How the bonds a drug forms with copovidone set the loading at which its dispersion stops releasing well

Background and motivation

An amorphous solid dispersion (ASD) is, ideally, a molecular-level amorphous mixture of a poorly soluble drug and a polymer, and it has become one of the preferred ways to raise the absorption of such drugs. In some drug–polymer systems, however, something more happens: the polymer controls the drug release rate, which can then be much faster than that of the amorphous drug alone, and the drug concentration can exceed the amorphous solubility (Indulkar 2019, ref 1; Saboo 2019, ref 2). Water-saturated particles of the neat amorphous drug can only reach the amorphous solubility, which is the highest free-drug concentration possible without crystallization. Above that level the solution separates, by nucleation and growth or by spinodal decomposition, into a disordered drug-rich phase that is initially nanometre-sized.

These drug-rich colloids matter because there is "emerging evidence" that they improve bioavailability compared with formulations that do not produce them (refs 3–5, including Wilson 2018 and Kesisoglou 2019). It has been suggested that the colloids act as a reservoir that keeps the solution at maximum supersaturation by replacing drug removed by absorption (Indulkar 2016, ref 6), and that they may drift into the unstirred water layer next to the gut wall (Sugano 2010, ref 7). The catch is that the colloids form only when release is polymer-controlled, and that happens only below a threshold drug loading (DL). For copovidone (PVPVA) ASDs, at or below that loading drug and polymer leave the compact together, at about the rate of the neat polymer, which can be orders of magnitude faster than the neat amorphous compound (refs 1, 2, 8). Above it the polymer leaves faster than the drug, a drug-rich layer builds up on the surface, release becomes drug-controlled, slow and incomplete, and typically no colloids form.

The loading that marks this switch from congruent to incongruent release has been called the limit of congruency (LoC) (Indulkar 2019, ref 1). The transition happens over a narrow range of DL, and the drop in performance is abrupt. Because higher loadings reduce pill burden, a high LoC is desirable. Yet the LoC is specific to each drug–polymer pair: for PVPVA it has been reported between 5 and 25% DL (refs 1, 2, 8), "with little understanding" of how drug and polymer properties produce this pattern.

Drug–polymer intermolecular interactions were an obvious candidate. They are known to shape other ASD properties. Hydrogen bonding improved drug–polymer miscibility and the processability of polyvinylpyrrolidone (PVP) hot-melt extrudates (Chan 2015, ref 9), and it slows drug crystallization in the ASD matrix (refs 10–13, including Rumondor 2009). Ionic interactions delay crystallization by lowering molecular mobility (Mistry 2015, ref 14). Hydrophobic drug–polymer interactions that persist in water help maintain supersaturation after dissolution (Chen 2019, ref 15; Ilevbare 2013, ref 16). And the same group had just shown that drugs carrying halogen atoms can form halogen bonds with the vinylpyrrolidone (VP) carbonyl of PVPVA in ASDs (Que 2020, ref 17). What had hardly been studied was how any of these interactions affects the LoC. The authors call this "a knowledge gap for ASD formulation design".

The paper sets out to close part of that gap with a deliberately narrow chemical design. Four compounds share one phthalide (phenolphthalein) scaffold but differ in two kinds of donor group: phenolic OH groups, which can hydrogen-bond to the PVPVA carbonyl, and iodine atoms, which can halogen-bond to the same acceptor. TIPP carries both, PHPH only OH, Me-TIPP only iodine, and Me-PHPH neither (Figure 1). PVPVA was chosen for its wide use in commercial ASDs (refs 18, 19). Interactions were read by infrared (IR) spectroscopy and X-ray photoelectron spectroscopy (XPS), release by surface-normalized dissolution, and colloids by dynamic light scattering (DLS) and nanoparticle tracking analysis (NTA). A second aim was to see how hydrogen and halogen bonds compete in a compound that can form both.

Background concepts

Congruent release and the limit of congruency. In a release experiment on an ASD compact, both components are assayed: the drug and the polymer. Each is expressed as a percentage of its own total amount in the compact (Eq. 1 of the paper):

Here $m_{\text{released}}$ is the amount of a component (drug or PVPVA) found in the medium at a given time and $m_{\text{total}}$ is the total amount of that component in the compact. In plain terms: the drug curve and the polymer curve are put on the same 0–100% scale, so that if drug and polymer leave the compact in the same proportion as they were mixed, the two curves overlap. That overlap is what "congruent release" means. In this paper the drug amount includes both molecularly dissolved drug and drug held in colloids, because the colloids were dissolved with an organic solvent before the high-performance liquid chromatography (HPLC) assay. The LoC is defined operationally as the highest DL at which drug and polymer released simultaneously and at least 80% of the drug was released in 90 min.

