**A New Measure for Assessing the Intensity of Addiction Memory in Illicit Drug Users: The Addiction Memory Intensity Scale**

#### **Jia-yan Chen 1, Jie-pin Cao 2,3, Yun-cui Wang 1,4, Shuai-qi Li 1,2 and Zeng-zhen Wang 1,2,\***


Received: 18 October 2018; Accepted: 21 November 2018; Published: 22 November 2018

**Abstract:** Disrupting the process of memory reconsolidation could be a promising treatment for addiction. However, its application may be constrained by the intensity of addiction memory. This study aimed to develop and initially validate a new measure, the Addiction Memory Intensity Scale (AMIS), for assessing the intensity of addiction memory in illicit drug users. Two studies were conducted in China for item analysis (*n* = 345) and initial validation (*n* = 1550) of the AMIS. The nine-item AMIS was found to have two factors (labelled Visual Clarity and Other Sensory Intensity), which accounted for 64.11% of the total variance. The two-factor structure provided a reasonable fit for sample data and was invariant across groups of different genders and different primary drugs of use. Significant correlations were found between scores on the AMIS and the measures of craving. The AMIS and its factors showed good internal consistency (Cronbach's *α*: 0.72–0.89) and test-retest reliability (*r*: 0.72–0.80). These results suggest that the AMIS, which demonstrates an advantage as it is brief and easy to administer, is a reliable and valid tool for measuring the intensity of addiction memory in illicit drug users, and has the potential to be useful in future clinical research.

**Keywords:** addiction; memory; assessment; substance use disorder

#### **1. Introduction**

Addiction memory is conceptualized as a pathological memory related to addictive behaviors [1,2]. Even after long-term abstinence, addiction memory can be reactivated upon re-exposure to substance-related cues and associated with a craving that results in relapse [2,3]. Recently, several studies have found that disrupting the reconsolidation of drug-related memory can reduce craving, attention bias, or drug-using behavior [4–7], indicating that addiction memory can be manipulated during reconsolidation. However, several "boundary conditions" may constrain the application of memory reconsolidation in addiction treatment [8,9]. One of the constraints is the intensity of memory. It has been shown that memories with a higher level of intensity may be more resistant to disruption and require a longer re-exposure (or a shorter time interval between memory acquisition and re-exposure) to induce memory lability [10–12]. Given that, a valid measurement for assessing the intensity of addiction memory is needed before examining the impact of memory intensity on reconsolidation propensity.

Previously, a single-item visual analogue scale (VAS) was used in some studies to assess the vividness of addiction memory [13,14]. Although the VAS has been widely used to evaluate the subjective experience, such as craving [15], this single-item measure may fail to reflect the complex nature of the intensity of addiction memory. Regarding instruments measuring the intensity of addiction memory, we believe that the use of a multi-item scale or questionnaire would be beneficial for research in that it could help to clarify the underlying dimensions [16]. Furthermore, although there have been several scales and questionnaires developed to assess the severity of addiction [17–21], we should note that these instruments were not originally designed to measure the intensity of addiction memory. It is unknown which items or dimensions of the tools may reflect the memory strength, which raises concern about the utility of these indexes in assessing the intensity of addiction memory. Given these concerns, a new measure is needed to evaluate the intensity of addiction memory explicitly.

Studies on the phenomenology of memory, which mainly focus on one's subjective experience retrieved from the memory [22,23], open an opportunity to assess the intensity of addiction memory. Although little is known about the phenomenological characteristics of addiction memory, previous studies have suggested a relationship between addiction memory and autobiographical memory, especially episodic memory [2,3,24]. Similar to the autobiographical memory, addiction memory is related to personal experiences derived from an individual's history of drug use. Thus, research on autobiographical memory can be used as a frame of reference. There have been several studies presenting a diverse variety of phenomenological characteristics of autobiographical memory, such as vividness and sensory details [22,25–30]. Although the classification of phenomenological traits is still under discussion, the characteristics involving the sensory-perceptual information could probably be used as a central measure of memory intensity. The sensory-perceptual details are the primary information retrieved from one's autobiographical (episodic) memories [31]. In the self-memory system, the sensory-perceptual information of autobiographical memory can be directly retrieved when there are event-related cues [31], meaning that the sensory details would be re-experienced when recalling the memory. In that, people may feel a high level of sensory intensity if they retrieve detailed sensory information. Additionally, the visual imagery is presented predominately and correlated with other sensory details (e.g., olfactory, gustatory) in episodic information [32,33]. In sum, a strong memory tends to be visually vivid, full of sensory details. Since the addiction memory is characterized by powerful imprints of the information about psychoactive substances [2,3,24], we infer that the strength of sensory-perceptual information could be used for measuring the intensity of addiction memory.

Although the studies on autobiographical memory provide a research frame for understanding the properties of addiction memory, there is one crucial limitation when applying the measurements of autobiographical memory in assessing the intensity of addiction memory. It should be emphasized that the measures were developed to evaluate the autobiographical memories of individuals about their life events but not specifically their addiction memories. To the best of our knowledge, there is no scale or questionnaire specifically designed for assessing the phenomenological characteristics in the strength of addiction memory. Therefore, our study aimed to fill this gap first through the development and initial validation of a scale specifically for evaluating the intensity of addiction memory in illicit drug users. In our research, a Likert-type scale, which we labelled the Addiction Memory Intensity Scale (AMIS), was initially developed. We conducted two studies: one for item analysis and one for initial validation of the AMIS. We hypothesized that the scale would be a reliable and valid tool for assessing the intensity of addiction memory in illicit drug users.

#### **2. Materials and Methods**

#### *2.1. Initial Scale Development*

The development of the AMIS began with a review of the literature on the phenomenology of autobiographical memory. Instruments that measured the phenomenological characteristics of autobiographical memory [22,25–30], especially the items involving sensory-perceptual information, were collected as the reference. Regarding the domains of memory intensity, we agree with the previous studies that the strength of sensory-perceptual details can be either considered as one characteristic [27] or classified into visual and non-visual information [22,29]. Therefore, it is prudent to generate different items for visual and non-visual information of addiction memory intensity and to test the factor structure of the AMIS (i.e., whether these two components represent different constructs).

For generating items unique to measuring the intensity of addiction memory, in-depth interviews were conducted with twelve illicit drug users with substance use disorders. The respondents were asked to recall and describe their experiences of using drugs, and then to describe their subjective experience during retrieval of the memories of using drugs. The interview transcripts were reviewed, coded, and classified into the themes of visual and non-visual information via a thematic framework analysis. Forty-four items were then initially generated based on the abovementioned instruments and interviews.

Five psychologists with professional experience in mental health and addiction treatment were invited to a consultation meeting to review the preliminary instrument. Twenty illicit drug users in the drug rehabilitation centers were then interviewed about the understandability and acceptability of the scale and each of the items. Based on the feedback from the psychologists and illicit drug users, items that were ambiguous, repetitive, not understood, or not acceptable were either revised or removed, yielding a 20-item draft of the AMIS. The AMIS draft was piloted in a sample of 345 illicit drug users to evaluate the appropriateness of the items further (see Study-1). The results of Study-1 were used for item selection. After the item selection, a final version of the AMIS was developed, consisting of nine items.

In the present study, the AMIS was specifically developed to measure the intensity of addiction memory in illicit drug users. Given the fact that most of the illicit drug users in China have a low education level, we agree with the previous studies [34–36] that suggest that a fully labelled five-point Likert scale may improve the respondents' ability to discriminate among categories and reduce response bias, without lowering the reliability of the instrument. Therefore, regarding the response options, five-point response categories were labelled (1 = strongly disagree, 2 = disagree, 3 = unsure, 4 = agree, 5 = strongly agree) and used on all the items. The total AMIS score can be obtained by computing the mean of the items; thus, the range of possible scores is from 1 to 5.

#### *2.2. Participants*

Two studies were conducted from January 2015 to March 2018 at drug rehabilitation centers in China. Participants were illicit drug users (substance use disorders as diagnosed by the DSM-IV) with the age of at least 18 years old. Illicit drug users who had difficulty in answering the survey as a result of illiteracy, withdrawal symptoms, cognitive disorders, or other psychiatric disorders were excluded from the studies. The study protocols were approved by the Institutional Review Board of the School of Public Health, Tongji Medical College, Huazhong University of Science and Technology (approval file number: [2015]#15). Each participant in the studies provided signed informed consent.

#### 2.2.1. Participants in Study-1

For item analysis, Study-1 was conducted at two drug rehabilitation centers in the city of Wuhan, China. Based on the lists provided by the rehabilitation centers, the illicit drug users were randomly selected by the principal researcher (Z.-Z.W.) using a list of computer-generated random numbers. Ultimately, a total of 345 participants were selected.

#### 2.2.2. Participants in Study-2

For initial validation of the AMIS, Study-2 was conducted at six drug rehabilitation centers in two cities (Wuhan and Zhongshan) in China. To avoid repetition, we excluded the individuals who participated in Study-1. Finally, a total of 1550 participants were recruited at the rehabilitation centers.

#### *2.3. Measurements for Testing Concurrent Validity*

In Study-2, the Obsessive Compulsive Drug Use Scale (OCDUS) [37] and a craving visual analogue scale (VAS) were used for concurrent validation since previous studies have indicated that addiction memory is related to craving [38–41]. Given that individuals who experienced a longer term of drug use may undergo more repeated exposures and thus strengthen their intensity of memory [42], the participants' duration of illicit drug use was also used for concurrent validation.

#### 2.3.1. Obsessive Compulsive Drug Use Scale (OCDUS)

The OCDUS was used to measure general craving in the past week. The Chinese version of OCDUS is a 12-item questionnaire and uses a three-factor structure: "interference of drugs," "frequency of craving," and "control of drugs" [37].

#### 2.3.2. Visual Analogue Scale (VAS)

The VAS was used to measure instant craving. The VAS is a 10-cm line with "not at all" on the left and "extremely" on the right. Participants were asked to rate their craving for drugs at present on the VAS.

#### 2.3.3. Duration of Illicit Drug Use

The participants were asked to answer a single question adapted from the Chinese version of Addiction Severity Index-V [17] to report their duration of illicit drug use ("How many years in your life have you regularly used the illicit drugs (e.g., heroin, amphetamines, ketamine, cannabis, cocaine, or more than one substance)?").

#### *2.4. Procedure*

The 20-item draft of the AMIS was administered in Study-1. Before answering the scale, participants were instructed to recall memories about their experiences of drug use ("Please recall your experiences of using drugs. Please try your best to recall the experiences in detail, for example, when and where it happened, whom you were with, and what you felt then."). In Study-2, the nine-item AMIS was applied. Participants were given the instruction mentioned above and then completed the scale. Subsequently, they were asked to answer the OCDUS and the VAS. To avoid leaving the participants in a vulnerable state, they were given relaxation techniques after finishing the surveys.

#### *2.5. Data Analysis*

The samples available for data analysis numbered 343 in Study-1 and 1420 in Study-2. In Study-1, two participants were unwilling to answer the survey and did not provide any information. Thus, they were excluded from the study. In Study-2, one-hundred-and-thirty questionnaires with missing items were excluded from data analysis. The characteristics of participants who were excluded were not significantly different from those included in the data analysis (Supplementary materials, Table S1). All statistical analyses were performed in SAS 9.4 (SAS Institute Inc., Cary, NC, USA) [43], and the significance level was set at *α* = 0.05 (two-tailed probability).

#### 2.5.1. Study-1: Item Analysis

The critical ratio, item-total correlation, factor loading, and coefficient of stability were calculated for each item. Regarding the coefficient of stability, thirty of the participants were randomly selected using a list of computer-generated random numbers. They were retested two weeks after the first test. The results of the item analysis were used comprehensively for item selection. Items that meet two or more deletion criteria were removed [44]. The deletion criteria were as follows [45]: the critical ratio <3.00, the item-total correlation coefficient <0.40, factor loading <0.45, and coefficient of stability <0.50.

#### 2.5.2. Study-2: Initial Validation of the AMIS

For cross-validation, both exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) were used to validate the factor structure of the AMIS. The sample data were randomly divided into two parts before the factor analyses were conducted. One subset was for EFA (*n* = 710), and the other was for CFA (*n* = 710).

The principal component analysis combined with oblique (direct oblimin) rotation was performed in EFA. The eigenvalue and the scree plot were used to assist in retaining the number of components.

The maximum likelihood estimation was conducted in CFA. The model chi-square statistic, comparative fit index (CFI), standardized root mean square residual (SRMR), root mean square error of approximation (RMSEA), and non-normed fit index (NNFI) were reported. The chi-square statistic was not used as a viable fit index because of its excessive sensitivity to sample size [46,47]. Thereby the acceptable fit-index criteria used in CFA were as follows [46,47]: CFI > 0.90, SRMR < 0.08, RMSEA < 0.10, and NNFI > 0.90. Additionally, the expected cross-validation index (ECVI) was tested to assist model comparison. The model with a lower ECVI value was considered to be better-fitting [46,47].

The impact of gender or primary illicit drug of use on the measurement invariance was tested in a series of multi-group CFAs. The measurement invariance could be justified if [48,49]: the overall fit of the model was acceptable, and the differences in CFI and NNFI between the constrained and unconstrained model were <0.01 and <0.05 respectively. Of the participants recruited, only 1.1% were ketamine users (see Table 1). Due to the insufficient sample size of this subgroup, the ketamine users were excluded from the multi-group CFA models when comparing different primary drugs of use.



Note: ATS = amphetamine-type stimulants; SD = standard deviation.

The inter-factor correlation and the concurrent validity of the AMIS were estimated using the Pearson correlation coefficient, using a full sample (*n* = 1420). When assessing the correlations between the AMIS and the VAS, the VAS scores were found to follow a non-normal distribution; thus, Spearman correlation coefficients were calculated instead.

To evaluate whether the AMIS could discriminate between the high and moderate-to-low levels of craving, we performed a series of independent-samples *t*-tests. According to the previous study [50], the participants who scored >5 on the VAS were considered to have a high level of instant desire for drugs. Of the OCDUS, the following cut-off values indicate the high level of related contents: "interference of drugs" >15, "frequency of craving" >15, and "control of drugs" >6 [37].

Cronbach's *α* coefficient was computed to evaluate the internal consistency of the AMIS in a full sample (*n* = 1420). Test-retest reliability of the AMIS was obtained from a retest conducted in a subset of sixty participants. The participants were retested two weeks after their first test. The test-retest reliability was assessed using the Pearson correlation coefficient.

#### **3. Results**

#### *3.1. Characteristics of the Participants*

The characteristics of the participants in our study are shown in Table 1.

#### *3.2. Item Analysis*

The results of the item analysis are shown in Table 2. Three items showed an item-total correlation coefficient of <0.40, eleven items showed a factor loading of <0.40, and ten items showed a coefficient of stability of <0.50. Based on the abovementioned criteria for item selection, nine items were ultimately retained in the AMIS.



Itemsthatmeetthedeletioncriteria;AMIS AddictionMemoryIntensityScale;SDstandard

#### *3.3. Exploratory Factor Analysis (EFA)*

Table 3 displays the factor structure of the AMIS. The Kaiser–Meyer–Olkin measure of sampling adequacy (0.90) and Bartlett's test of sphericity (*χ*<sup>2</sup> = 2910.00, *p* < 0.001) indicated a suitable correlation matrix for factor analysis. The pattern matrix showed a two-factor structure with each eigenvalue >1 (4.73 and 1.04). The scree plot also suggested this factor solution (Figure 1).

**Table 3.** Exploratory factor analysis of the Addiction Memory Intensity Scale (AMIS; *n* = 710) 1.


Note: <sup>1</sup> The pattern matrix was presented; factors were extracted by principal component analysis and were rotated by oblique (direct oblimin) rotation. <sup>2</sup> Numbering reflects the order of items on the original scale. The AMIS was originally developed in Chinese, and the English version presented here was translated only for the publication of our study. AMIS = Addiction Memory Intensity Scale.

**Figure 1.** Scree plot based on the principal component analysis (*n* = 710).

The two-factor solution of the AMIS accounted for 64.11% of the total variance. Factor-1, labelled Visual Clarity, had shown greatest loadings on six items and measured the intensity of one's visual information when retrieving the memories of drug use. Factor-2, labelled Other Sensory Intensity, had shown greatest loadings on three items and evaluated the intensity of one's non-visual sensations and feelings when recalling the experiences of using drugs.

Item-15 was suspected to be a cross-loading item (the difference in factor loading was 0.22, slightly over 0.20). Thus, we tried to delete this item and see how the factor solution and internal consistency were affected. After the factor analysis and sensitivity analysis (Supplementary materials, Table S2), we found that removing this item hardly changed the two-factor structure, but would reduce the internal consistency of the scale. Therefore, item-15 was retained in the AMIS.

#### *3.4. Confirmatory Factor Analysis (CFA)*

The two-factor structure of the AMIS from EFA provided an acceptable fit for the other subset of sample data (see Table 4, Model 1). An alternative one-factor model was also constructed to assess the intensity of sensory-perceptual information. All items in this model served as indicators of one single factor. The one-factor model showed an acceptable-to-poor fit to the sample data, with the RMSEA value failed to meet its fit-index criterion (see Table 4, Model 2). Due to the lower ECVI value, the two-factor solution was favored.

**Table 4.** Confirmatory factor analysis of the Addiction Memory Intensity Scale (AMIS; *n* = 710).


Note: \* *p* < 0.01; AMIS = Addiction Memory Intensity Scale; CFI = comparative fit index; SRMR = standardized root mean square residual; RMSEA = root mean square error of approximation; NNFI = non-normed fit index; ECVI = expected cross-validation index.

All of the model-fit indexes of the two-factor multi-group models met the acceptance criteria (see Table 4, Model 3 and Model 4). When constraining all parameters to be equal across groups, the differences between the constrained and unconstrained models showed ΔCFIs of <0.01 and ΔNNFIs of <0.05 (across gender: ΔCFI = −0.001, ΔNNFI = 0.007; across primary drug of use: ΔCFI = 0.000, ΔNNFI = 0.006). Also, no significant difference in model fits was observed between the constrained and unconstrained models (across gender: Δ*χ*<sup>2</sup> (7) = 10.71, *p* = 0.152; across primary drug of use: Δ*χ*<sup>2</sup> (7) = 5.57, *p* = 0.591). These results suggest measurement invariance across the groups.

