Bile salts as crystallization inhibitors: why the choice of bile salt changes how long a supersaturated drug solution lasts

Background and motivation

For a drug with adequate permeability, solubility in the intestine limits absorption, and only a higher concentration of molecularly dissolved drug raises the rate of passive permeation through the membrane; formulations that create supersaturated solutions can therefore improve absorption, provided the drug does not crystallize first (Introduction, refs 1, 7, 8) C6. Supersaturation arises when an amorphous solid dispersion (ASD) dissolves, or when a weakly basic drug passes from the acidic stomach into the intestine and loses its charge; either way the supersaturated solution is unstable and the drug tends to crystallize C7. Earlier work on keeping drugs supersaturated concentrated on polymers such as cellulose derivatives, and small molecules as nucleation inhibitors had received little attention (refs 9-12) C7.

Bile salts are the natural surfactants of the gut. They are steroid derivatives with facial polarity: the hydroxyl groups sit on one face of the ring system and the non-polar groups on the other (ref 13) R38. Sodium taurocholate (STC) is the model bile salt of biorelevant dissolution media such as FaSSIF, where it was chosen to reproduce the solubilizing power of intestinal fluid (refs 14, 66) R38. An earlier study from the same group (Chen 2015, ref 15) had found that STC lengthened the nucleation induction time of 11 structurally diverse compounds and that sodium glycocholate (SGC) and sodium glycodeoxycholate (SGDC) also delayed the nucleation of celecoxib and nevirapine R30. Whether the many other bile salts of human bile behave the same way, and whether one can stand in for another, was not known.

The question matters for prediction. In vitro precipitation tests tend to overestimate the precipitation seen in vivo (Carlert 2010, ref 16) R31. The authors name two possible reasons: the gut dilutes continuously while in vitro tests are static, and the biorelevant medium "might be oversimplified" C8. If bile salts differ strongly as crystallization inhibitors, a medium that contains only STC could misjudge how long a supersaturating dosage form stays supersaturated in the gut.

The paper therefore compares 13 bile salts (six unconjugated, four glycine-conjugated, three taurine-conjugated; Table 1) as nucleation inhibitors for three poorly soluble drugs, celecoxib (COX), nevirapine (NVP) and flibanserin (FLB) M4-M16·M1-M3. Celecoxib was chosen as a compound that is neutral over the physiological pH range, nevirapine and flibanserin as weak bases with pKa values below 6.5 C9. The authors also set out to explore whether the bile salts' solubilizing capacity or their structure explains the differences, and they add molecular dynamics (MD) simulations of nevirapine with four bile salts to look at drug-bile salt interactions at the molecular level C23·C29.

Background concepts

Supersaturation and its apparent measure. A solution is supersaturated when the drug is dissolved at a concentration above its equilibrium crystalline solubility. The paper defines the apparent supersaturation ratio (Methods, equation 1):

Here $C$ is the actual drug concentration in the supersaturated solution and $C_s$ is the equilibrium crystalline solubility of the drug in the presence of the bile salt at 1 mg/mL (0.1% w/v) E1. In plain terms: the added concentration divided by how much crystalline drug the same medium can dissolve. The thermodynamically correct measure is the ratio of activities (Discussion, equation 3):

where $a$ is the solute activity, $\gamma$ the activity coefficient, and the subscript $s^{*}$ marks the saturated solution in equilibrium with the crystal E3. In plain terms: the driving force for crystallization is how much higher the drug's activity is than at saturation; it equals C/Cs only if the activity coefficients of the supersaturated and the saturated solutions are the same, which the authors note is "commonly assumed" C47. A solubilizing additive raises the crystalline solubility and can thereby reduce the supersaturation even when the total drug concentration stays the same (Raina 2015, ref 26) R39. The same reference found that C/Cs "does not always give reasonable estimates" in the presence of surfactants R39. By classical nucleation theory, the nucleation rate depends strongly on supersaturation, so any comparison of inhibitors has to hold supersaturation constant E5·C47.

