Quantification of Cellular Proliferation in Tumor and Normal Tissues of Patients with Breast Cancer by [¹⁸F]Fluorothymidine-Positron Emission Tomography Imaging: Evaluation of Analytical Methods

Breast cancer is the leading cause of cancer death in women from the industrialized world. In patients with large tumors (>3 cm), preoperative chemotherapy or hormonal therapy is increasingly used to reduce the size of the primary tumor mass and any occult metastases in order to improve survival and/or as part of breast conservation surgery. There is an unmet need to develop imaging methodologies that can enable early and objective assessment of response to such neoadjuvant therapies. Cross-sectional imaging methodologies such as X-ray and computed tomography can give an indication of the sensitivity of tumors to treatment, but in most cases, only after several courses of treatment. Proliferation markers derived from tumor biopsies are both predictive of pathologic response and could measure disease outcome early enough (1-4 weeks) to allow treatment decisions to be altered (1–7). The drawbacks for using biopsies include sampling error, invasiveness, and the inability to perform multiple repeat procedures. In this article, we report the clinical development of 3′-deoxy-3′-[¹⁸F]fluorothymidine ([¹⁸F]FLT)–positron emission tomography (PET) for noninvasively quantifying proliferation in patients with breast cancer. This imaging method is likely to be superior to biopsy methods because it can report on the whole (heterogeneous) tumor, nodal disease, and metastases, and is less invasive for response monitoring compared with multiple biopsies.

[¹⁸F]FLT is a pyrimidine nucleoside that is a substrate for the cytoplasmic enzyme, thymidine kinase 1 (TK1; ref. 8). TK1 phosphorylates [¹⁸F]FLT into a highly charged product which is trapped and can be imaged using PET. A significant correlation between TK1 in breast tumors determined by immunohistochemistry and Ki-67 labeling index (Ki-67 LI) has been reported by He et al. (9). The activity of TK1 is S phase–specific (10–13), it increases in S phase (14) and is targeted for degradation in late M phase (15–18), such that newly divided G₁ cells have low TK1 levels (19). Our group has shown that TK1 and cofactor, ATP, regulate [¹⁸F]FLT uptake in tumors in vivo (20). Furthermore, our preclinical studies show that [¹⁸F]FLT-PET can be used to measure response to chemotherapy in mouse xenografts (21, 22). In parallel with these encouraging preclinical studies, a number of groups are developing [¹⁸F]FLT-PET clinically for imaging of cellular proliferation. Initial studies in lung cancer, colorectal cancer, and lymphoma (23–27) have shown high correlations between [¹⁸F]FLT uptake and Ki-67 immunostaining of biopsy material. To date, no systematic evaluation of analytic methods for determining “proliferation-related” PET parameters in breast cancer has been done. As a prelude to the use of [¹⁸F]FLT-PET to study drug response in patients, we have studied the kinetics of [¹⁸F]FLT in tissues by PET and simultaneously in blood by gamma counting to allow a number of kinetic parameters to be determined. We have compared these kinetic parameters to Ki-67 LI to see how the parameters relate to proliferation.

Study design. The aim of the study was to determine parameters that reflect the delivery (K₁) and retention [rate constant for the net irreversible clearance of radiotracer from plasma to tissue (K_i), fractional retention of radiotracer in tissue at 90 minutes relative to that at 1 minute (FRT), standardized uptake value at 90 minutes (SUV₉₀), and area under the radioactive concentration versus time curve (AUC)] of [¹⁸F]FLT in humans and to evaluate how these compare with proliferation as assessed by immunohistochemistry of biopsy material. To achieve this aim, we did PET scans on patients with locally advanced or metastatic epithelial breast cancer who were not receiving active treatment at the time of the study. Fifteen patients with 13 primary tumor areas, 6 nodal metastatic areas, and 1 pulmonary metastasis were studied; overall, multiple slices from 29 distinct tumor regions were studied by PET. Patients with histologically proven invasive breast cancer and measurable disease (≥2.5 cm) with the site(s) of measurable disease outside the liver or bone marrow (as previous [¹⁸F]FLT studies have shown high levels of physiologic uptake in these tissues; ref. 28), were eligible. The median time between PET studies and biopsy was 18 days.

