Residual nanocrystals in an "amorphous" hot-melt extrudate, seen by electron microscopy

Background and motivation

Amorphous solid dispersions (ASDs) are used to overcome the poor solubility and slow dissolution of many drugs: the drug is turned from crystal to amorphous and held dispersed in an amorphous polymer, so that the solubility advantage of the amorphous form can be realised (Introduction, refs 1–4). Hot melt extrusion (HME) makes ASDs with heat and shear: the drug melts and/or dissolves into the molten polymer (ref 5). The authors' premise is that with enough heat and mixing a homogeneous molecular dispersion "should" form without residual crystals, but if the transformation is incomplete, residual crystals "may" cause stability or dissolution problems (refs 6–9) C1.

The question is therefore how well residual crystals can be detected. X-ray powder diffraction (XRPD) is the standard method, but its sensitivity falls with dilution of the drug in the polymer, with poor crystal quality and with the measurement settings; polarized light microscopy (PLM) can be more sensitive but cannot resolve very small domains; differential scanning calorimetry (DSC) is limited by the mass fraction and size of the domains and by the heating itself (refs 10–16) C2. The abstract summarises these methods as typically having detection or quantification limits "on the order of 1%" C36. The XRPD used in this paper has a limit of detection of 0.4% and a limit of quantification of 1%, taken from the group's earlier work P3.

Transmission electron microscopy (TEM) had recently been shown to detect crystalline phases inside ASDs (Ricarte et al. 2015, ref 17; Mugheirbi et al. 2017, ref 18) C3. Ricarte et al. found crystals by TEM in an X-ray amorphous physical mixture containing 3% crystalline drug R18. The difficulty is the specimen: electrons pass only through material roughly 100 nm thick (refs 19, 20) C3.

The paper builds directly on the group's companion study (Moseson and Taylor 2018, ref 6), which constructed a temperature–composition phase diagram for indomethacin (IDM) and polyvinylpyrrolidone/vinyl acetate copolymer (PVPVA) and turned it into a "process operating space" for HME (Figure 1b) R1. That work gave 131 °C as the formulation critical temperature $T_c$ of a 1:1 blend: the lowest temperature at which a homogeneous molecular dispersion can in theory form R2. At $T_c$ the drug cannot melt (IDM melts at 161 °C) and must dissolve into the polymer melt, which is both thermodynamically and kinetically limited; the authors note that samples made at this lower bound carry the greatest risk of residual crystals M1·C4. They therefore made an extrudate exactly at $T_c$ with a long residence time, confirmed that routine tools call it amorphous, and then searched it with TEM (and scanning electron microscopy, SEM) for crystals below the limits of those tools A1·C4·C7.

Background concepts

Formulation critical temperature and the process operating space

For a crystalline drug mixed with a polymer, the temperature at which the drug's solubility in the polymer equals its content is the lowest temperature at which a fully dissolved, homogeneous mixture can exist. The paper calls this the solubility temperature or formulation critical temperature $T_c$ and takes 131 °C for 1:1 IDM–PVPVA from ref 6 R2. Figure 1b maps processing temperature against residence time. Above the drug's melting point (161 °C) the drug melts; between $T_c$ and the melting point lies a "dissolution" region, in which the crystals dissolve into the molten polymer if they are given enough time; below $T_c$, and at short residence times above it, the drug stays as a suspension of crystals; at the top is thermal degradation, and operating limits bound the short and long residence times R1. The extrudate studied here is marked at 131 °C and 20 min, on the lower edge of the dissolution region, and the boundary of that region reaches 131 °C at 10 min R1. The figure is a schematic summarising samples characterised in ref 6, not new data R1.

What TEM sees in an ASD, and how Fourier analysis helps

In bright-field TEM the image is formed by electrons that pass straight through a thin specimen. Amorphous material scatters diffusely and looks uniform; a crystal diffracts electrons out of the beam when its lattice planes are suitably oriented, so it looks darker, shows diffraction contrast, and at high magnification shows lattice fringes, the planes of the crystal seen edge-on. The paper uses exactly these signs (darker areas, diffraction contrast, lattice planes) to recognise crystalline domains in the extrudate R9.

