How a trace of HPMC keeps felodipine supersaturated: nucleation versus crystal growth

Background and motivation

Many drug candidates dissolve so poorly that the amount reaching solution in the gut limits absorption. One common answer is to give the drug in its amorphous form, which dissolves faster than the crystal and can produce a supersaturated solution, that is, a solution holding more drug than the crystal's solubility allows C1. A supersaturated solution is, however, thermodynamically unstable: sooner or later crystals form and grow, and the concentration falls back toward the crystalline solubility. The authors name crystallization from the supersaturated solution as one of the most important concerns for all supersaturating systems, and its prevention as a condition for using them to raise oral exposure (Brouwers et al., ref 14; refs 19–21) C1.

Polymers dissolved in the medium are known to slow this crystallization, and their use as crystallization inhibitors in drug delivery goes back decades (refs 9–13) C2. Crystallization, however, has two steps: new crystals must first nucleate, and existing crystals then grow. A polymer could act on either. The authors point out that much research has gone into identifying polymers that stabilise supersaturated solutions, but much less into quantifying how much a polymer changes the nucleation rate and how much it changes the growth rate (ref 14) C2. Without that split, it is unclear what an inhibitor actually does, and therefore which formulation risks it can and cannot cover.

Tools for the split existed in crystal-growth science. An empirical power law relates growth rate to supersaturation and gives a growth order that hints at the rate-limiting step (Mullin, ref 22; Mohan and Myerson, ref 23) C3. Kubota and Mullin developed a model that relates the slowing of growth to the surface coverage of an adsorbed impurity, built on the step-pinning picture of Cabrera and Vermilyea (refs 25–30) C9. To the authors' knowledge it had been used for inorganic crystals but "has not been used to date" for polymers acting on organic pharmaceutical crystals C10. For nucleation, Verdoes, Kashchiev and van Rosmalen showed how seeded growth rates and unseeded induction times measured under identical conditions give a relative nucleation rate (ref 32) C11.

The study therefore asks a narrow, quantitative question with one drug and one polymer: how strongly does predissolved hydroxypropylmethyl cellulose (HPMC) inhibit the nucleation and the crystal growth of felodipine, an antihypertensive drug with very low aqueous solubility, from supersaturated solution, and which of the two effects matters more for keeping the solution supersaturated A1·A2·A3? Growth was measured with seed crystals present and nucleation without them, and both were analysed with empirical and mechanistic models P4·P5·P8·P9·P10.

Background concepts

Supersaturation ratio. The driving force for crystallization is expressed as the supersaturation ratio (eq 1):

In plain terms: $S$ says how many times more drug is dissolved ($C$) than the crystal can hold at equilibrium ($C_{eq}$). For felodipine in the pH 6.8 phosphate buffer used here, $C_{eq}$ is about 0.5 µg/mL at 25 °C, so $S$ = 10 corresponds to 5 µg/mL M1·E1. The authors chose supersaturations up to about 10 because that is the theoretical and measured solubility advantage of amorphous felodipine over the crystal (Alonzo et al., ref 38) C42. The low solubility has a practical consequence that the whole design rests on: so little drug is dissolved that crystallizing it hardly changes the size or surface area of seed crystals during an experiment C5.

Nucleation, growth and the induction time. In a seeded experiment, crystals are already present, so drug leaves solution mainly by depositing on them; the early slope of concentration against time is then a measure of the crystal growth rate C13. In an unseeded experiment, crystals must first form. The induction time is the time needed for stable nuclei to form and grow to a detectable size C11. It depends on both nucleation and growth, which is why the paper needs the seeded growth rates to separate the two.

Growth order. The empirical growth law (eq 2) and its logarithmic form used for fitting (eq 3) are:

In plain terms: the growth rate $R_G$ rises as a power $g$ of the supersaturation ratio, scaled by a rate constant $k_G$; on log–log axes the data fall on a line whose slope is $g$ and whose intercept is $\ln k_G$ E2·E3. The authors use the common reading of $g$: a value of 1 means that diffusion of molecules to the crystal surface limits growth, and a value of 2 that their incorporation into the lattice (surface integration) limits it; measured values usually lie between 1 and 2 C3.

