How can you see the invisible? Microscopes can show you tiny details of cells, but there is a limit to their resolution. Siewert-Jan Marrink uses a supercomputer that visualizes the smallest details in cells by computing the interaction of molecules. It has earned him an important research grant from the NWO funding agency.

As a chemist, Siewert-Jan Marrink doesn’t like to get his hands dirty. He performs his experiments ‘in silico’, in hundreds or thousands of computer cores which may run his calculations for months on end. They show him the movement of molecules inside a small piece of cell membrane. ‘When I was a student, I hated those test tubes full of black goo that you had to clean at the end of the day’, Marrink jokes.

But clean hands aren’t his real motivation. ‘With computer simulations you can zoom in far beyond the capabilities of any microscope.’ His work, molecular dynamics, shows interactions that we wouldn’t know of otherwise. ‘But we also work with experimental groups that test our results.’

Marrink is continuing a tradition that started with Professor Herman Berendsen, now retired. Berendsen supervised the development of Gromacs , a programme that can calculate molecular interactions. ‘Basically, what it does is apply Newton’s Second Law’, Marrink says, with some understatement.

Gromacs is one of the leading molecular-simulation programmes. This kind of software uses classical mechanics to calculate interactions. ‘The software sees atoms or molecules as small balls, which is of course a simplification of reality. If you wanted the simulations to be more realistic, you’d have to use quantum chemistry as a basis, but then the calculations would take a lot of processing time, which would mean you could only simulate very short time spans.’

The art of simulation is therefore a balancing act: lots of detail will take a lot of time to compute, while a fast run may not provide enough resolution. Marrink has been working on this ever since he earned his PhD twenty years ago. ‘It requires a combination of intuition, hard thinking and rigorous testing to find the right balance’, he says.

Marrink developed the Martini parameter set , which defines the forces between particles. ‘Martini is what is known as a force field that provides input for programmes like Gromacs. It uses a coarse-grained model of molecules.’ This simplification of molecules makes it possible to simulate the interaction between dozens of proteins in a lipid membrane for a millisecond timescale.

So what results can he show? Marrink gives an example. ‘Many proteins in the membrane of mitochondria, the cell’s energy factories, are clustered in functional complexes. Experiments have shown that the fatty acid cardiolipin, which is part of the membrane, is necessary for the formation of these complexes. Our simulations showed that cardiolipin binds to specific sites on these proteins and pulls them together.’ Subsequent experiments confirmed the binding sites for cardiolipin.

Cellular membranes are central to Marrink’s work. They look dull in the textbooks – just a lipid bilayer in which a few proteins float around. ‘In the last decade, we have seen that membranes are much more complex’, Marrink explains. ‘A membrane can contain hundreds of different lipids, and proteins may constitute up to half the volume, often aggregated in functional complexes.’ The dynamics of all these components is a challenge for the simulation software. A dozen or so proteins is doable, but Marrink would like to simulate the interaction of hundreds of proteins. ‘And my ultimate goal would be to simulate an entire cell, with everything that takes place inside it.’

Marrink’s work explores two areas: improving the Martini force field and using it to find out how cell membranes work at a fundamental level. The latter is in close cooperation with several experimental research groups, at the University of Groningen and elsewhere. He has just received a EUR 780,000 research grant from NWO Chemical Sciences that will enable him to continue his work in both areas.

‘We do basic science, but the area of application is very wide.’ Hundreds of scientific papers have used Martini as a tool. The list of ‘hot papers’ on the Martini website shows that the software is applied in many fields, ranging from cell membranes to the lipids in tear fluid or the carbon compound graphene. ‘There is also a paper simulating lung surfactant, the fatty layer that is vital to proper lung function.’

There can be no question as to the relevance of Marrink’s work. Last year’s Nobel Prize for Chemistry was awarded to three scientists who developed computer simulations. ‘They are people that Berendsen collaborated with – and in my opinion, his work was as good as theirs’, says Marrink.

The Nobel laureates won their prize for multiscale models, which combine different levels of detail (from classical mechanics to quantum mechanics). ‘That is our aim for Martini: a model that can zoom in at different levels. Coarse-grained for speed, but with the possibility to switch to a higher resolution for part of the simulation, so that you can visualize crucial steps in a process. We’ve already got it working in a test system, so it’s within reach.’

Martini

Martini is a coarse-grained force field suitable for molecular dynamics simulations of biomolecular systems. The programme defines the forces between molecules, and can be used as input for Gromacs, the software platform originally designed at the University of Groningen, which is now maintained by experts from Sweden.

There are different versions of the Martini force field, such as a special version to simulate DNA and one in which the solvent (water) is ignored. The latter is called ‘Dry Martini’. So what’s the story behind the name? Marrink grins. ‘I’m from Groningen, and Saint Martin is its patron saint. And I do like a Martini, but that’s all there is to it.’ Which begs the question: shaken or stirred? ‘Shaken, as a rule.’

Simulations

For his simulations, Marrink uses the national supercomputing facility SARA in Amsterdam as well as a network of European supercomputers. He has his own computer cluster of several hundred CPUs at the University’s Centre for Information Technology. As a full simulation may take up to several months, it is important to test different parts before a full run. The most basic test can be run on a desktop. This is then scaled up to the computer cluster before being moved to the external supercomputers.