Lower Prefrontal Cortical Synaptic Vesicle Binding in Cocaine Use Disorder: An Exploratory ¹¹C-UCB-J PET Study in Humans

Abstract

Preclinical studies have revealed robust and long-lasting alterations in dendritic spines in the brain following cocaine exposure. Such alterations are hypothesized to underlie enduring maladaptive behaviors observed in cocaine use disorder (CUD). The current study explored whether synaptic density is altered in CUD. Fifteen individuals with DSM-5 CUD and 15 demographically matched healthy control (HC) subjects participated in a single ¹¹C-UCB-J positron emission tomography scan to assess density of synaptic vesicle protein 2A (SV2A). The volume of distribution (V_T) and the plasma free fraction-corrected form of the total volume of distribution (V_T/f_P) were analyzed in the anterior cingulate (ACC), dorsomedial and ventromedial prefrontal cortex (PFC), lateral and medial orbitofrontal cortex (OFC), and ventral striatum.

A significant diagnostic-group-by-region interaction was observed for V_T and V_T/f_P. Post-hoc analyses revealed no differences on V_T while for V_T/f_P showed lower values in CUD as compared to HC subjects in the ACC (−10.9%, p=0.02), ventromedial PFC (−9.9%, p=0.02), and medial OFC (−9.9%, p=0.04). Regional V_T/f_P values in CUD, though unrelated to measures of lifetime cocaine use, were positively correlated with the frequency of recent cocaine use (p=0.02–0.03) and negatively correlated with cocaine abstinence (p=0.008–0.03).

These findings provide initial preliminary in vivo evidence of altered (lower) synaptic density in the PFC of humans with CUD. Cross-sectional variation in SV2A availability as a function of recent cocaine use and abstinence suggests that synaptic density may be dynamically and plastically regulated by acute cocaine, an observation that merits direct testing by studies using more definitive longitudinal designs.

Graphical Abstract

This article studied whether synaptic density is altered in cocaine use disorder with 11C-UCB-J positron emission tomography It showed altered (lower) synaptic density in the medial prefrontal cortex of participants with cocaine use disorder in comparison to healthy controls Alterations in synaptic density among participants with cocaine use disorder showed correlations with frequency of recent cocaine use as well as hospital attained number of days abstinent.

Introduction

Cocaine use disorder (CUD) remains a public health concern for which no medication has an FDA-approved indication. While powerful preclinical models have advanced our understanding of molecular and cellular mechanisms of cocaine-related plasticity (1), a substantial gap remains in the clinical-translational testing and validation of such preclinical findings. Improved translation may facilitate medication development efforts for CUD.

Robinson and Kolb (2, 3) hypothesized that enduring drug-related behaviors result from drug-induced adaptations in synaptic structure (4). Specifically, they demonstrated (by Golgi-staining) robust and long-lasting (>1 month) increases in dendritic spines, post-synaptic markers of synapses (5, 6) in rodents after sensitizing regimens of amphetamine (2) and cocaine (3). Both layer 3/5 pyramidal cells of the medial prefrontal cortex (mPFC) and medium spiny neurons in the nucleus accumbens (NAc) were affected, and these cell types and brain regions have been strongly implicated in cocaine’s actions. In contrast, orbital frontal cortex (OFC), a region also involved in reward regulation, was oppositely affected (i.e., spines were decreased) (7–9). Thus, the mechanisms of aberrant structural synaptic density in CUD may not be uniform but rather regionally specific.

Since initial reports of altered synaptic/dendritic spine density, findings have been replicated in most (4, 10–16), but not all (17) preclinical studies, including by different groups (10, 12, 13, 15), after both experimenter-administered (2, 3) and self-administered drug (4, 7), in other species (18), and using more modern methods (10, 12, 13). However, to date, their relevance for the human condition of CUD has not been explored.

Macrostructural regional differences following cocaine exposure have also been reported in non-human primates, with reduced gray matter volume (GMV) in ventral and medial regions of the PFC and OFC initially that extend into dorsal and lateral regions with prolonged exposure (19). Lower ventral and medial PFC GMV have similarly been reported in CUD (20). Cocaine-related neurotoxicity may be linked to neuronal damage including synaptic structures (21), although brain structural alterations may also represent pre-existing vulnerability factors (22).

