Why mesoporous silica never gives back all of its drug: ritonavir adsorption on SBA-15 from supersaturated solution and the release floor it sets

Background and motivation

Up to 90% of drug candidates in development are estimated to have low aqueous solubility or high hydrophobicity (refs 1, 2). Ordered mesoporous silica (OMS), such as MCM-41 and SBA-15, has become a popular enabling carrier for such drugs (refs 3–6). It has specific surface areas (SSA) of 500–1000 m²/g and regular pores 2–30 nm wide, and drug loaded into the pores stays non-crystalline, a state the authors describe as thermodynamically stable because loading lowers the Gibbs free energy of the system (refs 6–9). Under non-sink conditions, water displaces drug from the silica surface and the medium becomes supersaturated (refs 6, 10–14). Since supersaturation is thought to be a main reason why such formulations improve absorption, the extent and duration of supersaturation they produce is the property that matters.

The difficulty is that OMS formulations do not release all of their drug. In the group's earlier study (Dening and Taylor 2018, ref 6), ritonavir loaded into SBA-15 supersaturated both phosphate buffer and fasted state simulated intestinal fluid (FaSSIF), but the percentage released fell significantly as the dose, and with it the theoretical supersaturation ratio $S$, was increased. Complete release was never reached, not even when the medium was diluted towards sink conditions. Other groups had reported incomplete release from OMS under supersaturating conditions as well (refs 3, 12–18), and Mellaerts et al. (2008, ref 12) had also put it down to a dynamic adsorption equilibrium between drug on the silica and drug in solution, shifting towards adsorption at higher dose and higher $S$. Because the solution is supersaturated, such an equilibrium is metastable: crystallization would reach a lower energy.

Why should adsorption grow with supersaturation? The large surface area of OMS gives it a high surface free energy, and drug adsorbed on that surface sits in a more stable state than drug in solution. The shift of the equilibrium as $S$ increases is therefore "likely attributable" to a fall in the solute chemical potential as drug adsorbs (Qian and Bogner 2012, ref 7; Rengarajan et al. 2008, ref 8; refs 9, 10). The consequence is that adsorption is an alternative route to crystallization for lowering the chemical potential of a supersaturated drug. For a formulator this means that the carrier itself competes with the solution for the drug it has just released.

Earlier work stopped short of the conditions that matter. Several studies had characterized the physical state of drug loaded in OMS and the drug–silica interactions in the solid state (refs 19–22), but the thermodynamics of loaded OMS in solution had hardly been examined. McCarthy et al. (2018, ref 23) studied drug adsorption onto porous SBA-15 and nonporous Aerosil 200 silica during dissolution under sink conditions, and Bui and Choi (2009, ref 24) measured the adsorption of five drugs from solution by SBA-15. Both found incomplete desorption, which they read as strong, not fully reversible binding to the silica surface. However, both worked only below the equilibrium crystalline solubility of the drug. To the authors' knowledge, nobody had measured how much drug OMS adsorbs from a supersaturated solution, which is exactly the range in which these formulations are meant to operate.

The paper sets out to close that gap and tests two hypotheses (Section 1). The first is that incomplete release under supersaturating conditions comes from adsorption to the OMS material, and that this adsorption becomes more favourable as supersaturation increases. It is tested with an adsorption isotherm of ritonavir on SBA-15 that reaches into the supersaturated range, and with probes of the interaction that include Fourier transform infrared (FTIR) spectroscopy and X-ray photoelectron spectroscopy (XPS). The second hypothesis is that surfactant molecules compete for the silica surface, displace adsorbed ritonavir and so increase release; it is tested with a neutral surfactant, vitamin E polyethylene glycol succinate (Vit E TPGS), and an anionic one, sodium dodecyl sulfate (SDS). Ritonavir suits the question: it is non-ionized at pH 6.8, hydrophobic (log P 5.98) and a slow crystallizer, so supersaturated solutions survive long enough to be studied. The authors present the work as the first report of drug adsorption by mesoporous materials under supersaturating conditions.

