Abstract
18F-FDG is in widespread use in cancer imaging but has limited utility in staging and monitoring of prostate cancer. 1-11C-Labeled acetate, a substrate for the citric acid cycle, is superior. The kinetics of prostate tumors were investigated. Methods: Ten patients with primary prostate cancer, 10 with recurrent tumor, and 2 men with benign prostate hypertrophy were studied. After administration of 5.5 MBq/kg 1-11C-acetate, dynamic PET of the pelvis was acquired for 20 min. Images were reconstructed with iterative algorithms, and corrections for attenuation and scatter were applied. Factor analysis produced factor images, representing iliac vessels and the prostate from which blood-input and tissue-output functions were derived with simple thresholding techniques. Five different kinetic models were applied to the dynamic data to estimate the rate constants. Results: The standard 3-compartment, 2-tissue model was able to describe 1-11C-acetate kinetics of the prostate. The model could be reduced to 3 parameters by setting the tissue blood fraction and release from the second tissue compartment (k4) to zero. Correction for metabolites appeared to be necessary. This reduced model performed marginally better than a 2-compartment model. A significant correlation was found between the influx rate constant (K) and acetate uptake (standardized uptake value) for primary tumors (r = 0.91), whereas there was no correlation for recurrent tumors (r = −0.17). Patlak graphical analysis provided accurate parameter estimates. Conclusion: A 3-compartment, 3-parameter model is able to describe adequately the acetate kinetics in prostate cancer. Significant differences between primary and recurrent cancer were found for transport k1, influx K, distribution volume Vd, as well as early (6–10 min) and late (15–20 min) 1-11C-acetate uptake.
Prostate cancer is common in the United States and the European Union and is the most frequently diagnosed cancer in men. Metabolic imaging with the glucose analog 18F-FDG is a routine procedure in the imaging work-up and monitoring of cancer. For prostate cancer, 18F-FDG is not suitable because of the physiologic excretion by the kidneys and the high concentration in the genitourinary (GU) tract. Dual-modality imaging with PET/CT would provide a possible solution, but prostate cancer usually has low glucose metabolism, making 18F-FDG an ineffective radiopharmaceutical. Acetate is a substrate for the tricarboxylic acid (TCA) cycle and has <5% excretion into the GU tract. 1-11C-Labeled acetate has been used for flow measurements, myocardial and lipid metabolism, and tumor imaging. Previously, we reported on the feasibility of using 1-11C-acetate for imaging of prostate cancer (1,2) and performed dosimetry (3).
Dimitrakopoulou-Strauss and Strauss have written a concise editorial on the relevant issues of 18F-FDG and 1-11C-acetate in prostate cancer imaging (4). The radiopharmaceutical 1-11C-acetate appears superior to 18F-FDG in staging primary and recurrent prostate cancer (5–7). However, benign prostate hyperplasia (BPH) has similar uptake as malignant neoplasia (4,8), just as 18F-FDG (9). Some of the discrepancies reported in the literature could be attributed to the acquisition protocol, static versus dynamic, and the timing of the images. Kato et al. (8) performed a dynamic acquisition in men with a normal prostate gland and found an age dependence of the acetate uptake. In addition, they found a similar acetate uptake in BPH and prostate cancer.
18F-Labeled choline has also been proposed for prostate cancer (10,11). In a direct comparison of 18F-choline and 1-11C-acetate, Kotzerke et al. (12) demonstrated that both tracers performed nearly identically.
1-11C-Acetate has been used to measure myocardial oxygen consumption, and the biochemical pathways are well known and have been published (13–18). In the current study, the 1-11C-acetate kinetics of prostate cancer were evaluated with image-derived time–activity curves—that is, tracer clearance from the blood and uptake in the prostate (or prostate bed). Kinetic modeling was used to estimate the rate constants between the various compartments (Fig. 1). Multicompartment models were tested as were graphical analysis with Patlak plots (19). Metabolite and partial-volume corrections were applied to estimate 1-11C-acetate transport across the cell membrane and influx into the tissue.
