Which polymer sits where: nanoscale infrared imaging of dextran and maltodextrin blends with PVP

Background and motivation

Whether two amorphous components mix into one phase matters in several kinds of pharmaceutical product. In an amorphous solid dispersion, a drug that is molecularly mixed with its polymer carrier is less likely to crystallize than one that has separated into a drug-rich phase, because more polymer is around each drug molecule to inhibit crystallization. In freeze-dried products, additives may help protect a protein, often the active agent, against chemical and physical changes during lyophilization and later storage. The authors name both kinds of product as systems where the miscibility of blends matters (Introduction, refs 1–5) C1.

The standard test is differential scanning calorimetry (DSC). A miscible blend is expected to show one glass transition temperature (Tg), and two Tg events are read as two phases. The authors list five ways this reading can fail. When the two components have similar Tgs, the transitions cannot be resolved. A phase-separated domain must exceed a minimal size, given as 10–50 nm for polymer blends, before it shows its own Tg. A transition can be too broad or too weak to see. Miscible polymer blends have been shown to give two Tgs. And DSC must heat the sample above its Tg, which can itself change how the components mix C2.

Other techniques fill some of these gaps but bring their own problems. Bulk mid-infrared spectroscopy and X-ray powder diffraction measure the whole sample, so they return a sum of the responses of both phases. Raman mapping can image phases, but the diffraction limit of light restricts it to domains of the order of micrometers. AFM and electron microscopy resolve much smaller structures but cannot tell which chemical phase is which. Local thermal analysis is destructive and needs many repeats C3. A method that combines the spatial resolution of AFM with chemical identification would close this gap. Tip-enhanced Raman spectroscopy and nanoscale mid-infrared spectroscopy (AFM-IR) are the two candidates named; the first may be better suited to crystalline materials, and initial AFM-IR work, including the group's own study of felodipine in poly(acrylic acid) (Van Eerdenbrugh 2012, ref 38), suggested that it is promising for phase separation in amorphous blends C4.

To test a miscibility method fairly, one needs blends of the same two components that range from miscible to clearly immiscible, so that chemistry does not change along with miscibility. Dextran and maltodextrin with poly(vinylpyrrolidone) (PVP) provide this. All three polymers are amorphous, so crystallization does not interfere, and the group's earlier DSC study (Van Eerdenbrugh and Taylor, Pharm. Res. 2012, ref 39) had shown that the molecular weight (MW) of the polymers moves 50/50 blends from completely miscible to virtually immiscible C5·R1. This paper uses that set of blends to ask whether AFM topography and AFM-IR agree with DSC, and what extra they reveal about the size, shape and composition of the domains.

Background concepts

Miscibility, phase separation and the single-Tg rule

Two amorphous polymers are miscible when they form one homogeneous phase at the scale that matters. When they are not, the blend separates into two phases, each rich in one polymer; one phase often forms discrete domains (droplets) inside a continuous phase (the matrix). DSC infers the number of phases from the number of glass transitions. This works only if each phase is large enough and different enough in Tg to give its own transition; the rule is an inference, not a direct observation of the phases C2.

How AFM-IR reads chemistry below the diffraction limit

In AFM-IR, a thin film sits on a zinc selenide (ZnSe) prism. An infrared laser fires short pulses (10 ns at 1 kHz) through the prism into the film. Where the film absorbs at the chosen wavenumber, it warms and expands slightly. An AFM tip in contact with the surface feels this as a kick that sets the cantilever ringing (the ringdown). The amplitude of the ringdown, plotted against wavenumber, is a local infrared spectrum. Because the tip, not the laser spot (about 50 µm), defines where the signal comes from, the spatial resolution is far better than the micrometer-scale diffraction limit of mid-infrared light C4·P6. The paper defines the signal precisely: the ringdown is converted to the frequency domain by a fast Fourier transform, and the integrated area around the cantilever's contact frequency, within a set window, is the response E2.

The model polymers

PVP was used in three Kollidon grades: K 12 (PVP12, 2,000–3,000 Da), K 25 (PVP25, 28,000–34,000 Da) and K 90 (PVP90, 1,000,000–1,500,000 Da). The sugar polymers, together called DEX, were a maltodextrin of 783 g/mol (DEX1) and dextrans of 6,000 (DEX6), 40,000 (DEX40) and 400,000–500,000 Da (DEX450); all MWs are the manufacturers' values (Table 1) M1·M2·M3·M4·M5·M6·M7. Dextran has a backbone of glucose units linked 1→6 with a few short 1→3-linked side chains; maltodextrin has a 1→4-linked backbone with 1→6-linked branches and is treated here as a low-MW analogue of dextran C6. In infrared, PVP is dominated by its carbonyl (C=O) stretching band near 1660–1680 cm⁻¹, while DEX has its strongest band in this range near 1350 cm⁻¹ R5.

Table 2 of the paper, taken from the earlier DSC study, gives the starting point: all DEX450 blends (with PVP90, PVP25 and PVP12) and both DEX40 blends (with PVP90 and PVP12) were immiscible at 50/50, and both DEX6 and both DEX1 blends were miscible. One entry is an inference: 50/50 DEX40–PVP12 showed a single Tg, but the 40/60 and 60/40 blends showed two, which the authors take as "strongly suggesting" that the 50/50 blend is immiscible as well R1·C8. From the same study they also note that PVP is less miscible in DEX than DEX is in PVP R2. Miscibility thus "tends to decrease" as DEX MW rises C7.

Methods in detail

The study has one experimental idea, applied systematically: image each blend film by AFM, take infrared spectra at chosen points on it, and compare the result with what DSC had said. The flow below shows the stages.

Making the films

Each blend started as an aqueous solution of 10% (w/v) total solids, made by mixing solutions of the pure polymers in volume ratios equal to the 50/50 weight ratio P1. For AFM-IR, 100 µL of solution was placed on a ZnSe attenuated total reflectance prism held in a custom holder and spun for 20 s at 3,100 rpm, leaving a thin film P4. The paper states no drying step for these films, and the AFM-IR measurements were made without humidity control; the reference films for bulk infrared, by contrast, were spun on KRS-5 at 8,000 rpm for 20 s, stored overnight under dry nitrogen (<10% relative humidity) and measured under dry air P2·P3·P4. Nine blends were made: DEX450 with PVP90, PVP25 and PVP12; DEX40, DEX6 and DEX1 each with PVP90 and PVP12 A1·A2.

Reference spectra

Bulk transmission spectra of the pure polymers were recorded from 500 to 4,000 cm⁻¹ at 4 cm⁻¹ resolution with 128 scans, background-subtracted and normalized P3. AFM-IR spectra of the same polymers were averaged over ten positions each. These spectra define the bands used later to tell the phases apart (Figure 2) R5.

Topography and local spectra

All AFM work was done in contact mode in the nanoIR instrument, with plane-fitted height images P5. For local spectra, the cantilever contact frequency was 132–137 kHz with a window of 20 or 40 kHz; the laser delivered about 0.5 mW at the prism in a spot of about 50 µm; spectra ran from 1200 to 1800 cm⁻¹ with a 4 cm⁻¹ data spacing, a spectral resolution of about 8 cm⁻¹ set by the laser line width, and 128 coaverages per point P6. In practice the authors drew a line of marked positions across a domain boundary on the height image, took one spectrum at each position, and averaged the spectra by phase. The spacing was about 2.5 µm for DEX450–PVP90 and DEX40–PVP90, about 4 µm for DEX450–PVP25 and about 0.5 µm for DEX40–PVP12, and the number of positions per phase ranged from 4 to 13 P6·R9·R12·R16·R22.

Chemical maps and the ratio image

For one blend, DEX40–PVP90, the authors mapped the infrared signal at two fixed wavenumbers. They chose 1280 and 1350 cm⁻¹ because pure DEX responds more strongly at 1350 than at 1280 cm⁻¹, and PVP the other way round C31. Each map covered 25 µm × 15 µm with 512 × 64 measurements, so points were about 50 nm apart along x and about 225 nm apart along y, with 16 coaverages per point and a scan rate of 0.08 Hz P7. Raw maps were then divided, point by point:

In plain terms: the ratio is high where PVP dominates and low where DEX dominates. The paper gives this relation in words, not as a numbered equation E1. The reason for using a ratio rather than either map alone is that local height and mechanical properties also change the measured intensity, an effect the group had seen in its felodipine–poly(acrylic acid) study; dividing two maps of the same spot cancels much of that C32.