Synaptic vesicle glycoprotein 2A (SV2A) is an essential presynaptic vesicular membrane protein that is nearly ubiquitously present in presynaptic nerve terminals in the human brain (23–25). Recently, radioligands have been developed with sufficient affinity and selectivity for SV2A to render them suitable for PET imaging (26). Among them, ¹¹C-UCB-J emerged as a PET radioligand with optimal properties for imaging synaptic vesicles in vivo (27, 28). In vivo measures of ¹¹C-UCB-J volume of distribution (V_T) have been positively correlated with ex vivo measures of densities of both SV2A and synaptophysin (a ‘gold-standard’ vesicular marker of synaptic number in vitro) (28). Our group has demonstrated the tracer’s superior imaging properties, test-retest reproducibility (29), and sensitivity for detecting lower SV2A in several disorders (e.g., temporal lobe epilepsy, Alzheimer’s and Parkinson’s disease, cannabis use disorder, and depression) (28, 30–33). As such, ¹¹C-UCB-J PET constitutes the first valid translational tool for measuring synaptic density in vivo.

The current study explored whether aberrant synaptic density was evident in humans with CUD as compared to healthy control (HC) subjects using ¹¹C-UCB-J PET. Given preclinical evidence of regional differences in synaptic density and potential influences of cocaine-related toxicity, we hypothesized a diagnostic-group-by-region interaction in which synaptic density in CUD would be lower in ventromedial regions of the PFC and OFC and greater in the dorsal PFC and striatum relative to HC. We also explored relationships between measures of synaptic density and cocaine use in CUD subjects.

Materials and Methods

Participants

Community-recruited subjects were 15 individuals with CUD and 15 age-, gender-, and race-matched HC subjects. Participants were 21–55 years of age, determined to be healthy by medical and laboratory examinations, free of psychotropic medications and anticoagulants and lacking medical contraindications to magnetic resonance (MR) imaging. Females had negative pregnancy tests. Participants provided voluntary written informed consent prior to study participation, which was approved by the Yale Human Investigation and Radiation Safety Committees.

¹¹C-UCB-J PET Imaging and Processing

¹¹C-UCB-J was prepared as previously described (27). PET imaging was performed on a Siemens high-resolution research tomograph (HRRT; Siemens/CTI, Knoxville, TN, USA). Before tracer injection, a 6-minute transmission scan was performed for attenuation correction. PET scans (slices=207, slice separation=1.2 mm, reconstructed image resolution ~3 mm (34)) were acquired in list-mode following a bolus administration of ≤ 20 mCi / 740 MBq of ¹¹C-UCB-J. Head motion was measured using an optical detector (Vicra, NDI Systems, Waterloo, Ontario, Canada) following established procedures (35, 36). Arterial cannulation was performed in order to measure the arterial input function, including the fraction of unmetabolized ¹¹C-UCB-J in plasma (37). Free radioligand fraction in plasma (f_P) was determined by ultrafiltration. PET data were reconstructed with corrections made for attenuation, normalization, scatter, randoms, and dead time using the Motion-compensation OSEM List-mode Algorithm for Resolution-recovery Reconstruction (MOLAR) algorithm (35). Voxel-based parametric images of V_T values were computed using a one-tissue compartment model (29). Regional V_T values obtained using parametric images matched extremely well with values from regional time-activity curve analysis (29) in the quantification of ¹¹C-UCB-J PET data. These parametric images were also used for exploratory voxel-wise analysis (see below).

High-resolution MR images were collected on a Siemens 3T Trio system (Siemens Medical Solutions, Malvern, PA) using a standard magnetization prepared rapid acquisition gradient echo (MPRAGE) sequence. For each subject, an average PET image, corresponding to the mean of the 0–10 min frames after injection, was aligned to each MR image using rigid registration with mutual information. A transformation of the MR image to Montreal Neurological Institute (MNI) space was determined using tissue segmentation (38) and an affine linear plus nonlinear registration (BioImage Suite 2.5). The inverse of this registration was applied to regions of interest (ROIs) in template space to obtain average V_T from the parametric images.

Experimental Design and Statistical Analyses

The current study was designed as a prospective, cross-sectional (i.e., single-scan) between-group study of synaptic (SV2A) density in a priori candidate brain regions (i.e., those implicated by preclinical studies) in 15 individuals with CUD (3 female) and 15 individually demographically matched HC subjects. The study did not meet clinical trial criteria and was not publicly registered.

