Acquisition of self-administration
All nicotine + cue rats successfully acquired and stabilized their self-administration behavior during the initial 12 sessions (FigS2a-b). The behavior was dependent on nicotine, as it differed significantly from the behavior observed in the saline + cue group (FigS2d-g and SI). Additionally, the plasma cotinine/nicotine ratio, measured in a subset of rats after a standard self-administration session, exhibited a positive correlation with the number of self-infusions (FigS2c). This finding is consistent with a faster nicotine metabolism (indicated here by a higher cotinine/nicotine ratio) promoting a higher nicotine intake, as shown previously [33, 34].
Contribution of nicotine and cue to self-administration
Both nicotine and the nicotine-paired cue contributed to self-administration behavior. The omission of the cue (CueOm) during session 13 (Fig. 2a) and the omission of nicotine (NicOm) during session 18 (Fig. 2b) led to modifications in instrumental responding when compared to the baseline, although in distinct ways. Specifically, the Global Effect of CueOm resulted in an approximately 40% reduction in self-infusions, while NicOm tended to amplify it [Omission type effect, F(1,61) = 83.7, p < 0.00001] (Fig. 2c). The Loading proportion increased in both conditions, with a higher loading effect in response to NicOm [Omission type effect, F(1,61) = 7.39, p < 0.01] (Fig. 2d, SI and FigS3). Large individual variations were observed in the effects of CueOm (FigS4a-b) and NicOm (FigS4c-d), including variations of opposed directions. These variations provide support for the hypothesis that individuals differ in how nicotine and the cue interact to influence their self-administration behavior.
The four variables of interest were normalized using Z-scores before applying dimensional and clustering methods.
Principal Component Analysis (PCA) - components underlying the four variables of interest
The suitability of our variables and dataset for PCA was confirmed through the Bartlett test (p < 0.001) and the Kaiser-Meyer-Olkin (KMO) value (> 0.5) (Table S1). Initially, a Pearson correlation matrix (Table S2) was employed to explore the relationships between the original variables prior to conducting the PCA itself. The Global Effect (GE) and Loading Effect (LE) variables were identified as non-redundant and capable of capturing different aspects of the omission effect. Notably, there was no proportional relationship between CueOm-GE and CueOm-LE (r2=-0.131, ns). Likewise, NicOm-GE and NicOm-LE displayed a weak association (r2=-0.379, p < 0.05). This observation was further corroborated by the PCA analysis, which revealed four components. Among the four components identified by the PCA, the inflection point on the scree plot and the eigenvalues indicated three primary components (F or Factors) that collectively explained about 88% of the total variance (F1: 42.5%, F2: 25.4%, and F3: 20%) (Fig. 2e). For both NicOm and CueOm, the two types of variables (GE and LE) exhibited disparate correlations with the factors (Fig. 2f left). Moreover, their loading patterns onto the main components were dissimilar, underscoring their differences (Fig. 2f right).
Analyzing the correlations between the original variables and the components, as well as the squared cosines of the variables, unveils distinct relationships (Fig. 2f). Specifically, F1 appears to capture the response to NicOm: this primary component (F1) displays substantial loadings for the two NicOm variables (Zscore_NicOm-GE and Zscore_NicOm-LE) (Fig. 2f right). F2 seems to encompass the loading effect regardless of the omission type: F2 is primarily associated with Zscore_CueOm-LE, while Zscore_NicOm-LE contributes nearly equally to both F2 and F1. Lastly, F3 seems to represent the global effect of CueOm, as indicated by its stronger correlation with Zscore_CueOm-GE.
Individual variations in the respective contribution of nicotine and the cue to self-administration: identification and characterization of two clusters (A and B) and relationships with the PCA factors
In order to explore whether distinct response patterns to CueOm and NicOm are present, we conducted Ascendant Hierarchical Clustering analysis (AHC) on the four normalized variables of interest. Our nicotine + cue dataset (n = 62) satisfied the requirements for minimum sample for clustering techniques (2d where d is the number of dimensions) [35]. AHC identified two clusters of individuals as the optimal choice, determined by the Hartigan method (Table S3 top). Cluster A included 39 rats, whereas Cluster B encompassed 23 rats. Analysis of variance reveals significant differences between the two clusters across the four variables of interest (Table S3 bottom). However, four rats classified in cluster A were subsequently reclassified due to receiving a negative silhouette score, which indicated misclassification.
While the two clusters differed in their response to CueOm and NicOm (Fig. 3a-d, FigS5, SI), they did not exhibit differences in the acquisition and maintenance of nicotine + cue self-administration (Fig. 3g-i).
We analyzed how the members of the two AHC-generated clusters contributed to the three main components isolated by the PCA. The members of the two clusters best segregated according to F1 coordinates, a component that translates the response to NicOm, as mentioned above (Fig. 3e). The two clusters differed by their fit with the three main factors [Cluster x Factor, F(2,120) = 5.58, p < 0.005], notably factors F1 and F2. As evaluated by the mean squared cosines, members of cluster B fitted better with F1 than with any other factors, while members of cluster A fitted equally with F1 and F2 (Fig. 3f).
