Raman imaging of drug delivery systems.

Advanced Drug Delivery Reviews 89 (2015) 21–41

Contents lists available at ScienceDirect

Advanced Drug Delivery Reviews journal homepage: www.elsevier.com/locate/addr

Raman imaging of drug delivery systems☆ Geoffrey P.S. Smith a, Cushla M. McGoverin b, Sara J. Fraser a, Keith C. Gordon a,⁎ a MacDiarmid Institute for Advanced Materials and Nanotechnology, Dodd-Walls Centre for Photonic and Quantum Technologies, Department of Chemistry, University of Otago, Dunedin, New Zealand b Dodd-Walls Centre for Photonic and Quantum Technologies, Department of Physics, University of Auckland, New Zealand

a r t i c l e

i n f o

Available online 26 January 2015 Keywords: Raman imaging Drug delivery Chemometrics Multivariate methods Pre-processing Solid dispersions Drug dissolution Polymeric microparticles

a b s t r a c t This review article includes an introduction to the principals of Raman spectroscopy, an outline of the experimental systems used for Raman imaging and the associated important considerations and limitations of this method. Common spectral analysis methods are briefly described and examples of interesting published studies which utilised Raman imaging of pharmaceutical and biomedical devices are discussed, along with summary tables of the literature at this point in time. © 2015 Elsevier B.V. All rights reserved.

Contents 1.

2.

3.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1. Principal of Raman spectroscopy. . . . . . . . . . . . . . . . . . . 1.2. Experimental systems — Raman microscopy . . . . . . . . . . . . . 1.3. Imaging methods . . . . . . . . . . . . . . . . . . . . . . . . . 1.4. Advantages of using Raman imaging . . . . . . . . . . . . . . . . . 1.5. Shortcomings of Raman imaging. . . . . . . . . . . . . . . . . . . 1.5.1. Spatial resolution . . . . . . . . . . . . . . . . . . . . . 1.5.2. Lack of Raman signal and fluorescence . . . . . . . . . . . . 1.5.3. Subsampling . . . . . . . . . . . . . . . . . . . . . . . 1.5.4. Spherical aberration/refraction . . . . . . . . . . . . . . . 1.5.5. Out of focus contributions and relative signal strength . . . . . Spectral analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1. Cosmic events and de-noising . . . . . . . . . . . . . . . . 2.1.2. Baseline correction and normalising . . . . . . . . . . . . . 2.2. Data exploration and quantification . . . . . . . . . . . . . . . . . 2.2.1. Univariate methods . . . . . . . . . . . . . . . . . . . . 2.3. Multivariate methods. . . . . . . . . . . . . . . . . . . . . . . . 2.3.1. Principal component analysis (PCA) . . . . . . . . . . . . . 2.3.2. Classical least squares (CLS). . . . . . . . . . . . . . . . . 2.3.3. Partial least squares (PLS) . . . . . . . . . . . . . . . . . 2.3.4. Multiple curve resolution/multivariate curve resolution (MCR) . 2.3.5. Band target entropy minimization (BTEM) . . . . . . . . . . 2.3.6. Classification and segmentation . . . . . . . . . . . . . . . Case studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Solid dispersions — including tablets . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

☆ This review is part of the Advanced Drug Delivery Reviews theme issue on “Pharmaceutical applications of Raman spectroscopy — from diagnosis to therapeutics”. ⁎ Corresponding author. Tel.: +64 3 4797599; fax: +64 3 479 7906. E-mail address: [email protected] (K.C. Gordon).

http://dx.doi.org/10.1016/j.addr.2015.01.005 0169-409X/© 2015 Elsevier B.V. All rights reserved.

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . .

22 22 22 23 24 24 24 25 25 26 26 28 29 29 30 30 30 30 30 31 31 31 31 31 32 32

22

G.P.S. Smith et al. / Advanced Drug Delivery Reviews 89 (2015) 21–41

3.2. Drug-eluting coatings . . . . . 3.3. Polymeric microparticles . . . 3.4. Other solid dosage forms . . . 3.5. Drug dissolution . . . . . . . 3.6. Percutaneous drug formulations 3.7. Cells. . . . . . . . . . . . . 4. Conclusions. . . . . . . . . . . . . References . . . . . . . . . . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

1. Introduction Raman spectroscopy is a technique in which scattered light is used to interrogate the nature of molecules within an irradiated volume. It is a technique that has developed significantly with advances in laser and detector technologies [1]. It is well suited to the analysis of materials with micrometre length scales, consequently Raman microscopes that are able to measure micron structures are now off-the-shelf instruments. In this review we explore the important aspects of Raman imaging with respect to biomedical devices, solid dispersions, percutaneous drug delivery, drug dissolution, intracellular drug distribution and microparticle composition. We include consideration of the important optical and physical properties associated with imaging as well as discuss the extensive use of chemometric techniques for image data analysis. We have focused on spontaneous Raman spectroscopy rather that the surfeit of interesting techniques that use non-linear optical responses, such as femtosecond stimulated Raman spectroscopy [2] and coherent anti-Stokes Raman spectroscopy (CARS) [3].

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

33 33 34 35 36 37 38 38

polarised (polarisation, α) to be related to the elasticity of the electron cloud of the molecule. In an analogous fashion to infrared spectroscopy, in which the change in dipole with vibration determines absorption strength, it is the change of polarisability with vibration that determines Raman activity. The scattering processes are described in Fig. 1. These show that scattering may be elastic (Rayleigh scattering) in which the vibrational state of the molecule is unchanged. Of the inelastic scattering processes, Stokes scattering red-shifts the laser photon by the vibrational energy (v = 0 → v = 1) and the anti-Stokes blue-shifts it (v = 1 → v = 0). In almost all normal Raman microscopes the Stokes scattering is that observed. The reason that the anti-Stokes lines are much weaker is that the population of molecules that are in the v = 1 state is lower thus these transitions occur less frequently. The population of the respective states is related to the energy between them (ΔE) by the Boltzmann distribution, Eq. (1). Nv¼1 −ΔE ¼ exp kT Nv¼0

ð1Þ

1.1. Principal of Raman spectroscopy Raman scattering is the inelastic scattering of light from a sample. It was first described in 1928 by Raman and Krishnan [4] from which it gets it name. It occurs because the scattered photons lose or gain energy from the molecules in the irradiated sample. The pattern of Raman scattering (the energies of the scattered photons and the intensities of those transitions) informs on the molecules present in the irradiated volume. The scattering phenomenon is a rare event and the majority of scattered photons are elastically scattered; this is called Rayleigh scattering. Rayleigh scattering has a probability of about 10−5 in 1 m of air. Importantly this process is strongly dependent on wavelength (λ−4) and it is Rayleigh scattering that makes the sky appear blue. The probability of Raman scattering occurring is 1:107 scattering events. So the strength of signal is very small and although an individual Raman scattering event occurs in a femtosecond (10−15 s) [5] to get appreciable signal from a sample (which is many events) can take nanoseconds. This has some very important consequences for imaging in that there exists a fundamental difference between Raman scattering and absorption or emission techniques. The latter have transition probabilities that range from 1:10 (transition event:photon number), for emission, to 1:105, for near-IR. This means that for the high probability methods the signal is likely to originate from the irradiated volume. However for Raman spectroscopy the intrinsic low probability of observing the signal means that the irradiated volume provides some of the signal but not all. Photon diffusion and scattering play a greater role in how the signal is observed. This problem becomes much more serious as attempts are made to collect data from deeper inside the sample [6]. This intrinsic problem is described in more detail in Section 1.5. An additional point to appreciate is that the intensities of Raman transitions vary greatly depending on the analyte material. The mechanism whereby Raman scattering occurs is related to how a compound interacts with the electric field component of the photon. The more easily the compound can be polarised — the more responsive the electrons within it are to the driving electric field of the photons — the more intense the Raman signal will be. We can consider the ability to be

At room temperature, for transitions at about 100 cm−1 NNv¼1 ¼ 0:62, v¼0 thus the anti-Stokes and Stokes lines have comparable intensity; but for 1000 cm−1 this ratio drops to 0.008. Molecules with electrons that are easy to polarise will give stronger Raman scattering than those held tightly. This has an important ramification for the study of pharmaceuticals in formulations and that is that most excipients are σ-bonded molecules like cellulose and starch and most APIs contain π-electrons. This means that APIs typically give stronger Raman signals than the excipients as shown in Fig. 2. This is exemplified by the data shown in Table 1 which shows the Raman cross sections ∂σ [7] for different molecules and as a function of excitation ∂Ω wavelength. These data show that the Raman cross sections vary dramatically with molecule, for example the benzene Raman transition at 992 cm−1 is ten times more intense than the 758 cm− 1 band of CHCl3. Furthermore the Raman cross section varies with λexc. Notably it becomes intense when the laser wavelength is coincident or approaching an electronic absorption energy of the analyte; this is the resonance Raman effect — it can enhance signals by 106 as seen for β-carotene. This can be problematic if colouring agents are added to formulation as by their nature such agents are highly absorbing in the visible region. 1.2. Experimental systems — Raman microscopy Raman microscopy is a term used to describe the fusion of Raman spectroscopy with optical microscopy. It can be performed using many of the various Raman spectroscopic techniques including dispersive Raman, FT-Raman, CARS and stimulated Raman scattering (SRS). Typically, the optical microscope is equipped with an excitation laser and spectrograph. The excitation laser beam is directed onto the sample through the objective lens and Raman scattering is produced. Scattered light is then collected using the same objective and directed to the spectrograph. Various different excitation laser wavelengths can be used depending on the sample and analytical question. Several other features


23

v’ = 2 v’ = 1

Electronic excited v’ = 0 state

ν˜LASER

νãnti-Stokes

ν˜Stokes

Wavenumber / cm-1

Virtual excited states

Ground state v=3 v=2

ν˜ VIB

v=1

IR

NIR

Rayleigh Stokes scattering scattering

Vibrational levels

v=0

Anti- Fluorescence Stokes scattering

Fig. 1. Depiction of the energy level changes in infrared (IR), near-infrared (NIR) absorption, Raman and Rayleigh scattering and fluorescence. For anti-Stokes scattering the laser photon (green) gains energy from the molecule and the scattered photon (blue) is at higher energy. For Stokes scattering the molecule gains energy from the photon and thus the scattered photon (red) is at a lower energy.

of Raman spectrometers may be altered: laser rejection filters, the use of pinhole or slit apertures, the method used for separation of the incoming wavelengths of light (polychromator, monochromator or just a simple diffraction grating), and the type of detector used (CCD or Ge detector in the case of FT-Raman). Fig. 3 shows the general layout of a typical Raman microscope.

1.3. Imaging methods Raman data for chemical mapping may be obtained in one of three ways, these methods use the fact that CCDs are matrix detectors. That is they possess pixels in a two-dimensional array (3 cm long and 0.6 cm high are typical dimensions with pixels, each 25 μm × 25 μm) that may be read out individually or averaged through binning. In the first of these methods the sample is translated to a specific point in coordinate space and a Raman spectrum is collected — the sample is then moved and another spectrum is measured. This is called point-by-point mapping [11], point mapping [12] point-scan [13] or whisker-broom [13] (or whiskbroom [14]). The wavenumber of Raman scattering is dispersed along the long axis of the CCD to allow for spectral discrimination and the vertical columns of pixels are added or binned. So each column of pixels measures a specific Raman shift or wavenumber e /cm−1) (Fig. 4a). (ν A second method selectively bins the CCD at certain set heights. Thus the height of the image entering the spectrograph and striking the CCD is differentiated by the specific binning of sections. The drawback of this is that the signal is weaker because the integration is over a smaller

Table 1 Raman scattering cross sections for a series of molecules in liquid or solution with differing λexc. Note that β-carotene is in pre-resonance at 514.5 nm hence the large β-value. Compound

Fig. 2. Raman spectra taken under identical conditions, λexc = 1064 nm, 85 mW; acquisition time 3 min; solid state samples: (a) corn starch; (b) cellulose; (c) carbamazepine. Spectra are offset for clarity.

Raman shift/cm−1

Benzene

992

CHCl3

758

Naphthalene in benzene Anthracene in benzene β-Carotene in benzene

1382 1402 1520

λexc/nm

∂σ ∂Ω

1030 [7]/cm2 sr−1 molecule−1

Reference

647 514.5 488.0 441.6 407 785.0 514.5 514.5

11 30 37 45 64 0.6 3.2 82 540 1.1 × 107

[8] [9] [9] [9] [10] [5] [5] [5] [5] [5]

24


Fig. 3. Schematic of a CCD-based Raman microscope.

number of pixels for each wavenumber. This is termed line-scanning [11,13] or push-broom [14], and is shown in Fig. 4b. The third method in which a map can be made is to utilise the twodimensional structure of the CCD fully. That is to image the irradiated volume on to the CCD. The shortcoming of this is that the CCD cannot be used for spectral discrimination; that must be accomplished before the irradiated volume is imaged (Fig. 4c). This is done by selecting a spectral slice of the Raman scattering and only allowing that wavenumber range to be imaged. Acousto-optical filters and liquid-crystal tunable filters may be used for this purpose [15–17]. This is termed global imaging [11,12], plane scan [13] or staring [14]. It is not uncommon to differentiate these methods based on whether the sample needs to be translated; if translation is required this is termed mapping and if not then the term imaging is more commonly used. The imaging experiments are rare and we have used the terms interchangeably in this paper. 1.4. Advantages of using Raman imaging There are many advantages to using vibrational spectroscopy, for example Raman spectroscopy, to image samples. Limited sample preparation is the most appealing feature of Raman spectroscopy. For example, confocal laser scanning microscopy (CLSM), often requires protein and fat samples to be dyed prior to imaging, whereas Raman spectroscopy requires no dye. This removes the possibility of a preparation technique altering the samples physical properties in an unknown way. Substances resistant to being stained by dyes, such as carbohydrates, are also problematic with CLSM as it makes obtaining distinct images of the substance difficult. Furthermore, Raman microscopy is non-destructive, so it is not necessary for samples to be dissolved or mixed with any other substances prior to imaging. This allows the same sample to be used for other analytical techniques once imaging has been performed. These advantages are also shared by near-infrared

and infrared imaging; for these techniques the transitions observed are absorptions and are much more intense than Raman scattering and thus can be easier to detect. Near-infrared (NIR) spectroscopy is widely used and NIR chemical imaging (called NIR-CI) is very well established in terms of instrumentation and imaging analysis [18–21]. Infrared microscopy is also a rapidly developing field [22–24]. 1.5. Shortcomings of Raman imaging There are a number of issues that can limit the effectiveness of Raman microscopy, these include: 1. 2. 3. 4. 5. 6.

spatial resolution; sample components having intrinsically low Raman signals; fluorescence; subsampling (although this may be desirable); spherical aberration/refraction; out of focus contributions and relative signal strength.

1.5.1. Spatial resolution Spatial resolution of Raman microscopy is defined by the size of the irradiated volume. This, in theory, may be calculated from the Abbe equation (Eq. (2)) which calculates the diffraction limited spot size. This relates spot size (d) to the irradiating wavelength (λ) and the numerical aperture (focal length) of the collecting optics.1 d¼

λ 2NA

ð2Þ

1 It is important to appreciate that the optical equations are for perfect systems; almost all pharmaceutical materials are opaque and scattering and photon diffusion further compromise the optical dimensions of irradiated volumes by diffusing focal depths and causing the signal from regions lateral to the focal point to become significant [6,25].


25

y / cm-1

x

Whisker-broom

(a) y

x

Push-broom

/ cm-1

(b) y

Staring

x

e ) from each Fig. 4. Examples of the different imaging methods. The x, y coordinates represent two dimensional coordinates of the sample the third dimension is the Raman spectrum (ν point. In whisker-broom (a) a single point is imaged (the blue dot) and the spectrum collected from that point; in push-broom a line of points, in this case the red, mauve, dark blue and blue dots, are imaged simultaneously. As the CCD is a matrix detector different heights within it measure the spectrum of each of the points and, as shown here, four spectra are collected e (shown in green) and this ν e is projected on to the detector with each pixel acting as a collection point. In this case 16 simultaneously; in staring mode a filter is used to select out a band of ν points are measured simultaneously and the intensity of the selected region is determined.

NA ¼ n sinθ:

ð3Þ

From this equation, it can be deduced that, for a specific laser wavelength, the resolution can be improved simply by increasing the numerical aperture (NA). This value can be calculated from Eq. (3), where n is the refractive index of air or immersion oil and θ is the half angle of the objective aperture. Using the Abbe equation, it is possible to calculate the expected resolution limit of a microscope. For example, a WITec confocal Raman microscope claims to have a spatial resolution of 300 nm when using a 532 nm laser wavelength and a 100 × objective with a NA of 0.9:

d¼

λ 2 NA

d¼

532 nm 2 0:9

d ¼ 296 nm: As expected, a theoretical spot size close to 300 nm was calculated. The resolution limit was shown experimentally by imaging 50 nm wide carbon nanotubes as the Raman microscope was unable to effectively resolve an object of that size and estimated the nanotube width to be ~300 nm (using full width at half maximum) [26]. This illustrates the importance of the diffraction limited spot size as it has a direct influence over the attainable resolution of a Raman microscope.

1.5.2. Lack of Raman signal and fluorescence Perhaps the most obvious problem one might face when using Raman spectroscopy is when a sample does not produce Raman scattering. This is a consequence of polarisability that has already been discussed in Section 1. Other than the lack of Raman scattering, fluorescence is possibly the most common issue encountered when using Raman microscopy. This occurs when the energy of the incident photon of light has enough energy to excite the molecule to a higher electronic energy state rather than a virtual one. This gives the molecule the opportunity to relax to lower vibrational energy levels within the electronic excited state before emitting a photon and relaxing down to the ground electronic state. When fluorescence occurs, it generally produces much more signal than the Raman effect and results in Raman signal being obscured. In many cases fluorescence can be avoided by changing the excitation laser wavelength (either by increasing or reducing it), fluorescence quenching, photo-bleaching or pre-processing (Section 2.1) [27]. Arguably, increasing the excitation laser wavelength is the simplest method, and it works by preventing excitation to the higher electronic energy state, therefore allowing Raman scattering without the intrusion of fluorescence. 1.5.3. Subsampling Subsampling is another common problem when imaging using Raman microscopy as the field of view is relatively small and may not be representative of the entire sample. This problem has been noted previously in the context of pharmaceuticals [28,27,29]. Johansson et al. [29] assessed the effect of subsampling within pharmaceuticals.

26


Four sampling methods were used to test the subsampling error when quantifying the active pharmaceutical ingredient (API) amongst several excipients. These methods include: point irradiation (total area of 0.4% of the tablet was irradiated), small circle irradiation (8% of the tablet), large circle irradiation (16% of the tablet) and area irradiation (39% of the tablet). Spectra were collected from two sides of the tablet. The ratio of the 1612 cm−1 band with the average intensity of 250–1500 cm−1 was calculated for each spectrum. Subsampling error was measured as the difference between the spectra collected from the two sides (i.e. the difference in the ratios). The average sampling error was calculated from 20 tablets for each of the irradiation methods and gave errors of: 4.0 for point irradiation, 2.1 for the small circle, 1.4 for the large circle and 1.4 for the irradiated area. This highlights the importance of considering the effects of subsampling within pharmaceuticals. Of course sampling small domain sizes may be desirable to understand the microstructure of tablets and the dissolution of active ingredients from them [30] but an appreciation of domain size and spot size are important for establishing meaningful data. 1.5.4. Spherical aberration/refraction Optical issues pertaining to Raman microscopy have been investigated and reported on extensively by Everall [31,6,32–34]. This is perhaps an area that has received less attention than other limitations, such as fluorescence; it is a serious problem and is intrinsic to the Raman process because Raman scattering is a low probability event. Spherical aberration causes limitations to depth resolution in Raman microscopy because light in the centre of the objective lens is refracted significantly less when it passes through the sample than the light that arrives at the sample from the edge of the lens (illustrated in Fig. 5). This causes the peripheral light to be bent and focus deeper within the sample. As a result, Raman scattering then occurs deeper within the sample than intended. This becomes a bigger problem when one tries to image the objects deep within the sample, as the deeper you focus the worse the depth resolution becomes. Although experimental results show that some of this light will be excluded by the confocal aperture, it comes at the expense of signal strength [35]. This inability to recover all of the depth resolution causes objects to appear closer to the sample surface because, even near the surface, the spectrum will show the presence of the deeper objects. This has been elegantly illustrated in a study of film depth of polyethylene terephthalate (PET) underneath a 40 μm

acrylate coating via Raman depth profiling where the results indicated that the PET film depth was ~20 μm — half the true depth [33]. The effect of refraction can be remedied, however, through the use of immersion objectives and immersion oil with a refractive index close to that of the sample. Since the refractive index of the oil is much closer to that of the sample than air, light will bend less when entering the sample and should result in more accurate depth profiling. In the previously described paper, use of such an immersion objective measured the depth of the same PET sample to be ~37 μm — more accurate than the use of the non-immersion objective [33]. The potential shortcomings of using immersion oils have been noted [6]. Firstly, for samples such as pharmaceuticals, it can be difficult to remedy this issue as it is essential for the refractive index of the immersion oil to be close to that of the sample, which may not always be known. Secondly, pharmaceuticals are typically heterogeneous and will have changing refractive indexes between crystalline API and excipient, thus negating the effect of an improved refractive index at the sample surface. Thirdly, the oil may produce intrusive Raman bands. Finally, the oil may adversely interact with the sample. However, an ingenious solution to this final point has been described by Tomba and Pastor [36]. They forced a thin polyethylene film to conform to the sample surface using a vacuum before placing the oil on top. This prevented the undesirable contact between oil and sample. 1.5.5. Out of focus contributions and relative signal strength Despite the use of immersion oil and the appropriate objectives, the depth resolution of the Raman microscope may still not meet theoretical expectations. This can be attributed, in part, to out of focus contributions. Fig. 6 shows that light from the objective is focused at point P. The majority of Raman scattering will occur at this point (approximately 90%); [32] however, an amount of light passes though point P and irradiates the area below in a process often referred to as photon diffusion. This causes Raman scattering to occur below the focal point and, despite the confocal aperture stopping the majority of this scattered light, not all of it will be blocked. Light which passes back through point P will reach the detector (Fig. 7). This can be a significant problem when trying to analyse thin layers on top of a bulk substrate especially when the substrate is a better Raman scatterer [31,32]. This is because a considerably large portion of the substrate is irradiated, as shown in Fig. 8a. Therefore, if the substrate produces much stronger Raman signal than the

Fig. 5. A simplified illustration of the influence of refraction on depth resolution (DR). As the objective focusses deeper into the sample, the depth resolution becomes significantly worse. The image was adapted from pictures provided in references [31] and [6].


Fig. 6. Raman scattering from deeper regions within the sample will be recorded by the detector if it passes back through the focal point, P [31].

27

coating, the spectral features of the coating will be overwhelmed by that of the substrate. The problem is exacerbated if the focal point is set slightly higher than the surface layer. Fig. 8b shows that, in this case, the majority of signal is obtained from point P (which is likely to be set in the air or immersion oil) and the second highest amount of Raman scattering is likely to occur within the substrate (simply due to the relative size of the area irradiated). McAnally and Everall [32] demonstrated this in an experiment where a 15 μm thick layer of polyethylene (PE) coated a 100 μm thick layer of polyethylene terephthalate (PET). When the laser was focused at the surface or focused 10 μm into the sample, the spectrum was dominated by PE features. When focused 20 μm into the sample, PET features dominated the spectrum. However, interestingly, when the laser was focused 5 μm above the sample surface, the spectrum was dominated by PET, the bulk substrate. The PE coating was almost invisible in the spectrum which suggests, when analysing transparent samples, the substrate is likely to produce spectral features even when out of focus. Despite these experiments involving the use of transparent samples, they represent somewhat the ‘best case scenario’ as prediction of the behaviour of light in opaque and/or heterogeneous samples becomes much more difficult. It is for this reason that considering the relative signal strength of sample components and their relative distances from the focal point becomes extremely important. Out of focus contributions do not only originate from regions above or below the focal point, but also from surrounding regions in the XY plane. Whilst investigating the UV-cure of PE coated PET substrate via lateral profiling (the PET contribution was found to be too intense for depth profiling), an interesting phenomenon occurred when the laser focus moved off the edge of the sample at the acrylate/air interface [31]. Once again, the PET signal then became the dominant spectral feature when the focus was 2 μm away from the sample. This is thought to be caused by refraction which directs marginal rays into sample. Fig. 8d shows how spectral features of PET might be obtained in this way. Pelletier demonstrated that reducing these out of focus contributions is possible with minor adjustments to the Raman instrument [37]. This method is referred to as mapping enhanced by structured pupils (MESP). Fig. 9 shows how the use of apertures can effectively reduce

Fig. 7. Simplified illustration of how a confocal aperture removes unwanted signal from outside the focal point of a Raman microscope.

28


Fig. 8. These images show the regions which contribute the largest amount of recorded Raman signal. (a) The PE layer (light green) is the dominant contributor of Raman signal, (b) PET (dark green) produces the most Raman signal, (c) lateral scanning with PE producing the majority of the Raman scattering, (d) a theorised explanation of why PET contributes the most Raman signal when the microscope focuses beyond the edge of the sample during lateral scanning. Figures were adapted from images presented by Everall [31].

the signal obtained from out of focus contributions from below the focal point. Aperture 1 blocks the path of light which would illuminate region B, and aperture 2 blocks the scattered light coming from region A. Light which reaches point P is still able to produce scattering which passes aperture 2. This method does, however, significantly reduces the signal output but with the benefit of greater depth resolution. In turbid materials the shortcomings of imaging are further compromised. In a detailed study comparing Raman intensities from simple layered structures [25] it has been shown that it is possible to model reflected and transmitted Raman signals. In composite layers in which the concentrations of components vary with penetration depth the transmitted Raman signal originates from the centre with the reflected originating from the surface. Importantly the probability of Raman scattering is modelled and reveals that in reflective mode the most probable Raman scattering is from below the surface and the intensity rapidly falls off with penetration depth. In summary the low probability of Raman scattering creates uncertainty in establishing the location from which scattering is observed and thus uncertainty in determining analyte location. Although such data may be modelled, this requires considerable knowledge about the internal structure to derive effective

imaging solutions. Despite these difficulties Raman imaging is widely used and with developments in new techniques, such as spatially offset Raman and transmission Raman spectroscopy [38–44] it is beginning to be used in tomography applications [45,11]. 2. Spectral analysis It is clear that there exist a number of challenges in the collection of reliable information, in terms of spatial coordinates, using Raman spectroscopy. However the technique is used to construct data in which spectral information is attributed to a particular x, y, z coordinate volume. Such a data set is referred to as a hyperspectral cube [13]. Bearing in mind the challenges associated with correctly defining x, y, z coordinates using Raman spectroscopy the question arises as to how these data are processed and analysed to be meaningful. The hyperspectral cube may be considered as four-dimensional in that it is made up of sampled volumes, each with specific spatial x, y, z coordinates; each of these volumes, which we will call elements, contain spectral data [20]. The spectrum of each element is, to a first approximation, made up of the spectra of the pure components (ST) of the samples weighted by the concentration of each component in the element (C),


29

components of interest are identified (through PCA or other methods). The third is the “quantification” of components; this may be achieved by determining the quantities of components in each volume of interest or by classifying each volume as being due to one of the components. 2.1. Pre-processing

Fig. 9. A generalised illustration of Pelletier's method for improving depth resolution. The picture was adapted from the simplified image [31].

and the residual noise (E). The cube may be unfolded into a matrix of element labels and wavenumber (D) (Fig. 10) and is given by Eq. (4). T

D¼CS þE

ð4Þ

There are a number of aspects to the interpretation of such data sets. The first is pre-processing in which the raw hyperspectral cube is corrected and the spectral data therein normalised. The second is the exploration of the data set in which spectral signatures of the

Pre-processing of the Raman spectra to remove differences associated with the sampling arrangements and/or sample emission is an important step, particularly with multivariate data analysis techniques. Any multivariate analysis done on spectral data sets without adequate pre-processing may give models which contain dominant differences or separating features associated with the experimental setup whereby masking the useful chemical and/or physical information. For example, if a sample set has not been normalised, the first PC from PCA analysis may contain information on relative sample spectrum intensity, which is due to the sample focus rather than any chemical differences. The method of pre-processing of any Raman spectra is an important consideration for imaging, interpretation and quantification of the data. There are three main types of spectral artefacts in Raman spectra which require different methods of pre-processing. The major artefacts are baseline, noise and scale differences, which arise from a range of different experimental conditions and are described under each correction method category below. As this is only a very brief overview of pre-processing, there are many excellent articles and books outlining the importance of preprocessing and describing the various common techniques [18,46–49]. 2.1.1. Cosmic events and de-noising The CCDs used in Raman microscopy instruments are capable of detecting high energy subatomic particles, indeed they are used to detect such species in particle accelerators. As such particles irradiate the earth CCDs detect them and they manifest as intense signals on one or two pixels of the CCD. These are readily removed by most commercial software. As they can be 103–106 times more intense than the Raman scattering their removal is essential prior to further processing [50]. In addition it is possible for the CCD to contain defects, such as dead pixels; these should also be removed prior to further processing.

Fig. 10. A depiction of a three-dimensional coordinate data set. This is unfolded into matrix D which contains all of the spectral information collected. These data are related to the concentrations of the components C, in this case i–v, and the spectra of the pure components ST, and residual noise E.

30


The rapidity with which large-scale maps are collected can lead to noisy data; [20,51] it is common to smooth spectral data and algorithms [52,47] to accomplish this include Savitsky–Golay [53], Fourier and wavelet-based methods [54]. 2.1.2. Baseline correction and normalising Baselines in Raman spectra may contain useful information, such as the boson peak observed in amorphous drugs at low frequencies [55], or it may be unrelated to the chemical composition of the sample. These baseline effects can obscure useful [20] data and can be removed using a number of methods. These include linear [47] and rubberband [56] corrections. The use of derivatives where the ∂

ðIntensityÞ

or higher ν ∂e derivatives are evaluated can greatly improve background issues because e and are derivatised to zero. It is the backgrounds respond slowly with ν also possible to avoid baseline issues using spectroscopic means [27]; the most common are time-resolving the Raman spectra; that is data is collected with a pulsed laser in which the detection is synchronised with the laser pulse and the slower emission signal is time-gated out [57,58] or by spectral shifting in which the laser is tuned ±10 cm−1 e LASER and that at ν e LASER ± 10 cm−1 are and the spectra generated at ν subtracted [10,59,60]. This results in a difference spectrum in which the emission background which is unchanged by such a minor shift in excitation wavelength is subtracted out. These methods are yet to be exploited in Raman imaging of biomedical devices. In addition to the variation in baselines differences in the focus, surface density and texture may conspire to vary spectral intensities across the sample. This variation is unrelated to the constituents of the sample and correcting for it may be accomplished using scattering correction algorithms. Two widely used methods are standard normal variate (SNV) and multiplicative scatter correction (MSC) [52,61,62]. In SNV a data set with variable intensity and variance of each spectrum. This is accomplished through the use of Eq. (5): xSNV i; j ¼

xi; j −xi Si

ð5Þ

e j ; xi,j is where: xSNVi,j is the corrected Raman intensity for spectrum i at ν e the uncorrected Raman intensity at ν j ; xi is the mean Raman intensity for spectrum i; and Si is the standard deviation of Raman intensities for spectrum i. MSC uses a reference spectrum, often the mean spectrum returned by the data set. For each spectrum it then fits a regression line that adjusts the offset of the treated spectrum (i) and scales it using the regression coefficients ai and bi respectively (Eq. (6)): xi; j;MSC ¼

xi; j −ai bi

ð6Þ

where: xi,j,MSC is the corrected Raman intensity for spectrum i at νj cm−1; ai is the intercept derived form the linear regression of spectrum i versus the reference spectrum; xi,j is the uncorrected Raman intensity e j ; and bi is the slope from the linear regression of the for spectrum i at ν spectrum i relative to the reference spectrum. The use of both baseline correction and normalising methods is nicely demonstrated in a comparative study which examined the performance of a range of these techniques, including: SNV; MSC; first, second and third derivatives, in quantifying furosemide crystal polymorphs in powdered mixtures. This study used the parameters of standard error of prediction, covariance of the reference and standard value and the ratio performance deviation (RPD) to assess the best performing experiment and processing [49]. The RPD is very effective at “scoring” calibration models; it is generally accepted that RPD N 2.5 is acceptable but values N 5 are required for quality control and N10 are considered excellent [63]. The study of these polymorphs with Raman spectroscopy revealed that form I was detected best with MSC

correction alone, but form II used normalisation + second derivative and form III normalisation + first derivative. The RPDs in all cases were N9. It is clear from this work that some exploration of which method works best for the problem of interest is warranted to achieve optimal results. 2.2. Data exploration and quantification The methods whereby images are created from hyperspectral data sets range from simple univariate analysis to multivariate quantification, in which component concentrations are given for each element in the spatial cube (Fig. 10), and classification or segmentation in which elements are labelled as containing one species or another. 2.2.1. Univariate methods Univariate methods, such as single band intensity, integral or bandwidth are commonly used for Raman spectroscopic maps. This is a straightforward method to visually determine the spatial distribution of a single component. A major disadvantage is that most of the data is not used. There are some important considerations to consider when using this method. The first is the presence of a band (preferably strong) representing the chemical of interest in a spectral region which is not masked by other component spectra. The second is the need to ensure that the data is pre-processed (normalised) appropriately, so that differences in intensity associated with sample focus or surface texture are not contributing to the intensity of a single component, otherwise the false colour map may have a focus component to the image. Many studies use the intensity of a single band to map the distribution of a single chemical constituent [64–67]. A number of studies have used peak integrals to map the spatial distribution of a single constituent [64,68]. Balss and co-workers used integrals of bands were used to show the distribution of the polymer and API in drug eluting stents [64]. The bandwidth and/or band position can be used to look at how a single constituent is changing across a structure [64,69]. An important consideration in drug development is the polymorph or crystallinity of the drug of interest in a given formulation. The bandwidth of components can be indicative of the order/disorder in the structure, with amorphous compounds having broader bands than their crystalline counterparts. For example Vervaeck et al. used the bandwidth of the peak lying between 810 and 830 cm−1 to map the crystallinity of metoprolol tartrate in a prilled formulation, both immediately after processing and after storage [69]. Bivariate analysis involves using two data points, such as two different component band intensities or integrals, in a ratio to create the false coloured images. This method has the advantage of gaining insight into how the relative amounts of the two components are changing across a sample, meaning changes in focus are accounted for by the relative intensities (or integral area) of the two bands acting as an internal normalisation method. Multiple groups have used a bivariate method for mapping [64,70–72]. Balss and coworkers used band ratios in combination with an HPLC assays to quantify the concentration of the API sirolimus in drug eluting stents [64]. 2.3. Multivariate methods Multivariate methods are advantageous in that all the spectral information is utilised, the image is not based on one single component value. A wide variety of different multivariate analysis methods were used across the different articles reviewed, a selection of methods will be briefly described here. Multivariate analysis was used for creating both qualitative [73] and quantitative [74] Raman images. 2.3.1. Principal component analysis (PCA) PCA is a useful tool to reduce dimensionality in large datasets, allowing for human interpretation on datasets with hundreds of


31

into the original components and their associated concentrations. The general equation that MCR follows is

dimensions or variables, such as spectra, which may be simplified down to 2 or 3 dimensions called PCs [47]. PCA effectively decomposes the initial dataset such that

T

T

X¼TP þE

X ¼ YS þ E

ð7Þ

where X is the initial data matrix such as a series of spectra, Y is the concentration of the sources, ST is the shape of the sources (i.e. pure component spectra) and E is the error matrix [80,61]. Solving Eq. (9) can lead to multiple pairs for Y and ST as a result of rotational and intensity ambiguities. Rotational ambiguity can be described as multiple possibilities for the shape of the pure spectra and the intensity ambiguity is described as the differing magnitude of the sources. The transformation of Eq. (9) gives insight into the two sources of ambiguities:

where X is the initial dataset (matrix of i samples by k parameters), T is the scores matrix, PT is the loadings matrix and E is the residuals matrix, as illustrated in Fig. 11. The scores are the reduced dimensionality values for each given sample, i.e. the values assigned to each sample for a given PC. The loadings are effectively what variance each PC is describing and can be used to interpret the variances described across a given PC. The x-residuals are the remaining components not described by the PCs, often mainly noise, but can sometimes contain useful information. Sasic nicely demonstrates the used of PCA to map the spatial distribution of the API and excipients in a pharmaceutical tablet manufactured in two different ways. The excipient magnesium stearate was only able to be detected using PCA and not via univariate methods due to the low signal levels [73].

T −1 T −1 T 0 0T X ¼ YS ¼ Y TT S ¼ ðYTÞ T S ¼ Y S

2.3.3. Partial least squares (PLS) PLSR is a regression model which is set up using two data matrices. A model is created to find the coefficients in the first matrix X, for example Raman spectra, that best predicts the second matrix Y, for example concentration of a specific component. These model coefficients can then be applied to an unknown X dataset to predict a Y value [62]. PLS can be used for both quantification and classification by using the regression and discriminant analysis variations, respectively. PLS is an inverse method used for relating measurements and concentrations, inverse methods follow the same general equation relating a specific property such as concentrations (y) to the measurement matrix (X) and model coefficients (b) [47]:

2.3.5. Band target entropy minimization (BTEM) BTEM is a self-modelling curve resolution multivariate technique, no reference spectra or prior information is required to find the pure components [82–84]. The output of information consists of pure component spectra and the relative amounts of the pure components. Widjaja and co-workers have used BTEM extensively [83,85,82], some particularly interesting applications include the detection of trace amounts of crystallinity in amorphous systems [85] and identification of minor constituents in pharmaceutical tablets [82].

ð8Þ

where Y is the vector or matrix containing concentration values to be modelled, T is the scores calculated via PLS, B is a coefficient and F is the y-residuals computed from the PLS model [52].

2.3.6. Classification and segmentation The aforementioned PLS methods provide quantification for components in each element volume (Fig. 10), but it may also be useful to attribute an element to a particular component through classification or segmentation [20]. This may be achieved using K-means clustering of PCA data [86,87]. This simple method may be extended by using fuzzy

1

2.3.4. Multiple curve resolution/multivariate curve resolution (MCR) MCR is a method used to find linear combinations of different components in a system. MCR effectively decomposes the initial data

-1

PC2 0

PC1

PC2

-1

0 PC1

ð10Þ

where Y′ = YT and S′T = T−1ST describe the X matrix as the Y and ST matrices do, however the Y′ and S′T are not the sought solutions [80,61]. To remove ambiguity in the resulting Y and ST pairs, constraints may be placed on the problem. Common constraints which can be selected include non-negativity, unimodality and closure. Non-negativity constraint forces the values in the source concentrations (Y) and/or the source profiles (ST) to be greater than or equal to 0. Unimodality constraint only allows one maximum per sample, this is used for monotonic profiles where only one component is increasing or decaying. Closure constraint is used for closed reaction systems where the overall balance of the components remains constant for all samples, it forces the sum of the components to equal a constant value for all samples. Constraints should be selected based on the nature of the sample set being studied [80,61]. Vajna and coworkers give a nice example of MCR to map the major components in the extruded and sustained-release Isoptin formulations [81].

2.3.2. Classical least squares (CLS) CLS assumes that the system has a linear relationship between the measurement and concentration. With Raman spectroscopy there is a linear relationship between band intensity and concentration of the associated component giving that signal. CLS requires that the dataset is well controlled, with known pure components. Many studies use CLS to quantify components of the spectroscopic images [75–79].

y ¼ Xb

ð9Þ

1

X

=

T

·

PT

+

E

x-values matrix

=

Scores

·

Loadings

+

x-residuals

Fig. 11. Graphical representation of how PCA separates the initial series of spectra into scores, loadings and residual components.

32


clustering in which more than one component maybe classified in an element [87]. PLS-DA is a variation of PLS which can be used to discriminate between two classes. Instead of a range of values for the Y vector/matrix, each sample is given a value based on belonging (1) or not belonging (0) to the class being discriminated. The resulting output from PLS-DA gives a value for each sample and its associated error (deviation). 3. Case studies Raman microscopy may be non-destructive, label-free, and chemically selective, multiple components may be analysed from one measurement, and in a confocal arrangement Raman microscopy may be used to depth profile samples. These characteristics have led to the application of Raman microscopy to a wide range of drug delivery problems, for example, Raman microscopy mapping has been used to characterise constituent distribution in solid dosage forms [75,70,71, 76,78,88,81,89,90], identify API metabolites within cells [91] and better characterise API dissolution from solid formulations [92,30,93–95]. 3.1. Solid dispersions — including tablets The simplicity and convenience of oral drug administration is far removed from the often complex research invested into API release rates for new formulations. Various methods of solid dispersion manufacture (e.g. closed melting method [96], hot melt extrusion [71,81,68, 97] and electrospinning [75,76,97]) have been used to produce dosage forms with desirable drug release characteristics. Understanding the mechanisms of drug dissolution requires knowledge of constituent distribution and the API solid state within these dosage forms. Raman mapping has been applied extensively to solid dispersion for distribution and API solid state characterisation (Table 2). The solid states of troglitazone in troglitazone-polyvinylpyrrolidone solid dispersions have been identified using Raman mapping [96]. Troglitazone has two crystalline forms (RR/SS and RS/SR) which are diastereomer pairs, and each crystalline form has a unique Raman peak: 1730 cm−1 for RR/SS and 1685 cm−1 for RS/SR. The intensities of 1730 cm−1 and 1685 cm−1 were used to illustrate the distribution of each crystalline form and the full width at half maximum of the 1747 cm−1 peak was used to distinguish crystalline from amorphous regions. Raman spectra of amorphous compounds tend to have broad peaks relative to crystalline counterparts as a consequence of the less structured vibrational environments within the sample. Amorphous

troglitazone was observed in a solid dispersion manufactured at a heating temperature of 105 °C despite this temperature being lower than the melting point of both crystalline forms of troglitazone. These solid dispersions were formed using the closed melting method and during this sealed heating it was proposed that PVP was present in a rubbery state. Amorphous troglitazone may have formed when crystalline API dissolved in rubbery PVP. The Raman mapping data supported this theory as the distribution of the amorphous API and PVP substantially overlapped. Furthermore, there was less crystalline RR/SS troglitazone in this solid dispersion: the RR/SS crystalline form has a lower melting point would have dissolved more readily. As expected, solid dispersions manufactured using high water contents and temperatures had larger amorphous regions. Raman mapping recently demonstrated that heterogeneity and API crystallinity within an extruded solid dispersion influences API release rate across a range of conditions [81]. Two formulations of verapamil hydrochloride were compared: 1) Isoptin SR, a conventional sustained release formulation manufactured by wet granulation; and 2) Isoptin SR-E, a melt extruded formulation. Isoptin SR-E was previously shown to maintain the same dissolution characteristics in media with up to 40% ethanol, however, isoptin SR was incapable of maintaining sustained release characteristics under these conditions. The difference between release characteristics was proposed to be a consequence of composition, however, a Raman mapping comparison indicated that both formulations contained the same components. Three pure constituent spectra were predicted for isoptin SR-E and isoptin SR using MCRALS. For both isoptin SR-E and isoptin SR, the three predicted spectra were similar to the pure constituent spectra of the conventional formulation (verapamil HCl, Na-alginate and microcrystalline cellulose) indicating that both formulations contained the same constituents. The MCR-ALS spectra for the isoptin SR-E formulation were not as accurate as those for isoptin SR, as the predicted pure spectra of Na-alginate contained verapamil HCl peaks. This was due to a rotational ambiguity within the spectral dataset, and this implied that the distribution of these two components was correlated. The problem of rotational ambiguity was diminished by applying a contrast constraint to the MCR-ALS. This constraint forced iterative calculations to converge on spectra that were as different from each other as possible. The resulting estimated pure spectrum for Na-alginate did not contain any API peaks. The distribution associated with each estimated pure spectrum was assessed using the respective estimated concentrations, and homogeneity was determined by calculating the relative standard deviation of concentrations for each pixel in the area sampled. The isoptin SR-E formulation

Table 2 Solid dispersion characterisation studies which use Raman microscopy. Author/s

Year

Drug and investigation

Experiment type

Analysis method

Reference

Breitenbach et al.

