Whole-Body Glycolysis Measured by the Deuterated-Glucose Disposal Test Correlates Highly With Insulin Resistance In Vivo

  1. Carine Beysen, DPHIL1,
  2. Elizabeth J. Murphy, MD, DPHIL12,
  3. Tracey McLaughlin, MD3,
  4. Timothy Riiff, BS, MSE1,
  5. Cindy Lamendola, BA3,
  6. Holly C. Turner, BS1,
  7. Mohamad Awada, PHD1,
  8. Scott M. Turner, PHD1,
  9. Gerald Reaven, MD3 and
  10. Marc K. Hellerstein, MD, PHD24
  1. 1KineMed, Inc., Emeryville, California
  2. 2San Francisco General Hospital, University of California at San Francisco, San Francisco, California
  3. 3Stanford University, Palo Alto, California
  4. 4Nutritional Sciences and Toxicology, University of California at Berkeley, California
  1. Address correspondence and reprint requests to Marc Hellerstein, MD, PhD, Nutritional Sciences and Toxicology, 309 Morgan Hall, University of California at Berkeley, Berkeley, CA 94720. E-mail: march{at}nature.berkeley.edu

Abstract

OBJECTIVE—The purpose of this study was to compare an in vivo test of whole-body glycolysis, the deuterated-glucose disposal test (2H-GDT), with insulin sensitivity measured by the euglycemic-hyperinsulinemic glucose clamp and the steady-state plasma glucose (SSPG) test.

RESEARCH DESIGN AND METHODS—The 2H-GDT consists of an oral glucose challenge containing deuterated glucose, followed by measurement of heavy water (2H2O) production, which represents whole-body glycolytic disposal of the glucose load. 2H2O production is corrected for ambient insulin concentration as an index of tissue insulin sensitivity. The 2H-GDT was compared with euglycemic-hyperinsulinemic glucose clamps in healthy lean subjects (n = 8) and subjects with the metabolic syndrome (n = 9) and with the SSPG test in overweight (n = 12) and obese (n = 6) subjects.

RESULTS—A strong correlation with the clamp was observed for the 75-g and 30-g 2H-GDT (r = 0.95, P < 0.0001 and r = 0.88, P < 0.0001, respectively). The 2H-GDT and clamp studies revealed marked insulin resistance in subjects with metabolic syndrome compared with lean control subjects. The correlation with the clamp was maintained in each group (lean, r = 0.86, P < 0.01; metabolic syndrome, r = 0.81, P < 0.01) for the 75-g test. The 2H-GDT also correlated strongly with the SSPG test (r = −0.87, P < 0.0001) in overweight and obese subjects.

CONCLUSIONS—The 2H-GDT, which measures whole-body glycolysis in humans in a quantitative manner, correlates highly with the euglycemic-hyperinsulinemic glucose clamp and the SSPG test. Impaired insulin-mediated whole-body glycolysis is a feature of insulin resistance, which provides a means of assessing insulin sensitivity in vivo.

Insulin resistance is a highly prevalent condition (1) and an important risk factor for the development of type 2 diabetes (2,3) and cardiovascular disease (4,5). Insulin resistance has thus become a public health issue of central importance (6). Several approaches have been used for measurement of insulin resistance in vivo.

The gold standard technique, the euglycemic-hyperinsulinemic glucose clamp (7), measures insulin-mediated glucose disposal by peripheral tissues. The clamp is labor intensive and burdensome for subjects, precluding routine use. Other direct measures of insulin resistance, such as the steady-state plasma glucose (SSPG) test (8,9) and the frequently sampled intravenous glucose tolerance test (10), also primarily measure whole-body glucose disposal and are similarly labor intensive. Indirect markers of insulin resistance, such as serum insulin concentrations, BMI, waist circumference, and serum triacylglycerol concentrations, have limited utility. BMI, for example, is not a good surrogate for insulin resistance, as 16% of individuals with insulin resistance are lean, whereas 30% of insulin-sensitive individuals are obese or overweight (11). Other surrogate measures model the relationship between glucose and insulin. The homeostasis model assessment (HOMA) (12,13), quantitative insulin sensitivity check index (QUICKI) (14), and models based on the oral glucose tolerance test (15) have correlated well with the clamp in some studies, but, often, the correlation has been poor (r2 = ∼0.50) (1619), especially in normal-weight individuals (19). Attempts to define a clinical entity (i.e., the metabolic syndrome) (20) or combine parameters (e.g., BMI and HOMA of insulin resistance) (21) to establish insulin resistance have proven to be insensitive for detecting it (22,23) and do not provide a continuous measure for monitoring treatment response (21).

