[pjin]

From the whitepaper [1]:

  After an appropriate transport period, the brain is
  perfused, sectioned, suitably stained, and each section
  digitally imaged. These 2-D images are co-registered into a
  3-D computer stack that is subsequently registered to a
  common reference atlas. The resulting 3-D brain image is
  largely unlabeled (i.e., contains no signal of interest),
  except for the connections between the injected region and
  its target regions. Thus the labeled connections are
  clearly identifiable. A given region is injected in
  multiple animals to account for individual variability.

They do two passes, in each pass they stain mouse brain cells in a particular region, then they kill the mouse, slice its brain, and image the slices. While this method certainly works well on mice (I've seen surgical work done of mice first hand, neat stuff), sadly it doesn't quite scale up to imaging the human brain. But having all those (petabytes) of data has gotta be worth something.

[1] http://brainarchitecture.org/mouse/documentation/project-whi...

[DocSavage]

I must be missing something. When I think of connections, I think of synapses or gap junctions -- the coupling of one neuron to others through a signaling mechanism. Their tracers do the equivalent of segmentation of neurons, which doesn't give you the connections, just the morphology of neurons. You left out what seems to be a key part of their method -- use of retrograde tracers that are taken up by axon terminals. These retrograde tracers have to be injected into likely end points. Seems like its an interesting technique for mapping a limited number of connections but useless for producing a comprehensive connectome due to required injection density. Am I missing some aspect of this that allows generation of a connectome?

It's misleading to use the term "wiring diagram for entire mouse brain". An engineer wouldn't consider it a wiring diagram if you stripped out a majority of the contact descriptions and just said there's a wire here, here, and here.

[polyfractal]

Well, there are both retrograde and anterograde tracers. So you can trace from either target->source or source->target depending on your choice of tracer.

These tracers are used to map circuits because they are passed from one neuron to the next via functional synapses...which means that any neuron that is stained must have been in contact with a previous neuron, etc etc.

The "connectome" is built up by injecting a limited amount of tracer in a single region, in multiple animals, to provide a mapping of that region. To get the entire brain you need lots and lots of injections in a very large number of mice (and a lot of technicians slaving away over cryostats).

To your point about not labeling synapses and gap junctions: those are relatively unimportant when considering wiring diagrams of the brain. What is more important is knowing which partners a neuron synapses onto. You don't really care how many synapses are involved unless you are looking at single or clusters of neurons, nor can you really measure it without some other method (e.g. electrophysiology).

Caveat about tracers which often goes unannounced: the staining of a tracer in a secondary neuron is only as strong as the connection between the primary and the secondary. Which means that a neuron who synapses strongly will be much brighter than a neuron that synapses weakly. Similarly, tertiary (and quaternary, etc) neurons become progressively weaker stained as the exponential dilution of the tracer kicks in.