How dissolved polymers slow the growth of drug crystals: hydrophobicity, hydrogen bonds and the speed of growth

Background and motivation

Many new drug candidates dissolve too poorly to be absorbed well. Cosolvents, cyclodextrins and surfactants raise the apparent solubility, but they can also lower the free fraction of drug and with it the effective permeability across the gut wall, so their effect on the fraction absorbed is erratic (refs 11−19). C8 Delivering the drug in an amorphous form raises the solution concentration without that penalty, and absorption modelling has linked the resulting supersaturation to better bioavailability (refs 20−22). C8 The catch is that a supersaturated solution is unstable: the excess drug will crystallize unless both nucleation and crystal growth are held back for a physiologically relevant time.

The authors focus on growth. An amorphous formulation can carry trace crystals from manufacture or storage, or form them during dissolution, and these act as seeds; their absence "cannot be guaranteed", so a crystal growth inhibitor in solution is "highly desirable". C1·C40 Polymeric additives were already known to help. They had been proposed to act through intermolecular interactions in solution, mainly hydrogen bonding, and through steric hindrance (Raghavan 2001, ref 23; Gao 2004, ref 24). C6 To slow growth, a polymer has to adsorb on the crystal surface; adsorption is usually driven by hydrophobic interactions, by electrostatic attraction when polymer and surface carry opposite charges, and by hydrogen bonds when donor and acceptor groups are accessible (refs 23, 24, 50−55). C47

The study builds on the group's earlier work on ritonavir. In a structure−property study (Ilevbare 2012, ref 27), polymer hydrophobicity and the presence of ionized groups decided how well a polymer slowed ritonavir growth, and both very hydrophilic and very hydrophobic polymers failed. C7·R19 From those insights the Edgar group designed three second-generation cellulose esters, CA 320S Sub 0.63, CA 320S Sub 0.90 and CA 320S Seb (refs 28, 41). C20 Two further observations framed the question. Polymers had been only marginally effective at the very high supersaturations that amorphous forms create, and with seeds present single polymers could hardly stop desupersaturation of felodipine or ritonavir there (refs 41−43). C35 And a polymer that delays nucleation well need not slow growth well (refs 27, 42, 43, 49). C41

The paper therefore asks whether the polymer properties that slowed ritonavir growth also apply to structurally different drugs. It ranks sixteen polymers, including eleven synthesized cellulose ω-carboxyalkanoate esters with more chemical diversity than commercial cellulose derivatives, by how much they slow the growth of three chemically diverse, poorly water-soluble drugs: celecoxib, efavirenz and ritonavir. It then probes the role of hydrophobic interactions by changing the ionic strength, the role of hydrogen bonding by infrared spectroscopy, the role of polymer ionization by changing pH, and the limits of inhibition by raising the supersaturation up to the estimated amorphous solubility. A1·A2·A3·A4·A5·A6

Background concepts

Supersaturation. The driving force for crystal growth is the supersaturation. The paper expresses it as a ratio (Eq. 4):

Here $C$ is the concentration of the supersaturated solution and $C_0$ the equilibrium (crystalline) solubility at the same temperature. In plain terms: S = 2 means the solution holds twice as much drug as a saturated one. Strictly the driving force is set by thermodynamic activity; the authors assume concentration is a reasonable substitute because the solutions are extremely dilute. E4·C14 The crystalline solubilities at pH 6.8 and 37 °C are 1.5 µg/mL for celecoxib, 8.2 µg/mL for efavirenz and 1.3 µg/mL for ritonavir, so the same 10 µg/mL gives S of 6.6, 1.2 and 7.6. M1·M2·M3·R7

Seeded desupersaturation and the growth rate. A small, fixed amount of seed crystals is placed in buffer, a concentrated drug solution in methanol is added to create supersaturation, and the dissolved drug is followed by a UV probe. Because the seeds are already there, the fall in concentration over the first minutes measures growth of the existing crystals rather than nucleation of new ones. The initial bulk growth rate $R_g$ is the slope of the amount of drug in solution against time over the first 2 min. P10·E8 The effect of a polymer is summarized by the effectiveness ratio (Eq. 5):

