Dense connectomic reconstruction in layer 4 of the somatosensory cortex
Science, 29 Nov 2019:Vol. 366, Issue 6469, eaay3134DOI: 10.1126/science.aay3134
Dense connectomic reconstruction in layer 4 of the somatosensory cortex
Alessandro Motta, et.al.
Department of Connectomics, Max Planck Institute for Brain Research, D-60438 Frankfurt, Germany.
Probabilistic Numerics Group, Max Planck Institute for Intelligent Systems, D-72076 Tübingen, Germany.
[paraphrase]
The cerebral cortex of mammals houses an enormously complex intercellular interaction
network implemented with neuronal processes that are long and thin, branching, and
extremely densely packed. Early estimates indicated that 4 km of axons and 400 m of
dendrites are compressed into a cubic millimeter of cortical tissue. This high packing
density of cellular processes has made the locally dense mapping of neuronal networks in
the cerebral cortex challenging.
So far, reconstructions of cortical tissue have been either sparse or restricted to small
volumes of up to 1500 μm
3
. Consequently, the detailed network architecture of the cerebral
cortex is unknown. Particular open questions are to what degree local neuronal circuits are
explainable by geometric rules alone and on which spatial scales cortical connectivity is
only explainable by innervation preferences beyond such geometric models. Similarly,
although numerous cortical neuronal cell types have been described based on protein
expression, morphology, and electrophysiological characteristics, and these have been
shown to have particular synaptic target patterns, the inverse question—whether, at the
level of the dense cortical circuit, axons represent a continuum of synaptic preference or a
set of distinct innervation paradigms that would allow for a purely connectomic cell type
definition [as has been successful in the retina]—is still open. Next, at the level of synaptic
input to the primary dendrites of cortical excitatory cells, it is not known whether the
typically three to 10 primary dendrites of a cortical neuron that leave the cell body
homogeneously sample the available excitatory and inhibitory synaptic inputs or if
there is an enhanced heterogeneity of synaptic input composition, making it possible to
exploit the numerous mechanisms that have been discussed for the nonlinear integration
of local synaptic inputs. Finally, whereas the change of synaptic weights in response to
electrical and sensory stimulation has been widely studied and connectomic data consistent
with LTP have been described, the fraction of a given cortical circuit that is plausibly
shaped by processes related to Hebbian learning under undisturbed conditions is still
unknown.
Using FocusEM, we obtained the first dense circuit reconstruction from the mammalian
cerebral cortex at a scale that allowed the analysis of axonal patterns of subcellular
innervation, ~300 times larger than previous dense reconstructions from cortex. Inhibitory
axon types preferentially innervating certain postsynaptic subcellular compartments
could be defined solely on the basis of connectomic information. In addition to inhibitory
axons, a fraction of excitatory axons also exhibited such subcellular innervation
preferences. The geometrical arrangement of axons and dendrites explained only a
moderate fraction of synaptic innervation, revoking coarse random models of cortical
wiring. A substantial TC synapse gradient in L4 gave rise to an enhanced heterogeneity of
synaptic input composition at the level of single cortical dendrites, which was accompanied
by a reduced innervation from AD-preferring inhibitory inputs. The consistency of synapse
size between pairs of axons and dendrites signified fractions of the circuit consistent with
saturated synaptic plasticity, placing an upper bound on the “learned” fraction of the
circuit. FocusEM allowed the dense mapping of circuits in the cerebral cortex at a
throughput that enables connectomic screening.