When can a polymer stop drug crystals from growing? Ritonavir, cellulose adipates, pH and supersaturation

Background and motivation

Many new drug candidates are so poorly soluble in water that they are formulated as amorphous drug–polymer blends (amorphous solid dispersions). On dissolution these give a supersaturated solution, with drug concentrations above the crystalline solubility. That advantage lasts only until the drug crystallizes. Crystals that form after release but before the drug is taken up by the enterocytes cancel the gain of the formulation, so the polymers in such products are often chosen as crystallization inhibitors, acting on nucleation, on crystal growth, or on both. The authors note that such polymers "may significantly impact" how high and how long supersaturation lasts, and so bioavailability (Section 1; refs 1–5). C1·C2

The starting point is the group's earlier work on felodipine with hydroxypropyl methyl cellulose (HPMC) (Alonzo et al. 2012, ref. 6). There, preventing nucleation kept supersaturation better than inhibiting crystal growth, because HPMC could not inhibit growth at the high supersaturation an amorphous felodipine formulation generates; polymer effectiveness fell as supersaturation rose. R19·C3 The authors add a practical reason why growth inhibition still matters: preventing nucleation "may not always be possible", because an amorphous product can contain seed crystals left from manufacturing, and these would de-supersaturate the solution quickly unless an effective growth inhibitor were present; "small traces of crystalline material can thus potentially" have a large effect. C4

The polymers studied here are new cellulose ω-carboxyalkanoates, in particular cellulose adipates. In an earlier screen (Ilevbare et al. 2012, ref. 7) the group measured ritonavir growth with many additives at a supersaturation ratio of 7.6: moderately hydrophobic polymers worked better than highly hydrophilic or highly hydrophobic ones, cellulose polymers with more ionizable carboxylic acids worked better than neutral or less ionized ones, and CAP Adp 0.85 was the most effective of 19 synthesized and commercial cellulose derivatives, including HPMC and HPMC acetate succinate. R18·C5 Ritonavir was chosen for its very low aqueous solubility (log P 5.98); it crystallizes from supersaturated solution more slowly than felodipine, "probably" because of its higher molecular weight and conformational flexibility. M1·C8

The adipate esters were designed for amorphous solid dispersions. The tetramethylene chain of the adipate group makes the cellulose ester more hydrophobic to raise its affinity for hydrophobic drugs; the terminal carboxyl group ionizes at the neutral pH of the small intestine so that the matrix swells or dissolves and releases the drug, and it "may" also interact specifically with the drug. Because the gastrointestinal pH varies, the authors ask how these polymers work in their ionized and unionized states. C14

The paper states two hypotheses (Section 1). First, the effectiveness of the adsorbed polymer at the solid–liquid interface changes with pH, through the pH-dependent ionization of its pendent carboxylic acid groups. Second, the growth-inhibiting effectiveness of a polymer falls as supersaturation rises, as theory predicts. To test them, the authors measured ritonavir growth from seeds at three supersaturations and two pH values, examined the best polymer by adsorption, zeta potential and the Kubota–Mullin model, and located the supersaturation at which ritonavir's growth mechanism changes. C6·C7

Background concepts

Supersaturation. For dilute solutions the paper expresses the driving force for crystallization as the supersaturation ratio (eq 1) and its logarithm, the supersaturation (eq 2):

In plain terms: $S$ is how many times the solution concentration $c$ exceeds the equilibrium solubility of the crystal $c_{eq}$, and $\sigma$ is its natural logarithm. With the crystalline solubility of ritonavir, 1.3 ± 0.1 µg/mL (cited from ref. 7), initial concentrations of 5, 10 and 20 µg/mL give S 3.8, 7.6 and 15.3, and 20 µg/mL, the amorphous solubility, corresponds to σ 2.7. E1·E2·M1·R3 The printed S values are slightly below 5/1.3, 10/1.3 and 20/1.3 (3.85, 7.69, 15.4), which fits a solubility of about 1.31 µg/mL; the paper itself prints S 15.3 in most places and 15.4 once. E1·X19

How a crystal face grows. The overall growth rate is often written as an empirical power law of supersaturation (eq 3):

