PET Quantification in Healthy Humans of Cyclooxygenase-2, a Potential Biomarker of Neuroinflammation

Abstract

Cyclooxygenase-2 (COX-2) is present in a healthy brain at low densities but can be markedly upregulated by excitatory input and by inflammogens. This study evaluated the sensitivity of the PET radioligand [¹¹C]-6-methoxy-2-(4-(methylsulfonyl)phenyl)-N-(thiophen-2-ylmethyl)pyrimidin-4-amine ([¹¹C]MC1) to detect COX-2 density in a healthy human brain. Methods: The specificity of [¹¹C]MC1 was confirmed using lipopolysaccharide-injected rats and transgenic mice expressing the human COX-2 gene, with 120-min baseline and blocked scans using COX-1 and COX-2 selective agents. Twenty-seven healthy participants were injected with [¹¹C]MC1. Ten of these participants received 2 PET scans: a baseline study followed by blockade with celecoxib (600 mg orally), a preferential COX-2 inhibitor. Seventeen participants underwent test-retest imaging. All scans included concurrent arterial sampling. The tissue-to-plasma ratio at equilibrium (i.e., total distribution volume) was determined using a 2-tissue compartment model (2TCM). Results: In humanized transgenic COX-2 mice, 70%-90% of [¹¹C]MC1 brain uptake was blocked by nonradioactive MC1 and celecoxib (a COX-2 selective inhibitor) but not by PS13 (a COX-1 selective inhibitor), thereby confirming specific binding to human COX-2. Radioactivity in the human brain peaked at a concentration of about 4.0 SUV, indicating good passage through the blood-brain barrier. Values for the total distribution volume achieved stability after 80 min, indicating no radiometabolite contamination. Celecoxib reduced radioligand binding in neocortical areas by 25% but had little or no effect in subcortical regions and the cerebellum, which correlated with COX-2 messenger RNA expression levels. Binding site occupancy by celecoxib was virtually complete, as determined by the Lassen plots. Test-retest reliability was moderate (intraclass correlation coefficient, 0.65) but had relatively large variability (absolute retest variability, 20%). Reference tissue methods yielded results comparable to those of 2TCM but reduced retest variability by up to 75% and reduced intersubject variability (coefficient of variation) by about half. Thus, compared with 2TCM, which requires arterial blood, the reference tissue method is expected to significantly reduce the sample sizes required to detect statistically significant differences between groups. Conclusion: [¹¹C]MC1 has adequate sensitivity to measure the low density of COX-2 in a healthy human brain, suggesting it can also quantify the COX-2 elevations expected in human disorders associated with neuroinflammation.

Keywords: COX-2; PET; human brain; neuroinflammation.

Graphical abstract

FIGURE 1.

Coronal views of COX-2 staining in rat brain tissue collected 24 h after intracranial administration of 50 µg of lipopolysaccharide (LPS; A) and in healthy rat brain tissue (B). 3-mm scale bar is provided for reference. (C–F) Magnified views (×50) of various cortical and subcortical regions from both LPS-injected and healthy rat brains, showing LPS-injected right striatum (C), control striatum (D), LPS-injected right cortex (E), and control cortex (F). 100-µm scale bar is provided for reference; arrows point to COX-2–positive cells. (G) Relative COX-2 expression determined by comparing regional total number of COX-2–positive cells to area of region of interest (ROI) and normalizing against that of left subcortical (L_Subcortical) control region. ROIs are illustrated as follows: left cortical (L_Cortical) in green, L_Subcortical in gray, right cortical (R_Cortical) in purple, and right subcortical (R_Subcortical) in yellow. Error bars represent mean ± SD (n = 6).

FIGURE 2.

(A) [¹¹C]MC1 uptake in hCOX-2 mouse brain displayed prominent radioactivity blockable with 0.3 mg/kg nonradioactive MC1 or 0.3 mg/kg celecoxib; in contrast, there was minimal uptake in wild-type (WT) mouse brain. (B) PET-derived whole-brain time–activity curves for WT and hCOX-2 mice after intravenous administration of [¹¹C]MC1, both at baseline and after administration of 0.3 mg/kg cold MC1 or celecoxib. (C) Elevated human PTGS2 expression was observed in hCOX-2 mouse brain, whereas mouse Ptgs2 expression was similar between WT and hCOX-2 mouse brains; humanized mice still expressed native COX-2. Relative PTGS2 expression was measured with β-actin as reference. Quantitative polymerase chain reaction analysis of PTGS2 expression in mice performed by Epigen DX. Error bars represent mean ± SD (n = 3 mice/group).

FIGURE 3.

(A and B) Test–retest studies of 17 healthy human volunteers. (C and D) Baseline and blocked scans (after celecoxib, 600 mg orally; n = 10). (A and C) Concentration of parent radioligand [¹¹C]MC1 in arterial plasma measured ex vivo with radio–high-performance liquid chromatography. (B and D) Concentration of radioactivity in occipital cortex measured with PET. Inserts show later points of brain time–activity curves, which tend to reflect receptor or enzyme density better than early points. Mean brain time–activity curves suggest that COX-2 density was slightly decreased on retest (B, insert) and more substantially decreased by celecoxib (D, insert). However, these mean values may be misleading, and more accurate quantitation requires individual analysis. Mean V_T was insignificantly affected on retest but was significantly reduced by ketoprofen in region-selective manner (Table 1). Unit of concentration is SUV. Symbols and error bars represent mean and SD, respectively.

FIGURE 4.

(A) Brain time–activity curves fit well with 2TCM but not with 1TCM. 2TCM assumes that radioligand has both high-affinity (specific) and low-affinity (nondisplaceable) binding in brain. Concentration in human occipital cortex at baseline was expressed as SUV. Symbols and error bars represent mean and SD (n = 10), respectively. (B) Representative PET images show distribution of radioactivity in healthy human brain. Selective COX-2 ligand celecoxib (600 mg orally) was taken 1 h before [¹¹C]MC1 blocked scan. Highest specific binding was observed in occipital cortex, insula, and prefrontal cortex. Conc = concentration.

FIGURE 5.

(A) Correlation between SRTM-derived BP_ND and V_T/f_P with Pearson coefficient of r = 0.78 (R² = 0.6). (B) Correlation between Logan-RTM–derived distribution volume ratio (DVR) and V_T/f_P with Pearson coefficient of r = 0.98 (R² = 0.97). (C) Correlation between SUVR, derived from 60–90-min PET frames, and V_T/f_P with Pearson coefficient of r = 0.92 (R² = 0.85). For all panels, dots represent regional averages for 10 baseline and 17 test scans from retest study.

FIGURE 6.

Analysis of time stability for V_T of radioactivity within 4 brain regions during baseline (A) and blocked (B) scans. V_T was determined using 2TCM and normalized to its terminal value at 120 min. Symbol and error bars denote mean and SD from 10 healthy volunteers, respectively.

FIGURE 7.

PET protein atlases for various [¹¹C]MC1 positron PET imaging metrics targeting COX-2. (A and B) V_T (A) and BP_ND (B) through volume-based analysis (left) and surface-based analysis (right) tailored to respect cortical geometry of brain to enhance precision. (C) Between-subject variability for total COX-2 uptake, quantified as coefficient of variation, contrasting volumetric analysis (left) and surface-based analysis (right). Surface-based analysis respects unique cortical geometry of individual participants (left), thereby reducing partial-volume effects. Moreover, surface-based approach mitigates increased between-subject variability observed using volumetric analysis, which is exacerbated by differences in anatomy that lead to edge or boundary effects. By respecting cortical geometry, surface-based analysis reduces between-subject variability in cortex, consequently lowering number of participants needed in group analysis to detect similar statistical significance.