Amorphous solubility and drug-rich colloids. The amorphous solubility is the highest concentration of free drug a solution can hold before a drug-rich amorphous phase separates. The paper measures it by the solvent-shift method: a concentrated drug solution in an organic solvent is pumped at a steady rate into buffer, and the onset of light scattering (extinction at 360 nm, a wavelength the drug does not absorb) marks the point where a second phase appears. When an ASD releases drug faster than the amorphous solubility can accommodate, the excess drug forms nanometre-sized drug-rich colloids, and the medium turns turbid.

Hydrogen bonds and halogen bonds. A hydrogen bond here is the attraction between the O–H of a phenol (the donor) and the oxygen of a carbonyl group (the acceptor). A halogen bond is the attraction between the electron-poor tip of a halogen atom, called the sigma hole, and an electron-rich atom such as oxygen or nitrogen. Iodine is in general the best halogen-bond donor among the halogens. Both bond types need the same electron-rich acceptor, so where both are possible, competition is "probable, although they can coexist" (Aakeröy 2007 and 2015, refs 29, 32). Crystal engineering studies show that hydrogen bonding is generally favoured over halogen bonding, but halogen bonds tend to persist more in relatively polar environments, where hydrogen bonds are less favoured (Wilcken 2013, ref 33; Robertson 2017, ref 34). PVPVA offers two carbonyl acceptors: the C=O of the vinylpyrrolidone (VP) unit and the C=O of the vinyl acetate (VA) unit, present at a VP/VA mass ratio of 60:40.

Reading bonds from infrared spectra. A carbonyl group absorbs infrared light at a characteristic wavenumber. When the carbonyl oxygen accepts a hydrogen or halogen bond, the C=O bond weakens slightly and its band moves to lower wavenumber; the size of the shift is a rough, qualitative measure of how strong the interaction is. In dry PVPVA the VP carbonyl band sits at about 1684 cm⁻¹ and the VA carbonyl band at 1736 cm⁻¹. The model compounds also carry their own carbonyl, in the phthalide ring of the scaffold. In the neat compounds it can accept bonds from neighbouring drug molecules; when the polymer separates the drug molecules, those bonds break and a "free" drug carbonyl band grows or moves up.

Reading halogen bonds from X-ray photoelectron spectra. XPS measures the binding energy (BE) of core electrons in atoms within roughly the top 10 nm of a surface. For iodine the 3d level splits into two peaks, 3d_5/2 and 3d_3/2, in an intensity ratio of 3:2; this pair is called a doublet. If an iodine atom gains electron density, for example because its sigma hole is engaged by a carbonyl oxygen, its core electrons are held slightly less tightly and the doublet appears at lower binding energy. A second, lower-energy doublet in a spectrum therefore marks a population of iodine atoms in a more electron-rich environment, and the share of the iodine signal in that doublet can be converted into a count of such atoms.

Glass transition and how a glassy polymer dissolves. The glass transition temperature ($T_g$) is the temperature at which an amorphous solid changes from a rigid glass to a rubbery state; dry PVPVA has a $T_g$ of 104 °C (ref 24). The paper's Discussion uses a standard picture of how a glassy polymer dissolves (refs 15, 45–47): water penetrates and plasticizes it into a gel, and the chains then disentangle and diffuse away. The solvent quality for the polymer, often expressed by the Flory–Huggins χ parameter, governs both steps. Neat PVPVA sorbs so much water at high water activity that its $T_g$ falls below room temperature (Taylor 2001, ref 43), so in water it gels and dissolves quickly.

Methods in detail

The work splits into four contributions. The first is the compound set: four analogues on one scaffold, characterized for the bulk properties that might otherwise explain differences in release. The second reads the drug–polymer interactions in the dry solid by IR of films and XPS of compacts, including the competition between hydrogen and halogen bonding in TIPP. The third is the release study, in which drug and polymer are assayed side by side from surface-normalized compacts to locate each LoC, with the colloids sized and imaged. The fourth ties these together: the LoC per mole, a test of polymer control at the LoC, and a check of the LoC against log P, amorphous solubility, $T_g$ and molecular weight (MW).

Contribution 1: A matched set of four probe compounds

① Problem. Earlier LoC values for PVPVA came from structurally unrelated drugs, which differ at once in size, lipophilicity, $T_g$, solubility, ionization and interaction ability. To ask whether the type of drug–polymer interaction matters, the authors needed compounds that change the interaction while keeping as much else as possible the same.