Here $R_{g_0}$ and $R_{g_p}$ are the initial growth rates without and with polymer. In plain terms: $E_g$ = 1 means the polymer does nothing, 2 that growth is halved, 10 that it is ten times slower. The authors treat $E_g$ < 1 as ineffective and $E_g$ > 1 as some inhibition. E5·C17 The method relies on new crystals not forming: such secondary nucleation would also remove drug from solution and is watched for as a rise in turbidity, measured as extinction at 280 nm (ritonavir) or 350 nm (celecoxib, efavirenz), where the drugs do not absorb. P11

Solubility parameter as a hydrophobicity scale. To rank polymers and drugs by hydrophobicity, the authors estimate the solubility parameter δ (SP) by the Fedors group-contribution method (Eq. 1):

Here $\Delta e_i$ and $\Delta v_i$ are the contributions of each atom or group $i$ to the energy of vaporization and to the molar volume, $\Delta E_v$ is the energy of vaporization and $V$ the molar volume. In plain terms: the SP is the square root of the cohesive energy per unit volume, and it can be computed from the structural formula alone. A lower SP means a more hydrophobic compound. E1·C10 The drugs span 20.03 MPa^1/2 (ritonavir, the most hydrophobic, Log P 5.6) to 23.38 MPa^1/2 (celecoxib, the least hydrophobic, Log P 3.5), with efavirenz at 21.89 (Log P 4.7); the polymers span 19.62 (CAB Seb) to 28.39 MPa^1/2 (PVP). M1·M2·M3·M4·M19

Theoretical amorphous solubility. The highest supersaturation an amorphous form can produce is its own solubility. The paper estimates it from the crystalline solubility $C_{\mathrm{e}\mathrm{q}}$ (Eq. 2):

with the free energy difference between amorphous and crystalline drug estimated at 37 °C from the heat of fusion $\Delta H_f$ and the melting point $T_m$ (Eq. 3):

In plain terms: the amorphous drug is more soluble than the crystal by the Boltzmann factor of its excess free energy, and slightly less soluble than that because the amorphous solid takes up water, which lowers its activity; the water term $-I(a_2)$ comes from applying the Gibbs−Duhem equation to the water sorption isotherm of the amorphous drug (ref 33). E2·E3·E7·P6·P7 The estimates are 20.6 µg/mL (ritonavir), 19.8 µg/mL (efavirenz) and 22.6 µg/mL (celecoxib). Dissolving spin-coated amorphous films gave 19.8, 12.3 and 2.5 µg/mL; the paper does not comment on the large gap for celecoxib. M1·M2·M3·P9·X9

Salting out. Adding salt makes water a poorer solvent for hydrophobic groups, so hydrophobic parts of a polymer associate more strongly with a hydrophobic surface (ref 44). If a polymer reaches the crystal mainly by hydrophobic attraction, it "should" therefore work better at higher ionic strength. C27

Hydrogen bonds in the infrared. An N−H or O−H group that donates a hydrogen bond has a slightly longer, weaker X−H bond, so its stretching band moves to a lower wavenumber (a red shift); the region 3500−3200 cm⁻¹ is used to classify such bonds (refs 45, 46). C31 Celecoxib carries a sulfonamide N−H and efavirenz an amide (carbamate) N−H; PVP and PVPVA carry a carbonyl oxygen that can accept a hydrogen bond (Figure 13). C51

Ionization of the carboxylated cellulose esters. The adipate, suberate and sebacate side chains of the synthesized esters end in a carboxylic acid. They are nearly fully ionized (about 100%) at pH 6.8 and less than 25% ionized at pH 3.8 (refs 47, 48), while celecoxib stays un-ionized at both. R18·C33 The degree of substitution (DS) counts how many of the three hydroxyl groups per glucose unit carry a given group: for example CA 320S Sub 0.90 has 0.90 suberate and 1.82 acetate groups per unit (Table 5). M12

Methods in detail

The study has three parts: a single growth-rate assay applied to every drug−polymer pair, three probes of the forces behind the polymer effect, and a supersaturation series that tests the limits of inhibition. The flow below shows how they connect.

