A computational model investigating the plausibility of generating precisely timed and reliable bursts in songbird nucleus HVC using the inputs from the thalamic nucleus UVA

Zebra finch song is driven by precisely timed and highly reliable bursts of projection neurons in the sensory-motor nucleus HVC (proper name) that form ultra-sparse burst sequences. The neural mechanism for the bursts is currently debated. One view is that bursts are generated in HVC through a synaptic chain mechanism (Long et al., 2010 and Egger et al., 2020). An alternative is that the thalamic nucleus Uvaeformis (UVA) drives HVC bursts as part of a distributed network through the brainstem, UVA, HVC, and RA (Hamaguchi et al., 2016). Here we computationally assess the plausibility of the distributed model. Single neuron recordings in UVA during singing show that UVA neurons that project to HVC contain timing information during the song, but compared to HVC projection neurons, fire densely in time and are much less reliable. To examine if convergence of UVA projection neurons to HVC can produce the precision and reliability of the HVC bursts, we use a resampling technique to scale up the number of UVA neurons. This allows the number of sampled UVA neurons to match the number of UVA projection neurons while preserving the timing and reliability of recorded UVA neurons. Using the sampled UVA neurons, we create a procedure to train model HVC projection neurons. Each model HVC neuron is trained to burst at a putative time in response to direct excitatory UVA input and indirect inhibitory input through HVC interneurons. Training is performed with learning rules similar to those previously used to characterize learning capacities of neuronal circuits (Memmesheimer et al., 2014). HVC projection neurons are simulated with an integrate-and-fire model and a more biologically informed two-compartment model that includes a dendritic calcium spike. In both models, trained HVC projection neurons produce precise, sparse bursts only when the convergence from UVA to HVC is high enough to overcome the inconsistency of UVA firing patterns. However, this level of convergence is not supported by experiments that have traced the axons from UVA to HVC. When realistic convergence is used, trained HVC projection neurons are unable to reproduce the precision and reliability experimentally observed in HVC. Noise in HVC interneurons further exacerbates this issue. Our work casts doubt on the mechanism of UVA driving HVC bursts.