Introduction

Structural and functional adaptations within the neural reward circuitry are at the core of cocaine addiction (CA). These alterations promote craving for cocaine, while striving for naturally rewarding experiences is blunted [1,2,3,4]. This dysfunctional reward motivation provokes frequent substance taking, and thus can create a vicious circle of addictive behaviors. The vulnerability for relapse to substance use is linked to multiple changes in reward-relevant neurotransmitter systems, particularly within the nucleus accumbens (Nacc) that serves as an integrative hub for translating value representation into action initiation [5]. The current understanding of these alterations underlying addictive behaviors has been decisively shaped by insights from animal models. In particular, rodent studies of CA identified molecular mechanisms behind drug-seeking behavior. They indicate destabilizing consequences of chronic substance administration on glutamate (Glu) homeostasis, especially on the balance between synaptic and non-synaptic Glu in the Nacc [6, 7]. Usually, Glu signaling is dynamically regulated by astroglial cells which govern its release and uptake [8]. However, these glial regulatory activities are dysfunctional after chronic cocaine administration [8], leading to decreased basal Glu concentrations in the extrasynaptic space of the Nacc core during phases of abstinence [6]. In turn, relapse-provoking exposure to cocaine challenges or conditioned cues result in enhanced stimulation of Glu-ergic transmission [9]. Given that the glial regulation is disrupted, the concentration of synaptically released Glu rises in the Nacc core [6]. Furthermore, this increase in extrasynaptic Glu is associated with amplified substance-seeking [10].

Gaining a better understanding of the neurobiological adaptations underlying maladaptive substance-seeking in humans is necessary for the targeted development of pharmacotherapies for CA, which are lacking so far [11]. Despite the animal model-based understanding of substance-related adaptations within the Glu system in the Nacc, sparse evidence exists for the translational validity of these models into humans. Extensive efforts have been made to capture accumbal Glu homeostasis by applying proton magnetic resonance spectroscopy (1H-MRS) as a noninvasive technique to determine concentrations of target metabolites such as Glu in the brain. However, due to technical challenges of 1H-MRS application, specific evaluation of Glu homeostasis within the Nacc of humans with CA has not been achieved yet [12,13,14,15]. Conventional 1H-MRS measures yield rather low sensitivity, thus it has been challenging to assess the concentration of Glu in the human brain (around 6–12.5 mmol/L) that is ~4000–7000 times lower than the corresponding concentration of water [16]. To compensate for the low Glu concentration, larger 1H-MRS acquisition voxels are conventionally required to ensure sufficient signal-to-noise ratio (SNR). However, the size of the Nacc (1–2 cm3) requires a small 1H-MRS voxel to assess Glu specifically within this region [17]. Alternatively, to still enable region-specific measurements with sufficient SNR, many separate measurements can be assessed and averaged to compensate for the small-voxel size. However, signal averaging requires a steady measurement environment over time. Unfortunately, vessels and ventricles surrounding the Nacc generate pulsating motion and magnetic susceptibility changes, which distort MRS line shape and limit the achievable SNR. To overcome the addressed issues, we used the water peak with its high SNR as a reference for averaging the signals by applying non-water suppressed metabolite-cycling 1H-MRS [18,19,20] to increase the SNR, decrease the linewidth, and thereby enable good spectral quality.

A further difficulty for neurometablic comparison between clinical and healthy populations in general is that metabolic concentrations have previously been evaluated as a relative measure only, defined as the ratio of the metabolite of interest to either creatine or water. Since these reference molecules could potentially be altered in patients, absolute quantification of metabolites of interest is necessary to unambiguously identify pathophysiological shifts of metabolites of interest. Therefore, we applied the principle of reciprocity [21] to achieve an absolute quantification of metabolite concentrations instead of relative measures.

