The probe trial: when the platform disappears

Opening

It's the fifth morning. The pool is the same, the posters on the walls are the same, and the water is just as cloudy. Only one thing has changed. Overnight, someone reached in and lifted out the platform.

A mouse is lowered in at the far side. It heads off with purpose, straight across the pool, to a patch of water near the south-east wall. It slows. It turns. It sweeps back and forth over the exact spot where the platform used to be. There's nothing there, but it keeps looking, circling tighter and tighter.

To anyone watching, that empty patch of water is the most interesting place in the room. The mouse is showing us, as plainly as it can, that it remembers.

What a probe trial is

This is the probe trial, sometimes called the transfer test. It's the quiet moment at the end of water maze training when the reward disappears and memory has to speak for itself. Why take the platform away, and how do you score a search for something that isn't there?

The standard set-up

Picture the standard set-up first. During training, an animal learns to find a hidden platform in a fixed place, using cues around the room. Each trial ends when it climbs out. That's useful, but every trial also mixes two things together. One is memory for where the platform is. The other is the business of getting there, which depends on swimming speed, motivation and luck.

The probe trial pulls those apart. The platform is removed, so the trial can't end early with an escape. The animal is placed in the water, often at a new start point, and allowed to search for a fixed time. Thirty or sixty seconds are common choices. Then it's lifted out.

What's left is a record of where it chose to look. If it learned the location, its search should cluster around the old platform site. If it didn't, its path will spread more or less evenly across the pool.

A widely used protocol suggests giving the probe a full day after the last training session. That way it measures longer-term memory, not just what happened a few minutes earlier. Some labs also slip in short probes during training, before the first trial of the day, to watch memory build up.

Notice what the probe doesn't need. It doesn't need the animal to climb out, or to be fast, or even to be especially keen. It only needs the animal to swim and to choose where. That makes it a surprisingly gentle test, given how much it can reveal.

Part of the maze from the start

The probe has been part of the water maze from the start. Richard Morris and colleagues, in the early nineteen eighties, used it to show that rats were truly remembering a place. A methods paper Morris published in 1984 developed the procedure further. It described automatic tracking of a rat's position from video, and ways of testing working memory and non-spatial learning in the same pool. Once you could follow a swim path frame by frame, you could measure exactly where an animal searched.

The early measures were sensible and simple. Divide the pool into four imaginary quarters, called quadrants. Then count how much of the probe the animal spends in the quadrant that used to hold the platform. With four quadrants, chance is about twenty-five percent. A well-trained animal might spend twice that.

Put that in seconds. In a sixty-second probe, chance would be about fifteen seconds in the target quadrant. An animal that spends thirty seconds there is showing a clear preference. One that spends sixteen is telling you very little, even if its training times looked fine.

The second classic measure is platform crossings. Draw an outline where the platform used to be, and count how many times the animal swims right over it. A good memory should produce lots of crossings. A poor one, very few.

For years, those two measures dominated. Then a question started to nag. Which measure is actually best at spotting a real difference between groups of animals?

Which measure is most sensitive?

In 2009, Hamid Maei, Paul Frankland and colleagues at the Hospital for Sick Children in Toronto set out to answer it. They gathered five separate databases, holding more than sixteen hundred mouse probe tests. Then they simulated experiments again and again, drawing animals at random from those databases. It's a technique called Monte Carlo simulation. It let them ask, over and over, how often each measure found a real group difference.

They compared four measures. Quadrant time, which we've met. Zone time, which uses a smaller circle centred on the old platform. Crossings. And a fourth called proximity, which is simply the animal's average distance from the platform site across the whole probe.

The winner was proximity, and by a clear margin. In one comparison, with fifteen mice per group, proximity detected the group difference about seventy-two percent of the time. Quadrant time managed about forty-nine percent. Crossings came in at around thirteen percent. False alarms, meanwhile, were similar for all four.

Why would proximity do so much better? The authors offered a neat explanation. Quadrant time and crossings were invented when tracking was crude. Quadrant time throws away detail. A swim just outside the line counts for nothing, while a swim in the far corner of the right quadrant counts in full. Crossings are even harsher, since a near miss scores zero.

Proximity keeps the detail. Every moment of the swim contributes, and being close is rewarded more than being far away. Think of two mice. Both spend thirty percent of the probe in the target quadrant. One hovers tightly over the old platform site. The other drifts round the outer edge of that quadrant. Quadrant time can't tell them apart. Proximity can.

What a good probe shows

So what does a good probe trial show? It shows that an animal has a memory for a location, independent of how fast it gets there. Swim speed doesn't change where the animal chooses to search. That's why, in the protocol literature, a deficit in the probe is seen as strong evidence of a true memory problem rather than a performance problem.

It can also show how precise that memory is. A broad, loose search suggests the animal knows roughly where to go. A tight focus on the exact spot suggests a sharp, metric memory. Researchers interested in subtle effects, such as early ageing or mild genetic changes, often care about that difference most of all.

And the probe can show change over time. Run one at the end of training, then move the platform to the opposite side and train again. A second probe after that reversal shows whether the animal has updated its map, or is stuck on the old answer.

The same logic works at different scales. Researchers can compare the target with each of the other three quadrants, rather than just with chance. A strong memory shows up as one tall bar and three short ones. A weaker memory might show a preference for the target and its neighbours, which hints at a rough sense of direction without a precise location.

The limits of the probe

The probe isn't flawless. For one thing, every probe trial is a lesson. The animal searches and finds nothing. Do that too often, and it may start to learn that the platform has gone for good. Protocols warn against running too many probe trials for exactly that reason.

Time matters within the trial too. Search often starts focused and then spreads out, as the animal gives up on the old spot and explores elsewhere. Some evidence suggests the preference for the target quadrant fades after the first thirty seconds. A sixty-second probe may dilute the very effect it's looking for, so many labs look at the first half separately.

The choice of start point can bias things as well. Start an animal right beside the target quadrant, and it may wander in by chance. That's why probes usually begin from a new position, often on the opposite side.

Then there are animals that don't really search at all. Some float. Some hug the wall. Some circle at a fixed distance from the edge, which can rack up crossings without any real knowledge of place. A single summary number can hide all of this. Looking at the actual swim path, and at search strategy, is the only way to be sure what the animal was doing.

And proximity, for all its strengths, has a caveat. It depends on where the animal starts, because a mouse that begins far away carries a long first stretch into its average. Some analyses deal with that by ignoring the opening seconds or correcting for start distance.

Back to the empty patch

Back to that empty patch of water near the south-east wall, and a mouse sweeping back and forth over a platform that isn't there. It's a strangely moving sight. The animal is acting on a memory, and the probe trial turns that memory into a map we can read. The trick is choosing a measure that sees the whole search, not just the tidy bits.

hvsimage.com has a detailed guide to the probe measures, including proximity. You'll find the papers we mentioned, and a transcript, at water-maze.com.