Andres Reyes - Equipo 3 - Theory and Instrumentation) - Infrared Spectroscopy in Microbiology

Infrared Spectroscopy in Infrared (IR) spectra of intact microbial cells are highly
specific, fingerprint-like signatures that can be used to

Microbiology differentiate, identify, and classify diverse microbial species
and strains. Microbial IR spectra are also useful to (i)
detect in situ intracellular compounds or structures such
Peter Lasch and Dieter Naumann as inclusion bodies, storage materials, and endospores;
Robert Koch Institute, Berlin, Germany (ii) monitor and quantify metabolically released CO2
in response to various different substrates; and (iii)
characterize growth-dependent phenomena and cell–drug
1 Introduction and History 1 interactions. The characteristic information, useful for
microbial characterizations, is generally distributed over
1.1 Objectives 2
the entire IR region of the electromagnetic spectrum,
1.2 Advantages and Disadvantages of the
i.e. over the near-infrared (NIR), mid-infrared (MIR),
Method 2
and far-infrared (FIR). The spectral traits can be
1.3 Infrared Spectroscopy 3 extracted from the typically broad and complex spectral
1.4 Fourier-Transform Infrared Spectroscopy 3 contours applying resolution enhancement techniques,
2 Chemical Structures and Composition of Microbial difference spectroscopy, and pattern recognition methods
Cells 5 such as cluster analysis and artificial neural network
2.1 Bacterial Cells 5 (ANN) analyses. Additional applications arise by means
2.2 Yeast Cells 7 of specifically designed microscopes coupled to IR
3 The Main Biological Building Blocks, Assignment of spectrometers. IR spectra of microcolonies containing
IR-Bands 8 less than 103 cells can be obtained from colony replica
by a stamping technique that transfers microcolonies
4 Experimental Methodologies 11
growing on culture plates to a special IR sample
4.1 Infrared Measurement Techniques for Microbial holder. Using modern IR imaging approaches together
Samples 11 with mapping and video techniques, the fundamental
4.2 Sampling of Microbial Cells and Data tasks of microbiological analysis, namely detection and
Acquisition 13 differentiation of microorganisms, can be perspectively
4.3 Variability of Microbial Specimens and the integrated in one single apparatus.
Problem of Reproducibility 14
5 Data Treatment and Evaluation 15
5.1 Spectral Data Preprocessing 15 1 INTRODUCTION AND HISTORY
5.2 Unsupervised Classification Analysis 16
The 1980s and 1990s have witnessed the emergence
5.3 Supervised Classification Analysis 18
of sensitive, rapid, and increasingly precise physical
6 Applications 19 techniques for microbiological analysis. These new tech-
6.1 Characterization of Particular Cell Compounds niques range from mass spectroscopy (MS), molecular
and Structures 19 spectroscopy [including fluorescence, Fourier-transform
6.2 Differentiation, Classification, and Identification infrared (FTIR), and Raman spectroscopy], the applica-
of Microorganisms 22 tion of laser technologies, and flow cytometry to different
6.3 IR Microspectroscopy and IR Imaging of separation techniques such as gas chromatography and
Microbial Microcolonies 26 high-performance liquid chromatography.(1) This devel-
7 Perspectives 28 opment is paralleled by the dynamic advance of molecular
Acknowledgments 29 genetic techniques in microbiology, which presumably
will develop into extremely sensitive and specific tools for
Abbreviations and Acronyms 29
microbial characterizations in the future.(2)
Related Articles 29 In 1911, W.W. Coblentz was probably the first scientist
References 30 to suggest that biological materials can profitably be
analyzed by means of IR spectroscopy. The use of IR
spectroscopy as a means of differentiating and identifying
bacteria was extensively reported as early as in the
1950s and 1960s.(3,4) A critical review on this subject
Update based on the original article by Dieter Naumann, Encyclopedia of published in 1959 summarized that, although bacteria
Analytical Chemistry, © 2000, John Wiley & Sons, Ltd. exhibit IR spectra that are unique for individual strains,
Encyclopedia of Analytical Chemistry, Online © 2006–2015 John Wiley & Sons, Ltd.
This article is © 2015 John Wiley & Sons, Ltd.
This article was published in the Encyclopedia of Analytical Chemistry in 2015 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470027318.a0117.pub2
2 BIOMEDICAL SPECTROSCOPY
the identification of bacteria via IR techniques cannot selection of application examples, and literature quota-
be regarded as a practical scheme as it is a too time- tions mirror the author’s personal view of the present
consuming and impractical procedure.(3) Because of status in this dynamic field of research.
the limited specifications of IR spectrometers at that
time (sensitivity, reproducibility, and measurement time), 1.2 Advantages and Disadvantages of the Method
reports on IR applications to microorganisms became less
frequent in the 1960s and had virtually ceased by the mid- Among the most important aspects of IR spectroscopy for
1970s. The development of modern interferometric IR investigating intact cells of microorganisms, the following
spectroscopy, the availability of low-cost mini-computers, are noteworthy:
and powerful new algorithms of multivariate statistical Advantages:
analysis (MSA) and pattern recognition methodologies
such as hierarchical cluster analysis (HCA) and ANNs 1. The method is uniformly applicable to virtually all
have contributed to the revival of IR spectroscopy as a microorganisms that can be grown in culture. Biomass
means for characterizing microbial samples.(1) requirements can be scaled down to single colonies
The scientific basis for FTIR analyses of intact microor- taken directly from the culture plates. Using an
ganisms was developed between 1986 and 1992 by IR microscope, microcolonies as small as 10 μm
researchers at the Robert Koch Institute in Berlin in diameter, corresponding to a few hundred cells,
in collaboration with Bruker Optics and financially can be analyzed. Results are available within minutes
supported by the Ministry of Research and Tech- after obtaining adequate samples of the pure culture.
nology and the Ministry of Health of Germany.(5 – 7) 2. Detection, identification, and classification can be
The prototype of a dedicated FTIR system for microbial integrated in one single instrument when using an
characterizations was then tested by several microbial IR microscope. Results are available within one
laboratories within a joint pilot study between 1992 and working day in a clinical setting, including isolation,
1994.(8) On the basis of the progress achieved so far, cultivation, and identification.
a project funded by the European Union to develop 3. IR spectroscopy can classify microorganisms at very
sensitive FTIR and Raman microspectroscopic tech- different levels of taxonomic discrimination without
niques for the rapid characterization and identification of any preselection of strains by other taxonomic
microorganisms was conducted between 1998 and 2003. criteria. In contrast to most other techniques, IR
The outcome of this project was published in several spectroscopy is useful at the genus, species, and
papers.(9 – 12) strain levels. The specificity of the method is generally
A high-throughput FTIR method combined with extremely high allowing differentiations at the strain
an optimized microcultivation protocol suitable for and/or serogroup/serotype level.
rapidly characterizing microorganisms has been described 4. These advantages make the method applicable
recently,(13 – 16) which is very helpful for the high- to (i) very rapid identifications of life-threatening
throughput tracking of bacteria, yeast, fungi, and pathogens; (ii) epidemiological investigations,
mold contamination in the food-producing industries. conductance of case studies, screening of pathogens,
The methodology includes a protocol allowing liquid hygiene control, elucidation of infection chains,
cultivation of even molds and fungi in a high-throughput therapy control, and detection of recurrent infec-
way and an automated microtiter plate compatible tions; and (iii) characterization and screening of
way of handling liquid cultivation, sample preparation, microorganisms from the environment, biotech-
and subsequent FTIR measurement of the microbial nological process control, microbiological quality
samples. Currently, a fully automated prototype for control and contamination source tracking in the
robust and routine contamination source tracking in the food and pharmaceutical industries, and maintenance
food industry is under development.(15) of strain collections.
5. In a number of cases, in situ detection of specific
cell components is possible (e.g. storage materials,
1.1 Objectives
spore formation, and encapsulation of microorgan-
This article describes details on technological develop- isms). Furthermore, drug resistance and cell–drug
ments, data elaboration methodologies, and application interaction can be monitored and characterized.
examples in this field. Only results that are concerned with
samples of intact cells will be presented. No attempt will Notable disadvantages:
be made to comment on the use of IR spectroscopy for
the analysis of isolated cell fragments, compounds, macro- 1. Only microorganisms that can be grown in culture
molecules, or cell metabolites. Focal points of this article, and are available as pure cultures can be analyzed.
DOI: 10.1002/9780470027318.a0117.pub2
INFRARED SPECTROSCOPY IN MICROBIOLOGY 3
Mixed cultures can only be investigated with the help hence, one deals with the convenient values of 14 300 cm−1
of an IR microscope provided single colony growth is (∼700 nm) to 100 cm−1 (100 μm) (Figure 1) throughout
obtained. the IR region. Also per convention, the IR region of
2. Classification and identification of microorganisms is the electromagnetic spectrum is subdivided into the NIR,
based on the analysis of spectral fingerprints that, MIR, and FIR regions (Figure 1). An IR spectrum of
in some way, ‘express’ all cell constituents. Specific a sample is obtained by scanning the intensity of IR
information on one or only a few specific compounds radiation before and after passage of the IR beam through
present in the cells is generally not available. the sample. The IR spectrum is displayed by plotting the
3. Thus, stable results can only be obtained when quantity T = IS /IR as a function of wavenumbers ν, where
the microbiological parameters (culture medium, T is the transmittance, IS the intensity of the IR beam
cultivation time, and temperature) can be rigorously after, and IR before passing through the sample. In most
controlled. Spectral databases that are used in cases, the absorbance A is used (A = −log10 T, see
different laboratories require the use of appropriate Figure 2b), as the absorbance at a given wavelength is
standardized culture media available on the market. directly proportional to the concentration of a sample
4. Acceptance of the IR technique by the microbiolog- according to Beer’s law. The IR spectra of most materials
ical community presupposes that spectral information consist of a large number of absorption bands. These
can be transformed into classification schemes that bands originate from the interaction (energy exchange)
are compatible with those obtained by conventional between discrete light quanta and mechanical motions
chemotaxonomic or molecular genetic techniques. (vibrational and rotational modes) of the molecules that
This task can only be fulfilled by trained people in are excited by the absorption of IR radiation. As the
expert laboratories. constituents of a typical biological sample are present
5. While the strength of the technique is its ability to in a condensed phase (solids, liquids, or solutions), only
differentiate microorganisms very rapidly below the vibrational modes and no rotational modes are observed.
species level, classifications at the genus level cannot Consequently, IR spectra of biological specimens are only
be expected to be taxonomically relevant in all cases. vibrational spectra.(17)
1.3 Infrared Spectroscopy

1.4 Fourier-Transform Infrared Spectroscopy
The IR region of the spectrum extends from the
visible region until it overlaps the terahertz, or very With the advent of FTIR spectroscopy,(18) major
short microwave, range at wavelengths below 100 μm drawbacks of classical dispersive IR spectroscopy could
(Figure 1). The basic characteristic of this region is be circumvented. Briefly, FTIR no longer measures
that the radiation originates from thermal emission of one wavelength after the other. This means that
a hot source. Per convention throughout the IR region, FTIR spectrometers are not working like dispersive IR
the ‘wavenumber’, that is the number of waves per spectrometers that are equipped with prism and/or grating
centimeter, is used to characterize the radiation. The monochromators where the light energy emanated from
wavenumber unit is the reciprocal centimeter (cm−1 ) and, the IR source is strongly limited by entrance and exit
Wavelength 104 102 100 10−2 10−4 10−6 10−8

cm
λ
1m 1 mm 1 μm 1 nm
Wavenumber 1 10 100 1000 10000 cm−1

∼
ν
Frequency 108 1010 1012 1014 1016 1018 s−1

ƒ
Terahertz Visible
IR
Spectral Radio- Short- Micro- γ
UV X-rays
range waves waves waves rays
MIR
NIR
FIR
Figure 1 The electromagnetic spectrum.
DOI: 10.1002/9780470027318.a0117.pub2
IR source Interferometer Sample Detector PC

(a)
Interferogram, single-channel background Interferogram single-channel sample

Intensity
Intensity
FT FT
0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700
(b) Δx Δx
~ ~
Single-channel background spectrum IR (ν) Single-channel sample spectrum IS (ν)
Intensity
Intensity
4000 3000 2000 1000 4000 3000 2000 1000

−1)
Wavenumber (cm Wavenumber (cm−1)
~
IS (ν)
~ = − log
A (ν) 10 ~
IR (ν)
~
Spectrum A (ν)
Intensity
4000 3000 2000 1000

Wavenumber (cm−1)
Figure 2 The FT-IR spectrometer. (a) Block scheme of the basic components of an FTIR spectrometer. (b) Working principle
of FTIR spectroscopy: First a single-channel interferogram of the optical background is measured (top left panel), then a single-
channel interferogram with the sample in the beam path is measured (top right panel). From these two interferograms, the
corresponding single-channel background and single-channel sample spectra are calculated (see middle left and right panels). These
two single-channel spectra are used to calculate the final absorbance spectrum (see bottom panel) according to the given formula.
DOI: 10.1002/9780470027318.a0117.pub2
slits. FTIR, instead, applies interferometric modulation can be co-added and signal averaged to reduce the
of radiation. As in dispersive instruments, a hot white- random noise. Thus, very tiny signals may be extracted
light source, a so-called Globar, is used in the MIR. by decreasing the noise to a satisfactory level.
A widespread type of interferometer used in FTIR The Connes advantage The means by which the moving
is the so-called Michelson interferometer equipped mirror position is monitored involves the use of an
with a KBr beam splitter coated with a germanium internal HeNe reference laser. This laser, at the same
film to split the collimated IR beam into two parts. time, is used as an internal wavelength calibration
For rapid-scanning interferometers, liquid nitrogen- standard. Therefore, the wavelength accuracy of an
cooled mercury-cadmium-telluride (MCT) detectors are FTIR spectrometer is specified to lie below 0.01
used. For slower scanning types of interferometers, cm−1 , which is particularly useful when absorbance
less sensitive pyroelectric detectors, e.g. a deuterated subtraction and resolution enhancement techniques
triglycine sulfate (DTGS) detector element, are suitable. are employed or when spectra are stored for data
The moving mirror of the interferometer yields a comparison in spectral libraries to identify unknown
sinusoidal signal at the detector for each modulated compounds.
frequency. Thus, for a white-light source, all modulated The stray light advantage Because the interferometer
frequencies superimpose to a complex interferogram. modulates each IR frequency differently, there is no
This interferogram is amplified and digitized by the A/D equivalence to stray light in FTIR. This means that the
converter, computer stored, and finally transformed into absorbance values remain very linear even beyond two
a spectrum by fast Fourier-transform (FT) techniques. absorbance units in case of DTGS detectors.
Thus, the auxiliary computer in case of FTIR is a The data handling advantage As FTIR implies the use of
virtual necessity.(18) The interferometer can be thought highly efficient computers controlling data acquisition,
of as a means of encoding the signal as a time- spectra can now be easily stored on electronic
dependent signal instead of a frequency- or wavelength- devices for further computer-based data analyses
dependent signal. The most important feature of this (preprocessing and multivariate classification).
interferometric technique is that every individual data
point of an interferogram contains information over
the entire spectral region measured. In essence, the 2 CHEMICAL STRUCTURES AND
detector is always observing all spectral frequencies at COMPOSITION OF MICROBIAL CELLS
the same time. The Fourier transformation is simply a
mathematical means of converting the signals back from Understanding of IR spectra of microbial cells requires
the time domain into the frequency domain for better some general perception of its composition, major
interpretability. In contrast to classical dispersive IR cell types, and chemical structures present. At the
spectrometers, FTIR spectrometers are single-channel simplest level, all biological systems are composed
instruments (Figure 2a). The resulting absorbance of water, lipids, proteins, and carbohydrates. The
spectrum is obtained by two consecutive single-channel gross composition of bacterial (prokaryotic), yeast, and
measurements, respectively, with and without the sample mammalian (eukaryotic) cells is given in Table 1.
in the optical beam path (Figure 2).
This, along with some other features of the interferom-
2.1 Bacterial Cells
eter, leads to the well-known advantages of FTIR over
dispersive IR forms of spectroscopy(18) : In contrast to animal, plant, and yeast or fungal cells
(the so-called eukaryotic organisms), bacteria (also
The Jaquinot advantage There are no longer slits in called prokaryotes) exist in only a limited number
the interferometers that limit the amount of energy of morphological forms (e.g. rods, cocci, chains, and
getting to the detector. The resolution of the FTIR spirals). Their chemical composition and structures,
spectrum is solely defined by the path length of the however, vary considerably. The cytoplasmic structures
moving mirror away from the so-called zero point of bacteria are less organized (compartmentalized) and
of the interferometer. This advantage is particularly are simpler than those of yeasts and fungi but have
useful in cases where a small sample limits the optical complex and very diverse molecular and supramolecular
throughput. structures outside the cytoplasmic membrane. These
The Felgett advantage Because the detector is observing include the cell wall, outer membrane, capsules, and
all frequencies simultaneously, the signal is said to be sometimes specific layers such as the S-layers. Some
‘multiplexed’. One complete scan of the moving mirror bacteria are capable of sporulation or storage material
of the interferometer takes only a fraction of the time production. Many structural differences providing the
that a ‘dispersive scan’ would need. Repetitive scans possibility of differentiation between bacteria by IR
DOI: 10.1002/9780470027318.a0117.pub2
Table 1 Gross composition of pro- and eukaryotic cells

