Chlorine, bromine and iodine on a drug as weak binding partners for the PVP carbonyl in amorphous solid dispersions

Background and motivation

Many modern drug candidates dissolve poorly in water. One of the most popular remedies for oral delivery is the amorphous solid dispersion (ASD): a miscible blend of a lipophilic drug with a hydrophilic polymer, which can greatly improve bioavailability compared with the crystalline drug (Section 1). The specific interactions between drug and polymer are considered critical to how an ASD performs, and hydrogen bonding is by far the most studied of them. Several studies suggest that drug–polymer hydrogen bonds help delay crystallization in the solid state (Rumondor 2009; Taylor and Zografi 1997; Miyazaki 2004), help maintain the supersaturation generated when an ASD dissolves (Yokoi 2005), and may improve dissolution (Chen 2019). A formulator screening polymers therefore tends to ask first whether the drug has hydrogen-bond donors that can pair with acceptor groups on the polymer. A drug without such donors is usually assumed to interact with the polymer only through weaker, non-specific forces.

Halogen atoms offer another possibility, which the formulation literature had not examined. The halogen bond (XB) is the attraction between an electrophilic region on a carbon-bound halogen (Cl, Br, I and occasionally F) and a Lewis base. This looks counterintuitive, because both partners are electronegative. The explanation is that the electron density around a carbon-bound halogen is uneven: opposite the C–X bond lies an electron-poor region, the σ-hole, which can attract electron-rich atoms such as oxygen, nitrogen and other halogens (Cavallo 2016). The strength of the bond depends on the size of the σ-hole and usually follows I > Br > Cl > F. It can be tuned by electron-donating or electron-withdrawing substituents near the carbon that carries the halogen, and fluorine forms halogen bonds only next to strong electron-withdrawing groups (Metrangolo 2011). The bond is directional, with an ideal C–X···A angle close to 180° (Wilcken 2013). Over the half century since it was first described, halogen bonding has been used in crystal engineering, materials science, protein–ligand design, supramolecular gels, catalysis and pharmaceutical cocrystals (refs 17–27).

How strong can it be? In protein–ligand complexes, halogen-bond interaction energies were estimated at −1.97 to −5.4 kcal/mol (Riley and Hobza 2011). The strong end matches a typical hydrogen bond, such as the bond between the hydroxyl of methanol and the carbonyl of acetone at about −5.3 kcal/mol (Lommerse 1997). Crystal-engineering studies show that when both kinds of bond are possible, hydrogen bonding wins, although the two can coexist (Aakeroy 2015; Aakeröy 2007). Halogen bonds are also less sensitive to a polar environment and persist in it (Wilcken 2013; Robertson 2017). On the acceptor side, a survey of ligand–protein complexes in the Protein Data Bank found carbonyl oxygen to be the most common electron donor in halogen bonds, followed by nitrogen (Xu 2014; Shinada 2019). That observation points directly at the PVP-type polymers used in many ASDs, whose vinylpyrrolidone amide carbonyl is a known good hydrogen-bond acceptor (Yuan 2015).

Halogen bonds are most often detected by X-ray crystallography, which measures the distance between the halogen and its partner but requires a crystal. Solid-state and solution NMR, IR, Raman and X-ray photoelectron spectroscopy (XPS) have also been used, often with density functional theory (DFT) calculations as support (refs 19, 33–38). None of this had been applied to drug–polymer blends. The authors point out that more than 30% of drug molecules in development contain halogens (Xu 2014) and that medicinal chemists already exploit halogen bonds in drug design. Yet while hydrogen bonding and ionic interactions between drugs and polymers had been described (refs 10, 41–43), no drug–polymer halogen bond had been reported. The question left open was simple: does a halogenated drug, dispersed in an amorphous polymer, form a halogen bond with it?

Que, Taylor and colleagues at Purdue University and AbbVie set out to answer this with four model compounds that contain Cl, Br or I but no hydrogen-bond donor (Figure 1). Two are marketed drugs, clotrimazole and loratadine. Brotrimazole, the bromine analogue of clotrimazole, was made for this study, as was Me-DIBF, 3,3-bis(3,5-diiodo-4-methoxyphenyl)isobenzofuran-1(3H)-one, a methylated derivative of a protein inhibitor with four iodines. The main polymer was copovidone (PVPVA), present in many commercial ASDs, and the simpler relative povidone (PVP) was used with clotrimazole. Three techniques address three questions: XPS asks whether the halogen gains electron density in the dispersion, solution NMR asks how strongly the halogen compound binds a carbonyl, and IR asks which carbonyl of the polymer is the partner. The authors present the work as the first demonstration of drug–polymer halogen bonding in amorphous formulations (Section 1).

Background concepts

Who donates what in a halogen bond. The vocabulary is a frequent source of confusion. In the usual convention, the halogen atom is the halogen-bond donor, because it supplies the electron-poor σ-hole, and the electron-rich partner (here a carbonyl oxygen) is the halogen-bond acceptor, even though it is the partner that donates electron density. So in this paper the drug is the halogen-bond donor and the polymer carbonyl the acceptor. The paper itself follows this convention in Sections 1 and 3.6, but on two later pages it calls the π electrons of toluene and the EP carbonyls "halogen bond donors" (Sections 4.2 and 5), where "electron donors" is meant.Card check: the word "donor" on these two pages should be read as electron donor, that is, halogen-bond acceptor (X22). In this article, "donor" and "acceptor" always refer to the halogen bond.

