How phospholipid digests and personal bile salt levels change the crystallization of supersaturated atazanavir

Background and motivation

Many new drug candidates are poorly water-soluble, and supersaturating formulations are one way to raise their absorption. Unlike solubilization in micelles or cyclodextrins, supersaturation raises the concentration of free, molecularly dissolved drug, and the rate of transport across the gut wall follows the drug's thermodynamic activity, that is, the free concentration rather than the total (Miller 2012, ref 1) C1. Supersaturation arises from amorphous solids, from the digestion of lipid formulations, from salts and cocrystals, and, for weak bases, from the pH change on passing from the acidic stomach into the more neutral small intestine (refs 2–8) C1·C2. Atazanavir, a weakly basic antiretroviral drug used against HIV, is such a case: it dissolves well in the stomach, loses that advantage in the upper small intestine and may precipitate there, but for a while it is supersaturated C2. A supersaturating system only works while nucleation is held off long enough for absorption C2.

Excipients are known to change crystallization. Much less is known about the endogenous components of intestinal fluid, and there has long been a need for physiologically relevant media to test supersaturating formulations (Brouwers 2009, ref 9) C2. The standard fasted medium, FaSSIF, contains a single bile salt, sodium taurocholate (TCA, chosen for its low pKa), together with undigested phosphatidylcholine (PC) (Galia 1998, ref 11) C3. Crystalline solubility in FaSSIF correlates with that in fasted human intestinal fluid (FaHIF) with r² = 0.85 (Augustijns 2014, ref 12), and the newer FaSSIF-V3 adds sodium glycocholate (GCA), PC digests and cholesterol, chosen from surface tension measurements and the solubility of nine drugs (Fuchs 2015, ref 13) C3·R33. Taken together, these "indicate" that current media predict crystalline solubility well C3.

For a supersaturating formulation, however, two other quantities matter: the amorphous solubility and the time before crystals form. Indulkar et al. had shown that both the PC concentration and the bile salt composition change how fast drugs crystallize from supersaturated solution, and that induction times differ between FaSSIF and Composite-SIF, a medium with six bile salts modelled on the average fluid of healthy Western volunteers (Indulkar 2018, ref 15) C5. Individual bile salts differ as crystallization inhibitors (Li 2016, ref 24; Lu 2017, ref 25) C5.

Two physiological facts motivated this paper. First, PC is the main phospholipid in bile, but in the intestine pancreatic phospholipase A₂ hydrolyses it at the sn-2 position into lyso-PC and a free fatty acid such as sodium oleate, so lyso-PC is the main phospholipid of intestinal fluid (refs 7, 13, 18–20) C6. Digestion of formulation lipids was known to reduce solubilization and trigger supersaturation, but the effect of endogenous and dietary lipid digests on the stability of supersaturated solutions had not been studied C7. Second, each person's bile acid pool is different: bile acids are made in the liver and transformed by gut bacteria, and the size of the pool depends on microbial bile acid metabolism (Enright 2018, ref 26; refs 27, 28) C9. Cholesterol, solubilized by PC/bile salt mixed micelles and present in FaSSIF-V3, can also change crystalline solubility, and its level may vary with gender, disease or diet (refs 22, 23) C8.

The paper therefore set three aims: compare PC with its digests lyso-PC and sodium oleate; test increasing cholesterol; and test the bile acid pools of individual volunteers, all for the solubility and crystallization of atazanavir and with the goal of better biorelevant media for supersaturating formulations A2·A3·A4·A5·A7·A8·A9·A10.

Background concepts

Crystalline and amorphous solubility. The crystalline solubility is the equilibrium concentration over the crystal. The amorphous solubility is higher: it is the concentration at which the solution is in metastable equilibrium with water-saturated amorphous drug (Taylor 2016, ref 14) C4. If more drug is added, a drug-rich liquid phase separates as small colloidal droplets; this liquid–liquid phase separation (LLPS) can be seen as a rise in light scattering. The authors use the onset of this scattering, measured as UV extinction at a non-absorbing 450 nm, as the amorphous solubility in each medium C4·P13. A solution prepared exactly at the amorphous solubility is single-phase and maximally supersaturated, and the authors take it to represent "an equivalent extent of supersaturation" in every medium C10.

Nucleation-induction time. The main read-out is the time from creating the supersaturated solution to the first detectable crystals, followed as light scattering at 350 nm P14. The authors state that the measured time contains two parts:

In plain terms: the reported crystallization-induction time $t_{\mathrm{i}\mathrm{n}\mathrm{d}}$ is the true nucleation-induction time $t_n$ plus the time $t_g$ that crystals need to grow to a size the probe can detect; the two were not measured separately E1. Throughout the paper the reported time is called the nucleation-induction time (NIT).

Bile salts: trihydroxy and dihydroxy. The bile salts used are conjugates of three bile acids. Taurocholate (TCA) and glycocholate (GCA) carry three hydroxyl groups (trihydroxy); taurochenodeoxycholate (TCDCA), glycochenodeoxycholate (GCDCA), taurodeoxycholate (TDCA) and glycodeoxycholate (GDCA) carry two (dihydroxy) M7·M8·M9·M10·M11·M12. Chenodeoxycholate (CDCA) is made in the liver, whereas deoxycholate (DCA) is formed by gut bacteria, which remove a hydroxyl group from cholate with the enzyme 7α-dehydroxylase C51·C57. FaSSIF and FaSSIF-V3 contain trihydroxy bile salts only; Composite-SIF contains both kinds P2.

Mixed micelles and the bile salt to phospholipid ratio. Bile salts form mixed micelles with phospholipids, their digests and cholesterol. How much bile salt is present relative to phospholipid is expressed as the molar ratio

In plain terms: the total bile salt concentration $c_{\mathrm{B}\mathrm{S}}$ divided by the total phospholipid concentration, counting lyso-PC and PC (used at 9:1, as in FaSSIF-V3) E2. With the FaSSIF-V3 lipid fraction (350 µM phospholipid) the six volunteers' bile salt levels give ratios from 1.6 to 14.7; in a second design the lipids were scaled so that the ratio stayed at 8.6 P6·P7·E3.

PC and its digests. PC carries two fatty acid chains; lyso-PC carries one, and the released chain appears as a free fatty acid such as oleate (Figure 1) C6. Lyso-PC has a higher surface activity and a lower critical micelle concentration than PC, and oleate also lowers surface tension (Fuchs 2014, ref 18) R37. These surface properties are central to how the authors explain their results.

Methods in detail

The study has one drug and one logic: bring atazanavir to its amorphous solubility in a medium, record how long it takes to crystallize, and change one component of the medium at a time. All work was at 37 °C in the same pH 6.5 phosphate buffer (FaSSIF buffer: 0.35 g sodium hydroxide, 3.44 g sodium dihydrogen phosphate and 6.19 g sodium chloride per litre) P1·C10. The authors used this buffer also for FaSSIF-V3, replacing its original maleate buffer, whose strong UV absorbance would interfere with the UV-based measurements P1.

The media

Three published media served as the starting point: FaSSIF, Composite-SIF and FaSSIF-V3 (Table 1) P2. FaSSIF was made from the commercial powder. For Composite-SIF the six bile salts were dissolved in buffer, and 0.75 mM PC, first dissolved in dichloromethane, was added as a turbid emulsion; stirring at 600 rpm and 50–60 °C for 40 min drove off the solvent until the medium was clear and without solvent odour P2. FaSSIF-V3 was prepared after Fuchs et al.: GCA, TCA and sodium oleate dissolved in buffer at 37 °C, then PC, lyso-PC and cholesterol added from dichloromethane and the solvent removed in the same way P2. The modified media, FaSSIFMod and Composite-SIFMod, replace the PC of FaSSIF and Composite-SIF with the FaSSIF-V3 lipid fraction; SIF-2BAMod uses only two bile salts with the same lipids (Table 2) P2.

Table A. Composition of the published and modified media, all in mM (from Tables 1 and 2) P2.

