Abstract
While functional brain imaging studies in humans suggest that chronic cocaine use alters functional connectivity (FC) within and between key large-scale brain networks, including the default mode network (DMN), the salience network (SN), and the central executive network (CEN), cross-sectional studies in humans are challenging to obtain brain FC prior to cocaine use. Such information is critical to reveal the relationship between individual's brain FC and the subsequent development of cocaine dependence and brain changes during abstinence. Here, we performed a longitudinal study examining functional magnetic resonance imaging (fMRI) data in male rats (n = 7), acquired before cocaine self-administration (baseline), on 1 d of abstinence following 10 d of cocaine self-administration, and again after 30 d of experimenter-imposed abstinence. Using repeated-measures analysis of variance (ANOVA) with network-based statistics (NBS), significant connectivity changes were found between anterior insular cortex (AI) of the SN, retrosplenial cortex (RSC) of the DMN, somatosensory cortex, and caudate–putamen (CPu), with AI–RSC FC showing the most robust changes between baseline and 1 d of abstinence. Additionally, the level of escalated cocaine intake is associated with AI–RSC and AI–CPu FC changes between 1 d and 30 d of abstinence; further, the subjects’ AI–RSC FC prior to cocaine intake is a significant moderator for the AI–RSC changes during abstinence. These results provide novel insights into the roles of AI–RSC FC before and after cocaine intake and suggest this circuit to be a potential target to modulate large-scale network and associated behavioral changes in cocaine use disorders.
Significance Statement
This study examines the impact of chronic cocaine self-administration on brain network–level interactions involving the default mode network (DMN; retrosplenial cortex, RSC), salience network (SN; anterior insular, AI), and caudate–putamen (CPu). These brain regions have been associated with self-referential functions, emotion, memory, and coordination between internal and external stimuli and align with the “triple network model” of psychopathology and addiction. By identifying relationships between the escalated cocaine self-administration and the changes of functional connectivity across longitudinal measures, this study contributes to the future development of circuit-based treatment strategies and suggests AI and RSC connectivity to be an imaging circuit marker for cocaine use disorders.
Introduction
Cocaine is a highly addictive psychostimulant drug known for its potent reinforcing effects (Spronk et al., 2013; Volkow et al., 2019) and associations with multiple emotional and cognitive impairments (Goldstein et al., 2009; Wolf, 2010; Spronk et al., 2013; Potvin et al., 2014). Compulsive drug-seeking behavior and craving persist long into abstinence, which has been shown to contribute to the risk of relapse (Nestler, 2005; Li et al., 2016). Numerous clinical and preclinical studies have established a relationship between cocaine use and alterations in the mesocorticolimbic system, involving short-term changes to dopaminergic (Volkow and Morales, 2015) and glutamatergic signaling (Kalivas, 2009; Schmidt and Pierce, 2010), and longer-term neuroplastic changes (Kolb and Sharpless, 2003; Kauer and Malenka, 2007; Robison and Nestler, 2011; Waselus et al., 2013) during abstinence and relapse (McFarland et al., 2003; Wolf, 2016; Werner et al., 2020). Although these cocaine-induced signaling changes can affect motivated behaviors and the relevant large-scale brain networks, our understanding of spatial, temporal, and polarity of these changes on a whole-brain scale remains relatively limited.
Functional connectivity magnetic resonance imaging (fcMRI) has been extensively employed to study cocaine-induced network-scale changes in the brain (Kufahl et al., 2005; Worhunsky et al., 2013; Kober et al., 2016). By analyzing temporally synchronized spontaneous fluctuations across spatially distinct regions, fcMRI provides information representing functional connectivity (FC) throughout the brain (Biswal et al., 1995). Previous human neuroimaging studies have identified altered FC between several brain regions in cocaine users, including the ventral tegmental area, striatum, nucleus accumbens, prefrontal cortex, amygdala, hippocampus, insular cortex, and cingulate cortex (Gu et al., 2010; Kelly et al., 2011; Cisler et al., 2013; McHugh et al., 2013, 2014; Hu et al., 2015). Furthermore, these FC changes have been associated with behavioral outcomes, such as years of past cocaine use (Gu et al., 2010), recent/compulsive cocaine use (Hu et al., 2015), impulsivity (McHugh et al., 2013; Hu et al., 2015), and history of relapse (McHugh et al., 2013; Camchong et al., 2014). Nevertheless, the inherent complexities of human subject research make it challenging to isolate the direct impact of cocaine exposure on the brain (Greenland et al., 1999; Compton et al., 2007) from factors accompanying cocaine use. Consequently, FC changes may be influenced by differences in individual histories of cocaine use, abstinence, relapse, and treatment, as well as comorbid conditions, including polydrug use (Unterrainer et al., 2019; Loganathan et al., 2021). These measures are also commonly collected as self-report data, which has uncertain reliability and validity in drug-using populations (Napper et al., 2010). Moreover, cross-sectional human studies lacking baseline measurements cannot examine whether FC prior to drug use could serve as a potential risk factor contributing to individual susceptibility for developing cocaine dependence.
