Capturing Mitosis Genes in Action

During the one-hour drama that is human cell division, many genes enter and exit the stage. Until now, researchers did not know the identities of many of these actors, nor understand their various roles. Now, using a combination of high-throughput screening methods, time-resolved movies and a supervised machine-learning algorithm, researchers have identified 572 genes that are involved in mitosis in human cells. The raw data and images are available to the research community at


“Researchers can go to the database, do a clustering analysis, and extract the genes that are most intThis microscopy image captures the mitotic spindle (green) and the chromosomes (red) of a dividing cell. EMBL researchers videotaped mitosis for 22,000 different gene knockouts. Videos and data for all 22,000 genes are available at the mitocheck web site. Courtesy of Thomas Walter & Jutta Bulkescher / EMBL.eresting from their research question point of view,” says Jan Ellenberg, PhD, head of the Cell Biology and Biophysics Unit at the European Molecular Biology Laboratory and senior author on the paper published in Nature on April 1, 2010.


The research addressed an age-old problem in the study of cell division, Ellenberg says. “We didn’t know all the genes or the proteins involved,” he says. “So we decided that we had to do this gene discovery ourselves.”


First, Ellenberg and his colleagues in the Mitocheck consortium developed the technology to do systematic high throughput screens of multiple samples of all 22,000 human genes and then visually match each knockout to a phenotype. They relied on RNA interference to knock out each of the approximately 22,000 individual genes. They then printed more than 384 of these samples at a time on microarray chips. Because mitosis occurs transiently (approximately once every 24 hours), the researchers developed microscopes to capture movies of each sample from four such microarrays in parallel over the course of 48 hours.


Analysis of so much visual data—nearly 200,000 movies—required supervised machine learning. First, a human expert annotated examples of different morphologies observed within the movies. A computer then extracted a numerical signature with 200 different parameters that it correlated with those characteristics. After iterative training with movies of just 3000 different individual cells, the computer analyzed additional movies and identified phenotypes with 90 percent accuracy. The researchers also developed new distance measures for clustering algorithms to categorize the differences in cell division behavior.


The scale of these experiments and the use of time-lapse imaging over two days are “unparalleled and nothing short of phenomenal,” says Anne Carpenter, PhD, director of the Imaging Platform at the Broad Institute, who was not involved in the research. “[The insights into mitosis are] just the tip of the iceberg of the knowledge that will be extracted from this single experiment,” Carpenter says.


The researchers’ next project, called Mitosys, will explore the molecular activity of the 572 mitosis-related genes.

Post new comment

The content of this field is kept private and will not be shown publicly.
This question is for testing whether you are a human visitor and to prevent automated spam submissions.
Enter the characters shown in the image.