Why polymers struggle to keep amorphous curcumin from crystallizing: intramolecular hydrogen bonds versus ionic interactions

Background and motivation

Curcumin is the yellow–orange pigment of the turmeric rhizome (Curcuma longa) and has been studied for a long list of biological activities. As a drug candidate it has two problems. It is poorly soluble in water, about 11 ng/mL at pH 5.0 (Tonnesen et al., ref 9), and it degrades quickly at neutral or alkaline pH (refs 10, 11); both limit its bioavailability. Cyclodextrin complexes, nanoparticles and solid dispersions have all been tried. The authors regard the amorphous solid dispersion (ASD) as one of the most promising routes: curcumin melts at 183 °C, a sign of high lattice energy, so removing the crystal lattice should bring a large gain in solution concentration. That gain is lost if the drug crystallizes during storage. C4·M1

Whether an amorphous drug crystallizes depends on the storage environment (temperature and relative humidity, RH; Rumondor et al., refs 16, 17) and on the formulation. Polymers are added to slow crystallization, and the outcome depends both on the drug's own crystallization tendency and on how well the polymer inhibits it (Van Eerdenbrugh and Taylor, ref 18). In the classification of Van Eerdenbrugh et al. (ref 19), curcumin is an "intermediate crystallizer": it does not crystallize instantly from a rapidly evaporated solvent, but it is not a stable glass former either. C5·R21

The Taylor group had built up a working premise over several papers: polymers that interact more with a drug slow its crystallization more (Kestur and Taylor 2010, ref 20; Kestur et al. 2011, ref 21; Van Eerdenbrugh and Taylor 2011, ref 22), and monitoring those interactions by infrared (IR) spectroscopy "appears to be a reliable method" for predicting ASD physical stability (Wegiel et al. 2013, ref 23, a study of resveratrol dispersions). Curcumin is an awkward test case for this premise. In its crystal structures, the central β-diketone exists as a keto–enol tautomer whose enol hydrogen forms a strong intramolecular hydrogen bond to the neighbouring keto oxygen, and each phenol OH forms a second intramolecular hydrogen bond to the methoxy group next to it on the ring (Parimita et al., ref 27; Sanphui et al., ref 28). The groups a polymer would normally bind to are therefore already occupied. C6·C7·R19

The paper asks what this means for ASDs. The authors made curcumin dispersions with five polymers chosen for diversity in hydrogen-bonding potential, namely poly(vinylpyrrolidone) (PVP), Eudragit E100 (E100), carboxymethyl cellulose acetate butyrate (CMCAB), hydroxypropyl methyl cellulose (HPMC) and HPMC-acetate succinate (HPMCAS); measured curcumin–polymer interactions by mid-IR spectroscopy; followed crystallization during storage at six conditions by powder X-ray diffraction (PXRD); and asked whether the interactions explain the differences in stability. C8·A1

Background concepts

Keto–enol tautomerism and the resonance-assisted hydrogen bond. Curcumin's central unit can be written as a di-keto form, with two C=O groups separated by a CH₂, or as a keto–enol form, in which one C=O has become C–OH conjugated with the chain (Figure 1a, b). Single-crystal studies find only the keto–enol form, and calculations place it lower in energy than the di-keto form (Benassi et al., ref 30). In the stable polymorph the enol H sits symmetrically between the two oxygens, and the six-membered ring this bond closes allows electron delocalisation across it, which is why it is called a resonance-assisted hydrogen bond (Figure 1c; Gilli et al., ref 29). In IR terms, the keto–enol form has no free ketone band near 1700 cm⁻¹ but a strong band near 1626 cm⁻¹. R19·R2

Intramolecular versus intermolecular hydrogen bonds. An intramolecular hydrogen bond links two groups within one molecule; an intermolecular one links two molecules, for example a drug OH and a polymer C=O. A hydrogen-bond donor offers the H (an OH); an acceptor offers a lone pair (a C=O, an ether O, an amine N). Etter's rules for organic crystals (refs 37, 38), which the authors cite, state that intramolecular hydrogen bonds that close a six-membered ring are usually preferred over intermolecular ones. The keto–enol hydrogen bond drawn in Figure 1c closes such a six-membered ring; the phenol–methoxy bonds are further intramolecular contacts on each aromatic ring. C36·R19