Measuring activity with a membrane. When a supersaturated solution sits on one side of a membrane and the other side is kept at sink conditions, the diffusive flux follows Fick's first law (Methods, equation 2):

Here $J$ is the flux, $dM/dt$ the mass flow rate, $A$ the membrane area (called S in the text), $D$ the diffusion coefficient of the solute, $a$ its thermodynamic activity in the donor, ${\gamma }_m$ its activity coefficient in the membrane and $h$ the membrane thickness E2. In plain terms: for a fixed membrane and drug, everything on the right except the activity is constant, so the mass flow is directly proportional to the drug's activity in the donor; equal mass flow with and without a bile salt means equal supersaturation C14. (The Discussion refers to this relation as "equation 3"; it is equation 2 X5.)

Induction time. The nucleation induction time is the time between creating supersaturation and the first appearance of crystals. What is measured, $t_{\mathrm{i}\mathrm{n}\mathrm{d}}$, is the time for a critical nucleus to form, $t_n$, plus the time for crystals to grow to a detectable size, $t_g$; when $t_n\gg t_g$ the induction time is dominated by nucleation, an assumption the authors say "appears to hold true" and is standard practice (refs 10, 22-24) E4·C12. A longer induction time therefore means slower nucleation, and an additive that lengthens it is a nucleation inhibitor.

Bile salt structure. All 13 bile salts share the cholane steroid scaffold of Table 1. They differ in two ways. First, in the hydroxyl groups at four positions: R1 (C3) is always 3α-OH; R2 (C6), R3 (C7) and R4 (C12) carry OH or H, giving trihydroxy (cholates), dihydroxy (deoxy-, chenodeoxy-, urso- and hyodeoxycholates) and monohydroxy (lithocholate) members, and the ursodeoxycholates carry the 7-OH in the β orientation M4-M16. Second, in the head group R5 at C24: a carboxylate in the free (unconjugated) bile salts, an amide-linked glycine in the glycine conjugates, and an amide-linked taurine (sulfonate) in the taurine conjugates M4-M16. Taurine conjugates have the lowest pKa and the lowest partition coefficients, glycine conjugates come next, and free bile acids have the highest (ref 59) R37. The paper uses LogP'A, the experimentally determined octanol/water partition coefficient of the ionized bile acid (ref 67), as its measure of hydrophobicity; it runs from 2.65 for sodium deoxycholate (SDC) to -0.50 for STC and is not available for SLC, STDC and STCDC M4-M16.

Aggregation and the CMC. Bile salts aggregate stepwise. In the widely accepted model of Small (ref 50), primary aggregates of about 2-10 molecules form at low concentration through the hydrophobic convex β-faces; they offer hydrophobic binding sites that solubilize drugs, and at higher concentration they join into secondary aggregates through hydrogen bonds C44·R34. The critical micelle concentration (CMC) depends strongly on the method used; trihydroxy bile salts such as STC typically have CMCs of 10-15 mM, the dihydroxy ones normally below 5 mM, and most reported values exceed 2 mM R34. The literature ranges compiled in Table 1 run from 0.5-0.9 mM (SLC) to 7-19 mM (SUDC) M4-M16.

Liquid-liquid phase separation (LLPS). Above a certain concentration, a supersaturated drug solution separates into a drug-rich liquid phase and a drug-lean aqueous phase instead of crystallizing directly. The paper measures this LLPS concentration in pH 6.5 buffer by the onset of light scattering during slow addition of drug stock (ref 25): 42.8 ± 4.2 µg/mL for celecoxib, 874 ± 106 µg/mL for nevirapine and 57.3 ± 2.1 µg/mL for flibanserin M1·M2·M3·P6. The induction-time experiments were run below these values so that crystals, not a liquid phase, would form C13.

Methods in detail

The study has three parts: a survey of induction times for 13 bile salts with three drugs; a set of checks on whether the bile salts act simply by solubilizing the drug; and MD simulations that look for molecular differences between strong and weak inhibitors. The flow below summarizes it.

Contribution 1: An induction-time survey of 13 bile salts

Problem. To compare inhibitors, the drug concentration, the inhibitor concentration and the detection method have to be the same across systems, and the drug must be supersaturated enough to crystallize within hours but not so much that it separates into a liquid phase first.

What was done. Each bile salt was dissolved at 1 mg/mL (0.1% w/v) in 50 mM buffer at 37 °C P4. pH 6.5, chosen to represent the duodenum (ref 20), was used wherever possible; bile acids that would not dissolve were run at pH 6.8 (sodium chenodeoxycholate, SCDC) or in pH 10 carbonate buffer (sodium lithocholate SLC, sodium ursodeoxycholate SUDC, sodium hyodeoxycholate SHDC and sodium glycoursodeoxycholate SGUDC) C11·M4-M16. The drug was added from a concentrated stock: celecoxib from methanol to 20 µg/mL, nevirapine from DMSO to 800 µg/mL and flibanserin from methanol to 45 µg/mL, each into 50 mL of stirred buffer (300 rpm) P4. A fiber-optic UV dip probe recorded spectra from 200 to 450 nm every 10-60 s, following both an absorbing wavelength (celecoxib 259 nm, nevirapine 334 nm, flibanserin 280 nm), which falls as crystals grow and remove drug from solution, and a non-absorbing wavelength (350, 400 and 350 nm), where scattering from crystals raises the signal P4. For celecoxib the two signals gave the same onset R5. The induction time was taken as the intersection of the regression lines through the two linear parts of the absorbance trace (celecoxib, flibanserin) or of the extinction trace (nevirapine, whose high concentration interfered with the absorbance) P4. Every measurement was made in triplicate P4. In a separate experiment, the six most prevalent human bile salts were blended in human-bile proportions (60 mg SGC, 30 mg SGDC, 60 mg sodium glycochenodeoxycholate SGCDC, 20 mg STC, 20 mg sodium taurodeoxycholate STDC and 10 mg sodium taurochenodeoxycholate STCDC) and dissolved to 0.1% total for a celecoxib test P5.

Why it works. The drug levels were set below the LLPS concentrations and above the crystalline solubilities, so the measured event is crystal nucleation from a single liquid phase C13. Because the same probe, stirring, temperature and drug level were used for every bile salt, the induction times can be compared directly within each drug. Note by the AI writer (not discussed in the paper): the cosolvent carried in with the stock differs between drugs, about 2% v/v DMSO for nevirapine against about 0.2-0.3% methanol for celecoxib and flibanserin, so comparisons across drugs include this difference X20.

Contribution 2: Checking that the delay is not just solubilization

Problem. Bile salts are surfactants. If they raise the crystalline solubility of the drug, the supersaturation falls at the same added concentration, and nucleation slows for that reason alone; a longer induction time would then say nothing about inhibition C48.

What was done. Four checks were made. (1) The crystalline solubility of each drug was measured in each bile salt medium by shaking excess crystals at 37 °C for 48 h (1 h for flibanserin, to limit degradation), ultracentrifuging and assaying by HPLC; the solids left at the end were checked by X-ray powder diffraction and DSC and were celecoxib form III, nevirapine anhydrate and flibanserin form A, with or without STC P1·P2·R2. From these values C/Cs follows by equation 1 E1. (2) The drug concentration was varied with and without 0.1% STC (celecoxib 10-40 µg/mL, nevirapine 400-1200 µg/mL), and the STC concentration was varied from 0.05 to 0.6% at fixed drug, with the matching solubilities (Table 5) A4·A5·R11. (3) The CMC of STC and SGC in the presence of celecoxib was found from celecoxib solubility against bile salt concentration, as the intersection of two straight lines P7. (4) In a side-by-side diffusion cell with a regenerated cellulose membrane (6-8 kDa cutoff, 30 mm orifice), both chambers held pH 6.5 buffer with or without 0.1% STC or SGC; the donor held 20 µg/mL celecoxib or 800 µg/mL nevirapine, and the receiver concentration was followed by a UV probe to give the mass flow P8. Runs without bile salt were repeated with 5 µg/mL HPMCAS-MF to confirm that no crystallization occurred during the measurement; the mass flow did not change R15.

Why it works. Checks 1 and 2 test the solubility route directly; check 4 measures the activity of the drug and therefore the true supersaturation, without having to assume that C/Cs is correct (equation 2) C14. How it differs from earlier work. The earlier STC study (ref 15) compared induction times; here the diffusion-cell method of Raina et al. (ref 26) adds an independent, activity-based check of the driving force R39.

Contribution 3: Molecular dynamics of nevirapine with four bile salts

Problem. If solubilization does not explain the differences, the bile salts must interact differently with the drug or its nuclei. The authors ask whether MD of small systems can show which interactions separate a strong from a weak inhibitor; they state that the aim was "not to simulate the nucleation process" C29.

