- Email: [email protected]
Evaluation of a miniaturized thermal biosensor for the determination of glucose in whole blood
Clinica Chimica Acta 267 (1997) 225–237
Evaluation of a miniaturized thermal biosensor for the determination of glucose in whole blood a a b a, Ulrika Harborn , Bin Xie , Raghavan Venkatesh , Bengt Danielsson * a
Department of Pure and Applied Biochemistry, Chemical Center, P.O.B. 124, S-221 00 Lund, Sweden b PROMED, The Voluntary Health Services, Adyar, 600 113 Madras, India Received 6 March 1997; received in revised form 22 July 1997; accepted 30 July 1997
Abstract A miniaturized thermal biosensor has been evaluated as part of a flow-injection analysis system for the determination of glucose in whole blood. Glucose was determined by measuring the heat evolved when samples containing glucose passed through a small column with immobilized glucose oxidase and catalase. Samples of whole blood (1 ml) can be measured directly, without any pretreatment. The correlation in the response between the thermal biosensor, the Reflolux S meter (Boehringer Mannheim), the Granutest 100 glucose test kit (Merck Diagnostica) and the Ektachem (Kodak) instrument was evaluated. The influence of the hematocrit value and of possible interferences is reported. The correlation measurements show that the thermal biosensor calibrated with aqueous glucose standards generally gives lower values on blood glucose than the reference methods calibrated for serum or blood measurements. Mean negative biases range from 0.53 to 1.16 mmol / l. Differences in sample treatment clearly complicate comparisons and the proper choice of reference method. There was no influence from substances such as ascorbic acid (0.11 mmol / l), uric acid (0.48 mmol / l), urea (4.3 mmol / l) and acetaminophen (0.17 mmol / l) on the response to 5 mmol / l glucose. The hematocrit value does not influence the glucose determination, for hematocrit values of between 13 and 53%. 1997 Elsevier Science B.V. Keywords: Flow injection analysis; Thermal; Calorimetric; Biosensor; Glucose oxidase; Calibration aspects; Diabetes; Enzyme thermistor
1. Introduction The concept of a biosensor is defined as an analytical device that intimately *Corresponding author. Fax: 0046 46 22 24611. 0009-8981 / 97 / $17.00 1997 Elsevier Science B.V. All rights reserved. PII S0009-8981( 97 )00151-4
226
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
associates a biological sensing element with a transducer and responds selectively and reversibly to the concentration or activity of chemical species in biological samples [1]. In thermal biosensors, the enzymes or cells immobilized and filled into a column are the biological sensing material and the thermistors in close contact to the column are the transducers [2]. The heat evolved during the enzyme-catalyzed reaction is in proportion to the amount of substrate in the sample. Thermal biosensors have previously proved to work well with biological samples, such as serum [3] and fermentation broth [4]. Miniaturization of thermal biosensors has led to the use of small sample volumes, giving a wider linear range and a reduced response time [5]. In an earlier report, we demonstrated that samples of whole blood can be applied directly in a miniaturized flow system for the determination of glucose [6]. By using a sample volume as small as 1 ml, whole blood can be directly injected without any predilution. The massive interest in developing analytical methods for the determination of glucose in whole blood is due to the fact that many people suffering from diabetes are dependent on these methods in their daily lives. Many biosensors, mostly electrochemical, have been developed for this purpose [7]. The ultimate goal for biosensor research in this field is the development of an in vivo sensor in combination with an insulin pump for the continuous regulation of the glucose concentration in diabetic patients [8]. Measurement of whole blood glucose with a flow-injection thermal biosensor can lead to the development of miniaturized sensors for home monitoring as well as to equipment for the continuous monitoring of glucose. Automatic regulation of the insulin dosage needed for patient use or for diabetes research is also actively being considered. The introduction of new measurement technologies requires comparison with established methods for validation, but it may be difficult to find such methods that work in the same medium and with the same calibrators. This problem is discussed in the present paper in which the precision and accuracy of the thermal biosensor is compared with two commercially available methods for glucose determination in whole blood and one common clinical method for plasma glucose determination, which is corrected to give whole blood values. The effect of the hematocrit value on the glucose concentration determination is reported.
2. Materials and methods
2.1. The thermal biosensor The miniaturized thermal biosensor used in this study has been described elsewhere [9]. The most important part of the sensor is the enzyme column,
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
227
Fig. 1. Schematic diagram of the calorimeter of the sensor.
which is made of a thin-walled acid-proof steel tube (1.5 mm, I.D.; length 15 mm) containing porous beads with immobilized enzymes. Buffer enters the sensor through the equilibration coil, composed of acid-proof steel tubing wound inside the cylindrical heat-sink (Fig. 1). In order to minimize heat leakage from the column, the effluent from the column passes through steel tubing that is wound around a copper tube (adiabatic shield) surrounding the column. The heat evolved during the enzyme-catalyzed reaction is measured with thermistors placed on 0.2 mm (I.D.) gold tubes at the inlet and outlet of the column.
2.2. Experimental set-up A peristaltic pump (Gilson, Minipuls 2, France) continuously pumps 70 ml / min of 0.1 mol / l sodium phosphate-buffered saline, pH 7.0, containing 1 g of EDTA (ethylenediaminetetraacetic acid, disodium salt) and 4 g of sodium fluoride per litre (analysis buffer), through the sensor. Samples were injected through a 1-ml injection valve (C14W, VIGI AG, Valco Europe, Switzerland). The heat evolved in the enzyme column was detected with a Wheatstone bridge and the signal was amplified. During the experiments, the sensor was placed in a small aluminium box that was insulated with polyurethane foam.
