# MICROBIAL HYDROGEN METABOLISM

EDITED BY : Chris Greening and Eric Boyd PUBLISHED IN : Frontiers in Microbiology

#### 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 978-2-88963-540-5 DOI 10.3389/978-2-88963-540-5

#### 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

# MICROBIAL HYDROGEN METABOLISM

Topic Editors: Chris Greening, Monash University, Australia Eric Boyd, Montana State University, United States

Citation: Greening, C., Boyd, E., eds. (2020). Microbial Hydrogen Metabolism. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-540-5

# Table of Contents


Nadya Dragomirova, Patricia Rothe, Stefan Schwoch, Stefanie Hartwig, Constanze Pinske and R. Gary Sawers


Ying Teng, Yongfeng Xu, Xiaomi Wang and Peter Christie


Carlo R. Carere, Ben McDonald, Hanna A. Peach, Chris Greening, Daniel J. Gapes, Christophe Collet and Matthew B. Stott

*135 Putative Iron-Sulfur Proteins are Required for Hydrogen Consumption and Enhance Survival of Mycobacteria*

Zahra F. Islam, Paul R. F. Cordero and Chris Greening

## Editorial: Microbial Hydrogen Metabolism

Chris Greening<sup>1</sup> \* and Eric Boyd<sup>2</sup> \*

*<sup>1</sup> School of Biological Sciences, Monash University, Clayton, VIC, Australia, <sup>2</sup> Department of Microbiology and Immunology, Montana State University, Bozeman, MT, United States*

Keywords: hydrogen, hydrogen metabolism, hydrogenase, fermentation, respiration

#### **Editorial on the Research Topic**

#### **Microbial Hydrogen Metabolism**

Among the most ancient and widespread metabolic traits of microbial life is the ability to interconvert molecular hydrogen (H2; Lane et al., 2010; Schwartz et al., 2013; Peters et al., 2015). Two classes of metalloenzymes, [FeFe]-hydrogenase and [NiFe]-hydrogenase, catalyze the reversible oxidation of H<sup>2</sup> to electrons and protons (Volbeda et al., 1995; Peters et al., 1998); a third class of hydrogenase, termed [Fe]-hydrogenase or Hmd, catalyzes the reduction of the substrate methenyltetrahydromethanopterin with H<sup>2</sup> (Shima et al., 2008). The three classes of enzyme differ structurally and are phylogenetically unrelated. As such, they represent profound examples of convergent evolution (Wu and Mandrand, 1993; Vignais and Billoud, 2007; Greening et al., 2016).

#### Edited by:

*Patricia Coutinho Dos Santos, Wake Forest University, United States*

#### Reviewed by:

*Shawn E. McGlynn, Tokyo Institute of Technology, Japan*

#### \*Correspondence:

*Chris Greening chris.greening@monash.edu Eric Boyd eboyd@montana.edu*

#### Specialty section:

*This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology*

> Received: *20 November 2019* Accepted: *13 January 2020* Published: *30 January 2020*

#### Citation:

*Greening C and Boyd E (2020) Editorial: Microbial Hydrogen Metabolism. Front. Microbiol. 11:56. doi: 10.3389/fmicb.2020.00056*

Approximately a third of sequenced microorganisms, spanning at least 70 microbial phyla, encode hydrogenases and are thus predicted to be capable of interconverting H<sup>2</sup> (Peters et al., 2015; Greening et al., 2016). The earliest evolving hydrogenase enzymes harbor a [NiFe] cofactor and these are thought to have functioned oxidatively (Boyd et al., 2014; Weiss et al., 2016), with [FeFe]-hydrogenases thought to have emerged more recently (Mulder et al., 2010). Both [NiFe] and [FeFe]-hydrogenases have since diversified to function in aerobic and anaerobic, heterotrophic, and autotrophic, and chemotrophic and phototrophic metabolic backgrounds (Kovács et al., 2005; Tamagnini et al., 2007; Thauer et al., 2010; Schwartz et al., 2013; Koch et al., 2014; Schuchmann and Muller, 2014; Pinske and Sawers, 2016). Many bacteria and archaea oxidize H<sup>2</sup> as a low potential electron donor, an activity typically (albeit not exclusively) attributed to various lineages of [NiFe]-hydrogenase enzymes. Various bacteria, archaea, and microbial eukaryotes also evolve H<sup>2</sup> as a diffusible end product during fermentative metabolism through the activity of [FeFe]- or [NiFe]-hydrogenases (Horner et al., 2000; Kim and Kim, 2011; Marreiros et al., 2013; Schwartz et al., 2013; Pinske and Sawers, 2016). In many organisms, the ability to metabolize H<sup>2</sup> is a facultative trait that is regulated through the expression and maturation of hydrogenases (Schwartz et al., 2013; Greening and Cook, 2014). In such taxa, H<sup>2</sup> represents a substrate that organisms utilize to supplement their energy metabolism, thereby allowing for an expansion of their niche space in ecosystems where other sources of reductant are low or variable in supply (e.g., Amenabar et al., 2018).

The implications of H<sup>2</sup> in ecosystem level processes is increasingly being realized in both environmental and biomedical settings. A wide range of ecosystems have now been described where H<sup>2</sup> cycling supports the bulk of primary production and where it forms the basis by which species interact, leading to ecologically structured communities. Much of the research on H<sup>2</sup> metabolism to date has focused on ecosystems where H<sup>2</sup> is present at elevated concentrations due to biological

**4**

activity (e.g., anoxic sediments, gastrointestinal tracts; Sørensen et al., 1981; Wolf et al., 2016; Greening et al., 2019; Kessler et al., 2019) or geological activity (e.g., hydrothermal vents, subsurface systems; Petersen et al., 2011; Brazelton et al., 2012; Telling et al., 2015; Dong et al., 2019; Lindsay et al., 2019). More recently, it has been recognized that atmospheric H<sup>2</sup> can serve as source of reductant for aerobic soil microorganisms and that this can influence the composition of the atmosphere (Conrad, 1996; Constant et al., 2010; Ji et al., 2017; Cordero et al., 2019). In parallel, medical microbiologists have shown that H<sup>2</sup> metabolism is critical for the virulence of numerous pathogens, including Helicobacter, Clostridia, and Enterobacteriaceae (Kaji et al., 1999; Olson and Maier, 2002; Maier et al., 2004, 2013).

This special issue, featuring 10 articles from 46 different authors, explores microbial H<sup>2</sup> metabolism from the molecular to the ecosystem scale. In the area of anaerobic metabolism, there are articles exploring the metabolism of H2-metabolizing bacteria capable of sulfate reduction, acetogenesis, halorespiration, and fermentation. Two articles investigate H<sup>2</sup> oxidation in sulfatereducing bacteria using the model system Desulfovibrio vulgaris (Fauque et al., 1988; Caffrey et al., 2007). Smith et al. present a mathematical model of the growth and metabolism of this bacterium, whereas Löffler et al. investigate the kinetic isotope fractionation associated with its H<sup>2</sup> oxidation activity. A comprehensive review led by Schuchmann et al. covers recent advances in understanding clostridial H<sup>2</sup> metabolism; it details the discovery and characterization of multimeric electron-bifurcating [FeFe]-hydrogenases, including those associated with formate dehydrogenases (Schut and Adams, 2009; Schuchmann and Müller, 2012, 2013; Buckel and Thauer, 2018). Another article led by Dragomirova et al. focuses on heterologous expression of a [NiFe]-hydrogenase from dehalogenating Chloroflexi (Kublik et al., 2016; Hartwig et al., 2017), reporting another unexpected association with formate dehydrogenase activity. Pinske explores a third type of formate dehydrogenase-linked hydrogenase, namely the classical formate hydrogenlyase complex of Enterobacteriaceae (McDowall et al., 2014), and its association with two novel iron-sulfur proteins.

Several articles also investigate aerobic H<sup>2</sup> metabolism. Islam et al. report two other novel iron-sulfur proteins in mycobacteria, demonstrating that they are essential for the activity of the two high-affinity hydrogenases described in this

#### REFERENCES


lineage (Greening et al., 2014). Carere et al. meanwhile, build on the recent discovery that verrucomicrobial methanotrophs are facultative mixotrophs (Carere et al., 2017; Mohammadi et al., 2017) by showing resource allocation of Methylacidiphilum varies depending on H<sup>2</sup> availability. Three articles also explore H<sup>2</sup> metabolism at the ecosystem level. Adam and Perner explore the diversity of aerobic and anaerobic H<sup>2</sup> metabolism in deep-sea hydrothermal vent systems, whereas Meyer-Dombard et al. investigate the influence of H<sup>2</sup> on biogeochemical cycling in serpentinizing springs in the Philippines. Teng et al. review the previously underexplored area of H<sup>2</sup> metabolism in bioremediation, including in the reduction of organohalides, nitroaromatic compounds, and heavy metals (Chardin et al., 2003; Hong et al., 2008; Schubert et al., 2018).

In summary, this special Research Topic sheds light on the diverse role of H<sup>2</sup> in microbial metabolism and uncovers novel enzymes and pathways that mediate this process. This body of work highlights the intricate linkages between H<sup>2</sup> cycling and the cycling of various other compounds, including methane, formate, carbon dioxide, sulfate, and organohalides, among others. In turn, these findings pave way for future studies on the biochemistry, physiology, ecology, and industrial applications of microbial H<sup>2</sup> metabolism.

#### AUTHOR CONTRIBUTIONS

CG and EB drafted this editorial together and approve its submission.

#### FUNDING

This work was supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences under Award DE-SC0020246 (EB).

#### ACKNOWLEDGMENTS

We would like to acknowledge all authors, peer reviewers, guest editors, and the Frontiers team for their support in developing this Research Topic. The salary of CG is covered by an Australian Research Council DECRA Fellowship (DE170100310) and a National Health & Medical Research Council EL2 Fellowship (APP1178715).


complex in Dehalococcoides mccartyi strain CBDB 1 suggests a proteindependent respiratory electron transport chain obviating quinone involvement. Environ. Microbiol. 18, 3044–3056. doi: 10.1111/1462-2920.13200


**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 © 2020 Greening and Boyd. 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.

## The Ferredoxin-Like Proteins HydN and YsaA Enhance Redox Dye-Linked Activity of the Formate Dehydrogenase H Component of the Formate Hydrogenlyase Complex

#### Constanze Pinske\*

*Institute for Biology/Microbiology, Martin-Luther University Halle-Wittenberg, Halle, Germany*

#### *Edited by:*

*Chris Greening, Monash University, Australia*

#### *Reviewed by:*

*Robert Maier, University of Georgia, United States William Thomas Self, University of Central Florida, United States*

#### *\*Correspondence:*

*Constanze Pinske constanze.pinske@ mikrobiologie.uni-halle.de*

#### *Specialty section:*

*This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology*

> *Received: 29 March 2018 Accepted: 23 May 2018 Published: 11 June 2018*

#### *Citation:*

*Pinske C (2018) The Ferredoxin-Like Proteins HydN and YsaA Enhance Redox Dye-Linked Activity of the Formate Dehydrogenase H Component of the Formate Hydrogenlyase Complex. Front. Microbiol. 9:1238. doi: 10.3389/fmicb.2018.01238* Formate dehydrogenase H (FDH-H) and [NiFe]-hydrogenase 3 (Hyd-3) form the catalytic components of the hydrogen-producing formate hydrogenlyase (FHL) complex, which disproportionates formate to H<sup>2</sup> and CO<sup>2</sup> during mixed acid fermentation in enterobacteria. FHL comprises minimally seven proteins and little is understood about how this complex is assembled. Early studies identified a ferredoxin-like protein, HydN, as being involved in FDH-H assembly into the FHL complex. In order to understand how FDH-H and its small subunit HycB, which is also a ferredoxin-like protein, attach to the FHL complex, the possible roles of HydN and its paralogue, YsaA, in FHL complex stability and assembly were investigated. Deletion of the *hycB* gene reduced redox dye-mediated FDH-H activity to approximately 10%, abolished FHL-dependent H2-production, and reduced Hyd-3 activity. These data are consistent with HycB being an essential electron transfer component of the FHL complex. The FDH-H activity of the *hydN* and the *ysaA* deletion strains was reduced to 59 and 57% of the parental, while the double deletion reduced activity of FDH-H to 28% and the triple deletion with *hycB* to 1%. Remarkably, and in contrast to the *hycB* deletion, the absence of HydN and YsaA was without significant effect on FHL-dependent H2-production or total Hyd-3 activity; FDH-H protein levels were also unaltered. This is the first description of a phenotype for the *E. coli ysaA* deletion strain and identifies it as a novel factor required for optimal redox dye-linked FDH-H activity. A *ysaA* deletion strain could be complemented for FDH-H activity by *hydN* and *ysaA*, but the *hydN* deletion strain could not be complemented. Introduction of these plasmids did not affect H<sup>2</sup> production. Bacterial two-hybrid interactions showed that YsaA, HydN, and HycB interact with each other and with the FDH-H protein. Further novel anaerobic cross-interactions of 10 ferredoxin-like proteins in *E. coli* were also discovered and described. Together, these data indicate that FDH-H activity measured with the redox dye benzyl viologen is the sum of the FDH-H protein interacting with three independent small subunits and suggest that FDH-H can associate with different redox-protein complexes in the anaerobic cell to supply electrons from formate oxidation.

Keywords: [NiFe]-hydrogenase, formate hydrogenlyase, YsaA, HydN, formate dehydrogenase H, FDH-H, ferredoxin, FeS-cluster proteins

#### INTRODUCTION

Ferredoxins are small proteins containing non-heme iron as iron-sulfur (FeS) clusters and they serve as electron carriers within the cell (Bruschi and Guerlesquin, 1988; Beinert, 1990). The group of ferredoxin proteins is ubiquitously distributed, very diverse and different classes can be distinguished based on the FeS cluster type (Bruschi and Guerlesquin, 1988; Beinert et al., 1997). They have typical cysteine motifs in common that co-ordinate the FeS clusters. In addition to the ferredoxin protein itself, Fdx, which is involved in FeS cluster synthesis, ferredoxin-like proteins often have functions as small or β-subunits of modular respiratory complexes like the E. coli proteins NarH of the nitrate reductase, HybA of the H2-oxidizing hydrogenase 2, FdoH and FdnH of the formate dehydrogenases O and N, respectively, NrfC of the periplasmic nitrite reductase, HycB and HycF of the H2-producing formate hydrogenlyase (FHL) complex and HyfA of the FHL-homologous complex, FHL-2. These proteins are predicted to have a similar core-fold with alpha-antiparallel beta sandwiches to hold the FeS-clusters (Pfam domain Fer4). In addition to the above, a number of other ferredoxin-like proteins are known in E. coli. Occasionally, these proteins are required for full activity of a particular respiratory enzyme, but are not essential components of the final enzyme. Among those are the NapF, G and H proteins, which are required for the full activity of the periplasmic nitrate reductase, and HydN, which is required for full formate dehydrogenase H activity (FDH-H) (Maier et al., 1996; Brondijk et al., 2002; Nilavongse et al., 2006). Other ferredoxin-like proteins like AegA and YsaA are not located within or close to an operon encoding an enzyme with predicted oxidoreductase activity and no phenotype has yet been described for mutants lacking these genes (Cavicchioli et al., 1996).

The FHL complex in E. coli is the main route of H<sup>2</sup> production under fermentative growth conditions (Pinske and Sawers, 2016). The complex comprises a bis-molybdopterin guanine dinucleotide (Mo-bis-PGD)-dependent FDH-H that oxidizes formate to electrons and CO2. FDH-H is physically linked to the [NiFe]-hydrogenase protein HycE by 3 iron-sulfur (FeS) proteins. These are the FDH-H small subunit HycB, which has a ferredoxin-like fold and carries 4 FeS clusters, the ferredoxin-type protein HycF with 2 predicted FeS clusters, and the hydrogenase small subunit, HycG with a single FeS cluster. These five subunits are attached to the cytoplasmic side of the membrane by the HycC and HycD proteins. Notably, the fdhF gene, which encodes FDH-H, is located separately on the chromosome from the FHLencoding hyc-operon, but it belongs to the same regulon (Sawers, 2005; Pinske and Sawers, 2016). Furthermore, it was established that FDH-H is the most loosely attached protein component of the FHL complex (McDowall et al., 2015). In addition to the Mobis-PGD cofactor, the FDH-H protein contains a selenocysteine and a FeS cluster, the latter of which requires insertion prior to attachment of FDH-H to the FHL complex (Boyington et al., 1997). A further function for FDH-H has been proposed as part of an alternative FHL-2 complex, encoded by the hyfoperon and expressed under different growth conditions in E. coli (Andrews et al., 1997; Trchounian et al., 2012). While FDH-H activity is associated with fermentative growth, the other formate dehydrogenases in E. coli, FDH-N and FDH-O, are associated with respiratory formate oxidation. FDH-N is active during nitrate respiration while FDH-O is active in the presence of both oxygen and nitrate (Sawers et al., 1991).

It has been suggested that HydN forms an additional pool of small subunits for FDH-H when it is not incorporated into the FHL complex (Sargent, 2016). Due to a high similarity with YfrA from Proteus vulgaris, the probable FeS subunit of fumarate reductase, it was further suggested that HydN could be involved in the electron transfer from formate to fumarate. However, a 1hydN strain was investigated for its ability to transfer electrons from formate to fumarate, and proved to be unimpaired (Maier, 1997). Notably, the hydN gene is in an operon with the hypF gene (Maier et al., 1996) and HypF is one of the universal maturases that assemble the [NiFe]-cofactor. Due to its co-expression with hypF, the product of hydN was also suggested to be involved in H2-metabolism of the cell. Nevertheless, the co-occurrence of hydN and hypF is negligible in other organisms, while on the other hand hydN scores highly in its co-occurrence with hycB (0.778) and with fdhF (0.572) (Szklarczyk et al., 2015), suggesting a tight functional linkage.

HydN is predicted to harbor four [4Fe-4S] clusters and resembles a formate dehydrogenase small subunit similar to HycB of the FHL complex with which it shares 39% amino acid identity (52% similarity). Although both proteins share the same ferredoxin-like fold, suggesting a function in electron transfer, they cannot functionally replace one another (Maier, 1997). Interestingly, the hydN gene forms a transcriptional unit with the fdhF gene in the opportunistic pathogen Serratia liquefaciens. This FDH-H is similar to the E. coli protein, but it harbors a cysteine instead of the catalytic selenocysteine in its active site. A respective E. coli Cys variant is 20-fold less active than the SeCys protein (Pinske and Sawers, 2016). Nevertheless, fermentative gas production from glucose has been observed in S. liquefaciens (Brenner et al., 2006), but it was not yet identified whether this gas is CO<sup>2</sup> or H2.

Due to the weak phenotype of the hydN deletion strain (Maier et al., 1996), a new homology search was conducted to identify other possible homologs, and revealed the ferredoxintype protein YsaA (synonym YiaI) in E. coli, which shares 62% amino acid identity (72% similarity) with HydN and is therefore even more closely related to HydN than HycB. The ysaA gene co-occurrence with fdhF is 0.545 and thus in the same range as the hydN-fdhF co-occurrence (Szklarczyk et al., 2015). Therefore, the presence of YsaA might exhibit functional redundancy with HydN. Nothing is known about the ysaA gene and it is not located near any hydrogenase-associated genes. However, an interaction with NuoE, a protein of the closely to FHL related respiratory Complex I, which is required for the association of the diaphorase subunit, and with RclA (synonym YkgC) has been reported (Arifuzzaman et al., 2006). Furthermore, expression of the gene is 2.7 fold up-regulated under anoxic conditions (Kang

**Abbreviations:** FHL, formate hydrogenlyase; Mo-bis-PGD, molybdopterin guanine dinucleotide; BV, benzyl viologen; FDH-H, formate dehydrogenase H; MU, Miller Units.

et al., 2005). Therefore, we wanted to investigate the role of YsaA in H<sup>2</sup> metabolism and to establish the interaction network of these ferredoxin-like proteins based on the hypothesis described above.

#### MATERIALS AND METHODS

#### Growth Conditions and Strain Construction

Strains and plasmids used in this study are listed in **Table 1**. Strains were routinely grown in liquid LB medium in a shaking incubator or on LB agar plates. For determination of enzyme activities and bacterial two hybrid interactions, the bacteria were grown in buffered TGYEP medium supplemented with glucose: 1% (w/v) trypton, 0.5% (w/v) yeast extract, 0.8% (w/v) glucose, 0.1 M potassium buffer, pH 6.5; according to Begg et al. (1977). The medium was supplemented with 100 µg ml−<sup>1</sup> ampicillin to ensure plasmid maintenance. Furthermore, when required, chloramphenicol was used at 12.5 µg ml−<sup>1</sup> and kanamycin at 50 µg ml−<sup>1</sup> . The strain DHD-N was used as 1hydN mutant and was previously described (Maier et al., 1996). The ysaA deletion was constructed in BW25113 transformed with the lambda red recombinase plasmid pKD46 according to Datsenko and Wanner (2000) by using the oligonucleotides ysaA\_5'KO 5 ′ -CTCTGGCACTCTGCTGTTTTAGTGCAA AGGAGTGATCATG CCATGGTCCATATGAATATCCTCC-3′ and ysaA\_3′KO 5′ -CGCACTGTTCCGGCGTTGAGAAAC GCCGGAAAACGTTTCA GCGATTGTGTAGGCTGGA GCT-3′ to amplify the cat gene from pKD3. The mutation was subsequently moved to MC4100, DHD-N and JW2694 (1hycB) by P1vir phage transduction (Miller, 1972). The resistance cassette was eliminated using the pCP20 plasmid as described (Cherepanov and Wackernagel, 1995). The deletion of aegA was introduced by the method of Hamilton et al. (1989) by cloning the upstream region as KpnI/BamHI fragment with the oligonucleotides DaegA1\_KpnI 5′ -GCGGGTACCGCCTGA TACCACGGCAAATC-3′ and DaegA2\_BamHI 5′ -GCGGGA TCC CATAATAAAACGATTCATAAC-3′ and the downstream region as BamHI/HindIII fragment with the oligonucleotides DaegA3\_BamHI 5′ -GCGGGATCCCAGTCAAATCTCACT GATAG-3′ and DaegA4\_HindIII 5′ -GCGAAGCTTCGCCGG TTTTGATCATCTCC-3′ into pMAK705 and recombining with the desired target strain as described. Introduction of the 1hycB deletion in DHD-N backgrounds or vice versa has been done according to Datsenko and Wanner (2000) by introducing the lambda red recombinase on the pKD46 plasmid, growing competent cells and inducing them with 10 mM Larabinose before electroporation of a PCR fragment containing a kanamycin resistance cassette and the upstream and downstream regions of the gene. These PCR fragments were obtained after amplification of the corresponding regions from strain JW2694 (1hycB) or JW2683 (1hydN). All clones were verified using colony PCR.

#### Cloning of *ysaA* and *hydN*

Cloning of ysaA gene in the pJET1.2 cloning vector (commercially available from Thermo Fisher Scientific) was done by using the T18-ysaAFW\_HindIII 5′ -GCGAAG TABLE 1 | Strains and plasmids.


*<sup>a</sup>Further plasmids from the bacterial two hybrid system that were constructed here can be found in Table S2.*

CTTGATGAACCGGTTTATTATTGCG-3′ and T18C/T25 ysaARW\_KpnI 5′ -GCGGGTACCTTATCAAACAGGCTG CTGCCGTAGC-3′ oligonucleotides for amplification from chromosomal DNA of MC4100 by the Q5-DNA polymerase (NEB) according to the manufacturer's instructions. The hydN gene was cloned into pBluescriptSK(+) by using the oligonucleotides HydN\_FW\_BamHI 5′ -CGCGGATCCATG AACCGTTTCATCATTGC-3′ and HydN\_RW\_EcoRI 5′ -CGC GAATTCTTAGAACATCAGCGCCGT-3′ , digestion of both vector and PCR product with BamHI and EcoRI and subsequent ligation. All cloning products were verified by sequencing.

#### Bacterial Two Hybrid Interactions

The bacterial two-hybrid system (Karimova et al., 1998) was used to clone constructs by amplifying the respective gene fragment from chromosomal DNA of MC4100 and digesting both the pUT18 or the pT25 target vectors and the insert with the same restriction enzymes before ligation. The bacterial two hybrid vectors were constructed to yield functional in-frame fusion proteins and all inserts were verified by sequencing. The oligonucleotides used are listed in Table S1, where the restriction sites are given as part of the oligonucleotide name. Generally, the entire orf was amplified except for the start codons (T25 fusions) or stop codons (T18 fusions) and all newly generated constructs are listed in Table S2. Plasmids of pT25 origin were transformed together with a pUT18 plasmid into BTH101 and grown as anaerobic culture in TGYEP, pH 6.5 at 30◦C for 16h containing both ampicillin and chloramphenicol as antibiotics. Determination and calculation of β-galactosidase activity was done according to Miller (1972). Each experiment was performed three times independently, and the activity for each sample was determined in triplicate. Alternatively, a volume of 5 µl of the culture was spotted on McConkey plates containing 0.5% (w/v) maltose, the antibiotics ampicillin and chloramphenicol. The plates were then aerobically incubated at 30◦C for 16 h.

#### Enzymatic Assays

For determination of protein activities, the cells were grown in TGYEP medium, pH 6.5 as standing liquid cultures in 50 ml reaction tubes at 30◦C for 16 h. For induction of NAR and FDH-N enzyme synthesis, sodium nitrate was added to the anaerobic cultures to a final concentration of 1% (w/v). The cells were harvested by centrifugation, sonicated (20 W, 0.5 s pulses, 3 min) and the extracts used directly. Activities of the soluble formate dehydrogenase and total hydrogenase were measured as formatedependent benzyl viologen (BV) reduction (FDH-H) and H2 dependent BV reduction (hydrogenase), respectively at 600 nm as described in Pinske et al. (2011a). Detection of NAR and FDH-N activities by in-gel activity staining was done as described (Pinske and Sawers, 2012). Immunoblotting against FDH-H (1:3000) polypeptides was done after denaturing gel electrophoresis as described (Pinske et al., 2011a).

#### GC Analysis of Gases in Culture Headspace

For assessing the activity of the intact FHL complex, the H<sup>2</sup> content of the gas headspace was quantified. The H<sup>2</sup> content of the 10 ml gas phase of an overnight culture was measured by sampling 200 µl in a GC-2010. The system was equipped with a packed column (Shin Carbon Micropacked column ST80/100). The carrier gas was N<sup>2</sup> with a flow rate of 13.9 ml min−<sup>1</sup> , the injector was kept at 140◦C, the column at 110◦C and the TCD detector at 150◦C and 40 mA. Quantification was done with a calibration curve of known amounts of H2.

#### Bioinformatics

The unrooted tree was created based on an alignment conducted within Uniprot (Clustal Omega) and visualized with iTOL (Sievers and Higgins, 2013; Letunic and Bork, 2016).

### RESULTS AND DISCUSSION

### HydN has Paralogues in *E. coli*

A phylogenetic tree shows that YsaA and HydN are most closely related among all known ferredoxin-like proteins in E. coli (**Figure 1**). Together with the two proteins HycB and HyfA, the small subunits of the FHL and FHL-2 complexes, respectively, and the AegA, YgfS, and YgfT proteins of unknown function, they form a group of paralogous proteins (see Figure S1 for alignment). In contrast, the HycF proteins represents the FeS-protein that provides an electron relay between the small subunit HycB of FDH-H and Hyd-3; however, HycF is more distantly related to HydN and YsaA. It is located in the tree together with proteins that are known to be required for full activity of respiratory complexes like NapF and NapG, but also with the E. coli ferredoxin protein Fer and the pyruvate formate-lyase activator PflA, which is required to introduce a radical into pyruvate formate-lyase (Knappe et al., 1969). The more classical β-subunits of multi-subunit respiratory enzymes like hydrogenase 2 (HybA), formate dehydrogenases N and O (subunits FdnH and FdoH, respectively), and nitrate reductases (NarH and NarY) cluster separately. Proteins of unknown function can be found clustered together both with the βsubunits and with the HydN paralogues. Due to the characteristic architecture of these proteins and the incorporation of one or more FeS clusters, it seems probable that these proteins also have a role in electron transfer within the cell.

#### The 1*ysaA* Mutant has a Similar Phenotype to the 1*hydN* Mutant

Previous research from the Böck group had identified a hydN deletion strain, that resulted in reduced FDH-H activity to 38% compared to the parental, while total hydrogenase activity remained at approximately 80% (Maier et al., 1996). These findings could be reproduced in this study using both the same strain (DHD-N) used previously (Maier et al., 1996), as well as the 1hydN strain (JW2683) of the Keio collection (**Table 2**). The FDH-H activity was reduced to 59 and 34%, while total hydrogenase activity was reduced only to 74% in the 1hydN strains DHD-N and JW2683, respectively. Having an influence on the activity of an enzyme, but not completely abolishing it suggested that HydN exhibits some redundancy with another unknown protein. Therefore, the gene coding for the most similar protein YsaA was deleted by the Datsenko and Wanner method (Datsenko and Wanner, 2000) and the deletion moved into the MC4100 and DHD-N backgrounds, resulting in strains CPH010 and CPH011 (see **Table 1**). The strain CPH010 phenocopied the

1hydN strains and had reduced FDH-H activity to 57% of the parental while total hydrogenase activity was comparable with the MC4100 parental (**Table 2**). Remarkably, the double deletion strain had a reduction in FDH-H activity to 28% demonstrating that the effect of the hydN and ysaA deletion was additive.

Only 10% of FDH-H activity measured anaerobically as formate-dependent BV reduction remained detectable when hycB was deleted (**Table 2**). The formate-dependent BV reduction was not detectable in an fdhF deletion strain and therefore highly specific for FDH-H [data not shown and (Pecher et al., 1983)]. This is in agreement with a previous finding that the HycB, HycE, HycF, and HycG are equally required for full FDH-H activity (Sauter et al., 1992). In contrast to the hydN and ysaA deletion strains, the hycB deletion disrupted electron flow from FDH-H to hydrogenase 3 and the strain retained only residual hydrogenase activity due to hydrogenases 1 and 2 activities, and thus had a total hydrogenase activity that was reduced by more than 90% (Pinske et al., 2011b). The strain consequently failed to produce H2. The effect of the hycB deletion on FDH-H activity was even more dramatic when the hycB and ysaA deletions were combined, which resulted in a residual 3% FDH-H activity (**Table 2**). The same could be observed when the hycB and hydN deletions were combined in strain CPH018, which showed 2% remaining FDH-H activity and total hydrogenase activity as low as in the 1hycB strain JW2694. Most strikingly, the combination of all three deletions in strain CPH019 resulted still in a detectable FDH-H activity of about 1% compared to the parental.

Electron transfer from formate to BV is thought to occur via the FeS cluster in the FDH-H polypeptide, which is located within 15 Å from the protein's surface (Boyington et al., 1997). The isolated enzyme without the small subunit was also shown to be electrochemically active in both directions on an electrode (Bassegoda et al., 2014) and in the BV assay (Axley et al., 1990). The findings in the present study, however, strongly suggest that in extracts FDH-H activity requires the presence of the ferredoxin-like proteins and in the absence of these only 1% of the FDH-H activity can be assayed directly from the enzyme.

FDH-H delivers the electrons from formate oxidation that are required for H<sup>2</sup> production by the FHL complex. Alternative electron donors other than formate are not known for the FHL complex, but synthetic protein-fusion experiments with Thermotoga maritima ferredoxin have shown that HycB can receive electrons from other sources like pyruvate:ferredoxin oxidoreductase to some extent (Lamont et al., 2017). When cultures are grown on glucose, the H<sup>2</sup> from the FHL complex accumulates in a closed growth vessel and can partially be re-oxidized by hydrogenase 1 and 2 in E. coli. However, in our experience the amount of accumulated H<sup>2</sup> also roughly corresponds to the activity of the FHL complex and therefore is a good indicator of FHL activity (Lindenstrauß et al., 2017). By sampling the gas headspace it became clear that the hydN deletion reduced the FHL activity to 86% of the wild-type level, while the ysaA hydN double deletion retained 91% H<sup>2</sup> production and the ysaA deletion alone had no influence on the amount of H<sup>2</sup> produced (**Table 2**). Thus, the reduced FDH-H activity as measured by BV reduction in the hydN and ysaA deletion strains does not correlate with FHL activity as quantified by H<sup>2</sup> production, perhaps suggesting that this reduced activity might be derived from FDH-H associated with another complex or complexes.

To test whether the effect of deleting the ysaA or hydN on FDH-H activity is specific, a further candidate AegA from the HydN-YsaA family of proteins was targeted and its corresponding gene aegA was deleted. The aegA gene was deleted alone and in combination with hydN, ysaA or hycB, but the effect on FDH-H activity was not greater than in the respective strains without aegA deletion (**Table 2**). Similarly, the aegA deletion had no influence on H<sup>2</sup> production and total hydrogenase activity. This indicates that FDH-H cannot transfer electrons from formate via AegA to the redox dye BV.

Previous experiments also showed that the complementation of the hydN deletion did not restore full FDH-H activity (Maier et al., 1996), which could be confirmed here regardless of whether strain DHD-N or strain JW2683 was used. The 1hydN 1ysaA double deletion strain showed the same behavior like the two 1hydN mutants after complementation with phydN. Somewhat surprisingly, restoration of almost parental FDH-H activity could be achieved when the 1ysaA deletion strain was transformed with phydN (**Table 2**), which further substantiates at least partial redundancy of function between these proteins. The 1ysaA mutant could also be complemented with pysaA to parental levels of FDH-H activity. Notably, however, introduction of the pysaA


TABLE 2 | Influence of YsaA and HydN on formate dehydrogenase H, hydrogenase, and H2-production activities.

*The strains indicated were grown in TGYEP, pH 6.5 and assayed as described in the Materials and Methods section. The values are given for three independent biological replicates with standard deviations. n.d., not determined.*

plasmid into the 1hydN strain caused a further reduction of FDH-H activity. This shows that although the effect of the hydN and ysaA deletions on FDH-H activity is additive, the functions of the gene products are not identical. The use of two independent 1hydN strains indicates that the lack of complementation is not strain-specific and therefore is unlikely due to additional mutations. The hycB deletion cannot be complemented for H<sup>2</sup> production by plasmid-encoded HydN (Maier, 1997).

Based on the observation that ysaA and hydN had little influence on total hydrogenase activity and H2-production, the addition in trans of these genes was not expected to cause significant differences in these activities. The complementation with either phydN or pysaA caused a slight reduction in total hydrogenase activity in all strains tested. Another interesting effect of the complementation was visible with the GC-headspace H2-quantification. While the addition of pysaA had no influence on the amount of H<sup>2</sup> produced by the cells in comparison to the respective plasmid-free strains, the addition of phydN increased the H<sup>2</sup> amount slightly in all strains.

#### Interaction Network of YsaA and HydN

In vivo protein interactions can be monitored by the bacterial two-hybrid system (Karimova et al., 1998). Generally, a T18 and a T25 domain of the adenylate cyclase is fused to the target proteins C-terminally or N-terminally, respectively, and any interaction can be quantified by measurement of β-galactosidase activity. Growth of the cells under the conditions where proteins under investigation are also normally synthesized ensures that a protein interaction that requires a further unknown interaction partner can be identified. Therefore, the reporter strain was grown under anoxic conditions with glucose, where proteins of the mixedacid fermentation are synthesized. These conditions show that empty pT25 and pUT18 plasmids yielded 217 ± 11 Miller units (MU) of activity while the zip positive control resulted in 3,619 ± 548 MU. An activity of more than 600 MU generally reflects a real interaction (Karimova et al., 1998). As additional control the empty pT25 and pT18 plasmids were co-transformed with each of the pT18 or pT25 protein fusions constructed here, respectively. They showed that the empty pT18 plasmid is not able to activate β-galactosidase in combination with any pT25 fusions. However, the empty pT25 plasmid gave a signal of more than 800 MU for the T18-HydN and T18-HyfA constructs while the other constructs showed interactions with less than the 600 MU threshold established here. Semi-aerobic interactions were additionally assessed qualitatively by spotting the colonies on McConkey-Maltose plates and showed that not all interactions identified anaerobically are detectable aerobically (**Figure 2**).

Initially, the interaction of FDH-H and its predicted small subunit HycB was determined. Because the N-terminus of this 80 kDa FDH-H protein (encoded by fdhF) is located at a distance of 20 Å from the FeS cluster (Boyington et al., 1997) it is therefore likely to be able to interact with its cognate small

subunit. Therefore, a T25-FdhF fusion, in which the entire FDH-H protein was fused to the C-terminus of the T25 fragment, was constructed. The interaction with the C-terminal fusion of HycB (fused to T18) showed an activity of 800 MU, which was also within the same order of magnitude as the interaction with T18-HydN (728 MU) and T18-YsaA (694 MU) with T25-FdhF (**Table 3**). No other positive interactions of the FdhF protein were detectable here (**Figure 3**). The T18-HydN fusion, however, showed also an interaction with the empty T25 plasmid in this order of magnitude (892 MU) and the interaction will need to be evaluated carefully by additional methods. Although, the C-terminal fusion between FDH-H and HycB was previously shown to be successful in transferring electrons into the FHL complex (McDowall et al., 2015), the T18-FdhF showed neither an interaction with HycB nor with the other tested protein fusions. This could indicate that HydN and YsaA only transiently interact with FdhF before it assembles into the FHL complex, as we could show recently for the HycH protein interaction with the hydrogenase 3 of the FHL complex on the pathway of assembly into the complex (Lindenstrauß et al., 2017). But instead of a temporal order of interactions, these results could also be interpreted to indicate that FDH-H interacts with other protein complexes.

The strong interaction between the N-terminal fusions of HydN, YsaA and the HycB proteins and their corresponding C-terminal fusions favors the latter hypothesis (**Table 3**). These range between 2,413 MU for the HycB self-interaction and 4,608 MU for the T25-YsaA × T18-HydN interaction. Interestingly, the AegA protein participates in this interaction network by showing strong interactions with itself, HycB, HydN, and YsaA. The close proximity of the second ferredoxin-like protein in the FHL complex HycF to HycB is probably the reason this protein interacts with the same partners as HycB, with the exception of FDH-H. It is striking that no interaction between the HycB homolog, HyfA, of the FHL-2 complex and the FDH-H can be detected here. Although a missing bacterial two hybrid


*aShown are the mean Miller Units (MU) from 3 independent biological samples measured in duplicates to triplicates with their respective standard deviations. Cells were grown anaerobically in TGYEP, pH 6.5 until stationary phase. The zip plasmids and empty plasmids served as positive and negative controls, respectively (Karimova et al., 1998). The positive control yielded an activity of 3,619* ± *548 MU and empty pUT18 and pT25 vectors as negative control had an activity of 217* ± *11 MU Strong interactions (*<sup>&</sup>gt; *1,000 MU) are highlighted by an orange color, weak interactions (600–1,000 MU) are shown with a yellow background.*

interaction does not necessarily reflect the absence of a true interaction, the similarity of the HycB and HyfA proteins and the detectable interaction of the former, is a strong indication and these initial findings do not support the proposed interaction (Bagramyan and Trchounian, 2003) between FDH-H with the HyfA subunit of the FHL-2 complex. It should be noted, however, that AegA and YsaA interact in only one orientation with HycB and HycF of the FHL complex. Thus, further fusion constructs need to be tested.

proteins with unknown function (red).

It is noteworthy that YsaA also showed an interaction with FdnH and NarH, the two β-subunits of nitrate-dependent formate dehydrogenase (FDH-N) and nitrate reductase (NAR), respectively. The genes encoding these enzymes are transcribed at only a low level (Walker and DeMoss, 1991; Li and Stewart, 1992), therefore these subunits of FDH-N and NAR complexes are likely to be present in the assay. The HydN and AegA proteins do not interact with these nitrate-dependent proteins, supporting the specificity for the YsaA interaction. The FdnH and NarH proteins furthermore show a self-interaction as well as a crossinteraction, which can be explained for FdnH because the enzyme complex FdnGHI exists as a trimer of trimers (Jormakka et al., 2003) and although primarily coupled by a Q-cycle the special colocalisation of FDH-N and NAR activities was shown to be highly organized and proximal to each other in the membrane (Alberge et al., 2015).

The previously identified interaction between YsaA with NuoE, a subunit of Complex I, using Tap-tag technology could also be verified here (Arifuzzaman et al., 2006). Further interactions of NuoE with itself and NarH, as well as of

TABLE 3 |

Beta-galactosidase

 activities of bacterial two hybrid interactions.

NuoE with the FHL subunits HycB and HycF were also detected here. However, the Fdx protein did not show any interaction with the protein fusions tested here, showing that its interaction with the proteins of the FeS-insertion machinery of the cell is highly specific (Tokumoto et al., 2002). This further indicates that it is not the general fold of the protein that is important for the interaction, but particular residues on the surface of the respective protein. Moreover, there are significant differences in length in a loop region between the third and the fourth Cys-rich (Cx2Cx2−8Cx3C) motif and the C-terminal region of the HycB, HydN, and YsaA proteins, the former lies at the expected interface with the interaction partner and this loop could also play a role in discrimination of the correct target. All identified interactions are summarized in **Figure 3**.

#### FDH-H Protein Pattern Remains Unaltered

Remarkably, although the activity of FDH-H varies among the 1hydN and 1ysaA strains, western blot analysis of the polypeptide showed that both the migration pattern and the protein amount appear similar in the ysaA, hydN, and ysaA hydN deletion mutants when compared to the protein in the MC4100 parental strain; the same is the case for strains carrying the plasmids pysaA or phydN (**Figure 4A**). Therefore, the transcription or translation of the fdhF gene is not altered in the mutants.

A native PAGE showing the activities of hydrogenases 1 and 2 (**Figure 4B**) reveals that the migration pattern and in-gel enzyme activities are similar regardless of the presence or absence of either HydN or YsaA. These hydrogenases harbor the same [NiFe]-cofactor as HycE, the hydrogenase of the FHL complex. Thus the influence is strictly limited to FDH-H and not the hydrogenases in general.

#### The *hydN* and *ysaA* Deletions Have no Influence on Other Mo-bis-PGD–Dependent Activities

In order to investigate whether the influence of the hydN and ysaA mutations on FDH-H is generally related to Mo-bis-PGD cofactor insertion, the cytoplasmic NAR and periplasmic FDH-N activities, which are both Mo-bis-PGD-dependent, were analyzed. **Figures 4C,D** show that the amount and migration patterns of the nitrate reductase and formate dehydrogenase N activities were comparable to those in the wild type strain, regardless of whether HydN and YsaA are present or not. Both activities are strictly dependent on Fnr for gene expression, hence the 1fnr mutant RM102 served as a negative control.

### CONCLUSIONS

Of the three formate dehydrogenases encoded in the E. coli genome, FDH-H is the only cytoplasmic enzyme. An early mutant study by Mandrand-Berthelot showed that differences in the requirements for maturation of the cytoplasmic and periplasmic formate dehydrogenases exist (Mandrand-Berthelot et al., 1988). One of the isolated mutants had a mutation in fdhE and was deficient in the respiratory formate dehydrogenases N and O only; this mutation was without effect on transcription (Stewart et al., 1991). The mutation in another mutant was located in fdhD and had an effect on all three formate dehydrogenases, because FdhD, as was shown recently (Thomé et al., 2012; Arnoux et al., 2015), is involved in donating a sulfur

ligand to the Mo-bis-PGD. It is conceivable that HydN and YsaA could contribute specifically to the cofactor activation of the FDH-H protein, since NAR and FDH-N activities are not impaired in the mutant; however, it would be expected that a more pronounced phenotype of the individual mutants would be observed, making this explanation unlikely.

Based on the observation of reduced, but not completely abolished FDH-H activity in the ysaA, hydN, hycB and the double and triple deletion mutants, it appears that full redox dye-reducing activity of FDH-H can only be measured when all three proteins are present simultaneously. Therefore, either all three proteins are involved as components of the FHL complex, although only HycB is essential (Sauter et al., 1992; **Table 2)** or HydN and YsaA could represent alternative βsubunits of the FDH-H enzyme, linking formate oxidation to other protein complexes. Neither HydN nor YsaA could be identified as components of the purified FHL complex (McDowall et al., 2014). If HydN and YsaA were transient components of FHL, then it would be anticipated that additional copies of hydN in trans should compete for FDH-H protein with HycB and reduce the H2-production by the FHL complex. However, H<sup>2</sup> production by the FHL complex remained essentially unaffected by all these mutations, which renders this hypothesis improbable. The data currently suggest, therefore, that HydN/YsaA provide alternative electron transfer routes for FDH-H to non-hydrogen producing electron acceptor complexes.

The phylogenetic analysis shows that AegA and YgfST are closely related to YsaA and HydN. Although AegA interacts with the ferredoxin-like proteins, the lack of interaction with FDH-H reflects an absence of influence on FDH-H activity. The AegA protein is a fusion of two domains; an N-terminal ferredoxin-like domain and a C-terminal glutamate synthase

#### REFERENCES


domain (Cavicchioli et al., 1996). Identifying its role within the network of ferredoxin-like proteins will require further experiments.

Generally, these interaction data have to be carefully evaluated because some proteins require interaction partners and if the physiological partner is not present, they seem to interact non-specifically with a number of other proteins. Therefore, future protein-protein interaction experiments will have to be performed in vitro. So far, HydN has proved recalcitrant to overproduction and alternative purification strategies will have to be established to allow full biochemical characterization of this protein family.

#### AUTHOR CONTRIBUTIONS

CP planned and performed the experiments. CP has drafted and written the manuscript.

#### ACKNOWLEDGMENTS

Financial support for the publication came from the Open Access Publication Fund of the Martin-Luther-University Halle-Wittenberg. The project was funded by the Deutsche Forschungsgemeinschaft project PI 1252/2. Special acknowledgement goes to R. Gary Sawers for critical and helpful discussions and to Ute Lindenstrauß for valuable technical assistance.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.01238/full#supplementary-material


nitrate reductase, in ubiquinol oxidation. Mol. Microbiol. 44, 245–255. doi: 10.1046/j.1365-2958.2002.02875.x


**Conflict of Interest Statement:** The author declares 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 Pinske. 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.

## Insights Into the Redox Sensitivity of Chloroflexi Hup-Hydrogenase Derived From Studies in Escherichia coli: Merits and Pitfalls of Heterologous [NiFe]-Hydrogenase Synthesis

#### Edited by:

Chris Greening, Monash University, Australia

#### Reviewed by:

Bat-Erdene Jugder, Harvard Medical School, United States Rhiannon Mari Evans, University of Oxford, United Kingdom

#### \*Correspondence:

Constanze Pinske constanze.pinske@ mikrobiologie.uni-halle.de R. Gary Sawers gary.sawers@mikrobiologie. uni-halle.de

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 19 September 2018 Accepted: 05 November 2018 Published: 21 November 2018

#### Citation:

Dragomirova N, Rothe P, Schwoch S, Hartwig S, Pinske C and Sawers RG (2018) Insights Into the Redox Sensitivity of Chloroflexi Hup-Hydrogenase Derived From Studies in Escherichia coli: Merits and Pitfalls of Heterologous [NiFe]-Hydrogenase Synthesis. Front. Microbiol. 9:2837. doi: 10.3389/fmicb.2018.02837 Nadya Dragomirova, Patricia Rothe, Stefan Schwoch, Stefanie Hartwig, Constanze Pinske\* and R. Gary Sawers\*

Institute of Microbiology, Martin-Luther University Halle-Wittenberg, Halle, Germany

The highly oxygen-sensitive hydrogen uptake (Hup) hydrogenase from Dehalococcoides mccartyi forms part of a protein-based respiratory chain coupling hydrogen oxidation with organohalide reduction on the outside of the cell. The HupXSL proteins were previously shown to be synthesized and enzymatically active in Escherichia coli. Here we examined the growth conditions that deliver active Hup enzyme that couples H<sup>2</sup> oxidation to benzyl viologen (BV) reduction, and identified host factors important for this process. In a genetic background lacking the three main hydrogenases of E. coli we could show that additional deletion of genes necessary for selenocysteine biosynthesis resulted in inactive Hup enzyme, suggesting requirement of a formate dehydrogenase for Hup activity. Hup activity proved to be dependent on the presence of formate dehydrogenase (Fdh-H), which is typically associated with the H2-evolving formate hydrogenlyase (FHL) complex in the cytoplasm. Further analyses revealed that heterologous Hup activity could be recovered if the genes encoding the ferredoxin-like electron-transfer protein HupX, as well as the related HycB small subunit of Fdh-H were also deleted. These findings indicated that the catalytic HupL and electron-transferring HupS subunits were sufficient for enzyme activity with BV. The presence of the HupX or HycB proteins in the absence of Fdh-H therefore appears to cause inactivation of the HupSL enzyme. This is possibly because HupX or HycB aided transfer of electrons to the quinone pool or other oxidoreductase complexes, thus maintaining the HupSL heterodimer in a continuously oxidized state causing its inactivation. This proposal was supported by the observation that growth under either aerobic or anaerobic respiratory conditions did not yield an active HupSL. These studies thus provide a system to understand the redox sensitivity of this heterologously synthesized hydrogenase.

Keywords: hydrogen, formate, ferredoxin-like proteins, electron transfer, uptake hydrogenase, heterologous expression

### INTRODUCTION

fmicb-09-02837 November 19, 2018 Time: 14:41 # 2

The bacterial genus Dehalococcoides belongs to the phylum Chloroflexi and the type species D. mccartyi is completely dependent on hydrogen for growth (Löffler et al., 2013; Schubert et al., 2018). D. mccartyi synthesizes several types of [NiFe]-hydrogenase (Hyd), and the hydrogen-uptake (Hup) hydrogenase is thought to be the main enzyme involved in H2-driven organohalide respiration. As D. mccartyi lacks quinones (Kube et al., 2005; Schipp et al., 2013), a direct transfer of the electrons derived from H<sup>2</sup> oxidation by Hup via protein–protein interaction has been implicated (Kublik et al., 2016; Hartwig et al., 2017; Seidel et al., 2018). The Hup enzyme is found in a respiratory supercomplex comprising a two-subunit complex iron-sulfur molybdoprotein, OmeAB (organohalide molybdoenzyme) and one of a number of reductive dehalogenases (Rdh), which catalyze the reduction of particular organohalides that function as electron acceptors for the bacterium (Fincker and Spormann, 2017; Schubert et al., 2018). In addition, the ferredoxin-like protein HupX, which resembles electron-transferring subunits of oxidoreductases, is associated with the complex.

Hup comprises two structural components: the catalytic subunit HupL, containing the NiFe(CN)2CO cofactor and HupS, the small electron-transferring subunit, which is predicted to have three iron-sulfur clusters. The membrane-associated, ferredoxin-like protein HupX is encoded within the operon of the Hup hydrogenase, but seems to associate more tightly with the core OmeAB-Rdh complex (Hartwig et al., 2017; Seidel et al., 2018), suggesting that it is the main mediator of electron transfer and acts as a "connector" protein between HupSL and the rest of the complex. HupX is homologous to HybA, a component of the Hyd-2 H2-oxidizing hydrogenase of Escherichia coli (Sargent et al., 1998; Beaton et al., 2018) and recent studies have provided strong evidence indicating that HybA is responsible for coupling electron transfer to the quinone pool, as Hyd-2 has no membrane subunit with a recognized heme cofactor, necessary for electron transfer into the membrane (Dubini et al., 2002; Pinske et al., 2015; Beaton et al., 2018).

The ferredoxin-like family of electron–transfer proteins harbors four [4Fe-4S] clusters and an interaction network of several members of this family has been uncovered recently in E. coli (Pinske, 2018). One member is HycB, the small subunit of the formate dehydrogenase (Fdh-H) that forms one of the two catalytic centers of the formate hydrogenlyase (FHL) complex, and another is the related protein HydN, which is proposed to be involved in FHL complex assembly (Pinske, 2018). Generally, however, the physiological function of most members of this emerging superfamily of iron-sulfur-containing electron transfer proteins is not understood.

Due to the fact that D. mccartyi grows extremely slowly and produces limited amounts of biomass, making biochemical studies challenging, we have established a heterologous expression system for the synthesis of a functional Hup enzyme in E. coli (Hartwig et al., 2015b). It is hoped that this system will facilitate a detailed biochemical characterization of Hup. Despite the significant phylogenetic distance between D. mccartyi and E. coli, the Hyp maturation system responsible for [NiFe]-cofactor biosynthesis and insertion (Böck et al., 2006) is capable of recognizing the HupL apoprotein and generating an active enzyme when the complete operon encoding Hup is expressed under anaerobic conditions (Hartwig et al., 2015b).

As well as the three structural genes, the hupXSLhoxM operon (**Figure 1A**) also encodes a HupL-specific maturation endoprotease (HoxM). Initial characterization of the heterologously synthesized Hup enzyme identified a fast-migrating complex, mainly comprising HupS and HupL after native-PAGE, which migrated at a similar position as the complex present in crude extracts of D. mccartyi that contained HupSL and minor amounts of HupX (Hartwig et al., 2015b). This suggests that HupSL alone is capable of catalyzing H2-dependent reduction of the redox dye BV. The activity of the complex was oxygen-sensitive, even when synthesized anaerobically in the heterologous host (Hartwig et al., 2015b), suggesting that a cofactor in the enzyme is redox-sensitive. Whether this redox-sensitive cofactor is in HupL, HupS or HupX is unclear. Therefore, to address these questions, in the current study we decided to determine the conditions necessary for heterologous production of HupSL activity and whether any other components of the host's metabolism, other than the Hyp proteins, are required for activity to be visualized. Surprisingly, we found a strong dependence for HupSL activity on the Fdh-H enzyme of the FHL complex. This dependence on Fdh-H for activity proved to be linked to an involvement of ferredoxin-like electron transfer proteins and to the redox sensitivity of the HupSL heterodimer.

#### MATERIALS AND METHODS

#### Strains and Growth Conditions

The strains listed in **Table 1** were used in this study. For routine molecular biology studies, growth was on LB-agar plates or in LBbroth at 37◦C (Miller, 1972). Anaerobic growths were performed at 37◦C as standing liquid cultures and cells were usually grown in M9 minimal medium (47.6 mM Na2HPO<sup>4</sup> × 2 H2O, 22 mM KH2PO4, 8.4 mM NaCl, 20 mM NH4Cl, 2 mM MgSO4, 0.1 mM CaCl2, 0.1 mM thiamin dichloride, 0.2% w/v casamino acids) containing 0.8% (w/v) glucose, or 0.4% (v/v) glycerol plus 15 mM fumarate, or 0.8% (w/v) glucose plus 1% (w/v) nitrate, where indicated, as described (Sambrook et al., 1989). When growth in rich medium was performed, buffered TGYEP (1% w/v tryptone, 0.5% w/v yeast extract, 0.8% w/v glucose, 100 mM potassium phosphate, pH 6.5) was used (Begg et al., 1977). The growth medium was supplemented with trace element solution SLA (Hormann and Andreesen, 1989). When required, the antibiotic kanamycin or chloramphenicol was added to a final concentration of 50 or 25 µg ml−<sup>1</sup> , respectively. Cells were harvested anaerobically by centrifugation at 5,000 g for 15 min at 4◦C when cultures had reached an OD600 nm of between 0.8 and 1.2. Cell pellets were used immediately or stored at −20◦C until use.

#### Plasmid and Strain Construction

The hupX gene (cbdbA131) was amplified as a 1212 bp DNA fragment from chromosomal DNA isolated from D. mccartyi strain CBDB1 using Pfu DNA polymerase and the oligonucleotides hupX\_fw (5<sup>0</sup> - GGGGCATATGCCTAATGGAATGCTGATTG-3<sup>0</sup> ) and hupX\_re (5<sup>0</sup> -GGGGCTCGAGCTAGTGCTTGCCAGCCTTG-3 0 ) and cloned in plasmid pACYC-Duet-I. Plasmid phupSL was constructed by using pSHH18 (referred to as phupXSL throughout this study) as template in a PCR mutagenesis employing the Q5 <sup>R</sup> Site-Directed Mutagenesis Kit (New England Biolabs, NEB). Care was taken when deleting the hupX gene to ensure that the ribosome binding site for the downstream hupS gene remained intact by using the oligonucleotides hupSLhoxM\_fw (5<sup>0</sup> -ATGGAGTAGGAAAAATGTTTAATAC-3 0 ) and hupSLhoxM\_re (5<sup>0</sup> -TCCTGTTGCCCCCCTTGT-3<sup>0</sup> ) and by following the instructions given in the Q5 <sup>R</sup> Site-Directed Mutagenesis Kit.

E. coli strains were constructed using P1kc-mediated phage transduction (Miller, 1972) to introduce the respective defined deletion mutation from the appropriate donor strain obtained from the Keio collection (Baba et al., 2006) to generate the series of FTD147 mutants lacking the structural genes encoding the three formate dehydrogenases of E. coli. When multiple gene knock-outs were constructed, the plasmid pCP20 was used to remove the kanamycin antibiotic resistance cassette as described (Cherepanov and Wackernagel, 1995).

### Preparation of Crude Cell Extracts and Cell Fractionation

Unless otherwise stated, all experiments were performed in an anaerobic CoyTM chamber under an atmosphere of 95% nitrogen/5% hydrogen. For standard hydrogenase enzyme activity determination, E. coli cell paste was re-suspended at a ratio of 1 g cell wet weight to 3 ml 50 mM MOPS buffer, pH 7. Cells were disrupted by sonication (30 W power for 5 min with 0.5 s pulses). Unbroken cells and cell debris were removed by centrifugation for 30 min at 50,000 g and 4◦C. The resulting crude extract, unless otherwise stated, was used for all studies reported herein.

