| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Clinical Investigation |
1 Department of Radiology, University of Washington, Seattle, Washington; 2 Department of Neurology, University of Washington, Seattle, Washington; and 3 Department of Statistics, University College Cork, Cork, Ireland
Correspondence: For correspondence or reprints contact: Mark Muzi, MS, University of Washington–Radiology, Box 356004, 1959 N.E. Pacific St., Seattle, WA 98195-6004. E-mail: muzi{at}u.washington.edu
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONReferences |
|---|
Key Words: 3'-deoxy-3'-fluorothymidine • kinetic modeling • glioma • blood–brain barrier disruption • thymidine kinase 1
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONReferences |
|---|
Recently, 3'-deoxy-3'-18F-fluorothymidine (FLT), a TdR analog, has been developed as an alternative to TdR for imaging proliferation (5,8). 18F-FLT offers the advantages of a longer-lived label with high specificity for thymidine kinase 1 (TK1) in the cytosol and few labeled metabolites. TK1 is highly regulated during the cell cycle and is highly expressed during S phase. Further FLT tissue metabolism produces phosphorylated products (nucleotides), which are retained in cells at a rate proportional to TK1 activity (9). FLT is not significantly incorporated into DNA because it lacks the 3'-hydroxyl, which is essential for chain propagation. 18F-FLT labels the intracellular nucleotide pool and is subject to retrograde metabolism (9). FLT anabolism primarily occurs in the liver to produce a glucuronide conjugate, which is exported to the blood and cleared by the kidneys (10). 18F-FLT-glucuronide appears to be the only observed metabolite contaminating the blood pool (10).
We have previously described a 2-tissue compartment, 4-rate constant model (2C) for 18F-FLT that has been validated for somatic tissues and successfully applied to patient studies (11,12). However, because the intact BBB restricts the transport of modified pyrimidine nucleotides such as FLT (7,13), and tumor growth or therapy can disrupt the BBB (14), further analysis was necessary to demonstrate that the model could distinguish between increased transport across a damaged BBB and increased retention of 18F-FLT in proliferating tumor tissue.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONReferences |
|---|
|
Imaging Procedure
The PET studies were performed on an Advance PET tomograph (GE Healthcare) providing 35 image planes over a 15-cm axial field of view with a 4.25-mm spacing (16). While a 25-min transmission scan with a 68Ge rotating sector source was underway, intravenous and intraarterial lines were introduced for isotope injection and arterial sampling. Arterial samples of 1 mL were obtained using an automated blood sampler (17) at 8 x 15 s, 2 x 30 s, 5 x 1 min, 1 x 2 min, and 16 x 5 min. Images were acquired in 3-dimensional (3D) mode with a dynamic sequence: 10 x 10 s, 4 x 20 s, 3 x 40 s, 3 x 1 min, 5 x 2 min, 4 x 3 min, and 12 x 5 min time frames for a total of 90 min. After correction for scattered and random coincidences, images were reconstructed by the method of 3D reprojection (18) with 6-mm Hanning, 4.5-mm radial, and 6-mm smoothing filters, resulting in an approximately isotropic image resolution of 6 mm.
Blood Sampling and Metabolite Analysis
For each arterial blood sample, 0.2 mL plasma were assayed for radioactivity using a COBRA 
-counter (Packard Instruments). Because FLT has negligible serum protein binding (19), all of the activity associated with 18F-FLT in blood was assumed to be available for tissue uptake. An aliquot (0.4 mL) of the plasma from 8 arterial samples (5, 10, 15, 20, 30, 45, 60, and 90 min) was assayed for the relative amount of FLT and FLT-glucuronide as described previously (11,20). The fraction of total activity present as FLT in each blood sample was fitted to a monoexponential curve to provide a continuous function describing the fraction of plasma activity associated with FLT (11). The fractional FLT curve was applied to each total blood activity curve to give a metabolite-corrected input function for further modeling analysis of each patient dataset.
Image Processing
MRI was performed with a 1.5-T Signa (GE Healthcare) with a standard head coil. The protocol for all subjects included a T1-weighted sequence acquired in the transverse plane before and after administration of intravenous gadolinium (Gd). MR images were coregistered to summed PET images with a method based on mutual information criteria (11).