XPS binding energy and chemical shifts. X-ray photoelectron spectroscopy shines monochromatic X-rays on a sample and measures the kinetic energy of the electrons ejected from inner (core) shells. From that, the instrument reports the binding energy (BE), the energy needed to remove a given core electron, such as Cl 2p, Br 3d, I 3d or O 1s. The BE of an element shifts by a few tenths of an eV to a few eV with its chemical environment. A lower BE means the atom carries more electron density around it; a higher BE means it has lost some. A p or d core level appears as a pair of peaks, a doublet, because of spin–orbit splitting. The paper fixes the area ratios of the two members at 2:1 for p and 3:2 for d levels, and the splittings at 1.6 eV for Cl 2p, 1.0 eV for Br 3d and 11.5 eV for I 3d (Section 4.1). In a spectrum that contains two chemical environments of the same halogen, the fit therefore needs two doublets, and the area of each gives the share of halogen atoms in that environment. Here the doublet found in the neat compound is called the high-BE (neat-position) component, and new doublets below it are called low-BE components.

Two practical points matter for interpreting XPS of polymer tablets. First, insulating samples charge up during measurement, which shifts all peaks. The instruments use a charge neutralizer, and the energy scale is then corrected by setting a reference peak to a fixed value; this paper sets the C–C component of the C 1s peak to 284.8 eV at each spot (Section 3.5). If different regions of a sample charge differently, the correction cannot remove the difference, and displaced peaks can appear that have nothing to do with chemistry. Second, XPS samples only the outer surface, here roughly the top 10 nm (the information depth stated in Section 3.5), so it reports the composition and environment near the surface of the tablet rather than in the bulk.

NMR titration and the binding constant. When a molecule binds a partner weakly and rapidly in solution, NMR does not show separate peaks for the free and bound forms; it shows one peak whose position moves gradually as more partner is added. The authors add increasing amounts of the carbonyl compound 1-ethyl-2-pyrrolidone (EP) to a solution of the halogenated compound and follow the chemical shift of a nucleus near the halogen. The curve of shift change against EP concentration rises and bends over as more of the compound becomes bound, and fitting it to a 1:1 binding model gives the association constant K_F (Equation 1, Section 3.6):

Here [AD] is the concentration of the halogen-bonded complex, [A] the concentration of free acceptor (EP) and [D] the concentration of free donor (the halogenated compound), all in mol/L, so that K_F has units of M⁻¹. In plain terms: K_F measures how strongly the two molecules prefer to be paired rather than apart. A useful rule of thumb follows directly from the definition: half of the donor is bound when the free acceptor concentration equals 1/K_F. With K_F near 1 M⁻¹, half of the compound is bound only when EP is near 1 M, which marks a very weak interaction.

IR carbonyl bands. PVPVA carries two kinds of carbonyl: the amide C=O of its vinylpyrrolidone (VP) units, which absorbs around 1684 cm⁻¹, and the ester C=O of its vinyl acetate (VA) units at 1736 cm⁻¹ (Section 4.3). PVP contains only the VP amide. When a carbonyl oxygen shares electron density with a partner, the C=O bond weakens slightly and its stretching band moves to lower wavenumber, a red shift. For drug–PVP hydrogen bonds, red shifts as large as 20–30 cm⁻¹ have been reported (Rumondor 2009; Wegiel 2014; Kestur 2011). A band that does not move in the same spectrum serves as an internal control.

Van der Waals contact. Each atom has a contact radius; the sum of two atoms' radii is the distance at which they just touch. In a crystal structure, two atoms closer than this sum are taken as a sign of an attractive interaction. The paper uses 3.55 Å as the van der Waals sum for Br and C (SI p. 8).

Methods in detail

The work splits into four contributions. The first is the design of a set of halogenated model compounds without hydrogen-bond donors, including the synthesis of two of them, a crystal structure, and the preparation of dispersions and films. The second is XPS of the halogen and of the polymer oxygen, used to detect electron transfer and to count how many halogens are affected. The third is a solution NMR titration that measures the binding strength between the compounds and a monomer analogue of the VP unit, with solvent and carbonyl-free controls. The fourth is IR of the polymer carbonyl region, used to identify which carbonyl accepts the bond. The flow below shows how these pieces feed the authors' conclusion.

Contribution 1: A set of halogenated model compounds without hydrogen-bond donors

① Problem. To show a halogen bond to a polymer carbonyl, the usual explanation for a carbonyl shift, a hydrogen bond from the drug, must be ruled out. Many halogenated drugs also carry N–H or O–H groups that would hydrogen-bond to the same carbonyl, and the two effects would be impossible to separate by spectroscopy. A useful set of compounds should also vary the halogen while keeping the rest of the molecule the same, and include iodine, the halogen expected to bond most strongly.

② What they did. The authors bought clotrimazole (one Cl on a phenyl ring of a trityl–imidazole scaffold) and loratadine (one Cl, plus a carbamate carbonyl) (Figure 1a, 1c). They synthesized brotrimazole, 1-((2-bromophenyl)diphenylmethyl)-1H-imidazole, in three steps (Scheme 1, Section 3.1): Grignard addition of phenylmagnesium bromide (2.2 equiv) to methyl 2-bromobenzoate gave (2-bromophenyl)diphenylmethanol in 86% yield; thionyl chloride with a few drops of DMF converted it to the chloride, used crude; and reaction with imidazole (1.5 equiv) and triethylamine (3.0 equiv) in anhydrous DMF gave brotrimazole (3.0 g, 77.1%). Me-DIBF was made by O-methylation of the tetraiodo phenol with potassium carbonate and methyl iodide in DMF (95.6%, Scheme 2, Section 3.2). Identity was checked by ¹H and ¹³C NMR (Figures 1S–4S) and high-resolution mass spectrometry; purity is not stated. Brotrimazole crystals were grown from dichloromethane and ethyl acetate by hexane vapour diffusion, and two crystals were measured at 150 K with Mo Kα radiation (Section 3.3). The structure is monoclinic, space group P2₁/c, Z = 4, refined to R = 0.038 (Table S1).

