Diluting the drug at the particle surface: hierarchical co-precipitated amorphous dispersions of grazoprevir at high drug loading

Background and motivation

Amorphous solid dispersions (ASDs) are popular because the amorphous drug can supersaturate the gastrointestinal fluid and drive absorption (Taylor 2016, ref 12) C2. Most commercial ASDs are made by spray drying or hot melt extrusion (HME), but both have limits: spray drying needs a volatile solvent that dissolves the drug, and HME suits poorly drugs with a high melting point or thermal instability (refs 14, 15) C1. Co-precipitation is an alternative. A solution of drug and polymer is mixed into an anti-solvent under high shear in a rotor–stator device, and amorphous particles precipitate; the product is a co-precipitated amorphous dispersion (cPAD) (refs 16–21) C1·C3. Earlier work found that cPAD material resembles spray-dried and extruded material by DSC and PXRD (refs 22, 23) but has a larger specific surface area, which "could potentially confer a dissolution advantage" (Schenck 2019, ref 14; ref 24) C3.

The weak point of cPADs is wetting. They can wet and dissolve poorly, and at high drug loading (DL) poor wettability can interfere with in vitro release testing (Schenck 2019, ref 14; Frank 2022, ref 25) C4. One remedy embeds cPAD particles in a water-soluble matrix such as hypromellose (HPMC) in an ordered-mixtures fashion, with evaporative isolation; these HPMC-coated, "hierarchical" cPADs managed poor wetting well, but had been tested only at loadings similar to spray-dried dispersions (Schenck 2021, ref 16; refs 28, 29) C5. Whether the same idea can rescue release at high loading remained open C5.

High loading is attractive, because it saves excipient and shrinks the tablet, but high-loading ASDs typically release worse even when they are physically stable (Indulkar 2019, ref 31) C34. For copovidone (PVPVA) dispersions the Taylor group and others described two regimes: at low DL release is "polymer-controlled", drug and polymer leave together at the fast rate of the polymer; above a threshold, the limit of congruency (LoC), release becomes "drug-controlled" and drops sharply ("falling off the cliff"), attributed to a drug-rich barrier at the dissolving surface (refs 30–35; Indulkar 2019, ref 31; Saboo 2019, ref 34) C35. For enteric polymers such as hypromellose acetate succinate (HPMCAS) the picture is less clear: the polymer must ionize before it dissolves (ref 39), and more drug may produce a gradual, "sloped" loss of release rather than a cliff (Hiew 2022, ref 30; Yang 2021, ref 35; Chen 2015, ref 40) C36. Separately, drug enrichment at the particle surface relative to the bulk had been linked to impaired release (Bhardwaj 2018, ref 41; Li 2020, ref 42) C44.

The paper joins these threads. Its stated goal is to probe the surface of hierarchical cPAD particles and its effect on dissolution, "with a hypothesis that it is the DL on the surface, rather than bulk particle DL that determines the extent of release", and to compare a water-soluble matrix added as discrete particles with one applied as a continuous coat C6. The model drug is grazoprevir, an antiviral synthesized by Merck & Co.; the primary polymer is HPMCAS (AQOAT AS-LF); the secondary, water-soluble polymer is HPMC (PHARMACOAT 603) or hydroxypropyl cellulose (HPC, KLUCEL EXF) M1·M2·M3·M4.

Background concepts

Binary and hierarchical cPADs. In this paper a binary cPAD contains only grazoprevir and HPMCAS. A hierarchical cPAD is defined as "a core ASD particle with an additional water soluble component that is coating the particle surfaces" C7. The core is always made at a target of 75% grazoprevir and 25% HPMCAS; adding the secondary polymer at the same mass as the HPMCAS gives a target composition of 60% drug, 20% HPMCAS and 20% secondary polymer P6. The sample names encode the recipe: B-75-HCl/FD is a binary cPAD at 75% DL precipitated in 0.001 N HCl and isolated by filtration and drying; H-HPMC-HCl/RE60 is a hierarchical cPAD with HPMC, precipitated in 0.001 N HCl and isolated by rotary evaporation with a 60 °C bath; (j) marks jet-milled HPMC P6.

