- Email: [email protected]
Quantification of Liver Glucose Metabolism by Positron Emission Tomography: Validation Study in Pigs
GASTROENTEROLOGY 2007;132:531–542
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
PATRICIA IOZZO,*,‡ MIKKO J. JARVISALO,* JAN KISS,* RONALD BORRA,* GRATIAN A. NAUM,* ANTTI VILJANEN,* TAPIO VILJANEN,* AMALIA GASTALDELLI,‡ EMMA BUZZIGOLI,‡ LETIZIA GUIDUCCI,‡,§ ELISABETTA BARSOTTI,㛳 TIMO SAVUNEN,¶ JUHANI KNUUTI,* MERJA HAAPARANTA–SOLIN,* ELE FERRANNINI,‡,㛳 and PIRJO NUUTILA*,# *Turku PET Centre and Departments of ¶Surgery and #Medicine, University of Turku, Turku, Finland; and ‡PET Centre, Institute of Clinical Physiology, CNR National Research Council, §SSSUP-Medical Sciences Branch, and 㛳Department of Internal Medicine, University of Pisa School of Medicine, Pisa, Italy
See editorial on page 794. Background & Aims: The liver is inaccessible to organ balance measurements in humans. To validate [18F]fluorodeoxyglucose ([18F]FDG) positron emission tomography (PET) in the quantification of hepatic glucose uptake (HGU), we determined [18F]FDG modeling parameters, lumped constant (LC), and input functions (single arterial versus dual). Methods: Anesthetized pigs were studied during fasting (n ⴝ 6), physiologic (n ⴝ 4), and supraphysiologic (n ⴝ 4) hyperinsulinemia. PET was performed with C15O (blood pool) and [18F]FDG (glucose uptake). 6,6-Deuterated glucose ([2H]G) was coinjected with [18F]FDG and blood collected from the carotid artery and portal and hepatic veins to compute LC as ratio between tracers fractional extraction. HGU was estimated from PET images and ex vivo from high-performance liquid chromatography measurements of liver [18F]FDG versus [18F]FDG-6-phosphate and [18F]glycogen. Endogenous glucose production was measured with [2H]G and hepatic blood flow by flowmeters. Results: HGU was increased in hyperinsulinemia versus fasting (P < .05). Fractional extraction of [18F]FDG and [2H]G was similar (not significant), intercorrelated (r ⴝ 0.98, P < .0001), and equally higher during hyperinsulinemia than fasting (P < .05), with an LC of 0.98 ⴞ 0.10 and 1.18 ⴞ 0.26, respectively. [18F]FDGPET modeling provided HGU values that did not differ from, and were correlated with, those from ex vivo measurements (r ⴝ 0.61, P < .02); proportional estimates of liver perfusion and endogenous glucose production were also obtained. Single and dual input functions produced strongly intercorrelated results (r > 0.95, P < .0001), with a modest underestimation of
HGU by the former. Conclusions: [18F]FDG-PET– derived parameters provide accurate quantification of HGU and estimates of liver perfusion and glucose production. In the liver, LC of [18F]FDG is nearly unitary. Using a single arterial input introduces only a small error in estimation of HGU.
A
bnormalities in hepatic glucose uptake (HGU) have been implicated in the pathogenesis of liver steatosis, hypertriglyceridemia, and diabetes.1–3 Because of its anatomic location and dual blood supply, the liver is inaccessible to direct organ balance measurements in humans in vivo. Positron emission tomography (PET) in combination with [18F]fluorodeoxyglucose ([18F]FDG) has been previously evaluated by us and others for the assessment of HGU in humans4,5 and animals.6 These studies suggest that the methodology may be valid provided that its unknowns and assumptions are better understood and validated. The conversion of [18F]FDG to glucose uptake requires knowledge of the lumped constant (LC) term, which accounts for the potentially different affinity of the 2 molecules for cellular transport and phosphorylation processes. In extrahepatic organs, this term has been shown to range between 1.2 (in skeletal muscle7,8) and 0.52 (in the brain9). In some organs, it has been suggested to vary between fasting and hyperinsulinemic conditions.10 Although some previous in vitro evidence,11 toAbbreviations used in this paper: EGP, endogenous glucose production; [18F]FDG, [18F]fluorodeoxyglucose; [18F]FDG-6P, [18F]fluorodeoxyglucose-6-phosphate; FE, fractional extraction; [2H]G, deuterated glucose; HGU, hepatic glucose uptake; HKi, hepatic glucose influx rate constant; HPLC, high-performance liquid chromatography; LC, lumped constant; PET, positron emission tomography. © 2007 by the AGA Institute 0016-5085/07/$32.00 doi:10.1053/j.gastro.2006.12.040
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
Quantification of Liver Glucose Metabolism by Positron Emission Tomography: Validation Study in Pigs
532
IOZZO ET AL
GASTROENTEROLOGY Vol. 132, No. 2
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
Figure 1. Study design. Continuous lines (top) indicate plasma sampling for tracer concentration measurements after tracer administration (small arrows); boxes represent PET scanning periods (light gray), liver tissue sampling and organ explantation (dark gray), and study conditions/groups (bottom, white boxes); and thick arrows indicate time points for blood collection (black) and liver Doppler blood flow measurements (gray).
gether with the in vivo comparison of liver inward transport of 11C-methyl glucose and [18F]FDG in pigs,6 support the use of a unitary LC, this term has not been determined experimentally. The mathematical approaches used to quantify liver [18F]FDG uptake and phosphorylation include graphical analysis4,6 and 3-compartmental modeling with 4 rate constants (4k),4 – 6 the latter accounting for the potential dephosphorylation of [18F]FDG-6-phosphate ([18F]FDG6P) occurring in hepatocytes. Both provide an estimation of a composite parameter (ie, the fractional extraction of the tracer) intended as a unidirectional influx rate constant, which can be used to compute HGU once the LC is known. The first rate constant in compartmental modeling (K1), reflecting hepatic tracer delivery, is assumed to be proportional to organ perfusion,4 – 6 but validation of this concept has not been conclusively provided. The rationale of a 4k as opposed to a 3k or irreversible uptake model derives from the notion that the liver possesses the glucose-6-phosphatase enzyme, implying that the last reverse rate constant in the model (k4) may reflect the activity of this enzyme. However, hepatic glucose uptake and output are compartmentalized processes within the liver, and the appropriateness of graphical analysis in HGU computations suggests irreversible uptake, especially during insulin stimulation. To date, there are no data to document a relationship between the rate of hepatic or endogenous glucose production (EGP) and k4. Once the LC is known and the appropriateness of the model is validated, quantification of regional processes by PET imaging still requires knowledge of the input function, which represents the concentration of tracer available for organ extraction over the measurement time. Given the dual perfusion of the liver by the portal vein and hepatic artery, the error introduced by using a
single arterial input function should be considered when utilizing the methodology in humans, in whom only arterial blood can be accessed. Previous data in pigs suggest that the substitution may predominantly affect the estimation of initial steps in tracer uptake and liver vascular volume.6 The aim of this study was to validate the use of [18F]FDG-PET in the quantification of HGU in the fasting and insulin-stimulated state. To this purpose, pigs underwent arterial, portal, and hepatic vein catheterization in combination with the simultaneous administration and sampling of [18F]FDG and 6,6-deuterated glucose ([2H]G) across the organ, PET imaging of the liver blood pool and [18F]FDG kinetics, and ex vivo measurements of [18F]FDG, [18F]FDG-6P, and [18F]-labeled glycogen.
Materials and Methods Study Design The study design is schematized in Figure 1. After animal preparation, PET imaging was performed to measure liver blood content and glucose uptake during fasting (n ⫽ 6), physiologic (n ⫽ 4), and supraphysiologic (n ⫽ 4) euglycemic hyperinsulinemia. [2H]G and [18F]FDG were coinjected, and their concentrations were frequently measured in the carotid artery, portal vein, and hepatic vein. HGU was calculated from PET data by using graphical analysis and 3k versus 4k compartmental modeling. EGP was measured by stable isotope approaches ([2H]G). Immediately after the animals were killed, ex vivo [18F]FDG, [18F]FDG-6P, and [18F]-glycogen were assessed in liver biopsy specimens; the liver was then explanted to measure organ density. The protocol was
reviewed and approved by the Ethical Committee for Animal Experiments of the University of Turku.
Animal Preparation Fourteen anesthetized, weight-matched pigs were studied during fasting (weight, 29.8 ⫾ 0.6 kg), physiologic (1.0 mU · kg⫺1 · min⫺1; weight, 30.0 ⫾ 0.5 kg), or supraphysiologic euglycemic hyperinsulinemia (5.0 mU · kg⫺1 · min⫺1; weight, 30.3 ⫾ 0.5 kg). Animals were deprived of food on the day before the study at 5 PM. Anesthesia was induced by injection of 1.0 g ketamine into the neck muscles before transportation of the pigs to the operating room. Throughout the experiment, animals were kept anesthetized with ketamine and pancuronium (total of 1.5 g and 40 mg, respectively) and mechanically ventilated via tracheal intubation with oxygen and normal room air (regulated ventilation, 16 breaths/min). Catheters were placed in the femoral vein and carotid artery for the administration of glucose, insulin, [2H]G, and [18F]FDG and for sampling of arterial blood, respectively. Splanchnic vessels were accessed by subcostal incision; after dissection of the hepatogastric ligament, purse-string sutures were allocated to allow catheter insertion via a small incision in the hepatic and portal veins. Doppler flow probes were placed around the portal vein and hepatic artery; blood flow was monitored continuously and recorded 8 times (Figure 1). The surgical access was closed, and the distal catheter extremities were secured to the abdominal surface to avoid tip displacement. The animals were then transported to the PET Centre for tracer administration, liver imaging, and blood sampling.
PET Scanning Scans were performed using an ECAT 931-08/12 scanner (CTI Inc, Knoxville, TN) with a 10.5-cm axial field of view and a resolution of 6.7 mm (axial) ⫻ 6.5 mm (in-plane) full width at half maximum. In the hyperinsulinemic clamp studies, insulin was infused in a primecontinuous fashion for 240 minutes via the venous cannula and euglycemia was maintained by measuring arterial plasma glucose levels at 5- to 10-minute intervals and adjusting a variable rate of a 10% dextrose intravenous infusion accordingly.12 In the fasting studies, saline was infused instead of insulin and glucose. After acquisition of a transmission scan to correct for photon attenuation, 15O-labeled carbon monoxide (C15O) was administered by inhalation and a static scan was performed to image the liver blood pool; 3 blood samples were obtained at equilibrium during the imaging period to measure whole blood radioactivity concentrations. Approximately 60 minutes into the insulin or saline infusions, [18F]FDG13 (274 ⫾ 7 MBq) and [2H]G (fast, 451 ⫾ 9; clamp, 1043 ⫾ 26 mol) were rapidly coinjected, an 180-minute dynamic [18F]FDG PET scan was started (31 frames, 8 ⫻ 15, 2 ⫻3 0, 2 ⫻ 120, 1 ⫻ 180, 6 ⫻ 300, 8 ⫻
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
533
600, 4 ⫻ 900 seconds), and arterial, portal venous, and hepatic venous blood was frequently sampled (ie, once every imaging time frame) for the determination of plasma [18F]FDG radioactivity and [2H]G tracer-to-tracee ratio (by gas chromatography/mass spectrometry14). Tracer-to-tracee ratio was multiplied by steady-state plasma glucose levels to obtain absolute [2H]G concentrations. Vital signs, blood pressure, and heart rate were monitored throughout the study. Additional blood samples were obtained for the assessment of plasma glucose, lactate, fatty acid, and insulin levels, which were assayed as previously described.2,15
Ex Vivo Sample Processing At the end of the experimental period, animals were killed by potassium chloride injection and anesthetic overdose, the abdominal cavity was rapidly accessed, and liver specimens were obtained from 4 animals per group, weighed, measured for whole [18F] activity, and prepared for high-performance liquid chromatography (HPLC) analysis and glycogen precipitation. Then, the whole organ was explanted and weighed and its volume was measured by water displacement; liver density was calculated as the ratio of organ weight to volume. Liver samples were homogenized in 0.9 mol/L PCA, and the supernatants were neutralized with 5 mol/L KOH and then centrifuged and analyzed by HPLC as described earlier.16 The relative amounts of [18F]FDG and [18F]FGD-6P were calculated from the integrated HPLC peak areas and multiplied by the total radioactivity in 1 mL of tissue sample to obtain respective activity concentrations per unit volume. Liver glycogen was separated as previously described17 and counted for [18F] radioactivity.
