Metabolic Reconstruction and Modeling Microbial ... · 07/07/2016 · 7 1. Biosciences Division,...

1

Metabolic Reconstruction and Modeling Microbial Electrosynthesis 1 2Christopher W. Marshall1,3*, Daniel E. Ross7, Kim M. Handley4#, Pamela B. Weisenhorn1, 3Janaka N. Edirisinghe4, Christopher S. Henry4, Jack A. Gilbert1,3, Harold D. May5,6, and 4R. Sean Norman7* 5 61. Biosciences Division, Argonne National Laboratory, Argonne, Illinois, USA; 72. Department of Ecology and Evolution, The University of Chicago, Chicago, Illinois, 8USA; 93. Department of Surgery, The University of Chicago, Chicago, Illinois, USA; 104. Computing, Environment, and Life Sciences, Argonne National Laboratory, Illinois, 11USA; 125. Department of Microbiology and Immunology, Medical University of South Carolina, 13Charleston, South Carolina, USA; 146. Marine Biomedicine and Environmental Science Center, Medical University of South 15Carolina, Charleston, South Carolina, USA; 167. Department of Environmental Health Sciences, Arnold School of Public Health, 17University of South Carolina, Columbia, South Carolina, USA; 18 19#Present address: Kim M. Handley, School of Biological Sciences, The University of 20Auckland, Auckland, New Zealand. 21*Corresponding authors 22Correspondence: CW Marshall, Department of Surgery (Room O-200C, MC 5030), 23University of Chicago, 5841 S Maryland Ave., Chicago, IL 60637. E-mail: [email protected]. RS Norman, Department of Environmental Health 25Sciences, University of South Carolina, 931 Assembly St., Columbia, SC 29208. E-mail: [email protected] 27 28Running title: Modeling a Microbial Electrosynthetic Community 29 30 31Abstract 32Microbial electrosynthesis is a renewable energy and chemical production platform that 33relies on microbial taxa to capture electrons from a cathode and fix carbon. Yet the 34metabolic capacity of multispecies microbial communities on electrosynthetic 35biocathodes remains unknown. We assembled 13 genomes from a high-performing 36electroacetogenic culture, and mapped their transcriptional activity from a range of 37conditions. This allowed us to create a metabolic model of the primary community 38members (Acetobacterium, Sulfurospirillum, and Desulfovibrio). Acetobacterium was the 39primary carbon fixer, and a keystone member of the community. Based on transcripts 40upregulated near the electrode surface, soluble hydrogenases and ferredoxins from 41Acetobacterium and hydrogenases, formate dehydrogenase, and cytochromes of 42Desulfovibrio were essential conduits for electron flow from the electrode into the 43electrosynthetic community. A nitrogenase gene cluster with an adjacent ferredoxin and 44one of two Rnf complexes within the genome of the Acetobacterium were also 45upregulated on the electrode. Nitrogenase is known to serve as a hydrogenase, thereby 46it would contribute to hydrogen production by the biocathode. Oxygenases of 47microaerobic members of the community throughout the cathode chamber, including 48Sulfurospirillum and Rhodobacteraceae, were expressed. While the reactors were 49maintained anaerobically, this gene expression would support anaerobic growth and 50thus electrosynthesis by scrubbing small amounts of O2 out of the reactor. These 51

.CC-BY-NC-ND 4.0 International licenseacertified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under

The copyright holder for this preprint (which was notthis version posted July 7, 2016. ; https://doi.org/10.1101/059410doi: bioRxiv preprint

https://doi.org/10.1101/059410

http://creativecommons.org/licenses/by-nc-nd/4.0/

2

molecular discoveries and metabolic modeling now serve as a foundation for future 52examination and development of electrosynthetic microbial communities. 53 54Introduction 55

A microbial electrosynthesis system (MES) is a bioelectrochemical device that employs 56

microbes to transport electrons from a cathode to protons and/or CO2. Thus, the 57

microbes act as cathode catalysts to generate, or synthesize, a valuable product 58

(Rabaey and Rozendal, 2010). This technology has immense potential, therefore 59

understanding which microbes are capable of cathodic electron transfer and which 60

metabolic pathways are involved in CO2 conversion to chemicals are critical to improving 61

the performance of MESs. Furthermore, this work has broader environmental 62

implications including understanding ecological aspects of one carbon metabolism and 63

extracellular electron transfer relevant to global biogeochemical cycling. Technologically, 64

these MESs could have a significant impact on geoengineering applications such as 65

carbon capture and storage. 66

67

Advances in microbial electrosynthesis and related biocathode-driven processes 68

primarily focus on the production of three compounds: hydrogen 69

(electrohydrogenesis)(Rozendal et al., 2008), methane (electromethanogenesis) (Cheng 70

et al., 2009), and acetate (electroacetogenesis) (Nevin et al., 2010). However, microbial 71

electrosynthesis can be used to generate alcohols and various short-chain fatty acids.\ 72

(Steinbusch et al., 2010). A thorough evaluation of a carbon dioxide fixing aerobic 73

biocathode has also been reported (Wang et al., 2015). To manipulate (and optimize) 74

these systems we must understand the extracellular electron transfer (EET) enzymes or 75

molecules involved in cathode oxidation, and we must determine how these cathodic 76

electron transport components are coupled to energy conservation by the microbial cell. 77

While an increasingly robust body of literature has been established relating to microbial 78



https://doi.org/10.1101/059410


3

communities metabolizing with an anodic electron acceptor (Ishii et al., 2013, 2015; Kiely 79

et al., 2011), much is still unclear about the diverse metabolic capabilities of 80

microorganisms and communities growing on a cathode (Tremblay and Zhang, 2015). 81

We used metagenomics, metatranscriptomics, and metabolic flux modeling to elucidate 82

carbon flux and EET mechanisms. We have deciphered the complex interactions 83

between cooperating and competing members of the microbiome, producing 84

opportunities to improve microbial electrosynthesis as well as understand multi-species 85

biofilm communities. 86

87

Materials and Methods 88

Reactor design and operation 89

The reactors containing the microbial communities used for metagenomics and 90

metatranscriptomics in this study were described in detail in two previous studies 91

(Marshall et al., 2012, 2013). Of the five MES reactors previously described, two 92

reactors were selected for further analysis in this study. The first reactor, closed circuit 93

(CC), was operated for 184 days before termination (MES-BW4 from previous 94

manuscript is CC reactor here). The second reactor, open circuit (OC), was operated in 95

closed circuit mode for 141 days before it was left at open circuit for three hours prior to 96

termination, in order to control for biofilm effects of the closed circuit reactor (MES 1a 97

from previous manuscript is OC reactor. The initial source of microorganisms was from a 98

brewery wastewater retention basin that was then subjected to sequential transfers of 99

the graphite and supernatant in bioelectrochemical systems before inoculation into the 100

two MESs described in this study. The brewery wastewater inoculum was allowed to 101

establish on graphite granule cathodes and then repeatedly washed with fresh defined 102

medium containing per liter: 2.5 g NaHCO3, 0.6 g NaH2PO4 * H2O, 0.25 g NH4Cl, 0.212 g 103

MgCl2, 0.1 g KCl, 0.03 g CaCl2, 20 mL vitamin solution, and 20 mL mineral solution. 104



https://doi.org/10.1101/059410


4

Subsequent, sludge-free transfers were made into 150 mL volume cathode chambers of 105

MESs containing 30 g of graphite granules, 75 mL of medium, 80% N2 and 20% CO2 or 106

100% CO2 in the headspace, and poised at -590 mV vs. standard hydrogen electrode 107

(SHE) for the duration of the experiments unless otherwise noted. The medium during 108

the initial startup period did not contain sodium 2-bromoethanesulfonate (2-BES), but 10 109

mM 2-BES was periodically used to inhibit methanogenesis and maintain a 110

predominantly acetogenic culture later in the study, including at the time of sampling for 111

metagenome and metatranscriptome analyses (day 181). Upon termination of the 112

experiment, supernatant and bulk granular cathodes were removed for DNA and RNA 113

extraction. All samples were frozen in liquid nitrogen within 5 minutes of applied voltage 114

interruption, except OC which was deliberately left at open circuit for 3 hours prior to 115

freezing. A schematic of the experimental design can be found in Supplementary Figure 116

1. 117

118

DNA/RNA extraction and sequencing 119

DNA and RNA were processed as previously reported and is detailed in the 120

Supplementary Online Methods. Whole genome shotgun sequencing of the cathode-121

associated microbial community from the closed circuit reactor (CCc) and the 122

supernatant community from the closed circuit reactor (CCs) was accomplished with the 123

Illumina MiSeq instrument yielding 250 bp paired-end reads totaling 9,793,154 reads 124

with a total length of 2,325,061,111 bp for the attached microbial community, and 125

5,049,632 reads with a total length of 1,198,063,143 bp for the supernatant microbiome. 126

127

Metagenome assembly and binning 128



https://doi.org/10.1101/059410


5

Two metagenomes were assembled and annotated from the closed circuit reactor (CCc 129

and CCs). Near full-length 16S rRNA sequences from the CCc and CCs metagenomes 130

were assembled over 100 iterations of the Expectation-Maximization Iterative 131

Reconstruction of Genes from the Environment (EMIRGE) method (Miller et al., 2011), 132

following initial read mapping to a modified version of the dereplicated version of the 133

SILVA 108 small-subunit (SSU) rRNA database (Quast et al., 2013). During iterations 134

sequences 97% similar were clustered. Representative sequences were searched 135

against the SSURef_111_NR_trunc database using BLASTN. 136

137

SPAdes (Bankevich et al., 2012), Velvet+Metavelvet (Zerbino and Birney, 2008; Namiki 138

et al., 2012), Ray (Boisvert et al., 2012), and IDBA-UD (Peng et al., 2012) community 139

genome assemblies were compared based on total length of the assembly, n50 score, 140

and percent reads mapping back to the assemblies (Supplementary Table 1). Based on 141

the assembly results, the SPAdes assembly was used to further analyze the 142

metagenomes due to the highest percent of reads mapped to a threshold above 1kb 143

(cathode: 95.95%, supernatant: 91.61%) and total assembly size (cathode: 67,511,580 144

bp above 1kb, supernatant: 42,119,276 bp). Following assembly, the SPAdes contigs 145

were analyzed by Prodigal (Hyatt et al., 2010) to predict protein-encoding genes. 146

Prodigal-predicted amino acid sequences were then annotated using UBLAST searches 147

(usearch64(Edgar, 2010)) against the uniref90 database (Suzek et al., 2015) with an e-148

value cutoff of 100. These annotations, along with kmer coverage and GC content were 149

used for preliminary genome binning. Emergent self-organizing maps (ESOM) were 150

used to refine the metagenome bins based on tetranucleotide frequency and contig 151

coverage on contigs greater than 4kb (Dick et al., 2009). Smaller fragments (>1kb) were 152

then projected onto the 4kb ESOM. To further refine the bins the multi-metagenome 153

pipeline (Albertsen et al., 2013) was used, leveraging differential genome coverages 154



https://doi.org/10.1101/059410


6

between CCc and CCs to aid contig identification. Results were compared to ESOM 155

bins. Only contigs binned by both ESOM and multi-metagenome pipeline were used for 156

reassemblies of individual genome bins. All reads associated with the contigs in each of 157

the final bins were reassembled as single organism genomes by SPAdes using default 158

options with the addition of the –careful flag and kmer sizes of 65, 77, 99, and 127. All 159

contigs greater than 1kb were uploaded to RAST (Aziz et al., 2008) for final annotation 160

using RASTtk (Brettin et al., 2015). The re-assembled individual genome bins and one 161

large “undetermined” bin were compiled into a combined metagenome for transcript 162

mapping. See Supplementary Table 2 for bin summaries and completeness generated 163

by CheckM (Parks et al., 2015). The RASTtk annotations were compared to annotations 164

(Wu et al., 2011) from Pfam, COG, and a Blastx search of the non-redundant protein 165

sequences database (Camacho et al., 2009). The RAST-annotated protein encoding 166

gene (peg) tags are used as identifiers throughout the manuscript with the bin 167

abbreviation preceding the peg number. See Supplementary Table 3 for all annotations. 168

169

Metatranscriptome 170

Four total metatranscriptomes were recovered from CCc, CCs, OCc, and OCs. The 171

FASTX toolkit was used to filter the reads with a quality score lower than 30 and a length 172

less than 50bp. Quality filtered reads from each sample were mapped to the CCc 173

electrode metagenome contigs using BWA (Li and Durbin, 2009). Raw hit counts 174

mapped to each RAST-annotated protein encoding gene in the combined metagenome 175

were then used to calculate reads per kilobase gene length per millions of mapped reads 176

(RPKM) values. RPKM values for each gene were also normalized to bin coverage 177

based on single copy recA gene abundances (Supplementary Figure 6). This 178

compensated for the abundance variations of the taxa between samples. Quantile 179

normalization was used to evenly distribute the transcript values between sample 180



https://doi.org/10.1101/059410


7

conditions. Fold change for differential expression analysis between conditions was 181

calculated by dividing RPKM/RecA-condition1 by RPKM/RecA-condition2. To eliminate 0 182

value errors and over interpretation of very low count genes, an empirically determined 183

threshold value of 0.1 was added to all of the RPKM/RecA values. 184

185

Metabolic model reconstruction, gapfilling, and curation 186

To construct models of our three highest relative abundance genomes (Acetobacterium, 187

Sulfurospirillum, and Desulfovibrio), we imported the RAST-annotated versions of these 188

genomes into the DOE Systems Biology Knowledgebase (KBase). Once in KBase, we 189

used the Build Metabolic Model app, which is based on the ModelSEED algorithm 190

(Henry et al., 2010), to construct one draft model for each genome. Once the draft 191

models were complete, we curated the models based on literature data, KEGG 192

(Kanehisa and Goto, 2000), and MetaCyc (Karp et al., 2002)(Supplementary table 4c). 193

This curation primarily involved the addition of electron utilization pathways 194

(Acetobacterium), addition of electron transport chain reactions (Sulfurospirillum, and 195

Desulfovibrio), and adjustment of reaction directionality to prevent pathways from 196

proceeding in unphysiological directions. 197

198

Next, we applied the Gapfill metabolic model KBase app to identify and fill all the gaps in 199

the metabolic pathways of our models that might prevent them from producing biomass 200

while operating in one of our hypothesized metabolic roles within the microbial 201

community. In this gap-filling process, the model was augmented to include all the 202

reactions contained in the ModelSEED database (available for download from GitHub 203

(https://github.com/ModelSEED/ModelSEEDDatabase/tree/master/Biochemistry). 204

Additionally, all reactions determined to be thermodynamically reversible (Henry et al., 205

2006, 2009; Jankowski et al., 2008) were adjusted to be reversible. Finally, flux balance 206



https://doi.org/10.1101/059410


8

analysis (FBA) (Orth et al., 2010) was performed to generate a flux profile that produced 207

biomass while minimizing the flux through all reactions and reaction directions that were 208

not contained in the original model, consistent with previously published algorithms 209

(Dreyfuss et al., 2013; Latendresse, 2014). All reactions and reaction directions not 210

included in the original model that had a nonzero flux in were then added to the model 211

as the gap-filling solution. This subsequently permits the growth of the gap-filled model 212

in the specified condition. Acetobacterium was gapfilled in a condition that reflected the 213

only proposed metabolic role for this organism: direct consumption of electrons to fix 214

CO2 coupled to the production of acetate. The Sulfurospirillum and Desulfovibrio models 215

were gapfilled separately in two distinct conditions reflecting the potential roles proposed 216

for these organisms: (i) autotrophic growth on CO2 and H2; or (ii) heterotrophic growth on 217

acetate. This produced a total of four models: (i) autotrophic Sulfurospirillum; (ii) 218

heterotrophic Sulfurospirillum; (iii) autotrophic Desulfovibrio; and (iv) heterotrophic 219

Desulfovibrio. The detailed composition of the electrosynthetic, autotrophic, and 220

heterotrophic media formulations used in the gapfilling analysis are included in the 221

supplementary material (Supplementary Table 4). 222

223

All models were reconstructed, gapfilled, and curated in the KBase Narrative Interface, 224

and the associated narratives include all parameters and media formulations used 225

(https://narrative.kbase.us/narrative/ws.15248.obj.1). All model data is also available for 226

download from the same narratives. The reactions for each model and condition can 227

also be found in Supplementary Table 4a. 228

229

Prediction of metabolic activity with comparison to expression data 230

Flux balance analysis (FBA) was applied with each of our gapfilled metabolic models to 231

simulate the production of biomass on the same media formulations used in our 232



https://doi.org/10.1101/059410


9

gapfilling analysis (electrosynthetic, autotrophic, and heterotrophic media). Thus, we 233

simulated the growth of our gapfilled Acetobacterium model in our electrosynthetic 234

media; we simulated the growth of our autotrophic Sulfurospirillum and Desulfovibrio 235

models in our autotrophic media; and we simulated the growth of our heterotrophic 236

Sulfurospirillum and Desulfovibrio models in our heterotrophic media (all media 237

formulations are listed in Supplementary table 4b). We then maximized the production of 238

biomass in each model, subject to the constraints on nutrient uptake imposed by our 239

media formulations, and once a solution was produced, we minimized the sum of the 240

fluxes comprising the solution, in order to produce a single solution that is both simple 241

and distinct. 242

243

We validated our flux solutions by comparing the active reactions in each flux profile 244

against the actively expressed genes identified based on our transcriptomic data. In 245

order to conduct this comparison, we first needed to classify the genes in our three 246

species as either active or inactive based on the relative abundance of mapped RNA-247

seq reads. This process began by identifying an initial set of universal genes in each 248

species that can be assumed to be always active (e.g. DNA polymerase subunits and 249

tRNA synthetases). These genes are listed in Supplemental table 4d. We ranked these 250

always active genes based on the abundance of reads mapped to them, identifying the 251

number of reads mapped to the gene ranked at the 10th percentile in this list as the 252

threshold expression value. Then in the broader genome, we labeled any gene with 253

mapped reads that exceeded the threshold as active, with all other genes being inactive. 254

Finally, we calculated the binary distance between the transcriptome and flux profiles, 255

excluding any reactions that had been gapfilled, to determine which predicted flux 256

profiles yielded the greatest agreement (i.e. lower binary distance) with the 257

transcriptomic data. 258



https://doi.org/10.1101/059410


10

259

We carried out a comparative model analysis on thirty-three KEGG pathway categories 260

involving gene presence and absence, transcriptomic data from CCc, flux analysis, and 261

gap-filled reactions on each pathway category (Figure 3). Carbon fixation pathway 262

categories (i.e. reductive TCA cycle and Wood- Ljungdahl pathways) were lumped into a 263

single category called CO2 fixation. All of the metabolic models, their associated 264

genomes, and the FBA analysis generated in this study are presented using the KBase 265

Narrative Interface (NI) and are accessible at 266

https://narrative.kbase.us/narrative/ws.15248.obj.1. 267

268

Scanning electron microscopy - energy dispersive x-ray spectroscopy (SEM-EDX) 269

Cathodic graphite granules were fixed in 0.1 M sodium cacodylate buffer with 2% 270

gluteraldehyde for three hours, washed in 2.5% osmium tetroxide, and then dehydrated 271

with an ethanol dilution series using 0%, 25%, 50%, 75%, % and 100%. Samples were 272

stored in a desiccator and then coated with Au and Pd using a Denton Vacuum sputter 273

coater. Images were generated using a FEI Quanta 400 Scanning Electron 274

Microscope. 275

276

Data access 277

The metagenomic and metatranscriptome reads can be found in MG-RAST (Meyer et 278

al., 2008) under the ID numbers (and names) 4673464.3-4673467.3 279

(ARPA_metagenome) for the metagenome 280

(http://metagenomics.anl.gov/linkin.cgi?project=15936) and 4536830.3-4536839.3 281

(ARPA_metatranscriptome) for the metatranscriptomes 282

(http://metagenomics.anl.gov/linkin.cgi?project=6065). Assembled contigs for 283

Acetobacterium, Sulfurospirillum, and Desulfovibrio can be accessed at 284



https://doi.org/10.1101/059410


11

https://narrative.kbase.us/narrative/ws.15248.obj.1. Raw Illumina sequencing reads have 285

been deposited in the NCBI SRA database under Bioproject PRJNA245339. 286

Additionally, sequencing and assembly files can be found online at 287

https://github.com/sirmicrobe/electrosynthesis. 288

289

Results 290

Performance and compositions of microbial electrosynthesis systems 291

Two high performing MESs inoculated from an established electrosynthetic community 292

were poised at -590 mV vs. a standard hydrogen electrode (SHE), and operated for over 293

100 days, generating an average percentage of total products formed of 65% acetate, 294

34% hydrogen, 0.4% formate, 0.3% propionate and 0.2% butyrate from CO2. At the 295

conclusion of the experiment, one of the two reactors was left at open circuit for three 296

hours to distinguish between transcripts influenced by the supply of electrons from the 297

cathode compared to electrons supplied by hydrogen and/or other free metabolites in 298

the biofilm. Similarly, metatranscriptome samples were taken from the electrode surface 299

and compared to the supernatant. This led to four metatranscriptome samples from two 300

reactors: closed circuit cathode (CCc), closed circuit supernatant (CCs), open circuit 301

cathode (OCc), and open circuit supernatant (OCs). Coulombic efficiencies at time of 302

sampling were 77% and 73% from CC and OC, respectively (Figure 1). Thirteen genome 303

bins spanning 5 different phyla were recovered from the assembled metagenomes taken 304

from the closed circuit MES (Figure 2). Near-complete genomes (89-100% complete) 305

were produced for Sulfurospirillum str. MES7, Acetobacterium str. MES1, Desulfovibrio 306

str. MES5, Methanobacterium str. MES13, Bacteroides str. MES9, Geobacter str. MES3, 307

and Sphaerochaeta str. MES8 (Supplementary Table 2), and a second partially 308

complete (~65%) Desulfovibrio str. MES6 genome was further obtained. 309

Acetobacterium, Sulfurospirillum, and Desulfovibrio combined comprised 40-90% of the 310



https://doi.org/10.1101/059410


12

community in each condition, with a Rhodobacteraceae related organism also 311

consistently represented (4-20% relative abundance) (Figure 2, Supplementary Figure 312

2). The 13 annotated genome bins recovered from the CC reactor were used as the 313

basis for transcript mapping and analyses. 314

315

Description of the microbial catalysts 316

Three species, Acetobacterium str. MES1, Sulfurospirillum str. MES7, Desulfovibrio str. 317

MES5, comprised 72% of the total abundance on the active cathode (CCc). The 318

abundance, expression levels, and predominant activity in the reactor (acetogenesis) as 319

well as the known physiology of similar organisms indicates that the Acetobacterium 320

species is the main proponent of carbon fixation on the electrode and thus the keystone 321

species in the microbial community. Genomic and transcriptomic evidence indicates the 322

reconstructed Acetobacterium str. MES1 reduces CO2 to acetate through the Wood-323

Ljungdahl pathway (WLP). WLP-indicative acetyl-CoA synthase/carbon monoxide 324

dehydrogenase (ACS/CODH, aceto.peg.1971-1981) and 5-325

methyltetrahydrofolate:corrinoid iron-sulfur protein methyltransferase (aceto.peg.1978) 326

components were among the most highly expressed genes on the closed circuit 327

electrode surface (Figure 3). Furthermore, the canonical enzyme ACS/CODH had on 328

average >3-fold greater expression on the closed circuit electrode compared to the 329

supernatant as well as higher expression on CCc compared to OCc, indicating that the 330

electrode surface was the major source of acetate production. Interestingly, a 331

Clostridium-type chain elongation pathway(Bruant et al., 2010) was expressed in the 332

Acetobacterium str. MES1 to convert acetyl-CoA to butyrate (aceto.peg.1850-4, 2312), 333

which may explain the butyrate production by the electrosynthetic microbiome 334

(Supplementary Figure 3)(Marshall et al., 2013). 335

336



https://doi.org/10.1101/059410


13

Reducing equivalents for the Wood-Ljungdahl pathway, reduced ferredoxin and NADH, 337

were generated by a soluble uptake hydrogenase complex, hydABCDE 338

(aceto.peg.4025-9), which were the most highly expressed Acetobacterium str. MES1-339

associated hydrogenase genes on the electrode. The high expression of hydABCDE and 340

rnfABCDEG gene clusters, suggests Acetobacterium str. MES1 conserves energy 341

similar to A. woodii.(Schuchmann and Müller, 2014) The exception is that MES1 has two 342

Rnf complex copies, one with 15-fold greater gene expression on the electrode surface 343

(high: aceto.peg.2709-14 and low: aceto.peg.926-931). This expression difference is 344

similar to Azotobacter vinelandii whose copy expression disparity is driven by 345

ammonium depletion.(Curatti et al., 2005) Rnf expression positively correlates with 346

nitrogen fixation (nif) gene expression, which is likely because Rnf is required for stable 347

accumulation of the 4Fe4S clusters in nifH.(Jeong and Jouanneau, 2000; Curatti et al., 348

2005) Acetobacterium str. MES1 maintains a nitrogenase gene cluster (nifBEHN) 349

(aceto.peg.248-54) with an adjacent ferredoxin. In fact nifH was in the top 5 most 350

expressed Acetobacterium str. MES1 genes in all conditions, suggesting N-limitation 351

and/or an involvement in electron transfer (Figure 3). As ammonium is depleted, 352

nitrogenase converts protons to hydrogen in the absence of N2 (Ryu et al., 2014), which 353

may contribute to the measured hydrogenesis. This supports previous observations of a 354

lag time between fresh media exchange and the onset of hydrogenesis, and of a positive 355

correlation between Acetobacterium and hydrogen production (LaBelle et al., 2014). It is 356

also possible that electron transfer from the Rnf complex to the nitrogenase for proton 357

reduction is a source of hydrogen production by the electrosynthetic microbiome in this 358

and related studies (LaBelle et al., 2014), and is a potential target for biotechnological 359

exploitation (Ryu et al., 2014). 360

361



https://doi.org/10.1101/059410


14

Sulfurospirillum str. MES7 remained consistently abundant (>60% relative abundance) in 362

the supernatant despite frequent exchanges with fresh medium. However, the role of 363

Sulfurospirillum in electrosynthesis has remained elusive (Marshall et al., 2012, 2013). 364

Given its prevalence it must fix CO2 and/or utilize acetate as a carbon source. In support 365

of CO2 fixation, Sulfurospirillum spp. are hypothesized to fix CO2(Goris et al., 2014) and 366

other members of the epsilonproteobacteria can fix CO2 via the reductive tricarboxylic 367

acid (rTCA) cycle (Hügler et al., 2005). As shown here and a recent Sulfurospirillum 368

comparative genomic study (Ross et al., 2016), the Sulfurospirillum str. MES7 assembly 369

contains several genes necessary to overcome the irreversible enzymes of the TCA 370

cycle, including a 2-oxoglutarate oxidoreductase alpha, beta, delta, and gamma subunits 371

(EC 1.2.7.3, sulfuro.peg.1237-40) and fumarate reductase (EC 1.3.5.1, sulfuro.peg.592-372

4). No ATP citrate lyase or citryl-CoA lyase were found, but it does have genes for the 373

conversion of citrate to oxaloacetate + acetate (citrate lyase alpha and beta chains, and 374

citrate (pro-3S)-lyase, EC 4.1.3.6, sulfuro.peg.2133-6), suggesting the possibility of CO2 375

fixation through rTCA. This is also supported by the transcriptional activity of key genes 376

in the rTCA cycle (Supplementary Figure 4). In addition, acetate permease 377

(sulfuro.peg.1580) and acetate kinase (sulfuro.peg.1274) are both highly expressed, 378

indicating either that a supplementary organic carbon source might be required for 379

growth, or that Sulfurospirillum str. MES7 flexibly switches between CO2 fixation and 380

acetate oxidation in the MES reactor. 381

382

By far the most highly expressed genes relating to terminal electron accepting processes 383

in Sulfurospirillum str. MES7 (for any condition) were cytochromes, particularly 384

cytochrome c oxidases. C-type heme-copper oxidases are the final step in the catalytic 385

reduction of oxygen in certain bacteria (Lee et al., 2011) and ccoNOPQ is highly 386

expressed by Sulfurospirillum str. MES7 (sulfuro.peg.1785-8) in all MES conditions. The 387



https://doi.org/10.1101/059410


15

high expression of ccoNOPQ suggests Sulfurospirillum str. MES7 is growing 388

microaerobically in the MES, using oxygen as the terminal electron acceptor. This 389

supports the hypothesis that microaerophilic bacteria provide a supportive role to the 390

rest of the community by scrubbing low levels of oxygen that diffuse from the anode 391

chamber. 392

393

Desulfovibrio str. MES5 was the 3rd most abundant taxon, and likely converts CO2 to 394

formate and while using cytochromes, formate dehydrogenase, and/or hydrogenases to 395

accept electrons from the cathode. The high expression of formate dehydrogenase (3-396

fold, dsv-h.peg.194) and cytochromes (6-fold, dsv-h.peg.1195; 6-fold, dsv-h.peg.3868) 397

on the electrode compared to the supernatant supports this hypothesis. The lack of a 398

suitable terminal electron acceptor indicates that Desulfovibrio str. MES5 reduces 399

protons and generates hydrogen gas (Carepo et al., 2002), which is supported by the 400

comparatively high expression of hydrogenases (dsv-h.peg.740-1) on the electrode 401

surface (>3-fold CCc vs. CCs). Furthermore, Desulfovibrio spp. have been shown to 402

interact with a cathode to facilitate electrohydrogenesis.\ (Aulenta et al., 2012; Croese et 403

al., 2011; Yu et al., 2011). It is likely that small amounts of acetate from Acetobacterium 404

str. MES1 are used as a supplementary carbon source for Desulfovibrio str. MES5, but 405

based on the high hydrogenase and low acetate kinase expression pattern the majority 406

of reducing potential comes from the electrode. 407

408

While Acetobacterium str. MES1, Sulfurospirillum str. MES7, and Desulfovibrio str. 409

MES5 can explain most of the activity occurring in the microbial electrosynthesis 410

systems, the remaining genomes have broad metabolic capabilities, but are typically in 411

lower abundance (<7%). We hypothesize that they are consuming detritus, degradation 412

products, and/or short chain fatty acids generated by the abundant organisms. A pan-413



https://doi.org/10.1101/059410


16

genome associated with the Rhodobacteraceae family had an expression profile 414

consistent with acetate oxidation and cbb3-type cytochrome c oxidases that could be 415

indicative of microaerobic growth. The latter have been expressed in Rhodobacter sp. 416

growing in microaerobic and anaerobic conditions (Kaplan et al., 2005) and were also 417

found to be upregulated by Shewanella oneidensis MR-1 growing on electrodes 418

(Rosenbaum et al., 2012). While we are hypothesizing that these cytochrome c oxidases 419

in Rhodobacteraceae str. MES12 and Sulfurospirillum str. MES7 are involved in 420

microaerobic growth, given their prominence in this study and others involving 421

bioelectrochemical systems (Rosenbaum et al., 2012), their role in EET should be 422

investigated further. One of two Bacteroides genomes (str. MES10) was closely related 423

(93% sequence identity) to Proteiniphilum acetatigenes, which has been shown to 424

generate acetate when growing on wastewater cell debris(Chen, 2005) and Croese et al. 425

discovered a relatively high abundance of uncultured Bacteroides in a hydrogen 426

producing biocathode community (Croese et al., 2013), suggesting possible productive 427

roles for the Bacteroides str. MES9 and MES10 in the electrosynthetic microbiome. 428

Interestingly, Bacteroides str. MES9 expresses a butanol dehydrogenase (bacteroid-429

h.peg.1751) on the electrode surface, which suggests that changing the operating 430

conditions of the MES could produce biofuels and other valuable products (see 431

Bioprospecting section in supporting materials). Additionally, both Bacteroides genomes 432

and the Sphaerochaeta str. MES8 genome exhibited high expression of ferredoxin 433

genes (bacteroid-h.peg.3284, bacteroid-l.peg.2157, sphaer.peg.2014), which may be 434

used to shuttle electrons for community metabolism. 435

436

Metabolic model reconstructions and analysis 437

Genome-scale metabolic models were constructed for the three dominate species in the 438

electrosynthetic community (see methods): Acetobacterium str. MES1, Sulfurospirillum 439



https://doi.org/10.1101/059410


17

str. MES7, and Desulfovibrio str. MES5. These models were then used, in combination 440

with flux balance analysis (Orth et al., 2010), to predict the metabolic activity of each of 441

these species during microbial community growth on the electrode. The accuracy of 442

these flux predictions was evaluated by calculating the fraction of active reactions 443

predicted by the models that were associated with actively expressed genes, as 444

determined from the metatranscriptomic data (Table 1 and methods). In this analysis, 445

the Acetobacterium model displayed only one plausible flux profile, which involved 446

carbon fixation and acetate production via the Wood-Ljungdahl pathway using hydrogen, 447

raw electrons, or similar reducing equivalents as an electron source. With this flux 448

profile, there were 338 active reactions with associated genes in Acetobacterium, and for 449

258 (76%) of these reactions, at least one of the associated genes was actively 450

expressed. Most active reactions that lacked expression support were involved in either 451

amino acid biosynthesis or nucleotide metabolisms. The Sulfurospirillum model 452

predicted two alternative theories for the metabolic role of this species: (i) reduction of 453

CO2 through the reductive TCA cycle with hydrogen used as the reducing agent (67% of 454

active reactions associated with at least one expressed gene), and (ii) oxidation of 455

acetate coupled to O2 reduction (66% of active reactions associated with at least one 456

expressed gene). Given the nearly equal agreement of both of these operating with our 457

transcriptome data, and given the high abundance of Sulfurospirillum in our community, 458

it is possible Sulfurospirillum actually performs both roles depending on its context and 459

environment. The Desulfovibrio model also predicted two alternative theories for the 460

metabolic role of this species in our electrosynthetic microbiome: (i) conversion of CO2 461

and electrons/hydrogen to formate (75.7% of active reactions associated with at least 462

one expressed gene), or (ii) consumption of acetate (75.9% of active reactions 463

associated with at least one expressed gene). As with Sulfurospirillum, the expression 464



https://doi.org/10.1101/059410


18

data was equally supportive of both metabolic theories, indicating that Desulfovibrio also 465

performs a combination of carbon-fixation and acetate utilization. 466

467

In addition to evaluating the overall evidence supporting each of our predicted flux 468

profiles, the transcriptome data was also applied to evaluate the flux profiles on a 469

pathway-by-pathway basis, enabling a better understanding of model accuracy, and 470

revealing insights into potential interspecies interactions (see Methods and Figure 4). 471

Generally, this analysis showed a large degree of agreement between predicted flux 472

profiles and expression data, with some notable exceptions: (i) the terpenoid pathways 473

in all our models were gap-filled because the models all include ubiquinone as a 474

component of biomass, but there is little evidence for these pathways in our genomes 475

and it is likely these genomes are functioning anaerobically and do not have to produce 476

ubiquinone; (ii) the Acetobacterium and Desulfovibrio models both appear to overuse 477

their pentose-phosphate-pathways compared to what would be expected based on 478

expression data; (iii) the Desulfovibrio model overuses its thiamin pathway and should 479

actually obtain thiamin from an external source according to expression data; (iv) the 480

Sulfurospirillum model overuses its sulfur and nitrogen metabolism pathways; and (v) all 481

three models appear to underutilize the vitamin B6 and fructose and mannose 482

metabolism pathways. 483

484

The pathway-based model analysis (Figure 4) reveals interactions and dependencies 485

among the top genomes. Desulfovibrio appears to require an external source of 486

histidine, thiamin, riboflavin, and folate, while the other genomes could be sources of all 487

four of these compounds due to depletion or absence in the supplied medium. 488

Acetobacterium appears to be unique in making glutathione, and also has far more 489

representation in carbon fixation than the other genomes. In contrast, Sulfurospirillum 490



https://doi.org/10.1101/059410


19

appears unique in possessing an active glyoxylate metabolism and the most active TCA 491

cycle. It is interesting to note that all three genomes appear to produce most of their own 492

amino acids, vitamins, and nucleotides, with Desulfovibrio being the least prototrophic of 493

the three. Biochemical reaction equations, compounds, associated flux values, and 494

comparative genomics tools including hypothetical gene knockout experiments for all the 495

models can be found at the KBase Narrative Interface: 496

https://narrative.kbase.us/narrative/ws.15248.obj.1. 497

498

Discussion 499

Metabolic analysis and modeling of this unique microbial community capable of 500

converting CO2 into volatile fatty acid provides a detailed look into how communities 501

actively metabolize on an electrosynthetic biocathode. The genome-wide metabolic 502

capabilities of microorganisms in the community were determined, and a putative 503

metabolic model of the primary community members has been summarized (Figure 5). 504

CO2 fixation and the carbon flux through the electrosynthetic microbiome center on the 505

Wood-Ljungdahl pathway of the Acetobacterium genome, components of which were 506

more active on the closed circuit electrode compared to any other condition tested, 507

demonstrating the importance of electrode-associated growth by Acetobacterium. In 508

addition, reducing equivalents in the form of hydrogen and reduced enzymes (eg. 509

ferredoxin) also stem from the Acetobacterium, alongside major contributions from 510

Desulfovibrio and Sulfurospirillum. Other organisms in the community are contributing to 511

overall fitness by scrubbing toxic compounds (e.g. oxygen) and providing nutrient 512

exchange. Given the proper conditions and applied drivers, this electrosynthetic 513

microbial community has a wide range of biotechnological potential, including the 514

production of alcohols, 2,3-butanediol, and polyhydroxyalkonoates (see supplementary 515

note on bioprospecting and Supplementary Table 4 for more detailed information). 516



https://doi.org/10.1101/059410


20

517

Generally, the proposed mechanisms for cathodic electron transfer in the absence of 518

added mediators are analogous to anodic electron transfer, namely that cytochromes 519

are the primary conduit of electrons to and from the microbial cell (Holmes et al., 2006; 520

Leang et al., 2010; Ross et al., 2011; Inoue et al., 2011; Carlson et al., 2012). The role of 521

cytochromes on anodes has been primarily elucidated in pure culture, but 522

metatranscriptomic studies also point to cytochrome involvement in EET to anodes (Ishii 523

et al., 2013, 2015) and even less is known about EET in communities grown on 524

cathodes. In this study, Desulfovibrio str. MES5 has three upregulated cytochromes on 525

the electrode compared to the supernatant: two with a 6-fold increase (dsv-h.peg.1195 526

and dsv-h.peg.3868) and one with a 2.5-fold increase (dsv-h.peg.1225). The latter two 527

were upregulated greater than 2-fold on the closed circuit cathode compared to the open 528

circuit cathode. Additionally, a formate dehydrogenase from Desulfovibrio str. MES5 529

(dsv-h.peg.194) was upregulated nearly 3-fold on the electrode compared to 530

supernatant, >2-fold CCc versus OCc, and could explain formate transiently observed in 531

the supernatant (Supplementary Figure 3) (Marshall et al., 2013)(da Silva et al., 2013). 532

In addition to their involvement in electron transfer at cathode surfaces, cytochromes can 533

also act as natural mediators for hydrogenases (Pereira et al., 1998; Yahata et al., 2006) 534

and could shuttle electrons from the electrode or from electrode-attached hydrogenases 535

to the microbial community. Three NiFe hydrogenases were highly expressed by 536

Desulfovibrio str. MES5 on the electrode compared to the supernatant (dsv-h.peg.740-537

741, >3-fold and 6-fold increase on electrode compared [CCc] to supernatant [CCs]; and 538

dsv-h.peg.1193, 6-fold increase on electrode [CCc] compared to supernatant [CCs]). 539

One of the NiFe hydrogenases (dsv-h.peg.1193) was adjacent to the upregulated high 540

molecular weight cytochrome c (dsv-h.peg.1195). Given the high expression on the 541

cathode compared to supernatant, it is hypothesized that cytochromes, hydrogenases, 542



https://doi.org/10.1101/059410


21

and possibly formate dehydrogenases associated with Desulfovibrio str. MES5 act as 543

electron mediators between the electrode and the cell to deliver reducing equivalents. In 544

addition to the well-established role of cell-bound cytochromes in direct electron 545

transport with an electrode (Bond et al., 2012; Pirbadian et al., 2014), free cytochromes 546

have been shown to be integral to anodic electron transfer while still imbedded in the 547

biofilm matrix, and could be assisting electron transport to the community in this 548

electrosynthetic biofilm (Inoue et al., 2011). 549

550

The recent discovery that extracellular enzymes (likely hydrogenases and formate 551

dehydrogenases) may be present in cathodic biofilms and may catalyze electron transfer 552

into microbially-utilizable substrates (hydrogen and formate) adds a new and important 553

piece to solve the EET mechanistic puzzle (Deutzmann et al., 2015). Of the genes highly 554

expressed on the electrode, several redox-active electron carriers (particularly 555

hydrogenases and ferredoxins) had high differential expression on the poised electrode 556

compared to the supernatant and open circuit electrode. A Sulfurospirillum str. MES7 557

ferredoxin (sulfuro.peg.2444) is >2-fold higher on the electrode compared to the 558

supernatant (1.5x higher CCc vs. OCc) and the low abundance Bacteroides str. MES10 559

contains a ferredoxin (bacteroid-l.peg.2157) that is >9-fold upregulated on the electrode 560

compared to supernatant (1.7x higher CCc vs. OCc). Importantly, a ferredoxin from 561

Acetobacterium str. MES1 (aceto.peg.1713) was 7-fold higher on the electrode (CCc) 562

compared to both supernatant conditions (CCs and OCs). Additionally, the highly 563

expressed Acetobacterium str. MES1 hydrogenase gene cluster, hydABCDE 564

(aceto.peg.4025-4029), had >5-fold higher expression per cell on the electrode 565

compared to the supernatant (1.6x higher CCc vs. OCc), despite the presence of 566

hydrogen in the supernatant. The combination of soluble hydrogenases and ferredoxins 567

on the cathode could facilitate cathodic electron transfer into organisms like 568



https://doi.org/10.1101/059410


22

Acetobacterium that contain no cytochromes and no obvious means of electron transfer 569

through the cell envelope. 570

571

Despite the lack of a redox-active signal in cell-free cyclic voltammograms and a 572

catalytic wave characteristic of direct electron transfer in the electrosynthetic microbial 573

community (Marshall et al., 2013), SEM-EDX images show extracellular material with 574

elemental signatures of common redox-active enzymes (Supplementary Figure 5). 575

Additionally, no reduction in current or shift in the voltammetric peaks were observed 576

when the supernatant was replaced with fresh medium, suggesting tightly bound 577

electron transfer mechanisms. The discovery of metals such as iron and nickel on the 578

electrode surface could be concentrated enzymes, but microbial or pH induced 579

precipitation of the metals as (hydr)oxides, sulfides, carbonates, phosphates, perhaps 580

even as nanoparticles cannot be ruled out. The latter has been hypothesized by Jourdin 581

et al. who observed copper concentrated on the surface of a microbial electroacetogenic 582

cathode (Jourdin et al., 2016). 583

584

Based on the highly expressed redox active proteins on the electrode (all of the proteins 585

mentioned above are in the top 5% of total genes expressed on the electrode), and the 586

consistent hydrogen production in the reactors (Marshall et al., 2012, 2013; LaBelle et 587

al., 2014), extracellular enzymes and/or concentrated metals are likely functionalizing 588

the electrode while hydrogen, ferredoxins, and/or cytochromes act as shuttles for 589

different organisms in the biofilm. This would explain why microbes lacking an outer 590

membrane can thrive on an electrode, and why previous studies ran electrohydrogenic 591

biocathodes that could sustain hydrogen generation in the absence of a carbon source 592

for over 1000 hours (Rozendal et al., 2008). Further studies are underway to fully 593

characterize functionalization of electrodes by microbes so as to optimize this strategy. 594



https://doi.org/10.1101/059410


23

595

The modeling results demonstrate how metabolic models are useful for interpreting 596

expression data in order to predict the role of individual species participating within a 597

mixed microbial community. The model predictions: (i) confirmed the role of 598

Acetobacterium as the primary species utilizing electrons to reduce CO2; (ii) identified 599

the more complex behavior of Sulfurospirillum as both reducing CO2 and utilizing 600

acetate; and (iii) provided more depth and detail into our understanding of the 601

heterotrophic growth of Desulfovibrio. These models refined models should prove to be 602

a resource for ongoing efforts to understand, and ultimately design and refine 603

electrosynthetic microbiomes. 604

605

We have provided insights into the genetic basis of electrosynthetic microbial 606

communities, which can be used for the optimization of microbial electrosynthesis 607

through operational changes and eventual pathway engineering. Furthermore, it was 608

demonstrated that a diverse set of microorganisms could be active in limited niche space 609

with carbon dioxide as the only carbon source and the electrode as the only electron 610

donor. The discovery that predominant members of the community provide an 611

ecosystem service by scrubbing oxygen makes this, and similar, electrosynthetic 612

microbial communities valuable for practical application. If deployed in true carbon-613

capture situations such as power plants or industrial exhaust lines, oxygen scrubbing to 614

protect the electrosynthetic anaerobes will be important. Finally, the comprehensive 615

analysis of the genomes and development of metabolic models provide the framework to 616

boost production rates and elevate this system to a platform chemical synthesis 617

technology. 618

619Acknowledgements 620



https://doi.org/10.1101/059410


24

We thank Drew Latta for SEM-EDX work and Fangfang Xia and Sebastien Boisvert for 621initial metagenome assembly advice. C.W.M. was supported in part by an Argonne 622National Laboratory Director's Fellowship. The project was funded by the Department of 623Energy, Advanced Projects Research Agency – Energy (DE-AR0000089) and Office of 624Naval Research: Grant# N00014-15-1-2219. We also acknowledge the University of 625Chicago Research Computing Center for their support. 626 627Conflict of Interest 628The authors declare no competing financial interests. 629 630References 631 632AlbertsenM,HugenholtzP,SkarshewskiA,NielsenKL,TysonGW,NielsenPH.633(2013).Genomesequencesofrare,unculturedbacteriaobtainedbydifferential634coveragebinningofmultiplemetagenomes.NatBiotechnol31:533–538.635

AulentaF,CatapanoL,SnipL,VillanoM,MajoneM.(2012).LinkingBacterial636MetabolismtoGraphiteCathodes:ElectrochemicalInsightsintotheH2-Producing637CapabilityofDesulfovibriosp.ChemSusChem5:1080–1085.638

AzizRK,BartelsD,BestAA,DeJonghM,DiszT,EdwardsRA,etal.(2008).TheRAST639Server:RapidAnnotationsusingSubsystemsTechnology.BMCGenomics9:75.640

BankevichA,NurkS,AntipovD,GurevichAA,DvorkinM,KulikovAS,etal.(2012).641SPAdes:ANewGenomeAssemblyAlgorithmandItsApplicationstoSingle-Cell642Sequencing.JComputBiol19:455–477.643

BoisvertS,RaymondF,GodzaridisÉ,LavioletteF,CorbeilJ.(2012).RayMeta:644scalabledenovometagenomeassemblyandprofiling.GenomeBiol13:R122.645

BondDR,Strycharz-GlavenSM,TenderLM,TorresCI.(2012).OnElectronTransport646throughGeobacterBiofilms.ChemSusChem5:1099–1105.647

BrettinT,DavisJJ,DiszT,EdwardsRA,GerdesS,OlsenGJ,etal.(2015).RASTtk:A648modularandextensibleimplementationoftheRASTalgorithmforbuildingcustom649annotationpipelinesandannotatingbatchesofgenomes.SciRep5.e-pubaheadof650print,doi:10.1038/srep08365.651

BruantG,LévesqueM-J,PeterC,GuiotSR,MassonL.(2010).GenomicAnalysisof652CarbonMonoxideUtilizationandButanolProductionbyClostridium653carboxidivoransStrainP7T.PLoSONE5:e13033.654

CamachoC,CoulourisG,AvagyanV,MaN,PapadopoulosJ,BealerK,etal.(2009).655BLAST+:architectureandapplications.BMCBioinformatics10:421.656

CarepoM,BaptistaJF,PamplonaA,FauqueG,MouraJJG,ReisMAM.(2002).657HydrogenmetabolisminDesulfovibriodesulfuricansstrainNewJersey(NCIMB658



https://doi.org/10.1101/059410


25

8313)—comparativestudywithD.vulgarisandD.gigasspecies.Anaerobe8:325–659332.660

CarlsonHK,IavaroneAT,GorurA,YeoBS,TranR,MelnykRA,etal.(2012).Surface661multihemec-typecytochromesfromThermincolapotensandimplicationsfor662respiratorymetalreductionbyGram-positivebacteria.ProcNatlAcadSci109:6631702–1707.664

ChenS.(2005).Proteiniphilumacetatigenesgen.nov.,sp.nov.,fromaUASBreactor665treatingbrewerywastewater.IntJSystEvolMicrobiol55:2257–2261.666

ChengS,XingD,CallDF,LoganBE.(2009).DirectBiologicalConversionofElectrical667CurrentintoMethanebyElectromethanogenesis.EnvironSciTechnol43:3953–6683958.669

CroeseE,KeesmanKJ,Widjaja-Greefkes(Aura)H.C.A.,GeelhoedJS,PluggeCM,670SleutelsTHJA,etal.(2013).RelatingMECpopulationdynamicstoanode671performancefromDGGEandelectricaldata.SystApplMicrobiol36:408–416.672

CroeseE,PereiraMA,EuverinkG-JW,StamsAJM,GeelhoedJS.(2011).Analysisof673themicrobialcommunityofthebiocathodeofahydrogen-producingmicrobial674electrolysiscell.ApplMicrobiolBiotechnol92:1083–1093.675

CurattiL,BrownCS,LuddenPW,RubioLM.(2005).Genesrequiredforrapid676expressionofnitrogenaseactivityinAzotobactervinelandii.ProcNatlAcadSciUSA677102:6291–6296.678

DeutzmannJS,SahinM,SpormannAM.(2015).ExtracellularEnzymesFacilitate679ElectronUptakeinBiocorrosionandBioelectrosynthesis.mBio6:e00496-15.680

DickGJ,AnderssonAF,BakerBJ,SimmonsSL,ThomasBC,YeltonAP,etal.(2009).681Community-wideanalysisofmicrobialgenomesequencesignatures.GenomeBiol68210:R85.683

DreyfussJM,ZuckerJD,HoodHM,OcasioLR,SachsMS,GalaganJE.(2013).684Reconstructionandvalidationofagenome-scalemetabolicmodelforthe685filamentousfungusNeurosporacrassausingFARM.PLoSComputBiol9:e1003126.686

EdgarRC.(2010).SearchandclusteringordersofmagnitudefasterthanBLAST.687Bioinformatics26:2460–2461.688

GorisT,SchubertT,GadkariJ,WubetT,TarkkaM,BuscotF,etal.(2014).Insights689intoorganohaliderespirationandtheversatilecatabolismofSulfurospirillum690multivoransgainedfromcomparativegenomicsandphysiologicalstudies.Environ691Microbiol16:3562–3580.692



https://doi.org/10.1101/059410


26

HenryCS,DeJonghM,BestAA,FrybargerPM,LinsayB,StevensRL.(2010).High-693throughputgeneration,optimizationandanalysisofgenome-scalemetabolic694models.NatBiotechnol28:977–982.695

HenryCS,JankowskiMD,BroadbeltLJ,HatzimanikatisV.(2006).Genome-scale696thermodynamicanalysisofEscherichiacolimetabolism.BiophysJ90:1453–1461.697

HenryCS,ZinnerJF,CohoonMP,StevensRL.(2009).iBsu1103:anewgenome-scale698metabolicmodelofBacillussubtilisbasedonSEEDannotations.GenomeBiol10:699R69.700

HolmesDE,ChaudhuriSK,NevinKP,MehtaT,MethéBA,LiuA,etal.(2006).701MicroarrayandgeneticanalysisofelectrontransfertoelectrodesinGeobacter702sulfurreducens.EnvironMicrobiol8:1805–1815.703

HüglerM,WirsenCO,FuchsG,TaylorCD,SievertSM.(2005).Evidencefor704AutotrophicCO2FixationviatheReductiveTricarboxylicAcidCyclebyMembersof705theεSubdivisionofProteobacteria.JBacteriol187:3020–3027.706

HyattD,ChenG-L,LoCascioPF,LandML,LarimerFW,HauserLJ.(2010).Prodigal:707prokaryoticgenerecognitionandtranslationinitiationsiteidentification.BMC708Bioinformatics11:119.709

InoueK,LeangC,FranksAE,WoodardTL,NevinKP,LovleyDR.(2011).Specific710localizationofthec-typecytochromeOmcZattheanodesurfaceincurrent-711producingbiofilmsofGeobactersulfurreducens.EnvironMicrobiolRep3:211–217.712

IshiiS,SuzukiS,Norden-KrichmarTM,TenneyA,ChainPSG,ScholzMB,etal.713(2013).Anovelmetatranscriptomicapproachtoidentifygeneexpressiondynamics714duringextracellularelectrontransfer.NatCommun4:1601.715

IshiiS,SuzukiS,TenneyA,Norden-KrichmarTM,NealsonKH,BretschgerO.(2015).716Microbialmetabolicnetworksinacomplexelectrogenicbiofilmrecoveredfroma717stimulus-inducedmetatranscriptomicsapproach.SciRep5.e-pubaheadofprint,718doi:10.1038/srep14840.719

JankowskiMD,HenryCS,BroadbeltLJ,HatzimanikatisV.(2008).Groupcontribution720methodforthermodynamicanalysisofcomplexmetabolicnetworks.BiophysJ95:7211487–1499.722

JeongH-S,JouanneauY.(2000).EnhancedNitrogenaseActivityinStrainsof723RhodobactercapsulatusThatOverexpressthernfGenes.JBacteriol182:1208–7241214.725

JourdinL,LuY,FlexerV,KellerJ,FreguiaS.(2016).BiologicallyInducedHydrogen726ProductionDrivesHighRate/HighEfficiencyMicrobialElectrosynthesisofAcetate727fromCarbonDioxide.ChemElectroChemn/a-n/a.728



https://doi.org/10.1101/059410


27

KanehisaM,GotoS.(2000).KEGG:KyotoEncyclopediaofGenesandGenomes.729NucleicAcidsRes28:27–30.730

KaplanS,ErasoJ,RohJH.(2005).Interactingregulatorynetworksinthefacultative731photosyntheticbacterium,Rhodobactersphaeroides2.4.1.BiochemSocTrans33:73251–55.733

KarpPD,RileyM,PaleySM,Pellegrini-TooleA.(2002).TheMetaCycDatabase.734NucleicAcidsRes30:59–61.735

KielyPD,ReganJM,LoganBE.(2011).Theelectricpicnic:synergisticrequirements736forexoelectrogenicmicrobialcommunities.CurrOpinBiotechnol22:378–385.737

