Kinetic Modeling of 5-Fluorouracil Anabolism in Colorectal Adenocarcinoma: A Positron Emission Tomography Study in Rats¹

Fluoropyrimidines, particularly 5-FU³ and prodrugs of 5-FU, continue to be among the most effective agents against metastatic gastrointestinal, mammary, head/neck, and cervical carcinomas (1, 2). With the ultimate goal of optimizing treatment regimens for individual patients, a great deal of research has been directed at understanding and overcoming tumor resistance to 5-FU (1, 2, 3, 4), as well as predicting tumor response to 5-FU-based treatment (2, 5).

The metabolism (Fig. 1) and cytotoxic mechanisms of 5-FU are well understood. Fluorouracil inhibits cell proliferation by: (a) forming FdUMP, which in turn blocks TS, the enzyme that catalyzes de novo synthesis of the DNA precursor thymidylate (i.e., TMP); (b) forming defective, F-RNA, which ultimately interferes with protein synthesis; and (c) forming defective, fluorinated DNA, which results in single-strand breaks and DNA fragmentation (1, 6, 7). Of these mechanisms, TS inhibition is generally considered the most important for continuous-infusion regimens, whereas F-RNA formation may dominate in high-dose bolus treatments (6). 5-FU is toxic only when taken up by cells and anabolized to fluoronucleotides, which, in turn, may be incorporated into nucleic acids or bind to TS. The bioavailability of 5-FU is greatly limited by rapid catabolism in the blood, liver, and other organs. After i.v. injection in humans, the drug has a half-life in blood of only 8–20 min (8).

Because it is rapidly catabolized in the intestine and liver, 5-FU is usually given i.v. Efforts to develop reliable oral delivery of 5-FU-based therapy have focused mainly on suppressing DPD (9), the enzyme that catalyzes the first step in the catabolism of 5-FU and other nucleobases (Fig. 1). One effective suppressor of 5-FU catabolism is 5-ethynyluracil (eniluracil), a potent, mechanism-based, irreversible inactivator of DPD (10). Chemically, eniluracil differs from 5-FU by replacement of the fluorine atom with an ethynyl group (C≡C-H) at the 5 position of the pyrimidine ring.

Tumor responsiveness to 5-FU-based treatment is low (on the order of 20% for 5-FU as a single agent; ≤50% in combination with certain other drugs; Ref. 1), implying a need to screen candidates based on likelihood of positive response. Direct measurement of gene expression levels for TS and other enzymes affecting tumor response to 5-FU with an ultra-sensitive, cDNA amplification technique (reverse transcription-PCR) has been shown to accurately predict tumor nonresponse to 5-FU (2, 5). However, the invasiveness of the biopsy procedure generally limits application of direct tissue assays to a single usage in a single tumor per patient. Thus, there is a need to develop additional, more comprehensive, and less invasive methods for predicting tumor response.

It is possible to monitor 5-FU pharmacokinetics noninvasively using either MRS/5-[¹⁹F]FU or PET/5-[¹⁸F]FU. MRS/5-[¹⁹F]FU has been used to measure the relative concentration of 5-FU versus time in tumors near the surface of the body after high-dose, bolus administration of the drug to human patients (11). Tumors with 5-FU retention half-times <20 min as determined by MRS consistently fail to respond to 5-FU. Because of its very high sensitivity to the radiolabel, PET/5-[¹⁸F]FU has some important advantages over MRS/5-[¹⁹F]FU, including optional use of subpharmacologic doses of 5-FU, absolute quantitation with high temporal resolution, and three-dimensional imaging at any location within the body. Furthermore, it may be possible to monitor tumor anabolism of 5-FU with PET/5-[¹⁸F]FU, which in general has not been achievable with MRS/5-[¹⁹F]FU.

Unfortunately, the pharmacokinetics of 5-FU are unfavorable for imaging studies with 5-[¹⁸F]FU. Tumor uptake of ¹⁸F is often low relative to normal tissues (12), which hinders both tumor visualization and accurate measurement of tumor radioactivity. Rapid destruction of injected 5-[¹⁸F]FU gives rise to high levels of recirculating, labeled catabolites. This reduces tumor-to-normal tissue contrast, obscures image interpretation, and hinders modeling of the radiolabel kinetics (13). We and others have shown previously that tumor visualization as well as the quality of information obtained with PET/5-[¹⁸F]FU about tumor anabolism of the drug can be substantially improved by using a DPD inhibitor such as eniluracil to suppress catabolism of the radiotracer (13, 14).

The objective of the work described here is to explore the potential of PET imaging of 5-[¹⁸F]FU in the presence of an inhibitor of 5-FU catabolism to provide information about the pharmacokinetics of 5-FU in tumors. Specifically, we sought in an in vivo rat model to: (a) determine the ability of compartmental models to describe time-activity data from rat tumors; and (b) relate the “best” resulting compartmental model to specific physiological and metabolic processes that govern the kinetics of 5-FU in those tumors.

