Quantitative Analyses of ¹⁸F-FEDAA1106 Binding to Peripheral Benzodiazepine Receptors in Living Human Brain

Subjects

Seven healthy male volunteers (age, 20–31 y; mean age ± SD, 23.1 ± 4.0 y) participated in this study. All volunteers were free of any somatic, neurologic, or psychiatric disorders, and they had no history of current or previous drug abuse. This study was approved by the ethics and radiation safety committees of the National Institute of Radiologic Sciences, Chiba, Japan. Written informed consent was obtained from each subject.

Radioligands

DAA1106 (19) and DAA1123 (16) (N-(5-fluoro-2-phenoxyphenyl)-N-(2-hydroxy-5-methoxybenzyl)acetamide) as desmethyl precursors were kindly donated by Taisho Pharmaceutical Co. 2-¹⁸F-Fluoroethyl bromide for radiosynthesis was prepared by the reaction of ¹⁸F-F⁻ with 2-bromoethyl triflate using a newly developed automated system (16). ¹⁸F-FCH₂CH₂Br was purified by distillation and trapped in a solution of N,N-dimethylformanide (DMF) containing DAA1123 (1 mg) and NaH (6–8 μL, 0.5 g/20 mL DMF) at −15°C. After the radioactive reagent trapping ended, fluoroethylation required a further 10 min at 130°C. After adding CH₃CN/H₂O (6:4, 500 μL), the radioactive mixture was applied to a semipreparative high-performance liquid chromatography (HPLC) column (16). The radioactive fraction corresponding to ¹⁸F-FEDAA1106 was collected in a sterile flask. Radiochemical purity was >95% and specific radioactivities were 12–120 (mean ± SD, 54 ± 32) GBq/μmol at the time of injection. ¹⁸F-FEDAA1106 was injected intravenously at a dose of 176–190 MBq (mean ± SD, 183.0 ± 5.04).

PET and MRI Study

PET was performed using an ECAT EXACT HR+system (CTI-Siemens) in 3-dimensional (3D) mode, providing 63 planes and a 15.5-cm field of view. To minimize head movement during the brain scans, a head fixation device with an individual mouthpiece was used (Fixster Instruments). A 10-min transmission scan was performed with a ⁶⁸Ge-⁶⁸Ga source to correct for attenuation. PET was performed according to the following protocol: 20 s × 9 frames, 60 s × 5 frames, 120 s × 4 frames, 240 s × 11 frames, and 300 s × 12 frames. All emission scans were reconstructed with a Hanning filter cutoff frequency of 0.4 (full-width half-maximum = 7.5 mm). T1-weighted MRI was acquired by an Intera, 1.5-T system (Philips Medical Systems). T1-weighted images of the brain were obtained from all subjects. The scan parameters were 1-mm thick, 3D, T1 images with a transverse plane (repetition time/echo time, 21/9.2; flip angle, 30°; matrix, 256 × 256; field of view, 256 × 256 mm).

Arterial Blood Sampling and Metabolite Analysis

A catheter was kept in the radial artery during the PET scan. To obtain arterial input function, an automated blood sampling system was used during the first 12 min of each PET measurement. The concentration of radioactivity in arterial blood was measured every second. The time difference between the appearance of radioactivity in the brain and automated blood sampling was experimentally defined and considered in the calculation.

At the same time, arterial blood samples were taken manually from the open end of the catheter 40 times during the scan. Each blood sample was separated into plasma and blood cell fractions by centrifugation, and the concentrations of radioactivity in whole blood and in plasma were measured.

Arterial blood samples were also taken 11 times after injection to measure the ratio of the unchanged ligand to the whole radioactive substances in plasma. To each plasma sample, acetonitrile was added and then centrifuged. The obtained supernatant was subjected to radio-HPLC analysis (μBondapak C18 column; mobile phase, 65:35 acetonitrile/distilled water; Waters Corp.) to measure the ratio of the unmetabolized ligand to the whole radioactive agents. The plasma time–activity curve was corrected by the percentile amount of unchanged ligand in plasma.