Bacteriaa Yeastsa Animal Cellsa
Radius (μm) ∼1 ∼10 10–100
Volume (μm3 ) ∼1 ∼1000 >10 000
Surface/volume (μm−1 ) ∼1 ∼0.1 Tissue
Generation time (h) 0.2–10 2–10 ∼20
DNA base pairs ∼4 × 106 ∼20 × 106 500–5000 × 106
Number of ‘genes’ ∼4000 ∼20 000 >50 000
Size of ribosome 70 S 80 S 80 S
RNA % (w/w)b 5–15 3–10 10–20
DNA % (w/w)b 2–4 1–3 ∼5
Proteins % (w/w)b 40–60 40-50 ∼60
Lipids % (w/w)b 10–15 5–20 10–15
Carbohydrates % (w/w)b 10–20 10–25 6–8
a All numbers are rough estimations which depend on organism/organ, growth conditions, cell division cycle, and other factors.
b Content in % of cell dry weight
spectroscopy reside in the cell envelope, which is generally type of amphiphilic molecule, the lipopolysaccharides
defined as the cytoplasmic membrane plus the cell (LPSs) and the various pore-forming proteins, the porins.
wall.(19) The structure of LPS contains three basic regions: the
Most cell envelopes fall in two categories: the so- so-called O-specific side chain (a hetero-oligosaccharide,
called Gram-positive bacteria consisting only of the responsible for O-antigenicity), the inner and outer
cytoplasm, the cytoplasmic membrane, and the cell core-regions, and a lipid anchor, called lipid A. This outer
wall and the more complex Gram-negative bacteria membrane is the major permeability barrier that protects
containing an outer membrane in addition to the cell these cells against bile salts, degradation by digesting
envelope. Some bacterial species, the mycoplasms, lack enzymes, and prevents the cells from hydrophobic drugs
any cell wall at all, express however a rather rigid plasma and antibiotics to diffusing through this particular layer.
membrane. The mycobacteria, nocardia, corynebacteria, and some
The bacterial cell wall is a rigid, high-molecular related groups have very unusual cell envelopes that
network made up primarily of peptidoglycan, which form thick, wax-like layers around the outside of the
has a shape-giving function and protects the cells cell wall. Major compounds present in this impermeable
from osmotic disruption. Its primary structure consists
and rigid layer are complex, long-chain fatty acids, the
basically of disaccharide–pentapeptide subunits with
mycolic acids.(19,20)
unusual features such as the occurrence of alternating
Some bacteria form capsules (sometimes referred to
D- and L-amino acids and a γ-bonded D-glutamic
also as ‘slime-layers’) surrounding the cell envelope.
acid residue. Its structural variants are found to be
These are not essential structures and are frequently
different for various groups of bacteria. In addition, a
built up of negatively charged polysaccharide compounds.
lipoprotein has been found to be covalently bound to
the peptidoglycan of Gram-negative organisms. Many Some bacilli exhibit capsules composed of negatively
Gram-positive bacteria contain an additional polymer, charged homo-oligopolypeptides such as poly-D-glutamic
covalently bound to peptidoglycan, the so-called teichoic acids.
and teichuronic acids.(19) The teichoic acids are ribitol Many bacilli and clostridia may form endospores
or glycerol containing macromolecules, built up by that are modified cell structures that can survive
a phosphate-carrying backbone with side chains of under unfavorable environmental conditions. These
variable composition. Teichuronic acids and neutral endospores have two membrane-like layers. Between
polysaccharides are also sometimes found in the Gram- these two layers, a spore-specific peptidoglycan (the so-
positive cell wall. called cortex) is found which differs in primary and
Gram-negative bacteria exhibit an additional three-dimensional structure from the peptidoglycan of
membrane, the so-called outer membrane. This outer vegetative cells in that the muramic acid of peptidoglycan
membrane is an asymmetric membrane, the inner leaflet is modified to a lactam derivative and is less cross-linked.
of which contains only phospholipids with nearly the same A keratin coat is located on the cell exterior and it has
composition as found in the cytoplasmic membrane. The been established that large quantities of Ca+2 -dipicolinate
outer leaflet contains nearly exclusively one particular are related to heat resistance of endospores.(19)
DOI: 10.1002/9780470027318.a0117.pub2
2.2 Yeast Cells

The proteins, lipids, and polysaccharides that make up 2922
the membrane, cell wall, and capsules of capsulated yeasts
have a significant impact on the systematics and phylogeny 2852 1738 1091
of yeasts. Compared to bacteria, only a small number of 1468
1238
species of yeasts has been investigated in depth and even
fewer studies have focused on structural details of these DMPC
constituents.
The storage compounds in yeasts have been 1088
reviewed.(21) The principal, readily mobilizable reserve 1238
1715
polysaccharide in yeasts is glycogen (see also FTIR spec-
trum of glycogen in Figure 3). Glycogen occurs in yeast
cells in both the cytoplasm and the nucleoplasm either
in soluble form or as aggregates of spherical particles ct-DNA
having a diameter of 40–50 nm. The glycogen content of
Absorbance
yeast cells is highly dependent on the physiological state
1656
and may reach up to 20% of the dry weight of the cells.
1546
Glycogen may have a molecular weight up to 107 Da; it
contains three types of α-(1→4) linked chains of glycosyl 1400
units: short A-chains (side chains) attached to B-chains
(main chains) by α-(1→6) glycosidic linkages. The main Collag
chains carrying one or more side chains are attached by
α-(1→6)-glycosidic links. 1026
The cell walls of yeasts are composed of complex
polysaccharides and glycoproteins.(21,22) The principal
1082
low-molecular building blocks of these polysaccharides
are glucose and mannose followed by galactose, xylose, 1153
N-acetyl-D-glucosamine, and uronic acids. While the qual-
1417
itative composition of these compounds is a taxonomic 1362
marker, the quantitative composition varies with culti-
vation conditions. Major structures of yeast cell walls Glyc
are β-glucans in which the glycosyl units are mutually
linked by β-(1→3), β-(1→6), and possibly β-(1→2) glyco- 3000 2500 2000 1500 1000
sidic bonds. These structures are found with different
Wavenumber (cm−1)
molecular weights and branching and may form microfib-
rillar structures of crystalline nature.(22) Chitin, a linear
Figure 3 MIR spectra from examples of the main biological
β-(1→4) polymer of N-acetylglucosamine is a typical building blocks: Lipids, nucleic acids, proteins, and carbohy-
constituent of primary septa and budding yeast. Chitosan, drates. DMPC, dimyristoylphosphatidylcholin; ct-DNA, calf
a β-(1→4)-linked polymer of D-glucosamine may be thymus desoxyribonucleic acid; Collag, collagen type I; and
considered as a minor yeast polysaccharide but in some Glyc, glycogen. Technique: absorbance/transmission; nominal
physical resolution: 4 cm−1 ; number of scans: 128; detector:
dimorphic fungi belonging to the group of Zygometes; it DTGS; and spectrometer type: IFS 66.
may represent one of the principal wall components.
Mannans exist in yeasts as covalent complexes with
proteins. The yeast mannoproteins are large molecules In some yeasts, particularly of the genus Cryptococcus,
(molecular weight up to 500000 Da) consisting of a the polymers containing D-glucoronic acid residues are
covalently linked carbohydrate and a protein. The important constituents of extracellular capsules. For
polysaccharide portion contains up to 150 mannosyl units, example, the capsule of Cryptococcus laurentii is made
being connected via N-glycosidically linked polymannose up of an α-(1→3) linked mannose backbone with xylosyl
units. A second group of carbohydrate of yeast mannans and glucuronosyl residues as side groups.(22)
are short manno-oligosaccharides, O-glycosidically linked Yeast membranes contain a number of lipids
to serine and/or threonine residues of the polypeptide and pigments that are not present in prokaryotic
(protein).(21,22) cells (bacteria).(23) These are sterols, sphingolipids,
DOI: 10.1002/9780470027318.a0117.pub2
ergosterins, melanins, and some glycolipids. Culture the fundamental vibrational modes and – in the group
conditions have a marked influence on the total lipid frequency notion – can often be assigned to particular
content and lipid composition of yeasts. Factors control- functional groups. At wavenumbers lower than 1400
ling lipid content and composition are pH of the medium, cm−1 , IR bands tend to arise no longer from localized
temperature, and time of growth, and the ratio of N- vibrational modes but rather from skeletal and strongly
and C-sources. Sterols occur both in free form and as coupled modes, which are difficult to describe. For prac-
esters with long-chain fatty acids. Both forms are inter- tical purposes, rough band assignments can be obtained
convertible. Free sterols are associated with membrane from group frequency charts published in several bibli-
functions; sterol esters may fulfill a storage or ‘pool’ func- ographies that may be helpful to obtain correlations
tion. Common sterol molecules of yeasts are ergesterol, between partial structures and band frequencies. Several
lanosterol, episterol, zymosterol, and fecosterol. Major spectra descriptions and excellent structure–spectra
structures of sphingolipids found in yeasts are the sphin- correlations for the most important biological macro-
gosines, cerebrins (ceramides), sphingomyelines, and molecules can be obtained from the literature.(24) Efforts
cerebrosides. A typical membrane lipid in yeast is ergos- to interpret the IR-spectra of biological molecules are
terin. Its structure is similar to cholesterin, and it belongs mainly based on the analysis of known structures, normal
to the group of sterines. Further compounds frequently coordinate analysis, and isotope exchange experiments.
found in the membranes of yeasts are melanins that are Nucleic acids, proteins, lipids, and carbohydrates are
black pigments built up from tyrosine derivates.(19) constantly present in different amounts and large diver-
sity in microbial cells. Figure 3 shows the MIR spectra
of a typical phospholipid, DNA, protein, and carbohy-
3 THE MAIN BIOLOGICAL BUILDING drate structure. Some of the IR bands are numbered and
BLOCKS, ASSIGNMENT OF IR-BANDS tentative assignments can be taken from Table 2. Figure 4
shows IR spectra of fully hydrated samples of intact cells
IR absorption bands observed in the MIR region between of Escherichia coli that span the MIR and NIR range
approximately 800 and 4000 cm−1 mainly arise from from 1000 to 8000 wavenumbers. The MIR spectrum in
Table 2 Tentative assignment of some bands frequently found in microbial IR spectra (peak frequencies have obtained
from the second derivative spectra); asym = asymmetric; sym = symmetric; str = stretching; def = deformation
Band numbering Frequency (cm−1 ) Assignment
(cf. Figure 5)
∼3500 O–H str of hydroxyl groups
∼3300 N–H str (amide A) of proteins
3010 C–H str of C=CHX (methene)
1 2959 C–H str (asym) of –CH3
2934 C–H str (asym) of >CH2
2 2921 C–H str (asym) of >CH2 in fatty acids
3 2872 C-H str (sym) of -CH3
4 2852 C–H str (sym) of >CH2 in fatty acids
5 1741 >C=O str of esters
1715 >C=O str of esters, RNA/DNA
1695–1675 Amide I band components resulting from
antiparallel pleated sheets and β-turns of proteins
6 ∼1655 Amide I of α-helical structures
7 ∼1637 Amide I of β-pleated sheet structures
8 1548 Amide II
9 1515 ‘Tyrosine’ band
10 1468 C–H def of >CH2 (scissoring)
11 ∼1400 C=O str (sym) of COO−
1379 C-H def of -CH3 (umbrella)
1240–1310 Amide III band components of proteins
12 1220–1250 P=O str (asym) of >PO2 − phosphodiesters
13 900–1200 C–O–C, C–O dominated by ring vibrations of carbohydrates C–O–P, P–O–P
13 1085 P=O str (sym) of >PO2 −
968 C–O phophodiester residue (DNA)
720 C–H rocking of >CH2
600–900 ‘True’ fingerprint region
DOI: 10.1002/9780470027318.a0117.pub2
Fundamental vibrations absorbance spectra, respectively. Both the MIR and the
Stretching Deformation NIR regions provide a large number of spectral traits
that may potentially be useful to characterize complex
OH CH OH CH
biological materials such as microorganisms. Some rough
NH NH assignments in respect to OH, NH, and CH functional
groups are shown at the top of the figure.
Figure 5 (black curve) shows an MIR spectrum of a
typical microbial sample dehydrated to a thin film layer
on a ZnSe optical plate. While exact assignments to
Absorbance
specific structures are still a too complex task at present,

many spectral features of intact cells can be visualized
applying resolution enhancement techniques. In general,
50–70 spectral features are resolved by derivative or
Fourier self-deconvolution techniques (Figure 5, red
curve) and are the basis of assignments to functional
groups, known chemical compounds, partial structures,
or different conformational states of particular molecules
4000 3000 2000 1000
present. These spectral features constitute a spectral
(a) Wavenumber (cm−1) fingerprint useful to differentiate between different types
or states of cells.
Overtones Combinations
IR spectra of microbial specimens provide not
only a number of absorption bands that describe
OH CH OH CH
molecular composition of the cells. Many of these
NH NH bands are also sensitive to structural changes, various
intra- and intermolecular interactions including H-
bonding pattern, membrane constitution, lipid–protein
interaction, and conformational states such as different
secondary structures of proteins. The physical state of the
sample such as hydration or aggregation state, interaction
Absorbance
with ions, etc., has a strong influence on results. These

7500 7000 6500 6000 5500 4550 facts necessitate the rigorous standardization of sampling,
preparation, and data acquisition procedures.(5 – 8)
Tentative band assignments based on the systematic
comparison of resolution-enhanced microbial IR spectra
with those of the known building blocks constantly
present in intact cells can be derived from Table 2 (see
8000 7000 6000 5000 4000
also spectra of Figures 3 and 4 for comparison).
(b) Wavenumber (cm−1)
1. The region between 4000 and 3100 cm−1 is dominated