In plain terms: the growth rate $R_G$ rises with supersaturation, with a coefficient $k_g$ and an order $g$ that typically lies between 1 (diffusion of solute to the crystal is slowest) and 2 (incorporation into the lattice is slowest). The paper does not fit $k_g$ or $g$; it uses the law only as background. E3·C9 At low supersaturation a crystal face grows layer by layer. New layers start from screw dislocations (spiral growth) or from small two-dimensional (2D) nuclei on the face, and molecules join at the edges of the layer (steps) and at corners in these edges (kinks); the face growth rate then rises with the square of supersaturation. The size of the critical 2D nucleus falls as supersaturation rises. At a critical supersaturation, ${\sigma }_c$, it shrinks to the size of a single growth unit: 2D nucleation becomes unhindered, the face becomes rough, and molecules can attach almost anywhere. This change from layer growth to continuous growth is the kinetic roughening transition. Above it, the face growth rate depends linearly on supersaturation, transport of molecules to the surface becomes rate-limiting, and crystal faces tend to round off (Sections 2.2 and 4.2; refs 13–16, 31). C9·C34

Adsorption of the inhibitor. A polymer added at low concentration, where it changes neither the viscosity nor the solubility, acts by adsorbing on the growing crystal (Section 2.3). The fraction of surface sites it covers is the coverage ${\theta }_{eq}$ (eq 4):

In plain terms: $n_{ad}$ is the number of adsorption sites occupied per unit area and $n_{max}$ the number available; ${\theta }_{eq}$ runs from 0 with no inhibitor to 1 when every site is occupied, and complete coverage is not needed to stop growth (ref. 18). E4·C10 The simplest isotherm, Langmuir's, assumes a homogeneous surface; the authors expect anisotropic molecular crystals not to meet this, and earlier polymer adsorption data deviated from Langmuir behaviour at low concentration, "possibly" because polymer molecules are much larger than water molecules. They therefore use a generalized Langmuir isotherm for energetically heterogeneous surfaces (eq 5) in its Langmuir–Freundlich form, the special case in which the two heterogeneity parameters are equal (eq 6): C11·E5

In plain terms: coverage rises with polymer concentration $c$ along an S-shaped curve set by a Langmuir-type constant $K$ and a heterogeneity exponent $n$, and levels off at full coverage. E6

How adsorbed impurities slow growth. Adsorbed impurities can act on growth in four ways: step pinning, incorporation into the crystal, kink blocking and surfactant action (refs 27, 28). The authors exclude the last three by argument: surfactants usually speed growth by lowering interfacial energy; large polymers are unlikely to be incorporated; and kink blocking needs a short residence time on the surface, which does not fit the essentially irreversible adsorption of polymers (Section 2.4). That leaves step pinning, described by Cabrera and Vermilyea: an advancing step stops where it meets an adsorbed impurity and must curve around it, and it can pass only if the gap between neighbouring impurities is wide enough compared with the critical 2D nucleus. The Kubota–Mullin model turns this into an equation for the bulk growth rate (eq 7): C12

In plain terms: $R$ and $R_0$ are the bulk mass growth rates with and without polymer; growth slows in proportion to coverage, and the effectiveness factor $\alpha$, which is related to the size of the critical 2D nucleus, sets how strongly. In this model the coverage is also the ratio of the distance between active sites, $L$, to the distance between adsorbed impurities, $l$ (${\theta }_{eq}=L/l$). An $\alpha$ near 1 means growth stops only at full coverage, where the adsorbed molecules sit about one critical-nucleus diameter apart ($l=2{\rho }_c$ at ${\theta }_{eq}=1$, with ${\rho }_c$ the critical radius); an $\alpha$ above 1 means they sit closer than that ($l<2{\rho }_c$), so growth stops before the surface is fully covered. E7·E11·E13 The model needs a 2D nucleation barrier. When supersaturation rises to the kinetic roughening transition and that barrier vanishes, the authors state that the Kubota–Mullin model no longer applies. C13

Amorphous solubility. The highest supersaturation a dissolving amorphous form can produce is set by its "solubility". The authors estimate it with the approach of Murdande et al. (refs 40, 41) (eq 8), using the Hoffman equation for the free energy difference between the amorphous and crystalline forms (eq 9):

