COMEDIANS, choreographers and composers will all tell you the same: the
secret鈥檚 in the timing. It鈥檚 not just words or notes or pirouettes in isolation
that carry the message, but the rhythm, the pace鈥he pauses.
A small but vocal group of neuroscientists now argue that the same holds true
for the neurons that shuttle messages around the brain. They claim that their
experiments have resurrected a long-discarded theory that information is encoded
in the timing of nerve impulses. If this controversial 鈥渢emporal coding鈥� model
is correct, they say, then the brain is far faster and more powerful than anyone
ever envisaged.
For decades, most neuroscientists assumed that the exact timing of nerve
impulses in the brain didn鈥檛 matter. Instead, they claimed that each neuron鈥檚
message was carried in the average number of electrical impulses it sent to its
neighbour over a set period. According to this model, a neuron is like a bucket
beneath a hole in the roof. When it rains, the bucket fills drop by drop, but it
is only emptied when the water reaches the top. In the brain, the model says,
each neuron collects all the inputs from neighbouring cells, impulse by impulse,
and only passes its message to the next cell when a threshold is
reached鈥攖he so-called 鈥渋ntegrate-and-fire鈥� theory.
Advertisement
But according to the temporal coding enthusiasts (who include some of
neuroscience鈥檚 leading lights) the bucket theory is wrong, or at the very least
incomplete. They believe that the exact timing of each nerve impulse in relation
to the others encodes information in the brain, and that neurons effectively act
like parts of a telegraph system that read the incoming impulses, or 鈥渟pikes鈥�,
as if they were the dots and dashes of Morse code. Rather than waiting for a
threshold to be reached, each neuron translates the information it receives from
the other neurons moment-by-moment, quickly sending the message along to the
next 鈥渞eceiver鈥�.
With this system, each neuron has the capacity to encode up to a hundred
times more data per second, says theorist William Softky of the National
Institutes of Health in Bethesda, Maryland. Multiply that by the billions of
neurons in the brain, and the brain鈥檚 computational capacity soars to
unimaginable heights. And it doesn鈥檛 end there. Neurons that don鈥檛 need to spend
time waiting to reach a threshold of stimulation before firing can send
information faster, which also increases computational power. 鈥淚t鈥檚 a
two-for-one kind of deal,鈥� says Softky.
In fact, the controversial temporal coding theory has been around in various
shapes and forms for nearly fifty years鈥攊t has simply failed to gain
widespread acceptance. There have been two problems. First, there doesn鈥檛 seem
to be any need to imagine a brain that is more powerful than that allowed by the
integrate-and-fire bucket model, which is already pretty impressive. Second,
until recently all the available evidence suggested that the bucket model was
correct.
All that has changed, according to Softky, Terry Sejnowski, a neuroscientist
at the Salk Institute for Biological Studies in San Diego, and others. They say
that it wasn鈥檛 that the temporal coding model was wrong but that the experiments
designed to test it were inadequate. By studying how nerves in rats, flies and
frogs respond to very complex stimuli鈥攍ike those an animal encounters in
real life鈥攖hey have amassed fresh evidence that supports the idea of a
temporal code in the brain.
In most of the early experiments, researchers gave neurons very simple
electrical stimuli that were unlike most natural inputs. It was these
experiments that led to the belief that neurons operate on an integrate-and-fire
system. Neuroscientists were aware of one or two exceptions, however. For
example, they knew that circuits in the base of the brain of the barn owl work
out the direction of a sound from the difference in timing of the nerve impulses
from the left and right ears.
But the real controversy has always been whether the timing of impulses plays
a role in intellectual thought and consciousness, and in integrating information
from the eyes, ears and other sense organs to create the big picture of the
outside world鈥攊n short, whether the time code operates in the brain cortex
of animals like humans. The cortex is certainly the part of the brain most in
need of an efficient system for sifting and sending vast streams of encoded
information, but it is also the most complex and the least understood.
That didn鈥檛 stop Sejnowski and his student Zach Mainen, who is now a
neuroscientist at the Cold Spring Harbor Laboratory in New York, from
investigating whether a time code could operate in the cortex of rats. Last
year, they ran a series of experiments in which they attached electrodes to
neurons in pieces of rat cortex, and then subjected them to an electric stimulus
designed to imitate the kind of complex jumble of information a neuron in the
brain of a live animal might receive. In response to a given input, each neuron
reproduced exactly the same pattern of impulses with an accuracy to within a
millisecond. That showed for the first time that in the cortex it is possible
that information is transmitted from one neuron to the next in the precise
timing of the spikes. If the pattern of spikes had varied then the researchers
would have had to concede that temporal coding was impossible, says
Sejnowski.