2.3. Enzyme immobilization The glutaraldehyde activation step was performed according to Weetall [10] on pre-silanized spherical CPG (controlled pore glass) beads (Trisoperl, size 125–140 mm, pore size 50 nm; Schuller, Steinach, Germany). The enzymes, 5 mg of glucose oxidase (EC 1.1.3.4) from Aspergillus niger (166 U / mg, Sigma)
228
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
and 6 mg of catalase (EC 1.11.1.6) from bovine liver (19 900 U / mg, Sigma) were added to 500 mg (wet weight) of activated beads in 1 ml of 0.1 mol / l sodium phosphate buffer, pH 7.0 (coupling buffer). The glucose oxidase was added first and allowed to react for 1 h. Excess activated groups were blocked using 20 mg of bovine serum albumin (Fraction V, 99%, Sigma) in 1 ml of coupling buffer. The beads were finally washed in the coupling buffer followed by the buffer used for the glucose determinations.
2.4. Calibration A 40-mmol / l standard solution of D( 1 )-glucose (monohydrate, Merck) was prepared in the analysis buffer 24 h before use. Glucose standards, at concentrations of 1–20 mmol / l, were made from the standard solution and a calibration curve was constructed, based on the heat evolved as a function of glucose concentration.
2.5. Collection and pretreatment of blood samples Capillary blood was collected when using the Reflolux glucose meter as the reference instrument (see Section 2.6). The finger tip was punctured, using a small lancet, and 500 ml of blood were collected in test-tubes containing 2 mg of ¨ sodium fluoride and 14 IU of sodium heparin (25 000 U / ml; Loevens, Malmo, Sweden). Venous blood was used in all other experiments. Blood was drawn in vacutainer tubes (5 ml, Becton Dickinson) containing 20 mg of sodium fluoride and 143 IU of heparin. In order to obtain a set of whole blood samples with glucose concentrations in the linear range of the thermal biosensor with minimal dilution, small volumes of a highly concentrated glucose solution were added to a comparably large volume of blood. This glucose solution was prepared in the analysis buffer. Blood samples with different hematocrit values were obtained by adding or subtracting certain amounts of plasma from the original blood sample taken from the vein.
2.6. Reference methods Reflolux S (Boehringer Mannheim Scandinavia, Bromma, Sweden) is a portable glucose monitor that measures the colour change of test strips exposed to whole blood. The test strips contain glucose oxidase, catalase and colour reagents. A drop of blood is placed on the test strip. After the reaction is completed, the blood is wiped off with Kleenex tissue. The entire measurement takes 2 min and the glucose concentration is displayed in ‘‘mM’’. Previous studies of the Reflolux instrument compared to other portable meters [11] and to the Kodak Ektachem [12] have been performed. Before starting the experiment, the Reflolux was checked with the test solution, Reflolux II control for whole
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
229
blood measurement. For each analysis, 50 ml of blood was placed on the test strip. After 57 s, the blood was wiped off and the strip was inserted into the monitor and read after a total of 66 s. Granutest 100 (Merck Diagnostica, Darmstadt, Germany) is a spectrophotometric test for glucose in whole blood and is based on the reaction catalyzed by glucose dehydrogenase and the formation of NADH at 340 nm is monitored [13]. Venous blood was deproteinized by mixing 500 ml of 0.33 mol / l perchloric acid (Merck, 70–72%, pro analysis) with 50 ml of blood. After centrifugation (10,000 g.min), 100 ml of supernatant were added to 1 ml of the reaction solution. The absorbance at 340 nm was measured after 10 min against a blank, containing 100 ml of 0.33 mol / l perchloric acid in 1 ml of reaction solution. The difference in absorbance at 340 nm between the sample and the blank after a 10-min reaction time, multiplied by a suitable factor, was used to calculate the glucose concentration in mmol / l for whole blood samples. The Ektachem test kit (Kodak Clinical Diagnostics, Eastman Kodak, USA) measures the glucose concentration in plasma samples. The system is based on the glucose oxidase and peroxidase reactions. The chromogen formed in the peroxidase reaction is similar to that reported by Trinder [14]. The glucose concentration in whole blood was calculated as being 0.9 times the plasma concentration of glucose. The analyses for this study were performed at the University Hospital, Lund, Sweden.
2.7. Determination of the hematocrit value A glass capillary was filled with blood. After centrifugation, using the specified rotation speed and run time, the portion of the capillary filled with erythrocytes was measured and the hematocrit value calculated [15]. The measurements were performed at the University Hospital, Lund, Sweden.
2.8. Studies on various interferents The maximum physiological concentration of each interferent in blood [16] was added to a 5-mmol / l glucose solution in analysis buffer. The thermometric response of the glucose solution with the interferent was measured and compared with a 5 mmol / l glucose solution without the interferent. The interferents tested were ascorbic acid (0.11 mmol / l), uric acid (0.48 mmol / l), urea (4.3 mmol / l) and acetaminophen (0.17 mmol / l).
3. Results and discussion The difficulty in analyzing blood samples is due to the complex matrix of blood, which contains a variety of metabolites, proteins and hemocytes. In our
230
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
Fig. 2. Calibration curve showing the linear range for 1 ml samples.
system, undiluted blood passes through small diameter tubing and enters the enzyme column, where the reactions take place and heat is evolved. We have previously shown that at least 100 blood samples can be measured using the same enzyme column. This figure is a least ten-times higher for serum or plasma samples. The blood does not tend to clog the column immediately, but, after about 100 samples, the back pressure gradually increases until it becomes too high for the peristaltic pump employed. The analysis time for blood samples was 5 min compared to 4 min for glucose samples in analysis buffer. The linear range of the thermal biosensor was 0.5–16 mmol / l, with the sample volume chosen, as shown in Fig. 2.