The benchmark

The DSC classification against which everything is compared comes from the earlier study (ref 39), not from new measurements. Those samples were prepared by lyophilization and measured by heating, whereas the AFM-IR films were spin-coated at room temperature P9·P4. The authors themselves raise this difference when the two methods disagree (see Results) C36.

Results

Reference spectra: retained bands, a shifted carbonyl and an unequal signal

The bands of the pure polymers seen by bulk infrared reappear in AFM-IR, slightly shifted (Table A). The authors call the features "well retained" C9. The PVP carbonyl band moves from 1680–1682 cm⁻¹ in bulk to 1660 cm⁻¹ in AFM-IR; because the bulk films were measured under dry air and the AFM-IR films without humidity control, they consider hydrogen bonding with absorbed moisture "the most plausible explanation" R5·C10. Film water content was not measured.

Table A. Band positions of the pure polymers in the 1200–1800 cm⁻¹ range, as stated in the text (Section "DEX-PVP Systems as Model Systems", p.1462; spectra in Figure 2).

One observation does not show in the normalized spectra but shapes everything that follows: PVP films gave much stronger ringdown responses than DEX films, with maximum intensities differing by up to an order of magnitude R6. In a blend, PVP therefore speaks louder than DEX, and together with partial miscibility this makes the phases harder to distinguish chemically C11. The pure-polymer height images showed no contrast (Figure 2C for PVP90), and the authors use such featureless images as the reference for a single phase R7·C12.

The DEX450 series: large layers that shrink as PVP gets shorter

DEX450–PVP90, the blend of the two longest polymers, was phase separated with large domains by optical microscopy (images not shown), as in the earlier study R8·C13. On the AFM height image (Figure 3A, 0–500 nm scale) a thick band runs horizontally through the centre, with thinner domains above and below. Inside both, smaller domains of the other phase are visible, most clearly in the thinner domains (the arrow in Figure 3A); the authors read this as a double emulsion ("it appears") R9·C14. Spectra at the 7 positions on the thinner domains showed no DEX bands in the 1200–1500 cm⁻¹ region, so these domains are PVP-rich ("it is clear"); spectra at the 13 positions on the thick band carried the DEX band at 1352–1364 cm⁻¹, which "suggests" a DEX-rich phase R10·C15.

The DEX-rich spectra still show a strong PVP carbonyl band; in Figure 3B it is the tallest band, with the DEX-region bands at about half its height. The authors expected only a limited amount of PVP in the DEX-rich phase from the DSC work and warn that, because the PVP signal is disproportionate to its concentration, "it cannot be concluded" that the DEX phase contains a large amount of PVP. The band at 1352–1364 cm⁻¹ is the diagnostic sign of DEX R10·R6·C16.

DEX450–PVP25 looked the same with somewhat smaller domains: phase separated by optical microscopy (not shown), and on the height image (Figure 4A, 0–750 nm) a thick central domain between thinner ones, with inclusions in both phases that were less distinct than in DEX450–PVP90 R11·R12·C17. The thinner phase (6 positions) was PVP-rich and the thicker phase (4 positions) DEX-rich, again with a non-negligible PVP carbonyl band R13·C18.

DEX450–PVP12 was expected to be more miscible: in the DSC work its phases had Tgs further from those of the pure polymers (most of all the PVP-rich phase), and blends at lower ratios on both sides of the composition range were miscible R3·C19. The height images (Figure 5A, a 50 × 30 µm area on a 0–200 nm scale; Figure 5B, 5 × 3 µm on 0–100 nm) show the change: thin, submicrometer discrete domains now sit in a thicker continuous phase R14·C20. Read against the 1 µm scale bar in Figure 5B, these domains are about 0.2–0.7 µm across R14. The authors attribute the smaller domains to the shorter PVP, which "likely has higher mobility and lower viscosity" and so separates into smaller, more localized domains; mobility and viscosity were not measured C21.