Figure 4 Infrared absorbance spectra of a fully hydrated
sample of Escherichia coli (strain RKI/A 139) in the spectral by rather broad spectral features resulting from –OH
range between 1000 and 8000 cm−1 (a) MIR; (b) NIR. The (∼3400 cm−1 ) and N–H stretching modes (amide A
insets show the second derivatives as calculated from the original ∼3300 cm−1 and amide B ∼3030 cm−1 ).
absorbance spectra. Physical resolution applied was 6 cm−1 for 2. The region between 3100 and 2800 cm−1 exhibits
the MIR and 16 cm−1 for the NIR measurements, respectively. the C–H stretching vibrations of –CH3 and >CH2
Technique: A/T; cuvettes: CaF2 cell with 8 μm (MIR) and functional groups and, hence, is generally dominated
250 μm (NIR) optical pathlength, respectively; number of
scans: 64 (MIR, DTGS-detector), 512 (NIR, InSb-detector);
by the spectral characteristics of fatty acid chains of
and spectrometer type: IFS 66. the various membrane amphiphiles (e.g. phospho-
lipids) and by some amino acid side-chain vibrations.
Figure 4(a) was obtained on a hydrated film sample with Complementary information can be deduced from
a thickness of about 8 μm, whereas Figure 4(b) shows the the region between 1470 and 1350 cm−1 , where the
NIR-spectrum recorded on a sample with a film thick- various deformation modes of these functional groups
ness of about 250 μm. In both cases, CaF2 was used as are found. In rare cases, a weak band near 3015 cm−1
the optical cuvette material. The insets of the two figures is also observed, resulting from =C–H double bond
give the second derivatives as calculated from the original stretching modes of unsaturated fatty acid chains.
DOI: 10.1002/9780470027318.a0117.pub2
5 11
12 13
10
9
Absorbance 3 7
8
1
2 4
3100 3000 2900 2800 2700 1800 1600 1400 1200 1000 800
Wavenumber (cm−1) Wavenumber (cm−1)
Figure 5 (a and b) Tentative band assignment of some bands frequently observed in bacterial MIR-spectra (Staphylococcus aureus
SG 511). Black curve: Original absorbance spectrum; Red curve: Second derivative spectrum. (1) –CH3 asymmetric stretching; (2)
>CH2 asymmetric stretching; (3) –CH3 symmetric stretching; (4) >CH2 symmetric stretching; (5) >C=O stretching; (6) amide I
of α-helical structures; (7) amide I of β-sheet structures; (8) amide II; (9) tyrosine ring vibration band; (10) >CH2 bending; (11)
–COO− symmetric stretching; (12) >PO2 − asymmetric stretching; and (13) spectral range dominated by complex ring vibrations of
carbohydrates, P–O–P stretching, C–O–P stretching, >PO2 − symmetric stretching. Technique: A/T; number of scans: 64; nominal
physical resolution: 6 cm−1 ; and spectrometer type: IFS 28/B. (Reproduced with permission of Bruker Optics.)
3. The region between 1800 and 1500 cm−1 is dominated obtained from bands near 1740 cm−1 , resulting from
by the conformation-sensitive amide I and amide >C=O stretching vibrations of the ester functional
II bands, which are often the most intensive bands groups in lipids. Absorptions of nucleic acids can
in the spectra. As IR spectroscopy is an averaging also be expected in this spectral domain owing to
technique, the amide I and amide II bands cannot >C=O, >C=N, >C=C< stretching of the DNA or
provide structure information on a single protein; RNA heterocyclic base structures.(24) A band near
they rather indicate the predominance of α- or β- 1715 cm−1 , which is assigned to a >C=O stretching
structures present. Useful information can also be vibration, is routinely observed in the spectra of
Table 3 Optical properties of the most important, for IR spectroscopy materials

Material ZnSe ZnS Si Ge BaF2 KRS-5TM CaF2
(thickness) (d = 3 mm) (d = 2 mm (d = 2 mm) (d = 1,5 mm (d = 10 mm) (d = 1 mm) (d = 1 mm)
Transmission 0.5–20 0.4–14 1.2–15 1.8–23 0.15–12 0.45–45 0.12–12
range (μm)
Maximal 70 70 40-50 45 85 75 >90
transmission
(%)
Loss of reflection 30 25 46 53 6.5 28 12
(for two (10.6 μm) (10.6 μm) (5 μm) (10.6 μm) (5 μm) (10 μm) (4 μm)
surfaces) (%)
Solubility in water 1 × 10−3 6.9 × 10−6 Insoluble Insoluble 0.17 5 × 10−2 1.7 ×
at 300 K (g/100 10−3
g H2 O)
Refractive
index at
0.5 μm 2.66 2.42 3.42 4.01 1.48 2.73 1.44
5.0 μm 2.25 4.003 1.45 1.40
10.0 μm 2.41 2.20 1.40 2.37
20.0 μm 2.30 2.34
DOI: 10.1002/9780470027318.a0117.pub2
hydrated microbial cells and tissue material and is 4 EXPERIMENTAL METHODOLOGIES

known as a sensitive probe of base pairing in nucleic
acids. Weak features of nucleic acids between 1700 Major advantages of IR spectroscopy are that almost
and 1600 cm−1 are often overlapped by the much any kind of material can be measured and that the
stronger protein amide I bands, as the relative amount measurements are not limited to the physical state
of DNA or RNA per cell mass is mostly below 10% of the sample: samples may be solutions, dried films,
w/w (Table 1). Weak bands, which can be assigned suspensions, inhomogeneous solids, or powders.(17,18) In
to amino acid-side chain vibrations, occur near addition, there are no principal restrictions to recording
1498 cm−1 (phenylalanine), 1515 cm−1 (tyrosine), IR spectra of a given sample under very different
and between 1585 and 1570 cm−1 (aspartate and physicochemical conditions with respect to temperature,
glutamate carboxylate stretching). pressure, state of dispersion, pH, and so on. This is of
4. Complex absorption profiles are observed between advantage for biomedical analyses, as it is imperative
1500 and 1300 cm−1 arising predominantly from to test biological specimens under conditions that leave
>CH2 and >CH3 bending modes of lipids and the sample’s structures ‘as they are’, preferentially
proteins. A characteristic, but weak feature is often hydrated, unperturbed, and nondisintegrated. In case of
observed around 1400 cm−1 , which may be attributed microorganisms, samples are mostly measured as dried
to the symmetric stretching vibrations of –COO− films.
functional groups of amino acid side chains or free
fatty acids.
5. Around 1230 cm−1 superimposed bands typical of 4.1 Infrared Measurement Techniques for Microbial
different >P=O double-bond asymmetric stretching Samples
vibrations of phosphodiester, free phosphate, and
mono ester phosphate functional groups are In general, biological samples of microorganisms do
observed. In most cases, three to four different, not behave ideally. Different biological specimens
weakly pronounced features can be discriminated by should be compared under conditions where (i) IR
resolution enhancement, with the band near 1220 absorbances of the samples to be compared are not
cm−1 being most probably due to the phospho- too different, (ii) IR bands are not too intensive
diester functional groups of DNA/RNA polysac- in order to avoid detector nonlinearity effects, (iii)
charide backbone structures. The other >P=O Beer–Lambert law is at least approximately obeyed,
double-bond stretching frequencies are due to (iv) the signal-to-noise ratio is sufficiently high, and
head group vibrations of phospholipids or e.g. (v) varying baseline shifts caused by scattering at the
the phosphorus-containing carbohydrates such as sample surface and/or due to inhomogeneity within
teichoic acids, teichuronic acids, and lipoteichoic the sample itself are minimized. These requirements
acids (charged polymers, which may be present are best fulfilled using the traditional A/T, absorbance/
in substantial amounts in Gram-positive bacteria) reflectance (A/R), and the attenuated total reflection
(Section 2.1). techniques.(17,18,25,26)
6. The spectral region between 1200 and 900 cm−1 is
generally dominated by the symmetric stretching
vibration of PO2 − groups in nucleic acids and a 4.1.1 Absorbance/Transmission Measurements
complex sequence of peaks mainly due to C–O–C
and C–O–P stretching vibrations of various oligo- A/T spectra can be obtained from liquid solutions,
and polysaccharides. dispersions, or suspensions, from viscous or solid (dry)
7. The region between 900 and 600 cm−1 exhibits a films cast on suitable IR transparent plates. As water is
variety of weak, but extremely characteristic features the medium ubiquitously present in all biological samples,
superimposed on an underlying broad spectral water-insoluble and IR-transparent optical materials have
contour. This region may contain weakly expressed to be used. These are CaF2 , BaF2 , ZnSe, ZnS, KRS-
bands arising from aromatic ring vibrations of 5TM , Si, or Ge, which differ in refractive index, spectral
phenylalanine, tyrosine, tryptophan, and the various transmission range, water solubility, and more. Table 3
nucleotides. With the exception of only a few peaks surveys relevant optical properties of these materials.
(e.g. a band near 720 cm−1 , resulting from the Different cuvette systems have been described in the
>CH2 rocking modes of the fatty acid chains), valid literature for A/T-IR measurements of microbial samples.
assignments can hardly be achieved. Therefore, we Two versatile technical solutions for multisample routine
refer to this spectral domain as to the true fingerprint measurements of microbial samples (dried film samples)
region (Table 2). are shown in Figures 6 and 7.
DOI: 10.1002/9780470027318.a0117.pub2
(c) 1
MCT
detector
S
A/R
4
(b)
A/T
Multiplate 6
DTGS
xy-stage detector 3
2 2
(a)
Figure 6 The high-throughput multiplate extension (HTS- (a) 5

XT) from Bruker Optics (Germany). This multisample system
allows fully automated data acquisition from samples on IR
compatible microplates in a 96- or 384-well format. Microbial
samples can be measured as dry films in the A/R or A/T
measurement modes. (a) ZnSe and Ge 96-well microplates. (b)
optical path in the A/T mode. (c) HTS-XT module.
4.1.2 Absorbance/Reflectance Measurements
A/R measurements (sometimes also called ‘trans-

flectance’) can be performed on the same samples used for
A/T. The samples are placed on a light-reflecting instead (b)
of transmitting medium (e.g. silver, gold, or polished
stainless steel), which leads to a double passage or, for
Figure 7 (a) Schematic drawing of a multisample ATR-
special A/R devices, even a multipassage of IR radiation cuvette: (1) ZnSe prism with marked sample areas S; (2)
through the sample. demountable, hermetically sealed cartridge carrying the internal
reflection element (IRE) (1); (3) sliding carriage of the cartridge
by which the ZnSe prism may be moved through the IR beam
4.1.3 Attenuated Total Reflectance Measurements for consecutive measurements; (4) cartridge drive; (5) housing
with reflecting optics; and (6) IR beam. (b) The multisample
For an attenuated total reflectance (ATR) experiment, ATR cell.
the sample is no longer placed in the path of the
propagating IR beam as it is the case in A/T- or A/R
experiments but is brought into contact with the surface of
ATR cuvette suitable for the measurement of hydrated
an internal reflection element (IRE), where it can interact
and dried microbial samples is shown in Figure 7.(25,26)
with the IR radiation evanescing from the optically denser
IRE.(17) The penetration depth d of the electromagnetic
wave into the rarer medium (air or the sample) is defined
by the wavelength, the wavelength-dependent ratio n2 /n1 4.1.4 Diffuse Reflectance Measurements
of the refractive indices of the optical denser (n1 ) and the
rarer (n2 ) media and by the angle of incidence θ and is at Diffuse reflectance (DR) measurements of highly scat-
the order of a few micrometers. To give an example, the tering samples such as freeze-dried biological specimens,
penetration depth d at 10 μm wavelength (1000 cm−1 ) for surface-rich materials such as catalytic or chromato-
a ZnSe IRE (n1 = 2.41) and an angle of incidence α = 45◦ graphic carriers, and finely ground or powdered materials
is 1.64 μm (the refractive index of the sample is assumed can be performed using standard accessories supplied
to be n2 = 1.4). This means that only the thin layer which by most manufacturers of FTIR spectrometers.(7) A
is in immediate contact with the IRE contributes to the thin-layer chromatography accessory useful for FTIR
resulting ATR spectrum and allows characterization of measurements of dried microbial samples in DR has
dry and liquid samples. A specially designed multisample been described.(27)
DOI: 10.1002/9780470027318.a0117.pub2
4.1.5 Infrared Microspectroscopic Measurements modern FTIR imaging microspectrometer equipped with
a single detector and a focal plane array (FPA) detector
The introduction of IR microspectroscopy as a combina- used in the authors’ laboratory.
tion of IR spectroscopy and microscopy has considerably
increased the analytical sensitivity, specificity, and versa-
4.2 Sampling of Microbial Cells and Data Acquisition
tility of the IR spectroscopic technique and allows charac-
terization of microbial samples down to the subnanogram In order to obtain reproducible results, sampling of
level.(17,18,25,26) While standard bulk FTIR measurements biological specimens sample treatment procedures, IR
require between 107 and 108 individual microbial cells measurement techniques, and IR acquisition parameters
for collecting high-quality [signal-to-noise ratio (SNR)] have to be controlled and standardized rigidly. There is
spectra, IR microspectroscopy allows spectral measure- no simple and uniform answer to these requirements. For
ments at a similar SNR level with only 102 –103 cells. This microbial analyses, standardized experimental protocols
increase in sensitivity has been achieved by optimizing the including data acquisition and evaluation procedures
optical path of the microscope for IR light, e.g. by utilizing have already been published.(5 – 8) These standardization
exclusively reflective optical elements (gold mirrors, high efforts have been stimulated by the necessity to exchange
numerical aperture Cassegrain objectives) and the use of spectral data between different laboratories and construct
sensitive small-sized MCT (MgCdTe) detectors.(18) validated reference databases for routine identification of
In a standard IR microspectroscopic set-up with a microorganisms isolated in different laboratories.
single detector, the IR beam is focused at the sample IR spectra of microorganisms can be obtained with
plane by a Cassegrain objective while apertures restrict sufficient reproducibility from microorganisms provided
the size of the illuminated sample area. Measurements the microbiological parameters influencing cell growth
can be carried out in A/T or A/R (transflection) mode. (composition of growth media, incubation time, temper-
For measurements with very high spatial resolution, some ature of growth, etc.) can be controlled and standardized.
manufacturers offer specially designed ATR objectives in Compared with these requirements, sample collection,
which an IRE, mostly Ge, is placed in the focal point of sample preparation, and spectroscopic data acquisition
a Cassegrain objective. Figure 8 shows the example of a parameters (spectral resolution, scanning time, etc.) are
FPA detector
Single point CCD camera

detector
Aperture
Objective
IR source
Microscope stage
Beamsplitter
Fixed mirror
Moving mirror
Figure 8 Schematic view of a modern FTIR imaging micro-spectrometer (Agilent Cary 620 FTIR spectrochemical imaging
system). The imaging system is connected to a FTIR spectrometer and is equipped with a computer-controlled (x,y) stage. Samples
can be analyzed in the visible mode and in the A/T-IR mode (transmission or transflection). Data are recorded either by a single
point MCT detector or a multichannel focal plane array (FPA) detector. The FTIR imaging system in the author’s laboratory
is equipped with a 128 × 128 FPA that allows simultaneous acquisition of 16384 spectra from a sample area of 700 × 700 μm2 .
(Reproduced with permission from Ref. 28, Applied Spectroscopy, 66, 1091-1120 (2012).)
DOI: 10.1002/9780470027318.a0117.pub2
of secondary importance. Microbial samples suitable for • (RL1 ) that describes reproducibility of independent
IR measurements may be obtained either from liquid IR measurements on samples prepared e.g. from
cultures or directly from solid agar plates. These samples aliquots of one and the same aqueous suspension
and can be measured as hydrated pellets or dried films of microorganisms.
applying the A/T, A/R, or ATR techniques. Using the • (RL2, defined as RL3 in Ref. 8) that defines
A/T-technique, a typical simple protocol runs as follows. reproducibility of independent IR measurements
Subcultured microbial strains are cultivated on appro- on samples prepared from aliquots of microbial
priate solid nutrient agar plates. Time and temperature suspensions with the microorganisms grown on the
of growth depends on the type of microorganism tested same agar medium produced from different batches
(e.g. 24 h at 37◦ C for many human pathogens). Small over a sufficient period of time (e.g. 1 year).
amounts of the microorganisms are carefully removed
from the agar plate with a standard calibrated (e.g. 1 mm To calculate quantitative numbers for these RL,
in diameter) platinum loop and are suspended in 80 μL of an objective measure for description and comparison
distilled water. Subsequently, 30 μL drops of the suspen- of independent IR measurements is necessary. One
sions are transferred to a water-insoluble IR transparent possibility is the crosswise calculation of correlation
optical plate as small drops covering predefined sample coefficients ry1y2 (Pearson’s product–moment correlation
areas of the ZnSe plate of the multicuvette system shown coefficient) between pairs of measured spectra according
in Figure 6. The drop of microbial suspension is then to Equation (1).
dehydrated in a desiccator over a drying agent (P4 O10
Sicapent from Merck or Silicagel) applying a moderate
n
vacuum (≈3–5 kPa) to form transparent film disks suit- y1i · y2i − n · y1 · y2
i=1
able for A/T IR measurements. The optical plate is then ry1y2 = (1)

