Better, Cheaper Biofuels through Computational Analysis

Better, Cheaper Biofuels through Computational Analysis

Developing and applying a powerful arsenal of computational tools for producing biofuels.

It's tough to break down woody material into its component sugars—so tough that scientists describe this characteristic as the "recalcitrance of biomass." Making the process even more difficult is the fact that the natural cellulase enzymes that can do the job act very slowly.

This recalcitrance is actually part of a plant's natural defenses that keeps it from being decomposed. But it is also a major roadblock for researchers eager to find ways to extract a plant's sugars to develop ethanol, advanced biofuels, and other chemicals. Researchers are seeking different methods to deconstruct abundant, renewable lignocellulosic biomass to access its sugars as essential intermediates for conversion to biofuels.

To unlock these sugars, scientists at NREL are using innovative computational-based studies to explore solutions to this problem. The NREL researchers are improving our understanding of the scientific basis for plant recalcitrance, so that they can develop sustainable feedstocks that are easier to deconstruct. They are also working to learn more about cellulase and hemicellulase enzymes that can quickly and inexpensively produce sugars from lignocellulosic biomass such as corn cobs and stalks, switch grass, poplar trees, and woody residues.

Pursuing an in silico Approach

For more than 30 years, NREL researchers have made significant experimental advances in understanding the intricacies of polymers found in plant cell walls, as well as the difficulties involved in breaking down the polymers to fermentable sugars. But while experimental studies are critical, this research approach can face daunting challenges in trying to fully understand the complexity of polymers within plants, across spatial scales spanning the atomic to macro.

To accelerate this understanding, NREL scientists increasingly use computational (or "in silico") studies to complement their experimental work. Computational results can provide insights that are inaccessible through experiment, while also identifying the next set of experiments to pursue to confirm or reject given hypotheses. The in silico approach, which includes modeling, simulation, and visualization, targets:

  • Understanding basic phenomena—for example, what is the detailed nature of cell walls, microfibrils, cellulose, and hemicellulose? How are cell wall structure, chemistry, and enzyme activity interrelated?
  • Using the deepened understanding to develop improved processes to deconstruct cell wall polymers to sugars—for example, significantly boosting the amount of biomass substrate that is converted to sugars, measured against how long it takes and how much enzyme is required to make the conversion happen. Finding enzymes that are more effective, thermally stable, and lower in cost will facilitate industrial production of biofuels from lignocellulosic biomass.

Starting with Cellulose—One Tough Material

Cellulose is a carbohydrate polymer—long chains of linked glucose molecules called cellodextrin, with these long chains combining to form crystalline bundles. Researchers have characterized several different crystallographic forms of cellulose, including natural forms found in plant cell walls, and various crystalline forms created by treating natural cellulose with chemicals such as ammonia or ionic liquids.

Illustration of three horizontal slightly twisted layers of strands that have short, alternating, blue and red sections. Above and touching these bundles is another layer with one yellow and two blue and red strands. Four snapshots of the yellow strand are seen pulled upward from this bundle, like stages in peeling a portion of a banana peel, with the shortest strand to the left and the other three strands being progressively longer toward the right. Enlarge image

Yellow chain in simulation shows stages of pulling a single cellulose chain out of a crystalline cellulose microfibril. The natural twist in the microfibril is evident in the bottom three layers. NREL scientists calculated the thermodynamic barrier that an enzyme must overcome to decrystallize cellulose.
Credit: NREL

Recently, NREL scientists have used the RedMesa supercomputer at Sandia National Laboratories to determine the thermodynamic work required to remove a single cellodextrin chain from four different forms, or crystalline polymorphs, of cellulose. The computational analysis revealed that it takes about an equal amount of thermodynamic work to remove a cellodextrin chain from natural plant celluloses (polymorphs Iα and Iβ), whereas removing a cellodextrin chain from cellulose II and III—two polymorphs that come from chemically pretreated natural cellulose—is considerably easier. This implies that celluloses II and III will be less recalcitrant, or resistant, to enzyme deconstruction than natural plant cellulose. These results provide critical understanding that can help guide future development of biomass pretreatment processes and potentially enhanced biological catalysts and genetically engineered plants, which may significantly reduce the cost of producing lignocellulosic biofuels.

In other studies, NREL scientists modeled cellulose microfibrils to learn how temperature affects the structure of plant cell walls at the nanoscale. Individual microfibrils are cellulose chains twisted together into ropelike structures that can be thousands of nanometers long, but are only 3.5 nanometers thick. Heating, which is part of the biomass pretreatment process, causes many changes in the cell wall; but these are difficult to study experimentally and therefore have not been well understood at the molecular level. Through molecular simulations, however, researchers learned that heating is predicted to cause the cellulose microfibrils to straighten. As a result, the microfibrils' "twist" is compressed into short regions with reduced crystallinity that may be easier to break down. The simulations also provide an explanation for experimental observations that heat-treated cellulose deconstructs significantly faster at the beginning of hydrolysis (the chemical reaction that breaks the bonds in the cellodextrin chain) than at the end. NREL scientists hypothesize that the highly twisted regions are more susceptible to hydrolysis and break down first, while the straighter regions remain recalcitrant.

