Solid-state methods

How it was shown

Taylor and Zografi (1998) asked whether FT-Raman spectroscopy can quantify the degree of crystallinity, and so the amorphous content, of a drug powder across the whole composition range, including small amounts of either form taylor1998a C3, C4. The two reference forms were γ-indomethacin from Sigma, a fine powder below 150 µm taken as 100% crystalline, and amorphous indomethacin made by melting at 165 °C for 5 min, quench cooling in liquid nitrogen, warming under vacuum, light pulverizing below 150 µm and storage over P₂O₅ at −20 °C taylor1998a M1, M2, P1, C6. XRPD and DSC characterized both: sharp peaks against a halo, a glass transition of 45 °C for the glass, and fusion at 161 °C with 109 J/g for the crystal taylor1998a R1, M1, M2. Mixtures of known ratio were prepared geometrically to 500 mg in a Wig-L-Bug capsule without the ball, sealed in a glass vial of about 1 cm diameter and stored at −20 °C and 0% RH; agitating the crystal in the capsule did not change its DSC enthalpy of fusion taylor1998a P4, R2, C7. A Bruker RFS 100 FT-Raman system (1064 nm, spot of about 1 mm, about 450 mW, 100 scans from 4000 to 50 cm⁻¹ at 4 cm⁻¹, backscattered light, Ge detector) recorded each powder through the glass wall taylor1998a P6. The benzoyl carbonyl bands at 1698 cm⁻¹ (crystal) and 1680 cm⁻¹ (glass) were chosen for quantitation; because the amorphous band is broadened, each measured height contains a share of the other form, which the authors treat with the two-band model of Kontoyannis et al. taylor1998a C9, C10, E3. Twelve calibration mixtures covering 0–100% gave the last unknown ratio, and twelve mixtures not used in the calibration, with 1, 3 or 6 independent samples of five runs each, served as validation taylor1998a P7, P11, R6, R9.

What the data show

The band-height ratio of a mixture is

I_R = \frac{I_c^{1698}}{I_a^{1680}} = \frac{x_c K_c^{1698} + x_a K_a^{1698}}{x_c K_c^{1680} + x_a K_a^{1680}}

with x the fraction of each form and K a band factor taylor1998a E3. From ten spectra of each pure form the crystalline band contributes 0.052 of its own height at 1680 cm⁻¹ and the amorphous band 0.401 of its height at 1698 cm⁻¹; plotting (x_c-1)(I_R-0.401) against x_c(0.052\,I_R-1) for the calibration mixtures gave a line with correlation coefficient 0.993 and slope 2.72, which fixes

x_c = \frac{I_R - 0.401}{0.859\,I_R + 2.316}

taylor1998a R5, R6, E4, E5. Recalculated with this equation, the calibration mixtures gave slope 1.01, intercept −0.01 and r = 0.9996; the card notes that these are the very mixtures that set the slope, so the independent test is the validation set taylor1998a R7, X5. The validation mixtures gave slope 1.008 and intercept −0.0139, which the authors call "reasonably rugged" taylor1998a R8, C14. Table 1 lists the means: 98.7, 97.7, 96.5 and 93.5% for 99, 98, 97 and 93% crystalline, 96.3% for 96%, 73.6% for 75%, 54.0% for 55%, 21.3% for 25%, 5.9% for 7%, 2.2% for 3%, 1.0% for 2% and 0.9% for 1%, with standard deviations of 0.2–4.0 points taylor1998a R9. All seven compositions at or below 75% crystalline read low, by 0.1 to 3.7 points, and for 25, 7, 3 and 2% crystalline the whole five-run range lay below the actual value, whereas from 93 to 99% the means were within 0.5 points taylor1998a X4, R9. The 2% row (mean 1.0, SD 0.4, range 1.35–0.05) cannot be internally consistent for five runs taylor1998a X3. Ten measurements of each pure form gave a crystallinity SD of 0.2%, so the detection limit (3 SD) is 0.6% and the quantitation limit (10 SD) about 2% (synthesis: 3 × 0.2 = 0.6; 10 × 0.2 = 2); the authors find about 1% detection "reasonable" and the 2% limit too low because the mixture data vary more than the pure-form SD suggests taylor1998a R14, E7, E8, C21. Crystalline indomethacin milled for 5, 15 and 30 min gave spectra like those of the mixtures with more intensity at 1680 cm⁻¹ and Raman-equivalent crystallinities of 85.0, 77.5 and 63.5%; these are equivalents on a two-state calibration, particle size was not evaluated and no second technique was run on the milled samples taylor1998a R15, R16, C19, C20, X9.

Figure

The validation line has slope 1.008 and intercept −0.0139; look at how the points cluster at both ends of the range and at the low-end points that sit below the line taylor1998a R8, R9.

Why

The authors' explanation (their interpretation): the crystal has a narrow carbonyl band and the glass a broad one, so the crystalline contribution to the 1680 cm⁻¹ position is small (0.052) and the amorphous contribution to the 1698 cm⁻¹ position larger (0.401); the band-factor ratios are assumed to depend only on the scattering parameter of each band ("can be considered") taylor1998a C10, C11, C12, R5. For milled material they note that a physical mixture is a two-state model, each particle wholly amorphous or wholly crystalline, whereas mechanically processed material is more likely partly disordered throughout, so a mixture calibration can only give an equivalent value taylor1998a C17, X9. Not measured: the cause of the low readings at 2–25% crystalline, which the card lists as an open question taylor1998a X4.

Conditions and limits

One compound with two forms that differ in a carbonyl band; powders below 150 µm taylor1998a M1, M2; physical mixtures of two pure forms with no polymer; one composition (55%) for the error tests taylor1998a P8, P9, P10; no measurement of particle-size effects, which the authors warn can change Raman intensity, so standards should match the size distribution of the samples taylor1998a C19, C36, C37. Laser heating was tested only on a 55% mixture, without a sample temperature taylor1998a X10. The method was not tested in a drug–polymer dispersion.

Counterexamples and conflicts

None found in the library for the indomethacin result. In a polymer matrix the picture is harder: for naproxen in HPMCAS the Raman spectra of dispersions are "generally not simple to interpret", chemometrics was needed, and the Raman model built on blends gave an RMSEE of 2.22 against 7.4% for the model built on dispersion standards (F3) kestur2012 C25, R5, R15.

Transfer check

For another drug: (1) record the Raman spectra of the pure crystal and of the glass and look for a band pair, one for each form, and for their overlap; (2) build the calibration from physical mixtures with particle size matched to the samples; (3) validate on independent mixtures, especially below 25%, where this paper's readings ran low, and at the extremes; (4) take the detection limit from the mixtures, not only from the SD of a pure form; (5) average several spectra per sample (F2). A drug whose crystal and glass bands coincide, or whose polymer bands overlap them, is outside the evidence.

History

1998: Taylor and Zografi set the first benchmark in the library, about 1% detection of either form in a powder by FT-Raman, built on their 1997 spectra of the two indomethacin forms taylor1998a C5, C41. 2010: Rumondor and Taylor cite it as the earlier crystallinity quantitation in a pure drug before turning to dispersions rumondor2010b C5; Wanapun and co-workers take its "about 1%" as the limit of XRPD and spectroscopy that SHG should beat wanapun2010 R1, C1. 2012: Kestur and co-workers use Raman as a reference method with its limit "as low as 1%" for pure drug kestur2012 R17, C13.

How it was shown

The error tests followed the parameters of Bugay and co-workers for polymorph quantitation: (a) three consecutive scans of the 55% sample at 225, 450 and 900 mW, (b) ten consecutive scans without moving the sample, and (c) six independently prepared 55% mixtures, each removed from the instrument, shaken and rescanned five times, giving the range of the 30 values and, from them, the overall method RSD taylor1998a P8, P9, P10. The RSD is \mathrm{RSD}(\%) = \mathrm{SD}\times 100/\bar{x} taylor1998a E1.

What the data show

Absolute Raman intensities rose with laser power but the band ratio changed little: 54.4–56.6% across the three powers, a spread of 2.2 points (synthesis: 56.6 − 54.4), and 55.8–56.3% across ten scans, 0.5 points (synthesis: 56.3 − 55.8) taylor1998a R10, R11. Re-sampling the powder was a different order of magnitude: Table 2 lists values from 44.8 to 62.5%, a spread of 17.7 points (synthesis: 62.5 − 44.8 = 17.7), with sample means from 51.9 to 56.1%, SDs of the five scans of one sample from 1.4 to 5.3 points and SDs between the six samples at one scan number from 2.7 to 4.5 taylor1998a R12. The 30 values gave a mean of 54%, an SD of 4% and an RSD of 7.3% (recomputed 7.2% from the printed values, 7.4% from the rounded mean and SD) taylor1998a R13, X2. The authors read the near equality of the within-sample and between-sample spreads, together with the close sample means, as showing that the scatter comes from non-homogeneous mixing within a sample and not from non-reproducible preparation taylor1998a C25. They add that the standard deviations of Table 1 show no trend with crystallinity, so the RSD will depend on composition taylor1998a C26. Averaging helps: accuracy is said to improve from 5 measurements (one sample) to 15 (three samples) or 30 (six samples), though no composition was reported at all three levels and only Table 2 supports it at 55% taylor1998a C16, X6. One printed mean in Table 2 (3.1 for scan 2) is a misprint of 53.1 taylor1998a X1.

Figure

Values run from 44.8 to 62.5%; the scatter within one sample column is as large as the scatter between columns taylor1998a R12.

Why

The authors' explanation (their interpretation): the Raman beam samples only about 1 mm of the powder bed, so the area measured "can lead to sub-sampling", and how representative one scan is depends on particle size and mixing homogeneity (large particles near the spot size, wide size ranges, clusters of one component) taylor1998a C31. Re-mixing by agitation before each scan is not efficient compared with a rotating sample holder, and a device that averages different parts of the sample "would improve this method substantially" and lower the detection and quantitation limits, which depend on that variation taylor1998a C33, C34, C32. In processed material a distribution of disorder that depends on size reduction is "more likely" than uniform disorder, so the same sampling issue would apply to milled powder taylor1998a C35. None of this was tested with a different spot size or holder.

Conditions and limits

One composition, one drug, a spot of about 1 mm and a shaking protocol; the laser-power result holds for a mixture at up to 900 mW and says nothing about sample temperature taylor1998a X10. The 0.6% detection limit comes from the SD of a pure form, which contains no mixing error; at 55% the SD is about twenty times larger, while at 1–3% crystalline the printed SDs are 0.2–0.4 points taylor1998a R14, R9.

Counterexamples and conflicts

None found in the library. Later papers handled the issue by design rather than by testing it: the naproxen–HPMCAS Raman spectra were averaged over 6–10 regions of the powder bed in triplicate and the SHG images over three fields per sample, and SHG voxel intensities "can have a broad distribution" with crystal size and orientation kestur2012 P10, P6, C12; the flutamide and ezetimibe SHG means came from triplicate samples with three 1925 µm fields each correasoto2017 P3.

Transfer check

For another drug: (1) at the composition you care about, prepare several independent samples and rescan each after re-mixing; (2) compare the spread within and between samples before reading a single spot; (3) if the spread within a sample is as large as between samples, average more regions or use a rotating holder instead of more preparations; (4) do not quote a detection limit from the SD of a pure form alone.

History

1998: Taylor and Zografi measured the separate sources of error and named sampling as the largest taylor1998a C25. 2012 and 2017: the SONICC papers adopted multi-field averaging as part of the design kestur2012 P6; correasoto2017 P3, without repeating the within-sample against between-sample test.

How it was shown

Rumondor and Taylor (2010) asked whether the crystalline fraction of a drug in a partly crystallized dispersion is better measured from PXRD by PLS regression than by the ratio of Bragg-peak area to total area, because quantifying crystalline drug inside an otherwise molecularly mixed dispersion "has not been widely reported" rumondor2010b C5, C27. They made felodipine–PVP K29-32 dispersions by dissolving both in 1:1 ethanol:dichloromethane, rotary evaporating, drying under vacuum and grinding at below 10% RH, and confirmed by PXRD that they were X-ray amorphous rumondor2010b P1, R2, M2. Calibration samples were the dispersion plus crystalline felodipine, with an overall PVP content of 25, 50 or 75% w/w and 0, 10, 25, 50, 75 or 90% of the drug added as crystals, cryo-milled for 4 min, sieved through 150 µm and made in three independent replicates, 18 samples per PVP level rumondor2010b P2. This design rests on the earlier finding that the halo of a molecular-level dispersion is not a linear combination of the pure amorphous patterns, so the standards must contain the real dispersion rumondor2010b C8. PXRD (Shimadzu XRD-6000, Cu-Kα, 5–35° 2θ, 8°/min, 0.04° steps) was normalized to a same-day silicon standard and to unit area; the area method used Fourier-transform filtering to separate Bragg peaks from the halo and a trapezoidal total area, and PLS used SIMCA-P+ with two components per PVP level and a log transformation whose target is not stated rumondor2010b P3, P5, P6, P7, X4. A milling control showed that samples milled 2 and 6 min differed by less than 2% in calculated crystallinity rumondor2010b R5, C11.

What the data show

The area ratio is

\%\ \text{crystallinity} = \frac{100\,A_c}{A_c + A_a}

with A_c the Bragg-peak area and A_a the halo area between 5 and 35° 2θ rumondor2010b E1. Against known crystallinity it scattered widely: at 25% PVP the 90% samples came out at about 108–126%, and at 75% PVP the 0% samples ranged from about −12 to 14% and the 75% samples from about 47 to 119% (read from Fig. 2) rumondor2010b R7. The 95% prediction bands lay about ±21, ±39 and ±46 points from the fitted line (read from Fig. 2) rumondor2010b R8. Table 1 gives RMSEE of 11.2, 17.0 and 23.6% and R² of 0.93, 0.80 and 0.75 for 25, 50 and 75% PVP; the authors see the error grow as the Bragg-peak share of the pattern falls with polymer content rumondor2010b R9, C15. Other area-based calculations did not help (data not shown) rumondor2010b R10, C17. The PLS models gave RMSEE of 2.2, 1.9 and 4.7%, R² of 0.996, 0.997 and 0.978 and slopes of 0.998, 0.997 and 0.984; the gain is five- to nine-fold (synthesis: 11.2/2.2 = 5.1, 17.0/1.9 = 8.9, 23.6/4.7 = 5.0) rumondor2010b R9. PLS predictions lay close to the actual levels (90% samples at about 91–92% at 25% PVP; 75% samples at about 63–76% at 75% PVP) and the bands narrowed to about ±6, ±6 and ±10 points, the 75% PVP band being the widest, "most likely" because its Bragg peaks are weakest rumondor2010b R17, R18, C31. The fraction of the variation in crystallinity explained by the first component (R2Y) was 0.872, 0.948 and 0.511 at 25, 50 and 75% PVP, and the second component added 0.119, 0.0465 and 0.448 rumondor2010b R14. At 75% PVP the drug is 25% of the mass, so an RMSEE of 4.7% of the drug is about 1.2% of the sample (synthesis: 4.7 × 0.25 = 1.175) rumondor2010b E3, R19. Two cautions come from the card: the area-method RMSEE appears to describe scatter about a fitted line, because the raw root-mean-square error of Eq. 1 at 25% PVP read from the figure is about 26% against the tabulated 11.2%; and the Table 1 slopes are predicted-on-actual, the reciprocals of the lines drawn in Fig. 2 rumondor2010b X1, X2.

