Correspondence: David J. Moore, Center for Skin Science and Applied Dermatology, TRI/Princeton, 601 Prospect Avenue, Princeton, NJ, USA. Tel.: +1-609-430-4825; fax: +1-609-683-7149; e-mail: firstname.lastname@example.org
This primer describes and illustrates experimental protocols for both Fourier transform infrared (FTIR) spectroscopic imaging and confocal Raman mapping of ex vivo skin and thereby acquaints the reader with these measurement techniques, including the temporal and spatial limitations associated with each technique. The experimental conditions by which the unique ‘molecular histology’ information obtained from confocal Raman mapping and infrared spectroscopic mapping of ex vivo skin is generated will be described. Raman and FTIR spectra of tissue, when collected in spatially resolved arrays, permit the generation of ‘molecular images’ of tissue components and tissue organization without the use of fluorescent labels or chemical stains. To illustrate the molecular information from ex vivo skin that can be spectroscopically imaged with confocal Raman and infrared microspectroscopy, we have collected new data using both techniques and generated spectral images which illustrate the capacity of each technique to provide unique insights into skin histology, biochemistry and biophysics. Understanding the measurement possibilities and specific constraints of both approaches is a prerequisite to their meaningful use as powerful research tools in skin research.
Cet article décrit et illustre les protocoles expérimentaux pour les deux techniques d'imagerie, la transformée de Fourier infrarouge (FTIR) et l'imagerie spectroscopique Raman confocale, appliquées à la cartographie de la peau ex vivo, et familiarise ainsi le lecteur avec ces techniques de mesure, y compris avec les limites temporelles et spatiales associées à chaque technique. Les conditions expérimentales dans lesquelles l'unique “histologie moléculaire” des informations obtenues à partir de la Cartographie confocale Raman et infrarouge de la peau ex vivo est générée seront décrites. Les spectres Raman et FTIR du tissu, collectés dans des tableaux de résolution spatiale, permettent de produire des images “moléculaires” des composants du tissu et de l'organisation du tissu sans l'utilisation de marqueurs fluorescents ou de coloration chimique. Pour illustrer l'information moléculaire de la peau ex vivo qui peut être imagée avec les spectroscopies confocale Raman et infrarouge, nous avons recueilli de nouvelles données à l'aide des deux techniques et avons généré des images spectrales qui illustrent la capacité de chaque technique à fournir un éclairage unique sur l'histologie de la peau, la biochimie et la biophysique. Comprendre les possibilités de mesure et des contraintes spécifiques des deux approches est une condition préalable à leur utilisation significative en tant qu'outils puissants de recherche dans la peau.
Over the last 15 years, there has been an exponential increase in the use of spectroscopic imaging methods to address many biological and biophysical questions in skin science and dermatology. Most of the non-clinical applications of these imaging techniques in skin science have focused on questions concerning the dermal and transdermal delivery of molecules from topical cosmetic and pharmaceutical formulations. Complementing this work is a growing literature describing experiments utilizing confocal Raman and infrared imaging spectroscopy to map the relative concentration and distribution of endogenous molecules in skin including water and natural moisturizing factor (NMF) concentration gradients [1-4]. An upcoming review article in this journal will address the topic of confocal Raman and infrared spectroscopy imaging methods in dermal and transdermal delivery.
The current work is conceived as a primer that describes and illustrates ‘standard’ experimental protocols for both IR imaging and confocal Raman spectroscopy of ex vivo skin and thereby acquaints the reader with these measurement techniques. Explaining the temporal and spatial limitations associated with each technique will allow the reader to appreciate the experimental conditions by which the unique ‘molecular histology’ information obtained from confocal Raman and infrared spectroscopic mapping of ex vivo skin is generated. As we will describe, the information inherent in the Raman and IR spectra of tissue, when collected with high-resolution spatial information, presents the possibility of generating ‘molecular images’ of tissue components and tissue organization without the use of fluorescent labels or chemical stains. Understanding the measurement possibilities as well as the specific constraints of both confocal Raman and infrared imaging spectroscopy studies of ex vivo skin is a necessary requisite to their meaningful use as powerful research tools in skin research.
