Central Executive Network drives delta-9-tetrahydrocannabinol (THC)-induced nonlinear changes in large-scale functional connectivity in adolescent nonhuman primates

Abstract

Adolescent cannabinoid exposure has been implicated in enduring modifications to adult brain circuitry; however, well-controlled, systematic analyses investigating dose-dependent effects of chronic delta-9-tetrahydrocannabinol (THC) exposure on brain connectivity are lacking. It is hypothesized that large-scale intrinsic networks, such as default mode (DMN), central executive (CEN), and salience networks (SN), are critically involved in vulnerability to deficits in cognitive processing often associated with adolescent cannabis use. The present study aimed to elucidate the effects of chronic THC exposure on functional connectivity (FC) of these putative large-scale networks in nonhuman primates. Separate groups of adolescent squirrel monkeys (aged 2.0-yrs [female] and 2.5-yrs [male]) were administered intramuscular injections of vehicle or THC daily (0.32 or 3.2mg/kg) for 6-months during adolescence. Resting state functional connectivity from scans conducted in awake subjects was measured before dosing, at 6-months of chronic dosing, and 60-days following discontinuation of daily THC exposure. Utilizing two distinct analytical methodologies, we observed a non-linear, dosage-dependent alteration in DMN-CEN FC across scan intervals. Specifically, exposure to a low THC dosage increased FC during chronic exposure compared to both the pre-dosing and discontinuation periods. This pattern, however, was not observed in either the vehicle or high THC dosage groups. Dual-regression unveiled a similar non-linear effect within the CEN, but not DMN, suggesting the effect on DMN-CEN FC may be driven by modifications within the CEN. Taken together, these results suggest adolescent THC exposure differentially affects large-scale brain networks and contributes to a nuanced understanding of CEN’s role in disrupting brain connectivity following chronic THC exposure.

Introduction

The enduring effects of substance use during adolescence, especially the use of cannabis and its constituents, have sparked extensive debate and scholarly inquiry, with particular focus on the neurobiological consequences during this pivotal developmental phase [1–7]. Adolescence marks a time of intense brain maturation and heightened sensitivity to exogenous agents. Previous research has suggested that the interaction of delta-9-tetrahydrocannabinol (THC), the primary psychoactive constituent in cannabis, with the maturing endocannabinoid system may lead to persistent changes in brain connectivity and function. Research has shown that adolescent use of cannabinoids can have persistent effects on the brain that last into adulthood, especially in regions linked to higher-order functions and emotional regulation, such as the prefrontal cortex, amygdala, and hippocampus [1–7]. Such structural and functional changes have been linked to cognitive and emotional disturbances, including deficits in attention, learning, and self-control, as well as increased anxiety and increased risk of mood or other psychiatric conditions [8]. Moreover, studies have established a link between early cannabis use and increased probability of developing substance use disorders (SUD) [9–12]. Nevertheless, the intricate nature of these relationships, especially considering the availability of cannabis products with varying THC concentrations, remains elusive. There is a critical need for detailed research to delineate the subtle differences between use and misuse and to understand the full impact of cannabis on the maturing brain. Detailed investigations into the neurophysiological pathways triggered by adolescent cannabinoid exposure are required, with particular attention to the potential dose-dependent effects on neural development into adulthood.

Resting state functional connectivity (rsFC) analyses are useful for investigating the underlying neurophysiological mechanisms of adolescent exposure to THC due to its non-invasiveness which supports longitudinal designs with high translational relevance. Human studies employing rsFC analyses have been utilized to decode the neural substrates of SUD, with findings suggesting impact on key brain regions such as the amygdala, hippocampus, and various prefrontal areas [13, 14]. Collectively, these regions are components of larger neural networks such as Default Mode Network (DMN), Central Executive Network (CEN), and Salience Network (SN), each with distinct roles in cognitive and emotional processing. The DMN, for instance, has been implicated in introspection and self-referential thoughts, while the CEN is associated with higher-order cognitive functions that require focused attention. The SN is crucial for detecting and responding to emotionally salient stimuli [15–17]. The interaction between these networks, often referred to as the tripartite network [18, 19], is crucial for maintaining cognitive function and responding to environmental demands (Fig. 1A).

Fig. 1. Setup and pipeline for analyzing functional connectivity differences across THC exposure groups.