In order to perform sub-cellular fractionation, periplasmic, soluble and membrane fractions were isolated as described (Sawers et al., 1985).

#### TABLE 1 | Strains and plasmids used in this study.

fmicb-09-02837 November 19, 2018 Time: 14:41 # 4


<sup>a</sup>The series of FTD147 mutants with different combinations of Fdh gene mutations was constructed by transduction of mutations from the corresponding Keio collection of mutants (see Materials and Methods). <sup>b</sup>Note that for reasons of clarity, this plasmid was referred to as phupXSL throughout this study.

Determination of protein concentration was done as described (Lowry et al., 1951).

#### Non-denaturing Polyacrylamide Gel Electrophoresis and Activity-Staining

Unless otherwise specified, non-denaturing polyacrylamide gel electrophoresis (PAGE) was performed anaerobically. Separating gels included 0.1% (w/v) Triton X-100 as described (Ballantine and Boxer, 1985). The crude extracts, or sub-cellular fractions, were incubated with a final concentration of 4% (w/v) Triton X-100 prior to application (usually 50 µg of protein) to the gel, which included 6% (w/v) polyacrylamide. Hydrogenase activitystaining was done in 50 mM MOPS buffer pH 7.0, as described (Sawers et al., 1985; Pinske et al., 2012), and included 0.5 mM BV and 1 mM 2,3,5-triphenyltetrazolium chloride (TTC). Gels were incubated under an atmosphere of 100% hydrogen gas.

#### Hydrogenase Activity Assay

Measurement of hydrogenase enzyme activity using BV as electron acceptor was performed as described (Ballantine and Boxer, 1985; Pinske et al., 2011). Briefly, anaerobically prepared cuvettes (1.6 ml) were filled with 0.8 ml of H2-saturated, anaerobic 50 mM MOPS buffer, pH 7.0, including 4 mM BV and placed under a H<sup>2</sup> atmosphere. After baseline determination, the assay was initiated by adding enzyme sample (approximately 150 µg of protein). All assays were performed at 25◦C. The wavelength used was 600 nm and an ε<sup>M</sup> value of 7400 M−<sup>1</sup> cm−<sup>1</sup> was assumed for reduced BV. One million unit of enzyme activity corresponded to the reduction of 1 nmol of substrate min−<sup>1</sup> . Enzyme assays were performed in triplicate using three biological replicates.

#### Denaturing Polyacrylamide Gel Electrophoresis (PAGE) and Western Blotting

Polypeptides in crude extracts were separated by 12.5% (w/v) sodium dodecyl sulfate (SDS)-PAGE (Laemmli, 1970) and gels were either stained with Coomassie Brilliant Blue R or transferred to nitrocellulose membranes for western blotting, which was performed as described (Towbin et al., 1979). The antibodies used were either anti-Strep-tag (IBA Life Sciences), anti-Hyd-2 (Sargent et al., 1998), anti-HupL or anti-HupX peptide antibodies (Hartwig et al., 2017).

### RESULTS

#### HupL and HupS Are Sufficient for BV Reduction Activity

The hupXSLhoxM operon has been shown to be functional in anaerobically grown E. coli (Hartwig et al., 2015b). In order to determine whether all three structural components

(HupSL and HupX) are essential for the manifestation of the H2:BV oxidoreductase activity observed in that study, we constructed two additional plasmid derivatives, one of which carried only the hupX gene, while the other included hupSLhoxM but lacked hupX (**Figure 1A**). These plasmids, along with pSHH18 (hupXSLhoxM+; Hartwig et al., 2015b; here referred to as phupXSL in the aid of clarity), were introduced into FTD147, which lacks the genes encoding the catalytic subunits of Hyd-1, Hyd-2, and Hyd-3 (Redwood et al., 2008). After fermentative growth, crude extracts were separated in native-PAGE and stained for hydrogenase enzyme activity (**Figure 1B**). As anticipated, the plasmid encoding only HupX showed no hydrogenase enzyme activity in extracts of strain FTD147 (1hyaB1hybC1hycE), while both of the other plasmids resulted in a fast-migrating activity band corresponding to the HupSL heterodimer (**Figure 1B**). Notably, although the activity resulting from introduction of the plasmid lacking the hupX gene (phupSL in **Figure 1B**) was apparently weaker than that resulting from introduction of phupXSL, both enzyme activities showed very similar migration characteristics, indicating that HupX is neither necessary for the ability of the enzyme to reduce BV nor seems to co-migrate with HupSL in this particular activity band. This result correlates well with earlier mass spectrometric analyses of heterologously expressed enzyme, which identified mainly the HupL protein (Hartwig et al., 2015b).

#### Manifestation of Heterologous HupSL Enzyme Activity Requires Fermentative Growth Conditions

In order to optimize conditions for the analysis of heterologously produced HupSL activity, we tested different anaerobic growth conditions using FTD147 (1hyaB1hybC1hycE) transformed with either phupXSL or phupSL (**Figure 2A**). The activity band was slightly more intense when cells were grown with 0.8% w/v glucose compared with half that glucose concentration (0.4% w/v). Suprisingly, however, no HupSL activity could be detected when cells were grown under anaerobic respiratory conditions, with either glycerol and fumarate or glucose and nitrate. Western blot analysis of the extracts derived from anaerobically grown strains after separation by SDS-PAGE using peptide antibodies raised against HupL revealed that the HupL polypeptide could be detected in each extract (**Figure 2B**). This indicates that a lack of transcription of the hup genes under respiratory conditions was not the reason for absence of HupSL enzyme activity. Surprisingly, it was not possible to restore in vitro HupSL enzyme activity to these extracts, even by incubating the extracts under reducing conditions. This suggests that the HupSL enzyme was irreversibly inhibited under the oxidizing conditions that prevailed within the cells grown under these conditions.

Quantitative assessment of H2-dependent BV reduction activity in anaerobically prepared, concentrated crude extracts of FTD147 (1hyaB, 1hybC, 1hycE) transformed with phupXSL measured a low but detectable hydrogenase activity of approximately 60 mU/mg (**Table 2**), which is in good agreement with previously determined HupSL activity in E. coli extracts (Hartwig et al., 2015b). The phenotypically identical strain CP1170 (1hyaB, 1hybC, 1hycE) had a background activity of 10 mU/mg. Brief incubation of the extract from FTD147 + phupXSL in the presence of oxygen resulted in a reduction of the HupSL activity by 50% (**Table 2**).

#### HupSL Activity in E. coli Requires a Functional Selenocysteine-Insertion Machinery

The lack of HupSL enzyme activity after respiratory growth is reminiscent of the effects of these growth conditions on

FIGURE 2 | Heterologous HupSL enzyme activity is only detectable in fermentatively growing E. coli cells. (A) Crude extracts derived from FTD147 (1hyaB1hybC1hycE) transformed with plasmid phupXSL (1) or phupSL (2) after growth in M9 minimal medium with the indicated carbon sources were separated in native-PAGE and stained for hydrogenase enzyme activity, as described in the legend to Figure 1. (B) A western blot using peptide antibodies raised against HupL (Hartwig et al., 2017) is shown in which the same extracts used in part A were separated in 10% (w/v) SDS-PAGE. The molecular mass marker proteins are presented in kDa. MC4100 was the wild type strain.



<sup>a</sup>Crude extracts were exposed to air for 15 min prior to determination of hydrogenase activity. <sup>b</sup>Assays were performed in triplicate from 3 independent growth experiments. The activity is shown together with the standard deviation. <sup>c</sup>n.d, not determined.

appearance of E. coli Hyd-3 and Fdh-H enzyme activities (Sawers et al., 1985), with the exception that the effect on synthesis of the E. coli enzymes is at the transcriptional level due to depletion of the regulatory metabolite formate (Rossmann et al., 1991). Due to the fact that HupSL is naturally associated with a formate dehydrogenase-like enzyme, OmeAB (Kublik et al., 2016; Hartwig et al., 2017), we wished to examine the influence of the Fdh-O and Fdh-N enzymes, which are phylogenetically related to OmeAB, on HupSL enzyme activity. Initially, we introduced into strain FTD147 a mutation in the selC gene, which encodes the selenocysteinyl-tRNASec necessary for translation of special UGA codons as selenocyteine, and which, when deleted, renders all Fdhs inactive (Leinfelder et al., 1988). This would provide information on whether HupSL enzyme activity was influenced by defects in formate metabolism. The left panel shown in **Figure 3** shows a control for HupSL activity revealing that is was readily detectable in strain FTD150 (1hyaB1hybC1hycE1hyfG), which is identical to strain FTD147 with the exception that the gene encoding the catalytic subunit of Hyd-4 (Andrews et al., 1997) is also deleted (Redwood et al., 2008). Thus, both FTD150 and FTD147

FIGURE 3 | Appearance of HupSL enzyme activity is dependent on a functional selenocysteine biosynthetic apparatus. Crude extracts (100 µg of protein, except M4100 where 30 µg of protein were applied) of the indicated strains were separated in native-PAGE and stained for hydrogenase enzyme activity as described in the legend to Figure 1. Strains used include: MC4100 (wild type); FTD150 (1hyaB1hybC1hycE 1hyfG); FTD1471selC (1hyaB1hybC1hycE 1selC).

yield an identical phenotype with regard to the heterologous HupSL activity (see also below). Analysis of an extract of the FTD147 1selC mutant revealed that no HupSL activity was detectable (**Figure 3** right panel). The lack of selC was confirmed by the absence of the H2:BV oxidoreductase activity associated with Fdh-N/O in the strain (Soboh et al., 2011). This result confirms that HupSL enzyme activity is linked to formate metabolism, most likely through one of the formate dehydrogenases (Fdh) the bacterium synthesizes under anaerobic conditions. Introduction of a mutation in selB, which encodes the special translation factor required to decode the UGA codon as selenocysteine (Forchhammer et al., 1989), into FTD150 also revealed a similar lack of HupSL activity (data not shown), confirming that the phenotype was due to a lack of selenocysteine incorporation.

#### In-Gel HupSL Activity Depends on the Fdh-H Enzyme

Dependence on the selenocysteine biosynthetic machinery for appearance of HupSL enzyme activity suggests an involvement of one or more of the three Fdhs present in E. coli. To determine which of the three Fdhs is required for the appearance of heterologous HupSL activity, we constructed a series of strains (see **Table 1**) lacking one or more of the genes encoding the catalytic subunit of FdnG (of Fdh-N), FdoG (of Fdh-O), or FdhF (of Fdh-H) (Pinske and Sawers, 2016; **Figure 4**). Strain FTD150, which lacked all four hydrogenases and the quadruple and quintuple mutants of FTD147, which lacked Hyd-1, Hyd-2, Hyd-3 as well as either or both respiratory Fdhs (Fdh-N and Fdh-O), retained fully active HupSL (**Figure 4**). Hydrogenase activity in extracts derived from FTD1471fdnG1fdoG with plasmid phupXSL was approximately 60 mU, while the strain without plasmid had approximately half this activity (**Table 2**). Exposure of the crude extract from FTD1471fdnG1fdoG transformed with phupXSL to air resulted in a similar 50–60% reduction in hydrogenase activity as was observed with FTD147 containing phupXSL (**Table 2**). This result confirms that HupSL is oxygen-labile (Hartwig et al., 2015b).

The only strains that lacked a detectable HupSL enzyme activity band were those that lacked the fdhF gene, which encodes the Fdh-H component of the FHL complex (**Figure 4**). Assay of hydrogenase activity in extracts derived from FTD1471fdhF + phupXSL failed to show HupSL-dependent hydrogenase activity

(**Table 2**). These findings indicate that for full HupSL activity to be manifested, an active Fdh-H enzyme is required.

In order to determine what the link between the appearance of HupSL enzyme activity and Fdh-H might be, we first performed a western blot using anti-HupX antibodies and with crude extracts derived from some of the strains shown in **Figure 4**. Surprisingly, HupX was only detectable in extracts of strains lacking fdhF, which encodes the Fdh-H enzyme, and, as expected, only in those strains carrying the phupXSL plasmid (**Figure 5**). This suggests that when Fdh-H was absent, HupX was stably synthesized and

FIGURE 5 | HupX is only detectable in extracts of strains lacking the fdhF gene. Western blots using peptide antibodies raised against HupX or HupL are shown in which 50 µg of crude extracts (A) or subcellular fractions (B) were separated in 10% (w/v) SDS-PAGE. The migration positions of molecular mass marker proteins are shown in kDa. The strong signal migrating around 70 kDa represents GroEL, which cross-reacted with the HupX antibodies, and acted as a protein loading control. The asterisks denote the HupX polypeptide. In (B), the letters E, S, and M represent crude extract, soluble cytoplasmic fraction and membrane fraction, respectively. Strain used include: deletion derivatives of FTD147 (1hyaB1hybC1hycE); and deletion derivatives of FTD150 (1hyaB1hybC1hycE 1hyfG).

when Fdh-H was present in the cells, HupX became unstable and was presumably degraded.

#### The Influence of Fdh-H on HupSL Activity Is Indirect via the Ferredoxin-Like Proteins HupX and HycB

To examine whether HupX might influence HupSL activity, plasmids phupSL and phupXSL, both encoding HupL, HupS and the endoprotease HoxM, but only the latter also encoding HupX (**Figure 1A**), were introduced into strains FTD147 and FTD1471fdhF and enzyme activity was compared after anaerobic growth with glucose (**Figure 6A**). The products of both plasmids in strain FTD1471fdhF showed a strongly reduced activity of HupSL compared with the fdhF<sup>+</sup> strain FTD147. This result indicates that the dependence on Fdh-H for HupSL activity was retained in the absence of HupX.

A recent study in E. coli identified a flexible interaction network of ferredoxin-like proteins with Fdh-H, including its small, electron-transferring subunit, HycB (Pinske, 2018). HycB and HupX belong to this family but share only 24% amino acid identity (38% similarity) in a MuscleWS alignment. The HupX protein, however, cannot functionally replace HycB in formate-dependent BV reduction (data not shown). Due to the link between HupSL activity and Fdh-H demonstrated above, we therefore examined whether the presence of HycB influenced HupSL's ability to reduce BV. To do this, we analyzed HupSL activity in strain CP1170 (1hyaB, 1hybC, 1hycA-I), which is similar to FTD147 (1hyaB, 1hybC, 1hycE) with the exception that the complete hyc operon is deleted in CP1170, rather than only the hycE gene (**Table 1**). Introduction of plasmid phupXSL into CP1170 and its 1fdhF derivative CPH020 (1hyaB1hybC1hycA-I1fdhF) (**Table 1**), revealed that the dependence on Fdh-H for HupSL activity was retained (**Figure 4**, right side of right panel). However, introduction of plasmid phupSL (lacking hupX) into CPH020 (CP1170 1fdhF) revealed that HupSL activity in native PAGE was no longer reduced in the absence of Fdh-H (**Figure 6B**, lane 1, right panel). Introduction of phupSL into strain CPH008 (1hycA-I, 1fdhF) in which Hyd-1 and Hyd-2 are still active, but all structural components of the FHL complex are missing, demonstrated that HupSL activity was retained, and even slightly more intense (**Figure 6B**, lane 2 left panel). As a final control, we analyzed HupSL activity after introduction of phupSL into strain RT2 (1hyaB1hybC1hycA-I1fdhE1pflA), which lacks Hyd-1, Hyd-2 and Hyd-3, as well as Fdh-N/O (through the fdhE mutation; Mandrand-Berthelot et al., 1988; Lüke et al., 2008) and the formate-inducible Fdh-H, due to the lack of active PflB (due to the pflA mutation), which is required for formate production (Sawers and Böck, 1988; Rossmann et al., 1991). This strain was chosen because it uses a different combination of mutations to generate the same phenotype, i.e., no hydrogenase or formate dehydrogenase enzymes, and no HycB protein. HupSL activity was also retained in this genetic background confirming that strains devoid of HupX and HycB exhibit HupSL- and H2-dependent reduction of BV (**Figure 6B**, lane 1).

Surprisingly, however, when phupSL was introduced into strain RM220 (1pflAB), which generates considerably reduced levels of formate under respiratory conditions (Suppmann and

Frontiers in Microbiology | www.frontiersin.org

1fdhF); CPH020 (1hyaB1hybC1hycA-I1fdhF); RM220 (1pflA1pflB).

Sawers, 1994), and thus expresses only low levels of the formateinducible hyc operon (Rossmann et al., 1991), the anticipated high activity of HupSL was not observed (**Figure 6B**, lane 4). This result indicates that even the low levels of HycB produced in this strain (Rossmann et al., 1991) are likely sufficient to inhibit HupSL activity.

#### DISCUSSION

In this study, we analyzed how the fermentative metabolism of the E. coli host influenced the redox dye-reducing activity of a heterologously synthesized hydrogen-uptake [NiFe] hydrogenase from the Chloroflexi phylum. We had previously demonstrated that the host's Hyp-maturation machinery was capable of recognizing the large subunit precursor pre-HupL and of successfully introducing the bimetallic [NiFe]-cofactor, generating active enzyme (Hartwig et al., 2015b). We also showed in that particular study that the HupSL hydrogenase had H2:BV oxidoreductase activity, which could be identified after anaerobic native-PAGE. Here we made the surprising discovery that the appearance of this HupSL enzyme activity was apparently dependent on whether the host's Fdh-H enzyme was synthesized or not. Under conditions favoring synthesis of Fdh-H, HupSL activity was observed, while in strains unable to synthesize Fdh-H, due to a deletion of the fdhF structural gene or the selenocysteine insertion machinery, no, or substantially reduced, activity was detected. Notably, however, this lack of enzyme activity did not result from a lack of synthesis of the HupSL enzyme, but rather appears to be due to an inactivation of the enzyme.

A recent study revealed that Fdh-H interacts with at least three electron-transferring small subunits, all of which belong to the ferredoxin-like family of electron-transfer proteins and possibly facilitate the coupling of Fdh-H with different enzyme complexes (Pinske, 2018). We currently interpret our data to indicate that the apparent dependence on Fdh-H for HupSL activity is, in fact, indirect and likely due to Fdh-H sequestering these small subunits, in particular HycB of the FHL complex. Notably, the HupX protein, which is presumed to mediate electron transfer within the Hup-Ome-Rdh supercomplex in the natural host D. mccartyi (Schubert et al., 2018), also belongs to the ferredoxinlike superfamily and this protein's ability to interact with the HupSL heterodimer also appears to be influenced by the presence of Fdh-H. If Fdh-H is either genotypically or phenotypically (e.g., through strongly reduced formate synthesis; Rossmann et al., 1991) absent, the ferredoxin-like proteins HupX or HycB remain consequently unbound within the cell. We suggest that if HupX is also absent, HycB can transiently interact with or modulate the HupSL enzyme within the cell prior to separation in the native-PAGE, rendering the enzyme inactive. This inactivity could result from a loss of the ability of the heterodimer to reduce or interact with BV in the presence of H2. Alternatively, these ferredoxinlike proteins might act by sequestering the HupSL complex resulting in an inactive complex in the native-PAGE; or indeed a combination of both effects might be the cause (**Figure 7**). The consequence would be that HupSL activity becomes visible in the

FIGURE 7 | A schematic depiction of the current interpretation of our data is shown. The hypothesis states that the HupSL heterodimer is capable of interacting with and reducing BV and if HupX or HycB are free and not sequestered by Fdh-H they prevent HupSL transferring electrons to BV and instead deliver them to currently unknown acceptor complexes, or possibly to the quinone pool (?). HupX has been shown to interact with HupSL (Hartwig et al., 2017; Seidel et al., 2018), while this has not been so far shown for HycB.

absence of HycB despite simultaneous absence of Fdh-H. The redox-potentials of the ferredoxin-like proteins have not yet been determined.

The model shown in **Figure 7** presents a working hypothesis for how we currently interpret our data. Under fermentative growth conditions Fdh-H is present and is available to interact with HycB. The presence of Fdh-H also prevents HupX interacting with HupSL, possibly forming an interaction with HupX, allowing the HupSL heterodimer to interact with and reduce BV.

How does an ability to interact with HupX, and possibly HycB, interfere with electron transfer to BV by HupSL? The redox dye BV can accept electrons directly from the electron-transferring subunit HupS, as evidenced by the fact that deletion of the hupX gene does not significantly affect H2-dependent reduction of BV by HupSL. Moreover, previous mass spectrometric analysis of the Hup activity band isolated after native-PAGE revealed mainly HupL and HupS to be present, suggesting that HupX's interaction with the heterodimer is weak or transient (Hartwig et al., 2015b). Moreover, in the natural host D. mccartyi, HupX preferentially associates with OmeAB, the formate dehydrogenase homolog, and reductive dehydrogenases rather than with HupSL (Hartwig et al., 2015b; Seidel et al., 2018; Dragomirova and Sawers, unpublished observations), which supports the suggestion that the affinity of HupX for HupSL is low. We also observed using antibodies raised against HupX that it is only readily detectable in the membrane fraction of cells that lack Fdh-H, suggesting that when Fdh-H is present it is more readily degraded. How this apparent degradation is controlled is currently unclear.

The HupS subunit encodes a functional Tat-signal peptide allowing its transport across the cytoplasmic membrane (Hartwig et al., 2015b). Together with HupX, the HupSL complex could be sufficiently anchored in the membrane to transfer the electrons

derived from oxidation of H<sup>2</sup> to the quinone pool, which is also the function of another HupX homolog, the HybA protein of E. coli Hyd-2 (Pinske et al., 2015; Beaton et al., 2018).

Support for the oxidative inactivation of HupSL was provided by the demonstration of inactivation of the enzyme complex after growth of the E. coli host under respiratory conditions, with either O<sup>2</sup> (E <sup>o</sup><sup>0</sup> = +830 mV), NO<sup>3</sup> − (E <sup>o</sup><sup>0</sup> = +420 mV), or fumarate (E <sup>o</sup><sup>0</sup> = 0 mV) as electron acceptor. We have only been able to detect HupSL activity after fermentative growth (E <sup>o</sup><sup>0</sup> = -415 mV), strongly suggesting that the enzyme retains activity only under strongly reducing conditions, which are also likely to be those prevailing in the environmental conditions where D. mccartyi is found (Löffler et al., 2013).

These studies thus provide a platform to study how heterologously synthesized hydrogenases can be integrated into the host's anaerobic metabolism. Clearly, this work is at an early stage but one of the next steps will be to examine whether electrons derived from H<sup>2</sup> oxidation can be coupled to reduction of E. coli-typical electron acceptors, e.g., fumarate. Initial studies examining hydrogen-driven fumarate reduction by HupXLS yielded first indications that a weak, but unfortunately so far irreproducible, activity was detectable (Schwoch et al., unpublished data). Optimization of Hup enzyme synthesis and membrane integration will likely be required for this approach to be fruitful.

Because D. mccartyi species are not amenable to large-scale biochemical analysis, and are currently genetically intractable, using the E. coli Hyd- and Fdh-negative host strains developed

#### REFERENCES


here will provide a means of studying the biochemical mechanism(s) underlying the loss of HupSL activity in response to oxidizing redox conditions and whether this effect is linked to a particular iron-sulfur cluster, or clusters, in HupS, or whether the bimetallic cofactor in HupL is the target of irreversible inactivation. A recent study by Hartmann et al. (2018) indicates that, at least for certain [NiFe]-hydrogenases, the NiFe(CN)2CO cofactor is not sensitive to oxidative conditions, suggesting that it might indeed be the electron-transfer pathway that is disrupted by non-reducing redox potentials.

#### AUTHOR CONTRIBUTIONS

ND, PR, SS, SH, and CP designed and performed the experiments and analyzed the data. CP and RS conceived the study, interpreted the data and drafted the manuscript. All authors read and approved the manuscript.

#### FUNDING

This work was funded by the Deutsche Forschungsgemeinschaft as part of the research unit FOR1530: "Anaerobic Biological Dehalogenation: Organisms, Biochemistry, and (Eco)physiology" granted to RS and by the priority program SPP1927 "Iron Sulfur for Life" granted to both CP and RS. We acknowledge the financial support of the Open Access Publication Fund of the Martin Luther University Halle-Wittenberg.



**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 Dragomirova, Rothe, Schwoch, Hartwig, Pinske and Sawers. 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.

## Microbially Mediated Hydrogen Cycling in Deep-Sea Hydrothermal Vents

#### Nicole Adam and Mirjam Perner\*

Geomicrobiology, GEOMAR Helmholtz Centre for Ocean Research Kiel, Kiel, Germany

Deep-sea hydrothermal vents may provide one of the largest reservoirs on Earth for hydrogen-oxidizing microorganisms. Depending on the type of geological setting, hydrothermal environments can be considerably enriched in hydrogen (up to millimolar concentrations). As hot, reduced hydrothermal fluids ascend to the seafloor they mix with entrained cold, oxygenated seawater, forming thermal and chemical gradients along their fluid pathways. Consequently, in these thermally and chemically dynamic habitats biochemically distinct hydrogenases (adapted to various temperature regimes, oxygen and hydrogen concentrations) from physiologically and phylogenetically diverse Bacteria and Archaea can be expected. Hydrogen oxidation is one of the important inorganic energy sources in these habitats, capable of providing relatively large amounts of energy (237 kJ/mol H2) for driving ATP synthesis and autotrophic CO<sup>2</sup> fixation. Therefore, hydrogen-oxidizing organisms play a key role in deep-sea hydrothermal vent ecosystems as they can be considerably involved in light-independent primary biomass production. So far, the specific role of hydrogen-utilizing microorganisms in deep-sea hydrothermal ecosystems has been investigated by isolating hydrogen-oxidizers, measuring hydrogen consumption (ex situ), studying hydrogenase gene distribution and more recently by analyzing metatranscriptomic and metaproteomic data. Here we summarize this available knowledge and discuss the advent of new techniques for the identification of novel hydrogen-uptake and -evolving enzymes from hydrothermal vent microorganisms.

#### Edited by:

Chris Greening, Monash University, Australia

#### Reviewed by:

Xiyang Dong, University of Calgary, Canada Carlo Robert Carere, University of Canterbury, New Zealand

#### \*Correspondence:

Mirjam Perner mperner@geomar.de

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 03 August 2018 Accepted: 08 November 2018 Published: 23 November 2018

#### Citation:

Adam N and Perner M (2018) Microbially Mediated Hydrogen Cycling in Deep-Sea Hydrothermal Vents. Front. Microbiol. 9:2873. doi: 10.3389/fmicb.2018.02873 Keywords: hydrogen cycling, hydrogen consumption, hydrogenases, hydrogen oxidizers, hydrothermal vent

### INTRODUCTION

Hydrogen conversion, the reversible reaction of molecular hydrogen (H2) to protons and electrons, plays a major role for metabolic processes in microbial cells: generally, energy conservation and the recycling of reducing equivalents (in microbial fermentation or light-dependent photosynthesis) is accomplished by enzymatic hydrogen evolution (Vignais and Billoud, 2007; Hallenbeck, 2009). Enzymatically catalyzed hydrogen oxidation is widely distributed among prokaryotes, and can power the synthesis of energy-rich ATP, which is needed for autotrophic carbon fixation (Dilling and Cypionka, 1990; Bothe et al., 2010; Greening et al., 2016).

The thermal (4◦C to several 100s ◦C) and chemical (e.g., oxidized to reduced) gradients hallmarking deep-sea hydrothermal vent habitats have the potential to host one of the largest reservoirs of physiologically and phylogenetically diverse hydrogen-converting microorganisms

(**Figure 1**, Kelley et al., 2002; Perner et al., 2013b). As the fluids pass through the subsurface, they get enriched in various inorganic compounds, such as reduced minerals, sulfide and hydrogen (**Figure 1**). The actual hydrogen and sulfide concentrations of the emanating fluids strongly depend on the type of host rock underlying the respective vent system and the mixing ratio of seawater and fluids. Hydrothermal endmember fluids of basalt-hosted systems are usually characterized by greater sulfide than hydrogen concentrations, resulting from magma degassing and high-temperature-leaching from enclosing host rocks. In contrast, due to serpentinization processes, end-member fluids of ultramafic-hosted vent systems usually exhibit greater hydrogen (up to 1–10 M) and methane (on mM levels) concentrations than sulfide concentrations (Charlou et al., 2002; Kelley et al., 2005; Haase et al., 2007; Perner et al., 2013b). Correspondingly, sulfide oxidation in the sulfide-rich basalt-hosted and hydrogen oxidation and methanotrophy in the hydrogen-rich ultramafic-hosted systems are estimated to be the predominant sources of metabolic energy available in venting habitats (McCollom, 2007).

Since the discovery of hydrothermal vents in the late 70s (Corliss et al., 1979), numerous hydrogen-oxidizers have been isolated from thermally and chemically distinct deep-sea vent habitats (e.g., Campbell et al., 2006; Miroshnichenko and Bonch-Osmolovskaya, 2006; Nakagawa and Takai, 2008; Hansen and Perner, 2015; Nagata et al., 2017). Although considerable efforts have been undertaken to promote our understanding of the distribution and role of hydrogen-oxidizing organisms in these environments (Nealson et al., 2005; Campbell et al., 2006; Perner et al., 2010, 2013b; Petersen et al., 2011; Adam and Perner, 2018), our knowledge of the overall hydrogen utilization potential and microbial hydrogen pathways still remains limited. This review summarizes the work that has been done on hydrogen-metabolizing microorganisms colonizing hydrothermally influenced environments with respect to their diversity, hydrogen consumption rates in incubation experiments and protein biochemistry. Recent findings in the context of culture-independent metagenomic and metatranscriptomic approaches for the identification of novel hydrogen-converting enzymes are included. Finally, an outlook is given which techniques (e.g., in situ experiments) and work are needed to

FIGURE 1 | Overview of possible (microbially mediated) hydrogen cycling in hydrothermal vent systems. Hydrothermal emissions emanate from the subsurface either as high-temperature focused fluids causing the typical black or white smoker structures or low- to mid-temperature diffuse fluids (for example from mussel beds). Abiogenic hydrogen sources are displayed in red and orange (hydrothermal fluids) and biogenic hydrogen sources in the shape of green arrows.

advance our understanding of the role that hydrogen-cycling microorganisms play in hydrothermal vents.

#### HYDROGEN-PRODUCING AND -OXIDIZING MICROORGANISMS

It is well known that hydrogen-producing and -oxidizing microorganisms can coexist or even interact in a variety of anoxic habitats like sediments or intestinal tracts (Chassard and Bernalier-Donadille, 2006). At low hydrogen partial pressures (e.g., <100 Pa), hydrogen can be produced in the course of microbial fermentation processes (Kraemer, 2007; Hallenbeck, 2009) which is then oxidized by hydrogenotrophic microorganisms, especially methanogens. This interspecies hydrogen transfer thereby forms so-called syntrophic communities (hydrogen-producers and –consumers thrive in close proximity) and most likely represents an important hydrogen source in hydrogen-poor habitats (Bryant et al., 1977; Chassard and Bernalier-Donadille, 2006). Since fermentative hydrogen production can already be inhibited at relatively low hydrogen concentrations (i.e., on a nM level) (Wolin, 1976; Hoehler et al., 1998; Hallenbeck, 2009), the role, that microbially produced hydrogen plays in hydrothermal vent systems, remains enigmatic. Even the hydrogen levels of hydrogen-poor hydrothermal vent systems easily exceed those of habitats known to harbor fermentative bacteria like sediments (which are typically below 60 nM) (Novelli et al., 1987; Hoehler et al., 1998; Charlou et al., 2002; Perner et al., 2013b) and thus are likely above the inhibitory limit for biological hydrogen production. This may explain why studies on microbial hydrogen production in deep-sea hydrothermal vent systems have been largely neglected so far. However, hydrogen-evolving heterotrophic Archaea and Bacteria have been identified in hydrothermal fluid incubation experiments (Topcuoglu et al., 2016). The authors posited that in some of the micro niches represented by the culturing conditions, hyperthermophilic Euryarchaeota and thermophilic Firmicutes produced hydrogen as a waste product during fermentation which was consumed by hydrogenotrophic sulfate-reducing Bacteria or methanogenic Archaea (under distinct temperature regimes) (Topcuoglu et al., 2016). Hydrogenotrophic methanogens can use hydrogen to reduce CO<sup>2</sup> via the reductive acetyl-CoA pathway (Wood-Ljungdahl pathway), thereby forming methane (Ladapo and Whitman, 1990; Thauer, 1998). Acetogenic Bacteria (producing acetate from CO2) can compete with hydrogenotrophic methanogens in anoxic, hydrogen-rich habitats using the same electron donor (hydrogen) and carbon fixation pathway (Wood Ljungdahl pathway) (Chassard and Bernalier-Donadille, 2006). Due to a lower hydrogen threshold (minimum hydrogen concentration required for hydrogenotrophic growth) and a greater overall energy yield from the conversion of CO<sup>2</sup> to methane, methanogenic Archaea are usually the dominating group in this competition (Ragsdale and Pierce, 2008 and references therein). Moreover, acetogens (and methanogens) can be outcompeted by Bacteria with an even lower hydrogen threshold than methanogens, such as Campylobacterota, which are highly abundant at hydrothermal vent sites and take advantage of their versatile metabolisms (for details see below). Therefore, active acetogenic Bacteria are presumably less abundant in venting biotopes and have so far not been the focus of research related to hydrogen utilization in deep-sea hydrothermal vent environments.

Overall, sulfide and thiosulfate oxidation as well as hydrogen oxidation are among the chemosynthetic reactions which provide the greatest energy yields in hydrothermal vent biotopes (Amend and Shock, 2001; Fuchs et al., 2007). Although considerably more energy is yielded through oxidation of sulfide or thiosulfate than through hydrogen oxidation (free standard enthalpies are −797 kJ/mol H2S vs. −237 kJ/mol H<sup>2</sup> with O<sup>2</sup> as electron acceptor) (**Table 1**, Fuchs et al., 2007), the latter reaction is favorable for autotrophic carbon fixation. Since the redox-potential of hydrogen is more negative than that of the reducing equivalent NAD(P)/H, in contrast to sulfide, a reverse electron transport is not required in conjunction with hydrogen oxidation. Thus, only a third of the energy is required for fixing 1 mol of carbon when oxidizing hydrogen compared to sulfide (1060 kJ for hydrogen vs. 3500 kJ for sulfide) (Heijnen and Van Dijken, 1992). The individual fluid compositions of different hydrothermal systems may even increase this effect:


Electron acceptors are indicated by bold letters.

depending on hydrogen and sulfide concentrations as well as other abiotic factors, such as temperature and pressure, thermodynamic models for fluids of ultramafic vent fields predict that between 10 to 18 times more energy per kg of fluid can be yielded by hydrogen oxidation compared to sulfide oxidation (McCollom and Shock, 1997; McCollom, 2007; Petersen et al., 2011). The actual energy yields of the respective oxidation reactions strongly depend on the type of terminal electron acceptor used in the metabolism, where coupled to oxygen reduction the greatest energy amount is gained (**Table 1**, Conrad, 1996). Alternative electron acceptors commonly used by hydrogen-oxidizing microorganisms are sulfate, Fe (III) and nitrate (Vignais and Billoud, 2007), but also elemental sulfur and CO<sup>2</sup> as well as different metals, e.g., Mn (III/IV), U (VI), Cr (VI), Co (III) and Tc (VII), can be reduced by hydrogen-consumers (**Table 1**, Liu et al., 2002; Nakagawa and Takai, 2008). Due to mixing processes with oxygenated, ambient seawater, deep-sea hydrothermal fluids may contain numerous possible electron acceptors (primarily oxygen, nitrate, sulfate, elemental sulfur and iron). Their individual concentrations may vary strongly, depending on the geological setting of the vent system and the seawater mixing ratio.

Since covering all aspects of microbial hydrogen conversion at hydrothermal vents in detail would go beyond the scope of this review, we will here primarily focus on autotrophic hydrogen-oxidizers. Genes encoding hydrogen-oxidizing (or producing) enzymes have been identified via (meta-)genomic approaches in Alpha-, Beta-, Gamma-, and Deltaproteobacteria, Epsilonproteobacteria (in the following referred to as Campylobacterota as recently proposed by Waite, 2018), Firmicutes, Actinobacteria, Bacteroidetes, Aquificales and other, (less abundant) bacterial and also archaeal phyla in diverse habitats (cf. **Figure 2** and Greening et al., 2016). Consistent with the generally great abundance of Campylobacterota at hydrothermal vents (often constituting more than 90% of the microbial vent communities in incubation experiments or metagenomic studies) (e.g., Dahle et al., 2013; Perner et al., 2013a; McNichol et al., 2018), a large part of the hydrothermal vent-derived hydrogen oxidizing, autotrophic isolates are related to this class. They are characterized by versatile metabolisms and only a few isolates are strict hydrogen oxidizers (i.e., they are not capable of using any other tested organic or inorganic electron donor), such as the mesophilic Sulfurovum aggregans (Mino et al., 2014) or the thermophilic Caminibacter hydrogeniphilus (Alain et al., 2002). Overall, there is a trend in the use of alternative electron donors with respect to the thermal preferences: while thermophilic members of the order Nautiliales tend to use formate (e.g., Nagata et al., 2017), mesophilic Campylobacterota like Sulfurimonas paralvinellae have the ability to use different reduced sulfur species such as thiosulfate or elemental sulfur as energy sources (Takai et al., 2006). Based on their metabolic and physiological versatility, Campylobacterota occupy diverse niches and can dominate microbial communities in hydrothermal vent environments. The frequent isolation of H2-oxidizing Campylobacterota from deep-sea vents further emphasizes that this class may play a major role in hydrogen conversion and hydrogen-based primary production within hydrothermal habitats (Corre et al., 2001; Nakagawa et al., 2005; Campbell et al., 2006).

Hydrogen-oxidizing Deltaproteobacteria isolated from deep-sea vents – like Desulfonauticus submarinus – are commonly heterotrophic (Audiffrin et al., 2003), albeit representatives of this class were isolated from deep-sea vents that can couple hydrogen oxidation to autotrophic growth. Up to now, the vent-derived autotrophic, hydrogen-oxidizing Deltaproteobacteria are nearly all characterized as thermophiles with temperature optima between 50 and 61◦C, with the so far only exception being a Desulfobulbus species with a mesophilic temperature optimum (Sievert and Kuever, 2000; Slobodkin et al., 2013; Slobodkina et al., 2016; Han et al., unpublished). Notably, among them the thermophilic Desulfacinum hydrothermale belongs to the group of Sulfate Reducing Bacteria (SRB). Most members of the SRB (which are ubiquitously found in anoxic habitats) belong to the Deltaproteobacteria and the group of SRB is known for comprising autotrophs that couple the oxidation of hydrogen to the reduction of sulfate or other electron acceptors as thiosulfate or elemental sulfur (Sievert and Kuever, 2000; Muyzer and Stams, 2008). As the substrates for hydrogenotrophic growth of SRB are readily available in hydrothermal vent systems, deltaproteobacterial SRB may contribute to hydrogen consumption in anoxic hydrothermal vent habitats to a greater extent than previously thought. Still, this hypothesis needs to be proven by the continuing identification of such microorganisms. So far, further evidence for the presence of hydrogen-converting Deltaproteobacteria in hydrothermal fluids stems from metatranscriptomic data, where deltaproteobacterial genes encoding hydrogen-converting enzymes were identified (Fortunato and Huber, 2016). Gammaproteobacteria are also demonstrated to be relevant for microbially mediated hydrogen cycling. The gammaproteobacterial Thiomicrospira/Hydrogenovibrio/Thiomicrorhabdus genera (recently reclassified by Boden et al., 2017) can be significantly enriched in bacterial vent communities with relative abundances of up to 37% based on 16S rRNA sequencing (Brazelton and Baross, 2010; Brazelton et al., 2010; Perner et al., 2011a). For many years isolates of the Thiomicrospira group (some of which are regrouped into the Hydrogenovibrio genus) were described as conventional sulfur oxidizers (Brinkhoff et al., 1999; Takai et al., 2004a; Knittel et al., 2005), until the first sequenced genome of this group indicated genes encoding hydrogen-converting enzymes (Scott et al., 2006) and strains of this group were shown to use hydrogen (Hansen and Perner, 2015, 2016). Although other hydrogen-oxidizing, autotrophic, gammaproteobacterial isolates have not been recovered yet, there is evidence for the hydrogen-converting ability among this group, based on classical sulfur-oxidizing symbionts (Petersen et al., 2011) and widespread deep-sea bacteria of the SUP 05 clade (Anantharaman et al., 2016).

Besides members of the Proteobacteria, other Bacteria and also Archaea contribute to the hydrogen-oxidizing communities in deep-sea vents. Particularly (among the Bacteria) the deeply branching order of Aquificales hosts a wide range of hydrogen-oxidizing organisms of different families and genera (e.g., Desulfurobacteriaceae) that have been isolated from

FIGURE 2 | Phylogenetic relationship of "uptake" [NiFe]-hydrogenase large subunit structural genes. The phylogenetic tree was calculated for nucleotide sequences of the large subunit [NiFe]-hydrogenase genes of (primarily vent-derived) phylogenetically diverse Bacteria and Archaea. The scale bar denotes the number of substitutions per nucleotide position and bootstrap values are only indicated if greater than 80%. ClustalW alignments were performed prior to tree calculation using BioEdit (Hall, 1999) with the standard settings. The tree was calculated using seaview (Gouy et al., 2010) with maximum likelihood analysis (four rate classes) and bootstrap values were calculated with 100 replicates. The classification of the [NiFe]-hydrogenase genes was determined using the HydDB web tool (Sondergaard et al., 2016).

hydrothermal fields around the globe (L'Haridon et al., 2006; Ferrera et al., 2014). Although they differ in their specific growth requirements (e.g., temperature, pH, electron acceptors), they are all described as strict chemolithoautotrophs and thermophiles (Nakagawa et al., 2003; L'Haridon et al., 2006). The strictly anaerobic, vent-derived Desulfurobacterium thermolithotrophum from the family of the Desulfobacteraceae was the first known thermophilic bacterial isolate with the ability to act as a primary producer in the temperature range of 45–70◦C (L'Haridon et al., 1998). Notably, most members of the Aquificales, such as Thermovibrio ammonificans or Balnearium lithotrophicum, use hydrogen as the only energy source for autotrophic growth (Takai et al., 2003; Vetriani et al., 2004). Another thermophile-comprising phylum, Thermodesulfobacteria, usually is not considered as an important contributor of hydrogen-oxidizing vent-derived bacteria. Nonetheless, two of its five genera comprise thermophilic SRB isolated from hydrothermal vents, which share the ability to use hydrogen as the sole energy source for autotrophic growth (Jeanthon et al., 2002; Moussard et al., 2004; Alain et al., 2010).

Among the Archaea, thermophilic and hyperthermophilic methanogens are supposed to be the numerically largest and (in terms of the hydrogen consumption ability) most important group of hydrogen-oxidizers in hotter temperature regimes (Huber et al., 2002; Topcuoglu et al., 2016; Fortunato et al., 2018). Starting from the 1980's, shortly after the first discovery of hydrothermal vent systems, methanogenic isolates were repeatedly drawn from hydrothermal vents (e.g., Huber et al., 1982, 1989; Jones et al., 1989), some of them using hydrogen as the sole energy source for autotrophic CO<sup>2</sup> fixation and methane production (e.g., Huber et al., 1982; Jeanthon et al., 1998; L'Haridon et al., 2003; Takai et al., 2004b). As mentioned above, the hydrogen can stem from an abiogenic source (e.g., resulting from serpentinization processes) or be produced by hydrogen-evolving microorganisms (**Figure 1**, Ver Eecke et al., 2012; Toki et al., 2016; Topcuoglu et al., 2016). Methanogenic communities require greater hydrogen concentrations (e.g., ≥17 µM according to experiments with hyperthermophilic Methanocaldococcus species) to support chemolithoautotrophic growth than organisms coupling hydrogen oxidation to alternative electron acceptors, e.g., oxygen, nitrate, ferric iron and sulfate (Lovley and Goodwin, 1988; Hoehler et al., 1998; Ver Eecke et al., 2012). Besides methanogens, other hydrogen-oxidizing, autotrophs also exist among the Archaea: Fe(III)-reducing, hydrogen-oxidizing hyperthermophiles are encountered among the Euryarchaeota and Crenarchaeota (Slobodkina et al., 2009; Lin et al., 2016). These hydrogenotrophic Archaea thrive at lesser hydrogen concentrations than methanogens (Ver Eecke et al., 2009). Thus, they may be important contributors to microbial hydrogen consumption in venting environments that are characterized by elevated temperatures but lesser hydrogen levels. Nonetheless, the biogeochemical and ecological impact of these two groups still needs to be resolved.

Despite the large difficulties typically associated with taking samples from deep-sea hydrothermal vents and the culturing of vent-derived microorganisms, a large number of hydrogen-oxidizers has been isolated so far. However, a decreasing trend can be observed regarding the number of novel isolates from hydrothermal vent environments, which may be caused by insuperable obstacles in defining the appropriate culture conditions. More likely though, the laborious efforts in isolating (extremophilic) slow-growing microorganisms from hydrothermal vents have lessened due to the advent of cost-effective culture-independent techniques. For now, we have only gained a small-scale insight into the great diversity of microbial hydrogen uptake taking place at hydrothermal vents (see further discussions below).

#### HYDROGENASE GENES

The interconversion of molecular hydrogen to protons and electrons (H<sup>2</sup> ↔ 2H<sup>+</sup> + 2e−) is catalyzed by hydrogenase enzymes, which are widely distributed among Bacteria and Archaea. Hydrogenases are classified according to their catalytic center and to date three different types are known: (i) [NiFe]-hydrogenases, (ii) [FeFe]-hydrogenases and (iii) [Fe]-hydrogenases (Vignais and Billoud, 2007). [NiFe]-hydrogenases are usually involved in hydrogen sensing and consumption, [FeFe]-hydrogenases are the so-called "hydrogen-evolving" (producing) hydrogenases and [Fe]-hydrogenases play a key role in methanogenesis (Thauer, 1998; Vignais and Billoud, 2007). Among the [NiFe]-hydrogenases four groups are distinguished, that each can be further divided into several subgroups based on different parameters concerning the catalytic subunit like amino acid sequence phylogeny and reported biochemical properties. Group 1 and group 4 [NiFe]-hydrogenases are termed membrane-bound "H2-uptake" (consuming) and "hydrogen-evolving" hydrogenases, respectively, which are involved in energy metabolism. The group 2 encompasses mainly cytosolic hydrogen-sensing hydrogenases and some with so far unknown function and localization, while the cytosolic group 3 includes the F420-reducing hydrogenases from methanogens, the bifunctional NADP-coupled hydrogenases and the bifurcating, heterodisulphide-linked hydrogenases (Greening et al., 2016; Sondergaard et al., 2016).

The [FeFe]-hydrogenases can also be further distinguished in three groups (A-C), of which groups A and C are additionally subdivided into four and three subgroups, respectively. Notably, only group A1 hosts the prototypical "hydrogen-evolving" [FeFe]-hydrogenases (other group A hydrogenases are involved in electron bifurcation or have unknown functions). [FeFe]-hydrogenases of groups B and C are currently only assigned to putative functions involved in hydrogen sensing and hydrogen production (Sondergaard et al., 2016).

While [NiFe]- and [FeFe]-hydrogenases are present in diverse prokaryotes, [Fe]-hydrogenases are only found in methanogenic Archaea and cannot be subdivided into distinct groups (Greening et al., 2016; Sondergaard et al., 2016). In contrast to [NiFe]- and [FeFe]-hydrogensaes, they do not contain FeS-clusters and couple the oxidation of hydrogen to the reduction of methenyltetrahydromethanopterin.

This intermediary step is only required in the reduction of CO<sup>2</sup> to methane under nickel limiting conditions when [NiFe]-hydrogenases cannot be synthesized (Vignais and Billoud, 2007). Due to the habitat-specific conditions, "uptake" [NiFe]-hydrogenases (and primarily those of the prototypical group 1b, see **Figure 2**) likely are the most common and (concerning primary biomass production) important hydrogenase type in deep-sea hydrothermal vent systems. As hydrothermal fluids usually contain various minerals and metals, nickel (central component of the active center in [NiFe]-hydrogenases) limitation should not occur in these habitats. Moreover, given the elevated hydrogen concentrations (>µM levels), fermentative hydrogen production is likely limited or inhibited in hydrothermal vent systems, leading to the assumption that microbial hydrogen oxidation (catalyzed by [NiFe]-hydrogenases) may be the dominating process in these environments.

Hydrogenase genes (and those of [NiFe]-hydrogenases in particular) are usually arranged in gene clusters that differ in their size and gene patterns (**Figure 3**). Due to the highly specific and complex maturation processes involved in the biosynthesis of hydrogenases, the clusters (in addition to the catalytic subunits) commonly also comprise genes encoding proteins for electron transfer, regulation factors and maturation factors, but also hypothetical proteins and partner enzymes (Casalot and Rousset, 2001; Bock et al., 2006; Greening et al., 2016). Commonly, the heterologous expression of (hydrogenase) enzymes (i.e., the expression in a foreign host) is limited by promoter recognition, diverging codon-usage, translation and the incompatibility or a lack of the respective maturation and assembly apparatus (cf. Perner et al., 2011b). In E. coli for example, the exchange of a carboxy-terminal extension of the large subunit of a [NiFe]-hydrogenase with that from an isoenzyme resulted in the abortion of the protein maturation. This indicates the great specificity of the proteolytic cleavage by the endopeptidase HybD, which is a necessity to form an active hydrogenase (Theodoratou et al., 2000; Casalot and Rousset, 2001). Furthermore, nickel incorporation proteins (HypA) or the carbamoyltransferase HypF (involved in the formation of the active site) are of vital importance for the formation of a functional protein (**Figure 3**, Casalot and Rousset, 2001). Nevertheless, heterologous expression of [NiFe]-hydrogenases has successfully been demonstrated in the past: not only with genes of (phylogenetically) closely related organisms (Rousset et al., 1998) but also in a setup where the insert hydrogenase and the host stem from different bacterial classes (Adam and Perner, 2018).

Hydrogenase genes from hydrothermal vents have been targeted by PCR amplification (group 1 and F420-reducing [NiFe]-hydrogenases) (Takai et al., 2005; Perner et al., 2010; Petersen et al., 2011) or by direct sequencing of metagenomes (Perner et al., 2014; Pjevac et al., 2018) and metatranscriptomes (Dahle et al., 2013; Fortunato and Huber, 2016) (consisting of the whole genetic information) of vent-derived samples. However, compared to some enzymes like esterases, which are considered as one of the most important industrial biocatalysts (cf. Perner et al., 2011b), hydrogenases have only rarely been in the focus of metagenomic studies. Moreover, in most cases the metagenomic datasets were merely screened for the presence of [NiFe]-hydrogenase genes (e.g., Brazelton et al., 2012; Dahle et al., 2013; Perner et al., 2014; Fortunato and Huber, 2016). The majority of the identifiable [NiFe]-hydrogenase genes of these studies can be phylogenetically related to members of the Campylobacterota (see phylogenetic tree, **Figure 2**), other proteobacterial and also archaeal phyla. The frequent identification of Campylobacterota may be a consequence of the specific primer sets and the database entries that are available to identify hydrogenase genes. As only those genes can be identified that share sequence homologies to known hydrogenase sequences (or specific groups), it cannot be excluded that by applying PCR- and other sequence-based techniques an unintentional selection occurred. Still, the large number of campylobacterotal isolates also emphasizes the importance for primary biomass production and large abundance of this group in hydrothermal vent systems.

However, examples exist where no campylobacterotal genes could be identified: the [NiFe]-hydrogenase genes identified in the metagenome of a chimney sample from the hydrogen-rich, ultramafic Lost City hydrothermal field were primarily affiliated with betaproteobacterial [NiFe] hydrogenase genes, showing the greatest resemblance to the Ralstonia eutropha hydrogenase (Brazelton et al., 2012). Since R. eutropha and other Betaproteobacteria closely related to the hydrogenases found in the Lost City metagenome are aerobic or facultatively anaerobic, the authors assume that the respective hydrogen oxidizers thrive in the oxic-anoxic transition zone of the chimney. Surprisingly, in addition to these [NiFe]-hydrogenases, "hydrogen-evolving" [FeFe]-hydrogenases related to Clostridiales were identified, which are most likely associated with fermentation of organic compounds. Given that fermentative hydrogen production can be inhibited even at nM hydrogen concentrations (discussed earlier), microbial hydrogen production technically does not seem feasible in a hydrogen-rich vent environment like the Lost City hydrothermal field. Furthermore, the origin of the organic fermentation substrates has not been resolved yet. Still, there are indications that in the course of serpentinization processes abiogenic organic carbon might evolve that can be used in microbial fermentation. As most fermentation processes require anoxic conditions, it is proposed that anaerobic Clostridia colonize the anoxic, deepsubsurface of the Lost City vent system (eventually entrained to the surface as the fluids pass by). It may also be possible that the hydrogen produced by Clostridia is later consumed by the hydrogen oxidizing Betaproteobacteria (Brazelton et al., 2012 and references therein). Despite the metagenomic indications, clear evidence for hydrogen pathways (including the oxidation of microbially produced hydrogen) in this habitat is still missing (cf. **Figure 1**). So far, it remains unclear if the respective genes actually belong to living organisms and are transcribed and expressed as functioning enzymes. Recently, Pjevac et al. (2018) stated a large discrepancy in the hydrogenase frequency of metagenomes and metaproteomes of two distinct chimneys of the Roman Ruins vent field: From 160 phylogenetically diverse hydrogenase genes identified in the metagenomes only

five proteins were found in the respective metaproteomes, all belonging to campylobacterotal representatives. Accordingly, the great phylogenetic diversity (in general but also of the hydrogenase genes) does not coincide with the actual metabolic diversity of the microbial communities. This phenomenon may in part be explained by the fact that due to the great microbial diversity, protein quantities of hydrogenases probably lay below the detection threshold of the experimental setup. Additionally, many of the "uptake" hydrogenases are membrane-associated and a bias of the isolation method against such membrane-bound proteins has to be considered. Still, it cannot be excluded that a significant part of the hydrogenase genes present in the metagenome are not expressed and are therefore missing in the metaproteome (Pjevac et al., 2018).

A deepened insight into putatively active metabolic processes – and microbial hydrogen utilization – as well as possible regulating factors can also be gained by metaproteomic approaches (Anantharaman et al., 2013; Fortunato et al., 2018). A comparison of the metatranscriptomes of a plume and a background sample of the Guaymas basin demonstrated site-specific (up-regulated) transcript abundances of distinct [NiFe]-hydrogenase genes of plume-derived and epipelagic members of the sulfur-oxidizing SUP05 group of the Gammaproteobacteria. Combined with transcript abundances of other genes related to chemoautotrophy like a form II ribulose-1,5-bisphosphate carboxylase-oxygenase (RubisCO) and other genes of the Calvin-Benson-Bassham (CBB) cycle, the hydrogenase expression levels indicate that hydrogen oxidation strongly contributes to the energy budget of the SUP05 group thriving in deep-ocean habitats (Anantharaman et al., 2013). Gammaproteobacterial [NiFe]-hydrogenase genes and transcripts were – in addition to those of Deltaproteobacteria and Campylobacterota as well as the methanogenic F420-reducing hydrogenase - also detected in metagenomes and metatranscriptomes of low-temperature diffuse fluid samples derived from the Axial Seamount hydrothermal field (Fortunato and Huber, 2016). In this study meta-omics were coupled to additional, hydrogen-enriched, RNA stable isotope probing incubation experiments at different temperatures, integrating the influence of a thermal gradient on the chemoautotrophic microbial community. This thermal gradient was reflected by the hydrogen-oxidizing communities of the incubation experiments: while at 30◦C exclusively (mesophilic) Campylobacterota were found, at 55◦C thermophilic Campylobacterota and (to a lesser extent) methanogens dominated. The 80◦C incubations, however, were dominated by hyperthermophilic methanogens, indicating that methanogenesis was the main metabolism at this temperature and may play a significant role for primary production in subsurface habitats, characterized by greater temperatures and greater hydrogen concentrations (Fortunato and Huber, 2016). In a more recent metaproteomic study, Fortunato and co-workers compared the fluid communities of three hydrothermal vents of the Axial Seamount field, sampled on an annual basis over a period of 3 years. Fluids from Marker 33 and Marker 113 exhibited 10 to 30 times lesser hydrogen concentrations than that of the Anemone vent. However, the hydrogen-poor fluids exhibited greater abundances and expression levels of hydrogenase genes and a greater percentage of hydrogen-utilizing Campylobacterota, Aquificae and methanogens. Notably, methanogenic transcripts at Marker 113 ranged from 30 to 56% of all annotated transcripts, with a large portion of hydrogenase genes. Therefore, the lesser hydrogen concentrations are likely caused by a draw-down of hydrogen through microbial hydrogen oxidation. Overall, more than 90% of the intra- and inter-vent changes in the community compositions observed within this study could be explained with the geochemical variables determined for the different fluid, plume and background samples (e.g., temperature, pH, hydrogen, sulfide and nitrate concentrations) (Fortunato et al., 2018).