A Fast Fourier Transform (FFT) of a TEM image turns each set of regularly spaced fringes into a pair of sharp spots, placed at the spacing and orientation of the fringes; an amorphous region gives only a diffuse halo. Spots in the FFT therefore confirm crystallinity even when fringes are hard to see, and many spots spread round a ring indicate many small crystals in random orientations C3. Fourier filtering goes one step further: masking one symmetric spot pair and transforming back produces an image that contains only the periodic content of that spot pair, so the region where that particular lattice plane exists can be mapped across the crystal and shown in false colour (Gatan DigitalMicrograph, ImageJ) P8. The paper calls these "reconstructed" images; this article calls them lattice-plane maps.

The authors list why FFT often fails to confirm small crystals: the crystal may be oriented so that its spacings are not recorded, parts of the specimen may be too thick for a lattice image, the crystal may lie below the surface of the particle, its order may already be partly lost, and the amorphous drug–polymer matrix interferes C11.

Defects that appear in TEM images

Two kinds of feature carry the paper's argument about defects. An edge dislocation is an extra half-plane inserted into a crystal; in a fringe image or a lattice-plane map it shows as a fringe that stops inside the lattice. Bend contours are dark bands that cross a crystal image where a bent or strained crystal happens to satisfy the diffraction condition; they mark curvature. The paper interprets the bend contours it sees as curvature arising from strain at the crystal's interface (refs 22, 23) C17. A third feature is one lattice spacing appearing in several orientations within one region, which the authors read as planes sheared and rotated C19.

Why routine methods can miss small crystals

The paper gives three reasons for the limits of XRPD, each from the literature: dilution of the drug by the polymer lowers the intensity of the crystalline peaks (refs 11, 24); as crystallites shrink toward nanometres their peaks broaden (ref 25, the Scherrer relation); and strain from dislocations, stacking faults and grain boundaries further affects the peaks (ref 26) C26. Small, diluted and defective crystals, exactly what the paper finds, are therefore the hardest case for XRPD.

Methods in detail

The study has one formulation and one process condition. It contributes no new model; its contribution is a characterisation sequence that moves from bulk, low-resolution methods to nanometre-resolution imaging on the same extrudate A1.

Making the extrudate at the edge of the dissolution region

Indomethacin (ChemShuttle; melting point 161 °C as stated) and PVPVA (Kollidon VA64, BASF; Tg 104 °C as stated) were blended 1:1 by weight, corrected for water M1·M2·P1. The polymorphic form of the indomethacin and whether the two thermal values were measured in this work are not stated M1·M2. The blend was processed in a 5 mL co-rotating conveying twin-screw extruder (Xplore Pharma) at 20 rpm with the hopper cooled to 10 °C P1. The heating zone was set to 134 °C, which gave a bulk melt temperature of 131 °C at an in-line thermocouple placed after the screws, that is, $T_c$ P1·R2. The extruder has a recirculation channel (Figure 1a): with the valve open the melt circulated for 20 minutes before the valve was closed and the melt extruded; the rod then cooled at room temperature P1.

Why 20 minutes? An earlier extrudate made at 131 °C with 10 minutes of residence had been X-ray amorphous but showed some birefringence under PLM, that is, crystals below the XRPD detection limit of 0.4% but visible optically R4·C7. The residence time was doubled to dissolve those crystals C7. The paper does not show the data for the 10 minute sample or say whether it was made in this study or in ref 6 X9.

Part of the rod was cryomilled under liquid nitrogen (60 s at 10 Hz) and sieved into 100–250 µm and <100 µm fractions; rods and powders were kept in a refrigerator over desiccant until analysis P2.