Anti-solvent and isolation. The anti-solvent is the liquid in which the drug and HPMCAS are insoluble, so that they precipitate as amorphous particles. Two were used: acidified water (0.001 N HCl), in which HPMC and HPC dissolve, and methyl tert-butyl ether (MTBE), in which grazoprevir, HPMCAS and HPMC are all insoluble P6·M9. After precipitation the solids are isolated either by filtration and drying (FD), which drains most of the liquid and whatever is dissolved in it, or by rotary evaporation (RE), which leaves every dissolved solid on the particles C15. This matters for hierarchical particles, because the secondary polymer is dissolved in the liquid that FD removes.

Cloud point. HPC and HPMC dissolve in cold water but come out of solution on heating. The temperature at which a 1% solution turns turbid is the cloud point; for HPC in 0.001 N HCl it is about 40 °C P9·M4. The authors use it to argue that drying temperature can change how the secondary polymer deposits C25.

Amorphous solubility and the dissolution dose. The amorphous solubility is the highest concentration of dissolved drug before a drug-rich second phase forms; above it the solution scatters light. The paper measures it by two methods, UV extinction (adding a DMSO stock until scattering at 450 nm rises) and ultracentrifugation (adding drug, spinning out the second phase and assaying the supernatant) P2·P3. The dissolution dose is set to the amorphous solubility in fasted state simulated intestinal fluid (FaSSIF), so that 100% release means a solution at the amorphous solubility, 150 µg/mL P12.

Two-stage dissolution. The material is first stirred 30 min in 50 mL simulated gastric fluid (SGF, pH 1.2); then 50 mL double-strength FaSSIF is added and release is followed for 60 min more at pH 6.5 by in situ UV probes P12. Grazoprevir is a weak acid with a pKa of 4.68, so it dissolves poorly in the acid stage M1·C11.

Surface concentration by XPS. X-ray photoelectron spectroscopy (XPS) analyses only the outer few nanometres of a powder. Here the spectra of each sample are fitted with the spectra of the pure components, and the near-surface composition is computed assuming a homogeneous mixture within an information depth of about 10 nm; each value is the mean of five measurements P14. The paper reports the result as "surface grazoprevir concentration" in percent R14. Comparing it with the bulk drug loading shows whether drug is enriched or depleted at the surface.

Polymer-controlled and drug-controlled release. These terms come from copovidone ASDs. In polymer-controlled release the fast-dissolving polymer carries the drug out with it; in drug-controlled release the slowly dissolving drug sets the pace, and release is poor C35. The authors use this vocabulary to interpret their high-loading samples, but in this paper only drug release was measured C53·X10.

Methods in detail

The study has three parts: a reference curve of release against bulk loading from conventional solvent-evaporated ASDs; ten co-precipitated materials in which the presence, type and mode of addition of the secondary polymer, the anti-solvent and the isolation route are varied; and a link between release and surface composition by XPS, followed by a dog study of one binary and one hierarchical material A1·A2·A3·A4·A5·A6·A7.

The reference series: solvent-evaporated ASDs

Grazoprevir and HPMCAS were dissolved in 1:1 dichloromethane:methanol at 20 mg/mL solids and dried on a rotary evaporator with a 50 °C bath, then dried at least 24 h under vacuum, cryomilled and sieved below 250 µm. Eight loadings were made: 5, 10, 15, 20, 30, 50, 60 and 75% w/w P5. These materials are referred to as SE ASDs; SE20 is the 20% DL one.

The ten co-precipitated materials

For every cPAD, grazoprevir and HPMCAS were dissolved in tetrahydrofuran (THF) at 20% w/w solids (target 75% DL) and sheared into ten volumes of anti-solvent with an IKA ULTRA-TURRAX at 6000–12000 rpm for a few seconds P6. The secondary polymer, at 5 mg/mL, was handled in one of three ways: dissolved in the acidified-water anti-solvent before precipitation; suspended in MTBE (as received or after spiral jet milling); or dissolved in an acidified-water wash poured over the filtered wet cake P6·P8. The wash route was used because putting enough HPMC into the anti-solvent for a filtered product to reach 20% would have made it too viscous C16. Table A lists the materials with the drug assays of the isolated solids.

Table A. The co-precipitated materials, from Tables 1 and 4 of the paper; targets for hierarchical materials are 60% drug, 20% HPMCAS and 20% secondary polymer P6·R6.