Dispersions were made by dissolving drug and polymer in ethanol (tetrahydrofuran, THF, for Me-DIBF), removing the solvent on a rotary evaporator with a 60 °C water bath (40 °C for THF), and drying for one day in a vacuum oven (Section 3.4). Clotrimazole–PVPVA, clotrimazole–PVP and loratadine–PVPVA were prepared at 10, 20, 25, 30, 40 and 50 wt% drug loading (DL), and brotrimazole–PVPVA and Me-DIBF–PVPVA at 10, 20, 30 and 40 wt%. For IR, separate thin films were spin-coated from ethanol solutions with 100 mg/mL total solids onto KRS-5 windows, at 10–90 wt% DL in steps of 10% (Section 3.7). The paper reports no X-ray powder diffraction, calorimetry or other test showing that the dispersions and films were amorphous and single-phase.

③ Why it works. Without hydrogen-bond donors on the drug, any specific attraction between drug and polymer carbonyl cannot be a hydrogen bond donated by the drug. The clotrimazole–brotrimazole pair changes only one atom, Cl to Br, on the same scaffold, so differences between them can be attributed to the halogen or to how it sits in the molecule. Me-DIBF brings iodine. The two polymers form a second pair: PVP offers only the VP amide carbonyl, while PVPVA offers both the VP amide and the VA ester, so a comparison of their spectra can show which carbonyl takes part. The brotrimazole crystal structure, finally, shows how this molecule packs with itself, which the authors later use to explain its behaviour.

④ How it differs from earlier work. Earlier spectroscopic studies of drug–polymer interactions in ASDs, for example indomethacin with PVP (Taylor and Zografi 1997), chose drugs with hydrogen-bond donors on purpose. Here the donor is removed on purpose. Earlier halogen-bond studies, in turn, worked mainly with cocrystals and protein–ligand complexes, where crystallography can locate the contact directly. An amorphous polymer blend has no crystal lattice, so the evidence has to come from spectroscopy alone.

Contribution 2: XPS of the halogen and of the acceptor oxygen

① Problem. A probe is needed that reports the electronic state of the halogen itself, inside an amorphous solid. IR and NMR mostly see the partner groups or neighbouring atoms; XPS sees the halogen core levels directly. The question is whether the halogen becomes more electron-rich when dispersed in the polymer, and how large a share of the halogen atoms is affected at each loading.

② What they did. For XPS (Section 3.5), 100 mg of dispersion powder was pressed into an 8 mm tablet at 1500 psi for 60 s and measured on a Kratos Axis Ultra DLD spectrometer with monochromatic Al Kα radiation (1486.6 eV), using a built-in charge neutralizer. Five spots of 0.5 mm diameter were measured per tablet. High-resolution spectra were collected at 20 eV pass energy and survey spectra at 160 eV. After charge correction to the C–C peak at 284.8 eV, each halogen spectrum was fitted after Shirley background subtraction with Gaussian/Lorentzian doublets of fixed area ratio and splitting. The binding energies of the neat compounds and of the new components in the dispersions are listed in Table 1, and the share of halogen atoms in the low-BE components at each loading in Table S2 and Figure 3. For clotrimazole in PVP, the O 1s peak was also compared between neat PVP and the 50/50 wt% dispersion (Figure 6S); in this system, unlike the others, all the oxygen signal comes from the VP carbonyl.

The authors then turned the low-BE share into a count of occupied carbonyls (Figure 6). They assume that every halogen atom in the low-BE state is halogen-bonded and that each bonded halogen engages exactly one VP carbonyl. The paper states this rule only in words, in the caption of Figure 6. Written out in symbols, with the symbols chosen by the AI writer, it reads:

Here $n_{\mathrm{V}\mathrm{P}}$ is the number of moles of VP units in the polymer of a sample, $n_X$ the number of moles of halogen atoms contributed by the drug (one per molecule for clotrimazole, loratadine and brotrimazole, four for Me-DIBF), and $f_{\mathrm{l}\mathrm{o}\mathrm{w}}$ the low-BE share from Table S2. In plain terms: the number of bonded carbonyls equals the number of bonded halogens, and the plotted ratio compares the carbonyls left over with those in use. The paper does not print the molar masses or the VP content of PVPVA used in the calculation.Card check: assuming 60 wt% VP in PVPVA 64 and 111.14 g/mol per VP unit, the extractor reproduced 19 of the 21 plotted points within about 0.6; brotrimazole at 20 wt% (plotted about 15, recomputed about 10.8) and clotrimazole–PVPVA at 25 wt% (plotted about 8, recomputed about 9.4) do not follow from Table S2 (X15).

③ Why it works. A halogen whose σ-hole draws electron density from a partner should become, on balance, more electron-rich, so its core electrons should be bound less tightly and appear at lower BE. The partner, by the same reasoning, should lose some electron density and shift to higher BE. Comparing each dispersion with the neat compound isolates the change caused by the polymer, and the O 1s of PVP gives a view from the acceptor side that is free of other oxygen signals. If the interaction requires drug and polymer to meet, the bonded share should also fall as the drug loading rises and drug molecules meet each other more often, which gives a test inside the data.

④ How it differs from earlier work. XPS had been used to study halogen bonding in materials (ref 19) and to follow hydrogen-bonding and ionic interactions in polymer blends (refs 55, 56). The new step is to apply it to the halogen of a drug in an ASD and to use the share of the new component as a semi-quantitative measure of how much of the drug is engaged, as a function of loading. The authors judge that XPS "appears to be of particular utility" here, because it is one of few techniques that probe the halogen directly (Section 5).