The Kubota–Mullin impurity model. Kubota and Mullin describe growth in the presence of an adsorbing impurity by the ratio of the growth rate with impurity, $R$, to that without, $R_0$ (eq 5):

In plain terms: growth slows in proportion to the fraction of the crystal surface covered by impurity, ${\theta }_{eq}$, and the effectiveness factor $\alpha$ sets how far it slows at full coverage E5. The authors spell out the readings: $\alpha$ = 1 means growth stops at full coverage, $\alpha$ above 1 means it stops before full coverage, an $\alpha$ near 0 means no effect at any concentration, and an intermediate value such as 0.4 means an initial fall followed by a plateau above zero C25. The coverage is taken to follow a Langmuir isotherm (eq 6):

In plain terms: coverage rises with the impurity concentration $x$ and saturates at 1, with $K$ the adsorption equilibrium constant E6. The model was written for step and face growth rates; the authors assume, following Garside et al. (ref 29), that the same ratio applies to the overall mass growth rate that a bulk de-supersaturation experiment measures (eq 4) C4·E4:

In plain terms: the fractional slowing of a single growth step ($V$), of a crystal face ($G$) and of the whole crystal mass ($R$) are taken to be the same. Two further assumptions make this usable with seeds: the seed amount per unit surface area is the same with and without polymer, and the relative areas of the crystal faces do not change; the authors call the error from the second "likely to be small" C5·C6.

Step pinning and the effectiveness factor. The physical picture behind the model is the pinning mechanism of Cabrera and Vermilyea. Molecules adsorbed on a crystal face act as obstacles; an advancing growth step must bulge between them, and a curved step grows more slowly than a straight one (eq 8):

In plain terms: a step with radius of curvature $\rho$ grows at a fraction $1-{\rho }_c/\rho$ of the speed of a straight step $V_{\infty }$, and stops when its radius reaches the critical radius of a two-dimensional nucleus, ${\rho }_c$ E8·C9. The closer the adsorbed molecules sit, the tighter the steps must curve; if they are $2{\rho }_c$ apart, growth stops C9. The authors stress that this is not an increase in surface free energy caused by adsorption, which they call thermodynamically incorrect C9. Kubota and Mullin condensed the picture into an expression for $\alpha$ (eq 7):

In plain terms: an impurity is more effective when its adsorption sites lie closer together (small $L$, the separation of sites available for impurity adsorption) and less effective as supersaturation rises; $\gamma$ is the surface excess energy, $a$ the area occupied by one crystallizing molecule, $k$ the Boltzmann constant and $T$ the temperature E7. Because the critical radius shrinks as $S$ rises, any adsorbed impurity "should" lose its grip on growth at higher supersaturation, "regardless of the inherent ability" of the impurity C8. The authors keep $\ln S$ rather than the usual low-supersaturation approximation $S-1$, because their supersaturations are too high for it C7. None of $\gamma$, $a$ or $L$ is given in the paper, so eq 7 serves as a qualitative prediction, not a calculation E7.

From induction times to a nucleation rate. With seeds, a slower growth rate means a proportionally longer seeded induction time (eq 9); without seeds, the induction time depends on both nucleation and growth, so the nucleation-rate ratio follows from both kinds of experiment (eq 10):

In plain terms: $t_s$ and $t_{s,0}$ are the seeded induction times with and without inhibitor, replaced in practice by the growth-rate ratio; $t_u$ and $t_{u,0}$ are the unseeded induction times; $J$ and $J_0$ are the nucleation rates with and without inhibitor E9·E10. A longer unseeded induction time lowers the estimated nucleation rate with the fourth power, partly offset by the slower growth, which enters with the third power E10.

Methods in detail

The study has three experimental arms and two layers of analysis. Seeded de-supersaturation runs give growth rates at supersaturation ratios from 3 to 12 with and without HPMC; an adsorption isotherm gives the surface coverage of HPMC on felodipine crystals; unseeded runs give induction times. The growth data are analysed with the empirical power law and with the Kubota–Mullin model, and the growth and induction-time data together give the nucleation-rate ratio P4·P5·P6·P8·P9·P10.

Contribution 1: growth rates from seeded de-supersaturation

Problem. Felodipine's solubility is so low (about 0.5 µg/mL) that conventional ways of following crystal growth, such as weighing crystals or tracking size distributions, would see almost nothing. The authors therefore followed the drug remaining in solution, which they call the most practical option given the low solubility M1·P4.