1999

Mapping (lateral)

2007

Qualitative, bivariate (peak intensity ratio) Qualitative, bivariate (peak intensity ratio)

[71]

Karavas et al.

Vajna et al.

2011

2011

Almeida et al.

2012

Mapping (lateral)

Nagy et al.

2012 2014

Qualitative, multivariate (classical least squares) Qualitative, multivariate (classical least squares)

[76]

Balogh et al.

Used Raman mapping to assess size of polyethylene oxide crystal domains in a hot melt extrudates containing metoprolol tartrate, polyethylene oxide and ethylene vinyl acetate with varying levels of vinyl acetate. Compared spironolactone solid state and distribution in electrospun and extrudate formulations of Soluplus®. Raman mapping of melt electrospun fibres of carvediol and Eudragit® E PO containing one of three plasticisers (triacetin, tween 80 or polyethylene glycol 1500) was used to determine if carvediol recrystallize after a month of storage.

Qualitative, multivariate (multivariate curve resolution alternating least squares) Qualitative, multivariate (band target entropy minimization and target transformation factor analysis) Qualitative, univariate (peak area)

[81]

Widjaja et al.

The distribution and amorphous state of ibuprofen in a polyvinylpyrrolidone matrix was assessed. Raman mapping was used to determine if felodipine was molecularly dispersed or present as an amorphous nanodispersion in a polyvinylpyrrolidone matrix. Assessed the compositional and constituent distribution differences between wet granulated and extruded verapamil formulations. The presence of crystalline API was determined in hot melt extrudates of griseofulvin and fenofibrate.

Mapping (lateral)

Mapping (lateral)

Mapping (lateral)

Mapping (lateral) Mapping (lateral)

[70]

[85]

[68]

[75]


was less heterogeneous than isoptin SR. In addition, an orientation effect was observed in the heterogeneities of the extruded formulation and was a consequence of the extrusion direction. Higher resolution Raman maps were investigated to determine if the API was present as a solid solution with Na-alginate. The extrusion process lead to partial incorporation of amorphous verapamil HCl into a solid solution; some API was present in micron-sized crystals. The dissolution behaviour of isoptin SR-E was a consequence of 1) the solid solution of the API-polymer, and 2) protection of small regions of API crystals provided by Na-alginate. These two factors prevented an increased dissolution rate even in the presence of high ethanol concentrations. The introduction of electrospinning has recently augmented the melt extrusion manufacture process. Electrospinning produces fibres that primarily contain amorphous API; the high surface area to volume ratio of the fibres allow for rapid cooling, thereby reducing the opportunity for API recrystallisation [75,76]. Electrospinning follows melt extrusion and the need for solvent is eliminated, allowing for a simpler, safer, more economically viable method of manufacture. Two methods for producing solid dispersions were compared for the poorly water soluble API spironolactone, and the polymer matrix Soluplus®: melt extrusion, and melt extrusion followed by electrospinning [76]. Constituent distribution was determined from Raman mapping data using CLS based on reference spectra of pure components. The physical mixture of spironolactone and Soluplus® was heterogeneous, and the electrospun fibres were less heterogeneous than the melt extrudate. When the manufacture process for a formulation with 10 or 20% API involved electrospinning there were no spironolactone crystals in the final product; removing the electrospinning stage resulted in a final product with spironolactone crystals. Raman mapping evidence for the presence of crystals was mostly supported by DSC and XRD data, although both DSC and XRD analyses were not able to detect crystalline components in the 10% API extrudate samples. The detection limit of the Raman mapping method for spironolactone crystallinity was more sensitive than the DSC and XRD methods. 3.2. Drug-eluting coatings Drug-eluting coatings are applied to biomedical implants for sustained release of APIs to specific regions of the body. For example, coronary stents are often covered with polymer-coatings containing sirolimus, paclitaxel or zotarolimus to provide sustained local drug delivery for the prevention of restenosis [98]. Raman microscopy has been used to characterise the primary coating material (e.g. polymer [98] or diamond like carbon [99]), API distribution [64,100] or polymer thickness [101] (Table 3) in drug-eluting coatings. The efficacy of restenosis prevention for drug-eluting stents is determined by the stent platform, drug and polymer formulation, and drug release profile. Raman depth profiling has been used to study the distribution of the API and two matrix polymers in stent coatings, sirolimus, poly(ethylene-co-vinyl acetate) and poly(n-butyl methacrylate) respectively [64,100]. The depth profiling data was used to construct

33

quantitative models for constituent concentration. Multivariate PLS calibrations for the API and matrix polymer constituents were more accurate than those based on peak intensity ratios, particularly for the two polymers. Partial least squares models were calculated on the basis of averaged spectra from individually mapped areas [100]. Concentration predictions were made for each image pixel and showed that API rich regions were close to the coating–air interface, and poly(ethylene-co-vinyl acetate)-rich regions were close to the parylene-C layer (which coated the stainless steel stent). The decreased signal to noise ratio of pixel spectra relative to image averaged spectra resulted in a worse mass balance of the three components (ca. 90% for each pixel spectrum versus ca. 100% when using image averaged spectra). 3.3. Polymeric microparticles The popularity of polymeric microparticles as a method of controlled drug release has steadily increased. Raman mapping has been used to characterise both the distribution of components within microparticles, and the solid state nature of the microparticle API (Table 4). Maps generated with confocal Raman microscopy and time of flight secondary ion mass spectrometry have provided a three-dimensional understanding of protein and polymer organisation within protein– polymer micro-spheres [102]. Poly(lactic-co-glycolide) (PLGA) microspheres with a model protein (lysozyme) were produced using a double emulsion solvent process. ToF-SIMS mapping of sectioned microspheres indicated that pores were 2–18 μm in diameter and that lysozyme was present within some pores. Confocal Raman mapping data was recorded across a 40 × 40 μm area from between 20 and 30 μm beneath the sample surface in intervals of 1 μm. Pores within the polymer matrix varied in shape and ranged from 2 to 16 μm in diameter. Smaller pores were filled with protein, and large pores were only lined with protein. Confocal Raman mapping data demonstrated the presence of protein in large pores without sample destruction. In this study, the mass spectrometry technique could not determine if protein in large pores was covering the pore surface or distributed throughout the pore. In addition to describing spatial distributions, Raman mapping can be used to identify contaminants. Band target entropy minimisation (BTEM) is a method of analysis that does not require a priori spectral information, and when it is applied to spectroscopic data, BTEM can be used to establish if unexpected compounds are present within a sample. Four contaminants were identified in ternary-phase double-walled microparticles using BTEM [103]. A total of seven component pure spectra were predicted, three of which were the expected constituents of the microparticle: poly(D,L-lactide-co-glycolide) (PLGA), poly(L-lactide) (PLLA), and poly(ε-caprolactone) (PCL). The remaining predicted pure component spectra were consistent with polyglycolic acid (PGA), a breakdown product of PLGA, dichloromethane, a solvent used in the preparation of the microparticles, and two contaminants introduced during sample preparation before recording Raman data: copperphthalocyanine and calcite. The samples were cut with a razor blade on blue paper before Raman data collection, and during this process

Table 3 Studies of drug-eluting coatings which used Raman mapping. Author/s

Year


Experiment type

Analysis method

Reference

Balss et al.

2008

Mapping (axial)

Qualitative, univariate (peak area) Quantitative, bivariate (peak area ratio)

[64]

Balss et al.

2008

Mapping (axial)

Qualitative, univariate (peak area) Quantitative, multivariate (partial least squares)

[100]

Dong et al.

2009

Mapping (lateral and axial)

Qualitative, univariate (peak area)

[90]

Biggs et al.

2012

Assessed the axial distribution and predicted the concentration of sirolimus, poly(ethylene-co-vinyl acetate) and poly(n-butyl methacrylate) in stent coatings. Used partial least squares to quantitatively image the axial distribution of sirolimus, poly(ethylene-co-vinyl acetate) and poly(n-butyl methacrylate) in stent coatings. Imaged the distribution of sirolimus and arborescent polyisobutylene-block-polystyrene in films used to coat stents before and after a period of dissolution. Characterised the distribution of sirolimus, poly(ethylene-co-vinyl acetate) and poly(n-butyl methacrylate) in stent coatings before and after dissolution.

Mapping (lateral and axial)

Qualitative, multivariate (linear combinations of pure component spectra)

[92]

34


Table 4 Raman mapping of polymeric microparticles. Author/s

Year

Doub et al.

2007 Raman microscopy was used to characterise the size of ingredient specific components of aqueous nasal spray formulations consisting of beclomethasone dipropionate, microcrystalline cellulose, carboxymethylcellulose sodium (CMC), dextrose, benzalkonium chloride, polysorbate 80 and phenylethyl alcohol. 2010 Component distribution (poly(D,L-lactide-co-glycolide), poly(L-lactide) and poly(ε-caprolactone)) in double-layered microparticles was examined. In addition, four contaminants in the microparticles were identified. 2011 The axial distribution of hydrocortisone in a Eudragit L100 microparticle was assessed.

Widjaja et al.

Rizi et al.


Experiment Analysis method type

Reference

Mapping

Univariate (band intensity)

[104]

Mapping (lateral)

[103] Qualitative, multivariate (band target entropy minimization) [67] Qualitative, univariate (peak intensity) [105] Qualitative, visual comparison of spectra

Mapping (axial)

Sievens-Figueroa 2012 Raman spectra were examined to determine if API was amorphous or crystalline in drug nanoparticles et al. distributed through hydroxypropyl methyl cellulose films. Near infrared spectroscopy was used to image the film.

some copper-phthalocyanine and calcite from the paper was transferred to the microparticle. Whilst the presence of copper-phthalocyanine and calcite in the microparticles could have been avoided with a change in sample preparation prior to Raman data collection, the presence of dichloromethane and PGA was a consequence of the protocol used to synthesise the microparticles. These results suggested that the protocol for producing microparticles required adjustment to prevent the degradation of PLGA and incomplete evaporation of dichloromethane. 3.4. Other solid dosage forms The benefits of sustained drug release have propelled the evolution of solid dosage forms well beyond simple physical mixtures pressed

Point mapping

into tablets, with examples of more sophisticated solid dosage forms now including nanocrystalline dispersions [89], lipid matrix extrudates [30,93] and multi-layered tablets [95]. The efficacy of an advanced solid dosage form is determined by the carefully engineered drug release rate, and these release rates are in turn dependent on the physical state of the drug. Over time, an unstable, amorphous drug may crystallise and undergo a shift in solubility. The stability of solid dosage forms in storage may be studied using Raman mapping (Table 5). Raman mapping has provided insight into constituent distribution and API solid state in solid dosage forms produced by prilling. In the prilling process, constituents are melted together and then forced through a nozzle to form a liquid jet. Droplets are formed from this jet by applying vibrational energy, and these droplets subsequently cool

Table 5 Raman microscopic studies of other solid dosage forms. Author/s


Experiment type

Analysis method

Reference

Henson and Zhang 2006 Raman mapping was used to identify polymorphs within low dosage tablets.

Mapping (lateral)

[86]

Sasic

Mapping (lateral)

Qualitative, multivariate (partial least squares discriminant analysis, partial least squares and Euclidean distance classification) Qualitative, multivariate (principal components analysis)

Zoppetti et al.

Schoenherr et al. Sasic et al.

Vasanthavada et al. Vogt and Williams Breitkreitz et al.

Krier et al. Vervaeck et al.

De Bleye et al.

Year

2007 Assessed distribution of individual constituents in a five component tablet. 2007 Raman microscopic measurements were used to show freeze dried mixtures of progesterone and hydroxypropyl-p-cyclodextrin formed a solid inclusion complex. 2009 Polymorphic form and the distribution of an API in a liposomal dispersion were examined. 2011 The distribution of metformin HC1, hydroxypropyl cellulose and microcrystalline cellulose in granules resulting from wet granulation using varying amounts of water was assessed. 2011 Distribution and crystallinity of imatinib mesylate in high API dose formulation was assessed. 2012 The homogeneity of nanocrystalline ebselen in a PVP-VA matrix was assessed. 2013 Compound distribution as determined by Raman mapping was used to inform solvent choice and ratio of solvent to Gelucire® 44/14 for a formulation with the API atorvastatin calcium. 2013 Celecoxib distribution in ethyl-vinyl-acetate hot melt extruded implant was assessed. 2013 The crystallinity of metoprolol tartrate in a prilled fatty acid (stearic acid and behenic acid) formulation was assessed over time. 2014 Detection of 4-aminophenol impurity present at levels less than 0.1% w/w within paracetamol.