A second laboratory-scale data set points the same way for PXRD. Kestur and co-workers built OPLS models on cryomilled naproxen–HPMCAS blends and found PXRD slope 0.985, R² 0.981 and RMSEE 4.10, with 95% prediction bands of about ±8 units, no detectable peaks below 5 wt% naproxen, and values of about 101 and 108 at 100% crystalline; their Raman model on the same blends gave RMSEE 2.22 kestur2012 R5, R8, R9. Models built on dispersion-based standards were less precise: RMSEE 5% for PXRD and 7.4% for Raman kestur2012 R15. Correa-Soto and co-workers, working at low crystallinity with a fixed drug loading, used the narrow-window ratio A=\mathrm{AUC}_{\text{peak window}}/\mathrm{AUC}_{\text{total window}} and obtained linear XRPD calibrations (R² 0.99 over 1–5% for flutamide) correasoto2017 E3, R6; the area ratio in that work covered a small crystallinity range, not 0–90%.

Figure

Actual crystallinity is on the y-axis; compare the width of the prediction bands, which is much narrower for PLS rumondor2010b R8, R18.

Why

The authors' explanation (their interpretation): the area ratio assumes that the measured crystalline and amorphous intensities are proportional to the phase fractions, which fails when the halo of a molecular-level dispersion has its own shape; poor signal-to-noise, non-uniform samples and unequal scattering constants "may have" added error rumondor2010b C13, C18, C20. The first PLS component weighs Bragg peaks against the halo, which they read as Vainshtein's law (total diffracted intensity is conserved as drug passes from amorphous to crystalline at fixed composition); the second component is "interpreted as" shifts of the Bragg peaks and the halo plus slight peak broadening, and "It is thought" that these subtle variations, from small crystallites, strain or intimate mixing, explain the gain rumondor2010b C21, C22, C23, C32. Not measured: peak positions, widths or crystallite size; a one-component PLS model, although at 50% PVP the first component alone (R2Y 0.948) already exceeds the area-method R² of 0.80 rumondor2010b X6, X7. Slower scans to raise signal-to-noise were not an option, because Bragg peaks of amorphous–crystalline nifedipine mixtures grew between consecutive readings at 2 or 4°/min, read by the authors as X-ray-induced crystallization (data not shown) rumondor2010b R11, C19.

Conditions and limits

One drug–polymer pair, one PVP grade, one instrument and one milling protocol; the calibration crystals were cryo-milled, which may give strain or small crystallites that crystals grown in a stored dispersion do not have rumondor2010b X13. Within each model the overall drug:polymer ratio is fixed, so the amorphous phase loses drug as crystallinity rises (from 75% to about 23% drug at 25% PVP between 0 and 90% crystalline); a sample whose amorphous phase changes in another way, for example by water uptake, is outside the calibration rumondor2010b X8. Crystallinity here is per drug, not per sample (F4).

Counterexamples and conflicts

None found in the library for the ranking of PLS above the area ratio. The same card-level caveat applies as in the SONICC work: Kestur's dispersion-based PXRD model (RMSEE 5%) was less precise than the blend model (4.10), so a calibration that matches the dispersion does not remove error kestur2012 R5, R15. Sensitivity is a separate matter: PXRD showed no peaks below 5 wt% in the naproxen blends and none below 1% in the flutamide standards, and a calibration error of 2–5% is not a detection limit (F7) kestur2012 R8; correasoto2017 R7.

Transfer check

For another drug: (1) measure the crystalline drug in the real dispersion, not a mixture of amorphous drug, crystals and polymer; (2) keep the overall drug:polymer ratio of the test samples and spike crystals into the dispersion; (3) report the error on samples left out of the calibration, which this paper did not; (4) state crystallinity per drug or per powder and the drug loading (F4); (5) check that the crystals of the test samples resemble the calibration crystals in size and strain.

History

2009: the library already held the point that the halo of a molecular-level dispersion is not the sum of the pure halos, which motivated the design rumondor2010b C8. 2010: Rumondor and Taylor built the dispersion-plus-crystals calibration and the PLS model. 2012: Kestur and co-workers took over dispersion-based standards for PXRD and Raman, with OPLS, as the reference against which SONICC was judged kestur2012 C23, P4, P9, P11. 2017: Correa-Soto and co-workers returned to peak-window ratios for low-level PXRD calibration correasoto2017 E3.

How it was shown

This finding is read across papers. For each one the question was what the percentage in the abstract refers to and how its limit was obtained. Taylor and Zografi use x_c, the mole fraction of the crystalline form in a binary mixture of two pure forms, and take the limit from 3 and 10 times the standard deviation of ten measurements of a pure form taylor1998a E3, E7, E8. Wanapun and co-workers define a crystallinity limit as the volume of the smallest detectable crystal over a probed volume of 600\times600\times60\ \mu\text{m}^3 wanapun2010 E3. Rumondor and Taylor define it per drug and convert by the drug mass fraction rumondor2010b E3, X12. Kestur and co-workers use two bases within one paper and a limit from the single-voxel signal at signal-to-background 3 kestur2012 X11, E9, P13. Correa-Soto and co-workers define "% crystallinity" as the weight per cent of crystalline drug in the whole powder, give the conversion to per cent of the drug crystallized, and compute limits from the residual standard deviation of the calibration (ICH Q2(R1)) correasoto2017 E1, E5, P9. Two papers outside the chapter show the same pair: posaconazole residual crystallinity reported both as wt% of the ASD and as wt% of the drug, and a BCL/PVPVA hot-melt extrusion study that does not state which basis its "% crystallinity" uses shah2024 R10; moseson2021 X7.

What the data show

In the table, the first block is the conversion as printed and the second my own conversion of limits that the papers give in one basis.

The limit definitions differ as much as the bases. A 3 SD or 10 SD criterion on a pure form (Taylor) contains no sampling error; a regression-residual limit (Correa-Soto) contains the scatter of the standards but needs standards at low levels; a single-voxel limit (Kestur) is a calculation on a sample without crystals, whose lowest imaged level was 0.1%; and the size-based limit of Wanapun and co-workers depends on an unprinted scaling from signal-to-background to crystal size taylor1998a E7; correasoto2017 P9; kestur2012 X10; wanapun2010 X3. The authors of the SONICC papers themselves flag the consequences: Correa-Soto and co-workers say "readily detected" for 0.005% and 0.01% from visual inspection of standards, whereas the computed limits are 0.03–0.047% correasoto2017 C17, C36, X7. The 2.9 × 10⁷-fold gain over conventional methods in the SHG film paper uses a 0.1–10% range for conventional limits without a reference, and its upper bound has an arithmetic slip wanapun2010 R12, X1.

Figure

The y-axis is per cent of the drug crystallized, not of the powder; at 60 days and 82% RH the SHG value (about 11%) sits next to the XRPD value (grey dot, about 12%), which is about 3.5% of the powder (synthesis: 0.115 × 0.30) correasoto2017 R21, R22.

Why

Not a mechanism but a convention problem. The number of crystalline mass per powder mass equals the per-drug figure times the drug loading, so at drug loadings of 15–30% the same crystal content looks 3.3–6.7 times smaller on a powder basis than on a drug basis (synthesis: 1/0.30 = 3.3 to 1/0.15 = 6.7) correasoto2017 E1; rumondor2010b E3; kestur2012 X11.

Conditions and limits

Only printed values were converted; the dispersions differ in drug, polymer and loading, so the converted limits are not equal in meaning, only comparable in basis. The ezetimibe conversion is conditional, and for the BCL/PVPVA paper the basis is simply unknown correasoto2017 X6; moseson2021 X7.

Counterexamples and conflicts

None found; no paper claims that the bases are interchangeable. The conflicts are implicit: the 10 ppm of the naproxen work and the 0.03–0.047% of the flutamide and ezetimibe work differ by a factor of 30–47 (synthesis: 0.03/0.001 and 0.047/0.001) and were obtained with different definitions, drugs, loadings and instruments (F7) kestur2012 R7; correasoto2017 R8.

Transfer check

For another drug: (1) write down whether a percentage is mole, mass, volume, per drug or per powder, and the drug loading; (2) convert to crystalline drug mass per gram of powder before comparing with another paper; (3) state which definition a limit follows (3 SD of a pure form, regression residual, signal-to-background 3, visual) and test it with standards made in your matrix at levels near the limit.

History

1998: Taylor and Zografi use mole fractions of two pure forms. 2010: Wanapun and co-workers express a limit as a volume fraction; Rumondor and Taylor express errors per drug and convert once to mass. 2012: Kestur and co-workers carry two bases in one paper. 2017: Correa-Soto and co-workers define a powder basis and state the conversion; 2024: Shah and co-workers print both bases in one sentence wanapun2010 E3; rumondor2010b E3; kestur2012 X11; correasoto2017 E1; shah2024 R10.

How it was shown

Wanapun and co-workers asked whether SHG microscopy (SONICC) detects the first crystals in an amorphous drug earlier and more selectively than optical microscopy wanapun2010 C1, C3, C4. Griseofulvin (Hawkins, Tg 88.9 °C, intrinsically chiral) and chlorpropamide (Sigma-Aldrich, Tg 15.1 °C, achiral) were melted as 1–3 mg onto a cover glass, held molten for 2–4 min and quenched on a cold surface under a nitrogen purge; a film was accepted as amorphous when polarized light showed no birefringence wanapun2010 M1, M2, P1, P2. Films were held above Tg, griseofulvin at 120 °C and chlorpropamide at 30 °C, while a custom beam-scanning microscope (800 nm, 100 fs, 80 MHz, 40 mW, about 0.5 nJ per pulse; 10× objective, N.A. 0.3; depth of field about 60 µm; field 600 × 600 µm) recorded forward and backward SHG every 45 s (griseofulvin) or 25 s (chlorpropamide), and a CCD camera recorded bright-field images of the same field wanapun2010 P3, P4, P5, P6. Images were thresholded at a signal-to-background of 3 wanapun2010 P7. The authors argue that imaging did not perturb crystallization because both drugs are transparent at 800 and 400 nm and the pulses are far below the microjoule pulses known to affect nucleation; no laser-off control was run wanapun2010 C5, X12.

What the data show

In film (a), neither channel showed crystals at 0 min; one SHG-active domain appeared at 7.5 min with nothing in the optical image, about two domains at 13.5 min and about ten at 75 min, while the optical image showed no crystals until after 39 min and then only with low contrast, a gap of more than 31.5 min (synthesis: 39 − 7.5) wanapun2010 R2. Crystals that started later were also seen first by SONICC wanapun2010 R3. For chlorpropamide SHG appeared at several crystal centres after 125 s and optical microscopy first showed crystals after 500 s, a gap of about 375 s or 6 min (synthesis: 500 − 125 = 375 s, about 6.25 min); the film crystallized within about 30 min wanapun2010 R5. In film (b), prepared and monitored under comparable conditions, the optical image showed no crystals at 0 min but SONICC showed several submicrometre crystals scattered across the field; their number rose at 1.5 and 3 min, and by 12 min the SHG field was densely covered and the optical image showed crystal texture wanapun2010 R4. The authors state that crystallization with seeds was faster than in samples where none were detected; the comparison rests on film (a), which held about ten SHG domains at 75 min, and film (b), which was densely crystallized by 12 min wanapun2010 C13, R2, R4. They put the usual limits of other methods at about 1% crystallinity for XRPD and spectroscopy wanapun2010 R1.

Figure

White is SHG above signal-to-background 3; in row (a) one crystal is circled at 7.5 min while the optical image is blank, and in row (b) SHG points are present at 0 min wanapun2010 R2, R4.

Why

The authors' explanation (their interpretation): SHG arises only from order without an inversion centre, so a glass gives no coherent SHG and any bright pixel is a crystal wanapun2010 C4. Optical microscopy can tell a likely crystal from a scattering centre only above 5 µm, so the lag before crystals are seen, usually ascribed to nucleation, depends on the sensitivity of the method wanapun2010 C2, C21. Crystals at 0 min in film (b) are read as residual seeds that formed "presumably" through incomplete melting or entrainment of nuclei during cooling, and seeds are "one of the major factors" in the observed rate, especially early wanapun2010 C8, C9, C10. They conclude that SONICC "provides a selective way to verify" that a preparation destroyed all crystalline structure wanapun2010 C11. Not measured: melt temperature and hold time as variables, rate constants or crystal counts of the seeded against the unseeded film, and any structural identification of the crystals wanapun2010 X10, C9.

Conditions and limits

Two drugs, thin films without polymer, one film of each kind shown for the timing and no replicate count printed. Griseofulvin is intrinsically chiral; chlorpropamide has two SHG-active (α, γ) and two SHG-inactive (β, δ) polymorphs, so absence of SHG cannot prove that all crystalline structure was destroyed (F11) wanapun2010 C23, R13, X11. Film thickness, melt temperature and the cold-surface temperature are not printed wanapun2010 P1. The usual acceptance test for a melt-quenched glass, no birefringence under polarized light, did not exclude the seeds in film (b) (synthesis: from P2 and R4).

Counterexamples and conflicts

None found in the library. The seeded-faster reading is a single qualitative comparison, and the kinetics samples of the same paper do not say whether they held seeds at 0 min (F6) wanapun2010 X10, X13.

Transfer check

For another drug: (1) confirm that its crystal form is SHG-active (a non-centrosymmetric space group; F11); (2) image the freshly prepared glass, not only a polarized-light check, before starting a storage or inhibitor study; (3) record the time of the first SHG crystal and the first optical or XRPD crystal separately; (4) treat a lag time from a less sensitive method as an upper bound on the true onset.

History

2010: Wanapun and co-workers showed in the library that a crystal-selective image detects crystals earlier than optical microscopy and can reveal seeds that polarized light misses. 2012: Kestur and co-workers took the method from films to powders and dispersions and cited from it the 90 nm detection, the first evidence of linear scaling (a few PXRD comparisons) and the 100 nm coherence length kestur2012 R17, C3. 2017: Correa-Soto and co-workers applied it to spray-dried dispersions on storage (F10).

How it was shown

The same SONICC set-up as in F5 imaged griseofulvin films at 125 °C for several hundred minutes wanapun2010 P3, P4. ImageJ found contiguous areas above three times the noise in each frame and gave the area and mean SHG of every crystal, compiled as a "growth map"; growth rate is the slope of the square root of area against time in windows without coalescence; each detected particle was taken as one nucleation site, the count corrected whenever it fell because of coalescence and converted to volume with the 60 µm depth of field; the SHG integrated over the field, relative to the end of the experiment, gave the fraction crystallized α, fitted with the JMA equation at n = 2 wanapun2010 P7, P8, P9, C31, C33. Rate constants at five temperatures, read from the plot as about 115–130 °C because the temperatures are not printed, gave the activation energy from an Arrhenius plot wanapun2010 R19, X5.

What the data show

The fit equation is

\alpha = 1-\exp\!\left(-\big(k\,(t-t_0)\big)^{n}\right),\qquad \frac{\mathrm{d}\ln k}{\mathrm{d}(1/T)} = -\frac{\Delta E}{R},

with n=2 wanapun2010 E4, E5. At 210 min three sites covered 1400, 5300 and 109 × 10³ µm² with mean SHG of 421, 829 and 1685 counts; the largest trace reached about 350 × 10³ µm² by 600 min wanapun2010 R14. Growth rates from the square-root-of-area slopes were 1.54 ± 0.02 µm/min for a single crystallite (window about 78–129 min), 1.173 ± 0.005 µm/min after that crystallite merged with another, and 1.16 ± 0.02 µm/min for a third crystallite (standard errors of the fits) wanapun2010 R15, R16. The coalescence-corrected number density rose stepwise from about 45 mm⁻³ at about 62 min to about 600 mm⁻³ at about 265–285 min, and a linear fit gave 2.85 ± 0.03 mm⁻³ min⁻¹ crossing zero at about 70 min wanapun2010 R17. The fraction crystallized was about 0 until 50 min, 0.2 at 200 min, 0.5 at 290 min, 0.9 at 510 min and about 1 at 750 min; the fit gave k = 3.35(±0.01) × 10⁻³ min⁻¹ wanapun2010 R18. The Arrhenius slope was (−1.7 ± 0.2) × 10⁴ K, so ΔE = 140 ± 20 kJ/mol (synthesis: 1.7 × 10⁴ × 8.314 = 141 kJ/mol), against 167 kJ/mol by DSC, about 1.4 standard deviations away wanapun2010 R19, R20, X14. Caveats from the card: the ±0.03 on the nucleation rate is a fit error, the steps of about 46 mm⁻³ equal one crystallite in the 0.0216 mm³ field, so the rate rests on about 13 crystals and counting statistics alone give about ±28%; the exponent was fixed at 2 without fit statistics, and the extractor notes that constant nucleation with linear growth would usually imply n = 3 or 4 (the extractor's background knowledge, not the paper's); the 125 °C point of the Arrhenius plot (ln k about −5.4) does not match the text value (ln k = −5.70) wanapun2010 X9, X8, X6.

Figure

Panel (a) shows the growth of a single crystallite, (b) the stepwise crystal count with its linear fit and (c) the S-shaped fraction crystallized wanapun2010 R16, R17, R18.

Why

The authors' explanation (their interpretation): low background and high contrast make real-time monitoring of nucleation and growth "readily achievable", with nucleation rate, growth rates and overall kinetics from one sample wanapun2010 C27. The drop in growth rate after coalescence "may be described by" the loss of surface on which growth occurs; differences between crystallites "may arise from" intrinsic variability and orientation wanapun2010 C29, C30. For dendritic crystallites the mean SHG intensity is taken as proportional to growth in the vertical direction, but crystallite thickness was not measured wanapun2010 C28. The agreement of the activation energy with the DSC value "suggests" that SONICC and DSC probe similar ensemble-averaged phenomena, while SONICC adds data on individual crystallites wanapun2010 C34.

Conditions and limits

One drug, one run at 125 °C for the rates wanapun2010 P3, no polymer, and no statement of whether these samples held seeds at 0 min, although seeds sped crystallization in F5 wanapun2010 X13. The five temperatures and the rate constants other than 125 °C are not printed wanapun2010 X5. The method sees only SHG-active crystals (F11).

Counterexamples and conflicts

None found in the library. The activation energy lies outside the SONICC uncertainty band of the DSC value by about 1.4 standard deviations, which the card reads as approximate agreement wanapun2010 X14.

Transfer check

For another drug: (1) confirm that the crystal is SHG-active and that SHG intensity rises monotonically with crystal size in your system (chlorpropamide did not; F11); (2) count enough crystals for a rate, since about 13 gave an uncertainty near ±28% by counting alone; (3) fit n instead of fixing it and compare k and ΔE only with fits that treat n the same way; (4) say whether seeds were present at the start.

History

2010: Wanapun and co-workers made this the only study in the library that extracts nucleation, growth and JMA kinetics from one SHG stack. 2012: Kestur and co-workers used SONICC for conversion kinetics in powders at 60 °C, comparing it with PXRD and Raman, and noted "potential utility" as an online tool, without extracting nucleation or growth rates (F10) kestur2012 R16, C30.

How it was shown

This finding collects limits from the papers of the chapter and from the ASD-process and bioperformance papers that apply the same methods. The SHG limits were obtained in four ways: size-based back-calculation in films (Wanapun), a single-voxel calculation on a crystal-free sample (Kestur), the residual standard deviation of a regression (Correa-Soto), and visual inspection of low-level standards (Correa-Soto) wanapun2010 P10, E3; kestur2012 P13, E9; correasoto2017 P9, R5. The XRPD limits came from calibration curves with spiked standards, sometimes with a signal-to-noise criterion (3:1 and 10:1) moseson2021 P2; shah2024 P6; hate2019 R2.

What the data show

In naproxen–HPMCAS blends, 3D SHG images of 0.1% crystalline naproxen in 99.9% HPMCAS showed about six discrete crystals; the calculated limit is 10⁻³ % because the polymer gives a small but nonzero background, which the authors use to explain why the limit is higher than the parts-per-trillion level of amorphous griseofulvin films kestur2012 R6, R7, C10, C29. For the film work the limit is a back-calculation: a signal-to-background of 30 (griseofulvin) or 5 (chlorpropamide) at the point where forward and backward SHG separate, scaled to 3, gave crystals of 90 and 125 nm and limits of 3.4 × 10⁻⁹ % and 9 × 10⁻⁹ % by volume, against more than 6 × 10⁻⁴ % for optical microscopy, with the scaling not printed and a probed depth of 60 µm assumed wanapun2010 R10, R11, X2, X3, X4. In the spray-dried dispersions the SHG calibration fits had R² 0.98 and the XRPD fit R² 0.99, and the computed limits and quantitation limits were 0.03 and 0.08% (SHG) and about 0.4 and 1% (XRPD) for flutamide and 0.047 and 0.14% (SHG) and 0.7 and about 2% (XRPD) for ezetimibe correasoto2017 R6, R8. The authors' headline, that less than 0.01% crystallinity was readily detected against 1% and higher for XRPD, rests on visual inspection, while the computed limits are higher correasoto2017 C36, X7. Posaconazole extrudates made by hot-melt extrusion at 125 °C, below the temperature at which the drug dissolves completely in the polymer, carried 0.3, 3.2 and 19.3 wt% crystalline drug by SHG at 10, 25 and 50 wt% drug loading, while pXRD (LOD 0.3 wt%, LOQ 1.2 wt%) gave 2.4% at 25% loading and "below the limit of detection" at 10% shah2024 R10, R9, P6, X4. For a BCL–PVPVA extrudate series with an XRPD LOD of about 1% and LOQ of about 3% (signal-to-noise 3:1 and 10:1), Table I classifies S/180 as no crystals by any method, S/155 (below 1% by XRPD) as seen only by TEM, M/155 (3.5%) as seen by polarized-light microscopy and TEM but not DSC or SEM, and L/155 (30%) as seen by all methods moseson2021 P2, R6, X6. In an indomethacin–PVPVA extrudate with only a halo by XRPD and one Tg by DSC, TEM found discrete crystals below 100 nm and clusters of 5–10 nm domains moseson2018a R5, R9, C10; in a series of extrudates polarized-light microscopy found birefringence where XRPD saw an amorphous sample, and DSC detected a crystal dissolution endotherm (above about 1.5 J/g) only in four 2-min batches at or below 131 °C moseson2018 C27, R11, R12, C31.

Figure

The initial dispersion is dark, 0.005% crystalline drug shows sparse bright spots, and the density of spots rises through 0.01–4% to a dense field at 15% correasoto2017 R5.

Why

The authors' explanation (their interpretation): amorphous drug and polymer give almost no net SHG, so the background is the polymer's small nonzero signal and crystals stand out even in a powder that is 99.9% polymer, whereas in PXRD and Raman the crystal signal is diluted by the polymer kestur2012 C4, C8, C29, C10. XRPD sensitivity is further limited by peak broadening of nanometre crystallites and by strain and defects, so quantities from XRPD "may be systematically underestimated" moseson2018a C26, C27. DSC identifies crystals positively but insensitively because crystals present at the start can dissolve into the molten matrix during the heating run moseson2018 C31. Not measured: the intensities behind the 10 ppm calculation, the polymer background, and the scaling from signal-to-background to crystal size kestur2012 X10; wanapun2010 X3.

Conditions and limits

Limits depend on the polymer (background), the drug loading, the instrument, the sample thickness and the definition; the SHG limits hold only for SHG-active crystals above about 100 nm (F11). The XRPD limits of this table are for different drugs and sample forms and range, in the literature cited by Correa-Soto and co-workers, from about 0.2% (synchrotron with special processing) to 5–7% (standard laboratory instruments) correasoto2017 R25.

Counterexamples and conflicts

The 10 ppm of the naproxen work and the 0.03–0.047% of the flutamide and ezetimibe work differ by a factor of 30–47 (synthesis: 0.03/0.001 and 0.047/0.001); the first is a single-voxel calculation, the second a regression-residual limit, and the sample, drug loading and instrument differ kestur2012 R7; correasoto2017 R8. In the posaconazole paper the pXRD result at 10 wt% loading is stated three ways (below the quantification limit, below the detection limit, and "as low as 0.3%" although 0.3% equals the stated LOD) shah2024 X4.

Transfer check

For another drug: (1) say which definition a limit uses and test it with standards in your own matrix at levels around the limit; (2) expect an XRPD limit of about 0.3–1% and a quantitation limit of 1–5% in a polymer matrix, and higher in a capsule blend; (3) if a formulation must be shown crystal-free below that, add an orthogonal method that responds differently, such as SHG for SHG-active forms, polarized-light microscopy or TEM, and report which one saw what; (4) do not equate "below the XRPD limit" with "amorphous".

History

1998: Taylor and Zografi set a Raman limit of 0.6% with about 1% detected for a pure drug taylor1998a R14, C21. 2010: Wanapun and co-workers gave a model-based SHG limit near 10^{-9} % in films wanapun2010 R11. 2012: Kestur and co-workers gave 10 ppm for a drug in a polymer and no PXRD peaks below 5 wt% kestur2012 R7, R8. 2017: Correa-Soto and co-workers gave computed limits of 0.03–0.047% with visual detection at 0.005% correasoto2017 R8, R5. 2018: Moseson and co-workers showed by TEM and polarized-light microscopy that crystals below the XRPD limit exist in extrudates moseson2018a R9; moseson2018 C27. 2021 and 2024: XRPD, DSC, SEM, polarized-light microscopy and TEM compared on one extrudate series, and SHG with spiked standards in a hot-melt-extruded posaconazole ASD moseson2021 R6; shah2024 P7.

How it was shown

Kestur and co-workers asked whether SONICC can quantify crystalline drug in drug–polymer powders and whether its signal scales linearly with crystallinity, a point that was open because linear scaling in pure powders rested on a few PXRD comparisons and SHG from pristine crystals can scale quadratically kestur2012 C3, C5. Crystalline naproxen (Sigma-Aldrich) and HPMCAS (AQOAT AS-MF) were cryomilled together (SPEX 6750, 8 Hz, 5 min equilibration, cycles of 2 min milling and 1 min cooling, sieved below 105 µm) at 0–100% crystalline naproxen in triplicate, plus a prediction set at 3.5, 18 and 83% kestur2012 M1, M2, P1, A1, A2. SONICC (800 nm, 100 fs, 80 MHz, 20 mW, 10× objective, N.A. 0.3) imaged 1–2 mg of powder between a slide and a cover glass with 150 µm spacers in three fields, using backward (epi) SHG for quantification; the relative crystallinity was the field-averaged SHG minus the near-zero HPMCAS background over the same quantity for 100% crystalline naproxen kestur2012 P6, P7, E8. PXRD (full-pattern OPLS, SIMCA P+) and Raman (785 nm, 6–10 regions averaged, OPLS on 3120–500 cm⁻¹) were calibrated on the same blends kestur2012 P8, P9, P10, P11. A 75 wt% HPMCAS–25 wt% naproxen dispersion was made by rotary evaporation from 1:1 dichloromethane:ethanol; for it, PXRD and Raman were calibrated on the dispersion plus crystalline naproxen, while SONICC was normalized to the 75/25 physical mixture kestur2012 P2, P4, P7.

What the data show

Calibration points lay close to the identity line, for example about 3.5 and 5 at 5%, 13–15 at 15%, 44–52 at 50% and 88–97 at 90%, with 95% prediction bands of about ±4.5 units over 0–100% (SHG), about ±8 (PXRD) and about ±2.5 (Raman), read from the figures kestur2012 R4, R8, R9. The abstract quotes R² 0.99 and a "root mean square error of prediction" of 2.7% over 0.1–100 wt%, which the card reads as the rounded RMSEE; the text gives slope 0.92 for 0–25% against 0.941 in the table kestur2012 X4, X3. Calibrations were on weighed compositions that were not checked after milling kestur2012 X12. In the prediction set the SHG errors were about 10% at the lowest level and 5% or less at the others; PXRD gave 16.70 and 85.73 at 18 and 83% (no value is printed at 3.5%) and Raman 3.74, 19.92 and 88.75 kestur2012 R10, C15. In the spike test the stated recovery of about 93% is 95.1% (39.1/41.1) or 89.4% ((39.1 − 22.2)/(41.1 − 22.2)) depending on the definition, and the amount added and the storage history of the sample are not printed kestur2012 R11, X5, X13. The PXRD and Raman models built on dispersion standards had RMSEE of 5% and 7.4%, so crystallinity could not be estimated by either before the 2 h time point of the storage test (F10) kestur2012 R15, C17. Correa-Soto and co-workers fitted their SHG standards of flutamide in two ranges, 0.005–0.1% (R² 0.98, slope 1.4, intercept 0.01) and 0.5–5% (R² 0.98, slope 1.5, intercept 0.08); Shah and co-workers drew a linear SHG standard curve with R² 0.999 from posaconazole ASDs spiked at 0–25% correasoto2017 R6; shah2024 P7.

Figure

The points lie on the straight line over 0–100% with the 0–25% range enlarged in the insert; dashed lines are the 95% prediction bands kestur2012 R4.