Eniluracil (Glaxo Wellcome 776C85) was supplied under special agreement by Glaxo Wellcome, Inc. (Research Triangle Park, NC). 5-[¹⁸F]fluorouracil was prepared according to a published, one-pot synthesis (15). The radiochemical purity of the final product as determined by HPLC was ≥98%, and the specific activity ranged from 15 to 22 GBq/mmol (400–600 mCi/mmol or 3–5 mCi/mg). DTPA labeled with ^99mTc (radiochemical purity >98%) was obtained from Syncor (Van Nuys, CA). RBCs were pretinned and labeled in vitro by mixing rat blood with stannous chloride solution (Syncor) followed by incubation with ^99mTc pertechnetate (16); measured labeling efficiency was >95%. Nonradiolabeled chemicals used as references for identification of radiolabeled chromatographic peaks were purchased either from Sigma (St. Louis, MO; 5-FU, fluorouridine, fluorodeoxyuridine, FdUMP) or as special preparations from Sierra Bioresearch (Tucson, AZ; FUMP, fluorouridine triphosphate, and fluorodeoxyuridine triphosphate). Chemical purities of these reference compounds were all ≥96% except for FdUMP, which was 85% pure.

Female Fischer-344 albino rats (180–210 g) were obtained from Simonsen Laboratories (Gilroy, CA) and implanted with the Ward tumor, a colorectal carcinoma that originated via chemical induction in this strain of rat (17). The Fischer rat/Ward tumor model has been used extensively to study the therapeutic effects of 5-FU as well as a number of different prodrugs and biochemical modulators of 5-FU, including eniluracil (18, 19).

Several small pieces (each ≈1 mm³) of previously frozen tumor were implanted by trochar injection between the skin and gastrocnemius muscle. Implanted tumors reached 0.5 g in 12–20 days, grew in a well-encapsulated manner, and appeared to derive their blood supply from the overlying skin. Macroscopic, central necrosis began to appear when the tumors reached 1 g. Kinetic modeling studies were performed with tumors weighing 0.24–0.84 g (mean, 0.52 g).

All of the in vivo procedures were performed in accordance with a protocol approved by the University of Southern California Animal Care and Use Committee. Anesthesia was induced by i.p. injection of ketamine (80 mg/kg) + xylazine (5 mg/kg) and maintained with smaller doses as needed. The rats were allowed to breathe spontaneously without intubation. Arterial blood gases remained normal for >3 h with this regimen. Vascular access was obtained by surgical cut down and cannulation of one jugular vein and one carotid artery with 24-gauge, i.v. catheters. A special blood sampling technique was developed that, without overly exsanguinating the rats, provided both enough data points to adequately characterize the plasma TAC and sufficient blood to test for circulating metabolites (14). The rats were sacrificed by i.v. injection of pentobarbital.

Imaging studies were performed with an ECAT 953, whole-body PET scanner (CTI/Siemens, Knoxville, TN) having 31 transaxial planes, 43-cm diameter transaxial field of view, 10.8-cm axial field of view, and 6-mm intrinsic resolution in all three dimensions. The anesthetized, tumor-bearing rat was bound to a surgery board made of thin plastic and injected i.p. with eniluracil (1 mg/kg). After catheterization, the rat and surgery board were oriented transaxially and centered within the scanner’s field of view and secured to a specially designed Lucite shelf. With this orientation, the tomograph provided whole-body images of the rats (Fig. 2). After a transmission scan, ¹⁸F marker sources were taped over the tumors and imaged to aid identification of tumors in the 5-[¹⁸F]FU study. The marker sources were then removed and, about 1 h after the eniluracil injection, a dynamic emission scan was started as the rat was injected through the jugular catheter with 70–110 MBq (2–3 mCi) of 5-[¹⁸F]FU (5-FU dose ≤5 mg/kg). Injected active volume, flush volume, and injection rate were 1.0, 1.0, and 0.5 ml/min, respectively. Dynamic imaging continued for 2 h (45 frames: 1 × 30 s, 9 × 10 s, 11 × 30 s, 12 × 2 min, 6 × 5 min, and 6 × 10 min). Arterial blood samples (12 of 0.2 ml each for activity concentration plus 1 of 0.6 ml for activity concentration and metabolite analysis) were drawn at increasing intervals during the course of the imaging study.

At the end of the dynamic imaging procedure, the rat and surgery board were removed from the scanner, and the tumors were rapidly excised and frozen between blocks of dry ice to arrest metabolism. Frozen tumors were transferred to prechilled plastic tubes and kept immersed in liquid nitrogen until they were prepared for metabolite analysis, except for brief intervals during which the samples were weighed and counted to determine total activity concentrations. The rats were killed after tumor excision and freezing.