3.1. Correlation A new measurement technique has to be compared with existing techniques and must agree sufficiently well with these methods to be accepted as an alternative method. Table 1 shows data for the reference methods compared with that for the thermal biosensor. The repeatability, or the within-run precision, of the methods to be compared is important since it limits the extent of agreement
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
231
Table 1 Comparison of the properties of the different methods Method
Linear range (mmol / l)
Detection limit (mmol / l)
Precision (%)
Biosensor Reflolux Merck Ektachem
0.5–16 0.5–27.7 1.0–40.0 0.99–29.8
0.5 0.5 1.0 0.99
5 5 5 2
that is possible between the two methods [17]. For the biosensor, this was obtained by repeatedly measuring a 5-mmol / l blood sample for a total of twenty times. The detection limit equals the concentration that gives a signal-to-noise ratio of two. The correlation graph between the thermal biosensor and the Ektachem is shown in Fig. 3. This is the most common method for comparison. Data from the regression lines obtained, using the values from the thermal biosensor on the
Fig. 3. Correlation graph between the thermal biosensor and the Ektachem method.
232
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
Table 2 Data from regression lines between the thermal biosensor ( y-axis) and the reference methods (x-axis) Method
Correlation coefficient (r)
Regression coefficient (b)
Intercept (a)
Reflolux Merck Ektachem
0.976 0.994 0.993
0.896 1.022 0.907
0.36 2 0.73 2 0.27
y-axis and the values from the reference methods on the x-axis, is shown in Table 2. The number of samples in the correlation experiments was 27–30. Another approach for estimating the agreement between the two methods is the plot suggested by Bland and Altman [18], where the difference between the methods is plotted against their mean. Fig. 4a–c show Altman–Bland plots of the comparison between the thermal biosensor and the reference methods. The plots indicate an overall positive mean difference bias of between 0.59 and 1.16 for the reference methods compared to the thermal biosensor. The correlation between the biosensor and the Ektachem using plasma samples gives a slope of 1.02 and an intercept of 0.09. The Altman–Bland plot for plasma samples is shown in Fig. 5. The limits of agreement are 2 0.14 to 0.7 mmol / l for plasma samples, whereas whole blood gives limits of agreement from 0.13 to 2.33 mmol / l. The deviation in Fig. 4a is larger at higher concentrations, which indicates that the empirical factor of 0.9, used for converting plasma values to whole blood values, is too large. There is no similar concentration dependence when plasma samples are compared (Fig. 5).
3.2. Influence of the hematocrit value The hematocrit value is the volume fraction of whole blood consisting of erythrocytes. The normal value is 40%, and is generally higher for men than for women. The glucose concentration of whole blood is a combination of the glucose concentration in the plasma and in the erythrocytes. The glucose concentration in plasma is higher than that inside the erythrocytes. This is due to the fact that the water content of plasma is higher (93%) than that in erythrocytes (73%), the glucose concentration in the liquid in these two compartments being the same [19]. At normal hematocrit values, the glucose concentration in plasma has been found to be 14–16% higher than that for whole blood [20]. Furthermore, the hematocrit value has been shown to influence the glucose concentration obtained for whole blood [14], especially for test-strip methods, giving false low readings for hematocrits above 50% and false high readings for hematocrits below 40%. The central assumption in
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
233
Fig. 4. Altman–Bland plots of the correlation between the reference methods and the thermal biosensor. (a) Mean 5 1.16 mmol / l; limits of agreement, 0.13–2.33 mmol / l; (b) mean 5 0.64 mmol / l; limits of agreement, 2 0.68–1.96 mmol / l; (c) mean 5 0.59 mmol / l; limits of agreement, 2 0.15–1.33 mmol / l.
234
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
Fig. 5. Altman–Bland plot of the correlation between the thermal biosensor using plasma samples. Mean, 0–0.28; limits of agreement, 2 0.14–0.7 mmol / l.
flow-injection analysis is that the dispersion of the sample is constant and reproducible, so that calibrating solutions and unknown samples can be treated in the same way. A change in the heterogeneity of the sample, that is, in the hematocrit value, would affect the dispersion and thereby the measured glucose concentration [20]. An increased hematocrit would give an increased dispersion and a decreased value of the glucose concentration. In Table 3, the glucose concentration, measured with the thermal biosensor and the Ektachem, is shown at different hematocrit values. The hematocrit value probably affects both methods. In the Ektachem method, no compensation is made for hematocrits differing from the normal (40%) and the same factor is used for the relationship between the glucose concentration Table 3 Glucose concentration measured in whole blood at different hematocrit values Hematocrit (%)
Ektachem (mmol / l)
Biosensor (mmol / l)
13 18 19 23 28 33 37 40 45 53
8.3 7.6 8.4 7.9 7.8 7.4 7.2 7.0 6.5 6.6
7.8 7.7 8.9 8.1 7.8 6.6 6.6 6.5 6.3 6.0
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
235
measured in plasma and the value given for whole blood (whole blood concentration 5 0.9 3 plasma concentration). This means that the method gives false low readings at low hematocrits and high readings at high hematocrits, although the deviation is small. The glucose concentration obtained with the biosensor correlates well with the values obtained with the Ektachem. There appears to be no significant difference, as might have been expected, at either high or low hematocrit values.
3.3. Influence of interferences Many sensors, especially amperometric biosensors, are sensitive to certain interfering substances. The thermal biosensor was tested at maximum physiological concentrations of different substances in a 5-mmol / l glucose solution, which are known to interfere with other sensors. None of the substances, ascorbic acid (0.11 mmol / l), uric acid (0.48 mmol / l), urea (4.3 mmol / l) and acetaminophen (0.17 mmol / l), had any effect on the thermometric response to 5 mmol / l glucose.
