Network news

Hui Ge sifts through oceans of data to explore how genes collaborate

Consider the worm.

For a multi-cellular organism, Caenorhabditis elegans keeps things simple. A typical adult roundworm has 959 cells, no more and no less, and scientists have traced the exact lineage of each cell. The animal goes through life without a brain or much of a sex life (almost all are hermaphrodites).

But C. elegans also has about 19,000 genes—almost as many genes as humans. And just as in humans, no gene does its work alone. Instead, tasks are accomplished through highly complex networks of protein interactions.

Today's advances in systems biology began with genome sequencing. Hui Ge is among those creating more powerful next-generation platforms that integrate protein interactions and other kinds of high-volume analyses as well. Photo: Sam Ogden

This is the realm of systems biology, and of Whitehead Fellow Hui Ge, who studies embryonic development in the worm.

“I’ve always been interested in how a fertilized egg develops into a whole organism like us,” says Ge. She gets the big picture on this process by combining data from several high-throughput analysis techniques that cut wide swaths through the worm genome, and using advanced statistical methods to sort through the results.

This systems biology approach starts with the components of a system, and studies how those components work together to achieve a certain function, such as protein synthesis or protein degradation. “It’s important to know not just the individual components of a cell but how they are mapped together,” Ge notes. “It’s like a subway system; if you remove a station, the effect on the system will depend on its position.”

“These are data-driven approaches as opposed to the more traditional hypothesis-driven approaches,” she adds. “We put together all this high-throughput information, and then we can find predictions for uncharacterized genes and connections between them, which we can test. And this can be an iterative process in the lab. You make predictions and validate them, and then that validation can help your prediction techniques.”

Reading the "Interactome"

Ge graduated with a bachelor’s degree in biochemistry and molecular biology from the Beijing University in 1999, and won a scholarship to Harvard Medical School.

Ge and her co-workers in Mark Vidal's Harvard lab mapped out interactions between many proteins in the C. elegans work, in this 2004 Science paper. ↓ In a 2005 Nature paper, Ge and colleagues combined maps of protein interactions (blue), gene expression (red) and loss-of-function profiling from RNA interference studies (red). ↓ Next, the researchers filtered out a network "backbone" that grouped proteins shown to be interacting by at least two sets of data. ↓ When the researchers zeroed in on "sub-networks," they found that proteins whose functions were already known helped to characterize unknown proteins nearby. Images courtesy of Nature and Science

She arrived just as systems biology began to soar.

“I fell in love with the idea that you can study how the organism works, not just by studying individual genes but understanding what a lot of genes do at a time,” she says.

In a homework assignment for a class taught by Harvard’s George Church, Ge came up with a computational strategy that eventually turned into a paper in Nature Genetics.

The strategy was about correlating data from large-scale studies of genes with large-scale studies of protein interactions. More specifically, it was to correlate transcriptome data (reflecting which genes a cell expresses under certain conditions) with interactome data (mapping interactions between the proteins). Where the two data sets agreed, clearer pictures of protein roles would emerge.

“She was one of the first researchers to suggest putting data together from very different data sets to get something that was better than the sum of the parts,” says Harvard’s Marc Vidal, in whose lab Ge ended up.

Doing research on yeast, Ge and her co-workers demonstrated “the first global evidence that genes with similar expression profiles are more likely to encode interacting proteins,” as their paper put it. And they showed that the integrated data could help to improve hypotheses generated from either approach alone.

Getting worm

Also at the Vidal lab, Ge worked on a big project to map out much of the interactome of C. elegans. Eventually published in Science, this paper had no fewer than eight other co-first-authors. (The worm was the likely target because it was the first multi-cellular organism to be sequenced completely, in 1998.)

Next, Ge and colleagues tackled very early embryogenesis—the process by which the worm divides twice, into four differentiated cells, in the first hour after fertilization.

The scientists combined data from three sources: protein-protein interaction, gene expression, and loss-of-function profiling based on RNA interference. They then made predictions about how the embryonic “molecular machines” work. Testing 10 uncharacterized proteins by seeing where they popped up in live animals, the researchers found that the locations generally were consistent with the proteins’ predicted roles, findings reported in a 2005 Nature paper.

Completing her PhD in genetics, Ge was picked as a Whitehead Fellow. Before starting at the Institute, though, she spent six months at the lab of Harvard’s Craig Hunter, learning the craft of worm wet-lab work.

Healthy pairs of genes

At Whitehead, Ge and colleagues have embarked on two main projects with the worm, the first being further explorations of genetic interactions during embryonic development.

RNA interference studies have highlighted about 2,500 genes whose loss kills the worm embryo. That number seems pretty small compared to the total of around 19,000 genes, she says, and she suggests that it’s because genes can back up each other’s functions.

“We are combining the protein interaction map with the genetic interaction map to predict these pairs that give you a synthetic phenotype,” Ge says. “These genes are not functionally equivalent, but they complement each other’s function. Knowing these kinds of genetically buffering pairs is very important for understanding development but also for understanding disease.” She gives the example of the mammalian p53 tumor suppressor gene: “Even if you knock down p53, a mouse will not get cancer immediately. It increases the chance that when something else is damaged, the mouse will get cancer.”

The second major effort is to take a dynamic view of gene expression that integrates time and space information. “In multi-cellular organisms, at different locations, the molecular networks are actually different, because not all of the genes are expressed at the same time,” Ge says. One student in her lab, for instance, is adding data about location and trying to predict the genes that are expressed in certain tissues such as muscles or skin.

“Quantitative science is providing us with a process that helps us understand biology more quickly and in a systematic way,” says Ge. “Little by little, we are learning how to achieve these projects.”

Today’s attempts at detailed molecular modeling are “still first draft, still relatively fuzzy—just like genome sequencing once was,” acknowledges Ge’s mentor Vidal. “But they are really shaping up.”