On my desk sits a tattered photocopy of one of the pinnacles of modern developmental biology, the “embryonic lineage” paper by John Sulston, et al. (1983). In this paper, Sulston et al. completed a project begun in the late 19th century, namely to trace the complete genealogy of all cells in a nematode embryo. C. elegans, like many (although not all) nematodes, displays highly invariant development, each embryo developing to an adult with the same number of cells. The introduction of Nomarski DIC optics in the 1960s made it possible for the first time to trace all cells and their divisions in live embryos, a feat performed without significant aid from video recording, and well before the era of computer image analysis.
The ‘embryonic lineage’ paper remains a treasure trove of information and insights. Yet many aspects of development are not captured in the lineage, most significantly information on dynamic movements and neighbor relationships. Further, the extreme amount of time and patience required in manual cell lineage analysis meant that much of this knowledge was unused for several years until video recording allowed manual tracking of cells from a single embryo with tools such as Simi Biocell. A breakthrough was the development by Zhirong Bao and Bob Waterston’s lab of computer based tracking algorithms to automatically trace cell lineages from 4D movies of nuclear histone-GFP markers.
Completely automated nuclear tracking is highly efficient in the early embryo (up to 350 cells). The processes of most interest to my lab, including ectodermal morphogenesis and organogenesis, begin after 350 cells, and our efforts to automatically track cells in these later stages were unsuccessful. Essentially the embryo becomes too crowded with nuclei; also, compromises in illumination and image quality needed to avoid phototoxicity in long 4D confocal movies result in images that are not clear enough for complete automation to be efficient.
Claudiu Giurumescu, a postdoc in my lab, took a different approach to the problem of tracking nuclei in the crowded environment of the later embryo. Importantly, he decided to use a combination of automatic tracking and manual curation. The tracking relies on the predictable behavior of nuclei in worm embryos: most nuclei do not move around much on the time scale used in 4D movies. Claudiu devised algorithms that took advantage of this predictability to search locally for each nucleus at a given time point, based on the information on where the nucleus is at the previous time point. Of course, this means the user has to identify all nuclei at the first time point in the series, which is usually easy for an early embryo. As all nuclei are either tracked or flagged for curation at each time point, error propagation is minimized.
We first tried this approach on embryos imaged with conventional laser scanning confocal microscopy, and were able to successfully follow all nuclei up to the point at which embryonic muscle movements interfere with tracking, a time when all but four nuclei have been generated in the embryo. Sukryool (Alan) Kang, a student with Pam Cosman, played a major role in refining the visualization tools and in quantitative analysis of cell movements. The resulting dynamic models of the embryo can be visualized in a variety of ways, as shown (at rather low resolution) in our Supplementary Material, and at higher resolution on our lab web page: http://126.96.36.199/~wormlab/. Our Matlab code and user manuals are publicly available on Sourceforge. We are still refining the visualization tools and plan to integrate 4D movies more directly with the lineage tree.
We next wanted to assess the generality of our semiautomated approach. We first collaborated with Thomas Planchon and Eric Betzig (Janelia Farm), whom we had met one summer at the MBL in Woods Hole while they were demonstrating their novel structured illumination approach, Bessel beam microscopy. Bessel beam illumination has much higher z-resolution than standard confocal movies, with reduced phototoxicity. Fortunately, Bessel beam 4D movies of C. elegans embryos proved highly amenable to our semiautomated tracking.
How well does this approach work in samples where development is less predictable? To answer this we struck up a collaboration with Debbie Yelon’s lab, our neighbors at UCSD, who were interested in tracking nuclei in zebrafish cardiac morphogenesis. Using data generated by Josh Bloomekatz, our tracking algorithms were able to track large numbers of zebrafish nuclei with only minor modifications.
Our work adds another tool to the toolbox for anyone interested in tracking large sets of nuclei, or similar features, in complex samples. Fully automated tracking remains the method of choice in simple samples where nuclei are well separated and can be unambiguously tracked from frame to frame. Semiautomated tracking allows one to go further into development, and opens up the prospect of quantitative analysis of morphogenetic stages of embryogenesis.
Sulston JE, Schierenberg E, White JG, Thomson JN. The embryonic cell lineage of the nematode Caenorhabditis elegans. Dev Biol. 1983 Nov;100(1):64-119. PubMed PMID: 6684600. (Full text at WormAtlas)
Bao Z, Murray JI, Boyle T, Ooi SL, Sandel MJ, Waterston RH. Automated cell lineage tracing in Caenorhabditis elegans. Proc Natl Acad Sci U S A. 2006 Feb 21;103(8):2707-12. Epub 2006 Feb 13. PubMed PMID: 16477039; PubMed Central PMCID: PMC1413828.
Giurumescu, C.A., Kang, S., Planchon, T.A., Betzig, E., Bloomekatz, J., Yelon, D., Cosman, P. & Chisholm, A.D. (2012). Quantitative semi-automated analysis of morphogenesis with single-cell resolution in complex embryos, Development, 139 (22) 4279. DOI: 10.1242/dev.086256