| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Abstracts of Oak Ridge Posters |
1
National Research Council of Canada, Institute for Biodiagnostics, 435 Ellice Ave., Winnipeg, Manitoba, R3B 1Y6 Canada
2
Department of Clinical Chemistry, Health Sciences Centre, 820 Sherbrook St., Winnipeg, Manitoba, R2H 2A6 Canada
a author for correspondence: fax 204-984-5472, e-mail Anthony.Shaw{at}nrc.ca
Infrared (IR) spectroscopy offers an approach to clinical analysis that is conceptually very appealing. Whereas countless assays rely on the use of chemical agents to "recognize" the analyte of interest and to react with the analyte to produce specific color changes, IR-based analysis is founded on the rich IR absorption patterns that characterize the analytes themselves. These absorption patterns provide the basis to distinguish among the constituents and to separately quantify them. The most obvious distinguishing feature is that no reagents are required. In addition, IR-based analytical methods require very small sample volumes (typically microliters), show good precision over the entire physiological range, and are well suited for automation.
Several previous studies have illustrated potential roles for IR spectroscopy in the clinical laboratory. For example, six serum analytes have been shown to be suitable for IR-based analysis, namely albumin, total protein, glucose, triglycerides, urea, and cholesterol (1)(2)(3)(4)(5)(6). Studies of amniotic fluid have yielded IR models to quantify the lecithin/sphingomyelin ratio and the surfactant/albumin ratio, establishing IR spectroscopy as an attractive option for the assessment of fetal lung maturity (7)(8).
There are several approaches to IR-based analysis, with the first choice being whether to use the near-IR (750–2500 nm) or mid-IR (2.5–100 µm) spectral range. Near-IR spectroscopy has gained notoriety within the clinical chemistry community through the many efforts to develop a noninvasive blood glucose monitor based on this technology [see e.g., Refs.(9)(10)], and in that vein it has been shown that glucose concentrations can be recovered from the near-IR spectrum of native serum (3).
The main reason for the focus on near-IR spectroscopy is that tissue is
quite transparent to near-IR light, hence the attraction for in vivo
work. However, this is obviously not a factor for in vitro analysis.
The mid-IR spectrum offers some potential advantages. Near-IR
spectroscopy typically requires a sample volume of at least 0.1–0.2
mL, whereas a mid-IR assay can be carried out with 
10 µL. Although
water contributes enormous absorption bands in the mid-IR,
these can be eliminated by simply drying the sample to a film and using
the spectrum of the dry film as the basis for analysis
(6)(7)(8). This film may then be archived for subsequent
reanalysis.
The present study was conducted to evaluate the sensitivity and accuracy of mid-IR spectroscopy in the determination of urine urea, creatinine, and total protein. The IR-based quantification methods were calibrated by comparison with the results provided by standard clinical chemistry assays. To that end, urea [enzymatic (urease) conductivity], creatinine (Jaffé rate), and total protein (benzethonium chloride reaction) concentrations were determined for 200 urine samples. Urea concentrations were 40–440 mmol/L, creatinine concentrations were 1.5–18 mmol/L, and total protein was 0.02–20 g/L. Samples were prepared for IR spectroscopy by first adding 0.1 mL of aqueous (4 g/L) potassium thiocyanate solution to 0.5 mL of the urine sample. Duplicate films were prepared by drying 12 µL of this mixture onto IR-transparent BaF2 substrates, and mid-IR absorption spectra were acquired at ambient temperature for the dry films (Bio-Rad FTS-40A Fourier transform IR spectrometer operating at 4 cm-1 resolution, with 512 scans averaged for both the sample and background spectra). An isolated thiocyanate absorption at 2060 cm-1 then provided the basis to normalize all spectra to a common effective optical pathlength.