Getting to Sugars—Deconstruction by Enzymes

In nature, fungi and bacteria use cocktails of enzymes to digest biomass. Although there are numerous individual enzymes involved, the enzymes that deconstruct cellulose can be classified as one of three basic types.

The first type of enzyme is an endoglucanase. Endoglucanases attack single cellodextrin chains within the interwoven network of crystalline cellulose fibers. Endoglucanses cut the cellodextrin chains in the middle, creating two new chain ends in the process.

The second type starts working where the first leaves off. This enzyme, which is called an exoglucanase, attaches itself to one of the new loose ends and pulls the cellulose chain out of the crystal structure. The end of the chain threads into a tunnel or notch in the enzyme, and it is cut by the exoglucanase into cellobiose, a disaccharide containing two linked glucose molecules similar to table sugar.

The third type of enzyme, a beta-glucosidase, catalyzes the hydrolysis reaction between water and cellobiose units, and cuts them into their glucose molecules. This glucose can then be fermented into ethanol, butanol, hydrocarbons, or other biofuels.

Illustration showing horizontal green rectangle (cellulose microfibril) along the bottom with one single green strand pulled upward to the right from this rectangle. The single strand is surrounded by a cage-like cluster of light-blue ribbon and thread shapes (the CD) that has three dark blue, small, irregularly shaped masses attached to it, two to the right and one on the left. To the left of the cage-like structure, attached to it at the top left of the structure, is a thin thread that attaches to a smaller ribbon-like mass (the CBM) to the lower left that is sitting on top of the green rectangle. The thread has three yellow, small, irregularly shaped masses attached to it. Enlarge image

The cellobiohydrolase enzyme from T. reesei consists of three sub-domains: (left) a carbohydrate-binding molecule (CBM); (middle) a linker peptide with attached polysaccharides (yellow); and (right) a catalytic domain (CD) with attached polysaccharides (blue). A single cellulose chain from the cellulose microfibril (green) is hypothesized to thread into a tunnel in the CD; the chain is cleaved and the two-sugar product, cellobiose, is expelled from the right end of the tunnel.
Credit: NREL

The typical exoglucanase has three "sub-domains," shown in the following figure. To reduce the number of atoms involved in the simulations and thus reduce the computational resources required, many functions of the enzyme can be computationally modeled by focusing separately on each domain. For example, NREL researchers have examined the movement of the carbohydrate-binding molecule (CBM) on the surface of cellulose microfibrils by combining computational analysis with experimentally determined three-dimensional structures. A key result of this research is the discovery of attractive attachment locations, or "energy wells," for movement of CBMs on cellulose surfaces that are spaced one nanometer apart. These energy wells are a natural fit for the CBM, and their spacing equals the dimension of two glucose units in the cellulose crystal, which is the distance required for the intact enzyme to move to produce the cellobiose product.

Researchers studying the catalytic domain (CD) are also using computational analysis. Their technique estimates the energy barriers between elementary steps in the enzyme-catalyzed reaction, and identifies where the most significant energy barriers exist. Specifically, their free-energy calculations predict the energies required to thread the cellodextrin chain into the tunnel, hydrolyze the chain into cellobiose units, expel the cellobiose product, and advance the cellodextrin chain another cellobiose unit to prepare for another hydrolysis step. The researchers have gained further insights about what occurs at a molecular level using techniques such as rare-event simulation to study threading, quantum mechanical/molecular mechanics to study the hydrolysis mechanism, and steered molecular dynamics to study cellobiose product expulsion. In the study of cellobiose expulsion, computer simulations of the expulsion provided data that helped researchers suggest mutations in the exoglucanase catalytic domain to enhance the product expulsion step. These suggestions are presently being incorporated into engineered enzymes to test the predictions.

Unlike enzymes produced by the fungus T. reesei, those found in the bacterium Clostridium thermocellum form a cellulosome, which is a very large multi-enzyme complex. The cellulosome is like a "multi-member demolition crew" dedicated to the job of attaching to a plant cell wall and working synergistically to dismantle the cell wall's polymers into sugars.