Regions of interest (ROIs) for tumor and contralateral (C/L) brain regions (brain, gray and white matter) were identified on MRI T1+Gd or T2-weighted images and 18F-FLT images summed between 30 and 60 min, an interval when transport of the radiotracer from blood to tissue is predominantly unidirectional. The ROIs from contiguous slices were combined to create volumes of interest (VOIs) for each tissue type by means of Alice image-processing software (Perceptive Informatics, Inc.). Tumor regions were placed on all planes containing portions of the lesion as indicated from the MRI T1+Gd or 18F-FLT images.
Quantitative Analysis
Compartmental Modeling.
The previously described 2C kinetic model for FLT kinetics is illustrated in Figure 1 (12). Similar to the kinetic assumptions of the TdR model in brain tumors (21), FLT transport into the brain is influenced by blood flow and the BBB (13). Details of the FLT model have been reported previously (12).
|
Metabolic flux, KFLT, is estimated from parameters derived by fitting the FLT input function and the total blood activity curve to the tissue time–activity curve data. The flux constant is determined by the product of the rate constants (12,23):
 | (Eq. 1) |
Model Starting Parameters.
In our previous FLT tissue model (11), early analyses suggested that it was difficult to estimate K1 and k2 independently. Hence, the model was reparameterized using K1 and K1/k2 as floating variables, which has been an effective method for handling K1–k2 covariance (6,12,21). The individual rate constants, K1, K1/k2, k3, and k4, and the regional Vb, the fraction of vascular activity in the tissue VOI, were estimated during parameter optimization.
Starting parameters for FLT were based on FLT kinetic assessment of lung tumors (11) and TdR in gliomas (4). Prior studies in somatic tumors (11) have shown that estimates of the transport rate for FLT are lower than those for TdR (6). A lower K1 for FLT relative to TdR is expected, as cellular TdR transporters do not effectively transport analogs modified at the 3' position (24,25). Phosphorylation of FLT by TK1 (k3) is also anticipated to be lower than that for TdR, reflecting the differences in activity of TK1 for each substrate (26,27). Dephosphorylation of FLT-monophosphate, represented by k4 as estimated in lung tumors and muscle (11), is anticipated to occur at a similar level for gliomas and brain tissue. Thus, the initial model conditions were adjusted from the TdR starting points to reflect the less reactive behavior of FLT (Table 2).
|
Model Characteristics.
The proposed model was evaluated to determine the extent to which the information obtained from a typical imaging study is sufficient to produce a unique solution with identifiable parameters. To establish the most reliable approach for parameter estimation, the brain FLT model was characterized with respect to (a) parameter sensitivity, the degree to which a change in an individual input parameter results in a change in the output; (b) parameter identifiability, the ability of the model to estimate parameters independently; (c) susceptibility to noise, as determined by Monte Carlo error analysis; and (d) model accuracy, the ability to estimate key parameters accurately across the expected range of values. These methods have been described (6,12,21) and are not repeated here.
Parametric Image Analysis.
Parametric image maps of each rate constant were generated by mixture analysis (28,29). Mixture analysis applies the same biologically based model with identical floating parameters and blood input functions used in the analysis of tissue time–activity curves to the dynamic series of images for the production of regional parametric maps. The VOIs used to generate tissue time–activity curves were applied to the parametric images to determine the extent of FLT flux relative to Gd enhancement on MR images, as well as the precision and bias of the parametric image values relative to modeling analysis of tissue time–activity curves.
Model-Independent Analysis.
Simple, model-independent estimates of 18F-FLT uptake were assessed by the standard uptake value (SUV) determined from images obtained between 30 and 60 min after injection and by a modified graphical analysis (GA) (30), which corrects for blood metabolites. The GA determination of flux may be valid for a short interval after injection, when the assumption of unidirectional transfer of FLT from blood to tissue is applicable (31). Because of the restricted transport of FLT, tissue pools of precursor require a greater time to stabilize with respect to blood delivery. Therefore, 30 and 60 min were chosen as the time boundaries for the linear fit in GA after examination of linearity in CE brain tumor regions. This is a different time range than that determined for lung tumors (15–50 min), where initial transport of FLT into lung tumors is greater (11). The method also assumes that phosphorylated FLT nucleotides are completely retained in the tissue, an assumption that has yet to be validated in vivo and is not supported by cell culture experiments (9).