[73]

Point measurements

Qualitative, visual inspection of spectra

[106]

Mapping (lateral)

Qualitative, multivariate (augmented classical least squares) Qualitative, univariate (band intensity)

[78]

Mapping (lateral)

Mapping (lateral)

[88]

Mapping (lateral)

Qualitative, multivariate (augmented classical least squares) Qualitative, univariate (peak area)

[89]

Mapping (lateral)

Qualitative, multivariate (classical least squares)

[77]

Mapping (lateral)

Qualitative, multivariate (multivariate curve resolution) Qualitative, univariate (peak width)

[19] [69]

Qualitative, univariate (band intensity)

[107]

Mapping (lateral)

SERS mapping (lateral)

[79]


whilst falling through the prilling tower to form spherical particles. Prilling has recently been investigated as a manufacturing method for a lipidic-based sustained-release dosage forms [69]. Metoprolol tartrate was prilled with stearic acid or behenic acid to obtain spherically shaped particles with a narrow particle size distribution. Band broadening of the metoprolol tartrate peaks between 800 and 875 cm−1 and the absence of the metoprolol tartrate peak between 920 and 980 cm− 1 indicated that amorphous metoprolol tartrate was present within prilled formulations. In addition, a downshift of the C–N stretching vibration of metoprolol tartrate (1210 cm−1 to ca. 1206 cm−1) indicated there was an interaction between the fatty acid matrix and the API. Recrystallisation of metaprolol tartrate within this solid dosage form showed complete conversion after only two days of storage, with evidence drawn from Raman maps of crystallinity (i.e. width of the 820 cm− 1 metoprolol peak). Confirmation that metoprolol tartrate had recrystallized was provided by an analysis of the peak position of the C–N stretching vibration and the intensity of bands in the 920–980 cm−1 spectral range. Raman mapping has also been applied to high-dose polymericbased sustained-release formulations, which were produced by melt granulation. Release-modifying polymers typically make up at least half of the total weight of a polymeric-based sustainedrelease drug formulation [79]. This high polymer percentage is problematic for high-dose API systems as tablets made from these mixtures become too large (e.g. if 750–1000 mg of API is needed the total tablet weight becomes at least 1500–2000 mg). A high-dose system investigated with Raman spectroscopy involved imatinib mesylate, magnesium stearate, and hydroxypropylmethyl cellulose or ethyl cellulose. Components were heated to 185 °C and granulated: 185 °C is below the melting point of imatinib mesylate and above the glass transition temperature of either polymer. Raman mapping of tablets compressed from these granules showed that the API remained crystalline throughout the manufacturing process. The polymer in the respective formulations was present as a thin film

35

between API and magnesium stearate domains. Unfortunately, the polymer was difficult to resolve in the first suite of maps as the Raman signal of the polymer was substantially weaker than that of the API, and the polymer areas were small. A higher spatial resolution map of a smaller region of interest was recorded, revealing the pervasion of polymer throughout the formulation. This polymer pervasion accounted for the decreased friability, better compactibility and prolonged period of drug release of tablets formed from melt granulation when compared to those formed from wet granulation.

3.5. Drug dissolution The dissolution behaviour of a drug can be greatly influenced by the spatial distribution of each constituent within the formulation. Consequently, many Raman mapping studies have focused on the spatial arrangement of constituents within formulations (i.e. prior to dissolution) [75,70,71,76,81]. With knowledge of initial constituent distribution and individual constituent dissolution characteristics, a mechanism of formulation dissolution may be proposed. Several studies have gone beyond theorising about the dissolution mechanism and have applied Raman microscopy to the analysis of partially dissolved formulations to better understand dissolution behaviour (Table 6). Raman mapping has been used to assess constituent distribution in solid lipid extrudates before and after dissolution [30]. The lipid is a non-eroding matrix and therefore the dissolution medium must penetrate the matrix for the drug to dissolve and diffuse out of the matrix. Constituent distribution was assessed by tracking constituent-specific peaks for theophylline anhydrate (554 cm− 1), tripalmitin (1100 cm − 1 or 1130 cm − 1) and polyethylene glycol (844 and 860 cm− 1). Solubility information was gained by comparing the intensities of each peak before and after dissolution. In the solid lipid extrudate containing just tripalmitin and theophylline anhydrate there was a zone of drug loss in the outer area of the

Table 6 Raman mapping for better understanding of drug dissolution. Author/s

Year

Kang et al.

2006 Assessed distribution of paclitaxel in poly(ethyl-co-vinyl acetate) films after several intervals of dissolution.

Kang et al.

2007 Imaged the distribution of paclitaxel, polyethylene glycol and poly(lactide-co-glycolide) before and after dissolution in phosphate buffered saline. 2008 Imaged the distribution of paclitaxel in a poly(styrene-b-isobutylene-b-styrene) film before and after paclitaxel release. 2009 Imaged the distribution of sirolimus and arborescent polyisobutylene-block-polystyrene in films used to coat stents before and after a period of dissolution. 2010 The distribution of theophylline monohydrate in a tripalmitin, and a tripalmitin: polyethylene glycol extrudate was assessed before and after dissolution. 2011 Raman mapping was used to assess if a uniform receding API boundary was present in tripalmitin extrudates containing theophylline monohydrate, with and with polyethylene glycol. 2011 Raman mapping was used to examine the distribution of ibuprofen and metoclopramide HC1 in double-layered ternary-phase microparticles before and after dissolution. 2012 Characterised the distribution of sirolimus, poly(ethylene-co-vinyl acetate) and poly(n-butyl methacrylate) in stent coatings before and after dissolution. 2013 The migration of tamsulosin HCl dehydrate in a three-layered matrix tablet (other constituents were polyethylene glycol, polyethylene oxide, sodium chloride and magnesium sterate) was assessed using Raman spectroscopy at several time points during dissolution. 2013 Imaged the dissolution of theophylline anhydrate and the formation of theophylline monohydrate. Two dissolution media were used, water and methyl cellulose.

Kang et al.

Dong et al.

Windbergs et al. Haaser et al.

Lee et al.

Biggs et al.

Choi et al.

Fussell et al.


Experiment type Analysis method

Reference

CARS mapping (axial and lateral) CARS mapping (axial and lateral) CARS mapping (axial and lateral) Mapping (lateral and axial)

Qualitative, univariate (peak intensity), bivariate (intensity differences) Qualitative, univariate (peak intensity), bivariate (intensity differences) Qualitative, univariate (peak intensity), bivariate (intensity differences) Qualitative, univariate (peak area)

[108]

Mapping (lateral)


[93]

Mapping (lateral)


[30]

Mapping (axial)

Qualitative, multivariate (band target entropy minimisation)

[94]

Mapping (axial and lateral)

Qualitative, multivariate (linear combinations of pure component spectra) Qualitative

[92]

Univariate

[111]

Imaging (lateral)

Mapping

[109]

[110]

[90]

[95]

36


extrudate. The addition of polyethylene glycol to the formulation mix increased the drug dissolution rate and again an outer drugabsent zone was observed. In these extrudates the polyethylene glycol signal was weak after 10 min of dissolution and was not observed after 30 min. As expected, polyethylene glycol dissolved from the extrudate first, thereby increasing the drug surface area exposed to the dissolution medium and the drug dissolution rate. A uniform receding boundary of theophylline anhydrate was not observed in either form of the extrudate. The API dissolution was dependent on the tortuosity of pores and channels formed in the matrix during dissolution. Microparticle dissolution is often characterised by an initial burst of API release. Raman mapping has been used to understand the dissolution behaviour of a series of binary and ternary phase microparticles loaded with either ibuprofen, metoclopramide HCl, or both APIs [94]. The binary phase microparticles consisted of a PLGA core and PLLA shell, the ternary phase microparticles also included PCL in the shell. Raman mapping confirmed that ibuprofen was localised in the shell and metoclopramide HCl was localised in the core in both types of microparticle. Both BTEM analysis of Raman mapping data and a visual inspection of SEM data provided evidence that the release kinetics were a consequence of morphological differences and the rate of hydrolytic degradation. The release of ibuprofen was largely determined by microparticle morphology: poor dispersion of ibuprofen in the shell, as in the binary phase microparticles, resulted in uneven particle surfaces that allowed the ingress of water and therefore rapid release of ibuprofen. The release of ibuprofen was not substantially influenced by the presence or absence of metoclopramide HCl in the core. Metoclopramide HCl release rate was a consequence of hydrolytic degradation and morphology. After four days in the dissolution medium some of the PLGA core had hydrolytically degraded to polyglycolic acid in all microparticles and some metoclopramide HCl had diffused into the shell. After 11 days of dissolution metoclopramide HCl in ternary microparticles loaded with both APIs had diffused into the shell and was subsequently released into the dissolution medium. No ibuprofen was observed in these microparticles after 11 days. Metoclopramide HCl release from dual loaded binary microparticles was faster than ternary microparticles because the initial burst of ibuprofen release observed in binary phase microparticles created a porous shell for metoclopramide HCl to more easily diffuse through. Inclusion of PCL in the shell and retardation of this initial burst release allowed for more controlled release of metoclopramide HCl. Data from both atomic force microscopy and confocal Raman microscopy have been analysed in parallel, with the intention of correlating physical features to chemical signatures of a drug-eluting stent coating [92]. This study examined a CYPHER stent that had first been coated with parylene-C, and had then been coated with a mixture of poly(ethylene-co-vinyl acetate) [PEVA], poly(n-butyl methacrylate) [PBMA] and sirolimus. Atomic force microscopy and confocal Raman microscopy data recorded from the same sample before, during and after exposure to dissolution media could be correlated by using laser ablated spots that served as internal references. Three types of confocal Raman map were recorded, a surface map (40 × 40 μm), a subsurface map (40 × 40 μm, 2 μm below the surface) and a depth profile (40 × 20 μm) at time 0, 60 and 228 h. At time 0 h the coating was heterogeneous at the micron-scale, and by correlating the confocal Raman microscopy data with the atomic force microscopy measurements, enriched sirolimus regions could be observed as elevated topographical regions. The subsurface confocal Raman map showed that there were regions of high and low drug concentration. After 60 h the majority of sirolimus at the surface had been released and the drugrich protruding mounds observed at time 0 h were depressed regions and pores. At time 228 h no sirolimus Raman signal was observed, and the pores had deepened and widened relative to 60 h. By correlating atomic force microscopy and confocal Raman mapping data

chemical evidence was linked to physical properties of drug release from the medical device coating. 3.6. Percutaneous drug formulations Skin topical drug applications are an attractive method for avoiding several problems associated with injections (e.g. pain, possible infection, issues of compliance) [112]. Unfortunately from the perspective of epicutaneous applications, skin has been evolutionarily selected to be an effective barrier against chemical penetration (mammals have a relatively thick stratum corneum or outermost layer of skin). The production of formulations for transdermal drug delivery is further complicated by the high intra- and inter-person variability of stratum corneum thickness (the stratum corneum layer varies from 10 to 30 μm in most skin regions) [113]. Hence, both the dermal penetration rate and the depth of drug penetration are important considerations for the efficacy of a skin topical formulation. The tape-stripping method has traditionally been used to characterise formulations for transdermal drug delivery [114]. The tape-stripping method involves the serial removal of skin layers, and the component of interest is subsequently extracted from each layer and quantified. This method is both time consuming, destructive and lacks reproducible spatial resolution. In contrast, confocal Raman microscopy is a non-destructive method that is chemically selective and allows the observation of multiple components in one measurement. Two key advantages of CRM for characterising skin topical drug formulations are 1) the capacity to profile compound depth penetration in a single sample over multiple time points, and 2) the capacity for in vivo analyses. Confocal Raman microscopy has been used in both in vitro and in vivo studies of skin topical formulations to characterise both penetration rates and penetration depths (Table 7). Several studies have discovered that dermal drug profiling using CRM is impeded by signal attenuation, a consequence of reflectance, absorbance and elastic scattering within the tissue, and that the signal attenuation increases with depth into the tissue [115,116]. To mitigate the effect of signal attenuation during the analysis of spectral data, many studies normalise the intensity of an API-specific Raman spectral peak against the intensity of a Raman spectral peak attributed to skin [117,115]. Normalised spectral data can indicate the rate and the depth of penetration of an API, however, information about absolute concentration is not provided. Correlating this signal ratio to absolute concentration requires an assessment of intra- and inter-person variability of skin derived Raman peaks. Three Caucasian abdominal skin samples were examined in vitro to determine intra- and interindividual variability of Raman spectra [65]. Inter- and intraindividual variability of the intensities of four intrinsic skin peaks recorded from the skin surface were highly variable but not significantly different (995–1018 cm− 1 CC stretch of aromatic amino acids; 1288–1314 cm − 1 –CH2 deformation of alkyl chains in skin lipids; 1388–1497 cm − 1 CH deformations of both proteins and lipids; 1559–1721 cm− 1 amide I of proteins). At approximately 5 μm into the skin samples the lipid peak was no longer observable. All four peaks exhibited an exponential decay of intensity with depth, indicating a systematic bias introduced by Raman signal attenuation. The ratio of peak intensities is therefore a viable method for peak correlation to concentration, excepting the lipid peak 1288–1314 cm−1 if information from beyond 5 μm is needed. Modelling the relationship between Raman peak intensity and signal attenuation may remove the need for ratios between API-specific Raman peaks and skin-specific Raman peaks. To characterise the effect of signal attenuation on Raman peak intensity, a caffeine doped surrogate system with optical properties simulating the properties of human stratum corneum was studied [118]. An algorithm for signal attenuation correction was developed by modelling the intensity change of a caffeine band at 555 cm−1 with depth. This algorithm was subsequently applied to data collected from a human skin sample that had


37

Table 7 Confocal Raman microscopy studies for better understanding percutaneous drug delivery. Author/s

Year


Experiment type

Analysis method

Reference

Zhang et al.

2007


Qualitative, univariate (peak position), bivariate (peak area ratio)

[72]

Zhang et al.

2007


Qualitative, bivariate (peak intensity ratio)

[117]

Freudiger et al.

2008

[122]

2009

SRS mapping (axial and lateral) Mapping (axial)

Qualitative, univariate (peak intensity)

Melot et al.

2010


Qualitative, multivariate (fit of pure component spectra) Qualitative, univariate (peak area)

[115]

Gotter, Faubel and Neubert Saar et al.

2011



[124]

Franzen et al.

2013

Mapping (axial)


[118]

Belsey et al.