What was done. Nevirapine was paired with four free bile salts chosen as two contrasts: SHDC (a strong nevirapine inhibitor) against sodium cholate SC (a weak one), and SDC (strong) against SUDC (weak) C30. Crystal structures of nevirapine and SC were taken from the Cambridge Structural Database and SC was modified to the other bile salts; geometries were optimized by density functional theory (PBE0/6-311++G**) and parameterized with SwissParam for the CHARMM force field in GROMACS P9. Dimers (one drug, one bile salt, 2,100 SPC/E water molecules, one Na⁺, 4.0 nm box) and, for SDC and SUDC, tetramers (two of each, 7,020 waters, two Na⁺, 6.0 nm box) were run for 100 ns at 310 K, and 2000 snapshots were analysed P9. The analysis counted hydrogen bonds (3.0 Å and 20° cut-offs), clusters (molecules within 3.0 Å), the minimum drug-bile salt distance and the interaction energy, averaged over the trajectory rather than taken for the single most stable conformation P10·C29.

How it differs from earlier work. Interaction energies of drug complexes are usually reported for the most stable conformation (refs 42-45); here they are averages over the many conformations the pair adopts in water, and the propensity to stay in contact is analysed alongside C29.

Experimental analysis

Setup in brief. All induction times were measured at 37 °C with 0.1% (w/v) bile salt, in triplicate, against a drug-only control in pH 6.5 buffer P4·R6. The drug-only control exists only at pH 6.5, although four bile salts were tested at pH 10 and one at pH 6.8 A3·X12. Table A collects the induction times of Table 4 together with each bile salt's class, medium pH and LogP'A from Table 1.

Table A. Nucleation induction times with 0.1% (w/v) bile salt at 37 °C (from Table 4; drug at 20, 800 and 45 µg/mL). "Ionized" marks the footnote "ionization led to greater extent of solubilization"; "particles" marks "insoluble compounds were formed". Class, medium pH and LogP'A from Table 1 R6·R4·M4-M16.

Main result 1: most bile salts delay nucleation, by very different amounts. Without bile salt all three drugs crystallized within minutes: 5 ± 2 min for celecoxib, 2 ± 1 min for nevirapine and 5.4 ± 0.4 min for flibanserin R6. With 0.1% bile salt most induction times rose, and the spread is wide: for celecoxib from 3 ± 1 min (STDC) and 3.8 ± 0.4 min (SGDC), no longer than the control, to more than 16 h (SDC, STCDC, and at pH 10 SHDC and SGUDC) R6. The authors write that "it is apparent that most bile salts delayed nucleation", list SGDC and STDC (celecoxib), SLC (nevirapine) and STDC (flibanserin) as exceptions, and in their conclusions describe the effect as inhibition "to some extent" C2. Some systems formed insoluble particles at once: flibanserin with SDC, SCDC and SLC, and nevirapine with SLC, so no induction time could be read R4. For flibanserin the authors suggest the partly protonated drug "might" form insoluble salts with the anionic bile salts; the particles were not characterized C20. Four celecoxib systems and two points in Figure 2 ran past the end of the 16 h observation window, so a long delay and complete inhibition cannot be told apart there X23.

Main result 2: the best bile salt depends on the drug. The authors rank the most effective inhibitors as SDC and STCDC, then SC and SGC, for celecoxib; SHDC and SDC, then STCDC, then SCDC and SGC, for nevirapine; and STCDC, SHDC, SGUDC and SGC for flibanserin C21. These rankings leave out the systems with ionization or particles; SHDC and SGUDC (more than 16 h at pH 10) and SLC (364 min) would otherwise rank high for celecoxib, and SLC (623 min) for flibanserin X9. The clearest example of drug dependence is SC: 281 ± 40 min with celecoxib but only 15.6 ± 0.6 and 26 ± 12 min with nevirapine and flibanserin R6. STCDC is the one bile salt among the strongest for all three drugs R6. The authors conclude that the inhibitory effects "appear to vary greatly depending on the structure of both the bile salt and the drug" C3.

Table B. Crystalline solubility and membrane mass flow in the diffusion-cell media, donor 20 µg/mL celecoxib or 800 µg/mL nevirapine (from Table 6) R15.

