Why drug-loaded mesoporous silica releases only part of its dose: hydrogen bonding, solute activity and adsorption from supersaturated solution

Background and motivation

Ordered mesoporous silica is used as a carrier for poorly water-soluble drugs: drug loaded into the pores stays non-crystalline below a certain loading, and on contact with water it can produce supersaturated solutions (Section 2) C1. Two questions about such carriers were open when this paper was written. The first is how drugs actually bind silica. Some studies attribute the binding to hydrogen bonding alone, others to a mixture of hydrogen bonding and non-specific hydrophobic interactions, and electrostatic interactions matter for some drugs depending on their properties, the local pH and the silica surface chemistry C2. The second is why release is so often incomplete. Incomplete release has been reported even under sink conditions, including in earlier work by the first author (McCarthy 2018) R30.

The way the supersaturation question is framed comes from work on amorphous solid dispersions. The highest supersaturation a drug can reach is set in theory by the spinodal limit and in practice by its amorphous solubility, the largest free-drug concentration obtainable relative to the crystal (Taylor and Zhang 2016; Almeida e Sousa 2015) C3. Above it, a slowly crystallizing drug separates into a water-rich phase and a drug-rich phase, liquid–liquid phase separation (LLPS) C3. Only one earlier study had examined release from mesoporous silica under such supersaturating conditions. Dening and Taylor (2018) found that ritonavir was released from SBA-15 very rapidly, yet no LLPS formed even when the amount of drug added exceeded the amorphous solubility; release was incomplete and the supersaturation reached was reduced R28. The proposed explanation (Dening et al. 2019) was a dynamic equilibrium between drug in solution and drug adsorbed to the silica, so that the silica retains drug even at low solution concentrations C4. This fits the observation that mesoporous silica removes poorly soluble drugs from water at low concentration (Bui and Choi 2009) R31.

Adsorption isotherms of drugs on silica had mostly been measured below the crystalline solubility and analysed with the Langmuir and Freundlich equations (Section 2). For carriers designed to supersaturate, the relevant range lies above the crystalline solubility, up to the amorphous solubility, and there little was known about how the degree of supersaturation governs the tendency of a drug to adsorb. A useful idea came from older work on adsorption from solution onto carbon. Hansen and co-workers found that structurally related aliphatic alcohols and acids adsorb alike when compared on a reduced-concentration scale, that is, at equal solute activity, and Wurster and co-workers related diazepam adsorption on activated carbon to its solubility R29. Those studies used crystal or liquid solubilities, not supersaturated solutions R29.

The paper therefore compares two structural analogues with different hydrogen-bonding capability (Figure 1). Indomethacin (IND) has one hydrogen-bond donor, the carboxylic acid, and four acceptors; indomethacin methyl ester (INDME) has no donor and the same four acceptors M1·M2. The authors hypothesized that this difference would change adsorption C5. They measured adsorption isotherms from supersaturated solutions, tested the roles of hydrogen bonding and hydrophobic interactions by adding urea or NaCl, characterised the drug–silica interaction by infrared, fluorescence and X-ray photoelectron spectroscopy, and finally measured release from drug-loaded SBA-15 under non-sink conditions (Section 2) C5·C6.

Background concepts

Amorphous solubility and LLPS. When a poorly soluble drug is supersaturated in water, there is an upper concentration beyond which the solution separates into drug-rich droplets: the amorphous solubility, also called the LLPS onset concentration. In this paper it was measured in two ways at pH 2.2 and 37 °C, with 100 µg/mL HPMC (hydroxypropyl methylcellulose, Pharmacoat 606) present to inhibit crystallization P2·P3·M4. In the UV method, a methanol stock was pumped into buffer while the extinction at 450 nm, a wavelength where the drug does not absorb, was recorded; a rise in extinction with the drug's own absorbance staying high marks a new droplet phase rather than crystals P2. In the ultracentrifugation method, a supersaturated, opaque solution was spun at 35,000 rpm for 60 min and the clear supernatant assayed P3. The ratio of amorphous to crystalline solubility measures how much supersaturation is available.

Silica surface chemistry. The silica surface carries two kinds of group: siloxane bridges (Si–O–Si), whose oxygen lone pairs are tied up and cannot accept hydrogen bonds from adsorbates, and silanol groups (Si–OH), isolated, geminal or vicinal, which can donate or accept hydrogen bonds C8. At pH 2.2 the authors take the silanols to be predominantly protonated and neither drug to be ionized, so electrostatic interactions are not expected C7. Drugs are reported to bind silica by hydrogen bonding, ion exchange with hydrophilic groups, or hydrophobic interaction C8.