4. Conclusions The miniaturized thermal biosensor reported here is a flow-injection analysis system with which the glucose concentration in whole blood can be analysed directly. The measurement period is 5 min and at least 100 whole blood samples can be analysed using the same enzyme column. If serum or plasma samples are analyzed, the number of samples can be at least ten-times higher than for blood samples. The system is linear from 0.5 to 16 mmol / l glucose using 1 ml samples. This range may not be sufficient in all situations since the glucose concentration in diabetic patients can be much higher. The linear range can be extended to over 20 mmol / l by increasing the amount of enzyme in the column or by increasing the oxygen concentration in the buffer. The linear range of the flow-injection analysis system can also be increased if the sample volume is decreased. The limits of agreement between two methods indicate by how much a new method can differ from the reference method. An acceptable difference depends on the clinical use of the new method. The limits of agreement for the biosensor compared to the Ektachem are 0.13 to 2.33 mmol / l. A difference of 2 mmol / l is not always acceptable in clinical analysis. For plasma samples, the limits of agreement between the two methods are 2 0.14 to 0.7 mmol / l and the correlation is better than for whole blood samples. The glucose value for whole blood is always smaller when measured with the thermal biosensor compared to the reference methods, and this may indicate an incomplete analysis of the
236
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
glucose present in the erythrocytes, since the residence time in the enzyme reactor is only a few seconds. The correlation with methods such as the Ektachem for whole blood can be improved by choosing another method for calibration. Calibration using glucose dissolved in analysis buffer is sufficient for plasma samples but may not be sufficient for whole blood. One alternative calibration method is standard addition to whole blood and measurement of the recovery. Another way is calibration by comparison with another method, for example using the Ektachem test. By running a large number of samples in this way, it is possible to obtain a correction factor for bringing the biosensor data into closer agreement with a certain method. The hematocrit value does not seem to influence the obtained glucose concentration for hematocrit values between 13 and 53%. The heterogeneity of the sample in flow-injection analysis has previously been shown to have a great effect on the results, but, in the present system, the influence is limited. Another advantage is that the thermal biosensor is not affected by substances in concentrations known to interfere with other biosensors that have been used for glucose measurement.
Acknowledgements The financial support by Novo-Nordisk A / S, Denmark, The Swedish National Board for Technical Development, and SAREC is gratefully acknowledged.
References [1] Coulet PR. What is a biosensor? In: Blum J, Coulet PR, editors. Biosensors principles and applications. New York: Marcel Dekker, 1991. York. [2] Danielsson B. Calorimetric biosensors. J Biotechnol 1990;15:187–200. [3] Danielsson B, Gadd K, Mattiasson B, Mosbach K. Enzyme thermistor determination of glucose in serum using immobilized glucose oxidase. Clin Chim Acta 1977;81:163–75. [4] Rank M, Danielsson B, Gram J. Industrial on-line monitoring of penicillin V using a split-flow modified thermal biosensor. Anal Chim Acta 1993;281:521–6. [5] Xie B, Danielsson B, Winquist F. Miniaturized thermal biosensors. Sens and Act B 1993;15-16:443–7. [6] Xie B, Hedberg U, Danielsson B. Fast determinations of whole blood glucose with a calorimetric micro-biosensor. Sens and Act B 1993;15-16:141–4. [7] Pickup JC. Biosensors for diabetes mellitus. In: Wise DL, editor. Applied Biosensors. Boston; Butterworth, 1989. [8] Reach G, Wilson GS. Can continuous glucose monitoring be used for the treatment of diabetes. Anal Chem 1992;6:381–6. [9] Xie B, Harborn U, Mecklenburg M, Danielsson B. Urea and lactate determined in 1-mL whole-blood samples with a miniaturized thermal biosensor. Clin Chem 1994;40:2282–7.
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
237
[10] Weetall HH. Covalent coupling methods for inorganic support materials. Methods Enzymol 1976;44:139–41. [11] Brooks KE, Rawal N, Henderson A. Laboratory assessment of three new monitors of blood glucose: Accuchek II, Glukometer II and Glucoscan 2000. Clin Chem 1986;32:2195–200. [12] Devreese K, Leroux-Roels G. Laboratory assessment of five glucose meters designed for self-monitoring of blood glucose concentration. Eur J Clin Chem Clin Biochem 1993;31:29– 37. ¨ [13] Banauch D, Brummer W, Ebeling W, Metz H, Rindfrey H, Lang H et al. Glucose dehydrogenase for the determination of glucose concentrations in body fluids. Z Klin Chem Klin Biochem 1975;13:101–7. [14] Trinder P. Determination of glucose in blood using glucose oxidase with an alternative oxygen receptor. Ann Clin Biochem 1969;6:24–7. [15] Laurell C-B, Lundh B, Nosslin B. Klinisk kemi i praktisk medicin. Lund, Sweden: Studentlitteratur, 1976:271–272. [16] Moussy F, Harrison DJ, O’Brien DW, Rajotte RV. Performance of subcutaneously implanted needle-type glucose sensors employing a novel trilayer coating. Anal Chem 1993;65:2072–7. [17] Miller JC, Miller JN. Statistics for analytical chemistry. Chichester: Ellis Horwood, 1984. [18] Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;8:307–11. [19] Somogyi M. The distribution of sugar in normal human blood. J Biol Chem 1928;78:117–27. [20] Burrin JM, Alberti KGMM. What is blood glucose: Can it be measured?. Diabetic Med 1990;7:199–206.