Positron emission tomography imaging. [¹⁸F]FLT was synthesized by Hammersmith Imanet by radiofluorination of the 2,3′-anhydro-5′-O-(4,4′-dimethoxytrityl)-thymidine precursor using a method described previously (29). PET scanning was done for 95 minutes after a single bolus i.v. injection of a tracer dose of [¹⁸F]FLT [ranging between 153 and 380 MBq, specific activity 25-465 GBq/μmol, determined by high-performance liquid chromatography (HPLC)] over 30 to 60 seconds. All scans were done on an ECAT962/HR+ scanner (CTI/Siemens, Knoxville, TN), which allows simultaneous data acquisition to form 63 transaxial planes (axial field of view, 15.7 cm). Data were binned into 31 discrete time intervals of varying length (30 seconds × 10, 60 seconds × 5, 120 seconds × 5, 180 seconds × 5, 600 seconds ×6). From the PET image data, regions of interest on tumor and normal tissues (normal breast tissue) were defined manually using the Analyze image analysis software (Biomedical Imaging Resource, Mayo Foundation Rochester, MN). Normal lung tissue was used instead of normal breast tissue where there was no normal breast tissue in the field of view (as in the case of patient no. 2 with a lung metastasis, or in patient no. 10 with relapsed disease in the axilla).

Blood and metabolite analyses. In addition to the PET scanning, arterial blood sampling was done continuously for the first 10 minutes, discrete arterial samples (10 mL) were taken at baseline, 2.5, 5, 10, 20, 30, 45, 60, 75, and 90 minutes. Total blood radioactivity was monitored by gamma counting and [¹⁸F]FLT plasma parent fraction (and metabolite, [¹⁸F]FLT-glucuronide; ref. 30) was determined by reversed-phase HPLC with radiochemical detection. To determine total blood radioactivity, 1 mL each of blood and plasma were counted in a sodium iodide well counter (assembled in-house). Plasma samples (2 mL) were deproteinated by mixing with 5-fold of ice-cold acetonitrile (Fisher Scientific UK, Leicestershire, United Kingdom) and centrifuged at 3,000 × g for 3 minutes at 4°C (Hettich, Scientific Laboratory Supplies Ltd., Nottingham, United Kingdom). The resulting supernatant containing the extracted radioactivity was concentrated by rotary evaporation (Heifdoff, LabPlant, Huddersfield, United Kingdom), resuspended in 3 mL of HPLC mobile phase and filtered using a 0.2 μm diameter filter (Acrodisc, VWR International Ltd., Leicestershire, United Kingdom). Aliquots (1 mL) of the filtrate were separated on a μ-Bondapak C₁₈ column (300 × 7.8 mm i.d., 10 μm particle size; Waters, Elstree, Hertfordshire, United Kingdom) with a mobile phase composed of 10 mmol/L potassium dihydrogen phosphate buffer (pH 4.0) and acetonitrile (85:15 v/v) delivered at a flow rate of 3 mL/min. The eluate was monitored for radioactivity with a gamma detector (Raytek, Sheffield, United Kingdom) linked to a personal computer–based integrator (Laura, Lablogic, Sheffield, United Kingdom) that enabled correction for [¹⁸F]-radioactivity decay and background. For each chromatogram, peak areas for [¹⁸F]FLT and [¹⁸F]FLT-metabolite were determined and expressed as the percentage of the total peak area.

To confirm that the [¹⁸F]FLT metabolite was a glucuronide, arterial plasma samples taken at 60 minutes were treated with β-glucuronidase (5,000 units/mL) for 30 minutes at 37°C, or β-glucuronidase (5,000 units/mL) and a β-glucuronidase inhibitor [saccharic acid-1,4-lactone (0.5 mol/L)] for 30 minutes at 37°C and analyzed by the HPLC method described above.