Image Processing All sinograms were corrected for tissue attenuation, dead time, and decay and reconstructed through standard reconstruction algorithms in a 256 ⫻ 256 matrix. Final in-plane resolution of reconstructed and Hann-filtered images was ⬃10 mm at 10 cm from the center of the gantry. Regions of interest were drawn in the right lobe of the liver over 3– 6 transaxial planes; regions of interest were originally delineated on the [18F]FDG image to obtain hepatic [18F]FDG time-activity curves. The same regions of interest were copied in respective C15O images and visually inspected to rule out animal movement between the 2 imaging periods. The liver blood pool (as a fraction) was computed by dividing C15O radioactivity concentrations in 1 mL of tissue by C15O radioactivity concentrations in 1 mL of blood. Plasma and liver [18F]FDG time-activity curves were used to estimate HGU by using graphical analysis and compartmental modeling. In the former,18,19 a graph is generated by plotting
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
February 2007
534
IOZZO ET AL
Ct(t) ⁄ Cp(t) versus
GASTROENTEROLOGY Vol. 132, No. 2
兰 C ()d ⁄ C (t), t
0
p
p
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
where Ct and Cp are tissue and plasma radioactivity levels at each sampling time point (t). When tracer influx occurs, the 2 variables describe a linear relationship after a few minutes of equilibration, provided that an irreversible compartment is present. The influx rate constant (HKi) is then given by the slope of the linear fit of the data after excluding the first few values. In compartmental modeling,4 – 6 3 compartments, with either 3 (3k model, K1, k2, k3) or 4 (4k model, K1, k2, k3, k4) rate constants, were used to describe the kinetics of [18F]FDG in the liver. The 3 compartments describe plasma [18F]FDG, tissue [18F]FDG, and tissue [18F]FDG6P, respectively. Model parameters are interrelated through the following equations: dCe(t) ⁄ dt ⫽ K1Cp(t) ⫺ (k2 ⫹ k3)Ce(t) ⫹ k4Cm(t) dCm(t) ⁄ dt ⫽ k3Ce(t) ⫺ k4Cm(t) Ct(t) ⫽ (1 ⫺ VB)(Ce(t) ⫹ Cm(t)) ⫹ VBCp(t). Ct is activity in tissue at each sampling time point (t) and Cp is activity in plasma, and Ce and Cm are the concentrations of nonmetabolized and phosphorylated [18F]FDG in tissue, respectively; VB is the vascular volume fraction. Because free exchange of solutes and macromolecules between blood and extravascular space occurs in the liver, K1 should reflect blood flow to the organ (mL · min⫺1 · mL⫺1 of tissue) and k2 (min⫺1) is the reverse rate constant. The rate constants k3 (min⫺1) and k4 (min⫺1) reflect [18F]FDG phosphorylation and dephosphorylation, respectively; in the 3k model, k4 was set to 0. Kinetic parameters were derived by nonlinear least-square fit of data. In this case, the influx constant (HKi) was calculated as follows: Ki (mL · min⫺1 · mL⫺1) ⫽ (k1k3) ⁄ (k2 ⫹ k3) Rate-constant values were multiplied by steady-state plasma glucose concentrations and divided by the LC, determined as described in the following text to derive HGU (mol · min⫺1 · mL⫺1).
Validation Adequacy of model. In graphical analysis, the ap-
propriateness of the quantitative procedure was evaluated through the regression coefficient of the fit of measured values. In this analysis, linearity of measurements confirms the existence of an irreversible compartment.18,19 Adequacy of compartmental model configuration was based on the weighted sum of squares between model-predicted and
measured data, the Akaike information criterion20 (AIC ⫽ N ln RSS ⫹ 2 P), and the Schwartz criterion21 (SC ⫽ N ln RSS ⫹ P ln N), where RSS equals residual sum of squares, N is the number of data points, and P is the number of parameters included in the model. In comparing models with a different number of parameters, the usefulness of these criteria is given by the notion that the best model is not necessarily the one producing the smallest weighted sum of squares, because adding more parameters may reduce the weighted sum of squares. The Akaike information criterion and Schwartz criterion provide a balance between fitting precision and number of parameters. An inverse relationship links the adequacy of the model to the numerical scores derived from the previously described formulas. LC. The LC evaluates the difference between [18F]FDG and glucose uptake by the organ of interest; it is expressed as a ratio and used to convert the former into the latter. The hepatic LC was determined as previously described for the assessment of the LC in skeletal muscle7 from bolus tracer administration. Fractional extractions (FE) of [18F]FDG and [2H]G were calculated as described7 over different time lengths (60, 120, and 180 minutes) throughout the study period according to the formula FE[18F]FDG (or FE[2H]G) ⫽ 兺 (CAP ⫺ CV) ⁄ 兺 CAP, where CAP indicates the concentration of tracer entering the organ, calculated as the product of the arterial and portal tracer concentrations multiplied by respective contribution to liver perfusion, and CV is the concentration of tracer exiting the system through the hepatic vein. Data were not weighted for the duration of respective frames to avoid magnifying the larger error associated with the lower [2H]G concentrations occurring at later points, corresponding to longer intervals. LC was defined as the ratio between FE[18F]FDG and FE[2H]G Ex vivo versus PET modeling data. To test the appropriateness of the PET-derived data in estimating the rate of liver trapping of phosphorylated [18F]FDG, values of HKi and HGU obtained from graphical analysis and 3k modeling were compared with those calculated from ex vivo measurements. Because the latter are performed in blood-free tissue, for this comparison modeling of PET data was repeated after correcting liver timeactivity curves by their blood content by using the conventional formula Time-Activity CurveT ⫽ (Ct ⫺ Cp ⫻ Liver Blood Pool) ⁄ Tissue Fraction, in which Cp and Ct are plasma and whole tissue regions of interest radioactivity levels, the liver blood pool was obtained from C15O images, as described previously, and the tissue fraction (T) was computed as (1 ⫺ liver blood pool). Ex vivo
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
February 2007
535
Table 1. Biochemical Parameters During [18F]FDG-PET Imaging
Artery Glucose (mmol · L⫺1) Lactate (mmol · L⫺1) Insulin (mU · L⫺1) Free fatty acid (mmol · L⫺1) Portal vein Glucose (mmol · L⫺1) Lactate (mmol · L⫺1) Insulin (mU · L⫺1) Free fatty acid (mmol · L⫺1)
Fasting
Physiologic
Supraphysiologic
3.7 ⫾ 4 2.3 ⫾ 0.7 3⫾1 0.51 ⫾ 0.09
5.0 ⫾ 0.1a 4.0 ⫾ 0.5b 48 ⫾ 3b 0.07 ⫾ 0.01d
5.0 ⫾ 0.1a 2.0 ⫾ 0.4 234 ⫾ 35c 0.06 ⫾ 0.01d
3.9 ⫾ 0.5e 2.3 ⫾ 0.7 5 ⫾ 2f 0.51 ⫾ 0.09
5.0 ⫾ 0.0a 4.1 ⫾ 0.5b 32 ⫾ 2b,e 0.07 ⫾ 0.01d
4.9 ⫾ 0.1a 2.2 ⫾ 0.3e 205 ⫾ 25c,f 0.06 ⫾ 0.01d
ⱕ .05, ⱕ .1, and cP ⬍ .0001 versus fasting and physiologic hyperinsulinemia. dP ⬍ .001 versus fasting. eP ⱕ .03 and fP ⱕ .1 versus arterial values. aP
bP
derived HKiT and HGUT were obtained according to previously published methods,22,23 in which HKiT is given by the ratio [18F]FDG-6P Radioactivity ⁄
兰
0⫺180
Cp
and HGUT is calculated as the product of plasma glucose and HKi divided by LC. In the previously listed equations, Cp represents the dual liver input function. Liver plasma flow versus K1. Liver blood flow values, as determined by Doppler flowmeters, were multiplied by (1 ⫺ hematocrit) to obtain corresponding plasma flow figures. They were compared with K1 values obtained from 3k and 4k compartmental modeling. EGP versus k4. We also determined whether k4 is related to glucose release from the liver, such that changes in the k3/k4 ratio indicate the balance between phosphorylation and dephosphorylation. First, a comparison was made between these parameters in the different metabolic states of the study, based on the fact that hyperinsulinemia suppresses EGP and should therefore result in lower k4 values and higher k3/k4 ratios than during fasting. Second, an association was sought between k4 and EGP, as quantified by using arterial [2H]G measurements. Values of EGP in these animals have been previously reported.24 Dual versus single input function. Both linear and nonlinear model fitting procedures were repeated by using either a dual or a single input function. The dual input concentrations were calculated as the flowweighted input concentrations by using measured arterial and portal flow values: (Cartery ⫻ Hepatic Artery Flow ⫹ Cporta ⫻ Portal Vein Flow) ⁄ Whole Liver Flow,
where Cartery and Cporta refer to radioactivity (15O and [18F]) concentrations in the artery and portal vein, respectively. All parameters in each model were compared to evaluate the error introduced by using a single input function.
Statistical Analysis All data are presented as mean ⫾ SEM. Differences in paired data were evaluated using Student paired t test. One-way analysis of variance was used for unpaired group comparisons. Regression analyses were performed according to standard techniques. P ⬍ .05 was considered statistically significant.
Results The metabolic data of the study animals are given in Table 1. Plasma glucose levels were similar between the 2 hyperinsulinemic periods and slightly higher than during fasting conditions. Insulin levels were progressively higher in the fasting, physiologic, and supraphysiologic hyperinsulinemia experiments. During the imaging period, lactate levels did not differ significantly between the fasting and insulin-stimulated groups; however, as compared with respective baseline values, they were significantly suppressed during hyperinsulinemia (P ⱕ .002). Fatty acid levels were ⬎85% suppressed in the insulin infusion studies as compared with fasting values (P ⬍ .0001). Liver volume and weight were 758 ⫾ 30 mL and 814 ⫾ 26 g, respectively; thus, organ density was 1.08 ⫾ 0.03 g · mL⫺1. Liver blood flow, as determined by Doppler probes, was 990 ⫾ 124 and 852 ⫾ 120 mL · min⫺1 in the fasting and hyperinsulinemic states, respectively, and remained constant throughout repeated measurements during [18F]FDG scanning (coefficient of variation, 6.4% ⫾ 0.8%). The contributions of the portal and hepatic arterial vessels to total liver flow were 83% ⫾ 3% and 17% ⫾ 3%,
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
Hyperinsulinemia
536
IOZZO ET AL
GASTROENTEROLOGY Vol. 132, No. 2
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
Figure 2. Representative time course of plasma concentrations of [18F]FDG (left) and [2H]G (right) in the artery (white bars), portal vein (gray bars), and hepatic vein (black bars) during the experiments; the expanded representation of early time points in the histogram allows better visualization of rapidly changing tracer concentrations.
respectively, in all 3 metabolic conditions (not significant). Liver blood pool, as determined by C15O PET imaging, was 30% ⫾ 2% in the fasting experiment, 37% ⫾ 1% during physiologic hyperinsulinemia (P ⫽ .02 vs fasting), and 35% ⫾ 2% at supraphysiologic hyperinsulinemia (P ⫽ .07 vs fasting).
LC One representative set of arterial, portal, and hepatic venous concentrations of [18F]FDG and [2H]G in each experimental condition is given in Figure 2. Reliable tracer-to-tracee ratio measurements could not be ob-
tained in one case, which was excluded from paired analysis. Between the 2 hyperisulinemic states, FE values were overlapping (P ⱖ .60) and were therefore pooled for statistical analysis. No significant difference was found between FE of [18F]FDG and [2H]G (Table 2), showing a similar trend between the 2 tracers across different metabolic states and time intervals. FE measurements were not affected by the time over which the arterial-venous (A-V) differences were measured. Individual FE values of [18F]FDG and [2H]G were strongly correlated (r ⫽ 0.98, P ⬍ .0001), with an intercept of 0.01 and a slope of 0.96. The FE of both tracers was similarly higher during hy-
Table 2. FE of [18F]FDG and [2H]G and Calculated Liver LC Across Progressive Time Intervals [18F]FDG (FE) Fasting (n ⫽ 6) 60 min 120 min 180 min Average Hyperinsulinemia (n ⫽ 7)a 60 min 120 min 180 min Average
⫺0.51 ⫾ 0.37 ⫺0.49 ⫾ 0.35 ⫺0.48 ⫾ 0.34 ⫺0.49 ⫾ 0.35 0.22 ⫾ 0.02b 0.21 ⫾ 0.02b 0.21 ⫾ 0.02b 0.21 ⫾ 0.02b
[2H]G (FE) ⫺0.56 ⫾ 0.35 ⫺0.56 ⫾ 0.33 ⫺0.55 ⫾ 0.32 ⫺0.56 ⫾ 0.33 0.23 ⫾ 0.02b 0.23 ⫾ 0.02b 0.23 ⫾ 0.02b 0.23 ⫾ 0.02b
P value
LC ([18F]FDG/[2H]G)
.40 .39 .42 NS
1.22 ⫾ 0.27 1.19 ⫾ 0.27 1.14 ⫾ 0.25 1.18 ⫾ 0.26
.74 .53 .29 NS
1.03 ⫾ 0.14c 0.99 ⫾ 0.05c 0.93 ⫾ 0.11c 0.98 ⫾ 0.10c
aData from hyperinsulinemic experiments are pooled (P ⱖ .60 between 1.0 and 5.0 mU · kg⫺1 · min⫺1 insulin infusion experiments). Average values in each group are also shown, NS between time intervals. bP ⱕ .05, cP ⫽ .50 (NS) versus fasting.
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
537
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
February 2007
Figure 3. Relationship between parameters estimated with 3k and 4k compartmental modeling.
perinsulinemic than fasting conditions (P ⱕ .05). The calculated LC was 1.18 ⫾ 0.26 and 0.98 ⫾ 0.10 during fasting and hyperinsulinemia, respectively (P ⫽ .50 between the 2 conditions). These results were used in the computation of HGU.
ance between these factors resulted in higher fasting but similar insulin-mediated HGU values in the 4k versus 3k model (Figure 4); consequently, insulin-mediated HGU
Adequacy of Model and HGU Data obtained by using a dual input function as the reference method were examined. By graphical analysis, the linear fit approximated measured data very closely, with r values of 0.996 ⫾ 0.001 during fasting, 0.999 ⫾ 0.000 at physiologic hyperinsulinemia, and 0.999 ⫾ 0.001 at supraphysiologic hyperinsulinemia. In compartmental modeling, either the 3k or the 4k nonlinear fit to the data was good, with similarly low weighted sum of squares values (average values, ⬍6.5; not significant in each metabolic state) and slightly lower 4k Akaike information criterion and Schwartz criterion indexes during fasting (P ⬍ .05 vs 3k model). Relationships between individual parameters derived by different approaches are shown in Figure 3. All of them showed a high degree of correlation, although K1, k2, and k3 were on average 5.3%, 8.5%, and 31% higher in 4k than 3k modeling (P ⱕ .002), especially during fasting. The bal-
Figure 4. Comparison between HGU, as determined by irreversible or reversible (4k) modeling. *P ⫽ .01 versus 3k model.
538
IOZZO ET AL
GASTROENTEROLOGY Vol. 132, No. 2
Table 3. Liver Fractional Analysis (HKiT) and Glucose Uptake (HGUT) as Determined by Ex Vivo Measurements Versus Graphical Analysis or 3k Compartmental Modeling
HKiT (mL · min⫺1 · mL⫺1) Fasting Physiologic hyperinsulinemia Supraphysiologic hyperinsulinemia HGUT (mol · min⫺1 · mL⫺1) Fasting Physiologic hyperinsulinemia Supraphysiologic hyperinsulinemia
Ex vivo measurements
Graphical analysis
P value versus ex vivo
3k modeling
P value versus ex vivo
0.0054 ⫾ 0.0009 0.0046 ⫾ 0.0008 0.0064 ⫾ 0.0016
0.0054 ⫾ 0.0005 0.0056 ⫾ 0.0026 0.0060 ⫾ 0.0003
.80 .37 .53
0.0060 ⫾ 0.0012 0.0058 ⫾ 0.0006 0.0062 ⫾ 0.0003
.40 .30 .83
0.017 ⫾ 0.003 0.023 ⫾ 0.002 0.032 ⫾ 0.004
0.019 ⫾ 0.003 0.028 ⫾ 0.003 0.030 ⫾ 0.002
.55 .37 .53
0.021 ⫾ 0.003 0.029 ⫾ 0.003 0.031 ⫾ 0.002
.29 .30 .81
NOTE. Statistical values between metabolic states within each approach are reported in the text. CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
by graphical analysis, 3k and 4k modeling was 69% (P ⱕ .02), 67% (P ⬍ .02), and 20% higher than fasting values, respectively, with no difference between the 2 hyperinsulinemic states.
Ex Vivo Versus PET Modeling Data Percentages of liver [18F]FDG and [18F]FDG-6P in all studies were 34% ⫾ 3% and 66% ⫾ 3%, respectively. Radioactivity in glycogen was close to background, and below precision sensitivity, accounting for ⬍5% of tissue [18F]. Ex vivo HGUT was significantly higher during supraphysiologic hyperinsulinemia than in the fasting state (P ⬍ .05), with intermediate values in the physiologic hyperinsulinemic situation. HKiT and HGUT values determined by ex vivo FDG-6P measurements in the liver did not differ from those derived from graphical analysis or 3k modeling (P ⫽ NS) (Table 3). Individual values showed a strong correlation between the ex vivo and other approaches (r ⫽ 0.61, P ⱕ .02).
Liver Plasma Flow Versus K1 Average plasma flow and K1 values were not different and showed a tendency to correlate. The correlation became significant when accounting for the arterial input delay from the sampling site to the liver. Such correction did not influence HGU comparisons. It improved the approximation of K1 to plasma flow values (K1 ⫽ 0.80 ⫾ 0.15 and 0.91 ⫾ 0.17 mL · min⫺1 · mL⫺1 in 3k and 4k modeling, respectively; Doppler plasma flow ⫽ 0.90 ⫾ 0.08 mL · min⫺1 · mL⫺1; P ⫽ NS), leading to a closer correlation (r ⫽ 0.57, P ⫽ .03).
because EGP is also determined from arterial measurements.
Dual Versus Single Input Function No significant difference was observed when calculating liver blood pool from single arterial or dual C15O blood concentrations (P ⫽ .83); individual values were tightly intercorrelated (r ⫽ 0.99, P ⬍ .0001, slope ⫽ 1.05). Both HKi and HGU derived with a single versus dual input function were strongly intercorrelated (r ⬎ 0.95, P ⬍ .0001), with slopes ranging from 0.88 to 0.96 (HKi) or 0.85 to 0.98 (HGU) and intercepts ranging from 0.00003 to 0.0004 (HKi) or – 0.001 to 0.001 (HGU) across the 3 modeling approaches. However, a nearly systematic underestimation was observed when using the arterial input function, with a ratio of HGU (arterial) to HGU (dual) of 0.94 ⫾ 0.01, 0.90 ⫾ 0.01, and 0.96 ⫾ 0.03 with graphical analysis (Figure 6), 3k and 4k modeling, respectively. On average, only 1%–2% of the difference was accounted for by the variations between arterial and portal vein plasma glucose levels; k3 and k4 estimates were similar between the 2 input choices (P ⫽ NS),
EGP Versus k4 By using a dual input function, k4 decreased progressively and k3/k4 increased from fasting to physiologic and supraphysiologic hyperinsulinemia (Figure 5). The association between EGP and k4 (r ⫽ ⫹0.39, P ⫽ .16) or k3/k4 (r ⫽ ⫺0.50, P ⫽ .067) fell short of statistical significance. The correlation was stronger when using a single input function to model liver PET data (with k4, r ⫽ 0.51, P ⫽ .065; with k3/k4, r ⫽ ⫺0.55, P ⫽ .043), likely
Figure 5. Dephosphorylation rate contant (k4) (left) and phosphorylation-to-dephosphorylation ratio (k3/k4) (right), showing a progressive response to incremental hyperinsulinemia. *P ⫽ .01 versus fasting and P ⫽ .10 versus physiologic hyperinsulinemia; **P ⫽ .02 versus supraphysiologic and P ⫽ .06 versus physiologic hyperinsulinemia.
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
539
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
February 2007
Figure 6. Representative comparison of graphically derived HKi and HGU values by using the arterial and the dual input, showing strong correlations between the 2 approaches (left panels) with a slight systematic underestimation by the former (right panels).
whereas K1 and k2 estimates were significantly lower with the single input function (P ⬍ .05). K1 values derived from the use of the arterial input function were still strongly correlated with liver plasma flow (3k model: r ⫽ 0.67, P ⫽ .009; 4k model: r ⫽ 0.68, P ⫽ .007) despite a ⬃50% underestimation with either model.
Discussion To our knowledge, no previous study has compared the quantitative parameters derived from modeling of liver [18F]FDG data with conventional ex vivo and glucose flux measurements during fasting and insulin dose-response stimulation. Our data show that 3k modeling and graphical analysis provide accurate estimates of the retention of [18F]FDG-6P in the liver. The use of a 4k model additionally incorporates the proportion of [18F]FDG cycling through the glucose-6-phosphatase step. These notions have so far been assumed on the basis of their biological plausibility; they were documented here for the first time. In fact, [18F]FDG and [18F]FDG-6P were directly measured in liver biopsy specimens for comparison and the rate constant of tracer accumulation was computed from ex vivo liver samples according to classic formulas.22,23 In agreement with the inherent assumption that [18F]FDG-6P is the only trapped radiochemical, [18F]-labeled glycogen did not contribute to the hepatic tracer retention to any measurable extent; in the current experiments, the formation of metabolites, if any, as
previously described,25 was below the detection limit. EGP was related to the dephosphorylation rate constant k4 (positive) and to the balance between phosophorylation and dephosphorylation (k3/k4) (negative). The fact that such correlations were not very strong may be partly explained by the notion that EGP does not exactly equal hepatic glucose output, because the former includes the contribution of glucose released from extrahepatic organs, which is not negligible under the postsurgical conditions of the present study.26,27 In line with the expected action of insulin, k4 was progressively lower and the k3/k4 ratio was higher during fasting versus physiologic versus supraphysiologic hyperinsulinemia. On the whole, this finding indicates that k4 is a fair indicator of the inhibition of hepatic glucose release. Notably, k4 was always ⬍⬍k3, in line with the current knowledge that hepatic glucose uptake and output are compartmentalized processes, in which the bulk of substrate for organ release derives from gluconeogenesis, glycogenolysis, or glucose cycling through glycolysis, with a minimal contribution of the glucose that has just entered hepatocytes. Thus, only in the fasting state did k4 translate into a measurable HGU difference between reversible and irreversible modeling. As the dephosphorylation rate was suppressed by insulin, the difference fell below detection, although the process remained measurable. Confirming previous evidence by other investigators6 and our current hypothesis, K1 was related to hepatic plasma flow. However, it
540
IOZZO ET AL
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
should be noted that K1 is influenced by organ flow and glucose transport. Because a more direct assessment of hepatic blood flow by using perfusion tracers in PET imaging has proven challenging and awaits further methodological developments,28 K1 may be a plausible surrogate indicator of this process. Our data confirm that insulin stimulates HGU by either augmenting the rate of substrate trapping or decreasing the rate of glucose dephosphorylation and efflux from the organ. This result, despite the limitations due to the anesthesia and postsurgical stress, confirms our previous evidence in humans.4 Although HGU responds to insulin in a dose-dependent manner in the low hyperinsulinemic range,29 our data document a lack of dose dependency once physiologic postprandial insulin concentrations are reached, because in the current studies an additional 4- to 6-fold elevation in insulin levels during euglycemia had no effect. There have been no previous studies on liver LC for [18F]FDG. Munk et al6 reported similar hepatic influx rate constants of 3-O-11C-methylglucose and [18F]FDG by modeling of PET data. Because 3-O-11C-methylglucose is not a substrate of hepatic glucokinase, as confirmed by the same investigators in their previous ex vivo work,25 the above comparison was limited to the effect of liver blood flow and glucose transport, which are not ratelimiting processes for HGU. In the current study, the combined use of [2H]G and [18F]FDG extends the comparison to the full sequence of events leading to the uptake of these tracers, including their phosphorylation. We obtained LC values of approximately 1.0, with a small deflection from 1.18 during fasting to 0.98 at hyperinsulinemia, which did not further decline in response to supraphysiologic insulin infusions. Similar to the vast majority of PET studies, our experimental design precluded the use of continuous tracer infusion, which is the more usual format of tracer administration for arterialvenous balance computations. However, it was important to describe the kinetics of [18F]FDG by dynamic PET studies, in which a rapid injection is the method of choice, because the fast clearance of tracer from plasma coupled with its progressive tissue trapping provides the highest target-to-background signal and image quality. Therefore, we used the method that was previously applied to determine skeletal muscle LC,7 which is based on classic approaches for nonconstant indicator concentrations.30 The current approach, consisting of frequent blood sampling and use of integrated hepatic inflow and outflow data over a prolonged study period, requires that metabolic conditions and organ blood flow are known to be stable, as documented by repeated measurements in the current study. LC was calculated by relating the FE of [18F]FDG to that of simultaneously administered [2H]G, assayed in the same blood samples, after allowing for [18F] decay. Under these conditions, some negative FE values observed during fasting do not indicate that there
GASTROENTEROLOGY Vol. 132, No. 2
