As you read these words, your eyes scan the page, picking up patterns to which your mind assigns meaning. Meanwhile your heart contracts and relaxes, your diaphragm rises and drops to control your breathing, your back muscles tense to maintain your posture, and a thousand other basic tasks of conscious and subconscious life proceed, all under the coordinated control of roughly 86 billion neurons and an equal number of supporting cells inside your skull. To neuroscientists like us, even the simple act of reading a magazine is a wondrous feat—as well as an example of perhaps the hardest problem in science today: in truth, we cannot yet fully explain how the human brain thinks and why the brain of a monkey cannot reason as we do.

Neuroscientists have intensely studied the human brain for more than a century, yet we sometimes still feel like explorers who have landed on the shores of a newly discovered continent. The first to arrive plotted the overall boundaries and contours. In the early 1900s German scientist Korbinian Brodmann sliced up human brains and placed them under his microscope to examine the cerebral cortex—the exterior layers of gray matter that handle most perception, thought and memory. He parceled this cortex into several dozen regions based on the topology of the organ and how the cells in each area appear when labeled with various stains.

A view gradually took hold that each region, each cluster of cells of a particular kind, handles a specific set of functions. Some neuroscientists challenged this theory that function is parceled by location. But the parcellation model has returned to vogue with the emergence of new tools, most prominently functional magnetic resonance imaging (MRI), which records what parts of the brain “light up” (consume oxygen) as people read, dream or even tell lies. Researchers have been exploiting this technology to construct “maps” that relate what they see using these tools to real-world human behavior.

A newer school of thought, however, postulates that the brain is more like an informal social network than one having a rigid division of labor. In this view, the connections that a neuron has made with other brain cells determine its behavior more than its position does, and the behavior of any given region is influenced strongly by its past experience and current situation. If this idea is correct, we can expect to see overlapping activity among the particular locations that handle the brain's responsibilities. Testing this hypothesis will be tricky; brain circuits are hard to trace, and the billions of neurons in a human brain connect at perhaps 100 trillion links, or synapses. But projects are under way to develop the new tools needed for the job [see box on opposite page].

In 2003, as the Human Genome Project published the sequence of code letters in human DNA, we and our colleagues at the Allen Institute for Brain Science in Seattle saw an opportunity to use the new catalogue of 20,000 or so human genes and rapidly improving gene-scanning systems to look at the human brain from a new perspective—one that might inform this debate. We realized that by combining the tools of genetics with those of classic neuroscience, we could plunge deep into the jungle of the uncharted continent: we could actually map which parts of the genome are active, and which are dormant, throughout the entire volume of the brain. We expected that this map would show a very different set of genes turned on in, say, the part of the brain that handles hearing from those parts that control touch, movement or reasoning.

Our goal, which ultimately took nearly a decade to achieve, was to produce three-dimensional atlases plotting where individual genes operate in the brains of healthy humans and, for comparison, mice. (We are now working to add monkeys as well.) Such molecular maps provide invaluable benchmarks for what is “normal”—or at least typical—in much the same way that the reference DNA sequence produced by the Human Genome Project does. We expect these atlases to accelerate progress in neuroscience and drug discovery while allowing investigators to explore their fundamental curiosity about the structure of the human mind.

Already these new views of the inner working of human and rodent brains have produced some surprises. One big one: although every person is unique, the patterns of gene activity are remarkably similar from one human brain to the next.Despite our differences, people share a common genetic geography in their brains. Moreover, within each individual, we unexpectedly found no major differences in gene actions between the left brain and right brain. And although mice are used as proxies for humans in most neuroscience research and early drug trials, it is clear from these new results that, at a genetic level, humans are not simply large mice. This discovery calls into question the use of mice as models for understanding the neurobiology of our own species.

From Mouse to Human
No one had ever made a complete genetic map of a mammal's brain before. To work out the many details, we started small, with a mouse brain. Mice have about as many genes as humans do, but their brains are a mere 3,000th the mass of ours.

