EC₅₀ images reveal reproducible spatial variation in drug affinity across single- and repeat-dose occupancy studies

Abstract

We investigated spatial variation in the apparent affinity of the D2 antagonist drug, JNJ-37822681, for dopamine receptors across single-dose (SD) and repeat-dose (RD) PET occupancy studies. Traditional whole-brain EC₅₀ estimates overlook potential spatial variation in drug affinity. We reanalyzed PET occupancy data from two published studies in healthy male volunteers (SD: administering 1 dose between 2–20 mg; RD: administering 13 doses of 10 mg over 7 days). Voxel-level occupancy images were generated from binding potential maps using Lassen Plot Filter (LPF), clustered via SLIC-Occ, and fitted to one- and two-parameter E_max models to generate EC₅₀ images, revealing regional variation in apparent affinity within the striatum. The two-parameter model incorporated receptor upregulation in a joint analysis of SD and RD data. EC₅₀ images demonstrated reproducible spatial variation in drug affinity across striatal subregions; caudate and ventral striatum showed lower EC₅₀ values than putamen. Additionally, left-right asymmetries in EC₅₀ were detected, once the effects of upregulation were addressed. Our findings validate EC₅₀ images for capturing spatial variation in drug affinity and highlights the importance of accurate modeling in chronic dosing studies. LPF and SLIC-Occ provide a robust framework for analyzing occupancy data, aiding dose optimization for drug trials.

Introduction

PET occupancy studies are routinely used to quantify the target receptor engagement of drugs in vivo.^1–9 Traditionally, the drug-induced alterations in tracer distribution in multiple regions of interest in the brain are used to calculate a single occupancy value per scan. From these values a whole-brain drug affinity, or half maximal effective (plasma) concentration (EC₅₀), is estimated for a cohort of subjects by fitting the occupancy data across subjects with a binding (“E_max”) model. Based on the resultant EC₅₀, a drug dosing window is determined with the goal of achieving a desired therapeutic effect without causing undesired side effects. This information is used by pharmaceutical companies to decide on drug dosing in early-phase clinical trials of candidate drugs. However, this method cannot accommodate any existing spatial variation in the apparent drug affinity.

Recently, de Laat et al. introduced the Lassen Plot Filter (LPF), an extended version of the standard Lassen plot, to estimate voxel-level occupancy images and subsequently built on it to generate EC₅₀ images rather than a single EC₅₀ value. LPF algorithm was originally developed for volume of distribution (V_T) images. Here, we have adapted it to estimate occupancy images using binding potential (BP_ND) images. EC₅₀ images are created by applying an E_max model at each voxel to occupancy images corresponding to different plasma drug concentrations.^10,11 EC₅₀ images generated by the LPF could be used to identify areas of high and low apparent affinity – perhaps as a result of differing concentrations of receptor subtypes with different affinities for the drug. In a recent work, we introduced a functional clustering algorithm (“SLIC-Occ”) to increase the precision of the EC₅₀ images by clustering the occupancy images into super-voxels across space and plasma drug level. ¹² Combining LPF and SLIC-Occ one can generate precise EC₅₀ images that may reveal spatial variation in the apparent drug affinity within the brain. In this paper, we seek to demonstrate the performance and consistency of this pipeline in generating EC₅₀ images on real data acquired in human volunteers.

We take advantage of two similar archival occupancy studies using the same drug and the same tracer. The two published studies used conventional regional analysis of occupancy to investigate the striatal dopamine D2 receptor occupancy of JNJ-37822681, a selective D2 receptor antagonist, in healthy volunteers.^13,14 te Beek et al. estimated a single whole brain EC₅₀ of 14.5 ng/mL in putamen after a single dose of JNJ-37822681. ¹⁴ Using the same tracer and drug, Schmidt et al. estimated EC₅₀ values of 15.8 ng/mL and 26.0 ng/mL for single dose and repeat dose protocols, respectively. ¹³ The findings of the two single dose studies are quite consistent. The discrepancy between the single and repeat dose analyses has been explained by Rabiner and Gunn as reflecting upregulation of D2 receptors following repeated exposure to the drug. ¹⁵ Rabiner and Gunn addressed these biases by introducing a variant of the E_max model which includes a term for upregulation. After accounting for upregulation, the studies were shown to be in agreement at the whole-brain level. In the present work, we apply the model of Rabiner and Gunn at the voxel-level to our occupancy images. By applying the upregulation model appropriately, we can create EC₅₀ images that reconcile the single-dose data with the repeat-dose (i.e., multiple doses over time) data. In doing so, we confirm the reproducibility of the EC₅₀ images and reiterate the potential added value of voxel-by-voxel analysis of occupancy data via the LPF and SLIC-Occ. Through this study, we aim to demonstrate that voxel-level analyses of occupancy data can capture authentic and meaningful spatial variations in drug affinity, a factor that may be critical in optimizing dosing regimens for drugs targeting the central nervous system.