Tumor extracellular fluid and V_rbcs were measured in groups of rats other than those used for the PET/5-[¹⁸F]FU experiments described above. Rats bearing the Ward tumor were injected i.v. with the extracellular indicator ^99mTc-DTPA (20). After 10 min, a blood sample was taken by percutaneous cardiac puncture, tumors were excised, and the rat was sacrificed. After separation of plasma from RBCs by centrifugation, tumors and plasma samples were weighed and counted for radioactivity. Tumor V_ecf (ml/g tumor) was calculated as the ratio of tumor:plasma activity concentration assuming a plasma density of 1.03 g/ml (21). The procedure for measuring tumor V_rbc was the same as for V_ecf, with the following exceptions: ^99mTc-labeled whole blood was injected, RBC rather than plasma activity concentration was determined (with an approximate correction for plasma trapped in the centrifuged pellet), and V_rbc was calculated as the ratio of tumor:RBC activity concentration assuming an RBC density of 1.09 g/ml (21).

Samples were measured for ¹⁸F or ^99mTc activity in a gamma counter (Packard Instruments, Meriden, CT). The counter was calibrated periodically against the same ion chamber-type dose calibrator (Capintec, Ramsey, NJ) used to assay 5-[¹⁸F]FU injected into the rats.

Blood samples were immediately placed in ice and then centrifuged. An aliquot of plasma was removed from each sample, weighed, and counted for radioactivity. In samples drawn for metabolite analysis, the remaining plasma was prepared for HPLC analysis by acid extraction with perchloric acid and neutralization with potassium hydroxide. Details of the sample preparation and HPLC procedure have been reported previously (22). HPLC eluents were passed through in-line UV (Waters, Milford, MA) and radiation detectors (Magen Scientific Corp., New York, NY). Eluted fractions were also collected and assayed for radioactivity.

The procedure for preparing and assaying tumors for molecular distribution of ¹⁸F has been described in detail (14) and validated (22). Frozen tumors were immersed in liquid nitrogen together with perchloric acid (5 n, 1:1 v/w of frozen tumor) and ground into a fine powder. A portion of the powder was gradually thawed on ice with periodic, rapid mixing. The sample was then diluted with ice-cold, deionized, distilled water, and the acid-soluble supernatant and acid-insoluble pellet were separated by microcentrifugation. The AIF (i.e., the pellet) was rinsed and assayed for radioactivity. The ASF was buffered with potassium monohydrogen phosphate (K₂HPO₄) and then neutralized with potassium hydroxide. The resulting perchlorate salt was removed by centrifugation, and the desalted ASF was analyzed by ion-pair HPLC and multistep, linear-gradient elution as described previously.

Radiolabeled peaks were identified in reference to UV absorption chromatograms of standard solutions containing authentic, nonradiolabeled counterparts of the various metabolites expected in the experimental samples. The fraction of recovered activity associated with a given molecular species was multiplied by the percentage of total activity in the ASF to determine the overall percentage of that molecular species in the original experimental sample. Because they were not well separated in some runs, ribonucleic and deoxyribonucleic species were not differentiated in the quantitative analysis of the chromatograms.

Raw data from the PET studies were corrected for detector nonuniformity, random coincidence noise, scanner dead time, photon attenuation, and radioactive decay. Images were reconstructed (image matrix 128 × 128, pixel width 1.7 mm, voxel volume 9.8 mm³) by filtered backprojection using a Hann filter with a cutoff frequency of 0.5 cycles/pixel. (We also tried a ramp filter with cutoff frequency = 0.5 cycles/pixel. Although the ramp filter provided modestly better image resolution than the Hann filter, the images were also noisier, and this proved detrimental in fitting models to tumor TACs.)

Tumor ROIs were defined by comparing the marker images obtained before injection of 5-[¹⁸F]FU with the images obtained after injection of the radiotracer. Invariably, the latter images showed spherical or elliptical regions of elevated activity density centered within a few voxels of the centers of the marker images. TACs were obtained from small (30–70 mm³), circular ROIs centered well within the boundaries of visualized tumors to minimize count spillover from adjacent tissues (i.e., muscle, bone, and skin). Potential count spillover from bladder, which lay within 4 cm of the tumors and always contained a large amount of activity by 2 h, was shown to be negligible. The spatial distribution of counts from the bladder was estimated by imaging small ¹⁸F sources placed on the dorsal and ventral surfaces directly above and below the bladder of a rat in standard scanning geometry. The resulting count profile in combination with the relative count rates from bladder and tumors observed in the PET/5-[¹⁸F]FU studies implied that bladder contributions to tumor counts were generally about 1% and never >8% in our studies. To avoid potential errors in measured tumor activity concentration because of the limited spatial resolution of the images (partial volume effects), the TACs were normalized at 2 h to activity concentrations measured directly in excised tumors. Time-activity data were normalized to injected activity/rat body weight, and thus expressed in terms of SUVs.

