Lumefantrine amorphous solid dispersions made the cheap way: how the polymer's acid groups decide stability, and how they can cost release

Background and motivation

Lumefantrine is the long-acting partner of artemether in Coartem, a fixed-dose combination of 20 mg artemether and 120 mg lumefantrine, both crystalline, and the only artemisinin-based combination approved in the United States (Tan and Arguin 2017) R21. Artemether clears most of the parasites quickly; lumefantrine stays in the body to prevent recrudescence and resistance R21. The problem is lumefantrine's absorption: it is low, very variable and strongly dependent on food, so the tablets are taken after meals, and treatment failures have been attributed to poor absorption or dosing on an empty stomach (Borrmann 2010; Mizuno 2009; Repetto 2011) C4. Malaria patients are often nauseated and, in poor regions, may not have enough food, so a formulation that does not need food would help C4.

An amorphous solid dispersion (ASD) is the obvious tool. A lumefantrine ASD has already raised oral absorption over the crystalline reference in healthy volunteers (Jain et al. 2017); it was made by hot melt extrusion, but the polymer was not described and no physical stability testing was reported C1. The catch is manufacturing. Spray drying and melt extrusion, the two large-scale routes, need costly equipment, and the authors are interested in cheaper routes, such as wet granulation or anti-solvent precipitation, that could be used where large capital investment is not possible, for example for infectious-disease drugs in low-income countries C3. These cheaper routes remove the solvent slowly or expose the drug to water, so the authors expect them to work "likely" only for drugs with a low tendency to crystallize during manufacture C3.

Two earlier lines of work frame the study. First, the general ASD picture: a polymer should be miscible with the drug and interact with it, a high-Tg polymer raises the Tg of the dispersion and lowers mobility, and mixing lowers the drug's chemical potential, yet drug-rich and polymer-rich regions can still separate and crystallize (Qian 2010; Kothari 2015; Van den Mooter 2001; Purohit and Taylor 2015) C2. Second, a lumefantrine-specific finding: Song et al. (2016) showed with X-ray photoelectron spectroscopy (XPS) that in spray-dried lumefantrine ASDs at 40% loading, HPMCP and Eudragit L100 transferred a proton to about 75% of the drug's tertiary amine nitrogen, forming a drug–polymer salt, while HPMCAS transferred none; the extent followed neither the number nor the strength of the polymer's acid groups R22. Salt formation with an acidic polymer had also been linked to better physical stability for other basic drugs (Weuts 2005) C37.

The paper sets itself two goals: to find out whether simple, low-cost manufacturing can produce lumefantrine ASDs, and to characterise the phase behaviour of these ASDs during storage and release, both to understand the absorption gain reported earlier and to test how well they resist crystallization under accelerated storage A1·A2·A5. The authors chose five polymers: four bearing acid groups (HPMCP, Eudragit L100, cellulose acetate phthalate CAP, and HPMCAS) and the neutral polyvinylpyrrolidone–vinyl acetate copolymer PVPVA (Figure 1) M3·M4·M5·M6·M7. Artemether, the partner drug, was screened as well M2·R7.

Background concepts

Glass-forming ability. Not every drug can be held amorphous. Baird et al. (2010) sorted drugs into classes by heating, cooling and reheating the melt in a DSC. A class 1 compound crystallizes as soon as the melt is cooled; a class 3 compound cools to a glass and stays amorphous when the glass is reheated above its Tg. A class 3 drug is a good candidate for an ASD because it tolerates processing and storage better C7·C27.

Drug–polymer salt formation and how XPS sees it. Lumefantrine carries a tertiary amine, a weak base. Enteric polymers carry carboxylic acid groups: phthalate in HPMCP and CAP, succinate in HPMCAS, methacrylic acid in Eudragit L100 M1·M3·M4·M5·M6. When the acid hands its proton to the amine, the two form an ion pair in which the polymer is the counterion: a drug–polymer salt. In XPS, the N 1s photoelectrons of a positively charged nitrogen need more energy to escape, so a protonated nitrogen appears as a second peak at higher binding energy (here near 402 eV instead of 399 eV) R10·C14. Fitting the two peaks gives the fraction of drug that is protonated E2·R11. XPS samples only the top ~10 nm of a powder P15. In infrared spectra, proton transfer can show as a carboxylate band and as changes in the polymer's carbonyl band, which in the neat polymer is partly hydrogen-bonded to the polymer's own acid groups C11·C12.