Determination of kinetic parameters. A key aspect of this study was to determine the kinetic parameters that describe the delivery (K₁) and retention (K_i, FRT, SUV₉₀, and AUC) of [¹⁸F]FLT in man. The SUV₉₀ and AUC were calculated from the time versus radioactivity curve. For this, the time versus radioactivity curve was decay-corrected and normalized for injected radioactivity and body surface area (BSA) at each of the 31 midframe times according to the equation below:

where A is the total tissue radioactivity (kBq/mL) at time (t) in seconds, λ is the decay constant for [¹⁸F] (1.053 × 10⁻⁴/s), ID is the injected dose in kBq, and BSA was expressed in units of m². SUV₉₀ was the SUV at the last time frame (m²/mL). For comparison, we also determined SUV_21.5, the SUV at 21.5 minutes (m²/mL). AUC (m²/mL × s) was calculated as the integral of all counts from 0 to 95 minutes.

K_i was determined by the modification of a graphical analysis first described by Herholz and Patlak (31). The original Patlak model assumes that there is a single radiotracer and a single source for the radiotracer, the plasma. [¹⁸F]FLT, however, undergoes a predominantly hepatic metabolism to a glucuronide, which is circulated in blood to all tissues and eliminated. Determination of K_i for [¹⁸F]FLT, therefore, demands the use of an arterial plasma input function that is corrected for metabolites, as well as an algorithm that takes into account the contribution of metabolites to the exchangeable space within tissues (the metabolite does not contribute to the specific signal; glucuronidation occurs in the same position that is phosphorylated; ref. 7). K_i was calculated according to the following equation at steady state (32):

where V is defined below,

and A, K_x, C_Tot, C_x, C_m, τ, V_ox, V_om and V_b are the total tissue radioactivity ([¹⁸F]FLT and metabolite; kBq/mL), K_i for [¹⁸F]FLT (mL plasma/s/mL tissue), total blood radioactivity (kBq/mL), radioactivity of parent compound ([¹⁸F]FLT) determined by HPLC (kBq/mL), radioactivity of metabolite (kBq/mL), time interval from time (s) of injection, steady state space of exchangeable region occupied by parent [¹⁸F]FLT, steady state space of exchangeable region occupied by the metabolite and blood volume, respectively. The solution of Eq. B is made simple because the individual components of parameter V are not of interest (32). For comparison, K_i was also calculated using the parent plasma and total plasma input functions.

K₁ (mL plasma/s/mL tissue) was determined by a two-tissue compartmental model as previously described for 2-deoxy-2-[¹⁸F]-fluoro-d-glucose ([¹⁸F]FDG; refs. 7, 33). This was possible for [¹⁸F]FLT because the proportion of [¹⁸F]FLT metabolites during the delivery phase is negligible. Decay-corrected FRT was determined by a general deconvolution technique, spectral analysis (7, 34, 35). This method allows modeling of the relationship between parent plasma radioactivity and tissue radioactivity to obtain the unit impulse response function (IRF). This function is superior to tissue radioactivity per se as it takes into account the time-dependent contribution of radioactivity from plasma. In this case, tissue data can be expressed as:

where n is the number of identifiable kinetic components, β is a constant chosen to describe the entire spectrum of expected kinetic behavior of the radiotracer from the slowest possible clearance λ, to the fastest measurable dynamic (λ < βi < 1), and α is the intensity of the kinetic component at βi. The retention parameter, FRT, was defined as the IRF at 90 minutes relative to that at 1 minute.

Immunohistochemistry. To evaluate the relationship between PET parameters and direct measurement of proliferation, formalin-fixed tumor samples obtained from core biopsies within 3 months (median 18 days) preceding the PET scan were sectioned and immunostained with an anti-Ki-67 antibody, NCL-Ki-67-MM1 (Novocastra Laboratories, Newcastle upon Tyne, United Kingdom). The number of total and Ki-67-positive cells were manually counted in eight randomly selected fields of view using a BX51 Olympus microscope (Olympus Optical, Tokyo, Japan) at ×400 magnification and with the aid of Sigma Scan Pro 5 (Aspire Software International, Leesburg, VA). The Ki-67 LI was calculated as the ratio of the number of Ki-67-positive cells to the total number of cells.

Statistical analysis. A paired t test was used to assess the difference between tumor and normal tissue data. The association between PET parameters and Ki-67 LI was determined by calculating the Pearson correlation coefficient (95% confident; two-tailed). P ≤ 0.05 was considered significant. Statistical analysis was done using GraphPad Prism version 3.0 (GraphPad Software, San Diego, CA).