Both Patlak and compartmental analysis were applied to tumor TACs. Matlab, v. 5.1 (The Math Works, Natick, MA) was used to perform the Patlak transformation (23), whereas the numerical module of SAAM II, v. 1.1.1 (SAAM Institute, University of Washington, Seattle, WA) was used for linear regression analysis of the postequilibration phase of the transformed data. Compartmental models were fitted to experimental TACs by nonlinear, least-squares regression using the compartmental analysis module of SAAM II (24). Starting values for parameters were varied to verify that solutions corresponded to global rather than local minima in the parameter space. An objective goodness-of-fit criterion, the AIC (25), was used to compare different models. The AIC is defined as:

where the sum of squares of the residual errors RE is given by:

In these equations, N is the number of measurements, p is the number of estimated model parameters, Y_i is the model estimate of the TAC at the time of the i_th data point, y_i is the i_th observed data point, and w_i is the weight for the i_th data point. Note that AIC decreases as the fit improves. We found that weighting based on counting statistics produced results that were negligibly different from uniform weighting. The model fits reported herein were obtained with uniform weighting.

The general form of the compartmental model used in this study is diagrammed in Fig. 3. The sequence of 5-FU anabolism (Fig. 1), the known inability of nucleotides and nucleic acids to cross the cell membrane (4), and the slow turnover of the macromolecular pools (26) relative to the duration of the imaging procedure suggested that a catenary compartmental model, i.e., a linear sequence of compartments, with increasingly slower turnover rates might be useful in describing tumor incorporation of 5-[¹⁸F]FU. Patlak analysis consistently indicated that some of the radiolabel is effectively trapped in tumors within the 2-h duration of the experiments (Fig. 4). This implies that the last compartment in the model should be expected to have input, but no output.

The differential equations corresponding to the diagram of Fig. 3 are as follows:

where q_i(t) is the amount of activity (Bq/g of tissue) in compartment i at time t, K₁ is the clearance (ml/g of tissue/min) of radiotracer from the arterial plasma into the tumor ROI, C_p(t) (Bq/ml) is the activity concentration in arterial plasma at time t and the rate coefficient k_ij (min⁻¹) is the fractional content of compartment j that is transferred per unit time to compartment i. The rate coefficient k₀₁ corresponds to loss of activity from the tumor ROI.

The model does not account for activity in RBCs. We simulated the RBC component of the tumor TACs based on our measurements of plasma TACs, hematocrit, and tumor V_rbc (14), together with a published model of 5-FU exchange between plasma and RBCs (27). RBC-borne ¹⁸F was estimated to comprise <3% of tumor TAC amplitude at any time. Inclusion of an additive term of the form V_rbcC_rbc, where C_rbc = calculated activity concentration in RBCs, had negligible (i.e., <5%) effect on the fitted model parameters as well as clearances and distribution volumes (Eqs.F–K) derived therefrom.

Compartmental models were fitted to integral activity per frame versus time rather than to tumor TACs per se. This was done to avoid potential errors resulting from the conventional, arbitrary use of frame midtimes as the time values for PET-derived TACs. We transformed the tumor TACs into integral activities per frame by multiplying each activity value by the corresponding frame duration and assigned the integral activities to the end times of the dynamic frames.

For presentation, sets of parameters obtained by fitting models to integral data were used to compute corresponding instantaneous TACs, and the resulting model TACs were plotted together with the measured TAC data (e.g., as in Fig. 5). Plasma TACs were linearly interpolated to frame end times for compartmental fits and to frame mid times for computation of model TACs. For Patlak analysis, the difference in results obtained with integral and instantaneous TACs was found to be negligible, and TAC format was used to obtain all of the Patlak results presented herein.

For compartmental models, one factor observed to have significant effect on convergence of the fitting algorithm was noise in the arterial plasma input functions. Patlak analysis (23) is very sensitive to such noise, and provides an effective means of differentiating between noise and true fluctuations in the plasma TACs. In cases where the Patlak plot indicated noise in the plasma data, we consistently found that model convergence was improved by replacing the plasma TAC beyond its early peak with a decaying biexponential function fitted to that portion of the data.

Various physiologically significant, “derived” parameters are defined below for the five-parameter compartmental model in terms of the model parameters (28).

K₂ and K₃ are, respectively, the net clearances of radiotracer from arterial plasma into compartments 2 and 3. V₁ is the distribution volume of compartment 1, i.e., at steady state during constant infusion at an arterial activity concentration C_p and with k₂₁ = 0, the amount of activity in compartment 1 would equal V₁C_p. Similarly, V₂ is the distribution volume of compartment 2, and the steady state activity content of compartment 2 would equal V₂C_p in the absence of compartment 3. V₁(eq) and V₂(eq) each include the effects of the other two compartments; their products with C_p represent the actual steady state contents of compartments 1 and 2 during constant infusion.

An important part of the modeling process was to relate the compartments to physiological spaces and/or labeled molecular species. The estimated distribution volume of model compartment 1 was compared with the measured distribution volume of ^99mTc-DTPA in tumor by two-tailed Student’s t test. In relating compartments to molecular species, we expressed both model and molecular activities as SUVs, i.e., decay-corrected activity per g tumor weight divided by injected activity per g rat body weight. Scatter plots and linear regression analysis were used to compare the compartments of fitted models at 2 h after injection to various combinations of directly measured, labeled molecular species (see Fig. 6). Results of statistical tests were considered significant at the 5% level of probability.

