Anal Bioanal Chem (2009) 395:829–837

DOI 10.1007/s00216-009-3019-y

ORIGINAL PAPER

Collagen types analysis and differentiation

by FTIR spectroscopy
Karima Belbachir & Razia Noreen & Gilles Gouspillou &
Cyril Petibois

Received: 7 May 2009 / Revised: 17 July 2009 / Accepted: 29 July 2009 / Published online: 16 August 2009
# Springer-Verlag 2009

Abstract Abnormal formation and organization of colla-
gen network is commonly observed in many organ
pathologies, but analytical techniques able to reveal the
collagen biodistribution are still lacking. In this study,
Fourier-transform infrared (FTIR) spectroscopy has been
used to analyze type I, III, IV, V, and VI collagens, the most
important compounds of connective tissues. A robust
classification of 30 FTIR spectra per collagen type could
be obtained by using a combination of four spectral
intervals [ν(C=O) absorption of amide I (1,700–
1,600 cm−1), δ(CH2), and δ(CH3) absorptions (1,480–
1,350 cm−1), ν(C–N), and δ(N–H) absorptions of amide
III (1,300–1,180 cm−1), and ν(C–O) and ν(C–O–C)
absorptions of carbohydrate moieties (1,100–1,005 cm−1)].
Then, a submolecular justification of this classification
model was sought using a curve fitting analysis of the four
spectral intervals. Results demonstrated that every spectral
interval used for the classification contained highly dis-
criminant absorption bands between all collagen types
(multivariate analysis of variance, p<0.01; Dunnett's T3
post hoc test, p<0.05). All conditions seem thus joined to
make FTIR spectroscopy and imaging major tools for
implementing innovative methods in the field of molecular
Electronic supplementary material The online version of this article
(doi:10.1007/s00216-009-3019-y) contains supplementary material,
which is available to authorized users.
K. Belbachir : R. Noreen : G. Gouspillou : C. Petibois (*)
Université Victor Segalen Bordeaux 2,
146 rue Léo Saignat,
33076 Bordeaux-Cedex, France
e-mail: cyril.petibois@u-bordeaux2.fr
K. Belbachir : R. Noreen : C. Petibois
CNRS UMR 5248, CBMN,
2 rue Robert Escarpit,
33607 Pessac-Cedex, France
histology, which would be very helpful for the diagnosis of
a wide range of pathologies.
Keywords FTIR spectroscopy . Collagens .
Connective tissue . Molecular structure . Classification
Abbreviations
FTIR Fourier-transform infrared
RMSE Root mean square error
Introduction
Understanding the enormous complexity of the human body
is one of the great challenges facing science; however, a
major simplification is to divide the body into “cells” and
“extracellular matrix.” Cells perform the basic processes of
life including generating energy, transporting oxygen, and
making proteins, while the extracellular matrix (ECM)
provides the scaffolding to house the cells themselves and
also their support network of blood and lymph vessels.
Likewise, many painful and debilitating medical conditions
have their origin not in a malfunction of the cells but rather
in defects of the ECM. This is an insoluble macromolecular
network, whose structure varies with organ. However,
different ECMs comprise the same few types of macro-
molecules (mostly collagen, elastin, proteoglycans) plus
water (65%). ECM does not migrate, proliferate, synthesize
proteins, or even contain DNA. Although this field initially
focused on the more static structural nature of the matrix and
the adhesive properties of cell–cell and cell substratum
receptors, the past 10 years have seen an explosion in our
understanding of the dynamic nature of both the matrix and
receptor functions. Now, it is recognized that it gives and
830
takes of signals with cells and partly determines the state of
differentiation of cells. ECM is composed chiefly of
collagen, the body's most abundant structural protein and a
very strong “biopolymer,” the network organization of which
takes the form of long and cross-linked fibers that give
tissues such as skin and cartilage their tensile strength. There
are about 30 types of collagen types actually referenced,
which vary in amount in each of the body's tissues.
