Skip to main content

Predicting Sequential Learning from Oscillatory Activity in Human MEG-Data

Final Report Summary - PSLOAHMD (Predicting Sequential Learning from Oscillatory Activity in Human MEG-Data)

The aim of the present project is to investigate the functional relationship between sequence learning and the phase and amplitude of rhythmic brain activity in different frequency bands in humans. Brain rhythms are ubiquitous in the nervous system of many species and a growing body of evidence suggests that they play a functional role in higher cognitive processes such as attention and memory. The learning of sequences on the other hand is considered to underlie the acquisition of behavioral skills as it permits to link events together that unfold over time, thereby supporting the representation of causal relations between objects and events in our environment. The hippocampus has long been involved in learning and memory, thereby supporting that the hippocampus may play a causal role in the learning of sequences. This is supported by recent data from functional magnetic resonance imaging studies showing that the sequential order of learned information could be decoded from hippocampal activity recorded in humans. These data confirm the central role of the hippocampus in sequence learning, however, other brain regions are likely to play a central role in sequence learning as well and it remains unclear how the hippocampus interacts with other brain regions during sequence learning. Furthermore, another important question is how the temporal order of sequentially structured inputs is represented by neuronal activity. Neuronal oscillations are a prominent feature of hippocampal activity and sensory cortices and several electrophysiological studies in animals have implicated theta (5-7Hz) oscillations into memory and learning. In addition, emerging evidence from multi-site electrophysiological recordings in non-human primates suggest that beta (15-30Hz) and gamma (30-100Hz) oscillations occur in hippocampal and sensory circuits and that feedback and feed forward interactions could be dynamically routed through distinct frequency channels across the hierarchy of cortical layers. These data suggest that neuronal oscillations could provide an interesting mechanism to represent on one hand the sequential order of perceptual inputs in memory systems and on the other hand to route information between the hippocampus and other brain regions involved in sequence learning. The exact role of different frequency bands during sequence learning, however, remains unclear and remains to be examined.

The present study aims to address this question by examining the role of theta, beta and gamma oscillations during sequence learning in humans. Specifically, the present study will examine in depth how the phase of theta oscillations contributes to the consolidation of sequential inputs in memory, and whether the amplitude of beta and gamma oscillations are correlated with the prediction of expected events. Specifically, oscillatory activity was recorded in n= 22 healthy adult participants with magnetoencephalography (meg), which is a non-invasive surface recording technique that permits to track the synchronized activity of large neuronal populations with a temporal resolution in the sub-millisecond range and which offers a spatial resolution comparable to modern neuroimaging techniques such as functional magnetic imaging resonance (fMRI) in the millimeter range. Spectral analysis and beam forming techniques were used to extract and reconstruct the phase and amplitude dynamics of large-scale neuronal networks and to assess whether neuronal activity carried sequential information. To quantify sequence learning at the behavioral level, participants were asked to rate the familiarity of sequences of visual images that they had perceived during a previous learning phase. To relate the amplitude and phase of oscillatory activity in different frequency bands with sequence learning, we used spectral analysis techniques together with correlation analysis. This approach therefore allowed us to quantify the relationship between oscillatory neuronal activity that occurred while participants were learning sequences with the overall degree of familiarity with which they later rated the perceived sequences of images.

During the learning of sequences, we observed an elevation of activity at theta (5-7Hz), beta (15-35Hz) and gamma (60-80Hz) frequencies that was correlated with the familiarity of sequences and which supports the hypothesis that rhythmic neuronal activity may play a functional role during sequence learning. This is supported by several of our findings. First, a particularly strong correlation (r>0.6) between sequence familiarity and oscillatory activity was found in the beta frequency band. Second, in the gamma-band the power of learning-related activity was found to be negatively correlated (r= -0.4) with sequences familiarity. Third, the consistency with which the onset of sequential events was tracked by the phase of theta activity during sequence learning was also correlated with sequence familiarity (r = 0.4). Thus, the present findings suggest that oscillatory activity at theta, beta and gamma frequencies is functionally related to sequence learning thereby supporting the notion that the representation of sequences in human memory may rely on an oscillatory phase code.

Similarly, the opposite pattern of effects observed at beta and gamma frequencies is in line with recent evidence from electrophysiological studies in non-human primates suggesting that beta and gamma oscillations reflect different aspects of how the brain may predict expected events. Specifically, these data suggest that beta oscillations reflect the predicted activity that will be generated by an expected event, whereas gamma oscillations reflect the extent to which the predicted activity deviates from the actual activity generated by a stimulus. In agreement with these findings, one possible interpretation of the present results is that successful sequence learning may be correlated with optimized predictions that are reflected by elevated beta activity, whereas elevated gamma activity may reflect a higher level of prediction errors that could be associated with reduced sequence learning. Although encouraging, the present results require further validation by additional analyses to clarify the exact role of theta phase during sequence learning. These additional analyses are currently being conducted and the results are expected to be submitted for publication within the next six months.