Seeing the drug-rich liquid: polarity-sensitive fluorescent dyes as detectors of liquid–liquid phase separation in supersaturated drug solutions

Background and motivation

Many new drug candidates dissolve poorly in water, and formulations such as amorphous solid dispersions (ASDs) work by creating supersaturated solutions, in which more drug is dissolved than the crystalline solubility allows. Since the 1960s, solutions of this kind have repeatedly been seen to contain small, non-crystalline drug aggregates. Tachibana and Nakamura reported nano-sized aggregates from a beta-carotene–PVP dispersion in 1965 (ref 1); later reports covered mangostin–PVP, dispersions in hydroxypropylmethyl cellulose acetate succinate (HPMCAS) and hydroxypropylmethyl cellulose (HPMC), and ritonavir melt extrudates (refs 2–5). The same aggregates appear when supersaturation is created without an ASD, by lipid digestion, by a pH shift, or by diluting a solvent stock as in high-throughput screening (refs 6–12). C2

The authors treat these aggregates as the product of a well-defined phase transition, liquid–liquid phase separation (LLPS). LLPS occurs in solutions that have a miscibility gap and is known as an undesirable precursor to crystallization of proteins and small molecules (refs 12, 13). For high log P drugs the concentration at which LLPS starts has been linked to the solubility of the water-saturated supercooled liquid form of the drug, which the paper calls the "amorphous solubility". When LLPS happens, a single supersaturated phase splits into a dispersed, nano-sized drug-rich phase and a continuous aqueous phase; the solution then stays supersaturated until crystals appear, even though it is visibly turbid (refs 14–17). C3·C4 It has been "postulated" that the drug-rich phase may act as a reservoir that maintains supersaturation and so benefits absorption (Raina 2014, ref 14; Frank 2014, ref 18). C5

If LLPS matters for delivery, it has to be measured, and the existing tools have gaps. Static, quasi-elastic and dynamic light scattering and UV extinction (refs 14, 19–21) report that scattering centres have appeared and roughly how large they are, but not what they are made of: a crystal and a liquid droplet scatter alike. Polarized light microscopy (PLM) can confirm that a drug-rich phase is non-crystalline because it shows no birefringence, but it is not sensitive when the drug concentration is low or the colloids are very small (refs 14, 22). Small-angle neutron scattering, NMR and surface plasmon resonance have also been used for small-molecule aggregation (refs 23–25). C6

The idea the authors borrow comes from surfactant science. Fluorescent probes have long been used to find the critical micelle concentration, because a probe's spectrum changes when it moves from water into the less polar interior of a micelle (refs 37, 38). By analogy the authors hypothesise that probes of "moderate lipophilicity" will partition into the drug-rich phase formed on LLPS and change their emission intensity and wavelength. C7·C24 They test this with six probes and six dihydropyridine drugs that the group had already characterized for crystallization tendency and LLPS (refs 15, 16, 26–28). The fluorescence onsets are compared with a UV-extinction method and with amorphous solubilities estimated from thermal data, and the spectra are used to tell phase separation by crystallization from LLPS. P3·P6·P9

Background concepts

Supersaturation, amorphous solubility and LLPS. The crystalline solubility ${\sigma }_c$ is the concentration in equilibrium with the crystal. An amorphous form has higher free energy, so the concentration in equilibrium with it, the amorphous solubility ${\sigma }_a$, is higher. Below ${\sigma }_a$ a supersaturated solution is one phase; above it, the excess drug separates as a second, drug-rich liquid phase (LLPS). The drug-rich phase is a supercooled liquid that contains some water. The authors estimate ${\sigma }_a$ from ${\sigma }_c$ with the relation used in their earlier work (refs 14, 29), printed as Eq. 1 E1·P3:

