How surfactants change the timing, shape and composition of water-induced phase separation in ritonavir–PVPVA amorphous solid dispersions

Background and motivation

Many new drug candidates have very low aqueous solubility, which makes adequate oral absorption hard to achieve (Williams et al., 2013). Amorphous solid dispersions (ASDs), in which amorphous drug and polymer are mixed at the molecular level, raise the transient solubility and the dissolution rate, and surfactants are often added to make ternary ASDs, mainly for processability and wetting (Sjökvist et al., 1991; Won et al., 2005). On dissolution, some ASDs release drug above its amorphous solubility, and the excess forms drug-rich nanodroplets by liquid–liquid phase separation (LLPS) or glass–liquid phase separation (Ilevbare and Taylor, 2013; Mosquera-Giraldo and Taylor, 2015). Nanodroplets have been linked to better absorption in vivo (Wilson et al., 2018; Stewart et al., 2017), but other ASDs release drug slowly and poorly (Yang et al., 2021; Indulkar et al., 2019).

Several studies relate this difference to amorphous–amorphous phase separation (AAPS), in which a miscible ASD splits into a drug-rich and a polymer-rich phase, both still amorphous. The evidence points two ways. Systems that resist macroscopic AAPS, in which a continuous drug-rich layer forms on hydration, were reported to release better (Indulkar et al., 2019; Saboo et al., 2019; Purohit and Taylor, 2017); for the model system of this paper, rapid AAPS in ritonavir–PVPVA ASDs enriched drug at the water interface and slowed release of both drug and polymer, especially at high drug loading (Indulkar et al., 2019). Others proposed that a discrete drug-rich phase inside the matrix helps dissolution (Li et al., 2020; Harmon et al., 2016). The authors conclude that the kinetics and morphology of AAPS are "seemingly critical" for release, while the precise link "remains unresolved". AAPS also matters in storage, since crystallization can follow once the drug-rich phase has lost its polymer (Rumondor et al., 2009; Purohit and Taylor, 2015).

Surfactants change both release and stability. Que et al. (2019) found that SDS raised the drug loading at which ledipasvir–PVPVA ASDs still released drug and polymer rapidly and congruently from 7.5% to 30%, and TPGS and poloxamers improved release of several ASDs (Saboo et al., 2021; Lang et al., 2016; Solanki et al., 2019). In storage, surfactants sped crystallization in some systems (Mosquera-Giraldo et al., 2014; Baghel et al., 2018) and improved stability in others (Solanki et al., 2019; Siriwannakij et al., 2021). Beyond wetting and solubilization, a newer explanation is microstructure: Saboo et al. (2021) explained the better release of TPGS-containing ASDs at high loading by discrete rather than bicontinuous domains after water-induced phase separation. Surfactants are also known to change phase separation in colloids, polymer solutions and polymer blends (Laradji et al., 1992; Zhao and Chen, 2007; Zhu and Li, 2016). The authors therefore ask whether surfactants act on release partly through their effect on AAPS.

The second problem is analytical. Differential scanning calorimetry (DSC) detects phase separation only above a certain domain size and can remix the sample during heating (Qian et al., 2010). Atomic force and electron microscopy show domains but not their chemistry. Fluorescence imaging assumes that a hydrophobic dye goes to the drug-rich phase, but Saboo and Taylor (2017) found that fluorescein, a charged dye, stained the drug-rich phase of miconazole–PVPVA ASDs, so cross-validation "is necessary". Estimating local composition from local glass transition temperatures ($T_g$) needs a model of $T_g$ against composition, "a low feasibility approach" for three components. Solution NMR has shown that some surfactants enter drug-rich nanodroplets (Ueda and Taylor, 2020; Ueda and Taylor, 2021), and such data "can be used to provide predictions" of where a surfactant will sit in a hydrated ASD, but the distribution of a surfactant in a phase-separated ASD had never been measured directly.