Background concepts

The silica surface. The surface of silica carries three kinds of chemical group (Section 5.2, citing ref 66): siloxane bridges (Si–O–Si), which are hydrophobic; silanols (Si–OH), which are hydrophilic and are considered the main adsorption sites; and silicic acid groups, that is, ionized silanols. Not all silanols are equally reactive: about 20% have a pKa of 4.5 and the remaining 80% a pKa of 8.5 (ref 67), and the more reactive isolated silanols, single Si–OH groups not hydrogen-bonded to a neighbour, are considered responsible for strong and potentially irreversible binding of adsorbates. The point of zero charge of silica is about pH 2 (ref 41), so at pH 6.8 the surface is negatively charged. A drug can bind this surface in three main ways: by hydrogen bonding or ion exchange with silanols, or by non-specific hydrophobic interaction with siloxanes (ref 41). Ritonavir has pKa values of 1.8 and 2.6 (its two thiazole groups, ref 68) and is non-ionized at pH 6.8, so ion exchange is not expected.

Crystalline solubility, amorphous solubility and supersaturation. The crystalline solubility is the concentration in equilibrium with the crystal; for ritonavir in 50 mM phosphate buffer pH 6.8 at 37 °C it is 3.6 ± 0.6 μg/mL (ref 6). The amorphous solubility is the higher concentration at which the solution undergoes liquid–liquid phase separation (LLPS) and forms a drug-rich, water-saturated phase as droplets; for ritonavir in the same buffer it is 27.9 ± 0.3 μg/mL (ref 6). Between these two values the solution is supersaturated but still a single phase. The authors relate supersaturation to the chemical potential of the solute (Eq. 5):

Here $\mu$ is the chemical potential of the dissolved drug, $a$ its activity, $\gamma$ its activity coefficient and $C$ its concentration; the asterisk marks the same quantities at saturation, in equilibrium with the crystal; $R$ is the gas constant and $T$ the temperature. In plain terms: the supersaturation ratio is a measure of how far the drug's chemical potential in solution lies above that of the crystal, so a more supersaturated solution holds drug that is more eager to leave it, by crystallizing or, as this paper argues, by adsorbing. For dilute solutions the activity coefficients cancel and the ratio becomes a ratio of concentrations (Eq. 6):

In plain terms: $S$ is the concentration divided by the crystalline solubility $C^{*}$. A "theoretical $S$" is the value that would be reached if the whole dose dissolved. For example, a dose equal to the amorphous solubility, 28 μg/mL, divided by the crystalline solubility of 3.6 μg/mL gives a theoretical $S$ of 7.8, the starting value of the dilution experiment in Fig. 8.

The adsorption isotherm and batch depletion. An adsorption isotherm plots the amount adsorbed per gram of adsorbent at equilibrium against the concentration left in solution. The authors measured it by solution depletion, in which the drop in solution concentration gives the amount adsorbed (Eq. 1):

Here $Q_a$ is the quantity of ritonavir adsorbed at equilibrium (mg/g), $C_i$ the initial concentration added and $C_e$ the equilibrium concentration left in the supernatant (μg/mL), $V$ the solution volume (mL) and $m$ the mass of SBA-15. In plain terms: whatever disappeared from solution is assumed to be on the silica. With 40 mL of solution and 0.5 mg of silica, every 1 μg/mL taken out of solution corresponds to 80 mg of drug per gram of silica, so even small depletions translate into large adsorbed amounts. For solutions above the crystalline solubility, the authors note that the "equilibrium" reached is a metastable one that lasts only as long as no crystals form (Fig. 1, footnote 2).

Two meanings of "monolayer". The paper uses two different monolayer quantities. The theoretical monolayer surface coverage, 253 mg/g, was calculated in ref 6 by assuming that the whole SSA of SBA-15 is accessible and that ritonavir lies in the orientation that maximizes its contact with the surface. The BET monolayer capacity $Q_m$, 103 mg/g, is a fitted parameter: the amount in the first, most strongly held layer according to the BET isotherm (explained under Contribution 1). The authors consider it "feasible" that $Q_m$ reflects the number of specific binding sites, the silanols, occupied in the presence of co-adsorbed water (Section 5.1). Ritonavir uptake exceeded both values.

The probes used to identify the interaction. The ratio of the first to the third emission peak of pyrene, $I_1/I_3$, falls when pyrene sits in a less polar environment, such as a drug-rich phase (ref 27). Urea, a strong hydrogen-bond acceptor, should reduce adsorption that depends on hydrogen bonding (ref 37), and added salt, which "salts out" hydrophobic solutes, should increase adsorption that depends on hydrophobic forces (refs 38, 39). In FTIR, isolated silanols give a sharp O–H band at 3749 cm⁻¹ that moves into the broad 3300–3600 cm⁻¹ region when they donate a hydrogen bond. In XPS, in the authors' reasoning, a hydrogen bond lowers the O 1s binding energy of the donor oxygen (the silanol) and raises that of the acceptor oxygen (a ritonavir carbonyl).

