The extraction and application of temporal structures: a fMRI study

1075 

Abstract Submission 

Linda Sempf¹, Peter Vavra¹, Toemme Noesselt^1,2

¹Institute of Psychology, Otto-von-Guericke-University, Magdeburg, Saxony-Anhalt, ²Center for Behavioral Brain Sciences, Magdeburg, Saxony-Anhalt, Germany

Linda Sempf  
Institute of Psychology, Otto-von-Guericke-University
Magdeburg, Saxony-Anhalt

Peter Vavra  
Institute of Psychology, Otto-von-Guericke-University
Magdeburg, Saxony-Anhalt

Toemme Noesselt  
Institute of Psychology, Otto-von-Guericke-University|Center for Behavioral Brain Sciences
Magdeburg, Saxony-Anhalt|Magdeburg, Saxony-Anhalt, Germany

The comprehension of recurrent temporal structures in our environment allows us to recognize causal relationships between events and build expectations about the timing of future ones. Previous studies on temporal attention have often used symbolic cues to indicate the probability of a target to appear at a particular moment in time. However, temporal expectations can also be based on unconsciously perceived probability distributions - for instance, a recurrent temporal interval between two stimuli. After statistical learning, the first of two stimuli can then be used as a meaningful cue to predict the moment when the second stimulus will occur. While this memory-based predictive mechanism is often used to optimize everyday behaviour and has been studied behaviorally, it is not fully understood, how the brain recognizes, extracts, and neuronally represents temporal relationships and utilizes them to guide attention to specific points in time. Here, we characterized the neural underpinnings of statistical learning and utilizing temporal structures with functional magnetic resonance imaging (fMRI).

In a longitudinal fMRI study participants (n=36) were exposed to audiovisual cue-target combinations. Targets appeared either at early (500ms) or late (1500 ms) time points after the cue and the probability of the two cue-target intervals was manipulated block-wise (1:4 or 4:1, early : late ratio). The two blocks alternated and each consisted of a likely and an unlikely interval (e.g. early likely and late unlikely). On day 1 participants passively observed the AV-combinations inside the scanner (4 runs), on day 2 and 3 they were trained on visual target detection outside the scanner and on day 4 they first passively observed the parings (4 runs) before they actively detected targets again inside the scanner (4 runs). fMRI-data (voxel size: 2.2 mm3, 66 slices, 250 volumes/run) was collected on a 3T Prisma scanner. After preprocessing (slice timing, realignment, normalization, smoothing) and first level modelling (with hemodynamic response function plus derivatives for all experimental conditions (early/ late & likely/ unlikely)) and realignment parameters) second level models (flexible factorial designs) were used for population level inferences.

Behaviorally, participants responded faster to early targets when they were more likely confirming that temporal expectations were formed to guide behaviour. Neurally, targets in the early-likely context elicited enhanced responses in the left hippocampus (CA1 region) already during passive viewing on day 1. During the active task left inferior parietal lobule (IPL), bilateral prefrontal gyrus and retrosplenial cortex (RSC) showed enhanced fMRI-responses for likely>unlikely trials. In IPL this effect was more pronounced for likely early than likely late targets. In addition, the fMRI response in this area correlates significantly with subject-specific mean response times. In contrast, unlikely>likely events modulated visual areas and right temporoparietal junction, in accord with previous findings on prediction-error processing.

Our results indicate that the processing of temporal context and potentially the extraction of temporal structures is initially associated with activity within the CA1 hippocampal subregion, even during passive observation. In keeping with this, neurophysiological animal studies have reported that the CA1 region contains time-sensitive neurons. When temporal expectations have been established, IPL and RSC are more engaged during likely trials, reflecting the use of temporal expectancies during active task performance. Similar activation patterns have also been observed during episodic memory processing, as confirmed by an association test (Yarkoni et al., 2011). These results conform with the notion of an engagement of episodic memory when learning and using temporal information (Frings et al., 2020).

Space, Time and Number Coding

Long-Term Memory (Episodic and Semantic) ¹

Activation (eg. BOLD task-fMRI)

Attention: Auditory/Tactile/Motor ²

Cognition

FUNCTIONAL MRI

Learning

Other - Time Perception, Episodic Memory