Why

The authors' explanation (their interpretation): the net SHG of N sources depends on their phase relation, so that

I_{c}^{2\omega}=\left|E_a^{2\omega}+E_b^{2\omega}\right|^2 \quad\text{(in phase)},\qquad I_{i}^{2\omega}=\left|E_a^{2\omega}\right|^2+\left|E_b^{2\omega}\right|^2 \quad\text{(random phase)},

where the card notes that the paper prints source a twice kestur2012 E5, E6, X1. In-phase sources give a signal that rises with the square of their number density and, for a pristine crystal growing beyond the coherence length of about 100 nm along the beam, with volume to the power 4/3; lattice defects over more than 100 nm or polycrystals make each domain an independent source and the signal linear kestur2012 C5, E7. A straight calibration is therefore read as incoherent combination, and the authors state that "it is clear" that SHG scales linearly in these dispersions and "reasonable to expect" this to be the general trend kestur2012 C3, C20, C21. The scaling "may depend sensitively" on the crystals formed, whose perfection and domain size were not measured; SHG against laser power at the measured crystallinities was not measured either, and the 20 mW used was chosen so that the dynamic range stays linear kestur2012 C5, C6. Because amorphous material gives no SHG signal, SONICC can be calibrated with pure crystalline drug or a crystalline blend, without chemometrics, but this was not tested against dispersion standards kestur2012 C27, X15.

Conditions and limits

One drug, one polymer, one drug loading (25%) for the dispersion kestur2012 P2, cryomilled sieved powders below 105 µm kestur2012 P1. The authors warn that standards must match the particle size range of the samples and that crystals of a different size in the dispersion shift the reading; the random-orientation assumption "might not" hold for very sparse early crystals kestur2012 C26, C31, C33. Blend values are per cent crystalline naproxen in the blend, dispersion values per cent of the drug (F4) kestur2012 X11. Only SHG-active crystals are seen (F11).

Counterexamples and conflicts

Flutamide in a spray-dried dispersion: the SHG signal of crystals grown in the dispersion exceeded that of the fully crystalline standard, so the calibration could not be applied (F9) correasoto2017 R9, C20. The linearity result therefore holds where the calibration crystals are the crystals that grow.

Transfer check

For another drug: (1) make standards from the crystal form and size range that the sample will contain, in the same polymer and at the same drug loading, by spiking the amorphous dispersion; (2) check that a sample known to be fully crystalline gives a ratio of 1, not more (F9); (3) test linearity by regression in the dispersion itself, not only in the blend; (4) keep laser power in the range where the signal is linear.

History

2010: Wanapun and co-workers gave the first evidence of linear scaling in powders, from a few PXRD comparisons, with a coherence length of about 100 nm kestur2012 C3, R17. 2012: Kestur and co-workers extended it to a calibration over 0–100% with a test set, a spike test and PXRD and Raman in parallel kestur2012 R4, R5, R10, R11. 2017 and 2024: Correa-Soto and co-workers and Shah and co-workers used spiked standards for flutamide, ezetimibe and posaconazole, with linear fits in limited ranges correasoto2017 R6; shah2024 P7.

How it was shown

Correa-Soto and co-workers spray dried flutamide (15% drug) and ezetimibe (30% drug) with HPMCAS AS-MF from methanol, stored the powders open at 40 °C and 75 or 82% RH for up to 60 days, and followed them by SHG (1064 nm femtosecond laser, 50 mW for flutamide and 250 mW for ezetimibe, triplicate samples, three 1925 µm fields each), XRPD, 19F direct-polarization ssNMR on selected samples and SEM correasoto2017 P1, P2, P3, P4, P5, P11. Standards were made by mixing the amorphous dispersion with a mixture of crystalline drug and HPMCAS of the same composition, so that only the crystalline fraction changed; flutamide SHG intensities were divided by that of the 15% crystalline standard, taken as the fully crystalline sample, and ezetimibe intensities were converted to per cent of drug crystallized with the calibration correasoto2017 P6, P10, E2, C9. Flutamide is not chiral but crystallizes in a non-centrosymmetric space group (Pna2₁, form I), so its crystals emit SHG correasoto2017 C8, M1.

What the data show

For flutamide the intensity ratio at 75% RH was about 0.0003 at 4 days, 0.012 at 29 days, 0.10 at 48 days, 0.14 at 53 days, 0.69 at about 57 days and 1.03 at 60 days, and at 82% RH about 0.49 at 29 days, 1.10 at 48 and 53 days and 3.1 at about 57 and 60 days; the 60-day image at 82% RH was visibly brighter than the 15% standard, and after 58 days the detector saturated, so the values near 3.1 are lower bounds correasoto2017 R9, R10, R11, X10. XRPD gave 1–2% crystallinity for samples of days 35–53 (relative humidity not stated in the sentence) and 19F ssNMR gave 1, 3.8 and 4.6% at 35, 39 and 48 days; the ssNMR humidity is stated as 75% in one sentence and as "the higher relative humidity" in the next, and the ssNMR values at 39 and 48 days are 1.9 and 2.3 times the top of the XRPD range (synthesis: 3.8/2 and 4.6/2) correasoto2017 R12, R14, X3, X4. Read linearly against the 15% standard, the 82% RH ratios of 0.6–1.1 at days 35–53 would correspond to about 9–16% crystallinity, which no method supports correasoto2017 X4. The authors call applying the calibration "impractical" and, because XRPD ruled out a new polymorph, turn to habit correasoto2017 C20, C23. SEM showed that the fresh dispersion consists of collapsed, dimpled particles and the as-received flutamide of large blade-shaped crystals, whereas after 60 days at either humidity there were long, thin, often bent needles several times longer than the 40 µm scale bar, one with a seemingly hollow end and one circled growing outward from a particle surface correasoto2017 R15, R16, C25. For ezetimibe, whose crystals are less SHG-active (hence 250 mW), the SHG-derived fraction at 82% RH was about 0.4% at 26 days, 3.7% at 53 days and 11.3% at 60 days (text about 11.5%), next to XRPD at about 12%, a crystalline mass fraction of about 3.5% of the powder (synthesis: 0.115 × 0.30 = 3.45), and ssNMR 17%, about 1.5 times the other two; at 75% RH it stayed below 1%, about 0.60% at 60 days correasoto2017 R17, R21, R22, R23, C29, X5. Clear ezetimibe crystals appeared by SEM only after storage at 82% RH correasoto2017 R24.

Two further results in the chapter fit the pattern that the signal per mass depends on the crystals. In chlorpropamide films the forward and backward SHG of a pixel levelled off or dipped before rising again, "tentatively attributed" to conversion between SHG-active (α, γ) and SHG-inactive (β, δ) polymorphs, which no structural method checked wanapun2010 R7, R13, C24. Kestur and co-workers note that larger domains give more SHG up to about the coherence length, that crystallite size in the dispersion may differ from that in the standard, and that the random-orientation assumption might not hold for sparse crystals kestur2012 C31, C32, C33.

Figure

Panels A and B are 75% RH and C and D 82% RH, so the steep rise above 1 is visible in B and D; the circles in the SEM image mark a needle growing from a particle (A), a seemingly hollow end (B inset) and a bent needle (C) correasoto2017 R9, R16.

Why

The authors' explanation (their interpretation): SHG depends on crystal orientation because crystals are anisotropic, the total signal of a powder sums over the orientations present, and a habit with a different orientation distribution, for example by preferred orientation, "could" make the signal quite different for two morphologies of one polymorph; hence "it appears clear" that the ASD-grown crystals are more SHG-active than the calibration crystals, and the dramatic change in morphology "is assumed to be responsible" correasoto2017 C22, C24, C38. Hollow organic crystals have been reported only from solvent-mediated phase transformations; the authors say such a route is "clearly" not responsible here and "further studies are needed" correasoto2017 C39. They present the result as a calibration problem, a perennial concern for any system crystallizing in a complex environment, not as an inability of the technique to quantify correasoto2017 C40. Not measured: SHG of single ASD-grown and reference crystals, the orientation distribution, and the hollowness beyond one end view; the surface-origin of needles rests on circled examples at one time point correasoto2017 X13, X14.

Conditions and limits

Two drugs, one polymer, 40 °C, two humidities correasoto2017 P2; SHG settings, drug loading and data treatment differ between the drugs, so the comparison is within each drug correasoto2017 X9. The SHG ratios at 82% RH are at detector saturation, the error bars are not defined, and the calibration data, ssNMR spectra and the 82% RH XRPD patterns are in supporting information that the card could not read correasoto2017 X10, X11, X16.

Counterexamples and conflicts

The three methods disagree for flutamide (SHG ratio up to 3, XRPD 1–2%, ssNMR up to 4.6%), and ssNMR is 1.5 times SHG and XRPD for ezetimibe, so the "reasonable agreement" for ezetimibe rests on one time point at which both SHG and XRPD quantify correasoto2017 R9, R12, R14, X5. In naproxen–HPMCAS the calibration crystals and the crystals in the dispersion agreed only over about 24–54% crystallinity, with no regression (F8) kestur2012 X8.

Transfer check

For another drug: (1) prepare calibration crystals by crystallizing the drug in the formulation, not only from the as-received lot, and compare their SHG per mass with the lot; (2) a fully crystallized sample must give a ratio of at most 1, and a ratio above 1 means the calibration cannot be used for quantitation; (3) keep the detector in its linear range; (4) check one time point with an orthogonal method (XRPD, ssNMR) before reading SHG as an amount; (5) otherwise use SHG for detection and ranking only.

History

2012: Kestur and co-workers raised the dependence on crystal size and orientation as a caveat kestur2012 C31, C33. 2017: Correa-Soto and co-workers met it as a quantitative failure for flutamide and a success for ezetimibe, and linked it to habit. The habit link and the surface-growth observation have no later test in the chapter.

How it was shown

Kestur and co-workers stored the naproxen dispersion at 60 °C and withdrew powder at 0, 1, 2, 4, 6, 8, 12, 24 and 48 h for SONICC, PXRD and Raman; humidity and container are not printed kestur2012 P3, X13. Correa-Soto and co-workers stored spray-dried flutamide and ezetimibe in open dishes at 40 °C and 75 or 82% RH and sampled SHG more often than XRPD (for flutamide at 75% RH 15 against 7 time points) correasoto2017 P2, R1, X8. Outside the chapter, Shah and co-workers extruded posaconazole with HPMCAS at 125 °C (below the temperature at which the drug dissolves completely) and at 175 °C and measured residual crystals by SHG and pXRD shah2024 C18, R9, R10.

What the data show

Naproxen: PXRD showed no peaks before 2 h, a weak peak at about 19° 2θ at 2 h and peaks at about 19, 22.5 and 23.7° by 12 h; Raman showed a single broad band at about 856 cm⁻¹ at 0 h and a shoulder at 863 cm⁻¹ from 2 h; SONICC showed sparse bright points at 0 h and more and larger regions at 2 h kestur2012 R12, R13, R14. Crystallinity read from plots was, at 0, 1, 2, 4, 6, 8, 12, 24 and 48 h, about 24, 30, 31, 32.5, 34, 36, 38.5, 42.5 and 48% by SONICC, and from 2 h about 27, 34, 37, 40, 43, 48 and 54% by PXRD and about 26, 27, 32, 33, 35, 39 and 44% by Raman; crystallization was incomplete after 48 h; SONICC lies between the other two from 4 h but above both at 2 h kestur2012 R16, C18, C19, X14. The 24% at 0 h is about 6 wt% of the powder (synthesis: 0.24 × 0.25), close to the 5 wt% below which the PXRD blends showed no peaks, so it is not a trace-level case kestur2012 X9, R8. Flutamide: the SHG ratio was about 0.0009 at 8 days, 0.0079 at 18 days and 0.012 at 29 days at 75% RH and about 0.011 at 15 days and 0.031 at 18 days at 82% RH, then rose steeply; XRPD showed a small 8.6° 2θ peak at 53 days (about the size of the 1% standard) at 75% RH and no evidence until 29 days (abstract: 26) at 82% RH correasoto2017 R9, R3, R1, X1. Ezetimibe: fine SHG spots at 15 days at both humidities, XRPD halo only at 39, 48 and 60 days at 75% RH, and detection at 48 days with quantitation possible only at 60 days at 82% RH (the abstract gives 60 days) correasoto2017 R18, R19, R20, R1, X2. At 75% RH the ezetimibe SHG-derived fraction stayed below 1%, which the authors use to explain why XRPD saw nothing correasoto2017 R21, C31. Posaconazole: pXRD gave 23.1% and 2.4% crystallinity at 50 and 25 wt% loading and "below the limit of detection" at 10 wt% for the 125 °C extrudates, whereas SHG gave 19.3, 3.2 and 0.3 wt% of the ASD mass; ASDs with residual crystals "even below the detection limit of pXRD (as low as 0.3%)" showed a reduced extent or duration of supersaturation shah2024 R9, R10, C56. In tacrolimus capsule contents stored at 40 °C and 75% RH, an apparent "lag period" before XRPD showed crystallization, and the XRPD limit of detection was 10% crystallinity (synthesis: a limit this high does not detect early crystallization, so part of the lag reflects the detection limit of the method) purohit2018 R1, R2, C14.

Figure

Compare the first SHG needles at 11 days with the XRPD halo that persists to 48 days and the first small 8.6° peak at 53 days correasoto2017 R2, R3.

Why

The authors' explanation (their interpretation): in a dispersion the polymer dilutes the crystallizing drug and can interfere with the technique, so PXRD and Raman "most likely" miss the crystals that SHG sees; the dispersion halo is not the sum of the pure halos and Raman spectra of dispersions are not simple to interpret kestur2012 C2, C8, C23, C25. The period before XRPD detection "would normally be considered the induction time", but the SHG data show crystallization progressing in it, followed by a rapid rise "fairly typical" of a solid-state transformation correasoto2017 C19. The earlier film study had made the same point for the lag period before crystals are seen: it depends on the sensitivity of the method wanapun2010 C2. Trace crystals are expected to seed further crystallization in storage and dissolution kestur2012 C1; correasoto2017 C4, C33, and in a BCL–PVPVA extrudate series crystals detectable only by TEM changed crystallization kinetics, supersaturation and polymorphic form moseson2021 C38. Not measured: whether early SHG crystals predict loss of dissolution performance for these dispersions correasoto2017 C33.

Conditions and limits

Naproxen, flutamide, ezetimibe and posaconazole, all in HPMCAS and all with SHG-active crystals; accelerated storage in open dishes (40 °C, 75–82% RH) or at 60 °C kestur2012 P3; correasoto2017 P2. Detection times are the first positive sampled points: the ezetimibe 15 days is the first sample after the initial one, the flutamide SHG ratio already rose at 8 days, and XRPD was sampled less often than SHG correasoto2017 X8. The earlier detection in the naproxen dispersion is at about 24% of the drug, near the PXRD limit of the blends and not at trace level; the 0.1% and 10 ppm results come from physical mixtures kestur2012 X9. Higher humidity sped crystallization: at 82% RH there were far more flutamide crystals at each time correasoto2017 R4, C15.

Counterexamples and conflicts

None found in the library for earlier detection by SHG. For the size of the gap, the sampling and the abstract–text differences (26 or 29 days; 48 or 60 days) mean the numbers are upper bounds correasoto2017 X1, X2, X8.

Transfer check

For another drug: (1) confirm the crystals are SHG-active; (2) sample SHG and XRPD at the same time points from the first day; (3) read an XRPD-defined induction time as an upper bound on the onset and report the XRPD limit with it; (4) if an early SHG signal appears, follow it with a calibration made from crystals grown in the formulation (F9) before calling it an amount; (5) test whether the early crystals matter for dissolution before using them to reject a formulation.

