Assessment of cerebral drug occupancy in humans using a single PET-scan: A [¹¹C]UCB-J PET study

This study presents the first human results using a newly developed displacement model for measuring drug-target engagement during a single PET scan with drug intervention [17]. Our results demonstrate the ability of this model to quantify drug occupancy from a single [¹¹C]UCB-J PET scan with LEV intervention. Occupancy estimates derived from both the Single-step and the Numerical solution aligned well with those obtained through the Lassen plot method, the gold standard approach for quantifying occupancy based on V_T estimates from multiple PET experiments – one without drug intervention and one or more with drug intervention [7, 32]. The IC₅₀ values correspond to an approximately average concentration of 12.4 µg/mL for the Single-Step solution, 11.7 µg/mL for the Numerical solution, and 6.3 µg/mL for the Lassen plot when normalized against the sample durations.

The average IC₅₀ value in µg/mL, as determined using the Lassen plot method, is slightly higher compared to values presented in a previous study [6], which reported LEV IC₅₀ values of 3.58, 4.02, and 4.9 µg/mL also using the Lassen plot in humans. Despite the IC₅₀ value from the Lassen plot in this present study is 1.4 µg/mL higher than the highest value in the previous study [6], the values are still within the same order of magnitude. Even within the previous study [6], the IC₅₀ values vary by 1.3 µg/mL. The higher IC50 value observed in the present study using the Lassen plot method could be attributed to the broader range of LEV doses administered, facilitating a 100% blockade of specific binding, in contrast to the 88% achieved in the previous study [6]. In contrast, the average IC₅₀ values in µg/mL calculated from the displacement model are approximately twice as big. This might be due to LEV’s slower brain entry rate compared to BRIVA, which is approximately 30 min in humans [6]. It is possible that the steady state between LEV concentration in the brain versus plasma is not fully achieved in the latter half of the initial PET scan but is reached during the second PET scan. However, despite the difference in average IC₅₀ values derived from the total AUC of plasma LEV concentrations, there was no significant difference in the estimated occupancies or V_ND between the displacement model and the Lassen plot. This may indicate that the displacement model handles the dynamic changes of the peak LEV plasma concentrations after the drug intervention well. The discrepancies observed in the average IC₅₀ values for the displacement model vs. the Lassen plot likely stem from differences in sample durations and the timing of occupancy assessments between the displacement and the block PET scans. A way to adjust for this could be to apply a correction factor to normalize the IC₅₀ to a standard duration. Regardless of this, it is noteworthy that the 95% CI for the computed average IC₅₀ values overlap for the displacement model and the Lassen plot method. Additionally, even though the Lassen plot is considered the gold standard approach for calculating occupancy, new methods, such as a maximum-likelihood estimator, calculate IC₅₀ better than the Lassen plot in simulated data [33]. Future PET studies could investigate the computation of occupancy with maximum-likelihood estimation vs. the displacement model potentially yielding even more comparable IC₅₀ values. The well aligned IC₅₀ values, expressed as total AUC and ranging from 731 to 808 µg*min/mL, for both solutions to the displacement model and the Lassen plot, indicate that the LEV concentrations are comparable between the displacement and the block scan, but the time of the drug exposure is halved for the displacement scans. Estimating IC50 values based on total AUC serves as a clinically relevant proxy for a patient’s LEV steady-state plasma concentration, as per the equation: Steady-state concentration = AUC/dosing interval. Unfortunately, to our knowledge, no other study has computed IC₅₀ values for LEV using a PET displacement model, and thus the results presented in this paper cannot be compared.

Visually inspecting the three occupancy plots revealed more variability in occupancy estimates when using lower LEV doses for the Lassen plot. This trend corresponds with a previous study [17], where the displacement model performed better with higher doses of BRIVA, with the analytical Single-step solution indicating a slightly higher Emax at lower BRIVA doses.

Additionally, individual four exhibited the lowest V_ND of 0.77 and the highest SV2A occupancy of 92% derived from the Lassen plot method. Individual four also had relatively low V_ND values estimated with the Numerical and the Single-step solution, 2.78 vs. 1.94, respectively (Table 2). This individual had the longest interval between the first and the subsequent PET scan (112 days, supplementary Table s1.). Nevertheless, this participant’s body weight, injected dose of [¹¹C]UCB-J, and injected mass were within the same range for both scans (supplementary Table s1), and the test-retest data for [¹¹C]UCB-J is relatively stable over 28 days [34].

