It鈥檚 the stuff of dreams, at least as far as neuroscientists are concerned: a
living model of a real brain that they can grow in any lab. With careful
nurturing and development, such a model could reveal how the brain learns, how
it remembers and how it communicates. And that鈥檚 not all. Give engineers such
models and they could design all kinds of new gadgets鈥攕ensors, control
systems, maybe even living computers.
Far-fetched? Maybe not. The building blocks of the brain are nerve cells, or
neurons. 杏吧原创s have been able to grow them in the lab for years. What they
haven鈥檛 been able to do is control this growth. This is changing and the
implications are mind-blowing.
Researchers are working feverishly to learn how to grow small networks of
neurons and how to record the signals they send back and forth to each other. If
they can reproduce the same circuits over and over again, they will be able to
repeat experiments and eventually develop devices that rely on the special
properties of nerve cells. It鈥檚 a crucial goal, a huge step on the way to
understanding the brain鈥攁nd late last year, they achieved it for the first
time.
Advertisement
Such circuits should provide unprecedented insights into learning and memory.
They could also have all sorts of spin-offs. For example, circuits of living
nerve cells could act as sensitive chemical sensors or as the interface between
electronic devices and the human nervous system. Some researchers even think
that the day may not be far away when scientists grow living computing devices
out of brain cells.
Growing such circuits is alot harder than it seems. The process starts with
neurons. A single neuron is a complicated sliver of biology. Like other cells,
it has a nucleus, a kind of control centre, contained inside the cell body. But
unlike other cells, a neuron grows an elongation called an axon.
An axon is the biological equivalent of a telephone line鈥攄own its
length pass the electrical signals neurons transmit when they are activated.
These signals are called action potentials and they鈥檙e an all-or-nothing
proposition鈥攖hey are either on or off. Studded along the length of the
atom are connector ports called synapses which act like transmitters, sending
out chemical signals to other neurons. The cell body also puts out wiry
branching tendrils called dendrites which are festooned with synapses. These are
receivers that pick up chemical signals from the axons of other nerve cells.
Researchers have been growing nerve cells for decades. Early on they
discovered that it is all but impossible to culture adult neurons. Instead,
neurons are usually taken from embryos before they have sprouted axons and
dendrites, and grown in culture. Neurons, however, won鈥檛 grow all by themselves.
They depend on other cells called glia, which are the brain鈥檚 version of
Hollywood personal assistants鈥攖hey do whatever neurons need to get them
through the day. Those services include providing physical support during
development and forming the fatty sheath of myelin that helps impulses whizz
along the axon. To keep cultures alive for more than a month or so, researchers
have to provide glial cells. Over shorter periods, they can use a chemical bath.
Growing nerve cells is only the first stage in building reproducible circuits.
Researchers also need to be able to record the activity of the cells. The
standard practice for a single neuron is to use an intracellular electrode, a
glass pipette filled with a conducting solution that is inserted into the cell
body. It is a highly sensitive technique, but it disrupts the cell membrane and
kills the cell within a few hours. 鈥淭o map the culture鈥檚 connections, you have
to be able to stimulate one and record from the rest in a way that does not hurt
them,鈥� says Jerry Pine, a biophysicist at the California Institute of Technology
in Pasadena.
Pine is a pioneer of a different technique. He was among the first
researchers to measure the electrical activity of cells by growing them on top
of an array of electrodes. The array is placed on the bottom of a Petri dish,
and the neuron culture is grown on top of it.
Listening to the neurons chatter is far from straightforward. Neurons
communicate with one another chemically, squirting small molecules called
neurotransmitters across their synaptic gap. This signal can trigger an adjacent
cell, which triggers another and so on. The signal will propagate depending on
how the network is wired up and the electrode arrays are designed to listen in
to this electronic chatter. Even small networks of neurons generate lots of
information. It鈥檚 a bit like trying to understand what a roomful of people are
saying to each other at a cocktail party using an array of microphones suspended
from the ceiling鈥攊t鈥檚 hard to tell who鈥檚 talking to whom. 鈥淪ay you have
two neurons over the top of an electrode. The electrical signal from them will
be [combined], and you鈥檙e going to need a computer to sort them out,鈥� says Bruce
Wheeler, an electrical engineer at the University of Illinois at Urbana-Champaign.