The paper studies how surfactants change water-induced AAPS in kinetics, morphology and phase composition. "The hypothesis under consideration" is that the physicochemical properties of a surfactant set where it goes in a phase-separated system, and that this distribution sets the phase morphology and the ripening kinetics. The model system is ritonavir (RTV) with polyvinylpyrrolidone/vinyl acetate copolymer (PVPVA, Kollidon VA 64), which phase-separates on hydration (Indulkar et al., 2019) although dry RTV and PVPVA are miscible according to their solubility parameters (Zhao et al., 2019). Four surfactants span a wide range of hydrophilicity, given as hydrophilic–lipophilic balance (HLB) values from the literature: sodium dodecyl sulfate (SDS, anionic, HLB 40.0), polyoxyethylene sorbitan monooleate (Tween 80, 15.0), sorbitan monolaurate (Span 20, 8.6) and sorbitan trioleate (Span 85, 1.8), the last three non-ionic. A second aim is to cross-validate dye-based confocal imaging against label-free methods.

Background concepts

The LLPS concentration as a partitioning probe. AAPS happens inside the solid or hydrated matrix; LLPS happens in solution, when dissolved drug exceeds a threshold, the LLPS concentration, and drug-rich nanodroplets appear. In practice the threshold is detected as the onset of light scattering. Ueda and Taylor (2020) reported that an additive that dissolves in the droplets lowers this threshold, because mixing dilutes the drug in the droplets and lowers its chemical potential there. A fall in the LLPS concentration therefore signals that an additive partitions into the drug-rich phase; an unchanged value signals little or no partitioning.

Surfactant descriptors and zeta potential. The HLB is an empirical polarity scale: the higher the value, the more hydrophilic the surfactant. The critical micelle concentration (CMC) is the level above which micelles form and can solubilize drug, which would confound a partitioning test. In 50 mM phosphate buffer at pH 6.8 the cited CMCs are 487 µg/ml for SDS and 14 µg/ml for Tween 80 (Correa Soto et al., 2022); the Span CMCs could not be measured because the Spans are poorly soluble in buffer. The zeta potential is a practical measure of a particle's surface charge: an adsorbed neutral layer shields the charge and pulls the value toward zero, while an adsorbed anion makes it more negative.

Ternary phase diagrams. Water-induced AAPS involves drug, polymer and water and is described on a triangular diagram (Fig. 11 is a schematic for RTV–PVPVA–water). The binodal line separates the miscible region from the two-phase region, and inside it lies the spinodal. Between the two lines the mixture is metastable and separates by nucleation and growth, which gives isolated droplets; inside the spinodal it is unstable and separates spontaneously by spinodal decomposition, which first gives an interconnected, bicontinuous structure. At equilibrium the two phases have compositions on the binodal joined by a tie line, and their amounts follow the lever rule (equation E1 of the card, printed in words in Section 5.3):

Here $X$ is the overall composition, $A$ and $B$ are the compositions of the two coexisting phases at the ends of the tie line, and $\overline{BX}$ and $\overline{AX}$ are the lengths of the tie-line segments. In plain terms: the closer the overall composition lies to one end of the tie line, the more of that phase forms. The authors use this for morphology: the phase present in the smaller amount "is likely" to form the discrete phase, and bicontinuous domains are expected when both phases are ample. A third line, the vitrification line, marks where the mixture becomes glassy and too immobile to separate. As a dry film absorbs water its composition moves toward the water corner, and where this hydration path crosses the vitrification line, relative to the binodal and spinodal, decides the route of separation. Ripening, or coarsening, is the later growth and merging of domains, which reduces the interface between phases.

The membrane analogy. Porous polymer membranes are made by letting a non-solvent such as water enter a polymer solution, so that polymer, solvent and non-solvent separate. The authors treat water-induced AAPS as analogous, with polymer, drug and water in those three roles, because water is a non-solvent for the drug. Membrane terms such as "lacey" (a continuous phase with large pores) and "sponge-like" (a finer porous structure) are used in the paper for ASD morphologies.

Imaging and spectroscopy. Two dyes are used together in confocal fluorescence microscopy: hydrophobic Nile red, expected to stain the drug-rich phase red, and hydrophilic Alexa Fluor 488, expected to stain the polymer-rich phase green. Two-photon excited ultraviolet fluorescence (TPE-UVF) is label-free and uses the natural fluorescence of RTV, so RTV-rich regions look bright. Optical photothermal infrared (O-PTIR) spectroscopy detects IR absorption indirectly: a tunable IR laser heats the spot where it is absorbed, and the change in thermal expansion or refractive index is read by a visible 532 nm probe laser. Because the visible beam sets the spatial resolution, O-PTIR gives IR spectra and maps with submicron resolution, without contact or labels.

