Data from 62 individuals were analyzed (see Table 1). Forty-six participants who met criteria A for the diagnosis of PTSD underwent clinical screening using the Structured Clinical Interview for DSM-5 (SCID-5) [31] and the Clinician-Administered PTSD Scale for DSM-5 (CAPS-5) [32]. Based on the latter, n = 19 participants met the criteria for chronic PTSD diagnosis (“PTSD” group; more than 1 year since trauma), and n = 27 were exposed to trauma but did not meet PTSD diagnosis (“TC” group). In addition, 16 participants who did not meet criteria A and had no Axis I diagnosis were classified as healthy controls (“HC” group). Exclusion criteria included non-affective (major depressive disorder; MDD) major psychiatric or neurological illness, moderate/severe substance use, unstable psychotropic medication, significant bloodwork abnormalities or positive drug screen (participants were required to abstain from cannabis use for at least 28 days prior to the study), contraindications to MRI, pregnancy/lactation, and recent use of unapproved medications, for detailed exclusion criteria, please refer to the Supplementary Methods.Table 1

Demographic and clinical characteristics of study groups.

The study was approved by both Yale University (HIC#2000025067) and the VA Connecticut Healthcare System (HSS#IHR009) Institutional Review Boards. All participants gave informed consent and received monetary compensation for their participation. All the methods and procedures were performed in accordance with the relevant guidelines and regulations.

MRI data were obtained using a 3 T Siemens Prisma scanner at the Yale Magnetic Resonance Research Center (MRRC), equipped with a 32-channel receiver array head coil. High-resolution structural images were acquired via Magnetization-Prepared Rapid Gradient-Echo (MPRAGE) imaging (TR = 2.5 s, TE = 2.83 ms, FOV = 256 × 256 mm², matrix = 256 × 256 mm², slice thickness = 1.0 mm without gap, 160 slices, voxel size 1.0 × 1.0 × 1.0 mm³).

[¹¹C]OMAR, a radiotracer with high binding affinity and selectivity for brain CB1 receptors [33, 34], was synthesized using established protocols adapted for automated production on the GE TRACERlab FXC-Pro synthesis module (GE Healthcare, Milwaukee, WI, USA) [33]. Mean (SD) molar activity at the time of injection was 179 (113) MBq/nmol, and injected activity was 485 (147) MBq. Participants underwent dynamic PET scans using an HRRT scanner (Siemens Medical Systems, Knoxville, TN). Prior to each PET scan, a transmission scan was acquired with a ¹³⁷Cs point source for attenuation correction. PET emission data acquisition began with the administration of [¹¹C]OMAR via bolus injection (i.v.) over 1 min and continued for 120 min. Participant motion was captured with an optical system (Polaris Vicra, Northern Digital Incorporated, Waterloo, Ontario, Canada) positioned behind the PET scanner to record the position and orientation of an infrared reflective tool mounted to the subject’s head.

Radioactivity concentration in arterial whole blood was measured during the initial 7 min following [¹¹C]OMAR administration using either continuous measurement with an integrated peristaltic pump and radioactivity detection system (PBS101, Veenstra Instruments, Joure, The Netherlands) or rapid manual sampling. Discrete samples were manually drawn for all scans at specific time points (3, 5, 7, 10, 15, 20, 30, 45, 60, 75, 90, 105, and 120 min post-injection). Gamma counter measurements (Wizard 1480, PerkinElmer, Waltham, MA, USA) determined each sample’s whole blood and plasma radioactivity. This was then converted to concentration based on the sample weight and density. Additionally, samples collected at specific time points (5, 15, 30, 60, 90, and 120 min) were analyzed using column-switching high-performance liquid chromatography (HPLC) [35] to determine the fraction of unmetabolized radiotracer, as previously described [34]. The unmetabolized parent fraction was calculated as the ratio of radioactivity in fractions containing the parent compound to the total radioactivity collected, fitted with an inverted gamma function, and normalized by the time-varying extraction efficiency of radioactivity for the corresponding filtered plasma sample. Finally, the metabolite-corrected arterial plasma input function was obtained by multiplying the total plasma radioactivity concentration curve by the parent fraction curve on a point-by-point basis.

Dynamic list mode data were reconstructed with corrections for subject motion, attenuation, normalization, scatter, randoms, and dead time using an ordered subset-expectation maximization (OSEM) [36] algorithm (2 iterations, 30 subsets) histogrammed into 33 frames. Post-reconstruction software motion correction was performed on the dynamic images using a mutual-information algorithm (FSL-FLIRT version 3.2, Analysis Group, FMRIB, Oxford, UK) [37], employing frame-by-frame registration to an early summed image (0–10 min post-injection), which was also registered to the subject’s MR anatomical image (6-parameter affine registration). The subject-specific MR image was subsequently registered to the Anatomical Automatic Labeling (AAL) [38] atlas using a non-linear registration routine for the region of interest (ROI) delineation [39]. The final outcome measure was regional [¹¹C]OMAR volume of distribution (V_T), referred to as CB1R availability, because it is proportional to the number of CB1R available for [¹¹C]OMAR binding [40]. [¹¹C]OMAR V_T was estimated using the multilinear analysis-1 method (MA1) [41] with t* = 30 min. Whole brain CB1R availability was computed by creating a composite score from the four lobes and the cerebellum, weighted according to their relative sizes in MNI space, as defined by the AAL atlas.

