Close your eyes, while looking out of the corner of your eyes. Now rub on your eyeball lightly, and you should see simple patterns of light called flicker phosphenes. This is because you are stimulating a corner of your retina, even though you aren't really looking at anything. Similar patterns have been observed by people in various situations, for example, shortly after ingesting psychedelic drugs such as LSD or marijuana, or after being exposed to flickering light. Recently, Jack Cowan, a mathematician and neuroscientist at the University of Chicago, has built a neural network model that can literally trip out. His work provides answers to why these hallucinations occur, and how they are a natural consequence of the architecture of our brains.
Hallucinations come in an endless variety and it seems impossible to even try to characterize them. But in the 1920s, Heinrich Kluver, a neuroscientist at the University of Chicago interviewed several people who had taken the drug mescaline, and even took it himself. Keeping a commendably straight head, Kluver found patterns in the patterns. He classified the patterns into four distinct categories: honeycombs, cobwebs, tunnels and spirals. He called them form constants for their surprisingly common occurrence across all subjects, especially in the early stages of a trip.
Several decades later in the late 1970s, Cowan and his graduate student, Bard Ermentrout (who, incidentally, is a Hopkins alumnus) were working on pattern formation due to convection in heated oil, when they came across illustrations of Kluver's patterns. Patterns are formed in heated oil when the oil particles come under the influence of two opposing forces: one, due to heat, which causes them to move rapidly and rise up, and the other, due to diffusion that causes the particles to move slower. Turning up the heat in a pan of oil gives rise to stripes and honeycombed patterns.
The similarity was striking; so they began with Kluver's hypothesis that patterns of activity of neurons in the visual areas of the brain might induce the perception of these form constants. However, they also had to explain the fact that these hallucinations are perceived only under specific conditions and not all the time. Therefore, they argued, that ingesting psychedelic drugs would cause an instability in the neuronal activity, leading to spontaneous waves of activity. This is somewhat akin to turning up the heat in a pan of oil, which has been known to give rise to honeycomb and stripe-like patterns. They also conjectured that this spontaneous activity must arise in the primary visual cortex, called V1, located at the back of the head, which is the first area of processing to which neurons from the retina send their connections.
Cowan and Ermentrout developed a model which reflects a general property of the wiring of the brain: each neuron is under the influence of two opposing forces: one from its nearest neighbors, that tends to excite it beyond its critical threshold and cause it to produce activity, and the other, from its distant neighbors, which induces the opposite effect and inhibits it. Thus a neuron must be under a delicate balance between excitation and inhibition to prevent it from being constantly active, or from being completely silent.
Drugs such as LSD upset this balance. One of the effects of these drugs is to, in effect, reduce the threshold for the neurons to become active. As a result, the visual cortex produces spontaneous patterns of activity, unrelated to the outside world. These patterns of activity are presumably interpreted by the higher visual areas as being real visual phenomena. But why don't we see stripe-like patterns at all, if the analogy of the convection patterns is correct? Because these patterns occur in the visual cortex, not the retina. A lot of cortical processing is devoted to the center of the visual field, where our vision is sharp, and the number of neurons devoted to each point in the visual field falls off towards the periphery. (This is why our peripheral vision is poor; interestingly, we have many motion-sensitive neurons in the periphery, enabling us to detect - and turn towards moving objects in our periphery.) Using this transformation, Cowan and Ermentrout mapped these patterns back to the retina, and came up with spectacular results: stripes of activity in V1 would correspond to spirals, bull's eyes, tunnels and even checkerboards, depending on the orientation of these stripes.
However, their early model could not explain all the form constants, for instance some checkerboard patterns and cobwebs. Recently, in a report published in the Proceedings of the Royal Society, Cowan and his colleagues have made a more realistic model, incorporating a wealth of new data collected by neuroscientists in the recent years. They built a model of neurons with preferences for edges of different orientations, with the same general principle of network interconnections. Cowan's computer model reproduces most of the hallucinatory patterns that have been documented by Kluver and many others.
There are other interesting issues that need to be addressed. We have two area V1s, one in each hemisphere of the brain, and each receives input from one half of our visual field. So how do hallucinations look continuous across both halves of our visual field, if they are produced by spontaneous (and therefore unrelated) drug-induced activity? One possibility is that the two discontinuous sets of activity are integrated together by areas further up in processing that give rise to a continuous percept. Alternatively, continuity may be established by communication lines running between the two areas through the corpus callosum, a bundle of fibers running between the two halves of the brain. However, which of these (or any other) mechanisms really underlie this continuous percept are not understood at this point. There is another crucial link, which is to actually observe these patterns of activity in V1 during hallucinations.
What does this tell us about the brain? Early neuroscientists went to extraordinary lengths to characterize neuronal connectivity and single neuron activity. But we need more than just a knowledge of wiring, or of how individual neurons work: Cowan's work is just one of many studies that are bringing out a beautiful interplay between architecture and function in the brain, and are explaining how the collective behavior of a network of neurons can emerge as a consequence of their connectivity.
Further reading:
Bressloff PC, Cowan JD et al (2001) Geometric visual hallucinations, Euclidean symmetry and the functional architecture of the striate cortex. Phil. Trans. R. Soc. London B 356: 299-330.
---
SP Arun
[published in the Johns Hopkins Newsletter, September 14,2001]