Any measurement test must be based on the known physiology of insulin resistance. Most attention has been given to the effects of insulin resistance on glycogen storage in tissues such as skeletal muscle, but glycolytic disposal of glucose shares the steps of glucose transport and phosphorylation, which are likely to play key roles in insulin sensitivity (24), and glycolytic enzymes are sensitive to insulin. Moreover, at serum insulin concentrations in the “dynamic range” between basal and maximal (i.e., between ∼100 and 1000 pmol/l), glycolytic disposal of glucose closely parallels glycogen storage in clamp studies and is impaired to a similar extent by insulin resistance (25,26). Measurement of glycolysis has a major advantage over glycogen synthesis, however, because tissue glycogen is not readily accessible to sampling, whereas an immediate product of glycolysis (hydrogen atoms released to tissue water) can be tagged, traced, and sampled in body fluids.

Here, we describe a stable isotope-mass spectrometric test of whole-body glycolysis, the deuterated-glucose disposal test (2H-GDT), and compare this index of in vivo glucose utilization to established measures of insulin resistance. The 2H-GDT measures the rate of uptake, phosphorylation, and glycolytic metabolism of glucose in response to a physiologic glucose load. Our goal here was to determine whether insulin-mediated whole-body glycolysis is a quantitative measure of insulin resistance in humans, based on correlation with the clamp and SSPG test in subjects across a range of insulin sensitivities.

RESEARCH DESIGN AND METHODS

Model

The 2H-GDT consists of a 75- or 30-g glucose load containing 15 g of [6,6-2H2]glucose. For C-6–labeled glucose, >90% of C–H hydrogen atoms are lost to tissue water after glycolytic metabolism to pyruvate and oxaloacetate, but C–H hydrogen atoms are otherwise nonlabile and are retained in the glucose molecule (27). Complete glycolytic disposal of 15 g of [6,6-2H2]glucose results in the release of 0.0824 mol of 2H2O. Dispersion in the body water pool (∼2,200 mol/70-kg human) results in body water enrichment of about 0.0037% (250 δ units) in humans, well above the detection limit (∼1 δ) for 2H2O by isotope ratio-mass spectrometry (IR-MS). The 2H-GDT was designed to adhere to the following principles: 1) ambient glucose and insulin concentrations should reflect metabolic conditions present physiologically in daily life; 2) the test should primarily quantify insulin-mediated glucose utilization by tissues, rather than other aspects of insulin action, such as suppression of hepatic glucose production; and 3) the metabolic conditions should be comparable to those present in tests proven to be predictive for cardiovascular outcomes and type 2 diabetes risk (i.e., euglycemic-hyperinsulinemic clamp and SSPG test).

Some features of glucose homeostasis are worth noting. First, suppression of endogenous glucose production (EGP) is essentially complete during a clamp or SSPG test (28). Second, insulin concentrations influence the relative contributions from glycolytic and glycogenic routes of glucose disposal. At supraphysiologic concentrations (e.g., >1,000 pmol/l), glucose disposal is dominated by the pathway in the body with the greatest glucose utilization capacity (muscle glycogen storage), whereas glycolysis reaches a maximum and becomes less informative. At low insulin concentrations (e.g., <120–180 pmol/l), hepatic glucose production is not fully suppressed (29), and insulin-independent glucose utilization (e.g., by the brain) makes a proportionately greater contribution to glucose utilization. Accordingly, our criteria for a glycolysis-based test of insulin resistance are best fulfilled at serum insulin concentrations in the dynamic range, as achieved by a 75-g oral glucose load (15).

Operationally, any measure of tissue insulin sensitivity must either control insulin concentrations (e.g., the clamp or SSPG test) or correct for them (e.g., HOMA or frequently sampled intravenous glucose tolerance test). Glucose utilization rates cannot be compared directly as a metric of tissue insulin sensitivity if insulin levels are different. Accordingly, the observed glycolytic rate was corrected here for ambient insulin concentrations to calculate insulin sensitivity of tissues. If glucose concentrations are variable or out of the normal range, correction for glucose concentrations can also be performed to account for glucose effectiveness (30).