With this tailored 1H-MRS we first aimed at quantifying accumbal Glu concentrations in cocaine-addicted individuals CA (CAI) and in healthy control individuals (HCI) during abstinence to test the glutamate homeostasis hypothesis of addiction in humans. Further, to address the potential role of Glu-ergic adaptations during cocaine craving in humans, Glu levels were additionally assessed when the desire to use cocaine was provoked with cocaine cues. Moreover, we investigated the restoring effect of a N-acetylcysteine challenge (N-AC) [22], that has shown to successfully counterbalance disruptions in Glu homeostasis in animal models of addiction [23]. We expected that CAI would show decreased Glu concentrations during abstinence and increased concentrations during craving. For N-AC, we hypothesized that it has the potential to normalize Glu and craving levels.

Materials and methods

Participants

The study sample consisted of 26 CAI and 30 HCI who were recruited via online and in-house advertisement. Inclusion criteria for CAI were a minimum average cocaine use of 1 g per week within the last 6 months and a diagnosis of cocaine dependence according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) [24]. Exclusion criteria for CAI were other DSM-IV axis I psychiatric disorders including dependence of other psychoactive substances (excluding tobacco dependence and attention-deficit/hyperactivity disorder). Exclusion criteria of HCI were regular use of psychoactive substances with exception of current alcohol and irregular cannabis use. In addition, also DSM-IV axis I psychiatric disorders (excluding tobacco dependence) lead to exclusion of HCI. General exclusion criteria for all participants were contraindication for magnetic resonance imaging, allergy from N-AC, severe somatic diseases, neurological disorders, prior head injury, pregnancy, lack of contraception, family history of severe DSM-IV axis I psychiatric disorders, and concurrent participation in another clinical trial.

All participants gave written informed consent in accordance with the declaration of Helsinki before study participation and received financial compensation after study completion or termination. The study was approved by the Cantonal Ethics Committee of Zurich (KEK ZH No: 2014-0010).

1H-MRS data acquisition and analysis

1H-MRS measurements were performed on a Philips Achieva 3 T whole-body scanner equipped with a 32-channel receive-only phased-array head coil (Philips Healthcare, Best, The Netherlands). High-resolution anatomical images (1 × 1 × 1 mm3) were acquired using a standard T1-weighted 3D turbo field echo sequence. 1H-MRS spectra were obtained by means of non-water suppressed Point-RESolved Spectroscopy (PRESS) preceded by a metabolite-cycling pulse combined with inner-volume saturation (IVS) [19]. The PRESS localization sequence was set to repetition time of 2500 ms, echo time of 32 ms, and spectral bandwidth of 2000 Hz. The metabolite-cycling technique with application of a shift-sensitive asymmetric adiabatic inversion pulse allowed alternately inverting either up- or downfield metabolites without affecting the water signal [16, 18,19,20]. This enables simultaneous analyses of the water peak and metabolite signals after separating metabolite peaks from the water peak and sideband artefacts by either adding or subtracting consecutively acquired echoes. The unsuppressed water peak with its high SNR can be used as a reference for frequency alignment, phase, and eddy current correction of all individual signals before averaging to enable optimal spectral quality. After metabolite-cycling pulses, IVS was implemented to enable precise voxel placement, to reduce chemical shift displacement errors between metabolites, and to reduce flow artefacts by applying six saturation bands (Fig. 1a; blue) based on highly selective, broadband radiofrequency pulses with polynomial phase response. Based on the T1-weighted image, the 1H-MRS voxel (effective size after IVS: 9.4 × 18.8 × 8.4 mm3) was positioned and tilted to cover the left Nacc. In four blocks of 128 signals a total of 512 signals were achieved, which lasted 22.5 min.