Monomeric and micellar surfactant. Below its critical micelle concentration (CMC) a surfactant is present as single molecules (monomers); above it, micelles form that can solubilize drug. The cited CMCs in water are 212 μg/mL for Vit E TPGS and 2.4 mg/mL for SDS (refs 29, 30). Both were used at 28 μg/mL, well below their CMCs, and Vit E TPGS also at 500 μg/mL, above its CMC.

Methods in detail

The work falls into three contributions. The first is the adsorption isotherm itself, measured into the supersaturated range, and the choice of model to describe it. The second is a set of probes, in solution and in the solid state, meant to identify how ritonavir is held on the silica. The third connects the isotherm to release: the authors calculate how much drug is still on the silica during stepwise dilution and in surfactant media, and compare that amount with the BET monolayer capacity. The figure below shows how the pieces fit together.

Contribution 1: Measuring the isotherm into the supersaturated range and choosing a model

① Problem. The two earlier isotherm studies on SBA-15 (refs 23, 24) stayed below the crystalline solubility, but a supersaturating OMS formulation releases drug into concentrations several times higher. To test whether adsorption explains incomplete release, the adsorbed amount has to be known across the whole range that a release experiment passes through: from below 3.6 μg/mL up to the amorphous solubility of about 28 μg/mL. The practical problem is that supersaturated samples can crystallize, and crystals removed by centrifugation would be counted as adsorbed drug; the measurement has to be finished while the metastable state lasts.

② What they did. SBA-15 from Glantreo (pore diameter 7.1 nm, pore volume 0.80 cm³/g, SSA 586 m²/g) was used at 500 μg in 40 mL of 50 mM sodium phosphate buffer pH 6.8, stirred for 60 min at 37 °C and 300 rpm. A 10 mg/mL methanolic ritonavir stock was then added to give initial concentrations from 0.625 μg/mL (written as 0.6 in Section 3.2) to 30 μg/mL; the methanol content was 0.01–0.3%, which the authors state had a negligible effect on adsorption. After a further 30 min, a time that preliminary experiments had shown to be enough to reach equilibrium (data not shown), the samples were centrifuged at 21,100 × g for 15 min at 37 °C, and the supernatant was analysed by HPLC (UV detection at 240 nm). $Q_a$ followed from Eq. 1. Each of the 14 concentration levels was run in triplicate, although only the lowest and highest initial concentrations are printed; the rest are visible only in Fig. 1. The data were then fitted with three isotherm models. The first is the linear form of the Langmuir isotherm (Eq. 2):

Here $Q_m$ is the monolayer adsorption capacity (mg/g) and $K_L$ the Langmuir constant. In plain terms: the Langmuir model assumes a single layer of identical, independent sites, so uptake rises and then levels off at $Q_m$; if the model holds, $1/Q_a$ plotted against $1/C_e$ is a straight line. The second is the linear form of the Freundlich isotherm (Eq. 3):

Here $K_F$ and $n$ are the Freundlich constants. In plain terms: the Freundlich model is an empirical power law, a straight line on log–log axes; it allows for a heterogeneous surface and for multilayers but its constants have no direct physical meaning. The third, and the one the authors adopt, is the Brunauer–Emmett–Teller (BET) isotherm in the form adapted to adsorption from solution by Ebadi et al. (ref 28), with an infinite number of adsorbed layers (Eq. 4):

Here $Q_m$ is the capacity of the first layer (mg/g), $K_S$ the equilibrium constant of adsorption for the first layer and $K_L$ the equilibrium constant for the upper layers; in this equation $K_L$ has a different meaning from the Langmuir constant of Eq. 2. In plain terms: drug first binds directly to the surface, with a strength set by $K_S$, and further layers then stack on the first, with a weaker strength set by $K_L$. As $C_e$ approaches $1/K_L$, the term $(1-K_L{C}_e)$ approaches zero and the predicted uptake rises without limit, so the isotherm turns steeply upward near a saturation concentration, as a gas isotherm does near the saturation vapour pressure. The model assumes that the first layer is held most strongly, that adsorbed molecules do not interact laterally and that the surface is homogeneous (Sections 4.2 and 5.1). The paper prints $K_S$ and $K_L$ in "μg/mL⁻¹".Card check: for $K_L{C}_e$ and $K_S{C}_e$ to be dimensionless, the unit must be mL/μg, which also matches $1/K_L$ being given in μg/mL; the printed unit is a notation error (card note X1). In gas-phase BET, $1/K_L$ corresponds to the saturation vapour pressure; for adsorption from solution its meaning "has been debated" (ref 28). The authors propose a meaning in Section 5.1:

Here $C_0$ is the saturation concentration of the liquid drug phase. In plain terms: the authors suggest that the concentration at which the isotherm would shoot up is the concentration at which ritonavir would otherwise separate from water as its own liquid phase, that is, the amorphous solubility. They base this on Hansen et al. (1949, ref 57), who used the saturation solubility of poorly water-soluble liquids in the solution form of the BET equation. The nonlinear fit was done in GraphPad Prism 8.0.1.