To overcome the limitations of human studies, preclinical animal models such as rodent cocaine self-administration can provide valuable insight into the transition from drug-naive to chronic cocaine-taking behavior and the abstinence-relapse cycle (Kawa et al., 2019). Classical rodent studies have demonstrated FC changes related to cocaine exposure, including alternations to brain regions related to cognition and emotion (e.g., amygdala, hypothalamus, striatum, hippocampus, and thalamus; Lu et al., 2014; Orsini et al., 2018). Despite these seminal findings, longitudinal changes among critical nodes of large-scale brain networks before and after cocaine self-administration remain incompletely studied. It is also unclear whether certain patterns of brain FC prior to cocaine exposure can predict later changes following prolonged abstinence. Such knowledge is critical, particularly in light of the growing interest in understanding FC changes among three large-scale brain networks (Menon, 2011; Lee et al., 2021; Mandino et al., 2022; Chao et al., 2023; Menon et al., 2023), namely, the salience network (SN), default mode network (DMN), and central executive network [CEN, sometimes referred to as the lateral cortical network or LCN in rodent studies (Coletta et al., 2020; Lee et al., 2021), in neuropsychiatric disorders (Menon, 2011, 2018; Mohan et al., 2016), and in addiction (Sutherland et al., 2012; Volkow et al., 2019; Zhang and Volkow, 2019; Ersche et al., 2020)].
In this study, we used a recently established isotopic echoplanar imaging (EPI) fMRI protocol (Lee et al., 2021), data-driven independent component analysis (ICA; Beckmann et al., 2005; Lee et al., 2021), dual regression (Beckmann et al., 2009), and network-based statistics (NBS; Zalesky et al., 2010) to evaluate global network FC at baseline, 1 d of abstinence (1 d ABS), and after 30 d of experimenter-imposed abstinence (30 d ABS) and determined significant FC changes between those periods relative to water self-administration controls. We also explored how FC at baseline and individual differences in development of cocaine self-administration behavior contribute to those significant FC changes using a moderation analysis model (Baron and Kenny, 1986). Our findings offer insights into brain circuit and network changes following escalation of cocaine intake and prolonged (1 month) abstinence.
Materials and Methods
Experimental subjects
All procedures were conducted in accordance with the Guide for the Care and Use of Laboratory Animals, as adopted by the National Institutes of Health, and with approval of the Institutional Animal Care and Use Committee at the University of North Carolina (UNC). Adult (initially weighing ∼300 g) male Sprague Dawley rats (Charles River Laboratories) were individually housed and maintained on a 12 h light cycle. All animals were mildly water restricted to ∼20 ml/d during the self-administration portion of the experiment (plus additional fluids collected during the session). During the 30 d abstinence period, animals underwent mild food restriction to maintain ∼90% body weight. An experimental setup schematic and timeline of the surgical procedures and scans appears in Figure 1, and detailed descriptions of the experimental procedures are outlined below.
Experimental timeline schematic and self-administration training data. A, All animals underwent fcMRI scans at each of three time points: prior to self-administration, immediately following self-administration (within 24–48 h following the last session), and after 30 d of experimenter-imposed abstinence. Self-administration sessions were 6 h/d for 10 d. Cocaine rats were trained to nosepoke to self-administer intravenous cocaine (0.33 mg/infusion). Control rats were trained to nosepoke for a water reward (200 µL) and received yoked intravenous 0.9% saline infusions. All infusions were accompanied by a 20 s tone/house light stimulus. B, Mean self-administration rates over the 10 d of nosepoke training for cocaine (n = 7) or water (n = 6; *p < 0.05; **p < 0.005). C, A significant linear trend was found in both cocaine and water groups via LMER modeling. The reinforced nosepokes (Z) were converted to an escalation value for each subject to capture individual differences in self-administration training by subtracting the sum of self-administration days 1–6 from the sum of days 7–10. D, A significant difference (t value = 3.65; p = 0.004) was found in the cocaine self-administration escalation value between cocaine and water group. Error bar represents standard error (SE).
fcMRI procedures
Each animal underwent six consecutive 5 min fcMRI scans three times: ∼8 d before self-administration training (baseline), between 24 and 48 h after the 10th and final day of self-administration (1 d ABS), and after 30 d of experimenter-imposed abstinence (30 d ABS; Fig. 1A). On fcMRI experiment days, each rat was endotracheally intubated and initially ventilated with 2% isoflurane mixed with medical air. Each rat was then placed on a custom-built MR cradle lined with a warm water circulating pad, and rectal temperature was maintained at 37 ± 0.5°C. The ventilation rate and volume were adjusted via a capnometer to maintain end-tidal CO2 (EtCO2) within a range of 3.0 ± 0.2%. Heart rate and oxygen saturation (SpO2) were continuously monitored by a noninvasive, MR-compatible MouseOx Plus System and maintained within normal ranges (∼280 bpm and >96%, respectively). For fcMRI acquisition, light anesthesia was maintained by a low dose mixture of 0.5% isoflurane with intravenous dexmedetomidine (0.05 mg/kg/h) and pancuronium bromide (0.5 mg/kg/h); this combination, used previously (Decot et al., 2017), has been shown to retain robust network connectivity (Lu et al., 2012b; Hsu et al., 2016; Orsini et al., 2018; Lee et al., 2021). All fcMRI data were acquired using a 9.4 tesla Bruker BioSpec MR scanner with a 72 mm volume transmitter and four-channel phased array receiver. Blood oxygenation level-dependent (BOLD) functional resting-state scans were acquired using a single shot gradient echo-echoplanar imaging sequence (spectral width = 150 kHz; TR/TE = 2,000/11.2 ms; FOV = 2.88 × 2.88 × 1.28 cm3; slices = 32; matrix = 72 × 72; and spatial resolution = 0.4 mm isotropic; Lee et al., 2021).
Cocaine self-administration
After baseline fcMRI scans, rats were surgically implanted with MR-compatible intrajugular catheters and allowed to recover for 1 week prior to being randomly assigned to either the cocaine (n = 7) or water (n = 6) self-administration group. Catheter implantation surgeries were conducted as previously described (Carelli and Deadwyler, 1994). After 1 week of recovery, rats were lightly water deprived and then trained on a standard self-administration regimen similar to that described previously (Saddoris et al., 2011; Moschak et al., 2023). Each animal underwent 6 h self-administration sessions daily for 10 d. At the beginning of each session, rats were placed into a standard self-administration chamber (25 × 25 × 30 cm, stainless steel rod floor; Med Associates) and connected via their intrajugular catheter to tubing that interfaced with a syringe containing either cocaine (1.67 mg/ml in 0.9% saline) or 0.9% saline. At the onset of each session, a cue light illuminated within the nosepoke port. For the cocaine group, rats were trained on a fixed ratio 1 schedule of reinforcement where each nosepoke resulted in intravenous cocaine delivery (0.33 mg/infusion) over 6 s. Each infusion was signaled by termination of the cue light and the onset of a tone–house light stimulus over a 20 s interval. Nosepokes made during the 20 s postresponse interval were recorded but had no behavioral consequences. Following the 20 s interval, the tone–house light stimulus turned off and the cue light was again illuminated. To maintain similar motivated instrumental experience between groups, control rats underwent a self-administration procedure identical to cocaine self-administration animals; however, reinforced nosepokes delivered 200 µl of water into a food cup, and animals received yoked saline infusions (0.9% saline, 0.2 ml, for 6 s) contingent upon the delivery schedule of a paired rat in the cocaine self-administration group.
fcMRI data preprocessing
fcMRI data were preprocessed using the Analysis of Functional NeuroImages software (Cox, 1996) following standard pipelines. Briefly, preprocessing steps included skull stripping, slice timing correction, rigid body motion correction, alignment to a high-resolution template space (Valdés-Hernández et al., 2011), high-pass filtering (>0.01 Hz), and spatial smoothing (Gaussian kernel full-width at half-maximum, FWHM = 0.6 mm). Independent component analysis (ICA) was used to identify and remove physiological, movement, and thermal (i.e., machine) noise components (Rummel et al., 2013; Griffanti et al., 2017), and six head movement parameters were regressed from the signal. No significant motion and temporal signal-to-noise ratio changes were found across the 6 fcMRI scans as a function of scan day or self-administration group (Fig. S1). After fcMRI preprocessing, the six consecutive 5 min fcMRI scans for each rat were concatenated into one for each scan day. One rat in the water self-administration control group was excluded due to three incomplete fcMRI scans. Further, to assess the reproducibility of our FC measures across varying lengths of fMRI scans, we employed an iterative random sampling method combined with intraclass correlation coefficient (ICC) estimation (Extended Data Fig. 2). Our iterative random sampling approach revealed a clear trend of increasing reproducibility in FC estimation with the concatenation of more fMRI trials (Extended Data Fig. 2; Cho et al., 2021). Therefore, the six consecutive 5 min fcMRI scans for each rat were concatenated into one for each scan day to achieve better reliability.
Figure S1.
No significant motion effect was found across the 6 fcMRI scans as a function of scan day or selfadministration group. (x-axis represents the scan number). Download Figure S1, PDF file.
Functional connectivity analysis
Group-level independent component analysis (gICA)
Following data preprocessing, all fcMRI data (total 36 scans: 12 rats × 3 scans) were used to generate resting-state FC component maps by using group-ICA (MELODIC; FSL). We selected 20 components in accordance with previous rodent fcMRI studies (Becerra et al., 2011; Jonckers et al., 2011; Lu et al., 2012b; Hsu et al., 2016). The spatial distributions of individual functional components were reconstructed using dual regression (Beckmann et al., 2009; Zuo et al., 2010), and a one-sample t test was performed to generate the group component maps (t value >6, p < 0.0001; Extended Data Fig. 3).
Resting-state FC and network-based statistics
To investigate the FC among selected functional components, we extracted the subject-specific fcMRI time courses corresponding to each selected functional components from the multivariate spatial regression in the first stage of dual regression (Nickerson et al., 2017). Correlation coefficients (Pearson's correlation) were computed between the individual subject time courses for each functional component of interest. These correlation coefficients were then Fisher z-transformed before statistical comparison.
Statistics and moderation analysis
To determine significant changes in self-administration behavior across days, we conducted a nonparametric test (Mann–Whitney U test) to compare differences of mean self-administration rates over the 10 d of nosepoke training between cocaine (n = 7) and water (n = 6) groups. In order to capture the nuanced temporal shifts in self-administration behavior, particularly during the later phase of the training period (days 7–10) as opposed to the initial phase (days 1–6), we examined the escalation value. The concept of an escalation value or ratio has been employed in previous studies to track changes in drug self-administration behavior over time (Zhukovsky et al., 2019; Cocker et al., 2020). This measure was selected not only for its established relevance in capturing drug-seeking behavioral trajectories but also to provide a more granular view of individual rat behaviors over time. It encapsulates the relative change within each subject, offering insights that raw scores might miss, highlighting the unique behavioral patterns distinct to the later stages of self-administration.
To demonstrate the linear trend of cocaine and water self-administration behavior across the 10 d of conditioning, reinforced nosepokes per session were z-normalized in each group (subtract the mean from each of the individual data points and divide by the standard deviation) and input into a generalized linear mixed-effects (LMER) model in MATLAB (MathWorks). The experimental day was included as a fixed effect, and random intercept and subject effects were included to characterize temporal correlations. For comparison to FC changes, reinforced nosepokes (Z) were converted to an escalation value for each subject to capture individual differences in self-administration conditioning by subtracting the sum of self-administration days 1–6 from the sum of days 7–10.
We conducted a repeated-measures analysis of variance (ANOVA) in NBS (Zalesky et al., 2010) to identify the significant connectivity changes among baseline, 1 d ABS, and 30 d ABS time points. NBS was chosen for its enhanced sensitivity to detect interconnected subnetwork level changes, offering a more comprehensive perspective of FC discrepancies across conditions. Unlike traditional voxel-wise or node-wise methods that correct for multiple comparisons per connection and may miss subtle network variations, NBS focuses on identifying significant clusters of connections between groups. Furthermore, its nonparametric permutation testing approach, involving the establishment of a null distribution of test statistics without specific effect assumptions, is particularly suitable for fMRI data, ensuring robust results especially given our study's sample size. The primary threshold (i.e., the test statistic threshold), which limits the variance between intergroup connectivities, was set at F(2,18) = 3.56 (p < 0.05) for the cocaine group and F(2,12) = 3.89 (p < 0.05) for the saline group, and the significance threshold for the analysis was set at p < 0.05 corrected for multiple comparisons with a FWE rate approach and 5,000 randomizations were performed. We further conducted the post hoc analysis using Tukey's honestly significant difference (HSD), a single-step multiple-comparisons procedure and statistical test, to test the pairwise comparisons among baseline, 1 d ABS, and 30 d ABS on the connectivities with significant changes yielded by the NBS analysis.
To demonstrate the direct effect on the cocaine self-administration–related FC changes of functional component at 30 d ABS–1 d ABS from cocaine escalation ratios, we conducted a path analysis using the structural equation modeling method (Boucard et al., 2007) in AMOS 17.0 (SPSS). In assessing this direct effect, we applied a Bonferroni’s correction to account for multiple comparisons. This decision stemmed from the method's proficiency in understanding complex causal relationships. Moreover, the relationships between cocaine self-administration escalation values and FC changes at 30 d ABS–1 d ABS were further examined with a moderation analysis model (Baron and Kenny, 1986), which assesses the impact of an intermediary variable (the moderator) on the relationship between an independent and dependent variable. The moderation model used the FC of a given functional component at baseline as a moderator of the association between the changes in the component FC between 30 d ABS–1 d ABS and the cocaine escalation ratio. The moderation effect of each moderator for each brain–behavior relationship was evaluated in the interaction (Int) between independent and moderator variables. In this context, full moderation occurs when the interaction effect is significant with the inclusion of a moderator variable (p < 0.05), and the relationship between the independent and dependent variables with moderator (WM) is no longer significant (Baron and Kenny, 1986).