Main result 3: the delay is not a solubility effect. At 0.1% the bile salts raised crystalline solubility only a little: celecoxib from 1.2 ± 0.1 µg/mL in buffer to 2.7-3.5 µg/mL at pH 6.5-6.8, and flibanserin from 9.4 ± 1.9 to at most 15.2 ± 4.2 µg/mL, while nevirapine stayed between 91.8 and 112.5 µg/mL against 105.3 ± 5.1 µg/mL in buffer (Table 3) R1. The authors call this "slightly enhanced" and small compared with the solubilization seen above the CMC C17·C49. The higher celecoxib solubility at pH 10 (7.7-11.1 µg/mL) "could be due to" the roughly 7.5% of celecoxib that is ionized at that pH R1·R3·C18. Nevirapine is the cleanest test: its solubility is essentially unchanged, so its C/Cs stays between about 7.1 and 8.7 across all bile salts while its induction time runs from 0 to 577 min (Figure 5A) R13. The system with the highest C/Cs, SHDC, even has the longest induction time; the authors conclude that the induction times "do not correlate" with S' R13·C50. The diffusion cell settles the point for STC and SGC: the mass flow of celecoxib (0.41-0.46 µg/min) and nevirapine (20.9-23.4 µg/min) is the same within error with or without the bile salt (Table B) R15. Since mass flow measures the drug's activity, the supersaturation was the same, and the authors write "we can confirm" that induction times were considerably extended at the same driving force, which "suggest[s]" that solubilization "can be eliminated as the mechanism" at 0.1% bile salt C51.

Two qualifications belong here. Mass flow was measured for STC and SGC only; for the other eleven bile salts the solubilization argument rests on C/Cs X17. And for celecoxib, C/Cs and mass flow disagree: HPMCAS, STC and SGC raise celecoxib solubility from 1.2 to 2.4-2.9 µg/mL, which would lower C/Cs two- to 2.4-fold, yet the mass flow does not move (Table B) R15·X18. Note by the AI writer (not discussed in the paper): for celecoxib in these media, C/Cs therefore understates the supersaturation, which is the case Raina et al. warned about X18·R39. The solubility values themselves also differ between experiments: celecoxib with SGC gave 3.5 ± 0.6 µg/mL in Table 3 but 2.4 ± 0.2 µg/mL in Table 6 X7.

Main result 4: more STC lengthens induction times, more drug shortens them. At 0.1% STC, raising the drug concentration shortened the induction time of both drugs, but STC kept it longer than the control at every concentration (Figures 2A, 3A) R7·R9·C24. For celecoxib alone the induction time fell from about 18 min at 10 µg/mL to about 2 min at 40 µg/mL; with STC it fell from more than 960 min to about 16 min R7. Nevirapine with STC fell from about 500 min at 400 µg/mL to about 2.7 min at 1200 µg/mL; note that 1200 µg/mL lies above nevirapine's LLPS concentration of 874 ± 106 µg/mL R9·X21. Table C shows the effect of the STC concentration itself.

Table C. Induction time against STC concentration (read from Figures 2B and 3B, approximate; celecoxib 20 µg/mL, nevirapine 800 µg/mL) and equilibrium solubility at the same STC levels (from Table 5) R8·R10·R11.

For celecoxib the induction time stays at about 94-105 min between 0.1 and 0.3% STC and rises steeply only from 0.4%, reaching more than 960 min at 0.6%, while celecoxib solubility rises from 2.4 to 7.1 µg/mL over the series R8·R11·X15. The authors attribute the celecoxib delay to "both solubility enhancement and increased nucleation inhibitor concentration" C25. For nevirapine, solubility does not change up to 0.2% STC, so solubilization "was not responsible" there; the induction time rises to about 110 min at 0.4% STC, drops to about 50 min at 0.5% and rises again to about 68 min, which the authors relate to STC's stepwise aggregation ("might lead to variations") R10·R11·C26·C27. (The text cites this as "Figure 3C"; the data are in Figure 3B X2.) Table 5 also lists apparent supersaturation ratios; the nevirapine ratios follow from 800 µg/mL divided by the solubility, but two celecoxib ratios do not follow from 20 µg/mL divided by the tabulated solubility (13.6 printed against 16.7 at 0% STC, 4.3 against 4.9 at 0.5%) E1·X1.