A Effects of THC exposure on tripartite network connectivity. The present study aimed to characterize longitudinal changes in the NHP analogs of tripartite (DMN, CEN, SN) brain network functional connectivity (FC) in adolescent squirrel monkeys exposed to chronic THC and after a period of discontinuation. B Scanning Timeline. Total of three fMRI scans were performed for the study. Shortly after acclimating subjects to awake MRI procedures, a baseline scan was conducted prior to THC or vehicle administration in all three treatment groups. After 6 months of THC daily dosing, subjects were scanned (Chronic Scan) to assess the influence of chronic THC treatment. Sixty days after treatment discontinuation, subjects were again scanned (Discontinuation Scan) to assess the effect of acute discontinuation. C fMRI analysis pipeline. Additional details described in text.

Alterations in human tripartite dynamics, particularly those involving the DMN, have been noted in SUD and a variety of psychiatric conditions such as depression, anxiety, and schizophrenia [20–22]. For instance, individuals with depression often exhibit DMN hyperconnectivity, linked to rumination and negative self-referential thoughts [20]. Those with anxiety disorders may show disrupted connectivity affecting emotional regulation and threat perception [21]. In schizophrenia, abnormal network connectivity is associated with cognitive control deficits [22]. In the context of SUD, cocaine- and heroin-dependent individuals have been shown to exhibit reduced connectivity within these networks with links to impaired self-control and increased vulnerability to relapse [14, 16, 23–25]. Additionally, disrupted connectivity patterns have been observed in alcohol, cannabis, and opioid use disorders, highlighting unique and shared alterations in network interactions that may exacerbate deficits in inhibitory control and emotional regulation [26, 27], potentially increasing vulnerability to relapse. Interestingly, acute drug administration appears to induce contrasting rsFC patterns compared to those seen in abstinent individuals, suggesting that the brain’s response to drugs is not only complex but also dynamic, varying over time and with the state of drug exposure [28–30]. These dynamics involve transient increases in connectivity within the SN and CEN during acute administration, followed by reductions during withdrawal and abstinence [28]. However, not all substances share pharmacological mechanisms of action and the relevance of these findings to adolescent cannabis use - which has received less attention - may be limited. Further research is necessary to explore how adolescent THC exposure alters tripartite dynamics, as this may provide a better understanding of how functional connectivity between large-scale networks in the context of adolescent cannabis use may increase vulnerability to cognitive deficits or psychiatric conditions in adulthood.

The present study was designed to begin to address gaps in our understanding of how different levels of THC exposure affect brain FC over time, employing seed-based and ICA-based methods to assess putative nonhuman primate (NHP) tripartite (DMN, CEN, SN) networks using a highly translational NHP model of adolescent drug exposure. While the seed-based analysis is based on the manual selection of network regions-of-interest (ROIs), ICA is data-driven, which has advantages with respect to minimizing assumptions and for cross-validating against seed-based approaches. The close homology between NHPs and humans in neurobiological, genetic, pharmacokinetic, and pharmacodynamic characteristics [31–34] has established NHPs as a gold-standard model for studying both cannabinoids and neurodevelopment [35–39]. Their complex and highly evolved neural systems, which include similar cortical structures and connectivity patterns [29, 30, 40, 41], provide a means for investigating the effects of cannabinoids on brain development with greater translational relevance compared to other animal models. Moreover, the prolonged developmental timeline of NHPs, which mirrors key stages of human neurodevelopment—including critical periods of synaptic pruning and neural circuit maturation [42]—offers a unique opportunity to explore how early-life exposure to cannabinoids, such as THC, impacts the trajectory of brain development. These models are particularly valuable for understanding how cannabinoid exposure during sensitive developmental windows [43] may alter brain network connectivity.

Methods

Subjects

Twenty-two adolescent squirrel monkeys (Saimiri sciureus, 33.3 ± 1.6 months; 11 F) were randomly assigned to one of three groups to receive daily intramuscular (IM) injections of vehicle (V, n = 7, 3 F), low dose (LD, 0.32 mg/kg/day, n = 8, 4 F) or high dose (HD, 3.2 mg/kg/day, n = 7, 4 F) THC for six months (described below). The experimental protocol was approved by the Institutional Animal Care and Use Committee at McLean Hospital in a facility licensed by the US Department of Agriculture and conducted in accordance with guidelines provided by the Committee on Care and Use of Laboratory Animals of the Institute of Laboratory Animals Resources, Commission on Life Sciences.

