# ADVANCEMENTS IN BIOMASS FEEDSTOCK PREPROCESSING: CONVERSION READY FEEDSTOCKS

EDITED BY : J. Richard Hess, Allison E. Ray and Timothy G. Rials PUBLISHED IN : Frontiers in Energy Research and Frontiers in Bioengineering and Biotechnology

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88963-465-1 DOI 10.3389/978-2-88963-465-1

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# ADVANCEMENTS IN BIOMASS FEEDSTOCK PREPROCESSING: CONVERSION READY FEEDSTOCKS

Topic Editors:

J. Richard Hess, Idaho National Laboratory (DOE), United States Allison E. Ray, Idaho National Laboratory (DOE), United States Timothy G. Rials, The University of Tennessee, Knoxville, United States

Cover image: Kletr/Shutterstock.com

The success of lignocellulosic biofuels and biochemical industries depends upon an economic and reliable supply of quality biomass. However, research and development efforts have historically focused on the utilization of agriculturally-derived, cellulosic feedstocks without consideration of their low energy density, high variations in physical and chemical characteristics and potential supply risks in terms of availability and affordability. This Research Topic will explore strategies that enable supply chain improvements in biomass quality and consistency through blending, preprocessing, diversity and landscape design for development of conversion-ready, lignocellulosic feedstocks for production of biofuels and bio-products.

Biomass variability has proven a formidable challenge to the emerging biorefining industry, impeding continuous operation and reducing yields required for economical production of lignocellulosic biofuels at scale. Conventional supply systems lack the preprocessing capabilities necessary to ensure consistent biomass feedstocks with physical and chemical properties that are compatible with supply chain operations and conversion processes. Direct coupling of conventional feedstock supply systems with sophisticated conversion systems has reduced the operability of biorefining processes to less than 50%.

As the bioeconomy grows, the inherent variability of biomass resources cannot be managed by passive means alone. As such, there is a need to fully recognize the magnitude of biomass variability and uncertainty, as well as the cost of failing to design feedstock supply systems that can mitigate biomass variability and uncertainty. A paradigm shift is needed, from biorefinery designs using raw, single-resource biomass, to advanced feedstock supply systems that harness diverse biomass resources to enable supply chain resilience and development of conversion-ready feedstocks.

Blending and preprocessing (e.g., drying, sorting, sizing, fractionation, leaching, densification, etc.) can mitigate variable quality and performance in diverse resources when integrated with downstream conversion systems. Decoupling feedstock supply from biorefining provides an opportunity to manage supply risks and incorporate value-added upgrading to develop feedstocks with improved convertibility and/ or market fungibility. Conversion-ready feedstocks have undergone the required preprocessing to ensure compatibility with conversion and utilization prior to delivery at the biorefinery and represent lignocellulosic biomass with physical and chemical properties that are tailored to meet the requirements of industrially-relevant handling and conversion systems.

Citation: Hess, J. R., Ray, A. E., Rials, T. G., eds. (2020). Advancements in Biomass Feedstock Preprocessing: Conversion Ready Feedstocks. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-465-1

# Table of Contents

*06 Editorial: Advancements in Biomass Feedstock Preprocessing: Conversion Ready Feedstocks*

J. Richard Hess, Allison E. Ray and Timothy G. Rials


Amber Hoover, Rachel Emerson, Allison Ray, Daniel Stevens, Sabrina Morgan, Marnie Cortez, Robert Kallenbach, Matthew Sousek, Rodney Farris and Dayna Daubaras


Rebecca G. Ong, Somnath Shinde, Leonardo da Costa Sousa and Gregg R. Sanford


Philip Coffman, Nicole McCaffrey, James Gardner, Samarthya Bhagia, Rajeev Kumar, Charles E. Wyman and Deepti Tanjore

*86 Three Way Comparison of Hydrophilic Ionic Liquid, Hydrophobic Ionic Liquid, and Dilute Acid for the Pretreatment of Herbaceous and Woody Biomass*

C. Luke Williams, Chenlin Li, Hongqiang Hu, Jared C. Allen and Brad J. Thomas


Kunwar Paritosh, Monika Yadav, Sanjay Mathur, Venkatesh Balan, Wei Liao, Nidhi Pareek and Vivekanand Vivekanand

*135 Techno-Economic Analysis of Forest Residue Conversion to Sugar Using Three-Stage Milling as Pretreatment* Kristin L. Brandt, Johnway Gao, Jinwu Wang, Robert J. Wooley and Michael Wolcott *146 Blended Feedstocks for Thermochemical Conversion: Biomass Characterization and Bio-Oil Production From Switchgrass-Pine Residues Blends* Charles W. Edmunds, Eliezer A. Reyes Molina, Nicolas André, Choo Hamilton, Sunkyu Park, Oladiran Fasina, Sushil Adhikari, Stephen S. Kelley, Jaya S. Tumuluru, Timothy G. Rials and Nicole Labbé *162 Hot Water Extraction Improves the Characteristics of Willow and Sugar Maple Biomass With Different Amount of Bark* Obste Therasme, Timothy A. Volk, Antonio M. Cabrera, Mark H. Eisenbies and Thomas E. Amidon *175 Techno-Economic Assessment of a Chopped Feedstock Logistics Supply Chain for Corn Stover* Lynn M. Wendt, William A. Smith, Damon S. Hartley, Daniel S. Wendt, Jeffrey A. Ross, Danielle M. Sexton, John C. Lukas, Quang A. Nguyen, J. Austin Murphy and Kevin L. Kenney *189 Simulation Modeling for Reliable Biomass Supply Chain Design Under Operational Disruptions* Bhavna Sharma, Robin Clark, Michael R. Hilliard and Erin G. Webb *204 Effect of Non-Structural Organics and Inorganics Constituents of Switchgrass During Pyrolysis* Pyoungchung Kim, Choo Hamilton, Thomas Elder and Nicole Labbé *216 Wet Corn Stover Storage: Correlating Fiber Reactivity With Storage Acids Over a Wide Moisture Range* Dzidzor Essien, Megan N. Marshall, Tom L. Richard and Allison Ray *233 Wear Properties of Ash Minerals in Biomass* Jeffrey A. Lacey, John E. Aston and Vicki S. Thompson *239 A Multi-Criteria Decision Analysis Approach to Facility Siting in a Wood-Based Depot-and-Biorefinery Supply Chain Model* Natalie Martinkus, Greg Latta, Kristin Brandt and Michael Wolcott *255 High Throughput Screening Technologies in Biomass Characterization* Stephen R. Decker, Anne E. Harman-Ware, Renee M. Happs, Edward J. Wolfrum, Gerald A. Tuskan, David Kainer, Gbekeloluwa B. Oguntimein, Miguel Rodriguez, Deborah Weighill, Piet Jones and Daniel Jacobson *273 Bioconversion of Pelletized Big Bluestem, Switchgrass, and Low-Diversity Grass Mixtures Into Sugars and Bioethanol* Bruce S. Dien, Robert B. Mitchell, Michael J. Bowman, Virginia L. Jin, Joshua Quarterman, Marty R. Schmer, Vijay Singh and Patricia J. Slininger *288 Ensiled Wet Storage Accelerates Pretreatment for Bioconversion of Corn Stover* Dzidzor Essien and Tom L. Richard *305 Integration of Pretreatment With Simultaneous Counter-Current Extraction of Energy Sorghum for High-Titer Mixed Sugar Production* Daniel L. Williams, Rebecca G. Ong, John E. Mullet and David B. Hodge

# Editorial: Advancements in Biomass Feedstock Preprocessing: Conversion Ready Feedstocks

J. Richard Hess <sup>1</sup> \*, Allison E. Ray <sup>1</sup> \* and Timothy G. Rials <sup>2</sup> \*

1 Idaho National Laboratory, Idaho Falls, ID, United States, <sup>2</sup> The Center for Renewable Carbon, The University of Tennessee, Knoxville, Knoxville, TN, United States

Keywords: biomass variability, conversion-ready feedstocks, preprocessing, blending, fractionation, sorting, resource diversity, integrated biorefineries

**Editorial on the Research Topic**

**Advancements in Biomass Feedstock Preprocessing: Conversion Ready Feedstocks**

# INTRODUCTION

The success of lignocellulosic biofuels and biochemical industries depends on an economic and reliable supply of biomass that meets conversion quality standards. However, research and development have historically focused on the utilization of lignocellulosic biomass resources without consideration of their high variations in physical and chemical characteristics and potential supply risks in terms of availability and affordability (Ray et al., 2017). This Research Topic explored strategies aimed at enabling supply-chain improvements in biomass quality and consistency through preprocessing operations like blending, sorting, leaching, drying, and other quality-management approaches for development of "conversion-ready" lignocellulosic feedstocks for production of biofuels and bio-products. A conversion-ready feedstock refers to an industrialscale feedstock resource with chemical and physical properties that meet the design specifications

#### Edited and reviewed by:

Uwe Schröder, Technische Universitat Braunschweig, Germany

#### \*Correspondence:

J. Richard Hess JRichard.Hess@inl.gov Allison E. Ray allison.ray@inl.gov Timothy G. Rials trials@utk.edu

#### Specialty section:

This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research

Received: 01 November 2019 Accepted: 15 November 2019 Published: 03 December 2019

#### Citation:

Hess JR, Ray AE and Rials TG (2019) Editorial: Advancements in Biomass Feedstock Preprocessing: Conversion Ready Feedstocks. Front. Energy Res. 7:140. doi: 10.3389/fenrg.2019.00140 of biorefinery handling and conversion systems. Variability has proven a formidable challenge to integrated biorefineries, impeding continuous operation and reducing product yields required for economical production of lignocellulosic biofuels and chemicals at scale (U. S. Department of Energy, 2016). Simple or minimalistic biomass supply systems (herein referred to as "conventional") that deliver field-run biomass to the biorefinery, while minimizing operating costs, do not ensure that biomass resources consistently meet feedstock quality requirements. Conventional supply systems lack the preprocessing capabilities necessary to produce feedstocks with consistent physical and chemical properties that are compatible with conversion processes (Lamers et al., 2015). Further, the utility of conventional systems has been limited to high-yield regions like the U.S. Corn Belt (Argo et al.,

2013). Direct coupling of conventional biomass supply systems with sophisticated conversion processes is further complicated by the unique challenges of bulk-solids behavior in particulate systems (Bell, 2005) and the cohesive nature of compressible, elastic biomass solids (Hernandez et al., 2017; Xia et al., 2019). Recent reports indicate the operability of biorefining processes and process models remain at less than 50% due to the variable physicochemical properties of biomass (U. S. Department of Energy, 2016).

As the bioeconomy grows, the variability inherent to biomass cannot be managed solely by passive supply systems (Kenney et al., 2013; Searcy et al., 2015). Recognition and understanding of the magnitude of biomass variability and uncertainty (Williams et al., 2016) must be balanced with the cost of failing to design feedstock supply systems that can manage variations. Targeted preprocessing strategies aim to develop stable, cost-effective and quality-controlled biomass supply

**6**

systems (Li et al., 2016). For instance, blending and densification approaches (Tumuluru et al., 2011; Ray et al., 2017; Ou et al., 2018; Narani et al., 2019) have shown promise for diversifying biomass resources to enable supply-chain resilience for the development of consistent, high-quality feedstocks and move toward advanced feedstock-supply systems.

The significance of feedstock quality on overall processing efficiency and expense has been recognized for some time; however, affordable solutions to the range of challenges presented by the biomass-to-feedstock operation have been slow to emerge. Recent work in government and academic laboratories, as well as industrial facilities has identified more problematic areas and provided direction for innovation to mitigate many of those longrunning challenges. This special issue is a collection of 23 articles that report on strategies that enable supply-chain improvements in biomass quality and consistency to produce conversion-ready feedstocks for biorefining.

# INTEGRATED PREPROCESSING

Integrated preprocessing techniques represent the majority of the collection. Active biomass-preprocessing controls, like drying, sorting, sizing, fractionating, leaching, densifying, etc., integrated into biomass supply systems can reduce complications in downstream conversion systems. The collection discusses six preprocessing techniques for generation of conversionready feedstocks. This issue highlights research that employs biomass preprocessing through fractionation, leaching, sorting, blending and sizing processes to enhance material quality for downstream conversion. Williams et al. and Kim et al. both address how the separation of biomass components can optimize conversion. Therasme et al. explore leaching as a preprocessing step to improve biomass quality and create co-products, and Williams et al. report the benefits of both air classification and separation of biomass materials for downstream conversion. The second of the techniques addressed is the blending of feedstocks. Edmunds et al. described mixing different biomass types to mitigate rheological—and compositional variability challenges seen in individual biomass resources. The third technique discussed is preprocessing and pretreatment of biomass materials for conversion. Brandt et al. address mechanical biomass preprocessing methods, and Williams et al. report on chemical methods to prepare the biomass materials for conversion. Biomass densification, the fourth technique, is addressed by Dien et al., who report on preprocessing and pelletizing techniques that reduce supply-system losses. Storage impacts on biomass quality and downstream conversion are addressed in work from Essien et al., Essien and Richard, Wendt et al., and Wendt et al.; all four papers discuss the effectiveness of wet storage to preserve and improve biomass materials for conversion. High-throughput and rapid methods for assessing biomass attributes and their behavior are also explored and represent tools that advance understanding and development of conversion-ready feedstocks (Coffman et al.; Decker et al.).

#### SUSTAINABLE AND EFFICIENT FEEDSTOCK SUPPLY-TO-CONVERSION PRACTICES

As solutions to biomass variability issues are introduced, environmental and economic impacts become an issue of concern. The collection discusses research on creating moresustainable biorefining facilities and producing conversionready feedstocks from recyclables. Martinkus et al. address the cost advantage of assessing existing industrial facilities for repurposing to biorefinery applications. Other papers include discussions of how to leverage recycled materials as biomass feedstocks (Paritosh et al.; Xu et al.) and the viability of recycling materials already used in preprocessing (Chen et al.).

#### IMPACT OF WEATHER AND RELATED EXTERNAL FACTORS

Environmental and other external factors that impact biomass variability and conversion processes are also considered in the collection. Addressing external issues—including equipment wear from soil-accumulated inorganics in biomass (Lacey et al.), challenges in the supply-chain structure (Sharma et al.), harvesting timelines and appropriate equipment selection (Daniel et al.), and weather factors (Hoover et al.; Ong et al.) offer insights to process changes that could be implemented to improve the viability of biomass feedstock supply chains.

#### CONCLUSION

A paradigm shift is needed from biorefinery designs using raw biomass resources to advanced feedstock-supply systems that deliver commercial-scale biomass feedstocks that are conversion ready (Searcy et al., 2015). There is a need to fully recognize the magnitude of variability and uncertainty in biomass resources, as well as the cost of failing to design feedstock-supply systems that can mitigate that variability and uncertainty. As the recognition and need for conversion-ready biomass feedstocks increase, the research presented in this collection will prove increasingly beneficial. Considering conversion-ready feedstocksupply systems for biorefining allows greater opportunities to manage supply risks, including feedstock system designs that may incorporate the biomass preprocessing techniques presented in this collection. Such feedstock supply systems will provide value-added upgrading of the biomass to increase its convertibility and market fungibility. Taken together, these papers present relevant research that will enable stakeholders to better navigate the range of challenges related to consistently and economically producing conversion-ready feedstocks derived from lignocellulosic biomass on a commercial scale.

# AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

#### FUNDING

This research was supported by the U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Bioenergy Technologies Office, under DOE Idaho Operations

#### REFERENCES


Office Contract DE-AC07- 05ID14517. Dr. Rials would like to acknowledge support for this project provided by the Tennessee Agricultural Experiment Station and AgResearch. The authors wish to acknowledge Leslie Ovard, Amanda Sant, and Jessica McCord for editorial support.


**Disclosure:** The views and opinions of the authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.

**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Hess, Ray and Rials. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Compatibility of High-Moisture Storage for Biochemical Conversion of Corn Stover: Storage Performance at Laboratory and Field Scales

*Lynn M. Wendt1 \*, J. Austin Murphy1 , William A. Smith1 , Thomas Robb2 , David W. Reed1 , Allison E. Ray1 , Ling Liang3 , Qian He3 , Ning Sun3 , Amber N. Hoover1 and Quang A. Nguyen1*

*<sup>1</sup> Idaho National Laboratory, Idaho Falls, ID, United States, 2 Independent Researcher, Olathe, KS, United States, 3Advanced Biofuels Process Development Unit, Lawrence Berkeley National Laboratory, Emeryville, CA, United States*

#### *Edited by:*

*Abdul-Sattar Nizami, King Abdulaziz University, Saudi Arabia*

#### *Reviewed by:*

*Mohammad Rehan, King Abdulaziz University, Saudi Arabia Yong Xu, Nanjing Forestry University, China*

> *\*Correspondence: Lynn M. Wendt lynn.wendt@inl.gov*

#### *Specialty section:*

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Bioengineering and Biotechnology*

*Received: 13 December 2017 Accepted: 09 March 2018 Published: 26 March 2018*

#### *Citation:*

*Wendt LM, Murphy JA, Smith WA, Robb T, Reed DW, Ray AE, Liang L, He Q, Sun N, Hoover AN and Nguyen QA (2018) Compatibility of High-Moisture Storage for Biochemical Conversion of Corn Stover: Storage Performance at Laboratory and Field Scales. Front. Bioeng. Biotechnol. 6:30. doi: 10.3389/fbioe.2018.00030*

Wet anaerobic storage of corn stover can provide a year-round supply of feedstock to biorefineries meanwhile serving an active management approach to reduce the risks associated with fire loss and microbial degradation. Wet logistics systems employ particle size reduction early in the supply chain through field-chopping which removes the dependency on drying corn stover prior to baling, expands the harvest window, and diminishes the biorefinery size reduction requirements. Over two harvest years, in-field forage chopping was capable of reducing over 60% of the corn stover to a particle size of 6 mm or less. Aerobic and anaerobic storage methods were evaluated for wet corn stover in 100 L laboratory reactors. Of the methods evaluated, traditional ensiling resulted in <6% total solid dry matter loss (DML), about five times less than the aerobic storage process and slightly less than half that of the anaerobic modified-Ritter pile method. To further demonstrate the effectiveness of the anaerobic storage, a field demonstration was completed with 272 dry tonnes of corn stover; DML averaged <5% after 6 months. Assessment of sugar release as a result of dilute acid or dilute alkaline pretreatment and subsequent enzymatic hydrolysis suggested that when anaerobic conditions were maintained in storage, sugar release was either similar to or greater than as-harvested material depending on the pretreatment chemistry used. This study demonstrates that wet logistics systems offer practical benefits for commercial corn stover supply, including particle size reduction during harvest, stability in storage, and compatibility with biochemical conversion of carbohydrates for biofuel production. Evaluation of the operational efficiencies and costs is suggested to quantify the potential benefits of a fully-wet biomass supply system to a commercial biorefinery.

Keywords: Corn stover, biomass storage, ensiling, Ritter pile, field storage, feedstock reactivity

# INTRODUCTION

Biomass is recognized as a renewable and sustainable energy resource, and presently 130 million tonnes of agricultural residues are available in the U.S. with the potential to supply up to 180 million dry tonnes of biomass for conversion to bioenergy by 2040 (Langholtz et al., 2016). In addition, the use of agricultural residues and other waste products to produce energy is being explored throughout the world as a means to provide income to rural communities (Dale et al., 2016; Nizami et al., 2017). However, because agricultural residue harvest is seasonal, long-term storage methods are needed to continuously supply feedstock throughout the year at a profitable refinery. Multiple strategies exist for designing the most cost-effective herbaceous feedstock logistics supply chains (Hess et al., 2007; Kumar and Singh, 2017; Shah et al., 2017; Zandi Atashbar et al., 2017), yet common to all of these is the reliance on dry bales for long-term storage. Although dry bales may have low dry matter loss (DML) in storage (<10%) when moisture content is below 20% wet basis (wb), they represent a significant source of combustible material and a fire risk (Rentizelas et al., 2009). Storage-related fires can lead to the losses of thousands of tons of biomass, may contribute to local air and water quality concerns, and in some cases may prompt legal action by adjacent property owners (Dale, 2017). In addition, microbial degradation of biomass, leading to DML, is also a significant risk if the bales are processed under conditions too wet for proper dry storage (over 25% moisture, wb) or if the bales absorb excessive moisture during storage (Kenney et al., 2013; Smith et al., 2013). DMLs that occur during outdoor storage of bales have been shown to be a significant contributor to delivered feedstock costs (Sahoo and Mani, 2017).

Corn stover is one of the primary agricultural residues available for producing bioenergy, but there are challenges associated with dry storage as a result of available harvest techniques. For example, since the final moisture content of the stover is fixed by the timing of the grain harvest, it is often higher than optimal for dry storage, especially in northern latitudes (Shinners and Binversie, 2007). In geographical regions where double-cropping is practiced, stover is removed soon after harvest to facilitate winter crop planting, which limits the time available for drying (Heggenstaller et al., 2008). In a multi-pass corn stover harvest, the stover is left in the field to dry to a moisture content of about 15–20% (wb) prior to baling (Darr and Shah, 2012); however, sufficient drying time is not always possible due to regional seasonal rain (Shah et al., 2011; Smith et al., 2013; Oyedeji et al., 2017). Oyedeji et al. (2017) estimated that less than 40% of nationally available corn stover meets the 20% moisture threshold at the time of baling. Likewise, single-pass harvesting is not compatible with field drying because it is baled at the same time as the grain is harvested (Shinners et al., 2011), although this approach has the advantage of reducing soil contamination by preventing stover contact with the ground.

Wet anaerobic storage (i.e., ensiling) is an attractive alternative to dry bale storage and can protect the biomass from biologically-mediated DML while reducing the risk of catastrophic fire loss. Wet storage through ensiling commonly utilizes silage bags, bunkers, or drive-over piles for oxygen limitation availability, followed by biological anaerobic organic acid production through fermentation, which lowers the feedstock pH and effectively preserves biomass during storage (McDonald et al., 1991). Significant attention has been given to ensiling for preservation of forage for livestock feed with specific focus on the requirements that initiate oxygen exclusion, additives including mineral and organic acids to exclude unwanted microbial activity, low-cost sugars to encourage growth of fermentative organisms, and microbial inoculants that produce well-preserved silage (Yitbarek and Tamir, 2014). Ensiling of winter crops has recently received attention for its potential to provide a year-round source of biomass for on-farm anaerobic digesters to provide renewable electricity (Dale et al., 2016).

The preservation of corn stover through ensiling as a bioenergy feedstock has been evaluated in the laboratory (Richard et al., 2001; Chen et al., 2012; Liu et al., 2013), although field-scale trials have been limited to livestock forage applications (Shinners et al., 2007, 2011). However, corn stover is typically harvested at lower initial moisture content compared to feedstock dedicated for forage and, thus, contains increased interstitial oxygen that must be removed or biologically consumed in order to establish fermentation conditions. Despite this challenge, low-moisture ensiling at 40–50% moisture (wb) has successfully preserved dry matter in corn stover for >6 months (Shinners et al., 2011) and remains a potential solution for stabilizing corn stover considered too wet for stable dry bale storage.

The "Ritter" pile was designed for the preservation of wet fiber and digestion of the non-fiber component into pulp and/ or paper (Ritter, 1960). The Ritter pile, which has been applied to bagasse (Morgan et al., 1974; Kruger et al., 1981) and corn stover (Hettenhaus et al., 2000; Atchison and Hettenhaus, 2003), is constructed and compacted using a slurried biomass; water is recirculated continuously in the traditional Ritter pile and only during pile formation in the modified Ritter approach (Atchison and Hettenhaus, 2003). The pile remains saturated at 70–80% moisture (wb) for the duration of storage except when pretreatment by acid impregnation of the biomass is necessary (Humbird et al., 2011), whereupon a dewatering step is included. Fortunately, the dewatering stream contains nutrients that could be applied to cropland as irrigation water to offset fertilizer application (Hettenhaus et al., 2000). One disadvantage to the Ritter method is that it is very water intensive and requires extensive infrastructure for the slurrying and recirculating process, increasing the cost and reducing environmental sustainability.

Two anaerobic wet storage methods, ensiling and the modified-Ritter storage system, were evaluated in this study because they have tremendous potential to secure corn stover from losses due to fire or microbial degradation, which are both significant vulnerabilities of dry storage systems at industrially relevant scales. These storage studies were necessary to evaluate the applicability of ensiled storage for biochemical conversion pathways, where cellulose and hemicellulose in biomass are depolymerized to monomeric carbohydrates through a combination of pretreatment and enzymatic saccharification followed by fermentation to fuels and/or chemicals. Unfortunately, the majority of research on ensiling assesses carbohydrate composition in storage performance using forage industry-specific terms, which are not directly correlated to fuel yield at a biorefinery (Wolfrum et al., 2009). To better understand this response, laboratory-based storage studies were conducted to characterize the performance of each storage method as related to as-harvested material, followed by dilute acid or alkali pretreatment and subsequent enzymatic saccharification to sugar monomers. The laboratory-based storage studies were also performed under high-moisture aerobic conditions to determine the potential impact of aerobic degradation in storage. Finally, a field-scale study was conducted to demonstrate the effectiveness of the wet logistic system using ensiling.

# MATERIALS AND METHODS

# Storage Reactor and Field Experimentation

Laboratory-based anaerobic storage experiments were conducted to simulate ensiling in a drive-over pile and a modified-Ritter pile. For these experiments, Pioneer P1151 corn stover was harvested in September of 2014 in Stevens County, KS following the grain harvest using a forage harvester, and then *Lactobacillus plantarum* was applied to the stover prior to collection in a forage wagon. Stover was immediately compacted in 55 gal drums at an initial moisture content of 47% (wb), sent to Idaho Falls, ID in a refrigerated semi-trailer, and then stored at −20°C at a local food storage facility until use. For ensiling, duplicate 100 L reactors were each loaded with 5.9 ± 0.5 kg corn stover dry basis (db) at an initial moisture content of 47.3 ± 1.5% (wb; *n*= 4). To simulate the modified-Ritter pile formation, slurrying was performed by washing 22.5 kg stover (db) with the equivalent of 6.2 kg water per kg dry stover continuously for 1 h prior to loading duplicate reactors with 6.4 ± 0.5 kg stover (db) at 76.0 ± 0.6% (wb; *n* = 4) moisture. For both anaerobic storage methods, a pressure of 3.9 kPa was applied to the biomass at five separate 300 s intervals during loading in order to maintain homogeneity between reactors and a final packing density of 73.7 and 76.9 kg/m3 (db) for the ensiling and modified-Ritter reactors, respectively. Storage reactors were modified from a previous design (Wendt et al., 2014; Bonner et al., 2015) to maintain anaerobic conditions and equipped with an aluminized gas sampling bag to capture and quantify gas formation during the 110-day storage period. Moisture content and gravimetric DML were calculated based on randomly selected samples, as described previously (Wendt et al., 2014).

An outdoor storage pile experiment was performed concurrently with aerobic laboratory experiments using corn stover, Pioneer P1151AMX, which was harvested in Stevens County, KS in August 2015. The corn stover was harvested at an initial moisture content of 40.5 ± 1.6% (wb; *n* = 4) using a windrowing flailshredder (Hiniker model 5630HL, Mankato, MN, USA) followed by a forage chopper (John Deere model 7780, Moline, IL, USA) with a windrow pickup head (John Deere model 640°C, Moline, IL, USA). The stover was blown into a self-unloading silage truck wagon (Taller Fehr 11.5 m silage trailer, Cuauhtémoc Chihuahua, Mexico) and transported to a nearby storage site where a 272 t (db) pile was formed. Pile dimensions were 23 m wide, 3.5 m tall, and 53 m long; the pile was shaped and compacted using a tractor (John Deere model 9420, Moline, IL, USA). A longitudinal section of the pile was selected for real-time monitoring based on solar irradiation and spatial location, with specific locations shown in **Figure 1**. Gas sampling ports, temperature probes (Campbell Scientific, Logan, UT, USA), and "DML bags" were placed in 10 locations within this section of the pile 1 day after pile formation to monitor storage performance. DML bags consisted of 10 cm diameter Drain-Sleeve® filter fabric (Clariff Corporation, Midland, NC, USA) cut into 45 cm pieces, filled with corn stover obtained from the pile and secured with zip-ties at the ends. Hourly temperature measurements were recorded for the external air at each of the 10 zones and at an additional surface location on top of the pile. The pile was left initially undisturbed for about 3 weeks prior to covering with plastic wrap, which was later added to limit additional air infiltration. Interstitial gas evolution was sampled at 7 points during the storage period, and gas was transferred into aluminized bags following purging each sampling line with five volumes. After 6 months of storage, the pile was sampled for DML, pH, organic acids, moisture content, composition, and pretreatment and enzymatic hydrolysis. DML bags and moisture samples from each of the 10 zones were collected as well as a 200 kg sample from a section of the pile approximately 20 m east of the sampling zone, and this sample is available for public use through the INL Bioenergy Feedstock Library at https://bioenergylibrary.inl.gov.

The 2015 corn stover was shipped overnight to INL and stored aerobically for 111 days in laboratory reactors to simulate the outer regions of an uncovered pile. For the aerobic laboratory storage experiment, two air flow rates were evaluated in duplicate 100 L reactors as described previously (Wendt et al., 2014). Filtered room air was supplied to the reactors at a rate of 0.5 or 1.0 L/min, corresponding to complete air exchange every 200 or 100 min, respectively. For a 0.5 L/min airflow rate, 8.46 and 7.93 kg corn

stover (db) were added with the compaction described above to obtain a packing density of 101.41 and 95.10 kg/m3 (db), respectively. For an airflow of 1 L/min, 8.35 and 7.37 kg corn stover (db) additions resulted in 100.10 and 88.37 kg/m3 packing density, respectively.

#### Chemical Compositional Analysis

To test for microbially generated greenhouse gases or air pollutants, the following were analyzed: CO2, CH4, H2, N2O, CO, NOx. Gas exiting the storage reactors was analyzed using an automated gas chromatograph (MicroGC 3000, Agilent, Santa Clara, CA, USA) as described previously (Wendt et al., 2017). Nitrous gases were measured using Nitrous Gases 100/c and Nitrogen Dioxide 0.5/c Draeger tubes (Draeger Safety Inc., Pittsburgh, PA, USA). Organic acids were extracted from samples prior to and after storage and measured *via* high-performance liquid chromatography (HPLC) according to a method described previously (Wendt et al., 2017).

Chemical compositional analysis was determined for duplicate corn stover composite samples (prior to and after storage) according to standard biomass procedures developed by the National Renewable Energy Laboratory (NREL) (National Renewable Energy Laboratory, 2013), similar to what is reported elsewhere (Wolfrum et al., 2017). Extractives from water and ethanol were produced using an ASE 350 (Dionex, Sunnyvale, CA, USA). Following extraction, the biomass was subjected to a two-stage acid hydrolysis. The liquor from the extraction and subsequent acid hydrolysis was analyzed using HPLC with a refractive index detector (Agilent, Santa Clara, CA, USA). Monomeric sugars were analyzed with an Aminex HPX 87 P column (Bio-Rad, 300 × 7.8 mm, Hercules, CA, USA). Organic acids were analyzed with an Aminex HPX 87 H ion exclusion column (Bio-Rad, 300 mm × 7.8 mm, Hercules, CA, USA). Results for soluble and structural sugars from duplicate analysis runs were normalized to 100% recovery for a NIST Reference 8491 Bagasse sample that was analyzed alongside the samples.

To determine the yields of sugars available in biomass, sugar release was measured following sample pretreatment with dilute acid/dilute alkaline and subsequent enzymatic hydrolysis. Samples analyzed include as-harvested stover from 2014 and 2015, the ensiled and 1 L/min samples, and the field composite. Soluble and structural sugars were considered in the analysis of sugar yields. Dilute acid pretreatment was performed with a Dionex ASE 350 Accelerated Solvent Extractor (Dionex Corporation, Sunnyvale, CA, USA) at 10% (w/w) solids loading by adding 30 mL of 1% sulfuric acid (w/w) in 66-mL Dionium cells, using a method similar to that described by Wolfrum et al. (2013) with a 360 s ramp in temperature to 160°C followed by a 420 s incubation (severity factor = 2.61). The pretreatment liquor and rinse liquid were measured for monomeric and polymeric sugars as well as degradation products, as described above.

Dilute alkaline pretreatment was carried out in 25 mL Incoloy® tube reactors (Alloy Metal and Tubes, Houston, TX, USA) with a fluidized sand bath (Omega FSB-4, Stamford, CT, USA) to supply heat. Dry biomass (2 g) was soaked in a prepared sodium hydroxide solution overnight before being loaded into the tube reactors. Total weight per reaction was 20 g with a 10% (w/w) biomass loading and 0.05 g sodium hydroxide per g biomass. After 140°C pretreatment for 1,620 s (severity factor = 2.61), reactions were quenched by immersing tubes into an ice bath. Pretreatment liquor was collected after centrifugation at 3,000 × rpm (1,811 × *g*) for 300 s. Washing of solids was performed using 100 mM citrate buffer (pH 4.8) followed by centrifugation until the solids equilibrated at a pH of 4.8.

Enzymatic hydrolysis of dilute acid-pretreated biomass was performed using 1 g (db) of washed solids at 10% (w/w) solids loading and 50 mM citrate buffer, pH 4.8 in triplicate, similar to the methods described by Wolfrum et al. (2017). However, enzymatic hydrolysis of dilute alkaline-pretreated biomass was performed using the entire washed sample at a 5% (w/w) solids loading. Cellic® Ctec2 and Cellic® Htec2 enzyme complexes (Novozymes®, Franklinton, NC, USA) were added at a loading rate at 40 mg protein/g and 4 mg protein/g dry biomass, respectively. Sodium citrate buffer was supplemented with NaN3 to a final concentration of 0.02% in the biomass slurry to prevent microbial contamination. Flasks were incubated at 50°C and 200 rpm for 5 days. Carbohydrates released in dilute acid pretreatment were measured using HPLC with the HPX-87P column as described above, and fermentation inhibitors, including acetate, furfural, 5-hydroxymethylfurfural, and levulinic acid were measured using HPLC with the HPX-87H column, as described above. Carbohydrates released in dilute alkaline pretreatment and enzymatic hydrolysis were measured using HPLC and a refractive index detector (Thermo Fisher Scientific, Ultimate 3000, Waltham, MA, USA) equipped with the HPX-87H column described above. At least two parallel samples were used in all analytical determinations, and data were presented as the mean of replicates.

# Particle Size Distribution (PSD)

Particle size distribution was determined for four representative samples of dry corn stover using a Ro-Tap RX-29 (W.S. Tyler, Mentor, OH, USA) with sieve sizes of 6.35, 3.34, 0.84, 0.42, 0.25, 0.177, and 0.149 mm. The weight percent of corn stover on each sieve was calculated after a 600 s vibration operating time (ASAE Standards, 1992; 47th ed.).

# Statistical Analysis

Averages and one SD are presented with *n* = 2 unless otherwise noted. For chemical composition data of samples (two duplicates included per sample), an unequal variance *t*-test was performed in Microsoft Excel to determine significant differences between each stored sample and the corresponding as-harvested source biomass. For sugar release experiments (*n* = 3), single-factor oneway analysis of variance (ANOVA) was performed in SigmaPlot (version 13.0) to identify significant differences, and Tukey's honest significant difference test was performed for multiplelevel comparison of statistical equivalency if the ANOVA was significant at *p* < 0.05.

# RESULTS

To determine the impact of wet storage on biomass, corn stover was treated with two laboratory anaerobic wet storage methods (traditional ensiling and modified-Ritter storage) as well as two laboratory aerobic wet storage scenarios (0.5 and 1.0 L/min supplied airflow) that simulate wet aerobic field storage. Parameters including gas production, fermentation products, fate or loss of the total dry matter, chemical composition of the stored solids, and availability of sugars for fermentation were analyzed for each condition. In addition, a 272 t (db) corn stover drive-over pile was constructed to evaluate ensiling in the field.

#### Particle Size Reduction in the Field

The PSD of the forage-chopped corn stover used in the storage experiments was measured for the two harvest years (**Figure 2**). In the 2014 harvest, direct cutting and chopping (one-pass harvest)

of the standing corn stover left <38% of the stover retained on the 6 mm screen, with the majority of the stover retained on the 3.4 mm (30%) and 0.84 mm (22%) screens and 4.1% below the 0.25 mm screen was considered fines. A slightly different distribution was obtained from the harvesting operation in 2015, where the flail chopping windrower performed initial size reduction followed by a second size reduction in the forage chopping operation (two-pass harvest). A narrower PSD was produced with this harvest approach; 33% of the stover was retained on the 6 mm screen whereas 45 and 13% was retained on the 3.4 and 0.84 mm screens, respectively, and 6.5% of material considered fines. The geometric mean particle size was 4.73 ± 8.19 mm in the 2014 harvest compared to 4.81 ± 6.62 mm in the 2015 harvest.

#### Storage Biomass Gas production, Temperature, and Mass loss

CO2 released from biomass during storage can be linked to microbial degradation of sugars due to aerobic microbial respiration or to a lesser extent as a result of anaerobic fermentation (McGechan, 1990). After 110 days in storage, the modified-Ritter storage produced about three times more CO2 relative to ensiling (**Table 1**) that coincided with higher DML (9.89% modified-Ritter vs 5.75% silage; **Table 2**). Anoxic conditions were reached in the modified-Ritter method within 7 days, while 1% oxygen was still present after 80 days in samples following the traditional ensiling approach for storage. Permanent gases other than N2, O2, and CO2 (CO, H2, NOX, NO2, N2O, and CH4) produced throughout the duration of the experiments were measured for each of the anaerobic conditions (**Table 1**). CH4 and N2O were not detected in either the ensiled or modified-Ritter reactor storage method. NO2 was present in the modified-Ritter reactors at <0.1 ppm

TABLE 1 | Gas production in laboratory reactors after 110 days (anaerobic) or 111 days (aerobic) in storage.


*Values represent means with one SD in parentheses.*

*BD, below detection.*

TABLE 2 | Composition analysis including percent dry matter loss (DML) and organic acid production for corn stover after anaerobic storage by ensiling and the modified-Ritter method as well as field and aerobic storage.


*Values represent means with one standard deviation in parentheses. Not shown are organic acids (valeric, 2-methylbutyric, isobutyric, and isovaleric acid) below detection levels in the samples.*

*DML, dry matter loss; BD, below detection.*

and was not detected in the ensiling reactors. In addition, the modified-Ritter reactors produced threefold to fivefold more CO and H2 than the traditional ensiling material although both gases were in the low ppm range.

CO2 production ranged from 161 to 366 g/kg in the aerobic reactors and DML varied from 32.10 ± 1.54% to 29.52 ± 5.33% for the 0.5 and 1 L/min conditions, respectively. The other permanent gases listed in **Table 1** were not detected in the aerobically stored corn stover in the laboratory reactors due to the dilution of these gases with air flow through the reactors. Significant selfheating occurred in the aerobic reactors through the duration of storage (**Figure 3**) as a result of microbial respiration of the available carbohydrates. Temperatures peaked at 60°C in 1 L/min reactors within the first 10 days of storage, whereas the 0.5 L/min reactors reached only 57°C. Despite maximum temperatures being only 3°C different, CO2 production in the 1 L/min reactors over the first 10 days was over twice that of the 0.5 L/min condition (data not shown). For the duration of the experiments, the aerobic laboratory reactors generally remained at >50°C, whereas the anaerobic laboratory reactors never increased above ambient room temperature. The replicate in the 1 L/min condition with lower packing density exhibited a 5°C drop in temperature after 20 days in storage and cooled sooner than the replicate with higher packing density, corresponding to a lower dry matter loss rate (24.2 vs 34.9%) and contributing to the high SD in DML for the 1 L/min condition.

For the field experiment, the pile exhibited self-heating during the initial 3 weeks (**Figure 4**) with temperatures reaching 60°C and decreasing to 20–30°C upon covering the pile with a tarp. Covering the pile was accompanied by a reduction in oxygen levels to <3% and an increase in CO2 to levels of 18–22% (data not shown); variability was assumed to result from the inhomogeneous nature of the silage pile relative to the laboratory reactors. For the field samples, up to 400 ppm CO was detected initially, and CH4 (~200–300 ppm) was detected by the end of the experiment (data not shown). The field-ensiled material had a DML that was comparable to the laboratory anaerobic wet storage samples after 6 months of storage (4.36 ± 3.25%, *n* = 10)

despite the self-heating period that occurred during the first weeks of storage.

#### Fermentation Organic Acids and pH

Organic acid concentrations and pH were measured in samples collected at the end of the storage period to characterize fermentation products (**Table 2**). The ensiled corn stover had a pH of 4.62, lower than that observed for the other storage methods tested (modified-Ritter, pH = 5.18; aerobic, pH = 5.91 and 6.46, 0.5 L/ min and 1 L/min, respectively) and contained the highest level of lactic acid (3.10%). Samples from the field-stored silage pile had a similar composite pH (pH = 4.9, *n* = 10) to the ensiled material and lactic acid concentrations near 0.5%, however butyric acid (0.36%) was also present in many samples. Acetic acid (5.41%) and butyric acid (1.45%) were the primary fermentation products of the stover stored in the modified-Ritter system. Propionic (2.10%) and succinic acid (1.73%) were produced in the 0.5 L/ min condition, whereas these levels were reduced by nearly half in the 1 L/min scenario.

#### Ash, Structural, and Soluble Sugars

Chemical analyses were performed on samples from each storage experiment alongside the respective starting material (as-harvested) to determine the effect of storage on composition (**Tables 3** and **4**). Total ash is defined as the combination of structural, or physiological, ash and the extractable inorganics fraction that typically contains exogenous soil entrained during harvest and collection operations. Total ash was 19.4% in 2015 samples that made contact with the ground during the two-step harvest method as compared to 9.0% in the direct-harvested 2014 samples, resulting in a proportional decrease in compositional components (i.e. carbohydrates, lignin, and protein) relative to the total stover mass. Washing of the corn stover from the 2014 harvest to stimulate the modified-Ritter storage reduced the extractable ash content from 5.8 to 3.5% and total extractives


*Samples were collected for analysis before (as-harvested) and after storage experiments. Values represent means with one SD in parentheses. p-Values (italicized) in bold are* <*0.05 and considered statistically significant as measured by an unequal variance t-test comparing stored samples to corresponding performed on as-harvested material.*

TABLE 4 | Analysis of percentage of corn stover sugars stored under conditions that were anaerobic, aerobic and in the field.


*Samples were collected before (as-harvested) and after storage experiments. Values represent means with one SD in parentheses. p-Values (italicized) in bold are* <*0.05 and considered statistically significant as measured by an unequal variance t-test comparing stored samples to corresponding performed on as-harvested material.*

from 22.3 to 15.6%, meanwhile proportionally increasing the structural components, including the sugars and lignin. In the anaerobic storage experiments, both ensiling and modified-Ritter storage methods resulted in significant decreases in soluble glucan and xylan as well as total extractives. Other than protein levels and soluble galactose (ensiled only), no other statistically significant differences were observed as a result of ensiling or modified-Ritter storage.

Aerobic storage in the two laboratory conditions resulted in many statistically significant changes compared to the asharvested material; extractable inorganics decreased, acetate decreased, and lignin was enriched. Total extractives decreased from 28.7 to 19.2% in the 0.5 L/min scenario but increased to 33.5% in the 1 L/min case. Additional changes were observed in the carbohydrates in terms of soluble, structural, and total levels. For the 0.5 L/min scenario, structural glucan was enriched from 25.1 to 30.0% with a corresponding decrease in soluble glucan from 6.4 to 1.1%, resulting in no overall difference in total glucan. Total xylan, galactan, and arabinan were all significantly enriched as a result of this lower airflow condition, although only the increase in structural xylan and soluble arabinan was significant. For the 1 L/min scenario, total glucan was reduced from 31.5 to 28.9% due to the reduction of soluble glucan from 6.4 to 3.3%. Soluble xylan, galactan, and arabinan all increased significantly spurred by a decrease in the corresponding structural counterparts; however, the total levels were unchanged despite the fact that solubilization occurred.

Field-stored samples were taken from multiple zones in the pile and data were combined for simplicity (*n* = 6 zones). Many of the compositional changes observed in the field generally followed the trends of the 1 L/min condition, with a significant reduction of extractable ash, increase in total extractives, and reduction of total glucan compared to the as-harvested material and no significant difference compared to the 1 L/min condition (*p* > 0.05, not shown). Significant loss of acetate and structural xylan, galactan, and arabinan was also evident in the field sample. Notable differences were observed in field samples stored at 0.6–0.7 m depths compared to 1.2–1.3 m depths. For example, structural glucan was reduced from 25.1 to 22.6% at the 0.6–0.7 m depth (*p* = 0.025, *n* = 2) and structural xylan was reduced from 14.6 to 13.3% (*p* = 0.002, *n* = 2), but there was no significant difference in either component at 1.2–1.3 m depths with glucan at 24.4% (*p* = 0.34, *n* = 4) and xylan at 14.2% (*p* = 0.100, *n* = 4). Acetate was also reduced from 2.3% at the time of harvest to 1.6% at the 0.6–0.7 m depth (*p* = 0.003, *n* = 2) but only to 1.8% at the 1.2–1.3 m depth (*p* = 0.003, *n* = 4).

#### Pretreatment and Enzymatic Hydrolysis

The laboratory-ensiled corn stover, field-stored stover, and the stover from 1 L/min airflow laboratory condition, along with their as-harvested counterpart, were subject to biomass pretreatments using dilute acid or dilute alkali, both followed by enzymatic hydrolysis for determining total sugar release. **Figures 5** and **6** show the proportion of glucose, xylose, and fermentation inhibitors as a result of pretreatment and enzymatic hydrolysis. Total reactivity is also presented, which is a measure of total glucose and xylose yield from the total structural and non-structural carbohydrates in the corn stover.

Laboratory-ensiled corn stover from the 2014 harvest had minor yet statistically significant variations compared to as-harvested stover as a result of dilute acid pretreatment. For example, glucose was slightly higher for the as-harvested sample, but xylose yield following acid pretreatment was increased as a result of laboratory-ensiled storage. No changes were seen in subsequent enzymatic hydrolysis of these two samples, resulting in the net effect of no statistically measurable difference in feedstock reactivity as a result of laboratory ensiling with the dilute acid pretreatment approach. Within the 2015 samples, significant differences were observed as a result of dilute acid pretreatment and enzymatic hydrolysis. Stored samples released approximately half of the glucose after pretreatment compared to the as-harvested stover, yet increased glucose was released in enzymatic hydrolysis in the aerobic storage condition. This resulted in similar total glucose yields for the as-harvested and aerobically stored samples, significantly higher than the fieldstored material, and this trend was conserved in final feedstock reactivity measurements.

An evaluation of the samples collected from the 2015 harvest showed markedly different response to dilute acid pretreatment. First, there was higher total soluble glucose released and ultimately higher reactivity as compared to the 2014 season samples; however, this was not surprising considering that the soluble glucan levels in the 2015 as-harvested field samples were also several fold greater than the 2014 as-harvested samples (5.9 vs 1.0–1.6%, respectively, **Table 4**). However, the 2015 samples produced significantly more fermentation inhibitors than in 2014. In the as-harvested samples, soluble glucose was degraded to 5-HMF and levulinic acid (0.071 g inhibitor/g dry biomass), whereas in the field-ensiled sample less glucose was converted to 5-HMF (0.031 g inhibitor/g dry biomass) without levulinic acid formation (data shown represent the sum of 5-HMF and levulinic acid). By contrast, less than 0.004 g/g 5-HMF was produced from glucose degradation for the laboratory-ensiled samples (2014 harvest) and levulinic acid was not produced.

Dilute alkaline pretreatment and enzymatic hydrolysis results are presented in **Figure 6**. In the laboratory-ensiled samples, no difference in sugar release in pretreatment was observed compared to the as-harvested samples but total glucose and xylose release in enzymatic hydrolysis were significantly increased, resulting in higher feedstock reactivity. For the field-ensiled samples, there was a statistically significant reduction in glucose release after alkaline pretreatment compared to the as-harvested material, which is consistent with the dilute acid pretreatment. However, no other significant difference in sugar yield was measured due to field ensiling, resulting in no significant difference in feedstock reactivity in the field-stored vs as-harvested sample. Similar to the

FIGURE 5 | Sugars released from corn stover with dilute acid pretreatment and enzymatic hydrolysis for the as-harvested or stored samples (A1, glucose; A2, xylose; A3, reactivity). Error bars represent the standards of deviation (*n* = 3). Letters represent statistically distinct values as determined by Tukey's test.

FIGURE 6 | Sugars released from corn stover with dilute alkaline pretreatment and enzymatic hydrolysis for the as-harvested or stored samples (B1, glucose; B2, xylose; B3, reactivity). Error bars represent the standards of deviation (*n* = 3). Letters represent statistically distinct values as determined by Tukey's test.

2014 samples, aerobic storage led to increased glucose yield in enzymatic hydrolysis; however, a significant reduction of xylose release in enzymatic hydrolysis resulted in a final material that had a similar reactivity to the as-harvested stover.

#### DISCUSSION

#### Particle Size Reduction in the Field

Forage chopping is a common practice employed when making silage for livestock feed, and the size reduction serves to improve packing density and, thus, limit oxygen infiltration during storage. Likewise, forage chopping of high-moisture feedstock can be used for biofuels-related crops and can reduce the preprocessing required to meet <6 mm target size specification for a biorefinery (Humbird et al., 2011). For the one- and two-pass corn stover harvesting operations tested, only 38 and 33% of the biomass would need further size reduction to meet size specification, respectively. While the geometric means from the two harvest years are similar and are in the range of previous reports for forage-chopped corn stover (Shinners et al., 2011), the increased SD in the 2014 harvest suggests a wide particle size range compared to the 2015 harvest. Geometric mean is often reported in forage-specific literature, and yet full PSD information is necessary to meet size specifications at a biorefinery and to identify opportunities for fractional removal of on-specification materials. For example, Lisowski et al. (2017) cite the need for stringent PSDs when using biomass for bioenergy purposes and provide PSDs for a range of foragechopped high-moisture energy crops. For the two-pass system, placing the corn stover on the soil between the flail shredding and forage chopping operations resulted in more soil entrainment in the biomass, as evidenced by an increase in total ash into the system from 9% for the 2014 harvest and 19% for the 2015 harvest (**Table 3**). Therefore, while the two-pass harvesting operation resulted in better size reduction and would require less preprocessing at a biorefinery to meet target size specification, it also resulted in increased soil (ash) content that would require removal and disposal at the biorefinery.

#### Anaerobic Storage Performance

The preservation of biomass during storage is a key indicator in the success of a particular storage approach. Likewise, gas production is an important factor in evaluating a storage system for greenhouse gas and air pollutant potential and thus was measured in the wet storage conditions tested. The increased production of CO2 and other permanent greenhouse gasses in the modified-Ritter storage reactors along with higher DML suggested greater microbial activity than for the traditional ensiling material. The presence of CO has been documented in composting of green and municipal solid wastes and has been related to a physiochemical process that occurs in the presence of oxygen during the initial stages of composting (Hellebrand and Kalk, 2001; Phillip et al., 2011). Hellebrand and Kalk demonstrated elevated CO levels at the beginning of composting and after each aeration episode, while Phillip et al. showed that CO levels were elevated in sterilized compost compared to non-sterilized. For the corn stover experiments, the highest CO levels were measured in the first 6 days of storage in the anaerobic reactors as well as in the first 7 days of the field storage demonstration (data not shown). This further supports the claim that anaerobic storage conditions should be established for high-moisture biomass to prevent both degradation and greenhouse gas production.

Successful ensiling results from the fermentative production of lactic acid from soluble sugars and is carried out by homolactic and heterolactic bacteria that help acidify material and prevent other microorganisms from degrading the dry matter; other compounds, such as acetic acid, may also be produced depending on the predominant fermentation pathways (McGechan, 1990; McDonald et al., 1991). The ensiled material was characterized by significant lactic acid, while the modified-Ritter method was primarily acetic and butyric acid. Similarly, Morgan et al. (1974) and Atchison and Hettenhaus (2003) reported an absence of lactic acid bacteria and lactic acid in sugarcane bagasse stored using the Ritter methods but observed the presence of acetic, butyric and propionic acid, likely due to the presence of *Clostridia* and *Bacillus* species in the feedstock.

In the present study, the lack of lactic acid and presence of butyric acid in the modified-Ritter material suggested that the washing process, used to simulate the slurrying of biomass in the Ritter method, may have had a negative effect on the lactic acid fermentation process. While a lactic acid bacteria inoculant could be added to the modified-Ritter storage method after the simulated slurrying, there is little practicality of doing this during field-scale storage as it would be challenging and costly to execute. The washing step of the as-harvested material was successful in removing about 40% of the soil contamination, which is measured as extractable inorganics. Similarly, total extractives and soluble glucan and xylan were significantly reduced by approximately one third as a result of the washing step. While there are benefits to removing these soluble components, primarily ash, prior to conversion, the additional DML observed in the modified-Ritter storage suggests that washing removed the necessary soluble sugars for successful fermentation. Overall, the organic acid profiles along with the DML and gas production observation suggest that the lactic acid fermentation associated with the ensiling storage method resulted in superior performance compared to the modified-Ritter and aerobic storage methods in the laboratory.

Anaerobic storage conditions preserved not only dry matter compared to aerobic storage but also the primary compositional components including the structural sugars. Similar observations for glucan and xylan preservation have been reported when DML in storage remains at 5% (Liu et al., 2013). A slight reduction in soluble sugar components, primarily soluble glucan, was observed as a result of both anaerobic storage methods. This is expected since consumption of the soluble sugars by fermentative microorganisms is necessary to fuel the production of organic acids for biomass preservation. Overall, the anaerobic conditions had minimal effects on composition and also preserved material significantly, and of the two methods, ensiling is the recommended approach due to the reduced DML.

#### Aerobic Storage Performance

Aerobic storage in laboratory reactors resulted in approximately 30% total loss of matter over the 111-day storage period. Significant self-heating was observed in the aerobic conditions likely due to microbial respiration of available carbohydrates, with the higher airflow condition exhibiting slightly higher temperatures and producing over twice the CO2 relative to the low-flow condition. Similar temperature profiles have been reported in piled corn stover (Shinners et al., 2011) as well as in laboratory-stored corn stover (Wendt et al., 2014; Bonner et al., 2015) and pine chips (Bonner et al., 2015). Self-heating also occurred during the field demonstration prior to covering. The temperatures closest to the surface (about 0.6 m depth) were >55°C, similar to that of the laboratory 1 L/min airflow reactors (~50–60°C for much of the experimental time). Temperatures at depths of ~1.2 m were approximately 50°C, more similar to the 0.5 L/min air flow reactors (~50°C for the first 80 days). These results suggest that the initial maximum temperatures were related to oxygen availability in the pile. Upon covering the field pile to encourage ensiling, the pile temperatures decreased dramatically to 20–30°C (lower temperatures were closer to the surface) suggesting that aerobic microbial activity was suppressed and that anaerobic conditions were established. While the organic acid profile for the field-stored samples was more similar to the 0.5 L/min aerobic case, limiting oxygen to the pile was effective in limiting overall DML to <5%. Considering the 30% DML losses in the aerobic conditions, the low DML achieved in the field suggest anaerobic conditions were prevalent within the storage pile.

In contrast to the relative stability of biomass components (i.e., carbohydrates, lignin, and inorganic nutrients) observed in the anaerobic storage laboratory experiments, aerobic and field storage resulted in measurable compositional changes. Extractable inorganics decreased in all samples, likely due to the consumption of macronutrients to support microbial respiration. A significant reduction in acetate was observed in all samples, suggesting that a mild form of pretreatment likely occurred. Acetate is one of the first structural components to be released during degradation, as acetyl bonds in the hemicellulose (xylan, galactan, and arabinan) are cleaved by microorganisms attempting to break down hemicellulose and later cellulose (primarily glucan). An assessment of the shift of structural xylan, galactan, and arabinan to soluble forms, along with significant acetate reduction, suggested that hemicellulose depolymerization occurred during aerobic storage with the 1 L/min airflow. A different story is evident as a result of storage at the lower airflow rate of 0.5 L/min. While total glucan was preserved under this condition, >75% of the soluble glucan was lost and a concomitant enrichment of structural glucan occurred. Hemicellulose depolymerization was not evident under low-airflow storage condition, as measured by an enrichment of structural xylan. Overall, the structural sugar profiles suggested that the lower airflow condition preserved the sugars better than in the high airflow condition, which were characterized by higher microbial consumption of the structural sugars as well as mild pretreatment effects. Interestingly, hemicellulose depolymerization was seen in the field-stored samples, with solubilization of galactan and arabinan at all pile depths and significant glucan and xylan solubilization in the shallower samples only (0.6–0.7 m depths). These results suggested that the microbial-mediated self-heating that occurred during the initial 3 weeks of storage was sufficient to produce the mild pretreatment effects seen in the 1 L/min airflow condition in the laboratory.

Lignin content was enriched as a result of aerobic storage, where approximately 30% of dry matter was lost. Lignin content is often proportionally enriched as a result of storage loss due to the biodegradation of cellulose and hemicellulose and the inaccessibility of lignin to microorganisms, and similar results have been reported for aerobically stored corn stover (Athmanathan et al., 2015). Not surprisingly, lignin content was not enriched in the field-stored samples due to the low DML that occurred as a result of the anaerobic conditions created after covering.

## Sugar Release

The pretreatment conditions chosen for this study are industrially relevant conditions and were selected to deliver maximum hydrolysis of structural sugars in cellulose and hemicellulose to monomeric sugars for subsequent fermentation to fuels and/or chemicals. In this study, enzymatic hydrolysis followed either acid or alkaline pretreatments in order to determine total glucose and xylose yield as well as feedstock reactivity. Dilute acid pretreatment resulted in glucose and xylose release several fold higher compared to alkaline pretreatment, however subsequent enzymatic hydrolysis improved depolymerization such that total glucose and xylose releases were only marginally less for the alkaline method. Similar trends in dilute acid and alkaline pretreatment have been reported for corn stover (Duguid et al., 2009) and are a response to the different chemistries; acid based chemistries remove hemicellulose so that the cellulose is assessable to enzymatic depolymerization (Mosier et al., 2005), whereas alkaline methods solubilize lignin and leave hemicellulose less affected (Alvira et al., 2010). A portion of the soluble glucan in the sample sets collected from the 2015 season was over-pretreated in the dilute acid method, resulting in a significant formation of fermentation inhibitors in the as-harvested material. However, both field and aerobic storage were effective in lowering the inhibitor levels.

Laboratory-ensiling resulted in no change in feedstock reactivity or a slight increase in reactivity as a result of pretreatment with dilute acid and dilute alkali, respectively, followed by enzymatic hydrolysis. A similar result has been observed in ensiled sorghum subject to dilute alkali pretreatment (Sambusiti et al., 2012) and ensiled corn stover subjected to steam explosion and enzymatic hydrolysis (Liu et al., 2013). The field-stored sample showed slightly less feedstock reactivity in the dilute acid pretreatment approach, and no changes in reactivity were observed with the dilute alkali approach. While the field sample points to a slightly different trend compared to the laboratory-ensiled corn stover, it can be understood by assessing carbohydrate changes as a result of storage. Whereas the laboratory storage experiment was sealed from the atmosphere immediately upon setup, the field-stored pile self-heated for 3 weeks, resulting in lower soluble sugar levels compared to the as-harvested. Therefore, less sugar was available for release in the subsequent conversion tests. Delayed sealing has been widely reported to reduce forage quality for livestock feed (Henderson and McDonald, 1975), and similar practices of quick sealing upon storage pile formation are recommended for the bioenergy industry. However, this field demonstration showed that delayed pile coverage, which may occur as a result of competing priorities during harvest season, resulted in less than 5% DML and that the remaining material yielded equivalent amounts of glucan and xylan upon pretreatment and enzymatic hydrolysis relative to the unstored starting materials. This indicates that biomass producers and aggregators have an operational window in which to finalize storage pile construction and covering without incurring extensive material losses and quality changes.

Laboratory-based aerobic storage experiments (1 L/min airflow) using the same corn stover that was collected for the field study indicated that these samples were slightly more reactive than the field-stored stover, also evidenced by hemicellulose depolymerization (**Table 4**). However, when total structural sugar availability and release were considered, regardless of pretreatment chemistry, sugar release from these aerobically stored materials was similar to the as-harvested samples (Wolfrum et al., 2013). Herrmann et al. suggest that DML should be accounted for when considering final energy yield from stored biomass (Herrmann et al., 2011); this adjustment was not made in this study such that it would be possible to assess feedstock reactivity in terms of the vastly different storage histories a biorefinery might encounter. There is still debate as to whether the farmer or the biorefinery will pay for any penalties incurred as a result of DML; some models assume a set price for tonnage harvested whereas others pay only for the biomass delivered to the biorefinery gate. Regardless, the loss of >30% total dry matter and the corresponding reduction of carbohydrates as a result of aerobic storage would be unacceptable for the refinery and should be avoided.

The sugar release results provide strong support for the incorporation of wet stored biomass into commercial biochemical conversion processes. The nearly 100% release of sugars along with inhibitor formation suggests that the severity of the dilute acid pretreatment method was too high to accurately measure storage-related changes in this feedstock. This finding highlights two opportunities for cost reduction in conversion, either through reduced pretreatment severity through lower temperatures or acid levels, or for capturing the water extractives—including soluble carbohydrates—prior to pretreatment and defining a value added product stream for a conversion facility. Wolfrum et al. suggest that lowering severity levels is an effective approach to assessing biomass-related responses in conversion (Wolfrum et al., 2013). Other groups have suggested converting soluble sugars to valuable products during storage, for example to ethanol or volatile organic acids, and extracting them prior to conversion (Henk and Linden, 1996; Hamilton et al., 2016). Results of this study show that fermentable sugar release from wet-stored biomass was equivalent to that of freshly-harvested materials using two different commercially-relevant pretreatment methods. This suggests that anaerobic storage of high-moisture corn stover is a potential near-term solution to manage storage losses in commercial biorefinery logistics operations.

# CONCLUSION

A primary challenge associated with the dry bale logistics system for providing herbaceous materials for bioenergy is the loss of feedstock in storage due to microbial degradation and potential fires. This study demonstrated that wet anaerobic storage is an active management approach for corn stover to preserve biomass in the supply chain. Storage performance was measured in terms of total DML, compositional analysis (i.e., carbohydrates, organic acids, ash, lignin, etc.), gas production, and the potential for sugar release for conversion to biofuels. Both laboratory and field studies showed that long-term stability can be achieved with little effect on feedstock performance in terms of sugar release. Furthermore, this study confirmed that field-chopping and particle size reduction early in the supply chain removed the bulk logistics system's dependency on drying corn stover prior to baling and could be used to diminish the biorefinery size reduction requirements; in-field forage chopping was capable of reducing over 60% of the corn stover to a particle size of less than 6 mm. Additional opportunities beyond preservation are also possible with wet storage, for example with directed microbial preprocessing for improved convertibility. In summary, incorporating feedstock supply logistics systems centered around high-moisture biomass and wet anaerobic storage offer the potential for biorefineries to reduce the risks associated dry baled feedstock meanwhile providing a feedstock that is compatible with existing conversion technologies.

# AUTHOR CONTRIBUTIONS

LW, JM, and WS performed a number of experiments and drafted the manuscript. TR and QN executed the field experimentation. TR, LL, and QH also performed experiments. DR, NS, AR, AH, and QN performed data analysis and revised the manuscript.

# ACKNOWLEDGMENTS

The authors thank Karen Delezene-Briggs, Eric Fillerup, Sergio Hernandez, Sabrina Morgan, Kastli Schaller, and Brad Thomas at Idaho National Laboratory for their efforts in sample analysis and Vicki Thompson for review of the manuscript. Idaho National Laboratory and Lawrence Berkley National Laboratory would like to acknowledge core funding from the U.S. Department of Energy Office of Energy Efficiency and Renewable Energy's Bioenergy Technologies Office, as well funding from the American Recovery and Reinvestment Act for the Advanced Biofuels Process Development Unit. This work is supported by the U.S. Department of Energy, under DOE Idaho Operations Office Contract DE-AC07-05ID14517. Accordingly, the U.S. Government retains a non-exclusive, royalty-free license to publish or reproduce the published form of this contribution, or allow others to do so, for U.S. Government purposes. The views and opinions of the authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.

#### REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer MR and handling Editor declared their shared affiliation.

*Copyright © 2018 Wendt, Murphy, Smith, Robb, Reed, Ray, Liang, He, Sun, Hoover and Nguyen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Impact of Drought on Chemical Composition and Sugar Yields From Dilute-Acid Pretreatment and Enzymatic Hydrolysis of Miscanthus, a Tall Fescue Mixture, and Switchgrass

Amber Hoover <sup>1</sup> \*, Rachel Emerson<sup>1</sup> , Allison Ray <sup>1</sup> , Daniel Stevens <sup>1</sup> , Sabrina Morgan<sup>1</sup> , Marnie Cortez <sup>1</sup> , Robert Kallenbach<sup>2</sup> , Matthew Sousek <sup>3</sup> , Rodney Farris <sup>4</sup> and Dayna Daubaras <sup>1</sup>

#### Edited by:

Sachin Kumar, Sardar Swaran Singh National Institute of Renewable Energy, India

#### Reviewed by:

Yaoping Zhang, University of Wisconsin-Madison, United States Héctor A. Ruiz, Universidad Autónoma de Coahuila, Mexico

#### \*Correspondence:

Amber Hoover amber.hoover@inl.gov

#### Specialty section:

This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research

Received: 08 January 2018 Accepted: 30 May 2018 Published: 19 June 2018

#### Citation:

Hoover A, Emerson R, Ray A, Stevens D, Morgan S, Cortez M, Kallenbach R, Sousek M, Farris R and Daubaras D (2018) Impact of Drought on Chemical Composition and Sugar Yields From Dilute-Acid Pretreatment and Enzymatic Hydrolysis of Miscanthus, a Tall Fescue Mixture, and Switchgrass. Front. Energy Res. 6:54. doi: 10.3389/fenrg.2018.00054 1 Idaho National Laboratory, Idaho Falls, ID, United States, <sup>2</sup> Division of Plant Sciences, University of Missouri, Columbia, MO, United States, <sup>3</sup> Department of Agronomy & Horticulture, University of Nebraska-Lincoln, Lincoln, NE, United States, <sup>4</sup> Oklahoma Agricultural Experiment Station, Oklahoma State University, OK, United States

Environmental factors like drought impact the quality of biomass entering a bioconversion process. Drought often reduces the sugar content in lignocellulosic biomass, which could have economic impacts, particularly when compounded with losses in dry biomass yield; however, the effects on conversion efficiency are not completely understood. This study investigated how drought may impact biomass composition and sugar yields from dilute-acid pretreatment and enzymatic hydrolysis of Miscanthus, a tall fescue mixture, and switchgrass from Nebraska, Missouri, and Oklahoma, respectively, grown as part of Regional Feedstock Partnership field trials. Samples were grown and harvested in 2010 during non-drought conditions and in 2012 during extreme drought conditions. Non-structural glucose and proline were significantly greater in 2012 compared with 2010 for Miscanthus, which suggests drought stress occurred. Structural glucan and xylan were significantly decreased in 2012 for Miscanthus; however, reactivity and sugar yields from dilute-acid pretreatment and enzymatic hydrolysis were significantly greater in 2012 compared with 2010, suggesting that although structural sugars may decrease during drought conditions, sugar yields and reactivity may increase. For the tall fescue mixture, proline was greater, and structural sugars were lower in 2012, indicating drought stress, but minimal differences were observed in the conversion experiments. Few differences were observed for switchgrass composition and reactivity between years. The observed patterns are likely because of site-specific climatic conditions combined with the tolerance each species may have to drought. As drought occurrence and severity have increased, it is necessary to understand drought impacts to mitigate risks to future bioenergy industry growth.

Keywords: drought, conversion performance, composition, Miscanthus, switchgrass, tall fescue

**22**

# INTRODUCTION

Lignocellulosic biomass is a promising renewable resource that can help meet U.S. energy demands and support domestic production of fuels and chemicals (U. S. Department of Energy, 2016). A supply of high-quality, consistent feedstock is needed for each biorefinery, which can require upwards of 680,000 Mg of dry feedstock per year (Humbird et al., 2011; Turhollow et al., 2014). Valuation of biomass on cost per dry ton that ignores biomass quality could result in highly variable feedstock quality and performance contributing to market uncertainty (Sokhansanj et al., 2002; Kenney et al., 2013; Li et al., 2016; Williams et al., 2016). Understanding the impact of location, environmental conditions, harvesting methods, and other agronomic treatments on both biomass quality and quantity is therefore important for ensuring a high-quality, consistent feedstock supply to biorefineries.

Based on historical trends, corn yields have a greater risk of decline than oil imports, with weather being a primary source of volatility in corn and ethanol production (Eaves and Eaves, 2007). During every decade from 1900 to 2011, drought events covered at least 30% of the U.S. (Peterson et al., 2013), and drought has occurred more frequently in the last 30 years than over the previous 100 (Stone et al., 2010). In 2012, according to the U.S. Drought Monitor, the U.S. experienced drought conditions across 65% of the continental U.S.; the largest coverage since the 1950s (Rippey, 2015). The 2012 drought reduced U.S. grain yield from corn by 27% (Rippey, 2015), increased competition for corn grain, and impacted ethanol production and cost (Cushman, 2012). The National Centers for Environmental Information estimates that the 2012 drought caused \$30 billion in losses, primarily in agriculture (Rippey, 2015). Climate studies indicate that more extreme weather abnormalities will occur in the future, including increases in drought frequency and severity (Wehner et al., 2011; Hansen et al., 2012; Williams et al., 2014). The increases in extreme weather will necessarily impact the biomass supply for the biorefining industry, causing reductions in yields of agricultural crops and increases in the price of crops and biofuels (Langholtz et al., 2014). In modeled scenarios, biofuel production costs increased 35% from a wet year compared to a dry year (Morrow et al., 2014), demonstrating the significant economic risk from drought. In addition, proposed risk mitigation strategies, such as sourcing material from a larger area or storing biomass from non-drought years, showed little impact on reducing risk associated with drought conditions (Morrow et al., 2014).

It has been well documented that drought decreases crop yields, with examples in the literature for corn grain and stover (Earl and Davis, 2003; Sanford et al., 2016); Miscanthus (Sanford et al., 2016); mixed perennial grasses (Emerson et al., 2014); native grasses (Sanford et al., 2016); barley grain and tillers (Samarah, 2005); vegetative growth and yield of rice (Mostajeran and Rahimi-Eichi, 2009); and wheat grain (Keyvan, 2010). Drought sensitivity is present in the fast-growing, secondgeneration bioenergy crops, such as Miscanthus (Emerson et al., 2014), partly because of the limited trials for improvement of these crops (Karp and Shield, 2008; Oliver et al., 2009); however, individual species and cultivars vary in how they respond to drought conditions (Lewandowski et al., 2003; Karp and Shield, 2008). Miscanthus × giganteus has displayed a stronger response to water deficiencies than switchgrass (Heaton et al., 2004), which has often been reported to be drought tolerant (Lewandowski et al., 2003). Low water requirements for switchgrass have been reported in screening trials on different soil types across 31 U.S. states (Wright and Turhollow, 2010). In contrast, other studies found that switchgrass had low yields during extreme drought, and biomass production decreased 80% in one case; however, plant survival was high, and plants responded well to increased soil moisture following drought, indicating switchgrass may be well suited for many environments following breeding or other improvement studies (McLaughlin and Kszos, 2005; Barney et al., 2009). Similar patterns have been reported for tall fescue, where fall and spring growth increased following summer drought, but the increased growth was not enough to counter low summer yields (Horst and Nelson, 1979). Tall fescue is a cool-season grass that has been used for forage and erosion control and is grown on grasslands that may have potential for biofuel production (Tilman et al., 2006, 2009) as little land use alterations would be required (Walsh et al., 2003). Potential energy crops that have been reported to have some level of drought tolerance include sorghum, which had improved yields in the absence of recurring drought (Gill et al., 2014), and reed canary grass (Lewandowski et al., 2003).

Drought has clear negative consequences for crop yields, but the effect of drought on biomass quality for biofuel production is less well understood. Plant responses to drought include a variety of complex physiological and biochemical responses that could affect biomass composition (Farooq et al., 2009), and biomass composition is directly linked to recalcitrance and conversion performance. Plant nutrient intake is impacted by water stress since nutrients are generally taken up with water (Lipiec et al., 2013). Drought-induced reactive oxygen species cause membrane injuries, protein degradation, and enzyme inactivation (Liu and Huang, 2000; Zlatev and Lidon, 2012). Production of antioxidant metabolites, such as lignins, were reported to alleviate cellular injury (Wahid, 2007; Zlatev and Lidon, 2012). Drought-stressed plants maintain water absorption and cell turgor by synthesizing compatible solutes, thus retaining tissue metabolic activity and allowing for regrowth if conditions improve (Chaves et al., 2003). Compatible solutes include proteins and amino acids (Samuel et al., 2000; Hamilton and Heckathorn, 2001), carbohydrates (Vijn and Smeekens, 1999), organic acids (Farooq et al., 2009), and cyclitols (Anderson and Kohorn, 2001). Proline, a compatible solute, can significantly accumulate during water stress and has been suggested for use as a drought-injury sensor (Iannucci et al., 2000). Proline accumulation as a result of drought has been reported in wheat (Keyvan, 2010) and rice (Mostajeran and Rahimi-Eichi, 2009). Carbohydrates also serve a role in signal transduction that represses genes associated with photosynthesis and reserve mobilization (Koch, 1996). Drought induced accumulation of soluble carbohydrates has been reported in studies of switchgrass (Ong et al., 2016), wheat (Keyvan, 2010), and rice (Mostajeran and Rahimi-Eichi, 2009), but was not observed in shoots of soybeans (Al-Hakimi, 2006).

Glucan, xylan, and lignin are of particular importance in a bioconversion process. A reduction in cellulose, lignin, and soluble carbohydrates was reported for shoots of droughtstressed soybeans, but hemicellulose increased (Al-Hakimi, 2006). Reduced cellulose and increased hemicellulose content were also observed for Miscanthus (Van Der Weijde et al., 2017). Corn stover, Miscanthus, switchgrass, and mixed perennial grasses grown under drought conditions had significantly more extractives, which would include soluble carbohydrates, and lower glucan, xylan, and lignin (Emerson et al., 2014; Ong et al., 2016). Emerson et al. (2014) reported that mixed perennial grasses had similar trends, and although no change in glucan occurred, increased xylan was observed. Following the 2012 drought, compositional changes in glucan and xylan for corn stover, Miscanthus, and mixed perennial grasses decreased theoretical ethanol yield (L Mg−<sup>1</sup> ) by 10–15%. This decrease was extrapolated to an economic loss of greater than \$283 Mg−<sup>1</sup> of biomass, with an assumed minimum ethanol selling price of \$8.37 L−<sup>1</sup> (Tao et al., 2013; Emerson et al., 2014). Studies that investigate the impact of drought on bioconversion of feedstocks, particularly field-scale studies, are limited.

This study investigates the impacts of drought conditions on biomass composition and sugar yields from dilute-acid pretreatment and enzymatic hydrolysis of crops grown at field scale. The study capitalized on samples already being collected as part of the Regional Feedstock Partnership, which was an initiative started by the U.S. Department of Energy and the Sun Grant Institute to address the barriers associated with supplying a sustainable and reliable source of feedstock to a large-scale biorefining industry. The Regional Feedstock Partnership was a series of field trials that each lasted for five to seven years for nine biomass types. These field trials were ongoing during 2010, a year without drought, and 2012, during which a nationwide drought occurred. The crop types included in this study were Miscanthus grown in Nebraska, a tall fescue mixture harvested from Missouri, and switchgrass grown in Oklahoma. The goal of the study was to observe the similarities and differences for biomass collected in 2010 and 2012 for each crop type individually, as the crops were grown in different locations and cannot directly be compared. To the authors' knowledge, this is the first study investigating impacts of drought on dilute-acid pretreatment and enzymatic hydrolysis of biomass.

For each of the three species three aspects were considered to determine the location-specific drought conditions and the drought stress reactions for each of the biomass types. First, the precipitation and temperature trends during the growing season were considered. As plants that experience drought conditions can recover if favorable conditions return prior to harvest (McLaughlin and Kszos, 2005), it is important to look at the environmental conditions throughout the year prior to the time of harvest. Second, chemical changes including multiple types of compatible solutes, structural components, and inorganic elements were assessed. Compatible solutes are produced during drought stress to help maintain turgor pressure by lowering water potential (Chaves et al., 2003; Farooq et al., 2009). Studying the levels of compatible solutes, or osmolytes, has two purposes. First, knowledge of compatible solute levels can be used to help determine whether plants were drought stressed, explaining the patterns observed in other chemical-composition parameters or carbohydrate yields from conversion tests. Second, changes in composition of compatible solutes during drought could impact downstream processing of biomass to fuels. Increases in composition parameters, such as extractives, could be at the expense of more desirable characteristics, such as structural carbohydrates, and may lead to formation of microbial inhibitors (Ong et al., 2016). Structural components such as structural sugars and lignin are critical for understanding biomass value to a biorefinery. Ash in biomass feedstock can be an issue because it is not convertible into fuels. Ash causes many issues; it accumulates as waste in biorefineries, adding disposal costs; it can increase the neutralization capacity of the biomass during dilute-acid pretreatment; and it can cause wear on feeding and handling systems (Weiss et al., 2010). Last, conversion efficiencies, fermentation inhibitor formation, and calculated theoretical ethanol yields were used to determine each sample's reaction to drought and estimate the overall impact on a bioconversion process.

#### METHODS

#### Drought-Based Sample Selection

The Regional Feedstock Partnership field trials for a variety of energy crops began in 2008. Regional Feedstock Partnership sample sets were selected for analysis from field-study locations that had minimal to no drought during 2010 and also from locations with drought conditions during 2012 for three feedstocks, switchgrass, Miscanthus, and a tall fescue mixture. Drought conditions were determined according to the University of Nebraska-Lincoln U.S. Drought Monitor<sup>1</sup> . The five drought condition categories, from least to most severe, were abnormally dry, moderate drought, severe drought, extreme drought, and exceptional drought. Numerous values were used to calculate these drought conditions, including the Palmer Drought Index, the Climate Prediction Center Soil Moisture Model, the U.S. Geological Survey Weekly Streamflow, the Standardized Precipitation Index, drought duration, and additional regionspecific information<sup>1</sup> . Biomass was harvested in the fall for the selected sample sets; therefore, the end of October was used for observation of drought conditions in the counties of interest (**Figure 1**). In addition, drought-index data were analyzed from January 2010 until December 2012.

#### Biomass

Three replicate 10 m<sup>2</sup> plots of Miscanthus × giganteus were established in summer 2008 and grown with no nitrogen amendment in Saunders County, Nebraska, as part of the field trials conducted by the Regional Feedstock Partnership. Biomass was harvested annually, and samples used in this study were harvested on December 2, 2010, and November 8, 2012.

<sup>1</sup>US Drought Monitor, National Drought Mitigation Center, University Nebraska-Lincoln, http://droughtmonitor.unl.edu/MapsandDataServices/MapService.aspx

Dry biomass yield was determined by harvesting Miscanthus from the 10 m<sup>2</sup> plot, taking a subsample from the harvested biomass, drying the subsample at 60◦C for 48 h, weighing the dry subsample, and then extrapolating the dry weight of the subsample to the entire plot.

The tall fescue mixture was a cool-season perennial grass mixture (Lee et al., 2013) grown in Boone County, Missouri, on a site managed in accordance with Conservation Reserve Program regulations since 2004. Three replicate 0.4 ha plots were established in spring 2008 and grown with no nitrogen amendment as part of the Regional Feedstock Partnership field trials. Biomass was harvested from 0.2 ha split plots annually in the spring and fall. The samples used in this study were harvested on November 4, 2010, and October 29, 2012. Species composition measured in June 2009 and averaged across the three plots was 48% tall fescue (Schedonorus phoenix), 24% red clover (Trifolium pratense), 12% yellow sweetclover (Melilotus officinalis), and 8% white clover (Trifolium repens). Grass weed and broadleaf plants made up the remaining 8% of the plots. To determine dry biomass yield, the tall fescue mixture was harvested at a height of 10–15 cm, baled, and weighed. Bales were then subsampled using a 5 cm diameter and 50 cm long electric corer. Subsamples were dried at 60◦C for 48 h, and the subsample dry weight and bale weight were used to calculate dry biomass for the entire plot.

Three replicate 0.4 ha plots of "Blackwell" Panicum virgatum L. (switchgrass) were planted on September 2, 2008, and grown with no nitrogen amendment in Muskogee County, Oklahoma, as part of the Regional Feedstock Partnership field trials. Switchgrass was harvested annually, and samples in this study were harvested on October 28, 2010, and November, 5, 2012. Switchgrass was harvested at a height of 10–15 cm from 0.4 ha of each plot, baled, and weighed. Approximately 300 g cores were taken from the bales using a hay probe, and dry matter was determined by weighing the biomass from each core, drying the sample at 60◦C for 48 h, and then reweighing the sample.

All biomass samples were milled to pass through a 2 mm sieve using a Thomas Model 4 Wiley Mill (Thomas Scientific, Swedesboro, NJ).

#### Precipitation and Temperature Records

Precipitation and temperature are key contributing factors to drought conditions. Emerson et al. (2014) reported total monthly precipitation and average monthly maximum temperature for 2010 and 2012 along with thirty-year averages (1981–2010) from the nearest Midwestern Regional Climate Center2,3 for Saunders County, NE (station: Mead 6 S Station USC00255362), and Boone County, MO (station: Columbia Regional Airport Station USW00003945). Total monthly precipitation in 2010 and 2012, average monthly maximum temperature for 2010 and 2012, and thirty-year averages for precipitation and temperature (1981–2010) for Muskogee County, OK (station: COOP:346130), were obtained from the National Oceanic and Atmospheric Administration<sup>3</sup> . Data for Muskogee County, OK, is provided in Figure S1 in the Supplementary Material.

## Chemical Composition

Chemical composition was determined for all untreated biomass samples in duplicate according to the National Renewable Energy Laboratory's laboratory analytical procedures for standard biomass analysis (Sluiter et al., 2010). Briefly, the extractions detailed in the laboratory analytical procedures were done using an accelerated solvent extractor (ASE) 350 (ThermoFisher, Scientific, Waltham, MA) three times for both water and ethanol for all samples to ensure adequate and consistent extraction of the material. To determine non-structural sugars content in water extractions, the liquid samples were analyzed via highperformance liquid chromatography (HPLC). To determine total extractable sugars and organic acids content in water extractions, the liquid samples were acid hydrolyzed prior to HPLC. The water extractives were adjusted to 4% acid using 72% sulfuric acid and autoclaved at 121◦C for 1 h. Acid-hydrolyzed samples were filtered through a 0.2µm filter and analyzed for organic acids. The remaining sample was neutralized using calcium carbonate, filtered through a 0.2µm filter, and analyzed for nonstructural sugars. Sugars were analyzed on an Aminex HPX-87P column (BioRad Laboratories, Hercules, CA) with a column temperature of 85◦C using a refractive index detector, a mobile phase of 18 M ultrapure water and a flow rate of 0.6 mL min−<sup>1</sup> . Organic acids were analyzed on an Aminex HPX-87H column (BioRad Laboratories, Hercules, CA), with a column temperature of 55◦C, using a diode array detector, a mobile phase of 0.01 M sulfuric acid, and a flow rate of 0.6 mL min−<sup>1</sup> . Duplicate injections were performed for all samples. For the two-stage acid hydrolysis of the extracted solids, specified in the laboratory analytical procedure, the resulting hydrolysis liquor was analyzed for monomeric sugars and organic acids using the same HPLC methods previously described. Acid-soluble lignin was determined by measuring absorbance at 320 nm with an ultraviolet-visible spectrophotometer (Varian Cary 50, Agilent, Santa Clara, CA) and calculating the concentration using Beer's Law with an extinction coefficient of 30. Protein was determined by measuring percent nitrogen using a LECO TruSpec CHN (St. Joseph, MI) and then multiplying that value by a nitrogenprotein conversion factor of 4.6.

#### Amino Acids

Samples were milled to pass through a 0.08 mm sieve using a Retsch ZM 200 (Haan, Germany). Amino acids were analyzed based on three AOAC methods. AOAC 994.12 (AOAC International, 1997) was followed to measure all of the amino acids except for tryptophan, which required a base hydrolysis using AOAC 988.15 (AOAC International, 1988b), and methionine and cysteine, which required a pre-oxidation step using modified AOAC 985.28 (AOAC International, 1988a). Samples were analyzed using an ion-exchange post column ninhydrin method. Specifically, following hydrolysis of the samples according to the AOAC methods mentioned previously, the extracts were filtered to remove particulates and analyzed by HPLC. After the amino acids exited the HPLC column, they were reacted with ninhydrin prior to detection and quantification.

## Dilute-Acid Pretreatment and Enzymatic Hydrolysis

Conversion performance was determined using a benchscale, dilute-acid pretreatment and enzymatic hydrolysis assay. Chemical composition of the untreated biomass was determined as described in the previous methods section on composition. Laboratory-scale, dilute-acid pretreatment was performed using an ASE 350 (ThermoFisher Scientific, Waltham, MA) according to Wolfrum et al. (2013). Wolfrum et al. (2013) determined that 130◦C was the optimal temperature for comparing biomass reactivity after conducting pretreatment experiments with temperatures ranging from 110 and 200◦C; therefore, a pretreatment temperature of 130◦C was selected for this study. Experiments were performed using 66-mL zirconium cells and

<sup>2</sup>Midwestern Regional Climate Center, University of Illinois at Urbana-Champaign, http://mrcc.sws.uiuc.edu

<sup>3</sup>National Oceanic and Atmospheric Administration, National Climatic Data Center, www.nws.noaa.gov/climate

a 10% (w/w) solids loading with an acid-to-biomass loading of 0.08 g g−<sup>1</sup> . Each cell was filled with 3.0 +/− 0.03 g biomass and 30 mL of 1% sulfuric acid (w/w). Cells were subjected to a 7 min heating period followed by a 7 min static time with a reaction temperature of 130◦C. Then cells were purged for 200 s with nitrogen. The temperature was reduced to 100◦C and 100 to 150 mL of nanopure water was rinsed through the cell with a 200 s nitrogen gas purge. Aliquots of the liquors rinsed from the solid biomass were collected for determination of total and monomeric sugars in the pretreatment liquors using the same HPLC methods previously described in the composition section. Fermentation inhibitors—acetic acid, levulinic acid, 5-hydroxymethylfurfural, and furfural—in the pretreatment liquors were also measured using HPLC. Measurement of degradation products like 5-hydroxymethylfurfural from the degradation of hexose sugars and furfural from the degradation of pentose sugars can inform whether degradation of sugars occurred; however, the methods used in this study cannot determine the exact amount of degraded sugars, which could be done by measuring the sugar content in the solids before and after pretreatment. Three ASE cells were extracted for each sample, and the pretreatment liquors were removed for analysis as described above. The remaining pretreated solids for the three samples were then used as the triplicates for subsequent enzymatic hydrolysis. Compositional analysis was not conducted on the pretreated solids, because there was limited solid sample remaining after pretreatment to conduct this analysis. Therefore, enzymes were loaded on a per gram dry biomass basis described in the subsequent paragraph and in Wolfrum et al. (2013).

Enzymatic hydrolysis was conducted according to the procedure described in Selig et al. (2008) at 1.5% solid loading (w/v, on a 105◦C dry weight basis) in a 20 mL scintillation vial with a final reaction volume of 10 mL. Enzymes were added at 40 mg protein/g dry biomass for Cellic CTec2 (Novozymes, Franklin, NC) and 4 mg protein/g biomass for Cellic HTec2. Cellic CTec2 had a protein concentration of 155 mg mL−<sup>1</sup> and a cellulase activity of 99 FPU mL−<sup>1</sup> . HTec2 had protein concentrations of 175 mg mL−<sup>1</sup> and a cellulase activity of 77 FPU mL−<sup>1</sup> . The low percent solids and high enzyme loadings used in the study were intended to determine the maximum sugar release from each sample. Enzyme and substrate blanks were prepared as controls. After an incubation period of 120 h at 50◦C, aliquots of liquor were removed, filtered through a 0.2µm filter, and analyzed for monomeric sugars using Megazyme assay kits (D-Glucose [GOPOD Format] Kit for glucose, D-Xylose Assay Kit for xylose; Bray, Ireland). Sugar yields were calculated by dividing the sugar released in dilute-acid pretreatment and enzymatic hydrolysis liquors by the initial sugar content in the biomass sample.

#### Theoretical Ethanol Yield/Dry Biomass Yield

Theoretical ethanol yields (TEYs) were completed as described in Emerson et al. (2014) using grams of sugar from enzymatic hydrolysis described in the previous Dilute-acid Pretreatment and Enzymatic Hydrolysis section. Total TEY was defined as TEY per hectare of harvested biomass and was calculated as described in Emerson et al. (2014).

# Statistical Analysis

Statistical analysis was done in SigmaPlot 12.3 (Systat Software, Inc., San Jose, CA). A one-way, repeated measures analysis of variance (ANOVA) was used to determine the effect of year (2010, 2012) on components described in previous sections for Miscanthus, the tall fescue mixture, and switchgrass for each biomass type and component separately (n = 3). To determine whether residuals met ANOVA assumptions of normality and homogeneity of residuals, a Shapiro-Wilk test and an equal variance test were done, respectively (p > 0.05). To meet either assumption of normality or homogeneity of residuals, Miscanthus L-Serine was reciprocal transformed and nonstructural xylose and arabinose were natural-log transformed; switchgrass L-Aspartic acid was natural-log transformed; and the tall fescue mixture extracted succinic acid, non-structural galactose, and non-structural arabinose were natural-log transformed. Differences were considered significant if p ≤ 0.05. Non-transformed values were displayed in all tables and figures.

# RESULTS AND DISCUSSION

#### Miscanthus

#### Precipitation and Temperature

Precipitation for Miscanthus grown in Mead, NE, reported by Emerson et al. (2014), totaled 96 cm in 2010 and only 47 cm in 2012. Monthly precipitation in 2010 was on average greater than the 30-year average, but in 2012 the precipitation was lower than the 30-year average for the majority of the growing season, including the harvest month of November. The lack of precipitation, primarily between July and November, was coupled with temperatures greater than the 30-year average in 2012 for most months leading to the drought conditions displayed in **Figures 1A,B**.

#### Impacts on Biomass Composition

Proline, a known compatible solute that accumulates in plants during water stress, was three times greater following drought for Miscanthus, indicating that this crop experienced drought stress (p ≤ 0.05, **Table 1**). Glycine, phenylalanine, and tryptophan levels were also significantly greater in the droughtstressed Miscanthus, which corroborates previous results for drought-stressed switchgrass (Meyer et al., 2014; Ong et al., 2016). The other amino acids analyzed in this study have not necessarily been identified in literature as indicators of drought stress in plants; however, significant changes were seen in these components along with proline, phenylalanine, and tryptophan. Of the 14 other amino acids measured, 10 were significantly greater in 2012 compared with 2010 for Miscanthus (**Table 1**).

Non-structural glucose, another drought stress indicator, was over two and a half times greater for Miscanthus harvested after drought in 2012 compared with Miscanthus harvested in the non-drought year (**Table 2**). Similar changes in non-structural glucose have been reported by Timpa et al. (1986) for cotton plants and Kang et al. (2011) for alfalfa shoots. This trend was also present for fructose, but was not significant (**Table 2**). In addition, extracted glucose following acid hydrolysis, water extractives that include non-structural carbohydrates, and ethanol extractives were 3, 1.5, and 2 times greater, respectively, in the drought year (**Table 3**). The results exemplify how an increase in extractives, including compatible solutes, might affect downstream processing to fuels because there was a 10% higher extractives content and a corresponding 10% lower structural content (i.e., lignin, structural sugars) for Miscanthus in a drought year compared to a normal year.

Non-extractable ash was slightly greater (0.85%) in the 2012 drought compared to the non-drought year, 2010 (p ≤ 0.05; **Table 3**). This trend is not likely to be due to ash from soil as the samples were harvested consistently between years, and most of the soil was removed during the water and ethanol extractions. Ions like potassium help in osmotic adjustment in plant cells (Farooq et al., 2009), and may explain the minor increase in ash, but the elemental composition of the ash was not determined in this study. Additional experiments would need to be done to determine the physiological mechanisms behind these trends.

Miscanthus had significantly lower structural glucan and xylan during the drought year (**Table 3**). This result is similar to that previously reported for corn stover, Miscanthus, switchgrass, and

TABLE 1 | Percent of each amino acid in Miscanthus, the tall fescue mixture, and switchgrass grown in 2010 and 2012 [n = 3; mean (1 SD)].


Asterisks indicate significant differences from a one-way, repeated measures ANOVA comparing 2010 with 2012 for each feedstock separately (p ≤ 0.05). <sup>a</sup>Reciprocal transformed to meet assumptions of equal variance.

<sup>b</sup>Natural-log transformed to meet assumptions of equal variance.

TABLE 2 | Percent of each non-structural sugar in Miscanthus, the tall fescue mixture, and switchgrass grown in 2010 and 2012 [n = 3; mean (1 SD)].


Asterisks indicate significant differences from a one-way, repeated measures ANOVA comparing 2010 with 2012 for each feedstock separately (p ≤ 0.05). <sup>a</sup>Natural-log transformed to achieve assumptions of ANOVA.


TABLE 3 | Percent of each compositional-analysis component for Miscanthus, the tall fescue mixture, and switchgrass grown in 2010 and 2012 [n = 3; mean (1 SD)].

Asterisks indicate significant differences from a one-way, repeated measures ANOVA comparing 2010 with 2012 for each feedstock separately (p ≤ 0.05).

<sup>a</sup>Water extractives includes extractable glucose, xylose, galactose, and arabinose as well as portions of extractable ash and protein.

mixed perennial grasses (Emerson et al., 2014; Ong et al., 2016). The impact of drought stress on plant cellulose and hemicellulose sugars has varied in the literature, however. Drought-stressed plants had decreased cellulose sugars and increased hemicellulose sugars for soybeans (Al-Hakimi, 2006), mixed perennial grasses (Emerson et al., 2014), and Miscanthus (Van Der Weijde et al., 2017). Lower structural sugars have negative implications for potential biofuel yield if it is assumed 100% of the sugars can be converted to fuels; however, rarely are 100% of available sugars converted to fuels. The impact of known recalcitrance factors in biomass must be accounted for. Lignin, a component known to cause recalcitrance (Davin et al., 2008), was also lower for drought-impacted Miscanthus (**Table 3**), which could lead to decreased recalcitrance offsetting the negative impact of reduced structural carbohydrates on conversion yields. Decreased lignin after drought was also found in previous studies of corn stover, mixed perennial grasses, Miscanthus, and corn leaves (Vincent et al., 2005; Emerson et al., 2014). In addition, certain phenylpropanoids (monomeric units of lignin) and peroxidases were increased in maize during drought, indicating decreased structural lignin (Alvarez et al., 2008). Similar to structural sugars, trends from previous studies vary. For example, a study of 50 Miscanthus genotypes reported only a slight decrease in lignin for stem tissue from drought-treated plants compared to the control, and no difference was observed for leaf tissue (Van Der Weijde et al., 2017).

#### Impacts on Conversion

Glucose yields from dilute-acid pretreatment and enzymatic hydrolysis were significantly greater for Miscanthus grown during the 2012 drought compared with 2010, the non-drought year (p ≤ 0.05, **Table 4**). In addition, total xylose yield was greater after drought for enzymatic hydrolysis and pretreatment and enzymatic hydrolysis combined (p ≤ 0.05, **Table 4**), and this trend was present for xylose yield from pretreatment, but it wasn't significant (p > 0.05, **Table 4**). In 2012, on average 0.2 grams of glucose were released during pretreatment and enzymatic hydrolysis per gram of dry biomass, which is over twice the average glucose release in 2010 (0.1 g g−<sup>1</sup> ), and this trend was also present for xylose release (0.1 g g−<sup>1</sup> in 2012 versus 0.08 g g−<sup>1</sup> in 2010). These results are somewhat counterintuitive given the lower structural glucan and xylan in 2012; however, Miscanthus after the 2012 drought had significantly greater extractable glucose and lower lignin than Miscanthus from the non-drought year, which might explain the higher sugar release in 2012 Miscanthus (**Table 3**). In particular, the glucose released in pretreatment for 2010 samples averaged 0.018 g g−<sup>1</sup> (dry biomass), which is similar to the average non-structural glucose in the samples (0.014 g g−<sup>1</sup> ). In 2012, glucose released during dilute-acid pretreatment was 0.06 g g−<sup>1</sup> (dry biomass), mirroring the higher non-structural glucose content in the samples (0.04 g g −1 ). The reactivity experiment was designed to determine differences in reactivity based on consistent enzyme loadings per gram of dry biomass and not per gram of glucan in the pretreated biomass. This caused higher enzyme loadings for samples that had lower glucan contents in the drought years; however, the enzymatic hydrolysis glucose yield almost doubled in the drought year (**Table 4**), but milligrams of Cellic CTec2 per gram glucan only increased by 22% (**Table 5**). Similarly, milligrams of Cellic HTec2 per gram xylan increased by 25% in the drought year,


TABLE 4 | Glucose and xylose yield (%) and release (grams per gram dry biomass) from dilute-acid pretreatment (DAPT), enzymatic hydrolysis (EH), and both DAPT and EH (total) for Miscanthus, the tall fescue mixture, and switchgrass grown in 2010 and 2012 [n = 3; mean (1 SD)].

Sugar yields were calculated by dividing the sugar released in DAPT and EH by the initial sugar content in the biomass sample. Asterisks indicate significant differences from a one-way, repeated measures ANOVA comparing 2010 with 2012 for each feedstock separately (p ≤ 0.05).

TABLE 5 | Enzyme loadings for hydrolysis on a per gram dry biomass basis and per grams of sugar in the pretreated (PT) solids for Miscanthus, the tall fescue mixture, and switchgrass grown in 2010 and 2012 [n = 3; mean (1 SD)].


Asterisks indicate significant differences from a one-way, repeated measures ANOVA comparing 2010 with 2012 for each feedstock separately (p ≤ 0.05).

but enzymatic hydrolysis xylose yield increased by 46% (**Table 4**). The enzyme loadings explain part of the increase in sugar yields, but compositional differences likely play a role as well, given that lignin decreased by 18% in the drought year. This is the first study to show these relationships for dilute-acid pretreated Miscanthus; however, increased degradability of drought-treated Miscanthus has been previously observed when a mild alkali pretreatment was used prior to enzymatic saccharification (Van Der Weijde et al., 2017).

This study did not assess performance of drought-stressed biomass through fermentation, but changes in composition of compatible solutes during drought could impact downstream processing of biomass to fuels. A recent study by Ong et al. (2016) found increased fermentation inhibitors from soluble sugars that degraded into pyrazines and imidazoles in droughtstressed switchgrass pretreated using ammonia fiber expansion. Inhibitors formed by dilute-acid pretreatment are different from those formed in ammonia fiber-expansion pretreatment, but could still form due to the increase in soluble sugars. Acetic acid was the only inhibitor measured in the pretreatment liquors of Miscanthus, and it was significantly reduced by 0.002 g g-1 (dry biomass) in drought stressed Miscanthus (p ≤ 0.05, **Table 6**). Organic acids can cause enzyme inhibition leading to decreased carbohydrate yields (Li et al., 2016), and this may partially explain the lower sugar yields for Miscanthus in 2010. The lack of 5-hydroxymethylfurfural, furfural, and levulinic acid formation in the pretreatment liquors may be due to the low pretreatment temperature of 130◦C (**Table 6**). Understanding the changes in biomass properties due to the environmental conditions during the growing season may affect the selection of pretreatment, enzymatic hydrolysis, and fermentation conditions. More research is necessary to understand how soluble components formed during drought affect enzyme and fermentation-inhibitor formation in the variety of pretreatments used to breakdown the lignocellulosic structure during biochemical conversion processes.

#### Dry Biomass Yield/Ethanol Yield

The theoretical ethanol yield (TEY), based on glucose and xylose released from enzymatic hydrolysis of the dilute-acid pretreated biomass, was 76% greater for drought-impacted Miscanthus compared with biomass grown in a non-drought year (p ≤ 0.05, **Figure 2A**). In contrast, Emerson et al. (2014) actually reported a decrease in TEY following drought; however, this


TABLE 6 | Enzymatic hydrolysis and fermentation inhibitors (grams per gram dry biomass) measured in the dilute-acid pretreatment (DAPT) hydrolysates for Miscanthus, the tall fescue mixture, and switchgrass grown in 2010 and 2012 [n = 3; mean (1 SD)].

Asterisks indicate significant differences from a one-way, repeated measures ANOVA comparing 2010 with 2012 for each feedstock separately (p ≤ 0.05). <sup>a</sup>NP = not present.

analysis was based solely on composition. A decrease in structural carbohydrates was seen in the experiments in the present study, but the TEY is based on sugar yields from reactivity experiments, rather than calculated assuming 100% conversion of the analyzed composition. Average dry biomass yield collected from the field was similar for Miscanthus in 2010 and 2012. This may be due to the fact that only two of the three plots had yields 6 Mg ha−<sup>1</sup> lower in 2012, compared with 2010, as would be expected in drought-stressed plants (**Figure 2B**). It is unclear why one plot yielded 5 Mg ha−<sup>1</sup> greater after the 2012 drought, but there can be significant plot-to-plot variation in field studies. Total TEY (L ha−<sup>1</sup> ), calculated by multiplying TEY and dry biomass yields, can be used to understand the amount of ethanol a hectare of biomass might produce. Miscanthus had 61% higher total TEY following the drought; however, the difference was not significant (**Figure 2C**). This is because of the greater TEY in the drought year and minimal change in dry biomass (**Figures 2A-C**), and if all three replicate plots had lower dry biomass yield after drought the difference in total TEY between the 2 years would have been noticeably less.

# Tall Fescue Mixture

#### Precipitation and Temperature

Precipitation and temperature data for the tall fescue mixture grown in Columbia, MO, was reported by Emerson et al. (2014). Total yearly precipitation in Columbia, MO, was 116 cm in 2010 and 78 cm in 2012. Spring precipitation was greater than the 30-year average in 2012, but for every month from May through December, the precipitation was below the 30 year average and was also below 2010 monthly precipitation, except during October. The 2012 average maximum temperature was greater than the 30-year average from January through September while the 2010 average temperature was similar to the 30-year average. The sampled biomass was harvested in late October or early November, and the lack of precipitation and elevated temperatures are consistent with the drought conditions in **Figures 1C,D**.

#### Impacts on Biomass Composition

The tall fescue mixture had proline levels twice as high following drought, indicating that the crop was impacted by drought conditions (p ≤ 0.05, **Table 1**). Other amino acids—glycine, phenylalanine, and tryptophan—also had significantly greater levels in the tall fescue mixture after drought, which has been observed previously for drought-stressed switchgrass (Meyer et al., 2014; Ong et al., 2016). Eleven of the 14 other amino acids measured were significantly greater in 2012 compared to 2010 for the tall fescue mixture (**Table 1**). Acetic acid significantly increased in 2012 for the tall fescue mixture, corroborating identification of organic acids as chemicals involved in osmotic regulation (Farooq et al., 2009); however, formic acid significantly decreased in 2012 (**Table 7**).

Counter to the proline results, extractives and extractable sugars were similar for the tall fescue mixture harvested after drought in 2012 and in a non-drought year (**Table 3**). Nonstructural sugars were actually higher in the non-drought year; however, this increase was only significant for xylose (**Table 2**). This is in contrast to a previous lab study in which non-structural sugar concentrations in tall fescue were significantly greater after drought (Karsten and MacAdam, 2001), which indicated a plant osmotic response. The difference in these results could be that the tall fescue mixture plots in this study were composed of 48% tall fescue and 44% clover. White clover has been reported to hydrolyze carbohydrates less in response to osmotic stress than tall fescue (Karsten and MacAdam, 2001). The increase in proline concentration best explained white clover response to water stress (Turner, 1990). The different mechanism of response to water stress in a cool-season grass compared with a coolseason forb may explain the significant increase in proline after drought for the tall fescue mixture, combined with no increase in non-structural glucose.

The structural glucan and xylan were significantly lower for the tall fescue mixture during the drought year (**Table 3**). Similar to results for Miscanthus reported by Van Der Weijde et al. (2017), there was no reduction in lignin content in 2012 as has been reported in other studies (**Table 3**) (Vincent et al., 2005; Emerson et al., 2014). This may be due to more moderate drought conditions during 2012 at this site (**Figures 1E,F**).

#### Impacts on Conversion

An increase in glucose yield during 2012 was observed for the tall fescue mixture for enzymatic hydrolysis and total glucose yield (**Table 4**), but these patterns were not significant (p > 0.05). Total xylose yield for the tall fescue mixture was greater after drought for enzymatic hydrolysis and pretreatment and enzymatic hydrolysis combined (p ≤ 0.05, **Table 4**). These

trends were not evident for grams glucose and xylose released from pretreatment and enzymatic hydrolysis per gram dry biomass (p > 0.05, **Table 4**). As stated previously the reactivity experiment was designed to determine differences in reactivity based on consistent enzyme loadings per gram of dry biomass and not per gram of glucan in the pretreated biomass. Milligrams of Cellic CTec2 per gram glucan was 24% greater in the drought year compared with 2010, which is similar to the 17% increase in enzymatic hydrolysis glucose yield, although not significant (**Tables 4**, **5**). Similarly, milligrams of Cellic HTec2 per gram xylan increased by 43% in the drought year, and enzymatic hydrolysis xylose yield increased by 30% (**Tables 4**, **5**). The percent increase in enzyme loading per gram glucan in the pretreated biomass is similar to or greater than the percent increase in sugar yields in 2012 for the tall fescue mixture. In addition, it is theorized that the tall fescue mixture had moderate trends for conversion yields due to the less-severe drought conditions in the county where the tall fescue mixture was grown (**Figures 1E,F**). Finally, fermentation inhibitors−5 hydroxymethylfurfural and furfural—were not present in the pretreatment liquors, indicating that fermentation would not be impacted by these components for the tall fescue mixture grown in either year (**Table 6**). However, the tall fescue mixture did have levulinic acid levels in the pretreatment liquors that were three times higher in samples grown in drought conditions, but this difference was not statistically significant (p > 0.05, **Table 6**).

#### Dry Biomass Yield/Ethanol Yield

There was no significant difference between TEY in 2010 and 2012 for the tall fescue mixture, even though the average TEY did increase 21% (p > 0.05, **Figure 2A**). The dry biomass yields from the tall fescue mixture were 1.6 Mg ha−<sup>1</sup> lower (73%) in the drought year compared with the non-drought year, and this trend affected the total TEY, which was decreased by 67% in the drought year (p ≤ 0.05, **Figures 2B-C**).

#### Switchgrass

#### Precipitation and Temperature

Precipitation in Muskogee County, OK, where the switchgrass samples were grown, totaled 95 cm in 2010 and 77 cm in 2012 (Figure S1). Precipitation in 2012 was lower than the 30-year average from April through December, except for September, and the temperatures were higher than average during this time as well, except for October, leading to the drought conditions shown in **Figures 1E,F**. In 2010, precipitation was much closer to average during the growing season from May through July and in September; however, precipitation was similar to 2012 from September through October. Temperatures in 2010 were similar to the 30-year average for much of the growing season, with higher than average temperatures in June, August, and October.

#### Impacts on Biomass Composition

In the present study, switchgrass compatible solutes were not impacted by drought. Switchgrass proline levels indicate that it did not experience drought stress during the drought


TABLE 7 | Percent of each organic acid in the water extractions of Miscanthus, the tall fescue mixture, and switchgrass grown in 2010 and 2012 [n = 3; mean (1 SD)].

Asterisks indicate significant differences from a one-way, repeated measures ANOVA comparing 2010 with 2012 for each feedstock separately (p ≤ 0.05).

<sup>a</sup>NP = not present.

<sup>b</sup>Natural-log transformed to meet assumptions of ANOVA.

in 2012, as proline concentrations did not differ compared to 2010. All other amino acids measured for switchgrass were significantly greater, approximately 50%, in the nondrought year compared with the drought year (p ≤ 0.05, **Table 1**). In addition, organic acids were, in the cases of acetic and succinic acid, significantly lower after drought (**Table 7**). Switchgrass from 2010 and 2012 had similar nonstructural sugars, water extractives, extracted/acid-hydrolyzed sugar concentration, and ethanol extractives (**Tables 2**–**3**). In general, the cell-wall components measured for switchgrass were not altered during the 2012 drought, but glucan was significantly, by 1.7%, greater in 2012 (**Table 3**), which is a minimal increase given typical analytical error for this measurement (Templeton et al., 2010).

Switchgrass had little to no change in composition as a result of drought, and this may be due to the drought tolerance and low water requirements of this plant species (Lewandowski et al., 2003; Wright and Turhollow, 2010) and the weather patterns in 2010 and 2012. Precipitation was lower and temperatures were higher than average during the drought year in Muskogee County, OK (Figure S1), but the month before harvest, in September, rainfall was greater than average even though temperatures were still elevated. In addition, La Niña triggered drought conditions across the southern U.S. in late 2010 (Rippey, 2015); consequently, the Oklahoma field site had total yearly rainfall of 95 cm, which is less than the 30 year average (106 cm), and had monthly precipitation similar to 2012 in August, September, and October, while May through July precipitation was much lower in 2012 than in 2010. Switchgrass plants have been found to respond well to favorable conditions following drought; therefore, these similar climate patterns near the end of the growing season may be the reason for similar composition in both years, despite the fact that most of Oklahoma was still under drought conditions at the time of harvest in 2012 (**Figures 1E,F**) (McLaughlin and Kszos, 2005).

#### Impacts on Conversion

For switchgrass, pretreated glucose yield and enzymatic hydrolysis glucose yield were similar in 2010 and 2012, but total glucose yield was actually slightly higher in 2010 compared with 2012 (p ≤ 0.05, **Table 4**). This result is consistent with the changes in composition (**Table 3**). The xylose yield was similar for switchgrass grown in 2010 and 2012 (**Table 4**), which is consistent with the lack of differences in glucose yields and structural sugars in both years. There were also no differences in the enzymatic and fermentation inhibitors measured in the pretreatment liquors (p > 0.05, **Table 6**). This matches the composition and conversion results from this study, but inhibitors were increased for drought-impacted switchgrass pretreated with ammonia fiber expansion in a recent study (Ong et al., 2016). The difference likely lies in the timing and severity of the drought conditions experienced by the switchgrass samples as well as the pretreatment type and conditions used in each study.

#### Dry Biomass Yield/Ethanol Yield

There was no significant difference in TEY between 2010 and 2012 switchgrass (p > 0.05, **Figure 2A**). Switchgrass actually had slightly greater dry biomass yields in 2012 compared to 2010, and a higher total TEY following drought because of the higher dry biomass yield in 2012 (p ≤ 0.05, **Figures 2B,C**). This may be due to some level of drought tolerance for switchgrass described previously. Other more drought-tolerant crops, such as soybeans, only had a 9% reduction in yield in the U.S. during the 2012 drought because of their ability to shutdown processes during adverse conditions and reproduce when favorable conditions return (Rippey, 2015). Another contributing factor to the similar yields between 2010 and 2012 is that the non-drought year, 2010, had total yearly rainfall 11 cm less than the 30-year average, and monthly precipitation was similar to 2012 in the 3 months leading up to harvest (Figure S1). Finally, the time when the switchgrass crop was established may also be a contributing factor. The non-drought comparison year was 2010, which was the second year after establishment of the perennial switchgrass crop, and the 2012 drought occurred during the fourth year of the switchgrass crop. Yield of switchgrass generally increases each year until a peak potential is met, and this can often be about 3 years (Parrish and Wolf, 1992; Hong et al., 2014).

Overall, the results of this study indicate that drought conditions have the potential not only to decrease dry-biomass yield, but also to alter biomass chemical composition including components related to recalcitrance. Miscanthus results demonstrate that biomass can have a decrease in structural sugars during drought while simultaneously having lower recalcitrance components, making the remaining sugars more readily available for conversion. These results indicate that the severity of drought conditions may be mitigated by the decrease in recalcitrance for some biomass types, which could offset some of the loss of biomass yield. The tall fescue mixture collected after drought had higher proline concentration, lower structural sugars, and lower dry biomass yield, but minimal significant differences in sugar release following dilute-acid pretreatment and enzymatic hydrolysis in samples collected after drought, possibly as a result of less-severe drought conditions at the field site. Few differences were observed for switchgrass composition and conversion performance, possibly because of its tolerance to drought or the climatic conditions at the field site during 2010 and 2012. Complementary experiments are necessary to confirm the results of this observational study by controlling the drought conditions during plant growth and testing the impact on composition and conversion performance, including fermentation. In addition, more research and development is suggested to understand drought impacts on potential bioenergy crops and strategies to mitigate these impacts.

### AUTHOR CONTRIBUTIONS

AH, RE, AR, and DS formed the research objective. RK, MS, and RF conducted field experiments and collected biomass samples and yield data. AH, RE, AR, DS, SM, DD, and MC coordinated samples and conducted laboratory experiments and data analysis. AH performed statistical analysis and drafted the initial article. RE, AR, and MC contributed section content and edits to subsequent drafts, while all authors reviewed and provided feedstock on the submitted article.

#### REFERENCES


# FUNDING

This research was supported by the U.S. Department of Energy (DOE), Office of Energy Efficiency and Renewable Energy (EERE), Bioenergy Technologies Office (BETO), under DOE Idaho Operations Office Contract DE-AC07-05ID14517. The views expressed in the article do not necessarily represent the views of the U.S. Department of Energy or the United States Government. The U.S. Government retains and the publisher, by accepting the article for publication, acknowledges that the U.S. Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for U.S. Government purposes.

### ACKNOWLEDGMENTS

The authors would like to thank Roch Gaussoin from University of Nebraska and the following INL colleagues for their assistance: Garold Gresham, David Muth, Kastli Schaller, Karen Delezene-Briggs, and Matthew Bryant. The maps displayed in **Figure 1** were courtesy of the U.S. Drought Monitor, which is jointly produced by the National Drought Mitigation Center at the University of Nebraska-Lincoln, the United States Department of Agriculture, and the National Oceanic and Atmospheric Administration. This research was supported by the U.S. Department of Energy under Department of Energy Idaho Operations Office Contract No. DE-AC07-05ID14517.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00054/full#supplementary-material

Datasets used in this article are stored in the Bioenergy Feedstock Library (bioenergylibrary.inl.gov). Samples in the Bioenergy Feedstock Library are identified using a globally unique identifier (GUID). The GUIDs for this dataset are in Table S1.


and thermochemical conversion. Renew. Sustain. Energy Rev. 65, 525–536. doi: 10.1016/j.rser.2016.06.063


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Battelle Energy Alliance, LLC, contract manager for Idaho National Laboratory. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Recycling of Dilute Deacetylation Black Liquor to Enable Efficient Recovery and Reuse of Spent Chemicals and Biomass Pretreatment Waste

Xiaowen Chen<sup>1</sup> \*, Erik Kuhn<sup>1</sup> , Nick Nagle<sup>1</sup> , Robert Nelson<sup>1</sup> , Ling Tao<sup>1</sup> , Nathan Crawford<sup>2</sup> and Melvin Tucker <sup>1</sup>

*<sup>1</sup> National Renewable Energy Laboratory, National Bioenergy Center, Golden, CO, United States, <sup>2</sup> Thermo Fisher Scientific, Lafayette, CO, United States*

#### Edited by:

*Allison E. Ray, Idaho National Laboratory, United States*

#### Reviewed by:

*Muhammad Aziz, Tokyo Institute of Technology, Japan Wenjian Guan, Harvard University, United States*

> \*Correspondence: *Xiaowen Chen xiaowen.chen@nrel.gov*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *09 April 2018* Accepted: *30 May 2018* Published: *19 June 2018*

#### Citation:

*Chen X, Kuhn E, Nagle N, Nelson R, Tao L, Crawford N and Tucker M (2018) Recycling of Dilute Deacetylation Black Liquor to Enable Efficient Recovery and Reuse of Spent Chemicals and Biomass Pretreatment Waste. Front. Energy Res. 6:51. doi: 10.3389/fenrg.2018.00051* Deacetylation/dilute alkaline pretreatment followed by mechanical refining (DMR) has been proven as an effective process for biomass sugar liberation without severe chemical modification to lignin. Previous research has been focused on optimizing deacetylation conditions, reducing energy consumptions in mechanical refining, and improving sugar yields and titers in enzymatic hydrolysis. To successfully commercialize this process, another critical challenge is to develop a robust process to balance water usage, recover spent chemicals, and utilize waste carbons from the dilute deacetylation waste liquor. In this work, a new process modification and strategy is pioneered to recycle and reuse the weak black liquor (WBL) in order to reduce water, chemical, and energy usage while increasing both inorganic and organic contents in the WBLto facilitate downstream processing. Results suggest that the accumulation did not lower acetyl and lignin removal in alkaline pretreatment, resulting in comparable sugar yields in enzymatic hydrolysis. Sodium and potassium were found to be the two most important inorganic compounds in the recycled WBL. Moreover, the accumulated sodium and phenolic compounds did not inhibit the downstream ethanol fermentation processes. Finally, techno-economic analysis (TEA) showed a decrease in the minimum ethanol selling price (MESP) by ∼5 to 15 cents per gallon of ethanol resulting from the inclusion of the recycling of weak black liquor when compared to a conventional non-recycling process.

Keywords: deacetylation and mechanical refining, DMR, black liquor recycling, bioethanol production, minimum ethanol selling price

#### INTRODUCTION

The Deacetylation and Mechanical Refining (DMR) process deconstructs and fractionates biomass into low toxicity, high concentration sugar syrups that have been demonstrated to be highly fermentable to ethanol with high ethanol yields, titers, and productivities (Chen et al., 2016), as well as highly reactive, tractable lignin streams that have been demonstrated to be upgradable to biojet fuel blendstocks and other bioproducts (Laskar et al., 2014; Jeon et al., 2015; Wang et al., 2015, 2017). However, to successfully commercialize alkaline-based pretreatments, including dilute alkali deacetylation, alkali recovery, and recycling are equally critical as the sugar yields in terms of economic feasibility and environmental impacts.

The major role of alkali, primarily sodium hydroxide (NaOH), in deacetylation/dilute alkaline pretreatments is to catalyze the saponification of acetyl groups from hemicellulose and partially delignify the biomass (Chen et al., 2012). The liberated acetic acid will then be neutralized by sodium hydroxide to form sodium acetate, lowering the pH of the deacetylation liquor. In the deacetylation of corn stover feedstocks, ∼10–30% of the solids in the biomass were found solubilized, including 70–80% of the acetate, 2–10% of the xylan, 1% of the glucan, and 20–40% of the lignin, which forms a weak black liquor (WBL) along with spent chemicals (Chen et al., 2012). The WBL is then separated from the biomass by draining and washing potentially recovering the spent chemicals, especially the sodium, to reduce environmental impacts, capitial, operational, and disposal costs.

