Seeing the first crystals in spray-dried dispersions: SHG microscopy against XRPD

Background and motivation

About 75% of drugs in development fall into classes II and IV of the Biopharmaceutics Classification System, that is, they dissolve poorly in water, and an oral drug must dissolve and stay dissolved before it can cross the gut wall (Section 1; Di 2012, ref 1; Pouton 2006, ref 2). C1·R25.a One answer is the amorphous solid dispersion (ASD): the drug is dispersed, typically at the molecular level, in a polymer, and the amorphous form is "expected" to have a several-fold higher apparent solubility than the crystal; more than 20 ASD products were on the market when the paper was written (Kawabata 2011, ref 7; He 2015, ref 8). C2·R25.b

The price of the amorphous form is instability. Amorphous drug tends to crystallize during storage, heat and moisture speed this up, and once crystals form the solubility gain is lost (refs 7, 12, 13). C3 The paper's starting argument is that small amounts of crystal, left over from manufacturing or formed by nucleation during storage, are "expected" to keep growing, because crystallization lowers the free energy of the system. Detecting the first crystals early therefore matters for choosing formulations and process conditions and for judging how robust a dispersion is to storage (refs 14, 15). C4

The difficulty is that a dispersion dilutes the drug. Diffraction, calorimetry, near-infrared, infrared, Raman and solid-state NMR all have limits; vibrational spectra may not separate amorphous from crystalline drug, and the polymer can overlap with the drug's bands. As a rule of thumb, crystal contents below about 1 wt% of the formulation, or crystallites smaller than 10⁻⁵ cm, are hard to identify (refs 9, 10, 16, 17). C5·R25.c–d With drug loadings of only 10–20%, 1 wt% crystal in the powder can already mean that a sizeable fraction of the drug has crystallized before anyone notices. C5·R25.e

Second harmonic generation offers a way round this, because it is blind to the amorphous matrix. Earlier work with the SONICC instrument had detected 3.4 × 10⁻⁹ %v/v crystallinity in amorphous griseofulvin (Wanapun et al., ref 18) and 10 ppm crystalline naproxen in a binary mixture with 99.9% HPMCAS (Kestur et al., ref 19). C6·R25.f–g SHG has also been used to guide synchrotron XRD to 100 ppm crystalline ritonavir (Newman et al., ref 20) and Raman spectroscopy to paclitaxel crystals in a commercial freeze-dried product whose XRPD pattern showed no peaks (Schmitt et al., ref 9). C6·R25.i–j What had not been tested was whether SHG gives the same advantage in real spray-dried dispersion powders as they crystallize on storage, and whether its signal can be turned into an amount. The authors set out to evaluate SONICC for this purpose, with XRPD and solid-state NMR as reference techniques (Section 1). A1–A4

Background concepts

Second harmonic generation and SONICC

SHG is a nonlinear optical effect: light of a certain energy interacting with a non-centrosymmetric material produces light at twice that energy (refs 19, 21). Only non-centrosymmetric systems give coherent SHG; isotropic and centrosymmetric systems, including liquids, glasses and other amorphous materials, give negligible signal because the contributions cancel almost perfectly (ref 18). C7 The chirality of many drugs leads to crystalline forms that are typically SHG active (ref 9). C7 An SHG image of a dispersion is therefore a map of crystals on a dark background: the method reports whether certain crystal structures are present rather than molecular-level information, and it works only when no excipient interferes with the drug's SHG signal. C7·C11 SONICC (second order nonlinear optical imaging of chiral crystals) is the commercial SHG microscope used here, in a modified form. P3

The two model drugs were chosen with this in mind. Flutamide is not chiral, but its stable form I crystallizes in the orthorhombic, non-centrosymmetric space group Pna2₁; a metastable form II of unknown space group exists but converts to form I as soon as it forms (Ceolin 1995, ref 22). C8·M1 Ezetimibe's chemical name carries three stereocentres (3R, 4S, 3S), and the paper selected both drugs as compounds that show SHG activity. M2 Both molecules also contain fluorine, which HPMCAS does not, so ¹⁹F solid-state NMR sees only the drug and can split its spectrum into amorphous and crystalline parts. C8·P5

Two ways to express crystallinity

The paper uses "% crystallinity" for the weight per cent of crystalline drug in the whole powder (Section 3.1). For the 15% flutamide dispersion, 15% crystallinity therefore means that all the drug is crystalline. E1·R3 For ezetimibe, Figure 12 instead reports "% drug crystallized", the share of the drug that is crystalline. The two are linked by the drug loading:

In plain terms: the crystalline mass fraction of the powder, $w_{\mathrm{c}\mathrm{r}\mathrm{y}\mathrm{s}\mathrm{t}}$, equals the fraction of the drug that has crystallized, $f_{\mathrm{d}\mathrm{r}\mathrm{u}\mathrm{g}}$, times the drug loading DL. The paper applies it once: 11.5–12% of the drug crystallized in the 30% ezetimibe dispersion corresponds to about 3.5% crystalline material in the powder (11.5 × 0.30 = 3.45). E1·R22 Keeping the two bases apart matters later, because the detection limits are given in % crystallinity of the powder.

Detection and quantitation limits

The limit of detection (LOD) is the smallest amount that can be told apart from a blank, and the limit of quantitation (LOQ) the smallest amount that can be measured with acceptable precision. The authors computed both from the residual standard deviation of their calibration lines, following the ICH Q2(R1) guideline (ref 26). The formula is named but not printed. E5·P9 They also report what could simply be seen: whether a standard of known crystallinity gave a visible signal. These two kinds of limit, statistical and visual, give different numbers and should not be mixed. X7

XRPD as the reference

XRPD detects crystals through their Bragg peaks above the broad halo of the amorphous material. For a standard laboratory instrument the detection limit for organic crystals has been put at about 5–7%; with an area detector, synchrotron radiation and an algorithm that separates amorphous and crystalline intensity, 0.2% has been reached, and chemometric processing has detected about 0.5% crystalline tacrolimus in a solid dispersion (refs 16, 23, 29). R25.k–m·X12

Methods in detail

The study is a side-by-side comparison of detectors on the same stored powders, backed by calibration standards and two independent checks. The method flow:

Dispersions and storage

Flutamide (Sigma-Aldrich) and ezetimibe (Attix Pharmaceuticals) were each dissolved with HPMCAS (AS-MF grade, Shin-Etsu) in methanol at 10% w/w solids, stirred for 3 h, filtered through 0.45 µm and spray dried on a ProCepT dryer with a feed rate of 6 mL/min, inlet and outlet temperatures of 90 °C and 54 °C, an atomization air flow of 6.3 L/min and an inlet air flow of 0.35 m³/min (Section 2.2). P1·M1·M2·M3 The products were a 15% w/w flutamide dispersion ("FTM 15%") and a 30% w/w ezetimibe dispersion ("EZT 30%"). P1 The powders were stored in open dishes at 40 °C and either 75% or 82% relative humidity (RH) and sampled at intervals for up to 60 days. P2 SHG was sampled more often than XRPD: for flutamide at 75% RH, 15 SHG time points (0 to 60 days) against 7 XRPD time points (0, 15, 29, 39, 48, 53 and 60 days). R1.a–b

SHG imaging

The modified SONICC microscope uses a femtosecond fibre laser at 1064 nm (51 MHz repetition rate, 166 fs pulses) that is raster-scanned by a resonant mirror (8 kHz) and a galvanometer mirror, and collects the SHG light in reflection (epi detection) (Section 2.3). P3 Powder was held between two cover glasses with a thin spacer to keep the layer thickness uniform. Each sample was prepared three times and imaged in three fields of 1925 µm × 1925 µm, and the mean pixel intensity of the 9 images was used (ImageJ). P3 The laser power at the sample was 50 mW for flutamide and 250 mW for ezetimibe, with 447 ms exposure for both; a run took about 1.5 min per sample. P3 The higher power for ezetimibe was needed because its crystals are less SHG active. R17

XRPD

Patterns were collected on a Rigaku SmartLab at 40 kV and 44 mA, at 5 deg/min with a 0.04° step, from 5° to 25° 2θ for flutamide and 5° to 40° 2θ for ezetimibe; a run took about 6 and 9 min (Section 2.4). P4 Crystallization was followed with the strong flutamide reflection at 8.6° 2θ and the ezetimibe reflection at 19.5° 2θ. P4·M1·M2