In plain terms: the amorphous solubility is the crystalline solubility $C_{eq}$ multiplied by the factor by which the amorphous form is higher in free energy, $\Delta G_{a\to c}$ (estimated from the heat of fusion $\Delta H_{fus}$, the melting temperature $T_m$ and the temperature $T$), and reduced by the term $I(a_2)$, which accounts for the water that the amorphous solid takes up and which lowers its activity; $R$ is the gas constant. For ritonavir the inputs were $C_{eq}$ 1.3 µg/mL, $\Delta H_{fus}$ 60.4 kJ/mol and $T_m$ 121 °C from DSC; $I(a_2)$ came from the water sorption isotherm of amorphous ritonavir through the Gibbs–Duhem equation. E8·E9·M1·P4 The result was 20.6 µg/mL; an amorphous film dissolved at pH 6.8 and 37 °C gave 19.8 µg/mL, which it held for about 2 h before de-supersaturating rapidly toward the crystalline solubility. R1·R2 Note by the AI writer (not discussed in the paper): the paper prints the inputs of eq 9 but not $\Delta G_{a\to c}$ or $I(a_2)$; from the printed inputs, $\Delta G_{a\to c}$ is about 10.1 kJ/mol at 37 °C, so the free energy term alone would raise solubility about 50-fold, and the water term must bring this down to the reported ratio of about 16. E9

Methods in detail

The study has four strands: growth rates from seeded de-supersaturation at different supersaturations and pH values; adsorption and zeta potential of the best polymer; the growth rate over a wide range of supersaturation to locate the change of growth mechanism; and the Kubota–Mullin model applied below that point. The flow below shows how they connect.

Materials: one drug and six polymers

Crystalline ritonavir (Attix Corporation) was used as received and sieved below 250 µm to serve as seeds; its BET specific surface area was 1.265 m²/g. M1·P2 HPMC (606 grade, Shin-Etsu), which has no ionizable groups, served as the control polymer. M2 The five cellulose adipates were made by the Edgar group from cellulose acetate, cellulose acetate propionate and cellulose acetate butyrate (Figure 3); their degree of substitution (DS), the number of substituents per anhydroglucose unit (maximum 3), was determined by 1H NMR from the ratio of acyl to backbone proton integrals (Section 3.3, Table 1). M3–M7·P3 The pKa of the adipate carboxyl group is about 4.43 (ref. 45), so it is almost completely ionized at pH 6.8 and less than 25% ionized at pH 3.8 (ref. 46); ritonavir is predominantly unionized at both pH. C22

Table A. The polymers, ordered by carboxyl content (from Table 1; Ac acetate, Pr propionate, Bu butyrate). For CAB Adp 0.81 the printed parts add up to 2.94, not the printed total of 2.84. M2–M7·X6

Measuring crystal growth by seeded de-supersaturation

Problem. Growth rates must be measured at concentrations of a few µg/mL, and with seeds present so that nucleation does not interfere. How. Seeds were stirred in buffer at 37 °C (overhead stirrer, 400 rpm) until equilibrated; supersaturation was then created by adding a small volume of ritonavir dissolved in methanol, to initial concentrations between 1.5 and 20 µg/mL. A UV spectrometer with a fiber-optic probe (5 mm path) recorded spectra from 200 to 450 nm every 5 s, and second-derivative spectra were used to suppress scattering by particles. The slope of concentration against time over the first 2 min was taken as the initial growth rate. Polymers were pre-dissolved at 0.5–7.5 µg/mL (5 µg/mL in the pH comparison), at pH 3.8 (100 mM sodium acetate) or pH 6.8 (100 mM sodium phosphate), and every condition was run in triplicate. Ritonavir's equilibrium solubility is known not to change in 5 µg/mL polymer solutions (ref. 7). P7 The paper reports results in two equivalent forms: the effectiveness ratio $R_{g0}/R_{gp}$ (growth rate without polymer over growth rate with polymer; a value above 1 counts as effective), used in Figures 5 and 6, and its inverse $R/R_0$, used in the Kubota–Mullin analysis (Figure 10, Table 4). E12·E7 The mass of seeds and the solution volume are not printed. X12

Adsorption, zeta potential and crystal shape

Adsorption isotherms. About 4.0 mg of seed crystals were equilibrated overnight with 20 mL of CAP Adp 0.85 solution (2–30 µg/mL) at pH 3.8 or 6.8. The polymer left in the supernatant was measured with the phenol–sulfuric acid assay of Dubois et al. (yellow-orange colour read at 490 nm; calibration 2–50 µg/mL), and the amount adsorbed per unit crystal area followed from the loss from solution (eq 10): P9

In plain terms: the adsorbed amount $C_a$ (mg/m²) is the drop in polymer concentration, from the initial $C_i$ to the equilibrium $C_f$, times the solution volume $V$, divided by the seed mass $m$ times its BET area $A$. The paper's own readings are consistent with this: at the top of the pH 3.8 isotherm, about 40.8 mg/m² at about 19.7 µg/mL equilibrium concentration corresponds to an initial 30 µg/mL. E10 The isotherms were fitted with eq 6 in Origin. R10

