Tumor models and small-animal imaging
Nineteen 6- to 8-week-old severe combined immunodeficiency (SCID) mice were maintained in a strict defined-flora, pathogen-free environment in the AAALAC-accredited animal facilities at UCLA. Human glioblastoma cell line U87 and breast cancer cell line MDA-MB-231(MDA) were used as a SCID-hu tumor model. Tumor cells were injected (with MDA cells in 11 mice and with U87 cells in 8) subcutaneously as single-cell suspensions in phosphate buffer saline (PBS; about 2 × 106 MDA or about 6 × 105 U87 cells in 100 μL PBS). When the diameter of the tumor grew to approximately 2.5 mm, a PET scan was performed once a week on the same animal until the diameter of the tumor exceeded 10 mm. Tumor size was measured weekly with a caliper, and the volume was calculated as 0.5 × length × width2.
Small-animal PET scans were performed either on a microPET Focus 220 scanner running microPET Manager 2.4.1.1 or on an Inveon dedicated PET running IAW 1.5 (Siemens Preclinical Solutions, Knoxville, TN, USA), but the same scanner was used for multiple longitudinal PET scans of each mouse. These scanners provide the same SUV and % ID values for mice imaged sequentially in both systems (unpublished data). List-mode PET data were acquired for 60 min immediately after 18 F-FDG injection via a tail vein catheter (18.28 ± 1.19 MBq, approximately 60 μL) in a bolus. Frame durations of all the PET studies were 4 × 1 s, 15 × 0.5 s, 1 × 2 s, 1 × 4 s, 1 × 6 s, 1 × 15 s, 3 × 30 s, 1 × 60 s, 1 × 120 s, 3 × 180 s, 3 × 900 s, and 1 × 51 s. After the PET scan was completed, a 10-min CT scan was acquired with a small-animal CT scanner (MicroCAT II, Siemens Preclinical Solutions, Knoxville, TN, USA) for attenuation correction of the PET measurements.
Tail vein blood glucose levels (in mmol/L) were measured using a blood glucose meter (Abbott AlphaTRAK, Abbott Laboratories, Abbott Park, IL, USA) at the beginning and the end of each scan. A single blood sample (approximately 10 to 15 μL) was collected with a 1-cc syringe from the heart at the end of the study (approximately 70 to 80 min). The whole blood in the syringe was released to a pre-weighed test tube and weighed, and the radioactivity was counted in a gamma counter (WIZARD 3"; PerkinElmer Life Sciences, Turku, Finland).
All animal experiments were performed in accordance with institutional guidelines and protocols approved by the Animal Research Committee of the University of California, Los Angeles, USA.
Image analysis
Image analysis was performed using AMIDE (http://amide.sourceforge.net/). The 3D isocontour regions of interest (ROIs) were manually defined for the skeletal muscle of the left ventricle (LV), forelegs, liver, and tumor in each mouse on the last 15-min images (about 60 min post 18 F-FDG injection). Only tumors without an apparent necrotic center on the 18 F-FDG PET images were included in the analysis. The time-activity curves (TAC) in each tissue of interest were calculated. Experimental information for all studies is available online at the UCLA Mouse Quantitation Project website (http://dragon.nuc.ucla.edu/mqp/index.html).
For semi-quantitative analysis, SUV was calculated using the mean voxel value within the ROI of the last 15-min frame (Equation 1). SUVs of each study were recorded and were used for generating the TACs.
(1)
Plasma input function
The plasma TAC (TACp; the input function) was derived based on the method reported by Ferl et al. [12]. The method included the use of the early-time LV TAC (t < 1 min) (with corrections of delay, dispersion, partial volume effects, and red blood cell uptake) and one whole blood sample taken at the end of the study (approximately 70 min). For each study, a time-dependent plasma-to-whole blood 18 F-FDG equilibrium ratio, R
PB(t) (Equation 2), was used to convert the whole blood 18 F-FDG concentrations to those in plasma [11].
(2)
where t is time in minutes after tracer injection. The input function was assumed to be describable with four exponential components (Equation 3):
(3)
The sum of the first three exponential terms was used to describe the main part of the input function; the fourth exponential term was needed so that TACp was equal to 0 at time 0. All parameters were estimated by simultaneously fitting the plasma 18 F-FDG blood curve with Equation 3 and the muscle and liver TACs with two separate 4 K compartmental 18 F-FDG models as described by Ferl et al. [12]. The kinetic modeling program SAAM II [13] was used to solve the systems of differential equations and estimate parameters. The Bayesian maximum a posteriori parameter estimation in SAAM II was used to improve parameter identification and the accuracy of the predicted input function as described by Ferl et al. [12].
Kinetic analysis
Both the PET image and blood data were converted to absolute radioactivity concentration (Bq/mL) using a cross-calibration factor derived from cylinder phantom experiments. The 18 F-FDG uptake rate constant K
i
(K
i
= K
1
k
3/(k
2 + k
3)) was estimated via the Patlak graphical analysis [14] using the derived plasma input function and tumor TAC data by taking the slope of the linear portion from 15 to 60 min of the plot based on Equation 4:
(4)
(5)
where C
T(t) is the total 18 F-FDG concentration in the tissue of interest, C
1(t) is the free 18 F-FDG concentration in the tissue, C
p(t) is the 18 F-FDG concentration in the plasma, C
B(t) is the blood 18 F-FDG concentration in the vasculature, and V
B is the volume fraction of blood in the tissue. The early-time tissue data (for t < 15 min) were not used because only after an equilibration time (t*) will the Patlak plot become linear. The metabolic rate of glucose (MRGlu) was calculated as MRGlu = K
i
× [Glc]/LC [15], where [Glc] was the averaged blood glucose level of the two measurements at the beginning and the end of the scan, and LC is the lumped constant. Values of the LC were assumed to be 1.4 for U87 [16] and 1.0 for MDA. The intercept value (Int) shown in Equation 5 is related to the V
B and the distribution volume of the tracer in the reversible tissue compartment.
Data analysis and functional relationship determination
K
i
and SUVs were partial volume (PV)-corrected by dividing the values by the recovery coefficient (RC) of the tumor. As a first-order approximation, RC for a tumor was calculated based on a sphere of diameter = the spatial resolution of the microPET scanner, and an assumed background activity level of 10% of the activity level in the sphere. Four models (Equations 6 to 9) were tested to fit the relationship of glucose concentration and tumor diameter to quantitative PET measures, Y (SUV/RC or K
i
/RC).
(7)
(9)
The parameters a and c were to account for the effect due to tumor growth in size; b was equivalent to the half saturation glucose concentration of 18 F-FDG uptake [11]. Models were tested separately for U87 and MDA. To account for possible effects introduced through repeated measures, models were fit using standard least-squares method as well as mixed-effects method [17]. For models 3 and 4 (Equations 8 and 9), if fitted values for parameter b were not found to be significant, the model was rebuilt, setting b equal to 0. Mixed-effects models were built assuming no within-group covariance of random effects. For models with more than one fitted parameter, all possible combinations of random effects were tested. If more than one mixed-effect model showed significant improvement over the standard fixed-effect model, the optimal mixed-effect model was chosen by the likelihood ratio test if models were nested or by the corrected Akaike's information criterion (AICc), otherwise. After a fitting method for each model was chosen, the optimal model for each cell line and response variable was chosen based on AICc. Fittings and model comparison were done using the nlme package of the statistical software R [18].