However, showing that neurons in a Petri dish behave in a certain way is one
thing, but showing that the same holds true for neurons in a living, breathing
beast is quite another. Take flies. They spend a lot of time flying, and their
brains spend a lot of time rapidly processing the sensory information needed to
keep them on course as they buzz about.
Not only do the flies need to see where they are going, they also use visual
cues to check that they are flying straight and to ensure that they don鈥檛 go
into dangerous pitches and rolls. In fact, a fly鈥檚 neurons are so fast that it
takes the insect less than 30 milliseconds鈥攖he length of time a single
video frame appears on a TV screen鈥攖o see a movement, process the
information and send an order to its muscles to respond. A human would take at
least 150 milliseconds to do the same thing.
A system in which neurons must translate ever-changing visual stimuli into
the instructions needed to stop a fly crashing seemed just the sort that might
rely on a rapid-fire Morse code. And about five years ago, biophysicist Rob de
Ruyter and theoretical physicist William Bialek of the NEC Research Institute in
Princeton, New Jersey, decided to find out if that was indeed the case.
To get their experimental subjects to cooperate, Bialek and de Ruyter
anchored them to the lab bench with blobs of sealing wax. They then showed the
flies videos that simulated the types of movement flies might experience in
flight鈥攋iggling patterns of light and dark lines. While the flies watched
the video, the researchers recorded the electrical activity in a type of neuron
in their brains that extracts information from visual patterns for use in
controlling flight muscles.
After the video show, Bialek and de Ruyter fed a portion of the neuron鈥檚
spike train鈥攖he pattern of electrical impulses it generates鈥攖hrough
a simple decoding computer program. They found that the information contained in
a single neuron鈥檚 output could be used to accurately reconstruct the jiggling
movement seen in the original movie, showing that the timing of the nerve
impulses encodes information in a reliable form in the visual system of the fly.
In fact, the timing of each spike was so precise that when Bialek fiddled with
the timing and delayed a single spike by a couple of milliseconds, and then ran
the train through the decoding program, the reconstructed video changed
dramatically.
Bialek wasn鈥檛 surprised. 鈥淲hen the stimulus varies rapidly, individual spikes
need to count, because the sensory world may change drastically from spike to
spike,鈥� he argues. In real life, each fly neuron may only fire once or twice
before incoming stimuli have changed, so if the neurons wait to amalgamate
several impulses before sending the message on, as suggested by the
integrate-and-fire theory, heaps of vital information about the fly鈥檚 flight
path could be lost.
Instead, its neurons are working at maximum efficiency and, according to
Bialek鈥檚 calculations, are cramming as much information into a single spike
train as possible. 鈥淵ou couldn鈥檛 do much better if you replaced them with the
most physically perfect components at every stage. They鈥檙e doing as well as the
laws of physics will allow,鈥� agrees Softky.
The same goes for frogs鈥攃reatures a bit more complicated than flies,
brain-wise. Fred Rieke at Stanford University, California, and Deana Bodnar at
Cornell University in Ithaca, New York, told the March meeting of the American
Physical Society in St Louis how sedated bullfrogs serenaded by tones that
roughly mimic the sounds of a pond at night, produce spike trains in certain
neurons that are so invariable they can be used to work back to what the frog
had originally heard.
Despite their success in the hearing and visual systems of frogs and flies,
and in the brain cortex of rats, the temporal coding advocates have had a harder
time finding evidence that the neuronal Morse code is important in the thought
processes of primates.
Neuronal skirmish
In 1989, Michael Shadlen and William Newsome of Stanford University showed
that the spike trains of neurons in the cortex of male monkeys watching a moving
pattern on a television screen were never the same. In fact, Shadlen says, they
saw about as much variation as it is possible to see, which suggests that
information cannot be encoded in the timing of the spikes.
What is more, Shadlen, who is now at the University of Washington in Seattle,
says that he can explain the results entirely on the basis of the
integrate-and-fire model. 鈥淥ur current understanding of cortical circuitry
suggests that spike timing conveys little, if any, information,鈥� he says. He
also argues that the integrate-and-fire system can transmit information as
rapidly as the temporal code because in reality multiple neurons transmit the
same piece of information, allowing each neuron to detect changes in the rate of
incoming signals almost instantaneously.
But believers in temporal coding are not yet willing to accept defeat in
primates. Softky and others argue that there are other explanations for
Shadlen鈥檚 results. Monkeys are complex animals. Like humans, they may rely
heavily on mental associations to interpret the world around them. For instance,
in Shadlen鈥檚 experiment, the first time the monkeys see the video they may be
thinking about their mates; the next time, a recent skirmish with another
monkey. The two associations could trigger different spike patterns.