Every collagen type consists of three polypeptide chains,
each one composed of at least one Gly–X–Y sequence
structured in left-handed -like helices and where the X and
Y positions are often proline and hydroxyproline, respec-
tively. These three -like helices are organized together to
form the characteristic structure of collagens, a right-handed
triple helix [1]. However, distinctive features also character-
ize every collagen type. For example, type VI collagen is
made of one short central triple helical domain flanked by
large N- and C-terminal globular domains [2], whereas type I
collagen is mainly composed of a triple helical region
flanked by two small N- and C-telopeptides (Fig. 1).
Abnormal formation and organization of collagen net-
work is commonly observed in several pathologies such as
myopathies [2, 3], liver fibrosis [4], cardiac diseases [5], and
Ehlers–Danlos syndrome [6]. However, only a few analytical
methods are able to provide information about collagen type
biodistribution and assembly. In vivo imaging techniques,
such as MRI, may be used to study connective tissues, but
the spatial resolution achieved is not sufficient for determin-
ing the collagen networks organization [7]. Among the
available ex vivo analytical techniques, immunohistochem-
Fig. 1 Schematic representation
of main structural differences
between the most abundant col-
lagen types of extracellular ma-
trix in human tissues
K. Belbachir et al.
istry is the most widely used technique to study the
biodistribution of collagen types [8, 9]. However, this
technique suffers from the few parameters that may be
analyzed at the same time since antibody compatibility and
specificity are limited. Thus, alterations of collagen type
networks in organ pathologies cannot be completely studied
at present. There is thus a real need for implementing a
molecular imaging technique allowing the study and the fast
diagnosis of pathologies involving collagens type alterations.
Since last decade, Fourier-transform infrared (FTIR)
spectroscopy and imaging emerged and developed rapidly
in the ex vivo diagnostic field [10]. The FTIR technique is
based upon the absorption of IR radiation by vibrational
transitions in covalent bonds of the biomolecules in
presence. The intensities of IR absorptions provide quan-
titative information about the sample contents, depending
on the nature of the molecular bonds, their structure, and
their environment. In complex systems, such as biological
samples, the FTIR spectrum is the sum of the contributions
gathered mainly from the proteins, lipids, nucleic acids, and
carbohydrates. It has been shown that IR spectra provide
useful diagnostic information in the case of different
pathologies, such as tumors [11–13], multiple sclerosis
[14], and osteoarthritis [15]. Another advantage of the FTIR
spectroscopy approach is that a spectrum may be obtained
within a few seconds [16], and with the recent development
of FTIR spectroscopic imaging systems having multiple
detectors, the analysis of tissues became possible. Only a
few minutes are now required to obtain a high-quality and
functional FTIR image of a 1-mm2 area from a tissue
Collagens analysis by FTIR spectroscopy
section. Moreover, in a preliminary study, we demonstrated
that this technique allows the differentiation of types I and
IV collagens by their secondary structure parameters and
that these connective tissue components could be detected
in a biological sample by FTIR spectroscopic imaging [17].
Therefore, as a molecular probe of tissue composition,
FTIR imaging may favorably help histopathology in
detecting and diagnosing diseases in collagen-rich tissues
[15, 18, 19].
The aim of this study was to show that FTIR spectroscopy
is a useful analytical technique to differentiate several of the
most abundant collagen types found in connective tissues,
taking also into consideration the different structures found
in collagens (fibrillar or glomerular, glycosylated, N- and C-
terminal domains, disulfide bonds…etc. Fig. 1). A classifi-
cation method is first proposed to differentiate types I, III, IV,
V, and VI collagens. The IR spectral intervals used for this
classification were then studied to seek for the most
discriminant submolecular parameters that allowed this
collagen type differentiation.
Experimental procedures
FTIR spectra of collagens
All the collagen types used in this study originated from
human placenta. Pure products of types I, III, and IV
collagens were purchased as lyophilized powder (Sigma-
Aldrich, Ref. C7774, C4407, and C5533, respectively).