sealed in a gas-tight cuvette cartridge to control humidity n 2 n 2
and prevent the instrument from contamination and is y1i − n · y21 · y2i − n · y22
transferred to the automatic cuvette holder of the instru- i=1 i=1
ment. The physical parameters have to be kept constant
for all measurements to be compared. A suitable set of where y1i and y2i are the individual absorbance values of
physical parameters is as follows: 6 cm−1 nominal phys- the two spectra to be compared, n the number of data
ical resolution, Blackman–Harris three-term apodization points in the given frequency range, and y1 and y2 the
function, and a sufficient number of scans to reach an SNR arithmetic mean values of y1 and y2 .
better than 3000:1. It is advisable to take the single beam From the correlation coefficient ry1y2 , a so-called
reference spectrum through an empty place of the multi- differentiation index Dy1y2 may be defined according to
sample cuvette system directly before the single beam Equation (2):
sample spectrum is obtained. This eliminates virtually all
contributions from impurities on the optical materials and Dy1y2 = (1 − ry1y2 ) · 1000 (2)
minimizes problems arising from water vapor and CO2
bands caused by possible instabilities of the instrument where ry1y2 is the correlation coefficient. D may adopt
and dry air purging of the system.(5,6,25) values between zero and 2000, with zero for identical
spectra (or spectral ranges), 1000 for completely uncor-
related, and 2000 for inversely correlated spectra.(6,25,26)
4.3 Variability of Microbial Specimens and the
Mean D values between 0.4 and 0.8 for RL1 and
Problem of Reproducibility
between 7 and 10 for RL2 are obtained with Poisson-
The enormous diversity of microbial species and strains like distributions of calculated individual D values in
makes it an absolute necessity to perform measure- practice when analyzing the first derivatives in the spectral
ments on statistically significant numbers of samples on range between 900 and 1200 cm−1 of dried microbial film
hundreds of different species and strains of microor- samples with the cuvette system shown in Figure 6.(6,25,26)
ganisms. The most important factors influencing data RL1 is generally one order of magnitude lower than
quality is repeatability and reproducibility of measure- RL2 . Hence, the microbiological parameters (quality of
ments. First, different levels of reproducibility that define growth medium, cultivation, and sampling of bacterial
and limit the discriminative power of the IR technique biomass e.g. from the surface of solid agar plates) define
have to be considered. For quantification purposes, it is the repeatability of IR measurements on microorganisms.
useful to define different reproducibility levels (RLi ) of The comparison of strains from different species or genera
repetitive IR measurements on microbiologically iden- may yield D values higher than 300.(6,25,26) Obviously,
tical strains(6,17,18) : IR spectroscopy provides considerable spectral variance
DOI: 10.1002/9780470027318.a0117.pub2
that can be utilized to discriminate between different 5.1 Spectral Data Preprocessing
microorganisms.
Spectral preprocessing is an important first step in any
workflow of spectral spectra analysis and involves specific
processing procedures performed on the raw spectral
data.(33) It is now widely recognized that classification
5 DATA TREATMENT AND EVALUATION models developed on the basis of preprocessed data
generally perform better than models that solely rely
In many microbiological applications of IR spectroscopy, on the raw data.(33)
large data amounts, often from hundreds of measure- The main goals of data preprocessing can be summa-
ments on dozens of different microbial species and strains, rized as follows(33) :
are generated. These data must be inspected, assessed,
and analyzed. The use of specific multivariate classifica- • Improvement of the robustness and accuracy of
tion techniques that allow a comprehensive analysis of a subsequent quantitative or classification analyses
plurality of complex spectral patterns is therefore a neces- • Improved interpretability: raw data are transformed
sity. While univariate statistical analysis considers only a into a format that will be better understandable by
single property of a given object, i.e. one spectral feature both humans and machines
such as a single absorbance value at a given wavenumber, • Detection and removal of outliers and trends
multivariate methods can be used to evaluate large • Reduction of the dimensionality of the data mining
numbers of spectral features of the objects at the same task. Removal of irrelevant and redundant informa-
time. In this way, the interrelationships between the tion by feature selection.
object’s properties can also be considered. A particularly
important requirement for the IR analysis techniques Within the context of preprocessing spectral data from
is that it has to deal with an unusual large number of microbial samples, the following preprocessing routines
are of particular interest.
different categories or classes. Often, data from a multi-
tude of different microbial species and strains are needed
Quality tests These tests can be considered as outliers
to evaluate for reliable differentiation, classification, and
tests. Quality assessment of the raw IR spectra can be
identification. The arsenal of multivariate statistical tech-
carried out by defining the so-called quality criteria, or
niques provides ample methodologies for evaluation and
thresholds. Among others, such criteria can be specific
representation of such complex data structures. Out of a absorbance values in the amide I region, the SNR, IR
large number of pattern recognition techniques available, absorbance values of sharp water vapor features, and
two are of particular interest when dealing with IR spectra the presence or absence of optical fringes in the spectra
from microorganisms. These are unsupervised hierar- of microorganisms.(8)
chical cluster analysis (HCA) and supervised classification Calculation of derivatives Derivatives are popular means
analysis by artificial neural network (ANNs).(29 – 32) to enhance the resolution of IR spectra. Derivation is
Preprocessing is an extremely important first step thus routinely employed to resolve overlapping band
before any type of further analysis. It is one of components in the complex spectral profiles. Another
the main goals of data preprocessing to improve the advantage of derivative spectroscopy is that contribu-
robustness and accuracy of subsequent multivariate tions from baseline offsets or slopes are minimized.
pattern analysis.(33) One of these methods, HCA, is an This requires, however, a relatively high SNR. A
unsupervised classification method that attempts to find popular method for derivation of spectral data is the
out intrinsic group structures within the data without the Savitzky–Golay smoothing/derivative method.(34) This
need of any a priori class assignment, or partitioning method can be considered a band filtering technique
of data, into teaching, validation, and independent test that allows simultaneous filtering of unwanted high
data sets. ANN analysis, on the other hand, belongs (smoothing) and low (derivation) frequencies in the
to the group of supervised classification methods by experimental data.
which the class assignment of each individual object is Normalization Normalization is commonly applied to
required from the beginning. Partitioning of the whole minimize the effects of varying optical pathlengths
data into a teaching, validation, and independent test on the spectra. The result of normalization is a
data set is needed to avoid overfitting and ensure the spectrum that is scaled and offset corrected at the same
reliability of the classification results. The robustness time. Popular normalization methods are (i) Min–Max
of multivariate classification can be further enhanced normalization, (ii) 1-norm, (iii) 2-norm (also called
by using bootstrapping, jackknifing, or cross-validation vector normalization), and (iv) standard normal variate
methods such as the stringent leave-one-out technique. (SNV).(33)
DOI: 10.1002/9780470027318.a0117.pub2
Spectral window selection In biomedical application of is not available or not desired. Approaches for unsuper-
IR spectroscopy, spectral window selection is often vised classification include self-organizing maps (SOMs),
understood as a manual process guided by knowledge the adaptive resonance theory (ART), hidden Markov
and experience of the investigator. Window selection is models, and various methods of clustering (k-means,
carried out to reduce the data amount which is helpful fuzzy C-means, and hierarchical clustering). Cluster anal-
to speed up computationally intensive classification ysis aims at the classification of objects (spectra), i.e.
problems with lower computational efforts. The process at the description of the structure and property inter-
involves elimination of regions with no or only little relationships intrinsic to a given set of objects each
amount useful spectral information such as the region defined by a multiplicity of properties. The various
between 1800 and 2750 cm−1 or between 3400 and cluster analysis methods are based on different algo-
4000 cm−1 . rithms and may be based on different distance measures
Feature selection Feature selection is a process in which for determining the similarity or dissimilarity between
a subset of the original spectral variables, usually a set the individual objects. Cluster analysis techniques can be
of preprocessed IR absorbance values, is selected for a divided into hierarchical and nonhierarchical procedures.
specific purpose, e.g. to teach a neural network.(33) Examples for the nonhierarchical techniques are k-means
This is helpful to remove redundant or irrelevant and fuzzy-C-means clustering. However, in microbio-
information from the data and improves the robustness logical applications of IR spectroscopy, the hierarchical
and accuracy of subsequent supervised classification techniques are often favored, mostly because these tech-
analysis. niques tend to provide a high degree of correlation
between the IR-based microbial identification technique
In a practical set-up for microbial classification, prepro- and the established microbial taxonomy. The various
cessing of the raw IR spectra typically involves a number different HCA methods can be subdivided into divi-
of sequentially executed preprocessing steps(5,6,8,33) sive and agglomerative methods. Divisive techniques are
(Figure 9). The main advantage of such an approach more rapid, but only rarely applied, whereas agglom-
is that it allows addressing many of the different require- erative techniques are computationally demanding, but
ments for successive classification analysis. In the example
used more frequently. Agglomerative clustering proce-
of Figure 9, microbial IR spectra are first checked by
dures work according to the same hierarchical ascendant
individual quality tests (amide I intensity, SNR, water
principles: (i) in the first step of agglomerative HCA, a so-
vapor, etc.) and then subjected to first or second deriva-
called dissimilarity, or distance matrix is calculated which
tive Savitzky–Golay smoothing/derivative filtering. After
contains a complete set of (n × n) interspectral distances
that, spectra are normalized and a preselection of spec-
(n denotes the number spectra in the data set). The second
tral windows, carrying specific spectral information, is
step of agglomerative HCA involves the virtual clustering
performed.(5,6,8) For subsequent HCA, the spectral infor-
process. At this stage, (ii) the two most similar spectra,
mation of individual spectral windows can be rated by the
i.e. spectra with the smallest interspectral distance, are
so-called weighting factors that account for the specific
contributions of cellular compounds such as fatty acids determined and merged to form a new cluster. Then, (iii)
of the membrane or polysaccharides of the cell wall.(5,6,8) according to the predefined linkage method, the spec-
The choice between first or second derivatives of spectral tral distances between the new cluster and the remaining
windows boundaries and the weighting factors can be spectra are evaluated. The process is repeated n−1 times
considered a subjective process in which the experience until all objects are combined into one single cluster.
and the spectroscopic expertise of the investigator play The sequence of fusions can be represented graphically
the decisive role.(33) as a dendrogram.(35) It is noteworthy that agglomera-
For supervised classification, the sequence of prepro- tive hierarchical classification schemes cannot provide an
cessing steps often includes an additional feature selection objective criterion of best partitioning. In one way or
routine. In the example of Figure 9, discriminative other, the number of classes has to be determined by the
spectral features are selected from labeled spectral user, who needs at least some a priori knowledge about
subsets based on the covariance between the individual the inherent class structure of the data cloud. While it
features. is possible to provide the user with additional informa-
tion that facilitates the interpretation of the clustering
process (e.g. a curve showing the number of classes in an
5.2 Unsupervised Classification Analysis
HCA calculation versus intra-class variance), it remains
Unsupervised classification (also called data-driven) anal- a subjective decision whether the partition achieved will
ysis, is useful to investigate the hidden structure in be useful or not. HCA can be applied directly to prepro-
complex data in cases where an a priori class assignment cessed IR spectra (the objects) represented as points in
DOI: 10.1002/9780470027318.a0117.pub2
FT-IR measurements:
Staphylococcus aureus ATCC 6538 raw microorganism
spectra
1. Quality tests
(intensity, water vapor, and
SNR test)
Pseudomonas aeruginosa 2. Derivation

ATCC 27853 (second derivatives)
Absorbance
3. Normalization
(vector normalization)
Candida albicans ATCC 10231