Fig. 1: Proton magnetic resonance spectroscopy (1H-MRS) assessment of glutamate concentration in the nucleus accumbens.
figure 1

a Coronal, sagittal, and axial slice of T1-weighted images of the brain illustrating the tilted position of the 1H-MRS acquisition voxel centered on the left nucleus accumbens. The effective voxel size (white; 9.4 × 18.8 × 8.4 mm3) results from the overlap localized volumes of lactate/lipid (black) and water (green) defined by saturation bands (blue; for details see Supplementary Information). The shim box is indicated in orange. b Exemplary single spectrum measured therein (black), LCModel fit (red), fit-residuals (above; black), and glutamate and glutamine extracted (yellow).

Full size image

For quantification of spectral data, LCModel [25] with a set of basis spectra, consisting of 18 metabolites, was used (Fig. 1b). To account for potential cerebrospinal fluid fraction in the measured voxel, the T1-weighted images were segmented using SPM8 [26]. Absolute quantification of metabolites was performed using an approach exploiting the phantom replacement method. To ensure adequate correction of differences between the acquisition of the phantom and the in-vivo measurement including corrections for coil loading and B1 inhomogeneity, the principle of reciprocity was applied as described in more details by Zoelch et al. [21] and in the Supplementary Information. An in-house patch for the code of the Philips scanner, allowing for the use of the metabolite-cycling technique and the absolute quantification method, was used and is available upon request.

Craving paradigm

During the 1H-MRS measurements, all participants watched two customized films, one as a control condition and the other one to provoke the desire to use cocaine. The first film showed video clips of a chess game to assess basal Glu levels (neutral film), while the second film showed clips of cocaine preparation and consumption to measure craving-induced Glu (cocaine film). The film order was not randomized to avoid carry-over effects of cocaine craving. These films were specifically created for this study in cooperation with professional actors and filmmakers to enable (a) accurate comparison between equivalent experimental and neutral condition; in both films, an interaction between two male protagonists is featured and both clips incorporate the same objects except for the chessboard and the cocaine paraphernalia, respectively. A customized cocaine film further allows (b) to expose participants to locally characteristic paraphernalia, and thereby the stimulus overcomes cultural diversity in substance use that might otherwise compromise authenticity. These authentic visual stimuli were accompanied by auditory stimuli such as the clicking sound of the chess pieces or the grinding and sniffing of cocaine to allow an immersive experience. In both films, the video clips were interleaved with blank black screens of 2.5 min. Before film presentations, participants were asked in a written instruction to empathize with the scenes shown in the video clips and to recollect similar autobiographic experiences during black pauses.

Subjective craving was assessed on a visual analog scale ranging from 0 to 10 with intervals of 0.25 (0 = no craving; 10 = strong craving). Participants rated their craving on this scale with a computer mouse while lying in the scanner, directly before and after each film.

Pharmacological challenge

The pharmacological challenge comprised the cysteine prodrug N-AC that is approved as a mucolytic and to treat acetaminophen overdose. A number of animal studies administering N-AC, which binds to cysteine-glutamate antiporters, were able to regulate Glu synthesis and cycling, and thus successfully modulate alterations in the Glu homeostasis after chronic cocaine administration.

The challenge was given as a dose of 2400 mg/day on two consecutive days. The dosage was selected based on a study assessing the safety and tolerability of three doses, showing no significant difference between 1200, 2400, or 3600 mg/day in terms of side effects [27]. In addition, preliminary data suggests that the dosage of 2400 mg/day N-AC is potentially effective on craving, relapse, and on Glu levels in the cingulate gyrus in addiction [23].

Here, the main goal was to investigate the potential impact of N-AC on the Glu homeostasis in the Nacc. The treatment period of 2 days was therefore chosen to be short enough to maximize compliance and to minimize dropouts, yet long enough to see a potential effect according to previous reports [23]. The effect of N-AC was tested against a placebo, in the form of identical capsules of mannitol, to enable a randomized, double-blind, placebo-controlled, and crossover procedure. The first dose of either N-AC or placebo was handed out to our participants on their assessment visit. The sequence of N-AC and placebo sessions were randomized and counterbalanced to avoid order and sequence effects. Participants were instructed to take the first dose 1 day prior to the first measurement day, 1200 mg in the morning and in the evening. On the measurement day itself, the second dose of 2400 mg was administered on site 1 h prior to the 1H-MRS measurement, since peak plasma concentration of N-AC occurs ~1–2 h after ingestion [28]. After the first measurement day, participants were crossed over to receive the other compound in the same administration regimen for the second measurement. Both measurement sessions were performed the same way. To ensure complete washout of the compounds, the sessions were separated by an interval of 14 days (±4 days). For more details, see Supplementary Information and Supplementary Fig. S1.