The generation of confocal Raman and IR spectral maps and images of ex vivo skin, as shown throughout this chapter, can be accomplished with the use of both univariate analyses and multivariate statistical analysis of the spatially resolved spectral data. The decision as to which data reduction protocols to employ is dependent on several considerations, some choices and other necessities, a number of which are addressed in the upcoming sections.
A great many original technical papers have been published in recent years utilizing confocal Raman and infrared imaging spectroscopy to probe molecular structure and composition in biomedical samples. The current work will not review this extensive literature; such an undertaking is well beyond the scope and purpose of this primer. However, several recently published review articles and books address the topic and contain references to the primary literature [1, 5-11].
To satisfy our pedagogical purpose in this primer, and by way of illustrating the quantity of molecular information from ex vivo skin that can be spectroscopically imaged with confocal Raman and infrared microspectroscopy, we have collected and processed new experimental data using both techniques. From these new data, we have generated spectral images that illustrate the capacity of each technique to provide unique insights into skin histology, biochemistry and biophysics.
Materials and methods
Human abdominal skin specimens were obtained from dermatological offices (otherwise to be discarded) with informed consent and in accordance with institutional protocols. Specimens were fast-frozen in liquid nitrogen after the removal of subcutaneous fat and stored at −20°C for <1 year. Skin was microtomed to ∼7 μm thick perpendicular to the SC surface and placed on CaF2 IR windows for IR imaging measurements. For confocal Raman microscopic measurements, intact skin specimens were placed in a custom-built brass sample cell with the SC up and sealed with a glass coverslip, so that the SC was in good contact with the coverslip and sample hydration was maintained. Experimental protocols such as sample thickness for IR imaging, or sample shrinkage during the acquisition for Raman measurements, are specific to the experimental design and are discussed in several of the references.
FTIR Imaging Microspectroscopy
Infrared images were acquired with a Perkin Elmer Spotlight 300 system (Waltham, MA, USA) using the transmission mode and a 6.25 μm2 pixel size. The instrument has an essentially linear array of (16 × 1) detector elements and has previously been described in detail . Briefly, 32 scans with a spectral resolution of 4 cm−1 were averaged for each pixel. Each pixel in an IR image therefore consists of a complete mid-IR spectrum. Image dimensions are typically 0.5 mm parallel and 0.2 mm perpendicular to the SC surface (~2500 spectra) requiring ~2 h for acquisition. IR imaging yields spectra with a spatial resolution of ~10 μm and high signal-to-noise ratios. Visible images were also acquired using the Spotlight 300 system.
Confocal Raman microscopy
Raman spectra were acquired with a Kaiser Optical Systems, Inc. Raman Microprobe (Ann Arbor, MI, USA). The instrument has been described in detail elsewhere [13, 14]. Briefly, a 785-nm diode laser generates ~8 mW of single-mode power focused using a 100× oil immersion objective to a volume of ~2 μm3 within the sample. The backscattered radiation illuminates a near-IR CCD with a spectral resolution of 4 cm−1 and wavenumber range of 100–3450 cm−1. Spectra were collected using 60-s exposure, three accumulations, and cosmic ray correction. Confocal maps were obtained in a point-by-point mode using a step size of 4 μm perpendicular (with depth into the skin) and parallel to the SC surface resulting in planar images. We typically generate confocal Raman images of ex vivo skin with dimensions 20–40 μm parallel and 80–100 μm perpendicular to the SC surface, requiring 10–20 h for acquisition.
IR and confocal Raman spectral images were processed using ISys 3.1 software (Malvern Instruments, LTD, UK). Both multivariate statistics (factor analysis) and univariate analysis (e.g. peak intensity ratios or frequency position) were used to highlight the spatial distribution of specific components and molecular structure in skin regions. Prior to generating the images, linear baselines were applied to spectral regions of interest.
Univariate measurements included the following. The Spectral Moments routine in ISys was applied to calculate the centre of mass of specific bands and generate images of peak position. Peak position can yield information pertaining to variations in physical or chemical environment, hydrogen bonding changes or molecular structure information. Semi-quantitative images of specific species in skin were generated using peak height ratio mapping or peak integration mapping. Peak height and area ratio images generated from the Raman spectra presented herein account for confocal Raman peak intensity loss with depth. We use the ring breathing mode of the amino acid, Phe, as an ‘internal standard’ for this purpose, thereby normalizing, in effect, to skin protein concentration.
