Sunday, September 14, 2008

From Spikes to Decisions: Part 2

Spatial Awareness

Space is an interesting concept, one we often don’t think about. It’s something that has to be calculated to do something as simple as walk across the room, or as complicated as walking across a city. It’s an abstract thing, and by that I mean that we have no sensory receptors for space; it’s something that has to be generated as an idea in our heads. Psychologists call what’s generated an internal map.

In 1971, the neuroscientists John O’Keefe and Jonathan Dostrovsky found nerve cells that fire differently according to where an animal is in an enclosure. This was particularly significant because the rat’s position in the maze could be measured, which allowed the researchers to correlate specific cell firing with specific locations. They dubbed these neurons, found throughout the hippocampus, place cells.


Resorting to a graphical representation is probably the best way to illustrate place cell function. I’ve borrowed a handy flash animation from University of Bristol’s spatial memory page to help with that (doesn't work in an RSS reader):





As the animal walks around the enclosure and checks out its environment, certain cells will fire at certain times, and by charting this firing relative to the animal’s position, the researcher can work out the place field of the neuron. Once a rat has learned a space, it can return weeks later and place cells will fire in a very similar way to the first time it walked around the space. Yet if the researcher moves the distal (far away) sensory cues—by rotating colored shapes that mark certain walls while the rat is out of the enclosure or distracted—the firing of these cells will change in proportion to the amount of rotation.

The above examples show that a rat makes the primary judgment of where it is via distal sensory cues and that place cell firing is directly related to this decision. Itzhak Fried, a UCLA neurosurgeon, and his collaborators have demonstrated the same principles in human epileptic patients.

We’re visual creatures, and for most of us it may be hard to separate ideas of space from our visual world, but internal maps and place cell firing aren’t wholly dependent on sight. Bats, whales, and blind people all navigate through space without visual cues. It may come as no surprise then, that researchers have shown that distal visual cues aren’t the only thing controlling the firing of place cells in rats. The input is polysensory, integrating all of the five senses, motor output to the limbs, the vestibular system (loopy things in the ear) and, of course, memory.

An illustration of this polysensory input is an experiment where the walls were quickly dropped as a rat ran down a long hallway. The walls and the symbols that researchers placed on them were all the rat could see before, and it would have a completely different environment after dropping the walls. If distal cues were all that determined the rat’s place cell firing, then they would have seen a change, but the cells continued to fire as they did before the walls were dropped. In this instance, the rat wasn't fooled because the act of running continuously was enough to let the rat know it didn't go anywhere new, and the cells fired the same way as a result.

To sum up this section, when place cell firing is overlaid on both time and location of a rat’s movements, there is the ability to begin deciphering the neural code of internal maps, and then move both down and upstream of the hippocampus to start figuring out the inputs. By doing this, scientists have found the major determinant input of place cell firing, grid cells.

Reward Prediction

Dopamine was long thought to be the chemical responsible for pleasure, but it’s now thought of as responsible for “wanting” or motivational urges. It’s also involved in other things, like movement and suppression of the hormone prolactin, but I’ll only refer to dopamine in its reward related role.

By recording the action potentials (spikes or firing) of dopaminergic cells of the substantia nigra and ventral tegmental area, two inter-related areas in the midbrain, scientists have been able to find what’s known as the dopamine reward error prediction signal. All the figures I’ll use below come from Wolfrum Schultz’s scholarpedia page on reward signals.

The least ambiguous way I can think of to illustrate this principle is to have the reader imagine a monkey sitting in front of a screen, with a little tube at his mouth for administering drops of juice. On the screen, different pictures are being flashed and after certain ones he’ll receive a drop of juice. Researchers call the pictures that are paired with juice a conditioned stimulus (CS) because there’s nothing inherent in a random picture that tells the monkey a reward is coming; he has to learn the relationship.

In the figure below, you can see how these cells respond to a reward that the monkey didn’t know was coming:

In this figure and the ones that follow, the individual cells being recorded are shown in rows from top to bottom. The spikes of these cells are shown left to right as they occur over about 3 seconds. At the top they’re all summed, so it’s easier to see when a lot of them fire together.

Because the monkey doesn’t know that the picture means a reward is coming, the cells fire a lot in response to the surprise juice. This is pretty much the pattern they follow, until the monkey has learned which pictures come before a drop of juice. Once the monkey learns, the dopamine cells fire like this:

Two things are very clear here. The first is that the cells all fire at essentially the same moment in response to the conditioned stimulus. The second is that the reward elicits no change in the normal firing of the cells. From this figure, the idea that dopamine doesn’t correlate with the pleasure of a reward should make sense.

Dopaminergic firing will continue in this pattern if the experimental conditions stay the same, and a reward always follows a conditioned stimulus. The experiment can be modified, however, and a stimulus that used to signal a reward will no longer do so.

When this happens, the monkey’s dopaminergic cells show an interesting firing pattern:

The response to the CS is the same, but there’s a period where the neurons are completely shut down because there was no reward when it was expected.

All three of these responses are distinctly correlated with stimuli in the world, in the same way that place cells fire in specific locations.

These two wonders of the brain, place cells and the dopamine reward signal, represent the neurological correlates of mental percepts for which there are no sensory organs. Researchers have spent several decades refining knowledge of both so that we now know just how to find the cell activity and, for the most part, what it means.

Can we take these two signals and look at them simultaneously to discover anything useful?

This is the question that a recent paper by Adam Johnson and David Redish asks. It also looks at how you would frame and experiment to find out. I think it's a very exciting proposal, and I'll go over it in part 3 of this series, along with some of my own thoughts on the matter.

For more information on either of the topics I talked about above, I recommend checking out University of Bristol's spatial memory page, and Wolfrum Schultz's scholarpedia article on reward signals, which are both linked below.

References:

University of Bristol’s neuroscience website: http://www.bristol.ac.uk/synaptic/research/projects/memory/spatialmem.htm

The Mosers' annual review article on Place Cells, Grid Cells, and the Brain’s Spatial Representation System: dx.doi.org/10.1146/annurev.neuro.31.061307.090723

Schultz’s scholarpedia page:

http://www.scholarpedia.org/article/Reward_signals

Schultz’s annual review article on reward: dx.doi.org/10.1146/annurev.psych.56.091103.070229

No comments: