In 1987 Ash and Janle described a small ultrafiltration probe that can be placed in the subcutaneous space.1 A small polyester cuff stabilizes a 1 mm diameter tube through the skin, and under negative pressure as provided by a Vacutainer tube (Becton, Dickinson & Co., Franklin Lakes, NJ), the device will deliver 50 μl per hour of protein-free fluid for a month of use or more.2 This filtrate has the same glucose and small molecule concentration as plasma free water, after accounting for a modest time delay for convection of the water out of capillaries, through the subcutaneous space, into the hollow fibers and out to the point of collection.3–5 Ultrafiltration sampling is easier to perform than microdialysis since no perfusate is needed, but flow through the fibers is less constant.
For years we have looked for a method of measuring glucose in this filtrate that will provide stable readings for many days or weeks. Glucose oxidase electrodes measure the glucose level accurately, but recalibration is still needed at least daily.6 Over the years, we have investigated mid-infrared measurements, oncotic pressure (with a glucose-rejecting membrane), and vapor pressure (a measure of osmolality), and none seemed practical.
Recently, a sensor for fluid density based on a microelectromechanical system chip was described with very high accuracy (to 4–5 decimal points).7 In vitro tests confirmed accuracy of measurement of glucose in solutions within 20 mg/dl in single point measurements. We reasoned that, if we used a conductivity probe to measure effects on density by all plasma water electrolytes and eliminated most macromolecules by ultrafiltration, we could demonstrate a correlation between plasma water glucose and density. Glucose has a much higher concentration (in mass/volume) than any other plasma water component except electrolytes, urea, and (occasionally) ethanol.
In Vitro Tests
A model for the effects of varying glucose and electrolyte concentrations was needed. The density of aqueous solutions of both glucose and sodium chloride were found in the CRC Handbook of Chemistry and Physics.8 Test solutions of glucose and sodium chloride were made to understand the effect that varying concentrations have on the density of solutions. Standardized solutions of 2,000 mg/dl glucose and 0.4 M sodium chloride were used to obtain 12 samples with varying levels of glucose and sodium chloride within physiological ranges. To measure the density of the samples, a model 106A In-Line Density Meter from Integrated Sensing Systems, Inc. (ISSYS, Ypsilanti, MI) was used which required only 10 μl samples. A Myron L. Company Ultrameter Model 6P (Myron L. Co., Carlsbad, CA) was used for the conductivity measurements using 50 μl samples. The data for the conductivity of test solutions containing no glucose and varying levels of sodium chloride gave a direct linear relationship between conductivity and sodium chloride. Using this relationship, the sodium chloride concentrations from the CRC data were converted into conductivity readings. Based on the linear relationship between conductivity and density for sodium chloride solutions and the separate linear relationship between glucose and density, a multivariate equation was obtained that related the glucose concentration to density and conductivity.
This assumes that the contributions to density from glucose and sodium chloride are additive. The relationship can be seen in Figure 1. The glucose concentration increases as the density increases and decreases as the conductivity increases.
The level of electrolytes (sodium and associated anions) is an important contributor to the conductivity although glucose has only a small effect on conductivity. To verify this assumption, the conductivity for three different concentrations of sodium, between 0 and 200 mmol/L, and varying glucose levels were measured and are shown in Figure 2. The conductivity decreases only slightly as the glucose increases. The reason for this decrease is not currently understood but probably relates to interactions between glucose, electrolytes, and water. However, it is clear that the level of sodium and associated anions are important contributors to conductivity.
After IRB approval, blood plasma samples from seven patients with diabetes were obtained from a local hospital, before they were planned to be discarded. Samples were chosen from patients with diabetes having glucose in range of 50–400 mg/dl and several samples per day, in patients without renal failure (creatinine <2 mg/dl). For each patient, 3–12 samples contained sufficient volume for the test for a total of 48 samples tested. The hospital supplied data on sodium and glucose concentration of the samples. For the first two patients, the blood plasma was filtered using Microsep 30k and Nanosep 30k centrifugal devices (Pall Corporation, Inc., Ann Arbor, MI), giving a filtrate volume of a few microliters. Samples of blood plasma for patients 3–7 were filtered through Microsep 10k centrifugal devices. The smaller molecular weight filter was used to ensure that most of the albumin in the plasma was removed. For patients 1–6, the filtrate was drawn into the ISSYS model 106A In-Line Density Meter using a syringe. The temperature reading from the density meter was monitored to ensure a constant temperature within 0.5°C of room temperature. The density had minimal fluctuations, within 10 mg/ml. For patient seven, an ISSYS model FC6 In-Line Density Meter was used because it required a smaller sample volume. Both of these meters are accurate to more than four decimal places. A flow-through conductivity probe (Lazar Research Laboratories, Inc., Los Angeles, CA) was used at the 100× setting for all of the conductivity measurements. These conductivity readings had significant fluctuations before reaching a stable conductivity value which was recorded.
When a multivariable linear regression is fit to the data, the overall direction of the plane shows that the glucose increases as the density increases and decreases as the conductivity increases (Figure 3). The relationship between glucose and density and conductivity in plasma ultrafiltrate for patients 1–7 was the same as the predicted model. However the individual scatter of the points was very high. The correlation coefficient for this regression was 0.45, which is significant at α = 0.05 using a t test for zero correlation. However, the coefficient of determination was 0.20 which indicates that only 20% of the variability in the glucose concentration can be explained by the regression.