How. Felodipine from AstraZeneca was ground with mortar and pestle and sieved below 106 µm to make seeds M1·P1. Laser diffraction (Malvern Mastersizer, in buffer with 0.1% w/v HPMC, 2500 rpm, sonication) gave an average specific surface area of 0.804 ± 0.3 m²/g and a surface-weighted mean size of 7.78 µm P2·M5. About 2.5 mg of seeds were placed in 18 mL of 50 mM phosphate buffer, pH 6.8, with or without predissolved HPMC (Pharmacoat 606), and left for 40 min at 25 °C; supersaturation was then created by adding felodipine dissolved in methanol P4·M2·M4. Concentration was followed every 2 s with a six-channel fiber-optic UV system, using a partial least-squares model on second-derivative spectra between 270 and 440 nm (one component, R² 0.999, Q² 0.999) P3. Second derivatives remove the sloping baselines that suspended particles add to UV spectra (Van Eerdenbrugh et al., ref 33) P3. The initial growth rate was the slope of concentration against time over the first 5 min P4. Initial supersaturations were 3, 5, 6, 7, 8, 10 and 12, and initial HPMC concentrations 0.2, 0.5, 1, 1.5, 2.5 and 3.5 µg/mL; all runs without HPMC were done in triplicate, runs with HPMC in triplicate at S 5 and 10 and once at the other supersaturations P4·A1·A2. A check at 50 and 100 µg/mL HPMC and extended 60-min runs at several supersaturations were added P4·R10·R3.

Why it works. At these concentrations the seeds offer far more surface than any crystals that might nucleate, so the early fall in concentration reflects growth on the seeds C13. Two controls support the readings. HPMC at 1 mg/mL, which the authors describe as three orders of magnitude more than the levels used here, does not change felodipine's equilibrium solubility (Konno et al., ref 35), so the polymer does not shift the end point R21·C12. And felodipine nanoparticles, which would absorb like dissolved drug and inflate the apparent concentration, shift the 361 nm peak to the red; no such shift appeared in the spectra R20·R19.

How it differs. Rather than ranking polymers by how long a solution stays supersaturated, the seeded design isolates one process and puts a number on it at each supersaturation, which the models then interpret.

Contribution 2: adsorption of HPMC and the Kubota–Mullin analysis

Problem. To use eq 5, the coverage of HPMC on the felodipine surface must be known at each polymer concentration, and the growth-rate ratios must be related to the polymer actually left in solution after adsorption C23·P7.

How. HPMC adsorption on the seeds was measured by surface tension (Krüss K12 tensiometer): calibration solutions of 0.2–5 µg/mL HPMC, seeds equilibrated 16–24 h in HPMC solutions of various concentrations, and a correction from seeds in plain buffer P6. The equilibrium polymer concentration in the growth experiments could not be measured this way, so it was estimated by interpolation from the isotherm data; Figures 4 and 5 use this equilibrium concentration, all other figures the initial concentration P7. The isotherm was fitted with a modified Langmuir equation (eq 11, ref 36):

In plain terms: the extra exponent $n$ lets coverage rise in an S-shaped step rather than gradually; the fit gave $K$ = 58.1 and $n$ = 6.9 with an adjusted R² of 0.79, better than the plain Langmuir form E11·R7. Note by the AI writer (from the card's extractor notes): although eq 6 defines $x$ as a mole fraction, the plotted coverage curve is reproduced only when $x$ is the equilibrium concentration in µg/mL (for example $\theta$ = 0.631 at 0.6 µg/mL), so $K$ here carries concentration units X11·E11. Coverage from this fit was then inserted into eq 5, and $\alpha$ was chosen at each supersaturation so that the predicted ratio matched the measured one P9.

Why it works, and its weak point. The fitted isotherm predicts almost no adsorption below about 0.3 µg/mL, so eq 5 predicts no slowing there, while growth was in fact already slower R7·R8·C23. The authors therefore "artificially adjusted" the predicted curve by removing its flat low-concentration part, and made the same adjustment at the other supersaturations C23. They explain the misfit by the Langmuir assumptions: the seed surface is "likely to be quite heterogeneous", and a polymer, much larger than the solvent, "will not rigorously comply" C21. Their support for the coverage estimate is that complete monolayer coverage falls at concentrations similar to those of maximum growth inhibition C22.

How it differs. The model had been tested on inorganic crystals; here it is applied, to the authors' knowledge for the first time, to a polymer acting on an organic drug crystal, with overall mass growth rates in place of face growth rates C10·C4.