Zeta potential. 4.0 mg of crystals were dispersed overnight at 37 °C in 20 mL of 10 mM sodium citrate buffer with 5 µg/mL CAP Adp 0.85 at pH 3.8 or 6.8 and measured with a Malvern Nano-ZS. Note that this buffer differs from the 100 mM acetate and phosphate buffers of the growth tests. P8·X14

Crystal shape. Seeds before and after growth were imaged by field-emission SEM; length and width were measured on 50 crystals per sample in ImageJ, and the aspect ratio was taken from the average length and width (Section 3.2). P1·E15

Locating the critical supersaturation and applying the Kubota–Mullin model

The growth rate of ritonavir was measured at pH 6.8 at supersaturations from about σ 0.1 to 2.7, without and with CAP Adp 0.85 (Figure 8; the polymer concentration is not stated for this figure). P10·X12 Above the change of mechanism the growth rate should rise linearly with σ, so the authors fitted straight lines to the high-σ points and extrapolated them to zero growth; the intercept is ${\sigma }_c$ (method of ref. 14). E14 The Kubota–Mullin analysis was then done below ${\sigma }_c$, at σ 0.99 (initial ritonavir 3.5 µg/mL): the growth rate ratio was measured over a range of CAP Adp 0.85 concentrations at pH 6.8 (Figure 10), and at 5 µg/mL at both pH; coverage came from the Langmuir–Freundlich fit and $\alpha$ from eq 7 (Table 4). P11

Experimental analysis

Main result 1: the higher the supersaturation, the less the polymers slow growth. Without polymer, seeds lowered the ritonavir concentration measurably at every supersaturation, and faster from a higher start: within 4 min from about 4.8 to 2.3 µg/mL at S 3.8, from about 10.0 to 5.3 at S 7.6 and from about 20.5 to 8.3 at S 15.3 (Figure 4). R4·C15 Note by the AI writer: the caption of Figure 4 assigns the blue triangles to S 3.8 and the black squares to S 15.3, but the blue triangles start at about 20.5 µg/mL; read by initial concentration, the legend is reversed. X1 Three polymers were then tested at the same three supersaturations: the most effective (CAP Adp 0.85), a somewhat effective (CA 320S Adp) and a marginally effective one (CAB Adp 0.25), as ranked in the earlier screen. R18

Table B. Growth rate ratio $R_{g0}/R_{gp}$ (above 1 means slower growth), read from Figure 5 (approximate; triplicates). The error bars of CAB Adp 0.25 span 1. Polymer concentration and pH are not stated for this figure. R5·X12

Table B shows the expected trend clearly for the two effective polymers: CAP Adp 0.85 slows growth about 6.4-fold at S 3.8, about threefold at S 7.6 and only about 1.6-fold at S 15.3, and CA 320S Adp falls from about 4.8 to 1.3. R5.a–b The authors describe CAB Adp 0.25 as "clearly ineffective" at all supersaturations, CA 320S Adp as somewhat effective at the lower supersaturations but ineffective at S 15.3, and CAP Adp 0.85 as the most effective at every supersaturation, although it reduced the growth rate by only "~50%" at S 15.3. C16 Two details do not match the plot exactly. A ratio of about 1.56 means a growth rate about 64% of the control, a reduction of about a third rather than a half; and CA 320S Adp at S 15.3 (about 1.30) is above 1, the authors' own threshold for an effective inhibitor. X2·X3 The direction, however, is unambiguous, and the authors see it as qualitative agreement with theory. C17 Their practical conclusion is strong: for ritonavir "it is not possible to substantially delay crystal growth" at the supersaturation generated by dissolving the amorphous solid, so "it can be surmised" that preventing seed formation (nucleation) is very important, as was found for felodipine; the rise of growth rate with supersaturation despite the polymer "indicates" an inherent limit to how much growth these polymers can inhibit. C18·C19 Effective polymers "would be expected" to still slow the overall de-supersaturation, especially as supersaturation falls; this was not measured. C20

Main result 2: at pH 3.8 the adipates lose much of their effect; HPMC does not. At S 7.6 and 5 µg/mL polymer, six polymers were compared at pH 3.8 and 6.8 (Figure 6). R7

