This image could be hung in a gallery, but it started life as a tiny chunk of a woman’s brain. In 2014, a woman undergoing surgery for epilepsy had a tiny chunk of her cerebral cortex removed. This cubic millimeter of tissue has allowed Harvard and Google researchers to produce the most detailed wiring diagram of the human brain that the world has ever seen.
Biologists and machine-learning experts spent 10 years building an interactive map of the brain tissue, which contains approximately 57,000 cells and 150 million synapses. It shows cells that wrap around themselves, pairs of cells that seem mirrored, and egg-shaped “objects” that, according to the research, defy categorization. This mind-blowingly complex diagram is expected to help drive forward scientific research, from understanding human neural circuits to potential treatments for disorders.
“If we map things at a very high resolution, see all the connections between different neurons, and analyze that at a large scale, we may be able to identify rules of wiring,” says Daniel Berger, one of the project’s lead researchers and a specialist in connectomics, which is the science of how individual neurons link to form functional networks. “From this, we may be able to make models that mechanistically explain how thinking works or memory is stored.”
Jeff Lichtman, a professor in molecular and cellular biology at Harvard, explains that researchers in his lab, led by Alex Shapson-Coe, created the brain map by taking subcellular pictures of the tissue using electron microscopy. The tissue from the 45-year-old woman’s brain was stained with heavy metals, which bind to lipid membranes in cells. This was done so that cells would be visible when viewed through an electron microscope, as heavy metals reflect electrons.
The tissue was then embedded in resin so that it could be cut into really thin slices, just 34 nanometers thick (in comparison, the thickness of a typical piece of paper is around 100,000 nanometers). This was done to make the mapping easier, says Berger—to transform a 3D problem into a 2D problem. After this, the team took electron microscope images of each 2D slice, which amounted to a mammoth 1.4 petabytes of data.
Once the Harvard researchers had these images, they did what many of us do when faced with a problem: They turned to Google. A team at the tech giant led by Viren Jain aligned the 2D images using machine-learning algorithms to produce 3D reconstructions with automatic segmentation, which is where components within an image—for example, different cell types—are automatically differentiated and categorized. Some of the segmentation required what Lichtman called “ground-truth data,” which involved Berger (who worked closely with Google’s team) manually redrawing some of the tissue by hand to further inform the algorithms.
Digital technology, Berger explains, enabled him to see all the cells in this tissue sample and color them differently depending on their size. Traditional methods of imaging neurons, such as coloring samples with a chemical known as the Golgi stain, which has been used for over a century, leave some elements of nervous tissue hidden.
In the example above, Berger made the smallest cells blue and the biggest cells red, with all other cells between falling on a color spectrum. This helped researchers to identify the brain’s six cortical layers and white matter.
While researchers have been able to identify structures from the data, one ongoing difficulty of the project is proofreading the automatic segmentation. This involves individuals manually sifting through every part of the 3D map to check for segmentation errors. “This is a huge challenge for human beings, because now we’re generating datasets that are larger than a single human can experience,” says Lichtman.
In parts of the data that have been proofread, Berger says that particular cells seem “really interested in contacting.” The researchers have found examples of over 50 synapses to one singular neuron, which, according to Berger, is a phenomenon previously overlooked that could be integral to cortical processing.
On top of identifying structures and connections, researchers have identified abnormal cells. Berger said he came across an unidentifiable egg-shaped “object” (much smaller than a cell body but part of a cell) when attempting to systematically categorize each cell in the dataset. Other ambiguous cells include those seemingly mirrored in shape and “tangled” cells that wrap around themselves; until further research is done, these cells remain mysteries. However, they may not remain so for long.
The brain map has been made open access, which means that these images have opened up boundless possibilities for progress in neuroscience, particularly as this is the first publicly available wiring diagram of the human brain at subcellular level. Both Berger and Lichtman emphasized that they did not go into the project with concrete aims of discovery but rather wanted to create the “possibility to observe,” and from this, they hope (and expect) that “further insights will come” from both the Lichtman lab and external researchers.
Berger anticipates that advancements could be made in understanding and treating mental conditions such as schizophrenia. Potential future discoveries could also expand beyond the mind, as Berger thinks the functions of the biological brain may be used to improve deep-learning AI systems and their structures.
In terms of future projects, the Harvard Lichtman lab plans to continue its collaboration with Google to “factor this rendering up another scale of a thousand” by studying a whole mouse brain. The research lab is also working on more human brain samples, to expand research into other regions of the brain. This will enhance the already invaluable resource and its ability to inform and expand future discoveries.
This article appears in the September/October 2024 issue of WIRED UK magazine.