Kinetic model with 3 compartments and 2 tissues for prostate. CO2 concentration in tissue is low because of low intracellular pH. In our simplified model, intracellularly labeled CO2 (dotted gray box) is considered negligible. Blood-to-tissue transport is minimal as indicated by dotted gray arrow. Most of labeled CO2 is transported rapidly to blood (as part of k2). Metabolic pool consists of precursors for lipids, amino acids, fatty acids, and steroids that trap or retain 11C label. Labeled products of tricarboxylic acid (TCA) cycle also contribute to metabolic pool (as part of k3). Increase in TCA cycle metabolism would lead to increase of both k2 and k3.
MATERIALS AND METHODS
Patients
The study population consisted of 22 men: 10 with primary prostate cancer, 10 with a prostate-specific antigen (PSA) relapse after radical prostatectomy (recurrent cancer), and 2 with BPH. The average age was 65 y (range, 48–79 y). The Gleason score in primary cancer was, on average, 7.6 (median, 7.5) and was 7.1 (median, 7.0) for the recurrent cancer group. Serum PSA was, on average, 22.2 mg/mL (range, 4–75 mg/mL) for primary cancer, 12.7 mg/mL (range, 1–68 mg/mL) for recurrent cancer, and 1.4 mg/mL for BPH. The PSA velocity for the group with recurrent cancer (PSA relapse) was, on average, 3.0 mg/mL/mo (median, 0.4 mg/mL/mo; range, 0.1–16.1 mg/mL/mo). All subjects had fasted at least 6 h before imaging.
This study was approved by the local Office for Protection of Research Subjects and the Institutional Review Board (IRB) of the University of California at Los Angeles (UCLA), and all patients signed an informed consent to participate in the imaging study.
Image Acquisition
1-11C-labeled acetate was synthesized according to a previously reported procedure (20,21). A dose of 5.5 MBq/kg (0.15 mCi/kg) 1-11C-acetate was administered intravenously. PET was performed using an ECAT HR or HR+ system (Siemens Medical Solutions, Inc.). A transmission scan of 20-min duration was acquired first, in 2-dimensional (2D) mode. Subsequently, 1-11C-acetate was administered intravenously, and a dynamic acquisition was begun simultaneously in 2D mode. Twenty-eight frames were acquired of 12 × 10, 9 × 20, 5 × 60, and 2 × 300 s duration.
The images were reconstructed with iterative techniques, maximum a-posteriori maximization (MAP) (22,23) for the transmission scan and (ordered-subset maximum likelihood expectation maximization [OSEM] consisting of the following sequence of iterations and subsets: 1–16, 1–12, 1–4, 2–1) (24 and Supplemental Appendix [supplemental materials are available online only at http://jnm.snmjournals.org]) for the emission scan. Corrections for attenuation and scatter were applied. A gaussian kernel with 6-mm full width at half maximum was used as the postreconstruction smoothing filter. To get the same x- and y-pixel size, a zoom of 1.5 was used for the ECAT HR+ and 1.28 was used for the HR image reconstruction. The final volume-set had a matrix size of 128 × 128 and consisted of 47 (ECAT HR) or 63 (ECAT HR+) planes, resulting in a voxel size of 3.375 × 3.375 × 3.125 or 3.375 × 3.375 × 2.425 mm3, respectively.
Processing
Factor analysis (FA) was performed on the volume set of reconstructed images (2). To speed up processing, data were resampled into larger voxels by zooming, plane summation, and rebinning, yielding a nearly isometric dataset of 13.5 × 13.5 × 12.5 mm3 voxels (ECAT HR) or 13.5 × 13.5 × 12.1 mm3 voxels (ECAT HR+) that was used for further processing (Supplemental Appendix). Time–activity-curves were generated for every voxel within the pelvis, and FA with a positivity constraint was applied. For our implementation, 3 factors and their corresponding factor images were generated. Factor 1 represented the arterial structures, factor 2 the prostate, and factor 3 the remainder. Tumor was seen in the second factor image. A volume of interest (VOI) was created by simple thresholding of the corresponding factor images. For the iliac vessels, a threshold of 50% was used, which is usual at our institution, and for the tumors 50% was used. With the created VOI, an image-based time–activity curve was generated from the reconstructed dynamic dataset. The time–activity curve from the iliac vessels served as the blood curve, and the time–activity curve for tumor or prostate served as the tissue curve. After correcting these time–activity curves for partial-volume effects, they were used as input and output functions for kinetic modeling. There were 20 patients with cancer (10 primary and 10 recurrent), and 2 men without cancer (BPH). Thus, 22 time–activity curve pairs were available for analysis.