Contribution 3: nucleation from unseeded induction times

Problem. A nucleation rate cannot be read directly from a de-supersaturation curve, because the drop seen after the induction time combines nucleation and growth of the new crystals C11.

How. Unseeded runs at S 10 were made without HPMC and with 1 µg/mL HPMC (n = 3); the induction time was taken where a line fitted to the flat initial data meets a line fitted to the linear falling part P5. With the growth-rate ratio at the same supersaturation and polymer level standing in for the seeded induction-time ratio (eq 9), eq 10 gives $J/J_0$ P10·E9·E10.

Why it works. The approach (Verdoes et al., ref 32) uses only quantities measured under identical conditions, so it needs no absolute nucleation theory and no surface-energy values C11.

Experimental analysis

Main result 1: trace HPMC slows growth on seeds at moderate supersaturation. At an initial S of 5 (Figure 1), the polymer-free seeded solution fell from about 2.49 to 1.77 µg/mL in 5 min, while the solutions containing 0.2, 0.5, 1 or 3.5 µg/mL HPMC fell only to about 2.08–2.22 µg/mL R1. From this the authors "surmise" that predissolved HPMC hinders felodipine growth even at very low concentration C14. At an initial S of 10 (Figure 2), the picture changes at the lowest level: with 0.2 µg/mL the curve matched the polymer-free one (both about 5.08 to 2.92 µg/mL); with 0.5 µg/mL the mean curve was slower (about 3.81 µg/mL at 5 min) but the replicates spread widely (SD about 0.48 µg/mL); with 1, 1.5 and 3.5 µg/mL the solutions ended at about 3.85, 3.93 and 4.06 µg/mL, which the authors call a "significant reduction" R2·C15. Their summary is that very low amounts of dissolved polymer affect growth, and that the loss of effect at S 10 "indicates" a dependence on supersaturation "consistent with theory" C16. Note by the AI writer (from the card's extractor notes): at the first time point the HPMC series start slightly higher than the controls (by up to about 8%), which would, if anything, make them grow faster, so it does not inflate the inhibition X13·R1·R2.

Table A. Initial growth rate of felodipine on seeds (µg/mL/min; slope of the first 5 min) at each initial supersaturation ratio and initial HPMC concentration; ± is half the drawn error bar (SD, n = 3), shown in the paper only for the control and for S 10. Read from Figure 3, approximate R4.

Main result 2: growth always rises with supersaturation, and the polymer matters less at high S. Table A shows the growth rate rising with S in every series R4. HPMC lowered the rate "in almost every system", but with 0.2 µg/mL the rates at S 8, 10 and 12 were the same as or higher than without polymer C17·C19. One 0.2 µg/mL point (S 6, about 0.053 µg/mL/min) "appears to be a random outlier" and was removed from all later analysis C18·R4. On log axes (Figure 6), 1 µg/mL HPMC lowered ln growth rate by about 0.9 at S 3 (about −3.47 against −2.55) but by only about 0.2 at S 12 (about −0.66 against −0.48) R12; the authors say the figure "clearly illustrates" the polymer's diminishing effectiveness C31. Because the growth rate rises with S whatever the polymer, they conclude that there is "an inherent limit to how much growth inhibition can be provided by HPMC" for felodipine at a given S C20.

Main result 3: inhibition levels off where adsorption saturates. The isotherm (Figure 4) is scattered: about 60–167 µg/mL/m² (unit as printed) adsorbed at equilibrium concentrations up to 0.38 µg/mL, and about 285–534 between 0.62 and 1.12 µg/mL R6. The fitted coverage is about 0.33 at 0.5 µg/mL, 0.63 at 0.6 and 0.98 at 1.0 µg/mL R7. At S 8 (Figure 5), the measured growth-rate ratio fell from about 0.95 at 0.03 µg/mL to 0.66 at 0.11 µg/mL and stayed at about 0.54–0.58 from 0.32 to 0.70 µg/mL (estimated equilibrium concentrations) R8. In the check at much higher polymer levels (data not shown), the ratios were 0.45 and 0.61 at S 5, printed as values "for both concentrations", and 0.45 at 50 µg/mL and 0.50 at 100 µg/mL at S 10 R10. The authors find these "in reasonable agreement" with plateaus of 0.65 (S 5) and 0.55 (S 10) derived from their α estimates, and conclude that the effect of HPMC on growth "rapidly reaches a plateau" R11·C29. Since the isotherm reaches full coverage at about 1 µg/mL, where inhibition levels off, they write that the plateau "lends support to the proposed adsorption mechanism of inhibition" C30.