Cocaine-using participants met criteria for at least moderately severe CUD (≥ 4 diagnostic criteria) as determined by the Structured Clinical Interview for DSM-5 (SCID-5) (31). Seeking treatment was not a requirement for the study. CUD subjects resided on a research-dedicated inpatient unit for a period of 12–24 days prior to PET scanning. Urine toxicology was used to verify self-report of absence of drug use among HC subjects. CUD subjects also underwent urine toxicology to confirm recent cocaine use upon admission and continued abstinence during inpatient monitoring. CUD and HC participants lacked other substance use disorders (e.g., alcohol, opiates, or sedative hypnotics), except for tobacco, and other primary DSM-5 major psychiatric disorders (e.g., schizophrenia, bipolar disorder, major depression, etc.). HC subjects were matched to CUD subjects for age (± 5 years), gender, and race in order to minimize the effects of demographic factors on measures of synaptic density. A subset of HC subjects (N=8) participated in other parallel ¹¹C-UCB-J PET studies (29, 30, 39, 40). HC subjects participated as outpatients.

Primary analyses were restricted to six candidate ROIs representative of areas implicated in prior preclinical studies, including subregions of the medial prefrontal cortex (i.e., anterior cingulate cortex (ACC), dorsomedial prefrontal cortex (dmPFC) and ventromedial prefrontal cortex (vmPFC), lateral and medial subdivisions of the orbitofrontal cortex (lOFC and mOFC), and ventral striatum (VS)). We considered it important to examine these five sub-regions of the PFC taking into account preclinical literature showing regionally specific differences (i.e., increased synaptic density in dmPFC regions (2, 3) and decreased synaptic density in OFC regions (7–9) in stimulant-exposed rodents). The five cortical ROIs were derived from the automated anatomic labeling (AAL) template in Montreal Neurologic Institute (MNI) space (41) as described previously (42), and the VS ROI was drawn by hand according to published methods (43) (Supplementary Figure 1). To assess the potential confounding influence of non-specific binding, a seventh region, the centrum semiovale (CS), one rich in white matter and with negligible SV2A levels (28), was included in analyses. The CS was obtained from the AAL atlas and further optimized to minimize potential spill-in from gray matter regions with high radioactivity concentrations (40).

The plasma free fraction-corrected form of the total volume of distribution (V_T/f_P) was selected for main analyses. The free fraction of radioligand in plasma is the fraction of the ligand that is not bound to plasma proteins at equilibrium, i.e., that which is freely diffusible in plasma water (44). Many PET tracers are quantified with the volume of distribution (V_T), the ratio at equilibrium between brain and plasma tracer concentrations. However, it has long been established that for many tracers, V_T can be affected by the plasma free fraction (f_P). Although we used V_T/f_P to test the main hypothesis, we also conducted analyses using V_T. For completeness, we also evaluated the potential use of binding potential (BP_ND). We found very similar group differences with BP_ND as with V_T/f_P, further supporting findings reported below.

Prior to analysis, we examined the distribution of each outcome by diagnostic group and region and observed no outliers or obvious departures from normality. Linear mixed models were used to test the primary hypotheses of group-by-region effects on V_T/f_P in the six candidate ROIs and the CS. Residual plots were examined from each model which indicated the selected model fit the data well.

Each model included group (CUD, HC) as a between-subjects factor and region (ROI) as a within-subjects factor. Age was a significant predictor thus analyses were conducted including age as a covariate. Covariance structures considered included unstructured, compound symmetry, and heterogeneous compound symmetry. Information criteria confirmed an unstructured covariance model fit the data best for all outcomes. Values were considered significant at the uncorrected two-tailed alpha threshold of 0.05.

To assess potential influences of macrostructural differences, gray matter volume (GMV) in ROIs were obtained from the high-resolution MR images independently from PET processing using the computational analysis toolbox (CAT12) (http://www.neuro.uni-jena.de/software/). A similar mixed model was employed when analyzing regional GMV. In addition to age, total intracranial volume (TIV) also served as a covariate.