Kinetic Model
The complete model for myocardial 1-11C-acetate kinetics based on known biochemical pathways was published previously (13,14,25,26). For the prostate, the kinetic model has not been elucidated. Some consider the increased citrate oxidation in the TCA cycle in neoplasms as the cause of enhanced acetate uptake (27), based on the view of the prostate as a citrate-producing gland (28). Others attribute the increased uptake to acetate retention in the prostate due to enhanced lipid synthesis (4), whereas the contribution as precursor (e.g., to amino and fatty acids, steroids) is usually not considered (29). In the current approach, our previously published myocardial model was adapted to the standard 3-compartment, 2-tissue model in oncology (Fig. 1), as is generally used for 18F-FDG (30,31). Briefly, acetate is transported across the cell membrane and used as substrate for various intracellular processes—for example, inside mitochondria for energy metabolism, and in the cytosol for enhanced lipid synthesis, building-blocks for membranes, amino acids, and steroids. The incorporation into lipids, amino acids, and so forth has a turnover in the order of hours. Given our acquisition time of just 20 min, possible pathways are oxidation in the TCA cycle to CO2 and H2O and intracellular pooling for further metabolization.
Kinetic modeling using 3 compartments, as shown in Figure 1, yielded the 4 rate constants k1, k2, k3, and k4 (min−1), from which the influx rate constant K was calculated as k1·k3 (k2 + k3)−1 (min−1). The blood volume fraction in tissue was estimated as a fifth parameter, Vb (Supplemental Appendix). This 3-compartment 5-parameter (3C-5p) model was subsequently simplified by reducing the number of parameters (setting them to 0). This is similar to van den Hoff et al., who validated a simple 3-compartment. 3-parameter (3C-3p) model for the acetate metabolism of the myocardium (16). Further reduction yielded a 2-compartmental model, which estimated 2 rate constants plus the blood volume fraction in tissue Vb. The effects of correction for metabolites and partial volume were investigated. Subsequently, Patlak graphical analysis was performed to estimate the influx rate constant KPatlak and the distribution volume VPatlak.
Contrary to prior studies, no metabolites were measured. Instead, the data of other publications were used. Although the metabolic rate of prostate cancer may be low, significant amounts of CO2 are present in blood from tissues with active oxidative metabolism, which contaminate the input function (11C-CO2; Fig. 1). Both Buck et al. (15) and van den Hoff et al. (32) have shown that the metabolites from acetate in the blood follow a monoexponential function:Eq. 1The main difference between the published formulae concerns the time shift t0 used by van den Hoff et al. and constant a = 0.91 used by Buck et al. Because the metabolites appear after a short delay, the van den Hoff equation was used here—that is, a = 1, T1/2 = 6.69, and t0 = 0.48 min:
Eq. 2
It turned out that for our data, Equation 2 produced somewhat better fits than the Buck equation. At 15 min, the metabolites comprise approximately 80% of the activity in the vascular compartment.
Statistical Analysis
Results are presented as the mean ± 1 SD. Nonparametric tests were used to compare parameters between subgroups because of sample size—that is, the Sign test and the Wilcoxon–Mann–Whitney U test. Linear regression analyses were used to study the correlation between parameters, subgroups, tumor types, and patients. Differences between kinetic models were analyzed with the F test to correct for the different number of parameters involved in the various models (33,34). The Akaike information criterion (AIC) was calculated to directly compare the different models. The AIC is based on the concept of entropy and is a measure of the goodness of fit of an estimated statistical model. In the general form for normally and independently distributed model errors:Eq. 3with Np the number of parameters, Ni the number of observations, and RSS the residual sum of squares between measured and model parameters. This AIC penalizes overfitting of data (35).
