Which side chains make a cellulose polymer hold a drug in supersaturation: telaprevir, 18 new cellulose esters and their simulated shapes

Background and motivation

Many new drug candidates dissolve poorly in water, and a drug must be dissolved to be absorbed. Amorphous solid dispersions (ASDs), in which the drug is molecularly mixed into a polymer, raise the concentration the drug reaches in the gut, but the amorphous drug is thermodynamically unstable and tends to crystallize. The polymer is there in part to delay that crystallization, both in the solid and in the supersaturated solution that forms when the ASD dissolves (Introduction, refs 1–6). The authors start from a practical complaint: which polymer to use is still decided largely empirically, and most marketed ASDs rely on the same short list of commercial polymers (Vandecruys 2007, ref 7).

Earlier work had begun to change that. Ting et al. and Yin and Hillmyer made analogues of hydroxypropyl methyl cellulose (HPMC) and HPMC acetate succinate (HPMCAS) and showed that succinoyl, alcohol and thioether substituents mattered (refs 9, 11, 13). The Taylor and Edgar groups made cellulose ω-carboxyalkanoates, cellulose esters whose side chains end in a carboxylic acid, and measured how long they delayed nucleation of three drugs (Ilevbare 2013, ref 8). Some of these new polymers matched or beat the commercial ones, and the authors explained this by a good match between the solubility parameters (SP) of drug and polymer. The same polymers slowed crystal growth by adsorbing on the crystal surface, which atomic force microscopy confirmed, and growth inhibition was linked to the conformation of the adsorbed polymer (Ilevbare 2013, ref 22; Schram 2015, refs 23, 24). A carboxylic acid group was thought to make an inhibitor amphiphilic: once ionized, it faces the water while the hydrophobic parts of the polymer meet the drug (ref 8).

The weakness the authors point to is that those polymers were chemically similar: cellulose esters whose side chains differed in length but all ended in a carboxylic acid. Within such a family, SP and a few other properties (Tg, the presence of carboxylic or bulky groups) correlate with performance; across chemically different polymers no single property does, so testing a new polymer remains trial and error (Introduction, refs 8, 19, 25). New synthetic chemistry from the Edgar group, namely olefin cross-metathesis, hydroboration–oxidation and cross-metathesis followed by thiol–Michael addition (Meng 2014–2016, refs 14–17, 20), now allows many different functional groups to be attached to the same cellulose backbone under mild conditions. Simulation was the other new tool: earlier molecular dynamics (MD) work had related nucleation inhibition to a balance between an additive's affinity for itself and for the solute (Anwar 2009, ref 26), but few simulations addressed ASD polymers.

The paper therefore sets out to make cellulose polymers with chemically diverse substituents (carboxylic acids, esters, ethers, alcohols, amides, amines and sulfides), to test how long each keeps telaprevir from crystallizing out of a supersaturated solution, and to use quantum chemical calculations on monomers and MD of short oligomers to explain why some work and others do not. Telaprevir, a hepatitis C protease inhibitor marketed as an ASD (ref 33), is the model drug. The authors present the combination of experiment and simulation as a step toward designing crystallization inhibitors rationally rather than by screening.

Background concepts

Nucleation induction time. If a drug is added to buffer above its crystalline solubility, the solution is supersaturated and will eventually crystallize. The time until crystals are detected in an unseeded solution is the induction time. The paper writes it as the sum of two parts (Eq. 1):

Here $T_{\mathrm{i}\mathrm{n}\mathrm{d}}$ is the measured induction time, $T_n$ the time to form a critical nucleus and $T_{\mathrm{g}\mathrm{r}\mathrm{o}\mathrm{w}\mathrm{t}\mathrm{h}}$ the time for that nucleus to grow to a detectable size. In plain terms: a longer induction time can come from slower nucleation, slower early growth or both, and the measurement does not separate them. The paper treats a longer induction time as a more effective polymer, without defining a numerical threshold for "effective"; in practice the comparison is with telaprevir alone, which crystallized after about 14–15 min at 150 µg/mL (Figure 2).

Glass–liquid phase separation (GLPS). When the concentration of telaprevir in solution exceeds about 90–100 µg/mL, the drug separates into a second, drug-rich phase of spherical aggregates 200–400 nm in size (Mosquera-Giraldo 2015, ref 36). Because telaprevir has a high glass transition temperature, this phase is glassy rather than liquid, hence the name glass–liquid phase separation, a form of the liquid–liquid phase separation seen with other drugs (ref 37). The concentration at which it starts is the amorphous solubility. The SI schematic (Figure S1) places the crystalline solubility of telaprevir at about 4.7 µg/mL, without a source. The authors argue that drug-rich particles are "likely" to change nucleation by offering interfaces for heterogeneous nucleation, so they tested every polymer above the GLPS concentration (150 µg/mL) and several below it (80 µg/mL).

