Using a Single Brain Physical Slice Library vs. Digitally Imaging Whole Brain at Once

Table of Contents

Movies of Lathe Microtome cutting and tape collection in action!

SBPSL Full PowerPoint Presentation (Warning large file! *.ppt file is 29Mbytes)

Movie: Piloting down a virtual neuron's dendritic tree using "Dendritic Explorer" test program (49 Second *.avi file, 22 Mbytes)

Dendritic Explorer test program overview slide

Contacts

Using a Single Brain Physical Slice Library vs. Digitally Imaging Whole Brain at Once

A frequently asked question is “Why do you need to store all the ultramicrotomed brain slices, imaging these slices only when an individual researcher requests? Why not just image the entire brain sequentially and then provide the digital image data, much quicker, upon a researcher’s request?” This is a good question and gets to the heart of why the concept of a Single Brain Physical Slice Library is a preferred route to the production of a Synapse-Resolution Brain Connectivity Atlas. Storing the slices instead of directly imaging them only makes sense if the time to slice a brain volume is many times faster than the time to image (at synapse-resolution) that same brain volume. The argument made here is that this is the case for current technologies.

Estimate of (mouse) brain slicing time ~1year:

For a mouse brain of ~1cm³ total volume, the brain is first diced into 1000 1mm³ tissue cubes, then embeded 27 tissue cubes at a time into 38 cylindrical tissue blocks each ~25mm in diameter (maximum circumference = 79mm). Each of these tissue blocks is placed in turn on the Automated Taping Lathe Microtome and a 100nm thick spiral of tissue is cut by advancing the diamond knife 100nm for each revolution of the cylindrical tissue block. In order to successfully slice all the 1mm³ tissue cubes in the cylindrical tissue block, it must be rotated at least 10,000 times producing a tissue tape ~790 meters long. The final (Mouse) Single Brain Physical Slice Library would consist of storing 38 of these 790 meter long tissue tape cassettes for user requested loading onto a transmission electron microscope and tomographic imaging at 10nm voxel resolution.

If we assume a slicing speed of 1mm/sec (at the conservative end of the range used in normal ultramicrotome use) then each tape would require 790m x 1s/0.001m = 790,000 seconds or ~9 days. Microtome tissue block and tape setup time should be insignificant compared to this cutting time. To produce all 38 tissue tapes then, 9 x 38 = 342 days or ~ 1 year would be required.

Estimate of full (mouse) brain digital imaging time ~4000years:

It is somewhat difficult to estimate the speed at which tissue slices can be sequentially imaged in an electron microscope since today’s researchers are not preoccupied with raw imaging speed and the technology has not been pushed in the direction of increasing this speed. For the purposes of this discussion we wish not to give a conservative estimate of the time to digitally image an entire mouse brain at 10nm voxel resolution but to estimate a reasonable lower bound on this imaging time. If this lower bound is on the same order as the ~1 year slicing time calculated above then it may be more advantageous to digitally image the brain slices sequentially and provide the digital image data directly to researchers instead of storing the slices for later, user requested, imaging.

For a mouse brain of ~1cm³ total volume the total number of 10nm x 10nm x 10nm voxels is 1x10¹⁸. If such a brain were sliced into 100nm slices (as per the above method) then the 10nm resolution in depth would still require using an electron microscope equipped with a goinometric tilt stage for electron tomography. Applying the tomographic resolution equation of [Soto 1994], an electron tomographic single-axis tilt series of ~60 images of a 100nm thick tissue slice can provide 10nm resolution in the Z direction. Collecting these images with a 2000x2500 pixel CCD, one of the largest TEM CCD's available today [NCMIR], will allow a square field of view to be captured in the imaging series having extents of 20um x 25um. Thus, this imaged 20um x 25um x 100nm piece of tissue will provide 5x10⁷ voxels after requiring 60 images be snapped. Since the tissue must be physically rotated by the goinometer stage between each picture, there will be a finite amount of time needed for each image. A liberal estimate of the speed at which each image could be taken is ~100 msec per image (actually many times this long is usually needed), giving 6 seconds to image the full 5x10⁷ voxels. In order to image the entire mouse brain via this protocol would require at least 1.2x10¹¹ seconds or ~4,000 years!

Assuming the arguments above are valid, production of a set of tissue tapes for storage using a single Automatic Taping Lathe Microtome is feasible; however, digitally imaging all the slices directly using a single TEM is not. Using many TEMs in parallel or developing radically different imaging protocols and instrumentation could of course lower the full digital imaging time to a more practical scale; however, these modifications would require an extremely large-scale (i.e. large budget) project to achieve. The solution adopted by the Single Brain Physical Slice Library approach is to allow individual researchers to direct precious imaging time intelligently. The random access nature of the imaging allows the researchers to design sparse imaging experiments that map the intrinsically multi-scale neural circuits of the brain while only imaging perhaps 1/10,000 of the brain’s total volume.

The conclusion is that a mouse brain physical slice library could perhaps be created in approximately one year (after the microtome is fully developed and debugged), and a sparse synapse-resolution brain connectivity atlas could then be incrementally generated by a large group of remote researchers using intelligently-directed random access imaging. Several years of this sparse, directed imaging would yield as complete an understanding of the mouse neuroanatomy as a complete digital image set would have, but would have required a fraction of the time/cost. As for a human brain, both a longer project duration and larger budget would be warranted, therefore parallel slicing and imaging could be used. The fundamental disparity between slice time and imaging time discussed above would remain however. Slicing the brain onto tapes for later random access imaging (a physical slice library) would remain the better option.

NOTE: The above arguments would need reviewing if a new imaging technology were used, especially one that required ablating the surface of the brain to reveal subsequent layers (i.e. destructive imaging) which would make storage and random access impossible

[Soto 1994] Soto, G.E., Young, S.J., Martone, M.E., Deerinck, T.J., Lamont, S., Carragher, B.O., Hama, K., and Ellisman, M.H. 1994. Serial Section Electron Tomography: A Method for Three-Dimensional Reconstruction of Large Structures. Neuroimage 1, 230-243

Last Updated: 11/09/2003