Types V (BD Biosciences, Ref. 354246) and VI (Tebu-Bio,
Ref. 009-001-108) collagens with 95% purity threshold
were purchased as solubilized in acetic acid solution. Types
V and VI collagens were first lyophilized for 24 h (Heto
Power Dry LL1500, Thermo Electron Corp., France) before
a further 24-h desiccation in presence of KOH in a drying
vacuum to evaporate residual acetic acid. All dry collagen
samples were further gelled in water at 37°C for 1 h, and
pH was verified to be neutral. Every collagen gel was laid
down flat on three ZnSe windows using 5µl of water in
order to obtain homogenous films with adequate thickness
for transmission measurements by FTIR spectroscopy. Then
the samples were desiccated again in a drying vacuum to
evaporate water (2 mmHg, 30 min). All spectra were
obtained using a Spotlight 300 FTIR imaging system
equipped with a Spectrum One spectrometer (Perkin-Elmer,
France) and a mercury cadmium telluride detector. For
FTIR spectra acquisition, the system was used in point
mode (aperture of 250×250 μm) with a 2.0 cm−1 resolution
and using 30 scans in transmittance mode (n=35 spectra
per collagen type from the three ZnSe windows), and
spectra acquisitions were performed under complete N2
purge of the analytical system.
831
FTIR spectra classification
FTIR spectra classification was performed using a subrou-
tine of Opus 4.2 software (Bruker, Germany) with Ward's
algorithm [20], which is used to find the most homoge-
neous groups between spectra with the smallest intrinsic
growth in heterogeneity. This heterogeneity is determined
by the calculation of Euclidian distances for all dependent
variables (here, the IR absorption values for each wave
number of the spectral interval used) within spectra. To be
significant, heterogeneity between two clusters must be
above the sum of their heterogeneities. As previously
described [21], spectra classification may be performed on
the whole wave number range (4,000–600 cm−1) but also
on a defined spectral interval or using a combination of
spectral intervals. These data treatments were tested in
order to find a classification model allowing the differen-
tiation of the five collagen types. Thirty of the thirty-five
spectra for each collagen type were randomly selected to be
used for the initial classification method, and the five
remaining spectra were further used to test the classification
model obtained. To remove the global IR absorption
differences between spectra, which are due to sample
thicknesses in spectra acquisitions, and to emphasize
spectral features, vector normalization and first derivative
with a five-point smoothing were used for the classification.
Once a classification model was obtained with the
formation of five clusters, each one containing exclusively
the series of 30 FTIR spectra for a collagen type, five other
FTIR spectra per collagen type were inserted in the
database to test the classification model.
Spectral curve fitting
Curve fitting of FTIR spectra was performed using a
subroutine of Opus 4.2 software. This method was applied
to study the submolecular information, for each collagen
type, contained in every spectral interval used in the
classification model described above. The aim of the curve
fitting process is to study separately the absorption bands of
a sample. Using the Levenberg–Maquardt algorithm, every
absorption band is characterized by its area. The second
derivative spectrum was used to determine the number and
the position (maximal IR intensity wave number) of the
bands constituting every spectral interval used in the
classification model presented above [22]. With this
method, the bands constituting a given spectral interval
are revealed by the successive minima of the second
derivative spectrum. With the aim at standardizing all data
treatments applied on FTIR spectra, the bands revealed
from the study of the second derivative FTIR spectrum of
type I collagen (average of 35 FTIR spectra) were used as
curve fitting models, each one corresponding to one
832
spectral interval, which were further applied on all other
collagen types, thus ensuring a standardization of data
treatments among FTIR spectra of all collagen types. The
spectra used to establish the classification model (n=30 per
collagen type) were analyzed using these curve fitting
models obtained from the averaged type I collagen FTIR
spectrum. All bands were positioned at maximal intensities
revealed by second derivative spectrum, and shape param-
eters were set free to any combination of Gaussian/Lorentz
functions. Once curve fitting models were fixed for type I
collagen spectrum, every band parameter (maximal inten-
sity and shape parameters) were fixed to be applied to all
other spectra. To have correct band area determinations, all
curve fitting were performed on larger spectral intervals
than those used in the classification model, and free bands
were added at the ends of each spectral interval studied to
avoid anomalous band addition. These “external” bands
were not used for differentiation between collagen types.