Methods in detail

The work has four contributions: time-lapse imaging of AAPS in dye-labelled films at high humidity, with the dye assignment checked against label-free images; label-free O-PTIR spectra and maps that show the composition of each domain and where each surfactant sits; solution tests with RTV nanodroplets that give an independent reading of which phase each additive prefers; and an interpretive framework from a schematic phase diagram and membrane science.

Contribution 1: Watching AAPS happen in dye-labelled films, and checking what the colours mean

① Problem. To compare the kinetics and morphology of AAPS between formulations, one must follow the same film from its homogeneous to its separated state and know which phase each colour shows. Earlier fluorescence work relied on the hydrophobic-dye assumption, which failed for fluorescein in miconazole–PVPVA (Saboo and Taylor, 2017).

② What they did. ASDs were made by rotary evaporation: binary ASDs from 1 g of drug plus polymer in 10 ml of methanol, ternary ASDs from 100 parts of drug plus polymer and 5 or 10 parts of surfactant (1.05 or 1.10 g) in 10 ml of solvent, dichloromethane (DCM) for Span 85 and methanol for the rest. For imaging, 30 µl of a 10% w/v stock containing 0.1% w/w Nile red and 0.02% w/w Alexa Fluor 488 was spin-coated on a glass-bottom dish (300 rpm for 10 s, then 2000 rpm for 45 s). All films were heated at 100 °C for 5 min "to remove any compositional heterogeneities" and dried under vacuum overnight. A Nikon A1Rsi confocal microscope with an in situ humidity chamber held the film at 37 °C and high relative humidity (RH), generated by evaporating distilled water; the RH value is not printed. Nile red was excited at 561 nm and read at 570–620 nm (red, drug-rich), Alexa Fluor 488 read at 500–550 nm (green, polymer-rich).Card check: the Alexa Fluor 488 excitation is 407 nm in the Methods but 488 nm in the Results and the Fig. 2 caption; treated as unresolved (note X1). Images were taken every 15 s; Fig. 3 shows 0, 5, 10 and 30 min for six films, RTV–PVPVA 20-80 and 30-70 and 30-70-5 films with each surfactant (compositions in parts, drug–polymer–surfactant). Replicate films are not stated. To check the dyes, Fig. 2 compares a separated 30% drug loading (DL) film with a label-free TPE-UVF image of a 30% DL RTV–PVPVA dispersion from earlier work (Li et al., 2021); a second check came from O-PTIR (Fig. 3d compared with Fig. 9c).

③ Why it works. Two dyes of opposite polarity light up complementary regions, so each phase is seen in its own channel and the overlay tests consistency. Films that start transparent and homogeneous make the onset visible in time. A label-free image showing the same type of pattern rules out the possibility that the dyes create or invert the contrast in this drug–polymer pair; the authors conclude that the hydrophobic-dye assumption "is validated for this drug-polymer system".

④ How it differs from earlier work. Earlier ASD studies used single dyes or imaged structures after hydration without chemical identification. Here two dyes are used at once, the time course is recorded in situ, and the assignment is cross-checked by two label-free methods. The authors present this as "the foundation" for confocal studies of AAPS in RTV–PVPVA, and keep the caveat that other systems need their own cross-validation.

Contribution 2: Label-free O-PTIR spectra and maps to locate drug, polymer and surfactant

① Problem. Fluorescence shows two phases but not where the surfactant is, and the $T_g$ route to composition is impractical for three components. The authors state that "to date, there has been no direct measurements of the distribution of surfactants in phase separated ASD systems".

② What they did. Label-free films (100 µl on a 25 × 25 mm cover glass, 1000 rpm for 10 s then 3000 rpm for 30 s, same heating and drying) were stored at high RH over distilled water for "different time periods"; time, temperature and RH are not printed. A mIRage microscope used a 532 nm probe of about 5 mW and a quantum cascade IR laser covering 950–1800 and 2700–3000 cm⁻¹, with spot spectra at 1 cm⁻¹ spacing and maps at 50 or 100 nm pixels. Marker bands were: RTV 1712 cm⁻¹ (ester C=O), 1630 cm⁻¹ (aromatic C–C) and 1523 cm⁻¹ (N–H bend; marked 1530 cm⁻¹ in Fig. 9a and the maps); PVPVA 1737 cm⁻¹ (vinyl acetate C=O) and 1680 cm⁻¹ (pyrrolidone C=O); SDS 2852 cm⁻¹; Tween 80 2924 and 1352 cm⁻¹; Span 20 2855 cm⁻¹; Span 85 2852 cm⁻¹ (marked 2855 cm⁻¹ in Fig. 9e). Maps were band ratios: 1530/1680 cm⁻¹ for drug, 1680/1530 for polymer, 2852/1530 for SDS and 2927/1680 for Tween 80. Spot spectra (Fig. 9) come from 30-70-5 films; the SDS and Tween 80 maps (Fig. 10c–f) from 30-70-10 films; the Span 20 map caption gives no composition.Card check: text and figure band positions differ by 3–7 cm⁻¹ (1523/1530, 2852/2855, 2924/2927); band identities are unaffected (note X7).