Drug treatment

Subjects were treated intramuscularly with vehicle, a low dose (0.32 mg/kg), or a high dose (3.2 mg/kg) of THC once-daily at approximately 11am, 7 days/week over a 6-month period [44]. To minimize potential discomfort that might have followed initial injections of the high dose, subjects in that group were administered THC in progressively increasing doses: 0.32 mg/kg for three days, followed by 1.0 mg/kg for the next three days, before receiving the first dose of 3.2 mg/kg. A high dose of 3.2 mg/kg/day THC was selected as a translationally relevant dosage to model heavy cannabis use in adolescents. This dosage reflects real-world consumption patterns, given that an average joint contains approximately 300–400 mg of cannabis with a THC concentration of 20% in frequent users, resulting in about 80 mg of THC per joint. Accordingly, a dose of 3.2 mg/kg/day in monkeys approximates the THC intake of a heavy adolescent user consuming 1.5 joints per day. Δ⁹-tetrahydrocannabinol was provided by the NIDA Drug Supply Program (Rockville, MD) and prepared for intramuscular administration in a 20%:20%:60% vehicle mixture of 95% ethanol:Tween-80:0.9% saline. Vehicle-treated subjects received a single injection of 20:20:60 mixture of 95% ethanol, polysorbate-80 (Tween-80; sigma-Aldrich, St Louis, MO) and saline each day. Subjects were not engaged in any other experimental procedures during the 6-month period of drug treatment and 2-month discontinuation phase. Figure 1 shows additional details of the experimental timeline.

MRI data acquisition and preprocessing

Extensive behavioral training was used to acclimate subjects to the MR procedures as described previously [40, 43, 45, 46]. Images were acquired in awake subjects with a 9.4 Tesla/400 mm diameter MR system (Varian Direct Drive, Varian Inc, Palo Alto, CA). The system has a 116 mm inner diameter gradient with a maximum strength of 45 G/cm; a single loop head-encompassing transmit-receive RF coil was used. Scan sessions included automated image-based shimming <75 Hz. fMRI images were acquired with a whole-brain gradient-echo planar imaging (EPI) sequence with parameters: TE = 8 ms, TR = 1500 ms, and flip angle=90°; scan matrix was 64 on a 64 mm field of view with 54x1mm slices; scan time for 1200 volumes was 30-min. Distortion-matched anatomic images were acquired with a spin-echo EPI sequence (TE = 17.5 ms, TR = 1500 ms, flip angle=90°, averages=8; scan matrix=64 on a 64 mm field of view with 54x1mm coronal slices matched to the fMRI slices).

Data were visually checked, and quality control was evaluated with the MRI Quality Control tool (MRIQC [47]). Intensity spiking was analyzed and adjusted for with a threshold of 1mm framewise displacement (https://github.com/bbfrederick/spikefix). Preprocessing was conducted using FMRIB’s Software Library (FSL, Oxford University, UK) and consisted of removing the first 10 volumes from each scan to allow for data stabilization and head motion correction using MCFLIRT tool [48] in FSL; 12-motion correction positions were used as nuisance regressors. Spatial smoothing was conducted using a Gaussian kernel of 2.0 mm FWHM and temporal filtering was applied using a high pass filter with a 100 s cutoff (0.01 Hz). Session averaged functional volumes were aligned to the VALiDATe [49] T2w template using a 12 DOF affine transformation followed by adjustment of nonlinear distortion fields (jip analysis toolkit; www.nitrc.org/projects/jip).

Seed-based approach

Mapping of region-of-Interest (ROI) seeds to brain network voxels

Spatial maps for the networks of interest were generated by placing 1x1x1 mm³ spherical seeds bilaterally at the posterior cingulate cortex (PCC) and prefrontal cortex (PFC) for the DMN [15], the dorsolateral prefrontal cortex (dlPFC) and lateral posterior parietal cortex (PPC) for the CEN [16], and the dorsal anterior cingulate cortex (dACC) and anterior insula for the SN [17]. Homologous brain regions have been documented across humans and various NHP species, indicating a high degree of conservation in both structure and function [50–54]. Research has shown that key nodes of the DMN and their connectivity patterns in humans were also observed in NHP [32, 55], including squirrel monkeys [40]. While the DMN’s seed locations draw directly from comparative neuroimaging evidence, seeds for the CEN and SN networks were chosen through an extrapolative approach. Despite a relative paucity of direct evidence for these networks in NHPs, we leveraged existing evidence suggesting a high degree of conservation of brain networks across humans and NHPs.