Calibration standards, quantitation and limits

To mimic a partly crystallized dispersion, the standards were made by mixing the initially amorphous spray-dried powder (flutamide 15%, HPMCAS 85%) with a physical mixture of crystalline flutamide and HPMCAS in the same 15:85 ratio, so that the total drug stayed at 15% and only the crystalline share changed; the crystalline drug was the as-received material (Section 2.6). P6·C9 XRPD standards contained 1, 2, 3, 4 and 5% w/w crystalline flutamide in triplicate; SHG used these plus 0.005, 0.01, 0.05, 0.1 and 0.5% w/w, which showed no crystallinity by XRPD. P6·R7 Ezetimibe standards were made the same way, with XRPD calibration over 1–10%. P6

For XRPD the measured quantity was a peak-area ratio: the area of the peaks between 7.72° and 8.92° 2θ divided by the total area between 6.6° and 13.3° 2θ for flutamide, and 18.32–20.36° over 17.6–20.5° 2θ for ezetimibe. P7·E3 For SHG it was the mean image intensity. P8 LOD and LOQ came from the residual standard deviation of the calibration lines. P9·E5 For the flutamide time courses, the sample's SHG intensity was divided by that of the 15% crystalline standard, which stands for a fully crystallized sample:

In plain terms: $I_{\mathrm{s}\mathrm{a}\mathrm{m}\mathrm{p}\mathrm{l}\mathrm{e}}$ is the mean SHG intensity of the stored powder and $I_{\mathrm{s}\mathrm{t}\mathrm{d}}$ that of the standard in which all the drug is crystalline, so a sample whose drug has fully crystallized should give $R=1$ (Section 3.1, Figure 5). E2·P10 For ezetimibe the intensity was converted with the calibration line into % of drug crystallized (Figure 12). P10

Solid-state NMR and SEM

¹⁹F direct-polarization magic-angle-spinning spectra (Bruker Avance III HD, 400 MHz ¹H frequency, 12 kHz spinning, 294 K, 10 s recycle delay, TPPM proton decoupling at 83 kHz) were split into amorphous and crystalline components, using the spectra of the crystalline drug and of the initial dispersion as references and Teflon at −122.0 ppm as shift reference (Section 2.5). P5·M5 For flutamide the CF₃ group gives a single peak at −121.9 ppm when crystalline and −133.9 ppm when amorphous. M1 The calibration set was measured too, and ssNMR followed the expected 1–5% crystallinity with "a reasonably good correlation" (Figure S7A). R13·C10 Scanning electron microscopy (FEI NOVA nanoSEM, platinum coating, 5 kV, 500–2000×) imaged the as-received flutamide and the dispersions before and after 60 days of storage (Section 2.7). P11

Experimental analysis

SHG sees crystals weeks before XRPD

At 40 °C and 75% RH the fresh flutamide dispersion gave almost no SHG signal; after 11 days crystals were "clearly observed", and their number and size grew until a dense field of needle-like crystals covered the image at 60 days (Figure 1). R2·C12 XRPD of the same powder showed only the amorphous halo at 0, 15, 29, 39 and 48 days. A small 8.6° 2θ peak appeared at 53 days, "qualitatively" about as strong as in the 1% (w/w) standard, and by 60 days reflections at about 8.6, 17.2 and 18.8° 2θ were visible (Figure 2). R3·C13 At 82% RH flutamide crystallized faster ("Unsurprisingly"): SHG showed crystals after 8 days, with far more crystals than at 75% RH at each time (Figure 3), and XRPD detected crystallinity after 29 days (Figure S8). R4·R1.c–d·C15 The abstract gives 26 days for this XRPD detection; the supplementary figure that would settle it was not available. X1

Ezetimibe gave the same picture. Its crystals appeared in SHG at the 15-day point at both humidities as fine, point-like spots that became more numerous with time (Figures 9 and 11). R18·R20·R1.e·R1.g At 75% RH XRPD showed no peaks at 39, 48 or 60 days (Figure 10); at 82% RH XRPD detected crystallinity after 48 days (Figure S9) and could quantify it only for the 60-day sample. R19·R1.f·R1.h·C28 The abstract instead quotes "XRPD (60 days)" for ezetimibe. X2

Table A. First storage time at which crystals were detected (from the text and Figures 1–3 and 9–11). R1·X1·X2

The authors conclude that SONICC detected crystals at a "considerably earlier" time than XRPD for flutamide and at "much earlier time points" for both drugs. C14·C42 Note by the AI writer (not discussed in the paper): the detection days are first positive sampled points, so they are upper bounds. Ezetimibe's 15 days is the first sample after the start, and at 75% RH the flutamide SHG ratio already rose between 4 and 8 days (Figure 5A), before the 11 days at which crystals were called clearly visible. X8·R9.b–c

How little crystal SHG can see

In the SHG images of the flutamide standards (Figure 4), the amorphous dispersion was dark and the 0.005% standard already showed sparse bright spots, which the authors describe as "readily detected"; the spot density rose through 0.01, 0.1, 0.5, 1, 2 and 4% to a dense field at 15%. R5·C17 XRPD showed no peaks below 1% crystallinity. R7 The calibration lines and the limits computed from them are summarised in Table B.