Solute activity on the reduced-concentration scale. Two drugs with different solubilities cannot be compared fairly at equal concentration, because the less soluble one is closer to saturation. The paper follows Hansen and Craig and approximates the thermodynamic activity by the reduced concentration (Equation 2):

Here $C_e$ is the equilibrium concentration of free drug in solution. In plain terms: the activity is how far the solution has gone toward the amorphous solubility, which is given the value 1; the reference state is the water-saturated amorphous drug E2·C10. Hansen and Craig used the solubility of a liquid solute; for high-melting compounds such as these, the authors argue that the amorphous solubility "can reasonably be replaced" for it C10. The concentrations used for Figure 5 were 0.111 mM (IND) and 0.025 mM (INDME) P18.

Adsorption isotherms. An adsorption isotherm relates the amount adsorbed per gram of adsorbent, $Q_a$ (mg/g), to the equilibrium concentration $C_e$ (µg/mL). The paper fits three models, given in the Supporting Information. The Langmuir equation, used in its linear form, assumes a single layer of adsorbed molecules (Equation SE1):

In plain terms: adsorption saturates at the monolayer capacity $Q_m$, and the Langmuir constant $K_L$ sets how quickly it approaches that limit E6. The Freundlich equation, also linearised, is empirical and allows the amount adsorbed to keep rising (Equation SE2):

In plain terms: the logarithm of the amount adsorbed rises in a straight line with the logarithm of concentration, with intercept $K_F$ and slope $1/n$ E7. The BET equation in the form proposed by Ebadi and co-workers for adsorption from liquid solution allows multilayers (Equation SE3):

Here $K_S$ is the equilibrium constant for the first layer and $K_L$ the constant for the upper layers. In plain terms: the first layer forms with constant $K_S$, further layers stack up with constant $K_L$, and the amount adsorbed diverges as $K_L{C}_e$ approaches 1, so $1/K_L$ plays the role of a saturation concentration E8·C21.

Theoretical monolayer. To decide how much drug to load, and to draw a reference line on the isotherms, the authors estimate the mass of drug that covers the silica surface exactly once (Equation 3, after Dening and Taylor 2018):

Here $X_m$ is the drug needed for monolayer coverage (g per g of silica), SSA the specific surface area of the silica (m²/g; the factor ${10}^{20}$ converts m² to Ų), $MW$ the molecular weight of the drug, $S_c$ the area one molecule covers (Ų) and $N_A$ Avogadro's number. In plain terms: the surface area divided by the footprint of one molecule gives the number of molecules in one layer, and this number times the molar mass over Avogadro's number gives the mass E3. The footprint was taken as the product of the two longest molecular dimensions measured in the crystal structures (Figure 1: 13.0 × 9.4 × 4.0 Å for IND and 13.9 × 10.0 × 4.1 Å for INDME), and the SSA was the supplier's 586 m²/g P11·M1·M2·M3.

Pyrene as a polarity probe. The fluorescence spectrum of pyrene has vibronic peaks whose intensity ratio, the first over the third peak ($I_1/I_3$), falls when pyrene sits in a less polar, more hydrophobic environment P5·C16. A drop in $I_1/I_3$ therefore signals that pyrene has found hydrophobic material, such as drug-rich droplets or, the authors argue, layers of drug adsorbed on silica C16·C17.

Supersaturation ratio. The degree of supersaturation is expressed as $S$, approximated by $C/C_{cry}$, the solution concentration over the equilibrium crystalline solubility (Section 4.5) E4.

Methods in detail

The work has three parts that build on each other: first, identify how each drug binds the silica; second, measure how much drug adsorbs from supersaturated solution and compare the two drugs on a common scale; third, measure release from drug-loaded SBA-15 and ask whether adsorption accounts for what is not released. All experiments used McIllvaine buffer at pH 2.2 (the paper's spelling) and 37 °C, and, wherever drug was supersaturated, HPMC at 100 µg/mL to prevent crystallization M7·M4. The flow below summarises the work.

Contribution 1: identifying the drug–silica interaction with a donor/no-donor pair

Problem. If drugs bind silica through hydrogen bonds, it matters which partner donates. Silanols can donate or accept C8. IND can donate through its carboxylic acid and accept through its carbonyl and ether oxygens; INDME can only accept M1·M2. Comparing the two isolates the role of the drug's donor, and adding urea or salt tests hydrogen bonding against hydrophobic interaction.