The recovery of sodium hydroxide from the deacetylation/dilute alkaline pretreatment step can potentially utilize similar equipment and causticizing processes as used in the Kraft pulping industry. **Figure 1** shows the typical Kraft sodium hydroxide recovery process (Tran and Vakkilainnen, 2008). After Kraft pulping, a stream of WBL containing ∼15% total solids including spent chemicals and solubilized biomass is produced as a waste liquor stream from the digester. Then the weak black liquor is concentrated in multi-effect evaporators to more than 65–80% total solids to allow effective burning in a recovery boiler. The heavy black liquor is then sprayed and burned in the lower part of the recovery boiler where sodium salts are formed under an oxygen deficient environment. The formed sodium salt smelt, mainly containing sodium carbonate (Na2CO3) and sodium sulfide (Na2S), are leached in water forming the so-called green liquor, which later reacts with lime (calcium hydroxide) to regenerate sodium hydroxide in the causticizing process. The precipitated byproduct of calcium carbonate is then regenerated back to quicklime (CaO) in the lime kiln and slaked with water to form lime for recycle. The recovery efficiencies for sodium have been reported as high as 97% in the Kraft process (Tran and Vakkilainnen, 2008).

Although the Kraft sodium recovery process is a mature and commercially available process, the direct application of this technology to WBL from the deacetylation/dilute alkaline pretreatment step in a biorefinery industry remains challenging. DMR requires a much lower sodium hydroxide loading compared to the Kraft pulping process (0.4–2 wt% compared to 15–18 wt%) designed to remove most of the lignin and hemicellulose to make white paper. Recent research conducted at the National Renewable Energy Laboratory (NREL) demonstrated that deacetylation of corn stover with 0.5 wt% NaOH (50 kg/ton of biomass) followed by mechanical refining, resulted in a residual solids containing 10% lignin (1/2–2/3 of original lignin content of the biomass) and 30% xylan (90% of original xylan content), and can reach >80% glucose yields in enzymatic hydrolysis with an enzyme loading of 10 mg protein/g of cellulose (Chen et al., 2016). The lower sodium hydroxide loading due to the lower requirements of biomass delignification for deacetylation/dilute alkali pretreatments significantly reduces the energy demand and capacity required for the recovery boiler and energy demand required in the lime kiln. In addition, lower sodium hydroxide loading reduces the hemicellulose degradation due to peeling reaction and thus improve overall sugar yields. The lower loading of sodium hydroxide, lower temperatures used in the DMR process as compared to the Kraft process, results in less organic substances are extracted and/or hydrolyzed in the deacetylation step, leading to more dilute WBLs. In addition, previous reported deacetylation/dilute alkaline pretreatment used a solid-to-liquid ratio of 1:10 followed by an extensive washing step using the same solids-to-water ratio. The large quantity of water usage not only increases the operational costs due to evaporation energy required for recovery and significantly increases the capital and operational expenditures required of wastewater treatment. The dilute WBL from DMR process cannot be direcly combusted in recovery boiler due to the low organic contents/heating value present. Even with recent improvments for an advanced Kraft process, to include gasification of the black liquor and the co-prodution electricity (Naqvi et al., 2012; Darmawan et al., 2018), direct gasification of WBL could be problematic due to the presence of large amount of water. Similar reasons also make direct evaporation of WBL to the solids level suitable for combustion or gasification (≥65%) not economical nor energy efficient.

To provide a solution to address these issues, we pioneered a new process modification and strategy to recycle and reuse the WBL to reduce water, chemical, and energy usage while increasing the solids, sodium, acetate, sugars, and lignin contents in the WBL. **Figure 2** shows a schematic diagram of the black liquor recycling scheme. WBL from the first batch of deacetylation is separated from the biomass and additional water and sodium hydroxide are added for that required for deacetylation of the second batch of fresh corn stover biomass at the 1:10 solids-to-liquid ratio. The same procedure is repeated for the nth time (here in this stud biomassy n = 6) until the thickened black liquor (TBL) is purged. Meanwhile, a clean water stream is used to wash the deacetylated biomass in a counter-current

reverse batch-wise sequence. The washing is not performed once the soluble solids in the wash liquor is equivalent to the total solids of the deacetylation liquor at the ith stage (here i = 1), at which point the water was recycled five times until the final total soluble solids concentration reached ∼9 wt%.

As shown in **Figure 2**, WBL from the upstream deacetylation step is used as a base liquor to extract a new batch of biomass so that the extracted organic substances and spent sodium will accumulate. The accumulation of both chemicals and organic matter enables efficient chemical recovery, but could potentially reduce the acetyl and lignin removal, leading to significantly lower enzymatic digestibility of the extracted biomass. Our results suggest otherwise. The accumulated sodium and phenolic compounds could also inhibit the downstream fermentation and catalytic upgrading processes. Again, our results suggest otherwise. Therefore, in this study, we investigated the effects of recycling of the weak black liquor from each batch stage on acetyl and lignin removal, biomass digestibility, enzymatic hydrolysis yield, and hydrolyzate fermentability. The hydrolyzate fermentability was investigated using ethanologen (recombinant Zymomonas13-H-9-2). Finally, the process and economic benefits of recycling weak black liquor were simulated using Aspen plus and economic models.

# METHODS

## Feedstock

Corn stover harvested in 2009 in Hurley, South Dakota, and transported to Idaho National Laboratory (INL) was hammer milled and stored indoors. Upon receipt at NREL in 2013, the corn stover was further knife milled (Jordan Reduction Solutions, model 14 × 20, Birmingham, Alabama) to allow passing through a rejection screen with 19 mm (3/4-inch) round holes and stored indoors in 200 kg lots in super sacks at room temperature.

# Bench-Scale Deacetylation and Black Liquor Recycling

Corn stover deacetylation was performed in a 90-L paddle reactor following the sequence described in **Figure 1**. The individual operation as described earlier (Chen et al., 2012). Dry corn stover (3.0 dry kg) was added to the reactor along with 30 kg of white liquor or 30 kg of recycled black liquor with makeup water and sodium hydroxide. The white liquor was made by mixing 0.165 kg sodium hydroxide pellets with 29.6 kg of tap water (0.3 kg of additional water came in with the biomass). The 30 kg of black liquor in the WBL from a previous deacetylation batch run was mixed with makeup water and 0.165 kg of sodium hydroxide. Because the pH of the spent deacetylation weak black liquor was near neutral, indicating the NaOH was titrated by the acetic acid released, the same quantity of sodium hydroxide was required to be added in the makeup white liquor. The WBL was drained through an 8-inch screen (∼2 mm wire spacing) followed by pressing using a Vincent screw press (Vincent Corporation, model CP-4, Tampa, Florida). The makeup water was calculated based on the amount of WBL obtained, which is further discussed in the Results and Discussion section. The slurry for the next ith batch deacetylation step was heated to 80◦C and held for 2 h, and then the weak black liquor was allowed to drain overnight through an 8-inch screen located in the bottom port of the paddle mixer.

The screw-pressed deacetylated corn stover was washed following the sequence described in **Figure 2**. Batch 6 deacetylated corn stover solids were washed with 30 kg of fresh water in the paddle reactor for 30 min. The wash water was drained through the 8-inch screen and the slurry is further dewatered again through the Vincent screw press. The wash water was collected and weighed. Approximately 7 kg of fresh makeup water for each batch washing step was used to achieve a total of 30 kg of washing liquor for each batch of deacetylated biomass, keeping the solids-to-liquid ratio of 1:10. The washing was not performed on batch 1 since the solids of the recycled washing liquor was equivalent to the solids in the black liquor exiting from batch 1, and no further washing is expected to take place.

# Szego Milling

The refining with the Szego mill was, previously described (Chen et al., 2014), performed at a measured rotational speed of 1,160 rpm. The biomass was fed at a rate of ∼100 kg/h with approximate solids of 10%. The refining energy required for the first pass through the Szego mill was measured at ∼70 kWh/oven dried metric ton (ODMT) using a Fluke model 1,735 Power Logger (Fluke Corporation, Everett, Washington), while the second pass through the Szego mill was measured at ∼30 kWh/ODMT. Therefore, a total energy input of ∼100 kWh/ODMT was used in the two-stage refining step. The twostage refined solids were loaded into close-mesh laundry bags and dewatered using a GE washing machine on the spin cycle prior to enzymatic hydrolysis.

#### Enzymatic Hydrolysis

Prior to enzyme hydrolysis, the deacetylated and mechanically refined corn stover substrates were slowly and carefully titrated with ammonium hydroxide to a pH of 5.3 in a 30-liter industrial dough mixer. Then 4 kg of the pH adjusted substrate was autoclaved at 121◦C for 30 min. The cellulase/hemicellulase enzyme cocktails and additional water were added to the final required total solids concentrations at 20 wt% solids. The hydrolysis were carried out in 9-liter stainless steel roller bottle vessels with a mixing speed at 19 rpm. About 1,000 g of biomass substrate at the required solids loadings were.enzymatically saccharified using Novozymes Cellic CTec3 at a loading of 16 mg/g of cellulose along with Cellic HTec3 at 4 mg/g of cellulose at 50◦C for 120 h.

#### Ethanol Fermentation

To test the toxicity of the effect of residue salts in biomass on subsequent ethanol fermentation, the NREL engineered strain, Zymomonas mobilis 13-H-9-2 (Chen et al., 2016), was selected for this work. All fermentations were carried out at 300 ml working volumes in BioStat-Q Plus fermenters using enzymatic hydrolyzed DMR substrates. No extra washing of the substrates was performed. The fermentation conditions were as follows: temperature at 33◦C, pH at 5.8 (controlled with 4N potassium hydroxide [KOH]), and agitation speed at 300 rpm. Ethanol yields were calculated using the methods mentioned elsewhere(Mohagheghi et al., 2004).

#### Analytical Methods

All analytic methods used in this work were following the NREL standard Laboratory Analysis Procedure (LAP) including solids compositional analysis by LAP No. NREL/TP-510-42627 (Sluiter et al., 2008a,b), and soluble sugars, acetic acid, and degradation product determined by LAP No. NREL/TP-510-42623 (Sluiter et al., 2008a). Pretreated biomass insoluble solid concentrations were determined by a six-step washing and centrifugation procedure (Schell et al., 2003).

# Viscosity Measurement

Rheological experiments were conducted using a Bohlin Gemini HR nano stress-controlled rheometer (Malvern Instruments, Westborough, Massachusetts) equipped with a 40 mm, 4◦ stainless steel cone-and-plate geometry. Temperature control of ±0.1◦C was provided by the bottom, Peltier effect plate. The truncation gap height of the cone was specified at 150µm.

The viscosity of the black liquors was investigated as a function of temperature at 25, 50, 75, 90, and 100◦C (temperatures above 100◦C were explored, but the liquors began to boil and the resulting viscosity measurements were erroneous). Initially, liquor (sample volume of ∼1.5 mL) was loaded onto the rheometer at 25◦C and allowed to equilibrate for 2 min before commencing rheological measurements. After equilibration, the sample was preheated at 100 s−<sup>1</sup> for 1 min. Then data was collected every 10 s at a constant shear rate of 100 s−<sup>1</sup> for a total of 2 min. After shearing at 25◦C, the temperature was increased to 50◦C and again allowed to equilibrate for 2 min. The sample was then presheared and sheared at 100 s−<sup>1</sup> , collecting data every 10 s over a 2 min duration. At every new temperature, the sample was equilibrated for 2 min and the same preshear and shear procedure described above was followed. Experiments were repeated in duplicate to verify reproducibility.

# RESULTS AND DISCUSSION

# Effect of Recycling Black Liquor on Water Consumption

**Table 1** shows the water consumption in this black liquor recycling study. For batch 1, the deacetylation white liquor was made by adding 0.165 kg of sodium hydroxide to 29.6 kg of water and mixing. Approximately 0.3 kg of water was bound in the starting biomass charge. The white liquor was used to deacetylate 3.3 kg of corn stover (3.0 kg dry). After deacetylation, 23.7 kg of the weak black liquor was recovered and 5.4 kg of deacetylated corn stover at ∼50% solids was recovered. Therefore, in batch 2, 6.3 kg of makeup water and 0.165 kg of makeup sodium hydroxide was added to the weak black liquor from batch 1, because the pH was measured at 6.43 suggesting that all of the sodium hydroxide was neutralized by the acetic acid saponified during the deacetylation step. This deacetylation process was repeated 6 times in the current study and a total of 59.7 kg of water was used generating/regenerating the deacetylation white liquor.

In addition, wash water was used in a reverse counter-current direction to wash out the spent chemicals and dissolved organic substances starting with batch 6, as shown in **Figure 2** (also known as counter-current washing). The initial washing water added to batch 6 was 30 kg including water entrained in the biomass. The wash liquor was drained and further separated from biomass using a screw press to ∼50 wt% solids. This method is widely used in the pulp and paper industry, known as dilution/extraction. The washing efficiency is highly dependent on pulp consistency after pressing. A makeup clean wash water


was added to guarantee 30 kg of washing liquor in the next batch of washing. The washing was stopped at batch 1 because the washing liquor contained a soluble solids content of 2.3%, which was equivalent to the soluble solids content in the weak black liquor produced from batch 1. The total wash water added in all the batches was 58 kg.

Therefore, the total water consumption in this study is 118 kg for 18 kg of dry biomass, resulting in a water-to-biomass ratio of 6.5:1 (wt/wt). This ratio is much lower compared to the conventional single-stage batch deacetylation method used in past publications, which is ∼20:1 (wt/wt). This recycling strategy can be optimized with displacement washing, with lower water usage and improved washing efficiency, or by employing an inclined, continuous screw counter-current deacetylation with improved washing strategies.

This recycling strategy is found arguably better than the high solids deacetylation steps carried out at 30% total solids (Shekiro et al., 2016). In that study, using a 3:1 water-to-biomass ratio in the single-stage deacetylation step and a similar 3:1 ratio in the single-stage washing step, the total water-to-solid ratio is 6:1. However, the higher solids deacetylation scenario requires energy-intensive agitation, and difficult separations, and may not be practical in commercial-scale production.

#### Effect of Recycling Black Liquor on Deacetylation Process

Our goal for recycling of the deacetylation liquor is to increase the solids to 10–15% total solids to allow efficient evaporation and reduce energy consumption in the concentration of the black liquor. The weak black liquor from Kraft pulping contains ∼15% total solids—mainly lignin, degraded hemicelluloses (peeling reactions), and spent chemicals (sodium). The soluble solids of a typical weak black liquor from our deacetylation pretreatment is ∼2–3%. **Figure 3** shows the effect of black liquor recycling on total solids and soluble solids. Total solids contain soluble solids and insoluble biomass fines that are collected in the drain liquor and squeezate liquor from the screw press. The total solids increased from 3.5 to 10.3%, while soluble solids increased from 2.4 to 8.9%., Both total and soluble solids did not linearly increase with the increasing number of recycles. For the first three runs, the soluble solids increased by ∼2%, while the solids increased by an average 0.8% for the last three runs. The decreased rate of accumulation of soluble solids shows the dissolution reaction of biomass is inhibited by the accumulated dissolved components. We hypothesize three things. First, deacetylation and certain delignification reactions are reversible reaction under alkaline conditions. The accumulation of the reaction products by recycling black liquor will eventually reach the equilibrium point, thus stopping the reactions. Second, sugar dissolution is controlled by the solubility, especially for the xylooligosaccharides. We postulate that during deacetylation, some xylan is solubilized as xylooligosaccharides (measured by HPLC for monomeric and total sugar assays the soluble sugars are primarily oligosaccharides) and eventually reaches the solubility limitations, thus reducing xylan losses in subsequent deacetylation batches. Third, the dissolved lignin

and hemicellulose can be adsorbed by the biomass, thus affecting enzymatic hydrolysis. Due to electrostatic forces, the dissolved lignin-carbohydrate complexes compounds could precipitate on biomass surfaces, especially at the end of the deacetylation when the pH is reduced to near neutrality (Ban and van Heiningen, 2011). According to the Langmuir adsorption equation, the adsorption rate increases at a higher concentration of solids, resulting in lower soluble solids in the liquor phase.

**Figure 4** shows the accumulation of solids in the wash liquor following batch-wise counter-current washing. Because the added wash water washes the deacetylated biomass in a counter-current direction, batch 6 wash liquor contains the lowest soluble solids. And the soluble solids increased until the washing liquor cycles to batch 4 followed by a slight decrease due to makeup washing water and lower solids content in deacetylated black liquor in batch 3 and 2. The highest solids in the washing liquors occurs around batch 4. The washing is stopped at batch 1 where the soluble solids of the black liquor from batch 1 is only 2.4%, equivalent to the soluble solids in the wash liquors out of batch 2 at 2.3%.

**Figure 5** shows the alkali and alkaline earth metal accumulations in the black liquors as a function of recycle. The sodium concentration increases from 3.17 to 12 g/L. As can be seen in **Table 1**, the first batch of deacetylation is conducted with 0.165 kg sodium hydroxide in 30 kg of water, resulting in 3.17 g/L of sodium. Therefore, the analytical result from the inductively coupled plasma analysis for the first batch corresponded to the initial sodium concentration that had been loaded. The later recycle batches where limited amounts of makeup water were added due to absorption of water by the biomass decreased the amount of dilution of the sodium. Therefore, the sodium carried over from the last batch was diluted in the wash process, showing an incremental increase in the rate of sodium accumulation at ∼1.6 g/L/recycle.

Potassium accumulation in the deacetylation weak black liquor is derived from extracting the soluble ash components of biomass. For every batch recycle, ∼0.36 g/L of potassium was extracted. The total amount of potassium extracted is ∼3.6 g/kg of corn stover. The dissolution rate of potassium is not affected by the number of recycles. Potassium is a possible valuable byproduct of the biorefinery, where it can be added back to the soil as an essential fertilizing agent to supplement the world's dwindling supplies of potassium. Calcium is another soluble ash component dissolved from biomass in the deacetylation step. In this study the dissolution and accumulation of calcium is low, with an average calcium accumulation of 0.06 g/L, because most of the calcium is tied up in the insoluble structural ash component in the corn stover feedstock.

**Figure 6** shows the accumulation of sugars, lignin, and acetic acid in the black liquors with most of the soluble sugars in the oligomeric form. The dissolution and accumulation of glucan is the lowest among all the sugars solubilized, with the final glucan concentration of 0.9 g/L achieved after recycling the weak black liquors six times. The accumulation of glucan is also linearly correlated with the number of batch recycles. Soluble xylan shows the highest sugar concentrations at ∼6 g/L after the final recycle batch, while arabinan and galactan accumulate to 5 and 2 g/L, respectively. All three hemicellulose-derived oligomeric sugars

are not linearly correlated to recycle batch number, especially when the last three recycles were performed. We hypothesize that xylan, arabinan, and galactan are all polysaccharides with limited solubility. The dissolution of soluble sugars slows down with an increased number of recycles. At a higher number of recycles, the adsorption of dissolved hemicellulose onto solid biomass starts to take effect with increasing hemicellulose concentrations in the solution. The reduced dissolution of hemicelluloses with each new batch at increased black liquor recycle batch number is beneficial in preserving biomass sugars for downstream sugar utilization.

The accumulation of lignin is also shown in **Figure 6**, displaying a linear correlation between lignin concentration and the number of recycles. The lignin concentration was found to be as high as 20 g/L at the final batch, suggesting that lignin solubilization is not significantly affected by the number of recycles in our strategy. A similar linear trend is also found for acetate accumulation, resulting in ∼15 g/L of acetic acid found in the black liquor from the final sixth recycle batch. The linear correlation between the number of recycles and the accumulation of lignin and acetate in the recycle black liquor shows that black liquor recycling has little impact on acetate and lignin removal under the current experiment conditions. By increasing the number of WBL recycles, the lignin and acetate can increase in concentrations without significantly affecting the removal of lignin and acetate.

**Figure 7** shows the correlation between sodium and organic matter accumulation. Lignin and acetate accumulation is roughly linearly correlated with sodium accumulation, with correlation coefficients (R 2 ) equal to 0.9938 and 0.9899, respectively. For the dissolved sugars cases, the data indicate that sugar dissolution is more controlled by oligomer solubility than sodium concentration. It's also interesting to note that the acetic acid to sodium (w/w) ratio was found to be 1.22:1, suggesting a molar ratio of acetic acid to sodium of ∼1:2, indicating there

FIGURE 6 | Effects of weak black liquor recycling on the accumulation of sugars and acetic acid in the recycle liquors.

were other sodium salts formed in addition to sodium acetate formed by saponification of the acetyl groups and by the neutralization by sodium hydroxide in the deacetylation black liquor.

**Figure 8** shows the biomass component dissolution for every single stage of the deacetylation. The dissolution was calculated based on each component in the native biomass. The dissolution of glucan was ∼0.5–0.8% only for soluble glucose and sucrose and only the amorphous cellulose was dissolved during deacetylation. Xylan dissolution increased from 6 to 10% in the first two recycle batches and gradually decreased to 0.7% loss in the sixth batch. The dissolution of arabinan also displayed an initial increase from 35 to 49% in the first two batches followed by gradually decreasing from 44 to 27% in the next four batches. The increase of sugar dissolution from batch 1 to batch 2 was possibly due to the residual alkaline in the batch 1, which increased the initial and final pH of batch 2 leading to a increased sugar dissolution. The decrease from recycle batch 3 to recycle batch 6 suggests the sugar dissolution was inhibited by solubility and potentially reabsorption of the xylan polysaccharides (Fengel and Wegener, 1984). Lignin dissolution averaged 35% with a standard deviation of ±5%. The yields of acetic acid dissolution were all more than 120%. This >100% acetate removal shows a potentially underestimated acetyl group component from the analysis of the native biomass (Sluiter et al., 2008a). Alkaline conditions are more effective at saponifying and removing acetyl groups from biomass compared to acid hydrolysis of the ester functional groups. The scattered results of acetic acid were also caused by baseline interference raised by other co-eluted organic acids.

#### Effects of Recycling Weak Black Liquor on Sugar Yields During Enzymatic Hydrolysis

After deacetylation, corn stover was mechanically refined with two passes through a small Szego mill to optimize the enzymatic hydrolysis yields. In the past, we have obtained enzymatic hydrolysis yields close to 90%. However, the deacetylated corn stover in the previous work was refined using a 36′′ pilot-scale disc refiner, or a combination of the disc refiner and Szego

mill (Chen et al., 2014). We found that a significant amount of unreacted hard kernels in the pretreated corn stover that were unable to be disintegrated using a double pass through the Szego mill. The enzymatic hydrolysis using the unbroken deacetylated hard kernels showed only 33% glucose and 31% xylose yields after 7 days (data not shown here). Therefore, the presence of the hard kernel reduces the overall yields of sugar in enzymatic hydrolysis experiments. The residual hard kernels after high solids enzymatic hydrolysis are shown in **Figure 9**, as it is apparent they were not touched by mechanical refining or enzymatic digestion.

The effects of recycling WBL from deacetylation on sugar yields are shown in **Figure 10**. The glucose and xylose yields in enzymatic hydrolysis were in the range of 68–78% and 64– 74%, respectively. The sugar yields were all 10–15% lower than those previously reported for the multistage mechanical refining. As discussed above, the lower yields were mainly caused by the

FIGURE 9 | The undigested hard kernels after enzymatic hydrolysis of DMR corn stover feedstock.

undigested hard kernels that were not refined in the double pass through the Szego mill. Interestingly, these kernels were found to be highly digestible if they were first refined using a disc refiner, followed by refining in the Szego mill. We speculate that size reduction of all biomass anatomical fractions is needed to improve sugar yields.

Besides the lower yields caused by undigested corn kernels, recycling the weak black liquors did show some negative effects on sugar yields. The highest glucose yield occurred at batch 1 where clean white liquor was used and no washing was performed. The second highest glucose yield occurred at batch 6 where washing with clean water was performed. The lowest glucose yields occurred for batches 2, 3, 4, and 5, showing almost the same glucose yields across all batches with insufficient washing. The glucose and xylose yields for batches 2–6 followed a reversed trend of dissolved lignin in the surrounding liquor phase of the DMR corn stover slurry prior to enzymatic hydrolysis, showing the dissolved lignin after deacetylation will have a negative impact on sugar yields when it is carried over to the enzymatic hydrolysis stage. However, batch 1, where no washing is available, does not follow this hypothesis indicating the sugar yields are affected by other unknown reasons.

#### Effects of Recycling Black Liquors on Ethanol Fermentations

Sodium is one of the critical inhibitors for Zymomonas fermentation, as this microorganism has low tolerance to inorganic ions (Vriesekoop et al., 2002). The sodium from deacetylation will not be carried over to enzymatic hydrolysis and fermentation if one can use unlimited amounts of wash water to achieve very high washing efficiencies. However, it's not realistic in industrial applications due to the costs of both fresh water and wastewater treatment. A compromise is needed between washing efficiency and the amount of wash water used; the former directly affects the sodium content in biomass hydrolyzates. **Figure 11** shows the metal ions in the biomass hydrolyzate prior to fermentation. Sodium content is around 1.8 g/L in batch 1 hydrolyzate, which later reaches a maximum concentration in batch 3 at ∼1.9 g/L. The lowest sodium content is found in batch 6 at ∼1.4 g/L because it is washed with fresh water. The amount of sodium in the biomass hydrolyzate liquors shows ∼20% sodium loss based on initial sodium loading due to low efficiency washing.

Potassium is the second largest inorganic ion in the hydrolyzate liquors and keeps almost constant concentration of ∼0.4 g/L. Calcium, magnesium, and iron are all in lower concentrations ranging from ∼0.1 to 0.3 g/L.

The effect of WBL recycling on ethanol fermentations using rZymomonas is shown in **Figure 12**. Ethanol process yields were in the range between 85 and 92%, showing the current sodium content has limited impact on ethanol yields. However, due to the relatively low sugar concentrations in the hydrolyzate liquors, the final ethanol titers were all around 60 g/L (data not shown). It is not clear if this level of sodium will affect ethanol yield at higher sugar/ethanol concentrations. In addition, further investigation is needed for phenol-introduced inhibition, as it is reported that lignin-derived phenolic compounds are strong inhibitors to Acetone-Butanol-Ethanol (ABE) fermentation (Guan et al., 2018).

#### WBL Thickening and Its Effect on Viscosity

In the current NREL design case of hydrocarbon fuel production from biomass, the lignin content of the biomass is eventually combusted in a boiler to generate heat, power, and steam to supply the biorefinery process (Biddy and Jones, 2013). The lignin source from the DMR process came from two core unit operations: the deacetylation/dilute alkaline pretreatment and the solid liquid separation of enzymatic hydrolyzed biomass slurry. The deacetylation black liquor contains up to 30–40% of the original content of lignin in the biomass. Two strategies to increase lignin content in the WBL: Increasing the severity of alkaline pretreatment to release more lignin, increasing the lignin content in the black liquor could be much higher. Secondly, concentration of the solids in the WBL from the

deacetylation/alkaline pretreatment enabling the possibility of black liquor combustion or gasification.

There are two critical properties of the concentrated black liquor affecting the operation of the recovery boiler: (1) viscosity and (2) boiling point rise (BPR). In the Kraft process, the thick black liquor shows non-linear behavior with viscosity and BPR. If the WBL from the Kraft process is concentrated to ∼70% total solids, the viscosity of the TBL has been reported to be as high as 88 centipoise (cP) even at 127◦C. The high viscosity of the concentrated black liquor requires liquor heat treatment to lower the liquor viscosity and ease the operation.

In a biorefinery process, the biggest challenge of using black liquor from the deacetylation/alkaline pretreatment step is the high content of hemicellulose sugars in herbaceous biomass, which may dissolve in high yields during dilute alkali pretreatment, and result in a high viscosity concentrated black liquor. However, these soluble sugars make the black liquor hard to flow at higher solids. In this research, we investigated the viscosity of concentrated WBL from the deacetylation and recycling process to understand its potential impact on black liquor utilization and combustion. The thin black liquor from batch 6 (∼10 wt% solids) is concentrated in a rotary evaporator to a final solids concentration of ∼65% total solids, which is required as the minimum solids content for combustion in a recovery boiler.

**Figure 13** is a photograph of the visual observation of the concentrated black liquor (∼65% solids) at room temperature, indicating that the concentrated black liquor is flowable, but very viscous compared to water.

**Figure 14** shows the effect of temperature on the viscosity of the concentrated black liquor from the sixth recycle experiment at ∼65% total solids. At room temperature, the black liquor has a viscosity of 2,000 cP, which decreases to 154.8 cP at 100◦C. Higher temperature is also applied to TBL in an attempt to compare with the literature-reported viscosity of pulping black liquor at 127◦C. However, due to evaporation of the TBL using the atmospheric pressure cell, we were unable to accurately measure the viscosity of the TBL above 100◦C. Therefore, a pressurized cell will be used in the future.

#### Effect of WBL Recycling on Minimum Ethanol Selling Price (MESP)

Recycling of the dilute alkali deacetylation black liquor opens the door for efficient utilization of the waste lignin and acetate stream to produce value-added products. More importantly, recycling dilute alkali deacetylation black liquor reduces water and energy usage and thus reduces the production costs. Four potential scenarios were modeled in this report including: (1) deacetylation at 10% solids without washing and recycling; (2) deacetylation at 10% solids with washing and recycling; (3) deacetylation at 30% solids with washing but without recycling; and (4) deacetylation at 30% solids without washing and recycling. **Table 2** summarizes water and energy usage as well as final MESP of the four different scenarios calculated from the model.

Dilute alkali deacetylation at 10% solids has its pros and cons. The advantage of dilute alkali deacetylation under low solids conditions is that it requires much less mixing energy and is easier to be realized at commercial scale in a batch stirred tank reactor. On the other hand, the low solids process requires larger sizes of reactors, higher water usage, and higher steam usage. Deacetylation at 30% solids, however, is difficult to scale up to the industrial scale due to high requirements for mixing energy. In addition, deacetylation at 30% solids without washing will not be able to separate the dissolved acetate and lignin from the slurry solids, leading to lower sugar and product yields. Therefore, recycling the dilute black liquor is more likely to be practically implemented at the commercial scale.

As shown in **Table 2**, deacetylation at 30% solids with washing but without recycling has the highest water usage. At a 2,000-ton (dry) biomass/day plant, the water used in deacetylation stage is ∼750 ton/h. If deacetylated at 10% solids without washing and

FIGURE 13 | Visual observation of thickened black liquor.

TABLE 2 | Effect of recycling and washing on water and energy usage and MESP.


recycling, the water usage is similar to this amount. In another extreme case, when deacetylation occurs at 30% solids without washing, the water usage is less than half, showing ∼360 ton/h usage. As discussed earlier, the no washing strategy will cause lower yields in enzymatic hydrolysis and fermentation due to the inhibition effects of acetate and lignin. However, if the dilute alkali deacetylation black liquor is recycled, the water usage for deacetylation at 10% solids with washing is only slightly higher than that of the 30% no washing case. The recycling saves as high as 45% of the water as compared to the no recycling process.

The recycling of dilute black liquor also significantly reduces the energy usage. As the black liquor recycles, the heat energy is conserved in the majority of the deacetylation liquor without loss or forced cooling. In non-recycling cases, the steam energy to heat up a large quantity of water from 25 to 80◦C is mostly wasted during the washing step going to wastewater treatment. The energy saved by recycling the dilute alkali black liquor could be as high as 75%.

To have a fair comparison for the effect of recycling on economics, the sugar and ethanol yields are fixed in the four cases described above. Thus, MESP is not affected by revenue but solely affected by the operational cost. As shown in **Table 1**, the highest MESP is at 10% deacetylation with no washing and recycling, while the lowest MESP is at 10% deacetylation with washing and recycling. The cost savings calculated by the model range from 5 to 15 cents per gallon of ethanol. This result indicates that recycling of the dilute alkali black liquor could make the deacetylation process more economical.

# CONCLUSIONS

In summary, the recycling of the WBL in the DMR process increases the concentrations of extracted components from corn stover biomass, making the black liquor from deacetylation a more valuable stream with upgradable lignin, acetate, and sugars. The increased sodium concentration in TBL also makes sense for sodium recovery through a causticization process similar to the Kraft process. In addition, the recycling strategy shows minimal impacts on downstream fermentation, indicating that the microorganisms used in current ethanol and proposed hydrocarbon production processes can tolerate the level of sodium and phenolic compounds in the recycled DMR hydrolyzate liquors where the sodium and lignin are carried over by incomplete washing during the recycling of the WBL. Moreover, the low viscosity of the concentrated black liquor shows that the recycled and evaporated deacetylation black liquor has similar rheological properties compared to pulping black liquor, thus could be pumped and processed at high solids concentrations. Finally, techno-economic analysis shows the recycling of weak black liquor could make the deacetylation process more economical.

# AUTHOR CONTRIBUTIONS

XC designed and conducted the pretreatment experiment and drafted the manuscript. EK and NN help conduct the pretreatment and enzymatic hydrolysis experiment. RN conducted the fermentation. LT conducted TEA analysis. NC conducted the viscosity analysis. MT reviewed and revised the manuscript.

# ACKNOWLEDGMENTS

This work was authored by Alliance for Sustainable Energy, LLC, the manager and operator of the National Renewable Energy Laboratory for the U.S. Department of Energy (DOE) under Contract No. DE-AC36-08GO28308. Funding provided by U.S. Department of Energy Office of Energy Efficiency and Renewable Energy Solar Energy Technologies Office. The views expressed in the article do not necessarily represent the views of the DOE or the U.S. Government. The U.S. Government retains and the publisher, by accepting the article for publication, acknowledges that the U.S. Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for U.S. Government purposes.

# REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Chen, Kuhn, Nagle, Nelson, Tao, Crawford and Tucker. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Pre-senescence Harvest of Switchgrass Inhibits Xylose Utilization by Engineered Yeast

Rebecca G. Ong1,2 \* † , Somnath Shinde3,4†, Leonardo da Costa Sousa<sup>4</sup> and Gregg R. Sanford5,6

*<sup>1</sup> Department of Chemical Engineering, Michigan Technological University, Houghton, MI, United States, <sup>2</sup> DOE Great Lakes Bioenergy Research Center, Michigan Technological University, Houghton, MI, United States, <sup>3</sup> Department of Chemical and Biomolecular Engineering, University of Tennessee, Knoxville, TN, United States, <sup>4</sup> Department of Chemical Engineering, Michigan State University, East Lansing, MI, United States, <sup>5</sup> Department of Agronomy, University of Wisconsin-Madison, Madison, WI, United States, <sup>6</sup> DOE Great Lakes Bioenergy Research Center, University of Wisconsin-Madison, Madison, WI, United States*

#### Edited by:

*Allison E. Ray, Idaho National Laboratory, United States*

#### Reviewed by:

*Yong Xu, Nanjing Forestry University, China Jinxue Jiang, Washington State University, United States*

> \*Correspondence: *Rebecca G. Ong rgong1@mtu.edu*

*†These authors have shared co-authorship for this article.*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *11 April 2018* Accepted: *30 May 2018* Published: *19 June 2018*

#### Citation:

*Ong RG, Shinde S, da Costa Sousa L and Sanford GR (2018) Pre-senescence Harvest of Switchgrass Inhibits Xylose Utilization by Engineered Yeast. Front. Energy Res. 6:52. doi: 10.3389/fenrg.2018.00052* Proper timing of switchgrass harvest for bioenergy is important to maximize yield and optimize end use conversion. Proposed windows range from peak biomass to the following spring after overwintering in the field. There are various pros and cons associated with harvest timing: earlier harvests maximize yield but can remove nutrients from the field that may require replacement, while later harvests have reduced biomass yields due to weathering but maximize nutrient resorption in belowground tissues. Switchgrass composition changes during the harvest period, with losses of potential fermentation nutrients (amino acids and minerals), and sources of pretreatment-derived inhibitors (soluble sugars), which could affect downstream conversion by microorganisms. For this work we investigated whether switchgrass harvest could be timed to maximize beneficial impacts on fermentation. Switchgrass samples were harvested from five replicate field plots in Wisconsin, roughly every 2–3 weeks from peak biomass (Aug. 20) until after the killing frost (Nov. 7). Cell wall composition showed little consistent variation with harvest date while bulk biomass analysis showed a relative increase in cell wall content (lignin and structural sugars) and loss of extractives (minerals, protein, soluble sugars, and others). Following high or low severity AFEX pretreatment and high solids enzymatic hydrolysis (6% glucan loading), two field replicates were fermented using *Saccharomyces cerevisiae* 424A, a strain engineered to utilize xylose in addition to glucose. For both pretreatment severities, *S. cerevisiae* 424A grown in hydrolysates from the three earlier harvests utilized only a small fraction of available xylose, while almost complete utilization occurred within 96 hr for the last three harvest dates. Detailed analysis of the hydrolysate low molecular weight aromatics did not indicate any compounds potentially responsible for the inhibition, with most of the observed variation in their concentration due to pretreatment severity. Amino acid composition also did not appear to be limiting. Current indications point to a

**48**

plant-generated compound that degrades during senescence, which future work will attempt to identify. Ultimately this work demonstrates that, although an attractive option to maximize yield, harvesting switchgrass before it begins senescing could have a negative effect on downstream conversion processes.

Keywords: biomass composition, fermentation inhibition, harvest timing, lignocellulosic biofuel, switchgrass, xylose utilization

#### INTRODUCTION

Biofuels made from lignocellulosic biomass are a potential environmentally friendly, carbon neutral replacement for petroleum fuels (Robertson et al., 2017). As lignocellulosic biofuels are derived from the compounds in plant cell walls, they can be produced from a wide variety of feedstocks. It is estimated that the U.S. alone has the potential to generate as much as 1.5 billion tons of lignocellulosic biomass, a number that includes forestry residues; herbaceous crop residues, such as wheat straw and corn stover; and dedicated perennial energy crops, such as energy grasses and coppiced woody materials (Langholtz et al., 2016). All of these potential lignocellulosic crops have defined harvest timings, and for crop residues the timing of residue collection is linked to grain harvest, which makes it fairly inflexible. In contrast, harvest of energy grasses, such as switchgrass, can happen any time throughout the growing season or after. For a single harvest during the growing season, switchgrass harvest timings have been proposed anywhere from the point of peak aboveground biomass yield up to the following spring after overwintering in the field (Adler et al., 2006; Lindsey et al., 2013). This flexibility in harvest could be beneficial for farmers, allowing them to time their harvest according to the weather and accommodate other, more time-sensitive field work.

During the biological processing of lignocellulosic biofuels, choice of harvest timing for perennial feedstocks generally represents a tradeoff between biomass yield and quality (Gorlitsky et al., 2015). Following peak biomass, the plants tend to lose biomass due to three processes: resorption of nutrients to belowground tissues (Yang et al., 2009; Lindsey et al., 2013; Ashworth et al., 2017), weathering of dry and brittle aboveground biomass components (Adler et al., 2006; Anderson et al., 2013), and lodging, resulting in materials that are too low to the ground to harvest effectively (Adler et al., 2006; Anderson et al., 2013). In terms of biomass quality, earlier harvests have typically been reported as more digestible following biological conversion (Bals et al., 2010; Dien et al., 2013), however later harvests tend to have a greater proportion of structural carbohydrates (Adler et al., 2006; Anderson et al., 2013; Lindsey et al., 2013), the substrate for microbial-mediated biofuel production.

Resorption of nutrients is another important parameter that influences the choice of optimal harvest time. As switchgrass senesces, it remobilizes and resorbs translocatable nutrients into belowground tissues for use the following year (Adler et al., 2006; Yang et al., 2009; Lindsey et al., 2013). Earlier harvest timings may remove nutrients from the field before they can be fully resorbed, potentially compromising overwintering success and requiring additional fertilizer the following year, adding cost and negative environmental effects (Cahill et al., 2014). Ideally one could find an economically optimum harvest date that balances the cost of fertilizer replacement with decreased revenue from lower biomass yields later in the harvest window. Two studies have investigated this possibility with mixed results. One found that the date of profit-maximizing harvest is generally after peak biomass, and this is pushed later in the season for lower switchgrass prices (Cahill et al., 2014), however another study found no significant influence of harvest date on economic returns (Gamble et al., 2015).

While removal of nutrients with the harvested biomass is generally viewed negatively from both a crop production and a processing standpoint (Gorlitsky et al., 2015; Serapiglia et al., 2016a), this is not necessarily true for biological conversion pathways. The biological route of lignocellulosic fuel production uses a thermochemical pretreatment to open up the cell wall structure, followed by enzymatic hydrolysis to release cell wall sugars that are converted to fuel by fermentative microorganisms. Certain thermochemical pretreatments, such as ammonia fiber expansion (AFEX) are able to retain minerals and protein through the process (Lau and Dale, 2009; Lau et al., 2012; Ong et al., 2016). This means that the nutrients harvested with the biomass are not necessarily a hindrance to conversion, but can be utilized by the fermentation microorganism, eliminating the need for nutrient supplementation (Lau and Dale, 2009). However, this benefit of nutrients from earlier harvests may be offset by a higher soluble sugar content in the biomass. Recent work has found that pretreatment reaction products of soluble sugars, which are generally at higher levels earlier in the growing season (Adler et al., 2006; Lindsey et al., 2013), contribute to fungal inhibition during fermentation (Ong et al., 2016). It is possible that instead of having a positive impact on fermentation through increased nutrients, harvesting early could negatively impact fermentation through formation of inhibitors. In another study, earlier harvest date (July) of switchgrass had significantly lower xylose utilization by Saccharomyces cerevisiae compared to the later harvest (October) (Bals et al., 2010). The reason for this effect was not determined but potentially attributed to higher biomass extractives content in the early harvest switchgrass. Another study found that a different strain of S. cerevisiae had difficulty fermenting xylose, however this was observed for all switchgrass maturities (pre-boot, anthesis, and post-frost) (Dien et al., 2013).

**Abbreviations:** AFEX, ammonia fiber expansion pretreatment; HPLC, high performance liquid chromatograph; OD600, optical density at 600 nm; MRM, multiple reaction monitoring; s.d., standard deviation.

For this study we wanted to determine whether switchgrass harvest date could be timed to have the most beneficial impacts on fermentation while avoiding production of harmful degradation compounds. As it was possible that later harvest timings could have insufficient nutrients for the fermentation microbes, we evaluated the amino acid composition of the hydrolysate and stillage following fermentation. In order to obtain a more detailed map of harvest date impacts on switchgrass fermentation compared to what has previously been reported, switchgrass grown in Arlington, WI was harvested every 2–3 weeks from peak biomass (Aug. 20) until post-killing frost (Nov. 7), for a total of six harvest dates. Five field replicates were harvested and evaluated for their cell wall composition. Two of the field reps (R2 and R4) were further evaluated for their bulk biomass composition. These were then processed using AFEX pretreatment at two severities to evaluate whether it was possible to alleviate potential fermentation inhibition by adjusting pretreatment conditions. Pretreated samples were subjected to high solids enzymatic hydrolysis (6% glucan loading) followed by ethanolic fermentation using an engineered xyloseutilizing yeast strain, S. cerevisiae 424A.

# MATERIALS AND METHODS

# Switchgrass Growth, Harvest, and Processing

Switchgrass was cultivated at the Arlington Research Station (ARL, 43◦ 17′ 45 ′′ N, 89◦ 22′ 48′′W, 315 masl) in Arlington, Wisconsin. Switchgrass was sourced from plots 112, 206, 303, 412, and 510 in field 115. The soil type at ARL is Plano silt-loam, as previously described (Oates et al., 2016; Ong et al., 2016), and the mean temperature and precipitation were 5.6◦C and 782 mm, respectively for the entire year of 2014, and were 15.3◦C and 470 mm, respectively for the growing season from 22 May to 13 November 2014. Switchgrass was planted in June 2008, and samples were manually harvested in 2014 from five replicate field plots (R1 – R5), roughly every 2 weeks from peak biomass (20 August 2014) until after the killing frost (7 November 2014). Harvest dates were Aug. 20, Sept. 10 and 24, Oct. 6 and 22, and Nov. 7. Following harvest, all samples were dried in a 60◦C oven (∼48 h) and milled using a 18-7-301 Schutte-Buffalo hammer mill (Buffalo, NY) equipped with a 5 mm screen. Samples were stored at room temperature in sealed bags until used.

# Cell Wall and Bulk Chemical Composition of Untreated Switchgrass

Analyses were performed on extractives free alcohol insoluble residue (AIR), which provides a representation of the biomass cell wall composition. The AIR was isolated and polysaccharides (crystalline cellulose, xylose, arabinose, glucose, galactose, and mannose) were quantified based on an existing protocol (Foster et al., 2010b). Lignin monolignol (syringyl, guaiacyl, and phydroxyphenyl) concentrations were also evaluated using an existing protocol (Foster et al., 2010a). Because of the intensive time requirements for later experiments, field replicates R2 and R4 were chosen randomly for all further experiments. Composition analysis of the bulk biomass was performed according to the NREL laboratory analytical procedures for biomass composition analysis (Hames et al., 2008; Sluiter et al., 2008a,b,c, 2012), with some modifications, as described previously (Ong et al., 2016).

# Ammonia Fiber Expansion (AFEX) Pretreatment

Ammonia fiber expansion was used to pretreat switchgrass samples from two randomly chosen field replicates (R2 and R4). Pretreatment was conducted at two different severities: low (1.0 g NH<sup>3</sup> g <sup>−</sup><sup>1</sup> dry biomass, 0.5 g H2O g−<sup>1</sup> dry biomass, 80◦C) and high (1.5 g NH<sup>3</sup> g <sup>−</sup><sup>1</sup> dry biomass, 2.0 g H2O g−<sup>1</sup> dry biomass, 130◦C). The biomass was hand-mixed with the appropriate mass of water and then loaded into a 3.8 L stainless steel Parr reactor (Parr Instrument Co., Moline IL, USA) that was kept inside a walk-in fume hood. The reactor was sealed and liquid ammonia was added using a previously calibrated LEWA EK1 metering pump (Leonberg, Germany). The reactor temperature was quickly raised to the set point and then held at the set reaction temperature for 30 min residence time. At the end of the reaction, the ammonia was released from the reactor and filtered compressed air was passed over the biomass to facilitate removal of residual ammonia. The biomass was transferred to a custom vented acrylic drying box where it was dried using filtered compressed air until the moisture content was <12% (total weight basis). Following drying the biomass was stored in sealed bags until use.

# Enzymes

Cellic <sup>R</sup> CTec2 (138 mg protein/mL, batch number VCNI 0001), a complex blend of cellulase, β-glucosidase and hemicellulase, and Cellic <sup>R</sup> HTec2 (157 mg protein/mL, batch number VHN00001) were generously provided by Novozymes (Franklinton, NC, USA). The protein concentrations of the enzymes were determined by estimating the protein using the Kjeldahl nitrogen analysis method (AOAC Method 2001.11, Dairy One Cooperative Inc., Ithaca, NY, USA). The non-protein nitrogen was subtracted from the total nitrogen content and then the total was multiplied by a conversion factor of 6.25.

# High-Solid-Loading Enzymatic Hydrolysis

High solids-loading enzymatic hydrolysis [∼20% (w/w); 6% glucan loading] of AFEX pretreated switchgrass was performed to evaluate the effect of harvest timing on sugar release. Enzymatic hydrolysis was performed in 250 mL Erlenmeyer flasks with 100 mL reaction volume and the pH of hydrolysates was adjusted to 4.8 using 12 M HCl and/or 10 M KOH. Enzymes were added at a ratio of 70:30 CTec2:HTec2 on a protein mass basis with a total enzyme loading of 22 mg protein g−<sup>1</sup> glucan. To prevent contamination during EH and fermentation, Geneticin antibiotic at 100 mg/L was added prior to enzymatic hydrolysis. The flasks were incubated at 250 rpm and 50◦C for 96 h in an orbital shaking incubator (Innova <sup>R</sup> 44, New Brunswick Scientific, USA). Pretreated switchgrass was added in fed-batch mode at an interval of 2 h to improve mixing. After 96 h, the hydrolysate was separated from the residual solids by centrifugation at 4,000 rpm for 30 min (Eppendorf centrifuge 5810R, NY, USA). Liquid hydrolysate pH was adjusted to 5.50 ± 0.05 using either HCl or KOH as specified above, and then vacuum filtered through a 0.22µm sterile filtration cup (Millipore Stericup, Massachusetts, USA). The filtered hydrolysate was stored at 4◦C in a sterile bottle prior to fermentation. Samples for HPLC analysis were taken and stored at −20◦C prior to analysis.

## Oligosaccharide Analysis

Oligomeric sugar analysis was conducted on the hydrolysate liquid streams using an autoclave-based microplate acid hydrolysis method at a 1-mL scale. Hydrolysate samples (10x dilution) were mixed with 35 µL of 72% sulfuric acid in 5 mL open top centrifuge tubes capped with TPE mat, placed in a 96 well open-bottom microplate, screw-fitted into a metal plate and autoclaved at 121◦C for 1 h. Autoclaved samples were cooled in a refrigerator for 30 min, filtered into a 96 well microplate, and analyzed by HPLC. The concentration of oligomeric sugars was determined by subtracting the monomeric sugar concentration of the non-hydrolyzed samples from the total sugar concentration of the acid hydrolyzed samples. Sugar degradation was accounted for by running the appropriate sugar recovery standards along with the samples during acid hydrolysis.

#### Microorganism, Seed Culture Preparation, and Hydrolysate Fermentation

S. cerevisiae 424A(LNH-ST), which is a genetically modified yeast strain that ferments xylose (Ho et al., 1999), was provided by Prof. Nancy W. Y. Ho, Purdue University and used for fermentation of hydrolysates. A yeast extract-tryptone medium with 100 g/L glucose, 25 g/L xylose, 10 g/L yeast extract, and 20 g/L tryptone was used as the seed culture medium. Seed culture was prepared in a 250 mL Erlenmeyer flask with 100 mL medium inoculated from a frozen glycerol stock. After inoculation, the seed culture had an initial optical density at 600 nm (OD600) of 0.1. The flask was capped using a rubber stopper with a needle pierced through and was incubated at 30◦C and 150 rpm for 18 h. The seed culture OD<sup>600</sup> reached ∼12 at 18 h and was centrifuged at 4,000 rpm for 5 min. The resulting yeast cell pellets were used for inoculation of hydrolysate fermentation. The initial OD<sup>600</sup> for fermentation was 2.0. Hydrolysate fermentation was performed in a 125 mL Erlenmeyer flask with working volume of 30 mL at pH 5.5, 30◦C and 150 rpm for 96 h. A 0.5 mL sample was taken at different time points during fermentation, filtered, and stored at −20◦C prior to analysis.

#### HPLC Analysis

Hydrolysate and fermentation samples were diluted 10x before being analyzed by HPLC. Glucose, xylose, and arabinose concentrations were analyzed using a Shimadzu HPLC system equipped with a Bio-Rad Aminex HPX-87H column equipped with automatic sampler, column heater, isocratic pump and refractive index detector (RID). The column was maintained at 50◦C and eluted with 5 mM H2SO<sup>4</sup> in water at 0.6 mL/ min flowrate. Monomeric sugars were identified and quantified by comparison to authentic standards calibration curve. Fermentation samples were analyzed for ethanol and residual sugars with the above mentioned HPLC system equipped with an Aminex HPX-87H column maintained at 50◦C.

## Quantification of Imidazoles and Pyrazines in AFEX-Treated Switchgrass

AFEX-pretreated switchgrass was milled through a 2.0 mm screen using a Foss CyclotecTM mill (Eden Prairie, MN, USA). The milled biomass was extracted with acetone using an Accelerated Solvent Extractor (DionexTM ASE 200, ThermoFisher Scientific, USA) under the following conditions: 5 min heat, 5 min static, 150% flush volume, 120 s purge, two cycles, 1,500 psi, and 70◦C. Standards for the analyzed compounds were prepared in pure acetone in concentrations ranging from 0.00128 to 20 mg/L. Internal standards of 4-methylimidazole-d<sup>6</sup> (imidazole authentic standard) and 2-methylpyrazine-d<sup>6</sup> (pyrazine authentic standard) were obtained from C/D/N Isotopes (Pointe-Claire, Quebec, Canada) and added to each sample, standard, and blank at a final concentration of 6 mg/L. Samples were directly analyzed via GC–MS, without derivatization, based on the protocol from Chundawat et al. (2010), with the following modifications to the GC temperature program: 40◦C (2 min), from 5◦C/min to 150◦C (1 min hold), 8◦C/min to 200◦C (2 min hold), 20◦C/min to 260◦C (3 min hold). Response factors were calculated based on the peak area of the selected ion chromatogram (molecular ion; M+) of each compound relative to the area of the internal standard peak.

## Hydrolysate and Fermentation Broth Amino Acid Composition

Prior to amino acid quantification, 50 µL aliquots of samples were spiked with stable isotope labeled internal standards for the 20 common proteinogenic amino acids (Sigma-Aldrich Cell Free Amino Acid Mixture—13C,15N; P/N 767964-1EA) (with the exception of cysteine and methionine, which were not quantified) and processed by solid-phase extraction (Phenomenex Strata-X-C cartridges; P/N 8B-S029-HCH) to remove matrix interferents. SPE-processed samples underwent vacuum centrifugation before resuspension in 1 mL of Mobile Phase A. Samples were then analyzed via LC–MS/MS, based on the protocol from Gu et al. (2007), with the following modifications: mobile phase A was 10 mM instead of 1 mM (to reduce column equilibration time) and an LC gradient of 0.00–1.75 min (98% A); 1.76–8.00 min (linear ramp to 45% A); 8.01–9.00 min (10% A); and 9.01– 13.00 min (98% A). Response factors were calculated based on the peak area of the selected multiple reaction monitoring (MRM) chromatograms for each compound relative to the area of the MRM peak for each amino acid's stable isotope labeled internal standard. The data for glutamine was inconsistent for the hydrolysate samples, and so was not included in the reported data set.

#### Statistical Analyses

Statistics were conducted in R-Studio <sup>R</sup> , version 1.0.143 (Boston, MA). Confidence intervals and linear regressions (lm function) of xylose consumption vs. process ethanol yield (based on experimental and field replicates), cell wall composition data (based on technical replicates), oligomeric sugar concentrations (based on field and experimental replicates) and hydrolysate phenolic concentration (based on experimental and field replicates) were calculated and graphed using the ggplot2 package (v.2.2.1) (Wickham, 2009). Principal component analysis plots were constructed using the pca function in R-Studio and the ggbiplot package. Amino acid heat maps were constructed using the heatmap.2 function from the gplots package (v.3.0.1) (Warnes et al., 2009) and the colors were scaled using a logarithmic function.

# RESULTS

#### Xylose Utilization Is Inhibited for Early Harvest Dates

Evaluation of fermentation profiles reveal that although all of the glucose was consumed by 24 h for all samples, xylose consumption was significantly impaired for switchgrass samples harvested prior to October (**Figure 1** and Figure S1). This was not dependent on pretreatment condition, as both low (**Figure 1)** and high severity (Figure S1) AFEX conditions showed the same trend. Additionally, samples that were harvested in October or November generally had greater final ethanol concentrations compared to samples harvested prior to October. The process ethanol yield, or the amount of ethanol produced compared to that expected from complete conversion of sugars initially present in the hydrolysate, was linearly correlated with xylose consumption with little difference between pretreatment conditions (**Figure 2**) or field replicates (Figure S2). The experimental data separates cleanly into two groups: the materials with low xylose consumption and process ethanol yields, which were harvested from August to September, and the materials with high xylose consumption and process ethanol yields, which were harvested from October to November. More detailed evaluation of these results revealed a very large increase in specific xylose consumption between the early and late harvest periods for both AFEX severities (**Table 1**). Comparison of the metabolic ethanol yield (the ability of a microorganism to efficiently convert consumed sugars to ethanol) to the process ethanol yield showed that the metabolic ethanol yield was fairly similar between early and late harvest periods, in contrast to the process ethanol yield, which was much higher for the later harvest dates.

#### Cell Well Composition Does Not Vary Consistently With Harvest Date

The cell wall composition, evaluated on the basis of AIR, showed very few consistent trends with respect to harvest date across feedstocks. Of the compounds evaluated, crystalline cellulose and most of the hemicellulosic/pectic sugars (arabinose, glucose, and galactose) did not show any consistent trend across the field replicates nor any statistically significant variation with harvest date (Figure S3). When averaged across all field replicates, both the xylose content and the lignin monomer compositions of the cell wall showed minor

FIGURE 2 | Xylose consumption is correlated to process ethanol yields, which tend to be lower for harvest dates before October and show a similar trend regardless of pretreatment condition. Data points include both field (R2 and R4) and experimental replicates.




*Values are reported as the mean* ± *s.d. based on combined field replicates (n* = *2) and experimental replicates (n* = *2). Propagation of error was conducted to obtain s.d. values for all calculated values.*

\**The specific glucose consumption rate is in mM glucose consumed*·*OD-1 <sup>600</sup>*·*h -1 between 0 and 24 h of fermentation.*

*† The specific xylose consumption rate in mM xylose consumed*·*OD-1 <sup>600</sup>*·*h -1 between 0 and 24 h of fermentation.*

*‡ The metabolic yield is the ratio of sugars (glucose and xylose) consumed during fermentation to ethanol produced assuming 0.51 g ethanol/g sugars as the theoretical maximum.* §*The process yield is the ratio of sugars initially present in the hydrolysate (glucose and xylose) to ethanol produced assuming 0.51 g ethanol/g sugars as the theoretical maximum.*

statistically significant increasing trends with harvest date, though not all field replicates showed this trend (**Figure 3**). When looking at the averaged values, the lignin monomer composition may follow more of a parabolic trend, but this is less clear based on the individual field replicates. Of the cell wall components evaluated, only the mannose content showed any strong linear correlation with harvest date across all materials, and this was observed for all five field replicates (**Figure 3**).

### Bulk Biomass Composition Shows Consistent Trends With Respect to Harvest Date

Ash, protein, and extractives (water and ethanol soluble with the exception of water soluble sugars) all decreased with harvest date, and tended to plateau quickly following peak biomass, reaching a fairly consistent concentration by early September (**Figure 4**). Conversely, the cell wall components correspondingly made up a greater proportion of the plant biomass over time. Both the lignin and the polymeric sugar concentrations continued to increase between peak biomass and post-senescence. The water soluble mono- and disaccharides (sucrose, glucose, and fructose) showed a distinct trend compared to the other components analyzed. These sugars continued to increase following peak biomass, reached a peak in late September, after which they decreased rapidly until post-killing frost. The full biomass composition data, broken down by component and field replicate, is included in the supplementary information (Tables S1, S2).

# Enzymatic Hydrolysis Yields and Composition Were Predominantly Related to Pretreatment Severity Rather Than Harvest Date

Unlike fermentation, which showed a distinct pattern related to harvest date, the enzymatic hydrolysis conversions did not show this same trend. Although there were differences in hydrolysis conversions between harvest dates, there was no real trend observed in the effect on either monomeric and oligomeric glucose or xylose release from the pretreated biomass (**Figures 5A,B**). There was some minor difference in sugar release between AFEX pretreatment severities, but this was also not consistent across samples. In general, overall conversions tended to be higher at the higher pretreatment severity.

A significant proportion of released sugars remained as oligomers following enzymatic hydrolysis. Around 13–22% of released glucose, 28–50% of released xylose, and 57–85% of released arabinose were in oligomeric form in the hydrolysates. The amount of oligomeric arabinose in solution was strongly correlated with the oligomeric xylose concentration, but not the oligomeric glucose concentration (**Figure 5C**). There were on average 3.6 molecules of oligomeric xylose for every molecule of oligomeric arabinose in solution.

# Neither Quantified Inhibitors Nor Amino Acid Depletion Appear to Be Responsible for Inhibition

Of the more than 30 low molecular weight phenolic and nitrogenous compounds quantified in the hydrolysates, none

harvest date. Values are based on the amount in the alcohol insoluble residue (AIR). Data points represent technical replicates (*n* = 3). The "All" column aggregates the data for all five field replicates, with the red dot representing the average across all field and technical replicates. G, guaiacyl; S, syringyl.

showed a trend correlated with the reduced xylose consumption during fermentation, with a higher concentration for the three early harvest periods and a lower concentration for the three later harvest periods. Low molecular weight imidazoles and pyrazines were produced during the higher severity AFEX pretreatment, but not the low severity (**Figure 6A**). For the high severity pretreatment, the concentration in the pretreated biomass was only slightly correlated with the water soluble sugar content of the untreated biomass (**Figure 6B**).

Of all of the compounds quantified, only two varied linearly with harvest date for both pretreatment severities (p < 0.05), benzamide and feruloyl amide, which both decreased with harvest date (**Figure 7**). Seven compounds showed statistically significant trends (p < 0.05) with respect to harvest date across field replicates for one of the two pretreatment severities (low severity: 4-hydroxybenzamide, 4-hydroxybenzylalcohol, vanillic acid, ferulic acid, 3-hydroxybenzoic acid, and HMF; high severity: vanillyl alcohol). All of these compounds, except for 4 hydroxybenzamide, showed a decreasing trend with respect to harvest date.

More compounds (showed distinct differences in concentrations between pretreatment severities compared to harvest date (Figure S4, Table S3). There were a number of particularly interesting patterns observed related to pretreatment severity. Ferulate and p-coumarate in the hydrolysates were predominantly in the acid form at the lower severity and amide form at the higher severity. The ketones related to H-, G-, and S-lignin (4-hydroxyacetophenone, acetovanillone, and acetosyringone, respectively) were only generated at the high severity pretreatment condition. Syringamide and syringaldehyde were both generated at higher concentrations during the high severity pretreatment. When evaluated using principal component analysis, the majority of the differences between hydrolysates could be attributed to differences in pretreatment severity (48% of variability) (**Figure 8**). The second largest principal component (17% of variability) differentiated between peak biomass (8/20) and everything else. Additional principal components did not differentiate further by harvest date. The relative contribution of each analyzed compound to PC1 and PC2 is included in a rotation plot in the Supplemental Information (Figure S5).

Free amino acids were present in both the hydrolysate and post-fermentation stillage in µM to mM concentrations, depending on the specific amino acid (**Figure 9**). The amino

acids at the lowest concentration in both samples were the nitrogenous aromatic amino acids, histidine and tryptophan. In general, amino acid concentrations in the hydrolysate decreased with harvest date. However, this trend was not observed in the stillage, where the earlier harvest dates tended to have the lowest amino acid concentration. The higher severity AFEX hydrolysates tended to have higher amino acid concentrations in the hydrolysate compared to the low severity.

confusion in interpreting overall trends. The full data set is presented in Tables

#### DISCUSSION

S1, S2.

Unlike harvest of corn stover, which is linked to grain harvest, harvest of switchgrass can occur at any time throughout the growing season. However, timing is generally optimized to accommodate the competing interests of maximizing biomass yield and recycling nutrients to belowground tissues. Biomass yield is negatively affected by harvest delays, and as much as one-third of switchgrass can be lost between peak harvest and post-senescence (Gorlitsky et al., 2015; Serapiglia et al., 2016a,b; Ashworth et al., 2017). Most studies recommend late fall or winter harvest dates to avoid significant yield losses while achieving high nutrient resoprtion (Gorlitsky et al., 2015; Serapiglia et al., 2016a,b; Ashworth et al., 2017). However, biomass yields and nutrient resorption are not the only factors that may influence choice of harvest date. Harvest date can influence process ethanol yields, which is important due to the potential influence on the minimum ethanol selling price and biorefinery economics (Vicari et al., 2012). Although previous studies have shown an effect of switchgrass harvest date on

ethanol production (Bals et al., 2010; Dien et al., 2013), they evaluated only two or three harvest dates.

In our study we evaluated six harvest dates in the ∼2.5 month window between peak biomass and post killing frost and identified a rapid change that occurred in the switchgrass in the 2 week period between the last harvest in September and the first harvest in October. Prior to this period, xylose was poorly consumed, leading to low ethanol yields, however, during later harvests, xylose was almost entirely consumed during yeast fermentations and resulted in correspondingly

FIGURE 6 | Low molecular weight imidazoles and pyrazines form under the high but not the low temperature AFEX pretreatment (A) and are slightly correlated with soluble sugar content in the untreated biomass (B). The data are shown separately for both field replicates (R2 and R4). The linear regressions are based off of the combined field replicate data. The total water soluble sugars are the amount of water extractable monomeric and polymeric sugars in the untreated biomass.

higher ethanol yields. A previous study found a significantly lower xylose consumption for a July harvest date compared to harvest in October and hypothesized that this may have been due to a higher extractives content (Bals et al., 2010), though this was not confirmed or investigated further. Our analysis of the bulk biomass composition data for each harvest date did not reveal any components, including the total extractives, that varied in the same manner as the observed trend in xylose utilization. However, there was a peak in soluble sugar content (glucose, sucrose and fructose) that occurred in late September at the time of the change in xylose utilization. A fall peak in the soluble sugar content of switchgrass has been observed previously (Lindsey et al., 2013) and represents a source of readily accessible carbohydrates for the plant to use during grain fill (White, 1973). Once seed fill is largely complete, the plant prepares for senescence and rapidly translocates the sugars into the roots and crown for use during initial growth the following season (Sadeghpour et al., 2014; Gorlitsky et al., 2015). This indicates that around the time of the change we observed in xylose utilization efficiency, the switchgrass was in preparation for dormancy.

In terms of the general trends in biomass composition we found similar results as many others have previously reported for switchgrass: cell wall components (structural carbohydrates and lignin) tend to increase with harvest date, while protein, ash, and soluble sugars decline from peak biomass/anthesis to post-killing frost (Adler et al., 2006; Dien et al., 2006; Anderson et al., 2013; Lindsey et al., 2013; Serapiglia et al., 2016b). Based on the cell wall composition data, the apparent increase in cell wall components in the bulk biomass was not due to additional deposition of cell wall material but was instead due to loss of extractives and translocatable elements. The only strongly significant trend was an increase in mannose content with harvest date. Most grasses have very

each pretreatment condition across both field replicates (R2 and R4).

low mannose contents (Scheller and Ulvskov, 2010), however glucomannans are used as a seed storage polysaccharide in C4 grass species (Rodríguez-Gacio et al., 2012). For switchgrass, the entire aboveground biomass, including panicles, are harvested and milled, so it seems likely that the observed increase in mannose is associated with seed fill. This would indicate that the switchgrass continued to contribute to seed fill at least until the killing frost, when our sample collection ended.

Resorption of nutrients into belowground tissues with later harvest dates could result in a nutrient deficit during fermentation, however we found that none of the amino acids were particularly limiting (though this analysis did not evaluate the disulfide bridging enzymes, cysteine and methionine). As the amino acid contents tended to be higher for earlier harvests, and this is the area where the xylose inhibition was observed, amino acid depletion was unlikely the source of the observed effect on xylose consumption. The stillage results showed a lower amino acid concentration earlier in the growing season, however this may be an artifact of processing. Amino acids can be unstable during storage (Van Eijk et al., 1994), and the first three harvest date samples were stored for a longer time than the last three harvest dates. The levels of the two nitrogenous aromatic amino acids, tryptophan (indole) and histidine (imidazole) were extremely low, and it is known that these amino acids can degrade via low temperature Maillard reactions in reducing sugar solutions (Simpson et al., 1976; Baxter, 1995). It is possible that these amino acids react with biomass sugars either during pretreatment or enzymatic hydrolysis to form other compounds. However, even if this is the case, as long as there are other amino acids available, the yeast should be able to generate their own tryptophan and histidine biosynthetically. A previous study observed that the amino acid composition of AFEX-treated corn stover was sufficient to support growth

colored on a logarithmic scale and amino acids are grouped using a dendrogram. Data are averaged across field replicates (R2 and R4).

of fermentation organisms without additional supplementation (Lau et al., 2012).

A number of biomass-derived inhibitors have previously been shown to influence fermentation and we evaluated whether the concentration of any of these compounds matched the pattern observed in xylose utilization. Imidazoles and pyrazines are two classes of inhibitory compounds that are produced during Maillard reactions of soluble sugars with ammonia during AFEX pretreatment (Chundawat et al., 2010) and can contribute toward inhibition of yeast fermentation of switchgrass hydrolysates (Ong et al., 2016). Very low levels of imidazoles and pyrazines were detected in the low AFEX severity pretreatment, most likely due to the low temperature used (80◦C), which is a major factor in the formation of these compounds (Shibamoto and Bernhard, 1976; Klinger et al., 2013). As inhibition of xylose consumption was observed for both pretreatment severities, it is therefore unlikely that these compounds were the source of the inhibition we observed. Aromatic compounds, particularly the aldehydes, are routinely implicated in inhibition of yeast fermentation (Palmqvist and Hahn-Hägerdal, 2000; Almeida et al., 2011; Sarks et al., 2016; Xue et al., 2018). We quantified over 50 different low molecular weight aromatic compounds, however only nine varied in a statistically significant manner with harvest date and none corresponded to the observed pattern in xylose consumption. In general, there were lower levels of aromatic inhibitors present in the hydrolysates at the end of the harvest period compared to the beginning, but some compounds (feruloyl amide, ferulic acid, 3-hydroxybenzoic acid, and vanillyl alcohol) only had a high content for the earliest harvest date (8/20) with similar concentrations in all later harvests. This led to segregation of the earliest (8/20) harvest date hydrolysate from the others during principal component analysis.

Compared to harvest date, pretreatment conditions had a greater influence on the concentration of most of the aromatic compounds in the hydrolysates. Additionally, a number of interesting trends were observed with respect to pretreatment conditions. Although the low severity pretreatment operated at a higher ammonia:water ratio (2:1 vs. 0.75:1), it generated a greater acid to amide ratio for the hydroxycinnamates (ferulate and p-coumarate) compared to the high severity pretreatment condition. This may indicate that temperature is a key factor dictating conversion of ferulic acid and pcoumaric acid to the amide form, and the relative selectivity of ammonolysis and hydrolysis reactions (Chundawat et al., 2010). The acid form is generally more inhibitory compared to the amide form (Tang et al., 2015) and this could be one contributor toward the lower ethanol yields of the switchgrass pretreated at low severity. Additionally, the formation of ketones was only observed at the higher ammonia pretreatment severity. This could be due to either the high temperature or the greater proportion of water in the reaction that may facilitate formation of more oxygen-rich products.

Based on our results, delaying harvest until switchgrass remobilizes soluble sugars and begins to senesce may be sufficiently long enough to avoid detrimental effects on fermentation, while reducing potential yield losses due to weathering and lodging and allowing for resorption of most translocatable nutrients into belowground tissues. Unfortunately, our results showed no conclusive evidence for the source of inhibition of xylose utilization for the first three switchgrass harvests following peak biomass. This inhibition could not be attributed to soluble sugar degradation products (imidazoles and pyrazines), phenolic inhibitors, or amino acid depletion. What appears most likely is that the inhibition is coming from one or more compounds, perhaps generated for plant defense, that were dismantled or translocated as the switchgrass began to enter dormancy in late September/early October. Future work will attempt to determine the specific compounds that were responsible for the reduction in xylose utilization. Ultimately our results suggest that harvest should be delayed until after the biomass begins to enter dormancy, but it may not be necessary to delay until after a killing frost. Once the switchgrass began to enter dormancy, it had reached a stable concentration of minerals and nitrogen and had passed the peak in soluble sugar content. It may actually be fairly straightforward to monitor switchgrass for this transition either by tracking the color change as the plant dismantles chlorophyll and translocates nitrogen, or by evaluating the change in soluble sucrose concentration over time in the aboveground tissues. Most surprisingly we have found that the optimum harvest date from a processing standpoint coincides with the optimum from an agronomic standpoint, which is counter to what would be expected solely from the composition data. This emphasizes the need to fully process biomass samples from pretreatment through fermentation in order to evaluate their suitability as a biofuel feedstock. As we have previously observed, the standard methods for evaluating biomass composition are not always able to predict or explain variations in fermentation performance for agronomically diverse feedstocks (Ong et al., 2016).

# DATASETS ARE AVAILABLE ON REQUEST

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

# AUTHOR CONTRIBUTIONS

RO and GS conceived the overall project. RO, GS, SS, and LdC designed the experiments. SS conducted enzymatic hydrolysis and fermentation experiments. RO performed computational data analysis. RO and SS drafted the manuscript with edits and comments from GS and LdC.

# FUNDING

This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research under Award Number DE-SC0018409, and work funded by the DOE Great Lakes Bioenergy Research Center (DOE BER Office of Science DE-FC02-07E R64494).

# ACKNOWLEDGMENTS

We want to thank the following people for their work in biomass collection and processing: J. Sustachek, A. Miller, Z. Andersen, B. Faust, and J. Tesmer. We also thank the following people for their work in running experiments and generating data: M. Samghabadi for composition analysis, C. Donald Jr. for AFEX pretreatment, S. Smith for MS analyses on imidazole and pyrazine contents and amino acid composition, M. McGee for fermentation product analysis, and A. Higbee for hydrolysate phenolic and furanic compound analysis. We gratefully acknowledge Novozymes, who provided the enzymes used for this work and Dr. Nancy Ho, who provided the yeast strain through a cooperative agreement with MSU.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00052/full#supplementary-material

Table S1 | Bulk composition of untreated switchgrass (% of total dry biomass) - Field Replicate R2. Values are reported as the mean ± s.d. (*n* = 3 technical replicates).

Table S2 | Bulk composition of untreated switchgrass (% of total dry biomass) - Field Replicate R4. Values are reported as the mean ± s.d. (*n* = 3 technical replicates).

Table S3 | ANOVA *p*-values for phenolic linear regression models. Values are based on the combined field replicates (R2 and R3) across both high and low severity pretreatment conditions.

#### REFERENCES


Figure S1 | Xylose consumption is impaired during *Saccharomyces cerevisiae* 424A fermentation of switchgrass harvested prior to October for high severity AFEX pretreatments. High severity AFEX was at 1.5 g NH3 g−<sup>1</sup> dry biomass, 2.0 g <sup>H</sup>2O g−<sup>1</sup> dry biomass, 130◦C and 30 min residence time. Results are averaged across experimental replicates (*n* = 2) and field replicates (R2 and R4, *n* = 2). Error bars represent ± standard.

Figure S2 | Xylose consumption is correlated to process ethanol yields but is unrelated to switchgrass field replicate. Data points include both low and high severity pretreatment.

Figure S3 | Most cell wall carbohydrates show no consistent variation with harvest date across field replicates. Values are based on the amount in the alcohol insoluble residue (AIR). Data points represent technical replicates (*n* = 3). The "All" column aggregates the data for all five field replicates, with the red dot representing the average across all field and technical replicates. H = *p*-hydroxyphenyl.

Figure S4 | Most phenolic and furanic inhibitors did not vary significantly with harvest date. Separate linear regressions with 95% confidence intervals were plotted for each pretreatment condition across both field replicates (R2 and R4).

Figure S5 | PCA rotation plot showing the relative contribution of each hydrolysate compound to the principal components. Colors indicate compound functional group.


Fraction Process Samples. Laboratory Analytical Procedures (LAPs). National Renewable Energy Laboratory, Golden, CO.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Ong, Shinde, da Costa Sousa and Sanford. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Changing Times: Altering Establishment Spacing, Harvesting Frequency, and Harvesting Machines to Promote Increased Sawtimber Volumes

Marissa "Jo" Daniel <sup>1</sup> \*, Tom Gallagher <sup>1</sup> , Dana Mitchell <sup>2</sup> , Timothy McDonald<sup>3</sup> and Brian Via<sup>1</sup>

*<sup>1</sup> School of Forestry and Wildlife Science, Auburn University, Auburn, AL, United States, <sup>2</sup> USDA Forest Service, Southern Research Station, Auburn, AL, United States, <sup>3</sup> Bio-systems Engineering, Auburn University, Auburn, AL, United States*

#### Edited by:

*Allison E. Ray, Idaho National Laboratory, United States*

#### Reviewed by:

*Jaya Shankar Tumuluru, Idaho National Laboratory, United States Sudhagar Mani, University of British Columbia, Canada*

> \*Correspondence: *Marissa "Jo" Daniel mzd0060@auburn.edu*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *13 December 2017* Accepted: *08 June 2018* Published: *27 June 2018*

#### Citation:

*Daniel MJ, Gallagher T, Mitchell D, McDonald T and Via B (2018) Changing Times: Altering Establishment Spacing, Harvesting Frequency, and Harvesting Machines to Promote Increased Sawtimber Volumes. Front. Energy Res. 6:61. doi: 10.3389/fenrg.2018.00061* Today's landowners are faced with important decisions when establishing loblolly pine plantations in the Southeastern part of the United States with regards to planting dimensions and forest management techniques. Although recent studies are beginning to demonstrate the need for change from the old practices, suppressed biomass markets and prices are hindering the transition. This paper provided readers with an informational overview of the benefits of: incorporating an additional thinning regime for biomass, using alternate spacing methods such as FlexstandsTM and rectangularity, and using small-scale harvesting machines for conducting initial thinning's. The overview was supported with both a field study as well as a modeling tool which verified using one or all of the above mentioned techniques to increase total harvest volumes while minimizing residual stand damage. The modeling tool determined that final sawtimber volumes were increased by a minimum of 15 green tons per acre using one or more of the above techniques. When expanding this volume out to 20 acres, the minimum tract size harvested in the southeast using convention equipment, landowners could easily recover any losses incurred from the suppressed biomass markets minimizing overall risk and promoting the use of these alternative techniques.

Keywords: flex plantations, rectangularity, small-scale harvesting, woody biomass, Ptaeda modeling

## INTRODUCTION

Forest landowners in the southeastern part of the United States are faced with multiple challenges when it comes to harvesting loblolly pine (Pinus taeda L) from their land. First, tract sizes are shrinking as lands become more fragmentized making it hard for landowners to find loggers willing to harvest their land (Daniel, 2012; Aguilar et al., 2014; Butler and Butler, 2016a). Next, plantations that promote woody biomass harvesting are being encourage but there are minimal markets available to sell the product to, stumpage prices are minimal if existent for the product, and today's standard sized machines aren't able to cost-effectively harvest the product so loggers aren't willing to cut the biomass for the landowner (Botard et al., 2015; BBI International, 2017; Gallagher et al., 2017; Yu et al., 2017). Finally, in order for these plantations to pay for themselves, landowners need to produce the highest sawtimber volumes possible to mitigate the risk of such a long-term investment and incentivize them to re-establish the land back into timber rather than convert it to another use that provides greater financial or intrinsic value for them (Butler and Leatherberry, 2004; Butler, 2008; Aguilar et al., 2013).

With all the above mentioned challenges, it becomes confusing for a landowner when trying to decide how to establish and manage their loblolly pine plantation. This paper's objectives were to provide an informational overview of the benefits of: incorporating an additional thinning regime for biomass, using alternate spacing methods such as FlexstandsTM and rectangularity, and using small-scale harvesting machines for conducting initial thinning's to promote increased sawtimber volumes. The overview was supported with both a field study as well as a modeling tool which verified using one or all of the above mentioned techniques to increase total harvest volumes while minimizing residual stand damage.

#### BIOMASS HARVESTS

Biomass harvests differ from first thinning's in a variety of ways. A pine biomass harvest is typically conducted between years 5 to 9 whereas a first thinning is between years 10 and 16. This difference in age generally results in a difference in size, product class, and inadvertently delivered price (Gallagher et al., 2017). This smaller diameter creates more surface bark, limbs, and needles which are undesirable when making pulp because it requires additional chemicals to be used during the breakdown of cellulosic fibers. Biomass is therefore not often used during a pulpwood shortage (Bajpai, 2012). Pulpwood, however, can be a suitable alternative when biomass shortages arise so consequently market demand doesn't increase and neither does woody biomass's price.

There are over 90 pulp and paper mills in the Southeastern part of the United States compared to the 41 biomass facilities that can be found in the same region. Of these 41, 12 are biomass power facilities which together produce only 563.3 megawatts of power every year from a combination of pulpwood, woody biomass and logging residues. Sixteen of the 41 are pellet mills which are able to use both hardwood and softwood feedstock (pulpwood or woody biomass) to produce approximately 7 million green tons of pellets per year. Eleven of the 41 are pellet mills which use only softwood feedstock (pulpwood or woody biomass) to produce 4.5 million green tons of pellets per year, and 2 are pellet mills designated as woody biomass feedstock. These two mills produce over 2 million green tons of pellets annually by themselves (BBI International, 2017). In general, it can be seen that even the biomass designated facilities are being supplied still with pulpwood rather than wood biomass only, indicating that there is a plethora of market potential if biomass was readily available as a product.

According to Timber Mart-South, the 2016/2017 average delivered price for woody biomass was \$21.18 per green ton (Timber Mart South, 2018). This price appears high and comparable to the 2016/2017 average delivered market price for pulpwood of \$29.49 per green ton, however it is deceiving. Delivered prices for woody biomass are designated for "clean" chips that come from the mill and are being deliver to another facility. These chips do not have any bark, needles, small limbs, or dirt in them. Woody biomass that comes straight from the woods can either be transported "whole tree" with tops and limbs still attached to the main stem or as "dirty chips." Dirty chips indicate that the tree has been chipped in the woods and will have limbs, needles, bark, dirt, and the potential for other small objects mixed into the chips. The market for "dirty chips" is basically non-existent at this time, therefore, revenue from woody biomass is also not available (Mitchell and Gallagher, 2007).

Although incorporation of a biomass harvests is not currently a viable solution for increasing a landowner's revenue at the time of that harvest, an additional thinning can increase total stand yield by removing trees that would either die or plateau the stands growth (Dean and Baldwin, 1993; Sharma et al., 2002). The removal of biomass to decrease the stands overall density stocking allows trees to continue to grow at a competitive rate thereby inadvertently increasing the number of sawtimber trees available throughout the stand (Amateis et al., 2004). Planting with higher density stocking initially has also been shown to instigate greater competition between saplings encouraging straighter trees with less branches which eventually has the potential to lead to a higher quality final product (Amateis et al., 2009; Amateis and Burkhart, 2012; Gallagher et al., 2017).

# FLEXSTANDSTM

The concept of a FlexStandTM was coined by ArborGen: Global Reforestation Partner, a worldwide provider of both genetically enhanced and conventional tree seedlings. This silvicultural technique involves planting conventional biomass, open-pollinated (OP), trees in-between rows of genetically improved, mass-control pollinated (MCP), trees to provide landowners with an economical solution for growing and thinning Loblolly pine stands (ArborGen Inc., 2018). This unique plantation establishment method was designed to assist in risk mitigation for future timber markets by producing multiple products from the same stand. This technique also allows landowners the flexibility of altering their management decisions based on current and expected market dynamics.

FlexStandsTM are considered to be high-density plantings. Although planting strategies differ depending on landowner objectives, the overall concept is to plant a high number of trees per acre by alternating/interchanging row plantings between MCP trees and OP trees. The enhanced seedlings will be spaced anywhere from 6 to 10 feet apart down the rows however the non-modified seedlings will be spaced as close as 2 feet and as far apart as 6 feet in order to increase the density stocking of the stand. Rows are typically 10–12 feet apart but have been seen as close as 5 feet apart (ArborGen Inc., 2018). Research has shown that seedling growth is not detrimentally affected by the distance between trees for the first few years of growth. Rather, the closer the seedling spacing, the more the saplings tend to focus on bole growth rather than branches or needles thereby decreasing defects that can be found in the tree (Ma, 2014).

Altering seedling types throughout the stand by rows has also been proven to minimize the costs of planting to the landowner because they are no longer purchasing all genetically enhanced seedlings, of which half are eventually removed before growing to sawtimber size (Ma, 2014; ArborGen Inc., 2018). FlexStandsTM are also proven to reduce the loss of revenue for landowners compared to if they were to plant only one seedling type. Planting only OP trees reduces the final sawtimber size inadvertently decreasing overall revenue, whereas planting only MCP seedlings results in a significantly higher increase in initial costs which must be carried through to the final harvest that is not guaranteed to be more profitable (ArborGen Inc., 2018).

With the FlexStandTM system, a biomass harvest is conducted between years 6 and 9 removing all OP sapling rows in order to promote the continued growth of the stand. A pulpwood thinning is conducted around years 12–16 to once again keep the stand from stagnating in size with a final harvest being conducted between years 24 and 30 depending on tree diameters and market prices.

Revenue associated with conducting a first thinning with both the biomass and the pulpwood out of the FlexStandTM does not currently mitigate the associated harvesting costs. Incorporating a biomass thinning into the management regime beforehand, however, does increase the size and overall value of the final sawtimber trees by forcing them to grow straighter and with fewer branches for the first few years of their life which results in a higher value final product (ArborGen Inc., 2018). When considering overall profitability, FlexStandTM could be considered a potential solution if a biomass harvest is conducted within the conventional timber harvest as long as there were viable markets to send the products. Further promotion for the FlexStandTM could occur if harvesting and relocation costs could be reduced by using small-scale equipment.

# RECTANGULARITY

Similar to the idea that a FlexStandTM could be a viable option to modify planting establishment methods, rectangularity is also being studied for its feasibility to promote woody biomass in the South. Typically, conventional stand seedling establishments occur with a specific number of trees being planted per acre in a shape that resembles a square. With rectangularity, the same number of trees are planted per acre but the shape resembles a rectangle rather than a square. This configuration allows for wider spacing in-between the rows of trees making site preparation costs cheaper as well as increasing maneuverability of forestry equipment throughout the stand, inadvertently decreasing damage caused to residual trees (Amateis et al., 2004).

The concept of rectangularity has been intermittingly studied since the 1940's as researchers continue to contemplate the ideal plantation spacing for specific tree species (Sharma et al., 2002; Amateis and Burkhart, 2012). Almost all studies have shown that rectangularity has no effect on tree height, diameter, volume per acre, basal area per acre or even tree survival (Gerrand and Neilsen, 2000; Amateis et al., 2004, 2009; Brand, 2012). In fact, most studies have shown that age plays a more significant effect than rectangularity. Crown size and shape appear to be the only factors that should be taken into account when contemplating a rectangularity spacing.

Although there are a variety of spacing options with regards to rectangularity, the three most recognized coincide with 436 trees per acre (tpa), 605 tpa, and 908 tpa. A normal plantation spacing at 436 tpa would be 10 feet in-between-rows by 10 feet within-rows, compared to the rectangular option of 20 feet inbetween-rows by 5 feet within the rows. At 605 tpa, a normal spacing would be 9 feet by 8 feet whereas a rectangular spacing would be 12 feet by 6 feet. Finally, at 908 tpa, a normal spacing regime would be 8 feet by 6 feet compared the rectangular spacing of 12 feet by 4 feet (Sharma et al., 2002; Amateis et al., 2004, 2009).

As forestry equipment continues to grow in dimension, landowner holdings are decreasing in size. Rectangularity could provide a viable solution to the increasing amounts of damage unintentionally administered to residual trees when thinning's occur. With rectangularity, the need for small-scale harvesting equipment becomes less of an issue, allowing the equipment industry to continue to focus on producing larger more powerful machines. NIPF landowners would also benefit from this technique by being able to strategically plant rows in a manner which allowed for optimal growth and harvest of the tract in future years while allowing for machine maneuverability.

# SMALL SCALE HARVESTING

Non-industrial private forest landowners (NIPFs) account for 36% of all of the forest land, 1+ acres, in the United States. Of this percentage, 13% comes from landowners who reside in the Southeastern part of the United States (Butler and Butler, 2016a,b). According to the national survey conducted in 2006, the majority of acres owned by NIPF landowners is between 1 and 49 acres (Butler and Leatherberry, 2004; Butler, 2008). As woody biomass becomes a more desired commodity, the forest industry will begin to look for further resources to supply to their mills. In addition to experimenting with genetic improvements for tree growth and establishment/planting modifications, mills will likely turn to the NIPF landowners for greater contribution.

Research has shown that it is unprofitable for a logger to harvest trees on less than 20 acres because today's equipment is too expensive for the harvest to result in economically feasibility after relocation costs, capital investments, labor, and fuel expenses are withheld from revenue (Athanassiadis, 1997; Burdg and Gallagher, 2011). Additionally, upholding today's high standards for best management practices can become an issue due to the large size of standard machines which measure approximately 10–11 feet wide, can range from 20 to 30 feet in length, weigh between 30,000 and 50,000 lbs, and have 174–300 HP for engine power (Caterpillar, 2018; Deere and Company, 2018). Even though a majority of the fellerbunchers and skidders in the south have articulated steering, these equipment specifications can inflict significant damage on residual trees when working in minimal acreage, conducting pulpwood thinning's, or even biomass thinning's.

Ideally, the top leaders in the forest equipment industry would design feller-bunchers and skidders that met the economic and environmental requirements of harvesting an area that was less than 20 acres in size. These machines would need to be small enough to maneuver through narrow spaces and rows without causing significant residual damage. The machines would need to be capable of handling trees approximately 55 feet in height and 9 inches in diameter. Ultimately, producers must be able to provide these machines at a cost which makes harvesting small tracts profitable. Realistically, however, equipment continues to grow in size to meet the market demand for larger and more powerful machines. Until market demand increases for smaller machines, minimal advancements will be made by the industries leaders.

Although purchasing small-scale feller-bunchers and skidders in the United States is currently a daunting task, finding forestry attachments that connect to skid-steers, compact tracked loaders, and mini-excavators is not. The ability to connect to a variety of attachments, both forestry-related and otherwise, to complete the immediate task at hand has made these machines the most versatile options available on today's market. Because of the advancements that have been made on these machines in both horsepower and hydraulic pressure flow technology, manufacturers have been able to create a system called "high flow." This system allows operation of attachments requiring significant speed and/or torque such as the harvester saw-heads which were previously impossible on such small machines.

These small-scale machines are dimensionally smaller, ranging from 3 to 7 feet in width and 8 to 15 feet in length depending on make and model which suggests increased mobility in small tight areas. Machine weights and range from approximately 2,500 to 9,500 pounds for the skidsteers/compact tracked loaders and 8,500 to 18,500 pounds for the mini excavators. Machine engine power ranges from 65 to 106 hp for skid-steers/compact tracked loaders and 40 to 65 hp for the mini excavators (Caterpillar, 2018; Deere and Company, 2018). These specifications indicate that these machines can be transported with a pickup truck and trailer rather than with a semi and lowboy trailer as is required for standard forestry equipment, inadvertently decreasing transportation costs. These machines are also known for having a low ground pressure which minimizes ground disturbance making them environmentally friendly. Finally, initial purchase price differences between small-scale and standard forestry equipment can be as low as one quarter to as high as one half of the cost depending on make, model, and attachment configuration.

#### METHODS

#### Case Study Site Description:

The field study was conducted on the Solon Dixon Forestry Education Center in Covington County, Alabama. The site consisted of a total of approximately 2.66 acres on Dothan and Malbis sandy loams. Stand 1 was 1.02 acres in size and contained a loblolly pine plantation with 8 × 6ft spacing. Stand 2 was 1.64 acres in size and was considered a flex plantation stand. The spacing configuration consisted of every third row being 10ft by 4ft spacing planted with OP seedlings while all other rows were MCP seedlings planted with a 10ft by 8ft spacing. Both stands were established with their rows facing in an east-west direction with a 20-foot corridor separating the two stands. The stands were approximately 8 years old at the time of harvest, in May 2017, with minimal mortality found in either stand.

A Caterpillar 279D compact track loader machine with a Fecon FBS1400 Single Knife Tree Shear attachment head was used to remove every third row from both stands for a harvested basal area of 70. The track loader weighed approximately 10,000 lbs, had 73 HP engine power, was 6 feet wide and 7 feet long. The sheared trees were collected in the shear heads' accumulating arm until full where the bundle would then be laid down within the row. A turbo forest mini skidder was used to collect the bundles and remove them from the site. This machine weighed 7,500 lbs, had 50 HP engine power, was 6 feet wide, and 12 feet in length. No time study was conducted in this analysis so operational costs could not be calculated.

Approximately two bundles per row in stand 1 and three bundles per row in stand 2 were randomly selected to be measured for a total of 16 bundles in stand 1 and 12 bundles in stand 2. Individuals trees were measured out of each selected bundle. Overall, 88 trees were measured in stand 1 and 79 trees were measured in stand 2.

Data were recorded and analyzed in Microsoft Excel. Results for the field data were analyzed by grouping trees by dbh class using 1-inch intervals from 3 to 9 inches. Basal area was calculated per size class as was the overall basal area that was removed from each stand. The average weight per tree was calculated for each size class and protracted out to determine the overall tonnage harvested per size class for 1 acre. Total green tons removed per stand were calculated to use as a reference for comparison. A stump count was conducted in each row per stand to use a reference for actual tree removal data. Two-sample t-tests were conducted in Minitab to determine if there were statistical differences between the field data for total height, weight, basal area, or dbh between Stand 1 and Stand 2.

# Ptaeda Study Model Description:

A comparison model study was conducted using a loblolly pine plantation modeling tool named Ptaeda 4.0. Six separate models were run with this tool; one each for stands 1 (M1) and 2 (M3) with a biomass harvest at year 8, thinning's at year 16 and final harvests at year 28. A third model (M5) was run to simulate a rectangularity setting with 12 × 4ft spacing (908 tpa) that could be compared against Stand 1. This model followed the same parameters as the previous two with regards to thinning and harvest schedules. The other three model simulations (M2, M4, and M6) only conducted pulpwood thinning's at year 16 with final harvests at year 28.

Each model incorporated specific parameters relating the models as close to field conditions as possible. Stand information included site productivity of 85, total rotation lengths of 28 years, planting distances between trees and between rows of 8 × 6, 10 × 6, and 12 × 4. Site information included physiographic regions based in the Coastal Plain, well-drained drainage class, and no fertilization at planting. Merchandising options and limits resulted in pulpwood tops at 2 inches with minimum dbh at 5 inches, chip and saw tops at 4 inches with dbh at 8 inches, and sawtimber tops at 6 inches with dbh at 11 inches. All topwood from chip and saw and sawtimber product classes were added into the pulpwood product class. All trees were calculated using green weight (green tons/acre with bark) measurements. No economic parameters were designated. Mid-rotation treatments varied by harvest plan. Biomass harvests were conducted using a 3rd-row and low (70 basal area) thin method at year 8 with 16 year thinning's conducted with a targeted residual basal area of 70 square feet. One thinning harvests included a third row and low (70 basal area) thinning conducted at year 16 only.

Ptaeda data that was recorded into excel included: the site index, the treatment conducted, dominant height, average dbh, average height, average crown ratio, dbh class, tree number, basal area, total weight (green ton), pulpwood weight harvested (green ton), chip n saw weight harvested (green ton), and sawtimber weight harvested (green ton). Clark and Saucier were referenced to calculate the predicted green weight in pounds of total tree (wood, bark, and foliage) in the Coastal Plain, based on dbh size class for total tree height using the following equation:

$$Y = 0.23369 \* (dbh^2 \* 
to 1 \stackrel{
\text{left}}{
to} 
\text{!}
\tag{1}$$

This number was converted to green tons and then multiplied by the total number of trees in each size class to find total tons per size class. Weights were calculated for each treatment year both before and after each harvest treatment by dbh size class but only the harvested treatment weights were used to calculate price per green ton. Harvest weights were then calculated per product class following the previously mentioned mechanizing limits.

Biomass weight was calculated using the difference between Clark and Saucier total tree green weight from the Ptaeda model merchandized green weight in each dbh size class. These weights were summed to determine a biomass weight in green tons for tops, limbs, and needles. One inch through four inch dbh size class weights for total tree height from Clark and Saucier were also included when available to determine total biomass available in the woods for that harvest treatment year. Regardless of intentions to collect all biomass available, recent studies have estimated that approximately 30% of the biomass harvested remains in the woods (Lancaster, 2017). For this reason, 30% of the biomass harvest weight was removed from the final biomass tonnage values. New total weights for each harvest were calculated to incorporate this 30% loss in biomass harvest.

Price per green ton was calculated for each product class as was total revenue for each treatment. Revenue, net present value at 3% (NPV), internal rate of return (IRR) were calculated both with biomass as well as without biomass. Cost for stand establishments for the landowner was calculated using reference numbers from the "Costs and Trends of Southern Forestry Practices 2012" by the Alabama Cooperative Extension System (Dooley and Barlow, 2013). Item description prices were based on numbers for the southern coastal plain on a per acre basis and included chemical site preparation at \$89.41, burning after chemical site prep \$53.44, hand planting costs for bare root seedlings \$62.78, fertilizer at establishment \$104.95, and seedling costs per thousand \$48.69 per thousand. Logging costs were not calculated since the biomass harvest costs would need to be calculated using small-scale equipment and that information is not currently available. Additionally, logging costs are not typically incurred by the landowner directly, rather they are removed from the landowner's final revenue received from harvest.

### RESULTS

## Field Study

Eighty-eight trees were measured in Stand 1 out of the 232 that were harvested. Of the trees that were measured, 6 were within the 3-inch dbh class, 9 were in the 4-inch dbh class, 26 were within the 5-inch class, 34 within the 6-inch class, 12 were in the 7-inch class, 1 in the 8-inch class, and none were found to be within the 9-inch dbh class (See **Table 1**). The average dbh for the stand was 6 inches with an average height of 39 feet. The residual basal area was approximately 70 down from the original 120 before harvesting. Tree weights for each dbh class were averaged to calculate the average weight per tree in each dbh class as well as the average dbh weight per acre. A final weight for Stand 1 was calculated at 84,690 pounds or 42.35 green tons that were removed from a one-acre tract.

Seventy-Nine of the Five hundred and sixty eight harvested trees were measured in Stand 2. Of those trees 3 were within the 3-inch dbh class, 16 were in the 4-inch dbh class, 32 were within the 5-inch class, 16 within the 6-inch class, 10 were in the 7 inch class, 1 in the 8-inch class, and 1 was found to be within the 9-inch dbh class (See **Table 2**). The average dbh for the stand

TABLE 1 | Summary data for the 8 × 6 stand of harvested trees 1.


TABLE 2 | Summary data for the 10 × 6 harvested stand 2.


was also 6 inches with an average height of 40 feet. The residual basal area was approximately 70 down from the original 120 before harvesting. Tree weights for each dbh class were averaged to calculate the average weight per tree in each dbh class as well as the average dbh weight per acre. A final weight for Stand 2 was calculated at 105, 158 pounds or 52.58 green tons that were removed from a 1-acre tract.

Visual observations were made following the harvests to collect information concerning damage made to residual trees. For the purpose of the study, damage was classified as any scrape or mark that was longer than 6 inches and cut through the bark into the trees cambium layer. Less than 5% damage was found in Stand 1 and none of the damage appeared to be significant enough to cause mortality. Less than 1% damage was observed in Stand 2, also none of which appeared harmful enough to cause mortality.

Two-sample t-tests were conducted to determine if there were any significant differences between the two stands with regards to the tree weights, dbh, height, or basal area. All variables except tree weight were found to be statistically insignificant. Tree weights, however, had a P-value of 0.045 at the 95% significance level. An additional comparison was conducted to determine if Stand 1 trees weighed more than Stand 2 trees. Thist-test was also found to be significant at the 95% level with a P-value of 0.023.

#### Ptaeda Model Green Tonnage

The Ptaeda model that included biomass in the harvest for Stand 1 (M1) had an average dbh of 5 inches with an average tree height of 31.5 feet after year 8. There were approximately 39 total green tons harvested at that time, all of which were designated as biomass. At year 16 a 2nd thinning resulted in an average dbh of 9.27 inches and an average height of 62.7 feet. Approximately 88 total green tons were removed with 11 green tons coming from biomass, 5 green tons from pulpwood, 70 green tons from chip and saw, and 2 green tons coming from sawtimber. Final harvest at year 28 resulted in an average dbh of 11.58 inches and 83.7 feet for an average height. Approximately 165 total green tons were removed with 19 green tons coming from biomass, 0 green tons from pulpwood, 67 green tons from chip and saw, and 79 green tons coming from sawtimber. In total 291 green tons were removed from the stand during the three harvests, with 69 green tons coming from biomass, 5 green tons from pulpwood, 137 green tons from chip and saw, and 80 green tons from sawtimber (See **Figures 1**–**4**).

Conventional Ptaeda model for Stand 1 that did not include a biomass harvest, also known as one thinning (M2), had an average dbh of 8.31 inches and an average height of 60.1 feet after the 1st thinning at year 16. There were approximately 83 total green tons harvested, 11 green tons came from biomass, 40 green tons from pulpwood, 32 green tons from chip and saw, and 1 green ton from sawtimber. Final harvest at year 28 resulted in an average dbh of 10.45 inches and an average height of 81 feet. One hundred and fifty one total green tons were removed from the final harvest with 18 green tons coming from biomass, 3 green tons from pulpwood, 93 green tons from chip and saw, and 25 green tons from sawtimber. In total 234 green tons were harvested from the tract with 28 green tons coming from biomass, 42 green tons from pulpwood, 125 green tons from chip and saw, and 39 green tons from sawtimber.

Stand 2's biomass inclusion harvest (M3) resulted in an average dbh of 5.52 inches with an average tree height of 32 feet after year 8. There were approximately 37 total green tons harvested at that time, all of which were designated as biomass. At year 16 a 2nd thinning resulted in an average dbh of 9.66 inches and an average height of 63.4 feet. Approximately 87 total green tons were removed with 11 green tons coming from biomass, 0.5 green tons from pulpwood, 74 green tons from chip and saw, and 2 green tons coming from sawtimber. Final harvest at year 28 resulted in an average dbh of 12.06 inches and 84.7 feet for an average height. Approximately 164 total green tons were removed with 19 green tons coming from biomass, 0 green tons from pulpwood, 35 green tons from chip and saw, and 110 green tons coming from sawtimber. In total 288 green tons were removed from the stand during the three harvests, with 67 green tons coming from biomass, 0.5 green tons from pulpwood, 108 green tons from chip and saw, and 112 green tons from sawtimber.

Stand 2's one thinning harvest (M4) had an average dbh of 8.81 inches and an average height of 60.8 feet after the 1st thinning at year 16. There were approximately 85 total green tons harvested, 11 green tons came from biomass, 23 green tons from pulpwood, 51 green tons from chip and saw, and 0 green ton from sawtimber. Final harvest at year 28 resulted in an average dbh of 11.05 inches and an average height of 81.0 feet. Final harvest tonnage resulted in 158 total green tons being removed with 19 green tons being from biomass, 0 green tons from pulpwood, 79 green tons from chip and saw, and 60 green tons from sawtimber. In total 242 green tons were harvested from the tract with 30 green tons coming from biomass, 23 green tons from pulpwood, 130 green tons from chip and saw, and 60 green tons from sawtimber.

Stand 3's biomass inclusion harvest (M5) resulted in an average dbh of 4.87 inches with an average tree height of 31.2 feet after year 8. There were approximately 38 total green tons harvested at that time, all of which were designated as biomass. At year 16 a 2nd thinning resulted in an average dbh of 8.68 inches and an average height of 61.2 feet. Approximately 88 total green tons were removed with 12 green tons coming from biomass, 25 green tons from pulpwood, 50 green tons from chip and saw, and 1 green ton coming from sawtimber. Final harvest at year 28 resulted in an average dbh of 10.82 inches and 81.9 feet for an average height. Approximately 153 total green tons were removed with 18 green tons coming from biomass, 0 green tons from pulpwood, 89 green tons from chip and saw, and 46 green tons coming from sawtimber. In total 278 green tons were removed from the stand during the three harvests, with 67 green tons coming from biomass, 25 green tons from pulpwood, 138 green tons from chip and saw, and 48 green tons from sawtimber.

Conventional Ptaeda model for Stand 3 that did not include a biomass harvest, also known as one thinning (M6), had an average dbh of 6.57 inches and an average height of 41.7 feet after the 1st thinning at year 16. There were approximately 84 total green tons harvested, 11 green tons came from biomass, 43 green tons from pulpwood, 30 green tons from chip and

saw, and 0 green ton from sawtimber. Final harvest at year 28 resulted in an average dbh of 8.96 inches and an average height of 62.9 feet. One hundred and fifty four total green tons were removed from the final harvest with 19 green tons coming from biomass, 5 green tons from pulpwood, 97 green tons from chip and saw, and 33 green tons from sawtimber. In total 237 green tons were harvested from the tract with 30 green tons coming from biomass, 47 green tons from pulpwood, 127 green tons from chip and saw, and 33 green tons from sawtimber.

Although M1 had slightly higher total and biomass tonnage overall, M3 was a close second and produced the most sawtimber in the final harvest by 32 green tons over M1. All models which included the biomass thinning produced more total weight, biomass tonnage, and sawtimber tonnage than the alternative model with the same spacing. Model 5 had the smallest net gain in the above-mentioned product classes with approximately 15 more green tons of sawtimber, 37 more green tons of biomass, and 41 more green tons in overall volume than in comparison to M6. Stand's 1 and 3 had higher chip and saw volumes in the biomass thinning models, however, Stand 2 did not. M6 and M2 had the two highest pulpwood volumes which were expected since they produced the least in all other product classes. Overall M2, the conventional stand with regards to spacing and harvest regime, performed the worst with regards to total volume produced.

#### Ptaeda Model Prices

All costs occurred at establishment for all models, regardless of harvest type or spacing configuration. Differences in cost pricing were due to the number of trees planted therefore resulting in Stand 1 and Stand 3's costs to be –\$354.79 an acre at year 0 while Stand 2's costs were only –\$345.98 an acre at year 0 (See **Tables 3**– **5**). Information regarding small-scale machine harvesting costs are not currently available therefore only costs incurred by the landowner could be calculated.

Overall profit was calculated for each harvest regime within each stand. Additionally, profits were calculated to both include as well as exclude profits from woody biomass to demonstrate the differences in revenue and final profits. This exclusion of biomass prices still assumed that the biomass was harvested at each cut, however, no profit was received by the landowner for this product. Profits for M1 were \$4041.33 an acre when biomass was included and \$3972.56 an acre without biomass prices included. M2 received \$3158.49 an acre for harvests with biomass and \$3130.22 an acre for harvests without biomass payment.

Profits for M3 were \$4282.56 an acre when biomass was included and \$4215.89 an acre without biomass prices included. M4 received \$3577.75 an acre for harvests with biomass and \$3548.18 an acre for harvests without biomass payment. Profits for M5 were \$3476.81 an acre when biomass was included and \$3409.75 an acre without biomass prices included. M6 received \$3105.95 an acre for harvests with biomass and \$3075.88 an acre for harvests without biomass payment.

Overall profit with a net present value at 3% was calculated to demonstrate to landowners what today's value of harvesting would be for all six model types. Three percent was used specifically for landowners to be able to compare results against today's interest rates. NPV's for M1 was \$1842.66 an acre when biomass was included and \$1777.17 an acre without biomass prices included. M2 received \$1367.67 an acre for harvests with biomass and \$1347.22 an acre for harvests without biomass payment.

NPV's at 3% for M3 was \$1957.67 an acre when biomass was included and \$1894.45 an acre without biomass prices included. M4 received \$1581.25 an acre for harvests with biomass and \$1559.88 an acre for harvests without biomass payment. NPV's for M5 was \$1563.66 an acre when biomass was included and \$1499.70 an acre without biomass prices included. M6 received \$1339.14 an acre for harvests with biomass and \$1317.50 an acre for harvests without biomass payment.

Internal rates of return were calculated for each stand's model to demonstrate the exact discount rate that would be received


TABLE

3


Pricing

and

Revenues

for

Stand

1.


TABLE

4


Pricing

and

Revenues

for

Stand

2.


by a landowner when net present value for the investment was zero. This method of evaluating capital expenditure proposals was chosen to more accurately depict potential benefits for landowners with regards to their investments choices with the 6 model options. IRR for M1 was 12.02% when biomass was included and 11.74% without biomass prices included. M2 received 10.46% for harvests with biomass and 10.41% for harvests without biomass payment.

IRR for M3 was 12.36% when biomass was included and 12.09% without biomass prices included. M4 received 11.21% for harvests with biomass and 11.16% for harvests without biomass payment. IRR for M5 was 11.28% when biomass was included and 10.98% without biomass prices included. M6 received 10.34% for harvests with biomass and 10.29% for harvests without biomass payment.

Overall, M3 procured the highest values in all categories with M1 a close second. M3 was third with regards to highest overall values, however, it was first in the one thinning category indicating that Stand 2 produced the highest profits in total. Stand 1 had the greatest variation between biomass thinning values vs. one thinning values while Stand 3 had the least variation. Comparing IRR values for biomass thinning with biomass vs. one thinning without biomass resulted in Stand 1 having the greatest variation at 1.61%, Stand 2 with a variation of 1.2%, and Stand 3 with a variation of 0.99%. Similar trends can be seen when comparing profits without NPV and profits with a 3% NPV for biomass thinning's with biomass vs. one thinning's without biomass in all stands.

### DISCUSSION

ArborGen's high-density planting technique of using OP trees in-between rows of MCP improved trees provides landowners with an excellent solution for today's plantation establishment concerns. By inter-planting non-genetically enhanced trees to be harvested for biomass or pulpwood, landowners are able to save money while still promoting larger volumes in sawtimber harvests in the final year as was seen in both the field study and Ptaeda model. Stand 2 was able to produce 10 green tons more biomass per acre than Stand 1 in the field study and was only 2 green tons less in the biomass thinning Ptaeda model. When market prices increase for woody biomass in the southeastern part of the United States, FlexStandsTM will be a viable option for landowners to increase their revenue.

Until that time, adding a biomass thinning to a FlexStandTM has already shown to increase final sawtimber volumes, as was seen when comparing the additional 32 green tons gained per acre from M3 vs. M1 in the sawtimber product class. This is a significant amount of volume added on a per acre basis. When assuming landowner's minimum acreage is 20 acres and multiplying that by the additional 32 green tons, that's an additional 640 green tons of wood to be sold at sawtimber prices which can make a significant impact on a landowner's final revenue value. IRR was also seen to be 0.36% higher in comparison to a conventionally spaced tract of land, all of which can add up in the long run.

FlexStandsTM also positively promote the use of small-scale harvesting during the stands initial thinning's as was observed during the field study where less than 1% damage was incurred in Stand 2. This is believed to be due to the wider spacing configuration which allowed the smaller machines to maneuver in-between rows easier than in conventional spacing with standard sized machines. Although Stand 1 had less than 5% damage throughout the stand, all of the damage incurred was due to the narrow row widths. Having standard sized machines would have likely resulted in significantly higher damage percentages resulting in fewer trees reaching sawtimber status. Even though no field studies were conducted using rectangularity, it can be inferred from the field studies above that less damage would have incurred in-between rows since spacing widths are even wider.

The Ptaeda model study resulted in rectangularity being the least favored in comparison to all other stands, however, it should be noted that when a biomass thinning was included, M5 still produced the third largest tonnage for biomass and total volumes. M5 also came in fourth in sawtimber volumes behind M4 and was fourth largest in profit and IRR values indicating this method is still a plausible option for landowners to increase their overall volumes and revenues. This option is best suited for landowners who do not wish to use smallscale harvesting machines but instead would rather harvest with standard sized machines throughout the life-cycle of the plantation.

T-tests within the field study depicted no statistical difference between any of the variables except between the weight of Stand 1 and Stand 2 with Stand 1 weight being greater. An explanation for this difference is not currently available. All trees were planted the same time of year in similar site and soil conditions. They came from the same nursery, were both OP designated trees, and received the same moisture amounts once planted. DBH was also slightly greater for Stand 1 however this number was not found to be statistically significant. Interestingly, tree height averages were slightly higher for Stand 2, however this number was also not found to be significant. Further research needs to be conducted to understand the differences in weight between the two stands.

Overall, both the field and modeling study verified that harvesting with one thinning only and using a conventional planting establishment regime will result in lower total harvested volumes. Incorporating a biomass thinning into a stands management plan will produce the highest volumes in regards to overall biomass, sawtimber, and total harvest volumes. This management style will also provide landowners an additional year of revenue to assist with establishment costs and further minimize the risk of waiting for final harvest. Once the biomass market becomes viable, landowners and loggers alike will reap the benefits of the increased revenue.

# CONCLUSION

As times continue to constantly change, so do our techniques and technology we use for loblolly pine plantation establishment and thinning's. Incorporation of biomass thinning harvests, alternative plantation spacing dimensions, and small-scale harvesting machines during initial thinning's all have the potential to provide the landowner with increased total volumes and more specifically increased sawtimber volumes. This increase in volume not only benefits the landowner but also the logger harvesting the unit. The additional volume provides an alternate incentive for incorporating biomass harvests or high density

#### REFERENCES


plantings into plantation establishment until market prices rise for woody biomass.

#### AUTHOR CONTRIBUTIONS

MD wrote this article. TG is for proofreading and providing primary input. TM, DM, and BV all assisted in providing input regarding the experimental design of the study.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer, JT, and handling Editor declared their shared affiliation.

Copyright © 2018 Daniel, Gallagher, Mitchell, McDonald and Via. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# In Situ Rheological Method to Evaluate Feedstock Physical Properties Throughout Enzymatic Deconstruction

Philip Coffman<sup>1</sup> , Nicole McCaffrey 1,2,3, James Gardner <sup>1</sup> , Samarthya Bhagia4,5,6 , Rajeev Kumar 4,6,7, Charles E. Wyman4,5,6,7 and Deepti Tanjore<sup>1</sup> \*

*<sup>1</sup> Advanced Biofuels Process Development Unit, Lawrence Berkeley National Laboratory, Berkeley, CA, United States, <sup>2</sup> Department of Chemical and Biomolecular Engineering, University of California, Berkeley, Berkeley, CA, United States, <sup>3</sup> Bio-Manufacturing to Market Program at University of California, Berkeley, Berkeley, CA, United States, <sup>4</sup> Center for Environmental Research and Technology, University of California, Riverside, Riverside, CA, United States, <sup>5</sup> Department of Chemical and Environmental Engineering, Bourns College of Engineering, University of California, Riverside, Riverside, CA, United States, <sup>6</sup> BioEnergy Science Center (BESC), Oak Ridge National Laboratory, Oak Ridge, TN, United States, <sup>7</sup> Center for Bioenergy Innovation (CBI), Oak Ridge National Laboratory, Oak Ridge, TN, United States*

#### Edited by:

*Allison E. Ray, Idaho National Laboratory (DOE), United States*

#### Reviewed by:

*Suyin Gan, University of Nottingham Malaysia Campus, Malaysia Ruchi Agrawal, Indian Oil Corporation, India*

> \*Correspondence: *Deepti Tanjore dtanjore@lbl.gov*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

> Received: *04 April 2018* Accepted: *30 May 2018* Published: *02 July 2018*

#### Citation:

*Coffman P, McCaffrey N, Gardner J, Bhagia S, Kumar R, Wyman CE and Tanjore D (2018) In Situ Rheological Method to Evaluate Feedstock Physical Properties Throughout Enzymatic Deconstruction. Front. Energy Res. 6:53. doi: 10.3389/fenrg.2018.00053* Feedstock physical properties determine not only downstream flow behavior, but also downstream process yields. Enzymatic treatment of pretreated feedstocks is greatly dependent on upstream feedstock physical properties and choice of pre-processing Technologies. Currently available enzyme assays have been developed to study biomass slurries at low concentrations of ≤1% w/w. At commercially relevant biomass concentrations of ≥15% w/w, pretreated feedstocks have sludge-like properties, where low free water restricts movement of unattached enzymes. This work is an account of the various steps taken to develop a method that helps identify the time needed for solid-like biomass slurries transition into liquid-like states during enzymatic hydrolysis. A pre-processing technology that enables feedstocks in achieving this transition sooner will greatly benefit enzyme kinetics and thereby overall process economics. Through this *in situ* rheological properties determining method, we compared a model feedstock, Avicel®PH101 cellulose, with acid pretreated corn stover. Novozymes Cellic®CTec2 (80 mg protein/g glucan) can reduce 25% (w/w) Avicel from solid-like to liquid-like state in 5.5 h, as the phase angles rise beyond 45◦ at this time. The same slurry needed 5.3 h to achieve liquid-like state with Megazyme endoglucanase (40 mg protein/g glucan). After 10.8 h, CTec2 slurry reached a phase angle of 89◦ or complete liquid-like state but Megazyme slurry peaked only to 64.7◦ , possibly due to inhibition by cello-oligomers. Acid pretreated corn stover at 30% (w/w) with a CTec2 protein loading of 80 mg/g glucan exhibited a solid-like to liquid-like transition at 37.8 h, which reflects the combined inhibition of low water activity and presence of lignin. The acid pretreated slurry also never achieved complete liquid-like state due to the presence of biomass residue. This method is applicable in several scenarios comparing varying combinations of pre-processing technologies, feedstock types, pretreatment chemistries, and enzymes. Using this method, we can generate a process chain with optimal flow behavior at commercially-relevant conditions.

Keywords: biomass, pre-processing, pretreatment, enzymatic hydrolysis, rheology, phase angle

# INTRODUCTION

Interest in alternative sources of sugar is increasing as novel approaches to biologically convert polysaccharides from renewable sources to fuels, chemicals, and materials are being devised rapidly. Lignocellulosic feedstocks have emerged as viable candidates but their biochemical deconstruction to simpler sugars requires the execution of several unit processes: harvest, storage, pre-processing, pretreatment, hydrolysis (or saccharification), conversion to fuels/chemicals (biological or chemical), and separation and purification to biofuel molecules (Satlewal et al., 2017). Enzymes employed in saccharification treatments cleave cellulose to individual long glucan chains and further hydrolyze them to simpler (monomeric and dimeric) sugars. A cocktail of synergistic cellulases is required to perform these multiple biochemical reactions. The synergy among three types of enzymes: (i) endoglucanases, (ii) exoglucanases, and (iii) β-glucosidases, produced by Trichoderma reesei, has often been used as a model to explain the mechanisms of cellulases (Agrawal et al., 2015a,b). Endoglucanase is instrumental in breaking down insoluble glucan strands to soluble oligomers (with degree of polymerization, DP < 6) and thereby reducing the viscosity of cellulose-rich solids. Exoglucanases breakdown the long chain soluble oligomers to cellobiose, which is further broken down to individual glucose units by β-glucosidase. This understanding of cellulolytic capabilities of cellulases has been developed based on research conducted for over a century, with cellulases uniquely identified in shipworms and molds in 1912. The application of cellulases in releasing lignocellulosic sugars was reckoned only in 1960 when alkali pretreated jute was found to be much more susceptible to enzymatic digestion than untreated jute, primarily due to the increased porosity of cellulose (Euler, 1912; Basu and Ghose, 1960; Selby and Maitland, 1967).