Contribution 1: one growth assay for sixteen polymers and three drugs

① Problem. To compare polymers across drugs, the growth rate must be measured the same way for every pair, the seeds must be well defined, and the loss of drug must come from growth of the seeds rather than from new crystals.

② What they did. The seeds were used as received and sieved below 250 µm. PXRD showed celecoxib peaks at 5.4, 14.9, 16.2 and 22.5° 2θ and efavirenz peaks at 6.0, 14.1, 21.1 and 24.8° 2θ, matching patterns predicted from the Cambridge Structural Database for the stable Form III and Form I (Figure S1); melting points of 163.5 and 139.0 °C agreed with literature values. R1·M1·M3·C11 SEM showed plate-like celecoxib and rod-like efavirenz crystals (Figure S2), and BET nitrogen adsorption gave surface areas of 1.145 (celecoxib), 1.770 (efavirenz) and 1.265 m²/g (ritonavir). R2·R7·P4 For each run, 0.010 g of seeds was equilibrated in 50 mL of sodium phosphate buffer, pH 6.8, in a jacketed beaker at 37 °C, with or without 5 µg/mL of predissolved polymer, and stirred at 400 rpm by an overhead axial-flow impeller. P10 About 0.25 mL of a 4 mg/mL methanol stock was then added (1:200 methanol to buffer, which did not measurably change the solubility), and UV spectra were taken every 5 s by a fiber-optic probe. P10·R3 All runs were in triplicate. P10 At the polymer concentration used, the polymers do not change the crystalline solubility (ref 34). R4

The sixteen polymers were three synthetic ones (PVP, PVPVA, PAA), two commercial cellulose derivatives (HPMC, HPMCAS) and eleven cellulose ω-carboxyalkanoate esters made in the Edgar group (Tables 1 and 2, Figure 2). [M4−M19] The esters combine a cellulose backbone with acetate, propionate or butyrate groups and an adipate, suberate or sebacate side chain that ends in a carboxylic acid; they are not regioselectively substituted. M9 All sixteen were compared at 10 µg/mL drug, and their order by SP sets the order of the bars in Figures 4−6. A2·M4−M19

③ Why it works. At these low concentrations the turbidity of celecoxib and ritonavir solutions did not rise between 5 and 22 µg/mL, so the authors conclude that desupersaturation was "mainly due to the bulk crystal growth of the seed crystals"; for ritonavir, particle size and mass balance data (ref 43) support this. R5·R6·C12 Efavirenz behaved differently: above 10 µg/mL (S 1.2) turbidity rose as macroscopic crystals formed, a sign of secondary nucleation, so all its polymer experiments were run at S 1.2. R5·C13 ④ Difference from earlier work. The ritonavir study (ref 27) had a single drug; here the same panel is applied to three drugs that differ in hydrophobicity, ionization and, as it turns out, growth speed, so that drug-specific and general polymer properties can be told apart. Most of the ritonavir data at 10 µg/mL had been reported before and are included for comparison. X15

Contribution 2: three probes of the forces that make a polymer work

① Problem. A ranking shows which polymers work but not why. The authors' hypothesis is that adsorption on the crystal is driven by hydrophobic interaction for the cellulose polymers and by specific interactions where the structures allow. C47·C54

② What they did. First, the effectiveness of PVPVA and five cellulose polymers (CA 320S Adp, HPMC, CA 320S Sub 0.63, CA 320S Seb, CAP Adp 0.85) was measured with celecoxib and efavirenz in 50 and 100 mM sodium phosphate buffer, pH 6.8, on the reasoning that more salt strengthens hydrophobic association. P13·C27 Second, celecoxib and efavirenz seeds (about 10 mg) were held overnight in 20 mL of 100 µg/mL PVPVA in buffer, recovered by ultracentrifugation at 40 000 rpm (274 356g) for 15 min, dried and measured by ATR-FTIR (128 scans, 4 cm⁻¹), alongside seeds conditioned without polymer. P14 Third, the effectiveness of three carboxylated esters (CA 320S Sub 0.63, CAP Adp 0.85, CAB Adp 0.81) and of HPMC, which has nothing to ionize, was compared with celecoxib at pH 3.8 (100 mM sodium acetate) and pH 6.8 (100 mM sodium phosphate). P12