The chemometric procedure of factor analysis was carried out to condense the information acquired in the hundreds to thousands of spectra into a smaller set of variables with minimal loss in content. Thus, the current goal of applying factor analysis is to characterize the spatial distribution of heterogeneity, whether chemical or structural, in the skin specimens studied. The ISys factor analysis algorithm, previously described in detail [11, 14], requires principal component analysis (PCA) scores and loadings for its implementation. This is followed by the application of the score segregation routine where transformations between the PCA loadings and Beer's Law parameters are sought so that the resultant factor loadings resemble typical spectra, although not of pure components. Correlations between the factor loadings and each pixel spectrum are represented by factor scores. The default score segregation parameter of 10 was used, and three to five factors were generated in the current analyses using an iterative technique with a loading difference error of <10−12.
FTIR imaging microspectroscopy
Typical single-pixel (6.25-μm2) raw IR spectra from three different skin regions obtained using the PE Spotlight IR imaging mode are overlaid in Fig. 1. Spectra were acquired from a human skin section microtomed perpendicular to the SC surface to ~7 μm thick. Immediately apparent is the high signal-to-noise ratio in the spectra where bands arising from predominantly lipid and protein skin constituents are evident. Bands of interest are marked and elaborated upon in Table 1. Changes in band position, shape and intensity can be observed in comparing the spectra of different skin regions. Subtle changes in particular peak positions, such as with the methylene symmetric stretching band, reveal structural information regarding the skin lipid or protein constituents. To illustrate the unique information available from such measurements, we have analysed IR images within spectral regions known to be sensitive to particular aspects of molecular structure. The origins of many of the spectral variations, due to compositional or structural alterations, are described below.
Table 1. Pertinent IR and Raman band assignments (from various references)
Vibrational mode (functional group)
IR frequency (cm−1)
Raman shift (cm−1)
Amide A, B
Out of range of current instrument
Fermi resonance between N-H stretch and overtone of Amide II, sensitive to secondary structure
CH3 asym stretch
Predominantly due to protein
CH3 sym stretch
CH2 asym stretch
Predominantly due to lipid, frequencies qualitatively monitor acyl chain conformational order and packing
CH2 sym stretch
Due to ester carbonyl, sensitive to hydrogen bonding
Predominantly due to C=O stretch, sensitive to secondary and tertiary structures
Sensitive to conjugation
Predominantly due to N-H in-plane bend and C-N stretch, sensitive to secondary structure
CH2, CH3 bend
Methylene modes in IR sensitive to acyl chain packing
COO− sym stretch
Due to NMF components and amino acid side chains
IR marker for Pro in collagen
Raman marker for trans acyl chain
Predominantly due to C-N stretch and N-H in-plane bend, sensitive to secondary structure
C-C stretch (skeletal)
Raman marker for trans acyl chain
Raman marker for disordered chains
Raman marker for DNA
C-C stretch aromatic ring
Raman marker for Phe
940, 925, 880, 860
Raman marker for collagen (Pro & HYP residues)
C-C stretch pyrimidine ring
Raman marker for cytosine
Raman marker for cholesterol
Factor analysis of an IR imaging data set truncated to the CH stretching region (2828–3000 cm−1) is highlighted in Fig. 2A and C. Figure 2B shows a visible micrograph of the same human skin section (unstained) used in the IR imaging experiment. The SC is visible at the top of the section as a dark, thin (15–20 μm) covering beneath which lies the viable epidermis (VE) and dermis. The visible image, acquired using the PE Spotlight instrument, is not intended to be of sufficiently high quality so as to define particular skin features, but rather serves to illustrate an advantage of IR imaging. Parallel sectioning is not necessary, and a single section can be used to generate multiple images of any IR spectral or statistical parameter. This single skin section was used to generate the images shown in Figs 2-6.