Evaluation of a miniaturized thermal biosensor for the determination of glucose in whole blood a a b a, Ulrika Harborn , Bin Xie , Raghavan Venkatesh , Bengt Danielsson * a
Department of Pure and Applied Biochemistry, Chemical Center, P.O.B. 124, S-221 00 Lund, Sweden b PROMED, The Voluntary Health Services, Adyar, 600 113 Madras, India Received 6 March 1997; received in revised form 22 July 1997; accepted 30 July 1997
Abstract A miniaturized thermal biosensor has been evaluated as part of a flow-injection analysis system for the determination of glucose in whole blood. Glucose was determined by measuring the heat evolved when samples containing glucose passed through a small column with immobilized glucose oxidase and catalase. Samples of whole blood (1 ml) can be measured directly, without any pretreatment. The correlation in the response between the thermal biosensor, the Reflolux S meter (Boehringer Mannheim), the Granutest 100 glucose test kit (Merck Diagnostica) and the Ektachem (Kodak) instrument was evaluated. The influence of the hematocrit value and of possible interferences is reported. The correlation measurements show that the thermal biosensor calibrated with aqueous glucose standards generally gives lower values on blood glucose than the reference methods calibrated for serum or blood measurements. Mean negative biases range from 0.53 to 1.16 mmol / l. Differences in sample treatment clearly complicate comparisons and the proper choice of reference method. There was no influence from substances such as ascorbic acid (0.11 mmol / l), uric acid (0.48 mmol / l), urea (4.3 mmol / l) and acetaminophen (0.17 mmol / l) on the response to 5 mmol / l glucose. The hematocrit value does not influence the glucose determination, for hematocrit values of between 13 and 53%. 1997 Elsevier Science B.V. Keywords: Flow injection analysis; Thermal; Calorimetric; Biosensor; Glucose oxidase; Calibration aspects; Diabetes; Enzyme thermistor
1. Introduction The concept of a biosensor is defined as an analytical device that intimately *Corresponding author. Fax: 0046 46 22 24611. 0009-8981 / 97 / $17.00 1997 Elsevier Science B.V. All rights reserved. PII S0009-8981( 97 )00151-4
226
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
associates a biological sensing element with a transducer and responds selectively and reversibly to the concentration or activity of chemical species in biological samples [1]. In thermal biosensors, the enzymes or cells immobilized and filled into a column are the biological sensing material and the thermistors in close contact to the column are the transducers [2]. The heat evolved during the enzyme-catalyzed reaction is in proportion to the amount of substrate in the sample. Thermal biosensors have previously proved to work well with biological samples, such as serum [3] and fermentation broth [4]. Miniaturization of thermal biosensors has led to the use of small sample volumes, giving a wider linear range and a reduced response time [5]. In an earlier report, we demonstrated that samples of whole blood can be applied directly in a miniaturized flow system for the determination of glucose [6]. By using a sample volume as small as 1 ml, whole blood can be directly injected without any predilution. The massive interest in developing analytical methods for the determination of glucose in whole blood is due to the fact that many people suffering from diabetes are dependent on these methods in their daily lives. Many biosensors, mostly electrochemical, have been developed for this purpose [7]. The ultimate goal for biosensor research in this field is the development of an in vivo sensor in combination with an insulin pump for the continuous regulation of the glucose concentration in diabetic patients [8]. Measurement of whole blood glucose with a flow-injection thermal biosensor can lead to the development of miniaturized sensors for home monitoring as well as to equipment for the continuous monitoring of glucose. Automatic regulation of the insulin dosage needed for patient use or for diabetes research is also actively being considered. The introduction of new measurement technologies requires comparison with established methods for validation, but it may be difficult to find such methods that work in the same medium and with the same calibrators. This problem is discussed in the present paper in which the precision and accuracy of the thermal biosensor is compared with two commercially available methods for glucose determination in whole blood and one common clinical method for plasma glucose determination, which is corrected to give whole blood values. The effect of the hematocrit value on the glucose concentration determination is reported.
2. Materials and methods
2.1. The thermal biosensor The miniaturized thermal biosensor used in this study has been described elsewhere [9]. The most important part of the sensor is the enzyme column,
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
227
Fig. 1. Schematic diagram of the calorimeter of the sensor.
which is made of a thin-walled acid-proof steel tube (1.5 mm, I.D.; length 15 mm) containing porous beads with immobilized enzymes. Buffer enters the sensor through the equilibration coil, composed of acid-proof steel tubing wound inside the cylindrical heat-sink (Fig. 1). In order to minimize heat leakage from the column, the effluent from the column passes through steel tubing that is wound around a copper tube (adiabatic shield) surrounding the column. The heat evolved during the enzyme-catalyzed reaction is measured with thermistors placed on 0.2 mm (I.D.) gold tubes at the inlet and outlet of the column.
2.2. Experimental set-up A peristaltic pump (Gilson, Minipuls 2, France) continuously pumps 70 ml / min of 0.1 mol / l sodium phosphate-buffered saline, pH 7.0, containing 1 g of EDTA (ethylenediaminetetraacetic acid, disodium salt) and 4 g of sodium fluoride per litre (analysis buffer), through the sensor. Samples were injected through a 1-ml injection valve (C14W, VIGI AG, Valco Europe, Switzerland). The heat evolved in the enzyme column was detected with a Wheatstone bridge and the signal was amplified. During the experiments, the sensor was placed in a small aluminium box that was insulated with polyurethane foam.