③ Why it works. Each probe changes one interaction while leaving the others roughly in place: salt changes hydrophobic association, the FTIR looks for the signature of a hydrogen bond, and pH changes the charge of the side chains but not of celecoxib or HPMC. Note by the AI writer (not discussed in the paper): the probes are not perfectly clean. "Ionic strength" is given as buffer molarity, the pH comparison also swaps the buffer species (acetate for phosphate), and the growth rate without polymer at 50 mM or at pH 3.8 is not reported, so part of a change in $E_g$ could come from $R_{g_0}$ rather than from the polymer. X11 The buffer molarity of the main series is not stated either; its values equal the 100 mM bars of Figures 7 and 8 and the pH 6.8 bars of Figure 10, which points to 100 mM. X10

Contribution 3: the supersaturation series and the normalized growth rate

① Problem. Amorphous forms create supersaturations far above those at which most growth studies are run, and polymers had been found weak there. C35

② What they did. Growth rates of celecoxib (S 3.3, 6.6 and 14.7) and ritonavir (S 3.8, 7.6 and 15.4) were measured without polymer and with HPMC, CA 320S Sub 0.63 and CA 320S Seb; the highest levels, 22 and 20 µg/mL, match the theoretical amorphous "solubility" of 22.6 and 20.6 µg/mL. P16·M2·M3 Efavirenz was left out because of secondary nucleation. P16·R5 To compare the intrinsic growth speed of the three drugs, the authors divided each growth rate by the seed mass times the BET area:

normalized growth rate = $R_g/(m_{\mathrm{s}\mathrm{e}\mathrm{e}\mathrm{d}}\times A_{\mathrm{B}\mathrm{E}\mathrm{T}})$, with $m_{\mathrm{s}\mathrm{e}\mathrm{e}\mathrm{d}}$ = 0.010 g, giving mg min⁻¹ m⁻² (stated in words in the paper). E6·P15

③ Why it works. Normalizing by surface area removes the effect of different seed sizes and shapes, so the drugs can be compared at similar supersaturation. ④ Difference from earlier work. Earlier ritonavir studies had shown weak inhibition at high supersaturation with first-generation esters (ref 41); this series tests whether the second-generation esters and HPMC do better, and adds a second drug. C57·R17

Experimental analysis

Main result 1: the three drugs grow at very different speeds. Without polymer, the seeds removed 2.50 µg min⁻¹ of ritonavir at S 1.2, 10.5 µg min⁻¹ of celecoxib at S 1.3 and 52.5 µg min⁻¹ of efavirenz at S 1.2; per unit of seed area this is 0.20, 0.91 and 2.97 mg min⁻¹ m⁻² (Table 6). R7.a·R7.b·R7.e Efavirenz therefore grows about 3 times faster than celecoxib and about 15 times faster than ritonavir at similar supersaturation. R7·C15 Growth speeds up steeply with supersaturation, as expected from theory, and at S of about 7 the normalized rates of celecoxib and ritonavir become "very similar" (6.23 and 6.14 mg min⁻¹ m⁻²). R7.c·R7.f·C15·C36

Table A. Crystal growth rates without polymer, 0.010 g seeds, 37 °C, pH 6.8, mean ± SD of three runs (from Table 6). R7

Two details of Table A need care. The celecoxib normalized rates at 10 and 22 µg/mL (6.23 and 19.08) cannot be recomputed exactly from the printed rate, seed mass and BET area (they give 6.17 and 18.91); they correspond to an area of about 1.14 m²/g, a 1% difference that does not change the ranking. E6·X8 And the same celecoxib condition (10 µg/mL, no polymer) appears as 71.0 ± 0.06 µg/min in Table 6, 69.80 µg/min in the fit of Figure 3 and about 68 µg/min in Figure 11, so the very small SDs of Table 6 understate how reproducible a growth rate is between figures. R7.c·R8·R14.a·X7 The abstract describes all three low-S rates as "at a similar supersaturation ratio of ∼1.2", while the celecoxib row is at S 1.3. X6

