When Cortex Stops Making Sense
When cortex is useless, what do we learn?
The cortex, that many layered cake of neurons, occupies a special place in neuroscience. Our best guess at the root of our prodigious intellectual abilities — of language, of sculpture, of base-jumping in a kilt — is that our cortex is far more dense with neurons than any other species. Our dominant sense, sight, commands a third or more of the entire cortex. And its exquisite circuitry hints at devilishly complex computations, of probabilities, of inference, of algorithms that AIs can only dream of. If we want to understand the brain from a human perspective, the cortex is where we must start.
(The fact that it is also by far the easiest part of the brain to record from in both animals and humans, as it’s right at the top, is entirely coincidental, and shall not detain us further here.)
So it’s a right bugger when it turns out cortex is essentially useless.
Writing in Nature, Kate Hong and colleagues in Randy Bruno’s lab have reported just that: a mouse can happily use a single whisker to solve a problem even when the special bit of cortex representing that specific whisker is missing. Worse, the mouse can even learn to solve the problem from scratch, without apparently noticing it has a hole in its cortex where lots of important stuff should be happening.
Is cortex then just a blanket to keep the rest of the brain warm? (Peter Redgrave, on innumerable occasions). No, but damn we have to be careful about what we claim cortex is necessary for.
Hong and friends set out to answer a different question: what is the difference between temporarily and permanently turning off a bit of cortex? If you want to do such complex monkeying around with a bit of cortex, then you need a simple behaviour to study, and you need to be able to turn off a bit of cortex that should be involved in that behaviour.
Now you may know that mice have whiskers. The nice thing about whiskers is that each big whisker has a set of dedicated neurons in cortex, which tend to be active only when that big whisker is tweaked, dabbed, or smacks into something. And it is (relatively) easy to find those neurons, because you can tweak the whisker and find which bit of cortex lights up. The bit of cortex representing one big whisker is then a juicy target for turning it off to see what happens.
So Hong and friends set their mice the task of detecting a pole with one big whisker. On each trial, the mouse held down a lever to say it was ready to take the test. Then either the pole swung within range of their big whisker, or it swung the other way, out of reach. When the pole was in range, it meant reward was available. If the pole was out range it meant there was no reward available right now, and the mouse had to keep holding down the lever until the pole reset — then it could let go and, in its own time, start again.
The mouse’s test was thus to swish its whisker and see if it hit the pole. A hit should mean to the mouse “aha! I can go and get a reward”; a miss should mean “aha! There is no reward; I’ll wait right here and keep pressing down this lever for some arbitrary reason known only to those dorks over there in white coats”.
Mice had to learn all this by trial and error — pressing a lever, swishing the whisker, hitting a pole (or not), and letting go of the lever (or not). They learn this quickly. Quicker than some of us, I’d wager.
Once the mice had learnt, Hong and friends permanently turned off the big whisker’s cortical neurons by removing them. And lo, performance immediately went to pot. The mice made the wrong decisions about whether the pole was hit or not, and got far fewer rewards as a result.
Nothing unexpected there. Monkey around in cortex and it makes a mess of performance, big whoop. But now the kicker: the terrible performance was on the first day after removing the neurons. On the second day after removal, the mice recovered completely. They performed just as well as they’d done before anyone monkeyed around with their cortex. And carried on doing so for all the following days.
Which led Hong and colleagues to the crucial question: er, if mice can apparently re-learn the task in a matter of a day, did they need this bit of cortex to learn in the first place?
No, they didn’t. Removing the big whisker’s neurons from cortex before any training had no effect on learning at all. Mice learnt equally fast and equally well with or without that bit of cortex. Sheesh.
The result of this study may seem simple, but it has many ramifications. There are lessons here for any serious student of how brains work.
Lesson one is that establishing causality is a pain. You’d think that turning off a part of the brain and seeing that a behaviour breaks would be big kahuna levels of evidence that the brain part in question causes that behaviour. But no. Here Hong and co have shown us, again, that turning off some brain bit does not establish causality.
To reinforce how misleading this turning off can be, in a separate set of experiments Hong used the terrific toolbox of modern neuroscience to temporarily turn off the cortical neurons representing that big whisker, but only on some, randomly chosen, trials (AKA they expressed a light-sensitive opsin in the pyramidal cells of barrel C2, which suppressed spiking in those neurons when activated with a light). And performance went to pot on just the trials when the neurons were turned off, remaining just fine on the rest of the trials. Read uncritically, this would imply that the big whisker neurons were essential for doing the task — after all, whenever you turn them off, performance tanks. But we know this is not true: the same bit of cortex could be turned off completely, and yet the mice still could do the task.
Lesson two is that establishing causality is a pain. Even when the big whisker’s neurons were turned off or removed, the dip in performance was not back to random guessing. The mice still did better than chance, so something else in the brain was still able to cope with the task a bit. Indeed when Hong and co temporarily removed the big whisker itself, then the performance of the mice collapsed completely, and they did no better than guessing randomly where the pole was. Which, of course, is what they were doing.
Lesson three is that establishing causality is a PAIN. Notice that removing the big whisker neurons immediately impaired the mice who had already learnt the task. But removing the big whisker neurons before learning did not have any effect on learning itself. Just because a brain region seems to be involved in performing a task does not mean it is critical to learn that task. And vice versa: a brain region may be necessary to learn it, but not do it.
Lesson four is that ESTABLISHING CAUSALITY IS A PAIN. As previously noted by Otchy and co, when performance tanks because you’ve shut down a bit of brain, but recovers quickly, then it actually establishes the opposite of what you wanted to do. It means that turning of that bit of cortex interfered with the normal function of a different bit of the brain — the bit that could do the task perfectly well without cortex. So all you’ve shown is that the bit of cortex you turned off is annoying a different part of the brain. And when you shut off the normal signals from that bit of cortex, it throws another bit of brain out of whack.
Our final lesson is the deepest: degeneracy. Or, the brain has multiple solutions to the same problem. Signals from the whisker go to many places in the brain, not just cortex. They go to various bits of thalamus and brainstem, that then send the signals on to other complex bits of brain that aren’t cortex (like striatum or the superior colliculus). Hong and co’s data tells us that a task as simple as “is the pole there or not” doesn’t need the big whisker signals to be routed to the big whisker parts of cortex — there are other bits of brain that are equally capable of solving the problem.
It seems we’ve learnt a simple rule about the mouse brain: if cortex is there, use that; if not, something else can cope. The question is, how complex does a task have to be before only cortex can cope? The answer is almost certainly Jean-Pipsqueak Sartre’s hell: dealing with other mice.
Want more? Follow us at The Spike