Drug release from mesoporous silica: adsorption, supersaturation and the pull of an absorptive sink, studied with atazanavir in SBA-15

Background and motivation

Many drug candidates dissolve too poorly in gastrointestinal fluid to be absorbed well, and amorphous solids are one of the standard remedies (Williams 2013, ref 1). Because their free energy is higher than that of the crystal, they can produce a supersaturated solution, a concentration above the crystalline solubility, which raises the driving force for transport across the intestinal membrane (Taylor 2016, ref 2). The most common commercial way to exploit this is the amorphous solid dispersion (ASD), in which a hydrophilic polymer promotes release and delays crystallization in solution. Mesoporous silica-based drug delivery systems (MPS) are an alternative route to an amorphous drug product, and the authors note that more than 6000 articles on them appeared in two decades (Vallet-Regí 2018, ref 8).

In an MPS the drug is held inside an ordered silica such as MCM-41 or SBA-15. These materials have a periodic hexagonal arrangement of cylindrical pores of 2–50 nm and specific surface areas up to 1500 m²/g, covered with silanol groups that can adsorb drug molecules (Zhao 2000, ref 9). The drug is loaded by solvent impregnation, incipient wetness, melt impregnation or supercritical CO₂ impregnation (Mellaerts 2008, ref 14). Inside the pores the drug tends to stay amorphous: below a critical pore diameter the surface energy makes crystallization unfavourable (Prasad 1994, ref 15), and drug immobilized on the pore wall may crystallize more slowly (Rengarajan 2008, ref 16). A simple estimate of how much drug the surface can carry assumes a monolayer and uses the surface area, the drug's molecular weight and the area one molecule covers (Azaïs 2006, ref 23); it considers geometry only.

Release from silica usually starts with a burst, but the final extent varies. Release rate depends on surface area, pore size, drug–silica interaction strength, how well water penetrates the pores to displace adsorbed drug, solubility and diffusion (Wang 2009, ref 12), and higher drug loading "may slow down" water penetration because the pores become more hydrophobic (Smirnova 2003, ref 37). Complete release was reported mostly under sink conditions with respect to the crystalline solubility. Under supersaturating or non-sink conditions release often stopped early: itraconazole plateaued at 60% (Mellaerts 2008, ref 14), indomethacin from MCM-41 and Syloid 244 FP EU at 40–80% depending on preparation (Limnell 2011, ref 43), and ritonavir near neutral pH at 50% (Dening 2018, ref 44). In the ritonavir case some drug stayed adsorbed on the silica even with bulk aqueous phase present, and the formulation gave a much lower mass flow across a cellulose membrane than a maximally supersaturated solution. Incomplete release matters because the extent of supersaturation sets the rate of mass transport.

Incomplete release is typically attributed to drug–silica interactions that leave drug adsorbed on the surface, and the authors' group had used an adsorption isotherm, of type II, to explain incomplete ritonavir release from SBA-15 (Dening 2019, ref 51). The paper then raises a point that closed-vessel testing ignores. In the body, absorption keeps removing dissolved drug, and a lower luminal concentration "may alter" the amount adsorbed, following the isotherm and le Chatelier's principle. A missing in vitro–in vivo correlation for flubendazole silica formulations suggested to the authors that membrane absorption "may have an impact" on release (Vialpando 2016, ref 47). Yet MPS are generally tested in closed compartments without an absorptive sink. The group had built a flow-through apparatus with a hollow-fibre membrane that measures release and absorption together, in which absorption promoted dissolution of slowly releasing formulations and limited crystallization in supersaturated solutions (Hate 2017, ref 48; Hate 2018, ref 49; Hate 2019, ref 45).

The paper states two goals: to evaluate how strongly atazanavir (ATZ) adsorbs from aqueous solution onto SBA-15, and to determine whether an absorptive compartment during dissolution testing promotes release. The second goal is framed as a test of the hypothesis that "better overall drug release would be observed with continuous drug removal". ATZ is a poorly water-soluble weak base (pKa 4.49, log P 4.5) with a crystalline solubility of 5 µg/mL and an amorphous solubility of 78 µg/mL, both taken from the group's earlier work (ref 49). Its 10% HPMC dispersion forms a drug-rich liquid phase on dissolution. The authors loaded ATZ into SBA-15 (7.1 nm pores, 586 m²/g) at five drug loadings (DL), characterized the solid state, and measured release with and without the membrane sink, with drug already dissolved in the medium, after a start at gastric pH, and against an HPMC ASD of the same drug.