Dilute acid and ionic liquid pretreatments followed by enzymatic hydrolysis have enabled the release of >93 and 99% (of theoretical maximum, also referred to as TM) combined glucan and xylan yields from corn stover; however, these studies were conducted at low insoluble biomass concentrations (1 and 10% w/w, respectively) (Lloyd and Wyman, 2005; Li et al., 2010). Techno-economic analyses indicate that the use of higher insoluble biomass concentrations (>15% w/w) during deconstruction is critical for the economic success of commercial scale ethanol production from lignocellulosic sugars (Humbird et al., 2011; Agrawal et al., 2017). At these high insoluble concentrations (IC), with less water involved in the process, the size and thereby the capital and operating expenses of a refinery can be substantially reduced. However, high IC slurries pose significant processing problems, possibly resulting in substantial reductions of sugar yields and thereby defeating the economic cause altogether (Knutsen and Liberatore, 2010). The reduced presence of water exponentially increases viscosity of biomass, which poses a number of mass transfer challenges, especially reducing efficacy of enzymes (Geng et al., 2015). Figure S1 in Supplementary Information illustrates the impact of increasing IC on the ratio of available water to insolubles in the slurry. Video S1 was prepared to demonstrate the drastic change in rheology due to the reduced water availability on Avicel, a form of microcrystalline cellulose. Early on, Reese et al. stated that three main parameters influence enzymatic hydrolysis of cellulose: (1) hydratability, (2) particle size, and (3) pore size (Reese et al., 1950). Pretreated lignocellulosic biomass can be more hygroscopic than crystalline cellulose and can absorb water readily. However, at low water availability, pretreated biomass can be particulate in nature that can lead to the inhomogeneous application of enzymes at high IC, see Video S2. Pre-processing technologies are being devised to make feedstocks amenable to low water activity environments in pretreatment and enzymatic hydrolysis, to minimize particulate behavior and increase hydratability (Li et al., 2014). Biomass variability is another issue that pre-processing technologies can also address (Williams et al., 2016). However, such preprocessing technologies cannot be screened for performance in downstream operations with available analytical methods that are currently applicable for low concentration slurries alone (Hu et al., 2009; Goacher et al., 2012; Opitz et al., 2013; Malinowska et al., 2015).

Reese et al., as a first, correlated a drop in intrinsic viscosity with a drop in DP and progress of enzymatic hydrolysis of cellulose with enzymes from Trichoderma viride QM 6a (Reese et al., 1957). While viscosity provides useful information, rheological characterization can further this understanding through parameters, such as yield stress and viscoelastic properties (Funk and Dinger, 1994). Yield stress, the minimum stress required to induce flow in a material, has been studied substantially for both pretreated (Pimenova and Hanley, 2003; Knutsen and Liberatore, 2009; Stickel et al., 2009; Ehrhardt et al., 2010; Lavenson et al., 2011; Senturk-Ozer et al., 2011; Wiman et al., 2011; Samaniuk et al., 2012; Zhang et al., 2014) and enzymatically hydrolyzed (Roche et al., 2009; Viamajala et al., 2009; Geddes et al., 2010; Knutsen and Liberatore, 2010) feedstocks. But this parameter can be measured only on offline samples and not always in real time. While a few other authors (Knutsen and Liberatore, 2009; Stickel et al., 2009; Wiman et al., 2011) reported substantial studies on viscoelastic properties of biomass, which are ascertained through oscillatory testing, there have been no reports of real-time viscoelastic property measurements of feedstock hydrolysis. As such, no information on changes in physical properties of the high IC feedstock slurry occurring throughout an enzymatic treatment time is available. Viscoelastic parameters can differentiate solid-like and liquid-like behavior of a material and measuring them for enzymatic hydrolysis slurries gives an opportunity to quickly compare pre-processing technologies for viscosity reduction and faster transition to liquid-like behaviors in high IC slurries and thereby potential for improving biomass feed and overall economics. At the Advanced Biofuels Process Demonstration Unit (ABPDU), we developed an in situ viscoelastic property determining method that displays, in real time, the transition of high IC microcrystalline cellulose and other cellulose-rich materials from solid-like to liquid-like state during enzymatic hydrolysis with various cellulases.

## MATERIALS AND METHODS

## Rheology, Viscoelastic Properties, and Rheometer

Rheology can be used to classify materials as: elastic (ideal solids), viscous (ideal fluids), or viscoelastic, see Figure S2. Typically, most materials lie between the ideal scenarios of elastic and viscous states and are called viscoelastic materials, characterized through three interrelated viscoelastic parameters: elastic modulus, viscous modulus, and phase angle (G′ , G′′ , δ), see Equation 1. If stress and strain are plotted against a time period, the lag observed between these vectors is called phase angle, see Figure S2E. In simpler words, viscoelastic materials that are more viscous, or liquid-like, exhibit a phase angle above 45◦ , where the lag is higher and can rise up to 90◦ . For completely Newtonian-type fluids. This concept is applied throughout this study to depict changing viscoelastic properties of the high IC biomass slurries (Funk and Dinger, 1994).

$$\boldsymbol{G}^\* = \boldsymbol{G}' + i\boldsymbol{G}'' \tag{1a}$$

$$\mathbf{G}' = \frac{\sigma\_0}{\varepsilon\_0} \cos \delta \tag{1b}$$

$$G'' = \frac{\sigma\_0}{\varepsilon\_0} \sin \delta \tag{1c}$$

$$\tan \delta = \frac{G''}{G'} \tag{1d}$$

where, G<sup>∗</sup> , G ′ , and G′′ are complex, elastic, and viscous moduli; σ and ε are stress and strain, and δ is phase angle.

In this study, rheological characterization of all samples was conducted on a Kinexus Pro+ oscillatory stress controlled rheometer (Malvern Instruments Ltd., Worcestershire, UK) with Peltier cartridges, both cylinder and plate type. The cartridges are able to reach 200◦C with a stability of ±0.1◦C controlled by feedback from PT100 sensors and since cellulases are mostly active at 50◦C, all studies were conducted at this temperature in situ the rheometer. Rheometers are generally designed to test homogenous material where particle-edge effects, settling, and other such issues from insolubles in slurries are not expected to pose a problem. This study was focused on developing a novel method to study rheology of particulate pretreated biomass samples in real-time.

#### Substrates, Sample Preparation, and Enzymatic Hydrolysis

Preliminary study was performed on switchgrass (Panicum virgatum), generously provided by Dr. Putnam from UC-Davis, pretreated with an ionic liquid (IL), EmimAcetate (purity ≥90%, BASF, Ludwigshafen, Germany) (Li et al., 2013). Biomass was stored at 4◦C prior to milling and sieving to 20-20 mesh size. Later, Avicel <sup>R</sup> PH101 cellulose (11363 Fluka, Sigma Aldrich, Milwaukee, WI), with a particle size of 200µm, was chosen as a model material for this study. Two mixtures of (i) Avicel with xylan from beechwood (TCX0064, VWR International, TCI America, Portland, OR) at a 2:1 mass ratio and (ii) Avicel with xylan and alkali treated lignin (VWR International, TCI, Japan) at 5:3:2 mass ratio were also tested. Avicel <sup>R</sup> PH101 (11365 Fluka, Sigma Aldrich, Milwaukee, WI) with particle size of 50 µm were tested and compared with the model material. Finally, corn stover, kindly provided by Idaho National Laboratory, pretreated in dilute acid at UC-Riverside, and analyzed for composition at the ABPDU was used to test this method for application on lignocellulosic biomass, see Table S1 for composition (Sluiter et al., 2008a,b). Dilute acid pretreatment was carried out at 5% (w/w) biomass concentration on a dry basis at a temperature of 140◦C for 40 min. Corn stover was soaked overnight at room temperature in 1% (w/w) sulfuric acid and then pretreated in a 1 L Parr <sup>R</sup> (Parr Instruments, Moline, IL) batch reactor at an impeller speed of 200 rpm. The reactor was heated in a fluidized sand bath (SBL-2D, Techne Corp., Princeton, NJ) maintained at 350◦C for rapid heating. To stop the pretreatment, the reactor was quickly removed from the sand bath and quenched in water at room temperature. The pretreated corn stover was recovered over Whatman <sup>R</sup> No. 1 filter paper after washing with about 15 reaction volumes of deionized water.

Cellic <sup>R</sup> CTec2 and Cellic <sup>R</sup> Htec2 (Novozymes, Davis, CA), and endoglucanase derived from Aspergillus niger (EC 3.2.1.4, Megazyme International Ireland, Ireland) were tested at either 40 or 80 mg/g glucan. CTec2 protein concentration, per Pierce BCA protocol, was observed to be 190 mg protein/ml enzyme (Smith et al., 1985). Each rheological test was performed on a batch of sample prepared to be a total weight of 100 g. Solid components were analyzed for moisture content and weighed to achieve final predetermined dry weight. A liquid mixture containing 50 mL of 0.1 M citrate buffer (pH 5.4), 1 ml of 2% (w/w) sodium azide solution, and pre-determined enzyme loading was prepared prior to addition to solid fraction (Selig et al., 2008). The liquid mixture and solid components were added to a beaker and topped off with water to achieve the total weight of 100 g and a pH of 5.5. This batch was thoroughly mixed for homogeneity and equally divided into two portions, one was added to the cup of the rheometer for viscoelastic property measurements only. The other portion was added to a shake flask that mimicked the enzymatic hydrolysis treatment conditions (50◦C and corresponding mixing) in the rheometer. Samples for sugar analysis were drawn from the shake flask.

#### Method Development on the Rheometer

The geometric attachments on the rheometer tested in this study included: parallel plate and vane-in-cup geometries, see **Figure 1**. Linear viscoelastic regions were determined for pretreated biomass before enzyme addition. Stress and frequency sweeps were conducted to identify the linear viscoelastic region to be below 10 Hz and 10 Pa. It is essential for us to perform all rheological property measurements within the linear viscoelastic range as no universal rheological constitutive equation can predict the behavior of the material in the non-linear range. These parameters were used as a starting point for the method. Method development was particularly challenging due to bulk density of material, settling issues, sample weeping, evaporation, and choice of oscillation frequency and amplitude throughout the treatment process. Since the sample was enclosed inside the cup of the rheometer, most often, we were unable to detect a failure due to aforementioned issues until after the tests were

complete, usually after 24 h, making the method development process very time consuming. Trials conducted in developing the method are described in sections "Preliminary Study" and "Avicel as Model Material" through "The Challenge of Prolonged Viscoelastic Property Measurements" and **Table 1** is a summary of some key test results that led to go/no-go decisions through the method development process. While the progress of method development was protracted, much of the information from the failed trials could be applicable in comprehending viscoelastic properties of insolubles in several other scenarios.

#### Shake-Flask Internal Mixing and Sampling

The shake flask in the incubator with the biomass sample was initially set at 150 rpm. Since the mixing in this system is only external, a slower viscosity drop of the biomass sample was observed. On the other hand, faster viscosity drop was observed in the cup of the rheometer due to internal mixing from the vane geometry, internal mixing and a faster viscosity drop of the biomass were observed in the cup of the rheometer. To rectify this incoherence, up to 4 ball bearings were tested in the flask (see section Harmonizing with Shake Flask Studies). Every 3 h, approximately 0.5 g of sample was removed from the shake flask and exactly 0.25 g of this sample was weighed and diluted with water to 2.5 g on the balance, a 10 times mass dilution required to obtain enough liquid sample that could be filtered through 0.2µm PTFE centrifugal filters. This mass to mass sampling was essential for the initial high viscosity samples that did not yield any free water with the sugars. Even so, any sugarrich free water would be more concentrated with the soluble sugars and thereby not being representative of the entire batch (Kristensen et al., 2009). Samples were prepared in triplicate and stored at −20◦C until necessary for High Performance Anion Exchange Chromatography (HPAEC) analysis (Tanjore et al., 2009). Further dilutions were performed when required to fit the range of HPAEC standards and detection.

# HPAEC Sugar Measurement

Glucose (G1), cellobiose (G2), cellotriose (G3), cellotetraose (G4), cellopentaose (G5), cellohexaose (G6), and xylose (X1) were measured in this study. Determination of mono- and oligosaccharide content was performed on Dionex ICS-3000 <sup>R</sup> system (Thermo Scientific, Waltham, MA) equipped with a Dionex ED40 <sup>R</sup> electrochemical detector. Separation was obtained using CarboPac <sup>R</sup> PA200 analytical (3 mm × 250 mm) and guard (2 mm × 50 mm) columns heated to 30◦C (Thermo Scientific, Waltham, MA). Elution was performed using a gradient program (0–14% eluent B) over 18 min with 0.1 M NaOH (eluent A) and 0.5 M NaOAc containing 0.1 M NaOH (eluent B). Analytes were measured via pulsed amperometric detection (PAD) using a standard quadruple waveform.

Samples containing xylan required separation of glucose and xylose, which co-elute on the PA200 column. These samples were concurrently run on an Ultimate 3000 HPLC <sup>R</sup> system (Thermo Scientific, Waltham, MA) equipped with a Shodex Refractive Index <sup>R</sup> detector (Shoko Scientific Co., Ltd., Yokohama, Japan). Separation was obtained using an Aminex HPX-87H <sup>R</sup> analytical column (7.8 mm × 300 mm) (Bio-Rad, Hercules, CA) heated to 65◦C. Isocratic elution was performed using freshly prepared 0.005 M sulfuric acid over 10 min with a 4-min wash step with 0.025 M sulfuric acid and reequilibration between injections. Integration and analysis of samples was performed using Dionex Chromeleon <sup>R</sup> software (Thermo Scientific, Waltham, MA). Identification of monoand oligo-saccharide content was determined relative to known standards (Megazyme International Ireland, Wicklow, Ireland). All monomeric and oligomeric yields were calculated as a ratio to their polysaccharides, either glucan or xylan, see Table S2 in Supplementary for stoichiometric equations.

# RESULTS AND DISCUSSION

# Preliminary Study and Avicel as Model Material

With the availability of IL pretreated switchgrass (Li et al., 2013), we started rheological testing at 18% IC using the long 25 mm Vane-in-cup geometry at a 5 mm gap. In this preliminary study, only oscillation measurements were performed to determine the changing viscoelastic properties of the IL pretreated biomass. At the beginning, the material had a very low bulk density, 100 kg/m<sup>3</sup> , which gradually increased as the material was liquefied through enzymatic hydrolysis leading to a drop in the overall volume of the slurry. Although this test provided promising results during the initial stages of the experiment, the drop in the volume and thereby the level of biomass within the cup inevitably invalidated the use of long vane for reliable measurements. The vane geometry should be completely covered throughout the course of the study to avoid edge effects and provide repeatable TABLE 1 | Tabulation of a few tests that led to key decision points during method development.


and reliable data, see Figure S3 in Supplementary. Accordingly, a short vane (14 mm vane size) was purchased to develop the method without the risk of sample dropping below the vane's height in the cup. However, even the shorter vane was too long for these studies.

IL pretreatment renders biomass to be highly porous and hygroscopic. Even though the regenerated IL pretreated biomass had a large particle size (2–5 mm), the solid phase absorbed the aqueous phase almost instantly and remained suspended through enzymatic hydrolysis. However, crystalline cellulose, which represents the majority of commercially-available pretreated biomass, separates from the aqueous phase and settles. While an increase in bulk density of biomass through enzymatic hydrolysis could continue to be an issue with crystalline cellulose, several more variables exist that could influence establishing the applicable method to determine viscoelastic properties of all types of pretreated biomasses during enzymatic hydrolysis. As we have found and others reported (Palmqvist and Lidén, 2012; Tanjore et al., 2013), variations in feedstock type, pre-processing methods, pretreatment catalyst, and pretreatment severity all have a strong influence on the progress of enzymatic hydrolysis and cannot be underestimated while measuring viscoelastic properties. To develop a method that can be applied on samples of biomass from such a diverse set of unit operations, we chose Avicel as our model material. Without the influences of a particular pretreatment and/or choice of feedstock, a method based on Avicel, a more "pure" substrate, allowed us to closely examine enzymatic activity and its influence on viscoelastic properties and was more likely to yield reliable data from repeatable and consistent measurements.

#### Vane-In-Cup Geometry Was Most Suitable

We started with Avicel at a high IC of 35% (w/w), treated with 40 mg protein of CTec2/g glucan, and tested in 14 mm Vane-incup geometry with a 5 mm gap (z2). At 35% IC, Avicel slurry was not very amenable and would not uniformly spread in the cup of the reactor, see Video S1B in Supplementary. Several air gaps and drastic drop of slurry volume led us to lower insoluble concentrations. At 30% IC, even though Avicel had some free water in its system, it was much lower than in hydrolyzed IL pretreated biomass. This led to a lower drop in bulk density and thereby slurry height. However, a settling issue that we did not observe in IL pretreated switchgrass emerged during the test with 30% IC of Avicel. The high hygroscopicity and porosity of IL pretreated switchgrass allowed it to absorb and remain suspended uniformly in the slurry, which was not the case with the more granular and crystalline Avicel. Later, we found this suspension of biomass in slurries to be true for acid pretreated biomass as well. While IL pretreated switchgrass-water suspensions are apposite to consistent viscoelastic property measurements, we wanted to develop a method that was applicable to a broad array of feedstock-preprocessing combinations and thereby were bound to tackle the settling issue.

Initially, we attempted to manually mix the sample every hour, but settling reoccurred within 10 min of the mixing process. This mixing method was further erratic because of differences in manual mixing hour by hour, each time causing a small but noticeable change in phase angle. Since the initial consistency of sample was very thick at 30% IC, plate-on-plate geometry seemed to be a better choice to help overcome the settling issues Coffman et al. Rheological Characterization of Enzymatic Hydrolysis

and eliminate the effect of internal mixing occurring from the vane geometry in the cup. The very narrow gap between the plates (2 mm) on the rheometer allow for oscillation, but do not cause any internal mixing, much like in the shake flask. This method seemed to work very well for the first 12 h, but after this time period, as hydrolysis proceeded, the soluble and insoluble components of the mixture would separate. The solubles along with the water would leak out the sides of the plates. To be able to perform viscoelastic measurements longer than 12 h, we reverted to the vane-in-cup geometry. This time, to maintain the insolubles suspended in the mixture, continuous rotational viscosity measurements were incorporated into the sequence between intermittent oscillatory measurements. By combining both viscosity and viscoelastic measurements (30 s of oscillation followed by 5 min of rotational mixing), we created a continuous method that mitigated the settling issues. Though we obtained substantial viscosity data at a constant shear rate of 150 s−<sup>1</sup> , the primary objective of incorporating rotational measurements was to maintain continuous mixing in the rheometer cup and keep the insolubles suspended. However, this improved internal mixing of the biomass led to a poor correlation between rheology of the samples in the cup and shake flask.

### Harmonizing With Shake Flask Studies

To ensure that glucose measurements from shake flask would be representative of those from the rheometer cup, it was necessary that there was no discord between progresses of enzymatic hydrolysis in both vessels. Initially, Avicel samples in the rheometer cup exhibited a higher phase angle than the material in shake flask, indicating that it was more liquid-like, primarily due to internal mixing from the viscosity measurements. Even when operating at 200 rpm, the laboratory shaker could not deliver the increase in phase angle as the rheometer set up as the shake flask can only provide external mixing. This external mixing was detrimental toward drop in sample viscosity, especially when at high IC of 30%. The addition of 1–4 steel ball bearings in each shake-flask along with external shaking was tested to create a mechanism for internal mixing. But, even with four ball bearings, there was very little drop in viscosity as 30% IC would not allow the ball bearings to move, well up to 12 h. At this IC, the samples were very thick and had a very paste-like quality, so the ball bearings were simply trapped inside the sample and did not rotate until hydrolysis proceeded enough to release the liquid phase.

After performing several tests at the 28% IC, we reduced the insolubles in the sample to 25% IC. Only at 25% IC or less, the ball bearings were moving appropriately in the shake flask from the start of the experiment, enforcing this concentration as our final choice of IC. However, more than one ball bearing was leading to shear in the shake flask samples, now leading to a faster than preferable drop in viscosity and, once again, lack of correspondence with the behavior of the sample in rheometer. To avoid any shear related influences, we proceeded with a single ball bearing at 25% IC that matches the rheological profile in the rheometer. We proceeded with this set up to find out that our sample in the rheometer began to settle at the low initial IC even when the rotational sequence was set up as high as 200 s−<sup>1</sup> . The short vane was not effective in suspending the insolubles at 25% concentration. Luckily, the bulk density of Avicel at this concentration was much higher and the drop in slurry height was minimal. Consequently, we returned back to the long (25 mm) vane, which was adequate in preventing settling when used with the combined sequence of oscillatory and rotational parts at 150 s−<sup>1</sup> , as described earlier.

Despite the advantages of reduced settling and enhanced mixing, the long vane presented a new set of challenges. Applying the parameters optimized for the short vane on method with the long vane would consistently push the slurry phase angle to 90◦ (completely Newtonian) before the shake flask appeared liquid-like. With the short vane, our oscillatory measurements were conducted at oscillatory frequency and amplitude of 10 Hz and 10 Pa. After a few optimization experiments, we found that oscillation at 5 Hz and 5 Pa led to a reliable and replicable rheological method with the long vane, but only up to 24 h, beyond which, evaporation of water from the slurry in the rheometer cup annulled our efforts.

# The Challenge of Prolonged Viscoelastic Property Measurements

Most viscoelastic property measurements are performed off-line, within a time period of a few min. These offline measurements can still be conducted real time for a few predetermined samples, if the timeframe of test completion is within a few hours, e.g., hydrolysis of Solka Floc with cellulases. With the rate limiting action of endoglucanase, real time and in-line measurement of viscoelastic properties of crystalline cellulose during enzymatic hydrolysis for about 72 h was very challenging. Although very little volume loss was observed during the hydrolysis of Avicel at 25% IC, much of it was noticeably lost to evaporation after 24 h. Initially, to contain the water in the slurry, a layer of lighter density oil was added after setting the vane in the cup with the slurry (Klinkesorn et al., 2004). But this oil would gradually mix into the entire sample during rotational viscosity measurements. A heavier vacuum grease was the next alternative tested to create a seal sufficient to prevent evaporation (Sun and Gunasekaran, 2009). Although the grease worked a few times, ultimately the system was not replicable leaving us to seek out a different alternative. After several trials, we learned that, to obtain a replicable and reliable method and alleviate much of the difficulties we had previously experienced with moisture loss, we will have to curtail our measurements to a maximum of 24 h.

A time frame of 24 h was fast enough to avoid any uncertainty around evaporation yet slow enough to generate useful data needed to compare feedstocks, processing technologies, and enzymes. We doubled the enzyme loading to 80 mg protein/g glucan to evade any enzyme inhibition related issues and achieve the fastest possible hydrolysis for the substrate. Although this number represents a much higher enzyme loading than would be used in industrial applications, it provided a means to replicate the studies in the absence of evaporation issues. Oscillation frequency and amplitude of 5 Hz and 5 Pa were, once again, too vigorous for this higher enzyme loading where hydrolysis was complete within 2 h, while measurements using 1 Hz and 1 Pa required more than 12 h to reach a phase angle of ∼45◦ . Oscillation frequency and amplitude for this system was optimized down to 3 Hz and 3 Pa. Finally, we established a rheological test that included continuous 5 min of rotational viscosity measurement at 150 s−<sup>1</sup> followed by oscillatory test at 3 Hz and 3 Pa on 25% IC of Avicel at 50◦C with an 80 mg protein/g glucan CTec2 loading.

With a defined process for replicable hydrolysis occurring in similitude in both the shake flask and in the rheometer, we were now ready to apply it for the viscoelastic property characterization of different enzyme cocktails. As a first step, we simply triplicated our test using 25% IC of Avicel with 80 mg protein/g glucan enzyme loading using Novozymes CTec2. CTec2 was selected because it is a complete enzyme complex (endoglucanases, exoglucanases, and β-glucosidases) for cellulose breakdown. We found that the material transitioned into liquidlike state within 5.5 h and achieved near Newtonian-behavior at roughly 10 h, when the phase angle of the sample surpassed 45◦ and reached 90◦ , respectively, see **Figure 2A** with glucose concentrations a little lower than 31 and 44.8% (of TM), respectively. Another approach to find this point of transition is to follow the profile of elastic (G′ ) and viscous (G′′) moduli, see **Figure 2B**. G′ and G′′ represent the solid and liquid-like behaviors of the material. As long as G′ was higher than G′′, the material was still in the solid-like state. Once G′′ crossed over the profile of G′ and observed to be stronger than G′ , then the material had transitioned into the liquid-like state. Also, during this transition, complex modulus (G<sup>∗</sup> ), shifts from shadowing the profile of G′ to that of G′′. Throughout the entire hydrolysis process, this substrate-enzyme combination exhibited minimal detection of gluco-oligomers; cellobiose was the only quantifiable oligomer and was only detected during the first 3 h of hydrolysis. Total yield, as calculated by the sum of glucose and cellobiose at 96 h, was 85% (of TM).

## Rheological Testing Was Applicable With Other Enzymes and Substrates

In contrast to our method-establishing test, when Avicel at 25% IC was treated with only Megazyme's endoglucanase at 40 mg protein/ g glucan, a complete Newtonian-like behavior or 90◦ phase angle was never achieved, see **Figure 3A**. Since endoglucanase intrinsically hydrolyzes long chained polysaccharides only to cellobiose and not to glucose, we expected a slower transition from solid to liquid and a lower total yield. Surprisingly, as shown in the figure, we observed that endoglucanase alone can also surpass a phase angle of 45◦ in roughly 5.3 h. However, as expected, the final sugar yield (calculated by the sum of quantifiable oligomers) after 96 h of treatment was only 61% (of TM). Furthermore, the rise in phase angle begins to significantly taper off at roughly 7.5 h and never reaches complete Newtonian state. Again, this result was expected, reaffirming that endoglucanase alone is not sufficient for complete transition to Newtonian liquid-like behavior, let alone complete hydrolysis. The presence of a high concentration of oligomers (G2 = 13.8, G3 = 16.5, G4 = 1.1%, of TM) could have inhibited the activity of endoglucanase (Xue et al., 2015). Previously performed wet chemistry tests (Xiao et al., 2004; Hodge et al., 2009) and molecular simulations (Payne et al., 2011) have established such inhibition of enzyme activity, but through our method we were able to obtain this information in real time and potentially identify the onset of inhibition by comparing with CTec2 profile.

Another inhibition type detected was the effect of xylan and xylooligomers on enzymatic hydrolysis of cellulose. It is known that hemicellulose polysaccharides, their oligomers, and xylose inhibit cellulase enzyme (Qing et al., 2010; Kumar and Wyman, 2014). When Avicel was mixed with xylan in a composition similar to what is expected in plant cell-wall (2:1 ratio) and

treated with HTec2 along with CTec2, we initially observed a very high phase angle, shown in **Figure 3B**. This is a surprising find as hemicellulose is most hygroscopic among the three constituents of the plant cell wall (Acharjee et al., 2011). We expect that the lack of hydrogen bonding between xylan and glucan in our slurry mixtures makes it substantially different from pretreated biomass. Interestingly, from the start of the experiment, the phase angle of Avicel together with xylan mixture started to drop and stayed low for a 12 h time period and does not reach liquid-like state until after this time period. Xylose yields reached 62.1% (of TM) at 13 h but glucose yields were low at 29% (of TM). Glucose yields followed the expectation, per phase angle, i.e., glucose yield from Avicel at phase angle of 45◦ was about 30% (of TM). As expected, due to the enzyme complex, we did not see any oligomers in the samples, not even those from xylan. This finding bolstered our concept that viscoelastic properties of high IC slurries are good indicators of ongoing kinetics.

To further test this concept, we chose to vary composition of slurry by including lignin into the mix. Interestingly, Avicel with xylan and lignin at 28% IC had a profile somewhere between Avicel with xylan and Avicel alone, see Figure S4 in Supplementary. At initiation, the phase angle of this mixture was high but soon fell to near solid-like state, similar to that of the Avicel with xylan mixture, but then started to increase at a rate similar to that of Avicel, i.e., the hydrolysis process was more rapid in the presence of lignin. At low ICs, the addition of extracted lignin to an enzymatic hydrolysis slurry had shown to increase sugar yield from enzymatic hydrolysis (Nakagame et al., 2011). At high IC, improved rheology can be another motive for such additions. Apart from chemical composition, we also tested two separate particle sizes (200 and 50µm) of Avicel and as expected, the slurry with smaller particle sized Avicel reached liquid-like state in much lesser time. The 50µm Avicel required only half the time (0.67 h) of that of 200µm Avicel (1.42 h) to achieve liquid-like state.

To further our understanding of rheological changes of cellulose-rich lignocellulosic materials during enzymatic hydrolysis, we tested acid pretreated corn stover at 30% IC. Biomass itself is much more hygroscopic than Avicel, compare Videos S1, S3 in Supplementary. Even though acid pretreatment renders biomass less hygroscopic than IL pretreatment does, acid pretreated biomass is still more hygroscopic than Avicel. Therefore, interestingly, at 30% IC of acid treated corn stover, evaporation issues were substantially mitigated as water was very much trapped in the insolubles, see **Figure 4**. We were able to test the high IC corn stover hydrolysis for a prolonged time but due to low water activity along with presence of insoluble cellulose, we were not able to achieve near Newtonian behavior, see **Figure 5**. To estimate the maximum corn stover insolubles concentration that can be administered at the initiation of enzymatic hydrolysis, without creating mass transfer issues and affecting the kinetics of the process, it will be necessary to perform rheological tests at several high IC levels. But once such a concentration level that is conducive for enzymatic hydrolysis is established, we can continue to add untreated biomass into a partially hydrolyzed slurry to re-use the released cellulase enzymes and increase the concentration of sugars in the recovered aqueous phase.

#### Perspective

Our understanding of pre-processing technologies and pretreatments of biomass and their influence on enzymatic hydrolysis of the recovered cellulose-rich materials has increased substantially in the past decade. The effects of lignin modification, lignin re-deposition, xylooligomers concentration, and several other factors have been studied extensively, but typically only at low ICs (Selig et al., 2007; Qing et al., 2010; Sannigrahi et al., 2011). In this study, we have been able to show that viscoelastic property measurements of high IC slurries can potentially provide several useful insights that can improve the prospects of commercial cellulosic sugar production.

Techno-economic analyses indicate that, apart from IC, enzyme production and purification are some of the top contributors to cost of lignocellulosic sugars. Achieving high sugar yields during saccharification, and recycling enzyme after, can tip the scales for economic success of lignocellulosic sugarbased products. As biomass is hydrolyzed, viscosity drop follows and the enzyme attached to a binding site on glucan can be freed back into the aqueous phase. The released enzyme can

FIGURE 4 | Corn stover at various stages of treatment; (A) Untreated (89% w/w lnsolubles Concentration), (B) Dilute acid pretreated (36% w/w lnsolubles Concentration), and (C) Enzymatically hydrolyzed (Final Insolubles Concentration after 24 h was 28% w/w; Initial Insolubles Concentration was 30% w/w).

re-operate on remaining biomass by binding to a new site, possibly at a much faster rate due to a higher percentage of free water (Hodge et al., 2009; Geng et al., 2015). By comparing several pre-processed and pretreated biomass samples, we can identify the upstream technology that can enable a higher release of enzymes. Furthermore, after a certain level of hydrolysis is achieved, a pulsed feed of new pretreated biomass can be administered to the reactor. Such a fed-batch process can be effective only when care is taken that the new batch of hygroscopic biomass does not create enzyme mass transfer issues (Liu et al., 2015). With phase angle data from our rheological testing, along with further optimization, it will be possible to identify a suitable time for administering the new batch. Once again, if a pre-processing technology leads to faster addition of batches of pretreated biomass, it will be advantageous in providing higher concentrations of sugar in the liquid phase. Instead of measuring sugar release in real time, which is possible through already available in situ methods such as FT-NIR (Gierlinger et al., 2008), we developed a method to identify the time required to achieve liquid-like behavior, without performing any wet chemistry based analysis. This method is not mitigated by background effects that are inevitable when varying feedstock type and impact spectroscopic methods. Rheological behavior and hygroscopicity of biomass most definitely vary with feedstock type, harvest time, pre-processing methods, pretreatment chemistry, and other upstream events (Palmqvist and Lidén, 2012). However, feedstock or even enzyme variation is not a limiting factor for the method presented in this study as phase angle is a unit-less measurement of the behavior of a slurry. Real-time rheological testing can potentially provide unpredicted information that is not easily obtained through wet chemistry tests on samples that are taken on a predetermined schedule.

# CONCLUSIONS

We developed widely applicable real-time rheological assessment of cellulases in high IC slurries, which are particularly challenging to test through wet chemistry methods. Transitioning viscoelastic properties of cellulose-rich insolubles at high concentrations correlated well with kinetics expected from previously established theories around enzyme inhibitions due to lack of water activity and presence of oligomers, monomers, and lignin. Through our method, it is possible to test multiple pre-processing and pretreatment technologies and cellulases such as endoglucanases for their ability to quickly convert a high IC solid-like slurry to liquid-like state. Such a faster transition can lead to higher availability of free water, reduced mass transfer limitations, and improved kinetics.

## AUTHOR CONTRIBUTIONS

PC and JG designed analytical studies and PC performed them; DT and NM devised and performed the rheology experiments and performed them; RK, SB, CW devised pretreatment studies and SB performed them; PC, DT, RK, and JG wrote the manuscript. All authors read and approved the final manuscript.

#### FUNDING

This report is based upon work supported by the U.S. Department of Energy. We would like to thank Energy Efficiency and Renewable Energy (EERE) Division's BioEnergy Technologies Office (BETO) for the funding required

#### REFERENCES


to pursue this work. This work was also supported, in part, by the BioEnergy Science Center (BESC) (Contract DE-PS02-06ER64304) and Center for Bioenergy Innovation (CBI) administered by Oak Ridge National Laboratory (ORNL) from DOE BER Office of Science. Another form of support came from the Bio-Manufacturing to Market student internship program at UC Berkeley, which as funded by the DOE EERE under Award No. DE-EE0006026 as a member of The Advanced Manufacturing Jobs and Innovation Accelerator Challenge. The views and opinions of the authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.

#### ACKNOWLEDGMENTS

We thank Drs. Vicki Thompson, Kevin Kenney, and Chenlin Li from the Idaho National Laboratory (INL) for providing biomass feedstocks tested in this project. We appreciate the generous gift of Cellic CTec2 and HTec2 from Novozymes.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00053/full#supplementary-material


maltodextrin. Food Res. Int. 37, 851–859. doi: 10.1016/j.foodres.2004. 05.001


**Conflict of Interest Statement:** The handling Editor declared a past co-authorship with one of the authors, DT.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Coffman, McCaffrey, Gardner, Bhagia, Kumar, Wyman and Tanjore. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Three Way Comparison of Hydrophilic Ionic Liquid, Hydrophobic Ionic Liquid, and Dilute Acid for the Pretreatment of Herbaceous and Woody Biomass

C. Luke Williams\*, Chenlin Li, Hongqiang Hu, Jared C. Allen and Brad J. Thomas

*Idaho National Laboratory, Idaho Falls, ID, United States*

#### Edited by:

*Timothy G. Rials, University of Tennessee, Knoxville, United States*

#### Reviewed by:

*Shishir P. S. Chundawat, Rutgers University, The State University of New Jersey, United States Jinxue Jiang, Washington State University, United States*

> \*Correspondence: *C. Luke Williams luke.williams@inl.gov*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *15 March 2018* Accepted: *20 June 2018* Published: *09 July 2018*

#### Citation:

*Williams CL, Li C, Hu H, Allen JC and Thomas BJ (2018) Three Way Comparison of Hydrophilic Ionic Liquid, Hydrophobic Ionic Liquid, and Dilute Acid for the Pretreatment of Herbaceous and Woody Biomass. Front. Energy Res. 6:67. doi: 10.3389/fenrg.2018.00067* This paper examines the efficacy of ionic liquid (IL) pretreatment on seven different commercially harvested biomass types: corn stover, miscanthus, pine, sorghum, sugarcane bagasse, switchgrass, and wheat straw in an effort to improve the production of renewable fuels and chemicals from biomass derived sugars. Initial experiments screened the pretreatment of lodgepole pine, a particularly recalcitrant biomass feedstock, with nine different imidazolium based ionic liquids. After screening, one hydrophilic and one hydrophobic ionic liquid was selected for pretreatment tests on six commercially harvested biomasses. Ultimately, the hydrophilic ionic liquid functioned better for biomass pretreatment than the hydrophobic ionic liquid. These results were then compared to a traditional dilute acid pretreatment to examine the relative effectiveness of ionic liquid pretreatment across a variety of biomass and ionic liquid types. Total theoretical sugar yields after IL pretreatment varied widely by IL and biomass type and ranged from 4.9 to 90.2%. Dilute acid pretreatment showed consistent sugar yields for herbaceous material (from 71.4 to 80.8%) but low yield for lodgepole pine (22.8%). Overall, ILs showed the potential to reach slightly higher sugar yields than dilute acid and were particularly effective for woody feedstocks. More importantly, the sugar release kinetics for IL pretreatment were three times faster than dilute acid and gave maximum sugar yields after about 24 h. Additional characterization of IL treated materials included scanning electron microscopy (SEM), x-ray diffraction (XRD), and compositional analysis. SEM and XRD showed qualitative and quantitative reductions in cellulose crystallinity (respectively) that correlated well to improved sugar release during enzymatic hydrolysis for hydrophilic ionic liquids. However, reductions in crystallinity associated with hydrophobic ionic liquids resulted in lower sugar release during enzymatic hydrolysis. Compositional analysis generally showed increased sugars content for hydrophilic ILs and increased lignin content for hydrophobic ILs.

Keywords: ionic liquid, hydrophilic, hydrophobic, dilute acid, biomass pretreatment, herbaceous, woody

# INTRODUCTION

Clean and domestic energy is an important issue for the development of sustainable and secure global communities. The United States has great opportunities to encourage the development of renewable energy as increased demand from residential, commercial, industrial, and transportation sectors, combined with the available billion tons of domestic biomass resources, (Perlack et al., 2011) could spur the growth of green fuels, chemicals, and energy. Lignocellulosic biomass is uniquely suited to address fuel and energy production challenges given that it can be transformed into liquid fuels and chemicals (Johnson et al., 2007; Floudas et al., 2012; Williams et al., 2012) and reduces GHG emissions through the capture of CO<sup>2</sup> via photosynthesis. Additionally, utilizing waste biomass can create new jobs, as evidenced by the fact that in 2016 the Midwest US employed just over half a million people in clean energy industry jobs, (Trust, 2017) and second generation biorefineries are capable of generating further job growth (Miranowski, 2007). However, to achieve a robust second generation cellulosic refining industry two key factors need to be addressed: a stable supply of chemically consistent feedstock (Williams et al., 2016, 2017), and cost-effective pretreatment strategies due to greater hydrolysis recalcitrance than first generation feedstocks (Eranki et al., 2011; Banerjee et al., 2012).

While various biomass pretreatment strategies have been developed (Mosier et al., 2005; Alvira et al., 2010) this paper focuses on the comparison between a traditional dilute acid pretreatment and emerging ionic liquid pretreatments to understand, and evaluate, the effectiveness of ionic liquids. To date both dilute acid and IL pretreatment have been used for a variety of biomass types. A brief summary of previous work for both of these pretreatment strategies, and the results obtained, follows. Optimum dilute acid pretreatment conditions vary by feedstock (which explains the variety of conditions found in literature) but general conditions use millimeter size range particles, sulfuric acid at 0.5–2 wt%, temperatures from 100 to 190◦C, and reaction times between 5 and 60 min. The percent of total sugars yielded from the combined pretreatment and enzymatic hydrolysis of these samples tends to be in the range of 40 to 90% but often averages 65%. Materials pretreated under this range of conditions include corn stover, (Lloyd and Wyman, 2005) miscanthus, (Sørensen et al., 2008; Yoshida et al., 2008) spruce, (Larsson et al., 1999) sugarcane bagasse, (Jackson De Moraes Rocha et al., 2011) switchgrass, (Dien et al., 2006) and wheat straw, (Saha et al., 2005) among others. The range of conditions and yields given above are represented in the references for the individual feedstocks.

Many different ionic liquids have been investigated for lignocellulose degradation due to their tunable properties (based on a multitude of possible anion and cation combinations) (Brandt et al., 2013). IL properties reported to be important for pretreatment of lignocellulosic biomass include viscosity, melting point, dipolarity, hydrogen bond basicity, and hydrophobicity/hydrophilicity (Mäki-Arvela et al., 2010). Among the tested ILs, imidazolium based ILs have demonstrated high capability to dissolve/degrade lignocellulose (Cheng et al., 2012; Brandt et al., 2013; Gräsvik et al., 2014). Reaction conditions typically have 10 wt% solids with reaction temperatures from 70 to 160◦C and reaction times ranging from 30 min to a few hours. The percent of total sugars yielded ranges from about 60 to 90% (with an average around 70%) for most materials and ionic liquids tested. Materials treated with IL are more varied, and studies often treat blended feedstocks, such as; miscanthus/corn stover, (Shill et al., 2011) maple wood, (Lee et al., 2009) corn stover/pine/sugarcane bagasse, (Li et al., 2008) sugarcane bagasse with varying enzymes, (Qiu et al., 2012) wheat straw, (Li et al., 2009) and miscanthus (Brandt et al., 2011) (with a variety of ionic liquids that showed a strong dependence on the cation). Again, the range of conditions and yields given above are represented in the references for the individual feedstocks.

Biomass tested in this study includes corn stover, miscanthus, lodgepole pine, sorghum, sugarcane bagasse, switchgrass, and wheat straw, since they are commercially harvested and show great potentials for application in future biorefineries. The diversity represented in this selection spans pine as a softwood, sorghum, and corn stover stalks as representatives of grains, sugarcane bagasse as a special grass, with the other three feedstocks being traditional grasses. Nine imidazolium based ILs were selected for investigating performance variations between different types of ILs as listed in **Table 1**. These ILs cover a wide range of hydrophobicity. **Table 1** is organized first by the most hydrophilic cations, decreasing in the series, ethyl, butyl, and then arranged by the perceived increasing hydrophobicity of anions from Cl−, I−, OTf−, BF<sup>−</sup> 4 , to PF<sup>−</sup> 6 . The four ILs with Bmim<sup>+</sup> exhibit increasingly hydrophobic behavior as the anion varies from simple Cl−, to bulky OTf <sup>−</sup>, BF<sup>4</sup> <sup>−</sup>, and PF<sup>6</sup> <sup>−</sup>. On the other hand, if anions are same, the hydrophobicity of ILs depends on the length and structure of their cations' alkyl chain (Huddleston et al., 2001; Yee et al., 2013). The ILs with Cl<sup>−</sup> enable us to examine the effect of increasing alkyl chain length and structure from ethyl, to butyl, allyl, and butyl with two methyls on the performance of these ILs. Emim Ac has the highest basicity and has been demonstrated high capability to dissolve cellulose (Sun et al., 2009; Zavrel et al., 2009; Cheng et al., 2012; Brandt et al., 2013). Bmim Cl is one of the most studied ILs for dissolving cellulose and serves as a good reference (Sun et al., 2009; Gräsvik et al., 2014).

Published results encompass various biomass types with different treatment conditions including: type of ILs, feedstock type, feedstock particle size, particle water content, solids loading, dissolution time, and temperature (Mäki-Arvela et al., 2010; Gräsvik et al., 2014). This wide variety of reaction conditions and feedstocks causes great difficulty in comparing results. Attempts have been made to solve this problem for the ubiquitous dilute acid treatment by developing a generalized reaction severity parameters that takes into account for variations in pH, temperature, and reaction time (Pedersen and Meyer, 2010). However, a similar metric for ionic liquid pretreatment would be much more difficult to make given the almost infinite variety of ionic liquids available. Confounding this issue further is the

TABLE 1 | Ionic liquids used in the IL screening experiments.


wide array of conditions for enzymatic hydrolysis (with varying enzyme types and loadings) that occurs after pretreatment. This paper addresses these comparison issues by evaluating pretreatment effects of different ILs on different types of lignocellulosic biomass under identical conditions. Additionally, to aid the comparison with dilute acid pretreatment the IL pretreatment has been performed under conditions similar to what is now considered standard pretreatment conditions for dilute acid (Wolfrum et al., 2013). To date, only one paper has made a direct comparison between dilute acid and IL pretreatment for switchgrass, which showed that IL pretreatment has the potential to be far superior to dilute acid (Li et al., 2010).

This paper begins by screening nine different imidazolium based ILs for pretreatment of lodgepole pine. Next, the most promising and diverse candidates were tested across the variety of commercially harvested feedstocks and compared to a traditional dilute acid pretreatment. Additionally, a variety of physical and chemical characterization methods have been utilized to further understand the effects that diverse ionic liquids can have on this variety of biomass types.

#### MATERIALS AND METHODS

#### Biomass

Corn stover was collected by a single pass harvester in Boone County Iowa, Miscanthus was collected from Tift County Goergia, Sorghum was collected from Garvin County Oklahoma, Sugarcane Baggasse was collected from Pointe Coupee County Lousiana, switchgrass was collected from Garvin County Oklahoma, wheat straw was collected from Jefferson County Idaho, and Lodgepole pine was collected from Mineral County Montana. All biomass materials were processed on a Shute Buffalo mill through a ¼" screen before being ground to 500µm using a Wiley Mill and finally size reduced to 200µm on a Retsch Ultra Centrifugal Mill ZM 200. Ground samples were divided using a Retsch Sample Divider PT 100 to obtain a representative sample size for IL pretreatment. After deconstruction to such fine particle sizes biomass moisture content is ∼5% on a wet basis.

#### Dilute Acid Pretreatment

Dilute-acid pretreatment was performed using an ASE 350 (Accelerated Solvent Extractor, ThermoFisher Scientific, Waltham, MA, USA) using procedures developed at National Renewable Energy Laboratory (NREL) (Wolfrum et al., 2013). Experiments were performed using 66-mL zirconium cells, and a 10% (w/w) solids loading with an acid-to-biomass loading of 0.08 g g−<sup>1</sup> . Each cell was filled with 3.0 ± 0.03 g biomass and 30 mL of 1% sulfuric acid (w/w). Cells were subjected to a 7-min heating period followed by a 7-min static time with a reaction temperature of 160◦C, followed by 200 s of N<sup>2</sup> purge. The temperature was then reduced to 100◦C and 100 to 150-mL of nanopure water was rinsed through the cell with a 200 s N<sup>2</sup> gas purge and the rinsate was collected. Aliquots of the rinsate were collected for determination of total and monomeric sugars and organic acids using High Performance Liquid Chromatography.

# Ionic Liquid Pretreatment

Lodgepole pine samples were treated with nine different ionic liquids, summarized in **Table 1**, at 160◦C for 3 h. The dry biomass solids loadings was 10 wt% for all experiments at 15 g dry biomass and 135 g ionic liquid. After the reaction samples were washed three times with 300 mL of water (or ethanol in the case of ionic liquid BmimPF<sup>6</sup> due to the hydrophobic nature of the hexafluorophosphate anion), filtered through 5µm filter paper, and then dried at 40◦C for at least 48 h to recover the pretreated solids. After the initial IL screening experiments with pine the washing procedure was altered to 300 mL each of ethanol, acetone, then water for the feedstock screening experiments to better wash the biomass (particularly for the more hydrophobic ILs).

#### Characterization

Raw and pretreated solids were examined for structural and compositional alterations as well as altered reactivity. Structural analysis was performed by scanning electron microscopy (SEM) and x-ray diffraction (XRD). Biomass composition was analyzed using the National Renewable Energy Laboratory's standard laboratory analytical procedure (Sluiter et al., 2008) with the modification that the sample size was 0.2 mm instead of the usual 2 mm. Reactivity characterization was quantified using enzymatic hydrolysis. More detail about these procedures can be found below.

#### Scanning Electron Microscopy (SEM)

SEM images were taken for both untreated and pretreated materials using a JEOL Ltd. JSM 6610LV microscope. Prior to SEM analysis samples were prepared by mounting on standard aluminum pin stub mounts with biomass powder adhered via copper tape. Mounted samples were sputter coated with gold prior to analysis using a Hummer 6.2 Sputter System operating at 15 mA under 60 mTorr vacuum for 1 min. SEM images were acquired using a 15 kV accelerating voltage.

#### X-ray Diffraction (XRD)

XRD was performed using a Rigaku SmartLab diffractometer. Scans were collected at 40 kV and 44 mA with a step size of 0.01◦ . The degree of cellulose crystallinity can be inferred by the ratio of peak heights for the cellulose I in the 002 plane at 2θ = 22.5 and the amorphous cellulose at 2θ = 16.6 (Segal et al., 1959).

#### Enzymatic Hydrolysis

Enzymatic hydrolysis was conducted using a modified version of the procedure described in Selig et al. (2008). Pretreated solids were enzymatically hydrolyzed by adding biomass up to an equivalent of 1.0 g of dry solids to a 50 mL incubation flask with 5 mL of 0.1 M citric acid buffer (pH 4.8), 100 µL of 2 % sodium azide solution, and enough nanopure water to reach a final reaction volume of 10 mL. Enzymes were added at 40 mg g <sup>−</sup><sup>1</sup> dry biomass for Cellic <sup>R</sup> CTec2 (Novozymes, Franklin, NC, USA) and 4 mg g−<sup>1</sup> biomass for Cellic <sup>R</sup> HTec2 with enzyme and substrate blanks prepared as controls. To investigate sugar release kinetics, 150 µL aliquots of liquor were removed after 2, 4, 8, 24, 48, and 72 h of incubation at 50◦C, filtered through a 0.2µm filter, and analyzed for monomeric sugars using high performance liquid chromatography (Agilent HPLC Model 1260; Agilent Technologies; Santa Clara, CA). Sugars were analyzed on an Aminex HPX-87P column (BioRad Laboratories; Hercules, CA) with a column temperature of 85 ◦C using a refractive index detector, a mobile phase of 18 M ultrapure water, and a flow rate of 0.6 mL min−<sup>1</sup> . The sugars evaluated for this study, defined as the "total sugar yield" included the sum of released glucose, xylose, galactose, arabinose, mannose, and cellobiose divided by those sugars present in the pretreated material multiplied by 100. Duplicate injections were performed for each sample.

# RESULTS AND DISCUSSION

# IL Screening Experiments With Lodgepole Pine by SEM, XRD, and Enzymatic Hydrolysis

Initial IL screening experiments used lodgepole pine as a feedstock because of its recalcitrant nature. The pine samples were treated with the nine ILs shown in **Table 1** which was comprised of a set of imidazolium based ionic liquids. After IL treatment the pine was characterized using SEM (**Figure 1**) and XRD (**Figure 2**) for physical changes and examined for reactivity enhancement using enzymatic hydrolysis (**Figure 3**).

The raw pine shown in **Figure 1A** demonstrates clear wood fibers and texture, which serves as a point of comparison for the IL treated samples, while other ILs show different alterations on biomass structure. Emim Ac (**Figure 1B**) showed a significant amount of degradation. This degradation likely has a strong link to the high hydrogen-bond basicity of the IL. Hydrogen-bond basicity is strongly linked to the IL's capability to dissolve or swell cellulose/lignocellulose because it is important for the hydrogenbond accepting ability of the IL anions (Mäki-Arvela et al., 2010;

FIGURE 1 | SEM images of IL treated lodgepole pine. (A) Raw pine, (B) EmimAc, (C) EmimCl, (D) BmimCl, (E) AmimCl, (F) BdimCl, (G) MpimI, (H) BmimOTf, (I) BmimBF4, and (J) BmimPF6. The scale bar in each image is 50µm.

Zhang et al., 2010; Brandt et al., 2013; Cláudio et al., 2014; Gräsvik et al., 2014).

Cation hydrophobicity was another key parameter varied in this study. Four Cl<sup>−</sup> based ILs, Emim Cl, Bmin Cl, Amim Cl, and Bdim Cl, as seen in **Figures 1C–F** respectively, showed increasing cation hydrophobicity with increasing length and branching structure of alkyl substituents (Huddleston et al., 2001; Yee et al., 2013). It is clear that the IL with the shortest alkyl chain (Emim Cl) altered the biomass structure by the greatest degree. The effect of the other ILs could not be clearly distinguished from the SEM images. The final IL structural trend investigated was that of anion hydrophobicity. As seen in **Figures 1G–J**, Mpim I, Bmin OTf, Bmin BF4, and Bmim PF6, exhibit increasingly hydrophobic behavior as the anion varies from I, through OTf <sup>−</sup> and BF<sup>4</sup> − to PF<sup>6</sup> <sup>−</sup> (Huddleston et al., 2001; Yee et al., 2013). None of these bulky or increasingly hydrophobic anions appeared to have any significant effect on the biomass crystallinity. In fact, Bmim BF<sup>4</sup> and Bmim PF<sup>6</sup> appear to be even more crystalline than the original biomass.

These observations have demonstrated that hydrophilic ILs exhibit better capability for dissolving biomass than hydrophobic ones, when altering either the cation or anion. Hydrogen bond basicity also appears to have a significant impact on IL effectiveness. Clear trends emerge when comparing the hydrogen-bond basicity (seen in **Table 1**) across the suite of ILs tested. The hydrogen bond basicity of Emim Ac is 0.95, which is slightly higher than Emim Cl (0.87) and much higher than Bmim PF<sup>6</sup> (0.41), which is in the lowest part of range of ILs applied in this study and made material that appeared very crystalline. It is clear that the structure and components of the IL plays an important role in biomass dissolution by modulating the hydrophobicity/philicity and hydrogen-bond basicity of the IL.

The cellulose crystallinity, shown in the XRD spectrum of **Figure 2**, supports the visible changes seen with the SEM images. The XRD spectrum are color coded to denote the different aspects of the ILs in terms of altering the cation chain length or the anion hydrophobicity. The raw biomass spectrum can be seen in black while the most effective IL (with a short functional groups and hydrophilic anion, Emim Ac) can be seen in red. The ILs representing increasing cation hydrophobicity, and a constant Cl<sup>−</sup> anion, can be seen in blue progressing from Emim Cl, through Bmim Cl and Amim Cl to Bdim Cl.) The least effective ILs for biomass dissolution contained longer functional groups and hydrophobic anions. This can be seen in the green hued XRD spectra for Mpim I, Bmim OTf, Bmim BF4, and Bmim PF6. These XRD spectrum confirm that hydrophilic ILs with higher hydrogen bond basicity are more efficient in dissolving biomass than hydrophobic ILs.

Quantitative values for cellulose crystallinity for pine treated with nine different ionic liquids can be seen in **Table 2**. These values were calculated based on the method outlined by Segal et al. (1959). This analysis is based on evaluating the ratio of peak heights between the cellulose 002 peak (∼22◦ ) and the amorphous cellulose minimum (∼18◦ ). This method is useful for comparing differences between samples and is the most frequently used throughout the literature (at least in part due to the simplicity of use). However, this method has three shortcomings in its quantitative analysis as outlined by Park et al. (2010). These shortcoming include: (1) underestimation of the amorphous peak height (which leads to an overestimation of the CI), (2) contributions from crystalline cellulose, other than the 002 peak, are not accounted for, and (3) variation in peak width, which is also impacted by cellulose crystallinity, for the 002 peak is neglected. In terms of this work the Segal method can be used to compare differences by caution should be used when evaluating small differences in CI as they relate to enzymatic hydrolysis. The CI values in **Table 2** show that crystallinity is reduced for the Cl<sup>−</sup> anion ionic liquids and generally increased for the hydrophobic ionic liquids except, interestingly, for treatment with Bmim PF<sup>6</sup> where it appears that there was a slight decrease in crystallinity.

FIGURE 2 | XRD images of raw and ionic liquid treated lodgepole pine. Ionic liquid abbreviations can be seen in Table 1.

TABLE 2 | Crystallinity Index (CI) of pine treated with various ionic liquids.


*Crystallinity is determined using the method described in Segal et al. (1959). Samples labeled NA have no distinguishable amorphous cellulose peak so a CI could not be calculated.*

However, it should be noted that a crystallinity index for the Emim Ac, Emim Cl, and Bmim Cl could not be effectively calculated because the amorphous region has been convoluted with the cellulose 002 peak. It should be noted that directly relating cellulose crystallinity to enzymatic hydrolysis yields can be difficult given that cellulose accessibility can also be influenced by lignin and hemicellulose content/distribution, porosity, and particle size.

After examining the results of the SEM images and XRD patterns three ILs were selected for further reactivity characterization using enzymatic hydrolysis. The ILs used included Emim Ac, Emim Cl, and BmimPF6. Emim Ac and Emim Cl were selected to represent hydrophilic ILs because they showed the greatest overall amount of reduction in cellulose crystallinity and Bmim PF<sup>6</sup> was chosen as representative of hydrophobic ILs due to a longer functional groups, a hydrophobic anion, and the lowest hydrogen-bond basicity. The results from the enzymatic hydrolysis study can be seen in **Figure 3**. As expected, the untreated pine released very little sugar (about 14% of the theoretical maximum) and the pine treated with Emim Ac and Emim Cl released significant amounts of sugar (73 and 52% respectively). Interestingly, the pine treated

with Bmim PF<sup>6</sup> actually became more recalcitrant and showed a decreased sugar yield (9%).

Results from SEM, XRD and enzymatic hydrolysis show that the most effective ionic liquids for biomass dissolution has three attributes: (1) fewer functional groups (as seen with the ethyl instead of butyl groups on Emim Cl vs. Bmim Cl and only one methyl group instead of two on Bmim Cl vs. Bdim Cl), (2) polar protic anions (as seen with the acetate anion compared with the chlorine anion for Emim Ac vs. Emim Cl, and (3) hydrophilic anions (as seen with the acetate and chlorine anions compared with BF4, PF6, and to a lesser degree the OTf). These characteristics are in accordance with other studies where ILs that have relatively small cations (Zavrel et al., 2009; Mäki-Arvela et al., 2010) and small hydrogen-bond acceptor anions, (Zhang et al., 2005) are often efficient in dissolving cellulose. After initial screening tests two ILs were selected for further study: Emim Ac because it was highly effective for biomass dissolution, and Bmim PF<sup>6</sup> because of its' unique hydrophobic properties and to see if the decrease in sugar yields during enzymatic hydrolysis is consistent across many biomass types.

#### Comparison of Various Biomass Types Pretreated by Screened Ionic Liquids SEM

The screened ILs, EminAc and BminPF6, were used to pretreat seven biomass samples including corn stover, miscanthus, pine, sorghum, sugar cane bagasse, switchgrass, and wheat straw. These samples were examined by SEM, XRD, compositional analysis, and enzymatic hydrolysis in order to elucidate the different effect of hydrophilic and hydrophobic IL on various biomass types. The results from the IL pretreatment of the seven biomass types with Emim Ac and BmimPF<sup>6</sup> showed the same general trends as the pine sample. The SEM images for all of the IL treated biomass are shown in **Figure 4** where the raw biomass is in the left hand column, biomass treated with Emim Ac is in the middle column, and biomass treated with BmimPF<sup>6</sup> is in the right hand column. Untreated biomass shows a high degree of filamentous structure while Emim Ac treated biomass, for the most part, has a much more amorphous structure. Treatment with BmimPF<sup>6</sup> appears to make the material more crystalline for all biomass types tested in this study.

#### XRD

XRD spectrums for all untreated biomass, biomass pretreated with Emim Ac, and biomass pretreated with BmimPF<sup>6</sup> are presented in **Figures 5A–C,** respectively. Untreated biomass clearly exhibits a cellulose I peak in the 002 crystalline plane at 22 = 22.5◦ . Treatment with Emim Ac clearly shifts the main peak to 20.7◦ as cellulose I is transformed to cellulose II.(Sun et al., 2009) Interestingly, reacting higher ash biomass samples (lodgepole pine is low in ash) with BmimPF<sup>6</sup> adds several new peaks to the XRD spectrum as seen in **Figure 5C**. These peaks are likely the result of the hexafluorophosphate anion complexing with ash species (like sodium and potassium; Huber et al., 1997) in the biomass samples (Wang et al., 2017) which eliminates the recovery of the ionic liquid for recycle and would severely limit the use of these types of IL anions for biomass pretreatment.

Crystallinity index data for the herbaceous biomass can be seen in **Table 3**. As expected, the cellulose crystallinity for material treated with Emim Ac could not be calculated due to a complete removal of the amorphous cellulose minimum so this data was not added to the table. Interestingly, the crystallinity as measured by the method of Segal et al. generally decreased for the Bmim PF<sup>6</sup> samples. This indicates that an alternative method of measuring cellulose crystallinity would be useful in more carefully evaluating overall cellulose accessibility. Other potential methods include XRD peak deconvolution, and XRD based amorphous peak subtraction, and an NMR C4 peak subtraction method as outlined by Park et al. (2010).

#### Compositional Analysis of the Regenerated Biomass

The compositional analysis for the raw biomass, and the biomass after pretreatment with IL can be seen in **Table 4**. The changes in composition for the various biomass types can be generalized into two different groups. The first group comprises corn stover, miscanthus, sugarcane, switchgrass, and wheat straw. Interestingly, the second group contains pine and sorghum. General changes within these two groups, from the untreated case, is discussed for each type of ionic liquid.

After treatment with BmimPF<sup>6</sup> the first group of biomass showed an average decrease in glucan content of 31% with a standard deviation of 12%. Xylan content showed the same trend as glucan, but to a greater degree, and decreased by an average of 87 ± 12%. Given the large decreases in sugars after treatment with BmimPF<sup>6</sup> it can be expected that the lignin content should increase, and in fact it does increase by an average of 108%. Treatment with Emim Ac produced results that were opposite from the BmimPF6. In this case the lignin decreased by an average of 43 ± 15% while the glucan and xylan increased by 19.0 ± 15 and 39 ± 16% respectively. These compositional changes are generally in line with what previous literature has seen for treatment with Emim Ac (Brandt et al., 2013).

For the second group of biomass types (pine and sorghum) the Emim Ac and Bmim PF<sup>6</sup> showed similar trends. Both of these materials saw an increase (or very little change) in glucan,

FIGURE 4 | SEM images of biomass that is untreated (left), treated with Emim Ac (middle), or treated with Bmim PF6 (right). Biomass types are corn stover (A), miscanthus (B), pine (C), sorghum (D), sugar cane bagasse (E), switchgrass (F), and wheat straw (G). The scale bar in each image is 50µm.

xylan, and lignin content that corresponds well with the loss of the extractive from the untreated biomass (which is often around 10% for sorghum and 5% for pine) (Inl, 2015). Sorghum has rarely been investigated in the literature and appears to be resistant to dissolution by Bmim Cl (though the authors of that study also saw an increase in lignin content) (Zhang et al., 2011). While somewhat speculative, it is possible that the recalcitrance of sorghum is related to high amounts of p-coumaric and ferulic acids associated with the cell walls creating cross linkages between the lignin and hemicellulose (Billa et al., 1997). The p-coumaric acid would contribute to the formation of the p-hydroxyphenyl (H) lignol and provide recalcitrance in much the same way that the high guaiacyl (G) to syringyl (S) ratio causes recalcitrance in softwoods (Brandt et al., 2013).

While glucan, xylan, and lignin are important components of biomass structure another influential, but lower wt%, component is ash content. It can be seen in **Table 4** that the ash content of the biomass increased drastically for the majority of the samples treated with Bmim PF6. This increase in ash is concomitant with the new peaks appearing in the XRD data in **Figure 5**. These peaks are likely due to the fact that hydrophobic ILs exhibit a strong metal-complexing ability (Mehdi et al., 2010; Wang et al., 2017) which could lead to the hexafluorophosphate TABLE 3 | Crystallinity Index of herbaceous biomass types that are either untreated, or treated with Bmim PF6.


*Emim Ac treatemnt was left off the table because a CI could not be calculated due to the redcution in cellulose crystallintiy.*

anion binding with the ash present in the biomass, resulting in an increased ash content by making complexes like potassium hexafluorophosphate (Huber et al., 1997). However, while this metal complexation is a detriment to biomass processing the


TABLE 4 | Compositional analysis of several raw and IL treated biomasses where the abbreviations are as follows: Cnst, corn stover; Misc, miscanthus; Pine, lodgepole pine; Srgm, sorghum; SugCn, sugarcane bagasse; Swgr, switchgrass; Wht Str, wheat straw.

*The numbers in parenthesis are the 95% confidence interval.*

effect could have useful applications in other areas, like the recovery of rare earth elements (Wang et al., 2017).

Overall it can be seen that for less recalcitrant biomass treatment with Emim Ac generally delignifies biomass and increases sugar content while treatment with BmimPF<sup>6</sup> decreases sugars and increases lignin content. It has also been shown that sorghum is recalcitrant in much the same way as pine for treatment with both hydrophilic and hydrophobic ionic liquids. It was also shown that treatment with an IL that contains a metal complexing anion like PF<sup>6</sup> will increase overall ash content by binding with ash species and eliminating the possibility of recovering the IL.

#### Enzymatic Hydrolysis of the Biomass

Past research has shown that biomass structure and cellulose crystallinity play an important role in the effectiveness of enzymatic saccharification, where material with a less ordered structure often has a higher sugar release (Li et al., 2010). This is because cellulose crystallinity, lignin content, and hemicelluloselignin linkages all effect cell wall structure and altering these facets permits enzymatic access for hydrolysis. To investigate the effects of how much the altered structure of the biomass in this study would alter sugar release we performed enzymatic hydrolysis. This enzymatic hydrolysis was also compared to that of untreated biomass and a standard dilute acid pretreatment to assess the overall effectiveness of ionic liquid pretreatment. The enzymatic hydrolysis data for untreated, dilute acid treated, Bmim PF<sup>6</sup> treated, and Emim Ac treated biomass can be seen in **Figures 6A–D**, respectively. Two points should be noted in this figure: (1) that the legend visible in panel C applies to all of the panels in the figure, and (2) that the theoretical maximum for the sugar yield is the sum of release glucan, xylan, and cellobiose based on the composition data of the pretreated material. If the composition of the untreated material was used in all cases (assuming that the act of pretreating the material reduces the accuracy of the compositional analysis) then the data presented in **Figure 6** would change in the following ways: (1) the dilute acid pretreatment data would essentially remain the same, (2) the Bmim PF<sup>6</sup> data would decrease in yield by about 50%, and (3) the Emim Ac data would increase in overall yield by about 20%.

The data in **Figure 6A** indicates that the ability to release sugars from untreated biomass using enzymatic hydrolysis varies widely by biomass type. However, straight EH was more effective for corn stover and sorghum while being almost completely ineffective for lodgepole pine, miscanthus, and switchgrass. Dilute acid pretreatment on the other hand (panel B) is very consistent for herbaceous materials (including sorghum), yielding about 70% of the available sugars, while being almost completely ineffective for woody feedstocks like pine. Treatment with the hydrophobic ionic liquid Bmim PF<sup>6</sup> (**Figure 6C**) actually reduces the amount the amount of sugar recovered from the biomass. This is particularly interesting given the fact that the cellulose crystallinity, as determined by the XRD spectrum, showed a reduction in cellulose crystallinity. Traditionally,

a decrease in crystallinity is thought to be correlated with an increase in hydrolytic sugar release. However, this result contributes to the evidence that this trend does not always hold true (Park et al., 2010). It is possible that while the cellulose crystallinity has been disrupted the increase in apparent lignin content, as measured by the composition in **Table 4**, contributes to the resistance to enzymatic hydrolysis. Biomass pretreatment with Emim Ac (**Figure 6D**) shows significant potential across all biomass types (although the treatment seems to be highly variable compared with dilute acid). Perhaps most favorable aspect of IL pretreatment is the rapid sugar release kinetics. While the biomass treated with dilute acid continues to release sugar over the course of 3 days the Emim Ac treated material has reached the full sugar release in only 1 day (**Figure 6**). Shorter reaction time could reduce the size of the equipment needed which would improve mixing and heat transfer issues often associated with biomass processing. It is also worth noting that the Emim Ac treated biomass also had more sugars available,

arabinose, mannose, and cellobiose released divided by the sugars in the pretreated material.

based on the compositional analysis, so the fact that it reached such high theoretical yields, in such a short time, is even more impressive. Overall, it appears that using IL as a pretreatment instead of dilute acid provides significant improvements in sugar release kinetics and modest improvements in sugar yield.

# CONCLUSION

Pretreatment of pine,with a wide array of ionic liquids that contained varying cation and anion hydrophobicity, confirmed that hydrophilic ILs with polar protic anions work well for biomass deconstruction. The hydrophilic IL Emim Ac increases sugar content and decreases lignin content while the hydrophobic IL Bmim PF<sup>6</sup> decreases sugar content and increases lignin content for most herbaceous feedstocks. Interestingly, sorghum behaves similar to pine in terms of recalcitrance to IL pretreatment with both feedstocks not undergoing a large shift in the sugar content to lignin ratio. For all biomass types, treatment with Bmim PF<sup>6</sup> showed metal complexation with naturally occurring ash that increases the difficulty of IL recovery. In a three way comparison between hydrophilic IL, hydrophobic IL, and dilute acid pretreatment four conclusions can be drawn: (1) hydrophilic ILs exhibit sugar release rates that are three times faster than dilute acid, (2) total sugar release from hydrophilic IL treatment is highly variable, compared with dilute acid, but is better for recalcitrant feedstocks like pine, (3) hydrophilic ILs exhibit a slightly greater amount of cellulosic sugars release than dilute acid, and (4) hydrophobic ILs increase biomass recalcitrance which reduces sugar release.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

CW idea conception, majority of the writing and data analysis, some experiments. CL and HH idea conception, some writing. JA data analysis, majority of the experiments. BT some experiments.

#### ACKNOWLEDGMENTS

The authors would like to thank Kastli Schaller and Eric Fillerup for running the enzymatic hydrolysis and compositional analysis experiments. This research was supported by the US Department of Energy under Department of Energy Idaho Operations Office Contract No. DE-AC07-05ID14517.


**Conflict of Interest Statement:** The US Government retains and the publisher, by accepting the article for publication, acknowledges that the US Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for US Government purposes. The authors have no relevant affiliations, or financial involvement, with any organization or entity with a financial interest in, or financial conflict with, the subject matter or materials discussed in the manuscript.

The views and opinions of the authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.

Copyright © 2018 Williams, Li, Hu, Allen and Thomas. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Properties of Torrefied U.S. Waste Blends

Zhuo Xu<sup>1</sup> , Stas Zinchik <sup>1</sup> , Shreyas S. Kolapkar <sup>1</sup> , Ezra Bar-Ziv <sup>1</sup> \*, Ted Hansen<sup>2</sup> , Dennis Conn<sup>2</sup> and Armando G. McDonald<sup>3</sup>

*<sup>1</sup> Department of Mechanical Engineering, Michigan Technological University, Houghton, MI, United States, <sup>2</sup> Convergen Energy LLC, Green Bay, WI, United States, <sup>3</sup> Department of Forest, Rangeland and Fire Sciences, University of Idaho, Moscow, ID, United States*

Power generation facilities in the U.S. are looking for a potential renewable fuel that is sustainable, low-cost, complies with environmental regulation standards and is a drop-in fuel in the existing infrastructure. Although torrefied woody biomass, meets most of these requirements, its high cost, due to the use of woody biomass, prevented its commercialization. Industrial waste blends, which are also mostly renewable, are suitable feedstock for torrefaction, and can be an economically viable solution, thus may prolong the life of some of the existing coal power plants in the U.S. This paper focuses on the torrefaction dynamics of paper fiber-plastic waste blend of 60% fiber and 40% plastic and the characterization of its torrefied product as a function of extent of reaction (denoted by mass loss). Two forms of the blend are used, one is un-densified and the other is in the form of pellets with three times the density of the un-densified material. Torrefaction of these blends was conducted at 300◦C in the mass loss range of 0-51%. The torrefied product was characterized by moisture content, grindability, particle size distribution, energy content, molecular functional structure, and chlorine content. It was shown that although torrefaction dynamics is of the two forms differs significantly from each other, their properties and composition depend on the mass loss. Fiber content was shown to decrease relative to plastic upon the extent of torrefaction. Further, the torrefied product demonstrates a similar grinding behavior to Powder River Basin (PRB) coal. Upon grinding the fiber was concentrated in the smaller size fractions, while the plastic was concentrated in the larger size fractions.

#### Edited by:

*Allison E. Ray, Idaho National Laboratory (DOE), United States*

#### Reviewed by:

*Muhammad Aziz, Tokyo Institute of Technology, Japan Maria Puig-Arnavat, Technical University of Denmark, Denmark*

> \*Correspondence: *Ezra Bar-Ziv ebarziv@mtu.edu*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *12 April 2018* Accepted: *19 June 2018* Published: *13 July 2018*

#### Citation:

*Xu Z, Zinchik S, Kolapkar SS, Bar-Ziv E, Hansen T, Conn D and McDonald AG (2018) Properties of Torrefied U.S. Waste Blends. Front. Energy Res. 6:65. doi: 10.3389/fenrg.2018.00065* Keywords: waste, fiber, plastic, torrefaction, grindability, energy content, chlorine content, FTIR spectroscopy

# INTRODUCTION

The U.S. Environmental Protection Agency (EPA) has accelerated regulatory pressure on utilities burning pulverized coal by issuing carbon emission guidelines on June 18, 2014 (EPA, 2014). The EPA has proposed state by state goals to achieve CO<sup>2</sup> emission reductions; 30% from the power sector as compared to CO<sup>2</sup> emission levels in 2005 (EPA, 2015). The ultimate fate and form of the EPA proposed rule may not be known for some time until the rule-making process is complete but the past history of utility emissions regulation and Supreme Court decisions on EPA rulemaking authority indicate a high probability that some form of CO<sup>2</sup> regulation will be implemented (White, 2014). Internationally, the U.S. has announced the reduction of greenhouse gas emissions by 26–28% below 2005 levels by 2025 (Nakamura and Mufson, 2014).

Torrefied-biomass is a high-energy fuel that can be used in combustion, gasification, and pyrolysis, and is considered either fully or partially renewable and complies with the above EPA regulations (EPA, 2015). Kiel (Kiel, 2011) suggested the use of biomass for coal power plants. Potential users of torrefied biomass are suggested for refineries to produce bio-oil (Wang et al., 2016; De Rezende Pinho et al., 2017) and syngas producers (TRI, 2018). A considerable amount of studies, pilot-scale plants, patents and commercial efforts have been devoted to torrefaction and torrefied materials. The entries "torrefaction" and "torrefied" in the title, shows 790 papers, 19 reviews, and 50 patents, between 1990 and 2017. The 50 patents comprise many technologies for torrefaction, most of which are based on mechanical mixing. Although torrefaction technology is well developed, it has not yet moved to the commercial market. The consensus is that the main hindrance to the commercialization of this technology is the use of high-cost woody biomass as a feedstock (Kumar et al., 2017; Radics et al., 2017).

The use of wastes (for example, municipal solid wastes— MSW—or industrial manufacturing residuals—fiber and plastic blends) can be the answer to the deployment of this technology as tipping fees are paid for the waste destined for landfill. U.S. wastes possess substantial energy content that can be utilized for energy and power (US-EIA, 2010). Wastes, as a feedstock in torrefaction, has been suggested by Bar-Ziv and Saveliev (2013) and Bar-Ziv et al. (2016) and others, using regular torrefaction (Yuan et al., 2015), wet torrefaction (Mumin et al., 2017), and microwave torrefaction (Iroba et al., 2017a,b). Some difficulties have been recognized while using waste for torrefaction because of difficulties in conveying, pretreatment and potential emissions. Other hurdles were also identified while using waste feedstocks in torrefaction: (i) inconsistency in feedstock, (ii) possibility of high Cl, S, and N content, (iii) binders required for compaction of torrefied biomass (Bar-Ziv and Saveliev, 2013; Bar-Ziv et al., 2016), (iv) high moisture content in MSW and the like, and (v) high contaminant content that leads to emissions issues.

The EPA regulatory actions (EPA, 2014, 2015) regarding the use of alternative fuels raise the likelihood that torrefied waste will find a market to replace pulverized coal in energy production. One other recent development affecting the market for torrefied biomass from MSW was a memorandum from the EPA's Office of Air and Radiation addressing the framework for determining the carbon neutrality of biomass (McCabe, 2014).

There is a significant amount of waste in the U.S., which is being disposed of in landfills, that can be used as an energy source. **Table 1** summarizes the various wastes, totaling ∼110,000 ton per year, as well as their calorific values. This significant amount, if torrefied, can replace coal and be considered renewable and clean fuel. From an energy perspective, except plastic wastes with very high heat content ∼ 36 MJ/kg, the rest have heat values in the range 15–17 MJ/kg. The weighted average heat content in U.S. waste is ∼21 MJ/kg, which is comparable to that of Powder River Basin (PRB) coal that has a heat content of ∼17–19 MJ/kg (Luppens, 2011). This indicates that 1 dry ton of U.S. waste can replace 1 ton of PRB coal. With current coal consumption of ∼650,000 tons/d of coal in the US (with over 50% TABLE 1 | U.S. wastes, quantities, and heat content.


PRB coal) (US-EIA, 2018), U.S. waste could replace well over 15% of the U.S. coal.

The present paper deals with torrefaction of certain U.S. wastes, including plastics, which can be converted into drop-in fuels as a replacement of coal in coal power plants. Specifically, the paper deals with wastes blends from paper/carton (wood fibers) and plastics. As such, the torrefied fuel should be shown to match the characteristics and properties of coals.

#### MATERIALS AND METHODS

#### Materials

Convergen Energy (CE) developed a fuel engineering process: sorting and blending feedstocks of fiber and plastic, removing metal and shredding down to 25 mm by 1 mm flakes by which waste blends of fibers (from paper, label matrix residuals, and laminated non-recyclable papers/plastics and the like) and plastics, become uniform, flowable and consistent, with a bulk density in the range 200–300 kg/m<sup>3</sup> . CE also developed a pelletization process that produces pellets (12 mm OD and 50 mm long) that are rather uniform with a density of 750– 800 kg/m<sup>3</sup> and bulk density of 400–450 kg/m<sup>3</sup> . The binder for the CE palletization process was the plastic component in the blend. CE characterized their product for over 7 years with properties that showed rather consistent products. **Table 2** shows average properties of waste blends of 60% fiber with 40% plastics, with standard deviations of its product over a 7-year period. As seen, the properties in **Table 2** are indicative of reproducible and consistent material. This material was the feedstock in the torrefaction process, both in un-densified and densified forms.

In this study, both the un-densified as well as the densified material (pellets indicated above) were used. **Figure 1** shows both forms before torrefaction, used in this study: (a) un-densified CE material; and (b) CE pellets.

#### Waste and Product Characterization

The properties depicted in **Table 2** are part of the routine characterization of CE products, both before and after


pelletization. Other characterization methods are as follows. All data presented in this paper were averaged over 3–5 data points.

#### Grinding

Grindability is an important characteristic that has an essential impact on the applicability of torrefied material as a drop-in fuel in coal power plants. Typically, coal power plant use pulverizers of type MPS 89 (Storm, 2009), however, for the grinding tests, blade grinders (that operate at 24,000 rpm) were used. The grinding results presented in this paper are for comparison purposes. Two blade grinders were used in this study: Model CIT-FW-800 and Model CIT-FW-200. An on-line power meter— Wattsup pro was used for power vs. time measurements. Also, note that CE material was torrefied in both non-densified and densified (pellets) forms and grinding tests were carried out for both materials. Two types of grinding tests were performed as follows:

(1) A 100–200 g torrefied sample (either un-densified or pellet form) was placed in the grinder, which was continuously operated for up to 120 s time interval (to avoid damage to the motor); the power was measured continuously during the experiment. If necessary, grinding was repeated in a similar manner for a total of 1,800 s.

(2) A 100–200 g torrefied sample was placed in the grinder and operated for short time intervals – 15–30 s. After each grinding run (time interval) the pulverized material was sifted to seven sizes, in the range of 150–2,000µm, after which all size fractions were mixed and were further pulverized for another time interval. This process was repeated until the size fractions reached asymptotic values.

In both methods, the power was measured with and without the sample in the grinder. The power without the sample was subtracted from that with the sample, which provided the net power required to grind the sample. **Figure 2** shows a typical plot of power vs. time with and without a sample (in this case, 200 g of a torrefied non-densified material at 21.4% mass loss during torrefaction). Note that the startup is accompanied by an overshoot, in both cases.

#### Sifting

Sifting of the pulverized material was carried out in a W.S Tyler, RX-86 model sieve shaker. Seven size fractions were obtained with screen sizes of 75, 150, 180, 250, 425, and 850µm. At each time interval after grinding, all the material inside the grinder was taken out and put into the shaker to sift for an hour. The weights of all the screens before and after the sifting were measured. The

difference in these weights provided the sample weight of each size fractions.

#### Chloride and Chlorine

The chloride dissolved liquid samples from high shear mixing (described below) were diluted by a factor of hundred. Chloride was measured in this aqueous solution using Milwaukee Instruments, MI414 model Chloride Professional Photometer. Two cuvettes were used for the experiments. One is the blank sample filled with 10 ml of distilled water and another cuvette filled with 10 ml of diluted liquid sample. Then 0.5 ml of reagent-1 (Thiocyanate and Mercury) was added to both cuvettes, and after 30 s of swirling, 0.5 ml of reagent-2 (Nitric Acid) was added to both cuvettes. After another 30 s of swirling, the blank sample was first measured and zeroed, then the liquid sample was inserted in Chloride photometer which directly showed the chloride content of the liquid sample.

Total Chlorine in the solid phase was measured using the ASTM D4208-1 standard. The testing process included following key steps: The weighed solid sample was burned in a bomb filled with 2–3 MPa oxygen. After the combustion, a diluted base solution (2% Na2CO<sup>3</sup> solution) was added to the bomb to react with the chloride product. Water was then used to wash the inside cylinder wall of the bomb. All the washings were collected in a beaker and the ionic strength was adjusted using (NaNO<sup>3</sup> solution) (Zhu, 2014). The total chloride content of the solid material is determined by measuring the potential of the solution with a chlorine ion-selective electrode using a potentiometric titration (916 Ti-Touch) with silver nitrate solution.

#### Heat Content

Heat content was measured by Parr 6100 Compensated Jacket Calorimeter, where 1 g samples was placed inside sampling bowl/tray, and the sample was connected to the electric circuit using fuse string. This setup was put into a bomb and then filled with oxygen. The bomb was then put into a bucket with 2,000 ±0.5 g of distilled water. The process involved ignition of sample using an ignition circuit and subsequent measurement of temperature difference after the burning of the measured sample. The heating value was displayed by the calorimeter based on the calibration and temperature difference.

#### Moisture Content

Moisture content was measured using HFT-1000 moisture analyser. Around 1 g of sample was put into the analyser. After starting the analysis, the heating coil would heat up and the moisture inside the material would volatilize. The analyser would show the moisture content by measuring the difference of the weight before and after the experiment. Moisture content was measured before and after torrefaction. The values were rather consistent, before torrefaction moisture was in range 2–3% and after torrefaction, 0%.

#### Density Measurements

Density measurement of pellets was done using a scale (model A&D HR-60) with readability of 0.0001 g. The Archimedes' principle/buoyancy method was used for density measurement. A simple stand with suspended metal wire setup was used to dip the pellet in water. The procedure followed was as below:


#### FTIR

FTIR spectra were obtained on (i) 20 randomly selected pieces of mixed waste and (ii) screened fractions of the torrefied material (in triplicate) using a Nicolet-iS5 FTIR spectrometer, 64 scans, with an attenuated total reflectance accessory (ZnSe crystal, iD5) and data analyzed and averaged with the OMNIC v9.8 software and Aldrich, Hummel, and Nicolet spectral libraries. Carbonyl index (CI), cellulose index (CeI), and hydroxyl index (HI) were calculated as the ratio of the band intensity (absorbance) at 1,720, 1,024, and 3,342 cm−<sup>1</sup> , respectively, to the band 2,916 cm−<sup>1</sup> for the -CH2- groups (Wei et al., 2013).

#### Experiments

#### Torrefaction

Torrefaction experiments were carried out by placing a sample, motionless, at the center of a convection furnace, Lindenberg/Blue type BF51828C-1, with flow of inert gas, either N<sup>2</sup> or CO<sup>2</sup> to avoid oxidation of the material. For un-densified CE material, typically samples of 150 g were placed in a thin aluminum foil at the furnace center, with residence time in the range 1–40 min. For CE pellets, sample size was ∼300 g and torrefaction residence time was between 3 and 120 min.

#### Removal of Soluble Minerals

Soluble minerals in the torrefied material were removed by a method developed by Donepudi (Donepudi, 2017). In the present study, a 7.5 g torrefied sample was placed in a high shear mixer of Charles Ross & Son Company (Model HSM-100LSK-1) where water was added to the sample in 20:1 ratio by weight and the mixer was rotated at ∼7,000 rpm for 5 min. A suspension generated was filtered by 11µm porosity paper filter (Whatman 1001-0155 quantitative filter paper circles), followed by another filtration by 1.6µm porosity paper filter (Whatman 1820-047 glass microfiber binder free filter). The two filtration processes produced a transparent solution with no apparent suspend particles or colloids. The aqueous solution was measured for chloride as described above.

#### RESULTS

#### Torrefaction

As mentioned, all current torrefaction experiments were carried out by introducing un-densified material and pellets in a convective furnace at 300◦C, with the initial temperature of the particle, To, at ambient temperature. The material was placed in the furnace center and was kept stationary. In this case, the particle was heated by heat transported from the hot walls at temperature (Tw) to the particle surface by convection; the heat was then transported into the particle by conduction. Numerous torrefaction experiments were carried out for pellets as well undensified material. In both cases, the results show clear trends, with a delay in the onset of mass loss followed by an increase in the mass loss with time. The dynamic behavior in the two cases differed significantly from each other; for the un-densified material, the mass loss starts at around 3 min, whereas for the pellets, it starts at around 9 min. Further, for the un-densified material, mass loss increase with time was faster compared to pellets. This behavior was indicative to the heat-transferchemical-reaction system. To determine the regime that best fits the description of the system behavior, one should start with the analysis with Biot number (Bi) and thermal Thiele modulus (M); the former is related to the heating regime of the particle, and the latter relates to the propagation of the torrefaction reaction within the particle. The Bi and M, which are defined as:

$$Bi = \frac{h}{\lambda / L\_c} \tag{1}$$

$$M = \frac{R^\dagger}{\lambda / \langle \mathbf{c}\_{\mathbb{P}} L\_{\mathbf{c}}^2 \rangle} \tag{2}$$

where h is the convective heat transfer coefficient, λ is the particle thermal conductivity, L<sup>c</sup> is the particle characteristic length, R † is the torrefaction reaction rate within the particle, c<sup>p</sup> is the particle heat capacity, and ρ is particle density. The parameters required to determine Bi and M from Equations (1) and (2) are not easy to determine as the material is not well defined and therefore, can only provide an estimate. The value of heat transfer coefficient, h, was selected to be 10 (W/m<sup>2</sup> - K) and was the closest to the flow conditions prevailing in the furnace (Incropera and DeWitt, 2002). The value for thermal conductivity, λ, varies between 0.15 (W/m-K) for PVC, and 0.38 (W/m-K) for polyethylene (Incropera and DeWitt, 2002; Patterson and Miers, 2010); for biomass and fibers the values range in 0.03–0.29 (W/m-K) (Mason et al., 2016). A value of 0.2 (W/m-K) was selected which was an average of the above. Literature data on reaction rates of the material used were even more scattered than thermal conductivity, therefore they were measured by thermogravimetry in the furnace. The rate of mass loss of the CE material from both measurements at 300◦C was about 0.03%/s, where the material temperature has been equal to the wall temperature (Tw); using the density of each form to obtain a value of 0.2–0.3 (kg/m<sup>3</sup> -s) for the undensified material and 0.1–0.2 (kg/m<sup>3</sup> -s) for the pellets. In this study, the density was 1,150 (kg/m<sup>3</sup> ) for the un-densified material and 850 (kg/m<sup>3</sup> ) for the pellets. Heat capacity was both taken from the literature (Incropera and DeWitt, 2002) and measured to yield an acceptable value of 1,600 (J/kg-K) (Donepudi, 2017). The characteristic lengths of the two forms were measured (very accurately for the pellets and rather scattered for the un-densified material). **Table 3** summarizes all properties required for the determination of Bi and M, yielding values for (i) Bi of ∼0.1 for the un-densified material and ∼0.35 for the pellets and (ii) M of ∼0.01 for the un-densified material and ∼0.08 for the pellets. The values for Bi in the range 0.1–0.35 indicate that the rate of heat transfer by convection from the furnace walls to the particle was lower than the rate of heat transfer into the particle. The values of M are in the range 0.01–0.08 which indicate that the reaction rate was significantly slower than the heat transfer into the particle, and the particles equilibrate its temperature faster than the reaction rate. This analysis indicates that the reaction propagation was controlled by the rate of heat transfer from the furnace walls to the particle surface, after which the particle temperature equilibrates instantly.

Establishing that the torrefaction reaction rate was controlled by the heat transfer from the walls to the particle surface and that the particle temperature was uniform at all times, means that the reaction propagates with the rate of ramp-up of the particle temperature. To calculate the particle temperature, the equation of the heat rate, dQ(t)/dt, from the walls to the particle surface was needed to be solved, which was equal to

$$\frac{dQ\left(t\right)}{dt} = hA\left[T\_w - T\_s\left(t\right)\right] \tag{3}$$

where T<sup>w</sup> and Ts(t)=T(t) are wall and particle surface (or particle) temperatures, respectively. Q(t) is the heat required to increase the particle temperature, or

$$Q\left(t\right) = mc\_{\mathcal{P}}\left[T\left(t\right) - T\_o\right] + mh\_r \tag{4}$$

where m and c<sup>p</sup> are particle mass and specific heat capacity, respectively, T<sup>o</sup> is the particle core temperature, which is also equal to the initial temperature of the particle, and h<sup>r</sup> is enthalpy of reaction. It was a challenge to find values for h<sup>r</sup> as the torrefied material was not well defined, it comprises



fibers (mostly cellulose) and a large variety of plastic materials. Cellulose torrefaction in the 25–300◦C temperature range starts as an endothermic reaction and continues as an exothermic reaction (Bates and Ghoniem, 2013). Enthalpies of reaction for plastic in the same temperature range were always positive and vary in the range (12.55–147.86 J/kg) (Zhao et al., 2017), which is smaller than the value of cp(T-To) (∼400 kJ/kg) in Equation (4). Thus, for simplification, this term was ignored. Introducing Equation (4), without h<sup>r</sup> , into Equation (3) and integration from T<sup>w</sup> to T(t) yields

$$\frac{T\_{\text{w}} - T\_{\text{}}(t)}{T\_{\text{w}} - T\_{o}} = e^{-t/\tau} \tag{5}$$

where τ is a characteristic time, defined as

$$
\pi = \frac{mc\_p}{hA} \tag{6}
$$

For the pellets (cylinders), τcyl =dρcp/4h (d is cylinder diameter, ρ is particle density) and for the un-densified material (slab) it is τslab =dρcp/2h (d is slab thickness). Rearrangement of Equation (5) yields

$$T^\*\left(t\right) = 1 - (1 - \frac{T\_o}{T\_w})e^{-t/\tau} \tag{7}$$

T ∗ is defined as

$$T^\*\left(t\right) = \frac{T(t)}{T\_w} \tag{8}$$

To model the mass loss, the torrefaction reaction rate was assumed to be represented by a first order rate, which a rather common assumption in many torrefaction studies (Lédé, 2010; Funke et al., 2017), or

$$R^\dagger = \rho \frac{d\alpha(t)}{dt} = -\rho k \alpha(t) \tag{9}$$

where a=m/m<sup>o</sup> is ratio of mass-to-initial-mass, k is rate coefficient assumed to follow an Arrhenius behavior,

$$
\rho k(T) = A^\dagger e^{-T\_a/T(t)} \tag{10}
$$

where A † is a pre-exponential factor and T<sup>a</sup> is a characteristic temperature equals T<sup>a</sup> = Ea/R, E<sup>a</sup> is activation energy and R is gas constant. Introducing Equation (10) into Equation (9) and integrating yields an expression for the mass loss, 1-α, equals

$$1 - \alpha = 1 - (A^\dagger/\rho)e^{-T\_a/T(t)}\tag{11}$$

The required values for determining τ , Equation (6), for each case are given in **Table 3**. Introducing these values in Equation (6) yields τslab =184 (s) and τcyl = 475 (s), the subscript slab is for the un-densified material and cyl is for the pellets. Using these values, the particle temperatures were calculated and presented in **Figure 3**. As noted, the particle temperature in the un-densified case increases much faster than that of the pellets. Note from **Figure 3** the temperature of the un-densified material reaches the wall temperature after 10 min, whereas for the pellets, it reaches the wall temperature after 30 min.

The values for (A † /ρ) and T<sup>a</sup> were determined by fitting the model results for mass loss of Equation (11), using the temperature transients of Equation (7) (**Figure 3**), to the experimental results. **Figure 4** shows the measured mass loss vs. time data (scattered results) and the model results using Equation (11). Clearly, the model results yielded an excellent fit to the experimental data. The fitting process yielded for the un-densified material (slab) values of (A † /ρ) slab =1.23x10<sup>8</sup> and (Ta)slab =15,200 (K), and for the pellets (slab) values of (A † /ρ) slab =1.08x10<sup>8</sup> and (Ta)cyl =15,800 (K). The values of A † /ρ and T<sup>a</sup> for both forms of materials are very close to each other which is a strong indication that the model proposed here is representing the actual system behavior rather well.

#### Grinding Energy

The method of determining the grinding behavior has been explained above, with power that was continuously measured as a function of time during grinding for a given sample weight. Numerous grinding tests were conducted, in the mass loss range 10–51%, for the two forms of torrefied materials: un-densified and pellets. All net power transient results portrayed distinct behavior that showed two characteristic time: short and much longer. Further, the net grinding power transients for all samples fitted a double exponential rise of the form:

$$P(t) = a\_1(1 - e^{-t/\tau\_1}) + a\_2(1 - e^{-t/\tau\_2}),\tag{12}$$

where τ<sup>1</sup> and τ<sup>2</sup> are the short (1) and long (2) characteristic times, respectively, and a<sup>1</sup> and a<sup>2</sup> are the asymptotic values of the power for the short and long characteristic times, respectively.

**Figure 5** shows typical examples of the measured (symbols) net power vs. time of two 200 g samples during grinding of torrefied CE, un-densified material and pellets and fits (dashed

FIGURE 3 | Temperature transient for the un-densified material and the pellets, using Equation (7) and characteristic times of 160 (s) for the former and 475 (s) for the later.

lines) of the net power to Equation (12). In both cases, the short characteristic time was found τ1 = 9.2 s and characteristic time τ2 = 203 s.

All results for the torrefied samples and pellets in the range 10–51% mass loss were fitted to Equation (12) to yield: for the short characteristic time of τ1 = 9.1 ± 0.5 s, and for the long time it was τ2 = 203 ± 10 s with the respective asymptotic values of a1 = 378.1 W and a2 = 73.0 W that varied within ±5%. To demonstrate the general behavior of torrefied samples, **Figure 6** shows normalized net grinding power (by the asymptotic values) vs. time for the short time range, showing clearly identical behavior for all samples tested. The dashed line in the figure is a unity line that shows the normalized asymptotic value. The fact that the grinding dynamics is characterized by two characteristic times, that significantly differ from each other, indicates clearly that there are two materials. A detailed discussion of these two materials is given in the energy content section below.

As will be shown below, most of the material was ground in the short time range, thus a characteristic grinding energy can be determined by integrating the power over a certain time, which we selected as 1 τ g, 2 τ g, and 3 τ g (or, 8.1 s, 16.2 s, 24.3 s). **Table 4** shows the values of the specific grinding energy for three characteristic grinding time, 1 τ g, 2 τ g, 3 τ g, where

τ g =8.1 (s) in kJ/kg and in commonly used kWh/ton units. As expected, the specific grinding energies increases strongly with the integration time. The values determined here are similar to values obtained in other studies at 8.23 kWh/ton (Khalsa et al., 2016). For comparison, grinding characteristics of PRB were also studied with power vs. time results for a 200 g PRB coal sample shown in **Figure 7**. A fit of these results with a characteristic grinding time, τ g, of 8.1 was done and specific grinding energies were calculated as shown in **Table 4**. The values for the specific grinding energies for the torrefied (un-densified) material are within the experimental uncertainty to those of the PRB coal and smaller than the energy required to grind the torrefied biomass (Wang et al., 2017).

#### Sizing Distribution

Many sifting experiments were done as a function of grinding time (or grinding energy), where the samples were sifted in size range 150 µm−3 mm in 5 size fractions: x<150µm, 150<x<250µm, 250<x<425µm, 425<x<850µm, x>850µm (x denote size). It was observed that after reaching steady state (i.e., the net grinding power reached an asymptotic value), the size distribution did not change anymore. Therefore, most of the sifting experiments were done after reaching grinding steady state. The initial sample was around 100 grams, and after grinding and sifting, there was ∼1 g of sample loss during the transferring procedure, which occurred only once during the process. Therefore, loss was not more than 1%. Although there is scatter in the results, there are clear trends: the size fraction >850µm decreased with mass loss and the size fraction <150µm increased with mass loss and the size fractions in between did not change much with mass loss. Therefore, the behavior in two size fractions: under and above 850µm was further investigated. **Figure 8** shows size fraction as a function of mass loss for the torrefied un-densified material and pellets for these two size fractions. It is interesting to note that for each size fraction, the dependence on mass loss is rather similar (the line is a fit to a straight line). For the size under 850µm, its fraction starts at 82% for 4.5% mass loss and reaches almost 100% at 51% mass loss, the size fraction above 850µm balances the smaller size fraction. **Table 5** shows fraction >200 mesh of pulverized torrefied material at various mass losses. The table indicates that above 8.4% mass loss, after grinding the fraction of <200 mesh is >70%, which is consistent with of the typical coal power plant requirements (Helble et al., 1990).

## FTIR Spectroscopic Characterization

The CE waste mix plus fiber (20 random pieces selected) was analyzed by FTIR spectroscopy to determine their chemical identity with spectra library matching. The mix was shown to be comprised of three cellulose/paper, three polypropylene (PP), three polyethylene (PE), four polyethylene terephthalate (PET), silicone, three cellulose/silicone mix, two paper/acrylate mix and one nylon samples. A composite FTIR spectrum is shown in **Figure 9A** and shows the major bands associated with PE,


TABLE 5 | Fraction <200 mesh of torrefied material in various mass losses.


PP, PET and paper. No characteristic bands at 610 cm−<sup>1</sup> (C-Cl stretch) and 1,425 cm−<sup>1</sup> (C-H2 bending) were observed for polyvinylchloride (Krimm, 1963).

The major chemical changes that occurred upon torrefaction on densified and un-densified material and subsequent particle screening (<150µm, 150<x<250µm, 250<x<425µm, 425<x<850µm, and >850µm) after grinding were also monitored by FTIR spectroscopy. The spectra for the ground screened 425<x<850µm fraction for the densified torrefied (10, 20, and 42% mass loss) material as well as the CE-fiber mix are shown in **Figure 9A**. The spectra for the ground screened fractions for the un-densified torrefied (30% mass loss) material are shown in **Figure 9B**. Specific spectral bands can provide information on specific chemical changes that occur during thermal treatment (Balogun et al., 2017). All the samples had C-H stretching bands at assigned to methyl (2,960 and 2,870 cm−<sup>1</sup> ) and methylene (2,916 and 2,850 cm−<sup>1</sup> ) groups mainly associated with PP and PE plastic (Mayo, 2004a). In the ground screened torrefied material, plastic was generally concentrated in the larger sized fractions (425<x<850µm and >850µm) (**Figure 9B**). The O-H stretching band 3,100–3,600 cm−<sup>1</sup> was present in all samples and progressively decreased in intensity upon the extent of torrefaction due to dehydration reactions (Wang et al., 2014; **Figure 9B**). A broad carbonyl (C=O) band at 1,690–1,750 cm−<sup>1</sup> was observed and assigned to mainly an ester in linkage in PET and acrylate and an amide linkage in nylon (Mayo, 2004b). A small band at 1,505 cm−<sup>1</sup> was assigned to lignin from paper (Faix, 1992). The spectral region between 1,000 and 1,070 cm−<sup>1</sup> has been assigned to C–O stretching in wood cellulose and hemicellulose and decreased in intensity with torrefaction mass loss (Pandey, 1999). All samples were shown to have cis- and trans-vinylene bands at 727 and 974 cm−<sup>1</sup> , respectively (Miller, 2004).

The relative changes in carbonyl, cellulose and hydroxyl content to methylene groups (plastic) that occurred during torrefaction were examined by calculating CI, CeI and HI, respectively (**Figure 10**). Low values of CI, CeI and HI means that there was a higher level of polyolefin plastic in the material. The CI generally decreased for all torrefied samples with an increase in particle size (from <150µm to 425<x<850µm), except for the >850µm fraction (**Figure 10A**). For example, in the 30% mass loss torrefied material the CI decreased from 1.78 to 0.49 going from <150µm to >850µm particle size. For the low to moderate level of torrefaction (8–20% mass loss) the >850µm fraction the higher CI values could be associated with higher levels of PET plastic. Furthermore, the CI levels were also shown

index (CeI), and (C) hydroxyl index (HI) for ground screened fractions (<150µm, 150<x<250µm, 250<x<425µm, 425<x<850µm, and >850µm) of torrefied densified (D) and un-densified (U) mate.

to decrease, associated with cleavage of the ester linkages in PET/acrylates and removal of the volatile degradation products (Çepeliogullar and Pütün, 2014), with the extent of torrefaction. Generally, for both CeI (**Figure 10B**) and HI (**Figure 10C**) decreased for all torrefied materials as screened particle size increased (<150 to >850µm), suggesting that the cellulose fiber was mainly in the finer screened fractions. For example, in the 30% mass loss torrefied material the CeI and HI respectively decreased from 1.21 to 0.33 and 0.29 to 0.07 going from <150 to >850µm particle size. Again, at low-moderate torrefaction levels (8–20% mass loss), the CeI and HI levels were high, suggesting that undegraded paper fragments were collected in the >850µm fraction. Moreover, Both CeI and HI were shown to decrease as torrefaction severity increased. These findings support that the cellulose content decreased relative to plastic with the extent of torrefaction as a result of dehydration and degradation reactions (Wang et al., 2014).

#### Energy Content

The energy content was originally measured for un-sifted pulverized samples; however, it was discovered that scooping a sample of 1 g for the heat content test from a 200 g of the pulverized material gave very large scatter in the measured value. This was because the pulverized material has a large size distribution (as observed above) and the scooping did not necessarily give uniform size distribution. Therefore, it was decided to measure the heat content for five size fractions: x<150µm, 150<x<250µm, 250<x<425µm, 425<x<850µm, and x>850µm separately. Although the heat content for all sifted samples in these size fractions, for the sake of brevity heat content was shown for the following consolidated fractions: x<150µm, 150<x<850µm, x>850µm, and the calculated total heat content (from the fraction and heat content for each fraction). Heat content results presented here are dry- ash-free basis. **Figure 11** top-left is a plot of the heat content of the x<150µm fraction as a function of mass loss. The point at zero mass loss is the heat content of the blend prior to torrefaction and the dashed line is a linear trend line to lead the eye. Clearly, the main source of this fraction was pulp fibers that increase heat content with an increase in mass loss as predicted by Klinger et al. (Klinger et al., 2013, 2015a,b). **Figure 11** top-right is a plot of the heat content of the 150 µm<x<850µm fraction as a function of mass loss. The heat content does not seem to change with mass loss and has an average heat content of 35 ± 3 MJ/kg; this value was lower than that of plastic and it was assumed as a combination of fiber and plastic materials. **Figure 11** bottomleft is a plot of the heat content of the x>850µm fraction as a function of mass loss. The heat content does not seem to change with mass loss and has an average heat content of 41.5 ± 3.0 MJ/kg; this value was similar to most of the plastic material (Sonawane et al., 2017) and thus was attributed as plastic. **Figure 11** bottom-right is a plot of the total heat content, as calculated from all fractions, as a function of mass loss. The slope of heat content increase was identical to that of the fiber.

content.

Although the entire sample was pulverized, two materials (fibers and plastics) clearly retain their original structure which is indicated by the size distribution as shown above and the heat content as shown here. However, this material distinction diminishes as the torrefaction reaction proceeds (seen from the decrease of fraction x>850µm). To further quantify this process, a plot of the contribution of the <850µm fraction, which is a combination of torrefied material (from fibers) and fibers and the fraction >850µm, which was entirely from plastic. **Figure 12** shows results of the contribution to the total energy from each fraction, showing that the contribution from plastics was about 20% at about 5–8% mass loss and became zero at 50% mass loss, where the plastic lost its original integrity.

#### Chlorine Removal

There was evidence that at the working temperatures of the torrefaction experiments (300◦C) in this study, chlorine from the plastic materials should have been released as HCl (Saleh et al., 2014). Further, Bar-Ziv and Saveliev (2013) measured HCl in the torrefaction gas stream that was equivalent to the chlorine reduction in the solid phase. In the current study, numerous torrefaction experiments were performed as described above, and measured chlorine levels in the solid phase (see details above) with no evidence of any reduction of chlorine. This puzzling result can be explained by the way the current experiments were conducted, i.e., the sample was placed motionless. In this case, it was possible that in the time frame of the experiment, diffusion of HCl from the solid phase was so slow that it was not released during the experiment. However, in previous experiments by Bar-Ziv and Saveliev (2013), the material was torrefied in a stirred reactor (Zinchik et al., 2018) using much smaller size particles (∼1 mm) than in the present study and clearly showed that HCl was released.

As mentioned, high shear experiments with the torrefied material were conducted to obtain aqueous extracts which were filtered and measured for chloride in the solution and chlorine in the solid powder. **Figure 13** shows results of chlorine/chloride vs. mass loss; chlorine in solid after the high shear mixing and chloride in the filtrate (aqueous solution, adjusted for dilution). The scatter in the results was large and originate primarily from the fact that in these experiments, the samples were small (2–3 g) and the composition may differ significantly in its content and may not well represent the actual case. Nevertheless, there was a clear trend: (i) in the aqueous solution there was little-to-no chloride at zero mass loss (no torrefaction); (ii) the chloride in the aqueous solution increases gradually until ∼25% mass loss, after which it stays constant at an asymptotic value of 2,043 ± 207 ppm; (iii) chlorine in the solid phase has a value of 2,031 ± 129 at zero mass loss, then decreases gradually to ∼10% of the initial value.

## SUMMARY AND CONCLUSION

In the present study blends of fiber and plastic wastes at a ratio of 60:40 (fiber-to-plastic) were used as feedstock for torrefaction. Both the un-densified material and pellets were torrefied at 300◦C with different time periods. It was observed that the two forms have significantly different torrefaction dynamics. Undensified material takes less time to start torrefaction compared to the pellets, which is due to the faster heat transfer to the undensified material. The torrefied samples were characterized by moisture content, grindability, particle size distribution, energy content, molecular functional structure, and chlorine content. It was shown that although torrefaction dynamics is of the two forms differs significantly from each other, their properties depend on the mass loss. The fiber content was shown to decrease relative to plastic with the extent of torrefaction (mass loss) as determined by FTIR spectroscopy. Further, chemical (cellulose, hydroxyl, and carbonyl) changes were also shown to progressively decrease by torrefaction mass loss. Grinding characteristics, size distribution after grinding gave similar results as a function of mass loss during torrefaction, for the forms of material. Further, the torrefied product demonstrates

aqueous solution (adjusted for dilution).

a similar grinding behavior to PRB coal. The heat content of the material with size x>850µm is much higher than that of size x<150µm; the former attributed to the plastic material, whereas the latter was attributed to the fibers. The total heat content was shown to increase with mass loss. Chlorine in the torrefied samples was removed by a high shear mixing in aqueous solution showing that 5 min was sufficient to remove all chlorine after 30% mass loss. Overall, the waste blends studied in this paper showed that they can be used as drop-in fuel in coal power generation facilities, since this fuel is sustainable and low-cost, it also meets the environ mental regulation standard.

#### REFERENCES


# AUTHOR CONTRIBUTIONS

ZX carried out torrefaction experiments and characterization for heat and moisture content, grinding and size fractions. SZ helped in all above plus carried out some of the modeling work and took part in data analysis. SK carried our chloride extraction and characterization. AM analyzed the samples by FTIR spectroscopy and contributed to the manuscript in the appropriate sections and editing. TH and DC prepared the materials, characterized them and determined chlorine content in the solid phase. EB-Z supervised the entire work, analyzed the data, carried out the modeling work and wrote most of the paper.

value, pelletability and moisture adsorption. Biomass Bioenergy 106, 8–20. doi: 10.1016/j.biombioe.2017.08.008


for alternative fuel production and aluminum reclamation. Waste Manage. 67, 106–120. doi: 10.1016/j.wasman.2017.05.022


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Xu, Zinchik, Kolapkar, Bar-Ziv, Hansen, Conn and McDonald. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Preprocessing and Hybrid Biochemical/Thermochemical Conversion of Short Rotation Woody Coppice for Biofuels

C. Luke Williams\*, Rachel M. Emerson, Sergio Hernandez, Jordan L. Klinger, Eric P. Fillerup and Brad J. Thomas

Idaho National Laboratory, Idaho Falls, ID, United States

Preprocessing with air classification, followed by a hybrid biochemical/thermochemical conversion scheme, was utilized to improve the quality of short rotation woody coppice (SRWC) for biofuels production. Air classification improved sugar release during enzymatic hydrolysis by 6–12% for poplar and willow coppice respectively. Total theoretical sugar release for these hardwood coppices was ∼70%, which suggests that they could be utilized for biochemical conversion. Improved sugar yields after air classification were tied to compositional changes of reduced ash and extractives which can neutralize dilute acid pretreatment and inhibit fermentation. However, air classification was shown to have little to no effect on pyrolytic thermochemical conversion as it removed material without returning a significant improvement in liquid yield. It was also shown that pyrolysis of biochemical conversion lignin rich residue gives liquid yields comparable to whole tree (without any fractionation) pyrolysis, with a higher quality oil that has ∼60% reduced total acid number. Using this combined biochemical/thermochemical conversion strategy can improve yields of fermentable sugars and pyrolysis liquid above 80%, instead of the 60% yield of sugars or bio-oil when using a single conversion strategy. Overall, it has been shown that preprocessing and hybrid conversion pathways are a viable strategy for maximizing biorefinery viability.

Keywords: preprocessing, hybrid, biochemical, thermochemical, conversion, coppice, biofuels

# INTRODUCTION

In an effort to shift the world's energy paradigm away from finite petroleum resources there has been an increased focus on renewable energy, fuels, and chemicals. Biomass fills a unique role in this paradigm because, as the only carbon based renewable energy source, it is uniquely suited to the production of liquid fuels and chemicals. Other renewable energy sources, such as solar, wind, and hydro are effective for electricity production but are not amenable to the direct production of liquid and solid products. To produce these renewable fuels and chemicals it will be necessary to utilize the billion tons of residual biomass available in the United States (US Department of Energy, 2011). Currently, there exists two distinct pathways for biomass conversion; biochemical and thermochemical, and the choice of which one to use comes down to the inherent properties of the available biomass (Williams et al., 2017). This work makes an effort to bridge the gap between biochemical and thermochemical processes and identify strategies to effectively utilize short rotation woody coppice

#### Edited by:

Timothy G. Rials, University of Tennessee, Knoxville, United States

#### Reviewed by:

Selhan Karagoz, Karabük University, Turkey Abu Yousuf, Shahjalal University of Science and Technology, Bangladesh

> \*Correspondence: C. Luke Williams luke.williams@inl.gov

#### Specialty section:

This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research

Received: 15 March 2018 Accepted: 09 July 2018 Published: 30 July 2018

#### Citation:

Williams CL, Emerson RM, Hernandez S, Klinger JL, Fillerup EP and Thomas BJ (2018) Preprocessing and Hybrid Biochemical/Thermochemical Conversion of Short Rotation Woody Coppice for Biofuels. Front. Energy Res. 6:74. doi: 10.3389/fenrg.2018.00074 (SRWC), represented by hybrid poplar and shrub willow, for a hybrid (combined biochemical and thermochemical) conversion scheme.

Coppicing involves cutting the stand down just above the ground after 1 year to promote rapid growth of smaller stems that can be harvested every 3 years over a 20 year period (Volk et al., 2006). Coppicing also allows for high energy yields per acer (Njakou Djomo et al., 2015), maximized growth potential through active root systems (Mitchell, 1992; Al Afas et al., 2008), and an increased harvesting frequency as compared to short rotation forestry (Kauter et al., 2003). Coppice crops have recently been considered for both biochemical (Dou et al., 2017b) and thermochemical conversion (Dou et al., 2017a). Advantageously, these resources have compositional properties between traditional lignocellulosic (pine) and herbaceous (corn stover) feedstocks (Williams et al., 2016) which could make them well suited to a hybrid biochemical/thermochemical conversion. While clean chips of poplar and willow have been identified to have favorable carbohydrate contents for biochemical conversion processes (Volk et al., 2006; Sannigrahi et al., 2010), excessive leaf and bark material can negatively impact conversion (Wyman et al., 2009; Dou et al., 2017b). Additionally, higher lignin content of leaves and bark is a compositional aspect that should be considered, high lignin lends itself to thermochemical conversion processes; conversely, the higher ash contents in bark and leaves may cause problems with high temperature conversion (Carpenter et al., 2017; Zinchik et al., 2018).

One key benefit of a hybrid conversion pathway is the ability to more easily produce a combination of platform chemical precursors and fuel, derived from biomass based sugars and lignin rich residue respectively (De Wild et al., 2014; Ragauskas et al., 2014). Biomass derived sugars can either be fermented to ethanol and blended with fuel (the traditional use) or transformed into building blocks for high-value materials like p-xylene (Williams et al., 2012; Chang et al., 2014), toluene (Green et al., 2016), and butanol (Nigam and Singh, 2011). The lignin rich hydrolysis reside, on the other hand, is typically combusted for biorefinery power generation or used as a local animal feed supplement. Another option for this lignin rich residue is conversion into a bio-oil that could be further upgraded liquid transportation fuel. This paper will evaluate the effectiveness of converting hybrid poplar and shrub willow to sugars (through enzymatic hydrolysis) and bio-oil (through fast pyrolysis). In addition, SRWC anatomical fractions (separated by hand and using air classification as a preprocessing step) were passed through each conversion process to understand how anatomical fractionation alters conversion yields. Finally, a hybrid pathway, where the lignin rich residue from biochemical conversion was utilized for bio-oil production through fast pyrolysis, was also explored. Utilizing lignin rich residues from pulp and paper processes has only recently been demonstrated at scale (De Wild et al., 2014). To the best of the authors' knowledge, no studies have focused on the fast pyrolysis of lignin rich residue from a biochemical conversion process; in particular, not for emerging energy crops like poplar and willow.

# MATERIALS AND METHODS

#### Materials Sourcing

Hybrid poplar (P. deltoides × P. nigra.), from a 3 year coppice with 5 year old roots, was harvested July 14, 2017 from Kootenai County, Idaho using a New Holland FR 9080 forage harvester with a FB130 coppice header producing chip dimensions around 1 ′′ × 2 ′′ × ½′′. Shrub willow was harvested September 5, 2017 from Jefferson County, New York from Celtic Energy Farms plots using the same harvesting method as the poplar. Once received, approximately 200 kg of each of material was divided into representative subsamples using a custom rotary splitter. Each subsample consisted of 2.5–3.5 kg of wet material for air classification, anatomical fractionation, characterization, and conversion. Moisture content for air classification was 45.4% for hybrid poplar and 52.6% for shrub willow. Processed samples were dried at 105◦C for 24–30 h using a high performance horizontal air flow drying oven (SHEL LAB, model no. FX28-2, Cornelius, OR). Samples not processed within a week were stored frozen at −20◦C.

## Biomass Fractionation

Air classification (AC) was performed on wet chips using a 2x Air Cleaner equipped with an Iso-flow dewatering infeed shaker (Key Technologies, Walla Walla, WA). Air classification separates the samples into two fractions (light and heavy) by passing material over a screen covered fan. The lighter fraction (primarily low quality fines, leaves, and bark) is blown upward and pneumatically conveyed into a separate container than the heavy fraction, which passes over the fan. Several air velocities were tried until a flow rate was found that removed the majority of the leaves and bark but retained the majority of the chips (∼4.7 m/s).

Additionally, anatomical fractions were separated by hand to create a clean chip fraction (white chips without any bark), and an unclean mix (the fraction containing leaves, twigs, branches, free bark, chips with bark, and fines). This separation of clean chips and an unclean mix was chosen to represent the best possible fractionation scheme. Both air classification and anatomical fractionation was done in triplicate for all samples and the results of the different fractionation methods can be seen in **Table 1**. It should also be noted that this table also contains the results of the residual lignin rich residue that is left over after biochemical pretreatment.



Presented as the mean and (standard deviation). It should be noted that the light air classified sample is 100 minus the heavy fraction and the biochemical residue is the solids fraction left over after pretreatment and enzymatic hydrolysis.

#### Sample Preparation

Samples were ground to pass a 2.0 mm screen using a Thomas Model 4 Wiley knife mill (Thomas Scientific, model no. 3375-E55, Swedesboro, NJ) based on ASTM E1757-01 for compositional analysis and biochemical conversion. Samples were further ground to pass a 200µm sieve (Retsch ZM 200) for thermochemical conversion analyses. To prepare samples for pyrolysis approximately 1 g of material was compressed into a 16 mm square custom die with rounded corners at approximately 20,000 lb<sup>f</sup> for 1 min using a Carver press (3853-0C, Carver Inc.).

## Chemical Characterization

Compositional analysis was performed following the standard Laboratory Analytical Procedures for Compositional Analysis developed at NREL (Sluiter et al., 2010). The details of the procedure are summarized below. Extractives were removed using an accelerated solvent extractor (ASE) 350 (ThermoFisher, Scientific, Waltham, MA) three times with both water and



Whole tree material has undergone no separation step, the heavy AC fraction is air classified, and the clean chips were separated by hand. Values are given as a percent dry weight basis where the "total" value represents the final mass balance closure.

TABLE 3 | Pyrolysis yields for whole, hand fractionated, and air classified hybrid poplar for the mean and (standard deviation).


ethanol. Water soluble non-structural sugars and organic acids content was determined by high performance liquid chromatography (HPLC) after acid hydrolysis (using 72% sulfuric acid to make a 4% acid solution and autoclaving at 121◦C for 1 h followed by filtering through a 0.25µm filter). Organic acids were analyzed on an Aminex HPX-87H column (BioRad Laboratories, Hercules, CA), with a column temperature of 55◦C, using a diode array detector, a mobile phase of 0.01 M sulfuric acid, and a flow rate of 0.6 mL min−1. After neutralization with calcium carbonate, sugars were analyzed on an Aminex HPX-87P column (BioRad Laboratories, Hercules, CA) with a column temperature of 85◦C using a refractive index detector, a mobile phase of 18 M ultrapure water, and a flow rate of 0.6 mL min−<sup>1</sup> . Duplicate injections were performed for all samples. Acid-soluble lignin was determined by measuring absorbance at 320 nm with an ultraviolet-visible spectrophotometer (Varian Cary 50, Agilent, Santa Clara, CA) and calculating the concentration using Beer's Law with an extinction coefficient of 30. Protein was determined by measuring percent nitrogen using a LECO TruSpec CHN (St. Joseph, MI) and then multiplying that value by a nitrogen-protein conversion factor of 4.6.