Methods

Data acquisition

PET occupancy data from two previously published studies were made available to us by colleagues.^13,14 Both studies were designed to estimate the affinity of JNJ-37822681, a D2 receptor antagonist, using a single drug dose or a combination of single and repeat dose protocols. JNJ-37822681 has moderate affinity to dopamine D2 receptor, and low affinity to D1 and D3 receptors, serotonin 5HT2A and 5HT2C receptors, histamine H1 receptors and adrenergic α1A receptors. ¹⁶ In both studies, PET scans were performed on an ECAT EXACT HR+ scanner (Siemens Healthcare Knoxville, Tennessee, USA). [¹¹C]raclopride was used in both protocols. All the scans were acquired using a 60 min dynamic sequence comprised of 21 frames (6 × 5 s, 3 × 10 s, 4 × 60 s, 2 × 150 s, 2 × 300 s, 4 × 600 s). Below is a brief description of each of the datasets, additional information is available in the previously published studies.^13,14

Single-dose data

An open-label study was performed in twelve healthy male volunteers (18–34 years old) to determine occupancy of JNJ-37822681. ¹⁴ All subjects were scanned once at baseline (no drug) and at least once post-drug administration. Four of the twelve subjects were administered two different drug doses and scanned once following each drug dose, thus, there are twelve baseline and sixteen post-drug PET scans in the dataset. Administrations of the first and second drug doses were separated by a washout time of seven days. Each drug dose was in the range of 2–20 mg to achieve different plasma concentrations at the time of scan. Scans occurred 2 hrs. ± 30 min post-drug administration. Blood samples were collected before dosing and prior to, at the midpoint, and immediately after each scan. All PET sinograms were reconstructed using filtered back projection with 0.5 Hanning filter, resulting in spatial resolution of ∼7 mm at full width half maximum in the center of the field of view. In the present paper, we refer to these data as Single-Dose (SD), since all the subjects were scanned after administration of a single dose of the JNJ drug.

Repeat-dose data

An open-label study was performed in twelve healthy male volunteers (22–52 years old) to determine occupancy of JNJ-37822681. ¹³ All subjects were scanned once at baseline. Unlike the protocol from te Beek et al., in this study all the subjects were administered the same 10 mg drug dose twice daily for 6 days, and once on day 7 (13 total doses). Subjects were scanned at different times (2–58 hours) following administration of the 13^th drug-dose to achieve different plasma drug concentrations. Four of the twelve subjects were scanned twice after the 13^th drug dose; therefore, there are twelve baseline and sixteen post-drug scans. Blood samples were collected before dosing and prior to, at the midpoint, and immediately after each scan. All the emission PET scans were reconstructed using ordered subset expectation maximization (OSEM) with 16 subsets and four iterations with a zoom of 2.25 and 4 mm Gaussian post-smooth. In this work, we refer to these data as Repeat-Dose (RD), since all the subjects were scanned after administration of 13 doses of the JNJ drug.

Ethical approval and informed consent

We did not conduct any new experiment in this study. The SD study was approved by the medical ethics review committee of the VU University Medical Center in Amsterdam. Prior to medical screening, all volunteers gave written informed consent. ¹⁴ The RD study was conducted at the PET center of the University Medical Center Groningen (Groningen, The Netherlands). The study was approved by an ethics committee (Stichting Beoordeling Ethiek Biomedisch Onderzoek, Assen, the Netherlands) and all subjects gave prior written informed consent after receiving detailed information about the protocol. ¹³ We confirmed that the informed consent documents from both studies allowed for third-party use of the data for research purposes.