Spectral curve fitting quality was assessed by a root
mean square error (RMSE) value set at 1% of the total
spectral interval area. RMSE was determined as follows
(Eq. 1):
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u n 2
u1 X
RMSE ¼ t ð ai  pi Þ ð1Þ
n i¼1
where n is the number of points constituting the spectral
interval, a is the true absorbance value, and p is the
predicted absorbance value determined by the Levenberg–
Maquardt algorithm.
Statistics
All results obtained are expressed as mean ± SD.
Comparisons between collagen types were performed using
a multivariate analysis of variance (MANOVA; SPSS 15.0,
SPSS Inc., France) applied on the results obtained from the
spectral curve fitting of every spectral interval used in the
classification model. Multiple comparisons were performed
using a Dunnett's T3 Post Hoc analysis for exhibiting the
significant differences between collagen types. P values
were fixed at 0.05 and 0.01 to consider the significant level
of difference between series of data (details are given in
text).
Results
FTIR spectra of collagens
Mean FTIR spectra of types I, III, IV, V, and VI collagens
are presented in Figs. 2 and 5. All collagen types exhibited
comparable IR absorptions as revealed by the pick peaking
K. Belbachir et al.
Fig. 2 Mean FTIR spectra of the five collagen types. Maximal
intensity absorptions are revealed by a peak-picking with a sensitivity
of 1%
performed on the 1,800–900 cm−1 interval. All collagen
FTIR spectra exhibited absorptions at 1,035 and 1,079 cm−1,
which arise from the ν(C–O) and ν(C–O–C) absorptions of
the carbohydrate moieties [15]. Although a precise absorp-
tion bands assignation is not available for several spectral
intervals of collagen FTIR spectra, absorptions bands at
1,454, 1,403, 1,340, 1,282, 1,240, and 1,205 cm−1 may be
attributed to the δ(CH2), δ(CH3), ν(C–N), and δ(N–H)
absorptions of collagens. Amides I and II absorptions
could be found at 1,659 and 1,555 cm−1, respectively
[23, 24].
FTIR spectra classification
Results of FTIR spectra classification are shown in Fig. 3.
The smallest spectral interval combination allowing the five
collagen types classification required the use of the ν(C=O)
absorptions of the amide I (1,700–1,600 cm−1), δ(CH2) and
δ(CH3) absorptions (1,480–1,350 cm−1), ν(C–N) and δ(N–
H) absorptions of amide III (1,300–1,180 cm−1), and ν(C–
O) and ν(C–O–C) absorptions of carbohydrate moieties
(1,100–1,005 cm−1). Taken separately, the last three regions
allowed classification of collagen types but with smaller
heterogeneity between clusters and higher within clusters,
and no reliable results could be obtained for the amide I
spectral interval. However, with the combination of the four
spectral intervals, the internal heterogeneity within clusters,
each one corresponding to a collagen type, was found very
small (range, 0.4≤h≥0.6), whereas the global heterogeneity
between clusters was high (H=5; always equal or above the
sum of two successive clusters).
In order to test this classification model, five other FTIR
spectra per collagen type were added to the initial spectra
database. As shown in Fig. 4, all test spectra were classified
within their respective collagen type cluster. Heterogeneity
level between clusters was also higher (H=5.8), while
Collagens analysis by FTIR spectroscopy 833

Fig. 3 FTIR spectra classifica-

tion model for the five collagen
types. n=30 spectra/collagen
type; four spectral intervals,
1,700–1,600 cm−1, 1,480–
1,350 cm−1, 1,300–1,180 cm−1,
and 1,100–1,005 cm−1. h intra-
cluster heterogeneity, H inter-
cluster heterogeneity

heterogeneity within clusters remained globally stable
(range: 0.6≤h≥0.8). This result provided evidence that
adding new series of spectra strengthened again the potential
of the classification model. Moreover, the stability of this
classification was not altered whatever the order of intro-
duction of test spectra (added all together or type by type).