The assays already carry a result. The binaries came out close to target, which the authors read as no loss of drug or HPMCAS to the mother liquors R6.a–c·C14. Note by the AI writer: the text says the binaries were "within ± 1%" of target, but B-75-HCl/FD is 1.2 points (1.6% relative) below its 75% target X5. Among hierarchical materials, RE kept the dissolved HPMC (58.3% drug), whereas FD drained most of it: H-HPMC-HCl/FD assayed 73.4% drug, "suggesting that only 1.6 wt.% polymer was present" R6.d·R6.e·C15. The washed material reached 56.2% drug, slightly more polymer than targeted R6.j. So the material later called the best performer and dosed to dogs is, by composition, a ≈73% DL particle with a small amount of HPMC, not a 60% DL particle X1.

Solubility, dissolution and characterization

Crystalline solubility was measured by stirring excess drug 48 h at 37 °C, then filtering (0.2 µm) and ultracentrifuging (35,000 rpm, 30 min) before HPLC P1·P4. Amorphous solubility was measured by UV extinction (stock of 20 mg/mL grazoprevir in DMSO pumped at 25 µL/min, scattering at 450 nm) and by ultracentrifugation (150 µL of the same stock added to 50 mL medium), each in triplicate; 200 µg/mL copovidone was added to SGF to stop crystallization P2·P3·M5. Dissolution followed the two-stage protocol above with paddles at 100 rpm and 15 mg drug per vessel P12. The paper does not state how many vessels were run per sample, nor whether the powder mass was set from the target or the assayed loading X18·X2. Solid-state characterization comprised PXRD, modulated DSC (onset Tg) and laser diffraction particle size in Isopar G after sonication; SEM at 5000× showed particle morphology; the HPC cloud point was taken as the temperature at which the transmittance of a 1% solution in 0.001 N HCl began to drop when heated at 1 °C/min P11·P10·P7·P13·P9.

XPS

Powders were mounted on Cu tape and analysed with monochromatic Al Kα radiation; after charge correction to the C–C component of C 1s at 284.8 eV, the spectra were curve-fitted with the spectra of the reference compounds and converted to near-surface compositions with a Tougaard background, Scofield sensitivity factors and inelastic mean free paths, assuming a homogeneous mixture within the ≈10 nm information depth P14. Seven cPADs and SE20 were measured, five times each R14·P14. The two MTBE-made hierarchical materials and B-60-HCl/FD do not appear in the XPS figure R14·X6.

Dog study

Six male Beagle dogs, fasted overnight and pretreated with pentagastrin (0.006 mg/kg, about 30 min before dosing), received 100 mg capsules of B-75-HCl/FD (assay 73.8%) and H-HPMC-HCl/FD (assay 73.4%) in a crossover with a seven-day wash-out P15. Plasma was sampled from pre-dose to 24 h and analysed by LC-MS/MS (linear range 4–5000 ng/mL); AUC0–24h, Cmax and Tmax came from noncompartmental analysis, and concentrations at each time point were compared by one-tail paired t-tests P16·P17. The paper does not say whether "100 mg" is the drug dose or the powder mass X17.

Experimental analysis

Solubility sets the scale

Crystalline grazoprevir is practically insoluble in SGF and pH 6.5 buffer (below the 0.07 µg/mL detection limit) and dissolves to 1.9 ± 0.2 µg/mL in FaSSIF R1. The amorphous form reaches far higher concentrations (Table B). The authors note "considerable solubilization" in FaSSIF relative to buffer, and trace the low gastric value to the weak-acid character of the drug (pKa 4.68, "hence") R2·C11.

Table B. Amorphous solubility of grazoprevir at 37 °C, mean (SD), n = 3, from Table 2 R2.

The authors call the two methods "in good agreement" and set the dissolution dose at 150 µg/mL, the UV value in FaSSIF C12·P12. Note by the AI writer: the ultracentrifugation method as printed adds 3 mg grazoprevir to 50 mL, a nominal 60 µg/mL, which is below both FaSSIF values, so the 186.3 µg/mL ultracentrifugation value cannot come from the procedure as written; the added volume is probably misprinted or differed for FaSSIF X3. The dose itself rests on the UV value and is not affected.