## Thermochemical Conversion

Microwave enhanced fast pyrolysis was used for thermochemical conversion and is described here in brief with further details found in Klinger et al. (2016). The experimental apparatus consists of a quartz tube surrounded by an insulated microwave cavity (furnace) that can be independently heated (up to 550◦C) on one end and is cooled with dry ice on the other end, while being purged with N2. A 3 kW SM1250D model microwave generator (MKS Instruments, Andover, MA) was used in combination with a waveguide and microwave autotuners to focus a microwave beam onto 3 g samples (comprised of three one gram pellets described above) inside the cavity to achieve fast pyrolysis heating rates. An infrared camera (A655sc with 25◦ lens, FLIR) was used to monitor sample temperature during the reaction. In a typical experiment a region of interest at least 20 × 15 pixels (325 µm/pixel) was recorded to track the spatial and temporal temperature distribution. This spatial distribution varied by no more than 20–30◦C during any given video frame. A virtual instrument developed in LabVIEW (National Instruments, Austin, TX) was used to control the microwave power applied to the sample. Product vapors and gases are swept through the dry ice cooled section of the quartz tube to a filter that recovers condensable vapors while permanent gasses are passed through a digital gas flow meter (FMA-4312, Omega Engineering) and gas analyzer (7905A, Nova Analytical Systems). Gas analysis uses thermal conductivity for H2, an electrochemical sensor for O2, and NDIR infrared detection for CO, CO2, CH4, and total hydrocarbons (HC). Char is weighed after each reaction for a complete mass balance. Experiments were run in triplicate (or greater).

Bio-oil quality analysis was performed by ALS Environmental for carbon, hydrogen, nitrogen (CHN) using ASTM – D5291, for water by Karl Fischer using ASTM – E1064, and for acid number by ASTM – D3339.

# Biochemical Conversion

Dilute acid pretreatment was performed by accelerated solvent extraction using 1 wt% sulfuric acid at 160◦C. Enzymatic hydrolysis was conducted using a modified version of the procedure described in Selig et al. (2008). Pretreated solids were enzymatically hydrolyzed by adding 1.0 g of dry biomass (dried at 105◦C) to a 50 mL incubation flask with 5 mL of 0.1 M citric acid buffer (pH 4.8), 100 µL of 2% sodium azide solution, and the appropriate amount of nanopure water to reach a final reaction volume of 10 mL. Enzymes were added at 20 mg/g dry biomass for Cellic <sup>R</sup> CTec2 (Novozymes, Franklin, NC, USA) and 2 mg/g biomass for Cellic <sup>R</sup> HTec2. The density of all solutions and biomass were assumed as 1 g/mL. Enzyme and substrate blanks were prepared as controls. To investigate sugar release kinetics 150 µL aliquots of liquor were removed after 6, 12, 24, 48, 72, and 120 h of incubation at 50◦C, filtered through a 0.2µm filter, and analyzed for monomeric sugars using HPLC (Agilent HPLC Model 1260; Agilent Technologies; Santa Clara, CA). Sugars were analyzed on an Aminex HPX-87P column (BioRad Laboratories; Hercules, CA) with a column temperature of 85◦C using a refractive index detector, a mobile phase of 18 M ultrapure water, and a flow rate of 0.6 mL/min. Duplicate injections were performed for each sample. Each feedstock was analyzed through dilute acid pretreatment and enzymatic hydrolysis in triplicate. The sugars evaluated for this study, defined as the "Theoretical Yield" included the sum of released glucose, xylose, galactose, arabinose, mannose, and cellobiose divided by those sugars present in the pretreated material multiplied by 100.

# RESULTS AND DISCUSSION

#### Biochemical Conversion With Air Classification

A partial suite of the compositional analysis for the whole tree, air classified, and hand fractionated clean chips for hybrid poplar and shrub willow can be seen in **Table 2**. From this data it is clear that fractionation of SRWC provides several benefits for biochemical conversion. In this data, the clean chips material is comprised of the hand separated fraction of white wood chips that represent the best possible feedstock for biochemical conversion due to low extractives and high sugar. Air classified material approaches clean chips in terms of lower ash, fewer extractives, and increased sugars content; however, even with the removal of leaves, the lower quality of the remaining bark content attached to twigs, branches, and chips is still evident in the composition. Decreasing ash can lead to more effective dilute acid pretreatment, because ash can neutralize the acid, and reducing extractives typically increases enzymatic activity. Air classification boosted the hybrid poplar total sugar content above 59%, which is generally agreed upon as suitable for biochemical conversion (Davis et al., 2013). Additional compositional analysis data can be found in the supporting information for these same samples.

The effects of air classification and hand separation on enzymatic hydrolysis can be seen in **Figure 1** with hybrid poplar in panel A and shrub willow in panel B. The results presented here are for the theoretical yield, which is the amount of sugar released over the amount of sugar contained in the sample based on hydrolysis time. It is important to note that preprocessing steps like air classification can increase total sugar content, which means that hydrolyzing a gram of air classified material can ultimately release more sugar than hydrolyzing whole tree material, given the same theoretical yield. Unsurprisingly, for both materials the hand separated "clean" material gave the best conversion results with a yield of about 75% of the theoretical maximum for coppices. However, while clean chips of each coppice material gave similar maximum sugar yields the "whole" materials gave significantly different results; with hybrid poplar releasing 67% of the available sugars compared to 57% for the shrub willow. This drastic difference in sugar yields could be explained by the fact that the willow samples contained a greater amount of both ash and extractives, which inhibit pretreatment and hydrolysis, respectively. These results for hand separated coppice are in agreement with work performed by Dou et al. (2017b). It can also be seen that air classification improves enzymatic hydrolysis yield for both feedstocks. Improved results can be tied directly to decreased ash and extractives content in each of the samples (**Table 2**). Air classification removes basically all of the leafy material (and some of the bark in the case of hybrid poplar) which decreases the extractives content significantly. Error bars on the enzymatic hydrolysis data are significant (>5%) on a few of the data points due to the runs being performed in triplicate. These small sample size error bars make a strict interpretation of the results difficult. When considering potential improvements achievable by using air classification it is useful to consider the average error of for a larger enzymatic hydrolysis sample size. The average error for EH results across all 36 of the samples (which were run in triplicate) was 3.7%. If the sugar yield data is interpreted through a lens of ∼4% error it becomes clear that air classification makes a significant improvement. At 4% error air classified material is generally comparable to clean chips and better than whole tree material, especially for shrub willow.

# Thermochemical Conversion With Air Classification

Initial pyrolysis experiments investigated the potential benefits of air classification on fast pyrolysis. Theoretically, air classification could remove ash species and improve liquid yields. However, it should be noted that the thermochemical specification of <1% ash is not met (Jones et al., 2013) and more preprocessing would be required in order to upgrade SRWC to a conversion ready feedstocks. Results from the pyrolysis of whole tree hybrid poplar as well as hand separated material and both heavy and light air classified samples can be seen in **Table 3**. It can be seen that the whole hybrid poplar has a liquid yield of 59.81% while the clean chips and unclean mix yielded 65.28 and 50.84% liquid, respectively. The heavy fraction of material separated by air classification showed a significant decrease in ash based on compositional analysis (**Table 2**) but gave a liquid yield of 60.56% that was not statistically different than the whole tree sample. The light fraction of the air classified material had a liquid yield of 48.86% which was within error for the unclean mix sample. It

can also be seen in **Table 3** that char and gas yields for the air classified material followed the same trends shown by the hand separated material. Overall, pyrolysis of air classified material showed the expected trends based on the composition but there was no significant benefit to air classification when the total liquid yields were weighted by the mass fraction of material in each bin (i.e., **Table 1**). Additional data on the effect of varying air classification fan speed on pyrolysis yield can be seen in the supporting information; however, the results mirror what is seen in the manuscript.

# Feedstock Preprocessing Combined With Hybrid Conversion

Biochemical conversion exhibits a much greater sugar release after air classification preprocessing compared to the modest increase in pyrolytic liquid from thermochemical conversion of SRWCs. Additionally, the pyrolysis oil quality has potential to be improved when processing lignin rich leaf and bark residue separated by air classification and/or lignin rich residue from biochemical conversion.

Pyrolysis yields from the whole tree, hand separated clean chips and unclean mix, and biochemical conversion residue for both poplar and willow can be seen in **Table 4**. As expected, the clean chips had the highest liquid yields (of around 65%) while the unclean mix (which contains everything that is not clean chipped material) had the lowest liquid yields (around 50%). The whole tree material gave intermediate oil yields of 55– 60% for the willow and poplar, respectively. These intermediate oil yields are expected to be close to the yields for the clean chips given that the bulk each material is made of heartwood instead of bark and leaves. Leftover lignin rich residue from the biochemical conversion process gave oil yields similar to that of the whole tree material. However, char yields for the lignin rich residue were significantly higher than the whole material. This is likely the case because lignin is largely responsible for char-forming reactions when compared to cellulose and hemicellulose (Evans et al., 1986; Sharma et al., 2004; Patwardhan et al., 2011). Interestingly, the biochemical conversion residues produced less char residuals than previous studies on lignin, possibly because of the removal of extractives and inorganics. In this sense, the biochemical pathway acts as a pretreatment stage for the thermochemical conversion. In addition, previous studies have noted that differential thermogravimetric curves of lignin from enzymatic hydrolysis look more like natural lignin than organosolv lignin and may reflect the relatively high yield compared to prior work (Cho et al., 2012).

Apart from simply maximizing pyrolysis liquid yields it is important to understand something about the quality of the bio-oil. However, there is still very little consensus on what constitutes a high quality bio-oil; to date, primary measurements of bio-oil quality have centered on reduced acidity, as measured by the total acid number (TAN), reducing water content, and reducing oxygen content. For this work bio-oil quality metrics can be seen in **Table 5**. The TAN is highest for clean chips and whole tree materials and decreases slightly for unclean (leaf and bark rich) material. The TAN can be seen to have decreased significantly for the biochemical conversion residue. This indicates that the quality of the bio-oil produced from these waste products has at least some of the key qualities for a higher value oil. It also appears that the majority of the nitrogen resides in the lignin rich fractions of the coppice but there is no clear trend related to water content or carbon content based on the Karl Fisher or CHN analysis, noting that the relative measures of carbon are likely skewed by measuring the whole water-laden liquids. If the water measurements are directly factored out, the qualitative results are largely unchanged. For hybrid poplar the lignin-rich residues produced the highest carbon content liquid followed by the clean chips, while the highest fractions in willow was clean chips followed by lignin-rich residues. Improved measurement methods of water content, and



Mean (standard deviation).

TABLE 5 | Bio-oil properties from fast pyrolysis conversion.


accounting between bound feedstock moisture and chemically formed water through primary and secondary reactions, needs further future development.

It should be noted that the use of a combined bio/thermochemical process has the potential to increase the overall yield of valuable products. Simple biochemical conversion converts about 60% of the original mass to fermentable sugars while pyrolysis provides about 65% yield of liquid bio-oil products. By utilizing air classification as a preprocessing technique to remove material detrimental to biochemical conversion, followed by pyrolysis of the lignin rich biochemical residue (and air separated leaf/bark) in a hybrid conversion process, the combined yield of sugars and oil can be increased above 80% and significantly increase overall biomass utilization.

#### CONCLUSION

In conclusion, air classification is a simple preprocessing strategy that has the potential to improve sugars release during enzymatic hydrolysis by 6–12% for short rotation woody coppice. The total theoretical sugar release for these hardwood coppices was ∼70% which suggests that they could be utilized for biochemical conversion. Improved enzymatic hydrolysis yields after air classification can be tied to the compositional changes of reduced ash and extractives that can neutralize dilute acid pretreatment and inhibit fermentation. However, air classification has little to no effect on pyrolytic thermochemical conversion of coppice as it removes material without returning a significant improvement in oil yield. It was also shown that pyrolysis of biochemical conversion lignin rich residue gives liquid yields comparable to whole tree pyrolysis with a higher quality oil that has ∼60% reduced total acid number. Using this hybrid bio/thermochemical strategy can improve yields of fermentable sugars and oil to ∼80% instead of the 60% yield of sugars or oil when using a single conversion strategy. Overall, it has been shown that preprocessing and hybrid conversion pathways are a viable strategy for maximizing biorefinery viability.

# FUNDING

The US Government retains and the publisher, by accepting the article for publication, acknowledges that the US Government retains a non-exclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for US Government purposes. The authors have no relevant affiliations, or financial involvement, with any organization or entity with a financial interest in, or financial conflict with, the subject matter or materials discussed in the manuscript. The views and opinions of the authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.

# AUTHOR CONTRIBUTIONS

CW: idea conception, data analysis, primary writer; RE: idea contribution, data analysis, authored sections; SH: idea analysis and experimental work; EF and BT for experimental work and editing. JK: Idea refinement and experimental work for the thermochemical conversion section and draft editing.

#### ACKNOWLEDGMENTS

The authors would also like to thank Mark Eisenbies, Tim Volk, and Obste Therasme from the College of Environmental Science and Forestry, State University of New York and

#### REFERENCES


Brian Stanton and Rich Shuren from Greenwood Resources for providing materials. This research was supported by the US Department of Energy under Department of Energy Idaho Operations Office Contract No. DE-AC07- 05ID14517.

cropping systems for woody biomass production in the EU. Renew. Sust. Energy Rev. 41, 845–854. doi: 10.1016/j.rser.2014.08.058


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Williams, Emerson, Hernandez, Klinger, Fillerup and Thomas. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Organic Fraction of Municipal Solid Waste: Overview of Treatment Methodologies to Enhance Anaerobic Biodegradability

Kunwar Paritosh<sup>1</sup> , Monika Yadav <sup>1</sup> , Sanjay Mathur <sup>1</sup> , Venkatesh Balan<sup>2</sup> , Wei Liao<sup>3</sup> , Nidhi Pareek <sup>4</sup> and Vivekanand Vivekanand<sup>1</sup> \*

*<sup>1</sup> Center For Energy and Environment, Malaviya National Institute of Technology, Jaipur, India, <sup>2</sup> Biotechnology Division, Protein and Carbohydrate Research Laboratory, Department of Engineering Technology, College of Technology, University of Houston, Houston, TX, United States, <sup>3</sup> Anaerobic Digestion Research and Education Center, Biosystems and Agricultural Engineering, Michigan State University, East Lansing, MI, United States, <sup>4</sup> Department of Microbiology, School of Life Sciences, Central University of Rajasthan Bandarsindri, Kishangarh, India*

#### Edited by:

*J. Richard Hess, Idaho National Laboratory (DOE), United States*

#### Reviewed by:

*Abu Yousuf, Shahjalal University of Science and Technology, Bangladesh Muhammad Aziz, Tokyo Institute of Technology, Japan*

> \*Correspondence: *Vivekanand Vivekanand vivekanand.cee@mnit.ac.in*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *11 April 2018* Accepted: *12 July 2018* Published: *03 August 2018*

#### Citation:

*Paritosh K, Yadav M, Mathur S, Balan V, Liao W, Pareek N and Vivekanand V (2018) Organic Fraction of Municipal Solid Waste: Overview of Treatment Methodologies to Enhance Anaerobic Biodegradability. Front. Energy Res. 6:75. doi: 10.3389/fenrg.2018.00075* Organic fraction of municipal solid waste and its proper disposal is becoming a serious challenge around the world. Environmental pollution, public health risk, and scarcity of dumping land are the aftereffects of its improper disposal. Embodied energy recovery associated with the organic waste along with waste minimization may be achieved using anaerobic digestion. The chemical composition of the substrate plays a crucial role among the factors responsible for digestion performance and cumulative methane production. Treatment of substrate to enhance the digestion performance is gaining momentum in the recent years. This review provides an overview of different treatment methodologies including mechanical, thermal, chemical, biological, ultrasonic, and microwave approaches to enhance methane yield of anaerobic digestion of organic fraction of municipal solid waste (OFMSW). Environmental impact analysis of treatment techniques, along with comparison of treatment methodologies and techno-economic assessment, has also been discussed to provide a proper insight into the various processing methods.

Keywords: organic waste, biomass treatment, biogas, anaerobic digestion, environmental impact assessment

#### INTRODUCTION

The organic fraction of municipal solid waste (OFMSW) is widely used as a feedstock for anerobic digestion (AD) and promising source for biogas generation (van Lier et al., 2001; Abudi et al., 2016). AD is a natural practice that breaks feedstock into renewable fuel (here biogas) and digestate by microbial consortium in absence of oxygen. The biogas from AD comprises of CH<sup>4</sup> (50–70%), CO<sup>2</sup> (30–50%) and few notable impurities such as NH3, H2S, siloxane, halides and water vapors (Bhakov et al., 2014). Biogas generated by digesting OFMSW anaerobically may be utilized directly in combined heat and power (CHP) units, cooking or may be purified for transportation purpose (Parthiba Karthikeyan et al., 2017). The anaerobic treatment of digestion run-off can produce a sustainable residue to be employed as a soil amendment or re-condition certain type of soil (Castillo et al., 2006). Furthermore, another motivation to apply AD to treat the waste is that the producers can acquire emission reduction (CER) credits by clean development mechanism (CDM) defined in the Kyoto Protocol (International Panel on Climate Change (IPCC) Guidelines, 2006; Tong and Jaafar, 2006). Besides the environmental benefits of waste treatment (Edelmann et al., 2005), government agencies and private companies have extensively invested in AD technologies due to the increasing energy demand and environmental concerns such as global warming. Emphasis is mainly given to increase the generation of biogas, accomplish better rate of breakdown in the reactor, and subsequently minimized organic waste residue (Sonesson et al., 2000) to make the process economical.

Since the substrates used in AD are relatively complex, it has been reported that the hydrolysis is the rate-limiting phase among the four phases that take place during AD namely hydrolysis, acidogenesis, acetogenesis, and methanogenesis (Rafique et al., 2010; Ma et al., 2011). This remark was based on characterizing individual biodegradation kinetics of carbohydrate, lipids, and protein (**Table 1**). Hydrolysis step also generates by-products like volatile fatty acids (VFAs) that are undesirable for the bioreactor in excess amount (Neves et al., 2006; Lu et al., 2008) that inhibit methanogenic bacteria (Lu et al., 2008). Numerous studies have been reported using different treatment techniques to accelerate the hydrolysis phase (Val del Rio et al., 2011; Carlsson et al., 2012) without producing toxic by-products (Neves et al., 2006). Several methods have also been reported to recover components such as nitrogen and phosphorus from the waste and using them as soil amendments (Müller, 2001).

Complexity of organic waste may hinder the overall methanogenesis process (Kumar et al., 2018). This complexity may be altered by different pretreatment (Mechanical, thermal, biological, chemical, etc.) technique by breaking the molecular fencing around it. In a study performed by Cano et al. (2014), thermal hydrolysis of biological sludge, cow manure, grease waste, spent grains and OFMSW were attempted before AD. Results showed that methane production increased by 1.5 folds. Also, it was concluded theoretically, that integration of thermal hydrolysis may enhance up to 40% of plant's income. Biological co—pretreatment was attempted by Zhang et al. (2017) for enhanced methanogenesis. Food waste and waste activated sludge were investigated by the research group and results showed an increment of 25% in methane yield.

The objective of this paper is to discuss the complexity of the OFMSW and feasibility of various pretreatment methodologies to overcome its complex nature. Techno-economic aspects as well as environmental impacts of pretreatment are also discussed to explore the relation between pretreatment technique and its impact on the environment. Comparison of pretreatment is provided for the readers to get a deep insight of pretreatment methodology. At last, challenges associated with the biogas and its uses are discussed in detail for a deeper comprehension.


# CATEGORIZATION OF THE OFMSW

AD has been inclusive for the handling of the OFMSW from last few decades (Luning et al., 2003). Contemporary trend is focused toward the linking of anaerobic and aerobic processes (Pognani et al., 2009), to acquire a net energy gain by methane and the production of soil conditioner from the substrates. The OFMSW is assorted based on its composition, source and biological structure (**Tables 1**–**4**) (Abdullah et al., 2008) and hence, tremendous variations have been observed in its specific content depending on the source. The basis of categorization should include regional, seasonal and socio-economic impacts (Tyagi et al., 2018). OFMSW includes food waste, kitchen waste, leaf, grass clippings, flower trimmings and yard waste. Wasted food epitomizes a noteworthy proportion of organic material. Colonized and commercial kitchens leftovers also comprise of a considerable portion of OFMSW. Yard waste involves lignocellulosic-based materials explicitly grass off cuts and straw, leaves, weeds, bush and tree trimmings, whose production varies depending on the season.

### DIFFERENT METHODS USED TO TREAT OFMSW

Improvement of AD process depends on treatment that is a fundamental step and alters the substrate features. The biogas

TABLE 2 | Density of OFMSW based on the source (Campuzano and Simón, 2016).


TABLE 3 | Ultimate analysis of OFMSW at globe level (Campuzano and Simón, 2016).


*NA-Not Available.*


TABLE 4 | Bromatology of OFMSW in different parts of the world.

yield of OFMSW through AD is considerably affected by the substrate availability and mass transfer (Li and Noike, 1992). Treatment ensures increased accessibility of microbes to the nutrients and subsequent product generation (Naran et al., 2016). The basic target of treatment techniques i.e., mechanical, thermal, chemical, biological is to break the barrier (**Figure 1**) between the microbes and nutrient to accelerate the chemical reaction (**Figure 2**). Researchers associated with the treatment techniques of OFMSW have mostly opted for thermal, chemical and mechanical processes accounting for 24, 21, and 33% respectively (Mata-Alvarez et al., 2014). Hydrolysis phase is the most decisive step for the overall performance of the biogas digester as in this phase conversion of complex compounds to oligomeric and monomeric units occurs (Rafique et al., 2010; Paritosh et al., 2017a) and it may limit the rate of the subsequent reactions.

Also, bromatology of OFMSW plays a substantial role in the hydrolysis rate owing to varying properties of diverse substrates (**Table 1**).

#### Mechanical Treatment

Mechanical treatment crumbles compact assembly (**Figure 3**) of the substrates, consequently liberating OFMSW cell wall. Liberated cell wall has increased surface area for improved contact with microorganism for enhanced biomethanation (Kim et al., 2000; Elliot and Mahmood, 2012). Thus, mechanical treatment techniques such as (i) sonication, (ii) high-pressure homogenization and (iii) steeping or maceration are conducted to reduce the particle size of the substrate. It has been reported by Kim et al. (2000) that rate of utilization of feedstock is firmly depends on its size inversely for microbial consumption. Also, if the size of substrate is large, chemical oxygen demand (COD) solubilization decreases and it may result in declined production of biogas (Esposito et al., 2011).

Mechanical pretreatment strategy also affects the utilization pattern of substrate by microbes. Hartmann et al. (2000), discovered that while steeping the feedstocks, shearing effect played crucial role in increasing the gas production to the cutting of substrate fiber. In another reported study, the cell structure of the substrate was interrupted by sonication (Elliott and Mahmood, 2007). A group of researchers used Hollander beater to treat waste paper. The findings revealed that treating waste paper for 30 min had no significant impact on biogas yield. However, increasing time to 60 min augmented the biogas yield by 21% compared to untreated sample (Rodriguez et al.,

2017). Ultrasonic treatment has also been reported to perform the particle size reduction at a low frequency ranging from 20 to 40 kHz of sound waves, and enhance microbial penetration to the substrates (Chua et al., 2002).

A number of studies have verified that reduced particle size enhances the surface area accessibility to the microbes, resulting in improved nutrient availability and boosts up the anaerobic process (Mshandete et al., 2006). On the contrary, a decrease in particle size may increase the hydrolysis and acidogenesis that led to the generation of soluble organic material as VFAs, resulting in excessively high organic loads in AD reactor. Izumi et al. (2010), studied the effect of particle size reduction and solubilization on biogas production from food waste (FW). The study showed that grinding FW by bead milling (**Figure 4**) led

to 40% more solubilisation of COD; leading to 28% enhanced biogas production compared to untreated FW. However, extreme size reduction was reported to trigger VFA accumulation with decreased solubility as well as methane production. Also, the robust structure of lignocellulosic fraction of OFMSW makes its size reduction energy intensive. Accompanying chemical treatment processes prior to wood size reduction may provide a better alternative (Zhu and Pan, 2010).

High-pressure homogenizer expands the pressure up to a few hundred bars, prior to the homogenization of the substrate under depressurization step (Mata-Alvarez et al., 2000). The formed cavitation prompts internal energy, which disturbs the cell layers (Barjenbruch and Kopplow, 2003). These treatment techniques have been employed on different substrates including lignocellulosic biomass, excrement and wastewater treatment plant discharge. Some of the benefits of mechanical treatment include, no unpleasant odor problem, simple execution, better dewaterability of the anaerobic deposit and direct energy utilization (Perez-Elvira et al., 2006; Toreci et al., 2009). Mechanical treatment process like electroporation and liquefaction also aid in disrupting the cell wall (Shepherd, 2006) of OFMSW residues and have been demonstrated at lab scale. While, other mechanical treatments such as revolving drum, plate screen shredder and cylinder squeeze were effectively demonstrated at pilot scale (Carlsson and Anox Kaldnes, 2008). Some mechanical treatment practices involve crushing of the waste with high weight into an expulsion chamber to press-out the wet portion from the expulsion openings (Gonella, 2011). This technique has been in practice and is incorporated in an AD plant in Salerno (Italy) for treating OFMSW (Nayono et al., 2010).

Mechanical treatment viz. the rotational drum (**Figure 4**) was utilized as a compelling innovation for OFMSW, which improves the biogas generation in the range of 18–36% (Zhu et al., 2009; Subramani and Ponkumar, 2012). Both improvements; i.e., methane yield per gram of volatile solids and amount of methane in biogas have been reported (Davidson et al., 2007). However, Zhang and Banks (2013) reported no significant improvement in biogas yield employing rotational drum treatment. Hansen et al. (2007) reported the impacts of the similar treatment approaches on the nature and source of OFMSW. Interestingly, Bernstad et al. (2013) reported that the screw press treatment results in significant loss of biodegradable substrate after processing. Fantozzi and Buratti (2011) highlighted the significance of the inoculation mechanically treated waste substrates that provide large accessible surface area for microbes.

# Thermal Treatment

Thermal treatment is often referred as molding procedure for raw or processed waste as it enhances the dewater ability properties of the waste (Bougrier et al., 2007). Thermal treatment modifies the structure of the insoluble part of the substrate and makes them amenable for biodegradation. Kuo and Cheng (2007) contemplated thermal treatment of kitchen waste at different temperatures (37, 50, and 60◦C), to assess its impacts on hydrolysis. The treatment at 60◦C showed best results, accomplishing a hydrolytic proficiency of 27.3% and fat removal of 37.7%, as compared to untreated sample. The study demonstrated stable operation and positive execution with respect to treatment of kitchen waste. Comparable results were reported by Komemoto et al. (2009).

Thermal treatment aids into pathogen expulsion, enhances dewatering and reduces thickness of the digestate (Liu et al., 2012). A wide range of temperatures (50–250◦C) to upgrade the AD of diverse organic fraction of solid wastes have been worked-out by various groups of reseachers. The fundamental impact of thermal treatment is the disruption of cell wall and aiding into solubilization of organic material (Marin et al., 2010; Protot et al., 2011). COD solubilization and temperature have direct impact on biogas yield. Higher solubilization may likewise be accomplished at lower temperatures; however, the longer treatment time is required. Mottet et al. (2009) studied diverse thermal treatment techniques and reported no considerable distinction amongst steam and electric heating, while microwave heating has been observed to solubilize more biopolymers. The higher rate of solubilization with microwave treatment was due to the phenomenon of polarization of macromolecules present in the feedstock (Toreci et al., 2009; Marin et al., 2010). Thermal treatment atsignificantly high temperatures (>170◦C) may result into formation of bonds and agglomeration of the particles (Bougrier et al., 2006). Maillard reaction is reported to produce complex recalcitrant substrates from starch and amino acids, when subjected to high thermal treatment (150◦C) or longer treatment time at lesser temperature (<100◦C) (Carrere et al., 2010; Elliot and Mahmood, 2012).

Protot et al. (2011) recommended that thermal treatment at temperatures lowerto 100◦C did not bring about the debasement of complex particles, rather it essentially incites the separation of macromolecules. Barjenbruch and Kopplow (2003) acquired a comparable conclusion with treatment at 90◦C. The outcomes from these experiments demonstrated that the fibers were not broken down, rather, they were just mugged with thermal treatment.

Release of particulate sugars and solubilization of proteins was reported by Neyens and Baeyens (2003) by employing thermal treatment.

The impacts of thermal treatment on the chemical and physical properties of kitchen waste, vegetable waste and waste activated sludge were explored by Liu et al. (2012). Outcomes revealed that thermal treatment (175◦C, 60 min) diminished the viscosity and increased COD, dissolvable sugar and proteins. A reduction of 7.9 and 11.7% of methane was observed for kitchen and vegetable waste respectively. The authors ascribed this phenomenon to the arrangement of an intractable copolymer and melanoidin (Maillard reaction products). Under related working conditions (170◦C, 1 h), Qiao et al. (2011) observed that both biogas and methane generation from anaerobically treated FW and cow manure were diminished by 3.4 and 7.5% respectively as a result of lower pH and higher VFAs. Ma et al. (2011) reported a 24% enhancement in CH<sup>4</sup> production when FW was treated at 120◦C.

#### Chemical Treatment

A range of chemicals viz. acids, base, and oxidants (e.g., ozone, peroxide) have been successfully employed to break down natural constituents (Carrere et al., 2010). Chemical treatment is utilized to breakdown the linkages in plant cell wall by employing strong acids, alkali or oxidants. Alkali treatment is considered as the favored chemical treatment when compared to other treatment methods (Li et al., 2012). During alkali treatment, the primary responses that happen to the substrate are solvation and saponification, which instigates the swelling of solids (Carlsson et al., 2012). Subsequently, the surface region is expanded, and the substrates are effectively available to anaerobes (López Torres and Espinosa Lloréns Mdel, 2008; Modenbach and Nokes, 2012).

Ozonation is a technique by which biogas upgradation may be achieved along with enhanced hydrolysis rate. However, chemical treatment is believed to be less appropriate for easily biodegradable or less recalcitrant substrates. Feedstocks encompassing high amounts of starch, showed accelerated biodegradation resulting in access VFA, which may inhibit methanogenesis step (Wang L. et al., 2011). However, it may have beneficial outcomes on lignin rich substrates that has complicated network of lignin carbohydrate complex linkages (Fernandes et al., 2009). López Torres and Espinosa Lloréns Mdel (2008) reported an increased OFMSW AD efficacy after a primary treatment with Ca(OH)2. The addition of 62.0 mEq Ca(OH)2/L along with stirring for 6 h resulted in 11.5% increment in the soluble COD (sCOD) out of the aggregate COD. Moreover, the addition of the Ca(OH)<sup>2</sup> showed reduction in biomass solubilization, which could be ascribed to the arrangement of complex, non-dissolvable compounds. Neves et al. (2006) reported that both; a basic hydrolysis treatment and co-assimilation with kitchen waste were positive to upgrade the methane generation from grain (barley) waste, yet the best results have been observed with alkaline hydrolysis in terms of decrease in total solid (TS) and volatile solid (VS). Maintaining pH during AD process may be an expensive procedure while treating large amounts of waste. This is considered as one of the possible obstacles toward employment of alkaline treatment at large-scale operations. In addition, using alkaline treated samples, in continuous stirred type reactors (CSTRs) has resulted in decreasing acetate and glucose consumption by 5 and 50% respectively because of generation of toxic compounds during the saponification in spite of the fact that sodium or potassium were 0.21 mol/L in the medium (Ward et al., 2008).

In recent years, the utilization of ozone as treatment reagent is gaining a large attention due to its very powerful oxidant property, with a redox potential of 2.07. In addition, it has several advantages, which include: (i) on-site generation, (ii) leaving no traces in the substrate and (iii) do not produce toxic halogenated compounds. However, the adequately high ozone dose promotes, mineralization of the released cell compounds (Elliott and Mahmood, 2007). The effectiveness of ozone relies on the type of biodegradable waste, ozone concentration and pH. Most studies allude to the utilization of ozone for sludge treatment and then subjecting to AD process (Delgenès et al., 2003; Chu et al., 2008).

Solid waste ozonation was performed by Cesaro and Belgiorno (2013) and reported that an ozone dosage of 0.16 g O3/gTS could increase sCOD by 55%, and bringing about 37% increment in biogas production levels. However, higher ozone concentration showed no increase in methane production irrespective of substrate solubilisation (Bougrier et al., 2006; Baratharaj, 2013). Bioconversion process requires stable pH for optimal performance during AD (Pavlostathis and Gosset, 1985). However, chemical treatment employing ozonation have resulted into decrease in pH, hence use of alkali to neutralize the pH is considered as an essential step before the start of AD process.

Acid treatment is attractive for lignocellulosic substrates since it solubilizes hemicelluloses, condenses lignin that precipitates and aids in hydrolysis which functions optimally in acidic conditions (Mata-Alvarez, 2003; Mussoline et al., 2012). One major drawback of acid treatment is that employment of strong acids may result in the formation of inhibitory products such as furfural and hydroxymethylfurfural (HMF) (López Torres and Espinosa Lloréns Mdel, 2008; Mussoline et al., 2012). Other drawback associated with acidic treatment includes loss of fermentable sugars because of the continued degradation of complex substrates, the high cost of acids. Additional cost associated with base that is needed to neutralize treated substrate before the AD process has been reported to be a matter of concern (López Torres and Espinosa Lloréns Mdel, 2008; Kumar and Murthy, 2011). In few cases, adding acids can be effective for the anaerobic absorption of protein-rich substrates by generation of ammonia. However, this may hinder the biogas production (Hansen et al., 1998) due to the fact that microbes get inhibited in the presence of ammonia. In spite of the interesting results from anaerobic trials of chemically treated waste, plenty of room is available for improvement in pH restoration during pilot scale applications for economic viability as well as sustainability of the process.

#### Biological and Biochemical Treatment

Promoting microbial growth on biomass could notably enhance hydrolysis of the substrate. This could be accomplished by the application of commercially available enzymes as well as growing microbes on it (Yin et al., 2016). Fdéz-Güelfo et al. (2011a) reported that the addition of compost (made from manure) to industrial OFMSW helps in reducing the dissolved organic compound (DOC) by 61% and VS by 35% compared to control. As a result, the biogas and methane generation increased up to 60 and 73%, respectively. While comparing the use of manure compost, biological treatment (using fungus Aspergillus awamori) showed better hydrolysis results (Fdéz-Güelfo et al., 2011b). Several other studies concluded the superiority of employing biological treatment compared to thermochemicaltreated substrates for increased AD rate (Fdéz-Güelfo et al., 2011b; Fdez-Güelfo et al., 2012).

Charles et al. (2009) studied the impacts of pre-air circulation to OFMSW and its effect on AD and concluded that aerobic treatment constrained the amount of methane generated as opposed to fortifying the process. The term "micro aeration" was defined as the slightly introducing oxygen into an anaerobic process to empower aerobic organisms to act on organic substrates inside bioreactor. Researchers revealed that increased solubilization and fermentation was accomplished with micro aeration at 37.5 mLO2/d. About 21 and 10% higher methane generation was observed during anaerobic co-processing of brown water and FW respectively in the reactor when infused with sludge. Lim and Wang (2013) reported that oxygen introduction could stimulate VFA production due to improved activity of hydrolytic and acidogenic microbes. Pre-composting of OFMSW resulted in the significant loss in methane production as reported by Brummeler ten and Koster (1990) and Mshandete et al. (2005). On the other hand, Miah et al. (2005) studied the biogas generation from sewage sludge treated with aerobic thermophilic microbes, Geobacillus thermodenitrificans. The study reported significant production of biogas (70 ml/gVS) with 80–90% methane content in AD at 65◦C.

Few authors considered the hydrolysis and acidogenisis steps (first and second steps) of a two-stage AD process as a biological treatment option (Carrere et al., 2010; Ge et al., 2010, 2011a,b), while others consider it as a general setup of AD bioreactor and not an alternative of pretreatment (Carlsson et al., 2012). Physically isolating the acidogenic microbes from the methanogenic microbes may bring about increased methane generation. It has been stated that enhancing the hydrolysis stage of an AD bioreactor could excite the acidogens to produce enzymes with higher specificity for prolonged degradation of substrates (Parawira et al., 2005). In a study performed by Verrier et al. (1987), 2–phase biomethanation of vegetable waste was compared with mesophilic and thermophilic single stage CSTRs respectively. It was concluded that a two-stage reactor converted over 90% of easily biodegradable waste to biogas. Also, the mesophilic and thermophilic single stage CSTR's was not able to handle higher organic loading as compared to 2-phased one (Verrier et al., 1987).

Impact of pH value on two-stage AD was studied by Zhang et al. (2005). The research group recommended maintaining the pH at 7 while hydrolysis is being carried out in the reactor. This would enhance the overall organic solid consumption and help to enhance the biogas production. White rot fungi, known to consume lignin, leaving behind cellulose may be effectively applied for biological treatment. Group of researchers have already reported the high de-lignification productivity of various white rot fungi on different lignocellulosic biomass (Keller et al., 2003; Shi et al., 2009; Kumar and Wyman, 2010). Keller et al. (2003) concluded that treating lignocellulosic residues using fungi could bring several benefits such as (i) eco-friendly procedure, (ii) no chemical requirement, (iii) decreased energy input, (iii) operating at ambient conditions, (iv) inexpensive unit operations, (v) less by-product generation, (vi) no washing step and (vii) negligible inhibiting agent production. However, Shi et al. (2009) contemplated that only drawback of fungal treatment is that it's time-consuming process and require considerable amount of space and additional infrastructure to hold the substrate for a period of 20–30 days. Pasteurization of substrate prior to fungal mycelium inoculation will add additional cost to the process.

Researchers have also evaluated the role of extracellular hydrolytic enzymes to increase the yield and the rate of organic matter solubilization of lignocellusolic biomass during AD. Moreover, the possible positive effects of the hydrolytic enzymes by adding to municipal sludge for solubilization were known since 1999 (Delgenès et al., 2003). Nevertheless, selection of appropriate commercial enzyme preparation is a prerequisite to achieve noteworthy outcomes (Mallick et al., 2010). However, the cost of commercial enzymes is a potential barrier to make the process economically viable.

Enzymatic hydrolysis of OFMSW is quite challenging due to its heterogeneous and inconsistent composition. Hence, development and utilization of enzyme cocktails that break down complex substrates i.e. carbohydrates (cellulase, hemicellulase, pectinase), lipids (lipase, lipolytic acyl hydrolase, lipoxygenase) and proteins (protease) are required. The resultant products are simple molecules like sugars, fatty acids and amino acids, which may further be utilized as nutrient source by microbes. Masse et al. (2003) and Valladão et al. (2007) evaluated the impacts of enzymatic hydrolysis on the anaerobic digestion of fat present in poultry waste. Masse et al. (2003) reported that a little impact on fat molecule assimilation at 25◦C occurred by enzyme addition, while Valladão et al. (2007) reported enhancement in crude effluent anaerobic treatment by employing biomass degrading microbes as a part of the pre-hydrolysis step. Lipase was successfully employed to hydrolyze the fats present in dairy wastewater by Cammarota et al. (2001) using a bench scale UASB reactor. Leal et al. (2002), reported that utilizing the blend of enzymes and microbes in bioreactors to treat the fat was viable and hybrid treatment. Mendes et al. (2006) employed a cost efficient and easily accessible lipase preparation derived from an animal source to perform enzymatic hydrolysis of lipid-rich wastewater from a dairy farm. The hydrolysis was performed at varying incubation periods ranging from 4 to 24 h at a fixed temperature (35 ± 1 ◦C) and the treatment proficiency was confirmed by running relative biodegradability tests. All treated measures indicated higher response rate when compared to crude wastewater examined, as evident by enhanced levels of biogas generation and higher removal of organic matter.

Currently, researchers are focusing on development of temperature phased anaerobic digestion (TPAD) (**Figure 5**). This strategy often comprises of an essential reactor at thermophilic (or hyperthermophilic) temperature trailed by using mesophilic auxiliary rectors. Advantages using TPAD include, higher methane yield as well as a pathogen free high quality nutrient rich digestate that may be utilized as soil conditioner (Sung and Harikashan, 2001; Riau et al., 2010). Schmit and Ellis (2001) has reported that conventional AD processes was outperformed by TPAD. Lee et al. (2009) utilized FW and excess sludge using TPAD at 70◦C for 4 days in the main bioreactor, followed by a subordinate bioreactor that operates at different temperatures viz. 35, 55, and 65◦C. Wang F. et al. (2011) compared treatment of FW with polylactide using TPAD processing at two different conditions 80◦C (hyperthermophiles) and 55◦C (thermophiles) followed by digesting the materials using mesophilic reactor.

#### Ultrasound Treatment

Ultrasound treatment (**Figure 6**) depends on stable cavitation along with physical and chemical impacts in fluid dynamics (Dehghani, 2005). The physical impacts are created by the breakdown of cavitation air pockets, which thus delivers a lifted change in the chemical nature through the development of free radicals (Mason and Peters, 2002). These impacts may enhance anaerobic assimilation yields by physical crumbling and improvement of microbial pool (Kwiatowska et al., 2011), contingent upon the treatment conditions. It has been demonstrated that higher ultrasound power may denature the enzymes, while lesser ultrasound span brought better responses of the same enzymes (Yu et al., 2009). Chen L. et al. (2008) examined impacts of ultrasound procedure on hydrolysis and acidogenesis of organic waste and reported 53% degradation in volatile solids as compared to the control. The improvement could be attributed to desorption of VFA's that are present on substrate surfaces. Comparative results were obtained by treating a blend of mechanically sorted OFMSW and sewage waste with ultrasound, before anaerobic absorption (Cesaro et al., 2012). Elbeshbishy et al. (2011a) observed 27% improvement in hydrogen generation in a CSTR reactor using sonicated FW when compared to control. In another study (Elbeshbishy and Nakhla, 2011), enhanced performance could be achieved by incorporating ultrasound inside the hydrogen reactor. In this study, isolated organics were sonicated for 24 min using a lab-scale ultrasonic test followed by single-stage and twostage anaerobic digestion. In another report, sonicated biological hydrogen reactor was setup in the first stage for generating hydrogen, followed by blended tank reactor for methane production as the second stage.

The most astounding overall COD diminishing proficiency of 9.3% was achieved in sonicated biological hydrogen reactor (SBHR), while a total COD reduction of 6.4% was achieved in both reactors with either non-sonicated or sonicated feedstocks. Applying sonication inside the reactor demonstrated better results than sonication of the feedstock outside the reactor at a similar specific energy of 5,000 kJ/kg TS. The noteworthy increment in yield was observed when ultra-sonication was carried out inside the reactor such as (i) solubilisation of the particulate organics, (ii) expulsion of the dissolved gases, (iii) change in mass transfer and (iv) increment in the growth

rate of the microbes. Ultrasound process ended up being the most adaptable, as it was compelling with various sort of fat-dominated substrates, originated from meat processing units when compared to thermal, alkaline, acid and biological treatment processes (Luste et al., 2009). In a study performed by Cesaro and Belgiorno (2013), ultrasound appeared comparatively suitable for OFMSW when compared with ozonation in terms of 9% higher solubilization of organic matter with ultrasound treatment. Elbeshbishy et al. (2011b) reported about 16% higher biogas volume when subjected OFMSW to ultrasound treatment.

Shanthi et al. (2018) examined the effect of ultrasonication on fruit and vegetable waste assisted with sodium dodecyl sulfate. Result showed that optimum dose of sodium dodecyl sulfate was 0.035 g/g of suspended solid. Also, the energy ratio 0.9 for sodium dodecyl sulfate coupled ultrasonic treatment, which showed its energy efficiency.

#### Microwave Treatment

Microwave utilizes the capacity of direct interaction between an object and a connected electromagnetic field to expand heat (**Figure 7**; Hu and Wen, 2008). Due to this both thermal and non-thermal impacts were created in the aqueous solution. The electromagnetic field ruptures the crystalline structures and changes the super atomic structure of lignocellulosic materials, enhancing their reactivity (Tomas-Pejo et al., 2011).

The impact of microwave treatment on the anaerobic biodegradability of kitchen waste was contemplated by Marin et al. (2010). In this study, FW was treated at 175◦C to solubilise the sugars and proteins present. However, enhanced solubilisation did not result into improved anaerobic biodegradability, contingent upon the rate of heating. Pecorini et al. (2016) reported that microwave treatment prompted to a methane generation increment of 8.5% for OFMSW with lignocellulosic materials; while autoclave treatment had an expansion extending from 1.0 to 4.4%. Results exposed an upsurge of the soluble fraction after treatments. A significant

increase was detected in sCOD for treated substrates (up to 219. 8%). Significant improvement in methane generation prompted to the conclusion that subject the OFMSW to autoclave and microwave brought about the hydrolysis of a noteworthy fraction of non-biodegradable natural substances recalcitrance to anaerobic processing.

Shahriari et al. (2012) explored the impacts of microwave treatment, even in a blend with hydrogen peroxide, on OFMSW bound to anaerobic treatment. Experiments were performed by heating samples from room temperature to 115, 145, and 175◦C at a consistent temperature slope time of 40 min. For the consolidated microwave-chemical treatment, tests were blended with 0.38 and 0.66 gH2O2/gTS and heated up to 85◦C. Biochemical methane potential (BMP) was analyzed to appraise a definitive methane generation and also to assess any potential hindrance on anaerobic assimilation. Higher sCOD concentration was detected for substrates treated with microwave up to 175◦C, H2O<sup>2</sup> and joined alkaline/microwave process. These treatment conditions did not bring any subtle change in anaerobic absorption rate or in stabilization degree proposing that the solubilized compounds were less biodegradable (Marin et al., 2010).

#### High Voltage Pulse Discharge Treatment

High voltage pulse discharge (HYPD) has been utilized to treat biodegradable waste over the years (Lee and Chang, 2014). Zou et al. (2016) attempted HYPD treatment on FW. They discovered that the cumulative methane yield increased after the treatment. A total of 134% increase was recorded as compared to untreated one. The VS reduction was also improved after HYPD treatment. The FW treated with HYPD showed 54.3% VS reduction while VS reduction in untreated FW was 32.3%. In another study, grape pomace was treated with HYPD and compared with alkali, ultrasonic and acid treatments for enhanced methanogenesis by El Achkar et al. (2018) in a batch mode. Results showed that alkali treatment yielded highest methane.

# Effect of Additive and Trace Metals on Anaerobic Digestion

With a specific end goal to guarantee rational hydrolysis improvement, it is important to keep a dynamic harmony between every progression. Additives may help to achieve better dynamic stability in methanogenesis (Milán et al., 2001). Some of the literature reported the management and utilization of various additives. Zeolites have been considered as the most preferred additive among the widely recognized ones (Yadvika et al., 2004). Zeolites are small permeable crystalline solids with well-characterized structure; contain silicon, aluminum and oxygen, water and additionally different atoms inside the pores. Most of the present forms of zeolites are naturally occurring minerals and are widely mined all through the world, while others are manufactured and arranged for specific affairs and as per requirement (Montalvo et al., 2006). On account of their novel permeable properties, zeolites are utilized as a part of an assortment of utilizations with a worldwide market of million tons/year. The fervor of zeolites concerning anaerobic assimilation procedures is their ability to adsorb salts especially ammonia, which acts as an inhibitor in the AD process (Tada et al., 2005; Chen Y. et al., 2008).

Kim et al. (2000) contemplated hindrance because of the sodium particle dose on the AD bioreactor running on thermophilic zone of FW and stated that sodium at a concentration of >5 g/L brought about decrease in biogas generation. Sodium is more lethal to propionic acid utilizing microorganisms when contrasted with other VFA debasing microorganisms in the AD system (Soto et al., 1993). The inhibitory effect of potassium begins at the level of 400 mg/L. Interstingly, anaerobic microbes can endure up to 8 g/L of potassium (Bashir and Matin, 2004). It has been reported that the potassium is harmful to thermophilic microbes but may not endanger the reactors running on mesophilic or psychrophilic temperatures (Chen Y. et al., 2008).

The ideal quantity of calcium and magnesium ions has been accounted to be 200 and 720 mg/L, respectively (Kugelman and McCarty, 1965; Schmidt and Ahring, 1993). Higher amounts of calcium may result into the scaling in the reactor because of precipitation of salts, and subsequently decrease the methanogenic process (Zhang et al., 2005). Additionally, a high concentration of the magnesium (100 mM) can interrupt methanogenesis followed by hindrance to the transformation of acetic acid (Schmidt and Ahring, 1993).

Moreover, AD could be improved by supplementation of a range of metals ions viz cobalt, molybdenum, selenium, iron, tungsten, copper and nickel, which assume a part in numerous biochemical responses of the anaerobic process and also play role of co-factors for enzymes involved in AD process. Zhang and Jahng (2012) utilized supplements of Fe, Co, Mo, and Ni to balance out a single stage reactor treating FW and reported that Fe was the best metal for adjustment of the AD procedure. Facchin et al. (2013) supplemented a metal cocktail (Co, Mo, Ni, Se, and W) into AD reactor and achieved 45–65% higher methane yield from FW. Supplementing trace metals to organic waste AD may help to achieve higher biogas generation rates with elevated methane composition in biogas (Paritosh et al., 2017b).

# COMPARISON OF TREATMENT METHODS

Efficiencies, monetary attainability and ecological impacts are criteria for the selection of required treatment strategy prior to AD of OFMSW. There are different types of treatment technologies available based on the origin and variety of biomass. These technologies may be physical, mechanical, chemical, biochemical, thermal as well as combination of the treatment methods for disrupting the cell walls. Every technology has their own pros and cons and it may not be possible to recommend the treatment technology, which is suitable for all types of biomass, as every biomass is unique in terms of its chemical composition and hence a specific treatment method may be required for its disruption and obtain maximum energy. The effectiveness of any treatment strategy can be assessed through the increased methane yield and VS reduction. **Table 5** illustrates the comparative proficiency of treatment methods including physical, biological, chemical and use of the additives for improving the AD of OFMSW.

# ENVIRONMENTAL IMPACTS OF TREATMENT

Besides the energy balance and techno-economic investigation, environmental aspects i.e., pathogen expulsion, utilization of chemicals, the likelihood for a manageable utilization of the deposits and impacts on human health has to be considered while opting a treatment process (Stabnikova et al., 2008; Thorin et al., 2012; Di Matteo et al., 2017). The solids that are generated after AD process has the potential to be used as soil amendments, which is good for the environment. Life cycle assessments (LCA) will help to assess the efficiency and environmental impact of AD process. Only few researchers have assessed the environmental impacts of using treatment technologies before AD of solid wastes. Carballa et al. (2011) have performed the LCA to evaluate environmental attributes associated with the use of seven treatment technologies (alkaline, acid, thermal, thermal acid, freeze-thaw, pressurizedepressurize, and ozonation) for kitchen waste and sewage sludge. Impacts were analyzed with respect to the potential of abiotic resource depletion, eutrophication, global warming, human toxicity and terrestrial ecotoxicity. The researchers suggested that pressurize-depressurize and chemical treatment techniques outstripped ozonation, freeze–thaw and thermal strategies by having a minimum adverse environmental impact. Nwaneshiudu et al. (2016) assessed the environmental impact of mild bi-sulphite treatment of forest residues. The study reported less eutrophication impact of forest residues as compared to the effects from traditional beet and cane sugars, while the global warming impact falls within the range of conventional processes. However, the scarcity of literature regarding the evaluation of environmental impacts of treatment methodologies is inevitable. Adding environmental impact assessment in addition to technical and economic evaluation of treatment will help biogas production environmentally sustainable.

#### TABLE 5 | Comparison of various treatment techniques.


# TECHNO-ECONOMIC TRAITS OF TREATMENT

Development of different treatment methodologies has dramatically increased in the recent years. Mechanical innovations such as robotics have also been employed to minimize human contact with the waste. Combining with treatment, a segregation zone for workers to remove the inorganic materials manually is been practiced. Once the OFMSW has been stacked into the mechanical parcel system, human contact will be insignificant. OFMSW portion is easily breakable in diminutive particles compared to the inorganic materials. Balsari and Menardo (2011) reported that depending on lignocellulosic biomass composition, the energy consumption may differ between 5 and 80 kWh/ton. However, the volume of methane production from these treated wastes fluctuates between 14 and 26%.

The volume of methane production alone will not decide the profitability of the operation. One has to take into account the net energy gained (i.e., energy generated from methane, subtracted by amount of energy spent) and other costs associated with the processing of OFMSW. Ma et al. (2011) reported the effects of different types of treatments (viz. acid, thermal, freez-thaw etc.) on kitchen leftover and methane generation. The results recommended that nevertheless the sort of substrate, reduced energy utilization provide more financial advantage. In thermal treatment, energy demand relies on the required treatment temperature. It has been reported that microwave heating has advantages over conventional heating due to its ability of the direct core heating (Bordeleau and Droste, 2011). Yang et al. (2010) observed higher carbon and VS removal after thermal

treatment of sewage sludge in two stage AD. Authors have also concluded that net energy profit decreased as temperature of thermal treetmement increased.

Nasr et al. (2012) evaluated the balance of energy in twophase AD of thin stillage and reported that optimization of the two-step AD could result into an increase (∼18%) in energy balance. Lu et al. (2008) reported 2.17 kJ/day of energy surplus in a two-step reactor when compared with a single stage framework for treating sewage sludge. In UK (Avonmouth), Wessex water introduced an ultrasound framework for treating the domestic

and industrial sludge with a population equivalent of 1,200,000. TS and VS decrease of the untreated waste sludge in the digesters was 40 and 50% as compared to that of sonicated sludge, which was 60 and 70%, respectively. The above-mentioned framework was introduced in numerous plants in the UK, USA, and Australia (Hogan et al., 2004). On the contrary, Mottet et al. (2009) reported that neither microwave nor ultrasound treatment was energy friendly for treating blended waste, as the improved methane yields were insufficient to remunerate the required energy and maintain the energy balance of the system.

#### BIOGAS AND CHALLENGES

The intricacies of AD process and the process instability that involved in modern technologies are two major constraints affecting AD improvement. The prime goal of research and developments to produce bioenergy around the globe is to develop and establish mature AD technology as source of renewable energy in various sectors (viz. CHP, transportation etc.). **Figure 8** describes possible strategies and the importance of relation between industries and academic institutions for developing technology on biomethanation. For cost reduction, it is essential to identify the crucial steps and type (single-phase, two-phase, hybrid) of technology in AD process by taking into account; targeted substrate as well as cost of enzymes employed in the process (IEA Bioenergy Update, 37, 2008).

Rate of biomass conversion and its digestion process mainly affected by microbial community and biocatalyst present in the system. Microorganism and biocatalyst having low adoptability to the AD environment are supposed to be stressed and reduce the total output in a biogas plant. This results in several changes to develop integrated process to convert waste to energy that is economically viable. During enzymatic hydrolysis, different enzyme are required as the OFMSW has a complex composition. Utility requirements viz. heat, electricity, power is also of great concern; and its optimal use in the process is an engineering issue by nature. By-products such as digestate if used or sold as soil amendment to generate revenue may meet the partial utility costs.

Keys to cost effective biomethanation process are; combination of appropriate treatment methodology, presence of robust and dynamic microbial community, quality of substrate and reactor operating conditions. Incorporation of engineering and biology in unison may play a futuristic encouraging role to bring higher yield of methane at competitive price. Other opportunities to overcome the challenges include, developing biocatalysts at low production cost, integration of solar photovoltaics with the biogas plants for energy generation, cost effective treatment and enriched methane for use as bio-CNG. Developing fundamental knowledge on the above mentioned topics may fetch a better process stability for biogas production and utilization.

The very concept of biogas economy driven by biogas inherently depends on feedstock availability, logistics and government policy. Converting industrial waste to biogas and utilizing them for industrial energy will benefit both industry and environment. It is widely accepted that AD technology has robust commercial applicability around the globe. Present energy utilization behavior encompasses dispatched energy sources such as thermal power plants, solar photovoltaics, etc. In dispatched energy source, the required energy can be sent as per requirement at any stipulated time. The present energy supply is in transition period. A major driver for development of bioenergy market is greenhouse gas mitigation. Prior to this, main argument was logistics, storage and biomass supply security. Biomass storage and its improved logistics may address the energy demand renewably. India, on the other hand, initiated various programs related to biomass power. Ministry of new and renewable energy (MNRE), Government of India, administrating a MNRE–UNDP /GFF assisted project on "Removal of Barriers to Biomass Power Generation in India." Total installed biomass power in different states of India is shown in **Figure 9**. The primary goal of the project is to increase the use of biomass power and cogenerations based technologies in the country and enhance electricity supply through renewable energy sources. Central finance assistance (CFA) in the form of capital subsidy and financial incentives are also provided by MNRE.

#### CONCLUSION AND FUTURE ASPECTS

The demand for renewable energy and rise in global warming due to greenhouse gas emission has motivated many government agencies to identify new methods to sustainably produce biogas using AD system. Most treatment techniques have been assessed at batch scale to evaluate the biogas potential from treated OFMSW. However, very limited reports are available that demonstrate biomethanation process at pilot scale using newly developed methods and technologies. Several scientific modifications have to be encountered when the lab scale experiments are translated to large/industrial scale operations. Also, anaerobic biomethanation plants should be operated 24 × 7 in contrast to the one under the controlled lab-scale conditions. Generating energy from OFMSW using anaerobic assimilation research is mainly in two areas, (i) understanding the fundamental of treatment systems and (ii)

#### REFERENCES


assessment of the technical and monetary attainability of the joined treatment/anaerobic processing framework. Promising treatment method that is developed in lab scale should be further scaled-up in order to evaluate the overall cost of processing and sustainability. Different treatment procedures viz. physical, thermal, combined, organic or chemical have extensively been attempted at lab scale under defined conditions. Other processes that have been demonstrated at large scale include mechanical, thermal and thermochemical techniques. Anaerobic digestion offers benefits compared to the other disposal procedures for OFMSW considering a fundamental sensibility assessment. Treatment technologies still require advancements at different faces viz. higher biogas yield, efficient management of pathogen, reduction of digestate and reducing the hydraulic retention time as well as the techno-economic viability. The OFMSW offers huge biogas potential and promising opportunity for renewable energy generation, nutrient recovery as well as future research avenues in the area of sustainable waste management and treatment.

## AUTHOR CONTRIBUTIONS

KP, VB, and VV conceived the idea for writing the review and prepared basic content of the manuscript. MY tabulated the data from various references. SM contributed for Environmental Impact Assessment area of treatment. VB contributed for OFMSW information. WL helped in anaerobic digestion aspects. NP drew the figures. VV contributed in treatment ideas. KP and VV clubbed all the information together. All authors worked on text and language correction and reviewed the manuscript.

#### FUNDING

VV would like to gratefully acknowledge and thank Department of Biotechnology, Government of India (No. BT/RLF/Reentry/04/2013) for financially supporting this work. VB thanks University of Houston for supporting his research with start-up funds. WL thanks Michigan State AgBioReserach for supporting his research.

#### ACKNOWLEDGMENTS

KP and MY would like to thank MNIT Jaipur for financial support. Authors would like to thank reviewers for their comments and suggestions that has improved the quality of manuscript.


anaerobic solubilisation of food waste. Waste Manage. 29, 2950–2955. doi: 10.1016/j.wasman.2009.07.011


residues into fermentable sugars for biofuel production. Biotech. Biofuels 9:15. doi: 10.1186/s13068-016-0433-1


of cotton stalks for ethanol production. Biomass Bioenergy 33, 88–96. doi: 10.1016/j.biombioe.2008.04.016


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Paritosh, Yadav, Mathur, Balan, Liao, Pareek and Vivekanand. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Techno-Economic Analysis of Forest Residue Conversion to Sugar Using Three-Stage Milling as Pretreatment

Kristin L. Brandt <sup>1</sup> \*, Johnway Gao<sup>2</sup> , Jinwu Wang<sup>3</sup> , Robert J. Wooley <sup>4</sup> and Michael Wolcott <sup>1</sup>

*<sup>1</sup> Composite Materials and Engineering Center, Washington State University, Pullman, WA, United States, <sup>2</sup> Global Cellulose Fibers, International Paper, Federal Way, WA, United States, <sup>3</sup> Forest Products Laboratory, USDA Forest Service, Madison, WI, United States, <sup>4</sup> Biomass ad Infinitum LLC, Cedar Key, FL, United States*

#### Edited by:

*J. Richard Hess, Idaho National Laboratory (DOE), United States*

#### Reviewed by:

*Wenjian Guan, Harvard University, United States Jaya Shankar Tumuluru, Idaho National Laboratory (DOE), United States*

> \*Correspondence: *Kristin L. Brandt kristin.brandt@wsu.edu*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *09 April 2018* Accepted: *17 July 2018* Published: *14 August 2018*

#### Citation:

*Brandt KL, Gao J, Wang J, Wooley RJ and Wolcott M (2018) Techno-Economic Analysis of Forest Residue Conversion to Sugar Using Three-Stage Milling as Pretreatment. Front. Energy Res. 6:77. doi: 10.3389/fenrg.2018.00077* This study quantifies the cost of cellulosic sugar production using a fully-mechanical pretreatment process and fuel pellets as a co-product. The pretreatment reduces softwood forest harvest residuals to micron-sized amorphous particles. Energy consumption is minimized using a three-stage milling process. A techno-economic analysis was completed for a milling facility with saccharification and wood pellet manufacture. For the base case, concentrated sugar syrup can be produced for \$0.496/kg of sugar. Sensitivity analyses were used to determine cost controlling variables, optimize the sugar cost and found that siting for this technology needs to strongly consider electricity cost and to a lesser extent local feedstock availability. If the sugar produced in this process is used to generate biofuel and is qualified for RIN credits through a life-cycle analysis, the effective cost could be reduced by \$0.04–\$0.06/kg of sugar. An additional \$0.067/kg savings is possible if the biofuel facility is located adjacent or on-site; the finished sugar syrup would not have to be concentrated for transportation. An optimized scenario, including the RIN credit, dilute sugar syrup, and favorable energy costs and consumption, could reduce the cost to \$0.34/kg sugar compared to \$0.496/kg for the base case.

Keywords: cellulosic sugar, clean sugar, mechanical pretreatment, amorphization milling, technoeconomic analysis, three-stage milling, fuel pellets

#### INTRODUCTION

Conversion of non-food lignocellulosic material into clean sugars or transportation fuels has been widely studied, with researchers citing abundant and low-cost feedstocks. However, they present caution regarding lower yields and more difficult processes, especially with high lignin feedstocks (Piccolo and Bezzo, 2009; He and Zhang, 2011; Dutta et al., 2015). Cost effective processing of lignocellulosic biomass into sugars or biofuels is most likely to be achieved through the addition of co-products. The choice of co-products, along with their price, will help determine if a process will succeed financially by increasing income and lowering risk through product diversification (Crawford, 2013; Davis et al., 2013; Biddy et al., 2016).

Lignocellulosic biomass is comprised primarily of cellulose, hemicellulose, and lignin. Cellulose, a highly crystalline substance, is the major component. Pretreatment before a bioconversion process is required to disrupt the intricate hierarchical structures, thereby allowing enzymes access to the cellulose and hemicellulose for saccharification (Cadoche and López, 1989; McMillan, 1994; Sun and Cheng, 2002; Zhu et al., 2008). Disrupting this crystalline microstructure is difficult and requires complex, high-cost pretreatment options (Mosier et al., 2005; Crawford, 2013; Zhang et al., 2013; MYPP BETO, 2016). Although thermochemical pretreatment methods are effective for pretreating biomass (Zhu et al., 2009; Humbird et al., 2011; Gao et al., 2013), significant amounts of chemicals are needed. These chemicals include sulfuric acid and bisulfites for pretreatment as well as ammonium hydroxide, calcium hydroxide or other bases for overliming and/or neutralization processes. Sulfites, in particular, are capable of breaking down carbohydrates and separating lignin for hemicellulosic feedstocks (Zhu et al., 2015). Chemical pretreatments contaminate the biomass and chemical removal adds significant processing costs. Contaminates from some pretreatment methods, including sulfur, can cause catalyst poisoning in a sugar catalytic process (Elliott et al., 2004; Biddy and Jones, 2013; Guan and Blersch, 2018). In addition, thermochemical pretreatments require high temperatures that can create inhibitors, such as hydroxymethylfurfural (HMF) and furfural (Weil et al., 2002; Jönsson et al., 2013). These compounds are toxic to the fermentation organisms and their removal adds cost to the process.

The cost of equipment for thermochemical processes is scaled exponentially, which equates with larger facilities being more cost effective (Humbird et al., 2011; Davis et al., 2013; de Jong et al., 2015). However, the nature of milling equipment can result in a linear scale-up for equipment costs; grinding equipment has a physical size limit that once reached requires additional pieces of equipment have to be added to meet greater throughput values instead of simply purchasing larger equipment (Pirraglia et al., 2010a). While mechanical pretreatment may not offer the economy of scale, the modular nature of mechanical milling provides ample opportunity for distributed processing strategies with little influence on the conversion costs. The micronizing process increases the bulk density of the biomass to 0.46 g/cm<sup>3</sup> from 0.19 g/cm<sup>3</sup> (Wang et al., 2018). It is well documented that increased feedstock density can lower transport costs which helps to offset the traditionally increased operating and capital costs of a distributed processing model, especially for longer transport distances (Wright et al., 2008; Kim et al., 2011; You and Wang, 2011; Kim and Dale, 2015, 2016). If smaller, milling depots are located at existing wood processing facilities, the existing supply chain can be used to efficiently transport the feedstock. In addition, the core competencies of other wood processing facilities align with wood milling. The micronized wood could then be transported to a central saccharification facility.

In general, all biomass pretreatment starts with mechanical particle size reduction, followed by a thermochemical or a biochemical pretreatment process. A fully mechanical pretreatment has the advantages of eliminating chemical contamination and producing no inhibitors in the pretreated biomass and its final sugars and residue products, in addition to saving water in the pretreated biomass washing process. However, the required energy has been considered costprohibitive as a stand-alone alternative and has therefore been largely discounted (Cadoche and López, 1989; McMillan, 1994; Zhu and Pan, 2010; Zhang et al., 2016). For fine wood milling, McMillan (1994) used a ball mill to reduce wood to 50µm. However, the electricity usage was high at 2.8 kWh/kg. Coarse milling completed by Cadoche and López (1989), who reduced hardwood chips from 23 mm to 1,600µm using a hammermill or a knife mill, requires less energy, only 0.13 kWh/ kg. Esteban and Carrasco (2006) compared one and two stage milling of pine chips with hammermills. They measured the lowest energy use, 0.15 kWh/OD kg for 95% passing a 1,000 µm screen and at least 12% passing a 125µm screen, when they applied a two-stage milling process with a sorting step after the first milling stage. Reducing woody biomass to a finer size, a median size (D50) of 105µm by an attrition mill requires much more energy at 1.9 kWh/ kg (Repellin et al., 2010).

The current paper examines the economics of a mechanical pretreatment process that uses three stages of milling to produce clean feedstock for saccharification. Included in this work is a description of the process and economics of converting biomass, specifically softwood forest harvest residuals (FHR) into sugars and fuel pellets. In addition to providing clean material for conversion, the removal of chemicals and water from the pretreatment lowers the environmental impact and the waste water treatment costs (Jones et al., 2013).

The cost of sugar can be used as a first step to determine the cost of high-value products including, biojet fuel, and biochemicals. Humbird et al. (2011) included a section in the 2011 NREL report that stopped at sugar instead of continuing to ethanol at the request of non-ethanol biofuels industry stakeholders. Zhang et al. (2013) describes the need to use cost-intense biochemical pretreatments to convert lignocellulosic biomass into sugars and Wyman (2003) stressed the need to choose the scale and products that are made from cellulosic sugars to ensure financial viability. Davis et al. (2013) states sugar costs must be lowered to manufacture hydrocarbons from lignocellulosic material in an economically viable way and suggests milder pretreatments.

This objective of this paper is intended to provide economic analysis for converting softwood biomass into sugars using a fully mechanical pretreatment, information that is not well defined in the current literature. The analysis provides the financially controlling variables and possible cost-reduction strategies to make the sugar produced economically viable for biofuels or biochemicals. Using the three-stage milling process reduces the power consumption to covert softwood biomass to both clean sugar and lignin fuel pellets compared to other milling methodologies (Wang et al., 2018).

#### METHODS

For this paper, a techno-economic analysis of processing softwood FHR, referred to using the general term "biomass" into sugar and fuel pellets was completed in a single facility. The experimental results for micronizing the biomass, saccharification and manufacturing pellets used in this study are from literature sources. In the future, additional analyses should be completed to determine the impact of separating the mechanical pretreatment and saccharification into two facilities being managed by a single company, separate companies, or separating the pretreatment facility into multiple smaller wood micronizing depots. The base case used in this paper describes the process using values that are representative of Washington state and is used for comparison as variables are manipulated.

#### Mechanical Pretreatment-Micronizing

The micronizing process reduces softwood biomass from material that passes through a 45-mm round screen to amorphous particles (Wang et al., 2018) (**Figure 1**). The biomass arrives at the facility and is stored in open air piles until entering the process where the material is first screened to remove bark and inorganic material. This rejected material is sold as hog fuel (Marrs et al., 2016). The assumed initial biomass wet-basis moisture content of 35% is reduced using natural gas fired drum dryers to a nominal 10% before entering the milling process. To minimize energy use, the milling is divided into three stages: coarse milling, fine milling, and amorphous milling, which concludes with amorphorous cellulose. The use of multiple milling stages optimizes the process by matching the particle initial size and size reduction for each mill type. Separating the energy intense fine and amorphous milling stages from the coarse milling is an effective method to reduce total energy consumption (Gu et al., 2018; Wang et al., 2018). The first stage of milling, coarse milling, reduces the biomass to a median particle D<sup>50</sup> of 270µm using a hammermill and 0.19 kWh/OD kg. This material is fed into the fine milling stage where an air classifying mill (ACM) takes the 270µm material down to a median size of 63µm using an additional 0.70 kWh/OD kg. The final stage of milling achieves an amorphous state in the wood particles using a media mill to attain a final median particle size of approximately 25µm. The media mill consumes 0.57 kWh/OD kg in this milling stage (Wang et al., 2018). The final product is called micronized wood and is transferred to the saccharification department to be processed into a concentrated sugar syrup; the lignin residuals are manufactured into fuel pellets.

# Saccharification

The micronized wood is sent to saccharification where the amorphous wood is hydrolyzed by cellulolytic enzymes (cellulase and hemicellulose-xylanase). In laboratory testing, a high solids content of 15–21% micronized wood is easily suspended by standard mixing. The micronized wood was hydrolyzed enzymatically at 21% solids to achieve the base case sugar yield of 0.328 kg sugar/kg micronized wood (Gao and Neogi, 2015; Wang et al., 2018). Explicit saccharification process details are presented by Gao and Neogi (2015). This yield accounts for process losses. The sugar yield includes the C6 sugars, glucose, mannose and galactose, which can all be used by standard yeasts for biofuel fermentation.

After enzymatic hydrolysis, the hydrolyzed slurry is pumped to a belt filter press operation, where the clear hydrolysate is filter-pressed and separated from the hydrolyzed residuals. The residuals are further washed with water to recover additional sugar. The clear hydrolysate or sugar stream is sent directly to an evaporator to concentrate it into a high titer syrup (49% solids). If non-concentrated sugar stream is produced for an onsite biofuel facility, the dilute stream can be directly transferred to fermentation. The filter-pressed residuals are sent to the fuel pellet department.

#### Fuel Pellets

Fuel pellets were chosen as the co-product to be manufactured from the lignin waste stream. Consumption of fuel pellets in the

Brandt et al. Technoeconomic Analysis for Amorphized Wood to Sugar

US grew 200% from 2002 to 2006 and world consumption is predicted to grow from 16.4 million tons in 2010 to 123 million tons in 2020 (Pirraglia et al., 2010b). The United States produced 11.5 million tons of fuel pellets in 2017 (EIA, 2018c). The costs of biomass, labor, and drying dominate the economics of energy pellet manufacture (Mani et al., 2006; Pirraglia et al., 2010a; Qian and McDow, 2013). Using a waste stream for feedstock will decrease the risks of fuel pellet manufacture though consistent, economic feedstock (Qian and McDow, 2013).

The residuals from the saccharification process are composed primarily of high lignin content residuals with a small amount of undigested cellulose and hemicellulose. This residual stream also includes small amounts of hydrolysates, monomeric sugars, and enzymes (Gao and Neogi, 2015; Fish et al., 2016). The filter-pressed residuals from the saccharification are assumed in the analysis to be 38% solids, or 62% water. After entering the pelletizing process, the residuals are dried before being extruded through the pellet mills into pellets in which the lignin acts as the binder. The pellets are cooled and screened, before packaging and storage. The fines are recirculated back into the pelleting process.

The conversion of the residuals into pellets requires 0.15 kWh/OD kg of electricity to operate the pellet mills, cooler. and shaker (Pirraglia et al., 2010a). However, the majority of the energy consumed is to dry the residuals using natural gas. The energy values utilized were reported by DiGiacoma and Taglieri (2009) and converted from electricity to natural gas.

#### Financial

Following the financial analysis method outlined by Petter and Tyner (2014), both engineering and economic analyses were completed. The engineering analysis consists of a discounted cash flow rate of return (DCFROR) with the minimum sugar selling price (MSSP) set to attain an internal rate of return (IRR) of 10%. For this analysis the MSSP is the break-even price where the future sugar syrup and fuel pellet sales cover the present value of both capital and operating costs, or the net present value (NPV) is zero. This analysis does not include the impact of inflation but does include the cost of debt. The inclusion of the engineering analysis allows comparison to the economic analysis and with historical literature, which often uses DCFROR without accounting for inflation.

The economic analysis follows the method used by Petter and Tyner (2014); the real discount rate chosen is 10%, which corresponds to a nominal financial discount rate of 12.2%, assuming 2% inflation. The average inflation from 1997 through 2016 is 2.0% according to the Consumer Price Index from the US Bureau of Labor and Statistics; this value was assumed for the economic analysis. The MSSP was determined by setting the NPV calculated in the nominal economic analysis to zero.

The techno-economic analysis (TEA) was completed assuming the parameters defined in **Table 1** and the TEA boundaries begin with the delivered biomass cost and ends with concentrated sugar syrup, fuel pellets and hog fuel, but does not consider the cost to ship these products or the costs to convert the sugar into higher value products. The **Supplementary Material** for this paper includes the details of the financial analysis as well as a complete listing of capital and operating costs. The tax rate TABLE 1 | Assumed economic parameters for base case scenario.


*<sup>a</sup>Modified Accelerated Cost Recovery System (IRS, 2017), <sup>b</sup>percent of FCI spent during each year of construction, <sup>c</sup>milling equipment maintenance is 10% of TPEC (Mani et al., 2006) <sup>d</sup>FCI ratio factor for solid-liquid processing plant with the factor for WWT removed, e (Peters et al., 2003).*

of 16.9% is average US corporate tax rate paid from 2008 to 2012 (Bann et al., 2017). The facility is assumed to be an Nth plant, meaning the technology and manufacturing process has been successfully implemented for multiple plants.

The fixed capital investment (FCI) was estimated using the ratio factor for solid-liquid processing plants applied to total delivered equipment cost (TDEC). This methodology approximates total direct costs (TDC) and FCI from the TDEC of major equipment located inside the battery limits (ISBL). Total purchased equipment costs (TPEC) used are a combination of literature values and vendor quotations and were increased by 10% to account for the cost of delivery to obtain TDEC as appropriate. This methodology is used for preliminary and study estimates with a predicted accuracy of ± 20–30% (Peters et al., 2003). This aligns with this strategic study, which scales up the laboratory data to determine if mechanical pretreatment of biomass is economically viable. Ratio factors have been successfully applied to compare processes and for siting decisions (Wright et al., 2010; Zhang et al., 2013; Astonios et al., 2015; de Jong et al., 2015; Martinkus and Wolcott, 2017). The ratio factor approach is used to estimate the cost of items located outside of the battery limits (OSBL). One item included in the direct costs estimated by a ratio factor is service facilities. For a solid-fluid processing plant, the ratio factor for service facilities is 0.55 × TDEC. The process included in this paper has only a small waste water treatment (WWT) requirement. It was decided to assume the small WWT need could be outsourced and is included as an operating cost. The service facilities ratio factor was reduced by 0.04, which corresponds to removing the typical value for WWT. Total capital investment (TCI) is the sum of installed equipment costs, land, and working capital. Working capital is used to cover operating costs when production and sales do not



*Unit operation costs are TDEC.*

meet the cash demands. Peters et al. (2003) suggests having cash to cover 30-days operations or 10–20% of TCI. The value chosen is 20% of the yearly operating costs; this covers just over 2 months of costs and is 13% of TCI for the base case scenario.

#### Capital Costs

The facility TDEC is dominated by milling equipment. The cost of equipment is a combination of vendor quotes and literature values. The milling portion of this facility does not have the same economy of scale that is common to many chemical processing plants as milling equipment reaches a finite size before needing to add additional units. This property of scaling makes a distributed processing model viable. However, the economic impact of a distributed model this will depend on the availability of feedstock and the transportation logistics for each specific location. This model should be evaluated in future work.

The saccharification process converts the micronized biomass into sugar, using capital and operating costs that are scaled from Humbird et al. (2011) for both the concentrated and non-concentrated sugar syrup scenarios. The concentrated sugar titer for this study and Humbird et al. (2011) is 49%; the cost of concentrating was increased to match the lower nonconcentrated sugar titer in this study of 8.4%, compared to the Humbird et al. (2011) value of 12.7%. For the base case used throughout this document, facility and operating costs were calculated for concentrated sugar syrup, which is appropriate for shipping. However, if the syrup is to be further processed onsite, or at an adjacent facility, the additional costs to remove water can be avoided and the final product is a non-concentrated sugar solution. The differences in capital costs for concentrated and non-concentrated sugar syrup are detailed in **Table 2**.

Solid residues, comprised of mostly lignin, are removed after saccharification, dried and processed into fuel pellets to be sold as a co-product. The fuel pellet equipment costs were taken from Mani et al. (2006), updating costs for the year using CEPCI Index and scaled to the appropriate pellet capacity (Chemical Engineering, 2017). The TPEC is \$2.9 MM; this does not include size reduction equipment needed in traditional pellet manufacturing facilities.

TABLE 3 | Base case variables for micronized and saccharification facilities.


#### Operating Costs

The base case scenario assumes electricity and natural gas costs that are a five-year average of Energy Information Administration (EIA, 2018a,b) industrial rates for the state of Washington; the values used are 2011–2015. The feedstock cost used is the weighted average of the delivered biomass cost to the Longview, WA mill, the assumed mill location. The cost for delivered feedstock assumes a facility scale of 249 k bone dry metric ton (BDMT)/yr. This cost includes the cost of the biomass in the woods, on-site grinding, loading into chip trucks and transport to the micronizing facility (Martinkus et al., 2017). The delivered feedstock cost used in this study, calculated using the Martinkus model, to the micronizing facility is \$56.9/BDMT, which includes \$4.4/BDMT stumpage (US DOE, 2011).

For the base case, the MSSP was calculated assuming a 12.2% nominal financial discount rate, which corresponds to a 10% real discount rate with 2% annual inflation using an updated fuel pellet value from Pirraglia et al. (2010b). The major variables for the base case are listed in **Table 3**; sugar yield is based on actual polymer sugar hydrolysis yield, e.g. glucan conversion yield to glucose is 1.10.

The operating costs are dominated by electricity and feedstock, followed by pellet manufacture (**Figure 2**). The OPEX for pellet manufacture includes electricity and natural gas costs for this department, which are not included in the overall electricity and natural gas costs. For the base case, the milling energy use and yield scenario chosen is 1.46 kWh/OD kg, which corresponds to a sugar yield of 0.33 kg of sugar/kg of biomass (Wang et al., 2018). With the chosen energy requirement and yield combination, the electricity required to micronize the biomass was calculated as 20% of the yearly manufacturing costs. The additional 2% seen in the electricity category (**Figure 2**) is the electricity requirement for saccharification.

The third largest category is the combined OPEX costs for pellet manufacture, at 13%. Pellet manufacture operating costs were adapted from Pirraglia et al. (2010a) with the secondary drying energy values modified to the process from DiGiacoma and Taglieri (2009). Labor and natural gas both require 11% of the yearly OPEX, the lowest individual category is maintenance, at 10%. Labor costs were adapted from Marrs et al. (2016) and Jones et al. (2013) not including the operational labor for pellet manufacture that was scaled from Qian and McDow (2013). If the sugar is left as a non-concentrated syrup, the evaporator can

be removed, which reduces the yearly OPEX by \$3.5MM, or 5%, primarily by reducing the energy requirements (**Table 4**).

# RESULTS

A sensitivity analysis was completed to determine which variables have the greatest impact on MSSP. The values to test were set at assumed realistic, stretch limits for each variable. The outcomes of the analyses were examined alone and in combination to aid in determining the range of possible financial outcomes. All scenarios are compared to the base case.

# MSSP of Concentrated Sugar Syrup From Micronized Wood

Within the ranges evaluated, the two most influential variables for MSSP are the TPEC and electricity cost rate. Peters et al. (2003) reports a TPEC accuracy of ± 30% and this was the range evaluated for impact on MSSP. However, it was assumed that the TPEC estimate is realistic and therefore this variable is held constant at the base case level for all other analyses. Although the industrial electricity rate for the state of Washington state is \$0.0265 less than the US national average for the same timeframe, electricity rates within some regions of Washington state are well below the value used in the base case and are therefore considered in the optimized analysis. Although electricity use is not among the most influential variables, it does have a practical impact and could be a path to lower costs. The milling energy requirements in this paper are from pilot-scale equipment and these requirements may drop as the mass throughput of biomass increase to industrial rates.

The skill sets that are required for the micronized wood and saccharification departments are very different. Size reduction of biomass, especially the heterogeneous forest residuals feedstock, is a complex solids handling and milling process. Experts in wood science and machine processing will be required to consistently attain target moisture content and particle size while running the equipment efficiently. The saccharification



department will require chemical engineers that are experts in converting lignocellulosic materials to sugars. It is possible that a company suited to run a saccharification facility, and perhaps a downstream biofuel or biochemical facility will not have the in-house expertise or the desire to add this expertise, which will be required to effectively run the micronizing facility. This may create a situation where the pretreatment and saccharification are handled in separate facilities, even if located at a single location. It is also possible that adjacent locations could allow the saccharification facility to sell the waste stream to the micronizing facility, which may be best suited to manufacture fuel pellets. Alternatively, this waste could be used to fuel the biomass dryers, decreasing capital and operating costs while eliminating the co-product revenue.

## Sensitivity Analysis-Electricity Use, Electricity Rate

Seven key cost variables were analyzed for impact on final, concentrated MSSP. These variables are TPEC, electricity rate, assumed real discount rate, milling energy, feedstock cost, natural gas rate, and labor cost. The cost of enzymes was determined to be \$3.2MM per year, while not negligible, this cost was also not a controlling parameter of concentrated sugar cost. A tornado plot of these variables shows that the most important factors to focus on are electricity rate, sugar concentration, and electricity use (**Figure 3**). Feedstock cost, which is often a controlling variable, is overshadowed for this process as a result of intense energy use (Mani et al., 2006; DiGiacoma and Taglieri, 2009; Pirraglia et al., 2010a; Wright et al., 2010; He and Zhang, 2011; Humbird et al., 2011; Crawford, 2013; Davis et al., 2013; Jones et al., 2013; Qian and McDow, 2013; Zhang et al., 2013; Astonios et al., 2015; Dutta et al., 2015). The feedstock cost used is specific to the Pacific Northwest and will vary by region. Although the detailed methodology used to determine the feedstock cost in this paper has not been applied on a national level, the feedstock cost was varied to ± 50% to encompass all reasonable delivered feedstocks in the United States; the impact of this is illustrated in **Figure 4**.

In general, the literature uses a 10% IRR calculated using DCFROR analysis. This analysis assumes a 10% real discount rate for the base case. However, this may not be acceptable to investors. Zhang et al. (2013) suggested that 25% IRR may be necessary to attract capital to build an advanced biofuels facility and Crawford (2013) proposed an IRR of 15% to appeal

to investors. The cost of concentrated sugar would increase to \$0.56/kg and \$0.68/kg for 15 and 25% IRR, respectively.

Sugar concentration is discussed in a later section as it is driven by the location of post-processing facilities that are outside the scope of the current assessment. The reduction in electricity use applies to milling energy only and is a percent change to the kWh/OD kg biomass to reach the target micronized wood dimension. The reductions to electricity rate is the major cost driver, most of which is from the decrease in cost to micronizing the wood and could be attained through negotiations with local electricity suppliers.

# DISCUSSION

# Non-concentrated Sugar Sales

The cost to concentrate sugar from the non-concentrated stream exiting the hydrolysis unit operation is split between capital costs and operating costs, which are dominated by the purchase of natural gas. The cost to concentrate the sugar varies with each scenario, which for the base case, is 14% of the concentrated



sugar cost, or \$0.067/kg. A promising future scenario is to locate the biochemical or biofuel facility on-site or adjacent to the saccharification facility so that the sugar can be transferred without this added cost.

# Sugar Yield

The sugar yield, which reflects losses from washing, is controlled by multiple, interacting variables, making the impact of greater sugar yields inconsistent. Three realistic yield scenarios were analyzed: the base case and an assumed optimistic and pessimistic case; the descriptors are for the sugar yield only (**Table 5**). Increasing the sugar yield decreases the fuel pellet yield. However, the milling energy required increases (Wang et al., 2018) and the net impact on sugar cost is small, a decrease of just under \$0.01/kg, compared to the base case. If the cost of electricity changes the financial decision for targeting the base or optimistic case will also change. For the pessimistic scenario, the sugar yield drops, the fuel pellet yield increases, and the milling energy is the same. This leads to an increase in MSSP of \$0.17/kg, as a result of the value reduction caused by lower sugar yield.

# National Energy Impact

Average, industrial Washington state electricity and natural gas prices were used in the base case. This decision reflects the location-specific feedstock costs that were used. The feedstock cost methodology chosen has not been extended to a national level to date. However, electricity and natural gas rates are readily available. The average state industrial rates for 2011– 2015 were obtained from the EIA for the continental US (EIA, 2018a,b). The change in sugar price based on average electricity

TABLE 6 | Variables used to calculate the optimized non-concentrated sugar cost.


and natural gas prices on a state level are shown in **Figure 5**. The US Forest Service FSGeodata Clearinghouse forest coverage information was also included (USFS)<sup>1</sup> The map demonstrates the impact of state utility costs and shows areas where low costs align with forest coverage. It should be noted that rates within a state vary and specific locations will have to be analyzed for cost effectiveness. Additional feedstock work will be required to analyze the biomass cost at each potential location. The information in **Figure 5** can be used to choose the next region to study.

#### RIN-Enabled Sugar

The softwood biomass feedstock utilized in this study may be a Renewable Identification Number (RIN) enabled cellulosic feedstock, meaning biofuel produced from this sugar is eligible for RIN credits. A life cycle analysis will need to be completed to verify that eligibility. Providing RIN-enabled sugar to biofuel producers will cover some of the conversion costs or increase the return (Lane, 2013). The cellulosic waiver credit (CWC) is feedstock dependent and can be combined with D5 RINs if the fuel is advanced (Christensen et al., 2014). The CWC value is difficult to define as a result of low cellulosic biofuel production and the corresponding uncertainty in future values. However, the

average of the CWC values published by the US EPA for 2010– 2015 for jet fuel is \$0.35/liter (Environmental Protection Agency, 2016). The CWC, claimed by the fuel producer, however, it is reasonable to assume that a producer will be willing to pay more for sugar that is RIN-enabled. Mathematical conversion of sugar to jet fuel was completed using the conversion of 3 kg/L of jet fuel (Lane, 2013). If the CWC is divided evenly between the sugar and fuel producers, a \$0.06/kg MSSP increase is possible. This sugar to jet fuel yield may be process dependent; if a conservative conversion value of 4 kg/L is assumed the price differential drops to \$0.04/kg sugar. Conservatively using the lower jet fuel yield, the MSSP would drop to \$0.4559/kg sugar for the base case.

#### Optimized Sugar Cost

An optimized scenario was analyzed to determine the impact of selecting low values for multiple cost variables to manufacture non-concentrated sugar. The natural gas and electricity rates were both dropped to match Cowlitz County, WA values. This

<sup>1</sup>FS Geodata Clearinghouse. Forest Biomass Across the Lower 48 States and Alaska. United States Department of Agriculture Forest Service. Available online at: https://data.fs.usda.gov/geodata/rastergateway/biomass/index.php.

location was chosen to match the cost of feedstock, which was determined for a location inside this county (**Table 6**). The milling energy reduction is assumed to be attainable in the scale up from the small equipment used to attain the energy use values to industrial equipment. The RIN-enabled cost reduction is the 50/50 CWC value for the lower sugar to jet fuel yield. Together these values drop the non-concentrated MSSP to \$0.33/kg.

#### MSSP Comparison

The USDA publishes multiple annual sugar price indexes. The price for US raw sugar and glucose syrup were the closest to the product we are discussing. However, they are food grade products and although the lack of chemical pre-treatment means that this sugar could be food grade, approval by the Food and Drug Administration would be required. The values from 1990 to 2015 are shown in **Figure 6** along with the base case, optimized case and optimized with the real discount rate increased to 15% (USDA, 2016a,b). The cost of sugar from micronized wood for the three chosen scenarios is comparable with both of the USDA data sets. Zhang et al. (2013) converted red oak into monosaccharides, gasoline, diesel and hydrogen to attain an 11.4% IRR; the sugar cost used was \$0.64/kg. The prices calculated within this study illustrate that stopping at sugar is an economically viable pathway, depending on the market prices of downstream products.

## CONCLUSIONS

The results of the financial analysis of using mechanical pretreatment for conversion of softwood biomass, specifically FHR, into sugars is promising, with cost values in line with both US raw sugar and glucose syrup. The success of this technology will be controlled by electricity use and cost, siting decisions, downstream biofuel or biochemical production and the production of co-products. Investigation into higher-value lignin-based co-products could reduce the minimum sugar selling price. Additional analysis that will help better determine financial viability of micronizing wood include the use of micronizing depots and co-location cost reductions. These

#### REFERENCES


reductions may come in the form of overhead and fixed costs; however, they could also come from locating the facilities with a traditional wood-product facility to maximize the financial benefits of waste streams. Further cost reductions are possible with facility integration, co-location, siting at a closed mill or brownfield location.

## AUTHOR CONTRIBUTIONS

The study described in this work was designed by MW, JG, and JW and the data was largely compiled by KB, JG, and RW. The analysis was completed by KB, JG, RW, and MW. KB wrote the draft manuscript that was reviewed by MW, JG, RW, and JW.

## FUNDING

This work, as part of the Northwest Advanced Renewables Alliance (NARA), was supported by the Agriculture and Food Research Initiative Competitive Grant no. 2011-68005-30416 from the USDA National Institute of Food and Agriculture. Partial funding was also generously provided by the Sun Grant Program as part of a project titled Advancing the National Bioeconomy through Regional Sun Grant Centers award number 2014-38502-22598.

#### ACKNOWLEDGMENTS

The authors express the sincere appreciation to Wallace E. Tyner and Dawoon Jeong for their assistance in completing the economic analysis. We would like to acknowledge the delivered feedstock cost modeling completed by Natalie Martinkus.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00077/full#supplementary-material

Chemical Engineering (2017). Economic indicators. Chem. Eng. 124, 164.


**Disclaimer:** The views and opinions expressed in this article are those of the authors and do not necessarily reflect the official policy or position of any corporation and any agency of the U.S. government.

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer, JST, and handling Editor declared their shared affiliation.

Copyright © 2018 Brandt, Gao, Wang, Wooley and Wolcott. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Blended Feedstocks for Thermochemical Conversion: Biomass Characterization and Bio-Oil Production From Switchgrass-Pine Residues Blends

#### Edited by:

Gordon Graham Allison, Aberystwyth University, United Kingdom

#### Reviewed by:

Selhan Karagoz, Karabük University, Turkey Tianju Chen, Qingdao Institute of Bioenergy and Bioprocess Technology, Chinese Academy of Sciences (CAS), China

#### \*Correspondence:

Charles W. Edmunds cedmund1@utk.edu Nicole Labbé nlabbe@utk.edu

#### Specialty section:

This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research

Received: 11 April 2018 Accepted: 20 July 2018 Published: 17 August 2018

#### Citation:

Edmunds CW, Reyes Molina EA, André N, Hamilton C, Park S, Fasina O, Adhikari S, Kelley SS, Tumuluru JS, Rials TG and Labbé N (2018) Blended Feedstocks for Thermochemical Conversion: Biomass Characterization and Bio-Oil Production From Switchgrass-Pine Residues Blends. Front. Energy Res. 6:79. doi: 10.3389/fenrg.2018.00079 Charles W. Edmunds <sup>1</sup> \*, Eliezer A. Reyes Molina<sup>2</sup> , Nicolas André<sup>1</sup> , Choo Hamilton<sup>1</sup> , Sunkyu Park <sup>2</sup> , Oladiran Fasina<sup>3</sup> , Sushil Adhikari <sup>3</sup> , Stephen S. Kelley <sup>2</sup> , Jaya S. Tumuluru<sup>4</sup> , Timothy G. Rials <sup>1</sup> and Nicole Labbé<sup>1</sup> \*

<sup>1</sup> Center for Renewable Carbon, University of Tennessee, Knoxville, TN, United States, <sup>2</sup> Department of Forest Biomaterials, North Carolina State University, Raleigh, NC, United States, <sup>3</sup> Department of Biosystems Engineering, Auburn University, Auburn, AL, United States, <sup>4</sup> Idaho National Laboratory, Biofuels and Renewable Energy Technologies Division, Idaho Falls, ID, United States

An abundant, low-cost, and high-quality supply of lignocellulosic feedstock is necessary to realize the large-scale implementation of biomass conversion technologies capable of producing renewable fuels, chemicals, and products. Barriers to this goal include the variability in the chemical and physical properties of available biomass, and the seasonal and geographic availability of biomass. Blending several different types of biomass to produce consistent feedstocks offers a solution to these problems and allows for control over the specifications of the feedstocks. For thermochemical conversion processes, attributes of interest include carbon content, total ash, specific inorganics, density, particle size, and moisture content. In this work, a series of switchgrass and pine residues blends with varying physical and chemical properties were evaluated. Physical and chemical properties of the pure and blended materials were measured, including compositional analysis, elemental analysis, compressibility, flowability, density, and particle size distribution. To screen blends for thermochemical conversion behavior, the analytical technique, pyrolysis gas chromatography mass spectrometry (Py-GC/MS), was used to analyze the vapor-phase pyrolysis products of the various switchgrass/pine residues blends. The py-GC/MS findings were validated by investigating the bio-oils produced from the selected blends using a lab-scale fluidized-bed pyrolysis reactor system. Results indicate that the physical properties of blended materials are proportional to the blend ratio of pure feedstocks. In addition, pyrolysis of pine residues resulted in bio-oils with higher carbon content and lower oxygen content, while switchgrass derived pyrolysis products contained relatively greater amount of anhydrosugars and organic acids. The distribution of the pyrolysis vapors and isolated bio-oils appear to be a simple linear combination of the two feedstocks. The concentration of alkali and alkaline

**146**

earth metals (Ca, K, Mg, and Na) in the blended feedstocks were confirmed to be a critical parameter due to their negative effects on the bio-oil yield. This work demonstrates that blending different sources of biomass can be an effective strategy to produce a consistent feedstock for thermochemical conversion.

Keywords: fast pyrolysis, biofuels, biomass, ash content, yield, inorganic metals

# INTRODUCTION

Lignocellulosic biomass represents a renewable and sustainable resource that can be utilized to produce fuels, chemicals, and other products. Over the past few decades, research interest in bio-oil derived from fast pyrolysis of lignocellulosic biomass has attracted attention as a potential renewable and sustainable alternative for hydrocarbon transportation fuels or chemical products (Chiaramonti et al., 2007). Generally, bio-oil is an acidic, dark-colored liquid composed of hundreds of oxygencontaining compounds derived from the rapid depolymerization, dehydration, and fragmentation of the carbohydrate and lignin components present in the feedstock biomass (Mohan et al., 2006). The high acidity and low miscibility of bio-oil with petroleum derived products make their direct application difficult (Lehto et al., 2014), and thus upgrading of the bio-oil is a necessity (Zhang et al., 2007).

The United States have the potential to supply 1 billion ton per year of diverse biomass feedstocks including dedicated bioenergy crops, agricultural residues, and forestry residues (U. S. Department of Energy, 2016). However, there are challenges associated with the implementation and commercialization of biomass conversion technologies that need to be overcome for this to become a reality. A high quality, low-cost, and consistent supply of biomass is required to feed biorefineries (Thompson et al., 2016). Other issues in feedstock supply logistics include the regional and seasonal factors that affect the availability and supply of different biomass resources (Sultana and Kumar, 2011).

Blending multiple biomass sources offers a solution to the challenges associated with feedstock quality, variability, supply, and cost (Kenney et al., 2013; Thompson et al., 2013; Ray et al., 2017). Benefits of biomass blending or formulation include increasing the potential biomass supply in a given area surrounding a biomass processing facility or biorefinery. In addition, blending has been demonstrated to provide favorable feedstock flowability and pelleting characteristics (Yancey et al., 2013; Crawford et al., 2015). Recently, several different pretreatments and bioprocesses have been applied to blended feedstocks in the scope of biochemical conversion. These comprise pretreatments such as steam, SO2-catalyzed steam, dilute acid, ionic liquid, and bioconversion processes including enzymatic hydrolysis, fermentation, and organosolv fractionation (Shi et al., 2013; Astner et al., 2015; Vera et al., 2015; Wolfrum et al., 2017).

Blending different sources of lignocellulosic biomass to produce feedstocks for thermochemical conversion has also received recent attention. For example, Mahadevan et al. (2016a) reported that bio-oil from a blend of switchgrass and southern pine wood that contained greater proportions of switchgrass had the beneficial characteristics of lower acidity and viscosity; however, the bio-oil had higher water content. Carpenter et al. (2017) demonstrated that pyrolytic bio-oil yields and hydrodeoxygenated product yields from blended samples composed of clean pine, tulip poplar, and switchgrass in different combinations showed a linear trend based on the blends' components. In addition, Ren et al. (2017) reported on blending loblolly pine wood and bark for pyrolysis feedstocks, and indicated that as the ratio of bark increased, the organic yields decreased, char yields increased, and levoglucosan yields increased.

Due to the challenges that feedstock supply and logistics are facing for the biomass conversion industry, and the emergence of blending as a solution to these challenges, we recognize the need to integrate different aspects regarding feedstock processing, biomass chemistry, and thermochemical conversion. Therefore, in this work, we integrate several topics addressing the suitability of blending switchgrass and pine residues to produce feedstocks for thermochemical conversion. These topics include (a) biomass physical characteristics relevant to processability and feeding of the blends, (b) biomass chemical properties related to thermochemical conversion, (c) the analysis of vapor-phase pyrolysis products, and (d) pyrolysis product yields.

# MATERIALS AND METHODS

# Biomass Feedstocks

#### Feedstock Harvesting and Preprocessing

Switchgrass and pine residues feedstocks were selected for this study based on their potential suitability for bioenergy feedstocks (Perlack and Stokes, 2011). The switchgrass, (P. virgatum L.) cv. Alamo, was field-grown and harvested in Vonore, TN, and processed with a tubgrinder by Genera Energy Inc. (Vonore, TN). Switchgrass is denoted as "SG." The pine residues samples were harvested from forest stands near Auburn, AL, and consisted of composite samples of small diameter tree tops, limbs, and needles from 50 Loblolly pine trees. Two large batches (2,000– 3,000 kg) of pine residues were collected by Auburn University with two different tree top (or stem) diameters that were included as the pine residues material. Residues harvested from 2 in. (50.8 mm) diameter tree tops are denoted as "2PN" and residues generated from 6 in. (152.4 mm) diameter tree tops are denoted as "6PN." The pine residues were dried and both, pine residues and switchgrass, were hammer-milled to pass through a 3/16 in. (4.76 mm) screen at Herty Advanced Biomaterials (Savannah, GA). Prior to chemical analysis, representative samples were knife-milled using a Wiley mill to pass through a 0.425 mm (40-mesh) screen.

#### Blended Feedstock Preparation

Binary blends of (a) SG and 2PN, and (b) SG and 6PN were prepared by weighing the appropriate amount of each respective feedstock to achieve the desired weight ratio. Blends were prepared by mixing SG and PN in 12.5% (0.125 wt. fraction) increments, resulting in 7 blends in addition to the two pure feedstocks for each binary blend (**Table 1**). Samples designated for Py-GCMS experiments were further homogenized with a rotating ball mill (PM 100, Retsch, Germany) using zirconium oxide balls and milling cup. The ball mill program was designed as to not induce heat-related degradation to the biomass and consisted of 3 cycles of 60 sec of milling at 500 rpm with a 10 min rest time between cycles. Samples designated for fluidized bed pyrolysis experiments were milled further to pass through 0.5 mm screen using a Wiley mill.

#### Biomass Chemical Analysis

#### Ash, Structural Carbohydrates, and Lignin Content

The total ash content was measured by combustion at 575◦C following the standard laboratory procedure developed by the National Renewable Energy Laboratory (NREL; Sluiter et al., 2008). Biomass was sequentially extracted with water and ethanol using an ASE 350 accelerated solvent extractor (Dionex, Sunnyvale, CA), and the cellulose, hemicellulose, and lignin content of the extractives-free material were measured by following the standard NREL protocol (Sluiter et al., 2010). Sugars were quantified by high-performance liquid chromatography (Flexar, PerkinElmer, Shelton, CT) equipped with a deashing guard column (125-0118, Bio-Rad, Hercules, CA) and Aminex HPX-87P carbohydrate column (300 x 7.8 mm ID) with a column temperature of 85◦C. Deionized water was used as the mobile phase with a flow rate of 0.25 mL/min. Acid insoluble lignin was measured gravimetrically, and acid soluble lignin was measured with a UV/VIS spectrometer (Thomas Fisher Scientific, Pittsburgh, PA). Total lignin content is reported as the sum of acid soluble and acid insoluble lignin. Analysis was performed in triplicate.

#### Inorganic and CHN Analysis

The concentration of inorganics was measured by microwaveassisted acid digestion and inorganics were detected by inductively coupled plasma-optical emission spectroscopy (ICP-OES) in accordance with the US Environmental Protection Agency (EPA) standard protocol (EPA, 1996). Milled biomass (0.5 g) was added to a solution of HNO<sup>3</sup> (4 mL, 67-70%), of H2O<sup>2</sup> (3 mL, 35%), and of HF (0.2 mL, 48%) in a PTFE pressure tube, and 1200 W of microwave power was applied using a Multiwave 3000 microwave digester (Anton Paar, Richmond, VA) to achieve a digestion temperature of 180–210◦C. After digestion, the reaction solution was filtered with PTFE syringe filters (0.20µm), then analyzed by ICP-OES (Optima 7300 Dual View, Perkin Elmer, Shelton, CT). The carbon, hydrogen, and nitrogen content were measured in duplicate with an elemental analyzer (Vario MICRO cube, Elementar, Ronkonkoma, NY), and the oxygen content was calculated by difference. Analysis was performed in triplicate.

# Feedstock Physical Properties Characterization

The particle size distribution of each sample was measured using a Camsizer particle analyzer (Retsch Technology, Hann, Germany). Approximately 100 g of a material were loaded into the instrument hopper, then a vibratory feeder conveyed the biomass into the analyzer. The diameters in which 90%, 50%, and 10% of the particles are smaller than are reported as d90, d50, and d10. The span (defined below) was used as a measure of the variability, and hence the distribution, of the sizes of particles in each sample.

$$\text{Span} = \frac{d90 - d10}{d50} \tag{1}$$

Bulk density was determined by a bulk density measuring apparatus (Burrows Co., Evanston, IL). This method involved pouring the bulk sample into a container (volume of 1,137 mm<sup>3</sup> ) from a funnel (positioned at a height of 610 mm above the top edge of the container). The heap formed by the sample was then carefully leveled with the top of the surface of the container. The sample in the container was then weighed. Bulk density was calculated as the ratio of the sample mass to the container volume. The particle density of each sample was measured with an AccuPyc 1340 gas pycnometer (Micromeritics Instrument Corp., Norcross, GA). A model TD-12 automated tap density tester (Pharma Alliance Group Inc., Valencia, CA) was used to measure the tap density of the feedstock samples according to ASTM B527 (ASTM, 2015). Samples were weighted to a precision of 0.001 g using a digital balance.

The compressibility (CM) was determined by filling a compression cell (height = 101.83 mm and internal diameter = 49.55 mm) with biomass, and using a fitted piston (diameter = 49.00 mm) attached to the cross-head of a model TA-HD texture analyzer (Stable MicroSystems, Surrey, U.K.). The piston was operated at a 1 mm/s compression rate and 6 kPa consolidating pressure. The following equation was used to calculate compressibility (Littlefield et al., 2011):

$$CM = 100 \left(\frac{V\_i - V\_f}{V\_i}\right) \tag{2}$$

where V<sup>i</sup> and V<sup>f</sup> are the initial and final volume of the sample, respectively. All of the physical properties analyses were carried out in duplicate.

#### Pyrolysis Experiments Py-GC/MS and Multivariate Data Analysis

The composition of the vapor-phase pyrolysis products of the pure and blended feedstocks was screened by using analytical pyrolysis-gas chromatography-mass spectrometry (Py-GC/MS). The Py-GC/MS instrumentation consisted of an EGA/Py-3030 D micropyrolyzer (Frontier Lab, Japan) attached to a Clarus 680 gas chromatograph and Clarus SQ 8C mass spectrometer (PerkinElmer, Shelton, CT). Biomass samples (0.5 mg) were added to a metal sample cup and an autosampler was used to drop the samples into the pyrolysis furnace heated to 500◦C. After a pyrolysis furnace residence time of 12 s, the vapors were swept


TABLE 1 | Weight ratios of prepared switchgrass/pine residues blends and associated analyses performed for each blend ratio.

into the GC via direct attachment to the injection port [unpacked 2 mm quartz liner, split ratio 80:1, and injector temperature of 270◦C via the carrier gas (ultrahigh-purity helium, 99.9999%)]. An Elite 1701 MS gas capillary column (60 m length, 0.25 mm ID, and 0.25µm film thickness) was used with He carrier gas (1 cm<sup>3</sup> /min flow rate and 17.3 psi pressure). The GC furnace used a time temperature ramp program of 4 min at 50◦C, followed by a ramp of 5◦C/min to 280◦C, and then hold at 280◦C for 5 min. The MS was operated with an ionization energy of 70 eV and temperature of 280◦C. Two-hundred pyrogram peaks and associated peak areas were extracted using the TurboMass GC/MS software with a Signal/Noise ≥ 2,000 and identified with the National Institute of Standards and Technology (NIST) library. Five replicates per sample were performed.

In order to visualize trends in the vapor-phase pyrolysis products derived from pure and blended biomass samples, we performed the multivariate statistical analysis, principal component analysis (PCA) using the statistical software, The Unscrambler X ver. 10.4 (Camo software Inc., Woodbridge, NJ). In total, 166 pyrogram peaks were selected and assigned a peak number corresponding to its order based on retention time. Pyrogram peak areas were normalized based on the total peak area prior to PCA.

#### Fluidized-Bed Pyrolysis Reactor and Bio-Oil Characterization

A schematic of the lab-scale fluid-bed reactor is shown in **Figure 1** (Meng et al., 2012). The system consists of a screw feeder, an externally heated fluid bed, a char collection cyclone, and a bio-oil collection system consisting of two water-cooled condensers and an electrostatic precipitator (ESP). Sand was used in the fluidized bed as the heat transfer media. Nitrogen was introduced from the bottom of the reactor at a flow rate of 4.5 L/min, and a secondary nitrogen stream was introduced prior to the screw auger feeding the reactor with a flow rate of 6.5 L/min. These nitrogen streams maintain an oxygen-free environment to maximize the yield of bio-oil. The fluid-bed was maintained at 510◦C and the residence time in the reactor was calculated to be between about 1.2–1.5 s (Basu, 2010). The biomass feeding rate was 150 g/hr and the total run time was approximately 60 min. The pyrolysis vapors were rapidly quenched and collected in the two condensers which were chilled to 2–4◦C (Freel and Graham, 1998), and aerosol particles were collected in the ESP. The yields of biochar and bio-oil were measured gravimetrically, while the non-condensable gases (NCG) were measured by difference. Furthermore, the yield of reaction water was calculated by normalizing the measured water content in the bio-oil by the total bio-oil yield, and the bio-oil organic yield is calculated as the total bio-oil yield minus the reaction water. All pyrolysis runs were performed in duplicate.

The carbon, hydrogen, and nitrogen (CHN) content of bio-oil samples were measured with a CHN Elemental Analyzer (2400 Series II, Perkin Elmer, Shelton, CT), and the oxygen content was calculated by difference. The molecular weight of the produced bio-oils were determined by the gel permeation chromatography (GPC). The GPC instrument (Shimadzu, Kyoto, Japan) was equipped with two columns (Waters Styragel HR 5E and Styragel HR 1). Calibration was achieved with polystyrene standards using a refractive index detector. Approximately 3 mg of bio-oil sample were dissolved into 10 mL of tetrahydrofuran (THF) and then passed through a 0.2µm PTFE syringe filter. An aliquot of the solution (20 µL) was injected into the GPC, and THF was used as the eluent at flow rate of 0.7 mL/min. The numberaverage molecular weight (Mn) and average molecular weight (Mw) were calculated using the instrument software (Shimadzu LC solution-GPC Postrum).

# RESULTS AND DISCUSSION

# Physico-Chemical Characteristics of Pure and Blended Feedstocks

#### Feedstock Composition

Switchgrass (SG) and pine residues (PN) were used for this study based on their availability and suitability as bioenergy feedstocks in the southeastern United States. Switchgrass is recognized as a good bioenergy crop due to its high biomass output and ability to be grown on marginal soils (Mitchell et al., 2016). Pine residues are defined as small diameter (2 or 6 inches) tree tops, limbs, and needles that are generally considered as waste during harvesting

of pine forests. This material is potentially abundant and low-cost (Perlack and Stokes, 2011). We investigated two batches of pine residues, 6PN is derived from 6′′ diameter tree tops, limbs, and needles and has a higher proportion of clean wood compared to 2PN, which is derived from 2′′ tree tops, limbs, and needles, and has a higher proportion of bark and needles. We hypothesize that 6PN will be a higher value feedstock based on the higher proportion of clean wood, and associated decrease in ash and inorganics content compared to the 2PN feedstock.

The chemical composition, CHN content, and inorganic analysis for the three individual feedstocks are shown in **Tables 2**, **3**. The cellulose and hemicellulose content are the highest in SG, followed by 6PN, while 2PN contains the least. Among the three feedstocks, SG has the highest ash content at 1.30%, followed by 2PN (1.13%), and 6PN the lowest ash content at 0.76%. The concentration of the combined alkali and alkaline earth metals (AAEMs), which includes Ca, K, Mg, and Na, follows the trend of 2PN>6PN>SG. When only considering the pine residues feedstocks, 6PN contains lower ash, inorganics (especially the AAEMs), and extractives compared to the 2PN, which can be explained by the higher proportion of clean wood found in larger stem diameter pine residues materials (6′′ vs. 2′′ stem diameter).

Due to lessened ash and AAEMs content, we expect 6PN to be a higher quality feedstock compared to the 2PN material.

TABLE 2 | Chemical composition of feedstock biomass.


Value in parenthesis represent the standard deviation calculated from three replicates.

Switchgrass (SG), despite having a higher ash content than the 2PN and 6PN feedstocks, shows the lowest concentration of AAEMs. The higher ash content in switchgrass can be largely attributed to its high silicon content, which is commonly reported in switchgrass (Vassilev et al., 2010). Silicon is inert under pyrolysis conditions, and while high content is undesirable due to the reduced amount of carbon available for conversion, it is not detrimental to the conversion reaction like the AAEMs (Fahmi et al., 2008; Patwardhan et al., 2010; Mahadevan et al., 2016b).

Carbon content is the highest in 2PN (53.5%), followed by 6PN (52.0%), and SG (49.2%) and the same order is observed for nitrogen, where nitrogen content is higher in 2PN (0.5%) compared to 6PN (0.2%) and SG (0.2%). Higher carbon contents have been correlated with increased heating value and energy content in biomass (Tillman, 1978) while increased N content in feedstock has been associated with reduced bio-oil acidity; but also with reduced bio-oil stability (Mante and Agblevor, 2014).

The chemical composition of the feedstocks can have a major impact on the resulting pyrolysis product yields and bio-oil quality (Mohan et al., 2006). Under typical pyrolysis conditions, structural carbohydrates (cellulose and hemicellulose) are converted to anhydrosugars, organic acids, furans, ketones, aldehydes, char, and non-condensable gases, while lignin is converted into phenolic compounds, light organic oxygenates,


Value in parenthesis represent the standard deviation calculated from three replicates. \*Combined AAEM (alkaline and alkali earth metals) is the sum of Ca, K, Mg, and Na.

acids, char, and non-condensable gases (Carpenter et al., 2014). In addition, previous work demonstrates that the presence of ash, especially AAEMs, results in greater formation of organic acids, non-condensable gases, and char during pyrolysis (Fahmi et al., 2008; Patwardhan et al., 2010; Mahadevan et al., 2016b).

#### Physical Characteristics Relevant to Processability and Feeding

The mean values of the particle size characteristics, bulk, particle and tap densities, compressibility, and Hausner ratio (used as a measure of flowability) are summarized in **Table 4**. Tukey's multiple comparison tests (SAS Institute, Cary, NC) on these mean values show that switchgrass bulk and tap densities are significantly (p-value<0.05) lower than the pine residues samples (2′′ and 6′′). When switchgrass is blended with the pine residues samples, the bulk and tap densities of the resulting blends are almost proportional to the ratios of the pine sample in the blend. A similar trend is observed for the d50 particle size (increasing with pine amount). The reverse trend is seen for the span (an indication of the variability in the size of particles in a particular sample), and as the amount of pine residues increased, the span was reduced (**Figure 2**).

The higher span of switchgrass particles is probably due to the elongated nature and higher aspect ratio of switchgrass grinds that was observed. We hypothesize that the more elongated nature of switchgrass causes entangling of the particles in a bulk sample and therefore may contribute to the lower densities of bulk switchgrass (i.e., bulk and tap densities). Even though there are significant differences in the particle densities of the samples, there is not an observable trend regarding the effect of sample type (switchgrass vs. pine residues) or relative amount of switchgrass in a switchgrass/pine residues blend.

The Hausner ratio (defined as tap density/bulk density) can be used as an indicator of the flowability of milled materials (Bernhart and Fasina, 2009). The Hausner ratio of the samples and the blends varies slightly from 1.26 to 1.33. According to Figura and Teixeira (2007), this indicates that the samples may have difficulty at flowing. Investigating densification techniques,


\*SG, switchgrass; CM, compressibility; HR, hausner ratio; Values followed by the same letter are not significantly different at p<0.05.

such as pelletization, may be required to improve handling and flowability characteristics, and is discussed later. These results agree with Crawford et al. (2015), who demonstrated that flow properties of blends composed of switchgrass, corn stover, miscanthus, and hybrid poplar behaved in a linear fashion depending on the characteristics of the blend's individual components.

# Py-GC/MS of Blended Samples

We used micro-scale pyrolysis gas chromatography mass spectrometry (Py-GC/MS) to analyze the vapor-phase pyrolysis products of the various switchgrass/pine residues blends. Using a small-scale pyrolysis system allows us to rapidly screen a larger number of samples in order to investigate blends for thermochemical conversion behavior. Due to the ambiguity of visual comparison of the numerous peaks in the Py-GC/MS pyrograms among many different samples, we used principal component analysis (PCA) to identify the sources of maximum variation and visualize trends in the vapor-phase pyrolysis products from the pure and blended samples. In addition, percent normalized peak areas for several pyrolysis products were compared among pure and blended samples.

#### Principal Component Analysis

**Figure 3** shows the PCA scores and loadings plots for SG, 2PN, and their blends which vary by weight fraction increments of 0.125 (**Table 1**). The PCA for SG, 6PN, and their blends is provided in the Supplemental Material (**Figure S1**). The data input into this PCA consisted of 4–5 replicate pyrograms for each biomass sample, and 166 peaks were detected in each pyrogram. The same 166 peaks occurred in all sample pyrograms in varying intensities. The NIST library was used to identify as many compounds as possible based on the mass fragmentation pattern, and identified peaks are listed in **Table 5**. The pyrogram peaks areas were normalized based on total peak area prior to analysis; therefore, the PCA results indicate relative concentration of vapor-phase pyrolysis products.

PC1 and PC2 accounted for 88 and 3% of the total variance in the dataset, respectively, indicating that most of the differences in the pyrolysis product distribution are represented by PC1. The scores plot shows that the pyrograms from 2PN are plotted at the most negative location and SG at the most positive location along the PC1 axis. The blended samples are plotted on the PC1 axis between the switchgrass and pine residues feedstocks in order of their blend ratio, with the trend of increasing PC1 score with higher ratios of switchgrass (**Figure 3A**).

The loadings plot for PC1, shown in **Figure 3B**, indicate which specific pyrogram peaks (representing chemical compounds) are the most significant in explaining the trends observed in the scores plot. For example, peaks in the loadings plot with positive intensities occur in greater relative amounts in the sample pyrograms with positive scores, and likewise, loadings with negative peaks occur in greater relative abundance in samples falling on the negative quadrant of PC1 in the scores plot. Therefore, compounds with positive intensity in the loadings plot have relatively greater abundant in pure SG and blends with higher proportions of SG. These compounds include light oxygenates such as acetic anhydride (6); organic acids including acetic acid (17), acetoxyacetic acid (29), and succinohydrazide (59); furans comprising furfural (37) and 2,3-dihydrobenzofuran (107); phenols such as 2,6-dimethoxyphenol (syringol) (116), 4 allyl-2,6-dimethoxyphenol (156); and sugars derived compounds such as methyl-α-D-ribofuranoside (127) and levoglucosan (155).

Similarly, peaks in the loadings plot with negative intensity occur in relatively greater amounts in 2PN and blends with higher proportions of 2PN. Phenolic compounds made up the majority of these and include phenol (72), 2-methoxyphenol (guaiacol) (74), 2-methoxy-4 methylphenol (92), 2-methoxy-4-vinylphenol (108), eugenol (112), 2-Methoxy-4-[(1Z)-1-propen-1-yl]phenol (Isoeugenol) (126), vanillin (129), and coniferyl alcohol (165). Other compounds with negative loadings include CO<sup>2</sup> (1), methanol (3), 1-hydroxyacetone (20), and 1,2-cyclopentanedione (54).

The pyrolysis products distributions from the switchgrass and pine residues feedstocks are related to their chemical compositions. Switchgrass contains a higher amount of cellulose and hemicellulose (**Table 2**), which have been shown to thermally degrade to organic acids, aldehydes, furans, ketones, TABLE 5 | Vapor-phase pyrolysis product compounds identified using pyrolysis gas chromatography mass spectrometry (Py-GC/MS) in pure switchgrass, 2PN, and 6PN biomass samples under identical pyrolysis conditions.


(Continued)

#### TABLE 5 | Continued

#### TABLE 5 | Continued


(Continued)

Frontiers in Energy Research | www.frontiersin.org

109–111 29.42–29.74 ?

and Milne, 1987). In addition, the relatively higher amount of methanol and phenolic compounds in pine residues pyrolysis vapors may be caused by the higher lignin content in the pine residues feedstocks compared to switchgrass (**Table 2**; Jegers and Klein, 1985).

