Breaching the Biomass Problem
ORNL team provides supercomputer-aided insights for how to turn plants into fuel and materials
In a farewell nod to Titan, scheduled to be decommissioned in August 2019, we present a short series of features highlighting some of Titan’s impactful contributions to scientific research.
A visualization of a cellulose fiber (green) interacting with lignin (brown). Using supercomputers like the OLCF’s Cray XK7 Titan, ORNL researchers model biomass molecules at a level of detail that no experiment can rival. (Movie credit: Barmak Mostofian)
Humans have found uses for woody plants since time immemorial—tools, shelter, firewood. Today, scientists are working to add more major uses to the list: renewable, cost-competitive biofuel and value-added materials.
Turning trees and waste biomass into liquid fuels and other bioproducts has long been a goal of researchers and entrepreneurs, but for many years, the problem proved as unyielding as a stout oak in a storm. This resistance can be traced all the way to the molecules that make up the cell walls of plants, where energy-rich cellulose fibers intermix with hemicellulose and hardy lignin. Much biofuel research has focused on how to design “pretreatment” methods that efficiently separate these components so they can be turned into valuable products.
Using supercomputers, a team from the US Department of Energy’s (DOE’s) Oak Ridge National Laboratory (ORNL) has made several fundamental discoveries related to the challenges associated with breaking down biomass. In a recent paper published in Nature Reviews Chemistry, ORNL researchers Jeremy Smith and Loukas Petridis summarize key concepts derived from a decade of molecular-level simulations of biomass systems consisting of thousands to millions of atoms. In addition to determining the molecular structure and characteristics of biomass, the team also uncovered critical forces at work during biomass pretreatment, where heat, water, and solvents are applied in advance of enzymatic deconstruction.
“After 10 years, we’ve formed a coherent viewpoint on how things are working, and we’ve put it down in a succinct way so others can take advantage of the concepts,” said Smith, a University of Tennessee (UT)–ORNL Governor’s Chair and the director of the UT–ORNL Center for Molecular Biophysics. “This review focuses on one particularly important element that has emerged over the years as critical to turning plants into high-value chemicals and biofuels, and that is solvation.”
In recent years, new experimental solvents that efficiently break down the long chains of biomass fibers—called polymers—have generated excitement about reducing the production costs of cellulosic ethanol, a biofuel derived from woody plants. Why some solvents are more effective than others, however, is a question that requires a supercomputer to answer.
By carrying out simulations on the Titan supercomputer, a 27-petaflop Cray XK7 at the Oak Ridge Leadership Computing Facility (OLCF), a DOE Office of Science User Facility at ORNL, Smith’s team modeled solvent molecules interacting with biomass molecules at a level of detail that no experiment could rival.
Joining forces with an experimental group from the University of California, Riverside (UCR), Smith’s team demonstrated just how valuable high-resolution simulation can be. The team studied how interactions between the components of a two-part solvent play a central role in accelerating the breakdown of biomass into cellulose and lignin. The solvent, composed of water and the organic compound tetrahydrofuran (THF), is central to a UCR-developed biomass pretreatment process called Cosolvent Enhanced Lignocellulosic Fractionation, or CELF, that enables the enzymatic release of more than 90 percent of biomass lignin and sugars. But an explanation for how the cosolvent worked eluded researchers.
“I could show experimental data, but there wasn’t this molecular-level picture as to what’s going on in terms of first principles,” said UCR assistant research engineer Charles Cai, who led development of CELF pretreatment with support from the Sun Grant Institute and DOE’s BioEnergy Science Center.
The partnership with Smith’s team led to several publications combining molecular dynamics (MD) simulation results with experimental biomass reactions. The findings shed light on exactly how the cosolvent mixture works—namely, how it causes lignin to uncoil and open into extended shapes, and how THF and water separate on the surface of cellulose to coordinate cellulose’s water-driven breakdown.
“Just knowing there is a change of conformation, or measurable interaction, between the solvent molecules and substrate gives us a new lens to study how this system is working,” said Cai. “With MD simulation, we can actually look for that very special interaction between THF and water at different process conditions.”
A decade of simulation
Years of work realistically modeling and simulating multicomponent biomass on OLCF supercomputers paved the way for modeling solvent-biomass interactions.
In 2011, Smith’s team used the Jaguar supercomputer to resolve the structure of lignin, a molecule that reinforces plant structure and impedes degradation. In 2013, the team turned to Jaguar again and took a large step toward modeling multicomponent biological systems, simulating a 3.5-million atom system of cellulose and lignin and finding that less-ordered cellulose fibers bind less to lignin. The project allowed for an even more demanding simulation that leveraged the GPU-accelerated Titan supercomputer in 2015. That simulation—a 23.7-million atom system of pretreated biomass (cellulose and lignin) in the presence of enzymes—represented one of the largest biomolecular simulations ever, requiring millions of computing hours to complete.
“That was really pushing the edge in terms of system size that can be simulated,” Petridis said. “This is important because you need to have all these components present and talking to each other to have a realistic model of biomass.”
Each leap in complexity contributed precise molecular-level explanations that augmented fundamental understanding. For example, simulations showed that different plant polymers react differently to water. This attraction or aversion to water can have an outsized impact on the effectiveness of pretreatment and further explains the attractiveness of THF, which shields lignin from water and makes lignin easier to remove.
“Having only water or only THF would not work,” Smith said. “You need to mix pairs of solvents. It’s this mix that enables these cosolvent pretreatments to be efficient.”
With more than 10 years of biomass research under their belts, Smith and Petridis are preparing for the next 10 The ORNL team recently was awarded a new supercomputing allocation under DOE’s INCITE program granting it access to Titan and the OLCF’s newest leadership-class supercomputer, Summit.
Summit, a 200-petaflop system that features multiple GPUs on each node, will allow the team to simulate larger, more realistic biomass systems. It will also provide ample computational power to further investigate biomass-solvent interactions, with the goal of predicting how suitable a given solvent is for breaking down a given plant.
“The Holy Grail is to discover the perfect solvent that will completely deconstruct the plant cell walls and separate them in a form where they can be optimally used to produce biofuels and other high-value products,” Smith said. “Future simulations will help us discover the key qualities solvents need to accomplish this.”
Related publication: Petridis, Loukas, and Jeremy C. Smith. “Molecular-level driving forces in lignocellulosic biomass deconstruction for bioenergy.” Nature Reviews Chemistry (2018): 382–389. DOI: https://doi.org/10.1038/s41570-018-0050-6.
ORNL is managed by UT-Battelle, LLC for the Department of Energy’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit http://science.energy.gov/.