#### *3.5. Inter-Factor Correlation and Concurrent Validity*

There was a moderate, significant correlation between the two factors of the AMIS (*r* = 0.65, *p* < 0.001). The correlations between the AMIS and other measures are shown in Table 5. Significant correlations were found between OCDUS scores, VAS scores, duration of illicit drug use, and AMIS scores.


**Table 5.** Correlations between the Addiction Memory Intensity Scale and other measures (*n* = 1420).

Note: \* *p* < 0.01. OCDUS = Obsessive Compulsive Drug Use Scale; VAS = Visual Analogue Scale.

#### *3.6. Discriminant Validity*

The *t*-test results showed that, compared to the participants with moderate-to-low levels of craving, those with high levels of craving reported significantly higher scores on the AMIS (Table 6), indicating that the AMIS can discriminate between the high and moderate-to-low levels of craving.


**Table 6.** Discriminant validity of the Addiction Memory Intensity Scale (*n* = 1420).

Note: OCDUS = Obsessive Compulsive Drug Use Scale; VAS = Visual Analogue Scale.

#### *3.7. Reliability*

Respectively, Cronbach's *α* coefficients for the total, Visual Clarity, and Other Sensory Intensity scores on the AMIS were 0.89, 0.88, and 0.72, suggesting a good internal consistency in the scale.

Results from the retest showed that, between the test and the retest, the Pearson correlation coefficients for the AMIS and its two factors were good (Total: 0.80 (*p* < 0.01); Visual Clarity: 0.75 (*p* < 0.01); Other Sensory Intensity: 0.72 (*p* < 0.01)).

#### **4. Discussion**

This study develops and initially validates a new measure to assess the intensity of addiction memory in illicit drug users. The final version of the AMIS consists of nine items and shows a robust two-factor structure. Results of the multi-group CFA suggest measurement invariance of the two-factor structure across groups of different genders and different primary drugs of use. The moderate inter-factor correlation further confirms that the internal structure of the AMIS is reasonable. The validity of the AMIS is supported by other findings that there are significant correlations between the AMIS and other measures of craving and that the AMIS can discriminate between different levels of craving. Additionally, the AMIS and its two factors show good internal consistency and test-retest reliability. These results support the AMIS as a valid and reliable tool for measuring the intensity of addiction memory in illicit drug users.

The AMIS was initially developed to assess the intensity of sensory-perceptual information in addiction memory of the illicit drug users. Two phenomenological characteristics (Visual Clarity and Other Sensory Intensity) were identified in EFA. It is notable that the contents of these two factors are consistent with the theoretical dimensions constructed from our assumption. Thus, it is easy to interpret the factors clinically, as follows: Visual Clarity refers to the intensity of illicit drug users' visual information when retrieving the memories of drug use, while Other Sensory Intensity refers to the strength of their non-visual sensations and feelings. Moreover, Visual Clarity explained 52.51% of the total variance, which was higher than Other Sensory Intensity (11.60%). That is, the clarity of visual information accounts for a percentage of variance in the intensity of addiction memory that is greater than other sensory details. Similar to other studies that demonstrate that visual information predominates in the episodic memories [32,33], the results of our study support the significance of visual clarity in the intensity of addiction memory.

Although researchers have differentiated visual clarity from other sensory intensity [22,29], the two components can also be considered as one characteristic which reveals the strength of all sensory-perceptual details [27]. Thus, we compared the two-factor model to an alternative one-factor model. The two-factor solution showed an acceptable fit to the sample data and, due to its lower ECVI value, was considered to fit the data better. Furthermore, it is indicated in the cross-validation that the two-factor structure would be more stable. Therefore, the two-factor solution was preferred. Based on the results of factor analyses and inter-factor correlation analysis, the Visual Clarity and Other Sensory Intensity factors can be considered two related but distinct characteristics of addiction memory intensity.

Regarding the concurrent validity, scores of the AMIS and its factors showed significant correlations with the measures of craving, which is consistent with previous studies, suggesting that addiction memory is related to craving [38–41]. Although the results provide support for the validation of the AMIS, it should also be noted that there were significant but only modest correlations between the AMIS and the measures of craving. This may be due to the fact that addiction memory is long-lasting and resistant to being extinguished, whereas craving is considered to be a transient, fluctuant state that is sensitive to environmental influences [51–53]. In other words, individuals who retrieve vivid details from their memories of using drugs may, if there are other influences, report a low desire for drug use. In our study, the participants did rate extremely low scores on both the measures of craving (Supplementary materials, Table S3), especially the VAS (over one-third of the participants scored 0). Similar situations can be found in other studies. For example, Anton et al. [54] found that the high ratings on craving diminished when the patients were either hospitalized or successfully participated in treatment. Since the participants in our study were undergoing rehabilitation, it might be the engagement in treatment (which may result in a high self-efficacy or mastery of coping skills) that assisted them to prevent the escalation of desire for drug use. That is, it seems the interaction between craving and environmental influences that interfere with the association between the intensity of addiction memory and the level of craving. Significant correlations were also found between the AMIS scores and the duration of illicit drug use. However, the associations were quite modest, which may suggest that the participants' duration of illicit drug use, as measured in our study, is not particularly related to the intensity of addiction memory. It is possible that there may be other factors, such as the strength of the unconditioned stimulus [9], that moderate the associations between the intensity of addiction memory and the duration of illicit drug use. For example, a higher level of frequency or dose of drug use may elicit a stronger memory of addiction, which means the intensity of addiction memory may vary in degrees of frequency or dose of drug use but not merely depend on the duration of drug use. Future studies are needed to collect the participants' full history of drug use and to explore how the drug use history may affect the intensity of addiction memory.

Our study has several limitations that should be considered. First, the participants in our study were hospitalized in the drug rehabilitation centers and could not provide any information on current illicit drug use. Therefore, only craving was measured to evaluate the concurrent validity of AMIS. Second, although we have taken considerable effort to administer the AMIS to a diverse sample, the final version of the scale is a result of the sample used in the current study. Like other scales, the AMIS focused on a specific population (illicit drug users). Given the commonalities across addictions, the items of the AMIS may be adapted for patients with tobacco dependence, alcohol dependence, and so on. However, based on the current study, it is prudent not to draw an over-interpreted conclusion. Whether the AMIS can be decontextualized to produce such a broader application, is unknown. Moreover, although the measurement invariance indicated an absence of difference across primary illicit drug of use, the samples of illicit drug users in this study were primarily using heroin and amphetamine-type stimulants but no other illicit drugs such as cocaine and cannabis; the sample data was not collected from adolescents or non-Chinese speakers either. Thus, it is unknown whether the results of this study can be generalized to other populations of illicit drug users. Finally, although this study initially validates the utility of the AMIS, the scale should

further be validated in reconsolidation-based interventions to see whether it can predict responses to the interventions. Also, since the participants were not followed-up, other studies are required to determine the predictive validity of the AMIS.

Regardless of the limitations, this study presents a promising measure to assess the intensity of addiction memory in illicit drug users. The AMIS allows the clinicians and researchers to gather information regarding the strength of addiction memory from a multidimensional perspective. Measurement invariance of the factor structure indicates that the AMIS can be reliably utilized to evaluate the intensity of addiction memory in, at least, heroin and amphetamine-type stimulant users, and thus enable comparison of the memory intensity across illicit drug users. Furthermore, the AMIS may tailor interventions based on responses to the scale, especially the interventions based on memory reconsolidation, although future studies are needed to validate the responsiveness of the AMIS. Finally, the brief nature of the AMIS makes it easy to administer, so that the clinicians and researchers can use the instrument to measure the intensity of addiction memory efficiently.

#### **5. Conclusions**

In conclusion, the present study is the first to develop a scale for measuring the intensity of addiction memory in illicit drug users. The current evidence suggests that the AMIS is a reliable and valid tool with the advantages of being brief and easily applied, and has the potential to be useful in future clinical research.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2077-0383/7/12/467/s1. Table S1: Comparison of the participants' characteristics in Study-2; Table S2: Factor loading and internal consistency of the Addiction Memory Intensity Scale if item-15 is deleted; Table S3: Descriptive data for measures in Study-2.

**Author Contributions:** Conceptualization, J.-y.C., J.-p.C., Y.-c.W., S.-q.L., and Z.-z.W.; Data curation, Z.-z.W.; Formal analysis, J.-y.C. and J.-p.C.; Funding acquisition, J.-y.C. and Z.-z.W.; Investigation, J.-y.C., Y.-c.W., and S.-q.L.; Supervision, Z.-z.W.; Writing–original draft, J.-y.C.; Writing–review and editing, J.-y.C., J.-p.C., Y.-c.W., S.-q.L., and Z.-z.W.

**Funding:** This research was funded by the National Natural Science Foundation of China (grant number 81573236) and the Joint School Research Award Program for Pedagogy, Social Science, and Medical Science (grant number YX2015002). Neither of the funding organizations had a role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

**Acknowledgments:** Thanks to all staff members participated in our study for their efforts in the data collection.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Neurocircuitry of Reward and Addiction: Potential Impact of Dopamine–Glutamate Co-release as Future Target in Substance Use Disorder**

#### **Zisis Bimpisidis and Åsa Wallén-Mackenzie \***

Department of Organismal Biology, Uppsala University, S-752 36 Uppsala, Sweden **\*** Correspondence: asa.mackenzie@ebc.uu.se

Received: 9 October 2019; Accepted: 1 November 2019; Published: 6 November 2019

**Abstract:** Dopamine–glutamate co-release is a unique property of midbrain neurons primarily located in the ventral tegmental area (VTA). Dopamine neurons of the VTA are important for behavioral regulation in response to rewarding substances, including natural rewards and addictive drugs. The impact of glutamate co-release on behaviors regulated by VTA dopamine neurons has been challenging to probe due to lack of selective methodology. However, several studies implementing conditional knockout and optogenetics technologies in transgenic mice have during the past decade pointed towards a role for glutamate co-release in multiple physiological and behavioral processes of importance to substance use and abuse. In this review, we discuss these studies to highlight findings that may be critical when considering mechanisms of importance for prevention and treatment of substance abuse.

**Keywords:** addiction; reward; transgenic mice; optogenetics; self-administration; cocaine; amphetamine

#### **1. Dopamine and Substance Use Disorder**

Substance use disorder is a chronic, relapsing neuropsychiatric disease that occurs in a minority of recreational drug users [1]. In vulnerable individuals, the initial elevation of dopamine (DA) upon drug-consumption, which is believed to reflect the reinforcing effects of substances of abuse, will lead to characteristic behavioral patterns related to long-lasting alterations in the glutamatergic system [2–5]. Drug-seeking behavior and objectively reported craving in drug-addicts are both dependent on environmental cues associated with drug intake and can lead to relapse even after many years of abstinence [5,6].

All addictive substances influence the brain DA system which is implicated in reward processing, motivation and behavioral reinforcement [7,8]. The mesolimbic DA system, containing DA neurons located within the ventral tegmental area (VTA) and their projections to the nucleus accumbens (NAc) [9], has been shown to be particularly important for reward processing. DA release in the NAc, in particular in the medial (m) aspect of the shell (Sh) compartment, NAc mSh, has been implicated in the reinforcing properties of both natural and drug rewards [10]. DA modulates the activity of NAc neurons and their response to inputs coming from other limbic areas [1,11,12]. Abusive drugs increase DA levels preferentially in the NAc mSh as compared to the NAc core or the dorsal aspect of the striatum [1,13,14]. While these increases are related to the reinforcing properties of the drugs, the behavioral patterns observed in addicts after chronic drug use are related to more persistent neuroadaptations in the glutamatergic system. It has been shown that the psychostimulant cocaine rapidly induces synaptic plasticity in VTA DA neurons after acute, passive administration [15] and that persistent neuroadaptations can be observed in rats that have self-administered cocaine [16]. Unlike the VTA, synaptic plasticity in medium spiny neurons (MSNs) of the NAc is observed only after long-term cocaine use. Further, any changes in plasticity of the MSNs are contingent to a withdrawal

period [17] and occur on MSNs expressing the DA receptor subtype 1 (D1R) [18]. Overall, synaptic changes upon drug administration occur in a time-dependent manner, so that the first changes take place in the VTA, and, after repeated administration, in the NAc, a phenomenon believed to reflect the long-lasting behavioral consequences of chronic drug intake [5]. Comprehensive review articles of putative mechanisms, particular characteristics and regional differences in the context of synaptic plasticity observed in animal models of drug addiction have been published elsewhere and further details are beyond the scope of this review.

#### **2. DA-Glutamate Co-Release in the Mesolimbic System**

#### *2.1. DA Neurons in the VTA*

Neurons that produce DA and release it into the extracellular space, either in the synapse or along the axon, are commonly referred to as dopaminergic neurons, or simply DA neurons. Molecularly, DA neurons are defined by the presence of the rate-limiting enzyme tyrosine hydroxylase (TH), which enables the production of DA, and the absence of enzymes that convert DA into noradrenalin and adrenalin [19]. Within the ventral midbrain, DA neurons are located in the VTA, substantia nigra *pars compacta* (SNc) and the retrorubral field. The VTA in turn is composed of several subnuclei, at least in rodents: the rostral linear nucleus (RLi), interfascicular nucleus (IF), parabrachial pigmented nucleus (PBP), paranigral nucleus (PN), parainterfascicular nucleus (PIF), and caudal linear nucleus of the raphe (CLi). Also, the rostral-most aspect of the area, which goes under different names, such as VTA rostral and rostro-medial VTA, is most often included when discussing the VTA [9,20–23] (Figure 1).

**Figure 1.** Ample Vglut2 mRNA-positive cells throughout dorsal and ventral midbrain with more sparse expression within the dopaminergic area. Colorimetric in-situ hybridization showing overview

of Vglut2 (**A**,**B**) and Th (**C**,**D**) mRNA in midbrain sections of wildtype adult mouse at two rostro-caudal levels. (**A**,**B**) Vglut2 mRNA is abundant throughout the midbrain with strong signals in e.g., the red nucleus (RN), retrosplenial group of the cortex (RSG) and dentate gyrus (DG), and weaker signals in the ventral tegmental area (VTA) and substantia nigra *pars compacta* (SNc). (**C**,**D**) Th mRNA is selectively localized in dopaminergic neurons of the VTA and SNc; its mRNA signal is implemented to visualize these areas. Dotted square around the VTA and SNc (scale bar 500 mm) presented as closeups in (**A'**–**D'**; scale bar 200 mm). (**C'**,**D'**) SNc, substantia nigra *pars reticulata* (SNr) and subregions of VTA outlined in Th closeups and superimposed on Vglut2 closeups (**A'**,**B'**). (**C'**,**D'**) Th mRNA was strongly localized in the SNc and within the parabrachial pigmented area (PBP) and paranigral nucleus (PN) of the VTA with a weaker signal in the rostral linear nucles (RLi) and caudal aspect of the interfascicular nucleus (IF). (**A'**,**B'**) Within the VTA, Vglut2 mRNA was detected in the PBP, PN, RLi, and IF as well as within the medially-located subzone of the PBP (szPBP) while no Vglut2 mRNA was detected in the GABAergic SNr area. Additional abbreviations: Ctx, Cortex; IPN, interpeducular nucleus. Reprinted from Papathanou et al., 2018 [24].

#### *2.2. Vesicular Glutamate Transporters and the Concept of DA-Glutamate Co-Release*

The midbrain DA system was described in the 1960s [25–27] and has been extensively studied since then. In contrast, the glutamatergic system remained more elusive for many years, to a large extent due to the lack of molecular markers that could enable their reliable identification within complex neuronal networks. Around year 2000 came the first reports of identification of the molecules that transport glutamate into presynaptic vesicles, and thereby, allow this amino acid to be used as a bona fide neurotransmitter ready to be released into the synapse upon neuronal activation [28–32]. These vesicular glutamate transporters, VGLUTs, which exist in three subtypes (VGLUT1, VGLUT2 and VGLUT3), have been found throughout the brain in various anatomical patterns, and their presence defines a neuron's ability for exocytotic glutamate release upon neuronal depolarization. VGLUTs thereby serve as excellent molecular markers of glutamatergic neurotransmission. Together, VGLUT1 and VGLUT2 cover all classically described glutamatergic neurons. In addition, soon after their discovery, it was found that VGLUTs can be present within neurons assigned to another neurotransmitter phenotype than glutamatergic. For example, VGLUT3 was extensively detected in subtypes of serotonergic and cholinergic neurons, and VGLUT1 in inhibitory neurons (for review, see e.g., [33,34]). The discovery of the VGLUTs thereby opened up for a completely new view of many neuronal systems, as they could be shown to possess the molecular machinery for co-release of glutamate parallel to release of its "first" neurotransmitter, often referred to as the "primary" neurotransmitter. Glutamate co-release is a striking neuronal feature which can help explain some complex physiological features that could not be fully accounted for by the primary neurotransmitter. For example, fast excitatory neurotransmission originating from midbrain DA neurons had been recorded in striatal brain slices and mesencephalic cell cultures using electrophysiological setups, however, it was unknown how DA neurons could give rise to this type of signaling [35–39]. With the finding that some DA neurons contain VGLUT2, originally described in in-vitro cell-based systems [40], and subsequently confirmed in a number of histological studies as described below, a molecular basis for the electrophysiologically measurable fast excitatory post-synaptic currents could be initiated and formed.