History

2010: Wanapun and co-workers showed that the lag depends on the sensitivity of the method in films wanapun2010 C2, R2. 2012: Kestur and co-workers carried the comparison to a dispersion on storage. 2017: Correa-Soto and co-workers extended it to spray-dried powders of two drugs and two humidities and to the induction time. 2018 and 2021: Moseson and co-workers showed by TEM that crystals below the XRPD limit change dissolution outcomes of extrudates moseson2021 C38. 2024: Shah and co-workers used SHG in a hot-melt-extruded dispersion and linked residual crystals below the pXRD limit to lower supersaturation shah2024 C56.

How it was shown

Wanapun and co-workers read forward (F, transmission) and backward (B, epi) SHG of single pixels over time and computed coherence lengths from the wave-vector mismatch of 800 nm light and its 400 nm harmonic,

L_C^{f}=\frac{2\pi}{\left|\vec k^{2\omega}-2\vec k^{\omega}\right|},\qquad L_C^{b}=\frac{2\pi}{\left|\vec k^{2\omega}+2\vec k^{\omega}\right|},

with an assumed 5% refractive-index mismatch between 800 and 400 nm, griseofulvin n = 1.65 and chlorpropamide n = 1.51 wanapun2010 P5, P10, E1, E2, R9. The signal-to-background at the point where F/B leaves 1, scaled to 3, gave the smallest detectable crystal, assuming equal dimensions wanapun2010 C15, C19, R10. Kestur and co-workers took the epi coherence length of about 100 nm from that work and discussed size, orientation and crystal-form limits; Correa-Soto and co-workers selected two model drugs for their SHG-active crystal forms kestur2012 R17, C34, C37; correasoto2017 C8.

What the data show

The printed coherence lengths reproduce from the equations within rounding wanapun2010 E1, E2. For griseofulvin, F/B stayed close to 1 until 20 min, was about 3 at about 27 min and about 8 at about 52–57 min; for chlorpropamide it was about 1 from 5 to about 18 min (scatter 0.6–1.8), reached about 7 at about 24 min and about 4.3 at about 28 min; simulated F and B coincide for very short domains, B then oscillating while F rises wanapun2010 R6, R7, R8. The measured signal-to-background at the switch was 30 and 5, giving crystals of 90 and 125 nm and the limits of F7 wanapun2010 R10. The authors use F/B as an internal "ruler" of crystal size relative to L_C^b but note that it does not measure nucleation, since the critical nucleus is not related to L_C^b wanapun2010 C15, C18. Chlorpropamide's SHG levelled off or dipped, "tentatively attributed" to polymorph conversion wanapun2010 C24, R7. Which crystals emit: SHG arises only from systems that lack inversion symmetry, so amorphous materials, liquids, solutions and centrosymmetric crystals give no net coherent signal; many chiral compounds crystallize in non-centrosymmetric space groups and achiral compounds routinely do not; crystalline chiral excipients can interfere kestur2012 C4, C37. Flutamide form I (Pna2₁) is non-centrosymmetric although the molecule is not chiral; ezetimibe, with three stereocentres, was expected to crystallize in an SHG-active group and SHG was used to separate its crystals from undissolved amorphous material in dissolution media correasoto2017 C8, M1; elkhabaz2017 C8. Smaller crystals: SEM and TEM "may be suitable" for crystals smaller than SONICC can detect at relatively high number densities kestur2012 C36; TEM found clusters of 5–10 nm domains and discrete crystals below 100 nm in an extrudate that showed only a halo by XRPD (synthesis: crystals of that size lie below the SHG size estimate) moseson2018a R9, C10; kestur2012 C34.

Figure

Forward and backward signals coincide at first and then separate; the time at which F/B leaves 1 marks a crystal about L_C^b thick, and the chlorpropamide trace (b) shows the plateau and dip wanapun2010 R6, R7, R8.

Why

The authors' explanation (their interpretation): while a domain is much thinner than L_C^b the forward and backward signals are generated about equally; once it is thicker, forward SHG rises much faster, so F/B marks a size wanapun2010 C15. SONICC "is limited to molecules that are intrinsically chiral or adopt a polymorph with a chiral unit cell" wanapun2010 C38. Refractive indices of most APIs lie between 1.4 and 1.8, so the authors expect submicrometre detection "routinely" but measured only two drugs wanapun2010 C25. Not measured: a crystal of independently known size, polymorph identity of the growing chlorpropamide crystals, and the sign of the index mismatch, which makes L_C^b uncertain by about 5% wanapun2010 X2, C24, X15.

Conditions and limits

The 90–125 nm sizes are upper-bound estimates through an unprinted scaling wanapun2010 X2, X3. A dark SHG image means no SHG-active crystal above the limit, not no crystals: SHG-inactive polymorphs and crystals below the limit stay invisible wanapun2010 X11. The orientation assumption (randomly oriented crystals) and the size match to standards matter for quantitation (F8, F9) kestur2012 C26, C33.

Counterexamples and conflicts

None found in the library on the selectivity rule. The size claims are consistent but differently worded: a limit of "<150 nm" and a smallest detectable crystal of 90 nm in the film paper, "on the order of 100 nm" with smaller nuclei "unlikely" in the powder paper wanapun2010 C22, R10; kestur2012 C34.

Transfer check

For another drug: (1) establish the space group of the crystal form from a structure or literature before relying on SHG (SHG-active forms exist for achiral drugs, and inactive forms for chiral ones can exist as polymorphs); (2) test SHG on a known crystalline sample of the form that is expected to grow; (3) combine with XRPD or TEM when the question is whether any crystals are present; (4) treat a falling SHG signal in a polymorphic drug as possibly a change of form, not a loss of crystals.

History

2010: Wanapun and co-workers introduced the F/B ruler and the coherence-length estimate for drug films. 2012: Kestur and co-workers carried the 100 nm size limit and the chiral-space-group restriction into powders and dispersions. 2017: Correa-Soto and co-workers showed that an achiral drug can emit SHG, and that orientation can change the signal per mass (F9). The SHG-inactive polymorph issue has no test in the library beyond the chlorpropamide observation.

How it was shown

Van Eerdenbrugh and co-workers asked whether AFM combined with nanoscale mid-infrared spectroscopy can say which phase is which in a phase-separated amorphous film vaneerdenbrugh2012b C2, C3; vaneerdenbrugh2012c C4, C46. In AFM-IR a pulsed infrared laser heats the film through a ZnSe prism, the thermal expansion where the film absorbs rings the contact-mode cantilever, and the integrated Fourier-transform amplitude around the contact frequency (about 134–140 kHz) is the local signal; spectra covered 1200–1800 cm⁻¹ at 4 cm⁻¹ spacing with about 8 cm⁻¹ resolution and 128 pulses per point, references being averages of 10 vaneerdenbrugh2012b P6, E2; vaneerdenbrugh2012c P6, E2. Felodipine and PAA were dissolved in methanol (10% w/v solids), spin coated (100 µL, 3100 rpm, 20 s) at 25:75, 50:50 and 75:25 and imaged by AFM; felodipine has a low crystallization tendency, so it was assumed to stay amorphous, and PAA is amorphous; the pair was known to be miscible at 90:10 and to begin to separate at 30:70 vaneerdenbrugh2012b P1, P5, C4, R1. The 50:50 film (0.57 ± 0.02 µm thick on cover slips, density taken as 1.2 g/mL, six repeats) was chosen because its domains were smaller than those of 75:25 and therefore a harder test; eight spectra were taken, four on domains and four on the continuous phase of one 1.5 × 2.5 µm² area, and 30 × 30 pixel maps at 1500 and 1700 cm⁻¹ with 50 nm spacing were divided to give a ratio image vaneerdenbrugh2012b R8, R9, P7, E1, C11, P2. The dextran–PVP films were spun at 3100 rpm from 10% w/v aqueous solutions at 50/50 w/w with three PVP grades (K12, K25, K90; 2,000–3,000 to 1,000,000–1,500,000 Da) and four dextran or maltodextrin grades (783 Da to 400,000–500,000 Da), spectra being taken along lines of marked positions 0.5–4 µm apart and averaged per phase vaneerdenbrugh2012c P1, P4, P6, M1, M2, M3, M4, M5, M6, M7.

What the data show

Felodipine–PAA: the 25:75 film was smooth, the 50:50 film showed raised domains of about 100 nm to a few hundred nm in a continuous phase, and the 75:25 film domains of about 100 nm to several µm vaneerdenbrugh2012b R2, R3, R4. Nanoscale reference bands of felodipine lay within 9 cm⁻¹ of the bulk bands and kept the amorphous–crystalline differences between 1600 and 1700 cm⁻¹; the ring band near 1210 cm⁻¹ was weak (1204 cm⁻¹, amorphous) or absent (crystalline) and the PAA C=O band at 1712 cm⁻¹ was less dominant than in bulk vaneerdenbrugh2012b R5, R6, R7, C13, C15. All components respond at 1700 cm⁻¹ (C=O stretching), while felodipine responds much more than PAA at 1500 cm⁻¹ (N–H bending coupled with CH₂ vibrations; 1488–1496 cm⁻¹ by AFM-IR) vaneerdenbrugh2012b C17, R7. Each phase also showed bands of the other component, read as partial miscibility, with spectral mixing "cannot be excluded"; the local spectra matched amorphous felodipine, with no crystalline features vaneerdenbrugh2012b R9, R10, C20, C22, C23. Dextran–PVP: in DEX450–PVP90 the thicker central band (n = 13 spectra) carried the dextran band at 1352–1364 cm⁻¹ and the thinner top and bottom domains (n = 7) showed no dextran bands in 1200–1500 cm⁻¹; in DEX450–PVP25 the thicker phase (n = 4) was dextran-rich and the thinner (n = 6) PVP-rich; in DEX40–PVP90 the thick discrete domains (n = 4) were dextran-rich and the thinner matrix (n = 6) PVP-rich vaneerdenbrugh2012c R10, R13, R17. In the DEX40–PVP90 maps the 1280 cm⁻¹ map was brighter in the matrix, the 1350 cm⁻¹ map brighter on the domains, and the ratio map (1280/1350, scale 0.25–2.75) was low on the large domains and on the submicrometre domains in the matrix vaneerdenbrugh2012c R18. In DEX450–PVP12 and DEX40–PVP12 the spectral differences were subtle and in DEX6–PVP90 absent (F13, F15) vaneerdenbrugh2012c R15, R23, R20. In the same period, griseofulvin particles in an HPMC film were identified by drug bands at 1710, 1664 and 1625 cm⁻¹ that the matrix lacks (F14) harrison2013 R5, C13.

Figure

Red spectra (domains) have a 1500 cm⁻¹ band as tall as their 1700 cm⁻¹ band, green spectra (matrix) do not; panel (c) gives the averages vaneerdenbrugh2012b R9, R10.

Why

The authors' explanation (their interpretation): AFM-IR measures the photothermal expansion that follows mid-infrared absorption, so it combines chemical information with a resolution below the diffraction limit of mid-infrared light, which AFM, SEM and TEM lack and mid-IR imaging cannot reach vaneerdenbrugh2012b C2, C3; vaneerdenbrugh2012c C3, C4. A band that only the drug gives (1500 cm⁻¹), read against a band both components give (1700 cm⁻¹), tells the phases apart, and the dextran band near 1350 cm⁻¹ is the diagnostic peak for dextran vaneerdenbrugh2012b C17; vaneerdenbrugh2012c C16, C31. Not measured: the composition of either phase; the expected minimum of 25% felodipine in the PAA-rich phase follows from the miscibility limits and was not measured vaneerdenbrugh2012b C21, X5.

Conditions and limits

Spin-coated films on ZnSe (0.57 ± 0.02 µm for felodipine–PAA, measured on glass vaneerdenbrugh2012b R8); one film per blend; four to thirteen spectra per phase in one area; film age and humidity not stated for felodipine–PAA and not controlled for dextran–PVP, where the PVP carbonyl band sat at 1660 cm⁻¹ against 1680–1682 cm⁻¹ in bulk, ascribed to absorbed moisture vaneerdenbrugh2012b X3; vaneerdenbrugh2012c C10, X4. The thickness of the felodipine–PAA film was measured on glass, not on the ZnSe prisms that were imaged vaneerdenbrugh2012b X4. PAA bands overlap felodipine bands, so PAA in the domains is inferred from band heights and not detected by a unique band vaneerdenbrugh2012b X9. The method has not been tested on spray-dried particles or extrudates.

Counterexamples and conflicts

Outside the chapter, in telaprevir dispersions, AFM-IR could not separate the two phases of 50:50 telaprevir–HPMCAS (identical overlaid spectra, n = 5 per phase), possibly because of small domains and low spectral contrast from high miscibility, and the apparent drug enrichment of the dispersed phase with loading was judged "more likely an artefact of spectral mixing" (F14) li2016c R11, C46, C21.

Transfer check

For another drug: (1) record AFM-IR spectra of the pure drug (amorphous and crystalline) and of the polymer on the same instrument and substrate, because nanoscale and bulk spectra differ; (2) pick one band only the drug gives and one that both give; (3) take spectra on several domains and several matrix positions, and on more than one film; (4) read the result as "richer in" and check it with another method before quoting a composition; (5) expect lower-contrast pairs and domains below about 100 nm to fail (F14).

History

2012: Van Eerdenbrugh and co-workers made the first use of AFM-IR on a drug–polymer blend (felodipine–PAA), then on dextran–PVP blends, the second paper citing the first for the ratio-map rationale and the finding of drug-rich discrete domains vaneerdenbrugh2012c C44, C32. 2013: Harrison and co-workers extended it to drug particles in a film (F14). 2015 and 2016: AFM-nanoIR was used to judge miscibility of itraconazole–HPMC and telaprevir films, with the lower-viscosity component forming the discrete domains in the first and spectral mixing identified as a limit in the second purohit2015a C4, C37; li2016c C14, C21.

How it was shown

The evidence comes from the same experiments as F12 and F14, read for artefacts. In felodipine–PAA, a 1.5 × 1.5 µm² map at 1500 cm⁻¹ and one at 1700 cm⁻¹ of the 50:50 film were recorded at about 1 mW and compared with topography; the 1700 cm⁻¹ image was not shown vaneerdenbrugh2012b P7, R11, R12. In griseofulvin films of about 500 nm thickness a 1710 cm⁻¹ image was compared with the height image harrison2013 R2, R3, R4. In dextran–PVP, pure PVP and dextran films were compared for ringdown response, and spectra were taken at a domain, in the matrix and near a boundary of DEX450–PVP12 vaneerdenbrugh2012c R6, R15.