Figure 2A and C displays overlaid factor loadings and factor scores in the 2825–3000 cm−1 region, respectively. One score image is shown for each of the loadings, arranged in the same order, top to bottom. The method of colour coding for the score images is noted in the figure caption. IR skin spectra are rich in both methyl and methylene, symmetric and asymmetric, stretching bands (see Table 1) with substantial variation in the relative intensities of the bands observed in different skin regions. The methyl vibrations observed at ~2960 and 2870 cm−1 (asymmetric and symmetric, respectively) arise predominantly from protein, whereas the methylene stretching bands (~2920 and 2850 cm−1, respectively) are mostly due to the presence of lipid acyl chains. Thus, the SC region, which is known to be lipid rich, displays the highest factor scores (Fig. 2C, top score image) corresponding to the factor (Fig. 2A, black) with the highest methylene-to-methyl peak intensity ratio. Conversely, the dermis with its dense network of collagen fibres is highlighted with high scores (Fig. 2C, bottom image) for the factor (Fig. 2A, blue) with the highest methyl-to-methylene intensity ratio. The two remaining factors (red and green) display high scores in the viable epidermal and dermal regions, respectively. The compositional heterogeneity of skin tissue is highlighted by factor analysis in the CH stretching region.
Two univariate spectral parameters are imaged in Fig. 3 illustrating both the compositional variation and the lipid structural information inherent in the IR images. A simple peak height ratio (2852/2876 cm−1) image (Fig. 3A) reveals the relative lipid-to-protein ratio in the skin section with the SC possessing the highest ratio (2.4) and the dermis (0.2) the lowest. This is consistent with the factor analysis shown in Fig. 2. The second parameter images the spatial distribution of the lipid acyl chain conformational order in the tissue (Fig. 3B). CH2 stretching frequencies are well known to be sensitive to chain conformational order (trans-gauche isomerization) from the early seminal studies of Snyder and others [15, 16]. Significant variation in lipid conformational order is clearly displayed in the frequency image progressing from the SC where the chains are highly ordered (~2851 cm−1) to the dermis, where substantial disorder is found (>2854.5 cm−1). The accuracy of the peak picking routine is less reliable in particular regions of the dermis where the methylene symmetric stretching band is essentially non-existent (see Fig. 2A, blue factor). It should be noted that the range of frequencies is dependent on the peak picking method and algorithm used. For example, some centre of gravity algorithms use the top 15% (highest intensity region) of the peak to determine the maximum peak position and others allow the user to define the range over which to calculate the centre of mass. It is also common to peak pick inverted second-derivative spectra to eliminate baseline variation. The image in Fig. 3B was generated using baseline-corrected (3000–2828 cm−1) spectra, and the centre of mass of the peak was determined over the 2864–2840 cm−1 region. The thickness of the SC can also limit the accuracy of the peak picking routine as areas with very thin SC approach the spatial resolution (~10 μm) of the IR experiment.
The results from factor analysis in the Amide I and II spectral region (1720–1480 cm−1) are shown in Fig. 4. Again, the factor loadings are stacked in Fig. 4A, top to bottom, as their corresponding score images are stacked in Fig. 4B. The Amide I protein backbone mode arises predominantly from the amide bond carbonyl stretch (1690–1620 cm−1), whereas the Amide II protein backbone mode arises from both the N-H in-plane bend and C-N stretch (1560–1510 cm−1). Both vibrational modes are sensitive to protein secondary and tertiary structures. As such, analysis of this spectral region permits us to easily distinguish keratin-rich (SC and VE) from collagen-rich (dermis) areas (Fig. 4B, comparing the top two to the bottom image, respectively) in skin. Collagen is characterized by a major Amide I peak near 1660 cm−1 and a shoulder near 1636 cm−1, whereas keratin is characterized by a more symmetric single band at 1652 cm−1 (Fig. 4A, blue compared to red factor loading, respectively) . The factor (black) that maps most closely to the SC is broadened on the low-frequency side of the Amide I band compared to the red factor with high scores in the viable epidermis. This is almost certainly due to the presence of ceramides in the SC. Ceramides, which constitute ~50% of SC lipids, contain an amide moiety in their headgroup.