2014

[125]

2014

SRS and CARS mapping (axial and lateral) Mapping (axial)

Qualitative, univariate (peakintensity)

Franzen and Windbergs

The penetration of phosphorylated resveratrol into skin, and subsequent conversion to resveratrol was assessed using confocal Raman mapping. Used confocal Raman mapping to assess the distribution of prodrug and drug (l-ethyloxycarbonyl-5-fluorouracil and 5-fluorouracil respectively) after 20 h of permeation and metabolism into skin. Imaged the penetration of dimethyl sulfoxide and retinoic acid in mouse skin. The effect of penetration enhancers on the delivery of trans-retinol into human skin was examined. Used confocal Raman mapping to examine the extent of dithranol penetration into a DDC membrane at different time points. Assessed the penetration of ketoprofen, ibuprofen and co-solvent propylene glycol into mouse skin at different time intervals. Assessed signal attenuation with depth using a skin surrogate and caffeine, and calculated a signal correction algorithm. Imaged the permeation of ketoprofen, ibuprofen and co-solvent propylene glycol into skin. Assessment of the inter- and intra-individual variability of endogenous skin component Raman peaks

Qualitative (band area)

[65]

been bathed in a caffeine solution. The resulting intensity profile was nearly constant, except for the initial ca. 10 μm; this was not unexpected as the algorithm did not model the surrogate data well in this region. Application of this algorithm may allow quantification of API in skin tissue based on component specific Raman peak intensity only. The stratum corneum is the rate-limiting barrier to drug penetration and the thickness of this layer must be known to normalise drug permeation profiles in dermatopharmacokinetic studies. Water content increases from approximately 10% at the skin surface, to approximately 60% [113] near the interface between the stratum corneum and the underlying epidermis, which in turn contains an approximately constant water content of 60%. The thickness of the stratum corneum may therefore be determined from the inflection point in water concentration depth profiles. The water content is determined as the ratio of the water O–H stretching peak (3390 cm− 1) to the protein –CH3 peak (2935 cm−1) [119,120]. In vivo confocal Raman based measurements of stratum corneum thickness have been shown to be the same as those determined using confocal laser scanning microscopy [113]. The dermatopharmacokinetics of ibuprofen in polypropylene glycol, or in polypropylene glycol and water mixtures have been determined using in vivo confocal Raman measurements [121]. Skin samples were profiled to a depth of 40 μm in 4 μm steps and the collected ibuprofen signal was normalised to keratin (protein) to account for signal attenuation. To allow for comparison with previous tape-stripping data the ibuprofen signal was normalised to the highest ibuprofen signal observed across all samples. Each depth measurement was normalised to the respective stratum corneum thickness to remove the effect of inter-individual variability. Polypropylene glycol had a large effect on ibuprofen penetration, and the depth profiles observed were consistent with those reported from tape-stripping experiments. However, the absolute diffusion parameters calculated from tape-stripping data were higher than those determined from the Raman data. This difference may be a consequence of variation in stratum corneum thickness. In confocal Raman depth profiling experiments the stratum corneum thickness was determined from the same area that ibuprofen content was measured from; in tape-stripping experiments the stratum corneum thickness was determined from a region adjacent to the tape-stripping sampling site and therefore variation in thickness was not accounted for. Occlusion of the sampling sites occurred before either tape-

[123]

stripping or Raman measurements were recorded. This has been reported to increase stratum corneum hydration which may cause an increase in stratum corneum thickness. 3.7. Cells Confocal Raman mapping has been used to observe the uptake and metabolism of APIs by cells, and to differentiate between incorporation of nanoparticles into the cytoplasm from the adherence of nanoparticles to the cell membrane (Table 8). Raman mapping provides rich spatial information but is often time intensive, and cells are typically fixed prior to recording spectral data [66,91,126]. Spectral changes induced by formalin fixation should be considered with respect to the cellular changes of interest, i.e. the spectral changes induced by fixation should not obscure any spectral changes associated with a diseased state, or with the incorporation of a nanoparticle or API. Important regions for cell analyses include 400–1800 cm−1 and 2800–3100 cm−1 as cells are a mixture of lipids, proteins and nucleic acids [127,128]. Furthermore, many confocal Raman microscopy studies of cells use CaF2 as a growth substrate, and thus avoid fluorescent interference from glass [66,91,126,129]. The extent to which a cell can internalise a nanoparticle can be an important consideration for drug delivery, and Raman spectroscopy has usefully mapped nanoparticle distributions within cells. Poly(lactide-co-glycolide) nanoparticles that had been surface stabilised with either poly(ethylene imine) or bovine serum albumin were introduced into HepG2 cells and studied with Raman confocal microscopy [129]. Romero and colleagues examined the spectral differences associated with lipid bodies, cytoplasm and nucleus of cells, and by examining the C–H stretching region (2800–3100 cm−1) were able to show that the surface stabilised poly(lactide-co-glycolide) nanoparticles were most likely colocalised with lipid bodies. The incorporation of poly(lactide-co-glycolide) nanoparticles inside the cell was evident from a serial set of Raman spectra recorded at different depths within the cell. The surface of liposome nanoparticles may be modified with cationic compounds to increase efficacy and specificity of drug delivery to cells [126]. Confocal Raman mapping was used to assess the intra-cellular distribution of surface modified liposome nanoparticles. A key analytical

38


Table 8 Raman microscopy studies of APIs or particles in cells. Author/s

Year


Experiment type

Analysis method

Reference

van Apeldoorn et al. Mattaus et al.

2004

Mapping (lateral)

[66]


Qualitative, univariate (peak intensity), bivariate (peak intensity ratio) Qualitative, univariate (peak intensity)

Xu et al.

2009

CARS mapping (lateral)


[131]

Mouras et al.

2010

CARS mapping (lateral)


[132]

Romero et al.

2010

Point measurement

Qualitative, visual inspection of spectra.

[129]

Chernenko et al. Garrett et al.

2012

Observed poly(lactide-co-glycolide) microsphere degradation in macrophages. Assessed the intracellular distribution of deuterated liposomes (with and without a cell-penetrating peptide) in human breast adenocarcinoma MCF-7 cells at different treatment time points. Showed that poly(lactide-co-glycolide) nanoparticles were not incorporated in to KB human nasopharyngeal carcinoma cells. Imaged the uptake of doxorubicin into breast cancer tissue and MCF-7 cells. Assessed distribution of poly(lactide-co-glycolide) nanoparticles surface modified with either poly(ethylene imine) or bovine serum albumin in HepG2 cells. Imaged the intracellular distribution of three types of cationic modified deuterated liposomes in human HeLa cells. Imaged chitosan-based nanoparticle incorporation into tissue.


Qualitative, multivariate (vertex component analysis) Qualitative, univariate (peak intensity)

[126]

Ma et al.

2013


[134]

El-Mashtoly et al.

2014

Qualitative, univariate (peak area), multivariate (hierarchical cluster analysis)

[91]

2008

2012

Assessed the distribution of graphene oxide wrapped gold nanoparticles after seeding HeLa cells with these nanoparticles. Identified and assessed the distribution of erlotinib metabolite in human colorectal adenocarcinoma SW480 cells.

challenge of this study was that the Raman spectra of liposome nanoparticles greatly overlapped with the Raman spectra collected from cells. Accordingly, the liposomes were labelled with deuterium. A large relative change in mass occurs with deuteration and vibrational modes involving hydrogen are shifted. The most significant shift occurs with C–H stretching vibrations, and in deuterated compounds these stretching modes are observed in the 2050–3000 cm−1 spectral region. Vertex component analysis was used to further increase the contrast within the Raman maps. Regions attributed to liposomes, nuclei and cell body proteins, and membrane-rich organelles (e.g. mitochondria, Golgi body) were observed in the biochemical maps. In maps collected after 1 h of treatment, inclusion of the three types of liposome nanoparticles (surface modified with either 1,2-dioleoyl-3-trimethylammonium propane, stearyl amine or stearyltripheylphosphonium) was detected throughout the cell volume. More cationic liposome had been incorporated into cells compared with non-modified liposomes. Both the 1,2dioleoyl-3-trimethylammonium propane- and stearyl amine-modified liposomes were found in the periphery of the cell and this did not change after 3 or 6 h of incubation. The stearyltripheylphosphonium-modified liposomes were distributed in the cell periphery at 1 h, however, after 3 h of incubation these liposomes were colocalised with organelle areas. In addition, the vertex component analysis spectrum representing these organelle areas contained deuterated signals, which further indicated colocalisation of the organelles and stearyltripheylphosphoniummodified liposomes. This distribution agreed with previous studies that had shown an affinity between stearyltripheylphosphonium and mitochondria. The cellular uptake of an API that is not bound to a nanoparticle may also be examined with Raman confocal microscopy. Erlotinib is an epidermal growth factor receptor inhibitor used to treat metastatic nonsmall cell lung and pancreatic cancers [91]. This API is particularly well suited to distribution assessment by confocal Raman mapping as erlotinib contains an acetylene functional group. Acetylene stretching is a particularly strong Raman mode and is observed in a spectral region that is otherwise free from cell-specific peaks, 2095–2140 cm− 1. Peaks were observed in this spectral region in Raman spectra collected from cells that had been incubated with erlotinib for 12 h. However, there were significant differences between API specific peaks in APIcell spectra and pure erlotinib spectra, which suggested that erlotinib had been metabolised within the cell. Three metabolism pathways were suggested, two of which were excluded based on an analysis of the Raman spectra and functional groups that must be present within

CARS mapping (axial and lateral) SERS mapping (lateral) Mapping (axial and lateral)

[130]

[133]

the metabolite. The remaining and therefore proposed metabolic pathway was O-demethylation of the side chain. Good agreement between the erlotinib metabolite in cell spectra and a synthetic desmethylerlotinib spectrum supported this as the metabolic pathway. Hierarchical cluster analysis was applied to Raman spectral data recorded only from cells; background pixels were identified as having no Raman spectral intensity from 2800 to 3050 cm− 1 and were deleted from the dataset. After 12 h of incubation erlotinib was distributed in the cell periphery, the cluster analysis also identified the position of the nucleus, and several regions within the cytoplasm which represent various subcellular components. MCR-ALS, vertex component analysis and PCA were also applied to the spectral dataset, and all agreed with respect to erlotinib distribution. Depth profiling with confocal Raman microscopy confirmed that erlotinib was incorporated into the cell. 4. Conclusions Chemical mapping with Raman spectroscopy has been widely applied across the spectrum of pharmaceutical samples. Data in the majority of these studies were collected with point by point mapping systems, with chemical and distribution information revealed through univariate and multivariate statistical methods. Peak intensities, integrals and other univariate methods have illustrated the relative concentrations of constituents and the crystallinity of constituents. Likewise, multivariate methods have given insight into relative constituent concentration and crystallinity. Multivariate methods that do not require a priori information (e.g. multivariate curve resolution, band target entropy minimization) have proven advantageous for studies dealing with contaminant identification. Whilst few imaging studies have quantitatively assessed constituent distribution within samples, wider utilisation of Raman microscopes and decreased data collection times will likely encourage this field of research. It is anticipated that Raman mapping will become an increasingly routine analytical technique in pharmaceutical development and manufacture. References [1] X. Zhu, T. Xu, Q. Lin, Y. Duan, Technical development of Raman spectroscopy: from instrumental to advanced combined technologies, Appl. Spectrosc. Rev. 49 (2014) 64–82. [2] P. Kukura, D.W. McCamant, R.A. Mathies, Femtosecond stimulated Raman spectroscopy, Annu. Rev. Phys. Chem. 58 (2007) 461–488.

G.P.S. Smith et al. / Advanced Drug Delivery Reviews 89 (2015) 21–41 [3] F. El-Diasty, Coherent anti-Stokes Raman scattering: spectroscopy and microscopy, Vib. Spectrosc. 55 (2011) 1–37. [4] C.V. Raman, K.S. Krishnan, A new type of secondary radiation, Nature 121 (1928) 501–502. [5] R.L. McCreery, Raman Spectroscopy for Chemical Analysis, Wiley-Interscience, 2000. [6] N.J. Everall, Confocal Raman microscopy: common errors and artefacts, Analyst 135 (2010) 2512–2522. [7] B. Schrader, D.S. Moore, Nomenclature, symbols, units, and their usage in spectrochemical analysis—xviii. Laser-based molecular spectroscopy for chemical analysis-Raman scattering processes. (IUPAC recommendations 1997), Pure Appl. Chem. 69 (1997) 1451–1468. [8] C.J. Petty, P.J. Hendra, T. Jawhari, A standardized intensity scale for Fourier-transform Raman-spectra of liquids, Spectrochim. Acta A Mol. Biomol. Spectrosc. 47 (1991) 1189–1199. [9] H.W. Shrötter, H.W. Klöckner, A. Weber, Raman Spectroscopy of Gases and Liquids, Topics in Current Physics, Springer, New York, 1979. [10] M.O. Trulson, R.A. Mathies, Raman cross-section measurements in the visible and ultraviolet using an integrating cavity — application to benzene, cyclohexane, and cacodylate, J. Chem. Phys. 84 (1986) 2068–2074. [11] F.W.L. Esmonde-White, M.D. Morris, Emerging Raman applications and techniques in biomedical and pharmaceutical fields, biological and medical physics, biomedical engineering, Springer, New York, 2010. 97–110. [12] E. Lee, Raman imaging: techniques and applications, Vol. 264 of Springer Series in Optical Sciences, Springer-Verlag, Berlin Heidelberg, 2012. [13] J.M. Amigo, Practical issues of hyperspectral imaging analysis of solid dosage forms, Anal. Bioanal. Chem. 398 (2010) 93–109. [14] R.W. Kessler, Perspectives in process analysis, J. Chemom. 27 (2013) 369–378. [15] D. Pappas, B.W. Smith, J.D. Winefordner, Raman spectroscopy in bioanalysis, Talanta 51 (2000) 131–144. [16] H.R. Morris, C.C. Hoyt, P. Miller, P.J. Treado, Liquid crystal tunable filter Raman chemical imaging, Appl. Spectrosc. 50 (1996) 805–811. [17] H.R. Morris, C.C. Hoyt, P.J. Treado, Imaging spectrometers for fluorescence and Raman microscopy — acoustooptic and liquid-crystal tunable filters, Appl. Spectrosc. 48 (1994) 857–866. [18] A. Rinnan, F. van den Berg, S.B. Engelsen, Review of the most common preprocessing techniques for near-infrared spectra, Trends Anal. Chem. 28 (2009) 1201–1222. [19] F. Krier, J. Mantanus, P.-Y. Sacre, P.-F. Chavez, J. Thiry, A. Pestieau, E. Rozet, E. Ziemons, P. Hubert, B. Evrard, PAT tools for the control of co-extrusion implants manufacturing process, Int. J. Pharm. 458 (2013) 15–24. [20] A. de Juan, M. Maeder, T. Hancewicz, L. Duponchel, R. Tauler, Chemometric Tools for Image Analysis, Wiley-VCH Verlag, Weinheim, 2009. 65–109 (Ch. 2). [21] J.M. Prats-Montalbán, A. de Juan, A. Ferrer, Multivariate image analysis: a review with applications, Chemom. Intell. Lab. Syst. 107 (2011) 1–23. [22] M.J. Baker, J. Trevisan, P. Bassan, R. Bhargava, H.J. Butler, K.M. Dorling, P.R. Fielden, S.W. Fogarty, N.J. Fullwood, K.A. Heys, C. Hughes, P. Lasch, P.L. Martin-Hirsch, B. Obinaju, G.D. Sockalingum, J. Sule-Suso, R.J. Strong, M.J. Walsh, B.R. Wood, P. Gardner, F.L. Martin, Using Fourier transform IR spectroscopy to analyze biological materials, Nat. Protoc. 9 (2014) 1771–1791. [23] R. Bhargava, Infrared spectroscopic imaging: the next generation, Appl. Spectrosc. 66 (2012) 1091–1120. [24] T. van Dijk, D. Mayerich, P.S. Carney, R. Bhargava, Recovery of absorption spectra from Fourier Transform Infrared (FT-IR) microspectroscopic measurements of intact spheres, Appl. Spectrosc. 67 (2013) 546–552. [25] D. Oelkrug, E. Ostertag, R.W. Kessler, Quantitative raman spectroscopy in turbid matter: reflection or transmission mode? Anal. Bioanal. Chem. 405 (2013) 3367–3379. [26] WITec, High-resolution optical and scanning probe microscopy systems, WITec product catalogue (2012). [27] C.M. McGoverin, T. Rades, K.C. Gordon, Recent pharmaceutical applications of Raman and terahertz spectroscopies, J. Pharm. Sci. 97 (2008) 4598–4621. [28] S.J. Fraser, K.C. Gordon, Raman spectroscopy in the study of pharmaceuticals: the problems and solutions to sub-sampling and data analysis, Eur. Pharm. Rev. 19 (2014) 3–8. [29] J. Johansson, S. Pettersson, S. Folestad, Characterization of different laser irradiation methods for quantitative Raman tablet assessment, J. Pharm. Biomed. Anal. 39 (2005) 510–516. [30] M. Haaser, M. Windbergs, C.M. McGoverin, P. Kleinebudde, T. Rades, K.C. Gordon, C.J. Strachan, Analysis of matrix dosage forms during dissolution testing using Raman microscopy, J. Pharm. Sci. 100 (2011) 4452–4459. [31] N.J. Everall, Confocal Raman microscopy: performance, pitfalls, and best practice, Appl. Spectrosc. 63 (2009) 245a–262a. [32] G.D. McAnally, N.J. Everall, J.M. Chalmers, W.E. Smith, Analysis of thin film coatings on poly(ethylene terephthalate) by confocal Raman microscopy and surfaceenhanced Raman scattering, Appl. Spectrosc. 57 (2003) 44–50. [33] N.J. Everall, Modeling and measuring the effect of refraction on the depth resolution of confocal Raman microscopy, Appl. Spectrosc. 54 (2000) 773–782. [34] N.J. Everall, Confocal Raman microscopy: why the depth resolution and spatial accuracy can be much worse than you think, Appl. Spectrosc. 54 (2000) 1515–1520. [35] K.J. Baldwin, D.N. Batchelder, Confocal Raman microspectroscopy through a planar interface, Appl. Spectrosc. 55 (2001) 517–524. [36] J.P. Tomba, J.M. Pastor, Avoiding coupling fluid–sample interaction in confocal Raman depth profiling with immersion objectives, Appl. Spectrosc. 62 (2008) 817–819.

