Clinical Chemistry 45: 1530-1535, 1999;
(Clinical Chemistry. 1999;45:1530-1535.)
© 1999 American Association for Clinical Chemistry, Inc.
Determination of Glucose in Dried Serum Samples by Fourier-Transform Infrared Spectroscopy
Cyril Petibois1,
Vincent Rigalleau2,
Anne-Marie Melin3,
Annie Perromat3,
Georges Cazorla1,
Henri Gin2 and
Gérard Déléris3,a
1
Faculté des Sciences du Sport et de l'Education Physique, Université Victor Segalen, Bordeaux 2, France.
2
Service de Nutrition, Hopital du Haut-Levèque,
Bordeaux 33604, France.
3
Laboratoire de Chimie Bio-Organique, INSERM U443,
Université Victor Segalen, Bordeaux 2, France.
a Address correspondence to this author at: Université Victor Segalen, Bordeaux 2, Laboratoire de Chimie Bio-Organique, 146 Rue Léo Saignat, F 33 076, PB 12, Bordeaux, France. Fax 05-57-57-10-02; e-mail gerard.deleris{at}bioorga.u-bordeaux2.fr
 |
Abstract
|
|---|
Background: Practical improvements are needed to allow measurement
of glucose concentrations by Fourier- transform infrared (FT-IR)
spectroscopy. We developed a new method that allows determination of
the glucose concentration in dried sera.
Methods: We studied 32 serum samples after fourfold dilution and
desiccation before FT-IR analyses on a spectrometer operated at a
resolution of 2.0 cm-1. We integrated all spectral windows
at the surface of the spectrum in the C
O region. For comparison,
glucose was measured in the sera by a glucose oxidase method.
Results: One peak within the spectrum was most specific for
glucose (997–1062 cm-1). Its surface integration showed a
strong relationship with reference data (r = 0.998;
P <0.001). FT-IR analyses of five glucose solutions
were performed to determine its specific absorption at the same peak.
In this way, glucose concentrations in serum spectra could be measured.
For the first time while using FT-IR spectroscopy, no manipulation of
spectra nor use of internal standard was necessary to obtain results in
high accordance with glucose concentration measured by a conventional
(glucose-oxidase) method (Sy|x = 0.25
mmol/L; r = 0.998).
Conclusions: FT-IR spectroscopy appears to be an easy and
accurate method to determine glucose concentration and could be widely
used to simultaneously identify and quantify several metabolites in
biological fluids or tissues.
|
Introduction
|
|---|
The search for a simple and accurate analytical method to
determine glucose concentration is a problem of major importance for
clinical laboratories. Most physicochemical analytical methods, such as
Raman spectroscopy (1)(2), mass spectrometry
(3), and nuclear magnetic resonance spectroscopy
(4)(5), have been evaluated as tools to
determine glucose concentration in whole blood, plasma, or serum. With
Fourier-transform infrared (FT-IR) spectroscopy, the spectrum of any
molecule shows its characteristic absorptions. For glucose, the
following absorption bands may be found: 
(OH) between
3570 and 3120 cm-1, (
CH)
between 3085 and 3020 cm-1,
(CO) between 1230 and 1000
cm-1, and (COC) between 1275
and 800 cm-1 (6). The two latter
bands, the CO region, are known to be the most specific of this
molecule in complex spectra (7)(8).
When using this technique to determine glucose concentration, some
corrections of results are still necessary, such as normalizations
and/or standardizations of spectra. Ward et al. (9) obtained
accurate results in using partial least-squares (PLS) calibration but
determined that these were only reliable for single subjects. A
standard calibration could not be proposed, probably because of
differences in water IR absorption or protein content between samples.
The normalization of each baseline set at zero absorbance is another
methodological problem. Ward et al. (8) added glucose as an
internal standard in the blood samples of diabetic subjects and applied
PLS calibration to measure glucose concentrations. Several biases
appeared, such as the effects of temperature on the water IR absorption
in the solution sample, protein IR absorptions in the frequency range
of analysis (between 950 and 1200 cm-1), and
differences in blood composition specific to each subject.
