Seeing drug nanocrystals inside a polymer film with an AFM tip that feels infrared absorption

Background and motivation

Many dosage forms are composites of a drug and a polymer: the coatings of drug-eluting stents (Kamath 2006, ref 1) and the nanoparticle formulations that the New Jersey co-authors make by milling, for example drug nanoparticles held in HPMC films (Sievens-Figueroa 2012, ref 2) or coated onto carrier particles (Bhakay 2013, ref 3). In such products the polymer choice and the composite microstructure set the release kinetics and, in the end, the bioavailability, so the formulator needs to know where the drug is and in what form, down to the micro- and nanoscale (refs 4–7). C4

The standard chemical tool, Fourier transform infrared (FTIR) spectroscopy, is rich in chemical and structural information and is widely used on drug–polymer composites (Van Eerdenbrugh 2011, ref 4; Van Eerdenbrugh 2012, ref 8). Its weakness is space: the optical diffraction limit restricts conventional FTIR imaging to a resolution of 3–30 µm (refs 9–12), so anything that happens at the scale of a nanocrystal is averaged away. Atomic force microscopy (AFM) has the opposite profile: its topography and phase-contrast images show nanoscale structure, but they do not say what a feature is made of. C5·R14

Two ways of combining the two had appeared. Infrared scattering scanning near-field optical microscopy (sSNOM) reads the light scattered by a sharp AFM tip and had imaged polymer blends, polymer–inorganic systems and nanoparticles in biological materials at about 20 nm (refs 13–17). Its signal, however, is about six times weaker than that of absorption-based infrared spectroscopy (Levin 2005, ref 18), and turning the scattered light into something comparable with an FTIR spectrum takes substantial post-processing, especially for unknown samples (Knoll 1999, ref 19). C6·R15 Photothermal induced resonance (PTIR), also sold as AFM-IR, instead measures absorption itself. It gives spectra directly comparable with FTIR at nanoscale resolution and had been used on polymer blends (refs 8, 24–31) and living cells (refs 20–22, 32, 33). C7 In pharmaceutics, the Taylor group had applied it to phase separation in a drug–polymer blend (Van Eerdenbrugh 2012, ref 8) and to the miscibility of dextran or maltodextrin with poly(vinylpyrrolidone) (Van Eerdenbrugh 2012, ref 31).

What had not been tested is the case a formulator meets with nanocrystal products: discrete crystalline drug particles of around 100 nm embedded in a solid polymer. The paper sets out to evaluate PTIR for the chemical and mechanical characterisation of such particles, using griseofulvin (GF) milled into hydroxypropyl methyl cellulose (HPMC, hypromellose); to check that particle sizes taken from PTIR images agree with dynamic light scattering (DLS); and to record spectra well above the noise for particles of about 100, 350 and 800 nm. C1·C2·C27

Background concepts

How PTIR works. The sample is deposited on a zinc selenide (ZnSe) prism, which is transparent in the mid-infrared. A tunable optical parametric oscillator (OPO) laser sends nanosecond pulses through the prism. When the pulse wavelength matches an absorption band of the material under the tip, the material absorbs, heats and expands for a moment. An AFM cantilever in contact with the surface is kicked by this expansion and rings; the Fourier transform of its deflection gives its vibration modes, and the maximum amplitude of those modes, recorded as the laser is tuned, is a local infrared spectrum (Dazzi 2010, ref 23). C8 The authors describe these spectra as able to be "highly sensitive and quantitative" when read at the cantilever's harmonic modes. C8 In this paper the PTIR signal is the peak-to-peak amplitude of the cantilever oscillation, normalised by background spectra that track the laser power. P6 Holding the laser at one band and scanning the tip gives a chemical map; holding the tip still and sweeping the laser gives a point spectrum. P6·P7

