We first causally tested the influence of different housing conditions, i.e. EE vs SE (n=24 each) on behavioral traits that are related to personality factors relevant to addiction in humans, namely anxiety (14), sensation seeking (7, 11, 30), boredom susceptibility (12), reward sensitivity (31) and sign-tracking (32) (Fig. S1), that we propose to reflect a multidimensional behavioural profile in the rat (see SOM). EE abolished the drug use proneness/addiction-resilience trait of high locomotor reactivity to novelty (HR), decreased the anxiety-associated behaviors of head dipping on the elevated plus-maze (EPM); EE also disrupted the asymmetric conditioned approach behavior usually displayed by sign-trackers towards a food CS, in line with previous evidence that EE impairs the attribution of incentive salience to food-paired cues (33) (Fig. 1 and Fig. S2, S3). EE did not influence each trait independently, but it shaped their relative multidimensional configuration in three distinct multidimensional behavioural profiles identified at the population level by cluster analysis (Fig. 1B). In particular EE increased the probability of displaying a multidimensional behavioural profile driven by sweetness and novelty preference (model 3), as opposed to that driven by stress reactivity (11, 34-36) (model 1) that is preferentially displayed by SE rats (Fig. 1B).

We then tested whether the influence of housing conditions on these behavioral traits mediated the well-established effect of EE on the tendency to self-administer stimulant drugs (19-21). As previously described (21, 37), rats living under EE conditions were significantly less likely to acquire cocaine self-administration (SA) (Fig. 2A) than SE rats. However, the data also revealed that this EE effect was driven by a further decrease in the propensity to acquire cocaine SA in non-vulnerable populations, i.e. rats that have been shown either to have a low tendency to selfadminister cocaine, characterized by a low locomotor response to novelty (low responders, LR) (7, 11) or a low saccharine preference (LSP) (38), or rats with low novelty (LNP) that are resilient to the transition to compulsive cocaine self-administration (12)(Fig. 2B, 2C and Fig. S4-5). The increased weight of the addiction-like behavioral proneness conveyed by sweetness and novelty preference to the multidimensional behavioural profile shaped by EE, at the expense of locomotor reactivity to novelty which confers resistance to addiction-like behavior (7, 12) (Fig. 2C), suggests that while decreasing the tendency to self-administer cocaine, EE may make the switch from controlled to compulsive drug use more likely in vulnerable individuals. We therefore investigated, in a large heterogenous cohort, the influence of housing conditions on the individual tendency to develop addiction-like behavior following prolonged exposure to cocaine SA (8, 12, 39).

Using new methodologies that enable the maintenance of i.v. catheter patency for several months in group- housed rats, we trained a new cohort of 48 rats (EE vs SE n=24 each) to selfadminister cocaine for ~50 days before they were tested for their addiction-like behavior according to the stimulant addiction-relevant 3 criteria model (7, 8, 12, 39-41). This multidimensional model is based on the operationalization of the main behavioral characteristics of addiction in the DSM (9), namely high motivation for the drug, the persistence of drug taking despite adverse consequences and an inability to refrain from seeking the drug when signaled as unavailable (see (7, 39) and Fig. S6).

In order to determine the influence of housing conditions on the vulnerability to develop addictionlike behavior rats from both housing conditions were considered to form a large, heterogenous, cohort, and the individual tendency eventually to display each of the three addiction-like criteria was measured in that population irrespective of the housing conditions.

Regardless of their housing conditions, rats were identified as having 0, 1, 2 or 3 addiction-like behavioral criteria (Fig. 3A and S6, see SOM). As previously described (7, 39), the overall population was linearly distributed alongside an addiction severity index (8) in which only those rats showing the three addiction criteria (3-criteria or 3crit) had a score that was beyond the standard deviation of the cohort, while 0crit, addiction-resilient rats had a negative addiction severity score. As previously described (7, 12, 42, 43), the linear addiction severity index provides a criterion-dependent difference in the manifestation of each of the addiction-like behaviors so that 3crit addiction-vulnerable rats had scores significantly higher than those of 0crit addictionresilient rats in each of the addiction-like behaviors (Fig. S6A).

The individual vulnerability to develop addiction-like behavior, seen in 13.8% of the population, was significantly influenced by housing conditions. A retrospective analysis revealed that EE rats had much higher addiction scores than SE rats (Fig. 3B and S6B) and that the entire 3crit population was comprised exclusively of EE rats (Fig. 3C). The influence of housing conditions on addiction vulnerability was primarily driven by a greater tendency to show compulsivity and the persistence of drug seeking, and less so by an increased motivation for the drug (Fig. 3D and S6B). Importantly it was not due to any overall differences in drug intake (Fig. 3D and S6C) nor to any differences in pain sensitivity (44) (Fig. 3D).

This dimensional approach provides clear evidence that while decreasing the tendency to initially engage in drug use EE instead increases the likelihood of developing addiction-like behavior in vulnerable individuals.

We further investigated whether the facilitatory effect of EE on the development of compulsive cocaine intake also extended to alcohol. In a third experiment, the tendency of rats housed in EE vs SE (n = 12 each) to develop quinine-resistant (45-48), compulsive, alcohol intake was assessed after several months of intermittent access to alcohol in a two bottle-choice procedure (49, 50).

EE rats did not differ from SE rats in their alcohol intake over 47 daily sessions of intermittent or unpredictable access to a choice between alcohol and water (Fig. S7A). However, as compared to SE rats, EE rats showed an increased tendency to relapse to alcohol drinking following several weeks of imposed abstinence (Fig. 4A) and did so compulsively in that they showed persistence of drinking at relapse despite adulteration by the bitter tastant, quinine (47) (Fig. 4B and S7). Ultimately, EE rats were shown to be more vulnerable than SE rats in their development of two key behavioral features of alcohol use disorder (2, 51).

Together, these results reveal a bidirectional effect of housing conditions on the initial tendency to self-administer drugs and on the vulnerability to compulsively take drugs (2). The only difference between EE and SE rats in these parametrically controlled experiments was the enrichment of living context of the former and the resulting differential negative contrast with the drug taking context (Fig. S8) shown previously to influence the relative preference for cocaine or heroin (52). Thus, while these results may at first glance appear counter-intuitive, or at variance with the literature on the effects of environmental enrichment on drug related behaviors based on the comparison of EE and SI rats (see 37 for further discussion), they are highly consistent with previous evidence that EE rats are more sensitive to the motivational effects of cocaine than SE rats (22) and highlight the importance of the experiential factors inherent in the contrast between a rat’s living conditions and the drug-taking setting.

In contrast to the well-known ‘Rat Park’ experiment, in this study EE rats were provided access to cocaine or alcohol in a relatively impoverished drug setting, rather than in the enriched housing environment itself (53). Additionally, the “Rat Park” experiment did not investigate individual differences in addiction-relevant behaviors, such as compulsive drug intake, that emerge only after a prolonged period of training. In the present study EE rats, unlike SE rats, first encountered the drug in a state of relative social, sensory and cognitive deprivation because of the marked contrast between their enriched housing conditions and the drug SA context the features of which were similar to those of the SE (Fig. S8). This negative environmental contrast likely biases individuals to self-administer drug in the experiential setting to ameliorate an internal deficit/distress state (5, 54). In humans, this negative motivational state is frequently associated with the initiation of drug use and increases the vulnerability to addiction (4, 55) by moderating or exacerbating the behavioral traits that here we show are influenced by EE, namely sensation seeking, boredom susceptibility/novelty preference and anxiety (55-57).