Spectral curve fitting
The four spectral intervals used in our classification model
were then studied by curve fitting. The absorption bands of
every spectral interval were determined using the second
derivative spectrum of type I collagen (Fig. 5). The results
obtained from the averaged type I collagen FTIR spectrum
were further used as a model to apply on all FTIR spectra of
the five collagen types. The curve fitting procedure applied
on the 1,720–1,480 cm−1 spectral interval allowed determin-
ing the secondary structure parameters of collagen types as
previously described [17]: -like helix (1,658 cm−1), -sheets
(1,679, and 1,626 cm−1), -turns (1,691 and 1,669 cm−1),
triple helix (1,637 cm−1), side chains (1,608 cm−1), and
unordered structure (1,647 cm−1) [10, 25–27]. Secondary

Fig. 4 Test of the classification

model with five added spectra/
collagen type. h intra-cluster
heterogeneity, H inter-cluster
heterogeneity, “+” type VI, “−”
type V, “*” type III, “¤” type I,
“°” type IV
834 K. Belbachir et al.

Fig. 5 Curve fitting of the

1,720–1,480 cm−1 (a),
1,500–1,300 cm−1 (b), 1,350–
1,150 cm−1 (c), and 1,130–
950 cm−1 (d) spectral intervals.
The bands were determined us-
ing the second derivative spec-
trum of type I collagen. 1
1,692 cm−1; 2 1,681 cm−1; 3
1,669 cm−1; 4 1,658 cm−1; 5
1,647 cm−1; 6 1,637 cm−1; 7
1,626 cm−1; 8 1,610 cm−1; 9
1,465 cm−1; 10 1,452 cm−1; 11
1,440 cm−1; 12 1,424 cm−1; 13
1,405 cm−1; 14 1,390 cm−1; 15
1,375 cm−1; 16 1,346 cm−1; 17
1,284 cm−1; 18 1,260 cm−1; 19
1,243 cm−1; 20 1,234 cm−1; 21
1,203 cm−1; 22 1,081 cm−1; 23
1,065 cm−1; 24 1,050 cm−1; 25
1,033 cm−1; 26 1,022 cm−1; 27
1,012 cm−1

structure parameters were expressed as percentages of the
total amide I absorption (1,700–1,600 cm−1).
Curve fitting was also performed on the 1,500–1,300,
1,350–1,150, and 1,130–950 cm−1 spectral intervals, and
the absorption bands corresponding to the four spectral
intervals used in our classification model were numbered 9
to 16, 17 to 21, and 22 to 27, respectively, which were
further considered for data treatments. The area for every
absorption band was expressed as a percentage of the total
area of its corresponding spectral interval (Fig. 5). For all
curve fittings performed on the four spectral intervals
studied, the RMSE values were lower than 1%.
Statistical tests (MANOVA) performed on curve fitting
results showed that all spectral intervals used for the spectra
classification presented significant differences between the
five collagen types (p<0.01). Dunnett's T3 post hoc tests
allowed to determine the secondary structure parameters
and other absorption bands for collagen types discrimina-
tion (detailed statistical results are available as Electronic
Supplementary Material). Secondary structure parameters
(see Electronic Supplementary Material Table S1) revealed
that types I, III, and V collagens were differentiated from
other types by higher triple helix contents (p<0.05). Type I
could be differentiated from type III by a higher content of
-like helix (p<0.05) and from type V by a lower content
of -turns (p<0.05). Type V collagen presented higher -
turn content than type III (p<0.05). Type IV collagen
contained a higher percentage of -sheets than types I, III,
Collagens analysis by FTIR spectroscopy
and V but also a smaller one than type VI collagen (p<
0.05), which possessed the smallest content of triple helix
(p<0.05).
The curve fitting of the 1,500–1,300 cm−1 spectral
interval (see Electronic Supplementary Material Table S2)
revealed that all collagen types could be differentiated by
the contribution of the 1,452 cm−1 band (p<0.05) except
for types I and IV, which were differentiated by the
contribution of the 1,440 cm−1 band (p<0.05).