Here the spectra become harder to read. Figure 5C shows one spectrum in a discrete domain (green), one in the matrix (red) and one at a position near a phase boundary (yellow). The DEX band near 1350 cm⁻¹ is strongest in the matrix, weakest in the domain and intermediate near the boundary, so the discrete domains are PVP-rich and the matrix DEX-rich ("it is apparent") R15·C22. But all three spectra are dominated by the PVP carbonyl band and nearly overlap; the authors call the discriminating power "clearly challenged" and give three reasons: the appreciable mutual miscibility of the two phases, the larger PVP signal, and spectral mixing near phase boundaries, seen as intermediate spectra ("suggesting some extent of spectral mixing") R15·C23. They add that optimizing sample thickness "might" improve this, and judge the current measurements "adequate to enable chemical identification of each phase" C24·C25.

The PVP90 series: DEX-rich droplets in a PVP-rich matrix

DEX40–PVP90 is listed as immiscible in Table 2, and its DSC Tgs were very similar to those of the pure polymers R1·R4·C28. AFM shows round discrete domains of about 5 to 10 µm diameter that stand higher than the thinner continuous phase, and many smaller discrete domains, from a few hundred nanometers to a few micrometers, inside that phase (Figure 6A–C, 0–250 nm scale) R16. Spectra on the large domains (4 positions) were DEX-rich and spectra on the matrix (6 positions) PVP-rich (Figure 6D) R17·C29. Compared with the DEX450 blends, the arrangement is reversed: the DEX-rich phase is now the dispersed one. The authors conclude that "it appears that the higher MW polymer tends to form the continuous phase" C30.

This blend also gave the clearest imaging result (Figure 7). The 1280 cm⁻¹ map (0.15–0.95 V) was brighter in the matrix and the 1350 cm⁻¹ map (0.3–0.7 V) brighter on the domains, as expected for DEX-rich domains in a PVP-rich matrix R18·C33. The ratio map (0.25–2.75) shows the large domains as low-ratio (blue) islands in a high-ratio matrix, in "good agreement" with the height image R18·C33. In the magnified ratio map (Figure 7F), the small domains inside the matrix also show lower ratios than their surroundings, which the authors take as evidence that they too are DEX-rich, and as "a further demonstration of the submicrometer chemical imaging capabilities of the technique" R18·C34. The graphical abstract shows a topography and ratio image of this kind side by side.

DEX6–PVP90 is where the methods disagree. DSC had found the 50/50 blend completely miscible, but the AFM images (Figure 8A–B, on a 0–75 nm scale) show discrete domains, about 0.3–0.6 µm across when read against the scale bar, in a thinner continuous phase; the features in the centre and lower right of Figure 8B are damage from the cantilever R1·R19. The authors write that the images "contradict" the DSC result, "suggesting some degree of immiscibility at this composition" C35. The six local spectra, taken at differently coloured positions, showed no differences: each looks like the spectrum of pure PVP90, with little intensity in the DEX region R20. Mutual miscibility, the unequal response and spectral mixing "may underlie" this failure C38. From the thickness trend seen elsewhere, the authors conclude that the system "most likely" consists of DEX-rich domains in a PVP-rich matrix C37.

For the mismatch with DSC they offer two possible reasons, neither tested. Lyophilization for DSC and rapid evaporation of water at room temperature by spin-coating are different preparation regimes that "may influence both the thermodynamics and kinetics of phase separation"; and DSC is run at elevated temperatures, "which may promote mixing" C36.

DEX1–PVP90 gave homogeneous height images, like Figure 2C, and identical spectra at all positions; the data are not shown, and the blend is described as completely miscible, in line with DSC R21·C39.

The PVP12 series: PVP-rich droplets in a DEX-rich matrix

DEX40–PVP12, classed immiscible by DSC on the strength of its neighbouring compositions, showed discrete domains dispersed in a thicker continuous phase (Figure 9A, 0–750 nm; Figure 9B, 0–500 nm) R1·R22·C40. The authors describe these domains as smaller than those of DEX450–PVP12 and DEX40–PVP90 C27. Note by the AI writer: read against the scale bars, the DEX40–PVP12 domains (about 0.4–2 µm) are smaller than those of DEX40–PVP90 but not smaller than those of DEX450–PVP12 (about 0.2–0.7 µm) in the images shown (extractor note X1). The spectral differences between the domain (4 positions) and the matrix (6 positions) were subtle. The authors compared the relative intensity of the band below 1300 cm⁻¹ with that of the band near 1350 cm⁻¹: this ratio is relatively high for pure PVP12 and relatively low for pure DEX40, and on that basis the domains are PVP-rich and the matrix DEX-rich ("it appears") R23·C41. In both blend spectra of Figure 9C the band near 1440 cm⁻¹ is about as tall as the carbonyl band, and the difference between the two phases lies mainly in the height of the band near 1290 cm⁻¹ relative to the 1350–1380 cm⁻¹ region R23.