Neurons releasing more than one neurotransmitter have been given different names to describe their complexity. Based on the capacity of certain midbrain DA neurons to co-release glutamate, a property in which glutamate is released by a neuron upon depolarization, these neurons have been referred to as co-releasing, but also "bi-lingual", "combinatorial" and "dual-signaling" [19,22,33,34], alluding to their ability to "speak different languages". The ability of neurons to release two or more neurotransmitters has also been referred to as "multiplexed neurochemical signaling", or "multiplexed neurotransmission" [41]. In terms of function, the sorting of vesicular neurotransmitter transporters to subcellular domains will account for the inherent property of co-releasing two or more neurotransmitters which can occur from the same synaptic vesicle, or the same pool of synaptic vesicle

within an axon, or from distinct sets of synaptic pools located in different subdomains within axons (reviewed in [33]). The kind of synaptic mechanisms that any given type of co-releasing neuron utilizes needs to be defined experimentally, which can be challenging. While "co-release" generally refers to the release of two or more neurotransmitters from any given neuron, the concept of "co-transmission", which is often used to describe similar phenomena, has been defined more strictly: "*Co-transmission in the strictest sense implies that two neurotransmitters are released at the same time from a common pool of synaptic vesicles within one axon terminal.*" [33]. In the context of glutamate co-release from midbrain DA neurons, this type of more narrowly defined "co-transmission" from a common pool of synaptic vesicles within one axon terminal remains to be experimentally identified and defined. Instead, current data rather point towards the co-release of DA and glutamate from different microdomains in axonal terminals. With their discovery that VGLUT2 and VMAT2, the vesicular monoamine transporter, sort to distinct subpopulations of synaptic vesicles within a subset of mesoaccumbens axons in rodents, Zhang and colleagues recently concluded that " ... *our results do not support the hypothesis that axon terminals from these neurons co-release dopamine and glutamate from identical axonal terminals. Rather, our findings indicate that synaptic vesicles that release dopamine or glutamate from mesoaccumbens terminals in both adult rats and adult mice are located in distinct microdomains*" [42]. Based on existing data and current terminology, we have used the term "co-release" throughout this review to broadly describe the concept of DA and glutamate release from the same midbrain neuron without specifying signaling mechanisms.

In summary, while the midbrain DA system has been the focus of attention in the field of reward and addiction for many years, the "subdiscipline" of DA–glutamate co-release is considerably younger. Consequently, the putative role of DA–glutamate co-release in neurocircuitry and behavioral regulation has remained rather unexplored. However, several studies using experimental animals point towards an impact of DA–glutamate co-release in mechanisms of relevance for addiction, suggesting that glutamate co-release is worthwhile to explore for the benefit of new prevention and/or intervention strategies for substance use disorder. Several recent reviews describe DA–glutamate co-release from different research angles, including possible packaging/release mechanisms, putative role of VGLUTs in promoting vesicular packaging of the primary neurotransmitter, post-synaptic effects of the co-released glutamate and more (see e.g., [34,43–45]). In this review, we will focus our attention on behavioral regulation putatively mediated by DA–glutamate co-release as discovered using rodent models. The feature of DA–glutamate co-release may contribute to dopaminergic function in reward mechanisms and may thus be of importance when considering physiological and behavioral consequences of substance use and abuse.

#### *2.3. Expression Patterns of VGLUT2 in the Midbrain DA System and Validation of Glutamate Co-Release*

To begin to understand functional implications of DA–glutamate co-release, it is important to know where the neurons that possess this ability are located, as only then can neurocircuitry and behavioral roles be fully delineated. The presence of VGLUT2 within midbrain areas where DA neurons reside have been described in several studies, most in which in-situ hybridization for VGLUT2 mRNA has been combined with the detection of either TH mRNA or TH protein for the visualization of DA neurons. Analyses of mouse and rat midbrains have shown that VGLUT2 mRNA-positive cells are scattered throughout the VTA and the adjacently located SNc with more frequent appearance of VGLUT2/TH co-localizing neurons medially than laterally, primarily in the medial aspect of the VTA [46–48]. VGLUT2-positive neurons in the ventral midbrain have also been described in primates, including humans [49]. In rodents, neurons expressing both the VGLUT2 and TH genes comprise a minority of the total number of cells in the adult VTA expressing either the VGLUT2 or the TH gene [47,50,51]. In two recent studies, we could confirm these previous analyses by showing that the highest density of VGLUT2 neurons in the VTA was found in the RLi, followed by the PBP, PN and IF. In the RLi and within a small spatially-restricted area within the PBP, barely any TH-positive neurons can be found, but here VGLUT2 is high. We have called this VGLUT2-dense area within the PBP "the subzone of the PBP" (szPBP) to distinguish it from the remaining PBP which contains

VGLUT2 at lower density but TH at higher density [23,24] (Figure 1). Thus, the PBP in general has a THhigh/VGLUT2low profile, but the szPBP and RLi have the opposite profile, THlow/VGLUT2high.

VGLUT2-TH co-expressing neurons are generally sparse in adulthood, and a temporal regulation between birth and adulthood has been shown [52,53]. When addressing embryogenesis, we could readily identify VGLUT2-TH co-localization already at E12.5 [54] (Figure 2). To address the spatio-temporal profile of VGLUT2 in midbrain DA neurons in more detail, we recently performed a time-study including E14.5, newborn (postnatal day 3, P3) and adult mice (Figure 3) [24]. By co-localizing VGLUT2 mRNA with both TH and dopamine transporter (DAT) mRNAs, we found an interesting temporally dynamic pattern of expression. Almost no co-localization between VGLUT2 with either TH or DAT was detected at E14.5 (Figure 3A–C), while substantial co-localization was observed in the newborn mouse (Figure 3D–F). At P3, VGLUT2 mRNA showed prominent co-localization with both TH and DAT mRNA with higher density of VGLUT2/TH than VGLUT2/DAT double-positive neurons. This is explained by the lower expression of DAT than TH in medial aspects of the VTA, where VGLUT2 is at its highest. In addition, subareas within the VTA showed different amounts of co-localization. Primarily the PBP, but also the PN and IF showed co-localization of VGLUT2 with TH and DAT, respectively. The RLi, which shows the highest levels of VGLUT2 in the VTA, was almost devoid of co-localization of VGLUT2 with either TH or DAT due to the low abundance of these transcripts in this brain nucleus. While readily detected in the newborn mouse, the level of VGLUT2/TH and VGLUT2/DAT double-positive neurons was overall low in all VTA subareas in the adult mouse (Figure 3H,I), confirming previous studies.

**Figure 2.** Vglut2 mRNA and Tyrosine hydroxylase (TH) immunoreactivity in midbrain DA neurons. (**A**–**D**) In-situ hybridization for Vglut2 (VG2) mRNA (black) combined with fluorescent immunohistochemistry for TH (green) on sagittal sections of an E12.5 embryo. Vglut2 mRNA is detected in multiple regions in the embryo, including the ventral midbrain (MB) where DA neurons develop (**A**,**B**). Vglut2 mRNA is co-localized with TH-immunoreactivity in DA neurons within the MB (**C**,**D**). Arrows indicate Vglut2-positive cytoplasm (**B**) and Vglut2/TH double-positive neurons (**D**) (Magnification: B,D 300x). Reprinted from Birgner et al., 2010 [54].

**Figure 3.** Vglut2, Th and Dat mRNA co-localization within certain VTA dopamine (DA) neurons is sparse at E14.5, peaks around birth and is subsequently down-regulated in adulthood. Double fluorescent in-situ hybridization for Th (red), Dat (green) and Vglut2 (red) mRNA, respectively, on wildtype mouse midbrain sections. (**A**–**C**) Sagittal sections of E14.5 embryo. Dotted square around the

area of developing midbrain DA neurons (**A**–**C**) with close-ups in (**A'**–**C'**). (**A**) Th and Dat mRNA show co-localization (yellow) in the ventral midbrain (scale bar 500 mm). (**A'**) higher magnification of insets (scale bar 100 mm); (**B**,**B'**) Th and Vglut2 mRNA expression in the midbrain. (**C**,**C'**) Dat and Vglut2 mRNA show sparse detection in the midbrain. (**D**–**F**) Coronal sections of ventral midbrain in pups of postnatal day (P) 3. (**D**) Th and Dat show ample co-localization (yellow) in the lateral VTA and SNc (scale bar 250 mm, inset 100 mm). (**E**) Th and Vglut2 mRNA and (**F**) Dat and Vglut2 mRNA prominently co-localize (yellow) at this age in the IF, PBP and PN areas (arrows) but not in the RLi of the VTA. (**G**–**I**) Coronal sections of the adult midbrain (10 weeks; scale bar 250 mm, inset 100 mm). (**G**) Th and Dat mRNA co-localization (yellow) remains strong; whilst the level of co-localization between (**H**) Th and Vglut2 and (**I**) Dat and Vglut2 mRNAs is lower than at P3 (arrows). Yellow arrows show co-localization green (Dat) and red (Vglut2) channel, red arrows show red (Vglut2) channel (Postnatal Day (P) 3 *n* = 3; adult *n* = 3). Abbreviations: cf, cephalic flexture; Ctx, cortex; IF, interfascicular nucleus; IPN, interpeducular nucleus; LV, lateral ventricle; M, medulla; Mb, midbrain; PBP, parabrachial pigmented area; PN, paranigral nuclei; RLi rostral linear nucleus; SNc, Substantia nigra pars compacta; VTA, Ventral tegmental area. Reprinted from Papathanou et al., 2018 [24].

To further the understanding of VGLUT2 expression in developing DA neurons, we have recently addressed VGLUT2 in DA neurons from the time-point around mid-gestation when these neurons are born and can now show that most early differentiating midbrain DA neurons express VGLUT2 at stages E10–11 (Dumas and Wallén-Mackenzie, in press, 2019). Further, we show that this early and abundant VGLUT2-expression is subsequently down-regulated as embryonal development proceeds, providing an explanation for the higher appearance of VGLUT2 in E12.5 than E14.5 described above. By including this novel piece of data into a concept of VGLUT2 expression in midbrain DA neurons, it seems that most, if not all, DA neurons initially express VGLUT2 which is subsequently downregulated during embryonal development to be upregulated in subsets of VTA neurons around birth and again down-regulated in adulthood.

In summary, VGLUT2 co-localizes to a higher extent with TH and DAT in newborn than in adult mice. With the expression of VGLUT2 in DA neurons primarily during early embryogenesis and during a subsequent phase around birth, it is interesting to speculate around putative roles of VGLUT2 in the developing DA neuron. In addition to its temporally regulated expression during the normal life span starting from embryogenesis, it has been shown that VGLUT2 expression levels can be induced in mature organisms upon stress and injury [53,55], suggesting that a glutamate co-releasing phenotype can be acquired, or at least accentuated, at different time-points during life in response to particular experiences. VGLUT2 thus shows an interesting spatio-temporal regulation pattern that may have important implications for behavioral regulation and disorders of the DA system throughout life. Below, we will discuss conditional knockout (cKO) studies in mice that support this observation.

In terms of projections of DA–glutamate co-releasing neurons, DA neurons in the medial aspect of the VTA, where VGLUT2 levels are highest, are known to project to the NAc mSh [10]. Indeed, optogenetics-driven analyses have confirmed previous observations from electrical stimulations in slice and cell culture systems and further shown that that stimulation of DA neurons in the VTA gives rise to excitatory post-synaptic currents in MSNs primarily on the ventral, rather than the dorsal, striatum [56,57]. Perhaps most attention in the DA–glutamate field has been given to MSNs, but also other striatal neurons have been shown to receive input from DA–glutamate co-releasing neurons. Lately, Chuhma and colleagues demonstrated that midbrain DA neurons induce post-synaptic glutamatergic effects in all neuronal types present in the striatum (MSNs, cholinergic interneurons (ChIs also referred as CINs) and fast-spiking interneurons (FSI)), but that the effects are different depending on the anatomical region where the neurons are located, and, specifically for the NAc mSh, most profound on ChIs [58]. While most prominently projecting to the ventral striatum, post-synaptic glutamatergic effects by DA–glutamate co-releasing VTA neurons have also been observed in these

same neuronal cell types in the dorsal striatum. Interestingly, ChIs in the dorsomedial parts were shown to be affected differently than those located in the dorsolateral parts [59,60].

Having summarized current knowledge of anatomical positions of DA–glutamate co-releasing neurons in the VTA and their striatal target neurons, the next section will deal with behavioral consequences upon experimentally-achieved disruption of DA–glutamate co-release using transgenics-based approaches in mice.

#### *2.4. Behavioral Consequences upon Disrupted Dopamine–Glutamate Co-Release in Transgenic Mice*

The current literature contains several studies in which conditional knockout technology in mice has been used to understand if targeted disruption of VGLUT2 gene expression within DA neurons has any measurable effect on behavioral output. These studies, which are all different implementations of the Cre-LoxP system with the aim to delete glutamate co-release in DA neurons, have in common that they identify significant alterations in responsiveness to natural rewards and/or drugs of abuse (listed in Table 1). Together, these studies strongly implicate the importance of DA–glutamate co-release in mechanisms of relevance to addiction. We will discuss some of these studies in more detail as specific features might be of particular importance when considering DA–glutamate co-release in mechanisms of addiction. We will also go through putative caveats that might be important to bear in mind when interpreting these behavioral data, some of which are based on recent discoveries of so called "off-target" effects in transgenics methodology.

**Table 1.** List of publications in which dopamine-glutamate co-release has been targeted to probe its putative behavioral roles. Main results summarized. – indicates no change; ↑ increase; ↓ decrease. Abbreviations: RT-PP, real-time place preference; VTA, ventral tegmental area.



**Table 1.** *Cont.*

#### *2.5. DAT-Cre-Mediated Gene Targeting of VGLUT2 in Studies Aiming to Unravel the Importance of Glutamate Co-Release in Neurocircuitry of Reward and Addiction*

Parallel to findings of VGLUT2 gene expression in midbrain DA neurons, initially detected in the adult rodent VTA [46] and in DA cell cultures [40], our own histological studies identified VGLUT2 in TH-positive neurons of the ventral midbrain of the mouse not only in adulthood but also in the developing embryo already at gestational day 12.5 (E12.5) [54] (Figure 2). Such an early developmental expression suggested to us that VGLUT2 might be important for proper DA cell development, and consequently, for dopaminergic functions at adulthood. At that time, we took two Cre-LoxP approaches to disrupt VGLUT2 in midbrain DA neurons to address if DA–glutamate co-release had any influence on dopaminergic functions. In both approaches, a floxed VGLUT2 mouse line [66,67] was crossed with a Cre-driver transgenic mouse to direct the targeting event to DA neurons: (i) TH-Cre and (ii) DAT-Cre. Both these Cre-drivers are active during embryonal development, from stages around mid-gestation of the mouse embryo. We also tested the "opposite" strategy, to delete the Vesicular monoamine transporter (VMAT2) gene in cells positive for VGLUT2 using a VGLUT2-Cre transgenic mouse line to target a floxed VMAT2 allele. However, this strategy, in which we aimed to delete DA signaling rather than glutamatergic signaling from DA–glutamate co-releasing neurons, resulted in dead pups around birth. We speculated that the neonatal death was due to a breathing phenotype similar to what we observed when deleting VGLUT2 in all cells [66], but caused by monoaminergic rather than glutamatergic loss of function in the breathing circuitry. However, beyond VGLUT2 in midbrain DA neurons, VGLUT2 is expressed in additional monoaminergic neuronal populations in the medulla, including noradrenergic neurons of nuclei A1 and A2 and adrenergic neurons of C1, C2, and C3 nuclei (see e.g., [68,69]), and loss of VMAT2 in these neurons could cause the loss of a series of vital autonomic functions. Further investigations would have been required to fully address the mechanisms for neonatal lethality of targeted deletion of VMAT2 in VGLUT2-neurons, however, we did not perform such studies.

By focusing on gene-targeting of VGLUT2, using the TH-Cre approach to disrupt VGLUT2 expression, we realized early on that TH-Cre will direct targeting to VGLUT2-positive neurons beyond DA neurons, due to an early and non-monoaminergic-selective phase of TH promoter activity [70]. The *Vglut2f*/*f;TH-Cre* cKO mice showed interesting hippocampal phenotypes [70], but were not used in our laboratory for further analysis of DA–glutamate co-releasing neurons. Instead, the approach using DAT-Cre to direct the targeted deletion of VGLUT2 to DA neurons was more promising. *Vglut2f*/*f;DAT-Cre* cKO and control mice were analyzed in several behavioral, electrochemical and biochemical parameters [54]. We found that basal motor and memory functions were normal in these cKO mice, but that their risk-taking behavior was altered. We also found that in both home-cage and novel environments, *Vglut2f*/*f;DAT-Cre* cKO mice showed a greatly blunted overall response to the psychostimulant amphetamine, which exerts strong effects on DA release. Specifically interesting, in a home-cage environment in which all movements were automatically registered, cKO mice showed a strikingly different response than the control with lower total activity at several different doses analyzed (Figure 4). When analyzing horizontal and vertical locomotion parameters separately, it was apparent that the cKO and control mice had strikingly different dose-response curves in terms of behavioral output. At higher doses, locomotion and rearing decreased in the control mice as stereotypic behavior became dominating. Stereotypy was detected as bodily shakings, and in the automated recording of any movement displayed by the mice, these shakings were recorded and contributed to the parameter of total activity. In control mice, the onset of stereotypic behavior thus caused a high level of total activity despite a reduction in locomotion and rearing. In contrast to control mice, cKO mice showed less stereotypy but increased both locomotion and rearing with higher doses. Since both motor function and basal locomotor activity were normal in the cKO mice, these parameters did not contribute to the differential response observed between genotypes. The finding rather suggested that the induction of stereotypic behavior seen at the higher doses in control mice did not at all develop to the same extent in *Vglut2f*/*f;DAT-Cre* cKO mice. The absence of stereotypy likely contributed to the overall lower activity of the cKO mice compared to control mice despite the increase in locomotion and rearing displayed by cKO mice at higher amphetamine doses. In contrast to this remarkable response at higher doses, both locomotion and rearing parameters were lower in cKO than control mice at lower doses, a blunted type of locomotion which might reflect a lower release level of mesostriatal DA in the absence of VGLUT2 in midbrain DA neurons, or which might be directly caused by the reduction of VGLUT2-mediated glutamatergic neurotransmission in the circuitry. In this context, it might be important to bear in mind that VGLUT2 in DA neurons is most abundant during embryonal development (as discussed above), when the DAT-Cre-mediated targeting of VGLUT2 is initiated in the *Vglut2f*/*f;DAT-Cre* cKO mice, suggesting that behavioral manifestations displayed by the cKO mice might be due to neurocircuitry compensations. Developmental adaptations will be further discussed below. In summary, the results of this first study implementing DAT-Cre to disrupt VGLUT2 in DA neurons with the aim to probe the putative importance of DA-glutamate co-release, demonstrated a strikingly altered response curve to amphetamine in the absence of VGLUT2. This finding led us to suggest that VGLUT2 in DAT-Cre neurons is important for the behavioral response to the psychostimulant amphetamine [54]. As described below, this initial finding was substantiated with subsequent studies using additional types of rewards and behavioral paradigms.