Factor analysis (Fig. 5) was also conducted over the low wavenumber (fingerprint) region of the IR spectrum (1480–988 cm−1). The resultant five factors appear to be grouped into two main classes (Fig. 5A) with two spatially segregated in the epidermis and three in the dermis. The factors reveal distinct spectral features unique to different chemical species and possibly structural elements, as suggested by their spatial distribution displayed in the score images (Fig. 5B). A composite profile is observed within the band arising from the methyl and methylene bending modes at ~1450 cm−1, likely reflecting differences in the lipid-to-protein ratio similar to that presented in Fig. 3A. Furthermore, the carboxylate symmetric stretching mode (~1400 cm−1) is relatively stronger in the keratin-rich regions, a result of the higher percentage of carboxylate-containing amino acids (Asp and Glu) in keratin compared to collagen. Additionally, the breakdown products of the protein filaggrin, known as natural moisturizing factor (NMF), are rich in carboxylate. Because filaggrin processing takes place within the SC and may be as much as 10% of corneocyte dry weight, this component is likely partially responsible for the high score (Fig. 5B) in the SC region observed in the top image. On the other hand, a peak unique to collagen arising from the CH2 wagging mode of the amino acid proline is observed at ~1337 cm−1 in the three factors that map to the dermis . Finally, one of the most prominent bands in this spectral region is the Amide III mode, 1235–1275 cm−1 (see Table 1), its complex profile differs significantly when comparing the factors in the two main classes. In an earlier publication, we noted that the spatial distribution of factor scores in this spectral region appeared to mimic the distribution of different keratin types in a cutaneous wound healing model . Two of the above-mentioned spectral parameters will be further explored using univariate measures.
Images of particular univariate parameters in the 1480–1200 cm−1 spectral region are presented in Fig. 6. The first image (Fig. 6A) displays the spatial distribution of the relative intensity of Amide III sub-band components, specifically the 1292/1240 cm−1 peak height ratio. High values for this ratio are observed in the SC and suprabasal region of the VE, whereas low values map to the dermis. There appears to be a layer of mid-range values (cyan) that are spatially correlated with the basal VE region. It is tempting to speculate as to whether the 1292/1240 cm−1 peak height ratio, and Amide III spectral region in general, is sensitive to different keratin types with a higher ratio identifying K1/10 (suprabasal VE and SC) and a lower ratio indicating the presence of K5/14 in the basal layer of the VE. Similar patterns among different sub-bands are evident in the Amide III region for the dermis (Fig. 5A), possibly indicative of different collagen types or the presence of elastin. Future experiments with model compounds will aid in elucidating these spectra–structure correlations. Finally, Fig. 6B simply displays an image of the integrated area of the 1337 cm−1 band illustrating its clear specificity to collagen in the dermis.
Confocal Raman microspectroscopy
Unprocessed, confocal Raman spectra of the SC, VE and dermal regions of intact ex vivo human skin are displayed in Fig. 7 with several bands of interest marked. The spectra are separated into three wavenumber regions, so that the signal-to-noise ratio of the relatively weaker bands in the 1225–800 cm−1 region can be easily visualized (note: different scale bars for each section). Raman scattering is inherently much weaker than IR absorbance; however, the technique offers other advantages for skin tissue studies. First, the spatial resolution of the current Raman instrument using the 100× microscope objective with excitation at 785 nm is better than IR (~2 compared to ~10 μm). Second, spectra are acquired in a confocal manner eliminating the need for physical sectioning of the sample as is done for IR imaging microspectroscopy. This allows for the acquisition of confocal Raman spectra of human skin in vivo, an approach initially pioneered by Puppels and colleagues who subsequently developed a commercial in vivo confocal Raman instrument [1, 20]. As discussed in the introduction, the in vivo application of confocal Raman spectroscopy is outside the focus of this paper. A general word of caution is necessary regarding all confocal Raman experiments as the absolute depth determination in the confocal Raman measurement has some uncertainty due to refractive index variations in tissue or any sample. Following from suggestions by Everall in several papers [21, 22], this effect has been evaluated with our current confocal Raman instrument using a polymer laminate with chemically distinct layers of known dimension [13, 14]. An error of ~10% in determining position in the z-axis was observed using the oil immersion objective for the laminate. Although the value will differ for skin because its optical properties vary from those of the laminate, our measured SC and VE thicknesses are well within the anticipated range. As a result, we do not adjust the depth values from those given by the computer-controlled microscope stage.