Data Analysis

The MR images were coregistered to PET images using SPM99 (Wellcome Department of Cognitive Neurology). VOIs were defined over the dorsolateral prefrontal cortex, medial prefrontal cortex, temporal cortex, medial temporal cortex, parietal cortex, occipital cortex, anterior cingulate, posterior cingulate, striatum, thalamus, and cerebellum with a template-based method (20). The striatum and thalamus were refined manually by using coregistered images. Regional radioactivity of each brain region was obtained from these VOIs. Analysis of human data was implemented using MATLAB (The MathWorks) and PMOD (PMOD Technologies).

Theory

Compartment Models.

The compartment model for ¹⁸F-FEDAA1106 kinetic analysis is shown in Figure 1. K₁ is the rate constant for transfer from plasma to the free and nonspecific binding compartment C_f, k₂ is the rate constant for transfer from C_f to plasma, and k₃ and k₄ represent the binding and dissociation of the radioligand at the PBR site, respectively. The total distribution volume (DV) is represented by K₁/k₂(1 + k₃/k₄), and BP is defined as k₃/k₄ (21).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

Kinetic model for ¹⁸F-FEDAA1106. C_p = the radioactivity concentration in the metabolite-corrected plasma; C_f = the free and nonspecific binding compartment; C_s = the specific binding compartment; BBB = blood–brain barrier.

Simulation Study

Simulated time–activity curves of ¹⁸F-FEDAA1106 with several noise levels were generated to investigate the bias and variation of parameter estimation caused by statistical noise for each analysis method. A dynamic tracer concentration for ¹⁸F-FEDAA1106 was derived from the rate constant values (k values) given as true values and the measured input function according to the human PET protocol. The true k values used for the simulation were K₁ = 0.25, k₂ = 0.078, k₃ = 0.043, and k₄ = 0.0086 as determined from the human data of the dorsolateral prefrontal cortex by the nonlinear least-squares (NLS) method with a 2-tissue compartment model.

The noise ratio for each frame was determined according to the collected total count given by:Eq. 3Eq. 4where i is the frame number, C_t is the nondecaying tissue radioactivity concentration derived from the k values and the plasma input function, t_i is the midpoint time of the ith frame, Δt_i is the data collection time, λ is the radioisotope decay constant, and F is a scaling factor representing the sensitivity of the measurement system, introduced here to adjust the noise level. The noise was generated with random numbers based on gaussian distribution and added to the nondecaying tissue activity for each frame. The level of the noise for the dynamic data was expressed as the mean of percent noise described in Equation 3 from 1 to 120 min. In this simulation study, 1,000 noisy datasets were generated for each noise level of 1%, 2%, 3%, 4%, 5%, 7%, and 10%.

In these simulated time–activity curves, each k value was estimated by the NLS method, and the DV value was estimated by the Logan plot and MA methods. For NLS, initial k values varied between ±25% from the true k value, and parameter estimates were considered invalid if K₁, k₂, k₃, and k₄ were outside the range 0.0 < k <0.5, or k₃/k₄ was >3 times the true value. For Logan plot and MA, t* was fixed at 30, 40, 50, or 60 min. For Logan plot and MA methods, parameter estimates were considered invalid if DV was >3 times the true value. The reliability of the estimated parameters was evaluated by the mean and coefficient of variation (COV; SD/mean [%]) of the estimates, and the relationship between the reliability of parameter estimation and noise level was investigated for each analysis method. In the Logan plot and MA methods, the relationship between estimated DV and t* was also investigated with 1%, 3%, 5%, 7%, and 10% noise time–activity curves.

We also investigated the relationship between scan time and the reliability of each parameter for 60, 70, 80, 90, 100, 110, and 120 min using 5,000 simulated time–activity curves with 1%, 2%, 3%, 4%, or 5% noise. The simulations were performed on Dr.View (Asahi Kasei Information Systems Co.).

Human Study

Typical examples of time–activity curves for several regions are shown in Figure 2. Parameters estimated by each method are listed in Table 1. There was not much difference among regions in young healthy volunteers. Although the mean DV values estimated by GA and MA were about 20% smaller than those estimated by NLS in all regions, there were significant correlations between DV estimated by NLS, Logan plot, and MA in each region (Fig. 3). Significant correlation was also observed between BP estimated by NLS and DV estimated by each method in any of the regions (Fig. 4 for dorsolateral prefrontal cortex). In a small region, such as medial temporal and thalamus, DVs are underestimated in comparison with BPs. This may be due to a partial-volume effect caused by limited spatial resolution of the PET scanner. Typical examples of fitting curves by Logan plot and MA are shown in Figure 5. As shown, each curve generated by each method fits to the time–activity curve data very well.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2. 

