2016: Stitching Thought Together

Stitching Thought Together

PSC Powers Harvard’s, Allen Institute’s 3D-Reconstruction of Excitatory Visual Neuron Wiring

Why It’s Important: One of the mysteries of brain function is how we make sense of the jumble of images that confront our eyes.

Neuroscientists have discovered that most individual nerve cells in the brain respond to specific elements in the visual environment. For example, a nerve cell may fire in response to vertical lines—another, in response to horizontal or slanted lines. Researchers suspected that the mammalian cortex amplifies this signal by having nerve cells that respond to similar elements excite each other. This mutual excitation may help those elements stand out and prime the network for their further processing. But scientists had no anatomical evidence that this actually happens. 

How PSC Helped: A team led by Wei-Chung Allen Lee of Harvard University and R. Clay Reid of the Allen Institute for Brain Science in Seattle identified brain nerve cells that respond to visual elements in living mice. Then they took a series of millions of microscope images of ultrathin (about 40-nanometer-thick) tissue slices around these nerve cells. In a collaboration facilitated by PSC’s Art Wetzel, PSC computational scientist Greg Hood helped them to reconnect these images into a three-dimensional volume using PSC’s AlignTK software. But because these slices are fragile, microscopic tears and other artifacts happened, requiring manual intervention to correct. So researchers had to move back and forth between computation and manual “repair” of the images until the quality of the aligned volume was good enough to trace the connections between the nerve cells. The team reported in the journal Nature in March 2016 that mammalian excitatory nerve cells that respond to a given visual feature do indeed make more and larger connections to other excitatory nerve cells that are tuned to respond to similar visual elements.

wholefigure highres


Stylized drawing showing how the excitatory visual neurons connect with each other. Arrows represent excitatory connections, their thickness showing how many other neurons each connects with.

Colors code for what orientation of visual lines each neuron responds to; circles are target neurons, triangles large-caliber neuron extensions that exit the image, and diamonds are other neurons. Adapted by permission from Macmillan Publishers Ltd: Lee et al. (2016) Anatomy and function of an excitatory network in the visual cortex. Nature 532(7599):370-374.




In their Nature paper, the Harvard and Allen Institute group and coauthor Hood have produced a reconstruction of an excitatory nerve-cell network in the brain’s cortex at a subcellular level. The work was a tour de force of two-photon fluorescence microscopy (to identify the nerve cells responding to a given orientation of visual lines), electron microscopy and computation. To connect the electron microscope images into a 3D reconstruction, the scientists first had to merge thousands of individual images for each slice into a single large image of the entire slice. They next aligned the adjacent slices. They could then combine these pairwise alignments over the entire stack to calculate exactly how to correct the distortion present in each slice and place its corrected image back into an aligned stack.

The group only reconstructed 0.03 cubic millimeters of the mouse brain, a volume that would go into a teaspoon about 167,000 times. But this still resulted in about 10 million camera images, amounting to roughly 100 Terabytes of raw data—about the memory required for nearly 30 million high-resolution, large-format photographs.

Greg Hood’s work was supported by a National Institutes of Health grant to the National Center for Multiscale Modeling of Biological Systems (MMBIOS).


Sometimes a breakthrough in computational research pays an extra dividend.That’s what Charles Peskin and David McQueen of New York University got when they employed PSC’s Cray C90 supercomputer in a 1993 effort to modelthe human heart in three dimensions. Their main goal was to improve the simulated flow of blood through the aortic valve.


The C90 allowed them to increase the number of points in the mesh-like grid used in the calculations, and the extra resolution did indeed render the bloodmotion through the valve more realistically. But it also fixed problems they’d been having with the right side of their simulated heart—a breakthrough that made the whole heartbeat work correctly for the first time in 15 years of research. Their accomplishment gained them the 1994 Computerworld Smithsonian Award for Breakthrough Computational Science; Peskin also received the 1994 Sidney Fernbach award.