Degree of substitution and polymer names. Each anhydroglucose ring of cellulose carries three hydroxyl groups (positions 2, 3 and 6) that can be esterified. The degree of substitution (DS) is the average number of a given substituent per ring, so the DS values of all substituents plus the remaining hydroxyls add up to 3. The paper names its polymers XX-Xxx-000: the starting cellulose ester (CA cellulose acetate with acetate DS 1.82, CAP cellulose acetate propionate, CAB cellulose acetate butyrate, CP cellulose propionate), then the side chain (Adp adipate, Sub suberate, Seb sebacate, or the codes A1a to A7 drawn in Figure 1), then the side-chain DS times 100. CA-A1a-056 is therefore cellulose acetate carrying the A1a side chain at DS 0.56.

Solubility parameter. The SP is a measure of cohesive energy density, often used as a stand-in for hydrophobicity: a lower SP means a more hydrophobic material. The paper calculates SP for each new polymer with the Fedors group-contribution method, following Babcock et al. and the procedure in the SI of Dong et al. (refs 19, 25, 35); the calculation uses the energy of vaporization and molar volume of each group weighted by its DS and does not account for ionization (Table S1, footnote a). The equation itself is not printed.

Quantum chemical descriptors. Density functional theory (DFT) gives the electronic structure of a molecule. Here it is used with the PBE0 functional and the 6-31+G* basis set, and the solvent is represented by a polarizable continuum model (PCM) with the dielectric constant of water (78.39), that is, water as a featureless dielectric rather than explicit molecules. Two outputs matter. The molecular electrostatic potential (MEP), mapped on a surface of constant electron density, shows where a molecule carries partial negative or positive charge and so where electrostatic contacts and hydrogen bonds are likely. The free energy of solvation (ΔGsolv) is the free energy of moving the molecule from vacuum into the implicit water; a more negative value means the molecule is better solvated.

Radius of gyration. For the MD oligomers, compactness is measured by the radius of gyration (Eq. 2):

Here $N$ is the number of atoms, $m_i$ the mass of atom $i$, $𝐫$ its position (printed without a subscript) and ${𝐫}_{\mathrm{c}\mathrm{o}\mathrm{m}}$ the position of the centre of mass. In plain terms: Rg is the mass-weighted root-mean-square distance of the atoms from the centre of the molecule; a chain that folds onto itself has a smaller Rg than one that stays stretched out.

Solvent accessible surface area. If every atom is drawn as a sphere of its van der Waals radius and a probe sphere of radius 1.4 Å (a water molecule) is rolled over the surface, the surface traced by the probe centre is the solvent accessible surface (SAS). The paper divides its area by the number of atoms (SASA per atom), so molecules of different size can be compared. A chain whose parts pack against each other hides surface from the solvent and has a smaller SASA per atom.

Radial distribution function. To see how parts of an oligomer sit relative to each other, the paper computes radial distribution functions (RDFs) between groups of atoms (Eq. 3):

Here $N_{\alpha }$ and $N_{\beta }$ are the numbers of atoms of the two groups, ${𝐫}_{ij}$ the vector between atoms $i$ and $j$, $\delta$ the Dirac delta function and the angle brackets a time average over the trajectory. In plain terms: the function counts how often an atom of one group is found at distance $r$ from an atom of the other group, so a sharp peak means a well-defined separation and a broad, low distribution means the distance fluctuates widely. The printed form omits the usual normalization by shell volume and density, and the RDF plots carry no y-axis scale, so only the peak positions and shapes are compared below. Four kinds are used (Figure 11): between the terminal atoms of different side chains (terminus-to-terminus), between the ring carbons C5 to which the side chains are attached (C5-to-C5), between each terminal atom and its own C5 (single side chain average) and between all three terminal atoms and all three C5 atoms (total).

Methods in detail

The paper has four parts: a library of cellulose polymers in which one chemical feature changes at a time, an induction-time screen below and above the GLPS concentration, DFT calculations on single repeat units, and MD of short oligomers in water. The simulations were run after the screen, to explain its results.

Contribution 1: a polymer library in which one feature changes at a time

① Problem. Commercial cellulose derivatives such as HPMCAS carry several substituents at once, so their performance cannot be traced to one chemical group. To learn which group matters, the authors needed polymers that share a backbone and differ only in the end of one side chain, or only in its length.