2.3. Enzyme immobilization The glutaraldehyde activation step was performed according to Weetall [10] on pre-silanized spherical CPG (controlled pore glass) beads (Trisoperl, size 125–140 mm, pore size 50 nm; Schuller, Steinach, Germany). The enzymes, 5 mg of glucose oxidase (EC 1.1.3.4) from Aspergillus niger (166 U / mg, Sigma)
228
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
and 6 mg of catalase (EC 1.11.1.6) from bovine liver (19 900 U / mg, Sigma) were added to 500 mg (wet weight) of activated beads in 1 ml of 0.1 mol / l sodium phosphate buffer, pH 7.0 (coupling buffer). The glucose oxidase was added first and allowed to react for 1 h. Excess activated groups were blocked using 20 mg of bovine serum albumin (Fraction V, 99%, Sigma) in 1 ml of coupling buffer. The beads were finally washed in the coupling buffer followed by the buffer used for the glucose determinations.
2.4. Calibration A 40-mmol / l standard solution of D( 1 )-glucose (monohydrate, Merck) was prepared in the analysis buffer 24 h before use. Glucose standards, at concentrations of 1–20 mmol / l, were made from the standard solution and a calibration curve was constructed, based on the heat evolved as a function of glucose concentration.
2.5. Collection and pretreatment of blood samples Capillary blood was collected when using the Reflolux glucose meter as the reference instrument (see Section 2.6). The finger tip was punctured, using a small lancet, and 500 ml of blood were collected in test-tubes containing 2 mg of ¨ sodium fluoride and 14 IU of sodium heparin (25 000 U / ml; Loevens, Malmo, Sweden). Venous blood was used in all other experiments. Blood was drawn in vacutainer tubes (5 ml, Becton Dickinson) containing 20 mg of sodium fluoride and 143 IU of heparin. In order to obtain a set of whole blood samples with glucose concentrations in the linear range of the thermal biosensor with minimal dilution, small volumes of a highly concentrated glucose solution were added to a comparably large volume of blood. This glucose solution was prepared in the analysis buffer. Blood samples with different hematocrit values were obtained by adding or subtracting certain amounts of plasma from the original blood sample taken from the vein.
2.6. Reference methods Reflolux S (Boehringer Mannheim Scandinavia, Bromma, Sweden) is a portable glucose monitor that measures the colour change of test strips exposed to whole blood. The test strips contain glucose oxidase, catalase and colour reagents. A drop of blood is placed on the test strip. After the reaction is completed, the blood is wiped off with Kleenex tissue. The entire measurement takes 2 min and the glucose concentration is displayed in ‘‘mM’’. Previous studies of the Reflolux instrument compared to other portable meters [11] and to the Kodak Ektachem [12] have been performed. Before starting the experiment, the Reflolux was checked with the test solution, Reflolux II control for whole
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
229
blood measurement. For each analysis, 50 ml of blood was placed on the test strip. After 57 s, the blood was wiped off and the strip was inserted into the monitor and read after a total of 66 s. Granutest 100 (Merck Diagnostica, Darmstadt, Germany) is a spectrophotometric test for glucose in whole blood and is based on the reaction catalyzed by glucose dehydrogenase and the formation of NADH at 340 nm is monitored [13]. Venous blood was deproteinized by mixing 500 ml of 0.33 mol / l perchloric acid (Merck, 70–72%, pro analysis) with 50 ml of blood. After centrifugation (10,000 g.min), 100 ml of supernatant were added to 1 ml of the reaction solution. The absorbance at 340 nm was measured after 10 min against a blank, containing 100 ml of 0.33 mol / l perchloric acid in 1 ml of reaction solution. The difference in absorbance at 340 nm between the sample and the blank after a 10-min reaction time, multiplied by a suitable factor, was used to calculate the glucose concentration in mmol / l for whole blood samples. The Ektachem test kit (Kodak Clinical Diagnostics, Eastman Kodak, USA) measures the glucose concentration in plasma samples. The system is based on the glucose oxidase and peroxidase reactions. The chromogen formed in the peroxidase reaction is similar to that reported by Trinder [14]. The glucose concentration in whole blood was calculated as being 0.9 times the plasma concentration of glucose. The analyses for this study were performed at the University Hospital, Lund, Sweden.
2.7. Determination of the hematocrit value A glass capillary was filled with blood. After centrifugation, using the specified rotation speed and run time, the portion of the capillary filled with erythrocytes was measured and the hematocrit value calculated [15]. The measurements were performed at the University Hospital, Lund, Sweden.
2.8. Studies on various interferents The maximum physiological concentration of each interferent in blood [16] was added to a 5-mmol / l glucose solution in analysis buffer. The thermometric response of the glucose solution with the interferent was measured and compared with a 5 mmol / l glucose solution without the interferent. The interferents tested were ascorbic acid (0.11 mmol / l), uric acid (0.48 mmol / l), urea (4.3 mmol / l) and acetaminophen (0.17 mmol / l).
3. Results and discussion The difficulty in analyzing blood samples is due to the complex matrix of blood, which contains a variety of metabolites, proteins and hemocytes. In our
230
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
Fig. 2. Calibration curve showing the linear range for 1 ml samples.
system, undiluted blood passes through small diameter tubing and enters the enzyme column, where the reactions take place and heat is evolved. We have previously shown that at least 100 blood samples can be measured using the same enzyme column. This figure is a least ten-times higher for serum or plasma samples. The blood does not tend to clog the column immediately, but, after about 100 samples, the back pressure gradually increases until it becomes too high for the peristaltic pump employed. The analysis time for blood samples was 5 min compared to 4 min for glucose samples in analysis buffer. The linear range of the thermal biosensor was 0.5–16 mmol / l, with the sample volume chosen, as shown in Fig. 2.