Amorphous solubility and liquid–liquid phase separation. A dissolving ASD can raise the drug concentration far above the crystalline solubility, but only up to a ceiling: the amorphous solubility. Above it, the solution separates into a water-rich phase and a dispersed, drug-rich phase, a process called liquid–liquid phase separation (LLPS); the drug-rich phase appears as colloidal nanodroplets C15. Concentrations measured above the amorphous solubility after filtration therefore mean that some drug is present as such colloids C18.

The pyrene probe. Pyrene's emission spectrum has several vibronic peaks; the ratio of the third to the first peak, I3/I1, rises when pyrene sits in a less polar environment (Glushko et al. 1981) C15. Drug-rich nanodroplets are less polar than water and can take up pyrene, so I3/I1 stays flat while the drug is fully dissolved and rises once droplets form. Plotting I3/I1 against added drug therefore locates the amorphous solubility, and following I3/I1 during dissolution tells whether drug-rich droplets are present (Jackson et al. 2016) C15·C21.

Congruent and incongruent release. Release from an ASD is called congruent when drug and polymer go into solution in the same proportion, so the drug's release is governed by the polymer's dissolution. It is incongruent when the two leave at different rates. As the drug loading rises, release can switch from polymer-controlled to drug-controlled: a drug-rich layer forms at the surface and release slows sharply (Indulkar et al. 2019) C44. Drug-rich nanodroplets usually form only from ASDs with low loadings, typically 10–20% drug R23.

Non-sink dissolution and supersaturation. In a non-sink test, the medium receives far more drug than it could dissolve at the crystalline solubility, so any concentration above that solubility is supersaturation, and the degree of supersaturation is the concentration divided by the crystalline solubility E1·P7.

Methods in detail

The study has two kinds of sample, and keeping them apart matters for reading the results. Rotary-evaporated powders at two drug loadings were used for crystallinity, glass transition and storage; granules made by the cheap route, at one loading, were used for release. The method flow below shows the stages.

Making the dispersions and the granules

The rotary-evaporated dispersions were made at 6:4 and 4:6 w/w lumefantrine:polymer, that is 60% and 40% drug loading. Drug and polymer were dissolved in 8:2 v/v dichloromethane (DCM):methanol for CAP and HPMCP, in DCM alone for PVPVA, in 1:1 DCM:methanol for HPMCAS and in 1:1 DCM:ethanol for Eudragit L100, then evaporated on a rotary evaporator with a 40 °C water bath (Section 3.1) P1.

The granules are the paper's low-cost process. 1080 mg of polymer was dissolved in 10 mL of the same solvent system, 720 mg of lumefantrine was added (4:6), and the solution was poured slowly onto 3.6 g of a 1:1 w/w blend of microcrystalline cellulose (MCC) and anhydrous lactose while being mixed with a spatula as the solvent evaporated; the mass was then dried under vacuum overnight and cryomilled (Section 3.1) P2. No spray dryer, extruder or fluid bed is involved. Only the 4:6 ratio was granulated in this way; the 3:7 and 2:8 Lume:CAP samples used later are described only as "lower drug loaded dispersions" P3·X10.

Solid-state characterisation

X-ray powder diffraction (XRPD, 4–40° 2θ, Cu Kα) was run right after preparation and then on samples stored in open dishes at 40 °C/75% RH for up to 3 months P13·P14. Crystalline lumefantrine was heated in a DSC to 150 °C at 10 °C/min, cooled to −30 °C and reheated, which gives its melting point, its glass-forming behaviour and its Tg P4. The dispersions were run by modulated DSC (cooled to −10 °C, heated to 150 °C at 2 °C/min, modulation 1 °C every 60 s), and the onset Tg was read from the reversing heat flow, n = 3 P5. ATR-FTIR spectra (64 scans) were recorded for amorphous lumefantrine, the neat polymers and the 4:6 dispersions P6. XPS (monochromatic Al Kα, 20 eV pass energy for high-resolution spectra, charge neutralizer, C 1s C–C at 284.8 eV) was run on 4:6 ASDs; the N 1s envelope was fitted with Gaussian/Lorentzian components after a Shirley background to give the percentage of protonated drug P15·E2. The paper does not state whether the FTIR and XPS samples were rotary-evaporated powders or granules X5.