SIF-2BAMod lumps the six bile salts into two: TCA stands for all trihydroxy (TCA + GCA) and GCDCA for all dihydroxy bile salts of average fasted human fluid (Table 3) E4. Note by the AI writer (not discussed in the paper): the printed 1.89 and 1.11 mM follow from the average composition only if its six percentages, which add up to 98.93%, are first rescaled to 100% E4·X5.

The investigational media isolate one lipid at a time in 2.8 mM GCA/TCA (1:1), the bile salt pair of FaSSIF-V3: PC or lyso-PC at 0.01, 0.02, 0.04, 0.08, 0.16, 0.32 and 0.63 mM; sodium oleate at 0.08–0.32 mM on top of 0.32 mM lyso-PC (the lyso-PC was needed to measure the amorphous solubility); and cholesterol at 0.1–0.3 mM with 0.04 mM PC, 0.32 mM lyso-PC and 0.32 mM sodium oleate P3. A fourth set mimics PC digestion directly: 0.08 mM PC against 0.16 mM lyso-PC + oleate (1:1), and 0.32 mM PC against 0.63 mM lyso-PC + oleate, the amount in FaSSIF-V3 P4.

The volunteer media

The volunteer media copy the duodenal bile acid data of six healthy Western volunteers reported by Perez de la Cruz Moreno et al. (ref 30) P5. Their total bile salt ran from 570 to 5137 µM and their composition differed widely; V4, for example, had only 13.01% TCA and 37.18% GCDCA, while V5 had 59.59% TCA (Table 3) P5. Four designs were made from these data P5·P6·P7·P8:

Table B. The volunteer pools, ordered by bile salt level (from Tables 3, 4 and 5). With the fixed lipid fraction every medium holds 350 µM phospholipid, 315 µM sodium oleate and 200 µM cholesterol; at BS:PL 8.6 the lipids scale with bile salt and cholesterol stays at 200 µM P5·P6·P7·E2·E3.

• Bile acids only: each volunteer's size and composition in buffer P5.
• Fixed lipid fraction: each pool plus the FaSSIF-V3 lipids, so the BS:PL ratio follows the bile salt level P6.
• Constant ratio: the average composition at each volunteer's total, with lipids scaled to BS:PL 8.6, because the volunteers' own lipid levels were unknown P7.
• Composition only: each volunteer's composition at a fixed 3 mM total with the FaSSIF-V3 lipids P8.

Note by the AI writer (not discussed in the paper): the paper does not say where the ratio 8.6 comes from; it equals 3.0 mM bile salt over 0.35 mM phospholipid, the ratio of the Mod media, not the 8.7 of the average volunteer or the 8.0 of FaSSIF-V3 X6.

The measurements

Crystalline solubility. Excess crystalline atazanavir was stirred at 300 rpm in the medium for 48 h at 37 °C; the suspension was centrifuged twice at 14000 rpm for 30 min and the supernatant analysed by HPLC with UV detection at 210 nm (C18 column, 60/40 v/v acidified water/acetonitrile, calibration 0.16–2.5 µg/mL, r² > 0.99) P10·P12. In the HPMC media the drug was equilibrated in a shaking water bath and filtered through glass fibre filters instead P11.

Amorphous solubility. A 10 mg/mL methanolic stock was pumped into 15 mL medium at a rate that kept the experiment under 10 min, with stirring at 300 rpm at 37 °C, while a fibre-optic UV probe recorded extinction at 450 nm every 10–15 s. A calibration with standards was made before each run (less than 150 µL stock in total). The concentration at which extinction rose marked the formation of a drug-rich colloidal phase, taken as the amorphous solubility P13·C4.

Nucleation-induction time. The same stock was added to 15 mL medium to give exactly the amorphous solubility, a single-phase, maximally supersaturated solution, which was then held at 37 °C and 300 rpm while scattering at 350 nm was followed until crystals appeared P14. Results are mean ± SD of three or more runs, compared by unpaired t-tests or one-way ANOVA with Tukey's test P15.