39

[37] M.J. Pelletier, Control of out-of-focus light intensity in confocal Raman microscopy using optical preprocessing, Appl. Spectrosc. 63 (2009) 591–596. [38] K. Buckley, P. Matousek, Recent advances in the application of transmission Raman spectroscopy to pharmaceutical analysis, J. Pharm. Biomed. Anal. 55 (2011) 645–652. [39] J.C. Burley, A. Alkhalil, M. Bloomfield, P. Matousek, Transmission Raman spectroscopy for quality control in model cocrystal tablets, Analyst 137 (2012) 3052–3057. [40] M. Bloomfield, D. Andrews, P. Loeffen, C. Tombling, T. York, P. Matousek, Noninvasive identification of incoming raw pharmaceutical materials using Spatially Offset Raman Spectroscopy, J. Pharm. Biomed. Anal. 76 (2013) 65–69. [41] J.C. Burley, A. Aina, P. Matousek, C. Brignell, Quantification of pharmaceuticals via transmission Raman spectroscopy: data sub-selection, Analyst 139 (2014) 74–78. [42] N. Kellichan, A. Nordon, P. Matousek, D. Littlejohn, G. McGeorge, Variation in the transmission near-infrared signal with depth in turbid media, Appl. Spectrosc. 68 (2014) 383–387. [43] N.A. Macleod, A. Goodship, A.W. Parker, P. Matousek, Prediction of sublayer depth in turbid media using spatially offset Raman spectroscopy, Anal. Chem. 80 (2008) 8146–8152. [44] C.M. McGoverin, M.D. Hargreaves, P. Matousek, K.C. Gordon, Pharmaceutical polymorphs quantified with transmission Raman spectroscopy, J. Raman Spectrosc. 43 (2012) 280–285. [45] M.V. Schulmerich, J.H. Cole, K.A. Dooley, M.D. Morris, J.M. Kreider, S.A. Goldstein, S. Srinivasan, B.W. Pogue, Noninvasive Raman tomographic imaging of canine bone tissue, J. Biomed. Opt. 13 (2) (2008) (020506–020506-3). [46] M.J. Adams, Chemometrics in analytical spectroscopy, RSC Analytical Spectroscopy Monographs, 2nd edition, Royal Society of Chemistry, 2004. [47] K.R. Beebe, R.J. Pell, M.B. Seasholtz, Chemometrics a practical guide, WileyInterscience Series on Laboratory Automation, John Wiley and Sons Inc, United States of America, 1998. [48] T. Bocklitz, A. Walter, K. Hartmann, P. Roesch, J. Popp, How to pre-process Raman spectra for reliable and stable models? Anal. Chim. Acta. 704 (1–2) (2011) 47–56. [49] S. Schönbichler, L. Bittner, A. Weiss, U. Griesser, J. Pallua, C. Huck, Comparison of NIR chemical imaging with conventional NIR, Raman and ATR-IR spectroscopy for quantification of furosemide crystal polymorphs in ternary powder mixtures, Eur. J. Pharm. Biopharm. 84 (2013) 616–625. [50] P.Y. Sacré, P. Lebrun, P.F. Chavez, C.D. Bleye, L. Netchacovitch, E. Rozet, R. Klinkenberg, B. Steel, P. Hubert, E. Ziemons, A new criterion to assess distributional homogeneity in hyperspectral images of solid pharmaceutical dosage forms, Anal. Chim. Acta. 818 (2014) 7–14. [51] J.M. Shaver, Handbook of Raman spectroscopy from the research laboratory to the process line, Practical Spectroscopy, Ch. 7 Chemometrics for Raman Spectroscopy, Marcel Dekker Inc, 2001. 275–306. [52] Camo software AS, The Unscrambler® appendices: method references, PDF from CAMO website, http://www.camo.com/downloads/U9.6pdfmanual/ TheUnscramblerMethodReferences.pdf (URL http://www.camo.com/downloads/ U9.6pdfmanual/TheUnscramblerMethodReferences.pdf). [53] A. Savitsky, M.J.E. Golay, Smoothing differentiation of data by simplified least squares procedures, Anal. Chem. 36 (1964) 1627–1639. [54] D. Chen, Z. Chen, E. Grant, Adaptive wavelet transform suppresses background and noise for quantitative analysis by Raman spectrometry, Anal. Bioanal. Chem. 400 (2011) 625–634. [55] A. Hédoux, L. Paccou, Y. Guinet, J.F. Willart, M. Descamps, Using the low-frequency Raman spectroscopy to analyze the crystallization of amorphous indomethacin, Eur. J. Pharm. Sci. 38 (2009) 156–164. [56] M. Pirzer, J. Sawatzki, Method and Device for Correcting a Spectrum, April 2008. [57] P. Matousek, N. Everall, M. Towrie, A.W. Parker, Depth profiling in diffusely scattering media using Raman spectroscopy and picosecond Kerr gating, Appl. Spectrosc. 59 (2005) 200–205. [58] R. Littleford, P. Matousek, M. Towrie, A.W. Parker, G. Dent, R.J. Lacey, W.E. Smith, Raman spectroscopy of street samples of cocaine obtained using Kerr gated fluorescence rejection, Analyst 129 (2004) 505–506. [59] S.E.J. Bell, E.S.O. Bourguignon, A. Dennis, Analysis of luminescent samples using subtracted shifted Raman spectroscopy, Analyst 123 (1998) 1729–1734. [60] A. O'Grady, A.C. Dennis, D. Denvir, J.J. McGarvey, S.E.J. Bell, Quantitative Raman spectroscopy of highly fluorescent samples using pseudosecond derivatives and multivariate analysis, Anal. Chem. 73 (2001) 2058–2065. [61] CAMO, The Unscrambler X Supporting Information, Software Contents File, 2009–2012. [62] K.H. Esbensen, Multivariate Data Analysis — In Practice, 5th edition CAMO Software As, 2001. [63] P.C. Wiliams, D. Sobering, Comparison of commercial near infrared transmittance and reflectance instruments for analysis of whole grains and seeds, J. Near Infrared Spectrosc. 1 (1993) 25–32. [64] K.M. Balss, G. LIanos, G. Papandreou, C.A. Maryanoff, Quantitative spatial distribution of sirolimus and polymers in drug-eluting stents using confocal Raman microscopy, J. Biomed. Mater. Res. A 85A (2008) 258–270. [65] L. Franzen, M. Windbergs, Accessing Raman spectral variability in human stratum corneum for quantitative in vitro depth profiling, J. Raman Spectrosc. 45 (2014) 82–88. [66] A.A. van Apeldoorn, H.J. van Manen, J.M. Bezemer, J.D. de Bruijn, C.A. van Blitterswijk, C. Otto, Raman imaging of PLGA microsphere degradation inside macrophages, J. Am. Chem. Soc. 126 (2004) 13226–13227. [67] K. Rizi, R.J. Green, O. Khutoryanskaya, M. Donaldson, A.C. Williams, Mechanisms of burst release from pH-responsive polymeric microparticles, J. Pharm. Pharmacol. 63 (2011) 1141–1155.

40


[68] A. Almeida, L. Brabant, F. Siepmann, T. De Beer, W. Bouquet, L. Van Hoorebeke, J. Siepmann, J.P. Remon, C. Vervaet, Sustained release from hot-melt extruded matrices based on ethylene vinyl acetate and polyethylene oxide, Eur. J. Pharm. Biopharm. 82 (2012) 526–533. [69] A. Vervaeck, L. Saerens, B.G. De Geest, T. De Beer, R. Carleer, P. Adriaensens, J.P. Remon, C. Vervaet, Prilling of fatty acids as a continuous process for the development of controlled release multiparticulate dosage forms, Eur. J. Pharm. Biopharm. 85 (2013) 587–596. [70] E. Karavas, M. Georgarakis, A. Docoslis, D. Bikiaris, Combining SEM, TEM, and micro-Raman techniques to differentiate between the amorphous molecular level dispersions and nanodispersions of a poorly water-soluble drug within a polymer matrix, Int. J. Pharm. 340 (2007) 76–83. [71] J. Breitenbach, W. Schrof, J. Neumann, Confocal Raman-spectroscopy: analytical approach to solid dispersions and mapping of drugs, Pharm. Res. 16 (1999) 1109–1113. [72] G. Zhang, C.R. Flach, R. Mendelsohn, Tracking the dephosphorylation of resveratrol triphosphate in skin by confocal Raman microscopy, J. Control. Release 123 (2007) 141–147. [73] S. Šašić, An in-depth analysis of Raman and near-infrared chemical images of common pharmaceutical tablets, Appl. Spectrosc. 61 (2007) 239–250. [74] A. Miro, I. d'Angelo, A. Nappi, P. La Manna, M. Biondi, L. Mayol, P. Musto, R. Russo, M.I. La Rotonda, F. Ungaro, F. Quaglia, Engineering poly(ethylene oxide) buccal films with cyclodextrin: a novel role for an old excipient? Int. J. Pharm. 452 (2013) 283–291. [75] A. Balogh, G. Dravavoelgyi, K. Farago, A. Farkas, T. Vigh, P.L. Soti, I. Wagner, J. Madarasz, H. Pataki, G. Marosi, Z.K. Nagy, Plasticized drug-loaded melt electrospun polymer mats: characterization, thermal degradation, and release kinetics, J. Pharm. Sci. 103 (2014) 1278–1287. [76] Z.K. Nagy, A. Balogh, B. Vajna, A. Farkas, G. Patyi, A. Kramarics, G. Marosi, Comparison of electrospun and extruded soluplus (R)-based solid dosage forms of improved dissolution, J. Pharm. Sci. 101 (2012) 322–332. [77] M.C. Breitkreitz, G.P. Sabin, G. Polla, R.J. Poppi, Characterization of semi-solid self-emulsifying drug delivery systems (SEDDS) of atorvastatin calcium by Raman image spectroscopy and chemometrics, J. Pharm. Biomed. Anal. 73 (2013) 3–12. [78] C. Schoenherr, T. Haefele, K. Paulus, G. Francese, Confocal Raman microscopy to probe content uniformity of a lipid based powder for inhalation: a quality by design approach, Eur. J. Pharm. Biopharm. 38 (2009) 47–54. [79] M. Vasanthavada, Y. Wang, T. Haefele, J.P. Lakshman, M. Mone, W. Tong, Y.M. Joshi, A.T.M. Serajuddin, Application of melt granulation technology using twin-screw extruder in development of high-dose modified-release tablet formulation, J. Pharm. Sci. 100 (2011) 1923–1934. [80] R. Tauler, A. de Juan, Practical guide to chemometrics, 2nd edition, Ch. 11 Multivariate Curve Resolution, CRC Press, Taylor and Francis Group, 2006. 417–473. [81] B. Vajna, H. Pataki, Z. Nagy, I. Farkas, G. Marosi, Characterization of melt extruded and conventional isoptin formulations using Raman chemical imaging and chemometrics, Int. J. Pharm. 419 (2011) 107–113. [82] E. Widjaja, R.K.H. Seah, Application of Raman microscopy and band-target entropy minimization to identify minor components in model pharmaceutical tablets, J. Pharm. Biomed. Anal. 46 (2008) 274–281. [83] E. Widjaja, N. Crane, T.C. Chen, M.D. Morris, M.A. Ignelzi, B.R. McCreadie, Bandtarget entropy minimization (BTEM) applied to hyperspectral Raman image data, Appl. Spectrosc. 57 (2003) 1353–1362. [84] E. Widjaja, C.Z. Li, M. Garland, Semi-batch homogeneous catalytic in-situ spectroscopic data. FTIR spectral reconstructions using band-target entropy minimization (BTEM) without spectral preconditioning, Organometallics 21 (2002) 1991–1997. [85] E. Widjaja, P. Kanaujia, G. Lau, W.K. Ng, M. Garland, C. Saal, A. Hanefeld, M. Fischbach, M. Maio, R.B.H. Tan, Detection of trace crystallinity in an amorphous system using Raman microscopy and chemometric analysis, Eur. J. Pharm. Biopharm. 42 (2011) 45–54. [86] M.J. Henson, L. Zhang, Drug characterization in low dosage pharmaceutical tablets using Raman microscopic mapping, Appl. Spectrosc. 60 (2006) 1247–1255. [87] L. Zhang, M.J. Henson, S.S. Sekulic, Multivariate data analysis for Raman imaging of a model pharmaceutical tablet, Anal. Chim. Acta. 545 (2005) 262–278. [88] S. Šašić, S.I. Loranger, B.A. Johnson, Characterizing the structure of pharmaceutical granules obtained by wet granulation with varying amounts of water via Raman chemical imaging, Appl. Spectrosc. 65 (2011) 1291–1299. [89] F.G. Vogt, G.R. Williams, Analysis of a nanocrystalline polymer dispersion of ebselen using solid-state NMR, Raman microscopy, and powder X-ray diffraction, Pharm. Res. 29 (2012) 1866–1881. [90] J. Dong, J.D. Foley, C.D. Frethem, R.A. Hoerr, M.J. Matuszewski, J.E. Puskas, G. Haugstad, Multimodal dynamic imaging of therapeutic biomedical coatings in aqueous medium, Langmuir 25 (2009) 5442–5445. [91] S.F. El-Mashtoly, D. Petersen, H.K. Yosef, A. Mosig, A. Reinacher-Schick, C. Koetting, K. Gerwert, Label-free imaging of drug distribution and metabolism in colon cancer cells by Raman microscopy, Analyst 139 (2014) 1155–1161. [92] K.B. Biggs, K.M. Balss, C.A. Maryanoff, Pore networks and polymer rearrangement on a drug-eluting stent as revealed by correlated confocal Raman and atomic force microscopy, Langmuir 28 (2012) 8238–8243. [93] M. Windbergs, M. Haaser, C.M. McGoverin, K.C. Gordon, P. Kleinebudde, C.J. Strachan, Investigating the relationship between drug distribution in solid lipid matrices and dissolution behaviour using Raman spectroscopy and mapping, J. Pharm. Sci. 99 (2010) 1464–1475. [94] W.L. Lee, C. Loei, E. Widjaja, S.C.J. Loo, Altering the drug release profiles of double-layered ternary-phase microparticles, J. Control. Release 151 (2011) 229–238.