Yet, such clear-cut, proportional relations between abiotic environmental parameters and the corresponding microbial (metabolic) diversity are often difficult to draw. In particular, differing hydrogen-concentrations are often not directly reflected by the microbial community: varying hydrogen concentrations, for example, do not necessarily lead to differences in the diversity and abundance of hydrogenase genes. The hydrogenase distribution across differing hydrogen concentrations indicates that other environmental parameters also play a central role in the distribution of hydrogen oxidizing microorganisms (Perner et al., 2010, 2014). Other factors putatively influencing the diversity and abundance of hydrogenase genes observed in hydrothermal vent environments might be the kinetics and affinities of the respective enzymes. The K<sup>m</sup> values of [NiFe]-hydrogenases reported in the past show a great diversity ranging from 0.06 to 140 µM (Léger et al., 2004; van Haaster et al., 2005 and references therein). It may be assumed that organisms harboring high-affinity hydrogenases exhibiting low K<sup>m</sup> values can suppress hydrogen oxidizers that harbor hydrogenases with greater K<sup>m</sup> values, leading to a reduced diversity. However, the high-affinity, oxygen-tolerant [NiFe]-hydrogenases of group 1 h/5, which are widely distributed in soils (Constant et al., 2011), have not been identified in hydrothermal vent environments yet.

Furthermore, a metatranscriptomic study showed that increased hydrogenase gene expression is not limited to hydrothermal emission zones with elevated hydrogen concentrations but can also be observed at similar levels in intra-field water samples. The latter are not directly hydrothermally influenced but located in the vicinity of diffuse venting sites (Olins et al., 2017). Compared to background water samples, in most diffuse fluids and intra-field water samples the hydrogenase transcript levels were significantly enriched (Olins et al., 2017). The frequent identification of hydrogenase genes and elevated hydrogenase transcript abundances in hydrothermal vents and intra-field waters give evidence that hydrogen oxidation is of particular importance for primary biomass production in the different habitats surrounding hydrothermal vent orifices.

#### HYDROGEN CONSUMPTION MEASUREMENTS

Despite influences of individual fluid composition and seawater mixing ratios, compared to hydrogen-poor basalthosted systems, microbial hydrogen consumption rates of hydrogen-rich, ultramafic-hosted vent systems generally are expected to be greater. In fact, ex situ incubation experiments with symbiont-hosting mussel tissue from distinct vent systems revealed a 20- to 30-fold greater hydrogen consumption potential of symbionts from the hydrogen-rich ultramafic vent system relative to the hydrogen-poor basalt-hosted system (**Table 2**, Petersen et al., 2011). The respective CO2-fixation rates confirmed that hydrogen oxidation fueled autotrophy (**Table 2**, Petersen et al., 2011). Ex situ incubations with diverse hydrothermal fluids (and free-living microorganisms), however, could not confirm the thermodynamic estimates.



In most incubations, hydrogen consumption rates and biomass production were greater in the tested fluids from basaltic than from ultramafic systems. These observations may result from the specific conditions provided with the experimental setup, i.e., oxic and anoxic conditions, addition of 12–14 µM hydrogen (in solution) and incubation at 18◦C (Perner et al., 2010, 2011a, 2013b). Accordingly, altered incubation conditions may exhibit quite different hydrogen consumption rates. Similar incubation experiments, performed with only basalt-hosted hydrothermal emissions, were advanced by mimicking in situ pressure and temperature in gas-tight samplers (McNichol et al., 2016). Nitrate availability had a stimulating effect on the respective hydrogen consumption rates, ranging from 3.66 to 63.97 fmol H<sup>2</sup> cell−<sup>1</sup> h −1 (**Table 2**, McNichol et al., 2016), comparable to those of previous ex situ measurements ranging from 0.2 to 92.0 fmol H<sup>2</sup> cell−<sup>1</sup> h −1 (Perner et al., 2013b). Despite the efforts made to reproduce in situ conditions in ex situ incubations, it is impossible to simulate the dynamic nature of the (micro) habitats present in the hydrothermal vent systems. These are hallmarked by vast thermal and chemical gradients in venting habitats, ranging from several 100 s to 4 ◦C water temperature and from highly reduced to fully oxic, respectively. Therefore, incubations with more conditions than manageable would have to be set up to cover all the micro niches present in a hydrothermal venting biotope (Perner et al., 2010). Other methods to determine the microbial hydrogen oxidation potential, e.g., the tritium-based hydrogenase assay applied to subsurface sediments also show a great potential for hydrogen oxidation (Adhikari et al., 2016). Yet, they suffer from similar limitations as the hydrogen consumption measurements of hydrothermal vent samples: the incubation experiments are not conducted under in situ conditions and freezing of the samples prior to the assay cause additional deviation (Adhikari et al., 2016). Against this background, the development of in situ techniques for the determination of microbial hydrogen consumption rates is inevitable.

In situ measurements of hydrogen concentrations are already being done by employing in situ mass spectrometry (Wankel et al., 2011; Perner et al., 2013a) and has been used to draw conclusions on the impact of subsurface microbial activity on hydrogen concentrations of diffuse hydrothermal fluids. A discrepancy between the calculated and actually measured hydrogen concentrations of hydrothermal fluids, ranging from 50 to 80%, was attributed to microbial activity taking place below the seafloor (Wankel et al., 2011). Yet, a link to the microorganisms responsible for the presumable hydrogen consumption is missing. To provide this link, the existing measurement techniques could be amended by the recently established in situ fixation of fluids for later nucleotide extraction and metatranscriptomic and/or metagenomic analysis (Fortunato and Huber, 2016; Olins et al., 2017). Therefore, future in situ hydrogen measurements and consumption experiments would ideally combine monitoring of hydrogen and CO<sup>2</sup> concentrations, cell counting and fixation of (fluid) samples for metagenomic and metatranscriptomic analysis to cover the full hydrogen consumption potential of vent-associated microbial communities.

So far, ex situ hydrogen consumption measurements have been linked to unspecified Campylobacterota (McNichol et al., 2016), mesophilic Alpha-, Beta- and Gammaproteobacteria, mesophilic Campylobacterota, methanogens (Perner et al., 2010, 2011a) as well as a typically sulfur-oxidizing gammaproteobacterial symbiont (Petersen et al., 2011). First hints that another vent-inhabiting, sulfur-oxidizing Gammaproteobacterium might be able to oxidize hydrogen were gained from sequencing the hydrogenase gene cluster containing genome of Thiomicrospira crunogena (Scott et al., 2006) (now Hydrogenovibrio crunogenus) (Boden et al., 2017). Additionally, in some oxic, H2-amended incubation experiments, genes related to the sulfur-oxidizing gammaproteobacterial Hydrogenovibrio crunogenus were highly (8 to 23-fold) enriched compared to sulfide-spiked incubations of the same vent sample (Perner et al., 2011a). Still, it remained unclear whether Thiomicrospira strains actually express functional hydrogenases until Thiomicrospira SP-41's hydrogen consumption ability was discovered (Hansen and Perner, 2015). Further hydrogen consuming (previously classified as) Thiomicrospira strains were detected after offering diverse growth conditions and supplements (Hansen and Perner, 2016), demonstrating an unexpected potential for hydrogen consumption among these sulfur-oxidizing Gammaproteobacteria. The flexibility to use hydrogen as an alternative electron donor might also be a key to the success and dominance of (other) sulfur-oxidizing Gammaproteobacteria as observed in a variety of hydrothermal fluids (Perner et al., 2010; Olins et al., 2017).

A similar hydrogen consumption potential can also be expected for sulfur-oxidizing representatives of the order Campylobacterales: for example, the growth of a sulfur-oxidizing Sulfurimonas denitrificans isolate was significantly improved by the addition of hydrogen in growth experiments and hydrogen consumption measurements confirmed the utilization as electron donor (Han and Perner, 2014). Although S. denitrificans was originally isolated from Wadden Sea sediments (Timmer-Ten Hoor, 1975), numerous strains have also been identified in hydrothermally influenced habitats (cf e.g., Perner et al., 2013a). Furthermore, hydrogenase genes of a S. denitrificans strains were found in vent-derived metatranscriptomes (Fortunato and Huber, 2016). Given the great abundances of Sulfurimonas and other Campylobacterales genera like Arcobacter in hydrothermal fluids or plumes (Perner et al., 2010, 2013a; Fortunato and Huber, 2016), members of the Campylobacterales may contribute significantly to overall hydrogen consumption in deep-sea vent systems.

Although diverse archaeal hydrogen-consuming representatives have been isolated, much of the archaeal hydrogen consumption in hydrothermal vents can likely be assigned to methanogens, evidenced by incubation experiments and sequencing (Perner et al., 2010; Ver Eecke et al., 2012; Fortunato and Huber, 2016). The full potential of methanogenic hydrogen-based primary production, however, may be even greater than current incubation experiments suggest, occurring over a wider temperature and pressure range. By applying elevated (in situ) hydrostatic pressure of 20 MPa, the growth range of the hyperthermophilic vent-derived Methanopyrus

kandleri was expanded from 116◦C up to 122◦C and the temperature optimum was increased by 5◦C to 105◦C (compared to the standard 0.4 MPa incubations), while the carbon isotope fractionation of generated methane decreased (Takai et al., 2008). The small carbon isotope fractionation of biogenic methane could lead to a misinterpretation of data from hydrothermal vent environments: it could be identified as isotopically "heavy" methane from a magmatic source, thus diminishing the estimated methanogenic contribution (Takai et al., 2008 and references therein).

So far, incubation experiments with hydrothermal fluid samples have been performed with temperatures up to 80◦C (Fortunato and Huber, 2016), thus the conditions might not have been ideal for hyperthermophiles in the existing incubations and their abundances were underestimated.

Before genomic analyses and incubation experiments could link hydrogen consumption to the putatively responsible organisms, for many species such as Thiomicrospira sp. no hints for a potential hydrogen utilization were obvious. Matched with the still existing difficulties in the cultivation of vent inhabitants, a need for the implementation of culture-independent approaches becomes evident in order to identify novel hydrogen-oxidizing or -producing microorganisms and respective enzymes.

#### ACCESSING THE UNCULTURED MAJORITY AND THEIR HYDROGEN-CONVERTING POTENTIAL

Hydrogenase genes have been frequently identified in metagenomic deep-sea hydrothermal vent data sets. The [NiFe]-hydrogenase hit rate (i.e., the number of identified hydrogenase genes relative to all other genes in the data set) from a hydrothermal vent metagenome can be up to 40-fold higher than in metagenomes from other habitat types (Brazelton et al., 2012; Perner et al., 2014), revealing the importance of hydrogen-uptake in venting biotopes. However, these sequence-based approaches only indicate potential hydrogenase encoding genes. The functionality of the putative hydrogen-converting enzymes remains unclear, until hydrogen-uptake or –evolution is experimentally confirmed. Furthermore, hydrogen-converting enzymes lacking sequence homologies to known hydrogenases cannot be identified by sequence-based metagenomic approaches. Up to now, truly novel enzymes can only be found by screening metagenomes with activity-based approaches (Handelsman, 2004).

Until recently no activity-based screen existed, that could seek hydrogen-converting enzymes from the environment. However, a newly developed screen enables the search for environmental hydrogenases: It is based on the recombinant expression of metagenome-derived genes in a [NiFe]-hydrogenase deletion mutant of Shewanella oneidensis MR-1 (Adam and Perner, 2017). By applying this screen to metagenomic libraries of hydrothermal vent environments, hydrogen-converting clones were identified, whose metagenomic inserts largely do not share any sequence homology with known hydrogenases. Hydrogen-uptake activities of the clones exhibited up to

258 ± 19 nmol H<sup>2</sup> <sup>∗</sup>min−1∗mg−<sup>1</sup> of partially purified proteins at 55◦C, exceeding those of some cultured organisms (Adam and Perner, 2018). Given the difficulties and drawbacks associated with heterologous (hydrogenase) enzyme expression, a limitation in the hydrogenase detection ability of this screen is not surprising. For example, hydrogenases of Escherichia coli, Hydrogenovibrio sp., Thiobacillus denitrificans, Desulfovibrio vulgaris, and Aquifex aeolicus (and thus likely those of uncultured relatives from the environment) could not be identified with this host-vector system. Nevertheless, the [NiFe]-hydrogenases from Photobacterium leiognathi, Rhodobacter capsulatus, Sulfurimonas denitrificans and Wolinella succinogenes were successfully expressed and exhibited measurable activities that were up to 2.6-fold higher than that of the host's own hydrogenase (Adam and Perner, 2018). Still, it can be assumed that by establishing other hosts with varying growth optima for the activity-based screen, the detection range may be significantly improved. Given specific hydrogen-uptake activities of up to 48,700 ± 4,000 nmol H<sup>2</sup> <sup>∗</sup>min−1∗mg−<sup>1</sup> for the vent isolate Thioreductor micantisoli (Takai et al., 2005), the large potential for the discovery of (highly) active hydrogenases present in the enzymatic pool of vent environments is apparent.

The possibility of successfully expressing vent-derived hydrogen-converting enzymes in an "easily" culturable host may also open the door to biotechnological applications of these enzymes. Hydrogen-converting enzymes are of particular interest for the use in hydrogen production as a clean energy carrier and energy generation in biofuel cells (Armstrong et al., 2009; Chenevier et al., 2013). As shown for Escherichia coli's hydrogenases for example, these enzymes can be used for enhanced hydrogen production on surface-enlarged nanofiber electrodes (Schlicht et al., 2016). Due to the steep thermal and chemical gradients prevailing in vent environments, enzymes of exceptional stability under various conditions (e.g., temperature or oxygen contents) can be expected to be found. These would be the ideal candidates for biotechnological applications in the field of hydrogen production or energy generation in biofuel cells.

#### CONCLUSION

Hydrogen oxidation, catalyzed by phylogenetically diverse Bacteria and Archaea with versatile metabolic pathways, plays a major role for primary biomass production in chemically distinct deep-sea hydrothermal vent systems. However, the metabolic processes and biogeochemical interactions involved in hydrogen conversion are still not fully understood. Assessing the full hydrogen consumption potential of microbial vent communities has often proved to be difficult as incubation experiments but also metagenomic and metatranscriptomic approaches have their particular limitations: i.e., either in the reproducibility of optimal growth and hydrogen consumption conditions or in the lack of functional proof for the putative hydrogen conversion ability. The development of in situ hydrogen consumption measurement techniques that include sampling for subsequent molecular analyses would therefore considerably improve the exploration of hydrogen-converting communities in deep-sea vents. Since the culture-dependent and –independent approaches all exhibit individual limitations in identifying novel mechanisms of hydrogen-based metabolisms, the currently available techniques should ideally be combined to elucidate the full hydrogen utilization potential among the yet uncultured majority.

#### AUTHOR CONTRIBUTIONS

NA and MP wrote the manuscript.

fmicb-09-02873 November 21, 2018 Time: 19:51 # 14

#### REFERENCES


#### FUNDING

This work was supported by the research grant DFG PE1549-6/1 and PE1549-6/3 from the German Science Foundation.

#### ACKNOWLEDGMENTS

We greatly appreciate the funding of the DFG in the framework of the research grants PE1549-6/1 and PE1549-6/3.

chimneys. Proc. Natl. Acad. Sci. U.S.A. 107, 1612–1617. doi: 10.1073/pnas. 0905369107



chemolithoautotroph within the epsilonproteobacteria isolated from a deepsea hydrothermal vent chimney, and an emended description of the genus Sulfurovum. Int. J. Syst. Evolut. Microbiol. 64, 3195–3201. doi: 10.1099/ijs.0. 065094-0



**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 Adam and Perner. 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.

## Complex Multimeric [FeFe] Hydrogenases: Biochemistry, Physiology and New Opportunities for the Hydrogen Economy

Kai Schuchmann, Nilanjan Pal Chowdhury and Volker Müller\*

Molecular Microbiology and Bioenergetics, Institute of Molecular Biosciences, Johann Wolfgang Goethe University, Frankfurt am Main, Germany

Hydrogenases are key enzymes of the energy metabolism of many microorganisms. Especially in anoxic habitats where molecular hydrogen (H2) is an important intermediate, these enzymes are used to expel excess reducing power by reducing protons or they are used for the oxidation of H<sup>2</sup> as energy and electron source. Despite the fact that hydrogenases catalyze the simplest chemical reaction of reducing two protons with two electrons it turned out that they are often parts of multimeric enzyme complexes catalyzing complex chemical reactions with a multitude of functions in the metabolism. Recent findings revealed multimeric hydrogenases with so far unknown functions particularly in bacteria from the class Clostridia. The discovery of [FeFe] hydrogenases coupled to electron bifurcating subunits solved the enigma of how the otherwise highly endergonic reduction of the electron carrier ferredoxin can be carried out and how H<sup>2</sup> production from NADH is possible. Complexes of [FeFe] hydrogenases with formate dehydrogenases revealed a novel enzymatic coupling of the two electron carriers H<sup>2</sup> and formate. These novel hydrogenase enzyme complex could also contribute to biotechnological H<sup>2</sup> production and H<sup>2</sup> storage, both processes essential for an envisaged economy based on H<sup>2</sup> as energy carrier.

Keywords: hydrogenase, formate dehydrogenase, CO<sup>2</sup> reduction, electron bifurcation, hydrogen production, acetogenesis, clostridia, hydrogen storage

#### INTRODUCTION

Molecular hydrogen (H2) is only present in trace concentrations (550 parts per billion) in the Earth's lower atmosphere (Novelli et al., 1999). Nevertheless, it plays an essential part in the biogeochemical cycles of other elements such as carbon and is a major constituent of the microbial metabolism. For example, H<sup>2</sup> is an important electron donor for methane formation in anoxic environments (Schink, 1997; Thauer et al., 2008). Here, too, steady state concentrations are very low (pH<sup>2</sup> < 10 Pa) but the turnover of H<sup>2</sup> is very high (150 million tons of H<sup>2</sup> of biological origin are estimated to be produced in anoxic ecosystem annually to fuel methanogenesis) (Thauer et al., 2008, 2010). In anoxic ecosystems, the major role of H<sup>2</sup> is electron transfer between the different participants of the food chain, e.g., transfer of electrons generated by primary fermenters to methanogens (Stams and Plugge, 2009; Schink et al., 2017). To produce or consume H<sup>2</sup> nature has evolved complex metalloenzymes, hydrogenases, which catalyze one of the simplest chemical reaction, reversible

Edited by:

Chris Greening, Monash University, Australia

#### Reviewed by:

Armen Trchounian, Yerevan State University, Armenia Kyle Costa, University of Minnesota Twin Cities, United States

> \*Correspondence: Volker Müller vmueller@bio.uni-frankfurt.de

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 30 August 2018 Accepted: 13 November 2018 Published: 04 December 2018

#### Citation:

Schuchmann K, Chowdhury NP and Müller V (2018) Complex Multimeric [FeFe] Hydrogenases: Biochemistry, Physiology and New Opportunities for the Hydrogen Economy. Front. Microbiol. 9:2911. doi: 10.3389/fmicb.2018.02911 oxidation of H<sup>2</sup> into two protons and two electrons:

fmicb-09-02911 November 30, 2018 Time: 15:13 # 2

$$H\_2 \rightleftharpoons 2H^+ + 2e^- \tag{1}$$

Hydrogenases are widespread in nature and can be found in all domains of life. Based on their phylogeny, they can be classified into three distinct classes that are named by the metal ions contained in their active sites as [NiFe]-, [FeFe]-, and [Fe] hydrogenases (Vignais et al., 2001; Vignais and Colbeau, 2004). [NiFe] hydrogenases have been found in bacteria and archaea, [FeFe] hydrogenases in bacteria and some eukaryotes, and [Fe] hydrogenases only in archaea (Vignais and Billoud, 2007). Even though [NiFe]- and [FeFe] hydrogenases have evolved independently, the complex metal centers responsible for catalysis share many features. The metal ions are ligated by inorganic CO and CN<sup>−</sup> ligands and are bridged by sulfur atoms. H<sup>2</sup> can reach the active sites that are buried within the enzymes by a hydrophobic gas channel and the electrons that result from H<sup>2</sup> oxidation are in both enzyme classes transferred to an [4Fe– 4S] cluster adjacent to the metal center (Happe et al., 1997; Pierik et al., 1998; Fontecilla-Camps and Ragsdale, 1999). The third class of hydrogenases, the [Fe] hydrogenases which are only found in methanogenic archaea differ not only by their architecture of the active site, but also by the catalyzed reaction that does not result in the release of electrons to an iron–sulfur cluster, but the direct transfer to the cosubstrate methenyltetrahydromethanopterin (Shima and Thauer, 2007; Shima et al., 2008). In this review, we will mainly focus on hydrogenases of the [FeFe] class. Enzymes from this class are widespread in anaerobic prokaryotes and play important roles in the energy and carbon metabolism in anoxic ecosystems. Crystal structures have been reported for three enzymes, namely from Clostridium pasteurianum (Peters et al., 1998), Desulfovibrio desulfuricans (Nicolet et al., 1999) and the eukaryotic algae Chlamydomonas reinhardtii (Mulder et al., 2010) and show high similarity in the overall structure and the architecture of the active site. Details on the crystal structures and the reaction mechanism have been described in excellent and comprehensive reviews elsewhere and will therefore not been repeated here except for the architecture of the cofactors in the C. pasteurianum enzyme that we will discuss in the context of electron bifurcation later (Vignais and Colbeau, 2004; Thauer et al., 2010; Lubitz et al., 2014). Interestingly, despite the apparent high similarity of the "core" hydrogenase subunit responsible for H<sup>2</sup> oxidation and production, [FeFe] hydrogenases show a remarkable diversity with respect to the auxiliary subunits that can be found in most enzymes and an even higher diversity can be predicted from genome sequence data. The auxiliary subunits follow a very modular structure and add multiple functions to the core subunit. These functions can include electron transfer to soluble electron carriers, coupling of H<sup>2</sup> oxidation/production to other chemical reactions, coupling to energy conservation by coupling the electron transfer to the generation of a transmembrane ion potential, or utilization of the novel energetic coupling mechanism of flavin-based electron bifurcation (FBEB) to overcome energetic limitations of the electron transfer (Buckel and Thauer, 2018a,b; Müller et al., 2018). A notable diversity of multimeric [FeFe] hydrogenases can be found especially in the strictly anaerobic Gram-positive bacteria of the class Clostridia within the phylum Firmicutes (Calusinska et al., 2010; Schmidt et al., 2010). Recent discoveries have revealed multimeric hydrogenases with remarkable and so far undescribed functions and catalytic properties, solving important questions of the metabolism and ecosystem functioning in anoxic environment. In addition, recently described multimeric hydrogenases provide interesting opportunities for biotechnological applications. H<sup>2</sup> is a candidate energy carrier that could replace fossil fuels for storage and transportation of energy generated from renewable sources such as wind or solar power (Dunn, 2002; Brandon and Kurban, 2017). However, for an economically viable H<sup>2</sup> economy many obstacles need to be overcome such as efficient methods for H<sup>2</sup> production and technologies for storage and transportation of the volatile and explosive gas (Service, 2004; Preuster et al., 2017a).

Here, we review the recent findings of novel complex multimeric hydrogenases and especially their function in the microbial H<sup>2</sup> metabolism. These include enzymes using the novel energy coupling mechanism of FBEB enabling otherwise endergonic H<sup>2</sup> production from NADH or the otherwise endergonic electron transfer from H<sup>2</sup> to the iron–sulfur protein ferredoxin. The second part will focus on the recently discovered formate dehydrogenase coupled hydrogenases enabling direct CO<sup>2</sup> reduction with H<sup>2</sup> or H<sup>2</sup> evolution from formate. In addition, both hydrogenase types will be discussed in the context of their biotechnological potential for the H<sup>2</sup> economy.

#### H<sup>2</sup> FUNCTIONING AS ELECTRON CARRIER IN ANOXIC ECOSYSTEMS

H<sup>2</sup> can be utilized by many organisms as electron donor. Recently, it has been reported that microorganisms are even capable of utilizing the low atmospheric concentrations of H<sup>2</sup> (Greening et al., 2014a,b). Nevertheless, the most prominent functions of H<sup>2</sup> are found in anoxic ecosystems where it is rapidly produced and consumed by microorganisms resulting in a large turnover. H<sup>2</sup> connects different parts of the food web to allow full remineralization of organic material. When complex polymeric organic material (polysaccharides, proteins, nucleic acids, lipids) enter anoxic ecosystems it is typically first hydrolyzed by exoenzymes followed by the partial oxidation by primary fermenting microorganisms into different fermentation products such as lactate, alcohols, short chain fatty acids, acetate, formate, CO<sup>2</sup> and H2. In the absence of suitable external electron acceptors, the latter compounds (acetate, formate, CO<sup>2</sup> and H2) can be directly converted by methanogenic archaea into methane (Schink, 1997; Thauer et al., 2008). Further oxidation of the other compounds leads to an energetic problem that can only be solved by the concerted cooperation of secondary fermenting organisms with methanogens in a process call syntrophy (Schink, 2002; Sieber et al., 2012; Morris et al., 2013). Secondary fermenters oxidize their substrates typically to acetate as end product coupled to oxidation of protons to H<sup>2</sup> to reoxidize their electron carriers. However, under standard conditions these reactions are endergonic and do not provide

energy for the cell. During substrate oxidation, electrons are typically transferred to either ferredoxin, a small iron–sulfur cluster containing protein with a very negative redox potential (E◦ ' = −400 to −500 mV), or NAD<sup>+</sup> with a more positive redox potential (E◦ '[NAD+/NADH] = −320 mV) (Thauer et al., 1977). Proton reduction can be used to recycle these electron carriers. The redox potential of the H+/H<sup>2</sup> couple is, under standard conditions, −414 mV. Therefore, while H<sup>2</sup> production from reduced ferredoxin is an exergonic reaction, H<sup>2</sup> formation from NADH represents a strong energetic barrier for the cells since it is a highly endergonic reaction (Thauer et al., 1977). Two possible mechanisms have evolved to overcome this problem and allow H<sup>2</sup> formation from NADH. In the classical view, H<sup>2</sup> oxidizing methanogens lower the H<sup>2</sup> partial pressure to very low values (1–10 Pa H2) resulting in a more positive redox potential of the H+/H<sup>2</sup> couple (above −300 mV) thus rendering H<sup>2</sup> formation from NADH an exergonic reaction (Schink, 2002; Sieber et al., 2012; Morris et al., 2013). Recently, another mechanism has been discovered that solves the problem of NADH reoxidation with protons as acceptor within one enzyme. Complex multimeric hydrogenases are necessary to catalyze this reaction that solves the energetic problem by energetically coupling the endergonic electron transfer to a second exergonic redox reaction, a process called FBEB. As we will see later, these enzymes not only participate in the metabolism of secondary fermenting organisms but can also be found in primary fermenters, thereby increasing the energy yield that can be conserved from a given substrate, or, in the reverse, be found in acetogenic bacteria to allow ferredoxin reduction with H<sup>2</sup> as electron donor.

#### THE CONCEPT OF FLAVIN-BASED ELECTRON BIFURCATION

In 2008 a new energy coupling mechanism called FBEB was first discovered in an enzyme complex of an electrontransferring flavoprotein and a butyryl-CoA dehydrogenase (Etf/Bcd) (Herrmann et al., 2008; Li et al., 2008). In FBEB, an electron pair from an electron donor such as NADH is split toward two different one-electron acceptors, one with a more positive redox potential and another with a much lower redox potential than that of the electron donor. It was proposed and later proven that electron transfer to a positive redox potential (exergonic reaction) sustains the movement of an electron to a more negative redox potential (endergonic reaction). In case of Etf/Bcd, the positive electron redox potential acceptor was crotonyl-CoA (E<sup>0</sup> ' = −10 mV) and the negative redox potential acceptor was ferredoxin (E<sup>0</sup> ' = −420 mV). NADH (E<sup>0</sup> ' = −320 mV) is the electron donor. The complete reaction catalyzed by the protein complex is

$$\text{Crotonyl} - \text{CoA} + 2 \text{ NADH} + \text{Fd}\_{\text{ox}}$$

$$\rightarrow \quad \text{Butryyl} - \text{CoA} + 2 \text{ NAD}^+ + \text{Fd}\_{\text{red}} \tag{2}$$

The enzyme complex contained FAD as the only cofactor that was essential for the activity (Chowdhury et al., 2014; Demmer et al., 2017). Hence, the name originated as FBEB. FBEB was drawn in analogy to the quinone-based electron bifurcation (QBEB) of the cytochrome bc<sup>1</sup> complex of the respiratory chain, which was discovered 43 years ago by Peter Mitchell, in which the oxidation of reduced ubiquinone (UQH2) by the high potential cytochrome c<sup>1</sup> by one electron allows the reduction of lowpotential cytochrome b<sup>L</sup> and further UQ inside the membrane (Mitchell, 1975). The process is repeated twice that allows four protons to be released outside of the cell and doubling the amount of energy conserved and electrons finally flow down to oxygen (terminal electron acceptor) to reduce oxygen to water.

#### H<sup>2</sup> PRODUCTION FROM NADH: ELECTRON-BIFURCATING HYDROGENASES FOR H<sup>2</sup> EVOLUTION

Soon after the discovery of FBEB in Clostridia, the first hydrogenase was reported that utilizes the mechanism of FBEB, however, in the physiological context in the reverse direction (named electron confurcation) (Schut and Adams, 2009). The enzyme was discovered in the hyperthermophilic and anaerobic bacterium Thermotoga maritima. The bacterium ferments one mole of glucose by the classical Embden–Meyerhof–Parnas pathway to two moles of CO2, two acetate and four moles of H<sup>2</sup> (Schroder et al., 1994). During its metabolism, both NADH and reduced ferredoxin are generated, however, for decades the link to the oxidation of these electron carriers to H<sup>2</sup> production remained obscure (Wrba et al., 1990; Blamey and Adams, 1993). The trimeric [FeFe] hydrogenase was isolated and could be assayed by coupling reduction of viologen dyes with H2. However, the enzyme did not use either reduced ferredoxin or NADH as sole electron donor (Verhagen et al., 1999). Though reduced clostridial ferredoxin (E<sup>0</sup> ' = −420 mV) alone can reduce protons to H<sup>2</sup> (E<sup>0</sup> ' = −414 mV), NADH cannot. It was an enigma since the discovery of fermentative H<sup>2</sup> production to how H<sup>2</sup> is produced from NADH. The solution was FBEB: exergonic electron flow from reduced ferredoxin to H<sup>+</sup> that drives endergonic electron flow from NADH to H+, according to:

$$\text{NADH} + \text{Fd}\_{\text{red}} + 3\text{ H}^+ \rightleftharpoons \text{NAD}^+ + \text{Fd}\_{\text{ox}} + 2\text{ H}\_2 \quad \text{(3)}$$

The hydrogenase of T. maritima is now the classic example where reduced ferredoxin drives H<sup>2</sup> evolution from NADH. The enzyme oxidizes NADH and ferredoxin simultaneously in a 1:1 ratio to produce H2. In this case both electrons from NADH and reduced ferredoxin are converged to reduce protons to H2. This mode of electron converging from different sources of electron donor (NADH and Fdred) to a single electron acceptor (protons) is now called electron confurcation. A similar lifestyle and metabolic pathway is also observed in the rumen bacterium Ruminococcus albus (Zheng et al., 2014). Similar to T. maritima, when grown in continuous culture the bacterium produces the same amounts of H<sup>2</sup> and acetate from glucose (**Figure 1A**). Also, the acetogenic model bacterium Acetobacterium woodii is suggested to use a confurcating hydrogenase to evolve H<sup>2</sup> from organic substrates (Schuchmann and Müller, 2012; Bertsch et al., 2015; Kremp et al., 2018). This mode of H<sup>2</sup> evolution

by electron confurcating hydrogenases is important for energy conservation in these fermenting bacteria. ATP is synthesized by substrate level phosphorylation at the phosphoglycerate kinase and pyruvate kinase reactions and NADH is formed by NADspecific glycerinaldehyde-3-phosphate-dehydrogenase.The redox pool could for example be balanced by reducing pyruvate to lactate, where 2 NADH will be consumed. However, this pathway will produce 2 ATP less. Rather the bacteria maintain its redox balance by producing two moles of acetyl-CoA via the pyruvateferredoxin oxidoreductase and finally releases 2 ATP catalyzed by the acetate kinase. This mode of metabolism leaves 2 NADH and 2 reduced ferredoxin that are converted to H<sup>2</sup> by FBEB hydrogenase. Hence, the presence of FBEB hydrogenases allows NADH and ferredoxin to be reoxidized with H<sup>+</sup> as electron acceptor and thus increasing the ATP yield (Müller et al., 2018).

#### ELECTRON-BIFURCATING UPTAKE HYDROGENASES

In the autotrophic mode of life, organisms can synthesize most of their carbon compounds from CO<sup>2</sup> using H<sup>2</sup> or another electron donor as reductant. One such group are the acetogenic bacteria that are capable of producing acetate from two molecules of CO<sup>2</sup> with H<sup>2</sup> as the reductant via the reductive acetyl-CoA pathway (also known as Wood–Ljungdahl pathway, WLP) (Wood, 1991; Wood and Ljungdahl, 1991). Three model organisms that have been studied in detail are Moorella thermoacetica, A. woodii, and Clostridium ljungdahlii. As these bacteria grow simply on CO<sup>2</sup> and H<sup>2</sup> without any external carbon source added, this mode of metabolism must be coupled to net ATP formation.

A. woodii as a model organism has been studied in detail to answer how energy is conserved during acetogenesis (Schuchmann and Müller, 2014). This unearthed surprising enzyme complexes that are unique in their diversity of the reactions they catalyze as well as their working mechanisms. Three such enzyme complexes are the Rnf complex, an FBEB hydrogenase and a H2-dependent CO<sup>2</sup> reductase. A. woodii employs a sodium ion-dependent ferredoxin: NADoxidoreductase (Rnf complex) that couples the exergonic oxidation of reduced ferredoxin (E<sup>0</sup> ' ∼ −450 mV) with NAD<sup>+</sup> (E<sup>0</sup> ' = −320 mV) to the generation of a transmembrane electrochemical Na<sup>+</sup> gradient (**Figure 1B**). The difference in the redox potential of ferredoxin and NAD+/NADH allows the pumping of Na<sup>+</sup> over the cytoplasmic membrane resulting in a Na<sup>+</sup> gradient that then drives the synthesis of ATP via the well-characterized Na<sup>+</sup> F1FO-ATP synthase (Spruth et al., 1995). Therefore, to drive energy conservation catalyzed by the Rnf complex, ferredoxin needs to be reduced. A. woodii solves this problem by using a FBEB uptake hydrogenase (Schuchmann and Müller, 2012) which is quite similar to the electron confurcating hydrogenases from T. maritima or R. albus (Zheng et al., 2014).

HydD. The arrangement of the iron–sulfur clusters and the binding site for ferredoxin are chosen arbitrary. Fd, ferredoxin.

However, in case of chemolitotrophic growth, H<sup>2</sup> is oxidized to reduce both NAD<sup>+</sup> and ferredoxin. The hydrogenase couples the exergonic oxidation of H<sup>2</sup> (E<sup>0</sup> ' = −414 mV) with NAD<sup>+</sup> (E<sup>0</sup> ' = −320 mV) to the endergonic reduction of ferredoxin (E<sup>0</sup> ' ∼ −450 mV). Reduced ferredoxin is used in two different reactions, one to reduce CO<sup>2</sup> to CO (E<sup>0</sup> ' = −520 mV) in the WLP and secondly to transfer electrons via the Rnf to NAD+, resulting in a Na<sup>+</sup> gradient (Biegel and Müller, 2010, 2011; Biegel et al., 2011). However, reaching the very low reduction potential for CO<sup>2</sup> to CO reduction is difficult even at very high H<sup>2</sup> partial pressure (10<sup>5</sup> Pa). FBEB provides an elegant solution by providing reduced ferredoxin with a more negative redox potential than the initial electron donor H2. The same is true when A. woodii is grown on methanol, where FBEB hydrogenase provides the extra reduced ferredoxin which is then used for CO<sup>2</sup> reduction (Kremp et al., 2018).

The hydrogenase of A. woodii has been studied in the context of the autotrophic metabolism where its function is H<sup>2</sup> oxidation. Besides, further insights into the non-autotrophic metabolism of A. woodii have revealed that the enzyme may function as a H2-evolving (electron bifurcating) hydrogenase as well. For example, when A. woodii grows on ethanol, only NADH is produced when ethanol is oxidized to acetyl-CoA. One part of the NADH is oxidized at the Rnf complex reducing ferredoxin. This reduced ferredoxin is then used with the other part of NADH by the hydrogenase to produce H<sup>2</sup> for the first step of the WLP (**Figure 1B**) (Bertsch et al., 2015). So, the electronbifurcating/confurcating hydrogenase from A. woodii provides a nice example how anaerobes have evolved their metabolic enzymes which serves the purpose of both uptake and evolving H<sup>2</sup> when needed in two different modes of energy metabolism.

#### COFACTOR AND SUBUNIT ARCHITECTURE OF FBEB HYDROGENASES

The H<sup>2</sup> forming hydrogenases from T. maritima, R. albus, and M. thermoacetica (Wang et al., 2013b) and the H<sup>2</sup> uptake/forming hydrogenase from A. woodii possess quite similar subunit compositions (**Figure 2**). All four enzymes are composed of the three subunits Hyd A (∼64 kDa), Hyd B (∼65 kDa) and Hyd C (∼14 kDa). HydA from T. maritima is larger compared to its counterpart by having a size of 73 kDa putatively containing

an additional [2Fe–2S] cluster. Another exception is that the A. woodii hydrogenase has an extra subunit HydD (∼15 kDa) which is predicted to contain no cofactor (Schuchmann and Müller, 2012, 2014). On amino acid sequence comparison, HydA finds the closest similarity to the monomeric [FeFe] hydrogenase from C. pasteurianum (**Figure 3**). The first crystal structure of the [FeFe] hydrogenase from C. pasteurianum CpI was reported by Peters et al. (1998). The overall structure of the core domain consists of the H-cluster (the active site of [FeFe] hydrogenases catalyzing H<sup>2</sup> oxidation) including a diiron subcluster and one [4Fe–4S] cluster connected via a cysteine residue as found conserved in most of the [FeFe] hydrogenases. The diiron metals are coordinated by CN<sup>−</sup> and CO, while the proximal Fe is linked to a cubane [4Fe–4S] cluster via a cysteine. The [4Fe– 4S] is around 4 Å apart from the di-iron center. Apart from the H2-activating domain the domain interacting with the activesite domain contains two [4Fe–4S] clusters named as FS4A and FS4B (**Figure 3**). FS4A is 9 Å apart from the H-cluster thus the direct electron carrier to or from the H-cluster. FS4A is around 10 Å apart from FS4B cluster and thus in line to the putative electron transfer pathway. There are two additional domains, one containing a [2Fe–2S] cluster named FS2 (11 Å from FS4B), the other contains a single [4Fe–4S] cluster called FS4C. FS4C has an unusual three cysteine and one histidine ligation. The two clusters FS4C and FS2 exhibit a forked architecture, where FS2 is far away from the direct line of electron transfer from FS4C. Most probably, FS4C directly gets electron from FS4B and finally transfers electrons to ferredoxin. A recent study using protein–protein docking modeling and NMR studies of electron transfer complex formation between the photosynthetic electron-transfer ferredoxin (PetF) containing a [2Fe–2S] cluster and the hydrogenase HydA1 from the microalga C. reinhardtii revealed PetF to interact with HydA1 near to FS4C (Chang et al., 2007; Rumpel et al., 2015). The function of FS2 and the forked orientation of the possible electron transfer chains is puzzling, especially in the context of an electron bifurcating enzyme.

Assuming a similar structural organization and cofactor content in the FBEB hydrogenases, HydA is predicted as the catalytic subunit for H<sup>2</sup> oxidation. It contains the H-cluster which is the site for H<sup>2</sup> activation, one [2Fe–2S] and three [4Fe– 4S] clusters. HydB has closest similarity to the NADH binding subunit NuoF of NADH-quinone oxidoreductase from E. coli, and is predicted to contain one FMN, one [2Fe–2S] and three [4Fe–4S] clusters. HydC, which is related to NuoE, contains only one [2Fe–2S] cluster.

#### A POSSIBLE ELECTRON PATHWAY IN FBEB HYDROGENASES

How can the electron flow from H<sup>2</sup> to NAD<sup>+</sup> and H<sup>2</sup> to ferredoxin be energetically coupled within these multimeric hydrogenases? Until now the basis of FBEB has been revealed in other enzyme complexes like the Etf/Bcd complex from Acidaminococcus fermentans (Chowdhury et al., 2014, 2016), C. difficile (Demmer et al., 2017) and Megasphaera elsdenii (Chowdhury et al., 2015), LctBCD and CarCDE of A. woodii (Bertsch et al., 2013; Weghoff et al., 2015) and the Nfn transhydrogenase from T. maritima (Demmer et al., 2015), Pyrococcus furiosus (Lubner et al., 2017), FixABCX of Azotobacter vinelandii (Ledbetter et al., 2017) and HdrABC-MvhAGD from the thermophilic methanogenic archaeon Methanothermococcus thermolithotrophicus (Wagner et al., 2017). However, FBEB hydrogenases represent a very special case. FBEB enzymes so far all are proposed to contain a flavin having special redox properties. Flavins have three different redox potentials for the

FIGURE 4 | Models for the electron flow in electron bifurcating hydrogenases. In the scenario that electron-bifurcation is facilitated by a flavin (A) according to the standard model of FBEB, the enzymes would have to contain two additional flavins that cannot be predicted from the primary structure. Flavin α functions as switch from one electron transferring iron–sulfur clusters to the actual bifurcating flavin β. Here, the two electrons are bifurcated to other iron–sulfur clusters and to the final acceptors. Flavin γ functions as switch from the one electron transferring iron–sulfur clusters to NAD+. In the second scenario (B). The H-cluster is the site of electron bifurcation. Here only one flavin is required for the switch from the one electron transferring iron–sulfur clusters to NAD+. Fd, ferredoxin.

three possible redox reactions:

$$\text{FMN} + 1 \text{ } e^- \rightleftharpoons \text{FMN}^{\circ-} \quad \text{(E}\_1\text{)}\tag{4}$$

$$\text{FMN}^{\circ}{}^{\circ} + 1 \text{ } e^{-} + 2 \text{ } H^{+} \rightleftharpoons \text{FMNH}\_{2} \quad \text{(E}\_{2}\text{)}\tag{5}$$

$$\text{FMN} + 2\text{ }e^- + 2\text{ }H^+ \rightleftharpoons \text{FMNH}\_2 \quad \text{(E3)}\tag{6}$$

In "standard" flavins the midpoint potential of E<sup>1</sup> is more positive than E2. In FBEB enzymes the flavin is supposed to have a "crossed" redox potential meaning that E<sup>1</sup> is more negative than E<sup>2</sup> (Nitschke and Russell, 2012). The two-electron redox potential (E3) is supposed to be between, meaning the average, of E<sup>1</sup> and E2. Therefore, the flavin can be reduced in a two-electron transfer reaction by the electron donor (E3) followed by the first electron being transferred to the more positive electron acceptor (e.g., NAD+) (E2) leaving behind a highly reactive FMN◦<sup>−</sup> that can now transfer the second electron to the more negative electron acceptor (e.g., ferredoxin) (E1) (Demmer et al., 2017; Baymann et al., 2018; Buckel and Thauer, 2018a).

Based on the current knowledge, FBEB hydrogenases have only one predicted flavin in HydB which is required for the switch from a two-electron carrier (NADH) to a one electron carrier (iron–sulfur cluster), a typical function of "standard" flavins. Therefore, Buckel and Thauer (2013) proposed that FBEB hydrogenases need to have an additional "special" flavin to perform the electron bifurcation reaction. To function in

accordance to the standard model of FBEB, the electron flow would look like the following (**Figure 4A**): H<sup>2</sup> is oxidized at the H-cluster followed by the electron transfer to the iron–sulfur clusters. H<sup>2</sup> has a redox potential in between NAD<sup>+</sup> and ferredoxin, therefore, it must reduce the flavin with a two-electron transfer reaction to reach the flavin redox potential E3. However, since iron sulfur clusters can only transfer one electron at a time we therefore speculate that in this model a second flavin (flavin α) would be required for the additional one electron/two electron switch. Then, the "special" flavin (flavin β) can transfer the first electron to the iron sulfur cluster leading to NAD<sup>+</sup> leaving behind a highly reactive flavin radical that transfers the second electron to the iron–sulfur clusters leading to ferredoxin. The third flavin (flavin γ) is then needed for the second one electron/two electron switch from the iron sulfur clusters to NAD+.

Of the characterized FBEB enzymes HdrABC-MvhAGD from M. thermolithotrophicus, a complex of a multimeric heterodisulfide reductase and a [NiFe] hydrogenase represents a special case (Wagner et al., 2017). The flavin proposed to be responsible for electron bifurcation is assumed to receive both electrons from H<sup>2</sup> in two single electron transfer steps in contrast to a two electron transfer from a hydride donor. This is in contrast to the current model of the energetic landscape required for electron bifurcating flavins. However, if electron bifurcation is possible by a flavin connected to three one electron donors/acceptors in the form of iron–sulfur clusters, this could also be the case in FBEB [FeFe] hydrogenases and would render the presence of the hypothetically proposed flavin α redundant.

The aforementioned model assumes the presence of additional flavin binding sites that are not predicted by the amino acid sequence. The flavin content of all isolated FBEB hydrogenases could not be determined or not reported since the flavins are only loosely attached and must be added to the buffers for activity and got immediately lost when left out of the buffers. Assuming that the additional flavins do not exist, a completely new mechanism of FBEB must be present in FBEB hydrogenases. First Nitschke and Russell (2012) raised the possibility that also some metal centers could have crossed redox potentials and could potentially catalyze the same reaction as catalyzed by the flavin cofactor. Peters et al. (2018) took up this exciting possibility and proposed for electron bifurcating hydrogenases that since the FMN cannot be the site for electron bifurcation, the H-cluster could be the possible site for the electron bifurcation reaction. In their proposal, the electrons flow from the H-cluster directly to two different accepting iron–sulfur clusters with different redox potentials. The first string of Fe–S clusters transfers the electrons from the H-cluster to ferredoxin. The other string of iron–sulfur clusters plus the single flavin transfers the electrons from the H-cluster to NAD<sup>+</sup> (**Figure 4B**). However, so far no metal center has been shown to accommodate the special properties necessary for electron bifurcation, therefore, this model would lead to a completely new field of enzyme mechanism catalyzed by metal centers.

#### THE CURIOUS CASE OF THE NAD+-DEPENDENT HYDROGENASE FROM Syntrophomonas wolfei

Recently, a hydrogenase has been purified from the syntrophic bacterium S. wolfei that very much resembles all known FBEB hydrogenases discussed so far, however, does not show FBEB (Losey et al., 2017). Therefore, comparison of this enzyme to the other ones might be helpful to unravel the mechanism of FBEB hydrogenases. The purified recombinant hydrogenase (Hyd1ABC) of S. wolfei showed H<sup>2</sup> production from NADH alone, uncoupled to ferredoxin. The recombinant enzyme had a very high H2-dependent: methyl-viologen reducing activity (3,340 U/mg), H2-dependent NAD<sup>+</sup> reducing activity (94.5 U/mg) and catalyzed H<sup>2</sup> production from NADH with a specific activity of 6.6 U/mg. The enzyme is a trimeric protein complex composed of HydA1 (63 kDa), HydB1 (43 kDa), and HydC1 (17.5 kDa) (**Figure 5**). It contains five [4Fe–4S], two [2Fe–2S] clusters, and one H-cluster. The flavin content was determined to be 0.7 mol of FMN per mole of enzyme. The Hyd1ABC subunits share close similarity to those of the earlier discussed FBEB hydrogenases. HydA1 is, again, similar to the monomeric C. pasteurianum hydrogenase, putatively containing the H-cluster, 3 [4Fe–4S] clusters and one [2Fe–2S] cluster and is the site for H<sup>2</sup> formation. HydC1 has been predicted to contain one [2Fe–2S] cluster. Nevertheless, the question arose why this enzyme Hyd1ABC of S. wolfei, though having the same organization of subunits like FBEB hydrogenases, does not oxidize/reduce ferredoxin. One reason that the authors discuss is the lack of one [2Fe–2S] cluster putatively bound in a N-terminal domain of HydB of FBEB hydrogenases as well as the lack of one [4Fe–4S] cluster putatively bound in the C-terminal domain of HydB of FBEB hydrogenases. In fact, S. wolfei HydB1 is much smaller than HydB of other FBEB hydrogenases (43 kDa compared to 63 kDa). Electrons are supposed to be transferred from NADH via the FMN to the proximal [4Fe–4S] cluster, and further transferred via the [Fe–S] clusters in HydA1 to the H-cluster to reduce protons to H2. The S. wolfei hydrogenase

might be a step in evolution away from FBEB hydrogenases to "standard" hydrogenases or, possibly, the other way around. In any case, solving the structure of this enzyme and comparing it to a bifurcating hydrogenase might unravel factors essential for the energetic coupling within the bifurcating enzymes.

#### FORMATE DEHYDROGENASE COUPLED HYDROGENASES

Protons can be used by microorganisms as ubiquitously available electron acceptor to get rid of excess reducing equivalents. A similar function is played by CO2. In anoxic environments, CO<sup>2</sup> or HCO<sup>3</sup> <sup>−</sup> is ubiquitously available (50–100 mM HCO<sup>3</sup> <sup>−</sup> in lake sediments, 200-400 mM HCO<sup>3</sup> <sup>−</sup> in biogas reactors) and can be used as electron acceptor yielding formic acid/formate (Crable et al., 2011; Schink et al., 2017). Interestingly, the redox potential of the CO2/formate couple of −432 mV is very close to the H+/H<sup>2</sup> couple rendering both electron acceptors energetically very similar. Therefore, it is not surprising that both H<sup>2</sup> and formate have been observed as important electron carriers in anoxic environments with often interchangeable functions (Thiele and Zeikus, 1988; Stams and Plugge, 2009; Schink et al., 2017; Montag and Schink, 2018). However, for a long time little was known on how the two pools of high energy redox mediators are connected with each other. In the past years these two compounds gained increasing interest due to their potential as electron donors for biofuel production, as energy carriers for mobile applications or issues like H<sup>2</sup> storage. This has led to discoveries of novel enzymes but also so far unknown organisms that connect and utilize these two redox mediators.

#### FORMATE HYDROGEN LYASE OF E. coli

Already in 1932 Stephenson and Stickland discovered that whole cells of E. coli grown in the presence of formate decompose formate into H<sup>2</sup> + CO<sup>2</sup> (Stephenson and Stickland, 1931, 1932). They named the supposed enzyme system formate hydrogen lyase (FHL). When growing under anoxic conditions in the absence of an alternative electron acceptor E. coli produces formic acid during mixed acid fermentation from pyruvate catalyzed by pyruvate-formate lyase which is then exported from the cytoplasm by the channels FocA and/or FocB (Clark, 1989; Sawers et al., 2004; Wang et al., 2009; Lu et al., 2011; Trchounian and Trchounian, 2014; Hakobyan et al., 2018). Formic acid (pK<sup>S</sup> = 3.7) dissociates to formate and leads to an acidification of the environment. A drop in the pH together with the accumulation of formate leads to subsequent import of formate again and the induction of expression of the genes coding for the FHL enzyme. FHL then oxidizes formate to H<sup>2</sup> + CO<sup>2</sup> followed by reoxidation of H<sup>2</sup> by other membrane bound hydrogenases of E. coli that transfer the electrons to the quinone pool (Rossmann et al., 1991; Sawers, 1994; Sawers, 2005; Pinske and Sawers, 2016).

Stephenson and Stickland assumed the enzymes to be a combination of formate dehydrogenase and hydrogenase. Genetic and physiological studies have confirmed this assumption and shown that the FHL of E. coli consist of a formate dehydrogenase bound to a membrane integral multimeric hydrogenase located in the cytoplasmic membrane (Zinoni et al., 1986; Böhm et al., 1990; Rossmann et al., 1991; Sauter et al., 1992). However, isolation of the whole FHL complex was achieved only recently by McDowall et al. (2014). The enzyme complex consists of two membrane integral subunits and five soluble subunits. The soluble subunits HycE and HycG represent the large (65 kDa) and small (20 kDa) subunits of the [NiFe]-hydrogenase termed E. coli hydrogenase-3 (Hyd-3) (**Figure 6A**). Formate is oxidized by the subunit FdhF (also called formate dehydrogenase H) (Axley et al., 1990; Gladyshev et al., 1996; Boyington et al., 1997).

The FHL complex shows high similarity to a group of similar membrane bound hydrogenases now called energy-converting hydrogenases (Ech). These enzymes are found in many anaerobic and facultative anaerobic bacteria and H<sup>2</sup> formation is coupled to different electron donors such as reduced ferredoxin or CO, facilitated by auxiliary subunits (Fox et al., 1996; Künkel et al., 1998; Meuer et al., 1999; Sapra et al., 2000; Soboh et al., 2002). It has been assumed that Ech hydrogenases couple H<sup>2</sup> formation to the translocation of ions across the membrane. Experimental proof for this concept was established using inverted membrane vesicles of the methanogenic archaeon Methanosarcina mazei (Welte et al., 2010a,b). However, whether the FHL complex of E. coli is also coupled to energy conservation has been a matter of debate for many years (Bagramyan and Martirosov, 1989; Trchounian et al., 1997, 2000, 2011; Hakobyan et al., 2005; McDowall et al., 2014; Trchounian and Sawers, 2014; Pinske and Sargent, 2016).

FHL couples the two half reactions

formate + H + CO<sup>2</sup> + 2 e <sup>−</sup> + 2 H <sup>+</sup> E 0 0 = −432 mV (7)

and

$$2\ H^{+} + 2\ e^{-} \rightleftharpoons\ H\_{2}\qquad E^{0'} = \ -414\ mV\tag{8}$$

resulting in the net reaction

$$
\begin{aligned}
\text{formatate} + H^+ &\rightleftharpoons H\_2 + \text{ CO}\_2 \end{aligned} \tag{9}$$

Under standard conditions, the free energy change of Eq. (9) is only −3.5 kJ mol−<sup>1</sup> (Thauer et al., 1977). Thus, based on the 1G 0 ' value, the FHL reaction should be fully reversible under physiological conditions, however, the enzyme shows a strong bias toward formate oxidation. Values for in vitro turnover frequency (TOF) for FHL activity was reported between 1,200 h−<sup>1</sup> and 1,920 h−<sup>1</sup> (McDowall et al., 2014; Pinske and Sargent, 2016) whereas the reverse reaction, formate formation from H<sup>2</sup> + CO<sup>2</sup> was only observed with an apparent TOFs of only 202 h−<sup>1</sup> (measured in a discontinuous one point assay and reported as 3.25 µmol formate produced after 5 h with 370 µg of enzyme) (Pinske and Sargent, 2016). This bias could not be explained by the catalytic properties of FdhF and Hyd-3 that have been analyzed separately and show high activities in both reaction directions (Bassegoda et al., 2014; McDowall et al., 2014). In addition to both active sites a factor determining the bias in one or the other direction could be the thermodynamic landscape

of the connecting iron–sulfur clusters. In addition, the membrane attachment of this enzyme complex and the connection to the enzyme activity is still puzzling (Bagramyan and Martirosov, 1989; Trchounian et al., 1997, 2000, 2011; Hakobyan et al., 2005; McDowall et al., 2014; Trchounian and Sawers, 2014; Pinske and Sargent, 2016). From a physiological point of view, the bias could be a remnant of evolution since there is no physiological situation where E. coli would use the reverse reaction to produce formate from CO<sup>2</sup> since this organism is not able to utilize formate further in the metabolism.

#### HYDROGEN-DEPENDENT CO<sup>2</sup> REDUCTASES

For many years, FHL was the only enzyme complex known that connects the electron carriers H<sup>2</sup> and formate but this enzyme shows a strong bias toward formate oxidation. Did evolution also bring up enzymes adapted for the reverse reaction? Efficient catalysts for hydrogenation of CO<sup>2</sup> are highly soughtafter (Appel et al., 2013; Beller and Bornscheuer, 2014; Wang et al., 2015). These could be used for CO<sup>2</sup> conversion technologies for carbon capture, for CO2-based synthesis or for H<sup>2</sup> storage (Yishai et al., 2016; Preuster et al., 2017b; Bulushev and Ross, 2018). Many homogeneous and heterogenous chemical catalyst have been developed, however, often requiring high temperatures and pressures (Fujita et al., 2013; Wang et al., 2015). In biological systems, of the six pathways known for CO<sup>2</sup> fixation, only the WLP proceeds via direct reduction of CO<sup>2</sup> to formate catalyzed by formate dehydrogenases (Fuchs, 2011). In 2013, a formate dehydrogenase was isolated from the acetogenic bacterium A. woodii that was in complex with a hydrogenase, on first glance resembling the FHL of E. coli (Schuchmann and Müller, 2013). However, in contrast to FHL, the enzyme was lacking membrane integral subunits. When isolated, the enzyme contained only four subunits, one formate dehydrogenase, one hydrogenase and two iron–sulfur cluster rich subunits, identified as being encoded by genes found in one gene cluster (**Figure 6B**). The gene clusters contains in addition a gene coding for a second formate dehydrogenase, a gene for a third iron–sulfur cluster rich subunits and a gene designated as fdhD (Poehlein et al., 2012; Schuchmann and Müller, 2013). The formate dehydrogenase FdhF2 as found in the isolated enzyme is a 80.1 kDa protein with 59% identity to FdhH of the E. coli FHL aligning over the whole sequence of 713 amino acids. It has one predicted [4Fe–4S] cluster and a bis-PGD cofactor. ICP-MS identified 0.6 mol molybdenum but no tungsten per mol of isolated enzyme preparation agreeing with a mononuclear molybdenum bound to the bis-PGD cofactor thus classifying this enzyme as member of the dimethylsulfoxide reductase family within the Mo/W-bis PGD super-family (Schuchmann and Müller, 2013). At amino acid position 139, a selenocysteine is predicted which is in agreement with the determined selenium in the preparation. In contrast to E. coli FHL, the [NiFe] hydrogenase subunits of FHL are substituted by the [FeFe] hydrogenase subunit HydA2 (50.2 kDa). The sequence of this subunit contains the conserved regions typical for H-cluster binding and aligns to amino acid region 201–572 of the monomeric ferredoxin-dependent [FeFe] hydrogenase of C. pasteurianum (**Figure 3**). In contrast to the C. pasteurianum enzyme, the N-terminus containing one [2Fe– 2S]- and one [4Fe–4S] cluster is missing. In addition to the H-cluster including the adjacent [4Fe–4S] cluster, the enzyme is predicted to contain two additional [4Fe–4S] clusters. HydA2 and FdhF2 are supposed to be connected by the two small electron transferring subunits HycB2 and HycB3 (18.9 and 20.1 kDa, respectively) containing 4 [4Fe–4S] clusters each (**Figure 6B**).