Note by the AI writer (not discussed in the paper): "equivalent supersaturation" here means each solution sits at its own amorphous solubility. The drug concentration itself differs between media, from 41.63 to 100.3 µg/mL in the media reported here, and the ratio of amorphous to crystalline solubility ranges from about 30 in GCA/TCA to about 187 in buffer X8.

Experimental analysis

The published media all shorten the time to crystals

In plain buffer, atazanavir had a crystalline solubility of 0.4 ± 0.1 µg/mL and an amorphous solubility of 74.9 ± 1.9 µg/mL R1. (The discussion repeats the buffer value as 74.9 ± 2.3 µg/mL; the results section gives 1.9 X3.) All three biorelevant media raised the crystalline solubility significantly, to about 1.65 (FaSSIF), 1.45 (Composite-SIF) and 0.63 µg/mL (FaSSIF-V3) R1. The amorphous solubility rose to 98.6 ± 4.4 µg/mL in FaSSIF and 100.3 ± 3.6 µg/mL in Composite-SIF, which share 0.75 mM PC as their only lipid but differ in bile salts, and fell slightly to 61.1 ± 2.1 µg/mL in FaSSIF-V3 (Figure 2a) R1·C11.

The induction time dropped from 253 ± 65 min in buffer to about 48 min in FaSSIF (48 ± 6 min in the discussion), about 81 min in Composite-SIF and 23 ± 6 min in FaSSIF-V3 (Figure 2b); all three media differed from buffer and from each other R2. The authors conclude that "it is apparent" that all media accelerate crystallization C12. They attribute the lower solubilities of FaSSIF-V3 to its lipid fraction rather than to its bile salts, because the Mod media with the same lipids also showed lower values, while noting that these media are "not directly comparable" because their lipid molarities differ C13·C14.

Phospholipids and their digests

Solubility. The bile salt pair alone, 2.8 mM GCA/TCA, raised the crystalline solubility to 1.46 ± 0.28 µg/mL but lowered the amorphous solubility from 74.9 to 44.2 ± 2.3 µg/mL (Figure S.1) R3. Adding PC raised the amorphous solubility step by step, to about 58, 61, 64, 68, 75, 83 and 93 µg/mL from 0.01 to 0.63 mM PC, each step significant against GCA/TCA alone (Figure 3a); the crystalline solubility stayed at 0.7–1 µg/mL, slightly below the bile-salt-only value R4·C15. Lyso-PC changed the crystalline solubility only marginally (0.81–1.23 µg/mL) and raised the amorphous solubility, "in general" 1.8-fold over GCA/TCA, but without a concentration trend R5·C17. Panel 3(b) with these data is blank in the supplied file, so only these statements are available X1. Sodium oleate on top of 0.32 mM lyso-PC left the crystalline solubility unchanged and moved the amorphous solubility only marginally, between about 67.5 and 88 µg/mL (Figure 4a) R7·C23. When PC was replaced by the digest made from the same amount, the amorphous solubility fell, from 68.1 ± 2.1 to 41.63 ± 0.55 µg/mL and from 83.02 ± 2.2 to 67.7 ± 2.4 µg/mL (Figure 5a, c) R9·C24.

The authors explain the solubility pattern through the bile salts. Bile salts mix with amorphous drug aggregates and lower the drug's thermodynamic activity in the amorphous phase, which lowers the amorphous solubility; PC takes bile salts into mixed micelles, leaving fewer to mix with the drug, so the amorphous solubility rises with PC; lyso-PC "likely" does the same C16·C18.

Induction time. In GCA/TCA alone atazanavir crystallized after 272 ± 94 min R6. Every level of PC or lyso-PC from 0.01 to 0.63 mM shortened this significantly, and lyso-PC was always the stronger promoter R6·C19. Only the end points are printed, because the plotted series (Figure 3c, d) is blank in the supplied file X1:

Table C. Induction times with PC, lyso-PC and their digests (text values; plot readings approximate) R6·R8·R10.

