Control of the temporal interplay between excitation and inhibition by the statistics of visual input

Abstract

IntroductionIn the primary visual cortex (V1), single cell responses to simple visual stimuli (drifting gratings) are usually strong but with a high trial-by-trial variability. In contrast, when exposed to full field natural scenes with simulated eye movements, the firing patterns of these neurons are sparse but highly reproducible over trials [1]. So far the mechanisms behind these two distinct different response behaviours are not yet fully understood. Different mechanisms are candidates for controlling spike timing precision and models are needed to clarify their respective contribution, which may be of thalamic or intracortical origin. As a first step, we investigated which aspects of the neuronal dynamics can be explained by balanced feedforward excitation and inhibition and its dependency upon the spatiotemporal statistics of the different stimuli. We built a simple model of the early visual system (LGN, V1). The thalamocortical connectivity was similar to the gpush-pullh architecture used in [2], with additional depressing thalamocortical synapses [3]. The model was written in PyNN [4] using NEST [5] as simulator. Indeed, the model can reproduce the main response characteristics of firstorder thalamo-cortical neurons in layer 4 of cat V1. During drifting gratings, excitatory and inhibitory conductances of cortical neurons were anti-correlated [6,7], such that excitation can be freely integrated and induce multiple spikes. In contrast, during natural scenes the conductances were correlated, with inhibition lagging by few milliseconds [1,8]. This small lag between excitation and inhibition induces a strong selectivity to synchronous inputs, with a consequence that the responses became sparse and precise. However, some key aspects of the in vivo recordings in cat area V1 cannot be explained, such as selective reduction of stimulus-locked subthreshold noise during natural scene viewing, precise firing during fixational eye-movements and center-surround non-linearities, opening the door for future investigation about the role of intra-cortical recurrent connectivity in further shaping the neuronal responses to natural images. In conclusion, our study points that correlated inhibition can explain, at least in part, sparse and reliable spiking activity as observed in response to natural scenes. This is consistent with its role reported from other sensory modalities and cortical areas [8]. Thus correlated excitation and inhibition could be a general mechanism to induce sparse and precise spiking in the nervous system.

Functional consequences of correlated excitatory and inhibitory conductances in cortical networks

“…In particular, our study suggests that the brain takes advantages of a simple architecture to filter, process and propagate synchronous events in complex networks. By investigating how these properties are important in a more functional context, such as processing in the visual system (Kremkow et al 2008b), future work will be able to address the functional impact of such architecture onto the dynamics of information processing.…”

“…In the visual system, balancing feedforward inhibition was proposed as a candidate mechanism to explain the invariant orientation tuning observed with drifting gratings (Troyer et al 1998). Interestingly, it was shown in vivo (Marre et al 2005;Haider et al 2010) and in a model of the early visual system (Kremkow et al 2008b) that conductances are indeed correlated during natural viewing. In the model, the correlation of the conductances was related to the correlation of the contrast polarity in the receptive fields of cortical layer 4 neurons.…”

Statistical Learning Across Development: Flexible Yet Constrained

“…Teinonen et al (2009) examined statistical learning in sleeping newborns by presenting a continuous stream of three-syllable words in an artificial language similar to that employed by Saffran et al (1996a), in which the only cues to word boundaries were the conditional relations or frequencies of co-occurrence between syllables. Using electroencephalography, they measured newborns’ event-related potential (ERP) negativities to the first, second, and third syllables in the words.…”

Neural Processing of Repetition and Non-Repetition Grammars in 7- and 9-Month-Old Infants

“…Even before they begin speaking, however, research has shown that infants can pick up on important and complex linguistic cues from their auditory input, including rhythmical cues (Mehler et al, 1988), phonemic contrasts (e.g., Werker and Tees, 1984; Kuhl et al, 2006), structural regularities (e.g., Christophe et al, 1994), and transitional probabilities (e.g., Saffran et al, 1996; Teinonen et al, 2009; for a review of early infant language development, see Gervain and Mehler, 2010). …”