Reading drug–polymer interactions from IR spectra. When a drug OH hydrogen-bonds to a polymer, the O–H bond is weakened and its stretching band moves to lower wavenumber; the acceptor C=O band of the polymer moves to lower wavenumber as well. The authors take a lower OH wavenumber and a larger C=O shift as stronger or more extensive hydrogen bonding (ref 31). The usual reference for such shifts is the spectrum of a physical mixture of the amorphous drug and the polymer (refs 23, 25, 26); in this paper the shifts are reported relative to neat amorphous curcumin and the neat polymer. Acid–base interactions are seen differently: proton transfer to a basic amine removes the bands of the non-ionized amine, so their disappearance is evidence of ionization. C10·C12·R3·R4

Acid–base (ionic) interactions and pKa. E100 is a methacrylate copolymer carrying dimethylaminoethyl groups; it is a polybase whose basic monomer has an average pKa of 8.4 (ref 39). Curcumin's three ionizable groups have pKa values of 10.51, 9.88 and 8.38 (ref 40). The authors set these values side by side before concluding, from the IR data, that E100 forms ionic interactions with curcumin: proton transfer from curcumin's acidic phenols to E100's amines gives ion pairs instead of hydrogen bonds. Earlier studies linked such ionic drug–polymer interactions to better miscibility and physical stability (Miyazaki et al., ref 33; Weuts et al., ref 34; Yoo et al., ref 35). M5·M1·R23

Glass transition and the Fox equation. The glass transition temperature (Tg) marks where an amorphous solid softens; molecular mobility, and with it the crystallization rate, rises steeply above it. The Tg of an ideal two-component glass can be estimated from the Fox equation, the only numbered equation in the paper (Eq. 1): E1

Here $T_{g,\mathrm{m}\mathrm{i}\mathrm{x}}$ is the Tg of the mixture, $w_1$ and $w_2$ are the weight fractions of the two components and $T_{g1}$ and $T_{g2}$ their own glass transition temperatures. In plain terms: the reciprocal Tg of the blend is the weight-averaged reciprocal of the component Tgs, which is what free-volume additivity and ideal mixing predict; because it uses reciprocals, the temperatures must be absolute (kelvin). The authors follow the argument that strong specific interactions reduce free volume and push the measured Tg above this prediction (a positive deviation), whereas a net loss of interactions on mixing pushes it below (a negative deviation). E1·C40

PXRD, detection limits and polymorphs. An X-ray amorphous sample shows only a broad halo; crystals add sharp reflections. In this study spiking crystalline curcumin into a dispersion showed that crystals become visible at about 5–10 wt%, so "amorphous" here means below that level. Curcumin has three known polymorphs: the stable form I and the metastable forms II and III reported in 2011 by Sanphui et al. (ref 28). Form I has characteristic reflections at 8.9, 17.4, 23.5 and 25.1° 2θ; form II is identified by reflections at 13.9, 26.0 and 27.1° 2θ. Form II is reported to dissolve faster and to be more soluble than form I. P6·R15·R17·R20

Methods in detail

The study has three experimental arms and one literature comparison: IR spectroscopy of thin films to measure curcumin–polymer interactions, a storage study of milled powders to measure crystallization, thermal and chemical analysis of the same powders, and a comparison of curcumin–PVP spectra with resveratrol–PVP spectra from the group's earlier paper. P1·P5·P7·R9

Materials

Curcumin came from Spectrum Chemical Manufacturing Corp. The polymers were PVP K29/32 (Mw 58 000, Sigma-Aldrich), HPMC grade 606 (hypromellose USP substitution type 2910, Mw 35 600, Shin-Etsu), HPMCAS grade AS-MF (Mw 17 000, Shin-Etsu), Eudragit E100 (Mw 47 000, Rohm GmbH) and CMCAB 641-0.5 (Mw 35 000, Eastman). All polymers were dried over P₂O₅ for at least a week. The structures in Figure 2 show what each can offer curcumin: PVP has only the amide carbonyl of its pyrrolidone ring, a strong acceptor; HPMC has hydroxyl and ether groups but no carbonyl; HPMCAS and CMCAB add ester carbonyls (and, for HPMCAS and CMCAB, carboxylic acid substituents) to the cellulose backbone; E100 has ester carbonyls and basic dimethylamino groups. M1–M6

Contribution 1: measuring curcumin–polymer interactions in films

The problem. To test whether curcumin's internal hydrogen bonds leave room for polymer interactions, the interactions must be measured in a well-mixed amorphous state. What they did. Curcumin and polymer at several dry-weight ratios were dissolved together in 1:1 (w/w) dichloromethane/ethanol; two or three drops were placed on a thallium bromoiodide (KRS-5) crystal and spun at 500 rpm for 18 s and then 2500 rpm for 30 s. Spectra were recorded in absorbance from 4000 to 500 cm⁻¹ at 4 cm⁻¹ resolution with 128 co-added scans, under dry-air purge, keeping the absorbance in the region of interest between 0.6 and 1.2. Films of 20, 40, 60 and 80 wt% curcumin are shown for every polymer (Figures 4–8); peak positions are tabulated at 80% curcumin (Tables 1 and 2). The reference spectra are neat amorphous curcumin and the neat polymers. P1·P4·R3·R4

Why spin coating. Neat curcumin could be made amorphous by spin coating on a small scale, but not by rotary evaporation; a solvent-free route, cryomilling in liquid nitrogen for 30 cycles of 2 min at 10 strikes/s with 2 min cooling (1 h of milling), gave larger amounts of amorphous powder. The authors link this to curcumin being an intermediate crystallizer. R1·P3·C9

Contribution 2: the storage study

What they did. Curcumin and polymer at 3:1 by dry weight were dissolved in the same solvent mixture, rotary-evaporated with a 60 °C water bath, dried under vacuum at room temperature for 24 h and cryomilled (two cycles of 2 min at 2 strikes/s with 2 min cooling) to a uniform powder. These 75:25 curcumin–polymer powders and cryomilled amorphous curcumin were stored in desiccators at six conditions for up to 350 days and sampled periodically. P2·P3·P5

Table A. Storage conditions (Storage treatments, p.978). P5

Crystallinity was read by PXRD (Shimadzu XRD-6000, Cu-Kα, 5–35° 2θ at 8°/min, 0.04° steps), with the 2θ scale checked each day against the [111] reflection of a silicon standard (28.423–28.463°), whose intensity was also used to normalise that day's data. A sample with sharp peaks was called crystalline. Spiking crystalline material into a dispersion set the detection limit at about 5–10 wt% crystals. The read-out reported for each sample and condition is the number of days before crystalline material was seen (Table 3). The sampling interval and the number of samples per time point are not stated. P6·R13·X10

Contribution 3: Tg, the Fox comparison and chemical stability

Tg was measured by DSC (TA Q2000, nitrogen purge 50 mL/min, 3–7 mg in pin-holed aluminium pans): samples were heated to 120 °C, quench-cooled to 0 °C and reheated to 200 °C at 10 °C/min, and transitions were taken from the second heating. The PVP dispersions needed modulated DSC (2 °C/min, ±1 °C every 60 s) to resolve their Tg. Table 4 reports onset values and compares them with the Fox equation. Curcumin content was measured by UV-Vis absorbance at 424 nm against a standard curve, with differences tested by one-way ANOVA and Tukey's post hoc test (p < 0.05). P7·P9·P8

Contribution 4: a polyphenol comparison from earlier work

To test the idea that curcumin's internal bond is the problem, the authors set the carbonyl region of curcumin–PVP films beside that of resveratrol–PVP films from their 2013 resveratrol paper (Figure 14). Resveratrol carries phenol groups similar to curcumin's, and PVP had been "an extremely effective crystallization inhibitor" for it. The resveratrol spectra and stability outcome are cited, not re-measured. R8·R9·C32