Main result 2: a middle window of polymer hydrophobicity. Table B lists the effectiveness of every polymer, ordered by SP as in Figures 4−6. Among the cellulose polymers, those with SP between 20.56 and 23.28 MPa^1/2 slowed all three drugs, while the three most hydrophobic esters, CAB Adp 0.25, CAP Seb and CAB Seb (SP 19.62−20.05), left $E_g$ at about 0.85−1.3. R9·C43·M17·M18·M19 The authors' summary is that "in general" the most hydrophobic new esters were ineffective, the moderately hydrophobic cellulose polymers effective and the very hydrophilic synthetic polymers "highly variable"; they read this as further support that cellulose polymers interact with the drugs through nonspecific, hydrophobic interactions. C26·C43

Table B. Effectiveness $E_g$ at 10 µg/mL drug and 5 µg/mL polymer, read from Figures 4−6 (approximate); SP from Table 3. Most of the ritonavir values were reported earlier (refs 27, 29). R9·M4−M19

Three cautions apply to this table. The authors count 12 (celecoxib), 13 (efavirenz) and 10 (ritonavir) of the 16 polymers as effective "to different extents" and state the criterion as $E_g$ > 1; by the bars, 15 polymers exceed 1 for celecoxib, and the three counts match only if bars up to about 1.2 are treated as ineffective. C19·C17·X3 In the discussion the failing range is printed as "SP values greater than 20.56", which contradicts the paper's own rule that lower SP is more hydrophobic; the very hydrophobic polymers that fail all have SP below 20.56. C43·C10·X2 And the window is not a sharp cut for efavirenz, where CAB Adp 0.25 (SP 20.05) reached about 1.3 and is among the 13 counted as effective. R9.n·X2

Main result 3: the best polymer depends on the drug. For celecoxib, the moderately hydrophobic commercial polymers were best, HPMC (10.2) > HPMCAS > PVPVA, and the best new ester was CA 320S Sub 0.90 at about 7. C21·R9.f·R9.i·R9.c·R9.g For efavirenz, a chemically diverse set worked; PVPVA and CA 320S Sub 0.90 both cut growth about 5-fold. The text ranks "PVPVA ≈ CA 320S Sub 0.90 > CAP Adp 0.85 ≈ HPMCAS" but omits CA 320S Sub 0.63, which at about 3.8 is the third best. C22·R9.h·X4 For ritonavir, the second-generation esters dominated, CA 320S Sub 0.90 (12.8 in the text, about 12.6 in Figure 6) > CA 320S Sub 0.63 (about 10.8) > CA 320S Seb (about 5.7) > CAP Adp 0.85 > CAB Adp 0.81, and were "much more effective" than the commercial polymers and the best first-generation esters (about 3.0 and 2.7). C23·C24·C20 The authors read the pattern as matching polymer to drug: the more hydrophobic ritonavir and efavirenz pair best with the CA 320S suberates and sebacate, the least hydrophobic celecoxib with the "slightly more hydrophilic" HPMC. C44 The synthetic polymers split: PVP and PVPVA slowed celecoxib and efavirenz but not ritonavir, and PAA slowed none of them. C25·R9.a−c

Main result 4: salt strengthens the cellulose polymers but not PVPVA. Table C shows that going from 50 to 100 mM phosphate raised the effectiveness of all five cellulose polymers for both drugs, while PVPVA hardly changed. R10·C28·C29 HPMC changed most: from about 6.45 to 10.15 with celecoxib and from about 1.1 to 2.55 with efavirenz. The text gives the drop at the lower ionic strength as "56% and 36%" for celecoxib and efavirenz, but the bars give about 36% for celecoxib and 56% for efavirenz, so the two numbers appear to have been swapped. R11·C30·X1 The authors conclude that the importance of hydrophobic interactions for the cellulose derivatives "was verified", and that PVPVA's insensitivity is "suggesting" a different, non-hydrophobic interaction. C48·C29

