A unifying theory on the relationship between spike trains, EEG, and ERP based on the noise shaping/predictive neural coding hypothesis
Introduction
A critical issue in neuroscience has been cracking neural code used by spiking neurons in the brain (Shadlen and Newsome, 1994, Von der Malsburg, 1995, Arieli et al., 1996, Shin, 2002a). At one extreme, so-called ‘rate code’ (or signal-plus-uncorrelated noise model) says that both background EEG and temporal structure of spike train are uncorrelated noise which is just a nuisance for brain information processing (for review, see Arieli et al., 1996). The signal (or event-related potential (ERP)) is then recovered experimentally from the noise by averaging over repeated trials. This approach implicitly assumes that variability reflects ‘noise’, which uncorrected with signal and could be overcome by the brain by appropriate averaging over populations of neurons (Shadlen and Newsome, 1994, Arieli et al., 1996). However, Shin (2002a) argues that ‘rate code’ and the signal-plus-uncorrelated noise model resulted originally from misinterpretations (Von Neumann, 1958) of the quantization theory developed at 40 s (Bennett, 1948). At the other extreme, so-called ‘temporal code’ suggests that precise spike timing represents time-varying sensory, motor, or cognitive signals (Von der Malsburg, 1995, Shin, 2001a). According to this view, high frequency EEG components (e.g. gamma band fluctuation (30–100 Hz)) and irregular spike trains, even during spontaneous activity, is not noise but signal (Von der Malsburg, 1995) and the ERP is not a signal but a meaningless epiphenomenon resulting from ensemble averaging (Freeman, 2000b). Thus, both rate code and temporal code have failed to distinguish between signal and noise in spike train and EEG. As a result, neither rate code nor temporal code has successfully explained the causal relationship between spike train, EEG and ERP in a unified manner. In this report, firstly, I shall explain how the noise shaping/predictive neural coding hypothesis (Shin, 2002a) can provide a unified view on several issues related to neural coding such as irregularity, neural noise, synchrony, cortical gain control, and brain disorders (e.g. Epilepsy, Parkinsonian tremor, Alzheimer, and Schizophrenia). Secondly, based on the unified view, I shall propose causal relationships between theta oscillation, gamma band fluctuation, and P3(00) ERP responses. In addition, recent experimental results supporting the unified view shall be discussed.
Section snippets
Irregularity in spontaneous spike train: noise or signal
Cortical neurons in vivo usually fire irregularly in spontaneous and stimulus-evoked conditions, even when the nerve cells are thoroughly adapted to a steady stimulus of fixed amplitude. Why is it useful or necessary for these neurons to fire irregularly (Shadlen and Newsome, 1994, Arieli et al., 1996)? In the brain, the pathological condition usually has regular dynamics while the normal non-pathological condition has irregular dynamics (Degn et al., 1987). However, it is unknown why regular
Noise shaping neural coding
The most popular family of neuron models with spike-like behavior use variants of integrate-and-fire neuron models which have been used for modeling of a spiking cell with a long and distinguished history (Shin, 2001a). The output spike train of the integrate-and-fire neuron model is regular. The integrate-and-fire neuron transforms continuous somatic current signal into a frequency modulated spike train based on the regular carrier. Due to the regular carrier in the integrate-and-fire neuron,
Biologically realistic implementation of the noise shaping neural coding
The noise shaping neural coding can be implemented both in single neuronal level by spike-activated potassium currents/recurrent inhibitions and in network level by inhibitory synaptic couplings (Shin, 2001a, Shin, 2002a). Here, I shall discuss on three necessary components for implementing biologically plausible noise shaping neural coding: spike-activated potassium currents/inhibitions, temporal integration, and stochastic/deterministic noise.
Adaptive predictive coding and cortical gain control mechanisms in vivo
I have suggested that the brain may use both predictive coding and noise shaping coding in both cellular and network levels for efficient spatial and temporal information processing (for review, see Shin, 2002a). In general, noise-shaping coding conducts high-pass (or band pass) filtering of noise to reduce low-frequency signal band noise. But, predictive coding performs subtraction between signals and the predicted signals, equivalently, high-pass filtering of signal, to encode mainly changes
Noise shaping and predictive coding in spatial domain
I have so far focused on temporal aspect of neural coding: how can spiking neurons achieve an efficient precision for temporal information processing? However, the spatial aspect of neural coding is also important: how can spiking neurons achieve an efficient spatial precision for spatial information processing (e.g. receptive field properties, visual map and sensory-motor map)? The noise shaping and predictive coding in time domain can be extended to the spatial domain, which may be related to
Towards causal relationships between theta oscillation, gamma band fluctuation, and P3(00) ERP responses
It is widely believed that the human brain's electrical activity reflects higher cognitive functions such as attention, arousal, and even consciousness. However, it has been also suggested that many aspects of the generalized patterns of electrical activity of the hippocampal formation and neocortex (including theta oscillation and neocortical activation) are closely correlated with concurrent motor activity (Vanderwolf, 1975, Shin and Talnov, 2001). Interestingly, several investigators have
Conclusion
The origin and functional role of neural noise based on the noise shaping/predictive neural coding hypothesis is summarized as follows: (1) synaptic noise in vivo is related to the noise shaping/predictive neural coding; (2) the dominant frequency band and maximum power of the synaptic noise depends on firing frequency, noise shaping filter, membrane low-pass filtering characteristics, and synchronicity; (3) low frequency noise from irregular spontaneous spike train prevents from generating
References (61)
- et al.
Gamma, alpha, delta, and theta oscillations govern cognitive processes
Int. J. Psychophysiol.
(2001) A common network of functional areas for attention and eye movements
Neuron
(1998)A proposed name for aperiodic brain activity: stochastic chaos
Neural Netw.
(2000)Pulse frequency modulation in neural systems
Biophys. J.
(1971)Neuromodulation: acetylcholine and memory consolidation
Trends Cognit. Sci.
(1999)- et al.
Lack of the burst firing of thalamocortical relay neurons and resistance to absence seizures in mice lacking alpha(1G) T-type Ca(2+) channels
Neuron
(2001) - et al.
Aging cognition: from neuromodulation to representation
Trends Cognit. Sci.
(2001) Neural mechanisms underlying brain waves: from neural membranes to networks
Elctroenceph. Clin. Neurophysiol.
(1991)- et al.
Theta activity, virtual navigation and the human hippocampus
Trends Cognit. Sci.
(1999) - et al.
Noise neural codes and cortical organization
Curr. Opin. Neurobiol.
(1994)