③ Why it works. The measurement depends on the drug staying in solution: ritonavir is a known slow crystallizer (ref 60), it had earlier been shown not to crystallize during centrifugation (ref 25), and no crystallization was observed during the adsorption experiments. Depletion of solution can therefore be read as adsorption. The printed numbers are internally consistent.Card check: with $V/m$ = 80 mL/mg, the printed top value of 424 mg/g at $C_i$ 30 μg/mL gives $C_e$ = 24.7 μg/mL, which matches the "approximately 25 μg/mL" stated in Section 4.3 and the last point of Fig. 1. The choice of model follows from the data: uptake went above the 253 mg/g theoretical monolayer, so a model that caps uptake at one layer cannot fit. BET has a further advantage over Freundlich: its two main parameters, $Q_m$ and $1/K_L$, have physical meanings that can be compared with independent numbers, the release residue and the amorphous solubility.

④ How it differs from earlier work. Earlier isotherms of drugs on SBA-15 stayed below the crystalline solubility and were fitted with Langmuir or Freundlich models (refs 23, 24, 49, 50, 52–54). Here the isotherm runs into the supersaturated range, and the BET form for liquids, which had been used for adsorption from solution (refs 28, 32, 33), for a drug on OMS (ref 56) and for poorly water-soluble liquids up to their saturation concentration (ref 57), replaces the usual models. The authors call the Langmuir and Freundlich models "clearly not appropriate" for these data.

Contribution 2: Identifying how ritonavir is held on the silica

① Problem. Ritonavir is non-ionized at pH 6.8, so it can bind silica by hydrogen bonds to silanols or by hydrophobic contact with siloxane bridges. The authors "therefore hypothesized" a combination of both, and with a log P of 5.98 they expected the hydrophobic part to contribute (Section 5.2). The answer decides which additives could compete for the surface.

② What they did. Five probes were used. (a) Pyrene fluorescence (pyrene 0.5 μg/mL; excitation 332 nm; $I_1$ at 371–373 nm and $I_3$ at 382–384 nm; buffer at 37 °C) in four samples: buffer alone; SBA-15 at 12.5 μg/mL; ritonavir at 50 μg/mL without silica, which is above the LLPS concentration; and ritonavir at 30 μg/mL with SBA-15 at 12.5 μg/mL. In the last sample the equilibrium concentration is about 25 μg/mL, below the LLPS threshold, so the authors reason that no drug-rich droplets form in solution and any drop in $I_1/I_3$ must come from drug-rich regions inside the pores. (b) Urea: adsorption at an initial concentration of 27.5 μg/mL in buffer with and without 5 M urea. (c) Ionic strength: adsorption at initial concentrations of 2.5, 7 and 14 μg/mL in distilled water, 0.1 M NaCl and 0.5 M NaCl. (d) ATR-FTIR (Bruker Vertex 70; 4 cm⁻¹ resolution; dry-air purge) on crystalline ritonavir, physical mixtures of crystalline ritonavir with blank SBA-15 at 38% and 22% w/w, neat amorphous ritonavir, ritonavir-loaded SBA-15 at 38% and 22% w/w, and blank SBA-15. The higher load was prepared at 150% of the theoretical monolayer coverage so that drug–silica interactions would be detectable. (e) XPS (Kratos Axis Ultra DLD; monochromatic Al Kα; C–C component of C 1s set to 284.8 eV; sampling depth about 10 nm) on blank SBA-15, amorphous ritonavir and loaded SBA-15 at 22% w/w. The loaded solids were made by solvent incipient wetness impregnation as in ref 6: 25 mg ritonavir in 400 μL methanol was added dropwise to 75 mg SBA-15, then dried for 24 h at 40 °C and 48 h under vacuum at ambient temperature; drug content was checked by methanol extraction and HPLC, but the values are not printed.