At equal molar concentration, lyso-PC brought crystallization on at least 2-fold faster than PC, and the effect "did not appear to scale" with the amorphous solubility changes, which "suggests" that lyso-PC is the more potent inducer C20. Sodium oleate added a further, concentration-dependent acceleration R8·C23. Replacing PC by its digests cut the induction time 4- to 5-fold, and the conclusion states that the digests "were determined to be more potent inducers of crystallization than the parent compound" R10·C24. Note by the AI writer (not discussed in the paper): the PC values at 0.08 mM (about 69 min) and 0.32 mM (about 80 min) from the other panels do not fall steadily between the printed end points of 185 and 55 min, so the shape of the PC dose-response cannot be checked from this file X17.

Why do the digests act faster? The authors list several explanations without testing them. PC "may" accelerate crystallization by tying bile salts up in micelles so that fewer remain to inhibit nucleation; "alternatively" by adsorbing on the emerging nucleus and lowering the interfacial energy between nucleus and solution; a third possibility is a higher supersaturation C21. For lyso-PC, its higher surface activity is "likely" to remove more bile salt, but because the induction time with lyso-PC is even shorter than in plain buffer (253 ± 65 min), faster adsorption on the nucleus is "likely the predominant mechanism" C22. Overall "it is postulated" that lyso-PC and sodium oleate lower the interfacial energy more than PC, supported by a cited FaSSIF-V3 prototype in which PC hydrolysis products gave an air/liquid surface tension of 32.6 mN/m against 54.4 mN/m for the corresponding PC C26·R32. The conclusion calls this "one possible explanation" C26.

Cholesterol

Cholesterol at 0.1–0.3 mM did not change the crystalline solubility appreciably and raised the amorphous solubility from about 47 to 59–62 µg/mL, independent of concentration (Figure 6a) R11·C28. Its effect on crystallization was "slight": the FaSSIF-V3 level of 0.2 mM lengthened the induction time from 9 ± 2 to 23 ± 6 min, "partly" offsetting the phospholipids and oleate R12·C28. The authors' "possible explanation" is that cholesterol raises the surface tension modestly (Fuchs et al.) and so hinders the adsorption of pro-nucleating molecules C29·R32. Note by the AI writer (not discussed in the paper): only these two induction times are printed (Figure 6b is blank), and the cholesterol-free base of this series behaved differently from the nearly identical lyso-PC + oleate medium of the other series (about 47 against 67.7 µg/mL; 9 against 15 min), so the cholesterol effect is best read within its own series X18·X9.

The bile salt fraction

Already with undigested PC, the six-bile-salt Composite-SIF held atazanavir supersaturated longer than TCA-only FaSSIF (about 81 against 48 min), in agreement with Indulkar et al. R2·C31. The question was whether this holds with the more realistic digest lipids. In the three media with the same FaSSIF-V3 lipid fraction, the induction time was 39 ± 12 min in FaSSIFMod (3 mM TCA), 74 ± 11 min in Composite-SIFMod (six bile salts) and 23 ± 6 min in FaSSIF-V3 (2.8 mM GCA/TCA), all significantly different (Figure 7b) R14·C32. Crystalline solubility was 0.94 ± 0.03, 0.62 ± 0.00 and 0.64 ± 0.04 µg/mL and amorphous solubility 71.4 ± 1.9, 75.9 ± 1.2 and 61.1 ± 2.1 µg/mL R13. The authors call the FaSSIFMod and Composite-SIFMod amorphous solubilities "similar", although Figure 7a marks the difference as significant (P ≤ 0.05) C32·X14.

In the abstract, swapping GCA/TCA for the bile salts of other media "partially circumvented" the crystallization-inducing effect of the phospholipids and digests, and dihydroxy bile salts were "particularly significant" in slowing crystallization C33. In the discussion the authors attribute the differences between bile salts to their hydrophobicity, and point out that trihydroxy bile acids make up on average only 47% of the duodenal pool, so trihydroxy-only media should not be expected to behave like human fluid C45·R34. Note by the AI writer (not discussed in the paper): no medium changes the dihydroxy content alone. The closest pair, SIF-2BAMod (1.89 mM TCA + 1.11 mM GCDCA, 78 ± 16 min) against FaSSIFMod (3 mM TCA, 39 ± 12 min), also replaces a taurine by a glycine conjugate X19.

