Imaging Brain Tumor Proliferative Activity with [¹²⁴I]Iododeoxyuridine¹

A noninvasive measurement of tumor cell proliferation could be used in the evaluation of tumor growth and grade of malignancy and to identify the most rapidly proliferating regions of the tumor, which would provide spatial information for radiation treatment planning and stereotactic biopsies. It could also provide a new and more relevant measure of treatment response that could identify treatment“success” or “failure” much earlier than changes in tumor size or volume identified with CT³or MR or identify changes in glucose metabolism measured with PET(1, 2, 3). This may be particularly important in the evaluation of patients early in the course of treatment, particularly those patients being treated with cytostatic drugs.

The potential for obtaining functional images of DNA synthesis using PET and SPECT has been recognized for some time (4, 5, 6). PET and SPECT provide the opportunity to perform noninvasive measurements of uptake, distribution, and clearance of radiolabeled precursors of DNA in both tumor and normal tissues of patients with cancer. Our proposed use of radiolabeled IUdR([¹²⁴I]IUdR) is substantially different from other investigations that use [¹¹C]TdR as a radiolabeled precursor of DNA synthesis (7, 8). The application of a time-dependent strategy for “wash-out” of radiolabeled metabolites (mainly iodide) to reduce background activity and improve image specificity is feasible with[¹²⁴I]IUdR and PET imaging because of the 4.2-day physical half-life of ¹²⁴I. A “wash-out strategy” is not possible with compounds labeled with ¹¹C or ¹⁸F because of the short physical half-lives of these radionuclides (20 and 110 min,respectively). Alternatively, kinetic modeling based on knowing the time course of parent compound and radiolabeled metabolites in blood can be used to correct measured tissue activity for non-DNA incorporated radioactivity.

This work represents an extension of our previous studies using[¹³¹I]IUdR (9) and[¹²³I]IUdR (10) and SPECT imaging. These earlier studies demonstrated that improved tomographic sensitivity would be beneficial, if not necessary. Because of the low count rates in the SPECT images, it was apparent that IUdR-SPECT could only identify the most active regions of tumor proliferation. We decided to initiate the studies reported here using[¹²⁴I]IUdR to achieve better tomographic sensitivity and counting statistics than are possible with PET (in comparison with SPECT). In 20 patients with primary brain tumors, we demonstrate the first clearly defined relationships between[¹²⁴I]IUdR uptake and retention and several independent measures (indices) of tumor proliferative activity, i.e., we compare the image-derived values of the IUdR-DNA incorporation constant (Ki), the SUV, and the Tm:Br radioactivity concentration ratio to other independent indices of tumor proliferation (tumor type and grade, labeling index, and survival).

Patients with a diagnosis of primary brain tumor (17 at the time of first presentation, 3 at the time of recurrence) were brought to the Paul Scherrer Institute. Participating hospitals included the University Hospital of Zürich and the Cantonal Hospital of Aarau. Patients were >18 years of age, had a Karnofsky score of >50, and had no known medical contraindications to PET scans. All patients had an MRI or CT scan, usually with contrast enhancement, as part of their standard medical care; these images were used to define the location of the tumor. Two patients (nos. 17 and 20) with meningioma received iodinated contrast-enhanced CT scans 12 and 24 days, respectively,prior to their PET studies. All other patients were imaged using gadolinium-enhanced MRI, 5 ± 4 (SD) days prior to their PET studies. All patients or family representatives signed informed consent documents prior to being entered into the study. Patients were pretreated with oral potassium iodide solution (10 drops SSKI three times per day) for 3 days prior to[¹²⁴I]IUdR administration to block thyroid uptake of radiolabeled iodide, the major radiolabeled metabolite. Patients were fasted 4 h prior to[¹²⁴I]IUdR administration. All patients had a tissue diagnosis of their tumor to confirm the type and grade of malignancy.

Fourteen of the 20 patients underwent surgical resection of their tumor within 3 days of completing their PET study, and BrdUrd LIs were determined. In those patients who underwent partial, subtotal, or total tumor resection, multiple tissue specimens (up to six) were obtained from different tumor sites. The individual LI values were averaged to result in a “global” LI for each tumor. A “peak” LI (highest individual) value was also used in patients with GBM because the individual LI values frequently varied in specimens obtained from this tumor. Peak LIs in patients with GBM were also correlated with peak IUdR uptake measures in the same tumor, although there was no attempt to confirm spatial correspondence between the measures. In the single patient who underwent tumor biopsy (patient 12), only one tissue sample was available.

Patient follow-up has extended over a 17–30-month period, and survival data are available on 19 of the 20 patients. Living patients were documented as of January 1999 and include 10 of the 19 evaluable patients. Note that patient 19 died of a pulmonary embolus.

¹²⁴I was routinely produced by The Cyclotron Corporation CV 28 compact cyclotron in Essen, Germany, via the nuclear reaction ¹²⁴Te(d,2n) ¹²⁴I by irradiation of enriched ¹²⁴TeO₂ (89.6%) with 14 MeV deuterons. After thermodistillation (6 min at 740°C), the ¹²⁴I product was adjusted to a specific activity of 12 Ci/μmol by iodide carrier (11). The solution [0.02 m NaOH (12)] was obtained in a volume not higher than 100 μl. It was transported by car to the Paul Scherrer Institute. The product had the following contaminants[their average relative yield at the time of patient administration(45 h after target irradiation) is given in parentheses]: 13.2-h ¹²³I (0.15%), 60-d ¹²⁵I (1.89%), 13.0-d ¹²⁶I(1.43%), 12.4-h ¹³⁰I (0.87%), and 8.02-d ¹³¹I (0.35%; Ref. 13).