Enzymatic assays utilizing methyl viologen demonstrated that both FdhF2 as well as HydA2 were highly active in the isolated enzyme. H<sup>2</sup> oxidation was catalyzed with a TOF of 30,474 s−<sup>1</sup> (Schuchmann and Müller, 2013). The formate dehydrogenase subunit catalyzed formate oxidation with 1,693 s−<sup>1</sup> and CO<sup>2</sup> reduction with 372 s−<sup>1</sup> also showing a bias for formate oxidation,

FIGURE 7 | Distribution of HDCR-like gene clusters. Fully sequenced bacterial genomes have been searched for HDCR-like gene clusters (Schwarz et al., 2018). Based on the gene cluster arrangement, the enzymes can be classified in HDCR-like enzyme complex without and with electron bifurcating modules. Predicted cofactors based on the primary sequence of each subunit are shown in the right top corner. For clarity, single genes disrupting the gene cluster in some organisms such as maturation genes (e.g., fdhD) or transposase are not shown.

however, not as severe as reported for FdhH of the FHL that showed no measurable CO<sup>2</sup> reduction activity with soluble electron carriers. HydA2 has been analyzed by catalytic protein film electrochemistry by adsorbing the whole enzyme complex to an electrode (Ceccaldi et al., 2017). The hydrogenase is biased for proton reduction oxidizing H<sup>2</sup> 10 times slower than it reduces protons at pH 6 as is typical for [FeFe] hydrogenases. However, this ratio defining the bias decreases with increasing pH.

K<sup>M</sup> for H<sup>2</sup> was determined to 0.24 atm [K<sup>i</sup> (H2) = 6.4 atm], being three times lower than reported values for C. reinhardtii and Clostridium acetobutylicum and thus showing the highest affinity ever reported for a [FeFe] hydrogenase (Fourmond et al., 2013; Ceccaldi et al., 2017). In addition, HydA2 showed the surprising characteristic of being the first completely CO tolerant [FeFe] hydrogenase. Even though the enzyme is strongly inhibited by CO [Ki(CO) = 0.11 µM], it fully reactivates upon removal of CO which is in contrast to other [FeFe] hydrogenases where CO provokes irreversible damage to the H-cluster (Goldet et al., 2009; Baffert et al., 2011; Foster et al., 2012). This feature could be a result of the metabolism of A. woodii where CO is an intermediate of the reductive acetyl-CoA pathway (Schuchmann and Müller, 2014; Bertsch and Müller, 2015b).

Coupling of the formate dehydrogenase/hydrogenase complex of A. woodii was analyzed by incubating the isolated enzyme in the presence of formate. This lead to H<sup>2</sup> production with a TOF of 142,212 h−<sup>1</sup> (Schuchmann and Müller, 2013). Interestingly,

this enzyme is fully reversible also catalyzing hydrogenation of CO<sup>2</sup> at 30◦C and 0.8 bar of H<sup>2</sup> and 0.2 bar of CO<sup>2</sup> with a TOF of 101,600 h−<sup>1</sup> being significantly faster than the currently best known chemical catalysts (Hull et al., 2012; Jeletic et al., 2013; Wang et al., 2015; Eppinger and Huang, 2017). To distinguish the soluble enzyme complex from membrane bound FHL complexes that are weak catalysts for CO<sup>2</sup> hydrogenation the enzyme was named hydrogen-dependent CO<sup>2</sup> reductase (HDCR). As mentioned before, CO is a strong inhibitor of the hydrogenase activity (Goldet et al., 2009). Interestingly, the HDCR has an alternative electron entry site and can use reduced ferredoxin as electron donor to allow CO<sup>2</sup> reduction even in the presence of 1 atm CO. However, this activity is 17 times slower than the hydrogen-dependent activity with a TOF of only 6,095 h−<sup>1</sup> (Schuchmann and Müller, 2013). Noteworthy, further characterization of the HDCR revealed another interesting property of the enzyme. In vitro the HDCR of A. woodiireversibly polymerized into ordered filamentous structures of more than 0.1 µM in length (Schuchmann et al., 2016). Divalent cations could be identified to promote the polymerization process and it was observed that the polymerized form of the enzyme was more active. The in vivo significance of this observation is unresolved.

As described before, the HDCR gene cluster of A. woodii contains a gene, fdhF1, for a second formate dehydrogenase subunit with the corresponding electron transferring subunit hycB1. The deduced amino acid sequence of FdhF1 is to 80% identical to FdhF2 with the major exception being a cysteine in FdhF1 at position 139 where a selenocysteine is encoded in FdhF2. Metal-dependent formate dehydrogenases have been described with either selenocysteine or cysteine in the active site (Axley et al., 1990; Friedebold and Bowien, 1993; Raaijmakers et al., 2002; de Bok et al., 2003; Laukel et al., 2003). The lower pK<sup>a</sup> value of selenocysteine compared to cysteine is typically connected to a higher reactivity (Stadtman, 1996; Böck et al., 2005; Stock and Rother, 2009). Exchanging selenocysteine by cysteine in FdhH of E. coli let to a decrease in turnover number by over two orders of magnitude. However, the affinity for the substrate was increased in the cysteine variant as reflected by a lower K<sup>M</sup> value (Axley et al., 1991). There is no data on the different properties of FdhF1 and FdhF2 of A. woodii, however, the assumption is that in dependence on the presence of selenium in the environment the two subunits are produced differentially with the selenocysteine containing variant being the more active one. In another acetogenic bacterium, a spirochete from the termite gut, Treponema primitia, a gene cluster with similarity to the HDCR gene cluster was identified (Matson et al., 2010). It consists of the genes fdhFcys, two hycB genes, fdhD and hydA. Separated by 14 genes, a second fdhF gene is encoded, containing a putative selenocysteine codon and a SECIS element (fdhFsec). Transcript analysis revealed a differential expression of fdhFcys and fdhFsec in dependence of selenium availability with up to 40-fold change in transcript levels. Half-maximum decrease in transcript level of fdhFcys was observed with less than 50 pM sodium selenite, whereas 1.5 nM sodium selenite were required for half-maximum increase in fdhFsec transcript levels. Interestingly, only transcript levels downstream of the SECIS element were differentially expressed, whereas transcript upstream of the SECIS element did not show differential regulation (Matson et al., 2010).

#### PHYSIOLOGY AND DIVERSITY OF HDCR COMPLEXES

The first HDCR complex has been identified in the acetogenic bacterium A. woodii. Acetogenic bacteria utilize the WLP for energy conservation and carbon fixation, the only carbon fixation pathway that utilizes CO<sup>2</sup> by direct reduction to formate. Therefore, this first reduction step of CO<sup>2</sup> is essential for the metabolism of these bacteria. The redox potential of CO2/formate of −432 mV limits the number of possible electron donors for this reaction. The first formate dehydrogenase characterized in an acetogenic bacterium was isolated from Moorella thermoacetica (Yamamoto et al., 1983). This heterotetrameric enzyme was the first enzyme described to contain tungsten and catalyzes CO<sup>2</sup> reduction coupled to NADPH oxidation. The standard redox potential of NADP+/NADPH of −340 mV is too positive for CO<sup>2</sup> reduction, however, in anaerobes the intracellular ratio of NADP+/NADPH is 1/40 resulting in a redox potential of −370 mV (Sauer et al., 2004; Bennett et al., 2009). Insights into an increasing number of sequenced genomes of acetogenic bacteria has revealed that the genes encoding enzymes for the first reaction step of the WLP show a large diversity (Schuchmann and Müller, 2014; Bertsch and Müller, 2015a; Bengelsdorf et al., 2016). Therefore, the knowledge of the enzymes of M. thermoacetica could not be transferred to other acetogens. Characterization of A. woodii has revealed that it does not use NADPH but H<sup>2</sup> as electron donor for CO<sup>2</sup> reduction catalyzed by the HDCR. Energetically, the difference between NADPH and H<sup>2</sup> is very small under physiological conditions. H<sup>2</sup> is a stronger reductant under standard conditions (E<sup>0</sup> ' = −414 mV), however, at the minimum H<sup>2</sup> pressures required by A. woodii to perform acetogenesis (250 Pa) the redox potential is only −340 mV (Poehlein et al., 2012). Under these conditions, the equilibrium concentration of formate is 0.1 mM (CO<sup>2</sup> at 0.2 × 10<sup>5</sup> Pa, 30◦C), in the range of the K<sup>M</sup> value of next enzyme of the pathway, the formyl-THF synthetase (Poehlein et al., 2012). Coupling CO<sup>2</sup> reduction directly to H<sup>2</sup> oxidation also energetically couples it directly to the external H<sup>2</sup> partial pressure. A H<sup>2</sup> pressure of 250 Pa is very high compared to the values observed in methanogenic environments of 1 to 10 Pa H<sup>2</sup> (Conrad et al., 1986; Seitz et al., 1990). Therefore, CO<sup>2</sup> reduction by HDCR enzymes seems to be not competitive in methanogenic environments. However, the observation of reduced ferredoxin as alternative electron carrier could be a mechanism to overcome this limitation (in addition to the proposed bypass for CO inhibition of the hydrogenase as discussed before). On the other hand, utilizing reduced ferredoxin instead of H<sup>2</sup> would result, in A. woodii, in less ATP conserved (Schuchmann and Müller, 2013). The physiological relevance of the ferredoxin entry site of the HDCR has, however, not been studied yet.

The complete reversibility of the HDCR, as opposed to membrane bound FHL complexes, is a direct reflection of its physiological function. Besides H<sup>2</sup> + CO2, A. woodii can grow

for example with formate or methanol as sole carbon and energy sources (Schuchmann and Müller, 2016; Kremp et al., 2018). To convert these compounds, part of the substrate must be oxidized to CO<sup>2</sup> to provide reducing equivalents. In these scenarios, the HDCR must operate in reverse to oxidize formate to H<sup>2</sup> + CO<sup>2</sup> since it is the only formate dehydrogenase found in A. woodii (Poehlein et al., 2012).

H<sup>2</sup> and formate are widely used electron carriers in anoxic ecosystems and are product or substrate of many microorganisms. The discovery of the HDCR enzyme as soluble, not energetically coupled enzyme complex that connects these two pools raised the question of the distribution of similar enzymes in other organisms. When searching for gene clusters containing homologs of fdhF1, hycB1/3 and hydA2 of A. woodii we found 18 organisms encoding gene clusters encoding putative HDCRs (**Figure 7**), see also Schwarz et al. (2018). Of these putative enzymes only one from the acetogen Thermoanaerobacter kivui has been isolated and characterized (Schwarz et al., 2018). Of the 18 organisms, three are acetogenic bacteria that can grow with H<sup>2</sup> + CO<sup>2</sup> as substrates [Clostridium difficile (Köpke et al., 2013), T. primitia (Graber and Breznak, 2004), T. kivui (Leigh et al., 1981)] and 5 belong to the sulfate reducers [Desulfotalea psychrophila (Knoblauch et al., 1999), Desulfobacterium autotrophicum (Brysch et al., 1987), Desulfovibrio alaskensis (Feio et al., 2004), Desulfovibrio magneticus (Sakaguchi et al., 2002), Desulfovibrio salexigens (Postgate and Campbell, 1966)]. In these organisms the HDCRs could play a role in the WLP as in A. woodii to reduce CO<sup>2</sup> to formate either for carbon fixation alone (sulfate reducers) or for carbon fixation and energy conservation (acetogens). Many organisms from the genus Paenibacillus have putative HDCR gene clusters. Paenibacilli are found in many environments from polar to tropic regions, often found in soil where they have been reported to be associated with plant roots promoting plant growth, some species are pathogenic to honeybees or invertebrates and some are opportunistic pathogenic to humans (Grady et al., 2016). Species such as Paenibacillus polymyxa are facultative anaerobes and can ferment glucose under anoxic conditions by mixed acid fermentation to a mixture of products such as acetate, ethanol, lactate, formate, H<sup>2</sup> and CO<sup>2</sup> (Marwoto et al., 2004). Cell free extracts of P. polymyxa have been shown to contain hydrogenase activity and catalyze H<sup>2</sup> evolution from formate (Grau and Wilson, 1962). We conclude, that in Paenibacilli the HDCR could take over the function of the FHL complex as used by E. coli to detoxify formate produced during fermentation by oxidizing it to H<sup>2</sup> + CO2.

We identified two gene clusters that show notable differences to the standard HDCR gene cluster. In this case, the hydrogenase subunit resembles the hydrogenase subunit of electron bifurcating hydrogenases by containing two additional iron–sulfur clusters. The formate dehydrogenase subunit also contains three putative additional iron–sulfur clusters whereas genes for the Hyc proteins are missing (**Figure 7**). The additional iron–sulfur clusters in the Hyd and Fdh subunit could take over the function of electron transfer otherwise catalyzed by the Hyc proteins. In contrast, in five other organisms we identified putative HDCR gene cluster that are more complex than the "standard" HDCR and that have, in the case of Clostridium autoethanogenum, been shown to combine features of HDCRlike enzyme complexes and electron bifurcating hydrogenases (Wang et al., 2013a).

#### RAISING THE COMPLEXITY: FORMATE DEHYDROGENASE COUPLED ELECTRON BIFURCATING HYDROGENASES

In the acetogenic bacterium C. autoethanogenum a very complex enzyme has been isolated that indeed combines the features of the HDCR enzyme complex and hydrogenases utilizing FBEB (Wang et al., 2013a). This complex is encoded by seven genes found in one cluster on the chromosome. It putatively contains 19 iron–sulfur clusters that could connect the active site of a [FeFe] hydrogenase and a selenocysteine containing bis-PGD containing formate dehydrogenase (**Figure 8**). In addition, a flavin is predicted to be bound to the complex. Chemical analysis of the isolated complex confirmed the presence of FMN, selenium, tungsten and 60 mol iron of 76 predicted mol iron per complex. Molybdenum and FAD were not detected. The enzyme complex showed very interesting catalytic properties. When incubated with H<sup>2</sup> + CO<sup>2</sup> it catalyzed formate formation with 41 U mg−<sup>1</sup> and the reverse reaction, H<sup>2</sup> formation from formate with 40 U mg−<sup>1</sup> thus showing the typical reversible reaction of HDCR enzymes. However, when incubated with H2, NADP+, and oxidized ferredoxin the enzyme oxidized H<sup>2</sup> and reduced NADP<sup>+</sup> and ferredoxin simultaneously (32 U mg−<sup>1</sup> ). NADP+, NAD<sup>+</sup> or ferredoxin alone in the presence of H<sup>2</sup> are not reduced. H<sup>2</sup> could also be formed only in the presence of NADPH and reduced ferredoxin together (27 U mg−<sup>1</sup> ). In addition, NADP<sup>+</sup> and oxidized ferredoxin could also be reduced with formate, again only in the presence of both electron acceptors (15 U mg−<sup>1</sup> ). The same was true for the reverse reaction, CO<sup>2</sup> reduction with NADPH and reduced ferredoxin (7 U mg−<sup>1</sup> ). It has been further demonstrated that the electron-bifurcating reactions with two electron acceptors are strictly energetically coupled (Wang et al., 2013a). Taking these results together, the enzyme represents a combination of formate dehydrogenase coupled hydrogenase as in the case of the HDCR and subunits facilitating FBEB. Interestingly, electron bifurcation to NADP<sup>+</sup> and oxidized ferredoxin is catalyzed with two alternative electron donors, H<sup>2</sup> or formate. Puzzling as well is the observation that CO inhibits not only the hydrogenase activity but also reduction of NADP<sup>+</sup> and ferredoxin with formate, even though CO is not known to inhibit formate dehydrogenases. The physiological function is complex as well. The enzyme is highly produced in cells grown with syngas as substrate (42% CO, 36% N2, 20% CO2, and 2% H2) even though already CO concentrations of around 0.1% inhibit most activities of the enzyme complex to 50%. Wang et al. (2013a) propose that during balanced growth on CO the steady state concentration within the cell is much lower than external CO concentrations thus not inhibiting the enzyme completely. The authors propose the enzyme utilizes NADPH

and reduced ferredoxin for CO<sup>2</sup> reduction when cells grown on CO and the hydrogenase module has the function to protect the cells from over-reduction when NADP<sup>+</sup> and ferredoxin get too reduced during growth of C. autoethanogenum on CO. For a detailed discussion of this function we refer the reader to the original work of Wang et al. (2013a) since it is beyond the scope of this review.

In four other organisms, we identified gene clusters that encode for putative enzyme complexes resembling the electronbifurcating formate dehydrogenase/hydrogenase complex of C. autoethanogenum. These include the acetogen Clostridium carboxidivorans that can, as C. autoethanogenum utilize CO as carbon and energy source (Liou et al., 2005) and the amino acid fermenting bacteria Cloacibacillus porcorum (Looft

et al., 2013) and Peptoclostridium acidaminophilum (formerly Eubacterium acidaminophilum) (Zindel et al., 1988) (**Figure 7**). P. acidaminophilum uses formate as electron donor to reduce glycine, sarcosine, or betaine to acetyl phosphate (Hormann and Andreesen, 1989; Andreesen, 1994). In addition, it is able to grow in syntrophic culture on, e.g., alanine, valine, leucine or malate if a hydrogen- or formate-consuming bacterium is also present (Zindel et al., 1988). The genome encodes for two putative formate dehydrogenases of which one has been isolated and characterized (Graentzdoerffer et al., 2003). The purified enzyme contained the subunits FdhF (selenocysteine and tungsten bis-PGD cofactor containing formate dehydrogenase), two iron– sulfur cluster containing subunits and the flavin- and NAD(P) binding subunit HydB. The corresponding gene cluster encodes for a [FeFe] hydrogenase, however, hydrogenase activity was lost during purification and a hydrogenase subunit or hydrogenase activity could not be detected in the isolated enzyme. On the other hand, formate dehydrogenase activity was present and showed reversible catalysis of formate oxidation/CO<sup>2</sup> reduction (43 U mg−<sup>1</sup> /12 U mg−<sup>1</sup> , methyl viologen as electron donor/acceptor).

#### BIOTECHNOLOGICAL H<sup>2</sup> PRODUCTION AND STORAGE

H<sup>2</sup> has been considered for a long time to replace fossil fuels in the future to tackle the climate problem by decreasing emission of CO<sup>2</sup> (Ball and Wietschel, 2010). H<sup>2</sup> is a very simple molecule that can be produced by various methods including splitting of water that, if the energy is provided by renewable energy sources like solar or wind power does not lead to CO<sup>2</sup> emission. Consumption of H<sup>2</sup> for energy generation either by thermal combustion or H<sup>2</sup> fuel cells at the site of use does not release CO<sup>2</sup> as well. On the other hand, significant challenges are connected with a switch to the H<sup>2</sup> economy. Currently, 95% of world-wide H<sup>2</sup> is produced from fossil fuels by coal gasification or by steam reforming of natural gas followed by further processing to increase the H<sup>2</sup> yield (water-gas shift reaction) (Brandon and Kurban, 2017). These processes emit large quantities of CO<sup>2</sup> and are not sustainable. Another major drawback of H<sup>2</sup> is its very low density. On a gravimetric basis, the energy content of 33.3 kWh/kg H<sup>2</sup> is three times higher than that of gasoline. On a volumetric basis this situation is dramatically reversed to 3 Wh/L for gaseous H<sup>2</sup> versus 8,600 Wh/L to 9800 Wh/L for gasoline or diesel, respectively (Preuster et al., 2017a). Therefore, methods need to be developed for onboard storage of H<sup>2</sup> in mobile applications such as cars and large-scale storage of H2, e.g., for storing energy produced by wind or solar power at off peak times. Another challenge of H<sup>2</sup> concerns safety issues with respect to the high volatility and the formation of highly explosive gas mixtures if getting in contact with air.

A key technology to sustainably produce H<sup>2</sup> without carbon emission is electrolysis of water driven by electricity generated from renewable sources (Ursua et al., 2012). Electrocatalytic splitting of water releases only H<sup>2</sup> and O<sup>2</sup> thus generating H<sup>2</sup> free of contaminations such as CO which is important for H<sup>2</sup> fuel cell applications. This process has been used already in large scale, however, the low price of fossil fuels has rendered it uneconomic. In the near future, process efficiencies of 85–95% are expected with the current price of H<sup>2</sup> produced by electrolysis being approximately at 3 € per kg (Tremel et al., 2015).

Biological catalysis provides another alternative method for H<sup>2</sup> production. These processes can currently not be considered as the major technology to provide H<sup>2</sup> for the H<sup>2</sup> economy but could provide small but significant contributions to exploit other H<sup>2</sup> sources as well. H<sup>2</sup> production by biological systems can be classified into four general mechanisms, namely direct and indirect biophotolysis, photofermentation, and dark fermentation which have been extensively reviewed elsewhere (Manish and Banerjee, 2008; van Niel, 2016; Kumar et al., 2018). We want to focus here on dark fermentation which has shown the highest H<sup>2</sup> evolution rates so far (Rittmann and Herwig, 2012), biological alternatives to store H<sup>2</sup> and the new opportunities provided by the recent discoveries of novel types of hydrogenases.

Dark fermentation can be catalyzed by single organisms or complex consortia to extract energy from multiple substrates such as waste biomass. This process has been used successfully for many years in biogas plants that convert waste biomass to methane. Within this process, H<sup>2</sup> is produced by fermenting organisms but immediately consumed by methanogens. Selective inhibition of methanogens can stop H<sup>2</sup> consumption leading to H<sup>2</sup> accumulation (Schink, 1997; Catal et al., 2015). Alternatives are single organisms capable of producing large quantities of H<sup>2</sup> either naturally or by metabolic engineering. In nature, most H<sup>2</sup> is produced from the fermentation of carbohydrates (Thauer et al., 2008). Complete oxidation of 1 mole of hexoses such as glucose would yield a maximum of 12 moles of H<sup>2</sup>

$$1\,\mathrm{C}\_6\mathrm{H}\_{12}\mathrm{O}\_6 + 12\,\mathrm{H}\_2\mathrm{O} \rightleftharpoons 12\,\mathrm{H}\_2 + \,\mathrm{12}\,\mathrm{HCO}\_3^- \tag{10}$$

This reaction is endergonic [1G 0 ' = 13.6 kJ mol−<sup>1</sup> (Flamholz et al., 2012)] and thus not feasible in vivo. The maximum H<sup>2</sup> that can be produced from glucose by microorganisms is limited to 4 mol H<sup>2</sup> per mol of glucose known as the Thauer limit (Thauer et al., 1977). In practice, most H<sup>2</sup> yields of microorganisms are far below this theoretical limit due to thermodynamic reasons. Oxidation of glucose to 2 pyruvate by the Embden–Meyerhof– Parnas pathway transfers all electrons to NAD<sup>+</sup> producing in total 2 NADH. In contrast, oxidative decarboxylation of 2 pyruvate to acetyl-CoA and CO<sup>2</sup> can be used to reduce 2 ferredoxins as electron carrier. H<sup>2</sup> production from ferredoxin is exergonic whereas H<sup>2</sup> production from NADH is endergonic as described above. Thus, without a H<sup>2</sup> consuming partner organism, H<sup>2</sup> can only be produced from reduced ferredoxin resulting in a maximum of 2 H<sup>2</sup> produced per mol of glucose. The solution to this problem is provided by the new class of electronbifurcating hydrogenases that can produce H<sup>2</sup> from NADH as well, driven by the coupled production of H<sup>2</sup> from reduced ferredoxin (Buckel and Thauer, 2013). Organisms that harbor such enzymes such as T. maritima (Selig et al., 1997; Frock et al., 2012) or R. albus (Zheng et al., 2014) are able to reach the Thauer limit by fermenting glucose according to

fmicb-09-02911 November 30, 2018 Time: 15:13 # 16

$$\text{Glucose} + 2\text{ }H\_2\text{O} \rightleftharpoons 4\text{ }H\_2 + 2\text{ }Acetate\text{ }^- + 2\text{ }CO\_2\tag{11}$$

with a 1G 0 ' value of −250 kJ mol−<sup>1</sup> at 90◦C (growth temperature of T. maritima) and −215 kJ mol−<sup>1</sup> at 25◦C. An alternative strategy to reach these high H<sup>2</sup> yields has been observed in the hyperthermophilic archaeon Pyrococcus furiosus growing at 100◦C. In this organism the NAD+-dependent glyceraldehyde-3 phosphate dehydrogenase is replaced by a ferredoxin-dependent enzyme at the expense of this enzyme not being couple to ATP formation (Mukund and Adams, 1995). Thus, glucose oxidation results only in generation of reduced ferredoxin allowing full reoxidation by formation of H2.

Because of the good understanding of the metabolism and the well-established genetic tools, E. coli is the prime candidate for biotechnological applications. However, E. coli does not harbor an FBEB hydrogenase and is limited to producing 2 mol H<sup>2</sup> per mol glucose. Recently, it has been demonstrated for the first time that heterologous expression of an FBEB hydrogenase in E. coli is possible (Kelly et al., 2015). Since E. coli lacks pyruvate:ferredoxin oxidoreductase (PFO), this enzyme from T. maritima as well as ferredoxin from T. maritima had to be produced additionally in E. coli to achieve H<sup>2</sup> production. The amounts and rates of H<sup>2</sup> production were very low therefore this study can be seen as mere proof of principle that such a complex hydrogenase can be produced in E. coli, however, future studies need to prove that this approach indeed can significantly improve the H<sup>2</sup> yields of E. coli.

The recently discovered formate dehydrogenase coupled hydrogenases could also contribute to the H<sup>2</sup> economy in the part of H<sup>2</sup> storage and transportation. To overcome the very low volumetric energy density of gaseous H<sup>2</sup> many technologies are tested to store H<sup>2</sup> in a more compact form (Schlapbach and Zuttel, 2001; Preuster et al., 2017a,b). Physical options are compression and storage in high pressure tanks above 200 bar or storage of H<sup>2</sup> in its liquid form present at −253◦C, both technologies requiring high energy input reducing the efficiency of H<sup>2</sup> as energy carrier. A chemical alternative is reacting H<sup>2</sup> with other compounds to produce liquid organic H<sup>2</sup> carriers (LOHCs). For example, hydrogenation of CO<sup>2</sup> leads to formic acid that contains 4.4 wt % of H2. This is close to the 2017 H<sup>2</sup> storage target of 5.5 wt % in gravimetric energy density set by the United States Department of Energy at temperatures from −40 to 60◦C at a maximum pressure of 100 atm (Enthaler et al., 2010; Laurenczy, 2011; Kawanami et al., 2017). The volumetric capacity is 53 g H2/L formic acid thus one liter formic acid can store roughly 600 liter of gaseous H2. Formic acid is nontoxic and non-explosive, however, its corrosive nature requires special consideration for tanks and equipment. Formic acid can be decomposed to H<sup>2</sup> + CO<sup>2</sup> before H<sup>2</sup> is then used in a H<sup>2</sup> fuel cell. Chemical catalysts have been developed that catalyze this dehydrogenation with high activity and stability at ambient temperatures below 100◦C (Boddien et al., 2008, 2009). However, the initial hydrogenation of CO<sup>2</sup> still represents a challenge. Many homogenous and heterogenous chemical catalyst have been developed but are depending on either high temperatures and pressures, expensive bases or special media for high efficiencies (Hull et al., 2012; Jeletic et al., 2013; Wang et al., 2015; Eppinger and Huang, 2017). Highest rates for CO<sup>2</sup> hydrogenation (TOF 3,400 h−<sup>1</sup> ) with chemical catalyst at ambient conditions have been achieved with cobalt based catalysts in the presence of a special and expensive base (Verkade's base) (Jeletic et al., 2013). The newly discovered HDCR complexes could provide a biotechnological alternative or could serve as model to design more efficient catalysts. HDCR of A. woodii catalyzes CO<sup>2</sup> hydrogenation with a TOF of 101,600 h−<sup>1</sup> at 30◦C and 1 bar of pressure (Schuchmann and Müller, 2013). In contrast to other CO<sup>2</sup> reductases the enzyme directly utilizes H<sup>2</sup> thus not requiring soluble electron carriers. In addition, with even higher TOFs it catalyzes the reverse reaction as well, closing the cycle for a H<sup>2</sup> storage process. By coupling the HDCR with a ferredoxin dependent CO dehydrogenase, conversion of syngas (H2, CO, CO2) or CO alone to formate was achieved (Schuchmann and Müller, 2013). This is advantageous since CO is very toxic to H<sup>2</sup> fuel cells and can be removed by intermittent conversion of the gases to formate. Major drawbacks are the high oxygen sensitivity and inherent stability issues typical for biocatalysts such as narrow pH and temperature range. To overcome the requirement of enzyme isolation, a whole cell system has been established using A. woodii as catalyst for reversible H<sup>2</sup> storage (Schuchmann and Müller, 2013). Genetic manipulations were not required since competing pathways for product formation were inhibited by addition of a sodium ionophore that makes the cytoplasmic membrane permeable for sodium ions thus inhibiting energy conservation and ATP synthesis. Using this system, formate could be specifically produced from H<sup>2</sup> + CO<sup>2</sup> reaching final concentrations (up to ∼0.25 M formate) that where only limited by the thermodynamics of the reaction. Formate was produced with a rate of 120 mmol formate h−<sup>1</sup> g −1 . Formate decomposition was catalyzed by whole cells with activities up to 71 mmol H<sup>2</sup> h −1 g −1 (Kottenhahn et al., 2018). Yields of 1 mol H<sup>2</sup> per mol of formate were demonstrated. Recently, the first HDCR from a thermophile, namely the acetogen T. kivui, was isolated (Schwarz et al., 2018). This enzyme showed surprising activities with TOFs for CO<sup>2</sup> hydrogenation of 9,556,000 h−<sup>1</sup> at 60◦C and still 1,856,000 h−<sup>1</sup> at 30◦C. Formate dehydrogenation was catalyzed with a TOF of 9,892,000 h−<sup>1</sup> . This enzyme contained in contrast to the HDCR of A. woodii tungsten instead of molybdenum and no selenocysteine in the formate dehydrogenase subunit. Future insights into the structure and biochemistry of this enzyme will hopefully unravel the factors determining this tremendous CO<sup>2</sup> reductase activities not observed for other apparently very similar formate dehydrogenases.

One alternative for HDCR catalysis of CO<sup>2</sup> hydrogenation is the FHL enzyme complex as found in E. coli. For many years thought to work efficiently only in direction of formate oxidation, recent work has demonstrated the reversibility. Even though CO<sup>2</sup> reductase activity of the isolated enzyme are orders

of magnitude slower compared to HDCR enzymes, whole cell catalysis under high pressure showed promising results (Roger et al., 2018). Utilizing a genetically modified strain that is deficient in competing formate dehydrogenases, hydrogenases and pyruvate formate lyase that are otherwise present in E. coli and using 10 bar of H<sup>2</sup> + CO<sup>2</sup> (H2:CO<sup>2</sup> ratio 1:1) at 37◦C approximately 500 mM formate could be produced over a time course of 23 h. Yields reached 100% (produced formate per consumed CO2) and initial rates of 36 mmol formate produced h −1 g <sup>−</sup><sup>1</sup> were demonstrated. Even though the FHL enzyme has less advantageous catalytic properties in the isolated form and has the disadvantage of being a membrane integral enzyme complex, the catalytic rates observed with whole cells of E. coli in the highpressure system are in the same order of magnitude as whole cell catalysis with A. woodii, however, the latter being performed at ambient pressures.

How could formic acid be integrated in the H<sup>2</sup> economy? As shown in **Figure 9**, energy production from renewable sources such as solar or wind power will be more decentralized. Electricity produced could be converted electrochemically by water splitting into H2. Alternative H<sup>2</sup> sources can be biohydrogen produced for example from waste biomass and multiple sources of synthesis gas such as industrial off-gas or gasification of municipal waste or biomass. H<sup>2</sup> production from syngas are well established technologies utilized already as major pathway for H<sup>2</sup> production. In the next step, H<sup>2</sup> could be bound to CO<sup>2</sup> at decentralized facilities producing formic acid that could then be stored in bulk amounts for energy storage or distributed to the final customer. Formic acid could be used directly for energy generation in direct formic acid fuel cells (DFAFCs) that are potential power sources for portable devices (Aslama et al., 2012). DFAFCs are less well developed compared to H<sup>2</sup> fuel cells. Therefore, for utilization in H<sup>2</sup> fuel cells, H<sup>2</sup> must be generated first by dehydrogenation of formic acid again at the site of use. Combustion of formic acid directly or H<sup>2</sup> after formic acid dehydrogenation results in release of water and CO<sup>2</sup> only. CO<sup>2</sup> can be used for the next H<sup>2</sup> storage cycle; therefore, no net CO<sup>2</sup> is generated.

#### CONCLUSION

Enzymes utilizing H<sup>2</sup> are known for almost a century and since the first crystal structure of a hydrogenase in 1995 (Volbeda et al., 1995) a very large number of studies revealed many facets of the biochemistry, reaction mechanism and evolution of hydrogenases. Most knowledge has been gained about the core subunit of hydrogenases that harbors the active site for H<sup>2</sup> oxidation/H<sup>+</sup> reduction, however, in recent

#### REFERENCES


years more and more hydrogenases have been discovered that are part of large and very complex multimeric enzymes connecting H<sup>2</sup> oxidation to multiple functions. One of the major breakthroughs in (anaerobic) microbiology within the last decade was the discovery of the novel enzyme mechanisms of FBEB. This discovery solved many thermodynamic enigmas within the physiology of anaerobic microorganisms. Concerning H<sup>2</sup> production, unraveling of hydrogenases utilizing FBEB finally solved the long-standing questions how H<sup>2</sup> production from NADH is possible and, the other way around, how H<sup>2</sup> can be used to reduce ferredoxin, both highly endergonic reactions without energetic coupling to other redox reactions. These enzymes follow a modular architecture with the large subunit of [FeFe] hydrogenases HydA being coupled to other functional subunits. The same is true for hydrogenases coupled to formate dehydrogenases or the combination of hydrogenase, formate dehydrogenase and electron bifurcation. All these enzymes contain a large number of iron–sulfur clusters that are supposed to accomplish the electrical connection of the different active sites or the active sites and multiple electron acceptors. The biochemical characterization of all these enzymes needs now to be followed by detail insights into the structure and reaction mechanisms. FBEB hydrogenases constitute a fascinating puzzle on how the splitting and energetic coupling of the two electrons is achieved since the flavin to take over this function in other characterized FBEB enzymes is apparently missing. The idea that indeed a metal center can take over this function is a compelling thought. On the other hand, a detailed understanding of H2 dependent CO<sup>2</sup> reductases would not only be important from a biochemical point of view but could also help to provide better catalysts for economic problems such as H<sup>2</sup> storage. After more than two decades focusing on the major subunit of hydrogenases the next decade might reveal the fascinating complexity of the auxiliary subunits of this still highly important class of enzymes.

#### AUTHOR CONTRIBUTIONS

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

#### FUNDING

Work from the authors laboratory was funded by the Deutsche Forschungsgemeinschaft and the European Research Council (ERC) under the European Union's Horizon 2020 Research and Innovation Programme (Grant Agreement No. 741791).

chemical catalysis of CO<sup>2</sup> fixation. Chem. Rev. 113, 6621–6658. doi: 10.1021/ cr300463y


selenium. Proc. Natl. Acad. Sci. U.S.A. 88, 8450–8454. doi: 10.1073/pnas.88.19. 8450



reducing equivalents for nitrogenase catalysis. Biochemistry 56, 4177–4190. doi: 10.1021/acs.biochem.7b00389


unusual coordination to an active site Fe binuclear center. Structure 7, 13–23. doi: 10.1016/S0969-2126(99)80005-7




**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 Schuchmann, Chowdhury and Müller. 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.

## Function of Biohydrogen Metabolism and Related Microbial Communities in Environmental Bioremediation

Ying Teng<sup>1</sup> \*, Yongfeng Xu1,2, Xiaomi Wang<sup>1</sup> and Peter Christie<sup>1</sup>

<sup>1</sup> Key Laboratory of Soil Environment and Pollution Remediation, Institute of Soil Science, Chinese Academy of Sciences, Nanjing, China, <sup>2</sup> College of Resources and Environment, University of Chinese Academy of Sciences, Beijing, China

Hydrogen (H2) metabolism has attracted considerable interest because the activities of H2-producing and consuming microbes shape the global H<sup>2</sup> cycle and may have vital relationships with the global cycling of other elements. There are many pathways of microbial H<sup>2</sup> emission and consumption which may affect the structure and function of microbial communities. A wide range of microbial groups employ H<sup>2</sup> as an electron donor to catalyze the reduction of pollutants such as organohalides, azo compounds, and trace metals. Syntrophy coupled mutualistic interaction between H2-producing and H2-consuming microorganisms can transfer H<sup>2</sup> and be accompanied by the removal of toxic compounds. Moreover, hydrogenases have been gradually recognized to have a key role in the progress of pollutant degradation. This paper reviews recent advances in elucidating role of H<sup>2</sup> metabolism involved in syntrophy and hydrogenases in environmental bioremediation. Further investigations should focus on the application of bioenergy in bioremediation to make microbiological H<sup>2</sup> metabolism a promising remediation strategy.

#### Edited by:

Eric Boyd, Montana State University, United States

#### Reviewed by:

Karen Trchounian, Yerevan State University, Armenia Philippe Constant, Institut National de la Recherche Scientifique (INRS), Canada

> \*Correspondence: Ying Teng yteng@issas.ac.cn

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 31 October 2018 Accepted: 17 January 2019 Published: 14 February 2019

#### Citation:

Teng Y, Xu Y, Wang X and Christie P (2019) Function of Biohydrogen Metabolism and Related Microbial Communities in Environmental Bioremediation. Front. Microbiol. 10:106. doi: 10.3389/fmicb.2019.00106 Keywords: bioremediation, hydrogenase, H<sup>2</sup> consumption, H<sup>2</sup> metabolism, H<sup>2</sup> production

### INTRODUCTION

It is well established that the main sources of molecular hydrogen (H2) are geochemical and anthropogenic activities and the main sink is the biological consumption of H<sup>2</sup> in soil ecosystems. The H<sup>2</sup> cycle can influence air quality and climate indirectly via effects on the oxidative capacity of the atmosphere (Ehhalt and Rohrer, 2009). In addition, the H<sup>2</sup> cycle plays an important role in microbial metabolism due to numerous microbial processes that depend on H<sup>2</sup> production and consumption (Vignais and Billoud, 2007; Greening et al., 2015b). For example, most of the tropospheric H<sup>2</sup> is consumed by soils due to the capacity of the majority of H2-oxidizing bacteria displaying high affinity for H<sup>2</sup> in soils to recycle it (Constant et al., 2010). H<sup>2</sup> is also a key metabolic compound in many anoxic ecosystems and its oxidation may support deep subsurface lithoautotrophic microbial ecosystems (Chivian et al., 2008; Nyyssonen et al., 2014; Wu et al., 2015; Bagnoud et al., 2016). The activities of H2-producing and consuming microbes therefore shape the global H<sup>2</sup> cycle and may have vital relationships with the global cycling of other elements including carbon, sulfur, and nitrogen.

The first H2-oxidizing microorganisms were discovered in the 1900s (Kaserer, 1905; Stephenson and Stickland, 1931). The physical properties of H<sup>2</sup> (e.g., its diffusion coefficient, 4 × 10−<sup>9</sup> m2 s −1 , and redox potential, E<sup>0</sup> ' = −0.42 V, make it relatively active in biological processes

(Greening et al., 2016). H<sup>2</sup> has dual physiological functions in microorganisms. Firstly, microbial fermentation of H<sup>2</sup> produced by facultative or obligate fermenters can disperse excess reductant from fermentative metabolism, for example in Escherichia coli and Clostridium spp. (Trchounian et al., 2012, 2017a,b). Secondly, prokaryotic microorganisms with different metabolic processes such as hydrogen-oxidizing bacteria, methanogens and anoxygenic phototrophic bacteria can exploit H<sup>2</sup> as an energy source and reductant (Schwartz et al., 2013). There are also a wide range of microorganisms with the ability to metabolize H<sup>2</sup> such as aerobes and anaerobes and lithotrophs and phototrophs (Vignais and Billoud, 2007; Schwartz et al., 2013; Peters et al., 2015; Greening et al., 2016). Furthermore, recent studies show that some aerobic soil acidobacteria and actinobacteria can exploit low levels of H<sup>2</sup> for survival in addition to growth, which challenges the traditional belief that H<sup>2</sup> metabolism is restricted to high-H<sup>2</sup> and low-O<sup>2</sup> environments (Constant et al., 2010; Osborne et al., 2010; Greening et al., 2014, 2015a,b; Liot and Constant, 2016).

Hydrogenases catalyze microbial H<sup>2</sup> production and consumption and are reversible enzymes responsible for reversible or partial catalytic reactions as follows (Equation 1).

$$\text{H}\_2 \Leftrightarrow \text{H}^- + \text{H}^+ \Leftrightarrow 2\text{H}^+ + 2\text{e}^- \tag{1}$$

On the basis of the metal cofactors of their H2-binding sites, these hydrogenases can be divided into three categories, namely the [NiFe]−, [FeFe]−, and [Fe]-hydrogenases (Vignais and Billoud, 2007; Schilter et al., 2016). [NiFe]-hydrogenases are closely related to both H<sup>2</sup> production and consumption, while [FeFe]-hydrogenases are responsible mainly for the production of H<sup>2</sup> owing to their higher turnover rate and activity compared with [NiFe]-hydrogenases (Marshall et al., 2012). However, [Fe] hydrogenases have so far only been found in methanogenic archaea without cytochromes (Thauer et al., 2010). A recent study shows that the three different types of hydrogenase contain many subgroups based on the properties of metalloenzymes (such as metal-binding motifs, amino acid sequence phylogeny, reported biochemical characteristics and predicted genetic organization), and hydrogenase-encoding genes have also been identified in many microorganisms indicating a broad ecological distribution (Trchounian et al., 2011; Greening et al., 2016). Although the contribution of H<sup>2</sup> metabolism to the entire ecosystem function is recognized in several environments such as hydrothermal vents, anoxic sediments and animal guts (Vignais and Billoud, 2007; Schwartz et al., 2013; Greening et al., 2016), the functions of hydrogenases in ecosystems in general remain largely unknown.

H<sup>2</sup> metabolism plays a vital role in stability and performance in many microbial biotopes at ecosystem level (Marshall et al., 2012). It has been gradually recognized that hydrogenases may be used in bioremediation (Vignais and Billoud, 2007; Jugder et al., 2013). Numerous studies have shown that H<sup>2</sup> can be utilized as an electron donor for reductive dehalogenation by many microorganisms and the occurrence of hydrogenases involved has been reported in dehalogenated bacteria (Seshadri et al., 2005; Rahm et al., 2006; Vignais and Billoud, 2007). In addition, microbial hydrogenases have been used in the remediation of metal-containing industrial wastes for the reduction of potentially toxic metals (Li et al., 2018). Under the impact of hydrogenases, microbial metabolic activities can influence the cycling of belowground minerals and organic matter and play a positive role in the bioremediation of both organic and inorganic pollutants (Lovley, 1993, 2008; Lovley and Coates, 2000; Vignais and Billoud, 2007). Thus, the use of hydrogenases for the remediation of polluted soils might be a promising strategy. In this review we attempt to integrate our understanding of the role of H<sup>2</sup> metabolism in environment and environmental bioremediation processes and summarize the knowledge of H<sup>2</sup> metabolism and hydrogenases involved in bioremediation.

#### MICROBIAL H2-PRODUCING PROCESSES AND THEIR IMPACT ON THE ENVIRONMENT

#### Fermentative Hydrogen Production From Organic Compounds

H<sup>2</sup> is a key compound in the metabolism of many anaerobes, as well as a few aerobes, which owed the capacity to use this energy-rich molecule when it is available in the environment and derive electrons from its oxidation to drive energy generation. In the absence of external electron acceptors, many anaerobic bacteria can exploit carbohydrate rich substrates to produce H<sup>2</sup> by reducing protons continuously. As described previously (Das and Veziroglu, 2001 ˇ ; Das and Veziroglu, 2008; Hallenbeck, 2009), the fermentative process generating H<sup>2</sup> comprises two major pathways. In the first, butane 2,3 diol fermentation or mixed acid fermentation produces H<sup>2</sup> via formate decomposition where glucose is transformed to pyruvate and then releases electrons to produce H<sup>2</sup> under hydrogenase through a series of oxidation and reduction reactions (**Figure 1A**). The second is an NADH pathway in which H<sup>2</sup> is produced by the re-oxidation of NADH (**Figure 1B**). In the various pyruvate metabolic pathways, H<sup>2</sup> is usually produced by butyric acid fermentation, mixed acid fermentation, and bacterial ethanol fermentation (Ren et al., 2005). Fermentative microorganisms such as Clostridium spp. (e.g., C. butyricum and C. acetobutylicum) (Fang et al., 2006; Zhang et al., 2006), rumen flora (e.g., Butyrivibrio fibrisolvens, Eubacterium limosum, Megasphaera elsdenii, Ruminococcus flavefaciens, and Ruminococcus albus) (Miller and Wolin, 1973; Joyner et al., 1977; Miller and Wolin, 1979; Chaucheyras-Durand et al., 2010), Enterobacter spp. (e.g., E. cloacae and E. aerogenes) (Kumar and Das, 2000; Fabiano and Perego, 2002), Pyrococcus furiosus and Thermococcus litoralis (Malik et al., 1989; Rákhely et al., 1999; Schwartz et al., 2013) have been found to effectively produce H<sup>2</sup> via different pyruvate metabolic pathways. Hydrogenase enzymes also play an important role in fermentative H<sup>2</sup> production (Woodward et al., 2000; Trchounian et al., 2012). In general, H<sup>2</sup> production could be catalyzed by a soluble [FeFe]-hydrogenase or a special class of membrane bound [NiFe]-hydrogenase (Ech). For example, Escherichia coli can transform intermediary fermentation products to the gaseous products H<sup>2</sup> and CO<sup>2</sup> by formate hydrogenlyase reaction

FIGURE 1 | Syntrophic interactions between hydrogen-producing and hydrogen-consuming microbes in pollutant degradation. FM, Fermentative microbe; AOM, Anaerobic CO oxidation microbe; NFB, Nitrogen-fixing bacteria; OHRB, Organohalide-respiring bacteria; SRB, Sulfate-reducing bacteria; MRB, Metal-reducing bacteria; ADDB, Azo dyes decolorization bacteria. (A) The progress of formate oxidation coupled to H<sup>2</sup> formation in E. coli (derived from Sawers, 1994; Hallenbeck, 2009; Trchounian et al., 2012). The FHL complex consist of a Ech type membrane-bound H2-evolving [NiFe] hydrogenase coupled to a formate dehydrogenase (Fdh) and membrane integral subunits. (B) The progress of re-oxidation of NADH coupled to H<sup>2</sup> formation in T. roseopersicina (derived from Jenney and Adams, 2008; Maróti et al., 2010). The Hyd is a membrane-bound H2-evolving [FeFe] hydrogenase. Fd, ferredoxin. (C) The progress of anaerobic CO oxidation coupled to H<sup>2</sup> formation in C. hydrogenoformans (derived from Svetlichny et al., 1991; Soboh et al., 2002). The Ech is a membrane-bound H2-evolving [NiFe] hydrogenase. Fd, ferredoxin. (D) The progress of producing H<sup>2</sup> as a byproduct of N<sup>2</sup> fixation. The nitrogenase complex consist of a Fe protein and MoFe protein. (E) The role of H<sup>2</sup> in reductive dechlorination in Dehalococcoides spp. (derived from Jugder et al., 2016). MBH, membrane-bound uptake hydrogenase. Cyt, cytochrome. MK, menaquinone; MKH2, dihydromenaquinone. Rdh, reductive dehalogenase. R-Cl, organohalide. (F) The role of H<sup>2</sup> in reductive PTEs in Desulfovibrio fructosovorans (derived from Chardin et al., 2003; Cao et al., 2014). Hyb is a membrane-bound H2-uptake [NiFe] hydrogenase. Fd, ferredoxin. M, PTEs. MS, metal sulfides. (G) The role of H<sup>2</sup> in reductive azo compounds in Shewanella decolorationis (derived from Hong et al., 2007, 2008). Hya, membrane-bound uptake [NiFe]-hydrogenase. Cyt, cytochrome. MK, menaquinone; MKH2, dihydromenaquinone. AZR, azo reductase.R-N = N-R', azo compounds.

(**Figure 1A**) (Sawers, 1994). Soboh et al. (2004) report that a ferredoxin-dependent [NiFe]-hydrogenase and a NADHdependent [Fe]-hydrogenase may catalyze H<sup>2</sup> evolution from NADH in Thermoanaerobacter tengcongensis. Production of H<sup>2</sup> by fermentation in Thermotoga maritima is catalyzed by a heterotrimeric [FeFe]-hydrogenase and two cytoplasmic [NiFe] hydrogenases have been identified in Thiocapsa roseopersicina (**Figure 1B**) (Jenney and Adams, 2008; Maróti et al., 2010).

#### Anaerobic Carbonic Monoxide (CO) Oxidation

There are several microbes owing different types of hydrogenogens that grow anaerobically in the dark and can unitize CO as the sole energy source to produce H<sup>2</sup> (**Figure 1C**). Uffen (1976) and Fox et al. (1996a,b) showed that Rhodospirillum rubrum can produce H<sup>2</sup> by oxidation of CO with the reduction of protons under the catalysis of a complex enzyme consisting of a CO-insensitive [NiFe]-hydrogenase and carbon monoxide dehydrogenase. A typical example of this group is the strictly anaerobic Carboxydothermus hydrogenoformans which contains a multienzyme membrane-bound [NiFe]-hydrogenase (Ech) complex (**Figure 1C**) (Svetlichny et al., 1991; Soboh et al., 2002). These enzymes together can oxidize CO and subsequently reduce the protons derived from H2O to form molecular H2. Carboxydocella thermautotrophica (Sokolova et al., 2002), Thermosinus carboxydivorans (Sokolova et al., 2004), Thermincola carboxydiphila (Sokolova et al., 2005), and

Thermolithobacter carboxydivorans (Sokolova et al., 2007) are also thermophilic hydrogenogens.

#### Production of H<sup>2</sup> as a Byproduct of N<sup>2</sup> Fixation

Nitrogen fixation is one of the main processes of biogenic H<sup>2</sup> production and is catalyzed by nitrogenase (**Figure 1D**). Approximately 30–50% of the entire reduction power consumed by nitrogenase is laterally tracked to H<sup>2</sup> evolution (Brewin, 1984; Evans et al., 1987). However, H<sup>2</sup> is not both a competitive inhibitor of N<sup>2</sup> fixation and also represents a net loss of energy unless the H<sup>2</sup> can be reprocessed by means of the uptake hydrogenase (Kosourov et al., 2014). Many H2-utilizing microorganisms such as the aerobic H2-oxidizing bacteria in soils reduce the loss of energy (Stein et al., 2005; Maimaiti et al., 2007; Constant et al., 2008; Osborne et al., 2010; Annan et al., 2012; Greening et al., 2015b). Many rhizobia can symbiotically fix dinitrogen in the root nodules of legumes and produce H<sup>2</sup> concomitantly. The most-studied symbiotic nitrogen-fixing bacteria in legumes include Bradyrhizobium japonicum, Mesorhizobium mediterraneum, Sinorhizobium meliloti, and Rhizobium leguminosarum (Nour et al., 1995; Spaink, 2000; Laranjo et al., 2014). In addition, strains of Azotobacter (Cocking, 2003) and various cyanobacteria (e.g., Anabaena cylindrica, Nostoc muscorum, and Westiellopsis prolica) (Bulen et al., 1965; Fay, 1992; Nandi and Sengupta, 1998; Das and Veziroglu, 2001 ˇ ; Das and Veziroglu, 2008) can produce hydrogen through the nitrogen fixation process. Gest and Kamen (1949) report that Rhodospirillum rubrum can evolve significant amounts of H<sup>2</sup> in the light and this is termed the photoproduction of H<sup>2</sup> caused by nitrogenase-catalyzed reduction of protons (Bulen et al., 1965). Photoproduction of H<sup>2</sup> was subsequently discovered in other phototrophic bacteria such as Rhodobater capsulatus, Rhodobater sphaeroides, Rhodobater palustris, Thiocapsa roseopersicina, and Halobacterium halobium (Vincenzini et al., 1982; Gogotov et al., 1991; Khan and Bhatt, 1991; Krahn et al., 1996; Fascetti et al., 1998).

#### Effects of Microbial Hydrogen Production on Environment

Atmospheric H<sup>2</sup> is derived mainly from anthropogenic activities and oxidation of atmospheric methane (CH4) and non-methane hydrocarbons. An H<sup>2</sup> mixing ratio of 0.53 ppmv is typically found in the global atmosphere (Novelli et al., 1999) and participates in atmospheric chemical cycles of H2O and greenhouse gasses as well as various pollutants (Schlegel et al., 1976; Crutzen and Fishman, 1977; Salvi and Subramanian, 2015; Talibi et al., 2017). In addition, H<sup>2</sup> is a potential future energy carrier that may significantly affect the atmospheric H<sup>2</sup> budget when used on a large scale (Brenninkmeijer et al., 2003; Petersen et al., 2011). It has been estimated that the total amount of H<sup>2</sup> emissions into the troposphere each year is approximately 107 Tg (Rhee et al., 2006). Tromp et al. (2003) reported that the concentrations of stratospheric H2O and ozone and stratospheric temperatures would be affected by these H<sup>2</sup> emissions. Moreover, the potential impacts of an increase in anthropogenic H<sup>2</sup> emissions on the concentration of other trace gasses such as CH<sup>4</sup> and CO) are also proposed.

About 7–11% of the global H<sup>2</sup> pool is contributed by all oceanic, lake, and soil organisms (Schwartz et al., 2013). 'Hot spots' can be found in hypersaline cyanobacterial mats, with the release of H<sup>2</sup> concentrations between 16,000 and 90,000 ppmv (Nielsen et al., 2015), which might be the main source of H<sup>2</sup> emission from lake surface waters to the atmosphere. Numerous studies show that both cell counts of cyanobacteria and their N<sup>2</sup> fixation rates are correlated with the H<sup>2</sup> concentration of lake water (Conrad et al., 1983; Schütz et al., 1988; Schmidt and Conrad, 1993). Furthermore, the production of fermentation H<sup>2</sup> and organic acids is a key component in the biogeochemistry of microbial mats, which promotes close interactions between anoxygenic phototrophs, cyanobacteria and heterotrophic bacteria (Otaki et al., 2012; Lee et al., 2014; Nielsen et al., 2015). However, almost all of the H<sup>2</sup> produced from hypoxic sediments is also consumed by the sediments (Schwartz et al., 2013). The effects of hydrogen consumption on microbial communities in sediments therefore deserve further study.

The contribution of soils to the atmospheric H<sup>2</sup> reservoir is more complex because soils are the main sink of the global H<sup>2</sup> cycle, accounting for about 75 to 80% of atmospheric absorption (Constant et al., 2009; Ehhalt and Rohrer, 2009). However, nitrogen-fixing bacteria that form symbioses with legumes or free-living N<sup>2</sup> fixing bacteria can generate large amounts of H<sup>2</sup> as a by-product during N<sup>2</sup> fixation (Orr et al., 2011; Mus et al., 2016). It has been estimated that H<sup>2</sup> concentrations inside N2-fixing legume nodules range from 9,000 to 27,000 ppmv (Hunt et al., 1988; Witty, 1991; Witty and Minchin, 1998), so that diffusion losses during the growing season might reach 240,000 L H<sup>2</sup> (Dong et al., 2003). Thus, the intensity of these H<sup>2</sup> emissions to soils is determined by the hydrogen-metabolic capabilities of rhizobacterial symbionts (Hup<sup>+</sup> or Hup<sup>−</sup> genotypes) in nodules through an uptake [NiFe]-hydrogenase (Evans et al., 1988; Annan et al., 2012). In the Hup<sup>+</sup> legume rhizosphere the energy of H<sup>2</sup> can be recycled by the [NiFe]-hydrogenase, while H<sup>2</sup> is released into the surrounding soil in the Hup<sup>−</sup> legume rhizosphere. There is thus increasing evidence that H<sup>2</sup> released into surrounding soils plays a key role in increasing plant biomass via the enrichment of aerobic H2-oxidizing bacteria (HOB), or plant growth-promoting rhizobacteria (PGPR) in both legumes and non-legumes (Dong et al., 2003; Maimaiti et al., 2007). Different H<sup>2</sup> mixing ratios found in natural ecosystems may indeed lead to changes in soil microbial community structure and coordinated feedback of community functions. Constant et al. (2008) found that soil actinomycetes (such as Streptomyces sp. PCB7) are the main users of trace level of H<sup>2</sup> in soils and might be key contributors to the function of soils as a sink in the global H<sup>2</sup> cycle. Subsequently, Khdhiri et al. (2017) validated their own hypothesis by showing that the taxonomic response of the soil microbial community composition to H<sup>2</sup> exposure is inconsistent across land use types. Piché-Choquette et al. (2018) revealed that H<sup>2</sup> supports metabolic and energetic flexibility in microorganisms supplying a variety of ecosystem services via dose-response relationships between environmentally relevant H<sup>2</sup> concentrations and the biological sinks of H2, CH4, and CO in soils.

#### ROLE OF H<sup>2</sup> IN ENVIRONMENTAL BIOREMEDIATION

fmicb-10-00106 February 14, 2019 Time: 16:50 # 5

The H<sup>2</sup> produced both biogenically and abiogenically can be released and provided to support for the growth and metabolism of hydrogenotrophic prokaryotes (Karyakin et al., 2007). H<sup>2</sup> metabolism fulfills a critical role in the ecosystems of many microbial biotopes (Vignais and Billoud, 2007; Schwartz et al., 2013; Greening et al., 2016). It is currently considered that a wide range of microbial groups employ H<sup>2</sup> as an electron donor to catalyze the reduction of pollutants such as organohalides, azo compounds and potentially toxic elements.

#### Organohalides

Organohalides are recalcitrant, toxic, highly persistent, globally prevalent, and carcinogenic environmental contaminants. Organohalide-respiring bacteria (OHRB) have been isolated from polluted soils, sludges, sediments, aquifers, freshwaters, and marine habitats, and they are of considerable importance in bioremediation processes and natural halogen cycles (Zanaroli et al., 2015). Most OHRB discovered to date belong to Desulfomonile, Dehalococcoides, Dehalobacter, Desulfitobacterium, Desulfuromonas, and Sulfurospirillum (formerly Dehalospirillum) as reviewed by Jugder et al. (2015). Reductive dechlorination is an anaerobic respiration process that utilizes H<sup>2</sup> as electron donor to dehalorespire these halogenated organics (**Figure 1E**) (Zanaroli et al., 2015; Agarwal et al., 2017). The process of electron via electron transport phosphorylation from the oxidation of the H<sup>2</sup> to reductive dechlorination of organohalides involving membrane associated oxidoreductases (**Figure 1E**) (Jugder et al., 2016). Membrane-bound hydrogenases (MBH) are the initial oxidizers to take up the electrons released from molecular H2, which play a vital role in organohalide respiration (Jugder et al., 2013, 2015, 2016). The reductive dehalogenation of organohalides is typically catalyzed by dehalogenating enzyme systems coupled to ATP synthesis, reductive dehalogenases (Rdases) replace the halogen substituent with a hydrogen atom, reducing the toxicity and recalcitrance to biodegradation (**Figure 1E**) (Adrian and Loeffler, 2016; Gevorgyan et al., 2018). Sequencing data of genomes reveal that OHRB possess as many as 36 putative Rdases. After transformation to lower halogenated organics under anaerobic conditions, these toxic compounds can subsequently be mineralized by aerobic bacteria (Jugder et al., 2015).Desulfomonile tiedjei strain DCB-1 is one of the best-described dechlorinating anaerobes. The strain, first discovered by Suflita et al. (1982), reductively dechlorinates 3-chlorobenzoate while replacing the chlorine atom with hydrogen from H2, whereby providing energy for bacterial growth (Shelton and Tiedje, 1984; Dolfing and Tiedje, 1986; Dolfing and Tiedje, 1987). The strain was then noted to

consume H<sup>2</sup> with 3-chloro-, 3-bromo-, 3-iodobenzoate, tetrachloroethene (PCE), trichloroethene (TCE) (Cole et al., 1995), and chlorophenols (Mohn and Kennedy, 1992) as electron acceptors (DeWeerd et al., 1991). During the dehalogenation of 3-chlorobenzoate, formate was the most effective electron donor, followed by H2, pyruvate, and acetate.

Dehalococcoides strains are also some of the best known species capable of reductively dechlorinating a wide range of haloorganics including chlorinated benzenes, biphenyls, dioxins, ethenes, naphthalenes, and brominated diphenyl ethers. For example, tetrachloroethene is a commonly used solvent that possesses high toxicity and is a suspect carcinogen. The complete reductive dechlorination of tetrachloroethylene (PCE) and trichloroethylene (TCE) to non-toxic ethylene was first observed under methanogenic conditions by mixed cultures (Freedman and Gossett, 1989). Although H<sup>2</sup> also served as the electron donor, methanol was more effective in sustaining the reductive dechlorination process. Holliger et al. (1993) isolated an anaerobic bacterial culture, previously named as PER-K23, from an anaerobic packed-bed column. By using H<sup>2</sup> and formate as the only electron donors, PCE or TCE was reductively transformed to ethane via cis-1,2-dichloroethene (cis-1,2-DCE), chloroethene, and ethene, which was coupled to bacterial growth. The key role of hydrogenases in metabolizing these pollutants is underscored by the fact that both uptake (Hup type) and energy-conserving hydrogenases (Hyc or Ech type) were found in the genome of Dehalobacter restrictus PER-K23 (Rupakula et al., 2013). Maymó-Gatell et al. (1997) then isolated a dehalogenator, strain 195, and characterized it as Dehalococcoides ethenogenes. To date, Dehalococcoides species are the only bacteria known to be capable of completely dechlorinating chloroethylene. Genomic analysis of Dehalococcoides ethenogenes 195 showed that several hydrogenase genes including the membrane-bound periplasmic Hup, cytoplasmic Vhu, and membrane-bound Ech and Hyc [NiFe]-hydrogenases (Groups 1, 3, 4, and 4, respectively), and a membrane-bound Hym [Fe]-hydrogenases has potential roles in electron transport, which are capable of completing anaerobic dechlorination of the solvents PCE and TCE to vinyl chloride (VC) and ethane (Vignais et al., 2001; Morris et al., 2006).

Unlike other halorespiring bacteria, Dehalococcoides spp. use only H<sup>2</sup> as an obligate electron donor for the dechlorination reaction, and no other electron acceptors support growth. For example, D. ethenogenes strain 195 grew only on H<sup>2</sup> as electron donor for both bacterial growth and PCE reduction rather than formate, lactate, methanol, ethanol, glucose, pyruvate, or yeast extract (Maymó-Gatell et al., 1997). In addition, Dehalococcoides sp. CBDB1 was the first purified isolate of a bacterium relying on the energy obtained from stoichiometrical dehalorespiration of chlorobenzenes (CB) such as 1,2,3-trichlorobenzene (TCB) and 1,2,3,4-tetrachlorobenzene (TeCB) (Adrian et al., 2000). Both Dehalococcoides sp. 195 and CBDB1 exhibit reductive dehalogenation of chlorophenols (Adrian et al., 2007). Kube et al. (2005) compared the genome sequence of Dehalococcoides sp. CBDB1 with Dehalococcoides ethenogenes strain 195 and revealed that the hydrogenases previously described for strain

195 are also present in strain CBDB1. Chloroform (CF, CHCl3) is a non-polar solvent that is ubiquitous and is toxic to humans. The biodegradation of CF involves two processes, (1) dehalorespiration in which CF is dechlorinated to dichloromethane (DCM, CH2Cl2) by employing H<sup>2</sup> as electron donor under the action of uptake hydrogenase, and (2) a fermentative process in which DCM is transformed to H2, acetate and carbon dioxide. Lee et al. (2012) report the involvement of Dehalobacter in dehalorespiration of CF [Equation (2)].

$$\text{CHCl}\_3 + \text{H}\_2 \rightarrow \text{CH}\_2\text{Cl}\_2 + \text{H}^+ + \text{Cl}^-\tag{2}$$

Despite these findings in respiration of organohalides, there is no real consensus on the involvement of various membrane associated components.

#### Potentially Toxic Elements (PTEs)

Potentially toxic elements display environmental durability, biological accumulation, and potential biological toxicity. The remediation of PTEs can be achieved by sulfate-reducing bacteria (SRB) or metal-reducing bacteria that can utilize H<sup>2</sup> or other organic compounds as terminal electron donors to reduce the PTEs. Tebo and Obraztsova (1998) isolated the first sulfate-reducing bacterium from PTE-polluted sediments named Desulfotomaculum reducens sp. nov. strain MI-1, which can utilize H<sup>2</sup> as terminal electron donor and metals [such as Cr(VI), Mn(IV), Fe(III), and U(VI)] as electron acceptors accompanied by bacterial growth. Thus far, more than 40 SRB species have been identified, including Desulfobacter, Desulfovibrio, Desulfotomaculum and Desulfomicrobium, and others (Leloup et al., 2009; Mizuno et al., 2012; Hussain et al., 2016; Li et al., 2018). Subsequently, due to the advantages of SRB (no secondary pollution and strong adaptability), they have been used in the bioremediation of PTEs (Li et al., 2018). Generally speaking, there are two steps involved in the mechanism of SRB removal of PTEs from wastewaters: (i) SRB utilize sulfate as electron acceptor to oxidize simple organic compounds to generate bicarbonate ion and hydrogen sulfide under anaerobic conditions [Equation (3)], and (ii) the hydrogen sulfide generated reacts with dissolved PTE to form insoluble metal sulfide precipitates [Equation (4)] (Kieu et al., 2011; Singh et al., 2011; Li et al., 2017).

> 2CH2<sup>O</sup> <sup>+</sup> SO2<sup>−</sup> <sup>4</sup> <sup>→</sup> 2HCO<sup>−</sup> <sup>3</sup> +H2S (3)

$$\text{H}\_2\text{S} + \text{M}^{2+} \rightarrow 2\text{H}^+ + \text{MS(S)}\tag{4}$$

Where CH2O represents simple organic compounds (such as acetate and lactate), M represents PTEs, and MS represents metal sulfides. Because of their special characteristics with the corresponding metal sulfides readily forming precipitates, SRBs have been used to treat PTE-polluted wastewaters (e.g., uranium-containing, chromium-containing and antimonycontaining wastewaters, organochlorines, and other pollutants) (Li et al., 2018). Lovley and Phillips (1994) showed that the bioremediation effect of Desulfovibrio vulgaris which utilizes H<sup>2</sup> as the electron donor catalyzed by the c<sup>3</sup> cytochrome functions as a Cr(VI) reductase in Cr(VI)-contaminated waters was superior to the previously described Cr(VI) reductive microorganisms. Kieu et al. (2011) reported that the PTE removal efficiencies of Cu2+, Ni2+, Zn2<sup>+</sup> , and Cr6<sup>+</sup> in anaerobic semi-continuous stirred tank reactors containing a consortium of SRB reached 94–100% after 4 weeks under experimental conditions. In addition, several microbial genera reduced uranium to form easily precipitated reduced U(IV) species, and this has been used successfully in soil remediation (Phillips et al., 1995; Fredrickson et al., 2000; Valls and De Lorenzo, 2002).

Several uptake hydrogenases were considered to have potential application in the bioremediation of PTEs. The [NiFe] uptake hydrogenases in group 1 including membrane-bound respiratory uptake hydrogenases that couple H<sup>2</sup> oxidation to catalyze metal reduction (**Figure 1F**). For example, [NiFe]-uptake hydrogenase from SRB can reduce toxic chromate VI to form a less toxic product (Chardin et al., 2003). In addition, technetium VII is reduced by Desulfovibrio fructosovorans through this mechanism (Tabak et al., 2005), and hydrogenases involving in metal reduction have also been observed in other metals including ferrum (Fe) (Coppi et al., 2004), platinum (Riddin et al., 2009), and lead (Deplanche et al., 2010). A comprehensive analysis of the genome sequence of the metal-reducing bacterium (Shewanella oneidensis) has been conducted, and has predicted that an [Fe]-hydrogenase and several cytochromes are involved in the electron transport and metal reduction processes (Heidelberg et al., 2002). However, the potential application of microbes with different subgroup hydrogenases for PTE respiration is not enough, requiring further study including the biochemical investigations of these different subgroup hydrogenases.

#### Other Pollutants

Azo compounds undergo dissimilatory azoreduction by Shewanella decolorationis S12 under anaerobic conditions. This strain utilized azo compounds as carbon source for growth by azo reductase which is sustained by the H<sup>2</sup> supply. The strain also catalyzed H2-dependent reduction of Fe(III) and humic substances (Coppi et al., 2004; Hong et al., 2008). Brigé et al. (2008) show that Shewanella decolorationis MR-1 utilized azo dye amaranth as electron acceptor for microbial energy conservation. Mutambanengwe et al. (2007) show the decolorization of a wide range of azo dyes with sulfate-reducing microbes (SRM) and hydrogenases might be involved in the degradation process. A multicomponent electron transfer chain has been proposed to be involved in the extracellular reduction of azo compounds. The electron transfer components consist of the cytoplasm/outer membrane, periplasm, c-type cytochromes, and menaquinone (Hong et al., 2007; Brigé et al., 2008). Hya type [NiFe]-hydrogenase or Hyd type [Fe]- hydrogenase act as a critical hub mediating the oxidization of H<sup>2</sup> to provide electrons for azoreduction metabolism (**Figure 1G**) (Hong et al., 2008).

H2-dependent reduction has been reported in nitroaromatic compounds (Watrous et al., 2003). In a strict anaerobe, Clostridium acetobotulinicum, an [Fe]-hydrogenase is responsible for the reduction of nitro substituents of 2,4,6-trinitrotoluene

(TNT) to the corresponding hydroxylamine in an acidogenic environment.

#### Factors Affecting the Utilization of Hydrogen by Degrading Bacteria in the Environment

There are many factors affecting the utilization of H<sup>2</sup> by degrading bacteria in the environment such as H<sup>2</sup> source, H<sup>2</sup> transfer process and other environmental factors (including trophic hierarchies, external pH, osmotic coditions, concentration of carbon sources and their mixtures and microbial community and other physicochemical factors).

Methanogens were found to affect the interspecies H<sup>2</sup> transfer of dehalorespiring bacteria, which might promote or inhibit the dechlorination process (Smatlak et al., 1996; Fennell et al., 1997; Yang and McCarty, 1998). Johnson et al. (2008) demonstrated the dechlorination of stress-related net cell growth by Dehalococcoides ethenogenes strain 195 (DE195) which was isolated and then transited to a smooth phase. Although Methanobacterium congolense (MC) can compete with DE195 for hydrogen, adverse effects of the dechlorination rate were not observed (Men et al., 2012). This is mainly because the H<sup>2</sup> threshold required for dechlorination is very low, so that even though methane production consumes a large amount of H2, it does not compete for dechlorination (Yang and McCarty, 1998; Men et al., 2012). In syntrophic communities, H2 producing bacteria and H2-consuming methanogens perceive the redox conditions and affect each other's metabolism (Stams and Plugge, 2009). Several studies have shown that the reduction dechlorination can be promoted in some communities in the presence of methanogens (Vogel and McCARTY, 1985; Heimann et al., 2006; Kong et al., 2014). In addition, a recent study found that Methylobacter seemed to be tolerant to TCE and may play a vital role in TCE degradation (Kong et al., 2014). Although many studies have assessed the association between methanogens and dechlorination bacteria, the mechanism by which methanogens affect dechlorinating communities remains unclear.

The process of forming compact aggregates involves both physicochemical and biological interactions (Stams and Plugge, 2009). When the compact aggregates are formed in anaerobic bacteria and methanogenic archaea, the rate of H<sup>2</sup> transfer between two species increases significantly (Lettinga et al., 1988; Stams and Plugge, 2009). Several studies have also shown that the inter-microbial distances affect both their specific growth rates and biodegradation rates (Ishii et al., 2005; Stams et al., 2006; Stams and Plugge, 2009). Thus, forming compact aggregates might be an important factor influencing the biodegradation rates of degrading bacteria.

It is well known that trophic hierarchies occur because different functional members of the community provide each other with a matrix and basic cofactors, and eliminate inhibitory metabolites (Schink, 1997; Rittmann and McCarty, 2012). DeWeerd et al. (1991) reported that acetylene, molybdate, selenate, and metronidazole can inhibit dehalogenation, sulfite reduction and H<sup>2</sup> metabolism, indicating that the reduction of sulfite and dehalogenation may share part of the same electron transport chain. However, some environmental factors might accelerate the degradation of pollutants by promoting H<sup>2</sup> utilization. For example, cobalamin has a positive effect on the dechlorination process as a co-factor of the reductive dehalogenases (Yan et al., 2012). Desulfovibrio vulgaris Hildenborough (DVH) possesses the full set of genes required for the biosynthesis of adenosylcobalamin, a derivative of vitamin B12 which might result in an increased concentration of the corrinoid co-factor (vitamin B12) in co-cultures, taken up and utilized immediately by Dehalococcoides species (Rodionov et al., 2004). In addition, the main factors influencing H<sup>2</sup> utilization such as external pH, osmotic conditions, concentration of carbon sources and their mixtures, microbial community and other physicochemical factors mainly affected growth and the physiological activity including uptake hydrogenase and pollutant degrading enzymes of the degrading bacteria (Richter and Gescher, 2014; Trchounian and Trchounian, 2014, 2015; Trchounian et al., 2017a).

#### INTERSPECIES HYDROGEN TRANSFER DURING SYNTROPHIC GROWTH

Syntrophy coupling mutualistic interactions between hydrogen- /formate-producing and hydrogen-/formate-consuming microorganisms is essential for biofuel production, pollutant degradation, and global carbon cycling (Kleinsteuber et al., 2012; Sieber et al., 2012; Morris et al., 2013). When sulfate is limited or unavailable, SRBs can also mediate the transfer of H<sup>2</sup> between species, which provides the bacterial species with a very versatile metabolism adapted to complex ecological environments. Odom and Peck (1981) first documented the transfer of the redundant H<sup>2</sup> evolved from substrate fermentation by SRBs to other H<sup>2</sup> consuming bacteria. Using a defined two-member continuous culture, Drzyzga et al. (2001) demonstrated that the sulfate reducer Desulfovibrio sp. strain SULF1 can use the dehalorespiring Desulfitobacterium frappieri TCE1 as a 'biological electron acceptor' to sustain growth. They also noted that dehalogenation of tetrachloroethene (PCE) was inhibited at sulfate concentrations above 2.5 mM, while PCE was completely dehalogenated to cis-dichloroethene (cis-DCE) with 1 mM sulfate or without sulfate addition (Drzyzga and Gottschal, 2002). In this community, Desulfovibrio vulgaris Hildenborough (DVH) can grow syntrophically with Dehalococcoides ethenogenes strain 195 (DE195), thus enhancing the robustness of bacterial growth and the dechlorination activity of trichloroethene (Men et al., 2012). The syntrophical interaction with sulfate reducers has been shown to result in more effective transfer of H2, thereby facilitating faster dechlorination and more rubust growth of dehalogenating strains compared with gaseous H<sup>2</sup> (Men et al., 2012). The syntrophic relationship between methanogens and archaea also involves interspecies H<sup>2</sup> transfer in the process of converting long-chain fatty acids (Stams and Plugge, 2009). Subsequently, Ziels et al. (2017) found several formate hydrogenases and dehydrogenases in the enriched genome bins (GBs) of both their codigesters. In the process of CF dechlorination, interspecies H<sup>2</sup> transfer was observed in the

form of acetogenesis and methanogenesis by Lee et al. (2012), which required syntrophic partners to maintain low H<sup>2</sup> partial pressures.

The possible processes of syntrophic interactions between H2 producing and H2-consuming microbes in pollutant degradation are shown in **Figure 1**. Previous studies have shown that H2-forming bacteria and H2-utilizing bacteria sense redox conditions, influencing each other's metabolism in syntrophic communities (Stams and Plugge, 2009). Interspecies electron transfer mechanisms underlie thermodynamically favorable syntrophic processes (Gieg et al., 2014). In anoxic environments, butyrate oxidations involving energy-dependent reactions were possible to be applied in syntrophic degradation of organohalides. For example, the standard free reaction enthalpies (1G o0 ) of butyrate oxidations and organohalide degradations were as follows [Equation (5) Müller et al., 2010; Equation (6) Jugder et al., 2016]:

$$\text{Butyate}^- + 2\text{H}\_2\text{O} \rightarrow 2\text{Acetalate}^- + \text{H}^+ + 2\text{H}\_2\text{(}\Delta\text{G}^{\circ'}$$

$$=\ +48.3\text{K}\text{/mol}\tag{5}$$

H<sup>2</sup> + R − Cl → R − H + HCl(1G o 0 = − 131to − 192KJ/mol) (6)

Based on energy balance toward H<sup>2</sup> production and consumption analysis, we propose that the energy-transforming reactions between H<sup>2</sup> production and organohalide degradations might be involved in syntrophic H<sup>2</sup> production and consumption microorganisms. Dehalogenating microorganisms (such as Dehalococcoides sp. strain BAV1 and Dehalococcoides ethenogenes strain 195) can utilize acetate as carbon source and H<sup>2</sup> as electron donor when grown in isolation, exhibiting limited dechlorination activity and low growth rates (He et al., 2003a,b). Thus, a promising method might be to develop improved bioremediation strategies by enhancing the strong growth and dechlorination activity of dehalogenating microorganisms (Men et al., 2012). However, many interspecies H<sup>2</sup> transfer interactions are syntrophic, and thus only present in complex microbial communities but not in pure cultures. In complex microbial consortia, H<sup>2</sup> indirectly mediates electron shuttle between electron donors and acceptors. Hydrogenotrophic bacteria can profit from the H<sup>2</sup> produced from their syntrophic partners, thereby transforming pollutants. Thus, both H2-producing and H2-consuming microorganisms are essential for their own growth and might also promote the degradation of pollutants (Stams and Plugge, 2009).

#### CONCLUSION AND PERSPECTIVES

Metabolism of H<sup>2</sup> including H<sup>2</sup> production and H<sup>2</sup> consumption have been recognized as a potential driving force affecting the structure of microbial communities and may even change community functions. Although the contribution of H<sup>2</sup> metabolism to entire ecosystem processes is recognized in hydrothermal vents, anoxic sediments and animal guts (Vignais and Billoud, 2007; Schwartz et al., 2013), the role of H<sup>2</sup> metabolism and hydrogenases in ecosystems are not fully elucidated. Further advances in exploiting the function of biohydrogen metabolism and related microbial communities in environmental bioremediation are expected to result from (i) using metagenome sequencing, single-gene fluorescence in situ hybridization, the functional gene arrays (GeoChip) and in situ mass spectrometry to track the dynamics of pollutant-degrading bacteria involving in H<sup>2</sup> metabolism and the interplay between pollutant-degrading bacteria and H2-metabolism bacteria in degradation process; (ii) effects of soil conditions on H2-consuming microorganisms degrading pollutants; (iii) structural studies of hydrogenases or the synergistic action of other enzymes (such as ATPase and Rdase) involving in the process of environmental bioremediation and enhancing these enzymes activity through protein engineering; (iv) integrative analyses of genomic, transcriptomic, and epigenomic data in these environmental bioremediation process.

To date, environmentally friendly management techniques named "3B" techniques (biological carbon sequestration, bioenergy, and bioremediation) have been proposed to further enhance biodiversity and mitigate environmental stressors (Teng et al., 2012). Environmental H<sup>2</sup> is an energy source for aerobic H<sup>2</sup> oxidizers, sulfate reducers, acetogens and methanogens and is also a source of reducing power for anaerobic bacteria and anoxygenic phototrophs (Schwartz et al., 2013). Syntrophy coupling mutualistic interactions between H2-producing and H2-consuming microorganisms is not restricted to the transfer of reducing agents such as H<sup>2</sup> or formate, but can also involve the exchange of organic, sulfurous and nitrogenous compounds or the removal of toxic compounds. Nevertheless, there is still a considerable need for appropriate research initiatives to apply those microbial groups to the bioremediation of contaminated soils. However, soil is a complex and dynamic biological system. From the soil to the microorganism, bioavailability of pollutants involves a full process of adsorption and desorption, transport, and uptake by microorganisms which are also affected by the soil conditions such as soil organic matter, soil minerals, soil moisture, soil aggregates and so on (Ren et al., 2018; Teng and Chen, 2019).

Proton ATPase or other membrane bound secondary transporters affect hydrogenase activity and thus H<sup>2</sup> metabolism (Trchounian et al., 2011; Gevorgyan et al., 2018). So, structural studies of hydrogenases or other synergistic enzymes (such as ATPase and Rdase) involving in the process of environmental bioremediation are vital important in directing protein engineering, for example, in rendering these enzymes activity to promote the degradation efficiency of pollutants via identification of factors linked to the protein environment of the active site. Studies of H<sup>2</sup> metabolism and regulation will also be important in engineering microorganisms at the cellular level to maximize the degradation efficiency of pollutants. Since hydrogenases and other synergistic enzymes have been shown to play an important role in the degradation of pollutants, it is also tempting to consider that analysis of genomic, transcriptomic,

and epigenomic data of these enzymes in environmental bioremediation process will likely provide vital insights into the hydrogenase participates in degradation mechanism of pollutants.

In conclusion, this review provides a comprehensive framework for H<sup>2</sup> production and H<sup>2</sup> consumption in environmental bioremediation processes. The syntrophy coupling mutualistic interaction between H2-producing and H2 consuming microorganisms could be applied to the removal of toxic compounds. In addition, several uptake hydrogenases are also considered to have potential application in the bioremediation of those toxic compounds. The use of this bioenergy may provide a low-input and ecologically friendly bioremediation strategy for the future.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

YT, YX, and XW collected the data. YT and YX drafted the article. YT, XW, and PC critically revised the article.

#### FUNDING

This study was funded by the Outstanding Youth Fund of Jiangsu Province (No. BK20150049), the National Natural Science Foundation of China (Grant Nos. 41671327 and 41571308), and the Special Project on the Basis of the National Science and Technology of China (2015FY110700).





H2 storage. Annu. Rev. Biochem. 79, 507–536. doi: 10.1146/annurev.biochem. 030508.152103



**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 Teng, Xu, Wang and Christie. 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.

## The Effect of a Tropical Climate on Available Nutrient Resources to Springs in Ophiolite-Hosted, Deep Biosphere Ecosystems in the Philippines

D'Arcy R. Meyer-Dombard<sup>1</sup> \*, Magdelena R. Osburn<sup>2</sup> , Dawn Cardace<sup>3</sup> and Carlo A. Arcilla<sup>4</sup>

<sup>1</sup> Department of Earth and Environmental Sciences, The University of Illinois at Chicago, Chicago, IL, United States, <sup>2</sup> Department of Earth and Planetary Sciences, Northwestern University, Evanston, IL, United States, <sup>3</sup> Department of Geosciences, The University of Rhode Island, Kingston, RI, United States, <sup>4</sup> Director of Science and Technology-Philippine Nuclear Research Institute, Manilla, Philippines

#### Edited by:

Eric Boyd, Montana State University, United States

#### Reviewed by:

Ruth Hindshaw, University of Cambridge, United Kingdom Piyali Chanda, University of Wisconsin–Madison, United States

> \*Correspondence: D'Arcy R. Meyer-Dombard drmd@uic.edu

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 02 October 2018 Accepted: 26 March 2019 Published: 01 May 2019

#### Citation:

Meyer-Dombard DR, Osburn MR, Cardace D and Arcilla CA (2019) The Effect of a Tropical Climate on Available Nutrient Resources to Springs in Ophiolite-Hosted, Deep Biosphere Ecosystems in the Philippines. Front. Microbiol. 10:761. doi: 10.3389/fmicb.2019.00761 Springs hosted in ophiolites are often affected by serpentinization processes. The characteristically low DIC and high CH<sup>4</sup> and H<sup>2</sup> gas concentrations of serpentinizing ecosystems have led to interest in hydrogen based metabolisms in these subsurface biomes. However, a true subsurface signature can be difficult to identify in surface expressions such as serpentinizing springs. Here, we explore carbon and nitrogen resources in serpentinization impacted springs in the tropical climate of the Zambales and Palawan ophiolites in the Philippines, with a focus on surface vs. subsurface processes and exogenous vs. endogenous nutrient input. Isotopic signatures in spring fluids, biomass, and carbonates were examined to identify sources and sinks of carbon and nitrogen, carbonate geochemistry, and the effect of seasonal precipitation. Seasonality affected biomass production in both low flow and high flow spring systems. Changes in meteorological precipitation affected δ <sup>13</sup>CDIC and δ <sup>13</sup>CDOC values of the spring fluids, which reflected seasonal gain/loss of atmospheric influence and changes in exogenous DOC input. The primary carbon source in high flow systems was variable, with DOC contributing to biomass in many springs, and a mix of DIC and carbonates contributing to biomass in select locations. However, primary carbon resources in low flow systems may depend more on endogenous than exogenous carbon, even in high precipitation seasons. Isotopic evidence for nitrogen fixation was identified, with seasonal influence only seen in low flow systems. Carbonate formation was found to occur as a mixture of recrystallization/recycling of older carbonates and rapid mineral precipitation (depending on the system), with highly δ <sup>13</sup>C and δ <sup>18</sup>O depleted carbonates occurring in many locations. Subsurface signatures (e.g., low DOC influence on Cbiomass) were most apparent in the driest seasons and lowest flow systems, indicating locations where metabolic processes divorced from surface influences (including hydrogen based metabolisms) are most likely to be occurring.

Keywords: serpentinization, carbon isotopes, carbonates, carbon cycling, nitrogen cycling, deep biosphere

### INTRODUCTION

fmicb-10-00761 April 29, 2019 Time: 17:13 # 2

Recent interest in the terrestrial deep biosphere has been fueled by estimates of biomass (Kallmeyer et al., 2012; Lomstein et al., 2012; McMahon and Parnell, 2014) and reports of unique microbial communities and ecosystem functions (Biddle et al., 2012; Parkes et al., 2014; Solden et al., 2016). The overall impact of the deep biosphere on global biogeochemical cycling remains unknown (Menez et al., 2012), and direct access remains expensive with non-trivial logistics. Surface connected expressions of the terrestrial deep biosphere such as caves, wells, and springs are convenient and less expensive (compared to drilling based endeavors), but elicit questions about authenticity of a subsurface signature. Evidence of subsurface biosphere diversity and function may be overprinted or masked by the oxygenated, photosynthesis-driven, surface biosphere to an unknown degree.

Several studies have attempted to isolate a subsurface signature from terrestrial locations. Arguably the most success has come from studies with direct access to subsurface sampling. These works have relied on comparative metagenomics (e.g., Lau et al., 2014), geochemical modeling integrated with statistical analyses (e.g., Osburn et al., 2014), and innovative culturing techniques (e.g., Rowe et al., 2017) to distinguish subsurface contributions to nutrients, energy, diversity, and biomass. Isolating a subsurface signature when samples are obtained within the surface biome (i.e., springs and seeps) poses additional challenge. More frequently, only individual processes can be linked to the subsurface. For example, the widespread ability to fix nitrogen (Hamilton et al., 2011) or carbon (Osburn et al., 2011; Urschel et al., 2015) using non-photosynthetic pathways in terrestrial hydrothermal systems suggests these traits are maintained due to a general lack of reliable nutrient input from the surface biome.

Stable isotope chemistry of nutrient pools and resident biomass is a useful tool for deep subsurface biosphere applications. Kinetic isotope fractionation associated with biosynthetic machinery discriminates broadly against heavy isotopes, producing <sup>13</sup>C and <sup>15</sup>N- depleted biomass relative to sources. Fractionation varies by process for both carbon and nitrogen isotopes and much can be learned or inferred by comparing isotopic ratios of sources and resulting biomass. For example, it has been shown that different carbon fixation pathways fractionate carbon to differing degrees (Berg et al., 2010; Pearson et al., 2016), and the isotopic composition of nitrogen in biomass is affected by how the organism participates in the nitrogen cycle (Brunner et al., 2013; Mobius, 2013; Frey et al., 2014; Zhang et al., 2014). While there is still much to learn concerning the fractionation of carbon or nitrogen by specific groups of organisms under varying environmental conditions, as well as abiotic considerations (McCollom et al., 2010; Li et al., 2012; Wunderlich et al., 2012; Schrenk and Brazelton, 2013; Zhang et al., 2014) a holistic approach of comparing isotopic ratios of bulk carbon, nitrogen, and biomass can reveal broad ecosystem patterns.

This study focuses specifically on surface expressions (springs and seeps) of the deep biosphere sourced in ultramafic rock units in the Philippines. Springs and seeps emanating from ophiolites represent subsurface fluids produced from the process of "serpentinization" that are mixed with fluids of other sources. Briefly, serpentinization is the aqueous alteration of ultramafic rocks (part of the ophiolite body). In terrestrial ophiolites, the process is often recharged by groundwaters, and provides alteration products including mineralogically altered solids, fluids of distinctive geochemistry, gasses such as H<sup>2</sup> (and possibly CH4), and chemical energy abundant enough to fuel chemosynthetic metabolism. Other resources have thoroughly described and reviewed this process and the resulting potential subsurface habitats (McCollom and Bach, 2009; Schrenk and Brazelton, 2013). Serpentinization has been discussed as a possible process fueling life on other planetary bodies (e.g., Mars and icy ocean worlds – Ehlmann et al., 2010; Vance et al., 2016; Deamer and Damer, 2017), as well as a potential platform for the development of life on Earth (Sleep et al., 2011; Russell et al., 2014). In both of these extensions, it is assumed that serpentinizationdriven life support is divorced from potential carbon and energy sources supplied from photosynthetic processes. Further, serpentinizing springs and seeps are often noted to have low dissolved organic or inorganic carbon (e.g., Morrill et al., 2013; Szponar et al., 2013). The production of H<sup>2</sup> gas from the serpentinization process has indicated the potential for hydrogen driven metabolic processes in impacted environments, with an overlying assumption that the presence of hydrogen gas may indicate a more subsurface "signal." Therefore, separating a true subsurface signal from complicating surface influence at modern, terrestrial, serpentinizing seeps is essential for understanding their utility as a "portal" into the deep subsurface biosphere of modern, ancient, and astrobiological environments.

The isotopic ratios of carbon in total or dissolved inorganic carbon (TIC/DIC) and dissolved organic carbon (DOC) can be compared to that of biomass, allowing conclusions concerning the source of the carbon fueling the biomass production. This approach has been used many times in other terrestrial environments (e.g., Lang et al., 2012; Schubotz et al., 2013; Pearson et al., 2016). There have been several reports from terrestrial serpentinizing seeps and springs that highlight potential carbon and nutrient sources. Both DOC and TIC have been reported in low concentrations in springs located in the Tablelands (Newfoundland, Canada) and The Cedars (CA, United States) areas, with δ <sup>13</sup>CDOC ranging from <sup>−</sup>22 to <sup>−</sup>13h, and <sup>δ</sup> <sup>13</sup>CTIC ranging from −33 to <sup>−</sup>2<sup>h</sup> (Brazelton et al., 2013; Morrill et al., 2013; Szponar et al., 2013). Isotopic composition of carbon from both DIC and DOC in the Yanarta¸s/Chimera site in Turkey are comparable, although concentrations of both are higher than other reported locations (50 and ∼5 ppmC, respectively, Meyer-Dombard et al., 2015). Carbon isotopic compositions of biomass are seldom reported in general for these locations (however, see Meyer-Dombard et al., 2015), but that of the carbonates found in these systems has been widely reported and summarized. The comparison of δ <sup>13</sup>Ccarbonate and δ <sup>18</sup>Ocarbonate is useful for determining the sources of carbon and oxygen

that formed the carbonates as well as processes involved in that mineral precipitation, and several reports give useful comparisons of carbonates found in serpentinizing seeps from Oman, Costa Rica, Italy, and the western United States (e.g., Szponar et al., 2013; Mervine et al., 2014; Sanchez-Murillo et al., 2014; Falk et al., 2016). The δ <sup>13</sup>Ccarbonate and δ <sup>18</sup>Ocarbonate in carbonates from serpentinizing seeps range from ∼ −33 to ∼ +3<sup>h</sup> <sup>δ</sup> <sup>13</sup>C and ∼ −20 to ∼ <sup>+</sup>5<sup>h</sup> <sup>δ</sup> <sup>18</sup>O. This wide range of isotopic compositions makes it clear that the formation and history of carbonates in serpentinizing systems can follow a varied path, which can sometimes be clarified using "clumped isotope" techniques (e.g., Eiler, 2007).

Among the studied examples of terrestrial serpentinizing seeps, few data are from tropical biomes (Beccaluva et al., 1999; Sanchez-Murillo et al., 2014; Cardace et al., 2015; Woycheese et al., 2015). Tropical surface biomes that deliver significant meteorological precipitation and cover field locations in dense foliage may complicate the isolation of the deep subsurface signal from surficial components. Frequent or heavy precipitation may facilitate the incorporation of exogenous nutrients into the surface exposed seeps and springs, and impact the nutrient availability to the subsurface biosphere. Conversely, addition of meteoric water to the serpentinizing fluids associated with the seeps and springs may locally dilute endogenous nutrient or energy resources, changing the fluid-rock interactions happening in the subsurface. Either scenario could cause fluctuations in sources of energy and carbon between seasons, ultimately affecting the ability to identify true subsurface-driven processes such as hydrogen based metabolisms. While this may incur difficulty in separating a "true" subsurface signal in serpentinizing systems hosted in ophiolites located in tropical biomes, it also provides an opportunity to explore a largely unrecognized exchange between the surface and subsurface biosphere.


Data for samples during the dry (2012) and wet (2013) seasons have been previously reported (Cardace et al., 2015; Table 2), and are repeated here for seasonal comparison. Carbon isotope ratios expressed as compared to VPDB. nd, not determined; bdl, below detection limit. <sup>∗</sup>Data for rivers and well are provided as points of reference to the local water table. The rivers are immediately proximal to the sample locations, in both cases. The well "DH-4" was drilled ∼200 m from the NWD pool.

Our interest lies in investigating the effect of seasonal meteorological precipitation on the deep subsurface signature from spring locations in the Zambales and Palawan ophiolites in the Philippines. Geologic and geochemical descriptions of both ophiolite exposures and springs have been reported previously (Abrajano et al., 1989; Abrajano and Sturchio, 1990; Cardace et al., 2015; Ilao et al., 2018). Here, we look at carbon isotopic signatures in fluids and solid materials (sediments, biomass, and carbonates), and nitrogen isotopic signatures of biomass, over three precipitation-defined periods. Both the Zambales and Palawan ophiolites are located in Monsoon climatic regimes, with defined wet and dry seasons. In addition, we categorize our sample locations based on the flow rate of subsurface fluids emanating from the springs. The driving questions in this work concern whether increased seasonal precipitation will increase exogenous nutrient input to the surface expressions [springs], or conversely, dilute the available metabolic resources derived from subsurface processes. We hypothesized that seasonal precipitation would differentially impact systems with low flow vs. high flow of subsurface fluids, which is supported by the results given here.

#### MATERIALS AND METHODS

#### Description of Field Locations

Both field locations are within the monsoon climate zone of the Philippines. Locations were visited in October 2012 during the beginning of the dry season (193 mm/month average precipitation), September 2013 at the end of the wet season (346 mm/month average precipitation), and January 2017 when the least precipitation is received in our field areas (<20 mm/month average precipitation). Several sites in the Zambales and Palawan ophiolites were sampled, although not all sites were sampled in all three seasons. **Table 1** notes the season each location was sampled, and **Tables 1**, **2** note the depth below fluid surface and distance down the outflow channel, where applicable. In general, both fluids and solids were sampled when they were both accessible/removable. Solids included loose sediments, obvious biofilms, or carbonate features (not all were available at every sample location). **Table 2** also provides details on sample name and the type of sample collected. For example, at 10m down the outflow at site ML2, three different solid materials were collected in 2012; "gray sediment," "carbonate mound," and "rimstone" (**Table 2**). Samples with identical names and notations between multiple sample years were taken from the same location, as exactly as possible, based on photographic records of previous sampling efforts. We were unable to sample precipitation during the time that we were in the field.

Images of the sampling areas are provided in **Figure 1** with more detailed images and descriptions given in **Supplementary Figure S1** for reference. Discharge was estimated by catching runoff into a 500 ml wide mouthed bottle over a 30 s interval, repeated in triplicate and averaged. Where relevant, discharge was measured at multiple points. Sample sites are regarded as being high flow (>2 L/min), with discrete pools and runoff channels, or low flow (<2 L/min) with individual pools or a series of pools and associated carbonate terraces and similar features, but run off channels limited or not present. All locations featuring actively flowing run off channels were estimated to have a discharge rate of >2 L/min. In addition, low flow areas are also designated as "capped" or "uncapped," where "capped" refers to the presence of a carbonate film on the top of the pool. In such cases, the bottom of the pool is not visible, although the caps may be transient and potentially broken by meteorological precipitation or animal (including human) interaction. Examples of capped and uncapped pools are shown in **Figure 1**. High flow systems are expected to have an abundance of input from subsurface fluid and gas, and increasing interaction with surface conditions and atmospheric exchange as the fluids progress down the outflow channel. Low flow systems are expected to have slower fluid and gas input from the subsurface, and the degree of interaction with surface conditions depends on the presence of absence of a carbonate cap. We expect that pools with carbonate caps will have more limited atmospheric exchange, as the cap functions as a physical barrier at the surface of the pool.

#### Concentration and Isotopic Analysis of Dissolved Inorganic Carbon (DIC) and Dissolved Organic Carbon (DOC) in Spring Fluids

Amber I-CHEM vials and septa were pre-washed and pretreated as previously described (Cardace et al., 2015). A glass media bottle was used to collect sample from the springs, after being triple rinsed with sample. Once collected, sample was filtered through a Millipore Sterivex GV 0.22 µm filter unit for sterilizing aqueous solutions (Cat. No. SVGVL10RC/Lot No. 1515/00631) into the sample bottle. Some samples required pumping for efficient collection, and a Geotech Environmental Geopump peristaltic pump was used to fill the glass collection bottle. Viton Masterflex tubing (Cole Parmer, Vernon Hills, IL, United States) was used for pumping when collecting DIC samples, at a slow pumping rate of <100 ml/min and as short a pumping distance as possible to minimize gas loss. DOC bottles were filled only after the filter was conditioned with several liters of sample. Both DIC and DOC bottles were filled to the top with sample to exclude air bubbles, and were stored at 4◦C.

Samples were analyzed by the University of California, Davis, Stable Isotope Laboratory. Carbon isotopic ratios of carbon in DOC was analyzed with a O.I. Analytical Model 1030 TOC analyzer (Xylem Analytics, College Station, TX, United States), interfaced to a PDZ Europa 20–20 isotope ratio mass spectrometer (Sercon Ltd., Cheshire, United Kingdom) with a GD-100 Gas Trap Interface (Graden Instruments). Several replicates of reference materials were interspersed with samples, including IAEA-600, USGS-40, USGS-41, and Elemental Microanalysis reference materials.

Dissolved inorganic carbon samples were analyzed on a GasBench II system interfaced to a Delta V Plus IRMS (Thermo Fisher Scientific, Bremen, Germany). The fluids were added to


(Continued)

**87**


1969; Friedman and O'Neil, 1977).

∗Estimated

 using δ13C DIC or δ13C DOC from a proximal and directly related location.

(a) Map of the Philippines with sample regions. Marine Geoscience Data System (MGDS; www.marine-geo.org). (b) Cistern pool ML1, Manleluag area. Images of location ML2 and associated outflow sites can be found in Supplementary Figure S1. (c) Source pool of "Governor's Spring." Small white scale bar noted. White arrow at top right indicates flow direction. (d) North West Dugout Pool. White arrow indicates location of spring in (c). (e) Pinaduguan Falls. Inset shows close up of "Pig Pool" (PF1). (f) Poon Bato "PB2" location, with major features noted. Car battery for scale. (g) Poon Bato "PB1" location. Note calcite "cap" on top of the low flow pool (arrow). (h) Close up image of "Star Pool" in (f). Note lack of calcite "cap" (bottom of pool is visible).

a sealed, He-purged exetainer, and acidified to liberate all of the DIC as CO2. Reference materials for DIC analyses were lithium carbonate (Acros-1, Acros-2 Li2CO3, lots measuring δ <sup>13</sup>C −13.4 and −3.85, respectively, Thermo Fisher Scientific, St. Louis, MO, United States) dissolved in degassed deionized water and a deep seawater (both calibrated against NIST 8545). Final δ <sup>13</sup>C values are expressed relative to the international standard VPDB (Vienna PeeDee Belemnite).

#### Determination of Endmember Composition and Contributions

An estimation of subsurface endmember δ <sup>13</sup>CDIC was calculated following the method of Miller and Tans (2003), where δ <sup>13</sup>CDIC ∗ [DIC] is plotted as a function of [DIC], for each field area, and the slope of the resulting linear regression indicates an estimated δ <sup>13</sup>CDIC for the endmember fluid at that field site. These endmember values of δ <sup>13</sup>CDIC were then used to estimate the fraction of the subsurface endmember remaining following select microbial metabolic processes on the measured δ <sup>13</sup>CDIC pool. Here, a Rayleigh distillation model was used;

$$
\delta^{13} \text{C}\_{\text{observed}} = \delta^{13} \text{C}\_{\text{source}} + 10^3 (\alpha - 1) \ln(\text{f})
$$

Where δ <sup>13</sup>Cobserved is the measured δ <sup>13</sup>CDIC in the samples, and δ <sup>13</sup>Csource is either the estimated δ <sup>13</sup>CDIC−subsurface, or δ <sup>13</sup>CDOC, as specified below. Fractionation factors, α, were chosen for individual microbial processes. For chemoautotrophic

pathways, namely the acetyl Co-A and rTCA cycles, a range of α = 0.964–0.996 as reported in Hayes, 2001 was used. For bacterial photoautotrophy, a range of α = 0.978–0.99 as reported in Hayes, 2001 was used. For hydrogenotrophic methanogenesis, α = 0.945 from carbonate to methane (Waldron et al., 1998) and α = 0.942 from CO<sup>2</sup> to methane (Krzycki et al., 1987) were used. For each of the above processes utilizing the DIC pool as a carbon source, the δ <sup>13</sup>Csource used was the estimated subsurface endmember δ <sup>13</sup>CDIC as calculated for each location as in Miller and Tans, 2003. Processes that produce CO2, such as acetoclastic methanogenesis (α = 0.976, Waldron et al., 1998) and methanol methanogenesis (α = 0.932, Rosenfeld and Silvermann, 1959; Silverman and Oyama, 1968) were also considered. For these latter processes, the δ <sup>13</sup>Csource used was the measured δ <sup>13</sup>CDOC.

#### Carbon and Nitrogen Isotopic Ratios in Solid Materials

Solid samples, consisting of sediments, carbonate terrace materials, and any resident microbiota or biofilms, were collected using sterile technique into Whirlpac bags, and kept frozen at −20◦C until analysis. Samples were freeze dried, and then ground to a fine powder in glass or agate mortars. Mortars and Pestles were baked at 550◦C overnight prior to use. Samples were analyzed in the Osburn Isotope Geobiology Laboratory at Northwestern University.

Carbonate content, assayed as mg CO<sup>2</sup> and its respective δ <sup>13</sup>C value (hereafter δ <sup>13</sup>CCO3) was determined simultaneously via continuous flow on a Thermo Fisher Scientific GasBench II, coupled to a Delta V Plus isotope ratio mass-spectrometer (CF-IRMS) at the Northwestern Stable Isotope Facility. Prior determination of the CO<sup>3</sup> <sup>2</sup><sup>−</sup> content from gravimetric quantification guided the sample amount used for analysis, aiming for ∼10 µmol CO2. Samples were weighed into 12 mL Exetainer <sup>R</sup> vials, which were subsequently septumsealed and purged with UHP He for 7 min. Approximately 200 µL of 103% H3PO<sup>4</sup> was injected into each vial, and the samples placed into a thermos-stated block at 70◦C to allow CO<sup>2</sup> to evolve overnight. The isotopic composition, δ <sup>13</sup>CCO3 is corrected using the periodic sampling of CO<sup>2</sup> from the H3PO4-acidified CaCO<sup>3</sup> standards NBS18 (δ <sup>13</sup>CVPDB <sup>=</sup> <sup>−</sup>5.014h) and NBS19 (<sup>δ</sup> <sup>13</sup>CVPDB = 1.95h), and samples reported on the VPDB scale. Estimated precision (1 s.d.) on δ <sup>13</sup>CCO3 is <sup>±</sup> 0.06h.

For the determination of wt% organic C and organic N, acidified and decarbonated samples were weighed into tin capsules and combusted online in a Costech 4010 Elemental Analyzer, coupled to a Thermo Fisher Scientific Delta V Plus mass-spectrometer via a ConFloIV. Briefly, ∼10– 30 mg of powdered decarbonated samples were weighed, then combusted online in a column containing chromium (III) oxide and silvered cobaltous chloride, held at 980◦C. Product gasses were carried over hot Cu reduction column held at 705◦C to removed excess O<sup>2</sup> and convert nitrogen oxides to N2. Product CO<sup>2</sup> and N<sup>2</sup> were separated by a molecular sieve 5A GC column. The gasses were analyzed via CF-irms, and size corrected. Tank corrections were done by regular calibration against organic standards supplied by Indiana Biogeochemical laboratories (IU acetanilide and IU urea), and placed on the δ <sup>13</sup>CVPDB and δ <sup>15</sup>NAIR scales respectively.

#### Carbonate Mineralogy

An Olympus Terra X-ray diffractometer<sup>1</sup> , with the specifications equivalent to the CheMin tool developed for Mars exploration as described in Blake et al. (2012), was used for X-ray diffraction (XRD) analysis. The Terra engages a Co X-ray source and a cooled charge-coupled device (CCD) detector arranged in transmission geometry with the sample, with angular range of 5◦ to 50◦ 2θ with <0.35◦ 2θ resolution (Blake et al., 2012). X-ray tube voltage is typically 30 kV, with a power of 10 W, a step size of 0.05◦ , and an exposure time of 10 s per step. A minimum of 250 exposures were recorded prior to diffractogram interpretation.

Dry samples were powdered using an agate mortar and pestle, cleaned with isopropyl alcohol between samples. Powder was passed through a standard 150 µm sieve (100-mesh) prior to analysis. Powdered, sieved material was transferred with a spatula to the input hopper of the vibration chamber sample cell, and shaken into the space between two mylar window, to be agitated during analysis, presenting all planes with the mineral sample to the x-ray beam.

The resulting diffractogram was interpreted using XPowder software<sup>2</sup> , which is a commercially available peak search-andmatch program that queries the PDF2 database for reference mineral peak information (see text footnote 2). XPowder allows for identification of major minerals; trace minerals can be missed or masked by peaks of other minerals. Diffractograms convey mineral fingerprint information customarily by plotting of diffracted signal intensity on the y-axis against ◦ 2θ on the x-axis. An intensity peak is the result of constructive interference when Bragg's law (nλ = 2d sin θ, where n is the "order" of reflection, λ is the incident X-rays wavelength, d is spacing between atomic planes in a crystal structure, and θ is the incidence angle) is fulfilled by the incoming x-rays. For reference, for data collected using a Co x-ray tube, the three most prominent d-values for minerals of interest are as follows: serpentine (var. lizardite) Mg3Si2O5(OH)4, D1: 7.12 Å, D2: 2.379 Å, D3: 3.56 Å; serpentine (var. antigorite) (Mg,Fe++)3Si2O5(OH)4, D1: 7.29 Å, D2: 2.53 Å, D3: 3.61 Å; brucite Mg(OH)2, D1: 2.365 Å, D2: 4.77 Å, D3: 1.794 Å; hydrotalcite Mg6Al2(CO3)(OH)16·4(H2O), D1: 7.69 Å, D2: 3.88 Å, D3: 2.58 Å; portlandite Ca(OH)2, D1: 2.628 Å, D2: 4.9 Å, D3: 1.927 Å; calcite CaCO3, D1: 3.035 Å, D2: 2.285 Å, D3: 2.095 Å; magnesite MgCO3, D1: 2.742 Å, D2: 2.102 Å, D3: 1.700 Å; artinite Mg2(CO3)(OH)2·3(H2O), D1: 2.736 Å, D2: 5.34 Å, D3: 3.69 Å; chlorite (var clinochlore) (Mg,Fe++)5Al(Si3Al)O10(OH)8, D1: 7.16 Å, D2; 4.77 Å, D3: 3.58 Å; and smectite (var beidellite) Na0.5Al2(Si3.5Al0.5)O10(OH)2·n(H2O), D1: 2.55 Å, D2: 2.61 Å, D3: 4.52 Å.

In order to interpret co-occurring minerals in association with spring water, an Eh-pH diagram was constructed in Geochemist's Workbench Act 2. The system was modeled at a temperature

<sup>1</sup>https://www.olympus-ims.com/en/xrf-xrd/mobile-benchtop-xrd/terra/#! <sup>2</sup>http://www.xpowder.com/

of 25◦C, at a pressure of 1.013 bars, with log activity HCO<sup>3</sup> − set at −2.699, log activity Ca2<sup>+</sup> set at 2, log activity Fe2<sup>+</sup> set at −3, and unit activity of Mg2<sup>+</sup> and water. The log activity HCO<sup>3</sup> <sup>−</sup>is based on high CO<sup>2</sup> in to water mixture as low DIC spring water encounters high DIC surface water. The log activity Ca2<sup>+</sup> set at 2 represents generally observed molalities of [Ca2+] near 100 m, which correspond to activities of ∼100, thus log 100 = 2. The log activity Fe2<sup>+</sup> set at −3 represents generally observed molalities of [Fe2+] near 1 mmolal, which correspond to activities of ∼0.001, thus log 0.001 = −3. Unit activity of Mg2<sup>+</sup> conveys the Ca dominance of the aqueous system, about two orders of magnitude greater than Mg, thus Mg activity taken as one. Unit activity of water is appropriate for most lower salinity, low temperature waters (activity coefficients not impacted by high levels of solutes).

#### RESULTS

#### Dissolved Carbon (DIC and DOC)

Results of the analysis of DIC and DOC can be found in **Table 1**. **Figures 2**, **3** display the carbon isotopic ratios and concentrations of DIC and DOC, separated by flow regime (high flow in **Figure 1**, low flow in **Figure 2**) and season. Data are grouped in **Figure 2** by samples that are in or near the source pools, vs. those that are part of the extended outflow channel. Calculated estimates of potential subsurface endmember δ <sup>13</sup>CDIC are shown in **Supplementary Figure S3**, and are ∼ −21.8<sup>h</sup> for ML, ∼ −22.7<sup>h</sup> for GS, ∼ −12.8<sup>h</sup> for NWD, ∼ −13.26<sup>h</sup> for PB, and ∼ −22.5<sup>h</sup> for PF locations. The ratio of DOC:DIC at each major field area is considered in **Supplementary Figure S4**. We are lacking δ <sup>13</sup>CDIC from precipitation during the time periods that we were in the field, so estimates of input from this source of DIC were not considered.

#### DIC in High Flow Systems (ML, GS, MF)

General trends in DIC (**Figure 2**) in high flow systems include low concentrations of DIC (e.g., <1 ppmC) in source pools, continuing downstream to end with an often higher concentration signal (1–10 ppmC). The δ <sup>13</sup>CDIC at the source pools was more enriched relative to downstream locations at all sites and varied between <sup>−</sup>11 to ∼ −18<sup>h</sup> at ML, but <sup>∼</sup> <sup>−</sup>14 to <sup>−</sup>22<sup>h</sup> in high flow Palawan ophiolite locations. At the high flow system ML2, a sample was taken 1.5 m beyond the mouth of the source pool in 2012 and 2017, and the δ <sup>13</sup>CDIC became briefly more enriched than the source pool across that distance. Discrete source pools associated within a larger spring system [e.g., ML1-ML2 or GS-NWD] had highly variable δ <sup>13</sup>CDIC, with as much as a 7–8<sup>h</sup> difference in δ <sup>13</sup>C between them (**Figure 2**). Precipitation (meteorological) did not influence DIC concentration, but may influence the δ <sup>13</sup>CDIC in the high flow ML systems (the only high flow systems where data from multiple seasons are available). The δ <sup>13</sup>CDIC at ML1 was more <sup>13</sup>C enriched in drier seasons relative to the wet season, while the δ <sup>13</sup>CDIC at ML2 became more <sup>13</sup>C depleted in drier seasons relative to the wet season. The highest concentration of DIC in a high flow system was found in MF (dry season). It was concluded previously that this site is influenced in part by non-serpentinizing hydrothermal fluids (Cardace et al., 2015).

Subsurface endmember δ <sup>13</sup>CDIC for sites at ML and GS are depleted by <sup>∼</sup>13<sup>h</sup> relative to atmospheric DIC. A Rayleigh distillation model (**Supplementary Table S3**) predicts a wide range of potential δ <sup>13</sup>CDIC from the subsurface endmember for both areas (∼6–91%) after fractionation by microbial processes that consume DIC, after DIC production and fractionation by acetoclastic and methanol methanogenesis (54–93%). In contrast, the predicted remaining subsurface endmember at NWD is 0% for DIC consuming microbial processes, but 28–86% for acetoclastic and methanol methanogenesis.