Main result 4: the effectiveness factor falls with supersaturation, though not steadily. Table B lists the α values. The authors read them as a fall from 0.7 to 0.2 between S 3 and S 12, "indicating a reduction in the ability of HPMC to inhibit growth at the higher supersaturation", with intermediate values of 0.4–0.6 at S 5, 8 and 10 R9·C26. They say this "would be predicted a priori by eq 7", "suggesting" that the Kubota–Mullin model can approximate the felodipine–HPMC system, and call the weaker effect at high S their "most important experimental finding" C27. From the same values they "suggest" that adsorbed HPMC molecules sit relatively far apart on felodipine crystals, which limits their effect on growth C28.

Table B. Kubota–Mullin effectiveness factor at each supersaturation ratio, from Table 1 of the paper R9. For S 8 the Figure 5 caption gives 0.45 X1.

Three details of these numbers do not fit together, and they are worth knowing before using α outside the paper. Note by the AI writer (from the card's extractor notes): first, the α at S 8 is 0.4 in the text and in Table 1, but 0.45 in the Figure 5 caption, and the plotted model curve is reproduced with 0.45 (for example 0.716 at 0.6 µg/mL), not with 0.4 (0.748) X1·E5. Second, with eq 5 at full coverage the plateau should be $1-\alpha$, that is 0.5 at S 5 and 0.4 at S 10 from Table 1, not the 0.65 and 0.55 quoted "from the alpha estimates"; the quoted plateaus instead match the measured ratios at 3.5 µg/mL in Figure 3 (about 0.64 and 0.54) X2·R11. Third, the values do not fall steadily: α at S 10 (0.6) is higher than at S 8 (0.4) and S 5 (0.5), and runs with HPMC at S 3, 8 and 12 were single measurements X3·R9·P4. Eq 7 predicts α proportional to $1/\ln S$, a fall of about 2.3-fold from S 3 to S 12; the end values differ 3.5-fold X3·E7.

Main result 5: HPMC raises the growth order from 1.5 to about 2. Log–log fits of growth rate against S were linear, with R² of 0.94 or better for all data sets R12. Table C lists the slopes. Without polymer the order was about 1.50; with HPMC it was about 1.85–2.17, and the mean over all HPMC levels was 1.97 ± 0.12, which the authors take as the order with HPMC, "Assuming the rate limiting step in the growth is the same in all cases" R13·C32. They read the change as a shift from mixed diffusion and integration control toward integration control ("This suggests"; the abstract states it without hedge) C33. In the Kubota–Mullin picture, they add, slower growth comes from a smaller step radius and an unfavourable free-energy change on growth, which "would ultimately be linked" to the rate of incorporation of molecules, not to their diffusion to the surface C34.

Table C. Empirical growth parameters at each initial HPMC concentration; ln kG was calculated with g fixed at 1.5 without and 2 with HPMC and averaged over the supersaturations. Read from Figures 7 and 8, approximate R13·R14.

With g fixed at 1.5 without and 2 with HPMC, the rate constant calculated at each supersaturation showed no dependence on S, "As expected", only on HPMC concentration C35·C36. The authors describe the drop of kG at all polymer levels as "a significant reduction in growth rate" that "qualitatively agrees" with the Kubota–Mullin picture of an effect that plateaus quickly C37. Note by the AI writer (from the card's extractor notes): because the two kG values multiply different powers of S, their ratio is not a growth-rate ratio. The fall from about −4.24 to −5.75 at 1 µg/mL is a roughly 4.5-fold drop in kG, but at S 10 it corresponds to a growth-rate ratio of about 0.7, against the 0.607 actually measured; growth rates compared at equal S (Table A) are the better measure X8·R5.

Main result 6: HPMC delays nucleation strongly. In unseeded solutions at S 10 (Figure 9), the polymer-free solution began to lose drug after about 10–15 min and fell from about 5.35 to 1.54 µg/mL by about 150 min R15. With 1 µg/mL HPMC, the concentration stayed near 5.45 µg/mL for about an hour, was about 5.29 at 145 min, and then fell slowly to about 4.01 µg/mL at 349 min R15. The induction times were 15.3 and 136.5 min R16. Inserting the inverse of the growth-rate ratio at S 10 and 1 µg/mL (0.607, printed inverse 1.643) and the two induction times into eq 10 gives $J/J_0$ = 7.05 × 10⁻⁴, "a substantial (more than a thousand fold) reduction in the nucleation rate" R5·R17·C38. Recomputing gives 7.06 × 10⁻⁴ with $1/0.607$ = 1.647, so the printed result stands; the printed inverse is a rounding slip E10·X6.