According to this framework, rather than solely an animal’s housing conditions, it is the experiential factors in the drug setting, such as those related to high boredom susceptibility in humans or novelty preference in rats (12, 58, 59), that influence the vulnerability subsequently to compulsively take drugs. We therefore extended the study to causally test the hypothesis that the acquisition of alcohol drinking as a coping, stress-reducing, strategy is sufficient to exacerbate the vulnerability to switch to compulsive alcohol intake in rats living under similar housing conditions. The acquisition of adjunctive behaviors, such as schedule-induced polydipsia (SIP), has been shown across species to result in a profound decrease in the levels of stress-related hormones that are increased by intermittent food delivery (60, 61), thereby providing a valuable tool to investigate the influence of alcohol drinking as a coping strategy on the subsequent vulnerability to develop compulsive alcohol intake. In SIP procedures, most rats readily acquire an adjunctive, anxiolytic, water drinking response (62-64) (Fig. 4C), as a way of coping with distress. However, some rats never develop this adjunctive behavior (62-64). We hypothesized that these individuals might instead develop adjunctive alcohol, not water, drinking as a coping response. We further predicted that rats having selectively acquired adjunctive alcohol drinking would also go on to develop compulsive (quinine-resistant) alcohol intake (see SOM)..

In a fourth longitudinal experiment, alcohol was provided to a cohort of 48 SE rats (65) that had been trained to cope with the distress of SIP by adjunctively drinking water (61, 66, 67). A subpopulation of these rats also maintained their established coping (adjunctive drinking) strategy when alcohol was introduced in place of water (Water Copers, WC). However, as predicted, a distinct sub-population of rats never acquired adjunctive water drinking and only developed a coping strategy by drinking alcohol (Alcohol Copers, AC) (Fig. 4C).

AC rats, which did not differ from WC rats in terms of total alcohol intake over the course of the 20 days of exposure (Fig. S9 and SOM), also persisted in drinking alcohol when adulterated with quinine (Fig. 4D and S9). This differential vulnerability to develop compulsive alcohol drinking was not due to any differences in blood alcohol levels since they were similar in both groups (Fig. S9) and it was therefore specific to the experiential nature of the SIP context. Only rats that had acquired the adjunctive, anxiolytic behavior of drinking alcohol went on to drink alcohol compulsively.

Together these results show that a negative experiential context, whether inherent to the drug taking environment or triggered by a negative affective contrast from housing/living conditions at the onset of alcohol use, may be an important gateway to the development of compulsive drinking in individuals unable to learn to cope with negative states by alternative means (Fig. 4D and S9).

Following a week of habituation to the vivarium, forty-eight 6-7 week-old male SD rats were housed in EE (n=24) or SE (n=24) conditions (see SOM) until the end of the experiment. Subsequently, their sensitivity to a natural reward (83) was assessed in a two-bottle choice procedure with water and saccharine (Figure S1). Rats were then screened for their anxietyrelated behavior and their locomotor reactivity. The individual variability in ascribing incentive value to a CS and novelty preference was measured in an autoshaping (sign-tracking) (30, 84) and a novelty-induced preference task (12).

At day 96, rats underwent surgical implantation of an indwelling catheter in the right jugular vein for SA training. Five days later individual differences in the acquisition of cocaine SA were assessed, starting with six 3.5-hour daily sessions under a fixed ratio (FR) 1 schedule of reinforcement (FR1) followed by one under FR3 session and three under FR5. A normalized score for each individual for each trait (% time spent in the open arm (OA), saccharine preference (%), time spent in novel compartment (%), total distance travelled in the open field (OF) and contacts made with the CS-associated lever) was calculated by subtracting the mean of the population to each individual value and dividing the result by the standard deviation of the population (z-score) as previously described (42). The z-score of each rat was used within a multi-dimensional model (represented as radar plots) within which the contribution of each behavioral trait to the multidimensional behavioral characterization of each individual (referred to as rat multidimensional behavioural profile) was calculated as the percentage of the normalized score of all the behavioral traits.

These multidimensional behavioural profiles were subjected to a k-means cluster analysis (72, 85) performed on the value (percentage) of each of the 5 behavioral traits with the following parameters: maximization of distances between clusters, maximum 20 iterations, maximum of 3 clusters.

Following a week of habituation to the vivarium, forty-eight male SD rats were housed in EE (n=24) or SE (n=24) conditions until the end of the experiment (Figure S1). Five days following the implantation of an indwelling catheter into their right jugular vein, rats were trained to selfadminister cocaine for 5 sessions under an FR1 schedule of reinforcement followed by 2 sessions of FR3 and 40 sessions under a FR5 schedule of reinforcement.

The individual vulnerability to develop addiction-like behavior was assessed as previously described (7, 12, 40, 42, 43) by (i) the inability to refrain from responding despite the drug being signalled as unavailable, during so-called no-drug periods, (ii) a high motivation for cocaine, assessed over a single session under a progressive-ratio schedule of reinforcement (40) and (iii) the persistence of responding for cocaine despite punishment, as measured in a single 40 min long session during which instrumental responses were punished response-contingently (40). Animals were scored for each addiction-like behavior independently. If the score of the individual was in the 33% highest percentile of the distribution, that individual was considered positive for that criterion. Animals were then separated in four groups depending on the number of positive criteria met (from 0 to 3). An addiction score was calculated for each animal as the sum of the normalized scores (Z-scores) for each of the three addiction-like criteria (12, 42).

Following a week of habituation to the vivarium, twenty-four male SD rats were housed in EE (n=12) or SE (n=12) conditions until the end of the experiment (Figure S1). Rats were first exposed to 21 two-bottle choice sessions under intermittent access. Rats were subsequently exposed to 20 two-bottle choice sessions under unpredictable to promote binge-like drinking behavior followed by 6 sessions under intermittent access. Access to alcohol was then withdrawn for 20 days during which rats stayed in their home cage with access only to water. Rats were subsequently challenged with a relapse test which consisted of a first 4-hour two-bottle choice session followed by a 24-hour two-bottle choice session. The individual propensity to relapse was calculated as the ratio between alcohol intake at relapse and alcohol intake in the last 4-hour twobottle choice session prior to abstinence.

EE and SE rats were then divided into two subpopulations (n=6 each) displaying the same drinking behavior at every stage of the experiment (training pre-abstinence and during relapse). One subpopulation of EE and SE rats was exposed to two sessions during which alcohol was adulterated with quinine (0.1 g/L) (86) while the other subpopulation was exposed to nonadulterated alcohol. This between-subject approach was carried-out to control for any potential confounding influence of a habituation to the bitter taste of quinine (latent inhibition) from which a within-subject design may have suffered. Finally, to control for the specificity of the effect of quinine on alcohol intake, rats underwent two 4-hour water adulteration sessions during which they had access to a single bottle of water supplemented with quinine (0.1 g/L).

Following a week of habituation to the colony, forty-eight male SD rats were all housed in a SE (Figure S1). Rats were food restricted to 80% free-feeding body weight (62, 64) for a week before being trained in a schedule-induced polydipsia (SIP) procedure with water for twenty 1hour daily sessions, as previously described (62, 64). Then, water was substituted with 10% alcohol and rats were trained in twenty 1-hour daily SIP sessions with alcohol. The day following the last SIP session with alcohol, the individual tendency to persist in drinking alcohol despite adulteration with quinine (0.1 g/L) was assessed.

Rats then underwent five baseline SIP with alcohol sessions over which they recovered their preadulteration alcohol intake level. Blood samples were collected immediately after the last session to measure their BAL.

In order to test whether persistence of drinking alcohol despite adulteration by rats trained under the SIP procedure was dependent on the SIP context in which rats initiated their alcohol use, another cohort (n=24) was housed under similar conditions and was then given a two-bottle choice procedure under intermittent access for twelve 24-hour sessions, followed by twelve 1hour sessions. Rats were subsequently given eight additional 24-hour two-bottle choice sessions under intermittent access followed by a single 1-hour session during in which their individual tendencies to maintain alcohol drinking despite adulteration was measured.

The individual propensity to persist in drinking alcohol despite adulteration was calculated as the ratio between “alcohol + quinine” intake over the average alcohol intake of the first hour during the last three two-bottle choice procedure sessions under intermittent access.