Results
Self-administration behavior
Rats in both cocaine and water self-administration groups showed no significant difference in reinforced nosepokes from day 1 to day 6 of training (Mann–Whitney U test; p > 0.05). A significant difference in reinforced nosepokes was observed between groups from day 7 to day 10 (Mann–Whitney U test; p < 0.05; Fig. 1B). The total reinforced nosepokes (from day 1 to day 10) did not differ significantly between groups (Mann–Whitney U test; z-value = 0.64; p > 0.05). Moreover, the LMER model demonstrated a significant positive linear trend (t value = 2.34; p = 0.02) in normalized reinforced nosepokes across 10 d in the cocaine group, contrasting with a significant negative linear trend for the water group (t value = −4.85; p < 0.001; Fig. 1C). Consequently, the escalation value for reinforced nosepokes between days 1–6 (no group differences) and days 7–10 (significant group differences) was significantly greater for the cocaine self-administration group than that for water self-administration controls (two-sample t test; t value = 3.65; p = 0.004; Fig. 1D).
Functional connectivity changes following cocaine self-administration and abstinence
Using gICA, we derived functional components from all subjects (n = 12) at baseline. Twelve functional components were identified (Extended Data Fig. 3). Specifically, nine functional components were further classified by the anatomical region corresponding to their peak significance in the one-sample t test of dual regression (Extended Data Fig. 3; Beckmann et al., 2009) due to their prevalent interest in addiction and brain network literature (Seeley et al., 2007; Buckner et al., 2008; Hutchison et al., 2010; Becerra et al., 2011; Gozzi and Schwarz, 2016; Hsu et al., 2016; Tsai et al., 2020; Lee et al., 2021; Mandino et al., 2022; Oyarzabal et al., 2022; Chao et al., 2023; Menon et al., 2023). These regions include the following: anterior insula (AI) and nucleus accumbens (NAc) of the SN; cingulate cortex (Cg), retrosplenial cortex (RSC), and posterior parietal cortex (PPC) of the DMN; primary motor cortex (M1), anterior primary somatosensory cortex (aS1), and posterior S1 (pS1) of the LCN; and caudate–putamen (CPu; Lobo and Nestler, 2011; Yager et al., 2015). These data-driven regions of interest were used for subsequent analysis (Fig. 2A).
NBS analysis indicated significant FC changes in the cocaine group among baseline, 1 d ABS, and 30 d ABS conditions. A, Graphic illustration of the significant connectivity changes from NBS analysis. The gray line and red square (B) indicated the significantly altered connectivities among three conditions using NBS (p < 0.05). C, The post hoc analysis using Tukey's HSD for each identified connectivity in the cocaine group and the corresponding connectivities in the water group. No significant differences were found in the water group (*p < 0.05). D, Graphic illustration of the significant post hoc analyses in the identified connectivities.
We conducted the NBS analysis to compare between-component FC among baseline, 1 d ABS, and 30 d ABS data in cocaine and water groups. Significant changes in FC between functional components were observed in the cocaine group, but not in the water control group (Fig. 2B). Specifically, post hoc analysis using Tukey's HSD (q value) of 1 d ABS compared with baseline revealed a significant change in FC, reflected by a reduced negative correlation between SN (AI)–DMN (RSC) (q = 5.79; p < 0.05) in cocaine-exposed animals (Fig. 2C). Comparing 1 d ABS with 30 d ABS, a significant change in FC was found in SN (AI)–DMN (RSC) (reduced negative correlation; q = 4.68; p < 0.05) and LCN (pS1)–DMN (RSC) (reduced negative correlation; q = 3.51; p < 0.05), whereas the FC of CPu with SN (AI; enhanced positive correlation; q = 4.69; p < 0.05) and RSC (reduced negative correlation; q = 3.93; p < 0.05) also showed significant changes (Fig. 2C,D).
Relationship between escalated cocaine self-administration and functional connectivity following prolonged abstinence
Next, we investigated how individual differences in cocaine self-administration (escalation value; Fig. 1D) were reflected in the cocaine-induced FC changes at 30 d ABS. The general linear model revealed significant group-dependent differences in the relationship between individual subjects’ escalation value and FC changes of SN (AI)–DMN (RSC) (t value = 3.85; p = 0.004) and SN (AI)–CPu (t value = −2.58; p = 0.032) at 30 d ABS and 1 d ABS between water and cocaine groups (Fig. 3A). Specifically, a significant negative association was found in SN (AI)–DMN (RSC) whereas a positive association was found in SN (AI)–CPu between escalation value and FC changes at 30 d ABS–1 d ABS (Fig. 3B), with no significant associations observed in the water control group. Further, the two significant FC changes (AI–RSC and AI–CPu) at 30 d ABS–1 d ABS further exhibited a prominent negative correlation (Fig. 3C; r = −0.81; p = 0.014). Of note, no significant associations were observed between escalation values and identified FC changes at 1 d ABS–baseline (Extended Data Fig. 4).
The significant interaction effect between the reinforced nosepoke session changes (escalation value) with the connectivity of (A) SN (AI)–DMN (RSC) (t value = 3.85, p = 0.004) and SN(AI)–CPu (t-value = −2.58, p = 0.032) at 30 d ABS and 1 d ABS between the water and cocaine groups (*p < 0.05; **p < 0.01; corrected by Bonferroni’s correction). B, Graphic illustration of the significant connectivity changes of SN–DMN and SN–CPu from 1 d ABS to 30 d ABS. C, The FC changes of SN (AI)–DMN (RSC) and SN (AI)–CPu at 30 d ABS–1 d ABS showed significantly negative correlation in cocaine group (r = −0.81; p = 0.014).