Routine characterisation

XRPD used a Rigaku SmartLab in Bragg–Brentano geometry with Cu Kα radiation, scanning the 100–250 µm powder from 5 to 40° 2θ at 0.5° per minute in 0.02° steps; PVPVA, a physical mixture of unstated ratio and crystalline IDM were measured as references P3·A3. Modulated DSC (TA Q2000) ran 3–5 mg of the <100 µm powder from 0 to 180 °C at 5 °C/min with a modulation printed as "±0.796 every minute" (no unit given), followed by cooling and a second heating at 10 °C/min P4·X11. PLM used powder in mineral oil between crossed polarizers (20× objective), with at least five images per sample P5.

Preparing and imaging TEM specimens

No sectioning was done. Instead, the authors used the small fragments that had broken off the extrudate rods and collected inside the lid of the storage vial: a 300 mesh carbon-coated copper grid (5–6 nm carbon) was touched or gently wiped against the inside of the lid, and the fragments that stuck were imaged P6·M4. IDM particles were simply sprinkled onto a grid P6. Imaging was bright-field at 200 keV (FEI Tecnai G20, LaB6 source, 100 µm aperture), with spot size 1 for the drug crystals and spot size 3 for the extrudate; at least three grid squares were searched for particles thin enough for electrons P7. The microscope carried an energy-dispersive X-ray detector, but no EDX data are reported P7·X7. The electron dose, the magnifications and the number of extrudate particles examined are not stated ("multiple particles"); three are shown in Figures 4–6 P7·R9·X10.

Each micrograph was Fourier transformed; for selected spot pairs, inverse transforms gave the lattice-plane maps described above, false-coloured to show where each plane occurs (panels A–D in Figures 4d, 5c, 6d and 6f) P8.

SEM of the rod interior

A cross section of the rod was mounted with carbon tape, sputter-coated with platinum for 60 s and imaged in a field-emission SEM (FEI Nova nanoSEM) at 5 kV and about 5 mm working distance with the through-the-lens detector; IDM particles were imaged with the Everhart–Thornley detector P9. Domain lengths were measured on one SEM image (n = 100), and the 5–10 nm domains on one TEM image (n = 20); the measuring tool is not stated P10.

Experimental analysis

The starting crystals

Under PLM the IDM starting material consisted mainly of plate-like particles, tens of micrometres across, whose birefringence went dark on rotation; the authors describe them as single crystals coated with smaller crystals (Figure 2a) R3·C6. SEM showed small crystals sitting on the faces of larger plates and visible surface imperfections (Figure 2b) R3. A TEM image of one IDM particle, about 1.2 µm long, showed lattice planes in a variety of orientations and an FFT with discrete spots (Figure 2c–e) R3. The authors find these views consistent with one another: small crystals agglomerated on, or held electrostatically to, larger ones C6. These images serve as the reference for what intact drug crystals look like; note that they were taken at spot size 1, the extrudate at spot size 3 P7.

Routine tests: the extrudate looks amorphous

Figure 3 collects the routine evidence. The XRPD pattern of the extrudate powder is a single broad halo with a maximum near 21° 2θ and none of the sharp peaks that crystalline IDM and the physical mixture show (the strongest near 11.7, 19.7, 21.9, 26.7 and 29.4° 2θ, read from Figure 3a, approximate) R5. The reversing heat flow shows a single step, a composite Tg of 73.8 ± 2.6 °C (the statistic and n are not stated) R6·X11. The rod is clear (Figure 3d) R8. Under PLM the text reports "trace or no birefringence", while the caption of Figure 3c says that trace birefringence is observed in some images R7.

The authors conclude that these tools "supported this supposition", namely that the 20 minute residence had dissolved the crystals left after 10 minutes C7·C8. Note by the AI writer (not discussed in the paper). Read strictly, only XRPD, DSC and appearance are silent; the PLM caption already reports a faint positive signal, and the TEM and SEM results below show crystals in the same extrudate X3.

Table A. What each method reported for the same extrudate (from the text and Figures 3–7) P3·R5·R6·R7·R8·R9·R17.