③ Why it works. Each component has a band the others lack: PVPVA has nothing at 1352 cm⁻¹ or in the relevant C–H region, and SDS has a 2852 cm⁻¹ band absent from neat PVPVA, so a spot spectrum assigns its spot directly. For Tween 80 there is an internal check: neat RTV absorbs more at 2963 than at 2924 cm⁻¹, so a drug-rich spectrum in which 2924 cm⁻¹ is stronger signals added Tween 80. Note by the AI writer (not discussed in the paper): mapping a ratio of two bands from the same spot makes the maps less sensitive to local differences in film thickness or signal strength.

④ How it differs from earlier work. By the authors' account this is the first direct measurement of surfactant location in a phase-separated ASD. Unlike AFM-based IR used earlier for surfactant-containing ASDs (Saboo et al., 2021), it is contactless and non-destructive, and unlike $T_g$-based estimates it reads chemistry directly.

Contribution 3: Nanodroplet tests in buffer as an independent reading of phase preference

① Problem. Film data show where components end up, but not whether a surfactant sits on the surface of the drug-rich phase or inside it. Ueda and Taylor (2020, 2021) showed that nanodroplet measurements reveal which additives enter the drug-rich phase; the question is whether such tests agree with the films.

② What they did. All tests used 50 mM phosphate buffer, pH 6.8, at 37 °C with stirring at 300 rpm. For zeta potential, RTV nanodroplets were made by solvent shift of a 200 mg/ml methanolic stock to 200 µg/ml RTV, with PVPVA at 800 µg/ml or each surfactant at 10, 25, 50 and 100 µg/ml (Malvern NanoZS, n = 3). For the LLPS concentration, a 20 mg/ml methanolic RTV stock was pumped in at 20 µl/min and the onset of scattering at 430 nm recorded (n = 3), with PVPVA 800, SDS 100, and Tween 80, Span 20 or Span 85 at 10 µg/ml. SDS and Tween 80 were kept below their cited CMCs "to avoid solubilization by micelles"; for the Spans a low level of 10 µg/ml was chosen. Solution ¹H NMR at 800 MHz used RTV 200 µg/ml with SDS 100 or the other surfactants at 10 µg/ml, with and without PVPVA, and quantified dissolved surfactant from peak areas against an internal standard. For polymer depletion, RTV nanodroplets (200 µg/ml) were formed in buffer with 800 µg/ml PVPVA, with no surfactant or with Tween 80 50, Span 20 50, SDS 200 or Span 85 10 µg/ml; samples taken at 0–3 h were centrifuged at 21100 g for 30 min and the PVPVA left in the supernatant measured by size-exclusion HPLC (n not printed). DSC glass transitions of the ASD powders were also measured, but the values appear only in the unavailable Table S2.

③ Why it works. Each test reads a different place. Zeta potential reads the droplet surface. The LLPS concentration reads the droplet interior. NMR reads what stays dissolved: RTV inside the droplets is invisible because its mobility is too low, given the "relatively high $T_g$" of RTV, so a surfactant that moves into the droplets loses solution peak area, and its presence there can broaden the dissolved-RTV peaks by changing drug exchange between droplets and solution. Depletion reads whether PVPVA leaves solution with the droplets.

④ How it differs from earlier work. The NMR and LLPS methods come from Ueda and Taylor; the new step is to set them, with zeta potential and depletion, against direct measurements in separated films of the same system, turning a solution assay into a test of whether solution behaviour predicts film behaviour.

Contribution 4: A schematic phase diagram and membrane science to explain morphology

① Problem. The data show that surfactants change when and how the drug-rich phase appears, but not why, and the RTV–PVPVA–water phase diagram has not been measured.

