Olsen, SR. Bortone, DS. Adesnik, H. Scanziani, M. (2012) Gain control by layer six in cortical circuits of vision. Nature 483: 47-54.
This is from Massimo's lab. He is on my committee and one of my favorite scientists. Shawn Olsen is a friend, he's also really good.
L6 excitatory neurons project to superficial layers and primary sensory thalamic nuclei.
They target L6 excitatory neurons with genetic line, approx 65% of population is labeled. L6 neurons can be divided into two categories: apical dendrites that end in L4, and apical dendrites that end in L1. Project to the dLGN and nucleus reticularis (NRT). Non-expressing cells were morphologically distinct. Expressed ChR2, linear multi-channel probe to record across vertical depth.
Drifting gratings. Activation of L6 supressed activity throughout depth, reduced spontaneous activity. Photostimulation scaled the tuning curve (multiplicative gain), did not change preferred orientation, tuning width, or the orientation selectivity index in all other layers. Also photosuppressed L6 with Halo-rhodopsin and the tuning curves scaled up.
L6 activation suppresses activity in dLGN. Mainly through disynaptic inhibition onto dLGN relay neurons - recruitment of NRT inhibitory neurons and possibly local inhibitory interneurons in dLGN. Suppression of cortex as a whole (by photostimulating PV inhibitory neurons), strongly facilitated dLGN. (This negative feedback from cortex to dLGN is part of the idea that cortex is building a model, and sending back to LGN what it thinks the pixels should look like based on the model. This suppresses LGN in all of the places in which cortex's model is correctly predicting the pixels. The errors of the prediction are sent back to cortex to refine the model, and also will signal for plasticity).
L6 can inhibit all layers of cortex through local circuits without going through LGN. Mainly targets neurons within column, falls off as stimulation moves away from target neuron tangentially.
They estimated the transfer function from dLGN to V1 activity to assess the relative inhibitory effects of the two pathways. If the suppression of V1 is larger than predicted by this transfer function, then the rest of the suppression can be attributed to the local circuit.
So, a big key to how cortex works is the feedback control of multiplicative-like inhibition and additive-like excitation. The excitatory neurons represent the information through their tuning curves, and the population is normalized by the inhibitory neurons. The mulitplicative normalization allows the neurons to have the same population-code, but can stabilize the activity across cortex.