[¹²⁴I]IUdR was routinely synthesized by direct electrophilic labeling of 2′-deoxyuridine with ¹²⁴I (12). Six hundred μg of 2′-deoxyuridine were dissolved in 250 μl of 0.2 mphosphate buffer (pH 7.4). In an IodoGen-coated ReactiVial, the solution was allowed to react for 15 min at 65°C under stirring with 185 MBq (5 mCi) ¹²⁴I in 100–200 μl of solution(including the rinsing solution). The product was separated by SEP-PAK C-18 cartridges and was washed with 30 ml of water that eluted deoxyuridine, unbound iodide, and the phosphates. The labeled compound was then eluted with 2 ml ethanol, and the solvent was slowly evaporated at 65°C under nitrogen atmosphere. The residue was taken up with 8 ml of physiological saline solution and filtered through a 0.22-μm Millex GS sterile filter. The radiochemical yield was∼60%.

Quality control was performed by TLC (Silicagel 60,F₂₅₄, 5 × 20, acetone) directly after synthesis and by radio-HPLC (Ultrasphere RP 18, 5 μm; eluent of water:methanol, 80:20 v/v) shortly before administration. On the basis of >30 syntheses, the average contamination grade was determined to be 2.9% by TLC (mainly iodate) and 4.3% by HPLC. A purity grade of 95%[¹²⁴I]IUdR could be kept routinely prior to injection into patients. Analysis of [¹²⁴I]IUdR metabolites showed that the in vitro stability of the solution allowed a time window of 2 days following the end of synthesis for patient administration (13).

On the day of study, a radial artery and a peripheral hand or arm vein were catheterized for arterial blood sampling and radiopharmaceutical administration, respectively. Patients were positioned in parallel to the orbitomeatal line. Head movements were restricted by an individually formed plastic head support. Before tracer injection, a 10-min transmission scan was acquired in each scanning position (see below) using a[⁶⁸Ga]/[⁶⁸Ge] ring source. Patients received a 10-ml i.v. infusion of[¹²⁴I]IUdR (28.0–64.4 MBq) over 3 min. A dynamic series of arterial blood samples and image acquisitions were obtained over 48 min (time frames: 3 × 1 min,10 × 3 min, and 3 × 5 min) using an ECAT 933/04-16 (Siemens-CTI PET Systems, Knoxville, TN). Given the limited axial sampling of the ECAT 933/04-16 (seven slices over 5.6 cm with an in-plane resolution of 8–9 mm FWHM), the couch was moved, and the head was re-imaged (50 to 65 min) to obtain a 14-slice image set of the whole head. The initial dynamic sequence was always obtained with the bulk of the tumor within the axial field of view. At completion of the first day of study, contour marks of the patient’s head were administered to the plastic head support to allow repositioning of the head in the scanner with the same support at the time of the washout study. Twenty-four h after[¹²⁴I]IUdR administration, a repeat imaging session (60 min acquisition time, 30 min per 7-slice image set) was performed. Two venous blood samples were also taken at that time and were averaged. Reconstructed images were obtained from the PET data. Corrections were made for randoms, scatter, dead-time, detector inhomogeneity, and attenuation. Image data were calibrated to Bq/ml based on a series of [¹²⁴I] phantom studies and decay corrected to the time of injection.

The same data acquisition protocol (PET, arterial blood) and data analysis was used in these experiments as described for the[¹²⁴I]IUdR studies. Similar doses of[¹²⁴I]iodide were injected (50–62.4 MBq), as an i.v. infusion over 3 min.

In selected patients, both the 24-h and 0–65-min PET images were registered to the MR image. For this purpose, PET image data sets of both studies were processed to form a contiguous volume of 14 adjacent planes. Scalp contours were defined on each of the 14 image planes of this volume, and similar contours of the scalp were derived from the MR images. The PET to MR registration was conducted using the Pelizzari(head and hat) algorithm (14). The registered PET images were resliced using trilinear interpolation along the planes of the MR scan.

The plasma input function of [¹²⁴I]IUdR was determined from 21 arterial blood samples drawn from a catheterized radial artery during the initial scan. Samples were immediately placed on ice and centrifuged at 4°C to obtain cell-free plasma. Plasma and whole-blood radioactivity was measured in a well counter that was cross-calibrated with the PET camera and expressed as Bq/ml and percentage of injected dose/ml of plasma. Up to 13 plasma samples,collected during the first 15 min after[¹²⁴I]IUdR administration, were further analyzed for parent compound and radioactive metabolites using an HPLC-radiation detection system as described previously(15). Seven hundred μl of cell-free plasma was mixed with an equal volume of ice-cold perchloric acid (0.8 mperchloric acid, 1%Na₂S₂O₅,and 0.1% EDTA) for deproteinization. After centrifugation, 1 ml of the supernatant was directly injected into the HPLC system. Metabolite separation was accomplished with a reverse-phase column (Beckman Ultrasphere RP18, 5 μm, 250 × 4.6 mm) using methanol:water (20:80 v/v) with a flow rate of 1 ml/min as mobile phase. The retention time of the unchanged tracer (7 min) and the metabolites (unbound [¹²⁴I]iodide at 2.5 min and [¹²⁴I]iodouracil at 5 min) were verified using unlabeled standards and UV absorption at 254 nm. The integrated peaks of tracer and metabolites were expressed as percentage of total plasma activity at the sampling time.

