Scientists working with rapidly advancing computer technology and electron microscopes hope one day to map the billions of neuronal connections in the brain. The resulting map, or “connectome,” could help us understand memory, intelligence and mental disorders, Dr. Sebastian Seung writes.
Suppose that someone gave you a radio and asked you to figure out how it works. You could try measuring electrical signals inside it, but the measurements might not be sufficient. You might be more successful if you were also given a circuit diagram illustrating all the components of the radio and how they are connected to each other.
Now imagine that your goal is to discover how a brain works. A map of brain connections would be helpful for interpreting measurements of the signals transmitted between neurons. In the human brain, these signals travel in a complex network of 100 billion or so neurons, each of which is connected to 10,000 others.
Such a map of a brain, human or otherwise, does not yet exist. But as technology advances, researchers are setting their sights on the “connectome,” a word coined in a 2005 study by Olaf Sporns and colleagues to describe a complete map of connections in a brain or a piece of a brain.
Genome and Connectome
“Connectome” was coined in analogy with the “genome”—the entirety of an organism’s hereditary information—studied by biologists. To imagine how the story of the connectome will unfold over the next few decades, it’s helpful to recall the history of the genome.
In 1953 James Watson and Francis Crick proposed the double helix structure for DNA. The double helix consists of a long chain of repeated units called nucleotides, of which there are four types: A, C, G, and T. Hereditary information is written in DNA using this alphabet of four letters. In the human genome, the sequence of nucleotides is about one billion letters long. The reading of this sequence was finally completed by the Human Genome Project in 2003.
The story of the connectome began when scientists first realized that the brain comprises a network of neurons. This happened around 1900, well before the double helix of Watson and Crick. But the connectome story is still in the future, and I believe the discoveries that compose this saga will be among the great prizes of 21st-century neuroscience.
Revealing connectomes will be much more difficult than identifying genomes. But I and others are now optimistic that the connectome will eventually be transformed from dream into reality. A new field of neuroscience will be created: “connectomics.” This new field will be driven by new technologies, as we will see. It will take shape alongside other research approaches, and these multiple methods will provide better insight into the brain’s complex structure than any individual method can.
Three-Dimensional Nanoscale Imaging
Connectomics is more challenging than genomics; the structure of the brain is extraordinarily complex. You have probably seen images of neurons before. A single neuron has a fantastic shape, forking out many branches to form a tree-like structure. But if you have only seen pictures of neurons in isolation, you may not fully appreciate the complexity of brain structure.
Before researchers study a single neuron under a microscope, they inject it with a stain. The neurons around it remain invisible because without the stain they are transparent. This technique is valuable for seeing the shape of a single neuron clearly. However, it does not give an accurate impression of what the brain is really like, because neurons are not islands in the brain. Instead, their forking branches are tightly entangled with each other. The brain can be compared to a giant bowl of spaghetti, in which each strand has been replaced by a complex, branched noodle.
Because their branches are so tightly entangled, neurons are locked in a multi-way embrace. At a point of contact between a pair of neurons, they can form a synapse, a junction at which one neuron sends chemical messages to another. When a synapse exists, the pair of neurons is said to be “connected.” The term should not be taken too literally, as there is still a narrow gap separating the two neurons, and the molecules in chemical messages have to float across this gap. The term is used in the metaphorical sense of communication, just as two people talking on cell phones are said to be connected. The efficacy of communication between a pair of connected neurons is known as the “strength” of the connection. If two neurons are strongly connected, the messages between them come in loud and clear, but if they are weakly connected, the messages are faint. Entanglement increases contact points between neurons, providing more potential locations for synapses, which allow neurons to communicate, or be “connected.”
Although entanglement is a crucial aspect of brain structure, it’s impossible to see with an ordinary light microscope. According to the laws of physics, structures smaller than the wavelength of light cannot be seen clearly using such a microscope.* The thinnest branches of neurons are less than a tenth of a micron in diameter,† which is less than the wavelength of visible light. Luckily, another kind of microscope uses electrons rather than light, and yields images with much higher spatial resolution. With an electron microscope, the branches of neurons can be seen clearly, even when they are tightly packed together in the brain.
By itself, a microscope cannot be used to see the interior of the brain, which is essential for observing brain structure. So, to see every location in the brain, scientists slice brain tissue into thin sections with a knife. By the combined use of the knife and the microscope, a sequence of two-dimensional images is acquired. Together these images show the entire three-dimensional brain structure.
The images from three-dimensional electron microscopy are generally visualized in the same way they were acquired, as a temporal sequence of two-dimensional images—in other words, a movie. This video contains images from the laboratory of Dr. Winfried Denk, a German scientist working at the Max Planck Institute in Heidelberg. The images show the structure of the retina of a rabbit. The retina, located at the back of the eye, is a sheet of neural tissue that converts light into neural signals.