Increased relative yields of CO<sup>2</sup> and methanol in 2PN are indicated by negative PC1 loadings, which can be attributed to the higher ash and inorganics content in 2PN. Ash and particularly AAEMs are more reactive than other inorganics and effectively catalyze ring-breaking, depolymerization, and fragmentation reaction, which results in increased production of non-condensable gases, organic acids, and other light oxygenated compounds (Carpenter et al., 2014; Mahadevan et al., 2016b). Similar trends are observed in the PCA results obtained from the Py-GC/MS data for SG, 6PN, and their blends (**Figure S1**).

A linear trend across PC1 in relation to the blend ratio is observed in the PCA scores plots for both SG/2PN and SG/6PN blends. This implies that the pyrolysis products distribution in blended samples mainly behaves as a simple linear combination of the two feedstocks. This also demonstrates that there are no significant non-linear effects on the pyrolysis vapor products that occurred when using blended materials as pyrolysis feedstocks. Carpenter et al. (2017) reported a similar linear trend based on the blend components for yields of raw and upgraded bio-oil from blends of switchgrass, pine wood, tulip poplar, and oriented strand board. In addition, Mahadevan et al. (2016a) tested blends of southern pine and switchgrass as a pyrolysis feedstock, and reported similar blending effects for pyrolysis products yields and bio-oil characteristics.

#### Pyrogram Peak Area Analysis

To further investigate the effect of blending on the vaporphase products formation, we analyzed some specific compounds which were identified as having high variability in the PCA loadings plot. **Figure 4** shows the percent normalized Py-GC/MS pyrogram peaks areas for CO2, acetic acid, 2,3 dihydrobenzofuran, isoeugenol, levoglucosan, and coniferyl alcohol from the SG:2PN and SG:6PN blends. For all compounds shown, there is a distinct linear correlation, which further evidences that the pyrolysis products distribution of blended samples behave as a simple linear combination of its components. These trends are in agreement with the previously discussed PCA results (**Figure 3**). The relative production of CO<sup>2</sup> is higher in the SG/2PN blends, which is most likely caused by the higher ash and inorganics content in 2PN compared to 6PN. This is significant because higher yields of non-condensable gases, such as CO2, will result in reduced yields of bio-oil.

The relative yields of coniferyl alcohol are greater for the SG/6PN blends compared to the SG/2PN blends. This is probably related to the relatively lower inorganics content in the 6PN compared to 2PN. Mahadevan et al. (2016b) demonstrated that increased concentrations of AAEMs result in decreased coniferyl alcohol yields in pyrolysis vapors. When comparing the SG/2PN and SG/6PN blends, the trends and % peak areas for acetic acid, 2,3-dihydrobenzofuran, isoeugenol, and levoglucosan are very similar between the two. This indicates that differences in the chemical composition between the 2PN and 6PN feedstocks did not strongly influence the relative yields for these specific compounds.

# Fluidized-Bed Pyrolysis of Selected Blends

A fluidized-bed pyrolysis reactor was used to produce bio-oil from pure switchgrass, pure pine residues (2PN and 6PN), and 0.25/0.75, 0.5:0.5, and 0.75:0.25 (mass fraction) blends of the two.

The pyrolysis product yields are shown in **Figure 5**. The yield of biochar and reaction water range from 8.7 to 15.9% and 10.6 to 14.5%, respectively, and both increase as the pine residues content of the blends augment in both SG/2PN and SG/6PN blends. The yield of NCG ranges from 28.6 to 34.6% and decreases with increasing pine residues content in the blends. A decreasing trend of bio-oil organic yield is observed for the SG/2PN blends, while no clear trend is observed in SG/6PN blends. The content in AAEMs in the 2PN feedstock, compared to SG and 6PN, resulted in reduced bio-oil organic yields. Within error, there is a linear trend in the pyrolysis products yields, based on the blend proportions of the pure feedstocks. Other researchers have reported similar linear trends in pyrolysis products yields in blends from different lignocellulose species (Carpenter et al., 2017).

The content of carbon, hydrogen, nitrogen, and oxygen and molecular weight of the produced bio-oils are shown in **Table 6**. The carbon content and the average molecular weight (Mw) and number-average molecular weight (Mn) of bio-oil from the pure feedstocks is highest in 2PN, followed by 6PN, and lowest in SG. The inverse is observed for oxygen content where the trend is 2PN<6PN<SG. The bio-oil molecular weight results are likely related to the feedstock lignin content. During pyrolysis, lignin is decomposed into higher molecular weight phenolic compounds (Evans and Milne, 1987), thus the trend in bio-oil molecular weight is the same as the trend for feedstock lignin content (2PN>6PN>SG). In addition, most of the oxygen contained in the feedstock is preserved in the bio-oil product (Mohan et al., 2006), thus feedstock with higher oxygen content produced bio-oils with higher oxygen content. Our results are in agreement with previous research which has demonstrated a positive correlation between lignin content and bio-oil molecular weight, and a negative correlation between lignin content and bio-oil oxygen content (Fahmi et al., 2008). In addition, Mahadevan et al. (2016a) reported higher carbon content and lower oxygen content bio-oil produced from southern pine compared to bio-oil from switchgrass. Similar to the other measured properties, the carbon and oxygen contents and molecular weight of bio-oil from blended feedstock behave as a linear combination based on the feedstock's blend ratio.

To probe the relationship between the concentration of inorganics and pyrolysis product yields in the pure and blended samples, a Pearson correlation analysis (SAS Institute, Cary, NC) was performed (**Table S1**). Significant correlations (defined as having a p-value below 0.01) are found between several inorganics and the yield of pyrolysis products. For example, strong correlations are observed between the concentration of K and the yield of biochar, bio-oil organic fraction, and reaction water, having correlation coefficient (r) of 0.98, −0.83, and 0.92, respectively. In addition, the correlation plots for total AAEMs (the sum of Ca, K, Mg, and Na) and pyrolysis products yield

are shown in **Figure 6**. The correlation (r) between the AAEM concentration and bio-oil organic fraction "Organic," biochar, and water yield are strong with correlation coefficient of −0.80, 0.91, and 0.79, respectively. The correlation between AAEMs and non-condensable gases is moderate with r = −0.63. In addition, ash content is not significantly correlated with the pyrolysis products yields. Although high ash content is not desirable due to overall reduced carbon content in the biomass, it is not

necessarily the best indicator of biomass quality parameter to use when assessing biomass for conversion via pyrolysis.

Feedstock properties that affect pyrolysis products yields include the organic composition of the feedstocks (cellulose, hemicellulose, and lignin) as well as the inorganic composition (ash and alkali and alkaline earth metals). However, given our current results, it is difficult to delineate the effects of organic vs. inorganic constituents. Previous research has reported that the inorganic composition (mainly the content of reactive AAEMs) of the feedstock has a greater influence on the pyrolysis product yields than the lignin content (Fahmi et al., 2008). Alkali and alkaline earth metals have been shown to result in increased reaction water, increased biochar yields, and decreased organic liquid yields (Das et al., 2004; Fahmi et al., 2008). This is agreeable to our observation of increased biochar yield with increasing pine residues, as the pine residues used in this work have higher inorganics content. In addition, decreased bio-oil organic fraction yields in the case of SG/2PN blends are likely a result of the higher AAEM content of the 2PN feedstock (**Table 3**).

Based on previous work, the effects of cellulose and lignin content in pyrolysis feedstocks on the pyrolysis products yields are not as clear. For example, several studies have reported increasing bio-oil yields and decreasing gas and biochar yields as feedstock lignin content increased (Fahmi et al., 2008; Tröger et al., 2013). Conversely, Lv et al. (2010) reported higher biochar and gas yields with increasing lignin content. There are likely interactions between the organic constituents and the inorganics of biomass (mainly reactive alkali and alkaline earth metals) which make understanding the relationship between biomass composition and pyrolysis products yields complex and difficult to delineate (Couhert et al., 2009; Tanger et al., 2013).

#### Blend Discussion/Recommendation The Impact of Blending on Feedstock Physical and Chemical Properties and Densification

Woody and herbaceous biomass feedstocks both exhibit significant variability in their chemical and physical composition. Blending of different types of biomass can help to produce a consistent feedstock that meets specifications required for thermochemical conversion in terms of energy, volatiles, inorganics, and ash content (Tumuluru et al., 2012). Densification is often considered as a necessary step to increase


TABLE 6 | Elemental analysis and molecular weight of bio-oils produced from pure and blended switchgrass and pine residues feedstocks.

Value in parenthesis represent the standard deviation calculated from three replicates.

\*Mn, number-average molecular weight, M<sup>w</sup> = weight-average molecular weight.

the bulk density of the feedstock to facilitate its transportation and storage. In addition, densification has the benefit of improving particle uniformity, as well as improving handling and flowability characteristics (Tumuluru et al., 2011).

Lignin is considered as a natural binding agent and is an important component of biomass for densification. In general, grasses, which have lower lignin content, are difficult to pelletize, do not produce pellets with good density and durability, and require higher pelleting energy. In addition, many researchers have reported that grasses do not produce good quality pellets due to their low lignin and needle shape particles (Stelte et al., 2011). Woody biomass, given that is a cleanly harvested material, generally have higher lignin and lower ash content compared to straws or grasses. Therefore, the blending of straws or grasses with woody biomass can augment pellet properties and reduce the pelleting energy requirements (Tumuluru et al., 2012). The present research indicates that switchgrass has about 21% lignin content whereas the pine residues contain 37.5 and 35.9%, for 2PN and 6PN, respectively (**Table 2**). We hypothesize that blending switchgrass with pine residues can bring significant improvement in terms of lignin and particle size distribution and can help to produce good quality pellets with lower energy consumption. Thus, blending not only helps to achieve a specific chemical composition of feedstock but also has the potential to improve its mechanical preprocessing and flow characteristics.

Researchers has shown that the particle size distribution of milled material has a significant impact on the quality of the resulting pellets. For example, Tumuluru et al. (2011) reported a negative correlation between the material particle size and pellet density and durability. This phenomenon is caused by greater surface area in smaller particles resulting in reduced void space after densification. In addition, different densification techniques perform better with different particle sizes. For example, smaller particles are better suited for a pellet mill because the consolidation process is dependent on the inter-particle contact area, while larger particles are better suited for a briquette press as binding in this system is achieved by interlocking of particles (Tumuluru et al., 2012). It is critical to manage the particle size to meet the densification equipment requirements in order to produce high-quality densified materials while reducing energy consumption. Blending woody and herbaceous biomass can be utilized to alter the particle size distribution and produce feedstocks suitable for different densification systems.

According to Payne (1978), adding moisture through steam conditioning while pelleting of medium to fine-ground materials is suitable due to the high surface area involved, and this can result in greater gelatinization of starch and increased binding. The same authors also reported that a specific ratio of fines to medium sized particles can improve pelleting efficiency by reducing energy requirements during pelleting. In the present study, blending of pine residues and switchgrass at different ratios has resulted in different particle size distributions (**Table 4**). This change in the particle size can have significant impacts on pellet quality and energy consumption of pelleting process.

#### The Effect of Blended Feedstocks on Bio-Oil Yield and Quality

Results from our small-scale pyrolysis (Py-GCMS) experiments as well as the bench-scale fluidized-bed pyrolysis reactor experiments demonstrate that the vapor-phase pyrolysis product distribution and pyrolysis products yields behave as a simple linear combination of the two feedstocks. The trend of the pyrolysis behavior of blended samples being proportional to the pure feedstocks in which it is composed has been observed by several other researchers using several other types of lignocellulosic biomass, and is confirmed in this study. Thus, this behavior appears to be ubiquitous across blends created with different lignocellulosic species. In effect, predicting the pyrolysis behavior of blended feedstocks is straight-forward, which should make the utilization of blending as a feedstock preprocessing strategy more direct. In addition, we demonstrate that biomass ash content is not strongly correlated with pyrolysis product yields, while some specific inorganics show strong correlations. Therefore, the concentration of alkali and alkaline earth metals should be considered as more important for determining the quality of a feedstock.

This work demonstrates that blending switchgrass and pine residues is a promising method for producing consistent and high-quality feedstocks for thermochemical conversion. However, several aspects must be considered when defining the optimum feedstock blend ratios. These aspects include the availability and cost of the individual pure feedstocks, the final cost of the processed and blended feedstock, and the potential quality and value of the produced bio-oil. Future research is required to further assess the technical and economic aspects regarding harvesting, transportation, and processing of blended feedstocks.

# CONCLUSIONS

Feedstocks for thermochemical conversion were prepared by blending low-ash pine residues and relatively higherash switchgrass biomass in several different ratios. Chemical and physical properties and thermochemical reactivity were investigated. Results indicate that physical properties such as bulk and tap densities and particle size distribution are proportional to the ratio of switchgrass and pine residues in the blended sample. The vapor-phase pyrolysis products of pine residues are higher in CO<sup>2</sup> and phenolic compounds, while switchgrass produces more acetic acid, dihydrobenofuran, and levoglucosan. Lab-scale fluidized-bed pyrolysis experiments demonstrated that pyrolysis of pine residues resulted in bio-oils with greater carbon content and reduced oxygen content, and the bio-oil organic yield showed a strong correlation with the concentration of alkali and alkaline earth metals in the feedstock. Similar to the physical properties, the distribution of the vapor-phase pyrolysis products and pyrolysis products yields are proportional to the ratio of switchgrass and pine residues in the blended samples. These results indicate that blending different sources of biomass is a promising strategy to produce a consistent feedstock for thermochemical conversion.

# AUTHOR CONTRIBUTIONS

CE performed small-scale pyrolysis (Py-GCMS) experiments and data analysis, assisted in feedstock processing, and prepared and organized the manuscript. ER performed the fluidizedbed pyrolysis reactor experiments and bio-oil characterization. ER, SP, and SK assisted with pyrolysis data interpretation. CH conducted the chemical analysis of the biomass samples. NA and NL organized feedstock harvesting and processing and assisted with manuscript editing. OF and SA performed the physical characterization of the feedstocks. JT contributed discussion toward blending, physical properties, and densification of blended feedstocks. TR is the project director, and SK and NL are principal investigators, and assisted in project planning and experimental design. All authors jointly assisted in manuscript writing and editing.

#### ACKNOWLEDGMENTS

This work was completed under the DOE-funded Logistics for Enhanced-Attribute Feedstocks (LEAF) Project, and

#### REFERENCES


Figura, L. O., and Teixeira, A. A. (2007). Food Physics. Physical Properties – Measurement and Applications. Berlin; Heidelberg: Springer-Verlag.


this material is based upon work supported by the Department of Energy, Office of Energy Efficiency and Renewable Energy (EERE), under Award Number DE-EE0006639. We also acknowledge support from the USDA National Institute of Food and agriculture, Hatch Project 1012359.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00079/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Edmunds, Reyes Molina, André, Hamilton, Park, Fasina, Adhikari, Kelley, Tumuluru, Rials and Labbé. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Hot Water Extraction Improves the Characteristics of Willow and Sugar Maple Biomass With Different Amount of Bark

Obste Therasme<sup>1</sup> \*, Timothy A. Volk <sup>1</sup> , Antonio M. Cabrera<sup>2</sup> , Mark H. Eisenbies <sup>1</sup> and Thomas E. Amidon<sup>3</sup>

<sup>1</sup> Department of Forest and Natural Resources Management, SUNY College of Environmental Science and Forestry, Syracuse, NY, United States, <sup>2</sup> Vicerrectoría de Investigación y Postgrado - Centro del Secano, Universidad Católica del Maule, Talca, Chile, <sup>3</sup> Department of Paper and Bioprocess Engineering, SUNY College of Environmental Science and Forestry, Syracuse, NY, United States

Shrub willows are being developed as a short rotation woody crop (SRWC) that can grow on marginal agricultural land. Willow has a high net energy ratio (energy produced/ fossil fuel energy consumed), low greenhouse gas footprint and high carbohydrate production potential. Willow biomass can be combined with forest biomass, but willow often has a higher proportion of bark that creates challenges because it increases the ash content and decreases the melting point. Hot water extraction is a pretreatment that has been shown to improve the quality of chipped material while producing a marketable stream of byproducts. This study evaluated how the amount of bark (0, 33, 66, and 100%) on three willow cultivars and sugar maple impact the output of hot water extraction in terms of mass removal and extract composition, as well as its influence on the heating value, ash and elemental content. The hot water extraction process resulted in ash content up to 50% for sugar maple and willow, but there was variation among the willow varieties. The heating value after hot water extraction was about 5% higher because of the removal of mostly hemicelluloses, which have relatively low heating value. HWE led to significant reductions of calcium, potassium, magnesium and sulfur contents. The hot water extraction provides a fermentable sugar stream and other coproducts after multiple separation and treatment steps, and improves the characteristics of willow and sugar maple biomass for combined heat and power. This paper demonstrates how biomass with higher bark content can generate a useable sugar stream while improving the quality of the biomass for combined heat and power by managing its ash content while simultaneously producing other valuable products.

Keywords: SRWC, willow, hot water extraction, heating value, ash, alkali metals, biomass

# INTRODUCTION

Biomass feedstock in the United States has the potential to displace and supplement a significant portion of the present petroleum consumption in the form of biofuels, bioenergy and bioproducts (USDOE, 2011). However, the variability in the quality of these materials haslimited their adoption. In order for them to increase market share, effective and efficient pretreatment methods for

#### Edited by:

Timothy G. Rials, University of Tennessee, Knoxville, United States

#### Reviewed by:

Ming Zhai, Harbin Institute of Technology, China Richard Arthur Venditti, North Carolina State University, United States

> \*Correspondence: Obste Therasme otherasm@syr.edu

#### Specialty section:

This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research

Received: 11 April 2018 Accepted: 24 August 2018 Published: 11 September 2018

#### Citation:

Therasme O, Volk TA, Cabrera AM, Eisenbies MH and Amidon TE (2018) Hot Water Extraction Improves the Characteristics of Willow and Sugar Maple Biomass With Different Amount of Bark. Front. Energy Res. 6:93. doi: 10.3389/fenrg.2018.00093 lignocellulosic biomass need to be developed (Chaturvedi and Verma, 2013). The specific role of pretreatment steps varies among feedstocks and conversion pathways. Broadly, pretreatments are seen as a means to reduce variability in feedstock characteristics in order to deliver a uniform feedstock to end users. Pretreatments can also improve the efficiency of conversion by altering the physical and chemical attributes of woody biomass (Liu, 2015). Activities during pretreatment range from changing the particle size to the modification of the structural component of the biomass feedstock.

Hot Water Extraction (HWE) has been studied and proposed as a pretreatment step into a multi-product biorefinery. This process consists of cooking biomass materials, such as wood chips, at high temperature and pressure in water (Amidon and Liu, 2009). The treated lignocellulosic biomass, or extracted wood chip, can be valued as an improved-quality raw material for fuels, manufacture of wood products or wood derivatives (Amidon et al., 2011). Multiple separation steps and treatments of the extract solution yield wood-based chemicals and materials such as acetic acid, methanol, lignin and fermentable sugars. The effect of HWE on various biomass feedstocks has been extensively studied during the last decade (Amidon and Liu, 2009; Amidon et al., 2011; Duarte et al., 2011; Pu et al., 2011; Lu et al., 2012; Wang et al., 2012; Ma et al., 2013; Mante et al., 2014; Corbett et al., 2015). By removing some biomass components, HWE alters the chemical components of biomass and the surface morphology of biomass. A recent study (Ma et al., 2013) on HWE of bamboo has shown that oxygen/carbon ratio decreases respectively from 0.43 to 0.34 on the exterior surface and to 0.37 on the interior surface. Hemicellulose, which has a higher oxygen/carbon ratio than cellulose and lignin, is dissolved into the water during HWE process. As a pretreatment, HWE increases the porosity of the cell wall (Duarte et al., 2011), which enhances sugar recovery and enzymatic hydrolysis (Wang et al., 2012).

Shrub willow (Salix sp.) is a potential feedstock for the bioenergy and bioproducts industry. It has a high net energy ratio, high yield, rapid growth rate, the ability to resprout after multiple harvests, and is compositionally similar to other hardwood species (USDOE, 2011; Volk et al., 2014). Willow is generally harvested every 3–4 years for up to 7 rotations without replanting, and it is suitable to grow on marginal land across the Northeast, Midwest and parts of the Southeastern United States; it has been grown for decades in Europe. The net energy ratio, which is the ratio of the total energy content of biomass at the gate of the farm by the total fossil fuels consumed, was estimated to be 55:1 for a typical agricultural production of willow biomass (Heller et al., 2003). The first rotation yields of top willow cultivars range from 9.4 to 13.6 odt/ha.yr across multiple sites (Volk et al., 2011). Sleight et al. (2016) demonstrates that plots with low yields during the first rotation had a significant yield increase during the second rotation. Eisenbies (Eisenbies et al., 2015) reports the characteristics of willow biomass chips harvested at two sites are similar to other woody biomass; the ash content ranges from 0.8 to 3.5%; the moisture content ranges from 37 to 51% and the higher heating value ranges from 18.3 to 19.1 MJ/kg. The stem diameters of three willow cultivars, which were 4 years old on a 5 year root system, were less than 45 mm (Eisenbies et al., 2015). Bark content of willow crop varies with stem diameters and site conditions. For instance, as diameter of willow stems increases bark content decreases. Mean whole stem bark percentage of willow in two sites was 12.1% and 18.9% (Eich et al., 2015).

Willow is typically harvested using single-pass cut-and-chip forage harvesters that have been developed to harvest shortrotation crop. The machine can cut willow stems of up to 120 mm in diameter and produce 10–45 mm long chips (Eisenbies et al., 2014). Due to the small diameter of willow crop at harvest, the prospect of debarking this feedstock is challenging, even infeasible with current technology, and would increase the overall processing time and cost of this feedstock (Magagnotti et al., 2011). Thus, bioenergy and bioproducts industries will have access to willow as bark-on feedstocks. To date most of research on HWE of wood chips used debarked wood (Duarte et al., 2011; Lu et al., 2012; Mante et al., 2014). Forest residues with bark, often called dirty chips, are considerably cheaper than debarked chips and if the HWE process can be used to improve their quality and/or extend their storage, these materials could be effectively utilized at an acceptable cost. A further option with feedstocks would be to blend different sources of material to ultimately provide a year round supply with consistent quality and minimal cost variation (Williams et al., 2016). Examining the performance of HWE on bark-on woodchips, as well as woodchips with varying amounts of bark removed, is an important step in understanding how this feedstock and pretreatment could be incorporated into the feedstock supply chain.

The objective of this study was to evaluate the effect of hot water extraction on the current bark-on willow biomass in term of mass removal and composition of the extract, as well as its influence on the heating value and ash content of three willow cultivars using sugar maple as an operational control. A second objective was to compare the degree to which bark removal would impact changes in biomass quality (ash content, heating value, elemental content) when HWE is applied.

#### MATERIALS AND EXPERIMENTAL METHODS

#### Source of Material

Shrub willow used for this study was hand harvested as whole stems in SUNY ESF's genetic field station in Tully, NY. Sugar maple (Acer saccharum Marshall) stems were obtained from SUNY ESF's field station in Lafayette, NY. Three cultivars of willow (SX64 - Salix sachalinensis F. Schmidt, SV1 - Salix x dasyclados Wimm. and Sherburne - Salix miyabeana Seemen) were used in addition with sugar maple. Both shrub willow and sugar maple are hardwood species.

Bark was removed at four levels (0, 33, 66, and 100%) to simulate the effect of bark removal. Four replicates were prepared for each experimental level. The bark of willow stems and sugar maple stems were manually peeled and the ratio of bark to wood on each of the stems was determined. The stem and the bark were chipped separately and remixed using 0, 33, 66, and 100% of the predetermined proportions. The 0% sugar maple was used as the control because this is the standard used in previous HWE studies (Duarte et al., 2011; Lu et al., 2012; Mante et al., 2014). The sample with 0% bark refers to wood with no bark and sample with 100% bark refers to wood with all its bark attached. All samples were screened in a Williams's classifier conforming to mesh size of 3/8 to 7/8 inches prior to HWE. This was done to remove fine particles that would create blockages in the vessels used for HWE. Heating value, ash content, alkali metals, silica and sulfur content were determined on wood samples prior to HWE and after HWE.

#### Hot Water Extraction and Analysis

Hot water extraction was performed in a 4.7 L M/K digester using 500 g oven dried chips and 4:1 water to wood ratio at 160◦C for 120 min. At the completion of the extraction, the liquor was collected and the extracted chips were washed twice. For the washing process, the equivalent volume of the liquor was added to the digester and cooked at 80◦C for 15 min which fully removes dissolved substances from the mass of chips (Wang et al., 2017). The mass removal was reported as the ratio of the mass removed by HWE by the initial weight of oven dried wood chips. The hydrolyzate was subjected to chemical composition analysis as described in Kiemle et al. (2003) AND Mittal et al. (2009a). A sample of the hydrolyzate, for the maple and SV1 samples, was mixed with 96% sulfuric acid to obtain a 4wt.% sulfuric acid solution. The solution was autoclaved for 45 min at 121◦C. Then, 0.1 ml of 72% deuterated sulfuric acid was added to 1 g of filtrated sample. After the hydrolysis, sugars (xylose, glucose, arabinose, rhamnose and galactose) as well as acetic acid and furfural concentrations in the sample were determined by <sup>1</sup>H NMR (Mittal et al., 2009a). The higher heating value was determined on samples prior and following HWE in accordance with ASTM method D5865-13: Standard test method for gross calorific value of coal and coke by using a Parr 6200 Oxygen bomb calorimeter (ASTM, 2013). The ash content was determined by combustion in a muffle furnace in accordance with TAPPI method T 211 om-02: Ash in wood, pulp, paper and paperboard: combustion at 525◦C (TAPPI<sup>1</sup> ) on samples prior to HWE and following HWE. An acid (HCl) solution was prepared with the ash as described by Wild et al. (1979). Potassium, sodium, magnesium, calcium, silicon and sulfur were measured by Inductively Coupled Plasma Atomic Emission Spectroscopy technique (Schulte and Huang, 1985) on samples prior to HWE and following HWE.

A repeated analysis was performed using Statistical Analysis Software (SAS version 9.4) using the mixed procedure with an alpha level of 0.05.There were four varieties of biomass (three willow cultivars and one sugar maple), four bark levels (0, 33, 66, and 100%) and two periods (prior to HWE, following HWE). (Model Parameter = Variety Bark Period Variety∗Bark Variety∗Period Bark∗Period Variety∗Bark∗Period).

#### RESULTS AND DISCUSSIONS

#### Mass Removed by Hot Water Extraction (HWE) and Composition of the Hydrolyzate

The majority of HWE research to date has been conducted on woody biomass that has been debarked. In this study, the mass removal of 0% bark samples was not significantly different among the sugar maple and all three willow cultivars, and ranged from 20.0 to 21.7%. These mass removals were similar to other studies on HWE of hardwood species; HWE of debarked maple chips, under the same conditions described in this study, resulted in 21.1% mass removal (Duarte et al., 2011). Similarly, HWE of debarked aspen woodchips at 160◦C for 210 min with a water to wood ratio of 4:1 produced a mass removal of 21.4% (Lu et al., 2012). HWE conducted for only 1 h, under approximately the same conditions, yields lower mass removal than what is found in this present study. Guan et al. found mass removal of 15.7% when HWE is performed on debarked hybrid poplar for 1 h at 170◦C (Guan et al., 2018). HWE of sugar maple removed 14.2% of the initial weight, when operated at 160◦C for 1 h (Mittal et al., 2009a). Mass removal is impacted by many factors, including—but not limited to—temperature, water-towood ratio and duration of extraction (Lu et al., 2012).

As the bark content of sugar maple increased, mass removal increased (**Table 1**). However, the mass removal of each willow cultivar did not match sugar maple with increasing bark content. Variation of bark content did not change the mass removal of Sherbune (p > 0.19) as it did for SX64 and SV1. Bark-on SV1 differed significantly from both SX64 and sugar maple. Differences of mass removal were 3.22% between SX64 and SV1 (p < 0.0001), and 3.52% between sugar maple and SV1 (p < 0.0001). These results suggest the impact of bark content on the mass removed by HWE varies among willow cultivars and sugar maple. The reason for the varied responses among samples is not entirely clear but may be partially related to the initial ash content and ash removal (see section Ash Content, below), the bark to wood ratio, and the variability in relative amounts of hemicellulose, cellulose, lignin and extractives. The relative proportion of bark-to-wood decreases as the diameter of the willow stems increases (Eich et al., 2015). Bark contains higher ash and lignin (Mészáros et al., 2004), and higher extractives than wood. Chemical composition analysis of 25 willow cultivars shows that hemicellulose and cellulose are greater in debarked wood than whole stem biomass (100% bark attached) (Serapiglia et al., 2009).

The mechanism of HWE for hardwood species has been described (Mittal et al., 2009a,b; Liu, 2010). HWE uses the hydrolytic property of water and in situ-generated organic acid to activate the depolymerization reaction. The depolymerization of carbohydrates, mainly hemicellulose, leads to the formation of oligomers and monomers of xylose, mannose, galactose and glucose, in addition to acetic acid, formic acid and other degradation products. Glucose may in fact come either from the decomposition of cellulose or hemicellulose. However, xylose is derived from hemicellulose depolymerization. Furfural is the result of the degradation of xylose at high temperature and pressure. Also, the degradation of hexoses leads to the formation of hydroxymethyl furfural (HMF), which could further converted into levulinic acid and formic acid (Palmqvist and Hahn-Hägerdal, 2000).

Glucose, xylose, mannose, arabinose, rhamnose, galactose including both oligomers and monomers—in addition to acetate and furfural concentrations were determined in the hydrolyzate (**Table 2**). Only the concentration for sugar maple and SV1

<sup>1</sup>TAPPI Ash in wood, pulp, paper and paperboard: combustion at 525◦C.


were reported due to technical issues with the analysis of the hydrolyzate of SX64 and Sherburne. The total concentration, which is the sum of the concentration of these components, for sugar maple and SV1 was not significantly different (p > 0.47) at any of the bark percentages. Concurrently, the total concentration of sugars decreased steadily and at almost the same rate. The total concentration in the extract of sugar maple ranged from 45.29 g/L (0% bark) to 32.60 g/L (100% bark). Similarly, it ranged from 44.72 g/L (0% bark) to 28.97 g/L (100% bark) for SV1. As for other parameters reported in this study these responses may be different for other willow cultivars, particularly since previous studies have shown some degree of variation in cellulose, hemicellulose and lignin content across willow cultivars (Serapiglia et al., 2013).

Xylose was the predominant component in the hydrolyzate after HWE of sugar maple and willow woodchips. Total xylose concentration—in the form of monomer and oligomer—in the hydrolyzate after HWE ranged from 14.36 to 27.51 g/L for sugar maple and 15.94 to 23.93 g/L for SV1. Guan et al. (2018) found total xylose concentration of 16.8 g/L from HWE of hybrid poplar (hardwood), mostly present in the form of oligosaccharides, and 4.1 g/L from HWE of southern pine (softwood). It has been documented that glucomannan is the predominant unit of hemicellulose in softwood while Glucuronoxylan is the major one in hardwood (Liu, 2010). There was a steady decline in the xylose concentration in the extract as the amount of bark in the sample increases for both maple and SV1. There was no significant difference in the xylose concentration between maple and SV1 at any of the bark percentages (P > 0.16). The ratio of xylose to glucose was high across all the samples, ranging from 3.9 (SV1 0% bark) to 12.4 (Maple 0% bark). This confirms that hemicellulose is the main component that is removed during HWE of sugar maple and willow across the range of bark samples.

Acetic acid, furfural, and other carbohydrate and lignin degradation products can inhibit fermentation, but have value as byproducts. Bark content did not affect the formation of furfural during HWE of sugar maple and SV1 samples (p > 0.48). The mean of furfural concentration in the hydrolyzate after HWE for sugar maple and SV1 were respectively 1.09 and 0.96 g/L. Thomsen reported furfural concentration ranging from 0.03 to 1.2 g/L during hydrothermal treatment of wheat straw (Thomsen et al., 2009). The relative cellular redox activity of Candida shehatae, a xylose fermenting yeast strain, was reduced to 65% when furfural concentration was above 2 g/L and 55% when acetic acid concentration was 4 g/L (Zhao et al., 2005). Beside their inhibition capability on fermentation, furfural and acetic acid hold valuable promise for the development of a biorefinery that includes the process of HWE. Acetic acid has been used for centuries by society for food preservation and is used in the manufacture of a range of products. Furfural is the precursor of many valuable chemicals in the category of furyl, furfuryl, furoyl, furfuryldiene as well as dimethyl furan and ethyl levulinate (Yan et al., 2014). Furfuryl alcohol, a product of catalytic hydrogenation of furfural, is being used in the production of foundry resin and furan fiber reinforced plastics (Schneider and Phillips; Yan et al., 2014). Recently, a new fuel blend that contains furfuryl alcohol and ionic liquid has been developed to be used in missiles and satellite launch vehicles (Bhosale et al., 2016).

### Ash Content

Before HWE, the ash content of sugar maple woodchips was significantly less than all three willow cultivars for all bark contents (P < 0.001). There were general differences among the willow cultivars at each of the bark percentages with Sherburne > SX64 > SV1. However, there were no statistical differences with three pairs of samples among all willow and bark percentage combinations; 33% bark for SV1 and Sherburne (p = 0.097), 66% bark for SV1 and Sherburne (p = 0.798), and 100% bark SV1 and SX64 (p = 0.22). At 0% bark level, SV1 had significantly lower ash content than Sherburne and SX64. Sherburne had significantly higher ash content than SV1 and SX64 at 100% bark level (**Figure 1**). The ash content values for Sherburne and SX64 with 100% bark were very similar to the values reported by Serapiglia et al. (2013), despite different sampling and analytical techniques being used. However, the ash values for SV1 in this study were about three times higher than reported values for samples from the Tully site, but the ash content in this study was about 40% lower than samples from another site in Belleville in the Serapiglia et al. (2013) study. Ash content has been shown to vary by site and willow cultivar but the values here are in the range of values reported for hand harvested plots.

After HWE, ash contents of the three willow cultivars were more uniform at low bark level. There were no significant differences between the ash content of SV1, SX64 and Sherburne samples at 0% bark level (p > 0.11). But, at 100% bark level SX64 had significantly higher ash content than Sherburne, which itself had higher ash content than SV1. Sugar maple had significantly lower ash content than all three willow cultivars across the four

#### TABLE 2 | Chemical concentrations of the hydrolyzate<sup>a</sup> .


<sup>a</sup>Both monosaccharides and oligosaccharides are included.

<sup>b</sup>Glu, Glucose; Xyl, Xylose; Man, Mannose; Ara, Arabinose; Rha, Rhamnose; Gal, Galactose; Ac, Acetate; Fur, Furfural.

<sup>c</sup>Values in parenthesis are standard deviations (n = 4), except for Arabinose measurement on sugar maple with 33% bark on (n = 2).

levels of bark except for 100% SV1, which was not significantly different (p = 0.35).

Overall, the ash content of sugar maple and willow samples decreased after HWE (**Figure 1**). The mean ash content of sugar maple samples across all bark levels decreased from 0.56% prior to HWE to 0.40% after HWE. Similarly, SV1 and SX64 had lower ash content after HWE. But, it is important to notice that ash content of 0% bark sugar maple, SV1 and SX64 was not impacted by HWE (p > 0.77). These results suggest that inorganic elements that are transferred into the water during the process of HWE originated mostly from the bark structure. The reduction in ash levels varied among the three willow cultivars. SV1 had no or small reductions in ash content until the 100% bark sample where ash was reduced from 1.49 to 0.84%. Previous studies have shown that washing alone can reduce ash content in biomass (Jenkins et al., 1996; Deng et al., 2013). The reduction of ash generation has the advantage of allowing the removal of ash less frequently, decreasing the risk of slagging and reducing the emissions problem. It was suggested that removal efficiency of ash increases as the water temperature increases (Deng et al., 2013). In addition to the reduction of ash, which is an important step for thermochemical conversion of biomass, the HWE process provides a liquid stream that contains key components to be recovered as products.

HWE had a greater effect on bark-on wood compared to barkoff wood. Ash content for bark-off sugar maple was essentially unchanged 0.24 and 0.23% for 0% bark before and after HWE respectively; conversely, for 100% bark content ash content decreased from 0.93 to 0.69%. All three willow cultivars (SV1, SX64, and Sherburne) with 100% bark-on had significant ash reductions compared to their corresponding 0% bark samples. For instance, among all three willow cultivars and prior to HWE, ash content ranged from 0.54 to 1.06% for 0% bark samples, and from 1.27 to 2.23% for 100% bark samples. Following HWE, the reduced ash content ranged from 0.57 to 0.83% on debarked willow samples, and from 0.84 to 1.66% on willow samples with 100% bark attached. As with maple, the amount of ash removed by HWE was greater for samples with higher bark content. Based on these results, the ash content for most 0% bark samples would meet the 1.0% threshold set for class A1 graded wood chips; Sherburne would meet the A1.5 class (1.5% ash content threshold) (ISO, 2014). HWE resulted in grade increase for willow. Samples with 100% bark content at best qualify for a class B chip, with ash content less than 3.0%.

The willow used in this trial was hand harvested. Due to the logistics of mechanical harvesting, ash content at commercial scales could be slightly higher. Eisenbies et al. (2015) found that ash content of over 200 willow samples that were harvested with a commercial single pass cut and chip harvester range from 0.8 to 3.5%. Inorganic materials that contaminate the feedstock during harvesting and transportation operations could be easily washed out during the HWE. Beside the variability that can be caused by willow cultivar, bark content and harvesting method, ash content depend also on site condition (Stolarski et al., 2013).

#### Heating Value

Before HWE, there were significant differences between heating values of all three willow cultivars (p < 0.003) (**Figure 2**). Across all bark content (0–100%), the heating value of SV1 was about 2–3% lower than the heating value of SX64 and Sherburne. But, heating values of SV1 and sugar maple were not significantly different (p = 0.39). There were no significant differences in HHV for 0% bark (p = 0.91) and for 100% bark (p = 0.70) for SX64 and Sherburne. The difference observed among the heating values of different willow cultivars may be due to the difference in their structural composition. Serapiglia et al. (2009) studied the composition of 25 willow cultivars and found significant differences in their relative cellulose, hemicellulose and lignin contents. On a dry mass basis, SV1 had 1.6 percent point lower lignin content than SX64 (Serapiglia et al., 2009). Lignin has higher heating value than cellulose (White, 2007; Novaes et al., 2010), which itself has higher heating value than hemicellulose.

standard deviations).

Therefore, the lower heating value of SV1 aligns with the positively correlated relationship between higher heating value and lignin content in hardwood species (Demirba¸s, 2001).

Both the amount of bark content and the specific cultivar had an impact on the higher heating value after HWE. For bark content ranging from 0 to 100% the heating values of SV1 and SX64 were not significantly different, but were significantly higher than the heating value of Sherburne. The heating value of debarked sugar maple and sugar maple with all its bark attached were significantly higher respectively than debarked SV1 (p = 0.004) and SV1 with all its bark attached (p < 0.0001).

Following HWE the heating value ranged from 19.5 to 20.4 MJ/kg across all species and different amount of bark, which is an increase compared to higher heating values prior to HWE. Heating values of sugar maple, SV1 and SX64 increase significantly after HWE across all bark levels except for 0% bark SX64, which had no significant change in HHV (p = 0.51). The means of the heating values of sugar maple across all bark treatments increases after HWE by 5.7%. Similarly, after HWE SV1 and SX64 have an increase of 4.7 and 2.7% of their heating value. The higher heating value of mixed hardwood species that were extracted using a similar process (160◦C for 2 h) showed a 2.9% increase in HHV. The increase was slightly higher, 3.6%, at the conditions of 170◦C for 2 h (Pu et al., 2011). During HWE, hemicellulose molecules in willow and sugar maple structures are being extracted and partially hydrolyzed. As it is shown in **Table 1**, 19.2–23.7% of the woodchips were transferred into the hot water. The analysis of the composition of the hydrolyzate shows that xylose, which is produced from the hydrolysis of hemicellulose, is the major component of the hydrolyzate. As a consequence, after HWE the wood had a lower proportion of hemicellulose and a higher proportion of lignin, which results in an increase of the heating value.

The effect of bark content on heating value before and after HWE varies between sugar maple and willow cultivars. The heating values for sugar maple and SV1 before HWE are similar across all the bark treatments. Following HWE the heating values for sugar maple and SV1 are higher for all bark treatments but the heating values do not vary across the bark treatments for either sugar maple or SV1. In contrast, the heating value of SX64 decreases with increasing bark content before HWE and increases with bark content after HWE. Studies of the effect of bark ratio on heating value of pellets made from five evergreen Mediterranean hardwoods species found that the heating value of bark was statistically lower in three species and was higher in one than the heating value of their debarked wood (Barmpoutis et al., 2015).

The increase of the heating value of woody biomass, including short rotation willow, can add value to the biomass feedstock for biopower and heating applications. The moisture content of the biomass will impact the lower heating value of the chips being used. Practically speaking, net (lower) heating value describes better the amount of energy available to be used. When the

TABLE 3 | Calcium, sodium, sulfur and silicon contents of sugar maple and three willow cultivars with bark levels ranging from 0 to 100% before and after HWE.


\*Values in parenthesis are standard deviations.

moisture content of a fuel increases, the lower heating value decreases. In general, the moisture content of chips immediately following HWE is higher than the moisture of the non-extracted chips. However, further investigation is needed to understand the drying process of extracted woodchips. HWE woodchips contain less hemicellulose and have higher porosity (Duarte et al., 2011), which suggests a propensity for more rapid drying and a lower capacity to retain water.

Potassium Content Prior to HWE, the potassium content in all three willow cultivars increased significantly with increasing bark content (p < 0.0001) (**Figure 3**). Potassium content at 100% bark level was 1.6, 1.5, and 1.3 times higher than the potassium content at 0% bark level for SV1, SX64 and Sherburne respectively. The potassium content was significantly different among all four biomass categories across all bark levels except for at the 33% bark treatment for Sherburne and SX64 (p = 0.31), and 66% bark treatment for Sherburne and SX64 (p = 0.87). These results mirror the ash content data, SV1 with 0% bark had lower potassium content (0.68 ± 0.03 mg/g) than SX64 (0.91 ± 0.02 mg/g), which had lower potassium content than Sherburne (1.14 ± 0.04 mg/g). Observed values of potassium content for SV1, SX64 and Sherburne with all their bark attached are within the range of values reported in a previous study (Tharakan et al., 2003).

After HWE, potassium content decreased to a uniform level across the percent bark treatments independently of their content before extraction (**Figure 3**). Among all bark content, following the extraction, the potassium content of SV1 varied from 0.20 to 0.29 mg/g while it varied from 0.30 to 0.37 mg/g in SX64 and 0.38 to 0.47 mg/g in Sherburne. HWE removed 70% of potassium in willow cultivars and 83% of potassium in sugar maple. The considerable reduction in potassium content by HWE is the result of potassium being in true solution and ionizable form in many plant tissues (Morris and Sayre, 1935). Potassium content

after HWE of sugar maple and three willow cultivars (Ya, dependent variable after HWE; Yb, dependent variable before HWE; and X, percentage of ash).


TABLE 4 | Pearson correlation coefficients and P-values for ash, heating value and various elements.

\*Values in parentheses are P-values. \*\*Data points prior to HWE and following HWE are combined.

along with the concentrations of sodium, magnesium, calcium and silicon determine the melting behavior of ash; therefore, potassium removal is a highly desirable benefit that reduces the risk of slagging and hard deposit that accumulates in furnaces and boilers (Obernberger and Thek, 2004). Although potassium contents were uniform within a given cultivar after HWE, there were significant differences in potassium content among willow cultivars and sugar maple across all bark levels (p < 0.03).

#### Magnesium Content

Prior to HWE the magnesium contents among all three willow cultivars and sugar maple samples across all bark content were within the range of 0.20–0.28 mg/g (**Figure 4**). Magnesium content was not significantly different with increasing bark content for maple but changed for each of the three willow cultivars. For Sherburne, the magnesium concentration was the same for 0, 33, and 66% bark treatments but was significantly higher at the 100% bark treatment. For SX64, the 0% bark treatment had a lower Magnesium concentration compared to the other three treatments. SV1 followed a slightly different pattern with slightly lower magnesium concentration values for the 0 and 100% bark treatments compared to the 33 and 66% bark treatments. For the 100% bark treatment, Sherburne had a significantly higher magnesium concentration than the other two willow cultivars (SX64 and SV1) and maple. Tharakan et al. (2003) reported magnesium content of 0.23 mg g−<sup>1</sup> and 0.24 mg g −1 respectively for SV1 and SX64 cultivars with all their bark attached.

Similar to potassium content, the magnesium content of the three willow cultivars and sugar maple decreased after HWE to a stable value for bark treatments ranging from 0 to 66% (**Figure 4**). The average reduction of magnesium content was about 65% for sugar maple, 60% for SV1, and 55% for SX64 and 40% for Sherburne, for bark treatment ranging from 0 to 100%. Magnesium in the form of oxide plays a key role in the inhibition of alkali (K, Na) volatilization while burning biomass (Miles et al., 1996). Therefore, a reduction of magnesium would lead to an increase in the volatized alkali, but this effect would not be a concern as there was a greater decrease of alkali in the form of potassium.

#### Calcium Content

Calcium content of all three willow cultivars and sugar maple followed exactly the same trend as ash content. Calcium content of all three willow cultivars and sugar maple increased with increasing ash content (**Table 3**). Prior to HWE, sugar maple and willow samples with all their bark attached had 2.2–2.4 times more calcium than their corresponding samples with 0% bark. Sherburne with 100% bark treatment had the highest calcium content (7.932 ± 0.241 mg/g) and SV1 with 0% bark treatment had the lowest calcium (1.93 ± 0.04 mg/g), across all bark levels and willow cultivars (**Table 3**). Calcium concentrations reported in the literature (Tharakan et al., 2003) for SV1 and SX64 willow cultivars were slightly higher than those found in this present study.

The effect of HWE on calcium content varied among the three willow cultivars and bark treatment. For 100% bark treatment, following HWE, calcium content in SV1 and Sherburne decrease by more than half of their calcium content prior to HWE while SX65 had no significant change. Additionally, HWE removed half of the initial calcium content in sugar maple with 0% bark treatment. No significant removal of calcium was observed for all three willow cultivars with 0% bark treatment (P > 0.15).

#### Sulfur, Sodium and Silicon Contents

As with other elements, sulfur content varied among cultivars, amount of bark and HWE treatment, following the same trend as ash (**Figure 5**). Prior to HWE, willow and sugar maple samples had significantly higher sulfur content than their corresponding samples with 0% bark treatment (**Table 3**). Prior to HWE, 100% bark treatment had 1.5 times more calcium than 0% bark treatment for sugar maple. But, it ranged from 1.4 to 2 times for the three willow cultivars. After HWE, there were no significant differences between 0% bark Sherburne and 100% bark Sherburne (P = 0.93). But, 100% bark treatment samples were 1.4–1.6 times higher than 0% bark treatment samples for SX64 and SV1, following HWE. Additionally, sulfur removal by HWE was the highest for 100% bark treatment SV1 and Sherburne.

Unlike potassium, there were few differences in both sodium and silicon contents among willow cultivars and sugar maple samples, as well as different bark level and HWE treatment. Concentrations of sodium and silicon were the lowest for all three willow cultivars and sugar maple samples (**Table 3**), among all elements that were analyzed in this study.

### Correlations

Calcium, sulfur, potassium and magnesium contents in willow and sugar maple had strong positive correlations with ash content before and after HWE (**Table 4**, **Figure 5**). Calcium, potassium and magnesium were strongly correlated to each other. There was no evidence of linear correlation among silica with ash across all the samples. Silica content in maple and willow is known to be relatively low compared to other elements, but soil contamination during harvesting can cause spikes (Tharakan et al., 2003). There was no evidence of significant linear correlation between ash and heating value for sugar maple (p = 0.13). However, prior to HWE, ash, potassium and magnesium had a negative correlation with heating value for SX64 and Sherburne cultivars. Ash content had a stronger correlation with potassium and magnesium in samples with all their bark attached than samples with 0% bark. As observed, ash was the lowest in samples with no bark.

#### CONCLUSION

The degree to which bark content impacts mass removal, sugars concentration in hot water extracts, and ash content, varies among willow cultivars and sugar maple. As bark content increased, the mass removal increased for sugar maple, but

#### REFERENCES


decreased for SV1 and SX64 at 0 to 66% bark treatment level. However, with increasing bark content, the total sugars concentration in the extract decreased, and the ash content increased for willow (hardwood) and sugar maple (hardwood). Hot water extraction improves the properties of the biomass as fuel source for bioenergy, and provides an extract that is rich in C5/C6 sugars. The residual material from this process still has the physical characteristics of a wood chip, but has chemical properties that make it more attractive than the original woodchip. The ash content of the extracted woodchip is lower and its energy density is greater than the original woodchip. The response among two key elements of ash (K, Mg) was that they were reduced to a consistent level following HWE across cultivars and bark treatments. The HWE process has been tested on debarked wood in the past; this research demonstrates that woodchips with bark can be used effectively in this process and that extracted woodchips from bark-on wood have significantly lower ash content and higher heating value. This will create new opportunities to make use of dirty wood, which has a significantly lower cost than debarked wood, while still yielding valuable products and a residual woodchip with improved properties. In the case of industries such as wood pellet manufacture, this would provide a significant cost savings and increase their profit margins. This work also demonstrates that willow biomass can be improved by this process and with improved characteristics that make them appealing for the production of wood pellets in a growing market. This will create new market outlets for willow biomass crops that were previously not available because of its high ash content.

#### AUTHOR CONTRIBUTIONS

OT drafted the manuscript. AC performed a number of experiments and TA granted laboratory access for laboratory experiments. OT and ME performed the data analysis. TV, ME, OT, and TA revised the manuscript.

## FUNDING

Funding to complete this research was provided by US Department of Energy Bioenergy Technologies Office under award number DE-EE0002992, the New York State Energy Research and Development Authority (NYSERDA) Award 30713, and USDA National Institute of Food and Agriculture under award 20161000825635.


during the autohydrolysis of hardwoods. Bioresour. Technol. 100, 6398–6406. doi: 10.1016/j.biortech.2009.06.107


acid on Saccharomyces cerevisiae and Candida shehatae. Biotechnol. Lett. 27, 207–211. doi: 10.1007/s10529-004-7884-3

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Therasme, Volk, Cabrera, Eisenbies and Amidon. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Techno-Economic Assessment of a Chopped Feedstock Logistics Supply Chain for Corn Stover

Lynn M. Wendt <sup>1</sup> \*, William A. Smith<sup>1</sup> , Damon S. Hartley <sup>1</sup> , Daniel S. Wendt <sup>1</sup> , Jeffrey A. Ross <sup>2</sup> , Danielle M. Sexton<sup>2</sup> , John C. Lukas <sup>2</sup> , Quang A. Nguyen<sup>1</sup> , J. Austin Murphy <sup>1</sup> and Kevin L. Kenney <sup>1</sup>

*1 Idaho National Laboratory, Idaho Falls, ID, United States, <sup>2</sup> Harris Group, Seattle, WA, United States*

Storing corn stover in wet, anaerobic conditions is an active management approach to reduce the risk of significant aerobic degradation and catastrophic loss due to fire. An estimated 50% of the corn stover available in the U.S. is too wet at the time of harvest to be stored safely in baled formats and is compatible with wet, anaerobic storage through ensiling. A logistics system based on field-chopping and particle size reduction early in the supply chain removes the dependency on field-drying of corn stover prior to baling, allows for an expanded harvest window, results in diminished size reduction requirements at the biorefinery, and is compatible with ensiling as a storage approach. The unit operations were defined for this chopped logistics system, which included field chopping, bulk transportation to a biorefinery site, on-site preprocessing to meet biorefinery size and ash specifications, industrial-scale storage through ensiling, and delivery of corn stover at a rate of 2,000 tonnes per day for ∼50% of the year. The chopped system was compared to the conventional bale system for 30% moisture (wet basis) corn stover, a likely delivered moisture content for baled corn stover harvested wet. Techno-economic analysis showed that the chopped logistics system is cost competitive, costing only 10% more than the baled logistics system, meanwhile reducing the energy consumption by 48% and greenhouse gas release by 60%. In summary, a chopped logistics system utilizing on-site preprocessing and storage at a biorefinery gate is an economically viable approach to provide a stable source of corn stover for use when dry bales are not available, meanwhile reducing the risk of loss in long-term storage.

Keywords: corn stover, forage chopping, ensiling, techno-economic analysis, sustainability

# INTRODUCTION

Agricultural residues, such as corn stover, could potentially supply >180 million tonnes of biomass feedstock for bioenergy conversion by 2040 (Langholtz et al., 2016), resulting in the production of nearly 16 billion U.S. gallons of liquid transportation fuel for the United States based on recent yield estimates (Humbird et al., 2011). As with all agricultural products, seasonal harvest necessitates long-term storage in order to provide the emerging bioenergy industry with a continuous feedstock supply throughout the year. Presently, herbaceous feedstock supply logistics operations and associated models are centered around dry bales, with long-term storage existing as a field-side operation, in satellite storage facilities, or at a centralized storage facility at the biorefinery gate

#### Edited by:

*Timothy G. Rials, University of Tennessee, Knoxville, United States*

#### Reviewed by:

*Xifei Li, Xi'an University of Technology, China Burton C. English, University of Tennessee, Knoxville, United States*

> \*Correspondence: *Lynn M. Wendt lynn.wendt@inl.gov*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *11 May 2018* Accepted: *21 August 2018* Published: *18 September 2018*

#### Citation:

*Wendt LM, Smith WA, Hartley DS, Wendt DS, Ross JA, Sexton DM, Lukas JC, Nguyen QA, Murphy JA and Kenney KL (2018) Techno-Economic Assessment of a Chopped Feedstock Logistics Supply Chain for Corn Stover. Front. Energy Res. 6:90. doi: 10.3389/fenrg.2018.00090* (Hess et al., 2009; Shah et al., 2017; Zandi Atashbar et al., 2017). The challenge with agricultural residues is that harvest conditions are optimized for the primary product, and thus there is less flexibility to control properties such as moisture content. In the case of corn stover, one of the primary agricultural residues available for bioenergy use, harvesting occurs at optimal moisture contents for the corn grain, generally <25% moisture (wet basis, wb), which corresponds to stover moisture contents about twice as wet as the grain and can range from 40 to 75% moisture (wb) (Pordesimo et al., 2004; Shinners and Binversie, 2007). During multi-pass corn stover harvest, the stover is allowed to dry in the field until 15–20% moisture (wb) is reached, followed by windrowing and baling (Darr and Shah, 2012); however, in-field drying is not possible in all climates and in all harvest years. A recent study by Oyedeji et al. calculated that only 37% of corn stover harvested in the U.S. is <20% moisture (wb) (Oyedeji et al., 2017). Likewise, the moisture content of corn stover residue was shown to vary significantly in the Midwestern U.S. Corn Belt over two harvest years. During a dry harvest year, the majority of harvested stover met the 20% moisture target; yet in the prior year, high grain moisture and wet field conditions resulted in stover baled and stored at moisture contents exceeding the 20% target (Kenney et al., 2013b).

In situations where field drying is not sufficient for meeting the moisture target for baling, the resulting high-moisture bales can suffer dry matter losses ranging from 10 to 30% (Shah et al., 2011; Smith et al., 2013; Wendt et al., 2014). Aerobic microbial degradation by bacteria, yeast, and fungi consumes valuable carbohydrates, leaving behind material enriched in non-fermentable biomass components such as lignin and ash. Bales also are at risk of significant microbial degradation and associated losses if they get wet during storage (Kenney et al., 2013b; Smith et al., 2013). Sahoo and Mani identified significant cost implications when bales suffered elevated dry matter losses due to degradation (Sahoo and Mani, 2017). Bales degraded during storage also lose their physical structure, which results in increased physical material losses during handling and transportation. Twines in degraded square bales may become loose and often are not cut in bale de-stringers, and twines can cause fires in grinders if they are not removed prior to size reduction operations. Another significant issue associated with dry bales is the fire risk during storage and preprocessing, as the bales are a large source of combustible material. Storage fires may be caused naturally, by lightning strikes or wild-fires, or by human activities such as sparks from adjacent mowing, welding, off-road vehicle activity, or arson. Increased moisture in corn plants also relates negatively to grinding performance, resulting in reduced throughput and increased energy consumption in hammer-mills (Probst et al., 2013; Cao and Rosentrater, 2015). Drying between first and second stage hammer-milling is assumed as a requirement for maintaining operational efficiency in recent designs for preprocessing biomass (Kenney et al., 2013a), yet this results in significant cost implications (Yancey et al., 2013).

While wet climates are more commonly associated with the Corn Belt and northeastern U.S., moisture management in corn stover is also a consideration in southern geographical locations as well, specifically where double-cropping is available to farmers. An example is winter wheat in Kansas, where it is advantageous to harvest the corn crop early in the season and subsequently remove the stover from the field as soon after harvest as possible to facilitate planting the winter crop (Heggenstaller et al., 2008). Current estimates of the Billion Tons of bio-based resources available for conversion to biofuels neglect the influence of double-cropping, and enabling this practice could increase not only resources for bioenergy but for human consumption as well.

Wet anaerobic storage (i.e., ensiling) is an alternative to dry bale storage. Wet storage is one of the lowest risk and most potentially flexible biomass storage options available. It accommodates preprocessing early in the supply chain through forage chopping, utilizes low cost but scalable storage facilities, preserves biomass with minimal losses over time, and can accommodate wet or dry biomass (water can be added on site if needed). Ensiling effectively preserves biomass through storage conditions that limit oxygen availability, which encourages lactic acid bacteria to ferment soluble sugars into organic acids; the resulting low pH environment further preserves biomass from microbial degradation during storage (McDonald et al., 1991). Wet anaerobic storage can occur in a range of different formats including silage bags, bunkers, and drive-over piles, which provides flexibility for the end-user. Common among these storage options is oxygen limitation during construction, such as packaging the chopped biomass into plastic silage bags or by compacting with a tractor during construction of a drive-over pile or bunker. Centralized wet storage at a biorefinery gate is an additional storage option that has received little attention but is common in the pulp and paper industry as well as in the case of sugarcane bagasse that is used by the sugarcane industry to heat boilers.

Wet-based systems utilizing field chopping for particle size reduction and ensiling for stable storage have been evaluated for corn stover (Shinners et al., 2011), sorghum (Henk and Linden, 1996; Shinners and Binversie, 2003; Williams and Shinners, 2012), and grasses (Oleskowicz-Popiel et al., 2011) destined for bioenergy use. However, techno-economic assessments on fully wet, bulk logistics systems for corn stover are limited to a handful of studies (Turhollow and Sokhansanj, 2007; Cook and Shinners, 2011), and none of these studies assess centralized storage at the biorefinery gate. The primary drawback of chopped logistics systems is that the transportation costs increase compared to bale systems, as trucks reach the maximum allowable load weights before volumetric capacity has been reached due presence of water, such that long distance transportation is generally thought to be cost prohibitive. However, chopped, wet logistics systems offer multiple advantages over bale systems primarily due to the fact that they have consistently and successfully demonstrated biomass preservation of >95% over 6–12 months (Shinners et al., 2003, 2011) compared to the 10–30% seen in aerobic systems (Smith et al., 2013; Emery and Mosier, 2014; Wendt et al., 2014, 2018). Wet storage systems also reduce the fire risk relative to dry bales in storage and preprocessing. Storage of high-moisture materials result in a product that is more difficult to ignite, either accidentally, or intentionally. Centralized wet storage allows for greater access, control and restriction of operational activities that increase fire risks. Additionally, chopped biomass handling eliminates bale twine related issues, including failed cuts, reducing the risk of grinder fires.

This research defines an approach for managing the >50% of corn stover that Oyedeji et al (Oyedeji et al., 2017) assert is too wet for stable storage in bales and can be problematic for existing preprocessing approaches for low-moisture bales. The chopped logistics system scenario provides the quantity of feedstock sufficient to satisfy the biorefinery demand (2,000 dry tonnes of biomass per day) to the throat of the biochemical reactor. The chopped system was compared to a bale logistics system that utilizes higher moisture content bales, 30% wb, which require drying to 20% to maintain efficient preprocessing throughput. The existing low-moisture dry bale infrastructure was assumed to be utilized for bales that can be harvested later in the year, stored field-side at low moisture contents and with low corresponding dry matter losses, and delivered to the biorefinery at 20% moisture (wb). Laboratory- and field-based storage studies, described in Wendt et al. (Wendt et al., 2018), were used to characterize the performance of the industrial-scale ensiling storage and to inform the final design and associated techno-economic analysis of the chopped feedstock logistics systems. Costs, energy consumption, greenhouse gas emissions, and water consumption were evaluated for the chopped system in comparison to the 30% moisture bale logistics system to estimate the full economic and environmental impact of the described approach.

## METHODS

#### Biomass Supply Scenarios

In this study, a hybrid feedstock supply chain scenario was assumed where approximately half of harvested corn stover is compatible with stable storage in bales (<20% moisture, wb). The remaining half of the corn stover is not able to be fielddried to 20% moisture (wb) and assumed to be delivered to the biorefinery at 30% moisture (wb) bales, as reported in Cook et al. (Cook and Shinners, 2011). Alternatively, the chopped logistics system is used for the high-moisture fraction of corn stover. The chopped logistics system design includes infrastructure to store 200,000 dry tonnes at the biorefinery gate (delivers 188,500 tonnes after 5% dry matter loss) in order to maintain consistency with previously reported costs for large, scale corn stover storage piles (Turhollow and Sokhansanj, 2007). Additional feedstock would also be brought into the biorefinery during the harvest season to supply the bioreactor with freshly-harvested biomass during the week (131,429 tonnes) and during weekends (52,571 tonnes). Tonnages are presented in **Table 1**. Annual biorefinery demand to supply 2,000 tonnes (2,205 U.S. tons) per day at 96% uptime is 701,254 tonnes (773,000 U.S. tons) (Humbird et al., 2011).

#### Logistics Supply Chain Modeling

Both bale logistics scenarios are based on the parameters described in Jacobson et al. (Jacobson et al., 2014) for three-pass harvesting systems to deliver 2,000 dry tonnes of corn stover to the throat of a biochemical reactor per day with the exception that pelleting is omitted. Following the grain harvest, a flail shredder pulled by a tractor is used for initial stover harvesting and collects the material into a windrow, and then a baler collects the material from the windrow into bales. Feedstock yield is assumed at 2.7 tonnes/hectare with resulting moisture content assumed at either 20 or 30% (wb). Bales are stored field-side in tarped stacks, and an assumed 10 or 12% dry matter loss occurs during the year of storage for 20% moisture (Hartley et al., 2015) and 30% moisture bales (Vadas and Digman, 2013), respectively. Bales are transported an average of 82 km using flatbed trucks to a preprocessing facility at the biorefinery gate. The bales are stored field-side, and therefore transportation operations occur year round. Transportation and handling costs include all processes involved in the movement of material from multiple local locations to the biorefinery gate and include processes such as loading, trucking, rail transport, and unloading and conveyance. A two-stage hammer mill system is used to reduce particle size in order to meet reactor in-feed size requirements, with a rotary dryer used for wet bales to reduce moisture content from 30 to 20% (wb) between first and second stage grinding. This is accomplished in order to maintain operating efficiency in the second stage grinder (Kenney et al., 2013a).

The chopped logistics system was designed such that the corn stover is transported to the centralized facility at the biorefinery gate during the annual 92 day harvest period at a rate sufficient to both provide continuous feed to the biochemical reactor as well as construct the wet storage piles. Harvesting occurs over a 92 day period in the fall season with operations occurring 5 days per week and 16 h per day, and stover moisture is assumed to be 45% (wb) due to the extension of the harvesting window earlier in the season prior to dry down of the corn stover. Harvesting is identical to the baled logistic system up to the point of baling, which is replaced by collection using a forage harvester with a chopping head. The forage harvester is tended by highdump wagons pulled by tractors, which are used to transport and load the chopped material into waiting semi-trucks. Semi-trucks pulling open top possum belly trailers are used to transport the corn stover to the centralized preprocessing and storage facility co-located with the biorefinery. As in the baled logistics case, a yield of 2.7 tonnes/hectare and an average draw radius of 82 km are assumed. A comparison of the baled and chopped logistics systems is provided in **Table 2**.

#### Centralized Wet, Bulk Operations for the Chopped Logistics System

The unit operations in the centralized wet storage design include receiving, screening and shredding, storage pile formation and reclaiming, and delivery to the reactor throat (**Figure 1**), and general design parameters are presented in **Table 3**. The reactor feed requirements include up to 50% moisture content, a particle size of <2.54 cm (1 inch), and 5% or less structural ash. The centralized operations necessary to provide 2,000 dry tonnes per day of as-harvested biomass feedstock while simultaneously forming a 200,000 dry tonnes storage pile for later utilization were determined through development of a mass balance of material flows through the unit operations. All proposed unit


TABLE 1 | Annual tonnages required to supply a biorefinery with 2,000 dry tonnes corn stover per day.

\**Bale-based logistics systems do not remove soil contamination.*

operations are designed to include 20% surge capacity. Surge capacity for providing a continuous biorefinery feed stream is achieved through the use of day piles (interruptions of 8 h or less in the receiving operations) and reclaiming material from the ensiled biomass storage pile (interruptions of >8 h in the receiving operations), as described in the following paragraph. A listing of all equipment quantities, capacities, power requirements, and costs are presented in **Table S1**.

Incoming trucks enter past one of four gatehouses located at separate entry points to reduce traffic congestion, each gatehouse equipped with inbound and outbound electronic scales. Corn stover is emptied via truck tippers into receiving hoppers. Stover is then conveyed through a magnetic separator and delivered to one of two day piles that enables 24 h utilization of downstream equipment given the 16 h receiving period. The circular day piles are formed by stackers and recovered with screw reclaimers. Losses of 0.75% are assumed in the day piles due to biological degradation, as measured in previous studies of high-moisture corn stover stored aerobically (Wendt et al., 2014). Receiving and day pile construction design parameters are summarized in **Table 4**.

Eight multi-stage screening processes are operated in parallel. Each multi-stage screening process uses first-stage disc screening to separate the oversize fraction from the fraction that meets biorefinery size specification; the oversized fraction is subject to additional size reduction via shredding prior to being combined with the at-specification fraction. Soil is removed from the atspecification sized fraction via vibratory screening. Screening and shredding design parameters are summarized in **Table 5**.

Preprocessed, as-harvested stover that will directly be utilized by the biorefinery is conveyed to the feed day pile, which operates in a similar fashion to the receiving day pile by allowing for continuous corn stover being conveyed to the biorefinery at a rate of 2,000 dry tonnes/day. The remaining biomass is diverted to four storage piles with a total capacity of 200,000 dry tonnes. In addition to providing long-term storage, the storage piles provide biorefinery feed for time intervals in the harvest season when the receiving operations are offline for longer than 8 h (weekends and/or periods when weather or other factors prevent collection, transportation, or receiving operations from operating at full capacity). As such, a total of 255,800 dry tonnes are stored in the piles, and the first piles that are constructed are partially utilized and then filled again at the end of the harvest window. Storage piles are constructed using a pair of stacking conveyors, which move on rails and construct the entire length of one pile before beginning the second. Landfill compactors are continuously operated during pile construction to compress the stover as well as to mechanically exclude oxygen. Corn stover is removed from the storage pile using a scraper-style reclaimer mounted on rails. The reclaimer draws material off the pile in slices ∼1 m in width as it moves down the length of the pile to limit the amount of pile that is exposed to oxygen prior to use and to provide some mixing of the reclaimed material. The reclaimed stover from the storage pile is conveyed to the feed day pile for further mixing and 24 h queuing prior to being conveyed to the biorefinery. Dry matter losses of 5% are assumed for the ensiling storage design, as reported in Wendt et al. (2018). Storage pile design parameters are summarized in **Table 6**.

The storage pile area is lined with ultra-high-molecularweight (UHMW) polyethylene, a layer of sand and gravel to protect the liner, and a layer of sacrificial biomass to prevent the introduction of gravel into the reclaimed corn stover. The storage pile is designed with a water collection system to accommodate runoff due to rainfall. The surface under the piles is inclined to route drainage water into catch basins positioned along the centerline of the pile. The catch basins connect to a common drainage pipe that diverts water to the end of the pile and into a collection basin. Collection basin water may be used to adjust moisture content or to deliver additives, such as microbial inoculum or acids commonly used in ensiling, to the biomass during pile construction. Excess runoff water from the collection basins is stored onsite in UHMW-lined water collection ponds. It is assumed that excess recovered water from the storage systems can be field-applied to nearby farmland and that wastewater treatment is not required. Water management system design parameters are summarized in **Table S3**.

#### Techno-Economic Analysis

The Biomass Logistics Model (BLM) framework as described previously (Cafferty et al., 2013) (Lamers et al., 2015) was used to determine costs for harvest & collection and transportation in the chopped logistics systems as well as the corresponding costs for the baled logistics system (harvest & collection, fieldside storage, transportation, preprocessing, and dockage). Costs TABLE 2 | Comparison of dry and wet feedstock supply logistics systems.


\**Bale-based logistics system does not remove soil contamination.*

in the BLM model are calculated using American Society of Agricultural and Biological Engineers (ASABE) standards, which provide guidelines on ownership costs, hours of equipment use, and salvage value. This approach assumes costs and returns consistent with agricultural economic practices, where purchased machinery is replaced roughly every 10 years and labor costs are based on salaried or seasonal employees leveraged across multiple crops. Grower payment was based on anticipated levels required to incentivize farmer participation (Hartley et al., 2015).

The centralized preprocessing & storage operations in the chopped logistics system is located at a biorefinery, and therefore they are costed in a manner similar to other DOE Bioenergy Technology Office design cases including Humbird et al. (Humbird et al., 2011), which use a discounted cash flow analysis based on estimates of capital and operating costs. Industrial-scale operations, such as the centralized preprocessing & storage facility described here, require installation of permanent infrastructure and ancillary equipment as well as indirect cost items such as engineering and permitting. Likewise, full-time supervision and maintenance staff are required in order to meet the throughput and capacity requirements of a centralized processing facility.

The centralized preprocessing & storage total installed cost (TIC) was determined by multiplying the total purchase cost for each equipment item by the selected installation factor (**Table S1**). Centralized preprocessing and storage total direct costs (TDC) include TIC plus other direct costs including a warehouse (4% of TIC) and site development (10% of TIC), along with design-specific costs for the storage pile area and water collection ponds. Indirect costs are calculated as a percentage of TDC using the parameters defined by Lamers et al. for estimating biomass preprocessing facility costs, including engineering (4%), construction expenses (4%), contractor's fee

(2%), and contingency (5%) (Lamers et al., 2015). The fixed capital investment (FCI) is the sum of direct and indirect costs, which is combined with a 5% working capital cost to result in a total capital investment (TCI).

The centralized preprocessing & storage fixed operating costs include labor, land rent, maintenance and property insurance (**Table S2**). Labor and supervision personnel include a plant engineer, maintenance technicians, gatehouse attendants, shift operators, and loader drivers; all positions are either shared with the adjacent biochemical refinery or are seasonal hires. Labor and supervision costs were calculated using Bureau of Labor and Statistics rates (U.S. Department of Labor, 2015). A 90% labor burden charge is applied to the total salary cost to obtain the fully loaded salary cost. Land rent was assumed to be an annual rate of \$70.54 per hectares for 65 hectares based on 2014 cash rents for non-irrigated land in Stevens County, KS (U.S Department of Agriculture, 2015). Maintenance costs were assumed at 3% of total installed equipment costs, and property insurance was assumed at 0.7% of Fixed Capital Investment as in Humbird et al. (2011).

Centralized preprocessing & storage variable operating costs include energy and mobile equipment operating costs (**Table S2**). Power requirements for each equipment item are listed in **Table S1**. Annual electricity usage for each equipment item was calculated by applying an operating load factor of 0.7 to the product of the equipment power rating and specified annual operating hours. Electricity costs for operation of the warehouse and other supporting infrastructure was also considered. Total electricity cost was calculated assuming an electricity price of \$0.0665/kWh (U.S Energy Information Administration, 2016b). Mobile equipment operating costs include lease (Wyoming Machine Company, 2016), maintenance (Caterpillar Incorporated, 2012), repair (Jackson, 2010), and fuel (American Petroleum Industry, 2016; U.S Energy Information Administration, 2016a) for two landfill compactors used for 3 months a year for storage pile construction.

The centralized preprocessing & storage capital and operating costs were annualized using a discounted cash flow analysis, assuming identical economic parameters and discounting as described elsewhere (Humbird et al., 2011). This economic TABLE 3 | General design parameters for the preprocessing & storage operations of the chopped logistics system.


scenario assumes a 10% discount rate, 10% internal rate of return, and a 30 year plant life. Equity financing was assumed at 40% with a loan at 8% interest for 10 years assuming nth-plant designs. Depreciation was set at a 7-year schedule, and federal corporate taxes were assumed at 35%. Construction time was assumed at 3 years, with a 6 month start-up time assuming 50% production, 75% variable expenses, and 100% of fixed expenses. Total capital and operating costs are reported in **Table S4**. Feedstock sales to the conversion facility were considered income for the storage facility, and the feedstock pricing was varied to result in a net present value of zero. The total income was divided by the amount of corn stover purchased to calculate overall cost per dry tonne. The discounted cash flow analysis for the chopped logistics system is presented in **Table S5** to demonstrate the application of this costing approach.

#### Sensitivity Analysis

A sensitivity analysis was performed in order to determine the key drivers for cost and energy consumption of the chopped logistics system. High impact inputs were determined by varying each input factor in isolation and measuring the change in feedstock cost (\$/tonne) and energy consumption (MJ/tonne) in harvest and collection, transportation, and preprocessing TABLE 4 | Truck receiving/day pile design parameters for preprocessing & storage operations of the chopped logistics system.


TABLE 5 | Screening/shredding system design summary for preprocessing operations of the chopped logistics system.


and storage. Harvest yield was varied from 1.7 tonnes/hectare, a conservative estimate based on (Wendt et al., 2018), to 4.5 tonnes/hectare (Shah and Darr, 2016). Bulk density in transportation was varied from the 48 kg/m<sup>3</sup> , the density of 28 cm chopped corn stover (Shinners and Binversie, 2003), to the upper range of 96 kg/m<sup>3</sup> for freshly cut forage (Wiersma and Holmes, 2000). Harvest window was varied from the most active harvest period for corn grain of 42 days (U.S. Department of U.S Department of Agriculture, 2010) to a 123 day window that prioritizes grain harvest over stover harvest (Lizotte and Savoie, 2011). Size reduction requirements were based on (Wendt et al., 2018) and were either doubled as a conservative estimate or eliminated based on reports that forage chopping eliminates the need for additional size reduction (Lisowski et al., 2017). Dry matter loss ranged from 2 to 10%, typical ranges for silage (Borreani et al., 2018). Diesel and electricity cost were varied by 25%.


#### Sustainability Metrics

Energy consumption, weighted by type (diesel, natural gas, and electricity) for each unit operation, was input into Argonne National Laboratory's Greenhouse Gases, Regulated Emissions, and Energy use in Transportation (GREETTM) model (Wang et al., 2015). GREET computes fossil, petroleum, and total energy use (including renewable energy in biomass), emissions of GHGs (CO2, CH4, and N2O), and emissions of six air pollutants: carbon monoxide (CO), volatile organic compounds (VOCs), nitrogen oxides (NOx), sulfur oxides (SOx), and particulate matter with a diameter below 10 micrometers (PM10) and below 2.5 micrometers (PM2.5). Biogenic emissions of CO, NOx, and N2O measured in ensiled corn stover were also included (Wendt et al., 2018). Energy consumption and GHG release was summarized on a dry tonne basis for each unit operation presented.

#### RESULTS

#### Techno-Economic Analysis

In this study, a chopped feedstock logistics system for corn stover was defined that utilizes on-farm harvest and collection through forage chopping, transportation to a central facility located at a biorefinery gate, preprocessing to meet size and ash specifications, automated storage pile construction using industrial-scale ensiling to manage seasonal variability, and delivery of corn stover to a biorefinery reactor throat at a rate of 2,000 dry tonnes per day. This system was compared to a conventional baled system. Both logistics systems assume that a biorefinery will employ a third party aggregator to harvest and deliver the corn stover to the biorefinery gate in order to be consistent with recent models (Kemp and Stashwick, 2015; Shinners et al., 2017; Mertens et al., 2018). Operations within the biorefinery gate including preprocessing and the centralized wet storage system were considered add-ons to the biorefinery as described elsewhere (Humbird et al., 2011), with the chopped logistics systems employing additional supervisory staff and infrastructure due to the extensiveness of the automated design. This differs from the conventional bale system in that initial size reduction occurs prior to storage, the corn stover is stored onsite rather than farm-side or at satellite locations, and that the TABLE 7 | Comparison of costs in chopped and baled logistics systems (2015 US dollars per dry tonne).


\**Represents the combination of refinery storage and handling, preprocessing, and centralized storage.*

corn stover is stored anaerobically and at a moisture content that reduces the risk of fire and microbial degradation.

Costs for both the chopped logistics scenario and the 30% moisture bale system are compared in **Table 7**. A grower payment of \$37.64/tonne was used in each scenario for consistency and contributed to nearly 30% of the total costs. For the chopped logistics system, harvest & collection and transportation costs totaled \$44.68/tonne and contributed to 32% of the total costs. The centralized preprocessing & long-term storage costs for the chopped system were estimated at \$46.88/tonne and 21% of the total costs, of which size reduction, ash removal, and handling constituted ∼55% of the total and storage and queueing accounted for the remaining 45%. A quality dockage, described in the following paragraph, of \$8.84/tonne cost was applied to this system. Total costs for the chopped logistics system were \$137.86/tonne. In comparison, harvest, collection, field-side long-term storage, and transportation costs for the 30% moisture bale logistics system were estimated at \$41.95/tonne or 30% of the total. Biorefinery operations including short-term storage, handling, and preprocessing were \$27.47 or 22% of the total. Dockage in this system was \$18.62/tonne. Total costs for the 30% moisture bale logistics system were \$125.70/tonne, ∼10% lower than the chopped system.

Dockage cost is incurred in each logistics system in order to account for losses in the feedstock supply chain, primarily due to the displacement of biomass by contaminating soil and the loss of material due degradation in long-term storage. The ash specification at the biochemical refinery is 4.9% (Humbird et al., 2011) which is based on the physiological content of ash in corn stover (Weiss et al., 2010). Dockages account for both ash disposal costs as well as the purchase of additional feedstock to meet tonnage targets at the biorefinery. In this study, the chopped logistics system incurred a total dockage of \$8.84/tonne, whereas dockage in the bale system was \$18.62/tonne. This difference is related to the higher ash content and higher degradation rate in storage for the 30% moisture bale system. Ash contents of 12% are assumed for multi-pass corn stover bales delivered to the Wendt et al. Chopped Feedstock Logistics System TEA

biorefinery gate (Jacobson et al., 2014), whereas the preprocessing operations in the chopped system actively remove this ash related to soil contamination. Dry matter loss of 12% is assumed during aerobic bale storage in the 30% moisture case (Jacobson et al., 2014). However, dry matter loss is reduced to 5% in the chopped logistics system due to the use of ensiling as a storage approach (Wendt et al., 2018), further reducing total dockage in this system.

#### Sustainability Metrics

Energy consumption and GHG release were determined for the unit operations in each logistics system (**Table 8**). The energy consumption data indicate that the chopped harvest and collection operations are more energy intensive compared to the baled system because they involve in-field size reduction through forage chopping. Likewise, energy consumption and GHG release for transportation of the chopped biomass increases significantly compared to the bale system due to the lower bulk density of the chopped biomass compared to bales. However, preprocessing energy requirements account for the majority of the energy consumed in the bale logistics system. Of the 1670 MJ/tonne required in preprocessing for the bale system, roughly 80% are required for drying from 30 to 20% moisture (**Table S6**, Supplementary Information); the dryer energy consumption alone is greater than the total energy usage in the chopped logistics system. Overall, the energy consumption of the chopped logistics system is 48% less than that of the energy consumption of the 30% moisture bale logistics system. The reduction in overall energy consumption also leads to a reduction in total of GHG emissions of over 60% for the chopped system, as shown in **Table 8**.

Gasses released during storage are a concern from a GHG emission perspective, as they can potentially be significant GHG sources or air pollutants (Emery and Mosier, 2012, 2014). Storage studies performed in laboratory reactors provided gas release data for the storage conditions used in the chopped and bale logistics systems (Jacobson et al., 2014; Wendt et al., 2018). Biogenic storage gasses (CO, NOx) detected in laboratory experiments were added to the total GHG emissions but did not have a significant impact on overall GHG release. On the other hand, CO<sup>2</sup> released during storage as a result of microbial degradation can be significant, especially in the case of highmoisture aerobic storage. Biogenic CO<sup>2</sup> released during storage due to degradation would have ultimately been emitted during fuel combustion, and therefore it is not included in the overall GHG emissions listed in **Table 8**. However, carbon utilization efficiency is highly reduced in the aerobic bale storage case compared to the chopped system which utilizes ensiling to prevent excessive degradation. The chopped system resulted in 6.5 kg CO<sup>2</sup> released per dry tonne biomass while the bale logistics system released 159.7 kg CO<sup>2</sup> per dry tonne biomass assuming aerobic, field-side storage of corn stover bales at 30% moisture with 12% dry matter loss. In summary, aerobic storage in 30% moisture bales releases 25 times more biogenic CO<sup>2</sup> to the atmosphere compared to ensiling, resulting in poor carbon utilization and ultimately lower fuel yield.

Biochemical conversion of cellulosic biomass to fuels is a water intensive process considering that 30% solids content, or 70% moisture content (wb), is desired in dilute acid pretreatment (Humbird et al., 2011), yet incoming dry stored biomass is ideally <20% moisture (wb) in tarped, stacked bales (Darr and Shah, 2012). One advantage of the chopped logistics system is that the water is maintained within the biomass and therefore can reduce the water burden at the pretreatment reactor. Assuming 50% moisture content (wb) at the time of conversion, the chopped logistics scenario requires an additional 1,330 L of water per dry tonne biomass. In comparison, the baled logistics system would require 2,080 L per dry tonne biomass to increase moisture content from 20% (wb) to 30% solids in a pretreatment reactor. Overall, the chopped systems require less water input at the biorefinery gate, resulting in a small cost reduction. Assuming the reduction of water consumption necessary at the reactor throat for the chopped logistics systems compared to the baled system and a water cost of \$0.22/m<sup>3</sup> (Peters and Timmerhaus, 2003), a credit of \$0.19/tonne of corn stover was applied to the chopped logistics system as indicated in **Table 7**.

#### Sensitivity Analysis

A sensitivity analysis was performed in which yield, harvest window, bulk density in transportation, particle size reduction at the biorefinery gate, dry matter loss, electricity cost, and diesel fuel price were varied in isolation. Changes in cost and energy consumption in harvest & collection, transportation, and preprocessing & storage operations in the chopped logistics system were assessed as a result of changes to these variables (**Figure 2**). The largest reductions in the total feedstock cost resulted from eliminating particle size reduction beyond forage harvesting (\$7.80/tonne total feedstock cost decrease). The second largest cost driver is harvest yield, with significant cost fluctuations observed with an increase in the yield from 2.7 to 4.5 tonnes per hectare (\$7.74/tonne decrease) or a decrease in the yield to 1.7 tonnes per hectare (\$9.78/tonne increase). Bulk density during transportation was another primary cost driver; decreased costs were predicted by increasing bulk density from 73.7 to 96 kg/m<sup>3</sup> (\$6.78/tonne decrease). Likewise, transportation costs could increase when bulk density was decreased from 73.7 to 48 kg/m<sup>3</sup> (\$15.47/tonne increase). The influence of harvest yield and bulk density on total feedstock costs of the chopped system are consistent with reported cost drivers for baled corn stover (Shah and Darr, 2016).

The duration of the harvest window in harvest & collection impacted costs significantly, with a moderate cost decrease when the window was increased from 92 to 123 days (\$3.45/tonne decrease) and a drastic increase experienced with the decrease in the window from 92 to 42 days (\$8.98/tonne increase). Reported harvest windows for corn grain vary widely by state (U.S Department of Agriculture, 2010; Oyedeji et al., 2017) and the corn stover harvest expands that window based on the reliance of field drying in windrows; one advantage of the proposed chopped logistics system is that the harvest window can be expanded due to the fact that it is not reliant on field drying of the stover (Shinners and Binversie, 2003; Cook et al., 2014). The remaining variables presented in the sensitivity TABLE 8 | Estimated energy consumption and GHG release for corn stover in the chopped and baled logistics systems.


\**Represents the combination of refinery storage and handling, preprocessing, and centralized storage.*

analysis, including diesel and electricity price as well as dry matter loss in storage, resulted in fluctuations of < \$3/tonne. Overall, the results of the sensitivity analysis indicate that operations in harvesting and transportation operations have the greatest impact in cost fluctuations of the proposed chopped logistics system. Similarly, harvest yield and bulk density in transportation were the primary drivers in reducing energy consumption.

## DISCUSSION

Chopped logistics systems offer many benefits for the bioenergy industry, primarily the removal of dependence on field-drying to provide stable, low moisture storage conditions in bales. A recent analysis assessed the practicality of corn stover to be fielddried to a moisture content of 20% and determined that this is possible an average of only 36% of the time in the top 10 corn stover producing states, ranging from 16 to 67% depending on the state (Oyedeji et al., 2017). Bales with moisture content >20% are subject to increased dry matter loss during storage and require supplemental drying for the preprocessing grinding operations to maintain operating efficiency. The chopped logistics system does not require the feedstock to be dried prior to the preprocessing operations and can consistently reduce dry matter losses in long term storage by utilizing ensiling. Therefore, the chopped logistics system was evaluated for its overall competitiveness for providing the biorefinery feed that the bale logistics system would be unable to provide unless moisture contents of 20% or less could be achieved at the time of harvest.

The cost for the chopped logistics system described in this study was 10% higher than a bale-based logistics system with a moisture content of 30%. A number of differences are apparent when comparing the unit operation specific costs of each logistics system. Harvest and collection costs were reduced for the chopped system because of increased harvest efficiency for forage chopping compared to baling, which is consistent with previous reports (Shinners and Binversie, 2003; Cook and Shinners, 2011). However, transportation costs for the chopped system are almost double compared to the bale system due to the reduction in bulk density during truck transport. The dockage cost in the bale scenario is over twice that of the chopped system primarily due to the cost of procuring replacement feedstock either displaced by contaminating soil or lost during storage. While both systems experience a dockage for disposal of ash, a positive attribute of the chopped system is that it includes active ash removal during preprocessing such that the biorefinery receives a lower ash feedstock. Removing the ash prior to preprocessing and conversion has unquantified benefits in terms of reduction of abrasion and it also increases the amount of organic material that can enter the biorefinery, ultimately increasing conversion yield. Of similar importance, the reduced water demand in the chopped system has positive environmental implications due to the reduced water consumption at the biorefinery, and water shortages are predicted in at least a portion of the majority of U.S. States within the next 10 years (U.S Government Accountability Office, 2014).

The assumption in the 30% moisture bale logistics system that drying was necessary to maintain the efficiency and biomass throughput in the second stage grinding operation is consistent with previously reported approaches for managing moisture in preprocessing (Hess et al., 2009; Kenney et al., 2013a). Drying increases costs of this system by \$9.10/tonne compared to a 20% moisture system with no drying (**Table S6**, Supplementary Information), but it is necessary for maintaining efficiency in the two-stage hammer milling operations. Hammer milling operations are most effective with low moisture (<20%, wb) biomass because they use impact as a means to deconstruct biomass, and moisture increases the shear strength of the biomass such that it does not break upon impact (Himmel et al., 1985). Although hammer milling may not be the most effective approach for size reducing high moisture baled corn stover, it has been used in the first-of-a-kind cellulosic biorefineries and thus was used as the baseline in this analysis. Rotary shear milling is an alternative size reduction approach that is less affected by moisture and could be substituted for the second stage hammer mill in the 30% moisture bale scenario, which would eliminate the rotary dryer in the high-moisture bale scenario and result in a combined cost savings of \$9.24/tonne. However, rotary shearer mills are not yet commonly utilized at the commercial scale for biomass deconstruction and were therefore not considered in this analysis.

The centralized storage operation in the described chopped logistics system is associated with significantly higher storage and handling costs than the field-side storage operation in the 30% moisture bale scenario as well as in previously described chopped logistics systems utilizing on-farm storage through ensiling (Shinners and Binversie, 2003; Cook and Shinners, 2011; Vadas and Digman, 2013). The primary factor for increased cost in the described chopped logistics system is the high infrastructure costs of the automated pile formation and reclaiming equipment (> \$30,000,000 installed, **Table S1**). Significant cost savings could be achieved through less expensive automated designs or by utilizing the storage pile for multiple biomass sources harvested in different seasons. Commercially available front end loaders and silage facers could replace the automated design in this system, yet the reduction in capital costs would be replaced with higher operating costs. A similar chopped logistics system has been reported for corn stover stored in a 25,000 tonne pile, with identical unit operations for harvest, collection, transportation, and preprocessing of corn stover (Wendt et al., 2017). Manual storage pile formation and deconstruction was used in this prior study, which resulted in higher operating costs associated with labor and equipment rental but lower capital costs; final storage costs were reduced to \$15.61/tonne compared to \$21.10/tonne in the present study. Another contributing factor to the high cost of the chopped system are the costs associated with the centralized preprocessing and storage operation including sitedevelopment, a warehouse, and on-site water collection ponds, all of which are limited or eliminated with field-side, on-farm storage used in the bale scenario. The chopped system also requires additional land rent costs for the centralized operations; land rent costs vary widely across geographical region. This study assumes a conservative land rent cost for southwest Kansas; however, increasing the land rent cost to the average cost for non-irrigated land in Iowa (\$570.8/hectare) (U.S Department of Agriculture, 2017) only increases the total costs of the chopped system by \$0.10/tonne. Overall, a consistent, readily available feedstock source at the biorefinery gate reduces the risk that gaps in supply can occur, ultimately reducing risk, for which the cost implications have not yet been defined.

Chopped feedstocks logistics systems are widely used across the U.S, with an estimated 147 million U.S. tonnes of foragechopped wet biomass harvested annually as animal feed (U.S Department of Agriculture, 2016), yet only a handful of chopped logistics systems have been reported for corn stover. Sensitivity analysis of the chopped logistics system presented in this study identified two primary variables in harvest and collection operations that would result in significant cost reductions: (1) increased harvest yield, and (2) meeting biorefinery size specifications through forage chopping. Harvesting yields range between 1.4 to 5.9 tonnes/hectare for corn stover depending on factors specific to a geographical region, such including harvesting approach, growth year, and geographical location (Sokhansanj et al., 2002). In this study, a conservative yield of 2.7 tonnes/hectare was assumed, and increasing yields would specifically impact the transportation in the chopped logistics system due to a reduced supply radius for the biorefinery. Likewise, increasing size reduction during harvest and the corresponding increase in bulk density during transportation are important cost drivers for chopped logistics systems (Turhollow and Sokhansanj, 2007). The majority of economic assessments for chopped, wet logistics systems assume that forage harvesting is capable of doing all the necessary size reduction required (Shinners and Binversie, 2003; Vadas and Digman, 2013) or that the cost will be less than \$6/tonne (Cook and Shinners, 2011) (Turhollow and Sokhansanj, 2007). This study conservatively assumes size reduction at the biorefinery gate is necessary for ∼40% of the biomass, and \$7.80/tonne in cost reductions could be achieved by eliminating size reduction beyond forage chopping. Cook et al. suggest that costs for chopped logistics systems could be as low as \$86-\$113/tonne, which is achieved through a reduced grower payment, field-side storage in silage bags, and minimal preprocessing costs beyond forage chopping (Cook and Shinners, 2011). Overall, multiple opportunities exist for cost reductions in the chopped logistics system described in this study.

Although the chopped logistics system requires a price premium of 10% relative to 30% moisture bales, this system provides benefits that are less easily quantified in addition to the previously listed operational flexibility, energy use, and water utilization benefits. The proposed chopped logistics system is compatible with not only existing feedstock supply logistics operations and vendor recommended preprocessing and storage pile infrastructure, but also available conversion technologies, leading to a quick entry into the marketplace. Specifically, today's commercially-available forage choppers can harvest and concurrently size-reduce material in the field, and the chopped logistics system delivers a feedstock that is compatible with existing biochemical sugar conversion pathways and is on-spec with regard to size, reactivity, ash content, and tonnage. Furthermore, this system provides solutions to the material handling issues experienced at current commercial-scale cellulosic biorefineries, including a reduction in fire risk in storage and preprocessing, a reduction in fines generated during single-stage low moisture grinding, a narrow particle size distribution of delivered feedstock, and a reduction of soil contamination and fines production in the pretreatment reactor achieved through screening in preprocessing.

# CONCLUSION

A chopped logistics system for processing the high-moisture content portion of a biorefinery's annual corn stover supply was designed and compared to a 30% moisture bale corn stover logistics operation. The chopped logistics system provides a flexible method for utilizing biomass with high moisture content from non-ideal harvesting or field conditions. The dry matter loss associated with ensiled biomass in storage is significantly reduced compared to that of high-moisture bales stored aerobically, resulting in more efficient carbon utilization. The use of a chopped feedstock logistics system in combination with a dry bale system can mitigate risks associated with harvest or weather conditions that preclude all biomass from being stored field-side under conditions that maintain moisture contents of ≤20%, where dry matter loss is limited. Additionally, storage using ensiling in the chopped logistics system can reliably limit the risks of losses from fires in storage and preprocessing operations.

The net costs to the throat of the reactor for the chopped logistics system were 10% greater than the corresponding 30% moisture bale logistics system. Harvest and collection costs were lower for the chopped system compared to baled corn stover, but transportation and preprocessing costs were significantly greater for the chopped system. Cost offsetting benefits of the chopped system include removal of the ash in preprocessing, reduction of the dockage associated with dry matter lost due to degradation in aerobic storage, and credits for offsetting water use required in pretreatment. Direct benefits of the chopped logistics system also include reduced energy consumption and GHG emissions relative to the dry bale system due to the elimination of drying in order manage moisture during size reduction. The centralized preprocessing & storage design for the chopped logistics design presented in this study describes a baseline on which cost and performance improvements can be measured against. The system is flexible such that any equipment piece can be substituted to describe newly developing approaches for handling, size reducing, and storing corn stover. Further research is warranted to adapt existing high-moisture forage harvest, collection, storage, and preprocessing operations to serve the requirements of emerging biorefineries.

# AUTHOR CONTRIBUTIONS

LW, JR, DS, JL, DH, and DW designed the wet, bulk logistics system. DH and WS designed the dry bale systems. WS, KK, and QN advised on the selection of the wet and dry logistics systems. DH, LW, and JM performed sustainability analysis. LW drafted the manuscript with contributions from all co-authors.

# ACKNOWLEDGMENTS

The research was supported by the U.S. Department of Energy (DOE), Office of Energy Efficiency and Renewable Energy (EERE), Bioenergy Technologies Office (BETO), under Award No. DE-AC07-05ID14517. The views expressed in the article do not necessarily represent the views of the U.S. Department of Energy or the United States Government.

The authors thank Thomas Robb and Kirk Spikes for their contributions to the wet logistics design scenario and Vicki Thompson for her review of the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00090/full#supplementary-material

# REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Wendt, Smith, Hartley, Wendt, Ross, Sexton, Lukas, Nguyen, Murphy and Kenney. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Simulation Modeling for Reliable Biomass Supply Chain Design Under Operational Disruptions

Bhavna Sharma<sup>1</sup> , Robin Clark <sup>2</sup> , Michael R. Hilliard<sup>3</sup> and Erin G. Webb<sup>4</sup> \*

*<sup>1</sup> Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, United States, <sup>2</sup> QMT Group, Knoxville, TN, United States, <sup>3</sup> Energy & Transportation Science Division, Oak Ridge National Laboratory, Oak Ridge, TN, United States, <sup>4</sup> Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, United States*

#### Edited by:

*J. Richard Hess, Idaho National Laboratory (DOE), United States*

#### Reviewed by:

*Alok Satlewal, Indian Oil Corporation, India David R. Shonnard, Michigan Technological University, United States*

> \*Correspondence: *Erin G. Webb webbeg@ornl.gov*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *10 May 2018* Accepted: *05 September 2018* Published: *25 September 2018*

#### Citation:

*Sharma B, Clark R, Hilliard MR and Webb EG (2018) Simulation Modeling for Reliable Biomass Supply Chain Design Under Operational Disruptions. Front. Energy Res. 6:100. doi: 10.3389/fenrg.2018.00100* Lignocellulosic biomass derived fuels and chemicals are a promising and sustainable supplement for petroleum-based products. Currently, the lignocellulosic biofuel industry relies on a conventional system where feedstock is harvested, baled, stored locally, and then delivered in a low-density format to the biorefinery. However, the conventional supply chain system causes operational disruptions at the biorefinery mainly due to seasonal availability, handling problems, and quality variability in biomass feedstock. Operational disruptions decrease facility uptime, production efficiencies, and increase maintenance costs. For a low-value high-volume product where margins are very tight, system disruptions are especially problematic. In this work we evaluate an advanced system strategy in which a network of biomass processing centers (depots) are utilized for storing and preprocessing biomass into stable, dense, and uniform material to reduce feedstock supply disruptions, and facility downtime in order to boost economic returns to the bioenergy industry. A database centric discrete event supply chain simulation model was developed, and the impact of operational disruptions on supply chain cost, inventory and production levels, farm metrics and facility metrics were evaluated. Three scenarios were evaluated for a 7-year time-period: (1) bale-delivery scenario with biorefinery uptime varying from 20 to 85%; (2) pellet-delivery scenario with depot uptime varying from 20 to 85% and biorefinery uptime at 85%; and (3) pellet-delivery scenario with depot and biorefinery uptime at 85%. In scenarios 1 and 2, tonnage discarded at the field edge could be reduced by increasing uptime at facility, contracting fewer farms at the beginning and subsequently increasing contracts as facility uptime increases, or determining alternative corn stover markets. Harvest cost was the biggest contributor to the average delivered costs and inventory levels were dependent on facility uptimes. We found a cascading effect of failure propagating through the system from depot to biorefinery. Therefore, mitigating risk at a facility level is not enough and conducting a system-level reliability simulation incorporating failure dependencies among subsystems is critical.

Keywords: operational disruptions, biorefinery, pre-processing, biomass, depots, simulation

# INTRODUCTION

Lignocellulosic biomass (wood, grasses, or non-edible parts of plants) has been considered as a promising feedstock for production of biofuels. The main driving force behind push toward using lignocellulosic biofuels are (1) energy security by reducing dependence on foreign oil, (2) boost rural and national economy through technology deployment and generating employment opportunities, (3) protecting environment by reducing greenhouse gas emissions (Valdivia et al., 2016; Satlewal et al., 2018). While currently the lignocellulosic biofuel industry relies on a conventional bale-delivery system (**Figure 1A**) in which feedstock is harvested, baled, stored locally, and then delivered in a low-density format to the biorefinery (Lamers et al., 2015b), the conventional bale-delivery supply system is accompanied by operational disruptions at the biorefinery. The disruptions are mainly due to difficulties encountered with the handling of bales and quality variability (ash, composition, and moisture) in the biomass feedstock (U.S.Department of Energy, 2016). Operational disruptions include decreased facility uptime, production efficiencies, and increased maintenance costs. System disruptions are especially problematic in a bioenergy industry based on a low-value, high-volume product where margins are very tight. In addition, biomass supply uncertainty in the conventional bale-delivery system limits largescale implementation of biorefinery facilities.

An advanced pellet-delivery system (**Figure 1B**) is one alternative strategy proposed for dealing with decreasing feedstock quantity and quality uncertainty. The advanced pelletdelivery system consists of a network of distributed biomass processing centers called depots. In these depots, biomass is densified from large bales into stable, dense, and uniform material. The depot creates consistent physical and chemical characteristics commodities that meet biorefinery conversion specifications and improve the handling, transport, and storage of biomass material (Hess et al., 2009; Lamers et al., 2015b). This system offers advantages by addressing biorefinery operational and supply risks and providing a reliable biomass supply chain design. While the pre-processing depots are costly, their benefits have been found to outweigh costs (Jacobson et al., 2014). Existing studies, however, have not considered the impact of adding a depot on the entire biomass supply system. In a way, the operational disruptions caused by inconsistent biomass supply and quality at the biorefinery have now moved to the depot. It is not necessarily true that achieving reliability by assuming that failures are independent of the entire system can be accomplished by providing a uniform format product to the biorefinery from the depot. Our hypotheses for this study is that decoupling processing of hard-to-handle bales from the biorefinery to the depot reduces costs and improves performance by minimizing operational disruptions at the biorefinery.

There have been several studies on modeling Biomass Supply Chain (BSC) compiled in review papers (Sharma et al., 2013; Zandi Atashbar et al., 2017). Initially, BSC models were designed under deterministic settings, but when the dynamic nature of BSC along with its complexity and uncertainty was recognized, researchers incorporated risks in models to improve system resilience and reliability (Bai et al., 2015; Liu et al., 2017). The majority of studies have been based on mathematical programming and heuristics (Atashbar et al., 2016) and some studies involved simulation (Hansen et al., 2015; Wang et al., 2018) modeling. Compared to analytical modeling, simulation is a powerful approach to studying complex and dynamic systems and holds great potential for modeling dynamic BSC with parametric, internal, and external uncertainties. A simulation model can also capture temporal variations in failure or repair rates relatively easily (Wan et al., 2014). In this study we developed a revised version of Integrated Biomass Supply Analysis and Logistics (IBSAL) simulation model (Sokhansanj et al., 2006). The current version (IBSAL02.0) is database-centric with additional capabilities to simulate multi-biomass, -form, -product, -facility, and -year biomass supply chain.

A few studies have evaluated the impact of storage (Liu et al., 2017), biorefinery characteristics (Bai et al., 2015), and intermodal facilities disruptions (Marufuzzaman et al., 2014) due to flooding, hurricane, and drought, on costs and BSC network design. In the majority of static planning models, the failure probabilities were considered unchanged over time (Liu et al., 2017). To the best of our knowledge, existing approaches focus only on failures at a facility, however, impact of failure at a facility on all the entities in the supply chain is not evaluated and could significantly impact overall reliability of a system. Hansen et al. (2015) quantified feedstock supply risk by considering uncertainty in yield, dry matter loss, and ash for conventional bale-delivery and advanced pellet-delivery systems. An advanced pellet-delivery system was found to significantly reduce biomass supply risk by diversifying the supply portfolio and protecting the system against catastrophic supply disruptions, such as drought, flood, and pests. However, this study did not evaluate impact on costs and system performance due to facility disruptions in conventional bale and advanced pellet-delivery systems. Therefore, our research question is to determine how costs and operational management decisions for an advanced pellet-delivery system compare with those of a conventional bale-delivery system under facility disruptions. Unlike the descriptions in the above-mentioned articles, we developed a multi-form (bale, pellets) and multiyear database centric simulation model to evaluate the impact of operational disruptions on costs, inventory and production levels, and facility metrics while failure probability at facilities varies across a span of years.

# METHODOLOGY

**Figure 2** shows the methodology used in this study to quantify the impact of operational disruptions at facilities for conventional bale and advanced pellet-delivery systems. Since agricultural residues, particularly corn stover, is the feedstock of choice for near-term ethanol production (Jessen, 2015), a case study for an Iowa biorefinery using corn stover as feedstock is developed. The following three sub-sections: Inputs, Simulation Model-IBSAL 2.0, and Scenarios, along with a case study, describe

the methodology used to investigate the impact of operational disruptions.

# Inputs

#### Spatial Analysis

To incorporate the impact of spatial variation on biomass supply system management decision-making, spatial analysis was conducted using Python and Esri, ArcMap 10.3 (Esri, 2017) software to gather information on farm and facility locations, available corn-stover tonnage, biomass allocation, and distances between sites. Corn-specific land cover and corn yields were determined using the Cropland Data Layer (CDL) (USDA National Agricultural Statistics Service, 2016) and Iowa Soil Properties and Interpretations Database (ISPAID) (Miller et al., 2010) as described by (Brandes et al., 2016). The harvest index (pounds of corn grain divided by total pounds of above-ground biomass, such as corn stover plus grain Pennington, 2013) was assumed to be 0.5, reflecting a corn-stover yield equal to corn grain yield in tons per acre (Lorenz et al., 2010). Multicriteria suitability analysis was conducted to determine suitable biorefinery sites (Sharma et al., 2017). Potential depot/storage locations were restricted to points on a 5-mile grid to provide enough options for optimal depot location estimation for this analysis. Suitability analysis could be conducted to determine ideal depots sites. The ArcGIS location-allocation maximized capacitated coverage problem type was formulated using a US census road network (United States Census Bureau, 2016) dataset and facility locations to determine site facilities and allocate corn stover supply to the facilities in the most efficient manner. Location-allocation is a two-fold problem that simultaneously locates facilities and allocates farms to them. The maximize capacitated coverage problem type (Esri, 2016) chooses biorefinery, depot, and storage sites, such that all, or the greatest amount of corn stover supply, is utilized without exceeding the capacity of any facility. The outputs of spatial analysis were farm, biorefinery, depot, and storage locations, available corn stover tons, allocations of corn stover supply to facilities, and distances between facilities and farms.

#### Facility and Logistical Parameters

As described in **Figure 2**, facility and logistical parameters include facility capacity, breakdown/repair times, processing costs, crew composition, size, harvest/transport costs, storage dry-matter loss, work hours, and harvest window. These input parameters are specific to the case study under consideration and are described in detail in section Scenarios.

### Simulation Model-IBSAL 2.0

The Integrated Biomass Supply Analysis and Logistics (IBSAL) Model-2.0 is an updated version of the IBSAL simulation model (Sokhansanj et al., 2006; Ebadian et al., 2011) (**Figure 3**). This version of model is built around the ExtendSim database. The ExtendSim's integrated model database tends to promote better model design, increase model organization, and streamline modeling processes. In addition, database-centric simulation models require less time to build, modify, maintain, validate and verify and are more reliable, scalable, and flexible (Diamond et al., 2010).

IBSAL 2.0 is a dynamic discrete event simulation model capable of making operational decisions for both the conventional bale-delivery and the advanced pellet-delivery biomass supply chain configurations. The model is multibiomass, multi-product, multi-form, multi-facility, and multi-year supply chain simulation model (**Figure 3**). The model manages the biomass system from harvesting all the way through supply to the biorefinery reactor throat. In addition to biomass harvesting, handling, inventory, and fleet management, the model can test the reliability of the system by considering biomass quality (moisture and ash), biomass degradation (dry matter loss and discarded/expired biomass), and facility failures across a multi-year system.

Biomass loss in the model is done thorough both dry matter loss and discarded biomass. Biomass must be harvested and transported to the field side within a pre-defined harvest window. The biomass not harvested and transported to the field side within this window will be discarded or used for alternative markets. Biomass must also be transported from the field side to the intermediate storage depot within a second pre-defined window while the biomass not transported within this window will be discarded. While in the field, at the field side, or in storage, dry matter loss occurs periodically based on the dry matter loss schedule defined in the model database.

Four primary tasks are defined in the model. The four tasks include: harvesting the biomass, transporting the biomass to the field side, transporting from the field side to the depot, and transporting from the depot to the refinery. Each task is completed by a specific crew type. Each crew type has a unique capacity, performance, and cost parameters depending on the equipment they are assigned. There are multiple crews of each crew type. Each crew has been assigned to a specific depot (harvesting, in-field transportation, and transportation to depot) or biorefinery (transportation to refinery) to minimize traveling between tasks. For example, we would not want a crew who has just finished a farm for depot 1 to travel across the system basically long distance to work on a farm for depot 2.

The modeling style used for IBSAL 2.0 is flipped from a traditional modeling style. The entities moving through the model are resources, not products as in a traditional model. Product characteristics and variations throughout supply chain are tracked in the log table. This flipped modeling approach uses database transactions to manage inventory throughout the system. The model database in this environment is more than simply a container for input and output data. The model database is truly integrated with the model. This modeling style tends to be faster and provides more control when dealing with large amount of product being processed by specialized resources. The entities in this model include crews, depot processing lines, and the biorefinery processing lines. These entities flow through 4 primary constructs in the model. The crew entities flow through the farm construct and the transport construct, depending on the type of crew. The depot processing lines flow through the depot construct. The biorefinery processing lines flow through the biorefinery construct. This focus on populations of entities each transitioning between states to accomplish tasks essentially implements an agent-based approach within a traditional discrete-event modeling tool and has many of the properties of an agent-based model. For instance, after a crew finishes a task, the crew must decide on their next task based on the status of the world. The results emerge from these multiple loosely linked decisions.

#### Scenarios

**Table 1** presents the three scenarios analyzed in this study. These scenarios were developed to illustrate effects of operational disruptions (i.e., equipment failures) at the depot and biorefinery on the performance and behavior of the biomass supply chain system. The current state of the US biofuel industry located in highly productive Midwest corn regions relies on a conventional bale-delivery represented by scenario-1 (Cafferty et al., 2014; Lamers et al., 2015a,b) (**Table 1**). The second and third scenarios represent use of advanced pellet-delivery systems, with scenario 3 representing mature and resilient lignocellulosic biofuel production system that operates in steady state (**Table 1**). The uptime/throughput at the facility was assumed to increase from 20 to 85% (Foody, 2014). Due to feedstock flowability and variability challenges the uptime at biorefinery has been low (U.S.Department of Energy, 2016). These challenges need to be addressed to achieve an overall equipment effectiveness score of 85%, considered as world class for manufacturing industries TABLE 1 | Scenarios analyzed to evaluate the impact of operational disruptions.


(Muchiri and Pintelon, 2008). The system was evaluated over 7 years, including 5 years to reach 85% uptime at the facility (Foody, 2014) and 2 years at steady-state.

#### Case Study

To test the model, a case study was developed for a biorefinery in Iowa (**Figure 4**), the leading producer of corn stover in the US. The biorefinery site was selected by conducting a multi-criteria suitability analysis (Sharma et al., 2017). For commercial scale biorefineries of size 20–30 million gallons per year, the biomass demand is reported between 285,000 and 375,000 bone-dry tons per year (Berven, 2009; Rosen, 2012). Therefore, the biorefinery yearly corn stover demand was assumed to be 335,000 bone-dry tons with a collection radius of 35 miles (Ward, 2015). The stover yield was estimated using the methodology described in section Spatial Analysis. The moisture content in stover and stover losses throughout the system were assumed to be 20% (Darr and Shah, 2012) and 10% (Shah and Darr, 2016), respectively. Farms with a harvestable yield greater than 1.2 tons/acre were selected for analysis (Wirt) (**Figure 4** and **Table 2**). The harvest window and work hours at the farms were assumed to be 32 days per year and 9 h per day, respectively (Lorenz et al., 2010; Darr and Webster, 2014; Shah and Darr, 2016).

To achieve consistency for comparison among scenarios (section Scenarios), the bale storage site and depot site locations were assumed in this analysis to be same. The location-allocation model was used to determine two optimal storage/depot sites and farms allocated to these sites (**Figures 5A,B** and **Table 2**). Distances from each farm to the depot, farm to the biorefinery, farm to storage, and between farms were also estimated

and used as input to the model. All costs (**Table 3**) were estimated using the methodology described by Turhollow et al. (2009). Additionally, personal communication with A. Khanchi (personal communication, 2018) was used to estimate costs of processing and storage of pellets. The depot processing capacity was assumed to be 30 tons/h and the processing cost was estimated as \$21.42/ton. The biorefinery processing capacity was assumed to be 60 tons/h and the cost of drying and grinding biomass at a biorefinery was estimated as \$6.24/ton. Operating days per year and work hours per year at depot and biorefinery were assumed to be 351 days per year and 24 h per day, respectively. Storage costs for bale storage and pellet storage were estimated as \$7.36 and \$9.22/ton, respectively.

Four types of crews were considered in the model: (a) baling, (b) in-field transportation, (c) road transportation to the depot, and (d) road transportation to the biorefinery. Equipment considered included a large rectangular baler (3 × 4 × 8 feet), an in-field transporter (12 bales per load) (Stinger Inc., 2018), and a flat-bed trailer (36 bales per load) (McGill and Darr, 2014). Fixed and variable costs and effective field capacity and machine capacity were estimated using the methodology described by (Turhollow et al., 2009). Large square baler field speed, and efficiency were assumed to be 80% and 5 miles TABLE 2 | Summary of parameters for farms, storage, depot and biorefinery used as input for the simulations model.


TABLE 3 | Fixed and variable costs of equipment considered in this analysis.


per hour, respectively (ASAE, 2003). The inflation-adjusted list prices for large square baler, stinger stacker, and flat-bed trailertruck was 377,814, 225,000, and 252,120\$, respectively (Edwards, 2014; Stinger Inc., 2018). The crew composition (number of equipment) for this analysis was 10, 1, and 6 for baling, in-field transportation and road transportation, respectively. The crew composition (number of equipment) was determined using trial and error to ensure that all farms were harvested within the harvest window.

# RESULTS

# Farm Tons Harvested, Discarded, and Transported

**Figure 6** shows farm metrics for total corn stover tonnage harvested, transported, discarded, and lost as dry matter. The model was set up to reflect all the contracted farms (40% within

stover to depots.

a 35-mile buffer radius, Berven, 2009) were harvested each year (460,000 tons). In year one of the bale-delivery (biorefinery uptime: 20–85%) scenario, all harvested tonnage was transported because there was no storage at the beginning of the simulation run. In the second year, since storage tonnage was at its maximum capacity and uptime at the biorefinery was low (37%), only 50% of the harvested tonnage was transported (**Figure 6**). The tonnage transported from the farms increased as production at the biorefinery increased. The number of tons discarded at the field edge was high (50%) in the second year because of low throughput at the biorefinery and a short storage window (3 months) at the field edge. The model does not account for cost of discarding corn stover at the field edge before the beginning of next corn planting season. In this analysis we assume that corn stover is discarded however in reality it could be available for alternative markets such as animal feeding and bedding. This tonnage discarded at the field edge decreased as biorefinery uptime increased and storage space became available. A similar trend was observed in the pellet-delivery scenario-2 (depot uptime: 20–85%, biorefinery uptime: 85%). Dry matter loss was low because corn stover was stored at the field edge for a maximum of 3 months (**Figure 6**). For scenario-3 (pellet-delivery (depot uptime: 85%, biorefinery uptime: 85%), an average of 6,479 tons of stover discarded (**Figure 6**) could be attributed to less available bale storage area at the depot, resulting in more bales stored at the field edge and not transported to the depot before the start of the next corn planting season. Three ways to reduce tonnage discarded at the field edge are the following:

(Depot uptime: 85%, Biorefinery uptime: 85%) scenarios.

(A) the processing facility should have higher uptime; (B) fewer farms should be contracted at the beginning, and the number subsequently increased as uptime increases; (C) determining alternative corn stover market.

#### Inventory

Inventory management is crucial to provide a buffer against biomass supply fluctuations, but it also adds infrastructure and maintenance costs. Since inventory is a biorefinery's largest asset to protect against supply fluctuations, a key decision for a biorefinery is how much inventory to keep on hand. It is recommended that biomass supply inventories should be at 110– 130% of the biorefinery nameplate capacity to ensure a yearround feedstock supply and to buffer against biomass supply risks (Darr and Shah, 2012; Darr et al., 2014). For bale-delivery (biorefinery uptime: 20–85%) scenario-1, the storage site capacity was 115% of biorefinery demand, which takes into account dry matter losses in the system. The bale inventories at the storage sites and the biorefinery on-site storage site were 435,000 and 25,000 tons, respectively. At the biorefinery maximum 2 weeks of corn stover storage was considered in the model (Wirt). No yield variation over the years was assumed in the model hence additional biomass storage to buffer against supply uncertainties was not required. In pellet-delivery scenario-2 (depot uptime: 20–85%, biorefinery uptime: 85%) and pellet-delivery scenario-3 (depot uptime: 85%, biorefinery uptime: 85%), the bale storage site capacity was 67% of the biorefinery demand, determined by running simulations to ensure that the biorefinery would not be starved in scenario-3. This is in agreement with the finding that establishment of a pellet-delivery system reduces storage cost and storage size (Lamers et al., 2015b).

The storage sites were empty at the beginning of the first year in each scenario run, and corn stover inventory built up during the 32-day harvest window, then decreased depending on facility uptime (depot and biorefinery) (**Figure 7**). In the first year of bale-delivery (biorefinery uptime: 20–85%) scenario-1, the maximum bale (tonnage) inventory grew to about 85% of the biorefinery demand, while during the non-harvest time-period, the biorefinery removed corn stover from the inventory, reducing storage tonnage to 64% of biorefinery demand. In the second year of harvest, the inventory began building up during the harvest season and reached its maximum capacity of 435,000 tons. It then stays at that level for about 90 days due to low uptime at the biorefinery which utilizes 25,000 tons of corn stover from onsite storage. As biorefinery uptime increases, the minimum inventory level decreases and the inventory remains at the maximum level for a shorter period (**Figure 7**). A similar inventory trend was observed for pellet-delivery (depot uptime: 20–85%, biorefinery uptime: 85%) scenario-2 and pellet-delivery (depot uptime: 85%, biorefinery uptime: 85%) scenario-3 (**Figure 7**). In the pelletdelivery scenario, as the system stabilizes (85% uptime at the depot and biorefinery) a safety stock of 50,000 tons (∼1-month biorefinery demand) was maintained to mitigate risk of depot supply disruptions. The maximum tonnages of pellets stored in pellet-delivery scenario-2 and pellet-delivery scenario-3 were 2,304 and 2,382 tons, respectively. These values were low because the consumption rate at the biorefinery matched the rate of production at the depot. We demonstrated storage behavior across multiple years with varying uptime at the biorefinery, and in the future we will investigate the role of storage in mitigating supply under risk related to feedstock supply uncertainty.

**Figure 8** shows the total number of truckloads delivered both to the biorefinery and to intermediate storage or depot sites. Each truckload can transport 21 or 44 tons, respectively, of bales or pellets, and a significant transportation effort is required to move the large square bales from the farm boundaries to the intermediate storage/depot locations and then move bales/pellets from the intermediate (storage/depot sites) to the biorefinery. The average maximum number of semi-loads of baled corn stover transported from the farms to the intermediate sites (storage/depot sites) was 20,000. A similar estimate was reported by McGill and Darr (2014) for a biorefinery with a biomass demand of 400,000 dry tons. The average maximum number of semi-loads of bales and pellets transported from bale storage to biorefinery, and from depot to biorefinery, were 17,673 and 9,596, respectively. The relative intensity of truck traffic decreases in the pellet-delivery scenarios because of more efficient transport of high-density pellets. As biorefinery uptime increases, the number of semi-loads transported from intermediate sites (storage/depot) to the biorefinery also increases (**Figure 8**).