RESULTS
FA was successful in all 22 cases. Figure 2A displays typical images of factor 1, representing the vascular structures—for example, the iliac arteries—and Figure 2B shows the corresponding maximum-intensity-projection (MIP) image. Figure 2C shows the factor images of a prostate gland with a primary tumor, and Figure 2D shows the corresponding MIP image. Three factors were established, and the iliac vessels and the prostate were easily identified in factor 1 and factor 2, respectively. In one study, the prostate was not found in factor 2, but in factor 3, which turned out to be a BPH patient with unusually high uptake of acetate in the bowel. The 3 factors accounted for an explained variance of 83% ± 9% (range, 66%–95%). Factor 1 contributed, on average, 67% and factor 2 contributed, on average, 11%. The number of voxels available for the iliac vessels was similar for the 3 subgroups (21 ± 8). The number of tissue voxels (tumor/prostate) was smaller for both primary cancer and BPH (44 ± 47) compared with that for recurrent tumors (157 ± 145), underscoring the generally stronger signal (i.e., higher uptake) of primary cancer and benign hyperplasia.
Parametric images from 65-y-old man with primary prostate cancer. (A) Factor images representing vascular component; right and left iliac arteries are identified in 6 transverse planes. (B) Maximum-intensity-projection image of factor 1 in anterior–posterior view representing iliac arteries. (C) Transverse factor images representing tumor component. Note bilobular involvement. Arrow points to tracer in rectum. (D) Maximum-intensity-projection image of factor 2 in anterior–posterior view showing prostate. Note hot spot, left and superior of prostate, represents metastatic lymph node.
VOIs were created from the factor images and time–activity curves generated; typical examples of uptake curves are shown in Figure 3. One of the patients had significant bladder uptake, and image planes had to be constrained to define the prostate. This patient was suspected of having cystitis, but there was no clinical evidence at the time of the study. Two patients had some tracer in the urethra, but the mask of the thresholding algorithm was sufficient to delineate the prostate gland. One patient had high small-bowel uptake, and approximately half of the patients showed significant uptake in the rectum. In all cases, the bowel was at a sufficient distance from the prostate for accurate delineation. Time–activity curves of the vessels and tissues had the following area-under-the-curve (AUC): 24.6 ± 3.8 for the iliac arteries and 46.6 ± 8.6 for the tumor/prostate.
Time–activity-curves from iliac arteries (squares, solid line), metabolite-corrected input function (open squares, dashed line), and prostate tumor (triangles, dotted line). Uptake is expressed in SUV (standardized uptake value). (A) A 65-y-old man with primary prostate cancer (patient in Fig. 2). (B) A 57-y-old man with recurrent cancer; note lower SUV of tumor. (C) A 49-y-old volunteer with BPH.
1-11C-Acetate Tumor Kinetic Model
Compartmental models were used to estimate the kinetic parameters of the prostate tumors, and the average results are shown in Table 1. As an error estimate, the weighted residual sum of squares (WRSS) was calculated—that is, (fitted − measured value)2 weighted by frame duration:Eq. 4Reduction of the number of parameters involved did not have a dramatic impact on individual parameter estimates. The AIC showed that a 2-compartment model is insufficient to describe the data. The AIC appeared the lowest for the 3C-5p and 3C-3p models using metabolite correction. Reducing the number of parameters in the model from 4 to 3 showed a significant decrease in the AIC for 19 cases (P < 0.01 Sign test) and reduction from 5 to 3 parameters showed a decrease in 16 cases (P < 0.05 Sign test). The tissue blood fraction Vb is small and can easily be ignored as demonstrated in Table 1. A similar reasoning holds for k4. Corrections for partial volume are obviously necessary given the 4- to 7-mm size of the iliac arteries. Partial-volume correction for the blood curve mainly affects k1 and, therefore, the influx rate constant K. As reported previously (2), a recovery coefficient of 0.6 was selected for the vascular time–activity curve. To investigate the partial-volume effects for the output function in more detail, the volume of the tumor/gland was determined by adding all voxels within the VOI. The average volume was 12.5 ± 6.8 mL for primary cancer, 13.9 ± 10.3 mL for recurrent cancer, and 9.4 ± 6.2 mL for BPH. Assuming a sphere, the overall average volume of 12.8 mL translates to a diameter of 29.1 mm, which is >4 times the spatial resolution of our scanners. Therefore, 1.0 was selected as the recovery coefficient for the time–activity curve of tumor/prostate.