What the data show

Felodipine–PAA: the 1500 cm⁻¹ map showed higher amplitude over the raised domains (colour scale 0.20–0.75 V) and lower amplitude in the surrounding phase, matching the topography; the 1700 cm⁻¹ map showed similar patterns; in the 1500:1700 ratio map (scale 0.9–1.9) low values tended to lie in the continuous phase and higher values on the domains, but most pixels were in one colour class, low-ratio areas ran mainly along scan rows, some single pixels were extreme, and no overlay or correlation with the topography is shown vaneerdenbrugh2012b R11, R12, R13, X6. Griseofulvin–HPMC: infrared intensity in the 1710 cm⁻¹ image varied directly with local topography, tall features looked sharper and more contrasted than in the height image (an "unexpected" gain the authors explain by greater sensitivity to thickness), and low regions had lower contrast harrison2013 R4, C12, C16, C17. Dextran–PVP: PVP samples gave much stronger ringdown responses than dextran samples, up to an order of magnitude at the maximum intensity, a difference not visible in the normalized spectra; in DEX-rich-phase spectra the PVP carbonyl band at about 1660 cm⁻¹ was the largest band or as large as the dextran bands, so the authors say it "cannot be concluded" that the dextran phase holds much PVP; in DEX450–PVP12 the dextran band near 1350 cm⁻¹ was strongest in the matrix, weakest in a domain and intermediate near a boundary, with all three spectra dominated by the PVP carbonyl; in DEX6–PVP90 six spectra from different positions were alike and had the pattern of pure PVP90 vaneerdenbrugh2012c R6, R10, R13, R15, R20, C16. The dextran-rich phase was the thicker one in all five blends assigned spectrally, which the authors use as a rule, and then use to assign DEX6–PVP90, whose spectra do not differ; film thickness is not stated and "thicker" compares phases only within one image vaneerdenbrugh2012c C26, C37, X7. Outside the chapter, in telaprevir dispersions the local AFM-IR intensity "depends strongly" on thickness, so band ratios were compared, and spectral mixing through the film or laterally below the spatial resolution can put both components' bands into both phases li2016c C12, C14.

Figure

The 1500 cm⁻¹ amplitude (b) is high over the domains and follows the topography (a); in the ratio map (c) low values lie mostly in the matrix, but the map is noisier than (b) vaneerdenbrugh2012b R11, R13, X6.

Why

The authors' explanation (their interpretation): topography may also shape the thermal response, since the continuous phase forms valleys around the domains and roughness, probe–sample angle, sample thickness and thermal expansion coefficient all change the signal; assuming these effects do not depend on wavenumber, a ratio of two maps should isolate the chemical contrast ("it can be assumed") vaneerdenbrugh2012b C25, C26, C27. For PVP against dextran the much stronger PVP response and appreciable mutual miscibility make the phases harder to tell apart, and spectra near boundaries show some spectral mixing ("suggesting"); the three factors were not separated experimentally vaneerdenbrugh2012c C11, C23, C38. In the griseofulvin films the thickness dependence of the PTIR signal is taken from earlier work and said to be shown ("demonstrate") by the match with topography; the authors infer that resolution and contrast fall as the height difference between particle and matrix falls harrison2013 C12, C17. The wavelength independence of the topographic term was not tested vaneerdenbrugh2012b C26.

Conditions and limits

Three systems, each with one film or a few images; the maps are qualitative and the colour scales differ between figures, so amplitudes cannot be compared across samples vaneerdenbrugh2012b X8; vaneerdenbrugh2012c X7. Detection settings (contact frequency 134–140 kHz, windows of 20–60 kHz) differed between samples, so only band ratios within a spectrum are comparable vaneerdenbrugh2012b X8.

Counterexamples and conflicts

None found in the library that contradict the artefacts; the DEX6–PVP90 phase assignment by the thickness rule alone shows how a convention can substitute for a spectrum vaneerdenbrugh2012c C37, X2.

Transfer check

For another drug: (1) record the height image with every map and look for features that appear in both; (2) use ratios of a drug-specific and a shared band and check them on mixtures or films of known composition; (3) measure the film thickness on the same substrate; (4) take spectra near boundaries and on domains of different size to see how much spectral mixing there is; (5) do not infer phase composition from band heights when the components differ in absorptivity.

History

2012: Van Eerdenbrugh and co-workers named the topography effect and the ratio-map remedy in the first AFM-IR drug–polymer paper, then the response-size and mixing effects in the dextran–PVP paper. 2013: Harrison and co-workers showed the thickness dependence for drug particles. 2016: Li and co-workers made spectral mixing the explanation of an apparent composition trend li2016c C21.

How it was shown

Harrison and co-workers asked whether PTIR (AFM-IR) can chemically identify and size crystalline drug particles of about 100 nm embedded in a polymer film, below the 3–30 µm diffraction limit of conventional infrared microscopy harrison2013 R14, C1, C5. Sample 1 was a standard wet-stirred-media-milled suspension (10% griseofulvin, 2.5% HPMC E15LV, 0.5% SDS; median 335.7 nm by DLS) and sample 2 an intensified one (100 µm zirconia beads, HPMC E3, 0.2% SDS, 6 h; median 85 nm) harrison2013 P1, P2, R1. Each was diluted in HPMC/SDS solution and spin-cast on ZnSe to films of about 500 nm (sample 1, imaged at 1710 cm⁻¹) and about 100 nm (sample 2, imaged at 1625 cm⁻¹, 10 nm pixel spacing in the 2.5 µm images) harrison2013 P4, P5, R2, P6. Contact-resonance (stiffness) maps were recorded at the same time; point spectra were co-averaged over 128 or 256 laser pulses (3 and 6.5 min per spectrum); image analysis split one thresholded 2.5 µm PTIR image into four signal classes and converted projected areas to equivalent circular diameters, the circle whose area equals the projected area harrison2013 P7, P8, P9, E2.

What the data show

In sample 1, bright areas of the 1710 cm⁻¹ image coincided with agglomerates of about 0.6–0.9 µm in the height image and a particle spectrum showed bands at 1710, 1664 and 1625 cm⁻¹ that the matrix lacked (F12) harrison2013 R3, R4, R5. In sample 2, particles ranged from tens of nanometres ("likely" primary particles) to about 1 µm ("likely" agglomerates); in the magnified images a 90 nm particle (solid circle) showed as a small spot of higher signal beside a larger one in the PTIR map, and a 100 nm particle (dashed circle) was marked but not discussed harrison2013 R6, R7, R8, C19. In the stiffness maps the particles appeared "noticeably stiffer" than the surrounding HPMC with softer rims, reported qualitatively with no frequency values; the rims were hypothesized to be surfactant- and water-rich regions formed while the film dried harrison2013 R9, C21, C23, X11. Spectra from 3600 to 1000 cm⁻¹ at particles of about 800, 350 and 100 nm showed the matrix bands at about 3450, 2900, 1460, 1370 and 1150–1050 cm⁻¹ on all four spots and the drug bands near 1710, 1660, 1620 and 1590 cm⁻¹ only on particles; the signal-to-noise ratio at 1625 cm⁻¹ was 5:1, 9:1 and 15:1 and the ratio of drug to HPMC signal 5, 8 and 26 (method of calculation not stated) harrison2013 R12, R13, X6. The image-analysis histogram spanned about 36 to 570 nm with a median of 96 nm; the DLS curve spanned about 40 to 460 nm and peaked near 90 nm with a median of 85 nm; every histogram bar is a whole multiple of about 0.76 vol%, which fits a count of about 132 particles, and if weighted by volume the same counts would give a median near 400 nm (extractor's calculation) harrison2013 R10, X12. Resolution is also discussed in the felodipine–PAA work: the authors estimate about 100 nm for films about 500 nm thick, which equals the smallest 50:50 domains and two pixels at 50 nm spacing vaneerdenbrugh2012b C33, R3, X4. Outside the chapter, AFM-IR imaging failed for 50:50 telaprevir–HPMCAS, with domains 122 nm wide and 13 nm high, while contact-resonance stiffness imaging, independent of absorptivity and thickness, still gave contrast li2016c R7, R11, C46, C49.

Figure

Panels (a–c) show 10 µm views and (d–f) 2.5 µm views; the solid circle marks the 90 nm particle, which appears as a small spot of higher signal beside a larger particle in (e) harrison2013 R8, R9.

The drug bands in 1700–1600 cm⁻¹ appear on the three particle spectra; the spectra are drawn at different baselines without an absorbance scale harrison2013 R12, X6.

Why

The authors' explanation (their interpretation): the PTIR signal rises with local thickness, so tall agglomerates look sharper than in the height image, while low regions lose contrast; the close match of topography and PTIR in sample 1 "suggests" that resolution "may be limited" by the size of the tip, and thickness and thermomechanical variations of the matrix can reduce it by thermal artefacts harrison2013 C12, C14, C15, C16, C17. Smaller particles in a thinner film were therefore examined, and stiffness maps add contrast that does not depend on absorptivity harrison2013 C20, C22, C24. The agreement of the median sizes is read as "suggesting" that uneven absorptivity does not significantly distort PTIR in this system, a reading that rests on one image harrison2013 C2, C26, X5. The resolution "will vary for other samples" with absorptivity and thermomechanical properties harrison2013 C29. Not measured: a point spectrum of the 90 nm particle, crystallinity of the particles in the dry film (assumed from the milling route), and the composition of the soft rims harrison2013 X1, X2, C23.

Conditions and limits

One drug and one polymer with a drug band region (1700–1600 cm⁻¹) free of polymer absorption; the two samples differ in milling process, HPMC grade, SDS level, dilution, film thickness, imaging band and cantilever mode at once, so differences in contrast cannot be assigned to particle size alone harrison2013 X8, R12. The film drug content is not printed (roughly 27 and 7.5 wt% from the recipes, an extractor estimate) harrison2013 X7.

Counterexamples and conflicts

The abstract speaks of infrared spectra of 90 nm particles, the figures show spectra of particles of about 100 nm and larger harrison2013 X1. In the telaprevir films the method failed at a similar lateral size when the domains were also low and the contrast small li2016c C46.

Transfer check

For another drug: (1) find a drug band that the polymer does not absorb; (2) make the film as thin as the particle height allows and measure its thickness; (3) test the sizing against an independent method on the same sample and state the weighting of the histogram; (4) add a stiffness (contact-resonance) map when absorptivity contrast is low; (5) expect failure for domains of about 100 nm width that are also low in height or close in composition to the matrix.

History

2012: the first AFM-IR paper in the library estimated about 100 nm resolution for films about 500 nm thick vaneerdenbrugh2012b C33. 2013: Harrison and co-workers claimed the first chemical characterization of sub-100 nm organic crystals and added stiffness maps harrison2013 C1, C24. 2016: Li and co-workers cited the contact-resonance approach for domains where AFM-IR failed li2016c C49.

How it was shown

The DSC benchmark was cited from an earlier study of lyophilized 50/50 dextran–PVP samples, in which a single Tg meant a miscible blend and two Tg events a phase-separated one; the AFM-IR films were spin coated from aqueous solution at room temperature, with no drying step stated and no humidity control vaneerdenbrugh2012c P9, P4, P6, R1. AFM topography of the pure polymers (PVP90 shown) gave the featureless reference for a single phase; contrast in a blend was read as phase separation vaneerdenbrugh2012c R7, C12. Phase identity came from the AFM-IR spectra of F12 and, where the spectra did not differ, from the observation that dextran-rich phases are the thicker ones vaneerdenbrugh2012c C26, C37.

What the data show

The nine blends and their results:

Against the 50/50 DSC traces themselves, AFM differs in two of nine blends (DEX40–PVP12 and DEX6–PVP90), against the reported classification in one; the card asks that the comparison state which is meant vaneerdenbrugh2012c X5. Domain size fell with the molecular weight of either polymer within the DEX450 series (large layers with PVP90, submicrometre domains with PVP12) and the PVP90 series (5–10 µm with DEX40, about 0.3–0.6 µm with DEX6); the claim that DEX40–PVP12 domains are smaller than those of DEX450–PVP12 does not hold in the images shown (≈0.4–2 µm against ≈0.2–0.7 µm) vaneerdenbrugh2012c R9, R14, R16, R19, R22, C27, X1. Discrete PVP-rich domains in a dextran-rich matrix occurred with PVP12 and discrete dextran-rich domains in a PVP-rich matrix with PVP90 (phase inversion); the felodipine–PAA case fits, the drug (lower MW) forming the discrete domains and higher-MW PAA the continuous phase vaneerdenbrugh2012c C43, C44; vaneerdenbrugh2012b C19. In the cited DSC study PVP was less miscible in dextran than dextran in PVP, and in DEX450–PVP12 the phase Tgs deviated further from the pure polymers than with higher-MW PVP vaneerdenbrugh2012c R2, R3.

Figure

The film that DSC calls miscible shows bright domains on a 0–75 nm height scale, while the six spectra from different positions are alike and PVP-like vaneerdenbrugh2012c R19, R20.

Why

The authors' explanation (their interpretation): lower-MW PVP "likely" has higher mobility and lower viscosity than higher-MW PVP, which favours separation into smaller, more localized domains, and discrete domains form from the lower-MW component during spin coating ("may be due to" its mobility and viscosity), a process they call complex vaneerdenbrugh2012c C21, C45. For the DEX6–PVP90 mismatch they suggest that freeze-drying (DSC samples) against fast evaporation at room temperature (spin coating) "may influence" the thermodynamics and kinetics of phase separation and that DSC heating "may promote mixing" vaneerdenbrugh2012c C36. Neither mobility, viscosity, film water content nor a DSC of the spin-coated films was measured vaneerdenbrugh2012c C45, X4. Itraconazole–HPMC, studied by AFM-nanoIR later, was likewise described as a lower-viscosity component forming the discrete domains, without a viscosity measurement purohit2015a C37.

Conditions and limits

50/50 blends of one polymer pair, no drug, spin coated; film thickness not stated; the optical micrographs and the evidence for the three single-phase blends are "data not shown" vaneerdenbrugh2012c X7, X8. Spectra of DEX450–PVP12 and DEX6–PVP90 carry no stated n vaneerdenbrugh2012c X6.

Counterexamples and conflicts

The DSC verdict of one phase did not guarantee a homogeneous spin-coated film (DEX6–PVP90), and a single Tg at 50/50 did not mean a single phase (DEX40–PVP12) vaneerdenbrugh2012c R19, R1, X5; the same pattern recurs with itraconazole–HPMC (F19). The size trend fails for DEX40–PVP12 against DEX450–PVP12 vaneerdenbrugh2012c X1.

Transfer check

For another system: (1) compare the AFM-IR or AFM result on the product-relevant preparation (film, particle, extrudate) with DSC on the same material, since lyophilized, melt-quenched and spin-coated samples can differ; (2) expect, as a hypothesis, the lower-MW or lower-viscosity component to form the discrete phase if the blend separates, and test it; (3) state which DSC trace (50/50 or neighbouring compositions) the classification refers to; (4) report film thickness and humidity.

History

2012: Van Eerdenbrugh and co-workers tested AFM-IR against a DSC study of the same polymer pairs and reported agreement for eight of nine blends. 2015: Purohit and Taylor took the AFM-nanoIR route to itraconazole–HPMC films and found phase separation at all loadings although DSC of the powders gave one Tg purohit2015a C36, C23; purohit2017a R25. 2017: the same group used fluorescence and ssNMR on the same system (F19, F20).