Experimental analysis

Amorphous curcumin keeps its intramolecular hydrogen bond

The IR spectrum of crystalline curcumin (the stable, monoclinic form) has no strong carbonyl band near 1700 cm⁻¹, where a free ketone of the di-keto form would absorb, and a prominent band at 1626 cm⁻¹, consistent with the hydrogen-bonded keto–enol tautomer (Figure 3). The amorphous spectrum shows the same pattern of peaks between 1700 and 1500 cm⁻¹, "indicating" that amorphous curcumin is predominantly the intramolecularly hydrogen-bonded keto–enol form; its OH band peaks at 3400 cm⁻¹. The authors call this "to be expected", given the lower calculated energy of the keto–enol tautomer, and add that because the phenol OH groups are bonded to the ortho methoxy groups in the crystal, "it might be anticipated" that curcumin hydrogen-bonds only weakly with polymers. R2·R3.a·C11·C7

How strongly each polymer hydrogen-bonds to curcumin

Table B. OH and polymer carbonyl positions in spin-coated films at 80% w/w curcumin (from Tables 1 and 2). HPMC has no carbonyl group. R3·R4

The cellulosic polymers. With 20% HPMC, HPMCAS or CMCAB, curcumin's OH maximum moved from 3400 to 3386, 3394 and 3399 cm⁻¹. From these shifts the authors infer the weakest hydrogen bonding with CMCAB and the strongest with HPMC among the three. The carbonyl bands agree for the two polymers that have one: CMCAB's moved by about 4 cm⁻¹ and HPMCAS's by 8 cm⁻¹. In the spectra of Figures 4 and 5, the curcumin bands near 1626, 1600–1580 and 1510 cm⁻¹ grow as curcumin content rises while the polymer carbonyl band shrinks, without a separate lower-wavenumber carbonyl band. Overall, the authors write, the IR results "indicate weak hydrogen bonding" between the cellulosic polymers and curcumin. R3.b–d·R4.c–d·R5·C13

Note by the AI writer (not discussed in the paper). The OH positions in Table 1 are maxima of composite bands, and the neat cellulosic polymers have their own OH bands near 3450–3480 cm⁻¹ in Figures 4–6. With no physical-mixture spectra for comparison, shifts of 1–14 cm⁻¹ could partly reflect band overlap, so the order HPMC > HPMCAS > CMCAB is best read as tentative. X9

E100. With 20% E100 the OH maximum was 3396 cm⁻¹, a shift of about 4 cm⁻¹, and the E100 carbonyl moved by only 2 cm⁻¹ (1730 to 1728 cm⁻¹), the smallest shift of all; the authors read this as hydrogen bonding that is "limited or weak". R3.e·R4.e·C15

PVP. PVP behaved differently, as its amide carbonyl is a strong acceptor (Van Eerdenbrugh and Taylor, ref 22). At 80% curcumin the curcumin OH maximum lay around 3187 cm⁻¹, far below 3400 cm⁻¹ and the lowest of all films, which "suggests" the strongest or most extensive hydrogen bonds. A new shoulder appeared at 1655 cm⁻¹, 27 cm⁻¹ below the free PVP carbonyl at 1682 cm⁻¹, which "confirms curcumin–polymer hydrogen bonding", and the ratio of free to bonded carbonyl rose with polymer content (Figure 7). In the figure, the 80% curcumin OH band is a broad plateau from about 3380 to 3190 cm⁻¹ rather than a sharp peak, and the bonded carbonyl component becomes the broad maximum (about 1662 cm⁻¹) at 60% curcumin. R3.f·R4.a·R6·C14

E100 forms an acid–base pair with curcumin

Neat E100 shows two bands at 2820 and 2770 cm⁻¹ that are assigned to non-ionized dimethylamino groups (ref 32). As curcumin content rose these bands weakened quickly and, by the text, were "fully eradicated" at 60% curcumin (Figure 8, upper panel; in the figure they are clear at 20%, weak at 40% and at most a faint shoulder at 60%). Their loss "provides evidence for an acid–base interaction" between curcumin's acidic phenol OH groups and E100's amino groups, that is, for ionized amines. R7·C16