In this study, we assumed two storage sites in the bale-delivery scenario and these same storage sites were taken as depot sites in the pellet-delivery scenario to enable direct comparison between the scenarios and determination of the impact of operational disruptions. This assumption resulted in large bale storage sites being included in the model, although storage of large volume biomass increases fire risk, infrastructural requirements, and seasonal road traffic (Sahoo and Mani, 2017). In the future, we will examine the impact in meeting biorefinery cost, quantity, and quality specifications of the number of storage sites, their location, size, layout, and safety stock. We will also consider local fire code and safety regulations.

#### Number and Type of Crew Required

The number of crew required in each scenario is based on corn stover demand at the biorefinery, capacity of crew type, length of harvest season, length of time bales stored at field edge, daily working hours, distance between farms, and time required to move crews between the farms (Ebadian et al., 2017; Wang et al., 2017). It was assumed that bales could be stored at field edge for about 3 months until the beginning of preparation for the next corn planting season.

The number of baling, in-field transportation, and road transportation crews required for the bale-delivery (biorefinery uptime: 20–85%) scenario as estimated by the model were 430, 34, and 32, respectively. Similarly, baling, in-field transportation, and road transportation crews for pellet-delivery (depot uptime: 20–85%, biorefinery uptime: 85%) scenario-2 and pellet-delivery (depot uptime: 85%, biorefinery uptime: 85%) scenario-3 were 440, 30, and 24, respectively. In Iowa, the harvest window is very short, varying between 27 and 32 days (Darr and Webster, 2014; Shah and Darr, 2016), impacting the fleet size and associated costs. The crew compositions were determined using a trialand-error approach to ensure that all contracted farms would be harvested within the harvest window. The fleet sizes varied among the scenarios according to their designs (storage sites vs. depots). It was assumed that the biorefinery owns the equipment fleet, but existing equipment owned by farmers or custom harvest groups could also be used for fleet operations. The large number of baling crews (430–440) was mainly attributed to low baler productivity and the narrow harvest window (32 days).

In bale-delivery (biorefinery uptime: 20–85%) scenario-1, 460,000 tons of corn stover harvested in the first year were transported to the bale storage sites by the bale road transportation crew (**Figure 9**). Since the simulation began with an empty system, in the first year all harvested corn stover was transported to storage, while in the second year, all contracted acres were harvested, but biorefinery uptime was low, resulting in underutilization of the road transportation crew. The crew utilization did increase as biorefinery uptime increased. In pellet-delivery (depot uptime: 20–85%, biorefinery uptime: 85%) scenario-2, 286,977, and 156,433 tons, respectively, of corn stover were transported in the first and second years. The relatively low transportation tonnage was due to smaller bale storage size in the pellet-delivery scenario. As biorefinery uptime increases, road transportation crew utilization increases. In bale-delivery scenario-1, comparing the available 3,285 h per year (365 days per year, 9 days per hour) with actual task hours, the harvest and in-field transportation crew utilization levels were only 7.6 and 6.9%. Similarly, farm-to-storage and storage-to-biorefinery average crew utilization levels were 14.9 and 43.3%, respectively. Since we assumed that the biorefinery owns the equipment, low equipment utilization would represent an economic burden on the biorefinery and increase the per-unit biofuel cost. Achieving an optimal combination of custom harvest groups, farmer-owned equipment, and biorefinery-owned equipment could serve as a cost-cutting strategy for the biorefinery.

#### Facility Metrics

To understand the impact of operational disruptions on the biomass supply system, we present facility metrics (processed, starved, failed, warm-up and scheduled downtime) at the depot and biorefinery for three scenarios in **Figure 10**. In year one of the bale-delivery (biorefinery uptime: 20–85%) scenario-1, the uptime at biorefinery was only 20%; therefore, downtime was high and the amount processed at the biorefinery was low. The biorefinery failures were assumed to decrease over time. As the biofuel industry matures, improved equipment for feedstock handling and consistent quality attributes could resolve the failure issues. As uptime at biorefinery increased, the failed time

decreased, thus tonnage processed at the biorefinery increased (**Figure 10**). There was no starved time at the biorefinery as adequate corn stover was available for processing. Biomass supply uncertainty due to extreme weather events, yield and quality variability, and supplier reliability were not considered in this analysis.

Downtime at facility not only causes loss of production, it also impacts product quality, loss of customers, and increased costs. Downtime at a biorefinery is mainly due to biomass handling problems (flowability) and inconsistent biomass quality attributes (ash, moisture). Preprocessing of biomass at depots to produce uniform format feedstock "commodities" is strategy proposed to improve uptime at the biorefinery (Lamers et al., 2015b). Dedicated pre-processing depots do not eliminate the problem but add an entity (depots) in the supply chain to resolve the issue. In pellet-delivery scenarios, intermediate preprocessing depots were added to the supply chain. Reducing downtime and increasing uptime improved the efficiency of system (**Figure 10**). In pellet-delivery (depot uptime: 20–85%, biorefinery uptime: 85%) scenario-2, we observed a cascading effect of failures across facilities (depot to biorefinery). Even though the biorefinery uptime was 85% with depots in the system to address handling and quality issues, depots started with higher failure rates, thus starving the biorefinery. This indicates that having a depot is only justified if they consistently have higher uptimes and the cost of having a depot outweighs overall supply benefits. Resolving biomass handling and quality challenges could eliminate the need of having depots. In pellet-delivery scenario-3 (depot uptime: 85%, biorefinery uptime: 85%), the uptime at the depot and biorefinery were same, thus we observed stable system and consistency among processed tonnage at both facilities.

We recommend performing a system-level reliability analysis incorporating failure dependencies among subsystems. Biomass quality and handling are key issues for operational disruptions at biorefinery. Quality issues could be addressed at farm level. For example, farmers adapt harvesting processes that minimize ash and moisture levels in corn stover. In addition, farmers should be fully aware and educated about quality metrics, including ash and moisture, required by the biorefinery. They should realize if quality metrics are not met there could be price penalties or rejected loads. Feedstock handling is other major issue limiting processing capacity at the biorefinery. A comprehensive understanding and quantification of biomass physical properties to enable design of equipment and handling systems at the biorefinery is a potential solution to feedstock flowability challenges. In addition, detailed cost-benefit analysis for depots should be conducted.

#### Costs

**Figure 11** shows average delivered cost based on tonnage processed at the biorefinery for three scenarios. The delivered cost, for the bale-delivery (biorefinery uptime: 20–85%) scenario, averaged to \$380/ton in the first year. As biorefinery tonnage processing increased, the delivered costs decreased. In the years five through seven when biorefinery uptime was assumed to

FIGURE 11 | Average delivered cost per ton corn stover processed (\$/ton) for the bale-delivery (Biorefinery uptime: 20–85%), pellet-delivery (Depot uptime: 20–85%, Biorefinery uptime: 85%) and pellet-delivery (Depot uptime: 85%, Biorefinery uptime: 85%).

be 85%, the average delivered cost was \$99/ton. Harvest cost was the biggest contributor to the average delivered costs. In the model, it was assumed that all the contracted farms are harvested every year and the equipment fleet is owned by the biorefinery. Large number of balers were required to complete the harvesting operation within 32 days due to low machine capacity of balers. This mainly contributed to harvest being the largest cost contributor. In the pellet-delivery (depot uptime: 20-85%, biorefinery uptime: 85%) scenario-2, the delivered cost followed the same trend, decreasing over time with increasing production at the biorefinery. In the years five through seven when uptime at depot and biorefinery was assumed to be 85%, depot cost contributed 26% to the total delivered cost. The storage cost decreased as less tonnage was stored. In the pelletdelivery scenario-3 (depot uptime: 85%, biorefinery uptime: 85%), the average delivered cost was found to be \$116/ton. The delivered cost for the pellet-delivery scenario was about \$16/ton greater than the bale-delivery scenario with uptime at facilities (biorefinery/depot) being 85%. This analysis considers the production costs and does not account for depot and biorefinery capital costs. Pre-processing depots may be necessary to achieve quality, quantity, and reliability required for the growth of a US bioeconomy. However, detailed analysis should be conducted to evaluate if overall supply benefits of preprocessing depots outweighs costs.

#### CONCLUSIONS AND IMPLICATIONS

We have presented a simulation-driven study for analyzing the operational reliability of the biomass-to-biorefinery supply chain. Our approach considers the end-to-end biomass supply chain system as a series of connected components. It estimates independent failure modes for each facility (depot and biorefinery) and tracks how these failures then propagate throughout the entire system. As a case study, we evaluated our simulation framework for a biorefinery with corn stover demand of 335,000 dry tons and facility (depot/biorefinery) failure data varying each year. This allows us to investigate and gain insights on the impact of operational disruptions on costs, facility metrics, and inventory. Our main contribution is the development and implementation of a database centric discrete event model that simulates multi-biomass, multiform, multi-product, and multi-year biomass supply chains for operational level decision making. We found cascading effects of failures propagating through the system from depot to biorefinery. Operational disruptions caused by biomass quality and handling issues needs to be addressed. If pre-processing depots are required to achieve quality, quantity, and reliability, detailed analysis should be conducted to evaluate how overall supply benefits outweigh costs. Harvest cost was found to be the biggest contributor to the delivered costs. Achieving an optimal combination of custom harvest groups, farmer-owned equipment, and biorefinery-owned equipment could serve as a cost-cutting strategy for the biorefinery.

#### NOTICE OF COPYRIGHT

This manuscript has been authored by UT-Battelle, LLC, under contract DE-AC05-00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-publicaccess-plan).

## AUTHOR CONTRIBUTIONS

BS and RC developed the model, designed the scenarios, and performed model runs. BS with help from MH analyzed and interpreted the modeling results. BS prepared the manuscript with input from all authors. EW conceived the study and was in charge of overall direction and planning.

# FUNDING

This material is based upon work supported by the U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Bioenergy Technologies Office.

#### ACKNOWLEDGMENTS

This project was funded by United States Department of Energy, Bioenergy Technologies Office. We thank Dr. Mahmood Ebadian (Biomass and Bioenergy Supply Chain Specialist), Magen E. Shedden (Post-masters: Biomass logistics), Devita D. Amal (Ph.D. Candidate: Biomass supply risk) in the Environmental Sciences Division, Oak Ridge National Laboratory and Amit Khanchi (Post-doctoral Research Associate: Biomass postharvest), Iowa State University for their scientific guidance and reviewing the manuscript.

#### REFERENCES


Bai, Y., Li, X., Peng, F., Wang, X., and Ouyang, Y. (2015). Effects of disruption risks on biorefinery location design. Energies 8:1468. doi: 10.3390/en8021468


economic case for cropland diversification. Environ. Res. Lett. 11:014009. doi: 10.1088/1748-9326/11/1/014009


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Sharma, Clark, Hilliard and Webb. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Effect of Non-Structural Organics and Inorganics Constituents of Switchgrass During Pyrolysis

#### Pyoungchung Kim<sup>1</sup> \* † , Choo Hamilton<sup>1</sup> , Thomas Elder <sup>2</sup> and Nicole Labbé<sup>1</sup> \*

<sup>1</sup> Center for Renewable Carbon, The University of Tennessee, Knoxville, TN, United States, <sup>2</sup> U.S. Forest Service (USDA), Auburn, AL, United States

#### Edited by:

Allison E. Ray, Idaho National Laboratory (DOE), United States

#### Reviewed by:

Selhan Karagoz, Karabük University, Turkey Halil Durak, Yüzüncü Yil University, Turkey Tianju Chen, Qingdao Institute of Bioenergy and Bioprocess Technology (CAS), China

#### \*Correspondence:

Pyoungchung Kim pyoungchung@gmail.com Nicole Labbé nlabbe@utk.edu

#### †Present Address:

Pyoungchung Kim, TerraPower, LLC, Bellevue, WA, United States

#### Specialty section:

This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research

Received: 10 April 2018 Accepted: 31 August 2018 Published: 15 October 2018

#### Citation:

Kim P, Hamilton C, Elder T and Labbé N (2018) Effect of Non-Structural Organics and Inorganics Constituents of Switchgrass During Pyrolysis. Front. Energy Res. 6:96. doi: 10.3389/fenrg.2018.00096 Non-structural components, such as inorganics and organic extractives, present in switchgrass were extracted with water and ethanol, and the resulting non-structural components-free materials were pyrolyzed to investigate the effect of the inorganic species on the pyrolytic products. The extraction was performed for switchgrass materials harvested from three consecutive growing seasons, removing 8.5 wt% of the organic extractives in the first season biomass, and 5.8 and 6.3 wt% in the second and third season, respectively, on total dry basis of biomass. In addition to organic extractives, from 0.7 to 2.7 wt% of ash were extracted. Specifically, 99% and 59% of total K and Mg were removed from the switchgrass harvested in the second and third growing season. Thermogravimetric analysis demonstrated that a predominant reduction of K and Mg content in the biomass increased temperature at which mass loss rate is maximized in the decomposition of cellulose, hemicellulose, and lignin. The reduction of K and Mg content also affected pyrolytic products generated at 450◦C. The chromatographic peak area percentage of levoglucosan from the extracted samples in the second and third growing season was two to three times higher than that from the extracted samples in the first growing season, showing a strong negative correlation with K and Mg content, whereas most of the light oxygenated and furanic/pyranic/cyclopentanic products exhibited a strong positive correlation with K and Mg content. We concluded that the removal of non-structural components prior to thermochemical conversion processes such pyrolysis can potentially produce a valuable extractives stream while removing catalytic inorganics that negatively impact downstream pyrolysis process.

Keywords: switchgrass, non-structural components, extractives, catalytic pyrolysis, ash, inorganics

#### INTRODUCTION

A wide variety of lignocellulosic biomass is under consideration as renewable resources, including woody and herbaceous biomass, to produce chemicals, biomaterials, and biofuels using various thermochemical and biochemical conversion technologies. Of these technologies, pyrolysis is a thermochemical conversion that liquefies biomass in the absence of oxygen at mild operating temperatures (450–600◦C) (Kim et al., 2014). Conventionally, pyrolysis is used to directly convert biomass to a high yield bio-oil (60–75%) as well as a biochar fraction (15–25%) and noncondensable gases (10–20%) similar to syngas (Bridgwater, 2012). Bio-oil can be upgraded to biofuels and chemicals through hydrotreating or catalytic processes while non-condensable gases

**204**

can be burned for combined heat and power (CHP). In addition, biochar can be used as soil amendment or alternatively be combusted for CHP. However, during thermochemical conversion, ash naturally present in the biomass can damage processing systems by slagging and fouling of reactors and pipes as well as influence the composition and quality of the pyrolytic products (Froment et al., 2013). The most abundant alkali and alkaline earth metallic species, such as potassium, calcium, sodium, and magnesium in ash, act as a catalyst for cellulose and hemicellulose depolymerization and fragmentation, and generate a high yield of light oxygenated compounds such as acids, ketones, aldehydes, and furans (Lian et al., 2012). The large amount of oxygenated compounds present in bio-oils has been found to deactivate catalysts by coke deposition during refinery upgrading processes (Bridgwater, 2012) and to restrain ethanol production by inhibiting the growth of fermentative microorganisms (Lian et al., 2012). Therefore, many efforts have been devoted to reduce inorganics present in biomass via water, mild, or strong acid pretreatment prior to fast pyrolysis and to promote the yield and quality of bio-oil in the aspect of heating value, viscosity, and water content. Bio-oil also contains high amount of levoglucosan due to the decrease of glycolaldehyde yield after water or acid pretreatment (Mourant et al., 2011; Wang et al., 2012). Levoglucosan is a useful starting chemical that can be converted into platform chemicals such as levulinic acid, levulinic esters, and 5-(hydroxymethyl) furfural (Yin et al., 2016). Levoglucosan can also serve as a precursor for bioethanol production through hydrolysis followed by fermentation (Luque et al., 2016). However, high severity pretreatment conditions, including high temperature, long reaction time, and high acid concentration, lead to the decomposition of the biomass matrix through the hydrolysis of cellulose and hemicellulose (Eom et al., 2011).

Switchgrass (Panicum virgatum), a perennial grass native to North America, is an attractive candidate for bioenergy production (Kline et al., 2013). Compared to woody biomass, switchgrass is a fast-growing crop that can reach up to three meters in height in one growing season (Parrish and Fike, 2005). Switchgrass can be harvested either annually or semi-annually for ten years before it needs to be replanted, leading to one of the most dependable feedstocks with low production costs and high renewability (Parrish and Fike, 2005). The quality of biomass reflected by its chemical composition is a key factor that affects efficiency of bioenergy production. The chemical constituents of switchgrass, are generally structural polymers such as cellulose (30.0–43.9 wt%), hemicellulose (24.4–30.5 wt%), and lignin (18.6–25.4 wt%), and non-structural components such as extractives (4.7–15.6 wt%) and inorganic ash-forming elements (1.8–7 wt%) (Lemus et al., 2002; Kline et al., 2015). Interestingly, the content of extractives in switchgrass is similar or higher than that of woody biomass and is mostly composed of lipophilic and hydrophilic types such as waxes, oils, fats, resins, free sugars, chlorophyll, organic acids, alditols, and polyphenolics, making them a potential source of value-added co-products (Sannigrahi et al., 2010). These extractives are generally extracted with water and organic solvents such as ethanol, acetone, hexane, dichloromethane, or diethyl ether (Thammasouk et al., 1997; Uppugundla et al., 2009). For example, a standard nonstructural components extraction procedure using water and ethanol according to National Renewable Energy Laboratory (NREL) analytical procedure (NREL, 2008) extracts water soluble compounds such as non-structural sugars, inorganics, and ethanol soluble compounds such as chlorophyll, phytosterols, terpenes, and phenolic compounds, which have been reported to have a potential antimicrobial and antioxidant activity (Proestos et al., 2006; Uppugundla et al., 2009; Labbé et al., 2016).

Most of the aforementioned studies, however, have investigated the effect of removal of inorganics present in biomass via water or acid treatment on pyrolytic products, but have not taken the loss of non-structural components such as extractives into consideration although this biomass fraction, extractives, could be a potential high-value co-product if an integrated biorefinery is considered (Labbé et al., 2016). Therefore, the objective of this study was to evaluate the effect of non-structural inorganic species on pyrolytic products after the biomass, switchgrass, was subjected to an extraction step that was specifically designed to extract the organic extractives fraction of switchgrass. Switchgrass materials with a wide extractives and ash content range were initially treated with both water and ethanol, then analyzed by thermogravimetric analysis (TGA) and pyrolysis-gas chromatography/mass spectrometry (Py-GC/MS). The inorganic composition of switchgrass before and after extraction was determined to evaluate the catalytic effects of existing inorganic species on pyrolytic products using statistical methods such as principal component and Pearson's correlation analyses.

#### MATERIALS AND METHODS

#### Material

Alamo switchgrass samples were collected from a University of Tennessee-Knoxville experiment field established in 2011.

Eight switchgrass samples were harvested for three consecutive growing seasons in December of 2011 (year 1), 2012 (year 2), and 2013 (year 3). A total of 24 switchgrass samples were collected, dried, then ground to less than 0.42µm particle size using a Wiley mill (Thomas Scientific). The samples are referred as S1 to S8 for year 1, S9 to 16 for year 2, and S17 to S24 for year 3.

#### Sample Extraction and Characterization

The ground samples were extracted using an accelerated solvent extractor (ASE 350, Dionex) to remove non-structural components, i.e., extractives and inorganics, according to the National Renewable Energy Laboratory (NREL) protocol (NREL, 2008). Approximately 5 g of the ground material were loaded with glass beads (1:5 wt/wt) in a 34 mL extraction

cell. The sample-loaded cell was heated up to 100◦C, then extracted with water by 3 cycles for 10 min/cycle, and followed by ethanol with 3 cycles for 10 min/cycle at 1600 psi of pressure. After extraction, the extractives-free samples were dried at 40◦C for 7 days before further analyses.

Ash content of the unextracted and the extracted samples was measured according to ASTM D 1762-84. The inorganic composition was measured by inductively coupled plasmaoptical emission spectroscopy (ICP-OES, Optima 7300 DV spectrometer, PerkinElmer) after acid microwave digestion of the material (Kim et al., 2011).

The thermal behavior of the sample was investigated by thermogravimetric analysis (TGA) and pyrolysisgas chromatography/mass spectrometry (Py-GC/MS). A thermogravimetric analyzer (TGA Q50 TA Instrument) was employed to monitor weight loss as a function of temperature and to evaluate thermal decomposition of the unextracted and extracted switchgrass over time. Approximatively 6–9 mg of the sample were heated from 30 to 110◦C at a rate of 20◦C/min and held for 10 min to remove moisture from the sample under nitrogen gas flow (20 mL/min). Then, the sample was heated to 550◦C at 20◦C/min and held for 15 min. The resulting TG and differential TG (DTG) curves described the thermal characteristics of each sample.

The unextracted and extracted switchgrass samples were also analyzed using a micro-pyrolyzer (Frontier EGA/Py-3030 D) fitted with a Perkin Elmer Clarus 680 Gas Chromatograph coupled with a Clarus SQ 8C Mass Spectrometer. Approximatively, 0.2 mg of sample in a stainlesssteel pan was injected into the pyrolysis furnace which was heated to 450◦C for 12 s of residence time. The vapor produced in the furnace was swept into a GC-injection port (temperature 270◦C) by the GC carrier gas (helium, 1 mL/min) that passes through the furnace with a split ratio of 80:1. The vapor was then separated using a DB-1701 Agilent capillary column (60 m × 0.25 mm ID × 0.25µm film thickness). The GC oven temperature program was as follows: 4 min at 50◦C, ramp 5◦C per minute to 280◦C, and hold at 280◦C for 4 min. The generated compounds were analyzed using a mass spectrometer with a source temperature of 270◦C and a 70 eV electron ionization. Chromatographic peaks with S/N > 2000, including identifiable and non-identifiable peaks, were extracted using the TurboMass GC/MS software. These peaks were normalized to the peak area percentage based on the sum of the total peak areas. A total of 91 organic compounds were selected for statistical analysis.

#### Statistical Analysis

The relationships between the inorganics present in switchgrass and the resulting pyrolytic organic products were investigated using multivariate statistical analysis as well as Pearson's correlation analysis. Normalized peak areas % of the 91 organic compounds identified in each pyrogram were analyzed using principal component analysis (PCA, Unscrambler software ver. 9.0 CAMO) to determine similarities and differences in the pyrolytic organic compounds produced from the samples containing various levels of inorganics. PCA is a statistical procedure that transforms a large set of variables into a smaller set in a new system of axes of principal components (PCs). The PCA results show the trend of similarity and difference of the samples and the variables in the plot (Saporta and Niang, 2009). The scores plot visualizes the samples and helps to interpret the relationships among the samples while the loadings plot presents the contribution of each variable (chromatographic peaks) to each new PC and describes the relationship among the variables. The Pearson correlation coefficient (r) was calculated using the SigmaPlot 12 software and was considered significant when the p < 0.05.

### RESULTS AND DISCUSSION

## Non-structural Components of Switchgrass

A total of twenty four switchgrass samples, eight samples collected for three consecutive years, were extracted with water and ethanol (**Figure 1**). Non-structural components, including organic and inorganic constituents, were extracted from the biomass. The averaged extraction was 9.6 wt% for the first growing season biomass, 6.8, and 7.5 wt% for the second and third growing season biomass, respectively, on a dry basis of total

TABLE 1 | Pearson's correlation between inorganic compounds and temperatures at derivative TG peaks.


a r represents correlation coefficient.

<sup>b</sup>p represents a significant correlation if p < 0.05.

<sup>c</sup>ND represents no correlation (p > 0.05).

biomass. Organic extractives constituted a large portion of the extracts, with an average of 8.5 wt% in the first year, then 5.8 and 6.3 wt% in the second and third year, respectively. Many researchers have been highly interested in biomass extractives, including flavonoids such as rutin and quercetrin and phenolic compounds such as p-coumaric, sinapic, and ferulic acids, that are potential sources of valuable co-products and can be used in the pharmaceutical and cosmetic industries providing an additional income stream for biorefineries producing biofuel in a more integrated manner (Uppugundla et al., 2009; Michels and Wagemann, 2010). In addition to these valuable organic components, inorganics naturally present in switchgrass were also extracted, which corresponds to a mass loss ranging from 0.7 to 2.7 wt% (**Figure 1**). The content and composition of these fractions are known to vary with the age, genetics, tissue types, and harvesting method and time of the biomass (Kline et al., 2013). The switchgrass samples harvested for three consecutive years had very similar chemical composition including cellulose of 39.1–39.3%, hemicellulose of 29.4–31.2%, and lignin of 21.4– 21.7%, respectively. The reduction of non-structural components in switchgrass samples over the course of three years can be explained by the fact that switchgrass reached a more steady-state growth pattern in the second and third year (Baxter et al., 2016).

#### Inorganic Compounds in Switchgrass

As expected, the unextracted (untreated) switchgrass samples had a higher ash content with an average of 7.0 wt% in the first year and a 3.5 and 3.2 wt% in the second and third year, respectively (**Figure 2A** and **Supporting Information Table A1**). The extracted switchgrass (treated) samples had an average ash content of 5.9 wt% with a decrease of 16% compared to the unextracted samples in the first year, and 2.4 and 2.1 wt% with a removal of 33% in the second and third year, respectively. These data demonstrated that the solvent extraction could remove a fraction of the inorganic compounds, between 16 and 33% of the total ash present in switchgrass. Alkali and alkaline earth metals (AAEM), such as calcium, magnesium, and potassium, were the most abundant constituents found in the untreated switchgrass samples (**Figure 2B** and **Supporting Information Table A1**). For example, the first growing season switchgrass contained an average of 7,103 mg of K per kg of biomass, 5,261 mg/kg of Ca, and 2,559 mg/kg of Mg whereas the biomass in the second and third year had lower of K (an average of 4,923 and 3,931 mg/kg), Ca (2,018 and 1,522 mg/kg), and Mg (1,590 and 1,411 mg/kg), respectively. After the solvent extraction, the concentration of K in the extracted switchgrass samples was predominantly reduced to an average of 370 mg/kg, a 95% removal compared the untreated samples in the first year, and to 40 and 28 mg/kg with 99% removal for the second and third year biomass, respectively. The concentration of Mg decreased by more than 50% after the extraction; with 55% removal in the first year and 58% removal for the second and third year biomass. However, Ca was slightly extracted from the switchgrass with only a 12 and 9% removal for the second and third year, respectively.

The concentration of other minor inorganics, such as Al, Fe, and Mn present in switchgrass was slightly reduced with removal ranging between less than 1 and 28%. The extraction of silica (Si) was also low with the highest removal at 8%. These findings indicated that salts present in biomass could be extracted with water and ethanol extraction, whereas inorganics structurally bound to the carboxylic and/or phenolic groups in the organic matrix still remain in biomass (Nik-Azar et al., 1997).

#### Thermogravimetric Analysis

The thermal degradation profile of switchgrass and the corresponding extractives-free biomass were obtained by TGA from 30 to 550◦C in order to investigate the biomass decomposition temperature in function of ash content (**Figure 3**). The TG thermograms exhibited a shift of initial degradation temperature, at which drastic mass loss started, toward higher TG temperatures as the unextracted samples were extracted with water and ethanol (**Figure 3A**). Derivative TG (DTG) curves were calculated from the TG thermograms and presented the thermal decomposition peaks corresponding to temperatures at which the mass loss rate was maximum (**Figure 3B**) according to the devolatilization temperatures determined as in Grønli et al. (2002). The initial degradation temperature (Tonset) at which hemicellulose decomposition starts (**Figure 4A**) demonstrated that as ash content in the unextracted switchgrass samples decreased from 7.0 wt% in


the first year to 3.1–3.6 wt% in the second and third year, Tonset shifted to higher temperature, from 223 to 251◦C showing a strong negative correlation (r = −0.85) between the two parameters (**Table 1**). The decomposition temperature (Tshoulder) of hemicellulose (**Figure 4B**), at which hemicellulose decomposition is maximum, moved to higher temperature (from 284 to 312◦C), with a mild negative correlation of −0.56 with ash content. The decomposition temperature (Tpeak) (**Figure 4B**), where the maximum cellulose decomposition rate occurs, shifted from 323 to 346◦C also showing a weaker negative correlation (r = −0.42) with ash content. The beginning temperature, Toffset, of the final tailing region, dominated by the lignin decomposition, slightly moved from 364 to 379◦C with no correlation with ash content (**Figure 4A**).

The DTG curves of the extracted switchgrass samples exhibited a shift of hemicellulose (Tonset and Tshoulder), cellulose (Tpeak), and lignin (Toffset) decomposition peaks toward higher temperatures due to the reduction in ash content, demonstrating strong negative correlations between ash and these decomposition temperatures (**Table 1**). Interestingly, the extracted samples with high ash content of 5.9 wt% in the first year biomass exhibited a shift of the DTG peaks to temperatures higher than the unextracted switchgrass samples with low ash content ranging from 3.1 and 3.6 wt% in the second and

third year (**Figures 4A,B**). These results confirm that some specific inorganic species present in the unextracted switchgrass significantly affected the thermal decomposition of switchgrass structural components. Of the inorganic species affected by the solvent treatment, the major alkali and alkaline earth metals (AAEM), such as potassium and magnesium, were remarkably extracted with their content decreasing by 95–99% and 43–62 %, respectively. The drastic reduction of potassium and magnesium could be responsible for the shift of hemicellulose (Tonset and Tshoulder), cellulose (Tpeak), and lignin (Toffset) decomposition peaks toward higher temperatures (**Figures 4C–F**, **Table 1**). Therefore, it could be concluded that a reduction of potassium and magnesium content during an extractives extraction process significantly impacts the thermal decomposition of biomass confirming that alkali and alkaline earth metals act as catalyst that affect the rate of thermal degradation of biomass components in thermochemical processes (Kleen and Gellerstedt, 1995).

#### Py-GC/MS Analysis

Switchgrass samples with varying ash content and their corresponding extracted samples were pyrolyzed using a Py-GC/MS at 450◦C. A total of 91 organic compounds were identified to primarily estimate the changes of the yield of these volatile organic products (**Table 2**). Chromatographic peak area percentage of the identified organic compounds for the unextracted and extractives-free samples were subjected to a multivariate statistical technique, principal component analysis (PCA) (**Figure 5**). The PCA scores plot featured the distribution of the samples along PC1 and PC2 which accounted for 90 and 6% of the total variance of the chromatographic data set, respectively. The PCA scores plot (**Figure 5A**) demonstrated that the extractives-free switchgrass samples with low ash, including potassium and magnesium, clustered together in the negative quadrant of PC1 whereas the unextracted samples with high ash, potassium, and magnesium content, grouped in the positive section of PC1 (**Figure 5A**). The extracted samples from the first year biomass closely clustered to the untreated samples from the first year. This result can be explained by the presence of high ash content (5.9 wt%) and potassium (370 mg/kg), magnesium (1147 mg/kg), calcium (5187 mg/kg) in the extracted samples in the first year, which still participates in catalytic pyrolysis similar to that of the untreated samples in the first year. Meanwhile, the untreated and extracted samples in the second and third year were far apart from each other due to a significant reduction of inorganic compounds with low ash content (2.1–2.4 wt%), potassium (33–35 mg/kg), and magnesium (577–686 mg/kg) in the extracted samples.

The loadings plot of PC1 identified the chromatographic peaks that contributed the most to the separation along PC1 (**Figure 5B**) and revealed that during pyrolysis the extractivesfree samples with lower potassium and magnesium generated higher amount of mainly pentanedioic acid (**37**), 4-ethylguaiacol (**61**), 3-ethyl-3-heptanol (**74**), and levoglucosan (**86**). Whereas, the unextracted samples with higher ash content produced higher amount of low molecular weight species such as carbon dioxide (**1**), acetic acid (**10**), acetol (**11**), butanedial (**21**),

consecutive years. (A) year 1, (B) year 2, (C) year 3.

1,4-butanediol (**57**), furans/pyrans/cyclopentenones such as 2 hydroxycyclopent-2-en-1-one (**35**), and dihydrobenzofuran (**66**) (**Figure 5B**) confirming again the catalytic role of the inorganics during switchgrass pyrolysis (Shen and Gu, 2009).

Chromatographic peaks of all identified pyrolytic compounds, accounting for higher than 90% of total peak area in all chromatograms, were integrated, expressed as a percentage of the total pyrogram area, and classified into light oxygenates, furans/pyrans/cyclopentenones, anhydrosugars, phenolics, and carbon dioxides (**Figure 6**). The first thermal decomposition step of carbohydrates (cellulose/hemicellulose fraction) during switchgrass pyrolysis is the generation of anhydrosugars such as mainly levoglucosan and 1,4:3,6-dianhydro-α-dglucopyranose, which are produced by competitive reactions through dehydration and cleavage of glycosidic bond during pyrolysis (Patwardhan et al., 2009). The anhydrosugars yield TABLE 3 | Pearson's correlation between inorganic compounds in the unextracted and extracted switchgrass samples and the pyrolytic organic compounds produced by Py-GC/MS.


a r represents correlation coefficient.

<sup>b</sup>p represents a significant correlation (p < 0.05).

<sup>c</sup>ND represents no correlation (p > 0.05).

of the extracted switchgrass samples was higher than that of unextracted samples (**Figure 6**). In particular, the amount of anhydrosugars from the extracted samples in the second and third year became two to three times more than that of the untreated samples in the same years. This finding can be explained by various amounts of inorganics present in the switchgrass samples. As aforementioned in the **Table 1**, the extracted samples in the first year still contained high ash content (5.9 wt%) after the solvent extraction whereas the extracted samples in the second and third year had low ash content of 2.1 and 2.4 wt%, respectively. Pearson's correlation analysis also demonstrated that the yield of levoglucosan (**Table 3** and supporting information **Figure A1**) increased with a reduction of K and Mg content showing a negative correlation (r = −0.70) whereas the dehydrated levoglucosan, 1,4:3,6-diahydro-a-Dglucopyranose (Gardiner, 1966), decreased with a reduction of K and Mg contents with a positive correlation value of 0.80. Furthermore, the extracted samples in the second and third year had a small residual amount of K and Mg with a significant removal efficiency, which only interacts with the formation of levoglucosan from carbohydrates and do not have any effect on its secondary degradation into lower molecular compounds such as light oxygenated compounds and furans/pyrans/cyclopentenes (**Figure 6**). In particular, significant removal (97–99%) of potassium, which is known to possess a stronger catalytic strength than magnesium and calcium (Muller-Hagedorn et al., 2003), from switchgrass harvested in the second and third year via water-ethanol extraction significantly promoted the yield of levoglucosan (**Figure 6**).

As the yield of anhydrosugars increased with decreasing inorganics content, the amount of low molecular weight oxygenates such as acetic acid, butanedial, acetol, and other low carbon-atom fragments gradually decreased, exhibiting a stronger positive correlation with K (r = 0.70) than with Mg (r = 0.53) content (**Figure 6**, **Table 3**, and supporting information **Figure A2**). These light organic compounds are promoted by fragmentation of glucose rings, known as ring-scission (Zhou et al., 2014), at the loss of glycosidic bond breakage for levoglucosan formation in the presence of alkali and alkaline earth metals. The yield of furanic/pyranic/cyclopentanic compounds such as 2-furanmethanol, 3-methyl-1,2 cyclopentanediion, 2-methyl-3-oxy-γ-pyrone, and other compounds decreased with a reduction of inorganics content (**Figure 6**) showing a stronger positive correlation with K (r = 0.70) than Mg (r = 0.53) (**Table 3** and supporting information **Figure A3**). Moreover, 4-hydroxydihydro-2(3H)-furanone and 3,5-dimethoxycyclohexanol exhibited a strong negative correlation with K and Mg indicating the fate of the catalytic depolymerization reaction of carbohydrates in the presence of these inorganic species (Eom et al., 2011). Non-condensable gases (NCGs), such as carbon dioxide, also exhibited a strong positive correlation with K and Mg (**Table 3**). Therefore, a reduction of K and Mg via a simple solvent extraction process resulted in increasing yields of anhydrosugar-derivatives and decreasing low molecular weight species and NCGs. The yield of lignin-derived phenolic compounds decreased with a reduction of inorganic compounds (**Figure 6**). Specifically, monomeric phenolic compounds, such as phenol, guaiacol, and other compounds (Supporting information **Figure A4**), showed a strong positive correlation with K (r = 0.73) and Mg (r = 0.70), whereas 4-ethylguaiacol and 2-methoxy-5-[(1E)]-1-propenyl]phenol had a strong negative correlation (**Table 3**). These results indicated that inorganics promote the decomposition of the side chains of lignin structures via demethylation, demethoxylation, and dehydration reactions (Kleen and Gellerstedt, 1995; Mourant et al., 2011).

# CONCLUSIONS

This study concluded that a non-structural components extraction step using water and ethanol produced valuable extractives while simultaneously resulting in removing undesirable catalytic inorganics, which ultimately improves pyrolytic bio-oil quality. The organic extractives were extracted up to 8.5 wt% in the first season biomass and between 5.8 and 6.3 wt% in the second and third season biomass, respectively, on the biomass dry basis. During the extraction step, ash removal ranged from 0.7 to 2.7 wt% with a predominant reduction of potassium and magnesium content in switchgrass. Pyrolysis of the extractives-free switchgrass harvested from the second and third season produced anhydrosugars two and three times more than that from the first season and reduced the yield of furanic/pyranic/cyclopentanic and light oxygenated compounds. From this study, we demonstrated that the removal of nonstructural components increases the quality of switchgrass by decreasing its total ash content, specifically potassium and magnesium content. This approach could be applied to any biomass that contained detrimental amount of inorganics. The implementation of such step prior to a pyrolysis process of lignocellulosic biomass has a potential to produce valuable extractives while improving the quality of the feedstock for thermochemical processes.

# AUTHOR CONTRIBUTIONS

PK performed the extraction, collected the PyGC/MS data, investigated the correlation between the different set of data, and wrote the first draft of the manuscript. CH collected the ash and inorganic data and edited the manuscript. TE collected the TGA data and assisted in writing the manuscript. NL designed the study, contributed in data analysis, and was heavily involved in revising the manuscript. All authors approved the final document.

# ACKNOWLEDGMENTS

The authors would like to acknowledge the Southeastern Partnership for Integrated Biomass Supply Systems (IBSS), which is supported by Agriculture and Food Research Initiative Competitive Grant no. 2011-68005-30410 and 2013-67021-21158 from the USDA National Institute of Food and Agriculture, for support of this work.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00096/full#supplementary-material

# REFERENCES


**Conflict of Interest Statement:** Presently, PK is employed by TerraPower, LLC but he performed all the work while being employed by the University of Tennessee.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Kim, Hamilton, Elder and Labbé. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Wet Corn Stover Storage: Correlating Fiber Reactivity With Storage Acids Over a Wide Moisture Range

Dzidzor Essien<sup>1</sup> \*, Megan N. Marshall <sup>1</sup> , Tom L. Richard<sup>1</sup> and Allison Ray <sup>2</sup>

*<sup>1</sup> Department of Agricultural and Biological Engineering, Pennsylvania State University, University Park, PA, United States, 2 Idaho National Laboratory, Bioenergy Technologies Department, Idaho Falls, ID, United States*

Wet storage is synonymous with ensilage, a simple biotechnology that has been used to preserve forage for livestock feed for millennia. In this natural process, organic acids are produced by anaerobic microbial degradation of a small fraction of the biomass, and these acids reduce the pH to levels that minimize further microbial activity and can preserve the biomass for years as long as anaerobic conditions are maintained. These organic acids also result in mild pretreatment with potential to enhance downstream conversion processes, making this an effective storage strategy. However, the degree and significance of this natural pretreatment capability of ensiled storage on downstream processes has not previously been quantified across a range of storage conditions. In this study, the degree of pretreatment was investigated by measuring the reactivity of corn stover fiber to cellulolytic enzymes. Although the results indicated significant improvement in hydrolytic outcomes after wet storage, by a factor of up to 2.4, saccharification of cellulose to sugar monomers was still limited. The results also show that dominance of lactic acid in the ensilage process is key to wet storage (pretreatment) effectiveness as in the livestock feed industry. Lactic acid pKa value is lower than the pKa of other silage acids and lower than typical silage pH. This gives lactic acid the advantage of being in the more dissociated form, with more protons available to facilitate pretreatment hydrolysis. However, unlike the livestock feed industry, where quality feedstock is attainable within very narrow storage moisture range, for biofuel purposes, a wider range of 35–65% is appropriate in achieving a similar quality outcome. This is true both for the immediate fiber response to enzymes and with subsequent pretreatment. This wider moisture range implies more flexibility in harvest schedule without sacrificing feedstock quality, thus alleviating concerns over feedstock quality that biomass suppliers or biorefineries may have.

Keywords: corn stover, ensilage, wet storage, pretreatment, organic acids, hydrolysis, lactic acid, lignocellulose

# INTRODUCTION

Two main attractions of wet storage of biomass under anaerobic conditions are (1) minimizing dry matter loss, and (2) enhancing downstream pretreatment and conversion processes. In the first case, it is well-established that the production of organic acids and associated reduction of pH have bacteriostatic or bactericidal effects on most spoilage microbes, and the outcome is a

#### Edited by:

*Abdul-Sattar Nizami, King Abdulaziz University, Saudi Arabia*

#### Reviewed by:

*Ao Xia, Chongqing University, China Yu-Shen Cheng, National Yunlin University of Science and Technology, Taiwan*

> \*Correspondence: *Dzidzor Essien idd103@psu.edu*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *04 May 2018* Accepted: *25 September 2018* Published: *02 November 2018*

#### Citation:

*Essien D, Marshall MN, Richard TL and Ray A (2018) Wet Corn Stover Storage: Correlating Fiber Reactivity With Storage Acids Over a Wide Moisture Range. Front. Energy Res. 6:108. doi: 10.3389/fenrg.2018.00108*

**216**

well-preserved feedstock with reduced dry matter loss. For the second case, previous studies have suggested the potential of the organic acids produced during wet storage or ensilage to break some structural bonds or alter cell wall structure, thereby serving as a pretreatment mechanism. The purpose of pretreatment is to allow more effective enzymatic hydrolysis of sugar polymers into sugar monomers, as those sugar monomers can serve as a reactive intermediate for fermentations that produce biochemicals and biofuels (Tanjore and Richard, 2015). The various chemical, thermal, and biological pretreatment strategies that have been developed have high capital and operational costs, often on the order of one third of the total conversion cost (Brown and Brown, 2014; Eggeman and Elander, 2015). The idea of ensilage serving as an avenue for in situ pretreatment during biomass storage to enhance downstream fermentation processes was suggested by Linden et al. (1987) and Richard et al. (2001). Several more recent studies have documented promising effects (Chen C. et al., 2007; Digman et al., 2007; Ren et al., 2007; Thomsen et al., 2008; Oleskowicz-Popiel, 2010; Pakarinen et al., 2011), although the range of silage conditions and their impacts have not been fully explored. With few exceptions, storage moistures in these studies are within the range of traditional ensilage recommendations for livestock feed, while each study typically explores a single moisture level. See **Table 1** for a snapshot of storage conditions and key outcomes of these studies. Wet storage for bioenergy may require different recommendations than for livestock feeding, and the criteria for desired outcomes may also differ.

The pretreatment mechanism of organic acids during wet storage is hypothesized to be similar to that of dilute hydrochloric, sulfuric, or phosphoric acid, which act on glycosidic bonds as described below. These inorganic acids have been extensively explored for the pretreatment of lignocellulosic feedstocks under various temperatures, pressures and times (Jacobsen and Wyman, 2000; Tanjore and Richard, 2015; Kumar and Sharma, 2017). Various organic acids,—maleic, oxalic, succinic, fumaric, formic,—have also been explored for pretreatment of biomass (Kootstra et al., 2009; Lee and Jeffries, 2011; Marzialetti et al., 2011; Barisik et al., 2016). However, there are important differences between wet storage (organic) acid treatment and conventional (inorganic) pretreatment in terms of both temperature and time. In the case of inorganic acid pretreatment, the acid-amended biomass is exposed to high pressures and temperatures (typically 120–200◦C) for a period of minutes, whereas in wet storage the temperatures would be much lower (20–40◦C) but the duration much longer, on the order of months to years.

The reactivity of acids with biomass is random and targeted mainly at glycosidic bonds, which are found between and within the polysaccharide polymers, mainly cellulose, and hemicellulose chains. Dilute acid hydrolysis results in partial hydrolysis of glycosidic bonds, and can also result in degradation of lignin if temperatures are above 180◦C (Chesson, 1993; Donohoe et al., 2008). At ambient temperatures, which would be typical of most wet storage systems for biofuel production, acid hydrolysis is slow and limited to the amorphous regions where the hydrolytic reaction is initiated (Tímár-Balázsy and Eastop, 1998; Stoddart, 2007). Amorphous cellulose constitutes anything from 10 to 50% of total wood cellulose, depending on the feedstock (Jacobsen and Wyman, 2000). Cellulose is less susceptible to acid pretreatment because of its highly crystalline nature. In contrast, hemicellulose has a comparatively lower degree of polymerization and a highly branched, heterogeneous and fully amorphous structure that makes it more susceptible to acid reactions. Another factor contributing to hemicellulose's susceptibility is the presence of α-glycosidic bonds in the side chains, which are less stable than the β-bonds found in cellulose and the main chain of hemicellulose. The pretreatment capability of organic acids, produced under wet storage conditions, is therefore achieved mainly through hemicellulose hydrolysis.

A number of studies (McDonald et al., 1991; Muck, 1996; Richard et al., 2001; Ren et al., 2007) have shown the preferential degradation of hemicelluloses over cellulose during ensilage. The hydrolysis of hemicelluloses is an important pretreatment outcome of conventional acid or liquid hot water pretreatment, and the effectiveness of these pretreatment processes is associated with the amount of hemicellulose removed (Wyman, 1999; Sun and Cheng, 2002; Mosier et al., 2005). Conventional dilute acid pretreatment can result in up to 90% removal of xylose (the major constituent of hemicellulose), while liquid hot water pretreatment can remove up to 51% (Elander et al., 2005; Zheng et al., 2009). Since the structural sugars are bound or linked together throughout the plant cell wall, it can be expected that the degradation of hemicellulose will increase the surface area and allow enzymes access to other structural components through the openings created by their removal. Levels of hemicellulose degradation in silages can range from 0.5% or less (Muck, 1996) to as high as 54% depending on the crop type, storage conditions, and storage additives (Ren, 2006). These variations in hemicellulose degradation not only reflect the heterogeneity of biomass feedstocks, particularly the hemicellulose component whose constituents and arrangement vary from species to species, but also the sensitivity of wet storage to different conditions. This variability poses a potential challenge to wet storage as a complement to other pretreatment systems.

In addition to measures of hemicellulose degradation during wet storage, electron micrographs show evidence that the structure of ensiled corn stover differs from that of unensiled stover (Donohoe et al., 2009; Oleskowicz-Popiel, 2010). However, the impacts or significance of these structural changes on further downstream requirements has not been verified. Investigations into downstream processing of ensilage (Linden et al., 1987; Henk and Linden, 1994; Chen C. et al., 2007; Chen Y. et al., 2007; Digman et al., 2007) usually involve amended or enhanced silage that includes some form of additive—microorganisms, enzymes, chemicals, acids, sugars, etc. These studies do not have sufficient controls to make a case for the effectiveness of plain, unamended silage. For instance, Chen C. et al. (2007) compared hydrolytic and fermentation yields from enzyme amended silage to that of conventional, chemically treated feedstock, ignoring plain silage. Digman et al. (2007) analysis focus on chemically aided silage as compared to plain silage without relating it to unensiled feedstock. These studies have implicitly attributed the effectiveness of silage as a pretreatment or hydrolytic tool to


*(Continued)*

TABLE

1


Silage

studies

involving

enzymatic

hydrolysis

showing

storage

and

post

storage

treatments

and

outcomes.


these silage additives, and overlooked the potential of the plain, unamended ensilage process. Furthermore, the outcome of these amended silage depended on various factors that include the type of feedstock, the drying method, the type of additive, etc. As in Linden et al. (1987), a clear conclusion could not be made as the controls, silage without additives, were not consistent with respect to combined endogenous water soluble sugars and poststorage hydrolytic yields; unensiled (day 0) samples sometimes had higher yields, sometimes lower yield and in some cases similar yields.

As these prior studies suggest, an important attraction for wet storage is to enhance feedstock quality. The question, however, is what attributes of the wet storage system are responsible for quality enhancement. Feedstock quality can impact or influence the outcome of every downstream processing step even to the quality and quantity of the final biofuel end product (Li et al., 2016). Lactic acid is recognized as responsible for the quick reduction of pH that aids preservation with a resultant minimum dry matter loss (Muck, 2010). However, the pretreatment impacts of silage systems is generally attributed to organic acids without distinction. Understanding the critical and specific attributes that contribute to feedstock quality and identifying consistent quality indicators could bring provide a coherent knowledge framework to understand and improve wet storage strategies for biomass.

In addition to the technical effects, wet storage offers practical and economic opportunities for innovation and expanded farmer involvement in the cellulosic industry. The US Energy Policy Act of 2005 stipulates, in section 942(a)(4) and (e) (2), that funding priority would be given to projects in which the small feedstock producers are full participants or equity partners in the development of the cellulosic biofuel industry. Another important section of the Act is 946(a)(1) in which grants would be offered to agricultural producers who can demonstrate cost effective cellulosic biomass innovations in the preprocessing of feedstock, including chemical or biochemical treatments to add value and lower the cost of subsequent processing at the biorefinery. Although the more recent Energy Independence and Security Act of 2007 (EISA), does not contain these statements, the issue of involvement and quality enhancement still remain. Wet storage presents a platform for achieving these goals by providing opportunities for manipulation of some storage factors or the use of additives that can facilitate pretreatment or preprocessing of biomass to enhance its value at the refinery. This value-added processing can be performed at the farm level by feedstock producers, many of whom may already be familiar with the basics of ensilage, providing a simple but effective technology that can be adopted quickly at the scale needed for rapid growth of a large scale biofuel industry.

In this study, a wide range of moisture levels, outside tradition silage recommendations, are covered to accommodate varying moisture associated with flexible harvest schedules. Such moisture variations can occur with the percentage of residue removed, often associated with various heights of cut, anatomical fractions, harvest dates and/or selective harvesting. Moisture content from various anatomical fractions or different heights of cut at various harvest date can range from 10 to 80%

Continued

*and individually.*

*(1,596 U/g) administered*

 *Also for sugars release, reflect results for Ca(OH)2 and H2SO4 trt only.*

 *at 10 U/g cellulose. NG, not given; NA, Not applicable.*

 *500 CS contains hemicellulase*

 *(851 U/g), fungal alpha-amylase*

 *(7,023 U/g), bacterial alpha-amylase*

 *(8,087 U/g), and cellulase*

(Igathinathane et al., 2006). The specific objective of this study was to determine the fiber reactivity of plain silage at six different moisture levels and relate this to the corresponding organic acid profile. For reporting purposes, the organic acid profile relates to the types, amount and relative proportion of each acid present during storage. The fiber reactivity concept was adapted from Henk and Linden (1994) and refers to how easily the fiber can be converted to sugar through enzymatic reactions. The essence of the fiber reactivity test is to determine if organic acids produced during ensilage have any significant pretreatment effect on fiber structure. The response to enzymatic reaction would be an indication of the degree of pretreatment that has occurred during storage, and can be used in defining the pretreatment capability of wet storage. This study also explored whether organic acids all have the same pretreatment and/or preservative effect on feedstock during storage. To our knowledge there has been no prior study on the pretreatment effect of individual acids or any attempt to define which acids have the most impact on enhancing feedstock quality during storage. This research therefore addresses the general impact as well as the specific impact of acids produced endogenously during ensilage. Better understanding of these relationships could allow the organic acid profile produced during storage to act as quality indicator of biomass feedstock after wet storage.

#### MATERIALS AND METHODOLOGY

#### Stover Description and Storage

Corn stover was from two sources, one in the U.S. state of Iowa (IA stover) and one in Pennsylvania (PA stover). The former was obtained from the US Department of Energy's Idaho National Lab while the latter was obtained from the Penn State Dairy farm. The IA stover, planted from Pioneer brand 34A20 seed, was conventionally harvested after the grain harvest from a plot near Boone, IA. After field-drying for 3–5 days it was raked, baled, transported to Idaho and stored indoors with a tarp cover to prevent dust accumulation. The PA stover, planted from Dekalb DKC54-46 seed, was left standing in the field through winter and harvested and chopped in spring. Both stover types had their particle size reduced to 1 inch minus (≤25.4 mm), using a forage chopper, prior to these trials.

The particle size distribution was analyzed using the Penn State Forage Particle Separator. The PA stover had ∼26% of particles >19 mm, 26% were between 19 and 8 mm, 37% were between 8 and 1.18 mm and 11% were <1.18 mm. The IA stover had a similar distribution; 26% of particles >19 mm, 26% were between 19 and 8 mm, 31% were between 8 and 1.18 mm and 17% were <1.18 mm. A description of this particle size separation method can be found in Heinrichs and Kononoff (2002).

Corn stover from these two sources was presumed to have different compositional qualities. The (glucan; xylan; lignin; water soluble extractives) composition of IA and PA stover as received was (34.02 ± 0.14; 21.39 ± 0.18; 14.24 ± 0.13; 6.04 ± 0.23) and (34.35 ± 0.44; 20.49 ± 0.29; 14.96 ± 0.20; 3.82 ± 0.63), respectively. Significant components of water soluble extractives would include water soluble carbohydrates (monosaccharides e.g., glucose, xylose, sucrose—and oligomeric sugars); and sometimes organic acids; alditols; inorganic ions/cations; and tannins. The initial moisture contents of the IA and PA stover, ∼7 and 30% respectively, were adjusted within ±2 percentage units of the target moisture to six various moisture levels (25, 35, 45, 55, 65, and 75% wet basis). Moisture adjustment was carried out by spraying with appropriate amount of deionized water, mixing thoroughly and leaving over night at 4◦C to be well-absorbed into fibers. For PA 25% moisture, the stover as received was dried and rewetted to achieve the desired moisture. Corn stover was packed at a density of about 159 dry Kg/m<sup>3</sup> in 1 pint (0.00047 m3 ) glass canning jars that were tightly sealed to create anaerobic conditions and stored for 220 days at 37◦C. Experiments were performed in triplicate. After storage, samples were stored in the freezer at −20◦C until further analysis.

## Corn Stover Composition and Organic Acids

Corn stover composition before and after storage was determined by quantitative saccharification using adapted biomass analytical methods from the U.S. National Renewable Energy Laboratory (NREL) (Hames et al., 2008; Sluiter et al., 2008a,b,c). Soluble extracts were collected before and after storage for pH and organic acid measurements. Stover samples were mixed with deionized water at a ratio of 1:10, wet stover weight: water. The mixtures were shaken for 30 min at 200 rpm using a Barnstead SHKA 2000 open air platform shaker (Barnstead International, Dubuque, IA) after which the extracts were filtered through Whatman No.1 paper. The pH of extracts was determine using SevenEasy S20 pH meter (Mettler-Toledo International Inc., Columbus, OH) and calibrated with standard buffers 2, 4, and 7. The collected extracts were filtered again using 0.2µm PTFE filters, diluted 20-fold and analyzed using Dionex ICS 3000 ion exclusion chromatography (Thermo Fisher Scientific Inc., Dionex ICS 3000, Sunnyvale, CA) for types and amount of organic acids. Separation was performed at 30◦C using IonPac ICE-AS1 guard (4 × 50 mm) and analytical (4 × 250 mm) columns with 100 mM methanesulfonic acid at a flowrate of 0.16 mL/min. Organic acids were detected with a photodiode array detector (Dionex UVD 340U) at a wavelength of 210 nm. Thirteen different potential acids (lactic, acetic, butyric, pyruvate, isobutyric, valeric, isovaleric, propionic, tartaric, malic, formic, citric, succinic) were used as standards.

# Fiber Reactivity Across Moisture Levels

For the fiber reactivity assay, replicates of wet stored samples, "as is" were combined then washed in multiple steps so that the organic acids and background or residual sugar level were negligible (less than or about 0.01% on a dry matter basis) and did not interfere with the assay either by inhibiting enzyme activity or inflating hydrolytic yields. Also, the removal of all extracts provided a better basis for directly testing the fibers. Samples "as is" refer to silage without any post-storage processing like drying or further size reduction. By combining replicates before resampling we reduced the variability among replicates and the propagation of error that could result from inherent variability of the samples or other unknown factors.

The moisture content of washed samples was estimated using the microwave method (Jones et al., 2004) as a guide to how much wet material was required to achieve the target dry weight for the fiber reactivity test. Fiber reactivity was measured by how much glucose was released after 3 days of enzymatic hydrolysis without additional pretreatment. Enzyme loading rates, using a commercial cellulase (Spezyme CP, Genencor, Rochester, NY), of 0 and 15 FPU/ g glucan with corresponding commercial β-glucosidase (Novozyme 188, Novozymes A/S, Bagsvaerd, Denmark) loading rates of 0 and 60 CBU/g glucan was chosen for the IA and PA stover. However, PA stover had additional samples hydrolyzed at 2 and 5 FPU/ g glucan with corresponding β-glucosidase of 8 and 20 CBU/g glucan, to explore the sensitivity of ensiled and unensiled feedstock to these lower enzyme loading rates.

Hydrolysis was carried out in 50 ml centrifuge tubes with ∼1 dry gram of stover, at 15% solids loading. Tetracycline was added at a final concentration of 40µg/mL to prevent microbial growth during hydrolysis, and citric acid buffer (pH 4.5) was added to obtain a final concentration of 0.05 M to maintain the pH in the optimum range for enzyme activity. Samples were vortexed for ∼5 s before placing in a Barnstead Max Q5000 SHKE 5000- 7 floor shaker/ incubator (Barnstead International, Dubuque, IA) at 50◦C, 120 rpm for hydrolysis. A HOBO <sup>R</sup> U12-011H temperature data logger (Onset Computer Corp., Cape Cod, MA) was placed in incubator to monitor temperature over the 3-day hydrolytic period. Control samples included substrate blanks, enzyme blanks and unensiled feedstock as negative controls and Avicel (α-cellulose, which is pure insoluble cellulose) as a positive control.

## Fiber Reactivity After Liquid Hot Water Pretreatment

The purpose of testing pretreated samples was to determine if the impact of organic acids on feedstock during storage enhances or reduces requirements of subsequent pretreatment or if fiber reactivity of wet storage samples would be more pronounced or effective after pretreatment. Pretreatment was applied to washed PA samples. The pretreatment process employed was a liquid hot water (LHW) method using the Dionex Accelerated Solvent Extraction 350 (ASE 350) (Thermo Fisher Scientific Inc., Dionex ASE 350, Sunnyvale, CA) system at 190◦C, with one static cycle of 15 min and 0% flush. This temperature and time have been found to be the optimum condition for controlled pH LHW pretreatment of corn stover (Mosier et al., 2005). Hydrolysis was performed on the pretreated solids only (without the pretreatment liquid extract) in order to directly test the stover fiber without interference from extracted solubles, which may contain sugars, acids and other inhibitors.

In all cases, after hydrolysis, the samples were placed in a hot water bath at 95◦C for 10 min to prevent any further enzymatic reaction after which 20 ml of deionized water was added to each sample to facilitate sampling due to the high solids loading. The samples were vortexed to mix and then centrifuged to collect supernatant for sugar analysis. The supernatants were then filtered using 0.2µm nylon filters and stored at −20◦C until analyzed for sugars. See **Figure 1** for process chart.

#### Data Analysis

Filtered samples were diluted 5-fold and the amount of glucose released during hydrolysis was measured using a YSI 2700 SELECTTM biochemical analyzer (YSI Inc., Yellow Springs, OH) with 2% precision. Glucose yields from substrate and enzyme blanks, if any, were subtracted from sample yield to get actual glucose resulting from hydrolysis. For pretreated samples, glucan removed during pretreatment was subtracted from initial amount. Results were analyzed using statistical tools including principal component analysis (PCA), clustering analysis and analysis of variance (ANOVA). To analyze the effect of organic acids on fiber reactivity, PCA was first performed on raw data to see if storage samples could naturally be grouped into categories based on their organic acid profiles and to estimate the number of categories to specify for the cluster analysis. Ward's hierarchical clustering method was then used to group the samples into the appropriate number of clusters. Two other cluster analyses were performed using hydrolytic yields. One involved hydrolytic yields from the four cellulase enzyme loadings (0, 2, 5, and 15 FPU/g glucan) for the PA samples as variables for grouping and the other combined the IA and PA samples with two cellulase enzyme loadings (0 and 15 FPU /g glucan) for grouping. All correlation analyses were carried out using the Pearson Method. This gives the Pearson correlation coefficient, r, which is a dimensionless index that ranges from −1.0 to 1.0 and measures the degree of linear relationship between two data sets. The initial assumption of linearity is rejected if "r" is 0 or the p > 0.05. All statistical tests were conducted at a significance level, α, of 0.05. Statistical software used for these analyses are The Unscrambler <sup>R</sup> X 10.0 (CAMO Software Inc., Woodbridge, NJ) and Minitab 14 (Minitab Inc., State College, PA).

# RESULTS AND DISCUSSION

# Organic Acid Profile

The organic acids identified in the wet stored stover samples included lactic, acetic, butyric, isobutyric, and propionic acids. Low levels of tartaric and malic acids were also identified in a number of samples but were difficult to quantify due to their short retention times near the elution of the system void volume and hence were not used in the current analysis. The results indicated that the IA control (day 0) samples had essentially no organic acids (see **Figure 2**). All acids are reported on a dry matter basis. PA control (day 0) samples had only lactic and acetic acids, both of which constituted ≤0.6% dry mass of stover. These two acids were also present in all wet stored IA and PA samples, except for samples stored at 75% moisture, which had no lactic acid likely due to clostridia secondary fermentations (Jones et al., 2004). Lactic acid concentration was up to 4.9 and 2.2% in IA and PA stover, respectively, while acetic acid was up to 3.5 and 4.2%, respectively. Isobutyric acid was also present in all storage samples. Propionic and butyric acids were present only in high moisture storage samples. **Figure 2** shows the organic acid profile of IA and PA stover at the two storage durations and different

moisture levels. The lower acid content of the 45% PA stover could be due to compromised anaerobic storage conditions as reflected in high mean pH (5.12) compared to other storage samples (4.00–4.93) as shown in **Table 2**. In general total acid content increased linearly with moisture content. The correlation between total acids and moisture content was 0.813 and 0.819 for IA and PA samples, respectively, both with p < 0.0001. In terms of individual acids, all acids were significantly linearly correlated with moisture except lactic acid [lactic (r = −0.145, p = 0.398), acetic (r = 0.807, p < 0.001), propionic (r = 0.508, p = 0.002), isobutyric (r = 0.586, p < 0.0001), and butyric (r = 0.651, p < 0.0001)]. Lactic acid, however, had a second order polynomic acid-moisture relationship (R-square of means >90%). Analysis of variance showed the mean value of the various organic acids at the different moisture levels are all not equal [lactic (p = 0.003), acetic (p < 0.001), propionic (p = 0.010), isobutyric (p = 0.001), and butyric (p < 0.001)].

Studies on organic acids for biomass pretreatment have tested a wide range of pretreatment severities that are much higher than can be expected during wet storage: acid concentrations from 1 to 80% wt; temperatures of 100–210◦C; and retention time of 10 min to 1 h (Kootstra et al., 2009; Lee and Jeffries, 2011; Marzialetti et al., 2011; Barisik et al., 2016). Furthermore, the organic acids that are dominant during wet storage are not those typically explored for conventional pretreatment. There are limited studies on acetic acid, which is common in and sometimes dominant during ensilage, but most studies have focused on dicarboxylic acids because they are relatively stronger than the analogous monocarboxylic acids due to the two ionizable functional groups they have. More importantly, as a result of the two functional groups, these acids also have two pKa values, which could make them more suitable and efficient in hydrolytic pretreatment over a range of temperature and pH values (Lee and Jeffries, 2011). One unexplored area is the interaction of various organic acids on biomass structure, as could be possible in wet storage systems.

## Cluster Analysis

For PA organic acids, the first two principal components accounted for ∼82% of total variability. Approximately 41% of the samples were significantly associated with the first component, and about 53% were associated with the second component. Acetic, propionic, isobutyric, and butyric acids, i.e., volatile fatty acids, were the major contributors to component 1, while lactic acid was the main contributor, ∼86%, to component 2. This is also true for PCA of the combined PA and IA samples. These PCA results were supported by the strong correlation acetic acid has with isobutyric acid (0.912, p < 0.0001), propionic acid (0.643, p = 0.001) and butyric acid (0.770, p < 0.0001), all of which were strongly correlated with each other. No significant correlations exist between lactic acid and these acids. However, when considering only day 220 samples, there were some negative correlations between lactic and isobutyric (−0.446, p = 0.006), propionic (−0.360. p = 0.031) and butyric acids (−0.552, p < 0.0001) but still no correlation with acetic (−0.297, p = 0.079). Since lactic and acetic acid were not correlated, both as dominant acids could qualify as potential predictors of storage effect in defining relationship between storage acid and glucose yield, if multicollinearity is to be avoided.

Based on the PCA result, three categories were chosen for Ward's clustering analysis. The three categories resulting from the cluster analysis of PA samples were: (1) high amount of acids (5 to <10%) associated with high moisture wet stored samples



\**The stover pH was generally stable across the 220 day storage period, although in a few cases (e.g., IA 25% moisture) the pH dropped gradually over several weeks. For day 21 pH values, see Table B3 in Darku (2013b).*

(55–75% moisture), (2) a moderate amount of acids (2 to <5%) associated with low moisture wet stored samples (35 and 45% moisture), and (3) low amount of acids (<2%) associated with day 0 and 25% moisture anaerobic (wet stored) samples. These three groupings were maintained when combined with the IA samples, which had a slightly different pattern. The IA groupings were: (1) High amount of acids: wet stored samples with 75% moisture, (2) moderate amount of acids: wet stored sample with 35–65% moisture and (3) low amount of acids: day 0 and 25% moisture anaerobic (wet stored) samples.

# Fiber Reactivity of Pretreated Washed PA Stover

To evaluate the impact of pretreatment, xylan, and glucan removal were measured as well as enzymatic hydrolytic sugar yields after pretreatment. Xylan removal was used as an indicator of liquid hot water (LHW) effectiveness. When taking into account the amount of xylan degraded during storage, xylan removals in the day 220 samples were significantly higher (55.0% ± 7.2) than day 0 samples (47.2% ± 6.1) (p < 0.0001) as shown in **Figure 3**. This suggests that the percentage xylan removal had approached the limit of removal under LHW conditions or could be close to xylan removal under optimum pretreatment conditions. The idea of a limit to xylan removal in liquid hot water pretreatment is also based on results from Elander et al. (2005) and Zheng et al. (2009), in which xylan removal was always <55%. Xylan removals for day 220 samples were more variable than day 0 and not significantly different across moisture (p = 0.915). For day 0, there were no significant differences among the different moisture treatments except for the 35% moisture condition, which was lower than 65% moisture (p = 0.031).

As shown in **Figure 3**, the glucan removed during pretreatment was generally <4% for both day 0 and day 220. Day 0 samples had significantly higher glucan removal during pretreatment (p < 0.0001) than day 220. The relatively low glucan removed from wet storage samples could be a result of some glucan degradation (on average ∼1%) during storage, hence a lesser amount of readily degradable glucan was available for removal during pretreatment. For glucan

for 55% moisture, Day 0 missing).

removal there was no significant difference across moisture for day 0 samples (p = 0.545). Glucan removal at day 220 was also not significantly different across moisture levels except for 25% moisture (p = 0.030), which was only similar to the 35% moisture samples.

Glucose yields, from fiber reactivity studies, were calculated with reference to the mean glucan composition for each moisture level and reported as percent of theoretical yields. The glucan content of the initial biomass prior to the storage trials ranged from 33.50 to 35.63% for both IA stover and PA stover. Grams glucan were converted to glucose using a factor of 1.111. Glucan removed during pretreatment was subtracted from this original amount to determine the appropriate amount of enzymes to add and for calculation of theoretical yields. As expected, glucose yields increased significantly with enzyme loading (p < 0.001) (see **Figure 4**). For the PA stover, these values were 28.79% ±

4.33, 47.18% ± 3.45 and 84.18% ± 6.65 for 2, 5, and 15 FPU/g glucan, respectively. The results indicated that the glucose yields of day 220 and day 0 samples were not significantly different at the various enzyme loadings, p = 0.586 (**Figure 4**). Furthermore, glucose yields at various moisture levels for each enzyme loading were not significantly different for both day 0 (p = 0.994) and day 220 (p = 0.990). The implication is that xylan removal from the biomass during pretreatment did not serve as a good indicator for downstream glucan hydrolysis. This is understandable since conditions used for pretreatment were the optimum time and temperature for raw feedstock. Given the differences in xylan removal, it is possible that at a lower pretreatment severity (less time and/or a lower temperature), there might be significant differences in glucose yield between day 220 and day 0 samples.

#### Fiber Reactivity of Corn Stover Without Pretreatment

Fiber reactivity results showed the glucose hydrolysis yields for both stover types to be generally poor, <30% of theoretical. The glucose yields after hydrolysis were calculated using mean glucan composition for each moisture level as was done for pretreated samples. As expected, sugar yield increased with increased enzyme loading. At 15 FPU cellulase enzyme loading, the glucose yields of day 220 samples of both IA and PA stover were significantly higher (p < 0.0001) than corresponding day 0 samples. Average yields for day 220 were ∼23.61% ± 3.16 and 15.54% ± 5.71, respectively in contrast to 14.30% ± 4.83 and 11.05% ± 3.51 for day 0. At 5 FPU cellulase enzyme loading, glucose yields of PA stover at day 220 (12.56% ± 3.41) were significantly higher than yields at day 0 (9.60% ± 3.85) (p = 0.022). However, at 2 FPU there were no significant differences as a result of wet storage, 8.83% ± 2.80 vs. 6.93% ± 3.20 for day 220 and day 0, respectively (p = 0.065). Compared to yields obtained from Avicel hydrolysis (∼58% for 2 FPU, 65% for 5 FPU and 83% for 15 FPU), enzymatic access to stover cellulose was still limited after wet storage. This could be due to the low level of hemicellulose degradation during storage as compared to conventional pretreatment. On average 10% of hemicelluloses in IA stover was degraded during storage, in contrast to ∼55% removal during LHW pretreatment (of PA stover). In theory, the pretreatment capability of wet storage derives from hemicelluloses degradation as a result of organic acid interaction.

In this study there was no significant difference in storage (day 220) hemicellulose degradation across the tested range of storage moisture contents (p = 0.083) and no correlation between hemicellulose degradation and glucose yields (p = 0.90). However, if day 0 samples are also considered in the analysis, in addition to the day 220 samples, then correlation between hemicellulose degradation and glucose yield exists and is significant, 0.62 (p = 0.001). For day 220, the IA samples had statistically significant groups of [(25–45%) 65%], (55, 65%) and (55, 75%) in decreasing glucose yields (p < 0.0001). The overlaps defined using the Tukey test suggest glucose yields at all moisture levels were not that different except for 75% moisture, which is similar to 55% percent moisture but significantly differs from the remaining moisture contents. For the PA samples at an enzyme loading of 15 FPU, the groups were (45–65%), (35, 65%) and (25, 75%) (p < 0.0001). Yields from the 25 and 75% moisture samples were the lowest among this set. Thus, when these extreme moisture levels are avoided, a similar glucose yield could be expected from wet storage across a wide range of moisture contents. On the other hand, as expected, day 0 samples did not show any significant difference in glucose yields across moisture for both stover types (p = 0.346 for IA stover and p = 0.073 for PA stover). See **Figure 5** for glucose yields across moisture.

Cluster analysis was performed on glucose yields of the fiber reactivity test, initially assuming the same number of groups (three) as was determined for the organic acid cluster analysis. PA samples were clustered based on yields from the four cellulase enzyme loadings: 0, 2, 5, and 15 FPU /g glucan. The result was slightly different from the combined IA-PA set, which was based on two enzyme loadings (0 and 15 FPU/g glucan). The three PA stover glucose clusters consisted of (1) high glucose yield: wet storage (day 220) 35–65% moisture samples, (2) moderate glucose yield: all day 0 samples excluding 25% moisture, as well as 25 and 75% wet stored (day 220) samples; and (3) low glucose yield: day 0, 25% moisture. The combined IA-PA analysis yielded: (1) high glucose yield: wet storage with 25–65% moisture for IA; 35% for PA, (2) moderate glucose yield: wet storage moisture of 75% for IA; 45–65% for PA and (3) low glucose yield: day 0 samples including PA wet stored samples of 25 and 75% moisture. The cluster outcome also showed that for each stover type, two groups could be sufficient, defined as wet stored samples and day zero samples. In general, IA samples had higher glucose yields than PA samples (p > 0.0001). Wet stored IA samples were clustered as relatively high yielding while wet stored PA samples fall mainly under moderate glucose yield. This lower outcome for PA stover compared to IA stover could be due to the lower pH of the former and the presence of more volatile acids which have higher pKa. The implication is that the PA stover had comparatively fewer protons available during storage for the catalysis of hydrolytic pretreatment.

#### Relating Organic Acid Cluster to Fiber Reactivity Cluster

The relationship between the clusters of similar organic acid portfolios with clusters of similar glucose yields was examined using results from the fiber reactivity assay across different moisture contents for non-pretreated, anaerobically stored samples. For the PA stover clusters, high glucose yields were associated with high acid levels (excluding the 75% moisture sample) and moderate acid levels. Moderate glucose yields were associated with day 0 samples (excluding the 25% moisture sample) as well as 25% and 75% moisture samples of day 220. Low glucose was associated with day 0, 25% moisture samples. The relationships between the organic acid and fiber reactivity clusters of the combined IA and PA samples are shown in **Figures 6**, **7**. The pattern in **Figure 6** suggests some reasonably distinct range of glucose yields and associations with organic acid levels and also corroborates the advantage of anaerobic storage at moisture levels of 35–65%. Glucose yields across this full population of ensiled and unensiled samples were correlated with the total organic acids (r = 0.349, p = 0.002) with significant differences in the glucose yields of the various group (p < 0.0001). In general, high glucose yields were associated with moderate organic acid levels; moderate glucose yields were associated with high organic acid levels; and low glucose yield with little or no organic acids. Two exceptions were the PA and IA stover samples stored anaerobically at 25% moisture. Interestingly, although both clustered in the low range for organic acid concentrations, the IA stover produced a high glucose yield while the PA stover had a low glucose yield. This suggests an important effect of biomass harvest conditions and structural composition on subsequent hydrolysis. Across this full population of initial (day 0) and wet storage (day 220) samples, only lactic acid showed a significant correlation with glucose yield (r = 0.518, p < 0.0001). Earlier, under organic acid cluster analysis, acetic acid and lactic acid were suggested predictors for glucose based on their dominance but more importantly to avoid potential for multicollinearity. **Figure 8** shows how lactic acid serves as a more logical predictor relative to acetic.

When considering only stored samples (day 220), glucose was not correlated to total acids (−0.142, p = 0.409). The

contrary effects associated with low acid concentrations in the 25% moisture IA and PA samples were partially responsible, as were the moderate and low glucose yields for the highest acid concentrations, at 75% moisture. Moderate acid concentrations were associated with high glucose yields, suggesting an optimum level of total acids. Glucose yields showed some significant correlations with some individual acids; positively with lactic acid (0.366, p = 0.028) and negatively with propionic (−0.345, p = 0.039) and isobutyric (−0.340, p = 0.043). There was no correlation with neither acetic nor butyric acids (each: 0.299, p = 0.180). In **Table 3**, it can be observed that this overall relationship does not necessary hold within the cluster groups. Organic acids, specifically lactic and acetic acids were significantly and positively correlated to glucose yields in Group 2 (samples with moderate acid levels), even though lactic acid dominates. In Group 1 (samples with high acid levels), all acids were present, however none of them was correlated with the glucose yields. Acetic acid is dominant in this group, and even assuming the correlation p-values gave a false negative (i.e., assuming there is in fact a significant correlation contrary to what the p-values from the analysis indicate), acetic acid showed a very weak correlation compared to other acids. Glucose yields from Group 3 (samples with low acid levels) were also not affected by individual organic acids, but this was understandable since acids levels were negligible. Although glucose yields from Group 1 and 3 had no correlations with individual organic acids, Group 1 had significantly higher glucose yields than Group 3 when excluding the sole storage condition with high glucose yield in Group 3 (IA stover, 25% moisture at day 220) (p = 0.006). Otherwise there was no significant difference in the glucose yields of the two groups (p = 0.090).

Although Group 1 has a high total acid concentration, the relatively low yields compared to Group 2 cannot be attributed to the potential inhibitory nature of the acids since the samples were thoroughly washed before hydrolysis. The pH values during storage (4.07–4.78), and the acid profile of the high acid samples (in which volatile acids dominate with pKa of 4.76–4.88) suggest that the effect of the Group 1 acids on fiber structure could be negligible. This is because the level of dissociated acids, and hence protons to facilitate cleavage of bonds, were lower in Group 1 than in Group 2. On the other hand, for the cluster of samples with moderate total acid concentrations (Group 2), lactic acid with a pKa of 3.86 is dominant. In this cluster, the high pH relative to this pKa leads to a relatively high dissociation of lactic acid, resulting in more hydrogen ions being available for the hydrolytic reaction that alters the feedstock structure. This also may explain why lactic acid was the only acid correlated to glucose yields. In addition, this may explain why the IA stover, which produced more lactic acid during wet storage, had higher glucose yields than the PA stover. The overall effect of these pKa – pH relationships increases the natural pretreatment effect of the lactic acid dominated treatments in Group 2, while a much

lower effect is observed with the higher pKa acids that dominate in Group 1. Although high concentrations of these volatile acids are present in Group 1, their impact on glucose yield is minimal, with results similar to Group 3 which had little or no acid.

## HIGHLIGHTS AND PRACTICAL IMPLICATIONS OF THIS STUDY

Wet storage has been recommended over dry storage as a biofuel feedstock storage system suitable for warm, humid climates, typical of most of the major U.S. agricultural zones. These climatic conditions prolong field drying and result in extensive material deterioration before material reaches a state that is dry enough for storage. The narrow harvest window experienced as a result of weather conditions in major agricultural regions of the US, in particular the Midwest corn belt, necessitate that harvest of crop residue be carried out within 40 days of grain harvest before winter arrives with heavy snow and wind. Within these 40 days, biomass can experience significant moisture changes. This study corroborates and provides evidence that the adaption of ensilage technology, which has been proven in the livestock feeding industry but not commercially applied to cellulosic biofuels, provides an alternative to dry storage that resolves the conflict between a short harvest window and the need for prolonged field drying.

This research shows that although ensilage for the livestock feed industry vs. for biofuel production may have some common requirements, e.g., the need for lactic acid dominance, in other respects the requirements differ. For instance, whereas the livestock industry would recommend a very narrow moisture range for any particular storage configuration, a much wider range can be accommodated by the biofuel industry with similar end product quality. In this study, the storage configuration used, which is comparable to traditional upright silo, yielded similar results for moistures 35–65%. In traditional upright ensilage the recommended moisture is 60–65%. Even when considering all storage configurations for traditional ensilage, the moisture range is still relatively narrow, ranging from 50 to 70%. With biofuels, however, if extreme moisture levels (≤25 and ≥75%) are avoided, the quality of feedstock as indicated by its reactivity to enzymes was not significantly different. The implication is that even when harvesting heterogeneous moisture material, such as different anatomical fractions, various heights of cut, or when unable to harvest stover residue immediately after grain harvest, biorefineries can be optimistic about the storage outcome if moistures are within 35–65%.

Although previous studies have recommended wet storage as an alternative to dry storage, none of the commercial cellulosic plants have yet incorporated any wet storage solutions. Instead, every one of the biorefineries storing corn stover uses dry bales, and every one has had serious fires and lost massive quantities of biomass feedstock. While there is interest in wet storage as a solution to this challenge, there is limited understanding of how wet storage works in a biochemical conversion process, or how those outcomes change with slight

moisture variations compared with conventional dry storage. There also remains a perception that wet storage presents an economic disadvantage as high moisture is transported along the supply chain. The latter concern is addressed in Darku (2013a), which modeled farm gate cost of wet and dry storage feedstock and shows cost of ensiled feedstock at moisture content <40% to be comparable or better than dry bales. The former concern, which relates to feedstock quality, is addressed in this research.

In particular, the results of this study indicate that the lactic acid produced during wet storage stood out from other acids as the major contributor to feedstock quality. Samples with moderate organic acid concentrations (2 to ≤5% dry matter) correlated with high lactic acid and with high glucose yields, which could then be converted into biofuels or other products. These results suggest that lactic acid dominance in the ensilage process is key to wet storage pretreatment effectiveness, and is more important than high amounts of total acid.

#### CONCLUSIONS

There is evidence from the fiber reactivity test conducted in this study that wet storage, without any form of biological or chemical additives, does indeed have pretreatment capability. Although cellulose accessibility was limited, due to the low hemicellulose degradation during storage, the feedstock structure was altered enough to improve glucose yields over day 0 samples by a factor of ∼1.5–2.4. However, relative to conventional pretreatment and theoretical glucose yields, these yields were low, indicating that wet storage of corn stover biomass would still require post-storage pretreatment to achieve desirable hydrolytic yields. The glucose yield after enzymatic hydrolysis of wet stored samples was comparable across moisture except for the extreme moisture levels tested (25 and 75%). Generally, storage at 75% moisture had the lowest glucose yield. In the moderate moisture range from 35 to 65%, this similarity in yields also minimizes the expected complexity and questions about quality given the varied moisture conditions under which different farmers may harvest and store their biomass feedstock. Other crop management factors rather than storage moisture may be of more concern to quality outcome.

Cluster analysis comparisons indicated that moderate (total) acid levels corresponded with high glucose yields and viceversa, with significant but not very strong correlations. For wet storage pretreatment to proceed, organic acids must dissociate to provide protons that will catalyze the cleavage of bonds. Generally, samples with high total acid levels are high moisture samples (55–75%) with comparatively lower lactic acid amount, while samples with moderate acid levels had comparatively higher lactic acid concentration. Lactic acid was found to be TABLE 3 | Differences in the three organic acid groupings and correlation of individual acids with glucose.


◦*For IA stover, 55 and 65% moisture fall under group 2.* \**IA and PA Day 220 samples with 25% moisture fall under this group. (*\**)Independent variable (organic acid) is zero hence no basis for correlation. Values in bracket represent correlation coefficient with glucose yield from fiber reactivity test and corresponding p-values. Lactic, acetic ratio used because no correlation between the two; in addition, acetic correlated to all the other acids. Highlighted cells indicate main acid under group. For glucose yield, percent before values is storage moisture.*

the most influential acid because of its low pKa, which allows more dissociation at the relatively high pH typical of corn stover silages.

With respect to post-storage liquid hot water pretreatment, there was no significant difference between hydrolytic yields of day 220 and day 0 samples likely due to the pretreatment conditions employed. It is probable that less severe conditions would favor wet storage samples over day 0 samples as the former had more xylan. Also using fresh stover instead of rewetted dry stover could present different outcome in favor of the former. The lack of difference at lower enzyme loadings could be more of low reaction rate rather than maximum yield that could be obtained from the two durations.

#### AUTHOR CONTRIBUTIONS

DE and TR conceived the idea. DE designed, planned and performed the experiments. MM contributed to the experimental set-up, analytical methods used and provided some technical support. AR provided the methodology and settings for washing the feedstock. TR and MM provided critical feedback that shaped the research as it progressed. DE performed the data analysis, drafted the manuscript and designed the figures. All authors discussed the results and commented on the manuscript.

#### ACKNOWLEDGMENTS

This research was supported by the U.S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Bioenergy Technologies Office, under DOE Idaho Operations Office Contract DE-AC07- 05ID14517. Additional project support was provided by Penn State University and Agriculture and Food Research Initiative Competitive Grant No. 2012-68005-19703 from the USDA National Institute of Food and Agriculture. The authors are grateful to Kay DiMarco and the many students that helped out in the lab. The authors would also like to thank the Bioenergy Technologies Office for their support. The views and opinions of the authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.

# REFERENCES


Springer International Publishing), 387–419. doi: 10.1007/978-3-319-179 155\_19


Zheng, Y., Yu, C., Cheng, Y. S., Zhang, R., Jenkins, B., and VanderGheynst, J. S. (2011). Effects of ensilage on storage and enzymatic degradability of sugar beet pulp. Bioresour. Technol. 102, 1489–1495. doi: 10.1016/j.biortech.2010.09.105

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Essien, Marshall, Richard and Ray. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Wear Properties of Ash Minerals in Biomass

#### Jeffrey A. Lacey, John E. Aston and Vicki S. Thompson\*

*Biological and Chemical Processing Department, Idaho National Laboratory, Idaho Falls, ID, United States*

Ash in biomass is believed to damage biorefinery equipment due to its abrasive properties. All biomass contains at least some ash, or inorganic content, as a result of normal physiological processes. The concentration of biogenic ash in biomass is largely species dependent; however it can also be affected by weather patterns, irrigation, soil type, and fertilizer applications. Ash concentrations in harvested biomass can also be elevated due to the incorporation of soil and dust during the harvest and collection processes. While ash concentration in biomass is important, so also is the mineral form of the ash. Certain mineral forms of ash can be much harder than the steels used to construct biorefinery equipment and cause excessive wear. In this perspective, the relative concentrations of ash elements, mineral forms of ash, and the hardness of these minerals are considered to identify ash components of concern to biorefinery operators. Strategies are suggested to remove ash from harvested biomass to reduce the risk of excessive wear on biomass processing equipment.

#### Edited by:

*Timothy G. Rials, University of Tennessee, Knoxville, United States*

#### Reviewed by:

*Ben J. Stuart, Old Dominion University, United States Oladiran Fasina, Auburn University, United States*

> \*Correspondence: *Vicki S. Thompson Vicki.Thompson@inl.gov*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *15 May 2018* Accepted: *22 October 2018* Published: *09 November 2018*

#### Citation:

*Lacey JA, Aston JE and Thompson VS (2018) Wear Properties of Ash Minerals in Biomass. Front. Energy Res. 6:119. doi: 10.3389/fenrg.2018.00119* Keywords: biomass abrasion, biogenic ash, introduced ash, ash minerals, equipment wear

#### INTRODUCTION

Based on reports from the U.S. Department of Energy supported Integrated Biorefinery projects (IBRs), biomass feedstocks such as corn stover are causing more wear to equipment than anticipated. Anecdotal information from one IBR cited grinder blade replacement every few days, less than 10% of their expected lifetime. Pneumatic conveyance systems were sustaining damage, particularly at bends where holes were worn through ducting. This excessive wear resulted in processing equipment shut-downs that could range from a few hours to days (Nguyen, 2017, pers. comm.). The problem of biomass-caused wear has been known and reported on for several years (U.S. Department of Energy, 2010; Kenney et al., 2013; Le et al., 2014); however very little research has been published that helps to determine the cause of the wear and develop mitigation strategies. Biomass-caused wear has been identified as a serious issue facing the startup of cellulosic biorefineries and needs to be mitigated for successful plant operations (U.S. Department of Energy, 2016). A more complete understanding of the physical properties of biomass is required to (1) adjust properties of the incoming biomass to result in less wear, and/or (2) design equipment that is more resistant to wear from biomass. In this perspective paper, we will discuss the poorly understood relationships between biomass feedstocks and biorefinery equipment wear. The known wear-causing properties of biomass feedstocks will be reviewed and strategies to mitigate these properties will be discussed.

#### BIOMASS- AND SOIL-CAUSED WEAR

A thorough review of the literature pertaining to wear properties of biomass feedstocks yielded limited information. Two studies examined wear occurring on an aluminum feed chute for an automated rice grain sorter (Camacho et al., 2007, 2009). The first study demonstrated that the main wear mechanisms were erosion and abrasion (Camacho et al., 2009). It was unclear how much rice or rice husks contributed to these wear mechanisms, but an experiment designed to simulate the presence of rice husks (which are known to be abrasive, Poudel et al., 2012) showed an order of magnitude increase in wear volume compared to rice alone. This study also examined several surface treatments (hard anodized, hard anodized with PTFE and chrome plating) that increased surface hardness. These treatments reduced wear volumes by 50% in the absence of abrasive particles and by 78– 85% when abrasive particles were present. The second study demonstrated that erosive wear was predominant at the top of the chute where the grains initially impact, and abrasive wear was predominant lower on the chute in areas where the grains were sliding (Camacho et al., 2007).

Two other studies examined wear phenomena in the sugar cane industry. The first study, examining tribological aspects of shafts and bearings in sugar cane mills, found that a combination of bagasse and soil contamination entered bearing gaps and resulted in rapid abrasive wear and formation of grooves on the shaft journal (Rivas et al., 2006). Further analysis determined that both abrasion and adhesion mechanisms were responsible, with plastic deformation of grains forming laminated zones and generating debris contributing to abrasion. The second study examined wear occurring on the blades of mechanical sugarcane cutters and found that wear increased as the cane was cut closer to the ground with blades having to be replaced as often as every 20 min (Langton et al., 2007). This indicates that in addition to the wear being caused by the plant material, contaminating soil also plays a strong role in wear.

The wear of pelleting and briquetting dies has also been investigated. Sharma et al. (2016) found that the screw extruder of a briquetting system was worn out within hours when rice husks were used as a feedstock. Shrestha and Ghimire (2014) reported that it was the silica in rice husks that caused the most friction against the screw extruder and the briquette die. In their study, the screw had worn significantly within 4–5 h of processing rice husks. Reducing the silica content by blending the rice husks with lower ash biomass led to reduced wear on the dies and screw extruder. In another report, it was determined that the main cause of wear in a pellet extruder machine was due to the inclusion of sand with the biomass and high silica content of the wood being pelleted (de Wet et al., 2016).

#### SOURCES OF INORGANIC CONTENT IN BIOMASS

The ash content in biomass is derived from two primary sources: introduced ash collected during harvest and processing, and biogenic ash inside of the plant tissues due to normal biological TABLE 1 | Mohs relative hardness scale, examples of minerals of each hardness level, and common tests to determine mineral hardness.


processes. The sum of these two numbers is referred to as the "ash content" of the sample and can range from 0.1% (w/w, debarked wood chips) to 26% (w/w, rice husks), or higher as introduced ash concentrations increase (Tao et al., 2012). Examination of 840 samples of wheat, miscanthus and corn stover showed that 90% of the samples had ash contents between 3 and 14% (w/w) (Kenney et al., 2013).

#### Introduced Ash

Introduced ash added during harvest and collection of biomass is the most likely cause of excessive wear being observed in biorefineries (U.S. Department of Energy, 2016 Nguyen, 2017, pers. comm.). The amount of introduced ash included with the biomass can be dependent upon the method of harvest (Shinners et al., 2012; Williams et al., 2016), the harvester configuration (Bonner et al., 2014b; Lizotte et al., 2015), and the skill of the equipment operator (Bonner et al., 2014b). This added material can contain rocks, sand, soil, and dust. The most common parent minerals of soil formation include quartz (SiO2, Mohs hardness = 7), calcite (CaCO3, Mohs hardness = 3), feldspar (KAlSi3O8, Mohs hardness = 7), and mica (biotite, K(Mg,Fe)3AlSi3O10(OH)2, Mohs hardness = 2.5– 3; Miller and Donahue, 1990; Barthelmy, 2014). The Mohs hardness index is explained in **Table 1** with higher numbers indicating harder materials. The exact chemical characteristics of soil contamination in biomass is very location specific, and can even be field and sub-field specific (Bonner et al., 2014a). Of these soil forming minerals, quartz and feldspar are of most concern as they are both harder than mild steel and are more likely to damage biomass processing equipment.

#### Biogenic Ash

Biomass is never ash-free as plants require inorganic elements and minerals for normal growth and physiological functions (Epstein, 1994, 1999; Kochian, 2000). Essential elements include the macronutrients nitrogen, potassium, calcium, magnesium, phosphorous, and sulfur, and the micronutrients chlorine, boron, iron, manganese, zinc, copper, molybdenum, and nickel. Cobalt is required for all plants that fix nitrogen. Silicon and sodium are required by very few plants; however their presence in the soil can be beneficial as they are absorbed by the plant and incorporated into plant tissue. Aluminum can be present in low quantities in plants complexed with other elements such as silicon (Vassilev et al., 2013).

The presence of these elements is not necessarily predictive of wear, but rather the wear properties are defined by the type of mineral in which the element is present. As an example, elevated concentrations of silicon in biomass, if present as quartz (SiO2), could be quite damaging to processing equipment due to the hardness of the mineral (Mohs hardness = 7; Barthelmy, 2014). However, the same concentrations of silicon present in the biomass as the mineral kaolinite (Al2Si2O5(OH)4, Mohs hardness = 1.5–2) would be unlikely to cause damage to the processing equipment. The mineral content found in the plants is dependent upon several variables, including plant species, soil type, soil amendments and weather. There are several chemical and physical forms that inorganic minerals can take in the biomass, including amorphous crystals, precipitated salts, integrated structures, or complexed with other organic molecules within the lignocellulosic matrix.

# DISCUSSION

For these minerals to cause damage to equipment, the mineral must be present in sufficient quantities and must be at least as hard as steel to cause rapid damage to biorefinery equipment. Softer minerals may also cause wear to steel through long term exposure and high speed impacts; however only minerals at least as hard as mild steel were considered due to their potential to cause wear at an accelerated rate. **Table 2** shows plant and soil minerals with observed concentrations above 1% shaded green and minerals with Mohs hardness of 5 (mild steel) or above shaded blue. The names of minerals with both these properties have been shaded red. It is hypothesized that these redshaded minerals are the most likely to cause damage to biomass processing equipment, and all but three are silicates. As a basic example, in some biomass feedstocks 10% of the mass is ash, and over 50% of that ash can be silicon-based minerals. Thus, five percent of the material that is being handled in a refinery could be hard silicates that are damaging equipment.

#### Minimizing the Abrasive Properties of Biomass

Prior to introducing the material into the conversion reactors, there is an opportunity to improve the quality of the biomass by removing ash. Effective mechanical and chemical methods have been developed that can reduce the ash content of the biomass; however it is likely not economically feasible to remove all of the ash from the biomass. An understanding of the advantages and limits of these approaches will enable the selective removal of the most abrasive ash components found in the biomass using the most cost effective technologies.

#### Wear Minimization Through Mechanical Separations

Introduced ash can be efficiently removed via mechanical separations including size separations and air classification. TABLE 2 | Mineral compounds found in biomass [Adapted from (Vassilev et al., 2013), hardness values obtained from the "Minerology Database (10) (http:// webmineral.com/)].


*(Continued)*

#### TABLE 2 | Continued


*Mineral concentrations with observed concentrations above 1% have been highlighted in green, Mohs hardness values of 5 or above have been highlighted in blue, and minerals names which meet both criteria (*>*1% concentration, Mohs hardness* ≥*5) have been shaded in red.*

Air classification of forest residues was shown to concentrate 40% of the total ash into a small fraction that represented about 7% of the total biomass (Lacey et al., 2015). The concentrated ash was primarily introduced ash (enriched in silicon, aluminum, and iron); however the mineral compositions of this ash fraction were not determined. A similar study using a variety of feedstocks including corn stover, switchgrass, and grass clippings, and a combination of air classification and size fractionation was used to isolate fractions with high ash content. An effective separation of soil elements was evidenced by elevated concentrations of aluminum and iron in the lightest air classified fractions and smallest size fractions (Thompson et al., 2016). Others have also shown ash to be concentrated in the smallest size fractions (Smith et al., 2012; Zhang et al., 2012). Using both sieving (Liu and Bi, 2011; Lacey et al., 2016; Thompson et al., 2016) and trommel screens (Smith et al., 2012; Dukes et al., 2013; Greene et al., 2014), concentrations of introduced ash could be effectively reduced from biomass feedstocks.

#### Wear Minimization Through Chemical Preprocessing

Biogenic ash is often precipitated inside the cell walls or contained in water transport elements, but can also be incorporated into the cell walls as part of the physical support structure. Because of this, typical mechanical fractionation methods cannot effectively separate it from the bulk of the feedstock. If it is necessary to remove this type of ash, chemical preprocessing that solubilizes the ash would be required. The exact location and chemical state of the ash will determine the most effective removal methods. Physiological cations involved in nutrient transport and enzyme catalysis may be leached out via diffusion if an acid is added as a counter ion to facilitate ion-ion exchange (Schell et al., 2003; Li et al., 2010; Liu and Bi, 2011; Aston et al., 2016). Hot water "washes" cause auto catalysis of acetyl acid groups producing acetic acid (Lu et al., 2016). This has been observed to result in over 90% removal of alkaline earth and alkali metals at temperatures as low as 90◦C (Aston et al., 2016). Although dilute acid leaches or hot water washes are effective at removing specific cations, alkaline extractions will likely be required to remove both entrained and physiological silica since it causes structural changes that liberate this element (Hsieh et al., 2009; Cheng et al., 2011; Bazargan et al., 2015). Such an approach combined with mechanical separations and blending with lower ash materials may improve the economic feasibility within the feedstock supply chain (Lacey et al., 2015, 2016; Thompson et al., 2016).

#### CONCLUSIONS

While little is known about the specific wear properties of biomass, much is known about the minerals that can be present in biomass, and methods have been developed that are capable of efficiently removing introduced and biogenic ash. Additional work is needed to better characterize the minerals present in biomass feedstocks that are causing the most wear. With this understanding, biomass could be modified to alter its wear properties, or equipment could be selected that will be more compatible with the specific mineral content found in the biomass being processed.

#### AUTHOR CONTRIBUTIONS

JL is the primary author and contributed text on the literature review and on mitigation methods for ash removal. JA contributed text on mitigation methods for ash removal. VT is

#### REFERENCES


the corresponding author and contributed text on the literature review.

#### FUNDING

The research was supported by the U.S. Department of Energy (DOE), Office of Energy Efficiency and Renewable Energy (EERE), Bioenergy Technologies Office (BETO), under DOE Idaho Operations Office Contract DE-AC07-05ID14517.

anatomical and size fractionation. Biomass Bioenergy 90, 173–180. doi: 10.1016/j.biombioe.2016.04.006


SPORL pretreatment. Bioenergy Res. 5, 978–988. doi: 10.1007/s12155-012- 9213-3

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Lacey, Aston and Thompson. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Multi-Criteria Decision Analysis Approach to Facility Siting in a Wood-Based Depot-and-Biorefinery Supply Chain Model

#### Natalie Martinkus <sup>1</sup> , Greg Latta<sup>2</sup> , Kristin Brandt <sup>1</sup> and Michael Wolcott <sup>1</sup> \*

*<sup>1</sup> Composite Materials and Engineering Center, Washington State University, Pullman, WA, United States, <sup>2</sup> Department of Natural Resources and Society, University of Idaho, Moscow, ID, United States*

#### Edited by:

*J. Richard Hess, Idaho National Laboratory DOE, United States*

#### Reviewed by:

*Feni Agostinho, Universidade Paulista, Brazil Mary Biddy, National Renewable Energy Laboratory DOE, United States*

> \*Correspondence: *Michael Wolcott wolcott@wsu.edu*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *18 April 2018* Accepted: *30 October 2018* Published: *20 November 2018*

#### Citation:

*Martinkus N, Latta G, Brandt K and Wolcott M (2018) A Multi-Criteria Decision Analysis Approach to Facility Siting in a Wood-Based Depot-and-Biorefinery Supply Chain Model. Front. Energy Res. 6:124. doi: 10.3389/fenrg.2018.00124* As the lignocellulosic biofuels industry is still developing, reducing operational, and capital costs along the supply chain can increase the competitiveness of the final fuel price and investor willingness to commit funds. Capital cost savings may be realized through co-locating depots with active biomass processing plants, such as saw mills, and through repurposing existing industrial facilities, such as pulp mills, into a biorefinery. Operational cost savings may be gained through the selective siting of depots and biorefineries based on operational cost components that vary geospatially, such as energy rates and feedstock availability. Utilizing depots in a biofuel supply chain to procure and preprocess feedstock has additionally been found to mitigate supply risk in regions of low biomass availability, as well as reduce the biorefinery footprint. A multi-criteria decision support tool (DST) is utilized to assess existing industrial facilities for their potential role in a wood-based depot-and-biorefinery supply chain. Geospatial cost components are identified through techno-economic analyses for use as siting criteria for the depots and biorefinery. The "repurpose potential" of industrial facilities is assessed as a siting criterion for candidate biorefinery locations. A case study is presented in the Inland Northwest region of the United States to assess the usefulness of the tool in selecting industrial facilities for a configured depot-and-biorefinery supply chain. The results are compared against optimization runs of the candidate facilities to validate the depots selected by the DST. In two of the three supply chains, the DST selected the same or similar facilities as the optimization run for no net increase in annual cost. The third supply chain showed an ∼1% increase in annual cost over the optimized facilities selected.

Keywords: depots, wood-based biorefinery, facility siting, aviation biofuel, supply chain analysis, multi-criteria decision analysis, forest residue, decision support tool (DST)

**Abbreviations:** BDMT, Bone Dry Metric Tons; CapEx, Capital Expenditure; DST, Decision Support Tool; FIA, Forest Inventory and Analysis; GIS, Geographic Information System; LURA, Land Use Resource Allocation; MCDA, Multi-Criteria Decision Analysis; OpEx, Operational Expenditure; TEA, Techno-Economic Analysis; TTCM, Total Transportation Cost Model

# INTRODUCTION

Annual feedstock supply variability and significant upfront capital costs can affect investor willingness to commit funds for the construction and operation of cellulosic and advanced biorefineries (Coyle, 2010). To mitigate supply risk in regions of low biomass availability, depots have been found useful for procuring and preprocessing feedstock (Hansen et al., 2015; Lamers et al., 2015a,b). Capital construction cost savings may be found through repurposing existing facilities into a biorefinery when the infrastructure and equipment are compatible with the biorefinery design (Martinkus and Wolcott, 2017). Additionally, operational cost savings may be gained through the selective siting of depots and biorefineries based on location-specific costs such as energy rates and delivered feedstock cost (Martinkus et al., 2017b). Prioritizing capital and operational cost savings during site selection can enable biorefineries and depots to be constructed with reduced financial risk.

Potential biorefinery locations are often selected based solely on general location characteristics (e.g., population, proximity to rail, etc.) with the assumption that a greenfield site can be found within the geographic boundaries (Panichelli and Gnansounou, 2008; Parker et al., 2008; Stephen et al., 2010; Zhang et al., 2011; Ng and Maravelias, 2016). Others assume every pixel, grid point, or county centroid is a potential biorefinery location in a study region or along a roadway (Graham et al., 2000; Noon et al., 2002; Wilson, 2009; Kocoloski et al., 2011; You et al., 2012; Lewis et al., 2014). Still others (Ma et al., 2005; Sultana and Kumar, 2012) expand on the pixel approach by performing an exclusion analysis using rasterized layers to identify potential biomass-based facilities. These approaches may be adequate when performing a scoping analysis for biorefinery siting as they broadly assume a suitable location can be found within the pixels or region identified as optimal. However, a strategic siting analysis should focus on reducing capital and operational biorefinery costs to lessen the barrier to entry into the fuels marketplace. Research has found that location characteristics and economic determinants influence site selection (Kenkel and Holcomb, 2006; Stewart and Lambert, 2011; Fortenbery et al., 2013). Therefore, strategic siting decisions must include considerations for location-specific cost variables to reduce capital and operational expenses.

A centralized, or integrated, biorefinery includes all biomass processing units from preprocessing through conversion into fuel, while in a distributed biorefinery, some of the initial processes occur at a separate location, such as a depot. The integrated biorefinery's feedstock collection area is the immediate radius surrounding the biorefinery, whereas depots can draw biomass from geographically separate locations and marginal lands previously inaccessible (Argo et al., 2013). The technical feasibility of depots in a biorefinery supply chain model has been explored in depth by others for preprocessing and pretreating cellulosic material to reduce the biorefinery footprint and operational costs (Carolan et al., 2007; Eranki et al., 2011; Bals and Dale, 2012; Argo et al., 2013; Hansen et al., 2015; Kim and Dale, 2015; Lamers et al., 2015a,b). All of these studies assume the biorefineries and depots would be greenfield facilities. Ng and Maravelias (2016) assume depots could co-locate with farms for biomass drying and densification, yet they do not provide siting criteria for farm selection. All facilities these studies are sited in optimized locations based on minimizing transportation and feedstock costs without considering other facility expenses that may impart additional influence on the overall cost to procure and process feedstock.

A multi-criteria decision analysis (MCDA) approach can be useful for facility siting analyses that utilize disparate siting criteria. The most-often used MCDA tool in biorefinery siting is the Analytic Hierarchy Process (Saaty, 2008; Sultana and Kumar, 2012; Van Dael et al., 2012). Stakeholders perform pairwise comparisons between all criteria in a siting analysis to determine the relative importance of each criterion, from which criteria weights are derived. Individual facilities or sites are then scored based on their location-specific criteria multiplied by the respective criteria weights. While this method provides a quantitative facility scoring method, the criterion weights are inherently biased due to the relative criteria importance determined qualitatively by the stakeholders.

Perimenis et al. (2011) developed an MCDA tool using a modified version of the AHP to aid users in selecting biofuel production pathways from multiple feedstock and conversion pathways. Production pathway criteria included qualitative and quantitative measures, and a pairwise comparison was performed by the researchers to determine the criteria weights. The range of values for each criterion was translated into a given grade scale. Production pathway scores were developed by multiplying criterion scaled values by their respective criterion weight.

Martinkus et al. (2017b) built on this approach by using economic and social metrics in a facility siting decision support tool (DST) to assess existing industrial facilities for their potential role as a repurposed biorefinery. Operational cost components that vary geospatially (such as feedstock cost or energy rates) were selected as the economic siting criteria. The range of values for the siting criteria in each metric were translated into a grade scale to assess the list of existing facilities within the region for their compatibility with the "biorefinery design case." Economic siting criteria weights were developed through assessing the biorefinery techno-economic analysis (TEA) to determine the annual percentage cost of each geospatial operational cost out of the total geospatial operational costs. This approach to weight derivation removes inherent biases that may be present in pairwise comparisons. Existing facilities were scored based on how well their individual site characteristics compared to the biorefinery design case.

The aim of this research is to develop a methodology for assessing existing facilities for their potential role in a distributed supply chain for biomass processing and biofuel creation. The objectives are to (1) select existing facilities to serve as depots for a given potential biorefinery through considering location-specific operational costs, and (2) identify the depotand-biorefinery configuration that provides the least processing and transportation costs from an array of potential depot and biorefinery locations for a given end user.

Researchers have studied the benefits of repurposing facilities into biorefineries, applied MCDA to biorefinery site selection, and utilized depots in biorefinery supply chain analysis, but none to our knowledge have combined all three approaches into one siting model. The MCDA approach utilized in this work is presented as a decision matrix developed to assess industrial facilities and primary biomass processing facilities for their potential inclusion in a biofuel supply chain. Coupled with a transportation cost model, supply chains are developed, and the supply chain that procures and processes biomass into biofuel at the least-cost for a given end user can then be identified. A case study is presented in the Inland Northwest region of the United States, and the results are compared against an optimization routine to determine how well facilities are selected by the decision support tool.

#### METHODOLOGY

Facility siting for both the depots and biorefinery is comprised of a series of steps (**Figure 1**) that center around two multicriteria decision matrices, one for the assignment of depots to each potential biorefinery and one for the selection of a final biorefinery location based on the depots + biorefinery configuration. The following sections describe the general form of the decision matrix, a general overview of the depot-andbiorefinery site selection process, and the general form of the Total Transportation Cost Model that is used to develop delivered feedstock costs.

#### Generalized Form of Decision Matrix

The decision matrix presented here (**Table 1**) is based on work by Martinkus et al. (2017b). The decision matrix defines facility siting criteria, weights, and scale values. Criteria are selected as geospatial metrics important in the siting of a biorefinery or depot, weights define the relative importance of each criterion, and scale values provide a means for assessing existing facilities against the design biorefinery case based on location-specific values relative to the range of regional values present.

The rationale behind using geospatial cost components as siting criteria stems from the knowledge that costs for feedstock, energy, and labor vary between locales. The infrastructure present at each facility also varies by facility. The decision matrix allows candidate facilities to be assessed based on their assets or rates in relation to the range of regional values for each cost component. Here, the scale values, s, range from 1 to 5. A "5" indicates a facility rate or asset that provides the least cost for a criterion component, and a "1" indicates an asset that may add significant additional cost to the construction or operation of a facility.

An average annual cost is determined for each geospatial cost component, c<sup>i</sup> , through inputting regional average rates into the TEA and aggregating the total amount spent from all operational units in the facility. For example, a regional average electricity rate is input into the TEA, and the total annual amount of electricity used is summed over all units that require electricity. Weights, w<sup>i</sup> , are determined by calculating each cost component's percentage of the total annual cost of all geospatial

TABLE 1 | Generalized form of the decision matrix.


cost components used as siting criteria (n) (Equation 1). The maximum scale value in the decision matrix, smax, is used to normalize the weights. Each facility j's score, F<sup>j</sup> , is calculated by multiplying each weight by the location-specific scale value, sji, for each siting criterion using the Weighted Sum Method (Wang et al., 2009) (Equation 2).

$$w\_i = \left(\frac{c\_i}{\sum\_{i=1}^n c\_i}\right) \; \* \; \frac{100}{s\_{\text{max}}} \tag{1}$$

$$F\_{\vec{j}} = \sum\_{i=1}^{n} w\_i \quad \text{\*} \quad s\_{\vec{j}\vec{i}} \tag{2}$$

Each criterion's range of regional values is used to determine its scale value designation in the decision matrix. These criterion "bin" values (Bi) are determined by dividing the range of regional values (ai,max, ai,min) by the maximum scale value (smax) for each criterion i (Equation 3). For each criterion, the maximum scale value is assigned to the minimum or maximum range value that denotes the most positive influence on facility siting, such as high infrastructure repurpose potential or low electricity rate (Martinkus et al., 2017b). The subsequent scale values are calculated by either adding or subtracting B<sup>i</sup> , depending on the positive or negative influence of the criterion (**Table 1**) (Martinkus et al., 2017b). Where regional values are not available or possible, as in infrastructure assessments or delivered feedstock cost, bin values are determined from the range of facility values (Martinkus et al., 2017b).

$$B\_i = \frac{a\_{i,max} - a\_{i,min}}{S\_{max}} \tag{3}$$

#### General Overview of Depot and Biorefinery Site Selection Process

Potential depot and biorefinery locations are first identified in a region of interest based on their compatibility with the biorefinery feedstock and proximity to other facilities. For example, in a wood-based biofuel supply chain, potential depots are identified as active sawmills and potential biorefineries are identified as active or recently decommissioned pulp mills, since all are compatible with woody feedstock. All facilities are then assessed to ensure their site acreage is sufficient for the design depot and biorefinery footprints as well as for storing a percentage of the annual feedstock demand.

A Total Transportation Cost Model (TTCM) is used to determine the least-cost routes between biomass (feedstock) sources, depots, biorefineries, and the end user. The model is developed in a geographic information system (GIS) platform, and includes a networked road dataset with a cost per road segment for each truck type that will haul the biomass in its various forms along the supply chain. Point locations (or nodes) are added to represent the biomass sources, depots, biorefineries, and the end user. Results from the model include a total cost to transport the material along the shortest distance between each node of the supply chain.

Facility siting criteria selection and weight derivations for both the biorefinery and depot decision matrices are performed using their respective TEAs. Depot siting criteria are derived from the operational expenditures (OpEx) that vary geospatially, such as feedstock, energy, and labor. From the TTCM, a weighted average delivered feedstock cost is determined for each depot at a set annual feedstock demand. To develop criteria weights, the average of all depot weighted average delivered feedstock costs is determined and input into the depot TEA as the feedstock cost, along with regional average energy rates. The annual expense for each criterion is determined, and the percentage of each expense out of the sum of all geospatial expenses is calculated (Equation 1). These percentages are normalized and represent the criteria weights for use in the decision matrix.

One depot decision matrix is developed for each potential biorefinery to identify the top-ranking depots that will contribute biomass to the biorefinery most efficiently. Each potential depot is assessed using the depot decision matrix, and a scale value is assigned for each criterion in the matrix based on the depot's location-specific value relative to the range of values used to define the criterion. Each criterion scale value is multiplied by the respective criterion weight, and the numbers are summed to develop the facility score (Equation 2). The facilities are ranked from highest to lowest score, with the highest-ranking depots having lower energy, feedstock, and labor costs as compared to the range of potential locations assessed. The number of depots selected to provide preprocessed feedstock to a biorefinery is dependent on the size (i.e., feedstock demand) of the design depot modeled in the TEA and the annual feedstock demand of the biorefinery.

If top-ranking depots are located in close proximity to one another, only one will be chosen and the next-ranked depot on the list will be selected. This removes biomass competition between proximal facilities that would force feedstock costs higher. Each depot's delivered feedstock cost is initially determined without consideration for biomass competition from other depots. Once the final depots are selected for inclusion in a biorefinery supply chain, biomass source nodes must be assigned to each of the final depots to determine their final weighted average delivered feedstock cost. This can be performed using an optimization model.

A total cost for each depot-and-biorefinery supply chain is developed by first summing each selected depot's biomass processing cost with the cost to transport densified biomass to the biorefinery, then averaging the resulting costs, and finally, summing with the cost to transport biofuel to the end user. Each depot's processing cost is determined by inputting the depotspecific delivered feedstock cost and energy/labor rates into the TEA and calculating the minimum selling price of the densified feedstock.

The depot-and-biorefinery supply chain total cost is used as a siting criterion in the biorefinery decision matrix, along with other geospatial operational siting criteria identified in the biorefinery TEA, such as energy and labor. An additional siting criterion is included for assessing facility repurpose potential, as we assume that capital expenditure (CapEx) cost savings may be gained through repurposing an existing facility as opposed to constructing a greenfield biorefinery. Biorefinery siting criteria and weights are derived similarly as for the depot decision matrix, and the depot-and-biorefinery supply chains are assessed and scored similarly as well. The top-ranked depot-and-biorefinery supply chain procures, processes, and transports material at the least-cost for the region of interest. It must be noted that in this analysis depots are considered to be co-located with an existing facility, and the depot decision matrix may include a siting criterion to reflect facility repurpose potential if needed.

#### Total Transportation Cost Model, TTCM

The TTCM determines the delivered feedstock cost and volume between two nodes along the supply chain using a multipleorigin, multiple destination algorithm based on Dijkstra's algorithm for finding the shortest path between two points (Dijkstra, 1959; Esri, 2015). Nodes are locations of biomass procurement or processing, and linkages are the road or rail network that connect nodes.

A networked road dataset is utilized in a GIS environment, and includes a transport cost for each road or rail segment used in the analysis. The general form of the TTCM between two nodes for road or rail transport is represented in Equation (4) (Sultana and Kumar, 2012; Martinkus et al., 2017a).

$$TC\_{b\dot{j}} = \,^F\_b + \,^V\_{b\dot{j}} \tag{4}$$

where TCbj is the total delivered feedstock cost between nodes b and j, F<sup>b</sup> is the fixed cost associated with node b, and Vbj is the total variable transport cost for the least-cost route between nodes b and j. Each road segment is assigned a cost per unit of biomass based on the road type, truck type, biomass moisture content and density, and speed limit of the assumed truck type on the given road type. Variable cost is distance- and/or-time dependent, therefore Vbj is represented by an equation to solve for the cost along each road segment. Equation (5) represents the generic form of a variable cost equation, where 2∗V<sup>x</sup> is the roundtrip transport cost for road segment x, and n is the total number of road segments in the least-cost path between nodes b and j. If rail transport is used, V<sup>x</sup> is not doubled as we assume there is no backhaul.

$$V\_{bj} = 2\sum\_{\mathbf{x}=1}^{n} V\_{\mathbf{x}} \tag{5}$$

Equation (6) represents the weighted average delivered feedstock cost, WA<sup>j</sup> , to a given depot j at a set facility design capacity (Sultana and Kumar, 2012).

$$WA\_j = \frac{\left(\sum\_{i=1}^{\mathcal{V}} TC\_{bj} B\_b\right)}{B\_j} \tag{6}$$

where TCbj is the total delivered feedstock cost of biomass source point b to depot j, B<sup>b</sup> is the biomass volume at source point b, y is the total number of biomass source points supplying depot j to meet facility demand, and B<sup>j</sup> is the total volume of feedstock delivered to depot j (i.e., facility demand). Where rail and road are compared along a linkage, the minimum of the two transport methods is selected to provide the least-cost route along the supply chain.

#### CASE STUDY

The depot-and-biorefinery siting model is applied in the Inland Northwest region of the United States, including western Montana, the panhandle of Idaho, and eastern Washington (**Figure 2**). Forest residues as a by-product of logging operations are assessed as a feedstock for the creation of isoparaffinic kerosene, or aviation biofuel. In this region where the Federal Government is the major landholder, forest residue is generated in significantly less quantities than regions where forests are primarily privately owned and where the climate is wetter (Martinkus et al., 2017a). Therefore, depots are proposed in this region to preprocess/pretreat residue.

The depots are modeled here using a three-stage dry-milling process to create micronized wood particles (∼30 microns) as a means of pretreatment (Wang et al., 2018). This process is utilized as no chemicals are necessary to break down the crystalline microstructure of lignocellulose, thus producing a clean sugar with no potential for causing catalyst poisoning or enzyme inhibitors (Brandt et al., 2018). Additionally, water is not needed, which lowers the environmental impact and the need for a wastewater treatment plant at the depot.

The depots must collectively meet an annual biorefinery demand of 300,000 bone dry metric tons (BDMT), accounting for 9% losses through the depots. We assume one large depot with an annual demand of 180,000 BDMT is co-located at the biorefinery to capitalize on nearby biomass, while two smaller satellite depots each procure 60,000 BDMT of forest residue annually. The biorefinery will create ∼11.3 million liters of aviation biofuel per year using an enzymatic hydrolysis, fermentation, and separation process (Gao and Neogi, 2015; Hawkins and Ley, 2016). The biofuel will supplement the Spokane, WA regional annual jet fuel demand with ∼20% of their year 2025 demand (Macfarlane et al., 2011; Federal Aviation Administration, 2016). A petroleum terminal located near the town's airport and military base is assumed to receive, blend, and store the fuel.

Biorefinery site requirements include a minimum lot size of 40.5 ha, access to natural gas, and a rail spur. Three facilities are evaluated as potential biorefineries: a decommissioned pulp mill in Frenchtown, MT; an active kraft pulp mill in Lewiston, ID; and, a greenfield site in Spokane, WA. The greenfield site is included to assess the hypothesis that repurposed facilities provide significant capital cost savings over greenfields. Primary wood processors

(e.g., sawmills and plywood mills) are considered as potential depot locations. Satellite depot siting considerations include access to natural gas, at least 5 ha of unused land for depot site development, and a rail spur. Additionally, where multiple mills reside in the same town or in close proximity, one representative mill is selected for analysis. Of the 27 primary processors in the region, 11 meet the siting requirements. The IFG Lewiston saw mill is co-located with the Lewiston pulp mill, therefore it acts as the large co-located depot as well as a potential satellite depot in this analysis. A large theoretical depot is assumed at the Spokane greenfield and at the Frenchtown Mill for co-location with the biorefineries (**Figure 2**).