How it was shown

Purohit and Taylor needed to know whether a lipophilic probe at trace level reports the polarity of its surroundings in an amorphous solid before using it to follow phase separation (F17) purohit2015b C3, C4. Films of probucol and procaine were spin coated from methanol with the probe, as were films of ten polymers (PVP K-90, HPMC 606, HPMCAS-MF, PVPVA 64, PAA, PVAc, polyvinyl-2-pyridine, polyvinyl-4-pyridine, CAP-482 and CAB-ADP, the pure polymers from 1:1 methanol:dichloromethane) with prodan, and read on a spectrofluorophotometer; solubility parameters came from the literature or the Fedors group-contribution method purohit2015b P2, P4, M1, M3, M4, E2, M6, M7. Pyrene is read as the ratio \mathrm{I/III}=I_{\mathrm{I}}/I_{\mathrm{III}} of its first to third vibronic emission peak, lower in less polar surroundings purohit2015b E1, C4.

What the data show

The probe shifts between probucol and procaine were 22 nm for 4-Di-2-ASP, 30 nm for prodan and 28 nm for dansyl glycine (synthesis: 570 − 548, 448 − 418, 501 − 473) purohit2015b R1. In the pyrene spectra peak III was relatively stronger in probucol purohit2015b R2. Prodan emission maxima in the polymers were 460.0 (PAA), 450.7 (PVP K-90), 444.1 (polyvinyl-4-pyridine), 443.0 (PVPVA), 442.5 (HPMC), 441.8 (CAP-482), 441.5 (HPMCAS-MF), 440.0 (polyvinyl-2-pyridine), 434.0 (CAB-ADP) and 427.0 nm (PVAc), a span of 33 nm (synthesis: 460.0 − 427.0); the solubility parameters run 25.92, 25.60, 24.72, 22.84, 22.68, 22.48, 22.44, 21.26, 20.86 and 19.20, in exactly the same order, six of them calculated by the Fedors method and four cited, with no unit printed purohit2015b R3, M4, M5, M6, M7, M8, M9, M10, M11, M12, M13, E2. Pyrene gave about 1.30 in amorphous ritonavir, 1.81 in a PVP film and about 1.515 in HPMCAS purohit2015b R6, R13. In the itraconazole work, pyrene I1/I3 was 1.39 in amorphous itraconazole and 1.54 in HPMC, with about 1.90 for pyrene in water, and prodan emitted at about 435 nm in itraconazole and about 451 nm in HPMC purohit2017a R7, R8, R13. The two HPMC values for prodan differ by about 8.5 nm (synthesis: 451 − 442.5 = 8.5) although both papers name Shin-Etsu HPMC 606; the first was a pure film from 1:1 methanol:dichloromethane and the preparation of the references in the second is not described, and the second value is read from a figure purohit2015b M6, P2, R3; purohit2017a M2, R13, X9. For comparison with solution, in water, buffer, methanol and benzene the pyrene ratio fell from 1.6 to 1.07, and in a felodipine or nifedipine drug-rich phase three probes moved beyond their benzene values raina2015a C8, R1, R6, R16.

Figure

The emission maximum falls from PAA to PVAc in the same order as the solubility parameter purohit2015b R3.

Why

The authors' explanation (their interpretation): an environment-sensitive probe emits at shorter wavelength (blue shift) in less polar surroundings and at longer wavelength in more polar ones, and the pyrene ratio falls as polarity falls, so a lipophilic probe in a drug-rich domain reads a less polar environment; the ranking of the polymers "providing support for the supposition" that prodan measures the polarity of the local polymer environment purohit2015b C4, C5, C6. The solution-state explanation of such shifts is in Solution methods F4. Not measured: probe location in a two-phase film or a partition coefficient between phases purohit2015b C3.

Conditions and limits

Two drug hosts with a very large log P difference and ten polymers; each drug–probe pair needs its own pure-component references, because the same probe can give different values in nominally identical hosts (HPMC) and different probes were used for probucol (prodan) and ritonavir (pyrene) purohit2015b M15, M14, X12; purohit2017a X9. Six of the ten solubility parameters are calculated, and the probe list of the methods differs from the one reported (pyrenecarboxaldehyde named, no data) purohit2015b E2, X1.

Counterexamples and conflicts

The two prodan values for HPMC 606 differ by about 8.5 nm (above). In solution, a probe in a drug-rich phase can read beyond its benzene value, so a polarity scale built on polymers cannot be extended to drug-rich phases without a reference in the same state raina2015a R6, R9, R16.

Transfer check

For another drug–polymer pair: (1) measure the probe in films of the pure drug and the pure polymer prepared as the dispersion is prepared; (2) the signal of a miscible dispersion should lie between the two references; (3) do not compare absolute values across papers or batches of reference films; (4) expect water to move the probe in the opposite direction in single-component films (F17).

History

2013 and 2014: environment-sensitive probes had been used to detect a drug-rich phase in supersaturated solutions purohit2015b C4. 2015: Purohit and Taylor calibrated prodan and pyrene in solid films and then applied them to dispersions; Raina and co-workers published the solution-state calibration of six probes in the same year raina2015a C8. 2017: Purohit and co-workers used the ratio of pyrene in a drug and a polymer reference as the test for a miscible dispersion (F19) purohit2017a R24.

How it was shown

Purohit and Taylor spin coated drug–polymer films from methanol at 50 mg/mL with 0.1% w/w probe (prodan for probucol–PVP, pyrene for ritonavir–PVP and ritonavir–HPMCAS), dried them under vacuum and stored them at 97% RH over saturated K₂SO₄, reading the emission on a solid-sample stage from 1 h to 5 days; pure PVP, pure ritonavir and pure HPMCAS films were followed as controls purohit2015b P1, P5. The working hypothesis was that the lipophilic probe partitions into drug-rich domains when water induces phase separation in the film ("We hypothesize") purohit2015b C3. The complementary methods were fluorescence microscopy, polarized-light microscopy, transmission IR of films, modulated DSC and XRD of rotary-evaporated powders of unstated drug:polymer ratio purohit2015b P8, P9, P10, P11, P12, X8. The model for the mechanism is moisture-induced amorphous–amorphous phase separation, known before for probucol–PVP and for PVP dispersions generally, and absent in HPMCAS dispersions purohit2015b R24, R26, R27.

What the data show

Probucol–PVP 50:50: 437 nm at the start, about 426 nm at 1 h, 425–427 nm from 2 to 28 h, 430.5 nm at 40 h, 433 nm at 64 h and 434.5 nm at 5 days (Table 3: lowest 425, end point 435 nm); branched birefringent crystals were seen at 21 h; the 20:80 film moved less (443, 437, 445 nm) purohit2015b R4, R5, R15. The authors call the fall "gradual", but most of it is complete at the first reading purohit2015b C8, X6. Ritonavir–PVP: all three films started between pure PVP (about 1.81) and pure ritonavir (about 1.30); the 50:50 film fell to about 1.31 by 21 h and plateaued at 1.30, the 80:20 film reached about 1.32 by 64 h and 1.275 at 5 days, and the 20:80 film stayed near 1.50–1.55; ritonavir, a very slow crystallizer, did not crystallize, so no red shift followed purohit2015b R6, C12. Images of the 80:20 film after humidity showed many bright round domains of about 5 µm, "presumably" the drug-rich domains, which pure PVP and pure ritonavir films did not show purohit2015b R7, R8, C13. Water alone moved the signal the other way in single-component films over 5 days at 97% RH (prodan in PVP from about 450.5 to 455.3 nm; pyrene in ritonavir from about 1.305 to 1.365 and in HPMCAS from about 1.515 to 1.565), and the 50:50 ritonavir–HPMCAS film started at 1.42, did not fall below it and rose to 1.47 purohit2015b R9, R13, R15. In the IR spectrum of dry 80:20 ritonavir–PVP a new band at 1654 cm⁻¹ was assigned to hydrogen bonds from ritonavir NH and OH donors to the PVP carbonyl; after humidity and drying it was gone and bands at 1630 and 1717 cm⁻¹ (ritonavir-rich) and 1681 cm⁻¹ (PVP-rich) were present; the ritonavir–HPMCAS spectrum was unchanged purohit2015b R17, R18, C17, C18. DSC gave one Tg before and two after exposure (probucol–PVP after 6 h near 40 and 100 °C; ritonavir–PVP at 50 and 140 °C, the lower close to amorphous ritonavir), and after 30 h the probucol sample showed a Tg at 25 °C and a large endotherm near the probucol melting point; XRD showed probucol peaks at 30 h and no ritonavir crystals purohit2015b R19, R20, R21, R22.

Other dispersions fit the pattern. Ritonavir dispersions in PVP and PVPVA separated under humidity at all loadings and none in HPMCAS did purohit2017b R2, R3, R6, R4, R9, C1; celecoxib–PVP and PVP-containing ternary dispersions "appeared to" undergo partial AAPS at 94% RH before crystallizing, read from two Tgs and loss of the N–H band xie2016b C5, C32, C20; miconazole–PVPVA 50:50 films cast at 20% RH were uniform and those cast at 60% RH, or stored at 75% RH for 7 days, gave drug-rich domains saboo2017 R12, R13, R15; evacetrapib–PVPVA films that were miscible stored at 97% RH separated with the opposite microstructure, a drug-rich continuous phase with polymer-rich holes li2017 R19, C28, C33.

Figure

The time axes are categorical; lower values mean less polar surroundings, and the 50:50 and 80:20 ritonavir–PVP curves approach the pure-ritonavir value purohit2015b R4, R6.

Why

The authors' explanation (their interpretation): moisture-induced separation is "a classic case of solvent-induced immiscibility" in which PVP interacts favourably with water while the drug interacts unfavourably with it, so sorbed water replaces drug–polymer contacts (for ritonavir, its NH–carbonyl hydrogen bonds) and, given enough mobility, a drug-rich phase forms into which the probe partitions purohit2015b C24, C27. The red shift of probucol–PVP after 21 h is explained by crystallization expelling the probe into PVP-rich surroundings, with the end value below pure PVP because some probucol "might be" left amorphous purohit2015b C9. The ritonavir–HPMCAS control did not separate; the authors attribute the small rise to absorbed water making the film more hydrophilic purohit2015b C16. Competing readings in the card: water alone raises the prodan value in PVP by about 5 nm over 5 days, so part of the late probucol red shift may be hydration; the 1733 cm⁻¹ band assigned to ritonavir–HPMCAS interaction lies between the two component bands and may be an overlap purohit2015b X9, X10.

Conditions and limits

Three drug–polymer systems at 97% RH, one HPMCAS dispersion (ritonavir at 50:50), although the discussion speaks of HPMCAS dispersions in the plural purohit2015b P5, X7. The "extent" of separation compares the size of a probe shift, with different probes for the two drugs and no calibration of partitioning, so it holds only within a drug–probe pair purohit2015b X12. Table 3 differs from the plots by up to about 2 nm or 0.04 in I/III; the temperature, replicate numbers and the drug:polymer ratio of the DSC and XRD powders are not stated purohit2015b X2, X8. The XRD text says crystals emerged "longer than 30h" although the 30 h trace already shows peaks purohit2015b X11.

Counterexamples and conflicts

None found that contradict the PVP result. For water in the film as a cause of separation during preparation, see F19: in itraconazole–HPMC and evacetrapib films the amount of water present during spin coating decided whether domains formed, which is a different route to the same state mugheirbi2017 C1; li2017 C50.

Transfer check

For another drug–polymer pair: (1) prepare pure-drug, pure-polymer and dispersion films with a probe whose references bracket the dispersion; (2) expose all to the humidity of interest and follow the signal from the first hour, with pure-component films as the control for water alone; (3) confirm with a second method that does not use a probe (IR hydrogen-bond band, two Tgs, imaging); (4) expect PVP and PVPVA dispersions to separate and HPMCAS dispersions to resist, as in three papers of the library, and test whether your pair follows.

History

Before 2015 the library had probucol–PVP and PVP dispersions of other drugs separating in humid air purohit2015b R24, R27; the detection methods were IR, ssNMR, X-ray scattering, DSC and AFM, each with limits purohit2015b C2. 2015: Purohit and Taylor added a probe read-out, ritonavir–PVP and the HPMCAS control. 2017: Purohit and co-workers added PVPVA and the relative rates of dissolution and phase separation; Saboo and co-workers, Xie and co-workers and Li and co-workers added other drugs and the role of water in solvent and air purohit2017b C5; saboo2017 C41; xie2016b C32; li2017 C50.

How it was shown

Spin-coated 50:50 films of each system were immersed in 100 mM pH 6.8 phosphate buffer for 10 s to 30 min (ritonavir–HPMCAS to 15 min), taken out and read on a fluorimeter as undissolved film material; temperature is not stated, and the signal disappears as the film dissolves purohit2015b P6, R10. The reasoning is that if separation occurs at 97% RH it should occur faster in bulk water ("It is reasonable to expect") purohit2015b C25, P5.

What the data show

Probucol–PVP: about 437 nm at 0, about 428 at 10 s (error bar about 424–431), about 424 at 1–2 min, about 422.5 at 3 min, about 422 at 5 min, about 421.5 at 7 and 10 min, then about 423–424.5 from 15 to 30 min; Table 4 gives 425 nm as the lowest value, the same as at 97% RH, while the figure shows about 421.5 purohit2015b R10, R16, X3. Ritonavir–PVP: about 1.565 at 0, 1.49 at 10 s, 1.45–1.51 through 15 min, about 1.405 at 20 min and about 1.29 at 30 min, close to amorphous ritonavir purohit2015b R11. The largest ritonavir–PVP step therefore falls between 20 and 30 min, not within seconds, and the statement that the largest shifts occur within a few seconds holds for probucol–PVP only purohit2015b X4. Ritonavir–HPMCAS: about 1.425 at 0, 1.44 at 1 min and 1.50–1.53 from 2 to 15 min, near pure HPMCAS (about 1.525) and never below its start purohit2015b R14. Pellets of both PVP dispersions gave the same trends in wavelength and ratio (data not shown) purohit2015b R12. Outside the chapter, Purohit and co-workers found that ritonavir films with 50% drug in PVP and PVPVA phase separated within 10 s of contact with water and released drug under drug control at or below the amorphous solubility, while dispersions that stayed miscible during early dissolution formed nanodroplets, and HPMCAS dispersions did so at every loading, fresh or after humidity purohit2017b C5, C3, C53, C54, C6, C55, C30, R4, R9. Saboo and co-workers found that miconazole–PVPVA films in buffer gave a much coarser, connected pattern within 30 min, the drug-rich regions standing higher, than films exposed to humid air saboo2017 R18, R19, C28, C46.

Figure

Each point is measured on the undissolved film; probucol–PVP drops within 10 s whereas ritonavir–PVP drifts down and reaches the ritonavir value at 30 min purohit2015b R10, R11.

Why

The authors' explanation (their interpretation): bulk water would be expected to cause the changes seen at 97% RH purohit2015b P5, with faster kinetics, as observed; they propose that separation during dissolution "might be anticipated" to lead to crystallization within the matrix and lower supersaturation, and that it may explain why dissolution depends on polymer type and amount, none of which was measured purohit2015b C25, C31, C33. The alternative in the card: if PVP dissolves from the film faster than the drug, the residue becomes drug-rich and the probe would read a drug-like environment without separation inside a hydrated matrix; the final ritonavir–PVP value equals pure ritonavir just as the film has mostly dissolved, and no composition of the residue or second technique on immersed films is given purohit2015b X5.