Contribution 3: Solution NMR titration with a monomer analogue

① Problem. XPS can show that something happens to the halogen but not how strong the interaction is. A binding constant cannot be measured in a solid polymer directly, so the authors used a solution model: the halogenated compound and a small molecule that mimics the VP repeat unit.

② What they did. The titrant was EP, 1-ethyl-2-pyrrolidone, a monomeric analogue of the VP unit (Figure 1h), which the authors assume reports the donor–acceptor binding strength of the polymer. The solvent was deuterated cyclohexane, chosen to give sufficient solubility and to avoid halogen bonding between the solute and π electrons. A 14.5 mM solution of the compound was divided among eight NMR tubes of 0.6 mL, and EP was added at 11.0, 87.5, 173.2, 339.9, 654.6, 946.8 and 1218.9 mM (Section 3.6).Card check: the procedure describes the 14.5 mM stock as made in deuterated toluene, while cyclohexane is named as the solvent and Table 2 is in cyclohexane; the cyclohexane series is not described separately, and seven EP levels are listed for eight tubes (the SI captions add 0 mM) (X7). Spectra were recorded at 25 °C after at least 5 min of equilibration on a Bruker Avance-III-800 spectrometer with a cryoprobe. The ¹H shift of a ring proton near the halogen-bearing carbon and the ¹³C shift of the halogen-bearing carbon (a marked carbon of the bromophenyl ring for brotrimazole) were fitted to a 1:1 binding curve in OriginPro 2019 to give K_F by Equation 1. Each titration was repeated three times.Card check: the text and two SI captions call the monitored proton "para" to the halogen, but the SI structure drawings mark the H on the ring carbon next to the halogen-bearing carbon (X6). The same titrations were run in deuterated toluene (Table S3). As a control, ¹H spectra of clotrimazole, brotrimazole and loratadine were recorded with 654 mM 1-ethylpyrrolidine, which has the same ring and ethyl group as EP but no carbonyl, and compared with 654 mM EP (Figures 12S–14S). Me-DIBF was not titrated "due to the complex potential stoichiometry" of a molecule with four iodines (Section 4.2).

③ Why it works. The titration curve depends on how the fraction of bound compound changes as EP is added, so its curvature yields K_F independently of the size of the shift. Two controls make the result specific. If the shift arises from the carbonyl, a carbonyl-free analogue should cause little change; and if the halogen is the binding site, a solvent with π electrons that can itself interact with the halogen, such as toluene, should compete with EP and lower K_F. The extractor of the card checked the fit against two points read from the clotrimazole curve (about 0.034 ppm at 339.9 mM and 0.084 ppm at 1218.9 mM EP, Figure 4a), which give about 0.62 M⁻¹ in a 1:1 hyperbola, close to the reported 0.63 M⁻¹.

④ How it differs from earlier work. NMR titration is a standard way to measure weak association constants (Gurka and Taft 1969; Dalvit and Vulpetti 2016; Thordarson 2011). The new element is its use as a stand-in for a drug–polymer contact: the polymer is replaced by its repeat unit so that a binding constant can be measured at all. The price of this step is that the number refers to two small molecules in an apolar solvent, not to a drug in a glassy polymer.

Contribution 4: IR of the polymer carbonyl bands

① Problem. PVPVA offers two possible acceptors, the VP amide carbonyl and the VA ester carbonyl. XPS of PVPVA cannot separate their oxygen signals, so another method is needed to tell which one takes part.

② What they did. Films of pure PVPVA and of clotrimazole–, brotrimazole– and Me-DIBF–PVPVA at 10–90 wt% drug were spin-coated (50 µL of solution, 15 s at 50 rpm then 50 s at 2500 rpm) onto KRS-5 windows and measured in transmission on a Bruker Vertex 70 spectrometer with 128 scans at 4 cm⁻¹ resolution (Section 3.7, Figure 5). Loratadine was left out because its own carbamate carbonyl overlaps the polymer carbonyl region. No replicate spectra are reported.

③ Why it works. A carbonyl that donates electron density to a halogen should weaken slightly and absorb at lower wavenumber, while a carbonyl that is not involved should stay where it is. Because both carbonyl bands appear in the same spectrum, the VA band acts as an internal reference for the VP band. Loadings up to 90 wt% also make the drug-to-VP ratio far higher than in a typical ASD, which should make any shift easier to see.

④ How it differs from earlier work. IR carbonyl shifts are the standard tool for drug–PVP hydrogen bonds, which give red shifts of up to 20–30 cm⁻¹ (Rumondor 2009; Wegiel 2014; Kestur 2011). Here the expected effect is much smaller, and the method is used close to its resolution limit.

Putting the contributions together, the authors' reasoning about loading can be drawn as a simple chain (Sections 4.1 and 5). It is their interpretation, introduced in Section 4.1 with "We rationalized", not a separate measurement.

Experimental analysis

Setup in brief. XPS was run on five compound–polymer pairs: clotrimazole in PVPVA and in PVP, loratadine, brotrimazole and Me-DIBF in PVPVA, each at four to six loadings plus the neat compound. NMR titrations covered clotrimazole, loratadine and brotrimazole with EP in two solvents, with one carbonyl-free control per compound. IR covered three compounds in PVPVA films at nine loadings each. One crystal structure, brotrimazole, was solved. No property of the dispersions as formulations, such as physical stability, dissolution or supersaturation, was measured.

Result 1: every halogen gains new, lower-energy XPS components in the polymer. Neat clotrimazole shows a single Cl 2p doublet at 200.4 and 202.0 eV (Figure 2a). In PVPVA, two new doublets appear, at 198.1/199.8 eV and 196.7/198.3 eV. Clotrimazole in PVP and loratadine in PVPVA behave the same way (Figure 2b, 2d). Brotrimazole gains a Br 3d doublet at 67.1/68.1 eV, which becomes the dominant component in the dispersion (Figure 2c), and Me-DIBF gains an I 3d doublet at 618.1/629.6 eV (Figure 2e). Table A collects the values and the size of each shift.