Shadlen, however, doesn鈥檛 buy these explanations and instead likens the
search for meaning in the timing of spike trains to reading tea leaves. And his
experiment is not the only hurdle that the time code supporters have been forced
to sidestep or tackle head-on with new experiments. Most of the information that
is shunted around the brain of sophisticated creatures like frogs or humans
relies on the combined efforts of complex circuits of neurons, rather than
single neurons firing one at a time. In the cortex of the human brain, for
example, each neuron can receive inputs from as many as 10 000 other
neurons.
Handling such a hodgepodge of information isn鈥檛 a problem if neurons do
indeed operate in an integrate-and-fire fashion. Each simply collects the
incoming impulses until it reaches its threshold (or fills the bucket) and then
fires. If, on the other hand, neurons really do keep track of the precise
arrival time of incoming impulses鈥攁s the temporal coding theory
suggests鈥攔eceiving multiple inputs could get confusing.
Neuroscientist Markus Meister of Harvard University believes that he knows
why each neuron doesn鈥檛 crumble under the avalanche of incoming data. Meister
and neurobiologists Denis Baylor and Leon Lagnado at Stanford University moved
out of the cortex and back into the visual system for their experiments. They
removed the retinas from salamanders鈥� eyes, placed them in Petri dishes, and
exposed them to flashes of light or flickering checkerboard patterns. Then,
rather than record the electrical activity in a single neuron, the researchers
simultaneously recorded the responses from 50 neurons.
They found that neurons that live near one another in the retina tend to fire
synchronously in response to visual stimuli. In fact, half the spikes were
time-locked with spikes from neighbouring neurons. Meister and his colleagues
calculated that the retinal neurons fired within a millisecond of one another up
to twenty times more often than would be expected by chance.
No one knows how the neurons get in step in the first place, or even whether
synchrony occurs throughout the cortex, but if it does, it may prevent precisely
timed spikes from getting lost, because they would all arrive at the receiving
neuron at the same time. Sejnowski likens the process to tuning a radio receiver
to a foreign station. 鈥淵ou鈥檙e looking for a weak signal in a lot of background
noise,鈥� he says. 鈥淪ynchrony could be a way of amplifying the weak signal.鈥�
Although Bialek, Rieke, Sejnowski and their colleagues are piling up
circumstantial evidence to support the idea that neurons encode information in
the timing of their spike trains, a larger question still hangs in the air. In
their fly, frog and rat experiments, they showed that the timing of electrical
impulses in the brain carried information that could be extracted. But that does
not necessarily mean that the animals鈥攐r more properly their
neurons鈥攈ave access to that information. Indeed, it is just possible that
the encoded information could be a curious artefact created by the architecture
of the cortex.
And there鈥檚 the rub. 鈥淭here鈥檚 a lot of tantalising evidence, but so far no
real proof, that neurons use temporally coded information,鈥� says neuroscientist
David Ferster at Northwestern University in Evanston, Illinois, who studies the
visual cortex. 鈥淚鈥檇 be happy if they did, but right now I鈥檓 still an
补驳苍辞蝉迟颈肠.鈥�
Changing direction
To show that an animal actually uses temporally coded information, someone
needs to show that particular spike-train patterns can change an animal鈥檚
behaviour. The fly and frog studies were heading in the right direction, but
they didn鈥檛 go far enough. Bialek imagines showing more movies to
flies鈥攖his time with the flies suspended from thin wires so that they can
fly, albeit in a restricted manner. If Bialek could then show that the fly
attempts to change direction each time it watches a particular movie that
generates a particular spike train鈥攐r even better, each time it is
artificially stimulated to produce the spike train 鈥攈e鈥檇 be on his way to
demonstrating that neurons use a temporal code to relay useful information in
the brain.
But such experiments are up against the boundaries of what is now technically
feasible, and working out whether or not the brain of a living animal uses a
temporal code could remain theoretical for some time. 鈥淚n the absence of data,
it boils down to one question,鈥� says Softky. 鈥淒o you believe that nature is
efficient and subtle or noisy and messy?鈥� Rieke is more matter of fact: 鈥淚 think
it would be a little weird if the animal went through all the trouble of coding
this information and then didn鈥檛 use it.鈥�
Of course, it may not be a simple case of either/or. Perhaps the Morse code
model operates in some parts of the brain, and the bucket model in others. Or
maybe both systems operate in the same neuron at different times. 鈥淚t seems to
me that information in the cortex is encoded in the average firing rate of
neurons,鈥� says Shadlen. But, he allows, 鈥渋t could be that individual spikes add
information on top of that鈥�.
Sejnowski, meanwhile, takes a predictably hardline stance: 鈥淭he question is
not whether the brain uses temporal information,鈥� he says, 鈥渂ut how.鈥