Any early graduate student can recite the rules of the canonical cortical microcircuit, which seem to be replicated throughout the brain. Input recipient layer 4 (L4) sends projections to layers 2/3 (L2/3), which in turn drives layer 5 (L5). This working model exists with good reason. And yet seemingly every month (including this one) we read of departures from this standard.

On page 1631 of this issue, Pluta et al. demonstrate another non-traditional circuit in early sensory cortices of mice. Employing a remarkable combination of in vivo and in vitro extracellular and intracellular recordings along with optogenetic manipulation, the authors trace a clear and substantial pathway between L4 and L5 that bypasses superficial layers.

Optogenetic manipulation of L4 activity in vivo modulated responses in L2/3 in a way that is consistent with the canonical cortical circuit: suppression of L4 reduced L2/3 responses, whereas activation of L4 elevated L2/3 responses. These same manipulations concurrently resulted in the opposite change in L5, counter to what the conventional model would predict. Two major possibilities could potentially explain this sign flip: L2/3 cells might project onto local L5 inhibitory neurons or L4 might project directly to L5 inhibitory neurons.

The authors next made paired in vitro intracellular recordings to demonstrate direct connections between L4 pyramidal cells and L5 fast-spiking cells in somatosensory cortex. The authors used focal photo-stimulation to systematically map the laminar profile of the excitatory drive onto L5 fast-spiking cells. Although every cell measured received excitatory drive from within L5, there was also a significant input from L4 (see image). In fact, a substantial fraction of cells received their strongest input from L4. There was virtually no drive from L2/3 and, in fact, disynaptic inhibition from L4 to L5 was maintained even if L2/3 was completely removed from the preparation.

Taken together, these results strongly support a direct input from L4 pyramidal cells to L5 fast-spiking cells. This suggests a revision of the model of signal flow through the cortical layers: sensory input propagates through L4 to drive L2/3 while simultaneously suppressing L5. The authors provide data suggesting that this newly identified circuit serves to sharpen tuning in the spatial domain. We might next ask whether the L4 disynaptic inhibitory connection to L5 is also present in non-sensory cortical areas, and, if so, what computation might it serve?