The goal is to grow these patterns in such a way that one cell sits on top of
one electrode, rather like making everyone at the cocktail party take an
assigned seat. Pine throws such a party on a silicon chip that he calls the
Neurochip. The set of chairs are 16 tiny holes in the silicon, each with an
electrode at the bottom, arranged in a four by four array. Hippocampal neurons
from a rat embryo are forced to take their seats by a researcher, who loads one
into each well. 鈥淵ou get very good recording off each neuron,鈥� says Pine.
The problem is that neurons are not very good at sitting still. As they
develop, they move, sending out threads called processes to guide the way. 鈥淭hey
migrate a rather long way,鈥� says Pine. When that happens, the contact between
the electrode and the cell is lost. 鈥淲e built several generations of wells with
smaller and smaller holes, until finally we found the neurons would go out
through a hole one by three microns, which is tiny, and they would die from the
effort.鈥� Eventually, researchers in Pine鈥檚 lab figured out that the processes
won鈥檛 grow around sharp angles. So the latest Neurochip design鈥漜ages鈥� them in
with overhangs and corners.
But while this helps to anchor the cells over the electrode array, it does
not control the pattern of connections that form. 鈥淲hat you鈥檇 like to do is have
some control over who connects to whom,鈥� says David Kleinfeld, a biophysicist at
the University of California, San Diego. 鈥淵ou need to do some chemistry to
control where the neurons fit and how they grow.鈥�
This is where the idea of reproducible circuits becomes important. One way to
determine the pattern is to print out a chemical template that the cells can
follow as they grow. In a technique akin to making potato prints, Wheeler uses a
micrometre-scale polymer stamp to apply a chemical called polylysine to a
background of polyethylene glycol. When neurons are grown on this surface, they
stick to the polysyline, but not to the polyethylene. The biggest problem with
this technique is that the neurons tend to jump the rails after a week or so and
begin to ignore the template. When this happens, the network is doomed. But by
refining the technique, Wheeler has been able to grow networks that remain
stable for up to four weeks. 鈥淏asically, these things stay put,鈥� he says.
Of course, the overall shape of the network is only one part of the problem.
Because signals move through synapses in only one direction鈥攆rom the axon
of one neuron to a dendrite of another鈥攔esearchers also need to control
the direction in which the axons and dendrites grow. Wheeler has made promising
progress in this area. By varying the chemicals in his microstamped pattern, he
can control which parts of a cell stick to which parts of the pattern. Axons,
for example, stick to a combination of laminin and polylysine, while dendrites
seemed to like polylysine alone. That at least is the first step in determining
the direction of growth.
At George Washington University in Washington DC, James Hickman has gone even
further. Hickman creates a pattern of hydrophobic (water-repellent) and
hydrophilic (water-loving) chemicals on a sliver of silicon. The nerve cells
grow in the shape of the pattern created by the hydrophilic chemicals. Hickman
used layers of a hydrophilic molecule called DETA, which he cut into a pattern
using a laser (see Diagram).
The layout ensures that the axons grow in one
direction round the circuit while the dendrites grow in the other, so that the
axon of one neuron grows towards the dendrites of another and vice versa. Just
how the layout influences the growth isn鈥檛 clear but the result is a simple
brain circuit consisting of two neurons which can trigger each other.
Late last year, he grew his first circuit using this technique. Since the
shape and direction of growth are controlled, his circuit is entirely
reproducible and this is the real significance of the work. And although Hickman
was only able to measure how one cell triggered another using the destructive
glass pipette method, he now plans to grow the pattern on top of a
microelectrode array.