**Figure 4.** *Vglut2f*/*f;DAT-Cre* mice show blunted behavioral response to amphetamine, as measured in a home-cage environment. (**A** and **B**) Male and female knockout mice showed a reduced behavioral response to amphetamine as shown by significantly lower total activity at all three doses of amphetamine. (**C** and **D**) Male and female knockout mice display a shift in dose-response to amphetamine-induced locomotion. (**E** and **F**) Male and female knockout mice show a different dose-response profile regarding rearing behavior in response to amphetamine. A right shift in dose-response in locomotion is apparent in the *Vglut2f*/*f;DAT-Cre* mice. Data were analyzed with one-way ANOVA followed by Tukey's post-hoc test when appropriate. Data are presented as mean ± SEM (*n* = 9). \*, *P* < 0.05; \*\*, *P* < 0.01; \*\*\*, *P* < 0.001. Reprinted from Birgner et al., 2010 [54].

Using mouse genetics, several different DAT-Cre transgenic lines and floxed VGLUT2 lines have been produced and combined to create various *Vglut2f*/*f;DAT-Cre* cKO lines for the study of DA–glutamate co-release. By using different versions of DAT-Cre and floxed VGLUT2 mouse lines, the different *Vglut2f*/*f;DAT-Cre* cKO lines that exist are not identical but similar, a feature which could be of importance when analyzing the results generated. Also, experimental procedures can vary between laboratories. However, while some discrepancies can been detected when comparing results from different *Vglut2f*/*f;DAT-Cre* cKO mouse lines and laboratories, overall, the findings point towards the same conclusion which implies DA–glutamate co-release in mechanisms of relevance to addiction. In one important study, Hnasko and colleagues confirmed that midbrain cell cultures from one *Vglut2f*/*f;DAT-Cre* cKO mouse line lacked VGLUT2 in synaptic boutons of DA neurons and that cKO mice had reduced excitatory post-synaptic potentials (EPSCs) on MSNs of the NAc in response to electrical stimulation of the VTA [63]. Using optogenetic stimulation of DA neurons, Stuber and colleagues subsequently demonstrated that the EPSCs on NAc MSNs were completely abolished in these *Vglut2f*/*f;DAT-Cre* cKO

mice [57]. Behaviorally, *Vglut2f*/*f;DAT-Cre* cKO mice were confirmed to not have deficits in spontaneous locomotion or motor coordination [63], however, one study has reported reductions in spontaneous activity and disturbances in motor coordination in the rotarod motor test in one line of *Vglut2f*/*f;DAT-Cre* cKO mice [62]. Several studies have by now confirmed that *Vglut2f*/*f;DAT-Cre* cKO mice have blunted locomotor activity in response to acutely administered cocaine and amphetamine [54,62,63], but it has also been shown that cKO mice have normal behavioral sensitization after repeated injections of cocaine compared to their control littermates [63]. The cocaine-induced locomotor responses seem to be dissociated from the rewarding effects of cocaine since the cKO mice showed intact conditioned place preference (CPP) to the drug [63].

Following up on these initial studies in which the experimenter delivered drug injections to the rodent, we wished to advance the knowledge further by using operant self-administration methodology to come closer to the situation where the subject itself has the possibility to choose whether or not to administer a rewarding substance. Using the same *Vglut2f*/*f;DAT-Cre* cKO mouse line as in our previous experiments [54], we next analyzed the responses to both natural (sugar) and drug reward in such a self-administration paradigm to investigate how disruption of glutamate co-release from DA neurons is involved in reward processing. In the sugar self-administration test, in which nose-poking in an active nose-poke led to sucrose pellet delivery, *Vglut2f*/*f;DAT-Cre* cKO mice acquired the operant behavior similarly to control mice, but they self-administered significantly more sugar compared to controls. This was particularly evident when the task requirements were higher (FR5: 5 nose-pokes to receive a sucrose pellet) [61]. These responses were selective towards the highly palatable food and resistant to satiety; the cKO mice still consumed more calories from sucrose during the high-requirement task (FR5) compared to littermate controls [61]. When we then tested *Vglut2f*/*f;DAT-Cre* cKO mice in a cocaine self-administration paradigm, the mice displayed increased behavioral responding to receive lower doses of cocaine (Figure 5). Furthermore, during a cue-induced reinstatement phase, where nose-poking would lead only to presentation of cues associated with cocaine intake but not cocaine itself, cKO mice responded significantly more than controls just to receive the cues [61]. Taken together, these results demonstrate that mice lacking the ability to co-release glutamate from DA neurons display distinct patterns of behavior related to reward processing. While *Vglut2f*/*f;DAT-Cre* cKO mice show reduced locomotor activity in response to psychostimulants, specific aspects of the rewarding process associated to these drugs are intact. The animals form normal pavlovian associations in response to cocaine in CPP and they acquire similar rates of cocaine self-administration to control mice. In contrast, *Vglut2f*/*f;DAT-Cre* cKO mice seemed to be more sensitive to sugar and lower doses of cocaine, and strikingly, they were more susceptible to environmental stimuli associated with drug intake as shown by increased response in a cue-induced reinstatement phase during a cocaine self-administration paradigm.

**Figure 5.** Elevated operant responding for low-dose cocaine and drug-paired cues during extinction in cKO mice. (**A**) Nosepoke response of food-trained mice implanted with indwelling intravenous catheters and allowed to nosepoke for cocaine infusions (controls, *n* = 7; cKO, *n* = 7). (**B**) Total dose self-administered at the different cocaine concentrations in (**A**). (**C**) Cocaine seeking, i.e., responding for light and sound cues previously associated with cocaine in the absence of the drug, of mice from A that had been responding for 0.75 mg/kg per infusion (inf) and subsequently subjected to 21 d of forced cocaine abstinence (controls, *n* = 6; cKO, *n* = 5). Group data represent mean ± SEM. \**p* < 0.05; \*\**p* < 0.01 versus Ctrl. Reprinted from Alsiö et al., 2011 [61].

To address molecular mechanisms, we next implemented a series of biochemical analyses. We found distinct molecular alterations in certain anatomical areas within the reward system of *Vglut2f*/*f;DAT-Cre* cKO mice. Interestingly, these alterations could be observed both under baseline conditions and after cocaine administration and might explain the observed behavioral patterns discussed above. *Vglut2f*/*f;DAT-Cre* cKO mice displayed elevated numbers of D1 receptors (D1R) in dorsal striatal areas and D2 receptors (D2R) in the shell region of the NAc, detected as increased binding of radio-labeled D1R and D2R ligands, respectively [61]. Furthermore, we found that cKO mice showed elevated mRNA levels of the nuclear receptor and immediate early gene Nur77 under baseline conditions in NAc core and dorsal striatal areas. The same elevation was seen in levels of c-Fos mRNA, another immediate early gene commonly used to assess neuronal activation. The elevated levels of Nur77 under baseline conditions in cKO mice were comparable to the ones observed is control mice after cocaine administration (Figure 6) [61]. Nurr77 expression is thought to be tonically inhibited by DA in the striatum and its upregulation to be related with neuroadaptations induced by chronic administration of drugs of abuse [71]. It is possible that the enhanced presence of D2R in the NAc disrupts the DA-mediated tone on Nur77 expression. Noteworthy, the detection of elevated levels of D2R might also reflect an increase in autoreceptors in presynaptic terminals related to decreased DA release. Increased levels of striatal DA receptors and immediate early genes Nur77 and c-Fos might render *Vglut2f*/*f;DAT-Cre* cKO mice more susceptible to neuroadaptations related to chronic administration of drugs of abuse, or more sensitive to their rewarding properties. This could also be the case for natural rewards. In addition, several studies have identified reduced stimulation-induced DA levels in the striatum of these mice [61–63]. This reduction could be related to disrupted development of the DA system [62] or due to altered intracellular dynamics and reduced synaptic packaging of DA, the "primary" neurotransmitter, due to the absence of VGLUT2 in presynaptic terminals [63,72]. In both scenarios, less DA in the synaptic cleft would result in the cessation of the inhibitory signal to Nur77 expression which in turn may lead to the long-lasting neuroadaptations and behavioral manifestations that have been demonstrated in the studies discussed above. Together, these biochemical analyses demonstrated that gene-targeting of VGLUT2 in DAT-Cre neurons leads to molecular neuroadaptations that are similar to changes observed in control animals only upon systemic administration of addictive drugs (here demonstrated with cocaine). *Vglut2f*/*f;DAT-Cre* cKO mice showed baseline levels of Nur77 and c-Fos elevated to the extent that cocaine administration did not increase them further, instead, baseline expression was already at the level observed in control mice upon drug administration. Together, the analyses of D1R and D2R availability and levels of immediate early gene expression strongly argue for a potent role of VGLUT2 in maintaining molecular homeostasis: When VGLUT2 is removed in DAT-Cre neurons from embryogenesis, there is a shift in the abundance of a range of molecular players that are important for normal functions of the DA system and this molecular shift is, in turn, likely to contribute to the behavioral manifestations observed in various reward-related experimental paradigms in which the response of the cKO mice differ significantly from that of the control mice.

**Figure 6.** High expression of Nur77 under baseline conditions in cKO mice. (**A**–**D**) Quantitative in-situ hybridization of Nur77 mRNA levels in Ctrl and cKO mice treated with saline (Ctrl, *n* = 6; cKO, *n* = 7) or cocaine (Ctrl, *n* = 5; cKO, *n* = 4). (**A**) Levels of Nur77mRNAin AcSh and AcC (bregma, 1.70 mm). (**B**) Nur77mRNAlevels in medial (StM) and lateral (StL) rostral striatum (bregma, 1.70 mm). (**C**) Nur77 transcript levels in the caudodorsal striatum (bregma, 0.48 mm). (**D**) Nur77 levels in the caudoventral striatum (bregma, 0.48). (**E**,**F**) Representative autoradiograms of Nur77 mRNA in-situ hybridization signals in the rostral striatal area (**E**; bregma, 1.70 mm) and in the caudal striatal area (**F**; bregma, 0.48 mm) with schematic illustrations to the right showing the location of the analyzed regions. Group data represent mean ± SEM expressed as percentage of Ctrl (saline). \**p* < 0.05, \*\**p* < 0.01, and \*\*\**p* < 0.001 versus Ctrl saline group; #*p* < 0.05 versus Ctrl cocaine group. cc, Corpus callosum. Reprinted from Alsiö et al., 2011 [61].

Drug and natural rewards can have several sites of action and influence systems that are not directly related to DA. Experimentally, optogenetic stimulation of the DA system has been used to isolate and dissect out the role of DA neurons in reward responses [73–75]. Previous studies have demonstrated that optogenetic stimulation of VTA DA neurons in mice has potent reinforcing effects on behavior and can induce molecular and behavioral adaptations similar to those observed after cocaine self-administration [76]. In a recent study, *Vglut2f*/*f;DAT-Cre* cKO mice were addressed in a collaborative effort to further the understanding of DA-glutamate co-release by using optogenetic approaches to avoid non-selective effects of drug or natural rewards. In this study by Wang et al., we could confirm the reduction of glutamatergic neurotransmission in *Vglut2f*/*f;DAT-Cre* cKO mice using both patch-clamp electrophysiology in slice preparations and in-vivo amperometry in the living mouse [65] (see Viereckel et al., [77], for protocol for optogenetics-coupled glutamate electrochemistry in-vivo). By implementing an optogenetics-based intracranial self-stimulation behavioral paradigm, it was found that the acquisition of operant behavior to optogenetically self-stimulate DA neurons in the VTA remained intact in *Vglut2f*/*f;DAT-Cre* cKO mice. However, the rate of responding when parameters of stimulation changed (intensity of stimulation from 8 mW to 32 mW) was lower in *Vglut2f*/*f;DAT-Cre* cKO mice compared to littermate controls [65]. Further, cKO and control mice showed the same level of real-time place preference (RT-PP) for a compartment paired to optogenetic stimulation [65].

To further explore how gene-targeting of VGLUT2 in DAT-Cre neurons might influence dopaminergic function, we recently performed a small pilot study in which *Vglut2f*/*f;DAT-Cre* cKO and control mice were compared in an extended version of the optogenetic self-stimulation paradigm described by Wang et al. [65]. We increased the testing time compared to the previous study [65] and also added several new phases to the program in an attempt to model different aspects of food or drug self-administration: Both fixed and progressive ratios were analyzed, followed through with sessions of forced abstinence, reinstatement, extinction and cue-induced reinstatement, respectively (Figure 7A). Upon stereotaxic surgery to deliver optogenetic DNA constructs to the VTA, mice were first validated in the optogenetic RT-PP paradigm which confirmed the strong RT-PP displayed by both *Vglut2f*/*f;DAT-Cre* cKO and control mice, in accordance with the previous study [65]. Next, the mice were tested in the extended optogenetic self-stimulation program and we could again confirm that cKO and control mice showed the same acquisition rate of the self-stimulation behavior [65]. Further, our extended program found no differences between cKO and control mice upon testing in a single progressive ratio session, in a reinstatement session after a period of forced abstinence or in a 5-day extinction phase where lever-pressing no longer resulted in optogenetic stimulation or cue presentation. However, when the mice were analyzed in a cue-induced reinstatement session, where lever-pressing resulted in cue-light presentation but not laser-stimulation (Figure 7B), cKO mice responded by lever-pressing significantly more vigorously than their littermate controls (Figure 7C & D). While this extended version of the optogenetic self-stimulation program has only been addressed in a modest number of animals, the striking difference observed between cKO and control mice in their behavioral response to cue-presentation should be of interest to address further to fully validate the findings and find out any underlying mechanims.

In summary, taken together with the previous study [65], the presented results here show that the actual acquisition of operant responding in the optogenetics-based self-stimulation paradigm is not different in *Vglut2f*/*f;DAT-Cre* cKO mice compared to control mice. Beyond this observation, by covering several additional behavioral parameters, our pilot study demonstrated that targeted removal of VGLUT2 in DAT-Cre neurons is sufficient to induce profound changes in the animal's response to cues associated with reward delivery. When considering this new result in the context of our previous data pin-pointing that the same transgenic line of *Vglut2f*/*f;DAT-Cre* cKO mice displayed increased behavioral response to cues associated with cocaine, an interesting picture appears: Mice lacking the ability for DA–glutamate co-release demonstrate distinct behavioral disturbances related to increased sensitivity to reward-associated cues. This finding strongly implies a role for DA–glutamate co-release in cue-induced reinstatement which, given its importance for relapse in human addicts, should be of particular interest to explore further in the context of interventive strategies.

**Figure 7.** (**A**). Timeline of the intracranial optogenetic self-stimulation (ICSS) experiments. (**B**). Schematic representation of the ICSS procedure and the cue-induced reinstatement phase. (**C**). Active and inactive lever presses during the cue-induced reinstatement period for cKO and control mice. (**D**,**E**)**.** Average of inactive (**D**) and active (**E**) during the cue-induced reinstatement phase cKO and Ctrl mice. Data are presented as mean ± SEM. \*\*\* *p* < 0.001 vs. ctrl. Ctrl *n* = 3, cKO *n* = 3. Original data, Bimpisidis and Wallén-Mackenzie, pilot study, 2019.

#### *2.6. DAT-Cre-Mediated Gene Targeting of Phosphate-Activated Glutaminase (GLS1)*

As mentioned above, knocking out VGLUT2 in DA neurons during embryonal development caused reduced release of DA in striatal areas in adulthood [61–63]. This finding might be either the result of disturbances in the development of the DA system when VGLUT2 is no longer present in these neurons [62], due to changes in the presynaptic milieu as a result of VGLUT2 absence [63,72], or both. Mingote and colleagues [64] applied a genetic approach to reduce glutamate release from DA neurons independently of VGLUT2 with the idea to circumvent the effects on DA availability upon VGLUT2 targeting. By using a *DATIREScre*/+*::GLS1lox*/+ mouse line in which the expression of phosphate-activated glutaminase (GLS1) was disrupted on the heterozygotic level in DAT-Cre neurons, a conditional reduction of glutamate synthesis in DA neurons was achieved. Using this approach, glutamate release from DA neurons was targeted in a frequency-dependent manner; the maintenance of glutamate release was disturbed mainly during high frequencies, normally associated with reward processing. Behaviorally, mice with reduced glutamate synthesis in DA neurons exhibit attenuated amphetamine sensitization and potentiated latent inhibition [64]. Latent inhibition refers to the phenomenon during which animals display attenuated conditioning when they are repeatedly pre-exposed to the conditioned stimulus in neutral settings, a condition thought to model the inability of patients with schizophrenia to show selective attention and to ignore irrelevant stimuli [78]. Attenuated behavioral sensitization and potentiated latent inhibition were previously associated to a mouse line with reduction of glutamate synthesis throughout the brain [78], but Mingote and colleagues narrowed down their observations to prove that the effects are mediated by reduction of glutamate synthesis selectively from DAT-Cre neurons. Indeed, a mouse line with reductions of glutamate synthesis in forebrain areas did not demonstrate the same behavioral manifestations [64]. The role of DA–glutamate

co-release in latent inhibition should be of interest to further research in the context of substance use/abuse.

