Connectivity between neurons is established by tracing the axons, dendrites, and synapses through the imaged volume, and patterns of connectivity are compared with the functional properties of the neurons in the circuit. In this way, the relationship between how neuronal circuits process information and how their constituent neurons are connected to one another can be explored.
We currently use large-scale, high-throughput transmission electron microscopy (EM) of serial thin (<50 nm) sections of brain tissue, followed by reconstruction of the neurons within the EM-imaged volume, to map the anatomical connectivity of a set of neurons.
The advantage of EM is that it can resolve both the "wires" between neurons—their axons, dendrites, and dendritic spines—and the connections between the wires, which are composed of chemical synapses and gap junctions. The method used to prepare neural tissue for EM labels cellular membrane in a complete and unbiased fashion. This means that, in principle, we can start at a given "seed" neuron in an EM-imaged volume and trace out its complete dendritic and axonal arbors and, in the process, note all sites of input and output to the cell (chemical and electrical synapses).
The tracing process can be continued iteratively. The pre- or postsynaptic partners of the seed neuron can be reconstructed, and then their partners, and so on, until the connectivity underlying a given circuit has been mapped out. (This mapping strategy is similar to how Web crawlers deployed by search engine companies chart the connectivity of the World Wide Web, by traversing from one Web page to the next.) A key output of the tracing effort is a graph, in the mathematical sense, with neurons represented by vertices and connections represented as edges. Pairwise and higher order connectivity patterns can be extracted from the graph and related to cell type, neural geometry, and most importantly, function: the physiological properties of the neurons in the graph, and information processing at the level of the circuit.
The method for obtaining functional information about the neurons in a given circuit depends on the species and the specific physiological parameters of interest. My recent work in Clay Reid's lab at Harvard Medical School, in collaboration with other members of the lab and researchers in the Center for Brain Science at Harvard and the Pittsburgh Supercomputing Center, provides an early proof-of-principle of combined network anatomy and function in mouse primary visual cortex. We used in vivo two-photon calcium imaging to characterize the preferred stimulus orientations of a group of neurons in layer 2/3 of visual cortex. We then prepared the tissue for EM and cut serial thin sections through the cluster of physiologically characterized cells.
We imaged the thin sections using a custom high-throughput transmission electron microscope camera array (TEMCA). This resulted in a 10-terabyte EM-imaged volume, with each section represented by a 120,000 × 80,000 pixel composite image (4 nm/pixel), encompassing about 450 × 350 × 50 micrometers of brain tissue. This volume was sufficiently large that we could construct the proximal portions of the axonal and dendritic arbors of the physiologically characterized neurons. We then traced all the dendrites postsynaptic to the physiologically characterized cells' axons, and examined the patterns of convergence by similarly and differently tuned cells onto their post-synaptic targets. In this way we were able to explore whether a relationship existed between the structure of this partial connectivity graph of visual cortex and the orientation tunings of the cells within it.
Our prototype effort revealed a number of technical limitations.
Foremost was the size of the EM-imaged volume. Although it was unusually large by historical standards, it was just barely big enough to contain some interesting cortical circuitry. My lab at Janelia has two dedicated FEI T12 electron microscopes. One will be used to develop a next-generation camera array; the other will be used to prototype fast sample handling. With these optimizations, I believe that significantly larger volumes of neural tissue (large enough, for example, to span all the cortical laminae), could be imaged in a few months' time. I also intend to continue collaborating to develop workflow for efficient manual and semi-automated tracing of subsets of neural circuitry contained in multi-terabyte EM image volumes.
Future work will take advantage of the ongoing and rapid advances in physiological imaging methods. Since we did the calcium-imaging experiments in mouse primary visual cortex (March 2008), the state of the art has advanced significantly. I anticipate that future EM-imaged volumes will contain much larger numbers of physiologically characterized neurons, with more complete characterization of each one. The result will be a greatly enriched combination of anatomy and function. I am also interested in exploring new combinations of light and electron microscopy; for example, labeling a cell within the EM-imaged volume with a monosynaptic retrograde virus could provide both a valuable cross-check of the two imaging modalities and connectivity information about input from more distant regions of brain.
Overall, our goal is to explore the extent to which anatomical connectivity can be related to the functional properties of a circuit. Although it is unlikely that an anatomical correlate can be found to all of a circuit's physiological properties (e.g., up and down states, dynamics of intrinsic oscillations, and the like), knowing the structure of a circuit's connectivity might constrain hypotheses about how it processes information, generate new hypotheses, and help guide new experimental work.
We demonstrate a meaningful prospective power analysis for an (admittedly idealized) illustrative connectome inference task. Modeling neurons as vertices and synapses as edges in a simple random graph model, we optimize the trade-off between the number of (putative) edges identified and the accuracy of the edge identification procedure. We conclude that explicit analysis of the quantity/quality trade-off is imperative for optimal neuroscientific experimental design. In particular, identifying edges faster/more cheaply, but with more error, can yield superior inferential performance.
The last decade has seen a rapid increase in the number of tools to acquire volume electron microscopy (EM) data. Several new scanning EM (SEM) imaging methods have emerged, and classical transmission EM (TEM) methods are being scaled up and automated. Here we summarize the new methods for acquiring large EM volumes, and discuss the tradeoffs in terms of resolution, acquisition speed, and reliability. We then assess each method's applicability to the problem of reconstructing anatomical connectivity between neurons, considering both the current capabilities and future prospects of the method. Finally, we argue that neuronal 'wiring diagrams' are likely necessary, but not sufficient, to understand the operation of most neuronal circuits: volume EM imaging will likely find its best application in combination with other methods in neuroscience, such as molecular biology, optogenetics, and physiology.
In the cerebral cortex, local circuits consist of tens of thousands of neurons, each of which makes thousands of synaptic connections. Perhaps the biggest impediment to understanding these networks is that we have no wiring diagrams of their interconnections. Even if we had a partial or complete wiring diagram, however, understanding the network would also require information about each neuron's function. Here we show that the relationship between structure and function can be studied in the cortex with a combination of in vivo physiology and network anatomy. We used two-photon calcium imaging to characterize a functional property--the preferred stimulus orientation--of a group of neurons in the mouse primary visual cortex. Large-scale electron microscopy of serial thin sections was then used to trace a portion of these neurons' local network. Consistent with a prediction from recent physiological experiments, inhibitory interneurons received convergent anatomical input from nearby excitatory neurons with a broad range of preferred orientations, although weak biases could not be rejected.