4. Spectral window
selection
5. Feature
selection
3000 2500 2000 1500 1000 500 3000 2000 1000

Wavenumber (cm −1) Agglomerative Wavenumber (cm−1)
HCA ANN
(a) (b)
Figure 9 Sequence of spectral data preprocessing steps before classification of microorganisms. (a) Original raw absorbance and
vector-normalized second-derivative FTIR spectra of Staphylococcus aureus ATCC 6538 (red), Pseudomonas aeruginosa ATCC
27853 (blue), and Candida albicans ATCC 10231 (green). For better clarity spectra are shifted along the y-axis. (b) Schematics
of the sequence of data preprocessing steps carried out before unsupervised classification by hierarchical cluster analysis (HCA)
or supervised analysis by artificial neural networks (ANN). The workflow involves (1) tests for spectral quality, (2) application of
a Savitzky-Golay smoothing/derivative filter; (3) normalization; (4) spectral window selection, and (5) feature selection (only for
further ANN analysis).
the hyperdimensional spectra space with as many dimen- method has to be found out by trial and error. Satisfac-
sions as there are encoded absorbance values. Adequate tory results on clustering IR spectra of complex biological
distance measures used to calculate the interspectral molecules can be expected from HCA on the basis of a
distances can be Euclidean or Mahalanobis distances, combination of D-values (see Equation 2) as interspec-
Pearson’s product momentum correlation coefficient, tral distance measure and Ward’s or average linkage
and D-values (see Equations 1 and 2). Popular linkage algorithm.(5,6)
methods are average linkage, single and complete linkage, Figure 10 illustrates an example of unsupervised
and Ward’s algorithm. The decision what is the optimal classification analysis by agglomerative HCA with
distance measure and what is the best HCA linkage IR spectra from Gram-negative and Gram-positive
DOI: 10.1002/9780470027318.a0117.pub2
35 000
Heterogeneity 5.3 Supervised Classification Analysis
30 000
25 000
20 000
15 000
10 000
Contrary to unsupervised classification, supervised classi-
5000
fication requires a priori knowledge of the internal group
0
Cit. freundii or class structure of the spectral data to be analyzed.(31,32)
E. coli Gram
The term supervised, or concept-driven, classification, is
Ps. fluorescens negative thereby used to describe a group of techniques in which
Ps. aeruginosa teaching (training) data are analyzed by a learning algo-
Ent. faecalis rithm to create a classification function from the input
Str. pyogenes
data. The input data typically consist of pairs of input
Bac. subtilis
vectors (spectra) and the corresponding target values
St. aureus Gram (class assignments). During teaching, the information
positive
contained in the input vectors is analyzed, processed,
St. aureus and assigned to map the required output values. One of
the important properties of learning machines is their
St. epidermidis ability to generalize while the term generalization is
C. albicans
used to describe how effectively the supervised classi-
fier behaves in case of new, i.e. previously unseen data.
C. tropicalis
For training and for assessing the performance of the
C. kefyr supervised classifier, it is essential to split the existing
Yeasts
C. glabrata data into subsets of teaching, validation, and independent
C. krusei test data. The teaching subset contains sets of training
C. parapsilosis examples for which the target class membership func-
tion is known. Data comprising the (internal) validation
Ward’s algorithm Frequency ranges (weights) = set are required to estimate the generalization error,
Correlation with 1200–900 cm−1 (1.0) recognize overfitting, and optimize the architecture of the
scaling to first range 3000–2800 cm−1 (1.0) classifier. Independent test data must not be used during
first derivative 1500–1400 cm−1 (1.0)
teaching. The error on the independent test set gives the
Figure 10 Dendrogram of an unsupervised classification unbiased estimate of the generalization error. Among a
analysis by agglomerative hierarchical cluster analysis (HCA) on plurality of supervised classification techniques, support
IR spectra from Gram-negative and Gram-positive bacteria and vector machines (SVMs), decision tree learning methods,
yeasts. FTIR spectra were acquired from three independent and most notably methods based on ANN analysis have
microbial cultures of each microbial strain. For HCA, raw found broad application in the field of biomedical IR
microbial spectra were pre-processed by a (i) first derivative
Savitzky-Golay smoothing/derivative filter with 9 smoothing spectroscopy.
points, (ii) vector normalization and (iii) spectral window ANNs are known as self-adaptive, parallel machine-
selection in the 900-1200, 1400-1500 and 2800-3000 cm−1 regions, learning systems made up of simple processing units,
respectively. Hierarchical cluster analysis was done using a which have a natural propensity for storing experimental
combination of D-values as interspectral similarity measures knowledge and making it available for practical use.
and Ward’s algorithm as the linkage method. Pre-processing
and hierarchical clustering were performed using the software
ANNs are primarily used for pattern recognition
OPUS from Bruker Optics, Germany. purposes. They resemble the connectivity in brain in
two respects. The first is that knowledge is acquired
by the network from a learning process and in the
second the interneuron connection strengths, known as
bacteria and yeasts. In this example, raw microbial synaptic weight, store the acquired knowledge. Several
spectra were preprocessed by applying a first derivative models of information processing of biological systems
Savitzky–Golay smoothing/derivative filter with nine are used as an analogy. In nerve cells (neurons), the
smoothing points, subsequent vector normalization, and following abstract elements can be defined: (i) synapses
selection of spectral window in the 900–1200, 1400–1500, that have chemical signal substances; (ii) dendrites, where
and 2800–3000 cm−1 regions, respectively. HCA was the incoming signals are received and processed; (iii) the
carried out using D-values as interspectral similarity cell nucleus (kernel), which controls this process; and (iv)
measures and Ward’s algorithm. The dendrogram of axons, which propagate the impulse irritation forward.
Figure 10 demonstrates a large degree of similarity The incoming weights are calculated via different input
between the microbial taxonomy at the one hand and layers of adjacent cell connections. After summation,
the IR classification technique on the other. an impulse is induced according to the strength of the
DOI: 10.1002/9780470027318.a0117.pub2
weighted signals. The propagation function transports the teaching should be performed on a representative
impulse to a new cell or directly to the output layer. The random sampling of the data population. In order to
mathematical functions involved in this process are (i) estimate the generalization, the stringent leave-one-out
an activation function, (ii) a description of the activation cross-validation method may be used. Here, validation
state of the neuron, and (iii) an output function. The is performed on the basis of n − 1 of the cases
activation function describes how a new activation state (spectra) investigated and the one case left out is tested
is developed from a preceding state of the neuron. The (reclassified).
activation state defines the degree of activation and
the propagation function has to weigh and combine all
incoming signals. Therefore, a summation is often used 6 APPLICATIONS
as an algorithm. The value produced by the propagation
function is used as input by the activation function. The 6.1 Characterization of Particular Cell Compounds and
activation function then produces the output. Binary, Structures
linear, sigmoid, and logarithmic functions are used as
activation functions. The connecting network defines the The following section describes the characterization of
manner in which the architecture of the network operates. particular cell components identified and analyzed by
For feed-forward ANNs, the information propagates specific vibrational bands. In this context, resolution
from the input layer through the hidden layer(s) to enhancement techniques and/or difference spectroscopy
the output layer. During teaching, the reaction value are of particular help to identify compound-specific
(calculated output) is compared with the expected target vibrational features. IR spectra from a number of
value at the output of the net. As long as deviations microorganisms exhibit characteristic IR bands of vari-
occur, the extent of the deviation is used as a measure of able intensities that cannot be considered as spectral
how strong and where the weights are to be changed. variation due to experimental conditions or changes of
The second group of networks used is the feedback microbiological parameters. A detailed analysis of these
nets that can feed back the outcoming information from spectral features revealed the presence of particular cell
the neurons to a preceding layer. Examples are the so- constituents such as intracellularly accumulated storage
called Hopfield, Elman, and Jordan nets. A supervised materials, endospores, and cell surface structures such
ANN requires a desired output for each input vector as proteineacous and polysaccharidal capsules or the
(spectrum), which is then compared to the actual output detection of metabolically released CO2 in cells and cell
generated at a certain stage of analysis. A couple of cultures.(36,37)
learning strategies and learning rules are used to obtain
incremental changes in the weights in order to optimize 6.1.1 Cell Storage Materials
an error criterion. The learning strategy attempts to
minimize a global error function for the given set of Poly-β-hydroxy fatty acids (PHFs), to give an example,
teaching data. The process of computing local errors for are energy and carbon reserve compounds found in many
each processing element for optimization of the weights prokaryotes. Generally, PHFs are accumulated when
is continued interactively until the error is minimized to supply of energy and carbon is in excess. Under conditions
an acceptable value. The advantage of ANNs is that they of starvation, PHFs can be utilized and degraded by
can be used as self-learning parallel working machines. the microorganisms helping the cells to survive. It is
ANNs can be taught using a maximum of repetitive known that the survival rate is related to the amount
measurements on each of the objects to be compared. The of PHFs, which are intracellularly accumulated as small
redundant information is then ignored, only the significant granules. These granules can be easily detected light-
information is stored. This saves computer storage microscopically or by electron microscopy (marked by
capacity. The more data are used for teaching, the more G in Figure 11a). Poly-β-hydroxybutyrate (PHB), to give
precise will be the assignment of an unknown object to an example, is frequently found in bacteria (e.g. various
one of the predefined patterns. This self-learning capacity strains of Bacilli, Clostridia, Acetobacter, Legionella, and
increases the flexibility significantly mostly because the Pseudomonas). In most microbial IR spectra, the ester
diffuse, possibly small, but characteristic spectral variance carbonyl band at 1738 cm−1 is only a small, weakly
can be recognized correctly. Furthermore, systematic expressed shoulder. This band is caused predominantly
errors in measurements such as noise, baseline slopes, by the >C=O stretching vibration of ester bound fatty
and other unwanted spectral properties are represented acids of in the membrane forming ester bound lipids.
in the teaching examples and will be recognized during the Gram-negative bacteria generally show a stronger ester
evaluation process, which may be helpful to reduce the carbonyl band than the Gram-positive organisms because
probability of misclassifications. To this end, however, of the presence of an additional membrane layer, the
DOI: 10.1002/9780470027318.a0117.pub2
outer membrane. In spectra obtained from Legionella, (a)

Pseudomonas, or Bacillus strains, for instance, a rather
prominent ester >C=O stretching band, accompanied
by a number of additional bands between 900 and 1500
cm−1 , is occasionally observed, which is not permanently G
present throughout the cell cycle. For some Legionella
pneumophila strains, this band reaches a maximum
of intensity after 48 h, whereas at 120 h of growth
practically, no additional >C=O ester band can be
detected. Figure 11(b) shows the overlaid FTIR spectra 500 nm
of L. pneumophila strain RKI/II8 grown for 48 (red
curve, Figure 11a) and 120 h (black curve, Figure 11b),
0.60 (b)
1550
respectively. Inspection of the two spectra reveals Ix,1
Absorbance (AU)
0.50 α=
1738
characteristic differences, most prominent in the ester Ix,2
0.40
carbonyl stretching region around 1738 cm−1 . Figure 11(c)
0.30 Ia,2 a
shows the difference spectrum as calculated from both 0.20 Ib,2
Ia,1
curves. This difference spectrum closely resembles the 0.10 Ib,1
FTIR spectrum recorded from isolated and purified PHB. 0.00 b
At least 10 bands can be identified and are assigned 1800 1600 1400 1200 1000
to a typical polyester compound. Relative quantitative Wavenumber (cm−1)
determination of PHB is achieved by calculation of the 0.30
1738
(c)
Absorbance (AU)
ratio α = Ix1 /Ix2 (Figure 11b), where Ix1 is the intensity 0.25
of the ester carbonyl peak at 1738 cm−1 for spectrum
1101
1186
0.20
1059
x. The peak at 1738 cm−1 is used as a measure marker
1305
0.15
1135
1262
1383
976
1445
band of PHB content, and Ix2 denotes the intensity of the 0.10
amide II peak at 1550 cm−1 in spectrum x and is used as 0.05
an internal standard that measures the approximate total 0.00
cell mass.(36) 1800 1600 1400 1200 1000
−1)
Wavenumber (cm
6.1.2 Endospore Formation in Cells Figure 11 IR-spectra of a Legionella pneumophila strain

(isolate RKI II 8) that produces different amounts of the intra-
Spore formation in bacteria serves as a strategy for cellular storage material poly-β-hydroxybutyrate (PHB) acid as
a function of growth time. (a) Electron micrograph showing
survival under unfavorable conditions. Endospores are granular PHB inclusion bodies (G). (b) Spectra obtained after
resistant to heat, radiation, and chemicals and can 48 h (red curve), and after 120 h (black curve). (c) Difference
survive starvation for years. In contrast, to vegeta- spectrum (a minus b). Some absorption bands diagnostic for
tive cells, spores are not killed by standard steriliza- PHB are annotated. The Legionella pneumophila strain was
◦
tion techniques and can cause severe medical prob- grown on CYE agar plates at 37 C. The intensity of the ester
lems. The multistage transformation process that occurs carbonyl band at 1738 cm−1 and the amide II band near 1550
within the mother cell has been investigated thor- cm−1 can be used to determine the relative amount of PHB
present in the cells. Data acquisition parameters see legend to
oughly by a variety of molecular genetic, analytical, Figure 5. The EM image was kindly provided by Michael Laue
and structural techniques. FTIR spectroscopy can be (RKI, ZBS4)
used to monitor the multiphase process of sporulation,
avoiding or complementing more time-consuming proce-
dures such as staining, microscopic enumeration, and
isolation.
The IR spectra of some strains of bacilli and clostridia of the difference spectrum depicted in Figure 12 are
exhibited a couple of relatively weak extra bands diagnostic for dipicolinic acid (DPA), a compound
around 1279, 767, 728, 703, and 660 cm−1 (Figure 12). involved in sporulation. The strong >C=O stretching
Calculation of the difference spectra reveals prominent band of COOH of DPA (expected around ∼1705 cm−1 )
bands between 1650 and 1250 cm−1 . Light-microscopic cannot be discovered in the difference spectra, as chelate
examination of the same suspensions used for FTIR binding of Ca2+ ions in vivo results in the two stretching
spectroscopy reveals up to 50% free and mature vibration bands of the COO− group at ∼1605 and ∼1405
endospores. Five out of the ten most intensive bands cm−1 , respectively.(36)
DOI: 10.1002/9780470027318.a0117.pub2
Clostridium sordellii
DSM 2141
1
Absorbance
2343
2
2343
Absorbance
3
1
3000 2500 2000 1500
1616
1570
2 (a) Wavenumber (cm−1)

1378
1440
0.70
2343
703
767
660
728
0.65
Absorbance (AU)
1279
With glucose
0.60
0.55
3
0.50
Without glucose
2000 1750 1500 1250 1000 750 500 0.45
Wavenumber (cm−1) 2365 2355 2345 2335 2325 2315

(b) Wavenumber (cm−1)
Figure 12 IR spectra of Clostridium sordellii (strain DSM
2141) which may produce different amounts of endospores as
a function of growth time. Spectra were obtained from two Figure 13 CO2 production by yeast cells. (a) Infrared spectra
different cell populations grown for 48 hours (red curve 1) and of fully hydrated samples of a Candida dubliniensis clinical
24 hours (black curve 2) on Columbia blood agar plates at 37◦ C, isolate (strain M467II/97) (curve 1), pure water (curve 2),
respectively. The difference spectrum (1) minus (2) is shown by and (curve 3) the difference spectrum (1) minus (2). (b) Time
curve (3) at the bottom. Data acquisition parameters see legend evolution of CO2 -production after adding (red curves) and not
to Figure 5. (Adapted from Ref. 36. © Elsevier, 1995.) adding (blue curves) glucose to a yeast culture of Issatchenkia
occidentalis. Spectra of (a) were measured in A/T using a cuvette
equipped with CaF2 windows and an optical pathlength of 8 μm.
Kinetic spectra of (b) were obtained with a 250 μm pathlength
using CaF2 optical windows. Physical resolution: 4 cm−1 ; number
6.1.3 Metabolically Released CO2 of scans: 64; detector: DTGS; spectrometer: IFS 66 (Bruker
Optics, Germany).
The detection and quantitation of metabolically released
CO2 in bacteria and yeasts is an additionally interesting
item of the FTIR technique. Carbon dioxide produced
by the cells can be determined extremely sensitive,
as the CO2 band near 2343 cm−1 is present in a a substrate for CO2 production. Spectra of both series
spectral region where SNR is optimal and is usually are collected as a function of time in intervals of 8
devoid of overlapping spectral features of the biological min after addition of glucose to the yeast culture. A
material.(37) Figure 13(a) shows the spectrum of a fully control culture without glucose is measured identically.
hydrated sample of a Candida dubliniensis isolate (1), Obviously, the intensity of the peak at 2343 cm−1 can
pure water (2), and the difference spectrum 1 minus 2 be used to quantify CO2 evolution and the kinetics of
(3). In the spectral region between 2300 and 2400 cm−1 , CO2 release after addition of various different substrates
the asymmetric stretching band of CO2 hydrates can be to a yeast culture. Provided 13 C-labeled substrates are
detected near 2343 cm−1 . Apparently, the CO2 band available, the simultaneous detection of the kinetics of
monitors distinct levels of metabolic activity in the cells CO2 production from two different substrates can be
or within the culture. Figure 13(b) shows two series of determined as well, as the 13 CO2 peak is detected near
IR spectra obtained from yeast cultures of Issatchenkia 2277 cm−1 , quite separate from the 12 CO2 stretching band
occidentalis differing in supplementation with glucose as near 2343 cm−1 .(37)
DOI: 10.1002/9780470027318.a0117.pub2
6.2 Differentiation, Classification, and Identification of techniques, biochemical, or enzyme reaction patterns
Microorganisms are almost always based on MALDI-ToF MS or
ultimately molecular genetic techniques as the gold
6.2.1 Differentiation of Microorganisms
standard.(38)
Estimating the exact number of species of bacteria Microbial diversity is always structural and biochemical
is impossible with today’s technology. However, some diversity. IR spectroscopy of intact microorganisms
thousands of bacterial species are already named provides information on the structure and composition
systematically with each species comprising 5–1000 of the whole cell. Three typical MIR spectra of microbial
strains, each described systematically by a strain number samples, a Gram-positive (Staphylococcus aureus, strain
(corresponding numbers for yeasts and fungi are rarely ATCC 6538), Gram-negative (Pseudomonas aeruginosa,
available). Many have been described as human, plant, strain ATCC 27853), and yeast (Candida albicans, strain
or animal pathogens, which emphasizes that most ATCC 10231), isolate, each dehydrated to a thin film
efforts devoted to the isolation and characterization disk on a ZnSe optical substrate (Figure 9a). While
of microorganisms to date is related to the problem the spectrum 3 of the yeast strain differs significantly
of host–cell interaction. Probably, there are a million from the spectra 1 and 2 obtained from two different
or even more different bacterial species on earth bacteria strains, the latter two are seemingly similar.
which have not yet been recognized. Taxa, such as However, a more detailed analysis of these two spectra
families, genera, species, or strains, are defined as groups by focusing to selected spectral regions and calculating
of related microorganisms comprising different levels second derivatives proves the presence of distinct spectral
of discrimination. Modern methods used in microbial differences between these microorganisms as well (see
taxonomy in addition to classical microscopy, staining details in the left panel of Figure 9).
Heterogeneity Heterogeneity
3000
2500
2000
1500
1000
4500
4000
3500
3000
2500
2000
1500
1000
500
500
Candida
0
0
albicans
Control C
Candida
albicans 1580/95
Control A
Candida
glabrata 23 I/96
1810/95
Candida
kefyr 3142/94
Candida 317/95
krusei 723/95
2200/94
Candida 1890/95
parapsilosis
1887/95
2147/95
(a) Candida (b)
tropicalis 828/95
Ward’s algorithm Frequency ranges (weights) = Ward’s algorithm Frequency ranges (weights) =
Correlation with 1070–1014 cm−1 (1.0) Correlation with 891–859 cm−1 (3.0)
Scaling to first range 1171–1099 cm−1 (1.0) Scaling to first range 761–735 cm−1 (1.0)
second derivative 1420–1339 cm−1 (1.0) 991–889 cm−1 (1.0)
1662–1670 cm−1 (1.0) 1401–1369 cm−1 (1.0)
Figure 14 Classification of microorganisms at the species and at the strain level. (a) Dendrogram of a hierarchical cluster analysis
performed on 240 spectra of 6 different species belonging to the genus Candida. (b) Dendrogram obtained when cluster analysis
is performed on the spectra obtained from 13 different strains belonging to the species Candida albicans. Ward’s algorithm was
applied.(5) For calculating the distance matrix the Pearson’s correlation coefficient (D-values, see Equation 2) was used. The spectral
windows used to calculate the distance matrix can be seen at the bottom of the respective figure. Further experimental and data
evaluation parameters see legends to Figures 5 and 10.
DOI: 10.1002/9780470027318.a0117.pub2
6.2.2 Classification of Microorganisms perfectly between the six different species, as established
by molecular genetic techniques.
Because microbial IR spectra are complex spectroscopic As all cell components depend on the expression of
signals encoding the superposition of hundreds or even smaller or larger parts of the genome, the FTIR spectra of
thousands of bands that cannot be resolved by any microorganisms display specifically a complete phenetic
means, pattern recognition techniques have to be used and genetic fingerprint of the cells under study. This is
which consider the spectra as kind of one-dimensional why the specificity of the technique is very high, allowing
fingerprints rather than a combination of discrete differentiations at quite different taxonomic levels, even
band intensities, frequencies, and bandwidths. Figure 10 down to the subspecies, strain, and/or serogroup/serotype
shows the dendrogram of a classification obtained by level.(26,39 – 49) Figure 14(b) shows the dendrogram of a
HCA performed on IR spectra from very diverse cluster analysis performed on repetitive and independent
microorganisms. The structure of the dendrogram clearly measurements on 13 different strains of C. albicans. This
proves that a microbiologically meaningful classification dendrogram proves that a clear distinction between the
can be obtained on the basis of the information contained different strains of C. albicans is possible as well, simply
in the microbial IR spectra that. Moreover, it is interesting using systematically the information contained in the IR
to note that the IR technique is not restricted to the spectra. Similar results are not easily obtained by classical
analysis of bacteria. Yeasts and fungi can obviously be microbiological techniques. In general, species and strain
analyzed as well. Figure 14(a) shows the dendrogram of a differentiations are only obtained using two different
hierarchical classification trial based on 240 microbial IR independent techniques and are generally much more
spectra obtained from independent measurements on 29 time consuming.
different strains belonging to 6 different Candida species. Figure 15 shows another example providing clear
Six distinct clusters are recognized, which discriminate evidence that FTIR can classify microorganisms at the
Heterogeneity Heterogeneity
2000
7000
6000
5000
4000
3000
1000
350
300
250
200
150
100
50
0
0
Myc. smegmatis
Myc. fortuitum
Myc. phlei ATCC 6841
Myc. fortuitum
6550/01
Myc. fortuitum
Myc. fortuitum
7157/01
Myc. chelonae
Myc. fortuitum
Myc. abscessus
8728/02
Myc. chelonae
Myc. avium Myc. fortuitum
9644/02
Myc. intracellulare
Myc. fortuitum
Myc. marinum 8727/02
Myc. kansasii Myc. fortuitum