② What they did. TIPP (3′,3″,5′,5″-tetraiodophenolphthalein) and PHPH (phenolphthalein) were bought from Sigma-Aldrich. Me-TIPP and Me-PHPH, in which both phenolic OH groups are converted to methyl ethers, were synthesized in-house. For Me-PHPH the SI gives the procedure: the bisphenol was O-methylated with iodomethane and potassium carbonate in dimethylformamide overnight, giving a yellow thick oil in 96.2% yield, identified by nuclear magnetic resonance (NMR) and high-resolution mass spectrometry. Me-TIPP was made in the earlier study (ref 17). Glass transition temperatures were measured by differential scanning calorimetry (DSC; four heat–cool cycles at 10 °C/min under nitrogen, midpoint of the last heating). TIPP crystallizes too fast to be made amorphous on its own, so its $T_g$ was measured on a 90/10 TIPP/PVPVA dispersion, and the neat value was estimated with the Fox equation (ref 25). The paper names this equation without printing it; its standard form, which the card uses to check the value, is:

Here $T_{g,\mathrm{m}\mathrm{i}\mathrm{x}}$ is the glass transition temperature of the mixture, $w_1$ and $w_2$ are the mass fractions of the two components, $T_{g,1}$ and $T_{g,2}$ are their own glass transition temperatures, and all temperatures are in kelvin. In plain terms: the reciprocal $T_g$ of a mixture is the mass-weighted average of the reciprocal $T_g$ values of its parts. With 86.1 °C for the 90/10 dispersion and 104 °C for PVPVA, solving for neat TIPP gives 84 °C, "slightly lower" than the measured mixture value, as the authors expect. Amorphous solubility was measured by solvent shift at 37 °C in 50 mM phosphate buffer containing 1 mg/mL PVPVA, the highest polymer concentration reachable in the dissolution tests. The pH was 6.8 for three compounds and 2.5 for TIPP, chosen so that every compound stayed un-ionized (at least 2 units below the calculated pKa). Log P and pKa were calculated, not measured. Dispersions were made by dissolving compound and polymer in tetrahydrofuran (THF; TIPP, Me-TIPP; water bath 40 °C) or ethanol (PHPH, Me-PHPH; 60 °C), removing the solvent on a rotary evaporator and drying for 24 h under vacuum. Powder X-ray diffraction (PXRD) of the highest-loading dispersions (50% DL for PHPH and TIPP, 70% DL for Me-TIPP and Me-PHPH) showed only an amorphous halo (SI Figure 6S).

③ Why it works. The four compounds form a two-by-two design on the same scaffold: hydrogen-bond donor present or absent, halogen-bond donor present or absent (Figure 1). All four keep the same phthalide carbonyl and the same overall shape, and all are dispersed in the same polymer by the same route and tested in the same release set-up. If the LoC changes sharply when only the donor groups change, the donor groups are the first suspects. Me-PHPH, with no hydrogen- or halogen-bond donor, serves as the control: it can interact with PVPVA "primarily via nonspecific dispersive interactions".

④ How it differs from earlier work. The closest precedent was a single pair, indomethacin and its methyl ester, in which removing the hydrogen-bond donor gave a much higher LoC (Saboo 2020, ref 42). The present set adds a second kind of specific interaction, the halogen bond, and a compound that can form both, so that the two bond types can be compared and their competition examined. Note by the AI writer (not discussed in the paper): the substitutions also change MW, log P, $T_g$ and amorphous solubility, and methylation removes the drug–drug hydrogen bonds of TIPP and PHPH. Contribution 4 checks the first set of confounds, but not the last.

Contribution 2: Reading the interactions in the dry solid with IR and XPS

① Problem. The compound design predicts which bonds each compound can form, but whether they actually form in the dispersion, and how much, has to be shown. For TIPP, which carries both donors, the question is also which bond wins when both compete for the same carbonyl acceptors.

② What they did. For IR, solutions of polymer or compound plus polymer (50 mg/mL solids in THF or ethanol) were spin-coated onto thallium bromoiodide (KRS-5) windows, 15 s at 50 rpm then 50 s at 2500 rpm. This was done in a humidity-controlled glovebox "to prevent moisture-induced phase separation". Films at 10 to 90% DL in steps of 10 were measured in transmission (128 scans, 4 cm⁻¹ resolution), and the carbonyl region (1780–1620 cm⁻¹) is shown in Figure 2. The Me-TIPP spectra are adapted from the earlier paper (ref 17). For XPS, compacts of TIPP/PVPVA and Me-TIPP/PVPVA at 10, 20, 30 and 40% DL were made as for dissolution and ejected from the die, together with samples of the neat compounds. Spectra were taken on a Kratos Axis Ultra DLD instrument with monochromatic Al Kα radiation (1486.6 eV), at a pass energy of 20 eV (resolution about 0.35 eV), at five spots per tablet. The I 3d spectra were fitted with a doublet like that of the neat compound plus a second doublet at lower binding energy (SI Figure 4S), and the share of iodine in the lower-energy doublet was tabulated (SI Table 1S). The authors then converted this share into a molar amount of lower-energy iodine in 100 mg of ASD (Figure 3). The paper states this conversion only in words; the card reproduces the plotted values with:

Here $n_{\mathrm{I},\mathrm{l}\mathrm{o}\mathrm{w}}$ is the amount of lower-energy iodine in mmol per 100 mg of ASD, DL is the drug loading as a mass fraction, $M$ is the molar mass of the compound in g/mol, 4 is the number of iodine atoms per molecule (Figure 1), and $f_{\mathrm{l}\mathrm{o}\mathrm{w}}$ is the fraction of the iodine signal in the lower-energy doublet (Table 1S). In plain terms: the mass of drug in 100 mg of dispersion is turned into moles of drug, then into moles of iodine atoms, and then multiplied by the share of those iodine atoms that look electron-rich. For TIPP at 10% DL, 10 mg divided by 821.91 g/mol, times 4, times 0.559, gives about 0.027 mmol, which matches Figure 3.

③ Why it works. The two methods look at the same bond from opposite ends. IR reads the acceptor: a shift of the VP carbonyl band shows that this group accepts a bond, and an unchanged VA band shows that it does not. It also reads the drug carbonyl, so that drug–drug bonds broken by the polymer can be seen. XPS reads the halogen-bond donor directly: only iodine that has gained electron density moves to lower binding energy. A halogen bond would show up in both; a hydrogen bond only in IR. Comparing TIPP with Me-TIPP in XPS then measures how much halogen bonding TIPP gives up when its OH groups are also present.

④ How it differs from earlier work. The earlier paper (ref 17) established that Me-TIPP halogen-bonds to the VP carbonyl. Here the same methods are applied to a compound with both donors, so that competition can be quantified, and the solid-state interaction classes are carried forward to release rather than to physical stability, where drug–polymer interactions are usually discussed. The paper leaves open whether the Me-TIPP XPS rows are new measurements, and what solid form the "neat amorphous" TIPP reference had, since neat amorphous TIPP could not be made (story notes X7, X9).

Contribution 3: Releasing drug and polymer side by side to locate the LoC

① Problem. Locating an LoC needs three things: a release test in which the surface area is fixed, so that rates can be compared between loadings and compounds; separate assays for drug and polymer, so that congruence can be judged; and a drug assay that counts drug in colloids as released, so that the very phase separation that signals good performance does not hide it.

② What they did. About 100 mg of ASD powder was compressed at 1500 psi for 60 s in an 8 mm die, leaving a single face of 0.5 cm² exposed (SI Figure 5S). The die was mounted on a paddle in an Agilent intrinsic dissolution assembly and rotated at 100 rpm in 100 mL of 50 mM phosphate buffer at 37 °C. The medium was pH 6.8 for Me-TIPP, PHPH, Me-PHPH and neat PVPVA, and pH 2.5 for TIPP "to suppress ionization". Samples of 3 mL were taken at 10, 20, 30, 40, 50, 60 and 90 min and replaced with fresh medium. For the drug, part of each sample was mixed with dimethyl sulfoxide (DMSO; iodo compounds) or ethanol (non-iodo compounds) to dissolve any colloids and assayed by reversed-phase HPLC. For the polymer, the rest was filtered through a 0.2 μm filter and assayed by size-exclusion HPLC (Table 1). The loadings tested were 10, 15 and 20% for TIPP; 5, 10 and 15% for PHPH; and 15, 20, 30, 40 and 50% for Me-TIPP and Me-PHPH (Figure 4 and SI Figure 3S), with neat PVPVA as the reference (SI Figure 5S). The number of replicate runs is not printed (story note X13). Colloids in 30-min samples were sized by DLS (backscatter at 173°, triplicate, polydispersity index below 0.3) and imaged by NTA (a single image of a 100 by 80 μm field).

③ Why it works. Because only one face of known area is exposed, the release rate reflects the material rather than the particle size or the tablet geometry. Because drug and polymer are both expressed as percent of their own totals, congruent release appears directly as two overlapping curves, and incongruent release as a polymer curve that runs ahead of the drug curve. Because colloids are dissolved before the drug assay, drug held in colloids counts as released. The amorphous-solubility line on each panel of Figure 4 shows how far release exceeds what dissolved drug alone could hold.