② How. Eight commercial polymers were bought: poly(vinylpyrrolidone) PVP K29/32, polyacrylic acid (PAA), the vinylpyrrolidone–vinyl acetate copolymer PVPVA K28, cellulose acetate phthalate (CAPhth), HPMC 606 grade, HPMCAS-MF, Eudragit L100 (EUD L100) and pectin from citrus peel (Table 1 and Materials). Eighteen new polymers were made in four ways (Materials and Methods; Table S1):

• Polymers 1–6, the cellulose ω-carboxyalkanoates: a monobenzyl ester of adipic, suberic or sebacic acid chloride is attached to a commercial cellulose ester and the benzyl group is then removed by hydrogenolysis over Pd(OH)₂/C. This group also contains CP-Tod-202, whose TOD side chain is an ether ester without an acid (Figure 1).
• Reaction pathway 1 (polymers 7 and 11–14): commercial cellulose esters are first given a terminal olefin (pent-4-enoyl or undec-10-enoyl chains), which is cross-metathesized with acrylic acid, 2-hydroxyethyl acrylate, PEG methyl ether acrylate or acrylamide and then hydrogenated. This gives CA-Adp-056 (carboxylic acid end), CA-A2a-079 and CA-A2b-067 (amide ends on a short and a long spacer), CA-A3-056 (PEG methyl ether) and CA-A4-056 (2-hydroxyethyl ester).
• Reaction pathway 2 (polymers 15–18): cross-metathesis with 2-hydroxyethyl acrylate or [2-(acryloyloxy)ethyl]trimethylammonium chloride, followed by thiol–Michael addition of 3-mercaptopropionic acid or 2-mercaptoethanol across the double bond. The side chain then carries two functions: A5a and A5b a thioether-linked COOH branch plus a hydroxyethyl ester (short and long spacer), A6 a COOH branch plus a trimethylammonium ester, A7 an OH branch plus a trimethylammonium ester.
• Reaction pathway 3 (polymers 8–10): hydroboration–oxidation converts the terminal olefin quantitatively into an alcohol (A1a on a short spacer; A1b on a long spacer, on CAP and on CAB).

In Figure 1 the spacer is drawn as a bracketed methylene unit with subscript 2 (short, from the pentenoate) or 8 (long, from the undecenoate); the text calls these the "two carbon" and "eight carbon" chains. Tg was measured by DSC (heat–cool–heat at 20 °C/min, or modulated DSC when the transition was unclear) and molecular weight by SEC in DMAc/LiCl with light scattering, viscometry and refractive index detection (SI Section 3).

③ What the library achieves, and where it falls short. The pathway-1 polymers share the cellulose acetate backbone (acetate DS 1.82) and have side-chain DS between 0.56 and 0.79, so their end group is the main difference. The authors state that DS and molecular weight were "maintained as constant as possible" and that the modular metathesis chemistry holds degree of polymerization, DS and the position of other substituents constant. Table S1 shows the limits of this: Mw within pathway 1 ranges from 37 to 180 kDa, and Mw and dispersity are not reported for any of the pathway-2 polymers. Two synthetic limits also shaped the design. Cross-metathesis of the short-chain olefin with the trimethylammonium acrylate failed (CA-Pen079-TMA could not be made), which the authors attribute "probably" to repulsion between cations held close together on short chains; so A6 and A7 exist only with the long spacer. And the thiol–Michael step for A5b reached 87% conversion, leaving the actual A5b DS at 0.58 with 0.09 of the unreacted α,β-unsaturated A4b chain (Table S1, footnote c). Seven of the new polymers (marked b in Table S1) could not be held above 5 µg/mL in buffer: they precipitated on contact with water, so they were tested at 5 µg/mL only.

④ Difference from earlier work. Earlier cellulose inhibitors from these groups varied mainly the length of an acid-terminated chain (ref 10). This library varies the end group itself (acid, amide, ether, alcohol, quaternary ammonium) on one backbone, adds branched side chains with two functions, and pairs short and long spacers with otherwise identical chemistry.

Contribution 2: an induction-time screen below and above GLPS

① Problem. Telaprevir in an ASD can dissolve to a solution that contains drug-rich particles as well as dissolved drug (SI Section 2). A polymer might delay crystallization in a clear solution but not in one with particles, which could act as nucleation sites. The test therefore had to cover both regimes.

② How. The new polymers were first dissolved in a little THF or DMSO and diluted into 100 mM sodium phosphate buffer, pH 6.8; the commercial ones were dissolved directly in buffer. Telaprevir was then added from a 7 mg/mL methanol stock to 47 mL of the polymer solution in a jacketed flask at 37 °C, stirred at 300 rpm, to give 80 µg/mL (below GLPS) or 150 µg/mL (above GLPS). A UV fibre-optic probe read the absorbance at 270 nm (telaprevir maximum) and at 370 nm (no absorption, so only scattering) every minute, and the difference was converted to concentration with a standard curve (R² > 0.999). At 80 µg/mL, the induction time is where the 370 nm scattering rises and the 270 nm signal falls together. At 150 µg/mL the particles already scatter, so only the 270 nm signal was used. Each condition was run three times (Figures 2–4).