TEM: crystals inside an X-ray amorphous extrudate

Every extrudate particle shown contains both amorphous regions, homogeneous and continuous, and crystalline regions R9·C9. The authors describe the particles as broadly similar but each with its own morphology, and identify two populations of "highly defective" nanometre-scale crystallites: discrete crystals smaller than 100 nm, which they describe as caught "mid-dissolution", and nanocrystalline domains of about 5–10 nm, commonly in clusters R9·C10.

Particle 1 (Figure 4). The particle is about 200 nm across (read from the 50 nm scale bar, approximate) and contains amorphous material and both populations R10. One region (Figure 4b) holds a cluster of small domains; its FFT shows spots spread round a ring, which the authors take as confirmation that many of the domains are crystalline and randomly oriented R11·C12. Twenty domains measured 5.8 ± 1.2 nm (± SD) R11. The clustering "suggests" to the authors that these domains came from one crystal and have not yet spread into the matrix, and the random orientation, similar size and clustering together "suggests a mechanical origin" C13·C14. The same particle holds a discrete crystal of about 26 × 70 nm (Figure 4c) with clear lattice fringes beside an amorphous region, confirmed by the FFT R12·C15. The graphical abstract is an annotated version of this particle R10.

Particle 2 (Figure 5). This particle, about 290 nm across, consists mostly of areas whose internal structure resembles crystalline domains R13. In a representative region one lattice spacing appears in three orientations (lattice-plane maps A–C), which "suggests" to the authors that the crystal planes have been sheared with rotation in three distinct orientations R13·C19. The three maps cover different parts of the region, and map C shows a boundary that the authors mark as a "dissolution front"; they state that each plane has a different level of dissolution R13·C19.

Particle 3 (Figure 6). This larger fragment, about 0.4 × 0.5 µm, has darker crystalline regions separated by lighter amorphous channels, with an amorphous region at its lower edge (Figure 6a) R14. The authors write that polymer "appears to have advanced" into the crystalline regions along these channels and "hypothesize" that the domains may have come from one discrete crystal now split into smaller, partly dissolved pieces C20. Close-ups show crystals with bend contours (Figures 6b–d), which the authors consider "likely" to come from crystal edges formed by dissolution together with strain at the interface with the amorphous matrix R15·C21. One field about 50 nm wide (Figure 6e) shows both populations side by side: seven 5–10 nm domains circled in yellow and two larger crystals circled in orange R16. These small domains are more scattered than the tight cluster of particle 1, which the authors read, "if also formed by the mechanical breakage mechanism", as more time having passed since breakage R16·C22. The crystal in the lower right of Figure 6e is called "~60 nm"; it shows one lattice spacing in several orientations and bend contours on one edge, again read as shearing and interfacial strain R16·C23.

Note by the AI writer (not discussed in the paper). Against the 10 nm scale bar of Figure 6e, the orange circle around that crystal measures about 26 × 23 nm and touches the edge of the panel, so the "~60 nm" size cannot be checked from the image; part of the crystal may lie outside the field X2.

Defects in the discrete crystals

The lattice-plane maps of the 26 × 70 nm crystal (Figure 4d) separate four lattice periodicities R12. Map A shows dark interference fringes labelled bend contours; maps B and C show lattice defects such as edge dislocations; maps C and D are two lattice planes that run through almost the whole crystal; maps A and B together form a third, in-focus plane R12. From this the authors "hypothesize" that the crystal, though defective, is still a single crystal, and state that the planes show dislocations and different degrees of "thermally- and/or mechanically-induced structural deformation" C18·C16. The bend contours, in their reading, "indicate curvature of the crystal" arising from strain at its interface C17.

Table B. Microstructural features and how the authors interpret them R11·R12·R13·R14·R15·R16·C13·C14·C16·C17·C19·C20·C21·C22.

Two points about this evidence. First, the defects are marked only in the discrete crystals; no defect is marked in the 5–10 nm domains, although the abstract states that both domain types contain a high defect density C10·X5. Second, no defect density is measured; "highly defective" is a qualitative description X5.