Table C. Growth rate ratios at S 7.6 and 5 µg/mL polymer, read from Figure 6 (approximate; triplicates). The paper states only the 63% and 17% losses; the other three were calculated by the AI writer from the plot readings. R7·X17

At pH 6.8, where the adipate groups are ionized, the two polymers with the most carboxyl groups were the best inhibitors, ahead of HPMC. At pH 3.8 every carboxyl-bearing polymer lost effectiveness; the largest loss, 63%, was for CAP Adp 0.85, the polymer with the most ionizable groups and the best at pH 6.8, and the smallest, 17%, for CAB Adp 0.25. HPMC, with no ionizable groups, was comparable at both pH, "as expected". C23·C24 Note by the AI writer: between the two extremes the size of the loss does not follow the carboxyl content (CA 320S Adp, DS 0.67, lost about 35%, less than CAP Adp 0.33, DS 0.33, at about 44%), and at pH 3.8 two polymers fall below 1, that is, ritonavir grew slightly faster with them than without. The comparison also assumes that ritonavir's solubility, and so S 7.6, is the same at both pH; no solubility at pH 3.8 is reported. X17·X13 The authors note that this pH dependence "may have implications" for in vivo effectiveness, since pH varies along the gastrointestinal tract. C54

Main result 3: at pH 3.8 the polymer adsorbs more, yet inhibits less. Two explanations of the pH effect were open: less polymer adsorbs at pH 3.8, or the adsorbed polymer has a different conformation ("may be due either to reduced adsorption ... or a change in polymer conformation"). C26 The zeta potential of the seeds with 5 µg/mL CAP Adp 0.85 was −27.2 mV at pH 3.8 and −52.6 mV at pH 6.8, which the authors attribute to the stronger ionization of the adipate group at pH 6.8. R8·C25 The adsorption isotherms (Figure 7) decide between the explanations. At pH 3.8 the adsorbed amount rose to a plateau of about 41 mg/m² (about 40.8 mg/m² at about 19.7 µg/mL), at pH 6.8 to about 27 mg/m² (about 26.6 mg/m² at about 23.3 µg/mL). The Langmuir–Freundlich fits gave $K$ 0.0144 and $n$ 2.45 at pH 3.8 (adjusted R² 0.9957) and $K$ 0.0198 and $n$ 2.14 at pH 6.8 (R² 0.9621). R9·R10·C27 At the 5 µg/mL of the growth tests "approximately the same amount" of polymer adsorbed at pH 3.8; the paper does not name the comparator, but read as pH 3.8 against pH 6.8 it is consistent with Table 4 (coverage times plateau gives about 6.6 against 6.1 mg/m², an AI calculation). C28·X16 Hence the authors "can deduce" that the weak effect at pH 3.8 is not due to a lack of adsorption, and that effectiveness "may be more dependent" on the conformation of the polymer at the solid–liquid interface. C29

For the conformation argument they draw on poly(acrylic acid), whose repeat unit has a pKa of 4.76: fluorescence studies showed its chains stretching out as pH rises, and its mean square radius of gyration increased with pH and ionization (refs 50–52). R20 By analogy, they expect ionized cellulose ω-carboxyalkanoates to adopt a more extended chain, through repulsion between the charged groups and better interaction with water; an extended chain is "expected" to cover the crystal surface more effectively and block more growth sites, and the conformation "appears to impact" the ability of the polymer to block growth. C30·C31 They also name "alternative plausible explanations": the polymer may interact differently with incoming ritonavir molecules, or change the surface energy and solvation of the crystal face differently at the two pH values. C32

Main result 4: above a critical supersaturation the growth rate rises linearly and needle tips round off. Figure 8 shows growth rate against σ at pH 6.8 (Table D). Without polymer, growth was slow up to σ about 1.35 (about 0.06 to 0.71 µg mL⁻¹ min⁻¹) and then rose steeply and linearly to about 3.35 at σ 2.73. With CAP Adp 0.85, growth was about zero up to σ 0.99 and still small at 1.35 (about 0.10), then rose to about 2.19 at σ 2.73. R11

Table D. Growth rate against supersaturation at pH 6.8, read from Figure 8 (approximate; triplicates); σc as printed. The polymer concentration is not stated for this figure. R11·R12·X12