Intercomponent functional connectivity before cocaine exposure moderates the association between later cocaine intake and functional connectivity changes following abstinence
Path analysis between individual subjects’ escalation value and FC changes between 1 d ABS and 30 d ABS (30 d ABS–1 d ABS) revealed a significant negative association relative to SN (AI)–DMN (RSC) (coefficient = −0.028; SE = 0.007; p < 0.001; Fig. 4A), whereas AI–CPu exhibited a significantly positive association (coefficient = 0.032; SE = 0.013; p = 0.01; Fig. 4B). We conducted a moderation analysis to investigate how the FCs (AI–RSC and AI–CPu) at baseline moderated the relationship between cocaine escalation value and FC changes at 30 d ABS–1 d ABS. Importantly, in the relationship between cocaine self-administration changes (escalation value) and the SN (AI)–DMN (RSC) connectivity during cocaine abstinence (30 d ABS–1 d ABS), significant partial moderation effects were identified when incorporating baseline SN (AI)–DMN (RSC) and SN (AI)–CPu FC as moderators. No significant moderation effect was identified in the relationship between cocaine self-administration changes with the SN (AI)–CPu connectivity at 30 d ABS–1 d ABS when recruiting baseline SN (AI)–DMN (RSC) and SN (AI)–CPu FC as moderators.
Graphic illustration of the moderation analysis. Results depict the relationship between cocaine self-administration at 1 d ABS–baseline (escalation value) and identified connectivity changes at 30 d ABS–1 d ABS, (A) SN–DMN and (B) SN–CPu from Figure 4 (dependent variable) with the influence (moderation) of their connectivity at baseline (moderators). A Bonferroni’s correction was applied when assessing the direct effects to account for multiple comparisons. Path analysis using only cocaine-exposed rats (N = 7) identified a significant relationship between cocaine-dependent changes in these two connectivities. Specifically, the associations between the cocaine self-administration at 1 d ABS–baseline with the SN–RSC was partially moderated by both connectivity of SN–DMN and SN–CPu at baseline and whereas no significant moderation effect was found in the association AI–CPu at 30 d ABS–1 d ABS and cocaine self-administration at 1 d ABS–baseline. The partial moderation occurs when the relationship between the independent variable (cocaine self-administration at 1 d ABS–baseline) and the dependent variable (connectivity change at 30 d ABS–1 d ABS) is still significant with the inclusion of a moderator variable (WM) while the moderation (Interaction, Int) is significant. The direct effect (DF) represents the relationship between dependent and independent variable without moderator. Arrowheads of solid red lines represent significant moderation effects (n.s. p > 0.05; *p < 0.05; **p < 0.01; and ***p < 0.001).
Discussion
Our study sheds light on the impact of cocaine self-administration on the cocaine-naive brain via an unbiased, data-driven analytical approach. We reveal that cocaine exposure leads to significant neuronal component alterations in FC across multiple distinct brain regions. These changes were absent in the water control animals, suggesting that the observed alternations are specific to cocaine exposure. Previous studies using animal models had examined FC changes to acute cocaine administration during the period of abstinence (Lu et al., 2014; Murnane et al., 2015), neuronal activity changes following acute cocaine administration (Lu et al., 2007; Jasinska et al., 2015), or focused on extracting network graph features without identifying region-specific FC changes following cocaine self-administration (Orsini et al., 2018). These studies explored diverse aspects of cocaine addiction, such as the impact on FC between brain regions, the role of neural circuits in relapse, and cocaine's influence on reward-seeking behavior, learning, memory, and emotion processing. Our study addresses critical knowledge gaps by (1) linking the escalated cocaine self-administration behavior with brain FC changes, (2) identifying AI–RSC as a critical pathway showing changes over the period of cocaine self-administration and again across 30 d of abstinence, and (3) establishing AI–RSC FC in the cocaine-naive brain to be a significant moderator for the subsequent FC changes induced by cocaine self-administration and abstinence.
