Raman and Fourier transform IR (FTIR) microspectroscopic images of biological material (tissue sections) contain detailed information about their chemical composition. The challenge lies in identifying changes in chemical composition, as well as locating and assigning these changes to different conditions (pathology, anatomy, environmental or genetic factors). Multivariate data analysis techniques are ideal for decrypting such information from the data. This protocol provides a user-friendly pipeline and graphical user interface (GUI) for data pre-processing and unmixing of pixel spectra into their contributing pure components by multivariate curve resolution–alternating least squares (MCR-ALS) analysis. The analysis considers the full spectral profile in order to identify the chemical compounds and to visualize their distribution across the sample to categorize chemically distinct areas. Results are rapidly achieved (usually <30–60 min per image), and they are easy to interpret and evaluate both in terms of chemistry and biology, making the method generally more powerful than principal component analysis (PCA) or heat maps of single-band intensities. In addition, chemical and biological evaluation of the results by means of reference matching and segmentation maps (based on k-means clustering) is possible.
This is a preview of subscription content, access via your institution
Open Access articles citing this article.
Scientific Reports Open Access 15 November 2021
Tracking mineral and geochemical characteristics of Holocene lake sediments: the case of Hotagen, west-central Sweden
Journal of Soils and Sediments Open Access 26 July 2021
Scientific Reports Open Access 13 April 2021
Subscribe to Journal
Get full journal access for 1 year
only $8.25 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Tax calculation will be finalised during checkout.
Get time limited or full article access on ReadCube.
All prices are NET prices.
Geladi, P., Grahn, H. & Burger, J. in Techniques and Applications of Hyperspectral Image Analysis (eds. Grahn, H.F. & Geladi, P.L.M.) (John Wiley & Sons, 2007).
Hall, H., Cheung, J. & Ellis, B. Immunoprofiling reveals unique cell-specific patterns of wall epitopes in the expanding Arabidopsis stem. Plant J. 74, 134–147 (2013).
Wilson, S. & Bacic, A. Preparation of plant cells for transmission electron microscopy to optimize immunogold labeling of carbohydrate and protein epitopes. Nat. Protoc. 7, 1716–1727 (2012).
Fabian, H. et al. Diagnosing benign and malignant lesions in breast tissue sections by using IR-microspectroscopy. Biochim. Biophys. Acta 1758, 874–882 (2006).
Nijssen, A. et al. Discriminating basal cell carcinoma from its surrounding tissue by Raman spectroscopy. J. Invest. Dermatol. 119, 64–69 (2002).
Sobottka, S., Geiger, K., Salzer, R., Schackert, G. & Krafft, C. Suitability of infrared spectroscopic imaging as an intraoperative tool in cerebral glioma surgery. Anal. Bioanal. Chem. 393, 187–195 (2009).
Gorzsás, A., Stenlund, H., Persson, P., Trygg, J. & Sundberg, B. Cell-specific chemotyping and multivariate imaging by combined FT-IR microspectroscopy and orthogonal projections to latent structures (OPLS) analysis reveals the chemical landscape of secondary xylem. Plant J. 66, 903–914 (2011).
Gierlinger, N. & Schwanninger, M. Chemical imaging of poplar wood cell walls by confocal Raman microscopy. Plant Physiol. 140, 1246–1254 (2006).
Chang, S.-S., Salmén, L., Olsson, A.-M. & Clair, B. Deposition and organisation of cell wall polymers during maturation of poplar tension wood by FTIR microspectroscopy. Planta 239, 243–254 (2013).
Pesquet, E. et al. Non-cell-autonomous postmortem lignification of tracheary elements in Zinnia elegans. The Plant Cell 25, 1314–1328 (2013).
Tsai, A. et al. Constitutive expression of a fungal glucuronoyl esterase in Arabidopsis reveals altered cell wall composition and structure. Plant Biotechnol. J. 10, 1077–1087 (2012).
Horvath, L. et al. Distribution of wood polymers within the cell wall of transgenic aspen imaged by Raman microscopy. Holzforschung 66, 717–725 (2012).
Schmidt, M. et al. Label-free in situ imaging of lignification in the cell wall of low lignin transgenic Populus trichocarpa. Planta 230, 589–597 (2009).
Gierlinger, N., Keplinger, T. & Harrington, M. Imaging of plant cell walls by confocal Raman microscopy. Nat. Protoc. 7, 1694–1708 (2012).
Richter, S., Müssig, J. & Gierlinger, N. Functional plant cell wall design revealed by the Raman imaging approach. Planta 233, 763–772 (2011).
Gierlinger, N. et al. Cellulose microfibril orientation of Picea abies and its variability at the micron-level determined by Raman imaging. J. Exp. Bot. 61, 587–595 (2010).
Gierlinger, N., Schwanninger, M., Reinecke, A. & Burgert, I. Molecular changes during tensile deformation of single wood fibers followed by Raman microscopy. Biomacromolecules 7, 2077–2081 (2006).
Naumann, A., Navarro-Gonzalez, M., Peddireddi, S., Kues, U. & Polle, A. Fourier transform infrared microscopy and imaging: detection of fungi in wood. Fungal Genet. Biol. 42, 829–835 (2005).
Wilson, R.H. et al. The mechanical properties and molecular dynamics of plant cell wall polysaccharides studied by Fourier-transform infrared spectroscopy. Plant Physiol. 124, 397–405 (2000).
Faix, O. Classification of lignins from different botanical origins by FT-IR spectroscopy. Holzforschung 45, 21–27 (1991).
Kataoka, Y. & Kondo, T. Quantitative analysis for the cellulose Iα crystalline phase in developing wood cell walls. Int. J. Biol. Macromol. 24, 37–41 (1999).
Akerholm, M., Hinterstoisser, B. & Salmen, L. Characterization of the crystalline structure of cellulose using static and dynamic FT-IR spectroscopy. Carbohydr. Res. 339, 569–578 (2004).
Wetzel, D. in Infrared and Raman Spectroscopic Imaging (eds. Salzer, R. & Siesler, H.W.) (Wiley-VCH, 2009).
Gorzsás, A. & Sundberg, B. Chemical fingerprinting of Arabidopsis using Fourier transform infrared (FT-IR) spectroscopic approaches. Methods Mol. Biol. 1062, 317–352 (2014).
Jirasek, A., Schulze, G., Yu, M.M.L., Blades, M.W. & Turner, R.F.B. Accuracy and precision of manual baseline determination. Appl. Spectrosc. 58, 1488–1499 (2004).
Eilers, P.H.C. Parametric time warping. Anal. Chem. 76, 404–411 (2004).
Savitzky, A. & Golay, M.J.E. Smoothing + differentiation of data by simplified least squares procedures. Anal. Chem. 36, 1627 (1964).
MathWorks. MATLAB http://www.mathworks.com/help/matlab/index.html (2013).
Stenlund, H., Gorzsás, A., Persson, P., Sundberg, B. & Trygg, J. Orthogonal projections to latent structures discriminant analysis modeling on in situ FT-IR spectral imaging of liver tissue for identifying sources of variability. Anal. Chem. 80, 6898–6906 (2008).
Baranska, M., Schulz, H., Rosch, P., Strehle, M.A. & Popp, J. Identification of secondary metabolites in medicinal and spice plants by NIR-FT-Raman microspectroscopic mapping. Analyst 129, 926–930 (2004).
de Juan, A., Maeder, M., Hancewicz, T., Duponchel, L. & Tauler, R. in Infrared and Raman Spectroscopic Imaging (eds. Salzer, R. & Siesler, H.W.) Ch. 2, 65–106 (Wiley-VCH, 2009).
Bonnier, F. & Byrne, H.J. Understanding the molecular information contained in principal component analysis of vibrational spectra of biological systems. Analyst 137, 322–332 (2012).
Tran, T.N., Wehrens, R. & Buydens, L.M.C. Clustering multispectral images: a tutorial. Chemometrics Intellig. Lab. Syst. 77, 3–17 (2005).
Piqueras, S., Duponchel, L., Tauler, R. & de Juan, A. Resolution and segmentation of hyperspectral biomedical images by multivariate curve resolution-alternating least squares. Anal. Chim. Acta 705, 182–192 (2011).
Nascimento, J.M.P. & Bioucas-Dias, J.M. Vertex component analysis: a fast alogrithm to unmix hyperspectral data. IEEE Trans. Geosci. Remote Sens. 43, 898–910 (2005).
Bioucas-Dias, J.M. et al. Hyperspectral unmixing overview: geometrical, statistical, and sparse regression-based approaches. IEEEE J. Stars 5, 354–379 (2012).
Krafft, C. et al. Crisp and soft multivariate methods visualize individual cell nuclei in Raman images of liver tissue sections. Vib. Spectrosc. 55, 90–100 (2011).
Piqueras, S., Burger, J., Tauler, R. & de Juan, A. Relevant aspects of quantification and sample heterogeneity in hyperspectral image resolution. Chemometrics Intellig. Lab. Syst. 117, 169–182 (2012).
Zhang, X. & Tauler, R. Application of multivariate curve resolution alternating least squares (MCR-ALS) to remote sensing hyperspectral imaging. Anal. Chim. Acta 762, 25–38 (2013).
Jaumot, J., Gargallo, R., de Juan, A. & Tauler, R. A graphical user-friendly interface for MCR-ALS: a new tool for multivariate curve resolution in MATLAB. Chemometrics Intellig. Lab. Syst. 76, 101–110 (2005).
de Juan, A., Rutan, S.C. & Tauler, R. in Comprehensive Chemometrics (eds. Brown, S., Tauler, R. & Walczak, R.) 325–344 (Elsevier B. V., 2009).
Tauler, R., Smilde, A. & Kowalski, B. Selectivity, local rank, 3-way data-analysis and ambiguity in multivariate curve resolution. J. Chemom. 9, 31–58 (1995).
Windig, W. & Guilment, J. Interactive self-modeling mixture analysis. Anal. Chem. 63, 1425–1432 (1991).
Windig, W. Spectral data files for self-modeling curve resolution with examples using the Simplisma approach. Chemometrics Intellig. Lab. Syst. 36, 3–16 (1997).
Batonneau, Y., Laureyns, J., Merlin, J.C. & Bremard, C. Self-modeling mixture analysis of Raman microspectrometric investigations of dust emitted by lead and zinc smelters. Anal. Chim. Acta 446, 23–37 (2001).
de Juan, A., Maeder, M., Hancewicz, T. & Tauler, R. Local rank analysis for exploratory spectroscopic image analysis. Fixed size image window-evolving factor analysis. Chemometrics Intellig. Lab. Syst. 77, 64–74 (2005).
de Juan, A., Maeder, M., Hancewicz, T. & Tauler, R. Use of local rank-based spatial information for resolution of spectroscopic images. J. Chemom. 22, 291–298 (2008).
Li, J.F., Hibbert, D.B., Fuller, S., Cattle, J. & Way, C.P. Comparison of spectra using a Bayesian approach. An argument using oil spills as an example. Anal. Chem. 77, 639–644 (2005).
Mark, H. & Workman, J. Chemometrics in Spectroscopy (Elsevier, 2007).
Linusson, A., Wold, S. & Norden, B. Fuzzy clustering of 627 alcohols, guided by a strategy for cluster analysis of chemical compounds for combinatorial chemistry. Chemometrics Intellig. Lab. Syst. 44, 213–227 (1998).
Plomion, C., Leprovost, G. & Stokes, A. Wood formation in trees. Plant Physiol. 127, 1513–1523 (2001).
Felten, J. & Sundberg, B. in Cellular Aspects of Wood Formation Plant Cell Monographs (ed. Fromm, J.) 203–224 (Springer, 2013).
Alanentalo, T. et al. Tomographic molecular imaging and 3D quantification within adult mouse organs. Nat. Methods 4, 31–33 (2007).
Alanentalo, T. et al. Quantification and three-dimensional imaging of the insulitis-induced destruction of beta cells in murine type 1 diabetes. Diabetes 59, 1756–1764 (2010).
Hornblad, A., Cheddad, A. & Ahlgren, U. An improved protocol for optical projection tomography imaging reveals lobular heterogeneities in pancreatic islet and beta cell mass distribution. Islets 3, 204–208 (2011).
We thank the Vibrational Spectroscopy Core Facility of the Chemical Biological Centre at Umeå University for full Raman and FTIR microspectroscopy instrumentation access and resources dedicated to method development. We thank B. Sundberg at the Umeå Plant Science Centre for initializing the project using hybrid aspen and continued support. U. Ahlgren and C. Nord at the Umeå Centre for Molecular Medicine are acknowledged for providing the mouse pancreas sample. A. de Juan and R. Tauler acknowledge financial support of the European Union project CHEMAGEB (Chemometrics and high-throughput omics analytical methods for assessment of global change effects on environmental and biological systems) and Joaquim Jaumot from the Spanish government (grant CTQ2012-11572). J. Felten and H. Hall acknowledge support by grants from the Swedish Research Council, Formas, Vinnova and Bio4Energy, the Swedish Programme for renewable energy.
The authors declare no competing financial interests.