Table D. The inputs and result of the nucleation analysis; growth rates read from Figure 3 (approximate) R4·R5·R16·R17.

Main result 7: with seeds present, HPMC cannot hold supersaturation. The decisive comparison is Figure 10, which plots a seeded 60-min run and the unseeded run, both at S 10 with 1 µg/mL HPMC R18. The seeded solution fell from about 5.10 µg/mL at the first point to about 3.3 at 10 min, 2.6 at 25 min and 2.17 at 58 min; the unseeded solution stayed at about 5.4–5.46 µg/mL for about 100 min and was still about 5.03 µg/mL at 198 min R18. In the extended seeded runs at several supersaturations, including 5 and 10, the concentration fell below 2.5 µg/mL in less than 45 min and then levelled off above the solubility of 0.5 µg/mL R3. The authors state that the extent of de-supersaturation was similar with or without HPMC over the experiments C43. They conclude that "it becomes quite clear from Figure 10 that inhibiting nucleation has a far more significant impact on maintaining supersaturation than inhibiting growth does", and that even at 1 µg/mL, where growth inhibition at S 10 is already at its maximum, HPMC's ability to keep supersaturation through growth inhibition is "still quite poor" C40·C41.

Discussion

The authors' chain of reasoning. The results fit together in a simple sequence. HPMC adsorbs on felodipine crystal surfaces, with coverage reaching its maximum near 1 µg/mL R7·C30. Adsorbed molecules pin growth steps, which slows growth and makes incorporation at the surface rate-limiting C9·C33·C34. Because the critical step radius shrinks as supersaturation rises, pinning by relatively widely spaced HPMC molecules loses effect at high S C8·C27·C28. The same trace of polymer, however, strongly delays the appearance of new crystals R16·R17. What happens to a supersaturated solution then depends on whether crystals are already present.

The authors' argument as a causal chain, with the values from Figures 9 and 10 and the growth-rate ratios R5·R10·R11·R16·R17·R18.

What the authors conclude for formulation. "It can be concluded", they write, that for felodipine with HPMC inhibiting nucleation "would be critical" with formulations that supersaturate after oral administration, and that "preventing the formation of crystals altogether would be the only effective method for stabilizing felodipine with HPMC"; characterising nucleation therefore becomes a key part of formulation development C44. Where nucleation cannot be prevented, extensive growth inhibition "might be necessary"; for felodipine it "would be extremely important" to keep residual crystallinity in an amorphous formulation to a minimum, because residual seed crystals "could have a large impact" on the supersaturation maintained, even with an inhibitor like HPMC C45. In the conclusions the authors state that ensuring nucleation does not take place "is the key factor" for keeping felodipine solutions supersaturated for a biologically relevant time, and that this "should be given priority when making formulation decisions" C47.

Where the conclusions apply. The evidence covers one drug (felodipine), one polymer grade (HPMC, Pharmacoat 606) at 0.2–3.5 µg/mL with a check at 50 and 100 µg/mL, a pH 6.8 phosphate buffer at 25 °C, and supersaturation ratios of 3 to 12 M1·M2·P4·A1·A2. The nucleation result rests on one supersaturation (10) and one polymer level (1 µg/mL) R16·R17. The seed loading, about 2.5 mg in 18 mL, is a large crystal surface compared with what residual crystallinity in a product would supply; the paper does not test smaller seed amounts P4.

Open questions. The authors leave open which inhibitors could provide the extensive growth inhibition that "might be necessary" when nucleation cannot be avoided C45. Note by the AI writer (not discussed in the paper): several further questions follow from the design. How the nucleation-rate reduction depends on supersaturation and polymer level is unknown, because only one condition was measured X7. The step-pinning reading rests on fitted parameters; the spacing of adsorbed HPMC and the step geometry were not observed C28·C34. Whether the picture holds at 37 °C, in biorelevant media, or for other polymers and drugs was not studied. And the amount of residual crystallinity that would defeat HPMC in an amorphous felodipine product was not measured C45.

