In ’09 Anderson et al published a paper with a general framework for connectomics research. I’ve been planning to summarize this paper for well over a year, and it pains me to be so late to the scene, but better late than never. Since it is open access I will sample liberally from their prose. One might say that the authors make three key points:
1) Neural networks are complex, so physiology, confocal microscopy, and/or behavior can’t fully differentiate between different plausible circuitry landscapes. I’ll note that this is incredibly controversial! Here is their reasoning:
Why can’t we deduce networks from physiology, confocal imaging, or behavior? The answer is that potential network motifs derived by these methods are not unique…. A small network of two different bipolar cells (BCs) driving two GC [ganglion cell] channels, interconnected by one amacrine cell (AC) class can be connected in 90 formal motifs and at least 40 of these are biologically tenable….
The main scotopic signal flow network is rod → rod BC → rod AC, which then bifurcates into two synaptic arms that reenter the ON- and OFF-cone BC pathways. This motif was reported… using ssTEM [serial section transmisstion electron microscopy]. Subsequent physiological and genetic analyses provided correlative support for the anatomical model, but neither study would have uniquely yielded the correct topology…. despite five decades of robust physiology of retinal rod signaling, the discovery of a second scotopic pathway was also based on ssTEM…. It is unlikely that the day of ultrastructural discovery is past and we argue that it is just dawning.
Another example of a motif they say is often indistinguishable without ssTEM is the nested feedback synapse (see here, doi: 10.1002/1096-9861(20001002)425:4<560).
2) Reconstructing anatomical networks from ssTEM needs to be automated. This is relatively uncontroversial. The manpower required to make connections from these data sets is too massive to scale.
3) Reconstructing anatomical networks from ssTEM can be automated. This is moderately controversial and is the crux of their paper. The three elements they assume are:
a) A resolution “sufficient to unambiguously identify synaptic contacts and gap junctions… nominally 2 nm/pixel. This yields synaptic vesicles spanned by 8–10 pixels that are robust for circuitry tracing.”
b) Coverage of all of the relevant classes of neuron in that region that the researchers would like to uncover. They call this region a “canonical field.” This is mainly a practical concern and will vary based on the goals of a given undertaking.
c) Classifying the neurons (and, eventually, the glia) in the region of interest. This can be done based on various molecular markers. In particular, they use immunoglobulins (colored) to bind to small molecules on the surface of the neuron. They then use light microscopy and a computational classification system based on the binding to classify the neurons. See their set up:
Once the images have been acquired (as of publication, their system was acquiring 3,000 / day), the images are put through a series of parameter-based and/or user-specified transforms to fit them into a mosaic space. Importantly, they note that “our goal is not to render 3D ultrastructural images, but rather tabulate connections within the volume.” 3d requires too much computational power, such that at this point it would not be readily accessible for most, so they take the same approach that google earth takes and make a computationally efficient image pyramid for each tile and then transform the pyramids with the GPU.
The Robert Marc lab is doing fascinating research, looks to be invested in open access research, and their lab page has much more info for the curious soul, including a short, illustrative video of a volume that has been reconstructed.
Anderson JR, Jones BW, Yang J-H, Shaw MV, Watt CB, et al. (2009) A Computational Framework for Ultrastructural Mapping of Neural Circuitry. PLoS Biol 7(3): e1000074. doi:10.1371/journal.pbio.1000074