Valenced tactile information is evoked by neutral visual cues following emotional learning

Abstract

Learning which stimuli in our environment co-occur with painful or pleasurable events is critical for survival. Previous research has established the basic neural and behavioral mechanisms of aversive and appetitive conditioning; however, it is unclear precisely what information content is learned. Here we examined the degree to which aspects of the unconditioned stimulus (US)-sensory information versus affective salience-are transferred to the conditioned stimulus (CS). To decode what stimuli features (e.g., valence vs. discriminative somatosensation) are represented in patterns of brain activation elicited during appetitive (soft touch) and aversive (painful touch) conditioning to faces, a novel approach to using modeling with representational similarity analysis (RSA) based on theoretically driven representational patterns of interest (POIs) was applied to fMRI data. Once associations were learned through conditioning, globally, the CS reactivated US representational patterns showing conditioning-dependent reactivation in specific high-order brain regions: In the dorsal anterior cingulate cortex, the CS reactivated patterns associated with the affective salience of the US-suggesting that, with affective conditioning, these regions carry forward the affective associations of the experience.

Keywords: associative learning; conditioning; emotional learning; representational content; representational similarity analysis.

Fig. 1.

(A) Both experimental tasks (appetitive and aversive conditioning) followed the same general structure in which seven Unpaired Blocks were interleaved with six CS–US Paired Blocks. Each block contained one presentation of each of the three different stimuli: 2x CS+, CS-. In Unpaired Blocks, faces were presented by themselves, while in Paired Blocks, the CS+ faces were paired with appetitive brush or aversive pressure. This design with blocks of Unpaired CS was chosen to be able to extract representational patterns of CS that are not confounded by US presentation (Visser et al., 2013,2015). (B) Patterns of interest (POIs) demonstrate the representational pattern that would be observed in the experimental data if a region of interest (ROI) perfectly represented the theoretically derived constructs. Each POI is a similarity or correlation matrix with six different experimental conditions (2x CS+, CS- each in the appetitive and aversive task). Since the POIs represent theoretically derived constructs in a perfect way, correlations are either ± 1 or 0 (see legend for color coding). For a detailed description of all POIs, seeTable 1. (C) In a first analysis step, Paired blocks were analyzed such that for each ROI, the pattern of voxel beta weights in response to different task conditions (2x CS+, CS-) was extracted. Subsequently, patterns of voxel activation from one stimulus condition were correlated with those patterns obtained from each other task condition using representational similarity analysis (RSA) in order to obtain one similarity matrix for each ROI. Bayesian Information Criterion (BIC) was then used to find the combination of POIs that best fit the data extracted for each ROI. Finally, multiple regression was used to obtain beta coefficients for each POI in order to determine their individual contributions to the ROI activation pattern (Kryklywy, Ehlers, et al., 2023). (D) In the present study, in a separate analysis step, RSA was also performed on CS-only data. Again, voxel beta weights obtained in different task conditions were extracted for the different ROIs before correlating voxel activation from different task conditions using RSA and examined across early, mid, and late Unpaired trials in order to be able to look at similarity patterns over temporal development. In this Unpaired Block analysis: (1) For global US reactivation, for each given ROI, correlations between the reconstructed similarity matrix obtained during the Paired Block analysis (Fig. 1D) and similarity matrices obtained for early, mid, and late Unpaired conditioning trials were subject to a repeated measures ANOVA. (2) For US component reactivation, we compared the contribution (i.e., beta weights) of any identified POI to the ROI similarity pattern obtained from Paired Block data and Unpaired Block data in separate Bayesian linear models.

Fig. 2.

Visualization of the eight regions of interest (ROIs) included in the current analyses. (A) Sensory regions of interest include bilateral primary somatosensory cortex (pink), the primary/secondary visual cortex (yellow), and ventral visual structures (blue). (B) Integrative regions of interest include the amygdala (pink), the ventromedial prefrontal cortex (yellow), the dorsal anterior cingulate cortex (blue), as well as the anterior (green) and posterior (cyan) insula.

Fig. 3.

Stimulus ratings (likeability and trustworthiness) for CS+ and CS- stimuli pre- and postappetitive (A, B) and aversive (C, D) conditioning. The violin plots show the distribution of data points in the sample. The box plot contains the median as well as 25th and 75th percentiles. Outliers are printed as individual data points.

Fig. 4.

Pattern of Interest (POI) reactivation by reconstructed US (rUS) as well as early, mid, and late Unpaired data in different regions of interest (ROIs). Displayed are beta weights and 95% credible intervals for the reactivation of the combination of POIs that was identified to best fit the pattern of activation elicited by the US (seeFig. 1C). Two separate Bayesian linear models were performed on rUS and CS data, respectively, with the illustrated POIs as predictors. The contribution of any given POI at any given time point of CS-only conditioning is considered consistent with that POIs contribution to rUS data if the point estimate of the CS-only data (in red) falls within the credible interval of the rUS data (in blue). Consistent contribution is further indicated in bold font for the obtained beta weigh. Following the same logic, a point estimate of any given Unpaired conditioning time point that falls outside the credible interval of another Unpaired conditioning time point is considered to indicate statistical difference between those. S1 = primary somatosensory cortex; V1 = primary/secondary visual cortex; VVS = ventral visual structures; vmPFC = ventromedial prefrontal cortex; dACC = dorsal anterior cingulate cortex; aIns = anterior insula; pIns = posterior insula