Layer-specific control of cortical inhibition by NDNF interneurons

Accurate perception requires the integration of external (bottom-up) and internally generated (top-down) information. The main recipient of top-down projections in cortex is layer 1, which houses the dendrites of pyramidal cells (PC; Schuman et al., 2021). While layer 1 is devoid of excitatory cell bodies, it contains a variety of interneurons, including neurogliaform cells expressing neuron-derived neurotrophic factor (NDNF; Abs et al., 2018). NDNF interneurons are unique in that they provide slow inhibition, partially via non-synaptic volume release of GABA (Pardi et al. 2020). Yet, their contribution to cortical computations is still unclear. Here, we propose that NDNF interneurons control cortical inhibition in a layer-specific manner. Specifically, we suggest that NDNF-mediated volume release targets presynaptic GABA receptors at the outputs of somatostatin-expressing (SOM) interneurons in layer 1, leaving SOM outputs in lower levels unaffected. We demonstrate in a computational model how this mechanism gradually replaces SOM-mediated inhibition to PC dendrites with NDNF-mediated inhibition, which carries top-down rather than bottom-up information and is slower in time. The competition for dendritic inhibition stems from a mutual inhibition motif between NDNF interneurons and SOM outputs. Notably, it relies on presynaptic inhibition and does not require synaptic connections from NDNF to SOM interneurons. We show that the motif can become bistable such that top-down inputs to NDNF interneurons function as a switch for different circuit dynamics. Finally, we find that the connections of NDNF interneurons within the circuit introduce additional (dis-) inhibitory pathways, changing how the circuit responds to cell type-specific perturbations. Our model elucidates how NDNF interneurons restructure inhibitory circuitry in cortical layer 1. Because NDNF interneurons receive a broad range of top-down inputs, the model suggests a neural mechanism by which top-down information can modulate cortical processing on behaviourally relevant timescales.