② What they did. Fig. 11 sketches a binodal, a spinodal, a vitrification line, a tie line and four hydration paths (a to d) from different drug loadings. Path a crosses the vitrification line in the metastable region, so separation "could occur" by nucleation and growth of a drug-rich phase; paths b and c cross it in the unstable region, so separation "would occur" by spinodal decomposition followed by ripening; path d gives nucleation of a polymer-rich phase. The paths enter the two-phase region at different places, so the phase amounts, and by the lever rule the morphology, differ. The authors add kinetic results from membrane science: slower separation gives denser sponge-like structures, faster separation larger cavities, and slow precipitation with water from the vapour the most homogeneous microporous structure (Strathmann and Kock, 1977; Matsuyama et al., 1999); more polymer means more chain entanglement, higher viscosity and delayed demixing (Hołda et al., 2013); and surfactant additives can give stable, uniform, microporous membranes (Dubyaga et al., 1990).

③ Why it works. In both membranes and ASDs a non-solvent enters a mixture and drives separation, and the outcome depends on where the composition path meets the thermodynamic lines and how mobile the mixture is then. A surfactant can act on both: the phase diagram "will be likely altered", and plasticization "would also move" the vitrification line, so the same path could follow a different route.

④ How it differs from earlier work. Earlier ASD studies described morphology without tying it to routes on a phase diagram. The framework is qualitative, however: no line on Fig. 11 is measured or calculated, and the authors state that "the exact composition pathway and resultant morphology for polymer-drug-water systems have not been explored in depth to date". Combined with the location data, it gives the hedged mechanism below ("would be expected", "likely", "may").

Experimental analysis

Setup in brief. Two binary films (20% and 30% DL) and four ternary films at 30% DL with 5 parts of one surfactant were imaged by confocal time-lapse (Fig. 3); films with 10 parts of SDS or Tween 80 were used only for O-PTIR maps. Composition comes from separately prepared, label-free O-PTIR films (Figs. 9, 10). Solution tests used RTV nanodroplets at 200 µg/ml with additive levels that differ between tests (Figs. 4–8, Table 1). No dissolution or storage-stability test was performed.

Main result 1: the two dyes mark the right phases. In the separated 30% DL binary film (Fig. 2a–c) the Nile red channel shows a red network and the Alexa Fluor 488 channel green regions in the complementary areas. The label-free TPE-UVF image (Fig. 2d) shows bright and dark regions in the same kind of network, and O-PTIR later showed that the red round domains of the Tween 80 film give RTV spectra (Fig. 9c). Fluorescence and O-PTIR were applied to different films, so the check is one of pattern type, not a pixel-by-pixel overlay.

Table A. Time course of water-induced AAPS at high RH and 37 °C (Section 4.1, read from Fig. 3; red = drug-rich, green = polymer-rich), with the drug-rich phase geometry in the O-PTIR films (Section 4.6). Descriptions are qualitative; no domain sizes or onset times were quantified.

Main result 2: more drug, faster separation. The 20% DL film hardly separated within 30 min, while the 30% DL film was bicontinuous by 5 min and then coarsened. The authors attribute this to the higher polymer content at 20% DL, which means more chain entanglement, higher viscosity and delayed demixing, "as can be seen" from Fig. 3a and 3b; more coarsening of the mobile drug-rich phase at lower polymer content "may help to explain" larger pores at 30% DL and a more sponge-like structure at 20% DL. The words for these structures shift within the paper: the 30% DL film is "bicontinuous" in the Results but a continuous "lacey" drug-rich phase with polymer-rich pores in Section 5.3, and its O-PTIR drug map (Fig. 10a) shows a connected drug-rich network enclosing polygonal polymer-rich domains.Card check: classify the binary morphology from the images, not from the descriptors (note X12).

