Title: Hierarchical sparse coding in the sensory system of Caenorhabditis elegans
Summary: The big question in this paper is how do different neurons convey sensory information about various stimuli.
To start, they generated a library of 19 strains, in total expressing GCaMP3 in 28 different neurons or neuron groups. They then the calcium response from >5 animals from each strain in response to a panel of 13 stimuli. Those stimuli included isoamyl alcohol (on and off), diacetyl (on and off), salt (on and off), pH (low and high), osmotic stress (on and off), E. coli supernatant (on and off), and blue light (on only). I believe all of these have been previously tested on specific C. elegans neurons, but this sort of systematic all-vs-all study has never been undertaken. Clearly, this is direly needed information to parameterize models of pan-neuronal activity!
The methods are worth note. They immobilize worms in a microfluidic device, and then flow chemicals past their heads while optically recording GCaMP3 signals. The consider the neuron to respond if the mean GCaMP levels varied by more than 20% in the 7 seconds following a stimulus. Stimuli were considered as the onset of the odor/chemical pulse, as well as the offset (as some neurons have been shown to detect only rising edges and some detect falling edges). For two neurons in the tail for which none of their stimuli induced a response, they also loaded worms backwards (tail-first) and retested them (and again, no response), but this was not done for any of the other neurons.
They find what they refer to as a sparse hierarchical code, but I didn't get a clear sense of what they take that to mean. The sparse part presumably means that few neurons respond to any given stimuli, which seems to only be sort of the case. Probably half the neurons respond to E. coli supernatant (which presumably has a ton of stuff in it), but only two neurons respond to isoamyl alcohol. On the other hand, 4 neurons respond to diacetyl, 5 respond to pH, and 7-8 respond to NaCl or osmotic stress or light. The hierarchical bit is a little less explained, and I'm just honestly not sure what they mean by that.
They also note that none of the mechanosensory neurons seem to respond to any of the chemical (or blue light) stimuli. Vice versa, when they use channelrhodopsin to stimulate some set of mechanosensory neurons, they don't observe any response in AWC (the most chemosensory neuron that responds to almost anything chemical, including changes in fluid flow direction). They take this to mean that the mechanosensory and chemosensory neurons are indeed well-isolated from one another and capable of carrying clearly orthogonal signals.
I have a lot of questions about this paper.
- They don't show almost any raw data. It's hard to evaluate their claims that a neuron does or does not respond without seeing the calcium traces. They present the data as binary heatmap - e.g. there is an effect, or there isn't, with no measure of whether the result contains any ambiguity.
- Along those same lines, they dont discuss variability in responses across animals or across trials within a given animal. This would be useful to know! Do certain neurons plateau or not, or are their responses seemingly graded by an unknown variable factor? Or are they perfectly stereotyped as in AWC in (Gordus. et al. 2015).
- What do the temporal patterns look like for each neuron? Are they mostly gradient detectors (temporally high-pass-like)? Or do they encode DC components of the signal? This was extremely well-characterized for AWA in (Larsch et al.) - a paper discussed here a few days ago - and it seems like here they'd have a chance to investigate whether AWA's behavior is remotely typical or not.
- For each stimulus, they measured responses at a single concentration. Would this hierarchical pattern break down if they measured them at higher concentrations? To be fair, I think that they used reasonably high concentrations already, so maybe not, but it seems worth considering.
- Why not a few more chemicals? This is of course, an entirely unfair question to ask, because I'm sure collecting this dataset was already an incredible amount of work. However, I presume C. elegans can differentiate more chemicals than just IAA, diacetyl and salt (though perhaps it can't - I should find out). If so, can it resolve more chemicals than it has sensory neurons? One could imagine it could resolve 2^k chemicals with k neurons, although that presents a massive problem if it ever has to resolve more than a single chemical at once. If it might have to resolve up to k chemicals at once, then it should encode each chemical in a single neuron.
- Given the fact that these sensory neurons are expected to massively interconnected via both chemical synapses and gap junctions, which might actually transmit very quickly (certainly faster than 7 seconds), how many of these responses are indirect? Perhaps they could use a blocker of synaptic transmission for these experiments? (Not that I know of one, but I should look). Unc-13 or unc-18 mutants? Are there viable innexin mutants?