9097/02
(a) Myc. xenopi (b)
Ward’s algorithm Frequency ranges (weights) = Ward’s algorithm Frequency ranges (weights) =
correlation with 1200–900 cm−1 (1.0) correlation with 1200–900 cm−1 (1.0)
scaling tp first range 1500–1200 cm−1 (1.0) scaling to first range
first derivative 1700–1500 cm−1 (1.0) first derivative
Figure 15 Classification of microorganisms at the species and at the strain level. (a) Dendrogram of a hierarchical cluster
analysis performed on spectra obtained from 10 different species belonging to the genus Mycobacterium. (b) Dendrogram obtained
when cluster analysis is performed on the spectra obtained from seven different strains belonging to the species Mycobacterium
fortuitum.(39) The first derivatives of the spectra considering the spectral ranges 900–1200, 1200–1500, and 1500-1700 cm−1 (a), and
900-1200 cm−1 (b), were used, respectively. Spectra were obtained from microcolonies using an FTIR microscope and the stamping
technique as described in section 6.3 and in Ref. 39. (Figure has been adapted from Ref. 39.)
DOI: 10.1002/9780470027318.a0117.pub2
species and at the strain level at the same time, using 6.2.3 Identification of Microorganisms
systematically the information contained in the microbial
FTIR spectra. Figure 15(a) shows a classification result Identifying unknown microbial isolates implies the
in the form of a dendrogram obtained from FTIR creation of valid spectral reference databases that can
measurements on 28 different strains from 10 different be used in different laboratories. Various algorithms can
species of the genus Mycobacterium.(39) As in Figure 14, be used to identify unknown microbial strains on the basis
the dendrogram clearly provides evidence that FTIR of such a microbial reference database. First of all, it is
can differentiate Mycobacteria at the species level. Most pertinent that these reference data contain representative
interestingly however, the spectral information contained numbers of spectra covering all relevant spectral types to
in the spectra can also provide strain-specific spectral be identified. Unknown microbial samples will only be
traits as is exemplified in Figure 15(b). Simply using an correctly identified with validated reference databases
optimized combination of spectral windows i.e. spectral of microbial IR spectra. Identification is then achieved
information, different then that used for the species simply by comparing the IR spectrum of an unknown
differentiating dendrogram (see legend to Figure 15), microorganism with all entries of the reference database.
a superb differentiation at the strain level is obtained An already extensively tested algorithm is based on the
for the different strains of Mycobacterium fortuitum. calculation of the so-called differentiation indices (D
It should be accentuated here again, that for species values) as defined by formula 2.(5,6,8) The spectrum is
and strain differentiation the information contained in first subdivided into several spectral windows, selected
one and the same spectrum could be used without the such that they contain the most discriminative spectral
need of any additional measurements or by changing information. The combination of these spectral windows
the experimental method, which is generally the case is then used in a stepwise correlation procedure to
when conventional molecular or biochemical techniques determine the most similar spectrum contained in the
are used. database. Finally a list of most probable hits is reported by
Level 1:
Gram Gram
negative positive
Level 2:
E. coli S. aureus E. faecium group
E. cloacae P. aeruginosa CNS E. faecalis
E. aerogenes Streptococcus spp.
Level 3:
E. coli E. cloacae S. aureus CNS E. faecalis E. faecium Streptococcus

E. aerogenes group group spp.
Level 4:
E. cloacae E. aerogenes
Figure 16 Sequential identification scheme used for identification of different species found in blood cultures: A hierarchical
artificial neural network (ANN) was specifically trained to obtain optimal results for the identification of species frequently found
in sepsis patients.(12) As input data for training the sequential ANN model, spectral data obtained from the 125 reference strains
shown in Table 4 have been used. Spectra were obtained from microcolonies using the FTIR microscope shown in Figure 17, and
the stamping technique as described in Section 6.3. Dried microcolony spots deposited on a round CaF2 plate (Ø 25 mm) were
measured (Figure 17). The spots were obtained by the stamping technique as described in Section 6.3. Technique: A/T; number of
scans applied: 512; nominal physical resolution: 6 cm−1 ; FT-IR-microscope: IRscope II coupled to an IFS 28/B spectrometer.
DOI: 10.1002/9780470027318.a0117.pub2
the program.(6,45 – 49) These libraries contain thousands of Table 4 Microbial species used to construct the infrared
spectra of different species and strains especially relevant spectral reference database
for foodborne microorganisms in the food industry. Bacteria No. of strains used
Spectral reference libraries combined with ANN Staphylococcus aureus 13
analysis and feature extraction methodologies may Coagulase negative 18
even more efficiently be used to identify unknown staphylococci (CNS)
microorganisms.(50,51) The elaboration of such ANN Staphylococcus epidermidis 6
libraries is partitioned into several steps including data Staphylococcus schleiferi 2
Staphylococcus saprophyticus 2
preprocessing (normalization and feature extraction)
Staphylococcus haemolyticus 2
elaboration of adequate network architectures, training Staphylococcus capitis 2
functions, and learning parameters. The most important Staphylococcus warneri 2
idea of this approach is to first establish optimized ANN Staphylococcus hominis 2
libraries for subgroups of the whole reference library Streptococcus spp. 11
Streptococcus oralis 3
and then to connect them in the form of a multilay-
Streptococcus salivarius 2
ered neural net architecture. The user working with Streptococcus pneumonia 2
such types of libraries can solve complicated identifi- Streptococcus pyogenes 3
cation problems by running through all layers (neural Streptococcus variant Gr. A 1
networks of subgroups) of these reference databases in Enterococcus faecalis 6
Enterococcus gallinarum 2
a controlled multistack procedure. The information flux
Enterococcus casseliflavus 1
is controlled by one heading or top-level network whose Enterococcus faecium group 9
single task is to separate, for example, first between Enterococcus faecium 5
bacteria and yeasts, then between Gram-positive and Enterococcus hirae 2
Gram-negative bacteria, and in further steps between Enterococcus durans 2
a number of different genera. In this way, to give an Escherichia coli 10
Pseudomonas aeruginosa 5
example, a microbial IR identification library has been Enterobacter cloacae 6
constructed that comprises six different bacterial genera Enterobacter aerogenes 6
each containing a representative number of species and
strains within the genera Pseudomonas, Bacillus, Staphy-
lococcus, Streptococcus, Aeromonas, Mycobacterium, and
two subnets for the identification of different Candida
species and for the differentiation between fluconazole- Table 5 Clinical study: comparison of phenotypic
sensitive and fluconazole-resistant C. albicans strains, (BacT/Abbot) and FTIR results of blood samples from
respectively.(50,51) 121 patients
A more recent example of how a reference database FT-IR microscopy
of microbial FTIR spectra can be used to construct an
Correct ID False ID
efficient model for identification of unknown strains
is shown in Figure 16 on the example of species S. aureus 27 (96.4%) 1 x CNS
identifications for patient isolates from blood cultures CNS 52 (98.1 %) 1 x S. aureus
E. coli 12 (100 %)
in the context of sepsis in clinical settings.(12) This figure E. cloacae 2 (100 %)
shows a multilevel ANN model optimized for species P. aeruginosa 1 (100 %)
identifications on the basis of a data set of microbial E. faecalis group 6 (100 %)
reference spectra collected from 125 different strains E. faecium group 7 (100 %)
of Gram-positive and Gram-negative bacteria typically Streptococcus spp. 5 (100 %)
found in blood cultures (Table 4). In a prospective clinical
case study on 112 patient isolates, 110 (98.2%) isolates
(Table 5) could be correctly identified.(12) The potentials
of optimized ANN models for rapid identification
of microorganisms have been recognized by several
authors.(41,44,52,53) The superiority of this approach, to give using 277 isolates. While the relatively simple D-value
another example, was recently recognized in a publication approach gave only 85% correct identifications at the
by Rebuffo et al.(44) These authors developed an ANN species level, the optimized ANN model yielded 96%
model for the identification of Listeria isolates from correct results for all species in toto, whereas for the
food samples, including 243 strains from five different human pathogen Listeria monocytogenes, even 99.2%
Listeria species. This identification model was challenged correct identifications could be obtained.
DOI: 10.1002/9780470027318.a0117.pub2
6.3 IR Microspectroscopy and IR Imaging of Microbial of spectra reduces the measurement time considerably.
Microcolonies The optical configuration of FPA imaging systems does
not involve apertures, so that a much larger proportion
The light microscope has long been a standard analyt-
of the IR light is available for the measurements. The
ical instrument in microbiological routine and research
lateral spatial resolution in FPA imaging data is defined
laboratories. While light microscopy provides informa-
by physics (Abbe’s criterion) and the geometry of the
tion on shape, color, and morphology of the complex
FPA (see Ref. 54 for details).
biomedical samples, IR spectroscopy may give comple-
The microbiological characterization of unknown clin-
mentary information about the sample’s structure and
ical, food, water, or airborne microbial specimens include
composition at the molecular level. Thus, the combina-
the fundamental steps of detection, differentiation, and
tion of light microscopy with the sensitivity and specificity
identification of the prokaryotic cells. Different tech-
of IR spectroscopy has opened new avenues to charac-
niques are used to detect, count, and differentiate
terize microorganisms and to obtain structure information microorganisms. These techniques include the counting
on ‘what you see’. The marriage between spectroscopy of colony-forming units, measurement of optical density,
and microscopy has been promoted by the introduction use of cell counters or cell sorters, and application of a
of modern FTIR spectrometers and the development whole arsenal of techniques by which microbial cells can
of sensitive small-sized MCT detectors without which be differentiated including molecular genetic methodolo-
modern IR microspectroscopy would not be possible. IR gies.
microspectroscopy pushed the detection limits down to Some fundamental efforts have been made to integrate
the subnanogram level and opened up the field of spatial detection, differentiation, and identification in one
resolution and imaging to IR spectroscopy. While conven- single instrument using IR microspectroscopy.(25,26)
tional light microscopes are equipped with condensers When applying IR microspectroscopy to microbial
and objectives containing dispersive optical materials with samples, three questions have to be addressed: (i)
only limited transparency for mid-IR radiation, the IR what are the detection limits and what would be
microscope generally requires reflecting optics, including the minimum cultivation time allowing collection of
mirrors and all-reflecting Cassegrain objectives.(18,54) In suitable IR microspectra with an acceptable SNR, (ii)
single detector IR microscopes, an assortment of aper- how can subnanogram amounts of microbial material
tures placed in the plane of the real intermediate image be reproducibly sampled and measured by the IR
is used to mask the sample area of interest to the microspectrometer, and (iii) what is the taxonomic
required values. In this way, apertures can be used to resolution of the microspectroscopic characterization
define the lateral spatial resolution in IR microspec- method.
troscopy down to the order of the wavelength of the IR Figure 17 shows the IR microspectroscopy instrumenta-
light.(55,56) tion that is used in the laboratory of the authors to answer
Modern single-detector IR microscopes are equipped these questions. The figure displays a Bruker IRscope
with computer-controlled x,y microscope stages. Such II IR microscope (Bruker Optics Ettlingen/Germany),
stages are helpful to characterize sample spots either in which is connected with an IFS28/B FTIR spectrometer
an operator-controlled single spot measurement manner (Bruker). The microscope is equipped with two sensi-
or in the fully automated data acquisition mode. If tive broadband single element MCT detectors (detector
automated IR microscope measurements are conducted element size of 100 × 100 μm2 and 250 × 250 μm2 , respec-
in a stepwise manner, e.g. by raster scanning a rectangular tively), a software-controlled x,y-stage, a video camera
sample area, one can reassemble IR images from for documenting the samples in the visible mode, and
the resulting spectra. In these so-called IR mapping a custom-made purge box for flushing the microscope’s
experiments, single-point IR microspectra are recorded at optical compartment with dry air (Figure 17a). Figure 17
each point of a rectangular grid with the resulting lateral illustrates furthermore the way of how microbial micro-
resolution being determined by the distance between two colonies can be reproducibly transferred from solid agar
measuring points in the x-and y-directions, respectively, plates onto IR-transparent substrates (Figure 17b). For
the geometry of the aperture and Abbe’s criterion.(56) this purpose, a replica, or stamping technique has been
In addition to raster scanning sample areas with single- developed which involves spatially accurate transfer of
detector equipment, IR imaging data can be collected the upper cell layers by a specially designed stamping
also by means of FPA multielement detectors. IR imaging device (Figure 17c). For this purpose, round IR trans-
systems with sensitive FPA detectors allow simultaneous parent windows (CaF2 and ZnSe) are glued in Teflon
recording of many individual microscopic spectra from holders. These holders can be gently lowered onto
mostly square sample areas without the need for moving agar plates by means of the stamping device. The
the sample. The simultaneous recording of large numbers device is equipped with a plastic hose that prevents
DOI: 10.1002/9780470027318.a0117.pub2
Video camera
IR microscope
Dry air box
(x,y)-stage
(a) (b) (c) (d)
(x,y)-stage stepping motor controller
Figure 17 FTIR microspectroscopy instrumentation used in the laboratory of the authors for characterizing microbial
microcolonies. (a) Infrared microscope IRscope II from Bruker Optics, Germany connected to an IFS28/B FTIR spectrometer
(Bruker Optics, Germany). The microscope is equipped with sensitive single element MCT detectors, a software-controlled (x,y)
stage, video camera equipment, and a purge box for flushing the microscope’s optical compartment with dry air. (b) Teflon holders
with glued IR windows (CaF2 ) which are transparent in the IR showing spatially accurately transferrect imprints from microbial
microcolonies (see enlargement) (c) Stamping device for transferring microcolony material directly from agar plates onto IR
compatible material (CaF2 , ZnSe). The device consists of a holder with a distance scale which can be manually lowered by means of
a lever, a plastic tube with a device for fastening (d) Details of (c) plastic tube with a device for fastening the Teflon holders.
twisting or laterally displacement of the windows during characterize colony growth. The combined use of both,
stamping. To sample microbial cells for FTIR microspec- the light-microscopic and the FTIR microspectroscopic
troscopy, the following experimental protocol has been information, may strongly enhance the potentials of
established(25,26,57 – 60) : Aliquots of the microbial cell FTIR spectroscopy as applied to microorganisms, which,
suspension, sufficiently diluted to guarantee single colony however, will only be realized when imaging technologies
growth on solid agar plates, are plated and incubated over and multivariate pattern recognition techniques are used
a period of 6–8 h. After these growing times, microbial in tandem. In this way, the fundamental tasks of micro-
microcolonies have diameters between 30 and 150 μm and biological analysis, namely detection, differentiation, and
are generally not yet visible to naked eyes. Microcolony identification of microorganisms, can be integrated in
imprints are subsequently produced using the stamping one single apparatus. Using this methodology, various
device shown in Figure 17(c). The microcolony imprints examples of how FTIR microspectroscopy of microbial
are allowed to dry and are then transferred to the FTIR microcolonies can be used to rapidly and sensitively clas-
microspectrometer. Dried sample spots are measured sify and identify unknown microorganisms have been
automatically, or operator-controlled, by means of video published.(25,26,57 – 60)
techniques for specifying regions of interest and for A promising example of how the IR microspectroscopic
documentation purposes. In addition, the number of imaging technique in combination with sophisticated
colony spots can be counted and size classification of pattern recognition methodologies can be used to identify
these spots is also possible. The information accessible different bacteria in a mixed culture is shown in
from the light-microscopic data such as the number of Figure 18. The left column of this Figure 18(a) shows
colonies, their size, shape, and topography and from FTIR microphotographs of stamping imprints obtained from a
microspectroscopy (cell composition, structural data, and mixed culture of Escherichia coli, Bacillus subtilis, and
type-specific FTIR fingerprints) can be used as input Staphylococcus aureus (cultivation time: 8 h; cultivation
data to differentiate and classify microorganisms even temperature: 37◦ C). The size of the sample area is 700
from mixed cultures (<103 cells per colony spot) and to × 700 μm2 . FTIR microspectra were recorded from the
DOI: 10.1002/9780470027318.a0117.pub2
Microscope image Amide I intensity ANN image was approximately 12 μm at 1000 cm−1 . Chemical images
1
reassembled from the unprocessed raw spectral data in the
1 1
amide I regions (integrated area between 1620 and 1690
3
cm−1 ) are shown in the middle column of Figure 18(a). In
3 3
these images, red regions encode sample areas onto which
Intensity
relatively large amounts of microbial material has been
High Low transferred, while blue marked areas display regions with
no or only very few microbial cells. The right column of
3 3 3
Figure 18(a) illustrates the results of spectral fingerprint
2 2 2 image analysis by means of a multilayer perceptron
ANN. For this purpose, an ANN model was trained
and validated using preprocessed microbial reference
microspectra from preparations of E. coli, B. subtilis,
(a) 1- E. coli, 2 - B. subtilis, 3 - S. aureus and S. aureus and from empty sample areas. The outputs
of independent testing the ANN with spectral imaging
data sets are class membership functions that contain the
Absorbance
classification results for all individual pixel spectra. False