DEX6–PVP12 and DEX1–PVP12 were single-phase in all evaluations, with observations like those for DEX1–PVP90 (data not shown) R24·C42.

Across the blends: three patterns

Table B collects what each blend showed. Three patterns emerge, and the authors state all three.

Table B. Summary of the nine 50/50 blends, compiled by the AI writer from Table 2, Figures 3–9 and the text; the DEX6–PVP90 domain size is read against the scale bar of Figure 8B (approximate).

The DEX-rich phase is the thicker one. In all five blends where the phases were identified spectrally, the DEX-rich phase stood higher in the height image, whether it formed layers, the matrix or the domains. The authors state that DEX-rich phases "tend to form thicker" domains or layers upon spin-coating C26. They then use this rule to assign the phases of DEX6–PVP90, where the spectra could not C37.

Domains get smaller as either polymer gets shorter. With DEX450, the domains go from large layers (PVP90) through somewhat smaller ones (PVP25) to submicrometer domains (PVP12); with PVP90, from the DEX450 layers to 5–10 µm domains (DEX40) and to domains of about 0.3–0.6 µm (DEX6, read from Figure 8B). The authors summarize that domains "tend to become smaller" as PVP MW falls and that the same happens when DEX MW is lowered C27·R9·R14·R16·R19. The one comparison that does not follow in the images shown, DEX40–PVP12 against DEX450–PVP12, is noted above X1.

The shorter polymer forms the domains. Four blends have a clear matrix. In DEX450–PVP12 and DEX40–PVP12 the matrix is DEX-rich and the domains PVP-rich; in DEX40–PVP90 and DEX6–PVP90 the reverse holds, which the authors call phase inversion C43. Checked against the molecular weights of Table 1, "without exception" the discrete domains are rich in the polymer with the lowest MW C30. The same was seen for felodipine in poly(acrylic acid), where the drug formed the domains and the longer polymer the matrix C44. The authors' explanation is hedged: the overall structure formation during spin-coating "is complex", and the preference of lower-MW components for discrete domains "may be due to their higher mobility and lower viscosity" C45.

Discussion

What AFM and AFM-IR add to DSC

The authors' overall judgement is that "good agreement" exists between the miscibility inferred from AFM with AFM-IR and that from DSC, and that AFM-IR adds what neither DSC nor plain AFM can give: chemical information at submicrometer resolution, which they call "unique" C47·C49. In the results section they go further and write that the results "clearly indicate" that both AFM and AFM-IR "can be considered as powerful tools" for blend miscibility, able to tell miscible from immiscible blends, to describe the size and distribution of the phases, and to identify the major component of each phase. The abstract is more careful: the combination "appears to be a promising technique for the evaluation of miscibility in certain pharmaceutical blends", and the conclusions call nanoscale mid-IR "an important complement" to traditional probes C46·C48.

The scope of "identify" deserves attention. What AFM-IR delivers here is the major component of each phase, "at least to the extent of identifying the major component of a given phase", in the authors' words C46. It does not give phase compositions: the PVP signal is up to ten times stronger than the DEX signal, so band heights cannot be read as amounts, and the authors explicitly decline to infer the PVP content of the DEX-rich phase from its carbonyl band R6·C16.

When the methods disagree

Two blends show where DSC and AFM part ways, and they point in the same direction. DEX6–PVP90, completely miscible by DSC, shows domains in the spin-coated film R19·C35. DEX40–PVP12 is listed as immiscible in Table 2, but its own 50/50 DSC trace showed only one Tg; AFM shows two phases R1·R22·C8. In both cases the film looks less mixed than the DSC trace suggests. The authors discuss only DEX6–PVP90 and offer two hedged reasons: different sample preparation, which "may influence both the thermodynamics and kinetics of phase separation", and the heating in DSC, which "may promote mixing" C36. They do not test either, for example by imaging lyophilized material or running DSC on spin-coated films.