④ How it differs from earlier work. The set-up follows the group's earlier LoC studies of ritonavir, ledipasvir and other drugs in PVPVA (refs 1, 2, 8). What is new is not the test but the series of compounds put through it. Because the hydrogen-bonding compounds were tested at 5–20% DL and the others at 15–50% DL, each LoC is known only to within one step: 5–10 wt% for PHPH, 10–15 for TIPP, 30–40 for Me-TIPP and 40–50 for Me-PHPH (story note X11).

Contribution 4: Normalizing the LoC and testing for polymer control

① Problem. The iodo compounds weigh about two and a half times as much as the non-iodo ones, so at the same weight loading they supply far fewer molecules per gram of polymer. And the claim that release at the LoC is polymer-controlled needs a direct test.

② What they did. The authors expressed each LoC as moles of compound per 100 g of ASD (Figure 6b), stated in words and reproduced by the card as:

Here ${\mathrm{L}\mathrm{o}\mathrm{C}}_{\mathrm{w}\mathrm{t}}$ is the LoC in grams of compound per 100 g of ASD (numerically the weight-percent loading), $M$ is the molar mass in g/mol, and ${\mathrm{L}\mathrm{o}\mathrm{C}}_{\mathrm{m}\mathrm{o}\mathrm{l}}$ is in mol per 100 g of ASD. In plain terms: the weight loading at the LoC is divided by the molar mass to count how many molecules the polymer can carry before release fails. For Me-PHPH, 40 divided by 346.38 gives 0.115. They then compared the normalized release rates (mg/min/cm²) of drug and polymer from each ASD at its LoC, and of neat PVPVA, by two-sample t tests assuming unequal variances, at a 99% confidence level (Figure 7). How the rates were derived from the release curves, over which time window and with what normalization, is not described (story note X6). Finally, they set the LoC beside the bulk properties of the compounds: log P, amorphous solubility, $T_g$ and MW (Table 2).

③ Why it works. A bond is formed between molecules and functional groups, not between grams, so if interactions set the LoC, the mole basis is the scale on which the ranking should be judged. If drug and polymer at the LoC leave at the rate of neat PVPVA, then the polymer's dissolution is the rate-limiting process, and the question of the LoC becomes: at what loading does the drug start to slow the polymer down? The comparison with bulk properties rules alternative explanations in or out: if the LoC tracked log P or $T_g$, the interaction story would not be needed.

④ How it differs from earlier work. The earlier two-compound PVPVA study found no trend between the LoC and log P (Saboo 2019, ref 2). Here four compounds are compared, the mole-based LoC is introduced alongside the weight-based value, and the polymer-control claim is tested statistically.

Experimental analysis

Setup in brief. All experiments used the four compounds with PVPVA 64: IR on dry films (10–90% DL), XPS on compact surfaces (iodo compounds, 10–40% DL), and 90-min release from rotating compacts at 37 °C (pH 6.8; TIPP pH 2.5), with colloids examined after 30 min.

Table A. Properties of the four compounds and their LoCs in PVPVA (from Table 2 and Figure 6; log P calculated; Tg and amorphous solubility as mean ± standard deviation (SD), n = 3; amorphous solubility at 37 °C with 1 mg/mL PVPVA, pH 6.8 unless stated).

Main result 1: no bulk property tracks the LoC, but the donor groups do. Table A shows why the authors conclude that log P, amorphous solubility, $T_g$ and MW "do not show obvious correlations" with the LoC. The two iodo compounds share an amorphous solubility below 0.01 μg/mL and the highest calculated log P, yet their LoCs are 10 and 30 wt%. The two highest LoCs belong to the compounds with the lowest (Me-PHPH, 25.2 °C) and the highest (Me-TIPP, 122.7 °C) $T_g$. On lipophilicity the authors write that "it is clear that LoC does not correlate", in line with the earlier two-compound study (ref 2). What does separate the groups is the phenolic OH: the two compounds that carry it (PHPH and TIPP) have LoCs of 5 and 10 wt%, the two that do not (Me-TIPP and Me-PHPH) 30 and 40 wt%. On a mole basis the ranking is TIPP 0.012 < PHPH 0.016 < Me-TIPP 0.035 < Me-PHPH 0.115 mol per 100 g ASD, and the authors' wording is that interaction ability "appears to trend" with this ranking. They point out, "Interestingly", that TIPP, the compound able to form both bond types, has the lowest per-mole LoC.