Background concepts

Silanols and nanoconfinement. The silica surface is covered mainly with silanol groups (Si–OH). Because a silanol is both a hydrogen-bond donor and acceptor, hydrogen bonding is usually the main way guest molecules bind, and isolated (non-hydrogen-bonded) silanols are the most reactive adsorption sites. In infrared spectra an isolated silanol gives a small, sharp band at 3749 cm⁻¹; its loss on loading suggests that something is now bonded to it. ATZ has several nitrogen lone pairs and carbonyl groups that can accept hydrogen bonds (Fig. 1), so a shift of its carbonyl stretching bands (around 1700 cm⁻¹) is the other half of the evidence. "Nanoconfinement" means drug held in pores only a few nanometres wide; confinement is known to alter thermal events such as the glass transition and melting, or to remove them, whereas drug that overflows the pores behaves like the bulk amorphous material. The silica surface charge depends on pH: its isoelectric point is about pH 2, so below that it carries a small positive charge (ref 85).

Two solubilities and the ceiling on supersaturation. The crystalline solubility (5 µg/mL for ATZ) is the true equilibrium. The amorphous solubility (78 µg/mL) is the highest concentration the amorphous form can reach and marks the maximum supersaturation. Beyond it, the solution separates into a drug-lean aqueous phase and drug-rich nanodroplets, which is called liquid–liquid phase separation (LLPS); the nanodroplets can act as a reservoir that replenishes drug as it is absorbed (Indulkar 2016, ref 86). As a weak base, ATZ is far more soluble in acid. The release tests used a dose concentration of 75 µg/mL, which the authors equate with the amorphous solubility (citing the same ref 49 that elsewhere gives 78 ± 5 µg/mL). This dose is non-sink for the crystal but at-sink for the amorphous drug, so in principle the whole dose could dissolve "as long as crystallization does not occur".

Adsorption from solution and the isotherm. An adsorption isotherm plots how much drug the solid holds per gram, $Q_a$, against the concentration left in solution at equilibrium, $C_e$. The paper measures it by solution depletion: drug is added to a silica suspension, and whatever disappears from solution is counted as adsorbed. Written out from the authors' description (the card's E3):

In plain terms: the drop in concentration, times the liquid volume $V$, is the mass of drug taken up, and dividing by the mass of silica $m_{\text{SBA-15}}$ gives the amount adsorbed per gram. A type II isotherm rises, flattens, then rises again at higher concentration; the flat part is read as a completed first layer and the second rise as multilayer adsorption. The Brunauer–Emmett–Teller (BET) equation describes this shape. The paper uses its solution form (Eq. 1):

Here $Q_m$ is the monolayer coverage (mg/g), $K_{\mathrm{B}\mathrm{E}\mathrm{T}}$ is the equilibrium constant of adsorption and desorption, and $X$ is the relative concentration, the equilibrium concentration $C_e$ divided by the saturation concentration $C_s$, which the authors take as the amorphous solubility. In plain terms: adsorption grows slowly at low relative concentration and climbs without limit as the solution approaches saturation, and the single fitted number $Q_m$ tells how much drug forms the first, most strongly bound layer. The model assumes that the first layer binds most strongly, that adsorbed molecules do not interact sideways, and that the surface is homogeneous. The authors acknowledge that few systems meet these assumptions. The geometric ceiling for comparison comes from the monolayer estimate of Azaïs et al., stated in words on p.5:

Here SSA is the specific surface area of the silica (586 m²/g), $M_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$ the molecular weight of the drug, $A_{\mathrm{m}\mathrm{o}\mathrm{l}\mathrm{e}\mathrm{c}\mathrm{u}\mathrm{l}\mathrm{e}}$ the area one molecule covers and $N_A$ Avogadro's number. In plain terms: divide the total surface by the footprint of one molecule to get how many molecules fit side by side, then convert that number to a mass. The paper does not print the molecular weight of ATZ; with the printed surface area and footprint, its result of 278 mg/g implies about 705 g/mol (extractor check).

Absorptive dissolution testing. In the group's apparatus (Fig. 2) the dissolution vessel, called the donor, is pumped through a bundle of hollow cellulose fibres. Buffer flows along the other side of the fibre wall in a single pass, called the receiver, and a UV probe reads its concentration in line. The receiver concentration therefore tracks the rate of absorption at each moment, not a cumulative amount. A closed compartment is an ordinary dissolution test without the membrane, so dissolved drug accumulates. To compare the two set-ups the authors use total release (stated in words on p.9, the card's E4):

In plain terms: drug still dissolved in the donor, $m_{\mathrm{d}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{r}}$, plus drug already carried across the membrane, $m_{\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{s}\mathrm{f}\mathrm{e}\mathrm{r}\mathrm{r}\mathrm{e}\mathrm{d}}$, as a percentage of the dose, $m_{\mathrm{d}\mathrm{o}\mathrm{s}\mathrm{e}}$. Le Chatelier's principle then gives the hypothesis its logic: if adsorbed and dissolved drug are in equilibrium, removing dissolved drug should shift the balance towards desorption.

Methods in detail

The work falls into four contributions. The first measures how much ATZ SBA-15 adsorbs from aqueous solution and compares it with the geometric monolayer. The second prepares five drug loadings and uses four solid-state methods to judge where the drug sits: inside the pores and bonded to the wall, or outside them as bulk amorphous drug. The third, the core of the paper, measures release with and without an absorptive sink and with drug already dissolved, to test whether the solution concentration holds release back. The fourth places the silica formulation in a physiological and practical context, with a start at gastric pH and a head-to-head comparison with an HPMC ASD.