Extrapolating the linear high-σ parts to zero growth gave ${\sigma }_c$ 1.37 without polymer and 1.74 with CAP Adp 0.85, corresponding to initial ritonavir concentrations of 5.1 and 7.4 µg/mL. R12 Least-squares lines through the plotted points give about 1.38 and 1.78, close to the printed values. E14 Because ritonavir grows very anisotropically, the authors "believe" that this ${\sigma }_c$ is where the fast-growing faces, which dominate the growth rate, turn to continuous growth, and they attribute the higher ${\sigma }_c$ with polymer to CAP Adp 0.85 slowing these fastest-growing faces. C35·C36 SEM images (Figure 9) add a morphological sign: the seeds had flat-ended needles; after growth at 10 µg/mL (σ 2.0) the ends were slightly rounded, and after growth at 20 µg/mL (σ 2.7) they were clearly rounded. R13 The authors take the diminishing dependence of growth rate on supersaturation beyond ${\sigma }_c$, the linear growth at high supersaturation and the rounding of faces as "considerable support" for kinetic roughening and fast continuous growth at high supersaturation. C37 They also call ritonavir's ${\sigma }_c$ "very similar" to values reported for other organic compounds (Table 3). Note by the AI writer: Table 3 marks the values for the n-paraffins (0.99–1.32) and naphthalene (1.5) as percentages, that is σ of about 0.01–0.015, roughly 100 times below ritonavir's 1.37; only lysozyme (2.0), L-arginine trifluoroacetate (1.1) and glucose isomerase (5.0) are of the same order, and all were measured at about 20–25 °C rather than 37 °C. C38·R14·X10

Why does adsorbed polymer fail above ${\sigma }_c$? The authors' molecular picture: a molecule reaching the surface must find a site where it is strongly bound, such as a kink, or it returns to solution. In the rough regime the critical nucleus is tiny, growth units are incorporated at kinks, step edges and terraces, and the steps themselves merge and cease to exist; growth proceeds by random addition, so "it is likely" that favourable growth sites remain even with adsorbed polymer, whose density of interaction is insufficient to block every site (refs 58, 63). C49 As a test, growth at S 7.6 and pH 6.8 with 100 µg/mL CAP Adp 0.85, far above the adsorption plateau, still occurred, which they read as showing that even maximal adsorption does not block all active sites in the rough region. R17·C50 "We surmise", they conclude, that the change from slow layer growth to fast continuous growth is "most likely responsible" for the much lower polymer effectiveness at very high supersaturation. C51

Main result 5: below σc, step pinning describes the inhibition, and α depends on pH. At pH 6.8 and σ 0.99, below ${\sigma }_c$, the measured $R/R_0$ fell steeply with CAP Adp 0.85: to about 0.44, 0.23 and 0.09 at equilibrium polymer concentrations of about 0.14, 0.28 and 0.69 µg/mL, and to zero, complete inhibition, at initial concentrations of 5 and 7.5 µg/mL (equilibrium about 3.5 and 4.2 µg/mL) (Figure 10). R15·C39 The Kubota–Mullin curve, calculated with the Langmuir–Freundlich coverage and $\alpha$ 4.35, stayed near 1 at low concentration and agreed with the measurements only at 5 and 7.5 µg/mL. The authors attribute the low-concentration mismatch to the isotherm, which predicts almost no adsorption there, and see the agreement at the growth-test concentration as helping "support our supposition" that step pinning applies. R15·C40·C41

Table E. Kubota–Mullin values (from Table 4). $\alpha$ follows from eq 7: (1 − 0.80)/0.16 = 1.25 and (1 − 0)/0.23 = 4.35. R16·E7

Table E is the quantitative core of the pH argument. With about the same amount of polymer on the crystal, growth stopped completely at pH 6.8 but was only 20% slower at pH 3.8. R16 In Kubota–Mullin terms, $\alpha$ about 1 at pH 3.8 means the adsorbed molecules sit about one critical-nucleus diameter apart even at full coverage, so the polymer limits growth only when it covers the whole surface; $\alpha$ above 1 at pH 6.8 means they sit closer than that, so growth stops at a coverage of 0.23. More polymer is therefore needed at pH 3.8 to stop growth. E13·C44·C45 The authors call this result one that "strongly supports our contention" that the polymer is more condensed on the surface at pH 3.8, favouring contacts between its own segments, and extended at pH 6.8, "most likely" because of repulsion between ionized groups and better solvent interaction. C46 They also describe CAP Adp 0.85 as "an extremely effective growth inhibitor at this supersaturation" and agree with the suggestion that $\alpha$ is related to the chemical properties of the additive. C42·C43 Note by the AI writer: the coverage values reproduce from eq 6 at pH 6.8 only if the equilibrium concentration (about 3.5 µg/mL) is used rather than the initial 5 µg/mL named in the footnote of Table 4, and the pH 3.8 equilibrium concentration is not printed. E6·X9