ROIs were determined using PET image data acquired from summed 33–48 min postinjection image frames. Elliptical whole tumor ROIs were then drawn on two or more adjacent planes of the summed images; their positioning was guided by visual comparison with the T1- and T2-weighted MRI or contrast-enhanced CT scans. Particular attention was paid to avoid inclusion of necrotic or cystic areas into the ROIs(glioblastomas). Peak tumor ROIs were obtained from the GBM tumor regions with highest tracer retention at 24 h, because GBMs frequently demonstrated intratumoral heterogeneity of radioactivity concentration. Elliptical ROIs for normal brain were obtained contralateral to the side of the tumor ROIs. The ROIs were then applied to the full dynamic series and to the 24-h images, and decay corrected tissue time-activity curves were generated.

Three tumor proliferation parameters were calculated from the ROI data:(a) the IUdR-DNA incorporation clearance constant, Ki; (b) the SUVs in the 24-h image; and(c) the tumor:normal brain radioactivity ratio in the 24 h image, Tm/Br.

The calculation of the IUdR-DNA incorporation clearance constant, Ki, involves a two-step process. The first step involves an estimation of the IUdR-iodide tissue plasma volume (Vp) and tissue distribution volume (Ve) from a kinetic analysis of the 0–48-min dynamic uptake data (see the “Appendix”). The second step involves the calculation of Ki:

where Am^24h (Bq/ml) is the measured radioactivity in the tissue obtained from the PET images at 24 h after [¹²⁴I]IUdR administration, Cp^24h (Bq/ml) is the concentration of radioactivity measured in plasma at 24 h, and ∫ Cp^IUdRdt (Bq*min/ml) is the plasma IUdR concentration-time integral (input function). Cp^24h largely reflects the concentration of radiolabeled iodide. Eq. A assumes that the IUdR-DNA incorporated radioactivity is irreversibly trapped during the course of the study (16, 17).

The SUV is a commonly used parameter in the clinical evaluation and comparison of nuclear medicine studies; it is unitless and was calculated in the standard manner (18, 19):

where the injected dose (Bq) and body weight (g) are known. SUV corrects for differences in dose of radiopharmaceutical administered and body weight between individual patient studies; it provides a better index for a comparisons between different patients and for the comparison of serial studies in a single patient.

The tumor:normal brain radioactivity concentration ratio (Tm:Br) was also calculated. An advantage of SUV and Tm:Br for routine clinical studies is that the arterial input function does not have to be determined. The disadvantage in expressing results in terms of SUV and Tm:Br is that the measured tissue radioactivity is not corrected for residual, non-DNA incorporated radioactivity in the tissue[i.e., Cp^24h * (Ve + Vp)] or for differences in the[¹²⁴I]IUdR input function between different subjects (20).

The calculation of the fraction of total tissue radioactivity(Am^24h) in the vascular(Vp), extracellular (Ve), and DNA-incorporated compartments at 24 h is based on the estimates of Vpand Ve, and the concentration of radioactivity in plasma at 24 h (Cp^24h); where the percentage of activity in Vp = 100*(Vp*Cp^24h)/Am^24h,the percentage activity in Ve = 100*(Ve*Cp^24h)/Am^24h,and the percentage of activity in DNA = 100 − (% in Vp + % in Ve).

The LI was determined based on a modified assay first used to determine tumor potential doubling times (21). The method has been described in detail elsewhere (22). Multiple biopsy samples were derived from resected tumors from patients administered BrdUrd 20 min previously. The biopsies were immediately fixed in 50% ethanol at 4°C. A single-cell suspension for flow cytometric evaluation was obtained from each biopsy as described previously (22). The cell density of the final suspension was monitored using an electronic counter (Coulter counter). The pellet was then resuspended in 100 ml HBSS and 20 ml of anti-BrdUrd FITC antibody (Becton Dickinson, Basel, Switzerland), incubated at room temperature for 25 min with occasional manual agitation, washed, and stained with 20 ml of propidium iodide (1 mg/ml) prior to analysis using a FACScan flow cytometer (Becton Dickinson).

The Pearson product moment correlation coefficient was determined for combinations of LI with each of the three PET-derived measures(IUdR-DNA Ki, SUV, and Tm:Br). In addition, each of the PET-derived parameters plus LI was correlated with survival. Plots were generated for each of the above-described relationships.

Twenty patients with primary brain tumors completed two imaging sessions (0–1 h and at 24 h) after i.v. administration of[¹²⁴I]IUdR, as well as arterial blood sampling for measuring radiolabeled metabolites and determining the arterial input function of [¹²⁴I]IUdR. The tumor and patient characteristics of the study population are summarized in Table 1. Fourteen of the 20 patients had a surgical resection of their tumor within 3 days of the IUdR PET study, and LIs were determined in multiple tumor fragments. Three additional patients were injected with[¹²⁴I]iodide, and the pharmacokinetics of iodide uptake and clearance from tumor and brain tissue was studied in the same manner as that described above for[¹²⁴I]IUdR; these results are presented in the“Appendix.”