The curve fitting of the 1,350–1,150 cm−1 spectral
interval (see Electronic Supplementary Material Table S3)
demonstrated that the 1,234 cm−1 absorption band allowed
the differentiation of all collagen types (p<0.05), except for
types IV and V, which were differentiated by the 1,203 cm−1
absorption band (p<0.05).
The curve fitting of the 1,130–950 cm−1 interval (see
Electronic Supplementary Material Table S4) showed that
the 1,033 cm−1 absorption band allowed the differentiation
of the five collagen types (p<0.05), except for types V and
VI, which were differentiated by the 1,081 cm−1 absorption
band (p<0.05).
Discussion
The aim of this study was to determine which discriminant
submolecular parameters might be used for the analysis by
FTIR spectroscopy of the major collagen types that may be
found in connective tissues. To address this issue, it was
first proposed to use a classification method providing the
discrimination of five collagen types. FTIR spectra of the
types I, III, IV, V, and VI collagens were thus classified
using the complete mid-infrared wave number range
(4,000–600 cm−1), which failed to separate the five types.
This lack of performance in this classification could be
expected since most of IR absorptions in spectra are very
similar between collagen types. The use of a single spectral
interval for which the highest discriminant potential could
be expected, i.e., the areas 1,480–1,350 cm−1, 1,300–
1,200 cm−1, and 1,100–1,005 cm−1, seemed successful for
differentiating all collagen types. However, the use of a
single spectral interval could have been criticized since any
change in the whole spectrum intensity, which is strictly
correlated to the amount of organic contents absorbing the
IR radiation, leads to an absorption intensity increase for
most of spectral regions. This is particularly true for tissue
samples, whose contents (proteins, nucleic acids, lipids, and
sugars) give rise to major absorptions in the mid-infrared
interval. As a consequence, there is evidence that a
classification model of pure product spectra of collagens
and based on a single spectral interval is not likely to be
further transferred on the analysis of tissue samples. One
should also note that no correct classification could be
835
obtained using only the amide I spectral interval. Previous
studies have clearly shown that this spectral interval allows
characterizing secondary structure of these proteins [17, 25,
28], but a curve fitting method was employed to highlight
small intensity changes (a few percents) in two over the
eight IR bands found within the amide I spectral interval for
two collagen types (I vs. IV) [17]. Therefore, these small
absorption differences were not likely to be sufficient for a
successful classification of five collagen types while using
the whole amide I spectral interval. Thus, as already shown
for the study of complex biological samples [21], only a
classification based on a combination of several spectral
intervals is likely to allow reaching a high level of
discrimination. To this end, we used a combination of the
three absorption bands that provided a correct classification
of collagen types, which slightly increased the discriminant
potential of the classification model. However, we found
that adding also the amide I region improved significantly
the result of this classification model. This was probably
due to the introduction of discriminant structural informa-
tion, notably one of the most characteristic collagen
absorptions, the triple helix, which is clearly different
between collagen types [17, 29]. Thus, by using four
spectral intervals, each one providing a discriminant
information giving strength to the classification model, we
could obtain that all FTIR spectra per collagen type entered
in a single cluster. Furthermore, this classification model
exhibited a small internal heterogeneity for every cluster
while inter-clusters heterogeneity was high. To our knowl-
edge, this study is the first to show that five complex
biomolecules presenting small structural differences may be
differentiated using a simple and easy-to-run classification
method of FTIR spectra.
