The transcriptional response to acute cocaine is inverted in male mice with a history of cocaine self-administration and withdrawal throughout the mesocorticolimbic system

Abstract

A large body of work has demonstrated that cocaine-induced changes in transcriptional regulation play a central role in the onset and maintenance of cocaine use disorder. An underappreciated aspect of this area of research, however, is that the pharmacodynamic properties of cocaine can change depending on an organism’s previous drug-exposure history. In this study, we utilized RNA sequencing to characterize how the transcriptome-wide effects of acute cocaine exposure were altered by a history of cocaine self-administration and long-term withdrawal (30 days) in the ventral tegmental area (VTA), nucleus accumbens (NAc), and prefrontal cortex (PFC) in male mice. First, we found that the gene expression patterns induced by a single cocaine injection (10 mg/kg) were discordant between cocaine-naïve mice and mice in withdrawal from cocaine self-administration. Specifically, the same genes that were upregulated by acute cocaine in cocaine-naïve mice were downregulated by the same dose of cocaine in mice undergoing long-term withdrawal; the same pattern of opposite regulation was observed for the genes downregulated by initial acute cocaine exposure. When we analyzed this dataset further, we found that the gene expression patterns that were induced by long-term withdrawal from cocaine self-administration showed a high degree of overlap with the gene expression patterns of acute cocaine exposure - even though animals had not consumed cocaine in 30 days. Interestingly, cocaine re-exposure at this withdrawal time point reversed this expression pattern. Finally, we found that this pattern was similar across the VTA, PFC, NAc, and within each brain region the same genes were induced by acute cocaine, re-induced during long-term withdrawal, and reversed by cocaine re-exposure. Together, we identified a longitudinal pattern of gene regulation that is conserved across the VTA, PFC, and NAc, and characterized the genes constituting this pattern in each brain region.

Introduction

Cocaine use disorder is a chronic neuropsychiatric condition characterized by long-lasting alterations in the neural circuitry regulating reward, motivation, and cognitive function (Aguilar et al., 2009; Le Moal & Koob, 2007; Mews & Calipari, 2017). Although the behavioral consequences are well-characterized, the molecular mechanisms controlling these uniquely long-lasting changes remain opaque (Nestler & Lüscher, 2019). A defining characteristic of substance use disorder is that it extends beyond the acute pharmacological effects of the drug (Lüscher, 2016; Lüscher & Janak, 2021; Robinson et al., 2001). In cocaine use disorder, specifically, the molecular effectors recruited by acute cocaine are altered in both their magnitude (the same effector recruited differentially after repeated dosing) and identity (new effectors recruited by subsequent exposure) over the course of continued drug use (Nestler, 2005; Robison & Nestler, 2011; Wolf, 2016). This process of cocaine-induced transcriptional regulation is known to be particularly important in brain regions involved in reinforcement, such as the ventral tegmental area (VTA) (Campbell et al., 2021; Vallender et al., 2017) and nucleus accumbens (NAc), (López et al., 2021) and regions involved in cognitive function, such as the prefrontal cortex (PFC) (Freeman et al., 2010; Sadakierska-Chudy et al., 2017). Here, using computational analysis of RNA sequencing from these brain regions in male mice with a range of cocaine self-administration histories we identify how gene expression programs are regulated by cocaine exposure at different time-points.

A key feature of substance use disorders is that the pharmacodynamic properties of a drug are highly influenced by an organism's previous exposure to that drug (Ferris, Calipari, Yorgason, et al., 2013; Mews & Calipari, 2017). For example, repeated exposure to cocaine can cause sensitization or tolerance to the effects of cocaine on the brain and behavior depending on the pattern and duration of drug administration (Calipari, Beveridge, et al., 2013; Calipari, Ferris, et al., 2013; Calipari, Ferris, Melchior, et al., 2014; Ferris, Calipari, Melchior, et al., 2013; Ferris et al., 2012; Grimm et al., 2001). This drug-induced behavioral plasticity is mediated, at least in part, by drug-induced transcriptional plasticity throughout reward circuits in both humans and animal models (Bastle & Maze, 2019; Childress et al., 1999; Maze & Nestler, 2011; Nestler, 2005; Robbins et al., 1997). Even during periods of abstinence, when the drug has been cleared from the system, changes within the brain can cause plasticity that controls craving and drug seeking (Walker et al., 2018; Wolf, 2016). Importantly, all of the aforementioned effects persist long beyond the half-life of individual proteins (Horikawa & Nawa, 1998; Pratt et al., 2002), suggesting that cocaine induces long-lasting changes in transcriptional activity within the brain (Hope et al., 1994; Im et al., 2010; McClung et al., 2005). Although there is a wealth of data showing the critical role that transcription plays in cocaine reinforcement, many of these studies have focused on candidate genes within individual brain regions (Cadet et al., 2013; Cahill et al., 2018; Chandra et al., 2015; Feng et al., 2014; Freeman et al., 2008; Gancarz-Kausch et al., 2013; Graham et al., 2007; Larson et al., 2010; Lu et al., 2006; Walker et al., 2018; Werner et al., 2018). These approaches have delivered novel insight into the mechanisms by which cocaine alters transcriptional processes, but likely do not capture the full complexity of drug-induced transcriptional dysregulation. Further, precisely how drug-induced gene expression changes as a function of self-administration history remains poorly understood. Recently, we and others have generated several transcriptional datasets using RNA sequencing at several time points following acutely administered and self-administered cocaine in mice (Walker et al., 2018). This allows for comprehensive analysis of wide-scale gene expression networks that occur in response to acute non-contingent cocaine exposure, self-administration, withdrawal, and re-exposure.

Using these data (Walker et al., 2018) and building upon them with a series of novel analyses, we identify how the transcriptional profile recruited by acute cocaine exposure (via IP injection) is altered by a history of volitional cocaine self-administration in male mice. In this study, we characterized how transcription changes as a function of cocaine exposure history across the mesocorticolimbic system (VTA, NAc, PFC). We performed this analysis to understand: 1) how the transcriptional program induced by cocaine itself is changed by a history of cocaine self-administration and withdrawal and 2) how the expression of genes within these cocaine-recruited transcriptional programs changes throughout cocaine withdrawal. Together, we show that acute cocaine results in the opposite recruitment of transcriptome-wide patterns following a history of self-administration compared to drug-naïve controls. The discordant gene expression relationship between acute cocaine and cocaine re-exposure likely results from re-recruitment of the same genes induced by acute cocaine during long-term withdrawal (in the absence of cocaine itself) and the subsequent reversal of expression of these same genes upon re-exposure to cocaine.

Materials and Methods

All analyses in this manuscript were performed on a previously published and freely available RNA sequencing dataset [Gene expression omnibus accession number (GEO): GSE110344]. Thus, the experimental subjects, surgery procedure and behavioral task, and RNA sequencing procedures were extensively described previously (Walker et al., 2018). We have described key methodological details briefly below.

Experimental subjects.

Male 6-8 week-old C57BL/6J mice weighing 20-24 grams were maintained on a 12:12 hour reverse light-dark cycle at 22-25°C with ad libitum access to food and water, except during training and testing when access to food was restricted. During self-administration testing mice were food restricted to 95% of their free-feeding weight. Mice were housed 5 per cage prior to jugular vein catheterization surgeries, after which mice were housed individually. Following self-administration, animals in the withdrawal groups were rehoused with their original cage mates for the remainder of the experiment. All experiments were conducted in accordance with the guidelines of the Institutional Animal Care and Use Committee at Icahn School of Medicine at Mount Sinai.

Training, Surgery and Self-Administration.

Operant Training:

Mice were trained for food reinforcement in standard mouse operant chambers (Med Associates, St Albans, VT) equipped with two retractable levers, a cue light, and a house light. Illumination of the house light and extension of the levers signaled the beginning of the self-administration session. Active lever presses resulted in food pellet delivery followed by a 20 second time-out period during which a cue light was illuminated, and levers were retracted. Responding on the inactive lever was recorded but resulted in no programmed consequence. Responding on the active lever was reinforced on a fixed-ratio 1 (FR1) schedule. Animals were considered to have acquired when they exhibited stable responding on the active lever (60% active/total lever presses) and >10 lever presses per 1 hour session. Once the animals met acquisition criteria, they were moved onto an FR2 schedule to further confirm acquisition of the task.

Cocaine Self-Administration and Withdrawal:

Following food training, mice were implanted with a jugular catheter (0.3 mm inner and 0.6mm outer diameter) under ketamine (100 mg/kg IP)-xylazine (10 mg/kg IP) anesthesia. After a recovery period, mice began self-administration for either cocaine or saline (which served as the control for these studies). During self-administration cocaine was available on a fixed-ratio 1 schedule of reinforcement (FR1) where each active response resulted in a single (0.03ml) infusion of cocaine (0.5 mg/kg/infusion over 3.25 seconds) and the illumination of a discrete light cue above the active lever. Mice self-administered cocaine on this schedule for five consecutive days and were then switched to FR2 for 5 additional days (10 total days of self-administration) before completing this phase of the task. Mice were either sacrificed 24 hours following the final self-administration session (1 day withdrawal) or began a 30-day withdrawal/forced abstinence period. After 30 days of withdrawal/forced abstinence, mice were given an injection of either cocaine (10 mg/kg, IP) or saline, placed back in their original operant chamber with house light illuminated; however, the levers were not extended, and no other cues were presented. Animals were euthanized by cervical dislocation 1 hr after this final injection.

Experimental groups:

In the associated gene lists, animals are labeled based on their self-administration history. A schematic of the experimental design is provided in Fig 1a.

Figure 1: Cocaine self-administration alters cocaine-induced transcription across the mesocorticolimbic system.

(a) Mice underwent either saline or cocaine (0.5mg/kg/inj) self-administration (SA) for ten days followed by a 30-day withdrawal period. On day 30 mice were sacrificed 1 hour following either saline or cocaine injection (10 mg/kg, IP). To determine differential gene expression induced by acute cocaine (green), mice that administered saline and received a saline injection (cocaine-naïve mice) were compared to mice that self-administered saline and received a cocaine injection (first-time acute cocaine exposure). To determine differential gene expression induced by cocaine re-exposure (purple), mice that self-administered cocaine and received a cocaine injection (acute cocaine re-exposure) were compared to mice that self-administered cocaine and received a saline injection (30-day cocaine withdrawal) (b) Stratified Rank-rank hypergeometric overlap (RRHO) was used to compare the relationship between differential gene expression induced by acute cocaine and cocaine re-exposure in a threshold-free fashion. Results are displayed as a heatmap with more significantly overlapping gene expression denoted by warmer colors and less significantly overlapping gene expression denoted by cooler colors. Hotspots in the bottom left and top right quadrants of the heatmap indicate a concordant gene expression relationship between the two lists (i.e. the same genes are upregulated and downregulated in both lists) whereas hotspots in the bottom right and top left quadrants of the heatmap indicate a discordant gene expression relationship between the two lists (i.e. downregulated and upregulated genes in one condition are upregulated and downregulated, respectively, in the other list). Stratified RRHO comparisons of differential gene expression induced by acute cocaine and cocaine re-exposure were performed in the (c) ventral tegmental area (VTA; genes compared in x-axis and y-axis = 13,317, maximum −log10(hypergeometric p-value) = 63.5; number most significantly overlapping genes in top left quadrant = 3,486, number most significantly overlapping genes in bottom right quadrant = 2338), (d) nucleus accumbens (NAc; genes compared in x-axis and y-axis = 13,232, maximum −log10(hypergeometric p-value) = 129.3, number most significantly overlapping genes in top left quadrant = 3,044, number most significantly overlapping genes in bottom right quadrant = 2,580), (e) prefrontal cortex (PFC; genes compared in x-axis and y-axis = 13,092, maximum −log10(hypergeometric p-value) = 75.1, number most significantly overlapping genes in top left quadrant = 3,637, number most significantly overlapping genes in bottom right quadrant = 2,647), and (f) caudate putamen (CPU; maximum −log10(hypergeometric p-value) = 11.0). This analysis revealed that acute cocaine and cocaine re-exposure exhibited opposite patterns of transcriptional regulation of the same genes in the VTA, NAc, and PFC (c-e), but that this discordant gene expression relationship was not present in the CPU (f).

1. Sal-Sal: 10 days of saline self-administration, 30 days of withdrawal, and a saline injection (cocaine-naïve).

2. Sal-Coc: 10 days of saline self-administration, 30 days of withdrawal, and an IP cocaine injection (first-time acute cocaine exposure).

3. Coc-Sal: 10 days of cocaine self-administration, 30 days of withdrawal, and a saline injection (30-day cocaine withdrawal).

4. Coc-Coc: 10 days of cocaine self-administration, 30 days of withdrawal, and an IP cocaine injection (cocaine re-exposure).

Two additional groups did not undergo long-term withdrawal, but rather were sacrificed 1 day after the last self-administration session. These mice were matched for age and time in the animal colony.

1. Sal-None: 10 days of saline self-administration, sacrificed 24 hours following the last session.

2. Coc-None: 10 days of cocaine self-administration, sacrificed 24 hours following the last session.

RNA Isolation, Library Preparation, and RNA-Sequencing.

Mice were euthanized by cervical dislocation, and a number of brain regions were dissected and rapidly and frozen on dry ice as described (Walker et al., 2018). For the current study analyses was done on PFC, (caudate putamen) CPU, NAc, and VTA samples (from GEO: GSE110344). RNA was extracted from dissections obtained from individual animals and its integrity and concentration were assessed as described previously (Walker et al., 2018). Libraries were prepared using the TruSeq Stranded mRNA HT Sample Prep Kit protocol (Illumina, San Diego CA) and sequenced on a HighSeq2500 with 50 base pair single end reads.

Statistical and Bioinformatic Analysis.

Transcriptomic Analysis:

Pairwise differential expression comparisons were performed as reported previously (Bagot et al., 2016, 2017) using Voom Limma (Law et al., 2014) and a significance threshold of p < 0.05 was applied. All differential gene expression p-values in this manuscript refer to uncorrected p-values. Uncorrected p-values were used as input for Stratified Rank-Rank Hypergeometric Overlap (RRHO) analysis because this technique was specifically developed to enable threshold free gene expression comparisons. Uncorrected p-values were used for profiling expression of upstream regulators because this expression profiling was performed on groups of genes whose expression had been defined a priori, rather than hypothesis-free genome wide expression profiling. Differential gene expression lists for acute cocaine (sal-coc versus sal-sal; VTA, NAc, PFC, and CPU) short-term withdrawal (coc-none versus sal-none; VTA, NAc, and PFC), long-term withdrawal (coc-sal versus sal-sal; VTA, NAc, and PFC), and cocaine re-exposure (coc-coc versus coc-sal; VTA, NAc, PFC, and CPU) are provided in Extended Data 1-8 and 9-14.