#### *2.7. Some Caveats in the Implementation of Transgenics to Address Neuronal Function*

All studies discussed above implemented transgenic approaches that target glutamate co-release from early developmental stages with the common feature of DAT-Cre transgenic mouse lines to drive the recombination of floxed alleles. Since the activity of the endogenous promoters for VGLUT2 [54,70] and DAT [79] both have embryonal onset, it is possible that the observed phenotypes are the result of developmental adaptations resulting from the gene-targeting event of VGLUT2. This possibility has been discussed in the literature cited above, and indeed, it has also been experimentally shown in one *Vglut2f*/*f;DAT-Cre* mouse line that cKO mice have impaired DA neuron development; the number of midbrain DA neurons and their projections to striatal areas are significantly reduced when VGLUT2 is not present in DAT-Cre neurons [62]. Also, as discussed above, major molecular neuroadaptations occur as a consequence of VGLUT2 gene-targeting with the elevation of D1R and D2R availability and enhanced c-Fos and Nur77 expression levels at baseline conditions. Based on all of these observations, it is highly conceivable that VGLUT2 plays a developmental role in the establishment of the nervous system, a role that stretches beyond its role as pre-synaptic transporter of glutamate into synaptic vesicles. While it has been challenging to dissociate putative developmental adaptations following gene-targeting of VGLUT2 in DA neurons from effects that are solely dependent of loss of vesicular glutamate packaging, issues related to VGLUT2 were avoided using the GLS1-approach presented by Mingote and colleagues [64]. However, not only regulation of VGLUT2 but also the spatio-temporal regulation of the DAT promoter, the sequence of which is used to drive the expression of Cre recombinase in DAT-Cre mice, may also present challenges. Implementation of developmentally regulated promotors always carries the risk of neuroadaptations that might be responsible for behavioral phenotypes, thus leading to misleading conclusions. Furthermore, ectopic expression of Cre recombinase can lead to unwanted targeting [80–82]. In the context of DAT-Cre, we could recently show that several DAT-Cre mouse lines show ectopic expression of Cre recombinase in multiple brain areas that are not associated with monoaminergic neurotransmission. For example, certain amygdaloid subnuclei, septal nuclei and neurons in the lateral habenula, all of which contain VGLUT2 but not DAT, were strongly positive for the DAT-Cre transgenes [82]. Bearing this important observation in mind, it is possible that any results obtained using a DAT-Cre transgene to achieve gene-targeting of a floxed VGLUT2 or GLS1 allele, or any other floxed allele, might be dependent on recombination of the floxed gene in these non-monoaminergic areas. Possible "off-target" effects mediated by DAT-Cre are crucial to consider as they could have consequences on physiological and behavioral output, and hence on conclusions drawn from such experiments. Finding more selective tools to address DA–glutamate co-release in reward and addiction is highly relevant, not least in light of this new revelation.

#### *2.8. DA–Glutamate Co-Release in Neuronal Plasticity within the Ventral Striatum*

To address DA–glutamate co-release specifically in the adult mouse and to avoid any developmental effects upon VGLUT2-gene-targeting, we recently applied an inducible knockout approach [24]. In this study, we took advantage of a tamoxifen-inducible DAT-CreERT2 mouse line [83] in which DAT-Cre-expressing neurons will translocate Cre recombinase into the nucleus only upon tamoxifen administration. We could show that the spatial expression pattern of Cre recombinase in the inducible DAT-CreERT2 line is similar to the conventional DAT-Cre line implemented in our previous studies, while the temporal expression is regulated by the time at which the tamoxifen is provided. The new *Vglut2f*/*f;DAT-CreERT2* cKO mouse line was compared with our previously published *Vglut2f*/*f;DAT-Cre* cKO mouse line (described above) in a range of experiments in order to have the same VGLUT2 floxed allele targeted by the two different DAT-Cre-drivers. By comparative analysis of *Vglut2f*/*f;DAT-CreERT2* and *Vglut2f*/*f;DAT-Cre* cKO mice, we observed that when glutamate co-release was

disrupted in adulthood, the mice displayed strikingly different behaviors in response to amphetamine and cocaine compared to mice with a developmentally-induced VGLUT2 targeting event. Unlike the drug-induced locomotor response observed in the *Vglut2f*/*f;DAT-Cre* cKO mice, which was lower than that shown by control mice, the inducible *Vglut2f*/*f;DAT-CreERT2* cKO mice displayed a similar level of drug-induced locomotor activity as controls [24]. Additionally, when electrophysiological experiments were performed to investigate how reduced levels of glutamate release from DA neurons might affect the plasticity observed after chronic drug administration, we found that inducible *Vglut2f*/*f;DAT-CreERT2* mice displayed characteristic increases in markers of synaptic plasticity already under baseline conditions. *Vglut2f*/*f;DAT-CreERT2* mice showed an increased AMPA/NMDA ratio on D1R-expressing MSNs compared to controls, and these changes were sufficient to occlude further increases normally seen following chronic cocaine administration [24] (Figure 8A). Increased AMPA/NMDA ratio is indicative of the presence of GluR2-lacking subunits of the AMPA receptor that have higher peak conductance and Ca2<sup>+</sup> permeability and thus higher inward-rectifying properties (expressed as higher rectification index) [1]. While the inducible *Vglut2f*/*f;DAT-CreERT2* cKO mice showed an increased AMPA/NMDA ratio in baseline conditions, the rectification index of MSNs was unaltered (Figure 8B), reflecting mechanisms others than the replacement of GluR2 in AMPARs [24]. These mechanisms may include net increases in AMPA receptors, decreases in NMDA receptors, or both. Clearly, the mechanisms underlying the observed increases in baseline AMPA/NMDA ratios require further investigation to fully understand this type of regulation. However, already now, these data suggest that deletion of VGLUT2 in mature neurons induced changes in neuronal plasticity already under baseline conditions and that inducible *Vglut2f*/*f;DAT-CreERT2* cKO mice might have abnormal drug- or other stimuli-induced plasticity. More studies will be needed to find out how these observed changes might be of importance to addiction.

**Figure 8.** VGLUT2 targeting in mature DA neurons results in an elevated baseline AMPA/NMDA ratio. (**A**) AMPA/NMDA ratio and raw traces of cells from *Vglut2f*/*fDAT-CreERT* cKO and ctrl mice treated with saline or cocaine for NMDA current (blue); AMPA current (light gray). (**B**) Rectification index (RI) and

raw traces recorded cells from *Vglut2f*/*fDAT-CreERT* cKO and ctrl mice treated with saline or cocaine. eKO-DRD1 and eCtrl-DRD1 mice treated with saline or cocaine at −70 mV (blue), 0 mV (light gray) and +40 mV (dark gray). Two-way ANOVA Sidak post hoc (ANOVA ###*p* < 0.001; post hoc between genotype \**p* < 0.05 and post hoc between treatment of same genotype: \**p* < 0.05, \*\**p* < 0.01, \*\*\**p* < 0.001; RI saline: saline *Vglut2f*/*fDAT-CreERT* Ctrl, *n* = 6, *Vglut2f*/*fDAT-CreERT* cKO, *n* = 6; cocaine: *Vglut2f*/*fDAT-CreERT* Ctrl, *n* = 12; *Vglut2f*/*fDAT-CreERT* cKO, *n* = 7; AMPA/NMDA ratio saline:: *Vglut2f*/*fDAT-CreERT* Ctrl, *n* = 9, *Vglut2f*/*fDAT-CreERT* cKO, *n* = 9; cocaine: *Vglut2f*/*fDAT-CreERT* Ctrl, *n* = 13, *Vglut2f*/*fDAT-CreERT* cKO, *n* = 7). Whole-cell patch clamp experiments performed on slices 10 days after last saline or cocaine injection. Reprinted from Papathanou et al., 2018 [24].

In summary, the main take-home message appearing from the various studies knocking out VGLUT2 in DAT-Cre neurons is that detectable neurocircuitry and behavioral changes do indeed occur upon targeted disruption of VGLUT2. No matter which combination of DAT-Cre transgene and floxed VGLUT2 allele that have been combined to probe DA–glutamate co-release, behavioral manifestations all point towards functional aspects of particular interest to reward and substance use/abuse. The availability of VGLUT2 in DA neurons of human individuals [49] might thereby be an important aspect to address in terms of vulnerability and pre-disposition to addiction, as briefly discussed further below.

#### *2.9. Dopamine–Glutamate Co-Release: Implications for Reward Processing*

How can all these observations from experimental animals be brought together into a comprehensive model to understand the potential role of DA–glutamate co-release in addiction? Despite the fact that additional studies are necessary in order to fully understand the role of co-release in behavioral regulation even under normal conditions, let alone under conditions of substance abuse, the findings reported in the research field so far can be summarized to form hypotheses related to the observed manifestations.

Drugs that are abused by human individuals can change brain physiology in many ways and eventually lead to characteristic and chronic behavioral patterns clinically known as substance use disorder. An abundance of studies has implicated several major neurotransmitter systems, including the DA and glutamate systems, in these phenomena while the role of combined DA and glutamate effects derived from the same neuron, i.e., DA–glutamate co-release, has only recently begun to be explored. Most studies in the addiction field have focused on the MSNs, as these neurons constitute the majority of neurons in the striatum. Similarly, initial studies of post-synaptic effects of DA–glutamate co-release have mainly been focused on recordings of MSN activity [24,56,57,63]. In the context of MSNs, Adrover and colleagues concluded that acute cocaine administration attenuates the DA neuron-induced EPSCs on MSNs through D2R activation based on the observation that the D2R antagonist sulpiride could reverse these effects [84]. Further, repeated cocaine administrations altered glutamatergic synapses between VTA terminals and NAc shell neurons in a withdrawal-dependent manner. After one day of withdrawal, there were no significant effects on the synapses, but when the VTA-NAc shell synapses were investigated three weeks after the last cocaine injection, a small but significant reduction in probability of neurotransmitter release was observed. This finding might suggest altered plasticity on the presynaptic level on DA–glutamate co-releasing populations, however, through the approach utilized, it cannot be excluded that the observed synapses were solely glutamatergic [85].

As described above, synaptic glutamate release in the striatum exerts different actions in a regionand neuron-type-dependent manner. DA–glutamate co-release has been demonstrated to not show the same distribution pattern throughout the striatum [57,86] and glutamate co-released into the NAc mSh affects ChIs more than MSNs or fast-spiking interneurons [58]. ChIs only constitute about 1–2% of striatal neurons in rodents, but they exert diffuse and profound effects on the physiology of the area [87,88]. ChIs display distinct burst-pause firing patterns and through ACh release, they can modulate presynaptic neurons by acting both at nicotinic and muscarinic ACh receptors [87,89]. Burst-pause firing rates in ChIs coincide with changes in the firing rate of DA neurons in response to reward-related, salient events but they convey different information compared to DA neuron firing [90];

pauses in their synchronous activity is thought to provide an optimal window where DA release due to increases in action potential frequency will have highest efficiency to promote the conveyed messages [11]. The psychostimulant amphetamine which increases DA release in terminal regions, has different effects depending on the region of the striatum and the neuronal type investigated. Thus, it attenuates the "burst-pause" neuronal activity observed on ChIs in the mSh in a dose-dependent manner, as shown by Chuhma et al. [58]. Only high doses affect post-synaptic activity on ChIs in the dorsal striatum while they have little effects on NAc core and on the EPSCs of MSNs in NAc mSh [58]. While the effects of chronic administration of drugs were not investigated in the study by Chuhma and colleagues, these observations indicate that psychostimulants can induce plasticity in different ways depending on anatomical region that in some instances can lead to behavioral abnormalities associated with chronic drug intake.

While it is known that SNc DA neurons project to the dorsal striatum, it has also been well established that VTA DA projections to striatal areas follow a medio-lateral topographical organization in the sense that DA neurons of the medial aspect of the VTA project to the NAc mSh and DA neurons located in the lateral aspect of the VTA project to NAc core [91]. DA neurons in the medial aspect of the VTA, which project mainly to the NAc mSh, are the ones that show the highest VGLUT2 expression levels, as described above. These histological findings have recently been confirmed using novel approaches implementing dual intersectional systems combining promoter activity of both TH and VGLUT2 to achieve selectivity for DA–glutamate co-releasing neurons [44,92]. Using these new approaches, previous results could be confirmed, firmly demonstrating that TH+/VGLUT2+ double-positive cell bodies are located primarily in the medial VTA. Further, identification of NAc mSh projections showed that this region receives the highest percentage of DA–glutamate co-release compared to other parts of the striatum [44,92]. As discussed above, the NAc mSh is a region highly implicated in reward-related behavior and drug addiction.

Based on novel findings that glutamate released by DA neurons mostly affect ChIs in this area, physiological and behavioral effects of DA-glutamate co-release will be discussed in this context. As outlined in a recent comprehensive review by Mingote and colleagues [44], glutamate released from DA neurons modulates the activity of ChIs, initially by increasing their firing rate [58]. This increase is correlated to the release of ACh which acts through nicotinic pre-synaptic receptors to affect DA terminals and further increases DA levels [87,89]. Mingote and colleagues suggest that these subsequent DA increases are necessary to induce behavioral switching, meaning that the organism will be able to engage in alternative behavioral patterns not associated to reward occurrence [44] (Figure 9). This hypothesis is supported by experimental studies on extinction and behavioral observations on *DATIREScre*/+*::GLS1lox*/+ mice. As discussed above, behavioral studies of these mice showed that they exhibit potentiated latent inhibition compared to control mice [64]. Under normal conditions, control mice would switch their behavior when exposed to conflicting contingencies, but reduced glutamate from DA neurons in the NAc mSh in *DATIREScre*/+*::GLS1lox*/+ mice induced a potentiation of the preserved response resulting in enhanced latent inhibition [44,64]. Data from our *Vglut2f*/*f;DAT-Cre* cKO mice also support this model: *Vglut2f*/*f;DAT-Cre* cKO mice will continue operant responding in response to cocaine [61] or optogenetic stimulation cues (Figure 7) during extinction. It is possible that reduction of glutamate release from DA–glutamate co-releasing neurons projecting to NAc mSh prevents disinhibition of downstream circuits which under normal conditions would promote alternative strategies and eventually lead to more efficient reward obtaining behaviors. This is demonstrated through persevered behaviors that can be related to cue-induced relapse in human addicts. For example, pre-clinical studies have demonstrated higher levels of VGLUT2 in terminals of alcohol-preferring rats after alcohol deprivation [93], and severe alcohol use disorder has been associated with polymorphisms in VGLUT2 [94]. While not making a distinction between VGLUT2 in glutamatergic neurons versus DA–glutamate co-releasing neurons in these studies, it is interesting and provoking to consider that alterations in VGLUT2 levels can affect both types of glutamatergic transmission in NAc mSh. Finally, it is possible that disturbed DA–glutamate co-release is

responsible for drug-related behavioral manifestations due to overall altered DA tone and consequently to disturbances in behavioral switching.

**Figure 9.** Simplified schematic model of down-stream neurocircuitry effects upon dopamine (DA)–glutamate co-release in the ventral striatal area leading to behavioral output. DA–glutamate co-releasing neurons (red) are located primarily in the medial part of the ventral tegmental area (VTA; gray) from where they project to the nucleus accumbens medial shell (NAc mShell). Burst-firing of these neurons leads to a release of DA and glutamate from mesoaccumbal nerve terminals which subsequently act on DA and glutamate receptors located on cholinergic interneurons (ChIs) and medium spiny neurons (MSNs) in the NAc mShell. DA and glutamate neurotransmission via receptors located on ChIs leads to synchronized activity and acetylcholine (ACh) release. ACh subsequently acts on ACh receptors located in VTA presynaptic terminals to further increase neurotransmitter release. DA and glutamate release, together with increased release of GABA from GABAergic interneurons, also leads to inhibition of the GABAergic MSNs, which in turn leads to disinhibiton of target areas and allows for the occurence of behaviors associated with alternative strategies to obtain reward, a "behavioral switch". Drawing based on original illustration by Mingote et al., 2019 [44].

#### *2.10. Whole-Brain Analysis and Improved Selectivity in Animal Models Should Enhance Current Knowledge of DA-Glutamate Co-Releasing Neurons*

While most of the published studies regarding DA–glutamate co-release have focused on the projections from the VTA to the NAc, it has also been reported that co-releasing neurons of the VTA project to a broader set of limbic regions in the rodent brain. The medial prefrontal cortex (mPFC), amygdala and hippocampus have all been shown to receive DA–glutamate co-releasing fibers from the VTA with different synaptic strengths [39,47,86,95–97]. As these brain areas have been strongly implicated in reward processing and drug-induced behavioral adaptations, future studies could aim to investigate the role of DA–glutamate co-release in each of these areas both under normal conditions and in models of disorders, such as addiction and schizophrenia. Furthermore, and consistent with the regional heterogeneity observed within the striatum, neurons of dual DA–glutamate profiles might have different effects on neurocircuitry depending on the target area. For instance, DA neurons projecting to the mPFC have been shown to code for aversive events [98,99] while DA release in the mPFC has the potential to alter dopaminergic and behavioral responses of subcortical areas to both natural stimuli [100] and addictive drugs [101]. For example, one question remaining to be addressed is how mesocortical DA–glutamate co-releasing neurons might influence subcortical responses to salient stimuli. In summary, by opening up for studies covering a brain perspective beyond the striatum, it seems likely that the knowledge of how DA–glutamate co-release affects physiology and behavior of relevance to both health and disease could be increased further.

Another point worth considering to forward current knowledge is the availability of animal models. It has been discussed in the literature how the lack of appropriate animal models has made it challenging to fully address behavioral roles of DA–glutamate co-release [19,22]. While this is indeed true, as outlined above, by implementing Cre-Lox-transgenics in rodents, the results of several different studies converge towards a strong implication for a role of DA–glutamate co-release in reward processing of relevance to addiction [54,61–64]. To now reach further, a higher level of selectivity is on the wish-list. For example, the recently described intersectional approaches ([44,92] and described above) should prove useful for further investigation of DA–glutamate co-release. To advance selectivity when creating new animal models, a first step forward might be to increase the level of anatomical and molecular knowledge of neurons that possess the ability for co-release.