The glass transitions of the dispersions (SI Table 2S) add to the $T_g$ argument. Me-PHPH lowers the $T_g$ of PVPVA strongly, to 76.1, 66.5 and 56.0 °C at 30, 40 and 50% DL, yet it gives the highest LoC. The authors judge that "it seems unlikely" that the high LoC comes from the low $T_g$ "per se", because other low-$T_g$ compounds in PVPVA did not give high LoCs (refs 2, 41) and the high-$T_g$ Me-TIPP also has a high LoC. The wet $T_g$, which they suggest may matter more (ref 42), was not measured. Note by the AI writer (not discussed in the paper): the PHPH dispersion at 10% DL has a $T_g$ of 108.3 °C, above both neat PHPH (99.0 °C) and PVPVA (104 °C). A mixture $T_g$ above both components is often read as a sign of strong specific interaction between them, which would fit the IR evidence below. Table 2S gives no value for the PHPH LoC loading (5%) or for Me-TIPP at 40% (story note X10).

Table B. Solid-state interaction evidence (IR from Figure 2 and Section 4.2.1; iodine shares from SI Table 1S, mean values; molar amounts read from Figure 3, approximate). The VA carbonyl band at 1736 cm⁻¹ did not change notably in any series.

Main result 2: which compound forms which bond. Both phenols move the VP carbonyl band by up to 30 cm⁻¹ while the VA band stays put, so the VP carbonyl is the binding site; the authors write that the shift "confirms" hydrogen bonding. The Discussion gives the same shift as "from 1684 to 1650 cm⁻¹", that is 34 cm⁻¹, a difference at the precision of reading the band (story note X1). Me-TIPP shifts the band by at most 7 cm⁻¹, "suggesting" a weaker halogen bond, and Me-PHPH barely shifts it ("it is clear" that it forms no directional interactions). The drug carbonyl bands show the other side. Neat TIPP has bands at 1772 and 1756 cm⁻¹, "hypothesized" to be halogen-bonded and hydrogen-bonded carbonyl, and the 1756 band weakens as the loading falls. In PHPH a small non-bonded band at 1765 cm⁻¹ grows as polymer is added, while Me-PHPH has a single band at 1765 cm⁻¹ that hardly changes. The authors' reading is that drug–polymer bonds form "at the expense of" drug–drug hydrogen bonds.

XPS shows the halogen bond from the iodine side. The neat compounds give one I 3d doublet at 620.7 ± 0.01 and 632.2 ± 0.04 eV; the second peak is printed as "I 3d1/2", a misprint for 3d_3/2 (story note X8). Dispersed in PVPVA, both iodo compounds add a doublet about 2.6 eV lower, at 618.1 ± 0.05 and 629.6 ± 0.05 eV, which "suggests" more electron-rich iodine, "likely" from I···O=C halogen bonds to the VP carbonyl. Table B shows two trends. The share of iodine in this state falls as loading rises in both compounds; the paper does not comment on this, and the AI writer reads it as fewer polymer carbonyls being available per drug molecule. And above 10% DL TIPP holds only about half as much halogen-bonded iodine as Me-TIPP: about 0.017–0.026 against 0.042–0.050 mmol per 100 mg ASD. The authors attribute this, "most likely", to competition between the OH and iodine groups of TIPP for the same carbonyls. In the 60% DL TIPP spectrum the original VP band shifts slightly, as with Me-TIPP, while a new band appears near 1650 cm⁻¹, so both bonds "appear to be present", but hydrogen bonding "seems to be more prevalent, especially at higher DLs". The ordering of bond strength rests on the size of IR shifts; no bond energies or association constants were measured.

Table C. Percent release at 90 min around each LoC (read from Figure 4, approximate; the text values are given where the text states them). Neat PVPVA reached about 105% at 90 min (SI Figure 5S).

Main result 3: the release cliff. At low loadings drug and polymer leave together at a percent rate similar to neat PVPVA, which released about 95% by 60 min and about 105% by 90 min (the text says "100% release after 60 min"; story note X2). Lower loadings of Me-TIPP and Me-PHPH (15 and 20%) also release about 84–105% drug and 92–108% polymer (SI Figure 3S). Above the LoC, Table C shows how steep the fall is. TIPP one step above its LoC (15%) and Me-TIPP two steps above (50%) release no detectable drug in 90 min, and even the polymer is held back to about 13–18%, so the failure is not only a drug lagging behind a dissolving polymer. PHPH at 10% and Me-PHPH at 50% release some of both, at "notably slower" rates; PHPH at 10% ends close to its amorphous-solubility line (about 28% release equivalent). At or below each LoC the drug concentration exceeded the amorphous solubility and the medium turned turbid, a "visual indication" of drug-rich colloids. Me-PHPH at 40% is a borderline pass: about 82% drug against about 102% polymer, with visible undissolved compound left in the holder (story note X11). The authors conclude that "it appears" that small structural changes, "such as methylation of a phenol group", drastically change ASD dissolution.

