If you’ve ever done research, then you know about variability. Probably too much about it, in fact. Variability means your results change from day to day, or from cell to cell, or animal to animal. This does not make us scientists happy.
Especially for those of us who study animal behavior, which is particularly capricious, variability is the bane of our existence. Even when we treat animals exactly the same, they usually don’t behave exactly the same. Sometimes they behave completely differently.
For example, in grad school I used to measure the alcohol consumption of fruit flies: some of the flies would drink practically their entire body weight in alcohol, while other flies barely touched the stuff. And these were genetically identical flies who had grown up together in the same bottle and were being tested right next to each other at the same exact time.
This kind of variability was mysterious and intriguing, but mostly it was annoying. Since I had already taken great pains to treat the flies as identically as possible, the only thing I could do about this variability was to average the results across many animals. That’s what most researchers do. Add in enough data points, and your error bars eventually shrink to a reasonable size. Most of us don’t spend much time thinking about why the variability exists in the first place.
Why can’t animals make up their minds?
A sweet new Cell paper from Cori Bargmann’s lab at Rockefeller addresses the question of variability head on. The study focuses on one of the most mysterious forms of variability: sometimes the same animal behaves differently each time you test it, even if you conduct the tests back to back under identical conditions.
You might be able to explain variability across different animals by imagining that they have slightly different brains, perhaps influenced by subtle experiences or random events during their lifetime. But that can’t explain why the same animal behaves differently on different tests, repeatedly changing its mind each time. What’s going on in that animal’s brain?
The new study by Andrew Gordus and others in the Bargmann lab sheds light on how dynamic changes in the brain produce variability in behavior. Specifically, they investigated why C. elegans worms (which I’ve talked about before) show variable responses to an attractive odor called isoamyl alcohol.
When you give worms a puff of this yummy odor, they tend to crawl toward it. But not always. Sometimes they just keep wandering around as if they haven’t even realized the odor is there.
Your first thought might be that that’s exactly what’s going on: maybe sometimes the worms don’t sense the odor’s presence. But this isn’t the case. The sensory neuron in the nose that physically detects isoamyl alcohol, called AWC, responds reliably on every trial. So the worms always know the odor is there—it’s just that sometimes they don’t care.
The AWC sensory neuron connects to an command interneuron called AVA, which is in charge of actually controlling the worm’s movement. Turning on the odor alters the activity of AWC, which transmits the signal to AVA, which ultimately tells the worm to move toward the odor.
While the activity of AWC reliably reports whether the odor is there, the activity of AVA isn’t so dependable. AVA doesn’t always respond to the odor, and when it does it often takes a while to react. But AVA does reliably tell the worm what to do—either move forward or backward. So the reason the worm doesn’t respond reliably to the odor is because the signal from AWC to AVA is somehow getting disrupted. How is this happening?
Networking in the brain
In addition to the direct connection between AWC and AVA, they also connect indirectly via two other neurons called AIB and RIM. AIB and RIM show unreliable odor responses just like AVA.
Interestingly, the three unreliable neurons—AVA, AIB, and RIM—seem to be part of a neuronal network. Each neuron can exist in two different states, either a high activity or low activity state (termed ON or OFF, respectively), and the three neurons are always in the same state. They can spontaneously switch from one state to the other, but they always switch together. They also stick together when you turn on the odor: either all of the neurons respond or none of them do.
It turns out that the state of the network can actually explain why the neurons respond to the odor on some trials and not others: they respond if they’re in the ON state but not the OFF state. So if you know what state the network is in, the odor responses no longer seem so puzzlingly unpredictable. That’s how reliable signals from AWC get transformed into unreliable AVA responses—by acting on a network that happens to be in the OFF state.
Generating variability in a network
But where does this variability actually come from? Bargmann’s team investigated whether a specific part of the network might represent the source of variability. They noticed that occasionally the three unreliable neurons weren’t all in the same state (contrary to what I told you earlier)—sometimes AIB was ON while the other two neurons were OFF. If the network happened to be in that configuration when the odor was turned on, AIB responded more reliably than ever!
This suggests that AVA and/or RIM are actively promoting variability in AIB: when they’re OFF, AIB can respond to the odor much more reliably. To confirm this hypothesis, Bargmann’s group artificially silenced the activity of AVA and RIM. Silencing either AVA or RIM (or both) made AIB’s odor responses more reliable. Silencing RIM made the worm’s actual behavior more reliable, too.
Conversely, silencing AIB made the responses of the other neurons less reliable, suggesting that it normally promotes reliability. So while RIM and AVA are making AIB more variable, poor AIB is working hard to make them more reliable. It’s like when your kids are actively making a mess at the same time as you’re trying to clean the freaking house.
How worms decide
Now we can finally make a model for what’s happening when a worm smells the odor. The sensory neuron AWC responds reliably to the odor and transmits a faithful signal to AIB. But AWC also activates AVA and RIM, which introduce variability into the network and distort the original sensory signal. The transformed signal that ultimately reaches AVA is no longer reliable. Since AVA controls the worm’s movement, the variability in AVA’s response is reflected in the worm’s behavior.
We often imagine brain circuits to be linear pathways, with each set of neurons transmitting a clear signal to the next step in the pathway. Bargmann’s paper highlights how this isn’t really how the brain works (even a brain as tiny and simple as a worm’s!). Instead, neurons are organized into intricate networks that transform the signal in complex ways, with information flowing in multiple directions.
Changing minds for changing times
Overall, Bargmann’s new study sheds light on why a reliable signal from the environment doesn’t always produce reliable behavior. In this study, the state of the network determines whether the worm will respond to the odor. The network state can change in a matter of minutes or even seconds, which explains why a worm’s behavior seems so inconsistent. It’s not clear what the network state really means to the animal, but it probably means something—maybe it reflects whether the worm is feeling hungry, or excited, or lazy, or who knows what else.
As shown in this study, some neurons promote reliability in a network while others promote variability. Variability in a network leads to variable behavior. But why the heck would you want a system where responses are variable?
In a previous post about songbirds I discussed how variability can be important for learning—it ensures that you don’t get stuck trying the same thing over and over again. That idea is also relevant here.
Completely deterministic behavior might be good for responses that never need to change. Like if you smell the scent of a predator, you should always run away—there’s no nuance there. But most of the time your behavior depends on the situation. Both the external world and your internal needs are constantly changing. Variability makes our behavior flexible, allowing us to adapt to different situations. Thanks to Bargmann’s study, we now have a better understanding of how neuronal networks can generate this variability and allow for flexible behavioral responses.
Citation for the study:
Gordus A, Pokala N, Levy S, Flavell SW, Bargmann CI. Feedback from Network States Generates Variability in a Probabilistic Olfactory Circuit. Cell pii: S0092-8674(15)00184-1 (2015). [Epub ahead of print]