In plain terms: the amorphous solubility equals the crystalline solubility multiplied by the free-energy advantage of the amorphous form, $e^{\Delta G/RT}$, and by a correction $e^{-I({\alpha }_a)}$ for the drop in activity of the amorphous drug when it takes up water. Here $\Delta G$ is the free energy difference between amorphous and crystalline drug, $R$ the gas constant and $T$ the temperature; $I({\alpha }_a)$ is the correction term of Murdande et al. (refs 31–33), obtained from moisture sorption. The paper prints the correction in the display as $I(a_a)$ and in the text as $I({\alpha }_a)$; it is the same term. E1 $\Delta G$ comes from the Hoffman equation, Eq. 2 E2:

In plain terms: the free energy gap between the supercooled liquid and the crystal grows with the heat of fusion $\Delta H_f$ and with the undercooling $\Delta T=T_m-T$ below the melting temperature $T_m$. The heat of fusion and melting point were measured by differential scanning calorimetry (DSC); the paper does not print them, nor the sorption data or the resulting ${\sigma }_a$ values except as bars in Figure 8 and two values in the text. P4·P5·E2·R11

Environment-sensitive fluorescence. A fluorescent molecule absorbs light, rises to an excited state with a larger dipole moment, and later emits light as it returns to the ground state. Between absorption and emission it loses energy, by vibrational relaxation and by the reorientation of surrounding solvent dipoles around its new dipole (solvent relaxation). Because energy and wavelength are linked by Eq. 3 E3:

In plain terms: the more energy lost before emission, the lower the energy $E$ of the emitted photon and the longer its wavelength $\lambda$ ($h$ is Planck's constant, $\nu$ the frequency, $c$ the speed of light). The excited state lives for nanoseconds while solvent relaxation takes picoseconds, so the emission comes from the relaxed state. In water, a polar solvent, much energy is lost and the emission is red-shifted and weak; in a non-polar solvent such as benzene less is lost, so the emission is blue-shifted (a "hypsochromic" shift) and brighter (Figure 13). C25 Viscosity, the rate of solvent relaxation, conformational changes, charge transfer and proton transfer in the excited state, and probe–probe interactions can add to the effect; the authors note that probes with strong charge transfer in the excited state (ICT) have the largest dipole changes and are expected to be the most polarity-sensitive. C28

The six probes. The probes are pyrene, PRODAN (6-propionyl-2-dimethylaminonaphthalene), dansyl glycine, dapoxyl sulfonic acid sodium salt, DASPEI (2-(4-(dimethylamino)styryl)-N-ethylpyridinium iodide) and 4-Di-2-ASP (4-(4-(diethylamino)styryl)-N-methylpyridinium iodide) (Figure 1). M8–M13 Five of them have one broad emission band whose maximum moves with polarity. Pyrene is different: its emission has several sharp vibronic peaks, and the ratio of the first peak (I1, 373 nm) to the third (I3, 383 nm) falls as its surroundings become less polar, so pyrene is read as a ratio rather than a wavelength. C8·R1.a Note by the AI writer (not discussed in the paper): Figure 1 labels the dansyl structure "Dansyl Chloride", but the drawn molecule carries a glycine group, and every other part of the paper names dansyl glycine. X1

Extinction. Extinction is the loss of light passing through a sample, by absorption plus scattering. Read at 450 nm with a dip probe as drug is added, it stays near zero in a clear solution and rises once a light-scattering phase forms. That rise is the "onset", but it does not say whether the new phase is liquid or crystalline. P6·C9

Methods in detail

The study has four strands that meet in one comparison: a solvent calibration of the probes, a light-scattering onset with a visual test for crystals, a fluorescence onset, and a thermodynamic estimate of amorphous solubility. Two side studies use felodipine's own fluorescence and follow crystallization with pyrene. All aqueous work used 50 mM phosphate buffer at pH 6.8, at 25 and 37 °C. M16·A2