③ Why it works. Each probe predicts opposite outcomes for the two candidate mechanisms: urea should cut adsorption if hydrogen bonds matter, salt should raise it if hydrophobic forces matter, and the FTIR and XPS shifts are specific to a silanol acting as hydrogen-bond donor and a carbonyl acting as acceptor. The design also checks the probes themselves: pyrene in the LLPS sample shows that the probe reports a drug-rich phase, and the other controls are discussed under the experimental analysis. Ritonavir has no absorption at 3749 cm⁻¹, so any change in that band belongs to the silica.

④ How it differs from earlier work. Earlier studies characterized drug–silica interactions in loaded solids (refs 19–22). This paper adds probes in solution, including one in the supersaturated range (urea at 27.5 μg/mL), and combines them with solid-state spectroscopy. Note by the AI writer (not discussed in the paper): FTIR and XPS were done on dry solids loaded from methanol, which the authors themselves describe as ritonavir adsorbed "from organic solvent", not on silica that had adsorbed ritonavir from water (card note X13).

Contribution 3: Linking the isotherm to release

① Problem. An isotherm describes drug adsorbing onto bare silica from solution, whereas release starts from drug-loaded silica. To claim that adsorption explains incomplete release, the amount still on the silica during release has to be put on the same mg/g scale as the isotherm and compared with it. A second question is whether the bound drug can be moved: if it is held in a competitive adsorption equilibrium, another molecule that binds the same sites should push it off.

② What they did. First, the solubility of ritonavir was measured in each surfactant medium: the crystalline solubility by shaking excess form II crystals for 48 h at 37 °C, and the amorphous solubility by adding a methanolic drug solution and at once pelleting the drug-rich phase by centrifugation at 21,100 × g. Second, a sequential-dilution test, whose protocol and release data come from ref 6: loaded SBA-15 was dosed at the amorphous solubility (28 μg/mL) into 50 mL buffer at 37 °C, 0.5 mL samples were taken every 15 min for 180 min, and pre-warmed buffer was added every 30 min so that the theoretical $S$ fell stepwise through 7.8, 3.9, 2.6, 1.9, 1.3 and 1.0, the last step equal to the crystalline solubility (n = 2, mean ± range). This paper calculated how much drug remained adsorbed at each step. Third, release with surfactants: loaded SBA-15 was added to 75 mL buffer containing SDS 28 μg/mL, Vit E TPGS 28 μg/mL or Vit E TPGS 500 μg/mL, at 37 °C and 300 rpm in a jacketed vessel, at a dose such that 100% release would equal the amorphous solubility in that medium (28 μg/mL, or 77 μg/mL with micellar Vit E TPGS). Samples were taken over 120 min (n = 3); the surfactant-free profile is from ref 6. The drug remaining adsorbed was calculated at the time of maximum release: 120 min for buffer and SDS, 30 min for both Vit E TPGS media. The authors frame the results with three equilibria. The first, from Section 1, is the competition between water and drug:

In plain terms: water and drug compete for the same surface; lowering the solution concentration, for example by dilution, pushes the balance to the right and releases drug, and raising the supersaturation pushes it to the left. The second, from Section 5.3, replaces water by a surfactant:

In plain terms: a surfactant that can occupy the same sites trades places with the drug and frees it. The third describes micellar media:

In plain terms: micelles act as a second reservoir that takes drug out of free solution and so, indirectly, draws more drug off the silica. The equation used to calculate the drug remaining adsorbed is not printed; in effect, drug that is not in solution is counted as still on the silica.Card check: every bar in Fig. 8 fits remaining adsorbed ≈ 288 mg/g × (1 − fraction released), which implies a drug load of about 288 mg per g silica; the drug load of the release formulation is not stated in the paper (card notes X6, X7).

③ Why it works. If adsorption were a fully reversible equilibrium following the isotherm, each dilution step would lower the solution concentration, and drug would come off the silica until the bound amount matched the isotherm at the new concentration. If part of the drug is held irreversibly, release should stop at a floor that dilution cannot lower. Comparing the floor with $Q_m$ tests the authors' idea that the irreversible part is the first layer on the silanols. The surfactant pair separates two effects that could both raise release. Both monomeric surfactants raise the crystalline solubility (Table A below), but only the neutral one is expected to reach a negatively charged surface, so a difference between them points to competition at the surface rather than to solubilization alone.

④ How it differs from earlier work. In ref 6 and ref 12 the adsorption equilibrium was proposed qualitatively, as an explanation for release curves. Here the residual drug on the silica is expressed in the same units as the isotherm and set against a fitted parameter of that isotherm. Surfactants serve here as competitors for the silica surface, not only as solubilizers.