Table D. Release rates at the LoC (read from Figure 7, approximate; p-values printed, two-sample t tests with unequal variances) and size of the drug-rich colloids after 30 min (Table 3, mean ± SD, n = 3).

Main result 4: at the LoC the polymer sets the pace. In Table D every drug and polymer rate at the LoC lies between about 2.8 and 3.7 mg/min/cm², against about 3.0 for neat PVPVA, and none of the differences is significant at the 99% level. The authors read this as showing that the compounds' differences "appear to have no impact" on release below the LoC, "indicating that the polymer controls the compound release". They go further: the system is "agnostic" to the compound, drug release "can be predicted" from polymer release, and undesirable drug properties are "completely masked" by the formulation. Two cautions from the card apply here. For TIPP the drug–polymer difference has p = 0.022, which would be significant at the more usual 95% level (story note X4). And TIPP was dissolved at pH 2.5 while the neat-polymer reference ran at pH 6.8, so its polymer-versus-neat comparison spans two media (story note X5).

Colloids. The iodo compounds gave colloids about half the size of the non-iodo ones: 179 and 193 nm against 348 and 364 nm. The authors call them "clearly" smaller and note that smaller colloids "can potentially" improve absorption through faster replenishment of the dissolved drug (Kesisoglou 2019, ref 4). NTA images (Figure 5) were sparse for TIPP at 10% and PHPH at 5% DL but oversaturated for Me-TIPP and Me-PHPH at 30% DL, which the authors attribute to the higher loadings; no particle counts are given. The Methods describe the DLS samples as "just below the LoC" and the Discussion as "at their respective LoCs", but Table 3 uses the LoC itself for three compounds and 30% DL, 10 wt% below the LoC, for Me-PHPH (story note X3).

Controls and what they rule out. The design contains two methylation pairs and two iodination pairs. Removing the OH groups raises the LoC from 5 to 40 wt% (PHPH to Me-PHPH) and from 10 to 30 wt% (TIPP to Me-TIPP). Adding iodine changes the weight-based LoC much less (PHPH 5 to TIPP 10 wt%; Me-PHPH 40 to Me-TIPP 30 wt%). Together with Table A, this rules out lipophilicity, amorphous solubility and dry $T_g$ as sufficient explanations within this set. Note by the AI writer (not discussed in the paper): on the mole basis, the halogen-only compound sits between the groups rather than with the non-interacting one. Me-TIPP (0.035 mol per 100 g) carries about a third as many molecules at its LoC as Me-PHPH (0.115) and about twice as many as PHPH (0.016). Adding iodine to the non-interacting scaffold thus costs roughly 70% of the per-mole capacity, which fits the authors' phrase "strength and extent" better than a simple hydrogen-bond-or-not rule. What the controls do not rule out is that methylation also removes drug–drug hydrogen bonds (Table B, drug carbonyl bands), so the effect of the OH group cannot be assigned uniquely to drug–polymer bonding (story note X12).

Discussion

Where the result applies. The finding is a comparison within one series: four close analogues, one polymer, one preparation route and one release test, with all compounds un-ionized and the interactions read in the dry solid. Within those limits the pattern is large and consistent. The authors also cite an independent pair that points the same way, indomethacin, which hydrogen-bonds to PVPVA, against its methyl ester, which does not and has a much higher LoC (Saboo 2020, ref 42).

The proposed mechanism. Since release at the LoC runs at the neat-polymer rate, the authors turn to how PVPVA dissolves. Neat PVPVA sorbs water, gels and dissolves fast, but its dissolution changes once a critical amount of a small molecule is dispersed in it, and "it is apparent" that additives that hydrogen-bond to the polymer have more effect than weakly interacting ones, "at least" above 10 wt%. "We can rationalize" this, they write, as follows. A drug–polymer bond, "if it persists in the presence of water", occupies a site where water would otherwise solvate the polymer, so the matrix takes up less water. Lower water uptake by hydrogen-bonded dispersions has been seen in water-vapour sorption, compared with the physical mixtures (Crowley 2002, ref 44). Less water would mean less plasticization, so slower water penetration, and slower chain disentanglement; both are "anticipated" to slow polymer dissolution and, with it, drug release. This "could explain" the low LoCs of PHPH and TIPP. The chain is sketched below; its middle steps were not measured in this paper.