Main result 3: surfactants change when and how the drug-rich phase appears. At 30% DL the surfactants split into two groups (Table A). With SDS or Tween 80 the film still looked homogeneous at 5 min and by 30 min held discrete round drug-rich domains, larger with SDS. With Span 20 or Span 85 the film had separated by 5 min, like the surfactant-free film, and coarsened quickly into mixed discrete and continuous (Span 20) or continuous (Span 85) drug-rich regions. The authors call the morphology differences "Drastic" and conclude that SDS and Tween 80 "retarded" AAPS; ripening occurred in all surfactant films, least obvious with Tween 80 and most obvious with the Spans. Two qualifications apply. The authors allow that the SDS and Tween 80 films at 5 min either were homogeneous "or the domains had not ripened to sizes that were visible under the magnification used", and these two rows carry 10 µm scale bars against 50 µm for the others.Card check: read the kinetic claim as a delayed appearance of optically resolved domains (notes X2, X3). The difference in shape is large and clearly visible, so the morphological result is more secure than the kinetic one.

Table B. Solution tests with RTV nanodroplets in 50 mM phosphate buffer, pH 6.8, 37 °C. Zeta potentials read from Fig. 4 except the buffer value (n = 3); LLPS concentrations read from Fig. 5 (n = 3); NMR from Table 1; PVPVA (800 µg/ml added) read from Fig. 8 over the 0, 1, 2 and 3 h samples (n not printed). Values read from plots are approximate.

Main result 4: zeta potential separates surface adsorbers from the rest. RTV nanodroplets carried −10.8 mV in buffer. PVPVA and Tween 80 brought this close to zero, which the authors read as "consistent with" adsorption of a neutral layer that shields the surface charge; SDS made it more negative, to about −39.6 mV at 100 µg/ml, which "is caused by" adsorption of the anion. The Spans changed little at low concentration and became, in the authors' words, "slightly more negative" at higher concentration; since adsorbed Spans should have pulled the value toward zero, the authors write that this "indicates a lack of interaction" with the droplet surface, while partitioning into the droplets "is still a possibility".Card check: Span 85 at 100 µg/ml reaches about −21 mV, the same as SDS at 10 µg/ml, which the authors assign to adsorption, so the absence of Span surface interaction is not established by Fig. 4 (note X4).

Main result 5: Tween 80 and the Spans enter the drug-rich phase, PVPVA and SDS do not. The RTV LLPS concentration stayed at about 36 µg/ml with PVPVA and about 35 with SDS, but fell to about 26 with Tween 80, 13 with Span 20 and 15 with Span 85 (Fig. 5). The authors read this as "little-to-no distribution of PVPVA or SDS into RTV nanodroplets", while Tween 80 and the Spans "do partition into the drug-rich phase"; because the Spans lower the LLPS concentration, micellar solubilization is not the dominant behaviour. NMR agrees (Table 1). SDS changed nothing. Tween 80 peaks shifted slightly and lost area with nanodroplets present, explained as micelle formation plus partitioning of a small amount.Card check: Table 1 explains the Tween 80 shift at 10 µg/ml as micelle formation, although its cited CMC is 14 µg/ml; unresolved (note X5). Span 20 broadened the RTV peaks and lost more area than Tween 80 ("More significant partitioning"), and the Span 85 peaks almost vanished (Fig. 7), "indicating that a majority" of Span 85 partitioned.Card check: "a majority" in the text, "All" in Table 1; the quantification is in the unavailable Supporting Information (note X6). PVPVA did not alter RTV peaks or surfactant partitioning. Over 3 h the PVPVA left in the supernatant stayed between about 775 and 860 µg/ml in all media, so PVPVA does not mix with the RTV-rich phase, with or without surfactant (Fig. 8).

The surfactant levels differ between tests: Tween 80 was at 10 µg/ml for LLPS and NMR, 50 µg/ml for depletion and up to 100 µg/ml for zeta potential, so it was below its cited CMC in some tests and above it in others, and no level corresponds to the film composition.Card check: compare partitioning across surfactants only at matched concentrations (note X14). Note by the AI writer (not discussed in the paper): SDS showed no partitioning at 100 µg/ml, ten times the level of the others, so its classification is not weakened by the mismatch, but how strongly Tween 80 and the Spans partition at film-like ratios remains open.

Table C. Phase composition and surfactant location in separated films by O-PTIR (Section 4.6; spot spectra from Fig. 9, maps from Fig. 10). Storage times at high RH were not stated; the Tween 80 domain size is read from Fig. 10e and is approximate.