Table B. Calibration and limits in % crystallinity of the powder (text, Sections 3.1 and 3.2; the flutamide SHG limits come from the lower-concentration line; y-axis units of the lines and the ezetimibe line parameters are not printed). R6·R8

On these numbers the SHG detection limit is more than ten times lower than XRPD's for both drugs (0.03 against about 0.4% for flutamide, 0.047 against 0.7% for ezetimibe). R8 In the conclusions the authors put it as "less than 0.01% crystallinity could be readily detected" by SHG, against 1% and higher for XRPD, and they note that the result agrees with an earlier 10 ppm SHG detection limit for naproxen. C36·C35·R25.h Note by the AI writer: the 0.005% and "less than 0.01%" figures are visual detections of standards, whereas the statistical LODs are 0.03–0.047%; likewise the "1%" for XRPD is a visual threshold, against computed LODs of about 0.4–0.7%. X7 The introduction attributes the 10 ppm naproxen result to ref 19 (a binary mixture with HPMCAS) and the discussion to ref 18 (a dispersion made by rotary evaporation), so the citation should be checked before reuse. X12

Crystallization during the "induction time"

Figure 5 follows the flutamide SHG ratio $R$ through storage. Read from the plots, at 75% RH it rose from about 0.0003 at 4 days to about 0.0009 at 8 days, 0.0032 at 11 days, 0.0079 at 18 days and 0.012 at 29 days; at 82% RH it reached about 0.011 at 15 days and 0.031 at 18 days. R9.b–h Over this period XRPD showed nothing. R3·R1.d Later the ratio climbed steeply: at 75% RH to about 0.10 at 48 days, 0.14 at 53 days, 0.69 at about 57 days and 1.03 at 60 days; at 82% RH to about 0.26 at 26 days, 0.5–0.76 at 29–39 days, 1.1 at 48 and 53 days and about 3.1 at about 57 and 60 days. R9.g–o

Table C. Flutamide SHG intensity relative to the fully crystalline (15%) standard, selected days (read from Figure 5, approximate; error bars not defined in the paper). R9·X11

The authors' reading is that a small amount of crystallization takes place over the first 30 days at 75% RH or 20 days at 82% RH, below the XRPD detection limit. That period "would normally be considered the induction time for crystallization, but the SONICC data clearly shows that crystallization is indeed progressing"; afterwards the extent of crystallization rises rapidly, in "a fairly typical profile for a solid state transformation". C19 Because $R$ is a ratio of intensities and, as the next section shows, the flutamide crystals grown in the dispersion are brighter than the standard, these early values show that crystals are present and growing, but not how much drug has crystallized. C20

Why flutamide could not be quantified

A sample whose drug has fully crystallized should give $R=1$. The stored flutamide samples exceeded this: at 75% RH $R$ reached about 1.03 at 60 days, and at 82% RH it approached 3. The authors therefore call applying the calibration curve "impractical"; Figure 6 shows the 60-day sample at 82% RH visibly brighter than the 15% crystalline standard. R9·R10·C20·C16 After 58 days at 82% RH the samples saturated the linear range of the photomultiplier and data collection was stopped. R11 Figure 5D nevertheless plots points at about 57 and 60 days with the same mean of about 3.1, so these values are best read as saturated lower bounds. X10

The two other methods found only a few per cent crystallinity. XRPD placed the samples with a high SHG ratio, analysed on days 35–53, at 1–2% crystallinity, and ¹⁹F ssNMR gave 1%, 3.8% and 4.6% at 35, 39 and 48 days. R12·R14 On a 15% drug loading, an $R$ near 1 would correspond to 15% crystallinity, so SHG overstates the amount by a large factor. E1·E2 The authors note that the ssNMR values agree better with XRPD than with SONICC and conclude that "it appears clear" the crystals formed from the amorphous particles are more SHG active than the calibration crystals. C21·C22 Two caveats apply to this comparison. The ssNMR samples are described as stored at 75% RH in one sentence and at "the higher relative humidity" in the next; the comparison with the high-SHG-ratio samples of days 35–53 fits 82% RH better, because at 75% RH the first XRPD peak appeared only at 53 days. X3 And ssNMR at 39 and 48 days (3.8 and 4.6%) is about two times the top of the XRPD range, so the agreement is relative rather than close. X4