The CMCs of STC and SGC in the presence of celecoxib, found from the break in celecoxib solubility (Figure 4), were 0.48% and 0.46% (w/v), well above the 0.1% used in the survey R12·C28. Below the CMC, celecoxib solubility still rose slightly over the bile-salt-free value C28. Because nevirapine solubility does not change at 0.1%, the authors infer that all CMCs in the presence of nevirapine lie above 0.1% C45. With most reported CMCs above 2 mM and 0.1% STC equal to 1.86 mM, they consider most bile salt in the test solutions to be present as monomers or low-order aggregates C46·R29. Note by the AI writer (not discussed in the paper): 0.1% corresponds to 1.86-2.51 mM across the bile salts, and Table 1 lists literature CMCs at or below this range for SLC, STCDC, SGCDC, STDC and SGDC, so the monomer picture is least certain for these X16.

Main result 5: hydrophobicity orders free bile salts one way and glycine conjugates the other. Figure 6 arranges the induction times by decreasing LogP'A, leaving out the systems with ionization or particles R16. For all three drugs the authors see the inhibition of free bile salts "generally" increasing with LogP'A, and for the glycine conjugates the induction time rising as LogP'A falls; with only STC's LogP'A known, no trend can be judged for the taurine conjugates C53. Nevirapine shows it best: SDC (2.65) 523 min, SHDC (2.28) 577 min, SCDC (2.25) 186 min, SUDC (2.20) 19 min and SC (1.1) 15.6 min among the free bile salts, and SGDC (0.80) 21 min, SGCDC (0.45) 28 min, SGUDC (0.20) 61 min and SGC (-0.40) 158 min among the glycine conjugates R6·M4-M16. There are exceptions in each drug: for celecoxib SC (1.1) outlasts SCDC (2.25), 281 against 103 min; for nevirapine SUDC and SCDC differ almost ten-fold at nearly the same LogP'A; for flibanserin the glycine order fails (SGDC 15.9 min against SGCDC 11 min) X10. As a group, the free bile salts "tend to be better" than the more hydrophilic conjugates, in the authors' words, but the taurine conjugate STCDC is among the strongest inhibitors for every drug C52·X11. The abstract condenses this into "the R5 group and hydrophobicity of the bile salt appeared to be essential" C4. Note by the AI writer (not discussed in the paper): for one steroid, the order free, glycine, taurine is not the same across drugs (cholates: celecoxib 281, 124, 103 min; nevirapine 15.6, 158, 30 min), so the head group matters but not in a fixed direction X25.

Table D. MD of nevirapine with strong (SHDC, SDC) and weak (SC, SUDC) inhibitors, 100 ns at 310 K (text values; SDC and SUDC energies read from Figures 9A-B, approximate; strong and weak as assigned by the authors from Table 4) R17·R18·R20·R22·R23·R24·R25·R26.

Main result 6: in the simulations, contact and hydrogen bonds separate SHDC from SC, but energy alone does not rank bile salts. In the dimer simulations, nevirapine and SHDC stayed in close contact, most often 0.2-0.3 nm apart, while nevirapine and SC spent about a quarter of the trajectory 1.0-2.5 nm apart (Figure 7A) R17·C31. Hydrogen bonds formed between the hydroxyl or carboxyl groups of the bile salt and the nitrogen or carbonyl group of nevirapine; SHDC formed 192 over the trajectory, SC 10 (Figure 7B) R18·R21. The authors read this as a "drastic impact" of the number and position of hydroxyl groups C32. The NVP-SHDC energies spread over -80 to -20 kJ/mol and averaged around -60 kJ/mol, against -40 to -20 and around -30 kJ/mol for NVP-SC (Figures 7C, 7D) R19·R20. In the authors' reading, persistent van der Waals and hydrogen-bonded contact with SHDC, against sporadic mainly van der Waals contact with SC, "explain" the energies and give "a perspective" on the inhibition C34. The second pair broke this pattern: SDC (strong) and SUDC (weak) gave similar dimer energies, about -41.5 and -47.5 kJ/mol (Figure 9A) R22·C35. With two drug and two bile salt molecules, SUDC bound even more strongly (about -110 against -70 kJ/mol), formed a single four-molecule cluster in 86% of snapshots, and paired with itself 73% of the time against 5% for SDC, while nevirapine paired with itself about half the time in both systems (Figures 9B-E) R23·R24·R25·R26. The authors conclude that "a more favorable interaction energy is not a general indication of a better bile salt system" and that dimer energies alone are "insufficient to explain" the ranking; distances, hydrogen bonds, cluster sizes and larger systems all need to be considered C37·C39.