#### DIC in Low Flow Systems (PB, PF)

Two DIC datapoints are available for each of the two low flow, uncapped sites (**Figure 3**). The data range from <1 to ∼ 30 ppmC, and δ <sup>13</sup>CDIC ∼ −11 to <sup>−</sup>21h. This isotopic signature is comparable to that of the high flow systems, although DIC was more abundant in these lower flow, uncapped pools.

While limited data availability makes it difficult to identify broad patterns, there was a notable difference in the δ <sup>13</sup>CDIC between the wet (∼ −13h) and dry season (∼ −25h) in the low flow, capped systems (**Figure 3**). This was the most negative δ <sup>13</sup>CDIC found. Concentrations of DIC for both low flow, capped samples was 1–10 ppmC.

The subsurface endmember sources of δ <sup>13</sup>CDIC were predicted to be <sup>−</sup>13.26<sup>h</sup> to the PB area, and <sup>−</sup>22.5<sup>h</sup> to the PF area. These calculated endmembers are considered with caution, as they are based on very few data.

#### DOC in High Flow Systems (ML, GS, MF)

DOC concentrations in the high flow systems were highly variable and range from 0 to 2.7 ppmC (**Figure 2**). Carbon isotopic signatures were typically <sup>&</sup>lt;−25<sup>h</sup> although several samples taken in the very dry season had more positive values (see ML and GS, **Figure 2**). Notable outliers to these general trends included a groundwater well associated with site NWD with 7.8 ppmC, and the highly <sup>13</sup>C depleted DOC (−50.8h, relative to the DOC at all other locations) found 10.3 m down the outflow of GS. There was little apparent effect of meteorological precipitation on the concentration of DOC in the high flow systems, however, lack of rain may be linked to the most positive δ <sup>13</sup>CDOC values found (−19 to <sup>−</sup>22h) which occurred in the very dry season in both ML and GS systems.

#### DOC in Low Flow Systems (PB, PF)

The DOC in the fluids of the low flow, uncapped systems (**Figure 3**) had a more narrow isotopic range than that in the high flow systems. However, the DOC was <sup>13</sup>C enriched relative to that of many sites (−21 to <sup>−</sup>26h), resembling the range for the very dry season samples at high flow ML and GS. No direct seasonal comparison could be made for these sites, due to lack of data from multiple seasons for identical locations. However, the spring systems sampled in the very dry season (PF) had an order of magnitude less DOC than the spring system sampled in the dry season (PB).

are also separated by seasonal sampling, and sample names correspond to the names in the sample location pictures (Figure 1 and Supplementary Figure S1) and Table 1. Dashed lines separate discrete samples within a season. Data can also be found separated by discrete samples in Supplementary Figure S2.

Concentrations of DOC from the low flow, capped systems were between 0.1 and 0.3 ppmC, and the δ <sup>13</sup>C was −23 to <sup>−</sup>27<sup>h</sup> (**Figure 3**), more <sup>13</sup>C depleted than the uncapped, low flow counterparts. In the dry season, DOC was more <sup>13</sup>C enriched than in the wet season, possibly representing the "leftover" pool in the reservoir, which was not refreshed by new material brought

#### Geochemistry of Solids/Mineralogy Organic Carbon (Biomass) in High Flow Systems (ML, GS, NWD, MF)

The carbon isotopic signature of organic carbon in solids (biomass) from high flow systems was in the range of ∼ <sup>−</sup>21 to <sup>−</sup>28h, across all seasons and locations (**Figure 2** and **Table 2**). The samples from high flow systems also had the largest range in fractionation between biomass, DIC, and DOC (**Figure 4**) of all the systems investigated. Biomass in the high flow systems was slightly <sup>13</sup>C depleted relative to DOC (**Table 2** and **Figures 2**, **4**). There were exceptions, usually from samples near the end of the runoff channel which broaden the range of fractionation from DOC. The biomass samples in these systems were the most <sup>13</sup>C enriched relative to DOC of all the systems considered (**Figure 4**). In general, the abundance of biomass increased downstream. In addition, biomass was far less abundant in the very dry season, and more abundant in the wet season (**Figure 2**). The GS system outflow channel presented an atypical case compared to other high flow systems. While the relationship between the biomass and the DOC in the source pool at GS was similar to the other high flow systems, the biomass was as much as 10<sup>h</sup> <sup>13</sup>C enriched relative to other sites. In addition, the δ <sup>13</sup>Cbiomass and δ <sup>13</sup>CDOC in the outflow channel of GS were dissimilar, casting doubt that DOC was the source of the carbon for the biomass in the outflow sites.

in by meteorological precipitation.

FIGURE 4 | Ranges of measured δ <sup>13</sup>Corganic in each of the three types of sample systems (vs. VPDB), and calculated 113C of organic carbon in solids relative to DIC and DOC. Gray bars note the δ <sup>13</sup>Corganic range, and number of analyses per sample type. The sample types are also broken into seasons, key at right. The brown arrow indicates a single outlier data point of <sup>∼</sup> + 24h.

#### Organic Carbon (Biomass) in Low Flow Systems (PB, PF)

The δ <sup>13</sup>Cbiomass with the most negative values were found in the low flow, uncapped systems. All samples were <sup>&</sup>gt;4<sup>h</sup> <sup>13</sup><sup>C</sup> depleted relative to the DOC, and there was very little seasonal variability (**Figures 3**, **4**). These samples also had the most depleted δ <sup>13</sup>Cbiomass relative to DIC of all the systems sampled, with samples from the very dry season as much as ∼ −18<sup>h</sup> depleted relative to DIC (**Figure 4**). Biomass from the low flow, capped systems had a similar δ <sup>13</sup>Cbiomass as that from the high flow systems (between ∼ −24 to <sup>−</sup>29h), with the more negative δ <sup>13</sup>Cbiomass in samples from the dry season.

#### Carbonates

In the pH range of the spring locations, speciation of DIC is expected to vary as a function of pH; at pH ∼10.3 we can expect about equal proportions of bicarbonate and carbonate components of DIC, below pH = 10.3 bicarbonate will dominate over carbonate, and above pH = 10.3 carbonate is the dominant ion. Fractionation between the two species occurs, but is expected to be small across the temperature and pH ranges studied here. Fractionation as carbonate minerals precipitate will also vary with precipitation rate. Fractionations due to pH and temperature differences between samples/seasons are expected to be minimal, on the order of <sup>&</sup>lt;2h when considering fractionation between

FIGURE 5 | δ <sup>13</sup>C (vs. VPDB), and δ <sup>18</sup>O (vs. VPDB) of carbonates, given in <sup>h</sup>. Black symbols = dry 2012 season, gray symbols = wet 2013 season, and red symbols = very dry 2017 season. Data are found in Table 2. Fields indicate formation and fractionation processes discussed in Clark et al. (1992) and Falk et al. (2016). Dashed boxes indicate the position of fields outlining formation and fractionation processes as discussed in Clark et al. (1992) and Falk et al. (2016) for Oman serpentinization-associated carbonates. Shaded boxes show adjusted positions accounting for the up to <sup>∼</sup>5<sup>h</sup> difference in <sup>δ</sup> <sup>18</sup>O of seasonal and annual rainfall for our study areas. Isotopic data for precipitation were obtained from the Online Isotopes in Precipitation Calculator and are found in Supplementary Table S2 (Bowen and Revenaugh, 2003; Bowen et al., 2005; Bowen, 2019). (a,b) These fields represent the expected isotopic ranges for carbonates formed in equilibrium with soil CO2, and atmospheric CO2, respectively (purposefully no data in the "b" fields). (c) This field is interpreted by Falk et al. (2016) as containing carbonates that result from mixing of endmember fluids, or recrystallization and/or isotopic exchange in older carbonates. (d) Fossil travertine crusts from Oman (Clark et al., 1992), interpreted as shifting toward equilibrium values during secondary recrystallization of more depleted travertine. (e) Field containing modern crusts from Oman (Clark et al., 1992), blue triangles, provided for reference. (f) Carbonates potentially formed by the CO<sup>2</sup> hydroxylation process.

H2O and CO<sup>2</sup> due to pH (Halas et al., 1997, **Table 4**), and <1 to <sup>∼</sup>2<sup>h</sup> between bicarbonate and carbonate in our temperature range (Emrich et al., 1970; **Table 1**). The fractionation due to precipitation rates is also expected to be <sup>&</sup>lt;1h (Turner, 1982). There are no reports that consider all three of these variables on fractionation of DIC or carbonate minerals in the range of conditions of our study sites.

The δ <sup>13</sup>C of solid inorganic carbon (carbonates) were typically only slightly fractionated relative to DIC, showing the most similarity in the wet season and the most fractionation from DIC in the dry and very dry seasons (**Table 2** and **Figures 2**, **3**). Again, there were exceptions to this pattern. For example, the carbonates in wet season samples from the PB location (2013), which was low flow and capped, were nearly 15<sup>h</sup> <sup>13</sup>C depleted relative to the DIC (**Figure 3**). In these samples, δ <sup>13</sup>Ccarbonate was more isotopically similar to the δ <sup>13</sup>Cbiomass and δ <sup>13</sup>CDOC. A similar landscape of carbonates <sup>13</sup>C depleted relative to DIC was found at the GS site, both near the source and farther down the outflow (**Figure 2**).

FIGURE 6 | Eh-pH diagram illustrating likely carbonate mineralogy of GS area spring waters, which are Ca-dominated waters with near surface DIC values due to interaction with atmosphere, and bear low concentrations of dissolved Mg and very low concentrations of dissolved Fe. The system was modeled at a temperature of 25◦C, at a pressure of 1.013 bars, with log activity HCO<sup>3</sup> − set at –2.699, log activity Ca2<sup>+</sup> set at 2, log activity Fe2<sup>+</sup> set at –3, and unit activity of Mg2<sup>+</sup> and water. Note that artinite is a hydrated magnesium carbonate mineral [Mg2(CO3)(OH)2·3H2O], and calcite is taken to be pure CaCO3. For relevant high pH, low Eh environmental conditions (lower right plotted area), one expects mineral precipitation of artinite, grading to calcite as pH drops (possibly due to organic acid production by microbiology or influx of atmospheric CO2).

Carbonates from the Philippines had some of the most depleted δ <sup>13</sup>C and δ <sup>18</sup>O values reported from terrestrial serpentinizing environments (**Table 2** and **Figure 5**). Values of δ <sup>13</sup>C in our carbonates range from <sup>−</sup>28.1 to <sup>−</sup>9.3<sup>h</sup> VPDB, and values of δ <sup>18</sup>O range from <sup>−</sup>24.6 to <sup>−</sup>4.8<sup>h</sup> VPDB (5.5– 23.4h VSMOW). Equilibrium considerations in the formation of carbonates were calculated and are available in **Table 2**, including a starting δ <sup>13</sup>CCO<sup>2</sup> and δ <sup>18</sup>OCO2. The difference between the measured δ <sup>18</sup>Ocarbonate and the expected (calculated) δ <sup>18</sup>Ocarbonate is shown in **Figure 7**, for samples for which a δ <sup>18</sup>Owater was available. In all samples except one (from PB2), the measured δ <sup>18</sup>Ocarbonate was depleted in <sup>18</sup>O relative to the calculated, expected value for δ <sup>18</sup>Ocarbonate, with up to <sup>−</sup>17.39<sup>h</sup> difference. Carbonates depleted in <sup>18</sup>O relative to equilibrium with water indicate rapid mineral precipitation or precipitation far out of equilibrium.

Because the most negative δ <sup>13</sup>C carbonates were primarily from the GS area, the mineralogy of solid samples from those locations was further explored. The expected carbonate mineralogy is shown in the Eh/pH diagram in **Figure 6**; measured carbonate mineralogy is given in **Figure 8**. The model for carbonate stability for a Ca-dominated water with low Mg, low Fe, and near surface DIC values suggests that at the pH and Eh of GS locations, the expected stable magnesium and calcium carbonate minerals are artinite and calcite.

XRD analysis of GS location carbonates (**Figure 8**) showed that the outflow sites between 4.5 and 10.3 m are dominated by calcite, serpentinite, clay minerals such as smectite and chlorite, and other smaller proportions of carbonates such as aragonite, magnesite, and artinite, with possible portlandite [Ca(OH)2], suggested by variable right-side shoulder near the serpentine peak at ∼42.5 ◦ 2-theta. At the source pool alone, brucite [Mg(OH)2] and hydrotalcite [Mg6Al2CO3(OH)16·4H2O] are indicated in XRD results, giving evidence for mineral precipitation of hydroxide phases from this OH<sup>−</sup> dominated spring water where it emerges from the subsurface.

#### Nitrogen

The δ <sup>15</sup>N in solid samples (presumably from biomass) compared to the ratio of total carbon to total nitrogen (both as wt.%) are given in **Figure 9**, broken down by flow system type and season. A relationship between the abundance of nitrogen in biomass and a depleted <sup>15</sup>N isotopic signature relative to atmospheric may indicate diazotrophic activity, supplying freshly fixed nitrogen to the biomass (e.g., Loiacono et al., 2012 and sources within). In contrast, nitrogen limitation or an <sup>15</sup>N enriched nitrogen isotopic signature relative to atmospheric may indicate nutrient recycling, exogenous nitrogen addition from eukaryotic surface systems, or microbial nitrogen cycling functions such as nitrification or denitrification. A poorly fit relationship was present between the C:N and δ <sup>15</sup>N in the high flow systems during the seasons with more meteorological precipitation (**Figure 9A**), but not in the very dry season (**Figure 9B**). While there were fewer data points available for the lower flow systems (**Figure 9C**), data cluster according to the season the samples were obtained in. Very dry season samples had the most positive δ <sup>15</sup>N, with high nitrogen biomass (**Figure 9C**, blue field), dry season samples had the lowest wt% nitrogen in biomass, and δ <sup>15</sup>N data fall between <sup>±</sup> 2.3<sup>h</sup> (yellow field), while wet season samples were primarily δ <sup>15</sup><sup>N</sup> <sup>&</sup>lt; <sup>0</sup><sup>h</sup> and low C:N (red field).

FIGURE 8 | X-ray diffractograms convey mineralogical differences between representative samples of Governor's Spring solids, collected at spring source (green), two locations along the outflow path (red, purple), and at the base of the main flow channel (blue). Data were collected using an Olympus Terra XRD unit outfitted with Co tube (https://www.olympus-ims.com/en/xrf-xrd/mobile-benchtop-xrd/terra/#!), and peaks were identified using Xpowder (http://www.xpowder.com/). Peaks (intensity on the y-axis) indicate strongly diffracted x-rays. Peaks correspond to specific angles (◦2-theta, on x-axis) at which x-rays are diffracted by specific planes of atoms present in the mineral sample. At spring source (green diffractogram), serpentine peaks co-occur with strong brucite peaks, and associated hydrotalcite. Along outflow path (red and purple diffractograms), carbonate and serpentine minerals dominate, with possible minor brucite, portlandite, and magnesite. The lowest elevation site (blue diffractogram) shows carbonate minerals with a smectite group clay (broad peak, far left).

FIGURE 9 | Ranges of measured δ <sup>15</sup>N within organic fraction of solid samples, as a function of the C/N ratio in the samples, divided by 10. Data are found in Table 2. Three sampling seasons are shown; 2013 wet (gray), 2012 dry (black), and 2017 very dry (red) seasons. Circles indicate high flow systems, Triangles indicate low flow systems. The dashed line indicates the expected δ <sup>15</sup>N of atmospheric N2. Values below this line are interpreted as produced by fractionation during nitrogen fixation processes, and values above are likely influenced by other nitrogen cycling reactions and nitrogen recycling. (A) High flow systems, during the wet and dry seasons. (B) High flow systems during the very dry season. (C) Low flow systems, all seasons.

#### DISCUSSION

Surface derived carbon, such as DOC picked up from plant, animal, or soil sources during overland flow, is characteristically depleted in δ <sup>13</sup>C relative to atmospheric CO2. Examples of surface derived carbon can be found in **Table 2** – samples from PB1 ("Muddy pot, 2013") and PB2 (leaf litter and soil reference material, 2012), which range from δ <sup>13</sup>C = −14.7 to <sup>−</sup>24.6h. When such biomass is carried into the sample locations after being dissolved or transported as solids, it provides organic carbon with δ <sup>13</sup>C depleted with respect to atmospheric CO2. Microbial heterotrophy of organic carbon for biomass production results in very little fractionation of carbon (e.g., Hayes, 1993). In these systems, when the δ <sup>13</sup>Cbiomass is only barely fractionated with respect to δ <sup>13</sup>CDOC the assumption can be made that DOC was utilized to produce the biomass. Further, it is logical that the source of the DOC is likely primarily from surface environments, rather than produced within the pools, especially during periods of high precipitation. Care needs to be taken when interpreting DOC-biomass relationships in the low precipitation seasons. Measured δ <sup>13</sup>CDOC or δ <sup>13</sup>Cbiomass values with a large enrichment relative to source DOC are likely products of microbial carbon fixation, a result of recycling of carbon in a closed or semi-closed system, or a mix of heterotrophic and autotrophic growth.

The measured δ <sup>13</sup>CDIC can be the product of multiple processes, including fluid mixing, fractionation following biological activity, and kinetic effects to name a few. Removal of DIC from the source (whether subsurface or surface) by microbial carbon fixation or methanogenesis will change both the concentration of DIC and the δ <sup>13</sup>CDIC in the remaining DIC pool. Some autotrophic organisms are known to fractionate DIC by as much as 36h with specific carbon fixation pathways at 25–40◦C (Hayes, 2001; House et al., 2003). Likewise, the production of DIC by heterotrophic processes, or acetoclastic methanogenesis will supply DIC to the available pool that is fractionated relative to the source DOC.

Nitrogen fixation fractionates N<sup>2</sup> only slightly from atmospheric values, and δ <sup>15</sup>Nbiomass close to 0<sup>h</sup> can be inferred to be a product of nitrogen fixation (Delwiche and Steyn, 1970). Other nitrogen cycle processes produce more negative δ <sup>15</sup>Nbiomass, and values of δ <sup>15</sup>Nbiomas that are enriched relative to atmospheric N<sup>2</sup> can result from recycling fixed N<sup>2</sup> in a closed or partially closed system (e.g., Havig et al., 2011).

The climate of the Philippines affects the primary carbon sources for biomass in serpentinizing spring fluids in the Zambales and Palawan ophiolites. Our goal was to determine the primary carbon source for biomass in both the source pools (where a greater "subsurface" fingerprint might be presumed) and runoff channel locations. Our data indicate that factors that impact the primary carbon source include flow rate of the fluids, and degree of exogenous carbon input from meteorological precipitation-derived DIC and DOC.

Seasonal low meteorological precipitation affects the quantity of biomass present in the sediments at all field locations where multiple seasons were sampled (ML, PB1) – measured biomass was less abundant in the dry and very dry seasons (**Figures 2**, **3**). This decrease in biomass production can not be directly linked to a decrease in DOC concentrations within our sample set. With few exceptions (ML1, and some outflow channel sites of ML2), there is not an apparent decrease of DOC with decreased meteorological precipitation and the highest DOC:DIC ratios are found in very dry season samples (**Supplementary Figure S4**). However, as discussed below, in a few cases (such as ML during the very dry season) there is evidence of DOC recycling during the drier seasons relative to the wet season. A possible explanation is that biomass does increase in response to an increase in delivery of DOC via overland flow, and the DOC measured during the wet season is "leftover" DOC that has not been consumed. The dynamics of population and metabolic shifts that might be tied to DOC have not been studied previously in these systems.

#### High Flow Systems (ML, GS, NWD, MF)

Our results indicate that the primary source of carbon for microbial communities in high flow systems was variable with location and affected by season. Evidence for carbon limitation was found during the very dry season, and the best indication of mixotrophic communities (i.e., indication of both heterotrophic and autotrophic processes) was found at distance down outflow channels where δ <sup>13</sup>Cbiomass relative to DOC was the most positive (**Figure 4**). Both DIC and DOC were <1 ppmC in the source pools of most high flow systems (with the exception of MF and NWD). The DOC:DIC of high flow sites was <1 (with only three exceptions), and these represent the highest DOC:DIC ratios found in each sampling season (**Supplementary Figure S4**).

The carbon isotopic ratio data indicate the primary carbon source incorporated into biomass in the ML1 and ML2 systems was likely DOC. The δ <sup>13</sup>Cbiomass (**Figure 2**) in these two springs was only slightly depleted or slightly enriched with respect to DOC (**Figure 4**); in cases of δ <sup>13</sup>Cbiomass enrichment relative to DOC, a secondary carbon source of DIC from atmospheric influence or from microbial metabolic byproducts may be invoked. Sampling during the very dry season in the Manleluag area revealed a δ <sup>13</sup>CDOC more enriched than that in other seasons, suggesting carbon recycling in the outflow channel due to lower flow rates and less exogenous carbon delivered by meteorological precipitation, resulting in the residual DOC pool harboring more <sup>13</sup>C. Modeling of the degree of input from the subsurface δ <sup>13</sup>CDIC was inconclusive with a broad range of incorporation of subsurface source δ <sup>13</sup>CDIC possible (6–93% incorporation). Given that the measured δ <sup>13</sup>CDIC in the ML system is influenced by a wide range of processes we feel the approach used was insufficient to model these dynamics. Our available evidence indicates that carbonate production in the ML systems required a mix of carbon from DIC and atmospheric CO2. Carbonates analyzed from the source pools of ML1 and ML2 had enriched δ <sup>18</sup>O and δ <sup>13</sup>C compared to other samples (field "c," **Figure 5**). Carbonates in this range of δ <sup>18</sup>O and δ <sup>13</sup>C have been interpreted by others as carbonates formed by remineralization or resulting from mixing of endmember fluids. Farther down the outflow at ML2, carbonates are a mix of "fossil" carbonates and freshly formed carbonates with more depleted δ <sup>13</sup>C and δ <sup>18</sup>O than at the source pools (**Figure 5**, field "d").

The other major high flow system, the GS area and associated NWD, was only sampled in the very dry season and the source of carbon for biomass was variable by location. In the source pool of GS, both δ <sup>13</sup>CDOC and δ <sup>13</sup>Cbiomass were enriched compared to all other sample locations (**Figure 2**), and the DOC:DIC was the lowest observed (**Supplementary Figure S4**). The microbial community in the source pool of GS may build biomass from carbon fixation processes, thus producing organic acid byproducts that are <sup>13</sup>C enriched relative to the DOC or subsurface DIC endmember (**Supplementary Figure S3**), contributing to the measured δ <sup>13</sup>CDOC. Alternatively, the measured δ <sup>13</sup>CDOC may also be a consequence of the very dry season sample time, similar to that observed in the ML systems described above – a comparison with wet season δ <sup>13</sup>CDOC is not available for the GS system. As the fluid at GS flowed downstream, DIC and DOC did not become more abundant (in contrast to the outflow of ML2), and the δ <sup>13</sup>Cbiomass became more depleted relative to the source pool biomass. Rayleigh modeling of the potential input from a subsurface source of DIC was largely inconclusive for the GS system. DIC at NWD is only slightly enriched relative to DOC. Carbonates produced in and near the source pool at GS occur with brucite [Mg(OH)2], and are more <sup>13</sup>C

enriched than other samples (and <sup>∼</sup>10<sup>h</sup> <sup>13</sup>C enriched relative to the next site downstream), possibly representing mixing with endmember fluids or recrystallization of carbonates rather than freshly precipitated carbonates (**Figure 5**, fields "c, d"). However, **Figure 7** shows that the carbonates found in the GS system have the largest difference between the expected δ <sup>18</sup>O and the measured δ <sup>18</sup>O, indicating that rapid carbonate deposition is occurring at this location, even in the source pool. Biomass carbon at the bottom of the outflow channel of GS either incorporated carbon from the carbonates, or influenced the δ <sup>13</sup>Ccarbonates. These carbonates had the most negative <sup>13</sup>C found in our high flow systems, and there was only <sup>&</sup>lt;1–4h difference between <sup>δ</sup> <sup>13</sup>Cbiomass and δ <sup>13</sup>Ccarbonate. This could indicate that carbonate formed quickly in the outflow and microbial waste product DIC was a key source of carbon used to form the carbonate. Another possibility is that the microbial community utilized carbon from the carbonate to build biomass. There is precedent for this latter concept. In high pH serpentinizing systems found at The Cedars (United States), it has been shown that Serpentimonas isolates use CaCO<sup>3</sup> in carbon fixation (Suzuki et al., 2014). Future work will be needed to determine if this is a phenomenon is restricted to the low meteorological precipitation season at GS. Regardless, by 4.5 m down the outflow channel, rapid carbonate deposition has resulted in carbonates with some of the most depleted δ <sup>13</sup>C and δ <sup>18</sup>O relative to other carbonates reported from terrestrial serpentinizing systems.

Evidence for nitrogen fixation in the high flow systems (**Figures 9A,B**) was limited to the MF site (only sampled in the dry season), the source pool of ML1 in the very dry season, and the distant outflow points of ML2 (but excluding the 2012 dry season samples). These results indicate that the ML system communities have potential for nitrogen fixation under some environmental conditions, and more investigation is needed to determine which members of the community are capable of nitrogen fixation and what conditions enable the process. The MF location is slightly higher temperature than ML, with a likely hydrothermal mixing member (Cardace et al., 2015) and nitrogen fixation processes could be attributed to thermophilic members of the community (e.g., Hamilton et al., 2011; Loiacono et al., 2012).

#### Low Flow, Uncapped Systems (PB2, PB3, PF)

While few data are available for low flow systems in general, those data presented in **Figure 3** indicate that there may be a relationship between more seasonal meteorological precipitation and an increase in concentrations and volume of DIC, DOC, and biomass in the uncapped systems.

Low flow systems have some of the lowest DOC:DIC ratios of all the samples examined (**Supplementary Figure S4**), with little variability between seasons. While few data are available for low flow systems, the uncapped pools featured the most depleted δ <sup>13</sup>Cbiomass, relative to DOC, with the largest fractionation from DOC (up to 7h), regardless of season of sampling. To produce biomass depleted in <sup>13</sup>C relative to both DOC and DIC, microorganisms either have to use an unidentified, very δ <sup>13</sup>C negative source of carbon, or a large fractionation from DIC needs to occur. These results indicate that even when carbon from surface processes was available in the drier seasons, it is possible that the lower flow, uncapped systems received enough fluid and gas from the subsurface to support a microbial community that engaged in metabolic activities independent from exogenous carbon. During the dry season, these systems may depend on methane or other carbon-bearing gasses sourced from depth. Estimation of the subsurface endmember δ <sup>13</sup>CDIC does not produce results that support this hypothesis. However, it is possible that the estimation of the subsurface source δ <sup>13</sup>CDIC is inaccurate (**Supplementary Figure S3**) – only a few samples were available for each pool and the Miller-Tans analysis was performed with the PB area samples (capped and uncapped combined) considered as one "site," rather than separate locations that may have differing endmembers in reality. Alternatively, when exogenous carbon was less abundant in the very dry season, the low flow, uncapped systems may recycle carbon similarly as described above for the high flow systems – metagenomic/metatranscriptomic data could help to clarify the carbon flow for the very dry season.

Regardless of the season of sampling, the nitrogen cycle was dependent on surface-sourced nitrogen as no direct evidence from geochemistry points to active subsurface nitrogen fixation (**Figure 9**). The low flow, uncapped systems form carbonates slowly and are enriched in δ <sup>13</sup>C and δ <sup>18</sup>O relative to other carbonates sampled, suggesting that they have opportunity to undergo recrystallization as they shift toward equilibrium values (fields "c, d," **Figure 5**).

#### Low Flow, Capped Systems (PB1)

Seasonality affected the carbon isotopic ratio of DIC and DOC in the "capped" low flow pools, the abundance of biomass present, and possibly the source of carbon for the biomass. In the dry season, the pool fluid is separated from atmospheric and most surface influence by the physical barrier of the carbonate skin on the pool surface (**Figure 1g**). The low flow, capped systems appear to be forming carbonates at a rapid rate, both on the bottom of the pools and across the surface of the pool. Some of these data fall outside of identified fields in **Figure 5**, or near field "f," identified as potentially forming via a CO<sup>2</sup> hydroxylation process (Falk et al., 2016). Along with the high flow GS locations, these low flow capped carbonates are the most depleted in δ <sup>13</sup>C and δ <sup>18</sup>O of the samples, indicating fresh, fast formation. Under these conditions, new carbon and nutrients can only be obtained from slowly flowing gas and fluid from depth, or from solids already present in the pool. **Figure 3** shows how this impacted the DIC, which was more δ <sup>13</sup>C depleted than DOC. The DIC pool was likely influenced by metabolic byproducts from microbial metabolism depleted in <sup>13</sup>C relative to DIC from surface sources, which were trapped in the fluids under the carbonate cap. The residual DOC pool was also more <sup>13</sup>C enriched than in the wet season, indicating that heavier δ <sup>13</sup>CDOC was left behind in the DOC pool, non-replenished by surface DOC.

The carbon isotopic signature of the biomass was more <sup>13</sup>C depleted than both DIC and DOC, and could be influenced by incorporation of carbon-bearing gasses from depth, such as in the low flow, uncapped pools. We interpret the wet season δ <sup>13</sup>CDIC as incorporating atmospheric sources, indicating that the cap was at least periodically washed away by precipitation. Biomass in both the wet and dry season carried a nitrogen isotopic signature indicative of nitrogen fixation (**Figure 9C** and **Table 2**), in contrast to the uncapped low flow systems. It is unclear why the capacity for nitrogen fixation would be more prevalent in these low flow capped systems than the uncapped systems.

#### CONCLUSION

Our results allow us to broadly characterize the effect of climate and fluid flow on the carbon and nutrient sources of several serpentinization-driven ecosystems in the Zambales and Palawan ophiolites. Increased meteorological precipitation during wetter seasons neither significantly diluted nor added to the DOC and DIC concentrations in the source pools of high flow systems ML1 and ML2, or capped, low flow pool PB1 (the only sample locations where such a direct comparison is possible). Samples farther down outflow channels at ML did have higher concentrations of DOC in the wet season, suggesting that climate may have a larger impact on downstream systems than source pools of high flow systems. However, changes in meteorological precipitation did impact the carbon isotopic ratio of both DIC and DOC in ML and PB1 fluids, which reflected seasonal gain/loss of atmospheric influence on the δ <sup>13</sup>C of DIC, and changes in exogenous DOC input.

The primary carbon source in high flow systems was variable, with DOC contributing more to biomass in the ML system, and a mix of DIC and carbonates contributing to biomass in the GS system. Primary carbon resources in the low flow systems may depend more on endogenous than exogenous carbon. Partially, this may be due to smaller "footprints" of the lower flow systems, affecting the surface area available to receive exogenous materials either washed or dropped into the systems. Carbonate "caps" on the very lowest flow systems seasonally isolate the pools from both organic and inorganic exogenous carbon.

The search for a true "subsurface" signature in the Zambales and Palawan serpentinizing systems has concluded that the highest degree of subsurface influence is found in the low flow systems, and in select source pools of the high flow systems (namely GS). Rapid mineral precipitation of carbonate, and δ <sup>13</sup>Cbiomass that was depleted relative to δ <sup>13</sup>CDOC highlights potential true subsurface signals. Biomass can not be produced from surface derived δ <sup>13</sup>CDOC with a resulting depleted δ <sup>13</sup>C relative to that source – therefore a subsurface process can be assumed. The drier the climate, the more the subsurface carbon signature is apparent, making it more likely that processes such as hydrogen based metabolisms (methanogenesis or sulfate reduction, for example) are key in ecosystem functioning. In the very dry season, evidence for <sup>13</sup>CDOC pool enrichment relative to source DOC, without subsequent <sup>13</sup>Cbiomass enrichment indicates that more autotrophy and/or methane-driven (and hydrogen dependent) metabolic schemes were in action while DOC was limited.

Future sampling will focus on obtaining multi-season samples from all these locations, with endmember sampling, to further explore the validity of this conclusion. Higher flow systems at ML should be considered very carefully with respect to subsurface signatures, and the degree of surface impact on the geochemistry and microbiology. Previous work identified relict subsurface genetic capacity in the ML systems, and future work will focus on separating surface vs. subsurface function in microbial systems, with an eye to identifying populations that are actively using hydrogen-driven vs. Corganicdriven metabolic processes. High flow GS and lower flowing capped systems, including PB1 should be the focus of future subsurface biosphere investigations. Further consideration of carbon and nitrogen cycling potential will include insight from metagenomic datasets. These will clarify the presence/absence of the genetic capacity for the microbial communities found in these locations to participate in carbon and nitrogen cycling, identify metabolisms that use the abundant hydrogen present in these systems, and allow deeper interpretation of these geochemical data presented here.

#### AUTHOR CONTRIBUTIONS

All authors contributed to the data collection and interpretation in this manuscript.

#### FUNDING

This work was supported in part by NSF, grant numbers 1147334 and 114610 to DM-D and DC, respectively. Further funding was awarded to DM-D by the Illinois Space Grant Consortium in support of student research. MO was supported by NASA Exobiology grant #NNX15AM08G. Field support was funded by the Alfred P. Sloan Foundation, Deep Carbon Observatory-Deep Energy award number AWD05445 (UCLA Subaward Number 2090 G UA348) to DC.

#### ACKNOWLEDGMENTS

Dr. Andrew Masterson of the Northwestern Stable Isotope Facility contributed significantly to the isotope analyses presented here. DM-D would like to acknowledge several undergraduate students who helped with this work; Ester Yim, Michael Tanzillo, Alexandra Guzman Biehl, and Mary Chaudhry. This is EDGE lab contribution #6.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2019. 00761/full#supplementary-material

#### REFERENCES

fmicb-10-00761 April 29, 2019 Time: 17:13 # 18



**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 Meyer-Dombard, Osburn, Cardace and Arcilla. 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.

## H<sup>2</sup> Kinetic Isotope Fractionation Superimposed by Equilibrium Isotope Fractionation During Hydrogenase Activity of D. vulgaris Strain Miyazaki

#### Michaela Löffler, Steffen Kümmel, Carsten Vogt\* and Hans-Hermann Richnow

Department of Isotope Biogeochemistry, Helmholtz Centre for Environmental Research – UFZ, Leipzig, Germany

#### Edited by:

Eric Boyd, Montana State University, United States

#### Reviewed by:

Wolfgang Buckel, University of Marburg, Germany Christopher L. Hemme, University of Rhode Island, United States

#### \*Correspondence:

Carsten Vogt carsten.vogt@ufz.de

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 30 April 2019 Accepted: 20 June 2019 Published: 10 July 2019

#### Citation:

Löffler M, Kümmel S, Vogt C and Richnow H-H (2019) H<sup>2</sup> Kinetic Isotope Fractionation Superimposed by Equilibrium Isotope Fractionation During Hydrogenase Activity of D. vulgaris Strain Miyazaki. Front. Microbiol. 10:1545. doi: 10.3389/fmicb.2019.01545 We determined <sup>2</sup>H stable isotope fractionation at natural abundances associated with hydrogenase activity by whole cells of Desulfovibrio vulgaris strain Miyazaki F expressing a NiFe(Se) hydrogenase. Inhibition of sulfate reduction by molybdate inhibited the overall oxidation of hydrogen but still facilitated an equilibrium isotope exchange reaction with water. The theoretical equilibrium isotope exchange δ <sup>2</sup>H-values of the chemical exchange reaction were identical to the hydrogenase reaction, as confirmed using three isotopically different waters with δ <sup>2</sup>H-values of – 62, <sup>+</sup>461, and <sup>+</sup> <sup>1533</sup>h. Expected kinetic isotope fractionation of hydrogen oxidation by non-inhibited cells was also superimposed by an equilibrium isotope exchange. The isotope effects were solely catalyzed biotically as hydrogen isotope signatures did not change in control experiments without cells of D. vulgaris Miyazaki.

Keywords: hydrogenase, D. vulgaris strain Miyazaki, monitoring, GC-IRMS, equilibrium isotope fractionation, kinetic isotope fractionation, hydrogen isotopes

#### INTRODUCTION

Many microorganisms use hydrogen (H2) or protons (H+) as electron donors or acceptors, coupled to the oxidation or production of H2. The enzyme catalyzing H<sup>2</sup> oxidation or production is a metalloenzyme termed hydrogenase, for which several differently structured isoenzymes are known (Vignais and Billoud, 2007; Greening et al., 2016). The most abundant and commonly studied type of hydrogenase contains a NiFe(Se)-active center (Vignais and Billoud, 2007). The reaction catalyzed by hydrogenases can be formulated as follows (Eq. 1).

$$H\_2 \Leftrightarrow \ 2H^+ + 2e^- \tag{1}$$

Such a reaction usually leads to a kinetic isotope fractionation, which is defined as the ratio of the rate constants for light and heavy isotopes within the unidirectional reaction of H<sup>2</sup> to two protons releasing two electrons. The isotope fractionation is a result of the slightly lower activation energy needed to cleave and form bonds of lighter isotopes compared to heavy isotopes in this (bio-) chemical reaction. The kinetic isotope fractionation leads to the predominant reaction of light isotopomers, which implies that the remaining fraction would get heavier during the reaction. For the oxidation of molecular hydrogen at natural abundances, this would result in accumulation of

<sup>2</sup>H (deuterium) in the remaining fraction. However, during the oxidation of H<sup>2</sup> (Eq. 1), an isotopic exchange with water (Eq. 2) was observed simultaneously to the kinetic isotope effect (Arp and Burris, 1982; Vignais et al., 1997, 2002; Yang et al., 2012).

$$H^2H + H\_2O \Leftrightarrow H\_2 + H^2HO \tag{2}$$

During the isotope exchange of gaseous hydrogen with water, the heavy isotopes of molecular hydrogen (2H) exchange with the light hydrogen isotopes (1H) of the water, until an equilibrium isotope value for molecular hydrogen is reached. This is an inverse isotope effect compared to kinetic isotope fractionation, where the H<sup>2</sup> will become enriched in deuterium over time and thus, the δ <sup>2</sup>H-value will become more positive. If kinetic and equilibrium isotope fractionation take place in parallel, the oxidation and the isotope exchange reaction cannot be separated from each other. In order to understand overall isotope effects during H2-consumption, the equilibrium isotope fractionation of the isotopic exchange reaction must be studied separately. Therefore we designed experiments to study both H2-oxidation and isotope exchange with cell suspensions of Desulfovibrio vulgaris strain Miyazaki F, which expresses one of the best studied NiFe hydrogenases, as well as a NiFe(Se) suited for H<sup>2</sup> oxidation (Yagi et al., 1976; Deckers et al., 1990; Ogata et al., 2002; Foerster et al., 2003; Fichtner et al., 2006; Pandelia et al., 2010; Nonaka et al., 2013; Riethausen et al., 2013). We hypothesized that the addition of molybdate will inhibit electron flow to sulfate (Peck, 1959; Wolin and Miller, 1980) and thus, only the isotopic exchange reaction would be observable.

The aim of our study was to analyze the equilibrium isotope effect of the isotope exchange reaction and the kinetic isotope effect of the unidirectional oxidation reaction of hydrogen, in order to eventually monitor hydrogenase activity in environmental samples and settings, e.g., during hydrogen underground storage. This method, based on natural abundant stable hydrogen isotopes of gaseous samples, would allow in situ assessment of hydrogenase activity without further treatment of samples.

#### MATERIALS AND METHODS

#### Chemicals

All chemicals until otherwise stated were purchased from Merck Chemicals GmbH (Darmstadt, Germany). Deuterium-enriched waters were prepared by mixing 1 l sterilized tap water water (Merck Millipore, Germany) with either 250 µl or 100 µl <sup>2</sup>H2O (99.9%; Merck Chemicals, Germany).

#### Culture and Cultivation Conditions

Desulfovibrio vulgaris strain Miyazaki F (DSM 19637) was obtained from the DSMZ (Deutsche Sammlung von Mikroorganismen und Zellkulturen, Braunschweig, Germany). The strain was grown in a mineral medium for sulfate-reducers, which consisted of NH4Cl (0.3 g/l), KH2PO<sup>4</sup> (0.4 g/l), CaCl<sup>2</sup> (0.075 g/l), Na2SO<sup>4</sup> (2 g/l), MgSO<sup>4</sup> ∗ 7 H2O (1 g/l), 1 ml trace element solution SL-10, 0.1 ml selenite-tungstate solution, 4 ml 50 % Na-DL-lactate, 2 ml vitamin solution, 10 ml 1 M NaHCO<sup>3</sup> solution, 2 ml 1 M Na-acetate solution. 1 M L-cysteine solution was used for reduction. 0.1 mg/l resazurin was used as redox indicator. The selenite-tungstate solution contained per 100 ml: 0.5 g/l NaOH, 3 mg/l Na2SeO<sup>3</sup> ∗ 5 H2O and 4 mg/l Na2WO<sup>4</sup> ∗ 2 H2O. The vitamin solution contained per 100 ml: 1 mg biotin, 1 mg folic acid, 25 mg pyridoxine-HCl, 25 mg thiamine-HCl ∗ 2 H2O, 5 mg riboflavin, 25 mg nicotinic acid, 25 mg D-Capantothenate, 0.5 mg vitamin B12, 25 mg p-aminobenzoic acid and 25 mg lipoic acid. All components, except the last four, were mixed in sterilized tap water (Merck Millipore, Germany) and purged with 75% N<sup>2</sup> and 25% CO<sup>2</sup> until they became virtually oxygen-free and were autoclaved subsequently. The remaining components were added within an anaerobic glovebox (Toepffer Lab Systems, Germany).

Experiments with cell suspensions were performed with cells pre-grown on 28 mM lactate and 22 mM sulfate at 30◦C and 120 rpm on a Multitron incubation shaker (Infors, Germany). At optical densities above 0.2 absorbance with no further increase in cell density, 50 ml each were transferred into 120 ml serum bottles in the glovebox, which was filled with a mixture of N<sup>2</sup> (97–98%) and H<sup>2</sup> (2–3%) and crimped close with PTFE-coated chlorobutyl-isoprene septa (Thermo Fisher Scientific, Germany) before the start of the experiment.

Three sets of isotope fractionation experiments were designed with these cells: one set in which cells were inhibited by molybdate, one in which the headspace had been purged to reduce sulfide load, and one without further treatment. For the inhibited setups, six of the bottles each were treated with 20 mM molybdate. Therefore, 2 ml of a 0.5 M sodium molybdate solution was added to each bottle, yielding yellow-orange molybdosulfide-complexes (Wolin and Miller, 1980; Biswas et al., 2009). Both the purged and untreated setups consisted of five bottles of cell suspension each. The headspace in the purged setups was exchanged with N<sup>2</sup> to eliminate H2S-burden in the headspace and reduce potential inhibition by sulfides.

In all samples 12 ml H<sup>2</sup> were hereafter exchanged with the existing headspace with a syringe and an empty needle while bottles were tilted to remove headspace only. Bottles were kept at 30◦C and 120 rpm (Multitron incubation shaker, Infors, Germany). For abiotic controls cell suspensions were inactivated by incubation at 80◦C in a water bath for 20 min and subsequent addition of 0.3 ml 10 M NaOH, yielding pH 12.

All experimental setups were performed in two replicates using water at natural abundance (δ <sup>2</sup>H-H2O = <sup>−</sup>62h) and water enriched in <sup>2</sup>H by adding 100 µl <sup>2</sup>H2O resulting in an isotope composition of δ <sup>2</sup>H-H2O = <sup>+</sup>461<sup>h</sup> One experiment with molybdate for inhibition of sulfate reduction was conducted with enriched water (250 µl <sup>2</sup>H2O) to yield a final isotope composition of δ <sup>2</sup>H-H2O = <sup>+</sup> <sup>1533</sup>h. Bottles were sacrificed at different time points in a water bath at 80◦C for 20 min and stored upside down after the addition of 10 M NaOH before hydrogen isotopes of H<sup>2</sup> in the headspace were analyzed. The setup with enriched water (δ <sup>2</sup>H-H2O = <sup>+</sup> <sup>1533</sup>h) was sacrificed in different periods of time than the other experiments, in order to gain insight into the initial reaction. NaOH was added to terminate microbial activity and to remove CO<sup>2</sup> from the headspace by precipitation of sodium bicarbonate. CO2-removal was necessary for subsequent isotope analysis, as the column used for gas chromatography (GC) would retain it. 100 to 300 µl sample volume were transferred via syringe into 1 ml 3% (w/v) Zn-acetate and stored at −20◦C for downstream processing. Sulfides in the headspace of the bottles were eliminated due to precipitation as ZnS by the addition of 3 ml 3 (w/v) % Zn-acetate.

### Analytical Methods

fmicb-10-01545 July 9, 2019 Time: 15:27 # 3

#### Sulfide Measurements

Sulfides were measured as previously described (Cline, 1969; Kleinsteuber et al., 2008) against a blank control. Previously frozen samples were thawed and mixed with 4 ml water and 0.4 ml of Cline's solution. The mixtures were stored in the dark for 20 min before photometric measurement on an UV-1800 (Shimadzu, Germany) at 670 nm. Due to the high concentrations of sulfides, samples were diluted before measurement. Optical densities of cell suspensions were determined on an UV-1800 (Shimadzu, Germany) at 600 nm. 1 ml culture solution was transferred via syringe into a cuvette filled with a few mg of Nadithionite, closed with parafilm and shaken until dissolved and then measured using water as control.

#### Isotope Measurement

Hydrogen isotope measurements were performed on a GCisotope ratio mass spectrometer (IRMS) system (**Supplementary Figure 1**) (7890A, Agilent Technologies, Germany; GC-IsoLink II Thermo Fisher Scientific, Germany; Conflo IV Thermo Fisher Scientific, Germany; Finnigan MAT 253, Thermo Fisher Scientific, Germany) equipped with a J&W CP-Molsieve 5Å GC Column (50 m, 0.32 mm, 30 µm, Agilent Technologies, Germany). An empty ceramic tube was kept isothermal at 500◦C inside the GC-IsoLink and the GC column was kept at 40◦C during measurements. An additional cold trap operated with liquid nitrogen was installed after the GC to remove water to keep the water background within the ion source of the IRMS low. After 2 h of runtime, the column was heated to 250◦C and held for at least 10 min remove residual water vapor and residual CO<sup>2</sup> from the GC column.

0.2 ml headspace injections were made manually at split-ratio of 1:25 with a gas-tight syringe in an interval of roughly eight minutes needed for elution of H<sup>2</sup> and permanent gases. The analytical system was evaluated for reproducibility and isotope artifacts. It was found to be reproducible and deliver true values within the uncertainty of 0.7 <sup>±</sup> 0.4h.

Both the normal and the δ <sup>2</sup>H-enriched waters were measured with an elemental analysis-chromium/high temperature conversion (EA-Cr/HTC)-IRMS system (HEKAtech, Germany) coupled via the Conflo IV (Thermo Fisher Scientific, Germany) to the same IRMS instrument (Gehre et al., 2017). All results are reported in the delta notation (Eq. 3) and according to the guidelines for stable isotope measurements (Coplen, 2011).

$$8\left[\%\right] = \left(\frac{\text{R}\_{\text{sample}}}{\text{R}\_{\text{standard}}} - 1\right) \ast 1000\tag{3}$$

The ratio of heavy to light isotopes (Rsample) in a compound is reported in δ-notation; for hydrogen, samples will be compared to Vienna Standard Mean Ocean Water (VSMOW) as standard with an isotope ratio of 155.76 ± 0.1 ppm (Rstandard). A laboratory standard, made of 10% H<sup>2</sup> in N2, with a δ <sup>2</sup>H-value of <sup>−</sup>205<sup>h</sup> was used for reference. The auto-protonation factor (H<sup>3</sup> <sup>+</sup> factor) was determined daily and remained stable at 8.24 ± 0.05. Theoretical equilibrium isotope values, mainly dependent on temperature and the isotope signature of water, were calculated according to a formula previously described (Horibe and Craig, 1995).

#### Assessment of H<sup>2</sup> Concentrations

The concentration of H<sup>2</sup> was measured with the GC-IRMS. For concentration measurements a defined volume of 10% H<sup>2</sup> was prepared using 120 ml serum bottles which were purged with N<sup>2</sup> before. Twelve ml H<sup>2</sup> were exchanged with the gaseous phase by a syringe and an empty needle while the bottle was held upside down. This sample was prepared daily and used for external calibration of the concentration measurement. Therefore the sample was injected three times at the start of each run, and all following samples were normalized to the average area under the curve of all controls in this measurement period. Intensities of external calibration were stable over 20 days, leading to overall variations in H2-concentration of ± 1.5%. Initial concentrations of H<sup>2</sup> between setups and replicates varied about 5% of the response, probably dependent on handling speed and temperature in the laboratory during exchange of H2, as well as gas-tightness of the syringe used for injections.

#### RESULTS

#### Measurement of Hydrogen Stable Isotopes

The isotope signatures of the different waters were determined to be δ <sup>2</sup>H-H2O = <sup>−</sup><sup>62</sup> <sup>±</sup> <sup>2</sup>h, <sup>+</sup><sup>461</sup> <sup>±</sup> <sup>1</sup>h, and <sup>+</sup><sup>1533</sup> <sup>±</sup> <sup>2</sup>h, respectively. For the sterilized tap water in Leipzig (δ <sup>2</sup>H-<sup>H</sup>2O = <sup>−</sup>62h), an equilibrium <sup>δ</sup> <sup>2</sup>H-value for H<sup>2</sup> of – 744<sup>h</sup> was calculated. The enriched water setups (δ <sup>2</sup>H-H2O = + <sup>461</sup>h and <sup>δ</sup> <sup>2</sup>H-H2O = <sup>+</sup> <sup>1533</sup>h) were calculated to equilibrate with a theoretical value for H<sup>2</sup> of δ <sup>2</sup>H = <sup>−</sup>606<sup>h</sup> and δ <sup>2</sup>H = <sup>−</sup>317h, respectively.

Control experiments with H<sup>2</sup> in the headspace and sterilized tap water or culture medium showed no significant change in the isotopic signature over the experimental timeframe of 18 days (**Figure 1**), with stable δ <sup>2</sup>H-values of δ <sup>2</sup>H = <sup>−</sup>141.2<sup>h</sup> and δ <sup>2</sup>H = <sup>−</sup>142.2h. These were identical to the isotope value of the pure H<sup>2</sup> used (δ <sup>2</sup>H = <sup>−</sup>139h), taking usual uncertainties into account. The δ <sup>2</sup>H signature of the hydrogen gas used in the anaerobic glove box was substantially lighter (approx. δ <sup>2</sup>H = <sup>−</sup>645h) than the signature of the H<sup>2</sup> gas used in the whole cell experiments, leading to a shift of <sup>1</sup>2H = 65<sup>h</sup> (**Figure 1**) when mixed.

#### H2-Oxidation by Non-inhibited Cells Experiments With Reduced Sulfide Concentration

To reduce concentrations of sulfide which may affect the metabolism of D. vulgaris, the headspace of culture bottles were

purged with N2. The sulfide concentrations after purging were typically roughly 6 mM.

13.9 ± 0.7% and 12.2 ± 0.3% H<sup>2</sup> were oxidized within 4 days (**Table 1** and **Figure 2D**). Sulfide concentrations increased from 6.0 ± 2.2 mM and 6.4 ± 0.1 mM to 17.9 mM and 16.7 ± 0.2 mM, respectively, in 3 days (**Figure 2A**). Hydrogen isotope values changed slowly toward depletion in the beginning from δ <sup>2</sup>H = <sup>−</sup>309.0 <sup>±</sup> 0.7<sup>h</sup> and <sup>δ</sup> <sup>2</sup>H = <sup>−</sup>163.2 <sup>±</sup> 0.9h, before the reaction gained speed (**Table 2** and **Figure 3**), yielding δ <sup>2</sup>H = <sup>−</sup>678.6 <sup>±</sup> 1.5<sup>h</sup> and <sup>δ</sup> <sup>2</sup>H = <sup>−</sup>475.3 <sup>±</sup> 4.1<sup>h</sup> after 3 days.

#### Experiments With High Sulfide Concentration

In the experiments without further treatment to reduce the amount of sulfides present, 20.4 ± 1.0% H<sup>2</sup> and 19.5 ± 0.4% H2, respectively were consumed within 2 days in normal water (**Table 1** and **Figure 2E**). 17.8 mM and 16.5 mM sulfides were produced starting from 9.48 ± 1.7 mM and 11.2 ± 0.6 mM until day three (**Figure 2B**). The δ <sup>2</sup>H of hydrogen in the bottles' headspace was rapidly decreasing from δ <sup>2</sup>H = <sup>−</sup>220.4 <sup>±</sup> 0.7<sup>h</sup> and <sup>δ</sup> <sup>2</sup>H = <sup>−</sup>268.4 <sup>±</sup> 0.5<sup>h</sup> to δ <sup>2</sup>H = <sup>−</sup>702.1 <sup>±</sup> 5.7<sup>h</sup> and <sup>δ</sup> <sup>2</sup>H = <sup>−</sup>408.1 <sup>±</sup> 2.8<sup>h</sup> after 1.5 to 2 days (**Table 2** and **Figure 3**). Concentrations at the last measureable time-points were 1.9 ± 0.1% and 1.8 ± 0.1% H2.

#### H<sup>2</sup> Isotope Exchange by Cells Inhibited With Molybdate

Molybdate was used to inhibit the electron flow to the electron acceptor sulfate. In molydate-amended cultures, the H<sup>2</sup> concentrations were almost constant with 14.3 ± 1.3%, 13.9 ± 1.7% and 15.8 ± 3.3% H2. Thus, no consumption of H<sup>2</sup> in the molybdate setups was observed (**Table 1** and **Figure 2F**). Furthermore, the sulfide concentrations did not increase and were stable between 2.5 to 3.5 mM (**Figure 2C**), indicating that sulfate reduction to sulfide was completely inhibited. In all experimental setups inhibited by molybdate, hydrogen isotopes in the headspace were depleting in deuterium starting from δ <sup>2</sup>H = <sup>−</sup>246.5 <sup>±</sup> 0.7h, <sup>δ</sup> <sup>2</sup>H = <sup>−</sup>223.9 <sup>±</sup> 0.5h, and δ <sup>2</sup>H = <sup>−</sup>162.4 <sup>±</sup> 1.7<sup>h</sup> (**Table 2**) and stabilized at δ <sup>2</sup>H = <sup>−</sup>735.8 <sup>±</sup> 0.4h, <sup>δ</sup> <sup>2</sup>H = <sup>−</sup>599.4 <sup>±</sup> 0.1h, and δ <sup>2</sup>H = <sup>−</sup>327.3 <sup>±</sup> 0.3<sup>h</sup> for the differently enriched waters after 6 days (**Figure 3**).

#### DISCUSSION

No changes in the isotope signature of H<sup>2</sup> for both culture medium and water in the abiotic controls could be observed, even though H2-concentrations decreased with continuous sampling of the same bottles (**Supplementary Figure 2**). Most studies on hydrogen isotope exchange with water used platinum or palladium as a catalyst and subsequent equilibration times of a few hours were reported at, e.g., 20◦C (Crist and Dalin, 1934; Farkas and Farkas, 1934; Horiuti and Polanyi, 1934; Farkas, 1936; Suess, 1949; Rolston et al., 1976). It is therefore reasonable to assume that isotope exchange without a catalyst is too slow to be assessable in the experimental timeframe used in this study. Subsequently, it is unlikely that the isotope signal in environmental samples could be significantly affected by minerals, as catalysts are needed for accelerating the exchange reaction. But sampling itself might lead to bias, as it has been shown that use of steel can lead to the generation of molecular H<sup>2</sup> from water, with which it is equilibrated (Chapelle et al., 1997).

Addition of gaseous hydrogen to water or culture medium leads to an immediate shift in the isotope signature from δ <sup>2</sup>H = <sup>−</sup>139.0 <sup>±</sup> 0.8<sup>h</sup> to <sup>δ</sup> <sup>2</sup>H = <sup>−</sup>141.2 <sup>±</sup> 0.3<sup>h</sup> for water and δ <sup>2</sup>H = <sup>−</sup>142.2 <sup>±</sup> 0.7<sup>h</sup> for culture medium. This effect is probably due to a relatively higher solubility of <sup>2</sup>H in water, which leads to an isotope fractionation (Muccitelli and Wen, 1978). The observed shift in isotopic signature lies within usual standard deviations for hydrogen isotope measurements, and no further change in the isotope signal was observed after the initial shift (**Figure 1**). Albeit negligible compared to biocatalysis, isotope exchange takes place in abiotic controls within 18 days at 30◦C in the absence of a catalyst.

The addition of a cell suspension leads to changes in the isotope signature due to hydrogenase activity. An inverse isotope effect was observed in all three experimental setups. This is consistent with previously reported δ <sup>2</sup>H -values for H2 production from water for Shewanella oneidensis MR-1, which can express a NiFe- or a FeFe-hydrogenase (Kreuzer et al., 2014). Upon inhibition of dissimilatory sulfate reduction by molybdate,


TABLE 1 | Concentrationof H2in all experimental setups: molybdate-inhibited, headspace purged with N2to reduce sulfides and without additional treatment. fmicb-10-01545 July 9, 2019 Time: 15:27 # 5

setups without further treatment, (C,F): inhibition by molybdate) in either water with <sup>δ</sup> <sup>2</sup>H2O = <sup>−</sup>62<sup>h</sup> (gray) or with <sup>δ</sup> <sup>2</sup>H2O = <sup>+</sup>461<sup>h</sup> (black). The uncertainty of 2<sup>σ</sup> is shown.


observable isotope effects should be limited to isotope exchange, since H<sup>2</sup> was not consumed. Hydrogenases were previously described to facilitate isotope exchange of <sup>2</sup>H<sup>2</sup> and H2O or H<sup>2</sup> and <sup>2</sup>H2O (Hoberman and Rittenberg, 1943; Jouanneau et al., 1980; Arp and Burris, 1982; Vignais et al., 2000, 2002). We therefore assume that the hydrogenase in our experiments is solely responsible for the isotope exchange.