Each frame of the video shows one slice through the retina. As the video plays, you are looking deeper and deeper into the specimen of neural tissue. The neural tissue consists of many branches of neurons, tightly packed together. Therefore a cross section through the neural tissue reveals the cross sections of numerous branches, each a small white region surrounded by a dark contour.
In addition to the gray-scale images, notice the colored objects in the video. They are branches of neurons—the “wires” of the brain—traced through the images by computer. You can also inspect the neural branches in the accompanying images, which are still images from the video.
In principle, humans can do this tracing manually. In practice, doing so would be painstaking and laborious, as the branches of just a single neuron could take many hours to trace. Furthermore, even a small sample of neural tissue comprises many branches. It would take an incredible amount of human effort to manually trace all of the branches in just a cubic millimeter of neural tissue. So we are turning to computers to automate the process of tracing and reduce the amount of human effort required.
This research to study neural wiring is in its beginning stages, so the algorithms are still not very good. The computer makes many mistakes. Sometimes it merges two neurons into one object. Sometimes it splits one neuron into two. As in the evolving field of computer vision, a branch of artificial intelligence in which robots try to see and recognize the objects around them but still cannot reliably do so, we do not yet have computers that can accurately see the shapes of neurons. Unless we can make such computers, we cannot hope to replace human effort with computer automation to trace the shapes of neurons.
Nevertheless, I am optimistic that researchers will eventually solve these technical problems. Then we will be ready to find connectomes. We will start with the simplest of brains, or even small parts of brains. As our technical abilities develop, we will scale up to larger brains. The process will start with electron microscopy, which will generate three-dimensional images of a specimen of brain tissue. Each synapse will be identified, and the branches involved will be traced back to their parent neurons. By assigning each synapse to a pair of neurons, we will be able to map the connections in the brain. The connectome will begin to take shape.
Today a small community of scientists and engineers is working on developing these technologies. If they are successful, the new field of connectomics will change our understanding of the brain in myriad ways.
The Connection Theory of Memory
First, connectomics will help reveal how the brain stores and retrieves information about the past. Neuroscientists believe that memories are stored in the connections between neurons. According to this theory, connections change when a new memory is stored. That such changes can happen is not in doubt. Neuroscientists have found that new synapses can be created and that the strengths of existing synapses can be altered. What remains uncertain is whether these changes are indeed the basis of memory.
Although the connection theory of memory is widely believed, it has been difficult to test experimentally. One barrier has been the lack of good techniques for measuring whether two neurons are connected, and if so, how strongly. One important task of connectomics will be to determine the connectivity of brain areas that are involved in memory storage.
Take, for example, sequential memory, such as the notes of a piece played on the piano. A pianist is able to store such sequences in memory and recall them at will. Recently, sequential memory has been studied by neuroscientists in the brain of the zebra finch. This bird learns a single, highly repetitive song as a juvenile and sings it repeatedly as an adult. The avian brain area called HVC appears to be important for a bird’s memory of its song. Creating a lesion in the HVC in adult birds causes a loss of song. Furthermore, electrical recordings of “projection” neurons in the HVC, which send long branches (axons) to downstream brain areas, have revealed a precisely timed, repeated sequence of neural activations while the bird sings this song. The zebra finch song consists of repetitions of a single motif, which is about one second long. During a song motif, each projection neuron in the HVC is activated exactly once, for just a few milliseconds. For every repetition of a motif, the projection neurons are activated in the same sequence. The projection neurons of the HVC activate other neurons in an area called RA, which in turn activate the motor neurons that control the syrinx, the avian vocal organ. Therefore the sequential activation of HVC neurons is believed to be directly responsible for the sequence of movements that produces birdsong.
According to one theory, the memory of the sequence is stored in the connectivity of the HVC. Each projection neuron transmits an excitatory signal through its synapses onto the neurons that are just after it in the sequence. These synaptic connections cause the neurons to be activated in sequence, like a row of falling dominoes.
Currently, it is not feasible to make the measurements that would be required to test this theory about HVC connectivity. But connectomics will eventually make it possible to find out whether the memory of birdsong is indeed stored in a sequential organization of connectivity in HVC. I expect that it will also become possible to study the connectional basis of other types of memory as well.
The Connection Theory of Intelligence
Connectomics also holds the potential for addressing the biological bases of mental—cognitive and emotional—differences. Just as physical characteristics differ, so do mental characteristics. For instance, some people get angry easily, while others seem to show little emotion. The few who have prodigious capabilities, such as Mozart and Einstein, are called geniuses. What causes these mental differences?
In the 19th century, the English scientist Francis Galton proposed that smart people have larger brains. He tested his proposal by comparing the head sizes of students at Cambridge University with their grades. Modern researchers continue in the tradition of Galton but use the more sophisticated method of magnetic resonance imaging to study brain structure. They have confirmed a correlation between overall brain volume and intelligence as measured by IQ tests. However, the correlation is fairly weak.
As brain imaging methods have become more advanced, researchers have been making more precise measurements of brain structure. They are looking for some structural feature that is more strongly correlated with intelligence than is overall brain volume. Some have looked at the sizes of particular brain areas. Others have examined the brain’s white matter, the neural fibers that connect brain areas. Still others have looked at the changes in the thickness of the brain’s cortex during development.