Contribution 1: How much atazanavir does SBA-15 adsorb from solution?

① Problem. If adsorption explains incomplete release, the silica should hold a substantial amount of drug at the concentrations reached during a release test. That amount has to be measured in water, where water competes for the surface, and at supersaturated concentrations, where crystals removed by filtration would look exactly like adsorbed drug. A reference point is also needed: how much drug could a fully covered surface hold?

② What they did. For the reference point, the authors measured the molecular dimensions of ATZ on the conformer in its reported crystal structure (Cambridge Structural Database REFCODE LISTEP) and assumed that the two larger dimensions, 17.53 Å × 14.07 Å, define the contact area with the pore wall. With the surface area of 586 m²/g, the geometric monolayer came to 278 mg/g, or about 28% w/w. For the isotherm, SBA-15 was suspended at 300 or 25 µg/mL in an aqueous medium (the pH is not stated), and an aliquot of a concentrated methanolic ATZ stock was added, keeping methanol below 0.5%. HPMC at 0.1 mg/mL was present as a crystallization inhibitor. The suspensions were equilibrated for 2 h at 37 °C at concentrations between the crystalline and the amorphous solubility, filtered through 1 µm glass-fibre filters and analysed by HPLC. Two controls were reported without data: no crystallization was seen, and without HPMC, at low supersaturation, adsorption was unchanged. The 300 µg/mL data were fitted to the BET equation by non-linear least squares; Langmuir and Freundlich models fitted poorly (not shown). The number of isotherm replicates is not stated.

③ Why it works. Solution depletion measures adsorption directly in the medium of interest. HPMC removes the main artefact, loss of drug to crystals, and the no-HPMC control checks that the polymer does not itself change adsorption. Because concentration is scaled by the amorphous solubility ($X=C_e/C_s$), $X=1$ marks the highest activity an amorphous formulation can create. The equilibrium is metastable, since the solution is supersaturated with respect to the crystal, but it is the one relevant to an amorphous product. Two silica concentrations test whether the isotherm is a property of the surface alone.

④ How it differs from earlier work. The approach follows the group's ritonavir study (Dening 2019, ref 51), which used an isotherm to explain incomplete release from SBA-15. This paper applies it to ATZ, which the authors call a structural analogue of ritonavir, adds the geometric monolayer as a ceiling, and reports how the isotherm changes with silica concentration.

Contribution 2: Where the drug sits at each loading

① Problem. Release depends on whether the drug is inside the pores and bonded to the wall, or overflowing them as bulk amorphous drug, and the geometric monolayer of about 28% w/w suggests that the 5–50% series crosses from one regime to the other. Every formulation must also start amorphous, so that differences in release are not caused by crystals.

② What they did. ATZ was loaded by incipient wetness impregnation (Mellaerts 2008, ref 14): the volume of methanolic stock (10–35 mg/mL) needed for the target loading was added to SBA-15 and mixed vigorously with a spatula, then dried in an oven at 40 °C overnight and under vacuum for 48 h. Five loadings were made: 5, 10, 20, 30 and 50% w/w. Drug content was checked by HPLC after extraction in methanol (the centrifugation force is printed as "21,1000 × g", not a valid number; card note X1) and by TGA. Powder X-ray diffraction (PXRD, 5–40° 2θ) was run after drying and after storage in a desiccator at room temperature for up to 3 months, with a physical mixture containing 5% crystalline ATZ as a sensitivity control. Modulated DSC (2 °C/min from 10 to 220 °C) was compared with a 1:1 physical mixture of melt-quenched amorphous ATZ and SBA-15. TGA ran to 600 °C under nitrogen, with the degradation temperature taken from the derivative curve. ATR-FTIR spectra were compared with crystalline and melt-quenched amorphous ATZ.

③ Why it works. Each method answers a different question. PXRD detects crystals. MDSC shows whether the drug behaves like a bulk glass, with a glass transition, crystallization on heating and melting, or shows none of these, as expected for confined molecules. TGA reports the degradation temperature, which the authors expect to rise when drug–silica interactions raise the activation energy for degradation. FTIR identifies bonding partners: a shift of the ATZ carbonyl bands and loss of the isolated-silanol band point to hydrogen bonds between carbonyls and silanols.

④ How it differs from earlier work. The techniques and the confinement arguments are standard (refs 14, 16, 19, 64, 65); what this paper adds is that the same five loadings are then tested under dissolution–absorption.

Contribution 3: Release with and without an absorptive sink

① Problem. A closed-vessel plateau cannot tell drug that is unable to leave the silica from drug that stays because the surrounding solution is already concentrated. If an equilibrium between dissolved and silica-held drug holds release back, lowering the solution concentration should release more drug and raising it should release less.