is no uptake or that the organ is producing the tracer30 but are related to the dynamic nature of the experiments and the expected small differences between the large numbers typical of the fasting state. The acceleration of metabolism during insulin stimulation was documented by the significantly higher FEs, with identical figures and changes with the 2 tracers. Without providing exact glucose metabolic rates, the method adopted here is valid in comparing the fate of 2 metabolizable tracers. An alternative approach for the estimation of tissue LC is that of directly comparing the quantification of glucose uptake estimated from [18F]FDG PET imaging with the steadystate arterial-venous balance or microdialysis of unlabeled glucose. These methods may sound more straightforward, but they are not easily applicable to the liver, because the balance of glucose across this organ results from the sum of simultaneous substrate uptake and output. Although net HGU and EGP were available here, notably the latter includes the release of glucose from organs other than the liver. Because EGP surpasses HGU by several fold, as confirmed in our study, even a small error in the quantification of hepatic glucose release, such as is introduced by assuming its equivalence with EGP, prevents reliable comparison. In skeletal muscle, in which all methods have been used by different investigators,7,8 identical LC results were observed. Collectively, our data indicate that [18F]FDG LC in the liver is virtually constant through a broad range of insulin concentrations and that the uptake of [18F]FDG does not differ substantially from that of glucose. The comparison between modeling parameters estimated from the analysis of [18F]FDG PET data by using the arterial versus dual input function has been previously shown in pigs under fasting conditions.7 The investigators concluded that the use of arterial concentrations results in the underestimation of K1 and blood volume terms. In our hands, VB was dependent on the delay occurring between the arterial sampling site and the liver rather than on the input function (data not shown). On the contrary, the delay did not affect HGU and its correction did not normalize K1. We extended the experiments to physiologic and supraphysiologic hyperinsulinemia. The most relevant findings here were that the correlation between K1 and liver perfusion persisted with the use of an arterial input function, and HGU was minimally and systematically affected, without altering group comparisons. The application of this method in humans essentially depends on this result. The observed input-related difference of ⬃1–2 mol · min⫺1 in the estimation of whole liver HGU is considerably lower than its intersubject variability and likely of negligible clinical relevance. Hepatic pathophysiology was not evaluated in this study. Previous observations suggested the potential usefulness of this technique in understanding disease pathogenesis and identifying treatment targets, although the
interpretation of the results partly relied on plausible biological assumptions. The main aim of the study was to provide a thorough understanding of the parameters estimated with this technique to allow us to describe molecular mechanisms in a more specific and conclusive manner. The study provides some information on the physiology of liver glucose metabolism and insulin sensitivity, processes that may be impaired in most common liver disorders.31–36 The previously described aspects are prerequisites to the formulation and testing of a hypothesis concerning disease pathophysiology. In summary, irreversible models provide an accurate estimate of HGU, differing from overall unidirectional uptake only during fasting conditions, in which a small but significant loss of immediately phosphorylated glucose occurs. By increasing the complexity of the model, additional information on liver perfusion and glucose production can be obtained. The LC for the estimation of HGU with [18F]FDG was shown to be nearly unitary. The error introduced by using a single arterial input function in the computation of HGU seems to be clinically irrelevant, supporting the implementation of the methodology in humans. References 1. Basu A, Basu R, Shah P, Vella A, Johnson CM, Nair KS, Jensen MD, Schwenk WF, Rizza RA. Effects of type 2 diabetes on the ability of insulin and glucose to regulate splanchnic and muscle glucose metabolism: evidence for a defect in hepatic glucokinase activity. Diabetes 2000;49:272–283. 2. Iozzo P, Hallsten K, Oikonen V, Virtanen KA, Kemppainen J, Solin O, Ferrannini E, Knuuti J, Nuutila P. Insulin-mediated hepatic glucose uptake is impaired in type 2 diabetes: evidence for a relationship with glycemic control. J Clin Endocrinol Metab 2003; 88:2055–2060. 3. Ferre T, Riu E, Franckhauser S, Agudo J, Bosch F. Long-term overexpression of glucokinase in the liver of transgenic mice leads to insulin resistance. Diabetologia 2003;46:1662–1668. 4. Iozzo P, Geisler F, Oikonen V, Maki M, Takala T, Solin O, Ferrannini E, Knuuti J, Nuutila P. Insulin stimulates liver glucose uptake in humans: an 18F-FDG PET Study. J Nucl Med 2003;44:682– 689. 5. Choi Y, Hawkins RA, Huang SC, Brunken RC, Hoh CK, Messa C, Nitzsche EU, Phelps ME, Schelbert HR. Evaluation of the effect of glucose ingestion and kinetic model configurations of FDG in the normal liver. J Nucl Med 1994;35:818 – 823. 6. Munk OL, Bass L, Roelsgaard K, Bender D, Hansen SB, Keiding S. Liver kinetics of glucose analogs measured in pigs by PET: importance of dual-input blood sampling. J Nucl Med 2001;42: 795– 801. 7. Kelley DE, Williams KV, Price JC, Goodpaster B. Determination of the lumped constant for [18F] fluorodeoxyglucose in human skeletal muscle. J Nucl Med 1999;40:1798 –1804. 8. Peltoniemi P, Lonnroth P, Laine H, Oikonen V, Tolvanen T, Gronroos T, Strindberg L, Knuuti J, Nuutila P. Lumped constant for [(18)F]fluorodeoxyglucose in skeletal muscles of obese and nonobese humans. Am J Physiol Endocrinol Metab 2000;279:E1122– E1130. 9. Reivich M, Alavi A, Wolf A, Fowler J, Russell J, Arnett C, MacGregor RR, Shiue CY, Atkins H, Anand A, et al. Glucose metabolic rate kinetic model parameter determination in humans: the lumped constants and rate-constants for [18F]fluorodeoxyglu-
10.
11.
12.
13.
14. 15.
16.
17.
18.
19.
20. 21. 22.
23.
24.
25.
26.
27.
541
cose and [11C]deoxyglucose. J Cereb Blood Flow Metab 1985;5:179 –192. Botker HE, Bottcher M, Schmitz O, Gee A, Hansen SB, Cold GE, Nielsen TT, Gjedde A. Glucose uptake and lumped constant variability in normal human hearts determined with [18F]fluorodeoxyglucose. J Nucl Cardiol 1997;4:125–132. Sweet IR, Peterson L, Kroll K, Goodner CJ, Berry M, Graham MM. Effect of glucose on uptake of radiolabeled glucose, 2-DG, and 3-O-MG by the perfused rat liver. Am J Physiol 1996;271:E384 – E396. DeFronzo RA, Tobin JD, Andres R. Glucose clamp technique: a method for quantifying insulin secretion and resistance. Am J Physiol 1979;237:E214 –E223. Hamacher K, Coenen HH, Stöcklin G. Efficient stereospecific synthesis of no-carrier-added 2-[18F]-fluoro-2-deoxy-D-glucose using aminopolyether supported nucleophilic substitution. J Nucl Med 1986;27:235–238. Wolfe RR. Radioactive and stable isotopes in biomedicine. New York: Wiley-Liss, 1992. Muscelli E, Mari A, Natali A, Astiarraga BD, Camastra S, Frascerra SD, Holst JJ, Ferrannini E. Impact of incretin hormones on beta-cell function in subjects with normal or impaired glucose tolerance. Am J Physiol Endocrinol Metab 2006;291:E1144 – E1150. Suolinna EM, Haaparanta M, Paul R, Härkönen P, Solin O, Sipilä H. Metabolism of 2-[18F]fluoro-2-deoxyglucose in tumor-bearing rats: chromatographic and enzymatic studies. Nucl Med Biol 1986;13:577–581. Barrett EJ, Bevilacqua S, DeFronzo RA, Ferrannini E. Glycogen turnover during refeeding in the postabsorptive dog: implications for the estimation of glycogen formation using tracer methods. Metabolism 1994;43:285–292. Patlak CS, Blasberg RG, Fenstermacher JD. Graphical evaluation of blood-to-brain transfer constants from multiple-time uptake data. J Cereb Blood Flow Metab 1983;3:1–7. Patlak CS, Blasberg RG. Graphical evaluation of blood-to-brain transfer constants from multiple-time uptake data. Generalizations. J Cereb Blood Flow Metab 1985;5:584 –590. Akaike H. A new look at the statistical model identification. IEEE Trans Automat Contr 1974;AC-19:716 –723. Schwartz G. Estimating the dimension of a model. Ann Stat 1978;6:461–564. Kraegen E, James DE, Jenkins AB, Chisholm DJ. Dose-response curves for in vivo insulin sensitivity in individual tissues in rats. Am J Physiol Endocrinol Metab 1985;248:E353–E362. Haaparanta M, Paul R, Gronroos T, Bergman J, Kamarainen EL, Solin O. Microdialysis and 2-[18F]fluoro-2-deoxy-D-glucose (FDG): a study on insulin action on FDG transport, uptake and metabolism in rat muscle, liver and adipose tissue. Life Sci 2003;73: 1437–1451. Iozzo P, Gastaldelli A, Jarvisalo MJ, Kiss J, Borra R, Buzzigoli E, Viljanen A, Naum G, Viljanen T, Oikonen V, Knuuti J, Savunen T, Salvadori PA, Ferrannini E, Nuutila P. 18F-FDG assessment of glucose disposal and production rates during fasting and insulin stimulation: a validation study. J Nucl Med 2006;47:1016 –1022. Bender D, Munk OL, Feng HQ, Keiding S. Metabolites of (18)FFDG and 3-O-(11)C-methylglucose in pig liver. J Nucl Med 2001; 42:1673–1678. Stumvoll M, Meyer C, Mitrakou A, Nadkarni V, Gerich JE. Renal glucose production and utilization: new aspects in humans. Diabetologia 1997;40:749 –757. Joseph SE, Heaton N, Potter D, Pernet A, Umpleby MA, Amiel SA. Renal glucose production compensates for the liver during the anhepatic phase of liver transplantation. Diabetes 2000;49: 450 – 456.
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
February 2007
542
IOZZO ET AL
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
28. Ziegler SI, Haberkorn U, Byrne H, Tong C, Kaja S, Richolt JA, Byrne H, Tong C, Schosser R, Krieter H, Kaja S, Richolt JA, Lammertsma AA, Price P. Measurement of liver blood flow using oxygen-15 labelled water and dynamic positron emission tomography: limitations of model description. Eur J Nucl Med 1996;23: 169 –177. 29. Arora P, Basu R, Zangeneh F, Schweink W, Rizza R. The ability of insulin to stimulate the incorporation of extracellular glucose into hepatic glycogen is impaired in type 2 diabetes. Diabetes 2003; 52(Suppl 1):A59. 30. Zierler K. Theory of the use of arteriovenous concentration differences for measuring metabolism in steady and non-steady states. J Clin Invest 1961;40:2111–2125. 31. Hui JM, Sud A, Farrell GC, Bandara P, Byth K, Kench JG, McCaughan GW, George J. Insulin resistance is associated with chronic hepatitis C virus infection and fibrosis progression. Gastroenterology 2003;125:1695–1704. 32. Marchesini G, Babini M. Nonalcoholic fatty liver disease and the metabolic syndrome. Minerva Cardioangiol 2006;54:229 –239. 33. Yki-Jarvinen H. Fat in the liver and insulin resistance. Ann Med 2005;37:347–356. 34. Seppala-Lindroos A, Vehkavaara S, Hakkinen AM, Goto T, Westerbacka J, Sovijarvi A, Halavaara J, Yki-Jarvinen H. Fat accumulation in the liver is associated with defects in insulin suppression of glucose production and serum free fatty acids
GASTROENTEROLOGY Vol. 132, No. 2
independent of obesity in normal men. J Clin Endocrinol Metab 2002;87:3023–3028. 35. Nolte W, Hartmann H, Ramadori G. Glucose and insulin metabolism in cirrhosis. Adv Exp Med Biol 1997;420:229 –233. 36. Bahr MJ, Ockenga J, Boker KH, Manns MP, Tietge UJ. Elevated resistin levels in cirrhosis are associated with the proinflammatory state and altered hepatic glucose metabolism but not with insulin resistance. Am J Physiol Endocrinol Metab 2006;291: E199 –E206.
Received September 19, 2006. Accepted November 16, 2006. Address requests for reprints to: Patricia Iozzo, MD, PhD, Turku PET Centre, University of Turku, PO Box 52, FIN-20521, Turku, Finland. e-mail: [email protected]; fax: (39) 050 3152166. This work is part of the project “Hepatic and Adipose Tissue and Functions in the Metabolic Syndrome” (HEPADIP, see http://www. hepadip.org/), which is supported by the European Commission as an Integrated Project under the 6th Framework Programme (contract LSHM-CT-2005-018734). The study was further supported by grants from the Academy of Finland (206359 to P.N.), Finnish Diabetes Foundation (to P.I.), EFSD/Eli-Lilly (fellowship to P.I.), Sigrid Juselius Foundation (to P.I.), and Novo Nordisk Foundation (to P.N.). The authors thank the technical staff of the Turku PET Centre for the efforts and skills dedicated to this project.