The volunteers' bile acid pools

Without lipid. The six bile acid pools alone gave crystalline solubilities of 0.8 ± 0.1 to 1.26 ± 0.01 µg/mL and amorphous solubilities of 55.5 ± 5.9 to 83.7 ± 4.1 µg/mL (Figure 8a) R15. Induction times "varied", but the panel is blank in the supplied file; its caption says no statistics were done because the time in the V3 medium exceeded the experimental time-frame R16·X1. The authors see "no clear trend" and think a mix of concentration and composition effects "likely" underlies this, since bile salts are not interchangeable inhibitors C34·C35.

With the fixed FaSSIF-V3 lipid fraction. Here a clear pattern appeared. The amorphous solubility was 64.6 ± 1.3 to 67.8 ± 2.6 µg/mL in the three volunteers with low bile salt (0.57–2.2 mM) and 77.2 ± 1.5 to 81.5 ± 2.5 µg/mL in the three with high bile salt (3.6–5.1 mM) (Figure 9a) R17. The induction time was 23–33 min in the low group and 112–144 min in the high group, 6-fold between V6 (23 ± 2 min) and V5 (144 ± 19 min) R18. One-way ANOVA separates every low-group volunteer from every high-group volunteer (P < 0.001, Table S.2) R20. The authors read this as two "subtypes" of subjects and as induction time rising with total bile salt C36·C37.

Table D. Induction times in the volunteer media. The induction-time panels of Figures 9–11 are blank in the supplied file; the constant-ratio values other than V5 and V6 are read from the graphical abstract chart (approximate) R18·R22·R26·X2.

At a constant BS:PL ratio of 8.6. When the lipids were scaled with bile salt, the induction time still depended on bile salt "in general" (V6 45 ± 7 against V5 134 ± 23 min, 3-fold), but V4, the volunteer with least bile salt (0.57 mM) and least lipid (66 µM phospholipid), "deviated" and reached about 120 min R22·C38. Crystalline solubility was marginally lower in the low group (0.54 ± 0.12 to 0.6 ± 0.3 µg/mL against 0.74 ± 0.08 to 0.94 ± 0.03 µg/mL), and amorphous solubility "clustered" in a similar way (52.4 ± 2.5 to 71.6 ± 2.9 against 66.87 ± 0.63 to 84.1 ± 2.3 µg/mL) R21·C38. Note by the AI writer (not discussed in the paper): these amorphous ranges overlap, and Table S.3 does not separate V4 from V2 X14.

Composition at a fixed total. When all six compositions were made up to 3 mM with the FaSSIF-V3 lipids, solubility changed little and the induction time ran from 71 min (V1) to 110 min (V5), which the authors call "reasonably insensitive" to composition R25·R26·C39.

The authors' explanation is that more bile salt leaves more molecules free to inhibit crystallization "and/or" takes up more phospholipid and digest into micelles, away from the forming nucleus C40. Note by the AI writer (not discussed in the paper): with the fixed lipid fraction, bile salt level and BS:PL ratio rise together, and at constant ratio the medium with least bile salt but also least lipid (V4) was slow. The data therefore show a dependence on bile salt relative to the lipid present, which fits the second half of the authors' explanation X20·C40.

HPMC removes the difference

The two extreme volunteer media of the constant-ratio design, V5 (134 ± 23 min) and V6 (45 ± 7 min), were repeated with 1 mg/mL HPMC (Methocel E5). Both then stayed supersaturated for more than 360 min (Figure 12b) R28. HPMC raised the amorphous solubility of the V5 medium from about 84 to 91.5 µg/mL and left V6 at about 64 µg/mL; the crystalline solubility rose to about 3.5 and 2.1 µg/mL R27. Note by the AI writer (not discussed in the paper): the crystalline solubility in HPMC media was measured with a shaking bath and filtration rather than stirring and centrifugation, so this rise cannot be assigned to the polymer alone X13. The authors conclude that HPMC "could potentially" be used to circumvent inter-individual differences in vivo C43.