Main result 6: in the films, each surfactant sits where its hydrophilicity predicts. Every film gave a clean phase assignment: drug-rich spectra carried no PVPVA bands and polymer-rich spectra no RTV bands. SDS appeared only in the continuous, polymer-rich phase, and its map matched the polymer map. Tween 80 appeared in the round drug-rich domains and "appears not to be substantially present" in the polymer-rich phase. Span 20 was found in the drug-rich domains and in a mixed boundary zone, and Span 85 only in the continuous drug-rich phase. The authors call the SDS and Span results "expected" from hydrophilicity. The drug-rich phase was continuous only with Span 85 and discrete with SDS, Tween 80 and Span 20, matching the confocal grouping.

Comparison across methods. Tables B and C sort the additives the same way. PVPVA hardly enters RTV nanodroplets and is absent from drug-rich film spectra; Tween 80 and the Spans enter nanodroplets and are found in the drug-rich film phase; SDS does neither and is found only in the polymer-rich phase. The authors call this "compelling evidence" that surfactants prefer different phases, and attribute the weaker Tween 80 partitioning seen by NMR to the different water content of bulk solution and of a film's polymer-rich phase ("can be attributed to"). The absence of cross bands and of PVPVA depletion together "indicate" a large miscibility gap between RTV and PVPVA in the presence of water. The O-PTIR films, however, were prepared differently from the confocal films and stored for unstated times, and their Tween 80 domains (about 6–8 µm) are far larger than the very small domains at 30 min in Fig. 3d.Card check: do not match domain sizes or stages between confocal and O-PTIR films (note X9).

Controls and what they rule out. The 30% DL binary film is the reference for every surfactant film, so changes in onset and shape are attributed to the surfactant (note by the AI writer, not discussed in the paper: the Span 85 film was cast from DCM rather than methanol, so its solvent history differs). In solution, PVPVA acts as a comparator that adsorbs on the droplets (zeta near zero) without changing the LLPS concentration, and SDS as one that adsorbs strongly without partitioning, which shows that the surface and interior read-outs can be told apart. NMR with and without PVPVA rules out an effect of the polymer on surfactant partitioning, and the flat 3 h depletion series rules out slow uptake of polymer. The controls do not rule out that the kinetic difference reflects magnification, or that partitioning depends on surfactant concentration.

Discussion

Where the findings apply. The results come from one drug and one polymer, a single drug loading (30%) and a single surfactant level (5 parts) for the kinetics, and hydration by water vapour at 37 °C in thin films heated to 100 °C, not by liquid water acting on a tablet. Within these limits three things are shown well: the two-dye confocal method marks the correct phases in RTV–PVPVA, O-PTIR can place a surfactant in one phase or the other without labels, and nanodroplet tests in buffer predicted the film location of all four surfactants and of the polymer. Note by the AI writer (not discussed in the paper): SDS is both the most hydrophilic and the only charged surfactant, so charge and HLB cannot be separated here, and Tween 80 (HLB 15.0) still went to the drug-rich phase. In every case the nanodroplet tests, rather than the HLB number alone, predicted the film result.

What remains unresolved in the mechanism. For SDS and Tween 80 the key step is location at the drug–polymer interface. This is inferred from the zeta potential of nanodroplets in buffer, not measured in films, and O-PTIR placed SDS only in the polymer-rich phase. The same paragraph of Section 5.4 says that lower interfacial tension "allows for immediate phase separation via nucleation and growth", while the observation it explains is a delayed onset. For the Spans the key step is plasticization of the drug-rich phase, which rests on $T_g$ values found only in Table S2. Note by the AI writer (not discussed in the paper): O-PTIR found Span 20 "along the phase boundary", yet the mechanism for the Spans relies on the "absence of surfactant at the interface"; the boundary is a mixed zone a few microns wide rather than an adsorbed layer, so the statements need not conflict, but the paper does not reconcile them. The mechanism thus stays at the level the authors give it, concepts that "could be potentially" applied.

The missing link to release. No dissolution was measured. Because microstructure "is believed" to affect dissolution, the authors "expect very different dissolution performance between these ASD systems" and present the work as "a foundation". Note by the AI writer (not discussed in the paper): the cited work suggests a direction to test. Continuous drug-rich layers were linked to slow release of RTV–PVPVA (Indulkar et al., 2019), and discrete domains formed with TPGS to better release (Saboo et al., 2021). If those links hold here, the SDS and Tween 80 formulations would be expected to release better at 30% DL than the Span 85 formulation, and the effect of SDS on ledipasvir–PVPVA (Que et al., 2019) points the same way. This remains a hypothesis until release is measured.