Table A. Halogen core-level binding energies of the neat compounds and the new components in the dispersions (from Table 1; the shift column is the AI writer's subtraction). The loading and the number of samples behind the tabulated values are not stated; the spectra in Figure 2 are at 20–25 wt%.

Three features of Table A matter for its interpretation. All shifts lie between about 2.3 and 3.8 eV, which the authors summarize as 2–4 eV and describe as "significant and comparable to the binding energy shift produced by hydrogen bonding" (Section 5, citing Cecchet 2004). For the two chlorine drugs, two separate new doublets about 1.4 eV apart appear, and the paper does not assign them to different environments. And the uncertainty of the second clotrimazole–PVPVA doublet is printed as ± 0.4 eV in Table 1 but partly as ± 0.1 eV in the text.Card check: Table 1 gives 198.3 ± 0.4 eV for the second Cl 2p_1/2 peak, the text (Section 4.1) 198.3 ± 0.1 eV; the table value is used here (X1). The text also refers to these values as "Table 2" (X2). Neat brotrimazole is a special case: it already carries a lower-BE Br doublet holding 25% of the bromine. The authors' "probable explanation" is that brotrimazole exists in two energetically similar conformations. In the crystal, one phenyl ring and the imidazole swap places in 18.3% of molecules (18% in the main text), which puts the Br next to a phenyl ring, "in general" a better electron donor than imidazole, instead of the imidazole (Section 4.1, SI p. 8). The 25% and the 18.3% do not match, and the neat component is not subtracted from the brotrimazole shares in Table S2.

The authors read the shifts as electron gain: "the halogen atoms likely received electrons from carbonyl groups present in the polymers" (Section 4.1). Smaller shifts in the same direction have been reported for halogen bonds between iodine and nitrogen (refs 57, 58), which the authors take as "further support" (Section 5).

Result 2: the bonded share is largest at low loading. Table B gives the share of halogen atoms in the low-BE components at each loading, as printed in Table S2 and plotted in Figure 3.

Table B. Percentage of halogen atoms in the low-BE state versus drug loading (from Table S2; also Figure 3). The basis of the ± values (five spots, number of tablets) is not stated. "Not reported": listed in the methods (Section 3.4) but no data given.

From lowest to highest loading, the low-BE share falls in all five systems: from 61–88% at 10 wt% to 22–56% at 40 wt%. The authors state that "the trend among all ASDs" is a decrease with loading and that this "supports the idea that PVPVA is donating electrons to the halogens" (Section 4.1). The decrease holds end to end, but not step by step in three systems: clotrimazole/PVPVA rises from 53.6 to 56.7% between 25 and 30 wt%, clotrimazole/PVP from 40.9 to 51.6% between 30 and 40 wt%, and loratadine from 69.7 to 82.1% between 10 and 20 wt%. Several of these steps are within the stated spreads, which for clotrimazole/PVPVA at 10 wt% reach ± 18.4%. Two planned loadings, 25 wt% clotrimazole/PVP and 50 wt% loratadine/PVPVA, are listed in the methods but have no data and no explanation (card note X23).

Converted into carbonyl occupancy by the relation above, the same data give Table C, read from Figure 6.

Table C. Molar ratio of free to halogen-bonded VP carbonyls versus drug loading (read from Figure 6, approximate; "—" = not tested). The loratadine marker at 40 wt% is partly hidden. Two points (brotrimazole 20 wt%, clotrimazole/PVPVA 25 wt%) do not follow from Table S2 (card note X15).

At 10 wt%, between about 16 and 36 VP carbonyls are free for each bonded one; the authors describe the interaction here as "undersaturated" (Section 5). The ratio drops by 20–30 wt% and then levels. In the authors' words, "a maximum of about 1 halogen bonded carbonyl per 8 free carbonyls is observed, with no further decreases when additional amounts of drug are present", at "approximately the same ratio" in all systems. Read from the figure, the level lies between about 5 and 8 rather than at a fixed 8: clotrimazole/PVPVA reaches about 5 at 50 wt% and brotrimazole about 5 at 40 wt%.Card check: recomputing the ratio from Table S2 gives 5.2 (clotrimazole/PVPVA, 50 wt%) and 4.6 (brotrimazole, 40 wt%), so the plateau spans about 5–8, not 1:8 (X14). The authors offer two hedged explanations. The saturation "may be due, at least in part, to steric constraints", and the similar plateau in all systems "further suggests possible steric constraints". Me-DIBF, which already reaches about 7 at 20 wt%, carries four iodines per molecule, so "multiple halogen bonds can be formed for each molecule, likely leading to the observed rapid saturation" (Section 5).

What a plateau in this ratio means deserves one sentence of arithmetic. As the loading rises, the moles of drug per mole of VP rise steeply, so if the bonded share stayed constant, the ratio would keep falling. A constant ratio therefore means that the absolute number of bonded halogens per VP carbonyl stops increasing: extra drug added beyond about 30–40 wt% does not find new carbonyl partners. Note by the AI writer (not discussed in the paper): the same outcome would follow if the extra drug formed drug-rich domains or crystals with little polymer contact, which the paper did not test for (see Limitations).