Why the resolution is not set by the infrared wavelength. In conventional infrared microscopy the light itself must be focused onto the feature, which limits resolution to the micrometre scale. R14 In PTIR the local reading is taken by the tip in contact with the sample, and the paper treats spatial detail as a matter of the tip and the sample rather than the wavelength: the authors suggest that the resolution on these films "may be limited" by the size of the tip. C14 They also note that the signal grows with local film thickness, and that differences in the thickness and thermomechanical properties of the matrix around a particle can blur the picture through thermal artifacts (refs 26, 29, 32, 36). C12·C15

Contact resonance and stiffness maps. A cantilever touching a surface behaves like a spring and damper; its contact resonant frequency rises on stiffer material. Tracking that frequency while scanning gives a stiffness map at the same time as the PTIR map (Dazzi 2012, ref 12). C8·P8 In this paper the nominal contact resonance was 394 kHz and the stiffness images span 364 to 424 kHz, with darker areas softer. P8 The authors stress that a stiffness map does not depend on infrared absorptivity. C22

Wet stirred media milling. Wet stirred media milling (WSMM) grinds drug crystals with small beads in a stabiliser solution to make a nanosuspension. Here the stabiliser was HPMC, used as a neutral polymeric stabiliser, with sodium dodecyl sulfate (SDS) as a wetting agent and for its known synergistic stabilising action with cellulosic polymers (refs 2, 3, 34). C9·P1·P2 The same HPMC then serves as the film matrix after spin-casting. M2

Particle size by two methods. DLS measures particles moving in a liquid and reports a distribution of equivalent sphere diameters; the paper plots it as a percentage of the total volume and reports the median, d50. P3·C25 From an image, a particle's size can be expressed as an equivalent diameter: the diameter of a circular disc with the same area as the particle's projected area. E2·C25 The authors compare the two. A note by the AI writer (not discussed in the paper): a size distribution can be weighted by number (each particle counts once) or by volume (each particle counts by its volume); for the same particles the volume-weighted median is larger, often much larger when the distribution is broad, so a fair comparison needs the same weighting on both sides.

Methods in detail

The study has three parts: making two griseofulvin nanosuspensions of different size and casting them into thin HPMC films; imaging and recording spectra of the films by PTIR, with stiffness maps for the finer film; and checking the particle sizes read from one PTIR image against DLS. The flow below follows the paper's order.

Part 1: Two nanosuspensions and two thin films

Problem. To test resolution, the authors needed drug particles of known size, embedded in a polymer, in a layer thin enough for AFM. Griseofulvin, a poorly water-soluble drug (EP/BP grade; water solubility given as 8.9 mg/l), was the model; HPMC was both the stabiliser and the matrix. M1·M2

What they did. Sample 1 was a suspension of 10% GF, 2.5% HPMC E15LV and 0.5% (w/w) SDS made by the standard WSMM process of ref 2. P1 Sample 2 contained 10% GF, 2.5% HPMC E3 (the lower-viscosity grade of the same chemistry) and 0.2% (w/w) SDS and came from a "novel, intensified" process using wear-resistant yttrium-stabilized zirconia beads of 100 µm, milled for 6 h. P2·M3·M6 The authors state that, to their knowledge, a median below 100 nm had not been reached by WSMM before. C10 Both suspensions were stirred at 300 rpm for 30 min and sized by DLS after dilution with water (1:19 for sample 1, 1:30 for sample 2): d50 was 335.7 nm for sample 1 and 85 nm for sample 2. P3·R1

For the films, 0.5 mL of sample 1 was added to 4 mL of an aqueous solution of 2.5% HPMC E15LV and 0.5% SDS (called a 1:8 dilution), and 0.05 mL of sample 2 to 2 mL of the same solution (1:40). Each was spin-cast onto a ZnSe prism with a custom holder at 33.3 Hz (2000 rpm) for 20 s. P4 The film thickness, read from the drop of the tip at the edge of the coating, averaged about 500 nm for sample 1 and about 100 nm for sample 2; the authors take the lowest heights in the AFM images as this film level and the highest heights as embedded griseofulvin clusters. P5·R2·C11

Note by the AI writer (not discussed in the paper): the dry-film drug content is not printed. From the recipes, taking all liquids as 1 g/mL and the percentages as w/w, griseofulvin is roughly 27 wt% of the dry solids in the sample 1 film and roughly 7.5 wt% in the sample 2 film; the sample 2 film also contains HPMC E15LV and SDS from the dilution solution, so its matrix holds both HPMC grades (card note X7).