Table D. Crystallinity estimates by method (text; SHG ratio range read from Figure 5D, approximate; flutamide percentages are of the powder, ezetimibe percentages of the drug; the RH of the flutamide ssNMR samples is stated two ways). R9.j–m·R12·R14·R21.i·R23·X3

XRPD showed no new polymorph of flutamide, so the authors looked at crystal shape. C23 SHG intensity depends on how a crystal is oriented relative to the laser, and the total signal from a powder sums over all orientations present; if the orientation distribution changes with the crystal habit, for example through preferred orientation, the total SHG "could be quite different for two morphologies of the same crystal polymorph" (ref 27). C24 Note by the AI writer: the polymorph check rests on a few weak reflections in the stored-sample patterns (about 8.6, 17.2 and 18.8° 2θ at 53–60 days at 75% RH), and the 82% RH patterns are in the unavailable supplement. X15

What the crystals look like

Before storage the flutamide dispersion consisted of collapsed, dimpled particles typical of spray drying, and the as-received flutamide of large blade-shaped crystals (Figure 7). R15·C25 After 60 days at 75% or 82% RH the powders contained long, thin, acicular crystals several times longer than the 40 µm scale bar; one is circled growing outward from the surface of a particle (Figure 8A), the magnified end of another appears hollow (Figure 8B inset), and many are bent (Figure 8C). R16·C25 In the abstract the authors call this "an unexpected mechanism of crystallization, whereby flutamide crystals grew as needle-shaped projections from the surface of the spray dried particles". C26 For ezetimibe, SEM showed clear crystals only after storage at 82% RH; small bumps on the 75% RH particles "may be due to" crystals, and the 82% RH powder had partly sintered (Figure S11). R24·C32

Ezetimibe: quantitation that agrees

For ezetimibe the SHG calibration gave sensible amounts. At 82% RH the crystallized share of the drug stayed near zero at 15–18 days, rose to about 0.4–1.6% at 26–35 days and 3.7% at 53 days, and reached about 11.3% at 60 days in Figure 12B (text: about 11.5%), next to the XRPD value of about 12%. R21·R21.i That corresponds to about 3.5% crystalline material in the powder. R22·E1 At 75% RH the SHG value stayed below 1% of the drug (about 0.6% at 60 days), which the authors use to explain why XRPD saw nothing. R21·C31 They call SHG and XRPD "in reasonable agreement"; ssNMR "suggests" 17% of the drug had crystallized at the same time point (Figure S10). C29·C30·R23 Note by the AI writer: ssNMR is about 1.5 times the other two values, XRPD could quantify only this one time point, and if the ezetimibe SHG limits (0.047 and 0.14%) are on the powder basis that the paper defines, the 75% RH values up to 53 days lie below the LOQ. X5·X6

Discussion

The authors' explanation of the flutamide discrepancy

The authors connect the findings in a short chain. Calibration assumes that a given mass of crystal gives a given SHG intensity; for ezetimibe this held, for flutamide it did not.

In the discussion the authors write that this "dramatic change in morphology is assumed to be responsible for the discrepancy between the two techniques", and in the conclusions that the discrepancies "could be explained" by the unusual hollow acicular crystals. C38 They find the hollow crystals "quite unexpected": the few hollow organic crystals reported so far all came from solvent-mediated phase transformations, in which a new phase grows on the surface of the old one and leaves a hollow centre, and "Clearly, a solvent mediated phase transformation is not responsible" here; "further studies are needed". C39 They frame the outcome as a calibration problem rather than a limit of the method: "it is not that the technique is inherently incapable of yielding accurate quantitative data", but the calibration material does not represent what crystallizes from the dispersion, "a perennial and realistic concern with any system that undergoes crystallization in a complex environment". C40