Implications, as read by the AI writer. Note by the AI writer (not stated in the paper): for a drug–polymer pair that behaves like felodipine and HPMC, a seeded growth-rate assay alone would rank the polymer as a weak inhibitor and miss its main benefit, which shows only in unseeded experiments. The inhibition plateau near 1 µg/mL, together with the 50 and 100 µg/mL checks, suggests that adding more dissolved HPMC would not slow growth further for this pair R10·R11. In practice this places the burden on keeping crystals out: controls on residual crystallinity and on crystallization in the solid during storage may matter as much as the choice of polymer.

Limitations

Limitations stated by the authors. The authors acknowledge that the assumption of unchanged face-area ratios with and without inhibitor is "a potential source of error", though "likely to be small" C6. They state that the Langmuir isotherm does not strictly apply: the model "begins to breakdown" at low equilibrium concentrations, where it predicts no adsorption in conflict with the measured data, because the seed surface is "likely to be quite heterogeneous" and a polymer "will not rigorously comply" with the model C21. They had to adjust the predicted growth-rate curve "artificially" to remove its flat low-concentration part, and note that the adjustment was "necessary due to the limitations of the model" C23. They report that the error at 0.5 µg/mL HPMC and S 10 "was quite large" C15. And they removed one growth rate as an apparent outlier C18.

The AI writer's own judgement (1): the mechanistic layer is softer than the main result. The central result, the contrast between seeded and unseeded solutions with the same HPMC level, is direct and large: after an hour the two differ by more than 3 µg/mL, and the induction time rose about nine-fold R18·R16. The quantitative mechanistic layer rests on weaker ground. The α values come from fits with a coverage function that has an adjusted R² of 0.79 and was shifted by hand, so they are model-dependent X12. The α at S 8 differs between text and figure, the plateaus quoted from the α values do not follow from them, and α does not fall steadily with S X1·X2·X3. Eq 6 defines $x$ as a mole fraction, but the fitted constants only reproduce the plotted curve with $x$ in µg/mL, so $K$ cannot be compared with Langmuir constants from other work X11. The growth-order shift is seen at every HPMC level, but the growth-rate data behind the 0.5, 1.5 and 2.5 µg/mL fits are not shown R13·X10. The comparison of kG across different fixed growth orders overstates the growth-rate reduction X8.

The AI writer's own judgement (2): the size of the nucleation effect. The thousand-fold figure comes from one pair of induction times, each taken from the intersection of two fitted lines without a stated uncertainty, and the unseeded times enter eq 10 to the fourth power, so a 10% error in one of them changes $J/J_0$ by about 1.5-fold X7·E10. The order of magnitude is secure; the exact factor is not. The claim that seeded de-supersaturation was similar with and without HPMC over longer times rests on runs that are not shown: no seeded run without HPMC is plotted beyond 5 min X9. There are also small textual slips: the text refers to "1 mg/mL HPMC" for Figure 10 where the caption gives 1 µg/mL; the constant-surface-area argument names S 10 as the highest supersaturation although S 12 was used; and the printed inverse of 0.607 is 1.643 rather than 1.647 X5·X4·X6. None of these changes the main conclusion.

Conclusion

The paper separates, for one drug and one polymer, the two ways a dissolved polymer can slow crystallization from a supersaturated solution. Seeded experiments show that HPMC at µg/mL levels slows felodipine crystal growth, raises the growth order from about 1.5 to about 2 and loses effect as supersaturation rises, as the Kubota–Mullin adsorption model predicts in direction R4·R13·R9·C27. Unseeded experiments show that the same polymer level lengthens the induction time from 15.3 to 136.5 min, an estimated nucleation-rate ratio of 7.05 × 10⁻⁴ R16·R17. Put side by side at S 10 with 1 µg/mL HPMC, an unseeded solution stays supersaturated for hours while a seeded one loses most of its supersaturation within an hour R18. In the authors' words, the relative impact on nucleation "(factor of 1000) was much more pronounced than the impact on growth rate (factor of 2)" C46.

For formulation, the result moves the emphasis from slowing crystals to preventing them: for felodipine with HPMC, supersaturation lasts only as long as no crystals are present, which is why the authors ask for nucleation to be characterised and residual crystallinity in amorphous products to be minimised C44·C45·C47. How general this is, across drugs, polymers, temperatures and media, remains to be tested.