③ The two drops above GLPS. Without polymer at 150 µg/mL the apparent concentration falls twice: from about 162 µg/mL at the first reading to about 100 µg/mL within about 12 min, and then steeply between about 15 and 25 min, to about 9 µg/mL at 100 min (Figure S2b). At 80 µg/mL there is a single fall, from about 85 µg/mL after about 15 min to about 12 µg/mL at 100 min, matched by a rise in scattering (Figure S2a). The authors assign the first drop above GLPS to a change in the optical properties of the drug-rich aggregates as they agglomerate or grow, as reported for danazol (SI ref 5), and the second to crystallization. They tested this with a side-by-side diffusion cell (regenerated cellulose membrane, 6–8 kDa cut-off): with 5 and 50 µg/mL HPMC, telaprevir appeared in the receiver at nearly the same rate, about 1.74 and 1.65 µg/mL after 60 min, although only the 5 µg/mL solution showed the first drop (Figure S3). Since a loss of supersaturation by crystallization would lower the flux, the first drop is taken not to be crystallization, and the induction time is read at the second drop.

④ Other read-outs. Crystals from the 5 µg/mL runs were collected on coverslips, dried, sputtered with platinum and imaged by field-emission SEM (Figure 5). A separate test repeated two acid polymers in 100 mM sodium citrate buffer at pH 3.0, where the carboxylic acids are protonated (Figure 3), and a concentration series (5, 25 and 50 µg/mL) compared HPMCAS, CA-Sub-090 and PVP (Figure 4).

Contribution 3: DFT on single repeat units

① Problem. After the screen, three questions were open: why A5a and A5b, which differ mainly in spacer length, behave so differently; why the quaternary ammonium polymers, charged at any pH, fail; and how the functional groups change the shape of the polymer and its ability to meet the drug (Section "Computational Simulations").

② How. An average repeat unit of each polymer (a "monomer") was built in IQmol with the side chain on position 6 (R6), the most accessible position, and hydrogen on the other positions; the chain ends were capped with methyl groups. Carboxylic acids were modelled as anions, as at pH 6.8, and the adipate monomer was also modelled protonated (Adp-P). A conformational search with the CHARMM force field (random walk over selected torsions, 1000 cycles, acceptance within 6 kcal/mol of the best) gave a starting structure, which was optimized at the PBE0/6-31+G* level in PCM water in Q-Chem 4.3. A single-point calculation then gave the MEP and ΔGsolv. Twelve monomers were modelled (Sub, Adp, Adp-P, A1a, A1b, A2a, A2b, A4, A5a, A5b, A6, A7), plus telaprevir from its crystal structure (Figure 6, Figure S4).

③ What it can and cannot show. A monomer shows whether a side chain prefers to stretch out or fold back onto its own ring, and how well its charged or polar end is solvated. It cannot show how neighbouring side chains interact along a chain, which is why the authors add MD of oligomers. The monomer carries none of the acetate, propionate or butyrate esters that make up most of the substitution in the real polymers (DS 1.82–2.09).

Contribution 4: MD of oligomers in water

① Problem. Interactions among side chains on the same chain, and the overall shape of the polymer in water, need a model larger than one ring. Quantum calculations on such systems are too expensive, so classical MD was used.

② How. Each oligomer had seven rings, three of which carried the same side chain on R6 (stated DS around 0.42; Figure 8 shows the linear and branched layouts). After a conjugate-gradient minimization, SwissParam topologies were generated and the oligomer was solvated with about 21,000 SPC/E water molecules in a cubic box of edge up to 8.6 nm, with three Na⁺ ions for oligomers carrying carboxylates. GROMACS 5.0 with the CHARMM force field ran four equilibration stages (NVE 50 ps; NVT 5 ns at 310 K; two NPT stages of 5 ns at 1 bar) and a 40 ns NPT production run at 310 K and 1 bar, with particle-mesh Ewald electrostatics and 1 nm cut-offs. Seven oligomers were simulated: Adp, Adp-P, A1a, A2a, A5a, A5b and A6. From the trajectories the authors computed Rg (Eq. 2), SASA per atom and the four kinds of RDF described above (Figures 10–13).