Exploratory voxel-wise analysis was conducted to visualize group differences and explore ¹¹C-UCB-J distribution beyond regions of interest. Parametric V_T/f_P and V_T images were registered into standard space using the CAT12-derived transformations and smoothed with a 6mm FWHM Gaussian kernel. Gray-matter masked two-sample t-tests were performed in SPM12 (Wellcome Trust Centre for Neuroimaging, London, UK) including age as a covariate at an initial uncorrected voxel threshold of p<0.01 and uncorrected cluster threshold of p<0.05 (k>414). Group differences in V_T did not survive this threshold and were explored further at an uncorrected voxel threshold of p<0.05 with an extent threshold of k>200.

Correlation analysis was used to explore potential relationships between imaging outcome measures (e.g., V_T/f_P and GMV) and cocaine-use variables (i.e., chronicity, frequency, and abstinence) in ROIs. Chronicity was defined as lifetime years of cocaine use. Frequency was defined as number of days of cocaine use in the month prior to study enrollment. Abstinence was defined as days since last use determined by number of inpatient days between admission and ¹¹C-UCB-J PET scan. As such, abstinence in this study is not an assessment of disease severity but represents the length of inpatient stay without use. Statistical analyses were conducted using SAS 9.4 (SAS, Cary, NC, USA).

Results

Demographic, substance-related, and radiotracer characteristics are summarized by group in Table 1. CUD and HC groups were demographically well-matched for age, gender, and race. Age was a significant predictor of V_T/f_P (p<0.0001) and GMV (p=0.0001) and therefore included in the models. CUD subjects were less educated compared to HC subjects (p=0.002). Education was not associated with either V_T/f_P or GMV, and its inclusion in models did not alter the observed findings; therefore, education was excluded from the models for parsimony. There were no associations between tobacco use (i.e., cigarettes per day), alcohol use (i.e., drinks per drinking day), and cannabis use (i.e., “joints” per day of use) and V_T/f_P or GMV among CUD subjects (all |r|’s < 0.35 and all p’s > 0.19). Among ¹¹C-UCB-J measures, CUD subjects had higher injected mass compared to HC subjects but the difference did not reach statistical significance (p=0.09), and injected mass dose was not correlated with V_T/f_P or V_T among either CUD or HC groups.

Table 1.

Subject demographic, substance use, and ¹¹C-UCB-J PET scan variables

	HC (N=15)	CUD (N=15)	P-value
Age, years	43.4 ± 9.2	42.9 ± 7.5	0.88
Gender, female/male	3 / 12	3 / 12	1.00
Race, AA/CA/HS	9 / 5 / 1	9 / 5 / 1	1.00
Education, years	14.7 ± 2.1	12.5 ± 1.4	0.002*
Cocaine use
Chronicity, years	-	18.9 ± 9.3
Frequency, days/month	-	18.5 ± 6.9
Abstinence, days	-	15.5 ± 3.4
Other substance use
Alcohol, days/month	2.9 ± 3.1	7.5 ± 7.7	0.04*
Tobacco, smoking / non-smoking	-	12 / 3
Marijuana, use / non-use	-	3 / 12
11 C-UCB-J injection
Dose, MBq	492 ± 250	588 ± 176	0.24
Injected mass, μg/kg	0.012 ± 0.008	0.019 ± 0.014	0.09
Free fraction plasma (fP)	0.28 ± 0.02	0.28 ± 0.03	0.37

Values are N or mean ± SD Abbreviations: HC, healthy control; CUD, cocaine use disorder; AA, African American; CA, Caucasian; HS, Hispanic.

In the model for V_T/f_P, the overall group effect did not reach significance (F_1,27=3.08, p=0.09) but a significant interaction between group and ROI was observed (F_6,27=7.07, p=0.0001). Post-hoc contrasts, as well as effect size, are shown in Table 2 and depicted in Figure 1. Compared to HC subjects, lower V_T/f_P values were observed among CUD subjects in the ACC (group difference ± standard error [SE]: −7.7 ± 3.07, 95% confidence interval [CI]: −14.03, −1.43), vmPFC (−7.68 ± 3.28, 95% CI: −14.40, −0.95), and mOFC (−7.71 ± 3.68, 95% CI: −15.26, −0.16). There was no group difference in non-specific binding as assessed in the CS (−0.01 ± 0.75, 95% CI: −1.55, 1.53). For completeness, analyses were also conducted without including age as a covariate, generating similar results (Supplementary Table 1).

