Feature request: Link connectors similar to how treenodes are linked

[dwitvliet]

We often see big variation in synapse sizes, with some appearing in only one section while others appear in more than 20. By linking connector, it would be possible to give a rough size measure to synapses. This could be done by simply putting a connector in every section the synapse appears, and then linking the connectors to indicate that all they all belong to only one synapse, just as many linked treenodes form a skeleton.

[aschampion]

An alternative approach suggested for this is #1085, so that connectors would have a radius annotation and would be visible in the tracing layer within the radius. Of course synapses aren't spherical, so that isn't ideal either.

[dwitvliet]

@aschampion It is difficult to imagine a perfect solution. However, I do believe linking connectors would represent the real data better than simply setting a radius. For large dyadic and triatic synapses (very common in at least C . elegans), we often see different postsynaptic partners on different sections. Therefore, even though a synapses from neuron A is visible on, say, 20 sections, it might only be presynaptic to neuron B in 18 sections and neuron C in 3 sections. Linked connectors could here represent this differential innovation of neuron B and C, while a single sphere could not easily do the same.

[acardona]

@dwitvliet Having given some thought to this, linking connectors is reminiscent of what Scott Emmon's team did for Elegance, precisely to be able to approximately measure synapse sizes. This works well in C. elegans because of the primarily longitudinal profile of both neuronal arbors and synaptic surfaces, along the anterior-posterior axis. Like you stated, synapses in C. elegans can extend for 20 sections or more. In other systems, you would have to create branched, complex 3D trees of connectors to represent synapses, and these would not extend across so many serial sections.

In my view, connectors are topological elements, and conflating the geometry of the synapse with the topology of the connectivity is perhaps not the best approach. Instead, what you are asking for are two distinct issues:

1) The ability to represent the same synapse in multiple sections, for ease of reconstruction purposes. Primarily to not miss a synapse or re-annotate the same synapse twice. CATMAID currently has mitigating strategies for the re-annotation issue (it warns you), but not for the missing (other than review of the arbor).

2) The ability to measure the postsynaptic surface or the spread of the vesicle cloud associated with a synaptic surface. CATMAID currently does not have special-purpose tools for this. Volumetric representations and annotations are very much in the works, and perhaps we could dedicate some thought towards representing synaptic surfaces as part of that work.

For your immediate needs in C. elegans, we could give more thought to the tree-like connectors approach. Would there be a single presynaptic relation or many? All should be from the same skeleton, which has to be enforced. Would multiple postsynaptic relations onto postsynaptic arbors count as one, or as many? We have cases where two postsynaptic dendrites from the same neuron reach the same polyadic presynaptic site, and we must count them as two synapses, even if they share the presynaptic t-bar (with all that it implies physiologically). (Fortunately these cases are rare in Drosophila.)

[dwitvliet]

@acardona Thank you for putting so much thought into this. I definitely agree that keeping the two issues apart would be optimal. Another reason is that annotating the size of synapses simply by the number of sections can skew the data quite a bit when synapses extend far in the xy dimensions, but not in the z dimension (which we see fairly often, especially when the animal isn't cut perfectly perpendicular to its anterior-posterior axis).

The main advantage of the linked connectors is that it is simple and fast to implement. Although far from perfect, it could serve as a rough estimate of synapse size that could assist network analysis until a more well-thought-out volumetrically based solution is functional. We have an old instance of CATMAID where linking connectors was integrated for this reason. It is simply allows linking connectors through a parent column (similar to treenodes) and enforces the same presynaptic partners, as you also mentioned. If you think other people could benefit from such a thing, we would be more than happy to merge it with the most current CATMAID version and submit a pull request. We are aware that this isn't optimal and would require additional work to be fully integrated, as each connector would still be considered by CATMAID to be an independent synapse. However, it does currently benefit our annotation and analysis, so we thought it might be useful for other groups as well. (Additionally, we wouldn't have to worry about merge conflicts anymore)