By enhancing current knowledge of moleculary defined subpopulations in the VTA, it may be possible to increase the level of resolution in the anatomical-functional mapping of DA-glutamate co-releasing neurons. In this context, we have recently demonstrated that some, but not all, DA–glutamate co-releasing neurons in the VTA express the NeuroD6 gene, a finding which opens up for subtyping DA–glutamate co-releasing neurons based on molecular profiles beyond TH, DAT and VGLUT2 [102]. Since our identification of VGLUT2 expression in subsets of DA neurons from early embryonal development ([54] and discussed above), we have viewed the VGLUT2-positive DA neuronal population as a distinct subpopulation within the VTA, in accordance with current literature. Based on our initial study [54], we reasoned that the VTA might contain additional subpopulations that, if they could be identified by a molecular profile, could be used to dissect out the causality between distinct neuronal activity in the VTA and behavioral regulation of importance to reward and addiction. To search for additional gene expression patterns that, beyond VGLUT2, might distinguish DA neurons from each other, we performed a microarray analysis followed through with systematic histological validation. Through several steps of analyses, we could identify a number of gene expression patterns that were selective for neuronal groups within the VTA, suggesting that they might represent molecularly definable VTA subpopulations [23]. Similar types of studies have been performed in the pre-clinical DA field focused on Parkinson's disease, which have substantially enriched molecular knowledge of both the VTA and the adjacently located SNc [103,104]. Recently, the advancement of transcriptomics analyses has enabled further gene expression analysis of the VTA and SNc [105,106]. All studies mapping out gene expression patterns in a particular brain location are of

particular interest as they provide molecular tools that enable the anatomical-functional dissection of "subpopulations" (or "subgroups" or "subtypes") of neurons.

In our microarray screen comparing gene expression in the VTA and the SNc, we identified the gene encoding the neurogenic basic helix-loop-helix transcription factor NeuroD6 as enriched in the VTA [23]. NeuroD6, which has also been reported by others [105,107,108], was found primarily in DA neurons located in the medial aspect of the VTA, suggesting that NeuroD6 expression represents a distinct subpopulation within the group of medially positioned VTA DA neurons that primarily project to the NAc Sh [23]. We also found that some, but not all, NeuroD6 VTA DA neurons were positive for VGLUT2 (Figure 10). This finding leads us to propose that the DA–glutamate co-releasing phenotype can be dissociated further based on molecular identity. While anatomically interesting, the most important aspect of molecular profiling is when it can be coupled to functional output. To directly test if we could identify a distinct role in behavioral regulation mediated by the newly discovered NeuroD6 VTA subpopulation, we implemented optogenetics using transgenic NeuroD6-Cre mice (also known as NEX-Cre) to direct optogenetic activation selectively to this NeuroD6-positive subpopulation. NeuroD6-Cre mice were compared with DAT-Cre and VGLUT2-Cre mice in a range of optogenetics-based analyses [102]. The experiments provided evidence for both glutamate and DA release in the NAcSh upon optogenetic stimulation in the VTA and could also demonstrate that selective stimulation of NeuroD6 VTA neurons led to significant place preference in a similar, but not identical, manner as when the entire VTA DA neuronal population was activated [102] (Figure 11). Activation of the NeuroD6 VTA subpopulation, with its mixed DA–glutamate co-releasing and non-co-releasing properties, is thereby sufficient to induce a distinct behavioral response [102]. This new finding should be well-worth exploring further in the context of substance use/abuse. This study is an example of how molecular knowledge of distinct VTA neurons can be used to implement available animal models in a new way or to even create new animal models with spatial and temporal selectivity for distinct neuronal subgroups at a level required to advance current knowledge of co-releasing neurons. By establishing direct causality between distinct subgroups of DA–glutamate co-releasing neurons and behavioral regulation of importance to reward and addiction, future studies can use this kind of knowledge to understand how regulation of co-release could be implemented clinically for the benefit of future interventive strategies in substance use disorders.

#### *2.11. Concluding Remarks*

This review has summarized the main findings derived from behavioral analyses in genetically modified mice produced with the aim of pin-pointing the putative relevance of DA–glutamate co-release in behavioral regulation. While different kinds of transgenic mouse lines have been generated and analyzed in various methodological paradigms spanning from baseline locomotion to self-administration of abusive substances, all studies share the common conclusion that behaviors of particular interest for addiction are altered when glutamate co-release is disrupted *(*results summarized in Table 1 and illustrated in Figure 12). Many questions remain to be answered, but so far, the experimental findings point towards glutamate co-release as a putative target in the treatment of substance use disorder, including prevention of relapse. The study of DA–glutamate co-release for clinical purposes would benefit from a higher degree of selectivity in experimental approaches which could aid in determining how this type of signaling could be used as a tool in prevention and treatment of addiction.

**Figure 10.** NeuroD6 mRNA is found in a modest population of the VTA and co-localizes with dopaminergic markers and partially with a glutamatergic marker. (**A**–**G**) Double FISH in the ventral midbrain of adult wild-type mice detecting the following mRNAs. **A**, **A'**, NeuroD6 (red). (**B**,**B'**) Th (green). (**C**,**C'**) NeuroD6 (red) and Th (green). Th/NeuroD6 mRNA that overlap are shown in yellow.

Low magnification to the left; close-ups to the right. Schematic outline shows borders for SNc and subregions of VTA: PN, PIF, PBP, IF, RLi. (**D**) Quantification of percentage of NeuroD6-positive cells among all Th VTA cells; all NeuroD6 cells are positive for Th mRNA. (**E**) NeuroD6 (red) and Dat (green), inset with high magnification of Dat/NeuroD6 mRNA overlap (yellow). (**F**) NeuroD6 (red) and Vglut2 (green). (**G**) NeuroD6 (red) and Viaat (green), inset with high magnification of Viaat-negative/NeuroD6-positive (red). (**H**–**P**) Triple-labeling FISH in the ventral midbrain of adult wild-type mice detecting: (**H**) Th (blue); (**I**) NeuroD6 (red); (**J**) Vglut2 (green) mRNAs and their co-localization: (**K**) NeuroD6/Th; (**L**) NeuroD6/Vglut2; (**M**) Th/NeuroD6/Vglut2. Cellular closeups: (**N**) NeuroD6/Th (top), NeuroD6/Vglut2 (middle), and Th/NeuroD6/Vglut2 (bottom). Arrows point to NeuroD6 mRNA-positive cells. (**O**) Quantification of percentage of NeuroD6+/Th+/Vglut2+ and NeuroD6+/Th+/Vglut2- neurons of the VTA. (**P**) Schematic illustration of distribution pattern of NeuroD6+/Th+/Vglut2+ and NeuroD6+/Th+/Vglut2- neurons within the VTA (same as shown with experimental data in **M**). NeuroD6+/Th+/Vglut2- cells in magenta; NeuroD6+/Th+/Vglut2+ cells in cyan. VTA, ventral tegmental area; SNc, substantia nigra pars compacta; PBP, parabrachial pigmented nucleus; PN, paranigral nucleus; PIF, parainterfascicular nucleus; RLi, rostral linear nucleus; IF, interfascicular nucleus. FISH, fluorescent in situ; Dat, Dopamine transporter; Th, Tyrosine hydroxylase; Vglut2, Vesicular glutamate transporter 2; Viaat, Vesicular inhibitory amino acid transporter. Reprinted from Bimpisidis et al., 2019 [102].

**Figure 11.** Optogenetic activation of NeuroD6 VTA neurons induces place preference. (**A**) Schematic drawing of the real-time place preference (RT-PP) experimental setup. (**B**–**D**) average percentage of time spent in each compartment during 4 days of RT-PP ± SEM (bar graphs; right; \**p* < 0.05, \*\*\**p* < 0.001 vs. light-paired compartment; #*p* < 0.05, ##*p* < 0.01, ###*p* < 0.001 vs. unpaired compartment). DAT-Cre, *n* = 10; NEX-Cre, *n* = 5, high-power stimulation of bilaterally-injected NEX-Cre mice, *n* = 4. Reprinted from Bimpisidis et al., 2019 [102].

*J. Clin. Med.* **2019**, *8*, 1887

**Figure 12.** Schematic illustration of behaviors relevant to addiction confirmed in experimental mice to be altered upon disruption of dopamine–glutamate co-release (see text and Table 1). Further studies will be needed to outline how these behaviors relate to typical behaviors of addiction displayed by humans and how dopamine–glutamate co-release can be used as a future target for prevention and treatment.

**Author Contributions:** Conceptualization and funding acquisition, Å.W.-M.; Writing—original draft, review and editing, Z.B. and Å.W.-M.

**Funding:** This work was funded by Uppsala University and by grants to ÅWM from the Swedish Research Council (Vetenskapsrådet), Parkinsonfonden, Hjärnfonden and the Research Foundations of Bertil Hållsten, Åhlén and Zoologisk Forskning.

**Acknowledgments:** Present and previous colleagues in the Mackenzie lab as well as collaborators and colleagues over the world are thanked for contributions to the studies discussed in the text. While we have tried to include all relevant references, we apologize if any citation is wrong or missing. Figures from our own previous work have been reprinted with permission from the publishers.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


#### *J. Clin. Med.* **2019**, *8*, 1887


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Addiction as Learned Behavior Patterns**

**Andreas Heinz 1, Anne Beck 1, Melissa Gül Halil 1, Maximilian Pilhatsch 2, Michael N. Smolka 2,3 and Shuyan Liu 1,\***


Received: 24 June 2019; Accepted: 19 July 2019; Published: 24 July 2019

**Abstract:** Individuals with substance use disorders (SUDs) have to cope with drug-related cues and contexts which can affect instrumental drug seeking, as shown with Pavlovian-to-instrumental transfer (PIT) tasks among humans and animals. Our review addresses two potential mechanisms that may contribute to habitual or even compulsive drug seeking and taking. One mechanism is represented by Pavlovian and PIT effects on drug intake. The other is a shift from goal-directed to habitual drug intake, which can be accessed via model-based versus model-free decision-making in respective learning tasks. We discuss the impact of these learning mechanisms on drug consumption. First, we describe how Pavlovian and instrumental learning mechanisms interact in drug addiction. Secondly, we address the effects of acute and chronic stress exposure on behavioral and neural PIT effects in alcohol use disorder (AUD). Thirdly, we discuss how these learning mechanisms and their respective neurobiological correlates can contribute to losing versus regaining control over drug intake. Utilizing mobile technology (mobile applications on smartphones including games that measure learning mechanisms, activity bracelets), computational models, and real-world data may help to better identify patients with a high relapse risk and to offer targeted behavioral and pharmacotherapeutic interventions for vulnerable patients.

**Keywords:** substance use disorders; alternative reward; cue exposure; animal and computational models; behavioral control; craving and relapse; habit formation

#### **1. Introduction**

Drugs of abuse stimulate dopamine release and thus reinforce drug intake [1]. Wise originally suggested that dopamine release is tied to pleasure and hedonic changes that strongly reinforce the behavior of repetitive drug use [2]. Robinson and Berridge later suggested that dopamine release is more associated with reward motivation rather than mediating hedonic pleasure, contributing to "wanting" or "craving" instead of "liking" drugs of abuse [3]. This hypothesis was based on studies by Schulz and co-workers [4]. They found that phasic dopamine release is modulated by an unexpected reward and a conditioned stimulus, which in turn reliably predict reward. They suggested dopamine signals code reward prediction errors (i.e., the difference between received and predicted rewards) which drive reward-motivated behaviors. Accordingly, dopamine D2-receptor blockade in humans was associated with motivational deficits, but not anhedonia [5]. Based on the observations above, dopamine is not only associated with the encoding of unexpected rewards, but also the attribution of incentive salience to reward-related cues [3]. Further research is required to better understand how such learning mechanisms may shed light on drug seeking and intake. Specifically, recreational drug use elicits a rather strong dopamine release, thus reinforcing drug consumption [6]. Habitual drug use is characterized by a shift from ventral to dorsal striatal processing, including the dopaminergic

modulation in fronto-striatal brain circuitries [6]. Ultimately, drug consumption was independent of rewarding or aversive outcomes [6,7]. The pathways from the orbitofrontal cortex to the dorsal striatum play a key role in compulsive drug use, in spite of aversive consequences [8]. Obsessions and compulsions in obsessive compulsive disorder (OCD) differ from drug craving and intake [9]. However, drug addiction is characterized by compulsive drug intake and has substantial similarities with other disorders of compulsions, including OCD, on phenomenological and neurobiological levels [10,11]. In this review, we discuss two potential mechanisms that may contribute to habitual drug intake and, ultimately, drug seeking and taking. One mechanism is the stimulus response associations as represented by Pavlovian effects on drug intake and the other is a shift from goal-directed to habitual drug intake, which can be accessed via model-based versus model-free decision-making in respective learning tasks [11,12].

#### **2. Pavlovian Mechanisms in Addictive Behavior**

Drug-associated cues can elicit drug craving and promote drug seeking [3,13]. From a theoretical point of view, Pavlovian unconditioned cues, such as food, elicit unconditioned responses, including increased salivation and food craving. Conditioned cues, such as pictures of alcoholic beverages, may elicit drug craving as a conditioned response [14]. However, most drugs of abuse do not often come accidentally to an addicted person. Instead, patients with drug dependence actively search for available drugs. One of our patients described the situation with the following words: "When the evening comes and the sky turns grey, I pass by these bars with their warm yellow light and hear the clinging of glasses. I'm lost." In this context, conditioned cues include the clinging of glasses, certain colors of light in a bar, and the kind of loneliness while looking at the dark gloomy sky. These conditioned cues have been previously paired with positive/pleasant activities/evenings. Such conditioned contextual cues elicit drug craving and have an impact on goal-directed behavior; the afflicted person changes his or her direction, enters the bar, orders a drink, and consumes it. The implicated mechanism has been called Pavlovian-to-instrumental transfer (PIT) [15]. During PIT, a Pavlovian conditioned cue (e.g., the clinging of glasses) can have an impact on a series of obviously unrelated approach behavioral sequences, including entering a certain place, talking to bartenders, and ordering a drink. Regarding cue reactivity, imaging studies show that functional activation elicited by drug-associated cues, particularly in the medial prefrontal cortex, was correlated with a high risk of relapse for detoxified patients with alcohol use disorder (AUD) [16,17]. Moreover, naltrexone, which blocks μ-opioid receptors that have been reported to be elevated in AUD, also reduces cue-induced functional activation in the ventral striatum in AUD patients [18,19]. Another neurotransmitter system implicated in cue-induced brain activation in addictive disorders is the dopamine system. A low availability of dopamine D2-receptors in the ventral striatum is associated with increased functional activation elicited by alcohol cues in the medial prefrontal cortex [20]. Low dopamine D2-receptor availability following detoxification may represent a counter-regulatory new adaptation following excessive dopamine release due to the consumption of drugs of abuse and delayed recovery of dopamine D2-receptor sensitivity following detoxification was associated with poor treatment outcomes [21].

So how can alcohol cues trigger not only drug craving and functional activation in the ventral striatum, amygdala, and medial prefrontal cortex [17,22], but also bias complex goal-directed behavior toward drug seeking and intake? A subclass of environmental cues is called Pavlovian conditioned stimuli due to the ability to elicit a conditioned response, which is usually inborn (such as the production of saliva in a hungry dog or avoidance of malodors) and hence hard-wired in the central nervous system [23]. As suggested above, such Pavlovian conditioned stimuli can also impact ongoing instrumental behavior, even if the instrumental behavior was acquired independently of Pavlovian conditioning, a process called Pavlovian-to-instrumental transfer (PIT) [24]. In PIT, positively valued Pavlovian cues promote instrumental responses and approach behaviors (e.g., enhance the frequency of pressing a button) [24], while negatively valued Pavlovian cues promote inhibition or withdrawal actions (e.g., lower the frequency of pressing a button for instrumental approach or enhance the

frequency of pressing a button for instrumental withdrawal [25] (Figure 1)). Thus, in drug addiction, Pavlovian conditioned cues can bias instrumental behavior toward drug seeking and intake [26–28].

**Figure 1.** The Pavlovian-to-instrumental (PIT) effect. (**A**) The unrelated Pavlovian stimulus (conditioned stimulus (CS)) presented in the background is negatively valued because it has previously been paired with passive monetary loss. (**B**) The PIT effect is indicated by the number of button presses (instrumental response) as a function of the value of the respective Pavlovian background stimulus (−€2, −€1, €0, +€1, +€2). (**C**) Combining the shell with a positive Pavlovian cue in the background of the screen increases approach behavior (number of button presses) in the unrelated instrumental task. (**D**) The PIT effect was significantly stronger in subjects with alcohol use disorder (AUD) compared to healthy controls (HC).

In outcome-specific PIT, presenting a particular reward-predicting cue can selectively elevate instrumental responses that are associated with the same unique reward, while in general PIT, a rewardor loss-predicting cue can generally modify instrumental responses toward any outcome [24]. So-called single-lever PIT tasks (see Figure 1) usually reflect general PIT, while a full transfer task enables the disentanglement between general and outcome-specific PIT [24]. Like habits, PIT effects may help to prune a complex "decision tree" by biasing an individual to instrumental approaches or withdrawal behaviors in the presence of certain background stimuli [29]. Indeed, a general tendency to rely on habitual rather than complex goal-directed decision-making was associated with increased PIT effects in healthy volunteers [30]. Moreover, we observed PIT effects being modulated by personality traits, such as impulsive decision making, with the strongest PIT effects observed in high impulsive alcohol-dependent patients compared to low impulsive patients [31].

PIT effects may be specifically strong in stressful situations, when decisions have to be fast, and profit from an overall "atmospheric" evaluation of the dangerousness or safety of the current situation [32]. Various forms of stress promote substance use and relapse, as evidenced by a broad range of literature [33,34]. In this context, Quail and co-workers suggested that stress exposure modifies the influence of Pavlovian cues on behavior [35]. They observed that subjects reporting high stress were impaired to suppress instrumental responding under no-reward Pavlovian cues [35]. Moreover, acute stress selectively increased cue-triggered wanting independently of hedonic properties of the reward [36]. Stress exposure and long-term endocrine stress measures (e.g., hair cortisol) in addicts have so far not been studied with respect to PIT and its association with losing versus regaining control over drug intake. Moreover, we did not find gender and age effects [31,37,38], which would require further research.