Which medium predicts best?

Finally, candidate media were compared with the average-volunteer medium (Table 4, average row). All gave what the authors call a "reasonable estimation" of its amorphous solubility, about 71 µg/mL: about 61 (FaSSIF-V3), 71 (FaSSIFMod), 76 (Composite-SIFMod) and 63 µg/mL (SIF-2BAMod), although all except FaSSIFMod differ significantly from it (Figure 13a) R29·C47. For the induction time the authors state that "it is clear" that media with tri- and dihydroxy bile salts estimate the average volunteer better than trihydroxy-only media, and that SIF-2BAMod behaves like Composite-SIFMod C47. The panel with these data (Figure 13b) is blank in the supplied file, and the average volunteer's own induction time is not printed anywhere X11.

Table E. Induction times of candidate media against the cited duration of atazanavir supersaturation in the human duodenum (text values, p.36) R30·R31.

The authors anchor the choice on an in vivo study: after a standard dose with a glass of water, atazanavir stayed supersaturated in the duodenum of healthy volunteers for about 75 min (Hens 2020, ref 32) R31. Composite-SIFMod (74 ± 11 min) and SIF-2BAMod (78 ± 16 min) are close to this value, FaSSIF (48 ± 6 min) and FaSSIF-V3 (23 ± 6 min) are not, and the authors propose SIF-2BAMod as a cheaper and simpler stand-in for the six-bile-salt medium C49·C48. Note by the AI writer (not discussed in the paper): the in vivo value is a duration of measured luminal supersaturation after dosing, not an induction time at the amorphous solubility, so the close numbers are suggestive rather than a validation X12.

Discussion

The authors' picture

The paper's results fit one picture, which the authors build from several hedged explanations rather than from direct measurements C16·C21·C26·C40·C45:

Two observations support the adsorption idea in the authors' view: lyso-PC acts more strongly than its effect on amorphous solubility would suggest, and with lyso-PC the induction time falls below even that of plain buffer, which removing bile salt alone could not do C20·C22. The cited surface tension of a digest-containing medium (32.6 against 54.4 mN/m) points the same way R32·C26. None of these steps (interfacial energy, adsorption on nuclei, the distribution of bile salt between micelles and free solution) was measured C26·C40.

Links to earlier work and to people

The authors extend the bile salt result to earlier in vitro studies in which weak bases stayed supersaturated longer in fed-state than in fasted-state media (Kostewicz 2004, ref 40; Dai 2007, ref 41) R35. Those results had been put down to either more bile salt or more PC in the fed medium; the authors write that their outcomes "implicate augmented bile salt levels, but not PC" C42. They add that the real fed state is more complex, with a less acidic stomach (pH 3–7) and more digested lipid C42.

Because bile acid pool size depends on gut bacteria, the bile salt dependence "demonstrates a potential role" for the microbiota: antibiotics, which enlarged the bile acid pool in mice (Sayin 2013, ref 28), might change how long a formulation stays supersaturated, and it is "interesting to speculate" that some of the variability in response to supersaturating formulations comes from the gut microbiota–bile acid axis C41·R38. Disease could shift the balance of tri- and dihydroxy bile acids: deoxycholate accumulates in some gallstone patients and falls after antibiotics, while in cirrhosis its loss appears to be offset by liver-made chenodeoxycholate, so "clinically significant impacts upon crystallization are unlikely" there C51·R39. None of this was tested in the paper C41·C56.

What the authors recommend

• Include the most physiologically relevant lipids, that is the digests rather than PC, when testing supersaturating formulations ("it is advisable") C30.
• Use a mixture of tri- and dihydroxy bile salts; media with trihydroxy bile salts only "may not adequately predict" in vivo crystallization, which "may partly explain" why in vitro tests overestimate precipitation C46·C47. SIF-2BAMod (1.89 mM TCA + 1.11 mM GCDCA with the FaSSIF-V3 lipids) is proposed as practical C48.
• There "may not be one representative SIF": average and extreme media should both be used in development C50.
• Polymers such as HPMC "may compensate" for differences in bile acid concentration between people C44.
• Dynamic dissolution models that include secretion and absorption are recommended, because the bile salt and phospholipid effects depend on concentration C52.