The model compounds

The six drugs are 1,4-dihydropyridines that share one core and differ in their ester groups and in the substituted phenyl ring at position 4 (Figure 2): felodipine carries a 2,3-dichlorophenyl ring, nifedipine and nisoldipine an ortho-nitrophenyl ring, and cilnidipine, nimodipine and nitrendipine a meta-nitrophenyl ring. M1–M6 Felodipine came from Attix Pharmaceuticals and the other five from Euroasia. M1·M2 The polymer, used as a crystallization inhibitor, was HPMC (Pharmacoat 606), at 100 μg/mL in the experiments of Figures 6b and 7; its concentration is not given for Figures 5, 8 and 12. M7·X12 The group had shown earlier that nifedipine is a very fast crystallizer in water and that nimodipine crystallizes faster than the other four, though slower than nifedipine (refs 14, 44). R17

Calibrating the probes against solvent polarity

Each probe was dissolved at 1 mg/mL in dimethyl sulfoxide and diluted to 1 μg/mL in water, buffer, methanol, ethanol, isopropyl alcohol, benzene, toluene and hexane. Excitation scans gave each probe's excitation and emission maxima, and emission spectra were then recorded at the excitation maximum in 1 cm quartz cuvettes on a SPEX Fluoromax-3. P8 Table I lists the maxima in buffer, methanol and benzene; Figure 3 shows the spectra for six of the solvents. R1·R2

The light-scattering onset

Drug dissolved in methanol at 5, 10, 20 or 25 mg/mL was infused slowly by a syringe pump into buffer while a 1 cm dip probe read extinction at 450 nm. Extinction was plotted against concentration, and the concentration of the sharp rise was taken as the onset of a new phase (Figure 4). P6 The infusion rate is printed as "5, 10, 20 or 40 μg/mL" without a time unit, and the final methanol fraction is not given. X12 Whether the new phase was liquid or crystalline was decided by eye and microscope: LLPS when the solution showed no visible particles, a hazy bluish tinge like the Tyndall effect of colloids, and no birefringence under PLM; crystallization when particles appeared. P7·C9 Runs were made in buffer and with HPMC at both temperatures (Figure 5). R5

The fluorescence onset

Buffer containing 1 μg/mL probe was the test medium. Methanolic drug stocks of 10 μg/mL to 25 mg/mL were added in aliquots to make a series of solutions of rising drug concentration; each was stirred at 25 or 37 °C in a water-jacketed cuvette and its spectrum recorded at the probe's excitation maximum (3 nm slits, 100 ms integration). Each solution was prepared and measured within 1 min, so that crystallization would not change the spectrum. P9 The emission maximum and intensity were plotted against drug concentration on a log axis, and the onset was the concentration where they changed sharply (Figures 6 and 7). R6·R9 Felodipine at 25 °C was run with all six probes in triplicate (Table II); the six-drug comparison of Figure 8 used DASPEI; nifedipine at 37 °C was shown with 4-Di-2-ASP. R12·R11·R9

The thermodynamic estimate

Crystalline solubility was measured by shaking excess crystalline drug in 18 mL buffer for 48 h at 25 and 37 °C, protected from light, then ultracentrifuging and assaying the supernatant by HPLC at 360 nm (standard curve 0.05–50 μg/mL). P1·P2 Heat of fusion and melting temperature came from DSC at 10 °C/min in pinholed aluminium pans, and the water correction from moisture sorption on a TA Q5000; these fed Eqs. 1 and 2. P3·P4·P5 None of the inputs is printed, so the estimates cannot be recomputed from the paper. E1·E2

The two side studies

Felodipine, unlike the other five drugs, fluoresces strongly. Its emission was recorded in six solvents (Figure 10) and at 0.25–100 μg/mL in buffer at 25 and 37 °C (Figure 11) to test whether it can report its own phase separation; the excitation wavelength is not stated. P10·X12 Finally, pyrene I1/I3 was followed while nifedipine, with pre-dissolved HPMC, was raised to 400 μg/mL at 25 °C, past the LLPS onset and into the range where particles appeared (Figure 12). P11 The methods heading also announces crystallization induction times, but none are reported. X15