Stratified RRHO:

Stratified RRHO heatmaps and lists of genes corresponding to RRHO hotspots for the indicated comparisons were generated using the R package RRHO2 (Cahill et al., 2018; Plaisier et al., 2010). Stratified RRHO enables threshold-free comparisons (i.e. no differential gene expression p-value thresholds) of gene expression patterns and the identification of overlapping sets of genes between two conditions. Stratified RRHO is an improvement upon older implementations of RRHO, the enrichment and two-sided methods (Cahill et al., 2018). These previous implementations reliably detected concordant overlapping gene expression (the same genes upregulated and downregulated in two conditions), but not discordant overlapping gene expression (opposite gene expression regulation in two conditions). In contrast, Stratified RRHO, which is employed in the current study, reliably detects both concordant and discordant expression.

To perform Stratified RRHO, two lists of common genes to be compared are ranked by their degree of differential expression [−log10(uncorrected p-value) * sign(effect size)]; under this ranking system, upregulated genes will have a positive degree of differential expression whereas downregulated genes will have a negative degree of differential expression. The first ranked list lies along the x-axis with the leftmost gene having the largest degree of differential expression (most upregulated) and the rightmost gene having the smallest degree of differential expression (most downregulated). The second ranked list makes up the y-axis with the gene having the largest degree of differential expression (most upregulated) at the bottom and the gene having the smallest degree of differential expression (most downregulated) at the top. This ranking procedure along the x-axis and the y-axis yields four quadrants (schematic description in Figure 1B): a bottom left quadrant that contains genes that are upregulated in the first list and upregulated in the second list; a top right quadrant that contains genes that are downregulated in the first list and downregulated in the second list; a top left quadrant that contains genes that are upregulated in the first list and downregulated in the second list; and a bottom right quadrant that contains genes that are downregulated in the first list and upregulated in the second list. Next, the significance of overlap between the two conditions is determined within each of the four quadrants. Specifically, the significance of overlap for all possible combinations of genes from the x-axis and the y-axis at a defined step-length beginning from the outermost corner of each quadrant is denoted with the −log10(hypergeometric p-value). Results can then be visualized as a heatmap with more significantly overlapping gene expression denoted by warmer colors and less significantly overlapping gene expression denoted by cooler colors (see Figure 1c-f). Hotspots in the bottom left and top right quadrants of the heatmap indicate a concordant gene expression relationship between the two conditions (i.e. the same genes are upregulated and downregulated in both conditions) whereas hotspots in the bottom right and top left quadrants of the heatmap indicate a discordant gene expression relationship between the two conditions (i.e. downregulated and upregulated genes in one condition are upregulated and downregulated, respectively, in the other condition). The most significantly overlapping genes within each quadrant can then be extracted by selecting the combination of genes from the x-axis and the y-axis that yield the largest −log10(hypergeometric p-value) within the quadrant.

During the performance of Stratified RRHO, correction for multiple comparisons was performed on the hypergeometric p-values denoting the significance of gene expression overlap using the Benjamini–Yekutieli procedure (Plaisier et al., 2010) except for the comparisons of acute cocaine and long-term cocaine withdrawal in the VTA and PFC and the comparison of cocaine re-exposure and long-term cocaine withdrawal in the VTA, NAc, and PFC; in these cases the uncorrected hypergeometric p-values were so close to zero that applying the multiple comparisons correction procedure and performing the −log10() transformation produced undefined, infinitely large numbers. Lists of genes corresponding to RRHO hotspots used in the present study are provided in Extended Data 15-32.

Log odds method for overlap detection.

To improve the rigor of this work, we also verified overlap using the Log Odds Method (Cahill et al., 2018). This complementary implementation of Stratified RRHO calculates overlap independent of sample size. Thus, it can be used to ensure that overlapping gene expression detected via Stratified RRHO is biologically meaningful, rather than spurious overlap detected due to comparisons of overlap between two long lists of genes. All analyses resulted in the same conclusions between Log Odds and Stratified RRHO Methods; thus, stratified RRHO was used to identify gene lists for all subsequent analyses.

Gene list intersection and venn diagram generation:

The intersection of gene lists for Stratified RRHO hotspots across comparisons of acute cocaine and cocaine re-exposure, acute cocaine and long-term cocaine withdrawal, and cocaine re-exposure and long-term cocaine withdrawal within each brain region was computed using custom code written in python and proportional venn diagrams representing this intersection were generated using the python module Biovenn (Hulsen, 2021). This resulted in the generation of two longitudinally intersecting sets of genes each within the VTA, NAc, and PFC: one set upregulated by acute cocaine and long-term cocaine withdrawal and subsequently downregulated by cocaine re-exposure (“longitudinal gene set I”) and another set downregulated by acute cocaine and long-term cocaine withdrawal and subsequently upregulated by cocaine re-exposure (“longitudinal gene Set II”). Lists of genes making up longitudinal gene set I and longitudinal gene set II for each brain region are provided in Extended Data 33-38. The probability of these multi-list intersections and the fold of intersection enrichment were calculated using the R package SuperExactTest (Wang et al., 2015). For the intersection statistics associated with longitudinal gene set I and longitudinal gene set II, the total number of population genes was the number of genes in the different gene expression list in the respective brain regions (VTA: 13,317, NAc: 13,232, PFC: 13,092). For the intersection statistics associated with the inter-brain-region intersection of longitudinal gene set I and longitudinal gene set II, the total number of population genes was the total number of non-redundant genes sequenced across all three brain regions (14,290). All of the SuperExactTest output is provided in Extended Data 39.

Generation of line graphs representing relative expression data:

The log2(fold-change) of genes within longitudinal gene set I and longitudinal gene set II for the VTA, NAc, and PFC were selected from differential gene expression lists using custom code written in python. The median log2(fold-change) and interquartile range were then calculated and plotted using Graphpad Prism 9.4.1 (La Jolla, CA). The data used to generate line graphs is provided in Extended Data 40-45.

Biological Pathway Enrichment Analysis:

Genes within longitudinal gene set I and longitudinal gene set II for the VTA, NAc, and PFC were subjected to overrepresentation analysis using g:profiler version e107_eg54_p17_bf42210 (Raudvere et al., 2019), which was queried using the python package gprofiler-official (https://pypi.org/project/gprofiler-official/) to identify significantly enriched biological pathways from the Gene Ontology Database and Reactome Pathway Database (Gillespie et al., 2022). Full overrepresentation analysis results for longitudinal gene set I and longitudinal gene set II for the VTA, NAc, and PFC are provided in Extended Data 46-51.

Upstream Regulator Analysis:

Predicted protein Master Regulators of the genes within longitudinal gene set I and longitudinal gene set II for the VTA, NAc, and PFC were identified using Causal Network analysis within Ingenuity Pathway Analysis (IPA) Software (Qiagen, Fredrick MD). Using custom code written in python, the results obtained using IPA were then filtered to include upstream regulators of a non-pharmacological molecule type only ('G-protein coupled receptor', 'complex' ,'cytokine', 'enzyme', 'fusion gene/product', 'group', 'growth factor', 'ion channel', 'kinase', 'ligand-dependent nuclear receptor', 'mature microRNA', 'microRNA', 'peptidase', 'phosphatase', 'transcription regulator', 'translation regulator', 'transmembrane receptor', and 'transporter'). The upstream regulator analysis results containing Master Regulators of longitudinal gene set I and longitudinal gene set II for the VTA, NAc, and PFC are provided in Extended Data 70-75.