Within three years we had processed more than a million slices of mouse brain, dousing each one with visible markers that stick wherever a particular gene is expressed—meaning that the gene is used; it is copied from the DNA into a short piece of RNA called a transcript. RNA transcripts are intermediate steps en route to the final product encoded by a gene, which is usually a protein that does work in the cell, such as carrying out an enzymatic reaction or serving as some piece of cellular machinery. Some RNA transcripts do useful work directly without ever being translated into protein form, and we were able to look for about a thousand kinds of such noncoding RNAs in addition to all the protein-coding genes.

Beyond honing our techniques, the mouse project handed us one of our first surprises. Of course, as in a human, almost every cell in a mouse contains a complete set of chromosomes and thus at least one copy of every gene in the animal's genome. In mature cells, a sizable fraction of these genes are silent at any given moment—no RNA is being made from them. Yet when we completed the mouse atlas in 2006, we saw that many genes—more than four out of every five—possessed by mice were functioning somewhere in the animals' brains at the time they died. (Neurobiologists know that, for the most part, patterns of genetic activity shift during life on a timescale of hours and persist for many hours after death.) As we began to make plans to create a human brain atlas, we wondered whether human brains would show a similarly high level of genetic activity—and more important, whether the specific patterns of activity would closely resemble those that we observed in mice.

We received our first human brain in the summer of 2009, from a 24-year-old African-American man whose brain had been donated by his family, scanned by MRI to make a virtual 3-D model of the intact organ, and then frozen solid, all within 23 hours of his accidental death—fast enough to lock in the normal RNA patterns. Aside from asthma, he had been healthy.

To deal with the 3,000-fold increase in size from the mouse brain project, we switched to a different method to measure gene expression. The frozen brain was cut into thin slices, which were stained and photographed in high detail. Anatomists then used lasers to snip microscopic samples from about 900 structures that we had preselected at positions throughout the brain. Molecular biologists tested each sample using a DNA microarray, a mass-produced gadget that simultaneously measures the amount of RNA present from every individual protein-coding gene in the human genome.

After we had collected data in this way from the first brain, we put all the results together into a computer database. We could select any gene and see how much of its corresponding RNA was present in each of the 900 sampled structures and thus how actively that gene was being expressed in the hours before the donor died. As we chose one gene after another, it was a thrill to see very different patterns emerge. Now the real exploration could begin.

Shades of Gray Matter
Early on, as we analyzed the data on the first brain thoroughly, we saw unexpectedly that gene expression patterns in the left hemisphere were mirrored, almost exactly, in the right hemisphere. The idea that the left side of the brain is specialized for certain functions, such as math and language, and that the right side contributes more to artistic and creative thought may be well established in popular culture, but we saw no evidence of such differences in the genetic patterns within this brain. We confirmed that finding with the second brain we examined. The results were so conclusive that we have studied just one hemisphere on each of the four brains we have processed since; this discovery accelerated the construction of the atlas by a year or more.

As we had seen in mice, the great majority of genes—84 percent of the different kinds of RNA transcripts we looked for—were active somewhere within the six human brains. The organ performs an uncommonly wide span of jobs, and the atlas revealed that distinct collections of genes are at work in each major region, contributing to its particular functions.

The donors of the brains we studied included both men and women, young and old, black, white and Hispanic. Some of them had big brains; others were smaller. Despite these differences, all six brains had highly consistent patterns of gene activity. More than 97 percent of the time, when we saw lots of RNA being made from a gene in a section of one brain, the same was happening in a majority of the others.

We began examining the sets of genes active in various parts of the brain. For example, we compared the genes being used most heavily in the ancient midbrain, which humans share with reptiles, against those highly active in the cerebral cortex. Neurologists have long known that cells in the more primitive parts of the brain—structures such as the hypothalamus, the hippocampus and the pons (responsible for managing body temperature, hunger, spatial memory and sleep)—cluster into distinct nuclei that behave rather differently from one another. We found that many of these nuclei express distinct sets of genes. Within these primal structures is a cacophony of genetic voices clamoring at once.