Quantification methods were derived by using partial least-squares
regression (PLS) to establish relationships between the IR spectra and
the reference analyses. A training set of 133 specimens (266 spectra)
was used to calibrate quantification methods for each of the three
analytes. The test set, comprising the remaining 67 specimens (134
spectra), served to test the validity of the IR-based assays. The
accuracy of the PLS quantification models was improved by using
spectral subregions rather than the entire 800-5000
cm-1 range that was available. The appropriate
spectral regions for PLS were determined by first carrying out a series
of exploratory trials using limited spectral ranges and fine-tuning
those ranges based on the standard errors in the training and test
sets. The number of PLS factors in the final model was set at the point
where (a) the addition of more factors produced either no
improvement or a deterioration in the concentrations predicted for the
test set, and (b) the predicted concentrations were equally
accurate for the training and test sets. The final quantification
models were based on the spectral region 900-1500
cm-1 for protein (16 PLS factors), 1400–1800
cm-1 for creatinine (11 factors), and 3100–3550
cm-1 for urea (7 factors). Scatterplots
comparing the IR-predicted protein, creatinine, and urea concentrations
to the reference analyses for this set of test specimens are shown in
Fig. 1
. The IR-based analytical methods yielded creatinine
concentrations with a Sy|x [the root mean
square difference between IR-predicted and reference analyte
concentrations for the test set only] of 0.58 mmol/L
(r = 0.98) for creatinine, 14.1 mmol/L
(r = 0.98) for urea, and 0.48 g/L (r =
0.94) for protein.
|
The distribution of protein concentrations is skewed heavily, with the
majority of specimens showing concentrations well below 1 g/L (Fig. 1
, middle panel). As a result, the best approach to IR-based
protein quantification is to use two models rather than one. A second
PLS quantification model was optimized for those samples with
concentrations <1 g/L, yielding Sy|x = 0.13 g/L.
although this still falls short of the performance required for
accurate quantification at typical low protein concentrations, the
method is sufficiently accurate to serve as a coarse screening test.
The ultimate accuracy of the IR-based methods is influenced in part by the accuracy of the reference methods used to calibrate them. This is not a factor for the protein analysis, where the reference method is clearly more accurate than the IR-based method, but it may play a role for both urea and creatinine. This possibility is suggested by the precision of the IR-based assays: SDdup = 0.18 mmol/L for creatinine, 6.8 mmol/L for urea, 0.14 g/L for protein (including all samples), and 0.05 g/L for protein concentrations <1 g/L.1 At least part of the gap between the precision and accuracy of the urea (Sy|x = 14.5; SDdup = 6.8 mmol/L) and creatinine (Sy|x = 0.54; SDdup = 0.18 mmol/L) assays may be attributable to scatter in the reference methods themselves.
The mid-IR quantification methods presented here match or exceed the
performance of the near-IR methods presented previously
(12). Both approaches yield analyses that are accurate
enough to serve as a routine method for urine urea and creatinine
analyses. Although protein concentrations are too low for accurate
quantification using IR spectroscopy, the method may serve as a screen
to detect concentrations above 
0.5 g/L and to quantify at those
concentrations.
The practical implementation of this and other clinical IR-based assays requires two key developments. One of these is the discovery of an inexpensive substrate to substitute for the costly BaF2 windows that were used as part of this work. Although these windows can be cleaned and used repeatedly, this is probably impractical in high-volume laboratories. A surprising alternative has emerged recently, as we have shown recently that many analyses can be carried out using ordinary glass as the substrate, despite its limited transparency in the mid-IR region (13)(14). The stumbling block that remains in place is a practical one, that being automation of the method. The practical benefits of IR-based methods are being realized in an extraordinary range of analytical applications (15), and it would seem to be only a matter of time before these methods find their way into the clinical realm.
Footnotes
1 SDdup = (
d2/2n)1/2, where d is the difference between concentrations determined for duplicate aliquots and n is the number of samples (11). 
References
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  K.-Z. Liu, R. A. Shaw, A. Man, T. C. Dembinski, and H. H. Mantsch Reagent-free, Simultaneous Determination of Serum Cholesterol in HDL and LDL by Infrared Spectroscopy Clin. Chem., March 1, 2002; 48(3): 499 - 506. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