Kinetic Modeling of Prostate
Metabolite correction improved the tightness of the fits and reduced the error estimates. Most of the individual error estimates did not differ significantly; Table 1 denotes the exceptions. However, the error estimate for each model was lower with metabolite correction than without metabolite correction (P < 0.001 Sign test). The same is true for the AIC, which appeared consistently and for every patient lower for each specific model with versus without metabolite correction (P < 0.01 Sign test). Moreover, the influx rate constant K has a more realistic value with metabolite correction. Direct comparison of the 3C-3p versus the 2C-3p model revealed significant differences (P < 0.001) for 6 patients with the F test (34); group comparison was significantly different with both the Sign test and the U test (P < 0.01). Reducing the number of parameters in the model to 3 showed a decrease in the AIC for most cases. Thus, it appears that a 3-compartment, 3-parameter model with metabolite and partial-volume correction is adequate and sufficient to describe acetate kinetics in prostate cancer. Patlak graphical analysis produced similar results, and Kpatlak was close to K estimated with nonlinear regression. The distribution volume estimated with the Patlak plot was also within the range of modeling values, and no significant differences were found (Tables 1 and 2).
Three-Compartment, 2-Tissue Model with 3 Parameter Estimates, k1–k3, Influx Rate Constant K, and Distribution Volume Vd
Results of the parameters per subgroup in Figure 4 show a significant difference between primary and recurrent cancer for k1 and K but not for k2 and k3. Kpatlak also turned out to be significantly different for primary versus recurrent cancer. The relation between the influx rate K and the transport rate k1 is linear as shown in Figure 5, although the range for recurrent tumor is much smaller than that for primary cancer.
Average values of kinetic parameters for 3 subgroups: primary and recurrent cancer as well as BPH. Error bar denotes 1 SD. Primary and recurrent prostate cancer show significant differences (*P < 0.01) for k1 and K (Table 2).
Scatter diagram of transport rate constant k1 vs. influx rate constant K, both in min−1. Triangles indicate primary cancer and open boxes indicate recurrent cancer. Lines are results of linear regression analysis, r = 0.82 (black) and r = 0.84 (gray). Slopes of linear fits are not statistically different.
1-11C-Acetate Uptake Patterns
The tissue curve has a steep initial incline (Fig. 3) and approaches a plateau, reaching the maximum around 5 min. Analysis of the individual time–activity curves revealed that the early uptake (SUV-early), from 6 to 10 min, was, on average, similar to the late uptake, from 15 to 20 min (Table 2). None of the early-to-late ratios was significantly different from unity. Subgroup analysis revealed that there was significantly different uptake in primary versus recurrent prostate cancer with SUV-late 3.84 versus 1.91 (P < 0.001), respectively. The range of SUV-late was 1.8–5.5 for primary cancer, 1.5–2.6 for recurrent cancer, and 3.2–3.5 for BPH. The difference between BPH and recurrent cancer was significant (P = 0.02) but primary cancer could not be differentiated from BPH. A linear relationship between influx rate K and SUV-late was found for primary tumors (r = 0.91), but not for recurrent cancer (r = −0.17) (Fig. 6).
Scatter diagram of influx rate constant K in min−1 vs. 1-11C-acetate uptake in prostate (SUV). Black squares indicate primary cancer and open circles indicate recurrent cancer. Lines are results of linear regression analysis, r = 0.91 (squares) and r = −0.17 (circles).
DISCUSSION
In this study, the kinetics of 1-11C-acetate in prostate cancer and BPH were investigated, and the compartment model selected was that which described the data best. Our way of analysis is completely image-based and has minimal operator dependency. The FA takes all voxels within the visualized pelvis into account. The factor images are parametric images, representing structures of interest with similar activity changes as a function of time. The thresholding algorithm will automatically create the VOI in x-, y-, and z-dimensions from the factor images and generate the time–activity curves from the dynamic dataset. Uptake in the bladder posed a problem in 1 study. Although uptake in the rectum and the bowel was seen in most patients, the tomographic images were able to distinguish the structures easily in the spatial domain.