A multiple linear regression was performed on the data for patients 1–7. Patients 2 and 3 had significant correlation coefficients of 0.79 and 0.88, respectively, based on the t test for zero correlation. However, the correlation coefficients for the other five patients were not significantly different from zero using α = 0.05. Out of the two sets of data with significant correlation, only the linear regression for patient 3, which had the largest glucose range, had the same relationships as the predicted model.
The concentration of glucose in blood plasma cannot be predicted from only the density of the plasma ultrafiltrate. The total electrolyte concentration can be reasonably predicted from the conductivity of the filtered blood plasma in individual patients. It was hypothesized that estimating the electrolyte concentration from the conductivity would refine the density measurement enough to give an accurate glucose concentration estimate. Based on the data collected, the glucose level of a blood plasma sample from a patients with diabetes could not be accurately predicted by measuring the density and conductivity of filtered blood plasma. The measurements only significantly correlated in two out of seven patients with only one patient following the predicted model. Currently, these measurements are not a feasible approach to predicting blood glucose.
In short, plasma water density in patients with diabetes does not correlate with blood glucose levels in most patients, even after correction for changes in conductivity (a measure of blood electrolyte concentration). There are a number of reasons that could be postulated. One explanation is that there are random changes in other blood components such as amino acids, which disguise the density signal from glucose. These components are probably not small charged solutes or they would have been detected and corrected by the conductivity measurement. Rather than time specific, the appearance of chemicals seems to be patient-specific. Incidentally, the study was not controlled for whether the patients had kidney disease, which could change the levels of many organic compounds and the body’s response to osmolar changes. Alternatively, glucose metabolism or the lack of it might have differing effects on various metabolites in the blood, increasing some components while decreasing others. Glucose can be a diuretic, but usually this effect is not significant at glucose levels <400 mg/dl. Elevated glucose is known to correlate with idiogenic osmoles such as ketone bodies as measured in plasma. These osmoles would be expected to also increase plasma water density, but they would also transfer water from cells into plasma. There is evidence that intracellular osmolytes such as glutamine and taurine increased after treatment of DKA, especially in the brain.9 These osmoles could draw water into cells from the plasma. The effects of alcohol dehydrogenase and osmoregulation will of course dilute plasma water whenever glucose increases.10,11 However, all of these water transfers should either increase or decrease electrolyte concentrations, which should be reflected by our conductivity measurements. If glucose increases and electrolyte changes are accounted for by conductivity, we would still expect to be able to measure an increase in plasma water density due to glucose. Changes in urea concentrations due to plasma dilution or concentration should not be large enough to offset the expected changes in density due to glucose, and the urea level should change in the same direction as electrolytes during water addition or removal.
We write this brief report to relay an unexplained observation. Perhaps some reader can explain why conductivity and density measurements of an ultrafiltrate do not correlate with plasma glucose. In the meantime, we continue to search for other physical methods to measure glucose in plasma water, and await an electrochemical sensor with long term stability.
1. Janle-Swain E, Turek J, Van Vleet JF, Ash SR: Use of a capillary filtrate collector for monitoring glucose in diabetics. ASAIO Trans
33: 336–340, 1987.
2. Ash SR, Zopp WE, Truitt RB, Janle EM, et al
: Subcutaneous ultrafiltration fibers for chemical sampling of blood: The capillary filtrate collector (CFC). Am Soc Filtration
7: 316–319, 1993.
3. Ash SR, Poulos JT, Rainier JB, et al
: Subcutaneous capillary filtrate collector for measurement of blood glucose. ASAIO J
38: M416–420, 1992.
4. Ash SR, Rainier JB, Zopp WE, et al
: A subcutaneous capillary filtrate collector for measurement of blood chemistries. ASAIO J
39: M699–M705, 1993.
5. Janle EM, Cregor M: Ultrafiltrate and microdialysis DL probe in vitro recoveries: Electrolytes and metabolites. Curr Sep
15: 31–34, 1996.
6. Leegsma-Vogt G, Janle E, Ash SR, et al
: Utilization of in vivo ultrafiltration in biomedical research and clinical applications. Life Sci
73: 2005–2018, 2003.
7. Sparks D, Smith R, Straayer M, et
al: Measurement of density and chemical concentration using a microfluidic chip. Lab Chip
3: 19–21, 2003.
8. Lide, David R. (eds): CRC Handbook of Chemistry and Physics
, 86th ed. Boca Raton: CRC Press, 2005.
9. Silver SM, Clark EC, Schroeder BM, Sterns RH: Pathogenesis of cerebral edema after treatment of diabetic ketoacidosis. Kidney Int
51: 1237–1244, 1997.
10. Durr JA, Hoffman WH, Hensen J, et al
: Osmoregulation of vasopressin in diabetic ketoacidosis. Am J Physiol
259: E723–E728, 1990.
11. Thompson CJ, Davis SN, Baylis PH: Effect of blood glucose concentration on osmoregulation in diabetes mellitus. Am J Physiol
256: R597–R604, 1989.