After four to 6 days the isotope exchange reaction approximately reached equilibrium with differences of <sup>1</sup>2H = 9<sup>h</sup> and <sup>1</sup>2H = 6<sup>h</sup> and <sup>1</sup>2H = 10<sup>h</sup> compared to the theoretical values (**Figure 4**) in cultures inhibited by molybdate addition. Deviations from the theoretical values could be due to the higher measurement error and standard deviations of the water measurements used for calculation, as well as the high sensitivity of the equilibrium equation toward fluctuations in temperature.

It has been shown for the NiFe-hydrogenase from D. vulgaris Miyazaki that hydrogen oxidation is a two-step process, during which an enzyme-hydride-state forms (Lubitz et al., 2014). First, H<sup>2</sup> diffuses into the active center. Then, it is heterolytically cleaved, forming a proton and an enzyme-hydridecomplex. Afterward, electrons and protons are shuffled out of the protein structure (Lubitz et al., 2014). A study using D. vulgaris Hildenborough suggested that H<sup>2</sup> can be caged by the protein structure surrounding the active site when selenocysteine replaces cysteine in the active center of the protein structure (Gutiérrez-Sanz et al., 2013). In this case, substrate and products can accumulate within the protein structure of the NiFeSe-hydrogenase near the catalytic center and the two transfers of protons proceed faster than in NiFehydrogenases. The authors suggested that the accumulation and subsequent availability of substrate would correspond to a fast isotope exchange reaction (Gutiérrez-Sanz et al., 2013). This hypothesis connects protein structure and isotope effects. D. vulgaris Miyazaki expresses a structurally similar NiFeSehydrogenase best suited for hydrogen oxidation (Nonaka et al., 2013; Riethausen et al., 2013), and the corresponding fast isotope exchange reaction was measured in this study, where the isotope equilibrium was reached within 6 days. It might be possible to compare the equilibrium isotope effect of structurally different types of hydrogenases, such as NiFe- and FeFehydrogenases, in order to characterize the isotope exchange rate in future studies in more detail. For this, a different inhibition of electron flow might be needed, as molybdosulfide-complexes might not work for all microorganisms, as it specifically inhibits sulfate reduction. Even though it has been shown to also inhibit H<sup>2</sup> production from glucose (Wolin and Miller, 1980), it has not been further or sufficiently studied. Information on structurally different hydrogenases is crucial for using stable hydrogen isotopes as a monitoring tool to track in situ hydrogenase activity, e.g., during storage of hydrogen in underground reservoirs.

Even though H<sup>2</sup> was consumed in the experiment with purged headspaces, only minimal changes in the δ <sup>2</sup>H-values were observed in the beginning. During the substrate consumption, the hydrogen bond is cleaved, which is expected to result in a normal kinetic isotope effect. However, in our experiments,

fmicb-10-01545 July 9, 2019 Time: 15:27 # 7

FIGURE 3 | Isotope signal over time for three experimental setups in either water with δ <sup>2</sup> <sup>H</sup>2O = –62<sup>h</sup> (A) or with <sup>δ</sup> <sup>2</sup> <sup>H</sup>2O = +461<sup>h</sup> (B). The isotope equilibrium is reached after about 6 days and indicated by light gray bars. The uncertainty of 2σ is shown and often smaller than the symbol.

the isotope values approximate isotope equilibrium values with increasing time. We therefore suggest that the kinetic isotope effect of the hydrogen bond cleavage is superimposed by the equilibrium isotope exchange reaction. Only in the beginning of hydrogen oxidation, an effect of the kinetic isotope effect can be observed. Here, the equilibrium isotope effect has not yet completely superimposed the isotope signature, which results in seemingly stable isotope values.

The observed kinetic isotope fractionation effects could be due to shuffling of protons into the cell. Hydrogenase and cytochrome complexes are able to translocate protons (Ide et al., 1999; Dolla et al., 2000; Chang et al., 2004). For example, the reduction of sulfate needs two additional protons (Eq. 4).

$$4H\_2 + SO\_4^{2-} + 2H^+ \rightarrow H\_2S + 4H\_2O \tag{4}$$

During this process a slight isotope fractionation is expected, due to different diffusivity according to their molecular mass and tunneling effects in the hydrogenase structure (Cukier, 2004). Then, more <sup>2</sup>H<sup>+</sup> than <sup>1</sup>H<sup>+</sup> would be released from the hydrogenase, which would equivalent the expected kinetic isotope fractionation. This effect would counteract an equilibrium isotope exchange reaction. Not only the rate of equilibrium isotope exchange, but also the kinetic isotope fractionation could be affected by the protein structure. Therefore, further studies on both kinetic isotope effects and equilibrium isotope effects and their superimposition using structurally different hydrogenases are needed in order to use this concept as a monitoring or diagnostic tool.

Interestingly, cultures containing the highest sulfide concentrations tested in this study (10 mM) showed consumption of H2, but lacked the "isotopic lag phase" of the purged experiments with 6 mM starting concentration (**Figure 3**). This might be an indication that the amount of sulfides could affect the electron and proton flow, as H<sup>2</sup> is still consumed but the normal kinetic isotope fractionation of the H-H bond cleavage is immediately superimposed by the equilibrium isotope effect of the exchange reaction. During growth on lactate and sulfate, up to 52% of electrons flow into the production of H<sup>2</sup> and the remaining 48 % of electrons are coupled to sulfate reduction in D. vulgaris, yielding the potential to reduce approx. 8.9 mM sulfate from lactate and H<sup>2</sup> (Noguera et al., 1998) or 14 mM sulfate from lactate alone before the start of the hydrogen oxidation experiments. And with concentrations of 9.48 up to 11.2 mM sulfide in the setups without further treatment to reduce sulfides at the start of the experiment, inhibition is a consequential hypothesis.

#### CONCLUSION

The hydrogenase of Desulfovibrio vulgaris Miyazaki facilitates an equilibrium isotope exchange when consumption of H<sup>2</sup> is inhibited. Resulting δ <sup>2</sup>H-values of H<sup>2</sup> corresponds to theoretical thermodynamic isotope equilibria. During H2, the kinetic isotope

#### REFERENCES


fractionation, which should be observed due to bond-cleavage, is superimposed by the equilibrium isotope exchange. These results might differ for other microorganisms and structurally different hydrogenases. Equilibrium isotope exchange in the experiments with starting concentration of about 10 mM sulfides also indicates a possibility that sulfides could inhibit electron flow. This research is fundamental in nature and aims to build a better understanding of the isotope effects and processes associated with hydrogenases. The results of this study serve as a basis for future research on a simple monitoring tool for environmental, gaseous samples based on stable isotopes of hydrogen.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

#### AUTHOR CONTRIBUTIONS

ML planned and performed the experiments and data analyses and wrote the manuscript. SK helped to establish a GC-IRMS method for H2. CV and H-HR supervised the research and edited the manuscript.

#### FUNDING

This work was supported by the Federal Ministry for Economic Affairs and Energy (BMWi) within the funding initiative "Energiespeicher," project ANGUS II, Grant Number 03ET6122B.

#### ACKNOWLEDGMENTS

We would like to acknowledge Florian Tschernikl for help in the cultivation and Matthias Gehre for helpful inputs during method development and work in the isotope lab.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.01545/full#supplementary-material

amino acid radical initiation and propagation, and bond making/breaking reactions of water and oxygen. Biochim. Biophys. Acta Bioenerget. 1655, 13–28. doi: 10.1016/j.bbabio.2003.08.010



**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 Löffler, Kümmel, Vogt and Richnow. 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 Mathematical Model for the Hydrogenotrophic Metabolism of Sulphate-Reducing Bacteria

Nick W. Smith1,2,3,4, Paul R. Shorten1,3 \*, Eric Altermann2,3, Nicole C. Roy2,3,5 and Warren C. McNabb3,5

<sup>1</sup> AgResearch, Ruakura Research Centre, Hamilton, New Zealand, <sup>2</sup> AgResearch, Grasslands Research Centre, Palmerston North, New Zealand, <sup>3</sup> Riddet Institute, Massey University, Palmerston North, New Zealand, <sup>4</sup> School of Food and Advanced Technology, Massey University, Palmerston North, New Zealand, <sup>5</sup> High-Value Nutrition National Science Challenge, The University of Auckland, Auckland, New Zealand

#### Edited by:

Chris Greening, Monash University, Australia

#### Reviewed by:

Xiyang Dong, University of Calgary, Canada Rafael Muñoz-Tamayo, Institut National de la Recherche Agronomique (INRA), France

\*Correspondence: Paul R. Shorten Paul.shorten@agresearch.co.nz

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 30 April 2019 Accepted: 03 July 2019 Published: 17 July 2019

#### Citation:

Smith NW, Shorten PR, Altermann E, Roy NC and McNabb WC (2019) A Mathematical Model for the Hydrogenotrophic Metabolism of Sulphate-Reducing Bacteria. Front. Microbiol. 10:1652. doi: 10.3389/fmicb.2019.01652 Sulphate-reducing bacteria (SRB) are studied across a range of scientific fields due to their characteristic ability to metabolise sulphate and produce hydrogen sulphide, which can lead to significant consequences for human activities. Importantly, they are members of the human gastrointestinal microbial population, contributing to the metabolism of dietary and host secreted molecules found in this environment. The role of the microbiota in host digestion is well studied, but the full role of SRB in this process has not been established. Moreover, from a human health perspective, SRB have been implicated in a number of functional gastrointestinal disorders such as Irritable Bowel Syndrome and the development of colorectal cancer. To assist with the study of SRB, we present a mathematical model for the growth and metabolism of the well-studied SRB, Desulfovibrio vulgaris in a closed system. Previous attempts to model SRB have resulted in complex or highly specific models that are not easily adapted to the study of SRB in different environments, such as the gastrointestinal tract. We propose a simpler, Monod-based model that allows for easy alteration of both key parameter values and the governing equations to enable model adaptation. To prevent any incorrect assumptions about the nature of SRB metabolic pathways, we structure the model to consider only the concentrations of initial and final metabolites in a pathway, which circumvents the current uncertainty around hydrogen cycling by SRB. We parameterise our model using experiments with varied initial substrate conditions, obtaining parameter values that compare well with experimental estimates in the literature. We then validate our model against four independent experiments involving D. vulgaris with further variations to substrate availability. Further use of the model will be possible in a number of settings, notably as part of larger models studying the metabolic interactions between SRB and other hydrogenotrophic microbes in the human gastrointestinal tract and how this relates to functional disorders.

Keywords: hydrogen, gastrointestinal tract, IBS, hydrogen sulphide, mathematical modelling

### INTRODUCTION

fmicb-10-01652 July 17, 2019 Time: 14:37 # 2

Sulphate-reducing bacteria (SRB) play an important role in a variety of ecosystems, from marine sediments and oil fields to the human gastrointestinal tract (Muyzer and Stams, 2008; Carbonero et al., 2012a). The functional group of SRB has been reported to comprise 60 genera (Barton and Fauque, 2009), and is characterised by the ability to utilise sulphate as an electron acceptor during metabolism. The presence of these bacteria has both positive and negative implications on human activities, depending on the context. Much research has been performed on hydrogen sulphide (H2S) production in oil fields by SRB, which can lead to reduced oil quality and machinery corrosion (Magot et al., 2000), and in the treatment of industrial wastewater, as the sulphides SRB produce facilitate the removal of contaminating heavy metals (Kiran et al., 2017). Less clear are the implications of SRB in the human gastrointestinal tract (GIT). The SRB population size in the GIT has been measured at approximately 10<sup>7</sup> cells per gram of faeces (Doré et al., 1995), but varies between individuals (Nava et al., 2012) and between studies (Smith et al., 2018). These bacteria are widely studied due to their controversial role in a number of functional GIT disorders. Increased levels of colonic SRB and increased H2S concentrations have been linked to Irritable Bowel Syndrome, Inflammatory Bowel Disease and colorectal cancer [for a review, see Carbonero et al. (2012b)]. However, beneficial effects of H2S have also been investigated, such as its capacity to stimulate mucus production (Motta et al., 2015) and the potential influence of this molecule on blood pressure regulation (Tomasova et al., 2016). The important connexions between SRB, H2S and the host justify further research into the metabolism of these bacteria.

Another key molecule in SRB metabolism is elemental hydrogen. Alongside methanogens and reductive acetogens, SRB can metabolise free hydrogen present in the GIT, utilising it in the reduction of sulphate (Smith et al., 2018). The sulphate metabolised by SRB can be dietary or host-derived; cross-feeding by SRB on sulphate released during mucin metabolism by other GIT microbes has been well studied (Willis et al., 1996; Rey et al., 2013). High concentrations of hydrogen in the GIT are known to inhibit the metabolism of saccharolytic members of the microbiota (Wolin and Miller, 1983), therefore the presence of hydrogen cross-feeders is thought to increase the rate of carbohydrate breakdown by the wider microbial population. This has been shown in rodent models and linked to increased energy yield for the host (Samuel and Gordon, 2006; Rey et al., 2010).

Due to the importance of SRB in human health and nutrition, a greater understanding of their metabolism and growth dynamics is sought. To this end, we developed a mathematical model for the metabolite flux and population growth of the human SRB Desulfovibrio vulgaris, grown on substrates found in the GIT (Scanlan et al., 2009). Ours is not the first attempt to model SRB metabolism and growth and we compare the predictions of our model with that of the existing mathematical model of Noguera et al. (1998). Many other mathematical models of SRB have been published, but these are almost universally applied to address specific characteristics of SRB or for the investigation of competitive and syntrophic relationships between SRB and methanogens [for example, Robinson and Tiedje (1984), Okabe et al. (1995), Stolyar et al. (2007)]. The model of Noguera et al. (1998) is not targeted to a specific characteristic or environment, therefore is a good benchmark against which to compare our model. The existing model is more complex than that proposed here: it consists of ten ordinary differential equations for aqueous and gaseous metabolite concentrations and microbial growth and is dependent on 20 parameter values that are estimated either from separate experimental work or from model fitting. While the model considers many aspects of the metabolism of D. vulgaris, it is computationally intensive and requires greater knowledge of kinetic parameters than is often available in environments such as the GIT. Therefore, its structure is less readily compared or combined with other existing models for the GIT microbiota. We also found that this model shows sensitivity to the initial values for dissolved hydrogen and carbonate concentrations; values that are difficult to determine experimentally and physiologically. As we wish to study SRB in the GIT, we construct a simpler model requiring less inputs to later integrate into a larger microbiota model. Our SRB model considers solely the concentrations of the initial and final metabolites in a metabolic pathway, treating the intermediate metabolites and reactions as a "black box." We calibrate our model using existing experimental data for the monoculture growth of a D. vulgaris strain and use it to predict the dynamics of separate independent experiments with both the same bacterium and a different D. vulgaris strain.

#### MATERIALS AND METHODS

#### Assumptions

For this model it was assumed that the only metabolites involved in the metabolism of D. vulgaris are lactate, acetate, hydrogen, sulphate and hydrogen sulphide (H2S), as these metabolites represent important initial and final metabolites in the major metabolic pathways of D. vulgaris (Keller and Wall, 2011). Other metabolic pathways involving fermentation of alternative organic molecules, such as monosaccharides and fatty acids, and reduction of nitrogenous compounds have been studied in Desulfovibrio and other SRB genera, but appear to be of lesser importance and not widespread within the functional group (Barton and Fauque, 2009). While formate has been implicated in the metabolism of Desulfovibrio species elsewhere (da Silva et al., 2013; Junicke et al., 2015; Martins et al., 2015), here we have assumed that formate may be represented as hydrogen equivalents. This is supported by the similar reduction potentials of formate and hydrogen, allowing for interconversion of the two molecules at low energetic cost to the bacterium (Stams and Plugge, 2009; da Silva et al., 2013; Rabus et al., 2013). Formate concentrations also remained very low (<0.5 mM) in previous experiments with D. vulgaris Hildenborough grown on either lactate and sulphate or lactate and hydrogen (da Silva et al., 2013).

We assume that the medium in which D. vulgaris is grown contains in abundance all other molecules necessary for growth and that these are not significantly depleted during the

experiment. We further assume that D. vulgaris is able to oxidise lactate incompletely to acetate, with concurrent production of hydrogen (Keller and Wall, 2011). This hydrogen may then be utilised in the reduction of sulphate to H2S. We assume that all metabolites remain in the aqueous phase, with the exception of hydrogen, which may transfer between the aqueous and gaseous phases. We assume that all metabolites in the aqueous phase are available to the bacteria in a well-mixed solution. No spatial component is considered in the model.

The assumed stoichiometries for the two reactions, expressing all protons as hydrogen molecule equivalents, are as follows (Thauer et al., 1977; Noguera et al., 1998; da Silva et al., 2013):

$$\text{CH}\_3\text{CHOH}\text{COO}^- \text{ (Lactate)} + 2\text{ H}\_2\text{O}$$

$$\rightarrow \text{CH}\_3\text{COO}^- \text{ (Acetate)} + 2.5\text{ H}\_2 + \text{HCO}\_3^-$$

$$\text{SO}\_4^{2-} \text{ (Salplate)} + 5\text{H}\_2 \rightarrow \text{H}\_2\text{S} + 4\text{ H}\_2\text{O}$$

Note that the bicarbonate molecule (HCO<sup>3</sup> <sup>−</sup>) produced in the oxidation of lactate and the water molecules produced in the reduction of sulphate are not included in the model, as they play no further role in the metabolism of D. vulgaris. Moreover, we assume that the culture remains well buffered throughout the experiment, therefore pH is not altered by changing concentrations of bicarbonate or other metabolites. There have been reports of bicarbonate as a growth-limiting molecule for other bacterial strains (Dobay et al., 2018), but there is currently no evidence of this for SRB. We explain this further in the Discussion.

#### Mathematical Model

The model is based on Monod kinetics for bacterial growth in a batch culture environment (Monod, 1949). Monod kinetics was chosen due to the biological meaning associated with the parameters, as well as the ability to determine these values experimentally if required. The model considers the molar concentration of lactate, acetate, sulphate and H2S, as well as the molar concentration of hydrogen in the aqueous phase and the partial pressure of hydrogen in the gaseous phase, measured in atmospheres. It also considers the concentration of the bacterial population in the aqueous phase (mg L−<sup>1</sup> ). These units were chosen to align with data sources for both the calibration and validation of the model. **Figure 1** shows the general structure of the model.

Following Monod kinetics, we model the rate of change in lactate concentration (L; mM) by

$$\frac{dL}{dt} = -\frac{\mu\_{\text{max},L}X}{Y\_L} \left(\frac{L}{K\_L + L}\right) \tag{1}$$

where µmax,<sup>L</sup> denotes the maximum growth rate (h−<sup>1</sup> ) and Y<sup>L</sup> denotes the biomass yield (mg L−<sup>1</sup> mM−<sup>1</sup> ) of D. vulgaris when grown on lactate. K<sup>L</sup> is the Monod constant (mM) for this bacterium and substrate, also referred to as the half-saturation constant. This value is the concentration of substrate required for the bacterium to attain half of its maximum growth rate. X is the concentration of bacterial cells in the medium (mg L−<sup>1</sup> ).

It is known that high concentrations of hydrogen in the medium inhibit the metabolism of lactate by certain SRB, including D. vulgaris, although the mechanism is not clear (Pankhania et al., 1988; Junicke et al., 2015). As such, we add an inhibition term to our model that reduces the rate of lactate metabolism as the aqueous hydrogen concentration, Haq (mM), increases. Equation 1 then becomes

modelled. H2S: hydrogen sulphide.

$$\frac{dL}{dt} = -\frac{\mu\_{\max,L}X}{Y\_L} \left(\frac{L}{K\_L + L}\right) \left(1 - \frac{H\_{aq}}{H\_{\max}}\right) \tag{2}$$

where Hmax (mM) is the aqueous hydrogen concentration above which lactate degradation is completely inhibited. This formulation also ensures that the rate of lactate degradation reduces proportionally to the aqueous hydrogen concentration. To ensure that the model is robust to hydrogen concentrations above Hmax, we add the following condition:

$$\frac{dL}{dt} = 0 \quad \text{when } H\_{aq} > H\_{max}.$$

The sulphate concentration (S; mM) is given by

$$\frac{dS}{dt} = -\frac{\mu\_{\max,S}X}{Y\_S} \left(\frac{S}{K\_S + S}\right) \left(\frac{H\_{aq}}{K\_H + H\_{aq}}\right). \tag{3}$$

Sulphate and hydrogen are both required for the formation of H2S, hence the inclusion of the aqueous hydrogen concentration in Eq. 3. The equation is adapted from the model equations of Kettle et al. (2015) for multiple essential resources. µmax,<sup>S</sup> denotes the maximum growth rate (h−<sup>1</sup> ) and Y<sup>S</sup> is the biomass yield (mg L−<sup>1</sup> mM−<sup>1</sup> ) of D. vulgaris during sulphate reduction. K<sup>S</sup> and K<sup>H</sup> denote the Monod constants (mM) for sulphate and hydrogen, respectively.

We assume that the aqueous hydrogen concentration is influenced by hydrogen production during the oxidation of

lactate, hydrogen consumption in the reduction of sulphate, and liquid-gas transfer of hydrogen. The rate of change in the concentration of aqueous hydrogen is

$$\frac{dH\_{aq}}{dt} = -b\_{LH}\frac{dL}{dt} + b\_{HP}\frac{dS}{dt} - \frac{1}{\rho\_H}\frac{dH\_\mathcal{g}}{dt}\frac{V\_\mathcal{g}}{V\_{aq}}\tag{4}$$

where bLH is the stoichiometric constant for moles of hydrogen produced per mole lactate metabolised and bHP is the stoichiometric constant for moles of hydrogen required to reduce one mole of sulphate. H<sup>g</sup> is the gaseous hydrogen concentration, measured in atmospheres, and mass transfer between the aqueous and gaseous phases is assumed to be linear, with

$$\frac{dH\_{\text{g}}}{dt} = k\_{L}a \left(\rho\_{H}H\_{aq} - H\_{\text{g}}\right) \frac{V\_{aq}}{V\_{\text{g}}} \tag{5}$$

Equation 5 represents a simple mass transfer model as explained in Kadic and Heindel (2014). Briefly, net transfer between the two phases is determined by the concentration gradient, with the rate of transfer determined by the mass transfer coefficient, k<sup>L</sup> (calculated from the thickness of the film through which molecules must travel and the diffusivity of the molecule in question) and the surface area, a, across which mass transfer may occur. Although other, more complex models do exist for mass transfer between two phases, as only the gaseous hydrogen concentration data is available here, we are limited in our ability to parameterise a more complex model. Although the simplicity of this representation may result in sub-optimal representation of the hydrogen dynamics, we also seek to minimise the number of fitted parameter values in our model, and thus the film model described here is sufficient for our purposes. kLa has the unit h−<sup>1</sup> and V<sup>g</sup> and Vaq (mL) are the fixed volumes of the gaseous and aqueous phases, respectively. ρ<sup>H</sup> (atm mM−<sup>1</sup> ) is the Henry conversion constant for hydrogen. H<sup>g</sup> is measured in atmospheres, whereas Haq is given in mM concentration, therefore we adapt the gas transfer equation used in Muñoz-Tamayo et al. (2016) for our model, giving a ρ<sup>H</sup> value of 1.364 atm mM−<sup>1</sup> .

The rates of change in acetate (A) and H2S (P) concentrations are proportional to the rates of change in the concentrations of lactate and sulphate, respectively.

$$\frac{dA}{dt} = -b\_{LA}\frac{dL}{dt} \tag{6}$$

$$\frac{dP}{dt} = -b\_{SP}\frac{d\mathcal{S}}{dt} \tag{7}$$

where bLA and bSP are constants determined by the stoichiometries of each reaction stated in Section 2.1. Note that we take these stoichiometries directly from the literature and do not include in the model some fraction of substrate being used in the production of cell biomass. This assumption is made as, for the batch culture cases considered here, the experimentally observed stoichiometries of the metabolites closely matched those given in Section 2.1.

Finally, the concentration of bacterial cells in the medium, X (mg L−<sup>1</sup> ), is proportional to the change in lactate and sulphate concentrations, with consideration of the biomass yield terms (assuming the energy requirements for cell maintenance are negligible relative to the growth requirements).

$$\frac{dX}{dt} = -Y\_L \frac{dL}{dt} - Y\_S \frac{d\mathbf{S}}{dt} \tag{8}$$

The system consisting of Eq. 2–8 fully describes the metabolism of D. vulgaris under our set of assumptions. A summary of model notation is given in **Table 5**.

#### Data Capture

Time-course data was captured from the literature using image capturing and graphical input software in MATLAB (The MathWorks<sup>1</sup> ). The mathematical model of Noguera et al. (1998) was reconstructed using the information in the original publication. This information was near complete, the only exception being the absence of initial conditions for some of the model variables. We have therefore made some assumptions based on other information given in the paper, which has allowed us to reproduce good representations of the published model fits.

#### Model Fitting

In order to determine the values of several of the parameters used in the model, model fitting to existing experimental data was performed. Time-course data from Noguera et al. (1998) was collected and used to calibrate the model and estimate parameter values.

The parameter values in **Table 1** were generated by minimising the normalised sum of squared errors between the model prediction and the data. The optimisation was performed using the fminsearch routine in MATLAB (The MathWorks; see text footnote 1).

#### Statistical Analysis

All statistics were calculated in MATLAB using the captured data and corresponding model prediction. A Markov Chain Monte Carlo (MCMC) technique was implemented over 200,000 MCMC iterations. A non-parametric distribution was then fitted to the MCMC sample for each of the nine parameters estimated. The cumulative density function of this distribution was used to obtain a 95% confidence interval.

To compare the proposed model with the existing model of Noguera et al. (1998), we used the corrected Akaike Information Criterion (AICc) (Akaike, 1974; Hurvich and Tsai, 1989):

$$\text{AICc} = \, 2K - 2(\log(\mathcal{L}(\theta))) + \frac{2K(K+1)}{n - K - 1}$$

where n is the number of data points (63), K is the number of parameters of the model and log(L(θ)) is the log likelihood function for the model. Following Burnham and Anderson (2002), we make the substitution

$$\log(\mathcal{L}(\theta)) = -\frac{1}{2}n\log\left(\frac{\text{RSS}}{n}\right)^2$$

where RSS is the normalised residual sum of squares of the model fit to the data. Normalisation, i.e., division by the sample mean in

<sup>1</sup>www.mathworks.com

#### TABLE 1 | Model parameter values.

fmicb-10-01652 July 17, 2019 Time: 14:37 # 5


<sup>∗</sup>These estimates are obtained from different models and therefore a direct comparison cannot be made with the parameters estimated in the paper. They are listed here for reference.

the calculation of the RSS for each data set, was included to ensure the RSS value was not biassed by the scale on which each variable was measured. Finally, we also calculate the Akaike weight, w<sup>i</sup> , for each model as follows (Burnham and Anderson, 2002):

$$\mathbf{w}\_{i} = \frac{l\_{i}}{l\_{1} + l\_{2}}$$

where l<sup>i</sup> = exp − 1 2 (AICc<sup>i</sup> − AICcmin) . Here, i is the model index (1 for the existing model of Noguera et al. (1998), 2 for the model presented here) and AICcmin represents the minimum AICc value of the two models.

#### RESULTS

#### Model Calibration

Data from two separate experiments were used simultaneously to obtain parameter values for the model (Noguera et al., 1998). The first experiment involved the growth of D. vulgaris in medium supplemented with lactate and sulphate (**Figure 2**), while the second experiment took place in the absence of sulphate (**Figure 3**). Our mathematical model was able to describe the trends in growth and metabolite flux dynamics for both these experiments, giving comparable goodness of fit to the more complex model of Noguera et al. (1998; **Table 2**). The parameter values used are shown in **Table 1**.

The model of Noguera et al. (1998) uses seven model fitted parameters and a total of 20 parameters either fitted or estimated from previous experimentation, whereas our model uses nine fitted parameters and one estimated from previous experimentation, giving a total of 10. This discrepancy is due to the increased complexity of the former model, which additionally models the concentrations and gaseous partial pressures of CO2, H2S and bicarbonate, as well as the mass transfer of these molecules between the two phases, and the thermodynamics of each reaction modelled. **Table 3** details the values used for

the AICc calculation. The AICc value for our model was 263.2 compared to an AICc of 282.8 for the model of Noguera et al. (1998). This indicates the aptness of our model to the data considered, although more complex models may be better suited for larger and more complex data sets.

Some of the parameters shown in **Table 1** were fixed to values taken from the literature. The stoichiometric constants

17.3 mM lactate in the absence of sulphate (Noguera et al., 1998): continuous lines display the fit of the current model; dotted lines display the fit of the model described in Noguera et al. (1998). Analysis of model fit is presented in Table 2.

were fixed to correspond with the assumed stoichiometries of the reactions considered and the Henry constant for hydrogen was also obtained from the literature.

It is notable that the best fit parameter values for KL, K<sup>H</sup> and Hmax lie outside their respective MCMC generated 95% confidence interval. This is likely due to the difficulties in estimating half-saturation constants and maximum growth rates simultaneously, as we observed high correlation between these values. This has been observed in Monod model fitting elsewhere [for example, Muñoz-Tamayo et al. (2016)]. We therefore performed a second MCMC run in which the halfsaturation constants were fixed at values obtained from the experimental literature. A comparison of the newly generated confidence intervals for the remaining fitted parameters with the original values is shown in **Table 4**, but the intervals are similar. We therefore analysed the sensitivity of the model prediction to variations in each parameter value (**Supplementary Table S1**). The model prediction for growth in medium with no sulphate, shown in **Figure 3**, was not notably sensitive to small changes in any fitted parameter value except for Hmax, which determines the final partial pressure of gaseous hydrogen. Contrastingly, the model fit to gaseous hydrogen shown in **Figure 2** showed sensitivity to a number of parameters. Small variations in the maximum growth rates, half-saturation constant for lactate, yield values and the stoichiometric constants bLH and bHP, all resulted in relatively large changes in the quality of fit of the model to the gaseous hydrogen data. The change in the goodness of fit to the other data types was minimal. We also found that the model fit was only slightly sensitive to small changes in the initial conditions for lactate, sulphate and bacterial concentration and

#### TABLE 2 | Analysis of model fits to the calibration data.

fmicb-10-01652 July 17, 2019 Time: 14:37 # 7


†CCC: Concordance correlation coefficient (Lin, 1989).

TABLE 3 | AIC calculation values.


insensitive to such changes in the initial conditions for other metabolites. This was in contrast to the model of Noguera et al. (1998), which we found to be disproportionately sensitive to small changes in the initial conditions for dissolved hydrogen and carbonates: variables less likely to have a strong effect on culture dynamics than lactate, sulphate and bacterial concentrations.

#### Model Validation

The model was validated against a number of different experimental data sources (Noguera et al., 1998; da Silva et al., 2013). **Figure 4** shows the model simulation for gaseous hydrogen

dynamics in medium lacking lactate, where D. vulgaris may only perform sulphate reduction, until the available hydrogen is depleted [data from Noguera et al. (1998)].

**Figures 5**, **6** show the comparison between the model prediction and experimental data from further validation experiments, with altered initial conditions [data from Noguera et al. (1998)]. Unfortunately, for these and the experiments from which **Figures 3**, **4** were generated, data for aqueous metabolite concentrations and bacterial growth are unavailable, so we cannot verify the model predictions for these variables. We also have no information regarding the concentration of bacteria at the beginning of the experiment, therefore 9.4 mg L−<sup>1</sup> , the initial bacterial concentration in previous experiments, was assumed.

The model predicts the full utilisation of lactate and only partial consumption of sulphate in **Figure 5**, but is not able to capture the delay in hydrogen accumulation in the headspace observed in the first few hours of the experiment. The same


is true of **Figure 6**. Here, the model accurately predicts the final concentration of lactate remaining in the medium at the end of the experiment as 2.98 mM, compared to the observed value of 2.58 mM. However, the model overpredicts the gaseous hydrogen accumulation. Under our model assumptions we expect hydrogen to accumulate to the inhibitory level, whereas in the experiment hydrogen production was far lower. Given that the model accurately predicted the lactate degradation, this would imply that less hydrogen is produced under the conditions shown in **Figure 6** than under the assumed stoichiometry. Hydrogen accumulation was not measured after 48 h in the experiment, therefore it is not possible to know whether and at what point hydrogen accumulation peaks.

**Figure 7** shows the validation of the both our model and that of Noguera et al. (1998) against separate experimental data for D. vulgaris Hildenborough, taken from da Silva et al. (2013). The experimental starting concentration of bacteria was not stated for this data set, so we fitted this value to the data with all other parameters fixed at their previously determined values. This gave an initial bacterial concentration of 6.75 mg L−<sup>1</sup> for our model and 0.038 mg L−<sup>1</sup> for the model of Noguera et al. (1998). As shown in **Figure 7**, the models performed similarly with their respective initial bacterial concentrations, with the exception of the gaseous hydrogen prediction, and both accurately captured the rate of lactate degradation and acetate production with no alteration to the parameter values obtained during model calibration. The large discrepancy between the obtained initial bacterial concentrations for the two models prompted further investigation. The initial optical density (OD) recorded for this experiment was approximately 0.025 (da Silva et al., 2013). No calibration to other units was performed by these authors and few exist in the literature for Desulfovibrio strains, but Bernardez and de Andrade Lima (2015) suggested a conversion of: dry weight (mg) = exp (5.12 OD–4.987), which gives an approximate initial bacterial concentration for this experiment of 7.76 mg L−<sup>1</sup> . Although the conditions under which this conversion was derived differ from the experiment of da Silva et al. (2013), this estimate compares well to that of our model.

The final acetate measurement in **Figure 7** was not predicted by either model, and it is not clear to where the remaining carbon from lactate degradation was directed in this experiment. D. vulgaris has the potential to use acetyl-CoA, an intermediate on the lactate oxidation pathway, in the biosynthesis of certain branched-chain amino acids and fatty acids, as well as in an incomplete citric acid cycle (Heidelberg et al., 2004), but only the metabolites shown in **Figure 7** were measured. However, separate experiments by these authors with concentrated cell suspensions found the expected 1:1 ratio of lactate degraded to acetate produced (da Silva et al., 2013).

#### DISCUSSION

prediction of the current model. R

This model provides a simpler mathematical representation of SRB metabolism than is currently available in the literature (Noguera et al., 1998), with similar predictive capability. As such, it can be more easily adapted to specific strains and culture conditions, not limited to SRB of the human GIT. The inclusion of further characteristics of specific SRB strains could be realised with the addition of further terms to existing equations, or the inclusion of further equations if additional metabolites were considered. For example, complete growth inhibition of a SRB strain due to sulphide concentrations above 16.1 mM has been shown previously (Reis et al., 1992). Acetate inhibition has also been investigated for SRB, with approximately 54 mg L−<sup>1</sup> undissociated acetic acid (≈45.9 mM acetate) resulting in 50% growth inhibition (Reis et al., 1990). Both these concentrations are greater than those measured in the experiments used here, and the H2S concentration is greater than that reported in faeces (Magee et al., 2000). However, the model could be adjusted to include inhibition terms for acetate and H2S for application of the model to more extreme environments. These terms could take the form used here for hydrogen inhibition, but alternative inhibition terms could be more appropriate and should be assessed by model fitting (see Han and Levenspiel (1988) for a list of inhibition terms and a generalised form). At present, we are not aware of any timecourse data involving such concentrations of these metabolites with which to parameterise the model.

It would also be useful to investigate experimentally the influence of bicarbonate on the growth rate of SRB. Several human-associated bacterial strains have shown reduced growth rates when exposed to 100 mM of bicarbonate in monoculture (Dobay et al., 2018). This molecule was also shown to disrupt biofilm formation in selected strains. D. vulgaris is a biofilm forming organism (Clark et al., 2007), but no SRB were studied in the bicarbonate inhibition experiments, so we cannot make any inference about the influence of this molecule on growth rates in our model. However, following the expected stoichiometry of the D. vulgaris metabolic pathways, we would anticipate less than 20 mM of bicarbonate could be produced in the experiments of Noguera et al. (1998), and up to 40 mM in the experiments of da Silva et al. (2013), considerably lower than those found to be growth limiting in Dobay et al. (2018). Bicarbonate is secreted into the gastrointestinal lumen in humans, reaching comparable concentrations to those expected in these experiments: bicarbonate concentration at the start of the colon is estimated at around 30 mM (Gennari and Weise, 2008). Further experimental investigation is needed to determine

whether, and to what extent, bicarbonate may be growth limiting to SRB before it can be included in a model.

Time-course data is also unavailable for the use of acetate as a carbon source by SRB, which has been shown in the absence of lactate [for example, Pankhania et al. (1986)]. We expect that acetate uptake is occurring in the data shown in **Figure 4**, as it is the sole available carbon source in the medium, but this was not measured. Experiments measuring acetate concentrations over time when this is the sole carbon source are required to determine the parameter values of acetate utilisation via model fitting.

Modelling mass transfer in experiments such as those described here is challenging. Due to limited available experimental data, we chose to use a simple mass transfer model to minimise the number of fitting parameters required. Mass transfer is modelled under the assumption of linear dynamics, but without knowledge of the concentration of dissolved hydrogen it is unclear how much this assumption biases the model. The model may be more limited in its ability to accurately capture hydrogen transfer between phases than other, more complex model structures (Kadic and Heindel, 2014). This simple structure may be partially responsible for the sensitivity of the gaseous hydrogen model fit to small changes in some of the parameter values of the model. However, we believe that the model fit to the lactate, acetate and sulphate data are of greater importance than those of gaseous hydrogen and bacterial growth for several reasons. The apparent initial lag phase in the gaseous hydrogen data from the experiments considered here was not captured by our model, despite the good fit to the data for other metabolites. While the inclusion of a lag phase in the model could rectify this aspect, such an addition would complicate a model that we wish to keep parsimonious and we also do not have a probable physiological cause for such a lag. The experimental data shows large variation in gaseous hydrogen pressure between replicates in both the calibration and validation datasets. The data for the concentration of bacterial cells in the medium is similarly limited. Only two measurements were taken during the exponential growth phase in **Figure 2**, and the error on both of these measurements is greater than 25% of the mean value. It is also unclear how reliable the initial value for cell concentration is, since this was assumed from the inoculum rather than measured. Although our model proved only slightly sensitive to certain initial condition values, measuring the initial concentrations of both cells and metabolites would be of great value. The data for lactate, acetate and sulphate concentrations are more complete and more repeatable, encouraging emphasis on the model fit to these data.

Uncertainty remains in the field around the nature of hydrogen production and use by SRB. Previously, there have been arguments both for and against its status as a mandatory intermediate in the simultaneous oxidation of organic compounds and reduction of sulphate, as well as the role of various hydrogenase enzymes (Keller and Wall, 2011; Rabus et al., 2013). The importance of hydrogen in the reduction of sulphate has also been shown differ between SRB species [see review by Rabus et al. (2015)]. We believe that one of the strengths of the model is its avoidance of any biassing assumption about

#### TABLE 5 | Model notation.

fmicb-10-01652 July 17, 2019 Time: 14:37 # 11


the nature of these relationships by using our two hydrogen compartments, aqueous and gaseous, as a method to represent hydrogen equivalents that are immediately available for use in sulphate reduction or not, respectively.

The mathematical model presented here is simpler in its construction than previous attempts to capture SRB dynamics. Our model uses nine fitted parameters (10 parameters in total), compared to seven fitted and three experimentally estimated parameters (20 parameters in total) in Noguera et al. (1998), and seven differential equations compared with ten in Noguera et al. (1998). Our model also shows good fits to experimental data as assessed by common measures for model analysis for two D. vulgaris strains from several independent experiments under varied conditions. While the model of Noguera et al. (1998) considers more factors, including the thermodynamics of the conversions performed by the bacteria and the concentrations of a greater number of metabolites, these inclusions can be limiting when investigating the metabolism of SRB in environments where knowledge of these factors is not available. For example, application of the model of Noguera et al. (1998) to the human GIT would be challenged by host influences on variables. The applied model would need to consider appropriate representation of bicarbonate and CO<sup>2</sup> when including secretion and absorption by the host, as well as the implications of host metabolite absorption on the modelled thermodynamic inhibition of the metabolic reactions. By contrast, the relative simplicity of our model means it can more easily be adapted to the specific environmental conditions of the GIT and has greater flexibility for the inclusion of additional influences upon the metabolism of these bacteria. In this way the model could be adapted to provide a representative model for the SRB functional group in the GIT.

Regarding dynamics in the GIT, current existing data from rodent models support the increased efficiency of carbohydrate breakdown by saccharolytic bacteria in the presence of either a methanogen or acetogen due to hydrogen metabolism by these microbes (Samuel and Gordon, 2006; Rey et al., 2010). However, there is no such evidence for the SRB, although in theory the same role could be filled by these bacteria (Smith et al., 2018). This may be due to competition for other substrates, which could be investigated using the model presented here in combination with existing models for saccharolytic bacteria [such as Kettle et al. (2015)].

It is our intention to use the SRB model presented here as part of a larger model including other hydrogenotrophic and hydrogenogenic microbes of the human GIT, to examine the role of hydrogen in this environment. Mathematical models for the GIT microbiota are available (Muñoz-Tamayo et al., 2010; Kettle et al., 2015, 2017), but as yet do not consider the action of SRB. The inclusion of this functional group may further enhance their predictive capabilities and could eventually be used to address the role of the SRB in human nutrition and health. Such community modelling should not be limited to the GIT, as the combination of models such as that presented here with similar structures for methanogens and reductive acetogens may reveal information about the cross-feeding and competitive relationships between these hydrogenotrophs in other environments.

#### DATA AVAILABILITY

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

#### AUTHOR CONTRIBUTIONS

NS performed the mathematical modelling and drafted the manuscript. All authors contributed to model and manuscript revision, and read and approved the submitted version.

#### FUNDING

NS was supported by a Ph.D. fellowship from the Riddet Institute through funding provided by the New Zealand Tertiary Education Commission. We are also grateful to AgResearch for the resources provided to NS.

#### ACKNOWLEDGMENTS

fmicb-10-01652 July 17, 2019 Time: 14:37 # 12

The authors thank I. Vetharaniam and W. Young at AgResearch for their thoughtful and useful comments on the manuscript.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.01652/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 © 2019 Smith, Shorten, Altermann, Roy and McNabb. 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.

## Hydrogen Oxidation Influences Glycogen Accumulation in a Verrucomicrobial Methanotroph

Carlo R. Carere<sup>1</sup> \*, Ben McDonald<sup>2</sup> , Hanna A. Peach<sup>3</sup> , Chris Greening<sup>4</sup> , Daniel J. Gapes<sup>2</sup> , Christophe Collet<sup>2</sup> and Matthew B. Stott<sup>5</sup>

<sup>1</sup> Department of Chemical and Process Engineering, University of Canterbury, Christchurch, New Zealand, <sup>2</sup> Scion, Te Papa Tipu Innovation Park, Rotorua, New Zealand, <sup>3</sup> Geomicrobiology Research Group, Department of Geothermal Sciences, GNS Science, Taupo, New Zealand, ¯ <sup>4</sup> School of Biological Sciences, Monash University, Clayton, VIC, Australia, <sup>5</sup> School of Biological Sciences, University of Canterbury, Christchurch, New Zealand

#### Edited by:

Roland Hatzenpichler, Montana State University, United States

#### Reviewed by:

Ludmila Chistoserdova, University of Washington, United States Grayson L. Chadwick, California Institute of Technology, United States Lisa Y. Stein, University of Alberta, Canada

> \*Correspondence: Carlo R. Carere

carlo.carere@canterbury.ac.nz

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 06 June 2019 Accepted: 29 July 2019 Published: 16 August 2019

#### Citation:

Carere CR, McDonald B, Peach HA, Greening C, Gapes DJ, Collet C and Stott MB (2019) Hydrogen Oxidation Influences Glycogen Accumulation in a Verrucomicrobial Methanotroph. Front. Microbiol. 10:1873. doi: 10.3389/fmicb.2019.01873 Metabolic flexibility in aerobic methane oxidizing bacteria (methanotrophs) enhances cell growth and survival in instances where resources are variable or limiting. Examples include the production of intracellular compounds (such as glycogen or polyhydroxyalkanoates) in response to unbalanced growth conditions and the use of some energy substrates, besides methane, when available. Indeed, recent studies show that verrucomicrobial methanotrophs can grow mixotrophically through oxidation of hydrogen and methane gases via respiratory membrane-bound group 1d [NiFe] hydrogenases and methane monooxygenases, respectively. Hydrogen metabolism is particularly important for adaptation to methane and oxygen limitation, suggesting this metabolic flexibility may confer growth and survival advantages. In this work, we provide evidence that, in adopting a mixotrophic growth strategy, the thermoacidophilic methanotroph, Methylacidiphilum sp. RTK17.1 changes its growth rate, biomass yields and the production of intracellular glycogen reservoirs. Under nitrogen-fixing conditions, removal of hydrogen from the feed-gas resulted in a 14% reduction in observed growth rates and a 144% increase in cellular glycogen content. Concomitant with increases in glycogen content, the total protein content of biomass decreased following the removal of hydrogen. Transcriptome analysis of Methylacidiphilum sp. RTK17.1 revealed a 3.5-fold upregulation of the Group 1d [NiFe] hydrogenase in response to oxygen limitation and a 4-fold upregulation of nitrogenase encoding genes (nifHDKENX) in response to nitrogen limitation. Genes associated with glycogen synthesis and degradation were expressed constitutively and did not display evidence of transcriptional regulation. Collectively these data further challenge the belief that hydrogen metabolism in methanotrophic bacteria is primarily associated with energy conservation during nitrogen fixation and suggests its utilization provides a competitive growth advantage within hypoxic habitats.

Keywords: methanotroph, hydrogenase, glycogen, extremophile, methylacidiphilum

### INTRODUCTION

fmicb-10-01873 August 14, 2019 Time: 15:24 # 2

Aerobic methane oxidizing bacteria (methanotrophs) serve as the primary biological sink for the potent greenhouse gas methane (CH4) (Kirschke et al., 2013). Methanotrophs grow by oxidizing CH<sup>4</sup> to methanol with a particulate or soluble methane monooxygenase enzyme (pMMO/sMMO) and subsequently yield reducing equivalents (e.g., NADH) for cellular respiration and biosynthesis through the oxidation of methanol to carbon dioxide (CO2). The gammaproteobacterial (Type I) and alphaproteobacterial (Type II) methanotrophs generate biomass by assimilating the intermediates formaldehyde or formate via the ribulose monophosphate (RuMp) or serine pathways (Hanson and Hanson, 1996) respectively, whereas the verrucomicrobial methanotrophs oxidize methanol directly to formate (Keltjens et al., 2014) and generate biomass by fixing inorganic carbon (CO2) via the Calvin–Benson–Bassham cycle (Khadem et al., 2012b). Despite the apparent restriction of most methanotrophs to grow on one carbon compounds (C1), they thrive at the interface of various oxic/anoxic habitats (e.g., peat bogs, forest soils, wetlands, rice paddies and geothermal environments) (Dunfield et al., 2007; Singh et al., 2010; Knief, 2015), where the availability of oxidant (O2), energy and carbon resources for growth is likely to fluctuate. Given the methane monooxygenase reaction (CH<sup>4</sup> + O<sup>2</sup> + [NAD(P)H + H+]/QH<sup>2</sup> CH3OH + NAD(P)+/Q + H2O) and the aerobic respiratory chain require a continual source of reductant and oxidant, methanotrophic bacteria must regulate their carbon, energy and resource allocation to fulfill metabolic demands for cellular growth and persistence (Hanson and Hanson, 1996).

Many bacterial species, including methanotrophs, accumulate biopolymers (e.g., glycogen, polyhydroxyalkanoates), phospholipids, and intracellular osmolytes (e.g., ectoine, sucrose) (Strong et al., 2016) in response to unbalanced growth conditions. This allows resources to be strategically conserved for assistance in times of starvation. The biosynthesis of glycogen, a highly branched polysaccharide consisting of α-1,4 bonded glucose residues with additional α-1,6 branched sidechains, is a common metabolic strategy for carbon storage that is shared among evolutionarily distant species (Wilson et al., 2010). Glycogen production has been widely described within Type I methanotroph species (Linton and Cripps, 1978; Eshinimaev et al., 2002) and the production of this compound has recently been reported in the verrucomicrobial methanotroph, Methylacidiphilum fumarolicum SolV (Khadem et al., 2012a). The physiological role of glycogen production in methanotrophs is not precisely understood, although it is believed to serve a role in environmental survival during periods of starvation and has been implicated to symbiotic performance, colonization and virulence (Bonafonte et al., 2000; McMeechan et al., 2005; Bourassa and Camilli, 2009; Wilson et al., 2010). Although the accumulation of intracellular glycogen may occur optimally during exponential growth (Gibbons and Kapsimalis, 1963; Eidels and Preiss, 1970), its synthesis is typically associated with entry into stationary phase when growth is limited due to the limitation of some critical nutrient (i.e., nitrogen, phosphate) or in the presence of excess carbon (Wilson et al., 2010). In bacteria, the biosynthesis of glycogen occurs by utilizing ADP-glucose as the glycosyl donor for polymer extension (Preiss, 1984). The precise mechanisms governing glycogen biosynthesis in bacteria, however, remain obscure. It is likely energy availability and redox status play a primary role in regulating glycogen biosynthesis, as ATP acts as substrate for the ADP-glucose producing reaction catalyzed by glucose-1-phosphate adenylyltransferase (Preiss, 1984).

To remain competitive within dynamic environments (Knief et al., 2003; Tavormina et al., 2010), some methanotrophs supplement CH<sup>4</sup> usage with other energy-yielding strategies (Dedysh and Dunfield, 2010). Several recent studies have revealed a few strains, notably Methylocella silvestris, utilize a suite of carbon and energy substrates, including simple organic acids, alcohols and short-chain alkane gases (Dedysh et al., 2005; Crombie and Murrell, 2014). Aerobic H<sup>2</sup> metabolism has also been shown in a range of methanotrophs (Chen and Yoch, 1987; Shah et al., 1995; Hanczar et al., 2002) and a wide range of hydrogenases have been shown to be distributed in methanotroph genomes (Greening et al., 2016). While H<sup>2</sup> oxidation was originally implicated in energy conservation in response to N<sup>2</sup> fixation (Takeda, 1988), more recent findings indicate that H<sup>2</sup> serves a multifaceted role in the growth and survival of these bacteria. Of the verrucomicrobial methanotrophs, the activity of respiratorylinked group 1d hydrogenases can provide sufficient energy to sustain chemolithoautotrophic growth on H<sup>2</sup> alone (Mohammadi et al., 2016; Carere et al., 2017). Further, mixotrophic growth (H<sup>2</sup> and CH4) in the thermoacidophile Methylacidiphilum sp. RTK17.1 has been observed under O2 limiting conditions and is proposed to provide a competitive advantage over obligate methanotrophy at oxic/anoxic soil boundaries within geothermal environments (Carere et al., 2017). This suggests that the additional energetic input of H<sup>2</sup> may counter the effect of otherwise unbalanced growth conditions.

The influence of H<sup>2</sup> metabolism on the production of intracellular energy reservoirs, commonly associated with unbalanced growth, within methanotrophic bacteria has yet to be elucidated. In this work, we investigate the effect of H<sup>2</sup> metabolism on glycogen production within the methanotroph, Methylacidiphilum sp. RTK17.1. Chemostat cultivation was performed during O2-replete and O2-limited cultivation, in the presence of NH<sup>4</sup> <sup>+</sup> or N2, with or without H<sup>2</sup> in the headspace, to determine the influence of H<sup>2</sup> metabolism on observed growth rates, biomass production characteristics and transcriptional regulation. We show that cellular growth rates, molar growth yields, and the allocation of resources between protein and glycogen production vary depending on the supply of H2, O2, and nitrogen (as NH<sup>4</sup> <sup>+</sup> or N2). Transcriptome data provided a basis of findings, showing significant differential regulation of operons encoding the group 1d [NiFe]-hydrogenase, methane monooxygenases, and nitrogenase between the conditions. In turn, these findings enhance understanding of the physiological strategies that methanotrophs use to grow and survive in different environments.

#### MATERIALS AND METHODS

fmicb-10-01873 August 14, 2019 Time: 15:24 # 3

#### Chemostat Cultivation of Methylacidiphilum sp. RTK17.1

Chemostat cultivation was performed to investigate the influence of H<sup>2</sup> metabolism on the growth and production of intracellular glycogen reservoirs in Methylacidiphilum sp. RTK17.1 with respect to unbalanced growth conditions (O<sup>2</sup> and nitrogen limitation). As previously described, a 1 l bioreactor (BioFlo 110; New Brunswick Scientific, Edison, NJ, United States) equipped with an InPro 6810 Polarographic Oxygen Sensor (Mettler-Toledo, Columbus, OH, United States) was used for all cultivations (Carere et al., 2017). Cultures were continuously incubated at pH 2.5, 50◦C with stirring (800 rpm). The reactor was constantly maintained at a volume of 0.5 l with V4 mineral medium (Carere et al., 2017), prepared with or without NH4Cl (0.4 g l−<sup>1</sup> ) addition (as necessary), and supplied at a constant flow rate of 10 ml h−<sup>1</sup> (D = 0.02 h−<sup>1</sup> ). Custom gas mixtures were supplied to the chemostat at a rate of 10 ml min−<sup>1</sup> and contained approximately (v/v) 3% CH<sup>4</sup> and 26% CO<sup>2</sup> for all experiments; O<sup>2</sup> at (v/v) 14.1% and 3.5%, respectively for O2-replete and limiting conditions, and H<sup>2</sup> at 0.4% (v/v). The balance of all gas mixtures was made up with N2.

Cell densities were monitored at 600 nm using a Ultrospec 10 cell density meter (Amersham Bioscience, United Kingdom) with one unit of OD<sup>600</sup> equivalent to 0.43 g l−<sup>1</sup> cell dry weight for Methylacidiphilum sp. RTK17.1. Influent and effluent gas concentrations were monitored using a 490 micro GC equipped with a thermal conductivity detector (Agilent Technologies, United States). After achieving a steady-state condition as determined by OD600, gas concentrations were monitored over several days and used as a basis to calculate growth and specific gas consumption rates. Biomass samples of Methylacidiphilum sp. RTK17.1 were harvested during steady-state operation for subsequent transcriptome sequencing, biomass cell dry weight determinations, intracellular glycogen, total protein and amino acid levels measurements.

#### Transcriptome Sequencing

Cell culture samples for transcriptome sequencing were harvested (10 ml) from steady-state chemostat experiments, pelleted by centrifugation at 5,000 x g (15 min, 4◦C), suspended in 1 ml RNAlater Stabilization solution (Thermo Fisher Scientific) and then stored at −80◦C until required for further analysis, as per the manufacturer's recommended protocols. The extraction and sequencing of RNA was performed by Macrogen Inc. (Seoul, Korea). Briefly, isolation of mRNA was performed using the RNeasy Mini kit (Qiagen) according to the manufacturer's protocol. Following total RNA extraction, ribosomal RNAs were removed using the Ribo-Zero rRNA removal kit (bacteria) and the quality of the remaining RNA was assessed using an Agilent 2100 Bioanalyzer (Agilent). Library construction was performed using the TruSeq Stranded Total RNA Sample Prep (microbe) Kit (Illumina) and sequencing was performed using an Illumina HiSeq2500 platform. From this, an average of 7,988,451 raw untrimmed reads were obtained for each of the five conditions sampled. These reads were then analyzed using the Artificial Intelligence RNA-Seq pipeline (Sequentia Biotech, Barcelona, Spain), as described elsewhere (Vara et al., 2019), which were reduced to an average of 7,287,318 following quality filtering and trimming. Retained paired-end reads (100 bp) were then mapped to the genome of Methylacidiphilum infernorum strain V4 (GCA\_000019665.1) (Hou et al., 2008) using the 'different genotype' setting. An average of 80.01% reads were mapped to genes within the reference genome for the five experimental conditions (condition 1: O<sup>2</sup> limiting, N2, no H<sup>2</sup> addition; condition 2: O<sup>2</sup> limiting, N2, H<sup>2</sup> addition; condition 3: O<sup>2</sup> limiting, NH<sup>4</sup> <sup>+</sup>, H<sup>2</sup> addition; condition 4: O<sup>2</sup> replete, NH<sup>4</sup> <sup>+</sup>, H<sup>2</sup> addition; condition 5: O<sup>2</sup> replete, NH<sup>4</sup> +, no H<sup>2</sup> addition). Following this, differential gene expression profiles and accompanying statistical analysis was performed to investigate regulation using the edgeR (Robinson et al., 2010) tool available within the Artificial Intelligence RNA-Seq pipeline. Synonymous conditions were grouped as replicates for differential gene expression analysis during oxygen limiting (conditions 1, 2, and 3) and oxygen excess (conditions 4 and 5) growth. Likewise, conditions were grouped as replicates for differential gene expression analysis under nitrogen fixing (N2; conditions 1 and 2) and nitrogen excess (NH<sup>4</sup> <sup>+</sup>; conditions 3, 4, and 5) growth conditions, respectively. Where provided, expression values are given as FPKM (Fragments per Kilobase Million; **Supplementary Table S1**) (Mortazavi et al., 2008). Raw and processed transcriptome sequence files (accession numbers GSM3872525-GSM3872529) were subsequently deposited into the Gene Expression Omnibus (GEO<sup>1</sup> ) for archival storage.