We first assessed the association of CB1R availability ([¹¹C]OMAR V_T) with sex, age, and BMI in the whole brain and areas postulated to be at the core of PTSD (i.e., amygdala, hippocampus, and frontal cortex) using Bayesian robust regression. Sex, Z-transformed age, and Z-transformed BMI were used as predictors, and minimally informed priors were applied:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Intercept} \sim {Normal}(0,1)$$\end{document}Intercept~Normal(0,1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Sex} \sim {Normal}(0,1)$$\end{document}Sex~Normal(0,1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Age}}_{Z} \sim {Normal}(0,1)$$\end{document}AgeZ~Normal(0,1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{BMI}}_{Z} \sim {Normal}(0,1)$$\end{document}BMIZ~Normal(0,1)

To account for outliers, we modeled the outcome variable ([¹¹C]OMAR V_T) using a Student’s t distribution [42]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{T} \sim {Student}{\rm{\mbox{'}}}s\,t(\mu ,\nu ,\varepsilon )$$\end{document}VT~Student’st(μ,ν,ε)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu \sim {Inverse\; Gamma}(3,1)$$\end{document}ν~Inverse Gamma(3,1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon \sim {Exponential}(1)$$\end{document}ε~Exponential(1)

Next, we tested group effects on CB1R availability in the same brain regions using Bayesian robust regression analysis. Groups were coded as dummy variables, with healthy control (HC) males as the reference. Covariates included sex, Z-transformed age, and Z-transformed BMI. We also added interaction terms for each group with sex, following previous work [16]. Minimally informed priors were used:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Intercept} \sim {Normal}\left(0,1\right)$$\end{document}Intercept~Normal0,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${TC} \sim {Normal}\left(0,1\right)$$\end{document}TC~Normal0,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${HC} \sim {Normal}\left(0,1\right)$$\end{document}HC~Normal0,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Sex} \sim {Normal}\left(0,1\right)$$\end{document}Sex~Normal0,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Age}}_{Z} \sim {Normal}\left(0,1\right)$$\end{document}AgeZ~Normal0,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{BMI}}_{Z} \sim {Normal}\left(0,1\right)$$\end{document}BMIZ~Normal0,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${TC}* {Sex} \sim {Normal}\left(0,1\right)$$\end{document}TC*Sex~Normal0,1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${HC}* {Sex} \sim {Normal}\left(0,1\right)$$\end{document}HC*Sex~Normal0,1

As with the first analysis, V_T was modeled using a Student’s t distribution with the same priors:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{T} \sim {Student}{\rm{\mbox{'}}}s\,t(\mu ,\nu ,\varepsilon )$$\end{document}VT~Student’st(μ,ν,ε)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu \sim {Inverse\; Gamma}(3,1)$$\end{document}ν~Inverse Gamma(3,1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon \sim {Exponential}(1)$$\end{document}ε~Exponential(1)

For the symptom cluster analysis, CAPS scores from 44 participants (19 PTSD; 25 TC) were clustered based on the DSM-5 model and the more robust 8-factor model [30]. Using a similar model to the initial regression model, with Z-transformed age and BMI, and sex, using separate models, we estimated the contribution of each symptom cluster to the whole brain [¹¹C]OMAR volume of distribution (V_T).

A robust association was defined as one where the 89% Highest Posterior Density (HPD) interval for the slope did not include 0 [43–45]. All models converged with rHat < 1.01 and effective sample sizes > 1000. The analyses were conducted in Python 3.10.11 using ‘PyMC’ (v4.1.7) [46] and ‘ArviZ’ (v0.17.1) [47]. The No-U-Turn Sampler (NUTS) was employed for Markov chain Monte Carlo (MCMC) inference using default settings: 1000 draws, 1000 tuning steps, an 80% acceptance rate, and no thinning. Additional analyses, including NHPT, were performed with the ‘Pingouin’ package (v0.5.4). All code is available at https://github.com/KoremNSN/CB1rPTSD.

A robust regression analysis assessing the association between CB1R availability and age, sex, and BMI, independent of diagnosis/trauma exposure, revealed no evidence of such associations in the amygdala, hippocampus, frontal cortex, or whole brain composite score (See Table 2 for statistics).Table 2

Sex, Age, and BMI Beta Coefficients Contributions to [¹¹C]OMAR Volume of Distribution (V_T) Values in Regions of Interest.

A robust regression analysis assessing the relationship between CB1R availability and group revealed no evidence of a robust difference in CB1R availability between the groups. However, there was a robust interaction between sex and the TC group in the amygdala, showing increased CB1R availability in females, though this group consisted of only two individuals (see Table 3 for detailed statistics).Table 3

Group Level and Interaction Beta Coefficients Contributions to [¹¹C]OMAR volume of distribution (V_T) values in Regions of Interest.

Looking at associations between whole brain CB1R availability and symptom clusters, a robust regression analysis showed that only the Anhedonia/emotional numbing cluster was associated with CB1R availability (see Table 4 for detailed statistics).Table 4

Symptom Level Beta Coefficients Contributions to [¹¹C]OMAR volume of distribution (V_T) values in the whole brain.