Modularity is all you need

Notes

This is a short script documenting my thought process

Recording idea

The idea is to find a use case for an e phys recording.

Basically, from what we know, T3 talks to LC11 and LC17.

There is seemingly a gap junction component, and as such, when you use tetanus toxin, you are still able to get position tracking. The logic from Mark Frye’s lab is that you can get faster responses on discrete time components and then update faster.

Idea: are saccades slower in the non–tetanus toxin version compared to the tetanus version? The logic being, if you know more data is coming, do you wait?

Modularity hypothesis

Anyway, my main goal is to understand how the distribution of data is done and why. Basically, you send 0.4, 0.2… until you sum to 1 over partner neurons, and those partner neurons for some reason lead to differentially distinct responses. The obvious hypothesis is that the structure is key. The idea being that wiring in circuits is what causes different outputs, and as such, different mini circuits can trigger distinct outputs.

Nonetheless, that is too big of a goal for a single project, so we have to solve smaller pieces.

Note to self: Hebbian wiring?

LC10 subtypes

LC10a and d are quite interesting to study. The logic being that they come from the same family, but they lead to “different” outputs. That is to say, ‘a’ leads to approach behavior in regards to mate locating, and ‘d’ is associated with avoidance behaviour. Although seemingly opposite, these can be wrapped in the same space. The key to validating this would be to find the role of LC10b and LC10c. Further reasoning for my hypothesis comes from the T4 and T5 neurons, which are subdivided into subtypes of a, b, c, and d, where each responds to one of the four cardinal directions. As such, it would not be far-fetched to assume a similar circuit may be at play for other behaviour.

More support

More support: the visual system and olfactory system share the idea that you integrate simple inputs into a large glomerulus or glomeruli, which perform X computation and spit out output. Again, repetition of modules. There are likely layers of integration where, on the smaller layers, you collect from receptive fields and you may get the idea of redundancy, but that is not the intended purpose. You are already putting in, for example, 10 ATP; you do not care if you get 10 outputs or 11 if you only need 10. In other words, you already paid for what you needed; anything extra was not created on purpose. As such, you collect over receptive fields, then you integrate to get data.

Next step

Okay, now back to LC10b and c. The first step would be to perform connectomic analysis on T4 subtypes and confirm they share, say, rank 1 and 2 inputs; same with T5. If that is validated, we continue to analyze LC10 subtypes. The idea being that we want to confirm the same pattern, but now the odd question is: what could possibly be the function of the two subtypes?

Okay, enough for today!

References

• Ribeiro, I. M. A., Chen, W.-Q., Drummond, N., Sauter, M., Prech, S., & Borst, A. (2026). A multi-input optic glomerulus mediates opposing behavioral responses to visual objects. bioRxiv. https://doi.org/10.64898/2026.01.26.701771
• Frighetto, G., Dombrovski, M., Palacios Castillo, L. M., Meera, P., Mirshahidi, P. S., Sanfilippo, P., Vaccari, A., Kandimalla, P., Hartenstein, V., Kurmangaliyev, Y. Z., Zipursky, S. L., & Frye, M. A. (2025). Electrical synapses mediate visual approach behavior. bioRxiv. https://doi.org/10.1101/2025.10.14.682373