The qualitative and quantitative differential effects of CueOm and NicOm observed in the two clusters (Fig. 3) suggest that their self-administration behaviour is supported by different interactions between nicotine and the cue. In Cluster A, both nicotine and the cue contribute to, and are necessary, to support the behavior. In Cluster B, the cue alone is capable of supporting self-administration behavior. While an initial "extinction burst" is evident, the time course of infusions eventually follows a pattern similar to the baseline (Fig. 3d). Although nicotine alone (CueOm) in Cluster B exhibits an extinction-like profile (increased loading proportion and decreased maximal infusions) (Fig. 3c), this effect is less pronounced compared to Cluster A (Fig. 3a).
Psychopharmacological features of Clusters A and B
Response to the reinforcing effects of nicotine: Progressive ratio and dose-response curve for nicotine
To evaluate the primary reinforcing effects of nicotine, we conducted progressive ratio (PR) and FR3 dose-response tests in CueOm conditions.
Rats from ClusterB exhibited a higher breakpoint for nicotine self-administration (Fig. 4a-left) [Cluster, F(1,59) = 16.34, p < 0.0005] and sustained responding throughout the PR session (Fig. 4a-right). Data was averaged over the two sessions of PR as the difference between clusters was similar in the two sessions [Cluster x Session, F(1,59) = 0.37, p = 0.54, not shown].
Additionally, the dose-response curve for nicotine self-administration at FR3 tended to be shifted upward in rats from cluster B with maintenance of self-administration for the lower nicotine dose (Fig. 4b). These findings support the results from the CueOm test (Fig. 3b,d) and loading on PCA factor F1, indicating that rats in Cluster B appear more sensitive to the reinforcing effects of nicotine.
Effect of altering the contingency between nicotine and cue (Disconnection test)
To examine the nature of the interaction between nicotine and cue during self-administration, we conducted a disconnection test. In this test, the cue and nicotine delivery were dissociated and delivered through the active hole and previously inactive hole, respectively (Fig. 4c). Rats learned the new rule during the first 20 min of the first session when only the nicotine hole was active and delivered infusions, while visits to the cue hole had no scheduled consequence. During these initial 20 minutes, there were no significant differences between the two clusters in terms of total responding [Cluster effect, F(1,22) = 0.004, p = 0.95] and the time course of responding [Cluster x Time, F(1,22) = 0,32,p = 0.58] (FigS6). Notably, both clusters reached the same level of responding for the hole delivering nicotine. For this first session, there were no significant differences between the two clusters (FigS7).
Over the next five test sessions, there were no significant differences in total responding between rats from the two clusters [Cluster effect, F(1,20) = 0.09, p = 0.76; Cluster x Session, F(4,80) = 0.76, p = 0.55]. However, they did differ in the distribution of responses per hole over sessions [Session x Hole x Cluster, F(4,80) = 2.69, p < 0.05]. In Cluster A, the cues earned from responding in the active hole were maintained over the five test sessions, as was the number of nicotine infusions earned from the previously inactive hole (Fig. 4d).
In Cluster B, the cues earned from responding in the active hole decreased over sessions, while the number of nicotine infusions earned from the previously inactive hole was not significantly affected (Fig. 4e). Thus, in Cluster A, the reinforcing effects of the cue appear to depend on nicotine (Fig. 3a), but do not require nicotine to be contingently delivered (Fig. 4d). In Cluster B, the cue seemed to act as a conditioned reinforcer secondary to its contingent association with nicotine delivery. As classically observed [36], these conditioned effects were extinguished over time.
To gain further insight into the timing of seeking cues and nicotine, we calculated two mean time intervals: one between each cue and the next nicotine infusion, and the second between each nicotine infusion and the next cue. The two clusters exhibited significant differences [Cluster effect, F(1,62) = 4.63, p < 0.05], and this difference was primarily driven by the INF-Cue interval [Cluster x Interval, F(1,62) = 4.06, p < 0.05] (Fig. 4f). In Cluster A, we observed a time-balanced distribution of cues and nicotine infusions. The Cue-INF and INF-Cue intervals were similar (Fig. 4f) and correlated with each other, supporting the hypothesis that nicotine and cue were spaced evenly (Fig. 4g) [r = 0.36, r2 = 0.13, p < 0.05, reaching r = 0.68, r2 = 0.47, p < 0.0001 after the outliers test (Grubbs test statistic = 3.84, p < 0.0005 for INJ-Cue interval) and recoding]. This profile is in accordance with the reinforcement-enhancement tracking the circulating levels of nicotine [37].
In ClusterB (Fig. 4f-4h), however, the time distribution of cues and nicotine infusion was unbalanced and depended on which occurred first. The mean INF-Cue interval was significantly longer than the Cue-INF interval and they were uncorrelated, suggesting that nicotine was more reinforcing than the cue.
Varenicline effect on nicotine seeking in Clusters A and B
Varenicline had a differential effect on seeking behavior in clusters A and B when nicotine was omitted [Cluster effect, F(1,53) = 38.84, p < 0,00001; Cluster x Treatment, F(1,53) = 8,85, p < 0.005] (Fig. 4i).
As mentioned earlier, during nicotine omission sessions (NicOm), Clusters A and B exhibited significant differences compared to their respective baselines. Cluster A showed a decrease in drug-seeking behavior, while Cluster B showed an increase (p < 0.001) (Fig. 3b,d & Fig. 4i). Varenicline reduced this increase (NicOm + Var) in Cluster B, but it had no significant effect in Cluster A (Fig. 4i-left). Varenicline acted in opposite ways on the two clusters [Cluster effect, F(1,53) = 8.85, p < 0.005], decreasing seeking in Cluster B and tending to increase it in Cluster A (Fig. 4i-right).