Experimental analysis

Every probe emits at shorter wavelength in less polar solvents

Table I gives the calibration. From buffer to benzene the emission maximum fell from 575 to 472 nm for dapoxyl, from 520 to 463 nm for PRODAN, from 554 to 502 nm for dansyl glycine, from 610 to 585 nm for 4-Di-2-ASP and from 580 to 564 nm for DASPEI, with methanol in between; pyrene I1/I3 fell from 1.6 in buffer to 1.2 in methanol and 1.07 in benzene. R1 The full spectra of Figure 3 add ethanol and toluene and keep the same order, water and buffer at the long-wavelength end and toluene giving the largest blue shift (for PRODAN about 455 nm against about 521 nm in buffer). R2 The abstract summarises this as a "large hypsochromic shift" for each probe. C8

Table A. Excitation and emission maxima (nm) of the probes at 1 μg/mL, from Table I; the last column is the difference computed by the AI writer. R1

Note by the AI writer (not discussed in the paper): two details of the calibration deserve care. First, the discussion calls the styryl dyes DASPEI and 4-Di-2-ASP the most polarity-sensitive of the six probes because of their charge transfer, yet in Table I they show the smallest buffer-to-benzene shifts. C29·X4 Second, the pyrene ratios one reads from the normalized spectra of Figure 3b are about 1.5 in water and buffer and about 1.0 in benzene, slightly below the 1.6 and 1.07 of Table I; the Table I values are the ones to use. R2·X6

Light scattering finds a new phase; two drugs crystallize first

For nisoldipine at 25 °C (Figure 4) the extinction stayed near zero up to about 6 μg/mL and then rose linearly to about 0.25 at 16.4 μg/mL; the authors place the onset at "approximately" 7 μg/mL, where the solution turned hazy. R3·C10 Repeating this for all six drugs (Figure 5), every compound formed a hazy, birefringence-free liquid phase at both temperatures, except nifedipine without polymer at 25 and 37 °C and nimodipine without polymer at 37 °C, which crystallized; the authors say these crystallize "just below" the concentration needed for LLPS. R4·C11 With HPMC both drugs formed the liquid phase. R4

Table B. Equilibrium crystalline solubility and extinction onset (μg/mL) in pH 6.8 buffer, read from Figure 5 (approximate); "crystals" marks systems where crystallization was seen. R5·R4

Three points stand out in Table B. The crystalline solubilities are far below the onsets, for example felodipine about 0.7 against about 6.0 μg/mL at 25 °C. R5.b For the four drugs that phase separate on their own, HPMC moved the onset by at most about 1.5 μg/mL. R5 For the two that crystallize, the two values mark different events: the buffer value marks crystals and the HPMC value the liquid phase (nifedipine about 38 and about 49 μg/mL at 25 °C). R5.c·R4 Every onset was higher at 37 °C than at 25 °C. R5

Fluorescence onsets: felodipine and nifedipine at 37 °C

Figure 6 shows the method at work for felodipine with DASPEI at 37 °C. Without polymer the emission maximum stayed at about 577–579 nm from 1 to 5 μg/mL, then fell to about 575 nm at 7.5, about 571 nm at 10 and about 558 nm at 25 μg/mL, while the intensity rose from about 14,000 to about 97,000 cps; with 100 μg/mL HPMC the maximum fell in the same way (about 579 to 558 nm by 20 μg/mL). R6 The authors read the start of the blue shift, at approximately 8 μg/mL, as the point where the homogeneous solution becomes a two-phase system with drug-rich aggregates. At and above it the solutions were hazy, free of particles and without birefringence; extinction gave a new phase at around 9 μg/mL, and the estimated amorphous solubility was approximately 8 μg/mL, a "good correlation". R7·C12·C13 They add that the polymer changes neither the LLPS concentration nor the emission properties of the probe. C12 Note by the AI writer (not discussed in the paper): the emission maxima and the onset indeed agree, but at 1–2 μg/mL the intensity was about 24,000–26,000 cps with HPMC against about 14,000–19,000 cps without, within overlapping error bars. X13

