A striking feature of the brain is that, even in early sensory areas, the numbers of inputs from the senses are dwarfed by intrinsic feedback connections from within the brain itself. This contrasts sharply with classical models that see the brain as mostly responding to information from the outside world to generate behavioral responses.

As Marcus Raichle (the 2014 Kavli prize laureate in neuroscience) so elegantly phrases it, the intrinsic feedback connectivity is the Dark Matter in the quest to understand the brain. In a recent paper published online in Cerebral Cortex, Dr. WANG Wei's group at Institute of Neuroscience of Chinese Academy of Sciences used focally specific causal manipulation to study a key feedback connection in the visual system (the corticothalamic loop).

This massive reciprocal pathway from primary visual cortex (V1) back to the main thalamic visual relay, the lateral geniculate nucleus (dLGN), demonstrates that only 7% of the inputs to the relay cells come from eye, whereas more than 60% of the mono– and di–synaptic inputs to this "sensory" relay come from V1. Using microionophoresis with carefully controlled ejection currents of a metabotropic GABA_B antagonist, they were able to selectively enhance the visually driven response of neurons in a single column of the feedback layer (layer 6) of primary visual cortex (V1). Researchers performed simultaneous recording of the 2D receptive field (RF) of multiple neurons in the dLGN before, during, and after the focally specific modulation of the visual responses in layer 6 of V1.

Contemporary optogenetic studies of feedback in the mouse are largely constrained by current limitations of global optical stimulation, and have therefore yielded a wide diversity of potential effects. To truly understand sensory transformations, researchers need to be able to control the sensory–map topography of their perturbation.

In this study, researchers could precisely control the focal spatial parameters of the causal manipulation. They first found that enhancing the cortical feedback can cause both increases and decreases of response gain in simultaneously recorded dLGN relay neurons. This contrasts with earlier studies which found facilitation or suppression restricted to a retinotopic Mexican–hat arrangement.

They then discovered that the majority of the gain changes occurred only when retinotopic relationships between V1 and dLGN were parallel or orthogonal to the orientation preference of the V1 location. That cortical feedback drives both facilitation and suppression and is linked to its own sensory preferences enables a more flexible functional modulation of incoming sensory information.

Researchers also analyzed the fine spatial structure, including the RF size and position during cortical manipulation. They found that there were changes to RF position for some cells, which is consistent with the effects of spatial attention observed in higher cortical areas.

Predictive coding (the dominant variant of the "Bayesian brain") is a high–level network theory of brain function, applicable from synapticconnectivity to complex diseases such as schizophrenia, and provides a framework with which to understand the intrinsic connectivity patterns in the brain. Previous studies by the same authors have been influential for computational modelling of the visual system using predictive coding schemes. This study adds a further dimension to the potential computations enabled by top-down sensory pathways for models of visual perception.