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
PATRICIA IOZZO,*,‡ MIKKO J. JARVISALO,* JAN KISS,* RONALD BORRA,* GRATIAN A. NAUM,* ANTTI VILJANEN,* TAPIO VILJANEN,* AMALIA GASTALDELLI,‡ EMMA BUZZIGOLI,‡ LETIZIA GUIDUCCI,‡,§ ELISABETTA BARSOTTI,㛳 TIMO SAVUNEN,¶ JUHANI KNUUTI,* MERJA HAAPARANTA–SOLIN,* ELE FERRANNINI,‡,㛳 and PIRJO NUUTILA*,# *Turku PET Centre and Departments of ¶Surgery and #Medicine, University of Turku, Turku, Finland; and ‡PET Centre, Institute of Clinical Physiology, CNR National Research Council, §SSSUP-Medical Sciences Branch, and 㛳Department of Internal Medicine, University of Pisa School of Medicine, Pisa, Italy
See editorial on page 794. Background & Aims: The liver is inaccessible to organ balance measurements in humans. To validate [18F]fluorodeoxyglucose ([18F]FDG) positron emission tomography (PET) in the quantification of hepatic glucose uptake (HGU), we determined [18F]FDG modeling parameters, lumped constant (LC), and input functions (single arterial versus dual). Methods: Anesthetized pigs were studied during fasting (n ⴝ 6), physiologic (n ⴝ 4), and supraphysiologic (n ⴝ 4) hyperinsulinemia. PET was performed with C15O (blood pool) and [18F]FDG (glucose uptake). 6,6-Deuterated glucose ([2H]G) was coinjected with [18F]FDG and blood collected from the carotid artery and portal and hepatic veins to compute LC as ratio between tracers fractional extraction. HGU was estimated from PET images and ex vivo from high-performance liquid chromatography measurements of liver [18F]FDG versus [18F]FDG-6-phosphate and [18F]glycogen. Endogenous glucose production was measured with [2H]G and hepatic blood flow by flowmeters. Results: HGU was increased in hyperinsulinemia versus fasting (P < .05). Fractional extraction of [18F]FDG and [2H]G was similar (not significant), intercorrelated (r ⴝ 0.98, P < .0001), and equally higher during hyperinsulinemia than fasting (P < .05), with an LC of 0.98 ⴞ 0.10 and 1.18 ⴞ 0.26, respectively. [18F]FDGPET modeling provided HGU values that did not differ from, and were correlated with, those from ex vivo measurements (r ⴝ 0.61, P < .02); proportional estimates of liver perfusion and endogenous glucose production were also obtained. Single and dual input functions produced strongly intercorrelated results (r > 0.95, P < .0001), with a modest underestimation of
HGU by the former. Conclusions: [18F]FDG-PET– derived parameters provide accurate quantification of HGU and estimates of liver perfusion and glucose production. In the liver, LC of [18F]FDG is nearly unitary. Using a single arterial input introduces only a small error in estimation of HGU.
A
bnormalities in hepatic glucose uptake (HGU) have been implicated in the pathogenesis of liver steatosis, hypertriglyceridemia, and diabetes.1–3 Because of its anatomic location and dual blood supply, the liver is inaccessible to direct organ balance measurements in humans in vivo. Positron emission tomography (PET) in combination with [18F]fluorodeoxyglucose ([18F]FDG) has been previously evaluated by us and others for the assessment of HGU in humans4,5 and animals.6 These studies suggest that the methodology may be valid provided that its unknowns and assumptions are better understood and validated. The conversion of [18F]FDG to glucose uptake requires knowledge of the lumped constant (LC) term, which accounts for the potentially different affinity of the 2 molecules for cellular transport and phosphorylation processes. In extrahepatic organs, this term has been shown to range between 1.2 (in skeletal muscle7,8) and 0.52 (in the brain9). In some organs, it has been suggested to vary between fasting and hyperinsulinemic conditions.10 Although some previous in vitro evidence,11 toAbbreviations used in this paper: EGP, endogenous glucose production; [18F]FDG, [18F]fluorodeoxyglucose; [18F]FDG-6P, [18F]fluorodeoxyglucose-6-phosphate; FE, fractional extraction; [2H]G, deuterated glucose; HGU, hepatic glucose uptake; HKi, hepatic glucose influx rate constant; HPLC, high-performance liquid chromatography; LC, lumped constant; PET, positron emission tomography. © 2007 by the AGA Institute 0016-5085/07/$32.00 doi:10.1053/j.gastro.2006.12.040
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
Quantification of Liver Glucose Metabolism by Positron Emission Tomography: Validation Study in Pigs
532
IOZZO ET AL
GASTROENTEROLOGY Vol. 132, No. 2
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
Figure 1. Study design. Continuous lines (top) indicate plasma sampling for tracer concentration measurements after tracer administration (small arrows); boxes represent PET scanning periods (light gray), liver tissue sampling and organ explantation (dark gray), and study conditions/groups (bottom, white boxes); and thick arrows indicate time points for blood collection (black) and liver Doppler blood flow measurements (gray).
gether with the in vivo comparison of liver inward transport of 11C-methyl glucose and [18F]FDG in pigs,6 support the use of a unitary LC, this term has not been determined experimentally. The mathematical approaches used to quantify liver [18F]FDG uptake and phosphorylation include graphical analysis4,6 and 3-compartmental modeling with 4 rate constants (4k),4 – 6 the latter accounting for the potential dephosphorylation of [18F]FDG-6-phosphate ([18F]FDG6P) occurring in hepatocytes. Both provide an estimation of a composite parameter (ie, the fractional extraction of the tracer) intended as a unidirectional influx rate constant, which can be used to compute HGU once the LC is known. The first rate constant in compartmental modeling (K1), reflecting hepatic tracer delivery, is assumed to be proportional to organ perfusion,4 – 6 but validation of this concept has not been conclusively provided. The rationale of a 4k as opposed to a 3k or irreversible uptake model derives from the notion that the liver possesses the glucose-6-phosphatase enzyme, implying that the last reverse rate constant in the model (k4) may reflect the activity of this enzyme. However, hepatic glucose uptake and output are compartmentalized processes within the liver, and the appropriateness of graphical analysis in HGU computations suggests irreversible uptake, especially during insulin stimulation. To date, there are no data to document a relationship between the rate of hepatic or endogenous glucose production (EGP) and k4. Once the LC is known and the appropriateness of the model is validated, quantification of regional processes by PET imaging still requires knowledge of the input function, which represents the concentration of tracer available for organ extraction over the measurement time. Given the dual perfusion of the liver by the portal vein and hepatic artery, the error introduced by using a
single arterial input function should be considered when utilizing the methodology in humans, in whom only arterial blood can be accessed. Previous data in pigs suggest that the substitution may predominantly affect the estimation of initial steps in tracer uptake and liver vascular volume.6 The aim of this study was to validate the use of [18F]FDG-PET in the quantification of HGU in the fasting and insulin-stimulated state. To this purpose, pigs underwent arterial, portal, and hepatic vein catheterization in combination with the simultaneous administration and sampling of [18F]FDG and 6,6-deuterated glucose ([2H]G) across the organ, PET imaging of the liver blood pool and [18F]FDG kinetics, and ex vivo measurements of [18F]FDG, [18F]FDG-6P, and [18F]-labeled glycogen.
Materials and Methods Study Design The study design is schematized in Figure 1. After animal preparation, PET imaging was performed to measure liver blood content and glucose uptake during fasting (n ⫽ 6), physiologic (n ⫽ 4), and supraphysiologic (n ⫽ 4) euglycemic hyperinsulinemia. [2H]G and [18F]FDG were coinjected, and their concentrations were frequently measured in the carotid artery, portal vein, and hepatic vein. HGU was calculated from PET data by using graphical analysis and 3k versus 4k compartmental modeling. EGP was measured by stable isotope approaches ([2H]G). Immediately after the animals were killed, ex vivo [18F]FDG, [18F]FDG-6P, and [18F]-glycogen were assessed in liver biopsy specimens; the liver was then explanted to measure organ density. The protocol was
reviewed and approved by the Ethical Committee for Animal Experiments of the University of Turku.
Animal Preparation Fourteen anesthetized, weight-matched pigs were studied during fasting (weight, 29.8 ⫾ 0.6 kg), physiologic (1.0 mU · kg⫺1 · min⫺1; weight, 30.0 ⫾ 0.5 kg), or supraphysiologic euglycemic hyperinsulinemia (5.0 mU · kg⫺1 · min⫺1; weight, 30.3 ⫾ 0.5 kg). Animals were deprived of food on the day before the study at 5 PM. Anesthesia was induced by injection of 1.0 g ketamine into the neck muscles before transportation of the pigs to the operating room. Throughout the experiment, animals were kept anesthetized with ketamine and pancuronium (total of 1.5 g and 40 mg, respectively) and mechanically ventilated via tracheal intubation with oxygen and normal room air (regulated ventilation, 16 breaths/min). Catheters were placed in the femoral vein and carotid artery for the administration of glucose, insulin, [2H]G, and [18F]FDG and for sampling of arterial blood, respectively. Splanchnic vessels were accessed by subcostal incision; after dissection of the hepatogastric ligament, purse-string sutures were allocated to allow catheter insertion via a small incision in the hepatic and portal veins. Doppler flow probes were placed around the portal vein and hepatic artery; blood flow was monitored continuously and recorded 8 times (Figure 1). The surgical access was closed, and the distal catheter extremities were secured to the abdominal surface to avoid tip displacement. The animals were then transported to the PET Centre for tracer administration, liver imaging, and blood sampling.
PET Scanning Scans were performed using an ECAT 931-08/12 scanner (CTI Inc, Knoxville, TN) with a 10.5-cm axial field of view and a resolution of 6.7 mm (axial) ⫻ 6.5 mm (in-plane) full width at half maximum. In the hyperinsulinemic clamp studies, insulin was infused in a primecontinuous fashion for 240 minutes via the venous cannula and euglycemia was maintained by measuring arterial plasma glucose levels at 5- to 10-minute intervals and adjusting a variable rate of a 10% dextrose intravenous infusion accordingly.12 In the fasting studies, saline was infused instead of insulin and glucose. After acquisition of a transmission scan to correct for photon attenuation, 15O-labeled carbon monoxide (C15O) was administered by inhalation and a static scan was performed to image the liver blood pool; 3 blood samples were obtained at equilibrium during the imaging period to measure whole blood radioactivity concentrations. Approximately 60 minutes into the insulin or saline infusions, [18F]FDG13 (274 ⫾ 7 MBq) and [2H]G (fast, 451 ⫾ 9; clamp, 1043 ⫾ 26 mol) were rapidly coinjected, an 180-minute dynamic [18F]FDG PET scan was started (31 frames, 8 ⫻ 15, 2 ⫻3 0, 2 ⫻ 120, 1 ⫻ 180, 6 ⫻ 300, 8 ⫻
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
533
600, 4 ⫻ 900 seconds), and arterial, portal venous, and hepatic venous blood was frequently sampled (ie, once every imaging time frame) for the determination of plasma [18F]FDG radioactivity and [2H]G tracer-to-tracee ratio (by gas chromatography/mass spectrometry14). Tracer-to-tracee ratio was multiplied by steady-state plasma glucose levels to obtain absolute [2H]G concentrations. Vital signs, blood pressure, and heart rate were monitored throughout the study. Additional blood samples were obtained for the assessment of plasma glucose, lactate, fatty acid, and insulin levels, which were assayed as previously described.2,15
Ex Vivo Sample Processing At the end of the experimental period, animals were killed by potassium chloride injection and anesthetic overdose, the abdominal cavity was rapidly accessed, and liver specimens were obtained from 4 animals per group, weighed, measured for whole [18F] activity, and prepared for high-performance liquid chromatography (HPLC) analysis and glycogen precipitation. Then, the whole organ was explanted and weighed and its volume was measured by water displacement; liver density was calculated as the ratio of organ weight to volume. Liver samples were homogenized in 0.9 mol/L PCA, and the supernatants were neutralized with 5 mol/L KOH and then centrifuged and analyzed by HPLC as described earlier.16 The relative amounts of [18F]FDG and [18F]FGD-6P were calculated from the integrated HPLC peak areas and multiplied by the total radioactivity in 1 mL of tissue sample to obtain respective activity concentrations per unit volume. Liver glycogen was separated as previously described17 and counted for [18F] radioactivity.