Nifedipine is the test case for crystallization. Without polymer it crystallized below its LLPS concentration, and the probes showed no distinct change in wavelength or intensity; these data are described but not shown. The authors' explanation is that no drug-rich liquid phase formed for the probe to enter. R8·C14 With 100 μg/mL HPMC (Figure 7) the 4-Di-2-ASP maximum stayed at about 609–611 nm from 10 to 70 μg/mL, dropped to about 603 nm at 75 μg/mL and kept falling to about 579 nm at 200 μg/mL; the intensity stayed at about 7,000–8,000 cps up to 70 μg/mL and rose to about 30,500 cps at 200 μg/mL. R9 The three onsets were 70 μg/mL by fluorescence, 73 μg/mL by extinction and 73 μg/mL by the estimate, which the authors call "excellent agreement". R10·C15 Figure 5b shows the extinction onset with HPMC at about 67 μg/mL rather than 73; the text and Figure 8b agree on 73, so that is the value to use. R5.c·X7

Note by the AI writer (not discussed in the paper): in both systems the probe maximum went past its value in benzene, DASPEI to about 558 nm (benzene 564 nm) and 4-Di-2-ASP to about 579 nm (benzene 585 nm). The shift in the drug-rich phase is therefore not a simple reading on the solvent scale of Table I; the authors' own list of additional factors, viscosity among them, would be one way to account for it. X5·C28·C35

Six drugs, three methods

Figure 8 is the central comparison: for each drug and temperature, the estimated amorphous solubility, the extinction onset and the fluorescence onset (DASPEI per the text; nifedipine and nimodipine with HPMC). R11

Table C. LLPS onset (μg/mL) by three methods, read from Figure 8 (approximate); n behind the error bars is not stated. R11·X12

The authors describe "a close correlation" for each compound at each temperature, and the abstract "good agreement". C16 The onsets span almost two orders of magnitude, from below 2 μg/mL for cilnidipine to about 70 μg/mL for nifedipine at 37 °C, and the three methods track each other across that range. R11 Note by the AI writer (not discussed in the paper): read from the bars, the highest and lowest of the three values differ by a factor of about 1.2–1.7 at 25 °C and 1.0–1.2 at 37 °C for five drugs, and by about 3 for cilnidipine, whose values all lie below 2 μg/mL. Which probe gave the nifedipine and nimodipine bars is also not quite clear: the text names DASPEI, but the nifedipine value at 37 °C equals the 4-Di-2-ASP onset of Figure 7. X9·X11

When the drug itself fluoresces

Felodipine at 25 °C was run with all six probes (Table II). Four gave onsets of 5–6 μg/mL; dansyl glycine and dapoxyl gave none. R12

Table D. Felodipine LLPS onset at 25 °C with each probe, mean ± SD, n = 3, from Table II; pyrene read from the intensity at 373 nm. R12

The reason is felodipine's own emission, which overlaps the probe bands (Figure 9). With dansyl glycine the felodipine band near 440 nm (about 117,000 cps at 2 μg/mL) dwarfed the probe band near 555 nm (about 20,000–25,000 cps); with DASPEI the felodipine band reached about 170,000 cps at 12.5 μg/mL while the DASPEI band rose to about 40,000 cps. R13.c·R13.d·C17 The same overlap made the pyrene I1/I3 ratio unusable; instead the authors used the sharp fall of the individual pyrene peaks (from about 5.1 × 10⁶ cps at 0.1 μg/mL to about 1–1.7 × 10⁶ cps at 15–20 μg/mL) to place the onset at 6.00 ± 0.57 μg/mL. R13.a·R12.a·C18 They attribute the fall to quenching by felodipine, which "confirms" close contact between pyrene and felodipine after LLPS. C19 The four usable probes agree with each other ("good agreement") and with the DASPEI value of Figure 8a. C17·R11.b The authors mention possible Förster resonance energy transfer between felodipine and the probes as something they "hope to investigate". C34