The time course (0–48 min) of arterial plasma radioactivity was measured after i.v. [¹²⁴I]IUdR administration(Fig. 1, C and D). The total plasma activity peaked at the end of the infusion (3 min). HPLC analysis revealed very fast metabolism of [¹²⁴I]IUdR and revealed two radiolabeled metabolites, iodouracil and iodide. The fraction of iodouracil, the product of the initial degradation step catalyzed by nucleoside phosphorylases, remained small because of rapid dehalogenation to uracil and iodide. At ∼5 min after injection,[¹²⁴I]iodide was the major radiolabeled metabolite, and the concentrations of[¹²⁴I]IUdR and[¹²⁴I]iodide were approximately equal. After 10 min, [¹²⁴I]iodide accounted for >85% of total plasma radioactivity. The mean plasma half-time of[¹²⁴I]IUdR was calculated from exponential curve fits of the individual patient data (Fig. 1, C and D); the mean clearance half-time was estimated as 2.5 ± 0.5 min. These results are similar to those that we reported previously for [¹³¹I]IUdR(9). The input function (plasma concentration × time) of IUdR was calculated from the HPLC determined time course of [¹²⁴I]IUdR in plasma and was similar for all tumor groups (Table 2).

The initial uptake (0–48 min) of [¹²⁴I]IUdR derived radioactivity in tumor and brain was measured by dynamically acquired PET scans; typical results are plotted in Fig. 1, A,B and E, F for low- and high-grade tumors,respectively. The uptake profiles differed considerably in low-grade astrocytomas and glioblastomas and between normal brain and the higher grade tumors. The tissue and plasma radioactivity profiles were fitted to several possible pharmacokinetic models to account for IUdR and iodide transport as well as IUdR incorporation into tissue (tumor) DNA and yielded multiple solutions or highly unlikely rate constants with large error estimates. The inclusion of reasonable constraints was also applied to the model to reduce the number of parameter estimates, but a substantial improvement could not be obtained.

Therefore, we used a simpler, two-compartment model (blood and tissue) to estimate tissue plasma volume (Vp), the initial plasma clearance (influx) constant (K1), and the tissue distribution volume (Ve) of[¹²⁴I]IUdR-derived radioactivity (see the“Appendix”). Estimates of Vp and Ve were necessary to calculate the IUdR-DNA incorporation constant, Ki (Eq. A), from the images that were obtained at 24 h. Mean Vp, Ve, and K1 values for the different tumor groups and contralateral brain tissue are shown in the“Appendix” and can be compared with the values obtained in an additional three patients after[¹²⁴I]iodide injection (Appendix; ).

Imaging was also performed at 24 h (30-min imaging frame)after IUdR administration to allow for “washout” of exchangeable radiolabeled metabolites, predominantly iodide. Frequently, the pattern of radioactivity in the tumor was substantially different at 24 h(Fig. 2, C and F) compared with that at 1 h (Fig. 2, B and E). The level of radioactivity in the tumor at 24 h was also substantially less than that at 1 h,indicating that considerable “washout” of radioactivity had occurred (radioactivity that was not incorporated into DNA). Radioactivity retained in tumor tissue at 24 h after IUdR administration was shown previously to predominantly reflect IUdR-DNA incorporation in rodent tumors (17).

The clearance constant, Ki, defines the process of IUdR incorporation into DNA and was calculated by Eq. A (see “Materials and Methods” and “Appendix”). Mean Ki values for the different tumor groups are presented in Table 2, and individual patient data points are shown graphically in Fig. 3 A. Only the Ki value for meningiomas was not significantly different from zero. A comparison between low-grade gliomas (WHO grade II), anaplastic gliomas (WHO grade III), and GBM(WHO grade IV) revealed no significant differences in Ki,whereas the Ki of anaplastic gliomas and glioblastomas was significantly higher than the Ki of meningiomas.

Two other measures frequently applied in clinical nuclear medicine studies, SUV and the Tm:Br radioactivity concentration ratio, were calculated from the 24-h image data (Table 2; Fig. 3, B and C). The SUV for each of the different tumor groups was significantly higher than that of contralateral brain tissue, although these values include background radioactivity that did not completely clear during the 0–24-h postinjection period. The Tm:Br ratio was significantly greater than unity for all tumor groups, and significant differences between the tumor groups was observed, except between meningiomas and low-grade gliomas (Table 2). The expected progression in mean Ki, SUV, and Tm:Br values with tumor grade was observed (Fig. 3).

An independent measure of tumor proliferation, the BrdUrd LI, was determined on multiple tumor samples obtained at surgery in 14 patients(Table 2 and Fig. 3,D). The expected relationship between Ki, SUV, Tm:Br ratio, and the BrdUrd LI was observed in those patients who underwent surgery after the IUdR PET study (Fig. 4).

The relationship between Ki, SUV, Tm:Br ratio, and patient survival is shown in Fig. 5. The Tm:Br ratio demonstrated the highest correlation to the BrdUrd LI(Fig. 4,C) and to survival (Fig. 5 C).