Part 2: PTIR maps, spectra and stiffness

Instrument and settings. All measurements used an Anasys Instruments nanoIR AFM with C-450 silicon cantilevers in contact mode. Ten background spectra were averaged before each set of experiments to normalise the signal for laser power. The laser was tuned to a griseofulvin band, 1710 cm⁻¹ (a C=O band) for sample 1 and 1625 cm⁻¹ (a C=C band) for sample 2, with band assignments taken from Page 1962 (ref 35). P6·R16 Images of sample 1 were read on the cantilever's first mode (61.0 kHz) and images of sample 2 on its third mode (394 kHz), following Felts 2013 (ref 25). P6 Each image pixel co-averaged 16 spectra and images were scanned at 0.1–0.12 Hz. P6

Table A. Image settings (from the PTIR Imaging section of the methods). P6

Point spectra over 1800–1200 cm⁻¹ co-averaged 128 laser pulses per wavenumber in 4 cm⁻¹ steps and took 3 minutes (Figure 1c); spectra over 3600–1000 cm⁻¹ co-averaged 256 pulses in 6.5 cm⁻¹ steps and took 6.5 minutes (Figure 4). P7 The contact resonant frequency was recorded with the PTIR maps of sample 2 to give the stiffness images of Figures 2c and 2f. P8

Why this design. The coarser sample 1, with agglomerates near a micrometre, tests whether the PTIR map identifies the drug at all. The finer sample 2, in a thinner film and at a finer pixel spacing, probes how small a particle can still be seen. A1·A3 Note by the AI writer: the two samples differ at once in milling process, HPMC grade, SDS level, dilution, film thickness, imaging band and cantilever mode, so differences in contrast between Figures 1 and 2 cannot be assigned to particle size alone (card note X8).

Part 3: Particle size from a PTIR image

Problem. Because the PTIR signal depends on how much material absorbs under the tip, one may ask whether particle outlines in a PTIR map are distorted by uneven infrared absorptivity. The authors test this by measuring a size distribution from a PTIR map and comparing it with DLS of the same suspension. C2

What they did. In MATLAB, each pixel of the 2.5 × 2.5 µm map of Figure 2e was normalised by subtracting the image's minimum PTIR value and dividing by its maximum PTIR value. P9·E1 Areas were then sorted into four classes by normalised signal: large (0.45–1), medium large (0.38–0.45), medium small (0.31–0.38) and small (0.2–0.31) particles (Figure 3b). P9 Each projected area was converted into the diameter of a circular disc of equal area, and the distribution of these equivalent diameters was plotted against the DLS distribution, which assumes spherical particles (Figure 3a). E2·C25·R10

Note by the AI writer: as worded, dividing by the maximum after subtracting the minimum does not give exactly 0–1; the basis of the four thresholds, the number of particles and whether the nested bands of one bright area (Figure 3b shows red cores inside orange rings with green and light-blue rims) count as one particle or several are not stated (card note X3).

Experimental analysis

Main result 1: the drug band lights up the agglomerates of sample 1. The 2 × 2 µm height image of sample 1 (Figure 1a) shows several micron-sized agglomerates in the polymer; the three largest are about 0.6–0.9 µm across. R3 In the PTIR map at 1710 cm⁻¹, a band the authors describe as unique to the griseofulvin ketone groups, the same features are bright and the space between them is dark (Figure 1b). R4·R16 A spectrum taken on a particle shows the two C=O bands at 1710 and 1664 cm⁻¹ and the C=C band at 1625 cm⁻¹, the last being the strongest; a spectrum taken in the matrix shows none of them (Figure 1c). R5 The authors conclude that the agreement between the topographical and chemical images "confirms that the particles are GF". C13