In the past, we have demonstrated that FA can reliably generate an input function for various radiopharmaceuticals and organ systems (1,2,36). These studies have shown that kinetic parameters obtained with an FA-based input function are similar to those obtained with a sampled blood curve, and no statistically significant differences were found for the rate constants obtained with these time–activity curves. The use of FA as an “intrinsic” way to correct for partial-volume effects and methods to calibrate the input function were published elsewhere (37).
Corrections for metabolites and partial volume are necessary to produce an adequate fit between the measured and model-generated time–activity curve. Measuring the metabolites in blood allows one to individualize the correction, which is obviously superior. However, 2 theoretic methods have been published (15,32). We used the equation of van den Hoff et al. (Eq. 2) (32) to correct for the metabolites in blood and to generate a realistic input function. Another method is to “calibrate” or “scale” the metabolite curve with a late blood sample. We have used this technique in previous work (1,2,36,37) but not in the current experiments. It should be noted that the concentration of labeled CO2 is relatively low, as the blood is moving continuously and it is exhaled via the lungs. Assuming that the interindividual blood activity concentrations have a low variation coefficient allows one to use an analytic function for the metabolite correction. The CO2 level in tissue is a factor of 2.5 lower than that in blood because of the lower intracellular pH (38). In our model, the labeled CO2 pool in tissue was considered negligible and was ignored (Fig. 1). More research is needed to provide evidence for these assumptions. Possible variations due to food intake were controlled by imaging individuals after prolonged fasting.
Partial-volume correction was applied by using a recovery coefficient of 0.6 for the blood curve. This choice was based on the iliac artery size of 4–7 mm and on recovery coefficients determined for our PET system based on earlier work (2,39). For the output function we used a recovery coefficient of 1 because the tumors were, on average, >10 mL. With dual-modality scanners (PET/CT), correction for partial-volume effects can be performed on an individual basis. VOI generation may be simplified, by generating them directly from the high-resolution anatomic images, which are fused to the metabolic images. FA will remain important for the generation of the tumor time–activity curve, as it highlights the metabolically active, and presumably proliferating, part of the tumor (29). The thresholding algorithm selected the most active part of the tumor/prostate. Although there were no necrotic zones visible in our patients, it would be unlikely for necrotic tissue to have sufficient acetate uptake to reach half of the tumor maximum. The top half of voxels was selected to reduce fluctuations compared with the maximum SUV from 1 voxel—that is, leading to more reliable fitting and parameter estimation. Parameters derived from this part of the tumor turned out to correlate with the type of the neoplasm—that is, primary versus recurrent (Fig. 6). Further studies will be necessary to confirm these observations in larger patient groups.
Prostate Cancer Kinetics and Biology
A 3-compartment model with 3 parameters—that is, setting k4 to 0 and neglecting the tissue blood volume term—with metabolite correction was sufficient to describe the acetate kinetics. This simplified model had overall the lowest error (WRSS) and AIC. In other words, when increasing the parameters from 3 to 4 or 5, there is no improvement in the AIC. Directly comparing the 3C-3p and 3C-5p models with the F test shows no significant differences in any of the patients. Apparently, an improved fit is a trade-off for including more parameters. The model with the lowest number of parameters is selected to prevent overfitting of the data.