Experimental analysis

Setup in brief. All experiments used one drug, ritonavir (ChemShuttle), and one SBA-15, at 37 °C in 50 mM phosphate buffer pH 6.8, except the salt series, which used distilled water and NaCl solutions. Isotherm points, probes, solubilities and surfactant release were run in triplicate; the sequential-dilution series has n = 2. Differences were tested with unpaired Student's t-tests at p < 0.05. The solubility data come first, because each release test was dosed at the amorphous solubility of its own medium.

Table A. Crystalline and amorphous solubility of ritonavir at 37 °C (from Table 1; mean ± SD, n = 3).

Monomeric SDS and Vit E TPGS raised the crystalline solubility about 2- and 3-fold but left the amorphous solubility unchanged; micellar Vit E TPGS raised the crystalline solubility about 4-fold and the amorphous solubility by a factor of 2.7, to 77 μg/mL (Section 4.1). Two consequences follow for the release tests. A dose at the amorphous solubility gives a lower theoretical $S$ in the monomeric surfactant media than in plain buffer, because the crystalline solubility is higher. And the micellar test had to be dosed almost three times higher, at 77 μg/mL.

Table B. Adsorption isotherm of ritonavir on SBA-15 (C_e and Q_a read from Fig. 1; percentage adsorbed read from Fig. 7 and matched to the points by Q_a; approximate). Fit statistics: Langmuir R² 0.64, Freundlich R² 0.85, BET R² 0.94 with Q_m 103 mg/g, K_S 1.169, K_L 0.031 and 1/K_L 32.7 μg/mL (Section 4.2, Figs. S1 and S2).

Main result 1: uptake rises with supersaturation and passes both monolayers. Most points lie above the crystalline solubility of 3.6 μg/mL, and no crystallization was seen. Table B shows three regimes: a steep rise at the lowest concentrations, a gentler and scattered rise between $C_e$ of about 8.5 and 19.6 μg/mL, and a second steep rise near the amorphous solubility, where the last three points (about 262, 349 and 423 mg/g) exceed the 253 mg/g theoretical monolayer. The authors write that the data "clearly demonstrated" adsorption increasing with supersaturation and going "well past" the monolayer. The BET curve describes this shape best.Card check: with the printed parameters, the BET equation returns 430–436 mg/g at the top point against 424 measured, and about 141 mg/g at $C_e$ 10 μg/mL, on the plotted curve. The abstract says the data reach "slightly above the amorphous solubility"; that holds for the added concentration, but the highest equilibrium concentration was about 24.7 μg/mL, below the measured 27.9 μg/mL (card note X8). And $1/K_L$, 32.7 μg/mL, is 17% above the measured amorphous solubility and is extrapolated from points that all lie below both values, so the authors' "good agreement" is an approximate correspondence from one fit (card note X9). The fraction of added drug that adsorbed fell from about 77% to 18% and then stayed at 13–18%; the authors "hypothesize" a change of mechanism at $Q_m$, from site binding to multilayers.Card check: the fraction is already about 18% at Q_a about 71–72 mg/g, below Q_m, so in Fig. 7 the transition lies between about 60 and 71 mg/g, not at 103 mg/g (card note X10). Finally, at the top point the adsorbed drug fills about 0.4 cm³/g, about half of the 0.8 cm³/g pore volume, taking the amorphous density as 1.14 (90% of the mean crystal density).Card check: the paper prints these densities in "cm³/g"; read as g/cm³, the estimate reproduces (0.37 cm³/g, 46% of the pore volume) (card note X2).

Table C. Probes of the ritonavir–SBA-15 interaction (Table 2, Figs. 2–4 and 6, Figs. S4 and S5; NaCl values read from Fig. 3, approximate).

Main result 2: the probes point to hydrogen bonds with silanols. Urea cut adsorption by about a third and salt changed nothing. From Table C the authors conclude that their data "collectively indicates" hydrogen bonding with silanols as the main interaction at all concentrations studied, and they found the absence of a hydrophobic contribution "unexpected" for so lipophilic a drug. Each probe has a weakness, however. The urea control, 259 mg/g at an initial 27.5 μg/mL, lies about 90 mg/g below the Fig. 1 point at the same initial concentration (about 349 mg/g), so values near the top of the isotherm vary between runs (card note X4). Ritonavir's solubility in 5 M urea was not measured; if urea raises it, the drug's activity falls and, by the authors' own argument, adsorption would fall without any competition for hydrogen bonds (card note X12). In water the uptake was higher than in buffer at the same $C_e$ (about 182 mg/g at 4.7 μg/mL in water against about 72–106 mg/g at 4.1–6.3 μg/mL in buffer), so the salt series cannot be pooled with the isotherm (card note X5). The pyrene drop was not significant and would also occur if pyrene entered a single adsorbed layer (card note X11).