③ What is not covered. Neither the drug nor a crystal surface is in any simulation, so drug–polymer contacts are not computed. A7, the alcohol-plus-cation polymer, has a DFT monomer but no MD oligomer. The oligomers again carry no acetate esters. The paper does not say how many independent trajectories were run per oligomer or how the spread in Figure 10 was computed.

Experimental analysis

Setup in brief. All induction times below were measured at 37 °C in 100 mM sodium phosphate buffer at pH 6.8 (except the pH 3.0 citrate arm of Figure 3), with three replicates. The paper prints no tables of induction times, so every value in this section is read from the figures and is approximate. Telaprevir alone crystallized after about 14–15 min at 150 µg/mL (Figures 2 and 3).

Table A. Induction times (min) with the commercial polymers, read from Figure 2a (approximate, n = 3). The polymers are ordered by decreasing SP, pectin highest and EUD L100 lowest (cited from ref 24). Bars labelled ">700" are cut at 700 min.

Main result 1: among commercial polymers only HPMC and HPMCAS hold telaprevir. Below GLPS (80 µg/mL), HPMC and HPMCAS kept telaprevir in solution for more than 700 min, while the others stayed below about 90 min (Table A). Above GLPS (150 µg/mL) the times were shorter for both, which the authors read as the drug-rich particles speeding up crystallization. At 150 µg/mL, raising HPMCAS from 5 to 50 µg/mL raised the induction time from about 175 to about 483 min, whereas HPMC stayed at about 243 and 233 min, so the authors conclude that HPMCAS-MF is the more effective of the two at 50 µg/mL when drug-rich particles are present. Pectin, PAA and EUD L100, which all carry carboxylic acids, did almost nothing; earlier reports had also found PAA and EUD L100 ineffective (refs 8, 52, 53). The shorter time at 150 than at 80 µg/mL is clear for the polymers that work but within a few minutes for PVP, PAA and EUD L100. Two readings conflict within the paper: the 5 µg/mL HPMCAS point is about 175 min in Figure 2a but about 226 min in Figure 4, and the representative HPMC traces in Figure S3a begin to fall at about 210 min with 5 µg/mL but only at about 300 min with 50 µg/mL, although the mean bars in Figure 2a are similar.

Table B. Cellulose ω-carboxyalkanoates and CP-Tod-202: properties from Table S1 and induction times (min, µg/mL telaprevir + µg/mL polymer) read from Figure 2b (approximate, n = 3).

Main result 2: among the ω-carboxyalkanoates, many acid groups help. The two best, CAP-Adp-085 and CA-Sub-090, carry the acid-terminated chain at DS 0.85 and 0.90, have SP 22.46–23.18 MPa^1/2, and held telaprevir for more than 700 min at 80 µg/mL and for about 470–500 min at 150 µg/mL with 50 µg/mL polymer (Table B). CAP-Seb-024 has an SP (23.04) inside that range but carries the acid at DS 0.24 only, and failed (about 9 min). The authors read this as optimal SP plus a high number of acid groups ("suggest"), and conclude that SP alone is not sufficient. The text attributes part of CAP-Seb-024's failure to "other groups such as butyrate and acetate"; Table S1 lists propionate and acetate for this polymer, not butyrate. CP-Tod-202, which has no acid, gave about 61 min, close to CA-Adp-067 and CAP-Adp-085 at the same condition; it was run only at 150 µg/mL with 5 µg/mL polymer and is not discussed.

Table C. The chemically diverse polymers at 150 µg/mL telaprevir and 5 µg/mL polymer: SP from Table S1; induction time read from Figure 2c and ΔGsolv read from Figure 7 (both approximate). Side-chain codes as in Figures 1 and 7; "short" and "long" are the bracket indices 2 and 8 of Figure 1.

Main result 3: on one backbone, only the acid end works. In the pathway-1 set, which shares the cellulose acetate backbone, only CA-Adp-056 delayed nucleation (about 113 min, upper error bar about 145); the amide, PEG ether and hydroxyethyl ester ends stayed at about 7–14 min, the level of telaprevir alone (Table C). The authors write that "it is clear" the only effective polymer of this subset carried a terminal carboxylic acid. The alcohol-ended pathway-3 polymers were also ineffective (about 10–15 min), whether on a short or a long spacer.

Main result 4: the acid branch works on a short spacer only. Adding a thioether-linked COOH branch to the hydroxyethyl ester chain raised the induction time from about 7 min (A4) to about 90 min with the short spacer (A5a), but only to about 12 min with the long one (A5b). The authors conclude that such a branch "can drastically improve" effectiveness ("These results suggest") and that the chain length "has a major impact". The two quaternary ammonium polymers failed as well (about 13 and 18 min), which the authors attribute "either" to the cation or to the long chain; with no short-chain version available, the data cannot separate the two. Their summary is that branches and bulky groups "may improve" effectiveness, but only when key functional groups are present. Note that A5a and A5b differ in more than the spacer: their side-chain DS is 0.79 and 0.67 (0.58 actual plus 0.09 of A4b), their residual hydroxyl DS 0.39 and 0.51, their Tg 63 °C and 6 and 89 °C, and neither has a reported molecular weight (Table S1).