Pre-processing

We registered all PET images to the standardized brain template (AAL2, 2 mm isometric voxels) using SPM 12. First, each subject’s PET images at each time point were realigned to the subject’s mean PET image. Then, the subject’s mean PET image was registered to the subject’s specific brain extracted MRI. Lastly, subject-specific PET images were aligned with the template using the registration matrix derived from registration of the subject-specific MRI to the brain atlas template. All the co-registered images were visually inspected to make sure the registration process was applied correctly.

BP_ND images

There was no arterial blood sampling during imaging; therefore, the Multilinear Reference Tissue Model 2 (MRTM2) was used to calculate BP_ND for all the scans using cerebellum cortex as the reference tissue. ¹⁷ Since D2/3 receptors are mainly present in striatum (i.e., caudate, putamen and ventral striatum), we used the CIC Structural–Anatomical Striatal Atlas to mask the BP_ND images. ¹⁸

Occupancy images

The LPF was used to calculate the voxel-level occupancy images from the masked BP_ND images, by definition V_ND is zero. ¹¹ LPF uses a moving kernel to calculate each local voxel value for occupancy by applying the Lassen plot on the voxels in a local neighborhood (corresponding to the size of the kernel) in the corresponding pair of pre- and post-drug BP_ND images. The slope of the Lassen plot (i.e., occupancy value) is assigned to the voxel in the occupancy image corresponding to the origin of the kernel. In this study, a kernel size of 3 × 3 × 3 voxels was used to generate the voxel-level occupancy images. We then applied SLIC-Occ to the voxel-level occupancy images and generated super-voxels (i.e., clusters). ¹² SLIC-Occ is a functional clustering algorithm that generates clusters to minimize variation across occupancy and spatial distance. An initial starting number of clusters (K) and spatial weighting coefficient are required for SLIC-Occ to functionally segment the voxel-level occupancy images into clusters. A total initial cubic cluster size of 16 (i.e., K = 4,096 clusters) with a spatial weighting factor, m = 0.1 was used to generate the final clustered occupancy images. SLIC-Occ generated a total of 186 clusters strictly within the whole striatum that was composed of 2718 voxels. There were on average 14.5 ± 5.5 voxels per clusters in each of the occupancy images clustered by SLIC-Occ.

Model fitting

EC₅₀ values were calculated using 1-parameter fit for both SD and RD as shown in equation (1):

$$\mathit{Occ}=\frac{C}{C+{EC}_{50}}$$ (1)

where $C$ is plasma concentration of the drug, ${EC}_{50}$ is plasma concentration leading to 50% occupancy, and $\mathit{Occ}$ is the predicted occupancy.

Rabiner and Gunn introduced upregulation into E_max model in equation (1) to accommodate the prolonged effects of the drug on receptor density as shown in equation (2): ¹⁵

$$\mathit{Occ}=\frac{C-{EC}_{50}*\left(U_{\mathit{max}}-1\right)*\frac{C}{C+{EC}_{50}}}{C+{EC}_{50}}$$ (2)

In equation (2), $U_{\mathit{max}}$ is the maximum upregulation achievable. We used the upregulation model in equation (2) to estimate EC₅₀ from a joint-fit of both SD and RD data simultaneously. The $U_{\mathit{max}}$ was fixed at 1 for SD data and allowed to vary for RD data as follows:

$$\Phi =\underset{E{C}_{50},U_{\max }}{min}{\displaystyle \sum _{i=1}^n{\left(Occ-Oc{c}_{data}\right)}_i^2}\text{where,\hspace{0.17em}}\left[\begin{array}{l}{U}_{\max }=1forSD\hfill \\ U_{\max }estimatedforRD\hfill \end{array}\right]$$ (3)

Here ${\mathit{Occ}}_{\mathit{data}}$ is the occupancy estimate generated by the LPF, $\mathit{Occ}$ is as in equation (2), the index $i$ corresponds to each pre/post dose pair. A value of $\text{Φ}$ is minimized at each voxel.