LaBelleEV,MarshallCW,GilbertJA,MayHD.(2014).InfluenceofAcidicpHon738HydrogenandAcetateProductionbyanElectrosyntheticMicrobiome.PLoSONE9:739e109935.740

LatendresseM.(2014).Efficientlygap-fillingreactionnetworks.BMCBioinformatics74115:225.742

LeangC,QianX,MesterT,LovleyDR.(2010).Alignmentofthec-TypeCytochrome743OmcSalongPiliofGeobactersulfurreducens.ApplEnvironMicrobiol76:4080–4084.744

LeeHJ,GennisRB,ÄdelrothP.(2011).Entranceoftheprotonpathwayincbb3-type745heme-copperoxidases.ProcNatlAcadSci108:17661–17666.746

LiH,DurbinR.(2009).FastandaccurateshortreadalignmentwithBurrows–747Wheelertransform.Bioinformatics25:1754–1760.748

MarshallCW,RossDE,FichotEB,NormanRS,MayHD.(2012).Electrosynthesisof749CommodityChemicalsbyanAutotrophicMicrobialCommunity.ApplEnviron750Microbiol78:8412–8420.751

MarshallCW,RossDE,FichotEB,NormanRS,MayHD.(2013).Long-termOperation752ofMicrobialElectrosynthesisSystemsImprovesAcetateProductionbyAutotrophic753Microbiomes.EnvironSciTechnol47:6023–6029.754

MeyerF,PaarmannD,D’SouzaM,OlsonR,GlassEM,KubalM,etal.(2008).The755metagenomicsRASTserver–apublicresourcefortheautomaticphylogeneticand756functionalanalysisofmetagenomes.BMCBioinformatics9:386.757

MillerCS,BakerBJ,ThomasBC,SingerSW,BanfieldJF.(2011).EMIRGE:758reconstructionoffull-lengthribosomalgenesfrommicrobialcommunityshortread759sequencingdata.GenomeBiol12:R44.760



https://doi.org/10.1101/059410


28

NamikiT,HachiyaT,TanakaH,SakakibaraY.(2012).MetaVelvet:anextensionof761Velvetassemblertodenovometagenomeassemblyfromshortsequencereads.762NucleicAcidsRes40:e155–e155.763

NevinKP,WoodardTL,FranksAE,SummersZM,LovleyDR.(2010).Microbial764Electrosynthesis:FeedingMicrobesElectricityToConvertCarbonDioxideand765WatertoMulticarbonExtracellularOrganicCompounds.mBio1:e00103-10.766

OrthJD,ThieleI,PalssonBØ.(2010).Whatisfluxbalanceanalysis?NatBiotechnol76728:245–248.768

ParksDH,ImelfortM,SkennertonCT,HugenholtzP,TysonGW.(2015).CheckM:769assessingthequalityofmicrobialgenomesrecoveredfromisolates,singlecells,and770metagenomes.GenomeResgr.186072.114.771

PengY,LeungHCM,YiuSM,ChinFYL.(2012).IDBA-UD:adenovoassemblerfor772single-cellandmetagenomicsequencingdatawithhighlyunevendepth.773Bioinformatics28:1420–1428.774

PereiraIAC,RomãoCV,XavierAV,LeGallJ,TeixeiraM.(1998).Electrontransfer775betweenhydrogenasesandmono-andmultihemecytochromesinDesulfovibriossp.776JBICJBiolInorgChem3:494–498.777

PirbadianS,BarchingerSE,LeungKM,ByunHS,JangirY,BouhenniRA,etal.(2014).778ShewanellaoneidensisMR-1nanowiresareoutermembraneandperiplasmic779extensionsoftheextracellularelectrontransportcomponents.ProcNatlAcadSci780111:12883–12888.781

QuastC,PruesseE,YilmazP,GerkenJ,SchweerT,YarzaP,etal.(2013).TheSILVA782ribosomalRNAgenedatabaseproject:improveddataprocessingandweb-based783tools.NucleicAcidsRes41:D590–D596.784

RabaeyK,RozendalRA.(2010).Microbialelectrosynthesis—revisitingthe785electricalrouteformicrobialproduction.NatRevMicrobiol8:706–716.786

RosenbaumMA,BarHY,BegQK,SegrèD,BoothJ,CottaMA,etal.(2012).787TranscriptionalAnalysisofShewanellaoneidensisMR-1withanElectrode788ComparedtoFe(III)CitrateorOxygenasTerminalElectronAcceptor.PLoSONE7:789e30827.790

RossDE,FlynnJM,BaronDB,GralnickJA,BondDR.(2011).Towards791ElectrosynthesisinShewanella:EnergeticsofReversingtheMtrPathwayfor792ReductiveMetabolism.PLoSONE6:e16649.793

RossDE,MarshallCW,MayHD,NormanRS.(2016).ComparativeGenomicAnalysis794ofSulfurospirillumcavoleiMESReconstructedfromtheMetagenomeofan795ElectrosyntheticMicrobiome.PLOSONE11:e0151214.796



https://doi.org/10.1101/059410


29

RozendalRA,JeremiasseAW,HamelersHVM,BuismanCJN.(2008).Hydrogen797ProductionwithaMicrobialBiocathode.EnvironSciTechnol42:629–634.798

RyuM-H,HullNC,GomelskyM.(2014).MetabolicengineeringofRhodobacter799sphaeroidesforimprovedhydrogenproduction.IntJHydrogEnergy39:6384–6390.800

SchuchmannK,MüllerV.(2014).Autotrophyatthethermodynamiclimitoflife:a801modelforenergyconservationinacetogenicbacteria.NatRevMicrobiol12:809–802821.803

daSilvaSM,VoordouwJ,LeitaoC,MartinsM,VoordouwG,PereiraIAC.(2013).804FunctionofformatedehydrogenasesinDesulfovibriovulgarisHildenborough805energymetabolism.Microbiology159:1760–1769.806

SteinbuschKJJ,HamelersHVM,SchaapJD,KampmanC,BuismanCJN.(2010).807BioelectrochemicalEthanolProductionthroughMediatedAcetateReductionby808MixedCultures.EnvironSciTechnol44:513–517.809

SuzekBE,WangY,HuangH,McGarveyPB,WuCH,ConsortiumtheU.(2015).810UniRefclusters:acomprehensiveandscalablealternativeforimprovingsequence811similaritysearches.Bioinformatics31:926–932.812

TremblayP-L,ZhangT.(2015).Electrifyingmicrobesfortheproductionof813chemicals.MicrobPhysiolMetab201.814

WangZ,LearyDH,MalanoskiAP,LiRW,HerveyWJ,EddieBJ,etal.(2015).A815PreviouslyUncharacterized,NonphotosyntheticMemberoftheChromatiaceaeIs816thePrimaryCO2-FixingConstituentinaSelf-RegeneratingBiocathode.ApplEnviron817Microbiol81:699–712.818

WuS,ZhuZ,FuL,NiuB,LiW.(2011).WebMGA:acustomizablewebserverforfast819metagenomicsequenceanalysis.BMCGenomics12:444.820

YahataN,SaitohT,TakayamaY,OzawaK,OgataH,HiguchiY,etal.(2006).Redox821InteractionofCytochromec3with[NiFe]HydrogenasefromDesulfovibriovulgaris822MiyazakiF†.Biochemistry(Mosc)45:1653–1662.823

YuL,DuanJ,ZhaoW,HuangY,HouB.(2011).Characteristicsofhydrogenevolution824andoxidationcatalyzedbyDesulfovibriocaledoniensisbiofilmonpyrolyticgraphite825electrode.ElectrochimicaActa56:9041–9047.826

ZerbinoDR,BirneyE.(2008).Velvet:Algorithmsfordenovoshortreadassembly827usingdeBruijngraphs.GenomeRes18:821–829.828

829 830 831 832



https://doi.org/10.1101/059410


30

Table 1. Transcriptome support for metabolic activity predicted by models 833Species Description Active

reactions^ Supported reactions*

Expressed genes

Acetobacterium C02 fixation to acetate 338/1244 258/338 402/954 Sulfurospirillum Reduction of CO2 328/1143 219/328 252/661 Sulfurospirillum Oxidation of acetate 330/1142 218/330 252/661 Desulfovibrio Conversion of CO2 to formate 322/1124 244/322 425/759 Desulfovibrio Consumption of acetate 320/1124 243/320 425/759 ^Active reactions only include reactions with associated genes (gapfilled reactions 834filtered out) 835 *Supported reactions are active reactions (gapfilled reactions filtered out) associated 836with at least one actively expressed gene 837 838



https://doi.org/10.1101/059410


31

Figure 1. Reactor performance over time. (A) Acetate production in CC and OC – breaks 839in plotted lines indicate an exchange of the medium, (B) acetate production rate in CC 840and OC, and (C) accounting of the coulombs over a seven day period between medium 841exchanges in CC and OC, numbers are percent coulombic efficiency with total percent 842efficiency averaged over the 7 days in parentheses. 843

844

0

25

50

75

100

125

0 50 100 150Time (days)

Ace

tate

Pro

duct

ion

(mM

)

variableCC_acetate

OC_acetate

0

3

6

9

0 50 100 150Time (days)

Ace

tate

Pro

duct

ion

Rat

e (m

M d

ay −

1)

ReactorCC

OC

A B

C

0

1000

2000

3000

4000

178 180 182 184Time (days)

Cou

lom

bs

acetate_cchydrogen_cctotal_cccoulombs.consumed_ccacetate_ochydrogen_octotal_occoulombs.consumed_oc

123%

89%

81%

73%

73%

74%

77% (84%)

66%

72%

74%

73%

73%

72%

73% (72%)



https://doi.org/10.1101/059410


32

Figure 2. (A) Phylogenetic tree of the CCc microbial community using EMIRGE-based 845reconstructed 16S rRNA gene sequences (blue) and relative abundance values in 846parentheses and indicated by the relative size of the blue circle. Sequences were 847aligned using MUSCLE and the evolutionary history was inferred using the Maximum 848Likelihood method based on the Jukes-Cantor model. Bootstrapping support greater 849than 50% is indicated on the tree and is based on 1,000 iterations. (B) ESOM based on 850tetranucleotide frequency in the CC cathode metagenome. 851 852 853

854 855 856

13 Methanobacterium (2.9%)

CP000577 Rhodobacter sphaeroides

8 Sphaerochaeta (1.6%)

U13928 Geobacter sulfurreducens

JX575921 uncultured Petrimonas sp.