Table D. Sequential dilution of ritonavir-loaded SBA-15 dosed at 28 μg/mL (read from Fig. 8, approximate; n = 2). Release percentages are from ref 6; the remaining adsorbed amounts were calculated in this paper.

Main result 3: release stops when about one BET monolayer is left. Release rose from about 40% to 60% by 75 min and then, in the authors' words, "did not change", although each dilution made conditions more favourable for release. The calculated drug on the silica fell from about 170 mg/g to 100–110 mg/g and stayed there, next to the 103 mg/g of $Q_m$. The authors call this an "excellent correlation" and say Fig. 8 "clearly demonstrates" that adsorption is partially irreversible, a property they "hypothesized" to come from strong hydrogen bonds to silanols. Note by the AI writer (not discussed in the paper): in the last step the theoretical maximum concentration equals the crystalline solubility, so the solution held less than 3.6 μg/mL, and at such concentrations the adsorption isotherm gives no more than about 72 mg/g (Table B). The residue of 100–110 mg/g therefore sits above what reversible adsorption from solution would leave, which fits the authors' word "irreversible". It also means the release floor and $Q_m$ come from different experiments, one desorption and one adsorption, and their agreement is a correspondence between one dilution series and one fit.

Table E. Release of ritonavir from loaded SBA-15 in surfactant media and the calculated drug remaining on the silica (Section 4.8, Figs. 9 and 10, Fig. S6; mean ± SD, n = 3). Section 5.3 gives the buffer and SDS residues as 145 ± 8 and 135 ± 5 mg/g.

Main result 4: a neutral surfactant frees bound drug, an anionic one does not. Monomeric Vit E TPGS raised maximum release from 50% to 61% and lowered the calculated drug on the silica from above to below $Q_m$, which the authors say "demonstrates" that a neutral surfactant of similar size, with hydrogen-bond acceptor groups and at a similar molar concentration, can displace adsorbed ritonavir. SDS changed neither release nor the residue; the authors consider displacement "unlikely" because the anionic surfactant is repelled by the negatively charged silica. Micellar Vit E TPGS gave the most release, which "may be attributed" to displacement plus micellar solubilization. In both Vit E TPGS media the drug crystallized between 30 and 45 min, so it is "not clear" whether release had finished at 30 min.Card check: the buffer, SDS and dilution residues all imply a drug load of about 270–292 mg per g silica, whereas the two Vit E TPGS residues imply about 231 mg/g. With the same load of about 288 mg/g, 61% release would leave about 112 mg/g, above Q_m, and 69% release about 89 mg/g. The basis of the Vit E TPGS calculation is not given, so the "below Q_m" placement for monomeric Vit E TPGS is unverified (card note X7).

Controls and what they rule out. Pyrene with SBA-15 alone gave the same ratio as buffer, which rules out the probe binding hydrophobically to bare silica. The physical mixtures kept all crystalline carbonyl bands and the 3749 cm⁻¹ silanol band, which rules out the spectral changes arising from mere contact between drug and silica. SDS, which also raises the crystalline solubility, left release and residue unchanged, so solubilization by monomers alone does not explain the Vit E TPGS effect. What the design does not rule out is that the bound amounts are artefacts of the calculation basis, since drug on the silica was never measured directly after release.

Discussion

Where the picture applies. The evidence covers one non-ionized, hydrophobic, slowly crystallizing drug on one SBA-15 with 7.1 nm pores, in phosphate buffer pH 6.8 at 37 °C. Slow crystallization is a precondition: supersaturated isotherm points exist only while no crystals form, and for a fast crystallizer adsorption and crystallization would compete for the same excess drug. The authors expect similar behaviour for other poorly water-soluble drugs, because LLPS occurs for many compounds ("we can expect"), and preliminary results with two chemically different compounds "suggest" a similar pattern, but those data are not shown.

What remains open. The thermodynamic argument, that higher supersaturation means higher activity and so a stronger drive to adsorb, is reasoned rather than measured: drug activity in the adsorption samples was not determined. The steep final rise is explained by confinement, by analogy with capillary condensation, with the pores expected to fill before LLPS in bulk ("we believe"), but the drug inside the pores was not examined during aqueous adsorption. The adsorption of Vit E TPGS on the silica, the solid formed after desupersaturation and the free drug or membrane flux in micellar media were not measured either.