Integrated supplementary information
The important steps are labeled with gray ellipsoids and numbers that correspond to the steps in the PROCEDURE section and match the ones of Figure 1
Supplementary Figure 2 Demonstrating the effects of area normalization, using a Raman hyperspectral image of wood fibers in the cross-section of a hybrid aspen stem as an example.
The panels show representative steps of the workflow, using non-normalized (left column) and area normalized (right column) data. Other pre-processing steps were applied uniformly (AsLS baseline correction, lambda = 30,000; p=0.001; Savitzky-Golay smoothing, order = 1, frame = 3)
a and b) Pure component (C1-C5) distribution maps (concentration profiles), following MCR-ALS (5 components, iteration limit: 50, convergence limit: 0.1). The white diamonds mark the location of the purest pixel for each component. These locations do not change, but the distribution maps are markedly different in intensity. The non-normalized maps (left) contain larger intensity fluctuations due to fluorescence (especially noticeable for C3 and C4)
c and d) The corresponding pure spectral profiles (C1-5). Notice that the spectral profiles are identical, i.e. the normalization had no effects on the spectral profiles (no distortions).
e and f) Segmentation maps, following k-means clustering with 4 clusters, based on the concentration profiles in a and b, respectively. Please note that k-means clustering arbitrarily assigns the cluster numbers, thus they do not match in e and f. This results in a coloring difference between the two plots. Nevertheless, the different zones are almost identical, except for a few borderline pixels (pixels 1, 12, 17 and 21).
g and h) The corresponding centroid profiles, showing the contribution of different components to each cluster. The effect of normalization is very obvious, with more absolute amounts in the non-normalized centroid plots, and more relative (proportional) amounts in the normalized centroid plots. When signal intensity is comparable (cluster 2 for the non-normalized data and cluster 4 for the normalized data), the component profiles are very similar. In other cases, the normalization “stretches” the contributions and discrepancies occur (c.f. cluster 1 in both cases).
Supplementary Figure 3 Comparing the results of band intensity heat maps (left column) and principal component analysis (PCA, right column), followed by k-means clustering, to MCR-ALS.
The same, non-normalized input data was used as in Supplementary Figure 2, with the same pre-processing, to allow direct comparison to MCR-ALS. Bands were chosen to represent lignin (1600 and 1331 cm-1), cellulose (1380 and 1095 cm-1) and pectin (856 cm-1), according to N. Gierlinger et al.14 PCA analysis was carried out using five principal components, which described 97.9% percent of the variation. The number of principal components were chosen to facilitate direct comparison to MCR-ALS (number of components) and to band intensities (number of bands).
a and b) Distribution maps based on band intensities and scores, respectively. Note that the maps based on band intensities can be correlated to certain cell wall zones (influenced by total intensity variations) and the bands assigned to the same compounds provide similar maps. On the other hand, maps based on scores (b) are harder to interpret, except for the first principal component (PC1).
c) An example spectrum (taken from pixel 7), marking the band positions used in the band intensity analysis. Colors match the ones used in the centroid profile (part g). Note the extreme low intensity of the 856 cm-1 band, assigned to pectin, in this pixel.
d) Loadings of principal components 1-5 (PC1-5). Note that these are abstract representations of the spectral profiles, and thus spectral band identification is difficult, except for PC1. In addition, all loadings (including that of PC1) contain negative values, which makes the interpretation of the centroid profiles (part h) and reference spectrum matching extremely difficult.
e and f) Segmentation maps, following k-means clustering with 4 clusters, based on the distribution maps in a and b, respectively. Note that the identified zones are matching very closely in both cases, and even when compared to the MCR-ALS results (Supplementary Figure 2e), with the exception of a few borderline pixels.