Table C. Effectiveness $E_g$ at two phosphate buffer concentrations (Figures 7 and 8) and at two pH values (Figure 10, 100 mM acetate at pH 3.8 and 100 mM phosphate at pH 6.8); drug 10 µg/mL, polymer 5 µg/mL; read from the figures (approximate). A dash means not tested. R10·R13

Main result 5: PVPVA shifts the drugs' N−H bands. After overnight exposure to PVPVA, the N−H stretching peaks of celecoxib moved from 3331 to 3321 cm⁻¹ and from 3225 to 3217 cm⁻¹, and the N−H peak of efavirenz from 3316 to 3308 cm⁻¹; other peaks were essentially unchanged (Figure 9). R12 The authors write that these red shifts "suggest" specific interactions between the drugs and PVPVA, which "appear to contribute" to its effectiveness. C32 In the discussion they "surmise" that PVPVA adsorbs mainly through hydrogen bonds from the efavirenz amide and the celecoxib sulfonamide to the PVPVA carbonyl oxygen, electrostatic forces being unlikely because all three are un-ionized at pH 6.8, and that PVPVA beats PVP, which also carries a carbonyl, because it is more hydrophobic ("may be attributed"). C51·C52

Main result 6: the carboxylated esters need to be ionized. With celecoxib at pH 3.8, CA 320S Sub 0.63 fell from about 6.1 to 1.5, CAP Adp 0.85 from 3.4 to 1.2 and CAB Adp 0.81 from 2.4 to 1.1, a loss of 54−75% that was largest for the most effective ester (75%); HPMC stayed near 10 "as expected" (Table C). R13·C34 The same had been seen with ritonavir (ref 41). C34

Main result 7: inhibition fades with supersaturation but does not vanish. For celecoxib (Figure 11) the growth rate without polymer was about 24, 68 and 218 µg/min at S 3.3, 6.6 and 14.7; with HPMC it was about 1, 6 and 23 µg/min, with CA 320S Sub 0.63 about 3, 10 and 38 µg/min, and with CA 320S Seb about 4, 16 and 135 µg/min. R14 HPMC was the best polymer at every level and CA 320S Sub 0.63 "almost as effective"; growth was "virtually completely inhibited" at S 3.3. C37·C38 The text states that HPMC cut growth by a factor of 7.4 at S 14.7, against about 10 at S 6.6; the plotted points at S 14.7 give about 9.5, so the printed factor cannot be reproduced from the figure. C38·X5 For ritonavir (Figure 12), the rate without polymer was about 36, 78 and 182 µg/min at S 3.8, 7.6 and 15.4; HPMC gave about 11, 29 and 170 µg/min, "somewhat effective" at the lower levels and ineffective at the theoretical amorphous solubility, while CA 320S Sub 0.63 gave about 0.5, 7 and 63 µg/min, complete inhibition at S 3.8 and a factor of about 3.0 at S 15.4. R15·C39 That factor compares with about 1.4 for the earlier best ester, CAP Adp 0.85 (ref 41), and about 3.3 for a combination of CAP Adp 0.85 and poly(N-isopropylacrylamide) (ref 43). C57·R17 The discussion repeats these ritonavir results with S values of 3.4 and 15.8, which do not match the 3.8 and 15.4 of the results and of S = C/C0. C56·X6

Controls and what the data mean. HPMC serves as the built-in control in the pH test, PVPVA in the salt test, and the methanol carrier was checked for its effect on solubility. R13.a·R10.a·R3 The two probes point the same way for the cellulose polymers: they gain with salt and, for the carboxylated ones, lose when they are un-ionized, which fits adsorption driven by hydrophobic contact and helped by charged, extended side chains. R10·R13·C48·C59 Note by the AI writer (not discussed in the paper): the CA 320S Seb result in Figure 11 is a useful warning about rankings made at one supersaturation; it cut celecoxib growth about 4-fold at S 6.6 but only about 1.6-fold at S 14.7. R14.d