② What they did. The dissolution–absorption runs used the group's flow-through apparatus (Fig. 2; refs 45, 48, 49). The donor was 50 mL of pH 6.8, 50 mM phosphate buffer at 37 °C with 100 µg/mL HPMC already dissolved, pumped at 4 mL/min through a cellulosic hollow-fibre module (100 cm², 15 nm pores); a 10 µm cannula filter kept silica out of the fibres. Buffer flowed co-currently through the receiver side at 4 mL/min and was read in line by UV, and the donor was sampled by hand, filtered (1 µm) and analysed by HPLC. All runs were in triplicate at a 75 µg/mL dose. About 5–6% of the drug adsorbed on the membrane, and silica particles were still visible at the end of every experiment. The five loadings were run for 90 min (Fig. 8). The 20% loading was run for 240 min with the membrane and, for comparison, in a closed compartment of the same medium without it (Fig. 10). In a separate closed-compartment test, the 20% formulation was added to buffer that already contained 0, 20 or 40 µg/mL dissolved ATZ, and the plotted concentration is the measured value minus the pre-dissolved amount (Fig. 9). Temperature, agitation and replicates of the closed-compartment tests are not stated (card note X17).

③ Why it works. The two experiments push the same variable in opposite directions. The membrane removes drug continuously and keeps the donor concentration below what a closed vessel would reach; the pre-dissolved drug raises the solution concentration from the first minute. If release responds to the solution concentration, the sink should raise it and pre-dissolved drug should lower it. The total-release mass balance makes runs with and without a membrane comparable, and the card's extractor confirmed that it reproduces: for the 20% loading at 90 min, about 36.8% of the dose sat in the donor and about 29.7% had crossed the membrane (receiver curve integrated at 4 mL/min), summing to the plotted 66.5%.

④ How it differs from earlier work. Silica formulations are generally tested in closed compartments. The group's ritonavir study (Dening 2018, ref 44) measured flux across a membrane, but used it to show that incomplete release lowers flux. Here the membrane is used as a lever on release itself, and the pre-dissolved-drug test, run in the opposite direction, is new for this system. The idea that absorption may pull drug off the silica had been raised to explain a missing in vitro–in vivo correlation (Vialpando 2016, ref 47); this paper turns it into a direct in vitro comparison.

Contribution 4: Gastric start and comparison with an amorphous solid dispersion

① Problem. A fasted stomach is acidic, and ATZ, as a weak base, is far more soluble there, so a test at pH 6.8 alone may miss how the formulation performs in vivo. A second question is practical: even if the silica releases more under absorptive conditions, does it match a polymer-based ASD of the same drug?

② What they did. In the two-step test (Fig. 11), the 20% formulation was dissolved in 0.1 N HCl (pH 1) for 30 min; the pH was then raised to 6.8 by adding concentrated HCl and NaOH solutions (as printed), absorption was started, and the run continued to 120 min. The medium after the shift is not described. In the comparison (Fig. 12), 10% ATZ-MPS and a 10% ATZ-HPMC ASD were each dosed at 200 µg/mL, above the amorphous solubility so that LLPS could occur, and tested in the absorptive set-up for 90 min. The ASD's preparation is not described (its behaviour was reported in ref 49), nor is the medium pH. A preliminary two-step run with the 10% formulation is mentioned without data.

③ Why it works. Starting in acid dissolves much more of a weak base, and the shift to pH 6.8 then leaves the solution highly supersaturated, which tests whether a higher donor concentration raises the absorption rate. Dosing both formulations at 200 µg/mL lets each reach its own maximum concentration; a formulation that reaches the amorphous solubility should show LLPS as turbidity.

④ How it differs from earlier work. Here a two-step pH protocol is combined with an absorptive sink for a silica formulation, and two kinds of amorphous formulation are compared on flux, not only on dissolved concentration.

Experimental analysis

Setup in brief. One drug, one silica and five loadings (5–50% w/w) make up the study. Release tests ran at a 75 µg/mL dose in 50 mL of medium containing 100 µg/mL HPMC, and the ASD comparison at 200 µg/mL. The dissolution–absorption runs are triplicates; replicates of the isotherms and closed-compartment runs are not stated. The paper prints no data tables, so the table values below were read from the figures by the card's extractor and are approximate. Where the text gives a different number, the figure value is used and the difference is noted.

Result 1: little drug adsorbs from solution. With 300 µg/mL SBA-15, the isotherm had the type II shape: adsorption rose, levelled off, then rose again towards the amorphous solubility (Fig. 3a). The highest measured point was about 21 mg/g at about 64.5 µg/mL. The text gives the maximum as "equivalent to 25 mg/g" at the amorphous solubility, but no point is plotted there and the fitted curve ends at about 22.8 mg/g, so 25 mg/g is best read as an extrapolation (card note X3). The BET fit gave a monolayer coverage of 4.5 mg/g, against the geometric estimate of 278 mg/g: about sixty times less. Table A shows selected points for both silica concentrations.

Table A. Adsorption of ATZ onto SBA-15 after 2 h at 37 °C with 0.1 mg/mL HPMC, at two silica concentrations (read from Fig. 3a and 3b, approximate; the meaning of the error bars is not stated).