We also calculated the voxel-by-voxel percent change in BP_ND ( $\text{Δ}{BP}_{ND}$ ) between the baseline and post-drug scans as follow:

$$\text{Δ}{BP}_{ND}=\frac{{BP}_{ND}^{\mathit{baseline}}-{BP}_{ND}^{\hspace{0.17em}{\mathit{post}}_{–}\mathit{drug}}}{{BP}_{ND}^{\mathit{baseline}}}$$ (4)

The $\text{Δ}{BP}_{ND}$ was used in place of occupancy in equations (1) to (3) to estimate EC₅₀ images for SD, RD and jointly fitting SD and RD data without using the LPF. These EC₅₀ images were used to compare with the results of EC₅₀ using LPF.

Statistical analysis

To compare EC₅₀ values across different regions of interest (ROIs) — namely, the caudate, putamen, and ventral striatum — as well as between the left and right hemispheres of each ROI, we performed unpaired t-tests as follow:

$$t_{\mathit{statistic}}=\frac{\left(\mathrm{ }{\overline{m}}_1-\mathrm{ }{\overline{m}}_2\mathrm{ }\right)}{\sqrt{\mathrm{ }\frac{{\mathit{var}}_1}{n_1}-\frac{{\mathit{var}}_2}{n_2}\mathrm{ }}}$$ (5)

where $\overline{m}$ is the sample mean of the group, $\mathit{var}$ is the variance, and $n$ is the number of independent observations (i.e., number of effective voxels). The degrees of freedom ( $df$ ) for each analysis was estimated using Welch-Satterthwaite approximation as:

$$df=\frac{{\left(\frac{{\mathit{var}}_1}{n_1}+\frac{{\mathit{var}}_2}{n_2}\right)}^2}{\left(\frac{{\left(\frac{{\mathit{var}}_1}{n_1}\right)}^2}{n_1-1}\right)+\left(\frac{{\left(\frac{{\mathit{var}}_2}{n_2}\right)}^2}{n_2-1}\right)}$$ (6)

To account for the spatial resolution and smoothness of the imaging data, the total number of number of voxels in each group was adjusted by dividing it by the resolution element ( $\mathit{Resel}$ ) to reflect an effective voxel size. The $\mathit{Resel}$ of the system was calculated as follow:

$$\mathit{Resel}=\frac{{\mathit{FWHM}}^{\mathrm{\hspace{0.17em}3}}}{{\mathit{voxel}}_{–}{\mathit{size}}^{\mathrm{\hspace{0.17em}3}}}$$ (7)

where $\mathit{FWHM}$ is the full width half maximum of the scanner, and ${\mathit{voxel}}_{–}\mathit{size}$ is the reconstructed voxel size in the image. The spatial resolution of the reconstructed PET images was 2 mm isotropic voxel size, as compared to the HR+ scanner resolution of 4.5 mm $\mathit{FWHM}$ at the center and 7.8 mm at r = 20 cm. ¹⁹ Furthermore, the reconstructed PET images were run through LPF with a kernel size of 3 × 3 × 3 voxels. To account for the scanner resolution and the smoothing by the LPF kernel, the $\mathit{Resel}$ of the system was calculated to be 125 (voxels), selecting a $\mathit{FWHM}$ of 10 mm to be conservative. The number of effective voxels (i.e., $n$ in equations (5) and (6)) for each group was calculated by dividing the voxel count in each ROI by 125. The adjusted voxel count was used to calculate degrees of freedom in each t-test, thereby accommodating for the reduced spatial resolution of the scanner. SLIC-Occ averages the EC₅₀ values in a cluster and assigns the average EC₅₀ to each voxel in the cluster. ¹² Since the occupancy data were clustered using SLIC-Occ, we used the cluster-wide EC₅₀ values to calculate the variance and the mean in equations (5) and (6).

The analyses were conducted separately for each pairwise comparison of EC₅₀ values across ROIs and hemispheres. Significance was evaluated based on p-values derived from the t-distributions. Bonferroni correction was applied for multiple comparisons. We performed 7 unpaired t-tests in each data set; therefore, α = 0.0071 was selected as significance level (rather than 0.05).