Secondary result: the polymer changes crystal shape. Table 2 lists seed dimensions at S 7.6 (Table F). The authors read them as fastest growth along the needle axis without polymer, and as CAP Adp 0.85 inhibiting this lengthwise growth so that the crystals became more rod-like. R6·C21

Table F. Seed size before and after growth at an initial 10 µg/mL (S 7.6), 50 crystals per sample (from Table 2). R6

Note by the AI writer: the mean length with polymer (54.0 µm) is not shorter than without (50.4 µm); the aspect ratio fell because the crystals became nearly twice as wide (10.5 against 5.7 µm). The no-polymer rows are taken from ref. 7, and 54.0/10.5 gives 5.14 rather than the printed 5.2. The data therefore show wider crystals with polymer rather than a measured reduction in length growth. X4·X5

Discussion

The authors' chain of reasoning. The results fit together into one picture, which the paper builds step by step. A polymer adsorbs on the growing crystal and pins the advancing steps; whether this stops growth depends on how closely the adsorbed molecules sit compared with the critical 2D nucleus; ionization sets that spacing through the conformation of the chain; and supersaturation decides whether steps exist at all. C12·C45·C46·C51

Each link has a different evidential status. That growth inhibition weakens with supersaturation and with lower pH is measured directly and replicated (Figures 5, 6). R5·R7 That less adsorption is not the cause is measured (Figure 7), within the limits of the depletion method. R9·X15 The conformational explanation is inferred: no measurement of chain conformation on the crystal was made, and the support is the combination of higher adsorption with a smaller $\alpha$ and the analogy with poly(acrylic acid). C46·R20 The roughening explanation rests on the linear high-σ growth rate, its extrapolated intercept and the rounded needle tips; the step-pinning description below ${\sigma }_c$ rests on a single concentration per pH at which $\alpha$ is derived. R12·R13·R16

Where the conclusions apply. The study covers one drug (ritonavir, essentially unionized at both pH), one temperature (37 °C), seeds from one batch, initial concentrations from 1.5 to 20 µg/mL and polymer concentrations mostly at or below 7.5 µg/mL. M1·P7 The pH effect was measured only at S 7.6, and the supersaturation series and the ${\sigma }_c$ determination were done with buffers and polymer concentrations that the figure captions do not fully specify. R7·X12 Within these limits, the comparison with felodipine adds some generality: the same trend of weaker polymer inhibition at higher supersaturation was found there, and HPMC had a similar effect on both drugs at S about 7.5, "although this may be purely coincidental". C52·C53

What it suggests for formulation. The authors' message is direct: to keep the high supersaturation an amorphous formulation produces, an inhibitor must hinder the formation of seed crystals, "since it appears to be very difficult to substantially inhibit crystal growth under these conditions". C18 They also point out that the strong pH dependence of CAP Adp 0.85 "may have implications for in vivo effectiveness", and present the insights as a basis for designing new excipients with better crystallization-inhibiting properties. C54·C55 Note by the AI writer (not discussed in the paper): two practical consequences follow from the data. First, a growth-inhibitor ranking made at a moderate supersaturation such as S 7.6 can overstate what a polymer does near the amorphous solubility: CAP Adp 0.85 went from about a threefold slowdown at S 7.6 to about 1.6-fold at S 15.3. R5 Second, for an acidic polymer the test pH matters: at pH 3.8, roughly gastric to upper-intestinal acidity, the best adipate behaved like a weak inhibitor, while the neutral HPMC was unaffected. R7 Below ${\sigma }_c$, by contrast, the ionized polymer stopped growth completely at 5 µg/mL, so formulations that keep the dissolved drug well below the amorphous solubility give a growth inhibitor much more to work with. R16·R12

Open questions. The authors leave three questions open in their own text: whether the pH effect comes from conformation or from altered interaction with incoming ritonavir molecules or with the surface energy of the crystal face; whether the similarity of HPMC's effect on ritonavir and felodipine is a coincidence; and the face dependence of ${\sigma }_c$, which they handle by assuming that the measured value belongs to the fast-growing faces. C32·C53·C35 Further questions follow from what was not measured: the conformation of the adsorbed polymer at each pH; face-specific growth rates; nucleation itself, which the paper's main recommendation is about; and the Kubota–Mullin model above ${\sigma }_c$, which the abstract describes as "quantitatively inaccurate" there although no calculation above ${\sigma }_c$ is shown. C46·C36·C18·X11