Individual nine had 40 days between the two PET scans, with comparable body weight, injected dose of [¹¹C]UCB-J and injected mass (supplementary Table s1). An explanation for the discrepancy in the occupancy estimates between the displacement model and the Lassen plot for this individual may stem from the increased uncertainty when the models derive occupancies from lower doses of LEV. Individual nine received a LEV dose of 7 mg/kg. The same tendency is observed for individual ten, who received a LEV dose of 5 mg/kg, exhibiting a discrepancy between the occupancy estimates of 46% from the Single-step solution and 69% from the Lassen plot method. This tendency was also observed in our method paper validating the displacement model in simulated data [17]. The study presented skewed occupancies results for low doses of BRIVA for both solutions to the displacement model [17]. As previously discussed by Laurell and colleagues [17], this tendency can also be observed when applying the Lassen plot to low-dose drug administration [33]. One way to potentially improve this could involve employing a radioligand with a longer half-life, such as [¹⁸F]UCB-J [35]. This would extend the imaging window, ensuring sufficient radioactive tracer for drug displacement by the end of the PET scan, even with a smaller total injected dose. Future PET studies could evaluate the performance of the displacement model by comparing tracers with different half-lives.

In a previous PET study [6], four healthy individuals were i.v. administered 1500 mg LEV, resulting in occupancies ranging from 78 to 84% using the Lassen Plot method. The LEV dose, when recalculated in mg/kg from the supplementary material [6], corresponded to 18.2, 19.2, 28.6, and 25.4 mg/kg for each individual, respectively. The occupancies estimated by the displacement model in the present study, with comparable LEV doses, fell within the same range, ranging from 75 to 96%. Additionally, three healthy individuals in the same study [6] received a LEV i.v. dose of 250 and 600 mg, corresponding to a 3.6, 2.7, and 8 mg/kg dose with associated occupancies between 52 and 68%. Utilizing the displacement model, we estimated similar occupancies ranging between 46 and 57% with a LEV dose of 5 and 7 mg/kg. Naganawa and colleagues [9] estimated 76% occupancy with the Lassen plot method using 10 mg/kg i.v. LEV to block the [¹¹C]UCB-J signal. In the current study, a LEV dose of 10–13 mg/kg yielded a corresponding occupancy between 57 and 77% using the displacement model. Another study [8] presented a displacement model estimating LEV occupancy from the tracer and drug rate constants as well as free fractions of [¹¹C]UCB-J and the ASM and V_T of [¹¹C]UCB-J. The displacement model initially had to fit 13 unknown parameters when computing the relationship between [¹¹C]UCB-J binding and ASM uptake and clearance simultaneously [8]. Upon evaluation of the model the number of parameters was reduced to stabilize estimates including the rate constants K₁ for both displacement and block PET scans, V_T, free fraction of [¹¹C]UCB-J and the rate constant of the ASM \({K}_{1}^{D}\). When reanalyzing data from Finnema et al. [6], this alternative displacement model [8] estimated a mean occupancy of 84% when administering 1500 mg LEV, which is close to what we find in the present study. Thus, the displacement model presented in this paper offers a straightforward method for calculating drug occupancy, without direct estimation of drug uptake, yet giving similar occupancy results.

Finnema and colleagues have previously estimated an Emax of 88.4% for LEV, which was significantly different from 100% [6]. In the present study, multiple LEV doses were administered to explore the potential for LEV to achieve 100% blockage of specific binding. The Numerical solution yielded an estimated Emax of 102.6%, 97.8% for the Single-step solution, and 101% for the Lassen plot method. The slightly lower Emax estimated with the Single-step solution is expected, as this solution was initially developed for rapid-acting drugs where the single-step assumption is more appropriate [17]. The most prolonged te value for the Numerical solution was 43 min. for an individual receiving 5 mg/kg LEV, resulting in SV2A occupancies of 57%, 46%, and 69% with the Numerical solution, the Single-step solution and the Lassen plot, respectively. Unlike the pattern observed with a small dose of the fast-acting drug BRIVA [17], the Single-step solution tends to underestimate the occupancy with a low dose (5 mg/kg) of the slow-acting drug LEV when compared to the other two methods, suggesting that this model solution might be more appropriate for more rapid-acting drugsNonetheless, the application of the extra-sum-of-squares F-test revealed no significant differences when using a specific binding model with Emax constrained to 100% versus Emax unconstrained. Thus, as expected, the Single-step Emax of 97.8% was not significantly different from 100%.