In another experimental FT-IR method, Budinova et al. (10)
deposited whole blood or serum samples on polyethylene cards and added
KSCN as internal standard for peak absorbance comparisons. They
analyzed dried samples and obtained accurate results with the 900-1185
cm-1 spectral area integration and with respect
to the baseline. The method yielded reproducible results, but this
frequency range is not specific for glucose absorption, which appears
but is not alone between 900 and 1300 cm-1.
Furthermore, the use of an internal standard is still necessary to
estimate glucose concentration.
Heise and Bittner (11) concluded in similar experimentations
that each manipulation of the spectrum has to be considered as a source
of quantitative error affecting the prediction performances of the
method used. Recently, Shaw et al. (12) obtained glucose
concentration in accord with a reference clinical method using
multivariate calibrations and statistical methods (PLS calibration),
but an internal standard (KSCN) was still necessary because of the
difficulty to obtain a reproducible dried-sample thickness onto the
IR-transparent material. They had also to normalize their spectra (to a
common absorption intensity in the C
N stretching mode of the KSCN
internal standard). Although their normalized spectra provided equal or
better agreement in all analyses, the differences were significant for
glucose measurement. The normalization step was still necessary and
provided a marked improvement in comparison with nonnormalized spectra.
Thus, practical improvements are still necessary to measure glucose
concentration with a minimum of sample manipulations. The objective is
the specific measurement of glucose absorption while avoiding any use
of statistical normalization or internal calibration. To propose such a
method, (a) we sought an appropriate dilution to obtain the
highest spectral signal-to-noise ratio; (b) we investigated
the most specific peak of glucose in serum spectra; (c) we
measured glucose absorption at this peak with known glucose standards;
(d) we used the results to calculate glucose concentration;
and (e) we validated the method by comparison of our results
with results of a glucose-oxidase method.
|
Materials and Methods
|
|---|
subjects and sample collections
Blood samples from 32 diabetic patients were used for this study.
The mean age of the patients was 49.9 ± 15.8 years (range, 22–82
years). Sampling was done between 0730 and 0830, after an overnight
fast of 10 h. Blood samples for glucose concentration
determination were taken in two collection tubes (red top, Vacutainer;
Becton Dickinson), one for the FT-IR method, and the other one for the
enzymatic glucose oxidase method (13). Venous blood was
sampled with an antecubital venipuncture of the right arm using a
Teflon catheter and was immediately centrifuged for 10 min at
4000g. Samples for FT-IR analyses were then stored at
-20 °C before analysis.
analyses
After the samples returned to room temperature (~15 min at
20–25 °C), aliquots of 15, 20, and 25 µL were taken with a
micropipette (pipetman P20; Gilson) and diluted with 85, 80, and 75
µL, respectively, of water (pipetman P200; Gilson). SDs for the
micropipette sampling volumes were 0.6% (0.12 µL) for the P20 and
0.9% (0.73 µL) for the P200 for 20 and 80 µL, respectively, of
water. The diluted samples were homogenized with an agitator (Vortex
Reax 2000; Heidolf) at 1000g for 10 s. Then, 35 µL of
each solution was deposited exactly within the cell limits of a
zinc-selenide (ZnSe) wheel bearing 15 sample cells (Bruker). Cell
limits are materialized by a colored circle (interior diameter, 7 mm)
that the sample must touch but not cover. The wheel was subsequently
placed in a drying vacuum to evaporate water (45 min). A KBr cover was
then screwed onto the wheel to protect it. The wheel was finally put
into the analysis compartment of a Bruker IFS 28/B spectrometer
equipped with a Globar (MIR) source (7 V), a KBr beamsplitter, and a
DTGS/B detector (18–28 °C). For all experiments, we used a
resolution of 2.0 cm-1, and acquisitions were
performed using 32 scans. Beam diameter at the sample location was 6
mm. All analyses were performed in triplicate on three successive
wheels. The higher signal-to-noise ratio (731 ± 25) and
homogeneous baselines between samples were obtained using a 1:4
(vol/vol) serum dilution in water. The signal-to-noise ratio was only
325 ± 71 and 461 ± 101, respectively, with 15:100 and 1:3
dilutions. The between-run CV was also the lowest (1.5%) for
triplicate FT-IR measurements using the fourfold concentration of
dilution. This was used for the following experiments.