With respect to neurobiological correlates, animal experiments and human studies suggest that activation of the basolateral amygdala, the nucleus accumbens shell, and the ventrolateral putamen contribute to an outcome-specific form of PIT [15,39,40]. The central nucleus of the amygdala and the nucleus accumbens core are involved in the general form of PIT [15,39,40]. These neurobiological differences are in line with a goal-directed aspect of specific PIT compared to an arousing effect of general PIT. In the outcome-specific form of PIT, the Pavlovian cue has been conditioned with the same rewarding outcome that can also be gained when performing the instrumental response. For example, the smell of wine promotes ordering and consuming a glass of wine instead of lemonade. In the general form of PIT, the Pavlovian cue has been conditioned to a positive outcome that is not associated with

the outcome available by the instrumental action. For example, upbeat music played in a shopping mall motivates customers to spent more money. Thus, general PIT appears to promote instrumental actions by modulating arousal, while outcome-specific PIT may facilitate the retrieval of particular actions based on their outcomes [26].

In line with this, stronger general PIT effects elicited by positive non-drug cues and functional PIT-related brain activation in the nucleus accumbens were observed in prospective AUD relapsers [37,41]. This phenomenon of increased PIT effects was also observed in studies when animals were pretreated with drugs of abuse [24].

In smokers, tobacco-related PIT effects have been demonstrated in several studies in satiated and deprived smokers [42,43], but contrary to our findings in AUD patients, studies in smokers did not see stronger PIT effects in more dependent subjects or compared to non-dependent controls. In cocaine addicts, cocaine-paired cues can provoke the pursuit of cocaine through a Pavlovian motivational process [27]. In general, there are a limited numbers of studies examining whether different types of drug abuse, such as opioids and amphetamine, can support PIT [24]. Establishing these effects may deepen our understanding of the behavioral and neural processes underlying cue-motivated drug-seeking behavior.

The PIT effects of drug-related cues were also studied in subjects with AUD. Regarding alcohol versus water cues, we expected that alcohol cues would promote approach behaviors and predict poor treatment outcomes, as was the case with general PIT effects. The appetitive and aversive Pavlovian cues were passively conditioned with monetary reward or loss. Surprisingly, however, patients with poor treatment outcomes behaved similar to the healthy controls. Patients with good treatment outcomes who did not relapse in the follow-up period of three months showed a significant difference both in behavior and in functional brain responses to alcohol cues in a general PIT task [38]. They showed both an increased functional activation of the ventral striatum when confronted with these Pavlovian-conditioned alcohol cues, as well as an inhibition of approached behavior and increased withdrawal behavior in the presence of such alcohol cues [38]. Interestingly, alcohol-dependent patients with good treatment outcomes appeared to learn a specific inhibitory reaction to alcohol cues. At least, they significantly differed both from healthy controls and patients who later relapsed during the follow-up period. Increased activation of the ventral striatum may be due to salience attribution to alcohol cues, which apparently did not simply trigger approach behaviors, but instead enabled subjects to inhibit unrelated goal-directed behaviors. Thus, patients with good treatment outcomes could use alcohol cues as warning signs and—unlike the patient in the example explained above—resist drug-approach tendencies. For example, they may not enter the bar with the warm yellow light or avoid going to the supermarket where they used to buy their alcoholic beverages.

Patients may learn to use environmental cues as warning-signs and thus train to avoid rather than approach situations in which drugs are available. One training program targeting such drug-approach tendencies is the so-called Zooming Joystick Task. Patients with addictive disorders learn to push pictures of alcohol beverages away instead of pulling them toward themselves. Four training sessions appear to be sufficient to successfully reduce the relapse-risk during an one year follow-up period, with the number needed to treat (NNT) being around 10, suggesting that 10% of all patients would benefit from this intervention [44]. From a neurobiological perspective, such alcohol cues activate the medial prefrontal cortex and further brain areas, including the amygdala, implicated in PIT mechanisms; successfully learning to push alcohol cues away was associated with reduced amygdala activation in AUD patients [45,46]. The success of such training programs encourages studies to better understand the neurobiological correlates and to identify patients who may respond particularly well to such training programs.

In line with the key role of the amygdala and nucleus accumbens, behavioral PIT effects are understood as driven by bottom-up processes. Nevertheless, a conflict—like in a Stroop task—should be elicited in situations in which Pavlovian and instrumental cues are incongruent (i.e., collecting "good" shells when negatively valued context stimuli are shown, or leaving "bad" shells during presentation of positively valued contexts) and this conflict should trigger the allocation of top-down control. Indeed, the results of Sommer and co-workers [31] revealed that instrumental behavior during PIT is more error-prone when instrumental and Pavlovian cues are incongruent, in line with the assumption of such a conflict between Pavlovian and instrumental control (Figure 2). Importantly, the incongruence effect was more pronounced in AUD subjects than in controls, indicating that reduced interference control may impair goal-directed behavior, especially in AUD subjects.

**Figure 2.** Conflict between Pavlovian and instrumental control: Subjects with alcohol use disorder (AUD) compared to controls. ER = error rates.

#### **3. From Goal-Directed to Habitual Drug Seeking—The Importance of Contextual Cues**

Dual-process theories of learning and addiction propose that the development of drug addiction involves a shift from goal-directed to habitual control of action [6,7]. Animal models of drug addiction suggest that occasional drug use becomes habitual and ultimately compulsive (i.e., it is maintained in spite of aversive consequences) [7]. In humans, complex model-based behavior is reduced in patients with different substance use disorders (SUDs) as well as with OCD [11]. This may help to explain why aversive outcomes associated with drug consumption do not affect the respective behavior and enforce modification. Regarding patients with AUD, the results of the recent studies were inconsistent. In a study by Voon and co-workers [11], a shift was not observed from model-based toward model-free behavior in AUD patients, while such a shift was observed by Sebold and co-workers [47]. However, Sebold and co-workers did not replicate their previous findings in a larger independent sample [48]. There was no overall reduction in model-based behavior in patients with AUDs and in patients with poor treatment outcomes compared to patients with good treatment outcomes [48].

Model-based versus model-free behavior and goal-directed versus habitual behavior are assessed by different tasks. Model-based versus model-free behavior is assessed via taking complex decision-making processes into account, while goal-directed versus habitual behavior is operationalized via the impact of reward devaluation. Nevertheless, both tasks are intercorrelated in the sense that individuals who tend to behave in a model-based way also show stronger goal-directed behaviors, while individuals who tend to respond in a habitual way rely more strongly on model-free decision-making [49]. Therefore, failure to observe effects of a reduction in model-based behavior in AUD patients may challenge the assumption that these patients have a general tendency for habit-formation at the expense of goal-directed decision-making. However, Sebold and co-workers also observed that model-based versus model-free behavior can predict treatment outcomes when taking alcohol expectancy into account [48]. Patients with high alcohol expectancies showing low model-based behavior, thus shifting the balance toward model-free behavior, had poor treatment outcomes [48]. These findings suggest that shifts from goal-directed to habitual decision-making depend on contextual stimuli. It may be specifically relevant for a subset of behavior patterns

associated with drug seeking and drug consumption. Instead of searching for general tendencies to form habits, specific context-dependent learning mechanisms that may interfere with cognitive control and conscious decisions to remain abstinent must be identified. Cognitive abilities such as working memory have been discovered to interplay between these two behavioral systems [50,51]. Acute [52] or chronic [53] stress are thought to impair executive resources underlying working memory and were found to impair goal-directed decision-making, inducing a relative shift toward habitual behavioral control. Stress is also an important factor in the development and maintenance of AUD and has been shown to increase alcohol intake [54–56]. Human imaging studies revealed that acute stress enhanced stimulus–response learning, which was accompanied by increased amygdala activity during a spatial learning task [57], as well as biased choices for immediately rewarding food stimuli and increased functional connectivity between the ventromedial prefrontal cortex and amygdala and striatal regions encoding tastiness [58]. Therefore, the acute stress experience might promote loss of control over alcohol intake by diminishing goal-directed responses and promoting habitual actions, thus undermining the goal to stay abstinent by promoting habitual substance intake. We also observed that goal-directed decision-making was affected by increased life stressors [59], underlining the strong potential of interventions aimed at altering stress-related effects on losing and regaining control over substance use. In future studies, researchers could model learning and cognitive control systems in interaction with real-life monitoring of stressors, cue responsivity, and ecological momentary assessment of alcohol consumption.

#### **4. Summary and Outlook**

Human behavior is more flexible and dependent on context than previously assumed in straight-forward models (i.e., increased PIT effects and habitual decision-making in drug addiction). Researchers should consider contextual cues, such as expectancies and availabilities, mood states, individual stress-levels, and cognitive control processes. Modern technology allows ambulatory assessments, including reports of mood-states, recordings of geolocation, and psychomotor activity in real life [60]. An important future focus should be on the development and establishment of computational models for learning and decision-making in humans. To date, cue exposure in general has limited effects and individual differences in cue effects, including ambulatory assessments of learning mechanisms like PIT, may help to target those patients [61,62]. Thus, utilizing a model's predictions and real-world data may help to better identify patients with a high relapse risk and to offer specific behavioral or pharmacological interventions for vulnerable patients.

**Author Contributions:** Conceptualization, A.H.; Writing—original draft, A.H., S.L.; Writing—review and editing, A.H., A.B., M.G.H., M.P., M.N.S., and S.L.

**Funding:** This research was funded by the German Research Foundation (grant number SFB/TRR 265).

**Acknowledgments:** The authors would like to thank the German Research Foundation for grant.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **The Potential of Cannabidiol as a Treatment for Psychosis and Addiction: Who Benefits Most? A Systematic Review**

#### **Albert Batalla 1,\*,**†**, Hella Janssen 1,**†**, Shiral S. Gangadin 1,2 and Matthijs G. Bossong <sup>1</sup>**


Received: 24 June 2019; Accepted: 18 July 2019; Published: 19 July 2019

**Abstract:** The endogenous cannabinoid (eCB) system plays an important role in the pathophysiology of both psychotic disorders and substance use disorders (SUDs). The non-psychoactive cannabinoid compound, cannabidiol (CBD) is a highly promising tool in the treatment of both disorders. Here we review human clinical studies that investigated the efficacy of CBD treatment for schizophrenia, substance use disorders, and their comorbidity. In particular, we examined possible profiles of patients who may benefit the most from CBD treatment. CBD, either as monotherapy or added to regular antipsychotic medication, improved symptoms in patients with schizophrenia, with particularly promising effects in the early stages of illness. A potential biomarker is the level of anandamide in blood. CBD and THC mixtures showed positive effects in reducing short-term withdrawal and craving in cannabis use disorders. Studies on schizophrenia and comorbid substance use are lacking. Future studies should focus on the effects of CBD on psychotic disorders in different stages of illness, together with the effects on comorbid substance use. These studies should use standardized measures to assess cannabis use. In addition, future efforts should be taken to study the relationship between the eCB system, GABA/glutamate, and the immune system to reveal the underlying neurobiology of the effects of CBD.

**Keywords:** cannabidiol; CBD; cannabis; psychosis; schizophrenia; substance use disorders; addiction

#### **1. Introduction**

Schizophrenia is a complex mental disorder, which has a profound impact on patients. The burden of schizophrenia is explained by the early onset, often in early adulthood or late adolescence, its chronic course, and its relatively high prevalence [1]. The symptomatology is highly heterogeneous and often overlaps with comorbid disorders, such as affective or substance use disorders [2,3]. Psychotic symptoms are grouped into three dimensions: Positive symptoms (e.g., delusions, hallucinations), negative symptoms (e.g., blunted affect, anhedonia), and cognitive symptoms (e.g., attention, memory, executive functioning; see for reviews [4–6]). Different combinations of symptoms and comorbidity lead to different clinical profiles and treatment needs. However, the pharmacological treatment of schizophrenia is mainly based on dopamine blockade, the effect of which is limited to the positive symptoms [7]. Moreover, two-thirds of the patients experience a suboptimal response with dopaminergic treatment [8], and these results are even worse when comorbid substance use disorders (SUDs) are present [9]. Therefore, there is an urgent need for alternative and more effective pharmacological interventions aimed to reduce the burden of complex and overlapping symptom profiles.

One of these interventions may involve the endocannabinoid (eCB) system, which is a promising new pharmacological target in this respect. The eCB system consists of at least two types of receptors and their endogenous ligands (i.e., endocannabinoids; [10,11]). The cannabinoid receptors are predominantly present in the central nervous system, in particular, in several limbic and cortical brain structures [12]. The eCB system is a retrograde messenger system that regulates both excitatory glutamate and inhibitory GABA neurotransmission according to an 'on-demand' principle: Endocannabinoids are released when and where they are needed [10,11,13]. This endocannabinoid-mediated regulation of synaptic transmission is a widespread phenomenon in the brain and is thought to play an important role in higher brain functions, such as cognition, motor function, and processing of sensory input, reward, and emotions [14–17]. eCB receptors are also present on immune cells in the central nervous system (i.e., microglia), which suggests their involvement in processes such as cytokine release, immune suppression, and induction of both cell migration and apoptosis [18,19].

The role of the eCB system in the pathophysiology of schizophrenia has been suggested in an accumulating amount of evidence [20,21]. First, epidemiological studies suggest that cannabis use increases the risk for developing schizophrenia [22] and lowers the age of onset of the disorder [23,24]. This risk increases with a higher frequency of cannabis use (e.g., daily use), and with the consumption of more potent cannabis (i.e., a higher amount of Δ9-tetrahydrocannabinol; THC) [22,25–27]. Second, modulation of the eCB system by the administration of THC (i.e., the main psychoactive component in cannabis) to healthy volunteers showed that THC can induce positive psychotic symptoms, effects that resemble negative symptoms (e.g., blunted affect, lack of spontaneity) and deficits in cognition (reviewed in [28]). Importantly, in schizophrenia patients, enhanced levels of endocannabinoids were demonstrated in cerebrospinal fluid and blood [29–31], and increased CB receptor density and availability were shown in the brain [32,33].

In addition to its role in schizophrenia, there is overwhelming evidence that the eCB system is implicated in the pathophysiology of addiction, in particular in processes such as drug-seeking behaviour, reward, withdrawal, and relapse (see for reviews [34–37]). For example, animal studies have shown that addictive properties reflected in behaviours such as self-administration or conditioned place preference of opiates, nicotine, and alcohol are absent or attenuated in cannabinoid CB1-receptor knockout mice and after administration of CB1 antagonists [35]. In addition, whereas the drug seeking behaviour of drugs of abuse was blocked with CB1 antagonists, it was reinstated after the administration of CB1 agonists [34,36]. Finally, endocannabinoid concentrations are affected by active drug seeking behaviour and eCB signalling seems to modulate the rewarding effects of addictive drugs [38].

SUDs and psychotic disorders such as schizophrenia co-occur frequently. Prevalence rates of any SUD (excluding nicotine and caffeine) in patients with schizophrenia are up to 45% [39,40], with the most frequently used substances being cannabis and alcohol. Considering nicotine use disorders, the prevalence rates rise up to 60%–90% [40]. Persistent use of licit or illicit drugs has been associated with adverse consequences in the overall course of psychotic disorders, and increased morbidity and mortality [40]. In addition, SUDs are also related to poor medication adherence, increasing the risk of relapse [39]. For example, in patients with schizophrenia, cannabis use has been related to higher relapse rates, increased severity of symptoms, and poor outcome [41–45]. Despite the high co-occurring rates, patients with comorbid SUDs and psychotic disorders are often excluded from clinical trials, which limits the generalization of results and ignores the potential (positive or negative) effects of the intervention on substance use.

While THC can trigger both schizophrenia and SUD and worsen the course of both disorders, the non-psychoactive cannabinoid compound cannabidiol (CBD) may have opposite or even beneficial effects. For example, CBD may have the ability to counteract psychotic symptoms and cognitive impairment associated with cannabis use as well as with acute THC administration [46,47]. In addition, CBD may lower the risk for developing psychosis that is related to cannabis use [48]. These effects are possibly mediated by the opposite effects of CBD and THC on brain activity patterns in key regions implicated in the pathophysiology of schizophrenia, such as the striatum, hippocampus, and prefrontal cortex [28]. Therefore, CBD displays a highly favourable profile for development as a new antipsychotic agent [48]. Similarly, CBD may serve as a treatment for SUDs, since evidence from preclinical studies suggests that CBD reduces negative withdrawal effects, motivation for self-administration, and reinstatement of drug use [37]. As a result, CBD-containing compounds are increasingly being investigated in the context of substance abuse in humans as well.

The eCB system appears an interesting target for schizophrenia, SUDs, and their comorbidity, due to the implication of the eCB system in their pathophysiology and the beneficial effects of CBD in both disorders. However, one may expect that CBD treatment may be most effective in a subgroup of patients, for example patients who show alterations in the eCB system or have a specific symptom profile. CBD may restore an imbalance in the eCB system, which may result in clinical improvement. Although previous excellent reviews (e.g., [37,48,49]) described the potential of CBD as a treatment for psychosis and SUD, this review provides a detailed and up-to-date systematic literature overview of clinical studies that investigated the efficacy of CBD treatment for schizophrenia and/or SUD. In addition, this review examined whether there are specific subgroup of patients with schizophrenia, SUD, or both that may benefit the most from CBD treatment.

#### **2. Experimental Section**

Clinical trials and case reports published up to February 2019, which described the effects of CBD on the symptomatology of psychotic disorders (i.e., schizophrenia and related disorders), SUD, or both were included. Reviews, non-English articles, pre-clinical or animal studies, studies that investigate CBD tolerability and pharmacokinetics or compare the acute effects of CBD with THC, and articles describing psychiatric or neurologic disorders other than psychotic disorders and SUD were excluded.

A literature search was conducted in the PubMed database. The following two searches were used: (1) "(((cannabidiol [MeSH Terms]) OR CBD[Text Word])) AND ((((((Substance-Related Disorders[MeSH Terms]) OR addiction[Text Word]) OR addictive behavior[Text Word]) OR drug abuse[Text Word])) OR drug dependence[Text Word])", (2) "(((((((Schizophrenia Spectrum and Other Psychotic Disorders[MeSH Terms])) OR schizophrenia[Text Word]) OR schizophrenic[Text Word]) OR psychosis[Text Word]) OR psychotic[Text Word])) AND ((cannabidiol[MeSH Terms]) OR CBD[Text Word])".