Two features of Table A matter for the rest of the paper. First, adsorption is small: even at the highest concentration the silica holds about 2% of its own mass in drug. The authors' explanation is competition with water: drug adsorbs only if the free energy gained balances the cost of displacing water from the surface, so complete coverage by drug "is unlikely in the presence of water", and the low adsorption "may be due" to this competition. They attribute the misfit at low concentration to a possibly heterogeneous surface ("may result") and suggest "possibly" multilayer adsorption beyond 4.5 mg/g. Second, the isotherm depends on how much silica is present. With 25 µg/mL silica, the amount adsorbed per gram was several times higher at the same concentration (about seven to nine times near the top of the range), while the share of drug adsorbed from a 60 µg/mL solution was lower, 4% against 9%. The authors explain this by a constant adsorption capacity of the silica, and conclude that the amount of silica in a release test "should be considered" when adsorption is used to explain incomplete release.

Note by the AI writer (not discussed in the paper), from the card's extractor note X15. If adsorption per gram were a property of the surface, one isotherm would give the same mg/g at the same equilibrium concentration whatever the silica amount, so the constant-capacity argument does not explain the difference. With only about 4% of the drug removed, the 25 µg/mL data rest on depletions of a few µg/mL, close to the precision of the HPLC assay. The depletion relation reproduces the plotted points only approximately (9% of 60 µg/mL on 300 µg/mL silica gives 18 mg/g against about 16.2 plotted; 4% on 25 µg/mL silica gives 96 mg/g against about 116). The 25 µg/mL isotherm is best used qualitatively.

Result 2: low loadings look confined, high loadings look like bulk drug. All five formulations were amorphous by PXRD after preparation (Fig. 4), though the control with 5% crystalline ATZ shows only one weak peak near 8.5° 2θ, so crystallinity below roughly that level would not be seen (card note X5). Table B collects the solid-state data.

Table B. Solid-state behaviour of the ATZ-MPS formulations by drug loading (MDSC and FTIR reference values from the text; band positions and DTG minima read from Fig. 6b and Fig. 7, approximate; storage in a desiccator at room temperature).

The MDSC pattern in Table B is clear. At 5, 10 and 20% loading the drug shows neither a glass transition nor melting; at 30% both appear weakly, and at 50% they match the amorphous drug in the physical mixture, including crystallization on heating and melting at 208 °C. The authors read the absence of events at low loading as "consistent with nanoconfinement and drug-pore wall interactions", and the events at 30% and 50% as "most likely due to excess drug on the silica particles", that is, drug outside overfilled pores. They argue that drug inside the pores, even as multilayers, would not crystallize. FTIR points the same way. The isolated-silanol band is weak at 5% and 10% and gone from 20%, which "could indicate drug-silanol interactions", and the carbonyl band lies about 10 cm⁻¹ below its amorphous position at low loading but moves back as loading rises (the text's band naming is loose; card note X7). Physical stability fits the picture: only the 50% formulation crystallized within a week.

The TGA evidence is weaker than the text implies. The authors write that degradation temperatures are higher at low loading (5% and 10% on p.8; 5, 10 and 20% on p.14) and fall towards that of ATZ as loading rises, but the minima in Table B are not ordered up to 30%; only the 50% sample lies close to the drug alone (card note X6). The geometric monolayer of 278 mg per gram of silica equals 27.8% of the silica mass but 21.8% of drug plus silica, and the paper does not define the basis of "% w/w", so the monolayer is reached at about 22–28% loading (card note X4). Either way, the change from confined to bulk-like behaviour between 20% and 30% sits near this estimate.

Result 3: release falls as loading rises. Under dissolution–absorption at pH 6.8 (Fig. 8), all formulations released part of their dose at once and then more slowly, and none reached the amorphous solubility in the donor. Table C summarizes the loading series.

Table C. Dissolution–absorption of ATZ-MPS at a 75 µg/mL dose, pH 6.8, 37 °C, n = 3 (read from Fig. 8a–c, approximate). The silica concentration assumes that drug loading means drug/(drug + silica) and is the card's calculation, not a value in the paper.

Release and flux both fall with loading. At 5% and 10% the donor peaked at about 33–35 µg/mL around 30–50 min and then declined slowly; the authors explain the decline by release being "slower than the drug permeation rate" across the membrane. At 50% there was almost no burst, the donor climbed slowly to about 19 µg/mL, and the receiver showed a lag. The amount absorbed at high loading was about half that at low loading. The text rounds the 5% and 10% results to "approximately 80%" and gives the 5% burst as 30%; the figure shows about 73% and about 25% for the 5% formulation (card note X8). The trend is therefore flat at low loading and falls from 20% upwards.