In order to generate representational network masks from the seeds, we first computed a set of correlation maps of each voxel’s time series correlated with the average time series of a given ROI seed (AFNI 3dNetCorr) at the individual subject level. To produce network masks that are representative of the group of subjects, a group-level analysis using AFNI 3dLME which specified a multivariate random intercept random slope regression model using the test condition (Baseline, Chronic, and Discontinuation), and sex (male, female) as two categorical variables; motion and age as quantitative variables, was conducted. The model formula was “session*age + motion + sex”, with random slope conditions for age. The resulting brain masks were thresholded at the level of p = 5e-17 for the most significant correlations and flattened in binary scale. This corresponds to an AFNI q-value of q < 0.05, an FDR correction method that works along with the conventionally used p-value (voxel-wise correction) specifically for multiple comparison testing. This stringent threshold was used to minimize overlap between masks of the three networks. While some overlap among the masks was detected using the AFNI 3dcalc command, the results remained consistent regardless of whether the full mask or a mask excluding overlapping areas was used. Formal analyses were conducted using the full mask for each network, as generated from the seed-based analysis.

Multivariate GLM analysis for ROI-ROI FC analysis

Once network masks were created, FC was probed between the three brain networks by performing spatial and temporal correlational analyses (CONN Toolbox). Seed-based masks were used as seed ROIs within CONN toolbox [56] and Matlab scripts (Mathworks, Natick, MA) to calculate ROI-to-ROI FC. The ROIs employed for analyses included the DMN, CEN, SN masks, as well as the anatomical gray matter, white matter, and ventricle masks which were used as nuisance regressors. Confounding effects from white matter and CSF regions, regression of motion covariates, as well as constant and linear BOLD signal trends within each session, were estimated and regressed out using aCompCor [57], an anatomical principal component analysis-based noise correction procedure for denoising, using default parameters. Finally, the standard error (SE) of the beta coefficients was computed using the equation $SE\left(\beta \right)=\frac{\beta }{T}$, where β is the beta coefficient and T is the corresponding t-statistic, based on the Ordinary Least Squares (OLS) solutions of CONN Toolbox.

Large-scale network FC comparison across THC treatment groups

Z-transformed correlations were used as measures of ROI-to-ROI FC between the seed-derived tripartite brain masks (CONN toolbox). The raw Z-transformed correlations between the three networks were analyzed across test groups and scan timepoints to explore systematic differences across conditions. To better assess the amount of change from each group’s baseline to chronic and treatment discontinuation timepoints, change in FC strength between the scans was quantified. This was to ensure that the FC at the chronic or discontinuation phase were aligned to their respective baseline levels and to better evaluate the effect of THC from one scan to the next. To account for and regress out the effects of age and sex covariates, we performed a GLM analysis to isolate the effect of THC on the FC changes across the scanning timeline. Hypothesis testing for the GLM model was evaluated based on the Wilks’ Lambda statistics defined in the context of a Likelihood Ratio Test.

ICA-based approach

Preprocessed baseline fMRI scans to group-level probabilistic ICA

To enhance the rigor and reproducibility of the seed-based result, group-level probabilistic ICA maps from the preprocessed baseline fMRI scans were generated to identify clusters that are separated as independent components. Several model orders were tested in the ICA analysis spanning across 10, 12, 15 and 20 components (Melodic in FSL) [40]. Testing different numbers of components for the ICA was conducted to explore the large-scale networks, see [40], and to cross-compare with the seed-derived brain masks to best align the maps between the two analytical approaches. Spatial maps of components identified in ICA were then correlated (fslcc) with seed-derived masks to identify the highest correlated component mask for each network. The highest correlation for the DMN was found in ICA 15 (r = 0.22), CEN in ICA 20 (r = 0.45), and SN in ICA 20 (r = 0.14). The resulting highest correlated components were visually identified and determined to be consistent with previously reported tripartite networks in humans and rhesus monkeys [31–34, 41]. A table of component-wise correlations is shown in Supplementary Table 1.

