Which phase holds the drug? Nanoscale infrared imaging of a phase-separated felodipine–poly(acrylic acid) film

Background and motivation

Polymers are added to drug formulations to change how fast the drug is released, and how the drug is dispersed in the polymer decides much of that behaviour. A drug can be mixed with a polymer at the molecular level, as nanometre or micrometre domains, or as a mixture of these. Which of these forms depends on manufacturing, for example the evaporation rate in solvent-based processes (Weuts 2011, ref 1), and on formulation variables such as drug–polymer miscibility (Marsac 2006, ref 2). The dispersion can also change over time: during storage the drug may phase separate or crystallize (Rumondor 2009, ref 3; Van Eerdenbrugh and Taylor 2010 and 2011, refs 4, 5), and during release the structure changes in ways that depend on whether the polymer dissolves, erodes, degrades or responds to its environment. To design such systems on purpose, the authors argue, one needs a technique that can characterise the micro- and nanostructure of a dosage form unambiguously and at high spatial resolution. C1·C2

The available techniques fall on either side of that need. Electron microscopy and atomic force microscopy (AFM) resolve nanometre features (Marsac 2010, ref 6; Lauer 2011, ref 7), but they cannot say unambiguously which organic material a feature is made of: an AFM height image shows bumps and valleys, not chemistry. Mid-infrared (mid-IR) imaging and Raman mapping identify chemical composition well, but their spatial resolution is limited by the diffraction limit of the light they use (Van Eerdenbrugh and Taylor 2011, ref 8; Padilla 2011, ref 9). The structure of drug–polymer systems at the nanoscale has therefore been hard to characterise, although it is the level at which processing, formulation and ageing would have to be linked to product performance. C2

Nanoscale mid-IR spectroscopy, in which an AFM probe detects the photothermal expansion caused by mid-IR absorption, offers both: the chemical information of IR and a spatial resolution not limited by the wavelength of the light (refs 10–27). By 2012 it had been used on cells, hemoglobin, bacterial storage polymers and viruses, organometallic compounds in cancer cells, polymer samples, quantum dots and inorganic materials. To the authors' knowledge it had not been applied to the micro- and nanostructure of drug–polymer mixtures. C3

The study therefore asks whether nanoscale mid-IR can characterise the miscibility of a drug–polymer system. The model is a solvent-cast blend of felodipine with poly(acrylic acid) (PAA; Figure 1). Felodipine, a dihydropyridine, has a relatively low tendency to crystallize when made amorphous by solvent evaporation or melt cooling (Van Eerdenbrugh 2010, ref 28; Baird 2010, ref 29), so the authors assume it stays amorphous during the experiment; PAA is amorphous and does not crystallize. C4 Earlier work from the group, with several techniques, found the felodipine–PAA pair partially miscible: blends were miscible at low and at high polymer content and phase separated at intermediate compositions, with a 90:10 (w/w) felodipine–PAA blend miscible and a 30:70 blend starting to separate (Rumondor 2009, ref 30). R1·C5 The plan was to follow the dispersion state across compositions with standard AFM, pick the most informative ratio, and on that film collect local spectra and maps at characteristic wavenumbers. A1·C11

Background concepts

Miscibility and phase separation in an amorphous blend. A drug and a polymer are miscible at a given composition when they form one amorphous phase. If they are only partially miscible, a blend inside the immiscible range splits into two amorphous phases, one rich in drug and one rich in polymer. Each phase is itself a mixture whose composition lies at the edge of the miscible range, so the drug-rich phase still contains some polymer and the polymer-rich phase still contains some drug. In a film the minority phase usually appears as discrete domains in a continuous majority phase. The question this paper answers is which of the two phases in such a film is which, and whether that can be shown chemically at the scale of the domains.