Solution tests

For the amorphous solubility, a 1 mg/mL lumefantrine stock in dimethyl acetamide was added stepwise to 10 mL of 50 mM pH 6.8 phosphate buffer containing 0.2 µg/mL pyrene, covering 0.5 to 12 µg/mL drug; pyrene was excited at 332 nm and the emission recorded from 350 to 450 nm; the experiment was run twice and the mean plotted (Section 3.5) P9. Prodan (1 µg/mL) is listed as a second probe, but no prodan results are reported P11.

Release was tested under non-sink conditions. Granules containing about 20 mg of lumefantrine were compressed into a pellet with 10% croscarmellose sodium and dropped into 100 mL of pH 6.8 phosphate buffer at 37 °C, stirred magnetically at 150 rpm; pH 6.8 was chosen because most of the polymers are insoluble in acid (Section 3.4) P7. Samples were filtered through 1 µm glass syringe filters and read with a UV dip probe at 310 nm over 2 h P7. The buffer is given as 10 mM in the methods and as 50 mM in the captions of Figures 8–10 and Table 3, and the paper does not resolve this X1. After 2 h the medium was passed through a 0.22 µm cellulose acetate filter and the filtrate measured by dynamic light scattering (DLS) and electrophoretic light scattering for size and zeta potential, averaging three runs P12. A second release run was made with 0.2 µg/mL pyrene in the medium, with samples filtered through 0.45 µm PTFE filters because glass filters adsorbed the probe P10. For CAP, the polymer's own release was followed by UV absorbance at 270 nm, where both drug and polymer absorb but the drug, at about 1–1.5 µg/mL, contributes little P8·C22.

For orientation (note by the AI writer, from the printed amounts, not stated in the paper): the granules hold 720 mg of drug in 5.4 g of solids, about 13% lumefantrine, so each release test used roughly 150 mg of granules; with 20 mg of drug in 100 mL, complete dissolution would give 200 µg/mL, far above the amorphous solubility, which is what makes the test non-sink P2·P7.

Experimental analysis

Which drug can be made amorphous at all

Lumefantrine melted with an onset of 129 °C, formed a glass on cooling and did not crystallize when reheated; its onset Tg is 18 (±1) °C (Figure 4, Table 1) M1·R5. This is class 3 behaviour, good glass-forming ability and good glass stability, and the authors take it as making lumefantrine "a good candidate" for an ASD C7·C27. Artemether behaved the opposite way: it melted at 87 °C and crystallized on cooling at a small undercooling, from about 60 °C (class 1, Figure S1) R6. Attempts to make artemether dispersions with various polymers gave partial crystallization, so artemether was dropped R7. The authors explain this "presumably" by its rigid structure and lack of hydrogen-bonding groups, add that an artemether dispersion "would likely" have a low Tg, and note that artemether does not need solubility enhancement because it is well absorbed C26·C25.

Crystallinity after preparation and in storage

Right after rotary evaporation, all five 4:6 dispersions showed only an amorphous halo in XRPD (Figure 2a) R1. At 6:4, the CAP, HPMCP and L100 dispersions were still X-ray amorphous, but the HPMCAS and PVPVA dispersions showed sharp lumefantrine peaks, near 5.6, 15.0, 18.7, 21.6, 23.0 and 25.4° 2θ (read from Figure 2b, approximate) R2. The authors read this as HPMCAS and PVPVA "not effective" at inhibiting crystallization during solvent removal at the higher loading C5.

Storage in open dishes at 40 °C/75% RH separated the polymers further (Figure 3) R3·R4. Table A collects the time points.

