On the cell surface, anchored in the cell membrane, a wide array of proteins perform functions, which are vital for the cell. These proteins, collectively known as the surfaceome, are a cell’s antennae to the outside world, sending and receiving signals that enable it to communicate with other cells. They also serve as gate keepers for molecules, transporting materials into and out of the cell, and enable cells to attach themselves to other cells or structures.
The medical field has a keen interest in the surfaceome and uses it to treat diseases. Some two-thirds of known medications achieve their effect by slotting precisely into a specific surface protein and triggering a cellular signalling cascade.
Revamping the paradigm
Yet as practical and simple as that sounds, the current “one drug–one target” approach has a serious drawback: a given target surface protein can be found on many different cell types. Many drugs therefore attack not only the target cell type, but other cell types too. This is one reason why many drugs have unwanted side effects.
To find an alternative, researchers in Professor Bernd Wollscheid’s group at the ETH Institute of Molecular Systems Biology and the Department of Health Sciences and Technology (D-HEST) are investigating the distribution and organisation of proteins on cell surfaces. Their goal is to take a fresh approach hoping to identify more suitable targets for drug intervention.
To date, the variety of the surfaceome in human cells has hardly been researched. In an initial step to remedy this, doctoral student Damaris Bausch-Fluck and bioinformatician Ulrich Goldmann, both members of Wollscheid’s group, worked together to create an
in silico
inventory of these molecules. Their study was recently published in the journal
PNAS
.
The researchers made use of the benefits of machine learning in their work: First, they taught the computer to compile properties and features of surface proteins by feeding it with protein data collected in previous experiments. The computer turned presence of cell surface specific features into a score and then calculated a surfaceome score of the 20,000 or so proteins found in humans. Finally he predicted above which score a protein is likely to appear at the cell surface.
Predictions largely correct
In the end, the computer-generated inventory encompassed about 2,900 different proteins. In other words, out of all the proteins in a human cell, one in seven could appear on the cell surface. The newly developed algorithm achieved a high degree of accuracy in its predictions: a subsequent review of the experiment revealed that the computer was correct in more than 93 percent of cases.
In addition, the researchers were able to show that the number of surface proteins varies widely by cell type. Using publicly accessible data on cell lines, they were able to show that immune cells have only about 500 different surface proteins, whereas lung and brain cells have more than 1,000. But the cells that showed the greatest variety in surface proteins were primary stem cells, with about 1,800 different kinds. “Cell lines have a less complex surface proteome than cells that have just been removed from body tissue, since the interactions that cell lines undergo are less diverse,” Wollscheid says. The ETH researchers have stored their findings in a public database.
Organisation and distribution are key