One patient with a glioblastoma and poor renal function [patient 1, Table 1: creatinine 108 μmol (normal range 53–94), urea 10.8 mmol/l (1.8–8.2)] had substantially higher levels of plasma radioactivity at 24 h (predominantly[¹²⁴I]iodide). The plasma concentration was 1.08 SUV/ml, compared with a mean value of 0.31 ± 0.16 SUV/ml for the other 19 patients at 24 h. Not surprisingly, the tumor:plasma activity ratio was comparatively low (0.58), compared with a mean value (0.89 ± 0.19) for the other six GBM patients at 24 h. The Ki value calculated by Eq. A (2.9μl/min/ml; which corrects for plasma and non-DNA incorporated radioactivity) was more consistent with that of other patients with GBM(2.0 ± 0.3 μl/min/ml). Without this correction, the calculated value of Ki would be ∼8-fold higher and would reflect a substantial error by including non-DNA incorporated radioactivity. Of note, the “peak” Ki for this patient’s GBM (patient 1) was considerably higher (∼3-fold) than the“peak” values for other GBMs, suggesting that part of this tumor was growing very rapidly (which is consistent with a very short poststudy survival, 1.6 months; Table 1). The tumor and “peak”tumor SUVs, which are not corrected for non-DNA incorporated radioactivity or the arterial input function of IUdR, were very high(0.63 and 0.69 SUV, respectively) in this patient compared with that of the other six grade IV tumors (0.16 ± 0.06 and 0.23 ± 0.08 SUVs, respectively; Table 2). On the basis of our kinetic analysis of the data (see the “Appendix”), only 12 and 34% of measured radioactivity (SUV) in the “viable-appearing”and “peak” regions, respectively, of the tumor at 24 h reflects IUdR-DNA incorporated activity. This is because of the combined effects of a comparatively high blood volume(Vp = 0.075 ml/cc) and extracellular space (Ve = 0.435 ml/cc) and high blood radioactivity level (Cp = 1.08 SUV/ml) at the time of imaging. In contrast, the Tm:Br ratio for patient 1 was not substantially different from that of the other GBMs.

In one patient (no. 1), both IUdR and FDG imaging was performed;[¹⁸F]FDG preparation and the clinical imaging protocol have been described (23). The 24-h IUdR image is somewhat compromised by high levels of residual plasma radioactivity,as described above. Nevertheless, an interesting relationship is seen by a direct comparison of the IUdR and FDG images (Fig. 6, A and B). This comparison shows foci of both high IUdR and high FDG activity, but they are spatially separated in the tumor. A pixel-by-pixel scattergram of the registered IUdR and FDG images also demonstrates the differences in the two images (Fig. 6 C).

The first attempts to image brain tumors with radiolabeled halogenated pyrimidines were performed in the early 1960s(16) and 1970s (24). More recent studies involving 26 patients with different tumors were studied with planar gamma camera imaging at 24 and 48 h after the administration of 5–15 mCi of [¹³¹I]IUdR (25). In half the patients, there was some localization of radioactivity to at least one tumor site, and three of four brain tumors were visualized with high tumor:brain contrast. Our initial studies were directed at imaging brain tumor proliferation and used[¹³¹I]IUdR as well as[¹²³I]IUdR and SPECT (9, 10). These studies demonstrated that both [¹³¹I]IUdR and [¹²³I]IUdR SPECT imaging of tumor proliferation have low (marginal) sensitivity because of low count rates and can detect only the most active regions of tumor growth. To achieve substantially improved tomograph sensitivity, we initiated this study using [¹²⁴I]IUdR and PET imaging. The improvement in tomograph sensitivity between our initial study(9) and this study using an ECAT 933/04-16 tomograph is∼40-fold, with an improvement of 8.5-fold in the relative number of detected events. A 2–14-fold further improvement in tomograph sensitivity and relative number of detected events is obtained by current-generation PET tomographs compared with the ECAT 933/04-16.

The critical issue addressed in this study is whether the[¹²⁴I]IUdR PET images (and the derived values of Ki, SUV, and the Tm:Br activity ratio) correlate with other independent measures of tumor malignancy, including tumor grade,BrdUrd labeling index, and patient survival. We present the first such comparisons in 20 patients with primary brain tumors (see Figs. 3,4,5),and the results are both encouraging and discouraging. All three measures of tumor proliferation (Ki, SUV, and Tm:Br) yielded similar results. The expected correlation of these three measures with an independent assessment of tumor proliferation (LI) was particularly strong, and this was encouraging (see Fig. 4). Similarly, the correlation of these three measures with patient survival was good,although there was some scatter in the data and one patient death(pulmonary embolus) that was not related to his tumor. It should be noted that histological features (e.g., tumor type, grade,and presence of necrosis) as well as patient age are among the strongest predictors of survival in patients with primary brain tumors of glial origin (26, 27).

A discouraging aspect of our analysis was the finding that the Tm:Br ratio demonstrated the best (or nearly the best) correlation with the three independent measures of tumor malignancy. Our expectation was that Ki would provide the most accurate assessment of tumor proliferation. Reviewing the data provides some insight into this issue(see the “Appendix,” particularly the assessment of Veand Vp). The calculated percentage of measured (imaged)radioactivity at 24 h in the tissue compartments was different in the different tumor groups (Table 3), and this difference impacts on the calculation of Ki (Eq. A) as well as SUV and the Tm:Br ratio. The calculated fraction of reversible (non-DNA incorporated) radioactivity in the 20 patients reported in this study was substantially greater than that obtained in our previous studies in animals, where <10% of measured tumor radioactivity at 24 h was not incorporated in DNA(17).