The reference curve: release falls with bulk loading

In the acid stage every solvent-evaporated ASD released almost nothing, about 0–1%, which the authors find consistent with the low solubility of the protonated drug R7·C17. After FaSSIF was added release began immediately and was ordered by loading (Figure 2a) R7. At 90 min (Figure 2b, Table C) the 5% DL ASD released completely, 10 and 15% DL about 84 and 83%, 20% about 63%, 30% about 43%, and 50, 60 and 75% DL all about 27–30% R8. The authors describe a steep decline "until an apparent plateau at 50% DL", and note that this sloped shape, rather than a single cliff, fits what is expected for an enteric polymer C18·C36.

Table C. Release of solvent-evaporated grazoprevir–HPMCAS ASDs at 90 min, read from Figure 2b (approximate) R8.

Binary cPADs: co-precipitation alone does not fix high loading

All cPADs were amorphous by PXRD; the 60 and 75% DL binaries had single onset Tgs of 130.3 and 135.0 °C (grazoprevir 149.4 °C, HPMCAS 115.1 °C) R4·R3·M1·M2. In dissolution (Figure 3) B-60-HCl/FD released about 37.5% and B-75-HCl/FD about 17% at 90 min, both below half the dose R10.a·R10.b·C19. The powders aggregated and floated in the medium, which "could have resulted" in the poor release R9·C20. Against the reference curve (Figure 9), the 60% DL binary did "slightly better" than the 60% DL SE ASD (about 37.5 against 29%) R17·C38. The authors do not comment on 75% DL, where the binary from acidified water released less than the SE ASD (about 17 against 30%) R8.h·R10.b·X20. The 75% DL binary made in MTBE released only about 8% R14.h.

Hierarchical cPADs: how the polymer is added decides the outcome

With HPMC dissolved in the acidified-water anti-solvent, the particles wetted visibly better, especially in FaSSIF, and released about 71% (H-HPMC-HCl/FD) and 63% (H-HPMC-HCl/RE60) at 90 min, against about 17% for the matching binary R9·R10.b–d·C21. Applying the same polymer as a wash of the filtered wet cake gave about 47% R10.e. HPMC slurried in MTBE, intended to deposit as discrete particles, gave about 8%, and reducing the HPMC particle size by jet milling raised this only to about 15%; the authors conclude that particle size reduction "was insufficient to eliminate the gap" between a discrete particle coat and the continuous coat from acidified water R11·C23·C24. With HPC dissolved in acidified water, the material dried with a 60 °C bath released about 72% and the one dried with a 90 °C bath about 56% (about 30 against 18% at the first FaSSIF point) R12·C26. The authors had "postulated" this: HPC is insoluble above its ≈40 °C cloud point and "should" then form a discrete coat, while with a 60 °C bath evaporative cooling "would" keep the product below the cloud point C25·M4.

Under SEM (Figure 6) the acidified-water materials appear as heterogeneous, intertwined rods and flakes and the MTBE materials as smoother surfaces; the hierarchical MTBE particles were rougher than the binary MTBE particles but still relatively homogeneous; the washed particles looked like the pre-dissolved ones; and the HPC materials dried at 60 and 90 °C looked alike R13·C27. In the Discussion the authors describe the acidified-water materials as "highly porous" C51. The hierarchical particle chosen for the dog study was also smaller than its binary counterpart (10th/50th/90th percentiles 9.6/28.8/77.6 against 12.5/36.9/85.1 µm) R5·C13.

The link to surface composition

Figure 7 is the centre of the paper. It plots release at 90 min against the XPS surface grazoprevir concentration (Figure 7a) and against bulk loading (Figure 7b). Table D lists the values.

Table D. Surface and bulk grazoprevir against release at 90 min; surface and release read from Figure 7a (approximate; surface values are means of five XPS measurements), bulk values from Table 4 where assayed R14·R6.

Against bulk loading the samples show no order: at 56–75% bulk drug, release spans about 8–72% R14. Against surface concentration they fall into two groups: the hierarchical materials and SE20, with about 10–34% surface drug and about 47–72% release, and the two 75% DL binaries, with about 53 and 84% surface drug and about 17 and 8% release R14. The authors read this as "in general" more release at lower surface concentration C8. The MTBE binary's surface value of about 84% "could potentially account" for its low release, and the lower value for the acidified-water binary "suggests" that the anti-solvent set the surface composition C28·C29. Filtered particles carried more surface drug than rotary-evaporated ones (about 25 against 10% for the HPMC pair), which "could be due" to loss of dissolved polymer in the filtrate; washed particles carried more still (about 34%) and released less, which "implied" that the polymer adheres and forms a continuous coat best when it is already dissolved in the anti-solvent at precipitation R14·C49·C50.