Generation of volcano plots representing expression of upstream regulators:

Predicted upstream transcriptional regulators of longitudinal gene set I and longitudinal gene set II were mapped to their Ensembl gene IDs using g:profiler (Raudvere et al., 2019), which was queried using the python package gprofiler-official (https://pypi.org/project/gprofiler-official/). The intersection and non-intersection of the predicted protein upstream regulators of the genes within longitudinal gene set I and longitudinal gene set II for the VTA, NAc, and PFC was computed using custom code written in python. The log2(fold-change) and uncorrected p-value for the differential gene expression of these upstream regulators, if they were profiled in the RNA sequencing data, were selected following acute cocaine, short-term cocaine withdrawal, and long-term cocaine withdrawal, using custom code written in python. Uncorrected p-values were used because this expression profiling was performed on groups of genes whose expression had been defined a priori, rather than hypothesis-free genome wide expression profiling. Volcano plots representing the log2(fold-change) and uncorrected p-values based on differential gene expression for upstream regulators of longitudinal gene set I and longitudinal gene set II in the VTA, NAc, and PFC following acute cocaine, short-term cocaine withdrawal, and long-term cocaine withdrawal were generated using Graphpad Prism 9.4.1 (La Jolla, CA). The data used to generate volcano plots is provided in Extended Data 76-87. Genes encoding significantly regulated (uncorrected p < 0.05) predicted upstream transcriptional regulator log2(fold-change) and uncorrected p-value for pairwise differential comparisons, as well as their summary descriptions from the National Center for Biotechnology Information (Sayers et al., 2022), are provided in Extended Data 88-99.

Statistics and data availability:

Type I error rate (alpha) was set to 0.05 for all statistical tests. Data were analyzed using custom code written in python and R, as well as Graphpad Prism 9.4.1 (La Jolla, CA).

Results

Withdrawal from cocaine self-administration alters the transcriptional patterns induced by acute cocaine exposure.

To understand how a history of cocaine self-administration and withdrawal influenced cocaine-induced transcription across the mesocorticolimbic system, we compared the transcriptional program induced by acute cocaine (10 mg/kg, IP) in the VTA, NAc, PFC, and CPU of cocaine-naïve animals to the transcriptional program induced by re-exposure to the same dose of acute cocaine in animals with a history of cocaine self-administration (Fig. 1a). First, animals underwent cocaine self-administration (0.5 mg/kg/inj) for 10 consecutive days, followed by a 30-day withdrawal period. Saline controls had the same surgical procedure, exposure to operant conditioning chambers, and handling experience, but only experienced saline and thus were cocaine naïve. At the end of the 30-day withdrawal period mice were injected with either cocaine (10 mg/kg, IP) or saline. All mice were sacrificed 1 hour following the last injection. This resulted in four groups: saline self-administration with saline challenge (sal-sal, cocaine-naïve), saline self-administration with cocaine challenge (sal-coc, first-time acute cocaine exposure), cocaine self-administration with 30-day withdrawal (coc-sal, 30-day cocaine withdrawal), and cocaine self-administration with a cocaine challenge (coc-coc, acute cocaine re-exposure).

The goal of this series of analyses was to identify the sets of genes induced by initial acute cocaine exposure and understand how the expression of these genes was altered by self-administration and long-term withdrawal. To this end, differential gene expression lists were generated from the four groups described above to define the differential gene expression induced by initial acute cocaine exposure (hereafter referred to as “acute cocaine”) and the differential gene expression induced by re-exposure to acute cocaine (hereafter referred to as “cocaine re-exposure”). The acute cocaine differential gene expression list was generated by comparing the first-time acute cocaine exposure group (sal-coc) to the cocaine-naïve group (sal-sal) (Fig 1a, top, green; Extended Data 1-4). The cocaine re-exposure differential gene expression list was generated by comparing the acute cocaine re-exposure group (coc-coc) to the 30-day cocaine withdrawal group (coc-sal); this comparison enabled us to control for any latent transcriptional differences induced by withdrawal to isolate only differential gene expression induced by cocaine re-exposure (Fig 1a, bottom, purple; Extended Data 5-8).

The acute cocaine and cocaine re-exposure differential gene expression lists were then compared to assess acute cocaine-induced transcriptional programs in each brain region using Stratified RRHO (Fig. 1b) (Cahill et al., 2018; Plaisier et al., 2010). Stratified RRHO enables threshold-free comparisons of gene expression patterns and the identification of overlapping sets of genes between two conditions. To perform Stratified RRHO, lists of genes in two conditions are ranked by their degree of differential gene expression [−log10(uncorrected p-value) * sign(log2(fold-change))]. The first ranked list is positioned along the x-axis with the most upregulated gene on the left and the most downregulated gene on the right; the second ranked list is positioned on the y-axis with the most upregulated gene on the bottom and the most downregulated gene on the top. This ranking procedure along the x-axis and the y-axis yields four quadrants (Fig 1b): a bottom left quadrant that contains genes that are upregulated in the first list and upregulated in the second list; a top right quadrant that contains genes that are downregulated in the first list and downregulated in the second list; a top left quadrant that contains genes that are upregulated in the first list and downregulated in the second list; and a bottom right quadrant that contains genes that are downregulated in the first list and upregulated in the second list. The significance of gene expression overlap between the two conditions is then calculated within each of the four quadrants via hypergeometric test and denoted by −log10(hypergeometric p-value). Hotspots in the bottom left and top right quadrants of the heatmap indicate a concordant gene expression relationship between the two lists (i.e. the same genes are upregulated and downregulated in both lists) whereas hotspots in the bottom right and top left quadrants of the heatmap indicate a discordant gene expression relationship between the two lists (i.e. downregulated and upregulated genes in one condition are upregulated and downregulated, respectively, in the other list).

Stratified RRHO comparisons of the transcriptional programs induced by acute cocaine and cocaine re-exposure revealed a discordant gene expression relationship in the VTA (Fig. 1c; maximum −log10(hypergeometric p-value) = 63.5), NAc (Fig. 1d; maximum −log10(hypergeometric p-value) = 129.3) and PFC (Fig. 1e; maximum −log10(hypergeometric p-value) = 75.1). In contrast, this discordant relationship was not observed in the CPU (Fig. 1f; maximum −log10(hypergeometric p-value) = 11.0). This indicated that genes that were upregulated by acute cocaine were downregulated by cocaine re-exposure. Likewise, genes that were downregulated by acute cocaine were upregulated by cocaine re-exposure. In other words, acute cocaine and cocaine re-exposure exhibited opposite patterns of transcriptional regulation of the same genes in the VTA, NAc and PFC – but not CPU. The same discordant gene expression relationship was also observed in the VTA, NAc, and PFC – but not the CPU – when these comparisons were made using the Log Odds Method (Supplementary Fig. 1c-f). Further confirming that this effect was in fact biological and not simply a result of overlap occurring between large groups of genes, no overlapping relationship was observed between these conditions when the order of the genes in each list was randomly scrambled and the comparison was made using either Stratified RRHO (Supplementary Fig. 2a-d) or the Log Odds Method (Supplementary Fig. 2e-h). Therefore, the transcriptome-wide effects of acute cocaine across the VTA, NAc, and PFC were not just different, but opposite after long-term withdrawal from cocaine self-administration in male mice.

Transcription induced by long-term cocaine withdrawal is concordant with acute cocaine-induced transcription and discordant with cocaine re-exposure.

Because we observed a specific pattern present in only the VTA, NAc, and PFC, we continued our analysis on these brain regions. We wanted to understand the relationship between the transcriptional patterns induced by acute cocaine and the gene expression observed at other time points in an animal's self-administration history. For example, we were interested in determining if there was an overlap between gene expression induced by acute cocaine exposure and gene expression occurring at different withdrawal time points, when no cocaine was on board. These comparisons were made in the VTA, NAc, and PFC using Stratified RRHO (Fig. 2a) and confirmed via the Log Odds method (Supplementary Fig. 3).

Figure 2: Longitudinal comparisons of cocaine-induced transcription.

(a) Schematic of longitudinal comparisons for Stratified RRHO in the ventral tegmental area (VTA), nucleus accumbens (NAc), and prefrontal cortex (PFC). Comparisons were first made between differential gene expression induced by acute cocaine and differential gene expression induced by two withdrawal timepoints: short-term (1 day; panel b) withdrawal from cocaine self-administration or long-term (30-day; panel c) withdrawal from cocaine self-administration. Differential gene expression induced by long-term cocaine withdrawal was then compared to differential gene expression induced by cocaine re-exposure (panel d). (b) Comparison of differential gene expression induced by acute cocaine to differential gene expression induced by short-term withdrawal revealed weak overall overlap across the VTA (i; maximum −log10(hypergeometric p-value)= 53.3), NAc (ii; maximum −log10(hypergeometric p-value) = 14.4) and PFC (iii; maximum −log10(hypergeometric p-value) = 10.5). (c) Comparison of differential gene expression induced by acute cocaine to differential gene expression induced by long-term withdrawal revealed a strong concordant gene expression relationship across the VTA (i; genes compared in x-axis and y-axis = 13,317, maximum −log10(hypergeometric p-value) = 293.2, number most significantly overlapping genes in bottom left quadrant = 4,329, number most significantly overlapping genes in top right quadrant = 4,022), NAc (ii; genes compared in x-axis and y-axis = 13,232, maximum −log10(hypergeometric p-value) = 103.7, number most significantly overlapping genes in bottom left quadrant = 2,054, number most significantly overlapping genes in top right quadrant = 3,920), and PFC (iii; genes compared in x-axis and y-axis = 13,092, maximum −log10(hypergeometric p-value) = 491.7, number most significantly overlapping genes in bottom left quadrant = 5,493, number most significantly overlapping genes in top right quadrant = 2,691). (d) Comparison of differential gene expression induced by long-term cocaine withdrawal to differential gene expression induced by cocaine re-exposure revealed strong discordant gene expression relationship across the VTA (i; genes compared in x-axis and y-axis = 13,317, maximum −log10(hypergeometric p-value) = 593.7, number most significantly overlapping genes in top left quadrant = 3,973, number most significantly overlapping genes in bottom right quadrant = 4,333), NAc (ii; genes compared in x-axis and y-axis = 13,232, maximum −log10(hypergeometric p-value) = 495.6, number most significantly overlapping genes in top left quadrant = 4,117, number most significantly overlapping genes in bottom right quadrant = 3,301), and PFC (iii; genes compared in x-axis and y-axis = 13,092, maximum −log10(hypergeometric p-value) = 615.5, number most significantly overlapping genes in top left quadrant = 3,510, number most significantly overlapping genes in bottom right quadrant = 4,641). Together, these results indicated that differential gene expression induced by acute cocaine is re-induced after long-term cocaine withdrawal (in the absence of drug) and reversed by cocaine re-exposure.

First, we compared the transcriptional program induced by acute cocaine to the transcriptional program induced by short-term (1 day) cocaine withdrawal (Fig 2b; coc-none versus sal-none; Extended Data 9-11). This comparison revealed weak overlap overall between acute cocaine and short-term cocaine withdrawal across the VTA (Fig. 2bi; maximum −log10(hypergeometric p-value) = 53.3), NAc (Fig. 2bii; maximum −log10(hypergeometric p-value) = 14.4) and PFC (Fig. 2biii; maximum −log10(hypergeometric p-value) = 10.5). Weak gene expression overlap between acute cocaine and short-term cocaine withdrawal in the VTA, NAc, and PFC was confirmed via the Log Odds Method (Supplementary Fig. 3b).

Next, we compared the transcriptional program induced by acute cocaine to the transcriptional program induced by long-term (30 day) cocaine withdrawal (coc-sal versus sal-sal; Extended Data 12-14). This revealed a strong concordant relationship between gene expression induced by acute cocaine and long-term cocaine withdrawal across the VTA (Fig. 2ci; maximum −log10(hypergeometric p-value) = 293.2), NAc (Fig. 2cii; maximum −log10(hypergeometric p-value) = 103.7), and PFC (Fig. 2ciii; maximum −log10(hypergeometric p-value) = 491.7), suggesting that after long-term cocaine withdrawal (30-days) the same transcriptional program induced by acute cocaine is also induced by withdrawal across these brain regions. Discordant gene expression overlap between acute cocaine and long-term cocaine withdrawal in the VTA, NAc, and PFC was confirmed via the Log Odds Method (Supplementary Fig. 3c). Finally, Stratified RRHO comparison of long-term cocaine withdrawal and cocaine re-exposure revealed a strong discordant gene expression relationship across the VTA (Fig. 2di; maximum −log10(hypergeometric p-value) = 593.7), NAc (Fig. 2dii; maximum −log10(hypergeometric p-value) = 495.6), and PFC (Fig. 2diii; maximum −log10(hypergeometric p-value) = 615.5). The discordant gene expression relationship between long-term cocaine withdrawal and cocaine re-exposure in the VTA, NAc, and PFC was confirmed via the Log Odds Method (Supplementary Fig. 3d).

Together, the data above suggest that differential gene expression induced by acute cocaine is also induced after long-term cocaine withdrawal (in the absence of the drug) and then reversed by cocaine re-exposure (see Supplementary Fig. 4 for a full schematic of Stratified RRHO comparisons made in the present study). However, for this to be true, within each brain region, the same gene sets would need to be regulated along this pattern. We next tested this hypothesis.

Acute cocaine, long-term cocaine withdrawal, and cocaine re-exposure longitudinally regulate shared sets of genes.

We next determined whether the gene expression pattern we had observed via the Stratified RRHO comparisons reflects the regulation of the same sets of genes. Within each brain region, we extracted the most significantly overlapping genes from the hotspot quadrants in the RRHO comparisons of acute cocaine and cocaine re-exposure (Fig. 1c-e), acute cocaine and long-term cocaine withdrawal (Fig. 2ci-iii), and cocaine re-exposure and long-term cocaine withdrawal (Fig. 2di-iii). The most significantly overlapping genes within each quadrant were extracted by selecting the combination of genes from the x-axis and the y-axis that yielded the largest −log10(hypergeometric p-value) within each hotspot quadrant; Figure 2a and b provide a visualization of the RRHO hotspot quadrants from which these genes were extracted. Lists of genes extracted from these RRHO hotspots are provided in Extended Data 15-32.