Visual distribution of 3′-deoxy-3′-[¹⁸F]fluorothymidine in the body. All of the patients had measurable disease in the breast or axilla, except for patient no. 2, who had a secondary lung metastasis (from breast cancer). The mean dose of activity injected was 302 MBq (range, 153-380 MBq; radiochemical purity, 100%, as determined by HPLC; ref. 29). [¹⁸F]FLT localized mainly in tumor, liver, and vertebrae (Fig. 1A). The high signal intensity in the liver and vertebrae has been attributed to glucuronidation and bone marrow cell proliferation, respectively (36). [¹⁸F]FLT tumor localization was lower in an inflammatory tumor (Fig. 1D and G) compared with a lobular carcinoma (Fig. 1B and E), which in turn had lower [¹⁸F]FLT uptake than a ductal carcinoma (Fig. 1C and F). In large tumors (>5 cm), there was marked heterogeneity; regions of obvious necrosis were excluded from the analysis. Radiotracer localization also differed between primary tumors and metastases from the same patient. Localization of radioactivity in normal tissues such as breast and lung was generally low.

Quantitative 3′-deoxy-3′-[¹⁸F]fluorothymidine–positron emission tomography kinetic parameters in tumors and normal tissues. Quantification of tissue kinetic parameters allows the generation of physiologic information of universal comprehension. Using arterial plasma samples from five patients, we were able to biochemically confirm that the sole plasma metabolite detected by HPLC was a glucuronide (Table 1; Fig. 2). We have determined parameters that describe delivery and retention of [¹⁸F]FLT-derived radioactivity in humans, and for the first time, we have employed a rigorous form of the Patlak model described for 2-[¹¹C]thymidine studies to determine K_i (32). The proposed compartmental model for [¹⁸F]FLT is shown in Fig. 3A. In this model, [¹⁸F]FLT-glucuronide derived predominantly from the blood (transported from the liver) can occupy an exchangeable space but does not contribute to the specific signal derived from the phosphorylation of [¹⁸F]FLT to [¹⁸F]FLT-phosphate. Determination of plasma-tissue exchange rate constants was made possible because of the ability to perform online arterial sampling and HPLC profiling of parent [¹⁸F]FLT levels. [¹⁸F]FLT was slowly metabolized in all patients such that only 71.54 ± 1.50% of [¹⁸F]FLT remained at 90 minutes post-injection, (coefficient of variation, 8.09%; Fig. 3B). Using the modified Patlak approach, a linear phase of the curve was discernible at steady state (Fig. 3C), indicating that there was unidirectional uptake of the radiotracer; the slope, K_i, was higher in tumor than in normal tissues. The spectral model also fitted the data and showed that there was a retention component in tumor but less so in normal breast (asymptotic nature of the IRF; Fig. 3D). As seen in Fig. 3E, spectral analysis shows all the dynamics, fast, and slow components. A trapped component is only seen in tumor and not normal breast, corroborating the low K_i of normal breast (Table 2; Fig. 3E).

A summary of the parameters from all 29 individual tumor sites from 15 patients is shown in Table 2. The delivery and retention of [¹⁸F]FLT were higher in tumors than in normal tissues as assessed by paired t test [K_i (P < 0.0001), SUV_21.5 (P < 0.0001), SUV₉₀ (P < 0.0001), AUC (P < 0.0001), FRT (P < 0.0001), and the delivery of [¹⁸F]FLT K₁, (P < 0.0001)]. K₁ values were low and similar to the low blood flow values seen previously in breast tumors (37). The variance for K_i and K₁ model fits ranged between 1% to 9% and 4% to 21%, respectively. When parent plasma (corrected for [¹⁸F]FLT glucuronide in plasma but not for the presence [¹⁸F]FLT-glucuronide in the exchangeable space of tissues) and total plasma input functions (uncorrected for [¹⁸F]FLT-glucuronide in plasma and tissue) were used for Patlak analysis, the variance for K_i model fits ranged between 1% to 8% and 1% to 10% for parent plasma and total plasma K_i's, respectively.