Limitations

Limitations stated by the authors. The authors acknowledge that a single model of growth inhibition cannot apply at all supersaturations, because the growth mechanism changes, and that the Kubota–Mullin model "may not be applicable" at the high supersaturations generated by amorphous solids. C33 They state that the Langmuir–Freundlich fit does not describe adsorption at low polymer concentration, so the calculated growth rate ratio is poor at low coverage. C40 They note that ${\sigma }_c$ depends on the crystal face and that their value is taken to reflect the fast-growing faces ("we believe"). C35 And they hedge their explanations: the pH effect "is thought to be" due to conformation, with "alternative plausible explanations"; the high-supersaturation failure is what "we surmise" is "most likely responsible". C46·C32·C51

The AI writer's own judgement (1): the quantitative model tests are weaker than they look. The value $\alpha$ = 4.35 at pH 6.8 is calculated from the same 5 µg/mL measurement with which the Kubota–Mullin curve is then said to agree ($R/R_0$ = 0 at ${\theta }_{eq}$ = 0.23 gives $\alpha$ = 1/0.23), so the agreement at that point is built in. Because growth stopped completely, 4.35 is also the smallest $\alpha$ consistent with the data; any larger value would give the same zero. The independent low-concentration points (measured $R/R_0$ about 0.44 to 0.09 against calculated about 1.00 to 0.96) do not fit, which the authors assign to the isotherm. X8·R15·R16 The coverage values themselves reproduce from eq 6 only at the equilibrium concentration, which for pH 3.8 is not printed, and the seed mass and solution volume of the growth tests are not given. X9 Likewise, the abstract's statement that the Kubota–Mullin model was "quantitatively inaccurate" above the critical supersaturation is argued from theory; the model was applied only at σ 0.99. X11 Finally, the statement that ritonavir's ${\sigma }_c$ is "very similar" to that of other compounds holds only for the Table 3 entries not printed in percent. X10

The AI writer's own judgement (2): what the measurements can and cannot separate. Adsorption was measured as polymer lost from solution, without a crystal-free control; any loss of the less ionized polymer at pH 3.8 by other routes (aggregation, phase separation, adsorption to vessel walls) would be counted as adsorption and would raise the pH 3.8 plateau on which the conformational argument relies. X15 The zeta potentials were measured in 10 mM citrate buffer, not in the growth buffers, and without a polymer-free control, so they show a difference in surface charge but not how much of it comes from the polymer. X14 The growth "rate" is the de-supersaturation slope over the first 2 min; near the amorphous solubility, seed-induced secondary nucleation or formation of a drug-rich amorphous phase would also lower the concentration, and no measurement separates these from growth, so the steep high-σ rates are best read as overall de-supersaturation rates. X18 Several statements in the text go beyond the plotted data: the "~50%" reduction at S 15.3 (about a third on the plot), "ineffective" for ratios above 1, and lengthwise growth inhibition that the mean lengths do not show. X2·X3·X4 There are also small printing slips: the reversed colour legend of Figure 4, the reference to "equation 10" where eq 7 is meant (Figure 10), a total DS for CAB Adp 0.81 that does not equal its printed parts, and S printed as both 15.3 and 15.4. X1·X7·X6·X19 None of these changes the direction of the main findings; they mainly affect how firmly the mechanisms and the numbers can be quoted.

Conclusion

This paper asks what controls the ability of cellulose adipate polymers to slow ritonavir crystal growth, and answers with two measured dependences and two proposed mechanisms. Measured: inhibition falls steeply with supersaturation, from about a 6.4-fold slowdown at S 3.8 to about 1.6-fold at the amorphous solubility for the best polymer, CAP Adp 0.85; and the adipates lose much of their effect at pH 3.8, where their carboxyl groups are less ionized, although they adsorb more, while neutral HPMC is unaffected. R5·R7·R9 Proposed: above a critical supersaturation (σ 1.37 without, 1.74 with CAP Adp 0.85) ritonavir grows by continuous, rough growth that adsorbed polymer cannot block; and below it, the ionized polymer lies extended on the crystal and pins steps more effectively ($\alpha$ 4.35 against 1.25). R12·R16·C46·C51 For formulation, the authors' conclusion is that at the supersaturation of an amorphous product, nucleation, not growth, is the step a crystallization inhibitor has to stop. C18