What AFM shows and what it cannot. In contact-mode AFM a sharp tip on a cantilever scans the surface and records its height. Phase-separated films show domains as raised or lowered regions, and the tip–sample interaction may differ between them. The images give the size and arrangement of the domains but not their composition; the authors say plainly that the composition of the phases cannot be identified from AFM images alone. C9

How AFM-IR works. In the nanoIR instrument used here, a tunable pulsed IR laser (pulses of 10 ns at a repetition rate of 1 kHz) illuminates the film from below through a zinc selenide (ZnSe) prism, in attenuated total reflectance (ATR) geometry. Where the film absorbs at the chosen wavenumber it heats and expands within nanoseconds. The AFM tip, in contact with the surface, feels this expansion as a kick that sets the cantilever ringing at its contact resonance (134–140 kHz here). The ringdown is converted to the frequency domain by a fast Fourier transform (FFT), and the area under the FFT within a window around the contact frequency is taken as the response, called the amplitude. P6·E2 Sweeping the laser wavenumber at one tip position gives a local IR spectrum; fixing the wavenumber and scanning the tip gives an absorption map. The laser spot is about 50 µm across, so the light does not localise the measurement; the tip does. The authors state that the spatial resolution is set by the AFM tip size but deteriorates for thicker samples, and estimate it at about 100 nm for samples about 500 nm thick. P6·C33 The spectral resolution is set by the laser line width, about 8 cm−1. P6

Reading IR bands. An IR spectrum is a set of absorption bands, each from a molecular vibration. To use AFM-IR spectra, the authors first assign the bands of the pure components between 1200 and 1800 cm−1 (Table 1). For PAA they take the assignments from the literature (Dong 1997, ref 31): the C=O stretch of the carboxylic acid near 1709–1712 cm−1, a CH2 deformation near 1454–1456 cm−1, and a C–O stretch coupled with O–H in-plane bending near 1416–1424 cm−1. For felodipine they calculate the vibrational modes from a geometry optimised in vacuum with a Hartree–Fock model and the 6-311+G** basis set, and assign only the most intense bands; the key band near 1488–1497 cm−1 is N–H bending coupled with CH2 vibrations, and the ester C=O stretches lie between 1682 and 1701 cm−1. P4·R7 Amorphous and crystalline felodipine differ in the 1600–1700 cm−1 region (ester C=O stretches at 1680–1700 cm−1 and dihydropyridine ring stretches at 1620–1645 cm−1), in peak breadth, and in the relative intensity of the bands near 1280 and 1305 cm−1. R5·C15

Why a ratio of two maps. The AFM-IR amplitude depends not only on how strongly the material absorbs, but also on how much it expands, which depends on its thermal expansion coefficient, its thickness, the surface roughness and the angle between probe and sample. C25 If such factors act the same way at two wavenumbers, dividing a map at one wavenumber by a map at the other cancels them and leaves the chemical contrast. C26 This is the logic behind the ratio map described below.

Methods in detail

The study has four steps: make thin films across the composition range, look at them with standard AFM, build reference spectra of the pure components, and then take local spectra and maps on the one blend that AFM shows to be phase separated with the smallest domains.

Films and their thickness

Felodipine (a gift from AstraZeneca) and PAA (Sigma–Aldrich; molecular weight printed as "1.800 g/mol", which can only mean 1800 g/mol) were dissolved in methanol at felodipine:PAA ratios of 0:100, 25:75, 50:50, 75:25 and 100:0 (w/w), keeping total solids at 10% (w/v). M1·M2·P1·X1 After complete dissolution, 100 µL of solution was deposited on a ZnSe ATR prism fixed in a custom holder and spun for 20 s at 3100 rpm, which leaves a thin film on the prism ready for AFM and AFM-IR. P1 To know how thick such a film is, the same protocol was repeated on 18 × 18 mm² glass cover slips, and the thickness was calculated from the weight gain using a density of 1.2 g/mL for the mixture. Six repeats gave an average thickness of approximately 0.57 ± 0.02 µm for the 50:50 film. P2·R8·E3