Mind-blowing
With all the techniques in place, a new generation of devices based on living
circuits could be just around the corner. The possible applications are so
staggering that researchers all over the world are eagerly getting in on the
act. Japan, Europe and the US are all funding programmes. One of the most
promising applications is in basic neuroscience. 鈥淟earning and memory are
basically the result of changes in the connectivity of the brain network,鈥� says
Pine, 鈥渟o the basic rules for why and how the connections change is one of the
fundamental questions.鈥�
A neuron-chip interface could also improve prosthetics. Many motor prostheses
are controlled by body movement. For example, the position of the shoulder
relative to the sternum can be used to control an artificial hand. 鈥淏ut there is
a real need in the rehabilitation field for a richer, more versatile source of
commands to control these prostheses,鈥� says Bill Heetderks, deputy head of the
Neural Prosthesis Program at the National Institutes of Health.
The ideal source of those commands would be the nerve that controlled a limb
before it was lost. While a chip that physically connects to nerves is still a
long way off, the techniques that are being developed to study the output from
nerves could lead to breakthroughs in this field.
Another idea is that neurons could form the basis for a portable sensor.
Nerve cells have, after all, evolved to fire when they detect tiny quantities of
a chemical. For example, many chemical weapons are neurotoxins, so a biosensor
could pick up the first hint of one in the air. 鈥淪ome cells can pick up
concentrations at the one-molecule level because of protein binding,鈥� says
Randy Garrett, director of information systems at Rockwell Collins Advanced
Technology Center in Cedar Rapids, Iowa. But there鈥檚 a catch, he says. 鈥淵ou鈥檝e
got one cell that says, `Aha, I found anthrax toxin.鈥� The problem is, how are
you going to tell me?鈥� Some researchers have grafted genes for phosphorescence
into neurons, so that a cell lights up when it detects a given chemical.
It may even be possible to build computing devices out of combinations of
neurons. Nerve cells are, after all, cheap to make, small and extremely low
power鈥攖hey basically run off glucose. And Hickman points out that they
come in inhibitory or excitatory flavours that make the neurons they鈥檙e
connected to less or more likely to fire. Simple combinations of inhibitory and
excitatory neurons grown in the correct patterns could function like logic
gates. These living logic gates could be combined with others to form higher
units, much as transistors are combined to create microchips. But 鈥渨hat you can
get a couple of neurons to do is much more complicated than a logic gate鈥�, says
Hickman. He points out that simple organisms can carry out relatively complex
functions. The nematode worm Caenorhabditis elegans has only 302 neurons
but can move, react to stimuli and even carry out simple pattern recognition to
find food. In comparison, a silicon chip with a similar number of transistors is
almost useless.
Not everyone is convinced: 鈥淭he people who are really crazy are people who
think this is computationally interesting,鈥� Pine says. Neurons transmit messages
to each other on timescales of the order of milliseconds; in a massively
parallel computer, processors send messages in nanoseconds鈥攆aster by
orders of magnitude. 鈥淪ilicon goes really fast; biology, conversely, goes really
slow,鈥� says Hickman. 鈥淲hy does a supercomputer have to go fast? Because its
complexity is really low.鈥� A single neuron does a huge amount of complicated
processing in picoseconds, the theory goes, and then sends that processed
information to the next cell in the network. No one knows how neurons do it, so
even if brain circuits aren鈥檛 useful as computing devices, they should shed some
light on this problem.
All these applications are a long way off, but the first steps have been
taken.鈥漈he very first thing we鈥檙e going to have to be able to do is make the
same thing over and over, even if it鈥檚 the most fund mental device,鈥� he says.
鈥淲e鈥檝e been able to determine the polarity of neurons,鈥� says Hickman. 鈥淭hat we
can do now by surface chemistry. We can get signals in and out. We can make
networks.鈥� The question is, what will they be able to do next?