Dual regression to determine differences between network FC across THC treatment groups

Dual regression was conducted in two stages. In the first stage, subject-specific time series were found by regressing the group network maps into each subject’s 4D fMRI data (spatial regression). In the second stage, subject-specific time series were regressed back into the 4D fMRI data to find the subject-specific spatial maps (temporal regression). Randomise (FSL) was then used to identify clusters of voxels indicating differences across test groups (V, LD, HD) and scan timepoints (baseline, chronic, discontinuation). Specifically, 6 contrast conditions: baseline > chronic, baseline > discontinuation, chronic > discontinuation, chronic > baseline, discontinuation > baseline, and discontinuation > chronic were tested. Therefore, for each of the three networks (DMN, CEN, SN), and for each of the three test groups (V, LD, HD), six spatial maps showing the contrast between scan timepoints were produced. All contrasts were assessed based on two-sample unpaired t-test using 5000 permutations and under threshold-free cluster enhancement (tfce) at the alpha level of p = 0.05. Notably, the dual regression was performed using two different input types. To ensure that any significant result obtained from this analysis was not limited to the group ICA, we ran a second round of dual-regression analysis using the seed-based masks as input.

Results

Seed-based approach

The FC strength at the three scan timepoints (Baseline, Chronic, Discontinuation) was initially compared across all treatment groups (V, LD, HD) for the DMN-CEN, DMN-SN, and CEN-SN network pairs. This analysis revealed no significant differences in baseline FC between the three groups for any of the network pairs (see Supplementary Fig. 2 and Supplementary Table 2).

Further analysis of FC strength across the three test conditions within each treatment group showed a significant increase in FC strength between Baseline and Chronic conditions in the DMN-CEN FC for the LD group (T(20) = 2.99, two-sided p < 0.01). No significant differences were observed for the other conditions across the three FC pairs (Fig. 2 and Supplementary Fig. 2).

Fig. 2. Region-of-Interest (ROI) Seeds to Brain Network Voxels Mapping.

Spherical seeds (1x1x1 mm³) were placed bilaterally in PCC and PFC for DMN, dlPFC, and PPC for CEN, and dACC and Anterior Insula for SN. Using AFNI 3dNetCorr, the selected ROIs were then used to generate a brain mask of correlated regions at an individual level. Then, AFNI 3dLME command was used to run a group-level analysis with multivariate random intercept random slope regression model using the test condition (Baseline, Chronic, and Discontinuation), and sex (male, female) as two categorical variables, and motion, and age as quantitative variables. The model formula was “session*age + motion + sex”, with random slope conditions for age. The resulting brain masks were then thresholded at the level of p = 5e-17 for the most significant correlations and flattened in binary scale. AFNI 3dcalc command was used to parse out non-overlapping regions of DMN, CEN, and SN to be used for further correlational analyses in CONN toolbox.

To evaluate how THC exposure influenced FC strength between test conditions, differences were calculated for each condition across treatment groups. For the Chronic-Baseline difference in DMN-CEN FC, a statistically significant difference in GLM coefficients was observed between (1) the V and LD groups (T(19) = 2.39, two-sided p = 0.027) and (2) the LD and HD groups (T(19) = 2.61, two-sided p = 0.017), with the LD group exhibiting elevated FC compared to both the V and HD groups. Additionally, under the Discontinuation-Chronic condition for DMN-CEN FC, a significant difference was found between the LD and HD groups (T(19) = 2.14, p = 0.045), with HD showing elevated FC compared to LD (Fig. 3, Supplementary Table 2.1). No significant differences were identified for other conditions in the DMN-CEN FC. Similarly, there were no significant differences between treatment groups across scan conditions for the DMN-SN or CEN-SN network pairs.

Fig. 3. Seed analysis shows that DMN-CEN FC significantly changes based on timepoint and dosage of THC.

A Multivariate GLM analysis for ROI-ROI static FC analysis. To account for respective baseline levels, change in FC strength between the scans was quantified. When examining the chronic-baseline condition for the DMN-CEN FC, there was a statistically significant (T(19) = 2.29, p < 0.05, two-sided t-test) difference of GLM coefficient between (1) V and LD, and (2) LD and HD respectively, showing that the LD had an elevated FC level compared to both the V and HD. No significant differences for any conditions for DMN-CEN FC were identified. Similarly, there were no significant differences between the test groups across scan conditions for the DMN-SN or CEN-SN. Notably, we found a nonlinear THC dosage effect on DMN-CEN FC when examining the GLM coefficient trends across scan timepoints (Baseline, Chronic, Discontinuation) between the LD and HD. For the LD, FC followed an inverted-U shape across the three scan timepoints (Baseline, Chronic, Discontinuation), with a positive FC change from Baseline to Chronic, and a negative FC change from Chronic to Discontinuation scan timepoints. Meanwhile, the HD showed the opposite FC trend, with a negative FC change from Baseline to Chronic, and a positive FC change from Chronic to Discontinuation scan timepoint.