The cortex, on the other hand, is different in both its cellular structure and its genetic activity. The cortex consists of a wide variety of cell types arranged into a sheet with six layers of gray matter. It evolved relatively recently and expanded to become proportionally much more prominent in humans than in other animals; the gray matter is what gives rise to the unique complexity of human behavior and individual personality. We naturally wondered: In this most human part of the brain, does the complexity of function arise from huge differences among the genes being expressed in one part of the cortex versus another? Brodmann divided the cortex into dozens of well-defined parcels, after all, and we expected that the different roles each parcel plays in human behavior arise from correspondingly different suites of genes being put to use.

But the atlas suggests that the answer is no: gene activity in the cortex for any given cell type is remarkably homogeneous within the gray matter, all the way from the forehead to the back of the skull.

We did find that each cortical cell type has a distinct genetic signature. But remarkably few sharp boundaries show up in the genetic geography—with the notable exception of the visual cortex at the back of the brain, which processes input from the eyes. The cerebellum, which sits at the base of the brain and is another structure that expanded in humans recently, is similarly a sea of homogeneity.

These results are hard to reconcile with the Brodmann-inspired idea that the cortex divides neatly into parcels devoted to particular functions whose behavior is governed by the genes at work inside them. The atlas instead supports an alternative theory: genes define each of the various cell types, as well as the basic blueprint for a small column of cortex that arranges cells of those different kinds in a predefined way from the surface of the brain to the bottom of the cortex. But the cortex as a whole consists of many copies of that canonical column. How the cortex behaves overall appears to depend much more on the specific ways that neurons are wired into circuits—and the history of stimuli hitting those circuits—than it does on shifts in genetic activity from one Brodmann region to another.

More Like Monkeys
When we compared roughly 1,000 genes active in the cortex of both mouse and human, we were amazed to find that nearly a third of them are being expressed quite differently. Some genes are silent in one species but not the other, for instance, whereas many others are used at much different rates.

The degree of similarity between mouse and human matters because almost all neurological experiments and drug trials are performed first on mice. Rodents are cheap, grow quickly, and are easy to control and examine. Yet therapies that succeed in mice rarely translate directly to effective treatments in people. The variance in gene expression between the two species could help explain why that is.

In striking contrast, the data we have analyzed so far on rhesus macaque monkeys suggests that fewer than 5 percent of genes are expressed in their brains in a significantly different way than in ours. Our consortium's work on a brain atlas for monkeys is still under way, so that number may change as we gather more data. Nevertheless, the observation that genetic activity in human and monkey brains is so fundamentally similar again points to the wiring among the neurons of our brains, rather than the genetic activity within the cells, as the likely source of our distinctiveness as a species. Moreover, it is clear that we need to put more detailed information about the human brain in the hands of researchers and pharmaceutical companies to help them distinguish those drug targets that can be modeled in mice from those that should be studied in animals more closely related to humans.

Since we released the brain atlas for the mouse in 2007, it has been used in more than 1,000 scientific studies. For the human brain atlas, which was opened to public view with the first two brains in 2010, the next logical steps are to improve the resolution and scope of the map. We have learned that we will not ultimately understand the role that gene activity plays in brain function until we measure gene expression patterns in individual brain cells. Doing that is truly a grand challenge for an organ as large and complex as the human brain. But new technologies are emerging that allow neurogeneticists to measure protein-coding RNA from single cells. These tools also enable detection of all transcribed pieces of RNA, which could clarify whether RNAs that do not give rise to proteins—the so-called dark matter of the genome—play important roles in the brain.

To make it easy for scientists who are researching disorders of the brain, such as autism, Alzheimer's disease and Parkinson's disease, to use the atlas, the Allen Institute has made all our data—as well as a point-and-click viewer called Brain Explorer—freely available online. We hope these early attempts to understand human brain function through its genetic map will pave the way for others to build on it in unforeseen ways.