Obviously, Patlak graphical analysis produced adequate estimates of the influx rate constant and distribution volume for prostate cancer (Tables 1 and 2). Figure 4 shows that primary prostate cancer had the highest transport rate constant k1, and BPH was similar. Significant differences (P < 0.01) between primary and recurrent cancer were found for k1, K, Vd, SUV-early, and SUV-late. The correlation between transport k1 and influx K was high, 0.82 and 0.84, respectively (Fig. 5). A poor correlation was found between influx rate constant K and SUV for recurrent tumors (Fig. 6), suggesting that, in this case, acetate uptake does not provide a reasonable estimate of the influx rate constant K in clinical practice. Reasons for this poor correlation could be early detection of recurrence because of close follow up, changed anatomy leading to altered perfusion patterns in the former prostate bed, and technical factors such as partial volume. Recurrences in the prostate bed may be small and present as thin foci, making recovery of the activity difficult. Examining the correlation between SUV-late and the various parameters revealed a weak negative correlation with k1 (r = −0.24) for recurrent tumors, whereas primary prostate cancer shows a positive correlation (r = +0.74). This supports the hypothesis that perfusion and, hence, transport are affected differently for each tumor type. Overall transport and metabolism of tracer—that is, influx rate K—shows a strong correlation with the transport rate k1 for both tumor types (Fig. 5). Partial-volume effects alone cannot explain the differences between primary and recurrent tumors as found in Figure 6. Incomplete recovery of the blood activity would lead to an increase in k1 and K. However, as a group, the size of the iliac vessels will be similar in men with primary versus those with recurrent cancer, and partial-volume effects on the input function should be the same. Average tumor size was also similar for both groups. Assuming a very inhomogeneous uptake in recurrent tumors, would lead to a lower recovery coefficient if compared with a more homogeneous group of primary tumors. However, although the parameters values would be lower, a positive correlation between influx and uptake would still be expected. The lack of such correlation in recurrent tumors suggests other reasons the than partial volume. The range of K is narrow (Fig. 6) and small sample size and selection of tumors could have contributed to the lack of correlation. On the other hand, biologic differences between primary and recurrent tumor cancer cells—that is, different membrane transport channels, altered TCA cycle, and so forth—might have caused this discrepancy. More research is warranted to elucidate these differences.
The acetate uptake in our subgroups was similar to those of Oyama et al. (6) for primary cancer, ranging from 3.3 to 9.9 SUV, compared with our 1.8–5.5 SUV. The higher numbers suggest more advanced disease in their population. They found a relationship between tumor grade and uptake. In our primary cancer group, 9 men had a T2 tumor and 1 had a T3 tumor, whereas the recurrent group was split evenly with 5 patients in each group having T2 and T3 tumors. The late uptake was slightly higher for T3 tumors with 2.2 compared with 1.9 for T2 tumors (P = 0.2). In the article of Oyama et al. on recurrent cancer (7), no uptake measures were provided, but our results show generally lower SUVs. Obviously, this is dependent on the time of discovering the PSA relapse. Because our patients were closely monitored, with a serum PSA test every 3–4 mo, the detection of recurrence was rather early, which might explain the lower uptake. This is corroborated by the PSA velocity, which is a measure of tumor burden. In 6 of 10 men with recurrent cancer, the PSA velocity was <0.3 mg/mL/mo. Oyama et al. (6) also found a relation between age and uptake. We had only 2 patients younger than 50-y-old. Surprisingly, the uptake in the men with BPH was higher for the 78-y-old (SUV 3.8) compared with the 49-y-old (SUV 3.2).
As was pointed out by van den Hoff (17), the transport constant k1 is not equal to the tissue perfusion F, but to F/E, with E being the extraction fraction (Supplemental Appendix). In the myocardium, there is a steep drop-off of acetate during the initial minute. We did not observe this in our data; our initial sampling of 10 s per frame should have been sufficient to measure such an early effect. The prostate time–activity curve shows a steep incline and reaches the plateau before 5 min. A myocardial time–activity curve, on the contrary, drops to <20% of its peak at 20 min. The acetate turnover and incorporation into lipids, amino acids, or steroids is too long to be measured in our 20-min time span. Therefore, the plateau-shaped time–activity curve in prostate cancers suggests retention of acetate in tissue. The sharp contrast between the curves of the prostate and the myocardium suggests that the model for the prostate proposed by Fricke et al. (27), based solely on increased citrate oxidation (28), is invalid. A plateau-shaped acetate uptake/retention curve has been reported for many other tissues, such as pancreas, salivary glands, bowel, as well as solid cancers and lymphoma (4). As reported by Kato et al. (8), the early uptake is the same as the late uptake, and the early-to-late uptake ratio is not significantly different from unity for any of the patients. The suggestion that the ratio is higher for primary prostate cancer and could be used for differentiation (4) is not supported by our data or those from Kato et al. (8). In this respect, prostate cancer behaves similarly to the behavior of normal prostate tissue. The finding that BPH cannot be differentiated from primary cancer (Fig. 4) has been reported for both 18F-FDG and 1-11C-acetate (4). The early-to-late acetate uptake ratio is equal to unity. This is in stark contrast with the myocardium, which shows a rapid decline, and 80% of the 1-11C-label has disappeared by 20 min. The plateau curve suggests a prolonged retention of the 1-11C-label in benign prostate cells.