The authors offer a two-part explanation. At 5% and 10% loading, incomplete release "can be attributed largely to the drug-silica interactions". At 20–50%, it "is likely caused both by increased hydrophobicity of the formulation, as well as ATZ-SBA-15 interactions": as hydrophobic ATZ fills the pores, water enters more slowly, and poor wetting "may contribute" to the losses. The results section is more cautious, saying incomplete release "can likely be partially attributed" to adsorption, and the discussion adds that the drug left in the silica was "much greater than expected based on the adsorption isotherm, especially for higher drug loading formulations".

Note by the AI writer (not discussed in the paper), from the card's extractor note X14. The silica concentration in the release tests varies twentyfold with loading (Table C), and only the 20% formulation matches the 300 µg/mL isotherm. In the closed vessel that formulation had dissolved about 31 µg/mL at 90 min, leaving about 44 µg/mL in the solid, while the isotherm at 31 µg/mL gives roughly 7 mg/g, or only about 2 µg/mL adsorbed on 0.3 mg/mL silica. Measured adsorption thus accounts for about a twentieth of the retained drug. This supports the authors' statement that retained drug far exceeds the isotherm, but not the abstract's wording that closed-compartment release is incomplete "largely due to the adsorption tendency of ATZ".

Result 4: drug already in solution holds release back. In the closed vessel the 20% formulation was added to buffer containing 0, 20 or 40 µg/mL dissolved ATZ (Fig. 9). Table D gives the concentration released by the formulation itself and the total in solution.

Table D. Closed-compartment release of 20% ATZ-MPS at pH 6.8 with ATZ already dissolved in the medium; released values are the measured concentration minus the pre-dissolved concentration (read from Fig. 9, approximate; the dose is 75 µg/mL by the card's comparison with Fig. 10c).

Every run still showed a burst, which the authors take to mean that "the pre-dissolved drug did not show net adsorption onto the silica surface" ("suggested"). After the burst, release levelled off lower the more drug was already present: 40 µg/mL of dissolved drug halved the amount released at 90 min, although the total in solution was highest in that run. The authors read this as "a dynamic equilibrium between solution concentration and the drug release", in which a higher solution concentration lowers the driving force for drug to leave the silica surface. The 0 µg/mL curve is the same data set as the closed-compartment curve of Fig. 10c (31.2 µg/mL equals 41.6% of a 75 µg/mL dose), not an independent repeat (card note X9).

Result 5: the absorptive sink roughly doubles release. This is the central experiment (Fig. 10). The 20% formulation was followed for 240 min with and without the membrane. Table E gives the time course.

Table E. Total release of 20% ATZ-MPS at a 75 µg/mL dose and pH 6.8 with the absorptive membrane and in a closed compartment, and the donor concentration in the sink run (read from Fig. 10a and 10c, approximate).

With the sink, the donor stayed near 28–29 µg/mL from about 20 to 75 min and then fell to about 15 µg/mL at 240 min; the receiver peaked at about 3.6 µg/mL at 40–50 min and declined to about 1.8–1.9 µg/mL. Total release rose steadily to about 89.5% at 240 min, while the closed vessel reached about 40% by 60 min and crept up to about 46.5%, a gap of about 43 percentage points. The authors conclude that "removal of drug due to permeation across the membrane promotes additional drug release from the MPS" ("These observations suggest"), and in the discussion they explain it by le Chatelier's principle: lower drug concentration "may improve the release of drug in the lumen as the equilibrium shifts towards desorption".

Three corrections to the wording apply. The abstract states "complete release over a 240 min time period", but the figure shows about 89.5% (card note X10). The closed vessel is described as reaching "only 40% drug release, with a plateau reached after 60 min", but its curve keeps rising to about 46.5% (card note X11). And about 5–6% of the drug binds to the membrane; the paper does not say whether this is counted in total release, so the sink value may be understated by up to that amount (card note X18). None of these changes the direction or the size of the effect much.

Result 6: a gastric start raises flux, but the ASD gives more. Table F compares the pH-shift run with the one-step run and the silica formulation with the ASD.

Table F. Effect of a gastric start and comparison with an HPMC ASD under dissolution–absorption (read from Fig. 8, Fig. 11 and Fig. 12, approximate; receiver times for the two-step run counted from the start of absorption).

At pH 1 the 20% formulation released about 60% at the first sample and about 72–74% by 30 min. The text's "almost 80%" exceeds the figure (card note X12). The authors attribute this to the much greater solubility of the weak base at low pH ("can be attributed") and "to a small extent" to repulsion between protonated ATZ and the positively charged silica below pH 2. After the shift to pH 6.8 the donor fell from about 55 to 32 µg/mL as drug crossed the membrane, while total release crept up to about 88.6% ("only a small extent of additional release"). The receiver peak, about 6.4 µg/mL, was almost twice the one-step peak of about 3.6 µg/mL, and the authors state that the greater supersaturation "clearly" raised absorption.

