Multiscale and multimodal reconstruction of cortical structure and function
Our newest preprint! https://www.biorxiv.org/content/10.1101/2020.10.14.338681v2 EM and neural activity images from three million cubic microns of mouse visual cortex. Four scientific vignettes from organelles to synapses to circuit structure and function @BCMHouston @PrincetonNeuro @AllenInstitute @IARPAnews
The EM images (https://microns-explorer.org/) were previously used by two preprints (https://www.biorxiv.org/content/10.1101/2019.12.29.890319v1.abstract & https://www.biorxiv.org/content/10.1101/2020.03.31.018952v1.abstract) to address specific scientific questions. Here we describe the entire resource, and the “vignettes” illustrate the diversity of potential uses.
The reconstruction describes a 250×140×90 μm3 EM volume from L2/3 mouse V1. It contains 8 million segmented objects and 3 million synapse predictions. Proofreading corrected almost all of the cells with somas in the volume, as well as the synapses between pyramidal cells.
Our volume is large enough to capture several inhibitory cell somas. We can coarsely classify most of these by their axonal morphology and synaptic features. Example basket (yellow), chandelier (green), bipolar (pink) and Martinotti (blue) cells are shown on the right.
The segmentation also has several types of glia, including astrocytes, microglia (green), oligodendrocytes (yellow) and their precursor cells (magenta). Astrocytes are extremely dense, so non-astrocytic glia are shown below, with examples on the right.
We performed an automated reconstruction of the roughly 2.4 million mitochondria within the volume, and will release it as a new layer of the microns-explorer.org dataset.
The resource incorporates 2-photon calcium imaging data for a subset of the reconstructed pyramidal cells. These images were co-registered to our reconstruction, enabling discovery of structure-function relations (see Fig 6).
We believe that the resource will be useful to many labs for addressing a broad range of scientific questions. Please take a look at microns-explorer.org! We provide four “vignettes” to demonstrate diverse uses of this rich dataset.
A lot of great previous work studies how inhibitory cells target other cells, but less work analyzes their inputs (see side chain). We use automated synapse predictions to analyze the properties of synaptic input onto several classes of inhibitory cell. (Fig 3)
We observe striking differences in linear and surface synapse density between the types we can classify. We also see stark differences in synapse size across the somas and dendrites of several inhibitory classes. (Fig 3)
Previous work has found a connection between dendritic synapse density and mitochondrial coverage of cultured pyramidal cells (side chain). We see this correlation in reconstructed pyramidal cells, and show that this correlation shows some compartment specificity. (Fig 4)
The preceding vignettes illustrate the scientific power of “multiscale.” High resolution (nanometers) EM reveals synapses and mitochondria. Large field of view (hundreds of microns) enables annotation with contextual information (cell type and cellular compartment).
In the subgraph of pyramidal cells with >100 μm of axon, 3-cell motif counts deviate from Erdős-Rényi models (red cross vs dashed line) as previously reported. Correcting for each neuron’s degree (w/ configuration model) improves fit (red cross vs. white circle). (Fig 5)
The configuration model of random graphs was previously applied to biological networks including the C. elegans connectome. Our application here suggests that the issue of “nonrandomness” of cortical connectivity will be more subtle than expected.
This is the first EM reconstruction of a cortical circuit that is large enough to permit analysis of three-cell motifs including the rare ones. The milestone in scale was attainable because of the large field of view and acceleration of reconstruction by automation.
Cells receiving a larger fraction of their connections from nearby cells exhibit stronger and more reliable visual responses to their preferred stimuli, consistent with the theory that recurrent excitation acts to amplify responses. (Fig 6)
This last vignette, relating neural connectivity and activity, is “multimodal.” Our resource advances the trend of performing calcium imaging prior to EM, a powerful approach for relating structure and function of neural circuits.
Much of this data is already available on microns-explorer.org, and most of our analysis code is now also available as an interactive binder https://github.com/AllenInstitute/MicronsBinder. It’s never been easier to start digging into the data.
We’re grateful for support from @IARPAnews, @neurowitz, @RJVneurotech, @NINDSnews, @NatEyeInstitute, @NIMHgov, @NIH, Mathers & @Samsung Foundations, as well as @googlecloud, @awscloud, and @Intel.
Inhibitory Side Chain:
Gulyás et al. (1999) quantified the densities of excitatory and inhibitory inputs onto parvalbumin (PV), calretinin (CR), and calbindin (CB) expressing interneurons in the hippocampus https://www.jneurosci.org/content/19/22/10082. 9.1/15
Martina, Vida, and Jonas (2000) studied how action potentials propagate through the axons, somas, and dendrites of somatostatin-expressing interneurons in the hippocampus for different types of inputs https://science.sciencemag.org/content/287/5451/295.full. 9.2/15
Kawaguchi, Karube, and Kubota (2006) quantified dendritic branching and spine formation in several interneuron types. https://academic.oup.com/cercor/article/16/5/696/276955 9.3/15
Kameda et al. (2012) quantified how glutamatergic and GABAergic synapses target the dendrites and somas of PV cells. https://pubmed.ncbi.nlm.nih.gov/22429243/ 9.4/15
Hioki et al. (2013) measured the inputs of PV cells in mouse somatosensory cortex, and distinguished the inputs of other PV cells, somatostatin (SOM) cells, and vasoactive intestinal polypeptide (VIP) expressing cells https://www.jneurosci.org/content/33/2/544.short. 9.5/15
See Kubota (2014) for a review of the study of inhibitory outputs and their roles in cortical circuits. https://www.sciencedirect.com/science/article/abs/pii/S0959438813001955 9.6/15
Mitochondria Side Chain:
Manipulating the expression of GTP-ases important for mitochondrial fusion and fission (DRP1 and OPA1) modifies density of PSDs and spines https://www.cell.com/cell/fulltext/S0092-8674(04)01045-1, https://academic.oup.com/brain/article/136/5/1518/285839.
Dickey and Strack showed similar effects with phosphorylation to regulate Drp1 https://www.jneurosci.org/content/31/44/15716.