Image Processing All sinograms were corrected for tissue attenuation, dead time, and decay and reconstructed through standard reconstruction algorithms in a 256 ⫻ 256 matrix. Final in-plane resolution of reconstructed and Hann-filtered images was ⬃10 mm at 10 cm from the center of the gantry. Regions of interest were drawn in the right lobe of the liver over 3– 6 transaxial planes; regions of interest were originally delineated on the [18F]FDG image to obtain hepatic [18F]FDG time-activity curves. The same regions of interest were copied in respective C15O images and visually inspected to rule out animal movement between the 2 imaging periods. The liver blood pool (as a fraction) was computed by dividing C15O radioactivity concentrations in 1 mL of tissue by C15O radioactivity concentrations in 1 mL of blood. Plasma and liver [18F]FDG time-activity curves were used to estimate HGU by using graphical analysis and compartmental modeling. In the former,18,19 a graph is generated by plotting
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
February 2007
534
IOZZO ET AL
Ct(t) ⁄ Cp(t) versus
GASTROENTEROLOGY Vol. 132, No. 2
兰 C ()d ⁄ C (t), t
0
p
p
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
where Ct and Cp are tissue and plasma radioactivity levels at each sampling time point (t). When tracer influx occurs, the 2 variables describe a linear relationship after a few minutes of equilibration, provided that an irreversible compartment is present. The influx rate constant (HKi) is then given by the slope of the linear fit of the data after excluding the first few values. In compartmental modeling,4 – 6 3 compartments, with either 3 (3k model, K1, k2, k3) or 4 (4k model, K1, k2, k3, k4) rate constants, were used to describe the kinetics of [18F]FDG in the liver. The 3 compartments describe plasma [18F]FDG, tissue [18F]FDG, and tissue [18F]FDG6P, respectively. Model parameters are interrelated through the following equations: dCe(t) ⁄ dt ⫽ K1Cp(t) ⫺ (k2 ⫹ k3)Ce(t) ⫹ k4Cm(t) dCm(t) ⁄ dt ⫽ k3Ce(t) ⫺ k4Cm(t) Ct(t) ⫽ (1 ⫺ VB)(Ce(t) ⫹ Cm(t)) ⫹ VBCp(t). Ct is activity in tissue at each sampling time point (t) and Cp is activity in plasma, and Ce and Cm are the concentrations of nonmetabolized and phosphorylated [18F]FDG in tissue, respectively; VB is the vascular volume fraction. Because free exchange of solutes and macromolecules between blood and extravascular space occurs in the liver, K1 should reflect blood flow to the organ (mL · min⫺1 · mL⫺1 of tissue) and k2 (min⫺1) is the reverse rate constant. The rate constants k3 (min⫺1) and k4 (min⫺1) reflect [18F]FDG phosphorylation and dephosphorylation, respectively; in the 3k model, k4 was set to 0. Kinetic parameters were derived by nonlinear least-square fit of data. In this case, the influx constant (HKi) was calculated as follows: Ki (mL · min⫺1 · mL⫺1) ⫽ (k1k3) ⁄ (k2 ⫹ k3) Rate-constant values were multiplied by steady-state plasma glucose concentrations and divided by the LC, determined as described in the following text to derive HGU (mol · min⫺1 · mL⫺1).
Validation Adequacy of model. In graphical analysis, the ap-
propriateness of the quantitative procedure was evaluated through the regression coefficient of the fit of measured values. In this analysis, linearity of measurements confirms the existence of an irreversible compartment.18,19 Adequacy of compartmental model configuration was based on the weighted sum of squares between model-predicted and
measured data, the Akaike information criterion20 (AIC ⫽ N ln RSS ⫹ 2 P), and the Schwartz criterion21 (SC ⫽ N ln RSS ⫹ P ln N), where RSS equals residual sum of squares, N is the number of data points, and P is the number of parameters included in the model. In comparing models with a different number of parameters, the usefulness of these criteria is given by the notion that the best model is not necessarily the one producing the smallest weighted sum of squares, because adding more parameters may reduce the weighted sum of squares. The Akaike information criterion and Schwartz criterion provide a balance between fitting precision and number of parameters. An inverse relationship links the adequacy of the model to the numerical scores derived from the previously described formulas. LC. The LC evaluates the difference between [18F]FDG and glucose uptake by the organ of interest; it is expressed as a ratio and used to convert the former into the latter. The hepatic LC was determined as previously described for the assessment of the LC in skeletal muscle7 from bolus tracer administration. Fractional extractions (FE) of [18F]FDG and [2H]G were calculated as described7 over different time lengths (60, 120, and 180 minutes) throughout the study period according to the formula FE[18F]FDG (or FE[2H]G) ⫽ 兺 (CAP ⫺ CV) ⁄ 兺 CAP, where CAP indicates the concentration of tracer entering the organ, calculated as the product of the arterial and portal tracer concentrations multiplied by respective contribution to liver perfusion, and CV is the concentration of tracer exiting the system through the hepatic vein. Data were not weighted for the duration of respective frames to avoid magnifying the larger error associated with the lower [2H]G concentrations occurring at later points, corresponding to longer intervals. LC was defined as the ratio between FE[18F]FDG and FE[2H]G Ex vivo versus PET modeling data. To test the appropriateness of the PET-derived data in estimating the rate of liver trapping of phosphorylated [18F]FDG, values of HKi and HGU obtained from graphical analysis and 3k modeling were compared with those calculated from ex vivo measurements. Because the latter are performed in blood-free tissue, for this comparison modeling of PET data was repeated after correcting liver timeactivity curves by their blood content by using the conventional formula Time-Activity CurveT ⫽ (Ct ⫺ Cp ⫻ Liver Blood Pool) ⁄ Tissue Fraction, in which Cp and Ct are plasma and whole tissue regions of interest radioactivity levels, the liver blood pool was obtained from C15O images, as described previously, and the tissue fraction (T) was computed as (1 ⫺ liver blood pool). Ex vivo
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
February 2007
535
Table 1. Biochemical Parameters During [18F]FDG-PET Imaging
Artery Glucose (mmol · L⫺1) Lactate (mmol · L⫺1) Insulin (mU · L⫺1) Free fatty acid (mmol · L⫺1) Portal vein Glucose (mmol · L⫺1) Lactate (mmol · L⫺1) Insulin (mU · L⫺1) Free fatty acid (mmol · L⫺1)
Fasting
Physiologic
Supraphysiologic
3.7 ⫾ 4 2.3 ⫾ 0.7 3⫾1 0.51 ⫾ 0.09
5.0 ⫾ 0.1a 4.0 ⫾ 0.5b 48 ⫾ 3b 0.07 ⫾ 0.01d
5.0 ⫾ 0.1a 2.0 ⫾ 0.4 234 ⫾ 35c 0.06 ⫾ 0.01d
3.9 ⫾ 0.5e 2.3 ⫾ 0.7 5 ⫾ 2f 0.51 ⫾ 0.09
5.0 ⫾ 0.0a 4.1 ⫾ 0.5b 32 ⫾ 2b,e 0.07 ⫾ 0.01d
4.9 ⫾ 0.1a 2.2 ⫾ 0.3e 205 ⫾ 25c,f 0.06 ⫾ 0.01d
ⱕ .05, ⱕ .1, and cP ⬍ .0001 versus fasting and physiologic hyperinsulinemia. dP ⬍ .001 versus fasting. eP ⱕ .03 and fP ⱕ .1 versus arterial values. aP
bP
derived HKiT and HGUT were obtained according to previously published methods,22,23 in which HKiT is given by the ratio [18F]FDG-6P Radioactivity ⁄
兰
0⫺180
Cp
and HGUT is calculated as the product of plasma glucose and HKi divided by LC. In the previously listed equations, Cp represents the dual liver input function. Liver plasma flow versus K1. Liver blood flow values, as determined by Doppler flowmeters, were multiplied by (1 ⫺ hematocrit) to obtain corresponding plasma flow figures. They were compared with K1 values obtained from 3k and 4k compartmental modeling. EGP versus k4. We also determined whether k4 is related to glucose release from the liver, such that changes in the k3/k4 ratio indicate the balance between phosphorylation and dephosphorylation. First, a comparison was made between these parameters in the different metabolic states of the study, based on the fact that hyperinsulinemia suppresses EGP and should therefore result in lower k4 values and higher k3/k4 ratios than during fasting. Second, an association was sought between k4 and EGP, as quantified by using arterial [2H]G measurements. Values of EGP in these animals have been previously reported.24 Dual versus single input function. Both linear and nonlinear model fitting procedures were repeated by using either a dual or a single input function. The dual input concentrations were calculated as the flowweighted input concentrations by using measured arterial and portal flow values: (Cartery ⫻ Hepatic Artery Flow ⫹ Cporta ⫻ Portal Vein Flow) ⁄ Whole Liver Flow,
where Cartery and Cporta refer to radioactivity (15O and [18F]) concentrations in the artery and portal vein, respectively. All parameters in each model were compared to evaluate the error introduced by using a single input function.
Statistical Analysis All data are presented as mean ⫾ SEM. Differences in paired data were evaluated using Student paired t test. One-way analysis of variance was used for unpaired group comparisons. Regression analyses were performed according to standard techniques. P ⬍ .05 was considered statistically significant.
Results The metabolic data of the study animals are given in Table 1. Plasma glucose levels were similar between the 2 hyperinsulinemic periods and slightly higher than during fasting conditions. Insulin levels were progressively higher in the fasting, physiologic, and supraphysiologic hyperinsulinemia experiments. During the imaging period, lactate levels did not differ significantly between the fasting and insulin-stimulated groups; however, as compared with respective baseline values, they were significantly suppressed during hyperinsulinemia (P ⱕ .002). Fatty acid levels were ⬎85% suppressed in the insulin infusion studies as compared with fasting values (P ⬍ .0001). Liver volume and weight were 758 ⫾ 30 mL and 814 ⫾ 26 g, respectively; thus, organ density was 1.08 ⫾ 0.03 g · mL⫺1. Liver blood flow, as determined by Doppler probes, was 990 ⫾ 124 and 852 ⫾ 120 mL · min⫺1 in the fasting and hyperinsulinemic states, respectively, and remained constant throughout repeated measurements during [18F]FDG scanning (coefficient of variation, 6.4% ⫾ 0.8%). The contributions of the portal and hepatic arterial vessels to total liver flow were 83% ⫾ 3% and 17% ⫾ 3%,
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
Hyperinsulinemia
536
IOZZO ET AL
GASTROENTEROLOGY Vol. 132, No. 2
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
Figure 2. Representative time course of plasma concentrations of [18F]FDG (left) and [2H]G (right) in the artery (white bars), portal vein (gray bars), and hepatic vein (black bars) during the experiments; the expanded representation of early time points in the histogram allows better visualization of rapidly changing tracer concentrations.
respectively, in all 3 metabolic conditions (not significant). Liver blood pool, as determined by C15O PET imaging, was 30% ⫾ 2% in the fasting experiment, 37% ⫾ 1% during physiologic hyperinsulinemia (P ⫽ .02 vs fasting), and 35% ⫾ 2% at supraphysiologic hyperinsulinemia (P ⫽ .07 vs fasting).
LC One representative set of arterial, portal, and hepatic venous concentrations of [18F]FDG and [2H]G in each experimental condition is given in Figure 2. Reliable tracer-to-tracee ratio measurements could not be ob-
tained in one case, which was excluded from paired analysis. Between the 2 hyperisulinemic states, FE values were overlapping (P ⱖ .60) and were therefore pooled for statistical analysis. No significant difference was found between FE of [18F]FDG and [2H]G (Table 2), showing a similar trend between the 2 tracers across different metabolic states and time intervals. FE measurements were not affected by the time over which the arterial-venous (A-V) differences were measured. Individual FE values of [18F]FDG and [2H]G were strongly correlated (r ⫽ 0.98, P ⬍ .0001), with an intercept of 0.01 and a slope of 0.96. The FE of both tracers was similarly higher during hy-
Table 2. FE of [18F]FDG and [2H]G and Calculated Liver LC Across Progressive Time Intervals [18F]FDG (FE) Fasting (n ⫽ 6) 60 min 120 min 180 min Average Hyperinsulinemia (n ⫽ 7)a 60 min 120 min 180 min Average
⫺0.51 ⫾ 0.37 ⫺0.49 ⫾ 0.35 ⫺0.48 ⫾ 0.34 ⫺0.49 ⫾ 0.35 0.22 ⫾ 0.02b 0.21 ⫾ 0.02b 0.21 ⫾ 0.02b 0.21 ⫾ 0.02b
[2H]G (FE) ⫺0.56 ⫾ 0.35 ⫺0.56 ⫾ 0.33 ⫺0.55 ⫾ 0.32 ⫺0.56 ⫾ 0.33 0.23 ⫾ 0.02b 0.23 ⫾ 0.02b 0.23 ⫾ 0.02b 0.23 ⫾ 0.02b
P value
LC ([18F]FDG/[2H]G)
.40 .39 .42 NS
1.22 ⫾ 0.27 1.19 ⫾ 0.27 1.14 ⫾ 0.25 1.18 ⫾ 0.26
.74 .53 .29 NS
1.03 ⫾ 0.14c 0.99 ⫾ 0.05c 0.93 ⫾ 0.11c 0.98 ⫾ 0.10c
aData from hyperinsulinemic experiments are pooled (P ⱖ .60 between 1.0 and 5.0 mU · kg⫺1 · min⫺1 insulin infusion experiments). Average values in each group are also shown, NS between time intervals. bP ⱕ .05, cP ⫽ .50 (NS) versus fasting.