Table D. Induction times (min) at two pH values, 150 µg/mL telaprevir and 5 µg/mL polymer, read from Figure 3 (approximate, n = 3).

Main result 5: the acid polymers work only when ionized. At pH 3.0 both acid polymers lost their whole effect and fell to the level of telaprevir alone (Table D). The authors read this as polymers with a carboxylic acid end being effective only when ionized ("clearly shows"), in line with the earlier hypothesis that ionized carboxylates make a polymer amphiphilic (ref 8). The experiment changes the buffer species along with the pH, telaprevir alone also crystallized sooner at pH 3.0, and the degree of ionization of the polymers was not measured, so protonation is the most likely but not the only difference between the two arms.

Main result 6: the concentration effect depends on the polymer. From 5 to 25 to 50 µg/mL, HPMCAS gave about 226, 325 and 482 min and CA-Sub-090 about 106, 245 and 500 min, while PVP stayed at about 15 min at 5 and 50 µg/mL (Figure 4). HPMCAS was "considerably more effective" than CA-Sub-090 at 5 µg/mL; the two were equally effective at 50 µg/mL. The authors note that such low concentrations "might be present in vivo" after an ASD dissolves and expect the concentration dependence to "likely depend on the polymer evaluated". The dashed curves drawn through the points in Figure 4 are not described; at zero polymer they would give about 205 min (HPMCAS) and about 80 min (CA-Sub-090), far above the 14–15 min of telaprevir alone, so they are best read as visual guides.

Main result 7: SP does not predict effectiveness across chemistries. The effective and ineffective polymers overlap in SP. The clearest case is the A5a/A5b pair, SP 25.54 and 24.29 MPa^1/2, with about 90 and 12 min (Table C). In the results the authors write "We do not find any correlation" between SP and effectiveness in this diverse set; in the discussion this becomes "only a weak correlation", against stronger correlations within the more uniform polymer sets of earlier work (refs 8, 22). They add that SP is not a direct measure of hydrophobicity and that calculated SPs "appear to be insufficient" to describe amphiphilicity. They also report no correlation with Tg or molecular weight. No plot or statistic is shown for any of these correlations, and the "similar" SPs of A5a and A5b differ by 1.25 MPa^1/2, more than the 0.72 span the authors use to describe the two effective ω-carboxyalkanoates.

Main result 8: charge sets solvation, but solvation alone is not enough. In the DFT monomers, charged ends gave ΔGsolv of about −72 to −109 kcal/mol and neutral ends (protonated COOH, OH, NH₂) about −25 to −34 kcal/mol (Table C, Figure 7). Ionized and protonated adipate differ "drastically" (about −86 and −33), which the authors say "supports the idea that sufficient amphiphilicity is only achieved when the polymer is ionized" and explains the pH result; Adp-P sits with the alcohol and amide monomers, whose polymers were equally ineffective. A literature benchmark puts the accuracy of such values at about 5 kcal/mol (ref 54), smaller than the gap between charged and neutral monomers. But A5b, A6 and A7 are as well solvated as the effective Adp, Sub (about −76) and A5a, yet their polymers failed. The authors therefore describe a "weak correlation" between effectiveness and ΔGsolv and conclude that ΔGsolv "in isolation is not predictive"; it "may give a first indication" of amphiphilicity.

Main result 9: the long-spacer analogues fold onto themselves. The DFT monomers of A1a, A2a and A5a favour extended side chains, while those of A1b, A2b, A4a, A5b, A6 and A7 are globular, which the authors call a "predisposition for self-interaction"; in Adp and Sub the carboxylate bends back toward the ring through "potential" intramolecular hydrogen bonds (Figure 6). The MD oligomers are summarized below.

Table E. MD of seven-ring oligomers with three side chains, 40 ns at 310 K: mean lines read from Figure 10 and peak positions read from Figures 12 and 13 (approximate). The Figure 10 caption defines the boxes as standard deviation "centered on the median" and the whiskers as quartiles, which the drawing does not bear out, so only the means are used here.