Statistical analysis

All analyses were performed in a sample consisting of 26 CAI and 30 HCI, unless stated otherwise. For detailed description of data exclusion, see Supplementary Information. In order to allow the inclusion of all available data from every subject despite missing of single measurements, a multilevel approach was applied. Uneven data sets and irregular time intervals between measurement points were taken into account [28], since random coefficients allow modeling the time course as continuous variable based on the assumption that slopes and intercepts vary randomly across individuals [29]. The random coefficient model estimated the impact of CA (Table 1; CAI) and cocaine-cue stimulation (Table 1; cocaine film) as fixed effects on either the dependent variable Glu or craving to investigate their differences between groups and conditions (Figs. 2 and 3, respectively). The same model was estimated for Glx. We included time, age, sex, and duration of abstinence (analysis of urine samples and log-transformed hours since last cocaine use prior to measurement) as predicting variables in our multilevel models. None of these variables had an influence on Glu and craving levels before and after stimulation with the cocaine film, and were therefore omitted from the final models. All random coefficient models were fitted using SAS V9.4 with the function proc mixed following Brown and Prescott [29] or proc nlmixed when accounting for variance inhomogeneity. Variance was observed to be inhomogeneous for craving as dependent variable. As the results did not differ substantially, only the model assuming variance homogeneity between groups is presented. Examination of possible outliers, test for normal distribution, and all other statistical calculations were performed with SPSS 25.0. Since guidelines regarding effect sizes for multilevel models have not yet been established, we calculated effect sizes using Cohen’s d for group differences of main interest taking into account their respective sample sizes.

Results

Demographic characteristics and cocaine consumption

The groups were matched regarding age, sex, education, verbal intelligence, and smoking status. In CAI, mean cocaine consumption rate and dose within the last 6 months was 3.40 times per week (±1.89 s.d., range: 0.46–7) and 5.52 g per week (±10.03 s.d., range: 1–52.50). The main route of cocaine administration was intranasal (22 CAI), while 4 CAI were primarily inhaling cocaine. For full characterization of the final sample regarding demographics, matching, and substance use, see Supplementary Tables S1 and S2.

Craving induction

Overall, CAI showed substantially higher craving ratings than HCI (P < 0.0001, d = 1.67, 95% CI [1.06, 2.28], d calculated for pre neutral film HCI vs. CAI). Craving ratings increased strongly after presentation of the cocaine film only among CAI (P = 0.0002, d = 0.85, 95% CI [0.28, 1.42]), while HCI did not report any cocaine craving (Fig. 2).

Fig. 2: Craving as a function of cocaine-cue stimulation.
figure 2

Plot of mean craving ratings per group before and after the presentation of a neutral and a cocaine film represented as dots with standard error bars; blue = healthy control individuals, red = cocaine-addicted individuals. a Placebo condition: a random coefficient model showed significantly higher craving ratings in cocaine-addicted individuals (P < 0.0001, N = 26) at all measurement points compared with healthy control individuals (N = 30), and significantly increased craving ratings after the cocaine film in cocaine-addicted individuals compared with their ratings before the cocaine film (P = 0.0002). b N-acetylcysteine (N-AC) condition: no significant effects were observed for either group. For statistics, see Table 1.