Table 2 shows that k2 and k3 are similar for primary and recurrent tumors, whereas k1 and K are different, providing a means to differentiate them. Whereas k1 seems similar in primary and benign tumors, Table 2 suggests that k3 is twice as high in malignant tumors as BPH. Although our numbers are too small for conclusions, one could hypothesize that k3 might be able to differentiate between malignant and benign lesions. Table 2 shows that k2 is the highest for malignant tumors, about twice as high as BPH, and that Vd is the lowest for malignant tumors, less than half of BPH. Increased oxidation of acetate through the TCA cycle is manifested as an increased k2 or decreased Vd, which is the case in comparing malignant tumors versus BPH. This observation would be consistent with the work by Costello and Franklin (28), in which it was shown that, unlike normal prostate cells, which are citrate producing, prostate cancer cells are oxidizing cells (with increased Krebs cycle activity). Thus, it appears that k2, k3, or Vd might be able to discriminate benign from malignant tumors, whereas k1 or K should help differentiate between primary and recurrent tumors. Further investigations to pursue these findings are warranted.
We did not find a correlation between SUV and influx rate constant K for recurrent tumors. Although our sample is too small for firm conclusions, this finding is troublesome because it suggests that the uptake cannot be used to “characterize” or “grade” the type of recurrence. This might suggest that therapy monitoring with 1-11C-acetate is limited. For detection of recurrence, however, the presence or absence of acetate uptake is assessed and its utility in recurrent prostate cancer has been determined (5,7,27). After radical prostatectomy, either by surgery or radiation, no viable tissue is expected in the prostate bed. Thus, any acetate uptake in this region is considered to indicate recurrence of disease.
The field of view of our system is 16 cm, which is not suited for adequate lymph node imaging or staging. Four patients had a pathologic lymph node in the field of view, and 2 patients had a bone metastasis. Using the 3C-3p model, the kinetic parameters of the lymph nodes closely resembled those of the prostate cancer of the patient in question. The bone metastases, on the other hand, demonstrated a transport and influx constant twice as large as those of the corresponding prostate cancer. This latter finding might be interpreted as more aggressive behavior of the bone metastases than the primary tumor.
Study Limitations
The limitations of this study are the relatively small subgroups, heterogeneity of tumors, and different stages of disease. Despite small subgroups, robust effects were found for primary cancers, revealing both a significantly higher transport (k1) and influx (K) compared with those of recurrent prostate cancer. A linear relationship between transport and influx was also found for primary prostate cancer but not for recurrent cancer. The selection of a constant recovery coefficient (0.6 for the iliac vessels and 1.0 for the tumor/prostate) does not represent individual variations and may cause differences in the parameter estimates k1 and K. Individual correction for partial-volume effects—that is, PET/CT—would obviously be a superior approach. The main objective of our study was to describe the kinetics and to test the model structure. The standard model for 18F-FDG, with its known applicability under clinical circumstances, appears appropriate for 1-11C-acetate and could be reduced to 3 parameters by ignoring k4 and Vb.
CONCLUSION
The acetate kinetics of prostate cancer are adequately described with a 3-compartment model, simplified to the 3 parameters k1, k2, and k3, and including corrections for 11C-CO2 in blood and partial volume. FA provides image-derived blood clearance and tumor uptake curves that are user independent. Transport rate k1 and influx rate K, as well as SUV-early and SUV-late, are significantly different for primary cancer compared with recurrent cancer. The high correlation (r = 0.91) between 1-11C-acetate uptake in primary tumor and influx rate constant K indicates that simple uptake measurements (SUV) will be sufficient in clinical practice. The 2 men with BPH had parameters that resemble those of primary prostate cancer. K and SUV for PSA relapse cancer, on the contrary, do not show a significant correlation.
Footnotes
-
COPYRIGHT © 2008 by the Society of Nuclear Medicine, Inc.
References
- Received for publication June 21, 2007.
- Accepted for publication November 19, 2007.