Felodipine as its own indicator

Felodipine's emission is broad, with a maximum near 438–445 nm in water and buffer, about 435 nm in methanol and ethanol and about 425–435 nm in benzene and toluene (Figure 10), so it too senses polarity. R14·C20 In buffer (Figure 11) its maximum stayed at about 442–443 nm from 0.25 to 3 μg/mL at 25 °C, moved to about 440 nm at 5–7.5 μg/mL and to about 433 nm at 100 μg/mL; at 37 °C it stayed at about 442–444 nm up to 7.5 μg/mL, moved to about 439.5 nm at 10 μg/mL and to about 432.5 nm at 100 μg/mL. R15 The authors call felodipine a "self-indicator": wavelength and intensity stay "more or less constant" until the anticipated LLPS concentration, then the emission blue-shifts and the intensity rises sharply. C21 They add that local viscosity increases after LLPS and that hydrogen bonding of felodipine "may be higher" in the drug-rich phase, both contributing to the shift. C35 Note by the AI writer (not discussed in the paper): the wavelength change starts close to the onsets found by the other methods, but the intensity per μg/mL roughly doubles only above about 15–25 μg/mL at 25 °C, so the wavelength is the cleaner marker. X14

Following crystallization after LLPS

Figure 12 follows pyrene I1/I3 as nifedipine, with pre-dissolved HPMC, is raised at 25 °C. The ratio was about 1.79–1.84 from about 5 to 60 μg/mL, about 1.72 at 75, about 1.13 at 120 and about 0.76 at 200 μg/mL; it then rose to about 0.94 at 300 and about 1.60 at 400 μg/mL, while particles appeared in the solution. R16 The authors place LLPS at around 40 μg/mL and the onset of crystallization at around 200 μg/mL, and read the recovery as pyrene being expelled from the shrinking nifedipine-rich phase into the more polar water ("can be attributed"). C22·C23 Note by the AI writer (not discussed in the paper): in the plotted data the ratio first falls between about 60 and 75 μg/mL, not at 40; Figure 8a gives about 40 μg/mL by fluorescence but about 48–49 μg/mL by the other two methods, and the HPMC level of this experiment is not stated. X8·X12

Discussion: how the authors explain the signal

The authors' account joins the photophysics of the probes to the phase behaviour of the drugs. Below the amorphous solubility the probe sits in water and emits at its aqueous wavelength. Above it, a drug-rich supercooled liquid forms and the probe partitions into it, where it is "mainly surrounded by more non-polar drug molecules", with some water. The local polarity drops sharply, less energy is lost to solvent relaxation, and the emission moves to shorter wavelength and higher intensity. C26·C25 They prefer the wavelength as the read-out because, unlike intensity, it does not depend on the excitation intensity or the probe concentration. C27 For the two styryl dyes they suggest that the aminophenyl tail, which is known to embed in hydrophobic membrane regions, "likely" enters the drug-rich phase; PRODAN is known to undergo charge transfer, hydrogen bonding and a conformational change of its carbonyl group. C30·C31

The crystallizing systems are, in the authors' words, "important verification" of the approach: a probe that only responds to a drug-rich liquid should stay silent when the drug goes straight to crystals, and it did. This is what separates the method from light scattering, which registers any new phase. C40·R8 Why the two drugs crystallize first is explained by kinetics: nifedipine is a very fast crystallizer and nimodipine the next fastest of the six, so crystallization "precedes LLPS"; HPMC "is known to" impede nucleation of dihydropyridines and so lets both reach LLPS; and nimodipine's crystallization at 37 °C but not at 25 °C "can be attributed" to faster nucleation at the higher temperature. C37·C38·C39·R17 The last step of the chain, crystallization after LLPS, follows from the fact that LLPS solutions of high-melting compounds are metastable and crystallize at a time set by nucleation kinetics; as crystals consume the supersaturation, the drug-rich phase disappears and the probe returns to water, which reveals crystallization in a solution that is already turbid. C41·C42