What they did. Attenuated total reflectance FTIR spectra (Bruker Vertex 70, 600–4000 cm⁻¹, 64 scans, 4 cm⁻¹ resolution) were compared for amorphous IND and INDME, the drug-loaded SBA-15 and blank SBA-15 in the carbonyl region, 1600–1800 cm⁻¹ P4. X-ray photoelectron spectra (Kratos Axis Ultra DLD, monochromatic Al Kα, charge-referenced to the C–C component of C 1s at 284.8 eV, at least four spots per sample) recorded the Si 2p and O 1s regions of SBA-15, the two drugs and the two loaded samples P6. The paper does not state the loading of the samples used for FTIR and XPS, or how the amorphous reference drugs were made X20. Pyrene (0.25 µg/mL) was added to buffer alone, to SBA-15 at 50 µg/mL, to drug above its LLPS concentration, and to drug below it with SBA-15 P5·R6. For the urea test, adsorption was run at an initial activity of about 0.2 for both drugs (8 µg/mL IND, 2 µg/mL INDME), with and without 5 M urea; urea is a strong hydrogen-bond former and should compete with the drug for silanols if hydrogen bonding matters P9·C28. For the salt test, adsorption at 2 and 8 µg/mL of each drug was compared in distilled water, 0.1 M NaCl and 0.5 M NaCl, each with HPMC; if the drugs bound silica hydrophobically, salt should increase adsorption by salting out P10·C30.

Why it works. A hydrogen bond to a carbonyl oxygen weakens the C=O bond and lowers its stretching frequency, so a downward shift of a carbonyl band on silica identifies that carbonyl as a hydrogen-bond acceptor C12·C13. XPS adds the electronic picture: a hydrogen bond that moves electron density from the drug's carbonyl to the silica should lower the Si 2p binding energy, and the carbonyl oxygen's own O 1s component, normally the lowest-energy one in the drug, should move up and merge into the silica peak C14·C15. Urea and salt then test the two candidate interactions directly, one by competition and one by changing the solvent's hydrophobic driving force C28·C30. Difference from earlier work. Earlier studies of drug–silica binding usually examined one drug at a time C2; here a single methyl group switches the donor off while the acceptors stay, so the donor's contribution can be read directly C5.

Contribution 2: adsorption isotherms from supersaturated solution, compared on the activity scale

Problem. For a supersaturating carrier the interesting part of the isotherm lies between the crystalline and the amorphous solubility, a range that standard isotherm studies do not cover; and two drugs whose amorphous solubilities differ about fourfold cannot be compared fairly at equal concentration R1.

What they did. SBA-15 (5 mg) was stirred in 40 mL buffer with HPMC at 37 °C, drug was added from a 10 mg/mL methanol stock to initial concentrations of 1–60 µg/mL (IND) or 1–14 µg/mL (INDME), and after 8 h the suspension was centrifuged and the supernatant assayed by HPLC P7. The amount adsorbed came from the depletion of the solution (Equation 1):

Here $C_i$ is the initial drug concentration, $C_e$ the equilibrium concentration in the supernatant, $V$ the volume of solution and $m$ the mass of SBA-15. In plain terms: the drug that disappeared from solution is taken to be on the silica, expressed per gram of silica E1. Because supersaturated solutions can crystallize, they are not at true equilibrium; the authors call the state reached a transient or metastable equilibrium that lasts as long as no crystallization occurs, and none was observed during the experiments C9·R9. The isotherms were fitted with the three models above and replotted against activity by Equation 2 P17·P18. A desorption test followed: suspensions loaded at an initial activity of about 0.2 were diluted with fresh buffer to sink conditions ("33 % of the crystalline saturated solubility"), chosen so that complete desorption would give both drugs the same activity, and left 24 h P8.

Why it works. Converting concentration to activity removes the solubility difference between the two drugs. If both bind the surface the same way, their adsorption should then depend only on how close each solution is to its own amorphous solubility, which is what Hansen's carbon studies found for related solutes C10·R29. Difference from earlier work. Those studies used crystal or liquid solubilities as the reference; this paper uses the amorphous solubility and works in the supersaturated range, up to and beyond the amorphous solubility as initial concentration R29·R9.

Contribution 3: release from drug-loaded SBA-15 and its link to adsorption

Problem. If adsorption is strong near the amorphous solubility, a formulation that tries to reach that concentration should lose part of its dose to the carrier. This needs formulations that are amorphous, a dose that targets the amorphous solubility, and a way to check that what is not released is still on the silica rather than crystallized.