Our measurement and data reduction approach with confocal Raman presented herein is similar to that discussed for IR imaging. As such, Figs 8-12 illuminate particular aspects of skin composition and structure by applying factor analysis and univariate analysis approaches to particular spectral regions within a confocal Raman image of human skin. Many of the spectral features highlighted by factor analysis in the following figures may be observed in the appropriate single-pixel spectra shown in Fig. 7.
Confocal Raman spectra were acquired of an intact sample of ex vivo human skin using the protocol described in the Experimental section. Factor analysis of the CH stretching region (3036–2836 cm−1) was conducted, and the loadings (top to bottom) and respective score images (left to right) are displayed in Fig. 8A and B, respectively. The loadings highlight spectral differences found in this wavenumber region for the SC, VE and dermis. The top loading (black), displaying strong methylene stretching bands observed at ~2880 and 2850 cm−1 (asymmetric and symmetric, respectively), maps to the ~12-μm-thick SC region of the skin. The intensity ratio of the 2880/2850 cm−1 band is sensitive to lipid acyl chain packing with a higher ratio indicative of increased packing order . The factors that map to the VE and dermal regions are observed to have less intensity in the methylene bands compared to the methyl modes in the 2980–2930 cm−1 region (see Table 1), consistent with the IR results (see Fig. 2). The dermal region is indicated by the high scores in the bottom right-hand corner of the image on the right-hand side of Fig. 8B. Although the difference between the factors mapping to the VE compared to the dermis may seem minor, small-intensity shifts in this complex spectral region are sufficient for multivariate methods to distinguish the skin regions.
Factor analysis was also conducted over the 1730–1275 cm−1 region with the results displaying two factors specific to the SC and one to the VE region (Fig. 9). This assignment is confirmed by the presence of the small sharp band at ~1300 cm−1 (Fig. 9A) arising from the CH2 twisting mode of an all-trans acyl chain as is found in the ordered lipid lamellae of the SC . There is a broad underlying feature in the top factor (grey) from 1500–1300 cm−1 region due to the absorption band of glass. High scores for this factor map to the top pixel row in the left-hand image due to the presence of the glass cover slip and its close contact with the SC surface. Examination of the Amide I band profile comparing the top two to the bottom factor likely reflects compositional differences between the SC and VE due to ceramides, proteins and unsaturated or aromatic molecules. The CH2, CH3 bending modes are observed in the 1470–1430 cm−1 region and their relative intensity provides an estimate of the lipid/protein ratio, similar to the IR.
Figure 10 displays the results of factor analysis conducted over the 1144–995 cm−1 spectral region where the SC and VE regions are delineated from each other in the two score images on the left (Fig. 10B) corresponding to the top two factor loadings in Fig. 10A, respectively. Conversely, the image on the right displays high scores in somewhat more isolated pixels. Utilizing well-established spectra–structure correlations, spectral features in the factors can be assigned to specific cellular constituents in a way that is consistent with the spatial distribution of the high scores. Thus, the presence of the broad band at ~1090 cm−1 is tentatively assigned to the phosphodiester mode of DNA, possibly revealing the location of cell nuclei, and will be considered further when univariate analyses are presented (Fig. 12) [25, 26]. Common to all three factors is the sharp and intense band at ~1003 cm-1, which is due to Phe ring breathing mode. We commonly use this spectral feature arising from proteins as an internal standard to account for confocal loss of signal with depth. The remaining peak (1060 cm−1) and relative intensity variation (peak height ratio of 1130/1003 cm−1) differentiating the factor corresponding to the SC from that of the VE reflect the difference in lipid acyl chain conformational order between the two regions. Intensity in the 1060 and 1125 cm−1 bands arising from skeletal C-C stretching modes is a marker for all-trans acyl chains [27, 28] In contrast, a broad feature at ~1085 cm−1 is generally observed for gauche rotations (disordered) in acyl chains .
