News & Views | Published:

Bringing fly brains in line

Nature Methods volume 8, pages 461463 (2011) | Download Citation

Software for fast and accurate alignment of brain images is used to generate a partial brain atlas for Drosophila melanogaster and should enable circuit mapping.

Comparative brain anatomy is an old field that has undergone a major technology-driven boom in the last decade. The synergy of confocal microscopy with ever-improving and more affordable computer performance has led to the construction of several three-dimensional 'standard brains', formed by averaging brain images taken from multiple individual animals1. In this issue of Nature Methods, Peng et al.2 use brains from Drosophila melanogaster to demonstrate the utility of their latest computational tool, BrainAligner, to align large numbers of brain images with high quality and speed (Fig. 1).

Figure 1: Schematic showing that images of various parts of the fruit fly brain can be aligned with BrainAligner software to generate a neural map.
Figure 1

Early attempts to generate standard brains often required the sample brain images to be manually annotated before registration, which is both hugely time-consuming and prone to human error. Computer algorithms were subsequently developed to automatically extract common features from collections of images, but they required enormous computational power and time, and the standard brains generated were of variable quality.

The freely available BrainAligner software combines and optimizes some of these preexisting methods to provide fast and accurate automatic registration of brain images and should be useful to most investigators interested in neural circuit construction. In addition, the potentially high-throughput nature of BrainAligner allows one to efficiently handle large amounts of anatomical data, which is essential for grand-scale anatomy based screening.

In the paper published in this issue, Peng et al.2 use BrainAligner to construct a 'standard' or 'target' fly brain by statistically averaging images from 295 fly brains labeled with an antibody that recognizes a general neural process marker (the nc82 antibody that labels the presynaptic protein bruchpilot). They then manually annotated the target fly brain to add 172 conserved brain compartment positions as 'landmarks'. A partial brain atlas could then be constructed by automatically aligning each of 470 discrete expression patterns from 2,954 sample brains, into the target brain, guided by the landmarks in the nc82-stained neural tissue.

BrainAligner selectively aligns each sample using only the most reliable landmarks, allowing it to maintain accuracy despite variable nc82 staining quality, partial tissue damage and image distortion. Dropping landmarks that fall outside the normal statistical variance led to quantifiably less error-prone warping of samples (expressed as the percentage of landmarks used for each alignment).

The authors improved alignment speed 50-fold by implementing hierarchical interpolation to generate the final warping field. In other words, the pixel resolution in each dimension (that is, the voxel resolution) of the image stack was first reduced or 'downsampled' by four, and after warping, the original resolution was approximated by interpolating the remaining voxels. This allowed two three-dimensional image stacks of 1,024 × 1,024 × 256 voxels to be aligned in about 40 minutes on a standard computer. Therefore, using BrainAligner and our desktop computer it would take 90 days to repeat the 3,248 alignments performed by Peng et al.2—a remarkable 12-year improvement in processing time over previous methods.

What might your average neurobiology researcher use BrainAligner for? An obvious application would be an automatic search of brain images for enhancer trap lines that express in overlapping brain regions. Although neural cell body position is highly variable, BrainAligner can find similar neurons using their primary neural tracts.

One might also wish to construct one's own atlas of the entire brain or of neurons comprising individual regions of the brain. This can be done by double-staining brains with nc82 and aligning the new images with the target brain constructed by Peng et al.2 or by constructing one's own target brain—with landmarks generated using nc82 or another robust antibody—and then aligning new samples using that same set of reference points. The key is that once a specified 'target' brain with reference label is established, one can align images to it so long as the program can retrieve some of the landmarks in the target. If the standard developed by Peng et al.2 was adopted by the entire community, we could all in principle log our expression patterns to a common reference.

Identification of neural connectivity would be a big bonus of such a brain atlas project, but realistically it can only indicate neurons that are putative synaptic partners. Nevertheless, narrowing down to such putative partners for functional validation would expedite the process of circuit mapping. Although a map of physical connections can be useful, a full understanding of circuit function requires additional types of knowledge, for instance, about the neurotransmitters involved, the electrical properties of component neurons and the influence of modulatory systems.

Will BrainAligner become the software of choice for the community? History suggests that will depend on more than alignment quality and speed. BrainAligner is freeware, and it is well integrated with the V3D and AtlasViewer freeware developed by the same group3. Price is therefore not an issue, but documentation and technical support, platform dependence (BrainAligner is currently available in Macintosh and Linux formats) and the availability of updates could be. Furthermore, there are other promising ventures in various states of development such as Flybrain@Stanford4, BrainGazer5 and FlyCircuit6 that have similar goals, so only time will tell. Let the bidding begin!

References

  1. 1.

    Front. Syst. Neurosci. 4, 26 (2010).

  2. 2.

    et al. Nat. Methods 8, 493–498 (2011).

  3. 3.

    , , , & Nat. Biotechnol. 28, 348–353 (2010).

  4. 4.

    et al. Cell 128, 1187–1203 (2007).

  5. 5.

    et al. IEEE Trans. Vis. Comput. Graph. 15, 1497–1504 (2009).

  6. 6.

    et al. Curr. Biol. 21, 1–11 (2011).

Download references

Author information

Affiliations

  1. Wolf Huetteroth and Scott Waddell are in the Department of Neurobiology, University of Massachusetts Medical School, Worcester, Massachusetts, USA.

    • Wolf Huetteroth
    •  & Scott Waddell

Authors

  1. Search for Wolf Huetteroth in:

  2. Search for Scott Waddell in:

Competing interests

The authors declare no competing financial interests.

Corresponding author

Correspondence to Scott Waddell.

About this article

Publication history

Published

DOI

https://doi.org/10.1038/nmeth.1615

Newsletter Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing