Just below the LLPS concentration, where no droplets form, ritonavir took more than twice as long to crystallize, and neither SDS nor Tween 80 had "an appreciable impact". R16 Together with the SEM images, in which needles appear among the droplets, this is the evidence for the authors' view that crystallization starts in the droplets. C39·R4·R7

Discussion

Where ASD colloids come from, and what they cap

The authors' central interpretation is that the colloids seen when these rapidly dissolving ASDs dissolve "originate from LLPS", which they call "highly likely": the ASD supersaturates the medium, and once the concentration exceeds the LLPS concentration the excess drug separates as noncrystalline submicrometer droplets, provided crystallization does not happen first. C33 From this they draw a broader conclusion. There "appears" to be an upper concentration of free drug that ASD dissolution can reach; any drug dissolved beyond it forms droplets. This fits the earlier finding of Alonzo et al. that the diffusion rate of drug from dissolved ASDs never exceeded that from a solution of the pure amorphous drug, although the nominal ASD concentration was higher and colloids were present. "Thus, the LLPS concentration represents the upper limit of supersaturation that can be achieved using an amorphous solid dispersion formulation." C35·R19 The free drug concentration above the LLPS concentration was not measured in this paper; the upper-limit statement rests on the droplet onset and on the cited diffusion data. C35

How the additives act: the authors' chain

The authors' reading of how additives set droplet size and lifetime (from the Discussion). C33·C35·C38·C43

Coalescence. The authors observe that the uncharged polymers are "largely ineffective" against coalescence, whereas the charged additives, with the exception of SDS, are effective at the concentrations used, and conclude that "it is likely" that a major mechanism of stabilization against coalescence is charge repulsion between polymers adsorbed at the ritonavir/aqueous interface. C37·C38 The abstract states the observation as "in general". C37 They add that much smaller and more stable droplets were possible with charged additives, whereas ritonavir droplets without a charged additive coalesced rapidly. C56

Nucleation inside the droplets. From the SEM images and the slower crystallization at 17 µg/mL, the authors infer that "it appears" crystallization starts in the drug-rich phase rather than in the bulk solution. C39 Their argument uses classical nucleation theory: if droplets and the surrounding solution are in equilibrium, the supersaturation is the same in both, so the nucleation rate depends on the absolute concentration, which is much higher in the droplets, and on the interfacial tension between the nascent crystal and its surroundings, which "would be expected" to be lower against a water-saturated supercooled liquid than against a dilute aqueous solution. It is therefore "not unreasonable" that nucleation is favoured in the droplets. C40·E1 For ritonavir, droplet formation thus "appears to promote crystallization", contrary to the earlier suggestion that a colloidal phase stabilizes against crystallization (ref 38). C41

Promoters. SDS, Tween 80, HM-HEC, HM-HPMC, butyl-6-CO2HPull and PQ-10 all shortened the induction time relative to ritonavir alone. C42 The authors point to Davey and co-workers, who found amphiphile-directed nucleation in aqueous emulsions of an undercooled aromatic compound, with closely packing amphiphiles and structural mimics being the better promoters. R21 Because SDS and Tween 80 had no appreciable effect at 17 µg/mL, "it is apparent that the additives require the presence of the ritonavir−water interface in order to promote crystallization". C43 The discussion names the emulsion system "m-chlorobenzene", while the cited reference title gives m-chloronitrobenzene. X19

Structure and position of the polymers. The authors state that they do not have enough polymer structures for a general rationale of why some polysaccharides promote crystallization while others inhibit it. C45 They offer three tentative pictures. The uncharged cellulose ethers HM-HEC and HM-HPMC may not sit at the droplet surface but partly in the interior, which would leave them less able to prevent growth ("it is certainly possible"). C47 The butyl ether chains of the regioselectively substituted butyl-6-CO2HPull lie on one face of the chain and lack a carboxyl group, so they "may be" buried in the droplet interior; PQ-10, randomly substituted with more polar poly(ethylene oxide) side chains, "might interact" differently. C48 The cellulose ω-carboxyalkanoates, randomly substituted with ester-linked hydrophobic side chains that end in anionic carboxyl groups, are plausibly displayed on the droplet exterior ("It is not difficult to imagine"), which "could help to explain" their effectiveness against size growth. C49 Further studies, they write, are required to confirm whether and how these structural differences cause the differences in crystallization. C50

Note by the AI writer (not discussed in the paper). The surface-versus-interior picture for HM-HEC and HM-HPMC predicts poorer size control, but Table 5 classes both as effective size stabilizers and Table 4 gives them narrow size ranges before crystallization (292–395 nm). The explanation as written does not match the paper's own classification. X6

Droplet size and the gut

The ritonavir droplets (smallest about 200 nm) are larger than the aggregates reported for several other aggregate-forming compounds; the authors suggest this "may be attributed" to ritonavir's high hydrophobicity, large molecular size and/or un-ionized state at pH 6.8. C54·C55 Citing work on NNRTIs, whose aggregates grew with pH as these weak bases became less ionized, they note that the dependence of droplet size on drug and additive ionization, and thus on pH, "may have important implications" for absorption along the gastrointestinal tract. C57·R20 The larger droplets at higher ionic strength are "in line with expectations", since water-structuring salts enhance hydrophobic interactions. C58