However, the aim of this study is to provide evidence
that FTIR spectroscopy allows to isolate the discriminant
information at the submolecular level, i.e., the characteristic
IR absorptions, which might be further used for implement-
ing FTIR spectroscopic imaging as new analytical tools to
study the molecular organization of connective tissues. To
isolate this discriminant information from every spectral
interval used in the classification method we obtained, we
used a spectral curve fitting method. This one allowed a
separate analysis of the IR absorption bands contained in a
given spectral interval [17], revealing small structural
differences between molecules. To identify all absorption
bands present in a given spectral interval, the second
derivative spectrum of type I collagen was used as a model
to be applied on every FTIR spectrum of the database. As
indicated by the RMSE values obtained, all FTIR spectra
could be correctly curve-fitted. The curve fitting models
could thus be applied on every FTIR spectrum of the five
collagen types, which opens the way to the systematization
and automation of FTIR spectra treatment. Thus, it
836
becomes possible to make FTIR spectroscopic imaging a
functional imaging technique based on reliable molecular
parameters. It should be noted that this work is out of step
with the current use of the FTIR spectroscopic imaging
technique, which is based above all on the statistical
treatment of global absorptions [30], which relate to larger
families of biomolecules. Therefore, since tissues have
different molecular compositions, the results obtained using
global statistical methods on a particular tissue cannot be
transferred to another biological medium (another organ,
another tissue…). Another strong limit at using global
absorptions treated statistically, between healthy and path-
ological tissues for example, is that tissue composition
changes due to pathophysiological processes cannot be
managed, such as for tissue inflammation, edemas, cell
dehydration, molecular changes due to enzymatic and non-
enzymatic processes…etc. These water and molecular
content changes are evenly heterogeneous and unpredict-
able in the tissue volume and may thus change dramatically
the size and the absorption patterns of FTIR spectra
between the pixels of a FTIR tissue image. It follows that
the statistical treatments of such spectra will present
increasing roots of error with this increasing heterogeneity
of tissue contents.
In our submolecular approach, the analysis of the amide
I region allowed highlighting the adequacy between the
known structural biochemistry and the spectral character-
istics of the five collagen types. Type IV collagen presented
a higher percentage of -sheet and smaller percentages of
triple helix and -like helix compared to type I, which was
in accordance with previous works [17]. These differences
may be explained by the presence of globular domains at
the ends of the triple helix in the type IV collagen [31],
which are notably assembled by -sheet structures. Type III
collagen could be differentiated from types I and V by a
smaller percentage of -like helix and from the other types
by a higher percentage of triple helix. The type V collagen
presented a lower percentage of -sheets compared to the
other types, which could be explained by the particular
organization of its NH2-terminal domain, which is attached
to the triple helical domain by a kink structuring [32]. It
was also expected that type VI collagen presented the most
evident spectral differences regarding other collagen types.
Type IV collagen presented the smallest percentage of triple
helix and the highest percentage of -sheets, whose
characteristics could be attributed to the short central
domain organized in triple helix and two large N- and C-
terminal globular domains of this collagen type [2]. Thus,
the submolecular approach developed in this study high-
lighted IR absorption differences between collagen types
that could be clearly linked to the secondary structure
parameters of these five collagen types. It was also
demonstrated that the three other spectral intervals used in
K. Belbachir et al.
the classification model contained IR absorption highly
discriminant between all collagen types. Since collagen
structure has not been sufficiently described at the
submolecular level, the IR absorption band differences
found in these spectral intervals could not be precisely
assigned to substructure parameters characteristic to every
collagen type. Nevertheless, since the combination of these
spectral intervals strengthened the classification model, one
may consider that combining several discriminant IR
absorption bands will also strengthen a data treatment
method for FTIR spectroscopic imaging applications on
tissue analyses. There is also evidence that using only the
most discriminant IR absorptions isolated from these
spectral intervals will allow proposing data treatment
methods much more specific to given biomolecules. This
approach will first require chemometrics and software
developments, notably for obtaining that all raw spectral
data from a FTIR image are curve-fitted and that selected
IR absorption bands are further used to rebuild a functional
image providing the biodistribution of the data specific to a
given biomolecule.
Taking in consideration the results obtained in this study,
we suggest that the substructure differences revealed from
the FTIR spectral intervals that present a discrimination
potential between collagen types should be considered in
the identification of collagen types for analyzing connective
tissue organization and assembly in histological sections.
Conclusion
Our results gave molecular bases for a functional investi-
gation of connective tissues in normal and pathological
conditions. The combined use of the submolecular param-
eters which appeared the most discriminant should allow a
strong improvement of the identification capacity of the
collagen types composing a given tissue by FTIR spectro-
scopic imaging. All conditions seem thus joined to make
FTIR spectroscopy and its imaging-related technique major
tools for implementing innovative methods in the field of
molecular histology, which would be very helpful for the
diagnosis of a wide range of pathologies.