Results

Occupancy images

Figure 1 shows clustered occupancy images in the striatum at different plasma drug concentrations calculated by applying SLIC-Occ to voxel-level occupancy images generated by LPF for both SD and RD data. The drug occupancy in the striatum is increasing with increasing plasma drug concentration. Although the range of plasma concentrations for SD was slightly lower compared to the RD data, the SD protocol still appears to have achieved higher drug occupancies in the striatum.

Figure 1.

Axial view of the clustered occupancy images corresponding to different plasma concentrations for SD from te Beek et al. ¹⁴ (left) and RD from Schmidt et al. ¹³ (right) data.

EC₅₀ images

A 1-parameter E_max model (equation (1)) was used to estimate the EC₅₀ values at every cluster for both SD and RD and subsequently generate an EC₅₀ image for SD and RD occupancy data. Figure 2 shows the EC₅₀ image at multiple slices in axial and coronal view for SD and RD clustered occupancy data. Note the difference in scale between SD and RD EC₅₀ images. Figure 3 shows the EC₅₀ image for the joint-fit generated by combing both SD and RD. Clustering of SD and RD was performed separately using SLIC-Occ, but E_max fitting was performed simultaneously using the 2-parameter E_max model containing an upregulation term (equation (2)). Figure 4 shows the U_max image for RD data based on fitting all the data simultaneously using the upregulation model as indicated in equation (2) (aka a ‘joint-fit’).

Figure 2.

EC₅₀ images using 1-parameter E_max model fit (equation (1)) are shown for a) SD, and b) RD clustered occupancy data. Top rows show axial, and bottom rows show coronal views for multiple slices.

Figure 3.

EC₅₀ images using the joint-fit/upregulation model (equation (2)) with both SD and RD clustered occupancy data are shown in axial (top row) and coronal (bottom row) views for multiple slices as in Figure 2.

Figure 4.

U_max images for RD clustered occupancy data are shown in axial (top row) and coronal (bottom row) views for multiple slices based on joint-fit/upregulation model (equation (2)). Note that U_max for SD clustered occupancy data was set to 1.

Spatial variation in EC₅₀ values can be observed within the striatum (i.e., differences between putamen, caudate, and ventral striatum) in both SD and RD occupancy data. Figure 5 shows the EC₅₀ distribution of all the voxel values from clustered occupancy data within the whole striatum and its subregions (i.e., putamen in green, caudate in purple, and ventral striatum in gray) for SD, RD, and joint-fit. Overall, caudate and ventral striatum had lower EC₅₀ values (higher apparent drug affinity) compared to putamen. The putamen has a significantly higher EC₅₀ than caudate in both SD (p = 0.0067) and joint-fit (p = 0.005) data. The pattern of spatial differences is similar between the SD (equation (1)) data and the joint-fit (equation (2)). Whereas when the regional EC₅₀ values were compared in RD data, the pattern of spatial variation in EC₅₀ was not significantly different between any pair of ROIs.

Figure 5.

Histogram of EC₅₀ distribution from the clustered occupancy data for the whole striatum (i.e., putamen in green, caudate in purple, and ventral striatum in light gray) for SD (left), RD (middle), and Joint-Fit (right).

We also investigated left–right laterization of EC₅₀ within striatal ROIs. Figure 6 shows the EC₅₀ distribution of all the voxels from clustered occupancy data separated into left (shown in blue) versus right (shown in orange). The histograms of EC₅₀ for ventral striatum in both SD data (1-parameter fit) and joint-fit appear asymmetrical. However, this asymmetry was not statistically different after taking into account Resel size of the scanner and LPF smoothing. Asymmetry in the caudate for RD data seems pronounced but was not statistically significant. The mean and standard deviation of EC₅₀ values for whole striatum and its subregions separated into left and right for SD, RD, and joint-fit using upregulation model can be found in Table 1.

Figure 6.

Histograms of EC₅₀ values at every voxel from the clustered occupancy data for left and right putamen (left column), caudate (middle column) and ventral striatum (right column) shown for SD (top row), RD (middle row), and Joint-Fit (bottom row) data.

Table 1.