#### Characterization of Biomass

Effluent biomass, produced during experimental steady-state chemostat operation, was collected and stored at 4◦C over a period of 7 days for all biomass characterization studies. Following collection of approximately 2 l culture, cells were pelleted (5,000 × g, 20 min, 4◦C) and stored at −20◦C until required. Characterization of Methylacidiphilum sp. RTK17.1 biomass (crude protein, ash content, amino acid composition) was performed at the Massey University Nutrition Laboratory (accredited to ISO 17025; New Zealand) according to the official methods of analysis of the Association of Official Analytical Communities (AOAC, 2005) international. Briefly, total crude protein and ash content (% w/w) were determined via the Dumas method (AOAC method 968.06) (Ebeling, 1968) and furnace methods, respectively (AOAC method 942.05) (Thiex et al., 2012). Amino acid profile determination of acid-stable residues was performed via reverse-phase high performance liquid chromatography (HPLC) separation using AccQ derivatization of biomass (60–140 mg) samples following oxidization with performic acid and hydrolysis with hydrochloric acid as described in AOAC method 994.12 (AOAC, 2005).

The concentration of glycogen within crude cell extracts of Methylacidiphilum sp. RTK17.1 was determined using a Dionex ICS 3000 HPLC equipped with a Biorad Aminex HPX-87H column and a Shodex RI-101 refractive index detector. Triplicate

<sup>1</sup>https://www.ncbi.nlm.nih.gov/geo/

cell pellets, dried to constant weight (5–15 mg), were suspended in 1M NaOH (0.9 ml) and disrupted by boiling lysis for 1 h within sealed screw top Micro Tubes (Thermo Fisher Scientific). The efficacy (>99%) of cell lysis was confirmed microscopically. Crude extracts were then incubated with amyloglucosidase (35 U/ml crude) from Aspergillus niger (Sigma-Aldrich), following acidification with acetate buffer (0.1 ml) to pH 4.8, for 8 h at 45◦C to convert intracellular glycogen reservoirs into glucose. Following this, the resulting glucose was quantified by HPLC and normalized against a standard curve of known glycogen content (Sigma) that had been contemporaneously treated with amyloglucosidase. To account for non-glycogen derived glucose in cell samples, extracts that did not undergo amyloglucosidase treatment were similarly analyzed by HPLC in parallel. Total intracellular glycogen content was then expressed as a function of cell dry weight (% w/w). Observed differences in growth and biomass characteristics were analyzed for statistical significance using the Tukey's honest significance test tool available in Prism v7.0a (Graphpad Software, Inc.).

#### RESULTS

#### Hydrogen Availability Differentially Affects Growth and Biomass Allocation Depending on Oxygen and Nitrogen Availability

The growth characteristics and biomass production of Methylacidiphilum sp. RTK17.1 was compared in chemostats in four different conditions that differed in O2, nitrogen (NH<sup>4</sup> <sup>+</sup>/N2), and H<sup>2</sup> supply. Results from these experiments are provided in **Table 1**.

Under nitrogen-replete conditions (i.e., NH<sup>4</sup> <sup>+</sup> present in the medium), Methylacidiphilum sp. RTK17.1 displayed many of the same growth characteristics in response to O<sup>2</sup> availability as found previously (Carere et al., 2017) and results were consistent with similar studies on M. fumarolicum SolV (Khadem et al., 2012a). The observed rate of steady-state biomass production (mg l−<sup>1</sup> h −1 ) of Methylacidiphilum sp. RTK17.1) decreased by 33.1% following the transition from O2-replete to O2-limiting cultivation (p-value: <0.0001). This decrease in growth rate was accompanied by a 30.4% reduction in the volumetric rate (mmol l−<sup>1</sup> h −1 ) of CH<sup>4</sup> consumption and a 19-fold increase in the observed rate of H<sup>2</sup> oxidation (p-value: <0.0001). With respect to specific gas consumption rates normalized against biomass production (mmol gCDW−<sup>1</sup> h −1 ), observed rates of CH<sup>4</sup> consumption did not change significantly (pvalue: 0.988), whereas H<sup>2</sup> consumption increased 36-fold (pvalue: <0.0001). Intracellular glycogen content was observed to increase dramatically from 11.26 (± 0.14)% during O2-replete growth to 20.23 (± 0.77)% (w/w; p-value: < 0.0001) during O2-limiting cultivation.

We observed significant changes in these parameters when Methylacidiphilum sp. RTK17.1 was cultivated under O2 limitation depending on H<sup>2</sup> and nitrogen supply. In the presence of H2, observed rates of Methylacidiphilum sp. RTK17.1 biomass production decreased by 8.61% (p-value: 0.019) during N2 fixation compared to growth on NH<sup>4</sup> <sup>+</sup> replete media. In comparison, removal of H<sup>2</sup> from the headspace resulted in an 18.67% reduction in the observed rate of biomass production compared to cells cultured in NH<sup>4</sup> <sup>+</sup>-replete media (from 5.57 to 4.33 mg l−<sup>1</sup> h −1 ; **Table 1**; p-value: < 0.0001). These results suggest the activity of respiratory-linked aerobic H<sup>2</sup> oxidation may, at least partially, serve to offset the energetic demand imposed by N<sup>2</sup> fixation. In support of this, onset of nitrogen limitation in cultures actively respiring H<sup>2</sup> did not significantly increase the production of intracellular glycogen (p-value: 0.999). Following the removal of H<sup>2</sup> from the reactor feedgas, however, internal glycogen content significantly increased from 20.00 (± 2.93)% to 48.86 (± 4.32)% (w/w; p-value: < 0.0001). This corresponds to a 2.03 fold increase in the volumetric production rate of glycogen (**Table 1**). In the absence of CH4, at 4◦C, cellular glycogen reservoirs were depleted within 100 days (**Supplementary Figure S1**).

The total protein and amino acid content of Methylacidiphilum sp. RTK17.1 biomass produced during steadystate growth was also determined (**Figure 1**). A maximum total protein content of 53.9 (± 2.69)% (w/w) was achieved during growth under O2-replete conditions with NH<sup>4</sup> <sup>+</sup> as a readily available source of nitrogen. Compared to O2-replete growth, a minor but insignificant decrease in total protein content of biomass coincided with reduced O<sup>2</sup> availability (51.2 ± 2.56%), and in the transition from NH<sup>4</sup> <sup>+</sup> to N<sup>2</sup> as a source of nitrogen (51.9 ± 2.56%). Consistent with the observation that glycogen production increased in the absence of H<sup>2</sup> (under O2-limiting, N2-fixing growth conditions), the least total protein content of biomass (42.5 ± 8.94%) was observed in conditions where energy availability is likely to constrain cell growth (i.e., O2-limiting, N2-fixing, no H2; p-value: 0.018; **Figure 1A**). Observed changes in the concentration of specific amino acid residues under all of the growth conditions were generally consistent with the changes associated with total protein determinations (**Figure 1C**). For each of the experimental conditions, glutamic acid, leucine, aspartic acid, lysine and alanine were the amino acids in greatest abundance whereas methionine, histidine and cysteine were the least abundant residues.

#### Transcriptome Analysis Reveals Changes in Hydrogenase, Nitrogenase, and Methane Monooxygenase Expression Between the Conditions

Transcriptome analysis was performed on chemostat grown cultures of Methylacidiphilum sp. RTK17.1 to determine whether genes associated with energy metabolism (CH<sup>4</sup> or H2), glycogen synthesis, N<sup>2</sup> or CO<sup>2</sup> fixation were regulated in response to O<sup>2</sup> or nitrogen availability (**Figure 2**). For these experiments, to resolve the possible influence of H<sup>2</sup> on transcriptional responses, a fifth chemostat condition (condition 5: O<sup>2</sup> replete, NH<sup>4</sup> <sup>+</sup>, no H<sup>2</sup> addition) was added to the four growth conditions described in **Table 1**. In response to O<sup>2</sup> limitation, 36 genes were identified as significantly upregulated (p-value: <0.001; False discovery rate (FDR): < 0.05) and 36 genes were significantly downregulated


TABLE 1 | Growth and productivity characteristics of Methylacidiphilum sp. RTK17.1 during chemostat cultivation.

<sup>a</sup>Feedgas was continuously supplied at a rate of 10 ml min−<sup>1</sup> with the following compositions (where applicable;% v/v): O2-replete, 14.1%; O2-limiting, 3.5%, H<sup>2</sup> 0.4%. For all experiments excess CH<sup>4</sup> was supplied at 3.2%, CO<sup>2</sup> at 26% with the balance made up with N2. O<sup>2</sup> saturations in the medium were 57.5 and 0.17% for the O2-replete and O2-limiting conditions, respectively. <sup>b</sup>A dilution rate of 0.02 h−<sup>1</sup> (10 ml h−<sup>1</sup> ) was maintained for all experiments, with NH<sup>4</sup> <sup>+</sup> supplied at an influent concentration of 0.4 g l−<sup>1</sup> where applicable. <sup>c</sup>Replicate numbers for each growth condition correspond to n = 6, n = 6, n = 4 and n = 7 moving from the leftmost column to rightmost column, respectively. The standard deviation from the average of these measurements is shown in brackets.

(p-value: < 0.001; FDR: < 0.05). Subunits of the particulate methane monooxygenase operon (pmoCAB1), corresponding to M. infernorum V4 loci Minf\_1509–1511, displayed the greatest degree of transcriptional upregulation (average: 9.8 Log2FC, pvalue: < 0.0001) in response to O<sup>2</sup> limitation. In contrast, the closely related and immediately proximal pmoCAB2 operon (homologous to M. infernorum V4 loci Minf\_1506–1508) that was highly expressed during O<sup>2</sup> replete growth, was strongly downregulated during O2-limited growth (average: −4.0 Log2FC, p-value: < 0.0001). Interestingly, transcripts for a third putative and relatively divergent pmoCAB3 operon were not detected under the experimental conditions tested.

Expression of the complete complement of genes necessary for the oxidation of CH<sup>4</sup> to CO2, carbon assimilation via the Calvin-Benson-Bassham cycle, and ATP production via aerobic respiration were detected in all conditions tested (**Figure 2C** and **Supplementary Table S1**). However, with exception of a 1.78-fold (p-value: < 0.0001) downregulation of a moxY-like gene (Minf\_1448) that encodes a methanol utilization control sensor protein, no significant differences in expression level in these pathways were observed across conditions. With respect to glycogen metabolism, all genes necessary for glycogen synthesis, storage and degradation were expressed, but did not display evidence of transcriptional regulation (**Supplementary Table S1**). Consistent with the onset of H<sup>2</sup> oxidation in chemostat cultures, genes encoding for the large and small subunits of the Group 1d respiratory [NiFe] hydrogenase, and its associated cytochrome b subunit, were upregulated by 3.5-, 3.6-, and 2.9 fold respectively (to a maximum of 3105 FPKM) in response to O<sup>2</sup> limitation (**Figures 2A,C**). Likewise, genes associated with hydrogenase maturation, nickel incorporation and nickel transport were also significantly upregulated (**Supplementary Table S1**). In contrast, the cytosolic Group 3b [NiFe] hydrogenase was constitutively transcribed at low levels (FPKM < 125).

We further analyzed RNA-seq data to investigate the influence of nitrogen limitation on transcriptional regulation within Methylacidiphilum sp. RTK17.1. During chemostat cultivation in the absence of a supplied nitrogen source (NH<sup>4</sup> <sup>+</sup>), 66 genes were upregulated (p-value: < 0.001; FDR: < 0.02) and 13 genes were downregulated (p-value: < 0.001; FDR: < 0.05) when compared to nitrogen excess growth conditions. Consistent with the onset of N<sup>2</sup> fixation, a seven-gene operon encoding nitrogenase structural subunits and cofactor biosynthesis proteins (nifHDKENX, Minf\_1870–1876) was upregulated (average 6.4 Log2FC; **Figures 2B,C**). In addition, numerous other genes involved in nitrogenase transcriptional regulation (nifA, Minf\_0464), cofactor biosynthesis (nifB, Minf\_0453), stabilization (nifW, Minf\_0472), and nitrate (Minf\_1096) and NH<sup>4</sup> <sup>+</sup> transport (Minf\_1075) were also significantly upregulated (**Supplementary Table S1**). Hydrogenase expression was not significantly different between the conditions, being high for the Group 1d [NiFe]-hydrogenase and low for the Group 3b [NiFe]-hydrogenase.

#### DISCUSSION

The accumulation and storage of carbon and energy as polymeric reserves is a common strategy employed by microorganisms during unbalanced growth to fortify them against periods of environmental starvation (Wilson et al., 2010). Nitrogen limitation is often cited as triggering the accumulation of carbon-rich reserve polymers (Wanner and Egli, 1990), but there is a lack of detailed understanding of the underlying mechanisms responsible for their production. As with many heterotrophic species, methanotrophs often produce carbon-rich polymers; however despite the prevalence of glycogen production within the Gammaproteobacteria methanotrophs (Pieja et al., 2011a), research has primarily focused on the physiology of polyhydroxybutyrate storage in the Alphaproteobacteria methanotrophs (Pieja et al., 2011a,b; Sundstrom and Criddle, 2015). The requirement for organic carbon compounds to

provide both the respiratory energy and carbon necessary for anabolic processes for methanotrophs makes it difficult to untangle the roles of nitrogen, carbon, and energy availability (e.g., ATP) in the production of intracellular glycogen. However, as verrucomicrobial methanotrophs fix CO<sup>2</sup> for carbon and supplement their energy requirements via the oxidation of H<sup>2</sup> (Mohammadi et al., 2016; Carere et al., 2017), this affords an opportunity (obfuscated by the Type I and II methanotrophs) to investigate the influence of nitrogen, carbon, and energy availability independently.

Our findings show that H<sup>2</sup> oxidation influences the production of intracellular glycogen reservoirs in the thermoacidophilic methanotroph, Methylacidiphilum sp. RTK17.1. During chemostat experiments, the maximum glycogen content of Methylacidiphilum sp. RTK17.1 occurred within cells grown in the absence of H2, under nitrogen- and O2- limiting growth conditions [48.86 (± 4.32)%; **Table 1** and **Figure 1A**]. With respect to other studies reporting on the production of carbon storage polymers in methanotroph species, the glycogen content values observed for Methylacidiphilum sp. RTK17.1 are generally congruent. In the closely related thermoacidophile, Methylacidiphilum fumarolicum SolV, a maximum glycogen content of 36% (w/w) was observed (Khadem et al., 2012a) in nitrogen-limited, batch grown cells and a similar value (33% w/w) has been reported in the halotolerant methanotroph Methylotuvimicrobium alkaliphilum 20Z [formerly Methylomicrobium alkaliphilum 20Z (Khmelenina et al., 1999; Orata et al., 2018)]. A maximum of 42.8 (± 17.5)% (w/w) glycogen has been reported in the industrially promising methanotroph, Methylotuvimicrobium buryatense 5GB1 [formerly Methylomicrobium buryatense 5GB1 (Orata et al., 2018)], during batch-growth on methanol, with up to 13.1 (± 4.0)% (w/w) glycogen reported during O2-limited chemostat growth on methane (Gilman et al., 2015).

The variability in reported glycogen content within methanotrophs (intraspecies) and between experimental

in the presence of ammonium (NH<sup>4</sup> <sup>+</sup>) or under N2-fixing conditions. (A) Volcano plot showing differential gene expression changes following the transition from oxygen replete to O2-limited growth. (B) Volcano plot showing differential gene expression changes following the transition from nitrogen-replete (NH<sup>4</sup> <sup>+</sup>) to N2-fixing growth conditions. Both volcano plots compare data generated from the same five transcriptomes with fold-change values (log2FC) and false discovery rates (FDR) calculated using O2-replete and nitrogen excess as reference conditions, respectively. Each gene is represented by a gray dot and genes of interest are highlighted as per the legend. (C) Heat map of transcript abundance for key genes encoding the structural subunits of enzymes participating in methane oxidation (pmoBAC; particulate methane monooxygenase), methanol oxidation (xoxFJ; methanol dehydrogenase, pqqABCDE; pyrroloquinoline biosynthesis), formate oxidation (fdsDAB; formate dehydrogenase), carbon-dioxide fixation (cbbsSL; Rubisco), hydrogen metabolism (hyaBA and hyhSL; encoding group 1d and 3b [NiFe]-hydrogenases respectively), nitrogen fixation (nifXNEKDH; nitrogenase), and glycogen biosynthesis (glgAABCPPP1). The fragment counts per kilobase million transcripts (FPKM) are shown for steady-state cultures. O2-replete, O2-limiting, nitrogen-replete (NH<sup>4</sup> <sup>+</sup>), N2-fixing and the supplementation of H<sup>2</sup> into the feed gas during chemostat operation is indicated.

trials (interspecies) is almost certainly a consequence of both underlying physiological characteristics and the inherent challenges associated with characterizing dynamic batch growth environments. We therefore sought to perform a series of carbon-excess (CO<sup>2</sup> and CH4) steady-state experiments to gain insight into the mechanisms governing glycogen production in Methylacidiphilum sp. RTK17.1. With respect to O<sup>2</sup> and nitrogen limitation, rates of growth (inferred from observed biomass productivity rates), and changes to glycogen, and to a lesser extent total protein and amino acid contents, were consistent with a cell's expected response to unbalanced growth conditions. While previous studies have reported significant changes to the amino acid composition of Staphylococcus aureus cultures in response to variable environmental conditions (Alreshidi et al., 2015, 2016), we observed no change to the relative abundance of specific amino acid residues in Methylacidiphilum sp. RTK17.1 cultures under the conditions tested. Nevertheless, a 144% increase in glycogen content was observed during unbalanced growth following the removal of H<sup>2</sup> gas supply. Depriving Methylacidiphilum sp. RTK17.1 cultures of the respiratory energy gains afforded from H<sup>2</sup> gas oxidation is demonstrative of how this strain dynamically allocates carbon, nitrogen and energy resources.

The observation that Methylacidiphilum sp. RTK17.1 cells produce glycogen and grow more slowly in response to oxygen limitation is consistent with the occurrence of glycogen

within the obligate chemolithoautotroph Hydrogenovibrio marinus when grown on H<sup>2</sup> and CO<sup>2</sup> under O2-limiting conditions (Nishihara et al., 1993). Similarly, production of polyhydroxybutyrate (PHB) has been reported within heterotrophically grown cultures of Azotobacter beijerinckii in response to oxygen limitation (Senior et al., 1972). We speculate that in the absence of sufficient oxygen, both glycogen and PHB reserves likely serve to not only store carbon and energy, but to maintain intracellular redox state. It is also plausible that the anabolic activities required for cell division (i.e., protein, DNA and RNA synthesis) were constrained by ATP availability under O2-limiting conditions. Given protein synthesis requires approximately 19 times more ATP (mmol ATP (g macromolecule)−<sup>1</sup> ) than for saccharide polymerization (Stouthamer, 1979; Russell and Cook, 1995; Russell, 2007), even considering the ATP requirements of CO<sup>2</sup> fixation, glycogen biosynthesis likely represents an energetic 'cost' savings for Methylacidiphilum sp. RTK171 compared to the ATP-demands of cell growth. These modest increases to intracellular glycogen content in response to O<sup>2</sup> limitation are unlikely to negatively impact biomass yields (YATP; Russell and Cook, 1995) while also benefiting cell survivability during periods of starvation. The additional burden imposed by nitrogen limitation not only created unbalanced growth conditions with respect to carbon and nitrogen, but also increased the cell's ATP requirement via the nitrogenase reaction (N<sup>2</sup> + 8H<sup>+</sup> + 16ATP → 2NH<sup>3</sup> + H<sup>2</sup> + 16ADP). Under these growth conditions, glycogen accounted for nearly half of Methylacidiphilum sp. RTK17.1 cell mass. Supplementing CH<sup>4</sup> oxidation with an alternative source of respiratory energy (H2), however, was sufficient to offset the ATP burden imposed by N<sup>2</sup> fixation and consequently the production of intracellular glycogen was reduced and growth rates increased.

An alternative explanation for our chemostat observations is that synthesis of glycogen during energy-limiting conditions serves as a strategy for 'metabolic anticipation'. The combined conditions of low O2, nitrogen, and H<sup>2</sup> availability are highly limiting for a cell and further resource deprivation is likely to trigger a transition from growth to persistence. Thus, disproportionately allocating biomass into storage compounds under this condition may serve as a 'bet-hedging' strategy to enable longer-term survival when conditions worsen. Indeed, the synthesis and storage of intracellular carbon polymers is commonly associated with an increase in viability during periods of environmental starvation. As with PHB, glycogen catabolism supplies reduced electron carriers (e.g., NADH) into the respiratory chain, thereby enabling the continuation of metabolic processes in the absence of an exogenous energy supply (e.g., CH<sup>4</sup> or H2). A reduced lag phase following CH<sup>4</sup> starvation has previously been linked to the catabolism of glycogen reservoirs within the methanotroph M. fumarolicum SolV (Khadem et al., 2012a). Likewise, in Methylacidiphilum sp. RTK17.1 cultures, we interpret depletions in cellular glycogen content throughout prolonged incubations at 4◦C (in the absence of CH4) as evidence it was being consumed to promote survival (**Supplementary Figure S1**). Within oxic/anoxic habitats, it seems evident that Methylacidiphilum sp. RTK17.1 distributes carbon, energy and nitrogen resources during methanotrophic or mixotrophic growth to fulfill the metabolic demands imposed for cell persistence and/or proliferation (Hanson and Hanson, 1996). Similar phenomena of metabolic anticipation have been observed in other species, for example mycobacteria, which accumulate storage compounds such as triacylglycerols during the early hypoxic response (Daniel et al., 2004; Eoh et al., 2017).

While hydrogenase, methane monooxygenase and nitrogenase all displayed evidence of significant transcriptional regulation in response to O<sup>2</sup> and nitrogen limitation, the genes associated with glycogen metabolism were constitutively expressed. These results are consistent with previous findings within the verrucomicrobial methanotrophs (Khadem et al., 2010; Khadem et al., 2012a; Mohammadi et al., 2016; Carere et al., 2017) and suggests the enzymes associated with glycogen metabolism may be allosterically regulated in response to high carbon (i.e., fructose 1,6-bisphosphate) and/or energy contents (i.e., ATP/AMP), as described in other bacterial species (Wilson et al., 2010). Consistent with other verrucomicrobial methanotrophs (Dunfield et al., 2007; Pol et al., 2007; Op den Camp et al., 2009), Methylacidiphilum sp. RTK17.1 also possesses three phylogenetically distinct pmoCAB operons. Based on observed ratios of non-synonymous versus synonymous substitution rates in pmoA orthologs, it has been proposed that the pMMOs encoded in Methylacidiphilum spp. serve functionally distinct roles (Op den Camp et al., 2009). The observation that Methylacidiphilum sp. RTK17.1 transcriptionally regulates pMMO expression in response to oxygen availability therefore supports this hypothesis and is congruent with reports of differential expression in response to oxygen limitation (Khadem et al., 2012a) and during growth on methanol (Erikstad et al., 2012). Finally, it is noteworthy to include that the transcriptional upregulation of the Group 1d [NiFe] hydrogenase occurred in response to O<sup>2</sup> limitation; whereas nitrogenase upregulation was induced by nitrogen availability. The transcriptional decoupling of these two enzymes is further evidence that the physiological role of H<sup>2</sup> oxidation in methanotrophs (Mohammadi et al., 2016; Carere et al., 2017) is distinct from recycling H<sup>2</sup> produced during the nitrogen fixation reaction (Bont, 1976; Dixon, 1976; Chen and Yoch, 1987). Collectively, these findings indicate while H<sup>2</sup> oxidation is sufficient to partially offset the energetic costs associated with N<sup>2</sup> fixation, the regulation of this enzyme is transcriptionally uncoupled from nitrogen availability.

#### DATA AVAILABILITY

The raw and processed transcriptome sequence files (accession numbers GSM3872525–GSM3872529) were deposited into the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih. gov/geo/) for archival storage.

#### AUTHOR CONTRIBUTIONS

CRC, CG, and MS conceived the study. CRC, MS, CG, CC, DG, and BM contributed to the experimental design. CRC,

BM, CC, and HP conducted the bioreactor and wet lab experiments. CRC and HP performed the glycogen analysis. CRC, MS, and CG performed the transcriptome experiments and analysis. CRC, CG, MS, and DG wrote the manuscript with input from BM, HP, and CC.

#### FUNDING

This work was supported by separate MBIE Strategic Science Investment Funds granted to GNS Science (CRC, MS, and HP) and Scion (DG, BM, and CC). CRC and CG were further supported by The Royal Society of New Zealand (Marsden Grant GNS1601). CC, DG, BM, and MS were also supported by an MBIE Smart Ideas Grant (C05X1710). CG was further supported by an ARC DECRA Fellowship (DE170100310).

#### REFERENCES


#### ACKNOWLEDGMENTS

Ngati Tahu–Ng ¯ ati Whaoa are acknowledged as the iwi ¯ having mana whenua (customary rights) over the Rotokawa geothermal field, Methylacidiphilum sp. RTK17.1, and associated microorganisms. We thank Ngati Tahu–Ng ¯ ati Whaoa Runanga ¯ Trust and Tauhara North 2 Trust for their support of our research. We also thank Prof. Peter Dunfield for valuable comments during the preparation of this manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.01873/full#supplementary-material


fumariolicum' SolV. Microbiology 156, 1052–1059. doi: 10.1099/mic.0.036061- 0


**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 Carere, McDonald, Peach, Greening, Gapes, Collet and Stott. 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.

## Putative Iron-Sulfur Proteins Are Required for Hydrogen Consumption and Enhance Survival of Mycobacteria

Aerobic soil bacteria persist by scavenging molecular hydrogen (H2) from the

Zahra F. Islam† , Paul R. F. Cordero† and Chris Greening\*

School of Biological Sciences, Monash University, Clayton, VIC, Australia

#### Edited by:

Patricia Coutinho Dos Santos, Wake Forest University, United States

#### Reviewed by:

Oliver Lenz, Technische Universität Berlin, Germany Constanze Pinske, Martin Luther University of Halle-Wittenberg, Germany Katherine A. Black, Cornell University, United States

> \*Correspondence: Chris Greening

chris.greening@monash.edu

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Microbial Physiology and Metabolism, a section of the journal Frontiers in Microbiology

> Received: 25 April 2019 Accepted: 12 November 2019 Published: 22 November 2019

#### Citation:

Islam ZF, Cordero PRF and Greening C (2019) Putative Iron-Sulfur Proteins Are Required for Hydrogen Consumption and Enhance Survival of Mycobacteria. Front. Microbiol. 10:2749. doi: 10.3389/fmicb.2019.02749 atmosphere. This key process is the primary sink in the biogeochemical hydrogen cycle and supports the productivity of oligotrophic ecosystems. In Mycobacterium smegmatis, atmospheric H<sup>2</sup> oxidation is catalyzed by two phylogenetically distinct [NiFe]-hydrogenases, Huc (group 2a) and Hhy (group 1h). However, it is currently unresolved how these enzymes transfer electrons derived from H<sup>2</sup> oxidation into the aerobic respiratory chain. In this work, we used genetic approaches to confirm that two putative iron-sulfur cluster proteins encoded on the hydrogenase structural operons, HucE and HhyE, are required for H<sup>2</sup> consumption in M. smegmatis. Sequence analysis show that these proteins, while homologous, fall into distinct phylogenetic clades and have distinct metal-binding motifs. H<sup>2</sup> oxidation was reduced when the genes encoding these proteins were deleted individually and was eliminated when they were deleted in combination. In turn, the growth yield and long-term survival of these deletion strains was modestly but significantly reduced compared to the parent strain. In both biochemical and phenotypic assays, the mutant strains lacking the putative iron-sulfur proteins phenocopied those of hydrogenase structural subunit mutants. We hypothesize that these proteins mediate electron transfer between the catalytic subunits of the hydrogenases and the menaquinone pool of the M. smegmatis respiratory chain; however, other roles (e.g., in maturation) are also plausible and further work is required to resolve their role. The conserved nature of these proteins within most Hhy- or Huc-encoding organisms suggests that these proteins are important determinants of atmospheric H<sup>2</sup> oxidation.

Keywords: hydrogenase, Mycobacterium, atmospheric H2, iron-sulfur protein, hydrogen cycle

#### INTRODUCTION

Over the last decade, various studies have revealed that aerobic bacteria conserve energy during persistence through aerobic respiration of atmospheric hydrogen (H2) (Constant et al., 2010; Greening et al., 2014b, 2015a; Meredith et al., 2014; Liot and Constant, 2016; Islam et al., 2019). This process is now recognized to be important for biogeochemical and ecological reasons. Gasscavenging soil bacteria serve as the primary sink in the global hydrogen cycle and are responsible

for the net consumption of approximately 70 million tonnes of H<sup>2</sup> each year (Constant et al., 2009; Ehhalt and Rohrer, 2009; Greening et al., 2014a; Piché-Choquette et al., 2018). More recently, it has been inferred that this process supports the productivity and biodiversity of various ecosystems, especially low-carbon soils (Lynch et al., 2014; Kanno et al., 2015; Khdhiri et al., 2015; Greening et al., 2016; Ji et al., 2017; Bay et al., 2018; Kessler et al., 2019; Piché-Choquette and Constant, 2019). Atmospheric H<sup>2</sup> oxidation appears to be a widespread trait among soil bacteria. To date, bacteria from three phyla have been experimentally shown to oxidize atmospheric H2, Actinobacteriota (Constant et al., 2008, 2010; Greening et al., 2014a; Meredith et al., 2014), Acidobacteriota (Greening et al., 2015a; Myers and King, 2016), and Chloroflexota (Islam et al., 2019). However, genomic and metagenomic studies have indicated at least 13 other phyla encode enzymes that can mediate this process (Greening et al., 2016; Carere et al., 2017; Ji et al., 2017; Piché-Choquette et al., 2017).

The genetic basis and physiological role of atmospheric H<sup>2</sup> oxidation is now largely understood. This process has been most comprehensively studied in the genetically tractable soil actinobacterium Mycobacterium smegmatis (Greening and Cook, 2014). In this organism, atmospheric H<sup>2</sup> oxidation is mediated by two membrane-bound, oxygen-tolerant hydrogenases, Huc (group 2a [NiFe]-hydrogenase, also known as Hyd1 or cyanobacterial-type uptake hydrogenase) and Hhy (group 1h [NiFe]-hydrogenase, also known as Hyd2 or actinobacterialtype uptake hydrogenase) (Berney et al., 2014b). Additionally, M. smegmatis encodes a third [NiFe]-hydrogenase, Hyh (Hyd3), which mediates fermentative H<sup>2</sup> production during hypoxia (Berney et al., 2014a). Both H2-oxidizing hydrogenases contain a large subunit containing the [NiFe] active site (HucL, HhyL) and a small subunit containing three iron-sulfur clusters (HucS, HhyS), as well as potential additional subunits (Berney et al., 2014b; Cordero et al., 2019b). These two hydrogenases are upregulated in stationary-phase cells, including in response to organic carbon limitation (Berney and Cook, 2010; Berney et al., 2014b). Consistently, when the structural subunits of these hydrogenases are deleted, strains show reduced growth yield and impaired long-term survival during starvation (Berney and Cook, 2010; Berney et al., 2014a; Greening et al., 2014b). Similar findings have been made in Streptomyces avermitilis; the sole hydrogenase of this organism, Hhy, is exclusively expressed in exospores and strains lacking this enzyme exhibit severe survival defects (Constant et al., 2010; Liot and Constant, 2016). Given these findings, it is proposed that bacteria shift from growing on organic compounds to persisting on atmospheric trace gases. Indeed, theoretical calculations indicate that the energy derived from atmospheric H<sup>2</sup> oxidation (0.53 ppmv) can sustain the maintenance of 10<sup>7</sup> to 10<sup>8</sup> cells per gram of soil (Conrad, 1999).

Despite this progress, little is currently known about the biochemical basis of atmospheric H<sup>2</sup> oxidation. One outstanding question is how electrons derived from H<sup>2</sup> oxidation are transferred to the respiratory chain. Most classes of respiratory uptake hydrogenases are predicted to be co-transcribed with a cytochrome b subunit (Greening et al., 2016; Søndergaard et al., 2016). For example, such subunits interact with the prototypical oxygen-tolerant hydrogenases (group 1d [NiFe]-hydrogenases) of Escherichia coli and Ralstonia eutropha; they anchor the hydrogenase to the membrane and transfer electrons from the hydrogenase small subunit to the quinone pool (Frielingsdorf et al., 2011; Volbeda et al., 2013). However, we did not detect equivalent proteins in the operons encoding the structural subunits of Huc (MSMEG\_2261–2270) or Hhy (MSMEG\_2722 – 2718) in M. smegmatis (**Supplementary Figure S1**) (Berney et al., 2014b). Putative iron-sulfur proteins, tentatively annotated as HucE (MSMEG\_2268) and HhyE (MSMEG\_2718), were encoded downstream of the hydrogenase structural subunits and may potentially fulfill this role instead (Berney et al., 2014b; Greening et al., 2015b). In this work, we characterized the effects of deleting these genes on hydrogenase activity, growth, and survival in M. smegmatis. We also investigated their broader conservation in hydrogenase-encoding bacteria.

#### MATERIALS AND METHODS

#### Bacterial Strains and Growth Conditions

All bacterial strains and plasmids used in this study are listed in **Supplementary Table S1**. Escherichia coli TOP10 was maintained on lysogeny broth (LB) agar plates (10 g L−<sup>1</sup> tryptone, 5 g L−<sup>1</sup> NaCl, 5 g L−<sup>1</sup> yeast extract, 15 g L−<sup>1</sup> agar), while Mycobacterium smegmatis mc<sup>2</sup> 155 (Snapper et al., 1990) and derived mutants were maintained on LB agar plates supplemented with 0.05% (w/v) Tween 80 (LBT). For broth culture, E. coli was grown in LB. M. smegmatis was grown in either LBT or in Hartmans de Bont (HdB) minimal medium (Hartmans and De Bont, 1992) supplemented with 0.2% (w/v) glycerol. In all cases, liquid cultures were grown in rotary incubators at 37◦C with agitation (200 rpm).

#### Mutant Strain Construction

Allelic exchange mutagenesis was used to produce markerless deletions of the genes encoding two putative iron-sulfur proteins, hucE (MSMEG\_2268) and hhyE (MSMEG\_2718) (**Supplementary Figure S2**). Briefly, a fragment containing fused left and right flanks of the MSMEG\_2268 (1800 bp) and MSMEG\_2718 (3098 bp) genes were synthesized by GenScript. These fragments were cloned into the SpeI site of the mycobacterial shuttle plasmid pX33 (Gebhard et al., 2006) to yield the constructs pX33-hucE and pX33-hhyE (**Supplementary Table S1**). These constructs were propagated in E. coli TOP10 and transformed into wild-type M. smegmatis mc<sup>2</sup> 155 cells by electroporation. Gentamycin (5 µg mL−<sup>1</sup> for M. smegmatis or 20 µg mL−<sup>1</sup> for E. coli) was used in selective solid and liquid medium to propagate pX33. Creation of the double iron-sulfur cluster mutant (1hucE1hhyE) was achieved by transformation of 1hhyE electrocompetent M. smegmatis mc<sup>2</sup> 155 with the pX33 hucE construct. Briefly, to allow for permissive temperaturesensitive vector replication, transformants were incubated on LBT gentamicin plates at 28◦C until colonies were visible (5– 7 days). Resultant catechol-positive colonies were subcultured onto fresh LBT gentamicin plates and incubated at 40◦C for 3–5 days to facilitate integration of the recombinant plasmid

flanks into the chromosome. The second recombination event was facilitated by subculturing catechol-reactive and gentamicinresistant colonies onto LBT agar plates supplemented with 10% sucrose (w/v) and incubating at 40◦C for 3–5 days. Catechol-unreactive colonies were subsequently screened by PCR to discern wild-type revertants from 1hucE, 1hhyE and 1hucE1hhyE mutants. Primers used for the generation of mutants and for screening are listed in **Supplementary Table S2**.

#### Complementation Vector Construction

The genes for the putative iron-sulfur proteins were amplified by PCR and the resulting fragments were cloned into the constitutive expression plasmid pMV261 via PstI/HindIII site for hucE and BamHI/HindIII site for hhyE (Stover et al., 1991) to yield the constructs pMVhucE and pMVhhyE (**Supplementary Table S1**). Sequence fidelity of the genes was verified through Sanger sequencing and insertion of the genes into the vector was confirmed through restriction-digestion analysis (**Supplementary Figure S3**). The plasmid constructs were propagated in E. coli DH5α and transformed into M. smegmatis cells by electroporation. Vector pMVhucE was transformed into M. smegmatis wild-type and 1hucE strains, while pMVhhyE was transformed into wild type and 1hhyE mutant. In addition, an empty pMV261 was transformed into wild-type, 1hucE, and 1hhyE strains. These seven M. smegmatis strains were used for complementation experiments in respirometry and activity staining. Kanamycin (20 µg mL−<sup>1</sup> for M. smegmatis or 50 µg mL−<sup>1</sup> for E. coli) was used in selective solid and liquid medium to propagate pMV261. Primers used for the generation of the constructs are listed in **Supplementary Table S2**.

#### Respirometry Measurements

Cultures of wild-type, derived mutants, and complemented mutant strains of M. smegmatis were grown in 125 mL aerated conical flasks containing 30 mL HdB medium supplemented with 0.2% glycerol. Respirometry measurements were performed with mid-stationary phase cells, i.e., 72 h post ODmax (∼3.0). A Unisense H<sup>2</sup> microsensor electrode was polarized at + 800 mV for 1 h using a Unisense multimeter and calibrated against standards of known H<sup>2</sup> concentration. Gas-saturated PBS was prepared by bubbling the solution with 100% (v/v) of either H<sup>2</sup> or O<sup>2</sup> for 5 min. The 1.1 mL microrespiration assay chambers were sequentially amended with stationary-phase cultures (0.9 mL, OD<sup>600</sup> = 3.0), H2-saturated PBS (0.1 mL), and O2-saturated PBS (0.1 mL). Chambers were stirred at 250 rpm, 37◦C. Changes in H<sup>2</sup> concentration were recorded using Unisense Logger Software, and upon observing a linear change in H<sup>2</sup> concentration, rates of consumption were calculated over a period of 20 s, which corresponds to the most linear uptake of hydrogen by the cells. Oxidation rates were normalized against total protein concentration, which was determined by the bicinchoninic acid method (Smith et al., 1985) with bovine serum albumin standards.

#### Activity Staining

Cultures of wild-type, derived mutants, and complemented mutant strains of M. smegmatis were grown in 2.5 L aerated conical flasks containing 500 mL HdB medium supplemented with 0.2% glycerol. For Huc activity staining, cultures of wild-type, 1hucS, 1hucE, 1hucS1hhyL, 1hucE1hhyE, and complemented 1hucE and wild-type M. smegmatis (either with empty pMV261 or complementation vector pMVhucE) were harvested by centrifugation (10,000 × g, 10 min, 4◦C) at earlystationary phase (24 h post ODmax, ∼3.0) (Cordero et al., 2019b). For Hhy activity staining, cultures of wild-type, 1hhyL, 1hhyE, 1hucS1hhyL, 1hucE1hhyE, and complemented 1hhyE and wild-type M. smegmatis (either with empty pMV261 or complementation vector pMVhhyE) were harvested by centrifugation at mid-stationary phase (72 h post ODmax, ∼3.0) (Cordero et al., 2019b). Harvested cultures were washed in phosphate-buffered saline solution (PBS; 137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 2 mM KH2PO4, pH 7.4), and resuspended in 16 mL lysis buffer (50 mM Tris-Cl, pH 8.0, 1 mM PMSF, 2 mM MgCl2, 5 mg mL−<sup>1</sup> lysozyme, 40 µg mL−<sup>1</sup> DNase, 10% glycerol). Resultant cell suspensions were passed through a Constant Systems cell disruptor (40,000 psi, four times), with unbroken cells removed by centrifugation (10,000 × g, 20 min, 4◦C) to yield whole-cell lysates. Protein concentration was determined using a bicinchoninic acid assay with bovine serum albumin standards. Next, 20 µg of each whole-cell lysate was loaded onto two native 7.5% (w/v) Bis-Tris polyacrylamide gels prepared as described elsewhere (Walker, 2009) and run alongside a protein standard (NativeMark Unstained Protein Standard, Thermo Fisher Scientific) for 1.5 h at 25 mA. One gel was stained overnight at 4◦C with gentle agitation using AcquaStain Protein Gel Stain (Bulldog Bio) for total protein determination. The other gel was incubated for hydrogenase activity staining in 50 mM potassium phosphate buffer (pH 7.0) supplemented with 500 µM nitroblue tetrazolium chloride (NBT) in an anaerobic jar amended with an anaerobic gas mixture (5% H2, 10% CO2, 85% N<sup>2</sup> v/v) overnight at room temperature.

#### Growth and Survival Assays

Cultures of wild type and derived mutants of M. smegmatis were inoculated into 125 mL conical flasks containing 30 mL LBT medium (initial OD<sup>600</sup> of 0.001), in six biological replicates. Growth was monitored by measuring optical density at 600 nm (1 cm cuvettes; Eppendorf BioSpectrometer Basic); when OD<sup>600</sup> was above 0.5, cultures were diluted ten-fold in LBT before measurement. Specific growth rate during mid-exponential growth was calculated for each replicate using GraphPad Prism (non-linear regression, exponential growth equation, least squares fit). The long-term survival of the cultures was determined by counting colony forming units (CFU mL−<sup>1</sup> ) of cultures 21 days post-ODmax. Cultures were serially diluted in HdB containing no carbon source and spotted on to agar plates in technical quadruplicates. After incubation at 37◦C for 3 days, the resultant colonies were counted.

#### Sequence and Phylogenetic Analysis

Sequences homologous to M. smegmatis HucE (MSMEG\_2268) and HhyE (MSMEG\_2718) were retrieved by protein BLAST (Altschul et al., 1990) using the National Center

for Biotechnology Information (NCBI) Reference Sequence (RefSeq) database (Pruitt et al., 2007). The retrieved hits were cross-referenced with the hydrogenase database (HydDB) (Søndergaard et al., 2016) in order to determine which organisms co-encode HucE with HucL and HhyE with HhyL. For downstream phylogenetic and motif analysis, sequences were filtered to remove truncated HucE/HhyE proteins and retain one protein sequence per genus. This resulted in a representative subset of 52 full-length HucE and 26 full-length HhyE sequences. The retrieved sequences were aligned using ClustalW in MEGA7 (Kumar et al., 2016). The phylogenetic relationships of these sequences were visualized on a maximum-likelihood tree based on the Poisson correction method and bootstrapped with 100 replicates. In addition, WebLogo (Crooks et al., 2004) was used to analyze the conserved motifs containing cysteine and histidine residues predicted to bind iron-sulfur clusters. The web-based software Properon (M. Milton<sup>1</sup> ) was used to generate to-scale genetic organization diagrams of the group 1h and group 2a [NiFe]-hydrogenases, with genes labeled according to the nomenclature in HydDB (Søndergaard et al., 2016).

#### RESULTS

#### HucE and HhyE Are Predicted to Be Iron-Sulfur Proteins Associated With Group 2a and Group 1h [NiFe]-Hydrogenases

We investigated the diversity of putative iron-sulfur cluster proteins associated with [NiFe]-hydrogenases by conducting a homology-based search using the amino acid sequences of HucE (MSMEG\_2268) and HhyE (MSMEG\_2718) from M. smegmatis (**Supplementary Figure S1**). Homologous sequences were retrieved from 14 phyla and 104 genera of bacteria (**Supplementary Figure S4** and **Supplementary Table S3**).

The evolutionary relationships of these proteins were visualized on a maximum-likelihood phylogenetic tree (**Figure 1**). All retrieved sequences fall into two robustly supported clades, the HucE proteins associated with group 2a [NiFe]-hydrogenases (Huc) and the HhyE proteins associated with group 1h [NiFe]-hydrogenases (Hhy), that share approximately 27% amino acid identity. HhyE proteins were encoded by various atmospheric H<sup>2</sup> oxidizers, including Streptomyces (Berney and Cook, 2010), Rhodococcus (Meredith et al., 2014), Pyrinomonas (Greening et al., 2015a), and Thermogemmatispora (Islam et al., 2019). HucE proteins were encoded by various Cyanobacteria, which are known to recycle H<sup>2</sup> produced during the nitrogenase reaction via group 2a [NiFe]-hydrogenases (Houchins and Burris, 1981; Tamagnini et al., 2002), as well as genera capable of aerobic hydrogenotrophic growth such as Nitrospira (Koch et al., 2014), Pseudonocardia (Grostern and Alvarez-Cohen, 2013), and Acidithiobacillus (Schröder et al., 2007). Of the hydrogenase-positive species surveyed, 9.5% lacked HucE and

Multiple sequence alignments show that HucE and HhyE proteins contain highly conserved motifs potentially involved in binding iron-sulfur clusters (**Supplementary Figures S5, S6**). Both HucE and HhyE contain a CxxC motif within a domain homologous to NifU proteins (Yuvaniyama et al., 2000). The C-terminus of HhyE proteins contains two CxxC motifs typical of iron-sulfur proteins (e.g., rubredoxins). In contrast, the HucE proteins contain an C-terminal motif CxH(x15−18)CxxC that matches the signature motif of Rieske iron-sulfur clusters (Schmidt and Shaw, 2001) (**Supplementary Figure S6**). A subset of the species surveyed contain truncated HucE and HhyE proteins that contain the NifU-like domain, but lack the C-terminal domains (**Supplementary Figures S4, S5**).

#### HucE and HhyE Are Essential for H<sup>2</sup> Oxidation in Mycobacterium smegmatis

We used allelic exchange mutagenesis to generate markerless single and double mutants of the hucE and hhyE genes in M. smegmatis, i.e., 1hucE, 1hhyE, and 1hucE1hhyE. Gene deletion was confirmed by PCR targeting chromosomal sequences adjacent to the flanking regions used for homologous recombination (**Supplementary Figure S2**). Assays were used to compare H<sup>2</sup> oxidation of these strains with the wild-type strain and strains containing previously generated deletions of the hydrogenase structural subunits, i.e., 1hucS, 1hhyL, and 1hucS1hhyL, that lack hydrogenase activity (Berney and Cook, 2010; Berney et al., 2014b; Greening et al., 2014a).

We first used a H<sup>2</sup> electrode to measure rates of aerobic H<sup>2</sup> respiration mediated by whole cells of each strain. There were significant differences in the rate of H<sup>2</sup> oxidation for all deletion strains compared to the wild type (**Figure 2A**). Loss of hucE and hhyE resulted in reductions of 1.8-fold and 8.4-fold, respectively; such reductions were statistically indistinguishable from those observed in the mutants of the hydrogenase structural subunits hucS and hhyL. Deletion of both iron-sulfur proteins (1hucE1hhyE) or both hydrogenase structural subunits (1hucS1hhyL) caused complete cessation of H<sup>2</sup> oxidation, highlighting that these two hydrogenases are solely responsible for H<sup>2</sup> oxidation and that the putative iron-sulfur proteins are indispensable for this process. The low-level negative rates in 1hucE1hhyE and 1hucS1hhyL strains most likely reflect drift of the electrode rather than actual H<sup>2</sup> production by Hyh (Hyd3), since this hydrogenase is only upregulated during hypoxia (Berney et al., 2014a). We successfully complemented the 1hucE and 1hhyE strains by reintroducing the hucE and hhyE genes on the episomal plasmid pMV261 (Stover et al., 1991) (**Figure 2B**); in contrast, introducing the empty vector caused no effect and neither did introducing the complementation vectors in a wild-type background. This restoration of Huc and Hhy activities in complemented iron-sulfur protein deletion mutants strongly indicate that HucE and HhyE are essential for

HhyE, including Thermomicrobium (Islam et al., 2019) and Methylacidiphilum (Mohammadi et al., 2017) species known to synthesize mid-affinity group 1h [NiFe]-hydrogenases. In contrast, no HucE or HhyE sequences were retrieved from organisms that lack hydrogenases.

<sup>1</sup>https://zenodo.org/record/3519494#.Xb4CnJozZPY

The tree was constructed using the maximum-likelihood method (gaps treated with partial deletion), bootstrapped with 100 replicates, and rooted at the mid-point. The sequences used to create this tree are provided in Supplementary Table S3.

H<sup>2</sup> oxidation. Moreover, the similarity in H<sup>2</sup> oxidation rates between the strains containing deletions of the catalytic subunits, compared to the putative iron-sulfur proteins, is consistent with HucE and HhyE being functionally linked with the Huc and Hhy hydrogenases, respectively.

In an interrelated assay, we performed activity staining of the Huc and Hhy hydrogenases using whole-cell lysates of wild-type and deletion mutant strains, with and without the complementation vectors, in the presence of the artificial electron acceptor nitroblue tetrazolium chloride. In the Huc activity staining gel (**Figure 3A**), three bands were observed in the wholecell lysates of wild-type strains, with or without complementation vectors: the top high-MW band, middle mid-MW band, and bottom low-MW band. Both the high-MW and low-MW bands correspond to Huc activity (Cordero et al., 2019b) and these bands were not observed in strains lacking either hucS or hucE. However, Huc activity was restored when the 1hucE strain was complemented by episomal expression of hucE. For Hhy activity

FIGURE 2 | Hydrogen oxidation by wild-type, derived mutants, and complemented mutant strains of M. smegmatis. H<sup>2</sup> uptake by whole cells in mid-stationary phase (72 h post ODmax ∼3.0) was measured amperometrically using a Unisense H<sup>2</sup> electrode. (A) Comparison of the rates of H<sup>2</sup> oxidation between wild-type, single and double mutants of the iron-sulfur proteins (1hucE, 1hhyE, 1hucE1hhyE), and single and double mutants of hydrogenase structural subunits (1hucS, 1hhyL, 1hucS1hhyL). (B) Rates of H<sup>2</sup> oxidation in 1hucE and 1hhyE strains complemented with expression of hucE and hhyE, respectively. Controls include wild-type, 1hucE, and 1hhyE strains transformed with empty vector pMV261 and wild-type strain transformed with complementation vectors pMVhucE and pMVhhyE. Error bars show standard deviations of three biological replicates and values labeled with different letters are significantly different (p < 0.05) based on a one-way ANOVA.

staining (**Figure 3B**), a mid-sized MW band was be observed in all wild-type strains. This band, which is the same middle band observed in the Huc activity stain (**Figure 3A**), corresponds to Hhy activity (Greening et al., 2014a; Cordero et al., 2019b). No Hhy staining was detected with the loss of either hhyL or hhyE, but complementation of the 1hhyE strain with hhyE restored Hhy activity. The similarity in the staining bands observed between 1hucS and 1hucE strains or between 1hhyL and 1hhyE

indicate that the putative iron-sulfur proteins HucE and HhyE, like their respective hydrogenase core subunits HucS and HhyL, are important for hydrogenase activity. The artificial electron acceptor cannot compensate for the loss of HucE/HhyE and neither can HucE for HhyE nor HhyE for HucE. This further supports the model that HucE and HhyE form a functional association with Huc and Hhy, respectively.

#### HucE and HhyE Mutant Strains Have Significant Growth and Survival Defects

Previous genetic studies have shown that the hydrogenases modestly increase growth yield and long-term survival of M. smegmatis (Berney and Cook, 2010; Greening et al., 2014b). We therefore tested whether these findings extended to the putative iron-sulfur proteins by analyzing the growth rate, growth yield, and long-term survival of the seven aforementioned strains when cultured aerobically on rich media (LBT). In line with previous findings (Berney et al., 2014a; Greening et al., 2014b), no significant differences in specific growth rate were observed between the strains (**Figure 4A**). However, there was a 10% reduction in the specific growth yield of the HhyE mutant compared to the wild-type strain (ODmaxwt = 4.19 ± 0.21; ODmax 1hhyE = 3.81 ± 0.09; p = 0.008) (**Figure 4B**). This phenotype extended to the double mutant strain (1hucE1hhyE) and again phenocopied single and double mutants lacking the hhyL gene.

We also tested whether the strains were defective in longterm survival by counting colonies of aerobic cultures 21 days following ODmax. There were significant reductions in the survival of most strains compared to the wild-type (**Figure 4C**). Cell counts were approximately two-fold lower for the 1hhyE and 1hhyL strains (p < 0.02), and four-fold lower for the double mutant strains (p < 0.002), relative to the wild-type. These findings agree with previous reports that atmospheric H<sup>2</sup> oxidation by the hydrogenases enables M. smegmatis to survive energy starvation (Greening et al., 2014b) and further supports that the putative iron-sulfur proteins contribute to this function. For reasons currently unclear, no phenotypes were observed for the 1hucE strain.

#### DISCUSSION

In summary, this study shows that HucE and HhyE are required for the enzymatic activity and physiological function of the mycobacterial uptake hydrogenases. Strains lacking these proteins showed no hydrogenase activity in either amperometric or zymographic assays. Furthermore, they exhibited growth and survival phenotypes similar to those of knockouts of hydrogenase structural subunits (Berney and Cook, 2010; Greening et al., 2014b); as with the structural subunit mutants, these phenotypes are relatively minor, likely reflecting the numerous survival mechanisms present in M. smegmatis such as the ability to persist on carbon monoxide (Cordero et al., 2019a). Despite some sequence similarity between the two proteins, they are non-redundant, as there was no compensation in hydrogenase activity in

the single mutant strains. The genomic survey and phylogenetic analysis indicate that hucE and hhyE genes co-evolved with the genes encoding the structural subunits of the group 2a and group 1h [NiFe]-hydrogenases. Their detection in the genomes of most but not all characterized high-affinity H<sup>2</sup> oxidizers indicate they are important but overlooked mediators of atmospheric H<sup>2</sup> oxidation. They are also associated with the group 2a [NiFe] hydrogenases of H2-recycling Cyanobacteria and various aerobic hydrogenotrophic bacteria that are not currently known to oxidize atmospheric H2.

This study lends some support to the hypothesis that these proteins serve as the immediate electron acceptors for the group 2a and group 1h [NiFe]-hydrogenases. There are broadly five lines of evidence that support this hypothesis: (i) the presence of highly conserved motifs for binding iron-sulfur clusters, (ii) the essentiality of these proteins for the function of these hydrogenases, (iii) their association with the structural rather than maturation operons of the hydrogenases (Berney et al., 2014b), (iv) co-localization of HhyL, HhyS, and HhyE subunits on native polyacrylamide gels (Cordero et al., 2019b), and (v) their genomic association with hydrogenases that lack known electron transfer subunits (e.g., cytochrome b subunits). With the respect to the latter point, it is interesting that these proteins are conserved in Cyanobacteria, given the immediate electron acceptors of their uptake hydrogenases have long remained enigmatic (Tamagnini et al., 2002). It is also notable that HucE proteins encode the signature motifs of a Rieske iron-sulfur cluster. Given their unusual ligands, these clusters have a higher standard redox potential (Eo' > −150 mV) than most ironsulfur clusters (e.g., ferredoxins) (Brown et al., 2008). They would therefore be well-poised to accept the relatively high-potential electrons derived from atmospheric H<sup>2</sup> and transfer them to menaquinone. Consistently, zymographic studies suggest that the high-affinity hydrogenases operate at higher redox potential than prototypical hydrogenases, given they are reactive with the nitroblue tetrazolium (Eo' = −80 mV) but not viologen compounds (Eo' = −360 mV) (Pinske et al., 2012; Greening et al., 2014a).

While this study demonstrates HucE and HhyE are important for mycobacterial hydrogenase activity, further work is ultimately needed to resolve their respective function. While a role in electron transfer is most plausible, we have not demonstrated that these proteins interact with the hydrogenases and it is notable that the artificial electron acceptor nitroblue tetrazolium chloride cannot compensate for their absence. In this regard, other roles are also possible and compatible with the available evidence, for example as specific assembly factors and/or structural scaffolds for the hydrogenases. For example, it has been demonstrated that

#### REFERENCES


a rubredoxin-related protein is important for aerobic maturation of the group 1d [NiFe]-hydrogenase in R. eutropha (Fritsch et al., 2014). Furthermore, it is possible that other hypothetical proteins downstream of HucE and HhyE may also serve as electron acceptor candidates, in particular MSMEG\_2717 that shares homology to PHG067, the proposed electron acceptor of R. eutropha (Schäfer et al., 2013). Biochemical studies, including studying the redox chemistry of these proteins and their interactions with the as-yet-unpurified hydrogenases, are now required to distinguish these possibilities and develop a sophisticated understanding of their function.

#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

#### AUTHOR CONTRIBUTIONS

CG conceived, designed, and supervised the study. CG and ZI were responsible for the phylogenetic analysis and analyzed the data. ZI and PC were responsible for the knockout generation, hydrogen electrode measurements, and activity staining. PC was responsible for the complementation experiments. ZI was responsible for the phenotypic assays. ZI, PC, and CG wrote the manuscript.

#### FUNDING

This work was supported by an ARC DECRA Fellowship (DE170100310; awarded to CG), an Australian Government Research Training Program Stipend (awarded to ZI), and a Monash University Doctoral Scholarship (awarded to PC).

#### ACKNOWLEDGMENTS

We thank Dr. Rhys Grinter and the three peer reviewers for their helpful feedback.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02749/full#supplementary-material




**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 Islam, Cordero and Greening. 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.