Skip to main content

Temporal predictions in the auditory cortex: neural mechanisms and their specificity across species and stimulus domains

Periodic Reporting for period 2 - BrainPredictDynamics (Temporal predictions in the auditory cortex: neural mechanisms and their specificity across species and stimulus domains)

Reporting period: 2020-09-01 to 2021-08-31

A recent paradigm shift in neuroscience has brought a substantial change in the way we see the brain. Instead of viewing it as an immensely complex input-output transformation organ, there is a growing consensus that its overarching function is to predict its immediate environment and minimise surprise. While the implementation of predictive processing in the brain is still being elucidated, a key challenge is to provide a unifying mechanistic explanation of neural expectations ranging from microscopic single-cell physiology, via models of interacting populations of neurons, to macroscopic signals observed non-invasively. To ground the theoretical claims that prediction propagation is a general principle of brain function, we need to understand the commonalities and boundary conditions of expectation signals across species, multiple domains (e.g. expectations of content, timing, location), and levels of complexity, from basic sensory processing to higher cognition.
Temporal predictions, in particular, are a current frontier in investigating the predictive brain capacities. While the importance of timing in perception was early recognised as a key determinant of behaviour, only recently has it been brought back to the centre of neuroscience. Auditory and speech perception are the optimal testing ground for studying temporal dynamics of predictive processing in the brain, as most auditory signals only gain meaning as temporal sequences. Thus, the current project will address two key challenges: (1) to test competing theories of temporal expectation mechanisms in the auditory cortex, using invasive electrophysiology in rodents and non-invasive human neuroimaging, and (2) to establish whether basic mechanisms observed in rodents and humans generalise across stimulus domains, from simple acoustic tones to complex hierarchical speech signals.
Work carried out in the period covered by the report includes:
(1) Analysis of the neural correlates of temporal expectation in direct, in vivo microelectrode MUA and LFP recordings from auditory cortex in rats – based on neural omission responses to acoustic noise bursts presented at isochronous rates
(2) Mapping and quantification of the regional specificity of neural signatures of expectations along the auditory cortical hierarchy, using EEG in humans, and applying multivariate decoding techniques to infer the source of expectations based on the spatial pattern of EEG responses – manuscript published (Frontiers in Human Neuroscience)
(3) Decoding contents of expectations from LFP recordings in the auditory cortex of rats – results shared in a peer-reviewed article (Luo et al., Current Research in Neurobiology)
(4) Complete acquisition of a MEG dataset (N = 24) to measure the effect of expectation on the cortical processing of linguistic materials (phonemes, syllables, words) at the Max Planck Institute for Empirical Aesthetics, Frankfurt – delayed due to Covid-19; completed in March 2021
(5) Complete acquisition of a EEG dataset (N = 22) to measure the effect of expectation on the cortical processing of linguistic materials (phonemes, syllables, words) at City University of Hong Kong – delayed due to Covid-19; completed in March 2021

The results of the project have addressed both objectives and yielded novel insights in the neural mechanisms of temporal predictions, and the generalisation of mechanisms subserving expectation signalling across species (rats vs. humans) and stimulus domains (basic tones vs. complex linguistic stimuli):
(1) Using non-invasive recordings in awake humans performing an auditory task, we found that temporal predictions based on the rhythmic structure of the stimulus stream modulate neural activity in a manner consistent with classical gain control mechanisms. The results of this study were disseminated in a peer-reviewed publication (Auksztulewicz et al., J Neurosci 2019).
(2) Using invasive recordings in anaesthetised rats exposed to rhythmic auditory stimuli, we found that violations of temporal predictions (by omitting a stimulus expected at a given moment in time) result in omission-related activity in the auditory cortex. The results of this study were disseminated at an international conference (SFN 2019).
(3) Regarding the generalisation of mechanisms of expectation signalling across species, in two homologous studies (both disseminated in peer-reviewed publications) we have shown that prediction errors to violations of different auditory expectations is encoded in distributed population-level activity both in humans (An et al., Front Hum Neurosci 2021) and rats (An et al., Hear Res 2020).
(4) Regarding the generalisation of expectation signalling across stimulus domains, we found that the key findings are also preserved – specifically, across two studies in rats using more complex speech-like sequences of sounds, we found that differences in neural activity between predictable and unpredictable sounds were only inferred through multivariate analysis (decoding) but not through conventional univariate analysis, suggesting that also more complex stimulus predictions are encoded in a distributed manner (Luo et al., Curr Res Neurobiol 2021; Cappotto et al., Biorxiv 2021). However, some differences between species were also observed. In a currently analysed dataset of non-invasive recordings obtained in awake humans exposed to complex speech-like sequences of sounds, we found that temporal expectations embedded in sound sequences modulate low-frequency neural activity at a rate specific to the temporal regularities of the speech streams (data to be disseminated at ICON 2022 conference in Helsinki, postponed due to Covid-19). On the other hand, no such modulation was found in the auditory cortex of anaesthetised rats exposed to similar stimuli, suggesting differences between species (rats vs. humans) and/or attentional states (wakefulness vs. anaesthesia). This shows that while some principles of expectation signalling (distributed encoding) generalise across both species and stimulus domains, others (low-frequency neural entrainment) might not.
The project yielded novel scientific insights into the generalisation of neural mechanisms of predictions across species and stimulus domains, resulting in 4 peer-review publications and 2 preprints, despite delays related to the Covid-19 pandemic. Going beyond the state of the art, the research conducted in this action has allowed for a better understanding of the neural mechanisms of temporal predictions in terms of neural gain modulation in sensory regions; it has brought new evidence for the distributed nature of prediction signalling in cortical circuits in both humans and rodents; and it has concluded that predictive information can be decoded from neural activity across species and stimulus domains (simple tones vs. speech sequences). The project has also enabled training junior scientists (3 doctoral students co-supervised by the Global Fellowship recipient Dr. Ryszard Auksztulewicz) and a dissemination of project results through scientific conferences and popular events (e.g. arts and science panels at the Hong Kong Arts Centre and the art gallery Savvy Contemporary Berlin).
Example of an experiment used to address the scientific questions pertaining to this research grant.