Main result 7: the main human bile salts beat STC. Glycine conjugates dominate human intestinal fluid, and SGC makes up about 30% of human bile (refs 61-65) R35. Among the six most prevalent human bile salts, STCDC gave the longest induction times for all three drugs and SGC appeared to be second (Figure 5B) C62·R14. SGC outlasted STC for all three drugs in the authors' words: clearly for nevirapine (158 ± 17 against 30 ± 5 min) and flibanserin (56 ± 15 against 35 ± 7 min), but within the spread for celecoxib (124 ± 54 against 103 ± 35 min) C68·R6·X13. The blend of the six bile salts in human-bile proportions gave a celecoxib induction time of 210 ± 21 min, about twice the value with STC alone R28. The authors read this as a sign that "crystallization kinetics in vivo may not be adequately predicted by experiments in commonly used simulated media", while noting that their blend is still simplified and that lecithin "may further alter" solubility and kinetics C63·C64. The 0.1% used in the survey equals 1.86 mM STC, below the 3 mM of FaSSIF and inside the fasted human jejunal (1.52-5.30 mM) and duodenal (1.40-5.90 mM) ranges of Table 7 R29·X19.

Discussion

How the authors explain the inhibition. The authors' account runs as follows. At 0.1% most bile salts are below their CMC, present as monomers or small aggregates, and barely change the drug's solubility C46·C49. The membrane test shows that the supersaturation is unchanged, so the longer induction times reflect slower nucleation, not a lower driving force C51. The bile salts contact the drug through van der Waals forces and hydrogen bonds, to an extent that depends on their chemistry C5. Drawing on simulations by Anwar and colleagues (ref 60), in which additives with high affinity for the solute entered and disrupted the nucleus while amphiphilic additives only delayed nucleation at its surface, the authors propose that bile salts, being surfactant-like, "are likely to interact with the surface of the emerging nucleus, delaying rather than completely inhibiting nucleation" R36·C58. Which bile salts do this well is set by their structure: for unconjugated bile salts, higher hydrophobicity "may lead to enhanced interactions with the drug or drug clusters which delay nucleation"; for glycine conjugates the reversed trend "may indicate" that hydrogen bonding with the glycine head group matters; and the self-aggregation of SUDC "may indicate a lower tendency to interact with a nevirapine nucleus" C55·C57·C60.

Note by the AI writer (not discussed in the paper): each arrow in this chain beyond step 2 is an interpretation. The paper measures induction times, solubilities, mass flow and simulated contacts; it does not measure bile salt binding to drug clusters or nuclei, the aggregation state of each bile salt in the test solutions, or hydrogen bonding of the glycine head group with the drugs C55·C57·C58·C60. The MD covers nevirapine only, with one 100 ns trajectory per system X24.

Where the results apply. The comparison holds for three fast-crystallizing drugs at one supersaturation each, 37 °C, a vigorously stirred 50 mL vessel and one bile salt level of 0.1% (w/v); medium pH was 6.5 for most systems but 6.8 or 10 where the bile acid required it P4·A3. At pH 10 celecoxib is partly ionized, and the authors themselves mark those values as affected R3·C18. No lecithin, no mixed micelles and no ionic strength or dilution changes were studied C64.

What the authors draw for biorelevant testing. Because many bile salts retard nucleation, "the presence of bile salts in the human GI tract could delay the crystallization of supersaturated solutions of poorly soluble drug candidates" C71. The effect of bile salts on the duration of supersaturation "may be a key factor" in bioavailability, "especially" for BCS class 4 compounds with their longer transit times C41. Given the concentration dependence seen with STC, inhibition "may be more pronounced in vivo", where hydrodynamics differ from a stirred vessel, although the authors add that the rise with STC "may be due, in part, to solubilization" C67·C66. Their recommendation: "attention should be paid to bile salt selection for biorelevant dissolution media" when supersaturating dosage forms are evaluated, because bile salts "are not interchangeable" in their crystallization inhibition C69. More broadly, small-molecule crystallization inhibitors deserve evaluation for supersaturating dosage forms C72.