What is unresolved. No water uptake, gel-layer or wet-state spectroscopy was collected, so whether the hydrogen bonds survive hydration, and whether they lower water uptake in the medium, remains open. Methylation removes drug–drug hydrogen bonds too, so drug–drug bonding is an alternative cause that the design cannot exclude (story note X12). Note by the AI writer (not discussed in the paper): the authors' own background cites work showing that halogen bonds tend to persist more than hydrogen bonds in polar environments (refs 33, 34). Under a persistence argument this would lead one to expect the halogen bonds of Me-TIPP to hinder hydration at least as much, yet Me-TIPP has a high weight-based LoC. Its weaker bond, its fewer molecules per gram, or both may account for this; the paper does not address the point. Finally, whether the smaller colloids of the iodo compounds matter in vivo was not tested.

What it suggests for formulation. The authors' advice is to formulate PVPVA dispersions "ideally" below the LoC, and they note that a higher LoC allows a higher loading and a smaller dose form. Their central message is that hydrogen bonding, usually welcomed because it improves physical stability, "may be detrimental" to release, while "perhaps counterintuitively" weak or absent specific interactions allow good release at much higher loadings. The AI writer's reading (not stated in the paper): a strong carbonyl IR shift with PVPVA flags a possibly low LoC, not only a stability benefit, so stability and release should be screened together; LoCs of candidates that differ much in MW should also be compared per mole; and for a drug with a phenol-like donor, loadings of 5–15 wt% deserve fine steps.

Limitations

Limitations stated by the authors. The authors state few limitations explicitly, but they hedge the weakest links. The mechanism is conditional: the bond must persist "if" water is present, and the chain "could explain" the low LoCs. The effect of $T_g$ "is still elusive"; "it is possible" that the wet $T_g$ of the compound or of the hydrated ASD is the more relevant one, and "more studies around Tg impact are clearly needed". The mechanisms of any in vivo effect of the colloids are "still under investigation", and colloid size "may be" another factor to consider. The XPS quantification assumes a homogeneous mixture within the roughly 10 nm information depth. The conclusions are worded as suggestions: the LoC "may depend on the strength and extent" of specific interactions, and interactions "may be a dominant factor" in release.

The AI writer's own judgement (1): what the design can and cannot attribute. The card rates the overall evidence as moderate. What holds: IR and XPS agree on the interaction classes, the LoC gap between the two groups is large and survives the mole basis, and the derived numbers (the Fox estimate, the mole-based LoCs, the XPS molar amounts) reproduce. What weakens it: only four compounds and one polymer were tested, and replacing the phenolic OH removes both drug–polymer and drug–drug hydrogen bonds, so the lower LoCs can be attributed to the phenolic OH but not uniquely to drug–polymer bonding (story note X12). The water-uptake mechanism has no direct data in the paper. Bond strength was not measured; "strong" and "weak" are inferred from the size of IR shifts.

The AI writer's own judgement (2): how precise the LoC values are. Each LoC is known only to the tested step (story note X11), and Me-PHPH at 40% passes the 80% criterion narrowly, with undissolved compound left behind. The step size matters for the per-mole ranking. Computing the next, failed step with the card's molar masses, TIPP lies between 0.012 and about 0.018 mol per 100 g, and PHPH between 0.016 and about 0.031. The ranges overlap, so the statement that TIPP, with both bond types, has the lowest per-mole LoC is not resolved by these data; the separation between the two groups is. The polymer-control test depends on the 99% level and on a cross-pH comparison for TIPP, and neither the derivation of the Figure 7 rates nor the number of replicates is printed (story notes X4, X5, X6, X13). Several profiles exceed 100%, and the paper does not say whether the totals in Eq. 1 are nominal or assayed, so drug–polymer differences of a few percent are within the normalization uncertainty (story note X14).

Conclusion

The paper shows, within a tightly controlled series, that the kind of bond a drug forms with PVPVA goes with how much of it the polymer can carry while still releasing well. Two compounds with phenolic OH groups, which hydrogen-bond to the VP carbonyl, lose congruent release above 5 and 10 wt% loading. A compound that only halogen-bonds, and more weakly by IR, holds out to 30 wt%, and one with no specific interaction to 40 wt%. Lipophilicity, amorphous solubility, dry $T_g$ and MW do not explain this ordering. At the LoC all four release at the neat polymer's rate, which points to polymer dissolution as the step that drug–polymer bonds may slow.

The mechanism that would connect the dry-state bond to the release cliff, reduced water uptake by the polymer, is reasoned from the literature rather than measured, and the design cannot separate drug–polymer from drug–drug hydrogen bonding. The practical lesson is still useful: the interaction usually sought to protect an ASD against crystallization may cap its useful drug loading in PVPVA, and both should be screened before a loading is chosen.