Illustration showing light blue curved shape to right labeled Bacterial Cell. Three small black ovals labeled SLH lie along the bacterial wall shape, with one connected by a line to the left to a small black oval labeled Cohesin II. Touching this black oval to the left is a small dark blue oval labeled Dockerin II. To the left are about ten similar small dark blue ovals (indicated collectively as Cohesin I) that look like beads along one horizontal string. Perpendicular to each of most of these beads is a smaller gray oval bead (indicated collectively as Dockerin I) connected to a larger gray oval bead (indicated collectively as Catalytic Domain) labeled with a number. Enlarge image

Cartoon of cellulosome from C. thermocellum, where scaffoldin subunit (dark blue) contains nine cohesins and a carbohydrate-binding module (CBM). Cellulolytic enzymes (gray) bind to cohesin partners with their dockerins. Another set of dockerin/cohesin interactions connect the scaffoldin to the cell wall via a protein (SLH).
Credit: NREL

Structurally, cellulosomes are made up of a scaffoldin, which is a long flexible protein that has specific binding sites called "cohesin domains," and multiple enzymes that have dockerin modules that bind to the cohesins. The cellulosome self-assembles because the dockerin modules on the enzymes recognize complementary cohesin sites on the scaffoldin.

Researchers are not yet sure exactly how this particularly effective nanomachine works. They do know that Clostridium thermocellum secretes enzymes and scaffoldins that self-assemble outside the organism. Because NREL's researchers want to better understand this molecular self-assembly process—as well as the computational resources needed to perform an atomistic simulation—they created a "coarse-grained" computational model containing only features that significantly contribute to the properties they want to simulate. The researchers assumed that the subunits' molecular weight, flexibility, and affinity for each other would have the greatest influence on the self-assembly.

Growing NREL's Computational Arsenal

In pursuing their computational work, NREL scientists encounter modeling issues across multiple scales. Some research problems require analysis at the atomic level; others at the scale of macromolecular interactions; and others at a coarser-grained scale of an entire system. In terms of spatial dimensions, the scale of these models may span from less than a nanometer (10-9 m) to micrometers (10-6 m).

Time scale is also an issue. When extremely large numbers of calculations are necessary, the sheer magnitude of processing cycles constrains the simulation time. Depending on the problem being addressed and techniques used, the time scale may range from femtoseconds (10-15 s) to milliseconds (10-3 s).

Considering these spatial and time scales, NREL's ongoing goal is to develop better techniques that are coded into better applications that run on better computing systems. Better techniques imply more accurate, powerful means to describe underlying physical and chemical processes. As an example, using improved force-field parameters may result in a more accurate description of cellulose structure.

To improve the software they use, researchers have to find ways to translate their improved techniques to computer code that is also better designed, more efficient, and more flexible analytically. To achieve this, NREL is leading a project, funded by U.S. Department of Energy (DOE) Biological and Environmental Research and DOE Advanced Scientific Computing Research (SciDAC), to build better molecular dynamics codes.

Finally, high-performance computing (HPC) provides the all-important increase in computational power that researchers require. NREL is constructing an Energy System Integration Facility that will house a new HPC data center. When operational in 2012, the supercomputer will have a processing speed of one-half petaFLOPS—0.5 quadrillion or 1015 floating-point operations per second—with plans to expand to one petaFLOPS. This world-class HPC capability will permit more sophisticated models to be developed and run.

The scientists ran thousands of models that used different enzyme concentrations and ratios to simulate representative environments that the scaffoldin and secreted enzymes encounter near the bacterium cell wall. They compared their results directly with the results of experimental studies of binding affinity. The modeling yielded a significant counterintuitive result: larger enzyme subunits are assembled more quickly than small subunits onto the scaffoldins. The enhanced binding of larger enzymes was shown to result from their slower diffusion and greater flexibility. Consequently, they are more likely to entangle with and bind to the scaffoldin. In contrast, the smaller enzymes move more rapidly and are less likely to bind to the scaffoldin.

Researchers will apply what they learned about binding affinities to control the ratios and perhaps the locations of enzymes on the scaffoldin that have synergistic functions. They will use what they understand about the binding affinity to engineer the enzymes' desired catalytic function. Their ultimate goal is to make artificial, or engineered, cellulosomes that are designed to work more effectively on specific pretreated biomass feedstocks. To this end, NREL has constructed a minicellulosome—a scaled down but fully functional version of larger natural multi-enzyme complexes—to advance the concept of cellulosome designs and create enzymes that more effectively catalyze cell wall deconstruction.

Shedding Light on the Science to Spur the Technology

NREL researchers, working with academic and industrial partners, have shown that combining experimental studies with computational modeling produces understanding that they can successfully apply to large, complex biological systems. By attacking industry-relevant problems with exactly this sort of rational, integrated approach, they expect their work to lead to engineered and optimized enzymes able to deconstruct specific types of biomass. To achieve this goal, researchers will have to:

  • Better understand how cellulases and other enzymes work that degrade cell wall polymers,
  • Predict the best place to make changes within the enzymes to improve them,
  • Actually make the specific changes,
  • And, finally, test the products or processes to confirm the beneficial outcome.

As NREL sheds light on the intricacies of the science underlying the conversion of renewable biomass resources to biofuels and chemicals, the laboratory is helping lay the solid foundation for developing the next-generation technologies that will enable the industry to produce large volumes of biofuels cost effectively.

—Don Gwinner

Related Links

Biomass Research

Computational Science

Top of page