Reference spectra in bulk and at the nanoscale

For bulk spectra, films of pure PAA, pure amorphous felodipine and the 50:50 blend were spin coated in the same way on KRS-5 windows, and crystalline felodipine was ground with potassium bromide and pressed into a pellet. Spectra were collected in transmission on a Bio-Rad FTS 6000 (128 scans) and normalised after background subtraction. P3 For nanoscale spectra, the pure films on ZnSe were measured by AFM-IR, using a contact frequency of 140 kHz for amorphous and crystalline felodipine and 134 kHz for PAA, with detection windows of 60 and 20 kHz. Each nanoscale reference spectrum in Figure 3b is an average of 10 spectra. The laser power at 1500 and 1700 cm−1 was set to about 0.5 mW, spectra were recorded from 1200 to 1800 cm−1 at a spacing of 4 cm−1, and ringdowns from 128 laser pulses were averaged for each data point. P6·R6 How the crystalline felodipine sample for the nanoscale spectrum was prepared is not described; the methods give the KBr pellet only for the bulk spectrum. X2

Local spectra on the 50:50 film

On the 50:50 film, a 1.5 × 2.5 µm² area was imaged and eight positions were marked: four on raised discrete domains (positions 1–4) and four on the continuous phase between them (positions 5–8; Figure 4a). AFM-IR spectra were recorded at each position with a contact frequency of 137 kHz and a 40 kHz window, and the four spectra of each group were averaged (Figure 4c). P6·R9·A4 The 50:50 film was chosen because both AFM and the earlier miscibility study point to phase separation there, and because its domains (about 100 nm to a few hundred nanometres) are smaller than those at 75:25, which makes it the harder test of spatial resolution. C11·R3

Maps and the ratio image

Maps were recorded at 1500 and at 1700 cm−1 with a contact frequency of 137 kHz, a 20 kHz window, about 1 mW of laser power and a scan rate of 0.04 Hz. Spectra were spaced 50 nm apart on a 30 × 30 grid, which gives the 1.5 × 1.5 µm² area of Figure 5. P7 The two bands were chosen because all components respond at 1700 cm−1, while amorphous and crystalline felodipine respond much more than PAA at 1500 cm−1. C17 The ratio image, computed in Gwyddion, is defined in words in the paper as the intensities of the 1500 cm−1 map divided by those of the 1700 cm−1 map: P7·E1

In plain terms: at each pixel, the felodipine-specific signal is divided by a signal that drug and polymer both give, so a high ratio marks a felodipine-rich spot and a low ratio a PAA-rich spot, provided that topography and thermal expansion affect both maps in the same way, which the authors assume. E1·C26 Here "map intensity" is the AFM-IR amplitude at that pixel, the integrated FFT area of the cantilever ringdown described above. E2

Experimental analysis

AFM: domains appear and grow with felodipine content

Figure 2 shows the three blends over the same 5 × 3 µm² area and the same 250 nm height range. P5 The 25:75 film is smooth, with few topographical features and a uniform interaction with the tip, which the authors read as homogeneous mixing "at the scale of scrutiny", in agreement with the earlier finding that 30:70 only starts to phase separate. R2·R1·C6 The 50:50 film is more heterogeneous: discrete raised domains, the smallest about 100 nm across, sit in a continuous phase. R3·C7 At 75:25 the domains vary much more in size, from about 100 nm to several micrometres, dominate the image and often almost touch. R4 Several of the largest 75:25 domains also have a lower centre than their rim, which the text does not discuss. R4

Table A. Standard AFM topography of the three blends (Figure 2 and text; sizes as stated by the authors, not quantified further). R2·R3·R4

The authors take these images to "strongly suggest" that a phase-separated system evolves as felodipine content rises, and call the 75:25 film "apparently" phase separated. C8 Because the domains grow in area with felodipine content, they write that the domains "may be" felodipine rich and the continuous phase richer in PAA, but stress that AFM cannot confirm this. C10·C9 In the abstract the AFM series is summarised as suggesting "limited miscibility", in line with previous findings. C30

Nanoscale spectra reproduce bulk IR

Figure 3 and Table 1 compare bulk transmission spectra with averaged AFM-IR spectra. For felodipine the agreement is close: each nanoscale band lies within 9 cm−1 of the nearest bulk band, and some bulk doublets appear as one nanoscale band, for example the amorphous C=O stretches at 1682 and 1701 cm−1, which give one band at 1684 cm−1. R7.e–t·C13 Table B lists the bands with a single bulk value.