Table B. Bands in the sample 1 spectra (Figure 1c). The first three rows are from the text; the last row is read from Figure 1c (approximate). R5·R16

Table B makes a practical point that the paper shows but does not dwell on: below about 1500 cm⁻¹ both spectra carry the same broad bands near 1460 and 1340 cm⁻¹, so only the 1700–1600 cm⁻¹ region separates drug from matrix. R5 The authors make the same choice when they later describe the spectra as distinguishable from HPMC "for the spectral region 1700–1600 cm⁻¹". C27

Main result 2: the signal follows thickness. In Figure 1 the infrared intensity generally varies with local topography: thicker regions and taller structures give stronger signals, which the authors read as the known thickness dependence of the PTIR signal (Mayet 2008, ref 22). R4·C12 The tall agglomerates even look sharper and higher in contrast in the PTIR map than in the height map. The authors call this an "unexpected" increase in apparent resolution and explain it by the greater sensitivity of PTIR than contact-mode topography to sample thickness; the gain is seen only in the high regions. C16 The lower regions have lower contrast, "suggesting" that resolution and contrast fall as the height difference between matrix and particles falls. C17 The close match of topography and PTIR also "suggests" to them that the resolution "may be limited" by the size of the tip. C14 These observations are why they turned to smaller particles in a thinner film. C17

Main result 3: a 90 nm particle in the map of sample 2. The 10 × 10 µm height image of sample 2 (Figure 2a) shows many particles in the HPMC matrix. R6 In the 1625 cm⁻¹ PTIR map (Figure 2b) blue marks low infrared absorptivity, and pink, yellow and red mark the griseofulvin particles. R7·C18 Together, Figures 2a–c show a broad size distribution, from tens of nanometres, "likely" primary particles, to about 1 µm, "likely" agglomerates. C19 In the magnified 2.5 × 2.5 µm images, at a pixel spacing of about 10 nm, a 90 nm particle (solid white circle) is distinguishable in both the height and the PTIR maps, and the authors write that this "demonstrates that sub-100 nm GF particles can be chemically characterized" with PTIR. R8·C20 In Figure 2e this particle is a small spot of higher signal beside a larger particle; a 100 nm particle in a dashed circle is also marked in the figure but not discussed in the text. R8

The stiffness maps (Figures 2c and 2f) show the particles as distinct regions, which the authors describe as "noticeably stiffer" than the surrounding HPMC; they write that these maps "confirm the presence of crystalline GF" and resolve particles better, because the contact resonance shift does not depend on absorptivity. R9·C21·C22 Each particle is ringed by a softer zone. The authors say its nature "is not clear" and hypothesise that these are surfactant- and water-rich regions formed while the films dried. R9·C23 They conclude that PTIR measurements near the resolution limit "can be greatly enhanced" by also analysing these contact-resonance shifts. C24 No resonance frequencies for particles or matrix are given; in the figure, the particle regions are speckled mixtures of bright and dark pixels with dark rims, and in the 2.5 µm map the matrix shows a network of lighter lines. R9·X11

Main result 4: sizes from the PTIR map resemble DLS. Figure 3a sets the image-analysis histogram against the DLS curve of the same suspension. The DLS curve is broad, rises from about 40 nm, peaks at about 13.5 vol% near 90 nm and returns to zero near 460 nm, with a median of 85 nm. R10 The histogram has 19 bars from about 36 to 570 nm, with a median of 96 nm. R10 The authors find that "the general shape shows reasonable agreement with the DLS curve", "suggesting" that PTIR on sub-100 nm particles gives a reasonable representation of the particle distribution in spite of the heterogeneous sample; in the abstract they add that the PTIR measurements are therefore not significantly affected by uneven infrared absorptivity in this system. C26·C2

Table C. The size comparison of Figure 3a. d50 values from the text; ranges and peaks read from Figure 3a (approximate). R1·R10·P3·P9