In the comparison at 200 µg/mL, the silica formulation stayed at about 48–57 µg/mL, below the amorphous solubility, whereas the ASD rose to about 207 µg/mL within about 15 min, turned turbid and declined slowly. The receiver concentration was about twice as high for the ASD. The authors conclude that the ASD gave "substantially enhanced absorption", and that release driven by membrane removal "is insufficient to compensate for the lower extent of supersaturation achieved". The ASD's preparation and donor measurement are not described (its donor was plotted at about 1-min intervals, unlike the manual HPLC sampling in the methods), and its donor values include drug in LLPS droplets, so the receiver data are the safer basis for comparison (card note X19). A preliminary acid-first run with the 10% formulation (data not shown) reached the amorphous solubility without any sign of LLPS, from which the authors infer that the nanodroplet reservoir effect "may not occur" with silica formulations.

Comparisons across experiments and what the controls rule out. The release data form a consistent pattern around one variable, the ATZ concentration next to the silica: removing drug through the membrane raised release (Table E), adding dissolved drug lowered it (Table D), and starting in acid, where much more drug can dissolve, raised it (Table F). The controls remove several simpler explanations. The isotherm run without HPMC showed that the polymer does not change adsorption. Every formulation started amorphous by PXRD, so crystals present at the start do not explain the falling release with loading, at least above the detection limit. The 10 µm filter kept silica particles out of the hollow fibres. The closed-compartment and sink runs used the same medium and dose, so the difference between them is attributed to the membrane. What the controls do not rule out is whether the drug that comes out under the sink was adsorbed on the pore wall or simply held in the pores.

Discussion

Where the findings apply. The study covers one drug (a lipophilic weak base), one silica, one cellulose membrane and media containing 100 µg/mL HPMC at fasted-state pH, and the central sink result was measured at a single loading (20%) and pH (6.8). The strength of a membrane sink also depends on the apparatus: the 50 mL donor, the 100 cm² membrane and the 4 mL/min flows set how fast drug is removed relative to how fast it is released, and the gut will differ. Within these limits the paper shows a large effect: removing dissolved drug about doubles the release of ATZ from SBA-15 over 4 h. Whether it holds for other drugs, silicas or loadings is not tested.

What drives the sink effect is still open. The authors frame the result as desorption: removing drug lowers the solution concentration and, by le Chatelier's principle, shifts the adsorption equilibrium towards release. The pre-dissolved-drug experiment points the same way. Note by the AI writer, from the card's extractor note X16: both experiments change the bulk concentration, and that concentration controls any gradient-driven release, not only desorption. Drug held in the pores without being adsorbed, which must diffuse out, would also leave faster when the bulk concentration falls and slower when it rises. Since the drug on the silica was not measured, the data cannot separate the two. The size of the retained fraction adds to the doubt: the measured isotherm accounts for only about a twentieth of the drug kept back at 20% loading (card note X14), and the authors themselves call the retained drug "much greater than expected". On these numbers, adsorption at the measured level is unlikely to be the main brake. Note by the AI writer (not discussed in the paper): the isotherm was measured by adding dissolved drug to empty silica, whereas the formulations were loaded from methanol and dried, so drug that starts inside filled pores may be held in a state the isotherm does not describe.

Loading and pore capacity. The solid-state and release data line up around the pore capacity. Up to 20% loading the drug shows no thermal events and releases about 66–80% in 90 min; from 30% bulk-like drug appears, release falls to about 52% and 36%, and at 50% the formulation crystallized within a week. The geometric monolayer, about 22–28%, sits at the transition. The authors attribute the high-loading losses partly to hydrophobic, overfilled pores that wet poorly ("likely", "may contribute"), but wetting was not measured. Note by the AI writer: for loading choice, these data suggest that staying within the pore capacity serves both release and physical stability, and the equilibrium isotherm is a weaker guide.

The supersaturation ceiling. At pH 6.8 no silica formulation reached the amorphous solubility, even when dosed at 200 µg/mL, while the HPMC ASD exceeded it, underwent LLPS and gave about twice the flux. The authors' reading is careful: "it appears that the extent of supersaturation generated may be limited", as they had also seen for ritonavir. In this paper the membrane raised the fraction of the dose released over time, not the donor concentration, and the authors judge it "insufficient to compensate". Starting in acid was the one route to a higher concentration: the pH-shift run nearly doubled the receiver peak, and a preliminary run reached the amorphous solubility, though without LLPS, so the nanodroplet reservoir that supports flux from ASDs "may not occur" with silica. The authors also note the practical catch: ATZ is usually taken with food (ref 87), so "the advantage from dissolution at a low pH may not be obtained", a concern smaller for ASDs designed to release in the intestine.

What remains unresolved. The authors raise one open question explicitly: whether silica dissolution affects drug release, and whether adsorbed drug affects silica dissolution, is "currently unknown". Other gaps follow from the design: drug on the silica was not assayed after release; the isotherm was measured at two silica concentrations while the release tests contained about 75–1425 µg/mL; wetting was proposed but not measured; only fasted-state pH was used; and the ASD comparison covers one loading and dose.