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
537
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
February 2007
Figure 3. Relationship between parameters estimated with 3k and 4k compartmental modeling.
perinsulinemic than fasting conditions (P ⱕ .05). The calculated LC was 1.18 ⫾ 0.26 and 0.98 ⫾ 0.10 during fasting and hyperinsulinemia, respectively (P ⫽ .50 between the 2 conditions). These results were used in the computation of HGU.
ance between these factors resulted in higher fasting but similar insulin-mediated HGU values in the 4k versus 3k model (Figure 4); consequently, insulin-mediated HGU
Adequacy of Model and HGU Data obtained by using a dual input function as the reference method were examined. By graphical analysis, the linear fit approximated measured data very closely, with r values of 0.996 ⫾ 0.001 during fasting, 0.999 ⫾ 0.000 at physiologic hyperinsulinemia, and 0.999 ⫾ 0.001 at supraphysiologic hyperinsulinemia. In compartmental modeling, either the 3k or the 4k nonlinear fit to the data was good, with similarly low weighted sum of squares values (average values, ⬍6.5; not significant in each metabolic state) and slightly lower 4k Akaike information criterion and Schwartz criterion indexes during fasting (P ⬍ .05 vs 3k model). Relationships between individual parameters derived by different approaches are shown in Figure 3. All of them showed a high degree of correlation, although K1, k2, and k3 were on average 5.3%, 8.5%, and 31% higher in 4k than 3k modeling (P ⱕ .002), especially during fasting. The bal-
Figure 4. Comparison between HGU, as determined by irreversible or reversible (4k) modeling. *P ⫽ .01 versus 3k model.
538
IOZZO ET AL
GASTROENTEROLOGY Vol. 132, No. 2
Table 3. Liver Fractional Analysis (HKiT) and Glucose Uptake (HGUT) as Determined by Ex Vivo Measurements Versus Graphical Analysis or 3k Compartmental Modeling
HKiT (mL · min⫺1 · mL⫺1) Fasting Physiologic hyperinsulinemia Supraphysiologic hyperinsulinemia HGUT (mol · min⫺1 · mL⫺1) Fasting Physiologic hyperinsulinemia Supraphysiologic hyperinsulinemia
Ex vivo measurements
Graphical analysis
P value versus ex vivo
3k modeling
P value versus ex vivo
0.0054 ⫾ 0.0009 0.0046 ⫾ 0.0008 0.0064 ⫾ 0.0016
0.0054 ⫾ 0.0005 0.0056 ⫾ 0.0026 0.0060 ⫾ 0.0003
.80 .37 .53
0.0060 ⫾ 0.0012 0.0058 ⫾ 0.0006 0.0062 ⫾ 0.0003
.40 .30 .83
0.017 ⫾ 0.003 0.023 ⫾ 0.002 0.032 ⫾ 0.004
0.019 ⫾ 0.003 0.028 ⫾ 0.003 0.030 ⫾ 0.002
.55 .37 .53
0.021 ⫾ 0.003 0.029 ⫾ 0.003 0.031 ⫾ 0.002
.29 .30 .81
NOTE. Statistical values between metabolic states within each approach are reported in the text. CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
by graphical analysis, 3k and 4k modeling was 69% (P ⱕ .02), 67% (P ⬍ .02), and 20% higher than fasting values, respectively, with no difference between the 2 hyperinsulinemic states.
Ex Vivo Versus PET Modeling Data Percentages of liver [18F]FDG and [18F]FDG-6P in all studies were 34% ⫾ 3% and 66% ⫾ 3%, respectively. Radioactivity in glycogen was close to background, and below precision sensitivity, accounting for ⬍5% of tissue [18F]. Ex vivo HGUT was significantly higher during supraphysiologic hyperinsulinemia than in the fasting state (P ⬍ .05), with intermediate values in the physiologic hyperinsulinemic situation. HKiT and HGUT values determined by ex vivo FDG-6P measurements in the liver did not differ from those derived from graphical analysis or 3k modeling (P ⫽ NS) (Table 3). Individual values showed a strong correlation between the ex vivo and other approaches (r ⫽ 0.61, P ⱕ .02).
Liver Plasma Flow Versus K1 Average plasma flow and K1 values were not different and showed a tendency to correlate. The correlation became significant when accounting for the arterial input delay from the sampling site to the liver. Such correction did not influence HGU comparisons. It improved the approximation of K1 to plasma flow values (K1 ⫽ 0.80 ⫾ 0.15 and 0.91 ⫾ 0.17 mL · min⫺1 · mL⫺1 in 3k and 4k modeling, respectively; Doppler plasma flow ⫽ 0.90 ⫾ 0.08 mL · min⫺1 · mL⫺1; P ⫽ NS), leading to a closer correlation (r ⫽ 0.57, P ⫽ .03).
because EGP is also determined from arterial measurements.
Dual Versus Single Input Function No significant difference was observed when calculating liver blood pool from single arterial or dual C15O blood concentrations (P ⫽ .83); individual values were tightly intercorrelated (r ⫽ 0.99, P ⬍ .0001, slope ⫽ 1.05). Both HKi and HGU derived with a single versus dual input function were strongly intercorrelated (r ⬎ 0.95, P ⬍ .0001), with slopes ranging from 0.88 to 0.96 (HKi) or 0.85 to 0.98 (HGU) and intercepts ranging from 0.00003 to 0.0004 (HKi) or – 0.001 to 0.001 (HGU) across the 3 modeling approaches. However, a nearly systematic underestimation was observed when using the arterial input function, with a ratio of HGU (arterial) to HGU (dual) of 0.94 ⫾ 0.01, 0.90 ⫾ 0.01, and 0.96 ⫾ 0.03 with graphical analysis (Figure 6), 3k and 4k modeling, respectively. On average, only 1%–2% of the difference was accounted for by the variations between arterial and portal vein plasma glucose levels; k3 and k4 estimates were similar between the 2 input choices (P ⫽ NS),
EGP Versus k4 By using a dual input function, k4 decreased progressively and k3/k4 increased from fasting to physiologic and supraphysiologic hyperinsulinemia (Figure 5). The association between EGP and k4 (r ⫽ ⫹0.39, P ⫽ .16) or k3/k4 (r ⫽ ⫺0.50, P ⫽ .067) fell short of statistical significance. The correlation was stronger when using a single input function to model liver PET data (with k4, r ⫽ 0.51, P ⫽ .065; with k3/k4, r ⫽ ⫺0.55, P ⫽ .043), likely
Figure 5. Dephosphorylation rate contant (k4) (left) and phosphorylation-to-dephosphorylation ratio (k3/k4) (right), showing a progressive response to incremental hyperinsulinemia. *P ⫽ .01 versus fasting and P ⫽ .10 versus physiologic hyperinsulinemia; **P ⫽ .02 versus supraphysiologic and P ⫽ .06 versus physiologic hyperinsulinemia.
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
539
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
February 2007
Figure 6. Representative comparison of graphically derived HKi and HGU values by using the arterial and the dual input, showing strong correlations between the 2 approaches (left panels) with a slight systematic underestimation by the former (right panels).
whereas K1 and k2 estimates were significantly lower with the single input function (P ⬍ .05). K1 values derived from the use of the arterial input function were still strongly correlated with liver plasma flow (3k model: r ⫽ 0.67, P ⫽ .009; 4k model: r ⫽ 0.68, P ⫽ .007) despite a ⬃50% underestimation with either model.
Discussion To our knowledge, no previous study has compared the quantitative parameters derived from modeling of liver [18F]FDG data with conventional ex vivo and glucose flux measurements during fasting and insulin dose-response stimulation. Our data show that 3k modeling and graphical analysis provide accurate estimates of the retention of [18F]FDG-6P in the liver. The use of a 4k model additionally incorporates the proportion of [18F]FDG cycling through the glucose-6-phosphatase step. These notions have so far been assumed on the basis of their biological plausibility; they were documented here for the first time. In fact, [18F]FDG and [18F]FDG-6P were directly measured in liver biopsy specimens for comparison and the rate constant of tracer accumulation was computed from ex vivo liver samples according to classic formulas.22,23 In agreement with the inherent assumption that [18F]FDG-6P is the only trapped radiochemical, [18F]-labeled glycogen did not contribute to the hepatic tracer retention to any measurable extent; in the current experiments, the formation of metabolites, if any, as
previously described,25 was below the detection limit. EGP was related to the dephosphorylation rate constant k4 (positive) and to the balance between phosophorylation and dephosphorylation (k3/k4) (negative). The fact that such correlations were not very strong may be partly explained by the notion that EGP does not exactly equal hepatic glucose output, because the former includes the contribution of glucose released from extrahepatic organs, which is not negligible under the postsurgical conditions of the present study.26,27 In line with the expected action of insulin, k4 was progressively lower and the k3/k4 ratio was higher during fasting versus physiologic versus supraphysiologic hyperinsulinemia. On the whole, this finding indicates that k4 is a fair indicator of the inhibition of hepatic glucose release. Notably, k4 was always ⬍⬍k3, in line with the current knowledge that hepatic glucose uptake and output are compartmentalized processes, in which the bulk of substrate for organ release derives from gluconeogenesis, glycogenolysis, or glucose cycling through glycolysis, with a minimal contribution of the glucose that has just entered hepatocytes. Thus, only in the fasting state did k4 translate into a measurable HGU difference between reversible and irreversible modeling. As the dephosphorylation rate was suppressed by insulin, the difference fell below detection, although the process remained measurable. Confirming previous evidence by other investigators6 and our current hypothesis, K1 was related to hepatic plasma flow. However, it
540
IOZZO ET AL
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
should be noted that K1 is influenced by organ flow and glucose transport. Because a more direct assessment of hepatic blood flow by using perfusion tracers in PET imaging has proven challenging and awaits further methodological developments,28 K1 may be a plausible surrogate indicator of this process. Our data confirm that insulin stimulates HGU by either augmenting the rate of substrate trapping or decreasing the rate of glucose dephosphorylation and efflux from the organ. This result, despite the limitations due to the anesthesia and postsurgical stress, confirms our previous evidence in humans.4 Although HGU responds to insulin in a dose-dependent manner in the low hyperinsulinemic range,29 our data document a lack of dose dependency once physiologic postprandial insulin concentrations are reached, because in the current studies an additional 4- to 6-fold elevation in insulin levels during euglycemia had no effect. There have been no previous studies on liver LC for [18F]FDG. Munk et al6 reported similar hepatic influx rate constants of 3-O-11C-methylglucose and [18F]FDG by modeling of PET data. Because 3-O-11C-methylglucose is not a substrate of hepatic glucokinase, as confirmed by the same investigators in their previous ex vivo work,25 the above comparison was limited to the effect of liver blood flow and glucose transport, which are not ratelimiting processes for HGU. In the current study, the combined use of [2H]G and [18F]FDG extends the comparison to the full sequence of events leading to the uptake of these tracers, including their phosphorylation. We obtained LC values of approximately 1.0, with a small deflection from 1.18 during fasting to 0.98 at hyperinsulinemia, which did not further decline in response to supraphysiologic insulin infusions. Similar to the vast majority of PET studies, our experimental design precluded the use of continuous tracer infusion, which is the more usual format of tracer administration for arterialvenous balance computations. However, it was important to describe the kinetics of [18F]FDG by dynamic PET studies, in which a rapid injection is the method of choice, because the fast clearance of tracer from plasma coupled with its progressive tissue trapping provides the highest target-to-background signal and image quality. Therefore, we used the method that was previously applied to determine skeletal muscle LC,7 which is based on classic approaches for nonconstant indicator concentrations.30 The current approach, consisting of frequent blood sampling and use of integrated hepatic inflow and outflow data over a prolonged study period, requires that metabolic conditions and organ blood flow are known to be stable, as documented by repeated measurements in the current study. LC was calculated by relating the FE of [18F]FDG to that of simultaneously administered [2H]G, assayed in the same blood samples, after allowing for [18F] decay. Under these conditions, some negative FE values observed during fasting do not indicate that there
GASTROENTEROLOGY Vol. 132, No. 2
is no uptake or that the organ is producing the tracer30 but are related to the dynamic nature of the experiments and the expected small differences between the large numbers typical of the fasting state. The acceleration of metabolism during insulin stimulation was documented by the significantly higher FEs, with identical figures and changes with the 2 tracers. Without providing exact glucose metabolic rates, the method adopted here is valid in comparing the fate of 2 metabolizable tracers. An alternative approach for the estimation of tissue LC is that of directly comparing the quantification of glucose uptake estimated from [18F]FDG PET imaging with the steadystate arterial-venous balance or microdialysis of unlabeled glucose. These methods may sound more straightforward, but they are not easily applicable to the liver, because the balance of glucose across this organ results from the sum of simultaneous substrate uptake and output. Although net HGU and EGP were available here, notably the latter includes the release of glucose from organs other than the liver. Because EGP surpasses HGU by several fold, as confirmed in our study, even a small error in the quantification of hepatic glucose release, such as is introduced by assuming its equivalence with EGP, prevents reliable comparison. In skeletal muscle, in which all methods have been used by different investigators,7,8 identical LC results were observed. Collectively, our data indicate that [18F]FDG LC in the liver is virtually constant through a broad range of insulin concentrations and that the uptake of [18F]FDG does not differ substantially from that of glucose. The comparison between modeling parameters estimated from the analysis of [18F]FDG PET data by using the arterial versus dual input function has been previously shown in pigs under fasting conditions.7 The investigators concluded that the use of arterial concentrations results in the underestimation of K1 and blood volume terms. In our hands, VB was dependent on the delay occurring between the arterial sampling site and the liver rather than on the input function (data not shown). On the contrary, the delay did not affect HGU and its correction did not normalize K1. We extended the experiments to physiologic and supraphysiologic hyperinsulinemia. The most relevant findings here were that the correlation between K1 and liver perfusion persisted with the use of an arterial input function, and HGU was minimally and systematically affected, without altering group comparisons. The application of this method in humans essentially depends on this result. The observed input-related difference of ⬃1–2 mol · min⫺1 in the estimation of whole liver HGU is considerably lower than its intersubject variability and likely of negligible clinical relevance. Hepatic pathophysiology was not evaluated in this study. Previous observations suggested the potential usefulness of this technique in understanding disease pathogenesis and identifying treatment targets, although the
interpretation of the results partly relied on plausible biological assumptions. The main aim of the study was to provide a thorough understanding of the parameters estimated with this technique to allow us to describe molecular mechanisms in a more specific and conclusive manner. The study provides some information on the physiology of liver glucose metabolism and insulin sensitivity, processes that may be impaired in most common liver disorders.31–36 The previously described aspects are prerequisites to the formulation and testing of a hypothesis concerning disease pathophysiology. In summary, irreversible models provide an accurate estimate of HGU, differing from overall unidirectional uptake only during fasting conditions, in which a small but significant loss of immediately phosphorylated glucose occurs. By increasing the complexity of the model, additional information on liver perfusion and glucose production can be obtained. The LC for the estimation of HGU with [18F]FDG was shown to be nearly unitary. The error introduced by using a single arterial input function in the computation of HGU seems to be clinically irrelevant, supporting the implementation of the methodology in humans. References 1. Basu A, Basu R, Shah P, Vella A, Johnson CM, Nair KS, Jensen MD, Schwenk WF, Rizza RA. Effects of type 2 diabetes on the ability of insulin and glucose to regulate splanchnic and muscle glucose metabolism: evidence for a defect in hepatic glucokinase activity. Diabetes 2000;49:272–283. 2. Iozzo P, Hallsten K, Oikonen V, Virtanen KA, Kemppainen J, Solin O, Ferrannini E, Knuuti J, Nuutila P. Insulin-mediated hepatic glucose uptake is impaired in type 2 diabetes: evidence for a relationship with glycemic control. J Clin Endocrinol Metab 2003; 88:2055–2060. 3. Ferre T, Riu E, Franckhauser S, Agudo J, Bosch F. Long-term overexpression of glucokinase in the liver of transgenic mice leads to insulin resistance. Diabetologia 2003;46:1662–1668. 4. Iozzo P, Geisler F, Oikonen V, Maki M, Takala T, Solin O, Ferrannini E, Knuuti J, Nuutila P. Insulin stimulates liver glucose uptake in humans: an 18F-FDG PET Study. J Nucl Med 2003;44:682– 689. 5. Choi Y, Hawkins RA, Huang SC, Brunken RC, Hoh CK, Messa C, Nitzsche EU, Phelps ME, Schelbert HR. Evaluation of the effect of glucose ingestion and kinetic model configurations of FDG in the normal liver. J Nucl Med 1994;35:818 – 823. 6. Munk OL, Bass L, Roelsgaard K, Bender D, Hansen SB, Keiding S. Liver kinetics of glucose analogs measured in pigs by PET: importance of dual-input blood sampling. J Nucl Med 2001;42: 795– 801. 7. Kelley DE, Williams KV, Price JC, Goodpaster B. Determination of the lumped constant for [18F] fluorodeoxyglucose in human skeletal muscle. J Nucl Med 1999;40:1798 –1804. 8. Peltoniemi P, Lonnroth P, Laine H, Oikonen V, Tolvanen T, Gronroos T, Strindberg L, Knuuti J, Nuutila P. Lumped constant for [(18)F]fluorodeoxyglucose in skeletal muscles of obese and nonobese humans. Am J Physiol Endocrinol Metab 2000;279:E1122– E1130. 9. Reivich M, Alavi A, Wolf A, Fowler J, Russell J, Arnett C, MacGregor RR, Shiue CY, Atkins H, Anand A, et al. Glucose metabolic rate kinetic model parameter determination in humans: the lumped constants and rate-constants for [18F]fluorodeoxyglu-
10.
11.
12.
13.
14. 15.
16.
17.
18.
19.
20. 21. 22.
23.
24.
25.
26.
27.
541
cose and [11C]deoxyglucose. J Cereb Blood Flow Metab 1985;5:179 –192. Botker HE, Bottcher M, Schmitz O, Gee A, Hansen SB, Cold GE, Nielsen TT, Gjedde A. Glucose uptake and lumped constant variability in normal human hearts determined with [18F]fluorodeoxyglucose. J Nucl Cardiol 1997;4:125–132. Sweet IR, Peterson L, Kroll K, Goodner CJ, Berry M, Graham MM. Effect of glucose on uptake of radiolabeled glucose, 2-DG, and 3-O-MG by the perfused rat liver. Am J Physiol 1996;271:E384 – E396. DeFronzo RA, Tobin JD, Andres R. Glucose clamp technique: a method for quantifying insulin secretion and resistance. Am J Physiol 1979;237:E214 –E223. Hamacher K, Coenen HH, Stöcklin G. Efficient stereospecific synthesis of no-carrier-added 2-[18F]-fluoro-2-deoxy-D-glucose using aminopolyether supported nucleophilic substitution. J Nucl Med 1986;27:235–238. Wolfe RR. Radioactive and stable isotopes in biomedicine. New York: Wiley-Liss, 1992. Muscelli E, Mari A, Natali A, Astiarraga BD, Camastra S, Frascerra SD, Holst JJ, Ferrannini E. Impact of incretin hormones on beta-cell function in subjects with normal or impaired glucose tolerance. Am J Physiol Endocrinol Metab 2006;291:E1144 – E1150. Suolinna EM, Haaparanta M, Paul R, Härkönen P, Solin O, Sipilä H. Metabolism of 2-[18F]fluoro-2-deoxyglucose in tumor-bearing rats: chromatographic and enzymatic studies. Nucl Med Biol 1986;13:577–581. Barrett EJ, Bevilacqua S, DeFronzo RA, Ferrannini E. Glycogen turnover during refeeding in the postabsorptive dog: implications for the estimation of glycogen formation using tracer methods. Metabolism 1994;43:285–292. Patlak CS, Blasberg RG, Fenstermacher JD. Graphical evaluation of blood-to-brain transfer constants from multiple-time uptake data. J Cereb Blood Flow Metab 1983;3:1–7. Patlak CS, Blasberg RG. Graphical evaluation of blood-to-brain transfer constants from multiple-time uptake data. Generalizations. J Cereb Blood Flow Metab 1985;5:584 –590. Akaike H. A new look at the statistical model identification. IEEE Trans Automat Contr 1974;AC-19:716 –723. Schwartz G. Estimating the dimension of a model. Ann Stat 1978;6:461–564. Kraegen E, James DE, Jenkins AB, Chisholm DJ. Dose-response curves for in vivo insulin sensitivity in individual tissues in rats. Am J Physiol Endocrinol Metab 1985;248:E353–E362. Haaparanta M, Paul R, Gronroos T, Bergman J, Kamarainen EL, Solin O. Microdialysis and 2-[18F]fluoro-2-deoxy-D-glucose (FDG): a study on insulin action on FDG transport, uptake and metabolism in rat muscle, liver and adipose tissue. Life Sci 2003;73: 1437–1451. Iozzo P, Gastaldelli A, Jarvisalo MJ, Kiss J, Borra R, Buzzigoli E, Viljanen A, Naum G, Viljanen T, Oikonen V, Knuuti J, Savunen T, Salvadori PA, Ferrannini E, Nuutila P. 18F-FDG assessment of glucose disposal and production rates during fasting and insulin stimulation: a validation study. J Nucl Med 2006;47:1016 –1022. Bender D, Munk OL, Feng HQ, Keiding S. Metabolites of (18)FFDG and 3-O-(11)C-methylglucose in pig liver. J Nucl Med 2001; 42:1673–1678. Stumvoll M, Meyer C, Mitrakou A, Nadkarni V, Gerich JE. Renal glucose production and utilization: new aspects in humans. Diabetologia 1997;40:749 –757. Joseph SE, Heaton N, Potter D, Pernet A, Umpleby MA, Amiel SA. Renal glucose production compensates for the liver during the anhepatic phase of liver transplantation. Diabetes 2000;49: 450 – 456.
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
LIVER GLUCOSE METABOLISM BY [18F]FDG–PET
February 2007
542
IOZZO ET AL
CLINICAL–LIVER, PANCREAS, AND BILIARY TRACT
28. Ziegler SI, Haberkorn U, Byrne H, Tong C, Kaja S, Richolt JA, Byrne H, Tong C, Schosser R, Krieter H, Kaja S, Richolt JA, Lammertsma AA, Price P. Measurement of liver blood flow using oxygen-15 labelled water and dynamic positron emission tomography: limitations of model description. Eur J Nucl Med 1996;23: 169 –177. 29. Arora P, Basu R, Zangeneh F, Schweink W, Rizza R. The ability of insulin to stimulate the incorporation of extracellular glucose into hepatic glycogen is impaired in type 2 diabetes. Diabetes 2003; 52(Suppl 1):A59. 30. Zierler K. Theory of the use of arteriovenous concentration differences for measuring metabolism in steady and non-steady states. J Clin Invest 1961;40:2111–2125. 31. Hui JM, Sud A, Farrell GC, Bandara P, Byth K, Kench JG, McCaughan GW, George J. Insulin resistance is associated with chronic hepatitis C virus infection and fibrosis progression. Gastroenterology 2003;125:1695–1704. 32. Marchesini G, Babini M. Nonalcoholic fatty liver disease and the metabolic syndrome. Minerva Cardioangiol 2006;54:229 –239. 33. Yki-Jarvinen H. Fat in the liver and insulin resistance. Ann Med 2005;37:347–356. 34. Seppala-Lindroos A, Vehkavaara S, Hakkinen AM, Goto T, Westerbacka J, Sovijarvi A, Halavaara J, Yki-Jarvinen H. Fat accumulation in the liver is associated with defects in insulin suppression of glucose production and serum free fatty acids
GASTROENTEROLOGY Vol. 132, No. 2
independent of obesity in normal men. J Clin Endocrinol Metab 2002;87:3023–3028. 35. Nolte W, Hartmann H, Ramadori G. Glucose and insulin metabolism in cirrhosis. Adv Exp Med Biol 1997;420:229 –233. 36. Bahr MJ, Ockenga J, Boker KH, Manns MP, Tietge UJ. Elevated resistin levels in cirrhosis are associated with the proinflammatory state and altered hepatic glucose metabolism but not with insulin resistance. Am J Physiol Endocrinol Metab 2006;291: E199 –E206.
Received September 19, 2006. Accepted November 16, 2006. Address requests for reprints to: Patricia Iozzo, MD, PhD, Turku PET Centre, University of Turku, PO Box 52, FIN-20521, Turku, Finland. e-mail: [email protected]; fax: (39) 050 3152166. This work is part of the project “Hepatic and Adipose Tissue and Functions in the Metabolic Syndrome” (HEPADIP, see http://www. hepadip.org/), which is supported by the European Commission as an Integrated Project under the 6th Framework Programme (contract LSHM-CT-2005-018734). The study was further supported by grants from the Academy of Finland (206359 to P.N.), Finnish Diabetes Foundation (to P.I.), EFSD/Eli-Lilly (fellowship to P.I.), Sigrid Juselius Foundation (to P.I.), and Novo Nordisk Foundation (to P.N.). The authors thank the technical staff of the Turku PET Centre for the efforts and skills dedicated to this project.