Concerning SV2A-binding ASMs, LEV exhibits the slowest drug-target engagement, but whether the Numerical solution would still perform satisfactorily for even slower drugs remains to be elucidated. A potential new SV2A-binding ASM, Padsevonil, has a tenfold higher binding affinity to SV2A than BRIVA [36]. Future PET studies could investigate whether the Single-step solution performs better in estimating the drug occupancy for the rapid-acting Padsevonil relative to the Numerical solution and the Lassen plot. Notably, for the BRIVA interventions during the pig PET scans, the two distinct solutions to the displacement model yielded almost identical results [17].

A previous study in simulated and pig data [17] estimated an Emax for BRIVA to approximately 87% for both the Single-step and the Numerical solution. The study did not evaluate whether this was statistically different from 100%. In contrast, another PET study in humans [6] reported complete blocking of specific binding using BRIVA and the Lassen plot method. Based on our findings, we conclude that the displacement model shows consistent results with and performs equally well as the Lassen plot method using the more slow-acting drug LEV. Additionally, LEV can block 100% of the specific binding. In this latter study [6], LEV Emax was estimated to be 88.4% with a substantially smaller dose range. Thus, when establishing binding curves for a drug, the choice of analytical method and the dose range is important.

Previous studies have incorporated peak plasma concentrations in the Emax model due to the rapid displacement of the tracer by the drug [17]. In accordance with our expectations, we find a slower displacement of [11 C]UCB-J by LEV compared to BRIVA, and consequently, we used the AUC of the LEV plasma concentrations in the Emax model rather than the peak concentration. Notably, the AUC of the LEV plasma concentrations provided more comparable LEV concentrations between displacement and block scans, as opposed to the peak concentrations, which were much higher during the displacement scans and stable during the block scans.

Limitations of this study include considerations raised by Tuncel and colleagues, who assessed the 28 test-retest repeatability for [¹¹C]UCB-J V_T estimates, revealing bias of -7.7 for whole brain grey matter, -1.1 for hippocampus, and − 8.2 for medial temporal lobe at retest, averaged across nine healthy individuals [34]. This potential bias could impact our results for individuals scanned on two different days, whereas the test-retest V_Ts for healthy individuals scanned on the same day are good, with the exception of hippocampus, which was not accounted for in the present study [37]. We did not, however, find any significant difference in V_T estimates from individuals scanned twice on the same day vs. individuals scanned on two different days. Additionally, both baseline and postdrug V_T estimates in all included regions were computed using an arterial input function and the 1TCM, which is the preferred kinetic model for [¹¹C]UCB-J [7, 37]. Nevertheless, the Bland-Altman plots did reveal a greater variation in CV estimates among individuals scanned on two different days vs. individuals scanned twice on the same day. Another potential source of bias in the V_T estimates is that the baseline V_T estimates were calculated from only approximately 60 min of data, while the full 120-minute scan duration was used to calculate the postdrug V_T estimates for the block scan. Lastly, we would like to highlight that the occupancy estimates from the displacement model and the Lassen plot do not necessarily correspond to the same LEV exposure. As discussed above, the total AUC was computed over approx. 60 min. for the displacement scans and for the entire 120 min. of the block scans.

In conclusion, both solutions of the displacement model demonstrated comparable or improved performance compared to the Lassen plot method, with well-aligned estimations of SV2A occupancies and V_ND estimates. The reliable estimation of SV2A occupancy from a single [¹¹C]UCB-J PET scan in the living human brain obviates the necessity for multiple scans when assessing the LEV-SV2A target engagement. The average IC₅₀ values were 12.4 µg/mL for the Single-Step solution, 11.7 µg/mL for the Numerical solution, and 6.3 µg/mL for the Lassen plot. The next step is to apply this methodology in newly diagnosed patients with epilepsy from the BrainDrugs-Epilepsy cohort study [25] and investigate how SV2A-LEV target engagement is associated with LEV treatment efficacy, including seizure freedom and side effects during follow-up.