With the advent of connectomics, it will be possible to investigate a related but somewhat different hypothesis. Maybe intelligence depends not on the size of a brain but rather on how its connections are organized. Because most of the volume of the brain is devoted to the neurons and their synaptic connections, differences in connectivity could be crudely manifested as differences in brain volume. However, there could be many differences in connectivity that have no effect on brain volume and that are difficult or impossible to measure by current techniques.
It has been popular to think of the brain as an assembly of interacting modules, each of which is functionally specialized and localized in a distinct brain area. One could imagine that mental differences arise from differences in connectivity within modules, between modules or both. The connections between modules are like the superhighways of the brain. They are relatively wide and long compared to connections within modules, which are like local streets and alleys. Connectomics will provide improved methods of studying both types of connections. With these improved methods, I think we will be able to find out whether mental differences are related to differences in connectivity.
Even if mental differences are caused by brain differences, that does not mean they are exclusively or even primarily determined by genetics. Some are the results of life experiences rather than heredity. For example, although identical twins have identical genes, one may play the piano and the other may play the flute. Much as muscles can be changed by weight lifting, brains can be changed by practice.
In addition to studying differences in normal mental functioning, connectomics will address the causes of mental disorders. In some cases, these disorders are characterized by the death of neurons—neurodegenerative disorders such as Alzheimer’s disease, for instance.
But for other mental disorders, neuron loss appears far less significant. Rather, many scientists suspect that at least some mental disorders involve abnormalities in the connections of the brain, or “connectopathies.” Since the 19th century, when this hypothesis was first proposed, its popularity has waxed and waned more than once. Today it is popular again, as scientists use magnetic resonance imaging to look for abnormalities in neural networks—connectopathies—that may be associated with disorders like schizophrenia and autism. Investigators speculate that such disorders are developmental, perhaps caused by abnormalities in the processes by which the brain wires itself. Imaging research is complemented by progress in finding genes that are associated with mental disorders. Some of these genes could be involved in controlling brain wiring processes.
Although magnetic resonance imaging has the advantage that it can view live brains, its disadvantage is low spatial resolution. Connectomics will make it possible to map brain wiring at a much higher resolution, thus providing more refined methods of looking for connectopathies.
These proposed applications of connectomics to the human brain, with its 100 billion neurons and their connections, are very speculative. Finding the human connectome is not realistic now. Connectomics must begin instead with simple brains.
For example, the worm C. elegans is about one millimeter long. While it doesn’t have a brain like ours, it does have a nervous system containing 300 neurons and 7,000 synapses. Every connection of this nervous system was mapped in the 1970s and 1980s by a team of scientists. Because the analysis was done manually, without the aid of a computer, it took more than 10 years to complete.
With the technologies currently under development, the labor involved in slicing, imaging and analyzing the images will be reduced through automation. That will make it possible to find the connectomes of brains more complex than the C. elegans nervous system. At first these brains will be very small—the 100,000 or so neurons in the fruit fly brain, for example. Finding even this small bug’s connectome is an incredible challenge.
We will also try to determine the connectomes of small parts of large brains. A prime target is the retina, the sheet of neural tissue at the back of the eye that converts visual stimuli into neural signals. Although the retina is in the eye, it contains networks of neurons much like those in the brain and is actually considered part of the brain because of its common embryological origin.
If we are successful in meeting these challenges, then we will move on to the brains of larger animals, such as the mouse or bird. How far will technology progress? Will I live long enough to see the human connectome? Finding the human connectome will be one of the greatest computational challenges of all time, because of the difficulty of analyzing such a large amount of image data. A one-terabyte hard disk costs a few hundred dollars today; you would need 100 million terabytes to store the images from a human brain if they were taken at 20-nanometer resolution and not compressed by software. (There are 1 million nanometers in a millimeter.) Even with 10,000 microscopes working in parallel, it might take 30 years to collect all of the images. Analyzing the images could require a parallel supercomputer with millions of processors.
Of course it’s difficult to predict the future, but the rate of technological progress in other areas gives reason for optimism. Computer technology is still improving exponentially over time. The price of computation is halved every two years, and the price of storage is halved every year. Astonishingly, the pace of improvement in genomic technology is even faster than that. The price of finding a human genome has dropped by a factor of 10 every year for the past four years.
If computer technology were to improve at an exponential pace for several more decades, then we could be confident about eventually finding the human connectome. Some are predicting that the exponential party will come to an end in just one decade, but others believe that advances in nanoelectronics will keep it going for longer.
Therefore, humanity’s quest to understand the brain will go hand in hand with the quest to build ever more powerful “artificial brains”—computers. I am confident in the future of both endeavors. I believe that by the time the 21st century is recorded in the history books, one of its greatest triumphs will be the determination of the intricate and awesome structure that gives rise to the mind: the human connectome.