Note by the AI writer (not discussed in the paper; card note X12): every bar of the image-analysis histogram, read at high zoom, is a whole multiple of about 0.76 vol%, that is 1/132 of the total (bar heights of about 1.5, 2.3, 3.8, 4.5, 5.3, 8.3, 9.0 and 11.3 vol% correspond to 2, 3, 5, 6, 7, 11, 12 and 15 units). A volume-weighted histogram over 36–570 nm cannot be quantised like this, because particles in different bars have very different volumes; the pattern fits a count of about 132 particles. Their summed projected area, about 3.1 µm², would be about half of the 6.25 µm² map, which is plausible for Figure 3b, and their median read from the bars, about 100–105 nm, is close to the printed 96 nm. Weighted by volume, the same counts would have a median of roughly 400 nm. If the histogram is in fact a number distribution, Figure 3a compares a number-weighted image distribution with a volume-weighted DLS curve, and the agreement of 96 with 85 nm is not like for like. R10·X12

Main result 5: spectra of particles of about 100 nm stand above the noise. Figure 4 shows spectra from 3600 to 1000 cm⁻¹ taken on particles of about 800, 350 and 100 nm and on the HPMC matrix of sample 2, with the four spots marked on a 2.5 × 2.5 µm height image. R12·R17 All four spectra carry bands near 3450, 2900, 1460, 1370 and 1150–1050 cm⁻¹ that the HPMC spectrum shows too; only the particle spectra show the griseofulvin bands near 1710, 1660, 1620 and 1590 cm⁻¹, largest for the 800 nm particle. R12 The authors write that the griseofulvin peaks are "readily distinguishable" from the baseline noise for all three particles and from the HPMC spectrum in 1700–1600 cm⁻¹. C27 At 1625 cm⁻¹ they report the following values. R13

Table D. Signal-to-noise ratio and drug-to-matrix signal ratio, as printed in the text (Results and Discussion). R13

From these values the authors conclude that griseofulvin is "quantifiably different from the background signal" and that "these results indicate" that PTIR can chemically characterise local areas of heterogeneity "to at least the 100 nm scale" in drug–polymer systems. C27·C28 Note by the AI writer: the paper does not say how the S/N or the ratios were calculated; the HPMC spectrum shows no band at 1625 cm⁻¹ beyond noise, so the ratio is taken against a noise-level value; and Figure 4 draws the spectra at different baselines without a numerical scale, so the ratios cannot be checked from the figure (card note X6). The particle diameters for Figure 4 are given without a method, and since the inset height scale ends at 100 nm in a film of about 100 nm, they are lateral sizes; the 100 nm spot lies next to the large agglomerate used for the 800 nm spectrum (card note X9).

Discussion

What the paper establishes. Within its single system the core demonstration is direct. Griseofulvin has bands between 1700 and 1600 cm⁻¹ where HPMC does not absorb, and at those bands PTIR maps show the drug particles where AFM topography shows features, in both films and at two bands. R4·R5·R7·R12 The spatial detail reached, a 90 nm particle in a map and about 100 nm particles in spectra, is well below the 3–30 µm of conventional FTIR imaging. R8·R13·R14 The authors place this as the first chemical characterisation of sub-100 nm organic crystals. C1

Where the approach applies. The authors limit their own claim: the resolution of PTIR depends on the infrared absorptivity and the thermomechanical properties of the components, so "the true resolution will vary for other samples". C29 For materials with sufficient differences in "electronic and mechanical properties", they see "great promise" for both nanoparticle and nano- or microstructural characterisation of drug delivery systems, and they end by stating that PTIR "can be used to unambiguously chemically characterize drug formulations" at these length scales. C30·C3 Two conditions visible in the data frame this: a drug band free of polymer absorption (here only 1700–1600 cm⁻¹), and attention to film thickness, because the signal rises with local thickness. R5·R12·C12

What it suggests for formulation (AI reading, not stated in the paper). For nanocrystal formulations dried into films, coatings or matrices, the method can show whether milled particles stay single or form agglomerates in the dry product: Figures 1 and 2 show both in the same films. Brightness in a PTIR map is not simply drug amount, because thickness adds to it; an independent size or composition check, as the DLS comparison attempted here, remains necessary. And the stiffness map, although only qualitative here, is a cheap second channel that comes with every PTIR scan.