Acknowledgment The authors are indebted to the “Association
Française contre les Myopathies” (AFM) for their financial support.
References
1. Brodsky B, Ramshaw JA (1997) Matrix Biol 15:545–554
2. Lampe AK, Bushby KM (2005) J Med Genet 42:673–685
3. Passerini L, Bernasconi P, Baggi F, Confalonieri P, Cozzi F,
Cornelio F, Mantegazza R (2002) Neuromuscul Disord 12:828–835
4. Tsukada S, Parsons CJ, Rippe RA (2006) Clin Chim Acta 364:33–60
Collagens analysis by FTIR spectroscopy
5. Stempien-Otero A, Plawman A, Meznarich J, Dyamenahalli T,
Otsuka G, Dichek DA (2006) J Biol Chem 281:15345–15351
6. Whitelaw SE (2003) Pediatr Nurs 29:423–426
7. Sifre L, Berge P, Engel E, Martin JF, Bonny JM, Listrat A,
Taylor R, Culioli J (2005) J Agric Food Chem 53:8390–8399
8. Listrat A, Lethias C, Hocquette JF, Renand G, Menissier F,
Geay Y, Picard B (2000) Histochem J 32:349–356
9. Passerieux E, Rossignol R, Chopard A, Carnino A, Marini JF,
Letellier T, Delage JP (2006) J Struct Biol 154:206–216
10. Petibois C, Deleris G (2006) Trends Biotechnol 24:455–462
11. Cohenford MA, Rigas B (1998) Proc Natl Acad Sci USA
95:15327–15332
12. Yano K, Ohoshima S, Gotou Y, Kumaido K, Moriguchi T,
Katayama H (2000) Anal Biochem 287:218–225
13. Wehbe K, Pinneau R, Moenner M, Deleris G, Petibois C (2008)
Anal Bioanal Chem 392:129–135
14. Levine SM, Wetzel DL (1998) Free Radic Biol Med 25:33–41
15. Bi X, Li G, Doty SB, Camacho NP (2005) Osteoarthritis Cartilage
13:1050–1058
16. Petibois C, Deleris G, Piccinini M, Cestelli Guidi M, Marcelli A
(2009) Nat Photonics 3:179
17. Petibois C, Gouspillou G, Wehbe K, Delage JP, Deleris G (2006)
Anal Bioanal Chem 386:1961–1966
837
18. Boskey AL, Gadaleta S, Gundberg C, Doty SB, Ducy P,
Karsenty G (1998) Bone 23:187–196
19. Crupi V, De Domenico D, Interdonato S, Majolino D, Maisano G,
Migliardo P, Venuti V (2001) J Mol Struct 563–4:115–118
20. Ward JH (1963) J Am Stat Assoc 58:236–244
21. Petibois C, Cazorla G, Gin H, Deleris G (2001) J Lab Clin Med
137:184–190
22. Haris PI, Severcan F (2001) J Mol Catal B: Enzym 7:207–221
23. Camacho NP, West P, Torzilli PA, Mendelsohn R (2001)
Biopolymers 62:1–8
24. Liu KZ, Dixon IM, Mantsch HH (1999) Cardiovasc Pathol 8:41–
47
25. Goormaghtigh E, Ruysschaert JM, Raussens V (2006) Biophys J
90:2946–2957
26. Fabian H, Naumann D (2004) Methods 34:28–40
27. Troullier A (2000) Nat Struct Biol 7:78–86
28. Haris PI, Servecan F (1999) J Mol Catal B: Enzym 6:207–221
29. Lee SM, Lin SY, Liang RC (1995) Artif Cells Blood Substit
Immobil Biotechnol 23:193–205
30. Lasch P, Haensch W, Naumann D, Diem M (2004) Biochim
Biophys Acta 1688:176–186
31. Kalluri R (2003) Nat Rev Cancer 3:422–433
32. Birk DE (2001) Micron 32:223–237