Reversal learning: when the platform moves

Opening

Day six of training, and the mouse has the pool worked out. It slides off the gloved hand, turns once, and swims a straight line toward the north-east. The water is cool and opaque, the colour of weak tea with milk in it. The lights are low. Posters and a shelf hang on the walls around the tank. The mouse reaches the spot where the platform has been all week. Its paws find nothing. The platform is still in the pool. It has simply moved, to the far side, opposite the old place. The animal circles. It circles again. What it does in the next few minutes is the whole point of today.

What reversal learning is

This is reversal learning in the Morris water maze. The idea is simple. First, an animal learns where a hidden platform sits. Then, without warning, the platform is moved, usually to the opposite quadrant. The animal now has two jobs. It has to stop searching the old place. And it has to learn the new one.

A single reversal trial looks almost exactly like an ordinary training trial. The animal is released facing the wall, from one of several start points. It has up to a minute or so to find the platform. If it fails, the experimenter guides it there. It rests on the platform for a few seconds, so it can take in the room, and then it's lifted out. The difference is that the room, the cues and the routine are all the same as before. Only the answer has changed.

Researchers usually run a few days of reversal training, often with four trials a day. Many labs also add a probe trial at the end, with the platform removed, to see which place the animal now prefers. Some studies go further and move the platform again, and again, to see how quickly each new answer replaces the last.

Where reversal came from

The pool itself came from Richard Morris and colleagues, in the early nineteen eighties. The original design asked a clear question. Could a rat find a place it couldn't see, using only cues around the room? It could, and that became the backbone of decades of memory research. But almost from the start, people wanted more than a single answer learned once.

Reversal grew out of a practical worry. An animal that swims straight to the platform might be using a precise map of the room. Or it might have learned a habit, a fixed heading from the usual start points, that happens to work. Moving the goal pulls those apart. A flexible animal soon gives up on the old spot and starts exploring. A rigid one keeps returning to it, trial after trial, like someone who still reaches for a light switch on a wall that was redecorated months ago.

By the time Vorhees and Williams wrote their widely used protocol paper in 2006, reversal was a standard add-on. They described it as a way to test whether the animal could learn a new location, which they linked to cognitive flexibility. Many groups now treat it as a gentle stress test for the system. It asks not just whether memory was formed, but whether memory can be updated.

What a reversal shows

So what does a reversal actually show? Start with the most familiar number, escape latency. On the first reversal day, latency usually jumps. That's expected. The animal swims to the old place first, finds nothing, and only then starts searching. Over the next few days, latency falls again. A group whose latency stays high for longer may be slower to adapt.

Latency alone hides a lot, though. Picture two mice on day one of reversal. Both take about thirty seconds to reach the new platform. The first spends twenty of those seconds circling the old spot, then heads off and finds the new one. The second barely visits the old spot at all and simply searches widely before stumbling onto the goal. Same latency. Very different behaviour. One is holding on to the past. The other let go fast but hasn't yet learned the new place.

That's why reversal studies lean on measures that look at where the animal goes, not just how long it takes. Time in the old target quadrant, compared with time in the new one, shows the shift of preference across days. Crossings of the old platform spot show how often the animal checks the previous answer. Average distance from the old spot, and from the new one, gives a smooth picture of that shift without cutting the pool into rigid quarters. An analysis by Maei and colleagues in 2009 found that distance-based measures like these often picked up differences that quadrant time missed.

Path shape matters too. Researchers classify searches into strategies: a direct swim to the goal, a focused search around one spot, a broad scan of the pool, or hugging the wall, which they call thigmotaxis. In reversal, the interesting moment is the switch. Does the animal move from a focused search at the old place, to scanning, to a focused search at the new place? Or does it fall back on scanning and wall-hugging for days?

That idea of strategy came through clearly in a 2009 study by Garthe, Behr and Kempermann. They reduced the birth of new neurons in the adult hippocampus of mice, then looked closely at search patterns. The mice learned the first platform position fairly well. The difference showed up when the platform moved. Mice with fewer new neurons used fewer spatially precise strategies in the reversal phase. The standard latency curves were only mildly affected. The strategy analysis was where the story lived.

Here's a worked example. Suppose a probe trial after reversal training lasts sixty seconds. A well-adapted mouse spends about twenty-five seconds in the new target quadrant and about eight in the old one. Chance is fifteen seconds per quadrant. Another mouse spends eighteen seconds in each. It has learned that the new place matters, but it hasn't stopped believing in the old one. On a single quadrant score, those two might be called similar. Side by side, they're clearly not.

Where it can mislead

Reversal is useful, but it can mislead. The first catch is that reversal depends on the original learning. If one group learned the first position less well, it may seem more flexible simply because it had less to unlearn. Or it may look worse because it's still learning the basics. It's worth checking that both groups reached similar performance before the platform moved.

The second catch is time. Each reversal adds days to a study. Animals get older, more handled and more familiar with the room. Swimming ability, vision and motivation all still matter. A visible-platform session helps rule out an animal that simply can't see well or won't swim. Floating and long pauses should be measured separately, so a tired animal isn't scored as a stubborn one.

The third catch is interpretation. A slow reversal is often described as a problem with flexibility, and it may well be. But it can also reflect a stronger original memory, a different strategy that worked fine before, or plain anxiety in the pool. A fast reversal isn't always a good sign either. An animal with a weak first memory has very little to defend. That's why reversal results are best read alongside probe trials, path plots and strategy measures, not as one headline number.

Finally, reversal tests one kind of change: the same room, a new goal. It doesn't tell us much about how an animal copes with new cues, new rooms or new rules. For those questions, other designs, such as cue rotation or a fresh pool layout, may suit better.

Back to the empty corner

Back to that mouse, circling the empty north-east corner of the pool. Its first few loops aren't failure. They're evidence of a memory doing its job. What matters is what comes next: how long it keeps checking, when it starts to explore, and how quickly a new straight line appears across the water. Reversal turns one simple trick, moving a small platform, into a window on how memories bend.

You'll find the papers we mentioned, and a transcript, at water-maze.com. hvsimage.com describes reversal and multiple-target analyses in its water maze software, if you'd like to see how those measures are laid out.