How the authors' observations connect, as drawn by the AI writer from the paper's statements (C7, C21, C23, C38, C45). Green and red mark a clear and a weak measurement outcome, not a formulation outcome.

What it suggests for formulation

The authors frame the work around amorphous solid dispersions and freeze-dried products, but test only polymer–polymer blends. Note by the AI writer (not discussed in the paper): three practical points follow from the data. First, a single-phase DSC result on material made one way did not guarantee a single-phase film made another way; for dispersions produced by solvent evaporation, miscibility is best checked on material made by the same route. Second, in these blends the lower-MW component formed the dispersed domains; together with the felodipine–poly(acrylic acid) case cited by the authors, this suggests that a separating low-MW drug in a long polymer will tend to appear as drug-rich droplets rather than as the matrix. Third, AFM-IR is a qualitative phase-identity tool in this form; a quantitative reading of phase composition would need a calibration of the very different signal strengths of the components.

Limitations

Limitations stated by the authors. The authors are open about the limits of the spectra. In DEX450–PVP12 the discriminating power is "clearly challenged", because of appreciable mutual miscibility, the different magnitude of the response of the two components and spectral mixing near phase boundaries; the same factors "may underlie" the complete failure in DEX6–PVP90 C23·C38. A strong PVP carbonyl band in a DEX-rich phase does not mean a high PVP content C16. Sample thickness "might" be optimized to improve discrimination C24. They acknowledge that structure formation during spin-coating is complex, and that their DSC benchmark used lyophilized samples measured at elevated temperature C45·C36.

The AI writer's own judgement (1): the benchmark and the missing evidence. The comparison is between methods and also between samples. The DSC classification is cited from ref 39 and was obtained on freeze-dried material that was heated in the calorimeter, while the films were spin-coated with no drying step stated and probed without humidity control; the authors themselves attribute the carbonyl shift in the films to absorbed moisture, and film water content was not measured X4·C10. The single-phase verdicts for DEX1–PVP90, DEX6–PVP12 and DEX1–PVP12, and the optical micrographs of the two DEX450 blends, are "data not shown", so half of the agreement with DSC rests on statements rather than on displayed evidence X8. Counting the 50/50 DSC traces themselves rather than the reported classes, AFM differs from DSC in two of nine blends, not one X5.

The AI writer's own judgement (2): how far the generalisations reach. The rule that discrete domains are rich in the lower-MW polymer rests on four blends, of which three were identified by spectra and one, DEX6–PVP90, only by the thickness rule, which was itself derived from the other blends X2. The DEX6–PVP90 phase separation that contradicts DSC rests on topographic contrast alone X3. The size trend does not hold in the images for DEX40–PVP12 against DEX450–PVP12, and every size comes from one or two images per blend without a stated number of fields X1. Film thickness is not given, so "thicker" compares phases only within one image, and the number of spectra behind the DEX450–PVP12 and DEX6–PVP90 panels is not stated although Figure 5 is captioned as average spectra X7·X6. Finally, Figure 1 carries no panel letters although its caption names three panels; the structures can be told apart by their glycosidic links X9. None of this undermines the main demonstration, which is clear for the strongly separated blends: AFM-IR assigns the phases of DEX450–PVP90, DEX450–PVP25 and DEX40–PVP90 unambiguously, and the ratio map of DEX40–PVP90 extends that assignment to submicrometer domains R10·R13·R17·R18.

Conclusion

This paper tests AFM topography and nanoscale mid-infrared spectroscopy on nine 50/50 blends of PVP with dextran or maltodextrin whose miscibility, by DSC, ranged from complete to none. AFM reproduced the DSC verdict for eight blends and found two phases in the ninth, DEX6–PVP90, which DSC had called miscible C47·C35. Local spectra identified a PVP-rich and a DEX-rich phase wherever the phases differed enough, and a 1280/1350 cm⁻¹ ratio map resolved DEX-rich domains well below a micrometer R10·R17·R18·C34. Along the way, the films showed regular patterns: the DEX-rich phase was always the thicker one, domains shrank as polymer MW fell, and the shorter polymer formed the dispersed phase C26·C27·C30. The method's limit is equally clear: as lower MW makes the phases more alike, and because PVP gives a much stronger signal than DEX, the spectral contrast fades from clear to subtle to absent C23·C38·R6.