All subjects gave written, informed consent. Protocols were approved by the appropriate institutional review boards. Two studies were performed.

Euglycemic-hyperinsulinemic glucose clamp

Ten healthy nonobese subjects and 10 subjects with the metabolic syndrome were recruited and studied at the Diabetes and Glandular Disease Center in San Antonio, Texas (a fee for service clinical trial site). Metabolic syndrome was defined by the presence of three of five modified Adult Treatment Panel III criteria (20): blood pressure ≥130/85 mmHg or taking antihypertensive medication, fasting triacylglycerol concentrations ≥1.7 mmol/l (150 mg/dl) or treated with gemfibrozil or fenofibrate, fasting glucose concentrations ≥5.5 mmol/l (100 mg/dl), fasting HDL cholesterol ≤1.0 mmol/l (40 mg/dl) for men or ≤1.3 mmol/l (50 mg/dl) for women, and waist circumference ≥102 cm for men or ≥88 cm for women. Subjects with known type 2 diabetes, subjects with fasting glucose ≥7.0 mmol/l (126 mg/dl), or subjects being treated with medications known to alter insulin sensitivity were excluded. Subjects underwent a 75-g 2H-GDT followed within 2 weeks by a clamp study. Subjects then returned 6–8 months later for a 30-g 2H-GDT.

SSPG studies

Overweight (n = 12) and obese subjects (n = 6) screened for a weight loss study at Stanford University underwent both an SSPG test and a 75-g 2H-GDT. Subjects with BMI >35 kg/m2, fasting plasma glucose ≥7.0 mmol/l, variable weight, or a history of major organ disease and subjects taking drugs known to alter insulin sensitivity were excluded.

2H-GDT protocol

After a 10- to 12-h overnight fast, subjects drank 75 or 30 g of glucose, of which 15 g was [6,6-2H2]glucose, dissolved in 300 ml of water. Plasma samples for glucose, insulin, and 2H2O content were obtained at baseline and hourly for up to 4 h.

Euglycemic-hyperinsulinemic glucose clamp protocol

The clamps were performed as described by DeFronzo et al. (7). Briefly, after a 10-h overnight fast subjects received a primed, continuous insulin infusion (40 mU/m2 per min) for 110 min. Blood glucose concentrations were clamped at 5 mmol/l with an exogenous glucose infusion (20% wt/vol).

SSPG test protocol

Insulin-mediated glucose disposal was quantified as described previously (11). Briefly, subjects were admitted to the Stanford General Clinical Research Center after a 12-h overnight fast for 180-min infusions of octreotide (0.27μg/m2 per min), insulin (25 mU/m2 per min), and glucose (240 mg/m2 per min).

Total body water determination

Total body water (TBW) in the clamp study was measured by bioelectrical impedance analysis (Tanita model TBF 300A). As a comparison, TBW was also calculated from weight and height using the Hume formula (31). TBW for the SSPG study was calculated using the Hume formula.

Analytic determinations

For 2H2O content, 100-μl aliquots of plasma in the cap of an inverted vial were placed in a 70°C glass bead–filled heating block overnight. Water distillate inside the vial was then collected. Samples were run in triplicate.

Plasma glucose was measured by the glucose-oxidase method. Plasma insulin concentrations were measured by radioimmunoassay (Linco, St. Charles, MO).

IR-MS analyses

The deuterium content of plasma samples was determined using a ThermoFinnigan High Temperature Conversion/Elemental Analyzer coupled with a ThermoFinnigan MAT 253 isotope ratio-mass spectrometer via a Conflo-III Interface. The deuterium isotope abundance is first calculated in δ 2H values relative to the international Vienna Standard Mean Ocean Water standard and then transformed to atom percent excess by using a calibration curve of standards.

Calculations

2H2O enrichment from IR-MS was converted to millimoles by multiplying enrichment by the TBW pool size and dividing by 20 (molecular weight 2H2O). Plasma insulin and glucose areas under the curves (AUCs) were calculated using the trapezoidal method.