calculations
The spectral integrations of Budinova et al. (10)
avoided baseline variations, and this presents a particular interest
for the interpretation of spectral data. They integrated the area
between the spectral window and the cord running through the abscissa
limits of the spectrum in the chosen frequency range (950–1850
cm-1). We applied this method to integrate all
of the spectral windows found at the surface of the spectrum in the
CO region (Fig. 1
). This was done by the software Opus-Ident 2.2 (Bruker) after
we had chosen the abscissa limits. Comparisons were made between the
integration results (expressed in the Opus-Ident 2.2 software
integration units; U) and the glucose concentration reference values.
View larger version (18K):
[in this window]
[in a new window]
|
Figure 1. Coordinates of the peaks observed in the CO region
(1300–900 cm-1) of a serum sample spectrum.
The area integrated for each peak is above the cord running through the
two side points found at the surface of the spectrum after the abscissa
limits have been chosen.
|
|
spectral determination of glucose concentrations
For spectral discrimination between glucose and other hexoses that
could be present in blood in nonnegligible concentrations, we performed
the FT-IR analyses of 55 mmol/L solutions of D-fructose,
D-galactose, and D-mannose (Fig. 2
). To measure glucose absorption, we did the FT-IR analyses of
five glucose solutions in water (S1 to S5) containing 6.77, 5.44, 4.11,
2.76, and 1.38 mmol/L (±10-2 mmol/L),
respectively, of D-glucose. This calibration series allowed
measurement of the relative absorption of each glucose peak in the
1300–900 cm-1 region. To avoid an eventual
crystallization of the D-glucose during drying, which is a
potential source of error for an optical technique such as FT-IR
spectroscopy, we also used solutions of 20 µL serum and 80 µL of
each glucose solution. When the above calibrators were diluted with
20% serum, the final glucose concentrations in this unique serum
increased by 5.44, 4.33, 3.27, 2.22, and 1.11 mmol/L
(±10-2 mmol/L), and these later solutions were
named S6 to S10. Glucose absorbances in these solutions were determined
by subtracting the spectral area at peak 5 (997–1062
cm-1) of the serum diluted in water to the one
diluted in a glucose solution. Therefore, the added
D-glucose may be considered as an internal standard, as in
the study by Ward et al. (8). Glucose absorptions were then
expressed in concentrations per spectrometer integration digits:
mmol · L-1 · U-1.
View larger version (22K):
[in this window]
[in a new window]
|
Figure 2. FT-IR spectra of D-fructose,
D-galactose, D-glucose, and
D-mannose in the CO region (1200–900
cm-1).
Standard solutions (55 ± 0.1 mmol/L) were used. The peak located
at 1033 cm-1 is the most specific for glucose.
|
|
statistics
Data are expressed as mean ± SD. Linear regressions were
done for comparison of series of data. Reference and spectral data were
considered to be comparable when correlation coefficients
(r) obtained a probability of P <0.01. When the
data were comparable, dispersion data around regression lines were
estimated by the mean prediction errors
(Sy|x) and the variation coefficient
(VC, in percentages or standard units) around the regression
line. Lastly, for differences between the methods, a standard
t-test for paired data was done.
|
Results
|
|---|
clinical data
Subjects presented venous glucose concentrations of 11.15 ±
6.94 mmol/L. Six subjects had glucose concentrations <2.5 mmol/L, and
six others had glucose concentrations >19.5 mmol/L. The range was
1.83–23.14 mmol/L. This high heterogeneity was chosen for best
comparison with FT-IR analyses.
comparisons of enzymatic and ft-ir analyses
There were five principal peaks in the CO region of a serum
spectrum (Fig. 1
). Comparisons between reference glucose concentrations
and FT-IR absorbance areas of the peaks observed in the CO region
are presented in Table 1
. The mean glucose concentration was 11.2 ± 6.9 mmol/L
with the glucose oxidase method and 0.31 ± 0.19 U with FT-IR
spectroscopy under peak 5 (997–1062 cm-1). The
correlation between these data was r = 0.998,
P <0.001. The other peaks observed in the CO region
showed a lower correlation between spectral and biochemical results,
with P <0.05. Spectra were performed in triplicate; a high
reproducibility was obtained with a mean difference of 1.76% (0.013 U)
between FT-IR absorbance areas at the peak located between 997 and 1062
cm-1. This mean difference was not statistically
significant.