Implications for testing and formulation. The authors draw three practical conclusions. An absorptive compartment "can provide more in vivo relevant insights" than a closed-compartment test (Fig. 10c). Gastrointestinal pH "should be considered" when designing dissolution tests for silica formulations. And "design of an appropriate in vitro evaluation methodology is clearly critical". Note by the AI writer (not stated in the paper): a low closed-vessel plateau does not show that drug is permanently lost to the silica, so ranking silica formulations by such plateaus understates them, while the ASD comparison suggests that the real limit is the peak supersaturation the carrier can build. Useful next experiments would be to assay residual drug on the silica after a sink run, to measure isotherms at the silica concentration of each release test, to repeat the sink comparison at other loadings, and to test fed-state media.

Limitations

Limitations stated by the authors. The paper has no separate limitations section, but it states several caveats. Silica dissolution was "not considered herein", although silica "has a small but finite solubility" under the test conditions, and its role in release is "currently unknown". Physical stability was "not evaluated beyond 3 months". About 5–6% of the drug adsorbed on the hollow-fibre membrane. The BET equation rests on assumptions that "few systems conform to": multilayer adsorption with the strongest first layer, no lateral interactions and a homogeneous surface. The geometric monolayer is a "simplified geometric approach" that considers geometry only. The isotherm describes a metastable equilibrium. Several supporting results are "data not shown", including the isotherm control without HPMC, the crystallized 50% formulation and the preliminary two-step run. Finally, with food the low-pH advantage "may not be obtained".

The AI writer's own judgement (1): the mechanism is inferred, and the abstract is stronger than the data. The card rates the paper's overall strength as moderate. The sink effect itself is solid: it is large (about 43 percentage points at 240 min), the sink runs are triplicates, and the total-release mass balance reproduces. The explanation is weaker. The abstract states that absorptive conditions "promote drug desorption from the silica surface", while the discussion says only "may improve". No drug on the silica was measured, and diffusion of non-adsorbed drug out of the pores fits the same data (card note X16). The abstract also attributes incomplete closed-compartment release "largely" to adsorption, but closed-compartment data exist only for the 20% loading, the results section says "partially", and the isotherm accounts for about a twentieth of the retained drug (card notes X13, X14). Read strictly, the data support "a membrane sink releases much more drug" and do not show why.

The AI writer's own judgement (2): fewer independent experiments than figures, and rounded-up text values. The figures share data: the first 90 min of the sink run are the 20% curve of Fig. 8, and the 0 µg/mL curve of Fig. 9 is the closed-compartment curve of Fig. 10c (card note X9). The sink, loading and pre-dissolved results are therefore not independent repeats, and the key comparison rests on one loading at one pH. Several numbers in the text exceed the figures, as noted where they appear above: "complete release" (X10), the 40% plateau (X11), "almost 80%" at pH 1 (X12), 80% for the 5% loading (X8) and the 25 mg/g maximum (X3). The amorphous solubility is given as 75 µg/mL when the dose is set and 78 ± 5 µg/mL in the results, both citing ref 49 (X2). None of these reverses a conclusion, but each makes the result look slightly cleaner than the figures show.

The AI writer's own judgement (3): unstated conditions and weaker supporting evidence. Several conditions are not stated: temperature, agitation and replicates of the closed-compartment tests, the pH of the isotherm and ASD runs, and the medium after the pH shift (X17), as well as the ASD's preparation and measurement (X19). Membrane-bound drug may or may not be counted in total release (X18). The isotherm's dependence on silica amount is unexplained and the 25 µg/mL data are imprecise (X15). Among the solid-state data, the TGA trend is not ordered by loading (X6), PXRD cannot see crystallinity below roughly 5% (X5), the basis of "% w/w" is undefined (X4), and the assay's centrifugation force is misprinted (X1). These weaken the supporting arguments for confinement and adsorption more than the central sink result.

Conclusion

The paper asked whether adsorption onto silica limits atazanavir release from SBA-15, and whether an absorptive sink draws more drug out. The sink clearly works: at 20% loading, total release reached about 90% in 240 min with the membrane against about 46% in a closed vessel, and drug already in solution held release back. Adsorption from solution was small (BET monolayer 4.5 mg/g against a geometric 278 mg/g) and, by the extractor's calculation, accounts for only a small part of the retained drug. Release fell as loading rose, with the change from confined to bulk-like drug between 20% and 30% loading, near the geometric monolayer. A start at pH 1 raised flux through the higher solubility of the weak base.

Two qualifications frame the result. First, the mechanism is the authors' interpretation: desorption by le Chatelier's principle fits the data, but so does diffusion of non-adsorbed drug out of the pores, and the drug held on the silica was never measured. Second, even with the sink, the silica formulation did not reach the amorphous solubility and gave about half the flux of an HPMC ASD of the same drug. In the authors' words, membrane-driven release "is insufficient to compensate for the lower extent of supersaturation achieved". For formulation work, the paper argues for testing silica formulations with an absorptive sink and at relevant pH, and its ASD comparison adds that the supersaturation a carrier can generate, not only the fraction it releases, sets the flux.