Time–activity curves of dorsolateral prefrontal cortex, occipital cortex, and thalamus for ¹⁸F-FEDAA1106. (Inset) Graphs are time–activity curves for first 5 min.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3. 

Correlation between DV estimated by NLS and Logan plot (A) and NLS and MA (B).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4. 

Correlation between BP and DV estimated by NLS (A), Logan plot (B), and MA (C).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 5. 

Fitting curve by Logan plot (A) and MA (B). ○, Each data point; •, points after 40 min; bold line, fitting curve.

Simulation Study

Examples of simulated time–activity curves with several noise levels are shown in Figure 6. K₁, k₂, k₃, and k₄ were estimated by NLS, and DV was estimated by Logan plot and MA in these simulated time–activity curves.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 6. 

Time–activity curves with 1%, 5%, and 10% noise. (Inset) Graphs are time–activity curves for first 5 min.

By the NLS method, the biases of DV, BP, and K₁/k₂ were approximately 0.8%, 0.7%, and 0.2%, respectively, when the noise level was 3%. The variation of estimated DV and BP became larger as the noise increased (Table 2). When the noise level was 3%, the COVs of DV, BP, and K₁/k₂ were 5.8%, 5.6%, and 5.9%, respectively.

By the Logan plot method, the variation was a little smaller than by the other methods (Table 2). However, DV was underestimated even at a low noise level, and this bias became gradually larger with increasing noise. Bias and COV were affected by the starting point of linear regression, t*. Earlier t* induced a smaller COV and larger bias, whereas later t* induced a larger COV and smaller bias.

By the MA method, DV was also underestimated by about 10%–15%, but the bias did not change much with the noise level (Table 2). COVs of the estimated parameters were a little larger than by Logan plot. Bias and COV were greatly affected by t*. Earlier t* induced a smaller COV and larger bias, whereas later t* induced a larger COV and smaller bias, the same as by the Logan plot method. The variation by the NLS method was larger than by Logan plot or MA. But when the noise level was low, there was little difference in variation in comparison with Logan plot and MA.

In both Logan plot and MA, even though t* was increased from 30 up to 60 min, the estimated DV did not approach the true DV value.

The effect of the scan time on the bias and COV of BP and DV estimated by NLS was also investigated with 5,000 simulated time–activity curves with 1%, 2%, 3%, 4%, or 5% noise. The longer the scan time was, the smaller the bias and COV were (Fig. 7).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 7. 

Effect of scan time on COV (A) and bias (B).

¹⁸F-FEDAA1106 Binding in Human Brain

The estimated BP values for the healthy volunteers are listed in Table 1. There was not much difference among regions in young healthy volunteers. It is expected that the BP and DV values of ¹⁸F-FEDAA1106 will increase in case of neuronal damage or loss in several neurodegenerative and psychiatric disorders. In ¹¹C-PK11195 PET studies, semiquantification of specific binding was performed by normalizing the time–activity of VOIs by the activity of the cerebellum (13), cortical gray matter (15), and whole brain (14). However, in terms of the semiquantitative approach, the choice of the area used for normalization may affect the sensitivity for detecting changes. Moreover, as the uptake ratio of ¹¹C-PK11195 into the brain is very low and the ligand is excreted soon after the injection, the signal amplitude in the brain is not high.

¹⁸F-FEDAA1106 shows a 6-fold higher signal than ¹¹C-PK11195 (17,18) and a stable plasma concentration, so the application of ¹⁸F-FEDAA1106 can be expected to provide useful information. Moreover, ¹⁸F is convenient for long-time storage, and there may indeed be possibilities of its long-distance transportation to other facilities in the future.

