Dazzling data show proteins in action

Photo: Sam Rentmeester

Photo: Sam Rentmeester

A new technique for imaging biomolecules is generating more data than pathologists can handle. Dr Raf van de Plas of 3mE is developing algorithms for distilling useful information from terabytes of data.

The screen displays hundreds of tiny images of a mouse in longitudinal cross-section. Although they look alike, each one is unique. Each image shows the distribution of a particular protein, and each protein is separated from all others because it has a slightly different mass. ‘The set consists of 2000 images, each containing 122 thousand pixels’, explains Raf van de Plas. ‘Each image shows the distribution of a certain molecule. This image shows that this molecule is less common in the liver, but more common in the intestines. And this molecule does not occur in the brain, although it is present in the skin and bones’. The images are the result of a new imaging technique known as Mass Spectrometry Imaging, or MSI. Mass spectrometry is a commonly used analytic method that ranks molecules according to their mass. It is used in the identification, quantification and profiling of isotopes, molecules and molecule complexes in very small quantities of chemical and biological mixtures.

Crystal layer
Although the principles of mass spectrometry have been known for a century, it is only relatively recently that it has been possible to apply the technique to larger quantities of biological molecules (e.g. fatty lipids and proteins). In the past, the problem had been that the process of ionisation, which is necessary to separate molecules from there surroundings and apply a charge to them, would cause large biological molecules to be broken into pieces. In 2002, researchers received the Nobel prize in physics for an invention that made it possible to use mass spectrometry to examine biological macromolecules intact. In essence, the invention involves coating the biological sample with a layer of crystal, which absorbs the energy of the UV laser and transfers it to the underlying tissue. ‘It could be compared to the crumple zone in a car’, explains Van de Plas, in reference to the technique known as matrix-assisted laser desorption ionisation (MALDI). ‘Without the crumple zone, the shock would break the bones in your body. If the energy is distributed, however, you come out without a scratch’. This has also proven to be the case for biomolecules in a sample. They are released, receive a charge and are sorted and counted by a mass spectrometer according to their mass.

The spectrum of molecule mass ranges from 500 to 30,000, with a maximum resolution to distinguish between the various molecules. For each location of at least 5 x 5 micrometres in the tissue, the technique registers all detectable molecule masses across the entire spectrum. A slice of mouse brain measuring 1 x 2.5 centimetres could easily generate 1.5 terabytes of raw data – for a single experiment on one strip of tissue, that is 1.5 times more data than could fit on the hard drive of an iMac.

Although MSI is currently used primarily as a tool for research it also has clinical applications.

This is only the beginning, however, because one measurement is not enough. Multiple records of the same mouse (technical replicates) are thus needed, along with multiple records of different mice (biological
replicates), as it is impossible to conduct research and draw conclusions based on a single measurement from a single mouse. Comparisons between healthy mice and diseased mice are needed as well. It would also be nice not to be limited to a single 2D crosssection, but to have a 3D volume consisting of a hundred cross-sections. The size of the datasets generated with this technique could thus amount to several petabytes (millions of gigabytes).

Machine learning
Van de Plas completed his engineering degree in Leuven, where he also earned a Master’s degree in artificial intelligence, as well as a doctorate. He then started to work at the Mass Spectrometry Research Center at Vanderbilt University in Nashville, Tennessee. He has recently become active as an assistant professor within the Numerics for Control and Identification group at the Delft Center for Systems and Control (3mE faculty). In this facility, he is focusing on the development of algorithms that could derive useful information from gigantic datasets, like those from MSI. One of the directions that he is pursuing in this regard is machine learning. In machine learning, the computer interprets such matters as the distribution of proteins according to an anatomic atlas, stating in medical terms whether the quantity of particular molecules is increasing or decreasing in a particular organ. Another approach is that of correlated behaviour. If a researcher is interested in an increase or decrease of a particular protein, Van de Plas can show which other substances are associated with that protein, as their concentrations show a simultaneous increase or decrease. Although MSI is currently used primarily as a tool for research, it also has clinical applications. One example is the pathological tissue studies conducted in tumour operations. Here a pathologist assesses a piece of tissue in order to determine whether all of the edges of the surrounding the tumour material that has been removed consist of healthy tissue – if so, the tumour has been removed successfully. Van de Plas: ‘In some cases, MSI also reveals chemical changes in tissue that appears healthy to the naked eye’. Pathologists could make better assessments if they also had access to this biochemical information.


Stay informed about the research

Receive the Delft Outlook newsletter 4 times a year