What it suggests for formulation

The authors' own conclusions for formulation are that through proper selection of formulation components "it thus appears possible" to obtain about 250–350 nm ritonavir droplets from a dissolving ASD and to inhibit their coalescence and crystallization, that the composition of the amorphous formulation "will be critical" for the formation and longevity of the colloidal phase, and that spontaneous nanodroplet formation from ASDs "might be important" as a route to drug nanospecies for oral and other delivery. C61·C60·C52

Note by the AI writer (not discussed in the paper). Three practical points follow from the data. First, a surfactant added to an ASD for wetting can shorten the life of the droplet phase: here Tween 80 and SDS took the induction time from 119 min to 24–35 min, and in the ASD itself Tween 80 led to needles within 30 min. R12·R7 Second, an additive screen run below the LLPS concentration would have missed this, because the same surfactants had no effect at 17 µg/mL. R16 Third, size stabilization and crystallization inhibition must be tested separately: Tween 80 and butyl-6-CO2HPull gave the smallest droplets (178–208 nm) and the fastest crystallization (23.6 and 28.0 min), while HPMC gave large droplets and slower crystallization. R12·R14

Limitations

Limitations stated by the authors. They state that their data set has too few polymer structures to rationalize which polysaccharides promote and which inhibit crystallization, and that further studies are needed to confirm whether the proposed structural differences are responsible. C45·C50 They caution that DLS sizes after crystallization in the Tween 80 dispersions are "likely to be unreliable". C19 For the role of colloids in absorption and in preventing crystallization, they note that there is "little direct evidence". C31

The AI writer's own judgement (1): what is measured and what is inferred. The identification of the droplets as a noncrystalline, drug-rich liquid rests on four consistent methods and is the strongest part of the paper. R1·R2·R3·R4 The onset in the ASD test, however, is only bracketed between 10 and 20 µg/mL by the plotted data, and the 18.3 µg/mL quoted with it comes from a separate titration. X8 The mechanisms in the Discussion (charge repulsion of adsorbed polymer, polymer location on or inside the droplets, nucleation inside the droplets) were argued, not measured: no adsorption, surface charge, droplet composition or nucleation site was determined. C38·C47·C39 The cap on free drug at the LLPS concentration is likewise an inference from the droplet onset and cited diffusion data. C35

The AI writer's own judgement (2): how well the summaries fit the tables. The charge rule has exceptions in the paper's own Table 5: PAA, a carboxylic acid polymer whose charge is not discussed, is an ineffective size stabilizer, while the uncharged HM-HEC and HM-HPMC and the nonionic Tween 80 are effective ones. X5 CAAdP 0.85 raises the LLPS concentration by about 17% although the text calls all polymers similar. X7 HPMC is classed as an effective crystallization inhibitor on an induction time about 12% longer than without additive (134.0 ± 6.00 against 119.3 ± 8.14 min). X15 PAA is called a promoter in the text but is not flagged as one in Table 5. X16 CAAdB 0.81 is at least as effective as CAAdP 0.33, which the text names the most effective polymer. X17 Some size statements are looser than the table (initial sizes of "about 180–250 nm" against Table 4 lower bounds up to 331 nm; HPMCAS "below 350 nm" against 352 nm in Table 4). X14 The text counts fourteen screened additives where Table 4 lists fifteen. X4

The AI writer's own judgement (3): text and figures that disagree. The pyrene experiment is described with two different media and ritonavir concentrations. X1 The initial droplet sizes in water and buffer, and the "approximately 50% larger" ionic-strength effect, do not agree with each other or with the plotted first points. X2 The onset of RTV/PVP crystallization is placed anywhere from about 81 min (Figure 8 arrow) through about 90 min (text) to about 120 min (pyrene caption), depending on the method. X11 Sizes from DLS (Z-average, "median" and "mean" are all used) and from SEM of dried samples are not directly comparable: SEM gave about 500 nm in water where DLS gave about 240 nm. X12 The SEM time points differ from the stated sampling plan, and Table 2 prints a total degree of substitution of 2.84 for CAAdB 0.81 where its listed components sum to 2.94. X13·X3 None of these changes the main conclusions, but they limit how precisely the numbers can be quoted.

Scope. One drug, ritonavir, was studied at 37 °C with one additive concentration per experiment type, 5 µg/mL in the screen and 180 µg/mL in the titrations; the 17 µg/mL induction times behind the droplet-nucleation argument are single values without SD; and no permeation or absorption data were collected. X9·X18

Conclusion

The paper makes three contributions. First, it shows that the colloid formed when a ritonavir–PVP ASD dissolves is a noncrystalline, ritonavir-rich liquid that appears when the dissolved drug passes the LLPS concentration, close to the amorphous solubility, and disappears when the solution is diluted below it. C59·R1·R2 Second, it shows that common polymers at 180 µg/mL leave that concentration almost unchanged, which the authors read as the polymers staying out of the droplets. R10·C22 Third, it separates two properties of the droplet phase, size stability and resistance to crystallization, and shows that additives can help one, both or neither: hydrophobic, negatively charged cellulose esters such as CAAdP 0.33, CAAdB 0.81 and HPMCAS kept the droplets small and delayed crystallization, while Tween 80, SDS and several polysaccharide ethers accelerated crystallization, apparently by acting at the ritonavir–water interface. R12·R14·C30·C43 The authors' summary is that "the composition of the amorphous formulation will be critical in determining the formation and longevity of this colloidal phase". C60