En route from retina to cortex, visual information passes through the dorsolateral geniculate nucleus of the thalamus (dLGN), where extensive corticothalamic (CT) feedback has been suggested to modulate spatial processing. How this modulation arises from direct excitatory and indirect inhibitory CT feedback pathways remains enigmatic. Here we show that in awake mice, retinotopically organized cortical feedback sharpens receptive fields (RFs) and increases surround suppression in the dLGN. Guided by a network model indicating that widespread inhibitory CT feedback is necessary to reproduce these effects, we targeted the visual sector of the thalamic reticular nucleus (visTRN) for recordings. We found that visTRN neurons have large receptive fields, show little surround suppression, and exhibit strong feedback-dependent responses to large stimuli. These features make them an ideal candidate for mediating feedback-enhanced surround suppression in the dLGN. We conclude that cortical feedback sculpts spatial integration in dLGN, likely via recruitment of neurons in visTRN..

Neurons in the dorsolateral geniculate nucleus (dLGN) of the thalamus receive a substantial proportion of modulatory inputs from corticothalamic (CT) feedback and brain stem nuclei. Hypothesizing that these modulatory influences might be differentially engaged depending on the visual stimulus and behavioral state, we performed in vivo extracellular recordings from mouse dLGN while optogenetically suppressing CT feedback and monitoring behavioral state by locomotion and pupil dilation. For naturalistic movie clips, we found CT feedback to consistently increase dLGN response gain and promote tonic firing. In contrast, for gratings, CT feedback effects on firing rates were mixed. For both stimulus types, the neural signatures of CT feedback closely resembled those of behavioral state, yet effects of behavioral state on responses to movies persisted even when CT feedback was suppressed. We conclude that CT feedback modulates visual information on its way to cortex in a stimulus-dependent manner, but largely independently of behavioral state.

20En route from retina to cortex, visual information travels through the dorsolateral geniculate nucleus of the thalamus (dLGN), where extensive cortico-thalamic (CT) feedback has been suggested to modulate spatial processing. How this modulation arises from direct excitatory and indirect inhibitory CT feedback components remains enigmatic. We show that in awake mice topographically organized cortical feedback modulates spatial integration in dLGN by sharpening receptive fields (RFs) and increasing surround suppression. Guided by a network model revealing wide-scale inhibitory CT feedback necessary to reproduce these effects, we targeted the visual sector of the thalamic reticular nucleus (visTRN) for recordings. We found that visTRN neurons have large receptive fields, show little surround suppression, and have strong feedback-dependent responses to large stimuli, making them an ideal candidate for mediating feedback-enhanced surround suppression in dLGN. We conclude that cortical feedback sculpts spatial integration in dLGN, likely via recruitment of neurons in visTRN. 21 24 object recognition 1, 2 , inference of depth and 3D structure from 2D images 3, 4 and semantic segmentation 5, 6 . Inspired by the 25 early visual system, these deep convolutional neural networks process visual information feedforward through a hierarchy of 26 layers. These layers each contain small computational units that, similar to real neurons operating within their receptive field 27 (RF), are applied to small patches of the image and during learning acquire selectivity for certain visual features. Remarkably, 28 such feature maps found in artificial networks resemble those of real neurons 7-9 , and the activity of various layers along the 29 artificial feedforward hierarchy can predict responses of real neurons in various cortical visual areas 2,[9][10][11] . Besides providing a 30 framework for machine vision, the feedforward architecture is also powerful for biological vision, where the initial feedforward 31 sweep contains a significant amount of information, sometimes sufficient to drive perception [12][13][14] . 32 So why then, is feedback such a prominent and ubiquitous motif in the brain, where descending projections generally 33 tend to outnumber ascending afferents? For instance, most inputs to primary sensory cortical areas do not come from primary 34 thalamus, but from higher-order structures 15 . The same principle applies to cortico-thalamic (CT) communication: relay cells 35 in the dorsolateral geniculate nucleus (dLGN) of the thalamus receive only 5-10% of their synaptic inputs from retinal afferents, 36 whereas 30% originate from L6 cortico-thalamic (L6CT) pyramidal cells of primary visual cortex (V1) 16 . 37Similar to a lack of general consensus regarding the function of cortico-cortical feedback 17, 18 , how CT feedback influences 38 the representation of visual information remains poorly understood. Despite massive cortical input, dLGN RFs closely resemble 39 retinal RFs rather than cortical ones [19][20...

8Neurons in the dorsolateral geniculate nucleus (dLGN) of the thalamus are contacted by a large number of feedback synapses from cortex, whose role in visual processing is poorly understood. Past studies investigating this role have mostly used simple visual stimuli and anesthetized animals, but corticothalamic (CT) feedback might be particularly relevant during processing of complex visual stimuli, and its effects might depend on behavioral state. Here, we find that CT feedback robustly modulates responses to naturalistic movie clips by increasing response gain and promoting tonic firing mode. Compared to these robust effects for naturalistic movies, CT feedback effects were less consistent for simple grating stimuli. Finally, while CT feedback and locomotion affected dLGN responses in similar ways, we found their effects to be largely independent. We propose that CT feedback and behavioral state use separate routes to powerfully modulate visual information on its way to cortex.

Choosing an action in response to visual cues relies on cognitive processes, such as perception, evaluation, and prediction, which can modulate visual representations even at early processing stages. In the mouse, it is challenging to isolate cognitive modulations of sensory signals because concurrent overt behavior patterns, such as locomotion, can also have brainwide influences. To address this challenge, we designed a task, in which head-fixed mice had to evaluate one of two visual cues. While their global shape signaled the opportunity to earn reward, the cues provided equivalent local stimulation to receptive fields of neurons in primary visual (V1) and anterior cingulate cortex (ACC). We found that mice evaluated these cues within few hundred milliseconds. During this period, ;30% of V1 neurons became cue-selective, with preferences for either cue being balanced across the recorded population. This selectivity emerged in response to the behavioral demands because the same neurons could not discriminate the cues in sensory control measurements. In ACC, cue evaluation affected a similar fraction of neurons; emerging selectivity, however, was stronger than in V1, and preferences in the recorded population were biased toward the cue promising reward. Such a biased selectivity regime might allow the mouse to infer the promise of reward simply by the overall level of activity. Together, these experiments isolate the impact of task demands on neural responses in mouse cerebral cortex, and document distinct neural signatures of cue evaluation in V1 and ACC.