Result 3: the acceptor is the VP amide carbonyl. Three observations point to the VP oxygen. In the IR spectra (Figure 5), the VP carbonyl band of PVPVA moves from about 1684 to 1682 cm⁻¹ as the clotrimazole or brotrimazole loading rises from 10 to 90 wt%, and by 6 cm⁻¹ with Me-DIBF, while the VA band at 1736 cm⁻¹ does not move. The authors write that "the spectral shift is well resolved, despite the very small numerical change of only 2 cm⁻¹", and that it "suggests that the carbonyl group in VP is responsible" (Section 4.3); in the discussion, IR results "confirm" that VP is preferred over VA (Section 5).Card check: the 2 cm⁻¹ shift is half the 4 cm⁻¹ spectral resolution, with no replicate spectra or uncertainty reported; it is best treated as qualitative (X19). In the Me-DIBF films the absorbance above about 1750 cm⁻¹ grows with loading and the 90 wt% VP band is broader. Second, in PVP, where all oxygen is VP carbonyl oxygen, the O 1s maximum moves from about 530.8 to 531.0 eV with 50 wt% clotrimazole (Figure 6S), +0.2 eV, "suggesting a reduction in electron density consistent with the formation of a halogen bond" (Section 5). Third, in solution, 654 mM 1-ethylpyrrolidine caused "only minor proton shifts" in clotrimazole, brotrimazole and loratadine, against clear shifts with the same concentration of EP (Section 4.2). The authors say this "further supports the role of the VP oxygen as the halogen bond acceptor", and add that it is "not surprising", since carbonyl oxygen is a common halogen-bond acceptor and the VP carbonyl a good hydrogen-bond acceptor (Section 5).Card check: Figures 12S–14S show only the two spectra with additive, with no additive-free reference and no shift values. In Figure 12S the clotrimazole doublet sits near 7.30 ppm with both additives, while Figure 7S places the monitored proton near 7.29 ppm without EP and near 7.34 ppm at 654.6 mM EP, so the control cannot be checked from the SI (X9).

Result 4: the interaction is weak. Table D gives the binding constants from the titrations.

Table D. Binding constants K_F with EP at 25 °C (from Table 2, cyclohexane, n = 3 titrations, and Table S3, toluene). The ¹³C/¹H ratio is from the card's extractor note X8. Me-DIBF was not titrated.

The constants in cyclohexane lie between 0.56 and 1.11 M⁻¹. At the top of the titration, 1219 mM EP, the monitored proton had moved about 0.070–0.084 ppm downfield and the monitored carbon about 0.28–0.52 ppm (upfield for clotrimazole and brotrimazole; for loratadine the tracked ¹³C peak crosses a neighbouring peak near 654.6 mM). The authors call the ¹H and ¹³C values "similar" (Section 4.2). The ¹³C constant is, however, about 1.5 times the ¹H constant for every compound, beyond the stated spreads, so they agree in order of magnitude rather than in value. In toluene the constants drop to 0.21–0.39 M⁻¹, which the authors find "consistent with competition between the toluene π electrons and the carbonyl groups in EP for the halogen atoms", as reported for other halogen-bonded systems (Ma 2012).

How weak is this? Table E sets the paper's numbers beside the comparators it cites.

Table E. Strength indicators for the proposed halogen bond and the literature values the paper compares them with (Sections 1 and 5; Table 2; Figure 5; Figure 6S).

On the two measures that compare like with like, the solution binding constant and the IR shift, the authors conclude that the bond "appears to be similar to weak hydrogen bonding interactions" (Section 5). The IR shift is several times smaller than for drug–PVP hydrogen bonds. Note by the AI writer (not discussed in the paper): with K_F between 0.56 and 0.65 M⁻¹ from ¹H, the 1:1 definition implies that only about 40–45% of the compound was bound even at the highest EP concentration (about 50–58% on the ¹³C constants). The titration curves therefore cover only the lower half of the binding isotherm, a range in which the binding constant and the limiting shift are fitted together and are hard to separate. That may contribute to the ¹³C/¹H difference. By the standard relation between a binding constant and a free energy, a constant near 1 M⁻¹ also means a standard binding free energy close to zero at 25 °C; the protein–ligand interaction energies in the last row of Table E are of a different kind and cannot be compared with it directly.

Result 5: iodine perturbs the carbonyl most; chlorine and bromine look alike. Halogen-bond strength usually follows the polarizability of the halogen, I > Br > Cl. Here Me-DIBF shifts the VP band by 6 cm⁻¹ against 2 cm⁻¹ for clotrimazole and brotrimazole, and the two latter have similar K_F (0.63 versus 0.56 M⁻¹ by ¹H; 0.97 versus 0.81 M⁻¹ by ¹³C). The authors conclude that iodine "appeared to form a stronger interaction than either Cl or Br, which appeared similar in strength" (Section 6). To explain why Br is not stronger than Cl, they turn to the brotrimazole crystal (Figure 5S). Its two phenyl rings form an "open clam" with the Br between them, and Br sits 3.373 and 3.425 Å from carbons of a neighbouring molecule, inside the 3.55 Å van der Waals sum. In the published clotrimazole structure (Song and Shin 1998), the Cl lies on the other side of the phenyl rings and no such contact is seen. The authors suggest that the Cl ≈ Br observation "can perhaps be rationalized" this way: "the tendency for brotrimazole to self-interact may be greater compared to clotrimazole, and this interaction can compete with the brotrimazole-PVPVA intermolecular interaction" (Section 5). In the SI they add that Br is "potentially more structurally hindered" than Cl as a halogen-bond donor.

The authors also state that "halogen bonding extent did not increase when Cl was replaced with Br, as corroborated by NMR titration results" (Section 5). The XPS data in Table B do not show this directly: brotrimazole's low-BE share exceeds clotrimazole's in PVPVA at every common loading (87.9 versus 73.0% at 10 wt%, 71.1 versus 55.3% at 20 wt%, 61.9 versus 56.7% at 30 wt%, 55.9 versus 32.8% at 40 wt%). Part of that gap may come from the 25% low-BE Br already present in the neat compound, which is not subtracted, so the Cl-versus-Br comparison of extent remains unresolved (card note X16).