#### **3. Results**

The searches resulted in 214 articles, which included one duplicate (Figure 1). The articles were screened by two authors independently, according to the PRISMA guidelines [50]. After full-text screening, ten articles from the systematic search were included and six additional papers were selected through references in other papers. Of these 16 included articles, seven studies were related to CBD treatment for schizophrenia and eight studies described the treatment of SUD with CBD-containing compounds. Only one study assessed the effects of the treatment with medicinal cannabis for patients with a psychotic disorder and a comorbid cannabis use disorder.

**Figure 1.** Study inclusion process. CBD: Cannabidiol; THC: Δ9-Tetrahydrocannabinol.

#### *3.1. CBD—Psychosis*

Four randomized controlled trials (RCTs) and three case reports assessed the efficacy of CBD as a treatment for psychotic disorders (Table 1).

Zuardi et al. (1995) described a 19-year-old woman with schizophrenia who received progressive increase of CBD monotherapy for 26 days (maximum of 1500 mg/day) [51]. CBD treatment was associated with the improvement of symptomatology as measured with the Brief Psychiatric Rating Scale (BPRS). This improvement did not further increase on haloperidol treatment [51]. In a second case report of the same group, three treatment-resistant schizophrenia male patients were treated with CBD monotherapy for four weeks. The authors reported mild improvement of positive and negative symptoms of one patient after CBD treatment (BPRS score decreased from 29 to 22). Moreover, CBD was well tolerated and no side effects were reported [52]. Makiol and Klunge (2019 described a case of a 57-year-old woman with treatment-resistant schizophrenia, which persisted for 21 years [53]. On admission she had a total PANSS (Positive and Negative Syndrome Scale) score of 117 and a negative symptom score of 41. Adjunctive to treatment with clozapine (275 mg/day) and lamotrigine (225 mg/day), the patient received CBD 500 mg twice daily, which was increased to 750 mg twice daily after seven weeks. On discharge, the PANSS total score decreased to 68 and negative symptom score to 21, which the authors indicated as accomplishment of remission criteria with only mild negative symptoms. CBD did not affect clozapine levels and was well tolerated apart from a mild hand

tremor [53]. Leweke et al. (2012) performed a double-blind randomized controlled trial in which 39 acutely psychotic inpatients were treated with either CBD (*N* = 20) or amisulpride 800 mg (*N* = 19) for four weeks [54]. The authors did not provide information about illness duration before hospitalization. Both treatments were associated with clinical improvement, considering a decrease of positive and negative symptoms (change from baseline to 28-day assessment in positive PANSS score −9.0 ± 6.1 and −8.4 ± 7.5, and in negative PANSS score −9.1 ± 4.9 and −6.4 ± 6.0 after CBD and amisulpride, respectively; all comparisons *p* < 0.001). However, CBD treatment had a superior side-effect profile in terms of less severe changes in weight gain, extrapyramidal symptoms, prolactin levels, and sexual functioning. In addition, Leweke et al. (2012) measured anandamide levels in serum before and after treatment with CBD or amisulpride and the relationship with psychotic symptoms [54]. As compared to treatment with amisulpride, CBD showed a significant increase in anandamide levels and this was associated with the improvement of psychotic symptoms (i.e., decrease of total PANSS score). These findings suggest that anandamide levels could serve as a possible biomarker for the efficacy of CBD treatment [54]. In the largest randomized placebo-controlled trial to date, McGuire et al. (2018) assessed the effect of six-week treatment with CBD (1000 mg/day) added to antipsychotic medication in 88 moderately ill (total PANSS score >60) outpatients with schizophrenia [55]. After six weeks, positive symptoms (change from baseline −3.2 after CBD and −1.7 after placebo, treatment difference = −1.4, 95% CI = −2.5, −0.2) and global clinical impression significantly improved in the CBD group compared with placebo (treatment difference = −0.5, 95% CI = −0.8, −0.1 for improvement rates and = −0.3, 95% CI = −0.5, 0.0 for change in severity of illness) [55]. These case studies and RCTs suggest that CBD treatment for psychosis is beneficial and could possibly be as effective as antipsychotic medication.

Two RCTs showed less conclusive results of CBD treatment on positive, negative, and cognitive symptomatology. The randomized placebo-controlled trial by Hallak et al. (2010) presented the effect of acute treatment with single doses of CBD on selective attention as measured with the Stroop Colour and Word Test in a heterogeneous group of 28 schizophrenia patients (illness duration <5 years (*N* = 11), >5 years (*N* = 17)) [56]. These patients performed the Stroop test twice: The first time without the administration of any drug and the second time after oral administration of either placebo, 300 mg, or 600 mg CBD. After the two sessions, all groups showed improvement in cognitive performance (i.e., reduced number of errors during the Stroop test). Improvement was greater in the placebo and CBD 300 mg groups, compared with the patients who received CBD 600 mg. There was no effect of CBD treatment on both positive and negative symptoms [56]. Second, the most recent double-blind, randomized placebo-controlled trial by Boggs et al. (2018) examined treatment with oral CBD (600 mg/day) or placebo adjunctive to a stable dose of antipsychotic medication in 36 chronic schizophrenia patients (mean illness duration >25 years) [57]. Although positive, negative, general, and total PANSS decreased and cognitive performance increased over time in both groups, there were no significant differences between groups. Thus, in this trial, symptomatology and cognitive performance did not improve after adjunctive CBD treatment in schizophrenia outpatients who were receiving long-term polypharmacy for a myriad of psychiatric symptoms (Boggs et al., 2018) [57]. One possibility is that these results may be explained by the significant difference in the use of multiple antipsychotics between the placebo (38.9%) and CBD groups (11%) [57].

In summary, most of abovementioned studies provided evidence for the potential of CBD as an antipsychotic treatment, which could alleviate both cognitive and psychotic symptoms in patients with psychotic disorders. The studies that showed negative results provided either a single dose of CBD [56] or included chronic schizophrenia patients who received multiple types of antipsychotic medication [57].



#### *3.2. CBD—Substance Use Disorders*

To date, eight studies assessed the potential effect of CBD as treatment for SUD (Table 2). Six studies focussed on cannabis dependence and two on tobacco dependence.

#### 3.2.1. Cannabis Dependence

The treatment of cannabis dependence with a cannabis-extracted CBD/THC mixture (Sativex) was assessed in three clinical trials. In the first double-blind randomized controlled trial by Allsop et al. (2014), 51 inpatients with cannabis dependence received Sativex or placebo for six days along with cognitive behavioural therapy [58]. Immediately after treatment, Sativex significantly decreased cannabis withdrawal and craving symptoms and improved treatment retention rates. At 28 days, both groups showed a decrease in cannabis use, in the amount of cannabis-related problems, and in the severity of cannabis dependence from baseline to follow-up, but the differences between groups were no longer significant [58]. A second double-blind randomized placebo-controlled trial assessed the effects of an eight-week treatment with self-titrated or fixed doses of Sativex in nine subjects with cannabis dependence [59]. During treatment sessions, when cannabis use was not allowed, both fixed and self-titrated doses of Sativex reduced cannabis withdrawal symptoms, however, the high fixed dose seemed the most effective. Sativex did not influence cannabis craving. The same research group performed a larger double-blind randomized placebo-controlled trial in which 27 cannabis-dependent subjects were treated with self-titrated dosages of Sativex in combination with cognitive behavioural therapy over 12 weeks [60]. The abstinence rate did not change significantly between baseline and follow-up. Cannabis use, withdrawal, and craving symptoms reduced over time in both groups. Sativex was associated with a greater reduction in cannabis craving symptoms when compared with placebo.

While the previous studies used CBD/THC mixtures, the following three studies described treatments of cannabis dependence with pure CBD. The first study by Crippa et al. (2013), reported a 19-year-old female with cannabis dependence who was treated with oral CBD over 11 days [61]. The dose was 300 mg on day 1, 600 mg on days 2–10 and 300 mg on day 11. During treatment, cannabis withdrawal, anxiety, and dissociative symptoms progressively decreased. A six-month follow-up period revealed a relapse of cannabis use, however at a lower frequency than on admission [61]. In a second case report, Shannon and Opila-Lehman (2015) described the treatment with 24–18 mg CBD oil adjunctive to citalopram and lamotrigine for a 27-year-old male with cannabis disorder and bipolar disorder. During the use of CBD oil, the patient did not use cannabis, showed a decrease in anxiety, and demonstrated improved sleep quality [62]. Third, the open-label clinical trial by Solowij et al. (2018) assessed the effects of ten-week treatment with CBD (200 mg/day) on psychological symptoms, cognition, and plasma concentrations [63]. Twenty frequent and ongoing cannabis users, of which twelve were dependent users (severity dependence scale score ≥3) and ten were nondependent users (severity dependence scale score <3), participated in this trial. Between baseline and post treatment sessions, cannabis use and withdrawal did not change, but cannabis-related experiences (i.e., euphoria and feeling high) decreased. Anxiety, depressive, and psychotic-like symptoms showed greater reductions in dependent than nondependent users. Attentional switching, verbal learning, and memory improved in all participants. Remarkably, higher CBD plasma concentrations were associated with lower psychotic-like symptoms (total and negative), distress, anxiety, and severity of cannabis dependence [63]. These results suggest greater effects of CBD in dependent users which can possibly be detected through CBD plasma concentrations.

Taken collectively, CBD shows some promise in the treatment of cannabis dependence as it reduces behaviour relevant to addiction such as craving and withdrawal in almost all studies. Because double-blind placebo-controlled RCTs with pure CBD are lacking, the evidence for the efficacy of products containing a combination of CBD and THC in the treatment of cannabis dependence is more convincing.



 Cannabidiol; CBT: Cognitive behavioural therapy; RCT: Randomized clinical trial; THC: Δ9-Tetrahydrocannabinol.

#### 3.2.2. Tobacco Dependence

Morgan et al. (2013) assessed the effects of the optional use of an inhaler containing CBD (400 μg/dose) during one week in 24 individuals who smoked >10 cigarettes/day and intended to quit [64]. Results showed that CBD reduced the total number of cigarettes smoked during the treatment period. However, CBD did not have an effect on craving symptoms. In addition, craving was reduced in both groups at the end of treatment, but this did not maintain at follow-up [64]. Additionally, a second clinical trial into the efficacy of CBD treatment for tobacco dependence provided information about treatment outcomes related to motivation and evaluation. Hindocha et al. (2018) treated 30 tobaccodependent individuals with a single dose of CBD 800 mg [65]. Attentional bias to pictorial cigarette cues was measured using a visual probe and an explicit rating task. In addition, craving, withdrawal, and side effects were assessed. After overnight cigarette abstinence, CBD reduced attentional bias to cigarette cues and pleasantness of cigarette cues, which could suggest that CBD has a potential effect on the motivational aspects of addiction. In this trial, CBD did not have an effect on craving and withdrawal. Moreover, no significant differences were found between CBD and placebo on side effects [65].

#### *3.3. CBD—Psychosis and SUD*

Schipper and colleagues (2018) were the first who described the efficacy of CBD treatment for patients with a psychotic disorder and a comorbid treatment-resistant cannabis use disorder (Table 3) [66]. Seven hospitalized patients received eight weeks of treatment with Bedrolite, medicinal cannabis that contains 0.4% THC and 9% CBD, as add-on therapy to conventional antipsychotic medication. The medicinal cannabis was supposed to substitute street cannabis used by the patients but was only provided at fixed moments during the day. Doses ranged from 0.125 to 0.5 g daily (11–45 mg CBD), depending on dose and frequency of the use of street cannabis before admission. Treatment with CBD-rich medicinal cannabis did not affect psychosis- or dependence-related symptomatology. Patients preferred street cannabis over the medicinal cannabis and started to use additional street cannabis during the treatment program [66]. The most likely explanation for these negative results was the low THC concentration in Bedrolite as compared to street cannabis. As a result, the substitution of THC-rich street cannabis by medicinal cannabis with mainly CBD may have been too abrupt for most patients.



CBD: Cannabidiol; THC: Δ 9-tetrahydrocannabinol.

#### **4. Discussion and Conclusions**

The current review aimed to provide a detailed and up-to-date systematic literature overview of studies that investigated the efficacy of CBD treatment for schizophrenia and/or SUD. Based on this overview, a second aim was to examine whether there is a specific subgroup of patients with schizophrenia, SUD, or both that may benefit most from CBD treatment. In some but not all studies, CBD seemed effective as a treatment for psychosis and SUD. CBD may have the capacity to alleviate positive, negative, and cognitive symptoms in schizophrenia, as well as craving and withdrawal in SUD. Although most of the studies showed promising results, differences in study design, patient population, and use of concomitant medication make it difficult to define specific subgroups to whom CBD should be administered. In addition, CBD doses and administration were different between studies and most

of the reviewed studies did not describe the source of CBD (i.e., synthetic or cannabis extracted), which may have different efficacy. However, the results of the reviewed studies suggested some features that may contribute to the identification of patients who may benefit most from CBD treatment.

Research into CBD treatment for psychosis provided evidence for a few possible clinical and biological characteristics of the subgroup. The effects of CBD were studied in patients in both early and later stages of psychotic disorders. Overall, acutely psychotic and early onset patients demonstrated reductions of positive and negative symptoms [51,54], while treatment resistant and chronic patients showed less promising improvement [56,57]. Even though Makiol and Kluge (2019) described a chronic schizophrenia patient who exhibited great clinical improvement (i.e., change of total PANSS score: 49) [53], the majority of the results suggest that CBD may be more effective in the early stage of psychotic disorders. This is in line with previous studies suggesting that immune dysregulation (i.e., microglial activation) is mainly involved in the early stage of psychotic disorders [67,68]. As cannabinoid receptors are also present on microglia, it is possible that CBD exerts its effects by decreasing microglial activity [69]. Furthermore, anandamide levels in serum could serve as a possible biomarker for the efficacy of CBD treatment. For instance, Leweke et al. (2012) reported a significant increase in anandamide levels after CBD treatment, which was associated with the improvement of psychotic symptoms (i.e., decrease of total PANSS score) [54]. This is in concurrence with a previously reported inverse association between elevated anandamide levels in cerebrospinal fluid and psychotic symptoms in antipsychotic-naïve patients [29–31].

Research into CBD treatment for SUD primarily focussed on cannabis dependence. Taken collectively, CBD shows promise in the treatment of cannabis dependence as it reduces craving and withdrawal in almost all studies. However, these studies have heterogeneous study designs and administration methods. The differences in administration and dosages may provide a possible explanation for the different results observed in the included studies. For instance, THC/CBD mixtures might be more effective in reducing some features of cannabis dependence (i.e., craving, use and withdrawal) than pure CBD. Moreover, the level of cannabis dependence and intrinsic motivation for treatment, may help to define a possible subgroup of patients in which CBD is more effective. Dependent users (i.e., those with a severity dependence scale score ≥3), showed reduced anxiety, depression, and psychotic-like symptoms after a 10-week treatment with CBD, compared with nondependent users [63]. However, studies that include individuals with more symptoms at baseline can show greater reductions after treatment. Therefore, it is difficult to determine whether symptom severity is truly a patient characteristic that could predict better outcomes after CBD treatment. Solowij et al. (2018) also found that cannabis-related experiences decreased after treatment [63], which is in accordance with previous studies that indicate that CBD counteracts the effects induced by THC [46,47]. Intrinsic motivation for treatment seems an important aspect as well, as it may increase medication adherence [65]. Conversely, patients that do not seek treatment are less inclined to follow strict study protocols [66]. The majority of the discussed studies recruited individuals with cannabis dependence from the community, which suggests that these individuals were at least open for treatment. To a certain extent, this may explain why the study by Schipper et al. (2018) found that CBD administration was not effective [66], as they included individuals that did not seek treatment.

Considering the efficacy of CBD in both psychotic disorders and SUD, one can speculate that CBD should also be effective in the treatment of the comorbidity. However, only Schipper et al. (2018) studied this population, with negative results [66]. As discussed previously, these patients were treatment resistant for SUD and showed lack of motivation for treatment. An additional limitation of this study was the good baseline functioning in five out of seven patients. Moreover, this study administered CBD in a formulation that contained very little THC, which possibly explains why the participants preferred street cannabis.

Future studies could take these limitations into account and should focus on examining the effects of CBD in the different stages of psychotic disorders, considering the high prevalence of comorbid SUD. Studies into psychotic disorders could use CBD (i.e., either as monotherapy or add-on) to treat psychotic symptoms and to prevent relapse in early stages, while exploring the effects on comorbid substance use (e.g., cannabis). These studies should use standardized measures to assess cannabis use. In later stages and comorbid treatment-resistant SUD, CBD studies may aim to reduce cannabis use, using harm-reduction strategies (e.g., gradually shift the THC/CBD ratio in medicinal cannabis in favour of CBD) [70]. Currently, nine ongoing clinical trials that study the effects of CBD on psychotic disorders or SUD (including alcohol and cocaine misuse) are registered in clinicaltrials.gov, of which one (NCT03883360) includes patients with recent-onset psychotic disorder and cannabis use. Therefore, more results on this topic are expected in the near future.

It remains unclear if the efficacy of CBD in schizophrenia, addiction, and their comorbidity could be explained by shared or different biological mechanisms. To elucidate this, future efforts should be taken to study the relationship between the eCB system, GABA/glutamate, and the immune system. For example, neuroimaging studies (e.g., positron-emission tomography, PET and magnetic resonance spectroscopy, MRS) could measure CB1 receptor densities and markers for glia in patients with schizophrenia and/or SUD who were treated with CBD.

In conclusion, CBD treatment is a promising and novel tool with several potential applications in the treatment of psychotic disorders, substance use disorders, and their comorbidity. Large-scale trials are needed to establish its clinical utility.

**Author Contributions:** Conceptualization, A.B. and M.G.B.; methodology, H.J. and S.S.G.; investigation, H.J.; data curation, H.J. and S.S.G.; Writing—Original draft preparation, A.B., H.J. and S.S.G.; Writing—Review and editing, A.B. and M.G.B.; visualization, H.J. and S.S.G.; supervision, A.B. and M.G.B.

**Funding:** M. Bossong was supported by a Veni fellowship from the Netherlands Organization for Scientific Research (grant number 016.166.038).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Journal of Clinical Medicine* Editorial Office E-mail: jcm@mdpi.com www.mdpi.com/journal/jcm

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18