Statistical Analysis
The comparisons between 18F-FLT PET uptake parameters or model estimations were made using standard parametric statistical tests (Pearson correlation, paired Student t test). Statistical analyses were performed using the statistical software JMP (SAS Institute).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONReferences |
|---|
Sensitivity Analysis
Parameter sensitivity for other brain regions (gray and white matter) and noncontrast-enhancing (NCE) tumors were similar to C/L brain. In CE tumors, K1 has a greater contribution to 18F-FLT uptake than phosphorylation, k3. The sensitivity of K1 for CE tumor is similar to TdR in magnitude and time course (21), with a large impact on early model output that diminishes after 10 min. The k3 phosphorylation rate had the greatest sensitivity of any individual rate constant during the prolonged 18F-FLT uptake phase after blood–tissue equilibration (Fig. 2).
|
|
2% time–activity curve coefficient of variation [COV] at 60 min), there was a SE of <10% for the estimate of KFLT and K1 for CE tumors. The other parameters (Vd, k3, k4) had larger COV values and were less robust. The COV was greater in brain and NCE tumors (7% time–activity curve COV at 60 min) than that in CE tumors, reflecting the low level of 18F-FLT transport and uptake in a nonproliferating tissue with restricted access at the BBB.
|
|
|
|
|
|
K1 is higher for MRI CE tumors than for NCE tumors (P < 0.02) and KFLT is higher in high-grade tumors (mean, 0.018; n = 9) than that in lower-grade tumors (mean, 0.003; n = 3, P < 0.01). For CE tumors, we observed a finite rate of late label loss, with an average estimate for k4 of 0.025 min–1 (range, 0.017–0.032 min–1; n = 9). Consistent with this finding, a characteristic late downward curvature was observed in the GA plot for most CE gliomas. As previous reports have shown, ignoring k4 can lead to significant underestimation of flux (11,31). KFLT values estimated from compartmental modeling analysis were correlated with GA KFLT-GA (r = 0.84) and FLT maximum SUV (r = 0.71) and also highly correlated with K1 (r = 0.94).
Mixture Analysis
Mixture analysis–derived image maps of parameters were matched to ROI analysis for K1 (r = 0.98, SEE/mean = 5.9%, n = 10) and KFLT (r = 0.99, SEE/mean = 8.9%, n = 12) (Fig. 5). Parametric images from the mixture analysis method did not reveal significant transport (K1) or flux (KFLT) outside of the regions of contrast enhancement. Examples of patient parametric images of transport and metabolic flux appear in Figure 3.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONReferences |
|---|
Our results indicate that accurate assessment of proliferation in brain by 18F-FLT imaging requires analysis of uptake kinetics to separate transport effects from tissue retention due to metabolic trapping of FLT nucleotides. This process involves an assay of the FLT fraction in blood, as blood clearance varied widely among the patients. It is likely that the blood metabolite FLT-glucuronide does not enter normal cells (10) but may flow through the disrupted BBB into and out of the interstitial space during the imaging procedure. We assume the kinetics of this interaction and resulting tissue activity in normal and tumor regions can be accounted for by the vascular parameter Vb. Limited transit of FLT-glucuronide to and from the interstitial space is unlikely to affect model behavior, given the small quantity of the labeled metabolite present late in the imaging study.
As demonstrated in a patient with a recurrent pretreated low-grade oligodendroglioma with significant contrast enhancement (Fig. 3C), simple measures of tracer uptake, such as SUV, can be misleading when total 18F-FLT uptake is due in large part to transport across the BBB and not to trapping of FLT after phosphorylation by TK1. In addition, low transport can limit uptake even in proliferative tumors. Another patient (Table 1, patient. 2) with a grade III astrocytoma (determined by biopsy that showed 10% MIB-1 staining) had no contrast enhancement on MRI T1+Gd images. 18F-FLT uptake was similar to the low levels observed in normal brain. Simple measures of uptake that do not fully account for 18F-FLT transport, uptake, and loss from tissues can lead to incorrect interpretation of summed uptake images.