SEM of the rod interior

SEM of a cross section shows some regions with no clear surface features, which the authors call fully amorphous (Figure 7a), and other regions with embedded features that they call crystals (Figures 7b, 7c) R17·C24. In Figure 7c the features measured 38 ± 18 nm in length (n = 100) R17. In Figure 7b the largest feature is about 670 × 370 nm on the scale bar (approximate) R17. Domains below 10 nm could not be imaged R17.

The text gives the size range as 10–1000 µm, the caption as 10–1000 nm; the 500 nm and 250 nm scale bars and the 38 nm mean fit nanometres, so µm in the text is a misprint X1. SEM shows shape, not crystal structure, so the identification of these features as crystals rests on the TEM and PLM results rather than on SEM itself, even though the abstract says that PLM and SEM "corroborate" crystallinity C35·X8. The larger features also exceed the <100 nm TEM population and are not discussed further X8.

Table C. Sizes reported or read from scale bars; "about" values are read from the figures (approximate) R3·R10·R11·R12·R13·R14·R16·R17.

Discussion

How the authors explain the microstructure

The authors' explanation, in the section "Crystal dissolution initiated from defect sites", joins the images into one picture of what happens to a drug crystal in the extruder at $T_c$ C28·C29·C30. Drug crystals are "likely" to contain imperfections and local disorder at their surfaces and inside; these defects are "the likely sites" where a phase change starts under thermal or mechanical stress, because mixed crystalline and disordered material has a higher overall free energy (refs 28, 29) C28. Etching studies have shown that dissolution starts at defect sites and that its rate correlates with defect density, and crystal structure, solvent–solid interactions and solvent adsorption shape the surface dissolution mechanism (refs 30–34) C28. The highly defective domains "suggest" that defects, both those present in the starting crystals and those created by mechanical damage, propagate as the crystals dissolve into the amorphous matrix, so that crystals shrink both by dissolution of their outer surfaces and by fragmentation C29.

The authors' closing statement on mechanism is that, under HME at high temperature and shear, loss of crystallinity, which they equate with an increase in crystal defects, "is expected" as an intermediate stage of the crystalline-to-amorphous transformation by mechanical damage and dissolution into the molten polymer C30. In the conclusions they add that the residual crystals "show little similarity" to the bulk starting crystals, as a result of thermally and/or mechanically induced deformation and dissolution C32. None of these steps was followed in time: the picture is assembled from single images of fragments taken from one extrudate at one processing time C14·X10.

Why XRPD missed the crystals, and what "much greater sensitivity" means

The authors compare their result with Ricarte et al., who detected crystals by TEM in an X-ray amorphous physical mixture containing 3% crystalline drug, and state that "a much greater degree of sensitivity was found in this analysis" R18·C25. They then argue that in this ASD, and in many samples of their earlier work and the literature, crystalline content quantified by XRPD "may be systematically underestimated" because of dilution, small crystallite size and lattice strain (refs 6, 27) C27·C26.

Note by the AI writer (from the card's extractor notes). The direction of this argument is supported: XRPD shows a halo (limit of detection 0.4%) for an extrudate in which TEM finds crystals R5·P3·R9. Its size is not: no crystalline fraction was estimated by TEM or any other method, so the paper shows detection, not an amount, and the comparison with the 3% of Ricarte et al. is qualitative X4. The conclusions call the crystals "low levels", which is likewise not quantified C31·X4.

What the study suggests for formulation

The authors' own message is that an extrudate that XRPD and DSC call amorphous can still contain nanometre-scale crystals, and that high-resolution techniques such as TEM are "considered an important first step" towards understanding what residual crystal populations mean for ASD performance C31·C34. They leave open "whether or not any level, critical crystal size, or characteristic of residual crystals may be acceptable" without harming stability or dissolution, and note that the poor sensitivity of common techniques makes the question hard to answer C33.