3.1. Correlation A new measurement technique has to be compared with existing techniques and must agree sufficiently well with these methods to be accepted as an alternative method. Table 1 shows data for the reference methods compared with that for the thermal biosensor. The repeatability, or the within-run precision, of the methods to be compared is important since it limits the extent of agreement
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
231
Table 1 Comparison of the properties of the different methods Method
Linear range (mmol / l)
Detection limit (mmol / l)
Precision (%)
Biosensor Reflolux Merck Ektachem
0.5–16 0.5–27.7 1.0–40.0 0.99–29.8
0.5 0.5 1.0 0.99
5 5 5 2
that is possible between the two methods [17]. For the biosensor, this was obtained by repeatedly measuring a 5-mmol / l blood sample for a total of twenty times. The detection limit equals the concentration that gives a signal-to-noise ratio of two. The correlation graph between the thermal biosensor and the Ektachem is shown in Fig. 3. This is the most common method for comparison. Data from the regression lines obtained, using the values from the thermal biosensor on the
Fig. 3. Correlation graph between the thermal biosensor and the Ektachem method.
232
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
Table 2 Data from regression lines between the thermal biosensor ( y-axis) and the reference methods (x-axis) Method
Correlation coefficient (r)
Regression coefficient (b)
Intercept (a)
Reflolux Merck Ektachem
0.976 0.994 0.993
0.896 1.022 0.907
0.36 2 0.73 2 0.27
y-axis and the values from the reference methods on the x-axis, is shown in Table 2. The number of samples in the correlation experiments was 27–30. Another approach for estimating the agreement between the two methods is the plot suggested by Bland and Altman [18], where the difference between the methods is plotted against their mean. Fig. 4a–c show Altman–Bland plots of the comparison between the thermal biosensor and the reference methods. The plots indicate an overall positive mean difference bias of between 0.59 and 1.16 for the reference methods compared to the thermal biosensor. The correlation between the biosensor and the Ektachem using plasma samples gives a slope of 1.02 and an intercept of 0.09. The Altman–Bland plot for plasma samples is shown in Fig. 5. The limits of agreement are 2 0.14 to 0.7 mmol / l for plasma samples, whereas whole blood gives limits of agreement from 0.13 to 2.33 mmol / l. The deviation in Fig. 4a is larger at higher concentrations, which indicates that the empirical factor of 0.9, used for converting plasma values to whole blood values, is too large. There is no similar concentration dependence when plasma samples are compared (Fig. 5).
3.2. Influence of the hematocrit value The hematocrit value is the volume fraction of whole blood consisting of erythrocytes. The normal value is 40%, and is generally higher for men than for women. The glucose concentration of whole blood is a combination of the glucose concentration in the plasma and in the erythrocytes. The glucose concentration in plasma is higher than that inside the erythrocytes. This is due to the fact that the water content of plasma is higher (93%) than that in erythrocytes (73%), the glucose concentration in the liquid in these two compartments being the same [19]. At normal hematocrit values, the glucose concentration in plasma has been found to be 14–16% higher than that for whole blood [20]. Furthermore, the hematocrit value has been shown to influence the glucose concentration obtained for whole blood [14], especially for test-strip methods, giving false low readings for hematocrits above 50% and false high readings for hematocrits below 40%. The central assumption in
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
233
Fig. 4. Altman–Bland plots of the correlation between the reference methods and the thermal biosensor. (a) Mean 5 1.16 mmol / l; limits of agreement, 0.13–2.33 mmol / l; (b) mean 5 0.64 mmol / l; limits of agreement, 2 0.68–1.96 mmol / l; (c) mean 5 0.59 mmol / l; limits of agreement, 2 0.15–1.33 mmol / l.
234
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
Fig. 5. Altman–Bland plot of the correlation between the thermal biosensor using plasma samples. Mean, 0–0.28; limits of agreement, 2 0.14–0.7 mmol / l.
flow-injection analysis is that the dispersion of the sample is constant and reproducible, so that calibrating solutions and unknown samples can be treated in the same way. A change in the heterogeneity of the sample, that is, in the hematocrit value, would affect the dispersion and thereby the measured glucose concentration [20]. An increased hematocrit would give an increased dispersion and a decreased value of the glucose concentration. In Table 3, the glucose concentration, measured with the thermal biosensor and the Ektachem, is shown at different hematocrit values. The hematocrit value probably affects both methods. In the Ektachem method, no compensation is made for hematocrits differing from the normal (40%) and the same factor is used for the relationship between the glucose concentration Table 3 Glucose concentration measured in whole blood at different hematocrit values Hematocrit (%)
Ektachem (mmol / l)
Biosensor (mmol / l)
13 18 19 23 28 33 37 40 45 53
8.3 7.6 8.4 7.9 7.8 7.4 7.2 7.0 6.5 6.6
7.8 7.7 8.9 8.1 7.8 6.6 6.6 6.5 6.3 6.0
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
235
measured in plasma and the value given for whole blood (whole blood concentration 5 0.9 3 plasma concentration). This means that the method gives false low readings at low hematocrits and high readings at high hematocrits, although the deviation is small. The glucose concentration obtained with the biosensor correlates well with the values obtained with the Ektachem. There appears to be no significant difference, as might have been expected, at either high or low hematocrit values.
3.3. Influence of interferences Many sensors, especially amperometric biosensors, are sensitive to certain interfering substances. The thermal biosensor was tested at maximum physiological concentrations of different substances in a 5-mmol / l glucose solution, which are known to interfere with other sensors. None of the substances, ascorbic acid (0.11 mmol / l), uric acid (0.48 mmol / l), urea (4.3 mmol / l) and acetaminophen (0.17 mmol / l), had any effect on the thermometric response to 5 mmol / l glucose.
