Synthesis of higher order feature codes through stimulus-specific supra-linear summation

Abstract

11 How cortical circuits build representations of complex objects is poorly understood. The 12 massive dimensional expansion from the thalamus to the primary sensory cortex may enable 13 sparse, comprehensive representations of higher order features to facilitate object 14 identification. To generate such a code, cortical neurons must integrate broadly over space, yet 15 simultaneously obtain sharp tuning to specific stimulus features. The logic of cortical 16integration that may synthesize such a sparse, high dimensional code for complex features is 17 not known. To address this question, we probed the integration and population coding of 18 higher order stimuli in the somatosensory and visual cortices of awake mice using two-photon 19 calcium imaging across cortical layers. We found that somatosensory and visual cortical 20 neurons sum highly specific combinations of sensory inputs supra-linearly, but integrate other 21 inputs sub-linearly, leading to selective responses to higher order features. This integrative 22 process generates a sparse, but comprehensive code for complex stimuli from the earliest 23 stages of cortical processing. These results from multiple sensory modalities imply that input-24 specific supra-linear summation may represent a widespread cortical mechanism for the 25 synthesis of higher order feature codes. This new mechanism may explain how the brain 26

Mapping the integration of sensory information across fingers in human sensorimotor cortex

“…More broadly, our results are also in agreement with recent evidence from multi-whisker stimulation studies in rodents, where multi-whisker combinations are uniquely represented by neural populations in the rodent barrel cortex (Laboy-Juárez et al, 2019;Lyall et al, 2020).…”

Sensing and processing whisker deflections in rodents

“…This sensory information is then projected up to layers II and III for further processing (along with other local cortical areas), then down to layers V and VI for final output to more distant cortical areas, such as motor cortex (as well as sending feedback to sub-cortical areas) ( Radnikow, Qi & Feldmeyer, 2015 ). Layer IV excitatory cells typically have strong, narrow tuning to single whiskers while cells in supra- and infra-granular layers typically show broader, mixed-strength tuning (indicating tuning to more precise, higher-order features, and possibly common to multiple whiskers, as indicated by generally narrower receptive fields in layer IV compared to other layers ( Brecht, Roth & Sakmann, 2003 ; Brecht & Sakmann, 2002 ) and more complex sensory information generally being computed and integrated in cortex in layers other than layer IV ( Bale & Maravall, 2018 ; Lyall et al, 2020 ; O’Herron et al, 2020 )). Neurons across all layers, but particularly infragranular layers, can be tuned to temporal or qualitative features of whisker deflection, e.g., directional sensitivity or initial versus sustained parts of deflection.…”

“…On average, most of these interneurons will synapse onto three to six pyramidal neurons in layer II and III and make similar kinds and numbers of connections with layer V pyramidal neurons. These interneuronal connections onto layer II and III pyramidal cells, in combination with the intricate excitatory connection patterns from layer IV and thalamus, allows cells to be finely tuned to complex, higher-order features of sensory input ( Bale & Maravall, 2018 ; Lyall et al, 2020 ; O’Herron et al, 2020 ).…”

Recurrent pattern completion drives the neocortical representation of sensory inference

“…Although the classical definition of IC encoding neurons are neurons whose RF is contained entirely within the gap region of the IC 8,9 , neurons with RFs that overlap one of the inducer segments could still detect the presence of the IC. Recent work in the mouse visual system has demonstrated that V1 cortical neurons are highly sensitive to the specific content of 'extraclassical' stimuli, i.e., contextual information outside of their classical RFs [14][15][16] . No study has yet addressed whether ICs might represent a particularly important class of such extra-classical stimuli.…”