Discussion

The authors' picture. The discussion assembles the results into one chain. A polymer must adsorb on the crystal to slow its growth; adsorption "appears to be mainly driven by hydrophobic interactions (in particular for cellulose derivatives)", and specific interactions add to it "when the chemical structures of the drug and polymer favor their formation". C47·C54 Which polymer adsorbs depends on its balance of hydrophilic and hydrophobic groups relative to the drug. C49·C60

In this picture the very hydrophobic esters fail because they "are likely to interact more favorably with other polymer molecules in solution", which "may prevent" adsorption or leave only weak contact with the crystal. C50 PVPVA, a hydrophilic polymer, works for celecoxib and efavirenz because their N−H groups can hydrogen-bond to its carbonyl, and it fails for ritonavir. C51·C25 Ionization helps the carboxylated esters because, it "is thought", charged chains adsorb in a more extended conformation that blocks drug attachment more effectively (ref 41). C59 The authors add that the amphiphilic structure of the CA 320S suberates, a fairly hydrophilic cellulose acetate backbone with an ionizable fatty-acid side chain, "may promote" interaction with drugs. C46

The role of growth speed. The second idea in the discussion is that how much a polymer can do depends on how fast the drug grows. Even at S 1.2 the best polymers cut efavirenz growth only about 5-fold, whereas celecoxib (HPMC, about 10-fold) and ritonavir (CA 320S Sub 0.63 and 0.90, 11 to 13-fold) were slowed more at much higher supersaturations (6.6 and 7.6). C55·R9 From this the authors conclude that for efavirenz "it is clearly not feasible" to block growth appreciably at a supersaturation of interest, and more generally that for a given driving force "it may be more difficult to retard crystal growth with polymeric additives for compounds that grow at a fast rate". C55·C5 At high supersaturation, where growth is fast for every drug, inhibition is harder; still, "it is apparent" that growth can be slowed at the concentrations an amorphous form generates "if a polymer with superior crystallization inhibitory properties is present", "at least for two of the three model compounds". C58

Nucleation versus growth. The best nucleation inhibitors found in the companion induction-time study (ref 34) for ritonavir and efavirenz, CAB Adp 0.81, CAP Adp 0.33 and CAP Adp 0.85, are not the best growth inhibitors found here, CA 320S Sub 0.63 and 0.90; for celecoxib, HPMC led on both. R16·C41 The two properties have to be tested separately, and since trace crystals can never be ruled out, the authors argue that growth inhibition deserves a specific test. C40·C41

Where the approach applies. The results cover three neutral, poorly soluble drugs at pH 6.8 and 37 °C, sixteen polymers at a single concentration of 5 µg/mL, seeded growth over the first 2 min, and supersaturations up to the theoretical amorphous solubility for celecoxib and ritonavir but only S 1.2 for efavirenz. A1·A2·A3·P10 Within that scope, the CA 320S suberates were the most consistent polymers: about 6−7 for celecoxib, 3.8−4.9 for efavirenz and 10.8−12.6 for ritonavir. The authors call them "interesting candidate polymers for supersaturating dosage forms" and consider comparison with HPMC and HPMCAS warranted. R9.g·R9.h·C45 They close with the hope that the identified polymer properties "help to guide the development of new excipients with superior crystallization inhibitory properties". C62

Note by the AI writer (not discussed in the paper): what it suggests for formulation. The useful variable is hydrophobicity relative to the drug rather than hydrophobicity alone: HPMC suits the least hydrophobic drug, the CA 320S suberates the most hydrophobic one. A practical screen would start from polymers with SP close to the drug's, test them at the supersaturation the formulation will actually reach (Figure 11 shows how much of an advantage at S 6.6 can disappear at S 14.7), and check whether a carboxylated polymer will be ionized where it has to work. For a drug that grows as fast as efavirenz, keeping residual crystals out of the product may matter more than adding an inhibitor. R9·R14·R13·R7

