Skip to Main content Skip to Navigation

Neuronal Activity Propagation in the Brain: From Neuronal Avalanches to Synfire Chains and Gamma Oscillations

Abstract : The brain is thought to be a highly complex and adaptive system that transforms sensory input into meaningful behavior. Processing of environmental information takes place in a highly distributed network of specialized areas that are believed to coordinate their dynamics into coherent spatiotemporal patterns of activity. This coordination requires efficient communication between brain areas and has been hypothesized to be accomplished by different mechanisms. In this doctoral work, we experimentally test in detail the presence of neuronal avalanches, a candidate mechanism for communication, in cortical activity of anesthetized and awake animals. Moreover, we suggest a novel theoretical framework to bridge the conceptual gap between synfire chains and gamma oscillations, two other putative hypotheses for neuronal communication, and test its validity in numerical simulations of neuronal networks. In a first study, we recorded spontaneous activity with multi-electrode arrays (16 electrodes) in the primary visual cortex of (halothane) anesthetized cats and searched for power laws and long-range correlations in local-field potentials (LFP) as well as spiking activity. These two statistics can be a signature of critical dynamics, which is referred to as neuronal avalanches in neural networks. In spiking activity, we found power laws in avalanche size distributions in four out of seven recordings and a consistent power law exponent of -1.8. In the same recordings, lifetime distributions and inter-spike interval distribution of population spike trains were better fit by a power law than an exponential distribution. In the other three datasets, these distributions were rather curved and were equally well described by power law and exponential statistics. Average correlation as measured by average pairwise cross-correlation across all units was stronger and extended over a longer period of time in datasets with power laws than in those where power laws were absent. Finally, we sub-sampled the spiking data by randomly removing spikes and thereby transformed power laws into curved distributions. Surprisingly, power laws were present in the LFP in six out of seven datasets, even though the underlying spiking activity did not show evidence for power law scaling. In summary, we showed that spiking activity can sometimes have features of critical dynamics, while at other times evidence for neuronal avalanches is absent. Our sub-sampling analysis demonstrated that insufficient recording of spikes can indeed destroy power law statistics. However, we alternatively propose the existence of different cortical states with a varying correlation structure as a potential cause for critical statistics in some, but not in other recordings. In a second study, we further pursued the cortical state hypothesis and recorded multi-electrode array data from the primary visual cortex of (isoflurane) anesthetized cats (32 electrodes) and one awake monkey (96 electrodes) in the dark. We developed a novel method to separate different cortical states within short time windows based on LFP power spectra using principal component analysis and k-means clustering. We thus identified up to five different states with different spectral profiles. In spiking activity, these states showed marked differences in their collective dynamics with synchronized states displaying population bursts followed by periods of silence and desynchronized states exhibiting continuous and asynchronous population activity. Neuronal avalanche analysis of these different spiking dynamics indeed revealed that desynchronized activity is associated with more curved distribution, while synchronized activity shifts avalanche distributions closer to power laws, as measured by higher values of the scale parameter of fitted lognormal distributions. The same result was found for inter-spike interval statistics of the population spike train. In contrast, neuronal avalanche analysis of the LFP always approached power law distributions, even though the size of the tail was slightly modulated by the cortical state. In addition, we showed that correlations between electrodes in the LFP recordings were much higher than in spiking activity. Finally, these results were similar across different datasets within a species and did not differ between the anesthetized and awake recordings. An important difference was an increase of population firing rate in monkey recordings with desynchronization as opposed to a decrease in the anesthetized cat cortex during desynchronized states. In conclusion, we demonstrated a modulation of tails in neuronal avalanche statistics by the cortical state in spiking activity of both anesthetized and awake preparations. These findings suggest the presence of both critical and non-critical fluctuation regimes in neuronal network. The modulation of tails was in general much weaker for LFP, which was on average more correlated than unit activity. The origin of this high correlation in the LFP is unclear and might be partly due to volume conduction. Sub-sampling may not explain the absence of power law in the monkey recordings, as the firing rate increases with further desynchronization of population spiking activity. In a third study, we study two other prominent theoretical frameworks for neuronal communication, the synfire chain model and the communication through coherence (CTC) hypothesis. In these models, communication between remote brain areas is established through two different types of synchrony, which have hitherto believed to be distinct from each other. In the synfire chain, synchrony is generated through common drive from previously firing neurons, while synchrony in the CTC model is generated through oscillations of local neuronal populations. In this work, we suggest a novel framework, in which coherent oscillation are a manifestation of synchrony spreading across a diluted synfire chain architecture which does not have strong enough connections to transmit synchrony across its layers. These weak connections are compensated for by amplification and synchronization of excitation during repeated cycles of an oscillation. The amplification is caused by progressive synchronization of inhibition within the oscillation, which in turn increases the gain of excitatory input. We introduce the concept of oscillation chains which represents the spread of a coherent oscillation across layers of a weakly connected synfire chain. We hypothesize that these oscillations trigger synaptic potentiation and transform the oscillation chain into a synfire chain, once the synaptic weights are sufficiently strong. Synfire chains may thus be regarded as a special case of oscillation chains. We tested this framework in layered networks of integrate and fire neurons and indeed found the presence of oscillation chains that supported the spread of synchrony with connections, whose strength was insufficient for synfire chain transmission. Further analysis revealed that oscillation chains can be converted to synfire chains by increasing synaptic weights and the emergence of oscillation chains depends on resonance properties of the network. Taken together, these results propose a new interpretation of communication through synchrony in neuronal networks and link coherent oscillations with synfire chains and synaptic plasticity. In summary, we used experimental techniques and modeling to study different modes of communication in neuronal networks. Experimentally, we found that neuronal networks display both critically synchronized population activity and desynchronized dynamics indicating that communication based on criticality may change as a function of cortical state. Furthermore, we show in a modeling study that synchrony created in synfire chains may be equivalent to synchronization created within coherent oscillations. Each of these two modes of communication is used for different strengths of synaptic weights and synaptic plasticity may be the key to convert oscillation dynamics into synfire chains.
Complete list of metadata
Contributor : Gerald Hahn Connect in order to contact the contributor
Submitted on : Tuesday, February 11, 2014 - 9:27:45 PM
Last modification on : Tuesday, June 23, 2020 - 10:46:02 AM
Long-term archiving on: : Monday, May 12, 2014 - 2:35:29 PM


  • HAL Id : pastel-00945214, version 1



Gerald Hahn. Neuronal Activity Propagation in the Brain: From Neuronal Avalanches to Synfire Chains and Gamma Oscillations. Neurobiology. Ecole Polytechnique X, 2013. English. ⟨pastel-00945214⟩



Record views


Files downloads