Table A. Crystallinity of the rotary-evaporated dispersions by XRPD (Figures 2 and 3, text of Section 4.1) and onset Tg by modulated DSC (Table 1, n = 3, SD in parentheses; ^a melting seen on heating above Tg). Neat polymer Tg: HPMCP 138, L100 194, CAP 161, HPMCAS 117, PVPVA 101 °C R1·R2·R3·R4·R8·M4·M3·M6·M5·M7.

At 40% loading the result is clear-cut: PVPVA crystallized within a week, HPMCAS within a month, and HPMCP, L100 and CAP not within three months; the authors' summary is that the most stable systems were the 4:6 CAP, L100 and HPMCP dispersions R3·C6. At 60% loading the three acidic polymers that had worked at 40% all eventually crystallized, CAP and L100 after 2 weeks and HPMCP after 6 weeks, with the HPMCP peak growing little up to 3 months (data not shown) R4. In the printed 6-week HPMCP trace no peak stands clearly above the noise, so the 6-week time point rests on the text X14.

The DSC results were, in the authors' words, "in broad agreement" with XRPD C10. Footnote a of Table 1 marks dispersions that showed a melting endotherm on heating, read as recrystallization during the DSC scan: 6:4 CAP and the 4:6 HPMCAS and PVPVA dispersions; the text adds that 6:4 L100 also showed a small endotherm, that the 4:6 HPMCP, CAP and L100 dispersions showed none, and that HPMCP showed none even at 60% loading, which is why HPMCP "appeared" the most effective inhibitor R8·C8·C9. All 4:6 dispersions with a clear Tg lie far above the drug's 18 °C, from 76 °C (HPMCAS) to 163 °C (L100), the L100 value being highly variable (SD 28 °C) R8. Table 1 lists the 6:4 L100 dispersion as "Crystallized", which conflicts with Figure 2b (amorphous after preparation) and with the text; the table entry is probably wrong X2.

Drug–polymer interactions: FTIR and XPS

In the carbonyl region, every acidic-polymer dispersion shifted the polymer's carbonyl band to higher wavenumber (Figure 5) R9. Lumefantrine itself has no carbonyl group, so the band belongs to the polymer; the authors expect it to be hydrogen-bonded to the polymer's own carboxylic acid groups in the neat polymer, and read the upward shift as a change in this hydrogen bonding caused by the drug C11. A new band near 1560 cm−1 appeared with CAP, HPMCP and L100, which the authors assign "likely" to carboxylate, "suggesting" that some proton transfer "may have occurred" C12. The region of the acid OH vibrations was too broad to interpret C13.

XPS gave the more direct evidence (Figure 6) R10. Lumefantrine alone shows one N 1s peak at 399 eV; with each acidic polymer a second component appears at 402 eV, assigned to protonated nitrogen as in Song et al. 2016, and since the polymers contain no nitrogen, both components are the drug's R10·C14. Table B sets the protonation next to the infrared shifts.

Table B. Percentage protonation of lumefantrine from N 1s peak fitting (Table 2; deviation in parentheses, n not stated) and carbonyl band positions from FTIR (Figure 5, Section 4.3) R11·R9.

The protonation spans a factor of four, from 16% with HPMCAS to 61% with CAP, and the authors call the extents "clearly different" R11·C14. As Song et al. had found, the extent did not follow the number of proton donors on the polymer ("no apparent correlation") C35·R22. Note by the AI writer (not discussed in the paper): the 1560 cm−1 region also holds a band of amorphous lumefantrine itself (about 1565 cm−1 in Figure 5a), so the infrared assignment supports, but does not by itself prove, proton transfer; the XPS component is the firmer evidence X11.

Protonation and stability

Placing Table A beside Table B gives the paper's central stability argument. HPMCAS, the acidic polymer that protonated the least drug (16%), was the only acidic polymer whose 4:6 dispersion crystallized within three months; L100, HPMCP and CAP, at 33–61% protonation, did not R3·R11. The authors state that physical stability "correlates" with the extent of protonation and that the more extensive salt formation "appears to account" for the excellent stability of the HPMCP, CAP and L100 dispersions at 40 °C/75% RH; in the abstract the differences "could be largely rationalized" by salt formation, and in the conclusions salt formation "contributed" to the stability C36·C37. They "probably" attribute the weaker HPMCAS result to its low share of carboxylic acid groups (Table S1, not available here) C34. PVPVA, which cannot form a salt, has "limited opportunities" for specific interactions; together with its hygroscopicity and the high log P of lumefantrine, this "most likely" makes the dispersion prone to amorphous phase separation in moisture and then to crystallization C31.

Amorphous solubility

Adding lumefantrine to buffer with pyrene left I3/I1 at about 0.504–0.507 from 0.5 to 2 µg/mL; above 2 µg/mL the ratio rose to about 0.516 at 3 µg/mL, 0.528 at 4, 0.541 at 6, 0.550 at 8 and 0.569 at 10 µg/mL (Figure 7, read approximately, n = 2) R12. The authors explain the rise by pyrene partitioning into a growing amount of drug-rich nanodroplets and place the amorphous solubility between 2 and 3 µg/mL in 50 mM pH 6.8 phosphate buffer C16·R13. The crystalline solubility was below 80 ng/mL, the detection limit of their method, so the amorphous form offers at least a 25-fold solubility advantage (2/0.08; arithmetic by the AI writer from the printed bounds) M1·R13.

Release from the granules

The 4:6 granules separated into two groups (Figure 8) R14. Table C gives the concentrations.

Table C. Lumefantrine in solution (µg/mL) from 4:6 granule pellets in 100 mL pH 6.8 phosphate buffer at 37 °C, 1 µm glass filter; read from Figure 8 (approximate) and text values of Section 4.6 R14.

HPMCAS granules released quickly but stopped near 21 µg/mL; L100 granules rose to 28 µg/mL at 2 h; HPMCP granules released slowly but steadily and caught up with L100 by 2 h; PVPVA and CAP granules stayed at about 1–2 µg/mL R14·C17. The authors rank L100 fastest, then HPMCAS, and say CAP released the least, "not even achieving the amorphous solubility" C17. Two details of the figure do not match these words: at 15 min HPMCAS (about 20.5 µg/mL) was ahead of L100 (about 17.7), and at 2 h the CAP marker reads about 2.2 µg/mL against 1.5 in the text, slightly above PVPVA (about 1.6), so CAP and PVPVA cannot really be ranked X8·X3. Since the crystalline solubility is below 80 ng/mL, every acid-polymer ASD supersaturated; the authors state "at least 20 fold", but concentrations above 1 µg/mL over a solubility below 80 ng/mL only guarantee more than about 12-fold R19·C40·X4.

The excess drug is in nanodroplets

Concentrations far above 2–3 µg/mL mean that colloidal drug passed the filter C18. After 2 h the 0.22 µm filtrates of the L100, HPMCAS and HPMCP granules scattered light: Z-average 116 (16), 179 (15) and 116 (6) nm, with zeta potentials of −27 (2.6), −20 (1) and −22 (3) mV (Table 3, n = 3) R15. The CAP filtrate had too low a count rate to measure R15. The authors read the high charge as "most likely" due to ionized polymer and/or drug at the particle–water interface C20.

DLS shows particles but not whether they are amorphous. For that the authors repeated the release with pyrene in the medium (Figure 9) R16. With CAP, I3/I1 stayed at about 0.50–0.51, the value for fully dissolved drug; with HPMCP it rose from about 0.51 to 0.69 as the concentration rose from about 1.2 to 23 µg/mL; with L100 from about 0.55 to 0.67; and with HPMCAS, already at about 13–14 µg/mL at 15 min, it stayed at about 0.60–0.64 R16. The authors take the link between concentration and I3/I1 as confirming that the high concentrations come from amorphous nanodroplets formed by liquid–liquid phase separation, small enough to pass the 0.45 µm PTFE filter C21. They point out that this is unusual at 40% loading, since nanodroplets usually form only from ASDs with 10–20% drug; for lumefantrine ASDs the cut-off loading "appears to be much higher" C41·R23.