A second observation fits this reading. The E100 dispersion was rich dark red, whereas curcumin and the other four dispersions were orange (Figure 9). Its curcumin content was not lower than that of the other dispersions, so chemical degradation "did not appear to be the cause"; curcumin is known to darken at high pH (ref 36), and the authors attribute the colour to a change in curcumin's ionization state caused by the basic polymer, "in good agreement" with the IR. R10·R11·R25·C18·C20

Note by the AI writer. The text cites "Figure 8b" for the amine bands, but they lie in the OH-region panel; the panels of Figure 8 carry no letters. X4

Which dispersions stayed amorphous

Table C. Days before crystalline material was observed by PXRD in 75% w/w curcumin dispersions, with their Tg (from Table 3). ">": monitoring stopped at that time; (a) only partial crystallization; (b) crystallized into a metastable polymorph. R13

Three regimes appear. At the mildest condition (0% RH, 25 °C) every dispersion stayed amorphous for over 350 days and neat amorphous curcumin for over 100 days, so the polymers cannot be told apart. At the two harshest (90 °C, 0% RH; 60 °C, 98% RH) every sample crystallized within one day, again without differentiation. In between, the polymer decided almost everything: stability "depended greatly" on the polymer and on the condition. R13·C21·C23

PVP was the poorest inhibitor. It delayed crystallization only "to some extent": 6 days against 3 for neat curcumin at 50 °C, 22 days against 10 (text) or 14 (Table 3) at 84% RH, and 16 days against 1 at 75% RH and 40 °C, faster than every other dispersion at the same condition. R13.a·R13.c·C24·X2

E100 was the best. It was stable at 0% RH at 25 and 50 °C and was the only dispersion stable at both humid intermediate conditions: still amorphous after 350 days at 84% RH and 200 days at 75% RH, 40 °C, when monitoring stopped. The authors add that at the harsh conditions the E100 dispersion stayed X-ray amorphous "considerably longer" than the others. R13.b·C25

Note by the AI writer. Table 3 records 1 day for every sample at both harsh conditions, and no shorter-interval data are shown, so the claimed E100 advantage there cannot be checked from the paper. X6

The cellulosics were in between. They gave "some enhancement". HPMC was the most effective of them, stable at 0% RH at 25 and 50 °C, with first crystals after 75 days at 84% RH and 100 days at 75% RH, 40 °C. HPMCAS and CMCAB behaved alike: stable at 0% RH and 25 °C for up to 350 days, crystalline after 75 days at 0% RH and 50 °C, after 30 days at 75% RH, 40 °C and after 75 days at 84% RH. Their crystallization was only partial: at several intermediate conditions it stopped part-way and no further growth was seen over the study. After 250 days, for example, the CMCAB dispersion showed sharp peaks near 17.3 and 25.6° 2θ on the amorphous halo at 40 °C/75% RH and 25 °C/84% RH, a weak peak at 50 °C/0% RH and none at 25 °C/0% RH (Figure 12, read approximately). R13.d–f·R16·C26·C27

Neat amorphous curcumin, the negative control, crystallized as form I wherever it crystallized. At 50 °C and 0% RH its PXRD pattern was still a halo after 2 days, and the first form I reflections (about 14.6, 17.3, 24.6, 25.6 and 27.4° 2θ) appeared on day 3 and grew by day 6 (Figure 11, read approximately). R14·R15

Note by the AI writer. The day values are the first sampling times at which crystals were seen above a detection limit of about 5–10 wt%; with the sampling interval not stated, small differences such as 75 versus 100 days are not resolved. X10

Tg does not explain the ranking

All five dispersions had Tg values between 61.7 and 68.3 °C, against 43.6 °C for amorphous curcumin (Table 3). E100, the best inhibitor, and PVP, the worst, differ by 0.4 °C. The pure polymers span 41 °C (E100) to 159 °C (PVP), again without a trend in stability. The authors conclude that the stability differences "could not be explained by small differences in Tg values"; "Certainly, Tg does not explain" E100's high or PVP's low stability, and "Clearly", stability is hard to predict from Tg alone. R13·R18·C22

Table D. Onset Tg (°C) of the polymers and dispersions and Fox predictions as printed (from Table 4; curcumin 44 °C), with the AI writer's recalculation in the last two columns (not in the paper). R18·E1·X1

Table 4 compares the observed onset Tg with the Fox equation. As printed, the E100 dispersion lies 20 °C above its prediction, above the Tg of either pure component, which the authors call "consistent with" the ionic curcumin–E100 interaction. All other dispersions lie well below their predictions, by 15 (HPMCAS) to 33 °C (PVP), which "may suggest a net loss of interactions" arising from curcumin–polymer hydrogen bonding, as reported for other hydrogen-bonding systems (Shamblin et al., ref 48). R18·C41·C42

Note by the AI writer (not discussed in the paper). The printed predictions cannot be reproduced with the Fox equation as it must be used, in kelvin and at the stored composition of 75 wt% curcumin. They are reproduced exactly (41.7, 96.2, 92.1, 80.4 and 89.6 °C) only if the Celsius values are inserted directly and curcumin is given a weight fraction of 0.25, matching the Table 4 caption ("25% drug loading") but not the dispersions that were made and measured. Recalculated in kelvin at 75% curcumin, the predictions become 43.2, 66.6, 64.4, 58.5 and 63.1 °C, and the deviations shrink to −3.6 to +6.5 °C for the four hydrogen-bonding polymers. Only the large positive deviation of E100 (+18.8 °C) survives. The caption also describes the deviation as calculated minus observed, while the printed numbers are observed minus calculated. X1·E1

PVP binds curcumin far less than it binds resveratrol

Figure 14 places the PVP carbonyl region of curcumin–PVP films beside resveratrol–PVP films from the earlier study. With curcumin the PVP maximum stays at about 1682 cm⁻¹ at 20%, gains a shoulder near 1660 cm⁻¹ at 40%, and becomes a broad maximum near 1662 cm⁻¹ only at 60% curcumin. With resveratrol a shoulder near 1657 cm⁻¹ is visible at 10%, two maxima (about 1680 and 1655 cm⁻¹) at 30%, and the 1655 cm⁻¹ band dominates at 50% (values read from the figure, approximate). The authors read this as "much less hydrogen bonding" between PVP and curcumin, and note that the stability result is "in direct contrast" to resveratrol, for which PVP was "an extremely effective crystallization inhibitor". R8·R9·C32

Because the two polyphenols carry similar phenol groups, "it can be supposed" that curcumin's intramolecular hydrogen bond is "indeed the critical factor" in its weaker interaction with PVP. C33

Note by the AI writer. The comparison uses different weight fractions (20–60% curcumin, 10–50% resveratrol) and earlier data, and the text cites it as "Figure 11" (the curcumin PXRD figure). Within this paper PVP still gives curcumin the largest shifts of all five polymers, so the comparison shows a difference in degree between two polyphenols rather than a test of the intramolecular bond itself. X13·X4

Cellulosic polymers lead to metastable form II

After one day at 98% RH and 60 °C, curcumin crystallized from all three cellulosic dispersions as form II, identified by reflections at 13.9, 26.0 and 27.1° 2θ and by the absence of the 17.4° form I peak, whereas the E100 (labelled "EUD") and PVP dispersions gave form I (Figure 13). Form II also appeared in the HPMCAS dispersion at 0% RH and 90 °C, and it did not convert to the stable form during the study. The authors note that this metastable polymorph has a higher dissolution rate and solubility than form I (Sanphui et al., ref 28). R17·R20·C28

Note by the AI writer. In Figure 13 the CMCAB trace also shows a small peak near 17.2° 2θ inside the boxed form I region, so that product may contain some form I. The two form II peaks in the box appear at about 25.4 and 26.6° 2θ and look resolved, whereas the text gives 26.0 and 27.1° 2θ "not resolved by our instrument" (plot readings, approximate). The paper also gives the same CCDC reference code, BINMEQ, for form I and form II. X11·X12·X3