#### Total Transportation Cost Model, TTCM

U.S. Forest Service Forest Inventory and Analysis (FIA) plots represent biomass source nodes (United States Forest Service, 2016). The Land Use Resource Allocation (LURA) bioeconomic model determines the projected 20-year average annual forest residue volume available at each FIA plot based on future timber market influences (Martinkus et al., 2017a; Latta et al., 2018). Similar to Chung and Anderson (2012), each FIA point is assumed to be a forest landing and is projected onto the nearest road for use in the TTCM. Each road segment is assigned a variable cost based on the material being hauled and associated truck type. Fixed and variable cost calculations for the three supply chain linkages are discussed below. See Martinkus et al. (2017a) for more detail on supply curve development.

#### Biomass Source-to-Depot

Based on work by Zamora-Cristales et al. (2013) and Martinkus et al. (2017a), a 6x4 chip van truck pulling a 13.7 m (45 ft) long drop center trailer is assumed for wood chip transport. The wood chips are assumed to have a moisture content of 35%, which translates into a payload of 14.1 BDMT. Fixed costs include transporting unmerchantable residue to a forest landing (\$16.5/BDMT), grinding the residue into chips (\$22.4/BDMT), and loading the chips onto a waiting chip van (\$3.9/ BDMT). The speed limit of the networked road dataset was modified based on average chip van speed and tractor-trailer weight loaded and unloaded on paved (70 km/h), gravel (15 km/h), and dirt (10 km/h) roads (Zamora-Cristales et al., 2013). The roundtrip variable unit trucking cost for each road type is based on known truck operating costs when loaded and unloaded (Zamora-Cristales et al., 2013; Martinkus et al., 2017a) (Equation 7).

$$TC\_{bj} = 47.2 + 0.258 \sum\_{p=1}^{n} t\_p + 0.194 \sum\_{\mathbf{g}=1}^{n} t\_{\mathbf{g}} + 0.184 \sum\_{d=1}^{n} t\_d \tag{7}$$

where TCbj is the total delivered feedstock cost (\$/BDMT) along a least-cost route between FIA point b and depot j, t<sup>p</sup> is the travel time (min) along all paved road segments, t<sup>g</sup> is the travel time (min) along all gravel road segments, t<sup>d</sup> is the travel time (min) along all dirt road segments, and n is the total number of road segments by type in the route.

#### Depot-to-Biorefinery

A 30,280 liter liquid tanker truck is assumed for transporting micronized wood to the biorefinery. This truck type was selected as a sealed vessel is needed to transport the wood particles and it must be pneumatically loaded and unloaded. Micronized wood has a moisture content of 10% and bulk density of 585 kg/m<sup>3</sup> (Wang et al., 2018) which translates to a payload of 16.3 BDMT. The road network was modified with speeds representative of this truck type (interstate-−96.5 km/h, U.S. highways-−80.5 km/h, and local roads, state, and county highways-−48 km/h). Fixed and variable transportation costs are derived from Parker et al. (2008), with liquid truck capacity converted to dry capacity (Equation 8). The fixed cost represents loading and unloading wait time.

$$TC\_{jk,t} = 8.54 + 2\left[1.98 \sum\_{t=1}^{n} \varkappa\_t + 0.05 \sum\_{d=1}^{n} \varkappa\_d \right] \tag{8}$$

where TCjk,<sup>t</sup> is the total delivered feedstock cost (\$/BDMT) between depot j and biorefinery k using truck transport, x<sup>t</sup> is the travel time (hrs) along road segment x, x<sup>d</sup> is the distance (km) along road segment x, and n is the total number of road segments in the least-cost route.

Rail transport was also assessed for this linkage using an equation derived from Parker et al. (2008). A 124,740 liter rail tanker is assumed to haul the micronized wood with a payload of the 80.5 BDMT. The fixed cost includes loading, unloading, and a charge for use of the railcar (Equation 9).

$$TC\_{jk,r} = 50.13 + 0.023 \sum\_{r=1}^{n} \wp\_r \tag{9}$$

where TCjk,<sup>r</sup> is the total delivered feedstock cost (\$/BDMT) between depot j and biorefinery k using rail transport, y<sup>r</sup> is the distance (km) along rail segment y, and n is the total number of rail segments in the least-cost route.

#### Biorefinery-to-Petroleum Terminal

The same rail and truck tanker types are assumed to transport aviation biofuel to the petroleum terminal in Spokane, WA. Therefore, the equations are the same with unit costs modified based on biofuel volume (Equations 10, 11).

$$TC\_{kl,t} = 0.56 + 2\left[0.10\sum\_{t=1}^{n} \chi\_t + 0.003\sum\_{d=1}^{n} \chi\_d\right] \tag{10}$$

$$TC\_{kl,r} = 2.92 + \ 0.001 \sum\_{r=1}^{n} \nu\_r \qquad \text{(11)}$$

where TCkl,<sup>t</sup> (\$/BDMT) and TCkl,<sup>r</sup> (\$/BDMT) are the total delivered feedstock cost between biorefinery k and petroleum terminal l for truck and rail, respectively, x<sup>t</sup> is the travel time (hrs) along road segment x, x<sup>d</sup> is the distance (km) along road segment x, y<sup>r</sup> is the distance (km) along rail segment y, and n is the total number of road or rail segments in the least-cost route.

#### Satellite Depot TEA and Decision Matrix

**Table 2** lists the operational cost components as identified in the satellite depot TEA, which was developed through modifying Brandt et al. (2018)'s TEA based on depot size and regional energy rates. The Brandt et al. (2018) TEA may be viewed in that manuscript's **Supplemental Information**. The first four cost components vary geospatially, thus are the siting criteria in the depot decision matrix. The resulting annual geospatial operational costs are converted into percentages of the total cost and then into weights (**Table 2**, Equation 1). TEA inputs include average county-level energy data (2011-2015) (U.S. Energy Information Administration, 2014a,b), weekly labor rates (2012-2015) (U.S. Bureau of Labor Statistics, 2015), and the average of all weighted average delivered feedstock costs for 60,000 BDMT of forest residue to each potential satellite depot.

Each criterion's range of regional or depot-specific values is additionally used to determine the depot decision matrix bin values (Equation 3) for use in facility assessments. The region is defined as all counties from which biomass is utilized in the supply chain or a potential depot resides (**Figure 2**). The feedstock criterion is measured as the weighted average delivered feedstock cost, WA<sup>j</sup> , for 60,000 BDMT of forest residue to depot j plus the transportation cost to deliver micronized wood from depot j to biorefinery k, TCjk (Equation 12). One depot decision matrix is created for each potential biorefinery, since the transportation cost from the depots to each biorefinery changes based on biorefinery location. **Table 3** shows the decision matrix used to assess all potential depots for the Spokane biorefinery supply chain. See the **Supplemental Information** (SI) for the Lewiston and Frenchtown depot decision matrices, associated potential depot scaled values and final scores, and breakdown of transportation costs between biomass source points-to-depots, and depots-to-biorefineries.

$$TC\_{bk} = WA\_j + TC\_{jk} \tag{12}$$

While the regional weighted average delivered feedstock cost is used in the TEA, the "feedstock" criterion here gives preference to those facilities that are efficient at procuring biomass, processing it into micronized wood and transporting preprocessed material to a biorefinery. Depot-specific criterion values (**Table 4**) are translated into scale values based on their position relative to the range of regional values (**Table 5**). Facility scores are determined using Equation (2), and are ranked from greatest to least (**Table 6**). Facilities with the highest scores would operate at lesser annual operational costs than those with lower scores.

The top two depots for both the Spokane, WA and Lewiston, ID biorefineries are IFG Moyie Springs and Vaagen Brothers Lumber, and the top depots for Frenchtown, MT are Plum Creek Evergreen and Vaagen Brothers Lumber. IFG Moyie Springs was selected as a depot for both Spokane and Lewiston because of (B)

TABLE 2 | (A) Satellite depot operational costs and (B) conversion into decision matrix weights.



\**Values may not sum exactly due to rounding.*

TABLE 3 | Satellite depot decision matrix for spokane biorefinery.


its significantly lower electricity cost as compared to all other locations, and Vaagen Brothers Lumber was selected as a depot for all three cases because of the low feedstock transportation and lower electricity costs. Even though feedstock transportation costs are higher, Vaagen Brothers Lumber was selected as a depot for Frenchtown because of its lower electricity and natural gas costs as compared to other locations. Further details are provided in the **Supplemental Information**.

After the initial depot ranks are completed, an optimization model is then used to assign biomass source points to each of the top two satellite depots and biorefinery co-located depot to meet the overall biorefinery demand of 300,000 BDMT based on minimizing the total delivered feedstock cost from each biomass source point to the biorefinery gate (**Figure 3**).

#### Biorefinery TEA and Decision Matrix

The biorefinery TEA used in this analysis is a modified version a TEA developed by Marrs et al. (2016) for a large integrated biorefinery (757,500 BDMT/year of feedstock) that creates aviation biofuel using a mild bisulfite solution to pretreat wood chips. The major changes to the original TEA include removal of the chemical pretreatment technology to reflect pretreatment occurring at depots, and scaling each department to the smaller biorefinery size used in this analysis.

In addition to energy, labor, and feedstock as biorefinery siting criteria, the infrastructure present at each potential biorefinery is included as a siting criterion (**Table 7**). This criterion is quantified as the estimated percent reduction in capital construction costs from a greenfield biorefinery through repurposing an existing facility. Percent reductions are determined using a method presented by Martinkus and Wolcott (2017), where a factored analysis is used to estimate cost savings through similarities in infrastructure and assets to a greenfield biorefinery. The total delivered equipment cost for all major processing components in a greenfield biorefinery are required, and all other ancillary assets and infrastructure are estimated as percentages of the total delivered equipment cost. A yes/no analysis is employed for all facilities assessed, where equipment cost is assigned to those components in the assessed facility that are not similar to the greenfield, and no cost is assigned to equipment that is present. All costs are summed to determine the estimated percent savings over constructing a greenfield biorefinery.

Weights for the biorefinery decision matrix are developed using Equation (1). Regional average energy rates are input into the biorefinery TEA. The feedstock cost for the TEA (Equation 13) is the average of all potential biorefinery weighted average delivered feedstock costs (Equation 14).

$$TC\_{bjk} = PC\_j + TC\_{jk} \tag{13}$$

$$WA\_k = \frac{\left(\sum\_{i=1}^{\mathcal{Y}} TC\_{bjk} B\_j\right)}{B\_k} \tag{14}$$

where TCbjk is the total delivered feedstock cost from depot j to biorefinery k, PC<sup>j</sup> is the processing cost, or minimum selling price, of micronized wood at depot j, TCjk is the transportation

#### TABLE 4 | Location-specific criterion values for each potential depot.


TABLE 5 | Potential depot scaled values for spokane biorefinery supply chain.


cost to deliver micronized wood to biorefinery k, B<sup>j</sup> is the volume of feedstock at depot j, B<sup>k</sup> is the total volume of feedstock to meet biorefinery demand, and y is the total number depots in the depot-and biorefinery supply chain. PC<sup>j</sup> is determined from the depot TEA by inputting depot-specific weighted average delivered feedstock cost and energy rates, which results in a final unit cost to produce one BDMT of milled wood plus profit. To determine the "facility infrastructure" criterion weight, the design biorefinery's total capital cost for construction, as identified in the TEA, is converted to an annualized cost assuming a plant life of 20 years and a discount rate of 8% (Martinkus et al., 2017b).

One decision matrix is developed to assess all three depotand-biorefinery supply chains so the least-cost supply chain may be identified. Bin values for energy and labor rates are the same as for the depot decision matrix, as the region definition is the same. Feedstock is now measured by the weighted average total delivered feedstock cost of the two satellite depots and one large depot to the biorefinery gate (WA<sup>k</sup> ) plus the cost to transport biofuel to the petroleum terminal (TCkl). Processing costs at the biorefinery are not included as they are reflected in the siting criteria. The range of facility values for feedstock and infrastructure assessment are used for their respective bin value determinations. Scoring and ranking of the proposed biorefinery supply chains is performed using the biorefinery decision matrix (**Table 8A**).

#### RESULTS AND SENSITIVITY ANALYSIS

**Table 8** displays the biorefinery decision matrix and facility assessments. Spokane was found to be the least-cost location for processing feedstock into aviation biofuel, although it is a greenfield. Infrastructure was not found to be a significant annual expense due to pretreatment occurring in the depots and not at the biorefinery, as seen in the infrastructure criteria weight of 1.0. As evidenced in **Figure 3**, Spokane is centrally located among large amounts of biomass and near depots with lesser energy rates, while Frenchtown incurs the greatest expenses due to higher energy rates and its remote location, which increases transportation costs. It must be noted that the feedstock costs in **Table 8B** are higher than an integrated biorefinery would pay for raw feedstock due to the milled wood being



TABLE 7 | (A) Biorefinery operational expenses and (B) Conversion to decision matrix weights.



\**Values may not sum exactly due to rounding.*

pretreated and the bulk density increased at the depot. Thus, the cost for pretreatment is not represented in the capital and operational expenditures of the biorefinery, but in the cost for feedstock.

A sensitivity analysis was performed to assess the variables and assumptions that influence the decision matrix results, including biomass availability, bin designations, and weights. Additionally, the decision matrix results are compared against an optimization routine to assess the accuracy of the matrix at selecting depots for each supply chain, and the least-cost supply chain for supplying the Spokane market with aviation biofuel.

#### Sensitivity Analysis Biomass Availability

Biomass availability is a function of biomass supply and the trucking assumptions represented in the TTCM. Biomass supply is considered to have a medium uncertainty due to the many variables that comprise it, including the seasonality of harvest operations and the amount of residue available and accessible at varying slopes and distances from the forest landing (Miller and Boston, 2017). Similar to Martinkus et al. (2017a), a high feedstock cost scenario was run to determine the increase in feedstock cost each biorefinery may incur to meet annual demand during years of low biomass availability. A low-cost scenario was run assuming the use of a blower for loading ground chips into the chip van at the forest landing, which was found to increase payload by 25% over traditional gravity-fed loading methods (Zamora-Cristales et al., 2014). The resulting delivered feedstock costs for each top-ranked depot were input into the depot TEA along with their respective electricity and natural gas rates to determine the revised fixed cost at each depot. This fixed cost was summed with the transport cost for hauling the micronized wood to each respective biorefinery to determine the total delivered feedstock cost (Equation 13, **Figure 4**).

The fixed cost at the forest landing is based on an assumed off-road diesel cost, equipment types and efficiencies, and a landowner payment assuming a weak market for forest residue. The variable cost for residue transport to depots is based on a 30 year average diesel cost of \$0.93/liter and a set chip van size. Han and Murphy (2011) found that a 10 percent increase in fuel cost resulted in a 3 percent increase in total transportation costs. The fixed and variable costs for the forest-to-depot linkage are used with low uncertainty, as time-motion studies were performed to develop the costs (Zamora-Cristales et al., 2013, 2015). The fixed and variable costs for the linkages between the depots, biorefineries, and petroleum terminal are used with medium uncertainty, as one reference was used (Parker et al., 2008) to develop the tanker truck total cost equation and only a diesel cost of \$0.66/liter was provided as insight into their cost derivations. The rail costs from Parker et al. (2008) were compared against other sources (Gonzales et al., 2013; Lewis et al., 2015; USDA Agriculture Marketing Service, 2015) and determined to be midrange among all costs and therefore acceptable for use.

#### Bin Designations

The importance of region definition on bin designations was assessed through evaluating energy averages in the region defined

by all counties in Oregon, Washington, Idaho, and Montana (here called the Pacific Northwest, or PNW). The electricity average increases marginally, from \$0.061 to \$0.064/kWh for the PNW, with the increase in region boundary size, while the natural gas average decreases slightly from \$0.27 to \$0.25/k.c.m. for the PNW. While the means did not change significantly, the minimum and maximum values for each is noticeably different. The PNW electricity range was \$0.028 to \$0.131/kWh while the TABLE 8 | (A) Biorefinery decision matrix, (B) Location-specific values, and (C) Scaled values with final facility scores.



study region range was \$0.036/kWh to \$0.091/kWh. Similarly, the PNW natural gas range was \$0.16 to \$0.33/k.c.m while the study region range was \$0.19 to \$0.31/k.c.m. The difference in range affects bin designations in the decision matrices, and thus affects the spread of scaled values. If the PNW range was used in this analysis, the facility scaled values would lose their granularity due to the larger range of values, and thus more facilities would be assigned the same scale value. Electricity, natural gas, and labor rates all come from reputable government agencies and are considered to have low uncertainty.

#### Weighting Analysis

A weight sensitivity analysis was performed by varying the cost of feedstock, electricity, and natural gas rates (**Table 9**). In the depot decision matrix, the only major change in weights occurs when electricity is set at the lowest rate in the study region. Electricity is the highest annual cost in the depot TEA; by lowering the rate substantially, feedstock then becomes the highest annual cost. In the biorefinery TEA, feedstock cost is significantly higher than all other costs, therefore an increase in electricity or natural gas rates is not significant in overall weighting.

Labor was not assessed because national average salaries were used for the various positions identified in the TEAs (such as plant engineer, shift supervisor, yard employees, etc.) as opposed to using county-level labor rates to calculate annual wages. Additionally, changes in the percent reduction in infrastructure was not assessed. Both of these criteria are low cost components in the overall annual cost to operate the depot or biorefinery, therefore changes in their values would not significantly affect

the other weights, as evidenced by evaluating a high natural gas rate.

#### Validation of Decision Matrix Results

An optimization routine using binary integer programming was created to identify the top two satellite depots for each biorefinery as a comparison against the depots selected by the decision matrix. The **Supplemental Information** describes the optimization routine. The optimization model minimizes total cost for each scenario including the total annual cost to procure, purchase, process, and transport the biomass/biofuel to the end user. The difference between this minimum cost and the total cost associated with the depots selected by the decision matrix represents the opportunity cost, or forgone value, of the approach. The results are displayed in **Table 10**. The optimization model and depot decision matrix were in agreement for the depots selected for Lewiston, ID, and were only marginally different in total cost for the depots selected for Spokane, WA. The depots selected by the decision matrix for Frenchtown, MT would increase the total cost to operate the supply chain over the optimization model results by close to \$1 million dollars, however this equates to only a 1% increase in total annual cost over the optimized supply chain.

## DISCUSSION

The results of the optimization comparison (**Table 10**) show the strength of the decision matrix in its ability to take the complex issue of multiple cost variables at a given site and simplify it through applying weights to the costs based on their importance in the facility's operating budget. The Spokane supply chain had the least annual cost of the potential biorefineries, and thus would most likely be the best location for a biorefinery. The optimization run for Spokane selected the second and third depots from the decision matrix ranking list, for Frenchtown it selected the second and fourth depots from the list, and for Lewiston, the top two depots were identified as optimal (**Tables 6**, **10**). Vaagen Brothers and IFG Moyie were consistently selected as top depots for each supply chain due to their ease of procuring feedstock and low electricity rate, respectively. These results, combined with the percent difference from optimized cost (**Table 10**), validate the use of the decision matrix as a strategic-level facility siting tool for identifying the top depots that provide biomass to a facility at the least cost. Stakeholders interested in using the decision matrix for biorefinery supply chain development would engage representatives from the top few-selected depots in discussions around contracts and pricing, then from those conversations select the final depots for use in a depot-and-biorefinery supply chain.

Both the depot and biorefinery geospatial cost components comprise ∼80% of their respective total facility operational costs (**Tables 2**, **7**). Thus, their use as siting criteria is relevant for assessing each potential facility's location-specific cost components and the effect it has on the total annual amount spent on operational costs. The assessment of high-cost biomass gives an indication of the additional amount biorefineries may pay during years of low feedstock availability, and the low-cost assessment shows how technological advances can translate into reduced feedstock costs at the depot mill gate. **Figure 4** indicates that Frenchtown is the most vulnerable to years of high biomass cost since it exhibits the largest range in feedstock pricing. The Spokane biorefinery would operate at lower costs than Lewiston or Frenchtown, even during years of low biomass availability, as evidenced by the compactness of its biomass procurement and transporting in **Figure 4**.

Region definition plays an important role in bin designations and therefore in facility scores. When regions are defined to be much larger than the study area, the range of each bin may increase with an increase in the minimum and maximum values for each criterion. This may result in more facilities being assigned the same scale values, and thus will reduce the range of facility scores.

Just as rates and assets vary geospatially, the annual amount spent on each cost component varies based on the design of the depot and biorefinery. For example, wood-based integrated biorefineries require significant amounts of electricity and natural gas for preprocessing woody biomass into pulp (Zhu and Zhuang, 2012), whereas biorefineries in a depot model require less energy due to preprocessing occurring at depots. The feedstock cost at the biorefinery gate in this analysis was larger than would be expected at an integrated biorefinery; however, the feedstock cost reflects the cost of pretreatment occurring at the depots. While the feedstock cost is larger, the capital and operational costs of the biorefinery are less than an integrated biorefinery due to the disaggregation of pretreatment from the facility. This was evidenced by the selection of Spokane as the least-cost location to construct a biorefinery, even though it is a greenfield. The annualized infrastructure (capital cost) to construct the biorefinery was around 2% of the total operational costs (**Table 7**), whereas in an integrated biorefinery, infrastructure may play a larger role as pretreatment can be a significant capital cost depending on the conversion technology. Additionally, feedstock costs are a function of the size and number of depots used in the supply chain model. This analysis was based on the a priori assumption of two small satellite depots and one large colocated depot at the biorefinery. Additional modeling through either optimization runs or varying the assumption of depot size and number, and creating decision matrices that reflect these assumptions, will allow for identification of the depotand-biorefinery configuration that provides the least overall supply chain costs. As depots become larger, economies of scale allow for more efficient processing of biomass, which is seen in a lower minimum selling price of preprocessed biomass to the biorefinery. Larger depots may equate to fewer depots in a supply chain, which can lessen transportation costs to the biorefinery as well. Cost sharing that may occur between depots and primary processors through co-location was not modeled here. Significant depot operational cost savings may be gained through sharing of staff, energy, residuals, etc.; however, further research is necessary to quantify these potential cost savings.

Similar to Richardson et al. (2011), we found that fixed costs account for a significant portion of the delivered feedstock cost from the forest to the depot. Technological advances may reduce this cost. However, as the biomass market becomes commercialized, landowner payments may increase (U.S. Department of Energy, 2011) and processing costs may decrease due to economies of scale. Rail was found to be utilized only for the furthest depot from the biorefinery in each supply chain analysis. This is consistent with personal communication with members of the freight industry that say rail is not feasible until ∼320 km.

The TEAs are constructed using ratio factors from methodology presented by Peters et al. (2003) to estimate total capital investment. Operational and capital expenses are estimated from equipment quotes and literature references. An economic analysis with a real discount rate of 10 and 2%

#### TABLE 9 | Weighting sensitivity analysis for (A) depot and (B) biorefinery decision matrices.



*Average, study region and depot-specific averages; Low Cost Biomass, 25% increase in payload from loading ground chips with blower; High Cost Biomass, low yield; Low Electricity, lowest electricity rate in study region (*∼*30% below regional average); High Natural Gas, highest natural gas rate in study region (*∼*10% above regional average).*


inflation was used to determine minimum selling price of both the micronized wood and aviation biofuel (Petter and Tyner, 2014).

The decision matrices presented here require depot and biorefinery sizes to be selected a priori. The most likely scenario for use would be stakeholders that want to identify potential depots for pre-selected potential biorefineries with a given annual feedstock demand. If using an optimization routine is not feasible for selecting the optimal size and location of facilities, then multiple iterations of the decision matrices may be run to assess the change in cost of the depot-and-biorefinery supply chain with different sized facilities. This requires the use of multiple TEAs, as each TEA is built for a specific facility size. The biorefinery decision matrix can also be used independently for siting an integrated biorefinery, as performed by Martinkus et al. (2017b).

Overall, the proposed decision matrix design performs well as a facility site selection tool. The advantage of the tool is its simplicity in identifying, combining, and weighting cost components that most affect the overall cost to construct and operate a facility. While the decision matrix was shown to perform well against more a complex optimization model, one disadvantage is that optimization modeling is still necessary to assign biomass source points to the final selected depots to estimate the delivered feedstock cost. Additionally, a TEA for each facility type (i.e., depot, biorefinery) is needed to perform siting criteria identification and analysis. Many TEAs exist online for biorefinery types, yet not all technologies are represented and location specific details are not always stated explicitly. Finally, the repurpose potential of existing facilities may be difficult to estimate. Information used to determine the presence or absence of equipment and assets at each facility may be located in national/state sources or datasets and from aerial imagery (Martinkus and Wolcott, 2017).

# CONCLUSIONS

We propose a quantitative facility siting tool to identify the least-cost regional biorefinery supply chain from an array of potential depot and biorefinery locations. Co-location and repurpose strategies are assumed for existing primary processors and pulp mills, respectively, as a means to reduce capital and operational costs. Decision matrices provide a quantitative, transparent MCDA tool for assessing and prioritizing the various operational cost components that are present at each site. A total transportation cost model is utilized to quantify the feedstock procurement and processing costs at each linkage of the supply chain. An optimization routine validated the results of the decision matrices in the selection of depots for each biorefinery, and for selecting the least-cost depot-and-biorefinery supply chain.

By performing facility siting through assigning weights for feedstock, energy, and labor costs based on their importance in the operational costs of the facility, biorefinery supply chain development can be better directed to select facilities that provide the greatest cost reductions. Any cost reductions gained during the early phase of commercialization may translate into a more cost-competitive biofuel for cellulosic and advanced biorefineries.

#### AUTHOR CONTRIBUTIONS

NM wrote the paper, created the decision support tools, developed all datasets needed to populate the decision matrices, and ran the siting analyses and sensitivity analyses to determine the best location for a biorefinery. GL created the LURA model and ran the model to determine 20-year average forest residue volumes at all FIA points used as biomass source points. GL also created and ran the optimization model to validate the results from the decision matrices. KB

#### REFERENCES


developed the TEAs that were used in the siting analyses for the determination of operational and capital costs at each facility. MW provided project oversight and review of the paper.

#### ACKNOWLEDGMENTS

The authors gratefully acknowledge funding from the Northwest Advanced Renewables Alliance (NARA), supported by the Agriculture and Food Research Initiative Competitive Grant no. 2011-68005-30416 from the USDA National Institute of Food and Agriculture. This research was also funded by the U.S. Federal Aviation Administration Office of Environment and Energy through ASCENT, the FAA Center of Excellence for Alternative Jet Fuels and the Environment, project COE-2014-01 through FAA Award Number 13-C-AJFE-WaSU under the supervision of James Hileman and Nathan Brown. Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the FAA.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00124/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Martinkus, Latta, Brandt and Wolcott. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# High Throughput Screening Technologies in Biomass Characterization

Stephen R. Decker <sup>1</sup> \*, Anne E. Harman-Ware<sup>1</sup> , Renee M. Happs <sup>1</sup> , Edward J. Wolfrum<sup>1</sup> , Gerald A. Tuskan<sup>2</sup> , David Kainer <sup>2</sup> , Gbekeloluwa B. Oguntimein<sup>3</sup> , Miguel Rodriguez <sup>2</sup> , Deborah Weighill 2,4, Piet Jones 2,4 and Daniel Jacobson2,4

*<sup>1</sup> National Renewable Energy Laboratory, Biosciences Center, Golden, CO, United States, <sup>2</sup> Oak Ridge National Laboratory, Biosciences Division, Oak Ridge, TN, United States, <sup>3</sup> Department of Civil Engineering, Morgan State University, Baltimore, MD, United States, <sup>4</sup> The Bredesen Center for Interdisciplinary Research and Graduate Education, University of Tennessee, Knoxville, TN, United States*

#### Edited by:

*J. Richard Hess, Idaho National Laboratory (DOE), United States*

#### Reviewed by:

*Ruchi Agrawal, Indian Oil Corporation, India Alok Satlewal, Indian Oil Corporation, India*

> \*Correspondence: *Stephen R. Decker steve.decker@nrel.gov*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *01 June 2018* Accepted: *23 October 2018* Published: *27 November 2018*

#### Citation:

*Decker SR, Harman-Ware AE, Happs RM, Wolfrum EJ, Tuskan GA, Kainer D, Oguntimein GB, Rodriguez M, Weighill D, Jones P and Jacobson D (2018) High Throughput Screening Technologies in Biomass Characterization. Front. Energy Res. 6:120. doi: 10.3389/fenrg.2018.00120* Biomass analysis is a slow and tedious process and not solely due to the long generation time for most plant species. Screening large numbers of plant variants for various geno-, pheno-, and chemo-types, whether naturally occurring or engineered in the lab, has multiple challenges. Plant cell walls are complex, heterogeneous networks that are difficult to deconstruct and analyze. Macroheterogeneity from tissue types, age, and environmental factors makes representative sampling a challenge and natural variability generates a significant range in data. Using high throughput (HTP) methodologies allows for large sample sets and replicates to be examined, narrowing in on more precise data for various analyses. This review provides a comprehensive survey of high throughput screening as applied to biomass characterization, from compositional analysis of cell walls by NIR, NMR, mass spectrometry, and wet chemistry to functional screening of changes in recalcitrance via HTP thermochemical pretreatment coupled to enzyme hydrolysis and microscale fermentation. The advancements and development of most high-throughput methods have been achieved through utilization of state-of-the art equipment and robotics, rapid detection methods, as well as reduction in sample size and preparation procedures. The computational analysis of the large amount of data generated using high throughput analytical techniques has recently become more sophisticated, faster and economically viable, enabling a more comprehensive understanding of biomass genomics, structure, composition, and properties. Therefore, methodology for analyzing large datasets generated by the various analytical techniques is also covered.

Keywords: biomass recalcitrance, biomass compositional analysis, high throughput analysis, neural networks, biomass conversion

# INTRODUCTION

Crop breeding for improved traits has been in constant development since mankind first intentionally put seeds in the ground. Selection of better crops has been, until recently, a long-term strategy, with simple selection criteria of higher yield or increased tolerance to various stresses. With the advent of molecular genetics, plant breeding has accelerated significantly, with specific genes and entire metabolic pathways being added or deleted in a few days, however the results are not always selectable with simple phenotypic criteria. While plants continue to require weeks, months, or years to fully mature and exhibit stable traits, phenotypes such as reduced recalcitrance, altered polysaccharide and lignin content, and other changes in cell wall composition are not readily discernable. Technologies to rapidly evaluate engineered plants for non-phenotypically evident resultant traits has not kept pace with the generation of these mutations, nor has the analytical methodology needed to catalog, interpret, compare, and model the huge genetic information being uncovered gene by gene.

As plant engineering methodologies have evolved to generate more variants in shorter time, screening these expanded sample sets has ranged from cumbersome to nearly impossible, depending on the analyses and turnaround required. Simple changes to phenotype such as biomass yield can be readily evaluated by increasing the area planted to accommodate more variants and replicates and measuring height, width, mass, etc. by hand. Given enough researchers (most probably students), several thousand plants can be evaluated in this manner, however, understanding the more intricate specifics of compositional analysis, conversion potential, and chemical changes to components such as lignin require much more detailed analyses.

Over the past decade or so, multiple research groups have developed methods to address some of these complex measurements. In this review, we cover advances in measuring changes in plant cell wall chemistry and recalcitrance, developments in rapid sugar and fatty acid analysis, advances in spectroscopy applications and instrumentation to cell wall composition determination, and use of data analysis and systems biology modeling to develop understandings from the acquired data.

**Table 1** is a summary of HTP methods used in HTP analysis of biomass. Details for each method can be found in sections High Throughput Plant Cell Wall Compositional Analysis, HTP Recalcitrance Screening, NMR for Biomass Sugar and Fatty Acid Composition and in the associated references. Data analysis and systems biology modeling is handled separately in section Data Analysis and Systems Biology Modeling for High Throughput Biomass Analysis.

#### HIGH THROUGHPUT PLANT CELL WALL COMPOSITIONAL ANALYSIS

Compositional analysis of biomass is a complex, but necessary and important analytical technique. The ratio, composition, and content of the three major cell wall components (cellulose, hemicellulose, lignin), have a direct impact on the technoeconomics of biomass conversion to products and knowing the potential yields is the best way to standardize the cost analysis across different feedstocks. Several methods have been developed to deconvolute the complexity of the plant cell wall down to the subunit level. Such methods are laborious, slow, and employ a variety of harsh reagents requiring some degree of remediation (Elliston et al., 2015). Destructive methods, such as 2-stage acid hydrolysis and pyrolysis-Molecular Beam Mass Spectrometry (py-MBMS), rely on breaking apart the various polymers and measuring the content of the resultant subunits, which can then be mathematically reassembled to estimate the various polymer content. The use of simpler and more rapid spectroscopic methods have proved invaluable in analyzing biomass (Elliston et al., 2015). Non-destructive spectroscopic methods such as Near InfraRed (NIR) spectroscopy rely on specific bond detection and quantitation, which is then fed to a multivariate model in order to predict the content of the various polymers, using samples previously characterized by destructive methods to build the model. While this latter method is fast and non-destructive, it is not a primary measurement and requires significant upfront efforts to build good models. Any samples that fall outside the predictive range of the model cannot be characterized with any precision. While the destructive methods provide direct data, they can be very slow (2-stage acid hydrolysis), require detailed analytical methods to measure products, and can use a lot of sample material. The NIR and py-MBMS methods have been reviewed by Xiao et al. for readers interested in more details (Xiao et al., 2014).

#### Micro-Scale 2-Stage Acid Hydrolysis

Decades ago, the National Renewable Energy Laboratory established several standard Laboratory Analytical Procedures (LAPs) designed to standardize analytical methods in an emerging and very chemically complicated landscape of biomass conversion. Basing many of these LAPs on tried-and-true methods from pulp and paper and forage analysis, these protocols became a standard yardstick by which disparate labs could meaningfully measure complex parameters of biomass conversion. The procedures established were highly regimented and geared toward the bench-scale, i.e., using 1–10 g of material and requiring several days for completion. A reasonably welltrained technician could be expected to handle 40–50 samples per week for compositional analysis and enzyme hydrolysis and/or Simultaneous Saccharification and Fermentation (SSF). A large fraction of that time taken up by the manual manipulations of weighing the samples and taking representative samples, as well as the relatively long analysis times (30–60 min/sample) via HPLC.

The bench-scale compositional analysis LAP requires 0.3–3 g of biomass to be hydrolyzed first by 72% (w/w) H2SO<sup>4</sup> at 30◦C and then by 4% (w/w) H2SO<sup>4</sup> in an autoclave, sequentially hydrolyzing cellulose and hemicellulose to oligomers and then to monomers, but requiring the use of pressure-rated glass vessels (Sluiter et al., 2008). The severe conditions result in degradation of sugars to hydroxymethyl furfural, furfural, levulinic acid, and other products as they are released from the polymer matrix, necessitating the use of sugar recovery standards to account for those losses. The sugars are typically measured via HPLC with Pb-, Ca-, or H-based ion exchange resins used to resolve the various sugars. The best separation and quantitation is typically found using the Pb-based chemistry, however these columns require neutralization of samples (CaCO3), long run times (∼45– 60 min), and de-ashing guard columns to protect the analytical TABLE 1 | Summary of HTP methods in biomass conversion.


column Pb-groups from complexing with the neutralization CO−<sup>2</sup> 3 ions. For a sample set of 50, running them in triplicate and with various sugar, sugar recovery, and instrument validation standards, could easily tie up an HPLC for a week or longer. The soluble lignin was estimated by absorbance and the insoluble lignin, whether natural or precipitated by the severe conditions, was measured gravimetrically, adjusting for inorganic content as determined gravimetrically after ashing the sample.

In 2011, Selig et al. published the first HTP method for measuring glucan and xylan content in plant cell walls (Selig et al., 2011). The protocol was primarily a scaled-down version of the NREL LAP, with enzyme-linked spectrophotometric measurements replacing HPLC as the primary sugar analysis. This limited sugar detection to glucose and xylose only, but allowed for parallel processing of multiple samples and automated quantitation by microtiter plate (MTP) spectrophotometers while only using 50 mg of sample for each analysis. While liquid handling and absorbance reading steps were all automated using various robots, transfer between the robots was carried out by hand. Solids dispensing was automated into special dispensing plates that were used to manually add samples to the custom Hastelloy deep 96-well reactor plates. As manual mixing during the acid hydrolysis was not practical, sonication was used to maintain sample dispersity and special clamping mechanisms were used to seal the reactor during the digestion. Other adaptations included the inclusion of strong buffering of diluted aliquots instead of whole sample neutralization to minimize differences in pH which affected the enzyme-linked assays, and the use of centrifugation to minimize solids interference with pipetting by the robots. While the scaled-down version of the compositional analysis method was fast and reproducible, it did not track exactly 1:1 with standard bench-scale analysis (due mainly to the change in sugar detection and the simple error propagation of using small masses and volumes), and should therefore only be used in a comparative manner within a given sample set and not as a precise analytical tool. It also suffered from the requirement of highly specialized, expensive, and very heavy custom reactor plates. Subsequent refinements to the method substituted 96-well format glass tubes for the Hastelloy reactor plate, eliminating the expensive custom reactor, the need for problematic sealing films, and allowed for the use of an autoclave instead of a Parr reactor for the 2nd stage of hydrolysis. Extraction and enzymatic de-starching of the starting material, especially for herbaceous feedstocks, was also worked into the method after the initial publication (Decker et al., 2012).

DeMartini et al. (2010) developed a similar method around the same time, utilizing HPLC vials instead of Hastelloy reactors or 96-well format glass tubes (DeMartini et al., 2010). The use of readily available HPLC vials allowed for automated solid and liquid dispensing, as well as heating on the Symyx Core Module robot, however subsequent manipulations, such as centrifugation, transfer of supernatant to polypropylene tubes, neutralization with CaCO3, and analysis by HPLC required individual manual operations. The other advantage of the HPLC vials, however, was the ability to estimate insoluble lignin content gravimetrically after washing and drying the residual solids and subtracting previously determined ash content, even if the washing steps were carried out by hand.

Foster et al. (2010) developed a small-scale protocol for plant cell wall compositional analysis as well, however the protocol utilized numerous solvent addition and removal steps as well as sample extraction and de-starching. The starting material was mechanically sized-reduced biomass from essentially an automated small-scale ball mill. These complex steps precluded ready automation, but the protocol provided much of the same information as larger-scale standard analyses while using considerably less material (Foster et al., 2010).

Though several labs have developed HTP/micro-scale compositional analysis methods, it remains a major tenet of this work that these methods are not as precise and accurate as largerscale, lower throughput methodologies. Much of this is tied up in the heterogeneity of the samples, especially when a single particle of bark or rind can comprise 10% or more of a single sample or when automated dispensing results in size-fractionation of the bulk material during repeated sample aliquoting. Primarily, these methods are useful in ranking large samples sets in terms of cellulose content or percent theoretical conversion in subsequent enzymatic or microbiological conversion. They also result in significant savings in reagent costs, sample prep, and technician time, however the costs of the robots, especially those designed for accurate and reproducible solids dispensing in the 1–10 mg range, is prohibitive for many research groups.

### Pyrolysis-MBMS for Analysis of Cell Wall Composition and Other Components Present in Lignocellulosic Biomass (LCB)

The analysis of biomass by analytical pyrolysis techniques has been practiced for decades as it can provide a significant amount of information regarding the structure and composition of biomass as well as inform pyrolysis processes, conditions and upgrading strategies that are used to generate bio-oil. Fast pyrolysis, the rapid thermal decomposition of material in the absence of oxygen, produces analytes that originate from different components in the feedstock that can be analyzed using detectors such as mass spectrometers. Pyrolysismolecular beam mass spectrometry (py-MBMS) is an analytical technique that uses a pyrolyzer coupled to a molecular beam mass spectrometer to analyze all ions generated without chromatographic separation. Fast pyrolysis of samples is typically performed at temperatures between 300 and 700◦C in the timescale of < 1 min/sample. Electron ionization is used to generate ions with voltages ranging from <20 eV up to 70 eV and ions are detected after passing through a quadrupole which typically scans for the analysis of m/z 30–450. Py-MBMS has been used as a high throughput technique to analyze LCB for estimation of lignin content, syringyl/guaiacyl (S/G) ratios, sugar composition as well as diterpenoid resin acid content in biomass extract. The benefits of using py-MBMS for biomass analysis include rapid throughput (250 samples/day), minimal sample preparation and low sample amount requirements (10 mg). The high-throughput nature of py-MBMS has enabled studies across large populations and sample set sizes allowing for the incorporation of appropriate statistical and biological variations in data. Py-MBMS data can be analyzed using various statistical tools such as principle component analysis (PCA), partial least squares regression (PLS), clustering methods and other predictive analytics that provide both quality control measures and a means to understand the underlying spectral patterns as well as their sources and associations with other biomass properties and genomics.

Lignin content in biomass samples can be estimated relative to a standard of known Klason lignin content run in the same set of samples by py-MBMS through summation of ion intensities that originate from phenolic species generated during pyrolysis which may include m/z 120, 124, 137, 138, 150, 152, 154, 164, 167, 168, 178, 180, 181, 182, 194, 208, and 210 (Sykes et al., 2008). Relative syringyl (S) monomers in lignin are typically calculated by summation of ion intensities of m/z 154, 167, 168, 182, 194, 208, 210; whereas relative guaiacyl (G) monomeric values can be determined by summation of ion intensities of m/z 124, 137, 138, 150, 164, 178. S/G ratios are then determined by dividing the sum of S-based ions by the sum of G-based ions. **Table 2** summarizes the sources of various ions seen in py-MBMS spectra of biomass and example spectra of maize stems are shown in **Figure 1** (Penning et al., 2014a). Otherwise, PLS models can be used to determine lignin content in a set of samples relative to wet chemistry methods using a variety of standards. The analysis of lignin content and S/G ratios in lignocellulosic biomass by py-MBMS has been reported extensively in the literature and has been used in studies focused on the analysis of biomass recalcitrance, (Studer et al., 2011; Biswal et al., 2015, 2018; Decker et al., 2015; Sykes et al., 2015b) for genetic studies including as QTL and GWAS, (Wegrzyn et al., 2010; Penning et al., 2014a; Muchero et al., 2015) to determine within-plant variability and the effects of environmental conditions on cell wall structure (Sykes et al., 2008; Mann et al., 2009), and for the analysis of biomass either engineered or selected for different lignin content and monomer compositions (Penning et al., 2014b; Sykes et al., 2015b; Edmunds et al., 2017).

Structural carbohydrate or sugar composition has also been estimated in a variety of LCB types by py-MBMS. Ions that can originate from C5 sugars (xylose, etc.) include m/z 57, 73, 85, 96, 114 and ions attributed to C6 sugars (glucose, etc.) include m/z 57, 60, 73, 98, 126, 144. The pyrolysates from which the carbohydrate-derived ions originate include anhydrosugars, furans, low molecular weight aldehydes and ketones and other compounds as shown in **Table 2**. In a study by Sykes et al. (2015a) structural carbohydrates were determined using a standard method involving low-throughput two-stage acid hydrolysis of biomass followed by HPLC analysis of the hydrolysates. The HPLC method was used to build a PLS model with py-MBMS spectra for the prediction of sugar content in various types of biomass. The best PLS models incorporated different biomass types (hardwood, softwood, etc.) which extended the range in composition of each of the major sugars predicted. While models used to predict sugar composition within a biomass type were not ideal, py-MBMS analysis for sugar composition of biomass has a significant increase in throughput over traditional methods using twostage acid hydrolysis followed by HPLC analysis (Sykes et al., 2015a).

Diterpenoid resin acids from pine and pine extracts have also been analyzed using py-MBMS. Typically, diterpenoid resin acids are extracted from coniferous biomass and derivatized prior to analysis by GC. The derivatizing step of resin acids is cumbersome, time consuming and consumes an additional step involving a derivatizing reagent. Harman-Ware et al. (2017) developed a high-throughput method for the analysis of diterpenoid resin acids from the organic extract obtained from pine sapling cross sections using py-MBMS. While the py-MBMS analysis of pine biomass correlated spectral patterns indicative of variable levels of diterpenoid resin acids with GC data, it was not possible to use or spike standards for the quantification of these compounds in the biomass samples. Additionally, the low content (<5 wt%) of the resin acids made reliable quantification by analysis of the whole biomass difficult and the variability within trees complicated sampling methodology. Instead, total diterpenoid content was determined by py-MBMS analysis of the organic extract left after evaporation by means of an external calibration standard consisting of a mixture of components closely resembling the composition as determined by GC (Harman-Ware et al., 2017).

The analysis of biomass by py-MBMS is limited to non-volatile samples and analytes (for quantification) and small sample size. Therefore, harvesting, sample preparation, and variability of biomass are all experimental considerations that must be made in advance and understood when interpreting spectral data. Other considerations that must be made when interpreting spectra include the presence of inorganic salts and other compounds (proteins, lipids, etc.), the pyrolysis temperature, the fragmentation energy of the ionization source and other parameters that can influence the pyrolysis and subsequent spectra of a given sample. Also, it is not recommended to compare samples analyzed when instrument maintenance and tuning adjustments have been made as the spectra could change slightly. Typically, standards are run within a set and samples that need to be compared are run within a set without maintenance down time of the instrument between sample analyses. As no chromatographic separation has occurred, the sources of the ions present in the spectra must also be interpreted with caution as many analytes may produce the same ions.

Py-MBMS data has been used to analyze large data sets with a focus on a small number of ions present in the spectra. The mining of MBMS spectral data for clustering, mapping and making spectral associations with genotypes and phenotypes will extend the usefulness and capabilities of py-MBMS analyses as more genetic and phenotypic information about biomass is elucidated. Improvements in mass resolution and sensitivity would also extend capabilities for py-MBMS to analyze components that are difficult to identify and quantify and/or make up a smaller fraction of biomass such as metabolites, lipids, etc. While py-MBMS has proven to be instrumental in the elucidation of biomass characteristics relating particularly to lignin, there may still be untapped information present in the TABLE 2 | Summary of ion assignments in py-MBMS spectra of lignocellulosic biomass or biomass extracts.


135, 148, 195, 197, 213, 219, 237, 239, 254, 285, 287, 300, 302

Abietic acid, dehydroabietic acid, neoabietic acid, fragmentation ions from methyl losses, etc. Diterpenoid resin acids

spectra that could potentially inform other useful properties of biomass.

#### NIR Spectroscopy for Structural Components

As an alternative to the wet chemistry method, Near Infrared (NIR) spectroscopy has been used for decades for the rapid analysis of biomass, starting with the prediction of forage quality (Norris et al., 1976; Shenk et al., 1979; Abrams et al., 1987). Its use for the prediction of biomass composition in a biorefinery context originated later (Sanderson et al., 1996; Hames et al., 2003). An overview of the technique and a comprehensive review of its use in biomass conversion processes was recently published (Skvaril et al., 2017).

Rapid analysis using NIR is considered a secondary analytical method, because NIR spectra are correlated with primary biomass compositional analysis data using multivariate calibration algorithms to produce a calibration equation. This equation can then be used to predict the composition of samples in lieu of primary biomass compositional analysis. Robust calibration methods, including methods for estimating calibration and predicting have been developed (Martens and Næs, 1992; Martens and Martens, 2001; Olivieri Alejandro et al., 2006; Zhang and Garcia-Munoz, 2009).

Because the method relies only on the collection of a NIR spectrum for a given sample, the technique can provide compositional analysis data much more quickly than traditional chemical analysis methods. Many calibration models for biomass feedstocks and process intermediates have been published, including contributions from these authors (Wolfrum and Sluiter, 2009; Godin et al., 2011; Liu et al., 2013; Sluiter and Wolfrum, 2013; Payne and Wolfrum, 2015). These models have been used for the prediction of large numbers of samples, saving substantial amounts of time (and cost) compared to conventional laboratory analysis (Pordesimo et al., 2005; Templeton et al., 2009).

Care must be taken when using NIR spectroscopy for rapid biomass analysis. Because it is a secondary method, a prediction of biomass composition will be accurate only if the underlying sample set used to develop the calibration equation contains sufficient compositional variability compared to that expected in the sample population. Unknown samples that are not part of the calibration population are inevitably poorly predicted. If careful attention is not paid to this issue, unreliable and even misleading results will be obtained. Poorly predicted samples (flagged either as outliers or with large prediction uncertainties) can be excellent candidates for improving a calibration model. If these samples undergo compositional analysis using traditional analytical methods, they can be added

to the existing calibration model to extend its predictive range; by definition, these new samples are within the calibration population.

As discussed above, high-throughput analysis using NIR spectroscopy involves (1) collecting a NIR spectra of a biomass sample, and then (2) predicting one or more chemical properties of the biomass sample by comparing the collected spectra to a collection or library of samples for which spectral and chemical properties are known. There are opportunities to make improvements in both areas.

Spectrometer manufacturers are constantly improving the performance of NIR spectrometers, such as increasing the signal-to-noise ratio, the spectral resolution, or the spectral range of instruments, and it is likely that these improvements in performance will continue in the future. While these improvements require additional research and development on conventional spectrometers and are important advances, they represent incremental improvements to existing technology. There are two different and complementary approaches that offer opportunities for substantial improvement in high-throughput analysis using NIR spectroscopy: NIR hyperspectral imaging and low-cost, ultra-portable NIR spectrometers.

NIR hyperspectral image cameras are conceptually similar to conventional cameras except the pixels of the two-dimensional images produced consist of NIR spectra (Boldrini et al., 2012). Recent contributions have presented comprehensive reviews and discussions of applications of hyperspectral imaging for biomass analysis (Fahlgren et al., 2015; Eylenbosch et al., 2017). This is an active area of research, and the Department of Energy ARPA-E Transportation Energy from Renewable Agriculture (TERRA, https://arpa-e.energy.gov/?q=arpae-programs/terra) is currently supporting work in highthroughput biomass phenotyping in both laboratory and field environments using a variety of spectroscopy approaches, including hyperspectral imaging. Work sponsored by the ARPA-E TERRA will dramatically accelerate the application of hyperspectral imaging for biomass analysis in the coming years.

Several ultra-portable NIR spectrometers have been developed in recent years that have the potential to provide performance comparable to conventional NIR spectrometers in a much smaller form factor and at a much lower cost than traditional NIR spectrometers. Very-low cost, ultraportable, and ubiquitous NIR spectrometers could represent a compelling alternative to traditional NIR rapid analysis approaches for biomass analysis. The composition of biomass materials could be tracked essentially continuously across the value chain from harvest and collection, transport, storage, through conversion to fuels and chemicals. **Table 3** highlights four different instruments (listed alphabetically) that represent unique approaches to ultra-portable NIR spectroscopy.

While this is not meant to be an exhaustive list of all ultra-portable NIR spectrometers, the list does demonstrate the breadth of technical innovation in this area. For example, each of these instruments employs a different active optical element or light processing modality. The microNIR instrument uses a fixed Linear Variable Filter as the dispersive element in the optical path. The NeoSpectra is a Fourier Transform spectrometer, using a miniature Michelson Interferometer based on microelectro-mechanical system (MEMS) fabrication technology. The NIRONE uses a MEMS-based Fabry-Perot Interferometer (FPI) as a tunable optical filter. The NIRVASCAN instrument uses a fixed grating in combination with a digital micromirror device (DMD) consisting of several hundred thousand miniature mirrors acting as a wavelength filter. Each of these instruments has a different spectral range, signal-to-noise ratio, sample presentation geometry, and data collection and processing environment. It is beyond the scope of this work to provide a comprehensive comparison of the performance of each of these instruments; the suitability of a given instrument depends in large part on the potential application.

Some challenges remain for further development of these novel spectrometers, including the demonstration of adequate data collection and processing environments, long-term performance stability in real-world applications, the ability to develop useful and robust calibration models for use on these new platforms, and to perform accurate calibration transfer among spectrometers (Workman and Mark, 2013) so that calibration equations developed on a primary instrument can be used on multiple secondary instruments. Nonetheless, it is clear that there has been substantial developing in ultra-portable NIR spectroscopy in recent years, and this development will likely continue.

As mentioned above, the way NIR spectroscopy is used for rapid analysis (calibration model development and subsequent sample prediction) has evolved substantially over the last several decades, for example, with more robust PLS modeling algorithms and improved outlier detection and measures of prediction uncertainty. However, as NIR spectroscopy for rapid biomass analysis becomes more widely used (in part due to developments in hyperspectral imaging instruments and lowercost spectrometer technology), it will be possible to take fundamentally new approaches to deriving useful information from larger collections of spectral data. While the size of these data sets may never approach those of laboratory analytical techniques such as hyphenated chromatography (e.g., GC-GC-MS, LC-MS), they will likely be large enough and have enough variability to permit machine learning or neural net modeling approaches for classification applications, and real-time updating of classification and prediction modeling using cloud computing resources. Both the development of novel approaches for processing the data and the curation and management of the data itself will represent key technical challenges (and opportunities) in the future. In summary, rapid analysis using near Infrared (NIR) spectroscopy has proven to be a robust, reliable technique for high-throughput biomass characterization when used properly and with care. In the future new opportunities for the technique will develop because of improvements in the two complementary technologies that have made the technique useful in the past: NIR spectroscopy instrumentation and spectral data processing techniques, particularly machine learning approaches.


#### HTP RECALCITRANCE SCREENING

Measuring biomass recalcitrance across large numbers of natural and transgenic plant variants has long held the promise of identifying promising lignocellulosic biofuel feedstocks. The best screen should include both compoitonal analysis of the starting feedstock as well as sugar release after processing, in order to measure conversion efficiency as well as titer and yield (Sykes et al., 2015a). For conversion, either pretreatment, enzyme hydrolysis, or a combined approach may be used. More recently, microbial assays of a Consolidated Bioconversion Process (CBP) nature, in which cellulolytic microbes are used to hydrolyze lignocellulosic feedstocks and measured by product formation, have been employed. Both approaches can be used to screen a range of substrates for differences in recalcitrance by holding the catalyst(s) constant or to evaluate different catalysts (pretreatment conditions, enzyme systems, microbes) on a single defined substrate. In both systems, product detection is critical to evaluating the differences induced by the variables introduced. Many of the methods used to set up and analyze the experiments are the same or similar.

HTP screening methods are being increasingly applied to process development in biotechnology (Long et al., 2014; Back et al., 2016; Yang et al., 2017; Zutz et al., 2017). As new methodology is developed, HTP screening is increasingly employed to collect biological data that historically required extensive time and effort (Scheel and Lutke-Eversloh, 2013; Suzuki et al., 2015) and is widely used today in the development of fermentation process assays (Decker et al., 2003, 2009; Selig et al., 2010, 2011; Studer et al., 2010; Suzuki et al., 2015; Yang et al., 2017). There have been rapid developments of HTP techniques in recent years in micro-scale culturing, online analysis and monitoring, and real-time control, which have enabled increased systems automation (Yang et al., 2017). A key technology in applying HTP to microbiological screening has been the miniaturization of bioreactors, making large experimental cultivation economical and practical (Back et al., 2016; Velez-Suberbie et al., 2017). Both micro- and minibioreactors are critical to biotechnology process development. Recent HTP developments applied to biological research have been applied to developing more effective large-scale operations, greatly decreasing the time and expense compared to development at scale (Lattermann and Buchs, 2015) and it is likely that there is room for continued improvements (Long et al., 2014). As an example, microtiter plates (MTP) are simple, easy to shake, and inexpensive (Bharadwaj et al., 2011; Yang et al., 2017) and have been demonstrated to effectively replace shake flasks (Oguntimein et al., 2018). MTPs have been used to screen specific activities of enzyme variants using various techniques, such as protein quantitation by immunoturbidimetric (ITA) assays, (Yang et al., 2017), direct fluorescence resonance energy transfer for protease activity (Suzuki et al., 2015), cell free protein production (Casteleijn et al., 2013), fungal biosensor assay to detect estrogen activity (Zutz et al., 2017), protein purification and characterization for crystallographic studies (Kim et al., 2011), and enzyme-screening of ionic liquid pretreated lignocellulose (Bharadwaj et al., 2011). Despite these recent examples of MTP-based screening, few details are known regarding actual culture conditions inside the MTP, the technology to measure these details in real time in such numbers and small volumes remains lacking (Long et al., 2014).

Lignocellulosic biomass (LCB) has been the focus of research as a renewable source for second generation bioethanol production but selection and development of these substrates with high bioethanol yield requires the availability of reliable methods for compositional and structural characterization (Elliston et al., 2015). As discussed in Section High Throughput Plant Cell Wall Compositional Analysis, quick HTP analysis of the potential of LCB feedstocks is an important step in the development of second generation bioethanol. HTP screening allows the rapid investigation of a large set of samples at minimum cost. An assay used to determine bioethanol production from large numbers of LCBs must be robust, rapid, easy to perform, and must use modest amounts of the samples (Elliston et al., 2015). This section of the review focuses on the use of high throughput (HTP) pretreatment and enzyme hydrolysis as well as consolidated bioprocessing for the conversion of LCB into bioethanol.

The production of second generation biofuel involves a number of consecutive process operations, each with a combination of multitude steps. These operations can be delineated into pretreatment, hydrolysis, fermentation, and distillation and/or separation. The overall process design could be one of several general approaches, including separate hydrolysis and fermentation (SHF) or simultaneous saccharification and fermentation (SSF) for any given LCB substrate, but the most rapid, effective, and cost effective method to produce bioethanol for any approach requires the optimization of various process parameters. Bearing this in mind, Decker et al. (2009) and Gomez et al. (2010) developed methods for the rapid screening of biomass for the hydrolysis stage (Decker et al., 2009; Gomez et al., 2010), however further research is required to investigate downstream process impacts due to yeast or other micro-organisms. The potential effect of

fermentation inhibitors released during biomass processing on the final alcohol yields is also very critical. This may be process or substrate-specific (Pienkos and Zhang, 2009). In view of the importance of both separate hydrolysis and fermentation (SHF) and simultaneous saccharification and fermentation (SSF) methodologies, SSF methodology on a solid substrate has not been widely investigated at the much smaller HTP scale. The SSF approach introduces a complication as a result of CO<sup>2</sup> production during yeast fermentation which must be vented to reduce pressure in the reaction vessel while at the same

time controlling evaporation, all whilst potentially being stirred vigorously. The current paradigm, however, is focused more toward SHF (Waldron, 2014) and consolidated bioprocessing (CBP) (Oguntimein et al., 2018). This method simplifies the engineering requirement such as decreased capital and operational expenditures and reduces the potential for microbial contamination prior to the addition of yeast.

#### Substrate Preparation

The preparation of solid substrates in the case of second generation biofuels presents its own unique set of problems to HTP screening, primarily the requirement to accurately, repeatedly and rapidly dispense solid, heterogenous sample material by weight. Manual weighing is too labor intensive and time consuming, making it impractical. Recent solutions have included automatic weighing robots (Santoro et al., 2010), production of handbills (dry biomass sheets that can be subdivided to repeatable mass) (Berlin et al., 2006) and biomass slurry pipetting (Chundawat et al., 2008). Automatic weighing and dispensing robots for dry biomass are expensive and require dedicated operation. Grinding of sample is required, with suitable size ranges pre-determined for each biomass type. Typically, the grinding is accomplished off-deck before loading the samples onto the robot, however on deck grinding through ball-milling has been used in several labs (Foster et al., 2010; Santoro et al., 2010). Attention should be paid to avoid overmilling, which reduces structural recalcitrance factors in the material. Electrostatic forces generated during dispensing (by a rotating anti-bridging wire on plastic dispenser walls or steel balls grinding biomass in a plastic tube for example) must be addressed or else the sample will be errantly dispensed to the adjacent wells, inter-well spaces, or the bottom of the balance or robot. Sample heterogeneity is also a major issue, especially in small sample masses (Santoro et al., 2010).

The use of handbills requires additional equipment and expertise, so the use of filter paper as a universal substrate for the measurement of cellulase activity has been adapted by many groups (Ghose, 1987). Elliston et al. (2015) have investigated slurry pipetting techniques using a Tecan Freedom EvoTM liquid handling robot equipped with a multi-channel arm (Tecan Group Ltd, Mannedorf, Switzerland). Liquid transfer was applied in the study for HTP SSF analyses using a 96-well plate format. The major obstacles include evaporation of samples, with the small scales utilized in HTP (typically ≤1 mL) driving the requirement for effective sealing, especially when incubated for days at elevated temperatures. In order to measure evaporative loss, a 1.0 mL matrix storage tube plate dried to a constant weight at 50◦C was used with each well filled with 1.0 mL yeast nitrogen base (YNB) medium. The tubes were sealed with screw caps and incubated at 50◦C (enzyme optimum temperature) over 72 h. Evaporation rates were low and linear over 24, 48, and 72 h time points, with evaporative losses of 0.28, 0.60, and 0.91%, respectively (Elliston et al., 2015).

For pretreated substrates, several methods have been employed. In one of the earliest studies on HTP pretreatment and enzyme hydrolysis, Chundawat et al. (2008) utilized ammonia fiber expansion in a batch pretreatment of corn stover, dispensing the pretreated material into the wells of a 96-well microtiter plate and evaluating the effects of solids loading and particle size on digestibility (Chundawat et al., 2008). While this approach gives the advantage of potentially screening numerous enzyme combinations, the process is limited to one of a few types of biomass. A more versatile approach, pioneered by Studer et al. (2010), Decker et al. (2009), and Selig et al. (2010) was to carry out pretreatment in a HTP reactor plate containing parallel individual reaction chambers. Multiple biomass types can be screened in a massively parallel fashion, though conversion conditions are limited to a single pretreatment and digestion temperature/time/pressure combination for each plate. For reactor plates in a standard 96-well format, the added advantage of automated or semi-automated simultaneous liquid transfers for all wells greatly increases throughput while retaining the option to pipet individual enzyme or acid catalysts by well. Use of acid-resistant metallurgy and appropriate sealing systems allows for high temperature and acid-catalyzed thermochemical pretreatments (Decker et al., 2009; Selig et al., 2010; Studer et al., 2010).

### Effect of Solid Substrate Mass on Simultaneous Saccharification and Fermentation (SSF)

In one of the earliest biomass-related HTP methods, Decker et al. (2003) used Whatman number one filter paper and powdered celluloses dispensed as a slurry to automated the filter paper assay used to quantify cellulase activity as a precursor to SSF (Decker et al., 2003). In 2012 and 2014, Yee et al. developed a reducedbench-scale system for both SSF and Consolidated BioProcessing (CBP, see below) of biomass using bottles, measuring products by HPLC and substrate utilization by mass loss (Yee et al., 2012, 2014). Reducing SSF to HTP-friendly microtiter plate formatting, Elliston et al. (2015), using Whatman number one filter paper (FP) and office copier paper (OCP) (6 mm diameter) as substrates, found that the masses were highly repeatable; 2.36 mg ± 3.1% (w/w) and 2.19 mg ± 2.5% (w/w), respectively, with six sample points. The consistent thickness and density of these substrates, which enabled the high mass repeatability, allowed for rapid allocation of substrate to small matrix tubes. In a comparative experiment between shake flask (200 mL in 500 mL Erlenmeyer flasks), tubes (10 mL in 30 mL screw-capped culture bottles) and MTP format (1 mL in 1 mL screw-cap matrix storage) using yeast and mold media (YM) plus glucose [0.9% (w/v)], OCP [2.5% (w/v)], or FP [2.5% (w/v)], ethanol yield was

similar at each scale, validating the use of the HTP small scale method for screening yeasts by SSF (Elliston et al., 2015).

#### Effect of Microorganisms on Bioethanol Production From LCB

Oguntimein et al. (2018) demonstrated a HTP 96 well microplate assay to evaluate MTP- consolidated bioprocessing as a method to measure biomass conversion potential. Twenty milligrams of pre-milled switchgrass or avicel was dispensed into deep well MTPs (2.2 mL/well volume) 96 well microplates shown in **Figure 2** using a Powdernium <sup>R</sup> powder dispensing system (Symyx, Geneva, Switzerland). After additional of liquid medium, C. thermocellum 1hpt was inoculated into rows A-C while C. thermocellum LL1210 was inoculated into rows E-G). Sterile water as added to rows D and H. Plates were placed at a 45-degree angle on an orbital shaker (Cole Palmer Model 51300) set at 125 rpm in a 60◦C incubator in a Coy anaerobic chamber (5% H2, 10% CO2, and 85% N2, Coy Laboratories Products Inc., Grass Lake, MI).

Both C. thermocellum 1hpt and LL1210 strains metabolized Avicel, generating cellobiose, glucose, lactic acid, formic acid, acetic acid and ethanol in titers and ratios similar to that obtained in bench scale fermentations, demonstrating the applicability of a HTP method using CBP using C. thermocellum. Strain LL1210 generated higher ethanol titers than those of strain 1hpt, which is consistent based on earlier reports from larger-scale experiments (Dumitrache et al., 2016). The absolute titers for ethanol were lower than those produced in pH-controlled bioreactors and switchgrass generated lower ethanol concentrations than avicel (Oguntimein et al., 2018).

# Effect of Lignocellulosic Biomass

Lindedam et al. (2014) compared three HTP pretreatment and enzymatic hydrolysis systems (HTPH-systems) for screening lignocellulosic biomass by enzymatic saccharification to confirm that quantitative differences in substrate can be detected at a small scale. Twenty winter wheat cultivars grown at two sites in Denmark were hydrothermally pretreated and enzymatically digested in three separately engineered HTPH-systems at (1) University of California, Riverside, (2) National Renewable Energy Laboratory (NREL), Colorado, and (3) University of Copenhagen (CPH). All three systems delineated differential sugar release among the cultivars, though average extent of cellulose conversion varied at 57, 64, and 71% for Riverside, NREL and CPH, respectively. Riverside and NREL systems had the highest pair-wise correlation with glucose, while xylose yields correlated best between Riverside and CPH. All three systems agreed on Flair as the cultivar with the highest yield and Dinosor, Glasgow, and Robigus with the lowest. Despite the varied conditions between the three HTPH-systems which resulted in different absolute values, the correlation and rank ordering agreement between them clearly indicates that microscale combined thermochemical and enzymatic conversion can be used to identify recalcitrant phenotypes between varied feedstocks (Lindedam et al., 2014).

Elliston et al. used ethanol production under SSF conditions to assay the conversion of milled wheat straw pretreated under two different conditions different conditions (195◦C for 10 min and 210◦C for 10 min). Analysis of twelve replicates demonstrated the expected increase in ethanol yield for wheat straw pretreated at 210◦C for 10 min (80% of theoretical yield) compared to pretreatment at 195◦C for 10 min (64% of theoretical yield) (Elliston et al., 2015).

Zhang et al. also used HTP pretreatment and co-hydrolysis (HTPH) to rapidly identify promising Miscanthus genotypes, including hybrids of Miscanthus sacchariflorus/M. sinensis as well as M. lutarioriparius, highlighting the commercially promising hybrids. The results also indicated that, at least in Miscanthus, glucan plus xylan content influences both mass and theoretical yields, while lignin and ash contents had no measurable impact (Zhang et al., 2012).

Applying consolidated bioprocessing (CBP) in HTP assay format, Oguntimein et al. demonstrated the fermentation of switchgrass and Avicel by a parent strain of Clostridium thermocellum (1hpt) (Oguntimein et al., 2018). The HTP, MTPbased CBP assay produced ethanol levels similar to bench-scale Avicel and switchgrass fermentations (Dumitrache et al., 2016; Tian et al., 2016). According to the authors, additional studies are needed comparing the effect of biomass concentration on bioethanol production, to correlate the HTP CBP screen with other analytical methods such as quantitative saccharification of residual biomass, and evaluation of well pooling necessary to determine the extent of hydrolysis and fermentation. Screening on other biomass feedstocks such as corn stover and poplar is also needed to validate a broader application of the method. Lastly, evaluating multiple microbial strains will provide a more comparative picture of the impact of microbial factors on the assay specificity (Oguntimein et al., 2018).

This review reports on the factors influencing HTP assays methods for screening of otherwise recalcitrant lignocellulosic substrates for bioethanol production so that they can be performed efficiently and reproducibly in a laboratory setting. Current methods are influenced by the preparation of LCB, the type of LCB, weight of biomass and the type of fermenting organisms. Further studies are required to evaluate and optimize the interactions of these factors in order to have practical uses in the biorefining of biomass substrates for second generation biofuels. Most of the HTP methods have been developed to mimic large-scale operating conditions (Lattermann and Buchs, 2015). This trend will likely continue toward even smaller reactors, potentially even single-cell microfluidic chips (Oguntimein et al., 2018).

#### NMR FOR BIOMASS SUGAR AND FATTY ACID COMPOSITION

Detailed characterization is required for the continued development of biomass feedstocks possessing traits desirable for biofuels and bio-derived chemicals. Quick and precise identification of cell wall chemistry traits and composition due to both gene transformation and natural variation can be accomplished through high-throughput (HTP) characterization using nuclear magnetic resonance (NMR) spectroscopy. Traditionally, lignin and carbohydrate chemistries have been obtained through time and labor-intensive bench-scale HPLC and gravimetric determination (Sluiter et al., 2008, 2010). While carbohydrate characterization methods can incorporate automated sample preparation in 96-well plates, a timeconsuming HPLC analysis step for minor sugars composition is often the bottleneck in throughput (Selig et al., 2011).

Simple <sup>1</sup>H NMR methods provide a wealth of information about liquid samples, such as biomass hydrolysates, including both carbohydrate and hydrolysate by-product compositions. Mixture analysis of NMR spectra is well developed, with many software applications making analysis straightforward and reliable (da Silva Neto et al., 2009; Powers, 2009; Spraul et al., 2009; Da Silva et al., 2013). Traditionally, sugar analysis in biomass hydrolysates has been performed using integration of the anomeric proton region between 4.4 and 5.4 ppm against a reference standard (Kiemle et al., 2003; Mittal et al., 2009), but peak overlap occurs for several sugars and the large water peak at 4.8 ppm often makes integration of these peaks impossible. Shifting of the water peak above the anomeric proton region requires either highly acidic conditions, which cause NMR instrumentation issues (Kelly et al., 2002), or a belowfreezing temperatures of water, rendering them undesirable for high-throughput analysis. Gjersing et al. (2013) reported a method developed for high-throughput screening of biomass hydrolysates generated from two-stage acid hydrolysis using <sup>1</sup>H NMR spectra. HPLC-measured concentration data was used to construct a Partial Least Squares Regression (PLS) model for sugar composition using NMR spectra of aqueous hydrolysates. A model for each monomeric sugar can be used to determine concentrations within the hydrolysate mixture. In this method, 8 biomass feedstocks were used to construct a PLS model with HPLC data and NMR spectra. The fully crossvalidated model was used to predict sugar concentrations for 15 samples, including a feedstock that was not in the original model, and compared to HPLC-measured concentrations. The NMR based PLS model and HPLC-measured concentrations agreed, within error, demonstrating the applicability of the model (Gjersing et al., 2013). Use of a cryoprobe allowed for the NMR experiment time to be further reduced, and preliminary work indicates 4 min per sample—a dramatic increase in throughput compared to traditional HPLC methods. A combination of this NMR approach for analysis of carbohydrates combined with the high-throughput micro-scale hydrolysis preparation discussed previously (Selig et al., 2011) could provide a truly high-throughput screening method for biomass sugar composition.

Increasingly, metabolite profiles are being used to screen biological materials, including biomass populations to look for varying plant responses to stress (Ruan and Teixeira da Silva, 2011; Sun et al., 2016). Recent reports demonstrate that the laborious extraction procedures for metabolites has been improved, making metabolic analysis by NMR more practical (Fumagalli et al., 2009; Martineau et al., 2011; Rivas-Ubach et al., 2013). Additionally, high-resolution magic angle spinning (HR MAS) NMR provides the ability to use whole cell plant material to detect changes in abundant metabolites, which eliminates several preparation steps (Silva et al., 2012; Blondel et al., 2016).

Fatty acids from non-lignocellulosic biomass such as microalgae have become more routinely recognized as a feedstock for biofuel production (Fukuda et al., 2001; Chisti, 2007; Wijffels and Barbosa, 2010). Algal lipid composition varies greatly among species and additional variation is added when culture growth conditions are modified (Scott et al., 2010). Thus, it is imperative that analytical tools be developed for rapid screening of large numbers of samples necessary for comparative studies. Traditionally, protocols for analysis of lipids from microalgae involve time-consuming and labor-intensive extraction followed by chromatography (Bligh and Dyer, 1959; Jones et al., 2012). A fluorescent model was developed that suffered from many technical drawbacks to large-scale screening, including daily calibration of the fluorescent probe and specificity of cell response (Cooksey et al., 1987; Elsey et al., 2007; Chen et al., 2009). However, a simple <sup>1</sup>H NMR screening method was developed that allowed for assessment of major lipid classes from rough microalgae extracts (Nuzzo et al., 2013). The collection of a single NMR spectrum took only a few minutes, without purification of rough extracts, and employed the use of a reference electronic signal as an external standard, known as ERETIC (Akoka et al., 1999). This allowed for the quantification of several major lipid classes important for biodiesel synthesis, including total fatty acids, free fatty acids, triacylglycerols,

unsaturated fatty acids, and saturated fatty acids (Nuzzo et al., 2013).

Overall, there have been dramatic improvements in sample handling over the last two decades that have allowed traditional bench-scale methodologies to become high-throughput. As robotics and sample changers become more common place, methodologies can be upgraded to accommodate the demand for the large data sets involved in bioinformatics.

#### DATA ANALYSIS AND SYSTEMS BIOLOGY MODELING FOR HIGH THROUGHPUT BIOMASS ANALYSIS

There is a need for integrated biological models to capture the higher order complexity in the interactions that occur among cellular components. A full model of all of the higher order interactions of cellular and organismal components is one of the ultimate grand challenges of systems biology (Sweetlove et al., 2017). The ability to build such comprehensive models will usher in a new era in biology. Success in the construction and application of computational algorithms will enable new insights into the molecular mechanisms responsible for complex biological systems and related emergent properties; using technologies not previously available on a scale not feasible before. A full systems biology model of all of the higher order interactions of cellular and organismal components would lead to breakthroughs, which would have profound effects on the field (Sweetlove et al., 2017).

The cost of generating biological data is dropping exponentially, resulting in increased data that has far outstripped the predictive growth in computational power from Moore's Law. This flood of data has opened a new era of systems biology in which there are unprecedented opportunities to gain insights into complex biological systems. The dominant paradigm of high-throughput systems biology is the use of new technologies to generate massive amounts of data that can then be analyzed computationally for new insights and hypothesis generation. Solving such complex combinatorial problems will give us extraordinary levels of understanding of biological systems. Paradoxically, understanding higher order sets of relationships among biological objects leads to a combinatorial explosion in the search space of biological data. These exponentially increasing volumes of data, combined with the desire to model more and more sophisticated sets of relationships within a cell and across an organism (or in some cases even ecosystems), have led to a need for computational resources and sophisticated algorithms that can make use of such datasets. Thus, the bottleneck in biological science is often no longer data generation but rather the computational analysis.

Biological organisms, including plants, microbes, and humans, are derived from complex genetic systems that are composed of functional networks of interacting molecules, macromolecules, and even species (Foster et al., 2017). The subsequent phenotypes are the result of orchestrated, hierarchical, varied collections of expressed genomic variants regulated by and related to biotic and abiotic signals. However, at the individual organism level, the measured effects of these genomic variants can be viewed as the result of historic selective pressure and current environmental as well as epigenetic interactions. Thus, the co-occurrence of genome variants and the resulting complex phenotypes can be viewed in the context of genome-wide associations in several different ways. This phenomenon allows us to use vectors of genome variant-to-trait associations to detect the higher order interactions occurring in an organism across hierarchical phenotypes. A full model of all of the higher order interactions of cellular and organismal components is one of the ultimate grand challenges of systems biology.

We are attempting to do this for the bioenergy feedstock Populus trichocarpa (black cottonwood) and are currently using 10 million genome variants derived from the resequenced genomes of more than a thousand different genotypes and 160,000 phenotypes that have been measured across this population (including transcriptomics, metabolomics, microbiomics, and phenomics data).

#### Networks

Networks are useful tools for modeling and analyzing complex biological systems by representing biological objects as nodes, (e.g., genes, proteins or metabolites) and representing the relationships/interactions/similarities between them as edges (Barabasi and Oltvai, 2004). For example, networks can model co-expression relationships between genes, sequence similarity between genes, physical interactions between proteins and/or correlations between metabolites. Networks allow for biological datasets to be visualized in an intuitive manner and network visualization packages such as Cytoscape provide an interactive environment for network visualization. However, networks are not simply useful as a visualization tool. Networks provide a data structure that serves as a mathematical representation of a complex system, allowing further analysis to be performed on a dataset represented as a network. Datasets represented as networks are also very easily merged with other networks, thus constructing a useful tool for combining information from different data sources to create a combined and holistic environment for data interpretation (Shannon et al., 2003).

#### GWAS Network Construction

Phenotypes are often complex traits, in that they are influenced by the environment and potentially a large number of genes (Solovieff et al., 2013). GWAS attempts to associate the presence of SNPs with these complex traits (Visscher et al., 2012; Solovieff et al., 2013). This involves genotyping a large number of individuals in a population, measuring phenotypes across all of these individuals and statistically determining the association between the presence/absence of the genotyped markers or SNPs and each phenotype (Korte and Farlow, 2013). A general concern when conducting GWAS studies is that individuals within a population that can be genetically related and share causal alleles, which cause the phenotype, and non-causal alleles artifactually connected to the phenotype (Visscher et al., 2012; Korte and Farlow, 2013). These causal and non-causal alleles can be located near each other on a chromosome and could thus be in linkage disequilibrium (i.e., alleles which are correlated across a population and co-inherited. This linkage disequilibrium (LD) between causal and non-causal alleles across related individuals results in non-causal alleles being correlated with a phenotype when they have no actual effect on the phenotype. A common approach to correcting for this phenomenon is to take population structure into account to avoid artificially inflated p-values. Population structure is often estimated from a kinship matrix and incorporated into the model (Flint-Garcia et al., 2003).

Spurious phenotype-to-genotype associations can also result from outlier phenotype values, this is especially evident when using linear models to calculate such associations. We therefore often apply a median absolute deviation (MAD) from the median cutoff in order to determine if a given phenotype measurement is an outlier compared to measurement taken across the population (Leys et al., 2013).

Associations between genome variants (SNPs) and phenotypes are typically made with the use of a linear mixed model as found in EMMAX (Kang et al., 2010). This results in multiple individual tests being performed, thereby introducing a multiple hypotheses bias, i.e., type 1 error. This bias is often mitigated with one of several false discovery rate methods, including the Benjamini–Hochberg method (Benjamini and Hochberg, 1995).

Alternatively, networks can then be created in which the respective SNPs and phenotypes are nodes and an edge denotes a significant GWAS association between them, enabling the subsequent determination of whether or not a phenotype-associated SNPs reside within genes and create a subnetwork of gene-phenotype associations.

#### Layered Networks, LOE Scores, and New Potential Targets

Alternate lines of evidence about the relationships between genes, and between genes and phenotypes can be created using several sets of networks. We recently developed a Lines of Evidence scoring system (LOE scores) in order to quantify the number of lines of evidence connecting genes to phenotypes (**Figure 3**). The GWAS network layers provide functional information at various scales (from molecular to organismal to environmental), which reflect signaling cascades, biosynthetic pathway information, and various regulatory circuits. For example, the co-expression and co-methylation networks provide information from multiple regulatory layers within the cell and the SNP correlation network models putative co-evolution relationships between genes (Climer et al., 2014; Joubert et al., 2018; Weighill et al., 2018).

LOE (Lines of Evidence) scores for each gene can be calculated by starting with functions or topics of interest, revealing the strength of the evidence linking each gene to the function or topic of interest. The LOE breadth score quantifies types of LOE's (number of layers) that connect a gene and topic or function, and the LOE depth score quantifies the total number of functions/topic a gene is associated with. Individual layer LOE scores from each layer (e.g., co-expression or GWAS) indicate

the number of function/topic associations a gene has within that particular layer (Weighill et al., 2018).

This LOE approach provides a new approach for exploring the vast data collections that are occurring in biology today. Any known genes, phenotypes or annotation topics of interest can be provided as input. A rank-order list of new candidate genes that have multiple lines of evidence supporting their involvement in the area of interest can be created from LOE scores. And as such, this approach generates a prioritized list for genetic modification via transformation, genome editing, selective breeding etc. used to validate and/or manipulate a phenotype or set of phenotypes.

# Deeper Discoveries in Systems Biology: The Balance Between Type 1 and Type 2 Error

In a GWAS analysis that is done in isolation there is often large concern for false positives and stringent, riotous FDR thresholds are frequently applied. However, this overcompensation for type-1 error (avoiding false positives) will likely result in large type 2 errors (i.e., false negatives). If one's goal is to create a systems biology model that captures as many biological interactions (e.g., protein-protein interactions, epistatic and pleiotropic interactions, biosynthetic regulators, etc.) as possible, this is a heavy price to pay. We are now using a combination of relaxed FDR thresholds in combination with LOE on the resulting associations in order to strike an improved balance between type 1 and type 2 error, allowing for a more comprehensive models of the entire biological system. As such, our ability to reconstruct the entirety of a complex biological system improves as the number of population-scale endo-, meso- and exo-phenotypes are measured and combined with deep layers of experimental data collected on individual genotypes.

#### FUTURE PROSPECTS FOR HIGH THROUGHPUT BIOMASS AND DATA ANALYSIS

This review has covered recent developments for the high throughput analysis of biomass composition and other properties that have been made possible through the use of robotics and miniaturized equipment, sophisticated computational tools, rapid detection instrumentation and by reduction in sample size, preparation, man-power, and materials needed for analyses. There is still room for improvement in these processes, particularly in relation to increases in preparation throughput and accuracy of the results. Improvements to instrumentation will continue in effort to enhance sensitivity, resolution, dynamic range, and robustness. Additionally, smaller and portable instruments that can be brought to the field would aid in reducing sample collection resources and errors. Processing parameters, scalability and the effects of other variables, associated particularly in deconstruction and conversion analyses, still require further investigations for accuracy and applicability to large-scale conditions. Comprehensive analysis of data using sophisticated computational tools could extend the capabilities of associated analytical methods and instrumentation and provide a better understanding of biological systems as a whole.

# AUTHOR CONTRIBUTIONS

SD coordinated manuscript assembly and editing, provided content for sections on 2-stage acid hydrolysis and HTP recalcitrance screening, RH Provided content on NMR analysis of sugars and fatty acids, AH-W Provided content on py-MBMS, EW Provided content on NIR, DJ, DW, PJ, GT, DK: Provided content on data management, MR, GO Provided content on HTP recalcitrance screening.

# FUNDING

Funding was also provided by The Center for Bioenergy Innovation (CBI). U.S. Department of Energy Bioenergy Research Centers supported by the Office of Biological and Environmental Research in the DOE Office of Science.

# ACKNOWLEDGMENTS

Portions of the manuscript have been authored by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-publicaccess-plan).

This work was authored in part by Alliance for Sustainable Energy, LLC, the manager and operator of the National Renewable Energy Laboratory for the U.S. Department of Energy (DOE) under Contract No. DE-AC36-08GO28308. Funding provided by U.S. Department of Energy Office of Energy Efficiency and Renewable Energy Bioenergy Technologies Office. The views expressed in the article do not necessarily represent the views of the DOE or the U.S. Government. The U.S. Government retains and the publisher, by accepting the article for publication, acknowledges that the U.S. Government retains a nonexclusive, paidup, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for U.S. Government purposes.This research was also supported by the Department of Energy Laboratory Directed Research and Development funding (7758), at the Oak Ridge National Laboratory. Oak Ridge National Laboratory is managed by UT-Battelle, LLC, for the US DOE under contract DE-AC05-00OR22725.

This research described herein was supported by an award of computer time provided by the INCITE program and used resources of the Oak Ridge Leadership Computing Facility Decker et al. HTP Screening

(OLCF) and the Compute and Data Environment for Science (CADES) at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725.

Support for the Poplar GWAS dataset was provided by The BioEnergy Science Center (BESC) and The Center for Bioenergy Innovation (CBI). U.S. Department of Energy Bioenergy Research Centers supported by the Office of Biological and Environmental Research in the DOE Office of Science. The Poplar GWAS Project used resources of the

## REFERENCES


Oak Ridge Leadership Computing Facility and the Compute and Data Environment for Science at Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05- 00OR22725.

The JGI Plant Gene Atlas project conducted by the U.S. Department of Energy Joint Genome Institute was supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. Full Gene Atlas data sets are available at: http://phytozome.jgi.doe.gov.


differences between cultivars of lignocellulosic biomass for ethanol production. Biomass Bioenergy 66, 261–267. doi: 10.1016/j.biombioe.2014.03.006


stress and for the biotechnological improvement of xero-halophytes. Crit. Rev. Biotechnol. 31, 153–169. doi: 10.3109/07388551.2010.505908


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Decker, Harman-Ware, Happs, Wolfrum, Tuskan, Kainer, Oguntimein, Rodriguez, Weighill, Jones and Jacobson. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Bioconversion of Pelletized Big Bluestem, Switchgrass, and Low-Diversity Grass Mixtures Into Sugars and Bioethanol

#### Edited by:

*J. Richard Hess, Idaho National Laboratory (DOE), United States*

#### Reviewed by:

*Nick John Nagle, National Renewable Energy Laboratory (DOE), United States Nalladurai Kaliyan, University of Georgia, United States Kamalakanta Sahoo, Forest Products Laboratory (USDA), United States*

> \*Correspondence: *Bruce S. Dien bruce.dien@ars.usda.gov*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research*

Received: *08 May 2018* Accepted: *16 November 2018* Published: *04 December 2018*

#### Citation:

*Dien BS, Mitchell RB, Bowman MJ, Jin VL, Quarterman J, Schmer MR, Singh V and Slininger PJ (2018) Bioconversion of Pelletized Big Bluestem, Switchgrass, and Low-Diversity Grass Mixtures Into Sugars and Bioethanol. Front. Energy Res. 6:129. doi: 10.3389/fenrg.2018.00129* Bruce S. Dien1,2 \*, Robert B. Mitchell <sup>3</sup> , Michael J. Bowman<sup>1</sup> , Virginia L. Jin<sup>4</sup> , Joshua Quarterman<sup>1</sup> , Marty R. Schmer <sup>4</sup> , Vijay Singh2,5 and Patricia J. Slininger 1,2

*<sup>1</sup> Bioenergy Research, USDA-ARS, National Center for Agricultural Utilization Research, Peoria, IL, United States, <sup>2</sup> Members of the DOE Center for Advanced Bioenergy and Bioproducts Innovation, University of Illinois, Urbana-Champaign, Urbana, IL, United States, <sup>3</sup> Wheat, Sorghum and Forage Research Unit, USDA-ARS, University of Nebraska, Lincoln, NE, United States, <sup>4</sup> Agroecosystem Management Research Unit, USDA-ARS, University of Nebraska—East, Lincoln, NE, United States, <sup>5</sup> Department of Agricultural and Biological Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States*

Three crops of warm-season grasses are being developed for biomass production on northern rain-fed marginal farmland: big bluestem (BBS), switchgrass (SG), and a low diversity mixture of grasses (LDM). In this study, biomass harvested from established fields were compared for pelletization and subsequent conversion to sugars and ethanol. Each biomass was successfully pelletized to similar bulk densities without adding a binder at a commercial feed operation. Pelletizing increased the bulk density by 407% on average and was equally effective on all three biomass samples (528–554 kg/m<sup>3</sup> ). Chemical analysis of the samples indicated that glucan and xylan contents were slightly reduced during pelletizing (by 23 and 16 g/kg, respectively), as well as theoretical ethanol yields, which are based upon total carbohydrate contents. Pellets and milled straws were pre-treated with either liquid hot-water or low-moisture ammonium hydroxide (LMA) and subsequently hydrolyzed with cellulases. Glucose and total sugar yields were similar for non-pellets and pellets using either pre-treatment; carbohydrates present in pellets were more efficiently recovered compared to non-pellets. LMA pretreated samples were separately hydrolyzed and fermented to ethanol using *Scheffersomyces stipitis* yeast. Hydrolysis recovered 69.7–76.8% of the glucose and 66.5–73.3% of the xylose across all samples. Glucose yields were 251–279 g/kg, db and were significantly lower for SG as compared to the other biomass samples. Recovered sugars were fermented to ethanol at 77.7–86.7% of theoretical yield. Final ethanol yields (245.9–275.5 L/Mg, db) were similar for all of the grasses and estimated to equate to production levels for BBS, LDM, and SG of 1,952, 2,586, and 2,636 l of ethanol per ha, respectively.

Keywords: bioethanol, bioenergy crops, pellets, sugars, grasses

# INTRODUCTION

Perennial warm-season (C4) grasses have been proposed as a sustainable resource for producing sugars, chemicals, and fuels using thermo-chemical and biochemical processes. The U.S. has sufficient resources to grow an estimated 171 Tg yr−<sup>1</sup> of biomass from bioenergy grasses by 2022 (Turhollow et al., 2014). Perennial warm-season grasses are favored for biomass production because of their high productivities and low input requirements. In the case of switchgrass (SG, Panicum virgatum L.), it has been calculated that its conversion to fuel ethanol would reduce greenhouse gas emissions by 94% compared to gasoline, and the bioethanol would have a total net energy balance of 540% (Schmer et al., 2008). Perennial, warm season grasses are ideal for inclusion in rainfed systems with variable precipitation because of their high water use efficiency and developed deep rooting systems which also serve to stabilize and improve soil quality (Liebig et al., 2008; Blanco-Canqui et al., 2017). They are also nutrient-use efficient because nitrogen and essential elements are translocated to the soil, roots, and rhizomes during and after senescence (Vogel et al., 2002; Mulkey et al., 2008).

The United States Department of Agriculture has an ongoing long-term project to develop biomass varieties of native prairie grasses for growth on marginal lands in the Great Plains and Upper Midwest (Anderson et al., 2016). Development of suitable cultivars is particularly difficult because they need to perform well on marginal (e.g., low productivity, high erosion potential, low profitability) rain-fed farmland to minimize land use change from growing row crops. Therefore, in addition to high production, other important traits include drought resistance, quick, and dependable establishment, resistance to lodging, lownutrient demands, resistance to pathogens and insects, and high winter survival rates (e.g., over-wintering) (Sarath et al., 2008). Breeding and management research requires a long-term commitment because of the need to study establishment and multiple production years on these traits. U.S.D.A., through its northern bioenergy centers, is developing and evaluating three native prairie-based bioenergy cropping systems: switchgrass, big bluestem (Andropogon gerardii Vitman), and a low-diversity grass mixture (Anderson et al., 2016).

Of these perennial bioenergy crops, SG is perhaps the most highly developed as judged by past field studies (McLaughlin and Kszos, 2005; Moore et al., 2014), available biomass cultivars, traditional, and molecular breeding research (Lipka et al., 2014; Vogel et al., 2014; Biswal et al., 2018), and other available tools (e.g., NIR calibrations) (Vogel et al., 2011; Serapiglia et al., 2017). Big bluestem (BBS) is a bioenergy crop of growing interest because of its high productivity, good genetic diversity, and ecological role as the dominant prairie species (Moore et al., 2014; Zhang et al., 2015). Yet development of BBS as a bioenergy crop has lagged considerably behind that of SG (Zhang et al., 2015). Finally, there is interest in cultivating pastures containing multiple species to emulate natural prairies for possible ecological benefits (Tilman et al., 2006). However, it is a challenge to develop a mixture of grasses that maintains both diversity and high production over multiple years and cuttings. The crops selected here represent a significant advance over previously cultivated grasses. For example, Liberty SwitchgrassTM is the first bioenergy developed variety released by the USDA (Vogel et al., 2014), following a long breeding and selection process for high productivity and good over-wintering survival, and is only presently being evaluated in commercially relevant field plots. Likewise, the BBS crop used for this study is a mixture of two cultivars to maximize yield and stand health. There is also a need for further testing of polyculture plantings of grasses for bioenergy production given the sparsity of past studies using defined grass mixtures in very limited geographical growth regions (Robertson et al., 2017). Here, we describe a relevant mixture of grasses based on established high production grasses amenable to cutting and processing to value-added products. This unique low diversity mixture (LDM) is comprised of BBS, indiangrass (Sorghastrum nutans (L.) Nash), and sideoats grama [Bouteloua curtipendula (Michx.) Torr.].

Each has shown good establishment and persistence in production fields located on marginal rain-fed farmland located in northeastern Nebraska (Blanco-Canqui et al., 2017). Earlier research established that all three crops were equally beneficial for improving soil quality (Blanco-Canqui et al., 2017).

However, conversion of harvested biomass to ethanol requires transporting the biomass to the biorefinery gate, pre-treating it to deconstruct the cell wall structure, extracting the carbohydrates as fermentable sugars using cellulases and other related enzymes, and fermenting the freed sugars to ethanol or other biofuels (for a review: Dien and Bothast, 2009). Therefore, the suitability of a bioenergy crop depends on more than solely crop yield. This study evaluates the performance of these three promising bioenergy crops for pre-processing and bioconversion to ethanol.

A major impediment to using herbaceous bioenergy crops for industrial-scaled bioconversion is solving the logistical supply challenge of getting the biomass to the biorefinery gate (Hansen et al., 2015). Biorefineries are projected to be very large and to require considerable quantities of biomass to operate at cost-effective scales. For example, U.S. Department of Energy (DOE) cellulosic ethanol cost estimates are based upon a plant consuming 2,000 metric ton/day (Humbird et al., 2011). Meanwhile, chopped or baled grass is difficult to store, transport, and process on the front end because it has low-bulk density and poor flow properties. This challenge can be addressed by densifying and reforming the biomass. There are several available methods for densifying and reforming biomass of which the pellet mill and briquette press are the most popular that are compatible with biochemical conversion (Tumuluru et al., 2011). A pellet mill has several advantages in terms of low machine maintenance, efficient, and effective densification, and pellets have better handling properties compared to briquettes (Tumuluru et al., 2011). It is also the most common densification method used for biochemical processes. In this study, pellets were compared to non-pellets (e.g., baled biomass). Furthermore, this study is unique in that an ongoing commercial feed mill was used to manufacture the pellets. Use of a pre-existing commercial plant located nearby can save on capital costs (if also used for processing forages) and as done here allows for processing to be trialed on a multi-ton scale, which is also afforded by the scale of the crop field trial conducted here. A successful outcome from using a commercial feed mill is uncertain because unlike forages, which are harvested green (for example Guretzky et al., 2011), bioenergy crops (as discussed later) are typically harvested at senescence.

While none of the bioenergy cultivars evaluated here, and especially the LDM, have been previously evaluated in this format, the literature supports our choice to include a pelletizing step in this study. Numerous past studies have sought to evaluate the interaction of pelletization and bioconversion processes. Prior studies have included corn stover (Theerarattananoon et al., 2012; Ray et al., 2013; Bals et al., 2014; Hoover et al., 2014), switchgrass (Rijal et al., 2012; Wolfrum et al., 2017), or postharvest created mixes of biomass (Shi et al., 2013; Wolfrum et al., 2017). Two of these studies pelletized corn stover with ammonium fiber expansion (AFEX) pre-treatment (Bals et al., 2014; Hoover et al., 2014). Other studies used a conventional pelletizing process followed by either pre-treatments with diluteacid (Rijal et al., 2012; Ray et al., 2013; Wolfrum et al., 2017), ionic liquids (Shi et al., 2013), or ammonium hydroxide at room temperature (Rijal et al., 2012). Much of the literature concerns the influence of pellets on chemical composition and biochemical conversion. This is understandable given that pelletizing biomass involves heating and high-pressure steps likely to alter the chemical and certainly the physical nature of the biomass (Stelte et al., 2012). Results from compositional analyses are somewhat mixed with studies reporting either no effect by pelletization (Theerarattananoon et al., 2012; Wolfrum et al., 2017) or a slight reduction in glucose and/or xylose contents (Rijal et al., 2012; Ray et al., 2013; Shi et al., 2013; Bals et al., 2014). Results for sugar or ethanol yields are somewhat more promising with studies reporting either the same or slight increases in yields using pellets vs. non-pellets. To our knowledge, this is the first study that we are aware of to process bioenergy crops using a commercialscaled feed mill. It is also the first to report on processing of LDM biomass and a mixture of BBS.

Pre-treatment is the heart of the ethanol production process and is among its most expensive unit operations (Lynd et al., 2017). We have chosen to compare the crops using two pretreatment techniques: liquid hot water (LHW) and low-moisture ammonium hydroxide (LMA). Liquid hot-water, as its name implies, reacts biomass in the presence of hot-water (e.g., 180– 200◦C) (Mosier et al., 2005a). It has been a popular choice in the literature because it combines some of the advantages of an acid catalyzed pre-treatment without the disadvantages of using a mineral acid (Mosier et al., 2005b). It also generally produces lower amounts of inhibitors compared to dilute-acid pre-treatment. LMA is relatively new and involves treating lowmoisture biomass with ammonium hydroxide (Kim et al., 2016). It is of interest because it employs a static reactor and does not require a dewatering step. We included both pre-treatments to evaluate the biomass samples using both acid and base based pre-treatments.

To summarize, in this study established production fields of BBS, LDM, and SG were harvested, field dried, and formed into round bales. Biomass yields for BBS, LDM, and SG were 7.4, 9.4, and 9.6 Mg/ha, respectively. The bales were transported to a nearby alfalfa feed operation and converted into pellets. These pellets were characterized for bulk density and chemical composition. Ground and pelletized biomasses were pre-treated with LHW and LMA. LHW pre-treated samples were evaluated for enzymatic conversion to sugars and LMA pre-treated samples for conversion to sugars and ethanol, the latter using a yeast that is capable of fermenting both glucose and xylose. The goal was to determine which crop(s) is most suited as a feedstock for bioethanol production.

# MATERIALS AND METHODS

# Chemicals and Media

All media and chemicals were research grade and purchased from either Fisher Scientific or Sigma Aldrich Chemicals. The enzymes used were commercial formulations of cellulases (140 mg protein/ml CTEC3, Novozymes Inc.) and hemicellulases (HTEC2, 109 mg protein/ml). Enzyme activities were measured using previously reported methods (Dien et al., 2008). The yeast Scheffersomyces stipitis Y-50871 (ARS Culture Collection, Peoria, IL) was used for fermentations. This strain has been extensively evolved and selected for ethanol fermentation of glucose/xylose sugars prepared from alkaline pre-treated biomass (Slininger et al., 2015).

#### Biomass Samples and Pelletizing

The sample set included BBS, Liberty SG, and a LDM consisting of BBS, indiangrass, and "Butte" sideoats grama. BBS was seeded as an equal blend of "Bonanza" and "Goldmine" and indiangrass as an equal blend of "Scout" and "Warrior," both on a pure live seed basis. The grasses were harvested in November 2013 from established fields after the first killing frost using commercial field-scale equipment. Grass was harvested with a John Deer 4990 forage harvester equipped with a John Deere 990 rotary head into windrows then packaged into net-wrapped round bales, using a John Deere 569 MegaWide round baler with net wrap, immediately after harvest. Bale weights ranged from 1,450 to 1,750 pounds. Plots were located near Mead, NE (Blanco-Canqui et al., 2017).

Approximately 18 Mg of each feedstock was coarsely ground and pelletized by our commercial partner Dehy Alfalfa Mills (Lyons, NE). Large amounts of each biomass were processed to demonstrate they can be processed at a commercial plant. Round bales were transported to the plant where they were sequentially milled using a tub grinder, and hammer mill. Milled biomass was dried in a ring drier, steam conditioned, and pelletized using a ring die pellet mill (**Figure 1**). The pellet mill settings were ¼" (6.35 mm) pellet diameter with a length of 3/8"−1 1/4" (9.525–31.75 mm).

#### Hammer Milling, Energy Usage, and Particle Size Determinations

Samples were tested for fine grinding as is required for some biomass conversion technologies (e.g., fast pyrolysis). Pelletized and raw biomass samples were ground to pass through a 2.0 mm screen using a hammer mill (1100 W, model MHM4, Glen Mills, Clifton, NJ). Energy usage required for milling was determined using an electrical meter. The distribution of particle sizes was

determined by passing the ground biomass through a series of screens with increasingly fine meshes and afterwards weighing the amounts retained on each screen; particles that passed through the smallest screen size (e.g., 38µm) were classified as fines. Particle sizes were determined using a sonic shifter (ATM model LP3, AdvanTech, New Berlin, WI) equipped with U.S. no. 30, 40, 60, 120, 325, and 400 screens (600, 425, 250, 125, 45, and 38µm openings).

were formed without the use of a binder.

#### Physical and Chemical Characterization

Bulk densities were measured by filling a tared 600 ml Pyrex beaker with pellets and dividing the weight of the biomass by the filled volume. The volume of the beaker was determined based upon filling it with an equal volume of water. Each measurement was performed in triplicate.

Mass, length, and diameter distributions of pellets were determined by measuring 100 individual pellets using an analytical balance and calibrated caliper. Mean densities were determined by measuring the individual volume and mass for 18 pellets for each biomass. Volume was calculated by measuring the diameter and height using a calibrated caliper and calculated assuming each pellet was cylindrical in shape.

Moisture contents were measured by monitoring weight loss after samples had been dried at 105◦C for 18 h. Chemical composition of water/ethanol extracts, structural carbohydrates, and lignin analysis was conducted using the two-stage acid digestion method (Sluiter et al., 2008). Soluble sugars (sucrose, glucose, and fructose) and starch were measured as previously reported (Dien et al., 2006). Each sample was analyzed in duplicate.

# Pre-treatments

LHW pre-treatments were conducted as previously described (Serapiglia et al., 2017). Briefly, milled samples were pre-treated at 10% w/w solids in sealed serum bottles. Serum bottles (2 ea.) were placed in stainless steel reactors, which were fitted into an infrared reactor system and heated to 190◦C for 15 min before being water-cooled to 40◦C. Reactors were heated at 2.5◦C/min.

LMA pre-treatments were conducted as follows. Milled biomass (6.0 g dry basis) and concentrated ammonium hydroxide solution (4.9 ml of 28%w/w ammonia content) were placed in a steel mini-reactor and immediately sealed with a screw cap steel lid. The reactor was mixed at 30◦C for 40 min at 6 rpm (clockwise followed by counter-clockwise) using a computer controlled infra-red reactor system (Mathis Labomat model BFA12, Switzerland). Once mixed, the reactor was placed upright in a static heating oven set to 110◦C and incubated for 72 h. The reactor was removed from the oven, allowed to cool, and the contents transferred to a PyrexTM tray, which had been placed in a chemical fume hood. The ammonia was allowed to evaporate for 36–48 h. The biomass was briefly milled (e.g., 30–60 s) in a coffee mill to remove clumps that might have formed during pretreatment. An aliquot was used to determine moisture content prior to enzymatic digestion and subsequently discarded.

# Enzymatic Digestion and Fermentation

Enzymatic sugar release assays were modified from the methods of the National Renewable Energy Laboratory (NREL, Golden, CO, USA) and run at 3% w/w solids in 0.1 M sodium citrate buffer (pH = 4.8). Samples were saccharified using cellulase (CTEC3 26 mg/g glucan) and xylanase (HTEC2 0.96 mg/g forage). Thymol (25 µg/L) was added to prevent spoilage. The reactions were gently mixed (50 rpm) and incubated at 50◦C for 72 h.

Yeast were routinely grown in YPD (yeast extract, peptone and dextrose medium formulated per L: 10 g yeast extract, 20 g peptone, and 20 g dextrose) supplemented with agar when preparing solid media (Bacto agar, 15 g/l). YP (yeast extract and peptone) was sterilized by heating at 121◦C for 15 min in an autoclave (15 min cycle) and supplemented with filter sterilized glucose solution (400 g/l). Hydrolysates were filter sterilized and likewise supplemented with YP without inclusion of the glucose.

Separate hydrolysis and fermentation (SHF) was used for converting the biomass into ethanol. Hydrolysis and fermentation was performed using 50 ml Corning PyrexTM glass wide-mouth screw cap bottles. The bottles were sealed with solid orange caps during hydrolysis and substituted with silicon foam caps for fermentation. For hydrolysis, 3.75 g of LMA pretreated and air-dried biomass was diluted in 30.25 ml of dH2O and adjusted to a pH of 4.8–5.0. Cellulase (CTEC3, 36 mg/ g gucan) and xylanase (HTEC2, 4.2 mg/g forage) were added and biomass digested at 50◦C, while mixing at 150 rpm for 72 h using a shaker/incubator (Innova, New Brunswick Scientific, NJ, USA). Bottles were cooled to 30◦C before adding 2.75 ml of 10x YP and sodium phosphate buffer (pH 6.5, 0.1 M). The medium was inoculated with S. stipitis to a beginning OD<sup>600</sup> = 1.0. Fermentations were sampled at 72 and 96 h for sugars and fermentation products.

The seed culture was prepared by transfer of a colony freshly grown on solid media into 10 ml of YPD. The pre-seed culture was grown over-night and 1 ml transferred to a 50 ml culture, which likewise was grown overnight, and the cells harvested, and concentrated to an OD<sup>600</sup> = 50 with distilled H2O. Pre-cultures were grown at 30◦C with mixing at 150 rpm in Erlenmeyer flasks sealed with porous silicon plugs using a shaker/incubator.

#### Analytical Methods

Sugars and fermentation production were determined using high performance liquid chromatography (HPLC). Samples were injected (20 µL) onto a Spectra System HPLC system (Thermo Electron Corporation, CA, USA) equipped with an auto-injector, isocratic pump, column heater, refractive index detector, and computer running analysis software Chromquest ver 2.5 (Thermo Electron Corporation). Analytes were separated using either an organic acid column (Aminex HPX-87H Column, 300 × 7.8 mm, Bio Rad Laboratories, Inc., Hercules, CA, USA) with 5 mM sulfuric acid as the mobile phase at 65◦C or sugar column (Aminex HPX-87P Column, 300 × 7.8 mm, Bio Rad Laboratories) with water at 80◦C. Flow rates of the mobile phases were 0.6 ml/min for both column separations. Galactose and xylose were determined by values from the sugar column and correlated with those measured using the organic acid column, on which the two sugars co-elute. Glucose and arabinose were taken as the average between the two runs. Acetic acid, ethanol, and glycerol were measured using the organic acid column. Enzyme protein contents were measured using the Bradford protein assay (Bradford, 1976).

Yields and conversion efficiencies were calculated as described (Dien, 2010).

Results were tested for significant differences (p < 0.05) using a 2-way ANOVA on biomass type and form (SigmaPlot 13.0, Systat Software, Inc.). Pairwise multiple comparisons were analyzed using the Holm-Sidak method (SigmaPlot 13.0, Systat Software, Inc.). All experiments included 2 or 3 levels of replication as indicated in the text.

# RESULTS

#### Pelletization

Big bluestem, switchgrass, and a low-diversity prairie grass mixture, which consisted of BBS, indiangrass, and sideoats grama, were separately pelletized at a commercial pellet feed mill using ∼18 metric tons of each biomass. Differences in bulk density are visually contrasted in **Figure 2**. The straws were condensed from a beginning density of 128–133 kg/m<sup>3</sup> to a final density of 528–554 kg/m<sup>3</sup> , which is a 407% increase in bulk density across all biomasses. Pellet mean diameters (6.52– 6.58 mm) were close to the specified diameter (6.35 mm) with relatively low standard deviations (1.41–1.57%). A significant difference was detected between the mean diameter of LDM vs. SG pellets (p < 0.05) (**Table 1**). The distributions of pellet weights and lengths had large standard deviations and were skewed to the right (**Table 1**). For example, relative standard deviations for pellet weights were 36.2–40.0%. LDM pellets weighed more than BBS and SG pellets (p < 0.05), however significance differences were not observed among pellet lengths. Finally, BBS pellets were less dense (0.98 ± 0.08 g/ml, p < 0.05) than either LDM (1.06 ± 0.05) or SG (1.08 ± 0.07 g/ml) pellets.

#### Composition and Milling

Compositions of the grasses were measured before and after pelletization and had mass closures of 91.5–95.8% (**Table 2**). Glucan and xylan contents (337 and 221 g/kg, db, respectively) were within the range of previously reported results for post-frost harvested switchgrass (309–385 g/kg, db and 200–246 g/kg, db), albeit toward the lower end (Dien et al., 2006, 2013; Kim et al., 2011; Rijal et al., 2012; Wolfrum et al., 2017). BBS glucan and xylan contents (340.1 and 199.6 g/kg, db) are in agreement with NREL published results (362 and 212 g/kg, db) (Wiselogel et al., 1996).

Pelletization led to compositional changes for the grasses. Biomass samples are prepared for compositional analysis by extracting soluble material with water and ethanol solutions. It is notable that pelletizing increased water-ethanol extractable material by 3–5% w/w across all three of the biomasses. Pelletizing also decreased glucan and xylan contents by 1–3% w/w. Differences in glucan and xylan contents between nonpellets and pellets were statistically significant (p < 0.05) for each biomass type. However, acid insoluble and soluble lignin contents were unaffected by pelletization. Maximum or theoretical ethanol yields are calculated based upon carbohydrate contents (Dien, 2010). As a result, theoretical ethanol yields were lower for pellets (∼400 g/kg) vs. non-pellets (slightly >400 g/kg) (p < 0.05) (**Figure 3**). Maximum ethanol yields from pelleted biomass were comparable to each other (p > 0.05) and similar to that of corn stover, as calculated from average compositional data published by the U.S. DOE (Humbird et al., 2011). Reduction in carbohydrate contents represents an added cost incurred from pelletizing.

The pellets were ground using a hammer mill to reduce experimental sampling error with conversion. The pellets milled more uniformly than the non-pellets. The ground pellets contained 8.1 and 4.1% fewer 180 micron particles and fines than the non-pellets, respectively (**Table 3**). BBS pellets and nonpellets took much less energy to mill than LDM and SG (p < 0.05) even though all had similar lignin contents. While further testing is warranted, favorable milling characteristics might be an important advantage for bioconversion if grinding is required prior to pre-treatment.

## Pre-treatment and Low-Solids Enzymatic Conversion

The biomass samples were next evaluated for neutral sugar yields using either LHW or LMA pre-treatment followed by digestion with cellulases and hemicellulases. LHW (also termed hydrothermal) pre-treatment consists of simply heating the biomass to a high temperature in water. The natural acidity of water, which is enhanced as the water is heated, acts as a catalyst. Following the LHW pre-treatment, the whole hydrolysate is pH adjusted to pH 5.0 and digested with a commercial cellulose

(p < 0.001). Legend: dark green is ground; green is pelletized, and cross is the ratio of bulk densities (right hand axis).

and hemicellulose preparations to end hydrolyze cell wall carbohydrates to saccharides. Glucose yields (per kg of beginning biomass) were 262–357 g/kg and the yield of total fermentable sugars (e.g., glucose, galactose, and xylose) were 438–530 g/kg (**Figure 4A**). This represented 70–98% of the available glucan and 72–91% of the total fermentable sugars (**Figure 4B**). Glucose and total sugar yields were either not impacted or improved by pelletization (p < 0.05). Total sugar yields were ranked LDM > BBS > SG (p < 0.05; **Figure 4A**).

Ammonium based pre-treatments operate by a different chemistry than LHW. The alkaline pH partially dissolves hemicellulose and cleaves some lignin cross linkages. In the special case of true grasses (Poaceae or Gramineae), alkali is thought to also improve digestibility by saponification of arabinose ferulic acid ester bonds linking lignin and xylan (Vogel, 2008). Following LMA pre-treatment, ammonia was removed by evaporation and the hydrated whole hydrolysate treated with commercial enzymes to extract the monosaccharides. For



<sup>1</sup>*Based upon 100 pellets.*

<sup>2</sup>*BBS, big blue stem; LDM, low diversity mixture; SG, switchgrass.*

<sup>3</sup>*Different letters are significantly different (p* < *0.05) for log transformed mass.*

<sup>4</sup>*Different letters are significantly different (p* < *0.05) for diameter.*

<sup>5</sup>*No significant differences (p* < *0.05) found for length. Failed Shapiro-Wilk normality test.*

LMA pre-treated and saccharified samples, glucose yields (per kg of beginning biomass) were 272–326 g/kg and the yield of total fermentable sugars were 451–510 g/kg (**Figure 5A**). This represented 68–89% of the available glucan and 71–87% of the total fermentable sugars (**Figure 5B**). Glucose and total sugar yields were either not impacted or improved by pelletization (p < 0.05). Total sugar yields for pellets did not vary with species (p > 0.05) and were slightly higher for BBS and SG than LDM (p < 0.05) for non-pellets (**Figure 5A**).

Overall, yields were higher and more consistent across grass samples for the LMA vs. LHW pre-treatment. For either pretreatment, pelletization was either favorable or neutral for glucose and total sugar yields.

## Ethanol Fermentation

Next, ethanol yields were compared following SHF. To prepare biomass for ethanol fermentation, the different biomass feedstocks were pre-treated with LMA and the whole hydrolysate saccharified at 10% w/w solids. LMA was selected over LHW because sugar yields using the former were more consistent across the different biomass feedstocks. The hydrolysate was supplemented with nutrients, and fermented to ethanol using S. stipitis yeast.

Pre-treatment and enzymatic hydrolysis (e.g., prefermentation) recovered 69.7–76.8% and 70.1–77.1% of the available glucose and xylose, respectively (**Table 4**). Glucose yields were slightly but significantly (p < 0.05) lower for SG vs. the other two grasses, as was also observed for LHW pre-treatment. Xylose yields did not vary among species (p > 0.05).

Glucose and xylose yields [g/kg, db] were reduced 6.4 and 9.4, respectively, for pellets vs. non-pellets averaged across all biomasses (p < 0.05). This reduced yield can be directly attributed to the lower sugar contents observed for pellets vs. non-pellets because pellets were more digestible for glucose (p < 0.05, 3.3% greater glucose efficiency for pellets vs. non-pellets) and not statistically different for xylose efficiencies (p > 0.05). Overall, efficiencies at which free sugars were recovered were less than observed for the low-solids hydrolysates, which is to be expected because of increases in concentration of soluble enzyme inhibitors (acetic acid, lignin derived aromatics, etc.) and end-product inhibition.

Ethanol titers were 18.8–20.5 g/L of fermentation culture following the 96 h fermentations and yields were similar between pellets and non-pellets for the entire biomass set (**Table 4**). Glucose was exhausted within the first 48 h (data not shown) and residual xylose was on average 4.7 ± 2.0 g/L. Glucose was fermented prior to xylose (data not shown) as previously reported (Slininger et al., 2015). The ethanol yields were 77.7–86.7% of theoretical based on the beginning concentrations of glucose and xylose. Ethanol titers and yield efficiencies for the pellets and nonpellets were similar; indicating that pelletizing did not diminish the value of the grasses as a fermentation carbon source. Overall, process yields were 194.0–217.4 g of ethanol per kg of biomass (db). Ethanol yields and efficiencies were similar for all three grass crops.

# DISCUSSION

#### Composition and Pre-processing

Across the ground grasses, there was no significant differences for glucan or xylan contents. Switchgrass did contain more acetate (35.7 g/kg vs. 29.7 g/kg), which is important because it can adversely affect fermentation. In addition, lignin content was lower for LDM (169.3 g/kg) compared to the other two grasses. Most important, all three grass crops had similar theoretical ethanol yields on a mass basis and for non-pellets this value exceeded that of corn stover (**Figure 3**).

If the biomass needs to be trucked to the factory gate, densifying should allow for more efficient transport. Trucks operate most effectively when filled with cargo to their weight limit. For a truck with a trailer volume of 70–90 m<sup>3</sup> , this requires a minimum bulk density of 270–320 kg/m<sup>3</sup> (Thoreson et al., 2010). The mean bulk densities for the three pelletized biomass samples (528–554 kg/m<sup>3</sup> ) consistently exceed this limit (**Figure 2**). In contrast, prior to compacting (128–135 kg/m<sup>3</sup> ), the trucks would be operating at <50% efficiency. The denser pellets are also expected to be more convenient to store vs. round bales or chopped grass or at least this was found to be the case in this laboratory.

Bulk densities of grasses for different presentations are reviewed by Sokhansanj et al. (2009). Ground SG (1.5 mm loose fill) has a bulk density of 120 kg/m<sup>3</sup> , like that reported here. Round baled SG is compressed to 140–180 kg/m<sup>3</sup> . Pellets (6.24 mm diameter) are reported as 500–700 kg/m<sup>3</sup> , though a value below 400 kg/m<sup>3</sup> has also been reported (Gilbert et al., 2009). Pellet densities reported here fall within the accepted range. Pellet properties vary based upon biomass feed moisture and particle size as well as process temperature and pressure (Tumuluru et al., 2011). Presumably, properties would also vary with harvest maturity (Tumuluru et al., 2011) and in that regard


*<sup>a</sup>Compositions, except extractable material, are based upon duplicate samples and values are given on a dry weight basis. Beginning sample % dryness (g oven dry biomass/g beginning biomass) was: (BBS) 95.4%, (BBS-Pellets) 93.1%, (LDM) 98.2%, (LDM-Pellets) 92.8%, (SG) 96.4%, and (SG-Pellets) 94.8%. Standard deviations were 0.03–0.59%. b Interactions were not significant (p* < *0.05) for any biomass component. For glucan and xylan, main components were significant (p* < *0.001) and for Klason Lignin only differences*

*among grasses were different (p* = *0.005).*

*<sup>c</sup>Acid soluble lignin.*

SG pellets destined for feed would be harvested at an earlier maturity to improve nutritive value. However, post-frost harvest is favored for bioenergy crops because nutrients are translocated to the soil post-frost (Dien et al., 2006). In forage systems, grasses are often cut multiple times throughout a season because harvest maturity is known to play a significant role in determining feed quality. Likewise, harvest maturity can affect mass-based conversion yields (Dien et al., 2006, 2013). In this study, grasses were harvested after a killing frost. In temperate regions, a single annual harvest is required to maintain healthy highly productive stands of switchgrass (Sanderson et al., 1999; Monti et al., 2008;

carbohydrate composition is based upon results from a large sampling study (Humbird et al., 2011).

Mitchell and Schmer, 2012). Single cut systems also consume less energy than multiple harvest systems (McLaughlin and Kszos, 2005). While biomass yield is maximum at post-anthesis stage, post-frost harvest is recommended with ∼50–60% of shoot nitrogen remobilized to the rhizomes (Sarath et al., 2014). This minimizes soil nutrient removal (Dien et al., 2006; Sarath et al., 2014) and helps to ensure a healthy productive stand, especially when under drought conditions (Mitchell and Schmer, 2012).

Therefore, further studies are warranted to determine the effect of harvest maturity on pellet processing and feed value in the event of dual use for SG. BBS has similar reported bulk


TABLE 3 | Particle reduction using a hammer mill equipped with a 2 mm screen<sup>a</sup> .