color ANN images can be obtained by combining these
functions with the available spatial information: for ANN
imaging, or segmentation, color pixels are plotted at the
E.coli spatial coordinates at which the individual pixel spectra
RKI A139
were measured. The resulting segmentation maps thus
encode the spatial distribution of taxon-specific spectral
B. subtilis patterns. In the example of Figure 18a, right column,
RKI w2gr
blue colored regions 3 illustrate the presence of spectra
S. aureus
with the typical features of S. aureus, while red areas (1)
DSM 20231 encode regions where spectra of E. coli are found. Black
colored pixels denote regions with unclassifiable spectra.
3000 2500 2000 1500 1000 For a comparison, Figure 18(b) shows furthermore the
(b) Wavenumber (cm−1) intensity-normalized IR spectra of the three pathogens.
Although the spectral differences are comparably small,
Figure 18 Characterization of microcolony imprints using IR these differences can be amplified by an adequate spectral
microspectroscopic imaging data collected by means of a 128 preprocessing for a robust and reliable differentiation
× 128 focal plane array (FPA) IR imaging system, chemical and identification of the three microbial strains under
imaging and artificial neural network analysis. (a) left column: study. Currently, we are systematically examining larger
light microscopy images of microcolony imprints from mixed
cultures of Escherichia coli, Bacillus subtilis, and Staphylococcus spectral data sets to prove the hypothesis that spectra
aureus. Sample area: 700 × 700 μm2. ; central column: chemical obtained from microbial microcolonies can be used to
images obtained from FPA measurements by employing the differentiate, classify, and identify very small amounts of
amide I band intensity in the spectral region between 1620 microorganisms at the genus, species, and at the strain
and 1690 cm−1 (from unprocessed raw spectra); right column: level applying FTIR imaging approaches.(61,62)
an IR image segmentation approach using analysis of FTIR
spectral fingerprints by artificial neural networks (from pre-
processed spectra, see text for details). (b) Representative 7 PERSPECTIVES
intensity-normalized IR microspectra from strains of E. coli,
B. subtilis, and S. aureus. The color-coding is equivalent to The main advantages of IR spectroscopy which make
that used in the ANN segmentation approach. For more clarity it attractive are the extreme rapidity compared to
spectra were shifted along the absorbance axis.
conventional techniques, uniform applicability to very
diverse microorganisms, and a high specificity that allows
differentiations even down to subspecies level. The IR
two sample areas shown by employing an Agilent Cary methodology requires only low amounts of consumables,
620 FPA microspectroscopic imaging system (Agilent is computer compatible, and may thus promote results
Technologies, Santa Clara, USA), which is equipped and databases via data nets. The strength of the
with a 128 × 128 multielement MCT FPA detector. IR technique is its ability to conduct epidemiological
The pixel size of each single detector element was about case studies and large screening experiments very
5.5 μm, whereas the effective lateral spatial resolution quickly. Additional fields of application are the detection
DOI: 10.1002/9780470027318.a0117.pub2
of infection chains and the control of therapy, the The persons who have contributed to this work in the
maintenance of strain collections, and the differentiation laboratory of the author can be taken from the original
of microorganisms from the environment for which literature and from the acknowledgments given therein.
established systems are not yet available. In the food, Particularly acknowledged are Dieter Helm, Christian
water, and pharmaceutical industries, IR spectroscopy Schultz, Vesna Fijala, Siedy Sällström-Baum, Anh Ngo-
may contribute to improve microbiological quality Thi, and Carolin Kirschner. The assistance of Maren
control. For the control of biotechnological processes, it is Stämmler in many sample preparations, measurements
an alternative or additional technique to already existing and evaluations is much appreciated.
analytical tools. Drawing upon the knowledge obtained
to date, the serial type of a dedicated instrument for IR
characterizations of microorganisms is now available on
the market. ABBREVIATIONS AND ACRONYMS
New application possibilities in microbiology can be
expected from the use FTIR microspectroscopy. This new A/R absorbance/reflectance
technology will help not only to scale down the number of ANN artificial neural network
cells needed to a few hundred but also to analyze mixed ART adaptive resonance theory
cultures and to detect light-microscopic and spectro- ATR attenuated total reflectance
scopic features of microorganisms simultaneously. FTIR CNS Coagulase negative staphylococci
microspectroscopy may also be helpful to investigate the DPA dipicolinic acid
phenotypic heterogeneity e.g. within larger colonies, fruit DR Diffuse reflectance
bodies, or even in developing biofilms.(63) Prospectively, DTGS deuterated triglycine sulfate
FTIR microspectroscopy combined with FPA detectors FIR far-infrared
and microarray printing of microbial samples known from FPA focal plane array
DNA technologies will further scale down the number of FT Fourier-transform
cells needed for analysis and help to investigate mixed FTIR Fourier-transform infrared
cultures and perform population analysis.(61,62) HCA hierarchical cluster analysis
The sampling techniques, spectroscopic procedures, IR Infrared
and data evaluation strategies elaborated in the context IRE internal reflection element
of microbiological FTIR spectroscopy can be easily LPS lipopolysaccharide
carried over to characterize other microorganisms such as MCT mercury-cadmium-telluride
amoebae, and even plant or mammalian cells and tissues. MIR mid-infrared
Information density can be increased by combining MS mass spectroscopy
the spectral traits accessible from IR and Raman MSA multivariate statistical analysis
spectroscopy of microorganisms. The exploitation of the NIR near-infrared
complementarity of IR and Raman spectroscopy may PHB Poly-β-hydroxybutyrate
thus open up new avenues for biomedical applications PHF Poly-β-hydroxy fatty acid
in the future. As FTIR and FT-NIR Raman spectra of SNR signal-to-noise ratio
nearly identical samples of intact microorganisms can now SNV standard normal variate
be obtained with excellent reproducibilities, both FTIR SOM self-organizing map
and FT-NIR Raman spectroscopy can be combined for SVM support vector machine
the characterization of complex biological samples. A
particularly intriguing challenge in this context will be
the elaboration of user-friendly computer-based pattern-
recognition software, allowing the analysis of covariance RELATED ARTICLES
patterns in such complex mixed data sets, data reduction,
effective feature extraction, and optimal classification Biomedical Spectroscopy (Volume 1)
results useful for practical applications. Biomedical Spectroscopy: Introduction • Infrared Spec-
troscopy of Biological Applications: An Overview
Peptides and Proteins (Volume 7)
Fourier Transform Infrared Spectroscopy in Peptide and
ACKNOWLEDGMENTS
Protein Analysis
The excellent technical assistance of Angelika Brauer Chemometrics (Volume 11)
in preparing the manuscript is gratefully acknowledged. Clustering and Classification of Analytical Data
DOI: 10.1002/9780470027318.a0117.pub2
Infrared Spectroscopy (Volume 12) 10. C. Kirschner, K. Maquelin, P. Pina, N.A. Ngo Thi, L.-
Infrared Spectroscopy: Introduction • Spectral Data, P. Choo-Smith, G.D. Sockalingum, C. Sandt, D. Ami,
Modern Classification Methods for • Interpretation of F. Orsini, S.M. Doglia, P. Allouch, M. Mainfait, G.J.
Infrared Spectra, A Practical Approach • Theory of Puppels, D. Naumann, ‘Classification and Identification
Infrared Spectroscopy • Microspectroscopy • Spectral of Enterococci: A Comparative Phenotypic, Genotypic
Databases, Infrared and Vibrational Spectroscopic Study’, J. Clin. Microbiol.,
39, 1763–1770 (2001).
Raman Spectroscopy (Volume 15)
11. K. Maquelin, C. Kirschner, L.-P. Choo-Smith, N. van
Raman Spectroscopy: Introduction • Fourier Transform den Braak, H.P. Endtz, D. Naumann, G.J. Puppels,
Raman Instrumentation ‘Identification of Medically Relevant Microorganisms by
Vibrational Spectroscopy’, J. Microbiol. Methods, 51,
255–271 (2002).
REFERENCES
12. K. Maquelin, C. Kirschner, L.-P. Choo-Smith, N.A. Ngo-
Thi, T. van Vreeswijk, M. Stämmler, H.P. Endtz, H.A.
1. W.H. Nelson, Modern Techniques for Rapid Microbiolog-
Bruining, D. Naumann, G.J. Puppels, ‘Prospective Study
ical Analysis, VCH Publishers, New York, (1991).
of the Performance of Vibrational Spectroscopies for
2. P.R. Murray, E.J. Baron, J.H. Jorgensen, M.L. Landry, Rapid Identification of Bacterial and Fungal Pathogens
M.A. Pfaller (eds), ‘Manual of Clinical Microbiology’, Recovered from Blood Cultures’, J. Clin. Microbiol., 41,
in Section 3: Diagnostic Technologies in Clinical Micro- 324–329 (2003).
biology, 9th edition, American Society of Microbiology,
13. V. Shapaval, T. Møretrø, H.P. Suso, A.W. Asli, J.
Washington, DC, 2007.
Schmitt, D. Lillehaug, H. Martens, U. Böcker, A.
3. K.P. Norris, ‘Infra-red Spectroscopy and its Application to Kohler, ‘A High-throughput Microcultivation Protocol for
Microbiology’, J. Hyg., 57, 326–345 (1959). FTIR Spectroscopic Characterization and Identification of
4. J.W. Riddle, P.W. Kabler, B.A. Kenner, R.H. Bordner, Fungi’, J. Biophotonics, 3, 512–521 (2010).
S.W. Rockwood, H.J.R. Stevenson, ‘Bacterial Identifi- 14. V. Shapaval, B. Walczak, S. Gognies, T. Møretrø, H.P.
cation by Infrared Spectrophotometry’, J. Bacteriol., 72, Suso, I. Skaar, A.W. Åsli, A. Belarbi, A. Kohler, ‘FTIR
593–603 (1956). Spectroscopic Characterization of Differently Cultivated
5. D. Helm, H. Labischinski, G. Schallehn, D. Naumann, Food Related Yeasts’, Analyst, 138, 4129–4138 (2013).
‘Classification and Identification of Bacteria by Fourier- 15. V. Shapaval, T. Møretrø, H.P. Suso, J. Schmitt, D.
Transform Infrared Spectroscopy’, J. Gen. Microbiol., 137, Lillehaug, A. Kohler, A Novel Library-independent
69–79 (1991). Approach Based on FTIR Spectroscopy for Source
6. D. Helm, H. Labischinski, D. Naumann, ‘Elaboration Tracking of Moulds Contamination in Food Industry, Lett.
of a Procedure for Identification of Bacteria Using Appl. Microbiol. (Under Review).
Fourier-Transform Infrared Spectral Libraries: A Stepwise 16. A. Kohler, U. Böcker, V. Shapaval, A. Forsmark, M.
Correlation Approach’, J. Microbiol. Methods, 14, 127–142 Andersson, J. Warringer, H. Martens, S.W. Omholt, A.
(1991). Blomberg, High-throughput Biochemical Fingerprinting
7. D. Naumann, D. Helm, H. Labischinski, ‘Microbiological of Saccharomyces cervisiae by Fourier Transform Infrared
Characterizations by FT-IR Spectroscopy’, Nature, 351, Spectroscopy, Plos ONE (Submitted).
81–82 (1991). 17. B. Schrader (ed), Infrared and Raman Spectroscopy, VCH
8. D. Naumann, ‘FT-IR of Microorganisms at the Robert Publishers, New York, 1995.
Koch-Institute: Experiences Gained During a Successful 18. P.R. Griffiths, J.A. de Haseth, Fourier Transform Infrared
Project’, in Biomedical Vibrational Spectroscopy V: Spectrometry, 2nd edition, John Wiley & Sons, Inc.,
Advances in Research and Industry, Proceedings of Hoboken, NJ, 2007.
SPIE, eds A. Mahadevan-Jansen, W.H. Petrich, 68530G-
19. G. Seltmann, O. Holst, The Bacterial Cell Wall, Springer-
1–68530G-12, Vol. 6853, 2008.
Verlag Berlin Heidelberg, New York, (2002).
9. L.-P. Choo-Smith, K. Maquelin, T. van Vreeswijk, H.A.
20. P.J. Brennan, H. Nikaido, ‘The Envelope of Mycobacteria’,
Bruining, G. Puppels, N.A. Ngo Thi, C. Kirschner,
Annu. Rev. Biochem., 64, 29–63 (1995).
D. Naumann, D. Ami, A.M. Villa, F. Orsini, S.M.
Doglia, H. Lamfarraj, G.D. Sockalingum, M. Main- 21. V. Farkas, ‘Polysaccharide Metabolism’, in The Yeasts,
fait, P. Allouch, H.P. Endtz, ‘Investigating Micro- eds A.H. Rose, J.S. Harrison, Academic Press, London,
bial (Micro)colony Heterogeneity by Vibrational Spec- 317–366, Vol. 3, 1989.
troscopy’, Appl. Environ. Microbiol., 67, 1461–1469 22. F.M. Klis, ‘Review: Cell Wall Assembly in Yeast’, Yeast,
(2001). 10, 851–869 (1994).
DOI: 10.1002/9780470027318.a0117.pub2
23. C. Ratledge, C.T. Evans, ‘Lipids and Their Metabolism’, in 38. P. Lasch, D. Naumann, ‘Maldi-TOF Mass Spectroscopy
The Yeasts, eds A.H. Rose, J.S. Harrison, Academic Press, for Rapid Identification of Highly Pathogenic Microorgan-
London, 367–455, Vol. 3, 1989. isms’, in BSL3 and BSL4 Agents, Proteomics, Glycomics
24. J. Coates, ‘Interpretation of Infrared Spectra, a Practical and Antigenicity, eds J. Stulik, R. Toman, P. Butaye, R.G.
Approach’, in Encyclopedia of Analytical Chemistry, ed. Ulrich, Wiley-VCH Verlag GmbH, Weinheim, 179–212,
R.A. Meyers, John Wiley & Sons, Ltd, Chichester, 2011.
10815–10837, 2000. 39. C.A. Rebuffo-Scheer, C. Kirschner, M. Stämmler, D.
25. D. Naumann, H. Labischinski, P. Giesbrecht, ‘The Naumann, ‘Rapid Species and Strain Differentiation
Characterization of Microorganisms by Fourier-Transform of Non-tuberculous Mycobacteria by Fourier-Transform
Infrared Spectroscopy (FT-IR)’, in Modern Techniques for Infrared Microspectroscopy’, J. Microbiol. Methods, 68,
Rapid Microbiological Analysis, ed. W.H. Nelson, VCH 282–290 (2007).
Publishers, New York, 43–96, 1991. 40. L. Beutin, Q. Wang, D. Naumann, W. Han, G. Krause,
26. D. Naumann, H. Fabian, P. Lasch, FTIR Spectroscopy L. Leomil, L. Wang, L. Feng, ‘Relationship Between O
of Cells, Tissues and Body Fluids, in Biological and Antigen Subtypes, Bacterial Surface Structures and O
Biomedical Infrared Spectroscopy, eds A. Barth, P.I. Haris, Antigen Gene Clusters in Escherichia coli O123 Strains
IOS Press, pp. 312-354. Carrying Genes for Shiga-toxins and Intimin’, J. Med.
Microbiol., 56, 177–184 (2007).
27. M.K. Winson, R. Goodacre, É.M. Timmins, A. Jones,
B.J. Alsberg, A.M. Woodward, J.J. Rowland, D.B. Kell, 41. T. Grunert, M. Wenning, M.S. Barbagelata, M. Fricker,
‘Diffuse Reflectance Absorbance Spectroscopy Taking D.O. Sordelli, F.R. Buzzola, M. Ehling-Schulz, ‘Rapid and
in Chemometrics (DRASTIC). A Hyperspectral FT- Reliable Identification of Staphylococcus aureus Capsular
IR-Based Approach to Rapid Screening for Metabolite Serotypes by Means of Artificial Neural Network-
Overproduction’, Anal. Chim. Acta, 348, 273–282 (1997). assisted Fourier Transform Infrared Spectroscopy’, J. Clin.
Microbiol., 51, 2261–2266 (2013).
28. R. Bhargava, ‘Infrared Spectroscopy Imaging: the Next
Generation’, Appl. Spectrosc., 66, 1091–1120 (2012). 42. N.M. Amiali, M.R. Mulvey, J. Sedman, A.E. Simor, A.A.
Ismail, ‘Epidemiological Typing of Methicillin-resistant
29. B.F.J. Manly, Multivariate Statistical Methods, a Primer,
Staphylococcus aureus by Fourier Transform Infrared
Chapman & Hall, New York, 1986.
Spectroscopy’, J. Microbiol. Methods, 69, 146–153 (2007).
30. B.S. Everitt, Cluster Analysis, John Wiley & Sons, Ltd,
43. R. Davis, L.J. Mauer, ‘Subtyping of Listeria monocytogenes
Toronto, 1993.
at the Haplotype Level by Fourier Transform Infrared (FT-
31. J.E. Dayhoff, Neural Network Architectures, Nostrand IR) Spectroscopy and Multivariate Statistical Analysis’,
Reinhold, New York, 1990. Int. J. Food Microbiol., 150, 140–149 (2011).
32. J. Zupan, J. Gasteiger, Neural Networks for Chemists, 44. C.A. Rebuffo, J. Schmitt, M. Wenning, F. von Stetten,
Weinheim, VCH Publishers, 1993. S. Scherer, ‘Reliable and Rapid Identification of Listeria
33. P. Lasch, Spectral Pre-processing for Biomedical Vibra- monocytogenes and Listeria Species by Artificial Neural
tional Spectroscopy and Microspectroscopic Imaging, Network-based Fourier Transform Infrared Spectroscopy’,
Chemom. Intell. Lab. Syst., 117, 100-114 (2012). Appl. Environ. Microbiol., 72, 994–1000 (2006).
34. A. Savitzky, M. Golay, ‘Smoothing and Differentiation 45. M. Kümmerle, S. Scherer, H. Seiler, ‘Rapid and Reliable
of Data by Simplified Least Squares Procedures’, Anal. Identification of Food-Borne Yeasts by Fourier-Transform
Chem., 36, 1627 (1964). Infrared Spectroscopy’, Appl. Environ. Microbiol., 64,
35. P. Lasch, W. Petrich, ‘Data Acquisition and Analysis 2207–2214 (1998).
in Biomedical Vibrational Spectroscopy’, in Biomedical 46. H. Oberreuter, H. Seiler, S. Scherer, ‘Identification of
Applications of Synchrotron Infrared Microspectroscopy: Coryneform Bacteria and Related Taxa by Fourier-
A Practical Approach, RSC Analytical Spectroscopy transform Infrared (FT-IR) Spectroscopy’, Int. J. Syst.
Monographs No 11, ed. D. Moss, The Royal Society of Evol. Microbiol., 52, 91–100 (2002).
Chemistry, Cambridge, 192–225, 2011. 47. M. Wenning, V. Theilmann, S. Scherer, ‘Rapid Analysis
36. D. Helm, D. Naumann, ‘Identification of Some Bacterial of Two Food-borne Microbial Communities at the Species
Cell Components by FT-IR Spectroscopy’, FEMS Micro- Level by Fourier-transform Infrared Microspectroscopy’,
biol. Lett., 126, 75–80 (1995). Environ. Microbiol., 8, 848–857 (2006).
37. C.P. Schultz, H.H. Eysel, H.H. Mantsch, M. Jackson, 48. M. Wenning, S. Scherer, ‘Identification of Microorganisms
‘Carbon Dioxide in Tissues, Cells, and Biological Fluids by FTIR Spectroscopy: Perspectives and Limitations of
Detected by FTIR Spectroscopy’, J. Phys. Chem., 100, the Method’, Appl. Microbiol. Biotechnol., 97, 7111–7120
6845–6848 (1996). (2013).
DOI: 10.1002/9780470027318.a0117.pub2
49. M. Wenning, S. Scherer, D. Naumann, ‘Infrared Spec- 57. N.A. Ngo Thi, C. Kirschner, D. Naumann, ‘FT-IR
troscopy in the Identification of Microorganisms’, in Microspectrometry: A New Tool for Characterizing
Vibrational Spectroscopy for Medical Diagnosis, eds M. Micro-organisms’, in Biomedical Spectroscopy: Vibrational
Diem, P.R. Griffiths, J. Chalmers, John Wiley & Sons, Ltd, Spectroscopy and Other Novel Techniques, Proceedings of
71–96, 2008. SPIE, eds A. Mahadevan-Jansen, G.J. Puppels, 36–44,
50. T. Udelhoven, D. Naumann, J. Schmitt, ‘Development of Vol. 3918, (2000). doi: 10.1117/12.384961.
a Hierarchical Classification System with Artificial Neural 58. M. Wenning, H. Seiler, S. Scherer, ‘Fourier-transform
Networks and FT-IR Spectra for the Identification of Infrared Microspectroscopy, a Novel and Rapid Tool for
Bacteria’, Appl. Spectrosc., 54, 1471–1479 (2000). Identification of Yeasts’, Appl. Environ. Microbiol., 68,
51. T. Udelhoven, M. Novozhilov, J. Schmitt, ‘The NeuroDe- 4717–4721 (2002).
veloper: A Tool for Modular Neural Classification of 59. N.A. Ngo Thi, C. Kirschner, D. Naumann, ‘Character-
Spectroscopic Data’, Chemom. Intell. Lab. Syst., 66, ization and Identification of Microorganisms by FT-IR
219–226 (2003). Microspectrometry’, J. Mol. Struct., 661-662, 371–380
52. M. Wenning, N.R. Büchl, S. Scherer, ‘Species and (2003).
Strain Identification of Lactic Acid Bacteria Using 60. C.A. Rebuffo-Scheer, J. Dietrich, M. Wenning, S. Scherer,
FTIR Spectroscopy and Artificial Neural Networks’, J. ‘Identification of Five Listeria Species Based on Infrared
Biophotonics, 3, 493–505 (2010). Spectra (FTIR) Using Macrosamples Is Superior to
53. N.R. Büchl, M. Wenning, H. Seiler, H. Mietke-Hofmann, a Microsample Approach’, Anal. Bioanal. Chem., 390,
S. Scherer, ‘Reliable Identification of Closely Related 1629–1635 (2008).
Issatchenkia and Pichia Species Using Artificial Neural 61. L.C. Carranza, P.A. Alvarez, A. Ghetler, I. Iugovaz, J.
Network Analysis of Fourier-transform Infrared Spectra’, Sedman, C.D. Carrillo, A.A. Ismail, ‘Evaluation of FPA-
Yeast, 25, 787–798 (2008). FTIR Spectroscopy as a Tool in The Differentiation of
54. P.R. Griffiths, ‘Infrared and Raman Instrumentation Campylobacter jejuni from Campylobacter coli Isolated
for Mapping and Imaging’, in Infrared and Raman from Retail Chicken Samples’, Journal of Food Safety,
Spectroscopic Imaging, eds R. Salzer, H.W. Siesler, 32(3), 289–295 (2012).
Wiley-VCH Verlag GmbH, Weinheim, (2009). doi: 62. J. Kirkwood, S.F. Al-Khaldi, M.M. Mossoba, J. Sedman,
10.1002/9783527628230.ch1. A.A. Ismail, ‘Fourier Transform Infrared Bacteria Iden-
55. R.G. Messerschmidt, M.A. Harthcock (eds), Infrared tification with the Use of a Focal-Plane-Array Detector
Microspectroscopy, Marcel Dekker, Inc., New York, Basel, and Microarray Printing’, Appl. Spectrosc., 58, 1364–1368
1988. (2004).
56. P. Lasch, D. Naumann, ‘Spatial Resolution in Infrared 63. N.A. Ngo Thi, D. Naumann, ‘Investigating the Hetero-
Microspectroscopic Imaging of Tissues’, Biochim. Biophys. geneity of Cell Growth in Microbial Colonies by FTIR
Acta, 1758, 814–829 (2006). Microspectroscopy’, Anal. Bioanal. Chem., 387, 1767–1777
(2007).
DOI: 10.1002/9780470027318.a0117.pub2