glucose determination and measurement
Spectra of the four hexoses analyzed did not show similar peaks in
the CO region, in particular for glucose, which presents a strong
absorption band at 1033 cm-1, the only one found
in this location (Fig. 2
). Fig. 3
presents the spectra of a serum (glucose concentrations, 5.5
mmol/L) diluted in water (1:4, by volume) and the same serum diluted in
a 11 mmol/L glucose solution (1:4, by volume). The peak located at 1033
cm-1 presents the major increase of absorbance
in the CO region because of the added glucose. Changes in the
997–1062 cm-1 absorption band with glucose
concentration in aqueous solutions (S1–S5) and recovery of glucose in
serum samples using the same absorption band (S6–S10) are presented in
Tables 2 and
3, respectively. The mean glucose absorption per unit of spectral
area at peak 5 (997–1062 cm-1) was 7.27 ±
0.17 mmol · L-1 · U-1 for the glucose
solutions and 7.26 ± 0.11
mmol · L-1 · U-1 for the glucose
solutions within sera. The correlation between the two series of
calibrators was r = 0.997 (P <0.001). The
mean prediction error (Sy|x) of these
measurements was 0.09 mmol/L for glucose concentrations in calibrator
solutions and 0.11 mmol/L when glucose was added to the sera. We
subsequently used this value of 7.27
mmol · L-1 · U-1 to determine
directly each glucose concentration by integrating the peak 5 area
within the samples analyzed previously (and considering the 20%
dilution chosen for analyses). Values with these methods were very
close (11.15 ± 6.94 mmol/L vs 11.14 ± 6.98 mmol/L,
respectively, for reference and FT-IR methods). The correlation between
these data was r = 0.998 (P <0.001) with a
mean prediction error Sy|x = 0.25 mmol/L
(VC = 0.14%). Regression between both methods is
presented in Fig. 4
. Regression was very close to the expected relationship with a
slope of 1.006 ± 0.004 and an intercept of -0.049 ± 0.003
mmol/L.
View larger version (17K):
[in this window]
[in a new window]
|
Figure 3. Spectra of a serum diluted in water and diluted in a
glucose solution (11 mmol/L).
The peak located at 1033 cm-1 is the major IR glucose
absorbance in the CO region (900–1300 cm-1).
|
|
|
Discussion
|
|---|
The objective of this study was the specific and accurate
measurement of glucose concentration using FT-IR spectroscopy. The
following original steps were performed: (a) focus on the
most specific peak of glucose in the CO region; (b)
quantification of glucose absorption under this peak with known
calibrators; and (c) glucose concentration measurement
without spectral manipulations.
Until now, several studies had been performed on sera in which
different substances, such as D-glucose
(8)(9) or KSCN (10)(12),
were added. These two components behave as internal standards and thus
may limit the method for routine measurements, because they require
precise sample manipulations. Therefore, in our method, we tried to
minimize manipulations, which actually consist of only one simple,
fourfold dilution and drying the samples on the analysis wheel. The
chemical composition of the blood samples and their differences were
not altered. In contrast, Ward et al. (8) could not control
glucose absorption with their method, because their calibration was
done with wide-ranging absorptions of glucose in the spectrum and not
its specific absorption in a definite band. Their calibration,
performed on the basis of internal standards and general spectral
absorptions, also contained other variables, i.e., proteins and water.
Simple transmittance measurements of liquid blood or blood serum are
impossible in the mid-IR because of the major absorption of water in
the matrix (10), and wide absorption areas will
necessarily take proteins into account because these are major blood or
serum constituents (12). Finally, the results of Ward et al.
(8) were reliable only for single subjects, and they were
not able to propose a standard method for routine measurements. The
drying of the samples is of particular importance for this kind of
technique because it avoids the standardization difficulties
experienced by Ward et al. (8)(9) and by Heise
and Bittner (14), who analyzed samples in solution. The
latter could not control the effects of water absorption in spectra and
of room temperature on water.
Like Budinova et al. (10), we analyzed dried samples to be
free of any kind of interaction with changes in ambient conditions.
However, they dried whole blood or blood sera in the ambient conditions
of their laboratory. For whole blood, a time-dependent glycolysis is
known to occur when erythrocytes and serum are not separated, which
makes their method invalid for such samples. For blood sera, the drying
probably lasted several hours, which represents a limitation when
compared with clinical reference methods. Furthermore, an internal
standard (KSCN) was still necessary because of the thickness
variability of the samples applied on polyethylene cards.
Budinova et al. (10) integrated a large part of the CO
region (900–1185 cm-1) and Shaw et al.
(12) chose the 925-1250 cm-1 region,
which are too wide to be strongly specific for glucose. We expected to
find a spectral peak more specific of glucose than these large regions.
Actually, we found a statistically significant relationship
(P <0.05) with the reference data for all peaks of the
CO region and a strong relationship (P <0.001) for that
located at 1033 cm-1 (997–1062
cm-1). This is in agreement with observations
reported by Ward et al. (8), who observed the strongest
glucose absorbance at 1035 cm-1. However, other
blood hexoses are present at nonnegligible concentrations. To verify
whether the peak located at 1033 cm-1 could be
glucose specific, we also analyzed fructose, galactose, and mannose
with the same method of analysis. The spectra of these derivatives are
close to that of glucose. However, there were differences in the
absorption frequencies. In comparison with the glucose peak located at
1033 cm-1, fructose presents a peak at 1059
cm-1, galactose at 1064
cm-1, and mannose at 1075
cm-1. Thus, our data suggest that the peak
situated between 997 and 1062 cm-1 might
correspond to a more specific absorption of the
(COC) bond.
The second aim of the study was the quantification of the glucose
absorption on this peak, which was done with a five-calibrator series.
We found that glucose absorption per unit of spectral area was 7.27
mmol/L/U. This value was used to determine glucose concentration in
serum spectra. The strong relationship observed with the
reference method (P <0.001) demonstrated that FT-IR
spectroscopy is a clinically satisfactory method. Unlike Shaw et al.
(12), we obtained a standard error between reference and
FT-IR methods (0.24 and 0.27 mmol/L, respectively), which allows our
method to be more suitable for clinical routine measurements on
diabetic subjects or other patients with metabolic disorders. It is
possible that this very small difference from the reference method is
because of the sample thickness we found highly reproducible. This
thickness was so reproducible because of the high concentration of
dilution used (1:4, by volume). It is, therefore, possible that the
serum constituents were more evenly distributed onto the sample
location than the one obtained by Shaw et al. (12) with a
1:1 (by volume) concentration of dilution. With only 7 µL of such a
solution, spread on a large cell (13-mm diameter), it is unlikely that
the sample could be homogeneously distributed on the surface of the
BaF2 window. Differences between wetability of
the mineral ZnSe wheel used for our analyses and the organic
polyethylene-card support used by Budinova et al. (10) may
also explain the advantage of our protocol.
Current enzymatic methods require frequent calibration controls and
reagents, and this is very costly (15). Our method does not
require any reagent. Nevertheless, a longitudinal study in clinical
conditions is necessary to assess the practical advantages of such an
analysis. Moreover, other blood component concentrations could be
determined during the same analysis or later because of the
computerized preservation of spectral information. Furthermore, the
ZnSe holder simultaneously allows 15 analyses, which represents a clear
advantage in comparison with other instruments. It is conceivable that
an extension of the wheel would allow the simultaneous analysis of more
samples, and this could reduce proportionally the mean time and cost of
analysis.