PET imaging and tumor metabolite analysis were performed for a total of five tumors in four rats. Tumors were well visualized in the PET images relative to adjacent, normal tissues (Fig. 2). At 2 h, tumor SUV was 1.1 ± 0.2; tumor ratios to plasma, muscle, and bone were 2.5 ± 1.0, 2.3 ± 0.3, and 2.7 ± 0.9, respectively. (Data are mean ± SD, determined by direct assay of excised tissues.)

To detect possible spillover of counts into tumor ROIs from nontumor tissues, the shape of each tumor TAC was compared with that from a larger ROI (80–190 mm³) drawn concentrically about the ROI (30–70 mm³) used to obtain the tumor TAC. The TAC from the larger ROI was normalized to the tumor TAC at 2 h. In all of the cases, the normalized test TAC differed by no more than a few percent from the tumor TAC at any time, which indicated that the shapes of the tumor TACs were not significantly distorted by counts from activity outside the tumors.

Metabolite analysis of arterial plasma showed that 100% of circulating ¹⁸F was on 5-FU for at least the first 90 min after injection (14). Plasma AIF was found to contain negligible activity, implying that all of the plasma-borne 5-FU was free and available for extraction into tissue. Thus, the plasma TAC measured during each study was directly used to determine plasma 5-[¹⁸F]FU concentrations for kinetic analysis.

Analysis of the TAC data revealed that part of the ¹⁸F that entered tumors was effectively trapped there during the 2-h period of observation. For all of the studies, Patlak transformation of the tumor and plasma data showed an equilibration phase followed by linear net uptake of ¹⁸F by tumors (Fig. 4). Consistent with this, attempts to fit the tumor TACs with a six-parameter compartmental model (Fig. 3) produced convergence to a five-parameter model (k₂₃ = 0) for three cases and lack of convergence for the other two studies.

All five of the data sets were successfully fitted with a five-parameter model. For all of the studies, objective goodness-of-fit worsened (i.e., AIC increased) progressively as the model was reduced to two compartments and four parameters (k₃₂ = 0), and then two compartments and three parameters (k₃₂ and k₁₂ = 0; Fig. 5). One set of parameter starting values was used for all of the data sets. Identifiability was verified for the five-parameter model by showing convergence to the same solution with alternative sets of parameter starting values. The quality of the five-parameter fits was very good (Fig. 5,C). The magnitude of residual errors was small in all of the cases (<2% near and beyond the initial maxima of the TACs). Fitted parameters are summarized in Table 1. Variability of the primary compartmental model parameters among different tumors was relatively small (coefficients of variation <20%). A high degree of correlation was observed between K₁ and k₀₁, k₀₁ and k₁₂, and k₁₂ and k₃₂ (mean correlation coefficients = 0.97, 0.90, and 0.80, respectively). Correlation coefficients for all of the other parameter pairs were <0.80.

There was good agreement between Patlak and compartmental model analysis (Table 1). The Patlak slope or clearance parameter K agrees very closely with K₃, the clearance from plasma into compartment 3 of the 5-parameter model. V_d, the distribution volume of the exchanging spaces in the Patlak analysis, is nearly the same as V₁(eq) + V₂(eq), the combined equilibrium distribution volumes for compartments 1 and 2.

Having established that the five-parameter compartmental model of Fig. 3 explains the observed tumor TACs, we next sought to interpret the model in terms of the kinetics of 5-FU. The parameter K₁ is generally taken to represent the product of flow and single-pass extraction fraction from blood into tissue (28). Given the known leakiness of tumor capillaries (29), tumor single-pass extraction fraction for small molecules like 5-FU can be expected to approach unity. Consistent with this, the values obtained in this study for K₁ (0.06 − 0.11 ml/min/g) are typical of plasma flow in implanted rodent tumors (30). Furthermore, the distribution volume of compartment 1 (V₁ = K₁/k₀₁) in our studies is in the range of 0.3–0.4 ml/g, a very reasonable result for tumors if compartment 1 reflects the combined plasma and interstitial spaces (31, 32). We also found tumor V_ecf as measured with ^99mTc-DTPA to be in good quantitative agreement with the model-derived V₁ (Table 1). Therefore, we conclude that compartment 1 represents the combined plasma and interstitial spaces (i.e., the extracellular space) of the tumors. This suggests that compartments 2 and 3 are associated with the tumor intracellular space and that rate coefficients k₂₁ and k₁₂ reflect transport of radiolabel across tumor cell membranes.

We next sought to identify compartments 2 and 3 in terms of 5-FU metabolism. Results of our measurements of ¹⁸F molecular distribution in tumors have been reported in detail (14, 22). Correlations between labeled molecules and the five-parameter compartmental model are shown in Fig. 6. The amplitude (i.e., SUV) of compartment 3 at 2 h was found to be nearly identical to the combined SUV for FN3Ps and macromolecules in tumor tissue for all five of the tumors (Fig. 6,A). Thus, given the well-established metabolic pathways for 5-FU (Fig. 1), the data clearly show the observed “trap” (compartment 3) to be associated with conversion of FN2Ps to FN3Ps.