The calculated fraction of residual, exchangeable radioactivity measured (imaged) at 24 h in this study was substantial. For most of the tumors, the mean estimates ranged from 47 (GBM) to 93%(meningiomas), whereas the mean estimate for three anaplastic gliomas was considerably lower (15%). The differences in the exchangeable fraction of measured radioactivity between the different tumor groups will have differing effects on the values of Ki, SUV, and Tm:Br presented in Table 2 and Fig. 3. This is because of the substantial differences in the percentage of measured radioactivity in Vp, Ve, and DNA (Table 3). The presence of poor renal function and reduced renal clearance of radiolabeled iodide further amplifies the magnitude and fraction of exchangeable radioactivity in the tissue, even after a 24-h period of washout (patient no. 1, see above). A large residual fraction of exchangeable radioactivity in the tumor results in higher tissue background activity and a loss of IUdR-DNA image specificity, and it will amplify the SUV and the Tm:Br values.

The calculation of Ki attempts to correct for the exchangeable fraction of intravascular and extracellular radioactivity. A question could be raised whether this calculation is accurate, i.e., whether the estimates of Ve and Vp from a kinetic analysis of IUdR-derived radioactivity data are accurate. We suggest that our numbers for the exchangeable fraction of radioactivity in brain and low-grade and anaplastic tumors with low vascular permeability (low K1; see in the “Appendix”) may be an underestimate, whereas in high grade gliomas with comparatively high vascular permeability (high K1) the estimate of Ve and the exchangeable fraction of measured radioactivity may be somewhat overestimated. The basis for this assessment is discussed in the “Appendix.”

Gliomas can be very heterogeneous and may include areas of necrosis and exhibit a wide range of intratumoral cellular density (particularly high-grade gliomas), which will impact on the measurements. To address these issues in part, the grade IV gliomas were also analyzed with respect to the “peak” area of radioactivity observed in the 24-h images. The rationale for “peak” tumor assessments has been discussed previously (28, 29) and was performed with the goal of identifying the most proliferative region of the tumor and presumably reflecting a more homogenous component. The 24-h images of the anaplastic and low-grade gliomas were considerably more uniform,and no attempt was made to perform “peak” activity analysis of these tumors.

Some limitations of current methods to image tumor proliferation are related to the rapid metabolic degradation of the nucleoside(e.g., radiolabeled IUdR and TdR) in blood, because of hydrolysis of the glycosidic bond by TdR phosphorylase. TdR phosphorylase is homologous with platelet-derived growth factor and limits the magnitude of the input function. The arterial input function of IUdR (48 ± 12 SUV*min; Table 2) is low compared with the input functions of FDG [170 ± 30 SUV*min;calculated from previous FDG experiments (23)]. In addition, the fraction of tumor cells that are undergoing division and are in S-phase at the time of IUdR injection is low in brain tumors,which further limits the amount of IUdR incorporated into DNA. Thus,the level of radioactivity in the tumors reflecting DNA incorporation is relatively low (0.13–0.29 SUV; Table 1) compared with other tumor-imaging agents, such as FDG, where brain cortical values are substantially higher (3.8 ± 1.3 SUV; Ref.23).

A variable common to all cell metabolism, transport, or cell epitope imaging studies is cell density within the field of view. Glioblastomas can have substantial differences in regional morphology (ranging from densely packed viable cells with high proliferative indices to areas of low cell density, microcystic changes, and areas of frank necrosis or cyst formation). Low-grade gliomas can also vary with respect to cell density and microcystic changes, whereas anaplastic gliomas tend to be more homogeneous and more densely packed. Thus, the level of radioactivity in the images is influenced by several factors and should be considered as a reflection of the magnitude of proliferating cells within the field of view. For this reason, we elected to express our data for GBM as a “peak” ROI value as well as a whole tumor value(excluding frankly cystic and necrotic components identified on the MR image).

A morphologically heterogeneous tumor is a potential source of error,particularly when cystic or necrotic regions of a tumor cannot be clearly identified and excluded on the basis of the MR image. Fluid-containing cysts or necrotic tumor regions can act as“reservoirs” of residual (exchangeable) radioactivity that clear more slowly compared with other parts of the tumor. Some of these areas were initially mistaken for active proliferative regions on the 24-h PET images, prior to the registration and comparison with the MR images. Thus, frankly necrotic and cystic tumor regions must be identified and excluded from the analysis to avoid potential confusion in the interpretation of the PET images.

Nevertheless, there are specific advantages for[¹²⁴I]IUdR proliferation imaging. The long(4-day) physical half-life of ¹²⁴I, as well as other isotopes of iodine (¹²³I and ¹³¹I) that can be used to label IUdR, provides the opportunity to use a “washout” strategy and “late” imaging. We have described this imaging strategy previously (9, 17). A 24-h washout period was used in this study; iodide is cleared by the kidney and gastrointestinal tract, and >70% of radiolabeled iodide is excreted in the urine during the first 24 h(16, 30, 31). Thus, background radioactivity will be substantially lower 24 h after IUdR administration compared with that at 1 or 2 h. As described above, the 24-h washout period did not achieve the level of tissue clearance nor image specificity that was achieved in our previous animal experiments (17). We estimate that only 53–85% of glioma radioactivity measured at 24 h reflected IUdR-DNA incorporation; in meningiomas, it was only 7%.