Two comparisons carry the authors' argument. First, SE20 and the 60% DL H-HPMC-HCl/RE60 have a similar surface concentration (about 12 and 10%) and the same release (about 63%), which "shows that by using the hierarchical approach to modify particle surface, higher DLs can be used" R14.a·R14.b·C31. Second, the authors map surface concentration onto the bulk-loading curve of Figure 2: because a solvent-evaporated particle has about the surface composition expected from its bulk, the two 75% DL binaries, at about 53 and 84% surface drug, sit on the plateau where release is poor, so for surface concentrations of 50% or more release "appeared to be dominated by the drug dissolution rather than the polymer", and dilution below 50% is "anticipated" to improve release C46·C52·C53. They summarize that surface depletion of drug by a water-soluble coat "drove improved release performance" C45.

Note by the AI writer (from the card's extractor notes): within the low-surface group release does not follow surface concentration; H-HPMC-HCl/FD has more surface drug than H-HPMC-HCl/RE60 (about 25 against 10%) but releases more (about 71 against 63%), and H-HPC-HCl/RE90 has slightly less surface drug than H-HPC-HCl/RE60 but releases less (about 56 against 72%) X7·X8. SE20, the only solvent-evaporated sample measured, reads about 12% surface drug against 20% in the bulk, so the reference rests on one point X9. No sample lies between about 34 and 53% surface drug, so the 50% threshold is placed only within that gap X10. The Results also say B-75-HCl/FD released "2 times" more than B-75-MTBE/RE60, while the Discussion says it "did not perform significantly better"; no statistical test is given for either X11.

Exposure in dogs

Table E. Pharmacokinetics after 100 mg capsules in six dogs (crossover); AUC and Cmax mean (SD), Tmax median (range), n = 6, from Table 5 R15.

The hierarchical material gave a 1.8-fold AUC and 1.5-fold Cmax, and median Tmax moved from 3 to 2 h (Table E) R15. Mean plasma concentrations were significantly higher with the hierarchical material at 0.5 and 1 h (p < 0.05) and at 2 h (p < 0.01), and the 8 h value, which the authors treat as Ctrough, was also higher (Figure 8) R16·C32. The authors attribute the low exposure of the binary cPAD to "slow and limited dissolution" ("could be attributed") and see "a clear advantage" of modifying the particle surface C42·C43. Note by the AI writer: AUC and Cmax were not tested statistically, and the Cmax ranges overlap X19. Integrating the mean curves of Figure 8 (axis labelled nM) gives about 104 and 55 µM·h, both about 1.30 times lower than Table 5; that factor equals the conversion from ng/mL to nM for a molar mass of about 767, so the axis of Figure 8 may actually be in ng/mL. Table 5 should be used for exposure X4.

Discussion

The authors' causal chain

The paper's argument can be drawn as one chain that splits by how the secondary polymer meets the particles. Each step is the authors' reading; the hedge words in quotes are theirs.

The authors' summary of the chain is that "dilution of surface drug concentration shifts the release profile of a high DL ASD towards that observed for a lower DL ASD", and that the surface chemistry can be modified by pre-dissolving a secondary polymer in the anti-solvent C57. In the Conclusions they add that "a more uniform polymer coat needed to be formed around the primary particles" and that XPS showed the hierarchical approach "significantly reduced" surface drug C55.

What is measured and what is inferred

Three steps of the chain are measured: release (Figures 2–5), surface grazoprevir by XPS (Figure 7) and exposure in dogs (Table 5, Figure 8) R8·R10·R11·R12·R14·R15. Three are inferred. The coat itself, continuous or discrete, is never imaged or analysed: XPS is reported only as grazoprevir, without the surface fractions of HPMCAS, HPMC or HPC, and SEM at 5000× shows particle morphology, not a polymer layer X15·R13. Wetting is judged by eye X12. And the drug-controlled reading above 50% surface drug comes from the vocabulary of copovidone ASDs; polymer release was not measured X10.