Emerging evidence suggests that interactions between brain networks play crucial roles in neuropsychiatric disorders, including substance use disorder (Sutherland et al., 2012; Ersche et al., 2020). A “triple-network” model (Menon, 2011) describing the relationships between SN, DMN, and CEN (sometimes referred to as LCN in rodent studies (Coletta et al., 2020; Lee et al., 2021), has been increasingly adapted to understand addiction and associated behaviors (Sutherland et al., 2012). Among these three networks, the SN is suggested to dictate dynamic switching between the DMN and CEN, and the balance between SN and the other two networks, known for driving self-referential thought and executive control, respectively, is hypothesized to be the key to allocating cognitive resources in substance use disorder (Everitt and Robbins, 2005; Zhang and Volkow, 2019; Keeley et al., 2020). Within the SN, the AI has been identified as a causal outflow hub mediating dynamic switching between DMN and CEN activity (Craig, 2002; Sridharan et al., 2008; Menon and Uddin, 2010). In the context of this model, we observed that the AI, RSC, and CPu were significantly altered nodes following cocaine self-administration. Specifically, we found that the AI–RSC connectivity was decoupled in a form of losing anticorrelation at 1 d of abstinence compared with that at 30 d of abstinence and also the baseline cocaine-naive (pre-administration) period. The pronounced negative coupling between AI and RSC at baseline suggests an antagonistic relationship between these two network (i.e., SN and DMN) nodes as seen in other functional studies (Sridharan et al., 2008; Menon and Uddin, 2010; Menon, 2011; Nekovarova et al., 2014; Seeley, 2019; Chao et al., 2023). The DMN is linked to self-referential thinking and internal mental processes (Sheline et al., 2009), whereas the SN is responsible for detecting and processing salient external stimuli and allocating attentional resources (Menon and Uddin, 2010). The connectivity between AI and RSC (rodent analog of the posterior cingulate cortex in humans; Vann et al., 2009) plays a crucial role in interoceptive awareness, perception, and consciousness (Craig, 2009; Caseras et al., 2013; Gasquoine, 2014). However, the diminishing coupling AI–RSC FC following 10 d of cocaine self-administration indicates a disruption in the normal balance and segregation of these networks, which may lead to the enhanced saliency toward drug cues and shifting attention to internal thoughts and emotions (Naqvi and Bechara, 2009, 2010; Zhang and Volkow, 2019). This disruption might contribute to the maladaptive decision-making and attentional biases observed in addiction (Anderson, 2016; Guttman et al., 2018; Parvaz et al., 2021), as previous studies have shown that reduced negative SN–DMN coupling is associated with decreased attention (Stern et al., 2012; Beucke et al., 2014; Posner et al., 2017).
In addition to our finding between AI and RSC, we also found the decoupled AI–CPu connectivity at 1 d of cocaine abstinence compared with that at 30 d of abstinence. The CPu is modulated by dopamine (DA) projections primarily from the substantia nigra pars compacta and is implicated in motor and cognitive control processes (Graybiel, 2008; Kreitzer and Malenka, 2008; Balleine and O’Doherty, 2010). Moreover, the CPu is also involved in cognitive and executive functions, such as spatial working memory (Levy et al., 1997; Rodriguez-Moreno and Hirsch, 2009). Although the CPu is not considered the primary brain region responsible for the acute rewarding effects of drugs of abuse, it is involved in habit learning (Packard and Knowlton, 2002; Volkow et al., 2006) and is recruited during the development of compulsive drug-seeking (Everitt et al., 2008) and drug-craving behaviors (Heinz et al., 2004; Volkow et al., 2008). Among these FC changes after cocaine self-administration, we also observed a significant direct effect of the amount of cocaine self-administration on AI–RSC and AI–CPu FC changes through 30 d of abstinence. Notably, these relationships were significantly moderated by the FC at baseline. As escalated cocaine use leads to negative coupling of AI–RSC and positive coupling of AI–CPu during abstinence, the significant negative correlation between changes in AI–RSC and AI–CPu connectivity during cocaine abstinence further supports the notion that these disrupted networks return to baseline via rebalancing their interactions through the AI. These results also supports the involvement of AI (a key node of SN) in controlling dynamic activity between the RSC (a key node of DMN) and CPu by flexibly directing attention to internal or external events (Sridharan et al., 2008; Fedota et al., 2018).
Cocaine self-administration methodology allows researchers to understand how drug-associated contexts contribute to the persistence of drug-seeking (Quick and Shahan, 2009; Oleson and Roberts, 2012). Escalation of cocaine might be attributed to development of compulsive drug use, which has been shown to be mediated by a transition in behavioral control from ventral NAc to dorsal CPu (Belin and Everitt, 2008; Pacchioni et al., 2011). The increased linear trend of drug reinforcement frequency was observed following 10 d of cocaine self-administration, while a decreased linear trend was evident in the water control group. The contrasting trends observed between the cocaine and water groups underscore the intrinsic disparities in the reinforcing attributes of cocaine compared with a natural reward like water. Further, the decrease in reinforced nosepokes for the water group could be related to animals adjusting their behavior as they learn the task. Nevertheless, among the identified FCs, we observed a significantly negative correlation of the escalation value for cocaine with the connectivity of AI–RSC during abstinence (30 d ABS–1 d ABS), while a positive correlation with AI–CPu connectivity changes during abstinence. Importantly, the relationship of cocaine escalation value with AI–RSC during abstinence builds upon previous findings that SN and DMN integration is specifically dysregulated in cocaine addiction (Liang et al., 2015; Geng et al., 2017), suggesting a functional circuit mechanism by which the recovery of saliency detection and integration, along with poor cognitive control and decision-making during abstinence, may be associated with the amount of escalated cocaine intake. Interestingly, the significant relationship between FC of AI–CPu and escalation value fits well with previously reported evidence that postescalation lesion of AI restores control over cocaine use and prevents cue-induced reinstatement (e.g., relapse; Rotge et al., 2017). Furthermore, using a moderation analysis, we demonstrated that the relationships of cocaine escalation value with changes of AI–RSC and AI–CPu FCs during abstinence (30 d ABS–1 d ABS) were statistically moderated by the individual FC at baseline. In other words, we provide evidence that the AI–RSC FCs before cocaine self-administration may influence downstream reversal of this observed circuit connectivity changes following 30 d of abstinence in a manner depending on the level of cocaine exposure. These findings highlight the critical relationship between individual's AI and RSC as a consequence and moderator of cocaine-induced changes in the brain.