A similar washout strategy is not possible with[¹¹C]TdR imaging of tumor proliferation. Measurements of ¹¹C radioactivity in tumor and normal tissues within 1 h of methyl-[¹¹C]TdR administration include a large fraction of radiolabeled metabolites (6, 32). The 20-min physical half-life of ¹¹C is too short to allow for sufficient tissue and body clearance of these metabolites. Recently, it has been suggested that ¹¹C labeling of TdR on the 2′ position of the pyrimidine ring results in fewer radiolabeled metabolites and that an increased percentage of total tissue radioactivity is measured in the acid insoluble or DNA fraction at 1 h after 2′-[¹¹C]TdR (80%) compared with methyl-[¹¹C]TdR (40%) injection(33, 34). However, substantially greater amounts of ¹¹C-labeled carbon dioxide and bicarbonate are measured after administration of 2′-[¹¹C]TdR. It has been shown in the dog that the predominant fraction of blood radioactivity is[¹¹C]CO₂/[¹¹C]CO₃⁻3 min after 2′-[¹¹C]TdR administration;[¹¹C]CO₂/[¹¹C]CO₃⁻accounts for ∼70% of total blood activity between 5 and 60 min, and only 47% of total administered radioactivity is exhaled over 10 min(32). At least 25% of blood radioactivity will be reflected in background tissue radioactivity attributable to bicarbonate (33). To address these issues, various groups have resorted to more “sophisticated” analytic methods and complex modeling to obtain reliable measures and parametric images of tissue(tumor) proliferation (7, 8, 36).

It must be emphasized that tumor proliferation imaging is not limited to brain tumors and can be applied to most systemic solid tumors as well (7, 8, 36, 37, 38). It is also important to note that the spatial location of high[¹²⁴I]IUdR uptake does not necessarily correspond to the location of high metabolic activity measured with[¹⁸F]FDG in the same tumor (see Fig. 6). A comparison of the registered images presented in Fig. 6 shows that there are foci of high IUdR and high FDG activity, but they are spatially separated in the tumor. It should also be noted that regions of high FDG accumulation within experimental tumors have been shown to be associated with morphological changes of hypoxia-ischemia, impending necrosis, and the infiltration of macrophages (39, 40, 41). Tumor proliferation imaging provides significantly different information from that provided by FDG or other commonly used imaging agents. This was demonstrated in our initial studies comparing[¹³¹I]IUdR SPECT with[²⁰¹Tl]thallium SPECT and[¹⁸F]FDG PET imaging (9). Images obtained with IUdR, FDG, methionine, or thallium reflect different biochemical and biological processes that contribute to the accumulation and retention of these radiopharmaceuticals and provide substantially different information about tumor phenotype.

An important issue for future applications of tumor proliferation imaging is whether the values, or more specifically a change in the value of Ki, SUV, and Tm:Br, are predictive of treatment response, i.e., can tumor proliferation imaging with[¹²⁴I]IUdR, [¹¹C]TdR,or any other radiopharmaceutical predict treatment response “early”in the course of therapy, prior to changes in tumor volume or energy(glucose) metabolism? Early assessment is particularly relevant to the newer biological treatments involving cytostatic (in contrast to cytolytic) drugs, where the evaluation of treatment response can be protracted over months. An extended period of observation is often required before significant changes in the patient’s clinical status or in the radiographic evaluations are observed, and these criteria are currently used to define treatment response. Can tumor proliferation imaging accurately predict treatment response before the standard assessments of “time to progression of disease” and “survival”can be completed (38, 42)?

[¹²⁴I]IUdR imaging of brain tumor (and systemic tumor) proliferative activity is feasible with the current generation of PET tomographs, particularly if septa-out (3-dimensional)acquisitions are performed. The expected relationships between Ki, SUV, and Tm:Br and other measures of tumor proliferation(tumor type and grade, LI, and patient survival) were observed. However, greater image specificity and significance of the SUV and Tm:Br values would be obtained by facilitating renal clearance of radiolabeled iodide (hydration) and by imaging at later times to achieve greater washout and clearance of the exchangeable fraction of residual radioactivity in the tumors.

A two-compartment model (tissue and plasma) was used to fit the 0–48 min PET and arterial plasma radioactivity data after i.v. injection of [¹²⁴I]IUdR (20 patients). Estimates of the tissue plasma volume (Vp), tissue distribution volume (Ve), and the blood:tissue clearance(influx) constant (K1) were obtained, and the results are shown in . For the [¹²⁴I]iodide experiments, the 0–48-min and 0–24-h data sets for each of the three patients were fitted using the same two-compartment model; the results from the 0–24-h data set are shown in . A comparison of the 0–24-h fitted values and the 0–48-min fitted values (not shown) suggests that the values from the 0–48-min fit approximate the values from the 0–24-h fit when the permeability of the vessels (reflected in K1 and k2) is high,whereas in contralateral brain and in the anaplastic astrocytoma, where vascular permeability is comparably low, the value of Ve^iodide (Ki/k2) obtained from the fit of the 0–48-min data underestimates the value obtained from the fit of the 0–24-h data. The value of Ve^iodide for brain tissue estimated from the 0–24-h data set approximates the reported values of brain extracellular space, whereas brain Ve^iodide estimated from the 0–48-min data set was substantially lower (0.030 ± 0.003 ml/g). These differences are most likely attributable to the low count rate in many of the images and resultant noise in the images.