X82931 Sulfurospirillum multivorans

EU258744 Mycoplana peli

CP000148Geobacter metallireducens GS-15

AF357916 Sphaerochaeta globosa str. Buddy

AF192154 Desulfovibrio desulfuricans subsp. desulfuricans

KC488627 Ochrobactrum anthropi

GU356639 Rhizobium borbori

13 Methanobacterium2 (1.2%)

AB548675 Dysgonomonas gadei

AB245368 Pedobacter panaciterrae

12 Haematobacter [Rhodobacteraceae](6.8%)

AB669242 uncultured Bacteroidetes bacterium

1 Acetobacterium(30.6%)

AY693830 Porphyromonadaceae uncultured bacterium

3 Geobacter (1.7%)

AF169245 Methanobacterium formicicum

KF309178 Rhodobacter sp. W04

10 Petrimonas [Bacteroid-l](<1%)

AB603520 Desulfovibrio simplex

AB246781 Sulfurospirillum cavolei

AF093061 Methanobacterium palustre

FN646438 uncultured Bacteroidetes bacterium

CP002273 Eubacterium limosum KIST612

AB020205 Sphingobacterium multivorum

CP002987 Acetobacterium woodii DSM 1030

8 Sphaerochaeta2 (1%)

10 Paludibacter [Bacteroid-l](<1%)

AZAO01000007 Desulfovibrio termitidis HI1

KF836224 Sulfurospirillum uncultured bacterium

6 Desulfovibrio-l (4.8%)

2 Desulfuromonadales(<1%)

9 Bacteroid-h (3.1%)

5 Desulfovibrio-h (18.6%)

FN430569 Thioclava sp. JC56

7 Sulfurospirillum (23.9%)

AF276958 Methanobacterium curvum

X96955 Acetobacterium wieringae

CP001197 Desulfovibrio vulgaris str. Miyazaki F

U46860 Geobacter hydrogenophilus

4 Desulfovibrio-ul (<1%)

AB078842 Paludibacter propionicigenes

AF233586 Methanobacterium congolense

HQ012835 Acetobacterium sp. enrichment culture clone DhR2/LM-A07

11 Rhizobium(<1%)

JN944166 Sphaerochaeta sp. GLS2

100

99

10074

86

54

8687

100

10087

73

7354

100

100100

54100

62

80

99 100

100

70 100

100

83

59

100

100

91

77

99

9677

Tree scale: 0.1

A B

122

3

4

56

12

11

78

91013



https://doi.org/10.1101/059410


33

Figure 3. Expression profile and comparative expression of important genes. The green 857panel shows relative expression between microorganisms and the red panel compares 858differential expression between condition 859

860

4.63

CCc vs.CCs

CCc vs.OCs

CCc vs.OCc

Log2 Fold Change

2.91

2.29

1.51

0.37

1.57

0.78

0.57

1.09

1.55

1.18

1.40

2.60

3.06

2.23

3.06

3.65

2.23

3.23

−0.33

0.53

2.83

2.86

1.32

0.11

1.71

0.62

−1.05

1.77

0.48

0.30

1.19

2.23

2.08

1.75

3.58

3.39

2.57

2.46

−0.11

−0.83

electrode_v_sup_reca_log

electrode_v_bsup_reca_log

electrode_v_offelectrode_reca_lo

aceto.peg.1713;Ferredoxin

aceto.peg.4025−9;hydABCDE hydrogenase

aceto.peg.2709−14;rnfABCDEG


aceto.peg.1972−81;CODH/ACS

aceto.peg.250;Nitrogenase (molybdenum−iron) beta chain

sulfuro.peg.2373;Flagellin

sulfuro.peg.2444;Ferredoxin

dsv−h.peg.194−8;Formate dehydrogenase

dsv−h.peg.200;cytochrome c3


dsv−h.peg.3868;Split−Soret cytochrome c

dsv−h.peg.1193−7;NiFe hydrogenase/Carbonic anhydrase

dsv−h.peg.740−1;NiFe hydrogenase

dsv−h.peg.3404;Pyruvate formate−lyase

rhodo.peg.7409;Superoxide dismutase [Mn]

rhodo.peg.4106;Catalase

bacteroid−l.peg.2157;Ferredoxin

bacteroid−h.peg.3284;4Fe−4S ferredoxin iron−sulfur binding

sphaer.peg.2014;Ferredoxin

−4

−2

0

2

4

CCc CCs OCcOCc

Ferredoxin

hydABCDE hydrogenase

rnfABCDEG

rnfABCDEG

CODH/ACS

Nitrogenase beta chain

Flagellin

Ferredoxin

Formate dehydrogenase

cytochrome c3

cytochrome c3

Split−Soret cytochrome c

hydrogenase/Carbonic anhydrase

NiFe hydrogenase

Pyruvate formate−lyase

Superoxide dismutase [Mn]

Catalase

Ferredoxin

Ferredoxin

Ferredoxin

−4

−2

0

2

4

aceto.peg.1713

aceto.peg.4025−9

aceto.peg.2709−14

aceto.peg.926−31

aceto.peg.1972−81

aceto.peg.250

sulfuro.peg.2373

sulfuro.peg.2444

dsv−h.peg.194−8

dsv−h.peg.200

dsv−h.peg.1225

dsv−h.peg.3868

dsv−h.peg.1193−7

dsv−h.peg.740−1

dsv−h.peg.3404

rhodo.peg.7409

rhodo.peg.4106

bacteroid−l.peg.2157

bacteroid−h.peg.3284

sphaer.peg.2014

Expression (RPKM)

OCc OCs

aceto.peg.1713;Ferredoxin

aceto.peg.4025−9;hydABCDE hydrogenase



aceto.peg.1972−81;CODH/ACS

aceto.peg.250;Nitrogenase (molybdenum−iron) beta chain

sulfuro.peg.2373;Flagellin

sulfuro.peg.2444;Ferredoxin

dsv−h.peg.194−8;Formate dehydrogenase



dsv−h.peg.3868;Split−Soret cytochrome c

dsv−h.peg.1193−7;NiFe hydrogenase/Carbonic anhydrase

dsv−h.peg.740−1;NiFe hydrogenase

dsv−h.peg.3404;Pyruvate formate−lyase

rhodo.peg.7409;Superoxide dismutase [Mn]

rhodo.peg.4106;Catalase

bacteroid−l.peg.2157;Ferredoxin

bacteroid−h.peg.3284;4Fe−4S ferredoxin iron−sulfur binding

sphaer.peg.2014;Ferredoxin

05001000

7000

05001000

7000



https://doi.org/10.1101/059410


34

Figure 4. Pathway flux and model agreement with expression data. The degree of 861agreement between the model-based flux predictions and expression data for each of 862the three metabolic models is shown, both for the entire models and broken down by 863categories of metabolism. In the graph, reactions are divided into five categories based 864on their flux and the expression of their associated genes: (i) reactions that are active 865and associated with at least one expressed gene (dark blue); (ii) reactions that are 866inactive and associated only with unexpressed genes (dark red); (iii) reactions that are 867inactive and associated with one or more expressed genes (green); (iv) reactions that 868are active and associated only with unexpressed genes (purple); and (v) gapfilled 869reactions associated with no genes. The dark blue and dark red categories indicate 870agreement between the models and expression data; purple and green categories 871indicate disagreement. 872

8730% 50% 100% 0% 50% 100% 0% 50% 100%

En(remodel

Centralcarbonmetabolism

Fa7yacidmetabolism

Aminoacidmetabolism

Carbonfixa(on

Electrontransport

Nucleicacidbiosynthesis

Aminoacidbiosynthesis

Bvitaminbiosynthesis

Cellwallandlipidbiosynthesis

Fluxandexpression Nofluxorexpression Expression/Noflux Flux/Noexpression Gapfilling

Acetobacterium Sulfurospirillum Desulfovibrio



https://doi.org/10.1101/059410


35

Figure 5. Hypothetical model of key metabolic activities and interactions among 874dominant members of the electrosynthetic microbial community. 875

876 877 878 879 880 881 882 883884885886 887



https://doi.org/10.1101/059410


Metabolic Reconstruction and Modeling Microbial ... · 07/07/2016 · 7 1. Biosciences Division,...

Documents

Transcript of Metabolic Reconstruction and Modeling Microbial ... · 07/07/2016 · 7 1. Biosciences Division,...