What they did. IND and INDME were loaded onto SBA-15 from ethanol by solvent incipient wetness, at amounts equal to 25, 50, 75, 100, 125 and 150% of the theoretical monolayer from Equation 3, dried (40 °C for 24 h, then 48 h under reduced pressure), assayed after methanol extraction and examined by PXRD P12·P13. The 50% formulations, X-ray amorphous, were added to 75 mL buffer with HPMC at 37 °C so that complete release would give the amorphous solubility, 35.1 µg/mL (IND) or 9.0 µg/mL (INDME) by ultracentrifugation; samples were taken for 120 min P14. The supersaturation ratio for full release is $S=35.1/2.29=15.3$ for IND and $9.0/0.42=21.4$ for INDME R21·E4. After dissolution, the solid was filtered off, dried, examined by PXRD and assayed for retained drug, and then re-dissolved in fresh medium under the same conditions in a single run P14·P15. The authors describe the exchange between surface and solution as a displacement equilibrium (Equation 5, after Dening and Taylor):

In plain terms: drug on the surface can be displaced by solvent and drug in solution can displace adsorbed solvent, so the amount of drug held by the silica follows the state of the solution rather than being fixed E5·C47.

Why it works. If the unreleased drug is adsorbed in equilibrium with the solution, then the solid recovered after a release test should release again in fresh medium, and the drug it holds should be of the same order as the isotherm predicts at the dose concentration C38·C39. If instead the drug had crystallized in the pores, PXRD would show it and re-dissolution would be poor C36. Difference from earlier work. The ritonavir study proposed the equilibrium from one drug C4; here two drugs with different donors are tested, and the recovered solid is re-dissolved and compared with the isotherm R24·R26.

Experimental analysis

Solubility: the scale for everything else. Table A collects the solubilities on which the dose, the supersaturation ratios and the activity scale rest. Methylation lowers both solubilities: INDME is about five times less soluble as a crystal and about four times less soluble as an amorphous form R1. The UV and ultracentrifugation values agree within their standard deviations, which the authors take as confirmation of the amorphous solubility R1·C11. The amorphous form gives a 17.4-fold (IND) and a 22.5-fold (INDME) higher concentration than the crystal R2; from Table 1 the INDME ratio is 9.4/0.42 = 22.4 X17.

Table A. Solubilities from Table 1, with HPMC 100 µg/mL in the amorphous-solubility media; log P cited from Hansch and Leo R1·R2·M1·M2.

Two different amorphous solubilities are used later. The activity scale of Figure 5 uses 0.111 and 0.025 mM, which correspond to the UV values (39.9 and 9.4 µg/mL), while the release experiments were dosed at the ultracentrifugation values (35.1 and 9.0 µg/mL) P18·P14·X18.

Main result 1: both drugs bind silanols through their carbonyl groups. Table B gathers the spectroscopic evidence. In amorphous IND the authors assign 1682 cm⁻¹ to the benzoyl C=O, 1710 cm⁻¹ to the acid C=O of the hydrogen-bonded IND dimer, and a shoulder at 1732 cm⁻¹ to free acid C=O R3. On SBA-15 the first two bands move to 1669 and 1688 cm⁻¹ and the shoulder disappears, which the authors read as hydrogen bonding of IND to the silanols, "with the likely disruption of the indomethacin dimers" R3·C12. INDME, with no donor, shows the same behaviour: its benzoyl C=O moves from 1674 to 1663 cm⁻¹ and its ester C=O from 1735 to 1725 cm⁻¹ R3·C13. Blank SBA-15 does not absorb in this region R3.

Table B. Spectroscopic and competition evidence for carbonyl–silanol hydrogen bonding, from the text (Sections 4.2.1, 4.2.2, 4.3.3) and read from Figures 2, 3 and 7 R3·R4·R14.

Note by the AI writer (not discussed in the paper): when Figure 2 is read on its own axis, some loaded-IND peaks sit nearer about 1700 and 1676 cm⁻¹ than the printed 1688 and 1669 cm⁻¹, and the amorphous INDME benzoyl band nearer about 1683 than 1674 cm⁻¹; every carbonyl band still moves to lower wavenumber on silica, so the direction of the evidence stands while the size of some shifts is uncertain X1.