Felodipine needed its own explanation. Its strong fluorescence is attributed, from the literature, to hindered rotation of its ortho-chloro-substituted phenyl ring, which gives a long-lived excited conformer with a bi-exponential decay and a high quantum yield, together with hydrogen bonding of the ring N–H. C32·R18 The other five drugs do not fluoresce, which the authors explain by the nitro group "in position 2 of the phenyl ring", through vibrational coupling that "possibly" quenches emission or, as "an alternate hypothesis", through charge transfer to the nitrobenzene group. C33 Note by the AI writer (not discussed in the paper): Figure 2 draws the nitro group in position 2 only for nifedipine and nisoldipine; cilnidipine, nimodipine and nitrendipine carry it in position 3, so the explanation as worded covers only two of the five. X2

What it means for formulation

The authors state that environment-sensitive probes "can be used to determine" the amorphous solubility of a compound by generating the amorphous material in situ, that precipitation to an amorphous or a crystalline phase "can be differentiated" with probes combined with orthogonal techniques, and that the approach helps to characterize supersaturated solutions, which matter for delivering poorly water-soluble drugs. C44·C45·C46

Note by the AI writer (not discussed in the paper): the amorphous solubility is the most supersaturation a formulation such as an ASD can build before drug-rich droplets form, so a quick, low-material way to measure it is useful in screening. In this paper the probe onset agreed with the thermodynamic estimate, which needs melting, sorption and solubility data. Three practical points follow from the data. The drug's own fluorescence has to be checked first, since it can swamp a probe (felodipine blocked two of six) or serve as one. Fast crystallizers need a nucleation inhibitor, and the result then refers to that polymer-containing medium. And because each point is read within a minute, the method locates the liquid–liquid boundary before crystallization, not the long-term behaviour of the solution. R12·R4·P9

Limitations

Limits the authors state. The authors note that felodipine's autofluorescence "has posed several challenges", through overlapping emission and possible energy transfer to the probes, which they "hope to investigate in the near future". C34 They also present partitioning of the styryl tail and the role of viscosity and hydrogen bonding in hedged terms ("It is likely", "may be higher"). C30·C35 The paper does not state other limitations.

The AI writer's own judgement. The core comparison is convincing in direction: three independent routes give similar onsets from below 2 to about 73 μg/mL, and the crystallizing systems behave as the model predicts. R11·R8 Several things weaken it. Almost every number is a reading from a bar or a point; n is stated only for Table II, so the error bars of the figures are undefined. X12 The agreement at 25 °C is within a factor of about 1.7 for five drugs and about 3 for cilnidipine, looser than "close correlation" suggests. X9 The claim that each probe indicates LLPS rests on DASPEI for the six-drug comparison and on felodipine alone for the probe comparison, where two probes failed. X10 Partitioning of the probe into the drug-rich phase is stated as a result in the abstract but was never measured: there is no partition coefficient, emission lifetime or droplet size, and the probe maxima go past their benzene values. X17·X5 The crystallization test is one system, its plotted onset differs from the stated one, and the announced induction times are not reported. X8·X15 Small inconsistencies in the source (Figure 1 label, nitro positions, the abbreviation list, the Figure 5b nifedipine bar) do not change the conclusions but should be known to anyone reusing the values. X1·X2·X3·X7 Each fluorescence point was read within 1 min, while the extinction onset came from continuous infusion at an unstated time scale, so the crystal-or-liquid assignments belong to these short experiments. X16

Conclusion

The paper shows, for six dihydropyridines at two temperatures, that a fluorescent dye in the buffer can mark the concentration at which a supersaturated solution splits into a drug-rich liquid, and that this onset agrees with light scattering and with a thermodynamic estimate of amorphous solubility. R11·C16 The dye stays silent when the drug crystallizes instead, and pyrene shows crystals replacing the droplets in one system, so the method adds what scattering lacks: an indication of whether the new phase is liquid. R8·R16·C40·C42 The proposed reason, partitioning of the dye into the drug-rich phase, is plausible but inferred, and the quantitative agreement is looser than the wording suggests for the least soluble drug. C26·X17·X9