Mean (std) EC₅₀ values (ng/mL) of left and right striatum and its subregion (caudate, putamen, and ventral striatum) for SD, RD and joint-fit clustered occupancy data. p values are the result of unpaired t-test between each ROI or the left to right difference within the same ROI. Mean (std) of the $U_{\mathit{max}}^{RD}$ are also shown for the joint-fit analysis ( $U_{\mathit{max}}^{SD}=1$ ).

	Single-dose (n = 16)			Repeat-dose (n = 16)			Joint-fit (n = 16SD+16RD)
	Left	Right	p	Left	Right	p	Left	Right	p	$U_{\mathit{max}}^{RD}$
Putamen	16.2 (1.72)	15.2 (1.04)	0.286	28.7 (2.43)	29.3 (2.59)	0.653	16.4 (1.66)	15.3 (0.99)	0.191	1.59 (0.10)
Caudate	13.6 (1.26)	13.9 (1.21)	0.681	24.5 (3.23)	28.4 (2.15)	0.128	13.8 (1.14)	13.8 (1.29)	0.909	1.65 (0.19)
Ventral Striatum	13.3 (0.74)	15.4 (0.69)	0.325	27.6 (2.01)	28.1 (2.67)	0.676	13.4 (0.62)	15.7 (0.73)	0.304	1.63 (0.12)
Striatum	15.0 (2.02)	14.8 (1.22)	0.996	27.2 (3.29)	28.8 (2.51)	0.111	15.1 (1.98)	14.8 (1.30)	0.661	1.61 (0.14)
Statistics between ROIs
Put vs Caud			0.006			0.126			0.005
Put vs V. Str			0.168			0.444			0.209
Caud vs V. Str			0.501			0.746			0.584

Note: Values in italic and bold are statistically significant after correction for multiple comparison (p < 0.0071).

EC₅₀ images from $\text{Δ}{\mathrm{BP}}_{\mathrm{ND}}$

The 1-parameter E_max model (equation (1)) and the upregulation model (equation (2)) were also used to estimate the EC₅₀ images using $\text{Δ}{BP}_{ND}$ images of SD and RD at every voxel for SD, RD, and joint-fits. Figures S1, and S2 show the EC₅₀ images created from the $\text{Δ}{BP}_{ND}$ at same slices as in Figures 2 and 3 in axial and coronal view for SD, RD, and joint-fit. Similar spatial variations in EC₅₀ images calculated from the $\text{Δ}{BP}_{ND}$ were observed as in the EC₅₀ images generated using LPF.

Discussion

This study demonstrates that spatial variation in the apparent drug affinity of the D2 antagonist, JNJ-3782268, for dopamine receptors within the striatum is reproducible across independently executed occupancy studies. To back this claim, we compared results from the reanalysis of two different studies, one using a single drug-dose and the other a repeat dosing protocol.^13,14 Leveraging the LPF and SLIC-Occ clustering, we generated voxel-wise occupancy images at multiple plasma concentrations and then an EC₅₀ image for each study. The EC₅₀ images reveal regional differences in drug affinity regardless of experimental protocol. Different protocols (SD vs RD) required slightly different modeling approaches. Nevertheless, the consistency of our findings across datasets highlights the robustness of a voxel-based analysis of occupancy data in identifying spatial patterns in receptor binding that are ignored in any conventional whole-brain analysis of drug occupancy.

In occupancy studies, subjects are typically imaged both at baseline—when no drug is present—and following drug administration (whether as a single dose or multiple doses). Drug exposure can induce changes in receptor availability, such as receptor upregulation which means that the measured BP_ND at baseline is, in essence, invalid, and can no longer be compared directly to the post-drug BP_ND without correction. To address this discrepancy, Rabiner and Gunn ¹⁵ modified the conventional E_max model by incorporating a U_max term (see equation (2)) to correct for biases in occupancy measurements resulting from drug-induced receptor alterations. It is important to note that the U_max term does not have a discrete interpretation; rather, it reflects multiple factors influencing receptor dynamics after drug dosing, including but not limited to receptor upregulation, internalization or downregulation, changes in receptor isoforms, and potential modulation by endogenous dopamine tone. Thus, U_max represents the effective increase in the true BP_ND and reconciles the discrepancy between the baseline BP_ND and the BP_ND observed post-drug administration.