The 2H-GDT parameter of insulin sensitivity was calculated as 2H2O production (millimoles) per unit of insulin exposure (insulin AUC). M values (milligrams per kilogram per minute), SSPG results, and steady-state plasma insulin concentrations were determined during the last 30 min of the clamp. HOMA (12), QUICKI (14), and insulin sensitivity indices (Matsuda) (15) were calculated as referenced. Plasma glucose was determined during the last 30 min of the SSPG test; the higher the SSPG concentration, the more insulin resistant an individual is.

Statistical analysis

Differences between independent groups were determined by the Mann-Whitney U test. Correlations were calculated with the Pearson's correlation test. P ≤ 0.05 was considered statistically significant.

RESULTS

Euglycemic-hyperinsulinemic glucose clamp study

Healthy lean subjects and subjects with the metabolic syndrome were recruited to compare the 75-g 2H-GDT with the clamp across a range of insulin sensitivities. Technical problems occurred with the clamp in three subjects, preventing completion of the study, leaving 17 subjects. Fourteen subjects also had a follow-up 30-g 2H-GDT to determine whether a lower glucose load allowed a shorter 2H-GDT. Clinical characteristics of subjects are shown in Table 1.

During the clamp, SSPG concentrations were 5.1 ± 0.3 mmol/l for control subjects vs. 4.9 ± 0.3 mmol/l for subjects with metabolic syndrome and steady-state plasma insulin concentrations were 382 ± 54 vs. 502 ± 116 pmol/l, respectively (P < 0.05). Peak plasma insulin concentrations after the 75-g oral glucose challenge were 309 ± 123 pmol/l for control subjects and 837 ± 352 pmol/l for subjects with metabolic syndrome. After the 30-g oral glucose challenge, peak insulin concentrations were 155 ± 73 pmol/l for control subjects and 371 ± 191 pmol/l for subjects with metabolic syndrome. No significant difference between groups was seen in glucose AUC for the 75-g 2H-GDT, although a significant difference between groups was seen after a 30-g glucose load (2 h, P < 0.01).

The M value from the clamp was significantly lower in subjects with metabolic syndrome (Fig. 1A). The 75-g (Fig. 1B) and the 30-g (data not shown) 2H-GDT results revealed a proportional decrease in insulin sensitivity almost identical to that with the clamp in subjects with metabolic syndrome. The 75-g 2H-GDT correlated extremely closely with the M value across all subjects (r = 0.93 at 3 h and r = 0.95 at 4 h) (Figs. 1C and D). Use of the Hume formula instead of bioelectrical impedance analysis to determine TBW did not change correlations. A significant correlation with the M value was seen for the 30-g 2H-GDT as early as 1–2 h after the glucose load (r = 0.88) (Figs. 1E and Table 2), better than the 2-h correlation seen with the 75-g test (r = 0.75). Adjusting the 75-g 2H-GDT values for differences in glucose concentrations (glucose AUC) did not improve correlations, although a slight improvement was seen for 30-g 2H-GDT values (r = 0.92).

Correlations between the clamp, 2H-GDT, and other indexes of insulin resistance are shown in Table 2. When data from all subjects were included, the best correlation was with the 75-g 2H-GDT, whereas only modest correlations were seen for fasting indexes of insulin resistance and Matsuda insulin sensitivity indices. Considering subjects in the metabolic syndrome group alone, only the 75-g 2H-GDT correlated significantly with the M value. Correlation coefficients improved slightly when glucose AUC was included. For subjects in the control group alone, both the 75- and the 30-g 2H-GDT correlated significantly with the clamp. Insulin AUC had a weaker but significant correlation at 4 h after the 75-g glucose load. When differences in glucose AUC were taken into account in the control group, the correlation coefficients improved for the 30-g but not for the 75-g 2H-GDT.

SSPG study

Clinical characteristics of subjects who participated in the SSPG study are shown in Table 1. SSPG values in these obese and overweight subjects ranged from 2.7 to 14.9 mmol/l, indicating a broad range of insulin sensitivities. Previous studies have shown that SSPG values >10 mmol/l represent the top tertile denoting insulin resistance (32). Correlation between the SSPG value and the 2H-GDT (Fig. 1F) was excellent (r = −0.874).

CONCLUSIONS—

We demonstrate here that, in humans, insulin-mediated whole-body glycolysis correlates closely with insulin sensitivity in states of normal and reduced insulin sensitivity and may be useful as a potential metric of insulin resistance.

Insulin-mediated whole-body glycolytic disposal of a glucose load was measured by the 2H-GDT in insulin-resistant and insulin-sensitive individuals and correlated extremely well with M values during the clamp. Correlations between the clamp and 2H-GDT were stronger than correlations with indirect markers of insulin resistance and remained strong even in lean control subjects alone, for whom correlations with other markers have historically been poor (19). The SSPG test also correlated well with the 2H-GDT, whereas other markers of insulin resistance have traditionally correlated poorly (33).

Metabolic influences on glycolysis are worth considering. Factors that reduce glucose transport and/or phosphorylation should impair glycolysis and glycogen synthesis in parallel. In contrast, high-fat diet feeding to rats was reported (34) to reduce glycolytic disposal before reducing glycogen synthesis. It is possible that fatty acid oxidation products may inhibit glycolytic enzymes, such as phosphofructokinase, more than glucose phosphorylation or transport. We did not see dissociation between insulin-mediated glycolysis and insulin-mediated total glucose disposal in humans with insulin resistance, however, based on comparisons of the 2H-GDT to clamps and the SSPG test.

The 2H-GDT was designed to measure glycolysis, not to complete oxidation of a glucose load. Another stable isotope method, based on measurement of whole-body oxidation of [U-13C6] glucose to 13CO2, has recently been described for assessing insulin resistance. This approach gave a lower correlation with clamps (r = 0.69) (35) than the 2H-GDT. Several factors may account for the lower correlation. A 13CO2 collection represents oxidation at a single time point, whereas measurement of 2H2O production reveals integrated glycolytic flux over the preceding 3- to 4-h period. In addition, extensive exchange of the 13C label from 13CO2 into cellular metabolites reduces recovery of 13CO2 in breath in an unpredictable manner (36), whereas 2H2O distributes predictably into body water. Dietary and endogenous fatty acids may also affect pyruvate dehydrogenase activity, independent of insulin sensitivity (37), so that complete oxidation of glucose may be dissociated from glycolysis or glycogen synthesis. Finally, the low total glucose load given (15 g) for the breath 13CO2 test (35) interrogates a different physiological state and insulinemic level than is present in clamps (e.g., EGP or insulin-independent glucose utilization may confound results).

Hyperglycemia per se can stimulate uptake and glycolytic disposal of glucose (“glucose effectiveness”) (30), potentially increasing 2H2O production. Accordingly, we corrected for the glucose AUC, although this correction had, at most, a minor impact on calculated insulin sensitivity in the euglycemic subjects studied here. Conversely, hyperglycemia can also reduce 2H2O production. A glucose load mixes into the whole-body pool of free glucose of 15–20 g, so that the throughput of glucose after a 75-g glucose challenge is much higher than the pool size present before the load. If the fasting blood glucose concentration is 11 mmol/l, the pool size increases to 30–40 g; at >17 mmol/l, the fasting pool size of glucose becomes quantitatively significant in comparison with the glucose load. Thus, the presence of fasting hyperglycemia may dilute exogenous-labeled glucose and reduce recovery of 2H2O from an exogenous load, but this effect is modest unless hyperglycemia is severe.

The effects of EGP on the 2H-GDT are also worth considering. One of the advantages of the 2H-GDT (compared with glucose concentrations, for example) is that it primarily reflects glucose utilization, not hepatic insulin resistance. EGP might dilute exogenous [2H]glucose, however, and reduce recovery of 2H2O. An oral glucose load normally suppresses fasting EGP by ∼60%, from ∼2 to ∼0.8 mg · kg−1 · min−1 during the next 3–4 h (38). Endogenously produced glucose (roughly 14 g over 4 h) mixes with exogenous glucose (75 g), for a total flux rate of ∼90 g of glucose through the bloodstream. This, in turn, mixes with the ∼20 g of free glucose in the body before the glucose load. Total “exposure” to glucose over the 3–4 h after a 75-g glucose load is therefore normally about 110 g, with EGP providing about 11–14 g, so that dilution of labeled glucose by EGP normally affects 2H2O recovery only modestly. If a euglycemic, insulin-resistant subject starts with a higher EGP (2.5–3.0 mg · kg−1 · min−1, for example) and EGP is only suppressed by 40% (to 1.5–1.8 mg · kg−1 · min−1), EGP will be about twice normal (∼25–30 g over 4 h). Total flux of glucose will increase to ∼100–105 g/4 h, and the total glucose exposure will increase to 120–125 g. Thus, if EGP after a glucose load is twice normal, the net effect is to dilute exogenous label by about 15–20% and potentially reduce 2H2O production proportionately. This result is less than the differences observed between lean control subjects and subjects with metabolic syndrome (Fig. 1B). In diabetic patients, EGP may be even higher and less suppressible. Although we did not measure EGP here, it could be measured after a 2H-GDT based on the die-away curve of plasma [2H]glucose. Given these factors, the 2H-GDT is most easily interpreted in the presence of normal or near-normal glucose concentrations, which is the setting in which metrics of insulin resistance are most lacking. Independent validation of the 2H-GDT in diabetes will be required.