Table 2.

Regional synaptic density (V_T/f_P & V_T) in HC and CUD subjects

VT/fP	HC	CUD	Differences in LS means ± SE: CI	Effect Size (Cohen’s d)	P-value
ACC	71.1 ± 2.1	63.4 ± 2.1	−7.7 ± 3.1: −14.0, −1.4	0.90	0.02*
dmPFC	69.7 ± 2.1	66.1 ± 2.1	−3.5 ± 3.0: −9.8, 2.7	0.41	0.25
vmPFC	77.5 ± 2.3	69.8 ± 2.3	−7.7 ± 3.3: −14.4, −0.9	0.83	0.02*
lOFC	67.4 ± 2.1	63.3 ± 2.1	−4.1 ± 3.0: −10.2, 2.1	0.48	0.18
mOFC	77.8 ± 2.6	70.1 ± 2.6	−7.7 ± 3.7: −15.2, −0.1	0.75	0.04*
VS	88.6 ± 3.1	84.1 ± 3.1	−4.4 ± 4.3: −13.4, 4.4	0.36	0.31
CS	15.3 ± 0.7	15.3 ± 0.5	−0.01 ± 0.7: −1.5, 1.5	0.01	0.98
V T
ACC	19.5 ± 0.7	18.0 ± 0.7	−1.5 ± 0.9: −3.4, 0.5	0.55	0.13
dmPFC	19.1 ± 0.7	18.8 ± 0.7	−0.3 ± 1.0: −2.3, 1.7	0.1	0.80
vmPFC	21.2 ± 0.8	19.9 ± 0.8	−1.3 ± 1.1: −3.6, 0.9	0.44	0.22
lOFC	18.5 ± 0.7	18.0 ± 0.7	−0.4 ± 1.0: −2.5, 1.6	0.15	0.66
mOFC	21.3 ± 0.8	20.0 ± 0.8	−1.3 ± 1.2: −3.8, 1.1	0.40	0.30
VS	24.2 ± 0.9	23.9 ± 0.9	−0.3 ± 1.3: −3.0, 2.4	0.08	0.82
CS	4.2 ± 0.1	4.4 ± 0.2	0.2 ± 0.2: −3.5, 0.7	0.24	0.51

Values are least square (LS) means ± SE. Abbreviations: LS, least squares; SE, standard error; CI, confidence interval; HC, healthy control; CUD, cocaine use disorder; V_T, total volume of distribution; V_T/f_P, free fraction corrected total volume of distribution; ACC, anterior cingulate cortex; dmPFC, dorsomedial prefrontal cortex; vmPFC, ventromedial prefrontal cortex; lOFC, lateral orbitofrontal cortex; mOFC, medial orbitofrontal cortex; VS, ventral striatum. Comparisons were not significant using a Bonferroni-adjusted alpha level of 0.008.

^*Unadjusted p<0.05

Figure 1. Regional synaptic density (V_T/f_P) in HC and CUD subjects.

Values are raw individual values for each group (15 HC and 15 CUD). CS is presented as a control/region of interest representative of white matter. Abbreviations: HC, healthy control; CUD, cocaine use disorder; V_T/f_P, free fraction corrected volume of distribution; ACC, anterior cingulate cortex; vmPFC, ventromedial prefrontal cortex; mOFC, medial orbitofrontal cortex; CS, centrum semiovale.

In the model examining V_T, while the main effect of group was not significant (F_1,27=0.58, p=0.45), a significant interaction between group and ROI was observed (F_6,27=6.57, p=0.0002). However, post-hoc testing revealed no differences between the diagnostic groups within any ROI (range p=0.14–0.83) (Table 2). Analyses not using age as covariate showed similar results (Supplementary Table 1).

In the model examining BP_ND, results were similar to those from using V_T/f_P (whether or not age was used as a covariate) (Supplementary Table 2) but showed main overall effect of group (F_1,27=5.9, p=0.02).

Between-group differences on the primary measurement (V_T/f_P) did not survive corrections for multiple comparisons using a Bonferroni-adjusted alpha level of 0.008. Differences observed on our secondary measurement (BP_ND) did, however, survive this same adjustment.