Table B. Selected bands from Table 1 of the paper; (a) amorphous, (c) crystalline felodipine. Rows with two or three bulk values are left out here; they are in the card record. R7

Two differences stand out. The ring band near 1210 cm−1 is dominant in the bulk felodipine spectra but weak or absent in the AFM-IR spectra, and for PAA the carbonyl band at 1712 cm−1 is less dominant in the nanoscale spectrum than in bulk. C13·C12·R6 The authors offer three possible reasons ("might"): transmission (bulk) versus ATR (nanoscale) geometry; a laser spot size that may change with wavenumber, since the background is normalised by total power while the probe responds to power density; and very low laser power at 1210 cm−1, which may push the probe response below the baseline. C14 Note by the AI writer (not discussed in the paper): the unassigned PAA band moves from 1267 cm−1 in bulk to 1280 cm−1 by AFM-IR, a 13 cm−1 difference, larger than the stated spectral resolution of about 8 cm−1, although the authors describe the PAA peaks as in good agreement. X7

Two observations carry the rest of the paper. First, the amorphous–crystalline differences in the 1600–1700 cm−1 region survive at the nanoscale, so the solid form of felodipine can be judged locally. C15 Second, in the nanoscale spectra all components respond at 1700 cm−1, but amorphous and crystalline felodipine respond much more than PAA at 1500 cm−1; in Figure 3b the PAA trace is close to its baseline there. R6·C17 The bulk spectrum of the 50:50 film, judged by the 1600–1700 cm−1 region, peak breadth and the relative heights near 1280 and 1305 cm−1, "can be regarded" as a combination of the amorphous felodipine and PAA spectra, "suggesting" that no crystalline felodipine is present. R5·C16

Domains versus matrix: the local spectra

The eight local spectra of the 50:50 film (Figure 4b) fall into two groups. On the continuous phase (positions 5–8) the band near 1700 cm−1 is the dominant band of the 1200–1800 cm−1 range and the 1500 cm−1 band is relatively weak. On the discrete domains (positions 1–4) the 1500 cm−1 band is somewhat higher than the 1700 cm−1 band. The pattern holds in each individual spectrum, and the averages in Figure 4c show it clearly. R9·R10·C18

Table C. Local AFM-IR spectra of the 50:50 film (Figure 4). The band-height ratios are read from the average spectra of Figure 4c (approximate; heights above the minimum near 1550–1600 cm−1 of normalised, offset spectra); the paper states the comparison only in words. R9·R10

Because the 1500 cm−1 band is characteristic of felodipine and essentially absent in PAA, the authors "can now unambiguously conclude" that the discrete domains are a felodipine-rich phase embedded in a PAA-rich continuous phase. C19 The abstract phrases the second half more softly: the continuous phase "tended to be rich in PAA". C19 The conclusion is qualitative; no composition is derived for either phase. X5

Each phase holds some of the other, and the drug is amorphous

Both groups of spectra also show the characteristic bands of the other component: the matrix spectra keep a clear 1500 cm−1 felodipine band, and the domain spectra keep a 1700 cm−1 band. The authors read this as "in line with" a partially miscible system. R9·R10·C20 They argue that it is expected: since the 25:75 film looked homogeneous, the PAA-rich phase should hold at least 25% felodipine, and since a 90:10 felodipine–PAA dispersion was miscible in earlier work, the felodipine-rich phase can hold some PAA. C21·R2·R1 They add that "some prudency is in order", because a certain degree of spectral mixing between the two phases "cannot be excluded" even if the phases were completely separated. C22 Note by the AI writer (not discussed in the paper): in AFM-IR each PAA band lies close to a felodipine band (1712 near the felodipine C=O bands at 1684–1696, 1456 and 1424 near 1436–1444, and 1280 at 1280 cm−1), so PAA in the felodipine-rich domains can show only as a change in relative band heights, which the paper does not quantify. X9·R7