For the [¹²⁴I]IUdR studies, we fitted the tissue and plasma radioactivity profiles to several possible pharmacokinetic models to account for IUdR and iodide transport as well as IUdR incorporation into tissue (tumor) DNA. These fits yielded multiple solutions or highly unlikely rate constants with large error estimates. The inclusion of reasonable constraints were also applied to the model to reduce the number of parameter estimates, but a substantial improvement could not be obtained. Our inability to obtain reliable parameter estimates for IUdR-DNA incorporation in tumor tissue from pharmacokinetic modeling and fitting the data were primarily related to: (a) the low IUdR-DNA signal relative to the total signal measured during the initial 48-min period of imaging (when the images primarily reflected [¹²⁴I]iodide);and (b) the relatively low count rate, resulting in image frames with a low signal:noise ratio. A substantial improvement in scanner count rate (>5-fold) is expected using current-generation tomographs and 3-dimensional (septa out) image acquisitions.

The two-compartment model that we used to fit the 0–48-min[¹²⁴I]IUdR data set assumed similar transcapillary exchange parameters and distribution volumes (plasma and extracellular) for IUdR and iodide, and it also assumes that the IUdR-DNA component of measured tissue radioactivity during the 0–48-min period is small (see Fig. 1, A and B). The time course of total tissue and plasma radioactivity was used in the fits, and no attempt was made to account for different radiolabeled species (e.g., [¹²⁴I]IUdR,[¹²⁴I]IU, or[¹²⁴I]iodide) or irreversibly bound radioactivity (IUdR-DNA). The two-compartment model has the advantage of simplicity, and reasonably good and consistent fits of the data were obtained (see Fig. 1, E and F).

The model simplifications described above are reasonable given:(a) the octanol-water partition coefficients of IUdR and iodide;⁴(b) the fraction of IUdR binding to serum proteins[e.g., non-bound fraction = 56.0 ± 11.9%(n = 6)];⁵(c) the low fraction of measured radioactivity that represents IUdR-DNA during the 0–30-min period (see Fig. 1, A and B); and (d) the fact that beyond 5 min the dominant radioactive species in blood is[¹²⁴I]iodide (see Fig. 1, C and D). For high-grade gliomas, only 19 ± 18%of the radioactivity imaged at 1 h was estimated to be attributable to IUdR-DNA incorporation; for meningiomas, the value was<4%.

The tumor estimates of Ve and Vp obtained from the two-compartment fit of the 0–48-min[¹²⁴I]IUdR data are reasonable ,although they are not perfect. Nevertheless, they were used in the calculation of Ki (Eq. A). The 5-fold higher K1for brain in the [¹²⁴I]IUdR studies (compared with the [¹²⁴I]iodide studies) probably reflects the higher lipid solubility of IUdR (compared with iodide⁴), although IUdR is partially bound to plasma proteins (see above). Also of note is the 40% lower estimate of brain Ve in the IUdR study (Table ) compared with that in the iodide study and to other published estimates of brain extracellular space (43, 44). This is partially explained by the noisy data and shorter duration (0–48 min) of the IUdR data set used in the fit , compared with the longer duration (0–24 h)of the iodide data set as discussed above. Only ∼50% of brain radioactivity measured at 24 h is accounted for by using the values of Ve and Vp shown in . In contrast, ∼81 and ∼105% of measured brain radioactivity can be accounted for if the published values of brain extracellular space(Ve) for white matter (0.11 ml/g brain) and gray matter(0.15 ml/g brain), respectively, are used in these calculations. This comparison is appropriate and consistent with many studies showing little or no proliferation (DNA synthesis) of adult brain tissue(45,46) and very low rates of cell proliferation in brain vascular structures (49). This suggests that our Ve estimates in tumors with low vascular permeability (low K1 values) may also be low, and that the exchangeable(nonincorporated) radioactivity correction used in Eq. A is likely to be underestimated under these conditions. Because the residual(exchangeable) radioactivity in low-grade gliomas at 24 h may account for a sizable fraction of the measured radioactivity (Table 3),an underestimate in Ve would result in higher than expected Ki values. This was seen in a comparison of the values of Ki, Ve, and K1 for some low-grade (grade II) and anaplastic (grade III) gliomas in Tables 2. This observation illustrates the importance of using a “washout”strategy with [¹²⁴I]IUdR and suggests that a longer washout period would be beneficial with respect to increasing image specificity (e.g., lowering the fraction of exchangeable background radioactivity). Not surprisingly, the estimate of brain Vp in both studies was nearly identical.

Another issue to consider is the basis for the difference in the pattern of radioactivity distribution in the 1- and 24-h images (see Fig. 2). The pattern of radioactivity observed at 1 h generally resembled the pattern of blood-brain barrier disruption observed in the contrast-enhanced MR or CT images. This similarity is attributable to the fact that the distribution of IUdR-derived radioactivity is largely governed by the vascular volume (Vp), vascular permeability(e.g., K1), the size of the extracellular space(Ve) of the tumor, and the level of IUdR-derived radioactivity in blood ([¹²⁴I]iodide). As noted above, IUdR-DNA incorporation accounts for only a small fraction of the radioactivity in the 0–60-min images, and these images largely reflect[¹²⁴I]iodide (see Fig. 1, A and B).

The values of K1 varied substantially among the different tumor groups and contralateral brain tissue . Not surprisingly, meningiomas had the highest values, followed by GBM(grade IV). These K1 values are consistent with previous studies of brain tumor vascular permeability in patients using[⁶⁸Ga]-labeled EDTA (50),[⁸²Rb]rubidium (51, 52), and PET imaging. Similarly, the values of Ve and Vp for the different tumor groups are consistent with published values(50,51 52).

We gratefully acknowledge the help of Leo Wyer (Department of Medical Radiobiology, Paul Scherrer Institute) in radiosynthesis of[¹²⁴I]iododeoxyuridine.