Results of voxel-wise analyses of V_T/f_P and V_T are shown in Figure 3. Whole-brain analyses supported the ROI-based interaction findings that appear related to prefrontal group differences in V_T/f_P. The findings also demonstrate similar, though less robust, regional differences in V_T. Additional group differences of lower V_T/f_P and V_T were observed in regions including the inferior frontal, temporal and occipital regions (Supplemental Figures 2 and 3).

Figure 3. Exploratory voxel-wise analyses of 11C-UCB-J V_T/f_P and V_T between groups.

Sagittal views (x=−2) of exploratory voxel-wise analyses to visualize group differences of lower 11C-UCB-J in CUD relative to HC subjects. V_T/f_P (left) displayed at uncorrected thresholds of voxel p<0.01 and cluster p<0.05. V_T (right) displayed at uncorrected p<0.05, with an extent threshold of k>200.

Regarding GMV, neither the main effect of group (F_1,26=0.11, p=0.75) nor the group-by-ROI interaction (F_5,130=1.30, p=0.27) were significant (Supplementary Table 3). Consistent with negative omnibus results, exploratory post-hoc analyses revealed no group differences at any region (range p=0.14–0.78). As a further precaution, we also tested for potentially embedded relationships between the significant V_T/f_P measures and GMV. These results were also supportive of the primacy of our PET findings. Specifically, in contrast to HC subjects, where partial correlations between V_T/f_P and GMV (adjusted for TIV and age) revealed significant associations in ACC, vmPFC, mOFC and lOFC in HC subjects (p=0.006–0.01), none of the cortical regions showed a significant relationship among CUD subjects (Supplementary Table 4).

Associations between measures of synaptic density (V_T/f_P) and cocaine use are summarized in Supplementary Table 5. V_T/f_P was positively correlated with frequency of recent cocaine use in the ACC, dmPFC, vmPFC, and lOFC but negatively correlated with abstinence duration in the ACC, dmPFC, vmPFC, and VS. No associations between SV2A availability and cocaine chronicity were observed. Additional post-hoc partial correlations, controlling for age, showed trend positive correlations between V_T/f_P and frequency of recent cocaine use in the ACC and vmPFC. Post-hoc partial correlations showed negative correlations between V_T/f_P and abstinence duration in the ACC and vmPFC and trend negative correlations in the dmPFC and VS (Supplementary Table 6).

Similar to previous reports (45–47), chronicity of cocaine use was negatively correlated with GMV in the ACC, dmPFC and VS. Associations between SV2A availability/GMV and cocaine use variables for one representative region (ACC) are depicted in Figure 2.

Figure 2. Significant relationships between synaptic density (V_T/f_P), grey-matter volume (GMV) and cocaine use variables in anterior cingulate cortex (ACC).

V_T/f_P ¹¹C-UCB-J binding in the ACC was positively correlated with frequency (days per month) of cocaine use (Figure 2A) and negatively correlated with duration (days) of abstinence (Figure 2B) at the time of PET scanning. By comparison, gray-matter volume (GMV) in the ACC was negatively correlated with years of lifetime cocaine use (Figure 2C). GMV relationships plotted as partial correlations adjusting for total intracranial volume (TIV). Abbreviations: V_T/f_P, free fraction corrected volume of distribution; ACC, anterior cingulate cortex; GMV, gray matter volume.

Discussion

This study is the first to measure SV2A density in vivo and the first to suggest preliminary alterations in synaptic integrity in CUD. Lower SV2A in ventral and medial PFC regions is consistent with preclinical literature demonstrating stimulant- and cocaine-related decreases in synaptic density (7–9). Contrary to preclinical evidence and hypotheses, there was no evidence of greater SV2A in the dorsal PFC or striatum (2, 3) in individuals with CUD. The findings resonate with the importance of medial and ventral PFC regions in brain mechanisms of reward and decision-making (48–55) and with prior clinical studies in cocaine-addicted populations documenting deficits in PFC metabolism (56, 57) and neurocognitive function (58–61).