The final spectral region in which factor analysis was conducted (1014–800 cm−1) cleanly demarcates the SC, VE and dermal skin regions (Fig. 11A and B). Identification of the dermis in the lower right side of the rightmost score image, as suggested by factor analysis in the CH stretching region, is confirmed by the presence of the characteristic collagen doublet-of-doublets as shown in the blue factor in Fig. 11A . Only subtle spectral differences between the two factors (Fig. 11A, top two loadings) that map to high scores in the SC and VE are observed highlighting the power of factor analysis. Indeed, the most profound difference in these two factors is the relative intensity of the Phe band (1003 cm−1) compared to the series of bands between 975–800 cm−1 in the factor loadings.
Several univariate spectral parameters are imaged in Fig. 12 to demonstrate the consistency between the types of image analyses used herein. The integrated area of the Phe ring breathing mode is displayed in Fig. 12A illustrating essentially the confocal loss of signal with depth in skin. The other images in this figure have been constructed from the integrated area of particular bands ratioed to the Phe band area. Although the Phe content in the regions of skin analysed is not absolutely constant, we have empirically found through comparisons with factor analysis that the intensity of this band serves as a practical, reliable internal standard. It should be noted, however, that at depths greater than ~75 μm in ex vivo skin, the Phe signal has greatly diminished, making the area ratios less reliable. Figure 12B and C displays images of the spatial variation in lipid acyl chain conformational order by the means of two different band area ratios, that is, the CH2 twisting mode (~1300 cm−1) and the 1060 cm−1 skeletal C-C stretching mode ratioed to the Phe band, respectively. As anticipated and revealed in the images, the SC region displays the highest ratios and hence contains lipids with a high degree of acyl chain order. A second, smaller area (pocket of 2–3 pixels) with higher ratio values observed at a depth of 40–50 μm coincides with some of the high scores for the factor containing the broad phosphodiester band in Fig. 10 (bottom factor in (A), right-hand score image in (B)).The origin of the ordered lipid in this region is uncertain.
The image shown in Fig. 12D was constructed from the integrated band area ratio of the phosphodiester mode at ~1090 cm−1 to Phe. The spatial distribution of high scores in Fig. 12D is again somewhat similar to that shown for the ‘phosphodiester score image’ mentioned in the preceding paragraph. Although this is not unanticipated, the colour contrast is better in the score image (Fig. 10B, right-hand image) than in the area ratio image (Fig. 12D). To confirm the presence of DNA, which would reveal the location of cell nuclei, a second Raman band specific to DNA was sought. A somewhat broad, weak band, most likely arising from cytosine, was observed between 800–775 cm−1 (data not shown). An image of the integrated area ratio of the cytosine band over the Phe band is displayed in Fig. 12E, and the spatial distribution of area ratio values is quite similar to the images shown in Figs 12D and 10B, right-hand image.
It is clear from the FTIR and confocal Raman images presented in this general introduction to these methods that each measurement technique provides specific molecular and spatial information regarding the chemical composition, concentration and molecular organization of endogenous components in ex vivo skin samples. As illustrated throughout the above figures and data, the interplay between factor analysis and known spectra–structure correlations is extremely useful in identifying spectral regions of interest for further semi-quantitative univariate analysis. The design of more complex FTIR or confocal Raman spectroscopic imaging studies exploring ex vivo skin from specific anatomical sites, diverse skin donors, diseased or compromised skin, topically treated skin and many other variables can be built upon a foundation of measurements and methods such as those described above. It should be noted, however, that spectroscopic characterization of ex vivo skin beyond the molecular features discussed in this work is possible; thus, for example, images can be generated that chemically and structurally characterize tissue appendages or localized deposits of endogenous molecules within the skin. For example, we have detected and mapped cholesterol-rich regions within some skin samples (characterized by sharp bands at 700 and 605 cm−1 in the Raman spectra) which are sometimes co-localized with regions of ordered lipids. We have also imaged the molecular composition and organization of sebaceous glands in skin sections through hair follicles.
The possibilities for generating molecular histology images of ex vivo skin (either from sections or from intact skin) are nearly infinite and in significant part depend upon understanding the molecular information inherent in the spectra for skin. By design, many of the key references cited in this work are not recent FTIR imaging and confocal Raman studies, but refer to the preceding literature which developed the spectra–structure correlations that can be so powerfully exploited with these newer techniques.
We thank Professor Richard Mendelsohn of Rutgers University and, in particular, G. Yu and G. Mao for data acquisition.