Next, we examined whether the most significantly overlapping genes extracted from these hotspots intersected within each brain region. This analysis revealed two sets of genes in the VTA, NAc, and PFC intersecting above chance. The first set, hereafter referred to as “longitudinal gene set I”, was upregulated by acute cocaine, upregulated by long-term cocaine withdrawal, and downregulated by cocaine re-exposure [VTA: 2796 genes (Fig. 3c; intersection p-value < 0.0001, 7.6 intersection fold enrichment), NAc: 1515 genes (Fig. 3e; intersection p-value < 0.0001, 12.9 intersection fold enrichment), PFC: 3079 genes (Fig. 3g; intersection p-value < 0.0001, 5.7 intersection fold enrichment)]. The second set, hereafter referred to as “longitudinal gene set II”, was downregulated by acute cocaine, downregulated by long-term cocaine withdrawal, and upregulated by cocaine re-exposure [VTA: 1866 genes (Fig. 3d, intersection p-value < 0.0001, 8.9 intersection fold enrichment), NAc: 2034 genes (Fig. 3f, intersection p-value < 0.0001, 8.6 intersection fold enrichment), and PFC: 1942 genes (Fig. 3h, intersection p-value < 0.0001, 13.3 intersection fold enrichment)]. Here, the term longitudinal denotes the profiling of the same group of genes over time. Lists of genes making up longitudinal gene set I and longitudinal gene set II, as well as intersection statistics, are provided in Extended Data 33-38 and Extended Data 39, respectively. The median log2(fold-change) in expression of each of these gene sets is shown on the right of Fig. 3c-h (data used to generate line graphs is provided in Extended Data 40-45).

Figure 3: The same genes are induced by acute cocaine, re-induced during withdrawal from self-administration, and reversed by cocaine re-exposure.

(a-b) Schematic of Stratified Rank-Rank Hypergeometric Overlap (RRHO) hotspot quadrants that contain intersecting genes. The results of our Stratified RRHO comparisons indicated that differential gene expression induced by acute cocaine is re-induced after long-term (30 day) cocaine withdrawal (in the absence of drug) and reversed by cocaine re-exposure across the ventral tegmental area (VTA), nucleus accumbens (NAc), and prefrontal cortex (PFC). For this to be the case, however, the same genes would need to be present throughout this pattern within each of these brain regions. To determine if the same genes were induced at all three time points, we first extracted the most significantly overlapping genes within each hotspot quadrant of the Stratified RRHO comparisons of acute cocaine and cocaine re-exposure (dark gray), acute cocaine and long-term cocaine withdrawal (medium gray), and cocaine re-exposure and long-term cocaine withdrawal (light gray). We extracted the most significantly overlapping genes by selecting the combination of genes from the x-axis and the y-axis yielding the largest −log10(hypergeometric p-value) within each hotspot quadrant. We next examined the intersection of the most significantly overlapping genes from these three comparisons within the VTA, NAc, and PFC. Within each of these three brain regions, we identified two sets of genes that intersected above chance: (a) genes upregulated by acute cocaine (dark gray), upregulated by withdrawal (medium gray), and reversed/downregulated by cocaine re-exposure (light gray; intersection illustrated in red in venn diagrams below), or (b) genes downregulated by acute cocaine (dark gray), downregulated by withdrawal (medium gray), and reversed/upregulated by cocaine re-exposure (light gray; intersection illustrated in green in venn diagrams below). (c) Venn diagrams (left) denoting the intersection (red) of genes from hotspot quadrants illustrated in panel a, and line graphs (right) denoting the expression of these genes. (d) Venn diagrams (left) denoting the intersection (green) of genes from hotspot quadrants illustrated in panel b and line graphs (right) denoting the expression of these genes. The same pattern of above chance intersection was observed in the NAc (e, f) and PFC (g, h). Thus, the same sets of genes are induced by acute cocaine, re-induced by long-term withdrawal, and reversed by re-exposure across the VTA, NAc, and PFC. Line graphs reported as median log2(fold-change) ± interquartile range. Wilcoxin matched-pairs signed rank test was used to determine significance.

Further, when we examined the median log2(fold-change) in expression of longitudinal gene set I following short-term cocaine withdrawal, we observed that the genes upregulated by acute cocaine were downregulated relative to acute cocaine by short-term cocaine withdrawal–near baseline–before being upregulated again by long-term cocaine withdrawal in the VTA (Supplementary Fig. 5a), NAc (Supplementary Fig. 5c), and PFC (Supplementary Fig. 5e). Likewise, when we examined the median log2(fold-change) in expression of longitudinal gene set II following short-term cocaine withdrawal, we observed that the genes downregulated by acute cocaine were upregulated relative to acute cocaine by short-term cocaine withdrawal before being downregulated by long-term cocaine withdrawal in the VTA (Supplementary Fig. 5b), NAc (Supplementary Fig. 5d), and PFC (Supplementary Fig. 5f). This result was consistent with our Stratified RRHO comparison of the transcriptional program induced by acute cocaine to the transcriptional program induced by short-term cocaine withdrawal, which revealed weak overlap at this time point, but is nonetheless interesting.

Finally, although we observed the same pattern of transcriptional regulation across the VTA, NAc, and PFC, the genes making up longitudinal gene sets I and II were minimally non-overlapping across these brain regions (Supplementary Fig. 6a and b; Extended Data 39).

Longitudinal gene sets are involved in metabolism, regulation of gene expression, and neuronal processes.

After identifying these longitudinal gene sets, we next characterized the biological pathways associated with longitudinal gene set I (red) and longitudinal gene set II (green) in each brain region (Fig. 4a). Using over-representation analysis, we identified biological pathways from the Reactome Pathway Database (Gillespie et al., 2022) that were significantly enriched in each gene set. This analysis revealed that the most enriched biological pathways associated with longitudinal gene set I (top, red) and longitudinal gene set II (bottom, green) are involved in transcription, translation, and metabolism across the VTA (Fig. 4b), NAc (Fig. 4c), and PFC (Fig. 4d). One exception to this were the biological pathways associated with longitudinal gene set II in the PFC; the most enriched biological pathways in this gene set were involved in synaptic structure and function, chemical signal transmission, and postsynaptic signal receptors by potassium channels. Finally, these gene sets also exhibited lower enrichment levels for neuronal biological pathways (full overrepresentation analysis results are provided in Extended Data 46-51). Gene ontology enrichment analysis results for these gene sets in the VTA, NAc, and PFC are also provided for molecular function (Extended Data 52-57), biological process (Extended Data 58-63), and cellular component (Extended Data 64-69).

Figure 4: Enriched biological pathways and predicted upstream regulators of longitudinally regulated gene sets.

(a) Schematic of identified gene set I (red) and gene set II (green). (b-d) Biological pathways from the Reactome Pathway Database (some Reactome titles were shortened for figure presentation, full names are available for all groups in supplementary data) enriched in longitudinal gene set I and longitudinal gene set II in the ventral tegmental area (b; VTA), nucleus accumbens (c; NAc), and prefrontal cortex (d; PFC) were characterized. (e-g) Predicted upstream transcriptional regulators of longitudinal gene set I ands longitudinal gene set II in the VTA, NAc, and PFC were identified using Causal Network Analysis and filtered to include non-pharmacological upstream regulators only. Predicted upstream transcriptional regulators of longitudinal gene set I and longitudinal gene set II were cross-referenced against the differential gene expression lists for acute cocaine, short-term (1 day) cocaine withdrawal, and long-term (30 day) cocaine withdrawal to identify significantly regulated upstream regulators at these time points. Expression of these upstream transcriptional regulators is displayed via volcano plots; the color of the points on these plots denotes whether transcriptional regulators are predicted to act upstream of longitudinal gene set I only, longitudinal gene set II only, or both longitudinal gene set I and II. (e) Expression of predicted upstream regulators induced by acute cocaine in the VTA (i), NAc (ii), and PFC (iii). (f) Expression of predicted upstream regulators induced following short-term withdrawal from cocaine self-administration in the VTA (i), NAc (ii), and PFC (iii). (g) Expression of predicted upstream regulators induced by long-term withdrawal from cocaine self-administration in the VTA (i), NAc (ii), and PFC (iii).