4. Conclusions The miniaturized thermal biosensor reported here is a flow-injection analysis system with which the glucose concentration in whole blood can be analysed directly. The measurement period is 5 min and at least 100 whole blood samples can be analysed using the same enzyme column. If serum or plasma samples are analyzed, the number of samples can be at least ten-times higher than for blood samples. The system is linear from 0.5 to 16 mmol / l glucose using 1 ml samples. This range may not be sufficient in all situations since the glucose concentration in diabetic patients can be much higher. The linear range can be extended to over 20 mmol / l by increasing the amount of enzyme in the column or by increasing the oxygen concentration in the buffer. The linear range of the flow-injection analysis system can also be increased if the sample volume is decreased. The limits of agreement between two methods indicate by how much a new method can differ from the reference method. An acceptable difference depends on the clinical use of the new method. The limits of agreement for the biosensor compared to the Ektachem are 0.13 to 2.33 mmol / l. A difference of 2 mmol / l is not always acceptable in clinical analysis. For plasma samples, the limits of agreement between the two methods are 2 0.14 to 0.7 mmol / l and the correlation is better than for whole blood samples. The glucose value for whole blood is always smaller when measured with the thermal biosensor compared to the reference methods, and this may indicate an incomplete analysis of the
236
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
glucose present in the erythrocytes, since the residence time in the enzyme reactor is only a few seconds. The correlation with methods such as the Ektachem for whole blood can be improved by choosing another method for calibration. Calibration using glucose dissolved in analysis buffer is sufficient for plasma samples but may not be sufficient for whole blood. One alternative calibration method is standard addition to whole blood and measurement of the recovery. Another way is calibration by comparison with another method, for example using the Ektachem test. By running a large number of samples in this way, it is possible to obtain a correction factor for bringing the biosensor data into closer agreement with a certain method. The hematocrit value does not seem to influence the obtained glucose concentration for hematocrit values between 13 and 53%. The heterogeneity of the sample in flow-injection analysis has previously been shown to have a great effect on the results, but, in the present system, the influence is limited. Another advantage is that the thermal biosensor is not affected by substances in concentrations known to interfere with other biosensors that have been used for glucose measurement.
Acknowledgements The financial support by Novo-Nordisk A / S, Denmark, The Swedish National Board for Technical Development, and SAREC is gratefully acknowledged.
References [1] Coulet PR. What is a biosensor? In: Blum J, Coulet PR, editors. Biosensors principles and applications. New York: Marcel Dekker, 1991. York. [2] Danielsson B. Calorimetric biosensors. J Biotechnol 1990;15:187–200. [3] Danielsson B, Gadd K, Mattiasson B, Mosbach K. Enzyme thermistor determination of glucose in serum using immobilized glucose oxidase. Clin Chim Acta 1977;81:163–75. [4] Rank M, Danielsson B, Gram J. Industrial on-line monitoring of penicillin V using a split-flow modified thermal biosensor. Anal Chim Acta 1993;281:521–6. [5] Xie B, Danielsson B, Winquist F. Miniaturized thermal biosensors. Sens and Act B 1993;15-16:443–7. [6] Xie B, Hedberg U, Danielsson B. Fast determinations of whole blood glucose with a calorimetric micro-biosensor. Sens and Act B 1993;15-16:141–4. [7] Pickup JC. Biosensors for diabetes mellitus. In: Wise DL, editor. Applied Biosensors. Boston; Butterworth, 1989. [8] Reach G, Wilson GS. Can continuous glucose monitoring be used for the treatment of diabetes. Anal Chem 1992;6:381–6. [9] Xie B, Harborn U, Mecklenburg M, Danielsson B. Urea and lactate determined in 1-mL whole-blood samples with a miniaturized thermal biosensor. Clin Chem 1994;40:2282–7.
U. Harborn et al. / Clinica Chimica Acta 267 (1997) 225 – 237
237
[10] Weetall HH. Covalent coupling methods for inorganic support materials. Methods Enzymol 1976;44:139–41. [11] Brooks KE, Rawal N, Henderson A. Laboratory assessment of three new monitors of blood glucose: Accuchek II, Glukometer II and Glucoscan 2000. Clin Chem 1986;32:2195–200. [12] Devreese K, Leroux-Roels G. Laboratory assessment of five glucose meters designed for self-monitoring of blood glucose concentration. Eur J Clin Chem Clin Biochem 1993;31:29– 37. ¨ [13] Banauch D, Brummer W, Ebeling W, Metz H, Rindfrey H, Lang H et al. Glucose dehydrogenase for the determination of glucose concentrations in body fluids. Z Klin Chem Klin Biochem 1975;13:101–7. [14] Trinder P. Determination of glucose in blood using glucose oxidase with an alternative oxygen receptor. Ann Clin Biochem 1969;6:24–7. [15] Laurell C-B, Lundh B, Nosslin B. Klinisk kemi i praktisk medicin. Lund, Sweden: Studentlitteratur, 1976:271–272. [16] Moussy F, Harrison DJ, O’Brien DW, Rajotte RV. Performance of subcutaneously implanted needle-type glucose sensors employing a novel trilayer coating. Anal Chem 1993;65:2072–7. [17] Miller JC, Miller JN. Statistics for analytical chemistry. Chichester: Ellis Horwood, 1984. [18] Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;8:307–11. [19] Somogyi M. The distribution of sugar in normal human blood. J Biol Chem 1928;78:117–27. [20] Burrin JM, Alberti KGMM. What is blood glucose: Can it be measured?. Diabetic Med 1990;7:199–206.