Limitations

Limitations stated by the authors. The authors note that the efavirenz experiments ran at a very low initial supersaturation (S 1.2) because of secondary nucleation, and they restrict the claim of inhibition at amorphous-level supersaturation to "at least two of the three model compounds". C13·C22·C58 Their mechanistic statements carry hedges throughout: "appeared", "are likely to", "we surmise", "may be attributed", "It is thought". C2·C4·C51·C52·C59

The AI writer's own judgement (1): the mechanism is inferred, not measured. Every explanation in the paper rests on adsorption of the polymer on the crystal, but adsorption is never measured, nor is the conformation of the adsorbed chains. X14 The hydrophobic contribution is inferred from the salt effect, which was tested only with celecoxib and efavirenz; for ritonavir the support is the SP trend alone. X14 The hydrogen bond between the drug N−H and the PVPVA carbonyl rests on red shifts of 8−10 cm⁻¹ in the crystal N−H bands; the carbonyl region is not shown, the amount of PVPVA on the crystals is not known, and no spectra are reported for PVP, the cellulose polymers or ritonavir. X13 The salt and pH probes each change more than one variable (buffer molarity rather than computed ionic strength; acetate instead of phosphate at pH 3.8), and the growth rates without polymer under those conditions are not given. X11

The AI writer's own judgement (2): numbers in the text that do not match the figures. Several printed statements disagree with the paper's own data: the HPMC ionic-strength percentages appear swapped between celecoxib and efavirenz; the HPMC factor of 7.4 at S 14.7 does not match Figure 11 (about 9.5); the SP threshold for failing polymers is printed as "greater than 20.56" instead of below it; the efavirenz ranking omits CA 320S Sub 0.63; the counts of effective polymers do not follow the stated $E_g$ > 1 rule; the discussion gives ritonavir supersaturations of 3.4 and 15.8 instead of 3.8 and 15.4; and the same celecoxib growth rate appears as 71.0, 69.80 and about 68 µg/min in three places. X1·X5·X2·X4·X3·X6·X7 None of these reverses a conclusion, but they mean that exact factors quoted from the text should be checked against the figures. The buffer molarity of the main series is not stated. X10

The AI writer's own judgement (3): what "amorphous solubility" means here. The supersaturation series ends at the theoretical amorphous solubility. For celecoxib, the measured amorphous "solubility" was only 2.5 µg/mL, about a ninth of the 22.6 µg/mL estimate and below every concentration used with polymers; the paper does not discuss this, so it is unclear how closely the celecoxib solutions at 10 and 22 µg/mL represent what a dissolving amorphous celecoxib formulation would produce. X9 Most ritonavir effectiveness values were published earlier, and ritonavir values agree across figures only to about 20%. X15 Finally, the conclusion about fast-growing drugs rests on one drug (efavirenz) tested at one supersaturation and compared with the others at different supersaturations. X12

Conclusion

The paper makes three contributions. First, it ranks sixteen polymers, including eleven new cellulose ω-carboxyalkanoate esters, by one seeded-growth assay across three structurally different drugs, and shows that a middle window of hydrophobicity (SP about 20.6−23.3 MPa^1/2 for drugs of SP 20.0−23.4) separates the cellulose polymers that work from those that do not. R9·C43 Second, it probes why: salt strengthens the cellulose polymers but not PVPVA, PVPVA shifts the N−H bands of celecoxib and efavirenz, and the carboxylated esters lose most of their effect when un-ionized; the authors read these as hydrophobic adsorption, hydrogen bonding and an extended adsorbed chain, none of which was observed directly. R10·R12·R13·C48·C51·C59 Third, it shows that inhibition fades with supersaturation and with the drug's intrinsic growth speed, yet the best polymers still slowed celecoxib and ritonavir at their theoretical amorphous solubility. R14·R15·R7·C58

For formulation, the second-generation suberates CA 320S Sub 0.63 and 0.90 stand out as polymers that slowed all three drugs, and the authors propose them as candidates for supersaturating dosage forms. C45 The next steps the paper leaves open are to measure adsorption directly, to test fast-growing drugs at useful supersaturations and other polymer concentrations, and to set the growth results beside the nucleation results of the companion study for the same polymers. X14·X12·R16