Multi-layered maps of neuropil with segmentation-guided contrastive learning.

Maps of the nervous system that identify individual cells along with their type, subcellular components and connectivity have the potential to elucidate fundamental organizational principles of neural circuits. Nanometer-resolution imaging of brain tissue provides the necessary raw data, but inferring cellular and subcellular annotation layers is challenging. We present segmentation-guided contrastive learning of representations (SegCLR), a self-supervised machine learning technique that produces representations of cells directly from 3D imagery and segmentations. When applied to volumes of human and mouse cortex, SegCLR enables accurate classification of cellular subcompartments and achieves performance equivalent to a supervised approach while requiring 400-fold fewer labeled examples. SegCLR also enables inference of cell types from fragments as small as 10 µm, which enhances the utility of volumes in which many neurites are truncated at boundaries. Finally, SegCLR enables exploration of layer 5 pyramidal cell subtypes and automated large-scale analysis of synaptic partners in mouse visual cortex.

Millisecond timescale disinhibition mediates fast information transmission through an avian basal ganglia loop.

Avian song learning shares striking similarities with human speech acquisition and requires a basal ganglia (BG)-thalamo-cortical circuit. Information processing and transmission speed in the BG is thought to be limited by synaptic architecture of two serial inhibitory connections. Propagation speed may be critical in the avian BG circuit given the temporally precise control of musculature during vocalization. We used electrical stimulation of the cortical inputs to the BG to study, with fine time resolution, the functional connectivity within this network. We found that neurons in thalamic and cortical nuclei that are not directly connected with the stimulated area can respond to the stimulation with extremely short latencies. Through pharmacological manipulations, we trace this property back to the BG and show that the cortical stimulation triggers fast disinhibition of the thalamic neurons. Surprisingly, feedforward inhibition mediated by striatal inhibitory neurons onto BG output neurons sometimes precedes the monosynaptic excitatory drive from cortical afferents. The fast feedforward inhibition lengthens a single interspike interval in BG output neurons by just a few milliseconds. This short delay is sufficient to drive a strong, brief increase in firing probability in the target thalamic neurons, evoking short-latency responses. By blocking glutamate receptors in vivo, we show that thalamic responses do not appear to rely on excitatory drive, and we show in a theoretical model that they could be mediated by postinhibitory rebound properties. Such fast signaling through disinhibition and rebound may be a crucial specialization for learning of rapid and temporally precise motor acts such as vocal communication.

Structural correlates of efficient GABAergic transmission in the basal ganglia-thalamus pathway.

Giant inhibitory terminals with multiple synapses, the counterparts of excitatory "detonator" or "driver" terminals, have not been described in the forebrain. Using three-dimensional reconstructions of electron microscopic images, we quantitatively characterize a GABAergic pathway that establishes synaptic contacts exclusively via multiple synapses. Axon terminals of the nigrothalamic pathway formed, on average, 8.5 synapses on large-diameter dendrites and somata of relay cells in the ventromedial nucleus of the rat thalamus. All synapses of a given terminal converged on a single postsynaptic element. The vast majority of the synapses established by a single terminal were not separated by astrocytic processes. Nigrothalamic terminals in the macaque monkey showed the same ultrastructural features both in qualitative and quantitative terms (the median number of synapse per target was also 8.5). The individual synapses were closely spaced in both species. The nearest-neighbor synaptic distances were 169 nm in the rat and 178 nm in the monkey. The average number of synapses within 0.75 microm from any given synapse was 3.8 in the rat and 3.5 in the monkey. The arrangement of synapses described in this study creates favorable conditions for intersynaptic spillover of GABA among the multiple synapses of a single bouton, which can result in larger charge transfer. This could explain faithful and efficient GABAergic signal transmission in the nigrothalamic pathway in the healthy condition and during Parkinson's disease. In addition, our structural data suggest that the rodent nigrothalamic pathway can be a valid model of the primate condition, when the mechanism of GABAergic transmission is studied.