XPS points the same way. The Si 2p3/2 peak of SBA-15 at 103.8 eV moves to 103.4 eV with IND and 103.3 eV with INDME; the authors "believe" this reflects electron density passing from the drug carbonyls to the silica through hydrogen bonds R4·C14. The O 1s peak of both loaded samples also lies at lower binding energy than that of SBA-15 R5. In the pure drugs, the carbonyl oxygen gives a low-energy shoulder; on silica this shoulder is absent for INDME and weak for IND, which the authors read as both INDME carbonyls being hydrogen-bonded ("This suggests") and only some of the IND carbonyls ("suggesting") R5·C15. Urea, finally, cut adsorption of both drugs by about half at equal initial activity (p < 0.05), "further evidence" for hydrogen bonding R14·C29. The axis labels of Figure 7 read "0.1M Urea", while the methods and the caption say 5 M X8.

Main result 2: no hydrophobic contribution was found. In distilled water, 0.1 M NaCl and 0.5 M NaCl, IND adsorbed about 0.8–0.9 mg/g at 2 µg/mL and about 8–9 mg/g at 8 µg/mL; INDME adsorbed about 11 mg/g at 2 µg/mL and about 65–76 mg/g at 8 µg/mL (read from Figure 8, approximate) R15. None of the differences was significant (p > 0.05) R15. Since both drugs have log P above 4, hydrophobic binding to siloxane bridges had been "considered feasible"; the absence of a salt effect led the authors to conclude that neither drug binds SBA-15 hydrophobically ("suggests"; abstract "No evidence was obtained") C6·C31. Two details limit the comparison: the same concentrations were used for both drugs, which puts INDME at about four times the activity of IND, although the methods state the test was run at equal activity, and the media were water and NaCl solutions without a stated buffer or pH X9.

Main result 3: adsorption climbs steeply toward the amorphous solubility. For IND, adsorption stayed below about 30 mg/g up to an equilibrium concentration of about 17 µg/mL, reached about 100 mg/g at about 29 µg/mL and about 220–290 mg/g at about 35 µg/mL (Figure 4a) R7. INDME followed the same pattern over a fourfold shorter concentration range, from about 23 mg/g at 3.1 µg/mL to about 342 mg/g at 7.7 µg/mL (Figure 4b) R8. Even when the initial concentration exceeded the amorphous solubility, the equilibrium concentration stayed below it, which the authors attribute to the extent of adsorption; no crystallization occurred R9·C23. At the highest initial concentrations, adsorption began to exceed the theoretical monolayer, 0.287 g/g for IND and 0.269 g/g for INDME, drawn as dotted lines in Figure 4 R9·R16·C18.

Table C. Isotherm fits (Table 3) and the geometric monolayer (Equation 3, Section 4.4) R10·R11·R16.

Langmuir, which allows only one layer, fitted worst; Freundlich and BET, which allow multilayers, fitted better, and the authors read this as multilayer adsorption ("suggest") R10·C19·C43. The Langmuir and Freundlich constants were higher for INDME, which the authors read as stronger binding of INDME to silica ("suggests") R10·C20. The BET monolayer was 29 and 31 mg/g, and $1/K_L$, which has been suggested to equal the saturation solubility, came out at 40 and 8.5 µg/mL, close to the amorphous solubilities R10·R11·C21. The BET recomputation is exact: $1/0.025=40$ and $1/0.118=8.47$ E8.

Note by the AI writer (not discussed in the paper): three cautions apply to these numbers. First, the BET monolayer (about 30 mg/g) is a tenth of the geometric monolayer (287 and 269 mg/g); measured against the latter, the highest points (about 289 and 342 mg/g) are at, or only modestly above, one layer, so "multilayer" depends on which monolayer is meant X10. Second, with the stated 40 mL and 5 mg silica, Equation 1 caps the amount adsorbed at $C_i\times V/m$: for INDME at its highest initial concentration of 14 µg/mL that is 112 mg/g even if all drug adsorbed, yet Figure 4b shows values up to about 342 mg/g; IND's top point would need an initial concentration of about 71 µg/mL, above the stated 60 E1·X2. Some unstated difference in volume, silica mass or concentration basis must apply, so the absolute amounts are uncertain, while comparisons within a figure may still hold X2. Third, $K_L$ and $K_F$ were fitted on a concentration scale; INDME's fourfold lower solubility would by itself raise them, so the higher constants need not mean stronger specific binding X21.