g and h) The corresponding centroid profiles, showing the contribution of different components to each cluster. Interpreting the results (i.e. identifying the contribution of any particular chemical compound, such as lignin or cellulose to any given cluster) is practically impossible in the case of PCA (h). Interpretation of the band intensity results (g) is considerably easier, assuming diagnostic bands (dark and light blue: lignin, red and green: cellulose, purple: pectin). While interpretation is easy, the results are erroneous. The pattern in the varying lignin amounts is reasonable, due to the fact that the 1600 cm-1 band (dark blue, originating from aromatic skeletal vibrations), is not shifting or overlapping and is very diagnostic. However, the contribution of the other lignin assigned peak (1331 cm-1, light blue) is much less significant, despite the fact that this is a very intense band (see c) and follows the same pattern as the 1600 cm-1 band (see a). Conversely, the contribution of the extremely small (virtually non-existent) pectin band (purple) is overestimated. Comparing this to the contribution of the red (“noise”) component in the MCR-ALS model in Supplementary Figure 2g reveals the strength of MCR-ALS in filtering off noise and using pure spectral profiles. Most critically, the estimation of cellulose contribution by band intensities results in errors. According to the centroid profiles, the amount of cellulose is practically constant in all cell wall layers (see green and red contributions in clusters 1-3). Thus falsely suggests that the lignin to cellulose ratio is (correctly predicted to be significantly different between the middle lamella and the g-layer) is only different because the lignin amount varies. In contrast, the results obtained by MCR-ALS (Supplementary Figure 2g) clearly show increased amounts of cellulose (green) in the g-layer (cluster 1) as compared to the middle lamella (cluster 2), as well as correctly predicting the dramatic differences in lignin amounts (blue) in these zones. This again underlines the importance of using full spectral profiles instead of single band intensities and illustrates how MCR-ALS combines the strengths of the two methods: full spectral profiles are used as in the case of PCA, while interpretation is as easy as in the case of band intensities.
Supplementary Figure 4 Demonstrating the importance of visual inspection of reference spectra matching results.
MCR-ALS resolved pure component spectra from Raman microspectroscopic images of cross-sections of hybrid aspen wood fibers are used as examples (see “good example” in Figure 6), with spectra of pure cellulose, lignin and D-glucoronic acid used as references (see the Materials section). a) The percentage matches based on Euclidean distances (dot products)
b-f) Area normalized spectra plots showing each component (thick black line), cellulose (blue), lignin (red) and D-glucoronic acid (green). As can be seen, all matches to D-glucoronic acid can be discarded despite the high percentage hits, since none its characteristic bands match the component spectra (most notably the band at 1750 cm-1, resulting from -C=O stretching, is absent). Component 1 matching lignin and component 2 matching cellulose can be confirmed as legitimate matches. Component 3 and 4 can have contributions from cellulose too, while component 5 is clearly an unresolved mixture, containing both lignin and cellulose.
The .zip archive contains the raw data files and the corresponding white light images for all the examples discussed in the paper, including the mouse pancreas section (064x064_MousePancreas_FTIR.mat, 064x064_MousePancreas_FTIR.jpg) and both plant sections (BadExample_Aspen_Raman.txt, BadExample_Aspen_Raman.jpg, and GoodExample_Aspen_Raman.txt, GoodExample_Aspen_Raman.jpg), as well as Raman reference spectra for cellulose, lignin and D-glucuronic acid (Cellulose.txt, Lignin.txt and Dglucuronicacid.txt, respectively) used in the reference matching of the plant examples. (ZIP 9061 kb)
Screenshot of the main window of the graphical user interface (GUI). (PDF 1261 kb)
Demonstrating the effects of area normalization, using a Raman hyperspectral image of wood fibers in the cross-section of a hybrid aspen stem as an example. (PDF 721 kb)
Comparing the results of band intensity heat maps (left column) and principal component analysis (PCA, right column), followed by k-means clustering, to MCR-ALS. (PDF 564 kb)
Demonstrating the importance of visual inspection of reference spectra matching results. MCR-ALS resolved pure component spectra from Raman microspectroscopic images of cross-sections of hybrid aspen wood fibers are used as examples (see “good example” in Figure 6), with spectra of pure cellulose, lignin and D-glucoronic acid used as references (see the Materials section). (PDF 478 kb)
About this article
Cite this article
Felten, J., Hall, H., Jaumot, J. et al. Vibrational spectroscopic image analysis of biological material using multivariate curve resolution–alternating least squares (MCR-ALS). Nat Protoc 10, 217–240 (2015). https://doi.org/10.1038/nprot.2015.008
This article is cited by
Investigating photocatalytic degradation of o-nitrophenol and p-nitrophenol over efficient CoO-Fe2O3@SiO2@TiO2 nanocomposite: rank annihilation factor analysis approach
Chemical Papers (2022)
Scientific Reports (2021)
Scientific Reports (2021)
Tracking mineral and geochemical characteristics of Holocene lake sediments: the case of Hotagen, west-central Sweden
Journal of Soils and Sediments (2021)
Quantitative visualization of subcellular lignocellulose revealing the mechanism of alkali pretreatment to promote methane production of rice straw
Biotechnology for Biofuels (2020)