The A5b oligomer is the most compact by both measures, and A6 has the second-lowest surface area per atom (Table E). The authors take the low Rg and SASA of A5b and A6 as evidence that the long-chain polymers self-aggregate, in agreement with their globular monomers, while A5a stays extended. The C5-to-C5 RDFs "attest" to a stiff, extended backbone in A2a, Adp, Adp-P and A5a and a more flexible one in A1a, A5b and A6; the authors write that these results "strongly correlate" with Rg, "suggesting" that Rg reflects mainly the backbone and SASA the side chains. Between the side-chain ends, there "appears evidence" of hydrogen bonding in A1a, A5a and A5b, close contacts between COO⁻ and N⁺(CH₃)₃ in A6, and broad, repulsive distributions between same-charge ends in Adp, A5b and A6, but not in A5a, "probably" because its short chain limits torsion (Figure 12b). In A1a, A2a, Adp, Adp-P and A5a the single and total RDFs coincide, so each side chain stays away from the rings; in A5b and A6 the end side chains lean toward the centre and the central one bends sideways, placing a carboxylate close to the neighbouring rings although far from its own (Figure 13). One comparison complicates the picture: A1a, a short-chain alcohol, has the same mean Rg as A6 (about 1.02 nm); it differs from A6 only in surface area. And the ionized Adp side chain, globular as a monomer, is extended in the oligomer, which the authors attribute ("can be attributed") to charged neighbours that keep the carboxylates from bending back. Two peak positions in the text differ from Figure 13: the Adp peak (text about 0.8 nm, figure about 0.9 nm) and the A5b total maximum (text about 0.5 nm, figure about 0.7 nm). The contrast with about 1.5 nm for the A5b side chain to its own ring holds either way.

Crystal size and shape. SEM of the crystals left after the 5 µg/mL runs (Figure 5) shows radiating blades about 2–5 µm long without polymer and crossed prisms about 3–4 µm long with PVP, but plates of about 0.2–2 µm with HPMCAS-MF and rounded particles of about 0.1–0.2 µm with CA-Sub-090; HPMC left a layer of plate-like fragments. The authors call these "preliminary observations" and suggest that HPMCAS and CA-Sub-090 "may be" crystal growth inhibitors, to be tested in future work.

Discussion

The authors' explanation. The discussion combines the screen and the simulations into one chain. The flow below follows the authors' own wording; the numbers in brackets are this paper's data.

The authors' conclusion is that "the right balance between key substituent groups and the hydrocarbon side chain length needs to be met" ("this study suggests"), and that "it is essential" for an effective polymer to have both enough hydrophobic regions to drive interactions with the drug and enough hydrophilicity to be solvated. They place this alongside earlier observations: AFM showed HPMCAS as globules and ineffective against crystal growth at low pH but as extended chains and effective at neutral pH (ref 23); MD of chitosan gave a larger Rg for charged chains (ref 55); Anwar et al. proposed a balance of affinities for nucleation inhibitors (ref 26); and Mackenzie et al. found an optimal polymer hydrophobicity for drug encapsulation in nanoparticles (ref 30). "Our results support the supposition," they write, "that the solvated polymer conformation plays a major role" in effectiveness.

Where the explanation applies. The data come from one drug (telaprevir), one temperature (37 °C), mainly one pH (6.8), polymer concentrations of 5–50 µg/mL in solution, and nucleation from solution rather than from the solid ASD. The simulated objects are repeat units and seven-ring oligomers that carry only the functional side chain, not the acetate esters of the real polymers. Within that frame, the explanation accounts for three observations together: the pH effect, the failure of the neutral end groups, and the A5a/A5b contrast. It is less clear for the quaternary ammonium polymers, where the cation and the long spacer are confounded, and it says nothing about why carboxylated pectin, PAA and EUD L100 fail; for those the authors invoke hydrophilicity and the lack of branches and bulky groups ("probably").

Open questions. First, no simulation contains the drug, so the step from polymer shape to drug–polymer contact is inferred, as the authors themselves say ("by inference"). Second, whether the cation or the long chain makes A6 and A7 fail cannot be settled without short-chain trimethylammonium analogues, which could not be made. Third, the growth-inhibition role of HPMCAS and CA-Sub-090 rests on crystal sizes in SEM only. Fourth, the authors' own next step, polymers with ionized groups not hindered by intramolecular interactions, is left for future synthesis. Fifth, the claim that the combined approach is "readily extendable" to other drug delivery problems has not been tested on another drug.

Note by the AI writer (not discussed in the paper): what it suggests for formulation. Within this set, two levers decided whether a cellulose polymer held telaprevir in supersaturation: whether its acid groups are ionized and how long the hydrocarbon spacer carrying them is. A carboxylated polymer gave no inhibition at pH 3.0, so a formulation that relies on such a polymer is protected only once the polymer is ionized, which matters for gastric transit. Adding hydrocarbon length to make a polymer more hydrophobic can backfire if the chain then folds onto itself. And none of the new polymers improved on HPMCAS for telaprevir: at 150 µg/mL drug and 5 µg/mL polymer, the best new polymer reached about 113 min against about 175–226 min for HPMCAS and about 243 min for HPMC.