Note by the AI writer (not discussed in the paper): for a formulation team, the practical reading is that a precipitation test in an STC-only medium can misjudge in vivo crystallization in either direction. In Table A, the six human bile salts alone span 3 min to more than 16 h for celecoxib, and the blend gave twice the STC value; a quick check with SGC or with a glycine-taurine blend, next to the standard medium, would show how sensitive a given drug is R6·R28. The insoluble particles formed by flibanserin with some free bile salts point to a second, separate risk for basic drugs that a single-bile-salt medium would not reveal R4.

Open questions. The authors ask what the insoluble particles are, how lecithin would change the picture, and how molecular interactions affect crystallization; they write that "clearly more sophisticated modeling of the nucleation process is required" C20·C64·C22·C61. Further questions left by the data: whether solubilization is also excluded, by flux rather than C/Cs, for the bile salts other than STC and SGC; whether the hydrophobicity trends survive when all bile salts are compared at one pH; and whether the bile salt effects seen in buffer carry over to human intestinal fluid X17·X12.

Limitations

Limitations stated by the authors. The MD was not meant to simulate nucleation, and the authors call the interactions it shows complex, "making it challenging to deconvolute the molecular origin of the nucleation inhibition effects" C29·C70. They state that more sophisticated modelling of nucleation is needed C61. Their six-salt blend is "a simplified one", and lecithin "may further alter both solubility and crystallization kinetics" C64. The rise of induction time with STC concentration may be due in part to solubilization C66. Further investigation of the insoluble species is needed C20. CMC values of bile salts depend strongly on the method and conditions used (ref 55) R34.

The AI writer's own judgement (1): what the structure-activity trends can carry. The large differences between bile salts are solid: they are replicated, measured with one method, and span two orders of magnitude R6. The interpretations built on them are weaker. The LogP'A trends hold "generally" but have exceptions in every drug, the values often overlap within their standard deviations, and no correlation statistic is given X10. Bile salt class is confounded with medium pH, since four of the six free bile salts were tested at pH 6.8 or 10 and the only drug-only control is at pH 6.5 X12. The head-group effect is large but changes direction between steroids and drugs X25. And the rankings in the text quietly drop the ionized and particle-forming systems X9. The trends are therefore useful as hypotheses, not as rules for choosing a bile salt.

The AI writer's own judgement (2): what the supersaturation checks cover. The flux experiment is the strongest piece of the solubilization argument, but it was done for STC and SGC only X17. For celecoxib it also shows that C/Cs from equation 1 understates the supersaturation in bile salt and HPMCAS media, so the C/Cs values computed for celecoxib cannot be compared across media X18. Two printed celecoxib ratios in Table 5 do not follow from the stated concentrations X1. The claim that most bile salts act as monomers is least certain for SLC and the dihydroxy conjugates, whose literature CMCs lie near or below the 1.86-2.51 mM used X16. The MD rests on one trajectory per system and one drug, and one of its energy histograms does not sum to the stated number of frames X24·X22. Several editorial slips (a cited Figure 3C that does not exist, mass flow units of ng/s in the methods against µg/min in Table 6, the medium pH of SUDC given as 10 in Table 1 but 8 in SI Table A) do not change the conclusions X2·X5·X3.

Conclusion

The paper shows, with one consistent induction-time method, that bile salts at physiologically relevant concentrations are potent but highly variable nucleation inhibitors: at 0.1% they lengthened induction times from minutes to more than 16 h, depending on both the bile salt and the drug R6·C2·C3. For STC and SGC the delay occurs at unchanged supersaturation, as shown by membrane mass flow, so it is inhibition of nucleation, not solubilization R15·C51. Free bile salts tend to inhibit more as their hydrophobicity rises and glycine conjugates the reverse, with exceptions C53·X10. MD with nevirapine shows that close, hydrogen-bonded contact separates a strong from a weak inhibitor in one pair, while in another pair the weak inhibitor binds as strongly but self-associates, so interaction energy alone does not rank bile salts C39·C60. A blend of the main human bile salts doubled the celecoxib induction time of STC, and the authors conclude that bile salts in biorelevant media are not interchangeable R28·C63·C69.