Note by the AI writer (not discussed in the paper): for a formulator, FaSSIF-V3 gave the shortest induction time of all published media tested, so a formulation that holds atazanavir supersaturated in it has passed the harsher screen; the volunteer data add that media with little bile salt relative to lipid are the relevant stress case. In the two media tested, 1 mg/mL HPMC outweighed a 3-fold difference between subjects, which suggests that with an effective polymer the choice among these media matters less; that would need testing with the formulation's own polymer and amount.

Limitations

Stated by the authors

• FaSSIF-V3 and the PC-containing media are "not directly comparable" because their PC and hydrolysis-product molarities differ C14.
• Whether inter-individual differences in fasted bile salt levels change supersaturation and bioavailability in vivo needs future work C56.
• The next step is to measure solubility and crystallization of diverse drugs in fasted human fluid against SIF-2BAMod, and to characterise the solid form of the precipitates C53.
• The fed state, with a higher gastric pH and more digested lipid, is more complex than the media studied C42.

The AI writer's own judgement

• Blank panels. In the supplied pre-proof, most induction-time panels (Figures 3b–d, 4b, 5d, 6b, 8b, 9b, 10b, 11b, 13b) are blank; whole series exist only as end points or ranges in the text, and two panels can be recovered only from the graphical abstract X1·X2.
• Mechanisms not measured. Interfacial energy, adsorption on nuclei and micelle composition were not measured, and the authors keep several explanations open C21·C25·C26·C40.
• Bile salts alone. 2.8 mM GCA/TCA alone did not prolong the induction time over buffer (272 ± 94 against 253 ± 65 min), so "bile acids are crystallization inhibitors" holds as a comparison between lipid-containing media, not against buffer X10.
• The average-volunteer benchmark. The claim that mixed media best predict the average volunteer rests on a blank panel and an unprinted value; the in vivo comparison uses a different measure X11·X12.
• Confounded factors. Bile salt level and BS:PL ratio change together in the volunteer media, and the closest test of dihydroxy bile salts also changes conjugation X20·X19.
• Method and medium details. The cholesterol base medium differs from the other series, the HPMC crystalline solubility used a different separation, some group statements are looser than the plotted data, the basis of BS:PL 8.6 and the handling of the unlisted bile acid fraction are not stated, and the end of induction-time monitoring is not given X9·X13·X14·X6·X5·X15.
• Scope. One drug, one supersaturation per medium (its amorphous solubility), one polymer at one level, and no characterisation of the precipitate C10·X8·X15.

Overall the AI writer rates the evidence as moderate: the solubility data and the large contrasts (digests against PC, high against low bile salt, with and without HPMC) are direct and replicated, while the mechanisms, the dihydroxy effect in isolation and the in vivo relevance remain inferences R10·R18·R28·X1·X11.

Conclusion

This paper asks whether the simulated intestinal fluid used to test supersaturating formulations changes what one sees, and for atazanavir the answer is clearly yes. The digests of phosphatidylcholine, lyso-PC and sodium oleate, which are what the intestine actually contains, trigger crystallization 4 to 5 times faster than the same amount of undigested PC R10·C24. The bile salts work in the opposite direction: with the same lipids, a six-bile-salt mixture held supersaturation about twice as long as TCA alone, and in media modelled on six healthy volunteers, those with more bile salt stayed supersaturated 3 to 6 times longer R14·R18. A small amount of HPMC removed that difference R28. The authors propose a two-bile-salt medium with phospholipid digests, SIF-2BAMod, whose induction time (78 ± 16 min) is close to the duodenal supersaturation reported in humans R30·R31·C48. The mechanism they offer, stronger adsorption of the surface-active digests on forming nuclei, and the link to gut microbes and to variability between patients, remain proposals for further work C26·C41·C56.