On the basis of these findings, the 2H-GDT is of interest as an index of insulin resistance. The 2H-GDT correlates closely with the euglycemic-hyperinsulinemic glucose clamp and the SSPG test. In addition, this measurement approach has a fundamental advantage over measures of insulin resistance that are based on serum insulin concentrations, including derived parameters such as HOMA or QUICKI. The loss of β-cell function that develops in the progression to type 2 diabetes (39,40) represents a fundamental problem for use of insulin concentrations alone as markers of insulin resistance. An individual who exhibits progressively lower insulin levels over time might either have improving insulin sensitivity or be progressing to β-cell failure. Indeed, a low insulin sensitivity index is better at predicting cardiovascular disease (41) than high postchallenge insulin concentrations. In contrast, the 2H-GDT overcomes the problem of pancreatic response, because the ratio of 2H2O production to insulin AUC remains low even if insulin secretion is failing.

The 2H-GDT does have some limitations. IR-MS is not a routine technique in clinical laboratories and requires special expertise. Future analytic advances (e.g., laser-based spectroscopic instruments) may overcome this limitation. Other limitations include the 2–3 h required to complete the test, with blood samples being drawn during that time. Although the 2H-GDT is considerably easier than the clamp, it does not have the ease of a single blood sample.

In summary, the 2H-GDT, a measure of whole-body glycolysis in response to a physiological glucose load, has excellent correlation with the euglycemic-hyperinsulinemic glucose clamp and SSPG test in humans across a wide range of insulin sensitivities. Accordingly, impaired insulin-mediated whole-body glycolytic disposal of a glucose load is a feature of insulin resistance in humans and provides a quantitative metric of insulin sensitivity.

Figure 1—

M value from the euglycemic-hyperinsulinemic glucose clamp (A) and the 2H-GDT insulin sensitivity index (2H2O production/insulin [INS] AUC) (B) calculated from the 75-g 2H-GDT at 3 h in lean, control subjects (n = 8) and subjects with metabolic syndrome (n = 9). Correlation between the M value and the insulin sensitivity index for the 75-g 2H-GDT at 3 h (C) and 4 h (D) and for the 30-g 2H-GDT at 2 h (E). Correlation between the 75-g 2H-GDT at 3 h and the SSPG test (F) in overweight and obese subjects (n = 18). Data are means ± SD. ***P < 0.001 and **P < 0.01, significantly different from control subjects.

Table 1—

Clinical characteristics of subjects who participated in the euglycemic-hyperinsulinemic glucose clamp study and the SSPG study

Table 2—

Correlations between the euglycemic-hyperinsulinemic glucose clamp M value, the 2H-GDT, and surrogate measures of insulin sensitivity for subjects with metabolic syndrome, lean control subjects, and all subjects combined

Acknowledgments

We thank Mark Kipnes and the Diabetes and Glandular Disease Center in San Antonio, Texas, for performing the clamps and Alex Fong for his assistance.

Footnotes

  • Published ahead of print at http://care.diabetesjournals.org on 26 January 2007. DOI: 10.2337/dc06-1809.

    M.K.H. is on the Board of Directors for, serves as Chair of the Scientific Advisory Board of, and holds stock in KineMed.

    A table elsewhere in this issue shows conventional and Système International (SI) units and conversion factors for many substances.

    The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C Section 1734 solely to indicate this fact.

    • Accepted January 17, 2007.
    • Received August 29, 2006.

References

| Table of Contents

This Article

  1. Diabetes Care vol. 30 no. 5 1143-1149
  1. All Versions of this Article:
    1. dc06-1809v1
    2. 30/5/1143 most recent