Limitations of the current study
Our study provides novel insights into the neuronal alterations following cocaine self-administration and extended (1 month) experimenter-imposed abstinence; however, there are several limitations of the current design. First, the study included only male rats. While prior research has indicated higher acquisition and reinstatement rates in cocaine self-administering female rats compared with males (Lynch and Carroll, 1999; Becker and Koob, 2016; Moschak et al., 2023), including females would have added another level of complexity to our current experimental design, particularly problematic given the longitudinal and intricate behavior/imaging protocols we employed. Future studies are needed to examine potential sex differences in network connectivity during self-administration and following prolonged abstinence. Second, while our study has identified the influences of cocaine and water self-administration and extended abstinence on fcMRI metrics, another limitation may be related to the timing of water and food restrictions. The baseline fcMRI measurements were completed prior to these water and/or food restriction and may therefore offer a more “neutral” gauge of the brain's intrinsic FC. However, mild food and water restrictions may have influenced subsequent fcMRI metrics during the self-administration stage, and future studies are needed to address this potential concern. Third, while the insights derived from our research are undeniably valuable, our relatively small sample size may have potentially limited the broader applicability of our findings. To mitigate concerns linked to the small sample size, we employed the permutation test, a potent nonparametric technique less susceptible to sample size-related biases (Ernst, 2004; Önder, 2007). While this approach is well-established, it might not completely offset the inherent limitations of a small sample. Finally, our choice to focus on distinct time points postadministration (1 d and 30 d of abstinence) means we may not have captured the complete spectrum of neuronal transformations that arise over extended cocaine abstinence periods. Subsequent investigations that integrate both genders, larger sample sizes, and more diverse postadministration durations could render a more comprehensive understanding of the neural implications of cocaine.
Conclusions
Our findings expand upon earlier rodent research that primarily investigated the effects of cocaine exposure through seed-based FC (Lu et al., 2014; Orsini et al., 2018), fMRI activity changes (Febo et al., 2005; Lu et al., 2007, 2012a; Johnson et al., 2013), and graph-level findings in an awake, restrained condition (Orsini et al., 2018). We aimed to investigate the effect of chronic cocaine exposure on large-scale brain network communication (i.e., DMN, SN, and LCN) using fcMRI, focusing on distinct phases of addiction. We discovered disrupted network–level interactions involving the RSC (a node of DMN), AI (a node of SN), and CPu, which have been implicated in self-referential functions, emotion, and memory and are responsible for coordinating between internal and external stimuli. The differential interactions among SN, DMN, and LCN appear broadly consistent with the “triple network model.” Specifically, our study underscores the importance of addressing the impact of cocaine on AI–RSC and AI–CPu FC and provides insights into the role of baseline AI–RSC FC in cocaine use disorders. Our longitudinal design and network analysis approach offer novel insights into the underlying FC mechanisms of cocaine addiction and abstinence, paving the way for new treatment strategies and predictive biomarkers for addiction severity and treatment outcomes.
Footnotes
We thank the members of the University of North Carolina Center for Animal MRI (CAMRI) for inputs. We also thank Drs. Garret Stuber and Heather Decot in supporting experimental design and data collection and Xuefei Wang for assistance with data collection.
This research was supported by the Extramural Research Programs of United States National Institutes of Health: National Institute on Drug Abuse (R21DA057503), National Institute of Mental Health (R01MH126518, RF1MH117053, R01MH111429, and S10MH124745), National Institute of Neurological Disorders and Stroke (R21NS133913, R01NS128278, R01NS091236, and RF1NS086085), National Institute on Alcohol Abuse and Alcoholism (U01AA020023 and P60AA011605), Eunice Kennedy Shriver National Institute of Child Health and Human Development (P50HD103573), and NIH High-End Instrumentation Grant (S10OD026796).
The authors declare no competing financial interests.
- Correspondence should be addressed to should be addressed to Li-Ming Hsu at liming_hsu{at}med.unc.edu or Yen-Yu Ian Shih at shihy{at}unc.edu.
