Delta rhythms as a substrate for holographic processing in sleep and wakefulness

Abstract

We initially considered the theoretical properties and benefits of so-called holographic processing in a specific type of computational problem implied by the theories of synaptic rescaling processes in the biological wake-sleep cycle. This raised two fundamental questions that we attempted to answer by an experimental in vitro electrophysiological approach. We developed a comprehensive experimental paradigm based on a pharmacological model of the wake-sleep-associated delta rhythm measured with a Utah micro-electrode array at the interface between primary and associational areas in the rodent neocortex. We first verified that our in vitro delta rhythm model possessed two key features found in both in vivo rodent and human studies of synaptic rescaling processes in sleep: The first property being that prior local synaptic potentiation in wake leads to increased local delta power in subsequent sleep. The second property is the reactivation in sleep of neural firing patterns observed prior to sleep. By reproducing these findings we confirmed that our model is arguably an adequate medium for further study of the putative sleep-related synaptic rescaling process. In addition we found important differences between neural units that reactivated or deactivated during delta; these were differences in cell types based on unit spike shapes, in prior firing rates and in prior spike-train-to-local-field-potential coherence. Taken together these results suggested a mechanistic chain of explanation of the two observed properties, and set the neurobiological framework for further, more computationally driven analysis. Using the above experimental and theoretical substrate we developed a new method of analysis of micro-electrode array data. The method is a generalization to the electromagnetic case of a well-known technique for processing acoustic microphone array data. This allowed calculation of: The instantaneous spatial energy flow and dissipation in the neocortical areas under the array; The spatial energy source density in analogy to well-known current source density analysis. We then refocused our investigation on the two theoretical questions that we hoped to achieve experimental answers for: Whether the state of the neocortex during a delta rhythm could be described by ergodic statistics, which we determined by analyzing the spectral properties of energy dissipation as a signature of the state of the dynamical system; A more explorative approach prompting an investigation of the spatiotemporal interactions across and along neocortical layers and areas during a delta rhythm, as implied by energy flow patterns. We found that the in vitro rodent neocortex does not conform to ergodic statistics during a pharmacologically driven delta or gamma rhythm. We also found a delta period locked pattern of energy flow across and along layers and areas, which doubled the processing cycle relative to the fundamental delta rhythm, tentatively suggesting a reciprocal, two-stage information processing hierarchy similar to a stochastic Helmholtz machine with a wake-sleep training algorithm. Further, the complex valued energy flow might suggest an improvement to the Helmholtz machine concept by generalizing the complex valued weights of the stochastic network to higher dimensional multi-vectors of a geometric algebra with a metric particularity suited for holographic processes. Finally, preliminary attempts were made to implement and characterize the above network dynamics in silico. We found that a qubit valued network does not allow fully holographic processes, but tentatively suggest that an ebit valued network may display two key properties of general holographic processing.