Skip to main content
Premium Trial:

Request an Annual Quote

Stockholm Royal Institute of Technology Releases First Version of Cell Atlas Resource


NEW YORK (GenomeWeb) – Scientists from the Stockholm Royal Institute of Technology (KTH) have released the first edition of their Cell Atlas resource, which provides intracellular localization data for more than 12,000 proteins.

The researchers presented the atlas, which is part of KTH's larger Human Protein Atlas project, earlier this month at the American Society of Cell Biology annual meeting in San Francisco. According to Emma Lundberg, associate professor at KTH and director of the Cell Atlas project, the data compiled by the effort suggests a surprising diversity of functions for proteins, with roughly 50 percent of proteins contained in the atlas appearing in more than one cellular compartment.

"That is higher than expected," she said. While the result might in some cases be due to antibody cross-reactivity, the KTH researchers have other data that also supports the idea "that many proteins are localized to multiple compartments and might even be moonlighting, performing different functions," she added.

"Location is tied to function, so a mitochondrial protein will not, for instance, become a Golgi protein in another cell line," Lundberg said. "But it might be found in the mitochondria in one cell line and in the nucleoli in another cell line. Or a protein might be cytoplasmic in one cell line, cytoplasmic and nuclear in another cell line, and nuclear in a third cell line."

The KTH team generated protein localization data in 26 different cell lines. In addition to variation between cell lines, they also observed a significant amount of variation between cells in the same cell line, including cells in the same image.

"For 16 percent of all proteins, we see variation between cells within an image," Lundberg said. "So, for instance, the protein might be expressed in some cells and not others in the same cell line in the same image. Or the distribution of the protein might be different. It might be localized to the nucleus in one cell and to the cytoplasm in the cell next to that one."

Lundberg suggested this variation might be due in part to the cells being in different phases of the cell cycle, an idea she and her colleagues are currently exploring.

"The cell is highly complex, with proteins that are both redistributing [across different cell types] and that show variation at a single cell level, and it probably has a lot of functional implications," she said.

She noted, however, that the antibody-based approach used in developing this initial version of the Cell Atlas was limiting in terms of exploring this variation.

"Since we are using an antibody-based approach, we had to fix and permeabilize the cells, so we are limited to this static image," she said. "To really study dynamic effects, if, for instance, a protein redistributes [in response to stimuli] or if it just localizes to different places in different cells, you would have to go into different types of imaging."

She suggested that approaches like live cell microscopy and GFP-tagged proteins could provide a more dynamic look at protein localization. "This is something we are doing, but that will be Cell Atlas 2.0," she added.

Classification of the images comprising the atlas is a fairly massive undertaking, Lundberg said, noting that while the atlas itself currently contains 80,000 images, she and her colleagues have more than 500,000 total images for analysis.

To aid in the process, the KTH researchers tried a citizen science approach, using online gamers to help annotate their images. Working with Icelandic video gaming company CCP Games, Swiss firm Massively Multiplayer Online Science, and researchers at Reykjavik University, they embedded a mini-game called Project Discovery within CCP's existing Eve Online game. Players are able to earn currency for use in the larger game by localizing proteins in the Cell Atlas images to different organelles.

As a May article in Nature Biotechnology noted, scientists have previously used gamers for projects, but, Lundberg said, these efforts have typically used standalone games developed specifically for the given research purpose. The Cell Atlas game, she said, was likely the first citizen science game to be integrated into an existing widely-played game.

"We thought, [Eve Online] already has half a million players," Lundberg said. "What if we [integrate Project Discovery] so we can get better participation and also don't have to recruit players?"

The KTH team released the minigame in March and it was "immediately way more successful than we expected," Lundberg said. To date, around 150,000 people have played the game, providing 30 million image classifications and around 50 working years of manpower.

"It really shows that if you want to tap into people's brains and use their brainpower, gaming is a really good way to go," she said.

The question remains, though, of how good the gamers are at assigning protein localizations, Lundberg noted.

"The players have a different incentive [from the KTH researchers], so the question is, is the scientific data as good?" she said, adding that she and her colleagues are currently analyzing the data provided by the gamers in an attempt to answer that question.

The analysis is still ongoing, but information generated by the gamers has helped her team reclassify some of the localization patterns and is represented in the initial Cell Atlas release, she said.

"I can already say that it has been useful," she said, suggesting that the gamers perform at a level comparable to that of a good machine learning approach.

Ultimately, Lundberg said, she envisions a process wherein annotations from gamers, machine learning tools, and experts like herself are combined, one feeding into and refining another.

"We have a big dataset of hundreds of thousands of images that have been annotated by experts, machine learning approaches, and a lot of gamers," she said. "So we can actually see how gamers, computers, and experts perform compared to each other. And then you can use the gamer data to refine the machine learning algorithm and maybe feed it back to the gamers, and when you do things like that, I am sure you can get superior results. That's my guess."