Transport impediments may pose a difficulty, however, in using FLT to assess residual viable brain tumor after therapy. In this case, transport may be transiently high because of treatment effects, and flux may be low in successfully treated tumors. The estimated KFLT would then be mistakenly higher than the actual flux rate and could lead to a conclusion of residual tumor instead of successful treatment. One potential application of FLT may be for brain tumors with high initial K1 and KFLT. After treatment, these patients may show an early reduction in KFLT due to decreased proliferation. This hypothesis would need to be tested in serial 18F-FLT imaging studies over the course of therapy.
Compartmental modeling provides separate estimates of both transport and flux (trapping) to account for 18F-FLT uptake. Simulations suggest that the transport parameter can be estimated with <15% COV and metabolic flux with <5% COV. However, at the extremes of transport, modeling estimates of metabolic flux may be less accurate. For values of K1 close to that of normal brain and characteristic of NCE brain tumors, transport limits uptake and the flux cannot be measured independent of transport. Flux values for normal brain fall within the error of model estimates and reflect restricted access of FLT across the BBB. This also appears to be true for NCE or minimally CE brain tumors. Thus, FLT may be less useful in assessing proliferation in NCE tumors regardless of histopathology grading.
18F-FLT PET may also have difficulty in differentiating residual proliferating tumor from BBB breakdown in regions that are not highly proliferating. At the extreme of high K1 and low KFLT, estimates of KFLT are imprecise. In fact, simulations with high transport and no phosphorylation (k3 = 0) produced model estimates of flux that were nonzero. This may limit the applicability of FLT in circumstances of low flux, such as would be expected in radionecrosis. This is distinct from TdR imaging, where both flux and transport could be estimated over a wider range (21). The differences in model behavior result from differences in model transport (K1) and the retention or trapping (k3) rate, which are both higher for TdR than for the FLT analog.
We examined the relationship between 18F-FLT transport and metabolic flux using parametric image maps, which removes operator dependency in defining the ROI for kinetic analysis and allows visualization of the distribution of parameter values. Parametric image creation using mixture analysis produced regional maps of FLT kinetic parameters that were well correlated with time–activity curve parameter estimates of K1 and KFLT (Fig. 5) and coincided with areas of CE on MRI T1+Gd images. As expected, patients without tumor enhancement on T1+Gd images showed uptake of FLT similar to the low levels observed in normal brain, which lends credence to the interpretation of K1 as transport across the BBB. Additionally, parametric image maps of overall flux and transport visually coincided with the extent of Gd contrast enhancement, suggesting the large influence of transport on FLT distribution in the brain.
We observed both qualitative and quantitative differences in results for FLT and TdR brain tumor imaging. For TdR, both flux and transport could be estimated over a wider range. The differences in model behavior for FLT versus TdR result from differences in the transport rate (K1) and the retention or trapping rate (k3), which are both higher for TdR than for FLT. FLT may not be transported by the same system as TdR. The saturable active nucleoside transport system for TdR at the BBB has significantly reduced transport for deoxynucleoside analogs with substitutions at the 3' position (24,25). Reports have suggested that the observed concentration gradient across the BBB after injection of 18F-FLT involves an active efflux transporter pumping FLT out from the brain (13). It is interesting to note that when this barrier has been disrupted as in CE gliomas, TdR and FLT have similar transport values (Table 7). The average metabolic flux of FLT in this study relative to the average flux of TdR reported previously (4) (KFLT/KTdR; Table 7) for high-grade gliomas (0.56) and C/L brain (0.23) was in agreement with the TK1 phosphorylation ratio (PR) for FLT relative to TdR (PR = 0.3) (26,27). It was also comparable with the relative incorporation rate of FLT to TdR for cultured glioma cells in vitro (0.64) (32). These biologic factors underlie observed differences in transport and retention of these proliferation tracers.
One potential explanation for low transport and retention of 18F-FLT reported recently is competition with high levels of endogenous TdR, thus lowering the uptake of tracers such as FLT and TdR that use the exogenous pathway (33). Cellular assay studies on endogenous and exogenous TdR use reported that both pathways are used to a similar extent in tumor and normal cell lines (34), suggesting that reliance on the endogenous pathway does not restrict access to the exogenous pathway.