Data are presented as number of individuals, mean (for some radar plots) or mean +/- standard error of mean (SEM). Behavioral data were subjected to repeated measure, one-, two- factorial analyses of variance (ANOVAs) or Kruskal-Wallis or Mann-Whitney U wherever appropriate. The experimental groups were used as between-subject factor and, depending on the analysis, the arm (open/closed arms) and the time (number of sessions or session blocks) were used as within-subject factors. Upon confirmation of significant main effects differences among individual means were analyzed using the Newman-Keuls post-hoc test and planned comparisons wherever appropriate. a was set at 0.05 and the effect size was reported by the partial etasquared value (ηp2) (see SOM for more details).

Enrichment of the housing environment (EE) exacerbates the contribution of sweetness preference to a multidimensional behavioral model of personality diminishing the contribution of anxiety, sign-tracking and reactivity to novelty (A and B). These behavioral traits differed in their contribution to the three distinct, environment-dependent [EE vs SE: Chi2=8.43, p<0.05], multidimensional behavioural profiles identified from a cluster analysis [cluster x trait interaction: F8,180=14.373, p<0.0001, ηp2=0.39], thereby suggesting a complex multidimensional interaction between the environmental conditions and behavioral traits of vulnerability versus resilience to addiction. Quantitatively, EE did not influence the sweetness preference of rats [main effect of group: F1,46<1; session: F3,138=11.538, p<0.0001, ηp2=0.20 and group x session interaction: F3,138=1.7583, p>0.05] (C) nor their preference for novelty [main effect of group: F1,41=2.3124, p>0.05] (D). However, EE abolished the behavioral trait of resilience to addiction, namely high locomotor response to novelty (HR), in that EE rats showed a marked decrease in their locomotor reactivity to novelty as compared to SE rats [main effect of group: F1,46=19.274, p<0.0001, ηp2=0.30; time: F11,506=102.16, p<0.0001, ηp2=0.69; group x time interaction: F(11,506)=1.5757, p>0.05] (E). Similarly, EE prevented the development of asymmetrical approach behavior, biased towards the CS, characteristically displayed by sign tracker rats raised in a SE during the exploitation phase of an AutoShaping task [SE, left panel: phase x approach response: F1,23=24.459, p<0.0001, ηp2=0.52; EE, right panel: phase x approach response: F1,23=3.8233, p>0.05] (F). EE influenced behavioral manifestations of anxiety on the elevated plus maze (EPM) (G) such that, despite spending a similar percentage of time in the open arms of an EPM [main effect of group: F1,43=1.472, p>0.05], EE rats made more head dippings while on the open arms 2 than did SE rats [main effect of group: F1,43=3.8447, p<0.05, ηp =0.08]. *: p≤0.05.

EE rats showed a lower rate of acquisition of cocaine self-administration (SA) than SE rats [main effect of group: F1,46=4.8168, p<0.05, ηp2=0.10; session: F9,405=200.63, p<0.0001, ηp2=0.82 and group x session interaction: F9,405=3.0058, p<0.01, ηp2=0.06] (A, left panel). This effect was not simply driven by a small number of the EE population that failed to acquire SA. In fact, more than 50% of rats in the EE group stayed below the SA acquisition criterion (median number of cocaine infusions set at each session) (A, right panel). The high responder (HR) phenotype, which was almost non-existent in the EE population, violating the expected distribution typically observed in SE rats [Chi2=15.41, p<0.0001] (B, right panel), was the behavioral trait that best recapitulated the differential tendency to self-administer cocaine shown by EE rats [main effect of group: F1,22=7.1607, p<0.05, ηp2=0.25; session: F9,198=138.78, p<0.0001, ηp2=0.86 and group x session interaction: F9,198=3.7631, p<0.001, ηp2=0.15] (B, left panel). EE also decreased the propensity of addiction-resilient, LNP rats, and rats with low reward sensitivity, LSP rats, to acquire cocaine SA [environment x phenotype x session interaction: F9,162=5.2896, p<0.0001, ηp2=0.23 and F9,180=2.1008, p<0.05, ηp2=0.10, respectively]. Thus, LNP and LSP rats from the EE population received far fewer cocaine infusions than their SE counterparts [planned comparison: p<0.01 and p<0.05 respectively] (C and D, left panels), even though the qualitative nature of these traits was not influenced by housing condition [environment x phenotype interaction: F1,18<1; environment x phenotype x time interaction: F4,72=1.9406, p>0.05, C, right panel and environment x phenotype interaction: F1,20=1.7919, p>0.05; environment x phenotype x time interaction: F3,60<1, D, right panel]. *: p≤0.05.