What it suggests for formulation. The authors state that incomplete release has "significant implications" for OMS in oral delivery of drugs that rely on supersaturation, and that the gain in release from micelles "should be balanced" against their possible reduction of free drug and flux. Note by the AI writer (not discussed in the paper): if the residue is set by the binding sites of the silica, it scales with silica mass rather than with dose. For ritonavir on this SBA-15 it is about 0.1 g drug per g silica, so the lower the drug load per gram of silica, the larger the share that never leaves. The silica also caps the concentration reached: in buffer, release from a dose at the amorphous solubility levelled off at 14 μg/mL, half of it. Measuring a drug's isotherm on the intended carrier into the supersaturated range is a cheap early check, since here $Q_m$ matched the release floor. Additives that compete for silanols are a lever and anionic ones are not, but the gain was modest (50% to 61–69%) and both effective media crystallized within 45 min, so a crystallization inhibitor would be needed to keep it.

Limitations

Limitations stated by the authors. The BET model assumes a homogeneous surface, and the mixed silanol population of SBA-15 "is anticipated" to cause deviations at low activity, with deviations at high activity also common (Section 5.1). The pyrene evidence for multilayers was "not quite statistically significant" (p = 0.054). The urea experiment gave "indirect evidence" for hydrogen bonding from supersaturated solution. With Vit E TPGS, crystallization between 30 and 45 min means it is "not clear" whether release had finished. Micelles "may actually reduce" the free drug and hence flux. Isotherms of more poorly water-soluble drugs are needed to test whether the pattern, including $1/K_L$ equal to the amorphous solubility, is general, and desorption under different supersaturating conditions should be studied.

The AI writer's own judgement (1): the central match rests on calculated numbers. The card rates the overall evidence as moderate. The isotherm, the dilution plateau and the surfactant results fit together, and the isotherm values reproduce by mass balance and by the BET equation. But the 100–110 mg/g plateau is calculated from solution concentrations in one dilution series (n = 2, release data from ref 6), while $Q_m$ comes from an independent fit of a different experiment (card note X18). The drug load needed for that calculation is not stated: the formulation appears as 25 mg drug per 75 mg silica (333 mg/g silica), as 150% of the theoretical monolayer (about 380 mg/g silica) labelled 38% w/w, as 22% w/w, and as about 250 mg/g in Fig. 12, and the measured content is not printed (card note X6). The "below $Q_m$" result for monomeric Vit E TPGS depends on a lower implied load than all the other media (card note X7).

The AI writer's own judgement (2): the mechanism evidence is indirect and the scope is narrow. All spectroscopy was done on dried, methanol-loaded solids, not on silica holding drug in water; the urea effect may act through solubility; the salt series was run in water; and the pyrene result does not separate one layer from several (card notes X13, X12, X5, X11). The abstract extends the finding to OMS materials in general and to "a wide range of dissolution conditions", while the data cover one drug, one silica and one buffer (card note X17). There are also minor errors: unit misprints for $K_S$, $K_L$ and the densities, two slightly different residue values, a Si 2p shift of 0.5 eV that reads as 0.5–0.7 eV in Fig. S5, an undescribed amorphous reference, and a Fig. 12 panel titled "complete" release that shows drug left at $Q_m$ (card notes X1–X3, X14–X16). Taken together, the plateau–$Q_m$ match is a well-argued correspondence in one system, not a general law.

Conclusion

The paper measured, for the first time by the authors' account, how much drug a mesoporous silica takes up from supersaturated solution: SBA-15 adsorbed more ritonavir as supersaturation rose, up to 424 mg/g, beyond both the 253 mg/g theoretical monolayer and the 103 mg/g BET first-layer capacity, and urea, salt, FTIR and XPS were read together as hydrogen bonding to silanols. When release from loaded silica was pushed by dilution, it stopped near 60% with a calculated 100–110 mg/g still on the silica, close to $Q_m$; a neutral surfactant with hydrogen-bond acceptors lowered that residue and raised release, while an anionic one did not.

For formulation, the message is that the carrier is an active partner in the solution equilibrium. It competes with the solution for the drug, more strongly the more supersaturated the solution, and it keeps a site-bound fraction that dilution does not recover. Whether this floor equals $Q_m$ for other drugs and silicas, and whether it holds when measured directly on recovered silica, are the next tests.