Comparisons across systems and what the controls rule out. Several comparisons inside the data are informative. Clotrimazole behaves similarly in PVP and PVPVA: both give two new Cl doublets at nearly the same energies (Table A) and a falling low-BE share (Table B). PVP, which has more VP units per gram than PVPVA, gives the higher free-to-bonded ratio at 10 wt% (about 36 versus 22, Table C), as expected if only VP carbonyls count. The two chlorine drugs, clotrimazole and loratadine, which have quite different scaffolds, give similar shifts and similar binding constants, which argues that the halogen rather than another part of the molecule is the common factor. The controls rule out some alternatives. Hydrogen bonding from the drug is excluded by design, since no compound has a donor. The VA ester is unlikely to be the main acceptor in PVPVA, since its IR band does not move. The carbonyl-free 1-ethylpyrrolidine control argues against a non-specific effect of adding a large amount of an amine-type solute, and the lower constants in toluene fit a halogen-centred interaction. What the controls do not rule out is that some of the XPS change arises from the comparison itself, neat compound against dispersion, rather than from a contact with the carbonyl. No drug–polymer physical mixture and no charging test were measured (see Limitations).

Discussion

Where the conclusion applies. The evidence covers a specific situation: compounds with Cl, Br or I and no hydrogen-bond donor, dispersed in PVP-type polymers whose acceptor is the VP amide carbonyl, at 10–50 wt% for XPS and up to 90 wt% for the IR films, measured dry. The binding constants refer to a monomer analogue in cyclohexane or toluene at 25 °C. Within that frame, three independent methods agree in direction: the halogens become more electron-rich in the polymer, the VP carbonyl becomes slightly perturbed, and the compounds associate weakly with a VP analogue in solution. In the discussion the authors "propose that halogen bonding exists between the amide carbonyl, on the polymer, and halogen atoms, Cl, Br, or I, in ASDs", to a greater extent at low drug loading (Section 5). The abstract and conclusions state it more firmly, as "demonstrated" and "were found", and call it the first such report in drug–polymer formulations.

How strong the effect is in practice. For a formulation scientist, the size of the effect matters as much as its existence. Every quantitative indicator in the paper places the interaction at the weak end: K_F near 1 M⁻¹, IR shifts of 2–6 cm⁻¹ against 20–30 cm⁻¹ for drug–PVP hydrogen bonds, and an O 1s shift of 0.2 eV. At 40–50 wt% loading, only about one VP carbonyl in six to nine takes part, by the paper's own count. Note by the AI writer (not discussed in the paper): an interaction this weak is unlikely, on its own, to stabilize a dispersion against crystallization the way a strong drug–polymer hydrogen bond can be expected to. It may still contribute to miscibility or to the local environment of the drug, but the paper measured none of these properties. The authors' own outlook is hedged accordingly: halogen bonding "may be a common, but neglected" interaction in drug–polymer blends, "potentially important" for their properties and performance (abstract), and its impact on formulation performance "warrants further investigation" (Section 6).

What remains unresolved. The first open point is the origin of the XPS signal. The halogen shifts of 2–4 eV are large next to the 0.2 eV shift of the acceptor oxygen, the 2–6 cm⁻¹ IR shifts and a binding constant near 0.6 M⁻¹, and the two separate new chlorine doublets are not assigned. A physical mixture of crystalline drug and polymer, pressed and measured in the same way, would show whether the new components require molecular mixing. The second is the state of the samples: without X-ray diffraction or calorimetry, it is not known whether the dispersions at 40–50 wt% and the films at 60–90 wt% were amorphous and single-phase. The third is the solid-state structure of the contact: no solid-state NMR or DFT calculation locates a C=O···X contact in the dispersion itself. The fourth is the stoichiometry of Me-DIBF, which was not titrated, and the reason why bromine does not bond more than chlorine; the self-association of brotrimazole was inferred from its crystal, not measured in the amorphous state.

What it suggests for further work. The study opens several testable questions. Does halogen bonding survive when the same drug also carries a hydrogen-bond donor, given that hydrogen bonding wins when both are possible in crystals (Aakeroy 2015)? Does it survive water, which is always present in a stored or dissolving ASD, given the literature claim that halogen bonds persist in polar environments? Do other polymers with carbonyl acceptors, such as cellulose esters, behave similarly? And does a dispersion of a donor-free halogenated drug in PVPVA crystallize more slowly, or maintain supersaturation longer, than one in a polymer without a suitable acceptor? The paper also suggests a practical tool. Note by the AI writer (not discussed in the paper): the halogen XPS spectrum could serve as a quick screen of the halogen environment in a dispersion, provided that a physical-mixture reference and a check for differential charging are included. For polymer selection, a halogenated drug without hydrogen-bond donors is not necessarily limited to non-specific interactions with PVP-type polymers; it may have a weak specific partner, most engaged at low drug loading, which can be weighed alongside miscibility data.

Limitations

Limitations stated by the authors. The paper acknowledges few limitations explicitly, and most are stated as choices rather than weaknesses. IR spectra of loratadine–PVPVA "were not collected due to the overlap of the carbonyl group in loratadine with PVPVA carbonyl group region" (Section 3.7), so the acceptor identification rests on clotrimazole, brotrimazole and Me-DIBF. NMR titrations "were not performed for Me-DIBF due to the complex potential stoichiometry between halogen bond donor and acceptor" (Section 4.2), so the strongest-looking compound has no binding constant. The solution measurements rest on EP being "a monomeric analogue of PVP" and on cyclohexane being chosen "to avoid interference with halogen bonding" (Section 3.6). The interpretive steps carry the authors' own hedges: the halogens "likely" received electrons, the saturation "may be due, at least in part, to steric constraints", the Cl ≈ Br result "can perhaps be rationalized" by self-interaction, and in the discussion the overall conclusion is introduced with "we propose". The authors do not discuss sample-state verification, XPS controls or the precision of the IR peak positions.

The AI writer's own judgement (1): the XPS evidence lacks its key controls. The card's extractor rates the overall evidence as moderate. The strongest part, three probes agreeing in direction, is weakened most by what was not measured on the XPS side. No drug–polymer physical mixture was run, the physical form of the neat references (as received, presumably crystalline, or amorphous) is not stated, and no test for differential charging is reported. For compressed tablets of insulating organic material, regions of different composition can charge differently even with a neutralizer, and C 1s referencing corrects only the average. Without these controls, the 2–4 eV shifts could partly reflect solid form or charging rather than a contact with the carbonyl; their size, set against the 0.2 eV shift of the acceptor oxygen, makes the question pressing. The loading and the number of samples behind Table 1 are also not stated. These gaps do not refute the interpretation, since the IR and NMR evidence points the same way, but they mean the XPS shares in Table B and the ratios in Table C are best treated as within-paper relative measures.

The AI writer's own judgement (2): sample state and the meaning of the loading trend. No X-ray diffraction, calorimetry or other test shows that the dispersions were amorphous and single-phase. This matters most at 40–50 wt% for XPS and 60–90 wt% for the IR films, the loadings at which the bonded share falls and the ratio levels off. If part of the drug had separated into drug-rich domains or crystallized, the low-BE share would fall and the ratio would level for that reason alone; the authors' explanations, drug–drug proximity and steric constraints, are therefore not the only ones. The ratio in Figure 6 also mixes two kinds of information: a surface-weighted XPS share from the top 10 nm of a pressed tablet and the nominal bulk composition of the powder. Surface enrichment of either component would bias it. In addition, the claimed trend is not monotonic in three systems, the plateau spans about 5–8 rather than a fixed 8, and two plotted points do not follow from the tabulated shares (Tables B and C).

The AI writer's own judgement (3): the strength ranking and the solution numbers. The ranking I > Br ≈ Cl rests on IR peak positions alone, and the 2 cm⁻¹ shift is half the 4 cm⁻¹ spectral resolution, with no replicates. A peak maximum can in principle be located more finely than the nominal resolution, but without repeated spectra its precision is unknown. The IR band maximum is also a population average: Me-DIBF carries four iodines per molecule and has the lowest free-to-bonded ratio at 10–20 wt%, so its larger shift could reflect more carbonyls bonded rather than stronger individual bonds. The IR films and the XPS dispersions were made differently (spin-coated films versus rotary-evaporated, vacuum-dried powders, with Me-DIBF dispersions from THF) and cover different loadings, so they should not be paired at equal nominal loading. On the NMR side, the ¹³C and ¹H constants differ by about 1.5 times, the preparation of the cyclohexane series is not described separately from the toluene one, and the carbonyl-free control is shown without numbers or an additive-free reference. The solution constants also refer to a monomer analogue in an apolar solvent; how they translate to a drug in a glassy polymer, where the carbonyl concentration is several molar but mobility is low, is not addressed.

The AI writer's own judgement (4): editorial errors. The accepted manuscript contains several errors that do not affect the data but can confuse a reader. The materials text assigns Me-DIBF to Figure 1b and brotrimazole to Figure 1d, whereas the drawing and caption show the reverse (X3). Scheme 1 gives the solvent of the last brotrimazole step as dichloromethane, while the procedure uses anhydrous DMF (X4). The HRMS formula of Me-DIBF is printed as C22H15I2O4, although the four-iodine structure is C22H15I4O4, whose mass matches the printed calculated value (X5). The brotrimazole SI spectra contain peaks at positions matching ethyl acetate, a purification and crystallization solvent, that are absent from the printed peak lists, and purity is not stated (X21). Cross-references are off in places: binding energies said to be in "Table 2" are in Table 1, the toluene table is cited as "Tablet S3", the SI text cites "Figure 3S" for the crystal structure shown in Figure 5S, and the SI carries a different title from the article (X2).

Conclusion

The paper makes a clear and new point: halogenated drug-like compounds without hydrogen-bond donors are not inert towards PVP-type polymers. In dispersions of clotrimazole, loratadine, brotrimazole and Me-DIBF, every halogen gains new lower-energy XPS components, the VP amide carbonyl band of PVPVA shifts slightly while the VA ester band does not, the oxygen of PVP loses a little electron density, and the compounds bind a VP analogue in solution with constants of about 0.6–1.1 M⁻¹. The authors read these together as a halogen bond between the drug halogen and the VP carbonyl, "similar to weak hydrogen bonding", most extensive at low loading and levelling at about one bonded carbonyl per five to eight free ones. Their wording in the conclusions is that the compounds "were found to form halogen bonds with the amide carbonyl of the vinylpyrrolidone moiety in copovidone-based amorphous dispersions" (Section 6).

The conclusion is best held at the strength of its weakest link. The direction of the evidence is consistent across three methods, but the XPS shifts lack physical-mixture and charging controls, the amorphous state of the samples was not verified, and the comparative claims about saturation and about I > Br ≈ Cl rest on fitted shares and IR shifts near the resolution limit. No ASD property was measured, so whether this weak interaction matters for stability, supersaturation or release is still open. For formulation work, the paper is a reason to look for halogen–carbonyl contacts when a halogenated drug without donors is paired with PVP or PVPVA, and to expect them to be weak and most relevant at low drug loading.