Fig. 6 B shows that compartment 1 was linearly correlated with tumor 5-FU. Nucleosides comprised only a few percent of tumor activity. Nucleotides and macromolecules formed from 5-[¹⁸F]FU would have been confined to the intracellular space (4). Thus, the figure indicates that the extracellular space (compartment 1) contained 40–50% of tumor 5-FU.

Given the identifications of compartments 1 and 3, compartment 2 must, by process of elimination, represent intracellular free 5-FU plus FNs, FN1Ps, and FN2Ps. No other combination of metabolites is quantitatively consistent with the model. Additional evidence supporting this conclusion is given in Fig. 6 C, where compartment 2 is shown to be quantitatively consistent with the sum of intracellular 5-FU (= total 5-FU − extracellular 5-FU), FNs, FN1Ps, and FN2Ps.

Noninvasive measurement of tumor incorporation and retention of 5-FU and its metabolites could be helpful in predicting response to treatment, evaluating alternative methods of drug administration, modulating 5-FU kinetics to improve response, and detecting mechanisms of tumor resistance. Studies with MRS/5-[¹⁹F]FU have established that nonresponsive tumors can be accurately identified in human patients on the basis of short retention half-time for 5-FU after i.v. high-dose bolus administration (11). Tumor retention of radiolabel from 5-[¹⁸F]FU as measured by quantitative PET may also be a useful predictor of tumor response to 5-FU. Dimitrakopoulou-Strauss et al. (33) reported a highly positive correlation between tumor retention of radiolabel 2 h after i.v. or hepatic intra-arterial injection of 5-[¹⁸F]FU and change in tumor size during subsequent treatment by continuous i.v. or intra-arterial infusion of 5-FU. The same group has also used PET/5-[¹⁸F]FU to compare systemic i.v. with hepatic intra-arterial administration of 5-FU in patients with intrahepatic metastases (12). They concluded that the pharmacokinetics of 5-FU after intra-arterial administration could not be reliably predicted from an i.v. administration to the same patient. Investigators at Hammersmith Hospital, London, have used PET/5-[¹⁸F]FU to evaluate the effects of the DPD inhibitor eniluracil on ¹⁸F activity versus time in blood, liver, kidneys, and tumors in patients with gastrointestinal cancers (34). Eniluracil was found to effectively eliminate catabolism of 5-FU, resulting in prolonged appearance of circulating 5-FU, decreased retention of 5-FU and its metabolites in liver and kidney, and increased uptake and retention of 5-FU in tumors. In another PET/5-[¹⁸F]FU study of patients with gastrointestinal cancer, the Hammersmith group examined the effects of N-phosphonoacetyl-l-aspartate, IFN-α, and folinic acid on tumor and liver uptake and retention of ¹⁸F (35). N-Phosphonoacetyl-l-aspartate was found to increase tumor blood flow and radioactivity exposure, IFN-α altered tumor uptake of ¹⁸F, and folinic acid had no measurable effect in either tumor or liver. Pretreatment with the antifolate methotrexate is known to enhance conversion of 5-FU to FUMP (4). MRS has been used to demonstrate increased tumor retention of free 5-FU after pretreatment with high-dose methotrexate in human patients (36).

Arguably, the greatest utility of noninvasively monitoring 5-FU could be achieved by using kinetic modeling of dynamic data to determine rates of 5-FU uptake and metabolic conversion in tumors and normal tissues. Studies of this type might be especially useful in identifying specific mechanisms of tumor resistance and overcoming them by modulation of 5-FU kinetics. The ability of MRS to support dynamic studies is severely limited by difficulty of absolute quantitation and low sensitivity (achievable temporal resolution is on the order of several minutes; Ref. 11). PET, on the other hand, provides accurate absolute quantitation and temporal resolution on the order of 10 s, but is fundamentally limited by its inability to directly identify molecular association of the radiolabel.

A few kinetic modeling studies of 5-FU have been reported. El-Tahtawy and Wolf (37) used MRS to measure relative concentrations of α-fluoro-β-alanine, free 5-FU, and combined FNs and fluoronucleotides over a 2-h period in s.c. implanted rat tumors. Using a two-compartment model, they were able to quantify the increase in fractional conversion rate of 5-FU to FNs plus fluoronucleotides resulting from pretreatment with methotrexate. Kissel et al. (27) used a 6-compartment model (formulated from a 12-compartment model) of systemic 5-FU kinetics to estimate 5-FU TACs in plasma during and after i.v. infusion of 5-[¹⁸F]FU in patients with colon cancer metastatic to liver. The plasma TACs, in turn, were used as input functions for a two-compartment model that was fitted to tumor TACs measured with PET. (Radioactivity in tumor vascular plus interstitial spaces was modeled as the product of plasma activity concentration and a volume parameter.) The analysis was based on a number of assumptions, including confinement of catabolites of 5-[¹⁸F]FU to the vascular space within tumors, which is physiologically unlikely and, in fact, has been disproved (13, 14).