Main result 4: on the activity scale the two isotherms nearly coincide. On the concentration scale the drugs look very different: at an equilibrium concentration of about 6–7 µg/mL, IND has adsorbed about 12 mg/g and INDME about 118–154 mg/g R7·R8·X16. Divided by their amorphous solubilities, the data fall close together (Figure 5): both stay below about 0.06 mmol/g up to an activity of about 0.35, reach about 0.21–0.24 mmol/g at an activity of about 0.6–0.66, and rise steeply above 0.8, to about 0.62–0.80 mmol/g for IND at about 0.88–0.89 and about 0.92 mmol/g for INDME at about 0.86 (read from Figure 5, approximate) R12. The authors call the two curves "nearly congruent", especially at low activity, as Hansen found for related solutes C24·R29. Their reading is that solute activity determines the extent of adsorption for "these two structurally related molecules" ("suggesting"), and that because the acceptor carbonyls, not the donor, "appear to be key", removing the donor "may explain" why the two adsorb alike C25·C42.

Desorption. After dilution to sink and 24 h, 73% of the adsorbed IND and 57% of the INDME had desorbed (Figure 6), leaving 0.98 ± 0.06 and 4.16 ± 0.16 mg/g on the silica R13. The authors read this as INDME having a stronger attraction to silica at low concentration ("indicating") and adsorption of both drugs as not completely reversible, although the amounts remaining were low C26·C27. The basis of the percentages and the dilution factor are not stated; with these percentages, the remaining amounts imply about 3.6 and 9.7 mg/g before dilution, less than the 12.4 and 14.2 mg/g found in the urea controls at the same nominal activity X13. "Equal activity" here refers to the initial concentrations, not to the equilibrium concentrations after adsorption X12.

Loaded formulations. Table D shows that the actual loads fell short of the targets, by more for INDME R17. The loading efficiency was significantly higher for IND at 25% and 50% of a monolayer (about 73–78% against about 57%), with no significant difference from 75% upward (Figure 9) R18·C32. PXRD showed both drugs X-ray amorphous at 25% and 50%; crystalline peaks appeared from 75% for IND, which the authors assign to the γ polymorph, and distinct INDME peaks from 100% (Figure 10) R19·C33. At these loadings "some drug may be present outside of the SBA-15 pores" C34. Figure 10 shows no reference patterns, so the polymorph assignment cannot be checked from the paper R19.

Table D. Drug loads from Table 4 (n = 3) and the PXRD outcome from Figure 10 R17·R19.

Note by the AI writer (not discussed in the paper): the printed INDME target at 100% (27.97%) does not fit the other INDME rows, which are all multiples of 26.96–26.97% per 100%, nor the text's 0.269 g/g; Figure 9 shows an INDME efficiency at 100% of about 58%, which matches 15.60/26.97 rather than 15.60/27.97, so 27.97 is probably a misprint X3. Also, the theoretical monolayer is in grams of drug per gram of silica, but it is used directly as "% w/w" drug load; as a fraction of the loaded powder, 0.287 g/g would be 22.3% w/w, and the basis of the measured load is not stated X4.

Main result 5: release stops at about half the possible supersaturation. The 50% formulations released drug in a burst (Figure 11). IND was at about 15.9 µg/mL at 5 min and about 16.5 µg/mL at 15–30 min, then about 15.7 µg/mL to 120 min; INDME rose from about 2.1 µg/mL at 5 min to about 2.9–3.0 µg/mL at 30–60 min and eased to about 2.6 µg/mL at 120 min (read from Figure 11, approximate) R20. The maximum concentrations were 16.6 µg/mL (IND) and 2.9 µg/mL (INDME), seven times the crystalline solubility for both (S = 7.2 and 7.0) R20. Full release would have given S = 15.3 and 21.4 R21. The authors note that, as a percentage of dose, IND released more (45% against 35%), but in terms of supersaturation the two drugs did equally well R20·C35.

Table E. Release from 50% monolayer formulations and from the recovered solid (text, Figures 11 and 13, Table 5) R20·R21·R23·R24·R25·R26.

Note by the AI writer (not discussed in the paper): the percentages in Table E do not reconcile. Cmax over dose gives 47% (IND) and 32% (INDME); the right-hand % axes of Figure 11 put 100% at about 40 and 9.5 µg/mL and so give about 41% and 31%, as does the graphical abstract; and released plus retained is 58.8 + 45 = 103.8% for IND (stated 103%) but 61.8 + 35 = 96.8% for INDME (stated 94%) X5. The Cmax and supersaturation values are the safer numbers X5.