Andres Reyes - Equipo 3 - Theory and Instrumentation) - Infrared Spectroscopy in Microbiology

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Andres Reyes - Equipo 3 - Theory and Instrumentation) - Infrared Spectroscopy in Microbiology

Caricato da

Copyright:

Formati disponibili

Infrared Spectroscopy in Infrared (IR) spectra of intact microbial cells are highly

specific, fingerprint-like signatures that can be used to

1.3 Infrared Spectroscopy

Wavelength 104 102 100 10−2 10−4 10−6 10−8

Wavenumber 1 10 100 1000 10000 cm−1

Frequency 108 1010 1012 1014 1016 1018 s−1

Figure 1 The electromagnetic spectrum.

IR source Interferometer Sample Detector PC

Interferogram, single-channel background Interferogram single-channel sample

4000 3000 2000 1000 4000 3000 2000 1000

4000 3000 2000 1000

Table 1 Gross composition of pro- and eukaryotic cells

2.2 Yeast Cells

specific structures are still a too complex task at present,

with ions, etc., has a strong influence on results. These

1. The region between 4000 and 3100 cm−1 is dominated

Table 3 Optical properties of the most important, for IR spectroscopy materials

hydrated microbial cells and tissue material and is 4 EXPERIMENTAL METHODOLOGIES

Figure 6 The high-throughput multiplate extension (HTS- (a) 5

4.1.2 Absorbance/Reflectance Measurements

A/R measurements (sometimes also called ‘trans-

Single point CCD camera

Pseudomonas aeruginosa 2. Derivation

Candida albicans ATCC 10231

3000 2500 2000 1500 1000 500 3000 2000 1000

outer membrane. In spectra obtained from Legionella, (a)

6.1.2 Endospore Formation in Cells Figure 11 IR-spectra of a Legionella pneumophila strain

2 (a) Wavenumber (cm−1)

Wavenumber (cm−1) 2365 2355 2345 2335 2325 2315

Myc. kansasii Myc. fortuitum

E. coli E. cloacae S. aureus CNS E. faecalis E. faecium Streptococcus

Dry air box

(a) (b) (c) (d)

(x,y)-stage stepping motor controller

classification results for all individual pixel spectra. False

Potrebbero piacerti anche