Thread by @neuroecology

Animals will respond differently to the same inputs when they are hungry/sated, attentive/bored, etc. But how do we know what state the animal is in?

In this preprint with @jpillowtime @MurthyLab, we show how you can do that automatically! 👇

biorxiv.org/content/10.110…

@jpillowtime @MurthyLab Watch this dog - he's super excited and he's going to watch the ball, take in that sensory information, and transform it into motor commands to catch the ball

Now take a look at this doggo - he too sees a ball go up and down but he uses a different strategy...

What was the difference between the two of them? The sensory information or their internal state?

Animals can be in any number of states which we have no idea about and very little control over. After all, we *don't know* the possible states an animal can even be in!

We often pretend that when animals are faced with some task, they will respond in roughly one kind of way. You give it some kind of input, and we estimate its behavioral response as a function of that output – simple psychometric curves. But we all know this isn’t true!

And these states have a huge impact on the nervous system, whether through neuromodulator release, activation/suppression of particular neuron subtypes, or other reorganization of the nervous system

We study this with courtship in fruit flies. When a male really loves a female fly, he'll follow her and vibrate his wing to produce song. He has to closely monitor what the female is doing, and changes his song based on her movement. If she likes it, she will allow him to mate!

The song looks something like this, where he can sing a tone (sine song), one of two pulsatile (Pfast or Pslow) or no song at all - he can only ever sing one type at a time.

This is perfect for us because it gives us a discrete behavior to understand

We track each animal's movements (velocities, distances, angles, etc) and use those to predict the song. Will he sing more of one type when he moves faster?

Classic way to do this is to assume there is one way the animal responds to these cues ("GLM")

Our way is to assume there are MULTIPLE ways the animal can respond to these cues ("GLM-HMM") and let it find out what those are

It does much better at explaining behavior

Why does it do so much better? Because there are THREE different ways this male is responding to the female, something you would never pick up on otherwise

If you try to predict the animal's behavior using the wrong state at a particular moment in time, you'll do really poorly

Then we can go in and MANIPULATE THE NERVOUS SYSTEM. We can directly activate neurons WHILE the male is courting the female

We find multiple sets can activate song (left in picture)

We find only ONE that changes which state the animal is in (how it responds to incoming cues)

If we hadn't had this model, we would have seen this behavior and thought, "well that's weird. Optogenetics, huh?"

But with the model we can tell that this neuron isn't a song COMMAND neuron but a neuron that determines song STRATEGY