3′-Deoxy-3′-[¹⁸F]fluorothymidine uptake correlates with Ki-67 labeling index. Because [¹⁸F]FLT is to be used as a surrogate marker of cell proliferation, it was important to examine the association between its kinetic parameters and a standard measure of proliferation, Ki-67 LI. Of particular interest are associations with the kinetic parameters that describe retention because these are related to the phosphorylation of the radiotracer by TK1. Suitable histology was available in 12 of the 15 patients. An example of the difference in retention of [¹⁸F]FLT between tumors with low (patient nos. 1 and 11) and high (patient no. 4) Ki-67 LI is shown in Fig. 1. A strong correlation was found between Ki-67 LI and K_i (r = 0.92, P < 0.0001), FRT (r = 0.92, P < 0.0001), SUV_21.5 (r = 0.71, P = 0.0098), SUV₉₀ (r = 0.79, P = 0.0022), and AUC (r = 0.76, P = 0.004; Fig. 4). There was a nonsignificant correlation between Ki-67 LI and K₁ (r = 0.28, P = 0.38).

We have shown that [¹⁸F]FLT is delivered and retained to a higher degree by breast tumors compared with normal breast and that the retention of [¹⁸F]FLT correlates with Ki-67. [¹⁸F]FLT is a molecular imaging probe for noninvasively quantifying cellular proliferation in patients. It has the potential to play a key role in the management of breast cancer as a prognostic marker, as well as for monitoring the sensitivity of tumors to therapy. Due to these important applications, we have done a detailed analysis of [¹⁸F]FLT pharmacokinetics in breast cancer and normal tissue, and determined correlations between [¹⁸F]FLT retention and Ki-67. [¹⁸F]FLT-PET images showed high tumor-to-background contrast. Uptake was also seen in the liver and bone marrow due to catabolism of the radiotracer, as well as its uptake by the proliferating bone marrow cells (28, 36).

Using new and previously described analytic methodologies, we measured the delivery and retention of the radiotracer in tumors and normal tissues. The kinetics support a simple model of intracellular retention of the radiotracer by phosphorylation (Fig. 2A; refs. 20, 21, 27). The K₁ for [¹⁸F]FLT was of the same order of magnitude as blood flow in breast tumors and normal breast previously reported by Wilson et al. (37); the K₁ values were in general higher for tumor (range, 0.96-3.82 × 10⁻³ mL plasma/s/mL tissue) than normal breast or lung (range, 0.24-1.18 × 10⁻³ mL plasma/s/mL tissue). Given the low blood flow in breast tumors, it remains to be seen how generalizable the conclusion from the work will be for tumors with higher blood flow. For comparisons with blood flow, estimations of K₁ in this study may be suboptimal given the large time frames of 30 seconds at the start of the dynamic imaging protocol (designed to improve the signal to noise ratio). This may explain the large variance in K₁ values.

Because the proliferation-specific signal for [¹⁸F]FLT is the formation of [¹⁸F]FLT-phosphate, we hypothesized that retention of [¹⁸F]FLT will correlate with proliferation. Both reversible and irreversible kinetic components were observed in the [¹⁸F]FLT-derived spectrum of kinetic components. The spectral analysis plot shows the entire spectrum of kinetics (reversible/fast and irreversible/slow). This includes the blood volume component (a very fast component seen as a large peak in the higher end of the spectrum; Fig. 3E), fast tissue components (the rapidly decreasing aspect of the IRF curve), and slow tissue components (asymptotic aspect of the IRF; the slowest components in the spectrum of kinetic components are virtually irreversible). The reversible component detected by spectral analysis may include influx/efflux of [¹⁸F]FLT, as well as dephosphorylation of [¹⁸F]FLT. In this article, we showed irreversible uptake kinetics by graphical analysis. The graphical analysis measures the net irreversible transfer constant (K_i) in the presence of reversible components. These two findings, spectral versus graphical, are not mutually exclusive as data from the entire time course of the radiotracer are presented in the former.