Aboagye et al. (13) performed PET/5-[¹⁸F]FU with and without pretreatment with eniluracil in patients with intrahepatic metastases or pancreatic tumors. Spectral analysis and deconvolution were applied to data from the eniluracil experiments to determine tumor response functions after exposure to a unit impulse of 5-FU. The tumor response functions were in turn convolved with measured 5-FU plasma TACs to determine the 5-FU plus anabolite components of tumor TACs measured in the experiments without eniluracil. The analysis showed that, without suppression of 5-FU catabolism, >80% of tumor TAC area resulted from labeled catabolites. When subjected to Patlak transformation (23), tumor TACs from the eniluracil experiments exhibited linear postequilibration uptake consistent with a component of irreversible uptake (i.e., “trapping”) within the 90-min period of data acquisition. No attempt was made to identify a mechanism for the observed trapping phenomenon.

Our study is the first in which modeling of ¹⁸F tissue TACs has been compared with independent, “gold-standard” methods to validate a physiological and metabolic interpretation of 5-FU uptake and retention. The objective was to define, within the context of a particular experimental tumor and observational period, the extent to which the kinetics of 5-FU uptake and anabolism in tumors (i.e., the part of 5-FU kinetics that directly affects tumor response) can be determined with PET/5-[¹⁸F]FU and kinetic analysis. The results of the study were unequivocal. They showed that the best fit (based on AIC) to the TAC data were obtained with a three-compartment, five-parameter model. Compartmental and Patlak analysis were consistent in identifying a terminal, trapping process for ¹⁸F. Comparison with biochemical analysis of tumor tissue identified the trapped molecular species as FN3Ps plus macromolecules (mostly F-RNA; Ref. 14). Given the known metabolic pathways of 5-FU (Fig. 1) and the smallness of TS concentrations in tissue (14), it follows that the trapping step is the conversion of FN2Ps to FN3Ps. Direct measurements of tumor extracellular space using ^99mTc-DTPA agreed with V₁, thus indicating that compartment 1 corresponds to the tumor extracellular space. Identification of compartments 1 and 3 implied that compartment 2 represents intracellular free 5-FU, plus FNs, FN1Ps, and FN2Ps, an interpretation that was quantitatively consistent with the model estimate of V₁(eq) combined with the observed molecular distribution of radiolabel in tumor (Fig. 6 C).

The kinetic analysis and model interpretation yielded a large amount of information about tumor physiology and 5-FU kinetics in tumors, much of which can be compared with pre-existing information. Values obtained for model parameters K₁ and V₁ are typical, respectively, of plasma flow and extracellular distribution volumes for animal tumors (30, 31, 32), which reinforces identification of compartment 1 with tumor extracellular space. Model analysis showed fractional extraction of 5-FU from the extra- to intracellular space (= K₂/K₁, the ratio of net plasma-to-intracellular and plasma-to-extracellular clearances) to be ≈1/3 (Table 1), thus identifying cell transport as a significant barrier to cellular uptake of 5-FU. Modeling also indicated that conversion of FN2Ps to FN3Ps was slow compared with the turnover rate of compartment 2, because the ratio of net plasma-to-trap and plasma-to-intracellular clearances, K₃/K₂, was only about 0.1.

The identification of a compartment corresponding to the tumor extracellular space implies that egress of ¹⁸F from tumor was markedly delayed for those 5-FU molecules that entered the intracellular space. This is consistent with observations made with MRS/5-[¹⁹F]FU that some tumors have the ability to retain 5-FU for extended periods of time at concentrations well above that of plasma (11). The mixed molecular composition of compartment 2 indicates that intracellular pools of ¹⁸F-labeled 5-FU, FNs, FN1Ps, and FN2Ps rapidly equilibrated among themselves compared with efflux of 5-FU and FNs from cells. (Nucleotides are generally thought not to cross cell membranes.) The data suggest that membrane transport provided a mechanism for cellular retention of 5-FU, i.e., clearance from compartment 1 to compartment 2 (= k₂₁V₁) exceeded clearance from compartment 2 to compartment 1 by approximately a factor of two. [Clearance from compartment 2 to compartment 1 was estimated as k₁₂V₂ × the fraction of compartment 2 comprised of 5-FU and FNs. The latter factor was estimated from the measured molecular distribution of ¹⁸F in tumor (14) and the observation that approximately half of the 5-[¹⁸F]FU in tumor was intracellular (Fig. 6 B)]. Asymmetrical, concentrative transport of 5-FU by tumor cells is known to occur both by active transport (38) and in the presence of a positive extra- to intracellular pH gradient (39).