Several comparisons also change more than one thing at a time. The discrete-coat materials were made in MTBE, and MTBE alone, without any HPMC, gave the most drug-rich surface (about 84%) and the lowest release (about 8%) of all; Figure 4 compares them with the filtered acidified-water material rather than with its rotary-evaporated twin X14·R14.h. The hierarchical particle dosed to dogs was smaller than the binary one X13·R5. And the best-releasing material, H-HPMC-HCl/FD, contains only about 1.6 wt% HPMC by assay, yet released more than H-HPMC-HCl/RE60, which kept all of its HPMC and had a lower surface drug concentration C15·R14·X7. The drying-temperature result for HPC also does not fit the surface picture: the 90 °C material released less although its surface drug was not higher X8.

What it suggests for formulation

The authors state that hierarchical co-precipitation "holds great potential as a technique to push DL" and "holds great promise" for high-loading amorphous products with good wetting, dissolution and in vivo performance; that the secondary polymer should be dissolved in the anti-solvent used to precipitate the particles; and that ASD characterization should look beyond bulk analyses to the surface C41·C58·C50·C48. They also stress that manufacturing method and process parameters, not only loading, have "substantial impact" on release C37.

Note by the AI writer (not discussed in the paper): for a high-loading co-precipitated product, these data argue for checking surface composition (here by XPS) during process development, because bulk assay alone did not order release. The isolation step deserves the same attention as the precipitation step: filtration drains a dissolved polymer, rotary evaporation keeps it. That the filtered H-HPMC-HCl/FD, with about 1.6 wt% HPMC, released best and was the material taken into dogs suggests that how the polymer is placed matters more than how much is added; this would keep the overall loading high, but the paper did not test it directly. Slurrying the polymer as particles in a non-solvent did not reproduce the benefit, even after jet milling, so a simple dry or slurry blend should not be expected to substitute.

Limitations

Limits the authors state. The paper has no limitations section. Its hedges mark the softer links: poor wetting "could have resulted" in poor release; the surface concentration "could potentially account" for low release; the anti-solvent effect "suggests"; filtration losses "could be due"; adherence of the coat is "implied"; release above 50% surface drug "appeared to be dominated" by the drug; the drying-temperature mechanism was "postulated" C20·C28·C29·C49·C50·C53·C25.

The AI writer's own judgement. The release differences between binary and hierarchical materials are large, measured directly, and supported by a crossover study in the same dogs R10·R14·R15. The overall strength is moderate, for these reasons drawn from the card's extractor notes:

• The surface–release relation is two groups, not a graded correlation, with no sample between about 34 and 53% surface drug and a single solvent-evaporated reference point X7·X9·X10.
• Coat type and wetting are inferred, the discrete-coat samples were not measured by XPS, and their comparison is confounded by the anti-solvent X15·X12·X6·X14.
• H-HPMC-HCl/FD is described as a 60% DL hierarchical cPAD in the Discussion, but its assay is 73.4% with about 1.6 wt% HPMC; the axes of Figures 7b and 9 mix target and assayed loadings X1·X16.
• Dissolution replicate numbers and whether the dissolution dose was weighed from target or assayed loading are not stated; for H-HPMC-HCl/FD the two differ by a factor of 1.22 X18·X2.
• In dogs, AUC and Cmax were not tested statistically, particle size differs between the two materials, and the concentration axis of Figure 8 may be ng/mL rather than nM X19·X13·X4.
• The ultracentrifugation amorphous solubility in FaSSIF (186.3 µg/mL) exceeds the nominal amount the method adds (60 µg/mL) X3.
• One drug, one primary polymer and a 90-minute test; physical stability of the hierarchical particles on storage was not studied.

Conclusion

The paper shows, for grazoprevir and HPMCAS, that high-loading co-precipitated particles release poorly because, in the authors' reading, their surface is rich in drug, and that precipitating them into an anti-solvent that already holds dissolved HPMC or HPC gives hierarchical particles with a polymer-rich surface (about 10–25% surface drug by XPS), release at 90 min of about 56–72% instead of 8–17%, and a 1.8-fold AUC in dogs R14·R10·R15·C45. The way the polymer is added decides the outcome: dissolved in the anti-solvent works, suspended as particles or added as a wash does not, and HPC dried above its cloud point works less well R10·R11·R12·C50. The surface–release link is clear as two groups but not as a graded relation, and the coat itself was not observed, so the mechanism remains the authors' well-supported interpretation rather than a measured chain X7·X15.