Limitations

Limitations stated by the authors. The authors acknowledge several limits. The calculated SP does not account for ionization (Table S1), and "calculated SPs appear to be insufficient" to describe amphiphilicity. ΔGsolv carries an uncertainty of about 5 kcal/mol and correlates only weakly with effectiveness. The A6/A7 result is ambiguous because "it was impossible to synthesize short chain derivatives" with the trimethylammonium end. Seven of the new polymers could not be dissolved above 5 µg/mL in buffer. The crystal-growth observations are "preliminary". And they note that "it is difficult to make predictions based on evaluation of the monomer chemical structure, because it is unclear how conformation and chemistry will be changed upon oligomerization and solvation", which is why oligomers were simulated.

The AI writer's own judgement (1): how firm the carboxylate rule is. The rule that only carboxylate-bearing polymers worked rests mainly on Figure 2c and on the 50 µg/mL and 80 µg/mL conditions of Figure 2b. At 150 µg/mL with 5 µg/mL polymer, CP-Tod-202, which carries no acid, gave about 61 min, close to two polymers the authors list among the most effective (CA-Adp-067, about 55 min; CAP-Adp-085, about 70 min); it was not tested further. "Effective" is never defined numerically, and the new polymers called effective all stayed below HPMC and HPMCAS at the same condition. Molecular weight was not held constant (37–180 kDa within pathway 1, not reported for pathway 2), and the A5a/A5b pair also differs in DS, residual hydroxyl DS and Tg. The statements that SP, Tg and molecular weight do not correlate with effectiveness come without a plot or statistic. Every induction time is read from bar charts, and two of them disagree within the paper (HPMCAS at 5 µg/mL: about 175 min in Figure 2a, about 226 min in Figure 4). The bars labelled ">500" and ">700" are lower bounds whose run length is not stated.

The AI writer's own judgement (2): how far the simulations carry the mechanism. The simulations were run after the experiments to explain known results, so the abstract's description of simulation as "a promising predictive tool in the screening of polymers" has not been tested by a prediction made in advance. The models leave out the drug, so the hydrophobic interaction with telaprevir at the heart of the design rule is not computed; they also leave out the acetate, propionate or butyrate esters that make up most of the real substitution (DS 1.82–2.09). The failure of A7 is rationalized "by considering the results of both quantum chemical and classical simulations", but A7 was not among the MD oligomers. A low Rg is not specific to long chains: the short-chain alcohol A1a has the same mean Rg as A6. The printed RDF definition lacks the usual normalization and the RDF plots have no y scale, so only peak positions can be compared, and two of those differ between text and figure. The Figure 10 caption describes boxes and whiskers that the drawing does not match. The MD setup states three Na⁺ counterions for carboxylate oligomers but does not say how the three trimethylammonium groups of A6 were balanced. Several smaller inconsistencies in the source do not change the conclusions: Table S1 and Table S2 number A6 and A7 in opposite order (the equivalent names agree with Figure 1); Figure 1 draws A4b with the short spacer, although footnote c places it in the long-spacer CA-A5b-067; Figure 7 labels a monomer "A4" where Figures 1 and 6 use A4a; and the fourth ring of the branched oligomer in Figure 8b is labelled M1 instead of M4.

Conclusion

The paper contributes three things. First, a library of cellulose polymers in which the end group, a branch or the spacer length changes on a shared backbone, screened for how long each keeps telaprevir supersaturated below and above its amorphous solubility. Second, a clear experimental pattern within that library: the polymers that worked carried carboxylic acids, lost their effect when the acids were protonated at pH 3.0, and failed when the acid sat on a long hydrocarbon spacer; calculated SP did not sort the polymers. Third, a retrospective explanation from DFT and MD: charged ends are much better solvated, and long-spacer analogues fold onto themselves into compact, self-aggregated chains. In the authors' words, "the effective polymers contained carboxylate groups in combination with an optimal hydrocarbon chain length", and "the right balance between key substituent groups and the hydrocarbon side chain length needs to be met" ("this study suggests").

The experimental pattern is the firmer part. The mechanism is a plausible reading built on simplified models without the drug, and the step from polymer shape to drug–polymer interaction is inferred. For formulation scientists, the practical messages are that a crystallization inhibitor must be soluble and charged where it is used, that extra hydrophobic chain length can make a polymer collapse rather than help, and that for telaprevir HPMCAS remained the benchmark this generation of designed polymers did not beat.