All the retention parameters for [¹⁸F]FLT correlated with Ki-67. The correlations with Ki-67 increased in the order AUC < SUV < K_i, FRT. This is the first description of K_i for [¹⁸F]FLT calculated with full correction for plasma and tissue metabolites. The superior nature of this parameter may be due to the implementation of these corrections. Visvikis et al. did K_i calculations for [¹⁸F]FLT uptake in colorectal cancer without correction for the space of exchangeable region occupied by metabolites (38). In our breast cancer studies, this leads to overestimation of K_i by 12.5% (data not shown), although the correlation with Ki-67 (r = 0.94, P < 0.0001) does not change (Fig. 4E). Ignoring any contribution from the metabolites underestimates K_i by 16.9%, although this does not affect the correlation with Ki-67 either (r = 0.94, P < 0.0001; Fig. 4F). The small difference in K_i values when exchangeable space is taken into account (12.5%) indicates that the contribution to the exchangeable space by labeled metabolites is not substantial, supporting the notion that glucuronides are rapidly eliminated. These findings also suggest that correction of data for metabolites may not be required for larger scale clinical studies in the future.

The semiquantitative parameters, SUV₉₀ and AUC, were also highly correlated with Ki-67. SUV is the parameter which has been used by most investigators in other tumor types including lung and colorectal cancer, and in general, correlation coefficients of 0.7 to 0.92 have been reported (23, 24, 26, 27). Our results suggest that, although suboptimal, semiquantitative methods may still be useful in assessing [¹⁸F]FLT retention. There was no significant difference between SUV₉₀ and SUV_21.5, suggesting that SUV measurements can be done anytime between 21.5 and 90 minutes after [¹⁸F]FLT injection. In the only other breast tumor study published thus far, Smyczek-Gargya et al. used [¹⁸F]FLT-PET to image 12 patients with primary breast cancer (39). In that study, there was a lack of correlation between Ki-67 and [¹⁸F]FLT-derived SUV. The scanning protocol in that case, however, was different and the SUV value was not corrected for body surface area. SUV normalized to lean body mass or body surface area eliminates the weight dependence of SUV (40, 41). The weight dependence of [¹⁸F]FLT has not been reported. We thought it was prudent, however, based on [¹⁸F]FDG data, to perform a BSA correction which has been shown to be weight-independent in patients with breast cancer studied with [¹⁸F]FDG.

Proliferation, as determined by [¹⁸F]FLT-derived K_i, was higher in tumors than in normal breast or lung. The values differed between primary tumors and corresponding metastases. This is important from both a prognostic viewpoint as well as for therapy monitoring, given that all tumor masses can be studied at once. Shields et al. have reported K_i values for a lymphoma and sarcoma in the dog (36). Their tumor-derived K_i values in the dog, where metabolism of [¹⁸F]FLT is lower, are in agreement with our metabolite-corrected K_i values in humans. In conclusion, the key findings to emerge from this study are:

• Thirty percent of [¹⁸F]FLT is metabolized in patients to a single metabolite, FLT-glucuronide.

• Primary breast cancer and metastases show a wide range of proliferative capacity as determined by [¹⁸F]FLT-PET.

• The delivery and retention of [¹⁸F]FLT in tumors were higher than those of normal breast and lung tissue (P < 0.0001).

• Metabolite-corrected kinetic parameters of [¹⁸F]FLT retention are more sensitive than semiquantitative methods such as SUV. Overall, both types of analysis are useful in the evaluation of [¹⁸F]FLT and correlate significantly with Ki-67 (P <0.01).

• Semiquantitative parameters are equally useful when monitored early (21 minutes) or late (90 minutes) post-[¹⁸F]FLT injection.

This initial study strongly supports further evaluation of [¹⁸F]FLT-PET in a larger trial to fully assess the utility of [¹⁸F]FLT in breast cancer. The technology has the potential to quantify the sensitivity of tumors to therapy, an important consideration in the individualization of therapy in patients with cancer.

Grant support: The United Kingdom Medical Research Council. R.C. Coombes and E.O. Aboagye are supported by Cancer Research UK.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank the radiochemists, blood laboratory and quality control staff, and radiographers of Hammersmith Imanet for the production of [¹⁸F]FLT and their expert assistance, and the patients for participating in this study.