[95] D.H. Choi, K.H. Kim, J.S. Park, S.H. Jeong, K. Park, Evaluation of drug delivery profiles in geometric three-layered tablets with various mechanical properties, in vitro–in vivo drug release, and Raman imaging, J. Control. Release 172 (2013) 763–772. [96] N. Furuyama, S. Hasegawa, T. Hamaura, S. Yada, H. Nakagami, E. Yonemochi, K. Terada, Evaluation of solid dispersions on a molecular level by the Raman mapping technique, Int. J. Pharm. 361 (2008) 12–18. [97] D.N. Bikiaris, Solid dispersions, part II: new strategies in manufacturing methods for dissolution rate enhancement of poorly water-soluble drugs, Expert Opin. Drug Deliv. 8 (2011) 1663–1680. [98] K. Udipi, M. Chen, P. Cheng, K. Jiang, D. Judd, A. Caceres, R.J. Melder, J.N. Wilcox, Development of a novel biocompatible polymer system for extended drug release in a next-generation drug-eluting stent, J. Biomed. Mater. Res. A 85A (2008) 1064–1071. [99] M. Castellino, V. Stolojan, A. Virga, M. Rovere, K. Cabiale, M.R. Galloni, A. Tagliaferro, Chemico-physical characterisation and in vivo biocompatibility assessment of DLCcoated coronary stents, Anal. Bioanal. Chem. 405 (2013) 321–329. [100] K.M. Balss, F.H. Long, V. Veselov, A. Orana, E. Akerman-Revis, G. Papandreou, C.A. Maryanoff, Multivariate analysis applied to the study of spatial distributions found in drug-eluting stent, coatings by confocal Raman microscopy, Anal. Chem. 80 (2008) 4853–4859. [101] C. McManamon, P. Delaney, C. Kavanagh, J.J. Wang, S. Rasappa, M.A. Morris, Depth profiling of PLGA copolymer in a novel biomedical bilayer using confocal Raman spectroscopy, Langmuir 29 (2013) 5905–5910. [102] A. Rafati, A. Boussahel, K.M. Shakesheff, A.G. Shard, C.J. Roberts, X. Chen, D.J. Scurr, S. Rigby-Singleton, P. Whiteside, M.R. Alexander, M.C. Davies, Chemical and spatial analysis of protein loaded PLGA microspheres for drug delivery applications, J. Control. Release 162 (2012) 321–329. [103] E. Widjaja, W.L. Lee, S.C.J. Loo, Application of Raman microscopy to biodegradable double-walled microspheres, Anal. Chem. 82 (2010) 1277–1282. [104] W.H. Doub, W.P. Adams, J.A. Spencer, L.F. Buhse, M.P. Nelson, P.J. Treado, Raman chemical imaging for ingredient-specific particle size characterization of aqueous suspension nasal spray formulations: a progress report, Pharm. Res. 24 (2007) 934–945. [105] L. Sievens-Figueroa, A. Bhakay, J.I. Jerez-Rozo, N. Pandya, R.J. Romanach, B. Michniak-Kohn, Z. Iqbal, E. Bilgili, R.N. Dave, Preparation and characterization of hydroxypropyl methyl cellulose films containing stable BCS class II drug nanoparticles for pharmaceutical applications, Int. J. Pharm. 423 (2012) 496–508. [106] G. Zoppetti, N. Puppini, F. Ospitali, A. Fini, Pharmaceutics, preformulation and drug delivery — solid state characterization of progesterone in a freeze dried 1: 2 progesterone/HPBCD mixture, J. Pharm. Sci. 96 (2007) 1729–1736. [107] C. De Bleye, P.Y. Sacre, E. Dumont, L. Netchacovitch, P.F. Chavez, G. Piel, P. Lebrun, P. Hubert, E. Ziemons, Development of a quantitative approach using surfaceenhanced Raman chemical imaging: first step for the determination of an impurity in a pharmaceutical model, J. Pharm. Biomed. Anal. 90 (2014) 111–118. [108] E. Kang, H. Wang, I.K. Kwon, J. Robinson, K. Park, J.-X. Cheng, In situ visualization of paclitaxel distribution and release by coherent anti-Stokes Raman scattering microscopy, Anal. Chem. 78 (2006) 8036–8043. [109] E. Kang, J. Robinson, K. Park, J.X. Cheng, Paclitaxel distribution in poly(ethylene glycol)/poly(lactide-co-glycolic acid) blends and its release visualized by coherent anti-Stokes Raman scattering microscopy, J. Control. Release 122 (2007) 261–268. [110] E. Kang, H. Wang, I.K. Kwon, Y.-H. Song, K. Kamath, K.M. Miller, J. Barry, J.-X. Cheng, K. Park, Application of coherent anti-Stokes Raman scattering microscopy to image the changes in a paclitaxel-poly(styrene-b-isobutylene-b-styrene) matrix pre- and post-drug elution, J. Biomed. Mater. Res. A 87A (2008) 913–920. [111] A. Fussell, E. Garbacik, H. Offerhaus, P. Kleinebudde, C. Strachan, In situ dissolution analysis using coherent anti-Stokes Raman scattering (CARS) and hyperspectral CARS microscopy, Eur. J. Pharm. Biopharm. 85 (2013) 1141–1147. [112] O.G. Jepps, Y. Dancik, Y.G. Anissimov, M.S. Roberts, Modeling the human skin barrier — towards a better understanding of dermal absorption, Adv. Drug Deliv. Rev. 65 (2013) 152–168. [113] A. Boehling, S. Bielfeldt, A. Himmelmann, M. Keskin, K.P. Wilhelm, Comparison of the stratum corneum thickness measured in vivo with confocal Raman spectroscopy and confocal reflectance microscopy, Skin Res. Technol. 20 (2014) 50–57. [114] K. Myer, H. Maibach, Stratum corneum evaluation methods: overview, Skin Res. Technol. 19 (2013) 213–219. [115] M. Melot, P.D.A. Pudney, A.-M. Williamson, P.J. Caspers, A. Van Der Pol, G.J. Puppels, Studying the effectiveness of penetration enhancers to deliver retinol through the stratum cornum by in vivo confocal Raman spectroscopy, J. Control. Release 138 (2009) 32–39. [116] S. Tfaili, C. Gobinet, G. Josse, J.-F. Angiboust, M. Manfait, O. Piot, Confocal Raman microspectroscopy for skin characterization: a comparative study between human skin and pig skin, Analyst 137 (2012) 3673–3682. [117] G. Zhang, D.J. Moore, K.B. Sloan, C.R. Flach, R. Mendelsohn, Imaging the prodrug-todrug transformation of a 5-fluorouracil derivative in skin by confocal Raman microscopy, J. Invest. Dermatol. 127 (2007) 1205–1209. [118] L. Franzen, D. Selzer, J.W. Fluhr, U.F. Schaefer, M. Windbergs, Towards drug quantification in human skin with confocal Raman microscopy, Eur. J. Pharm. Biopharm. 84 (2013) 437–444. [119] P.J. Caspers, G.W. Lucassen, H.A. Bruining, G.J. Puppels, Automated depthscanning confocal Raman microspectrometer for rapid in vivo determination of water concentration profiles in human skin, J. Raman Spectrosc. 31 (2000) 813–818. [120] P.J. Caspers, G.W. Lucassen, E.A. Carter, H.A. Bruining, G.J. Puppels, In vivo confocal Raman microspectroscopy of the skin: noninvasive determination of molecular concentration profiles, J. Invest. Dermatol. 116 (2001) 434–442.

G.P.S. Smith et al. / Advanced Drug Delivery Reviews 89 (2015) 21–41 [121] R. Mateus, H. Abdalghafor, G. Oliveira, J. Hadgraft, M.E. Lane, A new paradigm in dermatopharmacokinetics — confocal Raman spectroscopy, Int. J. Pharm. 444 (2013) 106–108. [122] C.W. Freudiger, W. Min, B.G. Saar, S. Lu, G.R. Holtom, C. He, J.C. Tsai, J.X. Kang, X.S. Xie, Label-free biomedical imaging with high sensitivity by stimulated Raman scattering microscopy, Science 322 (2008) 1857–1861. [123] B. Gotter, W. Faubel, R.H.H. Neubert, FTIR microscopy and confocal Raman microscopy for studying lateral drug diffusion from a semisolid formulation, Eur. J. Pharm. Biopharm. 74 (2010) 14–20. [124] B.G. Saar, L.R. Contreras-Rojas, X.S. Xie, R.H. Guy, Imaging drug delivery to skin with stimulated Raman scattering microscopy, Mol. Pharm. 8 (2011) 969–975. [125] N.A. Belsey, N.L. Garrett, L.R. Contreras-Rojas, A.J. Pickup-Gerlaugh, G.J. Price, J. Moger, R.H. Guy, Evaluation of drug delivery to intact and porated skin by coherent Raman scattering and fluorescence microscopies, J. Control. Release 174 (2014) 37–42. [126] T. Chernenko, R.R. Sawant, M. Miljkovic, L. Quintero, M. Diem, V. Torchilin, Raman microscopy for noninvasive imaging of pharmaceutical nanocarriers: intracellular distribution of cationic liposomes of different composition, Mol. Pharm. 9 (2012) 930–936. [127] C. Krafft, T. Knetschke, A. Siegner, R.H.W. Funk, R. Salzer, Mapping of single cells by near infrared Raman microspectroscopy, Vib. Spectrosc. 32 (2003) 75–83. [128] D. Naumann, FT-infrared and FT-Raman spectroscopy in biomedical research, Appl. Spectrosc. Rev. 36 (2001) 239–298.

41

[129] G. Romero, I. Estrela-Lopis, J. Zhou, E. Rojas, A. Franco, C. Sanchez Espinel, A. Gonzalez Fernandez, C. Gao, E. Donath, S.E. Moya, Surface engineered poly(lactide-co-glycolide) nanoparticles for intracellular delivery: uptake and cytotoxicity — a confocal Raman microscopic study, Biomacromolecules 11 (2010) 2993–2999. [130] C. Matthaeus, A. Kale, T. Chernenko, V. Torchilin, M. Diem, New ways of imaging uptake and intracellular fate of liposomal drug carrier systems inside individual cells, based on Raman microscopy, Mol. Pharm. 5 (2008) 287–293. [131] P. Xu, E. Gullotti, L. Tong, C.B. Highley, D.R. Errabelli, T. Hasan, J.-X. Cheng, D.S. Kohane, Y. Yeo, Intracellular drug delivery by poly(lactic-co-glycolic acid) nanoparticles, revisited, Mol. Pharm. 6 (2009) 190–201. [132] R. Mouras, G. Rischitor, A. Downes, D. Salter, A. Elfick, Nonlinear optical microscopy for drug delivery monitoring and cancer tissue imaging, J. Raman Spectrosc. 41 (2010) 848–852. [133] N.L. Garrett, A. Lalatsa, D. Begley, L. Mihoreanu, I.F. Uchegbu, A.G. Schoetzlein, J. Moger, Label-free imaging of polymeric nanomedicines using coherent antiStokes Raman scattering microscopy, J. Raman Spectrosc. 43 (2012) 681–688. [134] X. Ma, Q. Qu, Y. Zhao, Z. Luo, Y. Zhao, K.W. Ng, Y. Zhao, Graphene oxide wrapped gold nanoparticles for intracellular Raman imaging and drug delivery, J. Mater. Chem. B 1 (2013) 6495–6500.

Biomedical Imaging in Implantable Drug Delivery Systems.

Drug Delivery Systems for Imaging and Therapy of Parkinson's Disease.

Drug delivery systems 5A. Oral drug delivery.

Drug delivery systems. 6. Transdermal drug delivery.

Implantable drug-delivery systems.

Confocal Raman investigation of diffusion processes in monolithic type transdermal drug delivery systems.

Thorough characterization of a Self-Emulsifying Drug Delivery System with Raman hyperspectral imaging: a case study.

Lipid-Based Drug Delivery Systems.

ATP-Responsive Drug Delivery Systems.

MEMS: Enabled Drug Delivery Systems.

Kidney-targeted drug delivery systems.

pH-responsive drug-delivery systems.

Implantable neurosurgical drug delivery systems.

Radiation sterilization of new drug delivery systems.

Evaluating drug delivery with salt formation: Drug disproportionation studied in situ by ATR-FTIR imaging and Raman mapping.

Ocular drug delivery systems: An overview.

Supramolecular hydrogels as drug delivery systems.

Marine Origin Polysaccharides in Drug Delivery Systems.

Drug delivery systems versus dosage forms.

Pulmonary drug delivery systems for tuberculosis treatment.

Drug delivery from gelatin-based systems.

Mucoadhesive systems in oral drug delivery.

Advanced drug delivery systems for therapeutic applications.

Implantable drug delivery systems: practical considerations.