The current study agrees very closely with the study in humans by Aboagye et al. (13) with regard to plasma-to-trap clearance of ¹⁸F in tumors. Our values, as determined by Patlak and compartmental analysis, were (3.5 ± 0.3) × 10⁻³ and (3.8 ± 0.3) × 10⁻³ ml plasma/g tissue/min, respectively. Aboagye et al. (13) reported a value of (3.6 ± 0.5) × 10⁻³ ml plasma/cm³ tissue/min, which was obtained by Patlak analysis of PET data from liver metastases and pancreatic tumors. This suggests that 5-FU kinetics in the rat colon tumor used in our study are similar to those in human tumors and, specifically, that the trapping event in the human study by Aboagye et al. (13) was also conversion of FN2Ps to FN3Ps.

Comparison of our study and that by Aboagye et al. (13) with other reports on PET/5-[¹⁸F]FU demonstrates that suppression of DPD activity greatly enhances the information that can be derived about 5-FU by kinetic analysis of tumor TACs. However, the relevance of this “PET/5-[¹⁸F]FU+catabolism suppression” technique depends on its ability to predict the kinetics of 5-FU uptake and anabolism in tumors during treatment. Clearly, the technique is applicable for treatments that use eniluracil or other CS agents. Additional work will be required to determine to what extent the PET/5-[¹⁸F]FU+CS technique is useful with 5-FU-based therapies that do not use DPD suppression (14, 40). One issue is that 5-FU is often given as a high-dose bolus. Because suppression of catabolism preserves 5-FU, only very low doses of 5-FU are permissible with the PET/5-[¹⁸F]FU+CS technique. Thus, unless the kinetics of 5-FU in tumors remain linear over the concentration range encountered after high-dose bolus injection, the information provided by the PET technique may not be particularly useful with high-dose regimens. Another consideration is that the PET/5-[¹⁸F]FU+CS technique is necessarily insensitive to DPD-mediated tumor resistance. Thus, if used to predict tumor response to 5-FU, the technique should be supplemented by a test that can detect tumor DPD, such as biopsy followed by reverse transcription-PCR assay.

Clinical application of the five-parameter compartmental model derived in this study might be limited by noise in tumor TACs. Because of much lower injected activity/body weight, and perhaps patient motion as well, data from human studies are often much noisier than in our rat experiments. Such data cannot be expected to support determination of five model parameters. On the other hand, the noise problem can be mitigated in several ways. It generally would be possible in human studies to improve counting statistics by using larger tumor ROIs than used here. Three-dimensional imaging with modern PET scanners is much more sensitive than the two-dimensional imaging used in the current study, and this should translate into reduced TAC noise. Image noise can also be reduced by using iterative reconstruction methods rather than filtered back projection, which was the only method of reconstruction available to us in the current study.

Investigators should be cautious in analyzing tumor TACs from 5-[¹⁸F]FU with simpler compartmental models than the three-compartment, five-parameter model presented here. We observed, for example, that the compartmental TACs obtained by fitting two-compartment, three-parameter and two-compartment, four-parameter models to our data did not correspond consistently to any distinct groupings of labeled metabolites or recognizable physiological spaces (data not shown). This suggests that interpretability is lost with an oversimplified model, i.e., one that fails to capture all of the information available in the true, noise-free TACs. To the extent that the Ward rat colon tumor is representative with regard to 5-FU kinetics in tumors, analysis with compartmental models simpler than that of Fig. 3 may be misleading. Comparison with Patlak analysis may be useful in determining whether a two-compartment, three-parameter model is oversimplified. In our study, that model overestimated plasma-to-trap clearance relative to Patlak analysis by about 50% (data not shown), whereas the three-compartment, five-parameter model agreed with Patlak within 10% (Table 1).

In summary, we found that a catenary, three-compartment, five-parameter model consistently provided high-quality fits to low-noise, 0–2 h TACs for 5-[¹⁸F]FU obtained in vivo from s.c.-growing rat colon tumors in the presence of eniluracil. The compartments of this model were shown by comparison with independent, invasive tissue assays to correspond to: (a) tumor extracellular space; (b) intracellular free 5-FU + FNs + FN1Ps + FN2Ps; and (c) nonexchanging intracellular FN3Ps + macromolecules. Estimates of plasma-to-trap clearance and exchanging volume of distribution obtained with the model agreed closely with corresponding quantities calculated by model-independent Patlak analysis. The three-compartment, five-parameter model provided better objective fits than two-compartment, three- or four-parameter models, and appeared to be degenerate with a catenary, three-compartment, six-parameter model. The results may be indicative of the full range of information that can be obtained from a 2-h PET study about the kinetics of 5-FU in tumors. This includes distribution volume of the extracellular space, cell transport, size and turnover rate of an intermediate intracellular pool, and formation of a long-lived intracellular pool comprising FN3Ps + macromolecules. Such information might prove useful in predicting tumor response to 5-FU, increasing delivery of 5-FU into tumor cells, and modulating 5-FU kinetics to overcome tumor resistance.

Eniluracil used in this study was a gift from Glaxo Wellcome, Inc., through Dr. Thomas Spector. Glaxo Wellcome also provided partial funding for the purchase and housing of rats used in the project. The Ward tumor was provided by Dr. Yousef Rustum, Sr. Vice President for Scientific Affairs, Roswell Park Cancer Institute, Buffalo, NY.