Unraveling the hidden systems of the brain

Slugging it out in neuroscience

Credit: Puzzle by Chameleon Design from the Noun Project (CC-BY-3.0)

If no one has any strenuous objections, I’d like to tell you about some work we’ve just had published. Oh, you do? OK, read this instead: it’s an alleged academic miserably failing to win an argument with themselves.
Still here? Then let me tell you why neuroscience lives in hope: we don’t need to understand every neuron to understand the brain.

This is a big deal. Brains contain millions or billions of neurons. If we have to understand them all to understand what the brain is doing then we’re hopelessly lost, doomed to glimpse but a minuscule fraction of what is going on. We’d be trying to understand the complexities of a football match by watching footage of just one of the players for 90 minutes (weirdly, there is an entire movie of just this happening, starring everyone’s favourite mercurial Frenchman prone to spectacular melt-downs in World Cup finals). The player runs a bit for no apparent reason, gesticulating wildly with their arms. Occasionally the ball appears in shot. Occasionally another player runs into the frame, only to fall over their outstretched leg. From this, we’d try to work out who won, who scored, or who had the most influence on the game. Hopeless.

So the question is: do we? Do we have to understand every neuron? Or can we understand the brain by understanding what those neurons are collectively trying to achieve?

I think we can. What we are searching for is the hidden system within each group of neurons. This hidden system is what the neurons are collectively trying to achieve: the state of this system is what they are trying to convey to other neurons, or to the muscles that make things happen.

Take a group of neurons that we know do some job. Say, the set of neurons that make your jaw move up and down and gently rotate while you chew a biscuit (today, a fig roll — it’s one of my Five A Day, right? You can argue about whether a fig roll is technically a biscuit or a cake in the comments). Here, the hidden system is the set of commands for co-ordinating muscles that make the up, down, and rotate movements. It seems unlikely that a single neuron knows any command. Rather, it is only when looking at the activity of the neurons together that we can see the hidden commands embedded within them.

This idea of a simpler hidden system within a large group of neurons might apply everywhere. To the groups of neurons that do chewing, breathing, and walking. To the groups that do seeing, remembering, and deciding. But even in small animals, like mice, these are huge groups of neurons, controlling complex behaviours. So, how do we test these ideas?

We start with something simple. In our case, the neurons that control how a sea-slug crawls.


A sea-slug, running away. Credit: Angela Bruno

Imagine you’re a sea-slug. You’re a foot long bag of muscle and slime, that inches its way slowly across the sea floor. Question for you: how do you run away? You run away by reaching your neck out as far as you can, planting it firmly on the floor, then hauling your slimy arse towards it as fast as you can. And repeat. For sometimes up to a quarter of an hour.

There are many great things about this escape “gallop”. One is that we know exactly which part of the sea-slug’s brain entirely controls this repeated escaping movement. Another is that this part of the brain doesn’t contain many neurons — about 1800. Another is that we can make a sea-slug do this anytime we want by tickling its tail nerve with electricity, to make it think something scary bad is happening to its rear-end. And the final great thing is that its neurons are bloody huge.

Which all means my epically wonderful collaborators in Chicago — Angela Bruno and Bill Frost — could record from up to 200 neurons from this bit of sea-slug brain while the sea-slug thought it was escaping. That’s about 10% of the entire circuit that controls the escaping movement. Better, because the neurons are so huge, they could record every single bit of activity — every single spike — from them.

You couldn’t imagine a better set of data to answer the question: do we need to understand every neuron? Here that question becomes: do we need to understand every sea-slug neuron to understand how the sea-slug brain makes it crawl?


No. No we do not.

It turns out that, instead of needing to keep track of hundreds of neurons, we only need to keep track of around 5 or 6 numbers. Those 5 numbers tell us the current value of 5 hidden patterns in that hundred-plus set of neurons.

(How? Each moment in time, a bunch of neurons have the same activity — they are correlated. So we can represent that bunch at that moment by a single number representing “this set of neurons are all currently doing the same thing”. If we look across all moments in time, then it turns out that we can account for about 80% of all neurons and all time by just 5 such numbers. There are basically just 5 patterns in the data. But, no single neuron need have exactly any of those patterns — they are instead made up of the repeated correlations between neurons.)

The really cool bit is what those 5 hidden patterns together describe (remember, these 5 hidden patterns are happening at the same time). They make a spiral. A helter-skelter of neural activity. The activity of the whole group of neurons loops around over and over again, but falls on each loop, slowly but inexorably reaching a point where it will just. Stop.

We had discovered a spiral attractor. The activity was looping and falling: that’s the spiral part. But it was always the same spiral (or very similar) each time we started crawling in the same slug. That is, the spiral shape attracted the activity towards it. And when the activity wandered off the spiral briefly, it always came back. Again, the spiral shape attracted the activity back towards it.

The simple, hidden system in the sea-slug neural circuit for moving is a five-dimensional spiral attractor. (Yeah, I know: extrapolating two and three dimensional shapes into five or more dimensions makes my brain ache too. That’s why we have maths. The maths says its looping and the loops are getting smaller. It’s a spiral in five dimensions).

And that, we think, is what defines the galloping escape. Each loop of the activity is one complete repetition of the stretch-neck-plant-head-haul-arse cycle. How fast the activity passes around that loop is how fast the sea-slug would move (if it could move, which it can’t in these recordings, otherwise it would be damn hard for Angela having to chase it around the room with a microscope). How much the loop gets smaller on each loop is how fast this gallop is wearing off: the faster the loop shrinks, the faster the looping will come to a stop: and so will the gallop.


We loved this work. We’re also very happy that our paper came out in a prestigious journal. It means that more neuroscientists are likely to find it and, hopefully, to enjoy it. But the name of this particular journal will be utterly meaningless to most of you. And that’s how it should be. You should judge the quality of a scientific study on its own merits, not where it is published. Judge a study on whether its methods were sound; on whether its results could plausibly follow from the methods used; on whether the conclusions drawn actually follow from the results. (Sounds basic, but over a decade of reviewing papers for journals informs me that apparently only a small fraction of working scientists seem to understand this).

Here’s the happiest bit.
If you want to read the paper, and check everything I just said, you can. The journal we published in is open access, so anyone, anyone at all, can go to the website and read it there, or download it for you pleasure (or to print out and wallpaper a spare room).

Better, if you want to check that our new ways of analysing data could do what we said they could do, then you can. All of our analysis code is here. Download it, play with it, break it.

Better, if you want to check that our analysis of the data gives the results we said it did, you can. All of the data used for the paper is here. Download it, play with it. Replicate what we did, or find something new to science.


So we think we’re a tiny step closer to working out, if not the brain, then what we need to know about the brain in order to understand it. That we don’t need to know what every neuron in a brain region is doing in order to understand what that brain region is doing. That we can look for a hidden system, a simpler set of patterns within the churning, broiling mass of unfettered neural activity.

Which raises the rather fundamental question: if all we need to understand the brain’s activity are these hidden patterns, is that all it needs too? Does a brain represent everything by the state of a set of hidden patterns, that stretch from its input from the senses to its output to the muscles? Yes.

If you liked this, please click the ♥ so other people can read about it on Medium

Follow The Spike for more dispatches from the frontline of neuroscience