Notably, a nonlinear THC dosage effect on DMN-CEN FC was identified when examining the GLM coefficient trends across scan timepoints (Baseline, Chronic, Discontinuation) between the LD and HD groups (Supplementary Fig. 2). For LD, FC followed an inverted-U shape across the three scan timepoints (Baseline, Chronic, Discontinuation), with a positive FC change from Baseline to Chronic (T(20) = 2.99, two-sided p < 0.01), and a negative FC change from Chronic to Discontinuation (trend toward significance; T(20) = 2.02, two-sided p = 0.056) scan timepoints. HD showed opposite FC trends, with a negative FC change from Baseline to Chronic (not significant; T(20) = 1.52, two-sided p = 0.14), and a positive FC change from Chronic to Discontinuation (not significant; T(20) = 1.71, two-sided p = 0.10) timepoints (Supplementary Fig. 2). This THC dosage effect is not only reflected by the FC levels at transitions between scan timepoints, but also based on the directional differences at each scan timepoint between the LD and HD (Supplementary Fig. 2). LD showed positive FC at Chronic, and a negative FC at Discontinuation scan timepoint, whereas the HD had a negative FC at Chronic, and positive FC at Discontinuation scan timepoint. The nonlinear THC dosage effect found in DMN-CEN FC was not evident in DMN-SN or CEN-SN FC.

ICA-based Approach

Based on the significant FC difference from the seed-based analysis between the LD and both the V and HD groups in DMN-CEN FC, we sought to determine whether a specific network was driving these connectivity changes under THC exposure. Using the group ICA as input to dual-regression analysis, it was found that the CEN brain mask yielded significant contrast (p < 0.05) supporting Chronic>Discontinuation, Chronic>Baseline, and Baseline>Discontinuation for the LD group only (Fig. 4). Similarly, using the seed-based network masks as input to the dual-regression, CEN brain mask yielded significant contrast (p < 0.05) supporting Chronic>Discontinuation, Chronic>Baseline (data not shown), and Baseline>Discontinuation for the LD group only (Fig. 4). Meanwhile, no significant trend was found for the DMN with either the group ICA or seed-based mask for dual-regression input. The dual-regression found a similar non-linear effect found from the seed-based analysis but only within the CEN and not DMN (Fig. 4). The significant contrasts found from the CEN dual-regression were localized primarily in the cerebellum (both the group-ICA and seed-mask based dual-regression) and prefrontal regions (group-ICA based dual-regression).

Fig. 4. ICA-based dual regression suggests that alterations to the DMN-CEN FC in the Low THC group is driven by the CEN but not DMN.

Based on the significant FC difference from the seed-based analysis between the LD against the V and HD in DMN-CEN FC, we further explored whether a particular network was primarily driving the changes in this connectivity under THC exposure. Using the group ICA as input to the dual-regression analysis, we found that the CEN brain mask yielded significant contrast (tfce, thresholded at p < 0.05) supporting A Chronic > Discontinuation, Chronic > Baseline (not shown in Fig. 4), and B Baseline > Discontinuation for the LD group only. Similarly, using the seed-based network masks as input to the dual-regression, CEN brain mask yielded significant contrast (tfce, thresholded at p < 0.05) supporting Chronic > Discontinuation, and Baseline > Discontinuation, but not for Chronic > Baseline in the LD group only. Meanwhile, no significant trend was found for the DMN with either the group ICA or seed-based mask for dual-regression input. Hence, the dual-regression analysis exhibits a similar non-linear effect found from the seed-based analysis but only within the CEN, and not within the DMN. The significant contrasts found from the CEN dual-regression were localized primarily in the cerebellum (both the group-ICA and seed-mask based dual-regression) and prefrontal regions (group-ICA based dual-regression).

Discussion

The present study investigated the effects of chronic THC exposure on rsFC of large-scale neural networks in adolescent squirrel monkeys. Specifically, we probed interactions among the putative large scale DMN, CEN, and SN brain networks, which have been shown to play distinct, yet interacting, roles in cognitive and emotional processing and have been implicated in several psychiatric conditions in humans [20–22]. The results reveal dose-dependent alterations in DMN-CEN functional connectivity (FC) across different fMRI scan timepoints, with one of the major findings being the non-linear effect of THC dosage on DMN-CEN FC observed across the three timepoints: Baseline, Chronic, and Discontinuation. For the LD group, changes in FC followed an inverted-U shaped pattern, increasing from Baseline to Chronic and decreasing from Chronic to Discontinuation. Conversely, the HD group exhibited a decrease from Baseline to the Chronic treatment condition which then increased from Chronic to Discontinuation. Dual-regression analysis unveiled a similar non-linear effect within the CEN, but not DMN, suggesting the non-linear dosage effect on DMN-CEN FC may be driven by modifications within the CEN. Notably, there were no significant changes in DMN-SN or CEN-SN FC across scan conditions.

The results of the present study highlight a distinct non-linear response to daily treatment during adolescence with different THC dosages. Notably, LD (0.32 mg/kg/day) resulted in increased FC between DMN and CEN during the treatment phase, which was not observed with HD (3.2 mg/kg/day) of THC. Additionally, transitioning from chronic administration to treatment discontinuation resulted in decreased FC between DMN and CEN in LD THC-treated subjects, while HD THC treatment produced increased FC. The changes in CEN FC with LD THC may be a compensatory mechanism, potentially enhancing the brain’s ability to integrate information across disparate neural regions. The compensatory effects of THC on brain function, and the differential effects of various dosages, are well-documented. For example, a low dose of THC was found to promote hippocampal neurogenesis, prevent neurodegeneration, and improve cognitive function in aged rats, while higher doses impaired short-term memory and cognitive performance [58]. Another study investigated the long-term effects of chronic cannabis use on brain function related to cognitive control [59]. Using fMRI and the Multi-Source Interference Task, cannabis users (more than 10 years of daily or near-daily use) were compared with non-users and no significant differences were found in task performance or brain activation magnitude between the groups. However, cannabis users exhibited greater connectivity between the prefrontal cortex and occipitoparietal cortex as cognitive control demands increased. This was positively associated with age of onset and lifetime cannabis exposure, suggesting compensatory mechanisms that may mitigate cannabis-related cognitive impairments. Acute administration of THC has also been shown to dose-dependently increase cerebral blood flow (CBF), especially in regions such as the anterior cingulate cortex, frontal cortex, and insula [60]. However, with chronic use, cannabis produces a reduction in CBF, particularly in the prefrontal cortex. This reduction is thought to reflect the development of tolerance, where the brain adapts to the continuous presence of THC by downregulating its response. This compensatory response may initially help preserve cognitive function but could lead to exhaustion of neural resources or destabilization of network interactions over prolonged exposure periods. The dynamic and dose-dependent nature of THC’s effects demonstrated in the present study, particularly during the critical period of adolescent brain development, provides potential targets for future clinical and preclinical studies. Overall, these results suggest that THC may exert differential effects depending on dose, which could be pivotal for developmental neurobiology, recreational use, and medicinal applications.

The present findings suggest that CEN-DMN FC, particularly within the CEN, was disrupted. Prefrontal-subcortical circuits that are part of the CEN and DMN undergo dynamic maturation during adolescence [43, 61, 62], and previous studies indicate that THC disrupts the development of neural circuits typically active during this period. For example, structural MRI studies in human adolescents have documented reductions in cortical thickness in prefrontal regions [63, 64] while fMRI studies show altered connectivity patterns in the brains of cannabis users, particularly decreased connectivity in the orbitofrontal cortex [65] and increased cerebellum connectivity [66], which may underlie changes in cognitive control and emotional regulation, respectively. The impact of THC on synaptic pruning may be a key component of THC’s enduring effects on adolescent brain structure and function. Indeed, previous studies have already shown that THC’s interaction with cannabinoid receptors disrupts natural pruning processes [67], leading to either excessive pruning or insufficient elimination of synaptic connections [68]. In human cannabis users, Blest-Hopley et al. [69] reported reduced activation in regions associated with the CEN, such as the dorsolateral prefrontal cortex, while conversely, increased activation in DMN regions, including medial prefrontal cortex and posterior cingulate cortex. Similarly, Ertl et al. [70] found no differences in DMN connectivity associated with cannabis use in adolescent and young adults, but cannabis use was associated with increased connectivity in CEN regions (dorsal anterior cingulate cortex, middle and posterior insula, temporo-parietal junction, and superior temporal gyrus). However, whereas the present findings suggest no significant effects with respect to SN, it should be noted that studies in humans have found altered FC in SN-CEN or SN-DMN [71, 72]. Taken together, these findings suggest that THC could act as a trigger or contributing factor in the development of deficits in CEN or CEN-DMN-dependent processes, particularly when exposure occurs during the vulnerable period of adolescence.