While the spatial variations in EC₅₀ were largely consistent across both SD and RD data, there was an initial discrepancy in the overall EC₅₀ values in the two datasets (SD and RD) when each was fitted separately using a 1-parameter model (equation (1)). The need to account for the effects of chronic drug dosing on occupancy data has already been established at the whole brain level by Rabiner and Gunn. ¹⁵ In the present study, we incorporated the effect of upregulation into the voxel-level E_max model and then applied it in a joint-fit of SD and RD data. In doing so, we showed that the SD and RD data show the same regional variations in EC₅₀ estimates. This adjustment of the E_max model to the experimental protocol emphasizes the importance of proper modeling of receptor dynamics in occupancy studies.

In this study, we introduce for the first time a U_max image (Figure 4) that captures changes in receptor dynamics following drug administration. While some voxels exhibited U_max values near or above 2, the average U_max across all striatal regions of interest was approximately 1.6, indicating a 60% upregulation. This finding is consistent with previous reports in animal experiments,^20–22 which demonstrated 30–60% upregulation of D2 receptors following antagonist administration. In the context of this paper, U_max represents the upregulation effect resulting from multiple doses; however, equation (2) and the U_max parameter could also be applied to estimate receptor downregulation, in which case U_max would range between 0 and 1.

In our analysis of EC₅₀ values across striatal subregions, we found a significant difference between the putamen and caudate (Table 1). Statistical analyses using unpaired t-tests confirmed this distinction. This result challenges the conventional assumption that a drug should exhibit uniform effective affinity across D2 receptors in the striatum. Previous studies using [¹¹C]-(+)-PHNO (a tracer with a 40-to-1 selectivity for D3 over D2 receptors) demonstrated zero to minimal binding changes in the putamen and caudate following administration of a D3 antagonist.^23,24 These findings suggest that in our study, which utilized raclopride (a tracer with approximately equal selectivity for D2 and D3 receptors) any observed binding, and thus displacement, is predominantly attributable to D2 receptors. Therefore, the significant variation in EC₅₀ between the putamen and caudate is unlikely to be explained by regional differences in D3 receptor contributions to binding but rather points to a potential difference in apparent D2 affinity across these subregions. Variations in D2 receptor isoforms (i.e., D2L and D2S) between regions may account for observed differences. If the JNJ-37822681 drug binds with different affinities to different D2 isoforms, this could result in regional variation effective affinity across the striatum. If confirmed, differences in receptor isoform expression and drug-affinity would be a proper avenue of investigation in future occupancy studies.

Differences between in vitro and in vivo affinity estimates must also be considered. Although, JNJ-37822681 has been shown to have a low affinity to D1 and D3, and other receptors, these affinities might be much different in vivo. ¹⁶ Kim et al. demonstrated that ¹¹C-LY2795050, a novel kappa-selective antagonist PET tracer, had 4.7-fold lower selectivity at kappa-opioid over mu-opioid in vivo compared to in vitro. ²⁵ Similarly, Slifstein et al. showed that the affinity of [¹⁸F]fallypride for D2 in vivo was substantially lower than in vitro. ²⁶ Such differences between in vitro and in vivo affinities of the JNJ drug might also play a role in our findings.

In this paper, we calculated EC₅₀ values using voxel-level occupancy estimated via the LPF and verified these results by calculating EC₅₀ values directly from $\text{Δ}{BP}_{ND}$ images. Consistent statistical findings from both methods confirmed that the putamen exhibited a higher EC₅₀ than the caudate, corresponding to a lower effective affinity for the JNJ-37822681 drug in the putamen. The similarity of results between the two approaches further validates our findings and underscores the robustness of the observed regional differences in effective affinity. Importantly, the LPF offers a significant advantage over $\text{Δ}{BP}_{ND}$ analysis in which it could also be applied to V_T images, provided an arterial input function is available. While the current study applies LPF to BP_ND images derived from a reference tissue model, the method was originally developed for V_T image and has been validated in that context.^10,11 This capability allows for voxel-level drug occupancy analysis in scenarios where traditional reference tissue models are not applicable, greatly expanding the versatility of the LPF method.

