ИСТИНА

Configural representations integrate multiple cues into a coherent context. People – as well as many animals – tend to rely on stimulus configurations rather than on single elements: this is highly adaptive since configural representations usually provide much more reliable predictions. Configural learning heavily relies on the hippocampus [e.g., Rudy and Sutherland, 1995; Sutherland and Rudy, 1989; Duncan, 2018; Stout et al., 2018]. Hippocampus is presumed to maintain configural representations via relational and spatial binding of related stimuli [Eichenbaum and Cohen, 2014; Monti et al., 2015]. In addition to the hippocampus, involvement of neocortical areas is assumed within the indexing hypothesis, according to which the hippocampus stores only relations between elements, which themselves are actually stored in neocortex [Teyler, DiScenna, 1986; Buzsáki et al., 2022]. The process of systems consolidation leads to the establishment of memory traces in the neocortex, with continued involvement of the hippocampus [Nadel et al., 2000]. Yet, little is known on neocortical involvement during the early stages of learning. fMRI studies of configural learning during the first day of learning report strong hippocampal activation – with some studies also mentioning neocortical activation, including lateral and medial prefrontal cortex, insular and frontal opercular cortex, anterior and posterior cingulate cortex, medial and lateral inferior parietal cortex [Baeuchl et al., 2015; Stout et al., 2019]. Electrophysiology, including electro- and magnetoencephalography, clearly testifies to some neocortical activity involved in configural learning. Still, such studies are scarce, and they do not fit into any unifying picture of configural learning. Additionally, virtually all previous studies of configural learning in humans relied on combinations of visual stimuli, making it resemble contextual learning of visual scenes [e.g., Stout et al., 2018; Kveraga et al., 2011]. This leaves out the general notion that the brain can support multimodal configural representations [Eichenbaum and Cohen, 2014; Monti et al., 2015]. For the current research, we used magnetoencephalography, a method with excellent temporal resolution and sufficiently good spatial resolution. We focused on the theta band, as theta oscillations are generally associated with memory formation [Takehara-Nishiuchi, 2000] and neuroplasticity [Buzsaki, 2002]. Theta-range synchronization between the hippocampus and the neocortex during learning and memory integration has been implied in a number of studies [Stout et al., 2018; Zielinski et al., 2020; Backus et al., 2016; Nardin et al., 2023; Benchenane et al., 2010]. The main question was to find whether theta oscillations may be indeed specifically involved in encoding of configural stimuli. To address this goal, we designed a novel experimental paradigm, which involved discrimination between intermodal (auditory and visual) configural and elemental stimuli, some of which were paired with aversive electrocutaneous stimulation, while others were not. The study involved 28 volunteer participants. Discrimination learning was successful for both configural and elemental stimuli, as shown by both explicit behavioral responses (pressing buttons to report anticipation of the aversive electrocutaneous stimulation) and implicit autonomic responses (anticipatory electrodermal activity). Magnetoencephalogram was continuously recorded while the participants were performing on the task. We found a robust and reliable theta-band power increase in response to reinforced configural stimuli compared to non-reinforced configural stimuli. The effects were found in dorsolateral and medial prefrontal cortex, cingulate cortex, medial temporal cortex, insula, left temporal-parietal-occipital junction, left inferior parietal cortex. Moreover, this increase was greater for reinforced configural stimuli than for reinforced elemental stimuli (both visual and auditory). These findings demonstrate that encoding of a configural stimulus during discriminative associative learning is associated with theta synchronization. Additionally, this study shows that multiple neocortical areas may be involved in configural encoding during the early stage of learning – i.e. before systems consolidation. The topography of the effects hints to associative areas involved in higher-order sensory processing, action planning and memory. In summary, this study appears to be the first to address multimodal configural learning in humans, and it is the first to provide clear evidence that cortical synchronization in theta range is involved in configural learning. We interpret the current findings as hinting to a possible mechanism for establishing network communication during stimulus encoding – in conditions, requiring multimodal feature binding.