Place among earlier SHG work

An earlier naproxen study (ref 18) had found good agreement between SONICC, X-ray and Raman crystallinity estimates; here ezetimibe showed reasonable agreement and flutamide did not. C37 The sensitivity advantage over XRPD, which needed about 1% or more crystallinity for either drug, is "in general consistent" with the literature, which the authors note depends on the compound and the equipment. C36 They describe ASD formulations as "undoubtedly a target area" for SHG, because typical drug loadings (about 15%) and the high detection limits of common methods make minor crystallinity hard to detect. C34·R25.e

What it suggests for formulation

The authors' own conclusion is practical: crystals in an ASD are "considered extremely deleterious", any crystalline material present "will likely grow with time during storage of the product" or deplete supersaturation during dissolution, and rapid identification of robust, non-crystallizing formulations avoids long assessment times. C33 Early SHG detection, which needs only a small amount of powder, "is likely to aid in the formulation development of poorly water soluble compounds", and despite the quantitation problems SHG is "an important technique in the toolbox" for ASD characterization. C43·C41

Note by the AI writer (not discussed in the paper): three practical points follow. First, in a stability study SHG answers "are there crystals yet?" weeks before XRPD, which allows faster ranking of polymers, drug loadings or storage conditions, but its intensity should be reported as detection rather than as % crystallinity unless it is calibrated with crystals grown in the same matrix. Second, an onset time defined by XRPD can hide crystallization that is already under way, so stability arguments built on an XRPD "induction time" may be optimistic for systems like these. Third, the flutamide result shows that crystals grown from a dispersion can differ in habit from the drug lot used for calibration, which is worth checking for any habit-sensitive method.

Limitations

Limits the authors state

The authors state that quantitation "proved challenging" for flutamide because the crystals that formed from the spray-dried particles had a different SHG activity from the calibration crystals, and that the calibration samples were not representative of the crystallizing material. C16·C40 They leave the origin of the hollow crystals open ("further studies are needed"). C39 For ezetimibe, XRPD quantification was possible only for the 60-day sample at 82% RH. C28

The AI writer's own judgement

The central claim, that SHG detects crystallization in these dispersions much earlier than XRPD, is well supported: it holds for both drugs at both humidities, in images and in intensity curves, and XRPD of the same powders stays amorphous for weeks after SHG shows crystals. R1·R2·R3·R4·R9·R18·R19·R20 The quantitative and mechanistic claims are weaker, for these reasons:

• Two XRPD detection times differ between the abstract and the text (26 against 29 days for flutamide at 82% RH; 60 against 48 days, or no detection at 75% RH, for ezetimibe). X1·X2
• Detection times are bounded by the sampling schedules, which were denser for SHG than for XRPD. X8
• The headline detection level ("less than 0.01%") is a visual detection of standards; the computed SHG LODs are 0.03–0.047%, and the basis (powder or drug) of the ezetimibe limits is not stated. X7·X6
• For flutamide, SHG, XRPD and ssNMR disagree, the RH of the ssNMR samples is stated two ways, and the highest SHG ratios were measured at detector saturation. X3·X4·X10
• The morphology explanation was not tested: no SHG was measured on single crystals and no orientation analysis was done, and hollowness rests on the end view of one crystal. The surface-growth "mechanism" rests on a few SEM images taken after 60 days. X14·X13
• Error bars in Figures 5 and 12 are not defined, and the calibration data, ssNMR spectra and 82% RH XRPD patterns are in the supplementary data, which were not available here. X11·X16
• Two drugs, one polymer, one temperature and two humidities were studied; SHG settings and data treatment differed between the drugs (50 against 250 mW; ratio against % drug crystallized), so detection times should be compared within each drug. X9

Conclusion

For spray-dried flutamide and ezetimibe dispersions in HPMCAS stored at 40 °C and high humidity, SHG microscopy detected the first crystals 3 to more than 6 weeks before XRPD and showed crystallization progressing during the period XRPD would call an induction time. R1·R9·C19 Its detection limits were more than ten times lower than XRPD's. R8 Turning SHG intensity into an amount worked for ezetimibe but failed for flutamide, whose crystals grown in the dispersion, hollow and needle-shaped, gave up to about three times the signal of a fully crystalline standard; the authors attribute this to crystal habit and see it as a calibration problem, a link they did not test directly. R21·R9·C38·C40 The paper's lasting lesson is about calibration as much as sensitivity: a highly sensitive crystal detector is only as quantitative as its standards are representative of the crystals that actually grow in the formulation.