Conditions and limits

Two PVP systems and one control, 50:50 films only, undissolved film read after removal from buffer purohit2015b P6; no dissolution profile, release rate or supersaturation was measured purohit2015b C33, X5. Different probes for the two drugs, and the two 50:50 ritonavir–PVP film series started at 1.57 (immersion) and 1.60 (humidity) purohit2015b X12.

Counterexamples and conflicts

None in the library contradict the direction; the open point is the interpretation of the residue. In Solution methods F10 the 10% drug-loading Kollidon VA64 ritonavir dispersion dissolved above the amorphous solubility and the 50% one did not, which is consistent with a high-loading matrix that separates and releases drug under drug control tres2018 R6, R9; purohit2017b C54.

Transfer check

For another drug: (1) pair the probe read-out of an immersed film with a measurement that does not depend on the probe, such as the composition of the residue, imaging of the wet film or the release profile; (2) check that a drug-rich residue is not produced by selective polymer loss; (3) test several drug loadings, because the effect of loading on separation in water is large (high loading separates, low loading may not).

History

2015: Purohit and Taylor offered the first probe read-out of separation in a hydrating film. 2017: Purohit and co-workers added dissolution profiles, nanodroplets and PVPVA, and Saboo and co-workers added confocal imaging and AFM of immersed films purohit2017b C5; saboo2017 R18. The link from separation to supersaturation remains an inference in the papers of this chapter.

How it was shown

Purohit and co-workers dissolved itraconazole (ITZ) and HPMC-606 (40:60 w/w) in 1:1 methanol:dichloromethane at 2% w/v solids, added 0.2% w/w of one probe to part of the solution and dried it by film casting (about 5 min), spin coating (about 30 s), rotary evaporation (50 °C, then cryomilling) or spray drying (ProCepT, inlet 75 ± 4 °C, outlet 50 ± 4 °C) purohit2017a P1, P2, P3, P4, P5, M2. Modulated DSC (2 °C/min) and XRD were run on the rotary-evaporated and spray-dried powders only; steady-state fluorescence of pyrene, prodan and R6G on films and powders; two-probe confocal imaging with a 60× objective (theoretical resolution about 135 nm); and 13C CP-MAS ssNMR with ¹H T₁ and T₁ρ relaxometry on the spray-dried powder and a cryomilled film-cast sample purohit2017a P6, P7, P9, P10, P11. Single-Tg miscibility criteria have limits that the authors cite: DSC resolves domains only above about 15–30 nm, has recorded one Tg for micron-scale separation, and heating towards Tg can remix the phases purohit2017a R28, R29, C25.

What the data show

Both powders were X-ray amorphous (halo at about 19° and 18° 2θ) and gave one broad Tg step with a midpoint of 87 °C, above that of amorphous ITZ (about 60 °C, cited); the spray-dried trace also showed a small sharp endotherm at about 165 °C, absent at 10 °C/min and in the rotary-evaporated trace, which the authors assign to crystallization during the slow scan because the powder showed ITZ peaks after heating at 2 °C/min purohit2017a R1, R2, R3, R4, R5, C4, C5. Pyrene I1/I3: ITZ alone 1.39, HPMC alone 1.54, rotary-evaporated 1.38, film-cast 1.39 and spray-dried 1.39, all with SD of 0.02 or less; a mixed dispersion should lie between 1.39 and 1.54 purohit2017a R7, R24. No spin-coated I1/I3 value is printed although the methods list one purohit2017a X2. Prodan emitted at about 435 nm in ITZ, about 451 nm in HPMC and about 438 nm in the dispersion film, with the shape of the ITZ spectrum purohit2017a R13. The spray-dried spectrum showed a broad band at about 445–470 nm, assigned to a pyrene–ITZ excimer or exciplex, with about 116 ITZ per pyrene making a pyrene–pyrene origin "unlikely"; reference spectra in this region are not shown purohit2017a R10, R11, C30, X7. In confocal images of cast and spin-coated films prodan stained discrete circular domains and R6G the continuous phase, so the discrete phase is drug-rich ("Thus it is clear") purohit2017a R16, R17, C13. Table II of the ssNMR results is in F20 purohit2017a R21. Parallel cases of one Tg and two phases: DEX6–PVP90 (DSC miscible, two phases by AFM) and DEX40–PVP12 (one Tg at 50/50, classed immiscible from neighbouring compositions) (F15) vaneerdenbrugh2012c R19, R1, X5, and the humidity samples of F17, where DSC showed one Tg before and two after exposure purohit2015b R19, R20. Earlier work on spin-coated ITZ–HPMC films had found them miscible by DSC criteria but phase separated by AFM, nano-infrared spectroscopy, fluorescence and nano-thermal analysis purohit2017a R25.

The conflict: Mugheirbi and co-workers made ITZ–HPMC films by spin coating and spray drying at controlled humidity and solvent water. Films spun at 7% RH were homogeneous by TEM; films at 20–40% RH or particles made with 1–12% water in the methanol contained domains, whose size rose with polymer hygroscopicity and humidity; they conclude that the pair is thermodynamically miscible when dry and kinetically trapped as two phases when made from a water-containing solvent, and that the earlier spray-dried result, phase separated at all loadings with no information on solvent water, was "probably" caused by water contamination of the spray solvent mugheirbi2017 R10, R26, C1, C23, C37, C39. Li and co-workers found that evacetrapib–HPMC and –PVPVA films spun at 18% RH were featureless from 10 to 90% drug loading and that 35–50% RH during spin coating, or 20 µL water per mL of stock, gave drug-rich droplets li2017 C50, C51, C33, R3, R12. Saboo and co-workers saw the same for miconazole–PVPVA, uniform at 20% RH and domains at 60% RH saboo2017 R12, R13, C41. The water content of the Purohit 2017 solutions and films was never measured, and the card records no humidity for the casting or spin coating purohit2017a P1, P2, P3.

Figure

One broad Tg step in each DSC trace (arrows) against prodan-stained discrete domains (blue) in an R6G-stained continuous phase (red); the scale bar is 20 µm in both columns purohit2017a R3, R16, R17.

Why

The authors' explanation (their interpretation): non-polar pyrene "will preferentially partition" into the less polar drug-rich phase "given sufficient mobility", so a partly or fully immiscible dispersion reads like the pure drug, whereas a molecularly mixed one reads in between; DSC gives one Tg because heating towards Tg raises mobility and miscibility, which "can potentially" remix the phases purohit2017a C7, C8, C25. Discrete circular domains are read as "suggesting" nucleation and growth during drying purohit2017a C34. The counter-evidence changes the question from which route to how much water: in Mugheirbi's work water, taken up from humid air or present in the solvent, drives the separation during film formation, with a thermodynamic as well as a kinetic role; Li and co-workers name water absorbed by methanol from humid air as the likely source at high RH; and in Saboo's work water acts as an antisolvent for the hydrophobic drug mugheirbi2017 C1, C39; li2017 C14, C17, C18; saboo2017 C42, C43. Not measured here: pyrene location, the water content of the preparations, DSC of the film-cast and spin-coated samples, the pure-reference preparation, and the structure during or after the DSC scan purohit2017a C7, X5, X9, C25.

Conditions and limits

One drug, one polymer, mainly one ratio (40:60), one solvent system; DSC and XRD cover two of the four routes; pyrene level is 0.1 or 0.2% w/w (unresolved) and the preparation of the reference samples is not described purohit2017a X1, X5, X9, P1. The separate samples for each method (probe-containing for fluorescence, probe-free for DSC, XRD and ssNMR; a Teflon-cast film for ssNMR, a quartz film for imaging) are not the same material purohit2017a X11.

Counterexamples and conflicts

Mugheirbi and co-workers: dry-made ITZ–HPMC films and particles had no domains, so the inherent immiscibility of the pair is contradicted when water is excluded; Li and co-workers: evacetrapib films miscible over 10–90% loading under dry spinning mugheirbi2017 R10, C1; li2017 C50, R3. The claim that all routes gave separation therefore stands for the conditions of the 2017 paper, whose humidity and solvent water are unknown.

Transfer check

For another drug–polymer pair: (1) never rely on one Tg; add a probe or imaging method and, for particles, ssNMR; (2) record and, where possible, control the humidity and the water content of the solvent in every film, spray-dried and rotary-evaporated sample (Li and co-workers name water absorbed by methanol from humid air as the likely source li2017 C14, C17, C18); (3) repeat the preparation dry and wet and see whether the phase state changes; (4) when a probe reads like the pure drug, ask whether the dispersion is the product form or a film made under uncontrolled humidity.

History

2012: AFM-IR showed that a DSC verdict of one phase can fail for films (F15). 2015: Purohit and Taylor found ITZ–HPMC films phase separated at every loading although DSC gave one Tg purohit2015a C36, C23. 2016: Tian and co-workers read a blue-shifted emission band as the sign of a drug-rich phase in indomethacin films tian2016 C16, C17. 2017: Purohit and co-workers extended ITZ–HPMC to powders with pyrene and ssNMR; in the same year Mugheirbi, Li and Saboo and co-workers tied separation in films and particles to water mugheirbi2017 C1; li2017 C50; saboo2017 C41.

How it was shown

The same 40:60 solution was dried at different rates: about 5 min of primary drying for cast films, about 30 s for spin coating (most solvent gone in under 10 s) and a few milliseconds in a spray dryer (cited) purohit2017a P2, P3, R31. Confocal imaging with prodan and R6G sized the film domains (ImageJ with watershed and a circularity filter for the spin-coated films); powders were measured by ssNMR, where spin diffusion averages the relaxation times of intimately mixed components, so equal relaxation times mean mixing over the spin-diffusion length

L=\sqrt{6Dt},

with D = 0.5 × 10⁻¹² (mobile) and 8.0 × 10⁻¹² cm²/s (rigid) giving a range for each relaxation time t purohit2017a P10, P12, E2, C2.

What the data show

Film domains: cast, "around 5 µm" in the Results and "~3–5 µm" in the Discussion, larger prodan regions being built of smaller aggregates; spin coated, "1 µm" and "~400 nm–1 µm", with ImageJ giving a smallest domain of 300 nm, a largest of 2.8 µm, D50 0.9 µm and D90 1.5 µm (number of domains not printed) purohit2017a R16, R17, R18. ssNMR: spray-dried powder, T₁ 3.5 (0.18) s for ITZ and 3.4 (0.05) s for HPMC, Δ 0.1 s, giving 32–130 nm; T₁ρ 9.8 (0.58) and 12.6 (0.22) ms, Δ 2.8 ms, giving 2–8 nm. Film-cast sample, T₁ 3.4 (0.12) and 2.9 (0.04) s, Δ 0.5 s, 29–129 nm; T₁ρ 6.6 (0.64) and 18.6 (0.35) ms, Δ 12 ms, 1–9 nm; these ranges reproduce from L=\sqrt{6Dt} (for example 12.6 ms with 8.0 × 10⁻¹² cm²/s gives 7.8 nm) purohit2017a R21, E2. The Results word the spray-dried powder as mixed at 30–130 nm and not at 2–8 nm, whereas Table III, Fig. 10, the Abstract and the Conclusions give 2–8 nm or below 10 nm; Table III lists 3–5 µm, 500 nm to 1 µm and 2–8 nm, and Fig. 10 reads about 4000, 500 and 10 nm on a log axis purohit2017a R22, R23, C20, C43, X3, X4. The general lengths quoted for T₁ (20–100 nm) and T₁ρ (1–20 nm) do not follow from the paper's own equation (1 s gives about 17–69 nm and 10 ms about 1.7–6.9 nm) purohit2017a R26, X10. The 13C CP-MAS spectra of the spray-dried and film-cast samples were almost identical, so the authors infer that evaporation rate did not change molecular structure, though dynamics "could potentially" differ purohit2017a R20, C18. Other scales in the library: confocal resolution about 135 nm (here) or a theoretical lower limit of about 150 nm for discrete domains (Saboo and co-workers), AFM-IR about 100 nm estimated for films about 500 nm thick (F14), and the DSC resolution of about 15–30 nm (cited) purohit2017a P10, R28; saboo2017 C37; vaneerdenbrugh2012b C33. In miconazole–PVPVA films the drug-rich pattern after immersion in buffer was much coarser than after humid storage, so the scale also depends on the route of water entry saboo2017 C46.

Figure

The bars come from different techniques (confocal imaging for the films, ssNMR for the powder) and the spray-dried bar is a bound from ssNMR, not a measured size purohit2017a R22, R23, X11.

Why

The authors' explanation (their interpretation): faster evaporation "is expected to" generate supersaturation faster and so raise the nucleation rate of the drug-rich phase, giving smaller domains; slowly cast films have time for domains to grow and aggregate, which "indicates" that they form before the solvent has fully evaporated; spray-dried domains should be "much smaller" than the 1 µm of spin-coated films because primary drying takes milliseconds, and the ssNMR result is "entirely consistent" with this purohit2017a C35, C36, C38, C41. Supersaturation, nucleation rate and domain formation during drying were not observed, only end-state images purohit2017a C34, C35, C36. ssNMR sets a length through spin diffusion, with two assumed diffusion coefficients; the authors note that a rigorous measurement is needed for a true domain size purohit2017a E2, C2. Fluorescence was "sensitive and rapid" for nanoscale phase separation but gives no size, and confocal imaging cannot resolve the spray-dried domains purohit2017a C44, C40, C32. Because the domain-formation step is where water acts (F19), part of the drying-rate trend may be an effect of how much water was present at each rate, which was not separated.

Conditions and limits

One drug–polymer pair, one solvent system, mainly 40:60; the film sizes come from quartz films and the powder sizes from a cryomilled Teflon-cast film and spray-dried powder; the film-cast ssNMR value is a lower bound only purohit2017a X11, P1. The number of domains analyzed is not printed purohit2017a R18. The effect of the domains on crystallization or dissolution was not tested purohit2017a C23.

Counterexamples and conflicts

Domain size and presence depended on water in the work of Mugheirbi and Li (see F19), so a trend with drying rate at unknown humidity is not separate from that mugheirbi2017 C23; li2017 C33. Within the paper, sizes for the same route differ between text, Table III and Fig. 10 purohit2017a X3.

Transfer check

For another drug–polymer pair: (1) decide which length scale matters (domains that confocal imaging can resolve, or smaller) and choose the test accordingly; (2) for particles, combine a probe test for presence with ssNMR for a bound on size; (3) report T₁ and T₁ρ of both components, the spin-diffusion coefficients assumed and the resulting range, not a single value; (4) compare drying routes at controlled humidity.

History

2015: Purohit and Taylor sized ITZ–HPMC domains at 500–1000 nm in films by AFM and fluorescence purohit2015a C36, C37. 2017: Purohit and co-workers added the drying-rate comparison, spray-dried powder and ssNMR; Saboo and co-workers gave a confocal resolution of about 150 nm for the same two-dye approach saboo2017 C37.