Low uptake of 18F-FLT could be due to predominant reliance on the de novo—versus the salvage—TdR pathways for incorporation into DNA. In a series of primary glioma and brain tissue specimens, Bardot et al. (35) observed a shift to the de novo pathway through a reduction in the ratio of TK to TdR synthase (TS) but that higher TK/TS ratios measured between normal brain and low-grade gliomas were statistically identical. This suggests that low 18F-FLT transport and retention in low-grade gliomas are not due to a predominant de novo synthesis of pyrimidines.
The simulation results show significant error in the estimation of the shape parameters (k2, k3, and k4). A recent study on 18F-FLT in brain tumors (36) reported values of k3 for brain and gliomas, similar to ours, and concluded that the data do not support the hypothesis that estimation of glioma proliferation by 18F-FLT is accurate. Model simulations suggest that estimates of this parameter do not possess the precision required with small numbers of patients to evaluate this relationship. The model estimation error for k3 was approximately 50% for high-resolution simulations (2% COV at 60 min). Brain or NCE tumor tissue activity curves possess larger errors (7% COV at 60 min), which most likely would result in a much greater error in estimating k3. Considering the variability in the estimation process, it is not surprising that k3 in tumor could not be differentiated from C/L brain.
An assessment of 18F-FLT uptake as an indicator of cell proliferation requires an independent measure of growth—for example, the determination of histopathologic proliferation markers such as Ki-67. A number of reports have found high correlations of Ki-67 with several measures of 18F-FLT uptake (reviewed by Mankoff et al. (2)). Individual rate parameters such as k3 are less robust than overall flux, which correlates to a modest degree (Spearman 
= 0.70, n = 12) with pathologic grade. However, it is well known that the degree of BBB disruption tends to be higher in more proliferative tumors; therefore, this correlation may arise on the basis of transport.
|
CONCLUSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONReferences |
|---|
Normal brain and NCE tumors with an intact BBB have very limited transport and cannot be adequately assessed for cellular proliferation by 18F-FLT. In addition, problems interpreting images are encountered for patients with high K1 and low KFLT, as might be encountered in radionecrosis. 18F-FLT may not be useful for NCE gliomas regardless of histopathologic grading or proliferation state. 18F-FLT PET may also have difficulty in differentiating residual proliferating brain tumor from BBB breakdown in regions that are not highly proliferating. 18F-FLT brain imaging might have potential use in managing gliomas with initially high K1 and high KFLT for evaluation of early response to new therapies. In that case, it is likely that an early posttreatment decline in KFLT would be observed before a change in K1.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONReferences |
|---|
Related articles in JNM:
This article has been cited by other articles:
|
|
 |
|
  J. R. Bading and A. F. Shields Imaging of Cell Proliferation: Status and Prospects J. Nucl. Med., June 1, 2008; 49(Suppl_2): 64S - 80S. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. Ullrich, H. Backes, H. Li, L. Kracht, H. Miletic, K. Kesper, B. Neumaier, W.-D. Heiss, K. Wienhard, and A. H. Jacobs Glioma Proliferation as Assessed by 3'-Fluoro-3'-Deoxy-L-Thymidine Positron Emission Tomography in Patients with Newly Diagnosed High-Grade Glioma Clin. Cancer Res., April 1, 2008; 14(7): 2049 - 2055. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. S. Bradbury, D. Hambardzumyan, P. B. Zanzonico, J. Schwartz, S. Cai, E. M. Burnazi, V. Longo, S. M. Larson, and E. C. Holland Dynamic Small-Animal PET Imaging of Tumor Proliferation with 3'-Deoxy-3'-18F-Fluorothymidine in a Genetically Engineered Mouse Model of High-Grade Gliomas J. Nucl. Med., March 1, 2008; 49(3): 422 - 429. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. A. Mankoff, J. F. Eary, J. M. Link, M. Muzi, J. G. Rajendran, A. M. Spence, and K. A. Krohn Tumor-Specific Positron Emission Tomography Imaging in Patients: [18F] Fluorodeoxyglucose and Beyond Clin. Cancer Res., June 15, 2007; 13(12): 3460 - 3469. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY | THE JOURNAL OF NUCLEAR MEDICINE |
, Low-grade gliomas; 
, high-grade tumors. (C) In plot of flux vs. transport parameter values (n = 12), correlation is quite linear for tumor, but well under the line of identity, which may indicate effect of limited access of 18F-FLT to brain and tumor regions. (D) Same data as plotted in C with an expanded scale near the origin.