Comparing the local spectra with the amorphous and crystalline references, the authors conclude that felodipine "has not yet crystallized", consistent with the bulk spectrum and with the low crystallization tendency of felodipine. C23 The time between film preparation and measurement is not stated. X3

The maps: why one wavenumber is not enough

Figure 5 shows a 1.5 × 1.5 µm² area: the topography (height scale 250 nm), the AFM-IR map at 1500 cm−1 (amplitude scale 0.20–0.75 V) and the 1500:1700 ratio map (scale 0.9–1.9). R11·R13 In the 1500 cm−1 map the amplitude is high (green to red) over the raised domains and low (blue) in the surrounding continuous phase, and the authors note that its patterns "correspond well" to the topography. R11·C24 The map at 1700 cm−1, however, shows similar patterns (data not shown). R12 A felodipine-specific band and a shared band giving the same picture means that the single-wavenumber contrast cannot be read as chemistry alone. The authors reason that topography "may" also affect the thermal response: the continuous phase forms valleys around the domains, and roughness, the probe–sample angle, the sample thickness and the thermal expansion coefficient of each material can all change the signal, with higher thermal expansion usually giving a larger IR response. C25

Assuming these effects are wavelength independent between 1500 and 1700 cm−1 ("it can be assumed"), the authors recommend that conclusions rest on relative responses from images at several wavenumbers rather than on one, here the 1500:1700 ratio. C26·C27 In the ratio map, blue regions "tend to" correspond to the continuous phase and green and red regions to the discrete domains. The authors see "clearly" a good correspondence with the topography, which they say "again confirms" felodipine-rich domains in a PAA-rich continuous phase and "further demonstrates" that the technique can characterise drug–polymer systems chemically at submicrometre resolution. R13·C28·C29 Note by the AI writer (from the card's extractor notes): the ratio map is noisier than the 1500 cm−1 map. Most pixels are green (about 1.4–1.6 on the colour bar), the low-ratio blue areas run mainly as horizontal bands along the scan rows, and a few single pixels are red; the correspondence with the topography is shown side by side, without an overlay, line profile or correlation. R13·X6

Discussion

What the paper establishes. The paper's argument runs in a clear order. AFM shows that domains form at 50:50 and 75:25 but not at 25:75, and their growth with felodipine content hints that they are drug-rich, which AFM itself cannot confirm. C8·C9·C10 Reference spectra show that AFM-IR reproduces bulk band positions closely enough to read with bulk assignments, and identify a band that only felodipine gives strongly (1500 cm−1) and one that both components give (1700 cm−1). C13·C17 Local spectra on the 50:50 film then show the felodipine band relatively strong on the domains and weak in the matrix, which the authors read as felodipine-rich domains in a PAA-rich continuous phase, both phases containing some of the other component. C19·C20 Finally, because single-wavenumber maps also carry topography, they propose ratio maps and find that the ratio map broadly follows the domains. C25·C27·C28 The authors' own summary is that combining AFM with the "photothermal AFM coupled IR method" made it possible to interrogate both the micro- and nanostructure and the chemical composition of the two phases, and that data processing produced chemical composition maps with submicrometre resolution. C31·C29

How this fits the earlier picture of felodipine–PAA. The AFM series agrees with the earlier miscibility study (ref 30): homogeneous at high polymer content, phase separated at intermediate compositions. C6·C30·R1 The new information is chemical: the discrete domains, not the matrix, are the felodipine-rich phase, and the felodipine in them is amorphous. C19·C23 The expected compositions of the two phases, at least 25% felodipine in the PAA-rich phase and some PAA in the felodipine-rich phase, follow from the miscibility limits but were not measured. C21