Was the missing drug crystallized? PXRD of the solid filtered off after dissolution showed only an amorphous halo, as before dissolution, and no crystalline drug precipitated from solution (Figure 12) R22. The authors conclude that the retained drug stayed amorphous on the surface and that crystallization does not explain the incomplete release C36. The recovered solid held 58.8% (IND) and 61.8% (INDME) of the dose, which the authors find consistent with the measured release R23·C37.

Main result 6: the recovered solid releases again to the same level. The harvested solid carried about 60% of its original drug load, so for the same dose the vessel held more silica: 38.19 mg against 21.71 mg (IND) and 5.61 mg against 3.72 mg (INDME) R23·R25. It released again to Cmax 16.2 and 3.5 µg/mL, profiles the authors report as not significantly different from the first run (p > 0.05) (Figure 13) R24. They read this as "further evidence of the existence of a dynamic equilibrium" between adsorbed and free drug, and note that the extent of release did not change although the silica was "approximately double" C38. Finally, the drug left on silica after release, 0.2 mmol/g (IND) and 0.3 mmol/g (INDME), is "reasonably comparable" to the 0.38 and 0.41 mmol/g adsorbed in the isotherm experiments at an initial concentration equal to the amorphous solubility R26·C39.

Note by the AI writer (not discussed in the paper): Figure 13 is a single run (n = 1), so the significance test cannot apply as stated X15. Table 5 is hard to reconcile with the rest: its supersaturation ratios differ from the text (INDME original 8.2 against 7.0), 2.93 mg IND in 75 mL is 39.1 µg/mL rather than the stated 35.1 µg/mL, the implied loads (11.9% IND, 15.6% INDME) differ from the measured loads of the 50% formulations (10.46% and 7.65%), and the silica rose 1.76-fold and 1.51-fold rather than doubling X6·X7. The retained amounts are about 53% and 73% of the isotherm values, obtained at a different silica-to-volume ratio, so "reasonably comparable" holds to within a factor of about two X22.

The authors compare the result with ritonavir from SBA-15, where the theoretical supersaturation ratio was 7.8 and only 3.9 was reached; for all three drugs the supersaturation reached was 50% or less of the maximum set by the amorphous solubility, which they say the dynamic equilibrium "clearly" limits R27·C48.

Discussion

The authors' chain of reasoning. The paper's argument runs in five steps, which the diagram below lays out. Silanols donate hydrogen bonds to the drugs' carbonyl acceptors, so a drug donor is not needed and both analogues bind alike C41·C42. Binding alike, the two drugs adsorb according to their activity in solution, not their concentration C25·C53. As the activity approaches 1, adsorption rises steeply, which the authors read as multilayers and as pores becoming "partially filled with drug", with water "most likely co-adsorbed" C43·C44·C45. In a release test, adsorbed and free drug exchange in a dynamic equilibrium (Equation 5) C47. What happens next depends on whether anything removes drug from solution.

The second branch is the authors' speculation, not a measurement: in vivo the driving force for membrane transport is the degree of (super)saturation, so absorption should pull drug off the silica toward complete release, but "the average supersaturation is anticipated to be lower, which in turn is expected to lead to a reduced membrane transport rate"; the very rapid dissolution "may" compensate C50. They conclude that absorptive dissolution testing "is likely necessary" to separate these kinetic processes C51.

Where the activity rule might apply. The authors extend their finding cautiously. If adsorption onto silica follows solute activity, it "can be predicted to some extent" from the solubility of the drug, or from its amorphous solubility where supersaturation is sustained; so "a general trend would be expected whereby less soluble compounds will adsorb more than higher solubility compounds" at equal concentration, because the less soluble one sits at higher activity C46. They add that this supposition "clearly needs to be tested further" with drugs of various properties, and that literature data appear to support it C46. The evidence in this paper comes from two closely related molecules at one pH on one silica A1·M3.

Implications the authors draw for formulation. Because the three drugs so far examined (IND, INDME and ritonavir) reached 50% or less of the theoretical supersaturation, the authors see "a potential challenge" for mesoporous silica against polymer-based amorphous solid dispersions, which may reach higher supersaturation at an equivalent dose C48·C49. They also emphasise that the solute activity, rather than the concentration as such, is what matters for adsorption C53.