Acute cocaine and cocaine withdrawal alters the expression of predicted upstream transcriptional regulators of longitudinal gene sets.

Our data thus far indicated that cocaine self-administration and withdrawal cause wide-scale changes in transcriptional regulation across the mesocorticolimbic system. Thus, we posited that there could be upstream transcriptional regulators within each brain region induced following acute cocaine exposure or withdrawal from cocaine self-administration that regulate these sets of genes in a coordinated fashion.

To this end, we first identified the predicted upstream regulators of longitudinal gene set I and longitudinal gene set II separately in the VTA, NAc, and PFC. Specifically, we identified molecular effectors, including G protein-coupled receptors, ion channels, enzymes, and nuclear receptors, that could explain the observed changes in gene expression. In the VTA, we identified 551 upstream regulators of longitudinal gene set I and 572 upstream regulators of longitudinal gene set II. In the NAc, we identified 563 upstream regulators of longitudinal gene set I and 530 upstream regulators of longitudinal gene set II. In the PFC, we identified 530 upstream regulators of longitudinal gene set I and 546 upstream regulators of longitudinal gene set II. Full lists of identified upstream regulators for longitudinal gene set I and longitudinal gene set II for the VTA, NAc, and PFC are provided in Extended Data 70-75.

Next, we cross-referenced these predicted upstream regulators against the differential gene expression lists for acute cocaine (Fig 1a, top, green), short-term cocaine withdrawal, and long-term cocaine withdrawal to identify upstream regulators significantly regulated (uncorrected p-value < 0.05 based on differential gene expression) at each time point. Uncorrected p-values were used because this is statistically appropriate as expression profiling was performed on groups of genes whose expression had been defined a priori. As mentioned above, we chose to profile upstream regulator expression at these time points to identify upstream transcriptional regulators induced by cocaine exposure and withdrawal. For this reason, we profiled upstream regulator expression following 1 day cocaine withdrawal even though we did not observe robust overlap in transcriptome-wide gene expression across the VTA, NAc, and PFC at this time point in our longitudinal RRHO comparisons (Fig. 2bi-iii). Further, because our aim was to characterize transcriptional dysregulation that would occur before re-exposure and could explain the effects on transcription we focused in the main figures on acute cocaine and withdrawal timepoints; however, we provide this analysis for the cocaine re-exposure group in Supplementary Fig. 7 and Extended Data 76-78.

Consistent with our hypothesis, we found that many of the predicted upstream regulators of longitudinal gene set I or longitudinal gene set II are significantly regulated (uncorrected p-value < 0.05) following acute cocaine exposure (VTA: Hspa8, Camk4, Trerf1, Nrip1, Psen2, Vav3, Slc30a7, Meis2, Itpr1, Fkbp4, Crebzf, Ppp3r1, Smarcd3; NAc: Vav3, Sik1, Lipe; PFC: Hspa8, Pcbd1, Gsr, Fkbp4, Txnrd1, Gtf2f2, Eef2k, Atp2b2), short-term cocaine withdrawal (VTA: Ptrf, Hmgb2, Gper1, Polk, Ptgs2, Casp3, Fdxr, Tdg; NAc: Sik1, Smad7, Star, Ptrf, Antxr1; PFC: Hspa8, Tead1, Cx3cr1, Zeb2, Ptrf, Smo, Gli3, Fzd7, Rrm2, Atp8b1, Chd7, Baiap2, Map4k4, Sp1, Galnt6, Foxo4, Carhsp1, Dlg3, Ube2g2, Rabgef1, Slc30a3, Tgfb2, Clec11a, Nfil3, Smad7, Otud7b, Itga6), and long-term cocaine withdrawal (VTA: Phb2, Zfp219, Agt, Nr1h3, Aip, Smarcc1, Cx3cr1, Pgr, Cdk12, Srebf1, Plagl1, Eif2ak4, Bckdha, Cyp27a1, Dmtf1; NAc: Mcl1, Zfp507, Sik1, Cacna1a, Lipe, Ebf1, Nfil3, Ptgs2, Crem, Irf2bp2; PFC: Slc30a5, Kcna3, Pcbd1, Adcy1, Txnrd1, Atp2b2). Some of the predicted upstream regulators significantly regulated at these time points are specific to longitudinal gene set I (red) or longitudinal gene set II (green); however, many are upstream of both gene sets (yellow). Significantly regulated upstream transcriptional regulators of both gene sets are particularly interesting, as they could mediate the wide-scale transcriptional dysregulation induced by acute cocaine exposure and withdrawal. The data used to generate volcano plots is provided in Extended Data 79-87. Genes encoding significantly regulated predicted protein upstream regulators and their summary descriptions from the National Center for Biotechnology Information (Sayers et al., 2022) are provided in Extended Data 88-99.

Discussion

Together, these results highlight several novel aspects of cocaine-induced dysregulation of transcription across the mesocorticolimbic system. We show that the transcriptional program induced by acute cocaine is reversed after a history of volitional cocaine self-administration and withdrawal in the VTA, NAc, and PFC, but not the CPU. Further, we found that the gene expression patterns induced by long-term withdrawal from cocaine self-administration showed a high degree of overlap with the gene expression patterns induced by acute cocaine - even though animals had not consumed cocaine in 30 days. These data show that long-term withdrawal recruits some of the same transcriptional patterns as acute exposure to the drug itself. Interestingly, cocaine re-exposure at this withdrawal timepoint reversed this expression pattern. Finally, while the gene sets that made up these patterns were largely non-overlapping across regions, the general transcriptional pattern was conserved across the VTA, NAc and PFC. Together, we provide evidence of robust withdrawal-induced reorganization of the transcriptome across the mesocorticolimbic system following cocaine self-administration.

While previous work using these datasets (Walker et al., 2018) outlined common gene expression patters across six brain regions, here we characterized how gene expression induced by a pharmacological challenge of acute cocaine is altered by a history of cocaine self-administration across the mesocorticolimbic system. We focused on the mesocorticolimbic system because we were specifically interested in cocaine-induced gene expression in the nucleus accumbens–a key center for integration of drug–associated learning–and the PFC and VTA, which project action-outcome information to the NAc (Chiara & Imperato, 1988; Hyman & Malenka, 2001; Kalivas, 2007; Koob et al., 1998; Lüscher & Janak, 2021; Mews & Calipari, 2017; Olds & Milner, 1954; Pascoli et al., 2015; Roberts et al., 1977). Further, rather than assessing gene expression across timepoints by identifying differentially expressed genes at an imposed significance threshold, we compared acute cocaine-induced gene expression from one time point to another using threshold-free Stratified RRHO and extracting the most-significantly overlapping genes across these conditions. Indeed, we found that the pharmacodynamic actions of cocaine on the transcriptional response in the NAc was changed over time in a way that could not be identified using other analyses.