Note by the AI writer (not discussed in the paper). For process design, the case is a cautionary one. The extrudate was made exactly at the temperature that the group's own process operating space marks as the lower edge of the dissolution region, with a residence time doubled to be safe, and it still held crystals R1·R2·P1·R9. A margin of temperature or time above $T_c$ may be needed when the target is a crystal-free product. Trace birefringence under PLM was the one routine signal that crystals remained, which argues for treating it as a positive finding rather than noise R7. Whether nanocrystals like these act as seeds for crystallization on storage or for precipitation during dissolution is the practical question, and this paper does not test it C33.

Limitations

Limits the authors state

The authors acknowledge that FFT often fails to confirm small crystals, for reasons of orientation, thickness, position below the surface, partial order and interference from the amorphous matrix, so TEM can miss crystalline domains C11. They state that SEM could not image domains smaller than 10 nm R17. They describe the significance of residual crystals for performance as an open question that this characterisation is only a "first step" towards answering C33·C34. Their mechanistic statements carry hedges throughout: "suggests", "we hypothesize", "likely", "may have", "is expected" C13·C14·C18·C20·C21·C30.

The AI writer's own judgement

• No amount of crystal. The paper establishes that crystals are present; it gives no crystalline fraction, so "low levels" and "much greater sensitivity" than 3% are not quantified X4.
• Sampling. TEM specimens were fragments collected from the vial lid; the number of fragments examined and how many contained crystals are not stated, and three are shown X10.
• Identity of the domains. No lattice spacing was indexed against IDM and no EDX result is shown, although a detector was fitted; that the crystals are IDM is inferred from the formulation X7.
• Where the defects come from. The starting crystals were imaged at spot size 1 and the extrudate at spot size 3, the electron dose is not given, and bending or damage of thin organic crystals during preparation or under the beam is not considered as a source of bend contours or disorder; the attribution of the defects to HME carries this caveat X6.
• Defects in the smallest domains. High defect density is shown for the discrete crystals but only asserted for the 5–10 nm domains X5.
• Inconsistencies in the text. The SEM range is given in µm in the text and nm in the caption (nm is right) X1; a crystal called "~60 nm" measures about 26 × 23 nm on its scale bar X2; the text says routine tests supported complete dissolution while its own caption reports trace birefringence X3; two cross-references print as "Error! Reference source not found." in this manuscript version X12.
• The SEM evidence. SEM features are called crystals without a crystallinity read-out, and the largest ones, up to about 0.7 µm, are larger than the TEM population and not discussed X8.
• Scope. One drug, one polymer, one composition and one process condition; the 10 minute comparison sample is reported without data A1·X9.

Overall the central observation, crystalline lattice fringes with FFT spots inside an extrudate that XRPD and DSC call amorphous, is direct and convincing for the fragments shown. The size classes are reasonably documented. The mechanism, fragmentation plus surface dissolution along defects, is a plausible reading of the images but remains untested R9·R11·R12·C29.

Conclusion

This short paper takes an IDM–PVPVA (1:1) extrudate made by HME at its formulation critical temperature of 131 °C with 20 minutes of recirculation, shows that XRPD (a halo), DSC (one Tg at 73.8 ± 2.6 °C) and visual inspection call it amorphous, and then finds by bright-field TEM with Fourier analysis that fragments of it contain crystals P1·R2·R5·R6·R8·R9. Two populations appear: discrete crystals below 100 nm, such as one of about 26 × 70 nm, and clusters of 5–10 nm domains, measured at 5.8 ± 1.2 nm R11·R12. Lattice-plane maps show edge dislocations, bend contours, sheared planes and a boundary marked as a dissolution front, and SEM shows embedded features of 38 ± 18 nm in parts of the rod R12·R13·R15·R17. The authors read the crystals as remnants caught part-way through dissolving into the melt and breaking up, a reading built from single images C29·C30. They argue that XRPD "may" systematically underestimate crystalline content and that high-resolution imaging is a first step towards deciding what level of residual crystallinity an ASD can tolerate, a question the paper leaves open C27·C33·C34.