After 47 days of cocaine SA under a FR-5 schedule of reinforcement, rats from a heterogenous cohort comprising similar numbers of individuals housed in SE or EE were tested for the three behavioral criteria of addiction-like behavior: high motivation for the drug measured under progressive ratio; the maladaptive engagement in drug seeking, measured as the inability to refrain from responding when the drug is signaled as unavailable; the persistence of drug seeking despite negative consequences, measured as resistance to punishment. Rats were stratified on the number of criteria they displayed, i.e. 0, 1, 2 or 3 criteria (A). The 3crit group (n=5, 14% of the entire population) was the only one having an average addiction score that was outside the standard deviation of the overall population (grey bar). In contrast, the 0crit group (n=12, 33% of the overall population) was the only one displaying highly negative scores confirming their resilient phenotype [main effect of crit: F3,32=17.955, p<0.0001, ηp2=0.63] (A). The difference between these groups was not attributable to a differential exposure to cocaine throughout their SA history [main effect of crit: F3,32<1]. However, retrospectively factoring housing conditions revealed that EE rats had significantly higher addiction scores than SE rats, the latter actually having negative scores [main effect of environment: F1,34=8.8255, p<0.01, ηp2=0.21] (B). All of the 3crit rats and the majority of 2crit rats were from the EE population, in clear contrast to the predominance of SE rats in the resilient, 0crit, population (C). The higher addiction scores of EE rats as compared to SE rats were attributable to each of the addiction-like behaviours [main effect of environment: F1,34=8.82, p<0.01, ηp2=0.21, behavior: F1,2<1 and environment x behavior interaction: F1,68<1], even if more pronounced differences were observed for compulsivity and persistence of drug seeking than for the motivation for the drug. The facilitation of the transition to addiction by EE was not attributable to a differential exposure to cocaine or to a differential sensitivity to pain [main effect of environment: Fs1,34<1] (D). *: p≤0.05.

After 20 days of abstinence following a 3-month history of exposure to alcohol in an intermittent two-bottle choice procedure, EE rats were more prone to relapse to alcohol drinking than SE rats [Mann-Whitney: U=37.00, p<0.05] (A). EE rats were also more prone to persist in drinking alcohol despite adulteration with quinine, thereby displaying compulsive drinking behavior [Kruskal-Wallis EE: H1,12=2.08, p>0.05; SE: H1,12=4.33, p<0.05] (B). In another cohort of outbred SE rats, introduction of the opportunity to drink alcohol as a means of coping with distress in a scheduleinduced polydipsia procedure resulted in a specific subpopulation of individuals (Alcohol Copers, AC) that developed adjunctive alcohol drinking behavior [main effect of group: F1,10=1.2771, p>0.05; session: F19,190=1.2348, p>0.05; group x session interaction: F19,190=1.4429, p>0.05; C, right panel] that they had failed to acquire when water was available, in marked contrast to rats that had acquired high levels of water intake (Water Copers, WC) [main effect of group: F1,10=33.619, p<0.001, ηp2= 0.77; session: F19,190=9.8985, p<0.0001, ηp2=0.50 and group x session interaction: F19,190=7.0845, p<0.0001, ηp2= 0.41; C, left panel]. Although both WC rats and AC rats consumed the same amount of alcohol overall [main effect of group: F1,10=1.2771, p>0.05], only the latter, i.e. those that had acquired a coping strategy by drinking alcohol, subsequently displayed compulsive alcohol drinking, being resistant to adulteration of alcohol with quinine [main effect of group: F1,10=8.3820, p<0.05, ηp2=0.46] (D). *: p≤0.05.