Open questions. The authors leave open what the soft rims around each particle are, and how resolution will change for other drug–polymer pairs. C23·C29 Beyond these, the paper does not show a spectrum of a sub-100 nm particle, does not measure the crystallinity of the embedded particles, and rests the size comparison on one map. X1·X2·X3

Limitations

Limitations stated by the authors. The authors note that variations in the thickness and thermo-mechanical properties of the matrix around a particle "can reduce" the resolution through thermal artifacts, that in sample 1 the PTIR contrast falls in low regions, and that the resolution "may be limited" by the tip size. C15·C17·C14 They state that resolution depends on absorptivity and thermomechanical properties and will vary for other samples. C29 They also leave the nature of the soft regions around the particles unexplained ("it is hypothesized" that they are surfactant- and water-rich). C23

The AI writer's own judgement (1): what was shown at 90 nm. The abstract says that "infrared spectra of 90 nm griseofulvin particles were obtained". C1 The 90 nm particle appears only in maps at one wavenumber and in the stiffness map; the smallest particle with a spectrum is the one of about 100 nm in Figure 4. R8·R12·X1 The data therefore show sub-100 nm chemical imaging at a single band, and spectra at about 100 nm.

The AI writer's own judgement (2): crystallinity is assumed. The title speaks of nanocrystals and the stiffness maps are said to "confirm the presence of crystalline GF". C21 No diffraction, thermal or polarised-light measurement of the milled or embedded particles is reported, and stiffness alone does not distinguish crystalline from amorphous drug. X2

The AI writer's own judgement (3): the size comparison is weaker than it looks. It rests on one 2.5 µm map, with thresholds whose basis is not stated and a segmentation whose handling of nested signal bands is unclear. X3 It compares a suspension in water with particles in a dried film, by two methods that measure different quantities. X4 The histogram's bar heights suggest a number count although the axis says vol%, which would make the comparison mix weightings; weighted by volume, the image distribution would sit far above the DLS curve. X12 The broader reading, that uneven absorptivity does not significantly affect PTIR in this system, therefore rests on a comparison that is itself uncertain, while the paper's own Figure 1 shows the signal following thickness. X5·C12

The AI writer's own judgement (4): reporting and scope. The S/N and drug-to-matrix ratios lack a stated method and cannot be checked from Figure 4. X6 The two samples differ in many factors at once, and no replicate images are reported. X8·X3 The closing statement that PTIR can "unambiguously" characterise drug formulations goes beyond one drug in one polymer, a gap the authors' own caveat acknowledges. X10·C29 None of this removes the central demonstration, which is direct, but it sets its limits: one system, single maps, and quantitative claims that need checking before reuse.

Conclusion

The paper makes three contributions. First, it shows that a PTIR map at a griseofulvin band identifies drug agglomerates in an HPMC film, confirmed by point spectra that carry the drug's C=O and C=C bands where the matrix has none. R4·R5·C13 Second, in a thinner film of a finer suspension it resolves a 90 nm particle in a map and records drug spectra from particles of about 100 nm with a stated S/N of 5:1, with contact-resonance stiffness maps as a second channel. R8·R9·R13 Third, it compares a size distribution from one PTIR map with DLS and finds medians of 96 and 85 nm, which the authors read as "reasonable agreement". R10·C26

For a formulation scientist the method answers a question that bulk tools cannot: where, and in what clusters, drug nanoparticles sit inside a dried polymer product. The conditions are a drug band clear of the polymer's spectrum and awareness that the signal follows thickness. R5·C12 The authors' general claim of "unambiguous" characterisation should be read as shown for griseofulvin in HPMC; extending it needs other drug–polymer pairs, solid-state proof that the particles are crystalline, and a size comparison with matched weighting. C3·X10·X2·X12