Previous work has explored the lateralization of dopamine receptors and neurotransmitter changes in the human brain.^23,27–32 Larisch et al. reported significantly higher D2 receptor binding in the right striatum than in the left using SPECT imaging in a cohort of 18 subjects (9 male). ³¹ Cho et al. identified a greater BP_ND in the right dorsal putamen compared to the left in a group of 25 healthy male subjects. ²⁸ Additional findings have been presented by Oberlin et al. showing a dopamine response specifically in the right ventral striatum to beer flavor among 28 male drinkers, ³² while Cosgrove et al. observed more dopamine activation specifically in the right ventral striatum of (8) male smokers while smoking a cigarette compared to (8) female smokers. ²⁹

In our study, we observed a lateralization of EC₅₀ values (although not significant after correcting for Resel size) within the striatum, specifically between the left and right ventral striatum, shown in Figure 6. The right ventral striatum exhibited higher EC₅₀ values than the left in SD data, a lateralized pattern that was also observed in all (RD and SD) data when both datasets were fitted jointly using an E_max model containing upregulation (equation (2)). Although, this lateralization wasn’t significantly different after correcting for Resel size, we saw a significant difference when the Resel size of 64 (FWHM = 8 mm) or the number of clusters was used as an independent measure in order to calculate the $df.$ Such an approach, although slightly less conservative than our Resel size estimate, is no less reasonable. The functional clusters behave independently across space and/or plasma concentration and so, are a reasonable way to assess the number of independent observations. While previous studies have primarily reported lateralization in receptor availability or dopamine response (e.g., BP_ND), the lateralization observed in our study reflects variation in apparent affinity (EC₅₀). These two measures capture different aspects of the dopaminergic system: receptor availability reflects binding capacity, whereas EC₅₀ reflects the dose required to achieve a given level of occupancy. Both measures may be influenced by receptor subtype composition, competition from endogenous dopamine, and off-target binding. Coupled with previous findings, our results suggest that asymmetries may extend beyond density to include receptor-ligand interactions. If confirmed, asymmetries in EC₅₀ would suggest possible noteworthy differences in receptor function across hemispheres. Further research is needed to uncover possible sources of lateralization, and its possible implications for drug development.

Occupancy studies provide valuable information to a drug-development team to make early decisions to proceed or not with the candidate drug into a larger and more expensive later phase study. ³ It is an accepted procedure to use EC₅₀ values to estimate the drug dose range for later phase studies. Conventionally, the whole brain drug dose is selected by estimating the EC₅₀ from all the region of the brain (for JNJ drug regions within striatum). However, if a smaller subregion (such as left, or right putamen or caudate) is the known target for the candidate drug, then it stands to reason that the EC₅₀ from that smaller subregion would be a more appropriate basis for dose selection.

A limitation of this study is the difference in design between the two cohorts. In the SD study, participants received a single drug dose ranging from 2 to 20 mg and were scanned at a consistent time post-administration. In contrast, the RD study involved administering the same drug dose (10 mg) across 13 doses, while the post-drug PET scans were performed at varying times following the final dose. We believe we have reconciled these differences via the E_max model. To be clear, our data-fitting approach assumed negligible receptor upregulation within the first few hours after the initial dose. This assumption was already established as valid by Rabiner and Gunn, but on the whole brain level. ¹⁵

Conclusion

This study successfully demonstrates the robustness of the LPF and SLIC-Occ clustering pipeline for generating reproducible EC₅₀ images from comparable but independent occupancy studies. We observed significant spatial variations and possible lateralization (left–right asymmetry) in effective drug affinity within the striatum for JNJ-37822681 in both studies. Our findings highlight the importance of selecting an appropriate E_max model that accounts for physiological receptor changes–such as upregulation or downregulation–that may occur with chronic dosing. Our findings should serve as a further prompt to researchers to explore regional variation in apparent drug affinity on drug dosing and drug development.

Ethics approval and consent to participate

This is a secondary analysis of data acquired under previous protocols which was approved by the ethical committees (See references 13 and 14).

Supplementary material

Supplemental material for this article is available online.