Note by the AI writer (not discussed in the paper): three practical readings follow if the authors' picture holds. First, designing a drug or prodrug without a hydrogen-bond donor will not reduce its loss to silica while carbonyl acceptors remain, since INDME, with no donor, adsorbed as much as IND on the activity scale R12·C42. Second, because the isotherm is steepest close to the amorphous solubility, each increment of dose in that region adds mostly adsorbed drug rather than supersaturation in a closed vessel R12. Third, a plateau in a closed non-sink test of a silica formulation reflects an adsorption equilibrium rather than a failure of release, so such tests favour polymer dispersions in a head-to-head comparison; a flux or absorptive test, as the authors suggest, is needed before ranking the two C50·C51.

Open questions. Several questions remain. The pH was 2.2 throughout, where IND is taken as unionized; at intestinal pH IND ionizes and the silanols are less protonated, which could bring in the electrostatic effects the design excluded C7. HPMC was present in every adsorption and release medium, and its own adsorption to silica, and any competition with the drug, were not examined M4. The absolute adsorbed amounts need a consistent volume and mass basis before the isotherms can be used quantitatively X2. Whether the steep part of the isotherm is truly multilayer adsorption, or reaches about one geometric monolayer, is open X10. No membrane-transport or absorptive-dissolution data were collected C50.

Limitations

Limitations stated or implied by the authors. The authors acknowledge that supersaturated solutions are not at true equilibrium, since crystallization can occur, and treat the adsorption state as a transient equilibrium that holds only while no crystallization occurs C9. They present the activity rule for other drugs as a supposition that "clearly needs to be tested further" C46. They describe the in vivo consequences as something that "can be speculated about", and state that absorptive dissolution testing "is likely necessary" C50·C51. They note that at loadings of 75% of a monolayer and above, some drug "may" lie outside the pores C34.

The AI writer's own judgement (1): what the mechanism evidence shows. The case for carbonyl–silanol hydrogen bonding rests on four independent but each indirect observations: carbonyl bands shift down, the Si 2p binding energy drops, the O 1s carbonyl shoulder disappears, and urea halves adsorption R3·R4·R5·R14. Together they are persuasive in direction. Their weak points are specific: some band positions in the text do not match Figure 2 X1; the loading of the spectroscopy samples is not stated X20; the urea concentration is labelled 0.1 M in Figure 7 and 5 M in the text X8. The case against hydrophobic binding is weaker: the salt test was not done at equal activity, its media had no stated buffer, and the INDME results at 8 µg/mL have large standard deviations (about 10–16 mg/g) X9·R15. The pyrene result supports multilayers only indirectly, because drug alone at the sub-LLPS concentrations used with silica was not measured X11. The claim that methyl esterification significantly lowers the melting point has no supporting data in the paper X14.

The AI writer's own judgement (2): the numbers behind adsorption and release. The paper's central observation, the overlap of the two isotherms on the activity scale, is clear in Figure 5 and robust to the scale of the y-axis R12. The absolute amounts are not: with the stated volume and silica mass, several plotted INDME values exceed what all the added drug could give, and the multilayer reading depends on a BET monolayer ten times smaller than the geometric one X2·X10. The release link is a correlation plus one re-dissolution run: the percentages of release, the supersaturation ratios and the amounts of Table 5 do not reconcile with each other or with the text, the re-dissolution was a single run reported with a significance test, and the agreement between retained drug and isotherm holds to within a factor of about two X5·X6·X7·X15·X22. None of these issues reverses the qualitative result: release stopped well short of the dose for both drugs, the retained drug stayed amorphous, and the recovered solid released again R20·R22·R23·R24. Overall, in the AI writer's judgement, the support is moderate.

Conclusion

The paper shows, for indomethacin and its methyl ester on SBA-15 at pH 2.2, that binding to the silica runs through the drugs' carbonyl acceptors and silanol donors, that removing the drug's hydrogen-bond donor does not reduce adsorption, and that the two drugs adsorb alike when compared at equal solute activity C41·C42·C24. Adsorption rises steeply as the solution approaches the amorphous solubility, and formulations dosed at that concentration released only part of their drug, reaching a supersaturation of 7.2 and 7.0 against 15.3 and 21.4 for full release, with about 60% of the dose remaining amorphous on the silica R7·R8·R20·R21·R22·R23. The authors conclude that the incomplete release "can be rationalized" by this strong adsorption in dynamic equilibrium with the free drug, and that solute activity, rather than concentration per se, is the quantity to watch C52·C53. Whether this costs absorption in vivo, and whether less soluble drugs generally adsorb more, are left for absorptive dissolution testing and for studies of more drugs C50·C51·C46.