Unlike FT-IR, Raman spectroscopy does not suffer fromwater interference as water is a very weak scatterer. ThereforeRaman measurements can be made directly from biofluidsand there are even reports of
measurements fromthe bladder and prostate,
However, the Raman effect is veryweak and typically only 1 in 10
photons undergo aninelastic light scattering event; therefore collection times tendto be relatively long. In addition, Raman spectra collected witha visible laser from biological samples can often be plaguedwith fluorescence which, whilst much broader than the sharpRaman peaks, can dominate the spectra and often need to beremoved mathematically. The Raman signal can be enhancedby one of two methods. The first is based on resonanceenhancement and this occurs because the laser wavelengthused to excite the Raman spectrum lies under an intenseelectronic absorption band of a chromophore. When thisoccurs an enhancement of Raman scattering is observed sothat some of the band intensities are increased by a factor of 10
. As this method uses chromophores, resonanceRaman spectroscopy can be directed to certain chemicalspecies; for example in the deep UV, 227 nm is used to enhanceselectively aromatic amino acids, whilst 244 nm causesenhancement predominantly from nucleic acids.
Thesecond method of enhancement is that of surface enhancedRaman spectroscopy (SERS). In SERS one relies on either theadsorption or close proximity of an analyte to a metal sub-strate. The substrate can take the form of a roughened metalsurface, a colloidal solution or a roughened electrode.
Theenhancement is in the order of 10
, and can be combinedwith a chromophore to effect surface enhanced resonanceRaman spectroscopy (SERRS).
Multivariate data analysis
The result of FT-IR and Raman spectroscopy is either aninfrared absorbance spectrum (wavenumber
relativeabsorbance) or a Raman profile (wavenumber shift
photoncount). These contain many overlapping bands and so datainterpretation can not be made by simple visual inspection andalternative approaches are needed. Thus it is essential thatthese vibrational spectroscopic measurements are combinedwith any of the rapidly growing arsenal of multivariateanalysis strategies, in what appears to be the exponentialphase of computational bioanalysis.
Multivariate data such as those generated from an infraredor Raman experiment consist of the results of observations of many different variables (wavenumbers or wavenumber shifts)for a number of individuals (objects;
diseased or healthysubjects). Each variable may be regarded as constituting adifferent dimension, such that if there are
variables (IR orRaman bands) each object may be said to reside at a uniqueposition in an abstract entity referred to as
-dimensionalhyperspace. This hyperspace is necessarily difficult to visualise,and the underlying theme of multivariate analysis (MVA) is thus
or dimensionality reduction. This dimensionalityreduction occurs in one of two ways; either using an unsuper-vised or supervised learning algorithms. Please see Fig. 3 for asummary of the main methods and the following reviews.
In general unsupervised methods such as principal compo-nent analysis (PCA) and hierarchical cluster analysis (HCA)are used to look at the ‘natural’ differences and similaritiesbetween these spectra. For most infrared and Raman spectrathe number of observations (samples) in the analysis is verysmall compared to the number of variables (wavenumbers orwavenumber shifts), which may be 100s–1000s. In this casemultivariate analysis by PCA and HCA is most appropriate.These methods are employed to discover structure in thesedata and can be used to ‘cluster’ samples into groups byproducing scatter plots (PCA) and dendrograms (tree-likefigures; HCA).
By contrast, supervised methods likediscriminant analysis (DA) and artificial neural networks(ANNs) are ‘calibrated’ with some known answer fromexisting knowledge of the sample. That is to say, one set of spectra were generated from the serum of people with cancer,
another set of spectra from healthy age-matched,gender-matched
, controls. This
knowledge is usedin the construction of the DA or ANN model, and musttherefore be correct else the model will not generalise well, andone is reminded of the adage ‘garbage in – garbage out’.
Thus one must suitably validate these methods.
Biological applications of vibrational spectroscopy
Within the biosciences, the applications of FT-IR have beennumerous and diverse. In the field of microbiology forexample, FT-IR has been used for the rapid and accurateidentification of bacteria to the sub-species level,
differentia-tion of clinically relevant species,
rapid enumerationof food spoilage bacteria
metabolicfootprinting of tryptophan-metabolism mutants
and discri-mination or identification of a range of bacterial genera,such as
However, during the last decade it has been recognised thatFT-IR, in combination with the appropriate multivariateanalysis strategies, has considerable potential as a metabolicfingerprinting tool for the rapid detection and diagnosis of disease or dysfunction, and indeed, a significant number of studies have been undertaken on tissues, cell and biofluids inan emergent area of research termed ‘infrared pathology’ byDiem and co-workers,
and now more commonly referred toas metabolic fingerprinting.
The remainder of this review willconcentrate on some of the many spectroscopic studiesundertaken thus far, related in the main to human disease,concerning analysis of cells, tissues and biofluids.Raman spectroscopy has charted similar areas, particularlywithin microbiology (for an excellent review see ref. 71). Theuse of resonance Raman spectroscopy in the deep UV was firstused by Nelson, Sperry and colleagues
in the early 1990sfor the characterization of bacteria. This approach has recentlybeen taken further with the development of suitable chemo-metric processing that have allowed for the identification of bacteria
and the effect of antibiotics on the bacterial cell.
Raman spectroscopy in the visible to mid-IR part of the EMspectrum has also been used for the characterization of microorganisms
including single cells.
Enhancementmethods based on surface enhanced Raman spectroscopy
This journal is
The Royal Society of Chemistry 2006