In conclusion, FT-IR spectroscopy allows accurate glucose
concentration determination. Actually, it could be used for the
simultaneous determination and quantification of a large variety of
metabolites present in biological fluids. An important development of
the method could be the determination of such derivatives from
microsamples, which could combine the clinical accuracy of the
measurements with the convenience of microsampling. Developments are
under way in our laboratory to achieve such improvements.
|
Acknowledgments
|
|---|
We are indebted to the Conseil Régional d`Aquitaine
and the Fédération Française de Rubgy for financial
support and technical assistance. We greatly acknowledge helpful
suggestions and comments from Dr. Matthias Boese (Bruker-Optik-GmbH,
Biomedical Products).
|
References
|
|---|
-
Dou X, Yamaguchi Y, Yamamoto H, Uenoyama H, Ozaki Y. Biological applications of anti-stokes Raman spectroscopy: quantitative analysis of glucose in plasma and serum by a highly sensitive multichannel Raman spectrometer. Appl Spectrosc 1996;50:1301-1306.
-
Berger AJ, Itzkan I, Feld MS. Feasibility of measuring blood glucose concentration by near-infrared Raman spectroscopy. Spectrochim Acta Part A Mol Spectrosc 1997;53:287-292.
-
Hannestad U, Lundblad A. Accurate and precise isotope dilution mass spectrometry method determining glucose in whole blood. Clin Chem 1997;43:794-800.
[Abstract/Free Full Text]
-
Pan S, Chung H, Arnold MA. Near-infrared spectroscopic measurement of physiological glucose levels in variable matrices of protein and triglycerides. Anal Chem 1996;68:1124-1135.
[Medline]
[Order article via Infotrieve]
-
Tarata T, Noguchi J, Fukushima K, Ishihara Y, Ohkochi S, Uedaira H. Proton NMR spectroscopic studies of serum as an aid to perioperative cellular metabolic monitoring. Physiol Chem Phys Med NMR 1995;270:121-129.
-
Gremlich HU. Infrared and Raman spectroscopy. In:
Ullmann's encyclopedia of industrial chemistry, Vol. B5. Weinheim,
Switzerland: VCH, 1994:429–69..
-
Bauer B, Floyd TA. Monitoring of glucose in biological fluids by FT-IR with a cylindrical internal reflectance cell. Anal Chim Acta 1987;197:295-301.
-
Ward KJ, Haaland DM, Robinson MR, Eaton RP. Post-prandial blood glucose determination by quantitative mid-infrared spectroscopy. Appl Spectrosc 1992;46:959-965.
-
Ward KJ, Haaland DM, Robinson MR, Eaton RP. Quantitative infrared spectroscopy of glucose in blood using partial least-square analyses. SPIE 1989;1145:607-608.
-
Budinova G, Salva J, Volka K. Application of molecular spectroscopy in the mid-infrared region to the determination of glucose and cholesterol in whole blood and in blood serum. Appl Spectrosc 1997;51:631-635.
-
Heise HM, Bittner A. Investigation of experimental errors in the quantitative analysis of glucose in human blood plasma by ATR-IR spectroscopy. J Mol Structure 1995;348:21-24.
-
Shaw RA, Kotowich S, Leroux M, Mantsch HH. Multivariate serum analysis using mid-infrared spectroscopy. Ann Clin Biochem 1998;35:624-632.
-
Bergmeyer HV. Methods of enzymatic analysis. New York:
Academic Press, 1974:316pp..
-
Heise HM, Bittner A, Koschinsky T, Gries FA. Ex-vivo determination of blood glucose by microdialysis in combination with infrared attenuated total reflection spectroscopy. Frensenius J Anal Chem 1997;359:83-87.
-
Marks V. Blood glucose: its measurement and clinical importance. Clin Chim Acta 1996;251:3-17.
[ISI][Medline]
[Order article via Infotrieve]
The following articles in journals at HighWire Press have cited this article:
|
|
|
|
 
C. Petibois, A. Cassaigne, H. Gin, and G. Deleris
Lipid Profile Disorders Induced by Long-Term Cessation of Physical Activity in Previously Highly Endurance-Trained Subjects
J. Clin. Endocrinol. Metab.,
July 1, 2004;
89(7):
3377 - 3384.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
C. Petibois, G. Cazorla, A. Cassaigne, and G. Deleris
Plasma Protein Contents Determined by Fourier-Transform Infrared Spectrometry
Clin. Chem.,
April 1, 2001;
47(4):
730 - 738.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Introduction