Implications for formulation and further work. The authors expect the Tween 80 and SDS ASDs, with slower AAPS, "to show better physical stability against both phase separation and crystallization during storage", and they note that partitioning into drug-rich nanodroplets "can be used to provide predictions" of surfactant location. The AI writer's reading adds three untested points: the LLPS shift and the loss of NMR signal are quick screens that could flag surfactants that will load the drug-rich phase before films or tablets are made; a surfactant that plasticizes the drug-rich phase could raise its crystallization risk; and a film uniform at 5 min under high RH says little about months of storage. Further work suggested by the data includes sub-micron imaging of the early stages at one magnification, partitioning tests at matched and film-like concentrations, direct detection of surfactant at the interface in hydrated films, confocal and O-PTIR on the same film, replicate films with quantified onset times, and other loadings, surfactant levels, drugs and polymers.

Limitations

Limitations stated by the authors. On kinetics, they allow that "the domains had not ripened to sizes that were visible under the magnification used" in the SDS and Tween 80 films at 5 min. On the dyes, they limit the validation to "this drug-polymer system" and state that "cross validation with other techniques is necessary". On the phase diagram, "the exact composition pathway and resultant morphology for polymer-drug-water systems have not been explored in depth to date", and the effect of a surfactant on it is only "likely". On partitioning, the Span CMCs could not be determined, and the different extent of Tween 80 partitioning in NMR and O-PTIR is attributed to hydration. More broadly, "the precise link between AAPS and drug release remains unresolved", and the membrane literature they cite shows that additive effects depend on surfactant concentration, which was not varied here.

The AI writer's own judgement (1): kinetics and reporting. The claim that SDS and Tween 80 "retarded" AAPS rests on one time-lapse series per film, judged by eye at different magnifications, with no sub-micron method at early times, no stated replicates and no quantified onset times or domain sizes (card notes X2, X3, X16). The safest reading is a delayed appearance of optically resolved domains. The RH is not printed for either imaging method, and the O-PTIR storage times are not given (X9). The morphological difference is large and is the stronger result.

The AI writer's own judgement (2): location and partitioning. Surfactant location is the best-supported finding, resting on direct band assignment in films, two co-location maps, and the LLPS and NMR tests in solution. Three details weaken it slightly: the SDS and Tween 80 maps come from 10-part films while the spectra and time-lapse use 5-part films (X8); the solution tests used mismatched surfactant levels (X14); and the paper is internally inconsistent on Span 85 partitioning ("a majority" versus "All") and on Tween 80 micelles below the cited CMC (X5, X6), with the settling data in the unavailable Supporting Information.

The AI writer's own judgement (3): the mechanism. The mechanism is weakly supported and the authors hedge it accordingly. The interfacial location of SDS and Tween 80 is inferred, not measured, in films (X10); the Span zeta shifts are larger than "slightly" (X4); the "immediate" nucleation-and-growth statement conflicts with the delayed onset (X11); plasticization depends on $T_g$ values outside the main text (X13); and the phase diagram is a schematic, with the binary morphology described in different words in different sections (X12). Overall the card rates the paper "moderate": where each surfactant goes is answered convincingly, how fast and why only provisionally, and what this means for release not at all.

Conclusion

For ritonavir–PVPVA films at 30% drug loading, a few parts of surfactant changed how water-induced amorphous–amorphous phase separation unfolded. SDS and Tween 80 delayed the appearance of visible domains and gave discrete, round drug-rich domains that ripened slowly, while Span 20 and Span 85 separated as quickly as the surfactant-free film and coarsened into irregular or continuous drug-rich regions. Label-free O-PTIR placed SDS in the polymer-rich phase, Tween 80 and Span 85 in the drug-rich phase, and Span 20 in the drug-rich domains and their boundary zone, with drug and polymer showing very low mutual miscibility in the presence of water. Simple nanodroplet tests in buffer gave the same phase preferences, and the two-dye confocal method was confirmed for this system.

The authors explain the differences with membrane-science concepts, interfacial surfactants slowing separation and stabilizing round domains, Spans plasticizing the drug-rich phase and speeding coarsening, but this explanation is hedged and rests partly on solution data and unavailable $T_g$ values. No dissolution or storage test was done, so the expected effects on release and stability remain to be shown; in the authors' words, the work "provides a foundation to study the impact of surfactants and AAPS on ASD dissolution".