What it suggests for formulation. The authors present the work as a proof of concept that "will provide the framework" for more in-depth studies relating microstructure, processing and product performance. C32 Note by the AI writer (not discussed in the paper): for amorphous solid dispersion development, the immediate value is diagnostic. AFM or DSC can show that a film has phase separated, but not which domains carry the drug or whether that drug is still amorphous; AFM-IR, read as band ratios, answers both at roughly the 100 nm scale for thin films. The method as used here gives enrichment, not composition: putting numbers on the phases would need calibration films of known composition and an estimate of how much signal comes from the neighbouring phase. The films were about 0.57 µm thick and spin coated; how far the approach carries to spray-dried particles or extrudates is not tested here. R8·X5

Limitations

Limits the authors state. The authors stress that AFM images alone cannot identify the composition of the phases. C9 For the local spectra they write that "some prudency is in order", since spectral mixing between the phases "cannot be excluded". C22 They give the spatial resolution as an estimate ("we estimate"), about 100 nm for samples about 500 nm thick, and note that it deteriorates for thicker samples. C33 They point out that single-wavenumber maps carry a topographic contribution, and their correction rests on the assumption that it is wavelength independent ("it can be assumed"). C25·C26 The causes of the bulk–nanoscale spectral differences are offered as possibilities ("might"). C14

The AI writer's own judgement (1): size of the evidence. The phase identification rests on eight spectra, four per phase, in one 1.5 × 2.5 µm² area, and on one 1.5 × 1.5 µm² map, of one 50:50 film; no replicate film is reported, and the 25:75 and 75:25 films were examined by AFM only. X5·R9·A4 The wording "unambiguously conclude" fits the qualitative contrast in Figure 4, which is consistent across all eight spectra, but not a statement about composition. C19·X5 The ratio map adds less than its description suggests: it is noisy, its low-ratio areas follow the scan rows, and its match to the topography is judged by eye. X6 The claim that each phase contains the other component is weaker still, because PAA has no band in this range that felodipine does not overlap, and because the authors themselves allow spectral mixing. X9·C22

The AI writer's own judgement (2): unstated conditions and small inconsistencies. The smallest 50:50 domains (about 100 nm) are about the size of the estimated resolution and two image pixels, so features at that scale sit at the limit of the method; the film thickness used for the resolution estimate was measured on glass cover slips rather than on the ZnSe prisms that carried the imaged films. X4·R8·P7 The time and conditions between film preparation and measurement are not given, although phase separation and crystallization can change with time, and the authors' own wording is that felodipine "has not yet crystallized". X3·C23 The preparation of the crystalline felodipine sample measured by AFM-IR is not described. X2 The detection settings differed between samples (contact frequencies of 134, 137 and 140 kHz with different windows, and about 1 mW for maps against about 0.5 mW for spectra), so only band ratios within a spectrum, not absolute amplitudes, can be compared across samples. X8 The PAA molecular weight is printed as "1.800 g/mol", which must be read as 1800 g/mol. X1 None of these changes the qualitative conclusion that the domains are felodipine-rich.

Conclusion

This short methods paper makes three contributions. It shows that AFM-IR spectra of a small-molecule drug and a polymer reproduce bulk IR band positions closely, including the bands that distinguish amorphous from crystalline felodipine. C13·C15 It uses a felodipine-specific band at 1500 cm−1 against a shared band at 1700 cm−1 to show, at single domains of a 50:50 felodipine–PAA film, that the domains are felodipine-rich and amorphous and the matrix PAA-rich, both phases holding some of the other component. C19·C20·C23 And it shows that single-wavenumber AFM-IR maps of a rough film follow topography, and proposes ratio maps at two wavenumbers to separate chemistry from topography. C25·C27·C28 The evidence is qualitative and comes from one film, but the approach, and the caution about single-wavenumber maps, apply to any amorphous dispersion film thin enough for AFM-IR. X5·C29