May 2012

Feasible fuels

Large-scale computing proves an important tool for testing hypotheses about cellulose-derived fuels.

This simulation of an enzyme found in while rot digesting a million atoms of plant cellulose models a process that humans would like to recreate to make fuel from wood waste. The National Renewable Energy Laboratory (NREL) create the visualization in collaboration with Cornell and Pennsylvania State universities. See the sidebar for a related video.

In the search for renewable energy, converting biomass into fuel products is one promising option for replacing petroleum. But to efficiently transform a basic starting material in biomass – cellulose – to a liquid fuel, researchers must first understand that material’s chemistry and physics.

With joint funding from DOE’s Biological and Environmental Research and SciDAC (Scientific Discovery through Advanced Computing) programs, researchers at the National Renewable Energy Laboratory (NREL) have taken advantage of large-scale computing resources to look closely as possible at cellulose.

Cellulose contains long chains of sugar molecules – the polymers that give plants their stiff structure. As a complex carbohydrate, it packs energy, but the sugars are strung together with bonds to produce long, almost flat ribbons. These linkages create long chains that clump together into insoluble, three-dimensional twisted structures called microfibrils. Microfibrils in plants are just 3.5 nanometers wide but can be tens of thousands of times longer than that, says James Matthews, an NREL research scientist and cellulose expert. Their small diameter makes microfibrils especially difficult to study in a laboratory. Says NREL colleague Michael Crowley, principal scientist in the Biomolecular Sciences Division, “This is a huge gap in knowledge in understanding these microfibrils – what they’re like and how they behave.”

Large-scale computing provides an important tool for understanding and testing hypotheses about these materials. Information from simulations can help laboratory researchers interpret their findings. But for theory and experiment to work well together, researchers must ensure that assumptions embedded in computation are consistent with the physics and chemistry observed in the laboratory.

In his initial cellulose studies, Matthews uncovered computational results that were unusual and didn’t match experimental data. These outliers occurred at the end of the simulations, at the longest time points. He and his colleagues wondered whether the results reflected unobserved properties of the small-diameter cellulose or if the computation didn’t accurately reflect real-life data. To distinguish between those possibilities, they needed to run longer computations.

Plant polymers are ‘a well-designed mess, but from our point of view it’s difficult to understand on a molecular level.’

Large-scale calculations of chemical structures require significant computer resources; as a result, researchers typically limit simulations to capture the shortest possible relevant period. Extending the duration of a theoretical computation doesn’t simply require running the calculation longer. The exponential increase in computational requirements often demands substantial new programming. Otherwise, someone might set up a problem that would require so much computing time and resources that answering the original question would be impractical or impossible.

Cellulose presented this type of problem. To begin solving it, Crowley, Matthews and their colleagues relied on DOE’s Advanced Scientific Computing Research program for the programming necessary to extend their calculation timeline by a factor of 1,000 – from nanoseconds of theoretical time to microseconds. They also compared results across several simulation programs and sugar models so that all these programs put out data consistent with each other and consistent with laboratory findings. The computations, supported by a National Science Foundation TeraGrid grant, were done on Ranger, a supercomputing cluster at the Texas Advanced Computing Center at the University of Texas at Austin. The team recently published its results in The Journal of Chemical Theory and Computation.

The researchers also studied how heating cellulose can change the overall structure of microfibrils. In a paper published last year in the Journal of Physical Chemistry B, they showed that heating altered the orientation of one of the groups on the repeating sugar molecules, which changed the overall packing of the polymer and caused the microfibril to locally untwist.

As a part of that larger investigation, they also learned how cellulose might be harnessed for renewable energy. The researchers already knew that cellulose, when heated to high temperature and then cooled, converts from one of its naturally occurring forms, called cellulose Iα, to the other form, Iβ. Although both forms occur in plants, enzymes and chemicals break down cellulose Ia more quickly, Matthews says. Therefore, if researchers could understand the keys to making this form of cellulose, maybe they could find ways to develop fuels more efficiently.

In a paper published in the journal Cellulose, the researchers compared these two forms and what happens when they’re heated to high temperatures. Scientists previously had assumed the reason heating and cooling transforms cellulose Iα to cellulose Iβ was simply because cellulose Iβ was more stable. Their simulations, however, indicated that the polymer’s high-temperature form closely resembled cellulose Iβ. These results provided clues that may help researchers break down cellulose more efficiently.

These lessons already are feeding back into gaining a better understanding about these molecules and other polymers produced in plant cells, such as lignin, pectin and hemicelluloses. “They’re kind of in a mess,” Crowley says. “It’s a well-designed mess, but from our point of view it’s difficult to understand on a molecular level.”

These studies also are helping with other NREL projects on enzymatic reactions to break down cellulose. “Enzymes are now a large part of cost of producing ethanol and other fuels,” Crowley says, so information about how to efficiently break down cellulose to sugars will help lower production costs. Computational data has enabled them to trust their enzyme research, which investigates why some enzymes are better than others in doing the same thing and how to tweak efficiency in breaking down sugars. “All of this depends on getting the substrate (cellulose) right,” Crowley says.

The results of these studies also have larger implications for the computational community as a whole. They contribute valuable information to community codes, or the software tools that computational researchers use to set up parameters and run molecular simulations. Thus, Crowley notes, the results are helpful not only to researchers studying biofuels but also to other fields of chemistry that take advantage of computation, especially the biomedical and pharmaceutical industries.

This research also highlights the synergy between SciDAC, which supported the programming work, and the Center for Direct Catalytic Conversion of Biomass to Biofuels (C3Bio), one of the DOE’s Energy Frontier Research Centers, which has supported the scientific research.

“What’s so nice about that joint project through SciDAC,” Crowley says, “is to be able to have the performance guided by what’s needed and then have the science enabled by the coding and performance.”