Note by the AI writer (not discussed in the paper): the concentrations in the pyrene run are clearly lower than in Figure 8 for the same polymers (at 2 h, HPMCAS about 12.8 against 21 µg/mL, L100 about 21.5 against 27 µg/mL), and the CAP concentrations in Figure 9a are plotted slightly below zero; the filter used for these concentrations and the replicate number are not stated X9. The trend of I3/I1 with concentration within this run is what carries the argument, and that trend is clear.

The CAP anomaly

CAP was the polymer that protonated the most drug, kept the 4:6 dispersion amorphous for 3 months, and yet released almost no drug R11·R3·R14. To find out why, the authors measured the polymer itself: CAP in solution rose from about 8.6 µg/mL at 15 min to about 49 µg/mL at 2 h, with wide error bars (about 33–65 µg/mL at 2 h; Figure 10) R17. Complete dissolution would give 300 µg/mL, so less than 20% of the CAP dissolved, although CAP is expected to dissolve at pH 6.8; the polymer still left faster than the drug, and the authors call the release incongruent R17·C23.

Lowering the loading changed the picture completely (Figure 11) R18. Table D gives the concentrations.

Table D. Lumefantrine in solution (µg/mL) from Lume:CAP samples at three loadings, read from Figure 11 (approximate). How the 3:7 and 2:8 samples were made is not described R18·X10.

The authors considered and rejected one explanation: strong interactions "could conceivably" give an insoluble drug–polymer complex, as PVP and polyacrylic acid precipitate together in water, but the much better release at 20% loading "indicates that insoluble complex formation is not the likely explanation" C43. Instead, they find it "likley" (sic) that release shifted from polymer-controlled to drug-controlled as the loading rose, the mechanism described for ritonavir–copovidone ASDs by Indulkar et al. (2019), and that this drop-off happens at a lower loading for CAP than for the other acidic polymers "could, however, be related" to its stronger drug–polymer interaction C44. They close the point with "Clearly more studies are warranted" C44. They also note that this runs against earlier work on salt-forming ASDs, in which ionic interactions improved release (Weuts 2005) C42.

Discussion

How the authors connect the findings. The argument runs from chemistry to stability to release. The chain below sets it out, with the outcome for each polymer.

The authors' conclusion follows from the two ends of this chain: the polymer that best promotes stability need not give the most release, and "a balance between the ability of a given polymer to promote physical stability and drug release may need to be sought" C47. In their conclusions, Eudragit L100 "appeared" to give the best balance at 40 wt% loading, while CAP "formed strong, ionic interactions with the drug and inhibited drug release" C46.

The process. The authors argue that drying a drug–polymer solution on an excipient powder removes the solvent much more slowly than spray drying, which raises the risk of crystallization, since faster evaporation favours miscibility C29. They nevertheless conclude that the wet granulation approach "is a viable manufacturing route as long as the appropriate polymers and drug loading are selected" C30. On the dosage form, a 120 mg dose at 40 wt% loading needs 300 mg of ASD, which they call reasonable, and a high loading is "preferable" for a dosage form of reasonable size R20·C38. PVPVA, a common choice for melt extrusion, and the route of the Jain et al. formulations, was "clearly not effective" here C31.

What it suggests for formulation (note by the AI writer, not stated in the paper). For a weakly basic drug, salt formation with an enteric polymer is a usable stabiliser, and XPS gives a quick, quantitative read of it. But the strongest salt former was the worst releaser at the loading of interest, so a polymer screen based on interaction strength alone would have picked the wrong polymer. The practical sequence the data support is: screen acidic polymers for protonation and accelerated stability, then test release at the target loading before choosing. Between the two polymers that passed both tests, HPMCP and L100 ended at nearly the same concentration at 2 h; HPMCP released more slowly but was the more stable at 60% loading, so the choice between them depends on how fast release must be and how much loading margin is wanted.

Open questions. The authors leave open why CAP's release falls off at a lower loading than the other acidic polymers, and whether its strong interaction is the cause ("Clearly more studies are warranted") C44. The data also leave open, in the AI writer's reading: whether the granules themselves, which dry slowly and contain crystalline lactose, are as stable as the rotary-evaporated powders X5; at what loading release falls off for L100, HPMCP and HPMCAS, which were tested at 40% only R14·R18; and whether protonation causes the stability or merely accompanies it, which would need protonation varied within one polymer C37·X6.

Limitations

Limitations stated by the authors. The authors are explicit about the CAP result: insoluble complex formation was ruled out, a shift to drug-controlled release is only "likley", the link to the strong interaction "could" hold, and "more studies are warranted" C43·C44. They acknowledge that the acid OH region of the infrared spectra could not be interpreted C13, that the carboxylate assignment is "likely" and proton transfer only "may have occurred" C12, and that the amorphous nature of the colloids needed a separate pyrene experiment beyond DLS C19·C21. They mention that slow solvent removal in granulation makes these routes suitable only for drugs with low crystallization tendency C3·C29.

The AI writer's own judgement (1): the granules were not the samples tested for stability. The title, the abstract and the conclusions speak of physically stable ASD granules made by solvent granulation, but every crystallinity, storage and Tg measurement in the paper was made on rotary-evaporated dispersions, as the captions of Figure 2, Figure 3 and Table 1 state; the granules were used only for release, DLS and pyrene tests, and the sample type for FTIR and XPS is not stated X5. Granules dry more slowly, which the authors themselves name as a crystallization risk, and they contain crystalline lactose, which would complicate XRPD C29. The stability claim for the granules is therefore an inference from powders of the same drug:polymer ratio, not a measurement.

The AI writer's own judgement (2): the protonation–stability correlation is a two-group contrast. At 40% loading, the three polymers at 33–61% protonation were all stable for 3 months and cannot be ranked against each other, so the "correlation" rests on HPMCAS (16%) against the rest X6. At 60% loading, where the polymers do separate, the order of first peaks (HPMCP last, CAP and L100 first) does not follow protonation (CAP 61% > HPMCP 48% > L100 33%), but protonation was measured only at 40% X6·R4. Neat lumefantrine itself reads 7 (5)% protonation, which sets a baseline for the method R11. The data fit the authors' idea, but they do not test it.

The AI writer's own judgement (3): inconsistencies in the printed values and conditions. The release buffer is 10 mM in the methods and 50 mM in the figure captions X1. Table 1 lists the 6:4 L100 dispersion as "Crystallized" although Figure 2b and the text show it amorphous X2. The text gives 1.5 µg/mL for CAP at 2 h where the plotted marker reads about 2.2 µg/mL X3. The stated "at least 20 fold" supersaturation is not supported by the stated bounds, which give more than about 12-fold X4. The pyrene release run gave lower concentrations than the main run, with CAP values below zero, without comment X9. The replicate number and the meaning of the error bars are not stated for the release data or for Table 2 X12. The 3:7 and 2:8 CAP samples are not described X10. The amorphous solubility was measured in polymer-free buffer with a dimethyl acetamide stock, at an unstated temperature, so release concentrations near 2–3 µg/mL cannot be placed above or below it with confidence X13. None of these changes the main ranking of the polymers, but each limits how far the numbers can be compared with other studies.

Conclusion

For lumefantrine at 40% drug loading, the polymer decides almost everything. The acidic polymers HPMCP, Eudragit L100 and CAP kept rotary-evaporated dispersions amorphous for three months at 40 °C/75% RH, HPMCAS did not, and the neutral PVPVA crystallized within a week R3·C6. XPS shows that the acidic polymers protonate part of the drug, from 16% with HPMCAS to 61% with CAP, and the authors argue that this salt formation "appears to account" for the stability R11·C37. Granules made by pouring the drug–polymer solution onto MCC and lactose released lumefantrine far above its 2–3 µg/mL amorphous solubility with L100, HPMCP and HPMCAS, as 116–179 nm drug-rich nanodroplets, at a loading higher than usual for nanodroplet formation R14·R13·R15·C41. CAP, the strongest salt former, released almost no drug at 40% loading but released well at 20%, which the authors tentatively read as an early switch to drug-controlled release R18·C44. Their conclusion is a trade-off: L100 "appeared" to give the best balance, and stability and release "may need to be" balanced when choosing a polymer C46·C47.