PET measures of SV2A density as assessed by V_T/f_P were altered in three (i.e., ACC, vmPFC, and mOFC) of the six pre-specified ROIs. SV2A differences were detected in the absence of differences in non-specific ¹¹C-UCB-J uptake (i.e., CS V_T/f_P) or brain structure (i.e., GMV) and reference region (i.e., CS) ¹¹C-UCB-J uptake. As such, our findings argue for a primary and specific alteration in synaptic density (e.g., vs. methodological artifacts of partial volume and/or population differences in non-specific binding). The statistically significant variation in PET measures of synaptic density as a function of subjects’ self-reported cocaine use (days per month) and abstinence (days since last use) argue further for pathophysiological relevance.

In contrast to preclinical models (where increases in dendritic spines were observed in the NAc of cocaine-exposed rodents), we did not find evidence of altered synaptic density in the VS. Furthermore, our findings in mPFC were opposite of those observed in laboratory animals (i.e., lower instead of higher). Reasons for these differences are unclear, but it is possible that the negative findings regarding the VS reflect a type II error influenced by the modest sample size. Differences in patterns of drug use between animal models and human CUD (62), differences in the organization, function, and/or correspondence of specific cortical subregions (63), and/or differences in the preclinical vs. clinical-translational methods employed are other possibilities. Given the ubiquity of SV2A in presynaptic terminals in the brain, in vivo ¹¹C-UCB-J binding lacks specificity for dendritic spine subtype. As such, our PET measures may reflect summative changes in diverse neuronal populations other than or in addition to cortical pyramidal and/or striatal medium spiny cells. Similarly, despite ex vivo validation of in vivo ¹¹C-UCB-J binding as a marker of synaptic density in primates (28), we cannot exclude the possibility that changes in SV2A primarily reflect alterations in vesicular number and/or the amount of SV2A protein per vesicle, as opposed to synaptic number (although we are unaware of preclinical data that support either of these possibilities). Reverse translational studies that directly compare cocaine’s effects on Golgi-stained spines, SV2A availability/UCB-J binding, and other presynaptic vesicular (e.g., synaptophysin) and/or post-synaptic dendritic (e.g., PSD95) markers will be informative in this regard. However, our negative VS findings are consistent with a recent PET study of the medium spiny neuron-specific phosphodiesterase type 10a (PDE10a) radiotracer, ¹¹C-IMA107, that similarly failed to reveal alterations in individuals with CUD (64).

Among our most intriguing findings are those that emerged from analyses of the relationships between synaptic density, GMV and CUD subjects’ cocaine use. First, and somewhat unexpectedly, despite lower SV2A availability in the PFC (ACC, vmPFC, and mOFC) in CUD, we found no evidence that changes in synaptic density were related to chronicity of cocaine use. In contrast, our measures of GMV replicated prior reports (45–47) of negative correlations with years of lifetime use. We did, however, identify a seemingly counterintuitive positive correlation between SV2A availability and cocaine use frequency as well as a negative correlation between SV2A availability and cocaine abstinence duration (Figure 2). Alternatively stated, despite having lower SV2A availability as a group, CUD individuals with more frequent recent cocaine use showed higher SV2A availability than those who used less often and/or less recently (raising the possibility that synaptic density measures might have been even lower with longer cocaine-free intervals). This hypothesis needs further testing with future studies exploring longitudinal trajectories of these measures.

In combination, these observations suggest a more complicated and dynamic model underlying CUD, cocaine use, and SV2A availability than one solely of synaptic loss secondary to neurotoxicity (something argued against by the absence of GMV changes) (65). Non-human primate studies have shown that changes in functional activity are pronounced during the initial stages of cocaine exposure, and do not continue to intensify after a year of cocaine exposure (19). Thus, while cocaine-use alters functional neurobiology, these changes may not worsen with more years of chronic use. More acutely, preclinical evidence indicates that cocaine exposure induces the generation of silent synapses which may be pruned away following a brief period of abstinence (66, 67). In CUD, similar short-term changes have been reported in prefrontal glucose metabolism during periods of acute (i.e., normal/elevated at ≤ 1 week) and sustained (i.e., lower at 2–4 weeks) abstinence (57, 68). These data of a highly dynamic neuronal environment associated with cocaine use are consistent with the observed positive relationships between cocaine-use frequency, recency, and greater SV2A availability in CUD subjects. Such cocaine use- and/or abstinence-related changes in synaptic density might also help to explain the lack of positive correlations between MR measures of GMV and PET measures of SV2A availability in most cortical candidate ROIs in CUD. Nonetheless, conceptualization of these changes as having a highly dynamic nature is again hypothetical and needs further testing in future studies.