The fact that the transcriptional response to acute cocaine is reversed after a history of cocaine self-administration is striking but not unsurprising in the broader context of research characterizing the effects of cocaine on reinforcement circuits. The dopamine system is the canonical neurotransmitter system targeted by stimulant drugs of abuse and serves as a major neuromodulatory signal in the VTA, NAc, and PFC (Calipari & Ferris, 2013; Goldman-Rakic, 1997; Nolan et al., 2020; Zachry et al., 2021). After long-term cocaine exposure, there are robust changes in dopamine release in striatal regions, alterations in the pharmacodynamic effects of cocaine at the dopamine transporter, and alterations in dopamine receptor signaling (Calipari, Beveridge, et al., 2013; Calipari, Ferris, & Jones, 2014; Ferris et al., 2012; Ferris, Calipari, Yorgason, et al., 2013; Ferris, Calipari, Melchior, et al., 2013; Ferris et al., 2015; Park et al., 2013; Tomasi et al., 2010; Volkow et al., 1997). Given the dramatic changes induced by long-term cocaine intake on its most basic pharmacodynamic target – the dopamine transporter – it is not surprising that cocaine effects on transcription also undergo wide-scale changes. However, to our knowledge, this study is the first to show that the transcriptome-wide effects of cocaine are reversed after a history of cocaine self-administration and withdrawal across several brain regions in the mesocorticolimbic circuit.

One particularly interesting finding is the fact that the genes induced by the first cocaine injection show highly overlapping patterns of expression with the genes induced during long-term withdrawal. It is tempting to speculate that the induction of these overlapping patterns could play a critical role in drug craving and seeking via the re-induction of cocaine-associated transcriptional patterns in the absence of cocaine itself during withdrawal. Preclinical studies have shown that animals exhibit increased responses to cocaine-associated cues with increasing withdrawal (Bossert et al., 2013; Gawin & Kleber, 1986; Grimm et al., 2001; Lu et al., 2003; Lu, Grimm, et al., 2004). These effects are mediated by molecular and synaptic adaptations induced across the mesocorticolimbic system by cocaine abstinence (Conrad et al., 2008; Lee et al., 2013; Lu, Dempsey, et al., 2004; Lu et al., 2009; Mameli et al., 2009; Wolf, 2016) and there is evidence for some aspects of this potentiation in response to cocaine-associated cues following abstinence occurring in human participants as well (Parvaz et al., 2016). Further, with respect to the reversal in transcriptional regulation following cocaine re-exposure observed in the present study, cocaine-dependent human participants report an immediate decline in subjective craving upon re-exposure to cocaine (Risinger et al., 2005). Thus, a potential, albeit speculative, possibility is that the sets of genes identified in the present study may play a role in some aspects of cocaine craving. It is also possible that the re-recruitment of transcriptional regulation at the long-term withdrawal point reflects cocaine craving or seeking elicited by re-exposure to the drug-associated context, as the animals in this group were returned to the operant chambers where they had previously self-administered cocaine before they were sacrificed. We defined this time point as long-term cocaine withdrawal a priori rather than cue exposure because by definition the animals were undergoing a period of withdrawal, but had never been exposed to discrete cues, nor had it been established empirically that the animals had formed an association between cocaine and the environment. In either case, these results provide novel information regarding changes in cocaine-associated transcription following withdrawal from cocaine self-administration.

An interesting finding observed through Stratified RRHO comparisons and profiling of the expression of longitudinal gene sets I and II is that the gene expression regulation induced by acute cocaine, re-induced by long-term withdrawal, and reversed by cocaine-re-exposure is not maintained throughout short-term (1 day) cocaine withdrawal; rather, this cocaine-induced gene expression seems to return to baseline after short-term withdrawal before being re-induced during long-term (30 day) cocaine withdrawal. In this study, we are agnostic as to the precise mechanistic underpinning of this return to baseline after short-term withdrawal, but this finding is consistent with the broader literature. Specifically, an emerging model is that persistent cellular and molecular dysregulation is initiated early during withdrawal and progresses in magnitude thereafter. For instance, silent synapses form early during withdrawal within the NAc and mature thereafter as withdrawal duration increases (Lee et al., 2013; Wolf, 2016). Further, Maze et al., 2010 report downregulation of the epigenetic regulator histone methyltransferase G9a in the NAc after short-term cocaine withdrawal, leading to latent changes in transcription (Maze et al., 2010). Taken with the above, the observed return to baseline after short-term cocaine withdrawal may reflect a period in which changes in transcriptional and epigenetic regulatory mechanisms have been initiated, but before more enduring cocaine-induced transcriptional dysregulation has emerged.

As a result of our upstream regulator analysis and expression profiling, we identified many predicted upstream transcriptional regulators of longitudinal gene sets I and II that are significantly induced after acute cocaine exposure or withdrawal in the VTA, NAc, and PFC. Many have not been explicitly interrogated for their role in cocaine-mediated reinforcement and are thus future candidates for investigation. Others, however, have been previously implicated in cocaine-mediated reinforcement. For instance, Smarcd3, which encodes a subunit of the SWI/SNF chromatin remodeling complex known to mediate behavioral responses to cocaine, was significantly upregulated after acute cocaine exposure in the VTA (Zayed et al., 2022). Camk4, which encodes a calcium/calmodulin-dependent protein kinase implicated in behavioral responses to cocaine, was also significantly upregulated after acute cocaine exposure in the VTA (Bilbao et al., 2008). In the NAc, Sik1, which encodes a kinase that is predicted to act upstream of both longitudinal gene sets I and II, was significantly upregulated by acute cocaine, 1-day cocaine withdrawal, and 30-day cocaine withdrawal. This finding is particularly notable because the kinase encoded by Sik1 has previously been characterized as a mediator of epigenetic adaptations induced by cocaine in the striatum (Dietrich et al., 2012). In addition, Crem, which encodes cAMP responsive element modulator, was significantly upregulated by 30-day cocaine withdrawal. The protein encoded by Crem is highly related in structure and function to CREB, a canonical regulator of cocaine-induced transcription in the NAc (Nestler et al., 2001).

In summary, we show that the transcriptional program induced by acute cocaine is reversed after a history of volitional cocaine self-administration in the NAc, VTA, and PFC. Further, we show that the discordant gene expression relationship between acute cocaine and cocaine re-exposure may be a consequence of re-induction of the transcriptional program induced by acute cocaine during long-term withdrawal from cocaine self-administration and the reversal of this transcriptional program upon re-exposure to cocaine. Finally, we found that there are intersecting sets of genes that follow this conserved transcriptional pattern across all three brain regions. Although novel at the transcriptome-wide level, these results parallel previously described neurobiological adaptations induced by cocaine and patterns of cocaine seeking and taking. Thus, these gene sets may underlie aspects of the cocaine-induced neural dysregulation that drives cocaine craving and relapse. Critically, however, these results illuminate that the neurobiological mechanisms underlying cocaine use disorder cannot be explained solely by the acute pharmacological effects of cocaine, as the pharmacodynamic properties of the drug can dramatically change as a consequence of long-term intake.

Highlights.

• Gene expression induced by acute cocaine is different after self-administration

• Genes upregulated by initial exposure are downregulated by re-exposure, and vice versa

• Acute cocaine responsive genes are induced by withdrawal and reversed by re-exposure

• The same genes are regulated across acute exposure, withdrawal, and re-exposure

• This pattern is present in the NAc, VTA, and PFC, but not the CPU