The regional effects of THC and its alterations in functional connectivity (FC) may, in part, be due to the differential cannabinoid receptor (CB1) expression patterns in the brain. CB1 receptors are highly expressed in regions critical for cognitive and emotional processing, such as the hippocampus, prefrontal cortex, and amygdala, with dense localization in inhibitory axon terminals, which significantly influence synaptic signaling and plasticity [73]. The prefrontal cortex, a region essential for executive functions and decision-making, is highly sensitive to THC’s effects. CB1 receptors are densely located in inhibitory interneurons within the prefrontal cortex, influencing local circuitry and synaptic plasticity. Additionally, regional stimulation of CB1 receptors by THC has been shown to induce differential effects in brain areas such as the prefrontal cortex and hippocampus, modulating behaviors such as anxiety and emotional regulation through synaptic signaling [74]. Synaptic pruning in the prefrontal cortex is especially dynamic during adolescence, and disruptions caused by THC can lead to either excessive pruning or insufficient elimination of synapses.

The use of preclinical models, particularly NHPs, offer significant advantages in this research. NHPs allow for a highly controlled, longitudinal evaluation of the pharmacodynamic and behavioral effects of drug exposure, providing invaluable insights often unattainable in human studies. This enables a clearer understanding of the direct effects of THC on brain function and development. Despite these strengths, several limitations deserve mention. First, while the present results point to changes in brain FC due to adolescent THC exposure, longitudinal behavioral correlates were not measured which limits our understanding of the impact of brain changes on behavior and cognition over time, and to establish a direct causal relationship between THC exposure during adolescence and specific behavioral outcomes in adulthood. Next, the present study utilized two analytic strategies to assess FC of putative tripartite networks [40]. However, the reproducibility of large-scale networks in animal models have been questioned and choosing seeds in other subregions of CEN, DMN, or SN structures may have produced different spatial maps [75, 76]. Additionally, the present study focused on evaluation of putative tripartite networks because of their role in critical cognitive and behavioral processes, however, the effects of THC on whole brain connectivity or brain dynamics identified using other analytic approaches may yield profitable insights into how THC produces effects in the brain. Finally, the present study focuses on the effects of THC in a controlled experimental setting and may not directly represent the effects of real-world cannabis use. In particular, the use of intramuscular injections of THC deviates from typical human cannabis consumption methods, such as inhalation or oral ingestion. This route of administration has been shown to elicit heightened stress responses and other physiological confounding variables, which may have influenced the present results [77]. Furthermore, while this study specifically focused on delta-9-THC, it is important to recognize that cannabis contains several other active constituents, such as delta-8 and delta-10 THC, as well as a wide variety of cannabinoids (including cannabidiol) and terpenes.

These results advance our understanding of the complex neurophysiological impact of THC on adolescent brain development by demonstrating dose-dependent and non-linear effects on FC within critical large-scale networks. These findings reveal effects of THC exposure on residual alterations in brain function caused by THC and emphasize the role of CEN and DMN associated brain regions in mediating these effects. The observed changes in FC not only correlate with structural and functional disruptions found in human studies, but also highlight the potential long-term risks associated with early cannabis exposure. Taken together, these findings underscore an urgent need for targeted preventive measures and therapeutic interventions to mitigate THC’s effects on the neurodevelopmental trajectory of adolescents.

Author contributions

Participated in research design: Byun, Deshpande, Kangas, and Kohut. Conducted experiments: Stover, Byun, Deshpande. Performed data analysis: Byun, Deshpande. Wrote or contributed to the writing of the manuscript: Byun, Deshpande, Stover, Kangas, Kohut

Funding

This project was supported by R01-DA047575 from the National Institute on Drug Abuse and S10RR019356 by the Counterdrug Technology Assessment Center, an office within the Office of National Drug Control Policy, via Contract No. DBK39-03-C-0075, awarded by the Army Contracting Agency. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or U.S. Government.

Data availability

Correspondence and requests for materials should be addressed to Dr. Stephen Kohut.

Competing interests

Over the past 3 years, BDK has had sponsored research agreements with BlackThorn Therapeutics, Compass Pathways, Delix Therapeutics, Engrail Therapeutics, Neurocrine Biosciences, and Takeda Pharmaceuticals. No funding from these entities was used to support the current work. All other authors have no conflicts of interest or relevant disclosures.

Supplementary information

The online version contains supplementary material available at 10.1038/s41386-025-02068-5.