Chemical stability

By the text, all dispersions were chemically stable for up to 200 days at every condition except 98% RH and 60 °C, and crystalline curcumin did not degrade at any condition. Figure 10 shows the E100 dispersion only: curcumin content stayed between about 96 and 100% for about 224 days under the two milder conditions plotted, but at 60 °C and 98% RH it fell to about 92% by day 28 and about 85% by day 56 (read from the figure, approximate). The authors set this loss aside because those samples had already crystallized after one day. R11·R12·C19

Note by the AI writer. One series in Figure 10 is labelled 25 °C (75% RH), a combination that is not among the storage conditions (25 °C was paired with 84% RH, and 75% RH with 40 °C). X5

Discussion

The authors' explanation

The Discussion builds one argument from these observations. Extensive intramolecular hydrogen bonding persists in every curcumin polymorph (centred in form I, off-centre in the planar forms II and III), "suggesting that such H-bonding is energetically favorable", and the IR spectra show the same arrangement in the amorphous form. It is then "reasonable to expect" that polymers able to interact only through hydrogen bonds "might be poor curcumin crystallization inhibitors", because little opportunity for strong or extensive hydrogen bonding remains. C29·C30

Why PVP fails. PVP has only acceptor groups, so it "suffer[s] the additional disadvantage" of being able to bind only a curcumin donor; polymers with both donors and acceptors "may gain an advantage" by binding curcumin's donors and acceptors and so "may thus be able to better disrupt curcumin self-associations". This "may well explain why curcumin–PVP was the least stable" dispersion. C31

Why the cellulosics help a little. They are "slightly better" inhibitors, "presumably because they can form both donor and acceptor hydrogen bond interactions with curcumin"; the interactions "appear to be fairly weak" in the IR spectra, so these dispersions still crystallize under challenging humidity and temperature. C34

Why E100 works. E100's ionic interactions with curcumin are not blocked by the internal hydrogen bonds. The authors place this in a line of earlier reports: salt formation between loperamide and poly(acrylic acid) correlated with miscibility and stability (Weuts et al., ref 34); proton transfer between poly(acrylic acid) and p-aminoacetanilide inhibited recrystallization (Miyazaki et al., ref 33); and ionic drug–polymer systems stayed amorphous for up to 50 days at 60% RH and 25 °C while non-ionic ones recrystallized (Yoo et al., ref 35). E100 itself forms ionic interactions with resveratrol, dexamethasone phosphate and indomethacin (refs 23, 41, 42). They conclude that ionic interactions "appear to be the reason" for the curcumin–E100 stability, and in the Conclusions state without a hedge that these interactions "were not hindered, and when formed, resulted in increased physical stability". C17·C37·R23·R24

Support from other molecules. A similar pattern had been seen with nimesulide, where chemically different polymers hardly differed in slowing crystal growth because they could not compete with the drug's intramolecular interactions (Kestur et al., ref 21), and the result is "in line with" Etter's rule that six-membered intramolecular hydrogen-bonded rings are typically preferred over intermolecular bonds (refs 37, 38). The authors generalise: "it appears that for a compound like curcumin, which forms intramolecular hydrogen bonds, hydrogen bonding with polymers is impeded and it is more difficult to inhibit crystallization". C36·C35·R22

What it suggests for formulation

Stated by the authors. The best polymers for ASDs of compounds prone to intramolecular hydrogen bonding are "those that allow the formation of interactions other then hydrogen bonds (e.g. ionic interactions)" ("This suggests"). Tg alone does not predict stability. And because the cellulosic polymers steer curcumin to the more soluble form II, "there might still be benefits from formulating with a cellulosic polymer" if the amorphous form cannot be kept, "thus leading to a possible increase in bioavailability", although this "clearly would not be ideal from a regulatory perspective". C44·C22·C43