Full size image

1H-MRS quality

According to general MRS quality criteria from the MR Spectroscopy Consensus Group [30], the upper limit for linewidth (full-width at half-maximum peak height) was defined as <12.7 Hz (0.1 ppm) [30]. The threshold for sufficient SNR was set >10. Based on this quality assessment, five measurements were excluded from analysis (for details see Supplementary Information). In the resulting data set, Cramér–Rao lower bounds (CRLB) for Glu were ≤11% for all measurements. CRLB serve as quality criterion to determine the reliability of the fit for each metabolite and are considered reliable when <20% [30].

For exemplary spectrum and overall MRS data quality see Fig. 1 and Supplementary Table S3. All parameters for spectral quality are further presented per condition in the Supplementary Tables S4–S7. Spectral quality between groups in the four different conditions did not differ (P > 0.2), thus enabled valid group comparisons between conditions (for statistics see Supplementary Tables S4–S7).

Glutamate concentration in the Nacc

We found strongly decreased basal Glu concentrations in CAI (P = 0.0047, d = −0.89, 95% CI [−1.47, −0.32]) compared with HCI during the neutral film (Table 1 and Fig. 3). In accordance with the described raise of craving level in CAI, we also observed elevated Glu concentrations in the Nacc of CAI during the cocaine film. An interaction effect showed an increase in Glu levels from the neutral to the cocaine film in CAI, while in HCI a reduction was observed (P = 0.013 dCAI = 0.44, 95% CI [−0.12, 1.00], dHCI = −0.50, 95% CI [−1.03, 0.02], Table 1 and Fig. 3). Also for Glx, a significant main effect of film was found, while trends were observed for the group and the interaction effect (for details see Supplementary Table S8). There was no significant correlation between Glu and craving levels in the cocaine film condition (P > 0.35).

Table 1 Random coefficients model for craving and for glutamate measures.
Full size table
Fig. 3: Impact of cocaine-cue stimulation on glutamate levels within the nucleus accumbens in cocaine addiction.
figure 3

Plot of mean glutamate concentrations in millimoles per litre (mmol/L) per group during presentation of a neutral and a cocaine film represented as dots with standard error bars; blue = healthy control individuals, red = cocaine-addicted individuals. a placebo condition: a random coefficient model showed significantly lower basal glutamate levels (neutral film, P = 0.0047) in cocaine-addicted individuals (N = 26) compared with healthy control individuals (N = 30), and significantly increased cue-induced glutamate levels (cocaine film) compared with baseline levels (neutral film, P = 0.013). b N-acetylcysteine (N-AC) condition: no significant effect was observed in cocaine-addicted individuals, while a significant main effect for N-AC was found for healthy control individuals. For statistics, see Table 1.

Full size image

Pharmacological modulation

Contrary to expectations, the pharmacological challenge showed neither an effect on craving nor on Glu levels in CAI. However, the cue-induced Glu increase in CAI while taking placebo (see Table 1 and Fig. 3) was no longer observed in the N-AC condition. Unexpectedly, the N-AC challenge showed a strong effect on Glu in HCI (P = 0.024, d = −0.82, 95% CI [−1.36, −0.28], d calculated for neutral film placebo vs. N-AC), contrary to the expected effect on CAI, leading to an overall reduction in Glu across both conditions (Supplementary Table S9). For an overview of adverse events during study duration, see Supplementary Table S10.

Discussion

Research focussing on the role of the dopamine system in addiction has significantly advanced our understanding of the disease, whereas it has not resulted in effective treatments yet (26). Here, we evaluated alterations in the Glu-ergic homeostasis in the Nacc in CAI, since animal models suggested Glu imbalance in the Nacc to underlie addiction-like behavior [6]. Therefore, we examined Glu concentrations in the Nacc of CAI by means of a dedicated 1H-MRS protocol. In accordance with animal models of addiction [6], our data provide the first evidence in humans that CA is characterized by profound Glu-ergic shifts within the Nacc. We show that Glu concentrations in the Nacc were reduced in CAI in the withdrawal state relative to HCI, whereas Glu levels in CAI were amplified by visual and auditory cocaine-related cues reliably inducing craving symptoms. These results are consistent with alterations in Glu transmission described in substance-seeking reinstatement models in rodents, and indicate that accumbal Glu signaling is generally inhibited after chronic cocaine use and is accelerated during cue-exposure. This Glu imbalance is in line with, and might be explained by, molecular mechanisms that have been observed in animals exposed to cocaine. After withdrawal from cocaine, the expression of Glu transporters 1 and regulation via glial glutamate-cystine transporters are consistently down-regulated [9, 31,32,33]. As extracellular Glu normally provides release-regulating tone on presynaptic metabotropic Glu receptors 2 and 3 (mGluR2/3) [34], the signaling of mGluR2/3 is diminished following chronic cocaine administration [9, 35]. Together, these adaptations in the Glu system lead to decreased basal Glu concentration in the extrasynaptic space of the Nacc core after chronic cocaine administration. In turn, during substance-seeking reinstatement, dysregulated Glu release combined with impaired Glu removal promotes an overflow of synaptically released Glu in the Nacc core [9]. Thus, our findings might reflect a common Glu-ergic dysfunction in the Nacc underlying the mammalian craving brain.

However, 1H-MRS does not allow differentiating intra- and extrasynaptic Glu, and therefore final inferences from Glu concentrations regarding directional Glu distribution within the different cellular compartments in humans are not possible yet. Spatial inferences are further restricted since voxel size and shape, which define the region wherein Glu is measured by 1H-MRS, cover the Nacc including both core and shell regions, whereas preclinical findings described Glu changes within the core only. On the contrary, this limited spatial specificity rather amplifies the probability for false negative results, not for false positives.

Given that the Nacc has been suggested to be also involved in salience detection [36] as well as in sleep–wake homeostasis [37], the present Glu decrease observed in HCI while watching the cocaine film might reflect a decrease of salience of the audio-visual inputs over time together with a fatigue-driven Glu reduction, since the cocaine film was always shown after the neutral film and might constitute a stimulus with low salience for HCI. In CAI, on the other hand, the expected higher salience of cocaine-related cues [38] together with heightened arousal might contribute to increased Glu levels during the cocaine film, analog to animals showing accelerated Glu release in the Nacc in response to cues linked with incentive salience [39].

Glu imbalance in CAI was not significantly modulated by the short-term intake of N-AC (Table 1). However, in the N-AC condition, the cue-induced Glu increase in CAI while taking placebo was no longer observed, whereas N-AC had no impact on craving ratings. Given that N-AC caused a significant reduction of Glu in the control group, which is in line with recent 1H-MRS Glu measures of the basal ganglia in a healthy population [40], N-AC seems, in principle, capable of affecting the Glu system of the human brain. Further, recent findings have shown beneficial effects of long-term N-AC interventions of 25 days (as compared with our two-day administration) on problems related to CA [41]. Thus, an adjustment of the duration and potentially of the dose of N-AC intervention might be needed for reducing craving in CAI. Lastly, since the mechanisms of N-AC are far from being fully understood [42], it remains challenging to interpret our unanticipated findings regarding the N-AC administration. Thus, the impact of numerous other pharmacological compounds targeting the Glu system [43] should be further studied by means of a small-voxel 1H-MRS approach as proposed here. We hypothesize that Glu levels within the human Nacc could serve as a noninvasive biomarker to predict individual treatment responses in future clinical trials. Namely, such a biomarker would allow identifying individuals who benefit the most from pharmacological interventions aiming at restoring disturbances within the Glu homeostasis in CAI, and thus would enable a personalized pharmacological treatment.

In sum, our findings contribute to the establishment of a translational framework of neurometabolic changes within the Glu system underlying CA across species that will enhance our understanding of addictive disorders, and, most importantly can thereby foster the development of urgently needed novel pharmacotherapies that target the Glu system.