Comparison of Various Analytic Methods for ¹⁸F-FEDAA1106

It is difficult to define a region where there are few PBRs because microglia are spread over the whole brain. This indicates that a reference method using a region with rare specific binding cannot be applied for the quantification of PBR. In this study, several quantification methods for receptor binding with arterial input function were applied for ¹⁸F-FEDAA1106.

In the simulation study, the COV of the Logan plot method was smaller than that with the other methods. However, the DV was underestimated by at least >10%, and the bias became larger as the noise increased. This bias becomes a problem when comparing the estimated DV among time–activity curves with different noise levels. This bias was partially caused by the noise in the tissue time–activity curve, and several methods for reducing the bias in the Logan plot have been suggested. By smoothing the time–activity curves with the iterative method termed generalized linear least squares, it was possible to reduce the noise-induced bias of the Logan plot (24). Another approach was rearranging Equation 1 of the Logan plot to a multilinear form, and it was demonstrated that this multilinear analysis reduced the noise-induced bias for both the 1-tissue compartment model and the 2-tissue compartment model—that is, the MA method (23). When the noise level was high, the bias of MA became smaller than that of the Logan plot, but, within the noise level of 5%, there was almost no difference between the 2 methods. Equation 2 of MA is multilinear, so the COV of the MA method was larger than that of the Logan plot. However, within the noise level of 5%, there was also almost no difference between these methods.

For tracers with early equilibrium time, enough points for linear regression can be obtained after reaching equilibrium, so either the Logan plot or MA is useful for DV estimation. However, these methods are not appropriate for slow kinetic tracers because sufficient points for linear regression cannot be obtained after equilibrium (23). It is very difficult to determine the optimal t* of the Logan plot or MA because a t* that is too early will cause underestimation, whereas the DV estimated with later t* is greatly affected by noise. In the simulation study, the DVs estimated by Logan plot and MA methods could not approach the true value, indicating that the equilibrium time t* was not early (Table 2). In human data, DVs estimated by Logan plot and MA with t* = 40 min were smaller than that by NLS. These results were consistent with the simulation result, showing that the optimal t* satisfying both the condition of equilibrium and the effect of noise did not exist in the ¹⁸F-FEDAA1106 120-min scan. Therefore, Logan plot and MA are not appropriate for the quantification of ¹⁸F-FEDAA1106. Another disadvantage of Logan plot and MA is that the only estimated parameter with these methods is DV. In general, k₃/k₄ and DV are considered to be parallel to each other on the assumption that there is little difference in K₁/k₂ among the regions and individuals. But because DV is the product of K₁/k₂ and (1 + k₃/k₄), the possibility that DV does not directly reflect the receptor density still remains.

The advantage of the NLS method is that every parameter can be estimated, with a disadvantage being that the COV is large in the simulation study (Table 2). But within a noise level of 3%, the variations of DV and BP were sufficiently small. In addition, the COV of DV estimated by the NLS method was smaller than those estimated by Logan plot and MA methods in human data (Table 1). Thus, we concluded that estimation by the NLS method is the most appropriate for the quantitative assessment of PBR with ¹⁸F-FEDAA1106.

To improve the reliability of parameter estimation, the ratio of K₁/k₂ is often used. However, in an ¹⁸F-FEDAA1106 study, K₁/k₂ cannot be constrained because the reference region cannot be determined.

Effect of Scan Time and Noise Level on BP Reliability

Because positron-emitting ligands have a minute-order half-life, the collected count in late life is relatively small. But, as ¹⁸F has an advantage over ¹¹C with a longer half-life (110 vs. 20 min), we can extend the duration of the scan time to >90 min. Therefore, the effect of the scan time on the bias and COV of BP and DV as estimated by NLS was investigated on 5,000 simulated time–activity curves with 1%, 2%, 3%, 4%, or 5% noise. The longer the scan time is, the smaller the bias and COV are (Fig. 7). Also, the lower the noise level is, the smaller the bias and COV are. The bias and COV of BP for a 60-min scan with 1% noise were nearly the same as those for a 120-min scan with 5% noise. In summary, the longer the scan time is, the smaller the COV and bias become. A longer scan time is needed for better validity of the parameters. However, for VOIs with a small noise level, the duration of scan time can be shortened by 10–60 min in practice—for example, in case of a patient's inability to remain still for a long time.