Note by the AI writer (not discussed in the paper). Three practical points follow from the data as they stand. First, the size of an IR hydrogen-bond shift is a poor guide to inhibitor performance for a drug whose donors and acceptors are already used internally: in this paper the order of shifts (PVP largest) is the reverse of the order of stability. Second, the E100 result points to acid–base pairing as a design lever for weakly acidic drugs; the closeness of the E100 monomer pKa (8.4) to curcumin's most acidic pKa (8.38) is the kind of property such a screen could start from, although the paper does not test it. Third, every stability result is for 75 wt% curcumin, a high drug loading; how the ranking of the five polymers would change at lower loadings is not known from this paper. X7·M5·M1·P5

Limitations

Limits the authors state. The authors acknowledge that no polymer differentiated curcumin at the mildest or the two harshest storage conditions, that PVP inhibited crystallization only "to some extent", that the cellulosic dispersions "fail to prevent curcumin crystallization under challenging humidity and temperature conditions", that E100 "was unable to prevent crystallization at the harsh conditions", and that relying on form II "clearly would not be ideal from a regulatory perspective". C23·C24·C34·C25·C43

The AI writer's own judgement.

• The paper's own data run against its premise. PVP shows by far the largest hydrogen-bond shifts (OH at 3187 cm⁻¹, carbonyl 27 cm⁻¹) and gives the poorest protection, while E100 shows the smallest hydrogen-bond shifts and gives the best. Within this paper, the extent of hydrogen bonding seen by IR does not track stability. The authors' acceptor-only and donor–acceptor explanations are plausible but no measurement tests them; no band was assigned to polymer donors bonding to curcumin acceptor sites. X7
• No direct control for the intramolecular bond. No curcumin analogue without the intramolecular hydrogen bond was measured. The only comparator is resveratrol–PVP, with spectra from an earlier study at different weight fractions. The causal role of the intramolecular bond is therefore the authors' inference. X13
• Ionic interaction inferred, not quantified. The acid–base interaction rests on the loss of two amine bands and a colour change; the extent of proton transfer was not measured and no polymer with E100's backbone but without amine groups was tested. C16·C37
• The Fox-deviation table is miscalculated. The printed predictions use Celsius temperatures and 25% curcumin; in kelvin at the stored composition the large negative deviations largely disappear, which removes the support for the "net loss of interactions" reading. X1
• Sample forms and compositions differ. Interactions were measured on spin-coated films at 80% curcumin, stability on milled powders at 75%. The sampling interval and replicates of the storage study are not stated, and the PXRD detection limit is about 5–10 wt%. X8·X10
• Several inconsistencies in the text. Neat curcumin at 84% RH is given 14 days in Table 3 and 10 days in the text; the E100 advantage at the harsh conditions is not visible in Table 3; one Figure 10 series has a condition that was not used; figure cross-references point to the wrong figures in three places; and the same CCDC code is given for form I and form II. X2·X6·X5·X4·X3
• Scope. One drug, one loading, five polymers, no dissolution or supersaturation measurements. P5·A1

Conclusion

This paper tested the group's premise that stronger drug–polymer interactions mean better crystallization inhibition on a drug that resists hydrogen bonding with polymers. Curcumin kept its intramolecularly bonded keto–enol form when amorphous, and four of the five polymers produced only small IR shifts. In 75% curcumin dispersions stored at the two humid intermediate conditions (84% RH, 25 °C; 75% RH, 40 °C), E100 kept curcumin amorphous beyond 350 and 200 days, the cellulosics for 30–100 days with only partial crystallization, and PVP for 16–22 days, while Tg was nearly the same for all. R2·R3·R4·R13

The authors explain the ranking by competition with curcumin's internal hydrogen bonds: hydrogen-bond-only polymers lose that competition, and the basic E100, which pairs with curcumin as acid and base, does not. They therefore suggest polymers that can form non-hydrogen-bond interactions, such as ionic ones, for compounds prone to intramolecular hydrogen bonding. The direction of the stability data is clear; the mechanism is an inference that this paper does not test directly, and the paper's own IR data, with PVP bonding most and protecting least, show that the extent of hydrogen bonding alone does not predict protection for this drug. C33·C37·C44·X7·X13