*<sup>a</sup>Size distributions are based upon triplicate samples.*

*<sup>b</sup>Amount of particles (%w/w) retained on sized screen.*

*<sup>c</sup>Energy required for grinding was significantly different across species (p* = *0.027).*

densities as SG: 46.6 kg/m<sup>3</sup> for chopped straw and 467–618 kg/m<sup>3</sup> for pellets (Theerarattananoon et al., 2012). An important advantage of manufacturing pellets is the possibility of making a uniform feedstock using multiple readily available sources of biomass (Shi et al., 2013; Wolfrum et al., 2017), for a highly favorable supply logistics outcome (Ray et al., 2017). This study demonstrates that it is likewise possible to manufacture pellets using heterogeneous grasses harvested from a field planted with multiple plant species.

Forming pellets led to small but significant (p < 0.05) decreases in glucans (2.6% on average) and xylans (1.7%) for all the feedstocks. Most, but not all (Theerarattananoon et al., 2012; Wolfrum et al., 2017), previous studies also reported decreased glucan (1–4%) and xylose (2–4%) contents for pellets compared to beginning straws (Rijal et al., 2012; Ray et al., 2013; Shi et al., 2013; Bals et al., 2014). Varied results are not surprising because pelletizing involves a combination of heat and high pressure to compress biomass and form durable pellets. Formation of durable pellets depends upon operating above the glass transition state for lignin and forming cross linkages between carbohydrates, lignin, and other plant cell components (Kaliyan and Morey, 2010; Tumuluru et al., 2011). Decreased glucan and xylan contents did not appear to be correlated with feedstock type. The techniques used here might also overlook changes in lignin composition. Combinations of carbohydrates and lignin extractives form compounds termed pseudo-lignin, which are detected as Klason lignin (Sannigrahi et al., 2011). Likewise, scale did not play a role, though this study is the first to use pellets made by an existing commercial plant. Finally, the reader is cautioned that observed decreases in carbohydrate contents are near the practical detection limits of standard analytical methods (1–3% relative standard deviations) used to determine composition (Templeton et al., 2010), which is compounded in the case of mixtures by the opportunity for additional sampling error.

#### Conversion of Biomass to Sugars and Ethanol

Conversion yields were next measured by pre-treating pellets and non-pellets and measuring glucose and xylose released following enzymatic hydrolysis. Two pre-treatments were selected for this study: liquid hot-water and low moisture ammonium pre-treatments. LHW involves pre-treating biomass at high temperatures solely with water (Mosier et al., 2005a). LHW was selected because it avoids the use of a mineral catalyst (e.g., sulfuric acid), has been reported to minimize the formation of furfural because xylan is not hydrolyzed all the way to xylose (Mosier et al., 2005b), and has been successfully applied in the past to switchgrass (Kim et al., 2011). LMA is similar to low moisture anhydrous ammonium pre-treatment (LMAA) (Yoo et al., 2011) but in which ammonium is added as a concentrated ammonium hydroxide solution. LMA was selected as a representative alkaline pre-treatment because it is simple to implement, and ammonia pre-treated biomass does not require extensive conditioning prior to fermentation (Dien et al., 2013). Biomassis treated as a moist solid and is incubated in a static oven following a brief mixing step. It is envisioned that the ammonia and biomass will be mixed before being incubated in a large static tank in order to save energy and capital costs (Nghiem et al., 2016). Two pre-treatment technologies were selected because they reflect two different chemistries (e.g., acid and base). LHW and LMA pre-treated samples were assayed for sugar yields in low-solids digestions. LMA pre-treated samples were also used for two-stage high solids hydrolysis and ethanol fermentation (SHF).

LHW proved to be an effective pre-treatment. Across all samples, 72.5–81.8% of total neutral carbohydrates were recovered as monosaccharides following treatment with cellulases and hemicellulases. A prior study also using LHW and switchgrass with similar analytical methods reported a higher glucose yield (>80% glucan vs. 70% here) and a comparable xylan yield (>80% xylan vs. 76% here) (Kim et al., 2011). The higher yield in this prior study could have arisen from using different SG varieties, a higher pre-treatment temperature (200◦C vs. 190◦C), and a longer enzymatic hydrolysis time (168 vs. 96 h).

This is the first study that we are aware of to apply LMA/LMAA to SG (Kim et al., 2016), and the glucose and xylose recoveries reported here (85% of glucan and 76% of xylan) are favorable compared to those reported using other ammonia based pre-treatments with SG. One study that used dilute ammonium hydroxide at lower solids (15%) and higher temperatures (170◦C for 20 min) than here, reported glucose

pelleted or non-pelleted samples (*p* < 0.05).

letters represent significant differences based upon pellet vs. non-pellets and different capital letters represent significant differences based upon species for either



*<sup>a</sup>Average* ± *standard deviation of triplicates.*

*<sup>b</sup>Glucose yields were different for crops (p* < *0.001) and form (p* = *0.022). Xyloses yields were different for form (p* = *0.013). Ethanol yields and titers were not significantly different (p* > *0.05) among the samples. Glucose efficiencies were different for species (p* < *0.001) and form (p* < *0.001). Xylose efficiencies were different for species (p* = *0.041). <sup>c</sup>Based upon beginning biomass (e.g., grams of product per kg of biomass).*

*<sup>d</sup>Fermentation Efficiencies are based upon final ethanol titers (adjusted for enzyme blank) and beginning glucose and xylose concentrations.*

and xylose hydrolysis efficiencies of 66.9–90.5% and 60.1–84.2%, respectively (Dien et al., 2013). LMA appears to be more effective than dilute ammonium hydroxide when pre-treating similar maturity SG; the higher values from the other study were for the SG samples harvested at mid-maturity. Another study that compared both AFEX (50% solids, 140–150◦C, 20 min) and soaking in aqueous ammonia (SAA) (11.5% solids ammonium hydroxide solution, 90◦C, 24 h) pre-treatments, observed slightly lower results: ∼80% glucose and ∼65% xylose yields (Kim et al., 2011). SAA has also been tested on pelletized SG (14% solids, 60◦C, 6 h, 0.9 g NH3: 1.0 g biomass) with glucose and xylose yields of up to 95.2 and 77.6%, respectively. Our yields for LMA SG pellets are slightly lower for glucose (88.2%) and slightly higher for xylose (81.1%). Still, LMA is viewed as a technical advance to SAA (Kim et al., 2016). The reduction of water afforded by using LMA (or AFEX) facilitates recovery of ammonia, avoids energy wasted heating excess water during pre-treatment, and reduces the water footprint of the process.

A significant advantage of LMA pre-treatment is that hydrolyzed sugars can be fermented to ethanol without requiring a conditioning step to remove inhibitors, as is often required in the case of dilute-acid pre-treatment. LMA biomasses were converted to ethanol using a two-stage process: biomass was hydrolyzed to sugars and the sugars fermented to ethanol. A two-stage process allowed for optimal temperatures to be applied for enzymatic hydrolysis (50◦C) and fermentation (30◦C). The overall ethanol conversion efficiencies on a beginning biomass basis were 63.7–71.1%, which includes inefficiencies incurred during hydrolysis and fermentation. Overall ethanol efficiencies represent an advance from a prior study by this laboratory using SG pre-treated with dilute ammonium hydroxide and a Saccharomyces cerevisiae engineered for xylose fermentation (41.0–56.4% of theoretical) (Dien et al., 2013).

#### Comparison of Sugar and Ethanol Yields for Pellets and Non-pellets

No reduction in conversion yields were observed for pelleted and non-pelleted samples for the low-solids digestion assays or for ethanol yields. However, a 2 and 5% reduction in glucose and xylose yields (g/kg biomass, db), respectively was significant (p < 0.05) following the high-solids LMA hydrolysis even though hydrolysis efficiencies were mostly higher for pellets vs. nonpellets. Differences in results between the low and high solids hydrolysates can be caused by concentration of soluble inhibitors in the latter. Ethanol yields were the same for pellets and nonpellets, which suggests that sugar hydrolysis continued during the fermentation and that the final extent of hydrolysis was the same for pellets and non-pellets.

This conclusion that pelletization does not reduce bioconversion yields agrees with other related studies. Studies that use dilute-acid or ionic liquid did not observe a significant difference in product yields between non-pellets and pellets (Rijal et al., 2012; Theerarattananoon et al., 2012; Ray et al., 2013; Shi et al., 2013; Wolfrum et al., 2017) with one exception. Corn stover pellets treated with a low severity dilute acid, washed, and converted using SSF gave a higher ethanol yield efficiency (84% of theoretical) vs. the ground straw (68% of theoretical) (Ray et al., 2013). However, in the same study, when a pilot scale dilute-acid reactor was used, no difference was observed under optimal reaction conditions. In general, high severity or more effective pre-treatments (e.g., ionic liquid) would hide subtle differences in biomass properties associated with pelletizing. A study applying ammonium pre-treatment conducted at ambient temperature using switchgrass did not reveal any differences in glucan conversion, but xylan conversion was increased 10% (Rijal et al., 2012). A previously mentioned study that used SAA reported a 76% improvement in glucose yield from pelletizing (Nahar and Pryor, 2014). Processes that used AFEX pre-processed pellets did not report improved hydrolysis sugar yields, perhaps, because AFEX is a highly effective pretreatment on herbaceous biomass in general. However, one of the AFEX studies emphasized that lower adsorption of water by the pellets compared to the straw led to a more efficient hydrolysis at high-solids because biomass did not need to be fed into the reactor over time (Bals et al., 2014). We have also observed noticeably more free water is present following highsolids dilute-acid pre-treatment of SG pellets vs. ground straw (personal observation by Dien). In summary, pelletizing does not lower product yields and might slightly promote conversion efficiency.

The aforementioned logistical advantages have to be balanced against the added cost and energy inputs associated with making pellets. A typical biorefinery is expected to be scaled to process 2,000 Mg/day of biomass. At this scale, baled SG is estimated to cost \$80.64 /Mg at the factory gate and baling and transport will consume equivalent to 8.5% of the higher heating value (HHV) of the biomass. In contrast to baled SG, pellets are estimated to cost less (\$71.76/Mg) and consume less energy (7.8% HHV) (Sokhansanj et al., 2009). The major energy used to make pellets goes to drying the biomass. It has been suggested adapting the process for bioenergy crops to increase drying efficiency can afford considerable energy and cost savings (Lamers et al., 2015). However, densifying the biomass may be necessary to reduce financial risks associated with relying on a regional feedstock. In other words, increased investor risk has an associated cost and eliminating this risk is expected to more than make up for the cost of pelletizing (Hansen et al., 2015).

#### Comparison of Feedstocks

There was no significant difference among the samples based upon ethanol yields following SHF. However, standard deviations tend to be higher for fermentation studies than for the beginning hydrolysis step perhaps because of the added yeast fermentation step. In this case, BBS and LDM had significantly higher glucose yields compared to that of SG. Xylose yields were similar among the grasses. The 3-year average biomass yields for BBS, LDM, and SG were 7.4, 9.4, and 9.6 Mg/ha. Therefore, differences in conversion yields are minor compared to those observed for biomass production. These crop productivities translate to ethanol production levels of 1,952, 2,586, and 2,636 l of ethanol per ha.

# CONCLUSION

Three new field grown bioenergy grass crops were compared for composition, response to commercial scaled pelleting, and processing to sugars and ethanol. LDM had a higher glucan content than either BBS or SG (p ≤ 0.01) but no differences were observed for total fermentable carbohydrates and ethanol yields per ton of biomass. On a land basis, the

### REFERENCES


estimated productivities were 1,952, 2,586, and 2,636 l of ethanol per ha.

The use of pellets compared to straw does not impact sugar and ethanol yields. This was found to be the case with LHW (sugars) and LMA (sugars and ethanol) pre-treatments. We did see a slight reduction in glucan and xylan contents associated with the pellets (p = 0.014), which agrees with most but not all prior reports. The conflicting conclusions might be attributed to (in this case) use of a commercial scale feed plant and likely variation in conditions used to form pellets across studies. This study demonstrates the promise of BBS, LDM, and SG for bioenergy production and notably demonstrates their processing from field through fermentation. Future research will be directed at evaluating additional production years of the LDM and to further improve biomass yields for BBS, so it is comparable to the other two crops used in this study.

## AUTHOR CONTRIBUTIONS

BD and RM performed several experiments and drafted the manuscript. PS aided in fermentations. VS performed the milling experiment. All authors gave critical comments and helped to prepare the final manuscript.

# FUNDING

This work was funded by the Agricultural Research Service (ARS, USDA), DOE Center for Advanced Bioenergy and Bioproducts Innovation (U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research under Award Number DE-SC0018420 and by the Agriculture and Food Research Initiative Competitive Grant no. 2011-68005-30411 from the USDA National Institute of Food and Agriculture. Any opinions, findings, and conclusions or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the views of the U.S. Department of Energy.

#### ACKNOWLEDGMENTS

Patricia J. O'Bryan and Victoria Nguyen for technical help, Debra Palmquist for statistical help, Novozymes Inc. for supplying the enzymes, Ben Fann, Bill Bickmeier, Jordan Leach, and Suanne Kallis for field work and transporting bales, and Dehy Alfalfa Mills for processing the biomass to pellets.


**Disclaimer:** The mention of trade names or commercial products in this article is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture. USDA is an equal opportunity provider and employer.

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Dien, Mitchell, Bowman, Jin, Quarterman, Schmer, Singh and Slininger. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Ensiled Wet Storage Accelerates Pretreatment for Bioconversion of Corn Stover

#### Dzidzor Essien\* and Tom L. Richard

*Department of Agricultural and Biological Engineering, Pennsylvania State University, University Park, PA, United States*

Organic acids produced during ensiled wet storage are beneficial during the storage process, both for biomass preservation, and to aid in mild *in-situ* pretreatment. However, there is concern these acids could later have negative impacts on downstream processes, especially microbial fermentation. Organic acids can inhibit microbial metabolism or growth, which in turn could affect biofuel productivity or yield. This study investigated the interaction of organic acids produced during ensiled storage with subsequent pretreatment of the resulting corn stover silage, as well as the potential for interference with downstream ethanol fermentation. Interaction with pretreatment was observed by measuring xylan and glucan removal and the formation of inhibitors. The results indicated that organic acids generally do not impede downstream processes and in fact can be beneficial. The levels of organic acids produced during 220 days of storage jar tests at 23◦C or 37◦C, and their transformation during pretreatment, remained below inhibitory levels. Concentrations of individual acids did not exceed 6 g per liter of the pretreated volume, and < 5% on a dry matter basis. Whereas, unensiled corn stover required 15 min of 190◦C pretreatment to optimize sugar release, ensiled corn stover could be treated equally effectively at a lower pretreatment duration of 10 min. Furthermore, the different organic acid profiles that accumulate at various storage moisture levels (35–65%) do not differ significantly in their impact on downstream ethanol fermentation. These results indicate biorefineries using ensiled corn stover feedstock at 35–65% moisture levels can expect as good or better biofuel yields as with unensiled stover, while reducing pretreatment costs.

#### Edited by:

*J. Richard Hess, Idaho National Laboratory (DOE), United States*

#### Reviewed by:

*Damian Joseph Allen, Purdue University, United States Raymond Huhnke, Oklahoma State University, United States*

> \*Correspondence: *Dzidzor Essien idd103@psu.edu*

#### Specialty section:

*This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Bioengineering and Biotechnology*

Received: *12 May 2018* Accepted: *27 November 2018* Published: *14 December 2018*

#### Citation:

*Essien D and Richard TL (2018) Ensiled Wet Storage Accelerates Pretreatment for Bioconversion of Corn Stover. Front. Bioeng. Biotechnol. 6:195. doi: 10.3389/fbioe.2018.00195* Keywords: acetic acid, biofuel, biomass, fermentation, inhibitors, pretreatment, wet storage, ensilage

# INTRODUCTION

Wet storage is the storage of biomass materials under anaerobic conditions that inhibit microbial biodegradation. This is most frequently accomplished by creating an oxygen barrier (a silo, tarp, etc.) and storing the biomass at moisture levels that permit acidogenic anaerobic microorganisms to grow and produce sufficient quantities of organic acids to reduce pH to levels below pH 5, where very little degradation occurs. This microbially enhanced wet storage process, also known as ensilage, has long been used for storage of herbaceous plants for livestock feeds, allowing long term biomass preservation. There are other wet storage systems that do not rely on in-situ microbial organic acid production, instead adding external acids, or alkali compounds to adjust pH or use other biocide strategies to reduce degradation. The primary alternative to wet storage is dry storage, which requires keeping moisture levels low enough—usually below 20%—to slow down and inhibit active microbial activity. Traditional dry storage of biomass feedstocks in bales and other formats is low cost and can be effective if the materials are kept dry but carries the risk of spontaneous or accidental fire outbreaks, narrows harvest windows (especially in humid climates), and can result in extensive contamination of the biomass with soil from field drying operations. If weather conditions are not ideal there can be substantial biomass losses during field drying, and the costs of collection increase. Wet storage systems can reduce these concerns and may also serve as an avenue for in situ pretreatment of the biomass to enhance downstream biofuel fermentation processes (Linden et al., 1987; Richard et al., 2001). With a mechanism similar to dilute acid pretreatment, organic acids produced during ensiled wet storage could serve as long-term, mildly acidic, low temperature pretreatment. Pretreatment has remained the most expensive step in the processing of cellulosic feedstock to biofuels and accounts for at least 20% and usually about a third of the total processing cost in most technoeconomic analyses (Wyman, 1999; Yang and Wyman, 2008; Brown and Brown, 2014; Eggeman and Elander, 2015). As a major cost component of biofuel production, any reduction in pretreatment requirements is likely to have commercial value. A feedstock delivery model by Darku (2013) showed that at storage moisture levels <40% the benefits of wet storage can result in feedstock delivery costs that are lower than dry storage, even considering the costs of transporting the water in the wet biomass.

Previous studies on wet storage have had inconsistent outcomes. Although most reported a favorable impact on downstream processing with reference to the controls (Thomsen et al., 2008; Wendt et al., 2018), some results in some studies show no impact depending on the feedstock or treatment (Linden et al., 1987; Chen et al., 2007; Zheng et al., 2012). However, in most prior research studies the biomass samples were microbially, enzymatically, or chemically treated to enhance storage, and few have looked at the natural ensilage process. In one study without additives, Thomsen et al. (2008) investigated ensiled wet storage using whole-crop maize silage for ethanol production. Although their results failed to demonstrate a pretreatment effect from ensilage, they did show that subsequent pretreatment sugar yields as well as ethanol yield were remarkably improved as a result of the ensilage process. However, the widespread applicability of this result could be confounded by the high starch content of the whole crop (grain and stover) maize feedstock. The differences among wet storage outcomes are dependent, among other factors, on feedstock type. Very few wet storage studies have investigated corn stover, which is the most abundant agricultural residue in the US. Most importantly, none of these previous studies explicitly analyzed the impact of the organic acids produced during ensiled storage, or any modifications of these storage acids during pretreatment and fermentation. In a number of these studies, the feedstocks were washed before subsequent processing, perhaps to prevent interference of the acids with the downstream process. The cost of such washing and the associated wastewater treatment would be hard to justify at a large commercial scale.

Although wet storage of biomass has potential benefits for downstream processing to biofuels, it is also known that most storage acids, including lactic acid, and acetic acid, can under some circumstances inhibit microbial activities including both metabolism and growth (Lund and Eklund, 2000; Deublein and Steinhauser, 2008) and hence negatively affect biofuel fermentations. The impact of such inhibition is dependent on the specific inhibitory compound, its concentration, and also the fermentation organism and conditions used for biofuel production. Although no prior investigations were found on the impacts of silage organic acids on biofuel production, their function in ensilage is to inhibit microbial degradation, and there are ample examples in the food industry of organic acids preserving food through organic acid inhibition or other antimicrobial effects (Lund and Eklund, 2000). Natural acid fermentations are used to preserve sauerkraut, pickles, yogurt, and silage, but unlike foods which will be digested in a mammalian gut, when silage is used as a biofuel feedstock this acidic condition could serve as a potential impediment to downstream fermentations to ethanol and other biofuels and biochemicals. There are a number of reports on negative effects of organic acids with specific reference to ethanol-producing microbes (Palmqvist et al., 1996; Koegel et al., 1997; Zaldivar and Ingram, 1999; Palmqvist and Hahn-Hägerdal, 2000; Klinke et al., 2004; Knauf and Kraus, 2006). Several of these studies focus on the yeast Saccharomyces cerevisiae, which is the most common microbe used in ethanol fermentation. These studies showed the inhibitory effect of organic acids on ethanol-producing microbes is dependent on the fermentation conditions, especially initial pH, extracellular-intracellular pH gradient, temperature, the presence of other chemicals, and the type and amount of organic acid present in both dissociated and undissociated forms. Importantly, Taherzadeh et al. (1997), Thomas et al. (2002), and Torija et al. (2003) observed that the effect of organic acids at low levels can sometimes be positive, stimulating growth of fermentative microbes, and ethanol production, and may be necessary for fermentation to proceed. For instance, Taherzadeh et al. (1997) observed that acetic acid could stimulate ethanol production during glucose fermentation at concentrations lower than 10 g L−<sup>1</sup> , or 5 g L−<sup>1</sup> of the undissociated form at pH 4.5. Torija et al. (2003) also observed that organic acids commonly present in grapes were responsible for the completion of the fermentation process as well as enhanced ethanol yields; none of the controls (i.e., without organic acid of any sort) were able to ferment all the sugars within 21 days.

Other studies have shown that the stimulating and inhibitory effects of organic acids are not restricted to ethanol fermentation. Organic acids can also stimulate butanol fermentation at low concentrations but can be inhibitory above certain threshold concentrations (Cho et al., 2009; Wang et al., 2011; Zhou et al., 2018). Currently, ethanol, and butanol are the only liquid biofuels produced biochemically at a commercial scale by living microorganisms. Enzyme-catalyzed biodiesel production is considered a biochemical process but does not use living organisms, nor do other green drop-in fuels that are produced through hydro/thermo-chemical processes. Because Saccharomyces cerevisiae is by far the most common biofuel fermenter in commercial use, it is the focus of the current study.

The main aim of this research was to investigate the individual and cumulative positive and/or negative effects of the organic acids produced during wet storage of corn stover. Specifically, we investigate how these acids interact with the hot water pretreatment process, the potential for reduced severity pretreatment, as well as their effect on ethanol yields. Pretreatment severity is a function of temperature and time (Chum et al., 1990) and reducing either of these would have a corresponding reduction in both capital and operating costs. Any observed potential for reduced severity pretreatment after wet storage would be an indirect measure of the upstream pretreatment capability of ensilage, where the organic acids produced during storage interact with structural bonds in feedstock throughout the storage process. During pretreatment the potential for reduced severity could also result from the interaction of these organic acids, acting as catalyst, with structural bonds during the pretreatment process. Acetic acid and other organic acids generated during hydrothermal pretreatment are recognized as catalyzing agents in enhancing water ionization and cleavage of acetyl group/hydrolysis of hemicellulose (Mosier et al., 2005a; Mohammad, 2008; Zheng et al., 2009). Organic acids, when compared to dilute inorganic acids (as used in "dilute acid pretreatment"), can minimize degradation of hydrolyzed sugars to inhibitors that can impact on subsequent ethanol yield. This has motivated investigations into the use of organic acids (e.g., acetic, lactic, formic, and maleic) as catalysts in hydrothermal pretreatment (Kootstra et al., 2009; Xu et al., 2009; Marzialetti et al., 2011). Organic acid interactions with poststorage pretreatment could therefore be positive.

However, as noted earlier in this introduction, these acids can interfere with the fermentation process. The downstream effect of these acids was determined by examining the inhibitory nature of organic acids on ethanol fermentation by Saccharomyces cerevisiae. The use of unensiled, washed, and unwashed silage with and without liquid hot water (LHW) pretreatment extracts provides evidence of the transformation dynamics of these organic acids during common biofuel unit operations, as well as the direct effect of the acids on ethanol fermentation.

# MATERIALS AND METHODS

#### Stover Description and Storage

Corn stover, Pioneer brand 34A20, was obtained from the U.S. Department of Energy's Idaho National Lab. The stover was harvested from the Boyd plot near Boone, IA and field-dried, raked, baled, transported to Idaho, and stored indoors with a tarp cover to prevent dust accumulation. Particle size was reduced to 1" minus (less or equal to 25.4 mm) before storage.

The corn stover had an initial air-dried moisture content of about 7% and was adjusted to six different moisture levels (25, 35, 45, 55, 65, and 75% wet basis) to initiate these wet storage experiments. The use of dry feedstock was to achieve better control and explore well defined feedstock moisture ranges in a controlled experimental environment. Tanjore et al. (2012) showed that the major change resulting from oven drying feedstock is the loss of water-soluble carbohydrates (WSC). In that study the higher WSC available in fresh silage was beneficial in generating a lower storage pH than for the rewetted dry stover, but the rewetted dry stover still achieved an acceptable pH range for effective ensiled storage. Alternatives to drying and rewetting, such as harvesting as drying occurs in the field over several weeks during plant senescence, would create its own set of issues. A difference in stover harvest date of even 2 weeks has been shown to have great influence on silage response and composition (Russell, 1986). In the present study moisture adjustment was accomplished by spraying with an appropriate amount of water, covering with plastic, and leaving the samples overnight for the moisture to be thoroughly absorbed into the fibers. The moisture adjustment was within ±2 percentage units of the target moisture level. For each adjusted moisture level, there are corresponding samples that were not ensiled and used as control (Day 0).

Corn stover was stored at a dry bulk density of about 160 Kg/m<sup>3</sup> in 470 ml glass jars that were tightly sealed to create anaerobic conditions. This density provides sufficient compaction to facilitate the silage process and is also comparable to corn stover dry bale densities and the average density used in conventional dry storage systems for hay. Storage duration was 220 days at two temperatures: ambient, which was ∼23 ± 1 ◦C, and 37◦C. The higher 37◦C temperature, which is observed in warm climates, accelerates fermentation. This allows for shorter experiments and also stresses some of the microorganisms. Higher temperatures, non-optimal moisture or substrate conditions can also encourage secondary fermentations from organisms such as clostridia which can reduce silage quality (Weinberg et al., 2001). Including the 37◦C treatment thus accomplishes several goals: to subject the silage process to this temperature stress, to create wide variations in silage outcomes, and to then observe how that wide variations in silage outcomes impact downstream processing.

Experiments were performed in triplicate. After storage, samples were dried in a HotPack convection oven at 55◦C, ground using a 2 mm screen on a Wiley Mill (Model 4, Thomas Scientific, Swedesboro, NJ) and stored at room temperature in sterile airtight Whirl-Pak bags (Nasco, Fort Atkinson, Wisc.) Prior to pretreatment, replicates from each storage condition were thoroughly mixed together to reduce variability among replicates before resampling. Composition of the feedstock before and after storage was measured in accordance with the NREL standard protocols with the exception that the feedstock drying temperature was 55◦C (Hames et al., 2008; Sluiter et al., 2008a,b,c). At this drying temperature volatilization of acids and alcohols was expected to be small, other than possibly methanol, or ethanol. However, these alcohols were not expected to be present in the ensiled biomass at levels that might be inhibitory to Saccharomyces cerevisiae (Driehuis and van Wikselaar, 2000).

## Organic Acid Measurements and Pretreatment

Collection of soluble extracts and measurements of pH and organic acids of feedstock were performed before and after storage. Samples were thoroughly mixed before sub-sampling, and deionized water was added at a ratio of 1:10, i.e., 5 g of sample to 50 ml of water. The mixtures were shaken for 30 min at 200 rpm using a Barnstead SHKA 2000 open air platform shaker (Barnstead International, Dubuque, IA) after which the extracts were filtered through Whatman No.1 paper. The pH of storage extracts was determined using a pH meter (SevenEasy S20, Mettler-Toledo International Inc, Columbus, OH). The collected extracts were filtered again using 0.2µm PTFE filters, diluted 20 fold and analyzed using Ion Exclusion Chromatography System (Dionex ICS 3000, Thermo Fisher Scientific Inc., Sunnyvale, CA) for types and amount of organic acids. Separation was performed at 30◦C using IonPac ICE-AS1 guard (4 × 50 mm) and analytical (4 × 250 mm) columns with 100 mM methanesulfonic acid eluent at a flow rate of 0.16 mL/min. Organic acids were detected with a photodiode array detector (Dionex UVD 340U, Thermo Fisher Scientific Inc., Sunnyvale, CA) at a wavelength of 210 nm. Thirteen different potential acids (lactic, acetic, butyric, pyruvate, isobutyric, valeric, isovaleric, propionic, tartaric, malic, formic, citric, succinic) were used as standards.

The impact of organic acids on liquid hot water (LHW) pretreatment requirements was investigated using washed and unwashed samples of dry ground ensiled (Day 220) and unensiled (Day 0) corn stover. Ensiled samples are the moisture adjusted stover stored under anaerobic conditions for 220 days, while the unensiled samples are the corresponding Day zero samples. Washed samples were washed with deionized water using an Accelerated Solvent Extraction (ASE) system (ASE 350, Thermo Fisher Scientific Inc., Dionex ASE 350, Sunnyvale, CA) set at 40◦C with three static cycles of 10 min each, 100% flush, and a purge time of 200 s for 66 ml cells. The purpose of washing was to remove all organic acids produced during storage in order to prevent any involvement or interaction with the pretreatment procedure. In this way, the washed samples served as the control against which unwashed samples were compared to assess the impact of organic acids on the pretreatment process. Washed samples also provided controls to understand whether any change in pretreatment outcome is as a result of acid interaction during storage or acid interaction during pretreatment, by comparison with unwashed samples of unensiled and ensiled stover, respectively. Only 37◦C samples were washed for comparison.

Liquid hot water (LHW) pretreatment of samples was also accomplished using the ASE 350 equipment, with each sample replicated four times. LHW pretreatment is a well-established and effective strategy that involves heating water-saturated or moist feedstock at high temperatures (160–220◦C) under high pressure to maintain the liquid state for a few minutes, without any chemical additives. Optimum conditions reported by Mosier et al. (2005b) for controlled pH LHW pretreatment were 190◦C for 15 min, and these conditions were used as the benchmark for reduced severity comparisons. This standard pretreatment condition with the ASE was defined as 190◦C, 1 static cycle of 15 min, 0% flush volume and a purge time of 120 s for 10 ml cells, using deionized water as the solvent. Each 10 ml ASE cell was filled with 1.5 dry gram of sample. Solids loading was 14– 20% and 13–15% for unwashed and washed samples, respectively. The solids loading is the percentage of dry solids to total liquids after pretreatment. The variability in solids loading is subject to the amount of water added by the ASE 350 during the filling and heating stage. At the end of the retention time, the liquid is purged out along with some other soluble and insoluble components and is described as pretreatment extract. The pH and organic acid composition of the pretreatment extract were determined using the same equipment and methodology used in storage extract described above. The potential for reduced severity pretreatment was investigated by comparing shorter retention times (5 and 10 min) with the standard 15 min, all at 190◦C.

Pretreatment extracts, 500 µl each, were diluted 30-fold and filtered through 0.2µm PTFE syringe filters prior to analyzing for organic acids and inhibitors caused by sugar degradation (5 -Hydroxymethyl furfural (HMF) and furfural), again using the Dionex ICS 3000 for ion exclusion chromatography. Separation and detection of organic acids followed the same method described above for the before and after storage samples. Inhibitors were also detected with the same photodiode array detector but at wavelengths of 270 nm.

Xylan and glucan removal during pretreatment was also determined. Xylan removal, a proxy for hemicellulose hydrolysis, was used as a comparative indicator of the relative effectiveness of the different pretreatment conditions. To measure the removal of these sugar polymers, the pH of the undiluted extracts was first measured. The pretreatment extracts had pH levels >3.5 but <5. For this pH range, the hydrogen ion concentrations had significant digits at the 4th or 5th decimal place. As a result, volume of acid required for the monosaccharide assay was practically the same. Based on the pH range, 52.3 µl of 72% w/w sulfuric acid (Sigma–Aldrich, St. Louis, MO) was added to 1.5 ml of each extract to obtain a final concentration of 4% sulfuric acid in 10-ml autoclave safe bottles. Bottles were tightly covered using rubber stoppers with crimped aluminum seals and placed in autoclave, together with sugar recovery standards, at 121◦C liquid setting for 1 h. The acid-hydrolyzed extracts were filtered through 0.2µm PTFE filters and diluted 400-fold. Monosaccharide composition was determined using Dionex ICS 3000 ion exclusion chromatography. Separation was by high pH anion exchange at 30◦C using CarboPac PA20 guard (3 × 30 mm) and analytical (3 × 150 mm) columns with 2 mM sodium hydroxide (NaOH) eluent at a flow rate of 0.5 ml/min. Detection of the monosaccharides was by pulsed amperometric [electrochemical] detection at gold working electrodes, using a quadruple waveform. Xylan and glucan removal were calculated from xylose and glucose concentrations, using conversion factors of 0.88 and 0.90, respectively. Equation 1 was used in calculating xylan removal. For glucan removal, the various xylan parameters were replaced by the relevant glucan parameters.

$$\% \text{ Xylan removel} = \frac{\text{x}\_{\text{p}} \times 0.88}{\frac{(X\_{\text{DM}} - \text{x}\_{\text{i}} \times 0.88)}{100} \times DM} \times 100 \tag{1}$$

Where

x<sup>p</sup> = Mass of xylose in pretreatment extract (g) x<sup>s</sup> = Xylose degraded during storage (% dry matter)

XDM = Original xylan content of corn stover before storage (% dry matter)

DM = Dry Matter (Here as dry mass of corn stover that was pretreated) (g).

# Simultaneous Fermentation and Saccharification

After pretreatment, the solids content of each pretreatment cell was directly transferred to a 50-ml centrifuge tube for fermentation. Pretreatment extracts were collected separately during the extraction process. For each storage condition investigated, two replicates were fermented with pretreatment extract, and two without extract. The pretreatment extract contains most of the inhibitory compounds, which are soluble. Thus, there were two steps in the overall process when inhibitors could be separated, first when washing samples after wet storage but before pretreatment, and second when extracting liquids after pretreatment. These two separations (or their absence, for the "unwashed" storage samples and "with extract" fermentation treatments) make it possible to determine the impact of organic acids and other inhibitors formed during storage and/or pretreatment separately with respect their contribution to fermentation inhibition. Although the wet storage organic acid profile may transform during pretreatment, the new postpretreatment organic acid profile is assumed to be influenced by, or a product of, the acids produced during storage.

Simultaneous fermentation and saccharification (SSF) was carried out in tightly sealed 50-ml centrifuge tubes. Samples fermented with pretreatment extract had a solids loading of 8.4% ± 0.1%, while washed and unwashed samples fermented without extract had a solids loading of 5.2% ± 0.1%. The solids loading was calculated as the ratio of dry mass of feedstock used in fermentation to the mass of total fermentation liquids. For samples without extract, some solids were lost in the pretreatment extract resulting in the lower solids loading. The fermentation broth contained the following components prepared in a cocktail before addition: Penicillin-Streptomycin at a final concentration of 30µg/mL (0.1% v/v) to prevent bacterial growth; citric acid buffer (pH 4.5) at 0.05M to maintain a pH of 4.8, which is in the optimum range for enzyme activity; Yeast peptone (YP) as a microbial nutrient at 1% broth volume; commercial cellulase (Spezyme CP, Genencor, Rochester, NY) at 15 filter paper units (FPU)/g glucan complemented with a commercial β-glucosidase (Novozyme 188, Novozymes A/S, Bagsvaerd, Denmark) at 60 cellobiase units (CBU)/g glucan. The microorganism used for fermentation, Saccharomyces cerevisiae NRRL Y-2034, was obtained from the USDA ARS culture [NRRL] collection. Saccharomyces cerevisiae Y-2034 is a wild type 6-carbon sugar fermenting yeast. The yeast was grown in YPD media (10 g/L yeast extract, 20 g/L peptone, and 50 g/L dextrose) for about 24 h after which the cells were centrifuged at 4,200 rpm for 5 min. The supernatant was discarded, and the cells were washed in 1 × PBS (Phosphate Buffer Solution: 138 mM sodium chloride, 2.7 mM potassium chloride, 12 mM sodium and potassium phosphates, pH 7.4). After washing, cells were resuspended in PBS and used as fermentation inoculant. Each tube was inoculated with appropriate volume of inoculant to obtain an initial OD<sup>600</sup> of 0.5. Fermentation tubes were vortexed for ∼5 s to mix contents before incubation for 72 h. The fermentation temperature and agitation, 35◦C, and 110 rpm, were achieved using a lateral motion hot water bath. However, the vertical placement of tubes in the bath did not provide the complete mixing intended by the agitation. Tubes were therefore removed twice (every 24 h) within the fermenting period and inverted twice to mix contents. Control samples included enzyme-yeast blanks and Avicel (α-cellulose). At the end of the fermentation period, samples were centrifuged, and the supernatant collected in micro-centrifuge tubes. The supernatant from each fermentation broth was diluted 9-fold and analyzed for ethanol using the YSI 2700 SELECTTM biochemical analyzer (YSI Inc., Yellow Springs, OH) with 2% precision.

# Data Analysis

Results were analyzed using statistical tools such as analysis of variance (ANOVA), Tukey's multiple comparison test, and regression analysis. All statistical tests were conducted using Minitab 14 (Minitab Inc., State College, PA) at a significance level, α, of 0.05. Results are reported in most cases as means along with the standard deviation of mean (±) as a measure of variability.

# RESULTS AND DISCUSSION

## Pretreatment pH

After pretreatment the pH of the biomass feedstock generally decreased relative to the pH before pretreatment. This result was expected, and can be attributed to deacetylation of xylan, which is the main component of herbaceous hemicellulose [68–72% in this study], at high temperatures leading to the formation of acetic acid (Zhou et al., 2010; Johnson et al., 2017). This acid in turn interacts with the pretreatment process by serving as hydrolytic catalyst, providing free protons. Since pH is an indication of hydrogen ion concentrations, the change in pH can thus indirectly indicate LHW pretreatment activity. Compared to unensiled stover, smaller differences were observed between pH after wet storage and subsequent pretreatment pH of unwashed ensiled feedstock. The average difference for unensiled feedstock was 2.2 pH units while feedstock ensiled at 23◦C and 37◦C had average differences of 0.08 and 0.18 pH units, respectively. Each one-unit difference in the pH corresponds to a 10-fold change in acidity or hydrogen ion concentration. Two factors—hemicellulose degradation during storage and the buffering capacity of organic acids—are likely responsible for the smaller differences in ensiled feedstock. It was observed that on average, 10% of hemicelluloses were degraded after 220 days of storage, mainly through xylan and acetyl degradation. The acetyl groups constitute 3.8% to 5% of the stover total dry matter and 12–15% of the hemicellulose fraction, and are the most susceptible components to the low temperature acid hydrolysis that occurs during ensilage. This susceptibility was evident from the large amount of the acetyl fraction degraded during storage, up to 49.47% ± 3.12 at 35% moisture, 37◦C (see **Supplementary Material**), with a maximum of ∼56%. This implies that fewer acetyl groups would be available in ensiled feedstock for conversion to acetic acid during pretreatment. Alternatively, the organic acids present in ensiled samples, up to 9.1% of total dry matter compared to <0.5% for unensiled samples, could serve as buffering agents, resisting pH change. This second factor is supported by the larger pH change in washed ensiled samples compared to unwashed samples. Without storage derived organic acids interfering with pretreatment, the decrease in storage pH of 0.34 mean pH units during pretreatment of washed samples was more than double that of the unwashed ensiled stover but still much smaller than that of unensiled feedstock. Assuming deacetylation is the main factor accounting for change in pH, the results from this study suggest that the theoretical maximum number of hydrogen ions that can be released from acetyl component of corn stover during storage and/or pretreatment may be enough to bring down the pH to a vicinity of pH 4. On average, the pH of washed ensiled samples (pH 4.08) was lower than unwashed samples (pH 4.24) (p = 0.001).

Although the pH of unensiled samples decreased more dramatically during pretreatment, mean resultant pH values were still higher (4.44 ± 0.17) than for the ensiled samples (4.26 ± 0.16) (p < 0.0001). Relating this to acetyl content, unensiled stoved had 128% more acetyl than ensiled stover, if using 56% upper limit degradation during storage. If all these acetyl groups in unensiled samples were removed during pretreatment, that would imply about 2.28 times the hydrogen ions compared to ensiled samples. From **Table 1**, it can be inferred that this drastic decrease in pH was a result of more acetyl in the unensiled feedstock, which was then available for hydrolysis to acetic acid during pretreatment. In general, pH decreased with increased pretreatment time. For unensiled samples the pH values at all three-time levels (5, 10, 15, min) were significantly different from each other (4.64 ± 0.09, 4.42 ± 0.08, 4.27 ± 0.06, respectively; p < 0.0001) and feedstock moisture had no significant impact. For ensiled samples, there was no significant difference between 10 and 15 min (4.23 ± 0.10 and 4.18 ± 0.16), both of which were lower than 5 min (4.37 ±0.14) (p < 0.0001). With respect to storage moisture, there was no significant difference in pH at all moisture levels except for 25% moisture, which was higher than the 45% and 55% moisture treatments (p < 0.0001). There was also no significant impact of storage temperature on pretreatment pH. In all, the pH values were moderate and conducive to both enzymatic hydrolysis and ethanol fermentation.

#### Glucan and Xylan Removal

Glucan removal during pretreatment in unensiled stover was ∼58% higher than ensiled (4.49% ± 0.65 vs. 2.84% ± 0.56; p < 0.0001) and storage temperature had no significant impact on amount removed (p = 0.157). Pretreatment time also had no significant impact on glucan removal (p = 0.742 ensiled, 0.525 unensiled). Similarly, xylan removal from ensiled stover was not significantly different across the various pretreatment times (p = 0.210), with the results indicating 5 min (27.67% ± 3.29) was just as effective as 15 min (28.68% ± 2.08). This result may have important commercial implications if the assumption that xylan removal reflects the extent of pretreatment is valid. In contrast, xylan removal from unensiled stover was significantly lower after 5 min of pretreatment (22.23% ± 2.06) compared to 10 (26.62% ± 2.31) and 15 min (27.30% ± 1.12) of pretreatment (p < 0.001). At the longest retention time of 15 min, xylan removals for ensiled and unensiled stovers were not significantly different from each other. This result suggests as retention time or pretreatment severity increases wet storage benefits for pretreatment are masked. Ensiled stover had significantly higher xylan removal, about 28% on average, compared to 25% for unensiled samples (p < 0.0001). The minimum xylan removed was 17% and maximum was 34% for unwashed samples.

When considering the effect of storage moisture content on glucan removal, the results indicated samples ensiled at 45– 75% moisture and subsequently pretreated were not significantly different from each other, while glucan removal was higher in samples ensiled at 25% and 35% moisture. For unensiled samples,

TABLE 1 | Relating hydrogen ion concentration to acetyl group hydrolysis during pretreatment.


\**The pH of ensiled and unensiled stover is mean pH of all sample without regards to moisture levels or temperature.*

\*\**Expected ratio if all acetyl in sample is completely hydrolyzed to hydrogen ions.*

*Shaded values highlights mean ratio of hydrogen ion concentration in pretreated extract and expected ratio form complete degradation of acetyl group (unensiled:ensiled).*

pretreatment of samples in the range of 25% to 55% moisture did not experience significantly different glucan removal. With respect to xylan removal, the effect of moisture was only observed in the ensiled samples. At 23◦C, xylan removal was highest at 35% moisture (30.84% ± 2.48) although this difference was only significant relative to the 25% and 65% moisture samples, and both of these treatment conditions were not significantly different from other moisture levels (p = 0.007). At 37◦C, xylan removal at 45%, and 55% moisture was only significantly higher than the 35% moisture treatment. These results showed storage temperature had some effect on xylan removal and revealed a significant interaction with storage moisture (p < 0.0001). Generally, samples stored at 23◦C experienced more xylan removal during pretreatment than samples stored at 37◦C (27.58% ± 3.17 vs. 25.05% ± 3.21; p = 0.001). The lower average xylan removal rate after ensilage at 37◦C could be biased by a few samples at the extremes of the storage moisture range, which had no lactic acid or lower lactic acid at 37◦C compared to 23◦C.

Washing ensiled samples appeared to increase xylan removal. For example, during pretreatment for 15 min the washed ensiled samples experienced more xylan removal than the unwashed ensiled samples, 36.36% ± 4.56, and 23.89% ± 2.73, respectively (p < 0.0001). Since washed samples do not contain organic acids, the implication is that contribution of silage organic acids to pretreatment is primarily during the storage process, rather than during the subsequent conventional pretreatment process. Although organic acids accelerate xylan removal during pretreatment, they may also interfere with, and limit xylan removal during conventional pretreatment. This acceleration and limitation was observed in data on the amount of xylan removed in 5 min compared to 15 min in unwashed stover. Under these circumstances xylan removal did not increase significantly with pretreatment time from 5 to 15 min, although higher removal from the washed samples indicated more xylan was potentially available for removal in unwashed samples. Despite this apparent interference, xylan removal in wet storage samples was still better [at shorter retention times] or comparable [at longer retention times] to unensiled samples, as previously discussed.

#### Organic Acids and Inhibitors From Pretreatment Organic Acids

Organic acids are a product of anaerobic fermentation and provide the primary mechanism of preservation during ensilage; some can also be formed during pretreatment. The main acids

identified in the pretreatment extracts of unwashed corn stover feedstock were lactic (≤4.0% DM), acetic (≤2.2% DM), and isobutyric (≤3.9% DM) acids (see **Figures 1**, **2**). Individually, none of these acids exceeded 6 g L−<sup>1</sup> (mass per pretreated volume). Low levels of tartaric, malic, formic, pyruvic were also detected. Wet storage and pretreatment conditions affected which acids were dominant, and these were different for different conditions. Lactic acid was the dominant acid in wet stored, pretreated feedstock (2.94% DM ± 0.81 [Ensiled] vs. 0.14% DM ± 0.31 [Unensiled]) while acetic acid was dominant in unensiled samples [1.06% DM ± 0.31 [Unensiled] vs. 0.64% DM ± 0.33 [Ensiled]]. Isobutyric acid was equally high after pretreatment for both before and after storage samples (1.61% DM ± 0.91 [Ensiled] vs. 1.42% DM ± 0.75 [Unensiled]).

For most ensiled treatments, the amount of lactic acid increased during pretreatment while the amount of acetic acid decreased. Samples with lower lactic acid (0.0 < 1.50% DM; 25% and 75% moisture) prior to pretreatment generated more lactic acid (up to 4.22% DM) and samples with higher lactic acid (1.94–3.20% DM) generated <1.40%, with ∼68% from this group having ≤ 0.7% above storage levels and 17% showing a slight decrease below storage levels (see **Supplementary Data**). Lactic acid could have been produced through hydrothermal deamination and hydroxylation of amino acids, or in small amounts via hydrothermal degradation of polysaccharides (Quitain et al., 2002; He et al., 2008). Another potential source of lactic acid generation may be a reaction of acetic acid formed during storage with formaldehyde, which is also produced during pretreatment (see Equation 2).

$$\begin{array}{cccc} \text{C}\_2\text{H}\_4\text{O}\_2 & + & \text{CH}\_2\text{O} & \rightarrow & \text{C}\_3\text{H}\_6\text{O}\_3\\ \text{Acotic acid} & + & \text{Formaldehydro} & \rightarrow & \text{Lactic acid} & \end{array} \tag{2}$$

At standard conditions, this reaction is spontaneous. This pathway is proposed based on the disappearance, during pretreatment, of some of the acetic acid present after wet storage. Also, compared to washed samples with no storage acids, hence no acetic acid, the amount of lactic acid generated during pretreatment was less than the amount produced during storage but washed out before pretreatment. Lactic acid in the pretreatment extract of washed samples was in general <0.5% DM. Formaldehyde, assumed in Equation 2 as reacting with acetic acid, can be produced from thermohydrolytic degradation of xylose (Schäfer and Roffael, 2000; Roffael and Hüster, 2012).

About half of the unwashed wet storage samples with an acetic acid concentration lower than 1% DM [mostly samples from 25 to 45% moisture] had nearly a percentage point decrease (up to 0.9% DM) in the original amount after pretreatment, while the other half showed an increase (up to 0.8% DM) above the wet storage amount. In contrast, wet storage samples with acetic acid concentrations >1% to 2.8% DM [55–75% moisture] showed acetic acid decrease, which was up to 2.3% of the dry matter in the higher acetic acid samples. Isobutyric acid in pretreatment extracts was high and was not significantly different for both unensiled and ensiled samples. In most cases, there was an increase above the initial wet storage amount (up to 3.1% DM) except for the retention time of 5 min, where 65 and 75% moisture samples showed a decrease in isobutyric acid.

In the pretreatment extracts from washed samples, malic acid was dominant (2.11% ± 1.73 DM), far more than acetic and lactic acids which averaged ≤1% DM and ≤0.51% DM, respectively. Formic acid amounts were comparable to lactic acid. Traces of succinic acid were observed in the 75% moisture samples. No isobutyric acid was present in pretreatment extracts from any washed samples. This lack of isobutyric acid in washed samples contrasts with the relatively high levels of isobutyric acid after pretreatment in the unwashed samples. The increase in the isobutyric acid concentration of unwashed samples could thus be

boxplot graphing tool as at least 1.5 interquartile ranges from edge of box).

Error bars are ± standard deviation of mean; *n* = 4 per treatment group.

due to interactions of organic acids or other extractives washed out of the feedstock. Unlike acetic and lactic acids, isobutyric acid is usually not reported in studies of hydrothermal processing of lignocellulosic materials, and a specific abiotic reaction that could occur during pretreatment has not been elucidated.

Total organic acid in pretreatment extracts were not significantly different at the various pretreatment retention times (p = 0.353) and various storage moisture levels (p = 0.306). However, wet storage did have an impact. Organic acids were virtually absent in control (Unensiled) samples prior to pretreatment. These unwashed unensiled samples behaved similarly to the washed wet stored samples, generating more acids during pretreatment than unwashed wet stored samples. On average, however, total acids from pretreated unwashed wet stored feedstock were still higher (∼6.54% DM) than for unwashed unensiled samples (∼4.46% DM) (p < 0.0001).

The unwashed samples also showed apparent changes in the organic acid profile after pretreatment. Butyric acid, present in a number of unwashed high moisture samples in amounts >1% DM, disappeared completely. Individually, the amounts of lactic acid and acetic acid in ensiled samples were not significantly different at the various pretreatment times (p = 0.405 and 0.118, respectively). For unensiled samples, the lactic acid concentration was not different for the various pretreatment times (p = 0.642). However, the acetic acid concentration at 15 min was significantly higher than at 5 and 10 min (p < 0.0001). With respect to temperature, samples stored at 23◦C on the whole had lower lactic acid (p = 0.001) but more acetic (p = 0.002) than 37◦C samples after pretreatment. Based on the amount of storage acetic and lactic acids at 23◦C and 37◦C, a similar inverse relation was observed as was described previously, in which samples with relatively lower initial [storage] acids had higher amounts generated during pretreatment than samples with higher initial [storage] acids and vice-versa.

#### Inhibitors

In addition to organic acids, two sugar degradation products 5- (Hydroxymethyl) furfural (HMF), and furfural were measured in pretreatment extracts. The former derives from hexoses and the later from pentoses, and both are known to inhibit many ethanologens including yeast. Generally, both of these inhibitors were higher in ensiled samples compared to unensiled samples and both were not affected by storage temperature (p > 0.5). On average, HMF concentrations were <0.05% DM (0.039% ± 0.018) and furfural concentrations were <0.5% DM (0.47% ± 0.27) in unwashed ensiled samples. These HMF and furfural concentrations were about 30 and 75% higher, respectively, than the amounts in unensiled samples. The ratio of xylan removed to glucan removed was approximately the same as the ratio of furfural to HMF produced, and this ratio was on the order of 10. On a per [pretreated] volume basis, concentration of HMF and furfural were 0.03 ± 0.05 g L−<sup>1</sup> extract and 0.48 ± 0.32 g L−<sup>1</sup> extract, respectively.

In contrast to unwashed samples, washed samples had no HMF in the pretreatment extracts, but far higher furfural concentrations than were observed in the unwashed samples, averaging 1.12% ± 0.26 (DM) or 1.5 g L−<sup>1</sup> extract at 15 min retention time. Glucan removal during pretreatment from washed samples was higher than that from unwashed samples (3.28% DM ± 0.38% vs. 2.81% DM ± 0.43%, p = 0.017). This higher glucan removal was expected to result in larger amounts of HMF in the pretreatment extracts from the washed samples, but this was not the case. This suggests that HMF was produced from degradation of preexisting glucose in the water-soluble components of corn stover or glucose hydrolyzed during wet storage rather than from the glucose produced from structural degradation during the pretreatment process. This result thus supports the hypothesis that decomposing valuable feedstock components to simpler, more bioavailable forms during wet storage increases their risk of being degraded in subsequent processing to less valuable forms. Alternatively, it is possible that other water-soluble compounds in the stover or produced during ensilage may catalyze glucose degradation or serve as reaction partners in the formation of HMF in unwashed samples. Their absence in the washed samples would therefore hinder the formation of HMF. The higher furfural could be attributed to the higher xylan removal from washed samples (p < 0.0001). At 15 min pretreatment retention time xylan removal was 36 and 24% of theoretical for washed and unwashed samples, respectively. Furfural generated from the pretreated washed samples previously stored under the extreme moisture conditions (25 and 75%) were lower [<1%-point, g furfural/g biomass] than amounts produced during pretreatment of washed samples from the mid-range 35–65% moisture wet storage samples, which were not significantly different from each other. As noted earlier, furfural is formed when xylose is degraded and from pretreatment extract analysis, samples at these extreme moistures had lower xylan removal, hence lower furfural.

The amounts of both the HMF and furfural inhibitors increased with pretreatment time as expected. For unensiled feedstock, amounts of HMF and furfural generated at each pretreatment retention time were significantly different from each other (p < 0.05) (see **Table 2**). In wet stored unwashed stover, HMF generated during 5 and 10 min of pretreatment were not significantly different and were lower than the amount generated in 15 min. Furfural, however, was different for all pretreatment times. For the unensiled vs. wet stored stover, there was more than a 100 and a 60% increase in furfural, respectively, for every 5 min increase in pretreatment time. Furfural produced during pretreatment was not significantly different for the various storage moistures. HMF was also not significantly different except for 45% moisture ensiled stover samples (which had higher amounts than observed for the 65% samples) and 35% moisture (which had higher amounts than observed for the 75% moisture samples of unensiled stover).

#### Fermentation

#### Ensiled vs. Unensiled

In most cases, there was no significant difference in ethanol yields of unensiled (Day 0) and ensiled (Day 220) unwashed stover. This was true whether the stover samples were fermented with or without their pretreatment extracts (p: with extract = 0.745, without extract = 0.235) (see **Table 3**). However, the pretreatment response of samples with or without wet TABLE 2 | Furfural and HMF generated during liquid hot water pretreatment of unwashed corn stover, 23◦C, averaged across all moisture levels (*N* = 24 per treatment group).


\**All intercepts set to zero.*

*R* <sup>2</sup> *was similar in zeroed and actual intercept except for unensiled furfural in which the actual equation [(0.0411* × *pretreatment time)*−*0.1426] had an R*<sup>2</sup> *of* ∼*95%.*

*†Analysis of water-soluble extracts collected before pretreatment showed there was no furfural or HMF present in the feedstock before pretreatment.*

TABLE 3 | Ethanol yields, on percent of theoretical basis, averaged for all moisture levels at each different pretreatment retention time\*.


\**Mean results along with standard deviation pooled across the six moisture levels.*

*Means without a common superscript letter in a row differ significantly as analyzed by two-way ANOVA and the TUKEY test; the same color down column indicates not significantly different. n* = *12 per treatment group. Empty cells* = *no data because these conditions were not part of this study design.*

storage was different in terms of the dominant acids, inhibitors produced, and xylan removed at shorter retention times. Organic acids, furfural, and HMF were higher in ensiled samples, and so was xylan removal. The first group can adversely affect fermentation yields for organisms such as yeast, depending on their concentrations and the pH of the fermentation broth, while xylan removal could make glucan more accessible, and thus favor ethanol yields. This is especially true in systems like this study, where a typical hexose fermenting yeast was used. Xylan removal may also be beneficial for fermentations with consolidated bioprocessing organisms such as Clostridium thermocellum, for which pentoses have been shown to be inhibitory (Verbeke et al., 2017). Ranges of the dominant acids in the fermentation broths, on mass per fermentation volume basis, were 0.00– 2.17 g L−<sup>1</sup> , 0.00–1.07 g/L, and 0.00–2.69 g L−<sup>1</sup> for isobutyric, acetic, and lactic acids, respectively. Acetic acid was dominant in unensiled feedstock while lactic was dominant in ensiled feedstock.

Acetic acid levels above 0.50 g L−<sup>1</sup> lead to intracellular accumulation that could affect cell growth, ethanol production or both, while lactic acid concentrations >8.0 g L−<sup>1</sup> could lead to cell death depending on type of yeast and intracellular pH (Narendranath et al., 2001; Ingledew, 2003). Although lactic acid concentrations in the present study were much lower than this 8 g L−<sup>1</sup> level, 92% of unensiled samples in this study had acetic acid concentrations >0.50 g L−<sup>1</sup> . However, the fermentation pH of 4.8 in this study is higher than the pH values of 3.0–4.0 in Narendranath et al. (2001) and Ingledew (2003). Thus, the effect of these acids on fermentation microbes in the present study is expected to be less than in these prior studies, due to the lower amount of undissociated acids. A more recent study conducted by Xu et al. (2010) found that acetic acid, which is more inhibitory than lactic acid, only inhibited ethanol yields when the concentration used in LHW pretreatment exceeded 6% DM. In addition, at 6% DM, 0.51 g L−<sup>1</sup> furfural, 0.07 g L−<sup>1</sup> HMF, and 4.5 g L−<sup>1</sup> acetic acid present in fermentation broth did not have any inhibitory effect on fermentation yields but was similar to yields of control samples without acetic acid. When acetic acid used in pretreatment was less than 6% DM, concentrations of ethanol were higher than the control, up to 8.63 g L−<sup>1</sup> compared to 7.63 g L−<sup>1</sup> in the control for extracted pretreatment liquor; and up to 33.72 g L−<sup>1</sup> compared to 21.95 g L−<sup>1</sup> for the solid fraction. The maximum ethanol yield from the pretreatment liquor was obtained when acetic acid was 1.0%–3.0% DM, equivalent to 1.5 g–2.5 g acetic acid L-−<sup>1</sup> in the fermentation broth.

Acetic acid, after pretreatment of unwashed samples, was generally <2% of the dry mass of corn stover. The maximum concentration was 1.1 g L−<sup>1</sup> fermentation volume, which was observed for 75% moisture unensiled stover pretreated for 15 min. Only 8% of the unensiled samples, which had higher acetic acid concentrations than ensiled samples, had acetic acid concentrations exceeding 1 g L−<sup>1</sup> (all at 65 and 75% moisture). Furfural and HMF concentrations in unwashed samples were 0.346 g L−<sup>1</sup> ± 0.03 and 0.028 g L−<sup>1</sup> ± 0.002, respectively. Although about 10% of the unwashed samples in this study had furfural levels exceeding 0.51 g L−<sup>1</sup> , the maximum HMF was below 0.06 g L−<sup>1</sup> . Maximum furfural concentration in unwashed samples was 0.73 g L−<sup>1</sup> for 55% moisture ensiled stover pretreated for 15 min. The implication is that none of the potential inhibitors produced during pretreatment affected fermentation yield.

Even at higher solids loading, the concentrations in themselves may not contribute to significant inhibition. Graves et al. (2006) observed that with 25–30% solids in corn ethanol fermentation, Saccharomyces cerevisiae could tolerate, at pH ≥ 5, more than double the projected lactic acid concentration this study predicts would occur at 30% solids. There are several factors that complicate the predictability of the effects of these organic acids or furfural on ethanol yield under high solids loading. These include osmotic stress due to higher sugar concentrations (Darku and Richard, 2011); lower enzyme adsorption rates (Kristensen et al., 2009); mass transfer limitations (Varga et al., 2004; Kristensen et al., 2009; Zhang et al., 2009); and ethanol inhibition from higher ethanol concentration (Mohagheghi et al., 1992). Furthermore, these inhibitors act individually and synergetically, and their tolerable concentrations and combined effects are influenced by other factors, mainly pH (Palmqvist et al., 1999).

Disregarding these high solids limitations and scaling up the results from this study to reflect 30% solids loading, similar yields can be expected from both ensiled and unensiled stover based on the previously discussed observations by Palmqvist et al. (1999) and Larsson et al. (1999). This is because although ensiled samples have high furfural, only 14% of these samples at 30% solids loading would contain furfural >2.4 g L−<sup>1</sup> (39% would be <1 g L−<sup>1</sup> ) and maximum acetic acid would be <4 g L−<sup>1</sup> (51% were ≤ 2 g L−<sup>1</sup> ). At these concentrations, a stimulating effect on ethanol yield would be expected. For unensiled samples, acetic acid levels were higher than in ensiled samples. Yet even then, at a hypothetical 30% solids loading, while 89% of the samples would have concentrations >2 g L−<sup>1</sup> , only 4% would exceed 4.8 g L−<sup>1</sup> . In contrast, furfural concentrations were lower than in ensiled samples; at 30% solids the maximum concentration would be <2 g L−<sup>1</sup> and 33% of the samples would have concentrations <1 g/L. With limited understanding of the complex interactions of inhibitors in fermentation broth coupled with the pH effect, it is possible that with higher solids, ensiled and unensiled feedstock could respond differently, with one having better outcomes over the other in terms of ethanol yield. While ethanol productivity will most certainly be affected by 15 min of pretreatment at higher solids loading, final yields may not be affected. Based on the current analysis, the ethanol fermentation outcomes of ensiled and unensiled samples at high solids would be balanced by lower acetic acid in the former and lower furfural in the later, likely resulting in similar yields.

Since wild type Saccharomyces cerevisiae is exclusively a 6 carbon fermenting yeast, the similarity in glucan composition of ensiled and unensiled samples also accounted for the similarity in ethanol yields. The assumption here is that there was no glucose degradation after structural decomposition during pretreatment. This assumption is based on the negligible amounts of HMF generated during pretreatment. At high temperature, under acidic conditions, glucans are degraded to HMF. In addition, washed samples did not have any HMF suggesting that HMF was likely from free water extractable glucose, which in the case of washed samples was washed out. The similarities in yields also suggest that although xylan removal was significantly higher in ensiled feedstock, it was not of practical significance for yeast hexose fermentation. Alternatively, it may be possible that glucan was more accessible in ensiled samples, but its utilization was hindered by inhibitors leading to coincidental similarities in yields with the unensiled samples. The following section, discussing fermentation yields with and without the pretreatment liquids, provides insight into this question.

#### Fermented With Pretreatment Extract vs. Without Extract

Ethanol yields for unwashed samples fermented with pretreatment extracts were significantly higher than samples fermented without extract (p = 0.041) (see **Table 3**), even after normalizing for the glucan removed during pretreatment that ended up in the extract. The amount of potential ethanol lost if extracts are discarded may seem negligible from a mass perspective. Glucan removal was low during pretreatment and on average could theoretically have yielded 0.0073 g ± 0.0002 of ethanol per sample, or 0.59 ± 0.04, g ethanol per liter fermentation volume. In terms of theoretical ethanol yield, this could potentially be up to 5% of the total ethanol that could be derived from the fermentation. Yields of samples fermented with and without extract were 50.57% ± 5.79 and 45.72% ± 6.01 g ethanol g−<sup>1</sup> biomass, or on a mass ethanol per fermentation volume basis, 6.68 g L−<sup>1</sup> ± 0.09 and 5.78 g L−<sup>1</sup> ± 0.11, respectively.

The lower yields of samples without extract could be due to the absence of organic acids at the low, stimulating levels previously discussed. Correlation and regression analysis show that acetic acid has a positive correlation and significant regression relationship with ethanol yield in unensiled samples, but not with ensiled samples. This lack of correlation may be a result of lower acetic acid levels in the ensiled samples, suggesting that there may be a lower limit below which acetic acid has no stimulating effect, just as there is an upper limit as observed in other studies.

Furfural and HMF, at the concentrations found in unwashed stover in this study, were positively and strongly correlated with ethanol yield. **Figures 3**, **4** show a graphical relationship between some of these inhibitors and ethanol yield. The regression equations for unensiled stover were reasonable as they show that without furfural or HMF, ethanol yield was similar to the mean yield in samples fermented without pretreatment extract, hence with no furfural or HMF. This result supports the observation by Palmqvist et al. (1999) that furfural serves as an ethanol stimulant when concentrations are ≤2 g L−<sup>1</sup> . As noted in the previous paragraph, the organic acids, HMF and furfural in the pretreatment extract were not likely to exert any inhibition to

fermentation and could be responsible for the higher yield in samples fermented with pretreatment extracts.

#### Storage at 23◦C vs. Storage at 37◦C

This section compares the ethanol yields of unwashed ensiled samples fermented with pretreatment extract at different storage temperatures. Samples stored at 23◦C had better ethanol yields than samples stored at 37◦C. These ethanol yields, reported as a percentage of theoretical ethanol yield, were 50.34% ± 7.20 at 23◦C vs. 42.45% ± 7.53 at 37◦C; p < 0.001). The difference, however, was due mainly to differences in yields at the extreme moisture levels, 25 and 75%. At these respective storage moistures of 25 and 75%, yields as percentage of theoretical were 47.99% ± 5.72 vs. 41.33% ± 3.76 and 49.12% ± 5.46 vs. 32.88% ± 6.56 for 23◦C vs. 37◦C. Although glucan composition before pretreatment and glucan removal during pretreatment were similar for samples ensiled at these two temperatures, xylan removal was significantly higher in samples stored at 23◦C. In addition, samples stored at 23◦C had higher acetic acid and lower lactic acid content in the pretreatment extracts than samples stored at 37◦C. Thomas et al. (2002) observed that while both acetic acid and lactic acid at low concentrations could enhance ethanol yields, given appropriate pH, acetic acid was a better stimulant and resulted in more ethanol production while lactic acid benefited cell growth more than ethanol yield. At pH 4.5, increasing acetic acid concentration to 16 g L−<sup>1</sup> did not have much inhibitory effect on yields (Thomas et al., 2002). Acetic acid levels in this study were lower and the pH of the fermentation media was approximately pH 4.8, so it is possible that this stimulating effect of acetic acid was responsible for the difference in yields.

#### Washed vs. Unwashed Samples

Samples stored at 37◦C were analyzed for the effect of washing vs. not washing before pretreatment. Although unwashed samples fermented with the pretreatment extract had higher yields as a percentage of theoretical ethanol yields (42.45% ± 7.53) than washed samples (39.31% ± 6.55), there was no significant difference between the two. This was in spite of the higher xylan removed in the washed samples (36% DM) compared to unwashed samples (24% DM). The organic acid profiles of these two groups after pretreatment were also very different, with the washed samples dominated by malic acid, which is less inhibitory than the isobutyric, lactic and acetic acids that were the main acids in the unwashed samples. However, washed

FIGURE 4 | Relationships between ethanol yield and concentration of some potential inhibitors in the fermentation volume of ensiled stover. (A) Furfural (B) HMF [Lactic acid (C) and Isobutyric (D) are included because of their high concentration in ensiled samples, although they do not show obvious correlation with ethanol yield, *p*-values are also <0.05] See Supplementary Materials for correlation coefficients and *p*-values.

samples contained more than twice the amount of furfural found in unwashed samples.

On a furfural mass per fermentation volume basis, washed samples had 0.82 g L−<sup>1</sup> ± 0.19 furfural while unwashed had 0.35 g L−<sup>1</sup> ± 0.18 at a pretreatment retention time of 15 min. The furfural concentrations in the washed samples were not expected to inhibit ethanol production from discussion under fermentation of ensiled and unensiled stover. It is, however, possible that higher levels beyond 0.51 g L−<sup>1</sup> (Xu et al., 2010) or 0.6 g L −1 (Palmqvist et al., 1999) could inhibit production in the presence of other compounds in the pretreatment extract. Boyer et al. (1992) found the growth rate of S. cerevisiae was not affected at furfural concentrations ≤1 g L−<sup>1</sup> and specific ethanol productivity (gram ethanol per gram feedstock per hour, g/g/h) was not affected at 1.5 g L−<sup>1</sup> . Even at a furfural concentration of 2 g L−<sup>1</sup> , while specific ethanol productivity was significantly reduced, the final ethanol yield was not (Boyer et al., 1992). In the present study, fermentation was not allowed to proceed to completion, and that could have partially masked any inhibitory effect of borderline inhibitory concentrations of furfural, which are assumed to be responsible for the lower trending (but not significantly lower) yields of washed samples. The inhibitory effect of these compounds is dependent on several factors including pH, inoculation rate, substrate concentration, and the presence of other compounds.

Comparing washed samples fermented with extract to those fermented without extract, it was observed that although the ethanol yields of the latter were on average higher (45.65 % ± 6.62 vs. 41.56% ± 6.21, of theoretical ethanol, at 15 min pretreatment time), the difference was also not significant (p = 0.132). For the same pretreatment duration, and 37◦C storage, ethanol yield of washed samples without extract were also not significantly different from that of unwashed samples fermented with extract, at 46.86% ± 11.29 of the theoretical ethanol yields. The implication is that although furfural may have some effect on fermentation at a concentration of 0.82 g L−<sup>1</sup> , this effect is not pronounced.

#### Effect of Storage Moisture and Pretreatment Retention Time

With respect to the main processing alternatives investigated in this study, storage moisture had no significant influence on ethanol yield among all the samples of unwashed stover, both unensiled and ensiled at 23◦C, whether fermented with

or without their pretreatment extracts (see **Figure 5**). However, for unwashed samples ensiled at 37◦C, there was a storage moisture effect. Although yields for 75% moisture samples [unwashed with extract] were not significantly different from 25% moisture, they were lower than yields from all other moisture treatments, averaging 37% of theoretical ethanol yield compared to yields >43% of theoretical for samples in the intermediate moisture range (p = 0.008). These high moisture samples were high in butyric and acetic acids and had no lactic acid prior to pretreatment. During pretreatment the butyric acid was eliminated, the acetic acid was reduced, and lactic acid was generated. The only marked differences between the 75% moisture samples and other samples were the lack of lactic acid and the high butyric acid concentrations. These differences are likely related, as lactic acid is known to be metabolized to butyric acid by clostridia during ensilage at high moisture levels (McDonald et al., 1991). This suggests that retaining high levels of lactic acid during storage, which is in a more dissociated form than the other organic acids due to storage pH, could have favorably impacted feedstock structure at other moisture levels. Although lactic acid at low concentrations can stimulate fermentation, these results as well as observations from previous paragraph suggest that maintaining lactic acid concentrations during storage is of greater value than generating lactic acid during pretreatment.

In contrast to storage moisture, pretreatment retention time frequently had a significant impact on ethanol yields. Generally, ethanol yields increased with pretreatment time. For unensiled samples, with and without their pretreatment extracts, yields at the various pretreatment times were significantly different from each other. Importantly, for ensiled samples, yields from the 10-min pretreatment duration was not significantly different from 15 min, except for the 23◦C unwashed samples fermented with extract. This suggests that ensilage may permit a reduction of pretreatment severity without sacrificing yield, potentially reducing conversion costs. These pretreatment time comparisons are detailed in **Table 3**. (For data used in used in analysis made in this study, section Results and discussion, see the **Supplementary Material**).

# CONCLUSIONS

The results of this study indicate that the organic acids produced during wet storage and/or pretreatment generally do not inhibit ethanol fermentation and instead can enhance the fermentation yield. These effects of organic acids can be observed at three levels: (1) at the storage level, they potentially alter feedstock structure, resulting in more xylan removal, or weaker linkages between components of the plant cell wall matrix, (2) during subsequent pretreatment, when organic acids can accelerate as well as limit xylan removal depending on the acids involved, and (3) during fermentation, when organic acids, and some other pretreatment products known be inhibitory to yeast at high concentrations can instead provide a minor benefit at low concentrations. The contributions of organic acids to the downstream conversion of corn stover feedstock to ethanol are greater at the storage level, and the partial pretreatment benefits are more pronounced during storage than in subsequent thermochemical pretreatment processing. The organic acids produced during this wet storage study were not inhibitory to S. cerevisiae, and interactions of the acids with conversion processes leading to the final ethanol products were mostly positive.

Lactic acid, which is less inhibitory than some other organic acids, was dominant in ensiled samples and its low pKa means more of it was in the dissociated form. Higher levels of dissociated acids mean more hydrogen ions that can interact favorably with structural bonds. When these effects occur at the storage level the disassociated acids cannot be easily assimilated into microbes even if retained in subsequent processes and are thus less likely to inhibit microbial growth or ethanol production.

An important observation of this study is that the acid profile that was generated during wet storage changed during pretreatment. When lactic acid is dominant during storage there is the potential for production of more acetic acid during LHW pretreatment. This sequence benefits the downstream fermentation, since acetic acid is a better ethanol stimulant than lactic acid. As has long been observed in the livestock forage industry (McDonald et al., 1991), lactic acid dominated silage has the best outcomes during both storage and in subsequent bioconversion. From an engineering design perspective, it would be useful to develop coupled ensilage/pretreatment systems that encourage more lactic acid production during storage and more acetic acid in subsequent pretreatment processing, as long as those acid concentrations are <5% DM.

Using both xylan removal and ethanol yield as proxies for pretreatment effectiveness, these results also provide evidence that pretreatment of ensiled stover could be carried out at shorter pretreatment times and thus lower severity than unensiled stover, and still be as effective. There was evidence from the xylan removal results that wet storage resulted in changes that rendered the feedstock more responsive to subsequent pretreatment process. Fermentation results also indicated ensiled stover could achieve similar ethanol yields with shortened pretreatment times. However, as pretreatment severity increases, the benefits derived from ensilage decrease. Xylan removal rates by themselves were not always predictive, providing an indication of pretreatment severity, and perhaps pretreatment effectiveness, but not necessarily fermentation outcomes, perhaps due to the presence of other compounds generated during storage or pretreatment.

In general, if extreme storage moistures (25 and 75%) are avoided, the impacts of moisture on subsequent process outcomes, especially respect to ethanol yields, are not significant. This is also good news for biorefineries, which thus do not have to be overly concerned with process adjustments to accommodate changes that might result from different storage moistures, which are difficult to control in an industrial feedstock supply chain.

This study documents multiple benefits and few limitations associated with wet storage of biomass for ethanol fermentation. As the feedstock supply chains for lignocellulosic biofuels expand to millions and eventually billions of tons worldwide, wet storage can be an effective strategy for both preserving biomass and enhancing downstream conversion processes.

## AUTHOR CONTRIBUTIONS

DE and TR conceived the idea. DE designed, planned and performed the experiments. TR provided valuable feedback that shaped the research as it progressed. DE performed the data analysis, discussed results with TR, drafted the manuscript and designed the figures. TR reviewed and edited the manuscript.

#### FUNDING

This research was supported by the U. S. Department of Energy, Office of Energy Efficiency and Renewable Energy, Bioenergy Technologies Office, under DOE Idaho Operations Office Contract DE-AC07-05ID14517. Additional project support was provided by Penn State University and Agriculture and Food Research Initiative Competitive Grant No. 2012- 68005-19703 from the USDA National Institute of Food and Agriculture.

#### REFERENCES


#### ACKNOWLEDGMENTS

The authors are grateful to Kay DiMarco and the many students that helped out in the lab. The authors would also like to thank the Bioenergy Technologies Office for their support. The views and opinions of the authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2018.00195/full#supplementary-material


maize for ethanol production. Appl. Biochem. Biotechnol. 148, 23–33. doi: 10.1007/s12010-008-8134-2


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Essien and Richard. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Integration of Pretreatment With Simultaneous Counter-Current Extraction of Energy Sorghum for High-Titer Mixed Sugar Production

#### Daniel L. Williams 1,2, Rebecca G. Ong2,3, John E. Mullet 2,4 and David B. Hodge5,6 \*

<sup>1</sup> Department of Chemical Engineering and Materials Science, Michigan State University, East Lansing, MI, United States, <sup>2</sup> DOE Great Lakes Bioenergy Research Center, Michigan State University, Madison, MI, United States, <sup>3</sup> Department of Chemical Engineering, Michigan Technological University, Houghton, MI, United States, <sup>4</sup> Department of Biochemistry & Biophysics, Texas A&M University, College Station, TX, United States, <sup>5</sup> Department of Chemical & Biological Engineering, Montana State University, Bozeman, MT, United States, <sup>6</sup> Division of Chemical Engineering, Luleå University of Technology, Luleå, Sweden

#### Edited by:

Allison E. Ray, Idaho National Laboratory (DOE), United States

#### Reviewed by:

Héctor A. Ruiz, Universidad Autónoma de Coahuila, Mexico Ning Sun, Lawrence Berkeley National Laboratory (LBNL), United States

> \*Correspondence: David B. Hodge david.hodge3@montana.edu

#### Specialty section:

This article was submitted to Bioenergy and Biofuels, a section of the journal Frontiers in Energy Research

Received: 25 May 2018 Accepted: 21 November 2018 Published: 14 January 2019

#### Citation:

Williams DL, Ong RG, Mullet JE and Hodge DB (2019) Integration of Pretreatment With Simultaneous Counter-Current Extraction of Energy Sorghum for High-Titer Mixed Sugar Production. Front. Energy Res. 6:133. doi: 10.3389/fenrg.2018.00133 Sorghum (Sorghum bicolor L. Moench) offers substantial potential as a feedstock for the production of sugar-derived biofuels and biochemical products from cell wall polysaccharides (i. e., cellulose and hemicelluloses) and water-extractable sugars (i.e., glucose, fructose, sucrose, and starch). A number of preprocessing schemes can be envisioned that involve processes such as sugar extraction, pretreatment, and densification that could be employed in decentralized, regional-scale biomass processing depots. In this work, an energy sorghum exhibiting a combination of high biomass productivity and high sugar accumulation was evaluated for its potential for integration into several potential biomass preprocessing schemes. This included counter-current extraction of water-soluble sugars followed by mild NaOH or liquid hot water pretreatment of the extracted bagasse. A novel processing scheme was investigated that could integrate with current diffuser-type extraction systems for sugar extraction. In this approach, mild NaOH pretreatment (i.e., <90◦C) was performed as a counter-current extraction to yield both an extracted, pretreated bagasse and a high-concentration mixed sugar stream. Following hydrolysis of the bagasse, the combined hydrolysates derived from cellulosic sugars and extractable sugars were demonstrated to be fermentable to high ethanol titers (>8%) at high metabolic yields without detoxification using a Saccharomyces cerevisiae strain metabolically engineered and evolved to ferment xylose.

Keywords: sorghum, sucrose extraction, decentralized biorefining, pretreatment, cellulosic biofuels

# INTRODUCTION

Ethanol derived from sucrose and starch crops has seen substantial growth in recent years (Babcock, 2012), and a non-trivial fraction of the agricultural output of some countries such as the U.S. and Brazil are currently devoted to supply these first-generation fuel ethanol processes. Continued growth in the global capacity for renewable biofuels and bio-based products requires alternative feedstocks such as lignocellulosic biomass. Technologies have begun to be commercialized that are able to utilize the polysaccharides that comprise cell walls of plants (i.e., lignocellulose) as a feedstock for fermentationderived ethanol (Schwab et al., 2016). However, the nascent cellulosic biofuel industry faces challenging economics, with a high cost of cellulosic sugars relative to starch and sucrose due to the high capital and operating costs associated with large centralized lignocellulosic biorefineries (Dale, 2017). Approaches to address this challenge include improving process economics through novel processing approaches, novel or improved biomass feedstocks, and the production of additional co-products from the biomass. The low bulk density of herbaceous feedstocks such as corn stover and switchgrass represents an important challenge for the logistics of feedstock transport and storage for commercial-scale biorefineries. A lignocellulosic biorefinery that uses a biological process consisting of pretreatment such as dilute acid or dilute aqueous ammonia, enzymatic hydrolysis, and fermentation to produce cellulosic ethanol will need to process on the order of 2,000 dry tons per day of biomass to generate comparable ethanol yields as a typical dry grind corn ethanol plant (50 MM gal/yr). This presents an enormous set of logistical and storage challenges for these low bulk density feedstocks. One processing approach that addresses this challenge is decoupling select components, such as feedstock handling and potentially processing such as pretreatment, from the centralized biorefinery (Thompson et al., 2013). Decentralized biomass aggregation and preprocessing facilities that feed into larger centralized biorefineries have been proposed as solutions to the challenges associated with feedstock logistics. As one example, decentralized, depot-scale preprocessing that employs AFEX pretreatment and pelletization could provide a higher bulk-density feedstock that facilitates transportation and storage, enabling year-round supply, decreasing the sensitivity to supply chain disruptions, and potentially yielding a product with additional applications, such as a high-digestibility ruminant feed (Campbell et al., 2013; Lamers et al., 2015).

Another approach to addressing the economic challenges of cellulosic ethanol is by developing feedstocks with improved agronomic or processing attributes. Sorghum (Sorghum bicolor L. Moench) is one potential advanced feedstock that has been proposed due to a number of positive attributes. Based on whether the plant partitions significant stocks of carbon into grain, stem sucrose, or structural biopolymers, sorghum is classified as grain sorghum, sweet sorghum, and forage/silage/energy sorghum, respectively (Rooney et al., 2007). Proposed feedstock benefits of sorghum include its genetic tractability and diversity (Mullet et al., 2014), high biomass productivity (Rooney et al., 2007; Olson et al., 2012), and, when utilized for biofuel production, substantial greenhouse gas reduction benefits relative to gasoline (Cai et al., 2013). Additionally, like sugarcane, some sorghums (e.g., sweet and forage/silage/energy) accumulate water-extractable sugars (sucrose, fructose, glucose, and starch) in stems that are readily fermentable to ethanol (Eggleston et al., 2013). Unlike sugarcane, however, some cultivars of sorghum are suitable for cultivation in temperate climates, including much of the continental U.S. (Rooney et al., 2007). Sorghum cultivars/hybrids may be highly suitable for marginal lands as they can require less than two thirds of the water and less fertilizer compared to corn, with some varieties showing high drought tolerance (Eggleston et al., 2013). Further environmental benefits include advantages in process water utilization relative to starch-derived ethanol or other cellulosic ethanol technologies in that, as with sugarcane ethanol, high-moisture biomass is transported to the processing plant, while water and nutrients can be returned to field as vinasse (Chum et al., 2014).

The ability of sucrose-accumulating sorghum varieties to partition a significant fraction of the fixed carbon into waterextractable sugars suggests that hybrid processing schemes can be developed that integrate first- and second-generation biofuels technologies. Like sugarcane, sucrose-accumulating sorghum varieties can lose sucrose content quickly post-harvest, often within hours of being cut (Eggleston et al., 2013), which would require that stalks be processed immediately, and sugars be stabilized or potentially fermented at decentralized biomass processing depots before being transported to a centralized biorefinery for further processing. Another consideration for these processes is that, like sugarcane, the harvest period defines a limited window of operation, therefore many sugarcane processing mills only operate for a portion of the year. This limited feedstock lifetime, low bulk density, and high moisture all suggest that sorghum may lend itself to decentralized processing. A number of biofuels processing technologies have been proposed specifically for sweet sorghum utilization, which could be integrated into decentralized depot-scale processing facilities, making use of both the soluble sugars and structural polysaccharides within the biomass. Examples of sweet sorghum utilization include sucrose extraction and fermentation followed by ensiling the bagasse prior to anaerobic digestion (Zegada-Lizarazu and Monti, 2015), solid-state fermentation of soluble sugars to ethanol followed by alkaline pretreatment, enzymatic hydrolysis, and fermentation (Li et al., 2013), direct anaerobic digestion (Matsakas et al., 2014), and integration of ensiling with enzymatic hydrolysis of cell wall polysaccharides for ethanol production (Henk and Linden, 1994).

Sorghum harvest and processing for sugar extraction has been demonstrated using technologies developed for sugarcane (Noah and Linden, 1989; Woods, 2000; Webster et al., 2004), however maximum juice purities (i.e., sucrose, glucose, and fructose as a percentage of total solubles) tend to be lower, 75% for sorghum (Eggleston et al., 2013), relative to 88% for sugarcane (Webster et al., 2004) although this depends on feedstock properties. Commercial sugar or ethanol mills utilizing sugarcane employ counter-current extraction of sucrose from shredded stalks using either tandem roller-mills or diffuser extraction (Koster, 1995). Diffuser extraction is the most common technology, with extraction efficiencies as high as 98% for sugarcane, however it has some disadvantages compared to other methods, including higher levels of impurities and clarification requirements for sucrose production, and longer start-up and shut-down periods. However, extraction using diffusers also has lower capital costs, operating costs, and energy requirements relative to extraction by tandem milling (Rein, 1995). The most utilized sugarcane diffuser commercially is the counter-current moving bed diffuser with typical design capacities ranging from 1,000 to 40,000 wet tons

sugarcane/day. For these processes, the sugarcane stems are first subjected to combinations of crushing, shredding, and/or milling with some diffuser designs extracting soluble sugar at this stage. The prepared sugarcane is next introduced to a porous screen and is conveyed along the screen while juice is collected under the screen in a series of 10 to 18 collection troughs. This juice is reintroduced to the moving bed of solids in the region one stage upstream from where it was collected, yielding a counter-current extraction process. Upon leaving the screen, the extracted bagasse is dewatered in roller mills (Baikow, 1967; Rein, 1995).

Besides their application for sugar extraction, counter-current solid-liquid extraction technologies have a wide range of existing and potential applications in biofuels processes, where the counter-counter action provides processing benefits over other process configurations such as sequential batch or cocurrent continuous processes. Examples include counter-current pretreatment by dilute acid (Lee et al., 1999) or liquid hot water (Thomsen et al., 2006) that have the benefits of minimizing soluble xylan exposure to high temperature and acid, reducing sugar degradation. Counter-current enzymatic hydrolysis of pretreated biomass has also been demonstrated as a concept to minimize product inhibition by high concentrations of sugars (Lonkar et al., 2017). Counter-current sucrose extraction from sugarcane is performed to maximize both sucrose concentrations and extraction efficiency by minimizing the soluble sugars remaining in the extracted bagasse. Recently, counter-current diffuser technology has been proposed as a pretreatment process (Borden et al., 2015). However, to our knowledge, combination of sucrose extraction and pretreatment using an integrated countercurrent extraction has not been previously investigated and will be addressed in the present work.

The motivation for the current work is the development of a hybrid route for pretreatment and soluble sugar extraction that is suitable for decentralized biorefining. This work investigates the integration of mild NaOH pretreatment with counter-current soluble sugar extraction from an energy sorghum, utilizing both extractable and structural carbohydrates for biofuel production. Considering that minimal work has been published on this topic, this work is intended as a proof of concept to demonstrate the potential for this approach. Specifically, we compare the separate extraction and pretreatment using LHW and NaOH with the integrated approach and finally demonstrate the fermentability of the undetoxified combined juice extract, pretreatment liquor, and enzyme hydrolysate utilizing a yeast strain metabolically engineered to ferment xylose.

# MATERIALS AND METHODS

#### Lignocellulosic Biomass Feedstock

Sorghum bioenergy hybrid TX08001 was used in this work. This bioenergy feedstock is a photoperiod-sensitive hybrid exhibiting high nitrogen use efficiency (Olson et al., 2013) and extended vegetative growth duration due to delayed flowering resulting in high biomass yields (Olson et al., 2012; Gill et al., 2014). Sorghum was grown at the Texas A&M University Farm, College Station, Texas as previously described (Olson et al., 2013) and was harvested prior to floral initiation after 120–150 days of vegetative growth. Following harvest, sorghum stems that account for ∼80% of total shoot biomass, were isolated, milled to pass a 5-mm screen, and dried to a moisture content of 7.3% and stored at 4◦C in sealable bags. Composition analysis of the untreated sorghum stems was performed based on the NREL procedures for biomass analysis (Hames et al., 2008; Sluiter et al., 2008a,b,c, 2012), with alterations as specified by Ong et al. (2016). Extractives-free bagasse for sequential extraction, pretreatment, and hydrolysis was prepared by extensive washing of ∼15 g of sorghum with 600 mL water at 80◦C followed by filtration through a plastic funnel fitted with a 200-mesh porous base. The bagasse was then allowed to air dry before composition analysis and further use in batch pretreatment and hydrolysis experiments. Structural carbohydrate and lignin content of the bagasse and pretreated biomass were determined using the NREL analytical protocols NREL/TP-510-42618 with modifications as described previously (Li et al., 2012).

# Liquid Hot Water and Alkaline Pretreatments of Extractives-Free Bagasse

Following extraction, air-dried bagasse was pretreated using two levels for pretreatment "severity" for both liquid hot water (LHW) and NaOH pretreatment. Rather than attempting to optimize the pretreatment conditions, these two levels were chosen to gauge the range of responses to pretreatment. The LHW pretreatments were performed at 15% solids (w/v) by adding 6 g (dry basis) of bagasse and a total of 40 mL of water (including biomass moisture) into a pressure tube (Ace Glass part number 8648-162) and sealed with a Teflon screwcap. For the "low severity" LHW pretreatment, the pressure tubes were placed in an autoclave for 1 h at 120◦C. For the "high severity" LHW pretreatment, a 10 L M/K Systems digester (M/K Systems, Inc., Peabody, MA) was used as described in our prior work (Stoklosa and Hodge, 2015). For this, the reactor was filled with 5 L of water and pressure tubes were immersed in this water. The reactor was heated at a rate of 1◦C/min until the target temperature of 160◦C was reached, whereupon the reactor was held at 160◦C for 1 h. After 1 h of reaction, the reactor was cooled at a rate of ∼1.3◦C/min until a temperature of 80◦C was reached. The LHW-pretreated bagasse was then diluted to 10% solids (w/v) without washing and placed in flasks in preparation for hydrolysis. NaOH pretreatment was performed at two levels of pretreatment "severity" (based on NaOH loading) in shake flasks at 15% solids (w/v) and NaOH loadings of 0.1 and 0.06 g NaOH/g bagasse. The flasks were incubated at 80◦C for 1 h in a water bath without stirring, and then diluted to 10% (w/w) solids without washing. The pH was adjusted to 5.5 using concentrated sulfuric acid in preparation for enzymatic hydrolysis.

# Counter-Current Sugar Extraction and Alkali Pretreatment

Counter-current sugar extraction and sugar extraction plus NaOH pretreatment were performed using a series of five funnels and filter flasks (**Supplemental Figure 1**). The 3.8 cm diameter funnels were fitted with a 200-mesh stainless steel porous base. For counter-current extraction, 5.0 g of milled sorghum (dry, total mass basis) was introduced at one end of the series of funnels, designated Stage 1, washed with the filtrate from the following stage (Stage 2), and then moved to the next stage (Stage 2). This process was repeated sequentially until the biomass reached Stage 5, at which point fresh imbibition water (50 mL) was passed through the extracted bagasse and subjected to filtration under mild vacuum. The filtrate from each stage was moved sequentially to lower stage numbers during the process (**Supplemental Figure 1**). For example, the filtrate collected from Stage 5 was used to extract the biomass on Stage 4 and the final recovered filtrate was collected after extraction of biomass on Stage 1. A complete cycle was performed (biomass passing from Stage 1 to Stage 5) before the process was assumed to be at "steady-state." Once this steady-state was reached, the process was continued until enough juice and bagasse had accumulated to perform the subsequent hydrolysis and fermentation experiments. For the integrated NaOH pretreatment and extraction, the filtrate from Stage 5 was mixed with 5 M NaOH to yield an alkali loading of 0.06 g NaOH/g extractives-free biomass, slurried with the bagasse from Stage 3, and incubated for 1 h at 80◦C before being subjected to filtration at Stage 4. The bagasse exiting Stage 5 was immediately prepared for enzymatic hydrolysis. The extraction juice recovered from Stage 1 was filter-sterilized (250 mL Stericup-GP, 0.22µm membrane, EDM Millipore) and stored at 4◦C until fermentation was performed. The concentrations of glucose, fructose, and sucrose in each of the filtrates was determined by HPLC (Agilent 1,100 Series) equipped with an Aminex HPX-87H column (Bio-Rad, Milford, MA) and operated at 65◦C and a flowrate of 0.6 mL/min with a mobile phase of 0.005 M H2SO<sup>4</sup> and detection by refractive index.

#### Enzymatic Hydrolysis

For the batch LHW- and alkali-pretreated bagasse samples, pretreated material was diluted to 10% (w/w) solids and the pH adjusted to 5.5 as needed. Following this, sodium citrate buffer (50 mM, pH 5.5), tetracycline (10 mg/L), and cycloheximide (10 mg/L), respectively, were added to the hydrolysates to inhibit microbial growth during hydrolysis. Cellic <sup>R</sup> CTec2 and HTec2 (Novozymes A/S, Bagsværd, Denmark) were added in a protein ratio of 2:1 for an enzyme loading of 15 mg enzyme/g glucan (12.6 FPU/g glucan for the CTec2). The total protein contents of enzyme cocktails used in determining enzyme loadings on biomass were quantified using the Bradford Assay (Sigma-Aldrich). Samples were then incubated at 50◦C for 7 days. For the combined sugar extraction and pretreatment process, the bagasse leaving Stage 5 was diluted to ∼18% solids (w/v) and the pH adjusted to 5.5 using concentrated sulfuric acid. Citrate buffer, antibiotics and enzymes were added and samples were incubated as described above without the cycloheximide, and then filtersterilized and stored at 4◦C until fermentation was performed. Sugar concentrations in the hydrolysates were determined by HPLC using the method described above and converted to glucose and xylose yields based on the glucan and xylan contents of the untreated bagasse.

#### Fermentation

Fermentation of the combined extracted juice from the integrated extraction and NaOH pretreatment and enzymatic hydrolysate was performed using Saccharomyces cerevisiae strain GLBRCY73 expressing xylose reductase (XYL1), xylitol dehydrogenase (XYL2), and xylulokinase (XYL3) from Scheffersomyces (Pichia) stipitis and evolved to grow on xylose as described in our prior work (Sato et al., 2014). Juice and hydrolysate were mixed in a 60:40 v/v ratio, which is equivalent to the relative abundance of these liquors from the procedure. A second liquor was generated by vacuum evaporation (Buchi Rotavapor R114) of the combined extraction liquor and hydrolysate to yield a sugar stream concentrated by ∼6-fold with 325 g/L of total mixed monosaccharides (primarily glucose, xylose, fructose, and sucrose). Yeast nitrogen base (YNB) and urea were added to sterile, mixed-sugar hydrolysates at concentrations of 1.67 and 2.27 g/L, respectively. Yeast seed cultures were prepared by inoculating 50 mL of YPD medium (10 g/L yeast extract, 20 g/L peptone and glucose) with the glycerol stock of GLBRCY73 and incubating for 24 h at 30◦C with 150 rpm orbital shaking. After 24 h, 10 mL of culture was transferred aseptically to 60 mL of the liquor/hydrolysate mixture in a shake flask, in duplicate. The flasks were covered with fermentation locks, sparged with N2, and incubated at 30◦C with orbital shaking at 150 rpm for 7 days. Samples were collected every 24 h to determine OD<sup>600</sup> by spectrophotometer and sugar and ethanol concentrations by HPLC (Agilent 1100 series) as described above.

# RESULTS AND DISCUSSION

#### Sorghum Composition

The composition of the TX08001 sorghum on a total dry mass basis is presented in **Table 1** and shows the content of both extractives as well as cell wall biopolymers. Prior research on sorghum stems showed that significant compositional and structural heterogeneity among cell and tissue types and stage of plant development (Li et al., 2018). The current analysis averages these diverse sub-cellular compositions and examines sugar content only during vegetative phase growth when stem sucrose levels are relatively low compared to levels that accumulate postanthesis (McKinley et al., 2016). At the growth stage analyzed, the combined structural and extractable carbohydrate contents account for 64% of the dry mass of the plant and that 18% of this sugar is water-extractable: primarily sucrose, glucose, fructose, and polymeric glucan (i.e., starch and mixed-linkage glucan). Notably, the water-extractable polymeric glucan represents 7.7% of the water-extractable carbohydrates. This composition can be contrasted to other graminaceous feedstocks such as corn stover and switchgrass that contain comparable or slightly less total carbohydrates on a dry mass basis, but which is overwhelmingly comprised of structural polysaccharides (Ong et al., 2016; Williams et al., 2017). The implication is that a significant fraction of the sorghum plant mass is in the form of readily assimilable carbohydrates for microbial conversion to biofuels, which would not require pretreatment and enzymatic hydrolysis for utilization, although requiring immediate utilization or stabilization following harvest. Another key result is that 57%

#### TABLE 1 | Composition of untreated sorghum on a total dry mass basis.


Values are the average of three process replicates. \*ND, not detected.

of the extractable sugars were monosaccharides (glucose and fructose), while only 33% was sucrose. This is consistent with prior studies showing that during the vegetative phase, sorghum stems accumulate high levels of glucose and fructose compared to sucrose (McKinley et al., 2018). Some sucrose may also be hydrolyzed/interconverted during processing, however, relative to sugar mills, preventing sucrose inversion and high sucrose purity are not critical for processes fermenting these sugars.

#### Integration Concepts for Counter-Current Carbohydrate Extraction and Cell Wall Deconstruction

Two potential concepts for integrating recovery and utilization of water-extractable sugars and plant cell wall-derived polysaccharides were evaluated (**Figure 1)**. The first concept uses separate unit operations for the extraction of sugars and subsequent deconstruction of the structural polysaccharides in the bagasse during pretreatment and enzymatic hydrolysis (**Figure 1A**). Utilizing counter-current extraction in this processing approach would yield a relatively clean, highconcentration of water-extractable sugars (and extractable starch) and, depending on the how the process is integrated, either separate or combined pretreatment liquors and hydrolysate sugar streams derived from the enzymatic hydrolysis of the pretreated bagasse. The water-insoluble hydrolysis residue would be primarily lignin with some unhydrolyzed carbohydrates that could be burned for process energy. The pretreatment liquors would contain a significant number of pretreatment-solubilized compounds including xylan/xylose, lignin (minimal for LHW), acetate, and degradation products derived from sugar and lignin depending on the pretreatment process.

The second concept integrates sugar extraction with mild alkali pretreatment in the same unit operation (**Figure 1B**). In this configuration, the process acts as a combined extraction process for soluble sugars and a leaching/lixiviation process

of the extracted cell wall biopolymers. This process would yield an extraction liquor that would contain, in addition to water-extractable sugars and starch, pretreatment-derived compounds such as Na+, lignin, acetate, and xylan, which would not be desirable in conventional sugar mill, but are not necessarily problematic for ethanol fermentation. A second stream would comprise the hydrolysate of cell wall polysaccharides (glucose and xylose) that would be relatively free from pretreatment-derived contaminants. If these approaches were performed at decentralized bioprocessing depots rather than at a centralized biorefinery, the extraction and pretreatment could be coupled with a subsequent drying and densification of the biomass (**Figure 1)**. This approach would enable the conversion of a high-moisture, low-bulk density, unstable (due to the degradable sugar content), seasonally available feedstock into a high-bulk density, stable, storable intermediate product that is amenable to further enzymatic deconstruction in a cellulosic biofuels process or, alternatively, may serve as a high-digestibility feed for ruminants or as a feedstock for anaerobic digestion as suggested in the prior literature (Bals and Dale, 2012).

Integration of the pretreatment with the soluble sugar extraction has the potential to yield a number of advantages relative to other processing configurations. First, by performing this approach in a counter-current manner with the sugar extraction primarily taking place in the initial stages and the pretreatment taking place on one or several of the later stages, the high concentrations of water-extractable carbohydrates are not subjected to the high pH values that would degrade reducing sugars (i.e., containing an anomeric carbon not involved in a glycosidic bond) such as glucose and fructose (De Bruijn et al., 1986). One advantage of this process is that three outcomes are achieved simultaneously: **(1)** the soluble sugars are extracted at high yields and concentrations, **(2)** the biomass is pretreated, and **(3)** the pretreated biomass is washed prior to enzymatic hydrolysis. Additionally, relative to a sequential extraction and pretreatment process (**Figure 1A**), the integrated process (**Figure 1B**) offers the potential for substantial savings in process water use as the water used for sugar extraction and pretreatment are the same and can be derived in part from high moisture biomass such that additional concentration of an extraction juice with a pretreatment liquor would not be necessary to achieve the same sugar concentrations.

A commercial sugarcane diffuser may be comprised of more than 10 extraction stages with typical residence times of 5 min per stage, for total material residence times in the diffuser ranging from 40–60 min at an extraction temperature of 80 to 85◦C (Buchanan, 1967; Love and Rein, 1980; Breward et al., 2012). Mild NaOH pretreatment may be performed in this temperature range with residence times of 1 h (Stoklosa and Hodge, 2012). The time required for the mild alkaline pretreatment is a combination of reaction rate and diffusion of alkali into the cell walls where diffusion has been proposed to be the rate-limiting step during the initial stages of delignification at low temperatures (Olm and Tistad, 1979). As the name implies, molecular diffusion of sucrose out of the biomass is the primary mechanism for mass transfer during extraction in a diffuser (Buchanan, 1968), and it is possible that the pretreatment may be complete after only 10–20 min. This could take place over multiple stages or stage hold-up times could be modified for longer times in the pretreatment section. Finally, in addition to the single concept outlined, many variations can be envisioned, such as inclusion of enzymes during extraction to hydrolyze starch, xylans, or mixed-linkage glucans (i.e., βglucans) or further integrate with enzymatic hydrolysis of cell wall polysaccharides in stages following the pretreatment.

# Separate Carbohydrate Extraction and Deconstruction

Counter-current extraction of water-extractable carbohydrates from the sorghum was performed in a laboratory approximation of a diffuser extraction process using 5 stages (**Figures 1A**, **2C**). The sugar concentration profiles using this extraction approach demonstrate that relatively high sugar concentrations (i.e., 80.1 g/L sucrose, glucose, and fructose) can be achieved (**Figure 2B**). It should be noted that 7.7% of the extractable carbohydrates were polymeric glucan (i.e., starch) that was not quantified in this set of studies. The mass of extractable sugars estimated to remain entrained in the bagasse in stream S<sup>5</sup> was >3% of the mass of the extractable sugars entering the process in stream S0. Thus, the calculated extraction efficiency of this process is high (i.e., 97%) and is comparable to that of commercial diffuser systems (Rein, 1995). One factor that contributes the high sugar concentrations in the extraction liquor is that dried sorghum at a moisture of ∼7% was used in this work, resulting in significant sorption of water in Stage 1 and a low volume of extraction juice recovery in stream L<sup>0</sup> (13.5 mL) relative to the volumes of juice in all other streams (43.5–49.5 mL for streams L1-L4, and 50 mL of imbibition water in stream L5). The estimated water content of the biomass for each stage following extraction (% of total mass) was found to be relatively consistent and ranged from 89% on the first stage of the extraction (i.e., Stage 1) to a maximum of 91% on the last stage (i.e., Stage 5). This can be contrasted to commercial diffuser processes for sucrose extraction from sugarcane, where the biomass is fed at 70% moisture (rather than 7%), reaches a moisture content of ∼85% during diffusion process (comparable to the results in the present study), leaves diffuser, and is dewatered to ∼30% moisture (Breward et al., 2012). The low moisture content of the entering sorghum (S0) that results in the substantial sorption of juice from stream L<sup>1</sup> is why there is the increase in sugar mass (but not concentration) between streams L<sup>0</sup> and L1. As the moisture content of the bagasse leaving the last stage (i.e., stream S5) in our work is 90%, pressing out more juice would obviously result in even high sugar extraction yields. It should also be noted that based on mass balances, the extraction process had not reached steadystate and consequently the estimated sugar extraction yields (**Figure 2B**) are higher than the theoretical yields based on the original composition (**Table 1**).

For biological deconstruction and conversion of plant cell wall polysaccharides, pretreatments are a necessary step to improve the accessibility of these polysaccharides to hydrolytic enzymes for the enzymatic generation of cellulosic sugars (Ong et al., 2014). While a wide range and combination of solvents, pH, and temperatures can and have been used in the past to effectively pretreat biomass, any chemical inputs into the process can become a liability. The pretreatment chemicals must be either minimized or recovered with high efficiency to minimize process costs, integration of pretreatment chemicals/solvents with downstream processes can be challenging, and any inorganics added during pretreatment may prevent significant recycling of process water.

Hydrothermal pretreatments have economic, technological, and environmental advantages in that, compared to all other pretreatments, these require no chemical inputs other than water or steam. These technologies have been extensively researched as pretreatments for the biochemical production of biofuels from bioenergy grasses and agricultural residues (Mosier et al., 2005; Ruiz et al., 2013, 2017). Hydrothermal pretreatments are typically performed at temperatures between 150 and 220◦C with either liquid hot water (LHW, also called autohydrolysis), or steam, depending on whether the pressure is above or below the vapor pressure of water at operating temperature. One of the main outcomes of hydrothermal pretreatments is lignin melting and redistribution throughout the cell wall and the solubilization of a portion of the hemicellulose fraction as oligomers and acetic acid (Ong et al., 2014). Alkaline pretreatments are another class of promising pretreatments for the liberation of cell wall polysaccharides. Soda pulping of graminaceous feedstocks such as de-pithed sugarcane is practiced commercially, NaOH pretreatments are known to be effective at delignifying graminaceous biomass at relatively mild (i.e., <130◦C) conditions, and these processes have been adapted to facilitate high sugar yields during enzymatic hydrolysis (Karp et al., 2014; Liu et al., 2014; Li et al., 2018). An additional advantage of mild alkaline pretreatments is that pretreatmentgenerated inhibitors are benign enough that a detoxification is not required to facilitate fermentation (Sato et al., 2014) as in other pretreatments such as dilute acid pretreatment (Jönsson et al., 2013). Furthermore, these pretreatments are effective at temperatures below 100◦C, which can overcome the operational challenges of feeding biomass into a pressurized reactor. Alkali also lends itself to integration with diffuser technologies as lime

is typically used in these processes for pH control. However, as a disadvantage, the use of NaOH during pretreatment necessitates the use of strategies such as alkali recovery to prevent accumulation of inorganics in recycled process water streams.

In the present work, the response of the extracted sorghum bagasse to deconstruction was assessed for separate extraction and either LHW or NaOH pretreatment (**Figure 1A)**. For this, extractives-free sorghum bagasse was subjected to pretreatment at two levels each for LHW (1 h and either 120◦C or 160◦C) and NaOH pretreatment (1 h, 80◦C, and either 0.06 or 0.10 g NaOH/g biomass) followed by enzymatic hydrolysis at an enzyme loading of 15 mg/g glucan for 7 days. As can be observed, the hydrolysis yields for the more severe conditions for both the NaOH and LHW pretreatments were higher than the lower severity conditions, while the NaOH pretreatment at an alkali loading of 0.10 g/g resulted in the highest glucose hydrolysis yields (**Figure 3A**). These hydrolysis yields are within the range identified in our prior work with TX08001 sorghum (Li et al., 2018), where mild NaOH pretreatment of sorghum fractionated by tissue type could result in glucose hydrolysis yields ranging from 53% to the theoretical maximum. It is well-understood that the general mechanisms of reducing recalcitrance in alkaline pretreatments is by delignification as well as minor xylan solubilization (Ong et al., 2014) and not surprisingly, the levels of delignification were highest for the NaOH pretreatment and increased with increasing severity (**Figure 3B**). In contrast, LHW pretreatment decreases plant cell wall recalcitrance by hydrolyzing xylan and melting and redistributing lignin throughout the cell wall (Ong et al., 2014) and results for mass loss (**Figure 3C**) and lignin removal (**Figure 3B**) agree with this mechanism whereby more mass is lost with increasing pretreatment severity although minimal lignin is removed.

## Counter-Current Integrated Extraction and Pretreatment

As a proof of concept, an integrated extraction and pretreatment was performed at one experimental condition corresponding to the configuration previously outlined (**Figures 1B**, **4C)** using 0.06 g NaOH/g extractives-free sorghum. For the integrated approach, sugar concentrations (**Figure 4A)** are slightly lower than in the extraction-only study (**Figure 2A**). Potential reasons for this discrepancy may be a combination of both the system not yet reaching steady-state and differences in water sorption between untreated and alkali-pretreated biomass. It is known that alkali-pretreated graminaceous biomass is capable of sorbing substantially more water than unpretreated biomass as demonstrated in our prior work with maize (Li et al., 2015) and sorghum (Li et al., 2018). Another important finding from this study is that the high pH (i.e., >12) reached during the

pretreatment (Stage 4 and leaving in stream L3) is not propagated through the other process stages. That is, stream L<sup>3</sup> is partially neutralized during Stage 3 by a combination of dilution with the entrained liquor in stream S<sup>2</sup> as well as consumption of alkali by reaction with the biomass in stream S<sup>2</sup> (e.g., through saponification of acetyl and hydroxycinnamoyl ester xylan and lignin). Furthermore, the high pH liquor entrained in the biomass entering Stage 5 (stream S4) is clearly partially neutralized by dilution with the fresh imbibition water in stream L5. The implications of this are (1) that neutralization with a strong mineral acid is not required for this process performed under these conditions and (2) as the majority of the soluble sugar extraction takes place in the first two stages, high concentrations of soluble sugars are never in contact with the high pH liquors. This is important as it is well-understood that under alkaline conditions, reducing sugars in the extraction liquors (i.e., glucose and fructose) are highly susceptible to enolization and subsequent degradation or aldol condensation (e.g., with phenolics) (Horváth et al., 2005), which could represent a significant loss of fermentable sugar. Finally, it was observed that the extraction liquor leaving the process (stream L0) dropped to a pH of 5.5 (**Figure 4B**). Fortuitously, this is the appropriate pH for fermentation by S. cerevisiae, indicating that no additional pH adjustment of this extraction liquor is necessary prior to fermentation.

## Fermentation of Combined Hydrolysate and Extracted Sugars

The sugar streams derived from the integrated extraction and deconstruction of bioenergy sorghum (**Figure 1B**) were next subjected to fermentation by engineered S. cerevisiae strain GLBRCY73. This strain was previously developed through a combination of rational metabolic engineering to incorporate a xylose fermentation pathway into a background strain demonstrated to exhibit superior growth in the presence of pretreatment-derived inhibitors and evolutionary engineering to improve the xylose utilization rate (Sato et al., 2014). Challenges in performing fermentation of lignocelluosic hydrolysates include hydrolysate toxicity, the need to co-ferment glucose and xylose, and the need to generate hydrolysates at high sugar concentrations to minimize separation costs and process water usage. Hydrolysate toxicity to fermentation depends on the pretreatment technology employed and even the feedstock used where inhibitors include organic acids, phenolics, furans and inorganics (Palmqvist and Hahn-Hagerdal, 2000; Luo et al., 2002). Major fermentation inhibitors in pretreatment liquors derived from the mild alkaline pretreatment of grasses include hydroxycinnamic acids (p-coumaric and ferulic acid), acetate, Na<sup>+</sup> as well as other unknown and poorly characterized and quantified extractives (Sato et al., 2014). As an example, prior research has clearly demonstrated that weak acids such as acetic acid are inhibitors of the rates of cell growth and ethanol fermentation and cell biomass yields in S. cerevisiae under anaerobic conditions (Narendranath et al., 2001). A key finding from prior work is that sugar hydrolysates combined with alkali pretreatment liquors are completely fermentable by yeast without detoxification (Liu et al., 2014; Sato et al., 2014).

In the present work, the sugar streams derived from the integrated pretreatment and sugar extraction were combined with the hydrolysate derived from the enzymatic hydrolysis of the extracted and pretreated sorghum bagasse. Combining extracted juice or syrup from sweet sorghum with lignocellulosic hydrolysate has the advantage of further diluting the pretreatment-derived inhibitors in the hydrolysate with the juice. Two sugar hydrolysates derived from these combined sugar streams were tested in this study. The first hydrolysate comprised the combined sugar streams at their original concentrations that were combined at an extraction juice to hydrolysate ratio of 60:40 (v/v), which dilutes both liquors to yield a total of 48.8 g/L mono- and disaccharides (not including starch). A second, more concentrated hydrolysate was generated by subjecting the first hydrolysate subjected to vacuum evaporation. Using this approach, a sugar syrup was produced with a more than 6-fold increase in concentration (i.e., 332 g/L total mono- and

disaccharides) relative to the first, more dilute hydrolysate and was employed to test the limits of fermentation in terms of ethanol tolerance. Importantly, fermentations were performed without detoxification directly on these hydrolysates with only nitrogen supplementation.

Fermentation kinetics for the two hydrolysates (i.e., the combined extraction filtrate plus the hydrolysate and the concentrated filtrate plus hydrolysate) (**Figure 5)** reveal several key findings. First, for the low-concentration combined hydrolysate (**Figure 5A**), all the sucrose, glucose, and fructose are utilized within the first 16 h, while ∼80% of the xylose was utilized after 7 days. This indicates that the inhibition of the fermentation by organic solubles as well as some sodium derived from the pretreatment in the combined juice/hydrolysate is minimal. This can be contrasted with fermentation of lignocellulose-derived hydrolysates utilizing other pretreatments such as dilute acid that typically require extensive detoxification or dilution (Jönsson et al., 2013) or organosolv pretreatments whereby the solvent must be removed from the biomass prior to biological conversion. Furthermore, a final ethanol titer of 21 g/L was obtained corresponding to a yield of 0.44 g ethanol generated per g sugar consumed, or ∼85% of the theoretical maximum for ethanol fermentation. This lower value for the metabolic product yield is consistent with our prior work that found yields slightly lower than typical yeast fermentations potentially due to higher glycerol yields in this strain (Liu et al., 2014; Sato et al., 2014).

Typical challenges for industrial "very high gravity" yeast fermentation of starch hydrolysates or sucrose include high initial osmotic stress for cells resulting in slow initial growth and inhibited growth and fermentation when high ethanol titers are reached later in the fermentation for batch fermentations (Häggström et al., 2014). Both these phenomena are apparent in the kinetic profile of the high-sugar hydrolysate (**Figure 5B**). It can be observed that yeast growth was notably slower than in the more dilute hydrolysate and 3 days was required for the cell density to reach the maximum level achieved in the low (i.e., an OD<sup>600</sup> of ∼6). This slower growth as well as the longer lag phase can presumably be attributed to the osmotic stress exerted by the high sugar concentrations as well as potentially the increased concentration of pretreatmentderived inhibitors. Sugar consumption was also significantly slowed and ∼15–25 g/L each of sucrose, fructose and glucose were still remaining after 7 days of fermentation, compared to complete consumption of these sugars in >18 h for the more dilute hydrolysate. These results for fermentation of the undetoxified high-sugar syrup presumably demonstrate the limits of ethanol tolerance for this strain and the final ethanol titer of 80 g/L was sufficient to halt growth and fermentation. The ethanol yield per consumed sugar is slightly lower than the yield for the low sugar concentration hydrolysate and may be attributed to the diversion of carbon in the substrate to glycerol, which is known to occur in S. cerevisiae to offset the inhibitory effects of high osmotic pressure (Nevoigt and Stahl, 1997).

# CONCLUSIONS

An innovative hybrid approach to sugar extraction and pretreatment of energy sorghum was investigated in this work to demonstrate the potential of integrating mild alkaline pretreatment with counter-current sugar extraction. Hybrid energy sorghum TX08001 was found to be comprised of 64% carbohydrates as a percentage of the total dry mass of the sorghum, of which 27% of this sugar is water-extractable sucrose, fructose, glucose, and unidentified glucan. Laboratory counter-current extraction approximating diffuser extraction used for sugarcane processing was demonstrated to be capable of extracting 97% of the water-extractable sugars in 5 extraction stages, yielding extractives-free bagasse and an extraction juice containing 80.1 g/L of sucrose, fructose, and glucose. Extractivesfree bagasse resulting from this process was susceptible to deconstruction by both NaOH and LHW pretreatments, with subsequent glucose hydrolysis yields ranging from 64.9% for the low severity LHW pretreatment (1 h at 120◦C) to 87.3% for the most severe NaOH pretreatment condition (1 h at 80◦C and 0.10 g NaOH/g biomass). It can be highlighted that these yields were achieved with the moderate enzyme of loadings of 15 mg protein/g glucan in the presence of all pretreatment-derived solubles. Not surprisingly, it was found that increasing NaOH loading during alkaline pretreatment improved lignin removal and hydrolysis yields and increasing temperature during LHW pretreatment improved hydrolysis yields.

Combined counter-current soluble sugar extraction and NaOH pretreatment at the mild conditions of 0.06 g NaOH/g extractives-free biomass (1 h at 80◦C) was next demonstrated. It was identified that, again, high sugar extraction yields and titers could be achieved while simultaneously pretreating the biomass and generating a pretreated, extractives-free biomass pulp that was susceptible to enzymatic hydrolysis. Interestingly, the high pH (>12) of the liquors extracted from the pretreatment stage of the process was completely neutralized over the subsequent extraction stages, such that the final liquor pH exiting the process was only 5.5 and that the high concentrations of water-extractable sugars were never subjected to alkaline pH at elevated temperatures, which would degrade any free reducing sugars in the extract. Furthermore, the pH of the extraction liquor is already optimal for fermentation and no additional acidification is necessary. Finally, the extraction/pretreatment liquor from the integrated process was combined with the enzymatic hydrolysate to yield a mixed sugar solution with a concentration of 48.8 g/L total fermentable sugars (sucrose, glucose, fructose, and xylose). This was found to be completely fermentable by a S. cerevisiae strain engineered for xylose fermentation. Following a more than 6-fold concentration of this sugar stream by vacuum evaporation, an ethanol titer of 80 g/L could be achieved, however a significant fraction (21%) of the sugars remained unfermented, presumably due to inhibition of yeast growth and fermentation due to ethanol toxicity, potentially in combination with pretreatment-derived inhibitors. Additional variations of this integrated approach for soluble sugar extraction and pretreatment can be envisioned and may offer potential for novel depot-scale processing configurations such as subsequent drying and densification of extracted, pretreated sorghum to convert an unstable, transiently available feedstock into a stable intermediate product that can be stored to provide a continuous supply to a centralized biorefinery.

# AUTHOR CONTRIBUTIONS

DH and DW planned the experiments, analyzed the results, and wrote the manuscript. JM provided critical insights and ideas into sorghum processing that were used in this study. RO performed the detailed compositional analysis of the unextracted sorghum. All authors read, edited, and approved the final draft of the manuscript.

# FUNDING

Research carried out in the laboratories of JM and RO was funded in part by the Great Lakes Bioenergy Research Center, U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research under Award Numbers DE-SC0018409 and DE-FC02-07ER64494. The authors thank Dr. Trey Sato (GLBRC) for provision of the yeast strain used in this work, Bill Rooney (Crop & Soil Sciences, Texas A&M) for providing

#### REFERENCES


and planting the energy sorghum hybrid TX08001, and Brian McKinley (Biochemistry and Biophysics, Texas A&M) for collecting the stem samples used for the analysis.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenrg. 2018.00133/full#supplementary-material


peroxide-pretreated biomass. Appl. Environ. Microbiol. 80, 540–554. doi: 10.1128/AEM.01885-13


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Williams, Ong, Mullet and Hodge. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.