The post-hoc model we propose hypothesizes that i) CUD subjects exhibit lower synaptic density within regions of the mPFC, and ii) these reductions in synaptic density are transiently “normalized” (i.e., increased) by ongoing, frequent cocaine intake. This decreased synaptic density may represent premorbid or comorbid vulnerabilities (69) that could motivate cocaine use to normalize deficits (70). It could also be a consequence of cocaine use as suggested by non-human primate models of cocaine administration (19, 71). Observed relationships point to the possibility that CUD-associated reductions in synaptic density in PFC, whatever their origin, are temporarily restored by regular cocaine use, only to gradually decline (i.e., return to abnormal) upon cessation.

Co-occurring diagnoses may suggest common vulnerabilities. Regarding depression (30), CUD participants did not meet DSM-5 criteria for depression and the depression item on the Cocaine Selective Severity Assessment (72) did not support depression diagnoses. PET measures were also not related to tobacco or cannabis use (33), though the number of individuals with cannabis use was very small.

Several study limitations deserve mention. Although we believe sample size for this study is appropriate for PET imaging standards, the sample is relatively small. Considering the preclinical literature showing differential effects across regions of the PFC, we based our hypotheses on a group-by-region interaction that importantly considered these PFC regions as separate ROIs. Importantly, we observed a significant group-by-region interaction providing justification for the post-hoc within-region analyses, which when conducted independently did not survive correction for multiple comparisons. Effect sizes listed in table 2 can be used for future calculations of sample sizes required in order to observe differences between groups. The numbers of subjects required range widely depending on the measure, region, and correction employed for multiple comparisons, among other factors. In contrast to positive findings with V_T/f_P, reductions in V_T did not reach statistical significance. While theoretical considerations argue in favor of V_T/f_P as the more appropriate measure (i.e., given the greater relevance of plasma free tracer levels for brain availability), it is unknown whether meaningful biological or stochastic factors are primarily explanatory in the current context. Our prior work with V_T/f_P has shown it to have excellent absolute test-retest variability in heathy human subjects (i.e., 6.7 ± 3.4% vs. 4.4 ± 3.2% for V_T) (29). Mean plasma free fraction was similar between groups; however, the variability of free fraction amongst subjects contributed to variability in V_T, which was reduced after correction by f_p, particularly among CUD subjects. Although there were very similar group differences with BP_ND as with V_T/f_P, the former was not the main outcome. The rationale for this was two-fold: to minimize number of uncorrected multiple comparisons and because there was higher inter-subject variability to CS measures, due it its small region size.

Although our pilot sample is moderately sized relative to prior PET imaging of CUD (73–75), it is nonetheless modest in size statistically. Our a priori hypothesis-driven approach restricted primary analyses to six pre-specified ROIs based on prior work (2–9). While findings are consistent with evidence that ventral and medial PFC regions are most impacted by cocaine use (19, 71), we cannot exclude the possibility that the lack of alterations in the dorsal, lateral, and striatal ROIs reflects type-2 errors and small sample size. Another intrinsic weakness of an approach based on ROI includes challenges in addressing questions of regional specificity and the extent to which our observations of synaptic abnormalities in the PFC may extend to other brain areas. However, the results from whole-brain voxel-based analyses on both V_T and V_T/f_P both support the specificity of our findings but also suggest possible involvement of other cortical regions. Studies using larger samples will be needed to clarify synaptic density differences in the PFC. Similarly, correlational analyses with clinical characteristics were exploratory and should be investigated in larger, longitudinal cohorts. Thus, future research in larger samples and longitudinal cohorts more extensively matched for education and other substance-related variables, using whole-brain, voxel-based approaches are warranted to address these limitations.

Findings of altered prefrontal cortical synaptic density provide intriguing preliminary translational support for a compelling preclinical hypothesis of CUD. Further investigation into potential relationships with clinical features, PFC-mediated neurocognition, and functional and metabolic activity could motivate the development of novel treatment approaches, including strategies based on synaptotropic mechanisms. Such work holds promise for advancing our understanding of the role of dynamic, drug-induced changes in synaptic function in CUD, and for identifying more effective treatments.